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