]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * net/sched/sch_api.c Packet scheduler API. | |
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 | * | |
11 | * Fixes: | |
12 | * | |
13 | * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired. | |
14 | * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support | |
15 | * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support | |
16 | */ | |
17 | ||
18 | #include <linux/module.h> | |
19 | #include <linux/types.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/string.h> | |
22 | #include <linux/errno.h> | |
23 | #include <linux/skbuff.h> | |
24 | #include <linux/init.h> | |
25 | #include <linux/proc_fs.h> | |
26 | #include <linux/seq_file.h> | |
27 | #include <linux/kmod.h> | |
28 | #include <linux/list.h> | |
29 | #include <linux/hrtimer.h> | |
30 | #include <linux/lockdep.h> | |
31 | ||
32 | #include <net/net_namespace.h> | |
33 | #include <net/sock.h> | |
34 | #include <net/netlink.h> | |
35 | #include <net/pkt_sched.h> | |
36 | ||
37 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid, | |
38 | struct Qdisc *old, struct Qdisc *new); | |
39 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
40 | struct Qdisc *q, unsigned long cl, int event); | |
41 | ||
42 | /* | |
43 | ||
44 | Short review. | |
45 | ------------- | |
46 | ||
47 | This file consists of two interrelated parts: | |
48 | ||
49 | 1. queueing disciplines manager frontend. | |
50 | 2. traffic classes manager frontend. | |
51 | ||
52 | Generally, queueing discipline ("qdisc") is a black box, | |
53 | which is able to enqueue packets and to dequeue them (when | |
54 | device is ready to send something) in order and at times | |
55 | determined by algorithm hidden in it. | |
56 | ||
57 | qdisc's are divided to two categories: | |
58 | - "queues", which have no internal structure visible from outside. | |
59 | - "schedulers", which split all the packets to "traffic classes", | |
60 | using "packet classifiers" (look at cls_api.c) | |
61 | ||
62 | In turn, classes may have child qdiscs (as rule, queues) | |
63 | attached to them etc. etc. etc. | |
64 | ||
65 | The goal of the routines in this file is to translate | |
66 | information supplied by user in the form of handles | |
67 | to more intelligible for kernel form, to make some sanity | |
68 | checks and part of work, which is common to all qdiscs | |
69 | and to provide rtnetlink notifications. | |
70 | ||
71 | All real intelligent work is done inside qdisc modules. | |
72 | ||
73 | ||
74 | ||
75 | Every discipline has two major routines: enqueue and dequeue. | |
76 | ||
77 | ---dequeue | |
78 | ||
79 | dequeue usually returns a skb to send. It is allowed to return NULL, | |
80 | but it does not mean that queue is empty, it just means that | |
81 | discipline does not want to send anything this time. | |
82 | Queue is really empty if q->q.qlen == 0. | |
83 | For complicated disciplines with multiple queues q->q is not | |
84 | real packet queue, but however q->q.qlen must be valid. | |
85 | ||
86 | ---enqueue | |
87 | ||
88 | enqueue returns 0, if packet was enqueued successfully. | |
89 | If packet (this one or another one) was dropped, it returns | |
90 | not zero error code. | |
91 | NET_XMIT_DROP - this packet dropped | |
92 | Expected action: do not backoff, but wait until queue will clear. | |
93 | NET_XMIT_CN - probably this packet enqueued, but another one dropped. | |
94 | Expected action: backoff or ignore | |
95 | NET_XMIT_POLICED - dropped by police. | |
96 | Expected action: backoff or error to real-time apps. | |
97 | ||
98 | Auxiliary routines: | |
99 | ||
100 | ---requeue | |
101 | ||
102 | requeues once dequeued packet. It is used for non-standard or | |
103 | just buggy devices, which can defer output even if netif_queue_stopped()=0. | |
104 | ||
105 | ---reset | |
106 | ||
107 | returns qdisc to initial state: purge all buffers, clear all | |
108 | timers, counters (except for statistics) etc. | |
109 | ||
110 | ---init | |
111 | ||
112 | initializes newly created qdisc. | |
113 | ||
114 | ---destroy | |
115 | ||
116 | destroys resources allocated by init and during lifetime of qdisc. | |
117 | ||
118 | ---change | |
119 | ||
120 | changes qdisc parameters. | |
121 | */ | |
122 | ||
123 | /* Protects list of registered TC modules. It is pure SMP lock. */ | |
124 | static DEFINE_RWLOCK(qdisc_mod_lock); | |
125 | ||
126 | ||
127 | /************************************************ | |
128 | * Queueing disciplines manipulation. * | |
129 | ************************************************/ | |
130 | ||
131 | ||
132 | /* The list of all installed queueing disciplines. */ | |
133 | ||
134 | static struct Qdisc_ops *qdisc_base; | |
135 | ||
136 | /* Register/uregister queueing discipline */ | |
137 | ||
138 | int register_qdisc(struct Qdisc_ops *qops) | |
139 | { | |
140 | struct Qdisc_ops *q, **qp; | |
141 | int rc = -EEXIST; | |
142 | ||
143 | write_lock(&qdisc_mod_lock); | |
144 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) | |
145 | if (!strcmp(qops->id, q->id)) | |
146 | goto out; | |
147 | ||
148 | if (qops->enqueue == NULL) | |
149 | qops->enqueue = noop_qdisc_ops.enqueue; | |
150 | if (qops->requeue == NULL) | |
151 | qops->requeue = noop_qdisc_ops.requeue; | |
152 | if (qops->dequeue == NULL) | |
153 | qops->dequeue = noop_qdisc_ops.dequeue; | |
154 | ||
155 | qops->next = NULL; | |
156 | *qp = qops; | |
157 | rc = 0; | |
158 | out: | |
159 | write_unlock(&qdisc_mod_lock); | |
160 | return rc; | |
161 | } | |
162 | EXPORT_SYMBOL(register_qdisc); | |
163 | ||
164 | int unregister_qdisc(struct Qdisc_ops *qops) | |
165 | { | |
166 | struct Qdisc_ops *q, **qp; | |
167 | int err = -ENOENT; | |
168 | ||
169 | write_lock(&qdisc_mod_lock); | |
170 | for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next) | |
171 | if (q == qops) | |
172 | break; | |
173 | if (q) { | |
174 | *qp = q->next; | |
175 | q->next = NULL; | |
176 | err = 0; | |
177 | } | |
178 | write_unlock(&qdisc_mod_lock); | |
179 | return err; | |
180 | } | |
181 | EXPORT_SYMBOL(unregister_qdisc); | |
182 | ||
183 | /* We know handle. Find qdisc among all qdisc's attached to device | |
184 | (root qdisc, all its children, children of children etc.) | |
185 | */ | |
186 | ||
187 | struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) | |
188 | { | |
189 | struct Qdisc *q; | |
190 | ||
191 | if (!(root->flags & TCQ_F_BUILTIN) && | |
192 | root->handle == handle) | |
193 | return root; | |
194 | ||
195 | list_for_each_entry(q, &root->list, list) { | |
196 | if (q->handle == handle) | |
197 | return q; | |
198 | } | |
199 | return NULL; | |
200 | } | |
201 | ||
202 | struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) | |
203 | { | |
204 | unsigned int i; | |
205 | ||
206 | for (i = 0; i < dev->num_tx_queues; i++) { | |
207 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
208 | struct Qdisc *q, *txq_root = txq->qdisc_sleeping; | |
209 | ||
210 | q = qdisc_match_from_root(txq_root, handle); | |
211 | if (q) | |
212 | return q; | |
213 | } | |
214 | return qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle); | |
215 | } | |
216 | ||
217 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) | |
218 | { | |
219 | unsigned long cl; | |
220 | struct Qdisc *leaf; | |
221 | const struct Qdisc_class_ops *cops = p->ops->cl_ops; | |
222 | ||
223 | if (cops == NULL) | |
224 | return NULL; | |
225 | cl = cops->get(p, classid); | |
226 | ||
227 | if (cl == 0) | |
228 | return NULL; | |
229 | leaf = cops->leaf(p, cl); | |
230 | cops->put(p, cl); | |
231 | return leaf; | |
232 | } | |
233 | ||
234 | /* Find queueing discipline by name */ | |
235 | ||
236 | static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) | |
237 | { | |
238 | struct Qdisc_ops *q = NULL; | |
239 | ||
240 | if (kind) { | |
241 | read_lock(&qdisc_mod_lock); | |
242 | for (q = qdisc_base; q; q = q->next) { | |
243 | if (nla_strcmp(kind, q->id) == 0) { | |
244 | if (!try_module_get(q->owner)) | |
245 | q = NULL; | |
246 | break; | |
247 | } | |
248 | } | |
249 | read_unlock(&qdisc_mod_lock); | |
250 | } | |
251 | return q; | |
252 | } | |
253 | ||
254 | static struct qdisc_rate_table *qdisc_rtab_list; | |
255 | ||
256 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab) | |
257 | { | |
258 | struct qdisc_rate_table *rtab; | |
259 | ||
260 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { | |
261 | if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { | |
262 | rtab->refcnt++; | |
263 | return rtab; | |
264 | } | |
265 | } | |
266 | ||
267 | if (tab == NULL || r->rate == 0 || r->cell_log == 0 || | |
268 | nla_len(tab) != TC_RTAB_SIZE) | |
269 | return NULL; | |
270 | ||
271 | rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); | |
272 | if (rtab) { | |
273 | rtab->rate = *r; | |
274 | rtab->refcnt = 1; | |
275 | memcpy(rtab->data, nla_data(tab), 1024); | |
276 | rtab->next = qdisc_rtab_list; | |
277 | qdisc_rtab_list = rtab; | |
278 | } | |
279 | return rtab; | |
280 | } | |
281 | EXPORT_SYMBOL(qdisc_get_rtab); | |
282 | ||
283 | void qdisc_put_rtab(struct qdisc_rate_table *tab) | |
284 | { | |
285 | struct qdisc_rate_table *rtab, **rtabp; | |
286 | ||
287 | if (!tab || --tab->refcnt) | |
288 | return; | |
289 | ||
290 | for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { | |
291 | if (rtab == tab) { | |
292 | *rtabp = rtab->next; | |
293 | kfree(rtab); | |
294 | return; | |
295 | } | |
296 | } | |
297 | } | |
298 | EXPORT_SYMBOL(qdisc_put_rtab); | |
299 | ||
300 | static LIST_HEAD(qdisc_stab_list); | |
301 | static DEFINE_SPINLOCK(qdisc_stab_lock); | |
302 | ||
303 | static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { | |
304 | [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, | |
305 | [TCA_STAB_DATA] = { .type = NLA_BINARY }, | |
306 | }; | |
307 | ||
308 | static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) | |
309 | { | |
310 | struct nlattr *tb[TCA_STAB_MAX + 1]; | |
311 | struct qdisc_size_table *stab; | |
312 | struct tc_sizespec *s; | |
313 | unsigned int tsize = 0; | |
314 | u16 *tab = NULL; | |
315 | int err; | |
316 | ||
317 | err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy); | |
318 | if (err < 0) | |
319 | return ERR_PTR(err); | |
320 | if (!tb[TCA_STAB_BASE]) | |
321 | return ERR_PTR(-EINVAL); | |
322 | ||
323 | s = nla_data(tb[TCA_STAB_BASE]); | |
324 | ||
325 | if (s->tsize > 0) { | |
326 | if (!tb[TCA_STAB_DATA]) | |
327 | return ERR_PTR(-EINVAL); | |
328 | tab = nla_data(tb[TCA_STAB_DATA]); | |
329 | tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); | |
330 | } | |
331 | ||
332 | if (!s || tsize != s->tsize || (!tab && tsize > 0)) | |
333 | return ERR_PTR(-EINVAL); | |
334 | ||
335 | spin_lock_bh(&qdisc_stab_lock); | |
336 | ||
337 | list_for_each_entry(stab, &qdisc_stab_list, list) { | |
338 | if (memcmp(&stab->szopts, s, sizeof(*s))) | |
339 | continue; | |
340 | if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) | |
341 | continue; | |
342 | stab->refcnt++; | |
343 | spin_unlock_bh(&qdisc_stab_lock); | |
344 | return stab; | |
345 | } | |
346 | ||
347 | spin_unlock_bh(&qdisc_stab_lock); | |
348 | ||
349 | stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); | |
350 | if (!stab) | |
351 | return ERR_PTR(-ENOMEM); | |
352 | ||
353 | stab->refcnt = 1; | |
354 | stab->szopts = *s; | |
355 | if (tsize > 0) | |
356 | memcpy(stab->data, tab, tsize * sizeof(u16)); | |
357 | ||
358 | spin_lock_bh(&qdisc_stab_lock); | |
359 | list_add_tail(&stab->list, &qdisc_stab_list); | |
360 | spin_unlock_bh(&qdisc_stab_lock); | |
361 | ||
362 | return stab; | |
363 | } | |
364 | ||
365 | void qdisc_put_stab(struct qdisc_size_table *tab) | |
366 | { | |
367 | if (!tab) | |
368 | return; | |
369 | ||
370 | spin_lock_bh(&qdisc_stab_lock); | |
371 | ||
372 | if (--tab->refcnt == 0) { | |
373 | list_del(&tab->list); | |
374 | kfree(tab); | |
375 | } | |
376 | ||
377 | spin_unlock_bh(&qdisc_stab_lock); | |
378 | } | |
379 | EXPORT_SYMBOL(qdisc_put_stab); | |
380 | ||
381 | static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) | |
382 | { | |
383 | struct nlattr *nest; | |
384 | ||
385 | nest = nla_nest_start(skb, TCA_STAB); | |
386 | NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts); | |
387 | nla_nest_end(skb, nest); | |
388 | ||
389 | return skb->len; | |
390 | ||
391 | nla_put_failure: | |
392 | return -1; | |
393 | } | |
394 | ||
395 | void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab) | |
396 | { | |
397 | int pkt_len, slot; | |
398 | ||
399 | pkt_len = skb->len + stab->szopts.overhead; | |
400 | if (unlikely(!stab->szopts.tsize)) | |
401 | goto out; | |
402 | ||
403 | slot = pkt_len + stab->szopts.cell_align; | |
404 | if (unlikely(slot < 0)) | |
405 | slot = 0; | |
406 | ||
407 | slot >>= stab->szopts.cell_log; | |
408 | if (likely(slot < stab->szopts.tsize)) | |
409 | pkt_len = stab->data[slot]; | |
410 | else | |
411 | pkt_len = stab->data[stab->szopts.tsize - 1] * | |
412 | (slot / stab->szopts.tsize) + | |
413 | stab->data[slot % stab->szopts.tsize]; | |
414 | ||
415 | pkt_len <<= stab->szopts.size_log; | |
416 | out: | |
417 | if (unlikely(pkt_len < 1)) | |
418 | pkt_len = 1; | |
419 | qdisc_skb_cb(skb)->pkt_len = pkt_len; | |
420 | } | |
421 | EXPORT_SYMBOL(qdisc_calculate_pkt_len); | |
422 | ||
423 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) | |
424 | { | |
425 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, | |
426 | timer); | |
427 | ||
428 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
429 | smp_wmb(); | |
430 | __netif_schedule(qdisc_root(wd->qdisc)); | |
431 | ||
432 | return HRTIMER_NORESTART; | |
433 | } | |
434 | ||
435 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) | |
436 | { | |
437 | hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
438 | wd->timer.function = qdisc_watchdog; | |
439 | wd->qdisc = qdisc; | |
440 | } | |
441 | EXPORT_SYMBOL(qdisc_watchdog_init); | |
442 | ||
443 | void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) | |
444 | { | |
445 | ktime_t time; | |
446 | ||
447 | wd->qdisc->flags |= TCQ_F_THROTTLED; | |
448 | time = ktime_set(0, 0); | |
449 | time = ktime_add_ns(time, PSCHED_US2NS(expires)); | |
450 | hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); | |
451 | } | |
452 | EXPORT_SYMBOL(qdisc_watchdog_schedule); | |
453 | ||
454 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) | |
455 | { | |
456 | hrtimer_cancel(&wd->timer); | |
457 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
458 | } | |
459 | EXPORT_SYMBOL(qdisc_watchdog_cancel); | |
460 | ||
461 | static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) | |
462 | { | |
463 | unsigned int size = n * sizeof(struct hlist_head), i; | |
464 | struct hlist_head *h; | |
465 | ||
466 | if (size <= PAGE_SIZE) | |
467 | h = kmalloc(size, GFP_KERNEL); | |
468 | else | |
469 | h = (struct hlist_head *) | |
470 | __get_free_pages(GFP_KERNEL, get_order(size)); | |
471 | ||
472 | if (h != NULL) { | |
473 | for (i = 0; i < n; i++) | |
474 | INIT_HLIST_HEAD(&h[i]); | |
475 | } | |
476 | return h; | |
477 | } | |
478 | ||
479 | static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) | |
480 | { | |
481 | unsigned int size = n * sizeof(struct hlist_head); | |
482 | ||
483 | if (size <= PAGE_SIZE) | |
484 | kfree(h); | |
485 | else | |
486 | free_pages((unsigned long)h, get_order(size)); | |
487 | } | |
488 | ||
489 | void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) | |
490 | { | |
491 | struct Qdisc_class_common *cl; | |
492 | struct hlist_node *n, *next; | |
493 | struct hlist_head *nhash, *ohash; | |
494 | unsigned int nsize, nmask, osize; | |
495 | unsigned int i, h; | |
496 | ||
497 | /* Rehash when load factor exceeds 0.75 */ | |
498 | if (clhash->hashelems * 4 <= clhash->hashsize * 3) | |
499 | return; | |
500 | nsize = clhash->hashsize * 2; | |
501 | nmask = nsize - 1; | |
502 | nhash = qdisc_class_hash_alloc(nsize); | |
503 | if (nhash == NULL) | |
504 | return; | |
505 | ||
506 | ohash = clhash->hash; | |
507 | osize = clhash->hashsize; | |
508 | ||
509 | sch_tree_lock(sch); | |
510 | for (i = 0; i < osize; i++) { | |
511 | hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) { | |
512 | h = qdisc_class_hash(cl->classid, nmask); | |
513 | hlist_add_head(&cl->hnode, &nhash[h]); | |
514 | } | |
515 | } | |
516 | clhash->hash = nhash; | |
517 | clhash->hashsize = nsize; | |
518 | clhash->hashmask = nmask; | |
519 | sch_tree_unlock(sch); | |
520 | ||
521 | qdisc_class_hash_free(ohash, osize); | |
522 | } | |
523 | EXPORT_SYMBOL(qdisc_class_hash_grow); | |
524 | ||
525 | int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) | |
526 | { | |
527 | unsigned int size = 4; | |
528 | ||
529 | clhash->hash = qdisc_class_hash_alloc(size); | |
530 | if (clhash->hash == NULL) | |
531 | return -ENOMEM; | |
532 | clhash->hashsize = size; | |
533 | clhash->hashmask = size - 1; | |
534 | clhash->hashelems = 0; | |
535 | return 0; | |
536 | } | |
537 | EXPORT_SYMBOL(qdisc_class_hash_init); | |
538 | ||
539 | void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) | |
540 | { | |
541 | qdisc_class_hash_free(clhash->hash, clhash->hashsize); | |
542 | } | |
543 | EXPORT_SYMBOL(qdisc_class_hash_destroy); | |
544 | ||
545 | void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, | |
546 | struct Qdisc_class_common *cl) | |
547 | { | |
548 | unsigned int h; | |
549 | ||
550 | INIT_HLIST_NODE(&cl->hnode); | |
551 | h = qdisc_class_hash(cl->classid, clhash->hashmask); | |
552 | hlist_add_head(&cl->hnode, &clhash->hash[h]); | |
553 | clhash->hashelems++; | |
554 | } | |
555 | EXPORT_SYMBOL(qdisc_class_hash_insert); | |
556 | ||
557 | void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, | |
558 | struct Qdisc_class_common *cl) | |
559 | { | |
560 | hlist_del(&cl->hnode); | |
561 | clhash->hashelems--; | |
562 | } | |
563 | EXPORT_SYMBOL(qdisc_class_hash_remove); | |
564 | ||
565 | /* Allocate an unique handle from space managed by kernel */ | |
566 | ||
567 | static u32 qdisc_alloc_handle(struct net_device *dev) | |
568 | { | |
569 | int i = 0x10000; | |
570 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0); | |
571 | ||
572 | do { | |
573 | autohandle += TC_H_MAKE(0x10000U, 0); | |
574 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) | |
575 | autohandle = TC_H_MAKE(0x80000000U, 0); | |
576 | } while (qdisc_lookup(dev, autohandle) && --i > 0); | |
577 | ||
578 | return i>0 ? autohandle : 0; | |
579 | } | |
580 | ||
581 | /* Attach toplevel qdisc to device queue. */ | |
582 | ||
583 | static struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue, | |
584 | struct Qdisc *qdisc) | |
585 | { | |
586 | struct Qdisc *oqdisc = dev_queue->qdisc_sleeping; | |
587 | spinlock_t *root_lock; | |
588 | ||
589 | root_lock = qdisc_root_lock(oqdisc); | |
590 | spin_lock_bh(root_lock); | |
591 | ||
592 | /* Prune old scheduler */ | |
593 | if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1) | |
594 | qdisc_reset(oqdisc); | |
595 | ||
596 | /* ... and graft new one */ | |
597 | if (qdisc == NULL) | |
598 | qdisc = &noop_qdisc; | |
599 | dev_queue->qdisc_sleeping = qdisc; | |
600 | dev_queue->qdisc = &noop_qdisc; | |
601 | ||
602 | spin_unlock_bh(root_lock); | |
603 | ||
604 | return oqdisc; | |
605 | } | |
606 | ||
607 | void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) | |
608 | { | |
609 | const struct Qdisc_class_ops *cops; | |
610 | unsigned long cl; | |
611 | u32 parentid; | |
612 | ||
613 | if (n == 0) | |
614 | return; | |
615 | while ((parentid = sch->parent)) { | |
616 | if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) | |
617 | return; | |
618 | ||
619 | sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); | |
620 | if (sch == NULL) { | |
621 | WARN_ON(parentid != TC_H_ROOT); | |
622 | return; | |
623 | } | |
624 | cops = sch->ops->cl_ops; | |
625 | if (cops->qlen_notify) { | |
626 | cl = cops->get(sch, parentid); | |
627 | cops->qlen_notify(sch, cl); | |
628 | cops->put(sch, cl); | |
629 | } | |
630 | sch->q.qlen -= n; | |
631 | } | |
632 | } | |
633 | EXPORT_SYMBOL(qdisc_tree_decrease_qlen); | |
634 | ||
635 | static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid, | |
636 | struct Qdisc *old, struct Qdisc *new) | |
637 | { | |
638 | if (new || old) | |
639 | qdisc_notify(skb, n, clid, old, new); | |
640 | ||
641 | if (old) | |
642 | qdisc_destroy(old); | |
643 | } | |
644 | ||
645 | /* Graft qdisc "new" to class "classid" of qdisc "parent" or | |
646 | * to device "dev". | |
647 | * | |
648 | * When appropriate send a netlink notification using 'skb' | |
649 | * and "n". | |
650 | * | |
651 | * On success, destroy old qdisc. | |
652 | */ | |
653 | ||
654 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, | |
655 | struct sk_buff *skb, struct nlmsghdr *n, u32 classid, | |
656 | struct Qdisc *new, struct Qdisc *old) | |
657 | { | |
658 | struct Qdisc *q = old; | |
659 | int err = 0; | |
660 | ||
661 | if (parent == NULL) { | |
662 | unsigned int i, num_q, ingress; | |
663 | ||
664 | ingress = 0; | |
665 | num_q = dev->num_tx_queues; | |
666 | if ((q && q->flags & TCQ_F_INGRESS) || | |
667 | (new && new->flags & TCQ_F_INGRESS)) { | |
668 | num_q = 1; | |
669 | ingress = 1; | |
670 | } | |
671 | ||
672 | if (dev->flags & IFF_UP) | |
673 | dev_deactivate(dev); | |
674 | ||
675 | for (i = 0; i < num_q; i++) { | |
676 | struct netdev_queue *dev_queue = &dev->rx_queue; | |
677 | ||
678 | if (!ingress) | |
679 | dev_queue = netdev_get_tx_queue(dev, i); | |
680 | ||
681 | old = dev_graft_qdisc(dev_queue, new); | |
682 | if (new && i > 0) | |
683 | atomic_inc(&new->refcnt); | |
684 | ||
685 | notify_and_destroy(skb, n, classid, old, new); | |
686 | } | |
687 | ||
688 | if (dev->flags & IFF_UP) | |
689 | dev_activate(dev); | |
690 | } else { | |
691 | const struct Qdisc_class_ops *cops = parent->ops->cl_ops; | |
692 | ||
693 | err = -EINVAL; | |
694 | ||
695 | if (cops) { | |
696 | unsigned long cl = cops->get(parent, classid); | |
697 | if (cl) { | |
698 | err = cops->graft(parent, cl, new, &old); | |
699 | cops->put(parent, cl); | |
700 | } | |
701 | } | |
702 | if (!err) | |
703 | notify_and_destroy(skb, n, classid, old, new); | |
704 | } | |
705 | return err; | |
706 | } | |
707 | ||
708 | /* lockdep annotation is needed for ingress; egress gets it only for name */ | |
709 | static struct lock_class_key qdisc_tx_lock; | |
710 | static struct lock_class_key qdisc_rx_lock; | |
711 | ||
712 | /* | |
713 | Allocate and initialize new qdisc. | |
714 | ||
715 | Parameters are passed via opt. | |
716 | */ | |
717 | ||
718 | static struct Qdisc * | |
719 | qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue, | |
720 | u32 parent, u32 handle, struct nlattr **tca, int *errp) | |
721 | { | |
722 | int err; | |
723 | struct nlattr *kind = tca[TCA_KIND]; | |
724 | struct Qdisc *sch; | |
725 | struct Qdisc_ops *ops; | |
726 | struct qdisc_size_table *stab; | |
727 | ||
728 | ops = qdisc_lookup_ops(kind); | |
729 | #ifdef CONFIG_KMOD | |
730 | if (ops == NULL && kind != NULL) { | |
731 | char name[IFNAMSIZ]; | |
732 | if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { | |
733 | /* We dropped the RTNL semaphore in order to | |
734 | * perform the module load. So, even if we | |
735 | * succeeded in loading the module we have to | |
736 | * tell the caller to replay the request. We | |
737 | * indicate this using -EAGAIN. | |
738 | * We replay the request because the device may | |
739 | * go away in the mean time. | |
740 | */ | |
741 | rtnl_unlock(); | |
742 | request_module("sch_%s", name); | |
743 | rtnl_lock(); | |
744 | ops = qdisc_lookup_ops(kind); | |
745 | if (ops != NULL) { | |
746 | /* We will try again qdisc_lookup_ops, | |
747 | * so don't keep a reference. | |
748 | */ | |
749 | module_put(ops->owner); | |
750 | err = -EAGAIN; | |
751 | goto err_out; | |
752 | } | |
753 | } | |
754 | } | |
755 | #endif | |
756 | ||
757 | err = -ENOENT; | |
758 | if (ops == NULL) | |
759 | goto err_out; | |
760 | ||
761 | sch = qdisc_alloc(dev_queue, ops); | |
762 | if (IS_ERR(sch)) { | |
763 | err = PTR_ERR(sch); | |
764 | goto err_out2; | |
765 | } | |
766 | ||
767 | sch->parent = parent; | |
768 | ||
769 | if (handle == TC_H_INGRESS) { | |
770 | sch->flags |= TCQ_F_INGRESS; | |
771 | handle = TC_H_MAKE(TC_H_INGRESS, 0); | |
772 | lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); | |
773 | } else { | |
774 | if (handle == 0) { | |
775 | handle = qdisc_alloc_handle(dev); | |
776 | err = -ENOMEM; | |
777 | if (handle == 0) | |
778 | goto err_out3; | |
779 | } | |
780 | lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); | |
781 | } | |
782 | ||
783 | sch->handle = handle; | |
784 | ||
785 | if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { | |
786 | if (tca[TCA_STAB]) { | |
787 | stab = qdisc_get_stab(tca[TCA_STAB]); | |
788 | if (IS_ERR(stab)) { | |
789 | err = PTR_ERR(stab); | |
790 | goto err_out3; | |
791 | } | |
792 | sch->stab = stab; | |
793 | } | |
794 | if (tca[TCA_RATE]) { | |
795 | err = gen_new_estimator(&sch->bstats, &sch->rate_est, | |
796 | qdisc_root_lock(sch), | |
797 | tca[TCA_RATE]); | |
798 | if (err) { | |
799 | /* | |
800 | * Any broken qdiscs that would require | |
801 | * a ops->reset() here? The qdisc was never | |
802 | * in action so it shouldn't be necessary. | |
803 | */ | |
804 | if (ops->destroy) | |
805 | ops->destroy(sch); | |
806 | goto err_out3; | |
807 | } | |
808 | } | |
809 | if ((parent != TC_H_ROOT) && !(sch->flags & TCQ_F_INGRESS)) | |
810 | list_add_tail(&sch->list, &dev_queue->qdisc_sleeping->list); | |
811 | ||
812 | return sch; | |
813 | } | |
814 | err_out3: | |
815 | qdisc_put_stab(sch->stab); | |
816 | dev_put(dev); | |
817 | kfree((char *) sch - sch->padded); | |
818 | err_out2: | |
819 | module_put(ops->owner); | |
820 | err_out: | |
821 | *errp = err; | |
822 | return NULL; | |
823 | } | |
824 | ||
825 | static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) | |
826 | { | |
827 | struct qdisc_size_table *stab = NULL; | |
828 | int err = 0; | |
829 | ||
830 | if (tca[TCA_OPTIONS]) { | |
831 | if (sch->ops->change == NULL) | |
832 | return -EINVAL; | |
833 | err = sch->ops->change(sch, tca[TCA_OPTIONS]); | |
834 | if (err) | |
835 | return err; | |
836 | } | |
837 | ||
838 | if (tca[TCA_STAB]) { | |
839 | stab = qdisc_get_stab(tca[TCA_STAB]); | |
840 | if (IS_ERR(stab)) | |
841 | return PTR_ERR(stab); | |
842 | } | |
843 | ||
844 | qdisc_put_stab(sch->stab); | |
845 | sch->stab = stab; | |
846 | ||
847 | if (tca[TCA_RATE]) | |
848 | gen_replace_estimator(&sch->bstats, &sch->rate_est, | |
849 | qdisc_root_lock(sch), tca[TCA_RATE]); | |
850 | return 0; | |
851 | } | |
852 | ||
853 | struct check_loop_arg | |
854 | { | |
855 | struct qdisc_walker w; | |
856 | struct Qdisc *p; | |
857 | int depth; | |
858 | }; | |
859 | ||
860 | static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); | |
861 | ||
862 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) | |
863 | { | |
864 | struct check_loop_arg arg; | |
865 | ||
866 | if (q->ops->cl_ops == NULL) | |
867 | return 0; | |
868 | ||
869 | arg.w.stop = arg.w.skip = arg.w.count = 0; | |
870 | arg.w.fn = check_loop_fn; | |
871 | arg.depth = depth; | |
872 | arg.p = p; | |
873 | q->ops->cl_ops->walk(q, &arg.w); | |
874 | return arg.w.stop ? -ELOOP : 0; | |
875 | } | |
876 | ||
877 | static int | |
878 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) | |
879 | { | |
880 | struct Qdisc *leaf; | |
881 | const struct Qdisc_class_ops *cops = q->ops->cl_ops; | |
882 | struct check_loop_arg *arg = (struct check_loop_arg *)w; | |
883 | ||
884 | leaf = cops->leaf(q, cl); | |
885 | if (leaf) { | |
886 | if (leaf == arg->p || arg->depth > 7) | |
887 | return -ELOOP; | |
888 | return check_loop(leaf, arg->p, arg->depth + 1); | |
889 | } | |
890 | return 0; | |
891 | } | |
892 | ||
893 | /* | |
894 | * Delete/get qdisc. | |
895 | */ | |
896 | ||
897 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
898 | { | |
899 | struct net *net = sock_net(skb->sk); | |
900 | struct tcmsg *tcm = NLMSG_DATA(n); | |
901 | struct nlattr *tca[TCA_MAX + 1]; | |
902 | struct net_device *dev; | |
903 | u32 clid = tcm->tcm_parent; | |
904 | struct Qdisc *q = NULL; | |
905 | struct Qdisc *p = NULL; | |
906 | int err; | |
907 | ||
908 | if (net != &init_net) | |
909 | return -EINVAL; | |
910 | ||
911 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) | |
912 | return -ENODEV; | |
913 | ||
914 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); | |
915 | if (err < 0) | |
916 | return err; | |
917 | ||
918 | if (clid) { | |
919 | if (clid != TC_H_ROOT) { | |
920 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { | |
921 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
922 | return -ENOENT; | |
923 | q = qdisc_leaf(p, clid); | |
924 | } else { /* ingress */ | |
925 | q = dev->rx_queue.qdisc_sleeping; | |
926 | } | |
927 | } else { | |
928 | struct netdev_queue *dev_queue; | |
929 | dev_queue = netdev_get_tx_queue(dev, 0); | |
930 | q = dev_queue->qdisc_sleeping; | |
931 | } | |
932 | if (!q) | |
933 | return -ENOENT; | |
934 | ||
935 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) | |
936 | return -EINVAL; | |
937 | } else { | |
938 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
939 | return -ENOENT; | |
940 | } | |
941 | ||
942 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) | |
943 | return -EINVAL; | |
944 | ||
945 | if (n->nlmsg_type == RTM_DELQDISC) { | |
946 | if (!clid) | |
947 | return -EINVAL; | |
948 | if (q->handle == 0) | |
949 | return -ENOENT; | |
950 | if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0) | |
951 | return err; | |
952 | } else { | |
953 | qdisc_notify(skb, n, clid, NULL, q); | |
954 | } | |
955 | return 0; | |
956 | } | |
957 | ||
958 | /* | |
959 | Create/change qdisc. | |
960 | */ | |
961 | ||
962 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
963 | { | |
964 | struct net *net = sock_net(skb->sk); | |
965 | struct tcmsg *tcm; | |
966 | struct nlattr *tca[TCA_MAX + 1]; | |
967 | struct net_device *dev; | |
968 | u32 clid; | |
969 | struct Qdisc *q, *p; | |
970 | int err; | |
971 | ||
972 | if (net != &init_net) | |
973 | return -EINVAL; | |
974 | ||
975 | replay: | |
976 | /* Reinit, just in case something touches this. */ | |
977 | tcm = NLMSG_DATA(n); | |
978 | clid = tcm->tcm_parent; | |
979 | q = p = NULL; | |
980 | ||
981 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) | |
982 | return -ENODEV; | |
983 | ||
984 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); | |
985 | if (err < 0) | |
986 | return err; | |
987 | ||
988 | if (clid) { | |
989 | if (clid != TC_H_ROOT) { | |
990 | if (clid != TC_H_INGRESS) { | |
991 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
992 | return -ENOENT; | |
993 | q = qdisc_leaf(p, clid); | |
994 | } else { /*ingress */ | |
995 | q = dev->rx_queue.qdisc_sleeping; | |
996 | } | |
997 | } else { | |
998 | struct netdev_queue *dev_queue; | |
999 | dev_queue = netdev_get_tx_queue(dev, 0); | |
1000 | q = dev_queue->qdisc_sleeping; | |
1001 | } | |
1002 | ||
1003 | /* It may be default qdisc, ignore it */ | |
1004 | if (q && q->handle == 0) | |
1005 | q = NULL; | |
1006 | ||
1007 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { | |
1008 | if (tcm->tcm_handle) { | |
1009 | if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) | |
1010 | return -EEXIST; | |
1011 | if (TC_H_MIN(tcm->tcm_handle)) | |
1012 | return -EINVAL; | |
1013 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
1014 | goto create_n_graft; | |
1015 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1016 | return -EEXIST; | |
1017 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) | |
1018 | return -EINVAL; | |
1019 | if (q == p || | |
1020 | (p && check_loop(q, p, 0))) | |
1021 | return -ELOOP; | |
1022 | atomic_inc(&q->refcnt); | |
1023 | goto graft; | |
1024 | } else { | |
1025 | if (q == NULL) | |
1026 | goto create_n_graft; | |
1027 | ||
1028 | /* This magic test requires explanation. | |
1029 | * | |
1030 | * We know, that some child q is already | |
1031 | * attached to this parent and have choice: | |
1032 | * either to change it or to create/graft new one. | |
1033 | * | |
1034 | * 1. We are allowed to create/graft only | |
1035 | * if CREATE and REPLACE flags are set. | |
1036 | * | |
1037 | * 2. If EXCL is set, requestor wanted to say, | |
1038 | * that qdisc tcm_handle is not expected | |
1039 | * to exist, so that we choose create/graft too. | |
1040 | * | |
1041 | * 3. The last case is when no flags are set. | |
1042 | * Alas, it is sort of hole in API, we | |
1043 | * cannot decide what to do unambiguously. | |
1044 | * For now we select create/graft, if | |
1045 | * user gave KIND, which does not match existing. | |
1046 | */ | |
1047 | if ((n->nlmsg_flags&NLM_F_CREATE) && | |
1048 | (n->nlmsg_flags&NLM_F_REPLACE) && | |
1049 | ((n->nlmsg_flags&NLM_F_EXCL) || | |
1050 | (tca[TCA_KIND] && | |
1051 | nla_strcmp(tca[TCA_KIND], q->ops->id)))) | |
1052 | goto create_n_graft; | |
1053 | } | |
1054 | } | |
1055 | } else { | |
1056 | if (!tcm->tcm_handle) | |
1057 | return -EINVAL; | |
1058 | q = qdisc_lookup(dev, tcm->tcm_handle); | |
1059 | } | |
1060 | ||
1061 | /* Change qdisc parameters */ | |
1062 | if (q == NULL) | |
1063 | return -ENOENT; | |
1064 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1065 | return -EEXIST; | |
1066 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) | |
1067 | return -EINVAL; | |
1068 | err = qdisc_change(q, tca); | |
1069 | if (err == 0) | |
1070 | qdisc_notify(skb, n, clid, NULL, q); | |
1071 | return err; | |
1072 | ||
1073 | create_n_graft: | |
1074 | if (!(n->nlmsg_flags&NLM_F_CREATE)) | |
1075 | return -ENOENT; | |
1076 | if (clid == TC_H_INGRESS) | |
1077 | q = qdisc_create(dev, &dev->rx_queue, | |
1078 | tcm->tcm_parent, tcm->tcm_parent, | |
1079 | tca, &err); | |
1080 | else | |
1081 | q = qdisc_create(dev, netdev_get_tx_queue(dev, 0), | |
1082 | tcm->tcm_parent, tcm->tcm_handle, | |
1083 | tca, &err); | |
1084 | if (q == NULL) { | |
1085 | if (err == -EAGAIN) | |
1086 | goto replay; | |
1087 | return err; | |
1088 | } | |
1089 | ||
1090 | graft: | |
1091 | err = qdisc_graft(dev, p, skb, n, clid, q, NULL); | |
1092 | if (err) { | |
1093 | if (q) | |
1094 | qdisc_destroy(q); | |
1095 | return err; | |
1096 | } | |
1097 | ||
1098 | return 0; | |
1099 | } | |
1100 | ||
1101 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, | |
1102 | u32 pid, u32 seq, u16 flags, int event) | |
1103 | { | |
1104 | struct tcmsg *tcm; | |
1105 | struct nlmsghdr *nlh; | |
1106 | unsigned char *b = skb_tail_pointer(skb); | |
1107 | struct gnet_dump d; | |
1108 | ||
1109 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); | |
1110 | tcm = NLMSG_DATA(nlh); | |
1111 | tcm->tcm_family = AF_UNSPEC; | |
1112 | tcm->tcm__pad1 = 0; | |
1113 | tcm->tcm__pad2 = 0; | |
1114 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; | |
1115 | tcm->tcm_parent = clid; | |
1116 | tcm->tcm_handle = q->handle; | |
1117 | tcm->tcm_info = atomic_read(&q->refcnt); | |
1118 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); | |
1119 | if (q->ops->dump && q->ops->dump(q, skb) < 0) | |
1120 | goto nla_put_failure; | |
1121 | q->qstats.qlen = q->q.qlen; | |
1122 | ||
1123 | if (q->stab && qdisc_dump_stab(skb, q->stab) < 0) | |
1124 | goto nla_put_failure; | |
1125 | ||
1126 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
1127 | TCA_XSTATS, qdisc_root_lock(q), &d) < 0) | |
1128 | goto nla_put_failure; | |
1129 | ||
1130 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) | |
1131 | goto nla_put_failure; | |
1132 | ||
1133 | if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || | |
1134 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || | |
1135 | gnet_stats_copy_queue(&d, &q->qstats) < 0) | |
1136 | goto nla_put_failure; | |
1137 | ||
1138 | if (gnet_stats_finish_copy(&d) < 0) | |
1139 | goto nla_put_failure; | |
1140 | ||
1141 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; | |
1142 | return skb->len; | |
1143 | ||
1144 | nlmsg_failure: | |
1145 | nla_put_failure: | |
1146 | nlmsg_trim(skb, b); | |
1147 | return -1; | |
1148 | } | |
1149 | ||
1150 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1151 | u32 clid, struct Qdisc *old, struct Qdisc *new) | |
1152 | { | |
1153 | struct sk_buff *skb; | |
1154 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1155 | ||
1156 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1157 | if (!skb) | |
1158 | return -ENOBUFS; | |
1159 | ||
1160 | if (old && old->handle) { | |
1161 | if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) | |
1162 | goto err_out; | |
1163 | } | |
1164 | if (new) { | |
1165 | if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) | |
1166 | goto err_out; | |
1167 | } | |
1168 | ||
1169 | if (skb->len) | |
1170 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); | |
1171 | ||
1172 | err_out: | |
1173 | kfree_skb(skb); | |
1174 | return -EINVAL; | |
1175 | } | |
1176 | ||
1177 | static bool tc_qdisc_dump_ignore(struct Qdisc *q) | |
1178 | { | |
1179 | return (q->flags & TCQ_F_BUILTIN) ? true : false; | |
1180 | } | |
1181 | ||
1182 | static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, | |
1183 | struct netlink_callback *cb, | |
1184 | int *q_idx_p, int s_q_idx) | |
1185 | { | |
1186 | int ret = 0, q_idx = *q_idx_p; | |
1187 | struct Qdisc *q; | |
1188 | ||
1189 | if (!root) | |
1190 | return 0; | |
1191 | ||
1192 | q = root; | |
1193 | if (q_idx < s_q_idx) { | |
1194 | q_idx++; | |
1195 | } else { | |
1196 | if (!tc_qdisc_dump_ignore(q) && | |
1197 | tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
1198 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) | |
1199 | goto done; | |
1200 | q_idx++; | |
1201 | } | |
1202 | list_for_each_entry(q, &root->list, list) { | |
1203 | if (q_idx < s_q_idx) { | |
1204 | q_idx++; | |
1205 | continue; | |
1206 | } | |
1207 | if (!tc_qdisc_dump_ignore(q) && | |
1208 | tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
1209 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) | |
1210 | goto done; | |
1211 | q_idx++; | |
1212 | } | |
1213 | ||
1214 | out: | |
1215 | *q_idx_p = q_idx; | |
1216 | return ret; | |
1217 | done: | |
1218 | ret = -1; | |
1219 | goto out; | |
1220 | } | |
1221 | ||
1222 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) | |
1223 | { | |
1224 | struct net *net = sock_net(skb->sk); | |
1225 | int idx, q_idx; | |
1226 | int s_idx, s_q_idx; | |
1227 | struct net_device *dev; | |
1228 | ||
1229 | if (net != &init_net) | |
1230 | return 0; | |
1231 | ||
1232 | s_idx = cb->args[0]; | |
1233 | s_q_idx = q_idx = cb->args[1]; | |
1234 | read_lock(&dev_base_lock); | |
1235 | idx = 0; | |
1236 | for_each_netdev(&init_net, dev) { | |
1237 | struct netdev_queue *dev_queue; | |
1238 | ||
1239 | if (idx < s_idx) | |
1240 | goto cont; | |
1241 | if (idx > s_idx) | |
1242 | s_q_idx = 0; | |
1243 | q_idx = 0; | |
1244 | ||
1245 | dev_queue = netdev_get_tx_queue(dev, 0); | |
1246 | if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0) | |
1247 | goto done; | |
1248 | ||
1249 | dev_queue = &dev->rx_queue; | |
1250 | if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0) | |
1251 | goto done; | |
1252 | ||
1253 | cont: | |
1254 | idx++; | |
1255 | } | |
1256 | ||
1257 | done: | |
1258 | read_unlock(&dev_base_lock); | |
1259 | ||
1260 | cb->args[0] = idx; | |
1261 | cb->args[1] = q_idx; | |
1262 | ||
1263 | return skb->len; | |
1264 | } | |
1265 | ||
1266 | ||
1267 | ||
1268 | /************************************************ | |
1269 | * Traffic classes manipulation. * | |
1270 | ************************************************/ | |
1271 | ||
1272 | ||
1273 | ||
1274 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
1275 | { | |
1276 | struct net *net = sock_net(skb->sk); | |
1277 | struct netdev_queue *dev_queue; | |
1278 | struct tcmsg *tcm = NLMSG_DATA(n); | |
1279 | struct nlattr *tca[TCA_MAX + 1]; | |
1280 | struct net_device *dev; | |
1281 | struct Qdisc *q = NULL; | |
1282 | const struct Qdisc_class_ops *cops; | |
1283 | unsigned long cl = 0; | |
1284 | unsigned long new_cl; | |
1285 | u32 pid = tcm->tcm_parent; | |
1286 | u32 clid = tcm->tcm_handle; | |
1287 | u32 qid = TC_H_MAJ(clid); | |
1288 | int err; | |
1289 | ||
1290 | if (net != &init_net) | |
1291 | return -EINVAL; | |
1292 | ||
1293 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) | |
1294 | return -ENODEV; | |
1295 | ||
1296 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); | |
1297 | if (err < 0) | |
1298 | return err; | |
1299 | ||
1300 | /* | |
1301 | parent == TC_H_UNSPEC - unspecified parent. | |
1302 | parent == TC_H_ROOT - class is root, which has no parent. | |
1303 | parent == X:0 - parent is root class. | |
1304 | parent == X:Y - parent is a node in hierarchy. | |
1305 | parent == 0:Y - parent is X:Y, where X:0 is qdisc. | |
1306 | ||
1307 | handle == 0:0 - generate handle from kernel pool. | |
1308 | handle == 0:Y - class is X:Y, where X:0 is qdisc. | |
1309 | handle == X:Y - clear. | |
1310 | handle == X:0 - root class. | |
1311 | */ | |
1312 | ||
1313 | /* Step 1. Determine qdisc handle X:0 */ | |
1314 | ||
1315 | dev_queue = netdev_get_tx_queue(dev, 0); | |
1316 | if (pid != TC_H_ROOT) { | |
1317 | u32 qid1 = TC_H_MAJ(pid); | |
1318 | ||
1319 | if (qid && qid1) { | |
1320 | /* If both majors are known, they must be identical. */ | |
1321 | if (qid != qid1) | |
1322 | return -EINVAL; | |
1323 | } else if (qid1) { | |
1324 | qid = qid1; | |
1325 | } else if (qid == 0) | |
1326 | qid = dev_queue->qdisc_sleeping->handle; | |
1327 | ||
1328 | /* Now qid is genuine qdisc handle consistent | |
1329 | both with parent and child. | |
1330 | ||
1331 | TC_H_MAJ(pid) still may be unspecified, complete it now. | |
1332 | */ | |
1333 | if (pid) | |
1334 | pid = TC_H_MAKE(qid, pid); | |
1335 | } else { | |
1336 | if (qid == 0) | |
1337 | qid = dev_queue->qdisc_sleeping->handle; | |
1338 | } | |
1339 | ||
1340 | /* OK. Locate qdisc */ | |
1341 | if ((q = qdisc_lookup(dev, qid)) == NULL) | |
1342 | return -ENOENT; | |
1343 | ||
1344 | /* An check that it supports classes */ | |
1345 | cops = q->ops->cl_ops; | |
1346 | if (cops == NULL) | |
1347 | return -EINVAL; | |
1348 | ||
1349 | /* Now try to get class */ | |
1350 | if (clid == 0) { | |
1351 | if (pid == TC_H_ROOT) | |
1352 | clid = qid; | |
1353 | } else | |
1354 | clid = TC_H_MAKE(qid, clid); | |
1355 | ||
1356 | if (clid) | |
1357 | cl = cops->get(q, clid); | |
1358 | ||
1359 | if (cl == 0) { | |
1360 | err = -ENOENT; | |
1361 | if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) | |
1362 | goto out; | |
1363 | } else { | |
1364 | switch (n->nlmsg_type) { | |
1365 | case RTM_NEWTCLASS: | |
1366 | err = -EEXIST; | |
1367 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1368 | goto out; | |
1369 | break; | |
1370 | case RTM_DELTCLASS: | |
1371 | err = cops->delete(q, cl); | |
1372 | if (err == 0) | |
1373 | tclass_notify(skb, n, q, cl, RTM_DELTCLASS); | |
1374 | goto out; | |
1375 | case RTM_GETTCLASS: | |
1376 | err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS); | |
1377 | goto out; | |
1378 | default: | |
1379 | err = -EINVAL; | |
1380 | goto out; | |
1381 | } | |
1382 | } | |
1383 | ||
1384 | new_cl = cl; | |
1385 | err = cops->change(q, clid, pid, tca, &new_cl); | |
1386 | if (err == 0) | |
1387 | tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS); | |
1388 | ||
1389 | out: | |
1390 | if (cl) | |
1391 | cops->put(q, cl); | |
1392 | ||
1393 | return err; | |
1394 | } | |
1395 | ||
1396 | ||
1397 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, | |
1398 | unsigned long cl, | |
1399 | u32 pid, u32 seq, u16 flags, int event) | |
1400 | { | |
1401 | struct tcmsg *tcm; | |
1402 | struct nlmsghdr *nlh; | |
1403 | unsigned char *b = skb_tail_pointer(skb); | |
1404 | struct gnet_dump d; | |
1405 | const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; | |
1406 | ||
1407 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); | |
1408 | tcm = NLMSG_DATA(nlh); | |
1409 | tcm->tcm_family = AF_UNSPEC; | |
1410 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; | |
1411 | tcm->tcm_parent = q->handle; | |
1412 | tcm->tcm_handle = q->handle; | |
1413 | tcm->tcm_info = 0; | |
1414 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); | |
1415 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) | |
1416 | goto nla_put_failure; | |
1417 | ||
1418 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, | |
1419 | TCA_XSTATS, qdisc_root_lock(q), &d) < 0) | |
1420 | goto nla_put_failure; | |
1421 | ||
1422 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) | |
1423 | goto nla_put_failure; | |
1424 | ||
1425 | if (gnet_stats_finish_copy(&d) < 0) | |
1426 | goto nla_put_failure; | |
1427 | ||
1428 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; | |
1429 | return skb->len; | |
1430 | ||
1431 | nlmsg_failure: | |
1432 | nla_put_failure: | |
1433 | nlmsg_trim(skb, b); | |
1434 | return -1; | |
1435 | } | |
1436 | ||
1437 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1438 | struct Qdisc *q, unsigned long cl, int event) | |
1439 | { | |
1440 | struct sk_buff *skb; | |
1441 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1442 | ||
1443 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1444 | if (!skb) | |
1445 | return -ENOBUFS; | |
1446 | ||
1447 | if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { | |
1448 | kfree_skb(skb); | |
1449 | return -EINVAL; | |
1450 | } | |
1451 | ||
1452 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); | |
1453 | } | |
1454 | ||
1455 | struct qdisc_dump_args | |
1456 | { | |
1457 | struct qdisc_walker w; | |
1458 | struct sk_buff *skb; | |
1459 | struct netlink_callback *cb; | |
1460 | }; | |
1461 | ||
1462 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) | |
1463 | { | |
1464 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; | |
1465 | ||
1466 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, | |
1467 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); | |
1468 | } | |
1469 | ||
1470 | static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, | |
1471 | struct tcmsg *tcm, struct netlink_callback *cb, | |
1472 | int *t_p, int s_t) | |
1473 | { | |
1474 | struct qdisc_dump_args arg; | |
1475 | ||
1476 | if (tc_qdisc_dump_ignore(q) || | |
1477 | *t_p < s_t || !q->ops->cl_ops || | |
1478 | (tcm->tcm_parent && | |
1479 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) { | |
1480 | (*t_p)++; | |
1481 | return 0; | |
1482 | } | |
1483 | if (*t_p > s_t) | |
1484 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); | |
1485 | arg.w.fn = qdisc_class_dump; | |
1486 | arg.skb = skb; | |
1487 | arg.cb = cb; | |
1488 | arg.w.stop = 0; | |
1489 | arg.w.skip = cb->args[1]; | |
1490 | arg.w.count = 0; | |
1491 | q->ops->cl_ops->walk(q, &arg.w); | |
1492 | cb->args[1] = arg.w.count; | |
1493 | if (arg.w.stop) | |
1494 | return -1; | |
1495 | (*t_p)++; | |
1496 | return 0; | |
1497 | } | |
1498 | ||
1499 | static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, | |
1500 | struct tcmsg *tcm, struct netlink_callback *cb, | |
1501 | int *t_p, int s_t) | |
1502 | { | |
1503 | struct Qdisc *q; | |
1504 | ||
1505 | if (!root) | |
1506 | return 0; | |
1507 | ||
1508 | if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) | |
1509 | return -1; | |
1510 | ||
1511 | list_for_each_entry(q, &root->list, list) { | |
1512 | if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) | |
1513 | return -1; | |
1514 | } | |
1515 | ||
1516 | return 0; | |
1517 | } | |
1518 | ||
1519 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) | |
1520 | { | |
1521 | struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); | |
1522 | struct net *net = sock_net(skb->sk); | |
1523 | struct netdev_queue *dev_queue; | |
1524 | struct net_device *dev; | |
1525 | int t, s_t; | |
1526 | ||
1527 | if (net != &init_net) | |
1528 | return 0; | |
1529 | ||
1530 | if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) | |
1531 | return 0; | |
1532 | if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) | |
1533 | return 0; | |
1534 | ||
1535 | s_t = cb->args[0]; | |
1536 | t = 0; | |
1537 | ||
1538 | dev_queue = netdev_get_tx_queue(dev, 0); | |
1539 | if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0) | |
1540 | goto done; | |
1541 | ||
1542 | dev_queue = &dev->rx_queue; | |
1543 | if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0) | |
1544 | goto done; | |
1545 | ||
1546 | done: | |
1547 | cb->args[0] = t; | |
1548 | ||
1549 | dev_put(dev); | |
1550 | return skb->len; | |
1551 | } | |
1552 | ||
1553 | /* Main classifier routine: scans classifier chain attached | |
1554 | to this qdisc, (optionally) tests for protocol and asks | |
1555 | specific classifiers. | |
1556 | */ | |
1557 | int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp, | |
1558 | struct tcf_result *res) | |
1559 | { | |
1560 | __be16 protocol = skb->protocol; | |
1561 | int err = 0; | |
1562 | ||
1563 | for (; tp; tp = tp->next) { | |
1564 | if ((tp->protocol == protocol || | |
1565 | tp->protocol == htons(ETH_P_ALL)) && | |
1566 | (err = tp->classify(skb, tp, res)) >= 0) { | |
1567 | #ifdef CONFIG_NET_CLS_ACT | |
1568 | if (err != TC_ACT_RECLASSIFY && skb->tc_verd) | |
1569 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0); | |
1570 | #endif | |
1571 | return err; | |
1572 | } | |
1573 | } | |
1574 | return -1; | |
1575 | } | |
1576 | EXPORT_SYMBOL(tc_classify_compat); | |
1577 | ||
1578 | int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, | |
1579 | struct tcf_result *res) | |
1580 | { | |
1581 | int err = 0; | |
1582 | __be16 protocol; | |
1583 | #ifdef CONFIG_NET_CLS_ACT | |
1584 | struct tcf_proto *otp = tp; | |
1585 | reclassify: | |
1586 | #endif | |
1587 | protocol = skb->protocol; | |
1588 | ||
1589 | err = tc_classify_compat(skb, tp, res); | |
1590 | #ifdef CONFIG_NET_CLS_ACT | |
1591 | if (err == TC_ACT_RECLASSIFY) { | |
1592 | u32 verd = G_TC_VERD(skb->tc_verd); | |
1593 | tp = otp; | |
1594 | ||
1595 | if (verd++ >= MAX_REC_LOOP) { | |
1596 | printk("rule prio %u protocol %02x reclassify loop, " | |
1597 | "packet dropped\n", | |
1598 | tp->prio&0xffff, ntohs(tp->protocol)); | |
1599 | return TC_ACT_SHOT; | |
1600 | } | |
1601 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd); | |
1602 | goto reclassify; | |
1603 | } | |
1604 | #endif | |
1605 | return err; | |
1606 | } | |
1607 | EXPORT_SYMBOL(tc_classify); | |
1608 | ||
1609 | void tcf_destroy(struct tcf_proto *tp) | |
1610 | { | |
1611 | tp->ops->destroy(tp); | |
1612 | module_put(tp->ops->owner); | |
1613 | kfree(tp); | |
1614 | } | |
1615 | ||
1616 | void tcf_destroy_chain(struct tcf_proto **fl) | |
1617 | { | |
1618 | struct tcf_proto *tp; | |
1619 | ||
1620 | while ((tp = *fl) != NULL) { | |
1621 | *fl = tp->next; | |
1622 | tcf_destroy(tp); | |
1623 | } | |
1624 | } | |
1625 | EXPORT_SYMBOL(tcf_destroy_chain); | |
1626 | ||
1627 | #ifdef CONFIG_PROC_FS | |
1628 | static int psched_show(struct seq_file *seq, void *v) | |
1629 | { | |
1630 | struct timespec ts; | |
1631 | ||
1632 | hrtimer_get_res(CLOCK_MONOTONIC, &ts); | |
1633 | seq_printf(seq, "%08x %08x %08x %08x\n", | |
1634 | (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1), | |
1635 | 1000000, | |
1636 | (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts))); | |
1637 | ||
1638 | return 0; | |
1639 | } | |
1640 | ||
1641 | static int psched_open(struct inode *inode, struct file *file) | |
1642 | { | |
1643 | return single_open(file, psched_show, PDE(inode)->data); | |
1644 | } | |
1645 | ||
1646 | static const struct file_operations psched_fops = { | |
1647 | .owner = THIS_MODULE, | |
1648 | .open = psched_open, | |
1649 | .read = seq_read, | |
1650 | .llseek = seq_lseek, | |
1651 | .release = single_release, | |
1652 | }; | |
1653 | #endif | |
1654 | ||
1655 | static int __init pktsched_init(void) | |
1656 | { | |
1657 | register_qdisc(&pfifo_qdisc_ops); | |
1658 | register_qdisc(&bfifo_qdisc_ops); | |
1659 | proc_net_fops_create(&init_net, "psched", 0, &psched_fops); | |
1660 | ||
1661 | rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL); | |
1662 | rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL); | |
1663 | rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc); | |
1664 | rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL); | |
1665 | rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL); | |
1666 | rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass); | |
1667 | ||
1668 | return 0; | |
1669 | } | |
1670 | ||
1671 | subsys_initcall(pktsched_init); |