1
2
3
4
5
6
7
8
9
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>
33
34
35const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops;
36EXPORT_SYMBOL(default_qdisc_ops);
37
38
39
40
41
42
43
44
45
46
47
48
49static inline struct sk_buff *__skb_dequeue_bad_txq(struct Qdisc *q)
50{
51 const struct netdev_queue *txq = q->dev_queue;
52 spinlock_t *lock = NULL;
53 struct sk_buff *skb;
54
55 if (q->flags & TCQ_F_NOLOCK) {
56 lock = qdisc_lock(q);
57 spin_lock(lock);
58 }
59
60 skb = skb_peek(&q->skb_bad_txq);
61 if (skb) {
62
63 txq = skb_get_tx_queue(txq->dev, skb);
64 if (!netif_xmit_frozen_or_stopped(txq)) {
65 skb = __skb_dequeue(&q->skb_bad_txq);
66 if (qdisc_is_percpu_stats(q)) {
67 qdisc_qstats_cpu_backlog_dec(q, skb);
68 qdisc_qstats_cpu_qlen_dec(q);
69 } else {
70 qdisc_qstats_backlog_dec(q, skb);
71 q->q.qlen--;
72 }
73 } else {
74 skb = NULL;
75 }
76 }
77
78 if (lock)
79 spin_unlock(lock);
80
81 return skb;
82}
83
84static inline struct sk_buff *qdisc_dequeue_skb_bad_txq(struct Qdisc *q)
85{
86 struct sk_buff *skb = skb_peek(&q->skb_bad_txq);
87
88 if (unlikely(skb))
89 skb = __skb_dequeue_bad_txq(q);
90
91 return skb;
92}
93
94static inline void qdisc_enqueue_skb_bad_txq(struct Qdisc *q,
95 struct sk_buff *skb)
96{
97 spinlock_t *lock = NULL;
98
99 if (q->flags & TCQ_F_NOLOCK) {
100 lock = qdisc_lock(q);
101 spin_lock(lock);
102 }
103
104 __skb_queue_tail(&q->skb_bad_txq, skb);
105
106 if (qdisc_is_percpu_stats(q)) {
107 qdisc_qstats_cpu_backlog_inc(q, skb);
108 qdisc_qstats_cpu_qlen_inc(q);
109 } else {
110 qdisc_qstats_backlog_inc(q, skb);
111 q->q.qlen++;
112 }
113
114 if (lock)
115 spin_unlock(lock);
116}
117
118static inline void dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
119{
120 spinlock_t *lock = NULL;
121
122 if (q->flags & TCQ_F_NOLOCK) {
123 lock = qdisc_lock(q);
124 spin_lock(lock);
125 }
126
127 while (skb) {
128 struct sk_buff *next = skb->next;
129
130 __skb_queue_tail(&q->gso_skb, skb);
131
132
133 if (qdisc_is_percpu_stats(q)) {
134 qdisc_qstats_cpu_requeues_inc(q);
135 qdisc_qstats_cpu_backlog_inc(q, skb);
136 qdisc_qstats_cpu_qlen_inc(q);
137 } else {
138 q->qstats.requeues++;
139 qdisc_qstats_backlog_inc(q, skb);
140 q->q.qlen++;
141 }
142
143 skb = next;
144 }
145 if (lock)
146 spin_unlock(lock);
147 __netif_schedule(q);
148}
149
150static void try_bulk_dequeue_skb(struct Qdisc *q,
151 struct sk_buff *skb,
152 const struct netdev_queue *txq,
153 int *packets)
154{
155 int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
156
157 while (bytelimit > 0) {
158 struct sk_buff *nskb = q->dequeue(q);
159
160 if (!nskb)
161 break;
162
163 bytelimit -= nskb->len;
164 skb->next = nskb;
165 skb = nskb;
166 (*packets)++;
167 }
168 skb_mark_not_on_list(skb);
169}
170
171
172
173
174static void try_bulk_dequeue_skb_slow(struct Qdisc *q,
175 struct sk_buff *skb,
176 int *packets)
177{
178 int mapping = skb_get_queue_mapping(skb);
179 struct sk_buff *nskb;
180 int cnt = 0;
181
182 do {
183 nskb = q->dequeue(q);
184 if (!nskb)
185 break;
186 if (unlikely(skb_get_queue_mapping(nskb) != mapping)) {
187 qdisc_enqueue_skb_bad_txq(q, nskb);
188 break;
189 }
190 skb->next = nskb;
191 skb = nskb;
192 } while (++cnt < 8);
193 (*packets) += cnt;
194 skb_mark_not_on_list(skb);
195}
196
197
198
199
200static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
201 int *packets)
202{
203 const struct netdev_queue *txq = q->dev_queue;
204 struct sk_buff *skb = NULL;
205
206 *packets = 1;
207 if (unlikely(!skb_queue_empty(&q->gso_skb))) {
208 spinlock_t *lock = NULL;
209
210 if (q->flags & TCQ_F_NOLOCK) {
211 lock = qdisc_lock(q);
212 spin_lock(lock);
213 }
214
215 skb = skb_peek(&q->gso_skb);
216
217
218
219
220 if (!skb) {
221 if (lock)
222 spin_unlock(lock);
223 goto validate;
224 }
225
226
227 *validate = false;
228 if (xfrm_offload(skb))
229 *validate = true;
230
231 txq = skb_get_tx_queue(txq->dev, skb);
232 if (!netif_xmit_frozen_or_stopped(txq)) {
233 skb = __skb_dequeue(&q->gso_skb);
234 if (qdisc_is_percpu_stats(q)) {
235 qdisc_qstats_cpu_backlog_dec(q, skb);
236 qdisc_qstats_cpu_qlen_dec(q);
237 } else {
238 qdisc_qstats_backlog_dec(q, skb);
239 q->q.qlen--;
240 }
241 } else {
242 skb = NULL;
243 }
244 if (lock)
245 spin_unlock(lock);
246 goto trace;
247 }
248validate:
249 *validate = true;
250
251 if ((q->flags & TCQ_F_ONETXQUEUE) &&
252 netif_xmit_frozen_or_stopped(txq))
253 return skb;
254
255 skb = qdisc_dequeue_skb_bad_txq(q);
256 if (unlikely(skb))
257 goto bulk;
258 skb = q->dequeue(q);
259 if (skb) {
260bulk:
261 if (qdisc_may_bulk(q))
262 try_bulk_dequeue_skb(q, skb, txq, packets);
263 else
264 try_bulk_dequeue_skb_slow(q, skb, packets);
265 }
266trace:
267 trace_qdisc_dequeue(q, txq, *packets, skb);
268 return skb;
269}
270
271
272
273
274
275
276
277
278
279
280bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
281 struct net_device *dev, struct netdev_queue *txq,
282 spinlock_t *root_lock, bool validate)
283{
284 int ret = NETDEV_TX_BUSY;
285 bool again = false;
286
287
288 if (root_lock)
289 spin_unlock(root_lock);
290
291
292 if (validate)
293 skb = validate_xmit_skb_list(skb, dev, &again);
294
295#ifdef CONFIG_XFRM_OFFLOAD
296 if (unlikely(again)) {
297 if (root_lock)
298 spin_lock(root_lock);
299
300 dev_requeue_skb(skb, q);
301 return false;
302 }
303#endif
304
305 if (likely(skb)) {
306 HARD_TX_LOCK(dev, txq, smp_processor_id());
307 if (!netif_xmit_frozen_or_stopped(txq))
308 skb = dev_hard_start_xmit(skb, dev, txq, &ret);
309
310 HARD_TX_UNLOCK(dev, txq);
311 } else {
312 if (root_lock)
313 spin_lock(root_lock);
314 return true;
315 }
316
317 if (root_lock)
318 spin_lock(root_lock);
319
320 if (!dev_xmit_complete(ret)) {
321
322 if (unlikely(ret != NETDEV_TX_BUSY))
323 net_warn_ratelimited("BUG %s code %d qlen %d\n",
324 dev->name, ret, q->q.qlen);
325
326 dev_requeue_skb(skb, q);
327 return false;
328 }
329
330 return true;
331}
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352static inline bool qdisc_restart(struct Qdisc *q, int *packets)
353{
354 spinlock_t *root_lock = NULL;
355 struct netdev_queue *txq;
356 struct net_device *dev;
357 struct sk_buff *skb;
358 bool validate;
359
360
361 skb = dequeue_skb(q, &validate, packets);
362 if (unlikely(!skb))
363 return false;
364
365 if (!(q->flags & TCQ_F_NOLOCK))
366 root_lock = qdisc_lock(q);
367
368 dev = qdisc_dev(q);
369 txq = skb_get_tx_queue(dev, skb);
370
371 return sch_direct_xmit(skb, q, dev, txq, root_lock, validate);
372}
373
374void __qdisc_run(struct Qdisc *q)
375{
376 int quota = dev_tx_weight;
377 int packets;
378
379 while (qdisc_restart(q, &packets)) {
380
381
382
383
384
385 quota -= packets;
386 if (quota <= 0 || need_resched()) {
387 __netif_schedule(q);
388 break;
389 }
390 }
391}
392
393unsigned long dev_trans_start(struct net_device *dev)
394{
395 unsigned long val, res;
396 unsigned int i;
397
398 if (is_vlan_dev(dev))
399 dev = vlan_dev_real_dev(dev);
400 else if (netif_is_macvlan(dev))
401 dev = macvlan_dev_real_dev(dev);
402 res = netdev_get_tx_queue(dev, 0)->trans_start;
403 for (i = 1; i < dev->num_tx_queues; i++) {
404 val = netdev_get_tx_queue(dev, i)->trans_start;
405 if (val && time_after(val, res))
406 res = val;
407 }
408
409 return res;
410}
411EXPORT_SYMBOL(dev_trans_start);
412
413static void dev_watchdog(struct timer_list *t)
414{
415 struct net_device *dev = from_timer(dev, t, watchdog_timer);
416
417 netif_tx_lock(dev);
418 if (!qdisc_tx_is_noop(dev)) {
419 if (netif_device_present(dev) &&
420 netif_running(dev) &&
421 netif_carrier_ok(dev)) {
422 int some_queue_timedout = 0;
423 unsigned int i;
424 unsigned long trans_start;
425
426 for (i = 0; i < dev->num_tx_queues; i++) {
427 struct netdev_queue *txq;
428
429 txq = netdev_get_tx_queue(dev, i);
430 trans_start = txq->trans_start;
431 if (netif_xmit_stopped(txq) &&
432 time_after(jiffies, (trans_start +
433 dev->watchdog_timeo))) {
434 some_queue_timedout = 1;
435 txq->trans_timeout++;
436 break;
437 }
438 }
439
440 if (some_queue_timedout) {
441 trace_net_dev_xmit_timeout(dev, i);
442 WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
443 dev->name, netdev_drivername(dev), i);
444 dev->netdev_ops->ndo_tx_timeout(dev);
445 }
446 if (!mod_timer(&dev->watchdog_timer,
447 round_jiffies(jiffies +
448 dev->watchdog_timeo)))
449 dev_hold(dev);
450 }
451 }
452 netif_tx_unlock(dev);
453
454 dev_put(dev);
455}
456
457void __netdev_watchdog_up(struct net_device *dev)
458{
459 if (dev->netdev_ops->ndo_tx_timeout) {
460 if (dev->watchdog_timeo <= 0)
461 dev->watchdog_timeo = 5*HZ;
462 if (!mod_timer(&dev->watchdog_timer,
463 round_jiffies(jiffies + dev->watchdog_timeo)))
464 dev_hold(dev);
465 }
466}
467
468static void dev_watchdog_up(struct net_device *dev)
469{
470 __netdev_watchdog_up(dev);
471}
472
473static void dev_watchdog_down(struct net_device *dev)
474{
475 netif_tx_lock_bh(dev);
476 if (del_timer(&dev->watchdog_timer))
477 dev_put(dev);
478 netif_tx_unlock_bh(dev);
479}
480
481
482
483
484
485
486
487void netif_carrier_on(struct net_device *dev)
488{
489 if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
490 if (dev->reg_state == NETREG_UNINITIALIZED)
491 return;
492 atomic_inc(&dev->carrier_up_count);
493 linkwatch_fire_event(dev);
494 if (netif_running(dev))
495 __netdev_watchdog_up(dev);
496 }
497}
498EXPORT_SYMBOL(netif_carrier_on);
499
500
501
502
503
504
505
506void netif_carrier_off(struct net_device *dev)
507{
508 if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
509 if (dev->reg_state == NETREG_UNINITIALIZED)
510 return;
511 atomic_inc(&dev->carrier_down_count);
512 linkwatch_fire_event(dev);
513 }
514}
515EXPORT_SYMBOL(netif_carrier_off);
516
517
518
519
520
521
522static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc,
523 struct sk_buff **to_free)
524{
525 __qdisc_drop(skb, to_free);
526 return NET_XMIT_CN;
527}
528
529static struct sk_buff *noop_dequeue(struct Qdisc *qdisc)
530{
531 return NULL;
532}
533
534struct Qdisc_ops noop_qdisc_ops __read_mostly = {
535 .id = "noop",
536 .priv_size = 0,
537 .enqueue = noop_enqueue,
538 .dequeue = noop_dequeue,
539 .peek = noop_dequeue,
540 .owner = THIS_MODULE,
541};
542
543static struct netdev_queue noop_netdev_queue = {
544 RCU_POINTER_INITIALIZER(qdisc, &noop_qdisc),
545 .qdisc_sleeping = &noop_qdisc,
546};
547
548struct Qdisc noop_qdisc = {
549 .enqueue = noop_enqueue,
550 .dequeue = noop_dequeue,
551 .flags = TCQ_F_BUILTIN,
552 .ops = &noop_qdisc_ops,
553 .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
554 .dev_queue = &noop_netdev_queue,
555 .running = SEQCNT_ZERO(noop_qdisc.running),
556 .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
557 .gso_skb = {
558 .next = (struct sk_buff *)&noop_qdisc.gso_skb,
559 .prev = (struct sk_buff *)&noop_qdisc.gso_skb,
560 .qlen = 0,
561 .lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.gso_skb.lock),
562 },
563 .skb_bad_txq = {
564 .next = (struct sk_buff *)&noop_qdisc.skb_bad_txq,
565 .prev = (struct sk_buff *)&noop_qdisc.skb_bad_txq,
566 .qlen = 0,
567 .lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.skb_bad_txq.lock),
568 },
569};
570EXPORT_SYMBOL(noop_qdisc);
571
572static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt,
573 struct netlink_ext_ack *extack)
574{
575
576
577
578 qdisc->enqueue = NULL;
579 return 0;
580}
581
582struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
583 .id = "noqueue",
584 .priv_size = 0,
585 .init = noqueue_init,
586 .enqueue = noop_enqueue,
587 .dequeue = noop_dequeue,
588 .peek = noop_dequeue,
589 .owner = THIS_MODULE,
590};
591
592static const u8 prio2band[TC_PRIO_MAX + 1] = {
593 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
594};
595
596
597
598
599
600#define PFIFO_FAST_BANDS 3
601
602
603
604
605
606struct pfifo_fast_priv {
607 struct skb_array q[PFIFO_FAST_BANDS];
608};
609
610static inline struct skb_array *band2list(struct pfifo_fast_priv *priv,
611 int band)
612{
613 return &priv->q[band];
614}
615
616static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc,
617 struct sk_buff **to_free)
618{
619 int band = prio2band[skb->priority & TC_PRIO_MAX];
620 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
621 struct skb_array *q = band2list(priv, band);
622 unsigned int pkt_len = qdisc_pkt_len(skb);
623 int err;
624
625 err = skb_array_produce(q, skb);
626
627 if (unlikely(err))
628 return qdisc_drop_cpu(skb, qdisc, to_free);
629
630 qdisc_update_stats_at_enqueue(qdisc, pkt_len);
631 return NET_XMIT_SUCCESS;
632}
633
634static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
635{
636 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
637 struct sk_buff *skb = NULL;
638 int band;
639
640 for (band = 0; band < PFIFO_FAST_BANDS && !skb; band++) {
641 struct skb_array *q = band2list(priv, band);
642
643 if (__skb_array_empty(q))
644 continue;
645
646 skb = __skb_array_consume(q);
647 }
648 if (likely(skb)) {
649 qdisc_update_stats_at_dequeue(qdisc, skb);
650 } else {
651 qdisc->empty = true;
652 }
653
654 return skb;
655}
656
657static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc)
658{
659 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
660 struct sk_buff *skb = NULL;
661 int band;
662
663 for (band = 0; band < PFIFO_FAST_BANDS && !skb; band++) {
664 struct skb_array *q = band2list(priv, band);
665
666 skb = __skb_array_peek(q);
667 }
668
669 return skb;
670}
671
672static void pfifo_fast_reset(struct Qdisc *qdisc)
673{
674 int i, band;
675 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
676
677 for (band = 0; band < PFIFO_FAST_BANDS; band++) {
678 struct skb_array *q = band2list(priv, band);
679 struct sk_buff *skb;
680
681
682
683
684 if (!q->ring.queue)
685 continue;
686
687 while ((skb = __skb_array_consume(q)) != NULL)
688 kfree_skb(skb);
689 }
690
691 for_each_possible_cpu(i) {
692 struct gnet_stats_queue *q = per_cpu_ptr(qdisc->cpu_qstats, i);
693
694 q->backlog = 0;
695 q->qlen = 0;
696 }
697}
698
699static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
700{
701 struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
702
703 memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1);
704 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
705 goto nla_put_failure;
706 return skb->len;
707
708nla_put_failure:
709 return -1;
710}
711
712static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt,
713 struct netlink_ext_ack *extack)
714{
715 unsigned int qlen = qdisc_dev(qdisc)->tx_queue_len;
716 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
717 int prio;
718
719
720 if (!qlen)
721 return -EINVAL;
722
723 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
724 struct skb_array *q = band2list(priv, prio);
725 int err;
726
727 err = skb_array_init(q, qlen, GFP_KERNEL);
728 if (err)
729 return -ENOMEM;
730 }
731
732
733 qdisc->flags |= TCQ_F_CAN_BYPASS;
734 return 0;
735}
736
737static void pfifo_fast_destroy(struct Qdisc *sch)
738{
739 struct pfifo_fast_priv *priv = qdisc_priv(sch);
740 int prio;
741
742 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
743 struct skb_array *q = band2list(priv, prio);
744
745
746
747
748 if (!q->ring.queue)
749 continue;
750
751
752
753 ptr_ring_cleanup(&q->ring, NULL);
754 }
755}
756
757static int pfifo_fast_change_tx_queue_len(struct Qdisc *sch,
758 unsigned int new_len)
759{
760 struct pfifo_fast_priv *priv = qdisc_priv(sch);
761 struct skb_array *bands[PFIFO_FAST_BANDS];
762 int prio;
763
764 for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) {
765 struct skb_array *q = band2list(priv, prio);
766
767 bands[prio] = q;
768 }
769
770 return skb_array_resize_multiple(bands, PFIFO_FAST_BANDS, new_len,
771 GFP_KERNEL);
772}
773
774struct Qdisc_ops pfifo_fast_ops __read_mostly = {
775 .id = "pfifo_fast",
776 .priv_size = sizeof(struct pfifo_fast_priv),
777 .enqueue = pfifo_fast_enqueue,
778 .dequeue = pfifo_fast_dequeue,
779 .peek = pfifo_fast_peek,
780 .init = pfifo_fast_init,
781 .destroy = pfifo_fast_destroy,
782 .reset = pfifo_fast_reset,
783 .dump = pfifo_fast_dump,
784 .change_tx_queue_len = pfifo_fast_change_tx_queue_len,
785 .owner = THIS_MODULE,
786 .static_flags = TCQ_F_NOLOCK | TCQ_F_CPUSTATS,
787};
788EXPORT_SYMBOL(pfifo_fast_ops);
789
790static struct lock_class_key qdisc_tx_busylock;
791static struct lock_class_key qdisc_running_key;
792
793struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
794 const struct Qdisc_ops *ops,
795 struct netlink_ext_ack *extack)
796{
797 void *p;
798 struct Qdisc *sch;
799 unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size;
800 int err = -ENOBUFS;
801 struct net_device *dev;
802
803 if (!dev_queue) {
804 NL_SET_ERR_MSG(extack, "No device queue given");
805 err = -EINVAL;
806 goto errout;
807 }
808
809 dev = dev_queue->dev;
810 p = kzalloc_node(size, GFP_KERNEL,
811 netdev_queue_numa_node_read(dev_queue));
812
813 if (!p)
814 goto errout;
815 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
816
817 if (sch != p) {
818 kfree(p);
819 p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL,
820 netdev_queue_numa_node_read(dev_queue));
821 if (!p)
822 goto errout;
823 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
824 sch->padded = (char *) sch - (char *) p;
825 }
826 __skb_queue_head_init(&sch->gso_skb);
827 __skb_queue_head_init(&sch->skb_bad_txq);
828 qdisc_skb_head_init(&sch->q);
829 spin_lock_init(&sch->q.lock);
830
831 if (ops->static_flags & TCQ_F_CPUSTATS) {
832 sch->cpu_bstats =
833 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
834 if (!sch->cpu_bstats)
835 goto errout1;
836
837 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
838 if (!sch->cpu_qstats) {
839 free_percpu(sch->cpu_bstats);
840 goto errout1;
841 }
842 }
843
844 spin_lock_init(&sch->busylock);
845 lockdep_set_class(&sch->busylock,
846 dev->qdisc_tx_busylock ?: &qdisc_tx_busylock);
847
848
849 spin_lock_init(&sch->seqlock);
850 lockdep_set_class(&sch->busylock,
851 dev->qdisc_tx_busylock ?: &qdisc_tx_busylock);
852
853 seqcount_init(&sch->running);
854 lockdep_set_class(&sch->running,
855 dev->qdisc_running_key ?: &qdisc_running_key);
856
857 sch->ops = ops;
858 sch->flags = ops->static_flags;
859 sch->enqueue = ops->enqueue;
860 sch->dequeue = ops->dequeue;
861 sch->dev_queue = dev_queue;
862 sch->empty = true;
863 dev_hold(dev);
864 refcount_set(&sch->refcnt, 1);
865
866 return sch;
867errout1:
868 kfree(p);
869errout:
870 return ERR_PTR(err);
871}
872
873struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
874 const struct Qdisc_ops *ops,
875 unsigned int parentid,
876 struct netlink_ext_ack *extack)
877{
878 struct Qdisc *sch;
879
880 if (!try_module_get(ops->owner)) {
881 NL_SET_ERR_MSG(extack, "Failed to increase module reference counter");
882 return NULL;
883 }
884
885 sch = qdisc_alloc(dev_queue, ops, extack);
886 if (IS_ERR(sch)) {
887 module_put(ops->owner);
888 return NULL;
889 }
890 sch->parent = parentid;
891
892 if (!ops->init || ops->init(sch, NULL, extack) == 0)
893 return sch;
894
895 qdisc_put(sch);
896 return NULL;
897}
898EXPORT_SYMBOL(qdisc_create_dflt);
899
900
901
902void qdisc_reset(struct Qdisc *qdisc)
903{
904 const struct Qdisc_ops *ops = qdisc->ops;
905 struct sk_buff *skb, *tmp;
906
907 if (ops->reset)
908 ops->reset(qdisc);
909
910 skb_queue_walk_safe(&qdisc->gso_skb, skb, tmp) {
911 __skb_unlink(skb, &qdisc->gso_skb);
912 kfree_skb_list(skb);
913 }
914
915 skb_queue_walk_safe(&qdisc->skb_bad_txq, skb, tmp) {
916 __skb_unlink(skb, &qdisc->skb_bad_txq);
917 kfree_skb_list(skb);
918 }
919
920 qdisc->q.qlen = 0;
921 qdisc->qstats.backlog = 0;
922}
923EXPORT_SYMBOL(qdisc_reset);
924
925void qdisc_free(struct Qdisc *qdisc)
926{
927 if (qdisc_is_percpu_stats(qdisc)) {
928 free_percpu(qdisc->cpu_bstats);
929 free_percpu(qdisc->cpu_qstats);
930 }
931
932 kfree((char *) qdisc - qdisc->padded);
933}
934
935static void qdisc_free_cb(struct rcu_head *head)
936{
937 struct Qdisc *q = container_of(head, struct Qdisc, rcu);
938
939 qdisc_free(q);
940}
941
942static void qdisc_destroy(struct Qdisc *qdisc)
943{
944 const struct Qdisc_ops *ops = qdisc->ops;
945 struct sk_buff *skb, *tmp;
946
947#ifdef CONFIG_NET_SCHED
948 qdisc_hash_del(qdisc);
949
950 qdisc_put_stab(rtnl_dereference(qdisc->stab));
951#endif
952 gen_kill_estimator(&qdisc->rate_est);
953 if (ops->reset)
954 ops->reset(qdisc);
955 if (ops->destroy)
956 ops->destroy(qdisc);
957
958 module_put(ops->owner);
959 dev_put(qdisc_dev(qdisc));
960
961 skb_queue_walk_safe(&qdisc->gso_skb, skb, tmp) {
962 __skb_unlink(skb, &qdisc->gso_skb);
963 kfree_skb_list(skb);
964 }
965
966 skb_queue_walk_safe(&qdisc->skb_bad_txq, skb, tmp) {
967 __skb_unlink(skb, &qdisc->skb_bad_txq);
968 kfree_skb_list(skb);
969 }
970
971 call_rcu(&qdisc->rcu, qdisc_free_cb);
972}
973
974void qdisc_put(struct Qdisc *qdisc)
975{
976 if (qdisc->flags & TCQ_F_BUILTIN ||
977 !refcount_dec_and_test(&qdisc->refcnt))
978 return;
979
980 qdisc_destroy(qdisc);
981}
982EXPORT_SYMBOL(qdisc_put);
983
984
985
986
987
988
989void qdisc_put_unlocked(struct Qdisc *qdisc)
990{
991 if (qdisc->flags & TCQ_F_BUILTIN ||
992 !refcount_dec_and_rtnl_lock(&qdisc->refcnt))
993 return;
994
995 qdisc_destroy(qdisc);
996 rtnl_unlock();
997}
998EXPORT_SYMBOL(qdisc_put_unlocked);
999
1000
1001struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
1002 struct Qdisc *qdisc)
1003{
1004 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
1005 spinlock_t *root_lock;
1006
1007 root_lock = qdisc_lock(oqdisc);
1008 spin_lock_bh(root_lock);
1009
1010
1011 if (qdisc == NULL)
1012 qdisc = &noop_qdisc;
1013 dev_queue->qdisc_sleeping = qdisc;
1014 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
1015
1016 spin_unlock_bh(root_lock);
1017
1018 return oqdisc;
1019}
1020EXPORT_SYMBOL(dev_graft_qdisc);
1021
1022static void attach_one_default_qdisc(struct net_device *dev,
1023 struct netdev_queue *dev_queue,
1024 void *_unused)
1025{
1026 struct Qdisc *qdisc;
1027 const struct Qdisc_ops *ops = default_qdisc_ops;
1028
1029 if (dev->priv_flags & IFF_NO_QUEUE)
1030 ops = &noqueue_qdisc_ops;
1031
1032 qdisc = qdisc_create_dflt(dev_queue, ops, TC_H_ROOT, NULL);
1033 if (!qdisc) {
1034 netdev_info(dev, "activation failed\n");
1035 return;
1036 }
1037 if (!netif_is_multiqueue(dev))
1038 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
1039 dev_queue->qdisc_sleeping = qdisc;
1040}
1041
1042static void attach_default_qdiscs(struct net_device *dev)
1043{
1044 struct netdev_queue *txq;
1045 struct Qdisc *qdisc;
1046
1047 txq = netdev_get_tx_queue(dev, 0);
1048
1049 if (!netif_is_multiqueue(dev) ||
1050 dev->priv_flags & IFF_NO_QUEUE) {
1051 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
1052 dev->qdisc = txq->qdisc_sleeping;
1053 qdisc_refcount_inc(dev->qdisc);
1054 } else {
1055 qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT, NULL);
1056 if (qdisc) {
1057 dev->qdisc = qdisc;
1058 qdisc->ops->attach(qdisc);
1059 }
1060 }
1061#ifdef CONFIG_NET_SCHED
1062 if (dev->qdisc != &noop_qdisc)
1063 qdisc_hash_add(dev->qdisc, false);
1064#endif
1065}
1066
1067static void transition_one_qdisc(struct net_device *dev,
1068 struct netdev_queue *dev_queue,
1069 void *_need_watchdog)
1070{
1071 struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
1072 int *need_watchdog_p = _need_watchdog;
1073
1074 if (!(new_qdisc->flags & TCQ_F_BUILTIN))
1075 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
1076
1077 rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
1078 if (need_watchdog_p) {
1079 dev_queue->trans_start = 0;
1080 *need_watchdog_p = 1;
1081 }
1082}
1083
1084void dev_activate(struct net_device *dev)
1085{
1086 int need_watchdog;
1087
1088
1089
1090
1091
1092
1093 if (dev->qdisc == &noop_qdisc)
1094 attach_default_qdiscs(dev);
1095
1096 if (!netif_carrier_ok(dev))
1097
1098 return;
1099
1100 need_watchdog = 0;
1101 netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
1102 if (dev_ingress_queue(dev))
1103 transition_one_qdisc(dev, dev_ingress_queue(dev), NULL);
1104
1105 if (need_watchdog) {
1106 netif_trans_update(dev);
1107 dev_watchdog_up(dev);
1108 }
1109}
1110EXPORT_SYMBOL(dev_activate);
1111
1112static void dev_deactivate_queue(struct net_device *dev,
1113 struct netdev_queue *dev_queue,
1114 void *_qdisc_default)
1115{
1116 struct Qdisc *qdisc_default = _qdisc_default;
1117 struct Qdisc *qdisc;
1118
1119 qdisc = rtnl_dereference(dev_queue->qdisc);
1120 if (qdisc) {
1121 bool nolock = qdisc->flags & TCQ_F_NOLOCK;
1122
1123 if (nolock)
1124 spin_lock_bh(&qdisc->seqlock);
1125 spin_lock_bh(qdisc_lock(qdisc));
1126
1127 if (!(qdisc->flags & TCQ_F_BUILTIN))
1128 set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
1129
1130 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
1131 qdisc_reset(qdisc);
1132
1133 spin_unlock_bh(qdisc_lock(qdisc));
1134 if (nolock)
1135 spin_unlock_bh(&qdisc->seqlock);
1136 }
1137}
1138
1139static bool some_qdisc_is_busy(struct net_device *dev)
1140{
1141 unsigned int i;
1142
1143 for (i = 0; i < dev->num_tx_queues; i++) {
1144 struct netdev_queue *dev_queue;
1145 spinlock_t *root_lock;
1146 struct Qdisc *q;
1147 int val;
1148
1149 dev_queue = netdev_get_tx_queue(dev, i);
1150 q = dev_queue->qdisc_sleeping;
1151
1152 root_lock = qdisc_lock(q);
1153 spin_lock_bh(root_lock);
1154
1155 val = (qdisc_is_running(q) ||
1156 test_bit(__QDISC_STATE_SCHED, &q->state));
1157
1158 spin_unlock_bh(root_lock);
1159
1160 if (val)
1161 return true;
1162 }
1163 return false;
1164}
1165
1166static void dev_qdisc_reset(struct net_device *dev,
1167 struct netdev_queue *dev_queue,
1168 void *none)
1169{
1170 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
1171
1172 if (qdisc)
1173 qdisc_reset(qdisc);
1174}
1175
1176
1177
1178
1179
1180
1181
1182
1183void dev_deactivate_many(struct list_head *head)
1184{
1185 struct net_device *dev;
1186
1187 list_for_each_entry(dev, head, close_list) {
1188 netdev_for_each_tx_queue(dev, dev_deactivate_queue,
1189 &noop_qdisc);
1190 if (dev_ingress_queue(dev))
1191 dev_deactivate_queue(dev, dev_ingress_queue(dev),
1192 &noop_qdisc);
1193
1194 dev_watchdog_down(dev);
1195 }
1196
1197
1198
1199
1200
1201 synchronize_net();
1202
1203
1204 list_for_each_entry(dev, head, close_list) {
1205 while (some_qdisc_is_busy(dev))
1206 yield();
1207
1208
1209
1210 netdev_for_each_tx_queue(dev, dev_qdisc_reset, NULL);
1211 if (dev_ingress_queue(dev))
1212 dev_qdisc_reset(dev, dev_ingress_queue(dev), NULL);
1213 }
1214}
1215
1216void dev_deactivate(struct net_device *dev)
1217{
1218 LIST_HEAD(single);
1219
1220 list_add(&dev->close_list, &single);
1221 dev_deactivate_many(&single);
1222 list_del(&single);
1223}
1224EXPORT_SYMBOL(dev_deactivate);
1225
1226static int qdisc_change_tx_queue_len(struct net_device *dev,
1227 struct netdev_queue *dev_queue)
1228{
1229 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
1230 const struct Qdisc_ops *ops = qdisc->ops;
1231
1232 if (ops->change_tx_queue_len)
1233 return ops->change_tx_queue_len(qdisc, dev->tx_queue_len);
1234 return 0;
1235}
1236
1237int dev_qdisc_change_tx_queue_len(struct net_device *dev)
1238{
1239 bool up = dev->flags & IFF_UP;
1240 unsigned int i;
1241 int ret = 0;
1242
1243 if (up)
1244 dev_deactivate(dev);
1245
1246 for (i = 0; i < dev->num_tx_queues; i++) {
1247 ret = qdisc_change_tx_queue_len(dev, &dev->_tx[i]);
1248
1249
1250 if (ret)
1251 break;
1252 }
1253
1254 if (up)
1255 dev_activate(dev);
1256 return ret;
1257}
1258
1259static void dev_init_scheduler_queue(struct net_device *dev,
1260 struct netdev_queue *dev_queue,
1261 void *_qdisc)
1262{
1263 struct Qdisc *qdisc = _qdisc;
1264
1265 rcu_assign_pointer(dev_queue->qdisc, qdisc);
1266 dev_queue->qdisc_sleeping = qdisc;
1267}
1268
1269void dev_init_scheduler(struct net_device *dev)
1270{
1271 dev->qdisc = &noop_qdisc;
1272 netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
1273 if (dev_ingress_queue(dev))
1274 dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
1275
1276 timer_setup(&dev->watchdog_timer, dev_watchdog, 0);
1277}
1278
1279static void shutdown_scheduler_queue(struct net_device *dev,
1280 struct netdev_queue *dev_queue,
1281 void *_qdisc_default)
1282{
1283 struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
1284 struct Qdisc *qdisc_default = _qdisc_default;
1285
1286 if (qdisc) {
1287 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
1288 dev_queue->qdisc_sleeping = qdisc_default;
1289
1290 qdisc_put(qdisc);
1291 }
1292}
1293
1294void dev_shutdown(struct net_device *dev)
1295{
1296 netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
1297 if (dev_ingress_queue(dev))
1298 shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
1299 qdisc_put(dev->qdisc);
1300 dev->qdisc = &noop_qdisc;
1301
1302 WARN_ON(timer_pending(&dev->watchdog_timer));
1303}
1304
1305void psched_ratecfg_precompute(struct psched_ratecfg *r,
1306 const struct tc_ratespec *conf,
1307 u64 rate64)
1308{
1309 memset(r, 0, sizeof(*r));
1310 r->overhead = conf->overhead;
1311 r->rate_bytes_ps = max_t(u64, conf->rate, rate64);
1312 r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK);
1313 r->mult = 1;
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327 if (r->rate_bytes_ps > 0) {
1328 u64 factor = NSEC_PER_SEC;
1329
1330 for (;;) {
1331 r->mult = div64_u64(factor, r->rate_bytes_ps);
1332 if (r->mult & (1U << 31) || factor & (1ULL << 63))
1333 break;
1334 factor <<= 1;
1335 r->shift++;
1336 }
1337 }
1338}
1339EXPORT_SYMBOL(psched_ratecfg_precompute);
1340
1341static void mini_qdisc_rcu_func(struct rcu_head *head)
1342{
1343}
1344
1345void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp,
1346 struct tcf_proto *tp_head)
1347{
1348
1349
1350
1351 struct mini_Qdisc *miniq_old =
1352 rcu_dereference_protected(*miniqp->p_miniq, 1);
1353 struct mini_Qdisc *miniq;
1354
1355 if (!tp_head) {
1356 RCU_INIT_POINTER(*miniqp->p_miniq, NULL);
1357
1358 rcu_barrier();
1359 return;
1360 }
1361
1362 miniq = !miniq_old || miniq_old == &miniqp->miniq2 ?
1363 &miniqp->miniq1 : &miniqp->miniq2;
1364
1365
1366
1367
1368
1369 rcu_barrier();
1370 miniq->filter_list = tp_head;
1371 rcu_assign_pointer(*miniqp->p_miniq, miniq);
1372
1373 if (miniq_old)
1374
1375
1376
1377
1378 call_rcu(&miniq_old->rcu, mini_qdisc_rcu_func);
1379}
1380EXPORT_SYMBOL(mini_qdisc_pair_swap);
1381
1382void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc,
1383 struct mini_Qdisc __rcu **p_miniq)
1384{
1385 miniqp->miniq1.cpu_bstats = qdisc->cpu_bstats;
1386 miniqp->miniq1.cpu_qstats = qdisc->cpu_qstats;
1387 miniqp->miniq2.cpu_bstats = qdisc->cpu_bstats;
1388 miniqp->miniq2.cpu_qstats = qdisc->cpu_qstats;
1389 miniqp->p_miniq = p_miniq;
1390}
1391EXPORT_SYMBOL(mini_qdisc_pair_init);
1392