OSDN Git Service

locks: don't check for race with close when setting OFD lock
authorJeff Layton <jeff.layton@primarydata.com>
Fri, 8 Jan 2016 12:30:43 +0000 (07:30 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 17 Jan 2018 08:35:27 +0000 (09:35 +0100)
commit 0752ba807b04ccd69cb4bc8bbf829a80ee208a3c upstream.

We don't clean out OFD locks on close(), so there's no need to check
for a race with them here. They'll get cleaned out at the same time
that flock locks are.

Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Acked-by: "J. Bruce Fields" <bfields@fieldses.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Mel Gorman <mgorman@suse.de>
fs/locks.c

index 8eddae2..b515e65 100644 (file)
@@ -2220,10 +2220,12 @@ int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
        error = do_lock_file_wait(filp, cmd, file_lock);
 
        /*
-        * Attempt to detect a close/fcntl race and recover by
-        * releasing the lock that was just acquired.
+        * Attempt to detect a close/fcntl race and recover by releasing the
+        * lock that was just acquired. There is no need to do that when we're
+        * unlocking though, or for OFD locks.
         */
-       if (!error && file_lock->fl_type != F_UNLCK) {
+       if (!error && file_lock->fl_type != F_UNLCK &&
+           !(file_lock->fl_flags & FL_OFDLCK)) {
                /*
                 * We need that spin_lock here - it prevents reordering between
                 * update of i_flctx->flc_posix and check for it done in
@@ -2362,10 +2364,12 @@ int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
        error = do_lock_file_wait(filp, cmd, file_lock);
 
        /*
-        * Attempt to detect a close/fcntl race and recover by
-        * releasing the lock that was just acquired.
+        * Attempt to detect a close/fcntl race and recover by releasing the
+        * lock that was just acquired. There is no need to do that when we're
+        * unlocking though, or for OFD locks.
         */
-       if (!error && file_lock->fl_type != F_UNLCK) {
+       if (!error && file_lock->fl_type != F_UNLCK &&
+           !(file_lock->fl_flags & FL_OFDLCK)) {
                /*
                 * We need that spin_lock here - it prevents reordering between
                 * update of i_flctx->flc_posix and check for it done in