[PATCH net v2] net/sched: serialize autonomous timers with qdisc resets

[PATCH net v2] net/sched: serialize autonomous timers with qdisc resets

[Kyle Zeng]

Autonomous qdisc timers can race with queue resets during graft and
destruction. A timer can read qdisc_sleeping before a graft, acquire
that root's lock after the graft, and modify queues protected by the
replacement root's lock. __qdisc_destroy() also resets detached qdiscs
without the tree lock before their destroy callbacks stop the timers.
In SFQ, either race can corrupt skb lists and flow-slot state.

Add a helper that acquires the sleeping-root lock under RCU and retries
if the root changed. Use it in the SFQ, RED, PIE, FQ-PIE and DualPI2
timers. Route default-qdisc attachment and scheduler shutdown through
dev_graft_qdisc(), so all live root replacements hold the previous
root's lock. A validated root then remains current until the timer
unlocks.

Hold the tree lock across the final qdisc reset and release it before
calling the destroy callback so synchronous timer cancellation can
complete. Detached TAPRIO children can have different sleeping roots,
so defer their final resets to taprio_destroy() and lock each subtree
separately. Keep those resets before hardware offload is disabled.

The issue was found by an LLM.

Fixes: 225d9b89c937 ("sch_sfq: rehash queues in perturb timer")
Fixes: 47e8dbb6e763 ("net/sched: do not reset queues in graft operations")
Signed-off-by: Kyle Zeng <kylebot@openai.com>
---
Changes in v2:
- Serialize the final qdisc reset with autonomous timers, addressing
  the remaining destruction race identified in review.
- Reset detached TAPRIO children under their individual root locks.
- Add the two Fixes tags.

v1: https://lore.kernel.org/netdev/20261030042312.74363-1-kylebot@openai.com/
Review: https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261030042312.74363-1-kylebot@openai.com

I have a reproducer for this issue and can share via DM

 include/net/sch_generic.h | 20 ++++++++++++++++++++
 net/sched/sch_dualpi2.c   |  3 +--
 net/sched/sch_fq_pie.c    |  3 +--
 net/sched/sch_generic.c   | 11 +++++++----
 net/sched/sch_pie.c       |  3 +--
 net/sched/sch_red.c       |  3 +--
 net/sched/sch_sfq.c       |  3 +--
 net/sched/sch_taprio.c    | 16 ++++++++++++++++
 8 files changed, 48 insertions(+), 14 deletions(-)

diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index 1acaadb3e2cf..e234256e6ae8 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -608,6 +608,26 @@ static inline spinlock_t *qdisc_root_sleeping_lock(const struct Qdisc *qdisc)
 	return qdisc_lock(root);
 }
 
+/* The caller must hold rcu_read_lock().  dev_graft_qdisc() changes the
+ * sleeping root under the old root's lock, so retry if it changed while
+ * we were waiting for that lock.
+ */
+static inline spinlock_t *qdisc_lock_root_sleeping(const struct Qdisc *qdisc)
+{
+	struct Qdisc *root;
+	/* Lock of the sleeping root observed under RCU. */
+	spinlock_t *lock;
+
+	for (;;) {
+		root = qdisc_root_sleeping(qdisc);
+		lock = qdisc_lock(root);
+		spin_lock(lock);
+		if (root == qdisc_root_sleeping(qdisc))
+			return lock;
+		spin_unlock(lock);
+	}
+}
+
 static inline struct net_device *qdisc_dev(const struct Qdisc *qdisc)
 {
 	return qdisc->dev_queue->dev;
diff --git a/net/sched/sch_dualpi2.c b/net/sched/sch_dualpi2.c
index 4947def7c49e..8a74ceaab29d 100644
--- a/net/sched/sch_dualpi2.c
+++ b/net/sched/sch_dualpi2.c
@@ -750,8 +750,7 @@ static enum hrtimer_restart dualpi2_timer(struct hrtimer *timer)
 	spinlock_t *root_lock; /* to lock qdisc for probability calculations */
 
 	rcu_read_lock();
-	root_lock = qdisc_lock(qdisc_root_sleeping(sch));
-	spin_lock(root_lock);
+	root_lock = qdisc_lock_root_sleeping(sch);
 
 	WRITE_ONCE(q->pi2_prob, calculate_probability(sch));
 	hrtimer_set_expires(&q->pi2_timer, next_pi2_timeout(q));
diff --git a/net/sched/sch_fq_pie.c b/net/sched/sch_fq_pie.c
index 5982847df8f8..a1b5126db640 100644
--- a/net/sched/sch_fq_pie.c
+++ b/net/sched/sch_fq_pie.c
@@ -394,8 +394,7 @@ static void fq_pie_timer(struct timer_list *t)
 	int max_cnt, i;
 
 	rcu_read_lock();
-	root_lock = qdisc_lock(qdisc_root_sleeping(sch));
-	spin_lock(root_lock);
+	root_lock = qdisc_lock_root_sleeping(sch);
 
 	/* Limit this expensive loop to 2048 flows per round. */
 	max_cnt = min_t(int, q->flows_cnt - q->flows_cursor, 2048);
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index d2e0d20f3ba1..fbbc07700160 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -1106,7 +1106,7 @@ struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
 }
 EXPORT_SYMBOL(qdisc_create_dflt);
 
-/* Under qdisc_lock(qdisc) and BH! */
+/* Under sch_tree_lock(qdisc) and BH! */
 
 void qdisc_reset(struct Qdisc *qdisc)
 {
@@ -1154,8 +1154,10 @@ static void __qdisc_destroy(struct Qdisc *qdisc)
 #endif
 	gen_kill_estimator(&qdisc->rate_est);
 
+	/* Autonomous timers can still run after the qdisc was detached. */
+	sch_tree_lock(qdisc);
 	qdisc_reset(qdisc);
-
+	sch_tree_unlock(qdisc);
 
 	if (ops->destroy)
 		ops->destroy(qdisc);
@@ -1213,6 +1215,7 @@ struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
 	struct Qdisc *oqdisc = rtnl_dereference(dev_queue->qdisc_sleeping);
 	spinlock_t *root_lock;
 
+	/* Pairs with qdisc_lock_root_sleeping() in asynchronous callbacks. */
 	root_lock = qdisc_lock(oqdisc);
 	spin_lock_bh(root_lock);
 
@@ -1236,8 +1239,8 @@ static void shutdown_scheduler_queue(struct net_device *dev,
 	struct Qdisc *qdisc_default = _qdisc_default;
 
 	if (qdisc) {
+		qdisc = dev_graft_qdisc(dev_queue, qdisc_default);
 		rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
-		rcu_assign_pointer(dev_queue->qdisc_sleeping, qdisc_default);
 
 		qdisc_put(qdisc);
 	}
@@ -1261,7 +1264,7 @@ static void attach_one_default_qdisc(struct net_device *dev,
 
 	if (!netif_is_multiqueue(dev))
 		qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
-	rcu_assign_pointer(dev_queue->qdisc_sleeping, qdisc);
+	dev_graft_qdisc(dev_queue, qdisc);
 }
 
 static void attach_default_qdiscs(struct net_device *dev)
diff --git a/net/sched/sch_pie.c b/net/sched/sch_pie.c
index 3b7863ffd284..9367065db4b1 100644
--- a/net/sched/sch_pie.c
+++ b/net/sched/sch_pie.c
@@ -429,8 +429,7 @@ static void pie_timer(struct timer_list *t)
 	spinlock_t *root_lock;
 
 	rcu_read_lock();
-	root_lock = qdisc_lock(qdisc_root_sleeping(sch));
-	spin_lock(root_lock);
+	root_lock = qdisc_lock_root_sleeping(sch);
 	pie_calculate_probability(&q->params, &q->vars, sch->qstats.backlog);
 
 	/* reset the timer to fire after 'tupdate'. tupdate is in jiffies. */
diff --git a/net/sched/sch_red.c b/net/sched/sch_red.c
index d7598214270b..a7601b879092 100644
--- a/net/sched/sch_red.c
+++ b/net/sched/sch_red.c
@@ -333,8 +333,7 @@ static inline void red_adaptative_timer(struct timer_list *t)
 	spinlock_t *root_lock;
 
 	rcu_read_lock();
-	root_lock = qdisc_lock(qdisc_root_sleeping(sch));
-	spin_lock(root_lock);
+	root_lock = qdisc_lock_root_sleeping(sch);
 	red_adaptative_algo(&q->parms, &q->vars);
 	mod_timer(&q->adapt_timer, jiffies + HZ/2);
 	spin_unlock(root_lock);
diff --git a/net/sched/sch_sfq.c b/net/sched/sch_sfq.c
index 8bbcfc9e85d9..2bd3fc65053a 100644
--- a/net/sched/sch_sfq.c
+++ b/net/sched/sch_sfq.c
@@ -611,8 +611,7 @@ static void sfq_perturbation(struct timer_list *t)
 
 	get_random_bytes(&nkey, sizeof(nkey));
 	rcu_read_lock();
-	root_lock = qdisc_lock(qdisc_root_sleeping(sch));
-	spin_lock(root_lock);
+	root_lock = qdisc_lock_root_sleeping(sch);
 	q->perturbation = nkey;
 	if (!q->filter_list && q->tail)
 		sfq_rehash(sch);
diff --git a/net/sched/sch_taprio.c b/net/sched/sch_taprio.c
index 4f69ef3d847a..b7489cf4bba9 100644
--- a/net/sched/sch_taprio.c
+++ b/net/sched/sch_taprio.c
@@ -2041,6 +2041,10 @@ static void taprio_reset(struct Qdisc *sch)
 
 	hrtimer_cancel(&q->advance_timer);
 
+	/* Destroy resets each detached subtree under its own root lock. */
+	if (!refcount_read(&sch->refcnt))
+		return;
+
 	if (q->qdiscs) {
 		for (i = 0; i < dev->num_tx_queues; i++)
 			if (q->qdiscs[i])
@@ -2063,6 +2067,18 @@ static void taprio_destroy(struct Qdisc *sch)
 	hrtimer_cancel(&q->advance_timer);
 	qdisc_synchronize(sch);
 
+	if (q->qdiscs) {
+		for (i = 0; i < dev->num_tx_queues; i++) {
+			struct Qdisc *child = q->qdiscs[i];
+
+			if (!child)
+				continue;
+			sch_tree_lock(child);
+			qdisc_reset(child);
+			sch_tree_unlock(child);
+		}
+	}
+
 	taprio_disable_offload(dev, q, NULL);
 
 	if (q->qdiscs) {

base-commit: b3dfbde62cf1981c4638939c159dfd087b7a72e9
-- 
2.55.0.openai.911.g06873a30d2f3