Re: [PATCH] net/sched: Forbid assigning mirred action to a filter attached to the egress

[Jamal Hadi Salim <jhs@mojatatu.com>]

[-- Attachment #1: Type: text/plain, Size: 3419 bytes --]

On Wed, Mar 20, 2024 at 3:34 PM Jamal Hadi Salim <jhs@mojatatu.com> wrote:
>
> On Wed, Mar 20, 2024 at 2:26 PM Eric Dumazet <edumazet@google.com> wrote:
> >
> > On Wed, Mar 20, 2024 at 7:13 PM Eric Dumazet <edumazet@google.com> wrote:
> > >
> > > On Wed, Mar 20, 2024 at 6:50 PM Jamal Hadi Salim <jhs@mojatatu.com> wrote:
> >
> > > Nope, you just have to complete the patch, moving around
> > > dev_xmit_recursion_inc() and dev_xmit_recursion_dec()
> >
> > Untested part would be:
> >
> > diff --git a/net/core/dev.c b/net/core/dev.c
> > index 303a6ff46e4e16296e94ed6b726621abe093e567..dbeaf67282e8b6ec164d00d796c9fd8e4fd7c332
> > 100644
> > --- a/net/core/dev.c
> > +++ b/net/core/dev.c
> > @@ -4259,6 +4259,8 @@ int __dev_queue_xmit(struct sk_buff *skb, struct
> > net_device *sb_dev)
> >          */
> >         rcu_read_lock_bh();
> >
> > +       dev_xmit_recursion_inc();
> > +
> >         skb_update_prio(skb);
> >
> >         qdisc_pkt_len_init(skb);
> > @@ -4331,9 +4333,7 @@ int __dev_queue_xmit(struct sk_buff *skb, struct
> > net_device *sb_dev)
> >                         HARD_TX_LOCK(dev, txq, cpu);
> >
> >                         if (!netif_xmit_stopped(txq)) {
> > -                               dev_xmit_recursion_inc();
> >                                 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
> > -                               dev_xmit_recursion_dec();
> >                                 if (dev_xmit_complete(rc)) {
> >                                         HARD_TX_UNLOCK(dev, txq);
> >                                         goto out;
> > @@ -4353,12 +4353,14 @@ int __dev_queue_xmit(struct sk_buff *skb,
> > struct net_device *sb_dev)
> >         }
> >
> >         rc = -ENETDOWN;
> > +       dev_xmit_recursion_dec();
> >         rcu_read_unlock_bh();
> >
> >         dev_core_stats_tx_dropped_inc(dev);
> >         kfree_skb_list(skb);
> >         return rc;
> >  out:
> > +       dev_xmit_recursion_dec();
> >         rcu_read_unlock_bh();
> >         return rc;
> >  }
>
> This removed the deadlock but now every packet being redirected will
> be dropped. I was wrong earlier on the tc block because that only
> works on clsact and ingress which are fine not needing this lock.
> Here's a variation of the earlier patch that may work but comes at the
> cost of a new per-cpu increment on the qdisc.
>
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -3789,6 +3789,11 @@ static inline int __dev_xmit_skb(struct sk_buff
> *skb, struct Qdisc *q,
>         if (unlikely(contended))
>                 spin_lock(&q->busylock);
>
> +       if (__this_cpu_read(q->recursion_xmit) > 0) {
> +               //drop
> +       }
> +
> +       __this_cpu_inc(q->recursion_xmit);
>         spin_lock(root_lock);
>         if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
>                 __qdisc_drop(skb, &to_free);
> @@ -3825,6 +3830,7 @@ static inline int __dev_xmit_skb(struct sk_buff
> *skb, struct Qdisc *q,
>                 }
>         }
>         spin_unlock(root_lock);
> +       __this_cpu_dec(q->recursion_xmit);
>         if (unlikely(to_free))
>                 kfree_skb_list_reason(to_free,
>                                       tcf_get_drop_reason(to_free));
>

And here's a tested version(attached) that fixes both the A->A and A->B->A.

cheers,
jamal

[-- Attachment #2: qdisc-rec-xmit-var-fix.patch --]
[-- Type: text/x-patch, Size: 2619 bytes --]

diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
index cefe0c4bdae3..f9f99df037ed 100644
--- a/include/net/sch_generic.h
+++ b/include/net/sch_generic.h
@@ -125,6 +125,8 @@ struct Qdisc {
 	spinlock_t		busylock ____cacheline_aligned_in_smp;
 	spinlock_t		seqlock;
 
+	u16 __percpu            *xmit_recursion;
+
 	struct rcu_head		rcu;
 	netdevice_tracker	dev_tracker;
 	/* private data */
diff --git a/net/core/dev.c b/net/core/dev.c
index 9a67003e49db..2b712388c06f 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3789,6 +3789,13 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
 	if (unlikely(contended))
 		spin_lock(&q->busylock);
 
+	if (__this_cpu_read(*q->xmit_recursion) > 0) {
+		__qdisc_drop(skb, &to_free);
+		rc = NET_XMIT_DROP;
+		goto free_skb_list;
+	}
+
+	__this_cpu_inc(*q->xmit_recursion);
 	spin_lock(root_lock);
 	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
 		__qdisc_drop(skb, &to_free);
@@ -3825,7 +3832,9 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
 		}
 	}
 	spin_unlock(root_lock);
+	__this_cpu_dec(*q->xmit_recursion);
 	if (unlikely(to_free))
+free_skb_list:
 		kfree_skb_list_reason(to_free,
 				      tcf_get_drop_reason(to_free));
 	if (unlikely(contended))
diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
index 65e05b0c98e4..6c3bc1aff89a 100644
--- a/net/sched/sch_api.c
+++ b/net/sched/sch_api.c
@@ -1260,6 +1260,7 @@ static struct Qdisc *qdisc_create(struct net_device *dev,
 	struct Qdisc *sch;
 	struct Qdisc_ops *ops;
 	struct qdisc_size_table *stab;
+	int cpu;
 
 	ops = qdisc_lookup_ops(kind);
 #ifdef CONFIG_MODULES
@@ -1376,11 +1377,22 @@ static struct Qdisc *qdisc_create(struct net_device *dev,
 		}
 	}
 
+	sch->xmit_recursion = alloc_percpu(u16);
+	if (!sch->xmit_recursion) {
+		err = -ENOMEM;
+		goto err_out5;
+	}
+	for_each_possible_cpu(cpu)
+		(*per_cpu_ptr(sch->xmit_recursion, cpu)) = 0;
+
 	qdisc_hash_add(sch, false);
 	trace_qdisc_create(ops, dev, parent);
 
 	return sch;
 
+err_out5:
+	if (tca[TCA_RATE])
+		gen_kill_estimator(&sch->rate_est);
 err_out4:
 	/* Even if ops->init() failed, we call ops->destroy()
 	 * like qdisc_create_dflt().
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index ff5336493777..afbbd2e885a4 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -1070,6 +1070,8 @@ static void __qdisc_destroy(struct Qdisc *qdisc)
 	module_put(ops->owner);
 	netdev_put(dev, &qdisc->dev_tracker);
 
+	free_percpu(qdisc->xmit_recursion);
+
 	trace_qdisc_destroy(qdisc);
 
 	call_rcu(&qdisc->rcu, qdisc_free_cb);