On 26/1/6 8:22, Jakub Kicinski wrote:

On Wed, 31 Dec 2025 11:54:19 +0800 Chen Zhen wrote:

quoted

skbuff: skb_under_panic: text:ffffffff95b33e66 len:90 put:14 head:ffff915ac1967440 data:ffff915ac196743e tail:0x58 end:0x180 dev:br0.10
------------[ cut here ]------------
kernel BUG at net/core/skbuff.c:197!
Call Trace:
 <TASK>
 skb_push+0x39/0x40
 eth_header+0x26/0xb0
 vlan_dev_hard_header+0x58/0x130 [8021q]
 neigh_connected_output+0xae/0x100
 ip6_finish_output2+0x2cc/0x650
 ? nf_hook_slow+0x41/0xc0
 ip6_finish_output+0x27/0xd0
 ndisc_send_skb+0x1d0/0x370
 ? __pfx_dst_output+0x10/0x10
 ndisc_send_ns+0x5a/0xb0
 addrconf_dad_work+0x2b5/0x380
 process_one_work+0x17f/0x320

Please run this stack trace thru script/decode_stacktrace
and you can cut off here, no need to include functions
below process_one_work, they are irrelevant.

quoted

 worker_thread+0x26d/0x2f0
 ? __pfx_worker_thread+0x10/0x10
 kthread+0xcc/0x100
 ? __pfx_kthread+0x10/0x10
 ret_from_fork+0x30/0x50
 ? __pfx_kthread+0x10/0x10
 ret_from_fork_asm+0x1b/0x30
 </TASK>

This bug can be easily reproduced by these steps:

 ip link add veth0 type veth peer name veth1
 ip link set veth0 up
 ip link set veth1 up
 ethtool -K veth0 tx-vlan-hw-insert off
 # vlandev.header_ops = vlan_header_ops, hard_header_len = 18(hard_header_len + VLAN_HLEN)
 ip link add link veth0 name veth0.10 type vlan id 10 reorder_hdr off
 ip addr add 192.168.10.1/24 dev veth0.10
 ip link set veth0.10 up
 # vlandev.hard_header_len = 14(hard_header_len)
 ethtool -K veth0 tx-vlan-hw-insert on
 # Panic!

Instead of putting this in the commit message please add a selftest
which will automatically catch re-occurrence of the issue.

quoted

The reason is that when NETIF_F_HW_VLAN_CTAG_TX is off, vlandev.hard_header_len will be set to
dev->hard_header_len since commit 029f5fc31cdb ("8021q: set hard_header_len when VLAN offload features
are toggled"), but the header_ops remains unchanged. Then neigh_connected_output() will call
vlan_dev_hard_header() and panic in skb_push() because reorder_hdr is off.

Please wrap commit messages at 70 columns.

Thanks for your patient review and I will send patch v2 soon.

Best regards,
Chen Zhen