Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
1da177e4 | 14 | #include <linux/bitops.h> |
1da177e4 LT |
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> | |
1da177e4 | 20 | #include <linux/errno.h> |
1da177e4 LT |
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> | |
1da177e4 LT |
27 | #include <net/pkt_sched.h> |
28 | ||
29 | /* Main transmission queue. */ | |
30 | ||
0463d4ae | 31 | /* Modifications to data participating in scheduling must be protected with |
5fb66229 | 32 | * qdisc_lock(qdisc) spinlock. |
0463d4ae PM |
33 | * |
34 | * The idea is the following: | |
c7e4f3bb DM |
35 | * - enqueue, dequeue are serialized via qdisc root lock |
36 | * - ingress filtering is also serialized via qdisc root lock | |
0463d4ae | 37 | * - updates to tree and tree walking are only done under the rtnl mutex. |
1da177e4 | 38 | */ |
1da177e4 | 39 | |
c716a81a JHS |
40 | static inline int qdisc_qlen(struct Qdisc *q) |
41 | { | |
c716a81a JHS |
42 | return q->q.qlen; |
43 | } | |
44 | ||
37437bb2 | 45 | static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q) |
c716a81a | 46 | { |
554794de JP |
47 | if (unlikely(skb->next)) |
48 | q->gso_skb = skb; | |
49 | else | |
50 | __skb_queue_head(&q->requeue, skb); | |
6c1361a6 | 51 | |
37437bb2 | 52 | __netif_schedule(q); |
c716a81a JHS |
53 | return 0; |
54 | } | |
55 | ||
d3b753db | 56 | static inline struct sk_buff *dequeue_skb(struct Qdisc *q) |
c716a81a | 57 | { |
554794de JP |
58 | struct sk_buff *skb = q->gso_skb; |
59 | ||
60 | if (!skb) | |
61 | skb = skb_peek(&q->requeue); | |
c716a81a | 62 | |
ebf05982 JP |
63 | if (unlikely(skb)) { |
64 | struct net_device *dev = qdisc_dev(q); | |
65 | struct netdev_queue *txq; | |
66 | ||
67 | /* check the reason of requeuing without tx lock first */ | |
68 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); | |
554794de JP |
69 | if (!netif_tx_queue_stopped(txq) && |
70 | !netif_tx_queue_frozen(txq)) { | |
71 | if (q->gso_skb) | |
72 | q->gso_skb = NULL; | |
73 | else | |
74 | __skb_unlink(skb, &q->requeue); | |
75 | } else { | |
ebf05982 | 76 | skb = NULL; |
554794de | 77 | } |
ebf05982 | 78 | } else { |
c716a81a | 79 | skb = q->dequeue(q); |
ebf05982 | 80 | } |
c716a81a JHS |
81 | |
82 | return skb; | |
83 | } | |
84 | ||
6c1361a6 | 85 | static inline int handle_dev_cpu_collision(struct sk_buff *skb, |
970565bb | 86 | struct netdev_queue *dev_queue, |
6c1361a6 | 87 | struct Qdisc *q) |
c716a81a | 88 | { |
6c1361a6 | 89 | int ret; |
c716a81a | 90 | |
c773e847 | 91 | if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) { |
6c1361a6 KK |
92 | /* |
93 | * Same CPU holding the lock. It may be a transient | |
94 | * configuration error, when hard_start_xmit() recurses. We | |
95 | * detect it by checking xmit owner and drop the packet when | |
96 | * deadloop is detected. Return OK to try the next skb. | |
97 | */ | |
c716a81a | 98 | kfree_skb(skb); |
6c1361a6 KK |
99 | if (net_ratelimit()) |
100 | printk(KERN_WARNING "Dead loop on netdevice %s, " | |
c773e847 | 101 | "fix it urgently!\n", dev_queue->dev->name); |
6c1361a6 KK |
102 | ret = qdisc_qlen(q); |
103 | } else { | |
104 | /* | |
105 | * Another cpu is holding lock, requeue & delay xmits for | |
106 | * some time. | |
107 | */ | |
108 | __get_cpu_var(netdev_rx_stat).cpu_collision++; | |
37437bb2 | 109 | ret = dev_requeue_skb(skb, q); |
c716a81a JHS |
110 | } |
111 | ||
6c1361a6 | 112 | return ret; |
c716a81a JHS |
113 | } |
114 | ||
10297b99 | 115 | /* |
83874000 | 116 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. |
6c1361a6 | 117 | * |
e2627c8c | 118 | * __QDISC_STATE_RUNNING guarantees only one CPU can process |
83874000 DM |
119 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for |
120 | * this queue. | |
6c1361a6 KK |
121 | * |
122 | * netif_tx_lock serializes accesses to device driver. | |
123 | * | |
83874000 | 124 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, |
6c1361a6 KK |
125 | * if one is grabbed, another must be free. |
126 | * | |
127 | * Note, that this procedure can be called by a watchdog timer | |
128 | * | |
129 | * Returns to the caller: | |
130 | * 0 - queue is empty or throttled. | |
131 | * >0 - queue is not empty. | |
132 | * | |
133 | */ | |
37437bb2 | 134 | static inline int qdisc_restart(struct Qdisc *q) |
1da177e4 | 135 | { |
37437bb2 | 136 | struct netdev_queue *txq; |
5f1a485d | 137 | int ret = NETDEV_TX_BUSY; |
eb6aafe3 | 138 | struct net_device *dev; |
7698b4fc | 139 | spinlock_t *root_lock; |
eb6aafe3 | 140 | struct sk_buff *skb; |
1da177e4 | 141 | |
6c1361a6 | 142 | /* Dequeue packet */ |
d3b753db | 143 | if (unlikely((skb = dequeue_skb(q)) == NULL)) |
c716a81a | 144 | return 0; |
f6a78bfc | 145 | |
5fb66229 | 146 | root_lock = qdisc_lock(q); |
7698b4fc DM |
147 | |
148 | /* And release qdisc */ | |
149 | spin_unlock(root_lock); | |
c716a81a | 150 | |
37437bb2 DM |
151 | dev = qdisc_dev(q); |
152 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); | |
eb6aafe3 | 153 | |
c773e847 | 154 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
c3f26a26 DM |
155 | if (!netif_tx_queue_stopped(txq) && |
156 | !netif_tx_queue_frozen(txq)) | |
fd2ea0a7 | 157 | ret = dev_hard_start_xmit(skb, dev, txq); |
c773e847 | 158 | HARD_TX_UNLOCK(dev, txq); |
c716a81a | 159 | |
7698b4fc | 160 | spin_lock(root_lock); |
c716a81a | 161 | |
6c1361a6 KK |
162 | switch (ret) { |
163 | case NETDEV_TX_OK: | |
164 | /* Driver sent out skb successfully */ | |
165 | ret = qdisc_qlen(q); | |
166 | break; | |
167 | ||
168 | case NETDEV_TX_LOCKED: | |
169 | /* Driver try lock failed */ | |
eb6aafe3 | 170 | ret = handle_dev_cpu_collision(skb, txq, q); |
6c1361a6 KK |
171 | break; |
172 | ||
173 | default: | |
174 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ | |
175 | if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit())) | |
176 | printk(KERN_WARNING "BUG %s code %d qlen %d\n", | |
177 | dev->name, ret, q->q.qlen); | |
178 | ||
37437bb2 | 179 | ret = dev_requeue_skb(skb, q); |
6c1361a6 KK |
180 | break; |
181 | } | |
c716a81a | 182 | |
c3f26a26 DM |
183 | if (ret && (netif_tx_queue_stopped(txq) || |
184 | netif_tx_queue_frozen(txq))) | |
37437bb2 DM |
185 | ret = 0; |
186 | ||
6c1361a6 | 187 | return ret; |
1da177e4 LT |
188 | } |
189 | ||
37437bb2 | 190 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 191 | { |
2ba2506c | 192 | unsigned long start_time = jiffies; |
2ba2506c | 193 | |
37437bb2 | 194 | while (qdisc_restart(q)) { |
2ba2506c HX |
195 | /* |
196 | * Postpone processing if | |
197 | * 1. another process needs the CPU; | |
198 | * 2. we've been doing it for too long. | |
199 | */ | |
200 | if (need_resched() || jiffies != start_time) { | |
37437bb2 | 201 | __netif_schedule(q); |
d90df3ad | 202 | break; |
2ba2506c HX |
203 | } |
204 | } | |
48d83325 | 205 | |
e2627c8c | 206 | clear_bit(__QDISC_STATE_RUNNING, &q->state); |
48d83325 HX |
207 | } |
208 | ||
1da177e4 LT |
209 | static void dev_watchdog(unsigned long arg) |
210 | { | |
211 | struct net_device *dev = (struct net_device *)arg; | |
212 | ||
932ff279 | 213 | netif_tx_lock(dev); |
e8a0464c | 214 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
215 | if (netif_device_present(dev) && |
216 | netif_running(dev) && | |
217 | netif_carrier_ok(dev)) { | |
e8a0464c DM |
218 | int some_queue_stopped = 0; |
219 | unsigned int i; | |
220 | ||
221 | for (i = 0; i < dev->num_tx_queues; i++) { | |
222 | struct netdev_queue *txq; | |
223 | ||
224 | txq = netdev_get_tx_queue(dev, i); | |
225 | if (netif_tx_queue_stopped(txq)) { | |
226 | some_queue_stopped = 1; | |
227 | break; | |
228 | } | |
229 | } | |
338f7566 | 230 | |
e8a0464c DM |
231 | if (some_queue_stopped && |
232 | time_after(jiffies, (dev->trans_start + | |
233 | dev->watchdog_timeo))) { | |
6579e57b | 234 | char drivername[64]; |
5337407c | 235 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit timed out\n", |
6579e57b | 236 | dev->name, netdev_drivername(dev, drivername, 64)); |
1da177e4 LT |
237 | dev->tx_timeout(dev); |
238 | } | |
e8a0464c DM |
239 | if (!mod_timer(&dev->watchdog_timer, |
240 | round_jiffies(jiffies + | |
241 | dev->watchdog_timeo))) | |
1da177e4 LT |
242 | dev_hold(dev); |
243 | } | |
244 | } | |
932ff279 | 245 | netif_tx_unlock(dev); |
1da177e4 LT |
246 | |
247 | dev_put(dev); | |
248 | } | |
249 | ||
1da177e4 LT |
250 | void __netdev_watchdog_up(struct net_device *dev) |
251 | { | |
252 | if (dev->tx_timeout) { | |
253 | if (dev->watchdog_timeo <= 0) | |
254 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
255 | if (!mod_timer(&dev->watchdog_timer, |
256 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
257 | dev_hold(dev); |
258 | } | |
259 | } | |
260 | ||
261 | static void dev_watchdog_up(struct net_device *dev) | |
262 | { | |
1da177e4 | 263 | __netdev_watchdog_up(dev); |
1da177e4 LT |
264 | } |
265 | ||
266 | static void dev_watchdog_down(struct net_device *dev) | |
267 | { | |
932ff279 | 268 | netif_tx_lock_bh(dev); |
1da177e4 | 269 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 270 | dev_put(dev); |
932ff279 | 271 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
272 | } |
273 | ||
bea3348e SH |
274 | /** |
275 | * netif_carrier_on - set carrier | |
276 | * @dev: network device | |
277 | * | |
278 | * Device has detected that carrier. | |
279 | */ | |
0a242efc DV |
280 | void netif_carrier_on(struct net_device *dev) |
281 | { | |
bfaae0f0 | 282 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
0a242efc | 283 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
284 | if (netif_running(dev)) |
285 | __netdev_watchdog_up(dev); | |
286 | } | |
0a242efc | 287 | } |
62e3ba1b | 288 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 289 | |
bea3348e SH |
290 | /** |
291 | * netif_carrier_off - clear carrier | |
292 | * @dev: network device | |
293 | * | |
294 | * Device has detected loss of carrier. | |
295 | */ | |
0a242efc DV |
296 | void netif_carrier_off(struct net_device *dev) |
297 | { | |
298 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) | |
299 | linkwatch_fire_event(dev); | |
300 | } | |
62e3ba1b | 301 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 302 | |
1da177e4 LT |
303 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
304 | under all circumstances. It is difficult to invent anything faster or | |
305 | cheaper. | |
306 | */ | |
307 | ||
94df109a | 308 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc) |
1da177e4 LT |
309 | { |
310 | kfree_skb(skb); | |
311 | return NET_XMIT_CN; | |
312 | } | |
313 | ||
94df109a | 314 | static struct sk_buff *noop_dequeue(struct Qdisc * qdisc) |
1da177e4 LT |
315 | { |
316 | return NULL; | |
317 | } | |
318 | ||
94df109a | 319 | static int noop_requeue(struct sk_buff *skb, struct Qdisc* qdisc) |
1da177e4 LT |
320 | { |
321 | if (net_ratelimit()) | |
94df109a TG |
322 | printk(KERN_DEBUG "%s deferred output. It is buggy.\n", |
323 | skb->dev->name); | |
1da177e4 LT |
324 | kfree_skb(skb); |
325 | return NET_XMIT_CN; | |
326 | } | |
327 | ||
20fea08b | 328 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
329 | .id = "noop", |
330 | .priv_size = 0, | |
331 | .enqueue = noop_enqueue, | |
332 | .dequeue = noop_dequeue, | |
333 | .requeue = noop_requeue, | |
334 | .owner = THIS_MODULE, | |
335 | }; | |
336 | ||
7698b4fc | 337 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc DM |
338 | .qdisc = &noop_qdisc, |
339 | }; | |
340 | ||
1da177e4 LT |
341 | struct Qdisc noop_qdisc = { |
342 | .enqueue = noop_enqueue, | |
343 | .dequeue = noop_dequeue, | |
344 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 345 | .ops = &noop_qdisc_ops, |
1da177e4 | 346 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
242f8bfe | 347 | .requeue.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
83874000 | 348 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 349 | .dev_queue = &noop_netdev_queue, |
1da177e4 | 350 | }; |
62e3ba1b | 351 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 352 | |
20fea08b | 353 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
354 | .id = "noqueue", |
355 | .priv_size = 0, | |
356 | .enqueue = noop_enqueue, | |
357 | .dequeue = noop_dequeue, | |
358 | .requeue = noop_requeue, | |
359 | .owner = THIS_MODULE, | |
360 | }; | |
361 | ||
30ee42be DM |
362 | static struct Qdisc noqueue_qdisc; |
363 | static struct netdev_queue noqueue_netdev_queue = { | |
364 | .qdisc = &noqueue_qdisc, | |
365 | }; | |
366 | ||
1da177e4 LT |
367 | static struct Qdisc noqueue_qdisc = { |
368 | .enqueue = NULL, | |
369 | .dequeue = noop_dequeue, | |
370 | .flags = TCQ_F_BUILTIN, | |
371 | .ops = &noqueue_qdisc_ops, | |
372 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
242f8bfe | 373 | .requeue.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
30ee42be DM |
374 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
375 | .dev_queue = &noqueue_netdev_queue, | |
1da177e4 LT |
376 | }; |
377 | ||
378 | ||
d3678b46 DM |
379 | static const u8 prio2band[TC_PRIO_MAX+1] = |
380 | { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 }; | |
381 | ||
382 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
383 | generic prio+fifo combination. | |
384 | */ | |
385 | ||
386 | #define PFIFO_FAST_BANDS 3 | |
387 | ||
388 | static inline struct sk_buff_head *prio2list(struct sk_buff *skb, | |
389 | struct Qdisc *qdisc) | |
390 | { | |
391 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
392 | return list + prio2band[skb->priority & TC_PRIO_MAX]; | |
393 | } | |
394 | ||
395 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc) | |
321090e7 | 396 | { |
d3678b46 | 397 | struct sk_buff_head *list = prio2list(skb, qdisc); |
1da177e4 | 398 | |
d3678b46 DM |
399 | if (skb_queue_len(list) < qdisc_dev(qdisc)->tx_queue_len) { |
400 | qdisc->q.qlen++; | |
821d24ae | 401 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 402 | } |
821d24ae TG |
403 | |
404 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
405 | } |
406 | ||
d3678b46 | 407 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc) |
1da177e4 | 408 | { |
d3678b46 DM |
409 | int prio; |
410 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
1da177e4 | 411 | |
d3678b46 DM |
412 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) { |
413 | if (!skb_queue_empty(list + prio)) { | |
414 | qdisc->q.qlen--; | |
415 | return __qdisc_dequeue_head(qdisc, list + prio); | |
416 | } | |
417 | } | |
f87a9c3d | 418 | |
1da177e4 LT |
419 | return NULL; |
420 | } | |
421 | ||
d3678b46 | 422 | static int pfifo_fast_requeue(struct sk_buff *skb, struct Qdisc* qdisc) |
1da177e4 | 423 | { |
d3678b46 DM |
424 | qdisc->q.qlen++; |
425 | return __qdisc_requeue(skb, qdisc, prio2list(skb, qdisc)); | |
1da177e4 LT |
426 | } |
427 | ||
d3678b46 | 428 | static void pfifo_fast_reset(struct Qdisc* qdisc) |
1da177e4 | 429 | { |
d3678b46 DM |
430 | int prio; |
431 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
432 | ||
433 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
434 | __qdisc_reset_queue(qdisc, list + prio); | |
435 | ||
821d24ae | 436 | qdisc->qstats.backlog = 0; |
d3678b46 | 437 | qdisc->q.qlen = 0; |
1da177e4 LT |
438 | } |
439 | ||
d3678b46 DM |
440 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
441 | { | |
442 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
443 | ||
444 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1); | |
445 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); | |
446 | return skb->len; | |
447 | ||
448 | nla_put_failure: | |
449 | return -1; | |
450 | } | |
451 | ||
452 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
453 | { | |
454 | int prio; | |
455 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
456 | ||
457 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
458 | skb_queue_head_init(list + prio); | |
459 | ||
460 | return 0; | |
461 | } | |
462 | ||
463 | static struct Qdisc_ops pfifo_fast_ops __read_mostly = { | |
464 | .id = "pfifo_fast", | |
465 | .priv_size = PFIFO_FAST_BANDS * sizeof(struct sk_buff_head), | |
466 | .enqueue = pfifo_fast_enqueue, | |
467 | .dequeue = pfifo_fast_dequeue, | |
468 | .requeue = pfifo_fast_requeue, | |
469 | .init = pfifo_fast_init, | |
470 | .reset = pfifo_fast_reset, | |
471 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
472 | .owner = THIS_MODULE, |
473 | }; | |
474 | ||
5ce2d488 | 475 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
bb949fbd | 476 | struct Qdisc_ops *ops) |
1da177e4 LT |
477 | { |
478 | void *p; | |
479 | struct Qdisc *sch; | |
3d54b82f TG |
480 | unsigned int size; |
481 | int err = -ENOBUFS; | |
1da177e4 LT |
482 | |
483 | /* ensure that the Qdisc and the private data are 32-byte aligned */ | |
3d54b82f TG |
484 | size = QDISC_ALIGN(sizeof(*sch)); |
485 | size += ops->priv_size + (QDISC_ALIGNTO - 1); | |
1da177e4 | 486 | |
0da974f4 | 487 | p = kzalloc(size, GFP_KERNEL); |
1da177e4 | 488 | if (!p) |
3d54b82f | 489 | goto errout; |
3d54b82f TG |
490 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
491 | sch->padded = (char *) sch - (char *) p; | |
1da177e4 LT |
492 | |
493 | INIT_LIST_HEAD(&sch->list); | |
242f8bfe | 494 | skb_queue_head_init(&sch->requeue); |
1da177e4 LT |
495 | skb_queue_head_init(&sch->q); |
496 | sch->ops = ops; | |
497 | sch->enqueue = ops->enqueue; | |
498 | sch->dequeue = ops->dequeue; | |
bb949fbd | 499 | sch->dev_queue = dev_queue; |
5ce2d488 | 500 | dev_hold(qdisc_dev(sch)); |
1da177e4 | 501 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
502 | |
503 | return sch; | |
504 | errout: | |
01e123d7 | 505 | return ERR_PTR(err); |
3d54b82f TG |
506 | } |
507 | ||
bb949fbd DM |
508 | struct Qdisc * qdisc_create_dflt(struct net_device *dev, |
509 | struct netdev_queue *dev_queue, | |
510 | struct Qdisc_ops *ops, | |
9f9afec4 | 511 | unsigned int parentid) |
3d54b82f TG |
512 | { |
513 | struct Qdisc *sch; | |
10297b99 | 514 | |
5ce2d488 | 515 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
516 | if (IS_ERR(sch)) |
517 | goto errout; | |
9f9afec4 | 518 | sch->parent = parentid; |
3d54b82f | 519 | |
1da177e4 LT |
520 | if (!ops->init || ops->init(sch, NULL) == 0) |
521 | return sch; | |
522 | ||
0fbbeb1b | 523 | qdisc_destroy(sch); |
3d54b82f | 524 | errout: |
1da177e4 LT |
525 | return NULL; |
526 | } | |
62e3ba1b | 527 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 528 | |
5fb66229 | 529 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
530 | |
531 | void qdisc_reset(struct Qdisc *qdisc) | |
532 | { | |
20fea08b | 533 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
534 | |
535 | if (ops->reset) | |
536 | ops->reset(qdisc); | |
537 | } | |
62e3ba1b | 538 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 539 | |
1e0d5a57 | 540 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 541 | { |
8a34c5dc DM |
542 | const struct Qdisc_ops *ops = qdisc->ops; |
543 | ||
1e0d5a57 DM |
544 | if (qdisc->flags & TCQ_F_BUILTIN || |
545 | !atomic_dec_and_test(&qdisc->refcnt)) | |
546 | return; | |
547 | ||
3a682fbd | 548 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
549 | qdisc_list_del(qdisc); |
550 | ||
175f9c1b | 551 | qdisc_put_stab(qdisc->stab); |
3a682fbd | 552 | #endif |
8a34c5dc DM |
553 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
554 | if (ops->reset) | |
555 | ops->reset(qdisc); | |
556 | if (ops->destroy) | |
557 | ops->destroy(qdisc); | |
558 | ||
559 | module_put(ops->owner); | |
560 | dev_put(qdisc_dev(qdisc)); | |
561 | ||
554794de | 562 | kfree_skb(qdisc->gso_skb); |
242f8bfe | 563 | __skb_queue_purge(&qdisc->requeue); |
83874000 | 564 | |
1da177e4 LT |
565 | kfree((char *) qdisc - qdisc->padded); |
566 | } | |
62e3ba1b | 567 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 568 | |
e8a0464c DM |
569 | static bool dev_all_qdisc_sleeping_noop(struct net_device *dev) |
570 | { | |
571 | unsigned int i; | |
572 | ||
573 | for (i = 0; i < dev->num_tx_queues; i++) { | |
574 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
575 | ||
576 | if (txq->qdisc_sleeping != &noop_qdisc) | |
577 | return false; | |
578 | } | |
579 | return true; | |
580 | } | |
581 | ||
582 | static void attach_one_default_qdisc(struct net_device *dev, | |
583 | struct netdev_queue *dev_queue, | |
584 | void *_unused) | |
585 | { | |
586 | struct Qdisc *qdisc; | |
587 | ||
588 | if (dev->tx_queue_len) { | |
589 | qdisc = qdisc_create_dflt(dev, dev_queue, | |
d3678b46 | 590 | &pfifo_fast_ops, TC_H_ROOT); |
e8a0464c DM |
591 | if (!qdisc) { |
592 | printk(KERN_INFO "%s: activation failed\n", dev->name); | |
593 | return; | |
594 | } | |
e8a0464c DM |
595 | } else { |
596 | qdisc = &noqueue_qdisc; | |
597 | } | |
598 | dev_queue->qdisc_sleeping = qdisc; | |
599 | } | |
600 | ||
601 | static void transition_one_qdisc(struct net_device *dev, | |
602 | struct netdev_queue *dev_queue, | |
603 | void *_need_watchdog) | |
604 | { | |
83874000 | 605 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
606 | int *need_watchdog_p = _need_watchdog; |
607 | ||
a9312ae8 DM |
608 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
609 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
610 | ||
83874000 | 611 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
8d50b53d | 612 | if (need_watchdog_p && new_qdisc != &noqueue_qdisc) |
e8a0464c | 613 | *need_watchdog_p = 1; |
e8a0464c DM |
614 | } |
615 | ||
1da177e4 LT |
616 | void dev_activate(struct net_device *dev) |
617 | { | |
e8a0464c | 618 | int need_watchdog; |
b0e1e646 | 619 | |
1da177e4 | 620 | /* No queueing discipline is attached to device; |
d3678b46 DM |
621 | create default one i.e. pfifo_fast for devices, |
622 | which need queueing and noqueue_qdisc for | |
623 | virtual interfaces | |
1da177e4 LT |
624 | */ |
625 | ||
e8a0464c DM |
626 | if (dev_all_qdisc_sleeping_noop(dev)) |
627 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); | |
1da177e4 | 628 | |
cacaddf5 TC |
629 | if (!netif_carrier_ok(dev)) |
630 | /* Delay activation until next carrier-on event */ | |
631 | return; | |
632 | ||
e8a0464c DM |
633 | need_watchdog = 0; |
634 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
8d50b53d | 635 | transition_one_qdisc(dev, &dev->rx_queue, NULL); |
e8a0464c DM |
636 | |
637 | if (need_watchdog) { | |
1da177e4 LT |
638 | dev->trans_start = jiffies; |
639 | dev_watchdog_up(dev); | |
640 | } | |
b0e1e646 DM |
641 | } |
642 | ||
e8a0464c DM |
643 | static void dev_deactivate_queue(struct net_device *dev, |
644 | struct netdev_queue *dev_queue, | |
645 | void *_qdisc_default) | |
b0e1e646 | 646 | { |
e8a0464c | 647 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 648 | struct Qdisc *qdisc; |
970565bb | 649 | |
970565bb | 650 | qdisc = dev_queue->qdisc; |
b0e1e646 | 651 | if (qdisc) { |
83874000 DM |
652 | spin_lock_bh(qdisc_lock(qdisc)); |
653 | ||
a9312ae8 DM |
654 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
655 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
656 | ||
f7a54c13 | 657 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 658 | qdisc_reset(qdisc); |
d3b753db | 659 | |
83874000 | 660 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 661 | } |
1da177e4 LT |
662 | } |
663 | ||
4335cd2d | 664 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
665 | { |
666 | unsigned int i; | |
667 | ||
668 | for (i = 0; i < dev->num_tx_queues; i++) { | |
669 | struct netdev_queue *dev_queue; | |
7698b4fc | 670 | spinlock_t *root_lock; |
e2627c8c | 671 | struct Qdisc *q; |
e8a0464c DM |
672 | int val; |
673 | ||
674 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 675 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 676 | root_lock = qdisc_lock(q); |
e8a0464c | 677 | |
4335cd2d | 678 | spin_lock_bh(root_lock); |
e8a0464c | 679 | |
b9a3b110 DM |
680 | val = (test_bit(__QDISC_STATE_RUNNING, &q->state) || |
681 | test_bit(__QDISC_STATE_SCHED, &q->state)); | |
e8a0464c | 682 | |
4335cd2d | 683 | spin_unlock_bh(root_lock); |
e8a0464c DM |
684 | |
685 | if (val) | |
686 | return true; | |
687 | } | |
688 | return false; | |
689 | } | |
690 | ||
1da177e4 LT |
691 | void dev_deactivate(struct net_device *dev) |
692 | { | |
e8a0464c | 693 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, &noop_qdisc); |
8d50b53d | 694 | dev_deactivate_queue(dev, &dev->rx_queue, &noop_qdisc); |
41a23b07 | 695 | |
1da177e4 LT |
696 | dev_watchdog_down(dev); |
697 | ||
ce0e32e6 | 698 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. */ |
d4828d85 | 699 | synchronize_rcu(); |
1da177e4 | 700 | |
d4828d85 | 701 | /* Wait for outstanding qdisc_run calls. */ |
4335cd2d DM |
702 | while (some_qdisc_is_busy(dev)) |
703 | yield(); | |
1da177e4 LT |
704 | } |
705 | ||
b0e1e646 DM |
706 | static void dev_init_scheduler_queue(struct net_device *dev, |
707 | struct netdev_queue *dev_queue, | |
e8a0464c | 708 | void *_qdisc) |
b0e1e646 | 709 | { |
e8a0464c DM |
710 | struct Qdisc *qdisc = _qdisc; |
711 | ||
b0e1e646 DM |
712 | dev_queue->qdisc = qdisc; |
713 | dev_queue->qdisc_sleeping = qdisc; | |
b0e1e646 DM |
714 | } |
715 | ||
1da177e4 LT |
716 | void dev_init_scheduler(struct net_device *dev) |
717 | { | |
e8a0464c | 718 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
8d50b53d | 719 | dev_init_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
1da177e4 | 720 | |
b24b8a24 | 721 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
722 | } |
723 | ||
e8a0464c DM |
724 | static void shutdown_scheduler_queue(struct net_device *dev, |
725 | struct netdev_queue *dev_queue, | |
726 | void *_qdisc_default) | |
1da177e4 | 727 | { |
b0e1e646 | 728 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 729 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
730 | |
731 | if (qdisc) { | |
f7a54c13 | 732 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 733 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 734 | |
1da177e4 | 735 | qdisc_destroy(qdisc); |
10297b99 | 736 | } |
b0e1e646 DM |
737 | } |
738 | ||
739 | void dev_shutdown(struct net_device *dev) | |
740 | { | |
e8a0464c | 741 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
8d50b53d | 742 | shutdown_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
547b792c | 743 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 744 | } |