[PATCH net v2 0/2] vsock/virtio: Fix data loss/disclosure due to joining of non-linear skb

[PATCH net v2 0/2] vsock/virtio: Fix data loss/disclosure due to joining of non-linear skb

[Michal Luczaj]

Loopback transport coalesces some skbs too eagerly. Handling a zerocopy
(non-linear) skb as a linear one leads to skb data loss and kernel memory
disclosure.

Plug the loss/leak by allowing only linear skb join. Provide a test.

Signed-off-by: Michal Luczaj <mhal@rbox.co>
---
Changes in v2:
- Point out virtio transports affected/unaffected [Stefano]
- Move and comment skb_is_nonlinear() check [Stefano]
- Describe test logic in detail, mention "virtio" in the name [Stefano]
- Test: call poll() with a proper timeout, drop recv_verify()
- Link to v1: https://lore.kernel.org/r/20260108-vsock-recv-coalescence-v1-0-26f97bb9a99b@rbox.co

---
Michal Luczaj (2):
      vsock/virtio: Coalesce only linear skb
      vsock/test: Add test for a linear and non-linear skb getting coalesced

 net/vmw_vsock/virtio_transport_common.c   |  6 ++-
 tools/testing/vsock/vsock_test.c          |  5 +++
 tools/testing/vsock/vsock_test_zerocopy.c | 74 +++++++++++++++++++++++++++++++
 tools/testing/vsock/vsock_test_zerocopy.h |  3 ++
 4 files changed, 86 insertions(+), 2 deletions(-)
---
base-commit: ffe4ccd359d006eba559cb1a3c6113144b7fb38c
change-id: 20260103-vsock-recv-coalescence-38178fafd10c

Best regards,
-- 
Michal Luczaj <mhal@rbox.co>

[PATCH net v2 1/2] vsock/virtio: Coalesce only linear skb

[Michal Luczaj]

vsock/virtio common tries to coalesce buffers in rx queue: if a linear skb
(with a spare tail room) is followed by a small skb (length limited by
GOOD_COPY_LEN = 128), an attempt is made to join them.

Since the introduction of MSG_ZEROCOPY support, assumption that a small skb
will always be linear is incorrect. In the zerocopy case, data is lost and
the linear skb is appended with uninitialized kernel memory.

Of all 3 supported virtio-based transports, only loopback-transport is
affected. G2H virtio-transport rx queue operates on explicitly linear skbs;
see virtio_vsock_alloc_linear_skb() in virtio_vsock_rx_fill(). H2G
vhost-transport may allocate non-linear skbs, but only for sizes that are
not considered for coalescence; see PAGE_ALLOC_COSTLY_ORDER in
virtio_vsock_alloc_skb().

Ensure only linear skbs are coalesced. Note that skb_tailroom(last_skb) > 0
guarantees last_skb is linear.

Fixes: 581512a6dc93 ("vsock/virtio: MSG_ZEROCOPY flag support")
Signed-off-by: Michal Luczaj <mhal@rbox.co>
---
 net/vmw_vsock/virtio_transport_common.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
index dcc8a1d5851e..26b979ad71f0 100644
--- a/net/vmw_vsock/virtio_transport_common.c
+++ b/net/vmw_vsock/virtio_transport_common.c
@@ -1359,9 +1359,11 @@ virtio_transport_recv_enqueue(struct vsock_sock *vsk,
 
 	/* Try to copy small packets into the buffer of last packet queued,
 	 * to avoid wasting memory queueing the entire buffer with a small
-	 * payload.
+	 * payload. Skip non-linear (e.g. zerocopy) skbs; these carry payload
+	 * in skb_shinfo.
 	 */
-	if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
+	if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue) &&
+	    !skb_is_nonlinear(skb)) {
 		struct virtio_vsock_hdr *last_hdr;
 		struct sk_buff *last_skb;
 

-- 
2.52.0

[PATCH net v2 2/2] vsock/test: Add test for a linear and non-linear skb getting coalesced

[Michal Luczaj]

Loopback transport can mangle data in rx queue when a linear skb is
followed by a small MSG_ZEROCOPY packet.

To exercise the logic, send out two packets: a weirdly sized one (to ensure
some spare tail room in the skb) and a zerocopy one that's small enough to
fit in the spare room of its predecessor. Then, wait for both to land in
the rx queue, and check the data received. Faulty packets merger manifests
itself by corrupting payload of the later packet.

Signed-off-by: Michal Luczaj <mhal@rbox.co>
---
 tools/testing/vsock/vsock_test.c          |  5 +++
 tools/testing/vsock/vsock_test_zerocopy.c | 74 +++++++++++++++++++++++++++++++
 tools/testing/vsock/vsock_test_zerocopy.h |  3 ++
 3 files changed, 82 insertions(+)

diff --git a/tools/testing/vsock/vsock_test.c b/tools/testing/vsock/vsock_test.c
index bbe3723babdc..27e39354499a 100644
--- a/tools/testing/vsock/vsock_test.c
+++ b/tools/testing/vsock/vsock_test.c
@@ -2403,6 +2403,11 @@ static struct test_case test_cases[] = {
 		.run_client = test_stream_accepted_setsockopt_client,
 		.run_server = test_stream_accepted_setsockopt_server,
 	},
+	{
+		.name = "SOCK_STREAM virtio MSG_ZEROCOPY coalescence corruption",
+		.run_client = test_stream_msgzcopy_mangle_client,
+		.run_server = test_stream_msgzcopy_mangle_server,
+	},
 	{},
 };
 
diff --git a/tools/testing/vsock/vsock_test_zerocopy.c b/tools/testing/vsock/vsock_test_zerocopy.c
index 9d9a6cb9614a..a31ddfc1cd0c 100644
--- a/tools/testing/vsock/vsock_test_zerocopy.c
+++ b/tools/testing/vsock/vsock_test_zerocopy.c
@@ -9,14 +9,18 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <sys/ioctl.h>
 #include <sys/mman.h>
 #include <unistd.h>
 #include <poll.h>
 #include <linux/errqueue.h>
 #include <linux/kernel.h>
+#include <linux/sockios.h>
+#include <linux/time64.h>
 #include <errno.h>
 
 #include "control.h"
+#include "timeout.h"
 #include "vsock_test_zerocopy.h"
 #include "msg_zerocopy_common.h"
 
@@ -356,3 +360,73 @@ void test_stream_msgzcopy_empty_errq_server(const struct test_opts *opts)
 	control_expectln("DONE");
 	close(fd);
 }
+
+#define GOOD_COPY_LEN	128	/* net/vmw_vsock/virtio_transport_common.c */
+
+void test_stream_msgzcopy_mangle_client(const struct test_opts *opts)
+{
+	char sbuf1[PAGE_SIZE + 1], sbuf2[GOOD_COPY_LEN];
+	unsigned long hash;
+	struct pollfd fds;
+	int fd, i;
+
+	fd = vsock_stream_connect(opts->peer_cid, opts->peer_port);
+	if (fd < 0) {
+		perror("connect");
+		exit(EXIT_FAILURE);
+	}
+
+	enable_so_zerocopy_check(fd);
+
+	memset(sbuf1, 'x', sizeof(sbuf1));
+	send_buf(fd, sbuf1, sizeof(sbuf1), 0, sizeof(sbuf1));
+
+	for (i = 0; i < sizeof(sbuf2); i++)
+		sbuf2[i] = rand() & 0xff;
+
+	send_buf(fd, sbuf2, sizeof(sbuf2), MSG_ZEROCOPY, sizeof(sbuf2));
+
+	hash = hash_djb2(sbuf2, sizeof(sbuf2));
+	control_writeulong(hash);
+
+	fds.fd = fd;
+	fds.events = 0;
+
+	if (poll(&fds, 1, TIMEOUT * MSEC_PER_SEC) != 1 ||
+	    !(fds.revents & POLLERR)) {
+		perror("poll");
+		exit(EXIT_FAILURE);
+	}
+
+	close(fd);
+}
+
+void test_stream_msgzcopy_mangle_server(const struct test_opts *opts)
+{
+	unsigned long local_hash, remote_hash;
+	char rbuf[PAGE_SIZE + 1];
+	int fd;
+
+	fd = vsock_stream_accept(VMADDR_CID_ANY, opts->peer_port, NULL);
+	if (fd < 0) {
+		perror("accept");
+		exit(EXIT_FAILURE);
+	}
+
+	/* Wait, don't race the (buggy) skbs coalescence. */
+	vsock_ioctl_int(fd, SIOCINQ, PAGE_SIZE + 1 + GOOD_COPY_LEN);
+
+	/* Discard the first packet. */
+	recv_buf(fd, rbuf, PAGE_SIZE + 1, 0, PAGE_SIZE + 1);
+
+	recv_buf(fd, rbuf, GOOD_COPY_LEN, 0, GOOD_COPY_LEN);
+	remote_hash = control_readulong();
+	local_hash = hash_djb2(rbuf, GOOD_COPY_LEN);
+
+	if (local_hash != remote_hash) {
+		fprintf(stderr, "Data received corrupted\n");
+		exit(EXIT_FAILURE);
+	}
+
+	close(fd);
+}
diff --git a/tools/testing/vsock/vsock_test_zerocopy.h b/tools/testing/vsock/vsock_test_zerocopy.h
index 3ef2579e024d..d46c91a69f16 100644
--- a/tools/testing/vsock/vsock_test_zerocopy.h
+++ b/tools/testing/vsock/vsock_test_zerocopy.h
@@ -12,4 +12,7 @@ void test_seqpacket_msgzcopy_server(const struct test_opts *opts);
 void test_stream_msgzcopy_empty_errq_client(const struct test_opts *opts);
 void test_stream_msgzcopy_empty_errq_server(const struct test_opts *opts);
 
+void test_stream_msgzcopy_mangle_client(const struct test_opts *opts);
+void test_stream_msgzcopy_mangle_server(const struct test_opts *opts);
+
 #endif /* VSOCK_TEST_ZEROCOPY_H */

-- 
2.52.0

Re: [PATCH net v2 1/2] vsock/virtio: Coalesce only linear skb

[Stefano Garzarella]

On Tue, Jan 13, 2026 at 04:08:18PM +0100, Michal Luczaj wrote:
>vsock/virtio common tries to coalesce buffers in rx queue: if a linear skb
>(with a spare tail room) is followed by a small skb (length limited by
>GOOD_COPY_LEN = 128), an attempt is made to join them.
>
>Since the introduction of MSG_ZEROCOPY support, assumption that a small skb
>will always be linear is incorrect. In the zerocopy case, data is lost and
>the linear skb is appended with uninitialized kernel memory.
>
>Of all 3 supported virtio-based transports, only loopback-transport is
>affected. G2H virtio-transport rx queue operates on explicitly linear skbs;
>see virtio_vsock_alloc_linear_skb() in virtio_vsock_rx_fill(). H2G
>vhost-transport may allocate non-linear skbs, but only for sizes that are
>not considered for coalescence; see PAGE_ALLOC_COSTLY_ORDER in
>virtio_vsock_alloc_skb().
>
>Ensure only linear skbs are coalesced. Note that skb_tailroom(last_skb) > 0
>guarantees last_skb is linear.
>
>Fixes: 581512a6dc93 ("vsock/virtio: MSG_ZEROCOPY flag support")
>Signed-off-by: Michal Luczaj <mhal@rbox.co>
>---
> net/vmw_vsock/virtio_transport_common.c | 6 ++++--
> 1 file changed, 4 insertions(+), 2 deletions(-)

Thank you for enriching the commit message!

LGTM!

Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>

>
>diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
>index dcc8a1d5851e..26b979ad71f0 100644
>--- a/net/vmw_vsock/virtio_transport_common.c
>+++ b/net/vmw_vsock/virtio_transport_common.c
>@@ -1359,9 +1359,11 @@ virtio_transport_recv_enqueue(struct vsock_sock *vsk,
>
> 	/* Try to copy small packets into the buffer of last packet queued,
> 	 * to avoid wasting memory queueing the entire buffer with a small
>-	 * payload.
>+	 * payload. Skip non-linear (e.g. zerocopy) skbs; these carry payload
>+	 * in skb_shinfo.
> 	 */
>-	if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
>+	if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue) &&
>+	    !skb_is_nonlinear(skb)) {
> 		struct virtio_vsock_hdr *last_hdr;
> 		struct sk_buff *last_skb;
>
>
>-- 
>2.52.0
>

Re: [PATCH net v2 2/2] vsock/test: Add test for a linear and non-linear skb getting coalesced

[Stefano Garzarella]

On Tue, Jan 13, 2026 at 04:08:19PM +0100, Michal Luczaj wrote:
>Loopback transport can mangle data in rx queue when a linear skb is
>followed by a small MSG_ZEROCOPY packet.
>
>To exercise the logic, send out two packets: a weirdly sized one (to ensure
>some spare tail room in the skb) and a zerocopy one that's small enough to
>fit in the spare room of its predecessor. Then, wait for both to land in
>the rx queue, and check the data received. Faulty packets merger manifests
>itself by corrupting payload of the later packet.
>
>Signed-off-by: Michal Luczaj <mhal@rbox.co>
>---
> tools/testing/vsock/vsock_test.c          |  5 +++
> tools/testing/vsock/vsock_test_zerocopy.c | 74 +++++++++++++++++++++++++++++++
> tools/testing/vsock/vsock_test_zerocopy.h |  3 ++
> 3 files changed, 82 insertions(+)

Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>

>
>diff --git a/tools/testing/vsock/vsock_test.c b/tools/testing/vsock/vsock_test.c
>index bbe3723babdc..27e39354499a 100644
>--- a/tools/testing/vsock/vsock_test.c
>+++ b/tools/testing/vsock/vsock_test.c
>@@ -2403,6 +2403,11 @@ static struct test_case test_cases[] = {
> 		.run_client = test_stream_accepted_setsockopt_client,
> 		.run_server = test_stream_accepted_setsockopt_server,
> 	},
>+	{
>+		.name = "SOCK_STREAM virtio MSG_ZEROCOPY coalescence corruption",
>+		.run_client = test_stream_msgzcopy_mangle_client,
>+		.run_server = test_stream_msgzcopy_mangle_server,
>+	},
> 	{},
> };
>
>diff --git a/tools/testing/vsock/vsock_test_zerocopy.c b/tools/testing/vsock/vsock_test_zerocopy.c
>index 9d9a6cb9614a..a31ddfc1cd0c 100644
>--- a/tools/testing/vsock/vsock_test_zerocopy.c
>+++ b/tools/testing/vsock/vsock_test_zerocopy.c
>@@ -9,14 +9,18 @@
> #include <stdio.h>
> #include <stdlib.h>
> #include <string.h>
>+#include <sys/ioctl.h>
> #include <sys/mman.h>
> #include <unistd.h>
> #include <poll.h>
> #include <linux/errqueue.h>
> #include <linux/kernel.h>
>+#include <linux/sockios.h>
>+#include <linux/time64.h>
> #include <errno.h>
>
> #include "control.h"
>+#include "timeout.h"
> #include "vsock_test_zerocopy.h"
> #include "msg_zerocopy_common.h"
>
>@@ -356,3 +360,73 @@ void test_stream_msgzcopy_empty_errq_server(const struct test_opts *opts)
> 	control_expectln("DONE");
> 	close(fd);
> }
>+
>+#define GOOD_COPY_LEN	128	/* net/vmw_vsock/virtio_transport_common.c */
>+
>+void test_stream_msgzcopy_mangle_client(const struct test_opts *opts)
>+{
>+	char sbuf1[PAGE_SIZE + 1], sbuf2[GOOD_COPY_LEN];
>+	unsigned long hash;
>+	struct pollfd fds;
>+	int fd, i;
>+
>+	fd = vsock_stream_connect(opts->peer_cid, opts->peer_port);
>+	if (fd < 0) {
>+		perror("connect");
>+		exit(EXIT_FAILURE);
>+	}
>+
>+	enable_so_zerocopy_check(fd);
>+
>+	memset(sbuf1, 'x', sizeof(sbuf1));
>+	send_buf(fd, sbuf1, sizeof(sbuf1), 0, sizeof(sbuf1));
>+
>+	for (i = 0; i < sizeof(sbuf2); i++)
>+		sbuf2[i] = rand() & 0xff;
>+
>+	send_buf(fd, sbuf2, sizeof(sbuf2), MSG_ZEROCOPY, sizeof(sbuf2));
>+
>+	hash = hash_djb2(sbuf2, sizeof(sbuf2));
>+	control_writeulong(hash);
>+
>+	fds.fd = fd;
>+	fds.events = 0;
>+
>+	if (poll(&fds, 1, TIMEOUT * MSEC_PER_SEC) != 1 ||
>+	    !(fds.revents & POLLERR)) {
>+		perror("poll");
>+		exit(EXIT_FAILURE);
>+	}
>+
>+	close(fd);
>+}
>+
>+void test_stream_msgzcopy_mangle_server(const struct test_opts *opts)
>+{
>+	unsigned long local_hash, remote_hash;
>+	char rbuf[PAGE_SIZE + 1];
>+	int fd;
>+
>+	fd = vsock_stream_accept(VMADDR_CID_ANY, opts->peer_port, NULL);
>+	if (fd < 0) {
>+		perror("accept");
>+		exit(EXIT_FAILURE);
>+	}
>+
>+	/* Wait, don't race the (buggy) skbs coalescence. */
>+	vsock_ioctl_int(fd, SIOCINQ, PAGE_SIZE + 1 + GOOD_COPY_LEN);
>+
>+	/* Discard the first packet. */
>+	recv_buf(fd, rbuf, PAGE_SIZE + 1, 0, PAGE_SIZE + 1);
>+
>+	recv_buf(fd, rbuf, GOOD_COPY_LEN, 0, GOOD_COPY_LEN);
>+	remote_hash = control_readulong();
>+	local_hash = hash_djb2(rbuf, GOOD_COPY_LEN);
>+
>+	if (local_hash != remote_hash) {
>+		fprintf(stderr, "Data received corrupted\n");
>+		exit(EXIT_FAILURE);
>+	}
>+
>+	close(fd);
>+}
>diff --git a/tools/testing/vsock/vsock_test_zerocopy.h b/tools/testing/vsock/vsock_test_zerocopy.h
>index 3ef2579e024d..d46c91a69f16 100644
>--- a/tools/testing/vsock/vsock_test_zerocopy.h
>+++ b/tools/testing/vsock/vsock_test_zerocopy.h
>@@ -12,4 +12,7 @@ void test_seqpacket_msgzcopy_server(const struct test_opts *opts);
> void test_stream_msgzcopy_empty_errq_client(const struct test_opts *opts);
> void test_stream_msgzcopy_empty_errq_server(const struct test_opts *opts);
>
>+void test_stream_msgzcopy_mangle_client(const struct test_opts *opts);
>+void test_stream_msgzcopy_mangle_server(const struct test_opts *opts);
>+
> #endif /* VSOCK_TEST_ZEROCOPY_H */
>
>-- 
>2.52.0
>

Re: [PATCH net v2 0/2] vsock/virtio: Fix data loss/disclosure due to joining of non-linear skb

[patchwork-bot+netdevbpf]

Hello:

This series was applied to netdev/net.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Tue, 13 Jan 2026 16:08:17 +0100 you wrote:
> Loopback transport coalesces some skbs too eagerly. Handling a zerocopy
> (non-linear) skb as a linear one leads to skb data loss and kernel memory
> disclosure.
> 
> Plug the loss/leak by allowing only linear skb join. Provide a test.
> 
> Signed-off-by: Michal Luczaj <mhal@rbox.co>
> 
> [...]

Here is the summary with links:
  - [net,v2,1/2] vsock/virtio: Coalesce only linear skb
    https://git.kernel.org/netdev/net/c/0386bd321d0f
  - [net,v2,2/2] vsock/test: Add test for a linear and non-linear skb getting coalesced
    https://git.kernel.org/netdev/net/c/a63e5fe09592

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html