diff options
| author | Kent Overstreet <kent.overstreet@gmail.com> | 2021-11-03 10:46:35 -0400 |
|---|---|---|
| committer | Kent Overstreet <kent.overstreet@linux.dev> | 2022-10-03 22:49:31 -0400 |
| commit | 96e9e05ab5df76fa9d3e84a1176989f161ec8ec5 (patch) | |
| tree | 566078ff0fc7c32fed7f302d2995a43ab6793ee4 | |
| parent | 8a9c0304a3464db98b06e99931de51dda0d9d12f (diff) | |
loop: Don't call disk_force_media_change from lo_release
This fixes a deadlock that affects bcachefs, where we call blkdev_put()
directly from put_super().
put_super
-> blkdev_put
-> lo_release
-> disk_force_media_change
-> __invalidate_device
-> get_super
where we self deadlock on s_umount.
It doesn't affect other filesystems because they use the bizzare sget()
mechanism for exclusion with other mounts, and thus don't need to close
block devices in .put_super().
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
| -rw-r--r-- | drivers/block/loop.c | 2 |
1 files changed, 0 insertions, 2 deletions
diff --git a/drivers/block/loop.c b/drivers/block/loop.c index ad92192c7d61..123c8837e57b 100644 --- a/drivers/block/loop.c +++ b/drivers/block/loop.c @@ -1171,8 +1171,6 @@ static void __loop_clr_fd(struct loop_device *lo, bool release) if (!release) blk_mq_unfreeze_queue(lo->lo_queue); - disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); - if (lo->lo_flags & LO_FLAGS_PARTSCAN) { int err; |