[PATCH net v3 0/2] packet: use consistent hard_header_len in send paths

[Qihang <q.h.hack.winter@gmail.com>]

AF_PACKET send paths read dev->hard_header_len at several stages of skb
allocation and construction. Concurrent netdevice reconfiguration can
make those reads inconsistent and cause skb headroom underflow.

Split the regular and TX_RING paths so each patch has one Fixes tag.
The separate SOCK_DGRAM consistency issue between hard_header_len and
header_ops->create remains outside this series.

Changes in v3:
- Split into two patches, one Fixes tag each.
- Add LL_RESERVED_SPACE_EX() for calculations using a saved length.
- Use int consistently for saved header lengths.
- Drop unnecessary READ_ONCE(header_ops).
- Snapshot hard_header_len per-frame in the TX_RING path.
- Reset copylen for each frame.

Link to v2: https://lore.kernel.org/netdev/20260726092110.80185-1-q.h.hack.winter@gmail.com/

Qihang (2):
  packet: use consistent hard_header_len in non-ring send paths
  packet: use consistent hard_header_len in TX_RING send path

 include/linux/netdevice.h |  6 +++--
 net/packet/af_packet.c    | 47 ++++++++++++++++++++++++---------------
 2 files changed, 33 insertions(+), 20 deletions(-)

-- 
2.50.1 (Apple Git-155)