feature/conn25: do not lock mu in packetFilterAllow (#22534)

packetFilterAllow holds c.mu while touching nothing it protects.
It only calls getIPSets, a func field assigned once in newConn25
that reads the config out of an atomic pointer.
Its siblings transitIPInPool and destinationInPools already call
it unlocked. Drop the acquisition, which besides being dead
was a lock-ordering hazard if getIPSets ever grows a lock of its own.

Updates https://github.com/tailscale/corp/issues/38924

Change-Id: I2942a9b0a6a57378473393c62611c557f07e71b3

Signed-off-by: Mazdak Nasab <mazdak.nasab@gmail.com>
This commit is contained in:
Mazdak Nasab authored and GitHub committed 2026-11-20 15:24:05 -07:00
1 parent 4f7439b415
commit 8dc50105e4
1 file changed
-2
-2
View File
@@ -152,8 +152,6 @@ func (c *connector) destinationInPools(dst netip.Addr) bool {
// packetFilterAllow returns true if the provided packet has a Src that is in
// the configured transit IP range for this connector, false otherwise.
func (c *connector) packetFilterAllow(p packet.Parsed) (bool, string) {
c.mu.Lock()
defer c.mu.Unlock()
ipSets := c.getIPSets()
if ipSets.v4Transit != nil && ipSets.v4Transit.Contains(p.Dst.Addr()) {
return true, packetFilterAllowReason