]>
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)); | |
5a0d2268 | 63 | if (!netif_tx_queue_frozen_or_stopped(txq)) { |
6252352d | 64 | q->gso_skb = NULL; |
bbd8a0d3 KK |
65 | q->q.qlen--; |
66 | } else | |
ebf05982 JP |
67 | skb = NULL; |
68 | } else { | |
c716a81a | 69 | skb = q->dequeue(q); |
ebf05982 | 70 | } |
c716a81a JHS |
71 | |
72 | return skb; | |
73 | } | |
74 | ||
6c1361a6 | 75 | static inline int handle_dev_cpu_collision(struct sk_buff *skb, |
970565bb | 76 | struct netdev_queue *dev_queue, |
6c1361a6 | 77 | struct Qdisc *q) |
c716a81a | 78 | { |
6c1361a6 | 79 | int ret; |
c716a81a | 80 | |
c773e847 | 81 | if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) { |
6c1361a6 KK |
82 | /* |
83 | * Same CPU holding the lock. It may be a transient | |
84 | * configuration error, when hard_start_xmit() recurses. We | |
85 | * detect it by checking xmit owner and drop the packet when | |
86 | * deadloop is detected. Return OK to try the next skb. | |
87 | */ | |
c716a81a | 88 | kfree_skb(skb); |
6c1361a6 | 89 | if (net_ratelimit()) |
cc7ec456 ED |
90 | pr_warning("Dead loop on netdevice %s, fix it urgently!\n", |
91 | dev_queue->dev->name); | |
6c1361a6 KK |
92 | ret = qdisc_qlen(q); |
93 | } else { | |
94 | /* | |
95 | * Another cpu is holding lock, requeue & delay xmits for | |
96 | * some time. | |
97 | */ | |
d6d9ca0f | 98 | __this_cpu_inc(softnet_data.cpu_collision); |
37437bb2 | 99 | ret = dev_requeue_skb(skb, q); |
c716a81a JHS |
100 | } |
101 | ||
6c1361a6 | 102 | return ret; |
c716a81a JHS |
103 | } |
104 | ||
10297b99 | 105 | /* |
bbd8a0d3 KK |
106 | * Transmit one skb, and handle the return status as required. Holding the |
107 | * __QDISC_STATE_RUNNING bit guarantees that only one CPU can execute this | |
108 | * function. | |
6c1361a6 KK |
109 | * |
110 | * Returns to the caller: | |
111 | * 0 - queue is empty or throttled. | |
112 | * >0 - queue is not empty. | |
6c1361a6 | 113 | */ |
bbd8a0d3 KK |
114 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
115 | struct net_device *dev, struct netdev_queue *txq, | |
116 | spinlock_t *root_lock) | |
1da177e4 | 117 | { |
5f1a485d | 118 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
119 | |
120 | /* And release qdisc */ | |
121 | spin_unlock(root_lock); | |
c716a81a | 122 | |
c773e847 | 123 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
5a0d2268 | 124 | if (!netif_tx_queue_frozen_or_stopped(txq)) |
fd2ea0a7 | 125 | ret = dev_hard_start_xmit(skb, dev, txq); |
572a9d7b | 126 | |
c773e847 | 127 | HARD_TX_UNLOCK(dev, txq); |
c716a81a | 128 | |
7698b4fc | 129 | spin_lock(root_lock); |
c716a81a | 130 | |
9a1654ba JP |
131 | if (dev_xmit_complete(ret)) { |
132 | /* Driver sent out skb successfully or skb was consumed */ | |
6c1361a6 | 133 | ret = qdisc_qlen(q); |
9a1654ba | 134 | } else if (ret == NETDEV_TX_LOCKED) { |
6c1361a6 | 135 | /* Driver try lock failed */ |
eb6aafe3 | 136 | ret = handle_dev_cpu_collision(skb, txq, q); |
9a1654ba | 137 | } else { |
6c1361a6 KK |
138 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
139 | if (unlikely (ret != NETDEV_TX_BUSY && net_ratelimit())) | |
cc7ec456 ED |
140 | pr_warning("BUG %s code %d qlen %d\n", |
141 | dev->name, ret, q->q.qlen); | |
6c1361a6 | 142 | |
37437bb2 | 143 | ret = dev_requeue_skb(skb, q); |
6c1361a6 | 144 | } |
c716a81a | 145 | |
5a0d2268 | 146 | if (ret && netif_tx_queue_frozen_or_stopped(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; | |
7fee226a | 182 | WARN_ON_ONCE(skb_dst_is_noref(skb)); |
bbd8a0d3 KK |
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 | { |
d5b8aa1d | 192 | int quota = weight_p; |
2ba2506c | 193 | |
37437bb2 | 194 | while (qdisc_restart(q)) { |
2ba2506c | 195 | /* |
d5b8aa1d | 196 | * Ordered by possible occurrence: Postpone processing if |
197 | * 1. we've exceeded packet quota | |
198 | * 2. another process needs the CPU; | |
2ba2506c | 199 | */ |
f0c50c7c | 200 | if (--quota <= 0 || need_resched()) { |
37437bb2 | 201 | __netif_schedule(q); |
d90df3ad | 202 | break; |
2ba2506c HX |
203 | } |
204 | } | |
48d83325 | 205 | |
bc135b23 | 206 | qdisc_run_end(q); |
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) { |
9d21493b | 254 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 255 | dev->name, netdev_drivername(dev), i); |
d314774c | 256 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 257 | } |
e8a0464c DM |
258 | if (!mod_timer(&dev->watchdog_timer, |
259 | round_jiffies(jiffies + | |
260 | dev->watchdog_timeo))) | |
1da177e4 LT |
261 | dev_hold(dev); |
262 | } | |
263 | } | |
932ff279 | 264 | netif_tx_unlock(dev); |
1da177e4 LT |
265 | |
266 | dev_put(dev); | |
267 | } | |
268 | ||
1da177e4 LT |
269 | void __netdev_watchdog_up(struct net_device *dev) |
270 | { | |
d314774c | 271 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
272 | if (dev->watchdog_timeo <= 0) |
273 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
274 | if (!mod_timer(&dev->watchdog_timer, |
275 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
276 | dev_hold(dev); |
277 | } | |
278 | } | |
279 | ||
280 | static void dev_watchdog_up(struct net_device *dev) | |
281 | { | |
1da177e4 | 282 | __netdev_watchdog_up(dev); |
1da177e4 LT |
283 | } |
284 | ||
285 | static void dev_watchdog_down(struct net_device *dev) | |
286 | { | |
932ff279 | 287 | netif_tx_lock_bh(dev); |
1da177e4 | 288 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 289 | dev_put(dev); |
932ff279 | 290 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
291 | } |
292 | ||
bea3348e SH |
293 | /** |
294 | * netif_carrier_on - set carrier | |
295 | * @dev: network device | |
296 | * | |
297 | * Device has detected that carrier. | |
298 | */ | |
0a242efc DV |
299 | void netif_carrier_on(struct net_device *dev) |
300 | { | |
bfaae0f0 | 301 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
302 | if (dev->reg_state == NETREG_UNINITIALIZED) |
303 | return; | |
0a242efc | 304 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
305 | if (netif_running(dev)) |
306 | __netdev_watchdog_up(dev); | |
307 | } | |
0a242efc | 308 | } |
62e3ba1b | 309 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 310 | |
bea3348e SH |
311 | /** |
312 | * netif_carrier_off - clear carrier | |
313 | * @dev: network device | |
314 | * | |
315 | * Device has detected loss of carrier. | |
316 | */ | |
0a242efc DV |
317 | void netif_carrier_off(struct net_device *dev) |
318 | { | |
b4730016 DM |
319 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
320 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
321 | return; | |
0a242efc | 322 | linkwatch_fire_event(dev); |
b4730016 | 323 | } |
0a242efc | 324 | } |
62e3ba1b | 325 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 326 | |
06c4648d IC |
327 | /** |
328 | * netif_notify_peers - notify network peers about existence of @dev | |
329 | * @dev: network device | |
330 | * | |
331 | * Generate traffic such that interested network peers are aware of | |
332 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
333 | * a device wants to inform the rest of the network about some sort of | |
334 | * reconfiguration such as a failover event or virtual machine | |
335 | * migration. | |
336 | */ | |
337 | void netif_notify_peers(struct net_device *dev) | |
338 | { | |
339 | rtnl_lock(); | |
340 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
341 | rtnl_unlock(); | |
342 | } | |
343 | EXPORT_SYMBOL(netif_notify_peers); | |
344 | ||
1da177e4 LT |
345 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
346 | under all circumstances. It is difficult to invent anything faster or | |
347 | cheaper. | |
348 | */ | |
349 | ||
94df109a | 350 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc) |
1da177e4 LT |
351 | { |
352 | kfree_skb(skb); | |
353 | return NET_XMIT_CN; | |
354 | } | |
355 | ||
94df109a | 356 | static struct sk_buff *noop_dequeue(struct Qdisc * qdisc) |
1da177e4 LT |
357 | { |
358 | return NULL; | |
359 | } | |
360 | ||
20fea08b | 361 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
362 | .id = "noop", |
363 | .priv_size = 0, | |
364 | .enqueue = noop_enqueue, | |
365 | .dequeue = noop_dequeue, | |
99c0db26 | 366 | .peek = noop_dequeue, |
1da177e4 LT |
367 | .owner = THIS_MODULE, |
368 | }; | |
369 | ||
7698b4fc | 370 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 371 | .qdisc = &noop_qdisc, |
9f3ffae0 | 372 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
373 | }; |
374 | ||
1da177e4 LT |
375 | struct Qdisc noop_qdisc = { |
376 | .enqueue = noop_enqueue, | |
377 | .dequeue = noop_dequeue, | |
378 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 379 | .ops = &noop_qdisc_ops, |
1da177e4 | 380 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 381 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 382 | .dev_queue = &noop_netdev_queue, |
7b5edbc4 | 383 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
1da177e4 | 384 | }; |
62e3ba1b | 385 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 386 | |
20fea08b | 387 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
388 | .id = "noqueue", |
389 | .priv_size = 0, | |
390 | .enqueue = noop_enqueue, | |
391 | .dequeue = noop_dequeue, | |
99c0db26 | 392 | .peek = noop_dequeue, |
1da177e4 LT |
393 | .owner = THIS_MODULE, |
394 | }; | |
395 | ||
30ee42be DM |
396 | static struct Qdisc noqueue_qdisc; |
397 | static struct netdev_queue noqueue_netdev_queue = { | |
398 | .qdisc = &noqueue_qdisc, | |
9f3ffae0 | 399 | .qdisc_sleeping = &noqueue_qdisc, |
30ee42be DM |
400 | }; |
401 | ||
1da177e4 LT |
402 | static struct Qdisc noqueue_qdisc = { |
403 | .enqueue = NULL, | |
404 | .dequeue = noop_dequeue, | |
405 | .flags = TCQ_F_BUILTIN, | |
406 | .ops = &noqueue_qdisc_ops, | |
407 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
30ee42be DM |
408 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
409 | .dev_queue = &noqueue_netdev_queue, | |
7b5edbc4 | 410 | .busylock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock), |
1da177e4 LT |
411 | }; |
412 | ||
413 | ||
cc7ec456 ED |
414 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
415 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
416 | }; | |
d3678b46 DM |
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 | ||
cc7ec456 | 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 | ||
cc7ec456 | 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 | ||
cc7ec456 | 482 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 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 | ||
cc7ec456 | 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 | ||
cc7ec456 | 513 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
d3678b46 DM |
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 | 528 | |
23624935 ED |
529 | /* Can by-pass the queue discipline */ |
530 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
531 | return 0; |
532 | } | |
533 | ||
6ec1c69a | 534 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 535 | .id = "pfifo_fast", |
fd3ae5e8 | 536 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
537 | .enqueue = pfifo_fast_enqueue, |
538 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 539 | .peek = pfifo_fast_peek, |
d3678b46 DM |
540 | .init = pfifo_fast_init, |
541 | .reset = pfifo_fast_reset, | |
542 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
543 | .owner = THIS_MODULE, |
544 | }; | |
b8970f0b | 545 | EXPORT_SYMBOL(pfifo_fast_ops); |
1da177e4 | 546 | |
5ce2d488 | 547 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
bb949fbd | 548 | struct Qdisc_ops *ops) |
1da177e4 LT |
549 | { |
550 | void *p; | |
551 | struct Qdisc *sch; | |
d276055c | 552 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 553 | int err = -ENOBUFS; |
1da177e4 | 554 | |
f2cd2d3e ED |
555 | p = kzalloc_node(size, GFP_KERNEL, |
556 | netdev_queue_numa_node_read(dev_queue)); | |
557 | ||
1da177e4 | 558 | if (!p) |
3d54b82f | 559 | goto errout; |
3d54b82f | 560 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
561 | /* if we got non aligned memory, ask more and do alignment ourself */ |
562 | if (sch != p) { | |
563 | kfree(p); | |
564 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
565 | netdev_queue_numa_node_read(dev_queue)); | |
566 | if (!p) | |
567 | goto errout; | |
568 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
569 | sch->padded = (char *) sch - (char *) p; | |
570 | } | |
1da177e4 LT |
571 | INIT_LIST_HEAD(&sch->list); |
572 | skb_queue_head_init(&sch->q); | |
79640a4c | 573 | spin_lock_init(&sch->busylock); |
1da177e4 LT |
574 | sch->ops = ops; |
575 | sch->enqueue = ops->enqueue; | |
576 | sch->dequeue = ops->dequeue; | |
bb949fbd | 577 | sch->dev_queue = dev_queue; |
5ce2d488 | 578 | dev_hold(qdisc_dev(sch)); |
1da177e4 | 579 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
580 | |
581 | return sch; | |
582 | errout: | |
01e123d7 | 583 | return ERR_PTR(err); |
3d54b82f TG |
584 | } |
585 | ||
3511c913 CG |
586 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
587 | struct Qdisc_ops *ops, unsigned int parentid) | |
3d54b82f TG |
588 | { |
589 | struct Qdisc *sch; | |
10297b99 | 590 | |
5ce2d488 | 591 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
592 | if (IS_ERR(sch)) |
593 | goto errout; | |
9f9afec4 | 594 | sch->parent = parentid; |
3d54b82f | 595 | |
1da177e4 LT |
596 | if (!ops->init || ops->init(sch, NULL) == 0) |
597 | return sch; | |
598 | ||
0fbbeb1b | 599 | qdisc_destroy(sch); |
3d54b82f | 600 | errout: |
1da177e4 LT |
601 | return NULL; |
602 | } | |
62e3ba1b | 603 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 604 | |
5fb66229 | 605 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
606 | |
607 | void qdisc_reset(struct Qdisc *qdisc) | |
608 | { | |
20fea08b | 609 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
610 | |
611 | if (ops->reset) | |
612 | ops->reset(qdisc); | |
67305ebc | 613 | |
bbd8a0d3 KK |
614 | if (qdisc->gso_skb) { |
615 | kfree_skb(qdisc->gso_skb); | |
616 | qdisc->gso_skb = NULL; | |
617 | qdisc->q.qlen = 0; | |
618 | } | |
1da177e4 | 619 | } |
62e3ba1b | 620 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 621 | |
5d944c64 ED |
622 | static void qdisc_rcu_free(struct rcu_head *head) |
623 | { | |
624 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
625 | ||
626 | kfree((char *) qdisc - qdisc->padded); | |
627 | } | |
628 | ||
1e0d5a57 | 629 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 630 | { |
8a34c5dc DM |
631 | const struct Qdisc_ops *ops = qdisc->ops; |
632 | ||
1e0d5a57 DM |
633 | if (qdisc->flags & TCQ_F_BUILTIN || |
634 | !atomic_dec_and_test(&qdisc->refcnt)) | |
635 | return; | |
636 | ||
3a682fbd | 637 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
638 | qdisc_list_del(qdisc); |
639 | ||
a2da570d | 640 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 641 | #endif |
8a34c5dc DM |
642 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
643 | if (ops->reset) | |
644 | ops->reset(qdisc); | |
645 | if (ops->destroy) | |
646 | ops->destroy(qdisc); | |
647 | ||
648 | module_put(ops->owner); | |
649 | dev_put(qdisc_dev(qdisc)); | |
650 | ||
554794de | 651 | kfree_skb(qdisc->gso_skb); |
5d944c64 ED |
652 | /* |
653 | * gen_estimator est_timer() might access qdisc->q.lock, | |
654 | * wait a RCU grace period before freeing qdisc. | |
655 | */ | |
656 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 657 | } |
62e3ba1b | 658 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 659 | |
589983cd PM |
660 | /* Attach toplevel qdisc to device queue. */ |
661 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
662 | struct Qdisc *qdisc) | |
663 | { | |
664 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
665 | spinlock_t *root_lock; | |
666 | ||
667 | root_lock = qdisc_lock(oqdisc); | |
668 | spin_lock_bh(root_lock); | |
669 | ||
670 | /* Prune old scheduler */ | |
671 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
672 | qdisc_reset(oqdisc); | |
673 | ||
674 | /* ... and graft new one */ | |
675 | if (qdisc == NULL) | |
676 | qdisc = &noop_qdisc; | |
677 | dev_queue->qdisc_sleeping = qdisc; | |
678 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
679 | ||
680 | spin_unlock_bh(root_lock); | |
681 | ||
682 | return oqdisc; | |
683 | } | |
b8970f0b | 684 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 685 | |
e8a0464c DM |
686 | static void attach_one_default_qdisc(struct net_device *dev, |
687 | struct netdev_queue *dev_queue, | |
688 | void *_unused) | |
689 | { | |
cc7ec456 | 690 | struct Qdisc *qdisc = &noqueue_qdisc; |
e8a0464c DM |
691 | |
692 | if (dev->tx_queue_len) { | |
3511c913 | 693 | qdisc = qdisc_create_dflt(dev_queue, |
d3678b46 | 694 | &pfifo_fast_ops, TC_H_ROOT); |
e8a0464c | 695 | if (!qdisc) { |
cc7ec456 | 696 | netdev_info(dev, "activation failed\n"); |
e8a0464c DM |
697 | return; |
698 | } | |
e8a0464c DM |
699 | } |
700 | dev_queue->qdisc_sleeping = qdisc; | |
701 | } | |
702 | ||
6ec1c69a DM |
703 | static void attach_default_qdiscs(struct net_device *dev) |
704 | { | |
705 | struct netdev_queue *txq; | |
706 | struct Qdisc *qdisc; | |
707 | ||
708 | txq = netdev_get_tx_queue(dev, 0); | |
709 | ||
710 | if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) { | |
711 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); | |
712 | dev->qdisc = txq->qdisc_sleeping; | |
713 | atomic_inc(&dev->qdisc->refcnt); | |
714 | } else { | |
3511c913 | 715 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT); |
6ec1c69a DM |
716 | if (qdisc) { |
717 | qdisc->ops->attach(qdisc); | |
718 | dev->qdisc = qdisc; | |
719 | } | |
720 | } | |
721 | } | |
722 | ||
e8a0464c DM |
723 | static void transition_one_qdisc(struct net_device *dev, |
724 | struct netdev_queue *dev_queue, | |
725 | void *_need_watchdog) | |
726 | { | |
83874000 | 727 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
728 | int *need_watchdog_p = _need_watchdog; |
729 | ||
a9312ae8 DM |
730 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
731 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
732 | ||
83874000 | 733 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
9d21493b ED |
734 | if (need_watchdog_p && new_qdisc != &noqueue_qdisc) { |
735 | dev_queue->trans_start = 0; | |
e8a0464c | 736 | *need_watchdog_p = 1; |
9d21493b | 737 | } |
e8a0464c DM |
738 | } |
739 | ||
1da177e4 LT |
740 | void dev_activate(struct net_device *dev) |
741 | { | |
e8a0464c | 742 | int need_watchdog; |
b0e1e646 | 743 | |
1da177e4 | 744 | /* No queueing discipline is attached to device; |
d3678b46 DM |
745 | create default one i.e. pfifo_fast for devices, |
746 | which need queueing and noqueue_qdisc for | |
747 | virtual interfaces | |
1da177e4 LT |
748 | */ |
749 | ||
6ec1c69a DM |
750 | if (dev->qdisc == &noop_qdisc) |
751 | attach_default_qdiscs(dev); | |
af356afa | 752 | |
cacaddf5 TC |
753 | if (!netif_carrier_ok(dev)) |
754 | /* Delay activation until next carrier-on event */ | |
755 | return; | |
756 | ||
e8a0464c DM |
757 | need_watchdog = 0; |
758 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
24824a09 ED |
759 | if (dev_ingress_queue(dev)) |
760 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
761 | |
762 | if (need_watchdog) { | |
1da177e4 LT |
763 | dev->trans_start = jiffies; |
764 | dev_watchdog_up(dev); | |
765 | } | |
b0e1e646 | 766 | } |
b8970f0b | 767 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 768 | |
e8a0464c DM |
769 | static void dev_deactivate_queue(struct net_device *dev, |
770 | struct netdev_queue *dev_queue, | |
771 | void *_qdisc_default) | |
b0e1e646 | 772 | { |
e8a0464c | 773 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 774 | struct Qdisc *qdisc; |
970565bb | 775 | |
970565bb | 776 | qdisc = dev_queue->qdisc; |
b0e1e646 | 777 | if (qdisc) { |
83874000 DM |
778 | spin_lock_bh(qdisc_lock(qdisc)); |
779 | ||
a9312ae8 DM |
780 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
781 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
782 | ||
f7a54c13 | 783 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 784 | qdisc_reset(qdisc); |
d3b753db | 785 | |
83874000 | 786 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 787 | } |
1da177e4 LT |
788 | } |
789 | ||
4335cd2d | 790 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
791 | { |
792 | unsigned int i; | |
793 | ||
794 | for (i = 0; i < dev->num_tx_queues; i++) { | |
795 | struct netdev_queue *dev_queue; | |
7698b4fc | 796 | spinlock_t *root_lock; |
e2627c8c | 797 | struct Qdisc *q; |
e8a0464c DM |
798 | int val; |
799 | ||
800 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 801 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 802 | root_lock = qdisc_lock(q); |
e8a0464c | 803 | |
4335cd2d | 804 | spin_lock_bh(root_lock); |
e8a0464c | 805 | |
bc135b23 | 806 | val = (qdisc_is_running(q) || |
b9a3b110 | 807 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 808 | |
4335cd2d | 809 | spin_unlock_bh(root_lock); |
e8a0464c DM |
810 | |
811 | if (val) | |
812 | return true; | |
813 | } | |
814 | return false; | |
815 | } | |
816 | ||
3137663d ED |
817 | /** |
818 | * dev_deactivate_many - deactivate transmissions on several devices | |
819 | * @head: list of devices to deactivate | |
820 | * | |
821 | * This function returns only when all outstanding transmissions | |
822 | * have completed, unless all devices are in dismantle phase. | |
823 | */ | |
44345724 | 824 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 825 | { |
44345724 | 826 | struct net_device *dev; |
3137663d | 827 | bool sync_needed = false; |
41a23b07 | 828 | |
44345724 OP |
829 | list_for_each_entry(dev, head, unreg_list) { |
830 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, | |
831 | &noop_qdisc); | |
832 | if (dev_ingress_queue(dev)) | |
833 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
834 | &noop_qdisc); | |
835 | ||
836 | dev_watchdog_down(dev); | |
3137663d | 837 | sync_needed |= !dev->dismantle; |
44345724 | 838 | } |
1da177e4 | 839 | |
3137663d ED |
840 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
841 | * This is avoided if all devices are in dismantle phase : | |
842 | * Caller will call synchronize_net() for us | |
843 | */ | |
844 | if (sync_needed) | |
845 | synchronize_net(); | |
1da177e4 | 846 | |
d4828d85 | 847 | /* Wait for outstanding qdisc_run calls. */ |
44345724 OP |
848 | list_for_each_entry(dev, head, unreg_list) |
849 | while (some_qdisc_is_busy(dev)) | |
850 | yield(); | |
851 | } | |
852 | ||
853 | void dev_deactivate(struct net_device *dev) | |
854 | { | |
855 | LIST_HEAD(single); | |
856 | ||
857 | list_add(&dev->unreg_list, &single); | |
858 | dev_deactivate_many(&single); | |
5f04d506 | 859 | list_del(&single); |
1da177e4 | 860 | } |
b8970f0b | 861 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 862 | |
b0e1e646 DM |
863 | static void dev_init_scheduler_queue(struct net_device *dev, |
864 | struct netdev_queue *dev_queue, | |
e8a0464c | 865 | void *_qdisc) |
b0e1e646 | 866 | { |
e8a0464c DM |
867 | struct Qdisc *qdisc = _qdisc; |
868 | ||
b0e1e646 DM |
869 | dev_queue->qdisc = qdisc; |
870 | dev_queue->qdisc_sleeping = qdisc; | |
b0e1e646 DM |
871 | } |
872 | ||
1da177e4 LT |
873 | void dev_init_scheduler(struct net_device *dev) |
874 | { | |
af356afa | 875 | dev->qdisc = &noop_qdisc; |
e8a0464c | 876 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
877 | if (dev_ingress_queue(dev)) |
878 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 879 | |
b24b8a24 | 880 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
881 | } |
882 | ||
e8a0464c DM |
883 | static void shutdown_scheduler_queue(struct net_device *dev, |
884 | struct netdev_queue *dev_queue, | |
885 | void *_qdisc_default) | |
1da177e4 | 886 | { |
b0e1e646 | 887 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 888 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
889 | |
890 | if (qdisc) { | |
f7a54c13 | 891 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 892 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 893 | |
1da177e4 | 894 | qdisc_destroy(qdisc); |
10297b99 | 895 | } |
b0e1e646 DM |
896 | } |
897 | ||
898 | void dev_shutdown(struct net_device *dev) | |
899 | { | |
e8a0464c | 900 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
901 | if (dev_ingress_queue(dev)) |
902 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
af356afa PM |
903 | qdisc_destroy(dev->qdisc); |
904 | dev->qdisc = &noop_qdisc; | |
905 | ||
547b792c | 906 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 907 | } |