Skip to contentSkip to conversation

net/packet: fix TOCTOU race on mmap'd vnet_hdr in tpacket_snd()

PATCH In review Zero Mark 2 participants 4 messages Apr 17 1 message not in the archive

Subsystems (3)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): CAHOBGNBvZOXGzzMDuHWw1RrRvbg4TZVH34jVDhc1nkHbW_URXA@mail.gmail.com

Zero Mark

author
#1 lore raw

PATCH 1/1 net/packet: fix TOCTOU race on mmap'd vnet_hdr in tpacket_snd()

In tpacket_snd(), when PACKET_VNET_HDR is enabled, vnet_hdr points directly into the mmap'd TX ring buffer shared with userspace. The kernel validates the header via __packet_snd_vnet_parse() but then re-reads all fields later in virtio_net_hdr_to_skb(). A concurrent userspace thread can modify the vnet_hdr fields (gso_type, gso_size, flags, csum_start, csum_offset) between validation and use, bypassing all safety checks.

This can lead to:
- Out-of-bounds checksum writes via crafted csum_start/csum_offset
- Malicious GSO segmentation parameters
- Kernel memory corruption and potential local privilege escalation

The non-TPACKET path (packet_snd()) already correctly copies vnet_hdr to a stack-local variable. All other vnet_hdr consumers in the kernel (tun.c, tap.c, virtio_net.c) also use stack copies. The TPACKET TX path is the only caller of virtio_net_hdr_to_skb() that reads directly from user-controlled shared memory.

Fix this by copying vnet_hdr from the mmap'd ring buffer to a stack-local variable before validation and use, consistent with the approach used in packet_snd() and all other callers.

Exploitation requires CAP_NET_RAW, which can be obtained without special privileges via user namespaces.

Confirmed with a PoC on Linux 6.8.0 (Ubuntu): kprobe tracing on skb_partial_csum_set captured 77 race wins in 500,000 iterations.

Affects all kernels since PACKET_VNET_HDR support was added to the TPACKET TX path (~v3.14).

Fixes 9ed988e5 ("packet: add vnet_hdr support for tpacket_snd")
---
 net/packet/af_packet.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)
net/packet/af_packet.c +2 -1
index abcdef012345..fedcba654321 100644
@@ -2725,7 +2725,8 @@ static int tpacket_parse_header(struct packet_sock *po, void *frame,
 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 {
 	struct sk_buff *skb = NULL;
 	struct net_device *dev;
-	struct virtio_net_hdr *vnet_hdr = NULL;
+	struct virtio_net_hdr vnet_hdr;
+	bool has_vnet_hdr = false;
 	struct sockcm_cookie sockc;
 	__be16 proto;

int err, reserve = 0;
@@ -2828,16 +2829,17 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
if (po->has_vnet_hdr) {
- vnet_hdr = data;
- data += sizeof(*vnet_hdr);
- tp_len -= sizeof(*vnet_hdr);
+ memcpy(&vnet_hdr, data, sizeof(vnet_hdr));
+ data += sizeof(vnet_hdr);
+ tp_len -= sizeof(vnet_hdr);
if (tp_len < 0 ||
- __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
+ __packet_snd_vnet_parse(&vnet_hdr, tp_len)) {
tp_len = -EINVAL;
goto tpacket_error;
}
copylen = __virtio16_to_cpu(vio_le(),
- vnet_hdr->hdr_len);
+ vnet_hdr.hdr_len);
+ has_vnet_hdr = true;
}
copylen = max_t(int, copylen, dev->hard_header_len);
skb = sock_alloc_send_skb(&po->sk,
@@ -2875,11 +2877,11 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
}

- if (po->has_vnet_hdr) {
- if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
+ if (has_vnet_hdr) {
+ if (virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le())) {
tp_len = -EINVAL; goto tpacket_error;
}
- virtio_net_hdr_set_proto(skb, vnet_hdr);
+ virtio_net_hdr_set_proto(skb, &vnet_hdr);
}

skb->destructor = tpacket_destruct_skb;
--
2.43.0

maintainer
#2 lore raw
Quoted 23 lines · Zero Mark

In tpacket_snd(), when PACKET_VNET_HDR is enabled, vnet_hdr points directly into the mmap'd TX ring buffer shared with userspace. The kernel validates the header via __packet_snd_vnet_parse() but then re-reads all fields later in virtio_net_hdr_to_skb(). A concurrent userspace thread can modify the vnet_hdr fields (gso_type, gso_size, flags, csum_start, csum_offset) between validation and use, bypassing all safety checks.

This can lead to:
- Out-of-bounds checksum writes via crafted csum_start/csum_offset
- Malicious GSO segmentation parameters
- Kernel memory corruption and potential local privilege escalation

The non-TPACKET path (packet_snd()) already correctly copies vnet_hdr to a stack-local variable. All other vnet_hdr consumers in the kernel (tun.c, tap.c, virtio_net.c) also use stack copies. The TPACKET TX path is the only caller of virtio_net_hdr_to_skb() that reads directly from user-controlled shared memory.

Fix this by copying vnet_hdr from the mmap'd ring buffer to a stack-local variable before validation and use, consistent with the approach used in packet_snd() and all other callers.

Quoted 5 lines

Exploitation requires CAP_NET_RAW, which can be obtained without special privileges via user namespaces.

Confirmed with a PoC on Linux 6.8.0 (Ubuntu): kprobe tracing on skb_partial_csum_set captured 77 race wins in 500,000 iterations.

No need to add such details on exploitability of bugs.

Quoted 4 lines

Affects all kernels since PACKET_VNET_HDR support was added to the TPACKET TX path (~v3.14).

Fixes 9ed988e5 ("packet: add vnet_hdr support for tpacket_snd")

This patch does not exist. Also 12-char SHA1.

I think this should be

Fixes 1d036d25e560 ("packet: tpacket_snd gso and checksum offload")
Quoted 1 line

Thanks for the fix!

Only it does not apply cleanly. Please mark fixes [PATCH net] and ensure that they apply to current netdev-net/main

https://www.kernel.org/doc/html/latest/process/maintainer-netdev.html

Quoted 20 lines
---
 net/packet/af_packet.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)
net/packet/af_packet.c +2 -1
index abcdef012345..fedcba654321 100644
@@ -2725,7 +2725,8 @@ static int tpacket_parse_header(struct packet_sock *po, void *frame,
 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 {
 	struct sk_buff *skb = NULL;
 	struct net_device *dev;
-	struct virtio_net_hdr *vnet_hdr = NULL;
+	struct virtio_net_hdr vnet_hdr;
+	bool has_vnet_hdr = false;
 	struct sockcm_cookie sockc;
 	__be16 proto;
net/packet/af_packet.c +0 -0
 	int err, reserve = 0;

net/packet/af_packet.c static int tpacket_snd(struct packet_sock *po, struct msghdr *msg) · line 2830

Quoted 5 lines
net/packet/af_packet.c +1 -3
@@ -2828,16 +2829,17 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 		if (po->has_vnet_hdr) {
-			vnet_hdr = data;
-			data += sizeof(*vnet_hdr);
-			tp_len -= sizeof(*vnet_hdr);
+			memcpy(&vnet_hdr, data, sizeof(vnet_hdr));

Move the tp_len < 0 check before memcpy

Quoted 33 lines
net/packet/af_packet.c +8 -6
+			data += sizeof(vnet_hdr);
+			tp_len -= sizeof(vnet_hdr);
 			if (tp_len < 0 ||
-			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
+			    __packet_snd_vnet_parse(&vnet_hdr, tp_len)) {
 				tp_len = -EINVAL;
 				goto tpacket_error;
 			}
 			copylen = __virtio16_to_cpu(vio_le(),
-						    vnet_hdr->hdr_len);
+						    vnet_hdr.hdr_len);
+			has_vnet_hdr = true;
 		}
 		copylen = max_t(int, copylen, dev->hard_header_len);
 		skb = sock_alloc_send_skb(&po->sk,
@@ -2875,11 +2877,11 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 		}
 
-		if (po->has_vnet_hdr) {
-			if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
+		if (has_vnet_hdr) {
+			if (virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le())) {
 				tp_len = -EINVAL;
 				goto tpacket_error;
 			}
-			virtio_net_hdr_set_proto(skb, vnet_hdr);
+			virtio_net_hdr_set_proto(skb, &vnet_hdr);
 		}
 
 		skb->destructor = tpacket_destruct_skb;
--

2.43.0

author
#3 lore raw

PATCH 1/1 net/packet: fix TOCTOU race on mmap'd vnet_hdr in tpacket_snd()

In tpacket_snd(), when PACKET_VNET_HDR is enabled, vnet_hdr points directly into the mmap'd TX ring buffer shared with userspace. The kernel validates the header via __packet_snd_vnet_parse() but then re-reads all fields later in virtio_net_hdr_to_skb(). A concurrent userspace thread can modify the vnet_hdr fields between validation and use, bypassing all safety checks.

The non-TPACKET path (packet_snd()) already correctly copies vnet_hdr to a stack-local variable. All other vnet_hdr consumers in the kernel (tun.c, tap.c, virtio_net.c) also use stack copies. The TPACKET TX path is the only caller of virtio_net_hdr_to_skb() that reads directly from user-controlled shared memory.

Fix this by copying vnet_hdr from the mmap'd ring buffer to a stack-local variable before validation and use, consistent with the approach used in packet_snd() and all other callers.

Fixes 1d036d25e560 ("packet: tpacket_snd gso and checksum offload")
---
 net/packet/af_packet.c | 21 +++++++++++++--------
 1 file changed, 13 insertions(+), 8 deletions(-)
net/packet/af_packet.c +13 -8
index 4b043241fd56..8e6f3a734ba0 100644
@@ -2718,7 +2718,8 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 {
 	struct sk_buff *skb = NULL;
 	struct net_device *dev;
-	struct virtio_net_hdr *vnet_hdr = NULL;
+	struct virtio_net_hdr vnet_hdr;
+	bool has_vnet_hdr = false;
 	struct sockcm_cookie sockc;
 	__be16 proto;
 	int err, reserve = 0;
@@ -2819,16 +2820,20 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 		hlen = LL_RESERVED_SPACE(dev);
 		tlen = dev->needed_tailroom;
 		if (vnet_hdr_sz) {
-			vnet_hdr = data;
 			data += vnet_hdr_sz;
 			tp_len -= vnet_hdr_sz;
-			if (tp_len < 0 ||
-			    __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
+			if (tp_len < 0) {
+				tp_len = -EINVAL;
+				goto tpacket_error;
+			}
+			memcpy(&vnet_hdr, data - vnet_hdr_sz, sizeof(vnet_hdr));
+			if (__packet_snd_vnet_parse(&vnet_hdr, tp_len)) {
 				tp_len = -EINVAL;
 				goto tpacket_error;
 			}
 			copylen = __virtio16_to_cpu(vio_le(),
-						    vnet_hdr->hdr_len);
+						    vnet_hdr.hdr_len);
+			has_vnet_hdr = true;
 		}
 		copylen = max_t(int, copylen, dev->hard_header_len);
 		skb = sock_alloc_send_skb(&po->sk,
@@ -2865,12 +2870,12 @@ static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
 			}
 		}
 
-		if (vnet_hdr_sz) {
-			if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
+		if (has_vnet_hdr) {
+			if (virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le())) {
 				tp_len = -EINVAL;
 				goto tpacket_error;
 			}
-			virtio_net_hdr_set_proto(skb, vnet_hdr);
+			virtio_net_hdr_set_proto(skb, &vnet_hdr);
 		}
 
 		skb->destructor = tpacket_destruct_skb;
Signature · 1 line
2.53.0
maintainer Reviewed-by on v1 1/1
#4 lore raw
Quoted 13 lines · Zero Mark

In tpacket_snd(), when PACKET_VNET_HDR is enabled, vnet_hdr points directly into the mmap'd TX ring buffer shared with userspace. The kernel validates the header via __packet_snd_vnet_parse() but then re-reads all fields later in virtio_net_hdr_to_skb(). A concurrent userspace thread can modify the vnet_hdr fields between validation and use, bypassing all safety checks.

The non-TPACKET path (packet_snd()) already correctly copies vnet_hdr to a stack-local variable. All other vnet_hdr consumers in the kernel (tun.c, tap.c, virtio_net.c) also use stack copies. The TPACKET TX path is the only caller of virtio_net_hdr_to_skb() that reads directly from user-controlled shared memory.

Quoted 6 lines

Fix this by copying vnet_hdr from the mmap'd ring buffer to a stack-local variable before validation and use, consistent with the approach used in packet_snd() and all other callers.

Fixes 1d036d25e560 ("packet: tpacket_snd gso and checksum offload")