Btrfs: fix regression in lock_delalloc_pages
authorLiu Bo <bo.li.liu@oracle.com>
Tue, 7 Mar 2017 02:20:56 +0000 (18:20 -0800)
committerChris Mason <clm@fb.com>
Fri, 17 Mar 2017 20:47:09 +0000 (13:47 -0700)
The bug is a regression after commit
(da2c7009f6ca "btrfs: teach __process_pages_contig about PAGE_LOCK operation")
and commit
(76c0021db8fd "Btrfs: use helper to simplify lock/unlock pages").

So if the dirty pages which are under writeback got truncated partially
before we lock the dirty pages, we couldn't find all pages mapping to the
delalloc range, and the bug didn't return an error so it kept going on and
found that the delalloc range got truncated and got to unlock the dirty
pages, and then the ASSERT could caught the error, and showed

-----------------------------------------------------------------------------
assertion failed: page_ops & PAGE_LOCK, file: fs/btrfs/extent_io.c, line: 1716
-----------------------------------------------------------------------------

This fixes the bug by returning the proper -EAGAIN.

Cc: David Sterba <dsterba@suse.com>
Reported-by: Dave Jones <davej@codemonkey.org.uk>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/extent_io.c

index 28e81922a21c1ecead950f50cf3e685ad03c57f6..8df797432740df92e9c35c67fa75ce0331c193ff 100644 (file)
@@ -1714,7 +1714,8 @@ static int __process_pages_contig(struct address_space *mapping,
                         * can we find nothing at @index.
                         */
                        ASSERT(page_ops & PAGE_LOCK);
-                       return ret;
+                       err = -EAGAIN;
+                       goto out;
                }
 
                for (i = 0; i < ret; i++) {