CVE-2026-98954: net: always dissect GSO packets in __virtio_net_hdr_to_skb()

In the Linux kernel, the following vulnerability has been resolved:

net: always dissect GSO packets in virtionethdrtoskb()

Commit 9e8db5913264 ("net: avoid false positives in untrusted gso validation") added a '&& skb->networkheader' check before flow-dissecting GSO packets without VIRTIONETHDRFNEEDSCSUM in virtionethdrtoskb(), because some callers (such as tungetuser(), tunxdpone(), virtnetreceivedone(), and rawverifyheader()) called virtionethdrtoskb() before initializing skb->networkheader and skb->dev.

Because allocskb() and buildskbaround() zero-initialize skb->networkheader to 0 (unlike macheader and transportheader which are initialized to ~0U), those four callers always had skb->networkheader == 0 and bypassed flow dissection in virtionethdrtoskb(). More generally, skb->networkheader is an offset from skb->head (where 0 is also a valid offset whenever skbheadroom(skb) is 0), not a boolean flag.

Whenever the 'if (gsotype && skb->networkheader)' branch was skipped, the fallback 'else if (gsotype)' only pulled nhminlen + thlen (40 bytes for TCPv4) without dissecting the packet, without validating ipproto or nproto, and without setting skb->transportheader.

If the packet has a malformed network header, it is not rejected and a subsequent skbprobetransportheader() also fails, leaving skb->transportheader at ~0U (0xffff). Similarly, if an IPv4 packet carries IP options (ihl > 5) or an IPv6 packet carries extension headers, pulling only nhminlen + thlen can leave the TCP header outside skb->head. In both cases, tcphdrlen(skb) in skbgsotransportseglen() reads out-of-bounds:

BUG: KASAN: slab-out-of-bounds in skbgsotransportseglen Read of size 2 by task poc/133 skbgsotransportseglen (net/core/gso.c:155) skbgsovalidatemaclen (net/core/gso.c:270) tbfenqueue (net/sched/schtbf.c:260) devqdiscenqueue (net/core/dev.c:4227) devqueuexmit (net/core/dev.c:4884)

In addition, checking virtionethdrmatchproto() only inside 'if (!skb->protocol)' before flow dissection both skipped validation when skb->protocol was pre-set by the caller and rejected VLAN-tagged frames whose outer L2 protocol is ETHP8021Q or ETHP8021AD.

Fix this by: 1. Initializing skb->dev and skb->networkheader (plus skb->protocol for IFFTUN) before virtionethdrtoskb() in tungetuser(), tunxdpone(), virtnetreceivedone(), and rawverifyheader(). In tungetuser(), drop the redundant skbresetmacheader(skb) in the IFFTUN case since virtionethdrtoskb() unconditionally resets macheader. 2. Removing '&& skb->networkheader' and the unvalidated 'else if (gsotype)' fallback in virtionethdrtoskb() so all GSO packets without VIRTIONETHDRFNEEDSCSUM are flow-dissected, have their transport header pulled into linear data, and have skb->transportheader set. 3. Moving the virtionethdrmatchproto() check to after skbflowdissectflowkeysbasic(), validating keys.basic.nproto against hdrgsotype.