[PATCH] Btrfs: wait for ordered extents before removing extent maps

[Filipe David Borba Manana <fdmanana@gmail.com>]

Wang Shilong got into a case where during inode eviction we were
removing an extent map while it was pinned. This triggered a warning
in remove_extent_mapping() because the extent map had the pinned
flag set:

[ 1209.102076]  [<ffffffffa04721b9>] remove_extent_mapping+0x69/0x70 [btrfs]
[ 1209.102084]  [<ffffffffa0466b06>] btrfs_evict_inode+0x96/0x4d0 [btrfs]
[ 1209.102089]  [<ffffffff81073010>] ? wake_atomic_t_function+0x40/0x40
[ 1209.102092]  [<ffffffff8118ab2e>] evict+0x9e/0x190
[ 1209.102094]  [<ffffffff8118b313>] iput+0xf3/0x180
[ 1209.102101]  [<ffffffffa0461fd1>] btrfs_run_delayed_iputs+0xb1/0xd0 [btrfs]
[ 1209.102107]  [<ffffffffa045d358>] __btrfs_end_transaction+0x268/0x350 [btrfs]

Therefore wait for any pending ordered extents, if any, which will
trigger calls to unpin_extent_cache(), before removing the extent maps.

Wang's solution of simply clearing the pinned bit wasn't enough, as after
unpin_extent_cache() will be called and trigger another WARN_ON() because
the lookup for the extent map returned NULL.

Thanks Wang for finding out this.

Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
---
 fs/btrfs/inode.c |    5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index e889779..c2933fb 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -4509,6 +4509,9 @@ static void evict_inode_truncate_pages(struct inode *inode)
 	ASSERT(inode->i_state & I_FREEING);
 	truncate_inode_pages(&inode->i_data, 0);
 
+	/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
+	btrfs_wait_ordered_range(inode, 0, (u64)-1);
+
 	write_lock(&map_tree->lock);
 	while (!RB_EMPTY_ROOT(&map_tree->map)) {
 		struct extent_map *em;
@@ -4566,8 +4569,6 @@ void btrfs_evict_inode(struct inode *inode)
 		btrfs_orphan_del(NULL, inode);
 		goto no_delete;
 	}
-	/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
-	btrfs_wait_ordered_range(inode, 0, (u64)-1);
 
 	if (root->fs_info->log_root_recovering) {
 		BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
-- 
1.7.9.5