diff options
| author | Alexander Leidinger <netchild@FreeBSD.org> | 2026-07-19 07:38:52 +0000 |
|---|---|---|
| committer | Alexander Leidinger <netchild@FreeBSD.org> | 2026-07-19 15:13:40 +0000 |
| commit | d6915bffb7b68d9b55fa3db4e5709463549c379e (patch) | |
| tree | 1ea1c32fd6f06e99c2476bc4d6cb6775c83de1a6 | |
| parent | 9b21a52495758294e3a50542a59bc47a0c184173 (diff) | |
nullfs: close a race when syncing inotify flags from the lower vnode
After a bypassed VOP, nullfs mirrors the lower vnode's inotify state
onto the upper vnode. The flags were checked with lockless reads
before being updated with the asserting flag set/unset primitives, so
two threads syncing the same vnode concurrently (or a sync racing a
watch being established) could both decide to make the same change;
the loser then trips the "flags already set" assertion on an
INVARIANTS kernel. On other kernels the race is harmless.
Keep the lockless check as the fast path, but re-make the decision
under the vnode interlock before actually changing the flags.
Reproduced in a 4-CPU VM with one thread cycling an inotify watch on
a lower-filesystem file while several threads stat(2) the same file
through a nullfs mount: the unpatched INVARIANTS kernel panics under
this load, the patched kernel runs it to completion.
Fixes: f1f230439fa4 ("vfs: Initial revision of inotify")
MFC after: 2 weeks
Differential Revision: D58344
Reviewed by: markj
Assisted-by: Claude Code (Fable 5)
| -rw-r--r-- | sys/fs/nullfs/null_vnops.c | 19 |
1 files changed, 14 insertions, 5 deletions
diff --git a/sys/fs/nullfs/null_vnops.c b/sys/fs/nullfs/null_vnops.c index 50d359c9b909..41823d0c6202 100644 --- a/sys/fs/nullfs/null_vnops.c +++ b/sys/fs/nullfs/null_vnops.c @@ -200,17 +200,26 @@ SYSCTL_INT(_debug, OID_AUTO, nullfs_bug_bypass, CTLFLAG_RW, * VOP_INOTIFY. * - If the lower vnode is watched, then the upper vnode should go through * VOP_INOTIFY, so copy the flag up. + * + * The lockless check is only a fast path: the decision to change a flag + * is re-made under the upper vnode's interlock, since another thread may + * set or clear the flag concurrently. */ static void null_copy_inotify(struct vnode *vp, struct vnode *lvp, short flag) { + if (__predict_true((vn_irflag_read(vp) & flag) == + (vn_irflag_read(lvp) & flag))) + return; + VI_LOCK(vp); if ((vn_irflag_read(vp) & flag) != 0) { - if (__predict_false((vn_irflag_read(lvp) & flag) == 0)) - vn_irflag_unset(vp, flag); - } else if ((vn_irflag_read(lvp) & flag) != 0) { - if (__predict_false((vn_irflag_read(vp) & flag) == 0)) - vn_irflag_set(vp, flag); + if ((vn_irflag_read(lvp) & flag) == 0) + vn_irflag_unset_locked(vp, flag); + } else { + if ((vn_irflag_read(lvp) & flag) != 0) + vn_irflag_set_locked(vp, flag); } + VI_UNLOCK(vp); } /* |