[PATCH] afs: handle async processing of F_SETLK with FL_SLEEP flag

Vasily Averin vvs at virtuozzo.com
Thu Dec 23 04:51:31 PST 2021


kernel export thread (nfsd/lockd/ksmbd) use F_SETLK cmd with set
FL_SLEEP flag to request asynchronous processing of blocking locks.

afs does not support it, does not check F_SETLK cmd and handles
FL_SLEEP flag like usual blocking lock request.

To work around the problem let's detect this situation and
drop FL_SLEEP flag before afs_do_setlk() execution.
Dropped flag should be restored back because some calling function
(nfsd4_lock) require it.

https://bugzilla.kernel.org/show_bug.cgi?id=215383
Signed-off-by: Vasily Averin <vvs at virtuozzo.com>
---
 fs/afs/flock.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/fs/afs/flock.c b/fs/afs/flock.c
index c4210a3964d8..297c859d1f54 100644
--- a/fs/afs/flock.c
+++ b/fs/afs/flock.c
@@ -778,11 +778,19 @@ int afs_lock(struct file *file, int cmd, struct file_lock *fl)
 	fl->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
 	trace_afs_flock_op(vnode, fl, afs_flock_op_lock);
 
-	if (fl->fl_type == F_UNLCK)
+	if (fl->fl_type == F_UNLCK) {
 		ret = afs_do_unlk(file, fl);
-	else
+	} else {
+		bool async = (fl->fl_flags & FL_SLEEP) && IS_SETLK(cmd);
+
+		if (async)
+			fl->fl_flags &= ~FL_SLEEP;
+
 		ret = afs_do_setlk(file, fl);
 
+		if (async)
+			fl->fl_flags |= FL_SLEEP;
+	}
 	switch (ret) {
 	case 0:		op = afs_flock_op_return_ok; break;
 	case -EAGAIN:	op = afs_flock_op_return_eagain; break;
-- 
2.25.1




More information about the linux-afs mailing list