btrfs: fix double-decrement of bytes_may_use in submit_one_async_extent()

[ Upstream commit 82323b1 ]

submit_one_async_extent() calls btrfs_reserve_extent(), which decrements
bytes_may_use. If the call btrfs_create_io_em() fails, we jump to
out_free_reserve, which calls extent_clear_unlock_delalloc().

Because we're specifying EXTENT_DO_ACCOUNTING, i.e.
EXTENT_CLEAR_META_RESV | EXTENT_CLEAR_DATA_RESV, this decreases
bytes_may_use again. This can lead to problems later on, as an initial
write can fail only for the writeback to silently ENOSPC.

Fix this by replacing EXTENT_DO_ACCOUNTING with EXTENT_CLEAR_META_RESV.
This parallels a4fe134 ("btrfs: fix a double release on reserved
extents in cow_one_range()"), which is the same fix in cow_one_range().

Fixes: 151a41b ("Btrfs: fix what bits we clear when erroring out from delalloc")
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Mark Harmstone <mark@harmstone.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

Author: maharmstone

fs/btrfs/inode.c

@@ -1288,7 +1288,7 @@ static void submit_one_async_extent(struct async_chunk *async_chunk,
 				     NULL, &cached,
 				     EXTENT_LOCKED | EXTENT_DELALLOC |
 				     EXTENT_DELALLOC_NEW |
-				     EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
+				     EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV,
 				     PAGE_UNLOCK | PAGE_START_WRITEBACK |
 				     PAGE_END_WRITEBACK);
 	free_async_extent_pages(async_extent);