seccomp: Fix ioctl number for SECCOMP_IOCTL_NOTIF_ID_VALID
[linux/fpc-iii.git] / net / sched / sch_generic.c
blob896c9037155a5d9b2b1b214284eeab053388d3b4
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * net/sched/sch_generic.c Generic packet scheduler routines.
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 * Jamal Hadi Salim, <hadi@cyberus.ca> 990601
7 * - Ingress support
8 */
10 #include <linux/bitops.h>
11 #include <linux/module.h>
12 #include <linux/types.h>
13 #include <linux/kernel.h>
14 #include <linux/sched.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/netdevice.h>
18 #include <linux/skbuff.h>
19 #include <linux/rtnetlink.h>
20 #include <linux/init.h>
21 #include <linux/rcupdate.h>
22 #include <linux/list.h>
23 #include <linux/slab.h>
24 #include <linux/if_vlan.h>
25 #include <linux/skb_array.h>
26 #include <linux/if_macvlan.h>
27 #include <net/sch_generic.h>
28 #include <net/pkt_sched.h>
29 #include <net/dst.h>
30 #include <trace/events/qdisc.h>
31 #include <trace/events/net.h>
32 #include <net/xfrm.h>
34 /* Qdisc to use by default */
35 const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops;
36 EXPORT_SYMBOL(default_qdisc_ops);
38 /* Main transmission queue. */
40 /* Modifications to data participating in scheduling must be protected with
41 * qdisc_lock(qdisc) spinlock.
43 * The idea is the following:
44 * - enqueue, dequeue are serialized via qdisc root lock
45 * - ingress filtering is also serialized via qdisc root lock
46 * - updates to tree and tree walking are only done under the rtnl mutex.
49 #define SKB_XOFF_MAGIC ((struct sk_buff *)1UL)
51 static inline struct sk_buff *__skb_dequeue_bad_txq(struct Qdisc *q)
53 const struct netdev_queue *txq = q->dev_queue;
54 spinlock_t *lock = NULL;
55 struct sk_buff *skb;
57 if (q->flags & TCQ_F_NOLOCK) {
58 lock = qdisc_lock(q);
59 spin_lock(lock);
62 skb = skb_peek(&q->skb_bad_txq);
63 if (skb) {
64 /* check the reason of requeuing without tx lock first */
65 txq = skb_get_tx_queue(txq->dev, skb);
66 if (!netif_xmit_frozen_or_stopped(txq)) {
67 skb = __skb_dequeue(&q->skb_bad_txq);
68 if (qdisc_is_percpu_stats(q)) {
69 qdisc_qstats_cpu_backlog_dec(q, skb);
70 qdisc_qstats_cpu_qlen_dec(q);
71 } else {
72 qdisc_qstats_backlog_dec(q, skb);
73 q->q.qlen--;
75 } else {
76 skb = SKB_XOFF_MAGIC;
80 if (lock)
81 spin_unlock(lock);
83 return skb;
86 static inline struct sk_buff *qdisc_dequeue_skb_bad_txq(struct Qdisc *q)
88 struct sk_buff *skb = skb_peek(&q->skb_bad_txq);
90 if (unlikely(skb))
91 skb = __skb_dequeue_bad_txq(q);
93 return skb;
96 static inline void qdisc_enqueue_skb_bad_txq(struct Qdisc *q,
97 struct sk_buff *skb)
99 spinlock_t *lock = NULL;
101 if (q->flags & TCQ_F_NOLOCK) {
102 lock = qdisc_lock(q);
103 spin_lock(lock);
106 __skb_queue_tail(&q->skb_bad_txq, skb);
108 if (qdisc_is_percpu_stats(q)) {
109 qdisc_qstats_cpu_backlog_inc(q, skb);
110 qdisc_qstats_cpu_qlen_inc(q);
111 } else {
112 qdisc_qstats_backlog_inc(q, skb);
113 q->q.qlen++;
116 if (lock)
117 spin_unlock(lock);
120 static inline void dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
122 spinlock_t *lock = NULL;
124 if (q->flags & TCQ_F_NOLOCK) {
125 lock = qdisc_lock(q);
126 spin_lock(lock);
129 while (skb) {
130 struct sk_buff *next = skb->next;
132 __skb_queue_tail(&q->gso_skb, skb);
134 /* it's still part of the queue */
135 if (qdisc_is_percpu_stats(q)) {
136 qdisc_qstats_cpu_requeues_inc(q);
137 qdisc_qstats_cpu_backlog_inc(q, skb);
138 qdisc_qstats_cpu_qlen_inc(q);
139 } else {
140 q->qstats.requeues++;
141 qdisc_qstats_backlog_inc(q, skb);
142 q->q.qlen++;
145 skb = next;
147 if (lock)
148 spin_unlock(lock);
149 __netif_schedule(q);
152 static void try_bulk_dequeue_skb(struct Qdisc *q,
153 struct sk_buff *skb,
154 const struct netdev_queue *txq,
155 int *packets)
157 int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
159 while (bytelimit > 0) {
160 struct sk_buff *nskb = q->dequeue(q);
162 if (!nskb)
163 break;
165 bytelimit -= nskb->len; /* covers GSO len */
166 skb->next = nskb;
167 skb = nskb;
168 (*packets)++; /* GSO counts as one pkt */
170 skb_mark_not_on_list(skb);
173 /* This variant of try_bulk_dequeue_skb() makes sure
174 * all skbs in the chain are for the same txq
176 static void try_bulk_dequeue_skb_slow(struct Qdisc *q,
177 struct sk_buff *skb,
178 int *packets)
180 int mapping = skb_get_queue_mapping(skb);
181 struct sk_buff *nskb;
182 int cnt = 0;
184 do {
185 nskb = q->dequeue(q);
186 if (!nskb)
187 break;
188 if (unlikely(skb_get_queue_mapping(nskb) != mapping)) {
189 qdisc_enqueue_skb_bad_txq(q, nskb);
190 break;
192 skb->next = nskb;
193 skb = nskb;
194 } while (++cnt < 8);
195 (*packets) += cnt;
196 skb_mark_not_on_list(skb);
199 /* Note that dequeue_skb can possibly return a SKB list (via skb->next).
200 * A requeued skb (via q->gso_skb) can also be a SKB list.
202 static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
203 int *packets)
205 const struct netdev_queue *txq = q->dev_queue;
206 struct sk_buff *skb = NULL;
208 *packets = 1;
209 if (unlikely(!skb_queue_empty(&q->gso_skb))) {
210 spinlock_t *lock = NULL;
212 if (q->flags & TCQ_F_NOLOCK) {
213 lock = qdisc_lock(q);
214 spin_lock(lock);
217 skb = skb_peek(&q->gso_skb);
219 /* skb may be null if another cpu pulls gso_skb off in between
220 * empty check and lock.
222 if (!skb) {
223 if (lock)
224 spin_unlock(lock);
225 goto validate;
228 /* skb in gso_skb were already validated */
229 *validate = false;
230 if (xfrm_offload(skb))
231 *validate = true;
232 /* check the reason of requeuing without tx lock first */
233 txq = skb_get_tx_queue(txq->dev, skb);
234 if (!netif_xmit_frozen_or_stopped(txq)) {
235 skb = __skb_dequeue(&q->gso_skb);
236 if (qdisc_is_percpu_stats(q)) {
237 qdisc_qstats_cpu_backlog_dec(q, skb);
238 qdisc_qstats_cpu_qlen_dec(q);
239 } else {
240 qdisc_qstats_backlog_dec(q, skb);
241 q->q.qlen--;
243 } else {
244 skb = NULL;
246 if (lock)
247 spin_unlock(lock);
248 goto trace;
250 validate:
251 *validate = true;
253 if ((q->flags & TCQ_F_ONETXQUEUE) &&
254 netif_xmit_frozen_or_stopped(txq))
255 return skb;
257 skb = qdisc_dequeue_skb_bad_txq(q);
258 if (unlikely(skb)) {
259 if (skb == SKB_XOFF_MAGIC)
260 return NULL;
261 goto bulk;
263 skb = q->dequeue(q);
264 if (skb) {
265 bulk:
266 if (qdisc_may_bulk(q))
267 try_bulk_dequeue_skb(q, skb, txq, packets);
268 else
269 try_bulk_dequeue_skb_slow(q, skb, packets);
271 trace:
272 trace_qdisc_dequeue(q, txq, *packets, skb);
273 return skb;
277 * Transmit possibly several skbs, and handle the return status as
278 * required. Owning running seqcount bit guarantees that
279 * only one CPU can execute this function.
281 * Returns to the caller:
282 * false - hardware queue frozen backoff
283 * true - feel free to send more pkts
285 bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
286 struct net_device *dev, struct netdev_queue *txq,
287 spinlock_t *root_lock, bool validate)
289 int ret = NETDEV_TX_BUSY;
290 bool again = false;
292 /* And release qdisc */
293 if (root_lock)
294 spin_unlock(root_lock);
296 /* Note that we validate skb (GSO, checksum, ...) outside of locks */
297 if (validate)
298 skb = validate_xmit_skb_list(skb, dev, &again);
300 #ifdef CONFIG_XFRM_OFFLOAD
301 if (unlikely(again)) {
302 if (root_lock)
303 spin_lock(root_lock);
305 dev_requeue_skb(skb, q);
306 return false;
308 #endif
310 if (likely(skb)) {
311 HARD_TX_LOCK(dev, txq, smp_processor_id());
312 if (!netif_xmit_frozen_or_stopped(txq))
313 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
315 HARD_TX_UNLOCK(dev, txq);
316 } else {
317 if (root_lock)
318 spin_lock(root_lock);
319 return true;
322 if (root_lock)
323 spin_lock(root_lock);
325 if (!dev_xmit_complete(ret)) {
326 /* Driver returned NETDEV_TX_BUSY - requeue skb */
327 if (unlikely(ret != NETDEV_TX_BUSY))
328 net_warn_ratelimited("BUG %s code %d qlen %d\n",
329 dev->name, ret, q->q.qlen);
331 dev_requeue_skb(skb, q);
332 return false;
335 return true;
339 * NOTE: Called under qdisc_lock(q) with locally disabled BH.
341 * running seqcount guarantees only one CPU can process
342 * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
343 * this queue.
345 * netif_tx_lock serializes accesses to device driver.
347 * qdisc_lock(q) and netif_tx_lock are mutually exclusive,
348 * if one is grabbed, another must be free.
350 * Note, that this procedure can be called by a watchdog timer
352 * Returns to the caller:
353 * 0 - queue is empty or throttled.
354 * >0 - queue is not empty.
357 static inline bool qdisc_restart(struct Qdisc *q, int *packets)
359 spinlock_t *root_lock = NULL;
360 struct netdev_queue *txq;
361 struct net_device *dev;
362 struct sk_buff *skb;
363 bool validate;
365 /* Dequeue packet */
366 skb = dequeue_skb(q, &validate, packets);
367 if (unlikely(!skb))
368 return false;
370 if (!(q->flags & TCQ_F_NOLOCK))
371 root_lock = qdisc_lock(q);
373 dev = qdisc_dev(q);
374 txq = skb_get_tx_queue(dev, skb);
376 return sch_direct_xmit(skb, q, dev, txq, root_lock, validate);
379 void __qdisc_run(struct Qdisc *q)
381 int quota = dev_tx_weight;
382 int packets;
384 while (qdisc_restart(q, &packets)) {
386 * Ordered by possible occurrence: Postpone processing if
387 * 1. we've exceeded packet quota
388 * 2. another process needs the CPU;
390 quota -= packets;
391 if (quota <= 0 || need_resched()) {
392 __netif_schedule(q);
393 break;
398 unsigned long dev_trans_start(struct net_device *dev)
400 unsigned long val, res;
401 unsigned int i;
403 if (is_vlan_dev(dev))
404 dev = vlan_dev_real_dev(dev);
405 else if (netif_is_macvlan(dev))
406 dev = macvlan_dev_real_dev(dev);
407 res = netdev_get_tx_queue(dev, 0)->trans_start;
408 for (i = 1; i < dev->num_tx_queues; i++) {
409 val = netdev_get_tx_queue(dev, i)->trans_start;
410 if (val && time_after(val, res))
411 res = val;
414 return res;
416 EXPORT_SYMBOL(dev_trans_start);
418 static void dev_watchdog(struct timer_list *t)
420 struct net_device *dev = from_timer(dev, t, watchdog_timer);
422 netif_tx_lock(dev);
423 if (!qdisc_tx_is_noop(dev)) {
424 if (netif_device_present(dev) &&
425 netif_running(dev) &&
426 netif_carrier_ok(dev)) {
427 int some_queue_timedout = 0;
428 unsigned int i;
429 unsigned long trans_start;
431 for (i = 0; i < dev->num_tx_queues; i++) {
432 struct netdev_queue *txq;
434 txq = netdev_get_tx_queue(dev, i);
435 trans_start = txq->trans_start;
436 if (netif_xmit_stopped(txq) &&
437 time_after(jiffies, (trans_start +
438 dev->watchdog_timeo))) {
439 some_queue_timedout = 1;
440 txq->trans_timeout++;
441 break;
445 if (some_queue_timedout) {
446 trace_net_dev_xmit_timeout(dev, i);
447 WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
448 dev->name, netdev_drivername(dev), i);
449 dev->netdev_ops->ndo_tx_timeout(dev);
451 if (!mod_timer(&dev->watchdog_timer,
452 round_jiffies(jiffies +
453 dev->watchdog_timeo)))
454 dev_hold(dev);
457 netif_tx_unlock(dev);
459 dev_put(dev);
462 void __netdev_watchdog_up(struct net_device *dev)
464 if (dev->netdev_ops->ndo_tx_timeout) {
465 if (dev->watchdog_timeo <= 0)
466 dev->watchdog_timeo = 5*HZ;
467 if (!mod_timer(&dev->watchdog_timer,
468 round_jiffies(jiffies + dev->watchdog_timeo)))
469 dev_hold(dev);
472 EXPORT_SYMBOL_GPL(__netdev_watchdog_up);
474 static void dev_watchdog_up(struct net_device *dev)
476 __netdev_watchdog_up(dev);
479 static void dev_watchdog_down(struct net_device *dev)
481 netif_tx_lock_bh(dev);
482 if (del_timer(&dev->watchdog_timer))
483 dev_put(dev);
484 netif_tx_unlock_bh(dev);
488 * netif_carrier_on - set carrier
489 * @dev: network device
491 * Device has detected acquisition of carrier.
493 void netif_carrier_on(struct net_device *dev)
495 if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
496 if (dev->reg_state == NETREG_UNINITIALIZED)
497 return;
498 atomic_inc(&dev->carrier_up_count);
499 linkwatch_fire_event(dev);
500 if (netif_running(dev))
501 __netdev_watchdog_up(dev);
504 EXPORT_SYMBOL(netif_carrier_on);
507 * netif_carrier_off - clear carrier
508 * @dev: network device
510 * Device has detected loss of carrier.
512 void netif_carrier_off(struct net_device *dev)
514 if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
515 if (dev->reg_state == NETREG_UNINITIALIZED)
516 return;
517 atomic_inc(&dev->carrier_down_count);
518 linkwatch_fire_event(dev);
521 EXPORT_SYMBOL(netif_carrier_off);
523 /* "NOOP" scheduler: the best scheduler, recommended for all interfaces
524 under all circumstances. It is difficult to invent anything faster or
525 cheaper.
528 static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc,
529 struct sk_buff **to_free)
531 __qdisc_drop(skb, to_free);
532 return NET_XMIT_CN;
535 static struct sk_buff *noop_dequeue(struct Qdisc *qdisc)
537 return NULL;
540 struct Qdisc_ops noop_qdisc_ops __read_mostly = {
541 .id = "noop",
542 .priv_size = 0,
543 .enqueue = noop_enqueue,
544 .dequeue = noop_dequeue,
545 .peek = noop_dequeue,
546 .owner = THIS_MODULE,
549 static struct netdev_queue noop_netdev_queue = {
550 RCU_POINTER_INITIALIZER(qdisc, &noop_qdisc),
551 .qdisc_sleeping = &noop_qdisc,
554 struct Qdisc noop_qdisc = {
555 .enqueue = noop_enqueue,
556 .dequeue = noop_dequeue,
557 .flags = TCQ_F_BUILTIN,
558 .ops = &noop_qdisc_ops,
559 .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
560 .dev_queue = &noop_netdev_queue,
561 .running = SEQCNT_ZERO(noop_qdisc.running),
562 .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
563 .gso_skb = {
564 .next = (struct sk_buff *)&noop_qdisc.gso_skb,
565 .prev = (struct sk_buff *)&noop_qdisc.gso_skb,
566 .qlen = 0,
567 .lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.gso_skb.lock),
569 .skb_bad_txq = {
570 .next = (struct sk_buff *)&noop_qdisc.skb_bad_txq,
571 .prev = (struct sk_buff *)&noop_qdisc.skb_bad_txq,
572 .qlen = 0,
573 .lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.skb_bad_txq.lock),
576 EXPORT_SYMBOL(noop_qdisc);
578 static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt,
579 struct netlink_ext_ack *extack)
581 /* register_qdisc() assigns a default of noop_enqueue if unset,
582 * but __dev_queue_xmit() treats noqueue only as such
583 * if this is NULL - so clear it here. */
584 qdisc->enqueue = NULL;
585 return 0;
588 struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
589 .id = "noqueue",
590 .priv_size = 0,
591 .init = noqueue_init,
592 .enqueue = noop_enqueue,
593 .dequeue = noop_dequeue,
594 .peek = noop_dequeue,
595 .owner = THIS_MODULE,
598 static const u8 prio2band[TC_PRIO_MAX + 1] = {
599 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
602 /* 3-band FIFO queue: old style, but should be a bit faster than
603 generic prio+fifo combination.
606 #define PFIFO_FAST_BANDS 3
609 * Private data for a pfifo_fast scheduler containing:
610 * - rings for priority bands
612 struct pfifo_fast_priv {
613 struct skb_array q[PFIFO_FAST_BANDS];
616 static inline struct skb_array *band2list(struct pfifo_fast_priv *priv,
617 int band)
619 return &priv->q[band];
622 static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc,
623 struct sk_buff **to_free)
625 int band = prio2band[skb->priority & TC_PRIO_MAX];
626 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
627 struct skb_array *q = band2list(priv, band);
628 unsigned int pkt_len = qdisc_pkt_len(skb);
629 int err;
631 err = skb_array_produce(q, skb);
633 if (unlikely(err)) {
634 if (qdisc_is_percpu_stats(qdisc))
635 return qdisc_drop_cpu(skb, qdisc, to_free);
636 else
637 return qdisc_drop(skb, qdisc, to_free);
640 qdisc_update_stats_at_enqueue(qdisc, pkt_len);
641 return NET_XMIT_SUCCESS;
644 static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
646 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
647 struct sk_buff *skb = NULL;
648 int band;
650 for (band = 0; band < PFIFO_FAST_BANDS && !skb; band++) {
651 struct skb_array *q = band2list(priv, band);
653 if (__skb_array_empty(q))
654 continue;
656 skb = __skb_array_consume(q);
658 if (likely(skb)) {
659 qdisc_update_stats_at_dequeue(qdisc, skb);
660 } else {
661 WRITE_ONCE(qdisc->empty, true);
664 return skb;
667 static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc)
669 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
670 struct sk_buff *skb = NULL;
671 int band;
673 for (band = 0; band < PFIFO_FAST_BANDS && !skb; band++) {
674 struct skb_array *q = band2list(priv, band);
676 skb = __skb_array_peek(q);
679 return skb;
682 static void pfifo_fast_reset(struct Qdisc *qdisc)
684 int i, band;
685 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
687 for (band = 0; band < PFIFO_FAST_BANDS; band++) {
688 struct skb_array *q = band2list(priv, band);
689 struct sk_buff *skb;
691 /* NULL ring is possible if destroy path is due to a failed
692 * skb_array_init() in pfifo_fast_init() case.
694 if (!q->ring.queue)
695 continue;
697 while ((skb = __skb_array_consume(q)) != NULL)
698 kfree_skb(skb);
701 if (qdisc_is_percpu_stats(qdisc)) {
702 for_each_possible_cpu(i) {
703 struct gnet_stats_queue *q;
705 q = per_cpu_ptr(qdisc->cpu_qstats, i);
706 q->backlog = 0;
707 q->qlen = 0;
712 static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
714 struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
716 memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1);
717 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
718 goto nla_put_failure;
719 return skb->len;
721 nla_put_failure:
722 return -1;
725 static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt,
726 struct netlink_ext_ack *extack)
728 unsigned int qlen = qdisc_dev(qdisc)->tx_queue_len;
729 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
730 int prio;
732 /* guard against zero length rings */
733 if (!qlen)
734 return -EINVAL;
736 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
737 struct skb_array *q = band2list(priv, prio);
738 int err;
740 err = skb_array_init(q, qlen, GFP_KERNEL);
741 if (err)
742 return -ENOMEM;
745 /* Can by-pass the queue discipline */
746 qdisc->flags |= TCQ_F_CAN_BYPASS;
747 return 0;
750 static void pfifo_fast_destroy(struct Qdisc *sch)
752 struct pfifo_fast_priv *priv = qdisc_priv(sch);
753 int prio;
755 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
756 struct skb_array *q = band2list(priv, prio);
758 /* NULL ring is possible if destroy path is due to a failed
759 * skb_array_init() in pfifo_fast_init() case.
761 if (!q->ring.queue)
762 continue;
763 /* Destroy ring but no need to kfree_skb because a call to
764 * pfifo_fast_reset() has already done that work.
766 ptr_ring_cleanup(&q->ring, NULL);
770 static int pfifo_fast_change_tx_queue_len(struct Qdisc *sch,
771 unsigned int new_len)
773 struct pfifo_fast_priv *priv = qdisc_priv(sch);
774 struct skb_array *bands[PFIFO_FAST_BANDS];
775 int prio;
777 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
778 struct skb_array *q = band2list(priv, prio);
780 bands[prio] = q;
783 return skb_array_resize_multiple(bands, PFIFO_FAST_BANDS, new_len,
784 GFP_KERNEL);
787 struct Qdisc_ops pfifo_fast_ops __read_mostly = {
788 .id = "pfifo_fast",
789 .priv_size = sizeof(struct pfifo_fast_priv),
790 .enqueue = pfifo_fast_enqueue,
791 .dequeue = pfifo_fast_dequeue,
792 .peek = pfifo_fast_peek,
793 .init = pfifo_fast_init,
794 .destroy = pfifo_fast_destroy,
795 .reset = pfifo_fast_reset,
796 .dump = pfifo_fast_dump,
797 .change_tx_queue_len = pfifo_fast_change_tx_queue_len,
798 .owner = THIS_MODULE,
799 .static_flags = TCQ_F_NOLOCK | TCQ_F_CPUSTATS,
801 EXPORT_SYMBOL(pfifo_fast_ops);
803 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
804 const struct Qdisc_ops *ops,
805 struct netlink_ext_ack *extack)
807 void *p;
808 struct Qdisc *sch;
809 unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size;
810 int err = -ENOBUFS;
811 struct net_device *dev;
813 if (!dev_queue) {
814 NL_SET_ERR_MSG(extack, "No device queue given");
815 err = -EINVAL;
816 goto errout;
819 dev = dev_queue->dev;
820 p = kzalloc_node(size, GFP_KERNEL,
821 netdev_queue_numa_node_read(dev_queue));
823 if (!p)
824 goto errout;
825 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
826 /* if we got non aligned memory, ask more and do alignment ourself */
827 if (sch != p) {
828 kfree(p);
829 p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL,
830 netdev_queue_numa_node_read(dev_queue));
831 if (!p)
832 goto errout;
833 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
834 sch->padded = (char *) sch - (char *) p;
836 __skb_queue_head_init(&sch->gso_skb);
837 __skb_queue_head_init(&sch->skb_bad_txq);
838 qdisc_skb_head_init(&sch->q);
839 spin_lock_init(&sch->q.lock);
841 if (ops->static_flags & TCQ_F_CPUSTATS) {
842 sch->cpu_bstats =
843 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
844 if (!sch->cpu_bstats)
845 goto errout1;
847 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
848 if (!sch->cpu_qstats) {
849 free_percpu(sch->cpu_bstats);
850 goto errout1;
854 spin_lock_init(&sch->busylock);
855 /* seqlock has the same scope of busylock, for NOLOCK qdisc */
856 spin_lock_init(&sch->seqlock);
857 seqcount_init(&sch->running);
859 sch->ops = ops;
860 sch->flags = ops->static_flags;
861 sch->enqueue = ops->enqueue;
862 sch->dequeue = ops->dequeue;
863 sch->dev_queue = dev_queue;
864 sch->empty = true;
865 dev_hold(dev);
866 refcount_set(&sch->refcnt, 1);
868 if (sch != &noop_qdisc) {
869 lockdep_set_class(&sch->busylock, &dev->qdisc_tx_busylock_key);
870 lockdep_set_class(&sch->seqlock, &dev->qdisc_tx_busylock_key);
871 lockdep_set_class(&sch->running, &dev->qdisc_running_key);
874 return sch;
875 errout1:
876 kfree(p);
877 errout:
878 return ERR_PTR(err);
881 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
882 const struct Qdisc_ops *ops,
883 unsigned int parentid,
884 struct netlink_ext_ack *extack)
886 struct Qdisc *sch;
888 if (!try_module_get(ops->owner)) {
889 NL_SET_ERR_MSG(extack, "Failed to increase module reference counter");
890 return NULL;
893 sch = qdisc_alloc(dev_queue, ops, extack);
894 if (IS_ERR(sch)) {
895 module_put(ops->owner);
896 return NULL;
898 sch->parent = parentid;
900 if (!ops->init || ops->init(sch, NULL, extack) == 0)
901 return sch;
903 qdisc_put(sch);
904 return NULL;
906 EXPORT_SYMBOL(qdisc_create_dflt);
908 /* Under qdisc_lock(qdisc) and BH! */
910 void qdisc_reset(struct Qdisc *qdisc)
912 const struct Qdisc_ops *ops = qdisc->ops;
913 struct sk_buff *skb, *tmp;
915 if (ops->reset)
916 ops->reset(qdisc);
918 skb_queue_walk_safe(&qdisc->gso_skb, skb, tmp) {
919 __skb_unlink(skb, &qdisc->gso_skb);
920 kfree_skb_list(skb);
923 skb_queue_walk_safe(&qdisc->skb_bad_txq, skb, tmp) {
924 __skb_unlink(skb, &qdisc->skb_bad_txq);
925 kfree_skb_list(skb);
928 qdisc->q.qlen = 0;
929 qdisc->qstats.backlog = 0;
931 EXPORT_SYMBOL(qdisc_reset);
933 void qdisc_free(struct Qdisc *qdisc)
935 if (qdisc_is_percpu_stats(qdisc)) {
936 free_percpu(qdisc->cpu_bstats);
937 free_percpu(qdisc->cpu_qstats);
940 kfree((char *) qdisc - qdisc->padded);
943 static void qdisc_free_cb(struct rcu_head *head)
945 struct Qdisc *q = container_of(head, struct Qdisc, rcu);
947 qdisc_free(q);
950 static void qdisc_destroy(struct Qdisc *qdisc)
952 const struct Qdisc_ops *ops = qdisc->ops;
953 struct sk_buff *skb, *tmp;
955 #ifdef CONFIG_NET_SCHED
956 qdisc_hash_del(qdisc);
958 qdisc_put_stab(rtnl_dereference(qdisc->stab));
959 #endif
960 gen_kill_estimator(&qdisc->rate_est);
961 if (ops->reset)
962 ops->reset(qdisc);
963 if (ops->destroy)
964 ops->destroy(qdisc);
966 module_put(ops->owner);
967 dev_put(qdisc_dev(qdisc));
969 skb_queue_walk_safe(&qdisc->gso_skb, skb, tmp) {
970 __skb_unlink(skb, &qdisc->gso_skb);
971 kfree_skb_list(skb);
974 skb_queue_walk_safe(&qdisc->skb_bad_txq, skb, tmp) {
975 __skb_unlink(skb, &qdisc->skb_bad_txq);
976 kfree_skb_list(skb);
979 call_rcu(&qdisc->rcu, qdisc_free_cb);
982 void qdisc_put(struct Qdisc *qdisc)
984 if (!qdisc)
985 return;
987 if (qdisc->flags & TCQ_F_BUILTIN ||
988 !refcount_dec_and_test(&qdisc->refcnt))
989 return;
991 qdisc_destroy(qdisc);
993 EXPORT_SYMBOL(qdisc_put);
995 /* Version of qdisc_put() that is called with rtnl mutex unlocked.
996 * Intended to be used as optimization, this function only takes rtnl lock if
997 * qdisc reference counter reached zero.
1000 void qdisc_put_unlocked(struct Qdisc *qdisc)
1002 if (qdisc->flags & TCQ_F_BUILTIN ||
1003 !refcount_dec_and_rtnl_lock(&qdisc->refcnt))
1004 return;
1006 qdisc_destroy(qdisc);
1007 rtnl_unlock();
1009 EXPORT_SYMBOL(qdisc_put_unlocked);
1011 /* Attach toplevel qdisc to device queue. */
1012 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
1013 struct Qdisc *qdisc)
1015 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
1016 spinlock_t *root_lock;
1018 root_lock = qdisc_lock(oqdisc);
1019 spin_lock_bh(root_lock);
1021 /* ... and graft new one */
1022 if (qdisc == NULL)
1023 qdisc = &noop_qdisc;
1024 dev_queue->qdisc_sleeping = qdisc;
1025 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
1027 spin_unlock_bh(root_lock);
1029 return oqdisc;
1031 EXPORT_SYMBOL(dev_graft_qdisc);
1033 static void attach_one_default_qdisc(struct net_device *dev,
1034 struct netdev_queue *dev_queue,
1035 void *_unused)
1037 struct Qdisc *qdisc;
1038 const struct Qdisc_ops *ops = default_qdisc_ops;
1040 if (dev->priv_flags & IFF_NO_QUEUE)
1041 ops = &noqueue_qdisc_ops;
1042 else if(dev->type == ARPHRD_CAN)
1043 ops = &pfifo_fast_ops;
1045 qdisc = qdisc_create_dflt(dev_queue, ops, TC_H_ROOT, NULL);
1046 if (!qdisc) {
1047 netdev_info(dev, "activation failed\n");
1048 return;
1050 if (!netif_is_multiqueue(dev))
1051 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1052 dev_queue->qdisc_sleeping = qdisc;
1055 static void attach_default_qdiscs(struct net_device *dev)
1057 struct netdev_queue *txq;
1058 struct Qdisc *qdisc;
1060 txq = netdev_get_tx_queue(dev, 0);
1062 if (!netif_is_multiqueue(dev) ||
1063 dev->priv_flags & IFF_NO_QUEUE) {
1064 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
1065 dev->qdisc = txq->qdisc_sleeping;
1066 qdisc_refcount_inc(dev->qdisc);
1067 } else {
1068 qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT, NULL);
1069 if (qdisc) {
1070 dev->qdisc = qdisc;
1071 qdisc->ops->attach(qdisc);
1074 #ifdef CONFIG_NET_SCHED
1075 if (dev->qdisc != &noop_qdisc)
1076 qdisc_hash_add(dev->qdisc, false);
1077 #endif
1080 static void transition_one_qdisc(struct net_device *dev,
1081 struct netdev_queue *dev_queue,
1082 void *_need_watchdog)
1084 struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
1085 int *need_watchdog_p = _need_watchdog;
1087 if (!(new_qdisc->flags & TCQ_F_BUILTIN))
1088 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
1090 rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
1091 if (need_watchdog_p) {
1092 dev_queue->trans_start = 0;
1093 *need_watchdog_p = 1;
1097 void dev_activate(struct net_device *dev)
1099 int need_watchdog;
1101 /* No queueing discipline is attached to device;
1102 * create default one for devices, which need queueing
1103 * and noqueue_qdisc for virtual interfaces
1106 if (dev->qdisc == &noop_qdisc)
1107 attach_default_qdiscs(dev);
1109 if (!netif_carrier_ok(dev))
1110 /* Delay activation until next carrier-on event */
1111 return;
1113 need_watchdog = 0;
1114 netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
1115 if (dev_ingress_queue(dev))
1116 transition_one_qdisc(dev, dev_ingress_queue(dev), NULL);
1118 if (need_watchdog) {
1119 netif_trans_update(dev);
1120 dev_watchdog_up(dev);
1123 EXPORT_SYMBOL(dev_activate);
1125 static void dev_deactivate_queue(struct net_device *dev,
1126 struct netdev_queue *dev_queue,
1127 void *_qdisc_default)
1129 struct Qdisc *qdisc_default = _qdisc_default;
1130 struct Qdisc *qdisc;
1132 qdisc = rtnl_dereference(dev_queue->qdisc);
1133 if (qdisc) {
1134 bool nolock = qdisc->flags & TCQ_F_NOLOCK;
1136 if (nolock)
1137 spin_lock_bh(&qdisc->seqlock);
1138 spin_lock_bh(qdisc_lock(qdisc));
1140 if (!(qdisc->flags & TCQ_F_BUILTIN))
1141 set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
1143 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
1144 qdisc_reset(qdisc);
1146 spin_unlock_bh(qdisc_lock(qdisc));
1147 if (nolock)
1148 spin_unlock_bh(&qdisc->seqlock);
1152 static bool some_qdisc_is_busy(struct net_device *dev)
1154 unsigned int i;
1156 for (i = 0; i < dev->num_tx_queues; i++) {
1157 struct netdev_queue *dev_queue;
1158 spinlock_t *root_lock;
1159 struct Qdisc *q;
1160 int val;
1162 dev_queue = netdev_get_tx_queue(dev, i);
1163 q = dev_queue->qdisc_sleeping;
1165 root_lock = qdisc_lock(q);
1166 spin_lock_bh(root_lock);
1168 val = (qdisc_is_running(q) ||
1169 test_bit(__QDISC_STATE_SCHED, &q->state));
1171 spin_unlock_bh(root_lock);
1173 if (val)
1174 return true;
1176 return false;
1179 static void dev_qdisc_reset(struct net_device *dev,
1180 struct netdev_queue *dev_queue,
1181 void *none)
1183 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
1185 if (qdisc)
1186 qdisc_reset(qdisc);
1190 * dev_deactivate_many - deactivate transmissions on several devices
1191 * @head: list of devices to deactivate
1193 * This function returns only when all outstanding transmissions
1194 * have completed, unless all devices are in dismantle phase.
1196 void dev_deactivate_many(struct list_head *head)
1198 struct net_device *dev;
1200 list_for_each_entry(dev, head, close_list) {
1201 netdev_for_each_tx_queue(dev, dev_deactivate_queue,
1202 &noop_qdisc);
1203 if (dev_ingress_queue(dev))
1204 dev_deactivate_queue(dev, dev_ingress_queue(dev),
1205 &noop_qdisc);
1207 dev_watchdog_down(dev);
1210 /* Wait for outstanding qdisc-less dev_queue_xmit calls.
1211 * This is avoided if all devices are in dismantle phase :
1212 * Caller will call synchronize_net() for us
1214 synchronize_net();
1216 /* Wait for outstanding qdisc_run calls. */
1217 list_for_each_entry(dev, head, close_list) {
1218 while (some_qdisc_is_busy(dev))
1219 yield();
1220 /* The new qdisc is assigned at this point so we can safely
1221 * unwind stale skb lists and qdisc statistics
1223 netdev_for_each_tx_queue(dev, dev_qdisc_reset, NULL);
1224 if (dev_ingress_queue(dev))
1225 dev_qdisc_reset(dev, dev_ingress_queue(dev), NULL);
1229 void dev_deactivate(struct net_device *dev)
1231 LIST_HEAD(single);
1233 list_add(&dev->close_list, &single);
1234 dev_deactivate_many(&single);
1235 list_del(&single);
1237 EXPORT_SYMBOL(dev_deactivate);
1239 static int qdisc_change_tx_queue_len(struct net_device *dev,
1240 struct netdev_queue *dev_queue)
1242 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
1243 const struct Qdisc_ops *ops = qdisc->ops;
1245 if (ops->change_tx_queue_len)
1246 return ops->change_tx_queue_len(qdisc, dev->tx_queue_len);
1247 return 0;
1250 int dev_qdisc_change_tx_queue_len(struct net_device *dev)
1252 bool up = dev->flags & IFF_UP;
1253 unsigned int i;
1254 int ret = 0;
1256 if (up)
1257 dev_deactivate(dev);
1259 for (i = 0; i < dev->num_tx_queues; i++) {
1260 ret = qdisc_change_tx_queue_len(dev, &dev->_tx[i]);
1262 /* TODO: revert changes on a partial failure */
1263 if (ret)
1264 break;
1267 if (up)
1268 dev_activate(dev);
1269 return ret;
1272 static void dev_init_scheduler_queue(struct net_device *dev,
1273 struct netdev_queue *dev_queue,
1274 void *_qdisc)
1276 struct Qdisc *qdisc = _qdisc;
1278 rcu_assign_pointer(dev_queue->qdisc, qdisc);
1279 dev_queue->qdisc_sleeping = qdisc;
1282 void dev_init_scheduler(struct net_device *dev)
1284 dev->qdisc = &noop_qdisc;
1285 netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
1286 if (dev_ingress_queue(dev))
1287 dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
1289 timer_setup(&dev->watchdog_timer, dev_watchdog, 0);
1292 static void shutdown_scheduler_queue(struct net_device *dev,
1293 struct netdev_queue *dev_queue,
1294 void *_qdisc_default)
1296 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
1297 struct Qdisc *qdisc_default = _qdisc_default;
1299 if (qdisc) {
1300 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
1301 dev_queue->qdisc_sleeping = qdisc_default;
1303 qdisc_put(qdisc);
1307 void dev_shutdown(struct net_device *dev)
1309 netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
1310 if (dev_ingress_queue(dev))
1311 shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
1312 qdisc_put(dev->qdisc);
1313 dev->qdisc = &noop_qdisc;
1315 WARN_ON(timer_pending(&dev->watchdog_timer));
1318 void psched_ratecfg_precompute(struct psched_ratecfg *r,
1319 const struct tc_ratespec *conf,
1320 u64 rate64)
1322 memset(r, 0, sizeof(*r));
1323 r->overhead = conf->overhead;
1324 r->rate_bytes_ps = max_t(u64, conf->rate, rate64);
1325 r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK);
1326 r->mult = 1;
1328 * The deal here is to replace a divide by a reciprocal one
1329 * in fast path (a reciprocal divide is a multiply and a shift)
1331 * Normal formula would be :
1332 * time_in_ns = (NSEC_PER_SEC * len) / rate_bps
1334 * We compute mult/shift to use instead :
1335 * time_in_ns = (len * mult) >> shift;
1337 * We try to get the highest possible mult value for accuracy,
1338 * but have to make sure no overflows will ever happen.
1340 if (r->rate_bytes_ps > 0) {
1341 u64 factor = NSEC_PER_SEC;
1343 for (;;) {
1344 r->mult = div64_u64(factor, r->rate_bytes_ps);
1345 if (r->mult & (1U << 31) || factor & (1ULL << 63))
1346 break;
1347 factor <<= 1;
1348 r->shift++;
1352 EXPORT_SYMBOL(psched_ratecfg_precompute);
1354 static void mini_qdisc_rcu_func(struct rcu_head *head)
1358 void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1359 struct tcf_proto *tp_head)
1361 /* Protected with chain0->filter_chain_lock.
1362 * Can't access chain directly because tp_head can be NULL.
1364 struct mini_Qdisc *miniq_old =
1365 rcu_dereference_protected(*miniqp->p_miniq, 1);
1366 struct mini_Qdisc *miniq;
1368 if (!tp_head) {
1369 RCU_INIT_POINTER(*miniqp->p_miniq, NULL);
1370 /* Wait for flying RCU callback before it is freed. */
1371 rcu_barrier();
1372 return;
1375 miniq = !miniq_old || miniq_old == &miniqp->miniq2 ?
1376 &miniqp->miniq1 : &miniqp->miniq2;
1378 /* We need to make sure that readers won't see the miniq
1379 * we are about to modify. So wait until previous call_rcu callback
1380 * is done.
1382 rcu_barrier();
1383 miniq->filter_list = tp_head;
1384 rcu_assign_pointer(*miniqp->p_miniq, miniq);
1386 if (miniq_old)
1387 /* This is counterpart of the rcu barriers above. We need to
1388 * block potential new user of miniq_old until all readers
1389 * are not seeing it.
1391 call_rcu(&miniq_old->rcu, mini_qdisc_rcu_func);
1393 EXPORT_SYMBOL(mini_qdisc_pair_swap);
1395 void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1396 struct mini_Qdisc __rcu **p_miniq)
1398 miniqp->miniq1.cpu_bstats = qdisc->cpu_bstats;
1399 miniqp->miniq1.cpu_qstats = qdisc->cpu_qstats;
1400 miniqp->miniq2.cpu_bstats = qdisc->cpu_bstats;
1401 miniqp->miniq2.cpu_qstats = qdisc->cpu_qstats;
1402 miniqp->p_miniq = p_miniq;
1404 EXPORT_SYMBOL(mini_qdisc_pair_init);