In cake_overhead(), packets with a single segment bypass multi-segment
overhead calculations:
if (segs == 1)
return cake_calc_overhead(q, len, off);
Commit c5d34f4583ea ("net_sched: cake: use qdisc_pkt_segs()") switched
cake to retrieve the cached segmentation count via qdisc_pkt_segs(skb)
instead of calculating it locally for dodgy GSO packets. If an skb with
segs == 0 reaches cake_overhead(), it skips the segs == 1 early return
and enters the multi-segment arithmetic:
len = shinfo->gso_size + hdr_len;
last_len = skb->len - shinfo->gso_size * (segs - 1);
return (cake_calc_overhead(q, len, off) * (segs - 1) +
cake_calc_overhead(q, last_len, off));
Because segs is an unsigned 16-bit integer, (segs - 1) underflows to
65535 (and is promoted to 4294967295 in the 32-bit unsigned arithmetic
above). This multiplies the per-segment overhead by UINT32_MAX, so
cake_overhead() returns a length close to 4.29 GB. cake_advance_shaper()
then charges that length to the shaper, stalling the CAKE dequeue path
for tens of seconds at 1 Gbit/s, and for minutes to hours at lower rates.
Fix this by returning early with cake_calc_overhead(q, len, off) whenever
segs <= 1.
Fixes: c5d34f4583ea ("net_sched: cake: use qdisc_pkt_segs()")
Cc: stable@vger.kernel.org
Acked-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: Yuchao Zhang <redacted>
---
v4:
- Add Acked-by: Toke Høiland-Jørgensen [off-list ref].
- Link to v3: https://lore.kernel.org/netdev/20260927131009.24250-1-ndaugoing@gmail.com/ (local)
v3:
- Split from v2 into a standalone patch with its own Fixes: tag
(c5d34f4583ea) per Simon Horman and Sashiko review.
- Link to v2: https://lore.kernel.org/netdev/20260922084124.36858-1-ndaugoing@gmail.com/ (local)
- Link to v1: https://lore.kernel.org/netdev/20260917122153.62722-1-ndaugoing@gmail.com/ (local)
net/sched/sch_cake.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/net/sched/sch_cake.c b/net/sched/sch_cake.c
index dc93267029e7..b0d604a7052a 100644
--- a/net/sched/sch_cake.c
+++ b/net/sched/sch_cake.c
@@ -1420,7 +1420,7 @@ static u32 cake_overhead(struct cake_sched_data *q, const struct sk_buff *skb)
WRITE_ONCE(q->avg_netoff, cake_ewma(q->avg_netoff, off << 16, 8));
- if (segs == 1)
+ if (segs <= 1)
return cake_calc_overhead(q, len, off);
/* borrowed from qdisc_pkt_len_segs_init() */
--
2.53.0