[PATCH v5 09/15] block: reject a zone append that is larger than a zone

[Niklas Cassel <cassel@kernel.org>]

A zone append cannot cross a zone boundary, so a request larger than a
zone can never be carried out.

Nothing rejects one. The block layer does not look at the length at all,
and raw_co_zone_append() writes qiov->size bytes at the write pointer
once it has validated the offset, so the write runs past the end of the
zone: on a disk with 8 MiB zones, an append of 8 MiB plus one block
fills the zone and advances the write pointer of the zone after it.

virtio-blk does compare the length, but against
BlockLimits.max_append_sectors, which is how much the backend takes in
one command and not how large a zone is. For a Linux block device that
value is never larger than a zone, so virtio-blk is protected by what
the backend happens to report rather than by the check it makes.

Check the length in bdrv_co_zone_append(), next to the check that an
append is a multiple of the sector size. Both are conditions of the
operation rather than of a medium, so the block layer is where they hold
for every caller and where no driver has to repeat them.

The exact bound is tighter: a request also has to fit in the part of the
zone that is still writable, which depends on the write pointer and so
stays with the drivers.

Signed-off-by: Niklas Cassel <cassel@kernel.org>
---
 block/io.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/block/io.c b/block/io.c
index b9d0b3fd7b..d7c403bc08 100644
--- a/block/io.c
+++ b/block/io.c
@@ -3365,6 +3365,18 @@ int coroutine_fn bdrv_co_zone_append(BlockDriverState *bs, int64_t *offset,
         return -EINVAL;
     }
 
+    /*
+     * An append cannot cross a zone boundary, so one that is larger than a
+     * zone can never be carried out, wherever the write pointer of the zone
+     * is. The exact bound is the part of the zone that is still writable,
+     * which only a driver can check, as only it holds the write pointer.
+     * zone_size is zero when the device is not zoned, which is reported as
+     * unsupported below.
+     */
+    if (bs->bl.zone_size && qiov->size > bs->bl.zone_size) {
+        return -EINVAL;
+    }
+
     bdrv_inc_in_flight(bs);
     if (!drv || !drv->bdrv_co_zone_append || bs->bl.zoned == BLK_Z_NONE) {
         co.ret = -ENOTSUP;
-- 
2.55.0