]>
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 | */ | |
d6d9ca0f | 99 | __this_cpu_inc(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 | |
bc135b23 | 208 | qdisc_run_end(q); |
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 | |
06c4648d IC |
330 | /** |
331 | * netif_notify_peers - notify network peers about existence of @dev | |
332 | * @dev: network device | |
333 | * | |
334 | * Generate traffic such that interested network peers are aware of | |
335 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
336 | * a device wants to inform the rest of the network about some sort of | |
337 | * reconfiguration such as a failover event or virtual machine | |
338 | * migration. | |
339 | */ | |
340 | void netif_notify_peers(struct net_device *dev) | |
341 | { | |
342 | rtnl_lock(); | |
343 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
344 | rtnl_unlock(); | |
345 | } | |
346 | EXPORT_SYMBOL(netif_notify_peers); | |
347 | ||
1da177e4 LT |
348 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
349 | under all circumstances. It is difficult to invent anything faster or | |
350 | cheaper. | |
351 | */ | |
352 | ||
94df109a | 353 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc) |
1da177e4 LT |
354 | { |
355 | kfree_skb(skb); | |
356 | return NET_XMIT_CN; | |
357 | } | |
358 | ||
94df109a | 359 | static struct sk_buff *noop_dequeue(struct Qdisc * qdisc) |
1da177e4 LT |
360 | { |
361 | return NULL; | |
362 | } | |
363 | ||
20fea08b | 364 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
365 | .id = "noop", |
366 | .priv_size = 0, | |
367 | .enqueue = noop_enqueue, | |
368 | .dequeue = noop_dequeue, | |
99c0db26 | 369 | .peek = noop_dequeue, |
1da177e4 LT |
370 | .owner = THIS_MODULE, |
371 | }; | |
372 | ||
7698b4fc | 373 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 374 | .qdisc = &noop_qdisc, |
9f3ffae0 | 375 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
376 | }; |
377 | ||
1da177e4 LT |
378 | struct Qdisc noop_qdisc = { |
379 | .enqueue = noop_enqueue, | |
380 | .dequeue = noop_dequeue, | |
381 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 382 | .ops = &noop_qdisc_ops, |
1da177e4 | 383 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 384 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 385 | .dev_queue = &noop_netdev_queue, |
1da177e4 | 386 | }; |
62e3ba1b | 387 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 388 | |
20fea08b | 389 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
390 | .id = "noqueue", |
391 | .priv_size = 0, | |
392 | .enqueue = noop_enqueue, | |
393 | .dequeue = noop_dequeue, | |
99c0db26 | 394 | .peek = noop_dequeue, |
1da177e4 LT |
395 | .owner = THIS_MODULE, |
396 | }; | |
397 | ||
30ee42be DM |
398 | static struct Qdisc noqueue_qdisc; |
399 | static struct netdev_queue noqueue_netdev_queue = { | |
400 | .qdisc = &noqueue_qdisc, | |
9f3ffae0 | 401 | .qdisc_sleeping = &noqueue_qdisc, |
30ee42be DM |
402 | }; |
403 | ||
1da177e4 LT |
404 | static struct Qdisc noqueue_qdisc = { |
405 | .enqueue = NULL, | |
406 | .dequeue = noop_dequeue, | |
407 | .flags = TCQ_F_BUILTIN, | |
408 | .ops = &noqueue_qdisc_ops, | |
409 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
30ee42be DM |
410 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
411 | .dev_queue = &noqueue_netdev_queue, | |
1da177e4 LT |
412 | }; |
413 | ||
414 | ||
d3678b46 DM |
415 | static const u8 prio2band[TC_PRIO_MAX+1] = |
416 | { 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 }; | |
417 | ||
418 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
419 | generic prio+fifo combination. | |
420 | */ | |
421 | ||
422 | #define PFIFO_FAST_BANDS 3 | |
423 | ||
fd3ae5e8 KK |
424 | /* |
425 | * Private data for a pfifo_fast scheduler containing: | |
426 | * - queues for the three band | |
427 | * - bitmap indicating which of the bands contain skbs | |
428 | */ | |
429 | struct pfifo_fast_priv { | |
430 | u32 bitmap; | |
431 | struct sk_buff_head q[PFIFO_FAST_BANDS]; | |
432 | }; | |
433 | ||
434 | /* | |
435 | * Convert a bitmap to the first band number where an skb is queued, where: | |
436 | * bitmap=0 means there are no skbs on any band. | |
437 | * bitmap=1 means there is an skb on band 0. | |
438 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
439 | */ | |
440 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
441 | ||
442 | static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv, | |
443 | int band) | |
d3678b46 | 444 | { |
fd3ae5e8 | 445 | return priv->q + band; |
d3678b46 DM |
446 | } |
447 | ||
448 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc* qdisc) | |
321090e7 | 449 | { |
a453e068 KK |
450 | if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) { |
451 | int band = prio2band[skb->priority & TC_PRIO_MAX]; | |
452 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
453 | struct sk_buff_head *list = band2list(priv, band); | |
1da177e4 | 454 | |
fd3ae5e8 | 455 | priv->bitmap |= (1 << band); |
d3678b46 | 456 | qdisc->q.qlen++; |
821d24ae | 457 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 458 | } |
821d24ae TG |
459 | |
460 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
461 | } |
462 | ||
d3678b46 | 463 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc* qdisc) |
1da177e4 | 464 | { |
fd3ae5e8 KK |
465 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
466 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 467 | |
fd3ae5e8 KK |
468 | if (likely(band >= 0)) { |
469 | struct sk_buff_head *list = band2list(priv, band); | |
470 | struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list); | |
471 | ||
472 | qdisc->q.qlen--; | |
473 | if (skb_queue_empty(list)) | |
474 | priv->bitmap &= ~(1 << band); | |
475 | ||
476 | return skb; | |
d3678b46 | 477 | } |
f87a9c3d | 478 | |
1da177e4 LT |
479 | return NULL; |
480 | } | |
481 | ||
99c0db26 JP |
482 | static struct sk_buff *pfifo_fast_peek(struct Qdisc* qdisc) |
483 | { | |
fd3ae5e8 KK |
484 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
485 | int band = bitmap2band[priv->bitmap]; | |
486 | ||
487 | if (band >= 0) { | |
488 | struct sk_buff_head *list = band2list(priv, band); | |
99c0db26 | 489 | |
fd3ae5e8 | 490 | return skb_peek(list); |
99c0db26 JP |
491 | } |
492 | ||
493 | return NULL; | |
494 | } | |
495 | ||
d3678b46 | 496 | static void pfifo_fast_reset(struct Qdisc* qdisc) |
1da177e4 | 497 | { |
d3678b46 | 498 | int prio; |
fd3ae5e8 | 499 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
500 | |
501 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 502 | __qdisc_reset_queue(qdisc, band2list(priv, prio)); |
d3678b46 | 503 | |
fd3ae5e8 | 504 | priv->bitmap = 0; |
821d24ae | 505 | qdisc->qstats.backlog = 0; |
d3678b46 | 506 | qdisc->q.qlen = 0; |
1da177e4 LT |
507 | } |
508 | ||
d3678b46 DM |
509 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
510 | { | |
511 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
512 | ||
513 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX+1); | |
514 | NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt); | |
515 | return skb->len; | |
516 | ||
517 | nla_put_failure: | |
518 | return -1; | |
519 | } | |
520 | ||
521 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
522 | { | |
523 | int prio; | |
fd3ae5e8 | 524 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
525 | |
526 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 527 | skb_queue_head_init(band2list(priv, prio)); |
d3678b46 DM |
528 | |
529 | return 0; | |
530 | } | |
531 | ||
6ec1c69a | 532 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 533 | .id = "pfifo_fast", |
fd3ae5e8 | 534 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
535 | .enqueue = pfifo_fast_enqueue, |
536 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 537 | .peek = pfifo_fast_peek, |
d3678b46 DM |
538 | .init = pfifo_fast_init, |
539 | .reset = pfifo_fast_reset, | |
540 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
541 | .owner = THIS_MODULE, |
542 | }; | |
543 | ||
5ce2d488 | 544 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
bb949fbd | 545 | struct Qdisc_ops *ops) |
1da177e4 LT |
546 | { |
547 | void *p; | |
548 | struct Qdisc *sch; | |
3d54b82f TG |
549 | unsigned int size; |
550 | int err = -ENOBUFS; | |
1da177e4 | 551 | |
5d944c64 | 552 | /* ensure that the Qdisc and the private data are 64-byte aligned */ |
3d54b82f TG |
553 | size = QDISC_ALIGN(sizeof(*sch)); |
554 | size += ops->priv_size + (QDISC_ALIGNTO - 1); | |
1da177e4 | 555 | |
0da974f4 | 556 | p = kzalloc(size, GFP_KERNEL); |
1da177e4 | 557 | if (!p) |
3d54b82f | 558 | goto errout; |
3d54b82f TG |
559 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
560 | sch->padded = (char *) sch - (char *) p; | |
1da177e4 LT |
561 | |
562 | INIT_LIST_HEAD(&sch->list); | |
563 | skb_queue_head_init(&sch->q); | |
79640a4c | 564 | spin_lock_init(&sch->busylock); |
1da177e4 LT |
565 | sch->ops = ops; |
566 | sch->enqueue = ops->enqueue; | |
567 | sch->dequeue = ops->dequeue; | |
bb949fbd | 568 | sch->dev_queue = dev_queue; |
5ce2d488 | 569 | dev_hold(qdisc_dev(sch)); |
1da177e4 | 570 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
571 | |
572 | return sch; | |
573 | errout: | |
01e123d7 | 574 | return ERR_PTR(err); |
3d54b82f TG |
575 | } |
576 | ||
bb949fbd DM |
577 | struct Qdisc * qdisc_create_dflt(struct net_device *dev, |
578 | struct netdev_queue *dev_queue, | |
579 | struct Qdisc_ops *ops, | |
9f9afec4 | 580 | unsigned int parentid) |
3d54b82f TG |
581 | { |
582 | struct Qdisc *sch; | |
10297b99 | 583 | |
5ce2d488 | 584 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
585 | if (IS_ERR(sch)) |
586 | goto errout; | |
9f9afec4 | 587 | sch->parent = parentid; |
3d54b82f | 588 | |
1da177e4 LT |
589 | if (!ops->init || ops->init(sch, NULL) == 0) |
590 | return sch; | |
591 | ||
0fbbeb1b | 592 | qdisc_destroy(sch); |
3d54b82f | 593 | errout: |
1da177e4 LT |
594 | return NULL; |
595 | } | |
62e3ba1b | 596 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 597 | |
5fb66229 | 598 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
599 | |
600 | void qdisc_reset(struct Qdisc *qdisc) | |
601 | { | |
20fea08b | 602 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
603 | |
604 | if (ops->reset) | |
605 | ops->reset(qdisc); | |
67305ebc | 606 | |
bbd8a0d3 KK |
607 | if (qdisc->gso_skb) { |
608 | kfree_skb(qdisc->gso_skb); | |
609 | qdisc->gso_skb = NULL; | |
610 | qdisc->q.qlen = 0; | |
611 | } | |
1da177e4 | 612 | } |
62e3ba1b | 613 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 614 | |
5d944c64 ED |
615 | static void qdisc_rcu_free(struct rcu_head *head) |
616 | { | |
617 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
618 | ||
619 | kfree((char *) qdisc - qdisc->padded); | |
620 | } | |
621 | ||
1e0d5a57 | 622 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 623 | { |
8a34c5dc DM |
624 | const struct Qdisc_ops *ops = qdisc->ops; |
625 | ||
1e0d5a57 DM |
626 | if (qdisc->flags & TCQ_F_BUILTIN || |
627 | !atomic_dec_and_test(&qdisc->refcnt)) | |
628 | return; | |
629 | ||
3a682fbd | 630 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
631 | qdisc_list_del(qdisc); |
632 | ||
175f9c1b | 633 | qdisc_put_stab(qdisc->stab); |
3a682fbd | 634 | #endif |
8a34c5dc DM |
635 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
636 | if (ops->reset) | |
637 | ops->reset(qdisc); | |
638 | if (ops->destroy) | |
639 | ops->destroy(qdisc); | |
640 | ||
641 | module_put(ops->owner); | |
642 | dev_put(qdisc_dev(qdisc)); | |
643 | ||
554794de | 644 | kfree_skb(qdisc->gso_skb); |
5d944c64 ED |
645 | /* |
646 | * gen_estimator est_timer() might access qdisc->q.lock, | |
647 | * wait a RCU grace period before freeing qdisc. | |
648 | */ | |
649 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 650 | } |
62e3ba1b | 651 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 652 | |
589983cd PM |
653 | /* Attach toplevel qdisc to device queue. */ |
654 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
655 | struct Qdisc *qdisc) | |
656 | { | |
657 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
658 | spinlock_t *root_lock; | |
659 | ||
660 | root_lock = qdisc_lock(oqdisc); | |
661 | spin_lock_bh(root_lock); | |
662 | ||
663 | /* Prune old scheduler */ | |
664 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
665 | qdisc_reset(oqdisc); | |
666 | ||
667 | /* ... and graft new one */ | |
668 | if (qdisc == NULL) | |
669 | qdisc = &noop_qdisc; | |
670 | dev_queue->qdisc_sleeping = qdisc; | |
671 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
672 | ||
673 | spin_unlock_bh(root_lock); | |
674 | ||
675 | return oqdisc; | |
676 | } | |
677 | ||
e8a0464c DM |
678 | static void attach_one_default_qdisc(struct net_device *dev, |
679 | struct netdev_queue *dev_queue, | |
680 | void *_unused) | |
681 | { | |
682 | struct Qdisc *qdisc; | |
683 | ||
684 | if (dev->tx_queue_len) { | |
685 | qdisc = qdisc_create_dflt(dev, dev_queue, | |
d3678b46 | 686 | &pfifo_fast_ops, TC_H_ROOT); |
e8a0464c DM |
687 | if (!qdisc) { |
688 | printk(KERN_INFO "%s: activation failed\n", dev->name); | |
689 | return; | |
690 | } | |
bbd8a0d3 KK |
691 | |
692 | /* Can by-pass the queue discipline for default qdisc */ | |
693 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
e8a0464c DM |
694 | } else { |
695 | qdisc = &noqueue_qdisc; | |
696 | } | |
697 | dev_queue->qdisc_sleeping = qdisc; | |
698 | } | |
699 | ||
6ec1c69a DM |
700 | static void attach_default_qdiscs(struct net_device *dev) |
701 | { | |
702 | struct netdev_queue *txq; | |
703 | struct Qdisc *qdisc; | |
704 | ||
705 | txq = netdev_get_tx_queue(dev, 0); | |
706 | ||
707 | if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) { | |
708 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); | |
709 | dev->qdisc = txq->qdisc_sleeping; | |
710 | atomic_inc(&dev->qdisc->refcnt); | |
711 | } else { | |
712 | qdisc = qdisc_create_dflt(dev, txq, &mq_qdisc_ops, TC_H_ROOT); | |
713 | if (qdisc) { | |
714 | qdisc->ops->attach(qdisc); | |
715 | dev->qdisc = qdisc; | |
716 | } | |
717 | } | |
718 | } | |
719 | ||
e8a0464c DM |
720 | static void transition_one_qdisc(struct net_device *dev, |
721 | struct netdev_queue *dev_queue, | |
722 | void *_need_watchdog) | |
723 | { | |
83874000 | 724 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
725 | int *need_watchdog_p = _need_watchdog; |
726 | ||
a9312ae8 DM |
727 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
728 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
729 | ||
83874000 | 730 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
9d21493b ED |
731 | if (need_watchdog_p && new_qdisc != &noqueue_qdisc) { |
732 | dev_queue->trans_start = 0; | |
e8a0464c | 733 | *need_watchdog_p = 1; |
9d21493b | 734 | } |
e8a0464c DM |
735 | } |
736 | ||
1da177e4 LT |
737 | void dev_activate(struct net_device *dev) |
738 | { | |
e8a0464c | 739 | int need_watchdog; |
b0e1e646 | 740 | |
1da177e4 | 741 | /* No queueing discipline is attached to device; |
d3678b46 DM |
742 | create default one i.e. pfifo_fast for devices, |
743 | which need queueing and noqueue_qdisc for | |
744 | virtual interfaces | |
1da177e4 LT |
745 | */ |
746 | ||
6ec1c69a DM |
747 | if (dev->qdisc == &noop_qdisc) |
748 | attach_default_qdiscs(dev); | |
af356afa | 749 | |
cacaddf5 TC |
750 | if (!netif_carrier_ok(dev)) |
751 | /* Delay activation until next carrier-on event */ | |
752 | return; | |
753 | ||
e8a0464c DM |
754 | need_watchdog = 0; |
755 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
8d50b53d | 756 | transition_one_qdisc(dev, &dev->rx_queue, NULL); |
e8a0464c DM |
757 | |
758 | if (need_watchdog) { | |
1da177e4 LT |
759 | dev->trans_start = jiffies; |
760 | dev_watchdog_up(dev); | |
761 | } | |
b0e1e646 DM |
762 | } |
763 | ||
e8a0464c DM |
764 | static void dev_deactivate_queue(struct net_device *dev, |
765 | struct netdev_queue *dev_queue, | |
766 | void *_qdisc_default) | |
b0e1e646 | 767 | { |
e8a0464c | 768 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 769 | struct Qdisc *qdisc; |
970565bb | 770 | |
970565bb | 771 | qdisc = dev_queue->qdisc; |
b0e1e646 | 772 | if (qdisc) { |
83874000 DM |
773 | spin_lock_bh(qdisc_lock(qdisc)); |
774 | ||
a9312ae8 DM |
775 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
776 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
777 | ||
f7a54c13 | 778 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 779 | qdisc_reset(qdisc); |
d3b753db | 780 | |
83874000 | 781 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 782 | } |
1da177e4 LT |
783 | } |
784 | ||
4335cd2d | 785 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
786 | { |
787 | unsigned int i; | |
788 | ||
789 | for (i = 0; i < dev->num_tx_queues; i++) { | |
790 | struct netdev_queue *dev_queue; | |
7698b4fc | 791 | spinlock_t *root_lock; |
e2627c8c | 792 | struct Qdisc *q; |
e8a0464c DM |
793 | int val; |
794 | ||
795 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 796 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 797 | root_lock = qdisc_lock(q); |
e8a0464c | 798 | |
4335cd2d | 799 | spin_lock_bh(root_lock); |
e8a0464c | 800 | |
bc135b23 | 801 | val = (qdisc_is_running(q) || |
b9a3b110 | 802 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 803 | |
4335cd2d | 804 | spin_unlock_bh(root_lock); |
e8a0464c DM |
805 | |
806 | if (val) | |
807 | return true; | |
808 | } | |
809 | return false; | |
810 | } | |
811 | ||
1da177e4 LT |
812 | void dev_deactivate(struct net_device *dev) |
813 | { | |
e8a0464c | 814 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, &noop_qdisc); |
8d50b53d | 815 | dev_deactivate_queue(dev, &dev->rx_queue, &noop_qdisc); |
41a23b07 | 816 | |
1da177e4 LT |
817 | dev_watchdog_down(dev); |
818 | ||
ce0e32e6 | 819 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. */ |
d4828d85 | 820 | synchronize_rcu(); |
1da177e4 | 821 | |
d4828d85 | 822 | /* Wait for outstanding qdisc_run calls. */ |
4335cd2d DM |
823 | while (some_qdisc_is_busy(dev)) |
824 | yield(); | |
1da177e4 LT |
825 | } |
826 | ||
b0e1e646 DM |
827 | static void dev_init_scheduler_queue(struct net_device *dev, |
828 | struct netdev_queue *dev_queue, | |
e8a0464c | 829 | void *_qdisc) |
b0e1e646 | 830 | { |
e8a0464c DM |
831 | struct Qdisc *qdisc = _qdisc; |
832 | ||
b0e1e646 DM |
833 | dev_queue->qdisc = qdisc; |
834 | dev_queue->qdisc_sleeping = qdisc; | |
b0e1e646 DM |
835 | } |
836 | ||
1da177e4 LT |
837 | void dev_init_scheduler(struct net_device *dev) |
838 | { | |
af356afa | 839 | dev->qdisc = &noop_qdisc; |
e8a0464c | 840 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
8d50b53d | 841 | dev_init_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
1da177e4 | 842 | |
b24b8a24 | 843 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
844 | } |
845 | ||
e8a0464c DM |
846 | static void shutdown_scheduler_queue(struct net_device *dev, |
847 | struct netdev_queue *dev_queue, | |
848 | void *_qdisc_default) | |
1da177e4 | 849 | { |
b0e1e646 | 850 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 851 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
852 | |
853 | if (qdisc) { | |
f7a54c13 | 854 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 855 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 856 | |
1da177e4 | 857 | qdisc_destroy(qdisc); |
10297b99 | 858 | } |
b0e1e646 DM |
859 | } |
860 | ||
861 | void dev_shutdown(struct net_device *dev) | |
862 | { | |
e8a0464c | 863 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
8d50b53d | 864 | shutdown_scheduler_queue(dev, &dev->rx_queue, &noop_qdisc); |
af356afa PM |
865 | qdisc_destroy(dev->qdisc); |
866 | dev->qdisc = &noop_qdisc; | |
867 | ||
547b792c | 868 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 869 | } |