xfs: don't perform discard if the given range length is less than block size
authorJie Liu <jeff.liu@oracle.com>
Wed, 20 Nov 2013 08:08:53 +0000 (16:08 +0800)
committerBen Myers <bpm@sgi.com>
Wed, 4 Dec 2013 21:42:52 +0000 (15:42 -0600)
For discard operation, we should return EINVAL if the given range length
is less than a block size, otherwise it will go through the file system
to discard data blocks as the end range might be evaluated to -1, e.g,
# fstrim -v -o 0 -l 100 /xfs7
/xfs7: 9811378176 bytes were trimmed

This issue can be triggered via xfstests/generic/288.

Also, it seems to get the request queue pointer via bdev_get_queue()
instead of the hard code pointer dereference is not a bad thing.

Signed-off-by: Jie Liu <jeff.liu@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ben Myers <bpm@sgi.com>
fs/xfs/xfs_discard.c

index 8367d6dc18c9df7f51cd565e11395cba24929a53..4f11ef0111395bbb7775407e9039e92cad0a932d 100644 (file)
@@ -157,7 +157,7 @@ xfs_ioc_trim(
        struct xfs_mount                *mp,
        struct fstrim_range __user      *urange)
 {
-       struct request_queue    *q = mp->m_ddev_targp->bt_bdev->bd_disk->queue;
+       struct request_queue    *q = bdev_get_queue(mp->m_ddev_targp->bt_bdev);
        unsigned int            granularity = q->limits.discard_granularity;
        struct fstrim_range     range;
        xfs_daddr_t             start, end, minlen;
@@ -180,7 +180,8 @@ xfs_ioc_trim(
         * matter as trimming blocks is an advisory interface.
         */
        if (range.start >= XFS_FSB_TO_B(mp, mp->m_sb.sb_dblocks) ||
-           range.minlen > XFS_FSB_TO_B(mp, XFS_ALLOC_AG_MAX_USABLE(mp)))
+           range.minlen > XFS_FSB_TO_B(mp, XFS_ALLOC_AG_MAX_USABLE(mp)) ||
+           range.len < mp->m_sb.sb_blocksize)
                return -XFS_ERROR(EINVAL);
 
        start = BTOBB(range.start);