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> |
07ce76aa | 28 | #include <linux/if_vlan.h> |
292f1c7f | 29 | #include <net/sch_generic.h> |
1da177e4 | 30 | #include <net/pkt_sched.h> |
7fee226a | 31 | #include <net/dst.h> |
1da177e4 LT |
32 | |
33 | /* Main transmission queue. */ | |
34 | ||
0463d4ae | 35 | /* Modifications to data participating in scheduling must be protected with |
5fb66229 | 36 | * qdisc_lock(qdisc) spinlock. |
0463d4ae PM |
37 | * |
38 | * The idea is the following: | |
c7e4f3bb DM |
39 | * - enqueue, dequeue are serialized via qdisc root lock |
40 | * - ingress filtering is also serialized via qdisc root lock | |
0463d4ae | 41 | * - updates to tree and tree walking are only done under the rtnl mutex. |
1da177e4 | 42 | */ |
1da177e4 | 43 | |
37437bb2 | 44 | static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q) |
c716a81a | 45 | { |
7fee226a | 46 | skb_dst_force(skb); |
6252352d | 47 | q->gso_skb = skb; |
53e91503 | 48 | q->qstats.requeues++; |
bbd8a0d3 | 49 | q->q.qlen++; /* it's still part of the queue */ |
37437bb2 | 50 | __netif_schedule(q); |
6252352d | 51 | |
c716a81a JHS |
52 | return 0; |
53 | } | |
54 | ||
d3b753db | 55 | static inline struct sk_buff *dequeue_skb(struct Qdisc *q) |
c716a81a | 56 | { |
554794de | 57 | struct sk_buff *skb = q->gso_skb; |
1abbe139 | 58 | const struct netdev_queue *txq = q->dev_queue; |
554794de | 59 | |
ebf05982 | 60 | if (unlikely(skb)) { |
ebf05982 | 61 | /* check the reason of requeuing without tx lock first */ |
1abbe139 | 62 | txq = netdev_get_tx_queue(txq->dev, skb_get_queue_mapping(skb)); |
73466498 | 63 | if (!netif_xmit_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 { | |
1abbe139 ED |
69 | if (!(q->flags & TCQ_F_ONETXQUEUE) || !netif_xmit_frozen_or_stopped(txq)) |
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); |
e87cc472 JP |
90 | net_warn_ratelimited("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()); |
73466498 | 124 | if (!netif_xmit_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 | 138 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
e87cc472 JP |
139 | if (unlikely(ret != NETDEV_TX_BUSY)) |
140 | net_warn_ratelimited("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 | |
73466498 | 146 | if (ret && netif_xmit_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 | { | |
07ce76aa | 211 | unsigned long val, res; |
9d21493b ED |
212 | unsigned int i; |
213 | ||
07ce76aa | 214 | if (is_vlan_dev(dev)) |
215 | dev = vlan_dev_real_dev(dev); | |
216 | res = dev->trans_start; | |
9d21493b ED |
217 | for (i = 0; i < dev->num_tx_queues; i++) { |
218 | val = netdev_get_tx_queue(dev, i)->trans_start; | |
219 | if (val && time_after(val, res)) | |
220 | res = val; | |
221 | } | |
222 | dev->trans_start = res; | |
07ce76aa | 223 | |
9d21493b ED |
224 | return res; |
225 | } | |
226 | EXPORT_SYMBOL(dev_trans_start); | |
227 | ||
1da177e4 LT |
228 | static void dev_watchdog(unsigned long arg) |
229 | { | |
230 | struct net_device *dev = (struct net_device *)arg; | |
231 | ||
932ff279 | 232 | netif_tx_lock(dev); |
e8a0464c | 233 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
234 | if (netif_device_present(dev) && |
235 | netif_running(dev) && | |
236 | netif_carrier_ok(dev)) { | |
9d21493b | 237 | int some_queue_timedout = 0; |
e8a0464c | 238 | unsigned int i; |
9d21493b | 239 | unsigned long trans_start; |
e8a0464c DM |
240 | |
241 | for (i = 0; i < dev->num_tx_queues; i++) { | |
242 | struct netdev_queue *txq; | |
243 | ||
244 | txq = netdev_get_tx_queue(dev, i); | |
9d21493b ED |
245 | /* |
246 | * old device drivers set dev->trans_start | |
247 | */ | |
248 | trans_start = txq->trans_start ? : dev->trans_start; | |
73466498 | 249 | if (netif_xmit_stopped(txq) && |
9d21493b ED |
250 | time_after(jiffies, (trans_start + |
251 | dev->watchdog_timeo))) { | |
252 | some_queue_timedout = 1; | |
ccf5ff69 | 253 | txq->trans_timeout++; |
e8a0464c DM |
254 | break; |
255 | } | |
256 | } | |
338f7566 | 257 | |
9d21493b | 258 | if (some_queue_timedout) { |
9d21493b | 259 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 260 | dev->name, netdev_drivername(dev), i); |
d314774c | 261 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 262 | } |
e8a0464c DM |
263 | if (!mod_timer(&dev->watchdog_timer, |
264 | round_jiffies(jiffies + | |
265 | dev->watchdog_timeo))) | |
1da177e4 LT |
266 | dev_hold(dev); |
267 | } | |
268 | } | |
932ff279 | 269 | netif_tx_unlock(dev); |
1da177e4 LT |
270 | |
271 | dev_put(dev); | |
272 | } | |
273 | ||
1da177e4 LT |
274 | void __netdev_watchdog_up(struct net_device *dev) |
275 | { | |
d314774c | 276 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
277 | if (dev->watchdog_timeo <= 0) |
278 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
279 | if (!mod_timer(&dev->watchdog_timer, |
280 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
281 | dev_hold(dev); |
282 | } | |
283 | } | |
284 | ||
285 | static void dev_watchdog_up(struct net_device *dev) | |
286 | { | |
1da177e4 | 287 | __netdev_watchdog_up(dev); |
1da177e4 LT |
288 | } |
289 | ||
290 | static void dev_watchdog_down(struct net_device *dev) | |
291 | { | |
932ff279 | 292 | netif_tx_lock_bh(dev); |
1da177e4 | 293 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 294 | dev_put(dev); |
932ff279 | 295 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
296 | } |
297 | ||
bea3348e SH |
298 | /** |
299 | * netif_carrier_on - set carrier | |
300 | * @dev: network device | |
301 | * | |
302 | * Device has detected that carrier. | |
303 | */ | |
0a242efc DV |
304 | void netif_carrier_on(struct net_device *dev) |
305 | { | |
bfaae0f0 | 306 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
307 | if (dev->reg_state == NETREG_UNINITIALIZED) |
308 | return; | |
0a242efc | 309 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
310 | if (netif_running(dev)) |
311 | __netdev_watchdog_up(dev); | |
312 | } | |
0a242efc | 313 | } |
62e3ba1b | 314 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 315 | |
bea3348e SH |
316 | /** |
317 | * netif_carrier_off - clear carrier | |
318 | * @dev: network device | |
319 | * | |
320 | * Device has detected loss of carrier. | |
321 | */ | |
0a242efc DV |
322 | void netif_carrier_off(struct net_device *dev) |
323 | { | |
b4730016 DM |
324 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
325 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
326 | return; | |
0a242efc | 327 | linkwatch_fire_event(dev); |
b4730016 | 328 | } |
0a242efc | 329 | } |
62e3ba1b | 330 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 331 | |
1da177e4 LT |
332 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
333 | under all circumstances. It is difficult to invent anything faster or | |
334 | cheaper. | |
335 | */ | |
336 | ||
94df109a | 337 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc * qdisc) |
1da177e4 LT |
338 | { |
339 | kfree_skb(skb); | |
340 | return NET_XMIT_CN; | |
341 | } | |
342 | ||
94df109a | 343 | static struct sk_buff *noop_dequeue(struct Qdisc * qdisc) |
1da177e4 LT |
344 | { |
345 | return NULL; | |
346 | } | |
347 | ||
20fea08b | 348 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
349 | .id = "noop", |
350 | .priv_size = 0, | |
351 | .enqueue = noop_enqueue, | |
352 | .dequeue = noop_dequeue, | |
99c0db26 | 353 | .peek = noop_dequeue, |
1da177e4 LT |
354 | .owner = THIS_MODULE, |
355 | }; | |
356 | ||
7698b4fc | 357 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 358 | .qdisc = &noop_qdisc, |
9f3ffae0 | 359 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
360 | }; |
361 | ||
1da177e4 LT |
362 | struct Qdisc noop_qdisc = { |
363 | .enqueue = noop_enqueue, | |
364 | .dequeue = noop_dequeue, | |
365 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 366 | .ops = &noop_qdisc_ops, |
1da177e4 | 367 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 368 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 369 | .dev_queue = &noop_netdev_queue, |
7b5edbc4 | 370 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
1da177e4 | 371 | }; |
62e3ba1b | 372 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 373 | |
20fea08b | 374 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
375 | .id = "noqueue", |
376 | .priv_size = 0, | |
377 | .enqueue = noop_enqueue, | |
378 | .dequeue = noop_dequeue, | |
99c0db26 | 379 | .peek = noop_dequeue, |
1da177e4 LT |
380 | .owner = THIS_MODULE, |
381 | }; | |
382 | ||
30ee42be DM |
383 | static struct Qdisc noqueue_qdisc; |
384 | static struct netdev_queue noqueue_netdev_queue = { | |
385 | .qdisc = &noqueue_qdisc, | |
9f3ffae0 | 386 | .qdisc_sleeping = &noqueue_qdisc, |
30ee42be DM |
387 | }; |
388 | ||
1da177e4 LT |
389 | static struct Qdisc noqueue_qdisc = { |
390 | .enqueue = NULL, | |
391 | .dequeue = noop_dequeue, | |
392 | .flags = TCQ_F_BUILTIN, | |
393 | .ops = &noqueue_qdisc_ops, | |
394 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
30ee42be DM |
395 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
396 | .dev_queue = &noqueue_netdev_queue, | |
7b5edbc4 | 397 | .busylock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock), |
1da177e4 LT |
398 | }; |
399 | ||
400 | ||
cc7ec456 ED |
401 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
402 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
403 | }; | |
d3678b46 DM |
404 | |
405 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
406 | generic prio+fifo combination. | |
407 | */ | |
408 | ||
409 | #define PFIFO_FAST_BANDS 3 | |
410 | ||
fd3ae5e8 KK |
411 | /* |
412 | * Private data for a pfifo_fast scheduler containing: | |
413 | * - queues for the three band | |
414 | * - bitmap indicating which of the bands contain skbs | |
415 | */ | |
416 | struct pfifo_fast_priv { | |
417 | u32 bitmap; | |
418 | struct sk_buff_head q[PFIFO_FAST_BANDS]; | |
419 | }; | |
420 | ||
421 | /* | |
422 | * Convert a bitmap to the first band number where an skb is queued, where: | |
423 | * bitmap=0 means there are no skbs on any band. | |
424 | * bitmap=1 means there is an skb on band 0. | |
425 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
426 | */ | |
427 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
428 | ||
429 | static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv, | |
430 | int band) | |
d3678b46 | 431 | { |
fd3ae5e8 | 432 | return priv->q + band; |
d3678b46 DM |
433 | } |
434 | ||
cc7ec456 | 435 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc) |
321090e7 | 436 | { |
a453e068 KK |
437 | if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) { |
438 | int band = prio2band[skb->priority & TC_PRIO_MAX]; | |
439 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
440 | struct sk_buff_head *list = band2list(priv, band); | |
1da177e4 | 441 | |
fd3ae5e8 | 442 | priv->bitmap |= (1 << band); |
d3678b46 | 443 | qdisc->q.qlen++; |
821d24ae | 444 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 445 | } |
821d24ae TG |
446 | |
447 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
448 | } |
449 | ||
cc7ec456 | 450 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc) |
1da177e4 | 451 | { |
fd3ae5e8 KK |
452 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
453 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 454 | |
fd3ae5e8 KK |
455 | if (likely(band >= 0)) { |
456 | struct sk_buff_head *list = band2list(priv, band); | |
457 | struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list); | |
458 | ||
459 | qdisc->q.qlen--; | |
460 | if (skb_queue_empty(list)) | |
461 | priv->bitmap &= ~(1 << band); | |
462 | ||
463 | return skb; | |
d3678b46 | 464 | } |
f87a9c3d | 465 | |
1da177e4 LT |
466 | return NULL; |
467 | } | |
468 | ||
cc7ec456 | 469 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 470 | { |
fd3ae5e8 KK |
471 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
472 | int band = bitmap2band[priv->bitmap]; | |
473 | ||
474 | if (band >= 0) { | |
475 | struct sk_buff_head *list = band2list(priv, band); | |
99c0db26 | 476 | |
fd3ae5e8 | 477 | return skb_peek(list); |
99c0db26 JP |
478 | } |
479 | ||
480 | return NULL; | |
481 | } | |
482 | ||
cc7ec456 | 483 | static void pfifo_fast_reset(struct Qdisc *qdisc) |
1da177e4 | 484 | { |
d3678b46 | 485 | int prio; |
fd3ae5e8 | 486 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
487 | |
488 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 489 | __qdisc_reset_queue(qdisc, band2list(priv, prio)); |
d3678b46 | 490 | |
fd3ae5e8 | 491 | priv->bitmap = 0; |
821d24ae | 492 | qdisc->qstats.backlog = 0; |
d3678b46 | 493 | qdisc->q.qlen = 0; |
1da177e4 LT |
494 | } |
495 | ||
d3678b46 DM |
496 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
497 | { | |
498 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
499 | ||
cc7ec456 | 500 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
1b34ec43 DM |
501 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
502 | goto nla_put_failure; | |
d3678b46 DM |
503 | return skb->len; |
504 | ||
505 | nla_put_failure: | |
506 | return -1; | |
507 | } | |
508 | ||
509 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
510 | { | |
511 | int prio; | |
fd3ae5e8 | 512 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
513 | |
514 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 515 | skb_queue_head_init(band2list(priv, prio)); |
d3678b46 | 516 | |
23624935 ED |
517 | /* Can by-pass the queue discipline */ |
518 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
519 | return 0; |
520 | } | |
521 | ||
6ec1c69a | 522 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 523 | .id = "pfifo_fast", |
fd3ae5e8 | 524 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
525 | .enqueue = pfifo_fast_enqueue, |
526 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 527 | .peek = pfifo_fast_peek, |
d3678b46 DM |
528 | .init = pfifo_fast_init, |
529 | .reset = pfifo_fast_reset, | |
530 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
531 | .owner = THIS_MODULE, |
532 | }; | |
b8970f0b | 533 | EXPORT_SYMBOL(pfifo_fast_ops); |
1da177e4 | 534 | |
23d3b8bf ED |
535 | static struct lock_class_key qdisc_tx_busylock; |
536 | ||
5ce2d488 | 537 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
bb949fbd | 538 | struct Qdisc_ops *ops) |
1da177e4 LT |
539 | { |
540 | void *p; | |
541 | struct Qdisc *sch; | |
d276055c | 542 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 543 | int err = -ENOBUFS; |
23d3b8bf | 544 | struct net_device *dev = dev_queue->dev; |
1da177e4 | 545 | |
f2cd2d3e ED |
546 | p = kzalloc_node(size, GFP_KERNEL, |
547 | netdev_queue_numa_node_read(dev_queue)); | |
548 | ||
1da177e4 | 549 | if (!p) |
3d54b82f | 550 | goto errout; |
3d54b82f | 551 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
552 | /* if we got non aligned memory, ask more and do alignment ourself */ |
553 | if (sch != p) { | |
554 | kfree(p); | |
555 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
556 | netdev_queue_numa_node_read(dev_queue)); | |
557 | if (!p) | |
558 | goto errout; | |
559 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
560 | sch->padded = (char *) sch - (char *) p; | |
561 | } | |
1da177e4 LT |
562 | INIT_LIST_HEAD(&sch->list); |
563 | skb_queue_head_init(&sch->q); | |
23d3b8bf | 564 | |
79640a4c | 565 | spin_lock_init(&sch->busylock); |
23d3b8bf ED |
566 | lockdep_set_class(&sch->busylock, |
567 | dev->qdisc_tx_busylock ?: &qdisc_tx_busylock); | |
568 | ||
1da177e4 LT |
569 | sch->ops = ops; |
570 | sch->enqueue = ops->enqueue; | |
571 | sch->dequeue = ops->dequeue; | |
bb949fbd | 572 | sch->dev_queue = dev_queue; |
23d3b8bf | 573 | dev_hold(dev); |
1da177e4 | 574 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
575 | |
576 | return sch; | |
577 | errout: | |
01e123d7 | 578 | return ERR_PTR(err); |
3d54b82f TG |
579 | } |
580 | ||
3511c913 CG |
581 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
582 | struct Qdisc_ops *ops, unsigned int parentid) | |
3d54b82f TG |
583 | { |
584 | struct Qdisc *sch; | |
10297b99 | 585 | |
5ce2d488 | 586 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
587 | if (IS_ERR(sch)) |
588 | goto errout; | |
9f9afec4 | 589 | sch->parent = parentid; |
3d54b82f | 590 | |
1da177e4 LT |
591 | if (!ops->init || ops->init(sch, NULL) == 0) |
592 | return sch; | |
593 | ||
0fbbeb1b | 594 | qdisc_destroy(sch); |
3d54b82f | 595 | errout: |
1da177e4 LT |
596 | return NULL; |
597 | } | |
62e3ba1b | 598 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 599 | |
5fb66229 | 600 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
601 | |
602 | void qdisc_reset(struct Qdisc *qdisc) | |
603 | { | |
20fea08b | 604 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
605 | |
606 | if (ops->reset) | |
607 | ops->reset(qdisc); | |
67305ebc | 608 | |
bbd8a0d3 KK |
609 | if (qdisc->gso_skb) { |
610 | kfree_skb(qdisc->gso_skb); | |
611 | qdisc->gso_skb = NULL; | |
612 | qdisc->q.qlen = 0; | |
613 | } | |
1da177e4 | 614 | } |
62e3ba1b | 615 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 616 | |
5d944c64 ED |
617 | static void qdisc_rcu_free(struct rcu_head *head) |
618 | { | |
619 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
620 | ||
621 | kfree((char *) qdisc - qdisc->padded); | |
622 | } | |
623 | ||
1e0d5a57 | 624 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 625 | { |
8a34c5dc DM |
626 | const struct Qdisc_ops *ops = qdisc->ops; |
627 | ||
1e0d5a57 DM |
628 | if (qdisc->flags & TCQ_F_BUILTIN || |
629 | !atomic_dec_and_test(&qdisc->refcnt)) | |
630 | return; | |
631 | ||
3a682fbd | 632 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
633 | qdisc_list_del(qdisc); |
634 | ||
a2da570d | 635 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 636 | #endif |
8a34c5dc DM |
637 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
638 | if (ops->reset) | |
639 | ops->reset(qdisc); | |
640 | if (ops->destroy) | |
641 | ops->destroy(qdisc); | |
642 | ||
643 | module_put(ops->owner); | |
644 | dev_put(qdisc_dev(qdisc)); | |
645 | ||
554794de | 646 | kfree_skb(qdisc->gso_skb); |
5d944c64 ED |
647 | /* |
648 | * gen_estimator est_timer() might access qdisc->q.lock, | |
649 | * wait a RCU grace period before freeing qdisc. | |
650 | */ | |
651 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 652 | } |
62e3ba1b | 653 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 654 | |
589983cd PM |
655 | /* Attach toplevel qdisc to device queue. */ |
656 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
657 | struct Qdisc *qdisc) | |
658 | { | |
659 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
660 | spinlock_t *root_lock; | |
661 | ||
662 | root_lock = qdisc_lock(oqdisc); | |
663 | spin_lock_bh(root_lock); | |
664 | ||
665 | /* Prune old scheduler */ | |
666 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
667 | qdisc_reset(oqdisc); | |
668 | ||
669 | /* ... and graft new one */ | |
670 | if (qdisc == NULL) | |
671 | qdisc = &noop_qdisc; | |
672 | dev_queue->qdisc_sleeping = qdisc; | |
673 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
674 | ||
675 | spin_unlock_bh(root_lock); | |
676 | ||
677 | return oqdisc; | |
678 | } | |
b8970f0b | 679 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 680 | |
e8a0464c DM |
681 | static void attach_one_default_qdisc(struct net_device *dev, |
682 | struct netdev_queue *dev_queue, | |
683 | void *_unused) | |
684 | { | |
cc7ec456 | 685 | struct Qdisc *qdisc = &noqueue_qdisc; |
e8a0464c DM |
686 | |
687 | if (dev->tx_queue_len) { | |
3511c913 | 688 | qdisc = qdisc_create_dflt(dev_queue, |
d3678b46 | 689 | &pfifo_fast_ops, TC_H_ROOT); |
e8a0464c | 690 | if (!qdisc) { |
cc7ec456 | 691 | netdev_info(dev, "activation failed\n"); |
e8a0464c DM |
692 | return; |
693 | } | |
1abbe139 ED |
694 | if (!netif_is_multiqueue(dev)) |
695 | qdisc->flags |= TCQ_F_ONETXQUEUE; | |
e8a0464c DM |
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 { | |
3511c913 | 712 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT); |
6ec1c69a DM |
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); | |
24824a09 ED |
756 | if (dev_ingress_queue(dev)) |
757 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
758 | |
759 | if (need_watchdog) { | |
1da177e4 LT |
760 | dev->trans_start = jiffies; |
761 | dev_watchdog_up(dev); | |
762 | } | |
b0e1e646 | 763 | } |
b8970f0b | 764 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 765 | |
e8a0464c DM |
766 | static void dev_deactivate_queue(struct net_device *dev, |
767 | struct netdev_queue *dev_queue, | |
768 | void *_qdisc_default) | |
b0e1e646 | 769 | { |
e8a0464c | 770 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 771 | struct Qdisc *qdisc; |
970565bb | 772 | |
970565bb | 773 | qdisc = dev_queue->qdisc; |
b0e1e646 | 774 | if (qdisc) { |
83874000 DM |
775 | spin_lock_bh(qdisc_lock(qdisc)); |
776 | ||
a9312ae8 DM |
777 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
778 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
779 | ||
f7a54c13 | 780 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 781 | qdisc_reset(qdisc); |
d3b753db | 782 | |
83874000 | 783 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 784 | } |
1da177e4 LT |
785 | } |
786 | ||
4335cd2d | 787 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
788 | { |
789 | unsigned int i; | |
790 | ||
791 | for (i = 0; i < dev->num_tx_queues; i++) { | |
792 | struct netdev_queue *dev_queue; | |
7698b4fc | 793 | spinlock_t *root_lock; |
e2627c8c | 794 | struct Qdisc *q; |
e8a0464c DM |
795 | int val; |
796 | ||
797 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 798 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 799 | root_lock = qdisc_lock(q); |
e8a0464c | 800 | |
4335cd2d | 801 | spin_lock_bh(root_lock); |
e8a0464c | 802 | |
bc135b23 | 803 | val = (qdisc_is_running(q) || |
b9a3b110 | 804 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 805 | |
4335cd2d | 806 | spin_unlock_bh(root_lock); |
e8a0464c DM |
807 | |
808 | if (val) | |
809 | return true; | |
810 | } | |
811 | return false; | |
812 | } | |
813 | ||
3137663d ED |
814 | /** |
815 | * dev_deactivate_many - deactivate transmissions on several devices | |
816 | * @head: list of devices to deactivate | |
817 | * | |
818 | * This function returns only when all outstanding transmissions | |
819 | * have completed, unless all devices are in dismantle phase. | |
820 | */ | |
44345724 | 821 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 822 | { |
44345724 | 823 | struct net_device *dev; |
3137663d | 824 | bool sync_needed = false; |
41a23b07 | 825 | |
44345724 OP |
826 | list_for_each_entry(dev, head, unreg_list) { |
827 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, | |
828 | &noop_qdisc); | |
829 | if (dev_ingress_queue(dev)) | |
830 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
831 | &noop_qdisc); | |
832 | ||
833 | dev_watchdog_down(dev); | |
3137663d | 834 | sync_needed |= !dev->dismantle; |
44345724 | 835 | } |
1da177e4 | 836 | |
3137663d ED |
837 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
838 | * This is avoided if all devices are in dismantle phase : | |
839 | * Caller will call synchronize_net() for us | |
840 | */ | |
841 | if (sync_needed) | |
842 | synchronize_net(); | |
1da177e4 | 843 | |
d4828d85 | 844 | /* Wait for outstanding qdisc_run calls. */ |
44345724 OP |
845 | list_for_each_entry(dev, head, unreg_list) |
846 | while (some_qdisc_is_busy(dev)) | |
847 | yield(); | |
848 | } | |
849 | ||
850 | void dev_deactivate(struct net_device *dev) | |
851 | { | |
852 | LIST_HEAD(single); | |
853 | ||
854 | list_add(&dev->unreg_list, &single); | |
855 | dev_deactivate_many(&single); | |
5f04d506 | 856 | list_del(&single); |
1da177e4 | 857 | } |
b8970f0b | 858 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 859 | |
b0e1e646 DM |
860 | static void dev_init_scheduler_queue(struct net_device *dev, |
861 | struct netdev_queue *dev_queue, | |
e8a0464c | 862 | void *_qdisc) |
b0e1e646 | 863 | { |
e8a0464c DM |
864 | struct Qdisc *qdisc = _qdisc; |
865 | ||
b0e1e646 DM |
866 | dev_queue->qdisc = qdisc; |
867 | dev_queue->qdisc_sleeping = qdisc; | |
b0e1e646 DM |
868 | } |
869 | ||
1da177e4 LT |
870 | void dev_init_scheduler(struct net_device *dev) |
871 | { | |
af356afa | 872 | dev->qdisc = &noop_qdisc; |
e8a0464c | 873 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
874 | if (dev_ingress_queue(dev)) |
875 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 876 | |
b24b8a24 | 877 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
878 | } |
879 | ||
e8a0464c DM |
880 | static void shutdown_scheduler_queue(struct net_device *dev, |
881 | struct netdev_queue *dev_queue, | |
882 | void *_qdisc_default) | |
1da177e4 | 883 | { |
b0e1e646 | 884 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 885 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
886 | |
887 | if (qdisc) { | |
f7a54c13 | 888 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 889 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 890 | |
1da177e4 | 891 | qdisc_destroy(qdisc); |
10297b99 | 892 | } |
b0e1e646 DM |
893 | } |
894 | ||
895 | void dev_shutdown(struct net_device *dev) | |
896 | { | |
e8a0464c | 897 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
898 | if (dev_ingress_queue(dev)) |
899 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
af356afa PM |
900 | qdisc_destroy(dev->qdisc); |
901 | dev->qdisc = &noop_qdisc; | |
902 | ||
547b792c | 903 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 904 | } |
292f1c7f | 905 | |
01cb71d2 ED |
906 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
907 | const struct tc_ratespec *conf) | |
292f1c7f | 908 | { |
01cb71d2 ED |
909 | memset(r, 0, sizeof(*r)); |
910 | r->overhead = conf->overhead; | |
130d3d68 | 911 | r->rate_bytes_ps = conf->rate; |
292f1c7f JP |
912 | r->mult = 1; |
913 | /* | |
130d3d68 ED |
914 | * The deal here is to replace a divide by a reciprocal one |
915 | * in fast path (a reciprocal divide is a multiply and a shift) | |
916 | * | |
917 | * Normal formula would be : | |
918 | * time_in_ns = (NSEC_PER_SEC * len) / rate_bps | |
919 | * | |
920 | * We compute mult/shift to use instead : | |
921 | * time_in_ns = (len * mult) >> shift; | |
922 | * | |
923 | * We try to get the highest possible mult value for accuracy, | |
924 | * but have to make sure no overflows will ever happen. | |
292f1c7f | 925 | */ |
130d3d68 ED |
926 | if (r->rate_bytes_ps > 0) { |
927 | u64 factor = NSEC_PER_SEC; | |
928 | ||
929 | for (;;) { | |
930 | r->mult = div64_u64(factor, r->rate_bytes_ps); | |
931 | if (r->mult & (1U << 31) || factor & (1ULL << 63)) | |
292f1c7f | 932 | break; |
130d3d68 ED |
933 | factor <<= 1; |
934 | r->shift++; | |
292f1c7f | 935 | } |
292f1c7f JP |
936 | } |
937 | } | |
938 | EXPORT_SYMBOL(psched_ratecfg_precompute); |