sch_generic.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848
  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. #include <linux/bitops.h>
  14. #include <linux/module.h>
  15. #include <linux/types.h>
  16. #include <linux/kernel.h>
  17. #include <linux/sched.h>
  18. #include <linux/string.h>
  19. #include <linux/errno.h>
  20. #include <linux/netdevice.h>
  21. #include <linux/skbuff.h>
  22. #include <linux/rtnetlink.h>
  23. #include <linux/init.h>
  24. #include <linux/rcupdate.h>
  25. #include <linux/list.h>
  26. #include <linux/slab.h>
  27. #include <net/pkt_sched.h>
  28. /* Main transmission queue. */
  29. /* Modifications to data participating in scheduling must be protected with
  30. * qdisc_lock(qdisc) spinlock.
  31. *
  32. * The idea is the following:
  33. * - enqueue, dequeue are serialized via qdisc root lock
  34. * - ingress filtering is also serialized via qdisc root lock
  35. * - updates to tree and tree walking are only done under the rtnl mutex.
  36. */
  37. static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
  38. {
  39. q->gso_skb = skb;
  40. q->qstats.requeues++;
  41. q->q.qlen++; /* it's still part of the queue */
  42. __netif_schedule(q);
  43. return 0;
  44. }
  45. static inline struct sk_buff *dequeue_skb(struct Qdisc *q)
  46. {
  47. struct sk_buff *skb = q->gso_skb;
  48. if (unlikely(skb)) {
  49. struct net_device *dev = qdisc_dev(q);
  50. struct netdev_queue *txq;
  51. /* check the reason of requeuing without tx lock first */
  52. txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
  53. if (!netif_tx_queue_stopped(txq) &&
  54. !netif_tx_queue_frozen(txq)) {
  55. q->gso_skb = NULL;
  56. q->q.qlen--;
  57. } else
  58. skb = NULL;
  59. } else {
  60. skb = q->dequeue(q);
  61. }
  62. return skb;
  63. }
  64. static inline int handle_dev_cpu_collision(struct sk_buff *skb,
  65. struct netdev_queue *dev_queue,
  66. struct Qdisc *q)
  67. {
  68. int ret;
  69. if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) {
  70. /*
  71. * Same CPU holding the lock. It may be a transient
  72. * configuration error, when hard_start_xmit() recurses. We
  73. * detect it by checking xmit owner and drop the packet when
  74. * deadloop is detected. Return OK to try the next skb.
  75. */
  76. kfree_skb(skb);
  77. if (net_ratelimit())
  78. printk(KERN_WARNING "Dead loop on netdevice %s, "
  79. "fix it urgently!\n", dev_queue->dev->name);
  80. ret = qdisc_qlen(q);
  81. } else {
  82. /*
  83. * Another cpu is holding lock, requeue & delay xmits for
  84. * some time.
  85. */
  86. __get_cpu_var(softnet_data).cpu_collision++;
  87. ret = dev_requeue_skb(skb, q);
  88. }
  89. return ret;
  90. }
  91. /*
  92. * Transmit one skb, and handle the return status as required. Holding the
  93. * __QDISC_STATE_RUNNING bit guarantees that only one CPU can execute this
  94. * function.
  95. *
  96. * Returns to the caller:
  97. * 0 - queue is empty or throttled.
  98. * >0 - queue is not empty.
  99. */
  100. int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
  101. struct net_device *dev, struct netdev_queue *txq,
  102. spinlock_t *root_lock)
  103. {
  104. int ret = NETDEV_TX_BUSY;
  105. /* And release qdisc */
  106. spin_unlock(root_lock);
  107. HARD_TX_LOCK(dev, txq, smp_processor_id());
  108. if (!netif_tx_queue_stopped(txq) && !netif_tx_queue_frozen(txq))
  109. ret = dev_hard_start_xmit(skb, dev, txq);
  110. HARD_TX_UNLOCK(dev, txq);
  111. spin_lock(root_lock);
  112. if (dev_xmit_complete(ret)) {
  113. /* Driver sent out skb successfully or skb was consumed */
  114. ret = qdisc_qlen(q);
  115. } else if (ret == NETDEV_TX_LOCKED) {
  116. /* Driver try lock failed */
  117. ret = handle_dev_cpu_collision(skb, txq, q);
  118. } else {
  119. /* Driver returned NETDEV_TX_BUSY - requeue skb */
  120. if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit()))
  121. printk(KERN_WARNING "BUG %s code %d qlen %d\n",
  122. dev->name, ret, q->q.qlen);
  123. ret = dev_requeue_skb(skb, q);
  124. }
  125. if (ret && (netif_tx_queue_stopped(txq) ||
  126. netif_tx_queue_frozen(txq)))
  127. ret = 0;
  128. return ret;
  129. }
  130. /*
  131. * NOTE: Called under qdisc_lock(q) with locally disabled BH.
  132. *
  133. * __QDISC_STATE_RUNNING guarantees only one CPU can process
  134. * this qdisc at a time. qdisc_lock(q) serializes queue accesses for
  135. * this queue.
  136. *
  137. * netif_tx_lock serializes accesses to device driver.
  138. *
  139. * qdisc_lock(q) and netif_tx_lock are mutually exclusive,
  140. * if one is grabbed, another must be free.
  141. *
  142. * Note, that this procedure can be called by a watchdog timer
  143. *
  144. * Returns to the caller:
  145. * 0 - queue is empty or throttled.
  146. * >0 - queue is not empty.
  147. *
  148. */
  149. static inline int qdisc_restart(struct Qdisc *q)
  150. {
  151. struct netdev_queue *txq;
  152. struct net_device *dev;
  153. spinlock_t *root_lock;
  154. struct sk_buff *skb;
  155. /* Dequeue packet */
  156. skb = dequeue_skb(q);
  157. if (unlikely(!skb))
  158. return 0;
  159. root_lock = qdisc_lock(q);
  160. dev = qdisc_dev(q);
  161. txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
  162. return sch_direct_xmit(skb, q, dev, txq, root_lock);
  163. }
  164. void __qdisc_run(struct Qdisc *q)
  165. {
  166. unsigned long start_time = jiffies;
  167. while (qdisc_restart(q)) {
  168. /*
  169. * Postpone processing if
  170. * 1. another process needs the CPU;
  171. * 2. we've been doing it for too long.
  172. */
  173. if (need_resched() || jiffies != start_time) {
  174. __netif_schedule(q);
  175. break;
  176. }
  177. }
  178. clear_bit(__QDISC_STATE_RUNNING, &q->state);
  179. }
  180. unsigned long dev_trans_start(struct net_device *dev)
  181. {
  182. unsigned long val, res = dev->trans_start;
  183. unsigned int i;
  184. for (i = 0; i < dev->num_tx_queues; i++) {
  185. val = netdev_get_tx_queue(dev, i)->trans_start;
  186. if (val && time_after(val, res))
  187. res = val;
  188. }
  189. dev->trans_start = res;
  190. return res;
  191. }
  192. EXPORT_SYMBOL(dev_trans_start);
  193. static void dev_watchdog(unsigned long arg)
  194. {
  195. struct net_device *dev = (struct net_device *)arg;
  196. netif_tx_lock(dev);
  197. if (!qdisc_tx_is_noop(dev)) {
  198. if (netif_device_present(dev) &&
  199. netif_running(dev) &&
  200. netif_carrier_ok(dev)) {
  201. int some_queue_timedout = 0;
  202. unsigned int i;
  203. unsigned long trans_start;
  204. for (i = 0; i < dev->num_tx_queues; i++) {
  205. struct netdev_queue *txq;
  206. txq = netdev_get_tx_queue(dev, i);
  207. /*
  208. * old device drivers set dev->trans_start
  209. */
  210. trans_start = txq->trans_start ? : dev->trans_start;
  211. if (netif_tx_queue_stopped(txq) &&
  212. time_after(jiffies, (trans_start +
  213. dev->watchdog_timeo))) {
  214. some_queue_timedout = 1;
  215. break;
  216. }
  217. }
  218. if (some_queue_timedout) {
  219. char drivername[64];
  220. WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n",
  221. dev->name, netdev_drivername(dev, drivername, 64), i);
  222. dev->netdev_ops->ndo_tx_timeout(dev);
  223. }
  224. if (!mod_timer(&dev->watchdog_timer,
  225. round_jiffies(jiffies +
  226. dev->watchdog_timeo)))
  227. dev_hold(dev);
  228. }
  229. }
  230. netif_tx_unlock(dev);
  231. dev_put(dev);
  232. }
  233. void __netdev_watchdog_up(struct net_device *dev)
  234. {
  235. if (dev->netdev_ops->ndo_tx_timeout) {
  236. if (dev->watchdog_timeo <= 0)
  237. dev->watchdog_timeo = 5*HZ;
  238. if (!mod_timer(&dev->watchdog_timer,
  239. round_jiffies(jiffies + dev->watchdog_timeo)))
  240. dev_hold(dev);
  241. }
  242. }
  243. static void dev_watchdog_up(struct net_device *dev)
  244. {
  245. __netdev_watchdog_up(dev);
  246. }
  247. static void dev_watchdog_down(struct net_device *dev)
  248. {
  249. netif_tx_lock_bh(dev);
  250. if (del_timer(&dev->watchdog_timer))
  251. dev_put(dev);
  252. netif_tx_unlock_bh(dev);
  253. }
  254. /**
  255. * netif_carrier_on - set carrier
  256. * @dev: network device
  257. *
  258. * Device has detected that carrier.
  259. */
  260. void netif_carrier_on(struct net_device *dev)
  261. {
  262. if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
  263. if (dev->reg_state == NETREG_UNINITIALIZED)
  264. return;
  265. linkwatch_fire_event(dev);
  266. if (netif_running(dev))
  267. __netdev_watchdog_up(dev);
  268. }
  269. }
  270. EXPORT_SYMBOL(netif_carrier_on);
  271. /**
  272. * netif_carrier_off - clear carrier
  273. * @dev: network device
  274. *
  275. * Device has detected loss of carrier.
  276. */
  277. void netif_carrier_off(struct net_device *dev)
  278. {
  279. if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) {
  280. if (dev->reg_state == NETREG_UNINITIALIZED)
  281. return;
  282. linkwatch_fire_event(dev);
  283. }
  284. }
  285. EXPORT_SYMBOL(netif_carrier_off);
  286. /* "NOOP" scheduler: the best scheduler, recommended for all interfaces
  287. under all circumstances. It is difficult to invent anything faster or
  288. cheaper.
  289. */
  290. static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc)
  291. {
  292. kfree_skb(skb);
  293. return NET_XMIT_CN;
  294. }
  295. static struct sk_buff *noop_dequeue(struct Qdisc * qdisc)
  296. {
  297. return NULL;
  298. }
  299. struct Qdisc_ops noop_qdisc_ops __read_mostly = {
  300. .id = "noop",
  301. .priv_size = 0,
  302. .enqueue = noop_enqueue,
  303. .dequeue = noop_dequeue,
  304. .peek = noop_dequeue,
  305. .owner = THIS_MODULE,
  306. };
  307. static struct netdev_queue noop_netdev_queue = {
  308. .qdisc = &noop_qdisc,
  309. .qdisc_sleeping = &noop_qdisc,
  310. };
  311. struct Qdisc noop_qdisc = {
  312. .enqueue = noop_enqueue,
  313. .dequeue = noop_dequeue,
  314. .flags = TCQ_F_BUILTIN,
  315. .ops = &noop_qdisc_ops,
  316. .list = LIST_HEAD_INIT(noop_qdisc.list),
  317. .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock),
  318. .dev_queue = &noop_netdev_queue,
  319. };
  320. EXPORT_SYMBOL(noop_qdisc);
  321. static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = {
  322. .id = "noqueue",
  323. .priv_size = 0,
  324. .enqueue = noop_enqueue,
  325. .dequeue = noop_dequeue,
  326. .peek = noop_dequeue,
  327. .owner = THIS_MODULE,
  328. };
  329. static struct Qdisc noqueue_qdisc;
  330. static struct netdev_queue noqueue_netdev_queue = {
  331. .qdisc = &noqueue_qdisc,
  332. .qdisc_sleeping = &noqueue_qdisc,
  333. };
  334. static struct Qdisc noqueue_qdisc = {
  335. .enqueue = NULL,
  336. .dequeue = noop_dequeue,
  337. .flags = TCQ_F_BUILTIN,
  338. .ops = &noqueue_qdisc_ops,
  339. .list = LIST_HEAD_INIT(noqueue_qdisc.list),
  340. .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock),
  341. .dev_queue = &noqueue_netdev_queue,
  342. };
  343. static const u8 prio2band[TC_PRIO_MAX+1] =
  344. { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 };
  345. /* 3-band FIFO queue: old style, but should be a bit faster than
  346. generic prio+fifo combination.
  347. */
  348. #define PFIFO_FAST_BANDS 3
  349. /*
  350. * Private data for a pfifo_fast scheduler containing:
  351. * - queues for the three band
  352. * - bitmap indicating which of the bands contain skbs
  353. */
  354. struct pfifo_fast_priv {
  355. u32 bitmap;
  356. struct sk_buff_head q[PFIFO_FAST_BANDS];
  357. };
  358. /*
  359. * Convert a bitmap to the first band number where an skb is queued, where:
  360. * bitmap=0 means there are no skbs on any band.
  361. * bitmap=1 means there is an skb on band 0.
  362. * bitmap=7 means there are skbs on all 3 bands, etc.
  363. */
  364. static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0};
  365. static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv,
  366. int band)
  367. {
  368. return priv->q + band;
  369. }
  370. static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc)
  371. {
  372. if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) {
  373. int band = prio2band[skb->priority & TC_PRIO_MAX];
  374. struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
  375. struct sk_buff_head *list = band2list(priv, band);
  376. priv->bitmap |= (1 << band);
  377. qdisc->q.qlen++;
  378. return __qdisc_enqueue_tail(skb, qdisc, list);
  379. }
  380. return qdisc_drop(skb, qdisc);
  381. }
  382. static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc)
  383. {
  384. struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
  385. int band = bitmap2band[priv->bitmap];
  386. if (likely(band >= 0)) {
  387. struct sk_buff_head *list = band2list(priv, band);
  388. struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list);
  389. qdisc->q.qlen--;
  390. if (skb_queue_empty(list))
  391. priv->bitmap &= ~(1 << band);
  392. return skb;
  393. }
  394. return NULL;
  395. }
  396. static struct sk_buff *pfifo_fast_peek(struct Qdisc* qdisc)
  397. {
  398. struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
  399. int band = bitmap2band[priv->bitmap];
  400. if (band >= 0) {
  401. struct sk_buff_head *list = band2list(priv, band);
  402. return skb_peek(list);
  403. }
  404. return NULL;
  405. }
  406. static void pfifo_fast_reset(struct Qdisc* qdisc)
  407. {
  408. int prio;
  409. struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
  410. for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
  411. __qdisc_reset_queue(qdisc, band2list(priv, prio));
  412. priv->bitmap = 0;
  413. qdisc->qstats.backlog = 0;
  414. qdisc->q.qlen = 0;
  415. }
  416. static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb)
  417. {
  418. struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS };
  419. memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1);
  420. NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
  421. return skb->len;
  422. nla_put_failure:
  423. return -1;
  424. }
  425. static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt)
  426. {
  427. int prio;
  428. struct pfifo_fast_priv *priv = qdisc_priv(qdisc);
  429. for (prio = 0; prio < PFIFO_FAST_BANDS; prio++)
  430. skb_queue_head_init(band2list(priv, prio));
  431. return 0;
  432. }
  433. struct Qdisc_ops pfifo_fast_ops __read_mostly = {
  434. .id = "pfifo_fast",
  435. .priv_size = sizeof(struct pfifo_fast_priv),
  436. .enqueue = pfifo_fast_enqueue,
  437. .dequeue = pfifo_fast_dequeue,
  438. .peek = pfifo_fast_peek,
  439. .init = pfifo_fast_init,
  440. .reset = pfifo_fast_reset,
  441. .dump = pfifo_fast_dump,
  442. .owner = THIS_MODULE,
  443. };
  444. struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
  445. struct Qdisc_ops *ops)
  446. {
  447. void *p;
  448. struct Qdisc *sch;
  449. unsigned int size;
  450. int err = -ENOBUFS;
  451. /* ensure that the Qdisc and the private data are 64-byte aligned */
  452. size = QDISC_ALIGN(sizeof(*sch));
  453. size += ops->priv_size + (QDISC_ALIGNTO - 1);
  454. p = kzalloc(size, GFP_KERNEL);
  455. if (!p)
  456. goto errout;
  457. sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p);
  458. sch->padded = (char *) sch - (char *) p;
  459. INIT_LIST_HEAD(&sch->list);
  460. skb_queue_head_init(&sch->q);
  461. sch->ops = ops;
  462. sch->enqueue = ops->enqueue;
  463. sch->dequeue = ops->dequeue;
  464. sch->dev_queue = dev_queue;
  465. dev_hold(qdisc_dev(sch));
  466. atomic_set(&sch->refcnt, 1);
  467. return sch;
  468. errout:
  469. return ERR_PTR(err);
  470. }
  471. struct Qdisc * qdisc_create_dflt(struct net_device *dev,
  472. struct netdev_queue *dev_queue,
  473. struct Qdisc_ops *ops,
  474. unsigned int parentid)
  475. {
  476. struct Qdisc *sch;
  477. sch = qdisc_alloc(dev_queue, ops);
  478. if (IS_ERR(sch))
  479. goto errout;
  480. sch->parent = parentid;
  481. if (!ops->init || ops->init(sch, NULL) == 0)
  482. return sch;
  483. qdisc_destroy(sch);
  484. errout:
  485. return NULL;
  486. }
  487. EXPORT_SYMBOL(qdisc_create_dflt);
  488. /* Under qdisc_lock(qdisc) and BH! */
  489. void qdisc_reset(struct Qdisc *qdisc)
  490. {
  491. const struct Qdisc_ops *ops = qdisc->ops;
  492. if (ops->reset)
  493. ops->reset(qdisc);
  494. if (qdisc->gso_skb) {
  495. kfree_skb(qdisc->gso_skb);
  496. qdisc->gso_skb = NULL;
  497. qdisc->q.qlen = 0;
  498. }
  499. }
  500. EXPORT_SYMBOL(qdisc_reset);
  501. static void qdisc_rcu_free(struct rcu_head *head)
  502. {
  503. struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head);
  504. kfree((char *) qdisc - qdisc->padded);
  505. }
  506. void qdisc_destroy(struct Qdisc *qdisc)
  507. {
  508. const struct Qdisc_ops *ops = qdisc->ops;
  509. if (qdisc->flags & TCQ_F_BUILTIN ||
  510. !atomic_dec_and_test(&qdisc->refcnt))
  511. return;
  512. #ifdef CONFIG_NET_SCHED
  513. qdisc_list_del(qdisc);
  514. qdisc_put_stab(qdisc->stab);
  515. #endif
  516. gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est);
  517. if (ops->reset)
  518. ops->reset(qdisc);
  519. if (ops->destroy)
  520. ops->destroy(qdisc);
  521. module_put(ops->owner);
  522. dev_put(qdisc_dev(qdisc));
  523. kfree_skb(qdisc->gso_skb);
  524. /*
  525. * gen_estimator est_timer() might access qdisc->q.lock,
  526. * wait a RCU grace period before freeing qdisc.
  527. */
  528. call_rcu(&qdisc->rcu_head, qdisc_rcu_free);
  529. }
  530. EXPORT_SYMBOL(qdisc_destroy);
  531. /* Attach toplevel qdisc to device queue. */
  532. struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
  533. struct Qdisc *qdisc)
  534. {
  535. struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
  536. spinlock_t *root_lock;
  537. root_lock = qdisc_lock(oqdisc);
  538. spin_lock_bh(root_lock);
  539. /* Prune old scheduler */
  540. if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
  541. qdisc_reset(oqdisc);
  542. /* ... and graft new one */
  543. if (qdisc == NULL)
  544. qdisc = &noop_qdisc;
  545. dev_queue->qdisc_sleeping = qdisc;
  546. rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
  547. spin_unlock_bh(root_lock);
  548. return oqdisc;
  549. }
  550. static void attach_one_default_qdisc(struct net_device *dev,
  551. struct netdev_queue *dev_queue,
  552. void *_unused)
  553. {
  554. struct Qdisc *qdisc;
  555. if (dev->tx_queue_len) {
  556. qdisc = qdisc_create_dflt(dev, dev_queue,
  557. &pfifo_fast_ops, TC_H_ROOT);
  558. if (!qdisc) {
  559. printk(KERN_INFO "%s: activation failed\n", dev->name);
  560. return;
  561. }
  562. /* Can by-pass the queue discipline for default qdisc */
  563. qdisc->flags |= TCQ_F_CAN_BYPASS;
  564. } else {
  565. qdisc = &noqueue_qdisc;
  566. }
  567. dev_queue->qdisc_sleeping = qdisc;
  568. }
  569. static void attach_default_qdiscs(struct net_device *dev)
  570. {
  571. struct netdev_queue *txq;
  572. struct Qdisc *qdisc;
  573. txq = netdev_get_tx_queue(dev, 0);
  574. if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) {
  575. netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL);
  576. dev->qdisc = txq->qdisc_sleeping;
  577. atomic_inc(&dev->qdisc->refcnt);
  578. } else {
  579. qdisc = qdisc_create_dflt(dev, txq, &mq_qdisc_ops, TC_H_ROOT);
  580. if (qdisc) {
  581. qdisc->ops->attach(qdisc);
  582. dev->qdisc = qdisc;
  583. }
  584. }
  585. }
  586. static void transition_one_qdisc(struct net_device *dev,
  587. struct netdev_queue *dev_queue,
  588. void *_need_watchdog)
  589. {
  590. struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping;
  591. int *need_watchdog_p = _need_watchdog;
  592. if (!(new_qdisc->flags & TCQ_F_BUILTIN))
  593. clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state);
  594. rcu_assign_pointer(dev_queue->qdisc, new_qdisc);
  595. if (need_watchdog_p && new_qdisc != &noqueue_qdisc) {
  596. dev_queue->trans_start = 0;
  597. *need_watchdog_p = 1;
  598. }
  599. }
  600. void dev_activate(struct net_device *dev)
  601. {
  602. int need_watchdog;
  603. /* No queueing discipline is attached to device;
  604. create default one i.e. pfifo_fast for devices,
  605. which need queueing and noqueue_qdisc for
  606. virtual interfaces
  607. */
  608. if (dev->qdisc == &noop_qdisc)
  609. attach_default_qdiscs(dev);
  610. if (!netif_carrier_ok(dev))
  611. /* Delay activation until next carrier-on event */
  612. return;
  613. need_watchdog = 0;
  614. netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog);
  615. transition_one_qdisc(dev, &dev->rx_queue, NULL);
  616. if (need_watchdog) {
  617. dev->trans_start = jiffies;
  618. dev_watchdog_up(dev);
  619. }
  620. }
  621. static void dev_deactivate_queue(struct net_device *dev,
  622. struct netdev_queue *dev_queue,
  623. void *_qdisc_default)
  624. {
  625. struct Qdisc *qdisc_default = _qdisc_default;
  626. struct Qdisc *qdisc;
  627. qdisc = dev_queue->qdisc;
  628. if (qdisc) {
  629. spin_lock_bh(qdisc_lock(qdisc));
  630. if (!(qdisc->flags & TCQ_F_BUILTIN))
  631. set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state);
  632. rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
  633. qdisc_reset(qdisc);
  634. spin_unlock_bh(qdisc_lock(qdisc));
  635. }
  636. }
  637. static bool some_qdisc_is_busy(struct net_device *dev)
  638. {
  639. unsigned int i;
  640. for (i = 0; i < dev->num_tx_queues; i++) {
  641. struct netdev_queue *dev_queue;
  642. spinlock_t *root_lock;
  643. struct Qdisc *q;
  644. int val;
  645. dev_queue = netdev_get_tx_queue(dev, i);
  646. q = dev_queue->qdisc_sleeping;
  647. root_lock = qdisc_lock(q);
  648. spin_lock_bh(root_lock);
  649. val = (test_bit(__QDISC_STATE_RUNNING, &q->state) ||
  650. test_bit(__QDISC_STATE_SCHED, &q->state));
  651. spin_unlock_bh(root_lock);
  652. if (val)
  653. return true;
  654. }
  655. return false;
  656. }
  657. void dev_deactivate(struct net_device *dev)
  658. {
  659. netdev_for_each_tx_queue(dev, dev_deactivate_queue, &noop_qdisc);
  660. dev_deactivate_queue(dev, &dev->rx_queue, &noop_qdisc);
  661. dev_watchdog_down(dev);
  662. /* Wait for outstanding qdisc-less dev_queue_xmit calls. */
  663. synchronize_rcu();
  664. /* Wait for outstanding qdisc_run calls. */
  665. while (some_qdisc_is_busy(dev))
  666. yield();
  667. }
  668. static void dev_init_scheduler_queue(struct net_device *dev,
  669. struct netdev_queue *dev_queue,
  670. void *_qdisc)
  671. {
  672. struct Qdisc *qdisc = _qdisc;
  673. dev_queue->qdisc = qdisc;
  674. dev_queue->qdisc_sleeping = qdisc;
  675. }
  676. void dev_init_scheduler(struct net_device *dev)
  677. {
  678. dev->qdisc = &noop_qdisc;
  679. netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc);
  680. dev_init_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc);
  681. setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev);
  682. }
  683. static void shutdown_scheduler_queue(struct net_device *dev,
  684. struct netdev_queue *dev_queue,
  685. void *_qdisc_default)
  686. {
  687. struct Qdisc *qdisc = dev_queue->qdisc_sleeping;
  688. struct Qdisc *qdisc_default = _qdisc_default;
  689. if (qdisc) {
  690. rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
  691. dev_queue->qdisc_sleeping = qdisc_default;
  692. qdisc_destroy(qdisc);
  693. }
  694. }
  695. void dev_shutdown(struct net_device *dev)
  696. {
  697. netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc);
  698. shutdown_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc);
  699. qdisc_destroy(dev->qdisc);
  700. dev->qdisc = &noop_qdisc;
  701. WARN_ON(timer_pending(&dev->watchdog_timer));
  702. }