[PATCH net v3 2/2] net/sched: sch_cake: validate transport header offset in cake_overhead()

[Yuchao Zhang <ndaugoing@gmail.com>]

In cake_overhead(), the header length up to the transport layer is
computed using logic borrowed from qdisc_pkt_len_segs_init():

	/* borrowed from qdisc_pkt_len_segs_init() */
	if (!skb->encapsulation)
		hdr_len = skb_transport_offset(skb);
	else
		hdr_len = skb_inner_transport_offset(skb);

However, cake_overhead() does not validate the computed offset:

1. When the transport header was never set, skb->transport_header holds
   the sentinel value ~0U. skb_transport_offset() returns ~65535.
   skb_header_pointer() subsequently fails, leaving hdr_len as ~65535,
   charging ~66 KB per segment to the shaper.
   Mirror qdisc_pkt_len_segs_init() by returning cake_calc_overhead()
   when unlikely(!skb_transport_header_was_set(skb)).

2. While qdisc_pkt_len_segs_init() runs at the start of __dev_queue_xmit(),
   packet headers may be adjusted before cake_overhead() is reached:
   - in sch_handle_egress() via tc/BPF egress filters;
   - inside cake_enqueue() via cake_classify() -> tcf_classify() (e.g.
     act_bpf, act_pedit, act_mpls, act_vlan).
   For example, bpf_skb_adjust_room(..., BPF_ADJ_ROOM_MAC) invokes
   bpf_skb_net_hdr_pop(), which pulls skb->data forward and re-syncs
   transport_header only when it aliased network_header, i.e. when no
   transport header had been parsed. If a transport header had been
   parsed, its offset is left where it was while skb->data moves forward,
   so skb_transport_offset() becomes old_offset - len and can turn
   negative. Since hdr_len was declared as unsigned int, a negative offset
   wraps around to near UINT_MAX, corrupting header length accounting.

Declare hdr_len as int and fall back to cake_calc_overhead(q, len, off) if
unlikely(hdr_len < 0).

Fixes: a729b7f0bd5b ("sch_cake: Add overhead compensation support to the rate shaper")
Cc: stable@vger.kernel.org
Signed-off-by: Yuchao Zhang <ndaugoing@gmail.com>
---
v3:
 - Split from v2 into a standalone patch with its own Fixes: tag
   (a729b7f0bd5b) per Simon Horman and Sashiko review.
 - Clarify header mangling ordering (sch_handle_egress() and cake_classify()
   before cake_overhead()) rather than inaccurate "post-enqueue mangling"
   wording per Sashiko review.
 - Link to v2: https://lore.kernel.org/netdev/20260922084124.36858-1-ndaugoing@gmail.com/
 - Link to v1: https://lore.kernel.org/netdev/20260917122153.62722-1-ndaugoing@gmail.com/

 net/sched/sch_cake.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/net/sched/sch_cake.c b/net/sched/sch_cake.c
index b0d604a7052a..45969c1b95fc 100644
--- a/net/sched/sch_cake.c
+++ b/net/sched/sch_cake.c
@@ -1413,10 +1413,11 @@ static u32 cake_calc_overhead(struct cake_sched_data *qd, u32 len, u32 off)
 static u32 cake_overhead(struct cake_sched_data *q, const struct sk_buff *skb)
 {
 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
-	unsigned int hdr_len, last_len = 0;
+	unsigned int last_len = 0;
 	u32 off = skb_network_offset(skb);
 	u16 segs = qdisc_pkt_segs(skb);
 	u32 len = qdisc_pkt_len(skb);
+	int hdr_len;
 
 	WRITE_ONCE(q->avg_netoff, cake_ewma(q->avg_netoff, off << 16, 8));
 
@@ -1424,10 +1425,16 @@ static u32 cake_overhead(struct cake_sched_data *q, const struct sk_buff *skb)
 		return cake_calc_overhead(q, len, off);
 
 	/* borrowed from qdisc_pkt_len_segs_init() */
-	if (!skb->encapsulation)
+	if (!skb->encapsulation) {
+		if (unlikely(!skb_transport_header_was_set(skb)))
+			return cake_calc_overhead(q, len, off);
 		hdr_len = skb_transport_offset(skb);
-	else
+	} else {
 		hdr_len = skb_inner_transport_offset(skb);
+	}
+
+	if (unlikely(hdr_len < 0))
+		return cake_calc_overhead(q, len, off);
 
 	/* + transport layer */
 	if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 |
-- 
2.53.0