Linux kernel -stable discussions
 help / color / mirror / Atom feed
* [PATCH 6.12.y] netfilter: flowtable: ensure sufficient headroom in xmit path
@ 2026-11-16 15:28 Antonio Quartulli
  2026-11-17  1:02 ` Sasha Levin
  0 siblings, 1 reply; 2+ messages in thread
From: Antonio Quartulli @ 2026-11-16 15:28 UTC (permalink / raw)
  To: stable
  Cc: Antonio Quartulli, Pablo Neira Ayuso, Florian Westphal,
	Jozsef Kadlecsik, netfilter-devel

commit ef4f741e8627512cb8c82f59a1fc7aacd854aadf upstream.

Check for headroom and call skb_expand_head() like in the IP output
path to ensure there is sufficient headroom for the mac header when
forwarding this packet as suggested by sashiko.

Fixes: b5964aac51e0 ("netfilter: flowtable: consolidate xmit path")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
[ aq: the upstream commit does not apply to 6.12: 6.12 predates
  b5964aac51e0, and FLOW_OFFLOAD_XMIT_NEIGH flows are still sent through
  neigh_xmit() from nf_flow_offload_ip_hook() and
  nf_flow_offload_ipv6_hook(). Do the same headroom check there, before
  skb->dev is switched to the output device, and in nf_flow_queue_xmit().

  Without it, a forwarded skb reaches the output device's
  header_ops->create() with only the headroom it was received with. A
  UDP socket behind a veth (e.g. a container) leaves 16 bytes; virtio_net
  leaves 26 for frames larger than GOOD_COPY_LEN. GRE, ip6gre and VLAN
  have since been hardened individually, 6LoWPAN has not:
  lowpan_header_create() stores struct lowpan_addr_info (32 bytes) in
  front of skb->data after a mere WARN_ON_ONCE(), i.e. an out-of-bounds
  write before the skb head. Seen on v6.12.117 with IPv6 forwarded by a
  flowtable from a veth to a 6LoWPAN interface:

    WARNING: CPU: 1 PID: 4719 at net/ieee802154/6lowpan/tx.c:21 lowpan_header_create+0x165/0x1b0 [ieee802154_6lowpan]
    Call Trace:
     neigh_connected_output+0xa3/0xf0
     neigh_xmit+0xbb/0x210
     nf_flow_offload_ipv6_hook+0x540/0x7c0
     nf_hook_slow+0x39/0xe0
     __netif_receive_skb_core.constprop.0+0xe4c/0x1010
     __netif_receive_skb_one_core+0x37/0xa0
     process_backlog+0x77/0x140

  Before e67c577d8989 ("ipv4: ip_gre: make ipgre_header() robust") the
  same path panicked with multipoint GRE (DMVPN) output devices:
  skb_under_panic in ipgre_header() via nf_flow_offload_ip_hook(). ]
Fixes: 97add9f0d66d ("netfilter: flow table support for IPv4")
Fixes: 0995210753a2 ("netfilter: flow table support for IPv6")
Fixes: 7a27f6ab4135 ("netfilter: flowtable: use dev_fill_forward_path() to obtain egress device")
Link: https://vyos.dev/T7549
Signed-off-by: Antonio Quartulli <antonio@mandelbit.com>
---
Reproducer (run on v6.12.117: WARNING without this patch, clean with
it). Needs CONFIG_MAC802154_HWSIM and CONFIG_IEEE802154_6LOWPAN, plus
nft, conntrack and python3. The 6LoWPAN neighbour is static and the
conntrack entry is inserted with SEEN_REPLY, so the very first packet
matches "ct state established" and the flow is offloaded; no 802.15.4
peer is needed. If udev renames wpan0, use the name of the first
ARPHRD_IEEE802154 (type 804) interface. Run it in a throwaway VM.

  modprobe mac802154_hwsim; modprobe ieee802154_6lowpan
  sysctl -w net.ipv6.conf.all.forwarding=1
  ip link set wpan0 up
  ip link add link wpan0 name lowpan0 type lowpan
  ip link set lowpan0 up
  ip -6 addr add fd00:b::1/64 dev lowpan0 nodad
  ip -6 neigh replace fd00:b::2 lladdr 02:11:22:33:44:55:66:77 dev lowpan0 nud permanent

  ip netns add lan
  ip link add lan0 type veth peer name lan1 netns lan
  ip link set lan0 up
  ip -6 addr add fd00:a::1/64 dev lan0 nodad
  ip -n lan link set lan1 up
  ip -n lan -6 addr add fd00:a::2/64 dev lan1 nodad
  ip -n lan -6 route add default via fd00:a::1

  nft add table inet t
  nft 'add flowtable inet t ft { hook ingress priority filter; devices = { lan0, lowpan0 }; }'
  nft 'add chain inet t forward { type filter hook forward priority filter; policy accept; }'
  nft add rule inet t forward ct state established flow add @ft
  conntrack -I -p udp -s fd00:a::2 -d fd00:b::2 --sport 40000 --dport 5000 \
      -r fd00:b::2 -q fd00:a::2 --reply-port-src 5000 --reply-port-dst 40000 \
      -t 120 -u SEEN_REPLY,ASSURED

  ip netns exec lan python3 -c 'import socket; s = socket.socket(socket.AF_INET6, socket.SOCK_DGRAM); s.bind(("fd00:a::2", 40000)); [s.sendto(b"x" * 64, ("fd00:b::2", 5000)) for _ in range(100)]'
  dmesg | grep -A12 lowpan_header_create

Without this patch: the WARNING above. With it: none, while the flow is
still offloaded ("conntrack -L" shows [OFFLOAD]). Replacing
"flow add @ft" with "accept" also avoids it, because
ip6_finish_output2() expands the headroom.

 net/netfilter/nf_flow_table_ip.c | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

--- a/net/netfilter/nf_flow_table_ip.c
+++ b/net/netfilter/nf_flow_table_ip.c
@@ -330,6 +330,22 @@
 	}
 }
 
+/* Make room for the link layer header of @dev, like ip_finish_output2() does.
+ * The flowtable fast path skips the IP output path, and a device with
+ * header_ops (e.g. an NBMA GRE tunnel) may need more headroom than the skb
+ * was received with. On failure the skb has been freed.
+ */
+static struct sk_buff *nf_flow_xmit_headroom(struct sk_buff *skb,
+					     struct net_device *dev)
+{
+	unsigned int hh_len = LL_RESERVED_SPACE(dev);
+
+	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops))
+		skb = skb_expand_head(skb, hh_len);
+
+	return skb;
+}
+
 static unsigned int nf_flow_queue_xmit(struct net *net, struct sk_buff *skb,
 				       const struct flow_offload_tuple_rhash *tuplehash,
 				       unsigned short type)
@@ -340,6 +356,10 @@
 	if (!outdev)
 		return NF_DROP;
 
+	skb = nf_flow_xmit_headroom(skb, outdev);
+	if (!skb)
+		return NF_STOLEN;
+
 	skb->dev = outdev;
 	dev_hard_header(skb, skb->dev, type, tuplehash->tuple.out.h_dest,
 			tuplehash->tuple.out.h_source, skb->len);
@@ -452,6 +472,9 @@
 	case FLOW_OFFLOAD_XMIT_NEIGH:
 		rt = dst_rtable(tuplehash->tuple.dst_cache);
 		outdev = rt->dst.dev;
+		skb = nf_flow_xmit_headroom(skb, outdev);
+		if (!skb)
+			return NF_STOLEN;
 		skb->dev = outdev;
 		nexthop = rt_nexthop(rt, flow->tuplehash[!dir].tuple.src_v4.s_addr);
 		skb_dst_drop(skb);
@@ -748,6 +771,9 @@
 	case FLOW_OFFLOAD_XMIT_NEIGH:
 		rt = dst_rt6_info(tuplehash->tuple.dst_cache);
 		outdev = rt->dst.dev;
+		skb = nf_flow_xmit_headroom(skb, outdev);
+		if (!skb)
+			return NF_STOLEN;
 		skb->dev = outdev;
 		nexthop = rt6_nexthop(rt, &flow->tuplehash[!dir].tuple.src_v6);
 		skb_dst_drop(skb);

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH 6.12.y] netfilter: flowtable: ensure sufficient headroom in xmit path
  2026-11-16 15:28 [PATCH 6.12.y] netfilter: flowtable: ensure sufficient headroom in xmit path Antonio Quartulli
@ 2026-11-17  1:02 ` Sasha Levin
  0 siblings, 0 replies; 2+ messages in thread
From: Sasha Levin @ 2026-11-17  1:02 UTC (permalink / raw)
  To: stable
  Cc: Sasha Levin, Antonio Quartulli, Pablo Neira Ayuso,
	Florian Westphal, Jozsef Kadlecsik, netfilter-devel

> Check for headroom and call skb_expand_head() like in the IP output
> path to ensure there is sufficient headroom for the mac header when
> forwarding this packet as suggested by sashiko.

Queued for 6.12 and 6.6, thanks.

-- 
Thanks,
Sasha

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-17  1:03 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-16 15:28 [PATCH 6.12.y] netfilter: flowtable: ensure sufficient headroom in xmit path Antonio Quartulli
2026-11-17  1:02 ` Sasha Levin

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox