Repository navigation
5.0.1: fixes from the review of 5.0.0 - #208
Conversation
From the Qodo review of #202: - Reload: a network slot that takes another network, or whose mode changes (tap, addresses = none), ends its old sessions (a generation per slot, netw.c); a client of a removed network no longer stays on in the one that took its slot, with its key (2, 6, 8). - Reload: client isolation turned on ends the network's mesh sessions, so no direct path set up before goes on past the server (2). - Reload: the gateway alias comes off the tunnel also for addresses = none or another pool size (7). - A TAP client keeps a given MTU (--mtu, mtu=, CONFIG) instead of 1500: mtu_given in the packet open parameters (3); test-e2e-tap checks it. - known_servers: one writer at a time (<file>.lock, flock) through a mkstemp file, not a fixed .new (11). - --bridge-mode nat: a remote device not heard for 10 minutes is not sent to any more, and its neighbour entry goes (12). - camex.ko: an skb queued while the offloads were on is linearised before an ordinary read (10). 1 and 9 are issues #206 and #207.
PR Summary by QodoFix reload session isolation and networking regressions in 5.0.1
AI Description
Merge Risk
|
Code Review by Qodo
1. Isolated clients keep direct peer paths
|
…hared neighbour entries - known_servers: when the lock cannot be taken (busy for 2 s, or not writable), the key is not kept this time instead of a rewrite that could drop another writer's line. - Isolation turned on: the mesh client's session ends with server_end_client(), so its address stays leased to it. - --bridge-mode nat: an expired remote removes its neighbour entry only when no live remote has that address.
|
/review |
|
On the Qodo review of ea265ee, all three fixed in 224c126:
Checked locally: |
|
/review |
Fixes from the Qodo review of #202 (5.0.0); version 5.0.1.
--mtu,mtu=, the server's CONFIG) instead of the LAN's 1500:mtu_givenin the packet open parameters.mode = tap,addresses = none), ends its old sessions, key and PAKE ones too (a generation per slot).addresses = noneor another pool size.camex.ko: an skb queued while the offloads were on is linearised before an ordinary read (CAMEX_IOC_VNET turned off).known_servers: one writer at a time (<file>.lock,flock), through amkstempfile instead of a fixed.new.--bridge-mode nat: a remote device not heard for 10 minutes is not sent to any more, and its neighbour entry goes.1 and 9 are #206 and #207; 4 and 5 are answered in #202.
Tests, locally in the CI's images: all 25 e2e tests of the workflow (
test-e2e-tap.shnow also checks that--mtu 1420stays on a TAP device),camex.koin QEMU (ubuntu:24.04: VM TEST PASS, version 5.0.1), Windows x86_64 and i686 under Wine,ci/build-musl.sh armwith the unit tests under qemu-arm,make test.