You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Named networks: an optional public network selector, so one CPace attempt tests one network #201
For the future. This is option D of the name-guessing discussion; options A–C (rate limits, name strength, ban) go into 5.0.0 on the mesh branch.
The problem
A client with --network NAME gives the server nothing but the name, and only inside CPace. So PAKE2 carries one answer for every network the server lists (at most 24). A single PAKE1 therefore tests one guessed name against all of them: a server with 24 networks gives a guesser 24 chances per attempt.
This is inherent to joining by the name alone: the server has to try the name against every network it has. The 5.0.0 limits count attempts, not networks, so they do not remove this factor.
Proposal
An optional, non-secret selector per network that the client sends with PAKE1, so the server answers for that one network only:
server: selector = <word> in [network NAME]. It is public, like a username, and is shown in --status;
client: --network <selector>:<name> (or --network-selector), and s= in the connect link;
with a selector, PAKE2 has one answer, or none for an unknown selector (an answer for a decoy network, so the reply does not reveal which selectors exist);
without a selector, the behaviour stays as now.
The selector must not be derived from the name: anything computed from the name would let a guesser test names offline against it.
Costs
One more thing to configure and to type on a device, which works against camex --server-host H --network NAME being the whole setup.
A protocol change, though a compatible one: an old server ignores the selector field if it lives in PAKE1's padding.
Worth it for open servers or servers with many networks. For the default install.sh setup (one network, a random 20-character name) it adds nothing.
@dimerr, a question on this one before it waits for later: do you want a network selector at all?
How it fits your use of named networks matters here:
How many networks does a server of yours hold? With one or a few, a guess tests one or a few names at a time, and the 5.0.0 limits (cpace_*, the 16-character minimum) cover it. With dozens, each PAKE1 tests a guess against up to 24 of them, and a selector brings that back to one.
Would you type a selector? It makes the client side --network <selector>:<name> (and s= in the connect link) instead of just --network NAME.
Should it wait? It is a compatible protocol change (a field in PAKE1's padding, old servers ignore it), so it does not have to go into 5.0.0. Or would you rather have it in the same release as the rest of named networks?
If none of this matches how you run servers, I'll keep the issue as deferred and close it if nobody asks for it.
Activity
compot-dev commentedon Nov 21, 2026
@dimerr, a question on this one before it waits for later: do you want a network selector at all?
How it fits your use of named networks matters here:
cpace_*, the 16-character minimum) cover it. With dozens, eachPAKE1tests a guess against up to 24 of them, and a selector brings that back to one.--network <selector>:<name>(ands=in the connect link) instead of just--network NAME.PAKE1's padding, old servers ignore it), so it does not have to go into 5.0.0. Or would you rather have it in the same release as the rest of named networks?If none of this matches how you run servers, I'll keep the issue as
deferredand close it if nobody asks for it.compot-dev commentedon Nov 21, 2026
@dimerr, a second ping: the question above is still open.
A yes or no is enough: do you want a network selector at all, and if so, before 5.0.0 or after it?
If there is no answer by 2026-11-23 18:00 UTC, the selector waits until after 5.0.0 and we decide its shape ourselves.