kernel export threads like nfsd, lockd and ksmbd can deadlock if
exported file system does not support asynchronous processing of
blocking locks.
Some time ago this problem was work arounded in fuse by commit
48e90761b570 ("fuse: lockd support"), however it was not fully
correct because of all locking requests was disables including
non-blocking (i.e. w/o FL_SLEEP in fl_flags) locks.
Now this check is incomplete: nfs v4 does not use lockd and handles
locking request via nfsd directly, recently added ksmbd uses
vfs_lock_file() too. However both these servers does not have
fl->fl_lmops->lm_grant defined.
Original problem was noticed again, and now it will be fixed on the
server side: all affected kernel threads will not use FL_SLEEP if
exported file system does not support asynchronous processing of
blocking locks.
According patches was submitted but not merged yet
[PATCH] nfs: block notification on fs with its own ->lock
[PATCH] ksmbd: force "fail immediately" flag on fs with its own ->lock
Signed-off-by: Vasily Averin <vvs@xxxxxxxxxxxxx>
---
fs/fuse/file.c | 5 -----
1 file changed, 5 deletions(-)
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 9d6c5f6361f7..06e0d7fa86f3 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -2509,11 +2509,6 @@ static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
pid_t pid_nr = pid_nr_ns(pid, fm->fc->pid_ns);
int err;
- if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
- /* NLM needs asynchronous locks, which we don't support yet */
- return -ENOLCK;
- }
-
/* Unlock on close is handled by the flush method */
if ((fl->fl_flags & FL_CLOSE_POSIX) == FL_CLOSE_POSIX)
return 0;
--
2.25.1
