Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * net/sched/sch_generic.c Generic packet scheduler routines. | |
4 | * | |
1da177e4 LT |
5 | * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> |
6 | * Jamal Hadi Salim, <hadi@cyberus.ca> 990601 | |
7 | * - Ingress support | |
8 | */ | |
9 | ||
1da177e4 | 10 | #include <linux/bitops.h> |
1da177e4 LT |
11 | #include <linux/module.h> |
12 | #include <linux/types.h> | |
13 | #include <linux/kernel.h> | |
14 | #include <linux/sched.h> | |
15 | #include <linux/string.h> | |
1da177e4 | 16 | #include <linux/errno.h> |
1da177e4 LT |
17 | #include <linux/netdevice.h> |
18 | #include <linux/skbuff.h> | |
19 | #include <linux/rtnetlink.h> | |
20 | #include <linux/init.h> | |
21 | #include <linux/rcupdate.h> | |
22 | #include <linux/list.h> | |
5a0e3ad6 | 23 | #include <linux/slab.h> |
07ce76aa | 24 | #include <linux/if_vlan.h> |
c5ad119f | 25 | #include <linux/skb_array.h> |
32d3e51a | 26 | #include <linux/if_macvlan.h> |
292f1c7f | 27 | #include <net/sch_generic.h> |
1da177e4 | 28 | #include <net/pkt_sched.h> |
7fee226a | 29 | #include <net/dst.h> |
e543002f | 30 | #include <trace/events/qdisc.h> |
141b6b2a | 31 | #include <trace/events/net.h> |
f53c7239 | 32 | #include <net/xfrm.h> |
1da177e4 | 33 | |
34aedd3f | 34 | /* Qdisc to use by default */ |
35 | const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops; | |
36 | EXPORT_SYMBOL(default_qdisc_ops); | |
37 | ||
1da177e4 LT |
38 | /* Main transmission queue. */ |
39 | ||
0463d4ae | 40 | /* Modifications to data participating in scheduling must be protected with |
5fb66229 | 41 | * qdisc_lock(qdisc) spinlock. |
0463d4ae PM |
42 | * |
43 | * The idea is the following: | |
c7e4f3bb DM |
44 | * - enqueue, dequeue are serialized via qdisc root lock |
45 | * - ingress filtering is also serialized via qdisc root lock | |
0463d4ae | 46 | * - updates to tree and tree walking are only done under the rtnl mutex. |
1da177e4 | 47 | */ |
70e57d5e | 48 | |
b88dd52c ED |
49 | #define SKB_XOFF_MAGIC ((struct sk_buff *)1UL) |
50 | ||
70e57d5e JF |
51 | static inline struct sk_buff *__skb_dequeue_bad_txq(struct Qdisc *q) |
52 | { | |
53 | const struct netdev_queue *txq = q->dev_queue; | |
54 | spinlock_t *lock = NULL; | |
55 | struct sk_buff *skb; | |
56 | ||
57 | if (q->flags & TCQ_F_NOLOCK) { | |
58 | lock = qdisc_lock(q); | |
59 | spin_lock(lock); | |
60 | } | |
61 | ||
62 | skb = skb_peek(&q->skb_bad_txq); | |
63 | if (skb) { | |
64 | /* check the reason of requeuing without tx lock first */ | |
65 | txq = skb_get_tx_queue(txq->dev, skb); | |
66 | if (!netif_xmit_frozen_or_stopped(txq)) { | |
67 | skb = __skb_dequeue(&q->skb_bad_txq); | |
68 | if (qdisc_is_percpu_stats(q)) { | |
69 | qdisc_qstats_cpu_backlog_dec(q, skb); | |
73eb628d | 70 | qdisc_qstats_cpu_qlen_dec(q); |
70e57d5e JF |
71 | } else { |
72 | qdisc_qstats_backlog_dec(q, skb); | |
73 | q->q.qlen--; | |
74 | } | |
75 | } else { | |
b88dd52c | 76 | skb = SKB_XOFF_MAGIC; |
70e57d5e JF |
77 | } |
78 | } | |
79 | ||
80 | if (lock) | |
81 | spin_unlock(lock); | |
82 | ||
83 | return skb; | |
84 | } | |
85 | ||
86 | static inline struct sk_buff *qdisc_dequeue_skb_bad_txq(struct Qdisc *q) | |
87 | { | |
88 | struct sk_buff *skb = skb_peek(&q->skb_bad_txq); | |
89 | ||
90 | if (unlikely(skb)) | |
91 | skb = __skb_dequeue_bad_txq(q); | |
92 | ||
93 | return skb; | |
94 | } | |
95 | ||
96 | static inline void qdisc_enqueue_skb_bad_txq(struct Qdisc *q, | |
97 | struct sk_buff *skb) | |
98 | { | |
99 | spinlock_t *lock = NULL; | |
100 | ||
101 | if (q->flags & TCQ_F_NOLOCK) { | |
102 | lock = qdisc_lock(q); | |
103 | spin_lock(lock); | |
104 | } | |
105 | ||
106 | __skb_queue_tail(&q->skb_bad_txq, skb); | |
107 | ||
cce6294c ED |
108 | if (qdisc_is_percpu_stats(q)) { |
109 | qdisc_qstats_cpu_backlog_inc(q, skb); | |
73eb628d | 110 | qdisc_qstats_cpu_qlen_inc(q); |
cce6294c ED |
111 | } else { |
112 | qdisc_qstats_backlog_inc(q, skb); | |
113 | q->q.qlen++; | |
114 | } | |
115 | ||
70e57d5e JF |
116 | if (lock) |
117 | spin_unlock(lock); | |
118 | } | |
119 | ||
9c01c9f1 | 120 | static inline void dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q) |
c716a81a | 121 | { |
9c01c9f1 | 122 | spinlock_t *lock = NULL; |
9540d977 | 123 | |
9c01c9f1 PA |
124 | if (q->flags & TCQ_F_NOLOCK) { |
125 | lock = qdisc_lock(q); | |
126 | spin_lock(lock); | |
9540d977 | 127 | } |
c716a81a | 128 | |
9540d977 WY |
129 | while (skb) { |
130 | struct sk_buff *next = skb->next; | |
131 | ||
132 | __skb_queue_tail(&q->gso_skb, skb); | |
133 | ||
9c01c9f1 PA |
134 | /* it's still part of the queue */ |
135 | if (qdisc_is_percpu_stats(q)) { | |
136 | qdisc_qstats_cpu_requeues_inc(q); | |
137 | qdisc_qstats_cpu_backlog_inc(q, skb); | |
73eb628d | 138 | qdisc_qstats_cpu_qlen_inc(q); |
9c01c9f1 PA |
139 | } else { |
140 | q->qstats.requeues++; | |
141 | qdisc_qstats_backlog_inc(q, skb); | |
142 | q->q.qlen++; | |
143 | } | |
9540d977 WY |
144 | |
145 | skb = next; | |
146 | } | |
9c01c9f1 PA |
147 | if (lock) |
148 | spin_unlock(lock); | |
a53851e2 | 149 | __netif_schedule(q); |
a53851e2 JF |
150 | } |
151 | ||
55a93b3e ED |
152 | static void try_bulk_dequeue_skb(struct Qdisc *q, |
153 | struct sk_buff *skb, | |
b8358d70 JDB |
154 | const struct netdev_queue *txq, |
155 | int *packets) | |
5772e9a3 | 156 | { |
55a93b3e | 157 | int bytelimit = qdisc_avail_bulklimit(txq) - skb->len; |
5772e9a3 JDB |
158 | |
159 | while (bytelimit > 0) { | |
55a93b3e | 160 | struct sk_buff *nskb = q->dequeue(q); |
5772e9a3 | 161 | |
55a93b3e | 162 | if (!nskb) |
5772e9a3 JDB |
163 | break; |
164 | ||
55a93b3e ED |
165 | bytelimit -= nskb->len; /* covers GSO len */ |
166 | skb->next = nskb; | |
167 | skb = nskb; | |
b8358d70 | 168 | (*packets)++; /* GSO counts as one pkt */ |
5772e9a3 | 169 | } |
a8305bff | 170 | skb_mark_not_on_list(skb); |
5772e9a3 JDB |
171 | } |
172 | ||
4d202a0d ED |
173 | /* This variant of try_bulk_dequeue_skb() makes sure |
174 | * all skbs in the chain are for the same txq | |
175 | */ | |
176 | static void try_bulk_dequeue_skb_slow(struct Qdisc *q, | |
177 | struct sk_buff *skb, | |
178 | int *packets) | |
179 | { | |
180 | int mapping = skb_get_queue_mapping(skb); | |
181 | struct sk_buff *nskb; | |
182 | int cnt = 0; | |
183 | ||
184 | do { | |
185 | nskb = q->dequeue(q); | |
186 | if (!nskb) | |
187 | break; | |
188 | if (unlikely(skb_get_queue_mapping(nskb) != mapping)) { | |
70e57d5e | 189 | qdisc_enqueue_skb_bad_txq(q, nskb); |
4d202a0d ED |
190 | break; |
191 | } | |
192 | skb->next = nskb; | |
193 | skb = nskb; | |
194 | } while (++cnt < 8); | |
195 | (*packets) += cnt; | |
a8305bff | 196 | skb_mark_not_on_list(skb); |
4d202a0d ED |
197 | } |
198 | ||
5772e9a3 JDB |
199 | /* Note that dequeue_skb can possibly return a SKB list (via skb->next). |
200 | * A requeued skb (via q->gso_skb) can also be a SKB list. | |
201 | */ | |
b8358d70 JDB |
202 | static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate, |
203 | int *packets) | |
c716a81a | 204 | { |
1abbe139 | 205 | const struct netdev_queue *txq = q->dev_queue; |
fd8e8d1a | 206 | struct sk_buff *skb = NULL; |
554794de | 207 | |
b8358d70 | 208 | *packets = 1; |
a53851e2 JF |
209 | if (unlikely(!skb_queue_empty(&q->gso_skb))) { |
210 | spinlock_t *lock = NULL; | |
211 | ||
212 | if (q->flags & TCQ_F_NOLOCK) { | |
213 | lock = qdisc_lock(q); | |
214 | spin_lock(lock); | |
215 | } | |
216 | ||
217 | skb = skb_peek(&q->gso_skb); | |
218 | ||
219 | /* skb may be null if another cpu pulls gso_skb off in between | |
220 | * empty check and lock. | |
221 | */ | |
222 | if (!skb) { | |
223 | if (lock) | |
224 | spin_unlock(lock); | |
225 | goto validate; | |
226 | } | |
227 | ||
4d202a0d ED |
228 | /* skb in gso_skb were already validated */ |
229 | *validate = false; | |
f53c7239 SK |
230 | if (xfrm_offload(skb)) |
231 | *validate = true; | |
ebf05982 | 232 | /* check the reason of requeuing without tx lock first */ |
10c51b56 | 233 | txq = skb_get_tx_queue(txq->dev, skb); |
73466498 | 234 | if (!netif_xmit_frozen_or_stopped(txq)) { |
a53851e2 JF |
235 | skb = __skb_dequeue(&q->gso_skb); |
236 | if (qdisc_is_percpu_stats(q)) { | |
237 | qdisc_qstats_cpu_backlog_dec(q, skb); | |
73eb628d | 238 | qdisc_qstats_cpu_qlen_dec(q); |
a53851e2 JF |
239 | } else { |
240 | qdisc_qstats_backlog_dec(q, skb); | |
241 | q->q.qlen--; | |
242 | } | |
243 | } else { | |
ebf05982 | 244 | skb = NULL; |
a53851e2 JF |
245 | } |
246 | if (lock) | |
247 | spin_unlock(lock); | |
e543002f | 248 | goto trace; |
4d202a0d | 249 | } |
a53851e2 | 250 | validate: |
4d202a0d | 251 | *validate = true; |
fd8e8d1a JF |
252 | |
253 | if ((q->flags & TCQ_F_ONETXQUEUE) && | |
254 | netif_xmit_frozen_or_stopped(txq)) | |
255 | return skb; | |
256 | ||
70e57d5e | 257 | skb = qdisc_dequeue_skb_bad_txq(q); |
b88dd52c ED |
258 | if (unlikely(skb)) { |
259 | if (skb == SKB_XOFF_MAGIC) | |
260 | return NULL; | |
70e57d5e | 261 | goto bulk; |
b88dd52c | 262 | } |
fd8e8d1a | 263 | skb = q->dequeue(q); |
4d202a0d ED |
264 | if (skb) { |
265 | bulk: | |
266 | if (qdisc_may_bulk(q)) | |
267 | try_bulk_dequeue_skb(q, skb, txq, packets); | |
268 | else | |
269 | try_bulk_dequeue_skb_slow(q, skb, packets); | |
ebf05982 | 270 | } |
e543002f JDB |
271 | trace: |
272 | trace_qdisc_dequeue(q, txq, *packets, skb); | |
c716a81a JHS |
273 | return skb; |
274 | } | |
275 | ||
10297b99 | 276 | /* |
10770bc2 | 277 | * Transmit possibly several skbs, and handle the return status as |
f9eb8aea | 278 | * required. Owning running seqcount bit guarantees that |
10770bc2 | 279 | * only one CPU can execute this function. |
6c1361a6 KK |
280 | * |
281 | * Returns to the caller: | |
29b86cda JF |
282 | * false - hardware queue frozen backoff |
283 | * true - feel free to send more pkts | |
6c1361a6 | 284 | */ |
29b86cda JF |
285 | bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q, |
286 | struct net_device *dev, struct netdev_queue *txq, | |
287 | spinlock_t *root_lock, bool validate) | |
1da177e4 | 288 | { |
5f1a485d | 289 | int ret = NETDEV_TX_BUSY; |
f53c7239 | 290 | bool again = false; |
7698b4fc DM |
291 | |
292 | /* And release qdisc */ | |
6b3ba914 JF |
293 | if (root_lock) |
294 | spin_unlock(root_lock); | |
c716a81a | 295 | |
55a93b3e ED |
296 | /* Note that we validate skb (GSO, checksum, ...) outside of locks */ |
297 | if (validate) | |
f53c7239 SK |
298 | skb = validate_xmit_skb_list(skb, dev, &again); |
299 | ||
300 | #ifdef CONFIG_XFRM_OFFLOAD | |
301 | if (unlikely(again)) { | |
302 | if (root_lock) | |
303 | spin_lock(root_lock); | |
304 | ||
305 | dev_requeue_skb(skb, q); | |
306 | return false; | |
307 | } | |
308 | #endif | |
572a9d7b | 309 | |
3dcd493f | 310 | if (likely(skb)) { |
55a93b3e ED |
311 | HARD_TX_LOCK(dev, txq, smp_processor_id()); |
312 | if (!netif_xmit_frozen_or_stopped(txq)) | |
313 | skb = dev_hard_start_xmit(skb, dev, txq, &ret); | |
c716a81a | 314 | |
55a93b3e | 315 | HARD_TX_UNLOCK(dev, txq); |
3dcd493f | 316 | } else { |
6b3ba914 JF |
317 | if (root_lock) |
318 | spin_lock(root_lock); | |
29b86cda | 319 | return true; |
55a93b3e | 320 | } |
6b3ba914 JF |
321 | |
322 | if (root_lock) | |
323 | spin_lock(root_lock); | |
c716a81a | 324 | |
29b86cda | 325 | if (!dev_xmit_complete(ret)) { |
6c1361a6 | 326 | /* Driver returned NETDEV_TX_BUSY - requeue skb */ |
e87cc472 JP |
327 | if (unlikely(ret != NETDEV_TX_BUSY)) |
328 | net_warn_ratelimited("BUG %s code %d qlen %d\n", | |
329 | dev->name, ret, q->q.qlen); | |
6c1361a6 | 330 | |
29b86cda JF |
331 | dev_requeue_skb(skb, q); |
332 | return false; | |
6c1361a6 | 333 | } |
c716a81a | 334 | |
29b86cda | 335 | return true; |
1da177e4 LT |
336 | } |
337 | ||
bbd8a0d3 KK |
338 | /* |
339 | * NOTE: Called under qdisc_lock(q) with locally disabled BH. | |
340 | * | |
f9eb8aea | 341 | * running seqcount guarantees only one CPU can process |
bbd8a0d3 KK |
342 | * this qdisc at a time. qdisc_lock(q) serializes queue accesses for |
343 | * this queue. | |
344 | * | |
345 | * netif_tx_lock serializes accesses to device driver. | |
346 | * | |
347 | * qdisc_lock(q) and netif_tx_lock are mutually exclusive, | |
348 | * if one is grabbed, another must be free. | |
349 | * | |
350 | * Note, that this procedure can be called by a watchdog timer | |
351 | * | |
352 | * Returns to the caller: | |
353 | * 0 - queue is empty or throttled. | |
354 | * >0 - queue is not empty. | |
355 | * | |
356 | */ | |
29b86cda | 357 | static inline bool qdisc_restart(struct Qdisc *q, int *packets) |
bbd8a0d3 | 358 | { |
6b3ba914 | 359 | spinlock_t *root_lock = NULL; |
bbd8a0d3 KK |
360 | struct netdev_queue *txq; |
361 | struct net_device *dev; | |
bbd8a0d3 | 362 | struct sk_buff *skb; |
32f7b44d | 363 | bool validate; |
bbd8a0d3 KK |
364 | |
365 | /* Dequeue packet */ | |
b8358d70 | 366 | skb = dequeue_skb(q, &validate, packets); |
32f7b44d | 367 | if (unlikely(!skb)) |
29b86cda | 368 | return false; |
10c51b56 | 369 | |
32f7b44d | 370 | if (!(q->flags & TCQ_F_NOLOCK)) |
6b3ba914 JF |
371 | root_lock = qdisc_lock(q); |
372 | ||
bbd8a0d3 | 373 | dev = qdisc_dev(q); |
10c51b56 | 374 | txq = skb_get_tx_queue(dev, skb); |
bbd8a0d3 | 375 | |
32f7b44d | 376 | return sch_direct_xmit(skb, q, dev, txq, root_lock, validate); |
bbd8a0d3 KK |
377 | } |
378 | ||
37437bb2 | 379 | void __qdisc_run(struct Qdisc *q) |
48d83325 | 380 | { |
3d48b53f | 381 | int quota = dev_tx_weight; |
b8358d70 | 382 | int packets; |
2ba2506c | 383 | |
b8358d70 | 384 | while (qdisc_restart(q, &packets)) { |
b8358d70 | 385 | quota -= packets; |
b60fa1c5 | 386 | if (quota <= 0) { |
37437bb2 | 387 | __netif_schedule(q); |
d90df3ad | 388 | break; |
2ba2506c HX |
389 | } |
390 | } | |
48d83325 HX |
391 | } |
392 | ||
9d21493b ED |
393 | unsigned long dev_trans_start(struct net_device *dev) |
394 | { | |
07ce76aa | 395 | unsigned long val, res; |
9d21493b ED |
396 | unsigned int i; |
397 | ||
07ce76aa | 398 | if (is_vlan_dev(dev)) |
399 | dev = vlan_dev_real_dev(dev); | |
32d3e51a CD |
400 | else if (netif_is_macvlan(dev)) |
401 | dev = macvlan_dev_real_dev(dev); | |
9b36627a FW |
402 | res = netdev_get_tx_queue(dev, 0)->trans_start; |
403 | for (i = 1; i < dev->num_tx_queues; i++) { | |
9d21493b ED |
404 | val = netdev_get_tx_queue(dev, i)->trans_start; |
405 | if (val && time_after(val, res)) | |
406 | res = val; | |
407 | } | |
07ce76aa | 408 | |
9d21493b ED |
409 | return res; |
410 | } | |
411 | EXPORT_SYMBOL(dev_trans_start); | |
412 | ||
cdeabbb8 | 413 | static void dev_watchdog(struct timer_list *t) |
1da177e4 | 414 | { |
cdeabbb8 | 415 | struct net_device *dev = from_timer(dev, t, watchdog_timer); |
1da177e4 | 416 | |
932ff279 | 417 | netif_tx_lock(dev); |
e8a0464c | 418 | if (!qdisc_tx_is_noop(dev)) { |
1da177e4 LT |
419 | if (netif_device_present(dev) && |
420 | netif_running(dev) && | |
421 | netif_carrier_ok(dev)) { | |
9d21493b | 422 | int some_queue_timedout = 0; |
e8a0464c | 423 | unsigned int i; |
9d21493b | 424 | unsigned long trans_start; |
e8a0464c DM |
425 | |
426 | for (i = 0; i < dev->num_tx_queues; i++) { | |
427 | struct netdev_queue *txq; | |
428 | ||
429 | txq = netdev_get_tx_queue(dev, i); | |
9b36627a | 430 | trans_start = txq->trans_start; |
73466498 | 431 | if (netif_xmit_stopped(txq) && |
9d21493b ED |
432 | time_after(jiffies, (trans_start + |
433 | dev->watchdog_timeo))) { | |
434 | some_queue_timedout = 1; | |
ccf5ff69 | 435 | txq->trans_timeout++; |
e8a0464c DM |
436 | break; |
437 | } | |
438 | } | |
338f7566 | 439 | |
9d21493b | 440 | if (some_queue_timedout) { |
141b6b2a | 441 | trace_net_dev_xmit_timeout(dev, i); |
9d21493b | 442 | WARN_ONCE(1, KERN_INFO "NETDEV WATCHDOG: %s (%s): transmit queue %u timed out\n", |
3019de12 | 443 | dev->name, netdev_drivername(dev), i); |
d314774c | 444 | dev->netdev_ops->ndo_tx_timeout(dev); |
1da177e4 | 445 | } |
e8a0464c DM |
446 | if (!mod_timer(&dev->watchdog_timer, |
447 | round_jiffies(jiffies + | |
448 | dev->watchdog_timeo))) | |
1da177e4 LT |
449 | dev_hold(dev); |
450 | } | |
451 | } | |
932ff279 | 452 | netif_tx_unlock(dev); |
1da177e4 LT |
453 | |
454 | dev_put(dev); | |
455 | } | |
456 | ||
1da177e4 LT |
457 | void __netdev_watchdog_up(struct net_device *dev) |
458 | { | |
d314774c | 459 | if (dev->netdev_ops->ndo_tx_timeout) { |
1da177e4 LT |
460 | if (dev->watchdog_timeo <= 0) |
461 | dev->watchdog_timeo = 5*HZ; | |
60468d5b VP |
462 | if (!mod_timer(&dev->watchdog_timer, |
463 | round_jiffies(jiffies + dev->watchdog_timeo))) | |
1da177e4 LT |
464 | dev_hold(dev); |
465 | } | |
466 | } | |
467 | ||
468 | static void dev_watchdog_up(struct net_device *dev) | |
469 | { | |
1da177e4 | 470 | __netdev_watchdog_up(dev); |
1da177e4 LT |
471 | } |
472 | ||
473 | static void dev_watchdog_down(struct net_device *dev) | |
474 | { | |
932ff279 | 475 | netif_tx_lock_bh(dev); |
1da177e4 | 476 | if (del_timer(&dev->watchdog_timer)) |
15333061 | 477 | dev_put(dev); |
932ff279 | 478 | netif_tx_unlock_bh(dev); |
1da177e4 LT |
479 | } |
480 | ||
bea3348e SH |
481 | /** |
482 | * netif_carrier_on - set carrier | |
483 | * @dev: network device | |
484 | * | |
989723b0 | 485 | * Device has detected acquisition of carrier. |
bea3348e | 486 | */ |
0a242efc DV |
487 | void netif_carrier_on(struct net_device *dev) |
488 | { | |
bfaae0f0 | 489 | if (test_and_clear_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
b4730016 DM |
490 | if (dev->reg_state == NETREG_UNINITIALIZED) |
491 | return; | |
b2d3bcfa | 492 | atomic_inc(&dev->carrier_up_count); |
0a242efc | 493 | linkwatch_fire_event(dev); |
bfaae0f0 JG |
494 | if (netif_running(dev)) |
495 | __netdev_watchdog_up(dev); | |
496 | } | |
0a242efc | 497 | } |
62e3ba1b | 498 | EXPORT_SYMBOL(netif_carrier_on); |
0a242efc | 499 | |
bea3348e SH |
500 | /** |
501 | * netif_carrier_off - clear carrier | |
502 | * @dev: network device | |
503 | * | |
504 | * Device has detected loss of carrier. | |
505 | */ | |
0a242efc DV |
506 | void netif_carrier_off(struct net_device *dev) |
507 | { | |
b4730016 DM |
508 | if (!test_and_set_bit(__LINK_STATE_NOCARRIER, &dev->state)) { |
509 | if (dev->reg_state == NETREG_UNINITIALIZED) | |
510 | return; | |
b2d3bcfa | 511 | atomic_inc(&dev->carrier_down_count); |
0a242efc | 512 | linkwatch_fire_event(dev); |
b4730016 | 513 | } |
0a242efc | 514 | } |
62e3ba1b | 515 | EXPORT_SYMBOL(netif_carrier_off); |
0a242efc | 516 | |
1da177e4 LT |
517 | /* "NOOP" scheduler: the best scheduler, recommended for all interfaces |
518 | under all circumstances. It is difficult to invent anything faster or | |
519 | cheaper. | |
520 | */ | |
521 | ||
520ac30f ED |
522 | static int noop_enqueue(struct sk_buff *skb, struct Qdisc *qdisc, |
523 | struct sk_buff **to_free) | |
1da177e4 | 524 | { |
520ac30f | 525 | __qdisc_drop(skb, to_free); |
1da177e4 LT |
526 | return NET_XMIT_CN; |
527 | } | |
528 | ||
82d567c2 | 529 | static struct sk_buff *noop_dequeue(struct Qdisc *qdisc) |
1da177e4 LT |
530 | { |
531 | return NULL; | |
532 | } | |
533 | ||
20fea08b | 534 | struct Qdisc_ops noop_qdisc_ops __read_mostly = { |
1da177e4 LT |
535 | .id = "noop", |
536 | .priv_size = 0, | |
537 | .enqueue = noop_enqueue, | |
538 | .dequeue = noop_dequeue, | |
99c0db26 | 539 | .peek = noop_dequeue, |
1da177e4 LT |
540 | .owner = THIS_MODULE, |
541 | }; | |
542 | ||
7698b4fc | 543 | static struct netdev_queue noop_netdev_queue = { |
3b40bf4e | 544 | RCU_POINTER_INITIALIZER(qdisc, &noop_qdisc), |
9f3ffae0 | 545 | .qdisc_sleeping = &noop_qdisc, |
7698b4fc DM |
546 | }; |
547 | ||
1da177e4 LT |
548 | struct Qdisc noop_qdisc = { |
549 | .enqueue = noop_enqueue, | |
550 | .dequeue = noop_dequeue, | |
551 | .flags = TCQ_F_BUILTIN, | |
10297b99 | 552 | .ops = &noop_qdisc_ops, |
83874000 | 553 | .q.lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.q.lock), |
7698b4fc | 554 | .dev_queue = &noop_netdev_queue, |
f9eb8aea | 555 | .running = SEQCNT_ZERO(noop_qdisc.running), |
7b5edbc4 | 556 | .busylock = __SPIN_LOCK_UNLOCKED(noop_qdisc.busylock), |
f98ebd47 ED |
557 | .gso_skb = { |
558 | .next = (struct sk_buff *)&noop_qdisc.gso_skb, | |
559 | .prev = (struct sk_buff *)&noop_qdisc.gso_skb, | |
560 | .qlen = 0, | |
561 | .lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.gso_skb.lock), | |
562 | }, | |
563 | .skb_bad_txq = { | |
564 | .next = (struct sk_buff *)&noop_qdisc.skb_bad_txq, | |
565 | .prev = (struct sk_buff *)&noop_qdisc.skb_bad_txq, | |
566 | .qlen = 0, | |
567 | .lock = __SPIN_LOCK_UNLOCKED(noop_qdisc.skb_bad_txq.lock), | |
568 | }, | |
1da177e4 | 569 | }; |
62e3ba1b | 570 | EXPORT_SYMBOL(noop_qdisc); |
1da177e4 | 571 | |
e63d7dfd AA |
572 | static int noqueue_init(struct Qdisc *qdisc, struct nlattr *opt, |
573 | struct netlink_ext_ack *extack) | |
d66d6c31 PS |
574 | { |
575 | /* register_qdisc() assigns a default of noop_enqueue if unset, | |
576 | * but __dev_queue_xmit() treats noqueue only as such | |
577 | * if this is NULL - so clear it here. */ | |
578 | qdisc->enqueue = NULL; | |
579 | return 0; | |
580 | } | |
581 | ||
582 | struct Qdisc_ops noqueue_qdisc_ops __read_mostly = { | |
1da177e4 LT |
583 | .id = "noqueue", |
584 | .priv_size = 0, | |
d66d6c31 | 585 | .init = noqueue_init, |
1da177e4 LT |
586 | .enqueue = noop_enqueue, |
587 | .dequeue = noop_dequeue, | |
99c0db26 | 588 | .peek = noop_dequeue, |
1da177e4 LT |
589 | .owner = THIS_MODULE, |
590 | }; | |
591 | ||
cc7ec456 ED |
592 | static const u8 prio2band[TC_PRIO_MAX + 1] = { |
593 | 1, 2, 2, 2, 1, 2, 0, 0 , 1, 1, 1, 1, 1, 1, 1, 1 | |
594 | }; | |
d3678b46 DM |
595 | |
596 | /* 3-band FIFO queue: old style, but should be a bit faster than | |
597 | generic prio+fifo combination. | |
598 | */ | |
599 | ||
600 | #define PFIFO_FAST_BANDS 3 | |
601 | ||
fd3ae5e8 KK |
602 | /* |
603 | * Private data for a pfifo_fast scheduler containing: | |
c5ad119f | 604 | * - rings for priority bands |
fd3ae5e8 KK |
605 | */ |
606 | struct pfifo_fast_priv { | |
c5ad119f | 607 | struct skb_array q[PFIFO_FAST_BANDS]; |
fd3ae5e8 KK |
608 | }; |
609 | ||
c5ad119f JF |
610 | static inline struct skb_array *band2list(struct pfifo_fast_priv *priv, |
611 | int band) | |
d3678b46 | 612 | { |
c5ad119f | 613 | return &priv->q[band]; |
d3678b46 DM |
614 | } |
615 | ||
520ac30f ED |
616 | static int pfifo_fast_enqueue(struct sk_buff *skb, struct Qdisc *qdisc, |
617 | struct sk_buff **to_free) | |
321090e7 | 618 | { |
c5ad119f JF |
619 | int band = prio2band[skb->priority & TC_PRIO_MAX]; |
620 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); | |
621 | struct skb_array *q = band2list(priv, band); | |
cce6294c | 622 | unsigned int pkt_len = qdisc_pkt_len(skb); |
c5ad119f | 623 | int err; |
821d24ae | 624 | |
c5ad119f JF |
625 | err = skb_array_produce(q, skb); |
626 | ||
092e22e5 DC |
627 | if (unlikely(err)) { |
628 | if (qdisc_is_percpu_stats(qdisc)) | |
629 | return qdisc_drop_cpu(skb, qdisc, to_free); | |
630 | else | |
631 | return qdisc_drop(skb, qdisc, to_free); | |
632 | } | |
c5ad119f | 633 | |
8a53e616 | 634 | qdisc_update_stats_at_enqueue(qdisc, pkt_len); |
c5ad119f | 635 | return NET_XMIT_SUCCESS; |
1da177e4 LT |
636 | } |
637 | ||
cc7ec456 | 638 | static struct sk_buff *pfifo_fast_dequeue(struct Qdisc *qdisc) |
1da177e4 | 639 | { |
fd3ae5e8 | 640 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
c5ad119f JF |
641 | struct sk_buff *skb = NULL; |
642 | int band; | |
ec323368 | 643 | |
c5ad119f JF |
644 | for (band = 0; band < PFIFO_FAST_BANDS && !skb; band++) { |
645 | struct skb_array *q = band2list(priv, band); | |
fd3ae5e8 | 646 | |
c5ad119f JF |
647 | if (__skb_array_empty(q)) |
648 | continue; | |
fd3ae5e8 | 649 | |
021a17ed | 650 | skb = __skb_array_consume(q); |
c5ad119f JF |
651 | } |
652 | if (likely(skb)) { | |
8a53e616 | 653 | qdisc_update_stats_at_dequeue(qdisc, skb); |
28cff537 | 654 | } else { |
90b2be27 | 655 | WRITE_ONCE(qdisc->empty, true); |
d3678b46 | 656 | } |
f87a9c3d | 657 | |
c5ad119f | 658 | return skb; |
1da177e4 LT |
659 | } |
660 | ||
cc7ec456 | 661 | static struct sk_buff *pfifo_fast_peek(struct Qdisc *qdisc) |
99c0db26 | 662 | { |
fd3ae5e8 | 663 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
c5ad119f JF |
664 | struct sk_buff *skb = NULL; |
665 | int band; | |
fd3ae5e8 | 666 | |
c5ad119f JF |
667 | for (band = 0; band < PFIFO_FAST_BANDS && !skb; band++) { |
668 | struct skb_array *q = band2list(priv, band); | |
99c0db26 | 669 | |
c5ad119f | 670 | skb = __skb_array_peek(q); |
99c0db26 JP |
671 | } |
672 | ||
c5ad119f | 673 | return skb; |
99c0db26 JP |
674 | } |
675 | ||
cc7ec456 | 676 | static void pfifo_fast_reset(struct Qdisc *qdisc) |
1da177e4 | 677 | { |
c5ad119f | 678 | int i, band; |
fd3ae5e8 | 679 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
d3678b46 | 680 | |
c5ad119f JF |
681 | for (band = 0; band < PFIFO_FAST_BANDS; band++) { |
682 | struct skb_array *q = band2list(priv, band); | |
683 | struct sk_buff *skb; | |
d3678b46 | 684 | |
1df94c3c CW |
685 | /* NULL ring is possible if destroy path is due to a failed |
686 | * skb_array_init() in pfifo_fast_init() case. | |
687 | */ | |
688 | if (!q->ring.queue) | |
689 | continue; | |
690 | ||
021a17ed | 691 | while ((skb = __skb_array_consume(q)) != NULL) |
c5ad119f JF |
692 | kfree_skb(skb); |
693 | } | |
694 | ||
04d37cf4 DC |
695 | if (qdisc_is_percpu_stats(qdisc)) { |
696 | for_each_possible_cpu(i) { | |
697 | struct gnet_stats_queue *q; | |
c5ad119f | 698 | |
04d37cf4 DC |
699 | q = per_cpu_ptr(qdisc->cpu_qstats, i); |
700 | q->backlog = 0; | |
701 | q->qlen = 0; | |
702 | } | |
c5ad119f | 703 | } |
1da177e4 LT |
704 | } |
705 | ||
d3678b46 DM |
706 | static int pfifo_fast_dump(struct Qdisc *qdisc, struct sk_buff *skb) |
707 | { | |
708 | struct tc_prio_qopt opt = { .bands = PFIFO_FAST_BANDS }; | |
709 | ||
cc7ec456 | 710 | memcpy(&opt.priomap, prio2band, TC_PRIO_MAX + 1); |
1b34ec43 DM |
711 | if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt)) |
712 | goto nla_put_failure; | |
d3678b46 DM |
713 | return skb->len; |
714 | ||
715 | nla_put_failure: | |
716 | return -1; | |
717 | } | |
718 | ||
e63d7dfd AA |
719 | static int pfifo_fast_init(struct Qdisc *qdisc, struct nlattr *opt, |
720 | struct netlink_ext_ack *extack) | |
d3678b46 | 721 | { |
c5ad119f | 722 | unsigned int qlen = qdisc_dev(qdisc)->tx_queue_len; |
fd3ae5e8 | 723 | struct pfifo_fast_priv *priv = qdisc_priv(qdisc); |
c5ad119f JF |
724 | int prio; |
725 | ||
726 | /* guard against zero length rings */ | |
727 | if (!qlen) | |
728 | return -EINVAL; | |
d3678b46 | 729 | |
c5ad119f JF |
730 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) { |
731 | struct skb_array *q = band2list(priv, prio); | |
732 | int err; | |
733 | ||
734 | err = skb_array_init(q, qlen, GFP_KERNEL); | |
735 | if (err) | |
736 | return -ENOMEM; | |
737 | } | |
d3678b46 | 738 | |
23624935 ED |
739 | /* Can by-pass the queue discipline */ |
740 | qdisc->flags |= TCQ_F_CAN_BYPASS; | |
d3678b46 DM |
741 | return 0; |
742 | } | |
743 | ||
c5ad119f JF |
744 | static void pfifo_fast_destroy(struct Qdisc *sch) |
745 | { | |
746 | struct pfifo_fast_priv *priv = qdisc_priv(sch); | |
747 | int prio; | |
748 | ||
749 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) { | |
750 | struct skb_array *q = band2list(priv, prio); | |
751 | ||
752 | /* NULL ring is possible if destroy path is due to a failed | |
753 | * skb_array_init() in pfifo_fast_init() case. | |
754 | */ | |
1df94c3c | 755 | if (!q->ring.queue) |
c5ad119f JF |
756 | continue; |
757 | /* Destroy ring but no need to kfree_skb because a call to | |
758 | * pfifo_fast_reset() has already done that work. | |
759 | */ | |
760 | ptr_ring_cleanup(&q->ring, NULL); | |
761 | } | |
762 | } | |
763 | ||
7007ba63 CW |
764 | static int pfifo_fast_change_tx_queue_len(struct Qdisc *sch, |
765 | unsigned int new_len) | |
766 | { | |
767 | struct pfifo_fast_priv *priv = qdisc_priv(sch); | |
768 | struct skb_array *bands[PFIFO_FAST_BANDS]; | |
769 | int prio; | |
770 | ||
771 | for (prio = 0; prio < PFIFO_FAST_BANDS; prio++) { | |
772 | struct skb_array *q = band2list(priv, prio); | |
773 | ||
774 | bands[prio] = q; | |
775 | } | |
776 | ||
777 | return skb_array_resize_multiple(bands, PFIFO_FAST_BANDS, new_len, | |
778 | GFP_KERNEL); | |
779 | } | |
780 | ||
6ec1c69a | 781 | struct Qdisc_ops pfifo_fast_ops __read_mostly = { |
d3678b46 | 782 | .id = "pfifo_fast", |
fd3ae5e8 | 783 | .priv_size = sizeof(struct pfifo_fast_priv), |
d3678b46 DM |
784 | .enqueue = pfifo_fast_enqueue, |
785 | .dequeue = pfifo_fast_dequeue, | |
99c0db26 | 786 | .peek = pfifo_fast_peek, |
d3678b46 | 787 | .init = pfifo_fast_init, |
c5ad119f | 788 | .destroy = pfifo_fast_destroy, |
d3678b46 DM |
789 | .reset = pfifo_fast_reset, |
790 | .dump = pfifo_fast_dump, | |
7007ba63 | 791 | .change_tx_queue_len = pfifo_fast_change_tx_queue_len, |
1da177e4 | 792 | .owner = THIS_MODULE, |
c5ad119f | 793 | .static_flags = TCQ_F_NOLOCK | TCQ_F_CPUSTATS, |
1da177e4 | 794 | }; |
1f27cde3 | 795 | EXPORT_SYMBOL(pfifo_fast_ops); |
1da177e4 | 796 | |
5ce2d488 | 797 | struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue, |
d0bd684d AA |
798 | const struct Qdisc_ops *ops, |
799 | struct netlink_ext_ack *extack) | |
1da177e4 LT |
800 | { |
801 | void *p; | |
802 | struct Qdisc *sch; | |
d276055c | 803 | unsigned int size = QDISC_ALIGN(sizeof(*sch)) + ops->priv_size; |
3d54b82f | 804 | int err = -ENOBUFS; |
26aa0459 JSP |
805 | struct net_device *dev; |
806 | ||
807 | if (!dev_queue) { | |
d0bd684d | 808 | NL_SET_ERR_MSG(extack, "No device queue given"); |
26aa0459 JSP |
809 | err = -EINVAL; |
810 | goto errout; | |
811 | } | |
1da177e4 | 812 | |
26aa0459 | 813 | dev = dev_queue->dev; |
f2cd2d3e ED |
814 | p = kzalloc_node(size, GFP_KERNEL, |
815 | netdev_queue_numa_node_read(dev_queue)); | |
816 | ||
1da177e4 | 817 | if (!p) |
3d54b82f | 818 | goto errout; |
3d54b82f | 819 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); |
d276055c ED |
820 | /* if we got non aligned memory, ask more and do alignment ourself */ |
821 | if (sch != p) { | |
822 | kfree(p); | |
823 | p = kzalloc_node(size + QDISC_ALIGNTO - 1, GFP_KERNEL, | |
824 | netdev_queue_numa_node_read(dev_queue)); | |
825 | if (!p) | |
826 | goto errout; | |
827 | sch = (struct Qdisc *) QDISC_ALIGN((unsigned long) p); | |
828 | sch->padded = (char *) sch - (char *) p; | |
829 | } | |
a53851e2 | 830 | __skb_queue_head_init(&sch->gso_skb); |
70e57d5e | 831 | __skb_queue_head_init(&sch->skb_bad_txq); |
48da34b7 FW |
832 | qdisc_skb_head_init(&sch->q); |
833 | spin_lock_init(&sch->q.lock); | |
23d3b8bf | 834 | |
d59f5ffa JF |
835 | if (ops->static_flags & TCQ_F_CPUSTATS) { |
836 | sch->cpu_bstats = | |
837 | netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu); | |
838 | if (!sch->cpu_bstats) | |
839 | goto errout1; | |
840 | ||
841 | sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue); | |
842 | if (!sch->cpu_qstats) { | |
843 | free_percpu(sch->cpu_bstats); | |
844 | goto errout1; | |
845 | } | |
846 | } | |
847 | ||
79640a4c | 848 | spin_lock_init(&sch->busylock); |
96009c7d PA |
849 | /* seqlock has the same scope of busylock, for NOLOCK qdisc */ |
850 | spin_lock_init(&sch->seqlock); | |
f9eb8aea | 851 | seqcount_init(&sch->running); |
f9eb8aea | 852 | |
1da177e4 | 853 | sch->ops = ops; |
d59f5ffa | 854 | sch->flags = ops->static_flags; |
1da177e4 LT |
855 | sch->enqueue = ops->enqueue; |
856 | sch->dequeue = ops->dequeue; | |
bb949fbd | 857 | sch->dev_queue = dev_queue; |
28cff537 | 858 | sch->empty = true; |
23d3b8bf | 859 | dev_hold(dev); |
7b936405 | 860 | refcount_set(&sch->refcnt, 1); |
3d54b82f | 861 | |
ab92d68f TY |
862 | if (sch != &noop_qdisc) { |
863 | lockdep_set_class(&sch->busylock, &dev->qdisc_tx_busylock_key); | |
864 | lockdep_set_class(&sch->seqlock, &dev->qdisc_tx_busylock_key); | |
865 | lockdep_set_class(&sch->running, &dev->qdisc_running_key); | |
866 | } | |
867 | ||
3d54b82f | 868 | return sch; |
d59f5ffa JF |
869 | errout1: |
870 | kfree(p); | |
3d54b82f | 871 | errout: |
01e123d7 | 872 | return ERR_PTR(err); |
3d54b82f TG |
873 | } |
874 | ||
3511c913 | 875 | struct Qdisc *qdisc_create_dflt(struct netdev_queue *dev_queue, |
d2a7f269 | 876 | const struct Qdisc_ops *ops, |
a38a9882 AA |
877 | unsigned int parentid, |
878 | struct netlink_ext_ack *extack) | |
3d54b82f TG |
879 | { |
880 | struct Qdisc *sch; | |
10297b99 | 881 | |
a38a9882 AA |
882 | if (!try_module_get(ops->owner)) { |
883 | NL_SET_ERR_MSG(extack, "Failed to increase module reference counter"); | |
166ee5b8 | 884 | return NULL; |
a38a9882 | 885 | } |
6da7c8fc | 886 | |
a38a9882 | 887 | sch = qdisc_alloc(dev_queue, ops, extack); |
166ee5b8 ED |
888 | if (IS_ERR(sch)) { |
889 | module_put(ops->owner); | |
890 | return NULL; | |
891 | } | |
9f9afec4 | 892 | sch->parent = parentid; |
3d54b82f | 893 | |
a38a9882 | 894 | if (!ops->init || ops->init(sch, NULL, extack) == 0) |
1da177e4 LT |
895 | return sch; |
896 | ||
86bd446b | 897 | qdisc_put(sch); |
1da177e4 LT |
898 | return NULL; |
899 | } | |
62e3ba1b | 900 | EXPORT_SYMBOL(qdisc_create_dflt); |
1da177e4 | 901 | |
5fb66229 | 902 | /* Under qdisc_lock(qdisc) and BH! */ |
1da177e4 LT |
903 | |
904 | void qdisc_reset(struct Qdisc *qdisc) | |
905 | { | |
20fea08b | 906 | const struct Qdisc_ops *ops = qdisc->ops; |
a53851e2 | 907 | struct sk_buff *skb, *tmp; |
1da177e4 LT |
908 | |
909 | if (ops->reset) | |
910 | ops->reset(qdisc); | |
67305ebc | 911 | |
a53851e2 JF |
912 | skb_queue_walk_safe(&qdisc->gso_skb, skb, tmp) { |
913 | __skb_unlink(skb, &qdisc->gso_skb); | |
914 | kfree_skb_list(skb); | |
bbd8a0d3 | 915 | } |
a53851e2 | 916 | |
70e57d5e JF |
917 | skb_queue_walk_safe(&qdisc->skb_bad_txq, skb, tmp) { |
918 | __skb_unlink(skb, &qdisc->skb_bad_txq); | |
919 | kfree_skb_list(skb); | |
920 | } | |
921 | ||
4d202a0d | 922 | qdisc->q.qlen = 0; |
c8e18129 | 923 | qdisc->qstats.backlog = 0; |
1da177e4 | 924 | } |
62e3ba1b | 925 | EXPORT_SYMBOL(qdisc_reset); |
1da177e4 | 926 | |
81d947e2 | 927 | void qdisc_free(struct Qdisc *qdisc) |
5d944c64 | 928 | { |
73c20a8b | 929 | if (qdisc_is_percpu_stats(qdisc)) { |
22e0f8b9 | 930 | free_percpu(qdisc->cpu_bstats); |
73c20a8b JF |
931 | free_percpu(qdisc->cpu_qstats); |
932 | } | |
22e0f8b9 | 933 | |
5d944c64 ED |
934 | kfree((char *) qdisc - qdisc->padded); |
935 | } | |
936 | ||
5362700c | 937 | static void qdisc_free_cb(struct rcu_head *head) |
3a7d0d07 VB |
938 | { |
939 | struct Qdisc *q = container_of(head, struct Qdisc, rcu); | |
940 | ||
941 | qdisc_free(q); | |
942 | } | |
943 | ||
86bd446b | 944 | static void qdisc_destroy(struct Qdisc *qdisc) |
1da177e4 | 945 | { |
8a34c5dc | 946 | const struct Qdisc_ops *ops = qdisc->ops; |
a53851e2 | 947 | struct sk_buff *skb, *tmp; |
8a34c5dc | 948 | |
3a682fbd | 949 | #ifdef CONFIG_NET_SCHED |
59cc1f61 | 950 | qdisc_hash_del(qdisc); |
f6e0b239 | 951 | |
a2da570d | 952 | qdisc_put_stab(rtnl_dereference(qdisc->stab)); |
3a682fbd | 953 | #endif |
1c0d32fd | 954 | gen_kill_estimator(&qdisc->rate_est); |
8a34c5dc DM |
955 | if (ops->reset) |
956 | ops->reset(qdisc); | |
957 | if (ops->destroy) | |
958 | ops->destroy(qdisc); | |
959 | ||
960 | module_put(ops->owner); | |
961 | dev_put(qdisc_dev(qdisc)); | |
962 | ||
a53851e2 JF |
963 | skb_queue_walk_safe(&qdisc->gso_skb, skb, tmp) { |
964 | __skb_unlink(skb, &qdisc->gso_skb); | |
965 | kfree_skb_list(skb); | |
966 | } | |
967 | ||
70e57d5e JF |
968 | skb_queue_walk_safe(&qdisc->skb_bad_txq, skb, tmp) { |
969 | __skb_unlink(skb, &qdisc->skb_bad_txq); | |
970 | kfree_skb_list(skb); | |
971 | } | |
972 | ||
3a7d0d07 | 973 | call_rcu(&qdisc->rcu, qdisc_free_cb); |
1da177e4 | 974 | } |
86bd446b VB |
975 | |
976 | void qdisc_put(struct Qdisc *qdisc) | |
977 | { | |
6efb971b CW |
978 | if (!qdisc) |
979 | return; | |
980 | ||
86bd446b VB |
981 | if (qdisc->flags & TCQ_F_BUILTIN || |
982 | !refcount_dec_and_test(&qdisc->refcnt)) | |
983 | return; | |
984 | ||
985 | qdisc_destroy(qdisc); | |
986 | } | |
987 | EXPORT_SYMBOL(qdisc_put); | |
1da177e4 | 988 | |
3a7d0d07 VB |
989 | /* Version of qdisc_put() that is called with rtnl mutex unlocked. |
990 | * Intended to be used as optimization, this function only takes rtnl lock if | |
991 | * qdisc reference counter reached zero. | |
992 | */ | |
993 | ||
994 | void qdisc_put_unlocked(struct Qdisc *qdisc) | |
995 | { | |
996 | if (qdisc->flags & TCQ_F_BUILTIN || | |
997 | !refcount_dec_and_rtnl_lock(&qdisc->refcnt)) | |
998 | return; | |
999 | ||
1000 | qdisc_destroy(qdisc); | |
1001 | rtnl_unlock(); | |
1002 | } | |
1003 | EXPORT_SYMBOL(qdisc_put_unlocked); | |
1004 | ||
589983cd PM |
1005 | /* Attach toplevel qdisc to device queue. */ |
1006 | struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
1007 | struct Qdisc *qdisc) | |
1008 | { | |
1009 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
1010 | spinlock_t *root_lock; | |
1011 | ||
1012 | root_lock = qdisc_lock(oqdisc); | |
1013 | spin_lock_bh(root_lock); | |
1014 | ||
589983cd PM |
1015 | /* ... and graft new one */ |
1016 | if (qdisc == NULL) | |
1017 | qdisc = &noop_qdisc; | |
1018 | dev_queue->qdisc_sleeping = qdisc; | |
1019 | rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc); | |
1020 | ||
1021 | spin_unlock_bh(root_lock); | |
1022 | ||
1023 | return oqdisc; | |
1024 | } | |
b8970f0b | 1025 | EXPORT_SYMBOL(dev_graft_qdisc); |
589983cd | 1026 | |
e8a0464c DM |
1027 | static void attach_one_default_qdisc(struct net_device *dev, |
1028 | struct netdev_queue *dev_queue, | |
1029 | void *_unused) | |
1030 | { | |
3e692f21 PS |
1031 | struct Qdisc *qdisc; |
1032 | const struct Qdisc_ops *ops = default_qdisc_ops; | |
e8a0464c | 1033 | |
3e692f21 PS |
1034 | if (dev->priv_flags & IFF_NO_QUEUE) |
1035 | ops = &noqueue_qdisc_ops; | |
546b85bb VP |
1036 | else if(dev->type == ARPHRD_CAN) |
1037 | ops = &pfifo_fast_ops; | |
3e692f21 | 1038 | |
a38a9882 | 1039 | qdisc = qdisc_create_dflt(dev_queue, ops, TC_H_ROOT, NULL); |
3e692f21 PS |
1040 | if (!qdisc) { |
1041 | netdev_info(dev, "activation failed\n"); | |
1042 | return; | |
e8a0464c | 1043 | } |
3e692f21 | 1044 | if (!netif_is_multiqueue(dev)) |
4eaf3b84 | 1045 | qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT; |
e8a0464c DM |
1046 | dev_queue->qdisc_sleeping = qdisc; |
1047 | } | |
1048 | ||
6ec1c69a DM |
1049 | static void attach_default_qdiscs(struct net_device *dev) |
1050 | { | |
1051 | struct netdev_queue *txq; | |
1052 | struct Qdisc *qdisc; | |
1053 | ||
1054 | txq = netdev_get_tx_queue(dev, 0); | |
1055 | ||
4b469955 | 1056 | if (!netif_is_multiqueue(dev) || |
4b469955 | 1057 | dev->priv_flags & IFF_NO_QUEUE) { |
6ec1c69a DM |
1058 | netdev_for_each_tx_queue(dev, attach_one_default_qdisc, NULL); |
1059 | dev->qdisc = txq->qdisc_sleeping; | |
551143d8 | 1060 | qdisc_refcount_inc(dev->qdisc); |
6ec1c69a | 1061 | } else { |
a38a9882 | 1062 | qdisc = qdisc_create_dflt(txq, &mq_qdisc_ops, TC_H_ROOT, NULL); |
6ec1c69a | 1063 | if (qdisc) { |
6ec1c69a | 1064 | dev->qdisc = qdisc; |
e57a784d | 1065 | qdisc->ops->attach(qdisc); |
6ec1c69a DM |
1066 | } |
1067 | } | |
59cc1f61 | 1068 | #ifdef CONFIG_NET_SCHED |
92f91706 | 1069 | if (dev->qdisc != &noop_qdisc) |
49b49971 | 1070 | qdisc_hash_add(dev->qdisc, false); |
59cc1f61 | 1071 | #endif |
6ec1c69a DM |
1072 | } |
1073 | ||
e8a0464c DM |
1074 | static void transition_one_qdisc(struct net_device *dev, |
1075 | struct netdev_queue *dev_queue, | |
1076 | void *_need_watchdog) | |
1077 | { | |
83874000 | 1078 | struct Qdisc *new_qdisc = dev_queue->qdisc_sleeping; |
e8a0464c DM |
1079 | int *need_watchdog_p = _need_watchdog; |
1080 | ||
a9312ae8 DM |
1081 | if (!(new_qdisc->flags & TCQ_F_BUILTIN)) |
1082 | clear_bit(__QDISC_STATE_DEACTIVATED, &new_qdisc->state); | |
1083 | ||
83874000 | 1084 | rcu_assign_pointer(dev_queue->qdisc, new_qdisc); |
3e692f21 | 1085 | if (need_watchdog_p) { |
9d21493b | 1086 | dev_queue->trans_start = 0; |
e8a0464c | 1087 | *need_watchdog_p = 1; |
9d21493b | 1088 | } |
e8a0464c DM |
1089 | } |
1090 | ||
1da177e4 LT |
1091 | void dev_activate(struct net_device *dev) |
1092 | { | |
e8a0464c | 1093 | int need_watchdog; |
b0e1e646 | 1094 | |
1da177e4 | 1095 | /* No queueing discipline is attached to device; |
6da7c8fc | 1096 | * create default one for devices, which need queueing |
1097 | * and noqueue_qdisc for virtual interfaces | |
1da177e4 LT |
1098 | */ |
1099 | ||
6ec1c69a DM |
1100 | if (dev->qdisc == &noop_qdisc) |
1101 | attach_default_qdiscs(dev); | |
af356afa | 1102 | |
cacaddf5 TC |
1103 | if (!netif_carrier_ok(dev)) |
1104 | /* Delay activation until next carrier-on event */ | |
1105 | return; | |
1106 | ||
e8a0464c DM |
1107 | need_watchdog = 0; |
1108 | netdev_for_each_tx_queue(dev, transition_one_qdisc, &need_watchdog); | |
24824a09 ED |
1109 | if (dev_ingress_queue(dev)) |
1110 | transition_one_qdisc(dev, dev_ingress_queue(dev), NULL); | |
e8a0464c DM |
1111 | |
1112 | if (need_watchdog) { | |
860e9538 | 1113 | netif_trans_update(dev); |
1da177e4 LT |
1114 | dev_watchdog_up(dev); |
1115 | } | |
b0e1e646 | 1116 | } |
b8970f0b | 1117 | EXPORT_SYMBOL(dev_activate); |
b0e1e646 | 1118 | |
e8a0464c DM |
1119 | static void dev_deactivate_queue(struct net_device *dev, |
1120 | struct netdev_queue *dev_queue, | |
1121 | void *_qdisc_default) | |
b0e1e646 | 1122 | { |
e8a0464c | 1123 | struct Qdisc *qdisc_default = _qdisc_default; |
970565bb | 1124 | struct Qdisc *qdisc; |
970565bb | 1125 | |
46e5da40 | 1126 | qdisc = rtnl_dereference(dev_queue->qdisc); |
b0e1e646 | 1127 | if (qdisc) { |
96009c7d PA |
1128 | bool nolock = qdisc->flags & TCQ_F_NOLOCK; |
1129 | ||
1130 | if (nolock) | |
1131 | spin_lock_bh(&qdisc->seqlock); | |
83874000 DM |
1132 | spin_lock_bh(qdisc_lock(qdisc)); |
1133 | ||
a9312ae8 DM |
1134 | if (!(qdisc->flags & TCQ_F_BUILTIN)) |
1135 | set_bit(__QDISC_STATE_DEACTIVATED, &qdisc->state); | |
1136 | ||
f7a54c13 | 1137 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 1138 | qdisc_reset(qdisc); |
d3b753db | 1139 | |
83874000 | 1140 | spin_unlock_bh(qdisc_lock(qdisc)); |
96009c7d PA |
1141 | if (nolock) |
1142 | spin_unlock_bh(&qdisc->seqlock); | |
b0e1e646 | 1143 | } |
1da177e4 LT |
1144 | } |
1145 | ||
4335cd2d | 1146 | static bool some_qdisc_is_busy(struct net_device *dev) |
e8a0464c DM |
1147 | { |
1148 | unsigned int i; | |
1149 | ||
1150 | for (i = 0; i < dev->num_tx_queues; i++) { | |
1151 | struct netdev_queue *dev_queue; | |
7698b4fc | 1152 | spinlock_t *root_lock; |
e2627c8c | 1153 | struct Qdisc *q; |
e8a0464c DM |
1154 | int val; |
1155 | ||
1156 | dev_queue = netdev_get_tx_queue(dev, i); | |
b9a3b110 | 1157 | q = dev_queue->qdisc_sleeping; |
e8a0464c | 1158 | |
32f7b44d PA |
1159 | root_lock = qdisc_lock(q); |
1160 | spin_lock_bh(root_lock); | |
e8a0464c | 1161 | |
32f7b44d PA |
1162 | val = (qdisc_is_running(q) || |
1163 | test_bit(__QDISC_STATE_SCHED, &q->state)); | |
e8a0464c | 1164 | |
32f7b44d | 1165 | spin_unlock_bh(root_lock); |
e8a0464c DM |
1166 | |
1167 | if (val) | |
1168 | return true; | |
1169 | } | |
1170 | return false; | |
1171 | } | |
1172 | ||
7bbde83b JF |
1173 | static void dev_qdisc_reset(struct net_device *dev, |
1174 | struct netdev_queue *dev_queue, | |
1175 | void *none) | |
1176 | { | |
1177 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; | |
1178 | ||
1179 | if (qdisc) | |
1180 | qdisc_reset(qdisc); | |
1181 | } | |
1182 | ||
3137663d ED |
1183 | /** |
1184 | * dev_deactivate_many - deactivate transmissions on several devices | |
1185 | * @head: list of devices to deactivate | |
1186 | * | |
1187 | * This function returns only when all outstanding transmissions | |
1188 | * have completed, unless all devices are in dismantle phase. | |
1189 | */ | |
44345724 | 1190 | void dev_deactivate_many(struct list_head *head) |
1da177e4 | 1191 | { |
44345724 | 1192 | struct net_device *dev; |
41a23b07 | 1193 | |
5cde2829 | 1194 | list_for_each_entry(dev, head, close_list) { |
44345724 OP |
1195 | netdev_for_each_tx_queue(dev, dev_deactivate_queue, |
1196 | &noop_qdisc); | |
1197 | if (dev_ingress_queue(dev)) | |
1198 | dev_deactivate_queue(dev, dev_ingress_queue(dev), | |
1199 | &noop_qdisc); | |
1200 | ||
1201 | dev_watchdog_down(dev); | |
1202 | } | |
1da177e4 | 1203 | |
3137663d ED |
1204 | /* Wait for outstanding qdisc-less dev_queue_xmit calls. |
1205 | * This is avoided if all devices are in dismantle phase : | |
1206 | * Caller will call synchronize_net() for us | |
1207 | */ | |
7bbde83b | 1208 | synchronize_net(); |
1da177e4 | 1209 | |
d4828d85 | 1210 | /* Wait for outstanding qdisc_run calls. */ |
7bbde83b | 1211 | list_for_each_entry(dev, head, close_list) { |
4eab421b MKB |
1212 | while (some_qdisc_is_busy(dev)) { |
1213 | /* wait_event() would avoid this sleep-loop but would | |
1214 | * require expensive checks in the fast paths of packet | |
1215 | * processing which isn't worth it. | |
1216 | */ | |
1217 | schedule_timeout_uninterruptible(1); | |
1218 | } | |
7bbde83b JF |
1219 | /* The new qdisc is assigned at this point so we can safely |
1220 | * unwind stale skb lists and qdisc statistics | |
1221 | */ | |
1222 | netdev_for_each_tx_queue(dev, dev_qdisc_reset, NULL); | |
1223 | if (dev_ingress_queue(dev)) | |
1224 | dev_qdisc_reset(dev, dev_ingress_queue(dev), NULL); | |
1225 | } | |
44345724 OP |
1226 | } |
1227 | ||
1228 | void dev_deactivate(struct net_device *dev) | |
1229 | { | |
1230 | LIST_HEAD(single); | |
1231 | ||
5cde2829 | 1232 | list_add(&dev->close_list, &single); |
44345724 | 1233 | dev_deactivate_many(&single); |
5f04d506 | 1234 | list_del(&single); |
1da177e4 | 1235 | } |
b8970f0b | 1236 | EXPORT_SYMBOL(dev_deactivate); |
1da177e4 | 1237 | |
48bfd55e CW |
1238 | static int qdisc_change_tx_queue_len(struct net_device *dev, |
1239 | struct netdev_queue *dev_queue) | |
1240 | { | |
1241 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; | |
1242 | const struct Qdisc_ops *ops = qdisc->ops; | |
1243 | ||
1244 | if (ops->change_tx_queue_len) | |
1245 | return ops->change_tx_queue_len(qdisc, dev->tx_queue_len); | |
1246 | return 0; | |
1247 | } | |
1248 | ||
1249 | int dev_qdisc_change_tx_queue_len(struct net_device *dev) | |
1250 | { | |
1251 | bool up = dev->flags & IFF_UP; | |
1252 | unsigned int i; | |
1253 | int ret = 0; | |
1254 | ||
1255 | if (up) | |
1256 | dev_deactivate(dev); | |
1257 | ||
1258 | for (i = 0; i < dev->num_tx_queues; i++) { | |
1259 | ret = qdisc_change_tx_queue_len(dev, &dev->_tx[i]); | |
1260 | ||
1261 | /* TODO: revert changes on a partial failure */ | |
1262 | if (ret) | |
1263 | break; | |
1264 | } | |
1265 | ||
1266 | if (up) | |
1267 | dev_activate(dev); | |
1268 | return ret; | |
1269 | } | |
1270 | ||
b0e1e646 DM |
1271 | static void dev_init_scheduler_queue(struct net_device *dev, |
1272 | struct netdev_queue *dev_queue, | |
e8a0464c | 1273 | void *_qdisc) |
b0e1e646 | 1274 | { |
e8a0464c DM |
1275 | struct Qdisc *qdisc = _qdisc; |
1276 | ||
46e5da40 | 1277 | rcu_assign_pointer(dev_queue->qdisc, qdisc); |
b0e1e646 | 1278 | dev_queue->qdisc_sleeping = qdisc; |
b0e1e646 DM |
1279 | } |
1280 | ||
1da177e4 LT |
1281 | void dev_init_scheduler(struct net_device *dev) |
1282 | { | |
af356afa | 1283 | dev->qdisc = &noop_qdisc; |
e8a0464c | 1284 | netdev_for_each_tx_queue(dev, dev_init_scheduler_queue, &noop_qdisc); |
24824a09 ED |
1285 | if (dev_ingress_queue(dev)) |
1286 | dev_init_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
1da177e4 | 1287 | |
cdeabbb8 | 1288 | timer_setup(&dev->watchdog_timer, dev_watchdog, 0); |
1da177e4 LT |
1289 | } |
1290 | ||
e8a0464c DM |
1291 | static void shutdown_scheduler_queue(struct net_device *dev, |
1292 | struct netdev_queue *dev_queue, | |
1293 | void *_qdisc_default) | |
1da177e4 | 1294 | { |
b0e1e646 | 1295 | struct Qdisc *qdisc = dev_queue->qdisc_sleeping; |
e8a0464c | 1296 | struct Qdisc *qdisc_default = _qdisc_default; |
b0e1e646 DM |
1297 | |
1298 | if (qdisc) { | |
f7a54c13 | 1299 | rcu_assign_pointer(dev_queue->qdisc, qdisc_default); |
b0e1e646 | 1300 | dev_queue->qdisc_sleeping = qdisc_default; |
1da177e4 | 1301 | |
86bd446b | 1302 | qdisc_put(qdisc); |
10297b99 | 1303 | } |
b0e1e646 DM |
1304 | } |
1305 | ||
1306 | void dev_shutdown(struct net_device *dev) | |
1307 | { | |
e8a0464c | 1308 | netdev_for_each_tx_queue(dev, shutdown_scheduler_queue, &noop_qdisc); |
24824a09 ED |
1309 | if (dev_ingress_queue(dev)) |
1310 | shutdown_scheduler_queue(dev, dev_ingress_queue(dev), &noop_qdisc); | |
86bd446b | 1311 | qdisc_put(dev->qdisc); |
af356afa PM |
1312 | dev->qdisc = &noop_qdisc; |
1313 | ||
547b792c | 1314 | WARN_ON(timer_pending(&dev->watchdog_timer)); |
1da177e4 | 1315 | } |
292f1c7f | 1316 | |
01cb71d2 | 1317 | void psched_ratecfg_precompute(struct psched_ratecfg *r, |
3e1e3aae ED |
1318 | const struct tc_ratespec *conf, |
1319 | u64 rate64) | |
292f1c7f | 1320 | { |
01cb71d2 ED |
1321 | memset(r, 0, sizeof(*r)); |
1322 | r->overhead = conf->overhead; | |
3e1e3aae | 1323 | r->rate_bytes_ps = max_t(u64, conf->rate, rate64); |
8a8e3d84 | 1324 | r->linklayer = (conf->linklayer & TC_LINKLAYER_MASK); |
292f1c7f JP |
1325 | r->mult = 1; |
1326 | /* | |
130d3d68 ED |
1327 | * The deal here is to replace a divide by a reciprocal one |
1328 | * in fast path (a reciprocal divide is a multiply and a shift) | |
1329 | * | |
1330 | * Normal formula would be : | |
1331 | * time_in_ns = (NSEC_PER_SEC * len) / rate_bps | |
1332 | * | |
1333 | * We compute mult/shift to use instead : | |
1334 | * time_in_ns = (len * mult) >> shift; | |
1335 | * | |
1336 | * We try to get the highest possible mult value for accuracy, | |
1337 | * but have to make sure no overflows will ever happen. | |
292f1c7f | 1338 | */ |
130d3d68 ED |
1339 | if (r->rate_bytes_ps > 0) { |
1340 | u64 factor = NSEC_PER_SEC; | |
1341 | ||
1342 | for (;;) { | |
1343 | r->mult = div64_u64(factor, r->rate_bytes_ps); | |
1344 | if (r->mult & (1U << 31) || factor & (1ULL << 63)) | |
292f1c7f | 1345 | break; |
130d3d68 ED |
1346 | factor <<= 1; |
1347 | r->shift++; | |
292f1c7f | 1348 | } |
292f1c7f JP |
1349 | } |
1350 | } | |
1351 | EXPORT_SYMBOL(psched_ratecfg_precompute); | |
46209401 JP |
1352 | |
1353 | static void mini_qdisc_rcu_func(struct rcu_head *head) | |
1354 | { | |
1355 | } | |
1356 | ||
1357 | void mini_qdisc_pair_swap(struct mini_Qdisc_pair *miniqp, | |
1358 | struct tcf_proto *tp_head) | |
1359 | { | |
ed76f5ed VB |
1360 | /* Protected with chain0->filter_chain_lock. |
1361 | * Can't access chain directly because tp_head can be NULL. | |
1362 | */ | |
1363 | struct mini_Qdisc *miniq_old = | |
1364 | rcu_dereference_protected(*miniqp->p_miniq, 1); | |
46209401 JP |
1365 | struct mini_Qdisc *miniq; |
1366 | ||
1367 | if (!tp_head) { | |
1368 | RCU_INIT_POINTER(*miniqp->p_miniq, NULL); | |
b2fb01f4 | 1369 | /* Wait for flying RCU callback before it is freed. */ |
ae0e3349 | 1370 | rcu_barrier(); |
46209401 JP |
1371 | return; |
1372 | } | |
1373 | ||
1374 | miniq = !miniq_old || miniq_old == &miniqp->miniq2 ? | |
1375 | &miniqp->miniq1 : &miniqp->miniq2; | |
1376 | ||
1377 | /* We need to make sure that readers won't see the miniq | |
ae0e3349 | 1378 | * we are about to modify. So wait until previous call_rcu callback |
46209401 JP |
1379 | * is done. |
1380 | */ | |
ae0e3349 | 1381 | rcu_barrier(); |
46209401 JP |
1382 | miniq->filter_list = tp_head; |
1383 | rcu_assign_pointer(*miniqp->p_miniq, miniq); | |
1384 | ||
1385 | if (miniq_old) | |
b2fb01f4 | 1386 | /* This is counterpart of the rcu barriers above. We need to |
46209401 JP |
1387 | * block potential new user of miniq_old until all readers |
1388 | * are not seeing it. | |
1389 | */ | |
ae0e3349 | 1390 | call_rcu(&miniq_old->rcu, mini_qdisc_rcu_func); |
46209401 JP |
1391 | } |
1392 | EXPORT_SYMBOL(mini_qdisc_pair_swap); | |
1393 | ||
1394 | void mini_qdisc_pair_init(struct mini_Qdisc_pair *miniqp, struct Qdisc *qdisc, | |
1395 | struct mini_Qdisc __rcu **p_miniq) | |
1396 | { | |
1397 | miniqp->miniq1.cpu_bstats = qdisc->cpu_bstats; | |
1398 | miniqp->miniq1.cpu_qstats = qdisc->cpu_qstats; | |
1399 | miniqp->miniq2.cpu_bstats = qdisc->cpu_bstats; | |
1400 | miniqp->miniq2.cpu_qstats = qdisc->cpu_qstats; | |
1401 | miniqp->p_miniq = p_miniq; | |
1402 | } | |
1403 | EXPORT_SYMBOL(mini_qdisc_pair_init); |