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 LT |
28 | #include <net/pkt_sched.h> |
29 | ||
30 | /* Main transmission queue. */ | |
31 | ||
0463d4ae | 32 | /* Modifications to data participating in scheduling must be protected with |
5fb66229 | 33 | * qdisc_lock(qdisc) spinlock. |
0463d4ae PM |
34 | * |
35 | * The idea is the following: | |
c7e4f3bb DM |
36 | * - enqueue, dequeue are serialized via qdisc root lock |
37 | * - ingress filtering is also serialized via qdisc root lock | |
0463d4ae | 38 | * - updates to tree and tree walking are only done under the rtnl mutex. |
1da177e4 | 39 | */ |
1da177e4 | 40 | |
37437bb2 | 41 | static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q) |
c716a81a | 42 | { |
6252352d | 43 | q->gso_skb = skb; |
53e91503 | 44 | q->qstats.requeues++; |
bbd8a0d3 | 45 | q->q.qlen++; /* it's still part of the queue */ |
37437bb2 | 46 | __netif_schedule(q); |
6252352d | 47 | |
c716a81a JHS |
48 | return 0; |
49 | } | |
50 | ||
d3b753db | 51 | static inline struct sk_buff *dequeue_skb(struct Qdisc *q) |
c716a81a | 52 | { |
554794de JP |
53 | struct sk_buff *skb = q->gso_skb; |
54 | ||
ebf05982 JP |
55 | if (unlikely(skb)) { |
56 | struct net_device *dev = qdisc_dev(q); | |
57 | struct netdev_queue *txq; | |
58 | ||
59 | /* check the reason of requeuing without tx lock first */ | |
60 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); | |
bbd8a0d3 KK |
61 | if (!netif_tx_queue_stopped(txq) && |
62 | !netif_tx_queue_frozen(txq)) { | |
6252352d | 63 | q->gso_skb = NULL; |
bbd8a0d3 KK |
64 | q->q.qlen--; |
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 | /* |
bbd8a0d3 KK |
105 | * Transmit one skb, and handle the return status as required. Holding the |
106 | * __QDISC_STATE_RUNNING bit guarantees that only one CPU can execute this | |
107 | * function. | |
6c1361a6 KK |
108 | * |
109 | * Returns to the caller: | |
110 | * 0 - queue is empty or throttled. | |
111 | * >0 - queue is not empty. | |
6c1361a6 | 112 | */ |
bbd8a0d3 KK |
113 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
114 | struct net_device *dev, struct netdev_queue *txq, | |
115 | spinlock_t *root_lock) | |
1da177e4 | 116 | { |
5f1a485d | 117 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
118 | |
119 | /* And release qdisc */ | |
120 | spin_unlock(root_lock); | |
c716a81a | 121 | |
c773e847 | 122 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
9a1654ba | 123 | if (!netif_tx_queue_stopped(txq) && !netif_tx_queue_frozen(txq)) |
fd2ea0a7 | 124 | ret = dev_hard_start_xmit(skb, dev, txq); |
572a9d7b | 125 | |
c773e847 | 126 | HARD_TX_UNLOCK(dev, txq); |
c716a81a | 127 | |
7698b4fc | 128 | spin_lock(root_lock); |
c716a81a | 129 | |
9a1654ba JP |
130 | if (dev_xmit_complete(ret)) { |
131 | /* Driver sent out skb successfully or skb was consumed */ | |
6c1361a6 | 132 | ret = qdisc_qlen(q); |
9a1654ba | 133 | } else if (ret == NETDEV_TX_LOCKED) { |
6c1361a6 | 134 | /* Driver try lock failed */ |
eb6aafe3 | 135 | ret = handle_dev_cpu_collision(skb, txq, q); |
9a1654ba | 136 | } else { |
6c1361a6 KK |
137 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
138 | if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit())) | |
139 | printk(KERN_WARNING "BUG %s code %d qlen %d\n", | |
140 | dev->name, ret, q->q.qlen); | |
141 | ||
37437bb2 | 142 | ret = dev_requeue_skb(skb, q); |
6c1361a6 | 143 | } |
c716a81a | 144 | |
c3f26a26 DM |
145 | if (ret && (netif_tx_queue_stopped(txq) || |
146 | netif_tx_queue_frozen(txq))) | |
37437bb2 DM |
147 | ret = 0; |
148 | ||
6c1361a6 | 149 | return ret; |
1da177e4 LT |
150 | } |
151 | ||
bbd8a0d3 KK |
152 | /* |
153 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. | |
154 | * | |
155 | * __QDISC_STATE_RUNNING guarantees only one CPU can process | |
156 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for | |
157 | * this queue. | |
158 | * | |
159 | * netif_tx_lock serializes accesses to device driver. | |
160 | * | |
161 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, | |
162 | * if one is grabbed, another must be free. | |
163 | * | |
164 | * Note, that this procedure can be called by a watchdog timer | |
165 | * | |
166 | * Returns to the caller: | |
167 | * 0 - queue is empty or throttled. | |
168 | * >0 - queue is not empty. | |
169 | * | |
170 | */ | |
171 | static inline int qdisc_restart(struct Qdisc *q) | |
172 | { | |
173 | struct netdev_queue *txq; | |
174 | struct net_device *dev; | |
175 | spinlock_t *root_lock; | |
176 | struct sk_buff *skb; | |
177 | ||
178 | /* Dequeue packet */ | |
179 | skb = dequeue_skb(q); | |
180 | if (unlikely(!skb)) | |
181 | return 0; | |
182 | ||
183 | root_lock = qdisc_lock(q); | |
184 | dev = qdisc_dev(q); | |
185 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); | |
186 | ||
187 | return sch_direct_xmit(skb, q, dev, txq, root_lock); | |
188 | } | |
189 | ||
37437bb2 | 190 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 191 | { |
2ba2506c | 192 | unsigned long start_time = jiffies; |
2ba2506c | 193 | |
37437bb2 | 194 | while (qdisc_restart(q)) { |
2ba2506c HX |
195 | /* |
196 | * Postpone processing if | |
197 | * 1. another process needs the CPU; | |
198 | * 2. we've been doing it for too long. | |
199 | */ | |
200 | if (need_resched() || jiffies != start_time) { | |
37437bb2 | 201 | __netif_schedule(q); |
d90df3ad | 202 | break; |
2ba2506c HX |
203 | } |
204 | } | |
48d83325 | 205 | |
e2627c8c | 206 | clear_bit(__QDISC_STATE_RUNNING, &q->state); |
48d83325 HX |
207 | } |
208 | ||
9d21493b ED |
209 | unsigned long dev_trans_start(struct net_device *dev) |
210 | { | |
211 | unsigned long val, res = dev->trans_start; | |
212 | unsigned int i; | |
213 | ||
214 | for (i = 0; i < dev->num_tx_queues; i++) { | |
215 | val = netdev_get_tx_queue(dev, i)->trans_start; | |
216 | if (val && time_after(val, res)) | |
217 | res = val; | |
218 | } | |
219 | dev->trans_start = res; | |
220 | return res; | |
221 | } | |
222 | EXPORT_SYMBOL(dev_trans_start); | |
223 | ||
1da177e4 LT |
224 | static void dev_watchdog(unsigned long arg) |
225 | { | |
226 | struct net_device *dev = (struct net_device *)arg; | |
227 | ||
932ff279 | 228 | netif_tx_lock(dev); |
e8a0464c | 229 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
230 | if (netif_device_present(dev) && |
231 | netif_running(dev) && | |
232 | netif_carrier_ok(dev)) { | |
9d21493b | 233 | int some_queue_timedout = 0; |
e8a0464c | 234 | unsigned int i; |
9d21493b | 235 | unsigned long trans_start; |
e8a0464c DM |
236 | |
237 | for (i = 0; i < dev->num_tx_queues; i++) { | |
238 | struct netdev_queue *txq; | |
239 | ||
240 | txq = netdev_get_tx_queue(dev, i); | |
9d21493b ED |
241 | /* |
242 | * old device drivers set dev->trans_start | |
243 | */ | |
244 | trans_start = txq->trans_start ? : dev->trans_start; | |
245 | if (netif_tx_queue_stopped(txq) && | |
246 | time_after(jiffies, (trans_start + | |
247 | dev->watchdog_timeo))) { | |
248 | some_queue_timedout = 1; | |
e8a0464c DM |
249 | break; |
250 | } | |
251 | } | |
338f7566 | 252 | |
9d21493b | 253 | if (some_queue_timedout) { |
6579e57b | 254 | char drivername[64]; |
9d21493b ED |
255 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
256 | dev->name, netdev_drivername(dev, drivername, 64), i); | |
d314774c | 257 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 258 | } |
e8a0464c DM |
259 | if (!mod_timer(&dev->watchdog_timer, |
260 | round_jiffies(jiffies + | |
261 | dev->watchdog_timeo))) | |
1da177e4 LT |
262 | dev_hold(dev); |
263 | } | |
264 | } | |
932ff279 | 265 | netif_tx_unlock(dev); |
1da177e4 LT |
266 | |
267 | dev_put(dev); | |
268 | } | |
269 | ||
1da177e4 LT |
270 | void __netdev_watchdog_up(struct net_device *dev) |
271 | { | |
d314774c | 272 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
273 | if (dev->watchdog_timeo <= 0) |
274 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
275 | if (!mod_timer(&dev->watchdog_timer, |
276 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
277 | dev_hold(dev); |
278 | } | |
279 | } | |
280 | ||
281 | static void dev_watchdog_up(struct net_device *dev) | |
282 | { | |
1da177e4 | 283 | __netdev_watchdog_up(dev); |
1da177e4 LT |
284 | } |
285 | ||
286 | static void dev_watchdog_down(struct net_device *dev) | |
287 | { | |
932ff279 | 288 | netif_tx_lock_bh(dev); |
1da177e4 | 289 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 290 | dev_put(dev); |
932ff279 | 291 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
292 | } |
293 | ||
bea3348e SH |
294 | /** |
295 | * netif_carrier_on - set carrier | |
296 | * @dev: network device | |
297 | * | |
298 | * Device has detected that carrier. | |
299 | */ | |
0a242efc DV |
300 | void netif_carrier_on(struct net_device *dev) |
301 | { | |
bfaae0f0 | 302 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
303 | if (dev->reg_state == NETREG_UNINITIALIZED) |
304 | return; | |
0a242efc | 305 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
306 | if (netif_running(dev)) |
307 | __netdev_watchdog_up(dev); | |
308 | } | |
0a242efc | 309 | } |
62e3ba1b | 310 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 311 | |
bea3348e SH |
312 | /** |
313 | * netif_carrier_off - clear carrier | |
314 | * @dev: network device | |
315 | * | |
316 | * Device has detected loss of carrier. | |
317 | */ | |
0a242efc DV |
318 | void netif_carrier_off(struct net_device *dev) |
319 | { | |
b4730016 DM |
320 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
321 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
322 | return; | |
0a242efc | 323 | linkwatch_fire_event(dev); |
b4730016 | 324 | } |
0a242efc | 325 | } |
62e3ba1b | 326 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 327 | |
1da177e4 LT |
328 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
329 | under all circumstances. It is difficult to invent anything faster or | |
330 | cheaper. | |
331 | */ | |
332 | ||
94df109a | 333 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc) |
1da177e4 LT |
334 | { |
335 | kfree_skb(skb); | |
336 | return NET_XMIT_CN; | |
337 | } | |
338 | ||
94df109a | 339 | static struct sk_buff *noop_dequeue(struct Qdisc * qdisc) |
1da177e4 LT |
340 | { |
341 | return NULL; | |
342 | } | |
343 | ||
20fea08b | 344 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
345 | .id = "noop", |
346 | .priv_size = 0, | |
347 | .enqueue = noop_enqueue, | |
348 | .dequeue = noop_dequeue, | |
99c0db26 | 349 | .peek = noop_dequeue, |
1da177e4 LT |
350 | .owner = THIS_MODULE, |
351 | }; | |
352 | ||
7698b4fc | 353 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 354 | .qdisc = &noop_qdisc, |
9f3ffae0 | 355 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
356 | }; |
357 | ||
1da177e4 LT |
358 | struct Qdisc noop_qdisc = { |
359 | .enqueue = noop_enqueue, | |
360 | .dequeue = noop_dequeue, | |
361 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 362 | .ops = &noop_qdisc_ops, |
1da177e4 | 363 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 364 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 365 | .dev_queue = &noop_netdev_queue, |
1da177e4 | 366 | }; |
62e3ba1b | 367 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 368 | |
20fea08b | 369 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
370 | .id = "noqueue", |
371 | .priv_size = 0, | |
372 | .enqueue = noop_enqueue, | |
373 | .dequeue = noop_dequeue, | |
99c0db26 | 374 | .peek = noop_dequeue, |
1da177e4 LT |
375 | .owner = THIS_MODULE, |
376 | }; | |
377 | ||
30ee42be DM |
378 | static struct Qdisc noqueue_qdisc; |
379 | static struct netdev_queue noqueue_netdev_queue = { | |
380 | .qdisc = &noqueue_qdisc, | |
9f3ffae0 | 381 | .qdisc_sleeping = &noqueue_qdisc, |
30ee42be DM |
382 | }; |
383 | ||
1da177e4 LT |
384 | static struct Qdisc noqueue_qdisc = { |
385 | .enqueue = NULL, | |
386 | .dequeue = noop_dequeue, | |
387 | .flags = TCQ_F_BUILTIN, | |
388 | .ops = &noqueue_qdisc_ops, | |
389 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
30ee42be DM |
390 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
391 | .dev_queue = &noqueue_netdev_queue, | |
1da177e4 LT |
392 | }; |
393 | ||
394 | ||
d3678b46 DM |
395 | static const u8 prio2band[TC_PRIO_MAX+1] = |
396 | { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 }; | |
397 | ||
398 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
399 | generic prio+fifo combination. | |
400 | */ | |
401 | ||
402 | #define PFIFO_FAST_BANDS 3 | |
403 | ||
fd3ae5e8 KK |
404 | /* |
405 | * Private data for a pfifo_fast scheduler containing: | |
406 | * - queues for the three band | |
407 | * - bitmap indicating which of the bands contain skbs | |
408 | */ | |
409 | struct pfifo_fast_priv { | |
410 | u32 bitmap; | |
411 | struct sk_buff_head q[PFIFO_FAST_BANDS]; | |
412 | }; | |
413 | ||
414 | /* | |
415 | * Convert a bitmap to the first band number where an skb is queued, where: | |
416 | * bitmap=0 means there are no skbs on any band. | |
417 | * bitmap=1 means there is an skb on band 0. | |
418 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
419 | */ | |
420 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
421 | ||
422 | static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv, | |
423 | int band) | |
d3678b46 | 424 | { |
fd3ae5e8 | 425 | return priv->q + band; |
d3678b46 DM |
426 | } |
427 | ||
428 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc) | |
321090e7 | 429 | { |
a453e068 KK |
430 | if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) { |
431 | int band = prio2band[skb->priority & TC_PRIO_MAX]; | |
432 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
433 | struct sk_buff_head *list = band2list(priv, band); | |
1da177e4 | 434 | |
fd3ae5e8 | 435 | priv->bitmap |= (1 << band); |
d3678b46 | 436 | qdisc->q.qlen++; |
821d24ae | 437 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 438 | } |
821d24ae TG |
439 | |
440 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
441 | } |
442 | ||
d3678b46 | 443 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc) |
1da177e4 | 444 | { |
fd3ae5e8 KK |
445 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
446 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 447 | |
fd3ae5e8 KK |
448 | if (likely(band >= 0)) { |
449 | struct sk_buff_head *list = band2list(priv, band); | |
450 | struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list); | |
451 | ||
452 | qdisc->q.qlen--; | |
453 | if (skb_queue_empty(list)) | |
454 | priv->bitmap &= ~(1 << band); | |
455 | ||
456 | return skb; | |
d3678b46 | 457 | } |
f87a9c3d | 458 | |
1da177e4 LT |
459 | return NULL; |
460 | } | |
461 | ||
99c0db26 JP |
462 | static struct sk_buff *pfifo_fast_peek(struct Qdisc* qdisc) |
463 | { | |
fd3ae5e8 KK |
464 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
465 | int band = bitmap2band[priv->bitmap]; | |
466 | ||
467 | if (band >= 0) { | |
468 | struct sk_buff_head *list = band2list(priv, band); | |
99c0db26 | 469 | |
fd3ae5e8 | 470 | return skb_peek(list); |
99c0db26 JP |
471 | } |
472 | ||
473 | return NULL; | |
474 | } | |
475 | ||
d3678b46 | 476 | static void pfifo_fast_reset(struct Qdisc* qdisc) |
1da177e4 | 477 | { |
d3678b46 | 478 | int prio; |
fd3ae5e8 | 479 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
480 | |
481 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 482 | __qdisc_reset_queue(qdisc, band2list(priv, prio)); |
d3678b46 | 483 | |
fd3ae5e8 | 484 | priv->bitmap = 0; |
821d24ae | 485 | qdisc->qstats.backlog = 0; |
d3678b46 | 486 | qdisc->q.qlen = 0; |
1da177e4 LT |
487 | } |
488 | ||
d3678b46 DM |
489 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
490 | { | |
491 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
492 | ||
493 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1); | |
494 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); | |
495 | return skb->len; | |
496 | ||
497 | nla_put_failure: | |
498 | return -1; | |
499 | } | |
500 | ||
501 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
502 | { | |
503 | int prio; | |
fd3ae5e8 | 504 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
505 | |
506 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 507 | skb_queue_head_init(band2list(priv, prio)); |
d3678b46 DM |
508 | |
509 | return 0; | |
510 | } | |
511 | ||
6ec1c69a | 512 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 513 | .id = "pfifo_fast", |
fd3ae5e8 | 514 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
515 | .enqueue = pfifo_fast_enqueue, |
516 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 517 | .peek = pfifo_fast_peek, |
d3678b46 DM |
518 | .init = pfifo_fast_init, |
519 | .reset = pfifo_fast_reset, | |
520 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
521 | .owner = THIS_MODULE, |
522 | }; | |
523 | ||
5ce2d488 | 524 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
bb949fbd | 525 | struct Qdisc_ops *ops) |
1da177e4 LT |
526 | { |
527 | void *p; | |
528 | struct Qdisc *sch; | |
3d54b82f TG |
529 | unsigned int size; |
530 | int err = -ENOBUFS; | |
1da177e4 LT |
531 | |
532 | /* ensure that the Qdisc and the private data are 32-byte aligned */ | |
3d54b82f TG |
533 | size = QDISC_ALIGN(sizeof(*sch)); |
534 | size += ops->priv_size + (QDISC_ALIGNTO - 1); | |
1da177e4 | 535 | |
0da974f4 | 536 | p = kzalloc(size, GFP_KERNEL); |
1da177e4 | 537 | if (!p) |
3d54b82f | 538 | goto errout; |
3d54b82f TG |
539 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
540 | sch->padded = (char *) sch - (char *) p; | |
1da177e4 LT |
541 | |
542 | INIT_LIST_HEAD(&sch->list); | |
543 | skb_queue_head_init(&sch->q); | |
544 | sch->ops = ops; | |
545 | sch->enqueue = ops->enqueue; | |
546 | sch->dequeue = ops->dequeue; | |
bb949fbd | 547 | sch->dev_queue = dev_queue; |
5ce2d488 | 548 | dev_hold(qdisc_dev(sch)); |
1da177e4 | 549 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
550 | |
551 | return sch; | |
552 | errout: | |
01e123d7 | 553 | return ERR_PTR(err); |
3d54b82f TG |
554 | } |
555 | ||
bb949fbd DM |
556 | struct Qdisc * qdisc_create_dflt(struct net_device *dev, |
557 | struct netdev_queue *dev_queue, | |
558 | struct Qdisc_ops *ops, | |
9f9afec4 | 559 | unsigned int parentid) |
3d54b82f TG |
560 | { |
561 | struct Qdisc *sch; | |
10297b99 | 562 | |
5ce2d488 | 563 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
564 | if (IS_ERR(sch)) |
565 | goto errout; | |
9f9afec4 | 566 | sch->parent = parentid; |
3d54b82f | 567 | |
1da177e4 LT |
568 | if (!ops->init || ops->init(sch, NULL) == 0) |
569 | return sch; | |
570 | ||
0fbbeb1b | 571 | qdisc_destroy(sch); |
3d54b82f | 572 | errout: |
1da177e4 LT |
573 | return NULL; |
574 | } | |
62e3ba1b | 575 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 576 | |
5fb66229 | 577 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
578 | |
579 | void qdisc_reset(struct Qdisc *qdisc) | |
580 | { | |
20fea08b | 581 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
582 | |
583 | if (ops->reset) | |
584 | ops->reset(qdisc); | |
67305ebc | 585 | |
bbd8a0d3 KK |
586 | if (qdisc->gso_skb) { |
587 | kfree_skb(qdisc->gso_skb); | |
588 | qdisc->gso_skb = NULL; | |
589 | qdisc->q.qlen = 0; | |
590 | } | |
1da177e4 | 591 | } |
62e3ba1b | 592 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 593 | |
1e0d5a57 | 594 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 595 | { |
8a34c5dc DM |
596 | const struct Qdisc_ops *ops = qdisc->ops; |
597 | ||
1e0d5a57 DM |
598 | if (qdisc->flags & TCQ_F_BUILTIN || |
599 | !atomic_dec_and_test(&qdisc->refcnt)) | |
600 | return; | |
601 | ||
3a682fbd | 602 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
603 | qdisc_list_del(qdisc); |
604 | ||
175f9c1b | 605 | qdisc_put_stab(qdisc->stab); |
3a682fbd | 606 | #endif |
8a34c5dc DM |
607 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
608 | if (ops->reset) | |
609 | ops->reset(qdisc); | |
610 | if (ops->destroy) | |
611 | ops->destroy(qdisc); | |
612 | ||
613 | module_put(ops->owner); | |
614 | dev_put(qdisc_dev(qdisc)); | |
615 | ||
554794de | 616 | kfree_skb(qdisc->gso_skb); |
1da177e4 LT |
617 | kfree((char *) qdisc - qdisc->padded); |
618 | } | |
62e3ba1b | 619 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 620 | |
589983cd PM |
621 | /* Attach toplevel qdisc to device queue. */ |
622 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
623 | struct Qdisc *qdisc) | |
624 | { | |
625 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
626 | spinlock_t *root_lock; | |
627 | ||
628 | root_lock = qdisc_lock(oqdisc); | |
629 | spin_lock_bh(root_lock); | |
630 | ||
631 | /* Prune old scheduler */ | |
632 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
633 | qdisc_reset(oqdisc); | |
634 | ||
635 | /* ... and graft new one */ | |
636 | if (qdisc == NULL) | |
637 | qdisc = &noop_qdisc; | |
638 | dev_queue->qdisc_sleeping = qdisc; | |
639 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
640 | ||
641 | spin_unlock_bh(root_lock); | |
642 | ||
643 | return oqdisc; | |
644 | } | |
645 | ||
e8a0464c DM |
646 | static void attach_one_default_qdisc(struct net_device *dev, |
647 | struct netdev_queue *dev_queue, | |
648 | void *_unused) | |
649 | { | |
650 | struct Qdisc *qdisc; | |
651 | ||
652 | if (dev->tx_queue_len) { | |
653 | qdisc = qdisc_create_dflt(dev, dev_queue, | |
d3678b46 | 654 | &pfifo_fast_ops, TC_H_ROOT); |
e8a0464c DM |
655 | if (!qdisc) { |
656 | printk(KERN_INFO "%s: activation failed\n", dev->name); | |
657 | return; | |
658 | } | |
bbd8a0d3 KK |
659 | |
660 | /* Can by-pass the queue discipline for default qdisc */ | |
661 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
e8a0464c DM |
662 | } else { |
663 | qdisc = &noqueue_qdisc; | |
664 | } | |
665 | dev_queue->qdisc_sleeping = qdisc; | |
666 | } | |
667 | ||
6ec1c69a DM |
668 | static void attach_default_qdiscs(struct net_device *dev) |
669 | { | |
670 | struct netdev_queue *txq; | |
671 | struct Qdisc *qdisc; | |
672 | ||
673 | txq = netdev_get_tx_queue(dev, 0); | |
674 | ||
675 | if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) { | |
676 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); | |
677 | dev->qdisc = txq->qdisc_sleeping; | |
678 | atomic_inc(&dev->qdisc->refcnt); | |
679 | } else { | |
680 | qdisc = qdisc_create_dflt(dev, txq, &mq_qdisc_ops, TC_H_ROOT); | |
681 | if (qdisc) { | |
682 | qdisc->ops->attach(qdisc); | |
683 | dev->qdisc = qdisc; | |
684 | } | |
685 | } | |
686 | } | |
687 | ||
e8a0464c DM |
688 | static void transition_one_qdisc(struct net_device *dev, |
689 | struct netdev_queue *dev_queue, | |
690 | void *_need_watchdog) | |
691 | { | |
83874000 | 692 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
693 | int *need_watchdog_p = _need_watchdog; |
694 | ||
a9312ae8 DM |
695 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
696 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
697 | ||
83874000 | 698 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
9d21493b ED |
699 | if (need_watchdog_p && new_qdisc != &noqueue_qdisc) { |
700 | dev_queue->trans_start = 0; | |
e8a0464c | 701 | *need_watchdog_p = 1; |
9d21493b | 702 | } |
e8a0464c DM |
703 | } |
704 | ||
1da177e4 LT |
705 | void dev_activate(struct net_device *dev) |
706 | { | |
e8a0464c | 707 | int need_watchdog; |
b0e1e646 | 708 | |
1da177e4 | 709 | /* No queueing discipline is attached to device; |
d3678b46 DM |
710 | create default one i.e. pfifo_fast for devices, |
711 | which need queueing and noqueue_qdisc for | |
712 | virtual interfaces | |
1da177e4 LT |
713 | */ |
714 | ||
6ec1c69a DM |
715 | if (dev->qdisc == &noop_qdisc) |
716 | attach_default_qdiscs(dev); | |
af356afa | 717 | |
cacaddf5 TC |
718 | if (!netif_carrier_ok(dev)) |
719 | /* Delay activation until next carrier-on event */ | |
720 | return; | |
721 | ||
e8a0464c DM |
722 | need_watchdog = 0; |
723 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
8d50b53d | 724 | transition_one_qdisc(dev, &dev->rx_queue, NULL); |
e8a0464c DM |
725 | |
726 | if (need_watchdog) { | |
1da177e4 LT |
727 | dev->trans_start = jiffies; |
728 | dev_watchdog_up(dev); | |
729 | } | |
b0e1e646 DM |
730 | } |
731 | ||
e8a0464c DM |
732 | static void dev_deactivate_queue(struct net_device *dev, |
733 | struct netdev_queue *dev_queue, | |
734 | void *_qdisc_default) | |
b0e1e646 | 735 | { |
e8a0464c | 736 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 737 | struct Qdisc *qdisc; |
970565bb | 738 | |
970565bb | 739 | qdisc = dev_queue->qdisc; |
b0e1e646 | 740 | if (qdisc) { |
83874000 DM |
741 | spin_lock_bh(qdisc_lock(qdisc)); |
742 | ||
a9312ae8 DM |
743 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
744 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
745 | ||
f7a54c13 | 746 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 747 | qdisc_reset(qdisc); |
d3b753db | 748 | |
83874000 | 749 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 750 | } |
1da177e4 LT |
751 | } |
752 | ||
4335cd2d | 753 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
754 | { |
755 | unsigned int i; | |
756 | ||
757 | for (i = 0; i < dev->num_tx_queues; i++) { | |
758 | struct netdev_queue *dev_queue; | |
7698b4fc | 759 | spinlock_t *root_lock; |
e2627c8c | 760 | struct Qdisc *q; |
e8a0464c DM |
761 | int val; |
762 | ||
763 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 764 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 765 | root_lock = qdisc_lock(q); |
e8a0464c | 766 | |
4335cd2d | 767 | spin_lock_bh(root_lock); |
e8a0464c | 768 | |
b9a3b110 DM |
769 | val = (test_bit(__QDISC_STATE_RUNNING, &q->state) || |
770 | test_bit(__QDISC_STATE_SCHED, &q->state)); | |
e8a0464c | 771 | |
4335cd2d | 772 | spin_unlock_bh(root_lock); |
e8a0464c DM |
773 | |
774 | if (val) | |
775 | return true; | |
776 | } | |
777 | return false; | |
778 | } | |
779 | ||
1da177e4 LT |
780 | void dev_deactivate(struct net_device *dev) |
781 | { | |
e8a0464c | 782 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, &noop_qdisc); |
8d50b53d | 783 | dev_deactivate_queue(dev, &dev->rx_queue, &noop_qdisc); |
41a23b07 | 784 | |
1da177e4 LT |
785 | dev_watchdog_down(dev); |
786 | ||
ce0e32e6 | 787 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. */ |
d4828d85 | 788 | synchronize_rcu(); |
1da177e4 | 789 | |
d4828d85 | 790 | /* Wait for outstanding qdisc_run calls. */ |
4335cd2d DM |
791 | while (some_qdisc_is_busy(dev)) |
792 | yield(); | |
1da177e4 LT |
793 | } |
794 | ||
b0e1e646 DM |
795 | static void dev_init_scheduler_queue(struct net_device *dev, |
796 | struct netdev_queue *dev_queue, | |
e8a0464c | 797 | void *_qdisc) |
b0e1e646 | 798 | { |
e8a0464c DM |
799 | struct Qdisc *qdisc = _qdisc; |
800 | ||
b0e1e646 DM |
801 | dev_queue->qdisc = qdisc; |
802 | dev_queue->qdisc_sleeping = qdisc; | |
b0e1e646 DM |
803 | } |
804 | ||
1da177e4 LT |
805 | void dev_init_scheduler(struct net_device *dev) |
806 | { | |
af356afa | 807 | dev->qdisc = &noop_qdisc; |
e8a0464c | 808 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
8d50b53d | 809 | dev_init_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
1da177e4 | 810 | |
b24b8a24 | 811 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
812 | } |
813 | ||
e8a0464c DM |
814 | static void shutdown_scheduler_queue(struct net_device *dev, |
815 | struct netdev_queue *dev_queue, | |
816 | void *_qdisc_default) | |
1da177e4 | 817 | { |
b0e1e646 | 818 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 819 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
820 | |
821 | if (qdisc) { | |
f7a54c13 | 822 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 823 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 824 | |
1da177e4 | 825 | qdisc_destroy(qdisc); |
10297b99 | 826 | } |
b0e1e646 DM |
827 | } |
828 | ||
829 | void dev_shutdown(struct net_device *dev) | |
830 | { | |
e8a0464c | 831 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
8d50b53d | 832 | shutdown_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
af356afa PM |
833 | qdisc_destroy(dev->qdisc); |
834 | dev->qdisc = &noop_qdisc; | |
835 | ||
547b792c | 836 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 837 | } |