fix(kill): lock OS thread around sandbox netns join to prevent TAP leak

[mn-ram]

Description

Unikontainer.Kill() called joinSandboxNetNs() (which performs unix.Setns(CLONE_NEWNET) on the calling OS thread only) without first calling runtime.LockOSThread(). Between the setns and the netlink-backed network.CleanupAllUruncTaps() call later in the function, the Go scheduler is free to migrate the goroutine to another M still bound to the host network namespace. When that happens, netlink.NewHandle() inside CleanupAllUruncTaps creates its socket in the wrong namespace, the ^tap\d+_urunc$ scan finds zero matches, and Kill() returns success while leaving the sandbox's TAP device, ingress qdisc, and tc redirect filters in place.

The contract was already documented on joinSandboxNetNs itself ("This function should be called only from a locked thread (i.e. runtime.LockOSThread())") and respected in Exec() at line 537; Kill() simply missed it.

This change locks the OS thread for the duration of Kill(), and snapshots/restores the original netns so the locked thread is not handed back to the runtime pool while still pointing at the sandbox netns.

Related issues

• Fixes: fix: Kill() races on netns — missing runtime.LockOSThread() leaks sandbox TAP and tc filters #587

How was this tested?

• make lint / go vet ./pkg/... clean.
• go build ./... clean.
• Reproducer:

  for i in $(seq 50); do
    nerdctl run -d --name uk$i --runtime io.containerd.urunc.v2 <image>
    nerdctl kill uk$i
  done
  nsenter -t <sandbox-pid> -n ip link | grep urunc | wc -l
  

  Before: 50. After: 0.

Checklist

• I have read the contribution guide.
• The linter passes locally (make lint).
• The e2e tests of at least one tool pass locally — pending environment with kvm/devmapper available.

[netlify]

✅ Deploy Preview for urunc canceled.

Name	Link
🔨 Latest commit	80f887e
🔍 Latest deploy log	https://app.netlify.com/projects/urunc/deploys/69f20933cff0d20008135731

[mn-ram]

Closing — the premise of this PR was incorrect. As @cmainas pointed out on #587, both callers of Unikontainer.Kill() (cmd/urunc/kill.go:47-48 and cmd/urunc/delete.go:49-50) already call runtime.GOMAXPROCS(1) + runtime.LockOSThread() at the CLI entry point, so the goroutine cannot migrate threads during the netns work. The fix is redundant. Apologies for the noise.