]>
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> | |
5a0e3ad6 | 27 | #include <linux/slab.h> |
1da177e4 | 28 | #include <net/pkt_sched.h> |
7fee226a | 29 | #include <net/dst.h> |
1da177e4 LT |
30 | |
31 | /* Main transmission queue. */ | |
32 | ||
0463d4ae | 33 | /* Modifications to data participating in scheduling must be protected with |
5fb66229 | 34 | * qdisc_lock(qdisc) spinlock. |
0463d4ae PM |
35 | * |
36 | * The idea is the following: | |
c7e4f3bb DM |
37 | * - enqueue, dequeue are serialized via qdisc root lock |
38 | * - ingress filtering is also serialized via qdisc root lock | |
0463d4ae | 39 | * - updates to tree and tree walking are only done under the rtnl mutex. |
1da177e4 | 40 | */ |
1da177e4 | 41 | |
37437bb2 | 42 | static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q) |
c716a81a | 43 | { |
7fee226a | 44 | skb_dst_force(skb); |
6252352d | 45 | q->gso_skb = skb; |
53e91503 | 46 | q->qstats.requeues++; |
bbd8a0d3 | 47 | q->q.qlen++; /* it's still part of the queue */ |
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)); | |
bbd8a0d3 KK |
63 | if (!netif_tx_queue_stopped(txq) && |
64 | !netif_tx_queue_frozen(txq)) { | |
6252352d | 65 | q->gso_skb = NULL; |
bbd8a0d3 KK |
66 | q->q.qlen--; |
67 | } else | |
ebf05982 JP |
68 | skb = NULL; |
69 | } else { | |
c716a81a | 70 | skb = q->dequeue(q); |
ebf05982 | 71 | } |
c716a81a JHS |
72 | |
73 | return skb; | |
74 | } | |
75 | ||
6c1361a6 | 76 | static inline int handle_dev_cpu_collision(struct sk_buff *skb, |
970565bb | 77 | struct netdev_queue *dev_queue, |
6c1361a6 | 78 | struct Qdisc *q) |
c716a81a | 79 | { |
6c1361a6 | 80 | int ret; |
c716a81a | 81 | |
c773e847 | 82 | if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) { |
6c1361a6 KK |
83 | /* |
84 | * Same CPU holding the lock. It may be a transient | |
85 | * configuration error, when hard_start_xmit() recurses. We | |
86 | * detect it by checking xmit owner and drop the packet when | |
87 | * deadloop is detected. Return OK to try the next skb. | |
88 | */ | |
c716a81a | 89 | kfree_skb(skb); |
6c1361a6 KK |
90 | if (net_ratelimit()) |
91 | printk(KERN_WARNING "Dead loop on netdevice %s, " | |
c773e847 | 92 | "fix it urgently!\n", dev_queue->dev->name); |
6c1361a6 KK |
93 | ret = qdisc_qlen(q); |
94 | } else { | |
95 | /* | |
96 | * Another cpu is holding lock, requeue & delay xmits for | |
97 | * some time. | |
98 | */ | |
dee42870 | 99 | __get_cpu_var(softnet_data).cpu_collision++; |
37437bb2 | 100 | ret = dev_requeue_skb(skb, q); |
c716a81a JHS |
101 | } |
102 | ||
6c1361a6 | 103 | return ret; |
c716a81a JHS |
104 | } |
105 | ||
10297b99 | 106 | /* |
bbd8a0d3 KK |
107 | * Transmit one skb, and handle the return status as required. Holding the |
108 | * __QDISC_STATE_RUNNING bit guarantees that only one CPU can execute this | |
109 | * function. | |
6c1361a6 KK |
110 | * |
111 | * Returns to the caller: | |
112 | * 0 - queue is empty or throttled. | |
113 | * >0 - queue is not empty. | |
6c1361a6 | 114 | */ |
bbd8a0d3 KK |
115 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
116 | struct net_device *dev, struct netdev_queue *txq, | |
117 | spinlock_t *root_lock) | |
1da177e4 | 118 | { |
5f1a485d | 119 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
120 | |
121 | /* And release qdisc */ | |
122 | spin_unlock(root_lock); | |
c716a81a | 123 | |
c773e847 | 124 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
9a1654ba | 125 | if (!netif_tx_queue_stopped(txq) && !netif_tx_queue_frozen(txq)) |
fd2ea0a7 | 126 | ret = dev_hard_start_xmit(skb, dev, txq); |
572a9d7b | 127 | |
c773e847 | 128 | HARD_TX_UNLOCK(dev, txq); |
c716a81a | 129 | |
7698b4fc | 130 | spin_lock(root_lock); |
c716a81a | 131 | |
9a1654ba JP |
132 | if (dev_xmit_complete(ret)) { |
133 | /* Driver sent out skb successfully or skb was consumed */ | |
6c1361a6 | 134 | ret = qdisc_qlen(q); |
9a1654ba | 135 | } else if (ret == NETDEV_TX_LOCKED) { |
6c1361a6 | 136 | /* Driver try lock failed */ |
eb6aafe3 | 137 | ret = handle_dev_cpu_collision(skb, txq, q); |
9a1654ba | 138 | } else { |
6c1361a6 KK |
139 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
140 | if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit())) | |
141 | printk(KERN_WARNING "BUG %s code %d qlen %d\n", | |
142 | dev->name, ret, q->q.qlen); | |
143 | ||
37437bb2 | 144 | ret = dev_requeue_skb(skb, q); |
6c1361a6 | 145 | } |
c716a81a | 146 | |
c3f26a26 DM |
147 | if (ret && (netif_tx_queue_stopped(txq) || |
148 | netif_tx_queue_frozen(txq))) | |
37437bb2 DM |
149 | ret = 0; |
150 | ||
6c1361a6 | 151 | return ret; |
1da177e4 LT |
152 | } |
153 | ||
bbd8a0d3 KK |
154 | /* |
155 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. | |
156 | * | |
157 | * __QDISC_STATE_RUNNING guarantees only one CPU can process | |
158 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for | |
159 | * this queue. | |
160 | * | |
161 | * netif_tx_lock serializes accesses to device driver. | |
162 | * | |
163 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, | |
164 | * if one is grabbed, another must be free. | |
165 | * | |
166 | * Note, that this procedure can be called by a watchdog timer | |
167 | * | |
168 | * Returns to the caller: | |
169 | * 0 - queue is empty or throttled. | |
170 | * >0 - queue is not empty. | |
171 | * | |
172 | */ | |
173 | static inline int qdisc_restart(struct Qdisc *q) | |
174 | { | |
175 | struct netdev_queue *txq; | |
176 | struct net_device *dev; | |
177 | spinlock_t *root_lock; | |
178 | struct sk_buff *skb; | |
179 | ||
180 | /* Dequeue packet */ | |
181 | skb = dequeue_skb(q); | |
182 | if (unlikely(!skb)) | |
183 | return 0; | |
7fee226a | 184 | WARN_ON_ONCE(skb_dst_is_noref(skb)); |
bbd8a0d3 KK |
185 | root_lock = qdisc_lock(q); |
186 | dev = qdisc_dev(q); | |
187 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); | |
188 | ||
189 | return sch_direct_xmit(skb, q, dev, txq, root_lock); | |
190 | } | |
191 | ||
37437bb2 | 192 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 193 | { |
2ba2506c | 194 | unsigned long start_time = jiffies; |
2ba2506c | 195 | |
37437bb2 | 196 | while (qdisc_restart(q)) { |
2ba2506c HX |
197 | /* |
198 | * Postpone processing if | |
199 | * 1. another process needs the CPU; | |
200 | * 2. we've been doing it for too long. | |
201 | */ | |
202 | if (need_resched() || jiffies != start_time) { | |
37437bb2 | 203 | __netif_schedule(q); |
d90df3ad | 204 | break; |
2ba2506c HX |
205 | } |
206 | } | |
48d83325 | 207 | |
e2627c8c | 208 | clear_bit(__QDISC_STATE_RUNNING, &q->state); |
48d83325 HX |
209 | } |
210 | ||
9d21493b ED |
211 | unsigned long dev_trans_start(struct net_device *dev) |
212 | { | |
213 | unsigned long val, res = dev->trans_start; | |
214 | unsigned int i; | |
215 | ||
216 | for (i = 0; i < dev->num_tx_queues; i++) { | |
217 | val = netdev_get_tx_queue(dev, i)->trans_start; | |
218 | if (val && time_after(val, res)) | |
219 | res = val; | |
220 | } | |
221 | dev->trans_start = res; | |
222 | return res; | |
223 | } | |
224 | EXPORT_SYMBOL(dev_trans_start); | |
225 | ||
1da177e4 LT |
226 | static void dev_watchdog(unsigned long arg) |
227 | { | |
228 | struct net_device *dev = (struct net_device *)arg; | |
229 | ||
932ff279 | 230 | netif_tx_lock(dev); |
e8a0464c | 231 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
232 | if (netif_device_present(dev) && |
233 | netif_running(dev) && | |
234 | netif_carrier_ok(dev)) { | |
9d21493b | 235 | int some_queue_timedout = 0; |
e8a0464c | 236 | unsigned int i; |
9d21493b | 237 | unsigned long trans_start; |
e8a0464c DM |
238 | |
239 | for (i = 0; i < dev->num_tx_queues; i++) { | |
240 | struct netdev_queue *txq; | |
241 | ||
242 | txq = netdev_get_tx_queue(dev, i); | |
9d21493b ED |
243 | /* |
244 | * old device drivers set dev->trans_start | |
245 | */ | |
246 | trans_start = txq->trans_start ? : dev->trans_start; | |
247 | if (netif_tx_queue_stopped(txq) && | |
248 | time_after(jiffies, (trans_start + | |
249 | dev->watchdog_timeo))) { | |
250 | some_queue_timedout = 1; | |
e8a0464c DM |
251 | break; |
252 | } | |
253 | } | |
338f7566 | 254 | |
9d21493b | 255 | if (some_queue_timedout) { |
6579e57b | 256 | char drivername[64]; |
9d21493b ED |
257 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
258 | dev->name, netdev_drivername(dev, drivername, 64), i); | |
d314774c | 259 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 260 | } |
e8a0464c DM |
261 | if (!mod_timer(&dev->watchdog_timer, |
262 | round_jiffies(jiffies + | |
263 | dev->watchdog_timeo))) | |
1da177e4 LT |
264 | dev_hold(dev); |
265 | } | |
266 | } | |
932ff279 | 267 | netif_tx_unlock(dev); |
1da177e4 LT |
268 | |
269 | dev_put(dev); | |
270 | } | |
271 | ||
1da177e4 LT |
272 | void __netdev_watchdog_up(struct net_device *dev) |
273 | { | |
d314774c | 274 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
275 | if (dev->watchdog_timeo <= 0) |
276 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
277 | if (!mod_timer(&dev->watchdog_timer, |
278 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
279 | dev_hold(dev); |
280 | } | |
281 | } | |
282 | ||
283 | static void dev_watchdog_up(struct net_device *dev) | |
284 | { | |
1da177e4 | 285 | __netdev_watchdog_up(dev); |
1da177e4 LT |
286 | } |
287 | ||
288 | static void dev_watchdog_down(struct net_device *dev) | |
289 | { | |
932ff279 | 290 | netif_tx_lock_bh(dev); |
1da177e4 | 291 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 292 | dev_put(dev); |
932ff279 | 293 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
294 | } |
295 | ||
bea3348e SH |
296 | /** |
297 | * netif_carrier_on - set carrier | |
298 | * @dev: network device | |
299 | * | |
300 | * Device has detected that carrier. | |
301 | */ | |
0a242efc DV |
302 | void netif_carrier_on(struct net_device *dev) |
303 | { | |
bfaae0f0 | 304 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
305 | if (dev->reg_state == NETREG_UNINITIALIZED) |
306 | return; | |
0a242efc | 307 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
308 | if (netif_running(dev)) |
309 | __netdev_watchdog_up(dev); | |
310 | } | |
0a242efc | 311 | } |
62e3ba1b | 312 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 313 | |
bea3348e SH |
314 | /** |
315 | * netif_carrier_off - clear carrier | |
316 | * @dev: network device | |
317 | * | |
318 | * Device has detected loss of carrier. | |
319 | */ | |
0a242efc DV |
320 | void netif_carrier_off(struct net_device *dev) |
321 | { | |
b4730016 DM |
322 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
323 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
324 | return; | |
0a242efc | 325 | linkwatch_fire_event(dev); |
b4730016 | 326 | } |
0a242efc | 327 | } |
62e3ba1b | 328 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 329 | |
1da177e4 LT |
330 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
331 | under all circumstances. It is difficult to invent anything faster or | |
332 | cheaper. | |
333 | */ | |
334 | ||
94df109a | 335 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc) |
1da177e4 LT |
336 | { |
337 | kfree_skb(skb); | |
338 | return NET_XMIT_CN; | |
339 | } | |
340 | ||
94df109a | 341 | static struct sk_buff *noop_dequeue(struct Qdisc * qdisc) |
1da177e4 LT |
342 | { |
343 | return NULL; | |
344 | } | |
345 | ||
20fea08b | 346 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
347 | .id = "noop", |
348 | .priv_size = 0, | |
349 | .enqueue = noop_enqueue, | |
350 | .dequeue = noop_dequeue, | |
99c0db26 | 351 | .peek = noop_dequeue, |
1da177e4 LT |
352 | .owner = THIS_MODULE, |
353 | }; | |
354 | ||
7698b4fc | 355 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 356 | .qdisc = &noop_qdisc, |
9f3ffae0 | 357 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
358 | }; |
359 | ||
1da177e4 LT |
360 | struct Qdisc noop_qdisc = { |
361 | .enqueue = noop_enqueue, | |
362 | .dequeue = noop_dequeue, | |
363 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 364 | .ops = &noop_qdisc_ops, |
1da177e4 | 365 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 366 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 367 | .dev_queue = &noop_netdev_queue, |
1da177e4 | 368 | }; |
62e3ba1b | 369 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 370 | |
20fea08b | 371 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
372 | .id = "noqueue", |
373 | .priv_size = 0, | |
374 | .enqueue = noop_enqueue, | |
375 | .dequeue = noop_dequeue, | |
99c0db26 | 376 | .peek = noop_dequeue, |
1da177e4 LT |
377 | .owner = THIS_MODULE, |
378 | }; | |
379 | ||
30ee42be DM |
380 | static struct Qdisc noqueue_qdisc; |
381 | static struct netdev_queue noqueue_netdev_queue = { | |
382 | .qdisc = &noqueue_qdisc, | |
9f3ffae0 | 383 | .qdisc_sleeping = &noqueue_qdisc, |
30ee42be DM |
384 | }; |
385 | ||
1da177e4 LT |
386 | static struct Qdisc noqueue_qdisc = { |
387 | .enqueue = NULL, | |
388 | .dequeue = noop_dequeue, | |
389 | .flags = TCQ_F_BUILTIN, | |
390 | .ops = &noqueue_qdisc_ops, | |
391 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
30ee42be DM |
392 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
393 | .dev_queue = &noqueue_netdev_queue, | |
1da177e4 LT |
394 | }; |
395 | ||
396 | ||
d3678b46 DM |
397 | static const u8 prio2band[TC_PRIO_MAX+1] = |
398 | { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 }; | |
399 | ||
400 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
401 | generic prio+fifo combination. | |
402 | */ | |
403 | ||
404 | #define PFIFO_FAST_BANDS 3 | |
405 | ||
fd3ae5e8 KK |
406 | /* |
407 | * Private data for a pfifo_fast scheduler containing: | |
408 | * - queues for the three band | |
409 | * - bitmap indicating which of the bands contain skbs | |
410 | */ | |
411 | struct pfifo_fast_priv { | |
412 | u32 bitmap; | |
413 | struct sk_buff_head q[PFIFO_FAST_BANDS]; | |
414 | }; | |
415 | ||
416 | /* | |
417 | * Convert a bitmap to the first band number where an skb is queued, where: | |
418 | * bitmap=0 means there are no skbs on any band. | |
419 | * bitmap=1 means there is an skb on band 0. | |
420 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
421 | */ | |
422 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
423 | ||
424 | static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv, | |
425 | int band) | |
d3678b46 | 426 | { |
fd3ae5e8 | 427 | return priv->q + band; |
d3678b46 DM |
428 | } |
429 | ||
430 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc) | |
321090e7 | 431 | { |
a453e068 KK |
432 | if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) { |
433 | int band = prio2band[skb->priority & TC_PRIO_MAX]; | |
434 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
435 | struct sk_buff_head *list = band2list(priv, band); | |
1da177e4 | 436 | |
fd3ae5e8 | 437 | priv->bitmap |= (1 << band); |
d3678b46 | 438 | qdisc->q.qlen++; |
821d24ae | 439 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 440 | } |
821d24ae TG |
441 | |
442 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
443 | } |
444 | ||
d3678b46 | 445 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc) |
1da177e4 | 446 | { |
fd3ae5e8 KK |
447 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
448 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 449 | |
fd3ae5e8 KK |
450 | if (likely(band >= 0)) { |
451 | struct sk_buff_head *list = band2list(priv, band); | |
452 | struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list); | |
453 | ||
454 | qdisc->q.qlen--; | |
455 | if (skb_queue_empty(list)) | |
456 | priv->bitmap &= ~(1 << band); | |
457 | ||
458 | return skb; | |
d3678b46 | 459 | } |
f87a9c3d | 460 | |
1da177e4 LT |
461 | return NULL; |
462 | } | |
463 | ||
99c0db26 JP |
464 | static struct sk_buff *pfifo_fast_peek(struct Qdisc* qdisc) |
465 | { | |
fd3ae5e8 KK |
466 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
467 | int band = bitmap2band[priv->bitmap]; | |
468 | ||
469 | if (band >= 0) { | |
470 | struct sk_buff_head *list = band2list(priv, band); | |
99c0db26 | 471 | |
fd3ae5e8 | 472 | return skb_peek(list); |
99c0db26 JP |
473 | } |
474 | ||
475 | return NULL; | |
476 | } | |
477 | ||
d3678b46 | 478 | static void pfifo_fast_reset(struct Qdisc* qdisc) |
1da177e4 | 479 | { |
d3678b46 | 480 | int prio; |
fd3ae5e8 | 481 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
482 | |
483 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 484 | __qdisc_reset_queue(qdisc, band2list(priv, prio)); |
d3678b46 | 485 | |
fd3ae5e8 | 486 | priv->bitmap = 0; |
821d24ae | 487 | qdisc->qstats.backlog = 0; |
d3678b46 | 488 | qdisc->q.qlen = 0; |
1da177e4 LT |
489 | } |
490 | ||
d3678b46 DM |
491 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
492 | { | |
493 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
494 | ||
495 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1); | |
496 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); | |
497 | return skb->len; | |
498 | ||
499 | nla_put_failure: | |
500 | return -1; | |
501 | } | |
502 | ||
503 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
504 | { | |
505 | int prio; | |
fd3ae5e8 | 506 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
507 | |
508 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 509 | skb_queue_head_init(band2list(priv, prio)); |
d3678b46 DM |
510 | |
511 | return 0; | |
512 | } | |
513 | ||
6ec1c69a | 514 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 515 | .id = "pfifo_fast", |
fd3ae5e8 | 516 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
517 | .enqueue = pfifo_fast_enqueue, |
518 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 519 | .peek = pfifo_fast_peek, |
d3678b46 DM |
520 | .init = pfifo_fast_init, |
521 | .reset = pfifo_fast_reset, | |
522 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
523 | .owner = THIS_MODULE, |
524 | }; | |
525 | ||
5ce2d488 | 526 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
bb949fbd | 527 | struct Qdisc_ops *ops) |
1da177e4 LT |
528 | { |
529 | void *p; | |
530 | struct Qdisc *sch; | |
3d54b82f TG |
531 | unsigned int size; |
532 | int err = -ENOBUFS; | |
1da177e4 | 533 | |
5d944c64 | 534 | /* ensure that the Qdisc and the private data are 64-byte aligned */ |
3d54b82f TG |
535 | size = QDISC_ALIGN(sizeof(*sch)); |
536 | size += ops->priv_size + (QDISC_ALIGNTO - 1); | |
1da177e4 | 537 | |
0da974f4 | 538 | p = kzalloc(size, GFP_KERNEL); |
1da177e4 | 539 | if (!p) |
3d54b82f | 540 | goto errout; |
3d54b82f TG |
541 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
542 | sch->padded = (char *) sch - (char *) p; | |
1da177e4 LT |
543 | |
544 | INIT_LIST_HEAD(&sch->list); | |
545 | skb_queue_head_init(&sch->q); | |
546 | sch->ops = ops; | |
547 | sch->enqueue = ops->enqueue; | |
548 | sch->dequeue = ops->dequeue; | |
bb949fbd | 549 | sch->dev_queue = dev_queue; |
5ce2d488 | 550 | dev_hold(qdisc_dev(sch)); |
1da177e4 | 551 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
552 | |
553 | return sch; | |
554 | errout: | |
01e123d7 | 555 | return ERR_PTR(err); |
3d54b82f TG |
556 | } |
557 | ||
bb949fbd DM |
558 | struct Qdisc * qdisc_create_dflt(struct net_device *dev, |
559 | struct netdev_queue *dev_queue, | |
560 | struct Qdisc_ops *ops, | |
9f9afec4 | 561 | unsigned int parentid) |
3d54b82f TG |
562 | { |
563 | struct Qdisc *sch; | |
10297b99 | 564 | |
5ce2d488 | 565 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
566 | if (IS_ERR(sch)) |
567 | goto errout; | |
9f9afec4 | 568 | sch->parent = parentid; |
3d54b82f | 569 | |
1da177e4 LT |
570 | if (!ops->init || ops->init(sch, NULL) == 0) |
571 | return sch; | |
572 | ||
0fbbeb1b | 573 | qdisc_destroy(sch); |
3d54b82f | 574 | errout: |
1da177e4 LT |
575 | return NULL; |
576 | } | |
62e3ba1b | 577 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 578 | |
5fb66229 | 579 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
580 | |
581 | void qdisc_reset(struct Qdisc *qdisc) | |
582 | { | |
20fea08b | 583 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
584 | |
585 | if (ops->reset) | |
586 | ops->reset(qdisc); | |
67305ebc | 587 | |
bbd8a0d3 KK |
588 | if (qdisc->gso_skb) { |
589 | kfree_skb(qdisc->gso_skb); | |
590 | qdisc->gso_skb = NULL; | |
591 | qdisc->q.qlen = 0; | |
592 | } | |
1da177e4 | 593 | } |
62e3ba1b | 594 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 595 | |
5d944c64 ED |
596 | static void qdisc_rcu_free(struct rcu_head *head) |
597 | { | |
598 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
599 | ||
600 | kfree((char *) qdisc - qdisc->padded); | |
601 | } | |
602 | ||
1e0d5a57 | 603 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 604 | { |
8a34c5dc DM |
605 | const struct Qdisc_ops *ops = qdisc->ops; |
606 | ||
1e0d5a57 DM |
607 | if (qdisc->flags & TCQ_F_BUILTIN || |
608 | !atomic_dec_and_test(&qdisc->refcnt)) | |
609 | return; | |
610 | ||
3a682fbd | 611 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
612 | qdisc_list_del(qdisc); |
613 | ||
175f9c1b | 614 | qdisc_put_stab(qdisc->stab); |
3a682fbd | 615 | #endif |
8a34c5dc DM |
616 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
617 | if (ops->reset) | |
618 | ops->reset(qdisc); | |
619 | if (ops->destroy) | |
620 | ops->destroy(qdisc); | |
621 | ||
622 | module_put(ops->owner); | |
623 | dev_put(qdisc_dev(qdisc)); | |
624 | ||
554794de | 625 | kfree_skb(qdisc->gso_skb); |
5d944c64 ED |
626 | /* |
627 | * gen_estimator est_timer() might access qdisc->q.lock, | |
628 | * wait a RCU grace period before freeing qdisc. | |
629 | */ | |
630 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 631 | } |
62e3ba1b | 632 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 633 | |
589983cd PM |
634 | /* Attach toplevel qdisc to device queue. */ |
635 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
636 | struct Qdisc *qdisc) | |
637 | { | |
638 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
639 | spinlock_t *root_lock; | |
640 | ||
641 | root_lock = qdisc_lock(oqdisc); | |
642 | spin_lock_bh(root_lock); | |
643 | ||
644 | /* Prune old scheduler */ | |
645 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
646 | qdisc_reset(oqdisc); | |
647 | ||
648 | /* ... and graft new one */ | |
649 | if (qdisc == NULL) | |
650 | qdisc = &noop_qdisc; | |
651 | dev_queue->qdisc_sleeping = qdisc; | |
652 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
653 | ||
654 | spin_unlock_bh(root_lock); | |
655 | ||
656 | return oqdisc; | |
657 | } | |
658 | ||
e8a0464c DM |
659 | static void attach_one_default_qdisc(struct net_device *dev, |
660 | struct netdev_queue *dev_queue, | |
661 | void *_unused) | |
662 | { | |
663 | struct Qdisc *qdisc; | |
664 | ||
665 | if (dev->tx_queue_len) { | |
666 | qdisc = qdisc_create_dflt(dev, dev_queue, | |
d3678b46 | 667 | &pfifo_fast_ops, TC_H_ROOT); |
e8a0464c DM |
668 | if (!qdisc) { |
669 | printk(KERN_INFO "%s: activation failed\n", dev->name); | |
670 | return; | |
671 | } | |
bbd8a0d3 KK |
672 | |
673 | /* Can by-pass the queue discipline for default qdisc */ | |
674 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
e8a0464c DM |
675 | } else { |
676 | qdisc = &noqueue_qdisc; | |
677 | } | |
678 | dev_queue->qdisc_sleeping = qdisc; | |
679 | } | |
680 | ||
6ec1c69a DM |
681 | static void attach_default_qdiscs(struct net_device *dev) |
682 | { | |
683 | struct netdev_queue *txq; | |
684 | struct Qdisc *qdisc; | |
685 | ||
686 | txq = netdev_get_tx_queue(dev, 0); | |
687 | ||
688 | if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) { | |
689 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); | |
690 | dev->qdisc = txq->qdisc_sleeping; | |
691 | atomic_inc(&dev->qdisc->refcnt); | |
692 | } else { | |
693 | qdisc = qdisc_create_dflt(dev, txq, &mq_qdisc_ops, TC_H_ROOT); | |
694 | if (qdisc) { | |
695 | qdisc->ops->attach(qdisc); | |
696 | dev->qdisc = qdisc; | |
697 | } | |
698 | } | |
699 | } | |
700 | ||
e8a0464c DM |
701 | static void transition_one_qdisc(struct net_device *dev, |
702 | struct netdev_queue *dev_queue, | |
703 | void *_need_watchdog) | |
704 | { | |
83874000 | 705 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
706 | int *need_watchdog_p = _need_watchdog; |
707 | ||
a9312ae8 DM |
708 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
709 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
710 | ||
83874000 | 711 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
9d21493b ED |
712 | if (need_watchdog_p && new_qdisc != &noqueue_qdisc) { |
713 | dev_queue->trans_start = 0; | |
e8a0464c | 714 | *need_watchdog_p = 1; |
9d21493b | 715 | } |
e8a0464c DM |
716 | } |
717 | ||
1da177e4 LT |
718 | void dev_activate(struct net_device *dev) |
719 | { | |
e8a0464c | 720 | int need_watchdog; |
b0e1e646 | 721 | |
1da177e4 | 722 | /* No queueing discipline is attached to device; |
d3678b46 DM |
723 | create default one i.e. pfifo_fast for devices, |
724 | which need queueing and noqueue_qdisc for | |
725 | virtual interfaces | |
1da177e4 LT |
726 | */ |
727 | ||
6ec1c69a DM |
728 | if (dev->qdisc == &noop_qdisc) |
729 | attach_default_qdiscs(dev); | |
af356afa | 730 | |
cacaddf5 TC |
731 | if (!netif_carrier_ok(dev)) |
732 | /* Delay activation until next carrier-on event */ | |
733 | return; | |
734 | ||
e8a0464c DM |
735 | need_watchdog = 0; |
736 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
8d50b53d | 737 | transition_one_qdisc(dev, &dev->rx_queue, NULL); |
e8a0464c DM |
738 | |
739 | if (need_watchdog) { | |
1da177e4 LT |
740 | dev->trans_start = jiffies; |
741 | dev_watchdog_up(dev); | |
742 | } | |
b0e1e646 DM |
743 | } |
744 | ||
e8a0464c DM |
745 | static void dev_deactivate_queue(struct net_device *dev, |
746 | struct netdev_queue *dev_queue, | |
747 | void *_qdisc_default) | |
b0e1e646 | 748 | { |
e8a0464c | 749 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 750 | struct Qdisc *qdisc; |
970565bb | 751 | |
970565bb | 752 | qdisc = dev_queue->qdisc; |
b0e1e646 | 753 | if (qdisc) { |
83874000 DM |
754 | spin_lock_bh(qdisc_lock(qdisc)); |
755 | ||
a9312ae8 DM |
756 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
757 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
758 | ||
f7a54c13 | 759 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 760 | qdisc_reset(qdisc); |
d3b753db | 761 | |
83874000 | 762 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 763 | } |
1da177e4 LT |
764 | } |
765 | ||
4335cd2d | 766 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
767 | { |
768 | unsigned int i; | |
769 | ||
770 | for (i = 0; i < dev->num_tx_queues; i++) { | |
771 | struct netdev_queue *dev_queue; | |
7698b4fc | 772 | spinlock_t *root_lock; |
e2627c8c | 773 | struct Qdisc *q; |
e8a0464c DM |
774 | int val; |
775 | ||
776 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 777 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 778 | root_lock = qdisc_lock(q); |
e8a0464c | 779 | |
4335cd2d | 780 | spin_lock_bh(root_lock); |
e8a0464c | 781 | |
b9a3b110 DM |
782 | val = (test_bit(__QDISC_STATE_RUNNING, &q->state) || |
783 | test_bit(__QDISC_STATE_SCHED, &q->state)); | |
e8a0464c | 784 | |
4335cd2d | 785 | spin_unlock_bh(root_lock); |
e8a0464c DM |
786 | |
787 | if (val) | |
788 | return true; | |
789 | } | |
790 | return false; | |
791 | } | |
792 | ||
1da177e4 LT |
793 | void dev_deactivate(struct net_device *dev) |
794 | { | |
e8a0464c | 795 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, &noop_qdisc); |
8d50b53d | 796 | dev_deactivate_queue(dev, &dev->rx_queue, &noop_qdisc); |
41a23b07 | 797 | |
1da177e4 LT |
798 | dev_watchdog_down(dev); |
799 | ||
ce0e32e6 | 800 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. */ |
d4828d85 | 801 | synchronize_rcu(); |
1da177e4 | 802 | |
d4828d85 | 803 | /* Wait for outstanding qdisc_run calls. */ |
4335cd2d DM |
804 | while (some_qdisc_is_busy(dev)) |
805 | yield(); | |
1da177e4 LT |
806 | } |
807 | ||
b0e1e646 DM |
808 | static void dev_init_scheduler_queue(struct net_device *dev, |
809 | struct netdev_queue *dev_queue, | |
e8a0464c | 810 | void *_qdisc) |
b0e1e646 | 811 | { |
e8a0464c DM |
812 | struct Qdisc *qdisc = _qdisc; |
813 | ||
b0e1e646 DM |
814 | dev_queue->qdisc = qdisc; |
815 | dev_queue->qdisc_sleeping = qdisc; | |
b0e1e646 DM |
816 | } |
817 | ||
1da177e4 LT |
818 | void dev_init_scheduler(struct net_device *dev) |
819 | { | |
af356afa | 820 | dev->qdisc = &noop_qdisc; |
e8a0464c | 821 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
8d50b53d | 822 | dev_init_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
1da177e4 | 823 | |
b24b8a24 | 824 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
825 | } |
826 | ||
e8a0464c DM |
827 | static void shutdown_scheduler_queue(struct net_device *dev, |
828 | struct netdev_queue *dev_queue, | |
829 | void *_qdisc_default) | |
1da177e4 | 830 | { |
b0e1e646 | 831 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 832 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
833 | |
834 | if (qdisc) { | |
f7a54c13 | 835 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 836 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 837 | |
1da177e4 | 838 | qdisc_destroy(qdisc); |
10297b99 | 839 | } |
b0e1e646 DM |
840 | } |
841 | ||
842 | void dev_shutdown(struct net_device *dev) | |
843 | { | |
e8a0464c | 844 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
8d50b53d | 845 | shutdown_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
af356afa PM |
846 | qdisc_destroy(dev->qdisc); |
847 | dev->qdisc = &noop_qdisc; | |
848 | ||
547b792c | 849 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 850 | } |