Skip to contentSkip to conversation

don't bother with vfsmount_lock in mounts_poll()

PATCH In review Al Viro 1 participant 1 message Oct 3, 2013 VFS

lore
Share & tools
lore

Al Viro

author
#0 lore raw

PATCH 11/17 don't bother with vfsmount_lock in mounts_poll()

wake_up_interruptible/poll_wait provide sufficient barriers; just use ACCESS_ONCE() to fetch ns->event and that's it.

---
 fs/proc_namespace.c |    8 ++++----
 1 files changed, 4 insertions(+), 4 deletions(-)
fs/proc_namespace.c +4 -4
index 5fe34c3..439406e 100644
@@ -20,15 +20,15 @@ static unsigned mounts_poll(struct file *file, poll_table *wait)
 	struct proc_mounts *p = proc_mounts(file->private_data);
 	struct mnt_namespace *ns = p->ns;
 	unsigned res = POLLIN | POLLRDNORM;
+	int event;
 
 	poll_wait(file, &p->ns->poll, wait);
 
-	br_read_lock(&vfsmount_lock);
-	if (p->m.poll_event != ns->event) {
-		p->m.poll_event = ns->event;
+	event = ACCESS_ONCE(ns->event);
+	if (p->m.poll_event != event) {
+		p->m.poll_event = event;
 		res |= POLLERR | POLLPRI;
 	}
-	br_read_unlock(&vfsmount_lock);
 
 	return res;
 }
Signature · 1 line
1.7.2.5