* [PATCH] virtio_blk: validate zone append sector from device
@ 2026-11-20 6:30 Annie Kim
2026-11-20 6:47 ` sashiko-bot
2026-11-20 13:29 ` Damien Le Moal
0 siblings, 2 replies; 3+ messages in thread
From: Annie Kim @ 2026-11-20 6:30 UTC (permalink / raw)
To: Michael S . Tsirkin, Jason Wang
Cc: Jens Axboe, Damien Le Moal, virtualization, linux-block,
linux-kernel, stable, Hyokyung Kim
From: Hyokyung Kim <pulpannie@gmail.com>
The device reports the starting sector for REQ_OP_ZONE_APPEND completions,
and virtio_blk copies it into req->__sector without checking that it
still refers to the requested zone. A buggy or malicious device can
therefore make upper layers believe the append landed at an arbitrary
location.
Validate successful zone append completions before updating
req->__sector. The returned sector must not precede the request start
and the entire request must fit within the zone writable capacity.
Reject invalid completions with BLK_STS_IOERR.
The polling path may batch successful completions, so perform the same
validation before admitting a request to the batch. Invalid append
completions then fall back to the normal completion path and are
completed with an error instead of being reported as successful.
Fixes: 95bfec41bd3d ("virtio-blk: add support for zoned block devices")
Assisted-by: Codex:GPT-5
Signed-off-by: Hyokyung Kim <pulpannie@gmail.com>
---
Tested on v7.4-rc3+ using modified QEMU backend that returned incorrect
zone-append sectors on a memory-backed null_blk zoned device.
drivers/block/virtio_blk.c | 50 +++++++++++++++++++++++++++++++-------
1 file changed, 41 insertions(+), 9 deletions(-)
diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 2b462005c..8f88570b4 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -331,15 +331,45 @@ static inline u8 virtblk_vbr_status(struct virtblk_req *vbr)
return *((u8 *)&vbr->in_hdr + vbr->in_hdr_len - 1);
}
-static inline void virtblk_update_zone_append_sector(struct request *req,
- struct virtblk_req *vbr)
+#ifdef CONFIG_BLK_DEV_ZONED
+static inline blk_status_t
+virtblk_update_zone_append_sector(struct request *req,
+ struct virtblk_req *vbr,
+ blk_status_t status)
{
struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
+ sector_t req_start, sector, zone_end;
+ unsigned int nr_sectors;
+
+ if (req_op(req) != REQ_OP_ZONE_APPEND || status != BLK_STS_OK)
+ return status;
+
+ req_start = blk_rq_pos(req);
+ sector = virtio64_to_cpu(vblk->vdev, vbr->in_hdr.zone_append.sector);
+ nr_sectors = blk_rq_sectors(req);
+ zone_end = req_start + disk_zone_capacity(vblk->disk, req_start);
+
+ if (zone_end < req_start || sector < req_start || sector >= zone_end ||
+ nr_sectors > zone_end - sector) {
+ dev_err_ratelimited(&vblk->vdev->dev,
+ "zone append sector %llu + %u outside valid range [%llu, %llu)\n",
+ (u64)sector, nr_sectors,
+ (u64)req_start, (u64)zone_end);
+ return BLK_STS_IOERR;
+ }
- if (req_op(req) == REQ_OP_ZONE_APPEND)
- req->__sector = virtio64_to_cpu(vblk->vdev,
- vbr->in_hdr.zone_append.sector);
+ req->__sector = sector;
+ return status;
}
+#else
+static inline blk_status_t
+virtblk_update_zone_append_sector(struct request *req,
+ struct virtblk_req *vbr,
+ blk_status_t status)
+{
+ return status;
+}
+#endif
static inline void virtblk_request_done(struct request *req)
{
@@ -349,7 +379,7 @@ static inline void virtblk_request_done(struct request *req)
virtblk_unmap_data(req, vbr);
virtblk_cleanup_cmd(req);
- virtblk_update_zone_append_sector(req, vbr);
+ status = virtblk_update_zone_append_sector(req, vbr, status);
blk_mq_end_request(req, status);
}
@@ -1242,7 +1272,6 @@ static void virtblk_complete_batch(struct io_comp_batch *iob)
virtblk_unmap_data(req, vbr);
virtblk_cleanup_cmd(req);
- virtblk_update_zone_append_sector(req, vbr);
}
blk_mq_end_request_batch(iob);
}
@@ -1260,11 +1289,14 @@ static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
struct request *req = blk_mq_rq_from_pdu(vbr);
- u8 status = virtblk_vbr_status(vbr);
+ blk_status_t status;
+
+ status = virtblk_result(virtblk_vbr_status(vbr));
+ status = virtblk_update_zone_append_sector(req, vbr, status);
found++;
if (!blk_mq_complete_request_remote(req) &&
- !blk_mq_add_to_batch(req, iob, status != VIRTIO_BLK_S_OK,
+ !blk_mq_add_to_batch(req, iob, status != BLK_STS_OK,
virtblk_complete_batch))
virtblk_request_done(req);
}
--
2.53.0
^ permalink raw reply related [flat|nested] 3+ messages in thread* Re: [PATCH] virtio_blk: validate zone append sector from device
2026-11-20 6:30 [PATCH] virtio_blk: validate zone append sector from device Annie Kim
@ 2026-11-20 6:47 ` sashiko-bot
2026-11-20 13:29 ` Damien Le Moal
1 sibling, 0 replies; 3+ messages in thread
From: sashiko-bot @ 2026-11-20 6:47 UTC (permalink / raw)
To: Annie Kim; +Cc: virtualization, Michael S. Tsirkin, Eugenio Perez
Thank you for your contribution! Sashiko AI review found 1 potential
issue to consider:
- [Medium] virtio_blk: double virtblk_update_zone_append_sector() call
in virtblk_poll()
It also noticed 1 related pre-existing issue this patch does not
introduce:
- [Critical] virtio-blk: null pointer dereference in virtblk_poll() [1]
--
commit 9e4aa32cc96237b908b67c8c0ca12ebb3020f3a1
Author: Hyokyung Kim <pulpannie@gmail.com>
virtio_blk: validate zone append sector from device
This commit validates the device-reported starting sector on
REQ_OP_ZONE_APPEND completions in both the normal and polled completion
paths before updating req->__sector, failing out-of-range completions with
BLK_STS_IOERR.
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 2b462005cdb68..8f88570b4f0e5 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -331,15 +331,45 @@ static inline u8 virtblk_vbr_status(struct virtblk_req *vbr)
> return *((u8 *)&vbr->in_hdr + vbr->in_hdr_len - 1);
> }
>
> -static inline void virtblk_update_zone_append_sector(struct request *req,
> - struct virtblk_req *vbr)
> +#ifdef CONFIG_BLK_DEV_ZONED
> +static inline blk_status_t
> +virtblk_update_zone_append_sector(struct request *req,
> + struct virtblk_req *vbr,
> + blk_status_t status)
> {
> struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
> + sector_t req_start, sector, zone_end;
> + unsigned int nr_sectors;
> +
> + if (req_op(req) != REQ_OP_ZONE_APPEND || status != BLK_STS_OK)
> + return status;
> +
> + req_start = blk_rq_pos(req);
> + sector = virtio64_to_cpu(vblk->vdev, vbr->in_hdr.zone_append.sector);
> + nr_sectors = blk_rq_sectors(req);
> + zone_end = req_start + disk_zone_capacity(vblk->disk, req_start);
> +
> + if (zone_end < req_start || sector < req_start || sector >= zone_end ||
> + nr_sectors > zone_end - sector) {
> + dev_err_ratelimited(&vblk->vdev->dev,
> + "zone append sector %llu + %u outside valid range [%llu, %llu)\n",
> + (u64)sector, nr_sectors,
> + (u64)req_start, (u64)zone_end);
> + return BLK_STS_IOERR;
> + }
>
> - if (req_op(req) == REQ_OP_ZONE_APPEND)
> - req->__sector = virtio64_to_cpu(vblk->vdev,
> - vbr->in_hdr.zone_append.sector);
> + req->__sector = sector;
> + return status;
> }
> +#else
> +static inline blk_status_t
> +virtblk_update_zone_append_sector(struct request *req,
> + struct virtblk_req *vbr,
> + blk_status_t status)
> +{
> + return status;
> +}
> +#endif
>
> static inline void virtblk_request_done(struct request *req)
> {
> @@ -349,7 +379,7 @@ static inline void virtblk_request_done(struct request *req)
> virtblk_unmap_data(req, vbr);
> virtblk_cleanup_cmd(req);
>
> - virtblk_update_zone_append_sector(req, vbr);
> + status = virtblk_update_zone_append_sector(req, vbr, status);
>
> blk_mq_end_request(req, status);
> }
[ ... ]
> @@ -1260,11 +1289,14 @@ static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
>
> while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
> struct request *req = blk_mq_rq_from_pdu(vbr);
> - u8 status = virtblk_vbr_status(vbr);
> + blk_status_t status;
> +
> + status = virtblk_result(virtblk_vbr_status(vbr));
> + status = virtblk_update_zone_append_sector(req, vbr, status);
>
> found++;
> if (!blk_mq_complete_request_remote(req) &&
> - !blk_mq_add_to_batch(req, iob, status != VIRTIO_BLK_S_OK,
> + !blk_mq_add_to_batch(req, iob, status != BLK_STS_OK,
> virtblk_complete_batch))
> virtblk_request_done(req);
> }
[Severity: Medium]
Does calling virtblk_update_zone_append_sector() in virtblk_poll() before
falling back to virtblk_request_done() run the validation and update a
second time after req->__sector has already been overwritten?
When a polled REQ_OP_ZONE_APPEND request is not batched (for example when
an I/O scheduler is attached so RQF_SCHED_TAGS is set, when req->end_io is
set, or when validation itself returns BLK_STS_IOERR), virtblk_poll()
calls virtblk_request_done(), which recomputes status from vbr and calls
virtblk_update_zone_append_sector() again with req_start = blk_rq_pos(req)
now equal to the updated mid-zone sector rather than the zone start:
include/linux/blkdev.h:disk_zone_capacity() {
sector_t zone_sectors = disk->queue->limits.chunk_sectors;
if (sector + zone_sectors >= get_capacity(disk))
return disk->last_zone_capacity;
return disk->zone_capacity;
}
For a valid append in the second-to-last zone landing past the start of
that zone on a disk with a smaller last zone, sector + zone_sectors >=
get_capacity(disk) is true on the second call, so disk_zone_capacity()
returns last_zone_capacity and falsely fails the request with
BLK_STS_IOERR when nr_sectors > last_zone_capacity. Even when all zones
have equal capacity, the second call shifts zone_end to sector +
zone_capacity and re-reads vbr->in_hdr from device memory, and any invalid
polled append logs dev_err_ratelimited() twice.
[Severity: Critical]
This problem wasn't introduced by this patch, but can virtblk_poll() hit a
NULL pointer dereference [1] on vq->lock if polled I/O runs after
init_vq() fails during device resume?
When virtio_find_vqs() fails in init_vq() (for instance if the hypervisor
reduces the ring size on resume), init_vq() frees vblk->vqs and sets it to
NULL while leaving the request queue quiesced but not frozen. Because
blk_mq_poll() does not check whether the queue is quiesced, a userspace
polled read (such as preadv2() with RWF_HIPRI) calls virtblk_poll(), where
get_virtio_blk_vq() indexes the NULL vblk->vqs pointer and
spin_lock_irqsave(&vq->lock, flags) dereferences it.
[1] https://sashiko.dev/#/bug/linux-c7b06d8d-b2d5-4a94-acf6-9b1cf7480add
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261120063039.4155597-1-pulpannie@gmail.com?part=1
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] virtio_blk: validate zone append sector from device
2026-11-20 6:30 [PATCH] virtio_blk: validate zone append sector from device Annie Kim
2026-11-20 6:47 ` sashiko-bot
@ 2026-11-20 13:29 ` Damien Le Moal
1 sibling, 0 replies; 3+ messages in thread
From: Damien Le Moal @ 2026-11-20 13:29 UTC (permalink / raw)
To: Annie Kim, Michael S . Tsirkin, Jason Wang
Cc: Jens Axboe, virtualization, linux-block, linux-kernel, stable
On 2026/11/19 22:30, Annie Kim wrote:
> From: Hyokyung Kim <pulpannie@gmail.com>
>
> The device reports the starting sector for REQ_OP_ZONE_APPEND completions,
> and virtio_blk copies it into req->__sector without checking that it
> still refers to the requested zone. A buggy or malicious device can
> therefore make upper layers believe the append landed at an arbitrary
> location.
>
> Validate successful zone append completions before updating
> req->__sector. The returned sector must not precede the request start
> and the entire request must fit within the zone writable capacity.
> Reject invalid completions with BLK_STS_IOERR.
>
> The polling path may batch successful completions, so perform the same
> validation before admitting a request to the batch. Invalid append
> completions then fall back to the normal completion path and are
> completed with an error instead of being reported as successful.
>
> Fixes: 95bfec41bd3d ("virtio-blk: add support for zoned block devices")
> Assisted-by: Codex:GPT-5
This tag is not necessary.
> Signed-off-by: Hyokyung Kim <pulpannie@gmail.com>
> ---
> Tested on v7.4-rc3+ using modified QEMU backend that returned incorrect
> zone-append sectors on a memory-backed null_blk zoned device.
>
> drivers/block/virtio_blk.c | 50 +++++++++++++++++++++++++++++++-------
> 1 file changed, 41 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 2b462005c..8f88570b4 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -331,15 +331,45 @@ static inline u8 virtblk_vbr_status(struct virtblk_req *vbr)
> return *((u8 *)&vbr->in_hdr + vbr->in_hdr_len - 1);
> }
>
> -static inline void virtblk_update_zone_append_sector(struct request *req,
> - struct virtblk_req *vbr)
> +#ifdef CONFIG_BLK_DEV_ZONED
> +static inline blk_status_t
> +virtblk_update_zone_append_sector(struct request *req,
> + struct virtblk_req *vbr,
> + blk_status_t status)
> {
> struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
> + sector_t req_start, sector, zone_end;
> + unsigned int nr_sectors;
> +
> + if (req_op(req) != REQ_OP_ZONE_APPEND || status != BLK_STS_OK)
> + return status;
I would move this if outside of the function and call this function under this
if. That would avoid passing status, make this function a void one, and avoid
unnecessary function calls for commands other than zone append when the inline
is not honored by the compiler.
> +
> + req_start = blk_rq_pos(req);
> + sector = virtio64_to_cpu(vblk->vdev, vbr->in_hdr.zone_append.sector);
> + nr_sectors = blk_rq_sectors(req);
req_start and nr_sectors could be initialized when declared.
Other than this, this looks fine to me.
> + zone_end = req_start + disk_zone_capacity(vblk->disk, req_start);
> +
> + if (zone_end < req_start || sector < req_start || sector >= zone_end ||
> + nr_sectors > zone_end - sector) {
> + dev_err_ratelimited(&vblk->vdev->dev,
> + "zone append sector %llu + %u outside valid range [%llu, %llu)\n",
> + (u64)sector, nr_sectors,
> + (u64)req_start, (u64)zone_end);
> + return BLK_STS_IOERR;
> + }
>
> - if (req_op(req) == REQ_OP_ZONE_APPEND)
> - req->__sector = virtio64_to_cpu(vblk->vdev,
> - vbr->in_hdr.zone_append.sector);
> + req->__sector = sector;
> + return status;
> }
> +#else
> +static inline blk_status_t
> +virtblk_update_zone_append_sector(struct request *req,
> + struct virtblk_req *vbr,
> + blk_status_t status)
> +{
> + return status;
> +}
> +#endif
>
> static inline void virtblk_request_done(struct request *req)
> {
> @@ -349,7 +379,7 @@ static inline void virtblk_request_done(struct request *req)
> virtblk_unmap_data(req, vbr);
> virtblk_cleanup_cmd(req);
>
> - virtblk_update_zone_append_sector(req, vbr);
> + status = virtblk_update_zone_append_sector(req, vbr, status);
>
> blk_mq_end_request(req, status);
> }
> @@ -1242,7 +1272,6 @@ static void virtblk_complete_batch(struct io_comp_batch *iob)
>
> virtblk_unmap_data(req, vbr);
> virtblk_cleanup_cmd(req);
> - virtblk_update_zone_append_sector(req, vbr);
> }
> blk_mq_end_request_batch(iob);
> }
> @@ -1260,11 +1289,14 @@ static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
>
> while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
> struct request *req = blk_mq_rq_from_pdu(vbr);
> - u8 status = virtblk_vbr_status(vbr);
> + blk_status_t status;
> +
> + status = virtblk_result(virtblk_vbr_status(vbr));
> + status = virtblk_update_zone_append_sector(req, vbr, status);
>
> found++;
> if (!blk_mq_complete_request_remote(req) &&
> - !blk_mq_add_to_batch(req, iob, status != VIRTIO_BLK_S_OK,
> + !blk_mq_add_to_batch(req, iob, status != BLK_STS_OK,
> virtblk_complete_batch))
> virtblk_request_done(req);
> }
--
Damien Le Moal
Western Digital Research
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-11-20 13:29 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-20 6:30 [PATCH] virtio_blk: validate zone append sector from device Annie Kim
2026-11-20 6:47 ` sashiko-bot
2026-11-20 13:29 ` Damien Le Moal
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.