commit 93ff8b411a2ac8eb6c3b58d90e2476d3e0a372ec Merge: 810152b20 ce894e20b Author: Mike Perry mikeperry-git@torproject.org Date: Wed Aug 29 17:10:06 2018 +0000
Merge branch 'ticket25573-034' into ticket25573-master
changes/ticket25573 | 5 + src/core/or/circuitbuild.c | 3 +- src/core/or/circuitbuild.h | 3 +- src/core/or/circuitlist.c | 9 + src/core/or/connection_edge.c | 226 ++++++++++++++ src/core/or/connection_edge.h | 11 + src/core/or/half_edge_st.h | 34 ++ src/core/or/origin_circuit_st.h | 4 + src/core/or/relay.c | 75 ++++- src/feature/client/circpathbias.c | 63 ++++ src/feature/client/circpathbias.h | 1 + src/lib/container/smartlist.c | 2 +- src/lib/container/smartlist.h | 2 +- src/test/test_relaycell.c | 640 +++++++++++++++++++++++++++++++++++--- 14 files changed, 1033 insertions(+), 45 deletions(-)
diff --cc src/core/or/circuitbuild.c index 5c3d20944,000000000..205160ed5 mode 100644,000000..100644 --- a/src/core/or/circuitbuild.c +++ b/src/core/or/circuitbuild.c @@@ -1,3010 -1,0 +1,3009 @@@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file circuitbuild.c + * + * \brief Implements the details of building circuits (by chosing paths, + * constructing/sending create/extend cells, and so on). + * + * On the client side, this module handles launching circuits. Circuit + * launches are srtarted from circuit_establish_circuit(), called from + * circuit_launch_by_extend_info()). To choose the path the circuit will + * take, onion_extend_cpath() calls into a maze of node selection functions. + * + * Once the circuit is ready to be launched, the first hop is treated as a + * special case with circuit_handle_first_hop(), since it might need to open a + * channel. As the channel opens, and later as CREATED and RELAY_EXTENDED + * cells arrive, the client will invoke circuit_send_next_onion_skin() to send + * CREATE or RELAY_EXTEND cells. + * + * On the server side, this module also handles the logic of responding to + * RELAY_EXTEND requests, using circuit_extend(). + **/ + +#define CIRCUITBUILD_PRIVATE + +#include "core/or/or.h" +#include "feature/client/bridges.h" +#include "core/or/channel.h" +#include "feature/client/circpathbias.h" +#define CIRCUITBUILD_PRIVATE +#include "core/or/circuitbuild.h" +#include "core/or/circuitlist.h" +#include "core/or/circuitstats.h" +#include "core/or/circuituse.h" +#include "core/or/command.h" +#include "app/config/config.h" +#include "app/config/confparse.h" +#include "core/mainloop/connection.h" +#include "core/or/connection_edge.h" +#include "core/or/connection_or.h" +#include "feature/control/control.h" +#include "lib/crypt_ops/crypto_rand.h" +#include "feature/dircache/directory.h" +#include "feature/client/entrynodes.h" +#include "core/crypto/hs_ntor.h" +#include "core/mainloop/main.h" +#include "feature/nodelist/microdesc.h" +#include "feature/nodelist/networkstatus.h" +#include "feature/nodelist/nodelist.h" +#include "core/crypto/onion.h" +#include "core/crypto/onion_tap.h" +#include "core/crypto/onion_fast.h" +#include "core/or/policies.h" +#include "core/or/relay.h" +#include "core/crypto/relay_crypto.h" +#include "feature/rend/rendcommon.h" +#include "feature/stats/rephist.h" +#include "feature/relay/router.h" +#include "feature/nodelist/routerlist.h" +#include "feature/nodelist/routerparse.h" +#include "feature/nodelist/routerset.h" +#include "feature/client/transports.h" + +#include "core/or/cell_st.h" +#include "core/or/cpath_build_state_st.h" +#include "core/or/entry_connection_st.h" +#include "core/or/extend_info_st.h" +#include "feature/nodelist/node_st.h" +#include "core/or/or_circuit_st.h" +#include "core/or/origin_circuit_st.h" +#include "feature/nodelist/microdesc_st.h" +#include "feature/nodelist/routerinfo_st.h" +#include "feature/nodelist/routerstatus_st.h" + +static channel_t * channel_connect_for_circuit(const tor_addr_t *addr, + uint16_t port, + const char *id_digest, + const ed25519_public_key_t *ed_id); +static int circuit_deliver_create_cell(circuit_t *circ, + const create_cell_t *create_cell, + int relayed); +static crypt_path_t *onion_next_hop_in_cpath(crypt_path_t *cpath); +STATIC int onion_append_hop(crypt_path_t **head_ptr, extend_info_t *choice); +static int circuit_send_first_onion_skin(origin_circuit_t *circ); +static int circuit_build_no_more_hops(origin_circuit_t *circ); +static int circuit_send_intermediate_onion_skin(origin_circuit_t *circ, + crypt_path_t *hop); +static const node_t *choose_good_middle_server(uint8_t purpose, + cpath_build_state_t *state, + crypt_path_t *head, + int cur_len); + +/** This function tries to get a channel to the specified endpoint, + * and then calls command_setup_channel() to give it the right + * callbacks. + */ +static channel_t * +channel_connect_for_circuit(const tor_addr_t *addr, uint16_t port, + const char *id_digest, + const ed25519_public_key_t *ed_id) +{ + channel_t *chan; + + chan = channel_connect(addr, port, id_digest, ed_id); + if (chan) command_setup_channel(chan); + + return chan; +} + +/** Search for a value for circ_id that we can use on <b>chan</b> for an + * outbound circuit, until we get a circ_id that is not in use by any other + * circuit on that conn. + * + * Return it, or 0 if can't get a unique circ_id. + */ +STATIC circid_t +get_unique_circ_id_by_chan(channel_t *chan) +{ +/* This number is chosen somewhat arbitrarily; see comment below for more + * info. When the space is 80% full, it gives a one-in-a-million failure + * chance; when the space is 90% full, it gives a one-in-850 chance; and when + * the space is 95% full, it gives a one-in-26 failure chance. That seems + * okay, though you could make a case IMO for anything between N=32 and + * N=256. */ +#define MAX_CIRCID_ATTEMPTS 64 + int in_use; + unsigned n_with_circ = 0, n_pending_destroy = 0, n_weird_pending_destroy = 0; + circid_t test_circ_id; + circid_t attempts=0; + circid_t high_bit, max_range, mask; + int64_t pending_destroy_time_total = 0; + int64_t pending_destroy_time_max = 0; + + tor_assert(chan); + + if (chan->circ_id_type == CIRC_ID_TYPE_NEITHER) { + log_warn(LD_BUG, + "Trying to pick a circuit ID for a connection from " + "a client with no identity."); + return 0; + } + max_range = (chan->wide_circ_ids) ? (1u<<31) : (1u<<15); + mask = max_range - 1; + high_bit = (chan->circ_id_type == CIRC_ID_TYPE_HIGHER) ? max_range : 0; + do { + if (++attempts > MAX_CIRCID_ATTEMPTS) { + /* Make sure we don't loop forever because all circuit IDs are used. + * + * Once, we would try until we had tried every possible circuit ID. But + * that's quite expensive. Instead, we try MAX_CIRCID_ATTEMPTS random + * circuit IDs, and then give up. + * + * This potentially causes us to give up early if our circuit ID space + * is nearly full. If we have N circuit IDs in use, then we will reject + * a new circuit with probability (N / max_range) ^ MAX_CIRCID_ATTEMPTS. + * This means that in practice, a few percent of our circuit ID capacity + * will go unused. + * + * The alternative here, though, is to do a linear search over the + * whole circuit ID space every time we extend a circuit, which is + * not so great either. + */ + int64_t queued_destroys; + char *m = rate_limit_log(&chan->last_warned_circ_ids_exhausted, + approx_time()); + if (m == NULL) + return 0; /* This message has been rate-limited away. */ + if (n_pending_destroy) + pending_destroy_time_total /= n_pending_destroy; + log_warn(LD_CIRC,"No unused circIDs found on channel %s wide " + "circID support, with %u inbound and %u outbound circuits. " + "Found %u circuit IDs in use by circuits, and %u with " + "pending destroy cells. (%u of those were marked bogusly.) " + "The ones with pending destroy cells " + "have been marked unusable for an average of %ld seconds " + "and a maximum of %ld seconds. This channel is %ld seconds " + "old. Failing a circuit.%s", + chan->wide_circ_ids ? "with" : "without", + chan->num_p_circuits, chan->num_n_circuits, + n_with_circ, n_pending_destroy, n_weird_pending_destroy, + (long)pending_destroy_time_total, + (long)pending_destroy_time_max, + (long)(approx_time() - chan->timestamp_created), + m); + tor_free(m); + + if (!chan->cmux) { + /* This warning should be impossible. */ + log_warn(LD_BUG, " This channel somehow has no cmux on it!"); + return 0; + } + + /* analysis so far on 12184 suggests that we're running out of circuit + IDs because it looks like we have too many pending destroy + cells. Let's see how many we really have pending. + */ + queued_destroys = circuitmux_count_queued_destroy_cells(chan, + chan->cmux); + + log_warn(LD_CIRC, " Circuitmux on this channel has %u circuits, " + "of which %u are active. It says it has %"PRId64 + " destroy cells queued.", + circuitmux_num_circuits(chan->cmux), + circuitmux_num_active_circuits(chan->cmux), + (queued_destroys)); + + /* Change this into "if (1)" in order to get more information about + * possible failure modes here. You'll need to know how to use gdb with + * Tor: this will make Tor exit with an assertion failure if the cmux is + * corrupt. */ + if (0) + circuitmux_assert_okay(chan->cmux); + + channel_dump_statistics(chan, LOG_WARN); + + return 0; + } + + do { + crypto_rand((char*) &test_circ_id, sizeof(test_circ_id)); + test_circ_id &= mask; + } while (test_circ_id == 0); + + test_circ_id |= high_bit; + + in_use = circuit_id_in_use_on_channel(test_circ_id, chan); + if (in_use == 1) + ++n_with_circ; + else if (in_use == 2) { + time_t since_when; + ++n_pending_destroy; + since_when = + circuit_id_when_marked_unusable_on_channel(test_circ_id, chan); + if (since_when) { + time_t waiting = approx_time() - since_when; + pending_destroy_time_total += waiting; + if (waiting > pending_destroy_time_max) + pending_destroy_time_max = waiting; + } else { + ++n_weird_pending_destroy; + } + } + } while (in_use); + return test_circ_id; +} + +/** If <b>verbose</b> is false, allocate and return a comma-separated list of + * the currently built elements of <b>circ</b>. If <b>verbose</b> is true, also + * list information about link status in a more verbose format using spaces. + * If <b>verbose_names</b> is false, give nicknames for Named routers and hex + * digests for others; if <b>verbose_names</b> is true, use $DIGEST=Name style + * names. + */ +static char * +circuit_list_path_impl(origin_circuit_t *circ, int verbose, int verbose_names) +{ + crypt_path_t *hop; + smartlist_t *elements; + const char *states[] = {"closed", "waiting for keys", "open"}; + char *s; + + elements = smartlist_new(); + + if (verbose) { + const char *nickname = build_state_get_exit_nickname(circ->build_state); + smartlist_add_asprintf(elements, "%s%s circ (length %d%s%s):", + circ->build_state->is_internal ? "internal" : "exit", + circ->build_state->need_uptime ? " (high-uptime)" : "", + circ->build_state->desired_path_len, + circ->base_.state == CIRCUIT_STATE_OPEN ? "" : ", last hop ", + circ->base_.state == CIRCUIT_STATE_OPEN ? "" : + (nickname?nickname:"*unnamed*")); + } + + hop = circ->cpath; + do { + char *elt; + const char *id; + const node_t *node; + if (!hop) + break; + if (!verbose && hop->state != CPATH_STATE_OPEN) + break; + if (!hop->extend_info) + break; + id = hop->extend_info->identity_digest; + if (verbose_names) { + elt = tor_malloc(MAX_VERBOSE_NICKNAME_LEN+1); + if ((node = node_get_by_id(id))) { + node_get_verbose_nickname(node, elt); + } else if (is_legal_nickname(hop->extend_info->nickname)) { + elt[0] = '$'; + base16_encode(elt+1, HEX_DIGEST_LEN+1, id, DIGEST_LEN); + elt[HEX_DIGEST_LEN+1]= '~'; + strlcpy(elt+HEX_DIGEST_LEN+2, + hop->extend_info->nickname, MAX_NICKNAME_LEN+1); + } else { + elt[0] = '$'; + base16_encode(elt+1, HEX_DIGEST_LEN+1, id, DIGEST_LEN); + } + } else { /* ! verbose_names */ + elt = tor_malloc(HEX_DIGEST_LEN+2); + elt[0] = '$'; + base16_encode(elt+1, HEX_DIGEST_LEN+1, id, DIGEST_LEN); + } + tor_assert(elt); + if (verbose) { + tor_assert(hop->state <= 2); + smartlist_add_asprintf(elements,"%s(%s)",elt,states[hop->state]); + tor_free(elt); + } else { + smartlist_add(elements, elt); + } + hop = hop->next; + } while (hop != circ->cpath); + + s = smartlist_join_strings(elements, verbose?" ":",", 0, NULL); + SMARTLIST_FOREACH(elements, char*, cp, tor_free(cp)); + smartlist_free(elements); + return s; +} + +/** If <b>verbose</b> is false, allocate and return a comma-separated + * list of the currently built elements of <b>circ</b>. If + * <b>verbose</b> is true, also list information about link status in + * a more verbose format using spaces. + */ +char * +circuit_list_path(origin_circuit_t *circ, int verbose) +{ + return circuit_list_path_impl(circ, verbose, 0); +} + +/** Allocate and return a comma-separated list of the currently built elements + * of <b>circ</b>, giving each as a verbose nickname. + */ +char * +circuit_list_path_for_controller(origin_circuit_t *circ) +{ + return circuit_list_path_impl(circ, 0, 1); +} + +/** Log, at severity <b>severity</b>, the nicknames of each router in + * <b>circ</b>'s cpath. Also log the length of the cpath, and the intended + * exit point. + */ +void +circuit_log_path(int severity, unsigned int domain, origin_circuit_t *circ) +{ + char *s = circuit_list_path(circ,1); + tor_log(severity,domain,"%s",s); + tor_free(s); +} + +/** Return 1 iff every node in circ's cpath definitely supports ntor. */ +static int +circuit_cpath_supports_ntor(const origin_circuit_t *circ) +{ + crypt_path_t *head, *cpath; + + cpath = head = circ->cpath; + do { + /* if the extend_info is missing, we can't tell if it supports ntor */ + if (!cpath->extend_info) { + return 0; + } + + /* if the key is blank, it definitely doesn't support ntor */ + if (!extend_info_supports_ntor(cpath->extend_info)) { + return 0; + } + cpath = cpath->next; + } while (cpath != head); + + return 1; +} + +/** Pick all the entries in our cpath. Stop and return 0 when we're + * happy, or return -1 if an error occurs. */ +static int +onion_populate_cpath(origin_circuit_t *circ) +{ + int r = 0; + + /* onion_extend_cpath assumes these are non-NULL */ + tor_assert(circ); + tor_assert(circ->build_state); + + while (r == 0) { + r = onion_extend_cpath(circ); + if (r < 0) { + log_info(LD_CIRC,"Generating cpath hop failed."); + return -1; + } + } + + /* The path is complete */ + tor_assert(r == 1); + + /* Does every node in this path support ntor? */ + int path_supports_ntor = circuit_cpath_supports_ntor(circ); + + /* We would like every path to support ntor, but we have to allow for some + * edge cases. */ + tor_assert(circuit_get_cpath_len(circ)); + if (circuit_can_use_tap(circ)) { + /* Circuits from clients to intro points, and hidden services to rend + * points do not support ntor, because the hidden service protocol does + * not include ntor onion keys. This is also true for Single Onion + * Services. */ + return 0; + } + + if (circuit_get_cpath_len(circ) == 1) { + /* Allow for bootstrapping: when we're fetching directly from a fallback, + * authority, or bridge, we have no way of knowing its ntor onion key + * before we connect to it. So instead, we try connecting, and end up using + * CREATE_FAST. */ + tor_assert(circ->cpath); + tor_assert(circ->cpath->extend_info); + const node_t *node = node_get_by_id( + circ->cpath->extend_info->identity_digest); + /* If we don't know the node and its descriptor, we must be bootstrapping. + */ + if (!node || !node_has_preferred_descriptor(node, 1)) { + return 0; + } + } + + if (BUG(!path_supports_ntor)) { + /* If we're building a multi-hop path, and it's not one of the HS or + * bootstrapping exceptions, and it doesn't support ntor, something has + * gone wrong. */ + return -1; + } + + return 0; +} + +/** Create and return a new origin circuit. Initialize its purpose and + * build-state based on our arguments. The <b>flags</b> argument is a + * bitfield of CIRCLAUNCH_* flags. */ +origin_circuit_t * +origin_circuit_init(uint8_t purpose, int flags) +{ + /* sets circ->p_circ_id and circ->p_chan */ + origin_circuit_t *circ = origin_circuit_new(); + circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_CHAN_WAIT); + circ->build_state = tor_malloc_zero(sizeof(cpath_build_state_t)); + circ->build_state->onehop_tunnel = + ((flags & CIRCLAUNCH_ONEHOP_TUNNEL) ? 1 : 0); + circ->build_state->need_uptime = + ((flags & CIRCLAUNCH_NEED_UPTIME) ? 1 : 0); + circ->build_state->need_capacity = + ((flags & CIRCLAUNCH_NEED_CAPACITY) ? 1 : 0); + circ->build_state->is_internal = + ((flags & CIRCLAUNCH_IS_INTERNAL) ? 1 : 0); + circ->base_.purpose = purpose; + return circ; +} + +/** Build a new circuit for <b>purpose</b>. If <b>exit</b> + * is defined, then use that as your exit router, else choose a suitable + * exit node. + * + * Also launch a connection to the first OR in the chosen path, if + * it's not open already. + */ +origin_circuit_t * +circuit_establish_circuit(uint8_t purpose, extend_info_t *exit_ei, int flags) +{ + origin_circuit_t *circ; + int err_reason = 0; + int is_hs_v3_rp_circuit = 0; + + if (flags & CIRCLAUNCH_IS_V3_RP) { + is_hs_v3_rp_circuit = 1; + } + + circ = origin_circuit_init(purpose, flags); + + if (onion_pick_cpath_exit(circ, exit_ei, is_hs_v3_rp_circuit) < 0 || + onion_populate_cpath(circ) < 0) { + circuit_mark_for_close(TO_CIRCUIT(circ), END_CIRC_REASON_NOPATH); + return NULL; + } + + control_event_circuit_status(circ, CIRC_EVENT_LAUNCHED, 0); + + if ((err_reason = circuit_handle_first_hop(circ)) < 0) { + circuit_mark_for_close(TO_CIRCUIT(circ), -err_reason); + return NULL; + } + return circ; +} + +/** Return the guard state associated with <b>circ</b>, which may be NULL. */ +circuit_guard_state_t * +origin_circuit_get_guard_state(origin_circuit_t *circ) +{ + return circ->guard_state; +} + +/** Start establishing the first hop of our circuit. Figure out what + * OR we should connect to, and if necessary start the connection to + * it. If we're already connected, then send the 'create' cell. + * Return 0 for ok, -reason if circ should be marked-for-close. */ +int +circuit_handle_first_hop(origin_circuit_t *circ) +{ + crypt_path_t *firsthop; + channel_t *n_chan; + int err_reason = 0; + const char *msg = NULL; + int should_launch = 0; + const or_options_t *options = get_options(); + + firsthop = onion_next_hop_in_cpath(circ->cpath); + tor_assert(firsthop); + tor_assert(firsthop->extend_info); + + /* Some bridges are on private addresses. Others pass a dummy private + * address to the pluggable transport, which ignores it. + * Deny the connection if: + * - the address is internal, and + * - we're not connecting to a configured bridge, and + * - we're not configured to allow extends to private addresses. */ + if (tor_addr_is_internal(&firsthop->extend_info->addr, 0) && + !extend_info_is_a_configured_bridge(firsthop->extend_info) && + !options->ExtendAllowPrivateAddresses) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Client asked me to connect directly to a private address"); + return -END_CIRC_REASON_TORPROTOCOL; + } + + /* now see if we're already connected to the first OR in 'route' */ + log_debug(LD_CIRC,"Looking for firsthop '%s'", + fmt_addrport(&firsthop->extend_info->addr, + firsthop->extend_info->port)); + + n_chan = channel_get_for_extend(firsthop->extend_info->identity_digest, + &firsthop->extend_info->ed_identity, + &firsthop->extend_info->addr, + &msg, + &should_launch); + + if (!n_chan) { + /* not currently connected in a useful way. */ + log_info(LD_CIRC, "Next router is %s: %s", + safe_str_client(extend_info_describe(firsthop->extend_info)), + msg?msg:"???"); + circ->base_.n_hop = extend_info_dup(firsthop->extend_info); + + if (should_launch) { + if (circ->build_state->onehop_tunnel) + control_event_bootstrap(BOOTSTRAP_STATUS_CONN_DIR, 0); + n_chan = channel_connect_for_circuit( + &firsthop->extend_info->addr, + firsthop->extend_info->port, + firsthop->extend_info->identity_digest, + &firsthop->extend_info->ed_identity); + if (!n_chan) { /* connect failed, forget the whole thing */ + log_info(LD_CIRC,"connect to firsthop failed. Closing."); + return -END_CIRC_REASON_CONNECTFAILED; + } + } + + log_debug(LD_CIRC,"connecting in progress (or finished). Good."); + /* return success. The onion/circuit/etc will be taken care of + * automatically (may already have been) whenever n_chan reaches + * OR_CONN_STATE_OPEN. + */ + return 0; + } else { /* it's already open. use it. */ + tor_assert(!circ->base_.n_hop); + circ->base_.n_chan = n_chan; + log_debug(LD_CIRC,"Conn open. Delivering first onion skin."); + if ((err_reason = circuit_send_next_onion_skin(circ)) < 0) { + log_info(LD_CIRC,"circuit_send_next_onion_skin failed."); + circ->base_.n_chan = NULL; + return err_reason; + } + } + return 0; +} + +/** Find any circuits that are waiting on <b>or_conn</b> to become + * open and get them to send their create cells forward. + * + * Status is 1 if connect succeeded, or 0 if connect failed. + * + * Close_origin_circuits is 1 if we should close all the origin circuits + * through this channel, or 0 otherwise. (This happens when we want to retry + * an older guard.) + */ +void +circuit_n_chan_done(channel_t *chan, int status, int close_origin_circuits) +{ + smartlist_t *pending_circs; + int err_reason = 0; + + tor_assert(chan); + + log_debug(LD_CIRC,"chan to %s, status=%d", + channel_get_canonical_remote_descr(chan), status); + + pending_circs = smartlist_new(); + circuit_get_all_pending_on_channel(pending_circs, chan); + + SMARTLIST_FOREACH_BEGIN(pending_circs, circuit_t *, circ) + { + /* These checks are redundant wrt get_all_pending_on_or_conn, but I'm + * leaving them in in case it's possible for the status of a circuit to + * change as we're going down the list. */ + if (circ->marked_for_close || circ->n_chan || !circ->n_hop || + circ->state != CIRCUIT_STATE_CHAN_WAIT) + continue; + + if (tor_digest_is_zero(circ->n_hop->identity_digest)) { + /* Look at addr/port. This is an unkeyed connection. */ + if (!channel_matches_extend_info(chan, circ->n_hop)) + continue; + } else { + /* We expected a key. See if it's the right one. */ + if (tor_memneq(chan->identity_digest, + circ->n_hop->identity_digest, DIGEST_LEN)) + continue; + } + if (!status) { /* chan failed; close circ */ + log_info(LD_CIRC,"Channel failed; closing circ."); + circuit_mark_for_close(circ, END_CIRC_REASON_CHANNEL_CLOSED); + continue; + } + if (close_origin_circuits && CIRCUIT_IS_ORIGIN(circ)) { + log_info(LD_CIRC,"Channel deprecated for origin circs; closing circ."); + circuit_mark_for_close(circ, END_CIRC_REASON_CHANNEL_CLOSED); + continue; + } + log_debug(LD_CIRC, "Found circ, sending create cell."); + /* circuit_deliver_create_cell will set n_circ_id and add us to + * chan_circuid_circuit_map, so we don't need to call + * set_circid_chan here. */ + circ->n_chan = chan; + extend_info_free(circ->n_hop); + circ->n_hop = NULL; + + if (CIRCUIT_IS_ORIGIN(circ)) { + if ((err_reason = + circuit_send_next_onion_skin(TO_ORIGIN_CIRCUIT(circ))) < 0) { + log_info(LD_CIRC, + "send_next_onion_skin failed; circuit marked for closing."); + circuit_mark_for_close(circ, -err_reason); + continue; + /* XXX could this be bad, eg if next_onion_skin failed because conn + * died? */ + } + } else { + /* pull the create cell out of circ->n_chan_create_cell, and send it */ + tor_assert(circ->n_chan_create_cell); + if (circuit_deliver_create_cell(circ, circ->n_chan_create_cell, 1)<0) { + circuit_mark_for_close(circ, END_CIRC_REASON_RESOURCELIMIT); + continue; + } + tor_free(circ->n_chan_create_cell); + circuit_set_state(circ, CIRCUIT_STATE_OPEN); + } + } + SMARTLIST_FOREACH_END(circ); + + smartlist_free(pending_circs); +} + +/** Find a new circid that isn't currently in use on the circ->n_chan + * for the outgoing + * circuit <b>circ</b>, and deliver the cell <b>create_cell</b> to this + * circuit. If <b>relayed</b> is true, this is a create cell somebody + * gave us via an EXTEND cell, so we shouldn't worry if we don't understand + * it. Return -1 if we failed to find a suitable circid, else return 0. + */ +static int +circuit_deliver_create_cell(circuit_t *circ, const create_cell_t *create_cell, + int relayed) +{ + cell_t cell; + circid_t id; + int r; + + tor_assert(circ); + tor_assert(circ->n_chan); + tor_assert(create_cell); + tor_assert(create_cell->cell_type == CELL_CREATE || + create_cell->cell_type == CELL_CREATE_FAST || + create_cell->cell_type == CELL_CREATE2); + + id = get_unique_circ_id_by_chan(circ->n_chan); + if (!id) { + static ratelim_t circid_warning_limit = RATELIM_INIT(9600); + log_fn_ratelim(&circid_warning_limit, LOG_WARN, LD_CIRC, + "failed to get unique circID."); + goto error; + } + + memset(&cell, 0, sizeof(cell_t)); + r = relayed ? create_cell_format_relayed(&cell, create_cell) + : create_cell_format(&cell, create_cell); + if (r < 0) { + log_warn(LD_CIRC,"Couldn't format create cell"); + goto error; + } + log_debug(LD_CIRC,"Chosen circID %u.", (unsigned)id); + circuit_set_n_circid_chan(circ, id, circ->n_chan); + cell.circ_id = circ->n_circ_id; + + append_cell_to_circuit_queue(circ, circ->n_chan, &cell, + CELL_DIRECTION_OUT, 0); + + if (CIRCUIT_IS_ORIGIN(circ)) { + /* Update began timestamp for circuits starting their first hop */ + if (TO_ORIGIN_CIRCUIT(circ)->cpath->state == CPATH_STATE_CLOSED) { + if (!CHANNEL_IS_OPEN(circ->n_chan)) { + log_warn(LD_CIRC, + "Got first hop for a circuit without an opened channel. " + "State: %s.", channel_state_to_string(circ->n_chan->state)); + tor_fragile_assert(); + } + + tor_gettimeofday(&circ->timestamp_began); + } + + /* mark it so it gets better rate limiting treatment. */ + channel_timestamp_client(circ->n_chan); + } + + return 0; + error: + circ->n_chan = NULL; + return -1; +} + +/** We've decided to start our reachability testing. If all + * is set, log this to the user. Return 1 if we did, or 0 if + * we chose not to log anything. */ +int +inform_testing_reachability(void) +{ + char dirbuf[128]; + char *address; + const routerinfo_t *me = router_get_my_routerinfo(); + if (!me) + return 0; + address = tor_dup_ip(me->addr); + control_event_server_status(LOG_NOTICE, + "CHECKING_REACHABILITY ORADDRESS=%s:%d", + address, me->or_port); + if (me->dir_port) { + tor_snprintf(dirbuf, sizeof(dirbuf), " and DirPort %s:%d", + address, me->dir_port); + control_event_server_status(LOG_NOTICE, + "CHECKING_REACHABILITY DIRADDRESS=%s:%d", + address, me->dir_port); + } + log_notice(LD_OR, "Now checking whether ORPort %s:%d%s %s reachable... " + "(this may take up to %d minutes -- look for log " + "messages indicating success)", + address, me->or_port, + me->dir_port ? dirbuf : "", + me->dir_port ? "are" : "is", + TIMEOUT_UNTIL_UNREACHABILITY_COMPLAINT/60); + + tor_free(address); + return 1; +} + +/** Return true iff we should send a create_fast cell to start building a given + * circuit */ +static inline int +should_use_create_fast_for_circuit(origin_circuit_t *circ) +{ + const or_options_t *options = get_options(); + tor_assert(circ->cpath); + tor_assert(circ->cpath->extend_info); + + if (!circuit_has_usable_onion_key(circ)) { + /* We don't have ntor, and we don't have or can't use TAP, + * so our hand is forced: only a create_fast will work. */ + return 1; + } + if (public_server_mode(options)) { + /* We're a server, and we have a usable onion key. We can choose. + * Prefer to blend our circuit into the other circuits we are + * creating on behalf of others. */ + return 0; + } + return networkstatus_get_param(NULL, "usecreatefast", 0, 0, 1); +} + +/** + * Return true if <b>circ</b> is the type of circuit we want to count + * timeouts from. + * + * In particular, we want to consider any circuit that plans to build + * at least 3 hops (but maybe more), but has 3 or fewer hops built + * so far. + * + * We still want to consider circuits before 3 hops, because we need + * to decide if we should convert them to a measurement circuit in + * circuit_build_times_handle_completed_hop(), rather than letting + * slow circuits get killed right away. + */ +int +circuit_timeout_want_to_count_circ(const origin_circuit_t *circ) +{ + return !circ->has_opened + && circ->build_state->desired_path_len >= DEFAULT_ROUTE_LEN + && circuit_get_cpath_opened_len(circ) <= DEFAULT_ROUTE_LEN; +} + +/** Decide whether to use a TAP or ntor handshake for connecting to <b>ei</b> + * directly, and set *<b>cell_type_out</b> and *<b>handshake_type_out</b> + * accordingly. + * Note that TAP handshakes in CREATE cells are only used for direct + * connections: + * - from Single Onions to rend points not in the service's consensus. + * This is checked in onion_populate_cpath. */ +static void +circuit_pick_create_handshake(uint8_t *cell_type_out, + uint16_t *handshake_type_out, + const extend_info_t *ei) +{ + /* torspec says: In general, clients SHOULD use CREATE whenever they are + * using the TAP handshake, and CREATE2 otherwise. */ + if (extend_info_supports_ntor(ei)) { + *cell_type_out = CELL_CREATE2; + *handshake_type_out = ONION_HANDSHAKE_TYPE_NTOR; + } else { + /* XXXX030 Remove support for deciding to use TAP and EXTEND. */ + *cell_type_out = CELL_CREATE; + *handshake_type_out = ONION_HANDSHAKE_TYPE_TAP; + } +} + +/** Decide whether to use a TAP or ntor handshake for extending to <b>ei</b> + * and set *<b>handshake_type_out</b> accordingly. Decide whether we should + * use an EXTEND2 or an EXTEND cell to do so, and set *<b>cell_type_out</b> + * and *<b>create_cell_type_out</b> accordingly. + * Note that TAP handshakes in EXTEND cells are only used: + * - from clients to intro points, and + * - from hidden services to rend points. + * This is checked in onion_populate_cpath. + */ +static void +circuit_pick_extend_handshake(uint8_t *cell_type_out, + uint8_t *create_cell_type_out, + uint16_t *handshake_type_out, + const extend_info_t *ei) +{ + uint8_t t; + circuit_pick_create_handshake(&t, handshake_type_out, ei); + + /* torspec says: Clients SHOULD use the EXTEND format whenever sending a TAP + * handshake... In other cases, clients SHOULD use EXTEND2. */ + if (*handshake_type_out != ONION_HANDSHAKE_TYPE_TAP) { + *cell_type_out = RELAY_COMMAND_EXTEND2; + *create_cell_type_out = CELL_CREATE2; + } else { + /* XXXX030 Remove support for deciding to use TAP and EXTEND. */ + *cell_type_out = RELAY_COMMAND_EXTEND; + *create_cell_type_out = CELL_CREATE; + } +} + +/** + * Return true iff <b>purpose</b> is a purpose for a circuit which is + * allowed to have no guard configured, even if the circuit is multihop + * and guards are enabled. + */ +static int +circuit_purpose_may_omit_guard(int purpose) +{ + switch (purpose) { + case CIRCUIT_PURPOSE_TESTING: + case CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT: + /* Testing circuits may omit guards because they're measuring + * liveness or performance, and don't want guards to interfere. */ + return 1; + default: + /* All other multihop circuits should use guards if guards are + * enabled. */ + return 0; + } +} + +/** This is the backbone function for building circuits. + * + * If circ's first hop is closed, then we need to build a create + * cell and send it forward. + * + * Otherwise, if circ's cpath still has any non-open hops, we need to + * build a relay extend cell and send it forward to the next non-open hop. + * + * If all hops on the cpath are open, we're done building the circuit + * and we should do housekeeping for the newly opened circuit. + * + * Return -reason if we want to tear down circ, else return 0. + */ +int +circuit_send_next_onion_skin(origin_circuit_t *circ) +{ + tor_assert(circ); + + if (circ->cpath->state == CPATH_STATE_CLOSED) { + /* Case one: we're on the first hop. */ + return circuit_send_first_onion_skin(circ); + } + + tor_assert(circ->cpath->state == CPATH_STATE_OPEN); + tor_assert(circ->base_.state == CIRCUIT_STATE_BUILDING); + + crypt_path_t *hop = onion_next_hop_in_cpath(circ->cpath); + circuit_build_times_handle_completed_hop(circ); + + if (hop) { + /* Case two: we're on a hop after the first. */ + return circuit_send_intermediate_onion_skin(circ, hop); + } + + /* Case three: the circuit is finished. Do housekeeping tasks on it. */ + return circuit_build_no_more_hops(circ); +} + +/** + * Called from circuit_send_next_onion_skin() when we find ourselves connected + * to the first hop in <b>circ</b>: Send a CREATE or CREATE2 or CREATE_FAST + * cell to that hop. Return 0 on success; -reason on failure (if the circuit + * should be torn down). + */ +static int +circuit_send_first_onion_skin(origin_circuit_t *circ) +{ + int fast; + int len; + const node_t *node; + create_cell_t cc; + memset(&cc, 0, sizeof(cc)); + + log_debug(LD_CIRC,"First skin; sending create cell."); + + if (circ->build_state->onehop_tunnel) { + control_event_bootstrap(BOOTSTRAP_STATUS_ONEHOP_CREATE, 0); + } else { + control_event_bootstrap(BOOTSTRAP_STATUS_CIRCUIT_CREATE, 0); + + /* If this is not a one-hop tunnel, the channel is being used + * for traffic that wants anonymity and protection from traffic + * analysis (such as netflow record retention). That means we want + * to pad it. + */ + if (circ->base_.n_chan->channel_usage < CHANNEL_USED_FOR_FULL_CIRCS) + circ->base_.n_chan->channel_usage = CHANNEL_USED_FOR_FULL_CIRCS; + } + + node = node_get_by_id(circ->base_.n_chan->identity_digest); + fast = should_use_create_fast_for_circuit(circ); + if (!fast) { + /* We know the right onion key: we should send a create cell. */ + circuit_pick_create_handshake(&cc.cell_type, &cc.handshake_type, + circ->cpath->extend_info); + } else { + /* We don't know an onion key, so we need to fall back to CREATE_FAST. */ + cc.cell_type = CELL_CREATE_FAST; + cc.handshake_type = ONION_HANDSHAKE_TYPE_FAST; + } + + len = onion_skin_create(cc.handshake_type, + circ->cpath->extend_info, + &circ->cpath->handshake_state, + cc.onionskin); + if (len < 0) { + log_warn(LD_CIRC,"onion_skin_create (first hop) failed."); + return - END_CIRC_REASON_INTERNAL; + } + cc.handshake_len = len; + + if (circuit_deliver_create_cell(TO_CIRCUIT(circ), &cc, 0) < 0) + return - END_CIRC_REASON_RESOURCELIMIT; + + circ->cpath->state = CPATH_STATE_AWAITING_KEYS; + circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_BUILDING); + log_info(LD_CIRC,"First hop: finished sending %s cell to '%s'", + fast ? "CREATE_FAST" : "CREATE", + node ? node_describe(node) : "<unnamed>"); + return 0; +} + +/** + * Called from circuit_send_next_onion_skin() when we find that we have no + * more hops: mark the circuit as finished, and perform the necessary + * bookkeeping. Return 0 on success; -reason on failure (if the circuit + * should be torn down). + */ +static int +circuit_build_no_more_hops(origin_circuit_t *circ) +{ + guard_usable_t r; + if (! circ->guard_state) { + if (circuit_get_cpath_len(circ) != 1 && + ! circuit_purpose_may_omit_guard(circ->base_.purpose) && + get_options()->UseEntryGuards) { + log_warn(LD_BUG, "%d-hop circuit %p with purpose %d has no " + "guard state", + circuit_get_cpath_len(circ), circ, circ->base_.purpose); + } + r = GUARD_USABLE_NOW; + } else { + r = entry_guard_succeeded(&circ->guard_state); + } + const int is_usable_for_streams = (r == GUARD_USABLE_NOW); + if (r == GUARD_USABLE_NOW) { + circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_OPEN); + } else if (r == GUARD_MAYBE_USABLE_LATER) { + // Wait till either a better guard succeeds, or till + // all better guards fail. + circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_GUARD_WAIT); + } else { + tor_assert_nonfatal(r == GUARD_USABLE_NEVER); + return - END_CIRC_REASON_INTERNAL; + } + + /* XXXX #21422 -- the rest of this branch needs careful thought! + * Some of the things here need to happen when a circuit becomes + * mechanically open; some need to happen when it is actually usable. + * I think I got them right, but more checking would be wise. -NM + */ + + log_info(LD_CIRC,"circuit built!"); + circuit_reset_failure_count(0); + + if (circ->build_state->onehop_tunnel || circ->has_opened) { + control_event_bootstrap(BOOTSTRAP_STATUS_REQUESTING_STATUS, 0); + } + + pathbias_count_build_success(circ); + if (is_usable_for_streams) + circuit_has_opened(circ); /* do other actions as necessary */ + + if (!have_completed_a_circuit() && !circ->build_state->onehop_tunnel) { + const or_options_t *options = get_options(); + note_that_we_completed_a_circuit(); + /* FFFF Log a count of known routers here */ + log_notice(LD_GENERAL, + "Tor has successfully opened a circuit. " + "Looks like client functionality is working."); + if (control_event_bootstrap(BOOTSTRAP_STATUS_DONE, 0) == 0) { + log_notice(LD_GENERAL, + "Tor has successfully opened a circuit. " + "Looks like client functionality is working."); + } + control_event_client_status(LOG_NOTICE, "CIRCUIT_ESTABLISHED"); + clear_broken_connection_map(1); + if (server_mode(options) && !check_whether_orport_reachable(options)) { + inform_testing_reachability(); + router_do_reachability_checks(1, 1); + } + } + + /* We're done with measurement circuits here. Just close them */ + if (circ->base_.purpose == CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT) { + circuit_mark_for_close(TO_CIRCUIT(circ), END_CIRC_REASON_FINISHED); + } + return 0; +} + +/** + * Called from circuit_send_next_onion_skin() when we find that we have a hop + * other than the first that we need to extend to: use <b>hop</b>'s + * information to extend the circuit another step. Return 0 on success; + * -reason on failure (if the circuit should be torn down). + */ +static int +circuit_send_intermediate_onion_skin(origin_circuit_t *circ, + crypt_path_t *hop) +{ + int len; + extend_cell_t ec; + memset(&ec, 0, sizeof(ec)); + + log_debug(LD_CIRC,"starting to send subsequent skin."); + + if (tor_addr_family(&hop->extend_info->addr) != AF_INET) { + log_warn(LD_BUG, "Trying to extend to a non-IPv4 address."); + return - END_CIRC_REASON_INTERNAL; + } + + circuit_pick_extend_handshake(&ec.cell_type, + &ec.create_cell.cell_type, + &ec.create_cell.handshake_type, + hop->extend_info); + + tor_addr_copy(&ec.orport_ipv4.addr, &hop->extend_info->addr); + ec.orport_ipv4.port = hop->extend_info->port; + tor_addr_make_unspec(&ec.orport_ipv6.addr); + memcpy(ec.node_id, hop->extend_info->identity_digest, DIGEST_LEN); + /* Set the ED25519 identity too -- it will only get included + * in the extend2 cell if we're configured to use it, though. */ + ed25519_pubkey_copy(&ec.ed_pubkey, &hop->extend_info->ed_identity); + + len = onion_skin_create(ec.create_cell.handshake_type, + hop->extend_info, + &hop->handshake_state, + ec.create_cell.onionskin); + if (len < 0) { + log_warn(LD_CIRC,"onion_skin_create failed."); + return - END_CIRC_REASON_INTERNAL; + } + ec.create_cell.handshake_len = len; + + log_info(LD_CIRC,"Sending extend relay cell."); + { + uint8_t command = 0; + uint16_t payload_len=0; + uint8_t payload[RELAY_PAYLOAD_SIZE]; + if (extend_cell_format(&command, &payload_len, payload, &ec)<0) { + log_warn(LD_CIRC,"Couldn't format extend cell"); + return -END_CIRC_REASON_INTERNAL; + } + + /* send it to hop->prev, because that relay will transfer + * it to a create cell and then send to hop */ + if (relay_send_command_from_edge(0, TO_CIRCUIT(circ), + command, + (char*)payload, payload_len, + hop->prev) < 0) + return 0; /* circuit is closed */ + } + hop->state = CPATH_STATE_AWAITING_KEYS; + return 0; +} + +/** Our clock just jumped by <b>seconds_elapsed</b>. If <b>was_idle</b> is + * true, then the monotonic time matches; otherwise it doesn't. Assume + * something has also gone wrong with our network: notify the user, and + * abandon all not-yet-used circuits. */ +void +circuit_note_clock_jumped(int64_t seconds_elapsed, bool was_idle) +{ + int severity = server_mode(get_options()) ? LOG_WARN : LOG_NOTICE; + if (was_idle) { + tor_log(severity, LD_GENERAL, "Tor has been idle for %"PRId64 + " seconds; assuming established circuits no longer work.", + (seconds_elapsed)); + } else { + tor_log(severity, LD_GENERAL, + "Your system clock just jumped %"PRId64" seconds %s; " + "assuming established circuits no longer work.", + ( + seconds_elapsed >=0 ? seconds_elapsed : -seconds_elapsed), + seconds_elapsed >=0 ? "forward" : "backward"); + } + control_event_general_status(LOG_WARN, "CLOCK_JUMPED TIME=%"PRId64 + " IDLE=%d", + (seconds_elapsed), was_idle?1:0); + /* so we log when it works again */ + note_that_we_maybe_cant_complete_circuits(); + control_event_client_status(severity, "CIRCUIT_NOT_ESTABLISHED REASON=%s", + "CLOCK_JUMPED"); + circuit_mark_all_unused_circs(); + circuit_mark_all_dirty_circs_as_unusable(); + if (seconds_elapsed < 0) { + /* Restart all the timers in case we jumped a long way into the past. */ + reset_all_main_loop_timers(); + } +} + +/** Take the 'extend' <b>cell</b>, pull out addr/port plus the onion + * skin and identity digest for the next hop. If we're already connected, + * pass the onion skin to the next hop using a create cell; otherwise + * launch a new OR connection, and <b>circ</b> will notice when the + * connection succeeds or fails. + * + * Return -1 if we want to warn and tear down the circuit, else return 0. + */ +int +circuit_extend(cell_t *cell, circuit_t *circ) +{ + channel_t *n_chan; + relay_header_t rh; + extend_cell_t ec; + const char *msg = NULL; + int should_launch = 0; + + if (circ->n_chan) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "n_chan already set. Bug/attack. Closing."); + return -1; + } + if (circ->n_hop) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "conn to next hop already launched. Bug/attack. Closing."); + return -1; + } + + if (!server_mode(get_options())) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Got an extend cell, but running as a client. Closing."); + return -1; + } + + relay_header_unpack(&rh, cell->payload); + + if (extend_cell_parse(&ec, rh.command, + cell->payload+RELAY_HEADER_SIZE, + rh.length) < 0) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Can't parse extend cell. Closing circuit."); + return -1; + } + + if (!ec.orport_ipv4.port || tor_addr_is_null(&ec.orport_ipv4.addr)) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Client asked me to extend to zero destination port or addr."); + return -1; + } + + if (tor_addr_is_internal(&ec.orport_ipv4.addr, 0) && + !get_options()->ExtendAllowPrivateAddresses) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Client asked me to extend to a private address"); + return -1; + } + + /* Check if they asked us for 0000..0000. We support using + * an empty fingerprint for the first hop (e.g. for a bridge relay), + * but we don't want to let clients send us extend cells for empty + * fingerprints -- a) because it opens the user up to a mitm attack, + * and b) because it lets an attacker force the relay to hold open a + * new TLS connection for each extend request. */ + if (tor_digest_is_zero((const char*)ec.node_id)) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Client asked me to extend without specifying an id_digest."); + return -1; + } + + /* Fill in ed_pubkey if it was not provided and we can infer it from + * our networkstatus */ + if (ed25519_public_key_is_zero(&ec.ed_pubkey)) { + const node_t *node = node_get_by_id((const char*)ec.node_id); + const ed25519_public_key_t *node_ed_id = NULL; + if (node && + node_supports_ed25519_link_authentication(node, 1) && + (node_ed_id = node_get_ed25519_id(node))) { + ed25519_pubkey_copy(&ec.ed_pubkey, node_ed_id); + } + } + + /* Next, check if we're being asked to connect to the hop that the + * extend cell came from. There isn't any reason for that, and it can + * assist circular-path attacks. */ + if (tor_memeq(ec.node_id, + TO_OR_CIRCUIT(circ)->p_chan->identity_digest, + DIGEST_LEN)) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Client asked me to extend back to the previous hop."); + return -1; + } + + /* Check the previous hop Ed25519 ID too */ + if (! ed25519_public_key_is_zero(&ec.ed_pubkey) && + ed25519_pubkey_eq(&ec.ed_pubkey, + &TO_OR_CIRCUIT(circ)->p_chan->ed25519_identity)) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Client asked me to extend back to the previous hop " + "(by Ed25519 ID)."); + return -1; + } + + n_chan = channel_get_for_extend((const char*)ec.node_id, + &ec.ed_pubkey, + &ec.orport_ipv4.addr, + &msg, + &should_launch); + + if (!n_chan) { + log_debug(LD_CIRC|LD_OR,"Next router (%s): %s", + fmt_addrport(&ec.orport_ipv4.addr,ec.orport_ipv4.port), + msg?msg:"????"); + + circ->n_hop = extend_info_new(NULL /*nickname*/, + (const char*)ec.node_id, + &ec.ed_pubkey, + NULL, /*onion_key*/ + NULL, /*curve25519_key*/ + &ec.orport_ipv4.addr, + ec.orport_ipv4.port); + + circ->n_chan_create_cell = tor_memdup(&ec.create_cell, + sizeof(ec.create_cell)); + + circuit_set_state(circ, CIRCUIT_STATE_CHAN_WAIT); + + if (should_launch) { + /* we should try to open a connection */ + n_chan = channel_connect_for_circuit(&ec.orport_ipv4.addr, + ec.orport_ipv4.port, + (const char*)ec.node_id, + &ec.ed_pubkey); + if (!n_chan) { + log_info(LD_CIRC,"Launching n_chan failed. Closing circuit."); + circuit_mark_for_close(circ, END_CIRC_REASON_CONNECTFAILED); + return 0; + } + log_debug(LD_CIRC,"connecting in progress (or finished). Good."); + } + /* return success. The onion/circuit/etc will be taken care of + * automatically (may already have been) whenever n_chan reaches + * OR_CONN_STATE_OPEN. + */ + return 0; + } + + tor_assert(!circ->n_hop); /* Connection is already established. */ + circ->n_chan = n_chan; + log_debug(LD_CIRC, + "n_chan is %s", + channel_get_canonical_remote_descr(n_chan)); + + if (circuit_deliver_create_cell(circ, &ec.create_cell, 1) < 0) + return -1; + + return 0; +} + +/** Initialize cpath->{f|b}_{crypto|digest} from the key material in key_data. + * + * If <b>is_hs_v3</b> is set, this cpath will be used for next gen hidden + * service circuits and <b>key_data</b> must be at least + * HS_NTOR_KEY_EXPANSION_KDF_OUT_LEN bytes in length. + * + * If <b>is_hs_v3</b> is not set, key_data must contain CPATH_KEY_MATERIAL_LEN + * bytes, which are used as follows: + * - 20 to initialize f_digest + * - 20 to initialize b_digest + * - 16 to key f_crypto + * - 16 to key b_crypto + * + * (If 'reverse' is true, then f_XX and b_XX are swapped.) + * + * Return 0 if init was successful, else -1 if it failed. + */ +int +circuit_init_cpath_crypto(crypt_path_t *cpath, + const char *key_data, size_t key_data_len, + int reverse, int is_hs_v3) +{ + + tor_assert(cpath); + return relay_crypto_init(&cpath->crypto, key_data, key_data_len, reverse, + is_hs_v3); +} + +/** A "created" cell <b>reply</b> came back to us on circuit <b>circ</b>. + * (The body of <b>reply</b> varies depending on what sort of handshake + * this is.) + * + * Calculate the appropriate keys and digests, make sure KH is + * correct, and initialize this hop of the cpath. + * + * Return - reason if we want to mark circ for close, else return 0. + */ +int +circuit_finish_handshake(origin_circuit_t *circ, + const created_cell_t *reply) +{ + char keys[CPATH_KEY_MATERIAL_LEN]; + crypt_path_t *hop; + int rv; + + if ((rv = pathbias_count_build_attempt(circ)) < 0) { + log_warn(LD_CIRC, "pathbias_count_build_attempt failed: %d", rv); + return rv; + } + + if (circ->cpath->state == CPATH_STATE_AWAITING_KEYS) { + hop = circ->cpath; + } else { + hop = onion_next_hop_in_cpath(circ->cpath); + if (!hop) { /* got an extended when we're all done? */ + log_warn(LD_PROTOCOL,"got extended when circ already built? Closing."); + return - END_CIRC_REASON_TORPROTOCOL; + } + } + tor_assert(hop->state == CPATH_STATE_AWAITING_KEYS); + + { + const char *msg = NULL; + if (onion_skin_client_handshake(hop->handshake_state.tag, + &hop->handshake_state, + reply->reply, reply->handshake_len, + (uint8_t*)keys, sizeof(keys), + (uint8_t*)hop->rend_circ_nonce, + &msg) < 0) { + if (msg) + log_warn(LD_CIRC,"onion_skin_client_handshake failed: %s", msg); + return -END_CIRC_REASON_TORPROTOCOL; + } + } + + onion_handshake_state_release(&hop->handshake_state); + + if (circuit_init_cpath_crypto(hop, keys, sizeof(keys), 0, 0)<0) { + return -END_CIRC_REASON_TORPROTOCOL; + } + + hop->state = CPATH_STATE_OPEN; + log_info(LD_CIRC,"Finished building circuit hop:"); + circuit_log_path(LOG_INFO,LD_CIRC,circ); + control_event_circuit_status(circ, CIRC_EVENT_EXTENDED, 0); + + return 0; +} + +/** We received a relay truncated cell on circ. + * + * Since we don't send truncates currently, getting a truncated + * means that a connection broke or an extend failed. For now, + * just give up: force circ to close, and return 0. + */ +int - circuit_truncated(origin_circuit_t *circ, crypt_path_t *layer, int reason) ++circuit_truncated(origin_circuit_t *circ, int reason) +{ +// crypt_path_t *victim; +// connection_t *stream; + + tor_assert(circ); - tor_assert(layer); + + /* XXX Since we don't send truncates currently, getting a truncated + * means that a connection broke or an extend failed. For now, + * just give up. + */ + circuit_mark_for_close(TO_CIRCUIT(circ), + END_CIRC_REASON_FLAG_REMOTE|reason); + return 0; + +#if 0 + while (layer->next != circ->cpath) { + /* we need to clear out layer->next */ + victim = layer->next; + log_debug(LD_CIRC, "Killing a layer of the cpath."); + + for (stream = circ->p_streams; stream; stream=stream->next_stream) { + if (stream->cpath_layer == victim) { + log_info(LD_APP, "Marking stream %d for close because of truncate.", + stream->stream_id); + /* no need to send 'end' relay cells, + * because the other side's already dead + */ + connection_mark_unattached_ap(stream, END_STREAM_REASON_DESTROY); + } + } + + layer->next = victim->next; + circuit_free_cpath_node(victim); + } + + log_info(LD_CIRC, "finished"); + return 0; +#endif /* 0 */ +} + +/** Given a response payload and keys, initialize, then send a created + * cell back. + */ +int +onionskin_answer(or_circuit_t *circ, + const created_cell_t *created_cell, + const char *keys, size_t keys_len, + const uint8_t *rend_circ_nonce) +{ + cell_t cell; + + tor_assert(keys_len == CPATH_KEY_MATERIAL_LEN); + + if (created_cell_format(&cell, created_cell) < 0) { + log_warn(LD_BUG,"couldn't format created cell (type=%d, len=%d)", + (int)created_cell->cell_type, (int)created_cell->handshake_len); + return -1; + } + cell.circ_id = circ->p_circ_id; + + circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_OPEN); + + log_debug(LD_CIRC,"init digest forward 0x%.8x, backward 0x%.8x.", + (unsigned int)get_uint32(keys), + (unsigned int)get_uint32(keys+20)); + if (relay_crypto_init(&circ->crypto, keys, keys_len, 0, 0)<0) { + log_warn(LD_BUG,"Circuit initialization failed"); + return -1; + } + + memcpy(circ->rend_circ_nonce, rend_circ_nonce, DIGEST_LEN); + + int used_create_fast = (created_cell->cell_type == CELL_CREATED_FAST); + + append_cell_to_circuit_queue(TO_CIRCUIT(circ), + circ->p_chan, &cell, CELL_DIRECTION_IN, 0); + log_debug(LD_CIRC,"Finished sending '%s' cell.", + used_create_fast ? "created_fast" : "created"); + + /* Ignore the local bit when ExtendAllowPrivateAddresses is set: + * it violates the assumption that private addresses are local. + * Also, many test networks run on local addresses, and + * TestingTorNetwork sets ExtendAllowPrivateAddresses. */ + if ((!channel_is_local(circ->p_chan) + || get_options()->ExtendAllowPrivateAddresses) + && !channel_is_outgoing(circ->p_chan)) { + /* record that we could process create cells from a non-local conn + * that we didn't initiate; presumably this means that create cells + * can reach us too. */ + router_orport_found_reachable(); + } + + return 0; +} + +/** Helper for new_route_len(). Choose a circuit length for purpose + * <b>purpose</b>: DEFAULT_ROUTE_LEN (+ 1 if someone else chose the + * exit). If someone else chose the exit, they could be colluding + * with the exit, so add a randomly selected node to preserve + * anonymity. + * + * Here, "exit node" sometimes means an OR acting as an internal + * endpoint, rather than as a relay to an external endpoint. This + * means there need to be at least DEFAULT_ROUTE_LEN routers between + * us and the internal endpoint to preserve the same anonymity + * properties that we would get when connecting to an external + * endpoint. These internal endpoints can include: + * + * - Connections to a directory of hidden services + * (CIRCUIT_PURPOSE_C_GENERAL) + * + * - A client connecting to an introduction point, which the hidden + * service picked (CIRCUIT_PURPOSE_C_INTRODUCING, via + * circuit_get_open_circ_or_launch() which rewrites it from + * CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT) + * + * - A hidden service connecting to a rendezvous point, which the + * client picked (CIRCUIT_PURPOSE_S_CONNECT_REND, via + * rend_service_receive_introduction() and + * rend_service_relaunch_rendezvous) + * + * There are currently two situations where we picked the exit node + * ourselves, making DEFAULT_ROUTE_LEN a safe circuit length: + * + * - We are a hidden service connecting to an introduction point + * (CIRCUIT_PURPOSE_S_ESTABLISH_INTRO, via + * rend_service_launch_establish_intro()) + * + * - We are a router testing its own reachabiity + * (CIRCUIT_PURPOSE_TESTING, via router_do_reachability_checks()) + * + * onion_pick_cpath_exit() bypasses us (by not calling + * new_route_len()) in the one-hop tunnel case, so we don't need to + * handle that. + */ +int +route_len_for_purpose(uint8_t purpose, extend_info_t *exit_ei) +{ + int routelen = DEFAULT_ROUTE_LEN; + int known_purpose = 0; + + if (circuit_should_use_vanguards(purpose)) { + /* Clients want an extra hop for rends to avoid linkability. + * Services want it for intro points to avoid publishing their + * layer3 guards. They want it for hsdir posts to use + * their full layer3 guard set for those connections. + * Ex: C - G - L2 - L3 - R + * S - G - L2 - L3 - HSDIR + * S - G - L2 - L3 - I + */ + if (purpose == CIRCUIT_PURPOSE_C_ESTABLISH_REND || + purpose == CIRCUIT_PURPOSE_S_HSDIR_POST || + purpose == CIRCUIT_PURPOSE_HS_VANGUARDS || + purpose == CIRCUIT_PURPOSE_S_ESTABLISH_INTRO) + return routelen+1; + + /* If we only have Layer2 vanguards, then we do not need + * the extra hop for linkabilty reasons (see below). + * This means all hops can be of the form: + * S/C - G - L2 - M - R/HSDir/I + */ + if (get_options()->HSLayer2Nodes && !get_options()->HSLayer3Nodes) + return routelen+1; + + /* For connections to hsdirs, clients want two extra hops + * when using layer3 guards, to avoid linkability. + * Same goes for intro points. Note that the route len + * includes the intro point or hsdir, hence the +2. + * Ex: C - G - L2 - L3 - M - I + * C - G - L2 - L3 - M - HSDIR + * S - G - L2 - L3 - M - R + */ + if (purpose == CIRCUIT_PURPOSE_S_CONNECT_REND || + purpose == CIRCUIT_PURPOSE_C_HSDIR_GET || + purpose == CIRCUIT_PURPOSE_C_INTRODUCING) + return routelen+2; + } + + if (!exit_ei) + return routelen; + + switch (purpose) { + /* These two purposes connect to a router that we chose, so + * DEFAULT_ROUTE_LEN is safe. */ + case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO: + /* hidden service connecting to introduction point */ + case CIRCUIT_PURPOSE_TESTING: + /* router reachability testing */ + known_purpose = 1; + break; + + /* These three purposes connect to a router that someone else + * might have chosen, so add an extra hop to protect anonymity. */ + case CIRCUIT_PURPOSE_C_GENERAL: + case CIRCUIT_PURPOSE_C_HSDIR_GET: + case CIRCUIT_PURPOSE_S_HSDIR_POST: + /* connecting to hidden service directory */ + case CIRCUIT_PURPOSE_C_INTRODUCING: + /* client connecting to introduction point */ + case CIRCUIT_PURPOSE_S_CONNECT_REND: + /* hidden service connecting to rendezvous point */ + known_purpose = 1; + routelen++; + break; + + default: + /* Got a purpose not listed above along with a chosen exit. + * Increase the circuit length by one anyway for safety. */ + routelen++; + break; + } + + if (BUG(exit_ei && !known_purpose)) { + log_warn(LD_BUG, "Unhandled purpose %d with a chosen exit; " + "assuming routelen %d.", purpose, routelen); + } + return routelen; +} + +/** Choose a length for a circuit of purpose <b>purpose</b> and check + * if enough routers are available. + * + * If the routerlist <b>nodes</b> doesn't have enough routers + * to handle the desired path length, return -1. + */ +STATIC int +new_route_len(uint8_t purpose, extend_info_t *exit_ei, smartlist_t *nodes) +{ + int num_acceptable_routers; + int routelen; + + tor_assert(nodes); + + routelen = route_len_for_purpose(purpose, exit_ei); + + num_acceptable_routers = count_acceptable_nodes(nodes); + + log_debug(LD_CIRC,"Chosen route length %d (%d/%d routers suitable).", + routelen, num_acceptable_routers, smartlist_len(nodes)); + + if (num_acceptable_routers < routelen) { + log_info(LD_CIRC, + "Not enough acceptable routers (%d/%d). Discarding this circuit.", + num_acceptable_routers, routelen); + return -1; + } + + return routelen; +} + +/** Return a newly allocated list of uint16_t * for each predicted port not + * handled by a current circuit. */ +static smartlist_t * +circuit_get_unhandled_ports(time_t now) +{ + smartlist_t *dest = rep_hist_get_predicted_ports(now); + circuit_remove_handled_ports(dest); + return dest; +} + +/** Return 1 if we already have circuits present or on the way for + * all anticipated ports. Return 0 if we should make more. + * + * If we're returning 0, set need_uptime and need_capacity to + * indicate any requirements that the unhandled ports have. + */ +MOCK_IMPL(int, +circuit_all_predicted_ports_handled, (time_t now, int *need_uptime, + int *need_capacity)) +{ + int i, enough; + uint16_t *port; + smartlist_t *sl = circuit_get_unhandled_ports(now); + smartlist_t *LongLivedServices = get_options()->LongLivedPorts; + tor_assert(need_uptime); + tor_assert(need_capacity); + // Always predict need_capacity + *need_capacity = 1; + enough = (smartlist_len(sl) == 0); + for (i = 0; i < smartlist_len(sl); ++i) { + port = smartlist_get(sl, i); + if (smartlist_contains_int_as_string(LongLivedServices, *port)) + *need_uptime = 1; + tor_free(port); + } + smartlist_free(sl); + return enough; +} + +/** Return 1 if <b>node</b> can handle one or more of the ports in + * <b>needed_ports</b>, else return 0. + */ +static int +node_handles_some_port(const node_t *node, smartlist_t *needed_ports) +{ /* XXXX MOVE */ + int i; + uint16_t port; + + for (i = 0; i < smartlist_len(needed_ports); ++i) { + addr_policy_result_t r; + /* alignment issues aren't a worry for this dereference, since + needed_ports is explicitly a smartlist of uint16_t's */ + port = *(uint16_t *)smartlist_get(needed_ports, i); + tor_assert(port); + if (node) + r = compare_tor_addr_to_node_policy(NULL, port, node); + else + continue; + if (r != ADDR_POLICY_REJECTED && r != ADDR_POLICY_PROBABLY_REJECTED) + return 1; + } + return 0; +} + +/** Return true iff <b>conn</b> needs another general circuit to be + * built. */ +static int +ap_stream_wants_exit_attention(connection_t *conn) +{ + entry_connection_t *entry; + if (conn->type != CONN_TYPE_AP) + return 0; + entry = TO_ENTRY_CONN(conn); + + if (conn->state == AP_CONN_STATE_CIRCUIT_WAIT && + !conn->marked_for_close && + !(entry->want_onehop) && /* ignore one-hop streams */ + !(entry->use_begindir) && /* ignore targeted dir fetches */ + !(entry->chosen_exit_name) && /* ignore defined streams */ + !connection_edge_is_rendezvous_stream(TO_EDGE_CONN(conn)) && + !circuit_stream_is_being_handled(TO_ENTRY_CONN(conn), 0, + MIN_CIRCUITS_HANDLING_STREAM)) + return 1; + return 0; +} + +/** Return a pointer to a suitable router to be the exit node for the + * general-purpose circuit we're about to build. + * + * Look through the connection array, and choose a router that maximizes + * the number of pending streams that can exit from this router. + * + * Return NULL if we can't find any suitable routers. + */ +static const node_t * +choose_good_exit_server_general(router_crn_flags_t flags) +{ + int *n_supported; + int n_pending_connections = 0; + smartlist_t *connections; + int best_support = -1; + int n_best_support=0; + const or_options_t *options = get_options(); + const smartlist_t *the_nodes; + const node_t *selected_node=NULL; + const int need_uptime = (flags & CRN_NEED_UPTIME) != 0; + const int need_capacity = (flags & CRN_NEED_CAPACITY) != 0; + const int direct_conn = (flags & CRN_DIRECT_CONN) != 0; + + connections = get_connection_array(); + + /* Count how many connections are waiting for a circuit to be built. + * We use this for log messages now, but in the future we may depend on it. + */ + SMARTLIST_FOREACH(connections, connection_t *, conn, + { + if (ap_stream_wants_exit_attention(conn)) + ++n_pending_connections; + }); +// log_fn(LOG_DEBUG, "Choosing exit node; %d connections are pending", +// n_pending_connections); + /* Now we count, for each of the routers in the directory, how many + * of the pending connections could possibly exit from that + * router (n_supported[i]). (We can't be sure about cases where we + * don't know the IP address of the pending connection.) + * + * -1 means "Don't use this router at all." + */ + the_nodes = nodelist_get_list(); + n_supported = tor_calloc(smartlist_len(the_nodes), sizeof(int)); + SMARTLIST_FOREACH_BEGIN(the_nodes, const node_t *, node) { + const int i = node_sl_idx; + if (router_digest_is_me(node->identity)) { + n_supported[i] = -1; +// log_fn(LOG_DEBUG,"Skipping node %s -- it's me.", router->nickname); + /* XXX there's probably a reverse predecessor attack here, but + * it's slow. should we take this out? -RD + */ + continue; + } + if (!node_has_preferred_descriptor(node, direct_conn)) { + n_supported[i] = -1; + continue; + } + if (!node->is_running || node->is_bad_exit) { + n_supported[i] = -1; + continue; /* skip routers that are known to be down or bad exits */ + } + if (node_get_purpose(node) != ROUTER_PURPOSE_GENERAL) { + /* never pick a non-general node as a random exit. */ + n_supported[i] = -1; + continue; + } + if (routerset_contains_node(options->ExcludeExitNodesUnion_, node)) { + n_supported[i] = -1; + continue; /* user asked us not to use it, no matter what */ + } + if (options->ExitNodes && + !routerset_contains_node(options->ExitNodes, node)) { + n_supported[i] = -1; + continue; /* not one of our chosen exit nodes */ + } + + if (node_is_unreliable(node, need_uptime, need_capacity, 0)) { + n_supported[i] = -1; + continue; /* skip routers that are not suitable. Don't worry if + * this makes us reject all the possible routers: if so, + * we'll retry later in this function with need_update and + * need_capacity set to 0. */ + } + if (!(node->is_valid)) { + /* if it's invalid and we don't want it */ + n_supported[i] = -1; +// log_fn(LOG_DEBUG,"Skipping node %s (index %d) -- invalid router.", +// router->nickname, i); + continue; /* skip invalid routers */ + } + /* We do not allow relays that allow single hop exits by default. Option + * was deprecated in 0.2.9.2-alpha and removed in 0.3.1.0-alpha. */ + if (node_allows_single_hop_exits(node)) { + n_supported[i] = -1; + continue; + } + if (node_exit_policy_rejects_all(node)) { + n_supported[i] = -1; +// log_fn(LOG_DEBUG,"Skipping node %s (index %d) -- it rejects all.", +// router->nickname, i); + continue; /* skip routers that reject all */ + } + n_supported[i] = 0; + /* iterate over connections */ + SMARTLIST_FOREACH_BEGIN(connections, connection_t *, conn) { + if (!ap_stream_wants_exit_attention(conn)) + continue; /* Skip everything but APs in CIRCUIT_WAIT */ + if (connection_ap_can_use_exit(TO_ENTRY_CONN(conn), node)) { + ++n_supported[i]; +// log_fn(LOG_DEBUG,"%s is supported. n_supported[%d] now %d.", +// router->nickname, i, n_supported[i]); + } else { +// log_fn(LOG_DEBUG,"%s (index %d) would reject this stream.", +// router->nickname, i); + } + } SMARTLIST_FOREACH_END(conn); + if (n_pending_connections > 0 && n_supported[i] == 0) { + /* Leave best_support at -1 if that's where it is, so we can + * distinguish it later. */ + continue; + } + if (n_supported[i] > best_support) { + /* If this router is better than previous ones, remember its index + * and goodness, and start counting how many routers are this good. */ + best_support = n_supported[i]; n_best_support=1; +// log_fn(LOG_DEBUG,"%s is new best supported option so far.", +// router->nickname); + } else if (n_supported[i] == best_support) { + /* If this router is _as good_ as the best one, just increment the + * count of equally good routers.*/ + ++n_best_support; + } + } SMARTLIST_FOREACH_END(node); + log_info(LD_CIRC, + "Found %d servers that might support %d/%d pending connections.", + n_best_support, best_support >= 0 ? best_support : 0, + n_pending_connections); + + /* If any routers definitely support any pending connections, choose one + * at random. */ + if (best_support > 0) { + smartlist_t *supporting = smartlist_new(); + + SMARTLIST_FOREACH(the_nodes, const node_t *, node, { + if (n_supported[node_sl_idx] == best_support) + smartlist_add(supporting, (void*)node); + }); + + selected_node = node_sl_choose_by_bandwidth(supporting, WEIGHT_FOR_EXIT); + smartlist_free(supporting); + } else { + /* Either there are no pending connections, or no routers even seem to + * possibly support any of them. Choose a router at random that satisfies + * at least one predicted exit port. */ + + int attempt; + smartlist_t *needed_ports, *supporting; + + if (best_support == -1) { + if (need_uptime || need_capacity) { + log_info(LD_CIRC, + "We couldn't find any live%s%s routers; falling back " + "to list of all routers.", + need_capacity?", fast":"", + need_uptime?", stable":""); + tor_free(n_supported); + flags &= ~(CRN_NEED_UPTIME|CRN_NEED_CAPACITY); + return choose_good_exit_server_general(flags); + } + log_notice(LD_CIRC, "All routers are down or won't exit%s -- " + "choosing a doomed exit at random.", + options->ExcludeExitNodesUnion_ ? " or are Excluded" : ""); + } + supporting = smartlist_new(); + needed_ports = circuit_get_unhandled_ports(time(NULL)); + for (attempt = 0; attempt < 2; attempt++) { + /* try once to pick only from routers that satisfy a needed port, + * then if there are none, pick from any that support exiting. */ + SMARTLIST_FOREACH_BEGIN(the_nodes, const node_t *, node) { + if (n_supported[node_sl_idx] != -1 && + (attempt || node_handles_some_port(node, needed_ports))) { +// log_fn(LOG_DEBUG,"Try %d: '%s' is a possibility.", +// try, router->nickname); + smartlist_add(supporting, (void*)node); + } + } SMARTLIST_FOREACH_END(node); + + selected_node = node_sl_choose_by_bandwidth(supporting, WEIGHT_FOR_EXIT); + if (selected_node) + break; + smartlist_clear(supporting); + /* If we reach this point, we can't actually support any unhandled + * predicted ports, so clear all the remaining ones. */ + if (smartlist_len(needed_ports)) + rep_hist_remove_predicted_ports(needed_ports); + } + SMARTLIST_FOREACH(needed_ports, uint16_t *, cp, tor_free(cp)); + smartlist_free(needed_ports); + smartlist_free(supporting); + } + + tor_free(n_supported); + if (selected_node) { + log_info(LD_CIRC, "Chose exit server '%s'", node_describe(selected_node)); + return selected_node; + } + if (options->ExitNodes) { + log_warn(LD_CIRC, + "No exits in ExitNodes%s seem to be running: " + "can't choose an exit.", + options->ExcludeExitNodesUnion_ ? + ", except possibly those excluded by your configuration, " : ""); + } + return NULL; +} + +/* Pick a Rendezvous Point for our HS circuits according to <b>flags</b>. */ +static const node_t * +pick_rendezvous_node(router_crn_flags_t flags) +{ + const or_options_t *options = get_options(); + return router_choose_random_node(NULL, options->ExcludeNodes, flags); +} + +/* + * Helper function to pick a configured restricted middle node + * (either HSLayer2Nodes or HSLayer3Nodes). + * + * Make sure that the node we chose is alive, and not excluded, + * and return it. + * + * The exclude_set is a routerset of nodes that the selected node + * must not match, and the exclude_list is a simple list of nodes + * that the selected node must not be in. Either or both may be + * NULL. + * + * Return NULL if no usable nodes could be found. */ +static const node_t * +pick_restricted_middle_node(router_crn_flags_t flags, + const routerset_t *pick_from, + const routerset_t *exclude_set, + const smartlist_t *exclude_list, + int position_hint) +{ + const node_t *middle_node = NULL; + + smartlist_t *whitelisted_live_middles = smartlist_new(); + smartlist_t *all_live_nodes = smartlist_new(); + + tor_assert(pick_from); + + /* Add all running nodes to all_live_nodes */ + router_add_running_nodes_to_smartlist(all_live_nodes, + (flags & CRN_NEED_UPTIME) != 0, + (flags & CRN_NEED_CAPACITY) != 0, + (flags & CRN_NEED_GUARD) != 0, + (flags & CRN_NEED_DESC) != 0, + (flags & CRN_PREF_ADDR) != 0, + (flags & CRN_DIRECT_CONN) != 0); + + /* Filter all_live_nodes to only add live *and* whitelisted middles + * to the list whitelisted_live_middles. */ + SMARTLIST_FOREACH_BEGIN(all_live_nodes, node_t *, live_node) { + if (routerset_contains_node(pick_from, live_node)) { + smartlist_add(whitelisted_live_middles, live_node); + } + } SMARTLIST_FOREACH_END(live_node); + + /* Honor ExcludeNodes */ + if (exclude_set) { + routerset_subtract_nodes(whitelisted_live_middles, exclude_set); + } + + if (exclude_list) { + smartlist_subtract(whitelisted_live_middles, exclude_list); + } + + /** + * Max number of restricted nodes before we alert the user and try + * to load balance for them. + * + * The most aggressive vanguard design had 16 nodes at layer3. + * Let's give a small ceiling above that. */ +#define MAX_SANE_RESTRICTED_NODES 20 + /* If the user (or associated tor controller) selected only a few nodes, + * assume they took load balancing into account and don't do it for them. + * + * If there are a lot of nodes in here, assume they did not load balance + * and do it for them, but also warn them that they may be Doing It Wrong. + */ + if (smartlist_len(whitelisted_live_middles) <= + MAX_SANE_RESTRICTED_NODES) { + middle_node = smartlist_choose(whitelisted_live_middles); + } else { + static ratelim_t pinned_notice_limit = RATELIM_INIT(24*3600); + log_fn_ratelim(&pinned_notice_limit, LOG_NOTICE, LD_CIRC, + "Your _HSLayer%dNodes setting has resulted " + "in %d total nodes. This is a lot of nodes. " + "You may want to consider using a Tor controller " + "to select and update a smaller set of nodes instead.", + position_hint, smartlist_len(whitelisted_live_middles)); + + /* NO_WEIGHTING here just means don't take node flags into account + * (ie: use consensus measurement only). This is done so that + * we don't further surprise the user by not using Exits that they + * specified at all */ + middle_node = node_sl_choose_by_bandwidth(whitelisted_live_middles, + NO_WEIGHTING); + } + + smartlist_free(whitelisted_live_middles); + smartlist_free(all_live_nodes); + + return middle_node; +} + +/** Return a pointer to a suitable router to be the exit node for the + * circuit of purpose <b>purpose</b> that we're about to build (or NULL + * if no router is suitable). + * + * For general-purpose circuits, pass it off to + * choose_good_exit_server_general() + * + * For client-side rendezvous circuits, choose a random node, weighted + * toward the preferences in 'options'. + */ +static const node_t * +choose_good_exit_server(origin_circuit_t *circ, + router_crn_flags_t flags, int is_internal) +{ + const or_options_t *options = get_options(); + flags |= CRN_NEED_DESC; + + switch (TO_CIRCUIT(circ)->purpose) { + case CIRCUIT_PURPOSE_C_HSDIR_GET: + case CIRCUIT_PURPOSE_S_HSDIR_POST: + case CIRCUIT_PURPOSE_HS_VANGUARDS: + /* For these three, we want to pick the exit like a middle hop, + * since it should be random. */ + tor_assert_nonfatal(is_internal); + /* Falls through */ + case CIRCUIT_PURPOSE_C_GENERAL: + if (is_internal) /* pick it like a middle hop */ + return router_choose_random_node(NULL, options->ExcludeNodes, flags); + else + return choose_good_exit_server_general(flags); + case CIRCUIT_PURPOSE_C_ESTABLISH_REND: + { + /* Pick a new RP */ + const node_t *rendezvous_node = pick_rendezvous_node(flags); + log_info(LD_REND, "Picked new RP: %s", + safe_str_client(node_describe(rendezvous_node))); + return rendezvous_node; + } + } + log_warn(LD_BUG,"Unhandled purpose %d", TO_CIRCUIT(circ)->purpose); + tor_fragile_assert(); + return NULL; +} + +/** Log a warning if the user specified an exit for the circuit that + * has been excluded from use by ExcludeNodes or ExcludeExitNodes. */ +static void +warn_if_last_router_excluded(origin_circuit_t *circ, + const extend_info_t *exit_ei) +{ + const or_options_t *options = get_options(); + routerset_t *rs = options->ExcludeNodes; + const char *description; + uint8_t purpose = circ->base_.purpose; + + if (circ->build_state->onehop_tunnel) + return; + + switch (purpose) + { + default: + case CIRCUIT_PURPOSE_OR: + case CIRCUIT_PURPOSE_INTRO_POINT: + case CIRCUIT_PURPOSE_REND_POINT_WAITING: + case CIRCUIT_PURPOSE_REND_ESTABLISHED: + log_warn(LD_BUG, "Called on non-origin circuit (purpose %d, %s)", + (int)purpose, + circuit_purpose_to_string(purpose)); + return; + case CIRCUIT_PURPOSE_S_HSDIR_POST: + case CIRCUIT_PURPOSE_C_HSDIR_GET: + case CIRCUIT_PURPOSE_C_GENERAL: + if (circ->build_state->is_internal) + return; + description = "requested exit node"; + rs = options->ExcludeExitNodesUnion_; + break; + case CIRCUIT_PURPOSE_C_INTRODUCING: + case CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT: + case CIRCUIT_PURPOSE_C_INTRODUCE_ACKED: + case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO: + case CIRCUIT_PURPOSE_S_CONNECT_REND: + case CIRCUIT_PURPOSE_S_REND_JOINED: + case CIRCUIT_PURPOSE_TESTING: + return; + case CIRCUIT_PURPOSE_C_ESTABLISH_REND: + case CIRCUIT_PURPOSE_C_REND_READY: + case CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED: + case CIRCUIT_PURPOSE_C_REND_JOINED: + description = "chosen rendezvous point"; + break; + case CIRCUIT_PURPOSE_CONTROLLER: + rs = options->ExcludeExitNodesUnion_; + description = "controller-selected circuit target"; + break; + } + + if (routerset_contains_extendinfo(rs, exit_ei)) { + /* We should never get here if StrictNodes is set to 1. */ + if (options->StrictNodes) { + log_warn(LD_BUG, "Using %s '%s' which is listed in ExcludeNodes%s, " + "even though StrictNodes is set. Please report. " + "(Circuit purpose: %s)", + description, extend_info_describe(exit_ei), + rs==options->ExcludeNodes?"":" or ExcludeExitNodes", + circuit_purpose_to_string(purpose)); + } else { + log_warn(LD_CIRC, "Using %s '%s' which is listed in " + "ExcludeNodes%s, because no better options were available. To " + "prevent this (and possibly break your Tor functionality), " + "set the StrictNodes configuration option. " + "(Circuit purpose: %s)", + description, extend_info_describe(exit_ei), + rs==options->ExcludeNodes?"":" or ExcludeExitNodes", + circuit_purpose_to_string(purpose)); + } + circuit_log_path(LOG_WARN, LD_CIRC, circ); + } + + return; +} + +/** Decide a suitable length for circ's cpath, and pick an exit + * router (or use <b>exit</b> if provided). Store these in the + * cpath. + * + * If <b>is_hs_v3_rp_circuit</b> is set, then this exit should be suitable to + * be used as an HS v3 rendezvous point. + * + * Return 0 if ok, -1 if circuit should be closed. */ +STATIC int +onion_pick_cpath_exit(origin_circuit_t *circ, extend_info_t *exit_ei, + int is_hs_v3_rp_circuit) +{ + cpath_build_state_t *state = circ->build_state; + + if (state->onehop_tunnel) { + log_debug(LD_CIRC, "Launching a one-hop circuit for dir tunnel%s.", + (rend_allow_non_anonymous_connection(get_options()) ? + ", or intro or rendezvous connection" : "")); + state->desired_path_len = 1; + } else { + int r = new_route_len(circ->base_.purpose, exit_ei, nodelist_get_list()); + if (r < 1) /* must be at least 1 */ + return -1; + state->desired_path_len = r; + } + + if (exit_ei) { /* the circuit-builder pre-requested one */ + warn_if_last_router_excluded(circ, exit_ei); + log_info(LD_CIRC,"Using requested exit node '%s'", + extend_info_describe(exit_ei)); + exit_ei = extend_info_dup(exit_ei); + } else { /* we have to decide one */ + router_crn_flags_t flags = CRN_NEED_DESC; + if (state->need_uptime) + flags |= CRN_NEED_UPTIME; + if (state->need_capacity) + flags |= CRN_NEED_CAPACITY; + if (is_hs_v3_rp_circuit) + flags |= CRN_RENDEZVOUS_V3; + if (state->onehop_tunnel) + flags |= CRN_DIRECT_CONN; + const node_t *node = + choose_good_exit_server(circ, flags, state->is_internal); + if (!node) { + log_warn(LD_CIRC,"Failed to choose an exit server"); + return -1; + } + exit_ei = extend_info_from_node(node, state->onehop_tunnel); + if (BUG(exit_ei == NULL)) + return -1; + } + state->chosen_exit = exit_ei; + return 0; +} + +/** Give <b>circ</b> a new exit destination to <b>exit</b>, and add a + * hop to the cpath reflecting this. Don't send the next extend cell -- + * the caller will do this if it wants to. + */ +int +circuit_append_new_exit(origin_circuit_t *circ, extend_info_t *exit_ei) +{ + cpath_build_state_t *state; + tor_assert(exit_ei); + tor_assert(circ); + + state = circ->build_state; + tor_assert(state); + extend_info_free(state->chosen_exit); + state->chosen_exit = extend_info_dup(exit_ei); + + ++circ->build_state->desired_path_len; + onion_append_hop(&circ->cpath, exit_ei); + return 0; +} + +/** Take an open <b>circ</b>, and add a new hop at the end, based on + * <b>info</b>. Set its state back to CIRCUIT_STATE_BUILDING, and then + * send the next extend cell to begin connecting to that hop. + */ +int +circuit_extend_to_new_exit(origin_circuit_t *circ, extend_info_t *exit_ei) +{ + int err_reason = 0; + warn_if_last_router_excluded(circ, exit_ei); + + tor_gettimeofday(&circ->base_.timestamp_began); + + circuit_append_new_exit(circ, exit_ei); + circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_BUILDING); + if ((err_reason = circuit_send_next_onion_skin(circ))<0) { + log_warn(LD_CIRC, "Couldn't extend circuit to new point %s.", + extend_info_describe(exit_ei)); + circuit_mark_for_close(TO_CIRCUIT(circ), -err_reason); + return -1; + } + + // XXX: Should cannibalized circuits be dirty or not? Not easy to say.. + + return 0; +} + +/** Return the number of routers in <b>routers</b> that are currently up + * and available for building circuits through. + * + * (Note that this function may overcount or undercount, if we have + * descriptors that are not the type we would prefer to use for some + * particular router. See bug #25885.) + */ +MOCK_IMPL(STATIC int, +count_acceptable_nodes, (smartlist_t *nodes)) +{ + int num=0; + + SMARTLIST_FOREACH_BEGIN(nodes, const node_t *, node) { + // log_debug(LD_CIRC, +// "Contemplating whether router %d (%s) is a new option.", +// i, r->nickname); + if (! node->is_running) +// log_debug(LD_CIRC,"Nope, the directory says %d is not running.",i); + continue; + if (! node->is_valid) +// log_debug(LD_CIRC,"Nope, the directory says %d is not valid.",i); + continue; + if (! node_has_any_descriptor(node)) + continue; + /* The node has a descriptor, so we can just check the ntor key directly */ + if (!node_has_curve25519_onion_key(node)) + continue; + ++num; + } SMARTLIST_FOREACH_END(node); + +// log_debug(LD_CIRC,"I like %d. num_acceptable_routers now %d.",i, num); + + return num; +} + +/** Add <b>new_hop</b> to the end of the doubly-linked-list <b>head_ptr</b>. + * This function is used to extend cpath by another hop. + */ +void +onion_append_to_cpath(crypt_path_t **head_ptr, crypt_path_t *new_hop) +{ + if (*head_ptr) { + new_hop->next = (*head_ptr); + new_hop->prev = (*head_ptr)->prev; + (*head_ptr)->prev->next = new_hop; + (*head_ptr)->prev = new_hop; + } else { + *head_ptr = new_hop; + new_hop->prev = new_hop->next = new_hop; + } +} + +#ifdef TOR_UNIT_TESTS + +/** Unittest helper function: Count number of hops in cpath linked list. */ +unsigned int +cpath_get_n_hops(crypt_path_t **head_ptr) +{ + unsigned int n_hops = 0; + crypt_path_t *tmp; + + if (!*head_ptr) { + return 0; + } + + tmp = *head_ptr; + do { + n_hops++; + tmp = tmp->next; + } while (tmp != *head_ptr); + + return n_hops; +} + +#endif /* defined(TOR_UNIT_TESTS) */ + +/** + * Build the exclude list for vanguard circuits. + * + * For vanguard circuits we exclude all the already chosen nodes (including the + * exit) from being middle hops to prevent the creation of A - B - A subpaths. + * We also allow the 4th hop to be the same as the guard node so as to not leak + * guard information to RP/IP/HSDirs. + * + * For vanguard circuits, we don't apply any subnet or family restrictions. + * This is to avoid impossible-to-build circuit paths, or just situations where + * our earlier guards prevent us from using most of our later ones. + * + * The alternative is building the circuit in reverse. Reverse calls to + * onion_extend_cpath() (ie: select outer hops first) would then have the + * property that you don't gain information about inner hops by observing + * outer ones. See https://trac.torproject.org/projects/tor/ticket/24487 + * for this. + * + * (Note further that we still exclude the exit to prevent A - B - A + * at the end of the path. */ +static smartlist_t * +build_vanguard_middle_exclude_list(uint8_t purpose, + cpath_build_state_t *state, + crypt_path_t *head, + int cur_len) +{ + smartlist_t *excluded; + const node_t *r; + crypt_path_t *cpath; + int i; + + (void) purpose; + + excluded = smartlist_new(); + + /* Add the exit to the exclude list (note that the exit/last hop is always + * chosen first in circuit_establish_circuit()). */ + if ((r = build_state_get_exit_node(state))) { + smartlist_add(excluded, (node_t*)r); + } + + /* If we are picking the 4th hop, allow that node to be the guard too. + * This prevents us from avoiding the Guard for those hops, which + * gives the adversary information about our guard if they control + * the RP, IP, or HSDIR. We don't do this check based on purpose + * because we also want to allow HS_VANGUARDS pre-build circuits + * to use the guard for that last hop. + */ + if (cur_len == DEFAULT_ROUTE_LEN+1) { + /* Skip the first hop for the exclude list below */ + head = head->next; + cur_len--; + } + + for (i = 0, cpath = head; cpath && i < cur_len; ++i, cpath=cpath->next) { + if ((r = node_get_by_id(cpath->extend_info->identity_digest))) { + smartlist_add(excluded, (node_t*)r); + } + } + + return excluded; +} + +/** + * Build a list of nodes to exclude from the choice of this middle + * hop, based on already chosen nodes. + */ +static smartlist_t * +build_middle_exclude_list(uint8_t purpose, + cpath_build_state_t *state, + crypt_path_t *head, + int cur_len) +{ + smartlist_t *excluded; + const node_t *r; + crypt_path_t *cpath; + int i; + + /** Vanguard circuits have their own path selection rules */ + if (circuit_should_use_vanguards(purpose)) { + return build_vanguard_middle_exclude_list(purpose, state, head, cur_len); + } + + excluded = smartlist_new(); + + /* For non-vanguard circuits, add the exit and its family to the exclude list + * (note that the exit/last hop is always chosen first in + * circuit_establish_circuit()). */ + if ((r = build_state_get_exit_node(state))) { + nodelist_add_node_and_family(excluded, r); + } + + /* also exclude all other already chosen nodes and their family */ + for (i = 0, cpath = head; cpath && i < cur_len; ++i, cpath=cpath->next) { + if ((r = node_get_by_id(cpath->extend_info->identity_digest))) { + nodelist_add_node_and_family(excluded, r); + } + } + + return excluded; +} + +/** Return true if we MUST use vanguards for picking this middle node. */ +static int +middle_node_must_be_vanguard(const or_options_t *options, + uint8_t purpose, int cur_len) +{ + /* If this is not a hidden service circuit, don't use vanguards */ + if (!circuit_purpose_is_hidden_service(purpose)) { + return 0; + } + + /* If we have sticky L2 nodes, and this is an L2 pick, use vanguards */ + if (options->HSLayer2Nodes && cur_len == 1) { + return 1; + } + + /* If we have sticky L3 nodes, and this is an L3 pick, use vanguards */ + if (options->HSLayer3Nodes && cur_len == 2) { + return 1; + } + + return 0; +} + +/** Pick a sticky vanguard middle node or return NULL if not found. + * See doc of pick_restricted_middle_node() for argument details. */ +static const node_t * +pick_vanguard_middle_node(const or_options_t *options, + router_crn_flags_t flags, int cur_len, + const smartlist_t *excluded) +{ + const routerset_t *vanguard_routerset = NULL; + const node_t *node = NULL; + + /* Pick the right routerset based on the current hop */ + if (cur_len == 1) { + vanguard_routerset = options->HSLayer2Nodes; + } else if (cur_len == 2) { + vanguard_routerset = options->HSLayer3Nodes; + } else { + /* guaranteed by middle_node_should_be_vanguard() */ + tor_assert_nonfatal_unreached(); + return NULL; + } + + node = pick_restricted_middle_node(flags, vanguard_routerset, + options->ExcludeNodes, excluded, + cur_len+1); + + if (!node) { + static ratelim_t pinned_warning_limit = RATELIM_INIT(300); + log_fn_ratelim(&pinned_warning_limit, LOG_WARN, LD_CIRC, + "Could not find a node that matches the configured " + "_HSLayer%dNodes set", cur_len+1); + } + + return node; +} + +/** A helper function used by onion_extend_cpath(). Use <b>purpose</b> + * and <b>state</b> and the cpath <b>head</b> (currently populated only + * to length <b>cur_len</b> to decide a suitable middle hop for a + * circuit. In particular, make sure we don't pick the exit node or its + * family, and make sure we don't duplicate any previous nodes or their + * families. */ +static const node_t * +choose_good_middle_server(uint8_t purpose, + cpath_build_state_t *state, + crypt_path_t *head, + int cur_len) +{ + const node_t *choice; + smartlist_t *excluded; + const or_options_t *options = get_options(); + router_crn_flags_t flags = CRN_NEED_DESC; + tor_assert(CIRCUIT_PURPOSE_MIN_ <= purpose && + purpose <= CIRCUIT_PURPOSE_MAX_); + + log_debug(LD_CIRC, "Contemplating intermediate hop #%d: random choice.", + cur_len+1); + + excluded = build_middle_exclude_list(purpose, state, head, cur_len); + + if (state->need_uptime) + flags |= CRN_NEED_UPTIME; + if (state->need_capacity) + flags |= CRN_NEED_CAPACITY; + + /** If a hidden service circuit wants a specific middle node, pin it. */ + if (middle_node_must_be_vanguard(options, purpose, cur_len)) { + log_debug(LD_GENERAL, "Picking a sticky node (cur_len = %d)", cur_len); + choice = pick_vanguard_middle_node(options, flags, cur_len, excluded); + smartlist_free(excluded); + return choice; + } + + choice = router_choose_random_node(excluded, options->ExcludeNodes, flags); + smartlist_free(excluded); + return choice; +} + +/** Pick a good entry server for the circuit to be built according to + * <b>state</b>. Don't reuse a chosen exit (if any), don't use this + * router (if we're an OR), and respect firewall settings; if we're + * configured to use entry guards, return one. + * + * Set *<b>guard_state_out</b> to information about the guard that + * we're selecting, which we'll use later to remember whether the + * guard worked or not. + */ +const node_t * +choose_good_entry_server(uint8_t purpose, cpath_build_state_t *state, + circuit_guard_state_t **guard_state_out) +{ + const node_t *choice; + smartlist_t *excluded; + const or_options_t *options = get_options(); + /* If possible, choose an entry server with a preferred address, + * otherwise, choose one with an allowed address */ + router_crn_flags_t flags = (CRN_NEED_GUARD|CRN_NEED_DESC|CRN_PREF_ADDR| + CRN_DIRECT_CONN); + const node_t *node; + + /* Once we used this function to select a node to be a guard. We had + * 'state == NULL' be the signal for that. But we don't do that any more. + */ + tor_assert_nonfatal(state); + + if (state && options->UseEntryGuards && + (purpose != CIRCUIT_PURPOSE_TESTING || options->BridgeRelay)) { + /* This request is for an entry server to use for a regular circuit, + * and we use entry guard nodes. Just return one of the guard nodes. */ + tor_assert(guard_state_out); + return guards_choose_guard(state, purpose, guard_state_out); + } + + excluded = smartlist_new(); + + if (state && (node = build_state_get_exit_node(state))) { + /* Exclude the exit node from the state, if we have one. Also exclude its + * family. */ + nodelist_add_node_and_family(excluded, node); + } + + if (state) { + if (state->need_uptime) + flags |= CRN_NEED_UPTIME; + if (state->need_capacity) + flags |= CRN_NEED_CAPACITY; + } + + choice = router_choose_random_node(excluded, options->ExcludeNodes, flags); + smartlist_free(excluded); + return choice; +} + +/** Return the first non-open hop in cpath, or return NULL if all + * hops are open. */ +static crypt_path_t * +onion_next_hop_in_cpath(crypt_path_t *cpath) +{ + crypt_path_t *hop = cpath; + do { + if (hop->state != CPATH_STATE_OPEN) + return hop; + hop = hop->next; + } while (hop != cpath); + return NULL; +} + +/** Choose a suitable next hop in the cpath <b>head_ptr</b>, + * based on <b>state</b>. Append the hop info to head_ptr. + * + * Return 1 if the path is complete, 0 if we successfully added a hop, + * and -1 on error. + */ +STATIC int +onion_extend_cpath(origin_circuit_t *circ) +{ + uint8_t purpose = circ->base_.purpose; + cpath_build_state_t *state = circ->build_state; + int cur_len = circuit_get_cpath_len(circ); + extend_info_t *info = NULL; + + if (cur_len >= state->desired_path_len) { + log_debug(LD_CIRC, "Path is complete: %d steps long", + state->desired_path_len); + return 1; + } + + log_debug(LD_CIRC, "Path is %d long; we want %d", cur_len, + state->desired_path_len); + + if (cur_len == state->desired_path_len - 1) { /* Picking last node */ + info = extend_info_dup(state->chosen_exit); + } else if (cur_len == 0) { /* picking first node */ + const node_t *r = choose_good_entry_server(purpose, state, + &circ->guard_state); + if (r) { + /* If we're a client, use the preferred address rather than the + primary address, for potentially connecting to an IPv6 OR + port. Servers always want the primary (IPv4) address. */ + int client = (server_mode(get_options()) == 0); + info = extend_info_from_node(r, client); + /* Clients can fail to find an allowed address */ + tor_assert_nonfatal(info || client); + } + } else { + const node_t *r = + choose_good_middle_server(purpose, state, circ->cpath, cur_len); + if (r) { + info = extend_info_from_node(r, 0); + tor_assert_nonfatal(info); + } + } + + if (!info) { + log_warn(LD_CIRC,"Failed to find node for hop #%d of our path. Discarding " + "this circuit.", cur_len+1); + return -1; + } + + log_debug(LD_CIRC,"Chose router %s for hop #%d (exit is %s)", + extend_info_describe(info), + cur_len+1, build_state_get_exit_nickname(state)); + + onion_append_hop(&circ->cpath, info); + extend_info_free(info); + return 0; +} + +/** Create a new hop, annotate it with information about its + * corresponding router <b>choice</b>, and append it to the + * end of the cpath <b>head_ptr</b>. */ +STATIC int +onion_append_hop(crypt_path_t **head_ptr, extend_info_t *choice) +{ + crypt_path_t *hop = tor_malloc_zero(sizeof(crypt_path_t)); + + /* link hop into the cpath, at the end. */ + onion_append_to_cpath(head_ptr, hop); + + hop->magic = CRYPT_PATH_MAGIC; + hop->state = CPATH_STATE_CLOSED; + + hop->extend_info = extend_info_dup(choice); + + hop->package_window = circuit_initial_package_window(); + hop->deliver_window = CIRCWINDOW_START; + + return 0; +} + +/** Allocate a new extend_info object based on the various arguments. */ +extend_info_t * +extend_info_new(const char *nickname, + const char *rsa_id_digest, + const ed25519_public_key_t *ed_id, + crypto_pk_t *onion_key, + const curve25519_public_key_t *ntor_key, + const tor_addr_t *addr, uint16_t port) +{ + extend_info_t *info = tor_malloc_zero(sizeof(extend_info_t)); + memcpy(info->identity_digest, rsa_id_digest, DIGEST_LEN); + if (ed_id && !ed25519_public_key_is_zero(ed_id)) + memcpy(&info->ed_identity, ed_id, sizeof(ed25519_public_key_t)); + if (nickname) + strlcpy(info->nickname, nickname, sizeof(info->nickname)); + if (onion_key) + info->onion_key = crypto_pk_dup_key(onion_key); + if (ntor_key) + memcpy(&info->curve25519_onion_key, ntor_key, + sizeof(curve25519_public_key_t)); + tor_addr_copy(&info->addr, addr); + info->port = port; + return info; +} + +/** Allocate and return a new extend_info that can be used to build a + * circuit to or through the node <b>node</b>. Use the primary address + * of the node (i.e. its IPv4 address) unless + * <b>for_direct_connect</b> is true, in which case the preferred + * address is used instead. May return NULL if there is not enough + * info about <b>node</b> to extend to it--for example, if the preferred + * routerinfo_t or microdesc_t is missing, or if for_direct_connect is + * true and none of the node's addresses is allowed by tor's firewall + * and IP version config. + **/ +extend_info_t * +extend_info_from_node(const node_t *node, int for_direct_connect) +{ + tor_addr_port_t ap; + int valid_addr = 0; + + if (!node_has_preferred_descriptor(node, for_direct_connect)) { + return NULL; + } + + /* Choose a preferred address first, but fall back to an allowed address. */ + if (for_direct_connect) + fascist_firewall_choose_address_node(node, FIREWALL_OR_CONNECTION, 0, &ap); + else { + node_get_prim_orport(node, &ap); + } + valid_addr = tor_addr_port_is_valid_ap(&ap, 0); + + if (valid_addr) + log_debug(LD_CIRC, "using %s for %s", + fmt_addrport(&ap.addr, ap.port), + node->ri ? node->ri->nickname : node->rs->nickname); + else + log_warn(LD_CIRC, "Could not choose valid address for %s", + node->ri ? node->ri->nickname : node->rs->nickname); + + /* Every node we connect or extend to must support ntor */ + if (!node_has_curve25519_onion_key(node)) { + log_fn(LOG_PROTOCOL_WARN, LD_CIRC, + "Attempted to create extend_info for a node that does not support " + "ntor: %s", node_describe(node)); + return NULL; + } + + const ed25519_public_key_t *ed_pubkey = NULL; + + /* Don't send the ed25519 pubkey unless the target node actually supports + * authenticating with it. */ + if (node_supports_ed25519_link_authentication(node, 0)) { + log_info(LD_CIRC, "Including Ed25519 ID for %s", node_describe(node)); + ed_pubkey = node_get_ed25519_id(node); + } else if (node_get_ed25519_id(node)) { + log_info(LD_CIRC, "Not including the ed25519 ID for %s, since it won't " + "be able to authenticate it.", + node_describe(node)); + } + + /* Retrieve the curve25519 pubkey. */ + const curve25519_public_key_t *curve_pubkey = + node_get_curve25519_onion_key(node); + + if (valid_addr && node->ri) + return extend_info_new(node->ri->nickname, + node->identity, + ed_pubkey, + node->ri->onion_pkey, + curve_pubkey, + &ap.addr, + ap.port); + else if (valid_addr && node->rs && node->md) + return extend_info_new(node->rs->nickname, + node->identity, + ed_pubkey, + node->md->onion_pkey, + curve_pubkey, + &ap.addr, + ap.port); + else + return NULL; +} + +/** Release storage held by an extend_info_t struct. */ +void +extend_info_free_(extend_info_t *info) +{ + if (!info) + return; + crypto_pk_free(info->onion_key); + tor_free(info); +} + +/** Allocate and return a new extend_info_t with the same contents as + * <b>info</b>. */ +extend_info_t * +extend_info_dup(extend_info_t *info) +{ + extend_info_t *newinfo; + tor_assert(info); + newinfo = tor_malloc(sizeof(extend_info_t)); + memcpy(newinfo, info, sizeof(extend_info_t)); + if (info->onion_key) + newinfo->onion_key = crypto_pk_dup_key(info->onion_key); + else + newinfo->onion_key = NULL; + return newinfo; +} + +/** Return the node_t for the chosen exit router in <b>state</b>. + * If there is no chosen exit, or if we don't know the node_t for + * the chosen exit, return NULL. + */ +const node_t * +build_state_get_exit_node(cpath_build_state_t *state) +{ + if (!state || !state->chosen_exit) + return NULL; + return node_get_by_id(state->chosen_exit->identity_digest); +} + +/** Return the RSA ID digest for the chosen exit router in <b>state</b>. + * If there is no chosen exit, return NULL. + */ +const uint8_t * +build_state_get_exit_rsa_id(cpath_build_state_t *state) +{ + if (!state || !state->chosen_exit) + return NULL; + return (const uint8_t *) state->chosen_exit->identity_digest; +} + +/** Return the nickname for the chosen exit router in <b>state</b>. If + * there is no chosen exit, or if we don't know the routerinfo_t for the + * chosen exit, return NULL. + */ +const char * +build_state_get_exit_nickname(cpath_build_state_t *state) +{ + if (!state || !state->chosen_exit) + return NULL; + return state->chosen_exit->nickname; +} + +/** Return true iff the given address can be used to extend to. */ +int +extend_info_addr_is_allowed(const tor_addr_t *addr) +{ + tor_assert(addr); + + /* Check if we have a private address and if we can extend to it. */ + if ((tor_addr_is_internal(addr, 0) || tor_addr_is_multicast(addr)) && + !get_options()->ExtendAllowPrivateAddresses) { + goto disallow; + } + /* Allowed! */ + return 1; + disallow: + return 0; +} + +/* Does ei have a valid TAP key? */ +int +extend_info_supports_tap(const extend_info_t* ei) +{ + tor_assert(ei); + /* Valid TAP keys are not NULL */ + return ei->onion_key != NULL; +} + +/* Does ei have a valid ntor key? */ +int +extend_info_supports_ntor(const extend_info_t* ei) +{ + tor_assert(ei); + /* Valid ntor keys have at least one non-zero byte */ + return !tor_mem_is_zero( + (const char*)ei->curve25519_onion_key.public_key, + CURVE25519_PUBKEY_LEN); +} + +/* Is circuit purpose allowed to use the deprecated TAP encryption protocol? + * The hidden service protocol still uses TAP for some connections, because + * ntor onion keys aren't included in HS descriptors or INTRODUCE cells. */ +static int +circuit_purpose_can_use_tap_impl(uint8_t purpose) +{ + return (purpose == CIRCUIT_PURPOSE_S_CONNECT_REND || + purpose == CIRCUIT_PURPOSE_C_INTRODUCING); +} + +/* Is circ allowed to use the deprecated TAP encryption protocol? + * The hidden service protocol still uses TAP for some connections, because + * ntor onion keys aren't included in HS descriptors or INTRODUCE cells. */ +int +circuit_can_use_tap(const origin_circuit_t *circ) +{ + tor_assert(circ); + tor_assert(circ->cpath); + tor_assert(circ->cpath->extend_info); + return (circuit_purpose_can_use_tap_impl(circ->base_.purpose) && + extend_info_supports_tap(circ->cpath->extend_info)); +} + +/* Does circ have an onion key which it's allowed to use? */ +int +circuit_has_usable_onion_key(const origin_circuit_t *circ) +{ + tor_assert(circ); + tor_assert(circ->cpath); + tor_assert(circ->cpath->extend_info); + return (extend_info_supports_ntor(circ->cpath->extend_info) || + circuit_can_use_tap(circ)); +} + +/* Does ei have an onion key which it would prefer to use? + * Currently, we prefer ntor keys*/ +int +extend_info_has_preferred_onion_key(const extend_info_t* ei) +{ + tor_assert(ei); + return extend_info_supports_ntor(ei); +} + +/** Find the circuits that are waiting to find out whether their guards are + * usable, and if any are ready to become usable, mark them open and try + * attaching streams as appropriate. */ +void +circuit_upgrade_circuits_from_guard_wait(void) +{ + smartlist_t *to_upgrade = + circuit_find_circuits_to_upgrade_from_guard_wait(); + + if (to_upgrade == NULL) + return; + + log_info(LD_GUARD, "Upgrading %d circuits from 'waiting for better guard' " + "to 'open'.", smartlist_len(to_upgrade)); + + SMARTLIST_FOREACH_BEGIN(to_upgrade, origin_circuit_t *, circ) { + circuit_set_state(TO_CIRCUIT(circ), CIRCUIT_STATE_OPEN); + circuit_has_opened(circ); + } SMARTLIST_FOREACH_END(circ); + + smartlist_free(to_upgrade); +} + diff --cc src/core/or/circuitbuild.h index 0c6f2f6ce,000000000..cee71b297 mode 100644,000000..100644 --- a/src/core/or/circuitbuild.h +++ b/src/core/or/circuitbuild.h @@@ -1,103 -1,0 +1,102 @@@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file circuitbuild.h + * \brief Header file for circuitbuild.c. + **/ + +#ifndef TOR_CIRCUITBUILD_H +#define TOR_CIRCUITBUILD_H + +struct ed25519_public_key_t; +struct curve25519_public_key_t; + +int route_len_for_purpose(uint8_t purpose, extend_info_t *exit_ei); +char *circuit_list_path(origin_circuit_t *circ, int verbose); +char *circuit_list_path_for_controller(origin_circuit_t *circ); +void circuit_log_path(int severity, unsigned int domain, + origin_circuit_t *circ); +origin_circuit_t *origin_circuit_init(uint8_t purpose, int flags); +origin_circuit_t *circuit_establish_circuit(uint8_t purpose, + extend_info_t *exit, + int flags); +struct circuit_guard_state_t *origin_circuit_get_guard_state( + origin_circuit_t *circ); +int circuit_handle_first_hop(origin_circuit_t *circ); +void circuit_n_chan_done(channel_t *chan, int status, + int close_origin_circuits); +int inform_testing_reachability(void); +int circuit_timeout_want_to_count_circ(const origin_circuit_t *circ); +int circuit_send_next_onion_skin(origin_circuit_t *circ); +void circuit_note_clock_jumped(int64_t seconds_elapsed, bool was_idle); +int circuit_extend(cell_t *cell, circuit_t *circ); +int circuit_init_cpath_crypto(crypt_path_t *cpath, + const char *key_data, size_t key_data_len, + int reverse, int is_hs_v3); +struct created_cell_t; +int circuit_finish_handshake(origin_circuit_t *circ, + const struct created_cell_t *created_cell); - int circuit_truncated(origin_circuit_t *circ, crypt_path_t *layer, - int reason); ++int circuit_truncated(origin_circuit_t *circ, int reason); +int onionskin_answer(or_circuit_t *circ, + const struct created_cell_t *created_cell, + const char *keys, size_t keys_len, + const uint8_t *rend_circ_nonce); +MOCK_DECL(int, circuit_all_predicted_ports_handled, (time_t now, + int *need_uptime, + int *need_capacity)); + +int circuit_append_new_exit(origin_circuit_t *circ, extend_info_t *info); +int circuit_extend_to_new_exit(origin_circuit_t *circ, extend_info_t *info); +void onion_append_to_cpath(crypt_path_t **head_ptr, crypt_path_t *new_hop); +extend_info_t *extend_info_new(const char *nickname, + const char *rsa_id_digest, + const struct ed25519_public_key_t *ed_id, + crypto_pk_t *onion_key, + const struct curve25519_public_key_t *ntor_key, + const tor_addr_t *addr, uint16_t port); +extend_info_t *extend_info_from_node(const node_t *r, int for_direct_connect); +extend_info_t *extend_info_dup(extend_info_t *info); +void extend_info_free_(extend_info_t *info); +#define extend_info_free(info) \ + FREE_AND_NULL(extend_info_t, extend_info_free_, (info)) +int extend_info_addr_is_allowed(const tor_addr_t *addr); +int extend_info_supports_tap(const extend_info_t* ei); +int extend_info_supports_ntor(const extend_info_t* ei); +int circuit_can_use_tap(const origin_circuit_t *circ); +int circuit_has_usable_onion_key(const origin_circuit_t *circ); +int extend_info_has_preferred_onion_key(const extend_info_t* ei); +const uint8_t *build_state_get_exit_rsa_id(cpath_build_state_t *state); +const node_t *build_state_get_exit_node(cpath_build_state_t *state); +const char *build_state_get_exit_nickname(cpath_build_state_t *state); + +struct circuit_guard_state_t; + +const node_t *choose_good_entry_server(uint8_t purpose, + cpath_build_state_t *state, + struct circuit_guard_state_t **guard_state_out); +void circuit_upgrade_circuits_from_guard_wait(void); + +#ifdef CIRCUITBUILD_PRIVATE +STATIC circid_t get_unique_circ_id_by_chan(channel_t *chan); +STATIC int new_route_len(uint8_t purpose, extend_info_t *exit_ei, + smartlist_t *nodes); +MOCK_DECL(STATIC int, count_acceptable_nodes, (smartlist_t *nodes)); + +STATIC int onion_extend_cpath(origin_circuit_t *circ); + +STATIC int +onion_pick_cpath_exit(origin_circuit_t *circ, extend_info_t *exit_ei, + int is_hs_v3_rp_circuit); + +#if defined(TOR_UNIT_TESTS) +unsigned int cpath_get_n_hops(crypt_path_t **head_ptr); + +#endif /* defined(TOR_UNIT_TESTS) */ + +#endif /* defined(CIRCUITBUILD_PRIVATE) */ + +#endif /* !defined(TOR_CIRCUITBUILD_H) */ diff --cc src/core/or/circuitlist.c index f39e05ecd,000000000..78ecadab7 mode 100644,000000..100644 --- a/src/core/or/circuitlist.c +++ b/src/core/or/circuitlist.c @@@ -1,2742 -1,0 +1,2751 @@@ +/* Copyright 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file circuitlist.c + * + * \brief Manage global structures that list and index circuits, and + * look up circuits within them. + * + * One of the most frequent operations in Tor occurs every time that + * a relay cell arrives on a channel. When that happens, we need to + * find which circuit it is associated with, based on the channel and the + * circuit ID in the relay cell. + * + * To handle that, we maintain a global list of circuits, and a hashtable + * mapping [channel,circID] pairs to circuits. Circuits are added to and + * removed from this mapping using circuit_set_p_circid_chan() and + * circuit_set_n_circid_chan(). To look up a circuit from this map, most + * callers should use circuit_get_by_circid_channel(), though + * circuit_get_by_circid_channel_even_if_marked() is appropriate under some + * circumstances. + * + * We also need to allow for the possibility that we have blocked use of a + * circuit ID (because we are waiting to send a DESTROY cell), but the + * circuit is not there any more. For that case, we allow placeholder + * entries in the table, using channel_mark_circid_unusable(). + * + * To efficiently handle a channel that has just opened, we also maintain a + * list of the circuits waiting for channels, so we can attach them as + * needed without iterating through the whole list of circuits, using + * circuit_get_all_pending_on_channel(). + * + * In this module, we also handle the list of circuits that have been + * marked for close elsewhere, and close them as needed. (We use this + * "mark now, close later" pattern here and elsewhere to avoid + * unpredictable recursion if we closed every circuit immediately upon + * realizing it needed to close.) See circuit_mark_for_close() for the + * mark function, and circuit_close_all_marked() for the close function. + * + * For hidden services, we need to be able to look up introduction point + * circuits and rendezvous circuits by cookie, key, etc. These are + * currently handled with linear searches in + * circuit_get_ready_rend_circuit_by_rend_data(), + * circuit_get_next_by_pk_and_purpose(), and with hash lookups in + * circuit_get_rendezvous() and circuit_get_intro_point(). + * + * This module is also the entry point for our out-of-memory handler + * logic, which was originally circuit-focused. + **/ +#define CIRCUITLIST_PRIVATE +#include "lib/cc/torint.h" /* TOR_PRIuSZ */ + +#include "core/or/or.h" +#include "core/or/channel.h" +#include "feature/client/circpathbias.h" +#include "core/or/circuitbuild.h" +#include "core/or/circuitlist.h" +#include "core/or/circuituse.h" +#include "core/or/circuitstats.h" +#include "core/mainloop/connection.h" +#include "app/config/config.h" +#include "core/or/connection_edge.h" +#include "core/or/connection_or.h" +#include "feature/control/control.h" +#include "lib/crypt_ops/crypto_rand.h" +#include "lib/crypt_ops/crypto_util.h" +#include "lib/crypt_ops/crypto_dh.h" +#include "feature/dircache/directory.h" +#include "feature/client/entrynodes.h" +#include "core/mainloop/main.h" +#include "feature/hs/hs_circuit.h" +#include "feature/hs/hs_circuitmap.h" +#include "feature/hs/hs_ident.h" +#include "feature/nodelist/networkstatus.h" +#include "feature/nodelist/nodelist.h" +#include "core/crypto/onion.h" +#include "core/crypto/onion_fast.h" +#include "core/or/policies.h" +#include "core/or/relay.h" +#include "core/crypto/relay_crypto.h" +#include "feature/rend/rendclient.h" +#include "feature/rend/rendcommon.h" +#include "feature/stats/rephist.h" +#include "feature/nodelist/routerlist.h" +#include "feature/nodelist/routerset.h" +#include "core/or/channelpadding.h" +#include "lib/compress/compress.h" +#include "lib/compress/compress_lzma.h" +#include "lib/compress/compress_zlib.h" +#include "lib/compress/compress_zstd.h" +#include "lib/container/buffers.h" + +#include "ht.h" + +#include "core/or/cpath_build_state_st.h" +#include "core/or/crypt_path_reference_st.h" +#include "feature/dircommon/dir_connection_st.h" +#include "core/or/edge_connection_st.h" ++#include "core/or/half_edge_st.h" +#include "core/or/extend_info_st.h" +#include "core/or/or_circuit_st.h" +#include "core/or/origin_circuit_st.h" + +/********* START VARIABLES **********/ + +/** A global list of all circuits at this hop. */ +static smartlist_t *global_circuitlist = NULL; + +/** A global list of all origin circuits. Every element of this is also + * an element of global_circuitlist. */ +static smartlist_t *global_origin_circuit_list = NULL; + +/** A list of all the circuits in CIRCUIT_STATE_CHAN_WAIT. */ +static smartlist_t *circuits_pending_chans = NULL; + +/** List of all the (origin) circuits whose state is + * CIRCUIT_STATE_GUARD_WAIT. */ +static smartlist_t *circuits_pending_other_guards = NULL; + +/** A list of all the circuits that have been marked with + * circuit_mark_for_close and which are waiting for circuit_about_to_free. */ +static smartlist_t *circuits_pending_close = NULL; + +static void circuit_free_cpath_node(crypt_path_t *victim); +static void cpath_ref_decref(crypt_path_reference_t *cpath_ref); +static void circuit_about_to_free_atexit(circuit_t *circ); +static void circuit_about_to_free(circuit_t *circ); + +/** + * A cached value of the current state of the origin circuit list. Has the + * value 1 if we saw any opened circuits recently (since the last call to + * circuit_any_opened_circuits(), which gets called around once a second by + * circuit_expire_building). 0 otherwise. + */ +static int any_opened_circs_cached_val = 0; + +/********* END VARIABLES ************/ + +or_circuit_t * +TO_OR_CIRCUIT(circuit_t *x) +{ + tor_assert(x->magic == OR_CIRCUIT_MAGIC); + return DOWNCAST(or_circuit_t, x); +} +const or_circuit_t * +CONST_TO_OR_CIRCUIT(const circuit_t *x) +{ + tor_assert(x->magic == OR_CIRCUIT_MAGIC); + return DOWNCAST(or_circuit_t, x); +} +origin_circuit_t * +TO_ORIGIN_CIRCUIT(circuit_t *x) +{ + tor_assert(x->magic == ORIGIN_CIRCUIT_MAGIC); + return DOWNCAST(origin_circuit_t, x); +} +const origin_circuit_t * +CONST_TO_ORIGIN_CIRCUIT(const circuit_t *x) +{ + tor_assert(x->magic == ORIGIN_CIRCUIT_MAGIC); + return DOWNCAST(origin_circuit_t, x); +} + +/** A map from channel and circuit ID to circuit. (Lookup performance is + * very important here, since we need to do it every time a cell arrives.) */ +typedef struct chan_circid_circuit_map_t { + HT_ENTRY(chan_circid_circuit_map_t) node; + channel_t *chan; + circid_t circ_id; + circuit_t *circuit; + /* For debugging 12184: when was this placeholder item added? */ + time_t made_placeholder_at; +} chan_circid_circuit_map_t; + +/** Helper for hash tables: compare the channel and circuit ID for a and + * b, and return less than, equal to, or greater than zero appropriately. + */ +static inline int +chan_circid_entries_eq_(chan_circid_circuit_map_t *a, + chan_circid_circuit_map_t *b) +{ + return a->chan == b->chan && a->circ_id == b->circ_id; +} + +/** Helper: return a hash based on circuit ID and the pointer value of + * chan in <b>a</b>. */ +static inline unsigned int +chan_circid_entry_hash_(chan_circid_circuit_map_t *a) +{ + /* Try to squeze the siphash input into 8 bytes to save any extra siphash + * rounds. This hash function is in the critical path. */ + uintptr_t chan = (uintptr_t) (void*) a->chan; + uint32_t array[2]; + array[0] = a->circ_id; + /* The low bits of the channel pointer are uninteresting, since the channel + * is a pretty big structure. */ + array[1] = (uint32_t) (chan >> 6); + return (unsigned) siphash24g(array, sizeof(array)); +} + +/** Map from [chan,circid] to circuit. */ +static HT_HEAD(chan_circid_map, chan_circid_circuit_map_t) + chan_circid_map = HT_INITIALIZER(); +HT_PROTOTYPE(chan_circid_map, chan_circid_circuit_map_t, node, + chan_circid_entry_hash_, chan_circid_entries_eq_) +HT_GENERATE2(chan_circid_map, chan_circid_circuit_map_t, node, + chan_circid_entry_hash_, chan_circid_entries_eq_, 0.6, + tor_reallocarray_, tor_free_) + +/** The most recently returned entry from circuit_get_by_circid_chan; + * used to improve performance when many cells arrive in a row from the + * same circuit. + */ +static chan_circid_circuit_map_t *_last_circid_chan_ent = NULL; + +/** Implementation helper for circuit_set_{p,n}_circid_channel: A circuit ID + * and/or channel for circ has just changed from <b>old_chan, old_id</b> + * to <b>chan, id</b>. Adjust the chan,circid map as appropriate, removing + * the old entry (if any) and adding a new one. */ +static void +circuit_set_circid_chan_helper(circuit_t *circ, int direction, + circid_t id, + channel_t *chan) +{ + chan_circid_circuit_map_t search; + chan_circid_circuit_map_t *found; + channel_t *old_chan, **chan_ptr; + circid_t old_id, *circid_ptr; + int make_active, attached = 0; + + if (direction == CELL_DIRECTION_OUT) { + chan_ptr = &circ->n_chan; + circid_ptr = &circ->n_circ_id; + make_active = circ->n_chan_cells.n > 0; + } else { + or_circuit_t *c = TO_OR_CIRCUIT(circ); + chan_ptr = &c->p_chan; + circid_ptr = &c->p_circ_id; + make_active = c->p_chan_cells.n > 0; + } + old_chan = *chan_ptr; + old_id = *circid_ptr; + + if (id == old_id && chan == old_chan) + return; + + if (_last_circid_chan_ent && + ((old_id == _last_circid_chan_ent->circ_id && + old_chan == _last_circid_chan_ent->chan) || + (id == _last_circid_chan_ent->circ_id && + chan == _last_circid_chan_ent->chan))) { + _last_circid_chan_ent = NULL; + } + + if (old_chan) { + /* + * If we're changing channels or ID and had an old channel and a non + * zero old ID and weren't marked for close (i.e., we should have been + * attached), detach the circuit. ID changes require this because + * circuitmux hashes on (channel_id, circuit_id). + */ + if (old_id != 0 && (old_chan != chan || old_id != id) && + !(circ->marked_for_close)) { + tor_assert(old_chan->cmux); + circuitmux_detach_circuit(old_chan->cmux, circ); + } + + /* we may need to remove it from the conn-circid map */ + search.circ_id = old_id; + search.chan = old_chan; + found = HT_REMOVE(chan_circid_map, &chan_circid_map, &search); + if (found) { + tor_free(found); + if (direction == CELL_DIRECTION_OUT) { + /* One fewer circuits use old_chan as n_chan */ + --(old_chan->num_n_circuits); + } else { + /* One fewer circuits use old_chan as p_chan */ + --(old_chan->num_p_circuits); + } + } + } + + /* Change the values only after we have possibly made the circuit inactive + * on the previous chan. */ + *chan_ptr = chan; + *circid_ptr = id; + + if (chan == NULL) + return; + + /* now add the new one to the conn-circid map */ + search.circ_id = id; + search.chan = chan; + found = HT_FIND(chan_circid_map, &chan_circid_map, &search); + if (found) { + found->circuit = circ; + found->made_placeholder_at = 0; + } else { + found = tor_malloc_zero(sizeof(chan_circid_circuit_map_t)); + found->circ_id = id; + found->chan = chan; + found->circuit = circ; + HT_INSERT(chan_circid_map, &chan_circid_map, found); + } + + /* + * Attach to the circuitmux if we're changing channels or IDs and + * have a new channel and ID to use and the circuit is not marked for + * close. + */ + if (chan && id != 0 && (old_chan != chan || old_id != id) && + !(circ->marked_for_close)) { + tor_assert(chan->cmux); + circuitmux_attach_circuit(chan->cmux, circ, direction); + attached = 1; + } + + /* + * This is a no-op if we have no cells, but if we do it marks us active to + * the circuitmux + */ + if (make_active && attached) + update_circuit_on_cmux(circ, direction); + + /* Adjust circuit counts on new channel */ + if (direction == CELL_DIRECTION_OUT) { + ++chan->num_n_circuits; + } else { + ++chan->num_p_circuits; + } +} + +/** Mark that circuit id <b>id</b> shouldn't be used on channel <b>chan</b>, + * even if there is no circuit on the channel. We use this to keep the + * circuit id from getting re-used while we have queued but not yet sent + * a destroy cell. */ +void +channel_mark_circid_unusable(channel_t *chan, circid_t id) +{ + chan_circid_circuit_map_t search; + chan_circid_circuit_map_t *ent; + + /* See if there's an entry there. That wouldn't be good. */ + memset(&search, 0, sizeof(search)); + search.chan = chan; + search.circ_id = id; + ent = HT_FIND(chan_circid_map, &chan_circid_map, &search); + + if (ent && ent->circuit) { + /* we have a problem. */ + log_warn(LD_BUG, "Tried to mark %u unusable on %p, but there was already " + "a circuit there.", (unsigned)id, chan); + } else if (ent) { + /* It's already marked. */ + if (!ent->made_placeholder_at) + ent->made_placeholder_at = approx_time(); + } else { + ent = tor_malloc_zero(sizeof(chan_circid_circuit_map_t)); + ent->chan = chan; + ent->circ_id = id; + /* leave circuit at NULL. */ + ent->made_placeholder_at = approx_time(); + HT_INSERT(chan_circid_map, &chan_circid_map, ent); + } +} + +/** Mark that a circuit id <b>id</b> can be used again on <b>chan</b>. + * We use this to re-enable the circuit ID after we've sent a destroy cell. + */ +void +channel_mark_circid_usable(channel_t *chan, circid_t id) +{ + chan_circid_circuit_map_t search; + chan_circid_circuit_map_t *ent; + + /* See if there's an entry there. That wouldn't be good. */ + memset(&search, 0, sizeof(search)); + search.chan = chan; + search.circ_id = id; + ent = HT_REMOVE(chan_circid_map, &chan_circid_map, &search); + if (ent && ent->circuit) { + log_warn(LD_BUG, "Tried to mark %u usable on %p, but there was already " + "a circuit there.", (unsigned)id, chan); + return; + } + if (_last_circid_chan_ent == ent) + _last_circid_chan_ent = NULL; + tor_free(ent); +} + +/** Called to indicate that a DESTROY is pending on <b>chan</b> with + * circuit ID <b>id</b>, but hasn't been sent yet. */ +void +channel_note_destroy_pending(channel_t *chan, circid_t id) +{ + circuit_t *circ = circuit_get_by_circid_channel_even_if_marked(id,chan); + if (circ) { + if (circ->n_chan == chan && circ->n_circ_id == id) { + circ->n_delete_pending = 1; + } else { + or_circuit_t *orcirc = TO_OR_CIRCUIT(circ); + if (orcirc->p_chan == chan && orcirc->p_circ_id == id) { + circ->p_delete_pending = 1; + } + } + return; + } + channel_mark_circid_unusable(chan, id); +} + +/** Called to indicate that a DESTROY is no longer pending on <b>chan</b> with + * circuit ID <b>id</b> -- typically, because it has been sent. */ +MOCK_IMPL(void, +channel_note_destroy_not_pending,(channel_t *chan, circid_t id)) +{ + circuit_t *circ = circuit_get_by_circid_channel_even_if_marked(id,chan); + if (circ) { + if (circ->n_chan == chan && circ->n_circ_id == id) { + circ->n_delete_pending = 0; + } else { + or_circuit_t *orcirc = TO_OR_CIRCUIT(circ); + if (orcirc->p_chan == chan && orcirc->p_circ_id == id) { + circ->p_delete_pending = 0; + } + } + /* XXXX this shouldn't happen; log a bug here. */ + return; + } + channel_mark_circid_usable(chan, id); +} + +/** Set the p_conn field of a circuit <b>circ</b>, along + * with the corresponding circuit ID, and add the circuit as appropriate + * to the (chan,id)->circuit map. */ +void +circuit_set_p_circid_chan(or_circuit_t *or_circ, circid_t id, + channel_t *chan) +{ + circuit_t *circ = TO_CIRCUIT(or_circ); + channel_t *old_chan = or_circ->p_chan; + circid_t old_id = or_circ->p_circ_id; + + circuit_set_circid_chan_helper(circ, CELL_DIRECTION_IN, id, chan); + + if (chan) { + chan->timestamp_last_had_circuits = approx_time(); + } + + if (circ->p_delete_pending && old_chan) { + channel_mark_circid_unusable(old_chan, old_id); + circ->p_delete_pending = 0; + } +} + +/** Set the n_conn field of a circuit <b>circ</b>, along + * with the corresponding circuit ID, and add the circuit as appropriate + * to the (chan,id)->circuit map. */ +void +circuit_set_n_circid_chan(circuit_t *circ, circid_t id, + channel_t *chan) +{ + channel_t *old_chan = circ->n_chan; + circid_t old_id = circ->n_circ_id; + + circuit_set_circid_chan_helper(circ, CELL_DIRECTION_OUT, id, chan); + + if (chan) { + chan->timestamp_last_had_circuits = approx_time(); + } + + if (circ->n_delete_pending && old_chan) { + channel_mark_circid_unusable(old_chan, old_id); + circ->n_delete_pending = 0; + } +} + +/** Change the state of <b>circ</b> to <b>state</b>, adding it to or removing + * it from lists as appropriate. */ +void +circuit_set_state(circuit_t *circ, uint8_t state) +{ + tor_assert(circ); + if (state == circ->state) + return; + if (PREDICT_UNLIKELY(!circuits_pending_chans)) + circuits_pending_chans = smartlist_new(); + if (PREDICT_UNLIKELY(!circuits_pending_other_guards)) + circuits_pending_other_guards = smartlist_new(); + if (circ->state == CIRCUIT_STATE_CHAN_WAIT) { + /* remove from waiting-circuit list. */ + smartlist_remove(circuits_pending_chans, circ); + } + if (state == CIRCUIT_STATE_CHAN_WAIT) { + /* add to waiting-circuit list. */ + smartlist_add(circuits_pending_chans, circ); + } + if (circ->state == CIRCUIT_STATE_GUARD_WAIT) { + smartlist_remove(circuits_pending_other_guards, circ); + } + if (state == CIRCUIT_STATE_GUARD_WAIT) { + smartlist_add(circuits_pending_other_guards, circ); + } + if (state == CIRCUIT_STATE_GUARD_WAIT || state == CIRCUIT_STATE_OPEN) + tor_assert(!circ->n_chan_create_cell); + circ->state = state; +} + +/** Append to <b>out</b> all circuits in state CHAN_WAIT waiting for + * the given connection. */ +void +circuit_get_all_pending_on_channel(smartlist_t *out, channel_t *chan) +{ + tor_assert(out); + tor_assert(chan); + + if (!circuits_pending_chans) + return; + + SMARTLIST_FOREACH_BEGIN(circuits_pending_chans, circuit_t *, circ) { + if (circ->marked_for_close) + continue; + if (!circ->n_hop) + continue; + tor_assert(circ->state == CIRCUIT_STATE_CHAN_WAIT); + if (tor_digest_is_zero(circ->n_hop->identity_digest)) { + /* Look at addr/port. This is an unkeyed connection. */ + if (!channel_matches_extend_info(chan, circ->n_hop)) + continue; + } else { + /* We expected a key. See if it's the right one. */ + if (tor_memneq(chan->identity_digest, + circ->n_hop->identity_digest, DIGEST_LEN)) + continue; + } + smartlist_add(out, circ); + } SMARTLIST_FOREACH_END(circ); +} + +/** Return the number of circuits in state CHAN_WAIT, waiting for the given + * channel. */ +int +circuit_count_pending_on_channel(channel_t *chan) +{ + int cnt; + smartlist_t *sl = smartlist_new(); + + tor_assert(chan); + + circuit_get_all_pending_on_channel(sl, chan); + cnt = smartlist_len(sl); + smartlist_free(sl); + log_debug(LD_CIRC,"or_conn to %s, %d pending circs", + channel_get_canonical_remote_descr(chan), + cnt); + return cnt; +} + +/** Remove <b>origin_circ</b> from the global list of origin circuits. + * Called when we are freeing a circuit. + */ +static void +circuit_remove_from_origin_circuit_list(origin_circuit_t *origin_circ) +{ + int origin_idx = origin_circ->global_origin_circuit_list_idx; + if (origin_idx < 0) + return; + origin_circuit_t *c2; + tor_assert(origin_idx <= smartlist_len(global_origin_circuit_list)); + c2 = smartlist_get(global_origin_circuit_list, origin_idx); + tor_assert(origin_circ == c2); + smartlist_del(global_origin_circuit_list, origin_idx); + if (origin_idx < smartlist_len(global_origin_circuit_list)) { + origin_circuit_t *replacement = + smartlist_get(global_origin_circuit_list, origin_idx); + replacement->global_origin_circuit_list_idx = origin_idx; + } + origin_circ->global_origin_circuit_list_idx = -1; +} + +/** Add <b>origin_circ</b> to the global list of origin circuits. Called + * when creating the circuit. */ +static void +circuit_add_to_origin_circuit_list(origin_circuit_t *origin_circ) +{ + tor_assert(origin_circ->global_origin_circuit_list_idx == -1); + smartlist_t *lst = circuit_get_global_origin_circuit_list(); + smartlist_add(lst, origin_circ); + origin_circ->global_origin_circuit_list_idx = smartlist_len(lst) - 1; +} + +/** Detach from the global circuit list, and deallocate, all + * circuits that have been marked for close. + */ +void +circuit_close_all_marked(void) +{ + if (circuits_pending_close == NULL) + return; + + smartlist_t *lst = circuit_get_global_list(); + SMARTLIST_FOREACH_BEGIN(circuits_pending_close, circuit_t *, circ) { + tor_assert(circ->marked_for_close); + + /* Remove it from the circuit list. */ + int idx = circ->global_circuitlist_idx; + smartlist_del(lst, idx); + if (idx < smartlist_len(lst)) { + circuit_t *replacement = smartlist_get(lst, idx); + replacement->global_circuitlist_idx = idx; + } + circ->global_circuitlist_idx = -1; + + /* Remove it from the origin circuit list, if appropriate. */ + if (CIRCUIT_IS_ORIGIN(circ)) { + circuit_remove_from_origin_circuit_list(TO_ORIGIN_CIRCUIT(circ)); + } + + circuit_about_to_free(circ); + circuit_free(circ); + } SMARTLIST_FOREACH_END(circ); + + smartlist_clear(circuits_pending_close); +} + +/** Return a pointer to the global list of circuits. */ +MOCK_IMPL(smartlist_t *, +circuit_get_global_list,(void)) +{ + if (NULL == global_circuitlist) + global_circuitlist = smartlist_new(); + return global_circuitlist; +} + +/** Return a pointer to the global list of origin circuits. */ +smartlist_t * +circuit_get_global_origin_circuit_list(void) +{ + if (NULL == global_origin_circuit_list) + global_origin_circuit_list = smartlist_new(); + return global_origin_circuit_list; +} + +/** + * Return true if we have any opened general-purpose 3 hop + * origin circuits. + * + * The result from this function is cached for use by + * circuit_any_opened_circuits_cached(). + */ +int +circuit_any_opened_circuits(void) +{ + SMARTLIST_FOREACH_BEGIN(circuit_get_global_origin_circuit_list(), + const origin_circuit_t *, next_circ) { + if (!TO_CIRCUIT(next_circ)->marked_for_close && + next_circ->has_opened && + TO_CIRCUIT(next_circ)->state == CIRCUIT_STATE_OPEN && + TO_CIRCUIT(next_circ)->purpose != CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT && + next_circ->build_state && + next_circ->build_state->desired_path_len == DEFAULT_ROUTE_LEN) { + circuit_cache_opened_circuit_state(1); + return 1; + } + } SMARTLIST_FOREACH_END(next_circ); + + circuit_cache_opened_circuit_state(0); + return 0; +} + +/** + * Cache the "any circuits opened" state, as specified in param + * circuits_are_opened. This is a helper function to update + * the circuit opened status whenever we happen to look at the + * circuit list. + */ +void +circuit_cache_opened_circuit_state(int circuits_are_opened) +{ + any_opened_circs_cached_val = circuits_are_opened; +} + +/** + * Return true if there were any opened circuits since the last call to + * circuit_any_opened_circuits(), or since circuit_expire_building() last + * ran (it runs roughly once per second). + */ +int +circuit_any_opened_circuits_cached(void) +{ + return any_opened_circs_cached_val; +} + +/** Function to make circ->state human-readable */ +const char * +circuit_state_to_string(int state) +{ + static char buf[64]; + switch (state) { + case CIRCUIT_STATE_BUILDING: return "doing handshakes"; + case CIRCUIT_STATE_ONIONSKIN_PENDING: return "processing the onion"; + case CIRCUIT_STATE_CHAN_WAIT: return "connecting to server"; + case CIRCUIT_STATE_GUARD_WAIT: return "waiting to see how other " + "guards perform"; + case CIRCUIT_STATE_OPEN: return "open"; + default: + log_warn(LD_BUG, "Unknown circuit state %d", state); + tor_snprintf(buf, sizeof(buf), "unknown state [%d]", state); + return buf; + } +} + +/** Map a circuit purpose to a string suitable to be displayed to a + * controller. */ +const char * +circuit_purpose_to_controller_string(uint8_t purpose) +{ + static char buf[32]; + switch (purpose) { + case CIRCUIT_PURPOSE_OR: + case CIRCUIT_PURPOSE_INTRO_POINT: + case CIRCUIT_PURPOSE_REND_POINT_WAITING: + case CIRCUIT_PURPOSE_REND_ESTABLISHED: + return "SERVER"; /* A controller should never see these, actually. */ + + case CIRCUIT_PURPOSE_C_GENERAL: + return "GENERAL"; + + case CIRCUIT_PURPOSE_C_HSDIR_GET: + return "HS_CLIENT_HSDIR"; + + case CIRCUIT_PURPOSE_C_INTRODUCING: + case CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT: + case CIRCUIT_PURPOSE_C_INTRODUCE_ACKED: + return "HS_CLIENT_INTRO"; + + case CIRCUIT_PURPOSE_C_ESTABLISH_REND: + case CIRCUIT_PURPOSE_C_REND_READY: + case CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED: + case CIRCUIT_PURPOSE_C_REND_JOINED: + return "HS_CLIENT_REND"; + + case CIRCUIT_PURPOSE_S_HSDIR_POST: + return "HS_SERVICE_HSDIR"; + + case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO: + case CIRCUIT_PURPOSE_S_INTRO: + return "HS_SERVICE_INTRO"; + + case CIRCUIT_PURPOSE_S_CONNECT_REND: + case CIRCUIT_PURPOSE_S_REND_JOINED: + return "HS_SERVICE_REND"; + + case CIRCUIT_PURPOSE_TESTING: + return "TESTING"; + case CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT: + return "MEASURE_TIMEOUT"; + case CIRCUIT_PURPOSE_CONTROLLER: + return "CONTROLLER"; + case CIRCUIT_PURPOSE_PATH_BIAS_TESTING: + return "PATH_BIAS_TESTING"; + case CIRCUIT_PURPOSE_HS_VANGUARDS: + return "HS_VANGUARDS"; + + default: + tor_snprintf(buf, sizeof(buf), "UNKNOWN_%d", (int)purpose); + return buf; + } +} + +/** Return a string specifying the state of the hidden-service circuit + * purpose <b>purpose</b>, or NULL if <b>purpose</b> is not a + * hidden-service-related circuit purpose. */ +const char * +circuit_purpose_to_controller_hs_state_string(uint8_t purpose) +{ + switch (purpose) + { + default: + log_fn(LOG_WARN, LD_BUG, + "Unrecognized circuit purpose: %d", + (int)purpose); + tor_fragile_assert(); + /* fall through */ + + case CIRCUIT_PURPOSE_OR: + case CIRCUIT_PURPOSE_C_GENERAL: + case CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT: + case CIRCUIT_PURPOSE_TESTING: + case CIRCUIT_PURPOSE_CONTROLLER: + case CIRCUIT_PURPOSE_PATH_BIAS_TESTING: + case CIRCUIT_PURPOSE_HS_VANGUARDS: + return NULL; + + case CIRCUIT_PURPOSE_INTRO_POINT: + return "OR_HSSI_ESTABLISHED"; + case CIRCUIT_PURPOSE_REND_POINT_WAITING: + return "OR_HSCR_ESTABLISHED"; + case CIRCUIT_PURPOSE_REND_ESTABLISHED: + return "OR_HS_R_JOINED"; + + case CIRCUIT_PURPOSE_C_HSDIR_GET: + case CIRCUIT_PURPOSE_C_INTRODUCING: + return "HSCI_CONNECTING"; + case CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT: + return "HSCI_INTRO_SENT"; + case CIRCUIT_PURPOSE_C_INTRODUCE_ACKED: + return "HSCI_DONE"; + + case CIRCUIT_PURPOSE_C_ESTABLISH_REND: + return "HSCR_CONNECTING"; + case CIRCUIT_PURPOSE_C_REND_READY: + return "HSCR_ESTABLISHED_IDLE"; + case CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED: + return "HSCR_ESTABLISHED_WAITING"; + case CIRCUIT_PURPOSE_C_REND_JOINED: + return "HSCR_JOINED"; + + case CIRCUIT_PURPOSE_S_HSDIR_POST: + case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO: + return "HSSI_CONNECTING"; + case CIRCUIT_PURPOSE_S_INTRO: + return "HSSI_ESTABLISHED"; + + case CIRCUIT_PURPOSE_S_CONNECT_REND: + return "HSSR_CONNECTING"; + case CIRCUIT_PURPOSE_S_REND_JOINED: + return "HSSR_JOINED"; + } +} + +/** Return a human-readable string for the circuit purpose <b>purpose</b>. */ +const char * +circuit_purpose_to_string(uint8_t purpose) +{ + static char buf[32]; + + switch (purpose) + { + case CIRCUIT_PURPOSE_OR: + return "Circuit at relay"; + case CIRCUIT_PURPOSE_INTRO_POINT: + return "Acting as intro point"; + case CIRCUIT_PURPOSE_REND_POINT_WAITING: + return "Acting as rendezvous (pending)"; + case CIRCUIT_PURPOSE_REND_ESTABLISHED: + return "Acting as rendezvous (established)"; + case CIRCUIT_PURPOSE_C_GENERAL: + return "General-purpose client"; + case CIRCUIT_PURPOSE_C_INTRODUCING: + return "Hidden service client: Connecting to intro point"; + case CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT: + return "Hidden service client: Waiting for ack from intro point"; + case CIRCUIT_PURPOSE_C_INTRODUCE_ACKED: + return "Hidden service client: Received ack from intro point"; + case CIRCUIT_PURPOSE_C_ESTABLISH_REND: + return "Hidden service client: Establishing rendezvous point"; + case CIRCUIT_PURPOSE_C_REND_READY: + return "Hidden service client: Pending rendezvous point"; + case CIRCUIT_PURPOSE_C_REND_READY_INTRO_ACKED: + return "Hidden service client: Pending rendezvous point (ack received)"; + case CIRCUIT_PURPOSE_C_REND_JOINED: + return "Hidden service client: Active rendezvous point"; + case CIRCUIT_PURPOSE_C_HSDIR_GET: + return "Hidden service client: Fetching HS descriptor"; + + case CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT: + return "Measuring circuit timeout"; + + case CIRCUIT_PURPOSE_S_ESTABLISH_INTRO: + return "Hidden service: Establishing introduction point"; + case CIRCUIT_PURPOSE_S_INTRO: + return "Hidden service: Introduction point"; + case CIRCUIT_PURPOSE_S_CONNECT_REND: + return "Hidden service: Connecting to rendezvous point"; + case CIRCUIT_PURPOSE_S_REND_JOINED: + return "Hidden service: Active rendezvous point"; + case CIRCUIT_PURPOSE_S_HSDIR_POST: + return "Hidden service: Uploading HS descriptor"; + + case CIRCUIT_PURPOSE_TESTING: + return "Testing circuit"; + + case CIRCUIT_PURPOSE_CONTROLLER: + return "Circuit made by controller"; + + case CIRCUIT_PURPOSE_PATH_BIAS_TESTING: + return "Path-bias testing circuit"; + + case CIRCUIT_PURPOSE_HS_VANGUARDS: + return "Hidden service: Pre-built vanguard circuit"; + + default: + tor_snprintf(buf, sizeof(buf), "UNKNOWN_%d", (int)purpose); + return buf; + } +} + +/** Pick a reasonable package_window to start out for our circuits. + * Originally this was hard-coded at 1000, but now the consensus votes + * on the answer. See proposal 168. */ +int32_t +circuit_initial_package_window(void) +{ + int32_t num = networkstatus_get_param(NULL, "circwindow", CIRCWINDOW_START, + CIRCWINDOW_START_MIN, + CIRCWINDOW_START_MAX); + /* If the consensus tells us a negative number, we'd assert. */ + if (num < 0) + num = CIRCWINDOW_START; + return num; +} + +/** Initialize the common elements in a circuit_t, and add it to the global + * list. */ +static void +init_circuit_base(circuit_t *circ) +{ + tor_gettimeofday(&circ->timestamp_created); + + // Gets reset when we send CREATE_FAST. + // circuit_expire_building() expects these to be equal + // until the orconn is built. + circ->timestamp_began = circ->timestamp_created; + + circ->package_window = circuit_initial_package_window(); + circ->deliver_window = CIRCWINDOW_START; + cell_queue_init(&circ->n_chan_cells); + + smartlist_add(circuit_get_global_list(), circ); + circ->global_circuitlist_idx = smartlist_len(circuit_get_global_list()) - 1; +} + +/** If we haven't yet decided on a good timeout value for circuit + * building, we close idle circuits aggressively so we can get more + * data points. These are the default, min, and max consensus values */ +#define DFLT_IDLE_TIMEOUT_WHILE_LEARNING (3*60) +#define MIN_IDLE_TIMEOUT_WHILE_LEARNING (10) +#define MAX_IDLE_TIMEOUT_WHILE_LEARNING (1000*60) + +/** Allocate space for a new circuit, initializing with <b>p_circ_id</b> + * and <b>p_conn</b>. Add it to the global circuit list. + */ +origin_circuit_t * +origin_circuit_new(void) +{ + origin_circuit_t *circ; + /* never zero, since a global ID of 0 is treated specially by the + * controller */ + static uint32_t n_circuits_allocated = 1; + + circ = tor_malloc_zero(sizeof(origin_circuit_t)); + circ->base_.magic = ORIGIN_CIRCUIT_MAGIC; + + circ->next_stream_id = crypto_rand_int(1<<16); + circ->global_identifier = n_circuits_allocated++; + circ->remaining_relay_early_cells = MAX_RELAY_EARLY_CELLS_PER_CIRCUIT; + circ->remaining_relay_early_cells -= crypto_rand_int(2); + + init_circuit_base(TO_CIRCUIT(circ)); + + /* Add to origin-list. */ + circ->global_origin_circuit_list_idx = -1; + circuit_add_to_origin_circuit_list(circ); + + circuit_build_times_update_last_circ(get_circuit_build_times_mutable()); + + if (! circuit_build_times_disabled(get_options()) && + circuit_build_times_needs_circuits(get_circuit_build_times())) { + /* Circuits should be shorter lived if we need more of them + * for learning a good build timeout */ + circ->circuit_idle_timeout = + networkstatus_get_param(NULL, "cbtlearntimeout", + DFLT_IDLE_TIMEOUT_WHILE_LEARNING, + MIN_IDLE_TIMEOUT_WHILE_LEARNING, + MAX_IDLE_TIMEOUT_WHILE_LEARNING); + } else { + // This should always be larger than the current port prediction time + // remaining, or else we'll end up with the case where a circuit times out + // and another one is built, effectively doubling the timeout window. + // + // We also randomize it by up to 5% more (ie 5% of 0 to 3600 seconds, + // depending on how much circuit prediction time is remaining) so that + // we don't close a bunch of unused circuits all at the same time. + int prediction_time_remaining = + predicted_ports_prediction_time_remaining(time(NULL)); + circ->circuit_idle_timeout = prediction_time_remaining+1+ + crypto_rand_int(1+prediction_time_remaining/20); + + if (circ->circuit_idle_timeout <= 0) { + log_warn(LD_BUG, + "Circuit chose a negative idle timeout of %d based on " + "%d seconds of predictive building remaining.", + circ->circuit_idle_timeout, + prediction_time_remaining); + circ->circuit_idle_timeout = + networkstatus_get_param(NULL, "cbtlearntimeout", + DFLT_IDLE_TIMEOUT_WHILE_LEARNING, + MIN_IDLE_TIMEOUT_WHILE_LEARNING, + MAX_IDLE_TIMEOUT_WHILE_LEARNING); + } + + log_info(LD_CIRC, + "Circuit %"PRIu32" chose an idle timeout of %d based on " + "%d seconds of predictive building remaining.", + (circ->global_identifier), + circ->circuit_idle_timeout, + prediction_time_remaining); + } + + return circ; +} + +/** Allocate a new or_circuit_t, connected to <b>p_chan</b> as + * <b>p_circ_id</b>. If <b>p_chan</b> is NULL, the circuit is unattached. */ +or_circuit_t * +or_circuit_new(circid_t p_circ_id, channel_t *p_chan) +{ + /* CircIDs */ + or_circuit_t *circ; + + circ = tor_malloc_zero(sizeof(or_circuit_t)); + circ->base_.magic = OR_CIRCUIT_MAGIC; + + if (p_chan) + circuit_set_p_circid_chan(circ, p_circ_id, p_chan); + + circ->remaining_relay_early_cells = MAX_RELAY_EARLY_CELLS_PER_CIRCUIT; + cell_queue_init(&circ->p_chan_cells); + + init_circuit_base(TO_CIRCUIT(circ)); + + return circ; +} + +/** Free all storage held in circ->testing_cell_stats */ +void +circuit_clear_testing_cell_stats(circuit_t *circ) +{ + if (!circ || !circ->testing_cell_stats) + return; + SMARTLIST_FOREACH(circ->testing_cell_stats, testing_cell_stats_entry_t *, + ent, tor_free(ent)); + smartlist_free(circ->testing_cell_stats); + circ->testing_cell_stats = NULL; +} + +/** Deallocate space associated with circ. + */ +STATIC void +circuit_free_(circuit_t *circ) +{ + circid_t n_circ_id = 0; + void *mem; + size_t memlen; + int should_free = 1; + if (!circ) + return; + + /* We keep a copy of this so we can log its value before it gets unset. */ + n_circ_id = circ->n_circ_id; + + circuit_clear_testing_cell_stats(circ); + + /* Cleanup circuit from anything HS v3 related. We also do this when the + * circuit is closed. This is to avoid any code path that free registered + * circuits without closing them before. This needs to be done before the + * hs identifier is freed. */ + hs_circ_cleanup(circ); + + if (CIRCUIT_IS_ORIGIN(circ)) { + origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); + mem = ocirc; + memlen = sizeof(origin_circuit_t); + tor_assert(circ->magic == ORIGIN_CIRCUIT_MAGIC); + + circuit_remove_from_origin_circuit_list(ocirc); + ++ if (ocirc->half_streams) { ++ SMARTLIST_FOREACH_BEGIN(ocirc->half_streams, half_edge_t*, ++ half_conn) { ++ tor_free(half_conn); ++ } SMARTLIST_FOREACH_END(half_conn); ++ smartlist_free(ocirc->half_streams); ++ } ++ + if (ocirc->build_state) { + extend_info_free(ocirc->build_state->chosen_exit); + circuit_free_cpath_node(ocirc->build_state->pending_final_cpath); + cpath_ref_decref(ocirc->build_state->service_pending_final_cpath_ref); + } + tor_free(ocirc->build_state); + + /* Cancel before freeing, if we haven't already succeeded or failed. */ + if (ocirc->guard_state) { + entry_guard_cancel(ô->guard_state); + } + circuit_guard_state_free(ocirc->guard_state); + + circuit_clear_cpath(ocirc); + + crypto_pk_free(ocirc->intro_key); + rend_data_free(ocirc->rend_data); + + /* Finally, free the identifier of the circuit and nullify it so multiple + * cleanup will work. */ + hs_ident_circuit_free(ocirc->hs_ident); + ocirc->hs_ident = NULL; + + tor_free(ocirc->dest_address); + if (ocirc->socks_username) { + memwipe(ocirc->socks_username, 0x12, ocirc->socks_username_len); + tor_free(ocirc->socks_username); + } + if (ocirc->socks_password) { + memwipe(ocirc->socks_password, 0x06, ocirc->socks_password_len); + tor_free(ocirc->socks_password); + } + addr_policy_list_free(ocirc->prepend_policy); + } else { + or_circuit_t *ocirc = TO_OR_CIRCUIT(circ); + /* Remember cell statistics for this circuit before deallocating. */ + if (get_options()->CellStatistics) + rep_hist_buffer_stats_add_circ(circ, time(NULL)); + mem = ocirc; + memlen = sizeof(or_circuit_t); + tor_assert(circ->magic == OR_CIRCUIT_MAGIC); + + should_free = (ocirc->workqueue_entry == NULL); + + relay_crypto_clear(ô->crypto); + + if (ocirc->rend_splice) { + or_circuit_t *other = ocirc->rend_splice; + tor_assert(other->base_.magic == OR_CIRCUIT_MAGIC); + other->rend_splice = NULL; + } + + /* remove from map. */ + circuit_set_p_circid_chan(ocirc, 0, NULL); + + /* Clear cell queue _after_ removing it from the map. Otherwise our + * "active" checks will be violated. */ + cell_queue_clear(ô->p_chan_cells); + } + + extend_info_free(circ->n_hop); + tor_free(circ->n_chan_create_cell); + + if (circ->global_circuitlist_idx != -1) { + int idx = circ->global_circuitlist_idx; + circuit_t *c2 = smartlist_get(global_circuitlist, idx); + tor_assert(c2 == circ); + smartlist_del(global_circuitlist, idx); + if (idx < smartlist_len(global_circuitlist)) { + c2 = smartlist_get(global_circuitlist, idx); + c2->global_circuitlist_idx = idx; + } + } + + /* Remove from map. */ + circuit_set_n_circid_chan(circ, 0, NULL); + + /* Clear cell queue _after_ removing it from the map. Otherwise our + * "active" checks will be violated. */ + cell_queue_clear(&circ->n_chan_cells); + + log_info(LD_CIRC, "Circuit %u (id: %" PRIu32 ") has been freed.", + n_circ_id, + CIRCUIT_IS_ORIGIN(circ) ? + TO_ORIGIN_CIRCUIT(circ)->global_identifier : 0); + + if (should_free) { + memwipe(mem, 0xAA, memlen); /* poison memory */ + tor_free(mem); + } else { + /* If we made it here, this is an or_circuit_t that still has a pending + * cpuworker request which we weren't able to cancel. Instead, set up + * the magic value so that when the reply comes back, we'll know to discard + * the reply and free this structure. + */ + memwipe(mem, 0xAA, memlen); + circ->magic = DEAD_CIRCUIT_MAGIC; + } +} + +/** Deallocate the linked list circ-><b>cpath</b>, and remove the cpath from + * <b>circ</b>. */ +void +circuit_clear_cpath(origin_circuit_t *circ) +{ + crypt_path_t *victim, *head, *cpath; + + head = cpath = circ->cpath; + + if (!cpath) + return; + + /* it's a circular list, so we have to notice when we've + * gone through it once. */ + while (cpath->next && cpath->next != head) { + victim = cpath; + cpath = victim->next; + circuit_free_cpath_node(victim); + } + + circuit_free_cpath_node(cpath); + + circ->cpath = NULL; +} + +/** Release all storage held by circuits. */ +void +circuit_free_all(void) +{ + smartlist_t *lst = circuit_get_global_list(); + + SMARTLIST_FOREACH_BEGIN(lst, circuit_t *, tmp) { + if (! CIRCUIT_IS_ORIGIN(tmp)) { + or_circuit_t *or_circ = TO_OR_CIRCUIT(tmp); + while (or_circ->resolving_streams) { + edge_connection_t *next_conn; + next_conn = or_circ->resolving_streams->next_stream; + connection_free_(TO_CONN(or_circ->resolving_streams)); + or_circ->resolving_streams = next_conn; + } + } + tmp->global_circuitlist_idx = -1; + circuit_about_to_free_atexit(tmp); + circuit_free(tmp); + SMARTLIST_DEL_CURRENT(lst, tmp); + } SMARTLIST_FOREACH_END(tmp); + + smartlist_free(lst); + global_circuitlist = NULL; + + smartlist_free(global_origin_circuit_list); + global_origin_circuit_list = NULL; + + smartlist_free(circuits_pending_chans); + circuits_pending_chans = NULL; + + smartlist_free(circuits_pending_close); + circuits_pending_close = NULL; + + smartlist_free(circuits_pending_other_guards); + circuits_pending_other_guards = NULL; + + { + chan_circid_circuit_map_t **elt, **next, *c; + for (elt = HT_START(chan_circid_map, &chan_circid_map); + elt; + elt = next) { + c = *elt; + next = HT_NEXT_RMV(chan_circid_map, &chan_circid_map, elt); + + tor_assert(c->circuit == NULL); + tor_free(c); + } + } + HT_CLEAR(chan_circid_map, &chan_circid_map); +} + +/** Deallocate space associated with the cpath node <b>victim</b>. */ +static void +circuit_free_cpath_node(crypt_path_t *victim) +{ + if (!victim) + return; + + relay_crypto_clear(&victim->crypto); + onion_handshake_state_release(&victim->handshake_state); + crypto_dh_free(victim->rend_dh_handshake_state); + extend_info_free(victim->extend_info); + + memwipe(victim, 0xBB, sizeof(crypt_path_t)); /* poison memory */ + tor_free(victim); +} + +/** Release a crypt_path_reference_t*, which may be NULL. */ +static void +cpath_ref_decref(crypt_path_reference_t *cpath_ref) +{ + if (cpath_ref != NULL) { + if (--(cpath_ref->refcount) == 0) { + circuit_free_cpath_node(cpath_ref->cpath); + tor_free(cpath_ref); + } + } +} + +/** A helper function for circuit_dump_by_conn() below. Log a bunch + * of information about circuit <b>circ</b>. + */ +static void +circuit_dump_conn_details(int severity, + circuit_t *circ, + int conn_array_index, + const char *type, + circid_t this_circid, + circid_t other_circid) +{ + tor_log(severity, LD_CIRC, "Conn %d has %s circuit: circID %u " + "(other side %u), state %d (%s), born %ld:", + conn_array_index, type, (unsigned)this_circid, (unsigned)other_circid, + circ->state, circuit_state_to_string(circ->state), + (long)circ->timestamp_began.tv_sec); + if (CIRCUIT_IS_ORIGIN(circ)) { /* circ starts at this node */ + circuit_log_path(severity, LD_CIRC, TO_ORIGIN_CIRCUIT(circ)); + } +} + +/** Log, at severity <b>severity</b>, information about each circuit + * that is connected to <b>conn</b>. + */ +void +circuit_dump_by_conn(connection_t *conn, int severity) +{ + edge_connection_t *tmpconn; + + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, circ) { + circid_t n_circ_id = circ->n_circ_id, p_circ_id = 0; + + if (circ->marked_for_close) { + continue; + } + + if (!CIRCUIT_IS_ORIGIN(circ)) { + p_circ_id = TO_OR_CIRCUIT(circ)->p_circ_id; + } + + if (CIRCUIT_IS_ORIGIN(circ)) { + for (tmpconn=TO_ORIGIN_CIRCUIT(circ)->p_streams; tmpconn; + tmpconn=tmpconn->next_stream) { + if (TO_CONN(tmpconn) == conn) { + circuit_dump_conn_details(severity, circ, conn->conn_array_index, + "App-ward", p_circ_id, n_circ_id); + } + } + } + + if (! CIRCUIT_IS_ORIGIN(circ)) { + for (tmpconn=TO_OR_CIRCUIT(circ)->n_streams; tmpconn; + tmpconn=tmpconn->next_stream) { + if (TO_CONN(tmpconn) == conn) { + circuit_dump_conn_details(severity, circ, conn->conn_array_index, + "Exit-ward", n_circ_id, p_circ_id); + } + } + } + } + SMARTLIST_FOREACH_END(circ); +} + +/** Return the circuit whose global ID is <b>id</b>, or NULL if no + * such circuit exists. */ +origin_circuit_t * +circuit_get_by_global_id(uint32_t id) +{ + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, circ) { + if (CIRCUIT_IS_ORIGIN(circ) && + TO_ORIGIN_CIRCUIT(circ)->global_identifier == id) { + if (circ->marked_for_close) + return NULL; + else + return TO_ORIGIN_CIRCUIT(circ); + } + } + SMARTLIST_FOREACH_END(circ); + return NULL; +} + +/** Return a circ such that: + * - circ->n_circ_id or circ->p_circ_id is equal to <b>circ_id</b>, and + * - circ is attached to <b>chan</b>, either as p_chan or n_chan. + * Return NULL if no such circuit exists. + * + * If <b>found_entry_out</b> is provided, set it to true if we have a + * placeholder entry for circid/chan, and leave it unset otherwise. + */ +static inline circuit_t * +circuit_get_by_circid_channel_impl(circid_t circ_id, channel_t *chan, + int *found_entry_out) +{ + chan_circid_circuit_map_t search; + chan_circid_circuit_map_t *found; + + if (_last_circid_chan_ent && + circ_id == _last_circid_chan_ent->circ_id && + chan == _last_circid_chan_ent->chan) { + found = _last_circid_chan_ent; + } else { + search.circ_id = circ_id; + search.chan = chan; + found = HT_FIND(chan_circid_map, &chan_circid_map, &search); + _last_circid_chan_ent = found; + } + if (found && found->circuit) { + log_debug(LD_CIRC, + "circuit_get_by_circid_channel_impl() returning circuit %p for" + " circ_id %u, channel ID %"PRIu64 " (%p)", + found->circuit, (unsigned)circ_id, + (chan->global_identifier), chan); + if (found_entry_out) + *found_entry_out = 1; + return found->circuit; + } + + log_debug(LD_CIRC, + "circuit_get_by_circid_channel_impl() found %s for" + " circ_id %u, channel ID %"PRIu64 " (%p)", + found ? "placeholder" : "nothing", + (unsigned)circ_id, + (chan->global_identifier), chan); + + if (found_entry_out) + *found_entry_out = found ? 1 : 0; + + return NULL; + /* The rest of this checks for bugs. Disabled by default. */ + /* We comment it out because coverity complains otherwise. + { + circuit_t *circ; + TOR_LIST_FOREACH(circ, &global_circuitlist, head) { + if (! CIRCUIT_IS_ORIGIN(circ)) { + or_circuit_t *or_circ = TO_OR_CIRCUIT(circ); + if (or_circ->p_chan == chan && or_circ->p_circ_id == circ_id) { + log_warn(LD_BUG, + "circuit matches p_chan, but not in hash table (Bug!)"); + return circ; + } + } + if (circ->n_chan == chan && circ->n_circ_id == circ_id) { + log_warn(LD_BUG, + "circuit matches n_chan, but not in hash table (Bug!)"); + return circ; + } + } + return NULL; + } */ +} + +/** Return a circ such that: + * - circ->n_circ_id or circ->p_circ_id is equal to <b>circ_id</b>, and + * - circ is attached to <b>chan</b>, either as p_chan or n_chan. + * - circ is not marked for close. + * Return NULL if no such circuit exists. + */ +circuit_t * +circuit_get_by_circid_channel(circid_t circ_id, channel_t *chan) +{ + circuit_t *circ = circuit_get_by_circid_channel_impl(circ_id, chan, NULL); + if (!circ || circ->marked_for_close) + return NULL; + else + return circ; +} + +/** Return a circ such that: + * - circ->n_circ_id or circ->p_circ_id is equal to <b>circ_id</b>, and + * - circ is attached to <b>chan</b>, either as p_chan or n_chan. + * Return NULL if no such circuit exists. + */ +circuit_t * +circuit_get_by_circid_channel_even_if_marked(circid_t circ_id, + channel_t *chan) +{ + return circuit_get_by_circid_channel_impl(circ_id, chan, NULL); +} + +/** Return true iff the circuit ID <b>circ_id</b> is currently used by a + * circuit, marked or not, on <b>chan</b>, or if the circ ID is reserved until + * a queued destroy cell can be sent. + * + * (Return 1 if the circuit is present, marked or not; Return 2 + * if the circuit ID is pending a destroy.) + **/ +int +circuit_id_in_use_on_channel(circid_t circ_id, channel_t *chan) +{ + int found = 0; + if (circuit_get_by_circid_channel_impl(circ_id, chan, &found) != NULL) + return 1; + if (found) + return 2; + return 0; +} + +/** Helper for debugging 12184. Returns the time since which 'circ_id' has + * been marked unusable on 'chan'. */ +time_t +circuit_id_when_marked_unusable_on_channel(circid_t circ_id, channel_t *chan) +{ + chan_circid_circuit_map_t search; + chan_circid_circuit_map_t *found; + + memset(&search, 0, sizeof(search)); + search.circ_id = circ_id; + search.chan = chan; + + found = HT_FIND(chan_circid_map, &chan_circid_map, &search); + + if (! found || found->circuit) + return 0; + + return found->made_placeholder_at; +} + +/** Return the circuit that a given edge connection is using. */ +circuit_t * +circuit_get_by_edge_conn(edge_connection_t *conn) +{ + circuit_t *circ; + + circ = conn->on_circuit; + tor_assert(!circ || + (CIRCUIT_IS_ORIGIN(circ) ? circ->magic == ORIGIN_CIRCUIT_MAGIC + : circ->magic == OR_CIRCUIT_MAGIC)); + + return circ; +} + +/** For each circuit that has <b>chan</b> as n_chan or p_chan, unlink the + * circuit from the chan,circid map, and mark it for close if it hasn't + * been marked already. + */ +void +circuit_unlink_all_from_channel(channel_t *chan, int reason) +{ + smartlist_t *detached = smartlist_new(); + +/* #define DEBUG_CIRCUIT_UNLINK_ALL */ + + channel_unlink_all_circuits(chan, detached); + +#ifdef DEBUG_CIRCUIT_UNLINK_ALL + { + smartlist_t *detached_2 = smartlist_new(); + int mismatch = 0, badlen = 0; + + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, circ) { + if (circ->n_chan == chan || + (!CIRCUIT_IS_ORIGIN(circ) && + TO_OR_CIRCUIT(circ)->p_chan == chan)) { + smartlist_add(detached_2, circ); + } + } + SMARTLIST_FOREACH_END(circ); + + if (smartlist_len(detached) != smartlist_len(detached_2)) { + log_warn(LD_BUG, "List of detached circuits had the wrong length! " + "(got %d, should have gotten %d)", + (int)smartlist_len(detached), + (int)smartlist_len(detached_2)); + badlen = 1; + } + smartlist_sort_pointers(detached); + smartlist_sort_pointers(detached_2); + + SMARTLIST_FOREACH(detached, circuit_t *, c, + if (c != smartlist_get(detached_2, c_sl_idx)) + mismatch = 1; + ); + + if (mismatch) + log_warn(LD_BUG, "Mismatch in list of detached circuits."); + + if (badlen || mismatch) { + smartlist_free(detached); + detached = detached_2; + } else { + log_notice(LD_CIRC, "List of %d circuits was as expected.", + (int)smartlist_len(detached)); + smartlist_free(detached_2); + } + } +#endif /* defined(DEBUG_CIRCUIT_UNLINK_ALL) */ + + SMARTLIST_FOREACH_BEGIN(detached, circuit_t *, circ) { + int mark = 0; + if (circ->n_chan == chan) { + + circuit_set_n_circid_chan(circ, 0, NULL); + mark = 1; + + /* If we didn't request this closure, pass the remote + * bit to mark_for_close. */ + if (chan->reason_for_closing != CHANNEL_CLOSE_REQUESTED) + reason |= END_CIRC_REASON_FLAG_REMOTE; + } + if (! CIRCUIT_IS_ORIGIN(circ)) { + or_circuit_t *or_circ = TO_OR_CIRCUIT(circ); + if (or_circ->p_chan == chan) { + circuit_set_p_circid_chan(or_circ, 0, NULL); + mark = 1; + } + } + if (!mark) { + log_warn(LD_BUG, "Circuit on detached list which I had no reason " + "to mark"); + continue; + } + if (!circ->marked_for_close) + circuit_mark_for_close(circ, reason); + } SMARTLIST_FOREACH_END(circ); + + smartlist_free(detached); +} + +/** Return a circ such that + * - circ->rend_data->onion_address is equal to + * <b>rend_data</b>->onion_address, + * - circ->rend_data->rend_cookie is equal to + * <b>rend_data</b>->rend_cookie, and + * - circ->purpose is equal to CIRCUIT_PURPOSE_C_REND_READY. + * + * Return NULL if no such circuit exists. + */ +origin_circuit_t * +circuit_get_ready_rend_circ_by_rend_data(const rend_data_t *rend_data) +{ + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, circ) { + if (!circ->marked_for_close && + circ->purpose == CIRCUIT_PURPOSE_C_REND_READY) { + origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); + if (ocirc->rend_data == NULL) { + continue; + } + if (!rend_cmp_service_ids(rend_data_get_address(rend_data), + rend_data_get_address(ocirc->rend_data)) && + tor_memeq(ocirc->rend_data->rend_cookie, + rend_data->rend_cookie, + REND_COOKIE_LEN)) + return ocirc; + } + } + SMARTLIST_FOREACH_END(circ); + return NULL; +} + +/** Return the first service introduction circuit originating from the global + * circuit list after <b>start</b> or at the start of the list if <b>start</b> + * is NULL. Return NULL if no circuit is found. + * + * A service introduction point circuit has a purpose of either + * CIRCUIT_PURPOSE_S_ESTABLISH_INTRO or CIRCUIT_PURPOSE_S_INTRO. This does not + * return a circuit marked for close and its state must be open. */ +origin_circuit_t * +circuit_get_next_service_intro_circ(origin_circuit_t *start) +{ + int idx = 0; + smartlist_t *lst = circuit_get_global_list(); + + if (start) { + idx = TO_CIRCUIT(start)->global_circuitlist_idx + 1; + } + + for ( ; idx < smartlist_len(lst); ++idx) { + circuit_t *circ = smartlist_get(lst, idx); + + /* Ignore a marked for close circuit or purpose not matching a service + * intro point or if the state is not open. */ + if (circ->marked_for_close || circ->state != CIRCUIT_STATE_OPEN || + (circ->purpose != CIRCUIT_PURPOSE_S_ESTABLISH_INTRO && + circ->purpose != CIRCUIT_PURPOSE_S_INTRO)) { + continue; + } + /* The purposes we are looking for are only for origin circuits so the + * following is valid. */ + return TO_ORIGIN_CIRCUIT(circ); + } + /* Not found. */ + return NULL; +} + +/** Return the first service rendezvous circuit originating from the global + * circuit list after <b>start</b> or at the start of the list if <b>start</b> + * is NULL. Return NULL if no circuit is found. + * + * A service rendezvous point circuit has a purpose of either + * CIRCUIT_PURPOSE_S_CONNECT_REND or CIRCUIT_PURPOSE_S_REND_JOINED. This does + * not return a circuit marked for close and its state must be open. */ +origin_circuit_t * +circuit_get_next_service_rp_circ(origin_circuit_t *start) +{ + int idx = 0; + smartlist_t *lst = circuit_get_global_list(); + + if (start) { + idx = TO_CIRCUIT(start)->global_circuitlist_idx + 1; + } + + for ( ; idx < smartlist_len(lst); ++idx) { + circuit_t *circ = smartlist_get(lst, idx); + + /* Ignore a marked for close circuit or purpose not matching a service + * intro point or if the state is not open. */ + if (circ->marked_for_close || circ->state != CIRCUIT_STATE_OPEN || + (circ->purpose != CIRCUIT_PURPOSE_S_CONNECT_REND && + circ->purpose != CIRCUIT_PURPOSE_S_REND_JOINED)) { + continue; + } + /* The purposes we are looking for are only for origin circuits so the + * following is valid. */ + return TO_ORIGIN_CIRCUIT(circ); + } + /* Not found. */ + return NULL; +} + +/** Return the first circuit originating here in global_circuitlist after + * <b>start</b> whose purpose is <b>purpose</b>, and where <b>digest</b> (if + * set) matches the private key digest of the rend data associated with the + * circuit. Return NULL if no circuit is found. If <b>start</b> is NULL, + * begin at the start of the list. + */ +origin_circuit_t * +circuit_get_next_by_pk_and_purpose(origin_circuit_t *start, + const uint8_t *digest, uint8_t purpose) +{ + int idx; + smartlist_t *lst = circuit_get_global_list(); + tor_assert(CIRCUIT_PURPOSE_IS_ORIGIN(purpose)); + if (start == NULL) + idx = 0; + else + idx = TO_CIRCUIT(start)->global_circuitlist_idx + 1; + + for ( ; idx < smartlist_len(lst); ++idx) { + circuit_t *circ = smartlist_get(lst, idx); + origin_circuit_t *ocirc; + + if (circ->marked_for_close) + continue; + if (circ->purpose != purpose) + continue; + /* At this point we should be able to get a valid origin circuit because + * the origin purpose we are looking for matches this circuit. */ + if (BUG(!CIRCUIT_PURPOSE_IS_ORIGIN(circ->purpose))) { + break; + } + ocirc = TO_ORIGIN_CIRCUIT(circ); + if (!digest) + return ocirc; + if (rend_circuit_pk_digest_eq(ocirc, digest)) { + return ocirc; + } + } + return NULL; +} + +/** We might cannibalize this circuit: Return true if its last hop can be used + * as a v3 rendezvous point. */ +static int +circuit_can_be_cannibalized_for_v3_rp(const origin_circuit_t *circ) +{ + if (!circ->build_state) { + return 0; + } + + extend_info_t *chosen_exit = circ->build_state->chosen_exit; + if (BUG(!chosen_exit)) { + return 0; + } + + const node_t *rp_node = node_get_by_id(chosen_exit->identity_digest); + if (rp_node) { + if (node_supports_v3_rendezvous_point(rp_node)) { + return 1; + } + } + + return 0; +} + +/** We are trying to create a circuit of purpose <b>purpose</b> and we are + * looking for cannibalizable circuits. Return the circuit purpose we would be + * willing to cannibalize. */ +static uint8_t +get_circuit_purpose_needed_to_cannibalize(uint8_t purpose) +{ + if (circuit_should_use_vanguards(purpose)) { + /* If we are using vanguards, then we should only cannibalize vanguard + * circuits so that we get the same path construction logic. */ + return CIRCUIT_PURPOSE_HS_VANGUARDS; + } else { + /* If no vanguards are used just get a general circuit! */ + return CIRCUIT_PURPOSE_C_GENERAL; + } +} + +/** Return a circuit that is open, is CIRCUIT_PURPOSE_C_GENERAL, + * has a timestamp_dirty value of 0, has flags matching the CIRCLAUNCH_* + * flags in <b>flags</b>, and if info is defined, does not already use info + * as any of its hops; or NULL if no circuit fits this description. + * + * The <b>purpose</b> argument refers to the purpose of the circuit we want to + * create, not the purpose of the circuit we want to cannibalize. + * + * If !CIRCLAUNCH_NEED_UPTIME, prefer returning non-uptime circuits. + * + * To "cannibalize" a circuit means to extend it an extra hop, and use it + * for some other purpose than we had originally intended. We do this when + * we want to perform some low-bandwidth task at a specific relay, and we + * would like the circuit to complete as soon as possible. (If we were going + * to use a lot of bandwidth, we wouldn't want a circuit with an extra hop. + * If we didn't care about circuit completion latency, we would just build + * a new circuit.) + */ +origin_circuit_t * +circuit_find_to_cannibalize(uint8_t purpose_to_produce, extend_info_t *info, + int flags) +{ + origin_circuit_t *best=NULL; + int need_uptime = (flags & CIRCLAUNCH_NEED_UPTIME) != 0; + int need_capacity = (flags & CIRCLAUNCH_NEED_CAPACITY) != 0; + int internal = (flags & CIRCLAUNCH_IS_INTERNAL) != 0; + const or_options_t *options = get_options(); + /* We want the circuit we are trying to cannibalize to have this purpose */ + int purpose_to_search_for; + + /* Make sure we're not trying to create a onehop circ by + * cannibalization. */ + tor_assert(!(flags & CIRCLAUNCH_ONEHOP_TUNNEL)); + + purpose_to_search_for = get_circuit_purpose_needed_to_cannibalize( + purpose_to_produce); + + tor_assert_nonfatal(purpose_to_search_for == CIRCUIT_PURPOSE_C_GENERAL || + purpose_to_search_for == CIRCUIT_PURPOSE_HS_VANGUARDS); + + log_debug(LD_CIRC, + "Hunting for a circ to cannibalize: purpose %d, uptime %d, " + "capacity %d, internal %d", + purpose_to_produce, need_uptime, need_capacity, internal); + + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, circ_) { + if (CIRCUIT_IS_ORIGIN(circ_) && + circ_->state == CIRCUIT_STATE_OPEN && + !circ_->marked_for_close && + circ_->purpose == purpose_to_search_for && + !circ_->timestamp_dirty) { + origin_circuit_t *circ = TO_ORIGIN_CIRCUIT(circ_); + + /* Only cannibalize from reasonable length circuits. If we + * want C_GENERAL, then only choose 3 hop circs. If we want + * HS_VANGUARDS, only choose 4 hop circs. + */ + if (circ->build_state->desired_path_len != + route_len_for_purpose(purpose_to_search_for, NULL)) { + goto next; + } + + /* Ignore any circuits for which we can't use the Guard. It is possible + * that the Guard was removed from the samepled set after the circuit + * was created so avoid using it. */ + if (!entry_guard_could_succeed(circ->guard_state)) { + goto next; + } + + if ((!need_uptime || circ->build_state->need_uptime) && + (!need_capacity || circ->build_state->need_capacity) && + (internal == circ->build_state->is_internal) && + !circ->unusable_for_new_conns && + circ->remaining_relay_early_cells && + !circ->build_state->onehop_tunnel && + !circ->isolation_values_set) { + if (info) { + /* need to make sure we don't duplicate hops */ + crypt_path_t *hop = circ->cpath; + const node_t *ri1 = node_get_by_id(info->identity_digest); + do { + const node_t *ri2; + if (tor_memeq(hop->extend_info->identity_digest, + info->identity_digest, DIGEST_LEN)) + goto next; + if (ri1 && + (ri2 = node_get_by_id(hop->extend_info->identity_digest)) + && nodes_in_same_family(ri1, ri2)) + goto next; + hop=hop->next; + } while (hop!=circ->cpath); + } + if (options->ExcludeNodes) { + /* Make sure no existing nodes in the circuit are excluded for + * general use. (This may be possible if StrictNodes is 0, and we + * thought we needed to use an otherwise excluded node for, say, a + * directory operation.) */ + crypt_path_t *hop = circ->cpath; + do { + if (routerset_contains_extendinfo(options->ExcludeNodes, + hop->extend_info)) + goto next; + hop = hop->next; + } while (hop != circ->cpath); + } + + if ((flags & CIRCLAUNCH_IS_V3_RP) && + !circuit_can_be_cannibalized_for_v3_rp(circ)) { + log_debug(LD_GENERAL, "Skipping uncannibalizable circuit for v3 " + "rendezvous point."); + goto next; + } + + if (!best || (best->build_state->need_uptime && !need_uptime)) + best = circ; + next: ; + } + } + } + SMARTLIST_FOREACH_END(circ_); + return best; +} + +/** + * Check whether any of the origin circuits that are waiting to see if + * their guard is good enough to use can be upgraded to "ready". If so, + * return a new smartlist containing them. Otherwise return NULL. + */ +smartlist_t * +circuit_find_circuits_to_upgrade_from_guard_wait(void) +{ + /* Only if some circuit is actually waiting on an upgrade should we + * run the algorithm. */ + if (! circuits_pending_other_guards || + smartlist_len(circuits_pending_other_guards)==0) + return NULL; + /* Only if we have some origin circuits should we run the algorithm. */ + if (!global_origin_circuit_list) + return NULL; + + /* Okay; we can pass our circuit list to entrynodes.c.*/ + smartlist_t *result = smartlist_new(); + int circuits_upgraded = entry_guards_upgrade_waiting_circuits( + get_guard_selection_info(), + global_origin_circuit_list, + result); + if (circuits_upgraded && smartlist_len(result)) { + return result; + } else { + smartlist_free(result); + return NULL; + } +} + +/** Return the number of hops in circuit's path. If circ has no entries, + * or is NULL, returns 0. */ +int +circuit_get_cpath_len(origin_circuit_t *circ) +{ + int n = 0; + if (circ && circ->cpath) { + crypt_path_t *cpath, *cpath_next = NULL; + for (cpath = circ->cpath; cpath_next != circ->cpath; cpath = cpath_next) { + cpath_next = cpath->next; + ++n; + } + } + return n; +} + +/** Return the number of opened hops in circuit's path. + * If circ has no entries, or is NULL, returns 0. */ +int +circuit_get_cpath_opened_len(const origin_circuit_t *circ) +{ + int n = 0; + if (circ && circ->cpath) { + crypt_path_t *cpath, *cpath_next = NULL; + for (cpath = circ->cpath; + cpath->state == CPATH_STATE_OPEN + && cpath_next != circ->cpath; + cpath = cpath_next) { + cpath_next = cpath->next; + ++n; + } + } + return n; +} + +/** Return the <b>hopnum</b>th hop in <b>circ</b>->cpath, or NULL if there + * aren't that many hops in the list. <b>hopnum</b> starts at 1. + * Returns NULL if <b>hopnum</b> is 0 or negative. */ +crypt_path_t * +circuit_get_cpath_hop(origin_circuit_t *circ, int hopnum) +{ + if (circ && circ->cpath && hopnum > 0) { + crypt_path_t *cpath, *cpath_next = NULL; + for (cpath = circ->cpath; cpath_next != circ->cpath; cpath = cpath_next) { + cpath_next = cpath->next; + if (--hopnum <= 0) + return cpath; + } + } + return NULL; +} + +/** Go through the circuitlist; mark-for-close each circuit that starts + * at us but has not yet been used. */ +void +circuit_mark_all_unused_circs(void) +{ + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, circ) { + if (CIRCUIT_IS_ORIGIN(circ) && + !circ->marked_for_close && + !circ->timestamp_dirty) + circuit_mark_for_close(circ, END_CIRC_REASON_FINISHED); + } + SMARTLIST_FOREACH_END(circ); +} + +/** Go through the circuitlist; for each circuit that starts at us + * and is dirty, frob its timestamp_dirty so we won't use it for any + * new streams. + * + * This is useful for letting the user change pseudonyms, so new + * streams will not be linkable to old streams. + */ +void +circuit_mark_all_dirty_circs_as_unusable(void) +{ + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, circ) { + if (CIRCUIT_IS_ORIGIN(circ) && + !circ->marked_for_close && + circ->timestamp_dirty) { + mark_circuit_unusable_for_new_conns(TO_ORIGIN_CIRCUIT(circ)); + } + } + SMARTLIST_FOREACH_END(circ); +} + +/** Mark <b>circ</b> to be closed next time we call + * circuit_close_all_marked(). Do any cleanup needed: + * - If state is onionskin_pending, remove circ from the onion_pending + * list. + * - If circ isn't open yet: call circuit_build_failed() if we're + * the origin. + * - If purpose is C_INTRODUCE_ACK_WAIT, report the intro point + * failure we just had to the hidden service client module. + * - If purpose is C_INTRODUCING and <b>reason</b> isn't TIMEOUT, + * report to the hidden service client module that the intro point + * we just tried may be unreachable. + * - Send appropriate destroys and edge_destroys for conns and + * streams attached to circ. + * - If circ->rend_splice is set (we are the midpoint of a joined + * rendezvous stream), then mark the other circuit to close as well. + */ +MOCK_IMPL(void, +circuit_mark_for_close_, (circuit_t *circ, int reason, int line, + const char *file)) +{ + int orig_reason = reason; /* Passed to the controller */ + assert_circuit_ok(circ); + tor_assert(line); + tor_assert(file); + + if (circ->marked_for_close) { + log_warn(LD_BUG, + "Duplicate call to circuit_mark_for_close at %s:%d" + " (first at %s:%d)", file, line, + circ->marked_for_close_file, circ->marked_for_close); + return; + } + if (reason == END_CIRC_AT_ORIGIN) { + if (!CIRCUIT_IS_ORIGIN(circ)) { + log_warn(LD_BUG, "Specified 'at-origin' non-reason for ending circuit, " + "but circuit was not at origin. (called %s:%d, purpose=%d)", + file, line, circ->purpose); + } + reason = END_CIRC_REASON_NONE; + } + + if (CIRCUIT_IS_ORIGIN(circ)) { + if (pathbias_check_close(TO_ORIGIN_CIRCUIT(circ), reason) == -1) { + /* Don't close it yet, we need to test it first */ + return; + } + + /* We don't send reasons when closing circuits at the origin. */ + reason = END_CIRC_REASON_NONE; + } + + if (reason & END_CIRC_REASON_FLAG_REMOTE) + reason &= ~END_CIRC_REASON_FLAG_REMOTE; + + if (reason < END_CIRC_REASON_MIN_ || reason > END_CIRC_REASON_MAX_) { + if (!(orig_reason & END_CIRC_REASON_FLAG_REMOTE)) + log_warn(LD_BUG, "Reason %d out of range at %s:%d", reason, file, line); + reason = END_CIRC_REASON_NONE; + } + + circ->marked_for_close = line; + circ->marked_for_close_file = file; + circ->marked_for_close_reason = reason; + circ->marked_for_close_orig_reason = orig_reason; + + if (!CIRCUIT_IS_ORIGIN(circ)) { + or_circuit_t *or_circ = TO_OR_CIRCUIT(circ); + if (or_circ->rend_splice) { + if (!or_circ->rend_splice->base_.marked_for_close) { + /* do this after marking this circuit, to avoid infinite recursion. */ + circuit_mark_for_close(TO_CIRCUIT(or_circ->rend_splice), reason); + } + or_circ->rend_splice = NULL; + } + } + + /* Notify the HS subsystem that this circuit is closing. */ + hs_circ_cleanup(circ); + + if (circuits_pending_close == NULL) + circuits_pending_close = smartlist_new(); + + smartlist_add(circuits_pending_close, circ); + mainloop_schedule_postloop_cleanup(); + + log_info(LD_GENERAL, "Circuit %u (id: %" PRIu32 ") marked for close at " + "%s:%d (orig reason: %d, new reason: %d)", + circ->n_circ_id, + CIRCUIT_IS_ORIGIN(circ) ? + TO_ORIGIN_CIRCUIT(circ)->global_identifier : 0, + file, line, orig_reason, reason); +} + +/** Called immediately before freeing a marked circuit <b>circ</b> from + * circuit_free_all() while shutting down Tor; this is a safe-at-shutdown + * version of circuit_about_to_free(). It's important that it at least + * do circuitmux_detach_circuit() when appropriate. + */ +static void +circuit_about_to_free_atexit(circuit_t *circ) +{ + + if (circ->n_chan) { + circuit_clear_cell_queue(circ, circ->n_chan); + circuitmux_detach_circuit(circ->n_chan->cmux, circ); + circuit_set_n_circid_chan(circ, 0, NULL); + } + + if (! CIRCUIT_IS_ORIGIN(circ)) { + or_circuit_t *or_circ = TO_OR_CIRCUIT(circ); + + if (or_circ->p_chan) { + circuit_clear_cell_queue(circ, or_circ->p_chan); + circuitmux_detach_circuit(or_circ->p_chan->cmux, circ); + circuit_set_p_circid_chan(or_circ, 0, NULL); + } + } +} + +/** Called immediately before freeing a marked circuit <b>circ</b>. + * Disconnects the circuit from other data structures, launches events + * as appropriate, and performs other housekeeping. + */ +static void +circuit_about_to_free(circuit_t *circ) +{ + + int reason = circ->marked_for_close_reason; + int orig_reason = circ->marked_for_close_orig_reason; + + if (circ->state == CIRCUIT_STATE_ONIONSKIN_PENDING) { + onion_pending_remove(TO_OR_CIRCUIT(circ)); + } + /* If the circuit ever became OPEN, we sent it to the reputation history + * module then. If it isn't OPEN, we send it there now to remember which + * links worked and which didn't. + */ + if (circ->state != CIRCUIT_STATE_OPEN && + circ->state != CIRCUIT_STATE_GUARD_WAIT) { + if (CIRCUIT_IS_ORIGIN(circ)) { + origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); + circuit_build_failed(ocirc); /* take actions if necessary */ + } + } + if (circ->state == CIRCUIT_STATE_CHAN_WAIT) { + if (circuits_pending_chans) + smartlist_remove(circuits_pending_chans, circ); + } + if (circuits_pending_other_guards) { + smartlist_remove(circuits_pending_other_guards, circ); + } + if (CIRCUIT_IS_ORIGIN(circ)) { + control_event_circuit_status(TO_ORIGIN_CIRCUIT(circ), + (circ->state == CIRCUIT_STATE_OPEN || + circ->state == CIRCUIT_STATE_GUARD_WAIT) ? + CIRC_EVENT_CLOSED:CIRC_EVENT_FAILED, + orig_reason); + } + + if (circ->purpose == CIRCUIT_PURPOSE_C_INTRODUCE_ACK_WAIT) { + origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); + int timed_out = (reason == END_CIRC_REASON_TIMEOUT); + tor_assert(circ->state == CIRCUIT_STATE_OPEN); + tor_assert(ocirc->build_state->chosen_exit); + if (orig_reason != END_CIRC_REASON_IP_NOW_REDUNDANT && + ocirc->rend_data) { + /* treat this like getting a nack from it */ + log_info(LD_REND, "Failed intro circ %s to %s (awaiting ack). %s", + safe_str_client(rend_data_get_address(ocirc->rend_data)), + safe_str_client(build_state_get_exit_nickname(ocirc->build_state)), + timed_out ? "Recording timeout." : "Removing from descriptor."); + rend_client_report_intro_point_failure(ocirc->build_state->chosen_exit, + ocirc->rend_data, + timed_out ? + INTRO_POINT_FAILURE_TIMEOUT : + INTRO_POINT_FAILURE_GENERIC); + } + } else if (circ->purpose == CIRCUIT_PURPOSE_C_INTRODUCING && + reason != END_CIRC_REASON_TIMEOUT) { + origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); + if (ocirc->build_state->chosen_exit && ocirc->rend_data) { + if (orig_reason != END_CIRC_REASON_IP_NOW_REDUNDANT && + ocirc->rend_data) { + log_info(LD_REND, "Failed intro circ %s to %s " + "(building circuit to intro point). " + "Marking intro point as possibly unreachable.", + safe_str_client(rend_data_get_address(ocirc->rend_data)), + safe_str_client(build_state_get_exit_nickname( + ocirc->build_state))); + rend_client_report_intro_point_failure(ocirc->build_state->chosen_exit, + ocirc->rend_data, + INTRO_POINT_FAILURE_UNREACHABLE); + } + } + } + + if (circ->n_chan) { + circuit_clear_cell_queue(circ, circ->n_chan); + /* Only send destroy if the channel isn't closing anyway */ + if (!CHANNEL_CONDEMNED(circ->n_chan)) { + channel_send_destroy(circ->n_circ_id, circ->n_chan, reason); + } + circuitmux_detach_circuit(circ->n_chan->cmux, circ); + circuit_set_n_circid_chan(circ, 0, NULL); + } + + if (! CIRCUIT_IS_ORIGIN(circ)) { + or_circuit_t *or_circ = TO_OR_CIRCUIT(circ); + edge_connection_t *conn; + for (conn=or_circ->n_streams; conn; conn=conn->next_stream) + connection_edge_destroy(or_circ->p_circ_id, conn); + or_circ->n_streams = NULL; + + while (or_circ->resolving_streams) { + conn = or_circ->resolving_streams; + or_circ->resolving_streams = conn->next_stream; + if (!conn->base_.marked_for_close) { + /* The client will see a DESTROY, and infer that the connections + * are closing because the circuit is getting torn down. No need + * to send an end cell. */ + conn->edge_has_sent_end = 1; + conn->end_reason = END_STREAM_REASON_DESTROY; + conn->end_reason |= END_STREAM_REASON_FLAG_ALREADY_SENT_CLOSED; + connection_mark_for_close(TO_CONN(conn)); + } + conn->on_circuit = NULL; + } + + if (or_circ->p_chan) { + circuit_clear_cell_queue(circ, or_circ->p_chan); + /* Only send destroy if the channel isn't closing anyway */ + if (!CHANNEL_CONDEMNED(or_circ->p_chan)) { + channel_send_destroy(or_circ->p_circ_id, or_circ->p_chan, reason); + } + circuitmux_detach_circuit(or_circ->p_chan->cmux, circ); + circuit_set_p_circid_chan(or_circ, 0, NULL); + } + } else { + origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); + edge_connection_t *conn; + for (conn=ocirc->p_streams; conn; conn=conn->next_stream) + connection_edge_destroy(circ->n_circ_id, conn); + ocirc->p_streams = NULL; + } +} + +/** Given a marked circuit <b>circ</b>, aggressively free its cell queues to + * recover memory. */ +static void +marked_circuit_free_cells(circuit_t *circ) +{ + if (!circ->marked_for_close) { + log_warn(LD_BUG, "Called on non-marked circuit"); + return; + } + cell_queue_clear(&circ->n_chan_cells); + if (circ->n_mux) + circuitmux_clear_num_cells(circ->n_mux, circ); + if (! CIRCUIT_IS_ORIGIN(circ)) { + or_circuit_t *orcirc = TO_OR_CIRCUIT(circ); + cell_queue_clear(&orcirc->p_chan_cells); + if (orcirc->p_mux) + circuitmux_clear_num_cells(orcirc->p_mux, circ); + } +} + +static size_t +single_conn_free_bytes(connection_t *conn) +{ + size_t result = 0; + if (conn->inbuf) { + result += buf_allocation(conn->inbuf); + buf_clear(conn->inbuf); + } + if (conn->outbuf) { + result += buf_allocation(conn->outbuf); + buf_clear(conn->outbuf); + conn->outbuf_flushlen = 0; + } + if (conn->type == CONN_TYPE_DIR) { + dir_connection_t *dir_conn = TO_DIR_CONN(conn); + if (dir_conn->compress_state) { + result += tor_compress_state_size(dir_conn->compress_state); + tor_compress_free(dir_conn->compress_state); + dir_conn->compress_state = NULL; + } + } + return result; +} + +/** Aggressively free buffer contents on all the buffers of all streams in the + * list starting at <b>stream</b>. Return the number of bytes recovered. */ +static size_t +marked_circuit_streams_free_bytes(edge_connection_t *stream) +{ + size_t result = 0; + for ( ; stream; stream = stream->next_stream) { + connection_t *conn = TO_CONN(stream); + result += single_conn_free_bytes(conn); + if (conn->linked_conn) { + result += single_conn_free_bytes(conn->linked_conn); + } + } + return result; +} + +/** Aggressively free buffer contents on all the buffers of all streams on + * circuit <b>c</b>. Return the number of bytes recovered. */ +static size_t +marked_circuit_free_stream_bytes(circuit_t *c) +{ + if (CIRCUIT_IS_ORIGIN(c)) { + return marked_circuit_streams_free_bytes(TO_ORIGIN_CIRCUIT(c)->p_streams); + } else { + return marked_circuit_streams_free_bytes(TO_OR_CIRCUIT(c)->n_streams); + } +} + +/** Return the number of cells used by the circuit <b>c</b>'s cell queues. */ +STATIC size_t +n_cells_in_circ_queues(const circuit_t *c) +{ + size_t n = c->n_chan_cells.n; + if (! CIRCUIT_IS_ORIGIN(c)) { + circuit_t *cc = (circuit_t *) c; + n += TO_OR_CIRCUIT(cc)->p_chan_cells.n; + } + return n; +} + +/** + * Return the age of the oldest cell queued on <b>c</b>, in timestamp units. + * Return 0 if there are no cells queued on c. Requires that <b>now</b> be + * the current coarse timestamp. + * + * This function will return incorrect results if the oldest cell queued on + * the circuit is older than about 2**32 msec (about 49 days) old. + */ +STATIC uint32_t +circuit_max_queued_cell_age(const circuit_t *c, uint32_t now) +{ + uint32_t age = 0; + packed_cell_t *cell; + + if (NULL != (cell = TOR_SIMPLEQ_FIRST(&c->n_chan_cells.head))) + age = now - cell->inserted_timestamp; + + if (! CIRCUIT_IS_ORIGIN(c)) { + const or_circuit_t *orcirc = CONST_TO_OR_CIRCUIT(c); + if (NULL != (cell = TOR_SIMPLEQ_FIRST(&orcirc->p_chan_cells.head))) { + uint32_t age2 = now - cell->inserted_timestamp; + if (age2 > age) + return age2; + } + } + return age; +} + +/** Return the age of the oldest buffer chunk on <b>conn</b>, where age is + * taken in timestamp units before the time <b>now</b>. If the connection has + * no data, treat it as having age zero. + **/ +static uint32_t +conn_get_buffer_age(const connection_t *conn, uint32_t now_ts) +{ + uint32_t age = 0, age2; + if (conn->outbuf) { + age2 = buf_get_oldest_chunk_timestamp(conn->outbuf, now_ts); + if (age2 > age) + age = age2; + } + if (conn->inbuf) { + age2 = buf_get_oldest_chunk_timestamp(conn->inbuf, now_ts); + if (age2 > age) + age = age2; + } + return age; +} + +/** Return the age in timestamp units of the oldest buffer chunk on any stream + * in the linked list <b>stream</b>, where age is taken in timestamp units + * before the timestamp <b>now</b>. */ +static uint32_t +circuit_get_streams_max_data_age(const edge_connection_t *stream, uint32_t now) +{ + uint32_t age = 0, age2; + for (; stream; stream = stream->next_stream) { + const connection_t *conn = TO_CONN(stream); + age2 = conn_get_buffer_age(conn, now); + if (age2 > age) + age = age2; + if (conn->linked_conn) { + age2 = conn_get_buffer_age(conn->linked_conn, now); + if (age2 > age) + age = age2; + } + } + return age; +} + +/** Return the age in timestamp units of the oldest buffer chunk on any stream + * attached to the circuit <b>c</b>, where age is taken before the timestamp + * <b>now</b>. */ +STATIC uint32_t +circuit_max_queued_data_age(const circuit_t *c, uint32_t now) +{ + if (CIRCUIT_IS_ORIGIN(c)) { + return circuit_get_streams_max_data_age( + CONST_TO_ORIGIN_CIRCUIT(c)->p_streams, now); + } else { + return circuit_get_streams_max_data_age( + CONST_TO_OR_CIRCUIT(c)->n_streams, now); + } +} + +/** Return the age of the oldest cell or stream buffer chunk on the circuit + * <b>c</b>, where age is taken in timestamp units before the timestamp + * <b>now</b> */ +STATIC uint32_t +circuit_max_queued_item_age(const circuit_t *c, uint32_t now) +{ + uint32_t cell_age = circuit_max_queued_cell_age(c, now); + uint32_t data_age = circuit_max_queued_data_age(c, now); + if (cell_age > data_age) + return cell_age; + else + return data_age; +} + +/** Helper to sort a list of circuit_t by age of oldest item, in descending + * order. */ +static int +circuits_compare_by_oldest_queued_item_(const void **a_, const void **b_) +{ + const circuit_t *a = *a_; + const circuit_t *b = *b_; + uint32_t age_a = a->age_tmp; + uint32_t age_b = b->age_tmp; + + if (age_a < age_b) + return 1; + else if (age_a == age_b) + return 0; + else + return -1; +} + +static uint32_t now_ts_for_buf_cmp; + +/** Helper to sort a list of circuit_t by age of oldest item, in descending + * order. */ +static int +conns_compare_by_buffer_age_(const void **a_, const void **b_) +{ + const connection_t *a = *a_; + const connection_t *b = *b_; + time_t age_a = conn_get_buffer_age(a, now_ts_for_buf_cmp); + time_t age_b = conn_get_buffer_age(b, now_ts_for_buf_cmp); + + if (age_a < age_b) + return 1; + else if (age_a == age_b) + return 0; + else + return -1; +} + +#define FRACTION_OF_DATA_TO_RETAIN_ON_OOM 0.90 + +/** We're out of memory for cells, having allocated <b>current_allocation</b> + * bytes' worth. Kill the 'worst' circuits until we're under + * FRACTION_OF_DATA_TO_RETAIN_ON_OOM of our maximum usage. */ +void +circuits_handle_oom(size_t current_allocation) +{ + smartlist_t *circlist; + smartlist_t *connection_array = get_connection_array(); + int conn_idx; + size_t mem_to_recover; + size_t mem_recovered=0; + int n_circuits_killed=0; + int n_dirconns_killed=0; + uint32_t now_ts; + log_notice(LD_GENERAL, "We're low on memory (cell queues total alloc:" + " %"TOR_PRIuSZ" buffer total alloc: %" TOR_PRIuSZ "," + " tor compress total alloc: %" TOR_PRIuSZ + " (zlib: %" TOR_PRIuSZ ", zstd: %" TOR_PRIuSZ "," + " lzma: %" TOR_PRIuSZ ")," + " rendezvous cache total alloc: %" TOR_PRIuSZ "). Killing" + " circuits withover-long queues. (This behavior is controlled by" + " MaxMemInQueues.)", + cell_queues_get_total_allocation(), + buf_get_total_allocation(), + tor_compress_get_total_allocation(), + tor_zlib_get_total_allocation(), + tor_zstd_get_total_allocation(), + tor_lzma_get_total_allocation(), + rend_cache_get_total_allocation()); + + { + size_t mem_target = (size_t)(get_options()->MaxMemInQueues * + FRACTION_OF_DATA_TO_RETAIN_ON_OOM); + if (current_allocation <= mem_target) + return; + mem_to_recover = current_allocation - mem_target; + } + + now_ts = monotime_coarse_get_stamp(); + + circlist = circuit_get_global_list(); + SMARTLIST_FOREACH_BEGIN(circlist, circuit_t *, circ) { + circ->age_tmp = circuit_max_queued_item_age(circ, now_ts); + } SMARTLIST_FOREACH_END(circ); + + /* This is O(n log n); there are faster algorithms we could use instead. + * Let's hope this doesn't happen enough to be in the critical path. */ + smartlist_sort(circlist, circuits_compare_by_oldest_queued_item_); + + /* Fix up the indices before we run into trouble */ + SMARTLIST_FOREACH_BEGIN(circlist, circuit_t *, circ) { + circ->global_circuitlist_idx = circ_sl_idx; + } SMARTLIST_FOREACH_END(circ); + + /* Now sort the connection array ... */ + now_ts_for_buf_cmp = now_ts; + smartlist_sort(connection_array, conns_compare_by_buffer_age_); + now_ts_for_buf_cmp = 0; + + /* Fix up the connection array to its new order. */ + SMARTLIST_FOREACH_BEGIN(connection_array, connection_t *, conn) { + conn->conn_array_index = conn_sl_idx; + } SMARTLIST_FOREACH_END(conn); + + /* Okay, now the worst circuits and connections are at the front of their + * respective lists. Let's mark them, and reclaim their storage + * aggressively. */ + conn_idx = 0; + SMARTLIST_FOREACH_BEGIN(circlist, circuit_t *, circ) { + size_t n; + size_t freed; + + /* Free storage in any non-linked directory connections that have buffered + * data older than this circuit. */ + while (conn_idx < smartlist_len(connection_array)) { + connection_t *conn = smartlist_get(connection_array, conn_idx); + uint32_t conn_age = conn_get_buffer_age(conn, now_ts); + if (conn_age < circ->age_tmp) { + break; + } + if (conn->type == CONN_TYPE_DIR && conn->linked_conn == NULL) { + if (!conn->marked_for_close) + connection_mark_for_close(conn); + mem_recovered += single_conn_free_bytes(conn); + + ++n_dirconns_killed; + + if (mem_recovered >= mem_to_recover) + goto done_recovering_mem; + } + ++conn_idx; + } + + /* Now, kill the circuit. */ + n = n_cells_in_circ_queues(circ); + if (! circ->marked_for_close) { + circuit_mark_for_close(circ, END_CIRC_REASON_RESOURCELIMIT); + } + marked_circuit_free_cells(circ); + freed = marked_circuit_free_stream_bytes(circ); + + ++n_circuits_killed; + + mem_recovered += n * packed_cell_mem_cost(); + mem_recovered += freed; + + if (mem_recovered >= mem_to_recover) + goto done_recovering_mem; + } SMARTLIST_FOREACH_END(circ); + + done_recovering_mem: + + log_notice(LD_GENERAL, "Removed %"TOR_PRIuSZ" bytes by killing %d circuits; " + "%d circuits remain alive. Also killed %d non-linked directory " + "connections.", + mem_recovered, + n_circuits_killed, + smartlist_len(circlist) - n_circuits_killed, + n_dirconns_killed); +} + +/** Verify that cpath layer <b>cp</b> has all of its invariants + * correct. Trigger an assert if anything is invalid. + */ +void +assert_cpath_layer_ok(const crypt_path_t *cp) +{ +// tor_assert(cp->addr); /* these are zero for rendezvous extra-hops */ +// tor_assert(cp->port); + tor_assert(cp); + tor_assert(cp->magic == CRYPT_PATH_MAGIC); + switch (cp->state) + { + case CPATH_STATE_OPEN: + relay_crypto_assert_ok(&cp->crypto); + /* fall through */ + case CPATH_STATE_CLOSED: + /*XXXX Assert that there's no handshake_state either. */ + tor_assert(!cp->rend_dh_handshake_state); + break; + case CPATH_STATE_AWAITING_KEYS: + /* tor_assert(cp->dh_handshake_state); */ + break; + default: + log_fn(LOG_ERR, LD_BUG, "Unexpected state %d", cp->state); + tor_assert(0); + } + tor_assert(cp->package_window >= 0); + tor_assert(cp->deliver_window >= 0); +} + +/** Verify that cpath <b>cp</b> has all of its invariants + * correct. Trigger an assert if anything is invalid. + */ +static void +assert_cpath_ok(const crypt_path_t *cp) +{ + const crypt_path_t *start = cp; + + do { + assert_cpath_layer_ok(cp); + /* layers must be in sequence of: "open* awaiting? closed*" */ + if (cp != start) { + if (cp->state == CPATH_STATE_AWAITING_KEYS) { + tor_assert(cp->prev->state == CPATH_STATE_OPEN); + } else if (cp->state == CPATH_STATE_OPEN) { + tor_assert(cp->prev->state == CPATH_STATE_OPEN); + } + } + cp = cp->next; + tor_assert(cp); + } while (cp != start); +} + +/** Verify that circuit <b>c</b> has all of its invariants + * correct. Trigger an assert if anything is invalid. + */ +MOCK_IMPL(void, +assert_circuit_ok,(const circuit_t *c)) +{ + edge_connection_t *conn; + const or_circuit_t *or_circ = NULL; + const origin_circuit_t *origin_circ = NULL; + + tor_assert(c); + tor_assert(c->magic == ORIGIN_CIRCUIT_MAGIC || c->magic == OR_CIRCUIT_MAGIC); + tor_assert(c->purpose >= CIRCUIT_PURPOSE_MIN_ && + c->purpose <= CIRCUIT_PURPOSE_MAX_); + + if (CIRCUIT_IS_ORIGIN(c)) + origin_circ = CONST_TO_ORIGIN_CIRCUIT(c); + else + or_circ = CONST_TO_OR_CIRCUIT(c); + + if (c->n_chan) { + tor_assert(!c->n_hop); + + if (c->n_circ_id) { + /* We use the _impl variant here to make sure we don't fail on marked + * circuits, which would not be returned by the regular function. */ + circuit_t *c2 = circuit_get_by_circid_channel_impl(c->n_circ_id, + c->n_chan, NULL); + tor_assert(c == c2); + } + } + if (or_circ && or_circ->p_chan) { + if (or_circ->p_circ_id) { + /* ibid */ + circuit_t *c2 = + circuit_get_by_circid_channel_impl(or_circ->p_circ_id, + or_circ->p_chan, NULL); + tor_assert(c == c2); + } + } + if (or_circ) + for (conn = or_circ->n_streams; conn; conn = conn->next_stream) + tor_assert(conn->base_.type == CONN_TYPE_EXIT); + + tor_assert(c->deliver_window >= 0); + tor_assert(c->package_window >= 0); + if (c->state == CIRCUIT_STATE_OPEN || + c->state == CIRCUIT_STATE_GUARD_WAIT) { + tor_assert(!c->n_chan_create_cell); + if (or_circ) { + relay_crypto_assert_ok(&or_circ->crypto); + } + } + if (c->state == CIRCUIT_STATE_CHAN_WAIT && !c->marked_for_close) { + tor_assert(circuits_pending_chans && + smartlist_contains(circuits_pending_chans, c)); + } else { + tor_assert(!circuits_pending_chans || + !smartlist_contains(circuits_pending_chans, c)); + } + if (origin_circ && origin_circ->cpath) { + assert_cpath_ok(origin_circ->cpath); + } + if (c->purpose == CIRCUIT_PURPOSE_REND_ESTABLISHED) { + tor_assert(or_circ); + if (!c->marked_for_close) { + tor_assert(or_circ->rend_splice); + tor_assert(or_circ->rend_splice->rend_splice == or_circ); + } + tor_assert(or_circ->rend_splice != or_circ); + } else { + tor_assert(!or_circ || !or_circ->rend_splice); + } +} diff --cc src/core/or/connection_edge.c index ab3c14d2c,000000000..2cc54f774 mode 100644,000000..100644 --- a/src/core/or/connection_edge.c +++ b/src/core/or/connection_edge.c @@@ -1,4219 -1,0 +1,4445 @@@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file connection_edge.c + * \brief Handle edge streams. + * + * An edge_connection_t is a subtype of a connection_t, and represents two + * critical concepts in Tor: a stream, and an edge connection. From the Tor + * protocol's point of view, a stream is a bi-directional channel that is + * multiplexed on a single circuit. Each stream on a circuit is identified + * with a separate 16-bit stream ID, local to the (circuit,exit) pair. + * Streams are created in response to client requests. + * + * An edge connection is one thing that can implement a stream: it is either a + * TCP application socket that has arrived via (e.g.) a SOCKS request, or an + * exit connection. + * + * Not every instance of edge_connection_t truly represents an edge connction, + * however. (Sorry!) We also create edge_connection_t objects for streams that + * we will not be handling with TCP. The types of these streams are: + * <ul> + * <li>DNS lookup streams, created on the client side in response to + * a UDP DNS request received on a DNSPort, or a RESOLVE command + * on a controller. + * <li>DNS lookup streams, created on the exit side in response to + * a RELAY_RESOLVE cell from a client. + * <li>Tunneled directory streams, created on the directory cache side + * in response to a RELAY_BEGIN_DIR cell. These streams attach directly + * to a dir_connection_t object without ever using TCP. + * </ul> + * + * This module handles general-purpose functionality having to do with + * edge_connection_t. On the client side, it accepts various types of + * application requests on SocksPorts, TransPorts, and NATDPorts, and + * creates streams appropriately. + * + * This module is also responsible for implementing stream isolation: + * ensuring that streams that should not be linkable to one another are + * kept to different circuits. + * + * On the exit side, this module handles the various stream-creating + * type of RELAY cells by launching appropriate outgoing connections, + * DNS requests, or directory connection objects. + * + * And for all edge connections, this module is responsible for handling + * incoming and outdoing data as it arrives or leaves in the relay.c + * module. (Outgoing data will be packaged in + * connection_edge_process_inbuf() as it calls + * connection_edge_package_raw_inbuf(); incoming data from RELAY_DATA + * cells is applied in connection_edge_process_relay_cell().) + **/ +#define CONNECTION_EDGE_PRIVATE + +#include "core/or/or.h" + +#include "lib/err/backtrace.h" + +#include "feature/client/addressmap.h" +#include "lib/container/buffers.h" +#include "core/or/channel.h" +#include "feature/client/circpathbias.h" +#include "core/or/circuitlist.h" +#include "core/or/circuituse.h" +#include "app/config/config.h" +#include "core/mainloop/connection.h" +#include "core/or/connection_edge.h" +#include "core/or/connection_or.h" +#include "feature/control/control.h" +#include "lib/crypt_ops/crypto_util.h" +#include "feature/relay/dns.h" +#include "feature/client/dnsserv.h" +#include "feature/dircache/directory.h" +#include "feature/dircache/dirserv.h" +#include "feature/hibernate/hibernate.h" +#include "feature/hs/hs_common.h" +#include "feature/hs/hs_cache.h" +#include "feature/hs/hs_client.h" +#include "feature/hs/hs_circuit.h" +#include "core/mainloop/main.h" +#include "feature/nodelist/networkstatus.h" +#include "feature/nodelist/nodelist.h" +#include "core/or/policies.h" +#include "core/proto/proto_http.h" +#include "core/proto/proto_socks.h" +#include "core/or/reasons.h" +#include "core/or/relay.h" +#include "feature/rend/rendclient.h" +#include "feature/rend/rendcommon.h" +#include "feature/rend/rendservice.h" +#include "feature/stats/rephist.h" +#include "feature/relay/router.h" +#include "feature/nodelist/routerlist.h" +#include "feature/nodelist/routerset.h" +#include "core/or/circuitbuild.h" + +#include "core/or/cell_st.h" +#include "core/or/cpath_build_state_st.h" +#include "feature/dircommon/dir_connection_st.h" +#include "core/or/entry_connection_st.h" +#include "core/or/extend_info_st.h" +#include "feature/nodelist/node_st.h" +#include "core/or/or_circuit_st.h" +#include "core/or/origin_circuit_st.h" ++#include "core/or/half_edge_st.h" +#include "core/or/socks_request_st.h" +#include "lib/evloop/compat_libevent.h" + +#ifdef HAVE_LINUX_TYPES_H +#include <linux/types.h> +#endif +#ifdef HAVE_LINUX_NETFILTER_IPV4_H +#include <linux/netfilter_ipv4.h> +#define TRANS_NETFILTER +#define TRANS_NETFILTER_IPV4 +#endif + +#ifdef HAVE_LINUX_IF_H +#include <linux/if.h> +#endif + +#ifdef HAVE_LINUX_NETFILTER_IPV6_IP6_TABLES_H +#include <linux/netfilter_ipv6/ip6_tables.h> +#if defined(IP6T_SO_ORIGINAL_DST) +#define TRANS_NETFILTER +#define TRANS_NETFILTER_IPV6 +#endif +#endif /* defined(HAVE_LINUX_NETFILTER_IPV6_IP6_TABLES_H) */ + +#ifdef HAVE_FCNTL_H +#include <fcntl.h> +#endif +#ifdef HAVE_SYS_IOCTL_H +#include <sys/ioctl.h> +#endif + +#if defined(HAVE_NET_IF_H) && defined(HAVE_NET_PFVAR_H) +#include <net/if.h> +#include <net/pfvar.h> +#define TRANS_PF +#endif + +#ifdef IP_TRANSPARENT +#define TRANS_TPROXY +#endif + +#define SOCKS4_GRANTED 90 +#define SOCKS4_REJECT 91 + +static int connection_ap_handshake_process_socks(entry_connection_t *conn); +static int connection_ap_process_natd(entry_connection_t *conn); +static int connection_exit_connect_dir(edge_connection_t *exitconn); +static int consider_plaintext_ports(entry_connection_t *conn, uint16_t port); +static int connection_ap_supports_optimistic_data(const entry_connection_t *); ++STATIC void connection_half_edge_add(const edge_connection_t *conn, ++ origin_circuit_t *circ); ++STATIC half_edge_t *connection_half_edge_find_stream_id( ++ const smartlist_t *half_conns, ++ streamid_t stream_id); + +/** Convert a connection_t* to an edge_connection_t*; assert if the cast is + * invalid. */ +edge_connection_t * +TO_EDGE_CONN(connection_t *c) +{ + tor_assert(c->magic == EDGE_CONNECTION_MAGIC || + c->magic == ENTRY_CONNECTION_MAGIC); + return DOWNCAST(edge_connection_t, c); +} + +entry_connection_t * +TO_ENTRY_CONN(connection_t *c) +{ + tor_assert(c->magic == ENTRY_CONNECTION_MAGIC); + return (entry_connection_t*) SUBTYPE_P(c, entry_connection_t, edge_.base_); +} + +entry_connection_t * +EDGE_TO_ENTRY_CONN(edge_connection_t *c) +{ + tor_assert(c->base_.magic == ENTRY_CONNECTION_MAGIC); + return (entry_connection_t*) SUBTYPE_P(c, entry_connection_t, edge_); +} + +/** An AP stream has failed/finished. If it hasn't already sent back + * a socks reply, send one now (based on endreason). Also set + * has_sent_end to 1, and mark the conn. + */ +MOCK_IMPL(void, +connection_mark_unattached_ap_,(entry_connection_t *conn, int endreason, + int line, const char *file)) +{ + connection_t *base_conn = ENTRY_TO_CONN(conn); + edge_connection_t *edge_conn = ENTRY_TO_EDGE_CONN(conn); + tor_assert(base_conn->type == CONN_TYPE_AP); + ENTRY_TO_EDGE_CONN(conn)->edge_has_sent_end = 1; /* no circ yet */ + + /* If this is a rendezvous stream and it is failing without ever + * being attached to a circuit, assume that an attempt to connect to + * the destination hidden service has just ended. + * + * XXXX This condition doesn't limit to only streams failing + * without ever being attached. That sloppiness should be harmless, + * but we should fix it someday anyway. */ + if ((edge_conn->on_circuit != NULL || edge_conn->edge_has_sent_end) && + connection_edge_is_rendezvous_stream(edge_conn)) { + if (edge_conn->rend_data) { + rend_client_note_connection_attempt_ended(edge_conn->rend_data); + } + } + + if (base_conn->marked_for_close) { + /* This call will warn as appropriate. */ + connection_mark_for_close_(base_conn, line, file); + return; + } + + if (!conn->socks_request->has_finished) { + if (endreason & END_STREAM_REASON_FLAG_ALREADY_SOCKS_REPLIED) + log_warn(LD_BUG, + "stream (marked at %s:%d) sending two socks replies?", + file, line); + + if (SOCKS_COMMAND_IS_CONNECT(conn->socks_request->command)) + connection_ap_handshake_socks_reply(conn, NULL, 0, endreason); + else if (SOCKS_COMMAND_IS_RESOLVE(conn->socks_request->command)) + connection_ap_handshake_socks_resolved(conn, + RESOLVED_TYPE_ERROR_TRANSIENT, + 0, NULL, -1, -1); + else /* unknown or no handshake at all. send no response. */ + conn->socks_request->has_finished = 1; + } + + connection_mark_and_flush_(base_conn, line, file); + + ENTRY_TO_EDGE_CONN(conn)->end_reason = endreason; +} + +/** There was an EOF. Send an end and mark the connection for close. + */ +int +connection_edge_reached_eof(edge_connection_t *conn) +{ + if (connection_get_inbuf_len(TO_CONN(conn)) && + connection_state_is_open(TO_CONN(conn))) { + /* it still has stuff to process. don't let it die yet. */ + return 0; + } + log_info(LD_EDGE,"conn (fd "TOR_SOCKET_T_FORMAT") reached eof. Closing.", + conn->base_.s); + if (!conn->base_.marked_for_close) { + /* only mark it if not already marked. it's possible to + * get the 'end' right around when the client hangs up on us. */ + connection_edge_end(conn, END_STREAM_REASON_DONE); + if (conn->base_.type == CONN_TYPE_AP) { + /* eof, so don't send a socks reply back */ + if (EDGE_TO_ENTRY_CONN(conn)->socks_request) + EDGE_TO_ENTRY_CONN(conn)->socks_request->has_finished = 1; + } + connection_mark_for_close(TO_CONN(conn)); + } + return 0; +} + +/** Handle new bytes on conn->inbuf based on state: + * - If it's waiting for socks info, try to read another step of the + * socks handshake out of conn->inbuf. + * - If it's waiting for the original destination, fetch it. + * - If it's open, then package more relay cells from the stream. + * - Else, leave the bytes on inbuf alone for now. + * + * Mark and return -1 if there was an unexpected error with the conn, + * else return 0. + */ +int +connection_edge_process_inbuf(edge_connection_t *conn, int package_partial) +{ + tor_assert(conn); + + switch (conn->base_.state) { + case AP_CONN_STATE_SOCKS_WAIT: + if (connection_ap_handshake_process_socks(EDGE_TO_ENTRY_CONN(conn)) <0) { + /* already marked */ + return -1; + } + return 0; + case AP_CONN_STATE_NATD_WAIT: + if (connection_ap_process_natd(EDGE_TO_ENTRY_CONN(conn)) < 0) { + /* already marked */ + return -1; + } + return 0; + case AP_CONN_STATE_HTTP_CONNECT_WAIT: + if (connection_ap_process_http_connect(EDGE_TO_ENTRY_CONN(conn)) < 0) { + return -1; + } + return 0; + case AP_CONN_STATE_OPEN: + case EXIT_CONN_STATE_OPEN: + if (connection_edge_package_raw_inbuf(conn, package_partial, NULL) < 0) { + /* (We already sent an end cell if possible) */ + connection_mark_for_close(TO_CONN(conn)); + return -1; + } + return 0; + case AP_CONN_STATE_CONNECT_WAIT: + if (connection_ap_supports_optimistic_data(EDGE_TO_ENTRY_CONN(conn))) { + log_info(LD_EDGE, + "data from edge while in '%s' state. Sending it anyway. " + "package_partial=%d, buflen=%ld", + conn_state_to_string(conn->base_.type, conn->base_.state), + package_partial, + (long)connection_get_inbuf_len(TO_CONN(conn))); + if (connection_edge_package_raw_inbuf(conn, package_partial, NULL)<0) { + /* (We already sent an end cell if possible) */ + connection_mark_for_close(TO_CONN(conn)); + return -1; + } + return 0; + } + /* Fall through if the connection is on a circuit without optimistic + * data support. */ + /* Falls through. */ + case EXIT_CONN_STATE_CONNECTING: + case AP_CONN_STATE_RENDDESC_WAIT: + case AP_CONN_STATE_CIRCUIT_WAIT: + case AP_CONN_STATE_RESOLVE_WAIT: + case AP_CONN_STATE_CONTROLLER_WAIT: + log_info(LD_EDGE, + "data from edge while in '%s' state. Leaving it on buffer.", + conn_state_to_string(conn->base_.type, conn->base_.state)); + return 0; + } + log_warn(LD_BUG,"Got unexpected state %d. Closing.",conn->base_.state); + tor_fragile_assert(); + connection_edge_end(conn, END_STREAM_REASON_INTERNAL); + connection_mark_for_close(TO_CONN(conn)); + return -1; +} + +/** This edge needs to be closed, because its circuit has closed. + * Mark it for close and return 0. + */ +int +connection_edge_destroy(circid_t circ_id, edge_connection_t *conn) +{ + if (!conn->base_.marked_for_close) { + log_info(LD_EDGE, "CircID %u: At an edge. Marking connection for close.", + (unsigned) circ_id); + if (conn->base_.type == CONN_TYPE_AP) { + entry_connection_t *entry_conn = EDGE_TO_ENTRY_CONN(conn); + connection_mark_unattached_ap(entry_conn, END_STREAM_REASON_DESTROY); + control_event_stream_bandwidth(conn); + control_event_stream_status(entry_conn, STREAM_EVENT_CLOSED, + END_STREAM_REASON_DESTROY); + conn->end_reason |= END_STREAM_REASON_FLAG_ALREADY_SENT_CLOSED; + } else { + /* closing the circuit, nothing to send an END to */ + conn->edge_has_sent_end = 1; + conn->end_reason = END_STREAM_REASON_DESTROY; + conn->end_reason |= END_STREAM_REASON_FLAG_ALREADY_SENT_CLOSED; + connection_mark_and_flush(TO_CONN(conn)); + } + } + conn->cpath_layer = NULL; + conn->on_circuit = NULL; + return 0; +} + +/** Send a raw end cell to the stream with ID <b>stream_id</b> out over the + * <b>circ</b> towards the hop identified with <b>cpath_layer</b>. If this + * is not a client connection, set the relay end cell's reason for closing + * as <b>reason</b> */ +static int +relay_send_end_cell_from_edge(streamid_t stream_id, circuit_t *circ, + uint8_t reason, crypt_path_t *cpath_layer) +{ + char payload[1]; + + if (CIRCUIT_PURPOSE_IS_CLIENT(circ->purpose)) { + /* Never send the server an informative reason code; it doesn't need to + * know why the client stream is failing. */ + reason = END_STREAM_REASON_MISC; + } + + payload[0] = (char) reason; + + /* Note: we have to use relay_send_command_from_edge here, not + * connection_edge_end or connection_edge_send_command, since those require + * that we have a stream connected to a circuit, and we don't connect to a + * circuit until we have a pending/successful resolve. */ + return relay_send_command_from_edge(stream_id, circ, RELAY_COMMAND_END, + payload, 1, cpath_layer); +} + +/* If the connection <b>conn</b> is attempting to connect to an external + * destination that is an hidden service and the reason is a connection + * refused or timeout, log it so the operator can take appropriate actions. + * The log statement is a rate limited warning. */ +static void +warn_if_hs_unreachable(const edge_connection_t *conn, uint8_t reason) +{ + tor_assert(conn); + + if (conn->base_.type == CONN_TYPE_EXIT && + connection_edge_is_rendezvous_stream(conn) && + (reason == END_STREAM_REASON_CONNECTREFUSED || + reason == END_STREAM_REASON_TIMEOUT)) { +#define WARN_FAILED_HS_CONNECTION 300 + static ratelim_t warn_limit = RATELIM_INIT(WARN_FAILED_HS_CONNECTION); + char *m; + if ((m = rate_limit_log(&warn_limit, approx_time()))) { + log_warn(LD_EDGE, "Onion service connection to %s failed (%s)", + (conn->base_.socket_family == AF_UNIX) ? + safe_str(conn->base_.address) : + safe_str(fmt_addrport(&conn->base_.addr, conn->base_.port)), + stream_end_reason_to_string(reason)); + tor_free(m); + } + } +} + +/** Send a relay end cell from stream <b>conn</b> down conn's circuit, and + * remember that we've done so. If this is not a client connection, set the + * relay end cell's reason for closing as <b>reason</b>. + * + * Return -1 if this function has already been called on this conn, + * else return 0. + */ +int +connection_edge_end(edge_connection_t *conn, uint8_t reason) +{ + char payload[RELAY_PAYLOAD_SIZE]; + size_t payload_len=1; + circuit_t *circ; + uint8_t control_reason = reason; + + if (conn->edge_has_sent_end) { + log_warn(LD_BUG,"(Harmless.) Calling connection_edge_end (reason %d) " + "on an already ended stream?", reason); + tor_fragile_assert(); + return -1; + } + + if (conn->base_.marked_for_close) { + log_warn(LD_BUG, + "called on conn that's already marked for close at %s:%d.", + conn->base_.marked_for_close_file, conn->base_.marked_for_close); + return 0; + } + + circ = circuit_get_by_edge_conn(conn); + if (circ && CIRCUIT_PURPOSE_IS_CLIENT(circ->purpose)) { + /* If this is a client circuit, don't send the server an informative + * reason code; it doesn't need to know why the client stream is + * failing. */ + reason = END_STREAM_REASON_MISC; + } + + payload[0] = (char)reason; + if (reason == END_STREAM_REASON_EXITPOLICY && + !connection_edge_is_rendezvous_stream(conn)) { + int addrlen; + if (tor_addr_family(&conn->base_.addr) == AF_INET) { + set_uint32(payload+1, tor_addr_to_ipv4n(&conn->base_.addr)); + addrlen = 4; + } else { + memcpy(payload+1, tor_addr_to_in6_addr8(&conn->base_.addr), 16); + addrlen = 16; + } + set_uint32(payload+1+addrlen, htonl(dns_clip_ttl(conn->address_ttl))); + payload_len += 4+addrlen; + } + + if (circ && !circ->marked_for_close) { + log_debug(LD_EDGE,"Sending end on conn (fd "TOR_SOCKET_T_FORMAT").", + conn->base_.s); ++ ++ if (CIRCUIT_IS_ORIGIN(circ)) { ++ origin_circuit_t *origin_circ = TO_ORIGIN_CIRCUIT(circ); ++ connection_half_edge_add(conn, origin_circ); ++ } ++ + connection_edge_send_command(conn, RELAY_COMMAND_END, + payload, payload_len); + /* We'll log warn if the connection was an hidden service and couldn't be + * made because the service wasn't available. */ + warn_if_hs_unreachable(conn, control_reason); + } else { + log_debug(LD_EDGE,"No circ to send end on conn " + "(fd "TOR_SOCKET_T_FORMAT").", + conn->base_.s); + } + + conn->edge_has_sent_end = 1; + conn->end_reason = control_reason; + return 0; +} + ++/** ++ * Helper function for bsearch. ++ * ++ * As per smartlist_bsearch, return < 0 if key preceeds member, ++ * > 0 if member preceeds key, and 0 if they are equal. ++ * ++ * This is equivalent to subtraction of the values of key - member ++ * (why does no one ever say that explicitly?). ++ */ ++static int ++connection_half_edge_compare_bsearch(const void *key, const void **member) ++{ ++ const half_edge_t *e2; ++ tor_assert(key); ++ tor_assert(member && *(half_edge_t**)member); ++ e2 = *(const half_edge_t **)member; ++ ++ return *(const streamid_t*)key - e2->stream_id; ++} ++ ++/** ++ * Add a half-closed connection to the list, to watch for activity. ++ * ++ * These connections are removed from the list upon receiving an end ++ * cell. ++ */ ++STATIC void ++connection_half_edge_add(const edge_connection_t *conn, ++ origin_circuit_t *circ) ++{ ++ half_edge_t *half_conn = NULL; ++ int insert_at = 0; ++ int ignored; ++ ++ /* Double-check for re-insertion. This should not happen, ++ * but this check is cheap compared to the sort anyway */ ++ if (connection_half_edge_find_stream_id(circ->half_streams, ++ conn->stream_id)) { ++ log_warn(LD_BUG, "Duplicate stream close for stream %d on circuit %d", ++ conn->stream_id, circ->global_identifier); ++ return; ++ } ++ ++ half_conn = tor_malloc_zero(sizeof(half_edge_t)); ++ ++ if (!circ->half_streams) { ++ circ->half_streams = smartlist_new(); ++ } ++ ++ half_conn->stream_id = conn->stream_id; ++ ++ // How many sendme's should I expect? ++ half_conn->sendmes_pending = ++ (STREAMWINDOW_START-conn->package_window)/STREAMWINDOW_INCREMENT; ++ ++ // Is there a connected cell pending? ++ half_conn->connected_pending = conn->base_.state == ++ AP_CONN_STATE_CONNECT_WAIT; ++ ++ /* Data should only arrive if we're not waiting on a resolved cell. ++ * It can arrive after waiting on connected, because of optimistic ++ * data. */ ++ if (conn->base_.state != AP_CONN_STATE_RESOLVE_WAIT) { ++ // How many more data cells can arrive on this id? ++ half_conn->data_pending = conn->deliver_window; ++ } ++ ++ insert_at = smartlist_bsearch_idx(circ->half_streams, &half_conn->stream_id, ++ connection_half_edge_compare_bsearch, ++ &ignored); ++ smartlist_insert(circ->half_streams, insert_at, half_conn); ++} ++ ++/** ++ * Find a stream_id_t in the list in O(lg(n)). ++ * ++ * Returns NULL if the list is empty or element is not found. ++ * Returns a pointer to the element if found. ++ */ ++STATIC half_edge_t * ++connection_half_edge_find_stream_id(const smartlist_t *half_conns, ++ streamid_t stream_id) ++{ ++ if (!half_conns) ++ return NULL; ++ ++ return smartlist_bsearch(half_conns, &stream_id, ++ connection_half_edge_compare_bsearch); ++} ++ ++/** ++ * Check if this stream_id is in a half-closed state. If so, ++ * check if it still has data cells pending, and decrement that ++ * window if so. ++ * ++ * Return 1 if the data window was not empty. ++ * Return 0 otherwise. ++ */ ++int ++connection_half_edge_is_valid_data(const smartlist_t *half_conns, ++ streamid_t stream_id) ++{ ++ half_edge_t *half = connection_half_edge_find_stream_id(half_conns, ++ stream_id); ++ ++ if (!half) ++ return 0; ++ ++ if (half->data_pending > 0) { ++ half->data_pending--; ++ return 1; ++ } ++ ++ return 0; ++} ++ ++/** ++ * Check if this stream_id is in a half-closed state. If so, ++ * check if it still has a connected cell pending, and decrement ++ * that window if so. ++ * ++ * Return 1 if the connected window was not empty. ++ * Return 0 otherwise. ++ */ ++int ++connection_half_edge_is_valid_connected(const smartlist_t *half_conns, ++ streamid_t stream_id) ++{ ++ half_edge_t *half = connection_half_edge_find_stream_id(half_conns, ++ stream_id); ++ ++ if (!half) ++ return 0; ++ ++ if (half->connected_pending) { ++ half->connected_pending = 0; ++ return 1; ++ } ++ ++ return 0; ++} ++ ++/** ++ * Check if this stream_id is in a half-closed state. If so, ++ * check if it still has sendme cells pending, and decrement that ++ * window if so. ++ * ++ * Return 1 if the sendme window was not empty. ++ * Return 0 otherwise. ++ */ ++int ++connection_half_edge_is_valid_sendme(const smartlist_t *half_conns, ++ streamid_t stream_id) ++{ ++ half_edge_t *half = connection_half_edge_find_stream_id(half_conns, ++ stream_id); ++ ++ if (!half) ++ return 0; ++ ++ if (half->sendmes_pending > 0) { ++ half->sendmes_pending--; ++ return 1; ++ } ++ ++ return 0; ++} ++ ++/** ++ * Check if this stream_id is in a half-closed state. If so, remove ++ * it from the list. No other data should come after the END cell. ++ * ++ * Return 1 if stream_id was in half-closed state. ++ * Return 0 otherwise. ++ */ ++int ++connection_half_edge_is_valid_end(smartlist_t *half_conns, ++ streamid_t stream_id) ++{ ++ half_edge_t *half; ++ int found, remove_idx; ++ ++ if (!half_conns) ++ return 0; ++ ++ remove_idx = smartlist_bsearch_idx(half_conns, &stream_id, ++ connection_half_edge_compare_bsearch, ++ &found); ++ if (!found) ++ return 0; ++ ++ half = smartlist_get(half_conns, remove_idx); ++ smartlist_del_keeporder(half_conns, remove_idx); ++ tor_free(half); ++ return 1; ++} ++ ++/** ++ * Streams that were used to send a RESOLVE cell are closed ++ * when they get the RESOLVED, without an end. So treat ++ * a RESOLVED just like an end, and remove from the list. ++ */ ++int ++connection_half_edge_is_valid_resolved(smartlist_t *half_conns, ++ streamid_t stream_id) ++{ ++ return connection_half_edge_is_valid_end(half_conns, stream_id); ++} ++ +/** An error has just occurred on an operation on an edge connection + * <b>conn</b>. Extract the errno; convert it to an end reason, and send an + * appropriate relay end cell to the other end of the connection's circuit. + **/ +int +connection_edge_end_errno(edge_connection_t *conn) +{ + uint8_t reason; + tor_assert(conn); + reason = errno_to_stream_end_reason(tor_socket_errno(conn->base_.s)); + return connection_edge_end(conn, reason); +} + +/** We just wrote some data to <b>conn</b>; act appropriately. + * + * (That is, if it's open, consider sending a stream-level sendme cell if we + * have just flushed enough.) + */ +int +connection_edge_flushed_some(edge_connection_t *conn) +{ + switch (conn->base_.state) { + case AP_CONN_STATE_OPEN: + case EXIT_CONN_STATE_OPEN: + connection_edge_consider_sending_sendme(conn); + break; + } + return 0; +} + +/** Connection <b>conn</b> has finished writing and has no bytes left on + * its outbuf. + * + * If it's in state 'open', stop writing, consider responding with a + * sendme, and return. + * Otherwise, stop writing and return. + * + * If <b>conn</b> is broken, mark it for close and return -1, else + * return 0. + */ +int +connection_edge_finished_flushing(edge_connection_t *conn) +{ + tor_assert(conn); + + switch (conn->base_.state) { + case AP_CONN_STATE_OPEN: + case EXIT_CONN_STATE_OPEN: + connection_edge_consider_sending_sendme(conn); + return 0; + case AP_CONN_STATE_SOCKS_WAIT: + case AP_CONN_STATE_NATD_WAIT: + case AP_CONN_STATE_RENDDESC_WAIT: + case AP_CONN_STATE_CIRCUIT_WAIT: + case AP_CONN_STATE_CONNECT_WAIT: + case AP_CONN_STATE_CONTROLLER_WAIT: + case AP_CONN_STATE_RESOLVE_WAIT: + case AP_CONN_STATE_HTTP_CONNECT_WAIT: + return 0; + default: + log_warn(LD_BUG, "Called in unexpected state %d.",conn->base_.state); + tor_fragile_assert(); + return -1; + } + return 0; +} + +/** Longest size for the relay payload of a RELAY_CONNECTED cell that we're + * able to generate. */ +/* 4 zero bytes; 1 type byte; 16 byte IPv6 address; 4 byte TTL. */ +#define MAX_CONNECTED_CELL_PAYLOAD_LEN 25 + +/** Set the buffer at <b>payload_out</b> -- which must have at least + * MAX_CONNECTED_CELL_PAYLOAD_LEN bytes available -- to the body of a + * RELAY_CONNECTED cell indicating that we have connected to <b>addr</b>, and + * that the name resolution that led us to <b>addr</b> will be valid for + * <b>ttl</b> seconds. Return -1 on error, or the number of bytes used on + * success. */ +STATIC int +connected_cell_format_payload(uint8_t *payload_out, + const tor_addr_t *addr, + uint32_t ttl) +{ + const sa_family_t family = tor_addr_family(addr); + int connected_payload_len; + + /* should be needless */ + memset(payload_out, 0, MAX_CONNECTED_CELL_PAYLOAD_LEN); + + if (family == AF_INET) { + set_uint32(payload_out, tor_addr_to_ipv4n(addr)); + connected_payload_len = 4; + } else if (family == AF_INET6) { + set_uint32(payload_out, 0); + set_uint8(payload_out + 4, 6); + memcpy(payload_out + 5, tor_addr_to_in6_addr8(addr), 16); + connected_payload_len = 21; + } else { + return -1; + } + + set_uint32(payload_out + connected_payload_len, htonl(dns_clip_ttl(ttl))); + connected_payload_len += 4; + + tor_assert(connected_payload_len <= MAX_CONNECTED_CELL_PAYLOAD_LEN); + + return connected_payload_len; +} + +/** Connected handler for exit connections: start writing pending + * data, deliver 'CONNECTED' relay cells as appropriate, and check + * any pending data that may have been received. */ +int +connection_edge_finished_connecting(edge_connection_t *edge_conn) +{ + connection_t *conn; + + tor_assert(edge_conn); + tor_assert(edge_conn->base_.type == CONN_TYPE_EXIT); + conn = TO_CONN(edge_conn); + tor_assert(conn->state == EXIT_CONN_STATE_CONNECTING); + + log_info(LD_EXIT,"Exit connection to %s:%u (%s) established.", + escaped_safe_str(conn->address), conn->port, + safe_str(fmt_and_decorate_addr(&conn->addr))); + + rep_hist_note_exit_stream_opened(conn->port); + + conn->state = EXIT_CONN_STATE_OPEN; + connection_watch_events(conn, READ_EVENT); /* stop writing, keep reading */ + if (connection_get_outbuf_len(conn)) /* in case there are any queued relay + * cells */ + connection_start_writing(conn); + /* deliver a 'connected' relay cell back through the circuit. */ + if (connection_edge_is_rendezvous_stream(edge_conn)) { + if (connection_edge_send_command(edge_conn, + RELAY_COMMAND_CONNECTED, NULL, 0) < 0) + return 0; /* circuit is closed, don't continue */ + } else { + uint8_t connected_payload[MAX_CONNECTED_CELL_PAYLOAD_LEN]; + int connected_payload_len = + connected_cell_format_payload(connected_payload, &conn->addr, + edge_conn->address_ttl); + if (connected_payload_len < 0) + return -1; + + if (connection_edge_send_command(edge_conn, + RELAY_COMMAND_CONNECTED, + (char*)connected_payload, connected_payload_len) < 0) + return 0; /* circuit is closed, don't continue */ + } + tor_assert(edge_conn->package_window > 0); + /* in case the server has written anything */ + return connection_edge_process_inbuf(edge_conn, 1); +} + +/** A list of all the entry_connection_t * objects that are not marked + * for close, and are in AP_CONN_STATE_CIRCUIT_WAIT. + * + * (Right now, we check in several places to make sure that this list is + * correct. When it's incorrect, we'll fix it, and log a BUG message.) + */ +static smartlist_t *pending_entry_connections = NULL; + +static int untried_pending_connections = 0; + +/** + * Mainloop event to tell us to scan for pending connections that can + * be attached. + */ +static mainloop_event_t *attach_pending_entry_connections_ev = NULL; + +/** Common code to connection_(ap|exit)_about_to_close. */ +static void +connection_edge_about_to_close(edge_connection_t *edge_conn) +{ + if (!edge_conn->edge_has_sent_end) { + connection_t *conn = TO_CONN(edge_conn); + log_warn(LD_BUG, "(Harmless.) Edge connection (marked at %s:%d) " + "hasn't sent end yet?", + conn->marked_for_close_file, conn->marked_for_close); + tor_fragile_assert(); + } +} + +/** Called when we're about to finally unlink and free an AP (client) + * connection: perform necessary accounting and cleanup */ +void +connection_ap_about_to_close(entry_connection_t *entry_conn) +{ + circuit_t *circ; + edge_connection_t *edge_conn = ENTRY_TO_EDGE_CONN(entry_conn); + connection_t *conn = ENTRY_TO_CONN(entry_conn); + + connection_edge_about_to_close(edge_conn); + + if (entry_conn->socks_request->has_finished == 0) { + /* since conn gets removed right after this function finishes, + * there's no point trying to send back a reply at this point. */ + log_warn(LD_BUG,"Closing stream (marked at %s:%d) without sending" + " back a socks reply.", + conn->marked_for_close_file, conn->marked_for_close); + } + if (!edge_conn->end_reason) { + log_warn(LD_BUG,"Closing stream (marked at %s:%d) without having" + " set end_reason.", + conn->marked_for_close_file, conn->marked_for_close); + } + if (entry_conn->dns_server_request) { + log_warn(LD_BUG,"Closing stream (marked at %s:%d) without having" + " replied to DNS request.", + conn->marked_for_close_file, conn->marked_for_close); + dnsserv_reject_request(entry_conn); + } + + if (TO_CONN(edge_conn)->state == AP_CONN_STATE_CIRCUIT_WAIT) { + smartlist_remove(pending_entry_connections, entry_conn); + } + +#if 1 + /* Check to make sure that this isn't in pending_entry_connections if it + * didn't actually belong there. */ + if (TO_CONN(edge_conn)->type == CONN_TYPE_AP) { + connection_ap_warn_and_unmark_if_pending_circ(entry_conn, + "about_to_close"); + } +#endif /* 1 */ + + control_event_stream_bandwidth(edge_conn); + control_event_stream_status(entry_conn, STREAM_EVENT_CLOSED, + edge_conn->end_reason); + circ = circuit_get_by_edge_conn(edge_conn); + if (circ) + circuit_detach_stream(circ, edge_conn); +} + +/** Called when we're about to finally unlink and free an exit + * connection: perform necessary accounting and cleanup */ +void +connection_exit_about_to_close(edge_connection_t *edge_conn) +{ + circuit_t *circ; + connection_t *conn = TO_CONN(edge_conn); + + connection_edge_about_to_close(edge_conn); + + circ = circuit_get_by_edge_conn(edge_conn); + if (circ) + circuit_detach_stream(circ, edge_conn); + if (conn->state == EXIT_CONN_STATE_RESOLVING) { + connection_dns_remove(edge_conn); + } +} + +/** Define a schedule for how long to wait between retrying + * application connections. Rather than waiting a fixed amount of + * time between each retry, we wait 10 seconds each for the first + * two tries, and 15 seconds for each retry after + * that. Hopefully this will improve the expected user experience. */ +static int +compute_retry_timeout(entry_connection_t *conn) +{ + int timeout = get_options()->CircuitStreamTimeout; + if (timeout) /* if our config options override the default, use them */ + return timeout; + if (conn->num_socks_retries < 2) /* try 0 and try 1 */ + return 10; + return 15; +} + +/** Find all general-purpose AP streams waiting for a response that sent their + * begin/resolve cell too long ago. Detach from their current circuit, and + * mark their current circuit as unsuitable for new streams. Then call + * connection_ap_handshake_attach_circuit() to attach to a new circuit (if + * available) or launch a new one. + * + * For rendezvous streams, simply give up after SocksTimeout seconds (with no + * retry attempt). + */ +void +connection_ap_expire_beginning(void) +{ + edge_connection_t *conn; + entry_connection_t *entry_conn; + circuit_t *circ; + time_t now = time(NULL); + const or_options_t *options = get_options(); + int severity; + int cutoff; + int seconds_idle, seconds_since_born; + smartlist_t *conns = get_connection_array(); + + SMARTLIST_FOREACH_BEGIN(conns, connection_t *, base_conn) { + if (base_conn->type != CONN_TYPE_AP || base_conn->marked_for_close) + continue; + entry_conn = TO_ENTRY_CONN(base_conn); + conn = ENTRY_TO_EDGE_CONN(entry_conn); + /* if it's an internal linked connection, don't yell its status. */ + severity = (tor_addr_is_null(&base_conn->addr) && !base_conn->port) + ? LOG_INFO : LOG_NOTICE; + seconds_idle = (int)( now - base_conn->timestamp_last_read_allowed ); + seconds_since_born = (int)( now - base_conn->timestamp_created ); + + if (base_conn->state == AP_CONN_STATE_OPEN) + continue; + + /* We already consider SocksTimeout in + * connection_ap_handshake_attach_circuit(), but we need to consider + * it here too because controllers that put streams in controller_wait + * state never ask Tor to attach the circuit. */ + if (AP_CONN_STATE_IS_UNATTACHED(base_conn->state)) { + if (seconds_since_born >= options->SocksTimeout) { + log_fn(severity, LD_APP, + "Tried for %d seconds to get a connection to %s:%d. " + "Giving up. (%s)", + seconds_since_born, + safe_str_client(entry_conn->socks_request->address), + entry_conn->socks_request->port, + conn_state_to_string(CONN_TYPE_AP, base_conn->state)); + connection_mark_unattached_ap(entry_conn, END_STREAM_REASON_TIMEOUT); + } + continue; + } + + /* We're in state connect_wait or resolve_wait now -- waiting for a + * reply to our relay cell. See if we want to retry/give up. */ + + cutoff = compute_retry_timeout(entry_conn); + if (seconds_idle < cutoff) + continue; + circ = circuit_get_by_edge_conn(conn); + if (!circ) { /* it's vanished? */ + log_info(LD_APP,"Conn is waiting (address %s), but lost its circ.", + safe_str_client(entry_conn->socks_request->address)); + connection_mark_unattached_ap(entry_conn, END_STREAM_REASON_TIMEOUT); + continue; + } + if (circ->purpose == CIRCUIT_PURPOSE_C_REND_JOINED) { + if (seconds_idle >= options->SocksTimeout) { + log_fn(severity, LD_REND, + "Rend stream is %d seconds late. Giving up on address" + " '%s.onion'.", + seconds_idle, + safe_str_client(entry_conn->socks_request->address)); + /* Roll back path bias use state so that we probe the circuit + * if nothing else succeeds on it */ + pathbias_mark_use_rollback(TO_ORIGIN_CIRCUIT(circ)); + + connection_edge_end(conn, END_STREAM_REASON_TIMEOUT); + connection_mark_unattached_ap(entry_conn, END_STREAM_REASON_TIMEOUT); + } + continue; + } + + if (circ->purpose != CIRCUIT_PURPOSE_C_GENERAL && + circ->purpose != CIRCUIT_PURPOSE_C_HSDIR_GET && + circ->purpose != CIRCUIT_PURPOSE_S_HSDIR_POST && + circ->purpose != CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT && + circ->purpose != CIRCUIT_PURPOSE_PATH_BIAS_TESTING) { + log_warn(LD_BUG, "circuit->purpose == CIRCUIT_PURPOSE_C_GENERAL failed. " + "The purpose on the circuit was %s; it was in state %s, " + "path_state %s.", + circuit_purpose_to_string(circ->purpose), + circuit_state_to_string(circ->state), + CIRCUIT_IS_ORIGIN(circ) ? + pathbias_state_to_string(TO_ORIGIN_CIRCUIT(circ)->path_state) : + "none"); + } + log_fn(cutoff < 15 ? LOG_INFO : severity, LD_APP, + "We tried for %d seconds to connect to '%s' using exit %s." + " Retrying on a new circuit.", + seconds_idle, + safe_str_client(entry_conn->socks_request->address), + conn->cpath_layer ? + extend_info_describe(conn->cpath_layer->extend_info): + "*unnamed*"); + /* send an end down the circuit */ + connection_edge_end(conn, END_STREAM_REASON_TIMEOUT); + /* un-mark it as ending, since we're going to reuse it */ + conn->edge_has_sent_end = 0; + conn->end_reason = 0; + /* make us not try this circuit again, but allow + * current streams on it to survive if they can */ + mark_circuit_unusable_for_new_conns(TO_ORIGIN_CIRCUIT(circ)); + + /* give our stream another 'cutoff' seconds to try */ + conn->base_.timestamp_last_read_allowed += cutoff; + if (entry_conn->num_socks_retries < 250) /* avoid overflow */ + entry_conn->num_socks_retries++; + /* move it back into 'pending' state, and try to attach. */ + if (connection_ap_detach_retriable(entry_conn, TO_ORIGIN_CIRCUIT(circ), + END_STREAM_REASON_TIMEOUT)<0) { + if (!base_conn->marked_for_close) + connection_mark_unattached_ap(entry_conn, + END_STREAM_REASON_CANT_ATTACH); + } + } SMARTLIST_FOREACH_END(base_conn); +} + +/** + * As connection_ap_attach_pending, but first scans the entire connection + * array to see if any elements are missing. + */ +void +connection_ap_rescan_and_attach_pending(void) +{ + entry_connection_t *entry_conn; + smartlist_t *conns = get_connection_array(); + + if (PREDICT_UNLIKELY(NULL == pending_entry_connections)) + pending_entry_connections = smartlist_new(); + + SMARTLIST_FOREACH_BEGIN(conns, connection_t *, conn) { + if (conn->marked_for_close || + conn->type != CONN_TYPE_AP || + conn->state != AP_CONN_STATE_CIRCUIT_WAIT) + continue; + + entry_conn = TO_ENTRY_CONN(conn); + tor_assert(entry_conn); + if (! smartlist_contains(pending_entry_connections, entry_conn)) { + log_warn(LD_BUG, "Found a connection %p that was supposed to be " + "in pending_entry_connections, but wasn't. No worries; " + "adding it.", + pending_entry_connections); + untried_pending_connections = 1; + connection_ap_mark_as_pending_circuit(entry_conn); + } + + } SMARTLIST_FOREACH_END(conn); + + connection_ap_attach_pending(1); +} + +#ifdef DEBUGGING_17659 +#define UNMARK() do { \ + entry_conn->marked_pending_circ_line = 0; \ + entry_conn->marked_pending_circ_file = 0; \ + } while (0) +#else /* !(defined(DEBUGGING_17659)) */ +#define UNMARK() do { } while (0) +#endif /* defined(DEBUGGING_17659) */ + +/** Tell any AP streams that are listed as waiting for a new circuit to try + * again. If there is an available circuit for a stream, attach it. Otherwise, + * launch a new circuit. + * + * If <b>retry</b> is false, only check the list if it contains at least one + * streams that we have not yet tried to attach to a circuit. + */ +void +connection_ap_attach_pending(int retry) +{ + if (PREDICT_UNLIKELY(!pending_entry_connections)) { + return; + } + + if (untried_pending_connections == 0 && !retry) + return; + + /* Don't allow any modifications to list while we are iterating over + * it. We'll put streams back on this list if we can't attach them + * immediately. */ + smartlist_t *pending = pending_entry_connections; + pending_entry_connections = smartlist_new(); + + SMARTLIST_FOREACH_BEGIN(pending, + entry_connection_t *, entry_conn) { + connection_t *conn = ENTRY_TO_CONN(entry_conn); + tor_assert(conn && entry_conn); + if (conn->marked_for_close) { + UNMARK(); + continue; + } + if (conn->magic != ENTRY_CONNECTION_MAGIC) { + log_warn(LD_BUG, "%p has impossible magic value %u.", + entry_conn, (unsigned)conn->magic); + UNMARK(); + continue; + } + if (conn->state != AP_CONN_STATE_CIRCUIT_WAIT) { + log_warn(LD_BUG, "%p is no longer in circuit_wait. Its current state " + "is %s. Why is it on pending_entry_connections?", + entry_conn, + conn_state_to_string(conn->type, conn->state)); + UNMARK(); + continue; + } + + /* Okay, we're through the sanity checks. Try to handle this stream. */ + if (connection_ap_handshake_attach_circuit(entry_conn) < 0) { + if (!conn->marked_for_close) + connection_mark_unattached_ap(entry_conn, + END_STREAM_REASON_CANT_ATTACH); + } + + if (! conn->marked_for_close && + conn->type == CONN_TYPE_AP && + conn->state == AP_CONN_STATE_CIRCUIT_WAIT) { + /* Is it still waiting for a circuit? If so, we didn't attach it, + * so it's still pending. Put it back on the list. + */ + if (!smartlist_contains(pending_entry_connections, entry_conn)) { + smartlist_add(pending_entry_connections, entry_conn); + continue; + } + } + + /* If we got here, then we either closed the connection, or + * we attached it. */ + UNMARK(); + } SMARTLIST_FOREACH_END(entry_conn); + + smartlist_free(pending); + untried_pending_connections = 0; +} + +static void +attach_pending_entry_connections_cb(mainloop_event_t *ev, void *arg) +{ + (void)ev; + (void)arg; + connection_ap_attach_pending(0); +} + +/** Mark <b>entry_conn</b> as needing to get attached to a circuit. + * + * And <b>entry_conn</b> must be in AP_CONN_STATE_CIRCUIT_WAIT, + * should not already be pending a circuit. The circuit will get + * launched or the connection will get attached the next time we + * call connection_ap_attach_pending(). + */ +void +connection_ap_mark_as_pending_circuit_(entry_connection_t *entry_conn, + const char *fname, int lineno) +{ + connection_t *conn = ENTRY_TO_CONN(entry_conn); + tor_assert(conn->state == AP_CONN_STATE_CIRCUIT_WAIT); + tor_assert(conn->magic == ENTRY_CONNECTION_MAGIC); + if (conn->marked_for_close) + return; + + if (PREDICT_UNLIKELY(NULL == pending_entry_connections)) { + pending_entry_connections = smartlist_new(); + } + if (PREDICT_UNLIKELY(NULL == attach_pending_entry_connections_ev)) { + attach_pending_entry_connections_ev = mainloop_event_postloop_new( + attach_pending_entry_connections_cb, NULL); + } + if (PREDICT_UNLIKELY(smartlist_contains(pending_entry_connections, + entry_conn))) { + log_warn(LD_BUG, "What?? pending_entry_connections already contains %p! " + "(Called from %s:%d.)", + entry_conn, fname, lineno); +#ifdef DEBUGGING_17659 + const char *f2 = entry_conn->marked_pending_circ_file; + log_warn(LD_BUG, "(Previously called from %s:%d.)\n", + f2 ? f2 : "<NULL>", + entry_conn->marked_pending_circ_line); +#endif /* defined(DEBUGGING_17659) */ + log_backtrace(LOG_WARN, LD_BUG, "To debug, this may help"); + return; + } + +#ifdef DEBUGGING_17659 + entry_conn->marked_pending_circ_line = (uint16_t) lineno; + entry_conn->marked_pending_circ_file = fname; +#endif + + untried_pending_connections = 1; + smartlist_add(pending_entry_connections, entry_conn); + + mainloop_event_activate(attach_pending_entry_connections_ev); +} + +/** Mark <b>entry_conn</b> as no longer waiting for a circuit. */ +void +connection_ap_mark_as_non_pending_circuit(entry_connection_t *entry_conn) +{ + if (PREDICT_UNLIKELY(NULL == pending_entry_connections)) + return; + UNMARK(); + smartlist_remove(pending_entry_connections, entry_conn); +} + +/* DOCDOC */ +void +connection_ap_warn_and_unmark_if_pending_circ(entry_connection_t *entry_conn, + const char *where) +{ + if (pending_entry_connections && + smartlist_contains(pending_entry_connections, entry_conn)) { + log_warn(LD_BUG, "What was %p doing in pending_entry_connections in %s?", + entry_conn, where); + connection_ap_mark_as_non_pending_circuit(entry_conn); + } +} + +/** Tell any AP streams that are waiting for a one-hop tunnel to + * <b>failed_digest</b> that they are going to fail. */ +/* XXXX We should get rid of this function, and instead attach + * one-hop streams to circ->p_streams so they get marked in + * circuit_mark_for_close like normal p_streams. */ +void +connection_ap_fail_onehop(const char *failed_digest, + cpath_build_state_t *build_state) +{ + entry_connection_t *entry_conn; + char digest[DIGEST_LEN]; + smartlist_t *conns = get_connection_array(); + SMARTLIST_FOREACH_BEGIN(conns, connection_t *, conn) { + if (conn->marked_for_close || + conn->type != CONN_TYPE_AP || + conn->state != AP_CONN_STATE_CIRCUIT_WAIT) + continue; + entry_conn = TO_ENTRY_CONN(conn); + if (!entry_conn->want_onehop) + continue; + if (hexdigest_to_digest(entry_conn->chosen_exit_name, digest) < 0 || + tor_memneq(digest, failed_digest, DIGEST_LEN)) + continue; + if (tor_digest_is_zero(digest)) { + /* we don't know the digest; have to compare addr:port */ + tor_addr_t addr; + if (!build_state || !build_state->chosen_exit || + !entry_conn->socks_request) { + continue; + } + if (tor_addr_parse(&addr, entry_conn->socks_request->address)<0 || + !tor_addr_eq(&build_state->chosen_exit->addr, &addr) || + build_state->chosen_exit->port != entry_conn->socks_request->port) + continue; + } + log_info(LD_APP, "Closing one-hop stream to '%s/%s' because the OR conn " + "just failed.", entry_conn->chosen_exit_name, + entry_conn->socks_request->address); + connection_mark_unattached_ap(entry_conn, END_STREAM_REASON_TIMEOUT); + } SMARTLIST_FOREACH_END(conn); +} + +/** A circuit failed to finish on its last hop <b>info</b>. If there + * are any streams waiting with this exit node in mind, but they + * don't absolutely require it, make them give up on it. + */ +void +circuit_discard_optional_exit_enclaves(extend_info_t *info) +{ + entry_connection_t *entry_conn; + const node_t *r1, *r2; + + smartlist_t *conns = get_connection_array(); + SMARTLIST_FOREACH_BEGIN(conns, connection_t *, conn) { + if (conn->marked_for_close || + conn->type != CONN_TYPE_AP || + conn->state != AP_CONN_STATE_CIRCUIT_WAIT) + continue; + entry_conn = TO_ENTRY_CONN(conn); + if (!entry_conn->chosen_exit_optional && + !entry_conn->chosen_exit_retries) + continue; + r1 = node_get_by_nickname(entry_conn->chosen_exit_name, + NNF_NO_WARN_UNNAMED); + r2 = node_get_by_id(info->identity_digest); + if (!r1 || !r2 || r1 != r2) + continue; + tor_assert(entry_conn->socks_request); + if (entry_conn->chosen_exit_optional) { + log_info(LD_APP, "Giving up on enclave exit '%s' for destination %s.", + safe_str_client(entry_conn->chosen_exit_name), + escaped_safe_str_client(entry_conn->socks_request->address)); + entry_conn->chosen_exit_optional = 0; + tor_free(entry_conn->chosen_exit_name); /* clears it */ + /* if this port is dangerous, warn or reject it now that we don't + * think it'll be using an enclave. */ + consider_plaintext_ports(entry_conn, entry_conn->socks_request->port); + } + if (entry_conn->chosen_exit_retries) { + if (--entry_conn->chosen_exit_retries == 0) { /* give up! */ + clear_trackexithost_mappings(entry_conn->chosen_exit_name); + tor_free(entry_conn->chosen_exit_name); /* clears it */ + /* if this port is dangerous, warn or reject it now that we don't + * think it'll be using an enclave. */ + consider_plaintext_ports(entry_conn, entry_conn->socks_request->port); + } + } + } SMARTLIST_FOREACH_END(conn); +} + +/** The AP connection <b>conn</b> has just failed while attaching or + * sending a BEGIN or resolving on <b>circ</b>, but another circuit + * might work. Detach the circuit, and either reattach it, launch a + * new circuit, tell the controller, or give up as appropriate. + * + * Returns -1 on err, 1 on success, 0 on not-yet-sure. + */ +int +connection_ap_detach_retriable(entry_connection_t *conn, + origin_circuit_t *circ, + int reason) +{ + control_event_stream_status(conn, STREAM_EVENT_FAILED_RETRIABLE, reason); + ENTRY_TO_CONN(conn)->timestamp_last_read_allowed = time(NULL); + + /* Roll back path bias use state so that we probe the circuit + * if nothing else succeeds on it */ + pathbias_mark_use_rollback(circ); + + if (conn->pending_optimistic_data) { + buf_set_to_copy(&conn->sending_optimistic_data, + conn->pending_optimistic_data); + } + + if (!get_options()->LeaveStreamsUnattached || conn->use_begindir) { + /* If we're attaching streams ourself, or if this connection is + * a tunneled directory connection, then just attach it. */ + ENTRY_TO_CONN(conn)->state = AP_CONN_STATE_CIRCUIT_WAIT; + circuit_detach_stream(TO_CIRCUIT(circ),ENTRY_TO_EDGE_CONN(conn)); + connection_ap_mark_as_pending_circuit(conn); + } else { + CONNECTION_AP_EXPECT_NONPENDING(conn); + ENTRY_TO_CONN(conn)->state = AP_CONN_STATE_CONTROLLER_WAIT; + circuit_detach_stream(TO_CIRCUIT(circ),ENTRY_TO_EDGE_CONN(conn)); + } + return 0; +} + +/** Check if <b>conn</b> is using a dangerous port. Then warn and/or + * reject depending on our config options. */ +static int +consider_plaintext_ports(entry_connection_t *conn, uint16_t port) +{ + const or_options_t *options = get_options(); + int reject = smartlist_contains_int_as_string( + options->RejectPlaintextPorts, port); + + if (smartlist_contains_int_as_string(options->WarnPlaintextPorts, port)) { + log_warn(LD_APP, "Application request to port %d: this port is " + "commonly used for unencrypted protocols. Please make sure " + "you don't send anything you would mind the rest of the " + "Internet reading!%s", port, reject ? " Closing." : ""); + control_event_client_status(LOG_WARN, "DANGEROUS_PORT PORT=%d RESULT=%s", + port, reject ? "REJECT" : "WARN"); + } + + if (reject) { + log_info(LD_APP, "Port %d listed in RejectPlaintextPorts. Closing.", port); + connection_mark_unattached_ap(conn, END_STREAM_REASON_ENTRYPOLICY); + return -1; + } + + return 0; +} + +/** How many times do we try connecting with an exit configured via + * TrackHostExits before concluding that it won't work any more and trying a + * different one? */ +#define TRACKHOSTEXITS_RETRIES 5 + +/** Call connection_ap_handshake_rewrite_and_attach() unless a controller + * asked us to leave streams unattached. Return 0 in that case. + * + * See connection_ap_handshake_rewrite_and_attach()'s + * documentation for arguments and return value. + */ +MOCK_IMPL(int, +connection_ap_rewrite_and_attach_if_allowed,(entry_connection_t *conn, + origin_circuit_t *circ, + crypt_path_t *cpath)) +{ + const or_options_t *options = get_options(); + + if (options->LeaveStreamsUnattached) { + CONNECTION_AP_EXPECT_NONPENDING(conn); + ENTRY_TO_CONN(conn)->state = AP_CONN_STATE_CONTROLLER_WAIT; + return 0; + } + return connection_ap_handshake_rewrite_and_attach(conn, circ, cpath); +} + +/* Try to perform any map-based rewriting of the target address in + * <b>conn</b>, filling in the fields of <b>out</b> as we go, and modifying + * conn->socks_request.address as appropriate. + */ +STATIC void +connection_ap_handshake_rewrite(entry_connection_t *conn, + rewrite_result_t *out) +{ + socks_request_t *socks = conn->socks_request; + const or_options_t *options = get_options(); + tor_addr_t addr_tmp; + + /* Initialize all the fields of 'out' to reasonable defaults */ + out->automap = 0; + out->exit_source = ADDRMAPSRC_NONE; + out->map_expires = TIME_MAX; + out->end_reason = 0; + out->should_close = 0; + out->orig_address[0] = 0; + + /* We convert all incoming addresses to lowercase. */ + tor_strlower(socks->address); + /* Remember the original address. */ + strlcpy(out->orig_address, socks->address, sizeof(out->orig_address)); + log_debug(LD_APP,"Client asked for %s:%d", + safe_str_client(socks->address), + socks->port); + + /* Check for whether this is a .exit address. By default, those are + * disallowed when they're coming straight from the client, but you're + * allowed to have them in MapAddress commands and so forth. */ + if (!strcmpend(socks->address, ".exit")) { + log_warn(LD_APP, "The ".exit" notation is disabled in Tor due to " + "security risks."); + control_event_client_status(LOG_WARN, "SOCKS_BAD_HOSTNAME HOSTNAME=%s", + escaped(socks->address)); + out->end_reason = END_STREAM_REASON_TORPROTOCOL; + out->should_close = 1; + return; + } + + /* Remember the original address so we can tell the user about what + * they actually said, not just what it turned into. */ + /* XXX yes, this is the same as out->orig_address above. One is + * in the output, and one is in the connection. */ + if (! conn->original_dest_address) { + /* Is the 'if' necessary here? XXXX */ + conn->original_dest_address = tor_strdup(conn->socks_request->address); + } + + /* First, apply MapAddress and MAPADDRESS mappings. We need to do + * these only for non-reverse lookups, since they don't exist for those. + * We also need to do this before we consider automapping, since we might + * e.g. resolve irc.oftc.net into irconionaddress.onion, at which point + * we'd need to automap it. */ + if (socks->command != SOCKS_COMMAND_RESOLVE_PTR) { + const unsigned rewrite_flags = AMR_FLAG_USE_MAPADDRESS; + if (addressmap_rewrite(socks->address, sizeof(socks->address), + rewrite_flags, &out->map_expires, &out->exit_source)) { + control_event_stream_status(conn, STREAM_EVENT_REMAP, + REMAP_STREAM_SOURCE_CACHE); + } + } + + /* Now see if we need to create or return an existing Hostname->IP + * automapping. Automapping happens when we're asked to resolve a + * hostname, and AutomapHostsOnResolve is set, and the hostname has a + * suffix listed in AutomapHostsSuffixes. It's a handy feature + * that lets you have Tor assign e.g. IPv6 addresses for .onion + * names, and return them safely from DNSPort. + */ + if (socks->command == SOCKS_COMMAND_RESOLVE && + tor_addr_parse(&addr_tmp, socks->address)<0 && + options->AutomapHostsOnResolve) { + /* Check the suffix... */ + out->automap = addressmap_address_should_automap(socks->address, options); + if (out->automap) { + /* If we get here, then we should apply an automapping for this. */ + const char *new_addr; + /* We return an IPv4 address by default, or an IPv6 address if we + * are allowed to do so. */ + int addr_type = RESOLVED_TYPE_IPV4; + if (conn->socks_request->socks_version != 4) { + if (!conn->entry_cfg.ipv4_traffic || + (conn->entry_cfg.ipv6_traffic && conn->entry_cfg.prefer_ipv6) || + conn->entry_cfg.prefer_ipv6_virtaddr) + addr_type = RESOLVED_TYPE_IPV6; + } + /* Okay, register the target address as automapped, and find the new + * address we're supposed to give as a resolve answer. (Return a cached + * value if we've looked up this address before. + */ + new_addr = addressmap_register_virtual_address( + addr_type, tor_strdup(socks->address)); + if (! new_addr) { + log_warn(LD_APP, "Unable to automap address %s", + escaped_safe_str(socks->address)); + out->end_reason = END_STREAM_REASON_INTERNAL; + out->should_close = 1; + return; + } + log_info(LD_APP, "Automapping %s to %s", + escaped_safe_str_client(socks->address), + safe_str_client(new_addr)); + strlcpy(socks->address, new_addr, sizeof(socks->address)); + } + } + + /* Now handle reverse lookups, if they're in the cache. This doesn't + * happen too often, since client-side DNS caching is off by default, + * and very deprecated. */ + if (socks->command == SOCKS_COMMAND_RESOLVE_PTR) { + unsigned rewrite_flags = 0; + if (conn->entry_cfg.use_cached_ipv4_answers) + rewrite_flags |= AMR_FLAG_USE_IPV4_DNS; + if (conn->entry_cfg.use_cached_ipv6_answers) + rewrite_flags |= AMR_FLAG_USE_IPV6_DNS; + + if (addressmap_rewrite_reverse(socks->address, sizeof(socks->address), + rewrite_flags, &out->map_expires)) { + char *result = tor_strdup(socks->address); + /* remember _what_ is supposed to have been resolved. */ + tor_snprintf(socks->address, sizeof(socks->address), "REVERSE[%s]", + out->orig_address); + connection_ap_handshake_socks_resolved(conn, RESOLVED_TYPE_HOSTNAME, + strlen(result), (uint8_t*)result, + -1, + out->map_expires); + tor_free(result); + out->end_reason = END_STREAM_REASON_DONE | + END_STREAM_REASON_FLAG_ALREADY_SOCKS_REPLIED; + out->should_close = 1; + return; + } + + /* Hang on, did we find an answer saying that this is a reverse lookup for + * an internal address? If so, we should reject it if we're configured to + * do so. */ + if (options->ClientDNSRejectInternalAddresses) { + /* Don't let clients try to do a reverse lookup on 10.0.0.1. */ + tor_addr_t addr; + int ok; + ok = tor_addr_parse_PTR_name( + &addr, socks->address, AF_UNSPEC, 1); + if (ok == 1 && tor_addr_is_internal(&addr, 0)) { + connection_ap_handshake_socks_resolved(conn, RESOLVED_TYPE_ERROR, + 0, NULL, -1, TIME_MAX); + out->end_reason = END_STREAM_REASON_SOCKSPROTOCOL | + END_STREAM_REASON_FLAG_ALREADY_SOCKS_REPLIED; + out->should_close = 1; + return; + } + } + } + + /* If we didn't automap it before, then this is still the address that + * came straight from the user, mapped according to any + * MapAddress/MAPADDRESS commands. Now apply other mappings, + * including previously registered Automap entries (IP back to + * hostname), TrackHostExits entries, and client-side DNS cache + * entries (if they're turned on). + */ + if (socks->command != SOCKS_COMMAND_RESOLVE_PTR && + !out->automap) { + unsigned rewrite_flags = AMR_FLAG_USE_AUTOMAP | AMR_FLAG_USE_TRACKEXIT; + addressmap_entry_source_t exit_source2; + if (conn->entry_cfg.use_cached_ipv4_answers) + rewrite_flags |= AMR_FLAG_USE_IPV4_DNS; + if (conn->entry_cfg.use_cached_ipv6_answers) + rewrite_flags |= AMR_FLAG_USE_IPV6_DNS; + if (addressmap_rewrite(socks->address, sizeof(socks->address), + rewrite_flags, &out->map_expires, &exit_source2)) { + control_event_stream_status(conn, STREAM_EVENT_REMAP, + REMAP_STREAM_SOURCE_CACHE); + } + if (out->exit_source == ADDRMAPSRC_NONE) { + /* If it wasn't a .exit before, maybe it turned into a .exit. Remember + * the original source of a .exit. */ + out->exit_source = exit_source2; + } + } + + /* Check to see whether we're about to use an address in the virtual + * range without actually having gotten it from an Automap. */ + if (!out->automap && address_is_in_virtual_range(socks->address)) { + /* This address was probably handed out by + * client_dns_get_unmapped_address, but the mapping was discarded for some + * reason. Or the user typed in a virtual address range manually. We + * *don't* want to send the address through Tor; that's likely to fail, + * and may leak information. + */ + log_warn(LD_APP,"Missing mapping for virtual address '%s'. Refusing.", + safe_str_client(socks->address)); + out->end_reason = END_STREAM_REASON_INTERNAL; + out->should_close = 1; + return; + } +} + +/** We just received a SOCKS request in <b>conn</b> to an onion address of type + * <b>addresstype</b>. Start connecting to the onion service. */ +static int +connection_ap_handle_onion(entry_connection_t *conn, + socks_request_t *socks, + origin_circuit_t *circ, + hostname_type_t addresstype) +{ + time_t now = approx_time(); + connection_t *base_conn = ENTRY_TO_CONN(conn); + + /* If .onion address requests are disabled, refuse the request */ + if (!conn->entry_cfg.onion_traffic) { + log_warn(LD_APP, "Onion address %s requested from a port with .onion " + "disabled", safe_str_client(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_ENTRYPOLICY); + return -1; + } + + /* Check whether it's RESOLVE or RESOLVE_PTR. We don't handle those + * for hidden service addresses. */ + if (SOCKS_COMMAND_IS_RESOLVE(socks->command)) { + /* if it's a resolve request, fail it right now, rather than + * building all the circuits and then realizing it won't work. */ + log_warn(LD_APP, + "Resolve requests to hidden services not allowed. Failing."); + connection_ap_handshake_socks_resolved(conn,RESOLVED_TYPE_ERROR, + 0,NULL,-1,TIME_MAX); + connection_mark_unattached_ap(conn, + END_STREAM_REASON_SOCKSPROTOCOL | + END_STREAM_REASON_FLAG_ALREADY_SOCKS_REPLIED); + return -1; + } + + /* If we were passed a circuit, then we need to fail. .onion addresses + * only work when we launch our own circuits for now. */ + if (circ) { + log_warn(LD_CONTROL, "Attachstream to a circuit is not " + "supported for .onion addresses currently. Failing."); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return -1; + } + + /* Interface: Regardless of HS version after the block below we should have + set onion_address, rend_cache_lookup_result, and descriptor_is_usable. */ + const char *onion_address = NULL; + int rend_cache_lookup_result = -ENOENT; + int descriptor_is_usable = 0; + + if (addresstype == ONION_V2_HOSTNAME) { /* it's a v2 hidden service */ + rend_cache_entry_t *entry = NULL; + /* Look up if we have client authorization configured for this hidden + * service. If we do, associate it with the rend_data. */ + rend_service_authorization_t *client_auth = + rend_client_lookup_service_authorization(socks->address); + + const uint8_t *cookie = NULL; + rend_auth_type_t auth_type = REND_NO_AUTH; + if (client_auth) { + log_info(LD_REND, "Using previously configured client authorization " + "for hidden service request."); + auth_type = client_auth->auth_type; + cookie = client_auth->descriptor_cookie; + } + + /* Fill in the rend_data field so we can start doing a connection to + * a hidden service. */ + rend_data_t *rend_data = ENTRY_TO_EDGE_CONN(conn)->rend_data = + rend_data_client_create(socks->address, NULL, (char *) cookie, + auth_type); + if (rend_data == NULL) { + return -1; + } + onion_address = rend_data_get_address(rend_data); + log_info(LD_REND,"Got a hidden service request for ID '%s'", + safe_str_client(onion_address)); + + rend_cache_lookup_result = rend_cache_lookup_entry(onion_address,-1, + &entry); + if (!rend_cache_lookup_result && entry) { + descriptor_is_usable = rend_client_any_intro_points_usable(entry); + } + } else { /* it's a v3 hidden service */ + tor_assert(addresstype == ONION_V3_HOSTNAME); + const hs_descriptor_t *cached_desc = NULL; + int retval; + /* Create HS conn identifier with HS pubkey */ + hs_ident_edge_conn_t *hs_conn_ident = + tor_malloc_zero(sizeof(hs_ident_edge_conn_t)); + + retval = hs_parse_address(socks->address, &hs_conn_ident->identity_pk, + NULL, NULL); + if (retval < 0) { + log_warn(LD_GENERAL, "failed to parse hs address"); + tor_free(hs_conn_ident); + return -1; + } + ENTRY_TO_EDGE_CONN(conn)->hs_ident = hs_conn_ident; + + onion_address = socks->address; + + /* Check the v3 desc cache */ + cached_desc = hs_cache_lookup_as_client(&hs_conn_ident->identity_pk); + if (cached_desc) { + rend_cache_lookup_result = 0; + descriptor_is_usable = + hs_client_any_intro_points_usable(&hs_conn_ident->identity_pk, + cached_desc); + log_info(LD_GENERAL, "Found %s descriptor in cache for %s. %s.", + (descriptor_is_usable) ? "usable" : "unusable", + safe_str_client(onion_address), + (descriptor_is_usable) ? "Not fetching." : "Refecting."); + } else { + rend_cache_lookup_result = -ENOENT; + } + } + + /* Lookup the given onion address. If invalid, stop right now. + * Otherwise, we might have it in the cache or not. */ + unsigned int refetch_desc = 0; + if (rend_cache_lookup_result < 0) { + switch (-rend_cache_lookup_result) { + case EINVAL: + /* We should already have rejected this address! */ + log_warn(LD_BUG,"Invalid service name '%s'", + safe_str_client(onion_address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return -1; + case ENOENT: + /* We didn't have this; we should look it up. */ + log_info(LD_REND, "No descriptor found in our cache for %s. Fetching.", + safe_str_client(onion_address)); + refetch_desc = 1; + break; + default: + log_warn(LD_BUG, "Unknown cache lookup error %d", + rend_cache_lookup_result); + return -1; + } + } + + /* Help predict that we'll want to do hidden service circuits in the + * future. We're not sure if it will need a stable circuit yet, but + * we know we'll need *something*. */ + rep_hist_note_used_internal(now, 0, 1); + + /* Now we have a descriptor but is it usable or not? If not, refetch. + * Also, a fetch could have been requested if the onion address was not + * found in the cache previously. */ + if (refetch_desc || !descriptor_is_usable) { + edge_connection_t *edge_conn = ENTRY_TO_EDGE_CONN(conn); + connection_ap_mark_as_non_pending_circuit(conn); + base_conn->state = AP_CONN_STATE_RENDDESC_WAIT; + if (addresstype == ONION_V2_HOSTNAME) { + tor_assert(edge_conn->rend_data); + rend_client_refetch_v2_renddesc(edge_conn->rend_data); + /* Whatever the result of the refetch, we don't go further. */ + return 0; + } else { + tor_assert(addresstype == ONION_V3_HOSTNAME); + tor_assert(edge_conn->hs_ident); + /* Attempt to fetch the hsv3 descriptor. Check the retval to see how it + * went and act accordingly. */ + int ret = hs_client_refetch_hsdesc(&edge_conn->hs_ident->identity_pk); + switch (ret) { + case HS_CLIENT_FETCH_MISSING_INFO: + /* Keeping the connection in descriptor wait state is fine because + * once we get enough dirinfo or a new live consensus, the HS client + * subsystem is notified and every connection in that state will + * trigger a fetch for the service key. */ + case HS_CLIENT_FETCH_LAUNCHED: + case HS_CLIENT_FETCH_PENDING: + case HS_CLIENT_FETCH_HAVE_DESC: + return 0; + case HS_CLIENT_FETCH_ERROR: + case HS_CLIENT_FETCH_NO_HSDIRS: + case HS_CLIENT_FETCH_NOT_ALLOWED: + /* Can't proceed further and better close the SOCKS request. */ + return -1; + } + } + } + + /* We have the descriptor! So launch a connection to the HS. */ + log_info(LD_REND, "Descriptor is here. Great."); + + base_conn->state = AP_CONN_STATE_CIRCUIT_WAIT; + /* We'll try to attach it at the next event loop, or whenever + * we call connection_ap_attach_pending() */ + connection_ap_mark_as_pending_circuit(conn); + return 0; +} + +/** Connection <b>conn</b> just finished its socks handshake, or the + * controller asked us to take care of it. If <b>circ</b> is defined, + * then that's where we'll want to attach it. Otherwise we have to + * figure it out ourselves. + * + * First, parse whether it's a .exit address, remap it, and so on. Then + * if it's for a general circuit, try to attach it to a circuit (or launch + * one as needed), else if it's for a rendezvous circuit, fetch a + * rendezvous descriptor first (or attach/launch a circuit if the + * rendezvous descriptor is already here and fresh enough). + * + * The stream will exit from the hop + * indicated by <b>cpath</b>, or from the last hop in circ's cpath if + * <b>cpath</b> is NULL. + */ +int +connection_ap_handshake_rewrite_and_attach(entry_connection_t *conn, + origin_circuit_t *circ, + crypt_path_t *cpath) +{ + socks_request_t *socks = conn->socks_request; + const or_options_t *options = get_options(); + connection_t *base_conn = ENTRY_TO_CONN(conn); + time_t now = time(NULL); + rewrite_result_t rr; + + /* First we'll do the rewrite part. Let's see if we get a reasonable + * answer. + */ + memset(&rr, 0, sizeof(rr)); + connection_ap_handshake_rewrite(conn,&rr); + + if (rr.should_close) { + /* connection_ap_handshake_rewrite told us to close the connection: + * either because it sent back an answer, or because it sent back an + * error */ + connection_mark_unattached_ap(conn, rr.end_reason); + if (END_STREAM_REASON_DONE == (rr.end_reason & END_STREAM_REASON_MASK)) + return 0; + else + return -1; + } + + const time_t map_expires = rr.map_expires; + const int automap = rr.automap; + const addressmap_entry_source_t exit_source = rr.exit_source; + + /* Now, we parse the address to see if it's an .onion or .exit or + * other special address. + */ + const hostname_type_t addresstype = parse_extended_hostname(socks->address); + + /* Now see whether the hostname is bogus. This could happen because of an + * onion hostname whose format we don't recognize. */ + if (addresstype == BAD_HOSTNAME) { + control_event_client_status(LOG_WARN, "SOCKS_BAD_HOSTNAME HOSTNAME=%s", + escaped(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return -1; + } + + /* If this is a .exit hostname, strip off the .name.exit part, and + * see whether we're willing to connect there, and and otherwise handle the + * .exit address. + * + * We'll set chosen_exit_name and/or close the connection as appropriate. + */ + if (addresstype == EXIT_HOSTNAME) { + /* If StrictNodes is not set, then .exit overrides ExcludeNodes but + * not ExcludeExitNodes. */ + routerset_t *excludeset = options->StrictNodes ? + options->ExcludeExitNodesUnion_ : options->ExcludeExitNodes; + const node_t *node = NULL; + + /* If this .exit was added by an AUTOMAP, then it came straight from + * a user. That's not safe. */ + if (exit_source == ADDRMAPSRC_AUTOMAP) { + /* Whoops; this one is stale. It must have gotten added earlier? + * (Probably this is not possible, since AllowDotExit no longer + * exists.) */ + log_warn(LD_APP,"Stale automapped address for '%s.exit'. Refusing.", + safe_str_client(socks->address)); + control_event_client_status(LOG_WARN, "SOCKS_BAD_HOSTNAME HOSTNAME=%s", + escaped(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + tor_assert_nonfatal_unreached(); + return -1; + } + + /* Double-check to make sure there are no .exits coming from + * impossible/weird sources. */ + if (exit_source == ADDRMAPSRC_DNS || exit_source == ADDRMAPSRC_NONE) { + /* It shouldn't be possible to get a .exit address from any of these + * sources. */ + log_warn(LD_BUG,"Address '%s.exit', with impossible source for the " + ".exit part. Refusing.", + safe_str_client(socks->address)); + control_event_client_status(LOG_WARN, "SOCKS_BAD_HOSTNAME HOSTNAME=%s", + escaped(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return -1; + } + + tor_assert(!automap); + + /* Now, find the character before the .(name) part. + * (The ".exit" part got stripped off by "parse_extended_hostname"). + * + * We're going to put the exit name into conn->chosen_exit_name, and + * look up a node correspondingly. */ + char *s = strrchr(socks->address,'.'); + if (s) { + /* The address was of the form "(stuff).(name).exit */ + if (s[1] != '\0') { + /* Looks like a real .exit one. */ + conn->chosen_exit_name = tor_strdup(s+1); + node = node_get_by_nickname(conn->chosen_exit_name, 0); + + if (exit_source == ADDRMAPSRC_TRACKEXIT) { + /* We 5 tries before it expires the addressmap */ + conn->chosen_exit_retries = TRACKHOSTEXITS_RETRIES; + } + *s = 0; + } else { + /* Oops, the address was (stuff)..exit. That's not okay. */ + log_warn(LD_APP,"Malformed exit address '%s.exit'. Refusing.", + safe_str_client(socks->address)); + control_event_client_status(LOG_WARN, "SOCKS_BAD_HOSTNAME HOSTNAME=%s", + escaped(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return -1; + } + } else { + /* It looks like they just asked for "foo.exit". That's a special + * form that means (foo's address).foo.exit. */ + + conn->chosen_exit_name = tor_strdup(socks->address); + node = node_get_by_nickname(conn->chosen_exit_name, 0); + if (node) { + *socks->address = 0; + node_get_address_string(node, socks->address, sizeof(socks->address)); + } + } + + /* Now make sure that the chosen exit exists... */ + if (!node) { + log_warn(LD_APP, + "Unrecognized relay in exit address '%s.exit'. Refusing.", + safe_str_client(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return -1; + } + /* ...and make sure that it isn't excluded. */ + if (routerset_contains_node(excludeset, node)) { + log_warn(LD_APP, + "Excluded relay in exit address '%s.exit'. Refusing.", + safe_str_client(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return -1; + } + /* XXXX-1090 Should we also allow foo.bar.exit if ExitNodes is set and + Bar is not listed in it? I say yes, but our revised manpage branch + implies no. */ + } + + /* Now, we handle everything that isn't a .onion address. */ + if (addresstype != ONION_V2_HOSTNAME && addresstype != ONION_V3_HOSTNAME) { + /* Not a hidden-service request. It's either a hostname or an IP, + * possibly with a .exit that we stripped off. We're going to check + * if we're allowed to connect/resolve there, and then launch the + * appropriate request. */ + + /* Check for funny characters in the address. */ + if (address_is_invalid_destination(socks->address, 1)) { + control_event_client_status(LOG_WARN, "SOCKS_BAD_HOSTNAME HOSTNAME=%s", + escaped(socks->address)); + log_warn(LD_APP, + "Destination '%s' seems to be an invalid hostname. Failing.", + safe_str_client(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return -1; + } + + /* socks->address is a non-onion hostname or IP address. + * If we can't do any non-onion requests, refuse the connection. + * If we have a hostname but can't do DNS, refuse the connection. + * If we have an IP address, but we can't use that address family, + * refuse the connection. + * + * If we can do DNS requests, and we can use at least one address family, + * then we have to resolve the address first. Then we'll know if it + * resolves to a usable address family. */ + + /* First, check if all non-onion traffic is disabled */ + if (!conn->entry_cfg.dns_request && !conn->entry_cfg.ipv4_traffic + && !conn->entry_cfg.ipv6_traffic) { + log_warn(LD_APP, "Refusing to connect to non-hidden-service hostname " + "or IP address %s because Port has OnionTrafficOnly set (or " + "NoDNSRequest, NoIPv4Traffic, and NoIPv6Traffic).", + safe_str_client(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_ENTRYPOLICY); + return -1; + } + + /* Then check if we have a hostname or IP address, and whether DNS or + * the IP address family are permitted. Reject if not. */ + tor_addr_t dummy_addr; + int socks_family = tor_addr_parse(&dummy_addr, socks->address); + /* family will be -1 for a non-onion hostname that's not an IP */ + if (socks_family == -1) { + if (!conn->entry_cfg.dns_request) { + log_warn(LD_APP, "Refusing to connect to hostname %s " + "because Port has NoDNSRequest set.", + safe_str_client(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_ENTRYPOLICY); + return -1; + } + } else if (socks_family == AF_INET) { + if (!conn->entry_cfg.ipv4_traffic) { + log_warn(LD_APP, "Refusing to connect to IPv4 address %s because " + "Port has NoIPv4Traffic set.", + safe_str_client(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_ENTRYPOLICY); + return -1; + } + } else if (socks_family == AF_INET6) { + if (!conn->entry_cfg.ipv6_traffic) { + log_warn(LD_APP, "Refusing to connect to IPv6 address %s because " + "Port has NoIPv6Traffic set.", + safe_str_client(socks->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_ENTRYPOLICY); + return -1; + } + } else { + tor_assert_nonfatal_unreached_once(); + } + + /* See if this is a hostname lookup that we can answer immediately. + * (For example, an attempt to look up the IP address for an IP address.) + */ + if (socks->command == SOCKS_COMMAND_RESOLVE) { + tor_addr_t answer; + /* Reply to resolves immediately if we can. */ + if (tor_addr_parse(&answer, socks->address) >= 0) {/* is it an IP? */ + /* remember _what_ is supposed to have been resolved. */ + strlcpy(socks->address, rr.orig_address, sizeof(socks->address)); + connection_ap_handshake_socks_resolved_addr(conn, &answer, -1, + map_expires); + connection_mark_unattached_ap(conn, + END_STREAM_REASON_DONE | + END_STREAM_REASON_FLAG_ALREADY_SOCKS_REPLIED); + return 0; + } + tor_assert(!automap); + rep_hist_note_used_resolve(now); /* help predict this next time */ + } else if (socks->command == SOCKS_COMMAND_CONNECT) { + /* Now see if this is a connect request that we can reject immediately */ + + tor_assert(!automap); + /* Don't allow connections to port 0. */ + if (socks->port == 0) { + log_notice(LD_APP,"Application asked to connect to port 0. Refusing."); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return -1; + } + /* You can't make connections to internal addresses, by default. + * Exceptions are begindir requests (where the address is meaningless), + * or cases where you've hand-configured a particular exit, thereby + * making the local address meaningful. */ + if (options->ClientRejectInternalAddresses && + !conn->use_begindir && !conn->chosen_exit_name && !circ) { + /* If we reach this point then we don't want to allow internal + * addresses. Check if we got one. */ + tor_addr_t addr; + if (tor_addr_hostname_is_local(socks->address) || + (tor_addr_parse(&addr, socks->address) >= 0 && + tor_addr_is_internal(&addr, 0))) { + /* If this is an explicit private address with no chosen exit node, + * then we really don't want to try to connect to it. That's + * probably an error. */ + if (conn->is_transparent_ap) { +#define WARN_INTRVL_LOOP 300 + static ratelim_t loop_warn_limit = RATELIM_INIT(WARN_INTRVL_LOOP); + char *m; + if ((m = rate_limit_log(&loop_warn_limit, approx_time()))) { + log_warn(LD_NET, + "Rejecting request for anonymous connection to private " + "address %s on a TransPort or NATDPort. Possible loop " + "in your NAT rules?%s", safe_str_client(socks->address), + m); + tor_free(m); + } + } else { +#define WARN_INTRVL_PRIV 300 + static ratelim_t priv_warn_limit = RATELIM_INIT(WARN_INTRVL_PRIV); + char *m; + if ((m = rate_limit_log(&priv_warn_limit, approx_time()))) { + log_warn(LD_NET, + "Rejecting SOCKS request for anonymous connection to " + "private address %s.%s", + safe_str_client(socks->address),m); + tor_free(m); + } + } + connection_mark_unattached_ap(conn, END_STREAM_REASON_PRIVATE_ADDR); + return -1; + } + } /* end "if we should check for internal addresses" */ + + /* Okay. We're still doing a CONNECT, and it wasn't a private + * address. Here we do special handling for literal IP addresses, + * to see if we should reject this preemptively, and to set up + * fields in conn->entry_cfg to tell the exit what AF we want. */ + { + tor_addr_t addr; + /* XXX Duplicate call to tor_addr_parse. */ + if (tor_addr_parse(&addr, socks->address) >= 0) { + /* If we reach this point, it's an IPv4 or an IPv6 address. */ + sa_family_t family = tor_addr_family(&addr); + + if ((family == AF_INET && ! conn->entry_cfg.ipv4_traffic) || + (family == AF_INET6 && ! conn->entry_cfg.ipv6_traffic)) { + /* You can't do an IPv4 address on a v6-only socks listener, + * or vice versa. */ + log_warn(LD_NET, "Rejecting SOCKS request for an IP address " + "family that this listener does not support."); + connection_mark_unattached_ap(conn, END_STREAM_REASON_ENTRYPOLICY); + return -1; + } else if (family == AF_INET6 && socks->socks_version == 4) { + /* You can't make a socks4 request to an IPv6 address. Socks4 + * doesn't support that. */ + log_warn(LD_NET, "Rejecting SOCKS4 request for an IPv6 address."); + connection_mark_unattached_ap(conn, END_STREAM_REASON_ENTRYPOLICY); + return -1; + } else if (socks->socks_version == 4 && + !conn->entry_cfg.ipv4_traffic) { + /* You can't do any kind of Socks4 request when IPv4 is forbidden. + * + * XXX raise this check outside the enclosing block? */ + log_warn(LD_NET, "Rejecting SOCKS4 request on a listener with " + "no IPv4 traffic supported."); + connection_mark_unattached_ap(conn, END_STREAM_REASON_ENTRYPOLICY); + return -1; + } else if (family == AF_INET6) { + /* Tell the exit: we won't accept any ipv4 connection to an IPv6 + * address. */ + conn->entry_cfg.ipv4_traffic = 0; + } else if (family == AF_INET) { + /* Tell the exit: we won't accept any ipv6 connection to an IPv4 + * address. */ + conn->entry_cfg.ipv6_traffic = 0; + } + } + } + + /* we never allow IPv6 answers on socks4. (TODO: Is this smart?) */ + if (socks->socks_version == 4) + conn->entry_cfg.ipv6_traffic = 0; + + /* Still handling CONNECT. Now, check for exit enclaves. (Which we + * don't do on BEGIN_DIR, or when there is a chosen exit.) + * + * TODO: Should we remove this? Exit enclaves are nutty and don't + * work very well + */ + if (!conn->use_begindir && !conn->chosen_exit_name && !circ) { + /* see if we can find a suitable enclave exit */ + const node_t *r = + router_find_exact_exit_enclave(socks->address, socks->port); + if (r) { + log_info(LD_APP, + "Redirecting address %s to exit at enclave router %s", + safe_str_client(socks->address), node_describe(r)); + /* use the hex digest, not nickname, in case there are two + routers with this nickname */ + conn->chosen_exit_name = + tor_strdup(hex_str(r->identity, DIGEST_LEN)); + conn->chosen_exit_optional = 1; + } + } + + /* Still handling CONNECT: warn or reject if it's using a dangerous + * port. */ + if (!conn->use_begindir && !conn->chosen_exit_name && !circ) + if (consider_plaintext_ports(conn, socks->port) < 0) + return -1; + + /* Remember the port so that we will predict that more requests + there will happen in the future. */ + if (!conn->use_begindir) { + /* help predict this next time */ + rep_hist_note_used_port(now, socks->port); + } + } else if (socks->command == SOCKS_COMMAND_RESOLVE_PTR) { + rep_hist_note_used_resolve(now); /* help predict this next time */ + /* no extra processing needed */ + } else { + /* We should only be doing CONNECT, RESOLVE, or RESOLVE_PTR! */ + tor_fragile_assert(); + } + + /* Okay. At this point we've set chosen_exit_name if needed, rewritten the + * address, and decided not to reject it for any number of reasons. Now + * mark the connection as waiting for a circuit, and try to attach it! + */ + base_conn->state = AP_CONN_STATE_CIRCUIT_WAIT; + + /* If we were given a circuit to attach to, try to attach. Otherwise, + * try to find a good one and attach to that. */ + int rv; + if (circ) { + rv = connection_ap_handshake_attach_chosen_circuit(conn, circ, cpath); + } else { + /* We'll try to attach it at the next event loop, or whenever + * we call connection_ap_attach_pending() */ + connection_ap_mark_as_pending_circuit(conn); + rv = 0; + } + + /* If the above function returned 0 then we're waiting for a circuit. + * if it returned 1, we're attached. Both are okay. But if it returned + * -1, there was an error, so make sure the connection is marked, and + * return -1. */ + if (rv < 0) { + if (!base_conn->marked_for_close) + connection_mark_unattached_ap(conn, END_STREAM_REASON_CANT_ATTACH); + return -1; + } + + return 0; + } else { + /* If we get here, it's a request for a .onion address! */ + tor_assert(addresstype == ONION_V2_HOSTNAME || + addresstype == ONION_V3_HOSTNAME); + tor_assert(!automap); + return connection_ap_handle_onion(conn, socks, circ, addresstype); + } + + return 0; /* unreached but keeps the compiler happy */ +} + +#ifdef TRANS_PF +static int pf_socket = -1; +int +get_pf_socket(void) +{ + int pf; + /* This should be opened before dropping privileges. */ + if (pf_socket >= 0) + return pf_socket; + +#if defined(OpenBSD) + /* only works on OpenBSD */ + pf = tor_open_cloexec("/dev/pf", O_RDONLY, 0); +#else + /* works on NetBSD and FreeBSD */ + pf = tor_open_cloexec("/dev/pf", O_RDWR, 0); +#endif /* defined(OpenBSD) */ + + if (pf < 0) { + log_warn(LD_NET, "open("/dev/pf") failed: %s", strerror(errno)); + return -1; + } + + pf_socket = pf; + return pf_socket; +} +#endif /* defined(TRANS_PF) */ + +#if defined(TRANS_NETFILTER) || defined(TRANS_PF) || \ + defined(TRANS_TPROXY) +/** Try fill in the address of <b>req</b> from the socket configured + * with <b>conn</b>. */ +static int +destination_from_socket(entry_connection_t *conn, socks_request_t *req) +{ + struct sockaddr_storage orig_dst; + socklen_t orig_dst_len = sizeof(orig_dst); + tor_addr_t addr; + +#ifdef TRANS_TPROXY + if (get_options()->TransProxyType_parsed == TPT_TPROXY) { + if (getsockname(ENTRY_TO_CONN(conn)->s, (struct sockaddr*)&orig_dst, + &orig_dst_len) < 0) { + int e = tor_socket_errno(ENTRY_TO_CONN(conn)->s); + log_warn(LD_NET, "getsockname() failed: %s", tor_socket_strerror(e)); + return -1; + } + goto done; + } +#endif /* defined(TRANS_TPROXY) */ + +#ifdef TRANS_NETFILTER + int rv = -1; + switch (ENTRY_TO_CONN(conn)->socket_family) { +#ifdef TRANS_NETFILTER_IPV4 + case AF_INET: + rv = getsockopt(ENTRY_TO_CONN(conn)->s, SOL_IP, SO_ORIGINAL_DST, + (struct sockaddr*)&orig_dst, &orig_dst_len); + break; +#endif /* defined(TRANS_NETFILTER_IPV4) */ +#ifdef TRANS_NETFILTER_IPV6 + case AF_INET6: + rv = getsockopt(ENTRY_TO_CONN(conn)->s, SOL_IPV6, IP6T_SO_ORIGINAL_DST, + (struct sockaddr*)&orig_dst, &orig_dst_len); + break; +#endif /* defined(TRANS_NETFILTER_IPV6) */ + default: + log_warn(LD_BUG, + "Received transparent data from an unsuported socket family %d", + ENTRY_TO_CONN(conn)->socket_family); + return -1; + } + if (rv < 0) { + int e = tor_socket_errno(ENTRY_TO_CONN(conn)->s); + log_warn(LD_NET, "getsockopt() failed: %s", tor_socket_strerror(e)); + return -1; + } + goto done; +#elif defined(TRANS_PF) + if (getsockname(ENTRY_TO_CONN(conn)->s, (struct sockaddr*)&orig_dst, + &orig_dst_len) < 0) { + int e = tor_socket_errno(ENTRY_TO_CONN(conn)->s); + log_warn(LD_NET, "getsockname() failed: %s", tor_socket_strerror(e)); + return -1; + } + goto done; +#else + (void)conn; + (void)req; + log_warn(LD_BUG, "Unable to determine destination from socket."); + return -1; +#endif /* defined(TRANS_NETFILTER) || ... */ + + done: + tor_addr_from_sockaddr(&addr, (struct sockaddr*)&orig_dst, &req->port); + tor_addr_to_str(req->address, &addr, sizeof(req->address), 1); + + return 0; +} +#endif /* defined(TRANS_NETFILTER) || defined(TRANS_PF) || ... */ + +#ifdef TRANS_PF +static int +destination_from_pf(entry_connection_t *conn, socks_request_t *req) +{ + struct sockaddr_storage proxy_addr; + socklen_t proxy_addr_len = sizeof(proxy_addr); + struct sockaddr *proxy_sa = (struct sockaddr*) &proxy_addr; + struct pfioc_natlook pnl; + tor_addr_t addr; + int pf = -1; + + if (getsockname(ENTRY_TO_CONN(conn)->s, (struct sockaddr*)&proxy_addr, + &proxy_addr_len) < 0) { + int e = tor_socket_errno(ENTRY_TO_CONN(conn)->s); + log_warn(LD_NET, "getsockname() to determine transocks destination " + "failed: %s", tor_socket_strerror(e)); + return -1; + } + +#ifdef __FreeBSD__ + if (get_options()->TransProxyType_parsed == TPT_IPFW) { + /* ipfw(8) is used and in this case getsockname returned the original + destination */ + if (tor_addr_from_sockaddr(&addr, proxy_sa, &req->port) < 0) { + tor_fragile_assert(); + return -1; + } + + tor_addr_to_str(req->address, &addr, sizeof(req->address), 0); + + return 0; + } +#endif /* defined(__FreeBSD__) */ + + memset(&pnl, 0, sizeof(pnl)); + pnl.proto = IPPROTO_TCP; + pnl.direction = PF_OUT; + if (proxy_sa->sa_family == AF_INET) { + struct sockaddr_in *sin = (struct sockaddr_in *)proxy_sa; + pnl.af = AF_INET; + pnl.saddr.v4.s_addr = tor_addr_to_ipv4n(&ENTRY_TO_CONN(conn)->addr); + pnl.sport = htons(ENTRY_TO_CONN(conn)->port); + pnl.daddr.v4.s_addr = sin->sin_addr.s_addr; + pnl.dport = sin->sin_port; + } else if (proxy_sa->sa_family == AF_INET6) { + struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)proxy_sa; + pnl.af = AF_INET6; + memcpy(&pnl.saddr.v6, tor_addr_to_in6(&ENTRY_TO_CONN(conn)->addr), + sizeof(struct in6_addr)); + pnl.sport = htons(ENTRY_TO_CONN(conn)->port); + memcpy(&pnl.daddr.v6, &sin6->sin6_addr, sizeof(struct in6_addr)); + pnl.dport = sin6->sin6_port; + } else { + log_warn(LD_NET, "getsockname() gave an unexpected address family (%d)", + (int)proxy_sa->sa_family); + return -1; + } + + pf = get_pf_socket(); + if (pf<0) + return -1; + + if (ioctl(pf, DIOCNATLOOK, &pnl) < 0) { + log_warn(LD_NET, "ioctl(DIOCNATLOOK) failed: %s", strerror(errno)); + return -1; + } + + if (pnl.af == AF_INET) { + tor_addr_from_ipv4n(&addr, pnl.rdaddr.v4.s_addr); + } else if (pnl.af == AF_INET6) { + tor_addr_from_in6(&addr, &pnl.rdaddr.v6); + } else { + tor_fragile_assert(); + return -1; + } + + tor_addr_to_str(req->address, &addr, sizeof(req->address), 1); + req->port = ntohs(pnl.rdport); + + return 0; +} +#endif /* defined(TRANS_PF) */ + +/** Fetch the original destination address and port from a + * system-specific interface and put them into a + * socks_request_t as if they came from a socks request. + * + * Return -1 if an error prevents fetching the destination, + * else return 0. + */ +static int +connection_ap_get_original_destination(entry_connection_t *conn, + socks_request_t *req) +{ +#ifdef TRANS_NETFILTER + return destination_from_socket(conn, req); +#elif defined(TRANS_PF) + const or_options_t *options = get_options(); + + if (options->TransProxyType_parsed == TPT_PF_DIVERT) + return destination_from_socket(conn, req); + + if (options->TransProxyType_parsed == TPT_DEFAULT || + options->TransProxyType_parsed == TPT_IPFW) + return destination_from_pf(conn, req); + + (void)conn; + (void)req; + log_warn(LD_BUG, "Proxy destination determination mechanism %s unknown.", + options->TransProxyType); + return -1; +#else + (void)conn; + (void)req; + log_warn(LD_BUG, "Called connection_ap_get_original_destination, but no " + "transparent proxy method was configured."); + return -1; +#endif /* defined(TRANS_NETFILTER) || ... */ +} + +/** connection_edge_process_inbuf() found a conn in state + * socks_wait. See if conn->inbuf has the right bytes to proceed with + * the socks handshake. + * + * If the handshake is complete, send it to + * connection_ap_handshake_rewrite_and_attach(). + * + * Return -1 if an unexpected error with conn occurs (and mark it for close), + * else return 0. + */ +static int +connection_ap_handshake_process_socks(entry_connection_t *conn) +{ + socks_request_t *socks; + int sockshere; + const or_options_t *options = get_options(); + int had_reply = 0; + connection_t *base_conn = ENTRY_TO_CONN(conn); + + tor_assert(conn); + tor_assert(base_conn->type == CONN_TYPE_AP); + tor_assert(base_conn->state == AP_CONN_STATE_SOCKS_WAIT); + tor_assert(conn->socks_request); + socks = conn->socks_request; + + log_debug(LD_APP,"entered."); + + sockshere = fetch_from_buf_socks(base_conn->inbuf, socks, + options->TestSocks, options->SafeSocks); + + if (socks->replylen) { + had_reply = 1; + connection_buf_add((const char*)socks->reply, socks->replylen, + base_conn); + socks->replylen = 0; + if (sockshere == -1) { + /* An invalid request just got a reply, no additional + * one is necessary. */ + socks->has_finished = 1; + } + } + + if (sockshere == 0) { + log_debug(LD_APP,"socks handshake not all here yet."); + return 0; + } else if (sockshere == -1) { + if (!had_reply) { + log_warn(LD_APP,"Fetching socks handshake failed. Closing."); + connection_ap_handshake_socks_reply(conn, NULL, 0, + END_STREAM_REASON_SOCKSPROTOCOL); + } + connection_mark_unattached_ap(conn, + END_STREAM_REASON_SOCKSPROTOCOL | + END_STREAM_REASON_FLAG_ALREADY_SOCKS_REPLIED); + return -1; + } /* else socks handshake is done, continue processing */ + + if (SOCKS_COMMAND_IS_CONNECT(socks->command)) + control_event_stream_status(conn, STREAM_EVENT_NEW, 0); + else + control_event_stream_status(conn, STREAM_EVENT_NEW_RESOLVE, 0); + + return connection_ap_rewrite_and_attach_if_allowed(conn, NULL, NULL); +} + +/** connection_init_accepted_conn() found a new trans AP conn. + * Get the original destination and send it to + * connection_ap_handshake_rewrite_and_attach(). + * + * Return -1 if an unexpected error with conn (and it should be marked + * for close), else return 0. + */ +int +connection_ap_process_transparent(entry_connection_t *conn) +{ + socks_request_t *socks; + + tor_assert(conn); + tor_assert(conn->socks_request); + socks = conn->socks_request; + + /* pretend that a socks handshake completed so we don't try to + * send a socks reply down a transparent conn */ + socks->command = SOCKS_COMMAND_CONNECT; + socks->has_finished = 1; + + log_debug(LD_APP,"entered."); + + if (connection_ap_get_original_destination(conn, socks) < 0) { + log_warn(LD_APP,"Fetching original destination failed. Closing."); + connection_mark_unattached_ap(conn, + END_STREAM_REASON_CANT_FETCH_ORIG_DEST); + return -1; + } + /* we have the original destination */ + + control_event_stream_status(conn, STREAM_EVENT_NEW, 0); + + return connection_ap_rewrite_and_attach_if_allowed(conn, NULL, NULL); +} + +/** connection_edge_process_inbuf() found a conn in state natd_wait. See if + * conn->inbuf has the right bytes to proceed. See FreeBSD's libalias(3) and + * ProxyEncodeTcpStream() in src/lib/libalias/alias_proxy.c for the encoding + * form of the original destination. + * + * If the original destination is complete, send it to + * connection_ap_handshake_rewrite_and_attach(). + * + * Return -1 if an unexpected error with conn (and it should be marked + * for close), else return 0. + */ +static int +connection_ap_process_natd(entry_connection_t *conn) +{ + char tmp_buf[36], *tbuf, *daddr; + size_t tlen = 30; + int err, port_ok; + socks_request_t *socks; + + tor_assert(conn); + tor_assert(ENTRY_TO_CONN(conn)->state == AP_CONN_STATE_NATD_WAIT); + tor_assert(conn->socks_request); + socks = conn->socks_request; + + log_debug(LD_APP,"entered."); + + /* look for LF-terminated "[DEST ip_addr port]" + * where ip_addr is a dotted-quad and port is in string form */ + err = connection_buf_get_line(ENTRY_TO_CONN(conn), tmp_buf, &tlen); + if (err == 0) + return 0; + if (err < 0) { + log_warn(LD_APP,"NATD handshake failed (DEST too long). Closing"); + connection_mark_unattached_ap(conn, END_STREAM_REASON_INVALID_NATD_DEST); + return -1; + } + + if (strcmpstart(tmp_buf, "[DEST ")) { + log_warn(LD_APP,"NATD handshake was ill-formed; closing. The client " + "said: %s", + escaped(tmp_buf)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_INVALID_NATD_DEST); + return -1; + } + + daddr = tbuf = &tmp_buf[0] + 6; /* after end of "[DEST " */ + if (!(tbuf = strchr(tbuf, ' '))) { + log_warn(LD_APP,"NATD handshake was ill-formed; closing. The client " + "said: %s", + escaped(tmp_buf)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_INVALID_NATD_DEST); + return -1; + } + *tbuf++ = '\0'; + + /* pretend that a socks handshake completed so we don't try to + * send a socks reply down a natd conn */ + strlcpy(socks->address, daddr, sizeof(socks->address)); + socks->port = (uint16_t) + tor_parse_long(tbuf, 10, 1, 65535, &port_ok, &daddr); + if (!port_ok) { + log_warn(LD_APP,"NATD handshake failed; port %s is ill-formed or out " + "of range.", escaped(tbuf)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_INVALID_NATD_DEST); + return -1; + } + + socks->command = SOCKS_COMMAND_CONNECT; + socks->has_finished = 1; + + control_event_stream_status(conn, STREAM_EVENT_NEW, 0); + + ENTRY_TO_CONN(conn)->state = AP_CONN_STATE_CIRCUIT_WAIT; + + return connection_ap_rewrite_and_attach_if_allowed(conn, NULL, NULL); +} + +/** Called on an HTTP CONNECT entry connection when some bytes have arrived, + * but we have not yet received a full HTTP CONNECT request. Try to parse an + * HTTP CONNECT request from the connection's inbuf. On success, set up the + * connection's socks_request field and try to attach the connection. On + * failure, send an HTTP reply, and mark the connection. + */ +STATIC int +connection_ap_process_http_connect(entry_connection_t *conn) +{ + if (BUG(ENTRY_TO_CONN(conn)->state != AP_CONN_STATE_HTTP_CONNECT_WAIT)) + return -1; + + char *headers = NULL, *body = NULL; + char *command = NULL, *addrport = NULL; + char *addr = NULL; + size_t bodylen = 0; + + const char *errmsg = NULL; + int rv = 0; + + const int http_status = + fetch_from_buf_http(ENTRY_TO_CONN(conn)->inbuf, &headers, 8192, + &body, &bodylen, 1024, 0); + if (http_status < 0) { + /* Bad http status */ + errmsg = "HTTP/1.0 400 Bad Request\r\n\r\n"; + goto err; + } else if (http_status == 0) { + /* no HTTP request yet. */ + goto done; + } + + const int cmd_status = parse_http_command(headers, &command, &addrport); + if (cmd_status < 0) { + errmsg = "HTTP/1.0 400 Bad Request\r\n\r\n"; + goto err; + } + tor_assert(command); + tor_assert(addrport); + if (strcasecmp(command, "connect")) { + errmsg = "HTTP/1.0 405 Method Not Allowed\r\n\r\n"; + goto err; + } + + tor_assert(conn->socks_request); + socks_request_t *socks = conn->socks_request; + uint16_t port; + if (tor_addr_port_split(LOG_WARN, addrport, &addr, &port) < 0) { + errmsg = "HTTP/1.0 400 Bad Request\r\n\r\n"; + goto err; + } + if (strlen(addr) >= MAX_SOCKS_ADDR_LEN) { + errmsg = "HTTP/1.0 414 Request-URI Too Long\r\n\r\n"; + goto err; + } + + /* Abuse the 'username' and 'password' fields here. They are already an + * abuse. */ + { + char *authorization = http_get_header(headers, "Proxy-Authorization: "); + if (authorization) { + socks->username = authorization; // steal reference + socks->usernamelen = strlen(authorization); + } + char *isolation = http_get_header(headers, "X-Tor-Stream-Isolation: "); + if (isolation) { + socks->password = isolation; // steal reference + socks->passwordlen = strlen(isolation); + } + } + + socks->command = SOCKS_COMMAND_CONNECT; + socks->listener_type = CONN_TYPE_AP_HTTP_CONNECT_LISTENER; + strlcpy(socks->address, addr, sizeof(socks->address)); + socks->port = port; + + control_event_stream_status(conn, STREAM_EVENT_NEW, 0); + + rv = connection_ap_rewrite_and_attach_if_allowed(conn, NULL, NULL); + + // XXXX send a "100 Continue" message? + + goto done; + + err: + if (BUG(errmsg == NULL)) + errmsg = "HTTP/1.0 400 Bad Request\r\n\r\n"; + log_warn(LD_EDGE, "Saying %s", escaped(errmsg)); + connection_buf_add(errmsg, strlen(errmsg), ENTRY_TO_CONN(conn)); + connection_mark_unattached_ap(conn, + END_STREAM_REASON_HTTPPROTOCOL| + END_STREAM_REASON_FLAG_ALREADY_SOCKS_REPLIED); + + done: + tor_free(headers); + tor_free(body); + tor_free(command); + tor_free(addrport); + tor_free(addr); + return rv; +} + +/** Iterate over the two bytes of stream_id until we get one that is not + * already in use; return it. Return 0 if can't get a unique stream_id. + */ +streamid_t +get_unique_stream_id_by_circ(origin_circuit_t *circ) +{ + edge_connection_t *tmpconn; + streamid_t test_stream_id; + uint32_t attempts=0; + + again: + test_stream_id = circ->next_stream_id++; + if (++attempts > 1<<16) { + /* Make sure we don't loop forever if all stream_id's are used. */ + log_warn(LD_APP,"No unused stream IDs. Failing."); + return 0; + } + if (test_stream_id == 0) + goto again; + for (tmpconn = circ->p_streams; tmpconn; tmpconn=tmpconn->next_stream) + if (tmpconn->stream_id == test_stream_id) + goto again; ++ ++ if (connection_half_edge_find_stream_id(circ->half_streams, ++ test_stream_id)) ++ goto again; ++ + return test_stream_id; +} + +/** Return true iff <b>conn</b> is linked to a circuit and configured to use + * an exit that supports optimistic data. */ +static int +connection_ap_supports_optimistic_data(const entry_connection_t *conn) +{ + const edge_connection_t *edge_conn = ENTRY_TO_EDGE_CONN(conn); + /* We can only send optimistic data if we're connected to an open + general circuit. */ + if (edge_conn->on_circuit == NULL || + edge_conn->on_circuit->state != CIRCUIT_STATE_OPEN || + (edge_conn->on_circuit->purpose != CIRCUIT_PURPOSE_C_GENERAL && + edge_conn->on_circuit->purpose != CIRCUIT_PURPOSE_C_HSDIR_GET && + edge_conn->on_circuit->purpose != CIRCUIT_PURPOSE_S_HSDIR_POST && + edge_conn->on_circuit->purpose != CIRCUIT_PURPOSE_C_REND_JOINED)) + return 0; + + return conn->may_use_optimistic_data; +} + +/** Return a bitmask of BEGIN_FLAG_* flags that we should transmit in the + * RELAY_BEGIN cell for <b>ap_conn</b>. */ +static uint32_t +connection_ap_get_begincell_flags(entry_connection_t *ap_conn) +{ + edge_connection_t *edge_conn = ENTRY_TO_EDGE_CONN(ap_conn); + const node_t *exitnode = NULL; + const crypt_path_t *cpath_layer = edge_conn->cpath_layer; + uint32_t flags = 0; + + /* No flags for begindir */ + if (ap_conn->use_begindir) + return 0; + + /* No flags for hidden services. */ + if (edge_conn->on_circuit->purpose != CIRCUIT_PURPOSE_C_GENERAL) + return 0; + + /* If only IPv4 is supported, no flags */ + if (ap_conn->entry_cfg.ipv4_traffic && !ap_conn->entry_cfg.ipv6_traffic) + return 0; + + if (! cpath_layer || + ! cpath_layer->extend_info) + return 0; + + if (!ap_conn->entry_cfg.ipv4_traffic) + flags |= BEGIN_FLAG_IPV4_NOT_OK; + + exitnode = node_get_by_id(cpath_layer->extend_info->identity_digest); + + if (ap_conn->entry_cfg.ipv6_traffic && exitnode) { + tor_addr_t a; + tor_addr_make_null(&a, AF_INET6); + if (compare_tor_addr_to_node_policy(&a, ap_conn->socks_request->port, + exitnode) + != ADDR_POLICY_REJECTED) { + /* Only say "IPv6 OK" if the exit node supports IPv6. Otherwise there's + * no point. */ + flags |= BEGIN_FLAG_IPV6_OK; + } + } + + if (flags == BEGIN_FLAG_IPV6_OK) { + /* When IPv4 and IPv6 are both allowed, consider whether to say we + * prefer IPv6. Otherwise there's no point in declaring a preference */ + if (ap_conn->entry_cfg.prefer_ipv6) + flags |= BEGIN_FLAG_IPV6_PREFERRED; + } + + if (flags == BEGIN_FLAG_IPV4_NOT_OK) { + log_warn(LD_EDGE, "I'm about to ask a node for a connection that I " + "am telling it to fulfil with neither IPv4 nor IPv6. That's " + "not going to work. Did you perhaps ask for an IPv6 address " + "on an IPv4Only port, or vice versa?"); + } + + return flags; +} + +/** Write a relay begin cell, using destaddr and destport from ap_conn's + * socks_request field, and send it down circ. + * + * If ap_conn is broken, mark it for close and return -1. Else return 0. + */ +MOCK_IMPL(int, +connection_ap_handshake_send_begin,(entry_connection_t *ap_conn)) +{ + char payload[CELL_PAYLOAD_SIZE]; + int payload_len; + int begin_type; + const or_options_t *options = get_options(); + origin_circuit_t *circ; + edge_connection_t *edge_conn = ENTRY_TO_EDGE_CONN(ap_conn); + connection_t *base_conn = TO_CONN(edge_conn); + tor_assert(edge_conn->on_circuit); + circ = TO_ORIGIN_CIRCUIT(edge_conn->on_circuit); + + tor_assert(base_conn->type == CONN_TYPE_AP); + tor_assert(base_conn->state == AP_CONN_STATE_CIRCUIT_WAIT); + tor_assert(ap_conn->socks_request); + tor_assert(SOCKS_COMMAND_IS_CONNECT(ap_conn->socks_request->command)); + + edge_conn->stream_id = get_unique_stream_id_by_circ(circ); + if (edge_conn->stream_id==0) { + /* XXXX+ Instead of closing this stream, we should make it get + * retried on another circuit. */ + connection_mark_unattached_ap(ap_conn, END_STREAM_REASON_INTERNAL); + + /* Mark this circuit "unusable for new streams". */ + mark_circuit_unusable_for_new_conns(circ); + return -1; + } + + /* Set up begin cell flags. */ + edge_conn->begincell_flags = connection_ap_get_begincell_flags(ap_conn); + + tor_snprintf(payload,RELAY_PAYLOAD_SIZE, "%s:%d", + (circ->base_.purpose == CIRCUIT_PURPOSE_C_GENERAL) ? + ap_conn->socks_request->address : "", + ap_conn->socks_request->port); + payload_len = (int)strlen(payload)+1; + if (payload_len <= RELAY_PAYLOAD_SIZE - 4 && edge_conn->begincell_flags) { + set_uint32(payload + payload_len, htonl(edge_conn->begincell_flags)); + payload_len += 4; + } + + log_info(LD_APP, + "Sending relay cell %d on circ %u to begin stream %d.", + (int)ap_conn->use_begindir, + (unsigned)circ->base_.n_circ_id, + edge_conn->stream_id); + + begin_type = ap_conn->use_begindir ? + RELAY_COMMAND_BEGIN_DIR : RELAY_COMMAND_BEGIN; + + /* Check that circuits are anonymised, based on their type. */ + if (begin_type == RELAY_COMMAND_BEGIN) { + /* This connection is a standard OR connection. + * Make sure its path length is anonymous, or that we're in a + * non-anonymous mode. */ + assert_circ_anonymity_ok(circ, options); + } else if (begin_type == RELAY_COMMAND_BEGIN_DIR) { + /* This connection is a begindir directory connection. + * Look at the linked directory connection to access the directory purpose. + * If a BEGINDIR connection is ever not linked, that's a bug. */ + if (BUG(!base_conn->linked)) { + return -1; + } + connection_t *linked_dir_conn_base = base_conn->linked_conn; + /* If the linked connection has been unlinked by other code, we can't send + * a begin cell on it. */ + if (!linked_dir_conn_base) { + return -1; + } + /* Sensitive directory connections must have an anonymous path length. + * Otherwise, directory connections are typically one-hop. + * This matches the earlier check for directory connection path anonymity + * in directory_initiate_request(). */ + if (purpose_needs_anonymity(linked_dir_conn_base->purpose, + TO_DIR_CONN(linked_dir_conn_base)->router_purpose, + TO_DIR_CONN(linked_dir_conn_base)->requested_resource)) { + assert_circ_anonymity_ok(circ, options); + } + } else { + /* This code was written for the two connection types BEGIN and BEGIN_DIR + */ + tor_assert_unreached(); + } + + if (connection_edge_send_command(edge_conn, begin_type, + begin_type == RELAY_COMMAND_BEGIN ? payload : NULL, + begin_type == RELAY_COMMAND_BEGIN ? payload_len : 0) < 0) + return -1; /* circuit is closed, don't continue */ + + edge_conn->package_window = STREAMWINDOW_START; + edge_conn->deliver_window = STREAMWINDOW_START; + base_conn->state = AP_CONN_STATE_CONNECT_WAIT; + log_info(LD_APP,"Address/port sent, ap socket "TOR_SOCKET_T_FORMAT + ", n_circ_id %u", + base_conn->s, (unsigned)circ->base_.n_circ_id); + control_event_stream_status(ap_conn, STREAM_EVENT_SENT_CONNECT, 0); + + /* If there's queued-up data, send it now */ + if ((connection_get_inbuf_len(base_conn) || + ap_conn->sending_optimistic_data) && + connection_ap_supports_optimistic_data(ap_conn)) { + log_info(LD_APP, "Sending up to %ld + %ld bytes of queued-up data", + (long)connection_get_inbuf_len(base_conn), + ap_conn->sending_optimistic_data ? + (long)buf_datalen(ap_conn->sending_optimistic_data) : 0); + if (connection_edge_package_raw_inbuf(edge_conn, 1, NULL) < 0) { + connection_mark_for_close(base_conn); + } + } + + return 0; +} + +/** Write a relay resolve cell, using destaddr and destport from ap_conn's + * socks_request field, and send it down circ. + * + * If ap_conn is broken, mark it for close and return -1. Else return 0. + */ +int +connection_ap_handshake_send_resolve(entry_connection_t *ap_conn) +{ + int payload_len, command; + const char *string_addr; + char inaddr_buf[REVERSE_LOOKUP_NAME_BUF_LEN]; + origin_circuit_t *circ; + edge_connection_t *edge_conn = ENTRY_TO_EDGE_CONN(ap_conn); + connection_t *base_conn = TO_CONN(edge_conn); + tor_assert(edge_conn->on_circuit); + circ = TO_ORIGIN_CIRCUIT(edge_conn->on_circuit); + + tor_assert(base_conn->type == CONN_TYPE_AP); + tor_assert(base_conn->state == AP_CONN_STATE_CIRCUIT_WAIT); + tor_assert(ap_conn->socks_request); + tor_assert(circ->base_.purpose == CIRCUIT_PURPOSE_C_GENERAL); + + command = ap_conn->socks_request->command; + tor_assert(SOCKS_COMMAND_IS_RESOLVE(command)); + + edge_conn->stream_id = get_unique_stream_id_by_circ(circ); + if (edge_conn->stream_id==0) { + /* XXXX+ Instead of closing this stream, we should make it get + * retried on another circuit. */ + connection_mark_unattached_ap(ap_conn, END_STREAM_REASON_INTERNAL); + + /* Mark this circuit "unusable for new streams". */ + mark_circuit_unusable_for_new_conns(circ); + return -1; + } + + if (command == SOCKS_COMMAND_RESOLVE) { + string_addr = ap_conn->socks_request->address; + payload_len = (int)strlen(string_addr)+1; + } else { + /* command == SOCKS_COMMAND_RESOLVE_PTR */ + const char *a = ap_conn->socks_request->address; + tor_addr_t addr; + int r; + + /* We're doing a reverse lookup. The input could be an IP address, or + * could be an .in-addr.arpa or .ip6.arpa address */ + r = tor_addr_parse_PTR_name(&addr, a, AF_UNSPEC, 1); + if (r <= 0) { + log_warn(LD_APP, "Rejecting ill-formed reverse lookup of %s", + safe_str_client(a)); + connection_mark_unattached_ap(ap_conn, END_STREAM_REASON_INTERNAL); + return -1; + } + + r = tor_addr_to_PTR_name(inaddr_buf, sizeof(inaddr_buf), &addr); + if (r < 0) { + log_warn(LD_BUG, "Couldn't generate reverse lookup hostname of %s", + safe_str_client(a)); + connection_mark_unattached_ap(ap_conn, END_STREAM_REASON_INTERNAL); + return -1; + } + + string_addr = inaddr_buf; + payload_len = (int)strlen(inaddr_buf)+1; + tor_assert(payload_len <= (int)sizeof(inaddr_buf)); + } + + log_debug(LD_APP, + "Sending relay cell to begin stream %d.", edge_conn->stream_id); + + if (connection_edge_send_command(edge_conn, + RELAY_COMMAND_RESOLVE, + string_addr, payload_len) < 0) + return -1; /* circuit is closed, don't continue */ + + if (!base_conn->address) { + /* This might be unnecessary. XXXX */ + base_conn->address = tor_addr_to_str_dup(&base_conn->addr); + } + base_conn->state = AP_CONN_STATE_RESOLVE_WAIT; + log_info(LD_APP,"Address sent for resolve, ap socket "TOR_SOCKET_T_FORMAT + ", n_circ_id %u", + base_conn->s, (unsigned)circ->base_.n_circ_id); + control_event_stream_status(ap_conn, STREAM_EVENT_SENT_RESOLVE, 0); + return 0; +} + +/** Make an AP connection_t linked to the connection_t <b>partner</b>. make a + * new linked connection pair, and attach one side to the conn, connection_add + * it, initialize it to circuit_wait, and call + * connection_ap_handshake_attach_circuit(conn) on it. + * + * Return the newly created end of the linked connection pair, or -1 if error. + */ +entry_connection_t * +connection_ap_make_link(connection_t *partner, + char *address, uint16_t port, + const char *digest, + int session_group, int isolation_flags, + int use_begindir, int want_onehop) +{ + entry_connection_t *conn; + connection_t *base_conn; + + log_info(LD_APP,"Making internal %s tunnel to %s:%d ...", + want_onehop ? "direct" : "anonymized", + safe_str_client(address), port); + + conn = entry_connection_new(CONN_TYPE_AP, tor_addr_family(&partner->addr)); + base_conn = ENTRY_TO_CONN(conn); + base_conn->linked = 1; /* so that we can add it safely below. */ + + /* populate conn->socks_request */ + + /* leave version at zero, so the socks_reply is empty */ + conn->socks_request->socks_version = 0; + conn->socks_request->has_finished = 0; /* waiting for 'connected' */ + strlcpy(conn->socks_request->address, address, + sizeof(conn->socks_request->address)); + conn->socks_request->port = port; + conn->socks_request->command = SOCKS_COMMAND_CONNECT; + conn->want_onehop = want_onehop; + conn->use_begindir = use_begindir; + if (use_begindir) { + conn->chosen_exit_name = tor_malloc(HEX_DIGEST_LEN+2); + conn->chosen_exit_name[0] = '$'; + tor_assert(digest); + base16_encode(conn->chosen_exit_name+1,HEX_DIGEST_LEN+1, + digest, DIGEST_LEN); + } + + /* Populate isolation fields. */ + conn->socks_request->listener_type = CONN_TYPE_DIR_LISTENER; + conn->original_dest_address = tor_strdup(address); + conn->entry_cfg.session_group = session_group; + conn->entry_cfg.isolation_flags = isolation_flags; + + base_conn->address = tor_strdup("(Tor_internal)"); + tor_addr_make_unspec(&base_conn->addr); + base_conn->port = 0; + + connection_link_connections(partner, base_conn); + + if (connection_add(base_conn) < 0) { /* no space, forget it */ + connection_free(base_conn); + return NULL; + } + + base_conn->state = AP_CONN_STATE_CIRCUIT_WAIT; + + control_event_stream_status(conn, STREAM_EVENT_NEW, 0); + + /* attaching to a dirty circuit is fine */ + connection_ap_mark_as_pending_circuit(conn); + log_info(LD_APP,"... application connection created and linked."); + return conn; +} + +/** Notify any interested controller connections about a new hostname resolve + * or resolve error. Takes the same arguments as does + * connection_ap_handshake_socks_resolved(). */ +static void +tell_controller_about_resolved_result(entry_connection_t *conn, + int answer_type, + size_t answer_len, + const char *answer, + int ttl, + time_t expires) +{ + expires = time(NULL) + ttl; + if (answer_type == RESOLVED_TYPE_IPV4 && answer_len >= 4) { + char *cp = tor_dup_ip(ntohl(get_uint32(answer))); + control_event_address_mapped(conn->socks_request->address, + cp, expires, NULL, 0); + tor_free(cp); + } else if (answer_type == RESOLVED_TYPE_HOSTNAME && answer_len < 256) { + char *cp = tor_strndup(answer, answer_len); + control_event_address_mapped(conn->socks_request->address, + cp, expires, NULL, 0); + tor_free(cp); + } else { + control_event_address_mapped(conn->socks_request->address, + "<error>", time(NULL)+ttl, + "error=yes", 0); + } +} + +/** + * As connection_ap_handshake_socks_resolved, but take a tor_addr_t to send + * as the answer. + */ +void +connection_ap_handshake_socks_resolved_addr(entry_connection_t *conn, + const tor_addr_t *answer, + int ttl, + time_t expires) +{ + if (tor_addr_family(answer) == AF_INET) { + uint32_t a = tor_addr_to_ipv4n(answer); /* network order */ + connection_ap_handshake_socks_resolved(conn,RESOLVED_TYPE_IPV4,4, + (uint8_t*)&a, + ttl, expires); + } else if (tor_addr_family(answer) == AF_INET6) { + const uint8_t *a = tor_addr_to_in6_addr8(answer); + connection_ap_handshake_socks_resolved(conn,RESOLVED_TYPE_IPV6,16, + a, + ttl, expires); + } else { + log_warn(LD_BUG, "Got called with address of unexpected family %d", + tor_addr_family(answer)); + connection_ap_handshake_socks_resolved(conn, + RESOLVED_TYPE_ERROR,0,NULL,-1,-1); + } +} + +/** Send an answer to an AP connection that has requested a DNS lookup via + * SOCKS. The type should be one of RESOLVED_TYPE_(IPV4|IPV6|HOSTNAME) or -1 + * for unreachable; the answer should be in the format specified in the socks + * extensions document. <b>ttl</b> is the ttl for the answer, or -1 on + * certain errors or for values that didn't come via DNS. <b>expires</b> is + * a time when the answer expires, or -1 or TIME_MAX if there's a good TTL. + **/ +/* XXXX the use of the ttl and expires fields is nutty. Let's make this + * interface and those that use it less ugly. */ +MOCK_IMPL(void, +connection_ap_handshake_socks_resolved,(entry_connection_t *conn, + int answer_type, + size_t answer_len, + const uint8_t *answer, + int ttl, + time_t expires)) +{ + char buf[384]; + size_t replylen; + + if (ttl >= 0) { + if (answer_type == RESOLVED_TYPE_IPV4 && answer_len == 4) { + tor_addr_t a; + tor_addr_from_ipv4n(&a, get_uint32(answer)); + if (! tor_addr_is_null(&a)) { + client_dns_set_addressmap(conn, + conn->socks_request->address, &a, + conn->chosen_exit_name, ttl); + } + } else if (answer_type == RESOLVED_TYPE_IPV6 && answer_len == 16) { + tor_addr_t a; + tor_addr_from_ipv6_bytes(&a, (char*)answer); + if (! tor_addr_is_null(&a)) { + client_dns_set_addressmap(conn, + conn->socks_request->address, &a, + conn->chosen_exit_name, ttl); + } + } else if (answer_type == RESOLVED_TYPE_HOSTNAME && answer_len < 256) { + char *cp = tor_strndup((char*)answer, answer_len); + client_dns_set_reverse_addressmap(conn, + conn->socks_request->address, + cp, + conn->chosen_exit_name, ttl); + tor_free(cp); + } + } + + if (ENTRY_TO_EDGE_CONN(conn)->is_dns_request) { + if (conn->dns_server_request) { + /* We had a request on our DNS port: answer it. */ + dnsserv_resolved(conn, answer_type, answer_len, (char*)answer, ttl); + conn->socks_request->has_finished = 1; + return; + } else { + /* This must be a request from the controller. Since answers to those + * requests are not cached, they do not generate an ADDRMAP event on + * their own. */ + tell_controller_about_resolved_result(conn, answer_type, answer_len, + (char*)answer, ttl, expires); + conn->socks_request->has_finished = 1; + return; + } + /* We shouldn't need to free conn here; it gets marked by the caller. */ + } + + if (conn->socks_request->socks_version == 4) { + buf[0] = 0x00; /* version */ + if (answer_type == RESOLVED_TYPE_IPV4 && answer_len == 4) { + buf[1] = SOCKS4_GRANTED; + set_uint16(buf+2, 0); + memcpy(buf+4, answer, 4); /* address */ + replylen = SOCKS4_NETWORK_LEN; + } else { /* "error" */ + buf[1] = SOCKS4_REJECT; + memset(buf+2, 0, 6); + replylen = SOCKS4_NETWORK_LEN; + } + } else if (conn->socks_request->socks_version == 5) { + /* SOCKS5 */ + buf[0] = 0x05; /* version */ + if (answer_type == RESOLVED_TYPE_IPV4 && answer_len == 4) { + buf[1] = SOCKS5_SUCCEEDED; + buf[2] = 0; /* reserved */ + buf[3] = 0x01; /* IPv4 address type */ + memcpy(buf+4, answer, 4); /* address */ + set_uint16(buf+8, 0); /* port == 0. */ + replylen = 10; + } else if (answer_type == RESOLVED_TYPE_IPV6 && answer_len == 16) { + buf[1] = SOCKS5_SUCCEEDED; + buf[2] = 0; /* reserved */ + buf[3] = 0x04; /* IPv6 address type */ + memcpy(buf+4, answer, 16); /* address */ + set_uint16(buf+20, 0); /* port == 0. */ + replylen = 22; + } else if (answer_type == RESOLVED_TYPE_HOSTNAME && answer_len < 256) { + buf[1] = SOCKS5_SUCCEEDED; + buf[2] = 0; /* reserved */ + buf[3] = 0x03; /* Domainname address type */ + buf[4] = (char)answer_len; + memcpy(buf+5, answer, answer_len); /* address */ + set_uint16(buf+5+answer_len, 0); /* port == 0. */ + replylen = 5+answer_len+2; + } else { + buf[1] = SOCKS5_HOST_UNREACHABLE; + memset(buf+2, 0, 8); + replylen = 10; + } + } else { + /* no socks version info; don't send anything back */ + return; + } + connection_ap_handshake_socks_reply(conn, buf, replylen, + (answer_type == RESOLVED_TYPE_IPV4 || + answer_type == RESOLVED_TYPE_IPV6 || + answer_type == RESOLVED_TYPE_HOSTNAME) ? + 0 : END_STREAM_REASON_RESOLVEFAILED); +} + +/** Send a socks reply to stream <b>conn</b>, using the appropriate + * socks version, etc, and mark <b>conn</b> as completed with SOCKS + * handshaking. + * + * If <b>reply</b> is defined, then write <b>replylen</b> bytes of it to conn + * and return, else reply based on <b>endreason</b> (one of + * END_STREAM_REASON_*). If <b>reply</b> is undefined, <b>endreason</b> can't + * be 0 or REASON_DONE. Send endreason to the controller, if appropriate. + */ +void +connection_ap_handshake_socks_reply(entry_connection_t *conn, char *reply, + size_t replylen, int endreason) +{ + char buf[256]; + socks5_reply_status_t status = + stream_end_reason_to_socks5_response(endreason); + + tor_assert(conn->socks_request); /* make sure it's an AP stream */ + + if (!SOCKS_COMMAND_IS_RESOLVE(conn->socks_request->command)) { + control_event_stream_status(conn, status==SOCKS5_SUCCEEDED ? + STREAM_EVENT_SUCCEEDED : STREAM_EVENT_FAILED, + endreason); + } + + /* Flag this stream's circuit as having completed a stream successfully + * (for path bias) */ + if (status == SOCKS5_SUCCEEDED || + endreason == END_STREAM_REASON_RESOLVEFAILED || + endreason == END_STREAM_REASON_CONNECTREFUSED || + endreason == END_STREAM_REASON_CONNRESET || + endreason == END_STREAM_REASON_NOROUTE || + endreason == END_STREAM_REASON_RESOURCELIMIT) { + if (!conn->edge_.on_circuit || + !CIRCUIT_IS_ORIGIN(conn->edge_.on_circuit)) { + if (endreason != END_STREAM_REASON_RESOLVEFAILED) { + log_info(LD_BUG, + "No origin circuit for successful SOCKS stream %"PRIu64 + ". Reason: %d", + (ENTRY_TO_CONN(conn)->global_identifier), + endreason); + } + /* + * Else DNS remaps and failed hidden service lookups can send us + * here with END_STREAM_REASON_RESOLVEFAILED; ignore it + * + * Perhaps we could make the test more precise; we can tell hidden + * services by conn->edge_.renddata != NULL; anything analogous for + * the DNS remap case? + */ + } else { + // XXX: Hrmm. It looks like optimistic data can't go through this + // codepath, but someone should probably test it and make sure. + // We don't want to mark optimistically opened streams as successful. + pathbias_mark_use_success(TO_ORIGIN_CIRCUIT(conn->edge_.on_circuit)); + } + } + + if (conn->socks_request->has_finished) { + log_warn(LD_BUG, "(Harmless.) duplicate calls to " + "connection_ap_handshake_socks_reply."); + return; + } + if (replylen) { /* we already have a reply in mind */ + connection_buf_add(reply, replylen, ENTRY_TO_CONN(conn)); + conn->socks_request->has_finished = 1; + return; + } + if (conn->socks_request->listener_type == + CONN_TYPE_AP_HTTP_CONNECT_LISTENER) { + const char *response = end_reason_to_http_connect_response_line(endreason); + if (!response) { + response = "HTTP/1.0 400 Bad Request\r\n\r\n"; + } + connection_buf_add(response, strlen(response), ENTRY_TO_CONN(conn)); + } else if (conn->socks_request->socks_version == 4) { + memset(buf,0,SOCKS4_NETWORK_LEN); + buf[1] = (status==SOCKS5_SUCCEEDED ? SOCKS4_GRANTED : SOCKS4_REJECT); + /* leave version, destport, destip zero */ + connection_buf_add(buf, SOCKS4_NETWORK_LEN, ENTRY_TO_CONN(conn)); + } else if (conn->socks_request->socks_version == 5) { + size_t buf_len; + memset(buf,0,sizeof(buf)); + if (tor_addr_family(&conn->edge_.base_.addr) == AF_INET) { + buf[0] = 5; /* version 5 */ + buf[1] = (char)status; + buf[2] = 0; + buf[3] = 1; /* ipv4 addr */ + /* 4 bytes for the header, 2 bytes for the port, 4 for the address. */ + buf_len = 10; + } else { /* AF_INET6. */ + buf[0] = 5; /* version 5 */ + buf[1] = (char)status; + buf[2] = 0; + buf[3] = 4; /* ipv6 addr */ + /* 4 bytes for the header, 2 bytes for the port, 16 for the address. */ + buf_len = 22; + } + connection_buf_add(buf,buf_len,ENTRY_TO_CONN(conn)); + } + /* If socks_version isn't 4 or 5, don't send anything. + * This can happen in the case of AP bridges. */ + conn->socks_request->has_finished = 1; + return; +} + +/** Read a RELAY_BEGIN or RELAY_BEGIN_DIR cell from <b>cell</b>, decode it, and + * place the result in <b>bcell</b>. On success return 0; on failure return + * <0 and set *<b>end_reason_out</b> to the end reason we should send back to + * the client. + * + * Return -1 in the case where we want to send a RELAY_END cell, and < -1 when + * we don't. + **/ +STATIC int +begin_cell_parse(const cell_t *cell, begin_cell_t *bcell, + uint8_t *end_reason_out) +{ + relay_header_t rh; + const uint8_t *body, *nul; + + memset(bcell, 0, sizeof(*bcell)); + *end_reason_out = END_STREAM_REASON_MISC; + + relay_header_unpack(&rh, cell->payload); + if (rh.length > RELAY_PAYLOAD_SIZE) { + return -2; /*XXXX why not TORPROTOCOL? */ + } + + bcell->stream_id = rh.stream_id; + + if (rh.command == RELAY_COMMAND_BEGIN_DIR) { + bcell->is_begindir = 1; + return 0; + } else if (rh.command != RELAY_COMMAND_BEGIN) { + log_warn(LD_BUG, "Got an unexpected command %d", (int)rh.command); + *end_reason_out = END_STREAM_REASON_INTERNAL; + return -1; + } + + body = cell->payload + RELAY_HEADER_SIZE; + nul = memchr(body, 0, rh.length); + if (! nul) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Relay begin cell has no \0. Closing."); + *end_reason_out = END_STREAM_REASON_TORPROTOCOL; + return -1; + } + + if (tor_addr_port_split(LOG_PROTOCOL_WARN, + (char*)(body), + &bcell->address,&bcell->port)<0) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Unable to parse addr:port in relay begin cell. Closing."); + *end_reason_out = END_STREAM_REASON_TORPROTOCOL; + return -1; + } + if (bcell->port == 0) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Missing port in relay begin cell. Closing."); + tor_free(bcell->address); + *end_reason_out = END_STREAM_REASON_TORPROTOCOL; + return -1; + } + if (body + rh.length >= nul + 4) + bcell->flags = ntohl(get_uint32(nul+1)); + + return 0; +} + +/** For the given <b>circ</b> and the edge connection <b>conn</b>, setup the + * connection, attach it to the circ and connect it. Return 0 on success + * or END_CIRC_AT_ORIGIN if we can't find the requested hidden service port + * where the caller should close the circuit. */ +static int +handle_hs_exit_conn(circuit_t *circ, edge_connection_t *conn) +{ + int ret; + origin_circuit_t *origin_circ; + + assert_circuit_ok(circ); + tor_assert(circ->purpose == CIRCUIT_PURPOSE_S_REND_JOINED); + tor_assert(conn); + + log_debug(LD_REND, "Connecting the hidden service rendezvous circuit " + "to the service destination."); + + origin_circ = TO_ORIGIN_CIRCUIT(circ); + conn->base_.address = tor_strdup("(rendezvous)"); + conn->base_.state = EXIT_CONN_STATE_CONNECTING; + + /* The circuit either has an hs identifier for v3+ or a rend_data for legacy + * service. */ + if (origin_circ->rend_data) { + conn->rend_data = rend_data_dup(origin_circ->rend_data); + tor_assert(connection_edge_is_rendezvous_stream(conn)); + ret = rend_service_set_connection_addr_port(conn, origin_circ); + } else if (origin_circ->hs_ident) { + /* Setup the identifier to be the one for the circuit service. */ + conn->hs_ident = + hs_ident_edge_conn_new(&origin_circ->hs_ident->identity_pk); + tor_assert(connection_edge_is_rendezvous_stream(conn)); + ret = hs_service_set_conn_addr_port(origin_circ, conn); + } else { + /* We should never get here if the circuit's purpose is rendezvous. */ + tor_assert_nonfatal_unreached(); + return -1; + } + if (ret < 0) { + log_info(LD_REND, "Didn't find rendezvous service (addr%s, port %d)", + fmt_addr(&TO_CONN(conn)->addr), TO_CONN(conn)->port); + /* Send back reason DONE because we want to make hidden service port + * scanning harder thus instead of returning that the exit policy + * didn't match, which makes it obvious that the port is closed, + * return DONE and kill the circuit. That way, a user (malicious or + * not) needs one circuit per bad port unless it matches the policy of + * the hidden service. */ + relay_send_end_cell_from_edge(conn->stream_id, circ, + END_STREAM_REASON_DONE, + origin_circ->cpath->prev); + connection_free_(TO_CONN(conn)); + + /* Drop the circuit here since it might be someone deliberately + * scanning the hidden service ports. Note that this mitigates port + * scanning by adding more work on the attacker side to successfully + * scan but does not fully solve it. */ + if (ret < -1) { + return END_CIRC_AT_ORIGIN; + } else { + return 0; + } + } + + /* Link the circuit and the connection crypt path. */ + conn->cpath_layer = origin_circ->cpath->prev; + + /* Add it into the linked list of p_streams on this circuit */ + conn->next_stream = origin_circ->p_streams; + origin_circ->p_streams = conn; + conn->on_circuit = circ; + assert_circuit_ok(circ); + + hs_inc_rdv_stream_counter(origin_circ); + + /* Connect tor to the hidden service destination. */ + connection_exit_connect(conn); + + /* For path bias: This circuit was used successfully */ + pathbias_mark_use_success(origin_circ); + return 0; +} + +/** A relay 'begin' or 'begin_dir' cell has arrived, and either we are + * an exit hop for the circuit, or we are the origin and it is a + * rendezvous begin. + * + * Launch a new exit connection and initialize things appropriately. + * + * If it's a rendezvous stream, call connection_exit_connect() on + * it. + * + * For general streams, call dns_resolve() on it first, and only call + * connection_exit_connect() if the dns answer is already known. + * + * Note that we don't call connection_add() on the new stream! We wait + * for connection_exit_connect() to do that. + * + * Return -(some circuit end reason) if we want to tear down <b>circ</b>. + * Else return 0. + */ +int +connection_exit_begin_conn(cell_t *cell, circuit_t *circ) +{ + edge_connection_t *n_stream; + relay_header_t rh; + char *address = NULL; + uint16_t port = 0; + or_circuit_t *or_circ = NULL; + origin_circuit_t *origin_circ = NULL; + crypt_path_t *layer_hint = NULL; + const or_options_t *options = get_options(); + begin_cell_t bcell; + int rv; + uint8_t end_reason=0; + + assert_circuit_ok(circ); + if (!CIRCUIT_IS_ORIGIN(circ)) { + or_circ = TO_OR_CIRCUIT(circ); + } else { + tor_assert(circ->purpose == CIRCUIT_PURPOSE_S_REND_JOINED); + origin_circ = TO_ORIGIN_CIRCUIT(circ); + layer_hint = origin_circ->cpath->prev; + } + + relay_header_unpack(&rh, cell->payload); + if (rh.length > RELAY_PAYLOAD_SIZE) + return -END_CIRC_REASON_TORPROTOCOL; + + if (!server_mode(options) && + circ->purpose != CIRCUIT_PURPOSE_S_REND_JOINED) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Relay begin cell at non-server. Closing."); + relay_send_end_cell_from_edge(rh.stream_id, circ, + END_STREAM_REASON_EXITPOLICY, NULL); + return 0; + } + + rv = begin_cell_parse(cell, &bcell, &end_reason); + if (rv < -1) { + return -END_CIRC_REASON_TORPROTOCOL; + } else if (rv == -1) { + tor_free(bcell.address); + relay_send_end_cell_from_edge(rh.stream_id, circ, end_reason, layer_hint); + return 0; + } + + if (! bcell.is_begindir) { + /* Steal reference */ + address = bcell.address; + port = bcell.port; + + if (or_circ && or_circ->p_chan) { + const int client_chan = channel_is_client(or_circ->p_chan); + if ((client_chan || + (!connection_or_digest_is_known_relay( + or_circ->p_chan->identity_digest) && + should_refuse_unknown_exits(options)))) { + /* Don't let clients use us as a single-hop proxy. It attracts + * attackers and users who'd be better off with, well, single-hop + * proxies. */ + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Attempt by %s to open a stream %s. Closing.", + safe_str(channel_get_canonical_remote_descr(or_circ->p_chan)), + client_chan ? "on first hop of circuit" : + "from unknown relay"); + relay_send_end_cell_from_edge(rh.stream_id, circ, + client_chan ? + END_STREAM_REASON_TORPROTOCOL : + END_STREAM_REASON_MISC, + NULL); + tor_free(address); + return 0; + } + } + } else if (rh.command == RELAY_COMMAND_BEGIN_DIR) { + if (!directory_permits_begindir_requests(options) || + circ->purpose != CIRCUIT_PURPOSE_OR) { + relay_send_end_cell_from_edge(rh.stream_id, circ, + END_STREAM_REASON_NOTDIRECTORY, layer_hint); + return 0; + } + /* Make sure to get the 'real' address of the previous hop: the + * caller might want to know whether the remote IP address has changed, + * and we might already have corrected base_.addr[ess] for the relay's + * canonical IP address. */ + if (or_circ && or_circ->p_chan) + address = tor_strdup(channel_get_actual_remote_address(or_circ->p_chan)); + else + address = tor_strdup("127.0.0.1"); + port = 1; /* XXXX This value is never actually used anywhere, and there + * isn't "really" a connection here. But we + * need to set it to something nonzero. */ + } else { + log_warn(LD_BUG, "Got an unexpected command %d", (int)rh.command); + relay_send_end_cell_from_edge(rh.stream_id, circ, + END_STREAM_REASON_INTERNAL, layer_hint); + return 0; + } + + if (! options->IPv6Exit) { + /* I don't care if you prefer IPv6; I can't give you any. */ + bcell.flags &= ~BEGIN_FLAG_IPV6_PREFERRED; + /* If you don't want IPv4, I can't help. */ + if (bcell.flags & BEGIN_FLAG_IPV4_NOT_OK) { + tor_free(address); + relay_send_end_cell_from_edge(rh.stream_id, circ, + END_STREAM_REASON_EXITPOLICY, layer_hint); + return 0; + } + } + + log_debug(LD_EXIT,"Creating new exit connection."); + /* The 'AF_INET' here is temporary; we might need to change it later in + * connection_exit_connect(). */ + n_stream = edge_connection_new(CONN_TYPE_EXIT, AF_INET); + + /* Remember the tunneled request ID in the new edge connection, so that + * we can measure download times. */ + n_stream->dirreq_id = circ->dirreq_id; + + n_stream->base_.purpose = EXIT_PURPOSE_CONNECT; + n_stream->begincell_flags = bcell.flags; + n_stream->stream_id = rh.stream_id; + n_stream->base_.port = port; + /* leave n_stream->s at -1, because it's not yet valid */ + n_stream->package_window = STREAMWINDOW_START; + n_stream->deliver_window = STREAMWINDOW_START; + + if (circ->purpose == CIRCUIT_PURPOSE_S_REND_JOINED) { + int ret; + tor_free(address); + /* We handle this circuit and stream in this function for all supported + * hidden service version. */ + ret = handle_hs_exit_conn(circ, n_stream); + + if (ret == 0) { + /* This was a valid cell. Count it as delivered + overhead. */ + circuit_read_valid_data(origin_circ, rh.length); + } + return ret; + } + tor_strlower(address); + n_stream->base_.address = address; + n_stream->base_.state = EXIT_CONN_STATE_RESOLVEFAILED; + /* default to failed, change in dns_resolve if it turns out not to fail */ + + /* If we're hibernating or shutting down, we refuse to open new streams. */ + if (we_are_hibernating()) { + relay_send_end_cell_from_edge(rh.stream_id, circ, + END_STREAM_REASON_HIBERNATING, NULL); + connection_free_(TO_CONN(n_stream)); + return 0; + } + + n_stream->on_circuit = circ; + + if (rh.command == RELAY_COMMAND_BEGIN_DIR) { + tor_addr_t tmp_addr; + tor_assert(or_circ); + if (or_circ->p_chan && + channel_get_addr_if_possible(or_circ->p_chan, &tmp_addr)) { + tor_addr_copy(&n_stream->base_.addr, &tmp_addr); + } + return connection_exit_connect_dir(n_stream); + } + + log_debug(LD_EXIT,"about to start the dns_resolve()."); + + /* send it off to the gethostbyname farm */ + switch (dns_resolve(n_stream)) { + case 1: /* resolve worked; now n_stream is attached to circ. */ + assert_circuit_ok(circ); + log_debug(LD_EXIT,"about to call connection_exit_connect()."); + connection_exit_connect(n_stream); + return 0; + case -1: /* resolve failed */ + relay_send_end_cell_from_edge(rh.stream_id, circ, + END_STREAM_REASON_RESOLVEFAILED, NULL); + /* n_stream got freed. don't touch it. */ + break; + case 0: /* resolve added to pending list */ + assert_circuit_ok(circ); + break; + } + return 0; +} + +/** + * Called when we receive a RELAY_COMMAND_RESOLVE cell 'cell' along the + * circuit <b>circ</b>; + * begin resolving the hostname, and (eventually) reply with a RESOLVED cell. + */ +int +connection_exit_begin_resolve(cell_t *cell, or_circuit_t *circ) +{ + edge_connection_t *dummy_conn; + relay_header_t rh; + + assert_circuit_ok(TO_CIRCUIT(circ)); + relay_header_unpack(&rh, cell->payload); + if (rh.length > RELAY_PAYLOAD_SIZE) + return -1; + + /* This 'dummy_conn' only exists to remember the stream ID + * associated with the resolve request; and to make the + * implementation of dns.c more uniform. (We really only need to + * remember the circuit, the stream ID, and the hostname to be + * resolved; but if we didn't store them in a connection like this, + * the housekeeping in dns.c would get way more complicated.) + */ + dummy_conn = edge_connection_new(CONN_TYPE_EXIT, AF_INET); + dummy_conn->stream_id = rh.stream_id; + dummy_conn->base_.address = tor_strndup( + (char*)cell->payload+RELAY_HEADER_SIZE, + rh.length); + dummy_conn->base_.port = 0; + dummy_conn->base_.state = EXIT_CONN_STATE_RESOLVEFAILED; + dummy_conn->base_.purpose = EXIT_PURPOSE_RESOLVE; + + dummy_conn->on_circuit = TO_CIRCUIT(circ); + + /* send it off to the gethostbyname farm */ + switch (dns_resolve(dummy_conn)) { + case -1: /* Impossible to resolve; a resolved cell was sent. */ + /* Connection freed; don't touch it. */ + return 0; + case 1: /* The result was cached; a resolved cell was sent. */ + if (!dummy_conn->base_.marked_for_close) + connection_free_(TO_CONN(dummy_conn)); + return 0; + case 0: /* resolve added to pending list */ + assert_circuit_ok(TO_CIRCUIT(circ)); + break; + } + return 0; +} + +/** Helper: Return true and set *<b>why_rejected</b> to an optional clarifying + * message message iff we do not allow connections to <b>addr</b>:<b>port</b>. + */ +static int +my_exit_policy_rejects(const tor_addr_t *addr, + uint16_t port, + const char **why_rejected) +{ + if (router_compare_to_my_exit_policy(addr, port)) { + *why_rejected = ""; + return 1; + } else if (tor_addr_family(addr) == AF_INET6 && !get_options()->IPv6Exit) { + *why_rejected = " (IPv6 address without IPv6Exit configured)"; + return 1; + } + return 0; +} + +/** Connect to conn's specified addr and port. If it worked, conn + * has now been added to the connection_array. + * + * Send back a connected cell. Include the resolved IP of the destination + * address, but <em>only</em> if it's a general exit stream. (Rendezvous + * streams must not reveal what IP they connected to.) + */ +void +connection_exit_connect(edge_connection_t *edge_conn) +{ + const tor_addr_t *addr; + uint16_t port; + connection_t *conn = TO_CONN(edge_conn); + int socket_error = 0, result; + const char *why_failed_exit_policy = NULL; + + /* Apply exit policy to non-rendezvous connections. */ + if (! connection_edge_is_rendezvous_stream(edge_conn) && + my_exit_policy_rejects(&edge_conn->base_.addr, + edge_conn->base_.port, + &why_failed_exit_policy)) { + if (BUG(!why_failed_exit_policy)) + why_failed_exit_policy = ""; + log_info(LD_EXIT,"%s:%d failed exit policy%s. Closing.", + escaped_safe_str_client(conn->address), conn->port, + why_failed_exit_policy); + connection_edge_end(edge_conn, END_STREAM_REASON_EXITPOLICY); + circuit_detach_stream(circuit_get_by_edge_conn(edge_conn), edge_conn); + connection_free(conn); + return; + } + +#ifdef HAVE_SYS_UN_H + if (conn->socket_family != AF_UNIX) { +#else + { +#endif /* defined(HAVE_SYS_UN_H) */ + addr = &conn->addr; + port = conn->port; + + if (tor_addr_family(addr) == AF_INET6) + conn->socket_family = AF_INET6; + + log_debug(LD_EXIT, "about to try connecting"); + result = connection_connect(conn, conn->address, + addr, port, &socket_error); +#ifdef HAVE_SYS_UN_H + } else { + /* + * In the AF_UNIX case, we expect to have already had conn->port = 1, + * tor_addr_make_unspec(conn->addr) (cf. the way we mark in the incoming + * case in connection_handle_listener_read()), and conn->address should + * have the socket path to connect to. + */ + tor_assert(conn->address && strlen(conn->address) > 0); + + log_debug(LD_EXIT, "about to try connecting"); + result = connection_connect_unix(conn, conn->address, &socket_error); +#endif /* defined(HAVE_SYS_UN_H) */ + } + + switch (result) { + case -1: { + int reason = errno_to_stream_end_reason(socket_error); + connection_edge_end(edge_conn, reason); + circuit_detach_stream(circuit_get_by_edge_conn(edge_conn), edge_conn); + connection_free(conn); + return; + } + case 0: + conn->state = EXIT_CONN_STATE_CONNECTING; + + connection_watch_events(conn, READ_EVENT | WRITE_EVENT); + /* writable indicates finish; + * readable/error indicates broken link in windows-land. */ + return; + /* case 1: fall through */ + } + + conn->state = EXIT_CONN_STATE_OPEN; + if (connection_get_outbuf_len(conn)) { + /* in case there are any queued data cells, from e.g. optimistic data */ + connection_watch_events(conn, READ_EVENT|WRITE_EVENT); + } else { + connection_watch_events(conn, READ_EVENT); + } + + /* also, deliver a 'connected' cell back through the circuit. */ + if (connection_edge_is_rendezvous_stream(edge_conn)) { + /* don't send an address back! */ + connection_edge_send_command(edge_conn, + RELAY_COMMAND_CONNECTED, + NULL, 0); + } else { /* normal stream */ + uint8_t connected_payload[MAX_CONNECTED_CELL_PAYLOAD_LEN]; + int connected_payload_len = + connected_cell_format_payload(connected_payload, &conn->addr, + edge_conn->address_ttl); + if (connected_payload_len < 0) { + connection_edge_end(edge_conn, END_STREAM_REASON_INTERNAL); + circuit_detach_stream(circuit_get_by_edge_conn(edge_conn), edge_conn); + connection_free(conn); + return; + } + + connection_edge_send_command(edge_conn, + RELAY_COMMAND_CONNECTED, + (char*)connected_payload, + connected_payload_len); + } +} + +/** Given an exit conn that should attach to us as a directory server, open a + * bridge connection with a linked connection pair, create a new directory + * conn, and join them together. Return 0 on success (or if there was an + * error we could send back an end cell for). Return -(some circuit end + * reason) if the circuit needs to be torn down. Either connects + * <b>exitconn</b>, frees it, or marks it, as appropriate. + */ +static int +connection_exit_connect_dir(edge_connection_t *exitconn) +{ + dir_connection_t *dirconn = NULL; + or_circuit_t *circ = TO_OR_CIRCUIT(exitconn->on_circuit); + + log_info(LD_EXIT, "Opening local connection for anonymized directory exit"); + + exitconn->base_.state = EXIT_CONN_STATE_OPEN; + + dirconn = dir_connection_new(tor_addr_family(&exitconn->base_.addr)); + + tor_addr_copy(&dirconn->base_.addr, &exitconn->base_.addr); + dirconn->base_.port = 0; + dirconn->base_.address = tor_strdup(exitconn->base_.address); + dirconn->base_.type = CONN_TYPE_DIR; + dirconn->base_.purpose = DIR_PURPOSE_SERVER; + dirconn->base_.state = DIR_CONN_STATE_SERVER_COMMAND_WAIT; + + /* Note that the new dir conn belongs to the same tunneled request as + * the edge conn, so that we can measure download times. */ + dirconn->dirreq_id = exitconn->dirreq_id; + + connection_link_connections(TO_CONN(dirconn), TO_CONN(exitconn)); + + if (connection_add(TO_CONN(exitconn))<0) { + connection_edge_end(exitconn, END_STREAM_REASON_RESOURCELIMIT); + connection_free_(TO_CONN(exitconn)); + connection_free_(TO_CONN(dirconn)); + return 0; + } + + /* link exitconn to circ, now that we know we can use it. */ + exitconn->next_stream = circ->n_streams; + circ->n_streams = exitconn; + + if (connection_add(TO_CONN(dirconn))<0) { + connection_edge_end(exitconn, END_STREAM_REASON_RESOURCELIMIT); + connection_close_immediate(TO_CONN(exitconn)); + connection_mark_for_close(TO_CONN(exitconn)); + connection_free_(TO_CONN(dirconn)); + return 0; + } + + connection_start_reading(TO_CONN(dirconn)); + connection_start_reading(TO_CONN(exitconn)); + + if (connection_edge_send_command(exitconn, + RELAY_COMMAND_CONNECTED, NULL, 0) < 0) { + connection_mark_for_close(TO_CONN(exitconn)); + connection_mark_for_close(TO_CONN(dirconn)); + return 0; + } + + return 0; +} + +/** Return 1 if <b>conn</b> is a rendezvous stream, or 0 if + * it is a general stream. + */ +int +connection_edge_is_rendezvous_stream(const edge_connection_t *conn) +{ + tor_assert(conn); + /* It should not be possible to set both of these structs */ + tor_assert_nonfatal(!(conn->rend_data && conn->hs_ident)); + + if (conn->rend_data || conn->hs_ident) { + return 1; + } + return 0; +} + +/** Return 1 if router <b>exit_node</b> is likely to allow stream <b>conn</b> + * to exit from it, or 0 if it probably will not allow it. + * (We might be uncertain if conn's destination address has not yet been + * resolved.) + */ +int +connection_ap_can_use_exit(const entry_connection_t *conn, + const node_t *exit_node) +{ + const or_options_t *options = get_options(); + + tor_assert(conn); + tor_assert(conn->socks_request); + tor_assert(exit_node); + + /* If a particular exit node has been requested for the new connection, + * make sure the exit node of the existing circuit matches exactly. + */ + if (conn->chosen_exit_name) { + const node_t *chosen_exit = + node_get_by_nickname(conn->chosen_exit_name, 0); + if (!chosen_exit || tor_memneq(chosen_exit->identity, + exit_node->identity, DIGEST_LEN)) { + /* doesn't match */ +// log_debug(LD_APP,"Requested node '%s', considering node '%s'. No.", +// conn->chosen_exit_name, exit->nickname); + return 0; + } + } + + if (conn->use_begindir) { + /* Internal directory fetches do not count as exiting. */ + return 1; + } + + if (conn->socks_request->command == SOCKS_COMMAND_CONNECT) { + tor_addr_t addr, *addrp = NULL; + addr_policy_result_t r; + if (0 == tor_addr_parse(&addr, conn->socks_request->address)) { + addrp = &addr; + } else if (!conn->entry_cfg.ipv4_traffic && conn->entry_cfg.ipv6_traffic) { + tor_addr_make_null(&addr, AF_INET6); + addrp = &addr; + } else if (conn->entry_cfg.ipv4_traffic && !conn->entry_cfg.ipv6_traffic) { + tor_addr_make_null(&addr, AF_INET); + addrp = &addr; + } + r = compare_tor_addr_to_node_policy(addrp, conn->socks_request->port, + exit_node); + if (r == ADDR_POLICY_REJECTED) + return 0; /* We know the address, and the exit policy rejects it. */ + if (r == ADDR_POLICY_PROBABLY_REJECTED && !conn->chosen_exit_name) + return 0; /* We don't know the addr, but the exit policy rejects most + * addresses with this port. Since the user didn't ask for + * this node, err on the side of caution. */ + } else if (SOCKS_COMMAND_IS_RESOLVE(conn->socks_request->command)) { + /* Don't send DNS requests to non-exit servers by default. */ + if (!conn->chosen_exit_name && node_exit_policy_rejects_all(exit_node)) + return 0; + } + if (routerset_contains_node(options->ExcludeExitNodesUnion_, exit_node)) { + /* Not a suitable exit. Refuse it. */ + return 0; + } + + return 1; +} + +/** If address is of the form "y.onion" with a well-formed handle y: + * Put a NUL after y, lower-case it, and return ONION_V2_HOSTNAME or + * ONION_V3_HOSTNAME depending on the HS version. + * + * If address is of the form "x.y.onion" with a well-formed handle x: + * Drop "x.", put a NUL after y, lower-case it, and return + * ONION_V2_HOSTNAME or ONION_V3_HOSTNAME depending on the HS version. + * + * If address is of the form "y.onion" with a badly-formed handle y: + * Return BAD_HOSTNAME and log a message. + * + * If address is of the form "y.exit": + * Put a NUL after y and return EXIT_HOSTNAME. + * + * Otherwise: + * Return NORMAL_HOSTNAME and change nothing. + */ +hostname_type_t +parse_extended_hostname(char *address) +{ + char *s; + char *q; + char query[HS_SERVICE_ADDR_LEN_BASE32+1]; + + s = strrchr(address,'.'); + if (!s) + return NORMAL_HOSTNAME; /* no dot, thus normal */ + if (!strcmp(s+1,"exit")) { + *s = 0; /* NUL-terminate it */ + return EXIT_HOSTNAME; /* .exit */ + } + if (strcmp(s+1,"onion")) + return NORMAL_HOSTNAME; /* neither .exit nor .onion, thus normal */ + + /* so it is .onion */ + *s = 0; /* NUL-terminate it */ + /* locate a 'sub-domain' component, in order to remove it */ + q = strrchr(address, '.'); + if (q == address) { + goto failed; /* reject sub-domain, as DNS does */ + } + q = (NULL == q) ? address : q + 1; + if (strlcpy(query, q, HS_SERVICE_ADDR_LEN_BASE32+1) >= + HS_SERVICE_ADDR_LEN_BASE32+1) + goto failed; + if (q != address) { + memmove(address, q, strlen(q) + 1 /* also get \0 */); + } + if (rend_valid_v2_service_id(query)) { + return ONION_V2_HOSTNAME; /* success */ + } + if (hs_address_is_valid(query)) { + return ONION_V3_HOSTNAME; + } + failed: + /* otherwise, return to previous state and return 0 */ + *s = '.'; + log_warn(LD_APP, "Invalid onion hostname %s; rejecting", + safe_str_client(address)); + return BAD_HOSTNAME; +} + +/** Return true iff the (possibly NULL) <b>alen</b>-byte chunk of memory at + * <b>a</b> is equal to the (possibly NULL) <b>blen</b>-byte chunk of memory + * at <b>b</b>. */ +static int +memeq_opt(const char *a, size_t alen, const char *b, size_t blen) +{ + if (a == NULL) { + return (b == NULL); + } else if (b == NULL) { + return 0; + } else if (alen != blen) { + return 0; + } else { + return tor_memeq(a, b, alen); + } +} + +/** + * Return true iff none of the isolation flags and fields in <b>conn</b> + * should prevent it from being attached to <b>circ</b>. + */ +int +connection_edge_compatible_with_circuit(const entry_connection_t *conn, + const origin_circuit_t *circ) +{ + const uint8_t iso = conn->entry_cfg.isolation_flags; + const socks_request_t *sr = conn->socks_request; + + /* If circ has never been used for an isolated connection, we can + * totally use it for this one. */ + if (!circ->isolation_values_set) + return 1; + + /* If circ has been used for connections having more than one value + * for some field f, it will have the corresponding bit set in + * isolation_flags_mixed. If isolation_flags_mixed has any bits + * in common with iso, then conn must be isolated from at least + * one stream that has been attached to circ. */ + if ((iso & circ->isolation_flags_mixed) != 0) { + /* For at least one field where conn is isolated, the circuit + * already has mixed streams. */ + return 0; + } + + if (! conn->original_dest_address) { + log_warn(LD_BUG, "Reached connection_edge_compatible_with_circuit without " + "having set conn->original_dest_address"); + ((entry_connection_t*)conn)->original_dest_address = + tor_strdup(conn->socks_request->address); + } + + if ((iso & ISO_STREAM) && + (circ->associated_isolated_stream_global_id != + ENTRY_TO_CONN(conn)->global_identifier)) + return 0; + + if ((iso & ISO_DESTPORT) && conn->socks_request->port != circ->dest_port) + return 0; + if ((iso & ISO_DESTADDR) && + strcasecmp(conn->original_dest_address, circ->dest_address)) + return 0; + if ((iso & ISO_SOCKSAUTH) && + (! memeq_opt(sr->username, sr->usernamelen, + circ->socks_username, circ->socks_username_len) || + ! memeq_opt(sr->password, sr->passwordlen, + circ->socks_password, circ->socks_password_len))) + return 0; + if ((iso & ISO_CLIENTPROTO) && + (conn->socks_request->listener_type != circ->client_proto_type || + conn->socks_request->socks_version != circ->client_proto_socksver)) + return 0; + if ((iso & ISO_CLIENTADDR) && + !tor_addr_eq(&ENTRY_TO_CONN(conn)->addr, &circ->client_addr)) + return 0; + if ((iso & ISO_SESSIONGRP) && + conn->entry_cfg.session_group != circ->session_group) + return 0; + if ((iso & ISO_NYM_EPOCH) && conn->nym_epoch != circ->nym_epoch) + return 0; + + return 1; +} + +/** + * If <b>dry_run</b> is false, update <b>circ</b>'s isolation flags and fields + * to reflect having had <b>conn</b> attached to it, and return 0. Otherwise, + * if <b>dry_run</b> is true, then make no changes to <b>circ</b>, and return + * a bitfield of isolation flags that we would have to set in + * isolation_flags_mixed to add <b>conn</b> to <b>circ</b>, or -1 if + * <b>circ</b> has had no streams attached to it. + */ +int +connection_edge_update_circuit_isolation(const entry_connection_t *conn, + origin_circuit_t *circ, + int dry_run) +{ + const socks_request_t *sr = conn->socks_request; + if (! conn->original_dest_address) { + log_warn(LD_BUG, "Reached connection_update_circuit_isolation without " + "having set conn->original_dest_address"); + ((entry_connection_t*)conn)->original_dest_address = + tor_strdup(conn->socks_request->address); + } + + if (!circ->isolation_values_set) { + if (dry_run) + return -1; + circ->associated_isolated_stream_global_id = + ENTRY_TO_CONN(conn)->global_identifier; + circ->dest_port = conn->socks_request->port; + circ->dest_address = tor_strdup(conn->original_dest_address); + circ->client_proto_type = conn->socks_request->listener_type; + circ->client_proto_socksver = conn->socks_request->socks_version; + tor_addr_copy(&circ->client_addr, &ENTRY_TO_CONN(conn)->addr); + circ->session_group = conn->entry_cfg.session_group; + circ->nym_epoch = conn->nym_epoch; + circ->socks_username = sr->username ? + tor_memdup(sr->username, sr->usernamelen) : NULL; + circ->socks_password = sr->password ? + tor_memdup(sr->password, sr->passwordlen) : NULL; + circ->socks_username_len = sr->usernamelen; + circ->socks_password_len = sr->passwordlen; + + circ->isolation_values_set = 1; + return 0; + } else { + uint8_t mixed = 0; + if (conn->socks_request->port != circ->dest_port) + mixed |= ISO_DESTPORT; + if (strcasecmp(conn->original_dest_address, circ->dest_address)) + mixed |= ISO_DESTADDR; + if (!memeq_opt(sr->username, sr->usernamelen, + circ->socks_username, circ->socks_username_len) || + !memeq_opt(sr->password, sr->passwordlen, + circ->socks_password, circ->socks_password_len)) + mixed |= ISO_SOCKSAUTH; + if ((conn->socks_request->listener_type != circ->client_proto_type || + conn->socks_request->socks_version != circ->client_proto_socksver)) + mixed |= ISO_CLIENTPROTO; + if (!tor_addr_eq(&ENTRY_TO_CONN(conn)->addr, &circ->client_addr)) + mixed |= ISO_CLIENTADDR; + if (conn->entry_cfg.session_group != circ->session_group) + mixed |= ISO_SESSIONGRP; + if (conn->nym_epoch != circ->nym_epoch) + mixed |= ISO_NYM_EPOCH; + + if (dry_run) + return mixed; + + if ((mixed & conn->entry_cfg.isolation_flags) != 0) { + log_warn(LD_BUG, "Updating a circuit with seemingly incompatible " + "isolation flags."); + } + circ->isolation_flags_mixed |= mixed; + return 0; + } +} + +/** + * Clear the isolation settings on <b>circ</b>. + * + * This only works on an open circuit that has never had a stream attached to + * it, and whose isolation settings are hypothetical. (We set hypothetical + * isolation settings on circuits as we're launching them, so that we + * know whether they can handle more streams or whether we need to launch + * even more circuits. Once the circuit is open, if it turns out that + * we no longer have any streams to attach to it, we clear the isolation flags + * and data so that other streams can have a chance.) + */ +void +circuit_clear_isolation(origin_circuit_t *circ) +{ + if (circ->isolation_any_streams_attached) { + log_warn(LD_BUG, "Tried to clear the isolation status of a dirty circuit"); + return; + } + if (TO_CIRCUIT(circ)->state != CIRCUIT_STATE_OPEN) { + log_warn(LD_BUG, "Tried to clear the isolation status of a non-open " + "circuit"); + return; + } + + circ->isolation_values_set = 0; + circ->isolation_flags_mixed = 0; + circ->associated_isolated_stream_global_id = 0; + circ->client_proto_type = 0; + circ->client_proto_socksver = 0; + circ->dest_port = 0; + tor_addr_make_unspec(&circ->client_addr); + tor_free(circ->dest_address); + circ->session_group = -1; + circ->nym_epoch = 0; + if (circ->socks_username) { + memwipe(circ->socks_username, 0x11, circ->socks_username_len); + tor_free(circ->socks_username); + } + if (circ->socks_password) { + memwipe(circ->socks_password, 0x05, circ->socks_password_len); + tor_free(circ->socks_password); + } + circ->socks_username_len = circ->socks_password_len = 0; +} + +/** Free all storage held in module-scoped variables for connection_edge.c */ +void +connection_edge_free_all(void) +{ + untried_pending_connections = 0; + smartlist_free(pending_entry_connections); + pending_entry_connections = NULL; + mainloop_event_free(attach_pending_entry_connections_ev); +} diff --cc src/core/or/connection_edge.h index 24968b277,000000000..1348dd49f mode 100644,000000..100644 --- a/src/core/or/connection_edge.h +++ b/src/core/or/connection_edge.h @@@ -1,248 -1,0 +1,259 @@@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file connection_edge.h + * \brief Header file for connection_edge.c. + **/ + +#ifndef TOR_CONNECTION_EDGE_H +#define TOR_CONNECTION_EDGE_H + +#include "lib/testsupport/testsupport.h" + +edge_connection_t *TO_EDGE_CONN(connection_t *); +entry_connection_t *TO_ENTRY_CONN(connection_t *); +entry_connection_t *EDGE_TO_ENTRY_CONN(edge_connection_t *); + +#define EXIT_CONN_STATE_MIN_ 1 +/** State for an exit connection: waiting for response from DNS farm. */ +#define EXIT_CONN_STATE_RESOLVING 1 +/** State for an exit connection: waiting for connect() to finish. */ +#define EXIT_CONN_STATE_CONNECTING 2 +/** State for an exit connection: open and ready to transmit data. */ +#define EXIT_CONN_STATE_OPEN 3 +/** State for an exit connection: waiting to be removed. */ +#define EXIT_CONN_STATE_RESOLVEFAILED 4 +#define EXIT_CONN_STATE_MAX_ 4 + +/* The AP state values must be disjoint from the EXIT state values. */ +#define AP_CONN_STATE_MIN_ 5 +/** State for a SOCKS connection: waiting for SOCKS request. */ +#define AP_CONN_STATE_SOCKS_WAIT 5 +/** State for a SOCKS connection: got a y.onion URL; waiting to receive + * rendezvous descriptor. */ +#define AP_CONN_STATE_RENDDESC_WAIT 6 +/** The controller will attach this connection to a circuit; it isn't our + * job to do so. */ +#define AP_CONN_STATE_CONTROLLER_WAIT 7 +/** State for a SOCKS connection: waiting for a completed circuit. */ +#define AP_CONN_STATE_CIRCUIT_WAIT 8 +/** State for a SOCKS connection: sent BEGIN, waiting for CONNECTED. */ +#define AP_CONN_STATE_CONNECT_WAIT 9 +/** State for a SOCKS connection: sent RESOLVE, waiting for RESOLVED. */ +#define AP_CONN_STATE_RESOLVE_WAIT 10 +/** State for a SOCKS connection: ready to send and receive. */ +#define AP_CONN_STATE_OPEN 11 +/** State for a transparent natd connection: waiting for original + * destination. */ +#define AP_CONN_STATE_NATD_WAIT 12 +/** State for an HTTP tunnel: waiting for an HTTP CONNECT command. */ +#define AP_CONN_STATE_HTTP_CONNECT_WAIT 13 +#define AP_CONN_STATE_MAX_ 13 + +#define EXIT_PURPOSE_MIN_ 1 +/** This exit stream wants to do an ordinary connect. */ +#define EXIT_PURPOSE_CONNECT 1 +/** This exit stream wants to do a resolve (either normal or reverse). */ +#define EXIT_PURPOSE_RESOLVE 2 +#define EXIT_PURPOSE_MAX_ 2 + +/** True iff the AP_CONN_STATE_* value <b>s</b> means that the corresponding + * edge connection is not attached to any circuit. */ +#define AP_CONN_STATE_IS_UNATTACHED(s) \ + ((s) <= AP_CONN_STATE_CIRCUIT_WAIT || (s) == AP_CONN_STATE_NATD_WAIT) + +#define connection_mark_unattached_ap(conn, endreason) \ + connection_mark_unattached_ap_((conn), (endreason), __LINE__, SHORT_FILE__) + +MOCK_DECL(void,connection_mark_unattached_ap_, + (entry_connection_t *conn, int endreason, + int line, const char *file)); +int connection_edge_reached_eof(edge_connection_t *conn); +int connection_edge_process_inbuf(edge_connection_t *conn, + int package_partial); +int connection_edge_destroy(circid_t circ_id, edge_connection_t *conn); +int connection_edge_end(edge_connection_t *conn, uint8_t reason); +int connection_edge_end_errno(edge_connection_t *conn); +int connection_edge_flushed_some(edge_connection_t *conn); +int connection_edge_finished_flushing(edge_connection_t *conn); +int connection_edge_finished_connecting(edge_connection_t *conn); + +void connection_ap_about_to_close(entry_connection_t *edge_conn); +void connection_exit_about_to_close(edge_connection_t *edge_conn); + +MOCK_DECL(int, + connection_ap_handshake_send_begin,(entry_connection_t *ap_conn)); +int connection_ap_handshake_send_resolve(entry_connection_t *ap_conn); + +entry_connection_t *connection_ap_make_link(connection_t *partner, + char *address, uint16_t port, + const char *digest, + int session_group, + int isolation_flags, + int use_begindir, int want_onehop); +void connection_ap_handshake_socks_reply(entry_connection_t *conn, char *reply, + size_t replylen, + int endreason); +MOCK_DECL(void,connection_ap_handshake_socks_resolved, + (entry_connection_t *conn, + int answer_type, + size_t answer_len, + const uint8_t *answer, + int ttl, + time_t expires)); +void connection_ap_handshake_socks_resolved_addr(entry_connection_t *conn, + const tor_addr_t *answer, + int ttl, + time_t expires); + +int connection_exit_begin_conn(cell_t *cell, circuit_t *circ); +int connection_exit_begin_resolve(cell_t *cell, or_circuit_t *circ); +void connection_exit_connect(edge_connection_t *conn); +int connection_edge_is_rendezvous_stream(const edge_connection_t *conn); +int connection_ap_can_use_exit(const entry_connection_t *conn, + const node_t *exit); +void connection_ap_expire_beginning(void); +void connection_ap_rescan_and_attach_pending(void); +void connection_ap_attach_pending(int retry); +void connection_ap_mark_as_pending_circuit_(entry_connection_t *entry_conn, + const char *file, int line); +#define connection_ap_mark_as_pending_circuit(c) \ + connection_ap_mark_as_pending_circuit_((c), __FILE__, __LINE__) +void connection_ap_mark_as_non_pending_circuit(entry_connection_t *entry_conn); +#define CONNECTION_AP_EXPECT_NONPENDING(c) do { \ + if (ENTRY_TO_CONN(c)->state == AP_CONN_STATE_CIRCUIT_WAIT) { \ + log_warn(LD_BUG, "At %s:%d: %p was unexpectedly in circuit_wait.", \ + __FILE__, __LINE__, (c)); \ + connection_ap_mark_as_non_pending_circuit(c); \ + } \ + } while (0) +void connection_ap_fail_onehop(const char *failed_digest, + cpath_build_state_t *build_state); +void circuit_discard_optional_exit_enclaves(extend_info_t *info); +int connection_ap_detach_retriable(entry_connection_t *conn, + origin_circuit_t *circ, + int reason); +int connection_ap_process_transparent(entry_connection_t *conn); + +int address_is_invalid_destination(const char *address, int client); + +MOCK_DECL(int, connection_ap_rewrite_and_attach_if_allowed, + (entry_connection_t *conn, + origin_circuit_t *circ, + crypt_path_t *cpath)); +int connection_ap_handshake_rewrite_and_attach(entry_connection_t *conn, + origin_circuit_t *circ, + crypt_path_t *cpath); + +/** Possible return values for parse_extended_hostname. */ +typedef enum hostname_type_t { + NORMAL_HOSTNAME, ONION_V2_HOSTNAME, ONION_V3_HOSTNAME, + EXIT_HOSTNAME, BAD_HOSTNAME +} hostname_type_t; +hostname_type_t parse_extended_hostname(char *address); + +#if defined(HAVE_NET_IF_H) && defined(HAVE_NET_PFVAR_H) +int get_pf_socket(void); +#endif + +int connection_edge_compatible_with_circuit(const entry_connection_t *conn, + const origin_circuit_t *circ); +int connection_edge_update_circuit_isolation(const entry_connection_t *conn, + origin_circuit_t *circ, + int dry_run); +void circuit_clear_isolation(origin_circuit_t *circ); +streamid_t get_unique_stream_id_by_circ(origin_circuit_t *circ); + +void connection_edge_free_all(void); + +void connection_ap_warn_and_unmark_if_pending_circ( + entry_connection_t *entry_conn, + const char *where); + ++int connection_half_edge_is_valid_data(const smartlist_t *half_conns, ++ streamid_t stream_id); ++int connection_half_edge_is_valid_sendme(const smartlist_t *half_conns, ++ streamid_t stream_id); ++int connection_half_edge_is_valid_connected(const smartlist_t *half_conns, ++ streamid_t stream_id); ++int connection_half_edge_is_valid_end(smartlist_t *half_conns, ++ streamid_t stream_id); ++int connection_half_edge_is_valid_resolved(smartlist_t *half_conns, ++ streamid_t stream_id); ++ +/** @name Begin-cell flags + * + * These flags are used in RELAY_BEGIN cells to change the default behavior + * of the cell. + * + * @{ + **/ +/** When this flag is set, the client is willing to get connected to IPv6 + * addresses */ +#define BEGIN_FLAG_IPV6_OK (1u<<0) +/** When this flag is set, the client DOES NOT support connecting to IPv4 + * addresses. (The sense of this flag is inverted from IPV6_OK, so that the + * old default behavior of Tor is equivalent to having all flags set to 0.) + **/ +#define BEGIN_FLAG_IPV4_NOT_OK (1u<<1) +/** When this flag is set, if we find both an IPv4 and an IPv6 address, + * we use the IPv6 address. Otherwise we use the IPv4 address. */ +#define BEGIN_FLAG_IPV6_PREFERRED (1u<<2) +/**@}*/ + +#ifdef CONNECTION_EDGE_PRIVATE + +/** A parsed BEGIN or BEGIN_DIR cell */ +typedef struct begin_cell_t { + /** The address the client has asked us to connect to, or NULL if this is + * a BEGIN_DIR cell*/ + char *address; + /** The flags specified in the BEGIN cell's body. One or more of + * BEGIN_FLAG_*. */ + uint32_t flags; + /** The client's requested port. */ + uint16_t port; + /** The client's requested Stream ID */ + uint16_t stream_id; + /** True iff this is a BEGIN_DIR cell. */ + unsigned is_begindir : 1; +} begin_cell_t; + +STATIC int begin_cell_parse(const cell_t *cell, begin_cell_t *bcell, + uint8_t *end_reason_out); +STATIC int connected_cell_format_payload(uint8_t *payload_out, + const tor_addr_t *addr, + uint32_t ttl); + +typedef struct { + /** Original address, after we lowercased it but before we started + * mapping it. + */ + char orig_address[MAX_SOCKS_ADDR_LEN]; + /** True iff the address has been automatically remapped to a local + * address in VirtualAddrNetwork. (Only set true when we do a resolve + * and get a virtual address; not when we connect to the address.) */ + int automap; + /** If this connection has a .exit address, who put it there? */ + addressmap_entry_source_t exit_source; + /** If we've rewritten the address, when does this map expire? */ + time_t map_expires; + /** If we should close the connection, this is the end_reason to pass + * to connection_mark_unattached_ap */ + int end_reason; + /** True iff we should close the connection, either because of error or + * because of successful early RESOLVED reply. */ + int should_close; +} rewrite_result_t; + +STATIC void connection_ap_handshake_rewrite(entry_connection_t *conn, + rewrite_result_t *out); + +STATIC int connection_ap_process_http_connect(entry_connection_t *conn); +#endif /* defined(CONNECTION_EDGE_PRIVATE) */ + +#endif /* !defined(TOR_CONNECTION_EDGE_H) */ diff --cc src/core/or/half_edge_st.h index 000000000,000000000..5ed24dabe new file mode 100644 --- /dev/null +++ b/src/core/or/half_edge_st.h @@@ -1,0 -1,0 +1,34 @@@ ++/* Copyright (c) 2001 Matej Pfajfar. ++ * Copyright (c) 2001-2004, Roger Dingledine. ++ * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. ++ * Copyright (c) 2007-2018, The Tor Project, Inc. */ ++/* See LICENSE for licensing information */ ++ ++#ifndef HALF_EDGE_ST_H ++#define HALF_EDGE_ST_H ++ ++#include "core/or/or.h" ++ ++/** ++ * Struct to track a connection that we closed that the other end ++ * still thinks is open. Exists in origin_circuit_t.half_streams until ++ * we get an end cell or a resolved cell for this stream id. ++ */ ++typedef struct half_edge_t { ++ /** stream_id for the half-closed connection */ ++ streamid_t stream_id; ++ ++ /** How many sendme's can the other end still send, based on how ++ * much data we had sent at the time of close */ ++ int sendmes_pending; ++ ++ /** How much more data can the other end still send, based on ++ * our deliver window */ ++ int data_pending; ++ ++ /** Is there a connected cell pending? */ ++ int connected_pending : 1; ++} half_edge_t; ++ ++#endif ++ diff --cc src/core/or/origin_circuit_st.h index e7b864e82,000000000..26cdf590f mode 100644,000000..100644 --- a/src/core/or/origin_circuit_st.h +++ b/src/core/or/origin_circuit_st.h @@@ -1,290 -1,0 +1,294 @@@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +#ifndef ORIGIN_CIRCUIT_ST_H +#define ORIGIN_CIRCUIT_ST_H + +#include "core/or/or.h" + +#include "core/or/circuit_st.h" + +struct onion_queue_t; + +/** + * Describes the circuit building process in simplified terms based + * on the path bias accounting state for a circuit. + * + * NOTE: These state values are enumerated in the order for which we + * expect circuits to transition through them. If you add states, + * you need to preserve this overall ordering. The various pathbias + * state transition and accounting functions (pathbias_mark_* and + * pathbias_count_*) contain ordinal comparisons to enforce proper + * state transitions for corrections. + * + * This state machine and the associated logic was created to prevent + * miscounting due to unknown cases of circuit reuse. See also tickets + * #6475 and #7802. + */ +enum path_state_t { + /** This circuit is "new". It has not yet completed a first hop + * or been counted by the path bias code. */ + PATH_STATE_NEW_CIRC = 0, + /** This circuit has completed one/two hops, and has been counted by + * the path bias logic. */ + PATH_STATE_BUILD_ATTEMPTED = 1, + /** This circuit has been completely built */ + PATH_STATE_BUILD_SUCCEEDED = 2, + /** Did we try to attach any SOCKS streams or hidserv introductions to + * this circuit? + * + * Note: If we ever implement end-to-end stream timing through test + * stream probes (#5707), we must *not* set this for those probes + * (or any other automatic streams) because the adversary could + * just tag at a later point. + */ + PATH_STATE_USE_ATTEMPTED = 3, + /** Did any SOCKS streams or hidserv introductions actually succeed on + * this circuit? + * + * If any streams detatch/fail from this circuit, the code transitions + * the circuit back to PATH_STATE_USE_ATTEMPTED to ensure we probe. See + * pathbias_mark_use_rollback() for that. + */ + PATH_STATE_USE_SUCCEEDED = 4, + + /** + * This is a special state to indicate that we got a corrupted + * relay cell on a circuit and we don't intend to probe it. + */ + PATH_STATE_USE_FAILED = 5, + + /** + * This is a special state to indicate that we already counted + * the circuit. Used to guard against potential state machine + * violations. + */ + PATH_STATE_ALREADY_COUNTED = 6, +}; + +/** An origin_circuit_t holds data necessary to build and use a circuit. + */ +struct origin_circuit_t { + circuit_t base_; + + /** Linked list of AP streams (or EXIT streams if hidden service) + * associated with this circuit. */ + edge_connection_t *p_streams; + ++ /** Smartlist of half-closed streams (half_edge_t*) that still ++ * have pending activity */ ++ smartlist_t *half_streams; ++ + /** Bytes read on this circuit since last call to + * control_event_circ_bandwidth_used(). Only used if we're configured + * to emit CIRC_BW events. */ + uint32_t n_read_circ_bw; + + /** Bytes written to on this circuit since last call to + * control_event_circ_bandwidth_used(). Only used if we're configured + * to emit CIRC_BW events. */ + uint32_t n_written_circ_bw; + + /** Total known-valid relay cell bytes since last call to + * control_event_circ_bandwidth_used(). Only used if we're configured + * to emit CIRC_BW events. */ + uint32_t n_delivered_read_circ_bw; + + /** Total written relay cell bytes since last call to + * control_event_circ_bandwidth_used(). Only used if we're configured + * to emit CIRC_BW events. */ + uint32_t n_delivered_written_circ_bw; + + /** Total overhead data in all known-valid relay data cells since last + * call to control_event_circ_bandwidth_used(). Only used if we're + * configured to emit CIRC_BW events. */ + uint32_t n_overhead_read_circ_bw; + + /** Total written overhead data in all relay data cells since last call to + * control_event_circ_bandwidth_used(). Only used if we're configured + * to emit CIRC_BW events. */ + uint32_t n_overhead_written_circ_bw; + + /** Build state for this circuit. It includes the intended path + * length, the chosen exit router, rendezvous information, etc. + */ + cpath_build_state_t *build_state; + /** The doubly-linked list of crypt_path_t entries, one per hop, + * for this circuit. This includes ciphers for each hop, + * integrity-checking digests for each hop, and package/delivery + * windows for each hop. + */ + crypt_path_t *cpath; + + /** Holds all rendezvous data on either client or service side. */ + rend_data_t *rend_data; + + /** Holds hidden service identifier on either client or service side. This + * is for both introduction and rendezvous circuit. */ + struct hs_ident_circuit_t *hs_ident; + + /** Holds the data that the entry guard system uses to track the + * status of the guard this circuit is using, and thereby to determine + * whether this circuit can be used. */ + struct circuit_guard_state_t *guard_state; + + /** Index into global_origin_circuit_list for this circuit. -1 if not + * present. */ + int global_origin_circuit_list_idx; + + /** How many more relay_early cells can we send on this circuit, according + * to the specification? */ + unsigned int remaining_relay_early_cells : 4; + + /** Set if this circuit is insanely old and we already informed the user */ + unsigned int is_ancient : 1; + + /** Set if this circuit has already been opened. Used to detect + * cannibalized circuits. */ + unsigned int has_opened : 1; + + /** + * Path bias state machine. Used to ensure integrity of our + * circuit building and usage accounting. See path_state_t + * for more details. + */ + path_state_bitfield_t path_state : 3; + + /* If this flag is set, we should not consider attaching any more + * connections to this circuit. */ + unsigned int unusable_for_new_conns : 1; + + /** + * Tristate variable to guard against pathbias miscounting + * due to circuit purpose transitions changing the decision + * of pathbias_should_count(). This variable is informational + * only. The current results of pathbias_should_count() are + * the official decision for pathbias accounting. + */ + uint8_t pathbias_shouldcount; +#define PATHBIAS_SHOULDCOUNT_UNDECIDED 0 +#define PATHBIAS_SHOULDCOUNT_IGNORED 1 +#define PATHBIAS_SHOULDCOUNT_COUNTED 2 + + /** For path probing. Store the temporary probe stream ID + * for response comparison */ + streamid_t pathbias_probe_id; + + /** For path probing. Store the temporary probe address nonce + * (in host byte order) for response comparison. */ + uint32_t pathbias_probe_nonce; + + /** Set iff this is a hidden-service circuit which has timed out + * according to our current circuit-build timeout, but which has + * been kept around because it might still succeed in connecting to + * its destination, and which is not a fully-connected rendezvous + * circuit. + * + * (We clear this flag for client-side rendezvous circuits when they + * are 'joined' to the other side's rendezvous circuit, so that + * connection_ap_handshake_attach_circuit can put client streams on + * the circuit. We also clear this flag for service-side rendezvous + * circuits when they are 'joined' to a client's rend circ, but only + * for symmetry with the client case. Client-side introduction + * circuits are closed when we get a joined rend circ, and + * service-side introduction circuits never have this flag set.) */ + unsigned int hs_circ_has_timed_out : 1; + + /** Set iff this circuit has been given a relaxed timeout because + * no circuits have opened. Used to prevent spamming logs. */ + unsigned int relaxed_timeout : 1; + + /** Set iff this is a service-side rendezvous circuit for which a + * new connection attempt has been launched. We consider launching + * a new service-side rend circ to a client when the previous one + * fails; now that we don't necessarily close a service-side rend + * circ when we launch a new one to the same client, this flag keeps + * us from launching two retries for the same failed rend circ. */ + unsigned int hs_service_side_rend_circ_has_been_relaunched : 1; + + /** What commands were sent over this circuit that decremented the + * RELAY_EARLY counter? This is for debugging task 878. */ + uint8_t relay_early_commands[MAX_RELAY_EARLY_CELLS_PER_CIRCUIT]; + + /** How many RELAY_EARLY cells have been sent over this circuit? This is + * for debugging task 878, too. */ + int relay_early_cells_sent; + + /** The next stream_id that will be tried when we're attempting to + * construct a new AP stream originating at this circuit. */ + streamid_t next_stream_id; + + /* The intro key replaces the hidden service's public key if purpose is + * S_ESTABLISH_INTRO or S_INTRO, provided that no unversioned rendezvous + * descriptor is used. */ + crypto_pk_t *intro_key; + + /** Quasi-global identifier for this circuit; used for control.c */ + /* XXXX NM This can get re-used after 2**32 circuits. */ + uint32_t global_identifier; + + /** True if we have associated one stream to this circuit, thereby setting + * the isolation parameters for this circuit. Note that this doesn't + * necessarily mean that we've <em>attached</em> any streams to the circuit: + * we may only have marked up this circuit during the launch process. + */ + unsigned int isolation_values_set : 1; + /** True iff any stream has <em>ever</em> been attached to this circuit. + * + * In a better world we could use timestamp_dirty for this, but + * timestamp_dirty is far too overloaded at the moment. + */ + unsigned int isolation_any_streams_attached : 1; + + /** A bitfield of ISO_* flags for every isolation field such that this + * circuit has had streams with more than one value for that field + * attached to it. */ + uint8_t isolation_flags_mixed; + + /** @name Isolation parameters + * + * If any streams have been associated with this circ (isolation_values_set + * == 1), and all streams associated with the circuit have had the same + * value for some field ((isolation_flags_mixed & ISO_FOO) == 0), then these + * elements hold the value for that field. + * + * Note again that "associated" is not the same as "attached": we + * preliminarily associate streams with a circuit while the circuit is being + * launched, so that we can tell whether we need to launch more circuits. + * + * @{ + */ + uint8_t client_proto_type; + uint8_t client_proto_socksver; + uint16_t dest_port; + tor_addr_t client_addr; + char *dest_address; + int session_group; + unsigned nym_epoch; + size_t socks_username_len; + uint8_t socks_password_len; + /* Note that the next two values are NOT NUL-terminated; see + socks_username_len and socks_password_len for their lengths. */ + char *socks_username; + char *socks_password; + /** Global identifier for the first stream attached here; used by + * ISO_STREAM. */ + uint64_t associated_isolated_stream_global_id; + /**@}*/ + /** A list of addr_policy_t for this circuit in particular. Used by + * adjust_exit_policy_from_exitpolicy_failure. + */ + smartlist_t *prepend_policy; + + /** How long do we wait before closing this circuit if it remains + * completely idle after it was built, in seconds? This value + * is randomized on a per-circuit basis from CircuitsAvailableTimoeut + * to 2*CircuitsAvailableTimoeut. */ + int circuit_idle_timeout; + +}; + +#endif diff --cc src/core/or/relay.c index 6e1adfaff,000000000..8b58725ad mode 100644,000000..100644 --- a/src/core/or/relay.c +++ b/src/core/or/relay.c @@@ -1,3097 -1,0 +1,3168 @@@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file relay.c + * \brief Handle relay cell encryption/decryption, plus packaging and + * receiving from circuits, plus queuing on circuits. + * + * This is a core modules that makes Tor work. It's responsible for + * dealing with RELAY cells (the ones that travel more than one hop along a + * circuit), by: + * <ul> + * <li>constructing relays cells, + * <li>encrypting relay cells, + * <li>decrypting relay cells, + * <li>demultiplexing relay cells as they arrive on a connection, + * <li>queueing relay cells for retransmission, + * <li>or handling relay cells that are for us to receive (as an exit or a + * client). + * </ul> + * + * RELAY cells are generated throughout the code at the client or relay side, + * using relay_send_command_from_edge() or one of the functions like + * connection_edge_send_command() that calls it. Of particular interest is + * connection_edge_package_raw_inbuf(), which takes information that has + * arrived on an edge connection socket, and packages it as a RELAY_DATA cell + * -- this is how information is actually sent across the Tor network. The + * cryptography for these functions is handled deep in + * circuit_package_relay_cell(), which either adds a single layer of + * encryption (if we're an exit), or multiple layers (if we're the origin of + * the circuit). After construction and encryption, the RELAY cells are + * passed to append_cell_to_circuit_queue(), which queues them for + * transmission and tells the circuitmux (see circuitmux.c) that the circuit + * is waiting to send something. + * + * Incoming RELAY cells arrive at circuit_receive_relay_cell(), called from + * command.c. There they are decrypted and, if they are for us, are passed to + * connection_edge_process_relay_cell(). If they're not for us, they're + * re-queued for retransmission again with append_cell_to_circuit_queue(). + * + * The connection_edge_process_relay_cell() function handles all the different + * types of relay cells, launching requests or transmitting data as needed. + **/ + +#define RELAY_PRIVATE +#include "core/or/or.h" +#include "feature/client/addressmap.h" +#include "lib/err/backtrace.h" +#include "lib/container/buffers.h" +#include "core/or/channel.h" +#include "feature/client/circpathbias.h" +#include "core/or/circuitbuild.h" +#include "core/or/circuitlist.h" +#include "core/or/circuituse.h" +#include "lib/compress/compress.h" +#include "app/config/config.h" +#include "core/mainloop/connection.h" +#include "core/or/connection_edge.h" +#include "core/or/connection_or.h" +#include "feature/control/control.h" +#include "lib/crypt_ops/crypto_rand.h" +#include "lib/crypt_ops/crypto_util.h" +#include "feature/dircache/directory.h" +#include "feature/relay/dns.h" +#include "feature/stats/geoip.h" +#include "feature/hs/hs_cache.h" +#include "core/mainloop/main.h" +#include "feature/nodelist/networkstatus.h" +#include "feature/nodelist/nodelist.h" +#include "core/crypto/onion.h" +#include "core/or/policies.h" +#include "core/or/reasons.h" +#include "core/or/relay.h" +#include "core/crypto/relay_crypto.h" +#include "feature/rend/rendcache.h" +#include "feature/rend/rendcommon.h" +#include "feature/relay/router.h" +#include "feature/nodelist/routerlist.h" +#include "feature/nodelist/routerparse.h" +#include "core/or/scheduler.h" +#include "feature/stats/rephist.h" + +#include "core/or/cell_st.h" +#include "core/or/cell_queue_st.h" +#include "core/or/cpath_build_state_st.h" +#include "feature/dircommon/dir_connection_st.h" +#include "core/or/destroy_cell_queue_st.h" +#include "core/or/entry_connection_st.h" +#include "core/or/extend_info_st.h" +#include "core/or/or_circuit_st.h" +#include "core/or/origin_circuit_st.h" +#include "feature/nodelist/routerinfo_st.h" +#include "core/or/socks_request_st.h" + +#include "lib/intmath/weakrng.h" + +static edge_connection_t *relay_lookup_conn(circuit_t *circ, cell_t *cell, + cell_direction_t cell_direction, + crypt_path_t *layer_hint); + +static void circuit_consider_sending_sendme(circuit_t *circ, + crypt_path_t *layer_hint); +static void circuit_resume_edge_reading(circuit_t *circ, + crypt_path_t *layer_hint); +static int circuit_resume_edge_reading_helper(edge_connection_t *conn, + circuit_t *circ, + crypt_path_t *layer_hint); +static int circuit_consider_stop_edge_reading(circuit_t *circ, + crypt_path_t *layer_hint); +static int circuit_queue_streams_are_blocked(circuit_t *circ); +static void adjust_exit_policy_from_exitpolicy_failure(origin_circuit_t *circ, + entry_connection_t *conn, + node_t *node, + const tor_addr_t *addr); + +/** Stop reading on edge connections when we have this many cells + * waiting on the appropriate queue. */ +#define CELL_QUEUE_HIGHWATER_SIZE 256 +/** Start reading from edge connections again when we get down to this many + * cells. */ +#define CELL_QUEUE_LOWWATER_SIZE 64 + +/** Stats: how many relay cells have originated at this hop, or have + * been relayed onward (not recognized at this hop)? + */ +uint64_t stats_n_relay_cells_relayed = 0; +/** Stats: how many relay cells have been delivered to streams at this + * hop? + */ +uint64_t stats_n_relay_cells_delivered = 0; +/** Stats: how many circuits have we closed due to the cell queue limit being + * reached (see append_cell_to_circuit_queue()) */ +uint64_t stats_n_circ_max_cell_reached = 0; + +/** Used to tell which stream to read from first on a circuit. */ +static tor_weak_rng_t stream_choice_rng = TOR_WEAK_RNG_INIT; + +/** + * Update channel usage state based on the type of relay cell and + * circuit properties. + * + * This is needed to determine if a client channel is being + * used for application traffic, and if a relay channel is being + * used for multihop circuits and application traffic. The decision + * to pad in channelpadding.c depends upon this info (as well as + * consensus parameters) to decide what channels to pad. + */ +static void +circuit_update_channel_usage(circuit_t *circ, cell_t *cell) +{ + if (CIRCUIT_IS_ORIGIN(circ)) { + /* + * The client state was first set much earlier in + * circuit_send_next_onion_skin(), so we can start padding as early as + * possible. + * + * However, if padding turns out to be expensive, we may want to not do + * it until actual application traffic starts flowing (which is controlled + * via consensus param nf_pad_before_usage). + * + * So: If we're an origin circuit and we've created a full length circuit, + * then any CELL_RELAY cell means application data. Increase the usage + * state of the channel to indicate this. + * + * We want to wait for CELL_RELAY specifically here, so we know that + * the channel was definitely being used for data and not for extends. + * By default, we pad as soon as a channel has been used for *any* + * circuits, so this state is irrelevant to the padding decision in + * the default case. However, if padding turns out to be expensive, + * we would like the ability to avoid padding until we're absolutely + * sure that a channel is used for enough application data to be worth + * padding. + * + * (So it does not matter that CELL_RELAY_EARLY can actually contain + * application data. This is only a load reducing option and that edge + * case does not matter if we're desperately trying to reduce overhead + * anyway. See also consensus parameter nf_pad_before_usage). + */ + if (BUG(!circ->n_chan)) + return; + + if (circ->n_chan->channel_usage == CHANNEL_USED_FOR_FULL_CIRCS && + cell->command == CELL_RELAY) { + circ->n_chan->channel_usage = CHANNEL_USED_FOR_USER_TRAFFIC; + } + } else { + /* If we're a relay circuit, the question is more complicated. Basically: + * we only want to pad connections that carry multihop (anonymous) + * circuits. + * + * We assume we're more than one hop if either the previous hop + * is not a client, or if the previous hop is a client and there's + * a next hop. Then, circuit traffic starts at RELAY_EARLY, and + * user application traffic starts when we see RELAY cells. + */ + or_circuit_t *or_circ = TO_OR_CIRCUIT(circ); + + if (BUG(!or_circ->p_chan)) + return; + + if (!channel_is_client(or_circ->p_chan) || + (channel_is_client(or_circ->p_chan) && circ->n_chan)) { + if (cell->command == CELL_RELAY_EARLY) { + if (or_circ->p_chan->channel_usage < CHANNEL_USED_FOR_FULL_CIRCS) { + or_circ->p_chan->channel_usage = CHANNEL_USED_FOR_FULL_CIRCS; + } + } else if (cell->command == CELL_RELAY) { + or_circ->p_chan->channel_usage = CHANNEL_USED_FOR_USER_TRAFFIC; + } + } + } +} + +/** Receive a relay cell: + * - Crypt it (encrypt if headed toward the origin or if we <b>are</b> the + * origin; decrypt if we're headed toward the exit). + * - Check if recognized (if exitward). + * - If recognized and the digest checks out, then find if there's a stream + * that the cell is intended for, and deliver it to the right + * connection_edge. + * - If not recognized, then we need to relay it: append it to the appropriate + * cell_queue on <b>circ</b>. + * + * Return -<b>reason</b> on failure. + */ +int +circuit_receive_relay_cell(cell_t *cell, circuit_t *circ, + cell_direction_t cell_direction) +{ + channel_t *chan = NULL; + crypt_path_t *layer_hint=NULL; + char recognized=0; + int reason; + + tor_assert(cell); + tor_assert(circ); + tor_assert(cell_direction == CELL_DIRECTION_OUT || + cell_direction == CELL_DIRECTION_IN); + if (circ->marked_for_close) + return 0; + + if (relay_decrypt_cell(circ, cell, cell_direction, &layer_hint, &recognized) + < 0) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "relay crypt failed. Dropping connection."); + return -END_CIRC_REASON_INTERNAL; + } + + circuit_update_channel_usage(circ, cell); + + if (recognized) { + edge_connection_t *conn = NULL; + + if (circ->purpose == CIRCUIT_PURPOSE_PATH_BIAS_TESTING) { - pathbias_check_probe_response(circ, cell); ++ if (pathbias_check_probe_response(circ, cell) == -1) { ++ pathbias_count_valid_cells(circ, cell); ++ } + + /* We need to drop this cell no matter what to avoid code that expects + * a certain purpose (such as the hidserv code). */ + return 0; + } + + conn = relay_lookup_conn(circ, cell, cell_direction, layer_hint); + if (cell_direction == CELL_DIRECTION_OUT) { + ++stats_n_relay_cells_delivered; + log_debug(LD_OR,"Sending away from origin."); + if ((reason=connection_edge_process_relay_cell(cell, circ, conn, NULL)) + < 0) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "connection_edge_process_relay_cell (away from origin) " + "failed."); + return reason; + } + } + if (cell_direction == CELL_DIRECTION_IN) { + ++stats_n_relay_cells_delivered; + log_debug(LD_OR,"Sending to origin."); + if ((reason = connection_edge_process_relay_cell(cell, circ, conn, + layer_hint)) < 0) { + /* If a client is trying to connect to unknown hidden service port, + * END_CIRC_AT_ORIGIN is sent back so we can then close the circuit. + * Do not log warn as this is an expected behavior for a service. */ + if (reason != END_CIRC_AT_ORIGIN) { + log_warn(LD_OR, + "connection_edge_process_relay_cell (at origin) failed."); + } + return reason; + } + } + return 0; + } + + /* not recognized. pass it on. */ + if (cell_direction == CELL_DIRECTION_OUT) { + cell->circ_id = circ->n_circ_id; /* switch it */ + chan = circ->n_chan; + } else if (! CIRCUIT_IS_ORIGIN(circ)) { + cell->circ_id = TO_OR_CIRCUIT(circ)->p_circ_id; /* switch it */ + chan = TO_OR_CIRCUIT(circ)->p_chan; + } else { + log_fn(LOG_PROTOCOL_WARN, LD_OR, + "Dropping unrecognized inbound cell on origin circuit."); + /* If we see unrecognized cells on path bias testing circs, + * it's bad mojo. Those circuits need to die. + * XXX: Shouldn't they always die? */ + if (circ->purpose == CIRCUIT_PURPOSE_PATH_BIAS_TESTING) { + TO_ORIGIN_CIRCUIT(circ)->path_state = PATH_STATE_USE_FAILED; + return -END_CIRC_REASON_TORPROTOCOL; + } else { + return 0; + } + } + + if (!chan) { + // XXXX Can this splice stuff be done more cleanly? + if (! CIRCUIT_IS_ORIGIN(circ) && + TO_OR_CIRCUIT(circ)->rend_splice && + cell_direction == CELL_DIRECTION_OUT) { + or_circuit_t *splice_ = TO_OR_CIRCUIT(circ)->rend_splice; + tor_assert(circ->purpose == CIRCUIT_PURPOSE_REND_ESTABLISHED); + tor_assert(splice_->base_.purpose == CIRCUIT_PURPOSE_REND_ESTABLISHED); + cell->circ_id = splice_->p_circ_id; + cell->command = CELL_RELAY; /* can't be relay_early anyway */ + if ((reason = circuit_receive_relay_cell(cell, TO_CIRCUIT(splice_), + CELL_DIRECTION_IN)) < 0) { + log_warn(LD_REND, "Error relaying cell across rendezvous; closing " + "circuits"); + /* XXXX Do this here, or just return -1? */ + circuit_mark_for_close(circ, -reason); + return reason; + } + return 0; + } + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Didn't recognize cell, but circ stops here! Closing circ."); + return -END_CIRC_REASON_TORPROTOCOL; + } + + log_debug(LD_OR,"Passing on unrecognized cell."); + + ++stats_n_relay_cells_relayed; /* XXXX no longer quite accurate {cells} + * we might kill the circ before we relay + * the cells. */ + + append_cell_to_circuit_queue(circ, chan, cell, cell_direction, 0); + return 0; +} + +/** Package a relay cell from an edge: + * - Encrypt it to the right layer + * - Append it to the appropriate cell_queue on <b>circ</b>. + */ +static int +circuit_package_relay_cell(cell_t *cell, circuit_t *circ, + cell_direction_t cell_direction, + crypt_path_t *layer_hint, streamid_t on_stream, + const char *filename, int lineno) +{ + channel_t *chan; /* where to send the cell */ + + if (circ->marked_for_close) { + /* Circuit is marked; send nothing. */ + return 0; + } + + if (cell_direction == CELL_DIRECTION_OUT) { + chan = circ->n_chan; + if (!chan) { + log_warn(LD_BUG,"outgoing relay cell sent from %s:%d has n_chan==NULL." + " Dropping. Circuit is in state %s (%d), and is " + "%smarked for close. (%s:%d, %d)", filename, lineno, + circuit_state_to_string(circ->state), circ->state, + circ->marked_for_close ? "" : "not ", + circ->marked_for_close_file?circ->marked_for_close_file:"", + circ->marked_for_close, circ->marked_for_close_reason); + if (CIRCUIT_IS_ORIGIN(circ)) { + circuit_log_path(LOG_WARN, LD_BUG, TO_ORIGIN_CIRCUIT(circ)); + } + log_backtrace(LOG_WARN,LD_BUG,""); + return 0; /* just drop it */ + } + if (!CIRCUIT_IS_ORIGIN(circ)) { + log_warn(LD_BUG,"outgoing relay cell sent from %s:%d on non-origin " + "circ. Dropping.", filename, lineno); + log_backtrace(LOG_WARN,LD_BUG,""); + return 0; /* just drop it */ + } + + relay_encrypt_cell_outbound(cell, TO_ORIGIN_CIRCUIT(circ), layer_hint); + + /* Update circ written totals for control port */ + origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); + ocirc->n_written_circ_bw = tor_add_u32_nowrap(ocirc->n_written_circ_bw, + CELL_PAYLOAD_SIZE); + + } else { /* incoming cell */ + if (CIRCUIT_IS_ORIGIN(circ)) { + /* We should never package an _incoming_ cell from the circuit + * origin; that means we messed up somewhere. */ + log_warn(LD_BUG,"incoming relay cell at origin circuit. Dropping."); + assert_circuit_ok(circ); + return 0; /* just drop it */ + } + or_circuit_t *or_circ = TO_OR_CIRCUIT(circ); + relay_encrypt_cell_inbound(cell, or_circ); + chan = or_circ->p_chan; + } + ++stats_n_relay_cells_relayed; + + append_cell_to_circuit_queue(circ, chan, cell, cell_direction, on_stream); + return 0; +} + +/** If cell's stream_id matches the stream_id of any conn that's + * attached to circ, return that conn, else return NULL. + */ +static edge_connection_t * +relay_lookup_conn(circuit_t *circ, cell_t *cell, + cell_direction_t cell_direction, crypt_path_t *layer_hint) +{ + edge_connection_t *tmpconn; + relay_header_t rh; + + relay_header_unpack(&rh, cell->payload); + + if (!rh.stream_id) + return NULL; + + /* IN or OUT cells could have come from either direction, now + * that we allow rendezvous *to* an OP. + */ + + if (CIRCUIT_IS_ORIGIN(circ)) { + for (tmpconn = TO_ORIGIN_CIRCUIT(circ)->p_streams; tmpconn; + tmpconn=tmpconn->next_stream) { + if (rh.stream_id == tmpconn->stream_id && + !tmpconn->base_.marked_for_close && + tmpconn->cpath_layer == layer_hint) { + log_debug(LD_APP,"found conn for stream %d.", rh.stream_id); + return tmpconn; + } + } + } else { + for (tmpconn = TO_OR_CIRCUIT(circ)->n_streams; tmpconn; + tmpconn=tmpconn->next_stream) { + if (rh.stream_id == tmpconn->stream_id && + !tmpconn->base_.marked_for_close) { + log_debug(LD_EXIT,"found conn for stream %d.", rh.stream_id); + if (cell_direction == CELL_DIRECTION_OUT || + connection_edge_is_rendezvous_stream(tmpconn)) + return tmpconn; + } + } + for (tmpconn = TO_OR_CIRCUIT(circ)->resolving_streams; tmpconn; + tmpconn=tmpconn->next_stream) { + if (rh.stream_id == tmpconn->stream_id && + !tmpconn->base_.marked_for_close) { + log_debug(LD_EXIT,"found conn for stream %d.", rh.stream_id); + return tmpconn; + } + } + } + return NULL; /* probably a begin relay cell */ +} + +/** Pack the relay_header_t host-order structure <b>src</b> into + * network-order in the buffer <b>dest</b>. See tor-spec.txt for details + * about the wire format. + */ +void +relay_header_pack(uint8_t *dest, const relay_header_t *src) +{ + set_uint8(dest, src->command); + set_uint16(dest+1, htons(src->recognized)); + set_uint16(dest+3, htons(src->stream_id)); + memcpy(dest+5, src->integrity, 4); + set_uint16(dest+9, htons(src->length)); +} + +/** Unpack the network-order buffer <b>src</b> into a host-order + * relay_header_t structure <b>dest</b>. + */ +void +relay_header_unpack(relay_header_t *dest, const uint8_t *src) +{ + dest->command = get_uint8(src); + dest->recognized = ntohs(get_uint16(src+1)); + dest->stream_id = ntohs(get_uint16(src+3)); + memcpy(dest->integrity, src+5, 4); + dest->length = ntohs(get_uint16(src+9)); +} + +/** Convert the relay <b>command</b> into a human-readable string. */ +static const char * +relay_command_to_string(uint8_t command) +{ + static char buf[64]; + switch (command) { + case RELAY_COMMAND_BEGIN: return "BEGIN"; + case RELAY_COMMAND_DATA: return "DATA"; + case RELAY_COMMAND_END: return "END"; + case RELAY_COMMAND_CONNECTED: return "CONNECTED"; + case RELAY_COMMAND_SENDME: return "SENDME"; + case RELAY_COMMAND_EXTEND: return "EXTEND"; + case RELAY_COMMAND_EXTENDED: return "EXTENDED"; + case RELAY_COMMAND_TRUNCATE: return "TRUNCATE"; + case RELAY_COMMAND_TRUNCATED: return "TRUNCATED"; + case RELAY_COMMAND_DROP: return "DROP"; + case RELAY_COMMAND_RESOLVE: return "RESOLVE"; + case RELAY_COMMAND_RESOLVED: return "RESOLVED"; + case RELAY_COMMAND_BEGIN_DIR: return "BEGIN_DIR"; + case RELAY_COMMAND_ESTABLISH_INTRO: return "ESTABLISH_INTRO"; + case RELAY_COMMAND_ESTABLISH_RENDEZVOUS: return "ESTABLISH_RENDEZVOUS"; + case RELAY_COMMAND_INTRODUCE1: return "INTRODUCE1"; + case RELAY_COMMAND_INTRODUCE2: return "INTRODUCE2"; + case RELAY_COMMAND_RENDEZVOUS1: return "RENDEZVOUS1"; + case RELAY_COMMAND_RENDEZVOUS2: return "RENDEZVOUS2"; + case RELAY_COMMAND_INTRO_ESTABLISHED: return "INTRO_ESTABLISHED"; + case RELAY_COMMAND_RENDEZVOUS_ESTABLISHED: + return "RENDEZVOUS_ESTABLISHED"; + case RELAY_COMMAND_INTRODUCE_ACK: return "INTRODUCE_ACK"; + case RELAY_COMMAND_EXTEND2: return "EXTEND2"; + case RELAY_COMMAND_EXTENDED2: return "EXTENDED2"; + default: + tor_snprintf(buf, sizeof(buf), "Unrecognized relay command %u", + (unsigned)command); + return buf; + } +} + +/** Make a relay cell out of <b>relay_command</b> and <b>payload</b>, and send + * it onto the open circuit <b>circ</b>. <b>stream_id</b> is the ID on + * <b>circ</b> for the stream that's sending the relay cell, or 0 if it's a + * control cell. <b>cpath_layer</b> is NULL for OR->OP cells, or the + * destination hop for OP->OR cells. + * + * If you can't send the cell, mark the circuit for close and return -1. Else + * return 0. + */ +MOCK_IMPL(int, +relay_send_command_from_edge_,(streamid_t stream_id, circuit_t *circ, + uint8_t relay_command, const char *payload, + size_t payload_len, crypt_path_t *cpath_layer, + const char *filename, int lineno)) +{ + cell_t cell; + relay_header_t rh; + cell_direction_t cell_direction; + /* XXXX NM Split this function into a separate versions per circuit type? */ + + tor_assert(circ); + tor_assert(payload_len <= RELAY_PAYLOAD_SIZE); + + memset(&cell, 0, sizeof(cell_t)); + cell.command = CELL_RELAY; + if (CIRCUIT_IS_ORIGIN(circ)) { + tor_assert(cpath_layer); + cell.circ_id = circ->n_circ_id; + cell_direction = CELL_DIRECTION_OUT; + } else { + tor_assert(! cpath_layer); + cell.circ_id = TO_OR_CIRCUIT(circ)->p_circ_id; + cell_direction = CELL_DIRECTION_IN; + } + + memset(&rh, 0, sizeof(rh)); + rh.command = relay_command; + rh.stream_id = stream_id; + rh.length = payload_len; + relay_header_pack(cell.payload, &rh); + if (payload_len) + memcpy(cell.payload+RELAY_HEADER_SIZE, payload, payload_len); + + log_debug(LD_OR,"delivering %d cell %s.", relay_command, + cell_direction == CELL_DIRECTION_OUT ? "forward" : "backward"); + + if (relay_command == RELAY_COMMAND_DROP) + rep_hist_padding_count_write(PADDING_TYPE_DROP); + + /* If we are sending an END cell and this circuit is used for a tunneled + * directory request, advance its state. */ + if (relay_command == RELAY_COMMAND_END && circ->dirreq_id) + geoip_change_dirreq_state(circ->dirreq_id, DIRREQ_TUNNELED, + DIRREQ_END_CELL_SENT); + + if (cell_direction == CELL_DIRECTION_OUT && circ->n_chan) { + /* if we're using relaybandwidthrate, this conn wants priority */ + channel_timestamp_client(circ->n_chan); + } + + if (cell_direction == CELL_DIRECTION_OUT) { + origin_circuit_t *origin_circ = TO_ORIGIN_CIRCUIT(circ); + if (origin_circ->remaining_relay_early_cells > 0 && + (relay_command == RELAY_COMMAND_EXTEND || + relay_command == RELAY_COMMAND_EXTEND2 || + cpath_layer != origin_circ->cpath)) { + /* If we've got any relay_early cells left and (we're sending + * an extend cell or we're not talking to the first hop), use + * one of them. Don't worry about the conn protocol version: + * append_cell_to_circuit_queue will fix it up. */ + cell.command = CELL_RELAY_EARLY; + --origin_circ->remaining_relay_early_cells; + log_debug(LD_OR, "Sending a RELAY_EARLY cell; %d remaining.", + (int)origin_circ->remaining_relay_early_cells); + /* Memorize the command that is sent as RELAY_EARLY cell; helps debug + * task 878. */ + origin_circ->relay_early_commands[ + origin_circ->relay_early_cells_sent++] = relay_command; + } else if (relay_command == RELAY_COMMAND_EXTEND || + relay_command == RELAY_COMMAND_EXTEND2) { + /* If no RELAY_EARLY cells can be sent over this circuit, log which + * commands have been sent as RELAY_EARLY cells before; helps debug + * task 878. */ + smartlist_t *commands_list = smartlist_new(); + int i = 0; + char *commands = NULL; + for (; i < origin_circ->relay_early_cells_sent; i++) + smartlist_add(commands_list, (char *) + relay_command_to_string(origin_circ->relay_early_commands[i])); + commands = smartlist_join_strings(commands_list, ",", 0, NULL); + log_warn(LD_BUG, "Uh-oh. We're sending a RELAY_COMMAND_EXTEND cell, " + "but we have run out of RELAY_EARLY cells on that circuit. " + "Commands sent before: %s", commands); + tor_free(commands); + smartlist_free(commands_list); + } + + /* Let's assume we're well-behaved: Anything that we decide to send is + * valid, delivered data. */ + circuit_sent_valid_data(origin_circ, rh.length); + } + + if (circuit_package_relay_cell(&cell, circ, cell_direction, cpath_layer, + stream_id, filename, lineno) < 0) { + log_warn(LD_BUG,"circuit_package_relay_cell failed. Closing."); + circuit_mark_for_close(circ, END_CIRC_REASON_INTERNAL); + return -1; + } + return 0; +} + +/** Make a relay cell out of <b>relay_command</b> and <b>payload</b>, and + * send it onto the open circuit <b>circ</b>. <b>fromconn</b> is the stream + * that's sending the relay cell, or NULL if it's a control cell. + * <b>cpath_layer</b> is NULL for OR->OP cells, or the destination hop + * for OP->OR cells. + * + * If you can't send the cell, mark the circuit for close and + * return -1. Else return 0. + */ +int +connection_edge_send_command(edge_connection_t *fromconn, + uint8_t relay_command, const char *payload, + size_t payload_len) +{ + /* XXXX NM Split this function into a separate versions per circuit type? */ + circuit_t *circ; + crypt_path_t *cpath_layer = fromconn->cpath_layer; + tor_assert(fromconn); + circ = fromconn->on_circuit; + + if (fromconn->base_.marked_for_close) { + log_warn(LD_BUG, + "called on conn that's already marked for close at %s:%d.", + fromconn->base_.marked_for_close_file, + fromconn->base_.marked_for_close); + return 0; + } + + if (!circ) { + if (fromconn->base_.type == CONN_TYPE_AP) { + log_info(LD_APP,"no circ. Closing conn."); + connection_mark_unattached_ap(EDGE_TO_ENTRY_CONN(fromconn), + END_STREAM_REASON_INTERNAL); + } else { + log_info(LD_EXIT,"no circ. Closing conn."); + fromconn->edge_has_sent_end = 1; /* no circ to send to */ + fromconn->end_reason = END_STREAM_REASON_INTERNAL; + connection_mark_for_close(TO_CONN(fromconn)); + } + return -1; + } + + if (circ->marked_for_close) { + /* The circuit has been marked, but not freed yet. When it's freed, it + * will mark this connection for close. */ + return -1; + } + +#ifdef MEASUREMENTS_21206 + /* Keep track of the number of RELAY_DATA cells sent for directory + * connections. */ + connection_t *linked_conn = TO_CONN(fromconn)->linked_conn; + + if (linked_conn && linked_conn->type == CONN_TYPE_DIR) { + ++(TO_DIR_CONN(linked_conn)->data_cells_sent); + } +#endif /* defined(MEASUREMENTS_21206) */ + + return relay_send_command_from_edge(fromconn->stream_id, circ, + relay_command, payload, + payload_len, cpath_layer); +} + +/** How many times will I retry a stream that fails due to DNS + * resolve failure or misc error? + */ +#define MAX_RESOLVE_FAILURES 3 + +/** Return 1 if reason is something that you should retry if you + * get the end cell before you've connected; else return 0. */ +static int +edge_reason_is_retriable(int reason) +{ + return reason == END_STREAM_REASON_HIBERNATING || + reason == END_STREAM_REASON_RESOURCELIMIT || + reason == END_STREAM_REASON_EXITPOLICY || + reason == END_STREAM_REASON_RESOLVEFAILED || + reason == END_STREAM_REASON_MISC || + reason == END_STREAM_REASON_NOROUTE; +} + +/** Called when we receive an END cell on a stream that isn't open yet, + * from the client side. + * Arguments are as for connection_edge_process_relay_cell(). + */ +static int +connection_ap_process_end_not_open( + relay_header_t *rh, cell_t *cell, origin_circuit_t *circ, + entry_connection_t *conn, crypt_path_t *layer_hint) +{ + node_t *exitrouter; + int reason = *(cell->payload+RELAY_HEADER_SIZE); + int control_reason; + edge_connection_t *edge_conn = ENTRY_TO_EDGE_CONN(conn); + (void) layer_hint; /* unused */ + + if (rh->length > 0) { + if (reason == END_STREAM_REASON_TORPROTOCOL || + reason == END_STREAM_REASON_DESTROY) { + /* Both of these reasons could mean a failed tag + * hit the exit and it complained. Do not probe. + * Fail the circuit. */ + circ->path_state = PATH_STATE_USE_FAILED; + return -END_CIRC_REASON_TORPROTOCOL; + } else if (reason == END_STREAM_REASON_INTERNAL) { + /* We can't infer success or failure, since older Tors report + * ENETUNREACH as END_STREAM_REASON_INTERNAL. */ + } else { + /* Path bias: If we get a valid reason code from the exit, + * it wasn't due to tagging. + * + * We rely on recognized+digest being strong enough to make + * tags unlikely to allow us to get tagged, yet 'recognized' + * reason codes here. */ + pathbias_mark_use_success(circ); + } + } + + /* This end cell is now valid. */ + circuit_read_valid_data(circ, rh->length); + + if (rh->length == 0) { + reason = END_STREAM_REASON_MISC; + } + + control_reason = reason | END_STREAM_REASON_FLAG_REMOTE; + + if (edge_reason_is_retriable(reason) && + /* avoid retry if rend */ + !connection_edge_is_rendezvous_stream(edge_conn)) { + const char *chosen_exit_digest = + circ->build_state->chosen_exit->identity_digest; + log_info(LD_APP,"Address '%s' refused due to '%s'. Considering retrying.", + safe_str(conn->socks_request->address), + stream_end_reason_to_string(reason)); + exitrouter = node_get_mutable_by_id(chosen_exit_digest); + switch (reason) { + case END_STREAM_REASON_EXITPOLICY: { + tor_addr_t addr; + tor_addr_make_unspec(&addr); + if (rh->length >= 5) { + int ttl = -1; + tor_addr_make_unspec(&addr); + if (rh->length == 5 || rh->length == 9) { + tor_addr_from_ipv4n(&addr, + get_uint32(cell->payload+RELAY_HEADER_SIZE+1)); + if (rh->length == 9) + ttl = (int)ntohl(get_uint32(cell->payload+RELAY_HEADER_SIZE+5)); + } else if (rh->length == 17 || rh->length == 21) { + tor_addr_from_ipv6_bytes(&addr, + (char*)(cell->payload+RELAY_HEADER_SIZE+1)); + if (rh->length == 21) + ttl = (int)ntohl(get_uint32(cell->payload+RELAY_HEADER_SIZE+17)); + } + if (tor_addr_is_null(&addr)) { + log_info(LD_APP,"Address '%s' resolved to 0.0.0.0. Closing,", + safe_str(conn->socks_request->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return 0; + } + + if ((tor_addr_family(&addr) == AF_INET && + !conn->entry_cfg.ipv4_traffic) || + (tor_addr_family(&addr) == AF_INET6 && + !conn->entry_cfg.ipv6_traffic)) { + log_fn(LOG_PROTOCOL_WARN, LD_APP, + "Got an EXITPOLICY failure on a connection with a " + "mismatched family. Closing."); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return 0; + } + if (get_options()->ClientDNSRejectInternalAddresses && + tor_addr_is_internal(&addr, 0)) { + log_info(LD_APP,"Address '%s' resolved to internal. Closing,", + safe_str(conn->socks_request->address)); + connection_mark_unattached_ap(conn, END_STREAM_REASON_TORPROTOCOL); + return 0; + } + + client_dns_set_addressmap(conn, + conn->socks_request->address, &addr, + conn->chosen_exit_name, ttl); + + { + char new_addr[TOR_ADDR_BUF_LEN]; + tor_addr_to_str(new_addr, &addr, sizeof(new_addr), 1); + if (strcmp(conn->socks_request->address, new_addr)) { + strlcpy(conn->socks_request->address, new_addr, + sizeof(conn->socks_request->address)); + control_event_stream_status(conn, STREAM_EVENT_REMAP, 0); + } + } + } + /* check if the exit *ought* to have allowed it */ + + adjust_exit_policy_from_exitpolicy_failure(circ, + conn, + exitrouter, + &addr); + + if (conn->chosen_exit_optional || + conn->chosen_exit_retries) { + /* stop wanting a specific exit */ + conn->chosen_exit_optional = 0; + /* A non-zero chosen_exit_retries can happen if we set a + * TrackHostExits for this address under a port that the exit + * relay allows, but then try the same address with a different + * port that it doesn't allow to exit. We shouldn't unregister + * the mapping, since it is probably still wanted on the + * original port. But now we give away to the exit relay that + * we probably have a TrackHostExits on it. So be it. */ + conn->chosen_exit_retries = 0; + tor_free(conn->chosen_exit_name); /* clears it */ + } + if (connection_ap_detach_retriable(conn, circ, control_reason) >= 0) + return 0; + /* else, conn will get closed below */ + break; + } + case END_STREAM_REASON_CONNECTREFUSED: + if (!conn->chosen_exit_optional) + break; /* break means it'll close, below */ + /* Else fall through: expire this circuit, clear the + * chosen_exit_name field, and try again. */ + /* Falls through. */ + case END_STREAM_REASON_RESOLVEFAILED: + case END_STREAM_REASON_TIMEOUT: + case END_STREAM_REASON_MISC: + case END_STREAM_REASON_NOROUTE: + if (client_dns_incr_failures(conn->socks_request->address) + < MAX_RESOLVE_FAILURES) { + /* We haven't retried too many times; reattach the connection. */ + circuit_log_path(LOG_INFO,LD_APP,circ); + /* Mark this circuit "unusable for new streams". */ + mark_circuit_unusable_for_new_conns(circ); + + if (conn->chosen_exit_optional) { + /* stop wanting a specific exit */ + conn->chosen_exit_optional = 0; + tor_free(conn->chosen_exit_name); /* clears it */ + } + if (connection_ap_detach_retriable(conn, circ, control_reason) >= 0) + return 0; + /* else, conn will get closed below */ + } else { + log_notice(LD_APP, + "Have tried resolving or connecting to address '%s' " + "at %d different places. Giving up.", + safe_str(conn->socks_request->address), + MAX_RESOLVE_FAILURES); + /* clear the failures, so it will have a full try next time */ + client_dns_clear_failures(conn->socks_request->address); + } + break; + case END_STREAM_REASON_HIBERNATING: + case END_STREAM_REASON_RESOURCELIMIT: + if (exitrouter) { + policies_set_node_exitpolicy_to_reject_all(exitrouter); + } + if (conn->chosen_exit_optional) { + /* stop wanting a specific exit */ + conn->chosen_exit_optional = 0; + tor_free(conn->chosen_exit_name); /* clears it */ + } + if (connection_ap_detach_retriable(conn, circ, control_reason) >= 0) + return 0; + /* else, will close below */ + break; + } /* end switch */ + log_info(LD_APP,"Giving up on retrying; conn can't be handled."); + } + + log_info(LD_APP, + "Edge got end (%s) before we're connected. Marking for close.", + stream_end_reason_to_string(rh->length > 0 ? reason : -1)); + circuit_log_path(LOG_INFO,LD_APP,circ); + /* need to test because of detach_retriable */ + if (!ENTRY_TO_CONN(conn)->marked_for_close) + connection_mark_unattached_ap(conn, control_reason); + return 0; +} + +/** Called when we have gotten an END_REASON_EXITPOLICY failure on <b>circ</b> + * for <b>conn</b>, while attempting to connect via <b>node</b>. If the node + * told us which address it rejected, then <b>addr</b> is that address; + * otherwise it is AF_UNSPEC. + * + * If we are sure the node should have allowed this address, mark the node as + * having a reject *:* exit policy. Otherwise, mark the circuit as unusable + * for this particular address. + **/ +static void +adjust_exit_policy_from_exitpolicy_failure(origin_circuit_t *circ, + entry_connection_t *conn, + node_t *node, + const tor_addr_t *addr) +{ + int make_reject_all = 0; + const sa_family_t family = tor_addr_family(addr); + + if (node) { + tor_addr_t tmp; + int asked_for_family = tor_addr_parse(&tmp, conn->socks_request->address); + if (family == AF_UNSPEC) { + make_reject_all = 1; + } else if (node_exit_policy_is_exact(node, family) && + asked_for_family != -1 && !conn->chosen_exit_name) { + make_reject_all = 1; + } + + if (make_reject_all) { + log_info(LD_APP, + "Exitrouter %s seems to be more restrictive than its exit " + "policy. Not using this router as exit for now.", + node_describe(node)); + policies_set_node_exitpolicy_to_reject_all(node); + } + } + + if (family != AF_UNSPEC) + addr_policy_append_reject_addr(&circ->prepend_policy, addr); +} + +/** Helper: change the socks_request->address field on conn to the + * dotted-quad representation of <b>new_addr</b>, + * and send an appropriate REMAP event. */ +static void +remap_event_helper(entry_connection_t *conn, const tor_addr_t *new_addr) +{ + tor_addr_to_str(conn->socks_request->address, new_addr, + sizeof(conn->socks_request->address), + 1); + control_event_stream_status(conn, STREAM_EVENT_REMAP, + REMAP_STREAM_SOURCE_EXIT); +} + +/** Extract the contents of a connected cell in <b>cell</b>, whose relay + * header has already been parsed into <b>rh</b>. On success, set + * <b>addr_out</b> to the address we're connected to, and <b>ttl_out</b> to + * the ttl of that address, in seconds, and return 0. On failure, return + * -1. + * + * Note that the resulting address can be UNSPEC if the connected cell had no + * address (as for a stream to an union service or a tunneled directory + * connection), and that the ttl can be absent (in which case <b>ttl_out</b> + * is set to -1). */ +STATIC int +connected_cell_parse(const relay_header_t *rh, const cell_t *cell, + tor_addr_t *addr_out, int *ttl_out) +{ + uint32_t bytes; + const uint8_t *payload = cell->payload + RELAY_HEADER_SIZE; + + tor_addr_make_unspec(addr_out); + *ttl_out = -1; + if (rh->length == 0) + return 0; + if (rh->length < 4) + return -1; + bytes = ntohl(get_uint32(payload)); + + /* If bytes is 0, this is maybe a v6 address. Otherwise it's a v4 address */ + if (bytes != 0) { + /* v4 address */ + tor_addr_from_ipv4h(addr_out, bytes); + if (rh->length >= 8) { + bytes = ntohl(get_uint32(payload + 4)); + if (bytes <= INT32_MAX) + *ttl_out = bytes; + } + } else { + if (rh->length < 25) /* 4 bytes of 0s, 1 addr, 16 ipv4, 4 ttl. */ + return -1; + if (get_uint8(payload + 4) != 6) + return -1; + tor_addr_from_ipv6_bytes(addr_out, (char*)(payload + 5)); + bytes = ntohl(get_uint32(payload + 21)); + if (bytes <= INT32_MAX) + *ttl_out = (int) bytes; + } + return 0; +} + +/** Drop all storage held by <b>addr</b>. */ +STATIC void +address_ttl_free_(address_ttl_t *addr) +{ + if (!addr) + return; + tor_free(addr->hostname); + tor_free(addr); +} + +/** Parse a resolved cell in <b>cell</b>, with parsed header in <b>rh</b>. + * Return -1 on parse error. On success, add one or more newly allocated + * address_ttl_t to <b>addresses_out</b>; set *<b>errcode_out</b> to + * one of 0, RESOLVED_TYPE_ERROR, or RESOLVED_TYPE_ERROR_TRANSIENT, and + * return 0. */ +STATIC int +resolved_cell_parse(const cell_t *cell, const relay_header_t *rh, + smartlist_t *addresses_out, int *errcode_out) +{ + const uint8_t *cp; + uint8_t answer_type; + size_t answer_len; + address_ttl_t *addr; + size_t remaining; + int errcode = 0; + smartlist_t *addrs; + + tor_assert(cell); + tor_assert(rh); + tor_assert(addresses_out); + tor_assert(errcode_out); + + *errcode_out = 0; + + if (rh->length > RELAY_PAYLOAD_SIZE) + return -1; + + addrs = smartlist_new(); + + cp = cell->payload + RELAY_HEADER_SIZE; + + remaining = rh->length; + while (remaining) { + const uint8_t *cp_orig = cp; + if (remaining < 2) + goto err; + answer_type = *cp++; + answer_len = *cp++; + if (remaining < 2 + answer_len + 4) { + goto err; + } + if (answer_type == RESOLVED_TYPE_IPV4) { + if (answer_len != 4) { + goto err; + } + addr = tor_malloc_zero(sizeof(*addr)); + tor_addr_from_ipv4n(&addr->addr, get_uint32(cp)); + cp += 4; + addr->ttl = ntohl(get_uint32(cp)); + cp += 4; + smartlist_add(addrs, addr); + } else if (answer_type == RESOLVED_TYPE_IPV6) { + if (answer_len != 16) + goto err; + addr = tor_malloc_zero(sizeof(*addr)); + tor_addr_from_ipv6_bytes(&addr->addr, (const char*) cp); + cp += 16; + addr->ttl = ntohl(get_uint32(cp)); + cp += 4; + smartlist_add(addrs, addr); + } else if (answer_type == RESOLVED_TYPE_HOSTNAME) { + if (answer_len == 0) { + goto err; + } + addr = tor_malloc_zero(sizeof(*addr)); + addr->hostname = tor_memdup_nulterm(cp, answer_len); + cp += answer_len; + addr->ttl = ntohl(get_uint32(cp)); + cp += 4; + smartlist_add(addrs, addr); + } else if (answer_type == RESOLVED_TYPE_ERROR_TRANSIENT || + answer_type == RESOLVED_TYPE_ERROR) { + errcode = answer_type; + /* Ignore the error contents */ + cp += answer_len + 4; + } else { + cp += answer_len + 4; + } + tor_assert(((ssize_t)remaining) >= (cp - cp_orig)); + remaining -= (cp - cp_orig); + } + + if (errcode && smartlist_len(addrs) == 0) { + /* Report an error only if there were no results. */ + *errcode_out = errcode; + } + + smartlist_add_all(addresses_out, addrs); + smartlist_free(addrs); + + return 0; + + err: + /* On parse error, don't report any results */ + SMARTLIST_FOREACH(addrs, address_ttl_t *, a, address_ttl_free(a)); + smartlist_free(addrs); + return -1; +} + +/** Helper for connection_edge_process_resolved_cell: given an error code, + * an entry_connection, and a list of address_ttl_t *, report the best answer + * to the entry_connection. */ +static void +connection_ap_handshake_socks_got_resolved_cell(entry_connection_t *conn, + int error_code, + smartlist_t *results) +{ + address_ttl_t *addr_ipv4 = NULL; + address_ttl_t *addr_ipv6 = NULL; + address_ttl_t *addr_hostname = NULL; + address_ttl_t *addr_best = NULL; + + /* If it's an error code, that's easy. */ + if (error_code) { + tor_assert(error_code == RESOLVED_TYPE_ERROR || + error_code == RESOLVED_TYPE_ERROR_TRANSIENT); + connection_ap_handshake_socks_resolved(conn, + error_code,0,NULL,-1,-1); + return; + } + + /* Get the first answer of each type. */ + SMARTLIST_FOREACH_BEGIN(results, address_ttl_t *, addr) { + if (addr->hostname) { + if (!addr_hostname) { + addr_hostname = addr; + } + } else if (tor_addr_family(&addr->addr) == AF_INET) { + if (!addr_ipv4 && conn->entry_cfg.ipv4_traffic) { + addr_ipv4 = addr; + } + } else if (tor_addr_family(&addr->addr) == AF_INET6) { + if (!addr_ipv6 && conn->entry_cfg.ipv6_traffic) { + addr_ipv6 = addr; + } + } + } SMARTLIST_FOREACH_END(addr); + + /* Now figure out which type we wanted to deliver. */ + if (conn->socks_request->command == SOCKS_COMMAND_RESOLVE_PTR) { + if (addr_hostname) { + connection_ap_handshake_socks_resolved(conn, + RESOLVED_TYPE_HOSTNAME, + strlen(addr_hostname->hostname), + (uint8_t*)addr_hostname->hostname, + addr_hostname->ttl,-1); + } else { + connection_ap_handshake_socks_resolved(conn, + RESOLVED_TYPE_ERROR,0,NULL,-1,-1); + } + return; + } + + if (conn->entry_cfg.prefer_ipv6) { + addr_best = addr_ipv6 ? addr_ipv6 : addr_ipv4; + } else { + addr_best = addr_ipv4 ? addr_ipv4 : addr_ipv6; + } + + /* Now convert it to the ugly old interface */ + if (! addr_best) { + connection_ap_handshake_socks_resolved(conn, + RESOLVED_TYPE_ERROR,0,NULL,-1,-1); + return; + } + + connection_ap_handshake_socks_resolved_addr(conn, + &addr_best->addr, + addr_best->ttl, + -1); + + remap_event_helper(conn, &addr_best->addr); +} + +/** Handle a RELAY_COMMAND_RESOLVED cell that we received on a non-open AP + * stream. */ +STATIC int +connection_edge_process_resolved_cell(edge_connection_t *conn, + const cell_t *cell, + const relay_header_t *rh) +{ + entry_connection_t *entry_conn = EDGE_TO_ENTRY_CONN(conn); + smartlist_t *resolved_addresses = NULL; + int errcode = 0; + + if (conn->base_.state != AP_CONN_STATE_RESOLVE_WAIT) { + log_fn(LOG_PROTOCOL_WARN, LD_APP, "Got a 'resolved' cell while " + "not in state resolve_wait. Dropping."); + return 0; + } + tor_assert(SOCKS_COMMAND_IS_RESOLVE(entry_conn->socks_request->command)); + + resolved_addresses = smartlist_new(); + if (resolved_cell_parse(cell, rh, resolved_addresses, &errcode)) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Dropping malformed 'resolved' cell"); + connection_mark_unattached_ap(entry_conn, END_STREAM_REASON_TORPROTOCOL); + goto done; + } + + if (get_options()->ClientDNSRejectInternalAddresses) { + int orig_len = smartlist_len(resolved_addresses); + SMARTLIST_FOREACH_BEGIN(resolved_addresses, address_ttl_t *, addr) { + if (addr->hostname == NULL && tor_addr_is_internal(&addr->addr, 0)) { + log_info(LD_APP, "Got a resolved cell with answer %s; dropping that " + "answer.", + safe_str_client(fmt_addr(&addr->addr))); + address_ttl_free(addr); + SMARTLIST_DEL_CURRENT(resolved_addresses, addr); + } + } SMARTLIST_FOREACH_END(addr); + if (orig_len && smartlist_len(resolved_addresses) == 0) { + log_info(LD_APP, "Got a resolved cell with only private addresses; " + "dropping it."); + connection_ap_handshake_socks_resolved(entry_conn, + RESOLVED_TYPE_ERROR_TRANSIENT, + 0, NULL, 0, TIME_MAX); + connection_mark_unattached_ap(entry_conn, + END_STREAM_REASON_TORPROTOCOL); + goto done; + } + } + + /* This is valid data at this point. Count it */ + if (conn->on_circuit && CIRCUIT_IS_ORIGIN(conn->on_circuit)) { + circuit_read_valid_data(TO_ORIGIN_CIRCUIT(conn->on_circuit), + rh->length); + } + + connection_ap_handshake_socks_got_resolved_cell(entry_conn, + errcode, + resolved_addresses); + + connection_mark_unattached_ap(entry_conn, + END_STREAM_REASON_DONE | + END_STREAM_REASON_FLAG_ALREADY_SOCKS_REPLIED); + + done: + SMARTLIST_FOREACH(resolved_addresses, address_ttl_t *, addr, + address_ttl_free(addr)); + smartlist_free(resolved_addresses); + return 0; +} + +/** An incoming relay cell has arrived from circuit <b>circ</b> to + * stream <b>conn</b>. + * + * The arguments here are the same as in + * connection_edge_process_relay_cell() below; this function is called + * from there when <b>conn</b> is defined and not in an open state. + */ +static int +connection_edge_process_relay_cell_not_open( + relay_header_t *rh, cell_t *cell, circuit_t *circ, + edge_connection_t *conn, crypt_path_t *layer_hint) +{ + if (rh->command == RELAY_COMMAND_END) { + if (CIRCUIT_IS_ORIGIN(circ) && conn->base_.type == CONN_TYPE_AP) { + return connection_ap_process_end_not_open(rh, cell, + TO_ORIGIN_CIRCUIT(circ), + EDGE_TO_ENTRY_CONN(conn), + layer_hint); + } else { + /* we just got an 'end', don't need to send one */ + conn->edge_has_sent_end = 1; + conn->end_reason = *(cell->payload+RELAY_HEADER_SIZE) | + END_STREAM_REASON_FLAG_REMOTE; + connection_mark_for_close(TO_CONN(conn)); + return 0; + } + } + + if (conn->base_.type == CONN_TYPE_AP && + rh->command == RELAY_COMMAND_CONNECTED) { + tor_addr_t addr; + int ttl; + entry_connection_t *entry_conn = EDGE_TO_ENTRY_CONN(conn); + tor_assert(CIRCUIT_IS_ORIGIN(circ)); + if (conn->base_.state != AP_CONN_STATE_CONNECT_WAIT) { + log_fn(LOG_PROTOCOL_WARN, LD_APP, + "Got 'connected' while not in state connect_wait. Dropping."); + return 0; + } + CONNECTION_AP_EXPECT_NONPENDING(entry_conn); + conn->base_.state = AP_CONN_STATE_OPEN; + log_info(LD_APP,"'connected' received for circid %u streamid %d " + "after %d seconds.", + (unsigned)circ->n_circ_id, + rh->stream_id, + (int)(time(NULL) - conn->base_.timestamp_last_read_allowed)); + if (connected_cell_parse(rh, cell, &addr, &ttl) < 0) { + log_fn(LOG_PROTOCOL_WARN, LD_APP, + "Got a badly formatted connected cell. Closing."); + connection_edge_end(conn, END_STREAM_REASON_TORPROTOCOL); + connection_mark_unattached_ap(entry_conn, END_STREAM_REASON_TORPROTOCOL); + return 0; + } + if (tor_addr_family(&addr) != AF_UNSPEC) { + /* The family is not UNSPEC: so we were given an address in the + * connected cell. (This is normal, except for BEGINDIR and onion + * service streams.) */ + const sa_family_t family = tor_addr_family(&addr); + if (tor_addr_is_null(&addr) || + (get_options()->ClientDNSRejectInternalAddresses && + tor_addr_is_internal(&addr, 0))) { + log_info(LD_APP, "...but it claims the IP address was %s. Closing.", + fmt_addr(&addr)); + connection_edge_end(conn, END_STREAM_REASON_TORPROTOCOL); + connection_mark_unattached_ap(entry_conn, + END_STREAM_REASON_TORPROTOCOL); + return 0; + } + + if ((family == AF_INET && ! entry_conn->entry_cfg.ipv4_traffic) || + (family == AF_INET6 && ! entry_conn->entry_cfg.ipv6_traffic)) { + log_fn(LOG_PROTOCOL_WARN, LD_APP, + "Got a connected cell to %s with unsupported address family." + " Closing.", fmt_addr(&addr)); + connection_edge_end(conn, END_STREAM_REASON_TORPROTOCOL); + connection_mark_unattached_ap(entry_conn, + END_STREAM_REASON_TORPROTOCOL); + return 0; + } + + client_dns_set_addressmap(entry_conn, + entry_conn->socks_request->address, &addr, + entry_conn->chosen_exit_name, ttl); + + remap_event_helper(entry_conn, &addr); + } + circuit_log_path(LOG_INFO,LD_APP,TO_ORIGIN_CIRCUIT(circ)); + /* don't send a socks reply to transparent conns */ + tor_assert(entry_conn->socks_request != NULL); + if (!entry_conn->socks_request->has_finished) { + connection_ap_handshake_socks_reply(entry_conn, NULL, 0, 0); + } + + /* Was it a linked dir conn? If so, a dir request just started to + * fetch something; this could be a bootstrap status milestone. */ + log_debug(LD_APP, "considering"); + if (TO_CONN(conn)->linked_conn && + TO_CONN(conn)->linked_conn->type == CONN_TYPE_DIR) { + connection_t *dirconn = TO_CONN(conn)->linked_conn; + log_debug(LD_APP, "it is! %d", dirconn->purpose); + switch (dirconn->purpose) { + case DIR_PURPOSE_FETCH_CERTIFICATE: + if (consensus_is_waiting_for_certs()) + control_event_bootstrap(BOOTSTRAP_STATUS_LOADING_KEYS, 0); + break; + case DIR_PURPOSE_FETCH_CONSENSUS: + control_event_bootstrap(BOOTSTRAP_STATUS_LOADING_STATUS, 0); + break; + case DIR_PURPOSE_FETCH_SERVERDESC: + case DIR_PURPOSE_FETCH_MICRODESC: + if (TO_DIR_CONN(dirconn)->router_purpose == ROUTER_PURPOSE_GENERAL) + control_event_bootstrap(BOOTSTRAP_STATUS_LOADING_DESCRIPTORS, + count_loading_descriptors_progress()); + break; + } + } + /* This is definitely a success, so forget about any pending data we + * had sent. */ + if (entry_conn->pending_optimistic_data) { + buf_free(entry_conn->pending_optimistic_data); + entry_conn->pending_optimistic_data = NULL; + } + + /* This is valid data at this point. Count it */ + circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), rh->length); + + /* handle anything that might have queued */ + if (connection_edge_package_raw_inbuf(conn, 1, NULL) < 0) { + /* (We already sent an end cell if possible) */ + connection_mark_for_close(TO_CONN(conn)); + return 0; + } + return 0; + } + if (conn->base_.type == CONN_TYPE_AP && + rh->command == RELAY_COMMAND_RESOLVED) { + return connection_edge_process_resolved_cell(conn, cell, rh); + } + + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Got an unexpected relay command %d, in state %d (%s). Dropping.", + rh->command, conn->base_.state, + conn_state_to_string(conn->base_.type, conn->base_.state)); + return 0; /* for forward compatibility, don't kill the circuit */ +// connection_edge_end(conn, END_STREAM_REASON_TORPROTOCOL); +// connection_mark_for_close(conn); +// return -1; +} + +/** An incoming relay cell has arrived on circuit <b>circ</b>. If + * <b>conn</b> is NULL this is a control cell, else <b>cell</b> is + * destined for <b>conn</b>. + * + * If <b>layer_hint</b> is defined, then we're the origin of the + * circuit, and it specifies the hop that packaged <b>cell</b>. + * + * Return -reason if you want to warn and tear down the circuit, else 0. + */ +STATIC int +connection_edge_process_relay_cell(cell_t *cell, circuit_t *circ, + edge_connection_t *conn, + crypt_path_t *layer_hint) +{ + static int num_seen=0; + relay_header_t rh; + unsigned domain = layer_hint?LD_APP:LD_EXIT; + int reason; + int optimistic_data = 0; /* Set to 1 if we receive data on a stream + * that's in the EXIT_CONN_STATE_RESOLVING + * or EXIT_CONN_STATE_CONNECTING states. */ + + tor_assert(cell); + tor_assert(circ); + + relay_header_unpack(&rh, cell->payload); +// log_fn(LOG_DEBUG,"command %d stream %d", rh.command, rh.stream_id); + num_seen++; + log_debug(domain, "Now seen %d relay cells here (command %d, stream %d).", + num_seen, rh.command, rh.stream_id); + + if (rh.length > RELAY_PAYLOAD_SIZE) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Relay cell length field too long. Closing circuit."); + return - END_CIRC_REASON_TORPROTOCOL; + } + + if (rh.stream_id == 0) { + switch (rh.command) { + case RELAY_COMMAND_BEGIN: + case RELAY_COMMAND_CONNECTED: + case RELAY_COMMAND_END: + case RELAY_COMMAND_RESOLVE: + case RELAY_COMMAND_RESOLVED: + case RELAY_COMMAND_BEGIN_DIR: + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, "Relay command %d with zero " + "stream_id. Dropping.", (int)rh.command); + return 0; + default: + ; + } + } + + /* either conn is NULL, in which case we've got a control cell, or else + * conn points to the recognized stream. */ + + if (conn && !connection_state_is_open(TO_CONN(conn))) { + if (conn->base_.type == CONN_TYPE_EXIT && + (conn->base_.state == EXIT_CONN_STATE_CONNECTING || + conn->base_.state == EXIT_CONN_STATE_RESOLVING) && + rh.command == RELAY_COMMAND_DATA) { + /* Allow DATA cells to be delivered to an exit node in state + * EXIT_CONN_STATE_CONNECTING or EXIT_CONN_STATE_RESOLVING. + * This speeds up HTTP, for example. */ + optimistic_data = 1; + } else if (rh.stream_id == 0 && rh.command == RELAY_COMMAND_DATA) { + log_warn(LD_BUG, "Somehow I had a connection that matched a " + "data cell with stream ID 0."); + } else { + return connection_edge_process_relay_cell_not_open( + &rh, cell, circ, conn, layer_hint); + } + } + + switch (rh.command) { + case RELAY_COMMAND_DROP: + rep_hist_padding_count_read(PADDING_TYPE_DROP); +// log_info(domain,"Got a relay-level padding cell. Dropping."); + return 0; + case RELAY_COMMAND_BEGIN: + case RELAY_COMMAND_BEGIN_DIR: + if (layer_hint && + circ->purpose != CIRCUIT_PURPOSE_S_REND_JOINED) { + log_fn(LOG_PROTOCOL_WARN, LD_APP, + "Relay begin request unsupported at AP. Dropping."); + return 0; + } + if (circ->purpose == CIRCUIT_PURPOSE_S_REND_JOINED && + layer_hint != TO_ORIGIN_CIRCUIT(circ)->cpath->prev) { + log_fn(LOG_PROTOCOL_WARN, LD_APP, + "Relay begin request to Hidden Service " + "from intermediary node. Dropping."); + return 0; + } + if (conn) { + log_fn(LOG_PROTOCOL_WARN, domain, + "Begin cell for known stream. Dropping."); + return 0; + } + if (rh.command == RELAY_COMMAND_BEGIN_DIR && + circ->purpose != CIRCUIT_PURPOSE_S_REND_JOINED) { + /* Assign this circuit and its app-ward OR connection a unique ID, + * so that we can measure download times. The local edge and dir + * connection will be assigned the same ID when they are created + * and linked. */ + static uint64_t next_id = 0; + circ->dirreq_id = ++next_id; + TO_OR_CIRCUIT(circ)->p_chan->dirreq_id = circ->dirreq_id; + } + return connection_exit_begin_conn(cell, circ); + case RELAY_COMMAND_DATA: + ++stats_n_data_cells_received; + if (( layer_hint && --layer_hint->deliver_window < 0) || + (!layer_hint && --circ->deliver_window < 0)) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "(relay data) circ deliver_window below 0. Killing."); + if (conn) { + /* XXXX Do we actually need to do this? Will killing the circuit + * not send an END and mark the stream for close as appropriate? */ + connection_edge_end(conn, END_STREAM_REASON_TORPROTOCOL); + connection_mark_for_close(TO_CONN(conn)); + } + return -END_CIRC_REASON_TORPROTOCOL; + } + log_debug(domain,"circ deliver_window now %d.", layer_hint ? + layer_hint->deliver_window : circ->deliver_window); + + circuit_consider_sending_sendme(circ, layer_hint); + + if (rh.stream_id == 0) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, "Relay data cell with zero " + "stream_id. Dropping."); + return 0; + } else if (!conn) { ++ if (CIRCUIT_IS_ORIGIN(circ)) { ++ origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); ++ if (connection_half_edge_is_valid_data(ocirc->half_streams, ++ rh.stream_id)) { ++ circuit_read_valid_data(ocirc, rh.length); ++ log_info(domain, ++ "data cell on circ %u valid on half-closed " ++ "stream id %d", ocirc->global_identifier, rh.stream_id); ++ } ++ } ++ + log_info(domain,"data cell dropped, unknown stream (streamid %d).", + rh.stream_id); + return 0; + } + + if (--conn->deliver_window < 0) { /* is it below 0 after decrement? */ + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "(relay data) conn deliver_window below 0. Killing."); + return -END_CIRC_REASON_TORPROTOCOL; + } + /* Total all valid application bytes delivered */ + if (CIRCUIT_IS_ORIGIN(circ) && rh.length > 0) { + circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), rh.length); + } + + stats_n_data_bytes_received += rh.length; + connection_buf_add((char*)(cell->payload + RELAY_HEADER_SIZE), + rh.length, TO_CONN(conn)); + +#ifdef MEASUREMENTS_21206 + /* Count number of RELAY_DATA cells received on a linked directory + * connection. */ + connection_t *linked_conn = TO_CONN(conn)->linked_conn; + + if (linked_conn && linked_conn->type == CONN_TYPE_DIR) { + ++(TO_DIR_CONN(linked_conn)->data_cells_received); + } +#endif /* defined(MEASUREMENTS_21206) */ + + if (!optimistic_data) { + /* Only send a SENDME if we're not getting optimistic data; otherwise + * a SENDME could arrive before the CONNECTED. + */ + connection_edge_consider_sending_sendme(conn); + } + + return 0; + case RELAY_COMMAND_END: + reason = rh.length > 0 ? + get_uint8(cell->payload+RELAY_HEADER_SIZE) : END_STREAM_REASON_MISC; + if (!conn) { ++ if (CIRCUIT_IS_ORIGIN(circ)) { ++ origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); ++ if (connection_half_edge_is_valid_end(ocirc->half_streams, ++ rh.stream_id)) { ++ ++ circuit_read_valid_data(ocirc, rh.length); ++ log_info(domain, ++ "end cell (%s) on circ %u valid on half-closed " ++ "stream id %d", ++ stream_end_reason_to_string(reason), ++ ocirc->global_identifier, rh.stream_id); ++ return 0; ++ } ++ } + log_info(domain,"end cell (%s) dropped, unknown stream.", + stream_end_reason_to_string(reason)); + return 0; + } +/* XXX add to this log_fn the exit node's nickname? */ + log_info(domain,TOR_SOCKET_T_FORMAT": end cell (%s) for stream %d. " + "Removing stream.", + conn->base_.s, + stream_end_reason_to_string(reason), + conn->stream_id); + if (conn->base_.type == CONN_TYPE_AP) { + entry_connection_t *entry_conn = EDGE_TO_ENTRY_CONN(conn); + if (entry_conn->socks_request && + !entry_conn->socks_request->has_finished) + log_warn(LD_BUG, + "open stream hasn't sent socks answer yet? Closing."); + } + /* We just *got* an end; no reason to send one. */ + conn->edge_has_sent_end = 1; + if (!conn->end_reason) + conn->end_reason = reason | END_STREAM_REASON_FLAG_REMOTE; + if (!conn->base_.marked_for_close) { + /* only mark it if not already marked. it's possible to + * get the 'end' right around when the client hangs up on us. */ + connection_mark_and_flush(TO_CONN(conn)); + + /* Total all valid application bytes delivered */ + if (CIRCUIT_IS_ORIGIN(circ)) { + circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), rh.length); + } + } + return 0; + case RELAY_COMMAND_EXTEND: + case RELAY_COMMAND_EXTEND2: { + static uint64_t total_n_extend=0, total_nonearly=0; + total_n_extend++; + if (rh.stream_id) { + log_fn(LOG_PROTOCOL_WARN, domain, + "'extend' cell received for non-zero stream. Dropping."); + return 0; + } + if (cell->command != CELL_RELAY_EARLY && + !networkstatus_get_param(NULL,"AllowNonearlyExtend",0,0,1)) { +#define EARLY_WARNING_INTERVAL 3600 + static ratelim_t early_warning_limit = + RATELIM_INIT(EARLY_WARNING_INTERVAL); + char *m; + if (cell->command == CELL_RELAY) { + ++total_nonearly; + if ((m = rate_limit_log(&early_warning_limit, approx_time()))) { + double percentage = ((double)total_nonearly)/total_n_extend; + percentage *= 100; + log_fn(LOG_PROTOCOL_WARN, domain, "EXTEND cell received, " + "but not via RELAY_EARLY. Dropping.%s", m); + log_fn(LOG_PROTOCOL_WARN, domain, " (We have dropped %.02f%% of " + "all EXTEND cells for this reason)", percentage); + tor_free(m); + } + } else { + log_fn(LOG_WARN, domain, + "EXTEND cell received, in a cell with type %d! Dropping.", + cell->command); + } + return 0; + } + return circuit_extend(cell, circ); + } + case RELAY_COMMAND_EXTENDED: + case RELAY_COMMAND_EXTENDED2: + if (!layer_hint) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "'extended' unsupported at non-origin. Dropping."); + return 0; + } + log_debug(domain,"Got an extended cell! Yay."); + { + extended_cell_t extended_cell; + if (extended_cell_parse(&extended_cell, rh.command, + (const uint8_t*)cell->payload+RELAY_HEADER_SIZE, + rh.length)<0) { + log_warn(LD_PROTOCOL, + "Can't parse EXTENDED cell; killing circuit."); + return -END_CIRC_REASON_TORPROTOCOL; + } + if ((reason = circuit_finish_handshake(TO_ORIGIN_CIRCUIT(circ), + &extended_cell.created_cell)) < 0) { + circuit_mark_for_close(circ, -reason); + return 0; /* We don't want to cause a warning, so we mark the circuit + * here. */ + } + } + if ((reason=circuit_send_next_onion_skin(TO_ORIGIN_CIRCUIT(circ)))<0) { + log_info(domain,"circuit_send_next_onion_skin() failed."); + return reason; + } + /* Total all valid bytes delivered. */ + if (CIRCUIT_IS_ORIGIN(circ)) { + circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), rh.length); + } + return 0; + case RELAY_COMMAND_TRUNCATE: + if (layer_hint) { + log_fn(LOG_PROTOCOL_WARN, LD_APP, + "'truncate' unsupported at origin. Dropping."); + return 0; + } + if (circ->n_hop) { + if (circ->n_chan) + log_warn(LD_BUG, "n_chan and n_hop set on the same circuit!"); + extend_info_free(circ->n_hop); + circ->n_hop = NULL; + tor_free(circ->n_chan_create_cell); + circuit_set_state(circ, CIRCUIT_STATE_OPEN); + } + if (circ->n_chan) { + uint8_t trunc_reason = get_uint8(cell->payload + RELAY_HEADER_SIZE); + circuit_clear_cell_queue(circ, circ->n_chan); + channel_send_destroy(circ->n_circ_id, circ->n_chan, + trunc_reason); + circuit_set_n_circid_chan(circ, 0, NULL); + } + log_debug(LD_EXIT, "Processed 'truncate', replying."); + { + char payload[1]; + payload[0] = (char)END_CIRC_REASON_REQUESTED; + relay_send_command_from_edge(0, circ, RELAY_COMMAND_TRUNCATED, + payload, sizeof(payload), NULL); + } + return 0; + case RELAY_COMMAND_TRUNCATED: + if (!layer_hint) { + log_fn(LOG_PROTOCOL_WARN, LD_EXIT, + "'truncated' unsupported at non-origin. Dropping."); + return 0; + } - circuit_truncated(TO_ORIGIN_CIRCUIT(circ), layer_hint, ++ ++ /* Count the truncated as valid, for completeness. The ++ * circuit is being torn down anyway, though. */ ++ if (CIRCUIT_IS_ORIGIN(circ)) { ++ circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), ++ rh.length); ++ } ++ circuit_truncated(TO_ORIGIN_CIRCUIT(circ), + get_uint8(cell->payload + RELAY_HEADER_SIZE)); + return 0; + case RELAY_COMMAND_CONNECTED: + if (conn) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "'connected' unsupported while open. Closing circ."); + return -END_CIRC_REASON_TORPROTOCOL; + } ++ ++ if (CIRCUIT_IS_ORIGIN(circ)) { ++ origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); ++ if (connection_half_edge_is_valid_connected(ocirc->half_streams, ++ rh.stream_id)) { ++ circuit_read_valid_data(ocirc, rh.length); ++ log_info(domain, ++ "connected cell on circ %u valid on half-closed " ++ "stream id %d", ocirc->global_identifier, rh.stream_id); ++ return 0; ++ } ++ } ++ + log_info(domain, + "'connected' received on circid %u for streamid %d, " + "no conn attached anymore. Ignoring.", + (unsigned)circ->n_circ_id, rh.stream_id); + return 0; + case RELAY_COMMAND_SENDME: + if (!rh.stream_id) { + if (layer_hint) { + if (layer_hint->package_window + CIRCWINDOW_INCREMENT > + CIRCWINDOW_START_MAX) { + static struct ratelim_t exit_warn_ratelim = RATELIM_INIT(600); + log_fn_ratelim(&exit_warn_ratelim, LOG_WARN, LD_PROTOCOL, + "Unexpected sendme cell from exit relay. " + "Closing circ."); + return -END_CIRC_REASON_TORPROTOCOL; + } + layer_hint->package_window += CIRCWINDOW_INCREMENT; + log_debug(LD_APP,"circ-level sendme at origin, packagewindow %d.", + layer_hint->package_window); + circuit_resume_edge_reading(circ, layer_hint); + + /* We count circuit-level sendme's as valid delivered data because + * they are rate limited. + */ + if (CIRCUIT_IS_ORIGIN(circ)) { + circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), + rh.length); + } + + } else { + if (circ->package_window + CIRCWINDOW_INCREMENT > + CIRCWINDOW_START_MAX) { + static struct ratelim_t client_warn_ratelim = RATELIM_INIT(600); + log_fn_ratelim(&client_warn_ratelim,LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Unexpected sendme cell from client. " + "Closing circ (window %d).", + circ->package_window); + return -END_CIRC_REASON_TORPROTOCOL; + } + circ->package_window += CIRCWINDOW_INCREMENT; + log_debug(LD_APP, + "circ-level sendme at non-origin, packagewindow %d.", + circ->package_window); + circuit_resume_edge_reading(circ, layer_hint); + } + return 0; + } + if (!conn) { ++ if (CIRCUIT_IS_ORIGIN(circ)) { ++ origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); ++ if (connection_half_edge_is_valid_sendme(ocirc->half_streams, ++ rh.stream_id)) { ++ circuit_read_valid_data(ocirc, rh.length); ++ log_info(domain, ++ "sendme cell on circ %u valid on half-closed " ++ "stream id %d", ocirc->global_identifier, rh.stream_id); ++ } ++ } ++ + log_info(domain,"sendme cell dropped, unknown stream (streamid %d).", + rh.stream_id); + return 0; + } + + /* Don't allow the other endpoint to request more than our maximum + * (i.e. initial) stream SENDME window worth of data. Well-behaved + * stock clients will not request more than this max (as per the check + * in the while loop of connection_edge_consider_sending_sendme()). + */ + if (conn->package_window + STREAMWINDOW_INCREMENT > + STREAMWINDOW_START_MAX) { + static struct ratelim_t stream_warn_ratelim = RATELIM_INIT(600); + log_fn_ratelim(&stream_warn_ratelim, LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Unexpected stream sendme cell. Closing circ (window %d).", + conn->package_window); + return -END_CIRC_REASON_TORPROTOCOL; + } + + /* At this point, the stream sendme is valid */ + if (CIRCUIT_IS_ORIGIN(circ)) { + circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), + rh.length); + } + + conn->package_window += STREAMWINDOW_INCREMENT; + log_debug(domain,"stream-level sendme, packagewindow now %d.", + conn->package_window); + if (circuit_queue_streams_are_blocked(circ)) { + /* Still waiting for queue to flush; don't touch conn */ + return 0; + } + connection_start_reading(TO_CONN(conn)); + /* handle whatever might still be on the inbuf */ + if (connection_edge_package_raw_inbuf(conn, 1, NULL) < 0) { + /* (We already sent an end cell if possible) */ + connection_mark_for_close(TO_CONN(conn)); + return 0; + } + return 0; + case RELAY_COMMAND_RESOLVE: + if (layer_hint) { + log_fn(LOG_PROTOCOL_WARN, LD_APP, + "resolve request unsupported at AP; dropping."); + return 0; + } else if (conn) { + log_fn(LOG_PROTOCOL_WARN, domain, + "resolve request for known stream; dropping."); + return 0; + } else if (circ->purpose != CIRCUIT_PURPOSE_OR) { + log_fn(LOG_PROTOCOL_WARN, domain, + "resolve request on circ with purpose %d; dropping", + circ->purpose); + return 0; + } + connection_exit_begin_resolve(cell, TO_OR_CIRCUIT(circ)); + return 0; + case RELAY_COMMAND_RESOLVED: + if (conn) { + log_fn(LOG_PROTOCOL_WARN, domain, + "'resolved' unsupported while open. Closing circ."); + return -END_CIRC_REASON_TORPROTOCOL; + } ++ ++ if (CIRCUIT_IS_ORIGIN(circ)) { ++ origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); ++ if (connection_half_edge_is_valid_resolved(ocirc->half_streams, ++ rh.stream_id)) { ++ circuit_read_valid_data(ocirc, rh.length); ++ log_info(domain, ++ "resolved cell on circ %u valid on half-closed " ++ "stream id %d", ocirc->global_identifier, rh.stream_id); ++ return 0; ++ } ++ } ++ + log_info(domain, + "'resolved' received, no conn attached anymore. Ignoring."); + return 0; + case RELAY_COMMAND_ESTABLISH_INTRO: + case RELAY_COMMAND_ESTABLISH_RENDEZVOUS: + case RELAY_COMMAND_INTRODUCE1: + case RELAY_COMMAND_INTRODUCE2: + case RELAY_COMMAND_INTRODUCE_ACK: + case RELAY_COMMAND_RENDEZVOUS1: + case RELAY_COMMAND_RENDEZVOUS2: + case RELAY_COMMAND_INTRO_ESTABLISHED: + case RELAY_COMMAND_RENDEZVOUS_ESTABLISHED: + rend_process_relay_cell(circ, layer_hint, + rh.command, rh.length, + cell->payload+RELAY_HEADER_SIZE); + return 0; + } + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "Received unknown relay command %d. Perhaps the other side is using " + "a newer version of Tor? Dropping.", + rh.command); + return 0; /* for forward compatibility, don't kill the circuit */ +} + +/** How many relay_data cells have we built, ever? */ +uint64_t stats_n_data_cells_packaged = 0; +/** How many bytes of data have we put in relay_data cells have we built, + * ever? This would be RELAY_PAYLOAD_SIZE*stats_n_data_cells_packaged if + * every relay cell we ever sent were completely full of data. */ +uint64_t stats_n_data_bytes_packaged = 0; +/** How many relay_data cells have we received, ever? */ +uint64_t stats_n_data_cells_received = 0; +/** How many bytes of data have we received relay_data cells, ever? This would + * be RELAY_PAYLOAD_SIZE*stats_n_data_cells_packaged if every relay cell we + * ever received were completely full of data. */ +uint64_t stats_n_data_bytes_received = 0; + +/** If <b>conn</b> has an entire relay payload of bytes on its inbuf (or + * <b>package_partial</b> is true), and the appropriate package windows aren't + * empty, grab a cell and send it down the circuit. + * + * If *<b>max_cells</b> is given, package no more than max_cells. Decrement + * *<b>max_cells</b> by the number of cells packaged. + * + * Return -1 (and send a RELAY_COMMAND_END cell if necessary) if conn should + * be marked for close, else return 0. + */ +int +connection_edge_package_raw_inbuf(edge_connection_t *conn, int package_partial, + int *max_cells) +{ + size_t bytes_to_process, length; + char payload[CELL_PAYLOAD_SIZE]; + circuit_t *circ; + const unsigned domain = conn->base_.type == CONN_TYPE_AP ? LD_APP : LD_EXIT; + int sending_from_optimistic = 0; + entry_connection_t *entry_conn = + conn->base_.type == CONN_TYPE_AP ? EDGE_TO_ENTRY_CONN(conn) : NULL; + const int sending_optimistically = + entry_conn && + conn->base_.type == CONN_TYPE_AP && + conn->base_.state != AP_CONN_STATE_OPEN; + crypt_path_t *cpath_layer = conn->cpath_layer; + + tor_assert(conn); + + if (conn->base_.marked_for_close) { + log_warn(LD_BUG, + "called on conn that's already marked for close at %s:%d.", + conn->base_.marked_for_close_file, conn->base_.marked_for_close); + return 0; + } + + if (max_cells && *max_cells <= 0) + return 0; + + repeat_connection_edge_package_raw_inbuf: + + circ = circuit_get_by_edge_conn(conn); + if (!circ) { + log_info(domain,"conn has no circuit! Closing."); + conn->end_reason = END_STREAM_REASON_CANT_ATTACH; + return -1; + } + + if (circuit_consider_stop_edge_reading(circ, cpath_layer)) + return 0; + + if (conn->package_window <= 0) { + log_info(domain,"called with package_window %d. Skipping.", + conn->package_window); + connection_stop_reading(TO_CONN(conn)); + return 0; + } + + sending_from_optimistic = entry_conn && + entry_conn->sending_optimistic_data != NULL; + + if (PREDICT_UNLIKELY(sending_from_optimistic)) { + bytes_to_process = buf_datalen(entry_conn->sending_optimistic_data); + if (PREDICT_UNLIKELY(!bytes_to_process)) { + log_warn(LD_BUG, "sending_optimistic_data was non-NULL but empty"); + bytes_to_process = connection_get_inbuf_len(TO_CONN(conn)); + sending_from_optimistic = 0; + } + } else { + bytes_to_process = connection_get_inbuf_len(TO_CONN(conn)); + } + + if (!bytes_to_process) + return 0; + + if (!package_partial && bytes_to_process < RELAY_PAYLOAD_SIZE) + return 0; + + if (bytes_to_process > RELAY_PAYLOAD_SIZE) { + length = RELAY_PAYLOAD_SIZE; + } else { + length = bytes_to_process; + } + stats_n_data_bytes_packaged += length; + stats_n_data_cells_packaged += 1; + + if (PREDICT_UNLIKELY(sending_from_optimistic)) { + /* XXXX We could be more efficient here by sometimes packing + * previously-sent optimistic data in the same cell with data + * from the inbuf. */ + buf_get_bytes(entry_conn->sending_optimistic_data, payload, length); + if (!buf_datalen(entry_conn->sending_optimistic_data)) { + buf_free(entry_conn->sending_optimistic_data); + entry_conn->sending_optimistic_data = NULL; + } + } else { + connection_buf_get_bytes(payload, length, TO_CONN(conn)); + } + + log_debug(domain,TOR_SOCKET_T_FORMAT": Packaging %d bytes (%d waiting).", + conn->base_.s, + (int)length, (int)connection_get_inbuf_len(TO_CONN(conn))); + + if (sending_optimistically && !sending_from_optimistic) { + /* This is new optimistic data; remember it in case we need to detach and + retry */ + if (!entry_conn->pending_optimistic_data) + entry_conn->pending_optimistic_data = buf_new(); + buf_add(entry_conn->pending_optimistic_data, payload, length); + } + + if (connection_edge_send_command(conn, RELAY_COMMAND_DATA, + payload, length) < 0 ) { + /* circuit got marked for close, don't continue, don't need to mark conn */ + return 0; + } + + if (!cpath_layer) { /* non-rendezvous exit */ + tor_assert(circ->package_window > 0); + circ->package_window--; + } else { /* we're an AP, or an exit on a rendezvous circ */ + tor_assert(cpath_layer->package_window > 0); + cpath_layer->package_window--; + } + + if (--conn->package_window <= 0) { /* is it 0 after decrement? */ + connection_stop_reading(TO_CONN(conn)); + log_debug(domain,"conn->package_window reached 0."); + circuit_consider_stop_edge_reading(circ, cpath_layer); + return 0; /* don't process the inbuf any more */ + } + log_debug(domain,"conn->package_window is now %d",conn->package_window); + + if (max_cells) { + *max_cells -= 1; + if (*max_cells <= 0) + return 0; + } + + /* handle more if there's more, or return 0 if there isn't */ + goto repeat_connection_edge_package_raw_inbuf; +} + +/** Called when we've just received a relay data cell, when + * we've just finished flushing all bytes to stream <b>conn</b>, + * or when we've flushed *some* bytes to the stream <b>conn</b>. + * + * If conn->outbuf is not too full, and our deliver window is + * low, send back a suitable number of stream-level sendme cells. + */ +void +connection_edge_consider_sending_sendme(edge_connection_t *conn) +{ + circuit_t *circ; + + if (connection_outbuf_too_full(TO_CONN(conn))) + return; + + circ = circuit_get_by_edge_conn(conn); + if (!circ) { + /* this can legitimately happen if the destroy has already + * arrived and torn down the circuit */ + log_info(LD_APP,"No circuit associated with conn. Skipping."); + return; + } + + while (conn->deliver_window <= STREAMWINDOW_START - STREAMWINDOW_INCREMENT) { + log_debug(conn->base_.type == CONN_TYPE_AP ?LD_APP:LD_EXIT, + "Outbuf %d, Queuing stream sendme.", + (int)conn->base_.outbuf_flushlen); + conn->deliver_window += STREAMWINDOW_INCREMENT; + if (connection_edge_send_command(conn, RELAY_COMMAND_SENDME, + NULL, 0) < 0) { + log_warn(LD_APP,"connection_edge_send_command failed. Skipping."); + return; /* the circuit's closed, don't continue */ + } + } +} + +/** The circuit <b>circ</b> has received a circuit-level sendme + * (on hop <b>layer_hint</b>, if we're the OP). Go through all the + * attached streams and let them resume reading and packaging, if + * their stream windows allow it. + */ +static void +circuit_resume_edge_reading(circuit_t *circ, crypt_path_t *layer_hint) +{ + if (circuit_queue_streams_are_blocked(circ)) { + log_debug(layer_hint?LD_APP:LD_EXIT,"Too big queue, no resuming"); + return; + } + log_debug(layer_hint?LD_APP:LD_EXIT,"resuming"); + + if (CIRCUIT_IS_ORIGIN(circ)) + circuit_resume_edge_reading_helper(TO_ORIGIN_CIRCUIT(circ)->p_streams, + circ, layer_hint); + else + circuit_resume_edge_reading_helper(TO_OR_CIRCUIT(circ)->n_streams, + circ, layer_hint); +} + +void +stream_choice_seed_weak_rng(void) +{ + crypto_seed_weak_rng(&stream_choice_rng); +} + +/** A helper function for circuit_resume_edge_reading() above. + * The arguments are the same, except that <b>conn</b> is the head + * of a linked list of edge streams that should each be considered. + */ +static int +circuit_resume_edge_reading_helper(edge_connection_t *first_conn, + circuit_t *circ, + crypt_path_t *layer_hint) +{ + edge_connection_t *conn; + int n_packaging_streams, n_streams_left; + int packaged_this_round; + int cells_on_queue; + int cells_per_conn; + edge_connection_t *chosen_stream = NULL; + int max_to_package; + + if (first_conn == NULL) { + /* Don't bother to try to do the rest of this if there are no connections + * to resume. */ + return 0; + } + + /* How many cells do we have space for? It will be the minimum of + * the number needed to exhaust the package window, and the minimum + * needed to fill the cell queue. */ + max_to_package = circ->package_window; + if (CIRCUIT_IS_ORIGIN(circ)) { + cells_on_queue = circ->n_chan_cells.n; + } else { + or_circuit_t *or_circ = TO_OR_CIRCUIT(circ); + cells_on_queue = or_circ->p_chan_cells.n; + } + if (CELL_QUEUE_HIGHWATER_SIZE - cells_on_queue < max_to_package) + max_to_package = CELL_QUEUE_HIGHWATER_SIZE - cells_on_queue; + + /* Once we used to start listening on the streams in the order they + * appeared in the linked list. That leads to starvation on the + * streams that appeared later on the list, since the first streams + * would always get to read first. Instead, we just pick a random + * stream on the list, and enable reading for streams starting at that + * point (and wrapping around as if the list were circular). It would + * probably be better to actually remember which streams we've + * serviced in the past, but this is simple and effective. */ + + /* Select a stream uniformly at random from the linked list. We + * don't need cryptographic randomness here. */ + { + int num_streams = 0; + for (conn = first_conn; conn; conn = conn->next_stream) { + num_streams++; + if (tor_weak_random_one_in_n(&stream_choice_rng, num_streams)) { + chosen_stream = conn; + } + /* Invariant: chosen_stream has been chosen uniformly at random from + * among the first num_streams streams on first_conn. + * + * (Note that we iterate over every stream on the circuit, so that after + * we've considered the first stream, we've chosen it with P=1; and + * after we consider the second stream, we've switched to it with P=1/2 + * and stayed with the first stream with P=1/2; and after we've + * considered the third stream, we've switched to it with P=1/3 and + * remained with one of the first two streams with P=(2/3), giving each + * one P=(1/2)(2/3) )=(1/3).) */ + } + } + + /* Count how many non-marked streams there are that have anything on + * their inbuf, and enable reading on all of the connections. */ + n_packaging_streams = 0; + /* Activate reading starting from the chosen stream */ + for (conn=chosen_stream; conn; conn = conn->next_stream) { + /* Start reading for the streams starting from here */ + if (conn->base_.marked_for_close || conn->package_window <= 0) + continue; + if (!layer_hint || conn->cpath_layer == layer_hint) { + connection_start_reading(TO_CONN(conn)); + + if (connection_get_inbuf_len(TO_CONN(conn)) > 0) + ++n_packaging_streams; + } + } + /* Go back and do the ones we skipped, circular-style */ + for (conn = first_conn; conn != chosen_stream; conn = conn->next_stream) { + if (conn->base_.marked_for_close || conn->package_window <= 0) + continue; + if (!layer_hint || conn->cpath_layer == layer_hint) { + connection_start_reading(TO_CONN(conn)); + + if (connection_get_inbuf_len(TO_CONN(conn)) > 0) + ++n_packaging_streams; + } + } + + if (n_packaging_streams == 0) /* avoid divide-by-zero */ + return 0; + + again: + + cells_per_conn = CEIL_DIV(max_to_package, n_packaging_streams); + + packaged_this_round = 0; + n_streams_left = 0; + + /* Iterate over all connections. Package up to cells_per_conn cells on + * each. Update packaged_this_round with the total number of cells + * packaged, and n_streams_left with the number that still have data to + * package. + */ + for (conn=first_conn; conn; conn=conn->next_stream) { + if (conn->base_.marked_for_close || conn->package_window <= 0) + continue; + if (!layer_hint || conn->cpath_layer == layer_hint) { + int n = cells_per_conn, r; + /* handle whatever might still be on the inbuf */ + r = connection_edge_package_raw_inbuf(conn, 1, &n); + + /* Note how many we packaged */ + packaged_this_round += (cells_per_conn-n); + + if (r<0) { + /* Problem while packaging. (We already sent an end cell if + * possible) */ + connection_mark_for_close(TO_CONN(conn)); + continue; + } + + /* If there's still data to read, we'll be coming back to this stream. */ + if (connection_get_inbuf_len(TO_CONN(conn))) + ++n_streams_left; + + /* If the circuit won't accept any more data, return without looking + * at any more of the streams. Any connections that should be stopped + * have already been stopped by connection_edge_package_raw_inbuf. */ + if (circuit_consider_stop_edge_reading(circ, layer_hint)) + return -1; + /* XXXX should we also stop immediately if we fill up the cell queue? + * Probably. */ + } + } + + /* If we made progress, and we are willing to package more, and there are + * any streams left that want to package stuff... try again! + */ + if (packaged_this_round && packaged_this_round < max_to_package && + n_streams_left) { + max_to_package -= packaged_this_round; + n_packaging_streams = n_streams_left; + goto again; + } + + return 0; +} + +/** Check if the package window for <b>circ</b> is empty (at + * hop <b>layer_hint</b> if it's defined). + * + * If yes, tell edge streams to stop reading and return 1. + * Else return 0. + */ +static int +circuit_consider_stop_edge_reading(circuit_t *circ, crypt_path_t *layer_hint) +{ + edge_connection_t *conn = NULL; + unsigned domain = layer_hint ? LD_APP : LD_EXIT; + + if (!layer_hint) { + or_circuit_t *or_circ = TO_OR_CIRCUIT(circ); + log_debug(domain,"considering circ->package_window %d", + circ->package_window); + if (circ->package_window <= 0) { + log_debug(domain,"yes, not-at-origin. stopped."); + for (conn = or_circ->n_streams; conn; conn=conn->next_stream) + connection_stop_reading(TO_CONN(conn)); + return 1; + } + return 0; + } + /* else, layer hint is defined, use it */ + log_debug(domain,"considering layer_hint->package_window %d", + layer_hint->package_window); + if (layer_hint->package_window <= 0) { + log_debug(domain,"yes, at-origin. stopped."); + for (conn = TO_ORIGIN_CIRCUIT(circ)->p_streams; conn; + conn=conn->next_stream) { + if (conn->cpath_layer == layer_hint) + connection_stop_reading(TO_CONN(conn)); + } + return 1; + } + return 0; +} + +/** Check if the deliver_window for circuit <b>circ</b> (at hop + * <b>layer_hint</b> if it's defined) is low enough that we should + * send a circuit-level sendme back down the circuit. If so, send + * enough sendmes that the window would be overfull if we sent any + * more. + */ +static void +circuit_consider_sending_sendme(circuit_t *circ, crypt_path_t *layer_hint) +{ +// log_fn(LOG_INFO,"Considering: layer_hint is %s", +// layer_hint ? "defined" : "null"); + while ((layer_hint ? layer_hint->deliver_window : circ->deliver_window) <= + CIRCWINDOW_START - CIRCWINDOW_INCREMENT) { + log_debug(LD_CIRC,"Queuing circuit sendme."); + if (layer_hint) + layer_hint->deliver_window += CIRCWINDOW_INCREMENT; + else + circ->deliver_window += CIRCWINDOW_INCREMENT; + if (relay_send_command_from_edge(0, circ, RELAY_COMMAND_SENDME, + NULL, 0, layer_hint) < 0) { + log_warn(LD_CIRC, + "relay_send_command_from_edge failed. Circuit's closed."); + return; /* the circuit's closed, don't continue */ + } + } +} + +/** The total number of cells we have allocated. */ +static size_t total_cells_allocated = 0; + +/** Release storage held by <b>cell</b>. */ +static inline void +packed_cell_free_unchecked(packed_cell_t *cell) +{ + --total_cells_allocated; + tor_free(cell); +} + +/** Allocate and return a new packed_cell_t. */ +STATIC packed_cell_t * +packed_cell_new(void) +{ + ++total_cells_allocated; + return tor_malloc_zero(sizeof(packed_cell_t)); +} + +/** Return a packed cell used outside by channel_t lower layer */ +void +packed_cell_free_(packed_cell_t *cell) +{ + if (!cell) + return; + packed_cell_free_unchecked(cell); +} + +/** Log current statistics for cell pool allocation at log level + * <b>severity</b>. */ +void +dump_cell_pool_usage(int severity) +{ + int n_circs = 0; + int n_cells = 0; + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, c) { + n_cells += c->n_chan_cells.n; + if (!CIRCUIT_IS_ORIGIN(c)) + n_cells += TO_OR_CIRCUIT(c)->p_chan_cells.n; + ++n_circs; + } + SMARTLIST_FOREACH_END(c); + tor_log(severity, LD_MM, + "%d cells allocated on %d circuits. %d cells leaked.", + n_cells, n_circs, (int)total_cells_allocated - n_cells); +} + +/** Allocate a new copy of packed <b>cell</b>. */ +static inline packed_cell_t * +packed_cell_copy(const cell_t *cell, int wide_circ_ids) +{ + packed_cell_t *c = packed_cell_new(); + cell_pack(c, cell, wide_circ_ids); + return c; +} + +/** Append <b>cell</b> to the end of <b>queue</b>. */ +void +cell_queue_append(cell_queue_t *queue, packed_cell_t *cell) +{ + TOR_SIMPLEQ_INSERT_TAIL(&queue->head, cell, next); + ++queue->n; +} + +/** Append a newly allocated copy of <b>cell</b> to the end of the + * <b>exitward</b> (or app-ward) <b>queue</b> of <b>circ</b>. If + * <b>use_stats</b> is true, record statistics about the cell. + */ +void +cell_queue_append_packed_copy(circuit_t *circ, cell_queue_t *queue, + int exitward, const cell_t *cell, + int wide_circ_ids, int use_stats) +{ + packed_cell_t *copy = packed_cell_copy(cell, wide_circ_ids); + (void)circ; + (void)exitward; + (void)use_stats; + + copy->inserted_timestamp = monotime_coarse_get_stamp(); + + cell_queue_append(queue, copy); +} + +/** Initialize <b>queue</b> as an empty cell queue. */ +void +cell_queue_init(cell_queue_t *queue) +{ + memset(queue, 0, sizeof(cell_queue_t)); + TOR_SIMPLEQ_INIT(&queue->head); +} + +/** Remove and free every cell in <b>queue</b>. */ +void +cell_queue_clear(cell_queue_t *queue) +{ + packed_cell_t *cell; + while ((cell = TOR_SIMPLEQ_FIRST(&queue->head))) { + TOR_SIMPLEQ_REMOVE_HEAD(&queue->head, next); + packed_cell_free_unchecked(cell); + } + TOR_SIMPLEQ_INIT(&queue->head); + queue->n = 0; +} + +/** Extract and return the cell at the head of <b>queue</b>; return NULL if + * <b>queue</b> is empty. */ +STATIC packed_cell_t * +cell_queue_pop(cell_queue_t *queue) +{ + packed_cell_t *cell = TOR_SIMPLEQ_FIRST(&queue->head); + if (!cell) + return NULL; + TOR_SIMPLEQ_REMOVE_HEAD(&queue->head, next); + --queue->n; + return cell; +} + +/** Initialize <b>queue</b> as an empty cell queue. */ +void +destroy_cell_queue_init(destroy_cell_queue_t *queue) +{ + memset(queue, 0, sizeof(destroy_cell_queue_t)); + TOR_SIMPLEQ_INIT(&queue->head); +} + +/** Remove and free every cell in <b>queue</b>. */ +void +destroy_cell_queue_clear(destroy_cell_queue_t *queue) +{ + destroy_cell_t *cell; + while ((cell = TOR_SIMPLEQ_FIRST(&queue->head))) { + TOR_SIMPLEQ_REMOVE_HEAD(&queue->head, next); + tor_free(cell); + } + TOR_SIMPLEQ_INIT(&queue->head); + queue->n = 0; +} + +/** Extract and return the cell at the head of <b>queue</b>; return NULL if + * <b>queue</b> is empty. */ +STATIC destroy_cell_t * +destroy_cell_queue_pop(destroy_cell_queue_t *queue) +{ + destroy_cell_t *cell = TOR_SIMPLEQ_FIRST(&queue->head); + if (!cell) + return NULL; + TOR_SIMPLEQ_REMOVE_HEAD(&queue->head, next); + --queue->n; + return cell; +} + +/** Append a destroy cell for <b>circid</b> to <b>queue</b>. */ +void +destroy_cell_queue_append(destroy_cell_queue_t *queue, + circid_t circid, + uint8_t reason) +{ + destroy_cell_t *cell = tor_malloc_zero(sizeof(destroy_cell_t)); + cell->circid = circid; + cell->reason = reason; + /* Not yet used, but will be required for OOM handling. */ + cell->inserted_timestamp = monotime_coarse_get_stamp(); + + TOR_SIMPLEQ_INSERT_TAIL(&queue->head, cell, next); + ++queue->n; +} + +/** Convert a destroy_cell_t to a newly allocated cell_t. Frees its input. */ +static packed_cell_t * +destroy_cell_to_packed_cell(destroy_cell_t *inp, int wide_circ_ids) +{ + packed_cell_t *packed = packed_cell_new(); + cell_t cell; + memset(&cell, 0, sizeof(cell)); + cell.circ_id = inp->circid; + cell.command = CELL_DESTROY; + cell.payload[0] = inp->reason; + cell_pack(packed, &cell, wide_circ_ids); + + tor_free(inp); + return packed; +} + +/** Return the total number of bytes used for each packed_cell in a queue. + * Approximate. */ +size_t +packed_cell_mem_cost(void) +{ + return sizeof(packed_cell_t); +} + +/* DOCDOC */ +size_t +cell_queues_get_total_allocation(void) +{ + return total_cells_allocated * packed_cell_mem_cost(); +} + +/** How long after we've been low on memory should we try to conserve it? */ +#define MEMORY_PRESSURE_INTERVAL (30*60) + +/** The time at which we were last low on memory. */ +static time_t last_time_under_memory_pressure = 0; + +/** Check whether we've got too much space used for cells. If so, + * call the OOM handler and return 1. Otherwise, return 0. */ +STATIC int +cell_queues_check_size(void) +{ + time_t now = time(NULL); + size_t alloc = cell_queues_get_total_allocation(); + alloc += buf_get_total_allocation(); + alloc += tor_compress_get_total_allocation(); + const size_t rend_cache_total = rend_cache_get_total_allocation(); + alloc += rend_cache_total; + const size_t geoip_client_cache_total = + geoip_client_cache_total_allocation(); + alloc += geoip_client_cache_total; + const size_t dns_cache_total = dns_cache_total_allocation(); + alloc += dns_cache_total; + if (alloc >= get_options()->MaxMemInQueues_low_threshold) { + last_time_under_memory_pressure = approx_time(); + if (alloc >= get_options()->MaxMemInQueues) { + /* If we're spending over 20% of the memory limit on hidden service + * descriptors, free them until we're down to 10%. Do the same for geoip + * client cache. */ + if (rend_cache_total > get_options()->MaxMemInQueues / 5) { + const size_t bytes_to_remove = + rend_cache_total - (size_t)(get_options()->MaxMemInQueues / 10); + alloc -= hs_cache_handle_oom(now, bytes_to_remove); + } + if (geoip_client_cache_total > get_options()->MaxMemInQueues / 5) { + const size_t bytes_to_remove = + geoip_client_cache_total - + (size_t)(get_options()->MaxMemInQueues / 10); + alloc -= geoip_client_cache_handle_oom(now, bytes_to_remove); + } + if (dns_cache_total > get_options()->MaxMemInQueues / 5) { + const size_t bytes_to_remove = + dns_cache_total - (size_t)(get_options()->MaxMemInQueues / 10); + alloc -= dns_cache_handle_oom(now, bytes_to_remove); + } + circuits_handle_oom(alloc); + return 1; + } + } + return 0; +} + +/** Return true if we've been under memory pressure in the last + * MEMORY_PRESSURE_INTERVAL seconds. */ +int +have_been_under_memory_pressure(void) +{ + return last_time_under_memory_pressure + MEMORY_PRESSURE_INTERVAL + < approx_time(); +} + +/** + * Update the number of cells available on the circuit's n_chan or p_chan's + * circuit mux. + */ +void +update_circuit_on_cmux_(circuit_t *circ, cell_direction_t direction, + const char *file, int lineno) +{ + channel_t *chan = NULL; + or_circuit_t *or_circ = NULL; + circuitmux_t *cmux = NULL; + + tor_assert(circ); + + /* Okay, get the channel */ + if (direction == CELL_DIRECTION_OUT) { + chan = circ->n_chan; + } else { + or_circ = TO_OR_CIRCUIT(circ); + chan = or_circ->p_chan; + } + + tor_assert(chan); + tor_assert(chan->cmux); + + /* Now get the cmux */ + cmux = chan->cmux; + + /* Cmux sanity check */ + if (! circuitmux_is_circuit_attached(cmux, circ)) { + log_warn(LD_BUG, "called on non-attached circuit from %s:%d", + file, lineno); + return; + } + tor_assert(circuitmux_attached_circuit_direction(cmux, circ) == direction); + + /* Update the number of cells we have for the circuit mux */ + if (direction == CELL_DIRECTION_OUT) { + circuitmux_set_num_cells(cmux, circ, circ->n_chan_cells.n); + } else { + circuitmux_set_num_cells(cmux, circ, or_circ->p_chan_cells.n); + } +} + +/** Remove all circuits from the cmux on <b>chan</b>. + * + * If <b>circuits_out</b> is non-NULL, add all detached circuits to + * <b>circuits_out</b>. + **/ +void +channel_unlink_all_circuits(channel_t *chan, smartlist_t *circuits_out) +{ + tor_assert(chan); + tor_assert(chan->cmux); + + circuitmux_detach_all_circuits(chan->cmux, circuits_out); + chan->num_n_circuits = 0; + chan->num_p_circuits = 0; +} + +/** Block (if <b>block</b> is true) or unblock (if <b>block</b> is false) + * every edge connection that is using <b>circ</b> to write to <b>chan</b>, + * and start or stop reading as appropriate. + * + * If <b>stream_id</b> is nonzero, block only the edge connection whose + * stream_id matches it. + * + * Returns the number of streams whose status we changed. + */ +static int +set_streams_blocked_on_circ(circuit_t *circ, channel_t *chan, + int block, streamid_t stream_id) +{ + edge_connection_t *edge = NULL; + int n = 0; + if (circ->n_chan == chan) { + circ->streams_blocked_on_n_chan = block; + if (CIRCUIT_IS_ORIGIN(circ)) + edge = TO_ORIGIN_CIRCUIT(circ)->p_streams; + } else { + circ->streams_blocked_on_p_chan = block; + tor_assert(!CIRCUIT_IS_ORIGIN(circ)); + edge = TO_OR_CIRCUIT(circ)->n_streams; + } + + for (; edge; edge = edge->next_stream) { + connection_t *conn = TO_CONN(edge); + if (stream_id && edge->stream_id != stream_id) + continue; + + if (edge->edge_blocked_on_circ != block) { + ++n; + edge->edge_blocked_on_circ = block; + } + + if (!conn->read_event) { + /* This connection is a placeholder for something; probably a DNS + * request. It can't actually stop or start reading.*/ + continue; + } + + if (block) { + if (connection_is_reading(conn)) + connection_stop_reading(conn); + } else { + /* Is this right? */ + if (!connection_is_reading(conn)) + connection_start_reading(conn); + } + } + + return n; +} + +/** Extract the command from a packed cell. */ +static uint8_t +packed_cell_get_command(const packed_cell_t *cell, int wide_circ_ids) +{ + if (wide_circ_ids) { + return get_uint8(cell->body+4); + } else { + return get_uint8(cell->body+2); + } +} + +/** Extract the circuit ID from a packed cell. */ +circid_t +packed_cell_get_circid(const packed_cell_t *cell, int wide_circ_ids) +{ + if (wide_circ_ids) { + return ntohl(get_uint32(cell->body)); + } else { + return ntohs(get_uint16(cell->body)); + } +} + +/** Pull as many cells as possible (but no more than <b>max</b>) from the + * queue of the first active circuit on <b>chan</b>, and write them to + * <b>chan</b>->outbuf. Return the number of cells written. Advance + * the active circuit pointer to the next active circuit in the ring. */ +MOCK_IMPL(int, +channel_flush_from_first_active_circuit, (channel_t *chan, int max)) +{ + circuitmux_t *cmux = NULL; + int n_flushed = 0; + cell_queue_t *queue; + destroy_cell_queue_t *destroy_queue=NULL; + circuit_t *circ; + or_circuit_t *or_circ; + int streams_blocked; + packed_cell_t *cell; + + /* Get the cmux */ + tor_assert(chan); + tor_assert(chan->cmux); + cmux = chan->cmux; + + /* Main loop: pick a circuit, send a cell, update the cmux */ + while (n_flushed < max) { + circ = circuitmux_get_first_active_circuit(cmux, &destroy_queue); + if (destroy_queue) { + destroy_cell_t *dcell; + /* this code is duplicated from some of the logic below. Ugly! XXXX */ + /* If we are given a destroy_queue here, then it is required to be + * nonempty... */ + tor_assert(destroy_queue->n > 0); + dcell = destroy_cell_queue_pop(destroy_queue); + /* ...and pop() will always yield a cell from a nonempty queue. */ + tor_assert(dcell); + /* frees dcell */ + cell = destroy_cell_to_packed_cell(dcell, chan->wide_circ_ids); + /* Send the DESTROY cell. It is very unlikely that this fails but just + * in case, get rid of the channel. */ + if (channel_write_packed_cell(chan, cell) < 0) { + /* The cell has been freed. */ + channel_mark_for_close(chan); + continue; + } + /* Update the cmux destroy counter */ + circuitmux_notify_xmit_destroy(cmux); + cell = NULL; + ++n_flushed; + continue; + } + /* If it returns NULL, no cells left to send */ + if (!circ) break; + + if (circ->n_chan == chan) { + queue = &circ->n_chan_cells; + streams_blocked = circ->streams_blocked_on_n_chan; + } else { + or_circ = TO_OR_CIRCUIT(circ); + tor_assert(or_circ->p_chan == chan); + queue = &TO_OR_CIRCUIT(circ)->p_chan_cells; + streams_blocked = circ->streams_blocked_on_p_chan; + } + + /* Circuitmux told us this was active, so it should have cells */ + if (/*BUG(*/ queue->n == 0 /*)*/) { + log_warn(LD_BUG, "Found a supposedly active circuit with no cells " + "to send. Trying to recover."); + circuitmux_set_num_cells(cmux, circ, 0); + if (! circ->marked_for_close) + circuit_mark_for_close(circ, END_CIRC_REASON_INTERNAL); + continue; + } + + tor_assert(queue->n > 0); + + /* + * Get just one cell here; once we've sent it, that can change the circuit + * selection, so we have to loop around for another even if this circuit + * has more than one. + */ + cell = cell_queue_pop(queue); + + /* Calculate the exact time that this cell has spent in the queue. */ + if (get_options()->CellStatistics || + get_options()->TestingEnableCellStatsEvent) { + uint32_t timestamp_now = monotime_coarse_get_stamp(); + uint32_t msec_waiting = + (uint32_t) monotime_coarse_stamp_units_to_approx_msec( + timestamp_now - cell->inserted_timestamp); + + if (get_options()->CellStatistics && !CIRCUIT_IS_ORIGIN(circ)) { + or_circ = TO_OR_CIRCUIT(circ); + or_circ->total_cell_waiting_time += msec_waiting; + or_circ->processed_cells++; + } + + if (get_options()->TestingEnableCellStatsEvent) { + uint8_t command = packed_cell_get_command(cell, chan->wide_circ_ids); + + testing_cell_stats_entry_t *ent = + tor_malloc_zero(sizeof(testing_cell_stats_entry_t)); + ent->command = command; + ent->waiting_time = msec_waiting / 10; + ent->removed = 1; + if (circ->n_chan == chan) + ent->exitward = 1; + if (!circ->testing_cell_stats) + circ->testing_cell_stats = smartlist_new(); + smartlist_add(circ->testing_cell_stats, ent); + } + } + + /* If we just flushed our queue and this circuit is used for a + * tunneled directory request, possibly advance its state. */ + if (queue->n == 0 && chan->dirreq_id) + geoip_change_dirreq_state(chan->dirreq_id, + DIRREQ_TUNNELED, + DIRREQ_CIRC_QUEUE_FLUSHED); + + /* Now send the cell. It is very unlikely that this fails but just in + * case, get rid of the channel. */ + if (channel_write_packed_cell(chan, cell) < 0) { + /* The cell has been freed at this point. */ + channel_mark_for_close(chan); + continue; + } + cell = NULL; + + /* + * Don't packed_cell_free_unchecked(cell) here because the channel will + * do so when it gets out of the channel queue (probably already did, in + * which case that was an immediate double-free bug). + */ + + /* Update the counter */ + ++n_flushed; + + /* + * Now update the cmux; tell it we've just sent a cell, and how many + * we have left. + */ + circuitmux_notify_xmit_cells(cmux, circ, 1); + circuitmux_set_num_cells(cmux, circ, queue->n); + if (queue->n == 0) + log_debug(LD_GENERAL, "Made a circuit inactive."); + + /* Is the cell queue low enough to unblock all the streams that are waiting + * to write to this circuit? */ + if (streams_blocked && queue->n <= CELL_QUEUE_LOWWATER_SIZE) + set_streams_blocked_on_circ(circ, chan, 0, 0); /* unblock streams */ + + /* If n_flushed < max still, loop around and pick another circuit */ + } + + /* Okay, we're done sending now */ + return n_flushed; +} + +/* Minimum value is the maximum circuit window size. + * + * SENDME cells makes it that we can control how many cells can be inflight on + * a circuit from end to end. This logic makes it that on any circuit cell + * queue, we have a maximum of cells possible. + * + * Because the Tor protocol allows for a client to exit at any hop in a + * circuit and a circuit can be of a maximum of 8 hops, so in theory the + * normal worst case will be the circuit window start value times the maximum + * number of hops (8). Having more cells then that means something is wrong. + * + * However, because padding cells aren't counted in the package window, we set + * the maximum size to a reasonably large size for which we expect that we'll + * never reach in theory. And if we ever do because of future changes, we'll + * be able to control it with a consensus parameter. + * + * XXX: Unfortunately, END cells aren't accounted for in the circuit window + * which means that for instance if a client opens 8001 streams, the 8001 + * following END cells will queue up in the circuit which will get closed if + * the max limit is 8000. Which is sad because it is allowed by the Tor + * protocol. But, we need an upper bound on circuit queue in order to avoid + * DoS memory pressure so the default size is a middle ground between not + * having any limit and having a very restricted one. This is why we can also + * control it through a consensus parameter. */ +#define RELAY_CIRC_CELL_QUEUE_SIZE_MIN CIRCWINDOW_START_MAX +/* We can't have a consensus parameter above this value. */ +#define RELAY_CIRC_CELL_QUEUE_SIZE_MAX INT32_MAX +/* Default value is set to a large value so we can handle padding cells + * properly which aren't accounted for in the SENDME window. Default is 50000 + * allowed cells in the queue resulting in ~25MB. */ +#define RELAY_CIRC_CELL_QUEUE_SIZE_DEFAULT \ + (50 * RELAY_CIRC_CELL_QUEUE_SIZE_MIN) + +/* The maximum number of cell a circuit queue can contain. This is updated at + * every new consensus and controlled by a parameter. */ +static int32_t max_circuit_cell_queue_size = + RELAY_CIRC_CELL_QUEUE_SIZE_DEFAULT; + +/* Called when the consensus has changed. At this stage, the global consensus + * object has NOT been updated. It is called from + * notify_before_networkstatus_changes(). */ +void +relay_consensus_has_changed(const networkstatus_t *ns) +{ + tor_assert(ns); + + /* Update the circuit max cell queue size from the consensus. */ + max_circuit_cell_queue_size = + networkstatus_get_param(ns, "circ_max_cell_queue_size", + RELAY_CIRC_CELL_QUEUE_SIZE_DEFAULT, + RELAY_CIRC_CELL_QUEUE_SIZE_MIN, + RELAY_CIRC_CELL_QUEUE_SIZE_MAX); +} + +/** Add <b>cell</b> to the queue of <b>circ</b> writing to <b>chan</b> + * transmitting in <b>direction</b>. + * + * The given <b>cell</b> is copied onto the circuit queue so the caller must + * cleanup the memory. + * + * This function is part of the fast path. */ +void +append_cell_to_circuit_queue(circuit_t *circ, channel_t *chan, + cell_t *cell, cell_direction_t direction, + streamid_t fromstream) +{ + or_circuit_t *orcirc = NULL; + cell_queue_t *queue; + int streams_blocked; + int exitward; + if (circ->marked_for_close) + return; + + exitward = (direction == CELL_DIRECTION_OUT); + if (exitward) { + queue = &circ->n_chan_cells; + streams_blocked = circ->streams_blocked_on_n_chan; + } else { + orcirc = TO_OR_CIRCUIT(circ); + queue = &orcirc->p_chan_cells; + streams_blocked = circ->streams_blocked_on_p_chan; + } + + if (PREDICT_UNLIKELY(queue->n >= max_circuit_cell_queue_size)) { + log_fn(LOG_PROTOCOL_WARN, LD_PROTOCOL, + "%s circuit has %d cells in its queue, maximum allowed is %d. " + "Closing circuit for safety reasons.", + (exitward) ? "Outbound" : "Inbound", queue->n, + max_circuit_cell_queue_size); + circuit_mark_for_close(circ, END_CIRC_REASON_RESOURCELIMIT); + stats_n_circ_max_cell_reached++; + return; + } + + /* Very important that we copy to the circuit queue because all calls to + * this function use the stack for the cell memory. */ + cell_queue_append_packed_copy(circ, queue, exitward, cell, + chan->wide_circ_ids, 1); + + /* Check and run the OOM if needed. */ + if (PREDICT_UNLIKELY(cell_queues_check_size())) { + /* We ran the OOM handler which might have closed this circuit. */ + if (circ->marked_for_close) + return; + } + + /* If we have too many cells on the circuit, we should stop reading from + * the edge streams for a while. */ + if (!streams_blocked && queue->n >= CELL_QUEUE_HIGHWATER_SIZE) + set_streams_blocked_on_circ(circ, chan, 1, 0); /* block streams */ + + if (streams_blocked && fromstream) { + /* This edge connection is apparently not blocked; block it. */ + set_streams_blocked_on_circ(circ, chan, 1, fromstream); + } + + update_circuit_on_cmux(circ, direction); + if (queue->n == 1) { + /* This was the first cell added to the queue. We just made this + * circuit active. */ + log_debug(LD_GENERAL, "Made a circuit active."); + } + + /* New way: mark this as having waiting cells for the scheduler */ + scheduler_channel_has_waiting_cells(chan); +} + +/** Append an encoded value of <b>addr</b> to <b>payload_out</b>, which must + * have at least 18 bytes of free space. The encoding is, as specified in + * tor-spec.txt: + * RESOLVED_TYPE_IPV4 or RESOLVED_TYPE_IPV6 [1 byte] + * LENGTH [1 byte] + * ADDRESS [length bytes] + * Return the number of bytes added, or -1 on error */ +int +append_address_to_payload(uint8_t *payload_out, const tor_addr_t *addr) +{ + uint32_t a; + switch (tor_addr_family(addr)) { + case AF_INET: + payload_out[0] = RESOLVED_TYPE_IPV4; + payload_out[1] = 4; + a = tor_addr_to_ipv4n(addr); + memcpy(payload_out+2, &a, 4); + return 6; + case AF_INET6: + payload_out[0] = RESOLVED_TYPE_IPV6; + payload_out[1] = 16; + memcpy(payload_out+2, tor_addr_to_in6_addr8(addr), 16); + return 18; + case AF_UNSPEC: + default: + return -1; + } +} + +/** Given <b>payload_len</b> bytes at <b>payload</b>, starting with an address + * encoded as by append_address_to_payload(), try to decode the address into + * *<b>addr_out</b>. Return the next byte in the payload after the address on + * success, or NULL on failure. */ +const uint8_t * +decode_address_from_payload(tor_addr_t *addr_out, const uint8_t *payload, + int payload_len) +{ + if (payload_len < 2) + return NULL; + if (payload_len < 2+payload[1]) + return NULL; + + switch (payload[0]) { + case RESOLVED_TYPE_IPV4: + if (payload[1] != 4) + return NULL; + tor_addr_from_ipv4n(addr_out, get_uint32(payload+2)); + break; + case RESOLVED_TYPE_IPV6: + if (payload[1] != 16) + return NULL; + tor_addr_from_ipv6_bytes(addr_out, (char*)(payload+2)); + break; + default: + tor_addr_make_unspec(addr_out); + break; + } + return payload + 2 + payload[1]; +} + +/** Remove all the cells queued on <b>circ</b> for <b>chan</b>. */ +void +circuit_clear_cell_queue(circuit_t *circ, channel_t *chan) +{ + cell_queue_t *queue; + cell_direction_t direction; + + if (circ->n_chan == chan) { + queue = &circ->n_chan_cells; + direction = CELL_DIRECTION_OUT; + } else { + or_circuit_t *orcirc = TO_OR_CIRCUIT(circ); + tor_assert(orcirc->p_chan == chan); + queue = &orcirc->p_chan_cells; + direction = CELL_DIRECTION_IN; + } + + /* Clear the queue */ + cell_queue_clear(queue); + + /* Update the cell counter in the cmux */ + if (chan->cmux && circuitmux_is_circuit_attached(chan->cmux, circ)) + update_circuit_on_cmux(circ, direction); +} + +/** Return 1 if we shouldn't restart reading on this circuit, even if + * we get a SENDME. Else return 0. +*/ +static int +circuit_queue_streams_are_blocked(circuit_t *circ) +{ + if (CIRCUIT_IS_ORIGIN(circ)) { + return circ->streams_blocked_on_n_chan; + } else { + return circ->streams_blocked_on_p_chan; + } +} diff --cc src/feature/client/circpathbias.c index 1ee29c639,000000000..9f2ed9347 mode 100644,000000..100644 --- a/src/feature/client/circpathbias.c +++ b/src/feature/client/circpathbias.c @@@ -1,1578 -1,0 +1,1641 @@@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file circpathbias.c + * + * \brief Code to track success/failure rates of circuits built through + * different tor nodes, in an attempt to detect attacks where + * an attacker deliberately causes circuits to fail until the client + * choses a path they like. + * + * This code is currently configured in a warning-only mode, though false + * positives appear to be rare in practice. There is also support for + * disabling really bad guards, but it's quite experimental and may have bad + * anonymity effects. + * + * The information here is associated with the entry_guard_t object for + * each guard, and stored persistently in the state file. + */ + +#include "core/or/or.h" +#include "core/or/channel.h" +#include "feature/client/circpathbias.h" +#include "core/or/circuitbuild.h" +#include "core/or/circuitlist.h" +#include "core/or/circuituse.h" +#include "core/or/circuitstats.h" +#include "core/or/connection_edge.h" +#include "app/config/config.h" +#include "lib/crypt_ops/crypto_rand.h" +#include "feature/client/entrynodes.h" +#include "feature/nodelist/networkstatus.h" +#include "core/or/relay.h" +#include "lib/math/fp.h" +#include "lib/math/laplace.h" + +#include "core/or/cell_st.h" +#include "core/or/cpath_build_state_st.h" +#include "core/or/crypt_path_st.h" +#include "core/or/extend_info_st.h" +#include "core/or/origin_circuit_st.h" + +static void pathbias_count_successful_close(origin_circuit_t *circ); +static void pathbias_count_collapse(origin_circuit_t *circ); +static void pathbias_count_use_failed(origin_circuit_t *circ); +static void pathbias_measure_use_rate(entry_guard_t *guard); +static void pathbias_measure_close_rate(entry_guard_t *guard); +static void pathbias_scale_use_rates(entry_guard_t *guard); +static void pathbias_scale_close_rates(entry_guard_t *guard); +static int entry_guard_inc_circ_attempt_count(entry_guard_t *guard); + +/** Increment the number of times we successfully extended a circuit to + * <b>guard</b>, first checking if the failure rate is high enough that + * we should eliminate the guard. Return -1 if the guard looks no good; + * return 0 if the guard looks fine. + */ +static int +entry_guard_inc_circ_attempt_count(entry_guard_t *guard) +{ + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + entry_guards_changed(); + + pathbias_measure_close_rate(guard); + + if (pb->path_bias_disabled) + return -1; + + pathbias_scale_close_rates(guard); + pb->circ_attempts++; + + log_info(LD_CIRC, "Got success count %f/%f for guard %s", + pb->circ_successes, pb->circ_attempts, + entry_guard_describe(guard)); + return 0; +} + +/** The minimum number of circuit attempts before we start + * thinking about warning about path bias and dropping guards */ +static int +pathbias_get_min_circs(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_MIN_CIRC 150 + if (options->PathBiasCircThreshold >= 5) + return options->PathBiasCircThreshold; + else + return networkstatus_get_param(NULL, "pb_mincircs", + DFLT_PATH_BIAS_MIN_CIRC, + 5, INT32_MAX); +} + +/** The circuit success rate below which we issue a notice */ +static double +pathbias_get_notice_rate(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_NOTICE_PCT 70 + if (options->PathBiasNoticeRate >= 0.0) + return options->PathBiasNoticeRate; + else + return networkstatus_get_param(NULL, "pb_noticepct", + DFLT_PATH_BIAS_NOTICE_PCT, 0, 100)/100.0; +} + +/** The circuit success rate below which we issue a warn */ +static double +pathbias_get_warn_rate(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_WARN_PCT 50 + if (options->PathBiasWarnRate >= 0.0) + return options->PathBiasWarnRate; + else + return networkstatus_get_param(NULL, "pb_warnpct", + DFLT_PATH_BIAS_WARN_PCT, 0, 100)/100.0; +} + +/* XXXX I'd like to have this be static again, but entrynodes.c needs it. */ +/** + * The extreme rate is the rate at which we would drop the guard, + * if pb_dropguard is also set. Otherwise we just warn. + */ +double +pathbias_get_extreme_rate(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_EXTREME_PCT 30 + if (options->PathBiasExtremeRate >= 0.0) + return options->PathBiasExtremeRate; + else + return networkstatus_get_param(NULL, "pb_extremepct", + DFLT_PATH_BIAS_EXTREME_PCT, 0, 100)/100.0; +} + +/* XXXX I'd like to have this be static again, but entrynodes.c needs it. */ +/** + * If 1, we actually disable use of guards that fall below + * the extreme_pct. + */ +int +pathbias_get_dropguards(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_DROP_GUARDS 0 + if (options->PathBiasDropGuards >= 0) + return options->PathBiasDropGuards; + else + return networkstatus_get_param(NULL, "pb_dropguards", + DFLT_PATH_BIAS_DROP_GUARDS, 0, 1); +} + +/** + * This is the number of circuits at which we scale our + * counts by mult_factor/scale_factor. Note, this count is + * not exact, as we only perform the scaling in the event + * of no integer truncation. + */ +static int +pathbias_get_scale_threshold(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_SCALE_THRESHOLD 300 + if (options->PathBiasScaleThreshold >= 10) + return options->PathBiasScaleThreshold; + else + return networkstatus_get_param(NULL, "pb_scalecircs", + DFLT_PATH_BIAS_SCALE_THRESHOLD, 10, + INT32_MAX); +} + +/** + * Compute the path bias scaling ratio from the consensus + * parameters pb_multfactor/pb_scalefactor. + * + * Returns a value in (0, 1.0] which we multiply our pathbias + * counts with to scale them down. + */ +static double +pathbias_get_scale_ratio(const or_options_t *options) +{ + /* + * The scale factor is the denominator for our scaling + * of circuit counts for our path bias window. + * + * Note that our use of doubles for the path bias state + * file means that powers of 2 work best here. + */ + int denominator = networkstatus_get_param(NULL, "pb_scalefactor", + 2, 2, INT32_MAX); + (void) options; + /** + * The mult factor is the numerator for our scaling + * of circuit counts for our path bias window. It + * allows us to scale by fractions. + */ + return networkstatus_get_param(NULL, "pb_multfactor", + 1, 1, denominator)/((double)denominator); +} + +/** The minimum number of circuit usage attempts before we start + * thinking about warning about path use bias and dropping guards */ +static int +pathbias_get_min_use(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_MIN_USE 20 + if (options->PathBiasUseThreshold >= 3) + return options->PathBiasUseThreshold; + else + return networkstatus_get_param(NULL, "pb_minuse", + DFLT_PATH_BIAS_MIN_USE, + 3, INT32_MAX); +} + +/** The circuit use success rate below which we issue a notice */ +static double +pathbias_get_notice_use_rate(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_NOTICE_USE_PCT 80 + if (options->PathBiasNoticeUseRate >= 0.0) + return options->PathBiasNoticeUseRate; + else + return networkstatus_get_param(NULL, "pb_noticeusepct", + DFLT_PATH_BIAS_NOTICE_USE_PCT, + 0, 100)/100.0; +} + +/** + * The extreme use rate is the rate at which we would drop the guard, + * if pb_dropguard is also set. Otherwise we just warn. + */ +double +pathbias_get_extreme_use_rate(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_EXTREME_USE_PCT 60 + if (options->PathBiasExtremeUseRate >= 0.0) + return options->PathBiasExtremeUseRate; + else + return networkstatus_get_param(NULL, "pb_extremeusepct", + DFLT_PATH_BIAS_EXTREME_USE_PCT, + 0, 100)/100.0; +} + +/** + * This is the number of circuits at which we scale our + * use counts by mult_factor/scale_factor. Note, this count is + * not exact, as we only perform the scaling in the event + * of no integer truncation. + */ +static int +pathbias_get_scale_use_threshold(const or_options_t *options) +{ +#define DFLT_PATH_BIAS_SCALE_USE_THRESHOLD 100 + if (options->PathBiasScaleUseThreshold >= 10) + return options->PathBiasScaleUseThreshold; + else + return networkstatus_get_param(NULL, "pb_scaleuse", + DFLT_PATH_BIAS_SCALE_USE_THRESHOLD, + 10, INT32_MAX); +} + +/** + * Convert a Guard's path state to string. + */ +const char * +pathbias_state_to_string(path_state_t state) +{ + switch (state) { + case PATH_STATE_NEW_CIRC: + return "new"; + case PATH_STATE_BUILD_ATTEMPTED: + return "build attempted"; + case PATH_STATE_BUILD_SUCCEEDED: + return "build succeeded"; + case PATH_STATE_USE_ATTEMPTED: + return "use attempted"; + case PATH_STATE_USE_SUCCEEDED: + return "use succeeded"; + case PATH_STATE_USE_FAILED: + return "use failed"; + case PATH_STATE_ALREADY_COUNTED: + return "already counted"; + } + + return "unknown"; +} + +/** + * This function decides if a circuit has progressed far enough to count + * as a circuit "attempt". As long as end-to-end tagging is possible, + * we assume the adversary will use it over hop-to-hop failure. Therefore, + * we only need to account bias for the last hop. This should make us + * much more resilient to ambient circuit failure, and also make that + * failure easier to measure (we only need to measure Exit failure rates). + */ +static int +pathbias_is_new_circ_attempt(origin_circuit_t *circ) +{ +#define N2N_TAGGING_IS_POSSIBLE +#ifdef N2N_TAGGING_IS_POSSIBLE + /* cpath is a circular list. We want circs with more than one hop, + * and the second hop must be waiting for keys still (it's just + * about to get them). */ + return circ->cpath && + circ->cpath->next != circ->cpath && + circ->cpath->next->state == CPATH_STATE_AWAITING_KEYS; +#else /* !(defined(N2N_TAGGING_IS_POSSIBLE)) */ + /* If tagging attacks are no longer possible, we probably want to + * count bias from the first hop. However, one could argue that + * timing-based tagging is still more useful than per-hop failure. + * In which case, we'd never want to use this. + */ + return circ->cpath && + circ->cpath->state == CPATH_STATE_AWAITING_KEYS; +#endif /* defined(N2N_TAGGING_IS_POSSIBLE) */ +} + +/** + * Decide if the path bias code should count a circuit. + * + * @returns 1 if we should count it, 0 otherwise. + */ +static int +pathbias_should_count(origin_circuit_t *circ) +{ +#define PATHBIAS_COUNT_INTERVAL (600) + static ratelim_t count_limit = + RATELIM_INIT(PATHBIAS_COUNT_INTERVAL); + char *rate_msg = NULL; + + /* We can't do path bias accounting without entry guards. + * Testing and controller circuits also have no guards. + * + * We also don't count server-side rends, because their + * endpoint could be chosen maliciously. + * Similarly, we can't count client-side intro attempts, + * because clients can be manipulated into connecting to + * malicious intro points. */ + if (get_options()->UseEntryGuards == 0 || + circ->base_.purpose == CIRCUIT_PURPOSE_TESTING || + circ->base_.purpose == CIRCUIT_PURPOSE_CONTROLLER || + circ->base_.purpose == CIRCUIT_PURPOSE_S_CONNECT_REND || + circ->base_.purpose == CIRCUIT_PURPOSE_S_REND_JOINED || + (circ->base_.purpose >= CIRCUIT_PURPOSE_C_INTRODUCING && + circ->base_.purpose <= CIRCUIT_PURPOSE_C_INTRODUCE_ACKED)) { + + /* Check to see if the shouldcount result has changed due to a + * unexpected purpose change that would affect our results. + * + * The reason we check the path state too here is because for the + * cannibalized versions of these purposes, we count them as successful + * before their purpose change. + */ + if (circ->pathbias_shouldcount == PATHBIAS_SHOULDCOUNT_COUNTED + && circ->path_state != PATH_STATE_ALREADY_COUNTED) { + log_info(LD_BUG, + "Circuit %d is now being ignored despite being counted " + "in the past. Purpose is %s, path state is %s", + circ->global_identifier, + circuit_purpose_to_string(circ->base_.purpose), + pathbias_state_to_string(circ->path_state)); + } + circ->pathbias_shouldcount = PATHBIAS_SHOULDCOUNT_IGNORED; + return 0; + } + + /* Completely ignore one hop circuits */ + if (circ->build_state->onehop_tunnel || + circ->build_state->desired_path_len == 1) { + /* Check for inconsistency */ + if (circ->build_state->desired_path_len != 1 || + !circ->build_state->onehop_tunnel) { + if ((rate_msg = rate_limit_log(&count_limit, approx_time()))) { + log_info(LD_BUG, + "One-hop circuit has length %d. Path state is %s. " + "Circuit is a %s currently %s.%s", + circ->build_state->desired_path_len, + pathbias_state_to_string(circ->path_state), + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state), + rate_msg); + tor_free(rate_msg); + } + tor_fragile_assert(); + } + + /* Check to see if the shouldcount result has changed due to a + * unexpected change that would affect our results */ + if (circ->pathbias_shouldcount == PATHBIAS_SHOULDCOUNT_COUNTED) { + log_info(LD_BUG, + "One-hop circuit %d is now being ignored despite being counted " + "in the past. Purpose is %s, path state is %s", + circ->global_identifier, + circuit_purpose_to_string(circ->base_.purpose), + pathbias_state_to_string(circ->path_state)); + } + circ->pathbias_shouldcount = PATHBIAS_SHOULDCOUNT_IGNORED; + return 0; + } + + /* Check to see if the shouldcount result has changed due to a + * unexpected purpose change that would affect our results */ + if (circ->pathbias_shouldcount == PATHBIAS_SHOULDCOUNT_IGNORED) { + log_info(LD_BUG, + "Circuit %d is now being counted despite being ignored " + "in the past. Purpose is %s, path state is %s", + circ->global_identifier, + circuit_purpose_to_string(circ->base_.purpose), + pathbias_state_to_string(circ->path_state)); + } + circ->pathbias_shouldcount = PATHBIAS_SHOULDCOUNT_COUNTED; + + return 1; +} + +/** + * Check our circuit state to see if this is a successful circuit attempt. + * If so, record it in the current guard's path bias circ_attempt count. + * + * Also check for several potential error cases for bug #6475. + */ +int +pathbias_count_build_attempt(origin_circuit_t *circ) +{ +#define CIRC_ATTEMPT_NOTICE_INTERVAL (600) + static ratelim_t circ_attempt_notice_limit = + RATELIM_INIT(CIRC_ATTEMPT_NOTICE_INTERVAL); + char *rate_msg = NULL; + + if (!pathbias_should_count(circ)) { + return 0; + } + + if (pathbias_is_new_circ_attempt(circ)) { + /* Help track down the real cause of bug #6475: */ + if (circ->has_opened && circ->path_state != PATH_STATE_BUILD_ATTEMPTED) { + if ((rate_msg = rate_limit_log(&circ_attempt_notice_limit, + approx_time()))) { + log_info(LD_BUG, + "Opened circuit is in strange path state %s. " + "Circuit is a %s currently %s.%s", + pathbias_state_to_string(circ->path_state), + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state), + rate_msg); + tor_free(rate_msg); + } + } + + /* Don't re-count cannibalized circs.. */ + if (!circ->has_opened) { + entry_guard_t *guard = NULL; + + if (circ->cpath && circ->cpath->extend_info) { + guard = entry_guard_get_by_id_digest( + circ->cpath->extend_info->identity_digest); + } else if (circ->base_.n_chan) { + guard = + entry_guard_get_by_id_digest(circ->base_.n_chan->identity_digest); + } + + if (guard) { + if (circ->path_state == PATH_STATE_NEW_CIRC) { + circ->path_state = PATH_STATE_BUILD_ATTEMPTED; + + if (entry_guard_inc_circ_attempt_count(guard) < 0) { + /* Bogus guard; we already warned. */ + return -END_CIRC_REASON_TORPROTOCOL; + } + } else { + if ((rate_msg = rate_limit_log(&circ_attempt_notice_limit, + approx_time()))) { + log_info(LD_BUG, + "Unopened circuit has strange path state %s. " + "Circuit is a %s currently %s.%s", + pathbias_state_to_string(circ->path_state), + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state), + rate_msg); + tor_free(rate_msg); + } + } + } else { + if ((rate_msg = rate_limit_log(&circ_attempt_notice_limit, + approx_time()))) { + log_info(LD_CIRC, + "Unopened circuit has no known guard. " + "Circuit is a %s currently %s.%s", + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state), + rate_msg); + tor_free(rate_msg); + } + } + } + } + + return 0; +} + +/** + * Check our circuit state to see if this is a successful circuit + * completion. If so, record it in the current guard's path bias + * success count. + * + * Also check for several potential error cases for bug #6475. + */ +void +pathbias_count_build_success(origin_circuit_t *circ) +{ +#define SUCCESS_NOTICE_INTERVAL (600) + static ratelim_t success_notice_limit = + RATELIM_INIT(SUCCESS_NOTICE_INTERVAL); + char *rate_msg = NULL; + entry_guard_t *guard = NULL; + + if (!pathbias_should_count(circ)) { + return; + } + + /* Don't count cannibalized/reused circs for path bias + * "build" success, since they get counted under "use" success. */ + if (!circ->has_opened) { + if (circ->cpath && circ->cpath->extend_info) { + guard = entry_guard_get_by_id_digest( + circ->cpath->extend_info->identity_digest); + } + + if (guard) { + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + if (circ->path_state == PATH_STATE_BUILD_ATTEMPTED) { + circ->path_state = PATH_STATE_BUILD_SUCCEEDED; + pb->circ_successes++; + entry_guards_changed(); + + log_info(LD_CIRC, "Got success count %f/%f for guard %s", + pb->circ_successes, pb->circ_attempts, + entry_guard_describe(guard)); + } else { + if ((rate_msg = rate_limit_log(&success_notice_limit, + approx_time()))) { + log_info(LD_BUG, + "Succeeded circuit is in strange path state %s. " + "Circuit is a %s currently %s.%s", + pathbias_state_to_string(circ->path_state), + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state), + rate_msg); + tor_free(rate_msg); + } + } + + if (pb->circ_attempts < pb->circ_successes) { + log_notice(LD_BUG, "Unexpectedly high successes counts (%f/%f) " + "for guard %s", + pb->circ_successes, pb->circ_attempts, + entry_guard_describe(guard)); + } + /* In rare cases, CIRCUIT_PURPOSE_TESTING can get converted to + * CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT and have no guards here. + * No need to log that case. */ + } else if (circ->base_.purpose != CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT) { + if ((rate_msg = rate_limit_log(&success_notice_limit, + approx_time()))) { + log_info(LD_CIRC, + "Completed circuit has no known guard. " + "Circuit is a %s currently %s.%s", + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state), + rate_msg); + tor_free(rate_msg); + } + } + } else { + if (circ->path_state < PATH_STATE_BUILD_SUCCEEDED) { + if ((rate_msg = rate_limit_log(&success_notice_limit, + approx_time()))) { + log_info(LD_BUG, + "Opened circuit is in strange path state %s. " + "Circuit is a %s currently %s.%s", + pathbias_state_to_string(circ->path_state), + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state), + rate_msg); + tor_free(rate_msg); + } + } + } +} + +/** + * Record an attempt to use a circuit. Changes the circuit's + * path state and update its guard's usage counter. + * + * Used for path bias usage accounting. + */ +void +pathbias_count_use_attempt(origin_circuit_t *circ) +{ + if (!pathbias_should_count(circ)) { + return; + } + + if (circ->path_state < PATH_STATE_BUILD_SUCCEEDED) { + log_notice(LD_BUG, + "Used circuit is in strange path state %s. " + "Circuit is a %s currently %s.", + pathbias_state_to_string(circ->path_state), + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state)); + } else if (circ->path_state < PATH_STATE_USE_ATTEMPTED) { + entry_guard_t *guard = entry_guard_get_by_id_digest( + circ->cpath->extend_info->identity_digest); + if (guard) { + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + pathbias_measure_use_rate(guard); + pathbias_scale_use_rates(guard); + pb->use_attempts++; + entry_guards_changed(); + + log_debug(LD_CIRC, + "Marked circuit %d (%f/%f) as used for guard %s.", + circ->global_identifier, + pb->use_successes, pb->use_attempts, + entry_guard_describe(guard)); + } + + circ->path_state = PATH_STATE_USE_ATTEMPTED; + } else { + /* Harmless but educational log message */ + log_info(LD_CIRC, + "Used circuit %d is already in path state %s. " + "Circuit is a %s currently %s.", + circ->global_identifier, + pathbias_state_to_string(circ->path_state), + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state)); + } + + return; +} + +/** + * Check the circuit's path state is appropriate and mark it as + * successfully used. Used for path bias usage accounting. + * + * We don't actually increment the guard's counters until + * pathbias_check_close(), because the circuit can still transition + * back to PATH_STATE_USE_ATTEMPTED if a stream fails later (this + * is done so we can probe the circuit for liveness at close). + */ +void +pathbias_mark_use_success(origin_circuit_t *circ) +{ + if (!pathbias_should_count(circ)) { + return; + } + + if (circ->path_state < PATH_STATE_USE_ATTEMPTED) { + log_notice(LD_BUG, + "Used circuit %d is in strange path state %s. " + "Circuit is a %s currently %s.", + circ->global_identifier, + pathbias_state_to_string(circ->path_state), + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state)); + + pathbias_count_use_attempt(circ); + } + + /* We don't do any accounting at the guard until actual circuit close */ + circ->path_state = PATH_STATE_USE_SUCCEEDED; + + return; +} + +/** + * If a stream ever detatches from a circuit in a retriable way, + * we need to mark this circuit as still needing either another + * successful stream, or in need of a probe. + * + * An adversary could let the first stream request succeed (ie the + * resolve), but then tag and timeout the remainder (via cell + * dropping), forcing them on new circuits. + * + * Rolling back the state will cause us to probe such circuits, which + * should lead to probe failures in the event of such tagging due to + * either unrecognized cells coming in while we wait for the probe, + * or the cipher state getting out of sync in the case of dropped cells. + */ +void +pathbias_mark_use_rollback(origin_circuit_t *circ) +{ + if (circ->path_state == PATH_STATE_USE_SUCCEEDED) { + log_info(LD_CIRC, + "Rolling back pathbias use state to 'attempted' for detached " + "circuit %d", circ->global_identifier); + circ->path_state = PATH_STATE_USE_ATTEMPTED; + } +} + +/** + * Actually count a circuit success towards a guard's usage counters + * if the path state is appropriate. + */ +static void +pathbias_count_use_success(origin_circuit_t *circ) +{ + entry_guard_t *guard; + + if (!pathbias_should_count(circ)) { + return; + } + + if (circ->path_state != PATH_STATE_USE_SUCCEEDED) { + log_notice(LD_BUG, + "Successfully used circuit %d is in strange path state %s. " + "Circuit is a %s currently %s.", + circ->global_identifier, + pathbias_state_to_string(circ->path_state), + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state)); + } else { + guard = entry_guard_get_by_id_digest( + circ->cpath->extend_info->identity_digest); + if (guard) { + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + pb->use_successes++; + entry_guards_changed(); + + if (pb->use_attempts < pb->use_successes) { + log_notice(LD_BUG, "Unexpectedly high use successes counts (%f/%f) " + "for guard %s", + pb->use_successes, pb->use_attempts, + entry_guard_describe(guard)); + } + + log_debug(LD_CIRC, + "Marked circuit %d (%f/%f) as used successfully for guard %s", + circ->global_identifier, pb->use_successes, + pb->use_attempts, + entry_guard_describe(guard)); + } + } + + return; +} + +/** + * Send a probe down a circuit that the client attempted to use, + * but for which the stream timed out/failed. The probe is a + * RELAY_BEGIN cell with a 0.a.b.c destination address, which + * the exit will reject and reply back, echoing that address. + * + * The reason for such probes is because it is possible to bias + * a user's paths simply by causing timeouts, and these timeouts + * are not possible to differentiate from unresponsive servers. + * + * The probe is sent at the end of the circuit lifetime for two + * reasons: to prevent cryptographic taggers from being able to + * drop cells to cause timeouts, and to prevent easy recognition + * of probes before any real client traffic happens. + * + * Returns -1 if we couldn't probe, 0 otherwise. + */ +static int +pathbias_send_usable_probe(circuit_t *circ) +{ + /* Based on connection_ap_handshake_send_begin() */ + char payload[CELL_PAYLOAD_SIZE]; + int payload_len; + origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); + crypt_path_t *cpath_layer = NULL; + char *probe_nonce = NULL; + + tor_assert(ocirc); + + cpath_layer = ocirc->cpath->prev; + + if (cpath_layer->state != CPATH_STATE_OPEN) { + /* This can happen for cannibalized circuits. Their + * last hop isn't yet open */ + log_info(LD_CIRC, + "Got pathbias probe request for unopened circuit %d. " + "Opened %d, len %d", ocirc->global_identifier, + ocirc->has_opened, ocirc->build_state->desired_path_len); + return -1; + } + + /* We already went down this road. */ + if (circ->purpose == CIRCUIT_PURPOSE_PATH_BIAS_TESTING && + ocirc->pathbias_probe_id) { + log_info(LD_CIRC, + "Got pathbias probe request for circuit %d with " + "outstanding probe", ocirc->global_identifier); + return -1; + } + + /* Can't probe if the channel isn't open */ + if (circ->n_chan == NULL || + (!CHANNEL_IS_OPEN(circ->n_chan) + && !CHANNEL_IS_MAINT(circ->n_chan))) { + log_info(LD_CIRC, + "Skipping pathbias probe for circuit %d: Channel is not open.", + ocirc->global_identifier); + return -1; + } + + circuit_change_purpose(circ, CIRCUIT_PURPOSE_PATH_BIAS_TESTING); + + /* Update timestamp for when circuit_expire_building() should kill us */ + tor_gettimeofday(&circ->timestamp_began); + + /* Generate a random address for the nonce */ + crypto_rand((char*)ô->pathbias_probe_nonce, + sizeof(ocirc->pathbias_probe_nonce)); + ocirc->pathbias_probe_nonce &= 0x00ffffff; + probe_nonce = tor_dup_ip(ocirc->pathbias_probe_nonce); + + tor_snprintf(payload,RELAY_PAYLOAD_SIZE, "%s:25", probe_nonce); + payload_len = (int)strlen(payload)+1; + + // XXX: need this? Can we assume ipv4 will always be supported? + // If not, how do we tell? + //if (payload_len <= RELAY_PAYLOAD_SIZE - 4 && edge_conn->begincell_flags) { + // set_uint32(payload + payload_len, htonl(edge_conn->begincell_flags)); + // payload_len += 4; + //} + + /* Generate+Store stream id, make sure it's non-zero */ + ocirc->pathbias_probe_id = get_unique_stream_id_by_circ(ocirc); + + if (ocirc->pathbias_probe_id==0) { + log_warn(LD_CIRC, + "Ran out of stream IDs on circuit %u during " + "pathbias probe attempt.", ocirc->global_identifier); + tor_free(probe_nonce); + return -1; + } + + log_info(LD_CIRC, + "Sending pathbias testing cell to %s:25 on stream %d for circ %d.", + probe_nonce, ocirc->pathbias_probe_id, ocirc->global_identifier); + tor_free(probe_nonce); + + /* Send a test relay cell */ + if (relay_send_command_from_edge(ocirc->pathbias_probe_id, circ, + RELAY_COMMAND_BEGIN, payload, + payload_len, cpath_layer) < 0) { + log_notice(LD_CIRC, + "Failed to send pathbias probe cell on circuit %d.", + ocirc->global_identifier); + return -1; + } + + /* Mark it freshly dirty so it doesn't get expired in the meantime */ + circ->timestamp_dirty = time(NULL); + + return 0; +} + +/** + * Check the response to a pathbias probe, to ensure the + * cell is recognized and the nonce and other probe + * characteristics are as expected. + * + * If the response is valid, return 0. Otherwise return < 0. + */ +int +pathbias_check_probe_response(circuit_t *circ, const cell_t *cell) +{ + /* Based on connection_edge_process_relay_cell() */ + relay_header_t rh; + int reason; + uint32_t ipv4_host; + origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); + + tor_assert(cell); + tor_assert(ocirc); + tor_assert(circ->purpose == CIRCUIT_PURPOSE_PATH_BIAS_TESTING); + + relay_header_unpack(&rh, cell->payload); + + reason = rh.length > 0 ? + get_uint8(cell->payload+RELAY_HEADER_SIZE) : END_STREAM_REASON_MISC; + + if (rh.command == RELAY_COMMAND_END && + reason == END_STREAM_REASON_EXITPOLICY && + ocirc->pathbias_probe_id == rh.stream_id) { + + /* Check length+extract host: It is in network order after the reason code. + * See connection_edge_end(). */ + if (rh.length < 9) { /* reason+ipv4+dns_ttl */ + log_notice(LD_PROTOCOL, + "Short path bias probe response length field (%d).", rh.length); + return - END_CIRC_REASON_TORPROTOCOL; + } + + ipv4_host = ntohl(get_uint32(cell->payload+RELAY_HEADER_SIZE+1)); + + /* Check nonce */ + if (ipv4_host == ocirc->pathbias_probe_nonce) { + pathbias_mark_use_success(ocirc); ++ circuit_read_valid_data(ocirc, rh.length); + circuit_mark_for_close(circ, END_CIRC_REASON_FINISHED); + log_info(LD_CIRC, + "Got valid path bias probe back for circ %d, stream %d.", + ocirc->global_identifier, ocirc->pathbias_probe_id); + return 0; + } else { + log_notice(LD_CIRC, + "Got strange probe value 0x%x vs 0x%x back for circ %d, " + "stream %d.", ipv4_host, ocirc->pathbias_probe_nonce, + ocirc->global_identifier, ocirc->pathbias_probe_id); + return -1; + } + } + log_info(LD_CIRC, + "Got another cell back back on pathbias probe circuit %d: " + "Command: %d, Reason: %d, Stream-id: %d", + ocirc->global_identifier, rh.command, reason, rh.stream_id); + return -1; +} + +/** ++ * Check if a cell is counts as valid data for a circuit, ++ * and if so, count it as valid. ++ */ ++void ++pathbias_count_valid_cells(circuit_t *circ, const cell_t *cell) ++{ ++ origin_circuit_t *ocirc = TO_ORIGIN_CIRCUIT(circ); ++ relay_header_t rh; ++ ++ relay_header_unpack(&rh, cell->payload); ++ ++ /* Check to see if this is a cell from a previous connection, ++ * or is a request to close the circuit. */ ++ switch (rh.command) { ++ case RELAY_COMMAND_TRUNCATED: ++ /* Truncated cells can arrive on path bias circs. When they do, ++ * just process them. This closes the circ, but it was junk anyway. ++ * No reason to wait for the probe. */ ++ circuit_read_valid_data(ocirc, rh.length); ++ circuit_truncated(TO_ORIGIN_CIRCUIT(circ), ++ get_uint8(cell->payload + RELAY_HEADER_SIZE)); ++ ++ break; ++ ++ case RELAY_COMMAND_END: ++ if (connection_half_edge_is_valid_end(ocirc->half_streams, ++ rh.stream_id)) { ++ circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), rh.length); ++ } ++ break; ++ ++ case RELAY_COMMAND_DATA: ++ if (connection_half_edge_is_valid_data(ocirc->half_streams, ++ rh.stream_id)) { ++ circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), rh.length); ++ } ++ break; ++ ++ case RELAY_COMMAND_SENDME: ++ if (connection_half_edge_is_valid_sendme(ocirc->half_streams, ++ rh.stream_id)) { ++ circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), rh.length); ++ } ++ break; ++ ++ case RELAY_COMMAND_CONNECTED: ++ if (connection_half_edge_is_valid_connected(ocirc->half_streams, ++ rh.stream_id)) { ++ circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), rh.length); ++ } ++ break; ++ ++ case RELAY_COMMAND_RESOLVED: ++ if (connection_half_edge_is_valid_resolved(ocirc->half_streams, ++ rh.stream_id)) { ++ circuit_read_valid_data(TO_ORIGIN_CIRCUIT(circ), rh.length); ++ } ++ break; ++ } ++} ++ ++/** + * Check if a circuit was used and/or closed successfully. + * + * If we attempted to use the circuit to carry a stream but failed + * for whatever reason, or if the circuit mysteriously died before + * we could attach any streams, record these two cases. + * + * If we *have* successfully used the circuit, or it appears to + * have been closed by us locally, count it as a success. + * + * Returns 0 if we're done making decisions with the circ, + * or -1 if we want to probe it first. + */ +int +pathbias_check_close(origin_circuit_t *ocirc, int reason) +{ + circuit_t *circ = ô->base_; + + if (!pathbias_should_count(ocirc)) { + return 0; + } + + switch (ocirc->path_state) { + /* If the circuit was closed after building, but before use, we need + * to ensure we were the ones who tried to close it (and not a remote + * actor). */ + case PATH_STATE_BUILD_SUCCEEDED: + if (reason & END_CIRC_REASON_FLAG_REMOTE) { + /* Remote circ close reasons on an unused circuit all could be bias */ + log_info(LD_CIRC, + "Circuit %d remote-closed without successful use for reason %d. " + "Circuit purpose %d currently %d,%s. Len %d.", + ocirc->global_identifier, + reason, circ->purpose, ocirc->has_opened, + circuit_state_to_string(circ->state), + ocirc->build_state->desired_path_len); + pathbias_count_collapse(ocirc); + } else if ((reason & ~END_CIRC_REASON_FLAG_REMOTE) + == END_CIRC_REASON_CHANNEL_CLOSED && + circ->n_chan && + circ->n_chan->reason_for_closing + != CHANNEL_CLOSE_REQUESTED) { + /* If we didn't close the channel ourselves, it could be bias */ + /* XXX: Only count bias if the network is live? + * What about clock jumps/suspends? */ + log_info(LD_CIRC, + "Circuit %d's channel closed without successful use for reason " + "%d, channel reason %d. Circuit purpose %d currently %d,%s. Len " + "%d.", ocirc->global_identifier, + reason, circ->n_chan->reason_for_closing, + circ->purpose, ocirc->has_opened, + circuit_state_to_string(circ->state), + ocirc->build_state->desired_path_len); + pathbias_count_collapse(ocirc); + } else { + pathbias_count_successful_close(ocirc); + } + break; + + /* If we tried to use a circuit but failed, we should probe it to ensure + * it has not been tampered with. */ + case PATH_STATE_USE_ATTEMPTED: + /* XXX: Only probe and/or count failure if the network is live? + * What about clock jumps/suspends? */ + if (pathbias_send_usable_probe(circ) == 0) + return -1; + else + pathbias_count_use_failed(ocirc); + + /* Any circuit where there were attempted streams but no successful + * streams could be bias */ + log_info(LD_CIRC, + "Circuit %d closed without successful use for reason %d. " + "Circuit purpose %d currently %d,%s. Len %d.", + ocirc->global_identifier, + reason, circ->purpose, ocirc->has_opened, + circuit_state_to_string(circ->state), + ocirc->build_state->desired_path_len); + break; + + case PATH_STATE_USE_SUCCEEDED: + pathbias_count_successful_close(ocirc); + pathbias_count_use_success(ocirc); + break; + + case PATH_STATE_USE_FAILED: + pathbias_count_use_failed(ocirc); + break; + + case PATH_STATE_NEW_CIRC: + case PATH_STATE_BUILD_ATTEMPTED: + case PATH_STATE_ALREADY_COUNTED: + default: + // Other states are uninteresting. No stats to count. + break; + } + + ocirc->path_state = PATH_STATE_ALREADY_COUNTED; + + return 0; +} + +/** + * Count a successfully closed circuit. + */ +static void +pathbias_count_successful_close(origin_circuit_t *circ) +{ + entry_guard_t *guard = NULL; + if (!pathbias_should_count(circ)) { + return; + } + + if (circ->cpath && circ->cpath->extend_info) { + guard = entry_guard_get_by_id_digest( + circ->cpath->extend_info->identity_digest); + } + + if (guard) { + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + /* In the long run: circuit_success ~= successful_circuit_close + + * circ_failure + stream_failure */ + pb->successful_circuits_closed++; + entry_guards_changed(); + } else if (circ->base_.purpose != CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT) { + /* In rare cases, CIRCUIT_PURPOSE_TESTING can get converted to + * CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT and have no guards here. + * No need to log that case. */ + log_info(LD_CIRC, + "Successfully closed circuit has no known guard. " + "Circuit is a %s currently %s", + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state)); + } +} + +/** + * Count a circuit that fails after it is built, but before it can + * carry any traffic. + * + * This is needed because there are ways to destroy a + * circuit after it has successfully completed. Right now, this is + * used for purely informational/debugging purposes. + */ +static void +pathbias_count_collapse(origin_circuit_t *circ) +{ + entry_guard_t *guard = NULL; + + if (!pathbias_should_count(circ)) { + return; + } + + if (circ->cpath && circ->cpath->extend_info) { + guard = entry_guard_get_by_id_digest( + circ->cpath->extend_info->identity_digest); + } + + if (guard) { + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + pb->collapsed_circuits++; + entry_guards_changed(); + } else if (circ->base_.purpose != CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT) { + /* In rare cases, CIRCUIT_PURPOSE_TESTING can get converted to + * CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT and have no guards here. + * No need to log that case. */ + log_info(LD_CIRC, + "Destroyed circuit has no known guard. " + "Circuit is a %s currently %s", + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state)); + } +} + +/** + * Count a known failed circuit (because we could not probe it). + * + * This counter is informational. + */ +static void +pathbias_count_use_failed(origin_circuit_t *circ) +{ + entry_guard_t *guard = NULL; + if (!pathbias_should_count(circ)) { + return; + } + + if (circ->cpath && circ->cpath->extend_info) { + guard = entry_guard_get_by_id_digest( + circ->cpath->extend_info->identity_digest); + } + + if (guard) { + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + pb->unusable_circuits++; + entry_guards_changed(); + } else if (circ->base_.purpose != CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT) { + /* In rare cases, CIRCUIT_PURPOSE_TESTING can get converted to + * CIRCUIT_PURPOSE_C_MEASURE_TIMEOUT and have no guards here. + * No need to log that case. */ + /* XXX note cut-and-paste code in this function compared to nearby + * functions. Would be nice to refactor. -RD */ + log_info(LD_CIRC, + "Stream-failing circuit has no known guard. " + "Circuit is a %s currently %s", + circuit_purpose_to_string(circ->base_.purpose), + circuit_state_to_string(circ->base_.state)); + } +} + +/** + * Count timeouts for path bias log messages. + * + * These counts are purely informational. + */ +void +pathbias_count_timeout(origin_circuit_t *circ) +{ + entry_guard_t *guard = NULL; + + if (!pathbias_should_count(circ)) { + return; + } + + /* For hidden service circs, they can actually be used + * successfully and then time out later (because + * the other side declines to use them). */ + if (circ->path_state == PATH_STATE_USE_SUCCEEDED) { + return; + } + + if (circ->cpath && circ->cpath->extend_info) { + guard = entry_guard_get_by_id_digest( + circ->cpath->extend_info->identity_digest); + } + + if (guard) { + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + pb->timeouts++; + entry_guards_changed(); + } +} + +/** + * Helper function to count all of the currently opened circuits + * for a guard that are in a given path state range. The state + * range is inclusive on both ends. + */ +static int +pathbias_count_circs_in_states(entry_guard_t *guard, + path_state_t from, + path_state_t to) +{ + int open_circuits = 0; + + /* Count currently open circuits. Give them the benefit of the doubt. */ + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, circ) { + origin_circuit_t *ocirc = NULL; + if (!CIRCUIT_IS_ORIGIN(circ) || /* didn't originate here */ + circ->marked_for_close) /* already counted */ + continue; + + ocirc = TO_ORIGIN_CIRCUIT(circ); + + if (!ocirc->cpath || !ocirc->cpath->extend_info) + continue; + + if (ocirc->path_state >= from && + ocirc->path_state <= to && + pathbias_should_count(ocirc) && + fast_memeq(entry_guard_get_rsa_id_digest(guard), + ocirc->cpath->extend_info->identity_digest, + DIGEST_LEN)) { + log_debug(LD_CIRC, "Found opened circuit %d in path_state %s", + ocirc->global_identifier, + pathbias_state_to_string(ocirc->path_state)); + open_circuits++; + } + } + SMARTLIST_FOREACH_END(circ); + + return open_circuits; +} + +/** + * Return the number of circuits counted as successfully closed for + * this guard. + * + * Also add in the currently open circuits to give them the benefit + * of the doubt. + */ +double +pathbias_get_close_success_count(entry_guard_t *guard) +{ + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + return pb->successful_circuits_closed + + pathbias_count_circs_in_states(guard, + PATH_STATE_BUILD_SUCCEEDED, + PATH_STATE_USE_SUCCEEDED); +} + +/** + * Return the number of circuits counted as successfully used + * this guard. + * + * Also add in the currently open circuits that we are attempting + * to use to give them the benefit of the doubt. + */ +double +pathbias_get_use_success_count(entry_guard_t *guard) +{ + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + return pb->use_successes + + pathbias_count_circs_in_states(guard, + PATH_STATE_USE_ATTEMPTED, + PATH_STATE_USE_SUCCEEDED); +} + +/** + * Check the path bias use rate against our consensus parameter limits. + * + * Emits a log message if the use success rates are too low. + * + * If pathbias_get_dropguards() is set, we also disable the use of + * very failure prone guards. + */ +static void +pathbias_measure_use_rate(entry_guard_t *guard) +{ + const or_options_t *options = get_options(); + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + if (pb->use_attempts > pathbias_get_min_use(options)) { + /* Note: We rely on the < comparison here to allow us to set a 0 + * rate and disable the feature entirely. If refactoring, don't + * change to <= */ + if (pathbias_get_use_success_count(guard)/pb->use_attempts + < pathbias_get_extreme_use_rate(options)) { + /* Dropping is currently disabled by default. */ + if (pathbias_get_dropguards(options)) { + if (!pb->path_bias_disabled) { + log_warn(LD_CIRC, + "Your Guard %s is failing to carry an extremely large " + "amount of stream on its circuits. " + "To avoid potential route manipulation attacks, Tor has " + "disabled use of this guard. " + "Use counts are %ld/%ld. Success counts are %ld/%ld. " + "%ld circuits completed, %ld were unusable, %ld collapsed, " + "and %ld timed out. " + "For reference, your timeout cutoff is %ld seconds.", + entry_guard_describe(guard), + tor_lround(pathbias_get_use_success_count(guard)), + tor_lround(pb->use_attempts), + tor_lround(pathbias_get_close_success_count(guard)), + tor_lround(pb->circ_attempts), + tor_lround(pb->circ_successes), + tor_lround(pb->unusable_circuits), + tor_lround(pb->collapsed_circuits), + tor_lround(pb->timeouts), + tor_lround(get_circuit_build_close_time_ms()/1000)); + pb->path_bias_disabled = 1; + return; + } + } else if (!pb->path_bias_use_extreme) { + pb->path_bias_use_extreme = 1; + log_warn(LD_CIRC, + "Your Guard %s is failing to carry an extremely large " + "amount of streams on its circuits. " + "This could indicate a route manipulation attack, network " + "overload, bad local network connectivity, or a bug. " + "Use counts are %ld/%ld. Success counts are %ld/%ld. " + "%ld circuits completed, %ld were unusable, %ld collapsed, " + "and %ld timed out. " + "For reference, your timeout cutoff is %ld seconds.", + entry_guard_describe(guard), + tor_lround(pathbias_get_use_success_count(guard)), + tor_lround(pb->use_attempts), + tor_lround(pathbias_get_close_success_count(guard)), + tor_lround(pb->circ_attempts), + tor_lround(pb->circ_successes), + tor_lround(pb->unusable_circuits), + tor_lround(pb->collapsed_circuits), + tor_lround(pb->timeouts), + tor_lround(get_circuit_build_close_time_ms()/1000)); + } + } else if (pathbias_get_use_success_count(guard)/pb->use_attempts + < pathbias_get_notice_use_rate(options)) { + if (!pb->path_bias_use_noticed) { + pb->path_bias_use_noticed = 1; + log_notice(LD_CIRC, + "Your Guard %s is failing to carry more streams on its " + "circuits than usual. " + "Most likely this means the Tor network is overloaded " + "or your network connection is poor. " + "Use counts are %ld/%ld. Success counts are %ld/%ld. " + "%ld circuits completed, %ld were unusable, %ld collapsed, " + "and %ld timed out. " + "For reference, your timeout cutoff is %ld seconds.", + entry_guard_describe(guard), + tor_lround(pathbias_get_use_success_count(guard)), + tor_lround(pb->use_attempts), + tor_lround(pathbias_get_close_success_count(guard)), + tor_lround(pb->circ_attempts), + tor_lround(pb->circ_successes), + tor_lround(pb->unusable_circuits), + tor_lround(pb->collapsed_circuits), + tor_lround(pb->timeouts), + tor_lround(get_circuit_build_close_time_ms()/1000)); + } + } + } +} + +/** + * Check the path bias circuit close status rates against our consensus + * parameter limits. + * + * Emits a log message if the use success rates are too low. + * + * If pathbias_get_dropguards() is set, we also disable the use of + * very failure prone guards. + * + * XXX: This function shares similar log messages and checks to + * pathbias_measure_use_rate(). It may be possible to combine them + * eventually, especially if we can ever remove the need for 3 + * levels of closure warns (if the overall circuit failure rate + * goes down with ntor). One way to do so would be to multiply + * the build rate with the use rate to get an idea of the total + * fraction of the total network paths the user is able to use. + * See ticket #8159. + */ +static void +pathbias_measure_close_rate(entry_guard_t *guard) +{ + const or_options_t *options = get_options(); + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + if (pb->circ_attempts > pathbias_get_min_circs(options)) { + /* Note: We rely on the < comparison here to allow us to set a 0 + * rate and disable the feature entirely. If refactoring, don't + * change to <= */ + if (pathbias_get_close_success_count(guard)/pb->circ_attempts + < pathbias_get_extreme_rate(options)) { + /* Dropping is currently disabled by default. */ + if (pathbias_get_dropguards(options)) { + if (!pb->path_bias_disabled) { + log_warn(LD_CIRC, + "Your Guard %s is failing an extremely large " + "amount of circuits. " + "To avoid potential route manipulation attacks, Tor has " + "disabled use of this guard. " + "Success counts are %ld/%ld. Use counts are %ld/%ld. " + "%ld circuits completed, %ld were unusable, %ld collapsed, " + "and %ld timed out. " + "For reference, your timeout cutoff is %ld seconds.", + entry_guard_describe(guard), + tor_lround(pathbias_get_close_success_count(guard)), + tor_lround(pb->circ_attempts), + tor_lround(pathbias_get_use_success_count(guard)), + tor_lround(pb->use_attempts), + tor_lround(pb->circ_successes), + tor_lround(pb->unusable_circuits), + tor_lround(pb->collapsed_circuits), + tor_lround(pb->timeouts), + tor_lround(get_circuit_build_close_time_ms()/1000)); + pb->path_bias_disabled = 1; + return; + } + } else if (!pb->path_bias_extreme) { + pb->path_bias_extreme = 1; + log_warn(LD_CIRC, + "Your Guard %s is failing an extremely large " + "amount of circuits. " + "This could indicate a route manipulation attack, " + "extreme network overload, or a bug. " + "Success counts are %ld/%ld. Use counts are %ld/%ld. " + "%ld circuits completed, %ld were unusable, %ld collapsed, " + "and %ld timed out. " + "For reference, your timeout cutoff is %ld seconds.", + entry_guard_describe(guard), + tor_lround(pathbias_get_close_success_count(guard)), + tor_lround(pb->circ_attempts), + tor_lround(pathbias_get_use_success_count(guard)), + tor_lround(pb->use_attempts), + tor_lround(pb->circ_successes), + tor_lround(pb->unusable_circuits), + tor_lround(pb->collapsed_circuits), + tor_lround(pb->timeouts), + tor_lround(get_circuit_build_close_time_ms()/1000)); + } + } else if (pathbias_get_close_success_count(guard)/pb->circ_attempts + < pathbias_get_warn_rate(options)) { + if (!pb->path_bias_warned) { + pb->path_bias_warned = 1; + log_warn(LD_CIRC, + "Your Guard %s is failing a very large " + "amount of circuits. " + "Most likely this means the Tor network is " + "overloaded, but it could also mean an attack against " + "you or potentially the guard itself. " + "Success counts are %ld/%ld. Use counts are %ld/%ld. " + "%ld circuits completed, %ld were unusable, %ld collapsed, " + "and %ld timed out. " + "For reference, your timeout cutoff is %ld seconds.", + entry_guard_describe(guard), + tor_lround(pathbias_get_close_success_count(guard)), + tor_lround(pb->circ_attempts), + tor_lround(pathbias_get_use_success_count(guard)), + tor_lround(pb->use_attempts), + tor_lround(pb->circ_successes), + tor_lround(pb->unusable_circuits), + tor_lround(pb->collapsed_circuits), + tor_lround(pb->timeouts), + tor_lround(get_circuit_build_close_time_ms()/1000)); + } + } else if (pathbias_get_close_success_count(guard)/pb->circ_attempts + < pathbias_get_notice_rate(options)) { + if (!pb->path_bias_noticed) { + pb->path_bias_noticed = 1; + log_notice(LD_CIRC, + "Your Guard %s is failing more circuits than " + "usual. " + "Most likely this means the Tor network is overloaded. " + "Success counts are %ld/%ld. Use counts are %ld/%ld. " + "%ld circuits completed, %ld were unusable, %ld collapsed, " + "and %ld timed out. " + "For reference, your timeout cutoff is %ld seconds.", + entry_guard_describe(guard), + tor_lround(pathbias_get_close_success_count(guard)), + tor_lround(pb->circ_attempts), + tor_lround(pathbias_get_use_success_count(guard)), + tor_lround(pb->use_attempts), + tor_lround(pb->circ_successes), + tor_lround(pb->unusable_circuits), + tor_lround(pb->collapsed_circuits), + tor_lround(pb->timeouts), + tor_lround(get_circuit_build_close_time_ms()/1000)); + } + } + } +} + +/** + * This function scales the path bias use rates if we have + * more data than the scaling threshold. This allows us to + * be more sensitive to recent measurements. + * + * XXX: The attempt count transfer stuff here might be done + * better by keeping separate pending counters that get + * transferred at circuit close. See ticket #8160. + */ +static void +pathbias_scale_close_rates(entry_guard_t *guard) +{ + const or_options_t *options = get_options(); + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + /* If we get a ton of circuits, just scale everything down */ + if (pb->circ_attempts > pathbias_get_scale_threshold(options)) { + double scale_ratio = pathbias_get_scale_ratio(options); + int opened_attempts = pathbias_count_circs_in_states(guard, + PATH_STATE_BUILD_ATTEMPTED, PATH_STATE_BUILD_ATTEMPTED); + int opened_built = pathbias_count_circs_in_states(guard, + PATH_STATE_BUILD_SUCCEEDED, + PATH_STATE_USE_FAILED); + /* Verify that the counts are sane before and after scaling */ + int counts_are_sane = (pb->circ_attempts >= pb->circ_successes); + + pb->circ_attempts -= (opened_attempts+opened_built); + pb->circ_successes -= opened_built; + + pb->circ_attempts *= scale_ratio; + pb->circ_successes *= scale_ratio; + pb->timeouts *= scale_ratio; + pb->successful_circuits_closed *= scale_ratio; + pb->collapsed_circuits *= scale_ratio; + pb->unusable_circuits *= scale_ratio; + + pb->circ_attempts += (opened_attempts+opened_built); + pb->circ_successes += opened_built; + + entry_guards_changed(); + + log_info(LD_CIRC, + "Scaled pathbias counts to (%f,%f)/%f (%d/%d open) for guard " + "%s", + pb->circ_successes, pb->successful_circuits_closed, + pb->circ_attempts, opened_built, opened_attempts, + entry_guard_describe(guard)); + + /* Have the counts just become invalid by this scaling attempt? */ + if (counts_are_sane && pb->circ_attempts < pb->circ_successes) { + log_notice(LD_BUG, + "Scaling has mangled pathbias counts to %f/%f (%d/%d open) " + "for guard %s", + pb->circ_successes, pb->circ_attempts, opened_built, + opened_attempts, + entry_guard_describe(guard)); + } + } +} + +/** + * This function scales the path bias circuit close rates if we have + * more data than the scaling threshold. This allows us to be more + * sensitive to recent measurements. + * + * XXX: The attempt count transfer stuff here might be done + * better by keeping separate pending counters that get + * transferred at circuit close. See ticket #8160. + */ +void +pathbias_scale_use_rates(entry_guard_t *guard) +{ + const or_options_t *options = get_options(); + guard_pathbias_t *pb = entry_guard_get_pathbias_state(guard); + + /* If we get a ton of circuits, just scale everything down */ + if (pb->use_attempts > pathbias_get_scale_use_threshold(options)) { + double scale_ratio = pathbias_get_scale_ratio(options); + int opened_attempts = pathbias_count_circs_in_states(guard, + PATH_STATE_USE_ATTEMPTED, PATH_STATE_USE_SUCCEEDED); + /* Verify that the counts are sane before and after scaling */ + int counts_are_sane = (pb->use_attempts >= pb->use_successes); + + pb->use_attempts -= opened_attempts; + + pb->use_attempts *= scale_ratio; + pb->use_successes *= scale_ratio; + + pb->use_attempts += opened_attempts; + + log_info(LD_CIRC, + "Scaled pathbias use counts to %f/%f (%d open) for guard %s", + pb->use_successes, pb->use_attempts, opened_attempts, + entry_guard_describe(guard)); + + /* Have the counts just become invalid by this scaling attempt? */ + if (counts_are_sane && pb->use_attempts < pb->use_successes) { + log_notice(LD_BUG, + "Scaling has mangled pathbias usage counts to %f/%f " + "(%d open) for guard %s", + pb->circ_successes, pb->circ_attempts, + opened_attempts, entry_guard_describe(guard)); + } + + entry_guards_changed(); + } +} diff --cc src/feature/client/circpathbias.h index c99d1277b,000000000..9ce4a6b23 mode 100644,000000..100644 --- a/src/feature/client/circpathbias.h +++ b/src/feature/client/circpathbias.h @@@ -1,28 -1,0 +1,29 @@@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file circuitbuild.h + * \brief Header file for circuitbuild.c. + **/ + +#ifndef TOR_CIRCPATHBIAS_H +#define TOR_CIRCPATHBIAS_H + +double pathbias_get_extreme_rate(const or_options_t *options); +double pathbias_get_extreme_use_rate(const or_options_t *options); +int pathbias_get_dropguards(const or_options_t *options); +void pathbias_count_timeout(origin_circuit_t *circ); +void pathbias_count_build_success(origin_circuit_t *circ); +int pathbias_count_build_attempt(origin_circuit_t *circ); +int pathbias_check_close(origin_circuit_t *circ, int reason); +int pathbias_check_probe_response(circuit_t *circ, const cell_t *cell); ++void pathbias_count_valid_cells(circuit_t *circ, const cell_t *cell); +void pathbias_count_use_attempt(origin_circuit_t *circ); +void pathbias_mark_use_success(origin_circuit_t *circ); +void pathbias_mark_use_rollback(origin_circuit_t *circ); +const char *pathbias_state_to_string(enum path_state_t state); + +#endif /* !defined(TOR_CIRCPATHBIAS_H) */ diff --cc src/lib/container/smartlist.c index 4b29d834d,000000000..64cabfcc6 mode 100644,000000..100644 --- a/src/lib/container/smartlist.c +++ b/src/lib/container/smartlist.c @@@ -1,866 -1,0 +1,866 @@@ +/* Copyright (c) 2003-2004, Roger Dingledine + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file smartlist.c + * + * \brief Higher-level functions for the "smartlist" resizeable array + * abstraction. + * + * The functions declared here use higher-level functionality than those in + * smartlist_core.c, and handle things like smartlists of different types, + * sorting, searching, heap-structured smartlists, and other convenience + * functions. + **/ + +#include "lib/container/smartlist.h" +#include "lib/err/torerr.h" +#include "lib/malloc/malloc.h" +#include "lib/defs/digest_sizes.h" +#include "lib/ctime/di_ops.h" +#include "lib/string/compat_ctype.h" +#include "lib/string/compat_string.h" +#include "lib/string/util_string.h" +#include "lib/string/printf.h" + +#include "lib/log/util_bug.h" + +#include <stdlib.h> +#include <string.h> + +/** Append the string produced by tor_asprintf(<b>pattern</b>, <b>...</b>) + * to <b>sl</b>. */ +void +smartlist_add_asprintf(struct smartlist_t *sl, const char *pattern, ...) +{ + va_list ap; + va_start(ap, pattern); + smartlist_add_vasprintf(sl, pattern, ap); + va_end(ap); +} + +/** va_list-based backend of smartlist_add_asprintf. */ +void +smartlist_add_vasprintf(struct smartlist_t *sl, const char *pattern, + va_list args) +{ + char *str = NULL; + + tor_vasprintf(&str, pattern, args); + tor_assert(str != NULL); + + smartlist_add(sl, str); +} + +/** Reverse the order of the items in <b>sl</b>. */ +void +smartlist_reverse(smartlist_t *sl) +{ + int i, j; + void *tmp; + tor_assert(sl); + for (i = 0, j = sl->num_used-1; i < j; ++i, --j) { + tmp = sl->list[i]; + sl->list[i] = sl->list[j]; + sl->list[j] = tmp; + } +} + +/** If there are any strings in sl equal to element, remove and free them. + * Does not preserve order. */ +void +smartlist_string_remove(smartlist_t *sl, const char *element) +{ + int i; + tor_assert(sl); + tor_assert(element); + for (i = 0; i < sl->num_used; ++i) { + if (!strcmp(element, sl->list[i])) { + tor_free(sl->list[i]); + sl->list[i] = sl->list[--sl->num_used]; /* swap with the end */ + i--; /* so we process the new i'th element */ + sl->list[sl->num_used] = NULL; + } + } +} + +/** Return true iff <b>sl</b> has some element E such that + * !strcmp(E,<b>element</b>) + */ +int +smartlist_contains_string(const smartlist_t *sl, const char *element) +{ + int i; + if (!sl) return 0; + for (i=0; i < sl->num_used; i++) + if (strcmp((const char*)sl->list[i],element)==0) + return 1; + return 0; +} + +/** If <b>element</b> is equal to an element of <b>sl</b>, return that + * element's index. Otherwise, return -1. */ +int +smartlist_string_pos(const smartlist_t *sl, const char *element) +{ + int i; + if (!sl) return -1; + for (i=0; i < sl->num_used; i++) + if (strcmp((const char*)sl->list[i],element)==0) + return i; + return -1; +} + +/** If <b>element</b> is the same pointer as an element of <b>sl</b>, return + * that element's index. Otherwise, return -1. */ +int +smartlist_pos(const smartlist_t *sl, const void *element) +{ + int i; + if (!sl) return -1; + for (i=0; i < sl->num_used; i++) + if (element == sl->list[i]) + return i; + return -1; +} + +/** Return true iff <b>sl</b> has some element E such that + * !strcasecmp(E,<b>element</b>) + */ +int +smartlist_contains_string_case(const smartlist_t *sl, const char *element) +{ + int i; + if (!sl) return 0; + for (i=0; i < sl->num_used; i++) + if (strcasecmp((const char*)sl->list[i],element)==0) + return 1; + return 0; +} + +/** Return true iff <b>sl</b> has some element E such that E is equal + * to the decimal encoding of <b>num</b>. + */ +int +smartlist_contains_int_as_string(const smartlist_t *sl, int num) +{ + char buf[32]; /* long enough for 64-bit int, and then some. */ + tor_snprintf(buf,sizeof(buf),"%d", num); + return smartlist_contains_string(sl, buf); +} + +/** Return true iff the two lists contain the same strings in the same + * order, or if they are both NULL. */ +int +smartlist_strings_eq(const smartlist_t *sl1, const smartlist_t *sl2) +{ + if (sl1 == NULL) + return sl2 == NULL; + if (sl2 == NULL) + return 0; + if (smartlist_len(sl1) != smartlist_len(sl2)) + return 0; + SMARTLIST_FOREACH(sl1, const char *, cp1, { + const char *cp2 = smartlist_get(sl2, cp1_sl_idx); + if (strcmp(cp1, cp2)) + return 0; + }); + return 1; +} + +/** Return true iff the two lists contain the same int pointer values in + * the same order, or if they are both NULL. */ +int +smartlist_ints_eq(const smartlist_t *sl1, const smartlist_t *sl2) +{ + if (sl1 == NULL) + return sl2 == NULL; + if (sl2 == NULL) + return 0; + if (smartlist_len(sl1) != smartlist_len(sl2)) + return 0; + SMARTLIST_FOREACH(sl1, int *, cp1, { + int *cp2 = smartlist_get(sl2, cp1_sl_idx); + if (*cp1 != *cp2) + return 0; + }); + return 1; +} + +/** + * Return true if there is shallow equality between smartlists - + * i.e. all indices correspond to exactly same object (pointer + * values are matching). Otherwise, return false. + */ +int +smartlist_ptrs_eq(const smartlist_t *s1, const smartlist_t *s2) +{ + if (s1 == s2) + return 1; + + // Note: pointers cannot both be NULL at this point, because + // above check. + if (s1 == NULL || s2 == NULL) + return 0; + + if (smartlist_len(s1) != smartlist_len(s2)) + return 0; + + for (int i = 0; i < smartlist_len(s1); i++) { + if (smartlist_get(s1, i) != smartlist_get(s2, i)) + return 0; + } + + return 1; +} + +/** Return true iff <b>sl</b> has some element E such that + * tor_memeq(E,<b>element</b>,DIGEST_LEN) + */ +int +smartlist_contains_digest(const smartlist_t *sl, const char *element) +{ + int i; + if (!sl) return 0; + for (i=0; i < sl->num_used; i++) + if (tor_memeq((const char*)sl->list[i],element,DIGEST_LEN)) + return 1; + return 0; +} + +/** Return true iff some element E of sl2 has smartlist_contains(sl1,E). + */ +int +smartlist_overlap(const smartlist_t *sl1, const smartlist_t *sl2) +{ + int i; + for (i=0; i < sl2->num_used; i++) + if (smartlist_contains(sl1, sl2->list[i])) + return 1; + return 0; +} + +/** Remove every element E of sl1 such that !smartlist_contains(sl2,E). + * Does not preserve the order of sl1. + */ +void +smartlist_intersect(smartlist_t *sl1, const smartlist_t *sl2) +{ + int i; + for (i=0; i < sl1->num_used; i++) + if (!smartlist_contains(sl2, sl1->list[i])) { + sl1->list[i] = sl1->list[--sl1->num_used]; /* swap with the end */ + i--; /* so we process the new i'th element */ + sl1->list[sl1->num_used] = NULL; + } +} + +/** Remove every element E of sl1 such that smartlist_contains(sl2,E). + * Does not preserve the order of sl1. + */ +void +smartlist_subtract(smartlist_t *sl1, const smartlist_t *sl2) +{ + int i; + for (i=0; i < sl2->num_used; i++) + smartlist_remove(sl1, sl2->list[i]); +} + +/** Allocate and return a new string containing the concatenation of + * the elements of <b>sl</b>, in order, separated by <b>join</b>. If + * <b>terminate</b> is true, also terminate the string with <b>join</b>. + * If <b>len_out</b> is not NULL, set <b>len_out</b> to the length of + * the returned string. Requires that every element of <b>sl</b> is + * NUL-terminated string. + */ +char * +smartlist_join_strings(smartlist_t *sl, const char *join, + int terminate, size_t *len_out) +{ + return smartlist_join_strings2(sl,join,strlen(join),terminate,len_out); +} + +/** As smartlist_join_strings, but instead of separating/terminated with a + * NUL-terminated string <b>join</b>, uses the <b>join_len</b>-byte sequence + * at <b>join</b>. (Useful for generating a sequence of NUL-terminated + * strings.) + */ +char * +smartlist_join_strings2(smartlist_t *sl, const char *join, + size_t join_len, int terminate, size_t *len_out) +{ + int i; + size_t n = 0; + char *r = NULL, *dst, *src; + + tor_assert(sl); + tor_assert(join); + + if (terminate) + n = join_len; + + for (i = 0; i < sl->num_used; ++i) { + n += strlen(sl->list[i]); + if (i+1 < sl->num_used) /* avoid double-counting the last one */ + n += join_len; + } + dst = r = tor_malloc(n+1); + for (i = 0; i < sl->num_used; ) { + for (src = sl->list[i]; *src; ) + *dst++ = *src++; + if (++i < sl->num_used) { + memcpy(dst, join, join_len); + dst += join_len; + } + } + if (terminate) { + memcpy(dst, join, join_len); + dst += join_len; + } + *dst = '\0'; + + if (len_out) + *len_out = dst-r; + return r; +} + +/** Sort the members of <b>sl</b> into an order defined by + * the ordering function <b>compare</b>, which returns less then 0 if a + * precedes b, greater than 0 if b precedes a, and 0 if a 'equals' b. + */ +void +smartlist_sort(smartlist_t *sl, int (*compare)(const void **a, const void **b)) +{ + if (!sl->num_used) + return; + qsort(sl->list, sl->num_used, sizeof(void*), + (int (*)(const void *,const void*))compare); +} + +/** Given a smartlist <b>sl</b> sorted with the function <b>compare</b>, + * return the most frequent member in the list. Break ties in favor of + * later elements. If the list is empty, return NULL. If count_out is + * non-null, set it to the count of the most frequent member. + */ +void * +smartlist_get_most_frequent_(const smartlist_t *sl, + int (*compare)(const void **a, const void **b), + int *count_out) +{ + const void *most_frequent = NULL; + int most_frequent_count = 0; + + const void *cur = NULL; + int i, count=0; + + if (!sl->num_used) { + if (count_out) + *count_out = 0; + return NULL; + } + for (i = 0; i < sl->num_used; ++i) { + const void *item = sl->list[i]; + if (cur && 0 == compare(&cur, &item)) { + ++count; + } else { + if (cur && count >= most_frequent_count) { + most_frequent = cur; + most_frequent_count = count; + } + cur = item; + count = 1; + } + } + if (cur && count >= most_frequent_count) { + most_frequent = cur; + most_frequent_count = count; + } + if (count_out) + *count_out = most_frequent_count; + return (void*)most_frequent; +} + +/** Given a sorted smartlist <b>sl</b> and the comparison function used to + * sort it, remove all duplicate members. If free_fn is provided, calls + * free_fn on each duplicate. Otherwise, just removes them. Preserves order. + */ +void +smartlist_uniq(smartlist_t *sl, + int (*compare)(const void **a, const void **b), + void (*free_fn)(void *a)) +{ + int i; + for (i=1; i < sl->num_used; ++i) { + if (compare((const void **)&(sl->list[i-1]), + (const void **)&(sl->list[i])) == 0) { + if (free_fn) + free_fn(sl->list[i]); + smartlist_del_keeporder(sl, i--); + } + } +} + +/** Assuming the members of <b>sl</b> are in order, return a pointer to the + * member that matches <b>key</b>. Ordering and matching are defined by a + * <b>compare</b> function that returns 0 on a match; less than 0 if key is + * less than member, and greater than 0 if key is greater then member. + */ +void * - smartlist_bsearch(smartlist_t *sl, const void *key, ++smartlist_bsearch(const smartlist_t *sl, const void *key, + int (*compare)(const void *key, const void **member)) +{ + int found, idx; + idx = smartlist_bsearch_idx(sl, key, compare, &found); + return found ? smartlist_get(sl, idx) : NULL; +} + +/** Assuming the members of <b>sl</b> are in order, return the index of the + * member that matches <b>key</b>. If no member matches, return the index of + * the first member greater than <b>key</b>, or smartlist_len(sl) if no member + * is greater than <b>key</b>. Set <b>found_out</b> to true on a match, to + * false otherwise. Ordering and matching are defined by a <b>compare</b> + * function that returns 0 on a match; less than 0 if key is less than member, + * and greater than 0 if key is greater then member. + */ +int +smartlist_bsearch_idx(const smartlist_t *sl, const void *key, + int (*compare)(const void *key, const void **member), + int *found_out) +{ + int hi, lo, cmp, mid, len, diff; + + tor_assert(sl); + tor_assert(compare); + tor_assert(found_out); + + len = smartlist_len(sl); + + /* Check for the trivial case of a zero-length list */ + if (len == 0) { + *found_out = 0; + /* We already know smartlist_len(sl) is 0 in this case */ + return 0; + } + + /* Okay, we have a real search to do */ + tor_assert(len > 0); + lo = 0; + hi = len - 1; + + /* + * These invariants are always true: + * + * For all i such that 0 <= i < lo, sl[i] < key + * For all i such that hi < i <= len, sl[i] > key + */ + + while (lo <= hi) { + diff = hi - lo; + /* + * We want mid = (lo + hi) / 2, but that could lead to overflow, so + * instead diff = hi - lo (non-negative because of loop condition), and + * then hi = lo + diff, mid = (lo + lo + diff) / 2 = lo + (diff / 2). + */ + mid = lo + (diff / 2); + cmp = compare(key, (const void**) &(sl->list[mid])); + if (cmp == 0) { + /* sl[mid] == key; we found it */ + *found_out = 1; + return mid; + } else if (cmp > 0) { + /* + * key > sl[mid] and an index i such that sl[i] == key must + * have i > mid if it exists. + */ + + /* + * Since lo <= mid <= hi, hi can only decrease on each iteration (by + * being set to mid - 1) and hi is initially len - 1, mid < len should + * always hold, and this is not symmetric with the left end of list + * mid > 0 test below. A key greater than the right end of the list + * should eventually lead to lo == hi == mid == len - 1, and then + * we set lo to len below and fall out to the same exit we hit for + * a key in the middle of the list but not matching. Thus, we just + * assert for consistency here rather than handle a mid == len case. + */ + tor_assert(mid < len); + /* Move lo to the element immediately after sl[mid] */ + lo = mid + 1; + } else { + /* This should always be true in this case */ + tor_assert(cmp < 0); + + /* + * key < sl[mid] and an index i such that sl[i] == key must + * have i < mid if it exists. + */ + + if (mid > 0) { + /* Normal case, move hi to the element immediately before sl[mid] */ + hi = mid - 1; + } else { + /* These should always be true in this case */ + tor_assert(mid == lo); + tor_assert(mid == 0); + /* + * We were at the beginning of the list and concluded that every + * element e compares e > key. + */ + *found_out = 0; + return 0; + } + } + } + + /* + * lo > hi; we have no element matching key but we have elements falling + * on both sides of it. The lo index points to the first element > key. + */ + tor_assert(lo == hi + 1); /* All other cases should have been handled */ + tor_assert(lo >= 0); + tor_assert(lo <= len); + tor_assert(hi >= 0); + tor_assert(hi <= len); + + if (lo < len) { + cmp = compare(key, (const void **) &(sl->list[lo])); + tor_assert(cmp < 0); + } else { + cmp = compare(key, (const void **) &(sl->list[len-1])); + tor_assert(cmp > 0); + } + + *found_out = 0; + return lo; +} + +/** Helper: compare two const char **s. */ +static int +compare_string_ptrs_(const void **_a, const void **_b) +{ + return strcmp((const char*)*_a, (const char*)*_b); +} + +/** Sort a smartlist <b>sl</b> containing strings into lexically ascending + * order. */ +void +smartlist_sort_strings(smartlist_t *sl) +{ + smartlist_sort(sl, compare_string_ptrs_); +} + +/** Return the most frequent string in the sorted list <b>sl</b> */ +const char * +smartlist_get_most_frequent_string(smartlist_t *sl) +{ + return smartlist_get_most_frequent(sl, compare_string_ptrs_); +} + +/** Return the most frequent string in the sorted list <b>sl</b>. + * If <b>count_out</b> is provided, set <b>count_out</b> to the + * number of times that string appears. + */ +const char * +smartlist_get_most_frequent_string_(smartlist_t *sl, int *count_out) +{ + return smartlist_get_most_frequent_(sl, compare_string_ptrs_, count_out); +} + +/** Remove duplicate strings from a sorted list, and free them with tor_free(). + */ +void +smartlist_uniq_strings(smartlist_t *sl) +{ + smartlist_uniq(sl, compare_string_ptrs_, tor_free_); +} + +/** Helper: compare two pointers. */ +static int +compare_ptrs_(const void **_a, const void **_b) +{ + const void *a = *_a, *b = *_b; + if (a<b) + return -1; + else if (a==b) + return 0; + else + return 1; +} + +/** Sort <b>sl</b> in ascending order of the pointers it contains. */ +void +smartlist_sort_pointers(smartlist_t *sl) +{ + smartlist_sort(sl, compare_ptrs_); +} + +/* Heap-based priority queue implementation for O(lg N) insert and remove. + * Recall that the heap property is that, for every index I, h[I] < + * H[LEFT_CHILD[I]] and h[I] < H[RIGHT_CHILD[I]]. + * + * For us to remove items other than the topmost item, each item must store + * its own index within the heap. When calling the pqueue functions, tell + * them about the offset of the field that stores the index within the item. + * + * Example: + * + * typedef struct timer_t { + * struct timeval tv; + * int heap_index; + * } timer_t; + * + * static int compare(const void *p1, const void *p2) { + * const timer_t *t1 = p1, *t2 = p2; + * if (t1->tv.tv_sec < t2->tv.tv_sec) { + * return -1; + * } else if (t1->tv.tv_sec > t2->tv.tv_sec) { + * return 1; + * } else { + * return t1->tv.tv_usec - t2->tv_usec; + * } + * } + * + * void timer_heap_insert(smartlist_t *heap, timer_t *timer) { + * smartlist_pqueue_add(heap, compare, offsetof(timer_t, heap_index), + * timer); + * } + * + * void timer_heap_pop(smartlist_t *heap) { + * return smartlist_pqueue_pop(heap, compare, + * offsetof(timer_t, heap_index)); + * } + */ + +/** @{ */ +/** Functions to manipulate heap indices to find a node's parent and children. + * + * For a 1-indexed array, we would use LEFT_CHILD[x] = 2*x and RIGHT_CHILD[x] + * = 2*x + 1. But this is C, so we have to adjust a little. */ + +/* MAX_PARENT_IDX is the largest IDX in the smartlist which might have + * children whose indices fit inside an int. + * LEFT_CHILD(MAX_PARENT_IDX) == INT_MAX-2; + * RIGHT_CHILD(MAX_PARENT_IDX) == INT_MAX-1; + * LEFT_CHILD(MAX_PARENT_IDX + 1) == INT_MAX // impossible, see max list size. + */ +#define MAX_PARENT_IDX ((INT_MAX - 2) / 2) +/* If this is true, then i is small enough to potentially have children + * in the smartlist, and it is save to use LEFT_CHILD/RIGHT_CHILD on it. */ +#define IDX_MAY_HAVE_CHILDREN(i) ((i) <= MAX_PARENT_IDX) +#define LEFT_CHILD(i) ( 2*(i) + 1 ) +#define RIGHT_CHILD(i) ( 2*(i) + 2 ) +#define PARENT(i) ( ((i)-1) / 2 ) +/** }@ */ + +/** @{ */ +/** Helper macros for heaps: Given a local variable <b>idx_field_offset</b> + * set to the offset of an integer index within the heap element structure, + * IDX_OF_ITEM(p) gives you the index of p, and IDXP(p) gives you a pointer to + * where p's index is stored. Given additionally a local smartlist <b>sl</b>, + * UPDATE_IDX(i) sets the index of the element at <b>i</b> to the correct + * value (that is, to <b>i</b>). + */ +#define IDXP(p) ((int*)STRUCT_VAR_P(p, idx_field_offset)) + +#define UPDATE_IDX(i) do { \ + void *updated = sl->list[i]; \ + *IDXP(updated) = i; \ + } while (0) + +#define IDX_OF_ITEM(p) (*IDXP(p)) +/** @} */ + +/** Helper. <b>sl</b> may have at most one violation of the heap property: + * the item at <b>idx</b> may be greater than one or both of its children. + * Restore the heap property. */ +static inline void +smartlist_heapify(smartlist_t *sl, + int (*compare)(const void *a, const void *b), + int idx_field_offset, + int idx) +{ + while (1) { + if (! IDX_MAY_HAVE_CHILDREN(idx)) { + /* idx is so large that it cannot have any children, since doing so + * would mean the smartlist was over-capacity. Therefore it cannot + * violate the heap property by being greater than a child (since it + * doesn't have any). */ + return; + } + + int left_idx = LEFT_CHILD(idx); + int best_idx; + + if (left_idx >= sl->num_used) + return; + if (compare(sl->list[idx],sl->list[left_idx]) < 0) + best_idx = idx; + else + best_idx = left_idx; + if (left_idx+1 < sl->num_used && + compare(sl->list[left_idx+1],sl->list[best_idx]) < 0) + best_idx = left_idx + 1; + + if (best_idx == idx) { + return; + } else { + void *tmp = sl->list[idx]; + sl->list[idx] = sl->list[best_idx]; + sl->list[best_idx] = tmp; + UPDATE_IDX(idx); + UPDATE_IDX(best_idx); + + idx = best_idx; + } + } +} + +/** Insert <b>item</b> into the heap stored in <b>sl</b>, where order is + * determined by <b>compare</b> and the offset of the item in the heap is + * stored in an int-typed field at position <b>idx_field_offset</b> within + * item. + */ +void +smartlist_pqueue_add(smartlist_t *sl, + int (*compare)(const void *a, const void *b), + int idx_field_offset, + void *item) +{ + int idx; + smartlist_add(sl,item); + UPDATE_IDX(sl->num_used-1); + + for (idx = sl->num_used - 1; idx; ) { + int parent = PARENT(idx); + if (compare(sl->list[idx], sl->list[parent]) < 0) { + void *tmp = sl->list[parent]; + sl->list[parent] = sl->list[idx]; + sl->list[idx] = tmp; + UPDATE_IDX(parent); + UPDATE_IDX(idx); + idx = parent; + } else { + return; + } + } +} + +/** Remove and return the top-priority item from the heap stored in <b>sl</b>, + * where order is determined by <b>compare</b> and the item's position is + * stored at position <b>idx_field_offset</b> within the item. <b>sl</b> must + * not be empty. */ +void * +smartlist_pqueue_pop(smartlist_t *sl, + int (*compare)(const void *a, const void *b), + int idx_field_offset) +{ + void *top; + tor_assert(sl->num_used); + + top = sl->list[0]; + *IDXP(top)=-1; + if (--sl->num_used) { + sl->list[0] = sl->list[sl->num_used]; + sl->list[sl->num_used] = NULL; + UPDATE_IDX(0); + smartlist_heapify(sl, compare, idx_field_offset, 0); + } + sl->list[sl->num_used] = NULL; + return top; +} + +/** Remove the item <b>item</b> from the heap stored in <b>sl</b>, + * where order is determined by <b>compare</b> and the item's position is + * stored at position <b>idx_field_offset</b> within the item. <b>sl</b> must + * not be empty. */ +void +smartlist_pqueue_remove(smartlist_t *sl, + int (*compare)(const void *a, const void *b), + int idx_field_offset, + void *item) +{ + int idx = IDX_OF_ITEM(item); + tor_assert(idx >= 0); + tor_assert(sl->list[idx] == item); + --sl->num_used; + *IDXP(item) = -1; + if (idx == sl->num_used) { + sl->list[sl->num_used] = NULL; + return; + } else { + sl->list[idx] = sl->list[sl->num_used]; + sl->list[sl->num_used] = NULL; + UPDATE_IDX(idx); + smartlist_heapify(sl, compare, idx_field_offset, idx); + } +} + +/** Assert that the heap property is correctly maintained by the heap stored + * in <b>sl</b>, where order is determined by <b>compare</b>. */ +void +smartlist_pqueue_assert_ok(smartlist_t *sl, + int (*compare)(const void *a, const void *b), + int idx_field_offset) +{ + int i; + for (i = sl->num_used - 1; i >= 0; --i) { + if (i>0) + tor_assert(compare(sl->list[PARENT(i)], sl->list[i]) <= 0); + tor_assert(IDX_OF_ITEM(sl->list[i]) == i); + } +} + +/** Helper: compare two DIGEST_LEN digests. */ +static int +compare_digests_(const void **_a, const void **_b) +{ + return tor_memcmp((const char*)*_a, (const char*)*_b, DIGEST_LEN); +} + +/** Sort the list of DIGEST_LEN-byte digests into ascending order. */ +void +smartlist_sort_digests(smartlist_t *sl) +{ + smartlist_sort(sl, compare_digests_); +} + +/** Remove duplicate digests from a sorted list, and free them with tor_free(). + */ +void +smartlist_uniq_digests(smartlist_t *sl) +{ + smartlist_uniq(sl, compare_digests_, tor_free_); +} + +/** Helper: compare two DIGEST256_LEN digests. */ +static int +compare_digests256_(const void **_a, const void **_b) +{ + return tor_memcmp((const char*)*_a, (const char*)*_b, DIGEST256_LEN); +} + +/** Sort the list of DIGEST256_LEN-byte digests into ascending order. */ +void +smartlist_sort_digests256(smartlist_t *sl) +{ + smartlist_sort(sl, compare_digests256_); +} + +/** Return the most frequent member of the sorted list of DIGEST256_LEN + * digests in <b>sl</b> */ +const uint8_t * +smartlist_get_most_frequent_digest256(smartlist_t *sl) +{ + return smartlist_get_most_frequent(sl, compare_digests256_); +} + +/** Remove duplicate 256-bit digests from a sorted list, and free them with + * tor_free(). + */ +void +smartlist_uniq_digests256(smartlist_t *sl) +{ + smartlist_uniq(sl, compare_digests256_, tor_free_); +} diff --cc src/lib/container/smartlist.h index 9705396ac,000000000..0f5af3a92 mode 100644,000000..100644 --- a/src/lib/container/smartlist.h +++ b/src/lib/container/smartlist.h @@@ -1,168 -1,0 +1,168 @@@ +/* Copyright (c) 2003-2004, Roger Dingledine + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +#ifndef TOR_SMARTLIST_H +#define TOR_SMARTLIST_H + +/** + * \file smartlist.h + * + * \brief Header for smartlist.c + **/ + +#include <stdarg.h> + +#include "lib/smartlist_core/smartlist_core.h" +#include "lib/smartlist_core/smartlist_foreach.h" +#include "lib/smartlist_core/smartlist_split.h" + +void smartlist_add_asprintf(struct smartlist_t *sl, const char *pattern, ...) + CHECK_PRINTF(2, 3); +void smartlist_add_vasprintf(struct smartlist_t *sl, const char *pattern, + va_list args) + CHECK_PRINTF(2, 0); +void smartlist_reverse(smartlist_t *sl); +void smartlist_string_remove(smartlist_t *sl, const char *element); +int smartlist_contains_string(const smartlist_t *sl, const char *element); +int smartlist_pos(const smartlist_t *sl, const void *element); +int smartlist_string_pos(const smartlist_t *, const char *elt); +int smartlist_contains_string_case(const smartlist_t *sl, const char *element); +int smartlist_contains_int_as_string(const smartlist_t *sl, int num); +int smartlist_strings_eq(const smartlist_t *sl1, const smartlist_t *sl2); +int smartlist_contains_digest(const smartlist_t *sl, const char *element); +int smartlist_ints_eq(const smartlist_t *sl1, const smartlist_t *sl2); +int smartlist_overlap(const smartlist_t *sl1, const smartlist_t *sl2); +void smartlist_intersect(smartlist_t *sl1, const smartlist_t *sl2); +void smartlist_subtract(smartlist_t *sl1, const smartlist_t *sl2); + +int smartlist_ptrs_eq(const smartlist_t *s1, + const smartlist_t *s2); + +void smartlist_sort(smartlist_t *sl, + int (*compare)(const void **a, const void **b)); +void *smartlist_get_most_frequent_(const smartlist_t *sl, + int (*compare)(const void **a, const void **b), + int *count_out); +#define smartlist_get_most_frequent(sl, compare) \ + smartlist_get_most_frequent_((sl), (compare), NULL) +void smartlist_uniq(smartlist_t *sl, + int (*compare)(const void **a, const void **b), + void (*free_fn)(void *elt)); + +void smartlist_sort_strings(smartlist_t *sl); +void smartlist_sort_digests(smartlist_t *sl); +void smartlist_sort_digests256(smartlist_t *sl); +void smartlist_sort_pointers(smartlist_t *sl); + +const char *smartlist_get_most_frequent_string(smartlist_t *sl); +const char *smartlist_get_most_frequent_string_(smartlist_t *sl, + int *count_out); +const uint8_t *smartlist_get_most_frequent_digest256(smartlist_t *sl); + +void smartlist_uniq_strings(smartlist_t *sl); +void smartlist_uniq_digests(smartlist_t *sl); +void smartlist_uniq_digests256(smartlist_t *sl); - void *smartlist_bsearch(smartlist_t *sl, const void *key, ++void *smartlist_bsearch(const smartlist_t *sl, const void *key, + int (*compare)(const void *key, const void **member)); +int smartlist_bsearch_idx(const smartlist_t *sl, const void *key, + int (*compare)(const void *key, const void **member), + int *found_out); + +void smartlist_pqueue_add(smartlist_t *sl, + int (*compare)(const void *a, const void *b), + int idx_field_offset, + void *item); +void *smartlist_pqueue_pop(smartlist_t *sl, + int (*compare)(const void *a, const void *b), + int idx_field_offset); +void smartlist_pqueue_remove(smartlist_t *sl, + int (*compare)(const void *a, const void *b), + int idx_field_offset, + void *item); +void smartlist_pqueue_assert_ok(smartlist_t *sl, + int (*compare)(const void *a, const void *b), + int idx_field_offset); + +char *smartlist_join_strings(smartlist_t *sl, const char *join, int terminate, + size_t *len_out) ATTR_MALLOC; +char *smartlist_join_strings2(smartlist_t *sl, const char *join, + size_t join_len, int terminate, size_t *len_out) + ATTR_MALLOC; + +/* Helper: Given two lists of items, possibly of different types, such that + * both lists are sorted on some common field (as determined by a comparison + * expression <b>cmpexpr</b>), and such that one list (<b>sl1</b>) has no + * duplicates on the common field, loop through the lists in lockstep, and + * execute <b>unmatched_var2</b> on items in var2 that do not appear in + * var1. + * + * WARNING: It isn't safe to add remove elements from either list while the + * loop is in progress. + * + * Example use: + * SMARTLIST_FOREACH_JOIN(routerstatus_list, routerstatus_t *, rs, + * routerinfo_list, routerinfo_t *, ri, + * tor_memcmp(rs->identity_digest, ri->identity_digest, 20), + * log_info(LD_GENERAL,"No match for %s", ri->nickname)) { + * log_info(LD_GENERAL, "%s matches routerstatus %p", ri->nickname, rs); + * } SMARTLIST_FOREACH_JOIN_END(rs, ri); + **/ +/* The example above unpacks (approximately) to: + * int rs_sl_idx = 0, rs_sl_len = smartlist_len(routerstatus_list); + * int ri_sl_idx, ri_sl_len = smartlist_len(routerinfo_list); + * int rs_ri_cmp; + * routerstatus_t *rs; + * routerinfo_t *ri; + * for (; ri_sl_idx < ri_sl_len; ++ri_sl_idx) { + * ri = smartlist_get(routerinfo_list, ri_sl_idx); + * while (rs_sl_idx < rs_sl_len) { + * rs = smartlist_get(routerstatus_list, rs_sl_idx); + * rs_ri_cmp = tor_memcmp(rs->identity_digest, ri->identity_digest, 20); + * if (rs_ri_cmp > 0) { + * break; + * } else if (rs_ri_cmp == 0) { + * goto matched_ri; + * } else { + * ++rs_sl_idx; + * } + * } + * log_info(LD_GENERAL,"No match for %s", ri->nickname); + * continue; + * matched_ri: { + * log_info(LD_GENERAL,"%s matches with routerstatus %p",ri->nickname,rs); + * } + * } + */ +#define SMARTLIST_FOREACH_JOIN(sl1, type1, var1, sl2, type2, var2, \ + cmpexpr, unmatched_var2) \ + STMT_BEGIN \ + int var1 ## _sl_idx = 0, var1 ## _sl_len=(sl1)->num_used; \ + int var2 ## _sl_idx = 0, var2 ## _sl_len=(sl2)->num_used; \ + int var1 ## _ ## var2 ## _cmp; \ + type1 var1; \ + type2 var2; \ + for (; var2##_sl_idx < var2##_sl_len; ++var2##_sl_idx) { \ + var2 = (sl2)->list[var2##_sl_idx]; \ + while (var1##_sl_idx < var1##_sl_len) { \ + var1 = (sl1)->list[var1##_sl_idx]; \ + var1##_##var2##_cmp = (cmpexpr); \ + if (var1##_##var2##_cmp > 0) { \ + break; \ + } else if (var1##_##var2##_cmp == 0) { \ + goto matched_##var2; \ + } else { \ + ++var1##_sl_idx; \ + } \ + } \ + /* Ran out of v1, or no match for var2. */ \ + unmatched_var2; \ + continue; \ + matched_##var2: ; \ + +#define SMARTLIST_FOREACH_JOIN_END(var1, var2) \ + } \ + STMT_END + +#endif /* !defined(TOR_CONTAINER_H) */ diff --cc src/test/test_relaycell.c index eb30cab0e,3f84ee830..63820c996 --- a/src/test/test_relaycell.c +++ b/src/test/test_relaycell.c @@@ -5,22 -5,18 +5,25 @@@
#define RELAY_PRIVATE #define CIRCUITLIST_PRIVATE -#include "or.h" -#include "main.h" -#include "config.h" -#include "connection.h" -#include "crypto.h" -#include "crypto_rand.h" -#include "circuitbuild.h" -#include "circuitlist.h" -#include "connection_edge.h" -#include "log_test_helpers.h" -#include "relay.h" -#include "test.h" +#include "core/or/or.h" +#include "core/mainloop/main.h" +#include "app/config/config.h" +#include "core/mainloop/connection.h" +#include "lib/crypt_ops/crypto.h" ++#include "lib/crypt_ops/crypto_rand.h" +#include "core/or/circuitbuild.h" +#include "core/or/circuitlist.h" +#include "core/or/connection_edge.h" +#include "core/or/relay.h" +#include "test/test.h" ++#include "test/log_test_helpers.h" + +#include "core/or/cell_st.h" +#include "core/or/crypt_path_st.h" +#include "core/or/entry_connection_st.h" +#include "core/or/origin_circuit_st.h" +#include "core/or/socks_request_st.h" ++#include "core/or/half_edge_st.h"
static int srm_ncalls; static entry_connection_t *srm_conn;
tor-commits@lists.torproject.org