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 | 32 | |
34aedd3f | 33 | /* Qdisc to use by default */ |
34 | const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops; | |
35 | EXPORT_SYMBOL(default_qdisc_ops); | |
36 | ||
1da177e4 LT |
37 | /* Main transmission queue. */ |
38 | ||
0463d4ae | 39 | /* Modifications to data participating in scheduling must be protected with |
5fb66229 | 40 | * qdisc_lock(qdisc) spinlock. |
0463d4ae PM |
41 | * |
42 | * The idea is the following: | |
c7e4f3bb DM |
43 | * - enqueue, dequeue are serialized via qdisc root lock |
44 | * - ingress filtering is also serialized via qdisc root lock | |
0463d4ae | 45 | * - updates to tree and tree walking are only done under the rtnl mutex. |
1da177e4 | 46 | */ |
1da177e4 | 47 | |
37437bb2 | 48 | static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q) |
c716a81a | 49 | { |
7fee226a | 50 | skb_dst_force(skb); |
6252352d | 51 | q->gso_skb = skb; |
53e91503 | 52 | q->qstats.requeues++; |
bbd8a0d3 | 53 | q->q.qlen++; /* it's still part of the queue */ |
37437bb2 | 54 | __netif_schedule(q); |
6252352d | 55 | |
c716a81a JHS |
56 | return 0; |
57 | } | |
58 | ||
d3b753db | 59 | static inline struct sk_buff *dequeue_skb(struct Qdisc *q) |
c716a81a | 60 | { |
554794de | 61 | struct sk_buff *skb = q->gso_skb; |
1abbe139 | 62 | const struct netdev_queue *txq = q->dev_queue; |
554794de | 63 | |
ebf05982 | 64 | if (unlikely(skb)) { |
ebf05982 | 65 | /* check the reason of requeuing without tx lock first */ |
1abbe139 | 66 | txq = netdev_get_tx_queue(txq->dev, skb_get_queue_mapping(skb)); |
73466498 | 67 | if (!netif_xmit_frozen_or_stopped(txq)) { |
6252352d | 68 | q->gso_skb = NULL; |
bbd8a0d3 KK |
69 | q->q.qlen--; |
70 | } else | |
ebf05982 JP |
71 | skb = NULL; |
72 | } else { | |
1abbe139 ED |
73 | if (!(q->flags & TCQ_F_ONETXQUEUE) || !netif_xmit_frozen_or_stopped(txq)) |
74 | skb = q->dequeue(q); | |
ebf05982 | 75 | } |
c716a81a JHS |
76 | |
77 | return skb; | |
78 | } | |
79 | ||
6c1361a6 | 80 | static inline int handle_dev_cpu_collision(struct sk_buff *skb, |
970565bb | 81 | struct netdev_queue *dev_queue, |
6c1361a6 | 82 | struct Qdisc *q) |
c716a81a | 83 | { |
6c1361a6 | 84 | int ret; |
c716a81a | 85 | |
c773e847 | 86 | if (unlikely(dev_queue->xmit_lock_owner == smp_processor_id())) { |
6c1361a6 KK |
87 | /* |
88 | * Same CPU holding the lock. It may be a transient | |
89 | * configuration error, when hard_start_xmit() recurses. We | |
90 | * detect it by checking xmit owner and drop the packet when | |
91 | * deadloop is detected. Return OK to try the next skb. | |
92 | */ | |
c716a81a | 93 | kfree_skb(skb); |
e87cc472 JP |
94 | net_warn_ratelimited("Dead loop on netdevice %s, fix it urgently!\n", |
95 | dev_queue->dev->name); | |
6c1361a6 KK |
96 | ret = qdisc_qlen(q); |
97 | } else { | |
98 | /* | |
99 | * Another cpu is holding lock, requeue & delay xmits for | |
100 | * some time. | |
101 | */ | |
d6d9ca0f | 102 | __this_cpu_inc(softnet_data.cpu_collision); |
37437bb2 | 103 | ret = dev_requeue_skb(skb, q); |
c716a81a JHS |
104 | } |
105 | ||
6c1361a6 | 106 | return ret; |
c716a81a JHS |
107 | } |
108 | ||
10297b99 | 109 | /* |
bbd8a0d3 KK |
110 | * Transmit one skb, and handle the return status as required. Holding the |
111 | * __QDISC_STATE_RUNNING bit guarantees that only one CPU can execute this | |
112 | * function. | |
6c1361a6 KK |
113 | * |
114 | * Returns to the caller: | |
115 | * 0 - queue is empty or throttled. | |
116 | * >0 - queue is not empty. | |
6c1361a6 | 117 | */ |
bbd8a0d3 KK |
118 | int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
119 | struct net_device *dev, struct netdev_queue *txq, | |
120 | spinlock_t *root_lock) | |
1da177e4 | 121 | { |
5f1a485d | 122 | int ret = NETDEV_TX_BUSY; |
7698b4fc DM |
123 | |
124 | /* And release qdisc */ | |
125 | spin_unlock(root_lock); | |
c716a81a | 126 | |
c773e847 | 127 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
73466498 | 128 | if (!netif_xmit_frozen_or_stopped(txq)) |
f663dd9a | 129 | ret = dev_hard_start_xmit(skb, dev, txq); |
572a9d7b | 130 | |
c773e847 | 131 | HARD_TX_UNLOCK(dev, txq); |
c716a81a | 132 | |
7698b4fc | 133 | spin_lock(root_lock); |
c716a81a | 134 | |
9a1654ba JP |
135 | if (dev_xmit_complete(ret)) { |
136 | /* Driver sent out skb successfully or skb was consumed */ | |
6c1361a6 | 137 | ret = qdisc_qlen(q); |
9a1654ba | 138 | } else if (ret == NETDEV_TX_LOCKED) { |
6c1361a6 | 139 | /* Driver try lock failed */ |
eb6aafe3 | 140 | ret = handle_dev_cpu_collision(skb, txq, q); |
9a1654ba | 141 | } else { |
6c1361a6 | 142 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
e87cc472 JP |
143 | if (unlikely(ret != NETDEV_TX_BUSY)) |
144 | net_warn_ratelimited("BUG %s code %d qlen %d\n", | |
145 | dev->name, ret, q->q.qlen); | |
6c1361a6 | 146 | |
37437bb2 | 147 | ret = dev_requeue_skb(skb, q); |
6c1361a6 | 148 | } |
c716a81a | 149 | |
73466498 | 150 | if (ret && netif_xmit_frozen_or_stopped(txq)) |
37437bb2 DM |
151 | ret = 0; |
152 | ||
6c1361a6 | 153 | return ret; |
1da177e4 LT |
154 | } |
155 | ||
bbd8a0d3 KK |
156 | /* |
157 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. | |
158 | * | |
159 | * __QDISC_STATE_RUNNING guarantees only one CPU can process | |
160 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for | |
161 | * this queue. | |
162 | * | |
163 | * netif_tx_lock serializes accesses to device driver. | |
164 | * | |
165 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, | |
166 | * if one is grabbed, another must be free. | |
167 | * | |
168 | * Note, that this procedure can be called by a watchdog timer | |
169 | * | |
170 | * Returns to the caller: | |
171 | * 0 - queue is empty or throttled. | |
172 | * >0 - queue is not empty. | |
173 | * | |
174 | */ | |
175 | static inline int qdisc_restart(struct Qdisc *q) | |
176 | { | |
177 | struct netdev_queue *txq; | |
178 | struct net_device *dev; | |
179 | spinlock_t *root_lock; | |
180 | struct sk_buff *skb; | |
181 | ||
182 | /* Dequeue packet */ | |
183 | skb = dequeue_skb(q); | |
184 | if (unlikely(!skb)) | |
185 | return 0; | |
7fee226a | 186 | WARN_ON_ONCE(skb_dst_is_noref(skb)); |
bbd8a0d3 KK |
187 | root_lock = qdisc_lock(q); |
188 | dev = qdisc_dev(q); | |
189 | txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb)); | |
190 | ||
191 | return sch_direct_xmit(skb, q, dev, txq, root_lock); | |
192 | } | |
193 | ||
37437bb2 | 194 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 195 | { |
d5b8aa1d | 196 | int quota = weight_p; |
2ba2506c | 197 | |
37437bb2 | 198 | while (qdisc_restart(q)) { |
2ba2506c | 199 | /* |
d5b8aa1d | 200 | * Ordered by possible occurrence: Postpone processing if |
201 | * 1. we've exceeded packet quota | |
202 | * 2. another process needs the CPU; | |
2ba2506c | 203 | */ |
f0c50c7c | 204 | if (--quota <= 0 || need_resched()) { |
37437bb2 | 205 | __netif_schedule(q); |
d90df3ad | 206 | break; |
2ba2506c HX |
207 | } |
208 | } | |
48d83325 | 209 | |
bc135b23 | 210 | qdisc_run_end(q); |
48d83325 HX |
211 | } |
212 | ||
9d21493b ED |
213 | unsigned long dev_trans_start(struct net_device *dev) |
214 | { | |
07ce76aa | 215 | unsigned long val, res; |
9d21493b ED |
216 | unsigned int i; |
217 | ||
07ce76aa | 218 | if (is_vlan_dev(dev)) |
219 | dev = vlan_dev_real_dev(dev); | |
220 | res = dev->trans_start; | |
9d21493b ED |
221 | for (i = 0; i < dev->num_tx_queues; i++) { |
222 | val = netdev_get_tx_queue(dev, i)->trans_start; | |
223 | if (val && time_after(val, res)) | |
224 | res = val; | |
225 | } | |
226 | dev->trans_start = res; | |
07ce76aa | 227 | |
9d21493b ED |
228 | return res; |
229 | } | |
230 | EXPORT_SYMBOL(dev_trans_start); | |
231 | ||
1da177e4 LT |
232 | static void dev_watchdog(unsigned long arg) |
233 | { | |
234 | struct net_device *dev = (struct net_device *)arg; | |
235 | ||
932ff279 | 236 | netif_tx_lock(dev); |
e8a0464c | 237 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
238 | if (netif_device_present(dev) && |
239 | netif_running(dev) && | |
240 | netif_carrier_ok(dev)) { | |
9d21493b | 241 | int some_queue_timedout = 0; |
e8a0464c | 242 | unsigned int i; |
9d21493b | 243 | unsigned long trans_start; |
e8a0464c DM |
244 | |
245 | for (i = 0; i < dev->num_tx_queues; i++) { | |
246 | struct netdev_queue *txq; | |
247 | ||
248 | txq = netdev_get_tx_queue(dev, i); | |
9d21493b ED |
249 | /* |
250 | * old device drivers set dev->trans_start | |
251 | */ | |
252 | trans_start = txq->trans_start ? : dev->trans_start; | |
73466498 | 253 | if (netif_xmit_stopped(txq) && |
9d21493b ED |
254 | time_after(jiffies, (trans_start + |
255 | dev->watchdog_timeo))) { | |
256 | some_queue_timedout = 1; | |
ccf5ff69 | 257 | txq->trans_timeout++; |
e8a0464c DM |
258 | break; |
259 | } | |
260 | } | |
338f7566 | 261 | |
9d21493b | 262 | if (some_queue_timedout) { |
9d21493b | 263 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 264 | dev->name, netdev_drivername(dev), i); |
d314774c | 265 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 266 | } |
e8a0464c DM |
267 | if (!mod_timer(&dev->watchdog_timer, |
268 | round_jiffies(jiffies + | |
269 | dev->watchdog_timeo))) | |
1da177e4 LT |
270 | dev_hold(dev); |
271 | } | |
272 | } | |
932ff279 | 273 | netif_tx_unlock(dev); |
1da177e4 LT |
274 | |
275 | dev_put(dev); | |
276 | } | |
277 | ||
1da177e4 LT |
278 | void __netdev_watchdog_up(struct net_device *dev) |
279 | { | |
d314774c | 280 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
281 | if (dev->watchdog_timeo <= 0) |
282 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
283 | if (!mod_timer(&dev->watchdog_timer, |
284 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
285 | dev_hold(dev); |
286 | } | |
287 | } | |
288 | ||
289 | static void dev_watchdog_up(struct net_device *dev) | |
290 | { | |
1da177e4 | 291 | __netdev_watchdog_up(dev); |
1da177e4 LT |
292 | } |
293 | ||
294 | static void dev_watchdog_down(struct net_device *dev) | |
295 | { | |
932ff279 | 296 | netif_tx_lock_bh(dev); |
1da177e4 | 297 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 298 | dev_put(dev); |
932ff279 | 299 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
300 | } |
301 | ||
bea3348e SH |
302 | /** |
303 | * netif_carrier_on - set carrier | |
304 | * @dev: network device | |
305 | * | |
306 | * Device has detected that carrier. | |
307 | */ | |
0a242efc DV |
308 | void netif_carrier_on(struct net_device *dev) |
309 | { | |
bfaae0f0 | 310 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
311 | if (dev->reg_state == NETREG_UNINITIALIZED) |
312 | return; | |
0a242efc | 313 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
314 | if (netif_running(dev)) |
315 | __netdev_watchdog_up(dev); | |
316 | } | |
0a242efc | 317 | } |
62e3ba1b | 318 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 319 | |
bea3348e SH |
320 | /** |
321 | * netif_carrier_off - clear carrier | |
322 | * @dev: network device | |
323 | * | |
324 | * Device has detected loss of carrier. | |
325 | */ | |
0a242efc DV |
326 | void netif_carrier_off(struct net_device *dev) |
327 | { | |
b4730016 DM |
328 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
329 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
330 | return; | |
0a242efc | 331 | linkwatch_fire_event(dev); |
b4730016 | 332 | } |
0a242efc | 333 | } |
62e3ba1b | 334 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 335 | |
1da177e4 LT |
336 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
337 | under all circumstances. It is difficult to invent anything faster or | |
338 | cheaper. | |
339 | */ | |
340 | ||
82d567c2 | 341 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc) |
1da177e4 LT |
342 | { |
343 | kfree_skb(skb); | |
344 | return NET_XMIT_CN; | |
345 | } | |
346 | ||
82d567c2 | 347 | static struct sk_buff *noop_dequeue(struct Qdisc *qdisc) |
1da177e4 LT |
348 | { |
349 | return NULL; | |
350 | } | |
351 | ||
20fea08b | 352 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
353 | .id = "noop", |
354 | .priv_size = 0, | |
355 | .enqueue = noop_enqueue, | |
356 | .dequeue = noop_dequeue, | |
99c0db26 | 357 | .peek = noop_dequeue, |
1da177e4 LT |
358 | .owner = THIS_MODULE, |
359 | }; | |
360 | ||
7698b4fc | 361 | static struct netdev_queue noop_netdev_queue = { |
7698b4fc | 362 | .qdisc = &noop_qdisc, |
9f3ffae0 | 363 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
364 | }; |
365 | ||
1da177e4 LT |
366 | struct Qdisc noop_qdisc = { |
367 | .enqueue = noop_enqueue, | |
368 | .dequeue = noop_dequeue, | |
369 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 370 | .ops = &noop_qdisc_ops, |
1da177e4 | 371 | .list = LIST_HEAD_INIT(noop_qdisc.list), |
83874000 | 372 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 373 | .dev_queue = &noop_netdev_queue, |
7b5edbc4 | 374 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
1da177e4 | 375 | }; |
62e3ba1b | 376 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 377 | |
20fea08b | 378 | static struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { |
1da177e4 LT |
379 | .id = "noqueue", |
380 | .priv_size = 0, | |
381 | .enqueue = noop_enqueue, | |
382 | .dequeue = noop_dequeue, | |
99c0db26 | 383 | .peek = noop_dequeue, |
1da177e4 LT |
384 | .owner = THIS_MODULE, |
385 | }; | |
386 | ||
30ee42be DM |
387 | static struct Qdisc noqueue_qdisc; |
388 | static struct netdev_queue noqueue_netdev_queue = { | |
389 | .qdisc = &noqueue_qdisc, | |
9f3ffae0 | 390 | .qdisc_sleeping = &noqueue_qdisc, |
30ee42be DM |
391 | }; |
392 | ||
1da177e4 LT |
393 | static struct Qdisc noqueue_qdisc = { |
394 | .enqueue = NULL, | |
395 | .dequeue = noop_dequeue, | |
396 | .flags = TCQ_F_BUILTIN, | |
397 | .ops = &noqueue_qdisc_ops, | |
398 | .list = LIST_HEAD_INIT(noqueue_qdisc.list), | |
30ee42be DM |
399 | .q.lock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.q.lock), |
400 | .dev_queue = &noqueue_netdev_queue, | |
7b5edbc4 | 401 | .busylock = __SPIN_LOCK_UNLOCKED(noqueue_qdisc.busylock), |
1da177e4 LT |
402 | }; |
403 | ||
404 | ||
cc7ec456 ED |
405 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
406 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
407 | }; | |
d3678b46 DM |
408 | |
409 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
410 | generic prio+fifo combination. | |
411 | */ | |
412 | ||
413 | #define PFIFO_FAST_BANDS 3 | |
414 | ||
fd3ae5e8 KK |
415 | /* |
416 | * Private data for a pfifo_fast scheduler containing: | |
417 | * - queues for the three band | |
418 | * - bitmap indicating which of the bands contain skbs | |
419 | */ | |
420 | struct pfifo_fast_priv { | |
421 | u32 bitmap; | |
422 | struct sk_buff_head q[PFIFO_FAST_BANDS]; | |
423 | }; | |
424 | ||
425 | /* | |
426 | * Convert a bitmap to the first band number where an skb is queued, where: | |
427 | * bitmap=0 means there are no skbs on any band. | |
428 | * bitmap=1 means there is an skb on band 0. | |
429 | * bitmap=7 means there are skbs on all 3 bands, etc. | |
430 | */ | |
431 | static const int bitmap2band[] = {-1, 0, 1, 0, 2, 0, 1, 0}; | |
432 | ||
433 | static inline struct sk_buff_head *band2list(struct pfifo_fast_priv *priv, | |
434 | int band) | |
d3678b46 | 435 | { |
fd3ae5e8 | 436 | return priv->q + band; |
d3678b46 DM |
437 | } |
438 | ||
cc7ec456 | 439 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc) |
321090e7 | 440 | { |
a453e068 KK |
441 | if (skb_queue_len(&qdisc->q) < qdisc_dev(qdisc)->tx_queue_len) { |
442 | int band = prio2band[skb->priority & TC_PRIO_MAX]; | |
443 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
444 | struct sk_buff_head *list = band2list(priv, band); | |
1da177e4 | 445 | |
fd3ae5e8 | 446 | priv->bitmap |= (1 << band); |
d3678b46 | 447 | qdisc->q.qlen++; |
821d24ae | 448 | return __qdisc_enqueue_tail(skb, qdisc, list); |
d3678b46 | 449 | } |
821d24ae TG |
450 | |
451 | return qdisc_drop(skb, qdisc); | |
1da177e4 LT |
452 | } |
453 | ||
cc7ec456 | 454 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc) |
1da177e4 | 455 | { |
fd3ae5e8 KK |
456 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
457 | int band = bitmap2band[priv->bitmap]; | |
1da177e4 | 458 | |
fd3ae5e8 KK |
459 | if (likely(band >= 0)) { |
460 | struct sk_buff_head *list = band2list(priv, band); | |
461 | struct sk_buff *skb = __qdisc_dequeue_head(qdisc, list); | |
462 | ||
463 | qdisc->q.qlen--; | |
464 | if (skb_queue_empty(list)) | |
465 | priv->bitmap &= ~(1 << band); | |
466 | ||
467 | return skb; | |
d3678b46 | 468 | } |
f87a9c3d | 469 | |
1da177e4 LT |
470 | return NULL; |
471 | } | |
472 | ||
cc7ec456 | 473 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 474 | { |
fd3ae5e8 KK |
475 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
476 | int band = bitmap2band[priv->bitmap]; | |
477 | ||
478 | if (band >= 0) { | |
479 | struct sk_buff_head *list = band2list(priv, band); | |
99c0db26 | 480 | |
fd3ae5e8 | 481 | return skb_peek(list); |
99c0db26 JP |
482 | } |
483 | ||
484 | return NULL; | |
485 | } | |
486 | ||
cc7ec456 | 487 | static void pfifo_fast_reset(struct Qdisc *qdisc) |
1da177e4 | 488 | { |
d3678b46 | 489 | int prio; |
fd3ae5e8 | 490 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
491 | |
492 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 493 | __qdisc_reset_queue(qdisc, band2list(priv, prio)); |
d3678b46 | 494 | |
fd3ae5e8 | 495 | priv->bitmap = 0; |
821d24ae | 496 | qdisc->qstats.backlog = 0; |
d3678b46 | 497 | qdisc->q.qlen = 0; |
1da177e4 LT |
498 | } |
499 | ||
d3678b46 DM |
500 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
501 | { | |
502 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
503 | ||
cc7ec456 | 504 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
1b34ec43 DM |
505 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
506 | goto nla_put_failure; | |
d3678b46 DM |
507 | return skb->len; |
508 | ||
509 | nla_put_failure: | |
510 | return -1; | |
511 | } | |
512 | ||
513 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt) | |
514 | { | |
515 | int prio; | |
fd3ae5e8 | 516 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 DM |
517 | |
518 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) | |
fd3ae5e8 | 519 | skb_queue_head_init(band2list(priv, prio)); |
d3678b46 | 520 | |
23624935 ED |
521 | /* Can by-pass the queue discipline */ |
522 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
523 | return 0; |
524 | } | |
525 | ||
6ec1c69a | 526 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 527 | .id = "pfifo_fast", |
fd3ae5e8 | 528 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
529 | .enqueue = pfifo_fast_enqueue, |
530 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 531 | .peek = pfifo_fast_peek, |
d3678b46 DM |
532 | .init = pfifo_fast_init, |
533 | .reset = pfifo_fast_reset, | |
534 | .dump = pfifo_fast_dump, | |
1da177e4 LT |
535 | .owner = THIS_MODULE, |
536 | }; | |
537 | ||
23d3b8bf ED |
538 | static struct lock_class_key qdisc_tx_busylock; |
539 | ||
5ce2d488 | 540 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
d2a7f269 | 541 | const struct Qdisc_ops *ops) |
1da177e4 LT |
542 | { |
543 | void *p; | |
544 | struct Qdisc *sch; | |
d276055c | 545 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 546 | int err = -ENOBUFS; |
23d3b8bf | 547 | struct net_device *dev = dev_queue->dev; |
1da177e4 | 548 | |
f2cd2d3e ED |
549 | p = kzalloc_node(size, GFP_KERNEL, |
550 | netdev_queue_numa_node_read(dev_queue)); | |
551 | ||
1da177e4 | 552 | if (!p) |
3d54b82f | 553 | goto errout; |
3d54b82f | 554 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
555 | /* if we got non aligned memory, ask more and do alignment ourself */ |
556 | if (sch != p) { | |
557 | kfree(p); | |
558 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
559 | netdev_queue_numa_node_read(dev_queue)); | |
560 | if (!p) | |
561 | goto errout; | |
562 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
563 | sch->padded = (char *) sch - (char *) p; | |
564 | } | |
1da177e4 LT |
565 | INIT_LIST_HEAD(&sch->list); |
566 | skb_queue_head_init(&sch->q); | |
23d3b8bf | 567 | |
79640a4c | 568 | spin_lock_init(&sch->busylock); |
23d3b8bf ED |
569 | lockdep_set_class(&sch->busylock, |
570 | dev->qdisc_tx_busylock ?: &qdisc_tx_busylock); | |
571 | ||
1da177e4 LT |
572 | sch->ops = ops; |
573 | sch->enqueue = ops->enqueue; | |
574 | sch->dequeue = ops->dequeue; | |
bb949fbd | 575 | sch->dev_queue = dev_queue; |
23d3b8bf | 576 | dev_hold(dev); |
1da177e4 | 577 | atomic_set(&sch->refcnt, 1); |
3d54b82f TG |
578 | |
579 | return sch; | |
580 | errout: | |
01e123d7 | 581 | return ERR_PTR(err); |
3d54b82f TG |
582 | } |
583 | ||
3511c913 | 584 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
d2a7f269 | 585 | const struct Qdisc_ops *ops, |
586 | unsigned int parentid) | |
3d54b82f TG |
587 | { |
588 | struct Qdisc *sch; | |
10297b99 | 589 | |
6da7c8fc | 590 | if (!try_module_get(ops->owner)) |
591 | goto errout; | |
592 | ||
5ce2d488 | 593 | sch = qdisc_alloc(dev_queue, ops); |
3d54b82f TG |
594 | if (IS_ERR(sch)) |
595 | goto errout; | |
9f9afec4 | 596 | sch->parent = parentid; |
3d54b82f | 597 | |
1da177e4 LT |
598 | if (!ops->init || ops->init(sch, NULL) == 0) |
599 | return sch; | |
600 | ||
0fbbeb1b | 601 | qdisc_destroy(sch); |
3d54b82f | 602 | errout: |
1da177e4 LT |
603 | return NULL; |
604 | } | |
62e3ba1b | 605 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 606 | |
5fb66229 | 607 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
608 | |
609 | void qdisc_reset(struct Qdisc *qdisc) | |
610 | { | |
20fea08b | 611 | const struct Qdisc_ops *ops = qdisc->ops; |
1da177e4 LT |
612 | |
613 | if (ops->reset) | |
614 | ops->reset(qdisc); | |
67305ebc | 615 | |
bbd8a0d3 KK |
616 | if (qdisc->gso_skb) { |
617 | kfree_skb(qdisc->gso_skb); | |
618 | qdisc->gso_skb = NULL; | |
619 | qdisc->q.qlen = 0; | |
620 | } | |
1da177e4 | 621 | } |
62e3ba1b | 622 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 623 | |
5d944c64 ED |
624 | static void qdisc_rcu_free(struct rcu_head *head) |
625 | { | |
626 | struct Qdisc *qdisc = container_of(head, struct Qdisc, rcu_head); | |
627 | ||
628 | kfree((char *) qdisc - qdisc->padded); | |
629 | } | |
630 | ||
1e0d5a57 | 631 | void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 632 | { |
8a34c5dc DM |
633 | const struct Qdisc_ops *ops = qdisc->ops; |
634 | ||
1e0d5a57 DM |
635 | if (qdisc->flags & TCQ_F_BUILTIN || |
636 | !atomic_dec_and_test(&qdisc->refcnt)) | |
637 | return; | |
638 | ||
3a682fbd | 639 | #ifdef CONFIG_NET_SCHED |
f6e0b239 JP |
640 | qdisc_list_del(qdisc); |
641 | ||
a2da570d | 642 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 643 | #endif |
8a34c5dc DM |
644 | gen_kill_estimator(&qdisc->bstats, &qdisc->rate_est); |
645 | if (ops->reset) | |
646 | ops->reset(qdisc); | |
647 | if (ops->destroy) | |
648 | ops->destroy(qdisc); | |
649 | ||
650 | module_put(ops->owner); | |
651 | dev_put(qdisc_dev(qdisc)); | |
652 | ||
554794de | 653 | kfree_skb(qdisc->gso_skb); |
5d944c64 ED |
654 | /* |
655 | * gen_estimator est_timer() might access qdisc->q.lock, | |
656 | * wait a RCU grace period before freeing qdisc. | |
657 | */ | |
658 | call_rcu(&qdisc->rcu_head, qdisc_rcu_free); | |
1da177e4 | 659 | } |
62e3ba1b | 660 | EXPORT_SYMBOL(qdisc_destroy); |
1da177e4 | 661 | |
589983cd PM |
662 | /* Attach toplevel qdisc to device queue. */ |
663 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
664 | struct Qdisc *qdisc) | |
665 | { | |
666 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
667 | spinlock_t *root_lock; | |
668 | ||
669 | root_lock = qdisc_lock(oqdisc); | |
670 | spin_lock_bh(root_lock); | |
671 | ||
672 | /* Prune old scheduler */ | |
673 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
674 | qdisc_reset(oqdisc); | |
675 | ||
676 | /* ... and graft new one */ | |
677 | if (qdisc == NULL) | |
678 | qdisc = &noop_qdisc; | |
679 | dev_queue->qdisc_sleeping = qdisc; | |
680 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
681 | ||
682 | spin_unlock_bh(root_lock); | |
683 | ||
684 | return oqdisc; | |
685 | } | |
b8970f0b | 686 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 687 | |
e8a0464c DM |
688 | static void attach_one_default_qdisc(struct net_device *dev, |
689 | struct netdev_queue *dev_queue, | |
690 | void *_unused) | |
691 | { | |
cc7ec456 | 692 | struct Qdisc *qdisc = &noqueue_qdisc; |
e8a0464c DM |
693 | |
694 | if (dev->tx_queue_len) { | |
3511c913 | 695 | qdisc = qdisc_create_dflt(dev_queue, |
6da7c8fc | 696 | default_qdisc_ops, TC_H_ROOT); |
e8a0464c | 697 | if (!qdisc) { |
cc7ec456 | 698 | netdev_info(dev, "activation failed\n"); |
e8a0464c DM |
699 | return; |
700 | } | |
1abbe139 ED |
701 | if (!netif_is_multiqueue(dev)) |
702 | qdisc->flags |= TCQ_F_ONETXQUEUE; | |
e8a0464c DM |
703 | } |
704 | dev_queue->qdisc_sleeping = qdisc; | |
705 | } | |
706 | ||
6ec1c69a DM |
707 | static void attach_default_qdiscs(struct net_device *dev) |
708 | { | |
709 | struct netdev_queue *txq; | |
710 | struct Qdisc *qdisc; | |
711 | ||
712 | txq = netdev_get_tx_queue(dev, 0); | |
713 | ||
714 | if (!netif_is_multiqueue(dev) || dev->tx_queue_len == 0) { | |
715 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); | |
716 | dev->qdisc = txq->qdisc_sleeping; | |
717 | atomic_inc(&dev->qdisc->refcnt); | |
718 | } else { | |
3511c913 | 719 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT); |
6ec1c69a | 720 | if (qdisc) { |
6ec1c69a | 721 | dev->qdisc = qdisc; |
e57a784d | 722 | qdisc->ops->attach(qdisc); |
6ec1c69a DM |
723 | } |
724 | } | |
725 | } | |
726 | ||
e8a0464c DM |
727 | static void transition_one_qdisc(struct net_device *dev, |
728 | struct netdev_queue *dev_queue, | |
729 | void *_need_watchdog) | |
730 | { | |
83874000 | 731 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
732 | int *need_watchdog_p = _need_watchdog; |
733 | ||
a9312ae8 DM |
734 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
735 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
736 | ||
83874000 | 737 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
9d21493b ED |
738 | if (need_watchdog_p && new_qdisc != &noqueue_qdisc) { |
739 | dev_queue->trans_start = 0; | |
e8a0464c | 740 | *need_watchdog_p = 1; |
9d21493b | 741 | } |
e8a0464c DM |
742 | } |
743 | ||
1da177e4 LT |
744 | void dev_activate(struct net_device *dev) |
745 | { | |
e8a0464c | 746 | int need_watchdog; |
b0e1e646 | 747 | |
1da177e4 | 748 | /* No queueing discipline is attached to device; |
6da7c8fc | 749 | * create default one for devices, which need queueing |
750 | * and noqueue_qdisc for virtual interfaces | |
1da177e4 LT |
751 | */ |
752 | ||
6ec1c69a DM |
753 | if (dev->qdisc == &noop_qdisc) |
754 | attach_default_qdiscs(dev); | |
af356afa | 755 | |
cacaddf5 TC |
756 | if (!netif_carrier_ok(dev)) |
757 | /* Delay activation until next carrier-on event */ | |
758 | return; | |
759 | ||
e8a0464c DM |
760 | need_watchdog = 0; |
761 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
24824a09 ED |
762 | if (dev_ingress_queue(dev)) |
763 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
764 | |
765 | if (need_watchdog) { | |
1da177e4 LT |
766 | dev->trans_start = jiffies; |
767 | dev_watchdog_up(dev); | |
768 | } | |
b0e1e646 | 769 | } |
b8970f0b | 770 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 771 | |
e8a0464c DM |
772 | static void dev_deactivate_queue(struct net_device *dev, |
773 | struct netdev_queue *dev_queue, | |
774 | void *_qdisc_default) | |
b0e1e646 | 775 | { |
e8a0464c | 776 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 777 | struct Qdisc *qdisc; |
970565bb | 778 | |
970565bb | 779 | qdisc = dev_queue->qdisc; |
b0e1e646 | 780 | if (qdisc) { |
83874000 DM |
781 | spin_lock_bh(qdisc_lock(qdisc)); |
782 | ||
a9312ae8 DM |
783 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
784 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
785 | ||
f7a54c13 | 786 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 787 | qdisc_reset(qdisc); |
d3b753db | 788 | |
83874000 | 789 | spin_unlock_bh(qdisc_lock(qdisc)); |
b0e1e646 | 790 | } |
1da177e4 LT |
791 | } |
792 | ||
4335cd2d | 793 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
794 | { |
795 | unsigned int i; | |
796 | ||
797 | for (i = 0; i < dev->num_tx_queues; i++) { | |
798 | struct netdev_queue *dev_queue; | |
7698b4fc | 799 | spinlock_t *root_lock; |
e2627c8c | 800 | struct Qdisc *q; |
e8a0464c DM |
801 | int val; |
802 | ||
803 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 804 | q = dev_queue->qdisc_sleeping; |
5fb66229 | 805 | root_lock = qdisc_lock(q); |
e8a0464c | 806 | |
4335cd2d | 807 | spin_lock_bh(root_lock); |
e8a0464c | 808 | |
bc135b23 | 809 | val = (qdisc_is_running(q) || |
b9a3b110 | 810 | test_bit(__QDISC_STATE_SCHED, &q->state)); |
e8a0464c | 811 | |
4335cd2d | 812 | spin_unlock_bh(root_lock); |
e8a0464c DM |
813 | |
814 | if (val) | |
815 | return true; | |
816 | } | |
817 | return false; | |
818 | } | |
819 | ||
3137663d ED |
820 | /** |
821 | * dev_deactivate_many - deactivate transmissions on several devices | |
822 | * @head: list of devices to deactivate | |
823 | * | |
824 | * This function returns only when all outstanding transmissions | |
825 | * have completed, unless all devices are in dismantle phase. | |
826 | */ | |
44345724 | 827 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 828 | { |
44345724 | 829 | struct net_device *dev; |
3137663d | 830 | bool sync_needed = false; |
41a23b07 | 831 | |
5cde2829 | 832 | list_for_each_entry(dev, head, close_list) { |
44345724 OP |
833 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, |
834 | &noop_qdisc); | |
835 | if (dev_ingress_queue(dev)) | |
836 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
837 | &noop_qdisc); | |
838 | ||
839 | dev_watchdog_down(dev); | |
3137663d | 840 | sync_needed |= !dev->dismantle; |
44345724 | 841 | } |
1da177e4 | 842 | |
3137663d ED |
843 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
844 | * This is avoided if all devices are in dismantle phase : | |
845 | * Caller will call synchronize_net() for us | |
846 | */ | |
847 | if (sync_needed) | |
848 | synchronize_net(); | |
1da177e4 | 849 | |
d4828d85 | 850 | /* Wait for outstanding qdisc_run calls. */ |
5cde2829 | 851 | list_for_each_entry(dev, head, close_list) |
44345724 OP |
852 | while (some_qdisc_is_busy(dev)) |
853 | yield(); | |
854 | } | |
855 | ||
856 | void dev_deactivate(struct net_device *dev) | |
857 | { | |
858 | LIST_HEAD(single); | |
859 | ||
5cde2829 | 860 | list_add(&dev->close_list, &single); |
44345724 | 861 | dev_deactivate_many(&single); |
5f04d506 | 862 | list_del(&single); |
1da177e4 | 863 | } |
b8970f0b | 864 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 865 | |
b0e1e646 DM |
866 | static void dev_init_scheduler_queue(struct net_device *dev, |
867 | struct netdev_queue *dev_queue, | |
e8a0464c | 868 | void *_qdisc) |
b0e1e646 | 869 | { |
e8a0464c DM |
870 | struct Qdisc *qdisc = _qdisc; |
871 | ||
b0e1e646 DM |
872 | dev_queue->qdisc = qdisc; |
873 | dev_queue->qdisc_sleeping = qdisc; | |
b0e1e646 DM |
874 | } |
875 | ||
1da177e4 LT |
876 | void dev_init_scheduler(struct net_device *dev) |
877 | { | |
af356afa | 878 | dev->qdisc = &noop_qdisc; |
e8a0464c | 879 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
880 | if (dev_ingress_queue(dev)) |
881 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 882 | |
b24b8a24 | 883 | setup_timer(&dev->watchdog_timer, dev_watchdog, (unsigned long)dev); |
1da177e4 LT |
884 | } |
885 | ||
e8a0464c DM |
886 | static void shutdown_scheduler_queue(struct net_device *dev, |
887 | struct netdev_queue *dev_queue, | |
888 | void *_qdisc_default) | |
1da177e4 | 889 | { |
b0e1e646 | 890 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 891 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
892 | |
893 | if (qdisc) { | |
f7a54c13 | 894 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 895 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 896 | |
1da177e4 | 897 | qdisc_destroy(qdisc); |
10297b99 | 898 | } |
b0e1e646 DM |
899 | } |
900 | ||
901 | void dev_shutdown(struct net_device *dev) | |
902 | { | |
e8a0464c | 903 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
904 | if (dev_ingress_queue(dev)) |
905 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
af356afa PM |
906 | qdisc_destroy(dev->qdisc); |
907 | dev->qdisc = &noop_qdisc; | |
908 | ||
547b792c | 909 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 910 | } |
292f1c7f | 911 | |
01cb71d2 | 912 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
3e1e3aae ED |
913 | const struct tc_ratespec *conf, |
914 | u64 rate64) | |
292f1c7f | 915 | { |
01cb71d2 ED |
916 | memset(r, 0, sizeof(*r)); |
917 | r->overhead = conf->overhead; | |
3e1e3aae | 918 | r->rate_bytes_ps = max_t(u64, conf->rate, rate64); |
8a8e3d84 | 919 | r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK); |
292f1c7f JP |
920 | r->mult = 1; |
921 | /* | |
130d3d68 ED |
922 | * The deal here is to replace a divide by a reciprocal one |
923 | * in fast path (a reciprocal divide is a multiply and a shift) | |
924 | * | |
925 | * Normal formula would be : | |
926 | * time_in_ns = (NSEC_PER_SEC * len) / rate_bps | |
927 | * | |
928 | * We compute mult/shift to use instead : | |
929 | * time_in_ns = (len * mult) >> shift; | |
930 | * | |
931 | * We try to get the highest possible mult value for accuracy, | |
932 | * but have to make sure no overflows will ever happen. | |
292f1c7f | 933 | */ |
130d3d68 ED |
934 | if (r->rate_bytes_ps > 0) { |
935 | u64 factor = NSEC_PER_SEC; | |
936 | ||
937 | for (;;) { | |
938 | r->mult = div64_u64(factor, r->rate_bytes_ps); | |
939 | if (r->mult & (1U << 31) || factor & (1ULL << 63)) | |
292f1c7f | 940 | break; |
130d3d68 ED |
941 | factor <<= 1; |
942 | r->shift++; | |
292f1c7f | 943 | } |
292f1c7f JP |
944 | } |
945 | } | |
946 | EXPORT_SYMBOL(psched_ratecfg_precompute); |