qcow2: Allow discard of final unaligned cluster

As mentioned in commit 0c1bd46, we ignored requests to
discard the trailing cluster of an unaligned image.  While
discard is an advisory operation from the guest standpoint,
(and we are therefore free to ignore any request), our
qcow2 implementation exploits the fact that a discarded
cluster reads back as 0.  As long as we discard on cluster
boundaries, we are fine; but that means we could observe
non-zero data leaked at the tail of an unaligned image.

Enhance iotest 66 to cover this case, and fix the implementation
to honor a discard request on the final partial cluster.

Signed-off-by: Eric Blake <eblake@redhat.com>
Message-id: 20170407013709.18440-1-eblake@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

1 files changed, 6 insertions, 1 deletions

diff --git a/block/qcow2.c b/block/qcow2.c
index 4ca4cf04b0..5c1573c999 100644
--- a/block/qcow2.c
+++ b/block/qcow2.c
@@ -2515,7 +2515,12 @@ static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
 
     if (!QEMU_IS_ALIGNED(offset | count, s->cluster_size)) {
         assert(count < s->cluster_size);
-        return -ENOTSUP;
+        /* Ignore partial clusters, except for the special case of the
+         * complete partial cluster at the end of an unaligned file */
+        if (!QEMU_IS_ALIGNED(offset, s->cluster_size) ||
+            offset + count != bs->total_sectors * BDRV_SECTOR_SIZE) {
+            return -ENOTSUP;
+        }
     }
 
     qemu_co_mutex_lock(&s->lock);