Btrfs: fix race of using total_bytes_pinned
authorLiu Bo <bo.li.liu@oracle.com>
Wed, 2 Jul 2014 08:58:01 +0000 (16:58 +0800)
committerChris Mason <clm@fb.com>
Thu, 3 Jul 2014 14:04:15 +0000 (07:04 -0700)
commitd288db5dc0110c8e0732d099aaf7a05e2ea0e0c8
tree8f25c749e2d6199d47888278c3a22b5dc066249a
parent130d5b415a091e493ac1508b9d27bbb85ba7b8c0
Btrfs: fix race of using total_bytes_pinned

This percpu counter @total_bytes_pinned is introduced to skip unnecessary
operations of 'commit transaction', it accounts for those space we may free
but are stuck in delayed refs.

And we zero out @space_info->total_bytes_pinned every transaction period so
we have a better idea of how much space we'll actually free up by committing
this transaction.  However, we do the 'zero out' part a little earlier, before
we actually unpin space, so we end up returning ENOSPC when we actually have
free space that's just unpinned from committing transaction.

xfstests/generic/074 complained then.

This fixes it by actually accounting the percpu pinned number when 'unpin',
and since it's protected by space_info->lock, the race is gone now.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/extent-tree.c