[PATCH 3/5] fs: make mount(2) remounts killable on s_umount

[Josef Bacik]

mount(2) with MS_REMOUNT takes s_umount exclusively, with MS_BIND as well
it takes it shared.  If the lock is held for a long time, by a syncfs(2) on
a device which doesn't answer say, the remount sits in D state and can't be
killed.

The one which wants the lock exclusively makes matters worse for everybody
else: while it is queued, every new reader of s_umount queues up behind it.
Killing it gets it out of the way, the rwsem wakes those behind a waiter
which gives up.

Nothing has been changed yet when the lock is taken, and both can fail.
Use the killable variants.

Signed-off-by: Josef Bacik <josef@xxxxxxxxxxxxxx>
---
 fs/namespace.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index e5feb4e48290..837f531e494b 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -3485,7 +3485,9 @@ static int do_reconfigure_mnt(const struct path *path, unsigned int mnt_flags)
 	 * We're only checking whether the superblock is read-only not
 	 * changing it, so only take down_read(&sb->s_umount).
 	 */
-	down_read(&sb->s_umount);
+	ret = down_read_killable(&sb->s_umount);
+	if (ret)
+		return ret;
 	lock_mount_hash();
 	ret = change_mount_ro_state(mnt, mnt_flags);
 	if (ret == 0)
@@ -3531,8 +3533,9 @@ static int do_remount(const struct path *path, int sb_flags,
 	fc->oldapi = true;
 
 	err = parse_monolithic_mount_data(fc, data);
+	if (!err)
+		err = down_write_killable(&sb->s_umount);
 	if (!err) {
-		down_write(&sb->s_umount);
 		err = -EPERM;
 		if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
 			err = reconfigure_super(fc);

-- 
2.55.0