[PATCH net-next v2 0/2] Guard against gso_segs overflows

[PATCH net-next v2 0/2] Guard against gso_segs overflows

[Alice Mikityanska]

From: Alice Mikityanska <alice@isovalent.com>

This series is a follow-up on the discussion:

https://lore.kernel.org/netdev/CAD0BsJWzSr2zduf5v3mVC4zd=Lj6ZAoC+V42-VBdg42aDY8XXw@mail.gmail.com/T/#m1e22fca273c36cc8844e516505d3251cc1418fea

skb_segment is patched to avoid possible overflows in partial GSO. The
primary possible source of too many GSO segments is also addressed:
virtio-net clamps gso_size to >=8 in TCP, as suggested by Eric.

v2 changes: dropped new checks in fastpath from skb_gro_receive and
__virtio_net_hdr_to_skb, replaced the latter with a silent clamp for
gso_size, kept the cheap guard in skb_segment.

v1: https://lore.kernel.org/netdev/20260723155145.158572-1-alice.kernel@fastmail.im/

Alice Mikityanska (2):
  net: Guard for gso_segs overflow in skb_segment
  virtio-net: Ensure that TCP packets don't overflow gso_segs

 include/linux/virtio_net.h | 4 ++++
 net/core/skbuff.c          | 2 +-
 2 files changed, 5 insertions(+), 1 deletion(-)

-- 
2.55.0

[PATCH net-next v2 1/2] net: Guard for gso_segs overflow in skb_segment

[Alice Mikityanska]

From: Alice Mikityanska <alice@isovalent.com>

skb_segment calculates 32-bit partial_segs as len / gso_size, and then
assigns it to the 16-bit gso_segs field. The division might overflow in
some edge cases where the SKB is BIG TCP (65536 <= len <= 8*65535), and
gso_size < TCP_MIN_GSO_SIZE = 8. While normally this can't happen due to
TCP_MIN_GSO_SIZE, an AF_PACKET PACKET_VNET_HDR socket can generate such
a malformed packet.

Blocking malformed virtio_net packets is implemented in the next patch,
but this patch clamps partial_segs in skb_segment itself for more
generic robustness. Should len / gso_size happen to be bigger than
65535 in partial GSO, skb_segment will now just produce more than two
output SKBs, all of which will be valid with gso_segs <= 65535.

Signed-off-by: Alice Mikityanska <alice@isovalent.com>
---
 net/core/skbuff.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index c82a1472a5ea..439cbfeb02bd 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -4860,7 +4860,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
 		 * doesn't fit into an MSS sized block, so take care of that
 		 * now.
 		 */
-		partial_segs = len / mss;
+		partial_segs = min(len / mss, GSO_MAX_SEGS);
 		if (partial_segs > 1)
 			mss *= partial_segs;
 		else
-- 
2.55.0

[PATCH net-next v2 2/2] virtio-net: Ensure that TCP packets don't overflow gso_segs

[Alice Mikityanska]

From: Alice Mikityanska <alice@isovalent.com>

The user can specify any gso_size in a packet crafted with an AF_PACKET
PACKET_VNET_HDR socket, even smaller than TCP_MIN_GSO_SIZE = 8. At the
same time, GSO_MAX_SIZE = 8 * GSO_MAX_SEGS = 8 * 65535. When the user
crafts a packet with gso_size < 8, there is a risk for partial GSO to
overflow the 16-bit gso_segs field when dividing the SKB length by
gso_size.

Adjust gso_size of TCP packets to be at least TCP_MIN_GSO_SIZE = 8. Keep
gso_size of UDP GSO packets, as gso_size=1 is valid and explicitly
tested at tools/testing/selftests/net/tun.c:649.

Signed-off-by: Alice Mikityanska <alice@isovalent.com>
Suggested-by: Eric Dumazet <edumazet@google.com>
---
 include/linux/virtio_net.h | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/include/linux/virtio_net.h b/include/linux/virtio_net.h
index f36d21b5bc19..c381b916c1b5 100644
--- a/include/linux/virtio_net.h
+++ b/include/linux/virtio_net.h
@@ -6,6 +6,7 @@
 #include <linux/ip.h>
 #include <linux/ipv6.h>
 #include <linux/udp.h>
+#include <net/tcp.h>
 #include <uapi/linux/tcp.h>
 #include <uapi/linux/virtio_net.h>
 
@@ -179,6 +180,9 @@ static inline int __virtio_net_hdr_to_skb(struct sk_buff *skb,
 			if (skb->ip_summed == CHECKSUM_PARTIAL &&
 			    skb->csum_offset != offsetof(struct tcphdr, check))
 				return -EINVAL;
+
+			BUILD_BUG_ON(TCP_MIN_GSO_SIZE * GSO_MAX_SEGS < GSO_MAX_SIZE);
+			gso_size = max(gso_size, TCP_MIN_GSO_SIZE);
 			break;
 		}
 
-- 
2.55.0

Re: [PATCH net-next v2 1/2] net: Guard for gso_segs overflow in skb_segment

[Paolo Abeni]

On 8/13/26 7:46 PM, Alice Mikityanska wrote:
> From: Alice Mikityanska <alice@isovalent.com>
> 
> skb_segment calculates 32-bit partial_segs as len / gso_size, and then
> assigns it to the 16-bit gso_segs field. The division might overflow in
> some edge cases where the SKB is BIG TCP (65536 <= len <= 8*65535), and
> gso_size < TCP_MIN_GSO_SIZE = 8. While normally this can't happen due to
> TCP_MIN_GSO_SIZE, an AF_PACKET PACKET_VNET_HDR socket can generate such
> a malformed packet.
> 
> Blocking malformed virtio_net packets is implemented in the next patch,
> but this patch clamps partial_segs in skb_segment itself for more
> generic robustness. Should len / gso_size happen to be bigger than
> 65535 in partial GSO, skb_segment will now just produce more than two
> output SKBs, all of which will be valid with gso_segs <= 65535.

Minor nit: I think it would make sense to re-order the patches.

> Signed-off-by: Alice Mikityanska <alice@isovalent.com>
> ---
>  net/core/skbuff.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index c82a1472a5ea..439cbfeb02bd 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -4860,7 +4860,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
>  		 * doesn't fit into an MSS sized block, so take care of that
>  		 * now.
>  		 */
> -		partial_segs = len / mss;
> +		partial_segs = min(len / mss, GSO_MAX_SEGS);

Since on top of patch 2/2 the min() should always be a no-op, what about
instead:

		if (WARN_ON_ONCE(len/mss > GSO_MAX_SEGS))
			return ERR_PTR(-EINVAL);

?

/P

Re: [PATCH net-next v2 1/2] net: Guard for gso_segs overflow in skb_segment

[Alice Mikityanska]

On Tue, Aug 18, 2026, at 16:38, Paolo Abeni wrote:
> On 8/13/26 7:46 PM, Alice Mikityanska wrote:
>> From: Alice Mikityanska <alice@isovalent.com>
>> 
>> skb_segment calculates 32-bit partial_segs as len / gso_size, and then
>> assigns it to the 16-bit gso_segs field. The division might overflow in
>> some edge cases where the SKB is BIG TCP (65536 <= len <= 8*65535), and
>> gso_size < TCP_MIN_GSO_SIZE = 8. While normally this can't happen due to
>> TCP_MIN_GSO_SIZE, an AF_PACKET PACKET_VNET_HDR socket can generate such
>> a malformed packet.
>> 
>> Blocking malformed virtio_net packets is implemented in the next patch,
>> but this patch clamps partial_segs in skb_segment itself for more
>> generic robustness. Should len / gso_size happen to be bigger than
>> 65535 in partial GSO, skb_segment will now just produce more than two
>> output SKBs, all of which will be valid with gso_segs <= 65535.
>
> Minor nit: I think it would make sense to re-order the patches.
>
>> Signed-off-by: Alice Mikityanska <alice@isovalent.com>
>> ---
>>  net/core/skbuff.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>> 
>> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
>> index c82a1472a5ea..439cbfeb02bd 100644
>> --- a/net/core/skbuff.c
>> +++ b/net/core/skbuff.c
>> @@ -4860,7 +4860,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
>>  		 * doesn't fit into an MSS sized block, so take care of that
>>  		 * now.
>>  		 */
>> -		partial_segs = len / mss;
>> +		partial_segs = min(len / mss, GSO_MAX_SEGS);
>
> Since on top of patch 2/2 the min() should always be a no-op,

Not sure; it'll be a no-op in this specific virtio_net scenario, but
what if there are more?

> what about instead:
>
> 		if (WARN_ON_ONCE(len/mss > GSO_MAX_SEGS))
> 			return ERR_PTR(-EINVAL);

Yeah, it looks like a good idea to add a WARN to let syzbot uncover more
possible cases, but:

1. I'd keep it non-failing rather than return an error: partial GSO can
   deal with smaller segment size pretty well, no need to fail.

2. I'd make it DEBUG_NET_WARN_ON_ONCE to avoid the penalty in
   production.

I can do that if it passes Eric's filter for too much code in fastpath.

> ?
>
> /P