v0.5.4

Added

• Quic protocol can now be disabled using the --disable-quic flag
• Mycelium can now be started with a configuration file using --config-file.
  If no configuration file is supplied, Mycelium will look in a default location
  based on the OS. For more information see README
• Windows installer for Mycelium. The .msi file can be downloaded from the release
  assets.
• Added flag to specify how many update workers should be started, which governs
  the amount of parallelism used for processing updates.
• Send a seqno request if we receive an unfeasible update for a subnet with no
  routes, or if there is no selected route for the subnet.
• New public peers in US, India, and Singapore.

Changed

• Increased the starting metric of a peer from 50 to 1000.
• Reworked the internals of the routing table, which should reduce memory consumption.
  Additionally, it is now possible to apply updates in parallel
• Periodically reduce the allocated size of the seqno cache to avoid wasting some
  memory which is not currently used by the cache but still allocated.
• Demote seqno cache warnings about duplicate seqno requests go debug lvl, as it
  is valid to send duplicate requests if sufficient time passed.
• Skip route selection after an unfeasible update to a fallback route, as the (now
  unfeasible) route won't be selected anyway.
• No longer refresh route timer after an unfeasbile update. This allows routes
  which have become unfeasible to gracefully be removed from the routing table
  over time.
• Expired routes which aren't selected are now immediately removed from the routing
  table.
• Changed how updates are sent to be more performant.
• A triggered update is no longer sent just because a route sequence number got
  increased. We do still send the update to peer in the seqno request cache.
• Reduced log level when a route changes next-hop to debug from info.

Fixed

• When running mycelium with a command, a keyfile was loaded (or created, if not
  yet present). This was not necessary in that context.
• Limit the amount of time allowed for inbound quic connections to be set up, and
  process multiple of them in parallel. This fixes a DOS vector against the quic
  listener.
• We now update the source table even if we don't send an update because we are
  sure the receiver won't select us as a next-hop anyway.

v0.5.3

Added

• On Linux and macOS, a more descriptive error is printed when setting up the tun
  device fails because a device with the same name already exists.
• Seqno request cache, to avoid spamming peers with duplicate seqno requests and
  to make sure seqno's are forwarded to different peers.
• Added myceliumd-private binary, which contains private network functionality.
• Added API endpoint to retrieve the public key associated with an IP.
• The CLI can now be used to list, remove or add peers (see mycelium peers --help)
• The CLI can now be used to list selected and fallback routes (see mycelium routes --help)

Changed

• We now send seqno requests to all peers who advertised a subnet if the selected
  route to it is lost as a result of the next-hop dying, or and update coming in
  which causes no routes to be feasible anymore.
• Switched from the log to the tracing ecosystem.
• Only do the periodic route announcement every 5 minutes instead of every minute.
• Mycelium binary is no longer part of the workspace, and no longer contains private
  network functionality.
• If a packet received from a peer can't be forwarded to the router, terminate the
  connection to the peer.

Fixed

• Manually implement Hash for Subnet, previously we could potentially have multiple
  distinct entries in the source table for the same source key.

v0.5.2

Added

• New CI workflow to build and test the mycelium library separately from the full
  provided binary build.

Changed

• Disabled the protobuf feature on prometheus, this removes protobuf related code
  and significantly reduces the release binary size.
• Changed log level when sending a protocol message to a peer which is no longer
  alive from error to trace in most instances.
• Improved performance of sending protocol messages to peers by queueing up multiple
  packets at once (if multiple are ready).
• Before trying to send an update we now check if it makes sense to do so.
• If a peer died, fallback routes using it are no longer retained with an infinite
  metric but removed immediately.
• No longer run route selection for subnets if a peer died and the route is not
  selected.
• If routes are removed, shrink the capacity of the route list in the route table
  if it is larger than required.
• Check if the originator of a TLV is still available before processing said TLV.
• The router now uses a dedicated task per TLV type to handle received TLV's from
  peers.
• Statically linking openssl is now a feature flag when building yourself.

Fixed

• If a peer died, unselect the selected routes which have it as next-hop if there
  is no other feasible route.
• Properly unselect a route if a retraction update comes in and there is no other
  feasible route.
• If the router bumps it's seqno it now properly announces the local route to it's
  peers instead of the selected routes
• Seqno bump requests for advertised local routes now properly bump the router
  seqno.

v0.5.1

Added

• The repo is now a workspace, and pure library code is separated out. This is mainly
  done to make it easier to develop implementations on different platforms.
• Link local discovery will now send discovery beacons on every interface the process
  listens on for remote beacons.
• Experimental private network support. See the private network docs
  for more info.
• You can now optionally expose Prometheus compatible metrics about the system by
  setting the --metrics-api-address flag.
• On Linux, you can now set an optional firewall mark by setting the --firewall-mark
  flag.
• Added a nix flake to the repo.

Changed

• We no longer create an outbound connection to a link local discovered IP if that
  IP is already known (usually as inbound address) with potentially a different
  port.

v0.5.0

Changed

• Connection identifier is now included in the error log if we can't forward a
  seqno request.
• Garbage collection time for source entries has been increased from 5 to 30 minutes
  for now.
• The router implementation has been changed to use regular locks instead of an
  always readable concurrency primitive for all but the actual routing table. This
  should reduce the memory consumption a bit.
• Public key and shared secret for a destination are now saved on the router, instead
  of maintaining a separate mapping for them. This slightly reduces memory consumption
  of the router, and ensures stale data is properly cleaned up when all routes to
  a subnet are removed.
• Hello packets now set the interval in which the next Hello will be sent properly
  in centiseconds.
• IHU packets now set the interval properly in centiseconds.
• IHU packets now set an RX cost. For now this is the link cost, in the future
  this will be set properly.
• Route expiration time is now calculated from the interval received in updates.
• Ip address derivation from public keys now uses the blake3 hash algorithm.

Fixed

• Don't try to forward seqno requests to a peer if we know its connection is dead.

v0.4.5

Changed

• Size of data packets is limited to 65535 bytes.
• Update interval is now expressed as centiseconds, in accordance with the babel
  RFC.
• Update filters now allow retractions for a route from any router-id.

Fixed

• The feasibility distance of an existing source key is no longer incorrectly updated
  when the metric increases.
• Source key garbage collection timers are properly reset on update even if the
  source key itself is not updated.
• Nodes now properly reply to route requests for a static route.
• A retraction is now sent as reply to a route request if the route is not known.

v0.4.4

Changed

• The amount of bytes read and written to a peer are now no longer reset after
  a reconnect (for outgoing connection).
• Renamed connectionTxBytes and connectionRxBytes on the peer stats struct
  to txBytes and rxBytes to better express that they are no longer tied to
  a single connection to the peer.

Fixed

• When joining a link local multicast group on an interface returns a
  Address already in use error, the error is now ignored and the interface is
  considered to be joined.
• When sending an update to a peer, the source table is now correctly updated before
  the update is sent, instead of doing a batched source table update afterward.

v0.4.3

Added

• Feature flag for message subsystem. It is enabled by default, but a user can
  make a custom build with --default-features-false which completely leaves out
  the message related code, should he desire this and have no need for it.
• Link local discovery now periodically checks for new IPv6 enabled interfaces
  and also joins the discovery multicast group on them.
• Trace logs are removed from release binaries at compile time, slightly reducing
  binary size.
• New --silent flag which disables all logging except error logs.

Changed

• Update GitHub CI action to use latest version of the checkout action.
• Update GitHub CI action to stop using deprecated actions-rs actions.
• Failing to join the link local discovery multicast group now logs as warning
  instead of error.
• Failing to join any IPv6 multicast group for link local peer discovery will no
  longer disable local peer discovery entirely.

Fixed

• Add proper validation when receiving an OOB ICMP packet.

v0.4.2

Fixed

• Make sure the HTTP API doesn't shut down immediately after startup.

v0.4.1

Added

• Admin API
  • Ability to see current peers and related info
  • Ability to add a new peer
  • Ability to remove an existing peer
  • List current selected routes
  • List current fallback routes
  • General node info (for now just the node subnet)

Changed

• The tokio_unstable config flag is no longer used when building.
• The key file is now created without read permissions for the group/world.

Removed

• .cargo/config.toml aarch64-linux target specific entries. Cross compilation for
  these platforms can use cross or entries in the global .cargo/config.toml of
  the developer instead.
• Sending SIGUSR1 to the process on unix based systems no longer dumps internal
  state, this can be accessed with the admin API instead.