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 | { |
c716a81a | 47 | if (unlikely(skb->next)) |
d3b753db | 48 | q->gso_skb = skb; |
c716a81a JHS |
49 | else |
50 | q->ops->requeue(skb, q); | |
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 | { |
6c1361a6 | 58 | struct sk_buff *skb; |
c716a81a | 59 | |
d3b753db DM |
60 | if ((skb = q->gso_skb)) |
61 | q->gso_skb = NULL; | |
c716a81a JHS |
62 | else |
63 | skb = q->dequeue(q); | |
64 | ||
65 | return skb; | |
66 | } | |
67 | ||
6c1361a6 | 68 | static inline int handle_dev_cpu_collision(struct sk_buff *skb, |
970565bb | 69 | struct netdev_queue *dev_queue, |
6c1361a6 | 70 | struct Qdisc *q) |
c716a81a | 71 | { |
6c1361a6 | 72 | int ret; |
c716a81a | 73 | |
c773e847 | 74 | if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) { |
6c1361a6 KK |
75 | /* |
76 | * Same CPU holding the lock. It may be a transient | |
77 | * configuration error, when hard_start_xmit() recurses. We | |
78 | * detect it by checking xmit owner and drop the packet when | |
79 | * deadloop is detected. Return OK to try the next skb. | |
80 | */ | |
c716a81a | 81 | kfree_skb(skb); |
6c1361a6 KK |
82 | if (net_ratelimit()) |
83 | printk(KERN_WARNING "Dead loop on netdevice %s, " | |
c773e847 | 84 | "fix it urgently!\n", dev_queue->dev->name); |
6c1361a6 KK |
85 | ret = qdisc_qlen(q); |
86 | } else { | |
87 | /* | |
88 | * Another cpu is holding lock, requeue & delay xmits for | |
89 | * some time. | |
90 | */ | |
91 | __get_cpu_var(netdev_rx_stat).cpu_collision++; | |
37437bb2 | 92 | ret = dev_requeue_skb(skb, q); |
c716a81a JHS |
93 | } |
94 | ||
6c1361a6 | 95 | return ret; |
c716a81a JHS |
96 | } |
97 | ||
10297b99 | 98 | /* |
83874000 | 99 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. |
6c1361a6 | 100 | * |
e2627c8c | 101 | * __QDISC_STATE_RUNNING guarantees only one CPU can process |
83874000 DM |
102 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for |
103 | * this queue. | |
6c1361a6 KK |
104 | * |
105 | * netif_tx_lock serializes accesses to device driver. | |
106 | * | |
83874000 | 107 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, |
6c1361a6 KK |
108 | * if one is grabbed, another must be free. |
109 | * | |
110 | * Note, that this procedure can be called by a watchdog timer | |
111 | * | |
112 | * Returns to the caller: | |
113 | * 0 - queue is empty or throttled. | |
114 | * >0 - queue is not empty. | |
115 | * | |
116 | */ | |
37437bb2 | 117 | static inline int qdisc_restart(struct Qdisc *q) |
1da177e4 | 118 | { |
37437bb2 | 119 | struct netdev_queue *txq; |
5f1a485d | 120 | int ret = NETDEV_TX_BUSY; |
eb6aafe3 | 121 | struct net_device *dev; |
7698b4fc | 122 | spinlock_t *root_lock; |
eb6aafe3 | 123 | struct sk_buff *skb; |
1da177e4 | 124 | |
6c1361a6 | 125 | /* Dequeue packet */ |
d3b753db | 126 | if (unlikely((skb = dequeue_skb(q)) == NULL)) |
c716a81a | 127 | return 0; |
f6a78bfc | 128 | |
5fb66229 | 129 | root_lock = qdisc_lock(q); |
7698b4fc DM |
130 | |
131 | /* And release qdisc */ | |
132 | spin_unlock(root_lock); | |
c716a81a | 133 | |
37437bb2 DM |
134 | dev = qdisc_dev(q); |
135 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); | |
eb6aafe3 | 136 | |
c773e847 | 137 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
c3f26a26 DM |
138 | if (!netif_tx_queue_stopped(txq) && |
139 | !netif_tx_queue_frozen(txq)) | |
fd2ea0a7 | 140 | ret = dev_hard_start_xmit(skb, dev, txq); |
c773e847 | 141 | HARD_TX_UNLOCK(dev, txq); |
c716a81a | 142 | |
7698b4fc | 143 | spin_lock(root_lock); |
c716a81a | 144 | |
6c1361a6 KK |
145 | switch (ret) { |
146 | case NETDEV_TX_OK: | |
147 | /* Driver sent out skb successfully */ | |
148 | ret = qdisc_qlen(q); | |
149 | break; | |
150 | ||
151 | case NETDEV_TX_LOCKED: | |
152 | /* Driver try lock failed */ | |
eb6aafe3 | 153 | ret = handle_dev_cpu_collision(skb, txq, q); |
6c1361a6 KK |
154 | break; |
155 | ||
156 | default: | |
157 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ | |
158 | if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit())) | |
159 | printk(KERN_WARNING "BUG %s code %d qlen %d\n", | |
160 | dev->name, ret, q->q.qlen); | |
161 | ||
37437bb2 | 162 | ret = dev_requeue_skb(skb, q); |
6c1361a6 KK |
163 | break; |
164 | } | |
c716a81a | 165 | |
c3f26a26 DM |
166 | if (ret && (netif_tx_queue_stopped(txq) || |
167 | netif_tx_queue_frozen(txq))) | |
37437bb2 DM |
168 | ret = 0; |
169 | ||
6c1361a6 | 170 | return ret; |
1da177e4 LT |
171 | } |
172 | ||
37437bb2 | 173 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 174 | { |
2ba2506c | 175 | unsigned long start_time = jiffies; |
2ba2506c | 176 | |
37437bb2 | 177 | while (qdisc_restart(q)) { |
2ba2506c HX |
178 | /* |
179 | * Postpone processing if | |
180 | * 1. another process needs the CPU; | |
181 | * 2. we've been doing it for too long. | |
182 | */ | |
183 | if (need_resched() || jiffies != start_time) { | |
37437bb2 | 184 | __netif_schedule(q); |
d90df3ad | 185 | break; |
2ba2506c HX |
186 | } |
187 | } | |
48d83325 | 188 | |
e2627c8c | 189 | clear_bit(__QDISC_STATE_RUNNING, &q->state); |
48d83325 HX |
190 | } |
191 | ||
1da177e4 LT |
192 | static void dev_watchdog(unsigned long arg) |
193 | { | |
194 | struct net_device *dev = (struct net_device *)arg; | |
195 | ||
932ff279 | 196 | netif_tx_lock(dev); |
e8a0464c | 197 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
198 | if (netif_device_present(dev) && |
199 | netif_running(dev) && | |
200 | netif_carrier_ok(dev)) { | |
e8a0464c DM |
201 | int some_queue_stopped = 0; |
202 | unsigned int i; | |
203 | ||
204 | for (i = 0; i < dev->num_tx_queues; i++) { | |
205 | struct netdev_queue *txq; | |
206 | ||
207 | txq = netdev_get_tx_queue(dev, i); | |
208 | if (netif_tx_queue_stopped(txq)) { | |
209 | some_queue_stopped = 1; | |
210 | break; | |
211 | } | |
212 | } | |
338f7566 | 213 | |
e8a0464c DM |
214 | if (some_queue_stopped && |
215 | time_after(jiffies, (dev->trans_start + | |
216 | dev->watchdog_timeo))) { | |
6579e57b AV |
217 | char drivername[64]; |
218 | printk(KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit timed out\n", | |
219 | dev->name, netdev_drivername(dev, drivername, 64)); | |
1da177e4 | 220 | dev->tx_timeout(dev); |
b4192bbd | 221 | WARN_ON_ONCE(1); |
1da177e4 | 222 | } |
e8a0464c DM |
223 | if (!mod_timer(&dev->watchdog_timer, |
224 | round_jiffies(jiffies + | |
225 | dev->watchdog_timeo))) | |
1da177e4 LT |
226 | dev_hold(dev); |
227 | } | |
228 | } | |
932ff279 | 229 | netif_tx_unlock(dev); |
1da177e4 LT |
230 | |
231 | dev_put(dev); | |
232 | } | |
233 | ||
1da177e4 LT |
234 | void __netdev_watchdog_up(struct net_device *dev) |
235 | { | |
236 | if (dev->tx_timeout) { | |
237 | if (dev->watchdog_timeo <= 0) | |
238 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
239 | if (!mod_timer(&dev->watchdog_timer, |
240 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
241 | dev_hold(dev); |
242 | } | |
243 | } | |
244 | ||
245 | static void dev_watchdog_up(struct net_device *dev) | |
246 | { | |
1da177e4 | 247 | __netdev_watchdog_up(dev); |
1da177e4 LT |
248 | } |
249 | ||
250 | static void dev_watchdog_down(struct net_device *dev) | |
251 | { | |
932ff279 | 252 | netif_tx_lock_bh(dev); |
1da177e4 | 253 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 254 | dev_put(dev); |
932ff279 | 255 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
256 | } |
257 | ||
bea3348e SH |
258 | /** |
259 | * netif_carrier_on - set carrier | |
260 | * @dev: network device | |
261 | * | |
262 | * Device has detected that carrier. | |
263 | */ | |
0a242efc DV |
264 | void netif_carrier_on(struct net_device *dev) |
265 | { | |
bfaae0f0 | 266 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
0a242efc | 267 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
268 | if (netif_running(dev)) |
269 | __netdev_watchdog_up(dev); | |
270 | } | |
0a242efc | 271 | } |
62e3ba1b | 272 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 273 | |
bea3348e SH |
274 | /** |
275 | * netif_carrier_off - clear carrier | |
276 | * @dev: network device | |
277 | * | |
278 | * Device has detected loss of carrier. | |
279 | */ | |
0a242efc DV |
280 | void netif_carrier_off(struct net_device *dev) |
281 | { | |
282 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) | |
283 | linkwatch_fire_event(dev); | |
284 | } | |
62e3ba1b | 285 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 286 | |
1da177e4 LT |
287 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
288 | under all circumstances. It is difficult to invent anything faster or | |
289 | cheaper. | |
290 | */ | |
291 | ||
94df109a | 292 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc) |
1da177e4 LT |
293 | { |
294 | kfree_skb(skb); | |
295 | return NET_XMIT_CN; | |
296 | } | |
297 | ||
94df109a | 298 | static struct sk_buff *noop_dequeue(struct Qdisc * qdisc) |
1da177e4 LT |
299 | { |
300 | return NULL; | |
301 | } | |
302 | ||
94df109a | 303 | static int noop_requeue(struct sk_buff *skb, struct Qdisc* qdisc) |
1da177e4 LT |
304 | { |
305 | if (net_ratelimit()) | |
94df109a TG |
306 | printk(KERN_DEBUG "%s deferred output. It is buggy.\n", |
307 | skb->dev->name); | |
1da177e4 LT |
308 | kfree_skb(skb); |
309 | return NET_XMIT_CN; | |
310 | } | |
311 | ||
20fea08b | 312 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
313 | .id = "noop", |
314 | .priv_size = 0, | |
315 | .enqueue = noop_enqueue, | |
316 | .dequeue = noop_dequeue, | |
317 | .requeue = noop_requeue, | |
318 | .owner = THIS_MODULE, | |
319 | }; | |
320 | ||
7698b4fc | 321 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc DM |
322 | .qdisc = &noop_qdisc, |
323 | }; | |
324 | ||
1da177e4 LT |
325 | struct Qdisc noop_qdisc = { |
326 | .enqueue = noop_enqueue, | |
327 | .dequeue = noop_dequeue, | |
328 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 329 | .ops = &noop_qdisc_ops, |
1da177e4 | 330 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 331 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 332 | .dev_queue = &noop_netdev_queue, |
1da177e4 | 333 | }; |
62e3ba1b | 334 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 335 | |
20fea08b | 336 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
337 | .id = "noqueue", |
338 | .priv_size = 0, | |
339 | .enqueue = noop_enqueue, | |
340 | .dequeue = noop_dequeue, | |
341 | .requeue = noop_requeue, | |
342 | .owner = THIS_MODULE, | |
343 | }; | |
344 | ||
30ee42be DM |
345 | static struct Qdisc noqueue_qdisc; |
346 | static struct netdev_queue noqueue_netdev_queue = { | |
347 | .qdisc = &noqueue_qdisc, | |
348 | }; | |
349 | ||
1da177e4 LT |
350 | static struct Qdisc noqueue_qdisc = { |
351 | .enqueue = NULL, | |
352 | .dequeue = noop_dequeue, | |
353 | .flags = TCQ_F_BUILTIN, | |
354 | .ops = &noqueue_qdisc_ops, | |
355 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
30ee42be DM |
356 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
357 | .dev_queue = &noqueue_netdev_queue, | |
1da177e4 LT |
358 | }; |
359 | ||
360 | ||
d3678b46 DM |
361 | static const u8 prio2band[TC_PRIO_MAX+1] = |
362 | { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 }; | |
363 | ||
364 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
365 | generic prio+fifo combination. | |
366 | */ | |
367 | ||
368 | #define PFIFO_FAST_BANDS 3 | |
369 | ||
370 | static inline struct sk_buff_head *prio2list(struct sk_buff *skb, | |
371 | struct Qdisc *qdisc) | |
372 | { | |
373 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
374 | return list + prio2band[skb->priority & TC_PRIO_MAX]; | |
375 | } | |
376 | ||
377 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc) | |
321090e7 | 378 | { |
d3678b46 | 379 | struct sk_buff_head *list = prio2list(skb, qdisc); |
1da177e4 | 380 | |
d3678b46 DM |
381 | if (skb_queue_len(list) < qdisc_dev(qdisc)->tx_queue_len) { |
382 | qdisc->q.qlen++; | |
821d24ae | 383 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 384 | } |
821d24ae TG |
385 | |
386 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
387 | } |
388 | ||
d3678b46 | 389 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc) |
1da177e4 | 390 | { |
d3678b46 DM |
391 | int prio; |
392 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
1da177e4 | 393 | |
d3678b46 DM |
394 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) { |
395 | if (!skb_queue_empty(list + prio)) { | |
396 | qdisc->q.qlen--; | |
397 | return __qdisc_dequeue_head(qdisc, list + prio); | |
398 | } | |
399 | } | |
f87a9c3d | 400 | |
1da177e4 LT |
401 | return NULL; |
402 | } | |
403 | ||
d3678b46 | 404 | static int pfifo_fast_requeue(struct sk_buff *skb, struct Qdisc* qdisc) |
1da177e4 | 405 | { |
d3678b46 DM |
406 | qdisc->q.qlen++; |
407 | return __qdisc_requeue(skb, qdisc, prio2list(skb, qdisc)); | |
1da177e4 LT |
408 | } |
409 | ||
d3678b46 | 410 | static void pfifo_fast_reset(struct Qdisc* qdisc) |
1da177e4 | 411 | { |
d3678b46 DM |
412 | int prio; |
413 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
414 | ||
415 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
416 | __qdisc_reset_queue(qdisc, list + prio); | |
417 | ||
821d24ae | 418 | qdisc->qstats.backlog = 0; |
d3678b46 | 419 | qdisc->q.qlen = 0; |
1da177e4 LT |
420 | } |
421 | ||
d3678b46 DM |
422 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
423 | { | |
424 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
425 | ||
426 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1); | |
427 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); | |
428 | return skb->len; | |
429 | ||
430 | nla_put_failure: | |
431 | return -1; | |
432 | } | |
433 | ||
434 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
435 | { | |
436 | int prio; | |
437 | struct sk_buff_head *list = qdisc_priv(qdisc); | |
438 | ||
439 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
440 | skb_queue_head_init(list + prio); | |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
445 | static struct Qdisc_ops pfifo_fast_ops __read_mostly = { | |
446 | .id = "pfifo_fast", | |
447 | .priv_size = PFIFO_FAST_BANDS * sizeof(struct sk_buff_head), | |
448 | .enqueue = pfifo_fast_enqueue, | |
449 | .dequeue = pfifo_fast_dequeue, | |
450 | .requeue = pfifo_fast_requeue, | |
451 | .init = pfifo_fast_init, | |
452 | .reset = pfifo_fast_reset, | |
453 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
454 | .owner = THIS_MODULE, |
455 | }; | |
456 | ||
5ce2d488 | 457 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
bb949fbd | 458 | struct Qdisc_ops *ops) |
1da177e4 LT |
459 | { |
460 | void *p; | |
461 | struct Qdisc *sch; | |
3d54b82f TG |
462 | unsigned int size; |
463 | int err = -ENOBUFS; | |
1da177e4 LT |
464 | |
465 | /* ensure that the Qdisc and the private data are 32-byte aligned */ | |
3d54b82f TG |
466 | size = QDISC_ALIGN(sizeof(*sch)); |
467 | size += ops->priv_size + (QDISC_ALIGNTO - 1); | |
1da177e4 | 468 | |
0da974f4 | 469 | p = kzalloc(size, GFP_KERNEL); |
1da177e4 | 470 | if (!p) |
3d54b82f | 471 | goto errout; |
3d54b82f TG |
472 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
473 | sch->padded = (char *) sch - (char *) p; | |
1da177e4 LT |
474 | |
475 | INIT_LIST_HEAD(&sch->list); | |
476 | skb_queue_head_init(&sch->q); | |
477 | sch->ops = ops; | |
478 | sch->enqueue = ops->enqueue; | |
479 | sch->dequeue = ops->dequeue; | |
bb949fbd | 480 | sch->dev_queue = dev_queue; |
5ce2d488 | 481 | dev_hold(qdisc_dev(sch)); |
1da177e4 | 482 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
483 | |
484 | return sch; | |
485 | errout: | |
01e123d7 | 486 | return ERR_PTR(err); |
3d54b82f TG |
487 | } |
488 | ||
bb949fbd DM |
489 | struct Qdisc * qdisc_create_dflt(struct net_device *dev, |
490 | struct netdev_queue *dev_queue, | |
491 | struct Qdisc_ops *ops, | |
9f9afec4 | 492 | unsigned int parentid) |
3d54b82f TG |
493 | { |
494 | struct Qdisc *sch; | |
10297b99 | 495 | |
5ce2d488 | 496 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
497 | if (IS_ERR(sch)) |
498 | goto errout; | |
9f9afec4 | 499 | sch->parent = parentid; |
3d54b82f | 500 | |
1da177e4 LT |
501 | if (!ops->init || ops->init(sch, NULL) == 0) |
502 | return sch; | |
503 | ||
0fbbeb1b | 504 | qdisc_destroy(sch); |
3d54b82f | 505 | errout: |
1da177e4 LT |
506 | return NULL; |
507 | } | |
62e3ba1b | 508 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 509 | |
5fb66229 | 510 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
511 | |
512 | void qdisc_reset(struct Qdisc *qdisc) | |
513 | { | |
20fea08b | 514 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
515 | |
516 | if (ops->reset) | |
517 | ops->reset(qdisc); | |
518 | } | |
62e3ba1b | 519 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 520 | |
10297b99 | 521 | /* this is the rcu callback function to clean up a qdisc when there |
1da177e4 LT |
522 | * are no further references to it */ |
523 | ||
524 | static void __qdisc_destroy(struct rcu_head *head) | |
525 | { | |
526 | struct Qdisc *qdisc = container_of(head, struct Qdisc, q_rcu); | |
8a34c5dc DM |
527 | const struct Qdisc_ops *ops = qdisc->ops; |
528 | ||
3a682fbd | 529 | #ifdef CONFIG_NET_SCHED |
175f9c1b | 530 | qdisc_put_stab(qdisc->stab); |
3a682fbd | 531 | #endif |
8a34c5dc DM |
532 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
533 | if (ops->reset) | |
534 | ops->reset(qdisc); | |
535 | if (ops->destroy) | |
536 | ops->destroy(qdisc); | |
537 | ||
538 | module_put(ops->owner); | |
539 | dev_put(qdisc_dev(qdisc)); | |
540 | ||
83874000 DM |
541 | kfree_skb(qdisc->gso_skb); |
542 | ||
1da177e4 LT |
543 | kfree((char *) qdisc - qdisc->padded); |
544 | } | |
545 | ||
5fb66229 | 546 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
547 | |
548 | void qdisc_destroy(struct Qdisc *qdisc) | |
549 | { | |
1da177e4 | 550 | if (qdisc->flags & TCQ_F_BUILTIN || |
85670cc1 | 551 | !atomic_dec_and_test(&qdisc->refcnt)) |
1da177e4 LT |
552 | return; |
553 | ||
30723673 DM |
554 | if (qdisc->parent) |
555 | list_del(&qdisc->list); | |
1da177e4 LT |
556 | |
557 | call_rcu(&qdisc->q_rcu, __qdisc_destroy); | |
558 | } | |
62e3ba1b | 559 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 560 | |
e8a0464c DM |
561 | static bool dev_all_qdisc_sleeping_noop(struct net_device *dev) |
562 | { | |
563 | unsigned int i; | |
564 | ||
565 | for (i = 0; i < dev->num_tx_queues; i++) { | |
566 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
567 | ||
568 | if (txq->qdisc_sleeping != &noop_qdisc) | |
569 | return false; | |
570 | } | |
571 | return true; | |
572 | } | |
573 | ||
574 | static void attach_one_default_qdisc(struct net_device *dev, | |
575 | struct netdev_queue *dev_queue, | |
576 | void *_unused) | |
577 | { | |
578 | struct Qdisc *qdisc; | |
579 | ||
580 | if (dev->tx_queue_len) { | |
581 | qdisc = qdisc_create_dflt(dev, dev_queue, | |
d3678b46 | 582 | &pfifo_fast_ops, TC_H_ROOT); |
e8a0464c DM |
583 | if (!qdisc) { |
584 | printk(KERN_INFO "%s: activation failed\n", dev->name); | |
585 | return; | |
586 | } | |
e8a0464c DM |
587 | } else { |
588 | qdisc = &noqueue_qdisc; | |
589 | } | |
590 | dev_queue->qdisc_sleeping = qdisc; | |
591 | } | |
592 | ||
593 | static void transition_one_qdisc(struct net_device *dev, | |
594 | struct netdev_queue *dev_queue, | |
595 | void *_need_watchdog) | |
596 | { | |
83874000 | 597 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
598 | int *need_watchdog_p = _need_watchdog; |
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 | ||
b0e1e646 DM |
643 | dev_queue->qdisc = qdisc_default; |
644 | qdisc_reset(qdisc); | |
d3b753db | 645 | |
83874000 | 646 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 647 | } |
1da177e4 LT |
648 | } |
649 | ||
e8a0464c DM |
650 | static bool some_qdisc_is_running(struct net_device *dev, int lock) |
651 | { | |
652 | unsigned int i; | |
653 | ||
654 | for (i = 0; i < dev->num_tx_queues; i++) { | |
655 | struct netdev_queue *dev_queue; | |
7698b4fc | 656 | spinlock_t *root_lock; |
e2627c8c | 657 | struct Qdisc *q; |
e8a0464c DM |
658 | int val; |
659 | ||
660 | dev_queue = netdev_get_tx_queue(dev, i); | |
e2627c8c | 661 | q = dev_queue->qdisc; |
5fb66229 | 662 | root_lock = qdisc_lock(q); |
e8a0464c DM |
663 | |
664 | if (lock) | |
7698b4fc | 665 | spin_lock_bh(root_lock); |
e8a0464c | 666 | |
e2627c8c | 667 | val = test_bit(__QDISC_STATE_RUNNING, &q->state); |
e8a0464c DM |
668 | |
669 | if (lock) | |
7698b4fc | 670 | spin_unlock_bh(root_lock); |
e8a0464c DM |
671 | |
672 | if (val) | |
673 | return true; | |
674 | } | |
675 | return false; | |
676 | } | |
677 | ||
1da177e4 LT |
678 | void dev_deactivate(struct net_device *dev) |
679 | { | |
e8a0464c | 680 | bool running; |
1da177e4 | 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. */ |
ce0e32e6 | 691 | do { |
e8a0464c | 692 | while (some_qdisc_is_running(dev, 0)) |
ce0e32e6 HX |
693 | yield(); |
694 | ||
695 | /* | |
696 | * Double-check inside queue lock to ensure that all effects | |
697 | * of the queue run are visible when we return. | |
698 | */ | |
e8a0464c | 699 | running = some_qdisc_is_running(dev, 1); |
ce0e32e6 HX |
700 | |
701 | /* | |
702 | * The running flag should never be set at this point because | |
703 | * we've already set dev->qdisc to noop_qdisc *inside* the same | |
704 | * pair of spin locks. That is, if any qdisc_run starts after | |
705 | * our initial test it should see the noop_qdisc and then | |
706 | * clear the RUNNING bit before dropping the queue lock. So | |
707 | * if it is set here then we've found a bug. | |
708 | */ | |
709 | } while (WARN_ON_ONCE(running)); | |
1da177e4 LT |
710 | } |
711 | ||
b0e1e646 DM |
712 | static void dev_init_scheduler_queue(struct net_device *dev, |
713 | struct netdev_queue *dev_queue, | |
e8a0464c | 714 | void *_qdisc) |
b0e1e646 | 715 | { |
e8a0464c DM |
716 | struct Qdisc *qdisc = _qdisc; |
717 | ||
b0e1e646 DM |
718 | dev_queue->qdisc = qdisc; |
719 | dev_queue->qdisc_sleeping = qdisc; | |
b0e1e646 DM |
720 | } |
721 | ||
1da177e4 LT |
722 | void dev_init_scheduler(struct net_device *dev) |
723 | { | |
e8a0464c | 724 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
8d50b53d | 725 | dev_init_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
1da177e4 | 726 | |
b24b8a24 | 727 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
728 | } |
729 | ||
e8a0464c DM |
730 | static void shutdown_scheduler_queue(struct net_device *dev, |
731 | struct netdev_queue *dev_queue, | |
732 | void *_qdisc_default) | |
1da177e4 | 733 | { |
b0e1e646 | 734 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 735 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
736 | |
737 | if (qdisc) { | |
5fb66229 | 738 | spinlock_t *root_lock = qdisc_lock(qdisc); |
17715e62 | 739 | |
b0e1e646 DM |
740 | dev_queue->qdisc = qdisc_default; |
741 | dev_queue->qdisc_sleeping = qdisc_default; | |
1da177e4 | 742 | |
cffe1c5d | 743 | spin_lock_bh(root_lock); |
1da177e4 | 744 | qdisc_destroy(qdisc); |
cffe1c5d | 745 | spin_unlock_bh(root_lock); |
10297b99 | 746 | } |
b0e1e646 DM |
747 | } |
748 | ||
749 | void dev_shutdown(struct net_device *dev) | |
750 | { | |
e8a0464c | 751 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
8d50b53d | 752 | shutdown_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
547b792c | 753 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 754 | } |