net_sched: add qdisc_dequeue_drop() helper

Some qdisc like cake, codel, fq_codel might drop packets
in their dequeue() method.

This is currently problematic because dequeue() runs with
the qdisc spinlock held. Freeing skbs can be extremely expensive.

Add qdisc_dequeue_drop() method and a new TCQ_F_DEQUEUE_DROPS
so that these qdiscs can opt-in to defer the skb frees
after the socket spinlock is released.

TCQ_F_DEQUEUE_DROPS is an attempt to not penalize other qdiscs
with an extra cache line miss.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20251121083256.674562-14-edumazet@google.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

2 files changed, 30 insertions, 5 deletions

diff --git a/include/net/pkt_sched.h b/include/net/pkt_sched.h
index 4678db45832a..e703c507d0da 100644
--- a/include/net/pkt_sched.h
+++ b/include/net/pkt_sched.h
@@ -114,12 +114,13 @@ bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
 
 void __qdisc_run(struct Qdisc *q);
 
-static inline void qdisc_run(struct Qdisc *q)
+static inline struct sk_buff *qdisc_run(struct Qdisc *q)
 {
 	if (qdisc_run_begin(q)) {
 		__qdisc_run(q);
-		qdisc_run_end(q);
+		return qdisc_run_end(q);
 	}
+	return NULL;
 }
 
 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index b8092d0378a0..c3a7268b567e 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -88,6 +88,8 @@ struct Qdisc {
 #define TCQ_F_INVISIBLE		0x80 /* invisible by default in dump */
 #define TCQ_F_NOLOCK		0x100 /* qdisc does not require locking */
 #define TCQ_F_OFFLOADED		0x200 /* qdisc is offloaded to HW */
+#define TCQ_F_DEQUEUE_DROPS	0x400 /* ->dequeue() can drop packets in q->to_free */
+
 	u32			limit;
 	const struct Qdisc_ops	*ops;
 	struct qdisc_size_table	__rcu *stab;
@@ -119,6 +121,8 @@ struct Qdisc {
 
 		/* Note : we only change qstats.backlog in fast path. */
 		struct gnet_stats_queue	qstats;
+
+		struct sk_buff		*to_free;
 	__cacheline_group_end(Qdisc_write);
 
 
@@ -218,8 +222,10 @@ static inline bool qdisc_run_begin(struct Qdisc *qdisc)
 	return true;
 }
 
-static inline void qdisc_run_end(struct Qdisc *qdisc)
+static inline struct sk_buff *qdisc_run_end(struct Qdisc *qdisc)
 {
+	struct sk_buff *to_free = NULL;
+
 	if (qdisc->flags & TCQ_F_NOLOCK) {
 		spin_unlock(&qdisc->seqlock);
 
@@ -232,9 +238,16 @@ static inline void qdisc_run_end(struct Qdisc *qdisc)
 		if (unlikely(test_bit(__QDISC_STATE_MISSED,
 				      &qdisc->state)))
 			__netif_schedule(qdisc);
-	} else {
-		WRITE_ONCE(qdisc->running, false);
+		return NULL;
+	}
+
+	if (qdisc->flags & TCQ_F_DEQUEUE_DROPS) {
+		to_free = qdisc->to_free;
+		if (to_free)
+			qdisc->to_free = NULL;
 	}
+	WRITE_ONCE(qdisc->running, false);
+	return to_free;
 }
 
 static inline bool qdisc_may_bulk(const struct Qdisc *qdisc)
@@ -1116,6 +1129,17 @@ static inline void tcf_kfree_skb_list(struct sk_buff *skb)
 	}
 }
 
+static inline void qdisc_dequeue_drop(struct Qdisc *q, struct sk_buff *skb,
+				      enum skb_drop_reason reason)
+{
+	DEBUG_NET_WARN_ON_ONCE(!(q->flags & TCQ_F_DEQUEUE_DROPS));
+	DEBUG_NET_WARN_ON_ONCE(q->flags & TCQ_F_NOLOCK);
+
+	tcf_set_drop_reason(skb, reason);
+	skb->next = q->to_free;
+	q->to_free = skb;
+}
+
 /* Instead of calling kfree_skb() while root qdisc lock is held,
  * queue the skb for future freeing at end of __dev_xmit_skb()
  */