* [PATCH net v3 1/4] virtio_net: reject stats replies for queues outside the request
2026-11-16 9:24 [PATCH net v3 0/4] virtio_net: fix out-of-bounds accesses and a heap info leak via the device stats reply Lianjie Wang
@ 2026-11-16 9:24 ` Lianjie Wang
2026-11-16 9:42 ` sashiko-bot
2026-11-16 9:24 ` [PATCH net v3 2/4] virtio_net: zero the stats reply buffer Lianjie Wang
` (2 subsequent siblings)
3 siblings, 1 reply; 9+ messages in thread
From: Lianjie Wang @ 2026-11-16 9:24 UTC (permalink / raw)
To: Michael S. Tsirkin, Jason Wang, Eugenio Pérez, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni
Cc: Lianjie Wang, Xuan Zhuo, netdev, virtualization, linux-kernel,
Michael Bommarito, Takahiro Shinagawa, stable
__virtnet_get_hw_stats() passes the vq_index of each record in the
device's statistics reply to virtnet_fill_stats() without checking it.
For ethtool -S, virtnet_fill_stats() uses vq_index as an offset into the
output array, which the ethtool core sized for curr_queue_pairs. A
record with the vq_index of a higher queue can make the driver write
past the end of the array.
For the per-queue netlink stats, the queue type also comes from
vq_index. A receive record in reply to a request for a transmit queue
makes the driver fill the struct netdev_queue_stats_tx on the stack as
if it were a struct netdev_queue_stats_rx. If the device offers
VIRTIO_NET_STATS_TYPE_RX_SPEED, this writes one counter to the eight
bytes after the tx struct.
Fix this by rejecting the reply when the vq_index of a record lies
outside the range of queues in the request. The control vq is checked
separately, since the vqs of the queue pairs beyond curr_queue_pairs lie
between that range and the control vq.
Claude Opus 5.5 was used to audit the driver and draft this patch. To
reproduce, the virtio-net device in QEMU was modified to offer device
statistics and to answer the request of ethtool -S with a record for a
nonexistent receive queue. On an unmodified kernel, KASAN reports a
write past the ethtool -S output array:
BUG: KASAN: vmalloc-out-of-bounds in virtnet_fill_stats+0x6c1/0xf20
Write of size 8 at addr ffffc90000085140 by task ethtool_stats/158
Call Trace:
virtnet_fill_stats+0x6c1/0xf20
virtnet_get_hw_stats+0x5d6/0x7c0
virtnet_get_ethtool_stats+0x11b/0x920
ethtool_get_stats+0x150/0x1f0
dev_ethtool+0xaaa/0x1be0
The buggy address belongs to a 1-page vmalloc region starting at 0xffffc90000085000 allocated at ethtool_get_stats+0x120/0x1f0
With this patch, the driver rejects the same reply. When ethtool -L
leaves queue pairs unused, the driver also rejects a record for the
first unused receive queue, while a well-formed reply fills the
statistics as before. The drivers/net/stats.py selftest passes when the
device answers with well-formed records.
Fixes: 941168f8b40e ("virtio_net: support device stats")
Fixes: d888f04c09bb ("virtio-net: support queue stat")
Cc: stable@vger.kernel.org
Signed-off-by: Lianjie Wang <lian@is.s.u-tokyo.ac.jp>
---
v3:
- No functional changes.
v2: https://lore.kernel.org/all/20261102205550.4148064-1-lian@is.s.u-tokyo.ac.jp/
- Add a comment and a sentence on why the control vq is checked apart
from the range of the other queues.
- Add the KASAN report and a test with unused queue pairs.
v1: https://lore.kernel.org/all/20261022182710.2055771-1-karin0.zst@gmail.com/
drivers/net/virtio_net.c | 13 +++++++++++--
1 file changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index ed20e76fbc4c..57d7a98ecb57 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -4973,7 +4973,8 @@ static void virtnet_fill_stats(struct virtnet_info *vi, u32 qid,
static int __virtnet_get_hw_stats(struct virtnet_info *vi,
struct virtnet_stats_ctx *ctx,
struct virtio_net_ctrl_queue_stats *req,
- int req_size, void *reply, int res_size)
+ int req_size, void *reply, int res_size,
+ int first_vq, int last_vq, bool enable_cvq)
{
struct virtio_net_stats_reply_hdr *hdr;
struct scatterlist sgs_in, sgs_out;
@@ -4994,6 +4995,13 @@ static int __virtnet_get_hw_stats(struct virtnet_info *vi,
for (p = reply; p - reply < res_size; p += le16_to_cpu(hdr->size)) {
hdr = p;
qid = le16_to_cpu(hdr->vq_index);
+ /* The vqs of the queue pairs beyond curr_queue_pairs lie
+ * between last_vq and the control vq.
+ */
+ if ((qid < first_vq || qid > last_vq) &&
+ !(enable_cvq && qid == vi->max_queue_pairs * 2))
+ return -EINVAL;
+
virtnet_fill_stats(vi, qid, ctx, p, false, hdr->type);
}
@@ -5073,7 +5081,8 @@ static int virtnet_get_hw_stats(struct virtnet_info *vi,
if (enable_cvq)
virtnet_make_stat_req(vi, ctx, req, vi->max_queue_pairs * 2, &j);
- ok = __virtnet_get_hw_stats(vi, ctx, req, sizeof(*req) * j, reply, res_size);
+ ok = __virtnet_get_hw_stats(vi, ctx, req, sizeof(*req) * j, reply,
+ res_size, first_vq, last_vq, enable_cvq);
kfree(req);
kfree(reply);
--
2.56.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* Re: [PATCH net v3 1/4] virtio_net: reject stats replies for queues outside the request
2026-11-16 9:24 ` [PATCH net v3 1/4] virtio_net: reject stats replies for queues outside the request Lianjie Wang
@ 2026-11-16 9:42 ` sashiko-bot
0 siblings, 0 replies; 9+ messages in thread
From: sashiko-bot @ 2026-11-16 9:42 UTC (permalink / raw)
To: Lianjie Wang; +Cc: virtualization, Michael S. Tsirkin, Eugenio Perez
> __virtnet_get_hw_stats() passes the vq_index of each record in the
> device's statistics reply to virtnet_fill_stats() without checking it.
>
> For ethtool -S, virtnet_fill_stats() uses vq_index as an offset into the
> output array, which the ethtool core sized for curr_queue_pairs. A
> record with the vq_index of a higher queue can make the driver write
> past the end of the array.
>
> For the per-queue netlink stats, the queue type also comes from
> vq_index. A receive record in reply to a request for a transmit queue
> makes the driver fill the struct netdev_queue_stats_tx on the stack as
> if it were a struct netdev_queue_stats_rx. If the device offers
> VIRTIO_NET_STATS_TYPE_RX_SPEED, this writes one counter to the eight
> bytes after the tx struct.
>
> [ ... ]
>
> Fixes: 941168f8b40e ("virtio_net: support device stats")
> Fixes: d888f04c09bb ("virtio-net: support queue stat")
> Cc: stable@vger.kernel.org
> Signed-off-by: Lianjie Wang <lian@is.s.u-tokyo.ac.jp>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116092445.3391334-1-lian@is.s.u-tokyo.ac.jp?part=1
^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCH net v3 2/4] virtio_net: zero the stats reply buffer
2026-11-16 9:24 [PATCH net v3 0/4] virtio_net: fix out-of-bounds accesses and a heap info leak via the device stats reply Lianjie Wang
2026-11-16 9:24 ` [PATCH net v3 1/4] virtio_net: reject stats replies for queues outside the request Lianjie Wang
@ 2026-11-16 9:24 ` Lianjie Wang
2026-11-16 9:42 ` sashiko-bot
2026-11-16 9:24 ` [PATCH net v3 3/4] virtio_net: keep reads of stats records within the " Lianjie Wang
2026-11-16 9:24 ` [PATCH net v3 4/4] virtio_net: report a failed stats command as an error Lianjie Wang
3 siblings, 1 reply; 9+ messages in thread
From: Lianjie Wang @ 2026-11-16 9:24 UTC (permalink / raw)
To: Michael S. Tsirkin, Jason Wang, Eugenio Pérez, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni
Cc: Lianjie Wang, Xuan Zhuo, netdev, virtualization, linux-kernel,
Michael Bommarito, Takahiro Shinagawa, stable
virtnet_get_hw_stats() allocates the reply buffer of the device
statistics command with kmalloc(), and __virtnet_get_hw_stats() parses
the whole buffer. When the device writes less than the buffer holds,
stale heap bytes are reported as counters through ethtool -S and the
per-queue netlink statistics.
Allocate the buffer with kzalloc(), so that the counters the device
didn't write are zeroed. An unwritten record header then has size 0, so
reject it when walking on that header to avoid an infinite loop.
The used length cannot replace the zeroing here, since the device can
also report a larger value than it wrote.
Claude Opus 5.5 was used to audit the driver and draft this patch. To
reproduce, the virtio-net device in QEMU was modified to offer device
statistics and to leave the counters of a receive record unwritten. On
an unmodified kernel, ethtool -S reported kernel pointers as rx0
counters. With this patch, the counters are zeroed, and a reply with
missing records or a record of size 0 is rejected. The
drivers/net/stats.py selftest passes when the device answers with
well-formed records.
Reported-by: Michael Bommarito <michael.bommarito@gmail.com>
Link: https://lore.kernel.org/all/20260711150754.2918392-1-michael.bommarito@gmail.com/
Reported-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/all/20261110001659.1654289-2-kuba@kernel.org/
Fixes: 941168f8b40e ("virtio_net: support device stats")
Cc: stable@vger.kernel.org
Signed-off-by: Lianjie Wang <lian@is.s.u-tokyo.ac.jp>
---
v3:
- New patch.
drivers/net/virtio_net.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 57d7a98ecb57..498014773242 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -4994,6 +4994,9 @@ static int __virtnet_get_hw_stats(struct virtnet_info *vi,
for (p = reply; p - reply < res_size; p += le16_to_cpu(hdr->size)) {
hdr = p;
+ if (!hdr->size)
+ return -EINVAL;
+
qid = le16_to_cpu(hdr->vq_index);
/* The vqs of the queue pairs beyond curr_queue_pairs lie
* between last_vq and the control vq.
@@ -5068,7 +5071,7 @@ static int virtnet_get_hw_stats(struct virtnet_info *vi,
if (!req)
return -ENOMEM;
- reply = kmalloc(res_size, GFP_KERNEL);
+ reply = kzalloc(res_size, GFP_KERNEL);
if (!reply) {
kfree(req);
return -ENOMEM;
--
2.56.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* Re: [PATCH net v3 2/4] virtio_net: zero the stats reply buffer
2026-11-16 9:24 ` [PATCH net v3 2/4] virtio_net: zero the stats reply buffer Lianjie Wang
@ 2026-11-16 9:42 ` sashiko-bot
0 siblings, 0 replies; 9+ messages in thread
From: sashiko-bot @ 2026-11-16 9:42 UTC (permalink / raw)
To: Lianjie Wang; +Cc: virtualization, Michael S. Tsirkin, Eugenio Perez
> virtnet_get_hw_stats() allocates the reply buffer of the device
> statistics command with kmalloc(), and __virtnet_get_hw_stats() parses
> the whole buffer. When the device writes less than the buffer holds,
> stale heap bytes are reported as counters through ethtool -S and the
> per-queue netlink statistics.
>
> Allocate the buffer with kzalloc(), so that the counters the device
> didn't write are zeroed. An unwritten record header then has size 0, so
> reject it when walking on that header to avoid an infinite loop.
>
> The used length cannot replace the zeroing here, since the device can
> also report a larger value than it wrote.
>
> Claude Opus 5.5 was used to audit the driver and draft this patch. To
> reproduce, the virtio-net device in QEMU was modified to offer device
> statistics and to leave the counters of a receive record unwritten. On
> an unmodified kernel, ethtool -S reported kernel pointers as rx0
> counters. With this patch, the counters are zeroed, and a reply with
> missing records or a record of size 0 is rejected. The
> drivers/net/stats.py selftest passes when the device answers with
> well-formed records.
>
> Reported-by: Michael Bommarito <michael.bommarito@gmail.com>
> Link: https://lore.kernel.org/all/20260711150754.2918392-1-michael.bommarito@gmail.com/
> Reported-by: Jakub Kicinski <kuba@kernel.org>
> Link: https://lore.kernel.org/all/20261110001659.1654289-2-kuba@kernel.org/
> Fixes: 941168f8b40e ("virtio_net: support device stats")
> Cc: stable@vger.kernel.org
> Signed-off-by: Lianjie Wang <lian@is.s.u-tokyo.ac.jp>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116092445.3391334-1-lian@is.s.u-tokyo.ac.jp?part=2
^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCH net v3 3/4] virtio_net: keep reads of stats records within the reply buffer
2026-11-16 9:24 [PATCH net v3 0/4] virtio_net: fix out-of-bounds accesses and a heap info leak via the device stats reply Lianjie Wang
2026-11-16 9:24 ` [PATCH net v3 1/4] virtio_net: reject stats replies for queues outside the request Lianjie Wang
2026-11-16 9:24 ` [PATCH net v3 2/4] virtio_net: zero the stats reply buffer Lianjie Wang
@ 2026-11-16 9:24 ` Lianjie Wang
2026-11-16 9:42 ` sashiko-bot
2026-11-16 9:24 ` [PATCH net v3 4/4] virtio_net: report a failed stats command as an error Lianjie Wang
3 siblings, 1 reply; 9+ messages in thread
From: Lianjie Wang @ 2026-11-16 9:24 UTC (permalink / raw)
To: Michael S. Tsirkin, Jason Wang, Eugenio Pérez, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni
Cc: Lianjie Wang, Xuan Zhuo, netdev, virtualization, linux-kernel,
Michael Bommarito, Takahiro Shinagawa, stable
__virtnet_get_hw_stats() only checks that each record of the statistics
reply starts within the reply buffer, while virtnet_fill_stats() reads
the whole payload of the record's type, so a record near the end of the
buffer can make the driver read past it and report the heap bytes
through ethtool -S and the per-queue netlink statistics.
Fix this by reserving zeroed space of the largest ctx->size[] after the
reply, so the buffer can hold any type of payload.
Records of an unexpected size are still parsed, since every record
starts within the reply and cannot cause out-of-bounds reads, and
rejecting them with checks per type was judged costly to maintain [1].
Claude Opus 5.5 was used to audit the driver and draft this patch. To
reproduce, the virtio-net device in QEMU was modified to offer device
statistics and to end its reply with a truncated receive record. On an
unmodified kernel, KASAN reports:
BUG: KASAN: slab-out-of-bounds in virtnet_fill_stats+0x66d/0xf20
Read of size 8 at addr ffff8880036f5940 by task ethtool_stats/159
Call Trace:
virtnet_fill_stats+0x66d/0xf20
virtnet_get_hw_stats+0x5d6/0x7c0
virtnet_get_ethtool_stats+0x11b/0x920
ethtool_get_stats+0x150/0x1f0
dev_ethtool+0xaaa/0x1be0
Allocated by task 159:
__kmalloc_noprof+0x1b7/0x4a0
virtnet_get_hw_stats+0x25e/0x7c0
The buggy address is located 8 bytes to the right of
allocated 56-byte region [ffff8880036f5900, ffff8880036f5938)
With this patch, the same reply triggers no KASAN report. The
drivers/net/stats.py selftest passes when the device answers with
well-formed records.
Reported-by: Michael Bommarito <michael.bommarito@gmail.com>
Link: https://lore.kernel.org/all/20260711150754.2918392-1-michael.bommarito@gmail.com/ [1]
Reported-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/all/20261110001659.1654289-2-kuba@kernel.org/
Fixes: 941168f8b40e ("virtio_net: support device stats")
Cc: stable@vger.kernel.org
Signed-off-by: Lianjie Wang <lian@is.s.u-tokyo.ac.jp>
---
v3:
- New patch.
drivers/net/virtio_net.c | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 498014773242..b9b6c915b6f3 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -5071,7 +5071,13 @@ static int virtnet_get_hw_stats(struct virtnet_info *vi,
if (!req)
return -ENOMEM;
- reply = kzalloc(res_size, GFP_KERNEL);
+ /* virtnet_fill_stats() parses the whole payload of a record's type,
+ * so reserve zeroed space after the reply to hold any payload.
+ */
+ reply = kzalloc(res_size + max3(ctx->size[VIRTNET_Q_TYPE_RX],
+ ctx->size[VIRTNET_Q_TYPE_TX],
+ ctx->size[VIRTNET_Q_TYPE_CQ]),
+ GFP_KERNEL);
if (!reply) {
kfree(req);
return -ENOMEM;
--
2.56.1
^ permalink raw reply related [flat|nested] 9+ messages in thread* Re: [PATCH net v3 3/4] virtio_net: keep reads of stats records within the reply buffer
2026-11-16 9:24 ` [PATCH net v3 3/4] virtio_net: keep reads of stats records within the " Lianjie Wang
@ 2026-11-16 9:42 ` sashiko-bot
0 siblings, 0 replies; 9+ messages in thread
From: sashiko-bot @ 2026-11-16 9:42 UTC (permalink / raw)
To: Lianjie Wang; +Cc: virtualization, Michael S. Tsirkin, Eugenio Perez
> __virtnet_get_hw_stats() only checks that each record of the statistics
> reply starts within the reply buffer, while virtnet_fill_stats() reads
> the whole payload of the record's type, so a record near the end of the
> buffer can make the driver read past it and report the heap bytes
> through ethtool -S and the per-queue netlink statistics.
>
> Fix this by reserving zeroed space of the largest ctx->size[] after the
> reply, so the buffer can hold any type of payload.
>
> Records of an unexpected size are still parsed, since every record
> starts within the reply and cannot cause out-of-bounds reads, and
> rejecting them with checks per type was judged costly to maintain [1].
>
> Claude Opus 5.5 was used to audit the driver and draft this patch. To
> reproduce, the virtio-net device in QEMU was modified to offer device
> [ ... ]
> Reported-by: Jakub Kicinski <kuba@kernel.org>
> Link: https://lore.kernel.org/all/20261110001659.1654289-2-kuba@kernel.org/
> Fixes: 941168f8b40e ("virtio_net: support device stats")
> Cc: stable@vger.kernel.org
> Signed-off-by: Lianjie Wang <lian@is.s.u-tokyo.ac.jp>
Sashiko has reviewed this patch and found no issues.
--
Sashiko AI review · https://sashiko.dev/#/patchset/20261116092445.3391334-1-lian@is.s.u-tokyo.ac.jp?part=3
^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCH net v3 4/4] virtio_net: report a failed stats command as an error
2026-11-16 9:24 [PATCH net v3 0/4] virtio_net: fix out-of-bounds accesses and a heap info leak via the device stats reply Lianjie Wang
` (2 preceding siblings ...)
2026-11-16 9:24 ` [PATCH net v3 3/4] virtio_net: keep reads of stats records within the " Lianjie Wang
@ 2026-11-16 9:24 ` Lianjie Wang
2026-11-16 9:42 ` sashiko-bot
3 siblings, 1 reply; 9+ messages in thread
From: Lianjie Wang @ 2026-11-16 9:24 UTC (permalink / raw)
To: Michael S. Tsirkin, Jason Wang, Eugenio Pérez, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni
Cc: Lianjie Wang, Xuan Zhuo, netdev, virtualization, linux-kernel,
Michael Bommarito, Takahiro Shinagawa
__virtnet_get_hw_stats() returns 0 when the VIRTIO_NET_CTRL_STATS_GET
command fails, so ethtool -S reports every hardware counter as 0
without the "Failed to get hw stats." warning.
Return -EINVAL instead, as virtnet_probe() does for
VIRTIO_NET_CTRL_STATS_QUERY.
Claude Opus 5.5 was used to draft this patch. To test, the virtio-net
device in QEMU was modified to offer device statistics and to fail the
command. With this patch, ethtool -S triggers the warning.
Reported-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/all/20261110001659.1654289-2-kuba@kernel.org/
Fixes: 941168f8b40e ("virtio_net: support device stats")
Signed-off-by: Lianjie Wang <lian@is.s.u-tokyo.ac.jp>
---
v3:
- New patch.
drivers/net/virtio_net.c | 9 +++------
1 file changed, 3 insertions(+), 6 deletions(-)
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index b9b6c915b6f3..185ac95b9b1b 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -4980,17 +4980,14 @@ static int __virtnet_get_hw_stats(struct virtnet_info *vi,
struct scatterlist sgs_in, sgs_out;
void *p;
u32 qid;
- int ok;
sg_init_one(&sgs_out, req, req_size);
sg_init_one(&sgs_in, reply, res_size);
- ok = virtnet_send_command_reply(vi, VIRTIO_NET_CTRL_STATS,
+ if (!virtnet_send_command_reply(vi, VIRTIO_NET_CTRL_STATS,
VIRTIO_NET_CTRL_STATS_GET,
- &sgs_out, &sgs_in);
-
- if (!ok)
- return ok;
+ &sgs_out, &sgs_in))
+ return -EINVAL;
for (p = reply; p - reply < res_size; p += le16_to_cpu(hdr->size)) {
hdr = p;
--
2.56.1
^ permalink raw reply related [flat|nested] 9+ messages in thread