btrfs: btrfs_issue_discard ensure offset/length are aligned to sector boundaries
authorJeff Mahoney <jeffm@suse.com>
Mon, 15 Jun 2015 13:41:15 +0000 (09:41 -0400)
committerChris Mason <clm@fb.com>
Wed, 29 Jul 2015 15:15:24 +0000 (08:15 -0700)
It's possible, though unexpected, to pass unaligned offsets and lengths
to btrfs_issue_discard.  We then shift the offset/length values to sector
units.  If an unaligned offset has been passed, it will result in the
entire sector being discarded, possibly losing data.  An unaligned
length is safe but we'll end up returning an inaccurate number of
discarded bytes.

This patch aligns the offset to the 512B boundary, adjusts the length,
and warns, since we shouldn't be discarding on an offset that isn't
aligned with our sector size.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Tested-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/extent-tree.c

index 16655bb5f2936af3fb85710505a38440bf270eda..7aa6ad1f014d7ed4aeaec32022439a2785de22a5 100644 (file)
@@ -1887,12 +1887,21 @@ static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
                               u64 *discarded_bytes)
 {
        int ret = 0;
+       u64 aligned_start = ALIGN(start, 1 << 9);
 
-       *discarded_bytes = 0;
-       ret = blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
-       if (!ret)
-               *discarded_bytes = len;
+       if (WARN_ON(start != aligned_start)) {
+               len -= aligned_start - start;
+               len = round_down(len, 1 << 9);
+               start = aligned_start;
+       }
 
+       *discarded_bytes = 0;
+       if (len) {
+               ret = blkdev_issue_discard(bdev, start >> 9, len >> 9,
+                                          GFP_NOFS, 0);
+               if (!ret)
+                       *discarded_bytes = len;
+       }
        return ret;
 }