]>
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 | { |
6252352d | 47 | q->gso_skb = skb; |
53e91503 | 48 | q->qstats.requeues++; |
37437bb2 | 49 | __netif_schedule(q); |
6252352d | 50 | |
c716a81a JHS |
51 | return 0; |
52 | } | |
53 | ||
d3b753db | 54 | static inline struct sk_buff *dequeue_skb(struct Qdisc *q) |
c716a81a | 55 | { |
554794de JP |
56 | struct sk_buff *skb = q->gso_skb; |
57 | ||
ebf05982 JP |
58 | if (unlikely(skb)) { |
59 | struct net_device *dev = qdisc_dev(q); | |
60 | struct netdev_queue *txq; | |
61 | ||
62 | /* check the reason of requeuing without tx lock first */ | |
63 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); | |
6252352d JP |
64 | if (!netif_tx_queue_stopped(txq) && !netif_tx_queue_frozen(txq)) |
65 | q->gso_skb = NULL; | |
66 | else | |
ebf05982 JP |
67 | skb = NULL; |
68 | } else { | |
c716a81a | 69 | skb = q->dequeue(q); |
ebf05982 | 70 | } |
c716a81a JHS |
71 | |
72 | return skb; | |
73 | } | |
74 | ||
6c1361a6 | 75 | static inline int handle_dev_cpu_collision(struct sk_buff *skb, |
970565bb | 76 | struct netdev_queue *dev_queue, |
6c1361a6 | 77 | struct Qdisc *q) |
c716a81a | 78 | { |
6c1361a6 | 79 | int ret; |
c716a81a | 80 | |
c773e847 | 81 | if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) { |
6c1361a6 KK |
82 | /* |
83 | * Same CPU holding the lock. It may be a transient | |
84 | * configuration error, when hard_start_xmit() recurses. We | |
85 | * detect it by checking xmit owner and drop the packet when | |
86 | * deadloop is detected. Return OK to try the next skb. | |
87 | */ | |
c716a81a | 88 | kfree_skb(skb); |
6c1361a6 KK |
89 | if (net_ratelimit()) |
90 | printk(KERN_WARNING "Dead loop on netdevice %s, " | |
c773e847 | 91 | "fix it urgently!\n", dev_queue->dev->name); |
6c1361a6 KK |
92 | ret = qdisc_qlen(q); |
93 | } else { | |
94 | /* | |
95 | * Another cpu is holding lock, requeue & delay xmits for | |
96 | * some time. | |
97 | */ | |
98 | __get_cpu_var(netdev_rx_stat).cpu_collision++; | |
37437bb2 | 99 | ret = dev_requeue_skb(skb, q); |
c716a81a JHS |
100 | } |
101 | ||
6c1361a6 | 102 | return ret; |
c716a81a JHS |
103 | } |
104 | ||
10297b99 | 105 | /* |
83874000 | 106 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. |
6c1361a6 | 107 | * |
e2627c8c | 108 | * __QDISC_STATE_RUNNING guarantees only one CPU can process |
83874000 DM |
109 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for |
110 | * this queue. | |
6c1361a6 KK |
111 | * |
112 | * netif_tx_lock serializes accesses to device driver. | |
113 | * | |
83874000 | 114 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, |
6c1361a6 KK |
115 | * if one is grabbed, another must be free. |
116 | * | |
117 | * Note, that this procedure can be called by a watchdog timer | |
118 | * | |
119 | * Returns to the caller: | |
120 | * 0 - queue is empty or throttled. | |
121 | * >0 - queue is not empty. | |
122 | * | |
123 | */ | |
37437bb2 | 124 | static inline int qdisc_restart(struct Qdisc *q) |
1da177e4 | 125 | { |
37437bb2 | 126 | struct netdev_queue *txq; |
5f1a485d | 127 | int ret = NETDEV_TX_BUSY; |
eb6aafe3 | 128 | struct net_device *dev; |
7698b4fc | 129 | spinlock_t *root_lock; |
eb6aafe3 | 130 | struct sk_buff *skb; |
1da177e4 | 131 | |
6c1361a6 | 132 | /* Dequeue packet */ |
d3b753db | 133 | if (unlikely((skb = dequeue_skb(q)) == NULL)) |
c716a81a | 134 | return 0; |
f6a78bfc | 135 | |
5fb66229 | 136 | root_lock = qdisc_lock(q); |
7698b4fc DM |
137 | |
138 | /* And release qdisc */ | |
139 | spin_unlock(root_lock); | |
c716a81a | 140 | |
37437bb2 DM |
141 | dev = qdisc_dev(q); |
142 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); | |
eb6aafe3 | 143 | |
c773e847 | 144 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
c3f26a26 DM |
145 | if (!netif_tx_queue_stopped(txq) && |
146 | !netif_tx_queue_frozen(txq)) | |
fd2ea0a7 | 147 | ret = dev_hard_start_xmit(skb, dev, txq); |
c773e847 | 148 | HARD_TX_UNLOCK(dev, txq); |
c716a81a | 149 | |
7698b4fc | 150 | spin_lock(root_lock); |
c716a81a | 151 | |
6c1361a6 KK |
152 | switch (ret) { |
153 | case NETDEV_TX_OK: | |
154 | /* Driver sent out skb successfully */ | |
155 | ret = qdisc_qlen(q); | |
156 | break; | |
157 | ||
158 | case NETDEV_TX_LOCKED: | |
159 | /* Driver try lock failed */ | |
eb6aafe3 | 160 | ret = handle_dev_cpu_collision(skb, txq, q); |
6c1361a6 KK |
161 | break; |
162 | ||
163 | default: | |
164 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ | |
165 | if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit())) | |
166 | printk(KERN_WARNING "BUG %s code %d qlen %d\n", | |
167 | dev->name, ret, q->q.qlen); | |
168 | ||
37437bb2 | 169 | ret = dev_requeue_skb(skb, q); |
6c1361a6 KK |
170 | break; |
171 | } | |
c716a81a | 172 | |
c3f26a26 DM |
173 | if (ret && (netif_tx_queue_stopped(txq) || |
174 | netif_tx_queue_frozen(txq))) | |
37437bb2 DM |
175 | ret = 0; |
176 | ||
6c1361a6 | 177 | return ret; |
1da177e4 LT |
178 | } |
179 | ||
37437bb2 | 180 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 181 | { |
2ba2506c | 182 | unsigned long start_time = jiffies; |
2ba2506c | 183 | |
37437bb2 | 184 | while (qdisc_restart(q)) { |
2ba2506c HX |
185 | /* |
186 | * Postpone processing if | |
187 | * 1. another process needs the CPU; | |
188 | * 2. we've been doing it for too long. | |
189 | */ | |
190 | if (need_resched() || jiffies != start_time) { | |
37437bb2 | 191 | __netif_schedule(q); |
d90df3ad | 192 | break; |
2ba2506c HX |
193 | } |
194 | } | |
48d83325 | 195 | |
e2627c8c | 196 | clear_bit(__QDISC_STATE_RUNNING, &q->state); |
48d83325 HX |
197 | } |
198 | ||
1da177e4 LT |
199 | static void dev_watchdog(unsigned long arg) |
200 | { | |
201 | struct net_device *dev = (struct net_device *)arg; | |
202 | ||
932ff279 | 203 | netif_tx_lock(dev); |
e8a0464c | 204 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
205 | if (netif_device_present(dev) && |
206 | netif_running(dev) && | |
207 | netif_carrier_ok(dev)) { | |
e8a0464c DM |
208 | int some_queue_stopped = 0; |
209 | unsigned int i; | |
210 | ||
211 | for (i = 0; i < dev->num_tx_queues; i++) { | |
212 | struct netdev_queue *txq; | |
213 | ||
214 | txq = netdev_get_tx_queue(dev, i); | |
215 | if (netif_tx_queue_stopped(txq)) { | |
216 | some_queue_stopped = 1; | |
217 | break; | |
218 | } | |
219 | } | |
338f7566 | 220 | |
e8a0464c DM |
221 | if (some_queue_stopped && |
222 | time_after(jiffies, (dev->trans_start + | |
223 | dev->watchdog_timeo))) { | |
6579e57b | 224 | char drivername[64]; |
5337407c | 225 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit timed out\n", |
6579e57b | 226 | dev->name, netdev_drivername(dev, drivername, 64)); |
1da177e4 LT |
227 | dev->tx_timeout(dev); |
228 | } | |
e8a0464c DM |
229 | if (!mod_timer(&dev->watchdog_timer, |
230 | round_jiffies(jiffies + | |
231 | dev->watchdog_timeo))) | |
1da177e4 LT |
232 | dev_hold(dev); |
233 | } | |
234 | } | |
932ff279 | 235 | netif_tx_unlock(dev); |
1da177e4 LT |
236 | |
237 | dev_put(dev); | |
238 | } | |
239 | ||
1da177e4 LT |
240 | void __netdev_watchdog_up(struct net_device *dev) |
241 | { | |
242 | if (dev->tx_timeout) { | |
243 | if (dev->watchdog_timeo <= 0) | |
244 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
245 | if (!mod_timer(&dev->watchdog_timer, |
246 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
247 | dev_hold(dev); |
248 | } | |
249 | } | |
250 | ||
251 | static void dev_watchdog_up(struct net_device *dev) | |
252 | { | |
1da177e4 | 253 | __netdev_watchdog_up(dev); |
1da177e4 LT |
254 | } |
255 | ||
256 | static void dev_watchdog_down(struct net_device *dev) | |
257 | { | |
932ff279 | 258 | netif_tx_lock_bh(dev); |
1da177e4 | 259 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 260 | dev_put(dev); |
932ff279 | 261 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
262 | } |
263 | ||
bea3348e SH |
264 | /** |
265 | * netif_carrier_on - set carrier | |
266 | * @dev: network device | |
267 | * | |
268 | * Device has detected that carrier. | |
269 | */ | |
0a242efc DV |
270 | void netif_carrier_on(struct net_device *dev) |
271 | { | |
bfaae0f0 | 272 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
0a242efc | 273 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
274 | if (netif_running(dev)) |
275 | __netdev_watchdog_up(dev); | |
276 | } | |
0a242efc | 277 | } |
62e3ba1b | 278 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 279 | |
bea3348e SH |
280 | /** |
281 | * netif_carrier_off - clear carrier | |
282 | * @dev: network device | |
283 | * | |
284 | * Device has detected loss of carrier. | |
285 | */ | |
0a242efc DV |
286 | void netif_carrier_off(struct net_device *dev) |
287 | { | |
288 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) | |
289 | linkwatch_fire_event(dev); | |
290 | } | |
62e3ba1b | 291 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 292 | |
1da177e4 LT |
293 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
294 | under all circumstances. It is difficult to invent anything faster or | |
295 | cheaper. | |
296 | */ | |
297 | ||
94df109a | 298 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc) |
1da177e4 LT |
299 | { |
300 | kfree_skb(skb); | |
301 | return NET_XMIT_CN; | |
302 | } | |
303 | ||
94df109a | 304 | static struct sk_buff *noop_dequeue(struct Qdisc * qdisc) |
1da177e4 LT |
305 | { |
306 | return NULL; | |
307 | } | |
308 | ||
94df109a | 309 | static int noop_requeue(struct sk_buff *skb, struct Qdisc* qdisc) |
1da177e4 LT |
310 | { |
311 | if (net_ratelimit()) | |
94df109a TG |
312 | printk(KERN_DEBUG "%s deferred output. It is buggy.\n", |
313 | skb->dev->name); | |
1da177e4 LT |
314 | kfree_skb(skb); |
315 | return NET_XMIT_CN; | |
316 | } | |
317 | ||
20fea08b | 318 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
319 | .id = "noop", |
320 | .priv_size = 0, | |
321 | .enqueue = noop_enqueue, | |
322 | .dequeue = noop_dequeue, | |
323 | .requeue = noop_requeue, | |
324 | .owner = THIS_MODULE, | |
325 | }; | |
326 | ||
7698b4fc | 327 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc DM |
328 | .qdisc = &noop_qdisc, |
329 | }; | |
330 | ||
1da177e4 LT |
331 | struct Qdisc noop_qdisc = { |
332 | .enqueue = noop_enqueue, | |
333 | .dequeue = noop_dequeue, | |
334 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 335 | .ops = &noop_qdisc_ops, |
1da177e4 | 336 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
242f8bfe | 337 | .requeue.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
83874000 | 338 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 339 | .dev_queue = &noop_netdev_queue, |
1da177e4 | 340 | }; |
62e3ba1b | 341 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 342 | |
20fea08b | 343 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
344 | .id = "noqueue", |
345 | .priv_size = 0, | |
346 | .enqueue = noop_enqueue, | |
347 | .dequeue = noop_dequeue, | |
348 | .requeue = noop_requeue, | |
349 | .owner = THIS_MODULE, | |
350 | }; | |
351 | ||
30ee42be DM |
352 | static struct Qdisc noqueue_qdisc; |
353 | static struct netdev_queue noqueue_netdev_queue = { | |
354 | .qdisc = &noqueue_qdisc, | |
355 | }; | |
356 | ||
1da177e4 LT |
357 | static struct Qdisc noqueue_qdisc = { |
358 | .enqueue = NULL, | |
359 | .dequeue = noop_dequeue, | |
360 | .flags = TCQ_F_BUILTIN, | |
361 | .ops = &noqueue_qdisc_ops, | |
362 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
242f8bfe | 363 | .requeue.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
30ee42be DM |
364 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
365 | .dev_queue = &noqueue_netdev_queue, | |
1da177e4 LT |
366 | }; |
367 | ||
368 | ||
d3678b46 DM |
369 | static const u8 prio2band[TC_PRIO_MAX+1] = |
370 | { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 }; | |
371 | ||
372 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
373 | generic prio+fifo combination. | |
374 | */ | |
375 | ||
376 | #define PFIFO_FAST_BANDS 3 | |
377 | ||
378 | static inline struct sk_buff_head *prio2list(struct sk_buff *skb, | |
379 | struct Qdisc *qdisc) | |
380 | { | |
381 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
382 | return list + prio2band[skb->priority & TC_PRIO_MAX]; | |
383 | } | |
384 | ||
385 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc) | |
321090e7 | 386 | { |
d3678b46 | 387 | struct sk_buff_head *list = prio2list(skb, qdisc); |
1da177e4 | 388 | |
d3678b46 DM |
389 | if (skb_queue_len(list) < qdisc_dev(qdisc)->tx_queue_len) { |
390 | qdisc->q.qlen++; | |
821d24ae | 391 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 392 | } |
821d24ae TG |
393 | |
394 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
395 | } |
396 | ||
d3678b46 | 397 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc) |
1da177e4 | 398 | { |
d3678b46 DM |
399 | int prio; |
400 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
1da177e4 | 401 | |
d3678b46 DM |
402 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) { |
403 | if (!skb_queue_empty(list + prio)) { | |
404 | qdisc->q.qlen--; | |
405 | return __qdisc_dequeue_head(qdisc, list + prio); | |
406 | } | |
407 | } | |
f87a9c3d | 408 | |
1da177e4 LT |
409 | return NULL; |
410 | } | |
411 | ||
d3678b46 | 412 | static int pfifo_fast_requeue(struct sk_buff *skb, struct Qdisc* qdisc) |
1da177e4 | 413 | { |
d3678b46 DM |
414 | qdisc->q.qlen++; |
415 | return __qdisc_requeue(skb, qdisc, prio2list(skb, qdisc)); | |
1da177e4 LT |
416 | } |
417 | ||
d3678b46 | 418 | static void pfifo_fast_reset(struct Qdisc* qdisc) |
1da177e4 | 419 | { |
d3678b46 DM |
420 | int prio; |
421 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
422 | ||
423 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
424 | __qdisc_reset_queue(qdisc, list + prio); | |
425 | ||
821d24ae | 426 | qdisc->qstats.backlog = 0; |
d3678b46 | 427 | qdisc->q.qlen = 0; |
1da177e4 LT |
428 | } |
429 | ||
d3678b46 DM |
430 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
431 | { | |
432 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
433 | ||
434 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1); | |
435 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); | |
436 | return skb->len; | |
437 | ||
438 | nla_put_failure: | |
439 | return -1; | |
440 | } | |
441 | ||
442 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
443 | { | |
444 | int prio; | |
445 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
446 | ||
447 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
448 | skb_queue_head_init(list + prio); | |
449 | ||
450 | return 0; | |
451 | } | |
452 | ||
453 | static struct Qdisc_ops pfifo_fast_ops __read_mostly = { | |
454 | .id = "pfifo_fast", | |
455 | .priv_size = PFIFO_FAST_BANDS * sizeof(struct sk_buff_head), | |
456 | .enqueue = pfifo_fast_enqueue, | |
457 | .dequeue = pfifo_fast_dequeue, | |
458 | .requeue = pfifo_fast_requeue, | |
459 | .init = pfifo_fast_init, | |
460 | .reset = pfifo_fast_reset, | |
461 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
462 | .owner = THIS_MODULE, |
463 | }; | |
464 | ||
5ce2d488 | 465 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
bb949fbd | 466 | struct Qdisc_ops *ops) |
1da177e4 LT |
467 | { |
468 | void *p; | |
469 | struct Qdisc *sch; | |
3d54b82f TG |
470 | unsigned int size; |
471 | int err = -ENOBUFS; | |
1da177e4 LT |
472 | |
473 | /* ensure that the Qdisc and the private data are 32-byte aligned */ | |
3d54b82f TG |
474 | size = QDISC_ALIGN(sizeof(*sch)); |
475 | size += ops->priv_size + (QDISC_ALIGNTO - 1); | |
1da177e4 | 476 | |
0da974f4 | 477 | p = kzalloc(size, GFP_KERNEL); |
1da177e4 | 478 | if (!p) |
3d54b82f | 479 | goto errout; |
3d54b82f TG |
480 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
481 | sch->padded = (char *) sch - (char *) p; | |
1da177e4 LT |
482 | |
483 | INIT_LIST_HEAD(&sch->list); | |
242f8bfe | 484 | skb_queue_head_init(&sch->requeue); |
1da177e4 LT |
485 | skb_queue_head_init(&sch->q); |
486 | sch->ops = ops; | |
487 | sch->enqueue = ops->enqueue; | |
488 | sch->dequeue = ops->dequeue; | |
bb949fbd | 489 | sch->dev_queue = dev_queue; |
5ce2d488 | 490 | dev_hold(qdisc_dev(sch)); |
1da177e4 | 491 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
492 | |
493 | return sch; | |
494 | errout: | |
01e123d7 | 495 | return ERR_PTR(err); |
3d54b82f TG |
496 | } |
497 | ||
bb949fbd DM |
498 | struct Qdisc * qdisc_create_dflt(struct net_device *dev, |
499 | struct netdev_queue *dev_queue, | |
500 | struct Qdisc_ops *ops, | |
9f9afec4 | 501 | unsigned int parentid) |
3d54b82f TG |
502 | { |
503 | struct Qdisc *sch; | |
10297b99 | 504 | |
5ce2d488 | 505 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
506 | if (IS_ERR(sch)) |
507 | goto errout; | |
9f9afec4 | 508 | sch->parent = parentid; |
3d54b82f | 509 | |
1da177e4 LT |
510 | if (!ops->init || ops->init(sch, NULL) == 0) |
511 | return sch; | |
512 | ||
0fbbeb1b | 513 | qdisc_destroy(sch); |
3d54b82f | 514 | errout: |
1da177e4 LT |
515 | return NULL; |
516 | } | |
62e3ba1b | 517 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 518 | |
5fb66229 | 519 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
520 | |
521 | void qdisc_reset(struct Qdisc *qdisc) | |
522 | { | |
20fea08b | 523 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
524 | |
525 | if (ops->reset) | |
526 | ops->reset(qdisc); | |
527 | } | |
62e3ba1b | 528 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 529 | |
1e0d5a57 | 530 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 531 | { |
8a34c5dc DM |
532 | const struct Qdisc_ops *ops = qdisc->ops; |
533 | ||
1e0d5a57 DM |
534 | if (qdisc->flags & TCQ_F_BUILTIN || |
535 | !atomic_dec_and_test(&qdisc->refcnt)) | |
536 | return; | |
537 | ||
3a682fbd | 538 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
539 | qdisc_list_del(qdisc); |
540 | ||
175f9c1b | 541 | qdisc_put_stab(qdisc->stab); |
3a682fbd | 542 | #endif |
8a34c5dc DM |
543 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
544 | if (ops->reset) | |
545 | ops->reset(qdisc); | |
546 | if (ops->destroy) | |
547 | ops->destroy(qdisc); | |
548 | ||
549 | module_put(ops->owner); | |
550 | dev_put(qdisc_dev(qdisc)); | |
551 | ||
554794de | 552 | kfree_skb(qdisc->gso_skb); |
242f8bfe | 553 | __skb_queue_purge(&qdisc->requeue); |
83874000 | 554 | |
1da177e4 LT |
555 | kfree((char *) qdisc - qdisc->padded); |
556 | } | |
62e3ba1b | 557 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 558 | |
e8a0464c DM |
559 | static bool dev_all_qdisc_sleeping_noop(struct net_device *dev) |
560 | { | |
561 | unsigned int i; | |
562 | ||
563 | for (i = 0; i < dev->num_tx_queues; i++) { | |
564 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
565 | ||
566 | if (txq->qdisc_sleeping != &noop_qdisc) | |
567 | return false; | |
568 | } | |
569 | return true; | |
570 | } | |
571 | ||
572 | static void attach_one_default_qdisc(struct net_device *dev, | |
573 | struct netdev_queue *dev_queue, | |
574 | void *_unused) | |
575 | { | |
576 | struct Qdisc *qdisc; | |
577 | ||
578 | if (dev->tx_queue_len) { | |
579 | qdisc = qdisc_create_dflt(dev, dev_queue, | |
d3678b46 | 580 | &pfifo_fast_ops, TC_H_ROOT); |
e8a0464c DM |
581 | if (!qdisc) { |
582 | printk(KERN_INFO "%s: activation failed\n", dev->name); | |
583 | return; | |
584 | } | |
e8a0464c DM |
585 | } else { |
586 | qdisc = &noqueue_qdisc; | |
587 | } | |
588 | dev_queue->qdisc_sleeping = qdisc; | |
589 | } | |
590 | ||
591 | static void transition_one_qdisc(struct net_device *dev, | |
592 | struct netdev_queue *dev_queue, | |
593 | void *_need_watchdog) | |
594 | { | |
83874000 | 595 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
596 | int *need_watchdog_p = _need_watchdog; |
597 | ||
a9312ae8 DM |
598 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
599 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
600 | ||
83874000 | 601 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
8d50b53d | 602 | if (need_watchdog_p && new_qdisc != &noqueue_qdisc) |
e8a0464c | 603 | *need_watchdog_p = 1; |
e8a0464c DM |
604 | } |
605 | ||
1da177e4 LT |
606 | void dev_activate(struct net_device *dev) |
607 | { | |
e8a0464c | 608 | int need_watchdog; |
b0e1e646 | 609 | |
1da177e4 | 610 | /* No queueing discipline is attached to device; |
d3678b46 DM |
611 | create default one i.e. pfifo_fast for devices, |
612 | which need queueing and noqueue_qdisc for | |
613 | virtual interfaces | |
1da177e4 LT |
614 | */ |
615 | ||
e8a0464c DM |
616 | if (dev_all_qdisc_sleeping_noop(dev)) |
617 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); | |
1da177e4 | 618 | |
cacaddf5 TC |
619 | if (!netif_carrier_ok(dev)) |
620 | /* Delay activation until next carrier-on event */ | |
621 | return; | |
622 | ||
e8a0464c DM |
623 | need_watchdog = 0; |
624 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
8d50b53d | 625 | transition_one_qdisc(dev, &dev->rx_queue, NULL); |
e8a0464c DM |
626 | |
627 | if (need_watchdog) { | |
1da177e4 LT |
628 | dev->trans_start = jiffies; |
629 | dev_watchdog_up(dev); | |
630 | } | |
b0e1e646 DM |
631 | } |
632 | ||
e8a0464c DM |
633 | static void dev_deactivate_queue(struct net_device *dev, |
634 | struct netdev_queue *dev_queue, | |
635 | void *_qdisc_default) | |
b0e1e646 | 636 | { |
e8a0464c | 637 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 638 | struct Qdisc *qdisc; |
970565bb | 639 | |
970565bb | 640 | qdisc = dev_queue->qdisc; |
b0e1e646 | 641 | if (qdisc) { |
83874000 DM |
642 | spin_lock_bh(qdisc_lock(qdisc)); |
643 | ||
a9312ae8 DM |
644 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
645 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
646 | ||
f7a54c13 | 647 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 648 | qdisc_reset(qdisc); |
d3b753db | 649 | |
83874000 | 650 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 651 | } |
1da177e4 LT |
652 | } |
653 | ||
4335cd2d | 654 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
655 | { |
656 | unsigned int i; | |
657 | ||
658 | for (i = 0; i < dev->num_tx_queues; i++) { | |
659 | struct netdev_queue *dev_queue; | |
7698b4fc | 660 | spinlock_t *root_lock; |
e2627c8c | 661 | struct Qdisc *q; |
e8a0464c DM |
662 | int val; |
663 | ||
664 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 665 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 666 | root_lock = qdisc_lock(q); |
e8a0464c | 667 | |
4335cd2d | 668 | spin_lock_bh(root_lock); |
e8a0464c | 669 | |
b9a3b110 DM |
670 | val = (test_bit(__QDISC_STATE_RUNNING, &q->state) || |
671 | test_bit(__QDISC_STATE_SCHED, &q->state)); | |
e8a0464c | 672 | |
4335cd2d | 673 | spin_unlock_bh(root_lock); |
e8a0464c DM |
674 | |
675 | if (val) | |
676 | return true; | |
677 | } | |
678 | return false; | |
679 | } | |
680 | ||
1da177e4 LT |
681 | void dev_deactivate(struct net_device *dev) |
682 | { | |
e8a0464c | 683 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, &noop_qdisc); |
8d50b53d | 684 | dev_deactivate_queue(dev, &dev->rx_queue, &noop_qdisc); |
41a23b07 | 685 | |
1da177e4 LT |
686 | dev_watchdog_down(dev); |
687 | ||
ce0e32e6 | 688 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. */ |
d4828d85 | 689 | synchronize_rcu(); |
1da177e4 | 690 | |
d4828d85 | 691 | /* Wait for outstanding qdisc_run calls. */ |
4335cd2d DM |
692 | while (some_qdisc_is_busy(dev)) |
693 | yield(); | |
1da177e4 LT |
694 | } |
695 | ||
b0e1e646 DM |
696 | static void dev_init_scheduler_queue(struct net_device *dev, |
697 | struct netdev_queue *dev_queue, | |
e8a0464c | 698 | void *_qdisc) |
b0e1e646 | 699 | { |
e8a0464c DM |
700 | struct Qdisc *qdisc = _qdisc; |
701 | ||
b0e1e646 DM |
702 | dev_queue->qdisc = qdisc; |
703 | dev_queue->qdisc_sleeping = qdisc; | |
b0e1e646 DM |
704 | } |
705 | ||
1da177e4 LT |
706 | void dev_init_scheduler(struct net_device *dev) |
707 | { | |
e8a0464c | 708 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
8d50b53d | 709 | dev_init_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
1da177e4 | 710 | |
b24b8a24 | 711 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
712 | } |
713 | ||
e8a0464c DM |
714 | static void shutdown_scheduler_queue(struct net_device *dev, |
715 | struct netdev_queue *dev_queue, | |
716 | void *_qdisc_default) | |
1da177e4 | 717 | { |
b0e1e646 | 718 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 719 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
720 | |
721 | if (qdisc) { | |
f7a54c13 | 722 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 723 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 724 | |
1da177e4 | 725 | qdisc_destroy(qdisc); |
10297b99 | 726 | } |
b0e1e646 DM |
727 | } |
728 | ||
729 | void dev_shutdown(struct net_device *dev) | |
730 | { | |
e8a0464c | 731 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
8d50b53d | 732 | shutdown_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
547b792c | 733 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 734 | } |