On Sun, Oct 23, 2011 at 1:26 PM, Roger Dingledine arma@mit.edu wrote:
[Quoting the original mail, but it's actually the file in git that I read and am commenting on.]
On Wed, Sep 21, 2011 at 02:13:18PM -0400, Nick Mathewson wrote:
The 'AllAddrs' option tells Tor that if no address is given in the PortDescription part, we should bind/advertise every one of our publicly visible unicast addresses; and that if a hostname address is given in the PortDescription, we should bind/advertise every publicly visible unicast address that the hostname resolves to. (Q: Should this be on by default?)
Yes, I think. And if it's on by default, does the option need to exist at all?
I actually don't think it should be on-by-default at this point. It'd change the behavior of ORPort 9001 in a possibly surprising way that I'm not sure we agree with. See below.
[...]
Example: We have a dynamic DNS provider that maps tornode.example.com to our current external IPv4 and IPv6 addresses. Our firewall forwards port 443 on those address to our port 1337.
ORPort 1337 no-advertise alladdrs ORPort tornode.example.com:443 no-bind alladdrs
This drives home the issue with alladdrs: what would we do if that flag isn't listed here?
The same as we currently do if you say "ORPort 1337": Try to figure out a single value for "our address," and advertise that address with that port, and not open or advertise any other ports.
(Q: Any reason to allow more than 2? Multiple interfaces, I guess.)
By the same logic that we chose not to allow bitmasks in addresses, it's easy to argue that we shouldn't list more than 2 addresses (one ipv4, one ipv6). But does limiting it to 2 in the spec simplify the design in any way, or just constrain us down the road?
Hm. In practice, keeping track of an IPv4 address and an IPv6 address seems to be enough for now. I'm going to suggest that we remove port ranges from the proposal too, and just have lists of ports, and start by having servers only list up to two addresses with one port each, but have clients able to parse more than that, in case we someday come up with a better approach.
An authority shouldn't list a node as Running unless every or-address line it advertises looks like it will work.
This part makes me sad -- I worry that we'll end up with situations where most addresses work but we discard the whole relay because of a network hiccup somewhere (e.g. between the directory authority and one of the relay's addresses). How much more would it complexify things if we list the ones we think are up in the consensus, and then the voting process decides which ones get advertised?
Well, it would complexify things somewhat, but I'm not so sure it's avoidable: It seems like it'll be pretty common to autoconfig a working ipv4 address but not a working ipv6 address, or vice versa.
But on the other hand, self-testing is *supposed* to prevent that.
Consensus directories and microdescriptors:
We introduce a new line type for microdescriptors and consensuses, "a". Each "a" line has the same format as an or-address line. The "a" lines (if any) appear immediately after the "r" line for a router in the consensus, and immediately after the "onion-key" entry in a microdescriptor.
We should clarify which flavors we mean when we say consensuses. That is, are we going to add "a" lines to the microdescriptor-flavor consensus too, even though clients will soon find the same lines when they fetch the microdescriptors? I think yes, on the theory that clients will find the addresses useful in fetching microdescriptors. But it feels like a shame to include this mostly static info in every consensus when it's only really helpful for initial bootstrap for a tiny subset of users.
I think so; without them, ipv6-only clients just can't work.
I could imagine an ipv6-micro flavored consensus which includes ipv6 addresses for the clients who need that, and then those clients fetch the normal consensus after that. But maybe I'm trying to optimize too much for bandwidth.
Adding new flavors can totally happen after the first implementation here.
We will have to define a new voting algorithm version; when using this version or later, votes should include a single "a" line for every relay that has an IPv6 address, to include the first IPv6 line in its descriptor. (If there are no or-address lines, then
You meant "If there are no IPv6 or-address lines", yes?
yes.
As with other data in the vote derived from the descriptor, the vote will include whichever set of "a" lines are given by the most authorities who voted for the descriptor digest that will be used for the router.
Just to clarify, we treat the whole set of "a" lines for the router as an atomic blob, and vote for the blob that is most common? We could also do something more fine-grained, but I don't think we want to.
Hmmm. I actually think being more fine-grained is right if we want to be able to vote on individual addresses working or not.
[...]
Client behavior:
I propose that initially we shouldn't change client behavior too much here.
(Q: Is there any advantage to having a client choose a random address? If so we can do it later. If not, why list any more than one IPv4 and one IPv6 address?)
Tor clients not running with bridges, and running with IPv4 support, should still use the address and ORPort as advertised in the router or r line of the appropriate directory object.
Tor clients not running with bridges, and running without IPv4 support, should use the first listed IPv6 address for a node, using the lowest-numbered listed port for that address. They should only connect to nodes with an IPv6 address.
A) What's the recommended way for the Tor client to discover that it doesn't have ipv4 support?
Not having a public unicast IPv4 address; having all attempts to connect to IPv4 addresses fail pretty early on.
B) What if the client supports ipv4 and ipv6 yet the public ipv4 relays are blocked? We need some way for the user to explicitly ask for ipv6, and ideally some way to auto detect that Tor should try the other.
Good idea.
[...]
We can make this work, though: let's allow nodes to list themselves with a magic IPv4 address (say, 127.1.1.1) if they have or-address entries containing only IPv6 address. We could give these nodes a new flag other than Running to indicate that they're up, and not give them the Running flag. That way, old clients would never try to use them, but new clients could know to treat the new flag as indicating that the node is running, and know not to connect to a node listed with address 127.1.1.1.
It would be nice to come up with a color for the fence that doesn't involve forever maintaining a separate Running6 flag and forever including a hack in the consensus. For example, what if we have an "r6" line that is like an "r" line except its address is v6? I think we can't do that because the "w", "p", etc lines in the r6 stanza would be interpreted by old clients as part of the previous router's stanza. Does that realization mean we should declare one of the lines in the router stanza "at end, exactly once" in the spec, so in the future we can add new types of stanzas? Does it mean we can't add a "w" line to the directory footer stanza because it would confuse clients who don't know there's a directory footer stanza? Whee.
I think that your long series of questions here basically comes out to "yeah". Did you have an idea on this point that _would_ work out? I guess we could do a new consensus flavor, but minimizing those is probably smart.