net_sched: ets: fix a race in ets_qdisc_change()

Gerrard Tai reported a race condition in ETS, whenever SFQ perturb timer
fires at the wrong time.

The race is as follows:

CPU 0                                 CPU 1
[1]: lock root
[2]: qdisc_tree_flush_backlog()
[3]: unlock root
 |
 |                                    [5]: lock root
 |                                    [6]: rehash
 |                                    [7]: qdisc_tree_reduce_backlog()
 |
[4]: qdisc_put()

This can be abused to underflow a parent's qlen.

Calling qdisc_purge_queue() instead of qdisc_tree_flush_backlog()
should fix the race, because all packets will be purged from the qdisc
before releasing the lock.

Fixes: b05972f ("net: sched: tbf: don't call qdisc_put() while holding tree lock")
Reported-by: Gerrard Tai <[email protected]>
Suggested-by: Gerrard Tai <[email protected]>
Signed-off-by: Eric Dumazet <[email protected]>
Link: https://patch.msgid.link/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Author: edumazet

net/sched/sch_ets.c

@@ -661,7 +661,7 @@ static int ets_qdisc_change(struct Qdisc *sch, struct nlattr *opt,
 	for (i = q->nbands; i < oldbands; i++) {
 		if (i >= q->nstrict && q->classes[i].qdisc->q.qlen)
 			list_del_init(&q->classes[i].alist);
-		qdisc_tree_flush_backlog(q->classes[i].qdisc);
+		qdisc_purge_queue(q->classes[i].qdisc);
 	}
 	WRITE_ONCE(q->nstrict, nstrict);
 	memcpy(q->prio2band, priomap, sizeof(priomap));