Skip to content

Latest commit

 

History

History
105 lines (91 loc) · 4.58 KB

TODO.md

File metadata and controls

105 lines (91 loc) · 4.58 KB
Raw

TODO

Packaging

  • Better pre/post inst/rm scripts for systemd service
    • Upgrading wireguard often leaves wirelink inactive
  • ChangeLog generation
  • Auto-tag and release from CI

Systemd Integration

  • sd_notify

Functionality

  • Synchronize activation of AIPs with peer
    • Send a special fact when adding AIPs, which triggers the remote end to process queued facts immediately, which should trigger that peer to also add AIPs if it didn't already.
  • Intelligent prioritization of peer EPs to try
    • EPs that are on a local subnet, then EPs on the internet, then everything else?
    • Mechanism? LRU penalty / bonus? I.e. add N ms to actual LU stamp for internet EPs, and M>n for non-matching LAN IPs?
  • Router detection: Inspect the AllowedIP facts other peers send about ourselves (need to change broadcastFacts so they send those to us)
  • Adjust fact sending and TTL behavior so that alive facts reliably expire before other facts, avoiding inconsistent behavior around trust and link resets.
  • Find better ways to interact with "basic" devices, e.g. Android clients.
    • The big problem here is how to have the basic client route traffic via a router until it has the direct peer connection working
    • ... assuming that there's a way to transfer all the core peer configs to the basic peer (the public keys and AIPs)
  • Symmetric NAT handling
    • Try to detect it by looking for multiple endpoint facts with the same or nearby IP but varying ports
    • If ports on the same IP are nearby, do some guessing? This seems unlikely to work, and may make a wreck of the translation table on the sender side.
  • Easy config generators
    • Generate a wirelink config from a (subset) of a live wireguard interface config
    • Generate a minimal wirelink config interactively, by prompting for the trusted peer's public key and endpoint
    • Generate / export a static config from all currently known facts, or merge known facts with a static config to make an expanded config
    • Generate a signed configuration (JWS?) from a trusted peer, and import it on a leaf with signature verification, so configs can be shared over untrusted media (e.g. HTTP)
    • Generate an importable config for "basic" devices from current state or config (see above on generating config from state)
  • CLI
    • Show wg state annotated with facts (peer names, alternate endpoints, trust levels)
    • Send commands to daemon, e.g. state queries, refresh boot id, reload config, manually add facts/settings
    • Adjust debug logging on the fly
    • Monitor debug data without logging
  • Android
    • Make a build of the wireguard android app with wirelink built-in
    • Auto-config from wireguard config
    • Managing the secondary daemon may be a bit tricky, does the main wireguard even run as a daemon or does it run in a JNI library?
    • Make a merged daemon/code-base that runs both wireguard and wirelink in a single process?

Security

  • Drop privileges after startup
    • This is obstructed by Go's lack of support: golang/go#1435
    • Worked around for now by having systemd units drop privileges
  • Trust delegation (i.e. implement SetTrust level)
  • Improved trust models
    • E.g. require a majority of trust sources to agree before adding a peer (Membership in this mode gets a bit more complicated)
    • Conflating router status with trust levels is getting confusing, consider removing this, or having it be a separate trust model from the config, and which is enabled is part of the config

Fancy

  • Use packet capture to detect when we are actually trying to talk to a peer
    • Use this to only do peer setup when we need it
  • Delete peers that are offline
    • E.g. by not broadcasting facts about if if it's unhealthy

Chatter Management

  • Tell routers who we want to talk to (a peer-valued fact)
    • Only tell leaves info about peers they want
    • Or by IP address (see packet capture): only enable local peers that we want to talk to or who want to talk to us (requires fact exchange about who we want to talk to so the remote end will add us)

Config

  • Allow configuring various parameters
    • MaxChunk, ChunkPeriod, AlivePeriod, FactTTL, port / offset
    • All of these are a bit tricky as they mostly need to be consistent across the network for things to work well (or in the case of the port, at all)
    • Some internal capability for this is present now to support time acceleration in tests, but it is not yet configurable.
  • Better logger, at least add some basic filtering support for debug mode