[PATCH bpf v2] lwt_bpf: restore reserved headroom after xmit program

[Junseo Lim]

ip_finish_output2() expands an skb to LL_RESERVED_SPACE(dev) before LWT
xmit. An LWT_XMIT BPF program can then modify the skb head and still
return BPF_OK, so bpf_xmit() rechecks the remaining headroom before the
skb continues to neighbour output.

That recheck uses dst->dev->hard_header_len. This is not enough for the
neighbour cached-header path: neigh_hh_output() copies the cached hardware
header using the aligned hh_cache size, HH_DATA_MOD for short headers or
HH_DATA_ALIGN(hh_len) otherwise.

On Ethernet, hard_header_len is 14 but the cached copy needs 16 bytes. If
an LWT_XMIT BPF program calls bpf_skb_change_head(skb, 1, 0), the skb can
still have 15 bytes of headroom after the program. The existing check
accepts that, after which neigh_hh_output() hits its headroom warning and
drops the skb.

Use LL_RESERVED_SPACE(dst->dev) in the post-BPF headroom check to match
the reservation made before LWT xmit.

Fixes: 3a0af8fd61f9 ("bpf: BPF for lightweight tunnel infrastructure")
Reported-by: Sechang Lim <rhkrqnwk98@xxxxxxxxx>
Suggested-by: Daniel Borkmann <daniel@xxxxxxxxxxxxx>
Signed-off-by: Junseo Lim <zirajs7@xxxxxxxxx>
---
This issue was found by a custom fuzzer developed by
Sechang Lim <rhkrqnwk98@xxxxxxxxx>.

Changelog:
v1 -> v2:
- Use LL_RESERVED_SPACE() instead of HH_DATA_ALIGN() to match the
  reservation made before LWT xmit.
  (Daniel Borkmann)
- Add a Reported-by tag.

v1: https://lore.kernel.org/all/20260727103005.897983-1-zirajs7@xxxxxxxxx/T/

 net/core/lwt_bpf.c | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/net/core/lwt_bpf.c b/net/core/lwt_bpf.c
index 652952d416f2..da49364ec63d 100644
--- a/net/core/lwt_bpf.c
+++ b/net/core/lwt_bpf.c
@@ -167,10 +167,10 @@ static int bpf_output(struct net *net, struct sock *sk, struct sk_buff *skb)
 	return dst->lwtstate->orig_output(net, sk, skb);
 }
 
-static int xmit_check_hhlen(struct sk_buff *skb, int hh_len)
+static int xmit_check_headroom(struct sk_buff *skb, int hroom)
 {
-	if (skb_headroom(skb) < hh_len) {
-		int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
+	if (skb_headroom(skb) < hroom) {
+		int nhead = hroom - skb_headroom(skb);
 
 		if (pskb_expand_head(skb, nhead, 0, GFP_ATOMIC))
 			return -ENOMEM;
@@ -282,7 +282,7 @@ static int bpf_xmit(struct sk_buff *skb)
 
 	bpf = bpf_lwt_lwtunnel(dst->lwtstate);
 	if (bpf->xmit.prog) {
-		int hh_len = dst->dev->hard_header_len;
+		int hroom = LL_RESERVED_SPACE(dst->dev);
 		__be16 proto = skb->protocol;
 		int ret;
 
@@ -298,9 +298,12 @@ static int bpf_xmit(struct sk_buff *skb)
 				return -EINVAL;
 			}
 			/* If the header was expanded, headroom might be too
-			 * small for L2 header to come, expand as needed.
+			 * small for the L2 header to come, expand as needed.
+			 * neigh_hh_output() copies the cached header in
+			 * HH_DATA_MOD aligned chunks, so match the reservation
+			 * made before LWT xmit.
 			 */
-			ret = xmit_check_hhlen(skb, hh_len);
+			ret = xmit_check_headroom(skb, hroom);
 			if (unlikely(ret))
 				return ret;
 
-- 
2.55.0