[tor-commits] [tor/master] Add rend failure cache

nickm at torproject.org nickm at torproject.org
Tue Aug 11 13:36:11 UTC 2015


commit ab9a0e340728abd96128da726f67b4ccca10ba52
Author: David Goulet <dgoulet at ev0ke.net>
Date:   Thu Jun 18 16:09:18 2015 -0400

    Add rend failure cache
    
    When fetching a descriptor, we know test every introduction points in it
    against our rend failure cache to know if we keep it or not. For this to
    work, now everytime an introduction points is discareded (ex: receiving a
    NACK), we note it down in our introduction cache.
    
    See rendcache.c for a detailed explanation of the cache's behavior.
    
    Fixes #16389
    
    Signed-off-by: David Goulet <dgoulet at ev0ke.net>
---
 src/or/rendcache.c  |  259 ++++++++++++++++++++++++++++++++++++++++++++++++++-
 src/or/rendcache.h  |   16 ++++
 src/or/rendclient.c |    6 ++
 3 files changed, 278 insertions(+), 3 deletions(-)

diff --git a/src/or/rendcache.c b/src/or/rendcache.c
index be94ef4..7a86546 100644
--- a/src/or/rendcache.c
+++ b/src/or/rendcache.c
@@ -9,7 +9,6 @@
 #include "rendcache.h"
 
 #include "config.h"
-#include "rendcommon.h"
 #include "rephist.h"
 #include "routerlist.h"
 #include "routerparse.h"
@@ -22,6 +21,33 @@ static strmap_t *rend_cache = NULL;
  * directories. */
 static digestmap_t *rend_cache_v2_dir = NULL;
 
+/** (Client side only) Map from service id to rend_cache_failure_t. This
+ * cache is used to track intro point(IP) failures so we know when to keep
+ * or discard a new descriptor we just fetched. Here is a description of the
+ * cache behavior.
+ *
+ * Everytime tor discards an IP (ex: receives a NACK), we add an entry to
+ * this cache noting the identity digest of the IP and it's failure type for
+ * the service ID. The reason we indexed this cache by service ID is to
+ * differentiate errors that can occur only for a specific service like a
+ * NACK for instance. It applies for one but maybe not for the others.
+ *
+ * Once a service descriptor is fetched and considered valid, each IP is
+ * looked up in this cache and if present, it is discarded from the fetched
+ * descriptor. At the end, all IP(s) in the cache, for a specific service
+ * ID, that were NOT present in the descriptor are removed from this cache.
+ * Which means that if at least one IP was not in this cache, thus usuable,
+ * it's considered a new descriptor so we keep it. Else, if all IPs were in
+ * this cache, we discard the descriptor as it's considered unsuable.
+ *
+ * Once a descriptor is removed from the rend cache or expires, the entry
+ * in this cache is also removed for the service ID.
+ *
+ * This scheme allows us to not realy on the descriptor's timestamp (which
+ * is rounded down to the hour) to know if we have a newer descriptor. We
+ * only rely on the usability of intro points from an internal state. */
+static strmap_t *rend_cache_failure = NULL;
+
 /** DOCDOC */
 static size_t rend_cache_total_allocation = 0;
 
@@ -32,6 +58,7 @@ rend_cache_init(void)
 {
   rend_cache = strmap_new();
   rend_cache_v2_dir = digestmap_new();
+  rend_cache_failure = strmap_new();
 }
 
 /** Return the approximate number of bytes needed to hold <b>e</b>. */
@@ -85,6 +112,82 @@ rend_cache_increment_allocation(size_t n)
   }
 }
 
+/** Helper: free a rend cache failure intro object. */
+static void
+rend_cache_failure_intro_entry_free(rend_cache_failure_intro_t *entry)
+{
+  if (entry == NULL) {
+    return;
+  }
+  tor_free(entry);
+}
+
+/** Allocate a rend cache failure intro object and return it. <b>failure</b>
+ * is set into the object. This function can not fail. */
+static rend_cache_failure_intro_t *
+rend_cache_failure_intro_entry_new(unsigned int failure)
+{
+  rend_cache_failure_intro_t *entry = tor_malloc(sizeof(*entry));
+  entry->failure_type = failure;
+  return entry;
+}
+
+/** Helper: free a rend cache failure object. */
+static void
+rend_cache_failure_entry_free(rend_cache_failure_t *entry)
+{
+  if (entry == NULL) {
+    return;
+  }
+
+  /* Free and remove every intro failure object. */
+  DIGESTMAP_FOREACH_MODIFY(entry->intro_failures, key,
+                           rend_cache_failure_intro_t *, e) {
+    rend_cache_failure_intro_entry_free(e);
+    MAP_DEL_CURRENT(key);
+  } DIGESTMAP_FOREACH_END;
+  tor_free(entry);
+}
+
+/** Helper: deallocate a rend_cache_failure_t. (Used with strmap_free(),
+ * which requires a function pointer whose argument is void*). */
+static void
+rend_cache_failure_entry_free_(void *entry)
+{
+  rend_cache_failure_entry_free(entry);
+}
+
+/** Allocate a rend cache failure object and return it. This function can
+ * not fail. */
+static rend_cache_failure_t *
+rend_cache_failure_entry_new(void)
+{
+  rend_cache_failure_t *entry = tor_malloc(sizeof(*entry));
+  entry->intro_failures = digestmap_new();
+  return entry;
+}
+
+/** Remove failure cache entry for the service ID in the given descriptor
+ * <b>desc</b>. */
+static void
+rend_cache_failure_remove(rend_service_descriptor_t *desc)
+{
+  char service_id[REND_SERVICE_ID_LEN_BASE32 + 1];
+  rend_cache_failure_t *entry;
+
+  if (desc == NULL) {
+    return;
+  }
+  if (rend_get_service_id(desc->pk, service_id) < 0) {
+    return;
+  }
+  entry = strmap_get_lc(rend_cache_failure, service_id);
+  if (entry != NULL) {
+    strmap_remove_lc(rend_cache_failure, service_id);
+    rend_cache_failure_entry_free(entry);
+  }
+}
+
 /** Helper: free storage held by a single service descriptor cache entry. */
 static void
 rend_cache_entry_free(rend_cache_entry_t *e)
@@ -92,6 +195,9 @@ rend_cache_entry_free(rend_cache_entry_t *e)
   if (!e)
     return;
   rend_cache_decrement_allocation(rend_cache_entry_allocation(e));
+  /* We are about to remove a descriptor from the cache so remove the entry
+   * in the failure cache. */
+  rend_cache_failure_remove(e->parsed);
   rend_service_descriptor_free(e->parsed);
   tor_free(e->desc);
   tor_free(e);
@@ -111,8 +217,10 @@ rend_cache_free_all(void)
 {
   strmap_free(rend_cache, rend_cache_entry_free_);
   digestmap_free(rend_cache_v2_dir, rend_cache_entry_free_);
+  strmap_free(rend_cache_failure, rend_cache_failure_entry_free_);
   rend_cache = NULL;
   rend_cache_v2_dir = NULL;
+  rend_cache_failure = NULL;
   rend_cache_total_allocation = 0;
 }
 
@@ -150,6 +258,127 @@ rend_cache_purge(void)
   rend_cache = strmap_new();
 }
 
+/** Lookup the rend failure cache using a relay identity digest in
+ * <b>identity</b> and service ID <b>service_id</b>. If found, the intro
+ * failure is set in <b>intro_entry</b> else it stays untouched. Return 1
+ * iff found else 0. */
+static int
+cache_failure_intro_lookup(const uint8_t *identity, const char *service_id,
+                           rend_cache_failure_intro_t **intro_entry)
+{
+  rend_cache_failure_t *elem;
+  rend_cache_failure_intro_t *intro_elem;
+
+  tor_assert(rend_cache_failure);
+
+  if (intro_entry) {
+    *intro_entry = NULL;
+  }
+
+  /* Lookup descriptor and return it. */
+  elem = strmap_get_lc(rend_cache_failure, service_id);
+  if (elem == NULL) {
+    goto not_found;
+  }
+  intro_elem = digestmap_get(elem->intro_failures, (char *) identity);
+  if (intro_elem == NULL) {
+    goto not_found;
+  }
+  if (intro_entry) {
+    *intro_entry = intro_elem;
+  }
+  return 1;
+not_found:
+  return 0;
+}
+
+/** Add an intro point failure to the failure cache using the relay
+ * <b>identity</b> and service ID <b>service_id</b>. Record the
+ * <b>failure</b> in that object. */
+static void
+cache_failure_intro_add(const uint8_t *identity, const char *service_id,
+                        unsigned int failure)
+{
+  rend_cache_failure_t *fail_entry;
+  rend_cache_failure_intro_t *entry;
+
+  /* Make sure we have a failure object for this service ID and if not,
+   * create it with this new intro failure entry. */
+  fail_entry = strmap_get_lc(rend_cache_failure, service_id);
+  if (fail_entry == NULL) {
+    fail_entry = rend_cache_failure_entry_new();
+    /* Add failure entry to global rend failure cache. */
+    strmap_set_lc(rend_cache_failure, service_id, fail_entry);
+  }
+  entry = rend_cache_failure_intro_entry_new(failure);
+  digestmap_set(fail_entry->intro_failures, (char *) identity, entry);
+}
+
+/** Using a parsed descriptor <b>desc</b>, check if the introduction points
+ * are present in the failure cache and if so they are removed from the
+ * descriptor and kept into the failure cache. Then, each intro points that
+ * are NOT in the descriptor but in the failure cache for the given
+ * <b>service_id</b> are removed from the failure cache. */
+static void
+validate_intro_point_failure(const rend_service_descriptor_t *desc,
+                             const char *service_id)
+{
+  rend_cache_failure_t *new_entry, *cur_entry;
+  /* New entry for the service ID that will be replacing the one in the
+   * failure cache since we have a new descriptor. In the case where all
+   * intro points are removed, we are assured that the new entry is the same
+   * as the current one. */
+  new_entry = tor_malloc(sizeof(*new_entry));
+  new_entry->intro_failures = digestmap_new();
+
+  tor_assert(desc);
+
+  SMARTLIST_FOREACH_BEGIN(desc->intro_nodes, rend_intro_point_t *, intro) {
+    int found;
+    rend_cache_failure_intro_t *entry;
+    const uint8_t *identity =
+      (uint8_t *) intro->extend_info->identity_digest;
+
+    found = cache_failure_intro_lookup(identity, service_id, &entry);
+    if (found) {
+      /* This intro point is in our cache, discard it from the descriptor
+       * because chances are that it's unusable. */
+      SMARTLIST_DEL_CURRENT(desc->intro_nodes, intro);
+      rend_intro_point_free(intro);
+      /* Keep it for our new entry. */
+      digestmap_set(new_entry->intro_failures, (char *) identity, entry);
+      continue;
+    }
+  } SMARTLIST_FOREACH_END(intro);
+
+  /* Swap the failure entry in the cache and free the current one. */
+  cur_entry = strmap_get_lc(rend_cache_failure, service_id);
+  if (cur_entry != NULL) {
+    rend_cache_failure_entry_free(cur_entry);
+  }
+  strmap_set_lc(rend_cache_failure, service_id, new_entry);
+}
+
+/** Note down an intro failure in the rend failure cache using the type of
+ * failure in <b>failure</b> for the relay identity digest in
+ * <b>identity</b> and service ID <b>service_id</b>. If an entry already
+ * exists in the cache, the failure type is changed with <b>failure</b>. */
+void
+rend_cache_intro_failure_note(unsigned int failure, const uint8_t *identity,
+                              const char *service_id)
+{
+  int found;
+  rend_cache_failure_intro_t *entry;
+
+  found = cache_failure_intro_lookup(identity, service_id, &entry);
+  if (!found) {
+    cache_failure_intro_add(identity, service_id, failure);
+  } else {
+    /* Replace introduction point failure with this one. */
+    entry->failure_type = failure;
+  }
+}
+
 /** Remove all old v2 descriptors and those for which this hidden service
  * directory is not responsible for any more.
  *
@@ -537,20 +766,44 @@ rend_cache_store_v2_desc_as_client(const char *desc,
              "the future.", safe_str_client(service_id));
     goto err;
   }
-  /* Do we already have a newer descriptor? */
+  /* Do we have the same exact copy already in our cache? */
   tor_snprintf(key, sizeof(key), "2%s", service_id);
   e = (rend_cache_entry_t*) strmap_get_lc(rend_cache, key);
-  if (e && e->parsed->timestamp >= parsed->timestamp) {
+  if (e && !strcmp(desc, e->desc)) {
+    log_info(LD_REND,"We already have this service descriptor %s.",
+             safe_str_client(service_id));
+    goto okay;
+  }
+  /* Verify that we are not replacing an older descriptor. It's important to
+   * avoid an evil HSDir serving old descriptor. We validate if the
+   * timestamp is greater than and not equal because it's a rounded down
+   * timestamp to the hour so if the descriptor changed in the same hour,
+   * the rend cache failure will tells us if we have a new descriptor. */
+  if (e && e->parsed->timestamp > parsed->timestamp) {
     log_info(LD_REND, "We already have a new enough service descriptor for "
              "service ID %s with the same desc ID and version.",
              safe_str_client(service_id));
     goto okay;
   }
+  /* Lookup our failure cache for intro point that might be unsuable. */
+  validate_intro_point_failure(parsed, service_id);
+  /* It's now possible that our intro point list is empty, this means that
+   * this descriptor is useless to us because intro points have all failed
+   * somehow before. Discard the descriptor. */
+  if (smartlist_len(parsed->intro_nodes) == 0) {
+    log_info(LD_REND, "Service descriptor with service ID %s, every "
+             "intro points are unusable. Discarding it.",
+             safe_str_client(service_id));
+    goto err;
+  }
+  /* Now either purge the current one and replace it's content or create a
+   * new one and add it to the rend cache. */
   if (!e) {
     e = tor_malloc_zero(sizeof(rend_cache_entry_t));
     strmap_set_lc(rend_cache, key, e);
   } else {
     rend_cache_decrement_allocation(rend_cache_entry_allocation(e));
+    rend_cache_failure_remove(e->parsed);
     rend_service_descriptor_free(e->parsed);
     tor_free(e->desc);
   }
diff --git a/src/or/rendcache.h b/src/or/rendcache.h
index f61f02a..21be9f9 100644
--- a/src/or/rendcache.h
+++ b/src/or/rendcache.h
@@ -10,6 +10,7 @@
 #define TOR_RENDCACHE_H
 
 #include "or.h"
+#include "rendcommon.h"
 
 /** How old do we let hidden service descriptors get before discarding
  * them as too old? */
@@ -31,6 +32,17 @@ typedef struct rend_cache_entry_t {
   rend_service_descriptor_t *parsed; /**< Parsed value of 'desc' */
 } rend_cache_entry_t;
 
+/* Introduction point failure type. */
+typedef struct rend_cache_failure_intro_t {
+  unsigned int failure_type;
+} rend_cache_failure_intro_t;
+
+/** Cache failure object indexed by service ID. */
+typedef struct rend_cache_failure_t {
+  /* Contains rend_cache_failure_intro_t indexed by identity digest. */
+  digestmap_t *intro_failures;
+} rend_cache_failure_t;
+
 void rend_cache_init(void);
 void rend_cache_clean(time_t now);
 void rend_cache_clean_v2_descs_as_dir(time_t now, size_t min_to_remove);
@@ -53,5 +65,9 @@ rend_cache_store_status_t rend_cache_store_v2_desc_as_client(const char *desc,
                                                 rend_cache_entry_t **entry);
 size_t rend_cache_get_total_allocation(void);
 
+void rend_cache_intro_failure_note(unsigned int failure,
+                                   const uint8_t *identity,
+                                   const char *service_id);
+
 #endif /* TOR_RENDCACHE_H */
 
diff --git a/src/or/rendclient.c b/src/or/rendclient.c
index 59e938e..0b22f46 100644
--- a/src/or/rendclient.c
+++ b/src/or/rendclient.c
@@ -1019,6 +1019,9 @@ rend_client_report_intro_point_failure(extend_info_t *failed_intro,
         tor_fragile_assert();
         /* fall through */
       case INTRO_POINT_FAILURE_GENERIC:
+        rend_cache_intro_failure_note(failure_type,
+                                      (uint8_t *) failed_intro->identity_digest,
+                                      rend_query->onion_address);
         rend_intro_point_free(intro);
         smartlist_del(ent->parsed->intro_nodes, i);
         break;
@@ -1034,6 +1037,9 @@ rend_client_report_intro_point_failure(extend_info_t *failed_intro,
                    intro->unreachable_count,
                    zap_intro_point ? " Removing from descriptor.": "");
           if (zap_intro_point) {
+            rend_cache_intro_failure_note(failure_type,
+                                          (uint8_t *) failed_intro->identity_digest,
+                                          rend_query->onion_address);
             rend_intro_point_free(intro);
             smartlist_del(ent->parsed->intro_nodes, i);
           }





More information about the tor-commits mailing list