On Mon, Oct 25, 2021 at 05:43:06PM +0800, Xie Yongji wrote:
> The block layer can't support the block size larger than
> page size yet. If an untrusted device presents an invalid
> block size in configuration space, it will result in the
> kernel crash something like below:
>
> [ 506.154324] BUG: kernel NULL pointer dereference, address: 0000000000000008
> [ 506.160416] RIP: 0010:create_empty_buffers+0x24/0x100
> [ 506.174302] Call Trace:
> [ 506.174651] create_page_buffers+0x4d/0x60
> [ 506.175207] block_read_full_page+0x50/0x380
> [ 506.175798] ? __mod_lruvec_page_state+0x60/0xa0
> [ 506.176412] ? __add_to_page_cache_locked+0x1b2/0x390
> [ 506.177085] ? blkdev_direct_IO+0x4a0/0x4a0
> [ 506.177644] ? scan_shadow_nodes+0x30/0x30
> [ 506.178206] ? lru_cache_add+0x42/0x60
> [ 506.178716] do_read_cache_page+0x695/0x740
> [ 506.179278] ? read_part_sector+0xe0/0xe0
> [ 506.179821] read_part_sector+0x36/0xe0
> [ 506.180337] adfspart_check_ICS+0x32/0x320
> [ 506.180890] ? snprintf+0x45/0x70
> [ 506.181350] ? read_part_sector+0xe0/0xe0
> [ 506.181906] bdev_disk_changed+0x229/0x5c0
> [ 506.182483] blkdev_get_whole+0x6d/0x90
> [ 506.183013] blkdev_get_by_dev+0x122/0x2d0
> [ 506.183562] device_add_disk+0x39e/0x3c0
> [ 506.184472] virtblk_probe+0x3f8/0x79b [virtio_blk]
> [ 506.185461] virtio_dev_probe+0x15e/0x1d0 [virtio]
>
> So this patch tries to use the block layer helper to
> validate the block size.
>
> Signed-off-by: Xie Yongji <xieyongji@xxxxxxxxxxxxx>
> ---
> drivers/block/virtio_blk.c | 7 +++++--
> 1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 303caf2d17d0..5bcacefe969e 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -815,9 +815,12 @@ static int virtblk_probe(struct virtio_device *vdev)
> err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
> struct virtio_blk_config, blk_size,
> &blk_size);
> - if (!err)
> + if (!err) {
> + if (blk_validate_block_size(blk_size))
> + goto out_cleanup_disk;
> +
Did you test this with an invalid blk size? It seems unlikely that it
fails properly the way you'd expect.
> blk_queue_logical_block_size(q, blk_size);
> - else
> + } else
> blk_size = queue_logical_block_size(q);
>
> /* Use topology information if available */
> --
> 2.11.0
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
