~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/sched/sch_generic.c

Version: ~ [ linux-5.14-rc3 ] ~ [ linux-5.13.5 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.53 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.135 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.198 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.240 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.276 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.276 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * net/sched/sch_generic.c      Generic packet scheduler routines.
  3  *
  4  *              This program is free software; you can redistribute it and/or
  5  *              modify it under the terms of the GNU General Public License
  6  *              as published by the Free Software Foundation; either version
  7  *              2 of the License, or (at your option) any later version.
  8  *
  9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
 10  *              Jamal Hadi Salim, <hadi@cyberus.ca> 990601
 11  *              - Ingress support
 12  */
 13 
 14 #include <linux/bitops.h>
 15 #include <linux/module.h>
 16 #include <linux/types.h>
 17 #include <linux/kernel.h>
 18 #include <linux/sched.h>
 19 #include <linux/string.h>
 20 #include <linux/errno.h>
 21 #include <linux/netdevice.h>
 22 #include <linux/skbuff.h>
 23 #include <linux/rtnetlink.h>
 24 #include <linux/init.h>
 25 #include <linux/rcupdate.h>
 26 #include <linux/list.h>
 27 #include <linux/slab.h>
 28 #include <linux/if_vlan.h>
 29 #include <net/sch_generic.h>
 30 #include <net/pkt_sched.h>
 31 #include <net/dst.h>
 32 
 33 /* Qdisc to use by default */
 34 const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops;
 35 EXPORT_SYMBOL(default_qdisc_ops);
 36 
 37 /* Main transmission queue. */
 38 
 39 /* Modifications to data participating in scheduling must be protected with
 40  * qdisc_lock(qdisc) spinlock.
 41  *
 42  * The idea is the following:
 43  * - enqueue, dequeue are serialized via qdisc root lock
 44  * - ingress filtering is also serialized via qdisc root lock
 45  * - updates to tree and tree walking are only done under the rtnl mutex.
 46  */
 47 
 48 static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
 49 {
 50         q->gso_skb = skb;
 51         q->qstats.requeues++;
 52         q->q.qlen++;    /* it's still part of the queue */
 53         __netif_schedule(q);
 54 
 55         return 0;
 56 }
 57 
 58 static void try_bulk_dequeue_skb(struct Qdisc *q,
 59                                  struct sk_buff *skb,
 60                                  const struct netdev_queue *txq,
 61                                  int *packets)
 62 {
 63         int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
 64 
 65         while (bytelimit > 0) {
 66                 struct sk_buff *nskb = q->dequeue(q);
 67 
 68                 if (!nskb)
 69                         break;
 70 
 71                 bytelimit -= nskb->len; /* covers GSO len */
 72                 skb->next = nskb;
 73                 skb = nskb;
 74                 (*packets)++; /* GSO counts as one pkt */
 75         }
 76         skb->next = NULL;
 77 }
 78 
 79 /* Note that dequeue_skb can possibly return a SKB list (via skb->next).
 80  * A requeued skb (via q->gso_skb) can also be a SKB list.
 81  */
 82 static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
 83                                    int *packets)
 84 {
 85         struct sk_buff *skb = q->gso_skb;
 86         const struct netdev_queue *txq = q->dev_queue;
 87 
 88         *packets = 1;
 89         *validate = true;
 90         if (unlikely(skb)) {
 91                 /* check the reason of requeuing without tx lock first */
 92                 txq = skb_get_tx_queue(txq->dev, skb);
 93                 if (!netif_xmit_frozen_or_stopped(txq)) {
 94                         q->gso_skb = NULL;
 95                         q->q.qlen--;
 96                 } else
 97                         skb = NULL;
 98                 /* skb in gso_skb were already validated */
 99                 *validate = false;
100         } else {
101                 if (!(q->flags & TCQ_F_ONETXQUEUE) ||
102                     !netif_xmit_frozen_or_stopped(txq)) {
103                         skb = q->dequeue(q);
104                         if (skb && qdisc_may_bulk(q))
105                                 try_bulk_dequeue_skb(q, skb, txq, packets);
106                 }
107         }
108         return skb;
109 }
110 
111 static inline int handle_dev_cpu_collision(struct sk_buff *skb,
112                                            struct netdev_queue *dev_queue,
113                                            struct Qdisc *q)
114 {
115         int ret;
116 
117         if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) {
118                 /*
119                  * Same CPU holding the lock. It may be a transient
120                  * configuration error, when hard_start_xmit() recurses. We
121                  * detect it by checking xmit owner and drop the packet when
122                  * deadloop is detected. Return OK to try the next skb.
123                  */
124                 kfree_skb_list(skb);
125                 net_warn_ratelimited("Dead loop on netdevice %s, fix it urgently!\n",
126                                      dev_queue->dev->name);
127                 ret = qdisc_qlen(q);
128         } else {
129                 /*
130                  * Another cpu is holding lock, requeue & delay xmits for
131                  * some time.
132                  */
133                 __this_cpu_inc(softnet_data.cpu_collision);
134                 ret = dev_requeue_skb(skb, q);
135         }
136 
137         return ret;
138 }
139 
140 /*
141  * Transmit possibly several skbs, and handle the return status as
142  * required. Holding the __QDISC___STATE_RUNNING bit guarantees that
143  * only one CPU can execute this function.
144  *
145  * Returns to the caller:
146  *                              0  - queue is empty or throttled.
147  *                              >0 - queue is not empty.
148  */
149 int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
150                     struct net_device *dev, struct netdev_queue *txq,
151                     spinlock_t *root_lock, bool validate)
152 {
153         int ret = NETDEV_TX_BUSY;
154 
155         /* And release qdisc */
156         spin_unlock(root_lock);
157 
158         /* Note that we validate skb (GSO, checksum, ...) outside of locks */
159         if (validate)
160                 skb = validate_xmit_skb_list(skb, dev);
161 
162         if (likely(skb)) {
163                 HARD_TX_LOCK(dev, txq, smp_processor_id());
164                 if (!netif_xmit_frozen_or_stopped(txq))
165                         skb = dev_hard_start_xmit(skb, dev, txq, &ret);
166 
167                 HARD_TX_UNLOCK(dev, txq);
168         } else {
169                 spin_lock(root_lock);
170                 return qdisc_qlen(q);
171         }
172         spin_lock(root_lock);
173 
174         if (dev_xmit_complete(ret)) {
175                 /* Driver sent out skb successfully or skb was consumed */
176                 ret = qdisc_qlen(q);
177         } else if (ret == NETDEV_TX_LOCKED) {
178                 /* Driver try lock failed */
179                 ret = handle_dev_cpu_collision(skb, txq, q);
180         } else {
181                 /* Driver returned NETDEV_TX_BUSY - requeue skb */
182                 if (unlikely(ret != NETDEV_TX_BUSY))
183                         net_warn_ratelimited("BUG %s code %d qlen %d\n",
184                                              dev->name, ret, q->q.qlen);
185 
186                 ret = dev_requeue_skb(skb, q);
187         }
188 
189         if (ret && netif_xmit_frozen_or_stopped(txq))
190                 ret = 0;
191 
192         return ret;
193 }
194 
195 /*
196  * NOTE: Called under qdisc_lock(q) with locally disabled BH.
197  *
198  * __QDISC___STATE_RUNNING guarantees only one CPU can process
199  * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
200  * this queue.
201  *
202  *  netif_tx_lock serializes accesses to device driver.
203  *
204  *  qdisc_lock(q) and netif_tx_lock are mutually exclusive,
205  *  if one is grabbed, another must be free.
206  *
207  * Note, that this procedure can be called by a watchdog timer
208  *
209  * Returns to the caller:
210  *                              0  - queue is empty or throttled.
211  *                              >0 - queue is not empty.
212  *
213  */
214 static inline int qdisc_restart(struct Qdisc *q, int *packets)
215 {
216         struct netdev_queue *txq;
217         struct net_device *dev;
218         spinlock_t *root_lock;
219         struct sk_buff *skb;
220         bool validate;
221 
222         /* Dequeue packet */
223         skb = dequeue_skb(q, &validate, packets);
224         if (unlikely(!skb))
225                 return 0;
226 
227         root_lock = qdisc_lock(q);
228         dev = qdisc_dev(q);
229         txq = skb_get_tx_queue(dev, skb);
230 
231         return sch_direct_xmit(skb, q, dev, txq, root_lock, validate);
232 }
233 
234 void __qdisc_run(struct Qdisc *q)
235 {
236         int quota = weight_p;
237         int packets;
238 
239         while (qdisc_restart(q, &packets)) {
240                 /*
241                  * Ordered by possible occurrence: Postpone processing if
242                  * 1. we've exceeded packet quota
243                  * 2. another process needs the CPU;
244                  */
245                 quota -= packets;
246                 if (quota <= 0 || need_resched()) {
247                         __netif_schedule(q);
248                         break;
249                 }
250         }
251 
252         qdisc_run_end(q);
253 }
254 
255 unsigned long dev_trans_start(struct net_device *dev)
256 {
257         unsigned long val, res;
258         unsigned int i;
259 
260         if (is_vlan_dev(dev))
261                 dev = vlan_dev_real_dev(dev);
262         res = dev->trans_start;
263         for (i = 0; i < dev->num_tx_queues; i++) {
264                 val = netdev_get_tx_queue(dev, i)->trans_start;
265                 if (val && time_after(val, res))
266                         res = val;
267         }
268         dev->trans_start = res;
269 
270         return res;
271 }
272 EXPORT_SYMBOL(dev_trans_start);
273 
274 static void dev_watchdog(unsigned long arg)
275 {
276         struct net_device *dev = (struct net_device *)arg;
277 
278         netif_tx_lock(dev);
279         if (!qdisc_tx_is_noop(dev)) {
280                 if (netif_device_present(dev) &&
281                     netif_running(dev) &&
282                     netif_carrier_ok(dev)) {
283                         int some_queue_timedout = 0;
284                         unsigned int i;
285                         unsigned long trans_start;
286 
287                         for (i = 0; i < dev->num_tx_queues; i++) {
288                                 struct netdev_queue *txq;
289 
290                                 txq = netdev_get_tx_queue(dev, i);
291                                 /*
292                                  * old device drivers set dev->trans_start
293                                  */
294                                 trans_start = txq->trans_start ? : dev->trans_start;
295                                 if (netif_xmit_stopped(txq) &&
296                                     time_after(jiffies, (trans_start +
297                                                          dev->watchdog_timeo))) {
298                                         some_queue_timedout = 1;
299                                         txq->trans_timeout++;
300                                         break;
301                                 }
302                         }
303 
304                         if (some_queue_timedout) {
305                                 WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
306                                        dev->name, netdev_drivername(dev), i);
307                                 dev->netdev_ops->ndo_tx_timeout(dev);
308                         }
309                         if (!mod_timer(&dev->watchdog_timer,
310                                        round_jiffies(jiffies +
311                                                      dev->watchdog_timeo)))
312                                 dev_hold(dev);
313                 }
314         }
315         netif_tx_unlock(dev);
316 
317         dev_put(dev);
318 }
319 
320 void __netdev_watchdog_up(struct net_device *dev)
321 {
322         if (dev->netdev_ops->ndo_tx_timeout) {
323                 if (dev->watchdog_timeo <= 0)
324                         dev->watchdog_timeo = 5*HZ;
325                 if (!mod_timer(&dev->watchdog_timer,
326                                round_jiffies(jiffies + dev->watchdog_timeo)))
327                         dev_hold(dev);
328         }
329 }
330 
331 static void dev_watchdog_up(struct net_device *dev)
332 {
333         __netdev_watchdog_up(dev);
334 }
335 
336 static void dev_watchdog_down(struct net_device *dev)
337 {
338         netif_tx_lock_bh(dev);
339         if (del_timer(&dev->watchdog_timer))
340                 dev_put(dev);
341         netif_tx_unlock_bh(dev);
342 }
343 
344 /**
345  *      netif_carrier_on - set carrier
346  *      @dev: network device
347  *
348  * Device has detected that carrier.
349  */
350 void netif_carrier_on(struct net_device *dev)
351 {
352         if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
353                 if (dev->reg_state == NETREG_UNINITIALIZED)
354                         return;
355                 atomic_inc(&dev->carrier_changes);
356                 linkwatch_fire_event(dev);
357                 if (netif_running(dev))
358                         __netdev_watchdog_up(dev);
359         }
360 }
361 EXPORT_SYMBOL(netif_carrier_on);
362 
363 /**
364  *      netif_carrier_off - clear carrier
365  *      @dev: network device
366  *
367  * Device has detected loss of carrier.
368  */
369 void netif_carrier_off(struct net_device *dev)
370 {
371         if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
372                 if (dev->reg_state == NETREG_UNINITIALIZED)
373                         return;
374                 atomic_inc(&dev->carrier_changes);
375                 linkwatch_fire_event(dev);
376         }
377 }
378 EXPORT_SYMBOL(netif_carrier_off);
379 
380 /* "NOOP" scheduler: the best scheduler, recommended for all interfaces
381    under all circumstances. It is difficult to invent anything faster or
382    cheaper.
383  */
384 
385 static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc)
386 {
387         kfree_skb(skb);
388         return NET_XMIT_CN;
389 }
390 
391 static struct sk_buff *noop_dequeue(struct Qdisc *qdisc)
392 {
393         return NULL;
394 }
395 
396 struct Qdisc_ops noop_qdisc_ops __read_mostly = {
397         .id             =       "noop",
398         .priv_size      =       0,
399         .enqueue        =       noop_enqueue,
400         .dequeue        =       noop_dequeue,
401         .peek           =       noop_dequeue,
402         .owner          =       THIS_MODULE,
403 };
404 
405 static struct netdev_queue noop_netdev_queue = {
406         .qdisc          =       &noop_qdisc,
407         .qdisc_sleeping =       &noop_qdisc,
408 };
409 
410 struct Qdisc noop_qdisc = {
411         .enqueue        =       noop_enqueue,
412         .dequeue        =       noop_dequeue,
413         .flags          =       TCQ_F_BUILTIN,
414         .ops            =       &noop_qdisc_ops,
415         .list           =       LIST_HEAD_INIT(noop_qdisc.list),
416         .q.lock         =       __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
417         .dev_queue      =       &noop_netdev_queue,
418         .busylock       =       __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock),
419 };
420 EXPORT_SYMBOL(noop_qdisc);
421 
422 static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
423         .id             =       "noqueue",
424         .priv_size      =       0,
425         .enqueue        =       noop_enqueue,
426         .dequeue        =       noop_dequeue,
427         .peek           =       noop_dequeue,
428         .owner          =       THIS_MODULE,
429 };
430 
431 static struct Qdisc noqueue_qdisc;
432 static struct netdev_queue noqueue_netdev_queue = {
433         .qdisc          =       &noqueue_qdisc,
434         .qdisc_sleeping =       &noqueue_qdisc,
435 };
436 
437 static struct Qdisc noqueue_qdisc = {
438         .enqueue        =       NULL,
439         .dequeue        =       noop_dequeue,
440         .flags          =       TCQ_F_BUILTIN,
441         .ops            =       &noqueue_qdisc_ops,
442         .list           =       LIST_HEAD_INIT(noqueue_qdisc.list),
443         .q.lock         =       __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock),
444         .dev_queue      =       &noqueue_netdev_queue,
445         .busylock       =       __SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock),
446 };
447 
448 
449 static const u8 prio2band[TC_PRIO_MAX + 1] = {
450         1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1
451 };
452 
453 /* 3-band FIFO queue: old style, but should be a bit faster than
454    generic prio+fifo combination.
455  */
456 
457 #define PFIFO_FAST_BANDS 3
458 
459 /*
460  * Private data for a pfifo_fast scheduler containing:
461  *      - queues for the three band
462  *      - bitmap indicating which of the bands contain skbs
463  */
464 struct pfifo_fast_priv {
465         u32 bitmap;
466         struct sk_buff_head q[PFIFO_FAST_BANDS];
467 };
468 
469 /*
470  * Convert a bitmap to the first band number where an skb is queued, where:
471  *      bitmap=0 means there are no skbs on any band.
472  *      bitmap=1 means there is an skb on band 0.
473  *      bitmap=7 means there are skbs on all 3 bands, etc.
474  */
475 static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0};
476 
477 static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv,
478                                              int band)
479 {
480         return priv->q + band;
481 }
482 
483 static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc)
484 {
485         if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) {
486                 int band = prio2band[skb->priority & TC_PRIO_MAX];
487                 struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
488                 struct sk_buff_head *list = band2list(priv, band);
489 
490                 priv->bitmap |= (1 << band);
491                 qdisc->q.qlen++;
492                 return __qdisc_enqueue_tail(skb, qdisc, list);
493         }
494 
495         return qdisc_drop(skb, qdisc);
496 }
497 
498 static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc)
499 {
500         struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
501         int band = bitmap2band[priv->bitmap];
502 
503         if (likely(band >= 0)) {
504                 struct sk_buff_head *list = band2list(priv, band);
505                 struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list);
506 
507                 qdisc->q.qlen--;
508                 if (skb_queue_empty(list))
509                         priv->bitmap &= ~(1 << band);
510 
511                 return skb;
512         }
513 
514         return NULL;
515 }
516 
517 static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc)
518 {
519         struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
520         int band = bitmap2band[priv->bitmap];
521 
522         if (band >= 0) {
523                 struct sk_buff_head *list = band2list(priv, band);
524 
525                 return skb_peek(list);
526         }
527 
528         return NULL;
529 }
530 
531 static void pfifo_fast_reset(struct Qdisc *qdisc)
532 {
533         int prio;
534         struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
535 
536         for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
537                 __qdisc_reset_queue(qdisc, band2list(priv, prio));
538 
539         priv->bitmap = 0;
540         qdisc->qstats.backlog = 0;
541         qdisc->q.qlen = 0;
542 }
543 
544 static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
545 {
546         struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
547 
548         memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1);
549         if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
550                 goto nla_put_failure;
551         return skb->len;
552 
553 nla_put_failure:
554         return -1;
555 }
556 
557 static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
558 {
559         int prio;
560         struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
561 
562         for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
563                 __skb_queue_head_init(band2list(priv, prio));
564 
565         /* Can by-pass the queue discipline */
566         qdisc->flags |= TCQ_F_CAN_BYPASS;
567         return 0;
568 }
569 
570 struct Qdisc_ops pfifo_fast_ops __read_mostly = {
571         .id             =       "pfifo_fast",
572         .priv_size      =       sizeof(struct pfifo_fast_priv),
573         .enqueue        =       pfifo_fast_enqueue,
574         .dequeue        =       pfifo_fast_dequeue,
575         .peek           =       pfifo_fast_peek,
576         .init           =       pfifo_fast_init,
577         .reset          =       pfifo_fast_reset,
578         .dump           =       pfifo_fast_dump,
579         .owner          =       THIS_MODULE,
580 };
581 
582 static struct lock_class_key qdisc_tx_busylock;
583 
584 struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
585                           const struct Qdisc_ops *ops)
586 {
587         void *p;
588         struct Qdisc *sch;
589         unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size;
590         int err = -ENOBUFS;
591         struct net_device *dev = dev_queue->dev;
592 
593         p = kzalloc_node(size, GFP_KERNEL,
594                          netdev_queue_numa_node_read(dev_queue));
595 
596         if (!p)
597                 goto errout;
598         sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
599         /* if we got non aligned memory, ask more and do alignment ourself */
600         if (sch != p) {
601                 kfree(p);
602                 p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL,
603                                  netdev_queue_numa_node_read(dev_queue));
604                 if (!p)
605                         goto errout;
606                 sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
607                 sch->padded = (char *) sch - (char *) p;
608         }
609         INIT_LIST_HEAD(&sch->list);
610         skb_queue_head_init(&sch->q);
611 
612         spin_lock_init(&sch->busylock);
613         lockdep_set_class(&sch->busylock,
614                           dev->qdisc_tx_busylock ?: &qdisc_tx_busylock);
615 
616         sch->ops = ops;
617         sch->enqueue = ops->enqueue;
618         sch->dequeue = ops->dequeue;
619         sch->dev_queue = dev_queue;
620         dev_hold(dev);
621         atomic_set(&sch->refcnt, 1);
622 
623         return sch;
624 errout:
625         return ERR_PTR(err);
626 }
627 
628 struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue,
629                                 const struct Qdisc_ops *ops,
630                                 unsigned int parentid)
631 {
632         struct Qdisc *sch;
633 
634         if (!try_module_get(ops->owner))
635                 goto errout;
636 
637         sch = qdisc_alloc(dev_queue, ops);
638         if (IS_ERR(sch))
639                 goto errout;
640         sch->parent = parentid;
641 
642         if (!ops->init || ops->init(sch, NULL) == 0)
643                 return sch;
644 
645         qdisc_destroy(sch);
646 errout:
647         return NULL;
648 }
649 EXPORT_SYMBOL(qdisc_create_dflt);
650 
651 /* Under qdisc_lock(qdisc) and BH! */
652 
653 void qdisc_reset(struct Qdisc *qdisc)
654 {
655         const struct Qdisc_ops *ops = qdisc->ops;
656 
657         if (ops->reset)
658                 ops->reset(qdisc);
659 
660         if (qdisc->gso_skb) {
661                 kfree_skb_list(qdisc->gso_skb);
662                 qdisc->gso_skb = NULL;
663                 qdisc->q.qlen = 0;
664         }
665 }
666 EXPORT_SYMBOL(qdisc_reset);
667 
668 static void qdisc_rcu_free(struct rcu_head *head)
669 {
670         struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head);
671 
672         if (qdisc_is_percpu_stats(qdisc)) {
673                 free_percpu(qdisc->cpu_bstats);
674                 free_percpu(qdisc->cpu_qstats);
675         }
676 
677         kfree((char *) qdisc - qdisc->padded);
678 }
679 
680 void qdisc_destroy(struct Qdisc *qdisc)
681 {
682         const struct Qdisc_ops  *ops = qdisc->ops;
683 
684         if (qdisc->flags & TCQ_F_BUILTIN ||
685             !atomic_dec_and_test(&qdisc->refcnt))
686                 return;
687 
688 #ifdef CONFIG_NET_SCHED
689         qdisc_list_del(qdisc);
690 
691         qdisc_put_stab(rtnl_dereference(qdisc->stab));
692 #endif
693         gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
694         if (ops->reset)
695                 ops->reset(qdisc);
696         if (ops->destroy)
697                 ops->destroy(qdisc);
698 
699         module_put(ops->owner);
700         dev_put(qdisc_dev(qdisc));
701 
702         kfree_skb_list(qdisc->gso_skb);
703         /*
704          * gen_estimator est_timer() might access qdisc->q.lock,
705          * wait a RCU grace period before freeing qdisc.
706          */
707         call_rcu(&qdisc->rcu_head, qdisc_rcu_free);
708 }
709 EXPORT_SYMBOL(qdisc_destroy);
710 
711 /* Attach toplevel qdisc to device queue. */
712 struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
713                               struct Qdisc *qdisc)
714 {
715         struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
716         spinlock_t *root_lock;
717 
718         root_lock = qdisc_lock(oqdisc);
719         spin_lock_bh(root_lock);
720 
721         /* Prune old scheduler */
722         if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
723                 qdisc_reset(oqdisc);
724 
725         /* ... and graft new one */
726         if (qdisc == NULL)
727                 qdisc = &noop_qdisc;
728         dev_queue->qdisc_sleeping = qdisc;
729         rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
730 
731         spin_unlock_bh(root_lock);
732 
733         return oqdisc;
734 }
735 EXPORT_SYMBOL(dev_graft_qdisc);
736 
737 static void attach_one_default_qdisc(struct net_device *dev,
738                                      struct netdev_queue *dev_queue,
739                                      void *_unused)
740 {
741         struct Qdisc *qdisc = &noqueue_qdisc;
742 
743         if (dev->tx_queue_len) {
744                 qdisc = qdisc_create_dflt(dev_queue,
745                                           default_qdisc_ops, TC_H_ROOT);
746                 if (!qdisc) {
747                         netdev_info(dev, "activation failed\n");
748                         return;
749                 }
750                 if (!netif_is_multiqueue(dev))
751                         qdisc->flags |= TCQ_F_ONETXQUEUE;
752         }
753         dev_queue->qdisc_sleeping = qdisc;
754 }
755 
756 static void attach_default_qdiscs(struct net_device *dev)
757 {
758         struct netdev_queue *txq;
759         struct Qdisc *qdisc;
760 
761         txq = netdev_get_tx_queue(dev, 0);
762 
763         if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) {
764                 netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
765                 dev->qdisc = txq->qdisc_sleeping;
766                 atomic_inc(&dev->qdisc->refcnt);
767         } else {
768                 qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT);
769                 if (qdisc) {
770                         dev->qdisc = qdisc;
771                         qdisc->ops->attach(qdisc);
772                 }
773         }
774 }
775 
776 static void transition_one_qdisc(struct net_device *dev,
777                                  struct netdev_queue *dev_queue,
778                                  void *_need_watchdog)
779 {
780         struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
781         int *need_watchdog_p = _need_watchdog;
782 
783         if (!(new_qdisc->flags & TCQ_F_BUILTIN))
784                 clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
785 
786         rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
787         if (need_watchdog_p && new_qdisc != &noqueue_qdisc) {
788                 dev_queue->trans_start = 0;
789                 *need_watchdog_p = 1;
790         }
791 }
792 
793 void dev_activate(struct net_device *dev)
794 {
795         int need_watchdog;
796 
797         /* No queueing discipline is attached to device;
798          * create default one for devices, which need queueing
799          * and noqueue_qdisc for virtual interfaces
800          */
801 
802         if (dev->qdisc == &noop_qdisc)
803                 attach_default_qdiscs(dev);
804 
805         if (!netif_carrier_ok(dev))
806                 /* Delay activation until next carrier-on event */
807                 return;
808 
809         need_watchdog = 0;
810         netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
811         if (dev_ingress_queue(dev))
812                 transition_one_qdisc(dev, dev_ingress_queue(dev), NULL);
813 
814         if (need_watchdog) {
815                 dev->trans_start = jiffies;
816                 dev_watchdog_up(dev);
817         }
818 }
819 EXPORT_SYMBOL(dev_activate);
820 
821 static void dev_deactivate_queue(struct net_device *dev,
822                                  struct netdev_queue *dev_queue,
823                                  void *_qdisc_default)
824 {
825         struct Qdisc *qdisc_default = _qdisc_default;
826         struct Qdisc *qdisc;
827 
828         qdisc = rtnl_dereference(dev_queue->qdisc);
829         if (qdisc) {
830                 spin_lock_bh(qdisc_lock(qdisc));
831 
832                 if (!(qdisc->flags & TCQ_F_BUILTIN))
833                         set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
834 
835                 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
836                 qdisc_reset(qdisc);
837 
838                 spin_unlock_bh(qdisc_lock(qdisc));
839         }
840 }
841 
842 static bool some_qdisc_is_busy(struct net_device *dev)
843 {
844         unsigned int i;
845 
846         for (i = 0; i < dev->num_tx_queues; i++) {
847                 struct netdev_queue *dev_queue;
848                 spinlock_t *root_lock;
849                 struct Qdisc *q;
850                 int val;
851 
852                 dev_queue = netdev_get_tx_queue(dev, i);
853                 q = dev_queue->qdisc_sleeping;
854                 root_lock = qdisc_lock(q);
855 
856                 spin_lock_bh(root_lock);
857 
858                 val = (qdisc_is_running(q) ||
859                        test_bit(__QDISC_STATE_SCHED, &q->state));
860 
861                 spin_unlock_bh(root_lock);
862 
863                 if (val)
864                         return true;
865         }
866         return false;
867 }
868 
869 /**
870  *      dev_deactivate_many - deactivate transmissions on several devices
871  *      @head: list of devices to deactivate
872  *
873  *      This function returns only when all outstanding transmissions
874  *      have completed, unless all devices are in dismantle phase.
875  */
876 void dev_deactivate_many(struct list_head *head)
877 {
878         struct net_device *dev;
879         bool sync_needed = false;
880 
881         list_for_each_entry(dev, head, close_list) {
882                 netdev_for_each_tx_queue(dev, dev_deactivate_queue,
883                                          &noop_qdisc);
884                 if (dev_ingress_queue(dev))
885                         dev_deactivate_queue(dev, dev_ingress_queue(dev),
886                                              &noop_qdisc);
887 
888                 dev_watchdog_down(dev);
889                 sync_needed |= !dev->dismantle;
890         }
891 
892         /* Wait for outstanding qdisc-less dev_queue_xmit calls.
893          * This is avoided if all devices are in dismantle phase :
894          * Caller will call synchronize_net() for us
895          */
896         if (sync_needed)
897                 synchronize_net();
898 
899         /* Wait for outstanding qdisc_run calls. */
900         list_for_each_entry(dev, head, close_list)
901                 while (some_qdisc_is_busy(dev))
902                         yield();
903 }
904 
905 void dev_deactivate(struct net_device *dev)
906 {
907         LIST_HEAD(single);
908 
909         list_add(&dev->close_list, &single);
910         dev_deactivate_many(&single);
911         list_del(&single);
912 }
913 EXPORT_SYMBOL(dev_deactivate);
914 
915 static void dev_init_scheduler_queue(struct net_device *dev,
916                                      struct netdev_queue *dev_queue,
917                                      void *_qdisc)
918 {
919         struct Qdisc *qdisc = _qdisc;
920 
921         rcu_assign_pointer(dev_queue->qdisc, qdisc);
922         dev_queue->qdisc_sleeping = qdisc;
923 }
924 
925 void dev_init_scheduler(struct net_device *dev)
926 {
927         dev->qdisc = &noop_qdisc;
928         netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
929         if (dev_ingress_queue(dev))
930                 dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
931 
932         setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
933 }
934 
935 static void shutdown_scheduler_queue(struct net_device *dev,
936                                      struct netdev_queue *dev_queue,
937                                      void *_qdisc_default)
938 {
939         struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
940         struct Qdisc *qdisc_default = _qdisc_default;
941 
942         if (qdisc) {
943                 rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
944                 dev_queue->qdisc_sleeping = qdisc_default;
945 
946                 qdisc_destroy(qdisc);
947         }
948 }
949 
950 void dev_shutdown(struct net_device *dev)
951 {
952         netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
953         if (dev_ingress_queue(dev))
954                 shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc);
955         qdisc_destroy(dev->qdisc);
956         dev->qdisc = &noop_qdisc;
957 
958         WARN_ON(timer_pending(&dev->watchdog_timer));
959 }
960 
961 void psched_ratecfg_precompute(struct psched_ratecfg *r,
962                                const struct tc_ratespec *conf,
963                                u64 rate64)
964 {
965         memset(r, 0, sizeof(*r));
966         r->overhead = conf->overhead;
967         r->rate_bytes_ps = max_t(u64, conf->rate, rate64);
968         r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK);
969         r->mult = 1;
970         /*
971          * The deal here is to replace a divide by a reciprocal one
972          * in fast path (a reciprocal divide is a multiply and a shift)
973          *
974          * Normal formula would be :
975          *  time_in_ns = (NSEC_PER_SEC * len) / rate_bps
976          *
977          * We compute mult/shift to use instead :
978          *  time_in_ns = (len * mult) >> shift;
979          *
980          * We try to get the highest possible mult value for accuracy,
981          * but have to make sure no overflows will ever happen.
982          */
983         if (r->rate_bytes_ps > 0) {
984                 u64 factor = NSEC_PER_SEC;
985 
986                 for (;;) {
987                         r->mult = div64_u64(factor, r->rate_bytes_ps);
988                         if (r->mult & (1U << 31) || factor & (1ULL << 63))
989                                 break;
990                         factor <<= 1;
991                         r->shift++;
992                 }
993         }
994 }
995 EXPORT_SYMBOL(psched_ratecfg_precompute);
996 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp