]>
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 | ---peek | |
101 | ||
102 | like dequeue but without removing a packet from the queue | |
103 | ||
104 | ---reset | |
105 | ||
106 | returns qdisc to initial state: purge all buffers, clear all | |
107 | timers, counters (except for statistics) etc. | |
108 | ||
109 | ---init | |
110 | ||
111 | initializes newly created qdisc. | |
112 | ||
113 | ---destroy | |
114 | ||
115 | destroys resources allocated by init and during lifetime of qdisc. | |
116 | ||
117 | ---change | |
118 | ||
119 | changes qdisc parameters. | |
120 | */ | |
121 | ||
122 | /* Protects list of registered TC modules. It is pure SMP lock. */ | |
123 | static DEFINE_RWLOCK(qdisc_mod_lock); | |
124 | ||
125 | ||
126 | /************************************************ | |
127 | * Queueing disciplines manipulation. * | |
128 | ************************************************/ | |
129 | ||
130 | ||
131 | /* The list of all installed queueing disciplines. */ | |
132 | ||
133 | static struct Qdisc_ops *qdisc_base; | |
134 | ||
135 | /* Register/uregister queueing discipline */ | |
136 | ||
137 | int register_qdisc(struct Qdisc_ops *qops) | |
138 | { | |
139 | struct Qdisc_ops *q, **qp; | |
140 | int rc = -EEXIST; | |
141 | ||
142 | write_lock(&qdisc_mod_lock); | |
143 | for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next) | |
144 | if (!strcmp(qops->id, q->id)) | |
145 | goto out; | |
146 | ||
147 | if (qops->enqueue == NULL) | |
148 | qops->enqueue = noop_qdisc_ops.enqueue; | |
149 | if (qops->peek == NULL) { | |
150 | if (qops->dequeue == NULL) { | |
151 | qops->peek = noop_qdisc_ops.peek; | |
152 | } else { | |
153 | rc = -EINVAL; | |
154 | goto out; | |
155 | } | |
156 | } | |
157 | if (qops->dequeue == NULL) | |
158 | qops->dequeue = noop_qdisc_ops.dequeue; | |
159 | ||
160 | qops->next = NULL; | |
161 | *qp = qops; | |
162 | rc = 0; | |
163 | out: | |
164 | write_unlock(&qdisc_mod_lock); | |
165 | return rc; | |
166 | } | |
167 | EXPORT_SYMBOL(register_qdisc); | |
168 | ||
169 | int unregister_qdisc(struct Qdisc_ops *qops) | |
170 | { | |
171 | struct Qdisc_ops *q, **qp; | |
172 | int err = -ENOENT; | |
173 | ||
174 | write_lock(&qdisc_mod_lock); | |
175 | for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next) | |
176 | if (q == qops) | |
177 | break; | |
178 | if (q) { | |
179 | *qp = q->next; | |
180 | q->next = NULL; | |
181 | err = 0; | |
182 | } | |
183 | write_unlock(&qdisc_mod_lock); | |
184 | return err; | |
185 | } | |
186 | EXPORT_SYMBOL(unregister_qdisc); | |
187 | ||
188 | /* We know handle. Find qdisc among all qdisc's attached to device | |
189 | (root qdisc, all its children, children of children etc.) | |
190 | */ | |
191 | ||
192 | static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle) | |
193 | { | |
194 | struct Qdisc *q; | |
195 | ||
196 | if (!(root->flags & TCQ_F_BUILTIN) && | |
197 | root->handle == handle) | |
198 | return root; | |
199 | ||
200 | list_for_each_entry(q, &root->list, list) { | |
201 | if (q->handle == handle) | |
202 | return q; | |
203 | } | |
204 | return NULL; | |
205 | } | |
206 | ||
207 | static void qdisc_list_add(struct Qdisc *q) | |
208 | { | |
209 | if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) | |
210 | list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list); | |
211 | } | |
212 | ||
213 | void qdisc_list_del(struct Qdisc *q) | |
214 | { | |
215 | if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) | |
216 | list_del(&q->list); | |
217 | } | |
218 | EXPORT_SYMBOL(qdisc_list_del); | |
219 | ||
220 | struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle) | |
221 | { | |
222 | struct Qdisc *q; | |
223 | ||
224 | q = qdisc_match_from_root(dev->qdisc, handle); | |
225 | if (q) | |
226 | goto out; | |
227 | ||
228 | q = qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle); | |
229 | out: | |
230 | return q; | |
231 | } | |
232 | ||
233 | static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid) | |
234 | { | |
235 | unsigned long cl; | |
236 | struct Qdisc *leaf; | |
237 | const struct Qdisc_class_ops *cops = p->ops->cl_ops; | |
238 | ||
239 | if (cops == NULL) | |
240 | return NULL; | |
241 | cl = cops->get(p, classid); | |
242 | ||
243 | if (cl == 0) | |
244 | return NULL; | |
245 | leaf = cops->leaf(p, cl); | |
246 | cops->put(p, cl); | |
247 | return leaf; | |
248 | } | |
249 | ||
250 | /* Find queueing discipline by name */ | |
251 | ||
252 | static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind) | |
253 | { | |
254 | struct Qdisc_ops *q = NULL; | |
255 | ||
256 | if (kind) { | |
257 | read_lock(&qdisc_mod_lock); | |
258 | for (q = qdisc_base; q; q = q->next) { | |
259 | if (nla_strcmp(kind, q->id) == 0) { | |
260 | if (!try_module_get(q->owner)) | |
261 | q = NULL; | |
262 | break; | |
263 | } | |
264 | } | |
265 | read_unlock(&qdisc_mod_lock); | |
266 | } | |
267 | return q; | |
268 | } | |
269 | ||
270 | static struct qdisc_rate_table *qdisc_rtab_list; | |
271 | ||
272 | struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab) | |
273 | { | |
274 | struct qdisc_rate_table *rtab; | |
275 | ||
276 | for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) { | |
277 | if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) { | |
278 | rtab->refcnt++; | |
279 | return rtab; | |
280 | } | |
281 | } | |
282 | ||
283 | if (tab == NULL || r->rate == 0 || r->cell_log == 0 || | |
284 | nla_len(tab) != TC_RTAB_SIZE) | |
285 | return NULL; | |
286 | ||
287 | rtab = kmalloc(sizeof(*rtab), GFP_KERNEL); | |
288 | if (rtab) { | |
289 | rtab->rate = *r; | |
290 | rtab->refcnt = 1; | |
291 | memcpy(rtab->data, nla_data(tab), 1024); | |
292 | rtab->next = qdisc_rtab_list; | |
293 | qdisc_rtab_list = rtab; | |
294 | } | |
295 | return rtab; | |
296 | } | |
297 | EXPORT_SYMBOL(qdisc_get_rtab); | |
298 | ||
299 | void qdisc_put_rtab(struct qdisc_rate_table *tab) | |
300 | { | |
301 | struct qdisc_rate_table *rtab, **rtabp; | |
302 | ||
303 | if (!tab || --tab->refcnt) | |
304 | return; | |
305 | ||
306 | for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) { | |
307 | if (rtab == tab) { | |
308 | *rtabp = rtab->next; | |
309 | kfree(rtab); | |
310 | return; | |
311 | } | |
312 | } | |
313 | } | |
314 | EXPORT_SYMBOL(qdisc_put_rtab); | |
315 | ||
316 | static LIST_HEAD(qdisc_stab_list); | |
317 | static DEFINE_SPINLOCK(qdisc_stab_lock); | |
318 | ||
319 | static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = { | |
320 | [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) }, | |
321 | [TCA_STAB_DATA] = { .type = NLA_BINARY }, | |
322 | }; | |
323 | ||
324 | static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt) | |
325 | { | |
326 | struct nlattr *tb[TCA_STAB_MAX + 1]; | |
327 | struct qdisc_size_table *stab; | |
328 | struct tc_sizespec *s; | |
329 | unsigned int tsize = 0; | |
330 | u16 *tab = NULL; | |
331 | int err; | |
332 | ||
333 | err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy); | |
334 | if (err < 0) | |
335 | return ERR_PTR(err); | |
336 | if (!tb[TCA_STAB_BASE]) | |
337 | return ERR_PTR(-EINVAL); | |
338 | ||
339 | s = nla_data(tb[TCA_STAB_BASE]); | |
340 | ||
341 | if (s->tsize > 0) { | |
342 | if (!tb[TCA_STAB_DATA]) | |
343 | return ERR_PTR(-EINVAL); | |
344 | tab = nla_data(tb[TCA_STAB_DATA]); | |
345 | tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16); | |
346 | } | |
347 | ||
348 | if (!s || tsize != s->tsize || (!tab && tsize > 0)) | |
349 | return ERR_PTR(-EINVAL); | |
350 | ||
351 | spin_lock(&qdisc_stab_lock); | |
352 | ||
353 | list_for_each_entry(stab, &qdisc_stab_list, list) { | |
354 | if (memcmp(&stab->szopts, s, sizeof(*s))) | |
355 | continue; | |
356 | if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16))) | |
357 | continue; | |
358 | stab->refcnt++; | |
359 | spin_unlock(&qdisc_stab_lock); | |
360 | return stab; | |
361 | } | |
362 | ||
363 | spin_unlock(&qdisc_stab_lock); | |
364 | ||
365 | stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL); | |
366 | if (!stab) | |
367 | return ERR_PTR(-ENOMEM); | |
368 | ||
369 | stab->refcnt = 1; | |
370 | stab->szopts = *s; | |
371 | if (tsize > 0) | |
372 | memcpy(stab->data, tab, tsize * sizeof(u16)); | |
373 | ||
374 | spin_lock(&qdisc_stab_lock); | |
375 | list_add_tail(&stab->list, &qdisc_stab_list); | |
376 | spin_unlock(&qdisc_stab_lock); | |
377 | ||
378 | return stab; | |
379 | } | |
380 | ||
381 | void qdisc_put_stab(struct qdisc_size_table *tab) | |
382 | { | |
383 | if (!tab) | |
384 | return; | |
385 | ||
386 | spin_lock(&qdisc_stab_lock); | |
387 | ||
388 | if (--tab->refcnt == 0) { | |
389 | list_del(&tab->list); | |
390 | kfree(tab); | |
391 | } | |
392 | ||
393 | spin_unlock(&qdisc_stab_lock); | |
394 | } | |
395 | EXPORT_SYMBOL(qdisc_put_stab); | |
396 | ||
397 | static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab) | |
398 | { | |
399 | struct nlattr *nest; | |
400 | ||
401 | nest = nla_nest_start(skb, TCA_STAB); | |
402 | if (nest == NULL) | |
403 | goto nla_put_failure; | |
404 | NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts); | |
405 | nla_nest_end(skb, nest); | |
406 | ||
407 | return skb->len; | |
408 | ||
409 | nla_put_failure: | |
410 | return -1; | |
411 | } | |
412 | ||
413 | void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab) | |
414 | { | |
415 | int pkt_len, slot; | |
416 | ||
417 | pkt_len = skb->len + stab->szopts.overhead; | |
418 | if (unlikely(!stab->szopts.tsize)) | |
419 | goto out; | |
420 | ||
421 | slot = pkt_len + stab->szopts.cell_align; | |
422 | if (unlikely(slot < 0)) | |
423 | slot = 0; | |
424 | ||
425 | slot >>= stab->szopts.cell_log; | |
426 | if (likely(slot < stab->szopts.tsize)) | |
427 | pkt_len = stab->data[slot]; | |
428 | else | |
429 | pkt_len = stab->data[stab->szopts.tsize - 1] * | |
430 | (slot / stab->szopts.tsize) + | |
431 | stab->data[slot % stab->szopts.tsize]; | |
432 | ||
433 | pkt_len <<= stab->szopts.size_log; | |
434 | out: | |
435 | if (unlikely(pkt_len < 1)) | |
436 | pkt_len = 1; | |
437 | qdisc_skb_cb(skb)->pkt_len = pkt_len; | |
438 | } | |
439 | EXPORT_SYMBOL(qdisc_calculate_pkt_len); | |
440 | ||
441 | void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc) | |
442 | { | |
443 | if (!(qdisc->flags & TCQ_F_WARN_NONWC)) { | |
444 | printk(KERN_WARNING | |
445 | "%s: %s qdisc %X: is non-work-conserving?\n", | |
446 | txt, qdisc->ops->id, qdisc->handle >> 16); | |
447 | qdisc->flags |= TCQ_F_WARN_NONWC; | |
448 | } | |
449 | } | |
450 | EXPORT_SYMBOL(qdisc_warn_nonwc); | |
451 | ||
452 | static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer) | |
453 | { | |
454 | struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog, | |
455 | timer); | |
456 | ||
457 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
458 | __netif_schedule(qdisc_root(wd->qdisc)); | |
459 | ||
460 | return HRTIMER_NORESTART; | |
461 | } | |
462 | ||
463 | void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc) | |
464 | { | |
465 | hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS); | |
466 | wd->timer.function = qdisc_watchdog; | |
467 | wd->qdisc = qdisc; | |
468 | } | |
469 | EXPORT_SYMBOL(qdisc_watchdog_init); | |
470 | ||
471 | void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires) | |
472 | { | |
473 | ktime_t time; | |
474 | ||
475 | if (test_bit(__QDISC_STATE_DEACTIVATED, | |
476 | &qdisc_root_sleeping(wd->qdisc)->state)) | |
477 | return; | |
478 | ||
479 | wd->qdisc->flags |= TCQ_F_THROTTLED; | |
480 | time = ktime_set(0, 0); | |
481 | time = ktime_add_ns(time, PSCHED_TICKS2NS(expires)); | |
482 | hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS); | |
483 | } | |
484 | EXPORT_SYMBOL(qdisc_watchdog_schedule); | |
485 | ||
486 | void qdisc_watchdog_cancel(struct qdisc_watchdog *wd) | |
487 | { | |
488 | hrtimer_cancel(&wd->timer); | |
489 | wd->qdisc->flags &= ~TCQ_F_THROTTLED; | |
490 | } | |
491 | EXPORT_SYMBOL(qdisc_watchdog_cancel); | |
492 | ||
493 | static struct hlist_head *qdisc_class_hash_alloc(unsigned int n) | |
494 | { | |
495 | unsigned int size = n * sizeof(struct hlist_head), i; | |
496 | struct hlist_head *h; | |
497 | ||
498 | if (size <= PAGE_SIZE) | |
499 | h = kmalloc(size, GFP_KERNEL); | |
500 | else | |
501 | h = (struct hlist_head *) | |
502 | __get_free_pages(GFP_KERNEL, get_order(size)); | |
503 | ||
504 | if (h != NULL) { | |
505 | for (i = 0; i < n; i++) | |
506 | INIT_HLIST_HEAD(&h[i]); | |
507 | } | |
508 | return h; | |
509 | } | |
510 | ||
511 | static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n) | |
512 | { | |
513 | unsigned int size = n * sizeof(struct hlist_head); | |
514 | ||
515 | if (size <= PAGE_SIZE) | |
516 | kfree(h); | |
517 | else | |
518 | free_pages((unsigned long)h, get_order(size)); | |
519 | } | |
520 | ||
521 | void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash) | |
522 | { | |
523 | struct Qdisc_class_common *cl; | |
524 | struct hlist_node *n, *next; | |
525 | struct hlist_head *nhash, *ohash; | |
526 | unsigned int nsize, nmask, osize; | |
527 | unsigned int i, h; | |
528 | ||
529 | /* Rehash when load factor exceeds 0.75 */ | |
530 | if (clhash->hashelems * 4 <= clhash->hashsize * 3) | |
531 | return; | |
532 | nsize = clhash->hashsize * 2; | |
533 | nmask = nsize - 1; | |
534 | nhash = qdisc_class_hash_alloc(nsize); | |
535 | if (nhash == NULL) | |
536 | return; | |
537 | ||
538 | ohash = clhash->hash; | |
539 | osize = clhash->hashsize; | |
540 | ||
541 | sch_tree_lock(sch); | |
542 | for (i = 0; i < osize; i++) { | |
543 | hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) { | |
544 | h = qdisc_class_hash(cl->classid, nmask); | |
545 | hlist_add_head(&cl->hnode, &nhash[h]); | |
546 | } | |
547 | } | |
548 | clhash->hash = nhash; | |
549 | clhash->hashsize = nsize; | |
550 | clhash->hashmask = nmask; | |
551 | sch_tree_unlock(sch); | |
552 | ||
553 | qdisc_class_hash_free(ohash, osize); | |
554 | } | |
555 | EXPORT_SYMBOL(qdisc_class_hash_grow); | |
556 | ||
557 | int qdisc_class_hash_init(struct Qdisc_class_hash *clhash) | |
558 | { | |
559 | unsigned int size = 4; | |
560 | ||
561 | clhash->hash = qdisc_class_hash_alloc(size); | |
562 | if (clhash->hash == NULL) | |
563 | return -ENOMEM; | |
564 | clhash->hashsize = size; | |
565 | clhash->hashmask = size - 1; | |
566 | clhash->hashelems = 0; | |
567 | return 0; | |
568 | } | |
569 | EXPORT_SYMBOL(qdisc_class_hash_init); | |
570 | ||
571 | void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash) | |
572 | { | |
573 | qdisc_class_hash_free(clhash->hash, clhash->hashsize); | |
574 | } | |
575 | EXPORT_SYMBOL(qdisc_class_hash_destroy); | |
576 | ||
577 | void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash, | |
578 | struct Qdisc_class_common *cl) | |
579 | { | |
580 | unsigned int h; | |
581 | ||
582 | INIT_HLIST_NODE(&cl->hnode); | |
583 | h = qdisc_class_hash(cl->classid, clhash->hashmask); | |
584 | hlist_add_head(&cl->hnode, &clhash->hash[h]); | |
585 | clhash->hashelems++; | |
586 | } | |
587 | EXPORT_SYMBOL(qdisc_class_hash_insert); | |
588 | ||
589 | void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash, | |
590 | struct Qdisc_class_common *cl) | |
591 | { | |
592 | hlist_del(&cl->hnode); | |
593 | clhash->hashelems--; | |
594 | } | |
595 | EXPORT_SYMBOL(qdisc_class_hash_remove); | |
596 | ||
597 | /* Allocate an unique handle from space managed by kernel */ | |
598 | ||
599 | static u32 qdisc_alloc_handle(struct net_device *dev) | |
600 | { | |
601 | int i = 0x10000; | |
602 | static u32 autohandle = TC_H_MAKE(0x80000000U, 0); | |
603 | ||
604 | do { | |
605 | autohandle += TC_H_MAKE(0x10000U, 0); | |
606 | if (autohandle == TC_H_MAKE(TC_H_ROOT, 0)) | |
607 | autohandle = TC_H_MAKE(0x80000000U, 0); | |
608 | } while (qdisc_lookup(dev, autohandle) && --i > 0); | |
609 | ||
610 | return i>0 ? autohandle : 0; | |
611 | } | |
612 | ||
613 | void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n) | |
614 | { | |
615 | const struct Qdisc_class_ops *cops; | |
616 | unsigned long cl; | |
617 | u32 parentid; | |
618 | ||
619 | if (n == 0) | |
620 | return; | |
621 | while ((parentid = sch->parent)) { | |
622 | if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS)) | |
623 | return; | |
624 | ||
625 | sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid)); | |
626 | if (sch == NULL) { | |
627 | WARN_ON(parentid != TC_H_ROOT); | |
628 | return; | |
629 | } | |
630 | cops = sch->ops->cl_ops; | |
631 | if (cops->qlen_notify) { | |
632 | cl = cops->get(sch, parentid); | |
633 | cops->qlen_notify(sch, cl); | |
634 | cops->put(sch, cl); | |
635 | } | |
636 | sch->q.qlen -= n; | |
637 | } | |
638 | } | |
639 | EXPORT_SYMBOL(qdisc_tree_decrease_qlen); | |
640 | ||
641 | static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid, | |
642 | struct Qdisc *old, struct Qdisc *new) | |
643 | { | |
644 | if (new || old) | |
645 | qdisc_notify(skb, n, clid, old, new); | |
646 | ||
647 | if (old) | |
648 | qdisc_destroy(old); | |
649 | } | |
650 | ||
651 | /* Graft qdisc "new" to class "classid" of qdisc "parent" or | |
652 | * to device "dev". | |
653 | * | |
654 | * When appropriate send a netlink notification using 'skb' | |
655 | * and "n". | |
656 | * | |
657 | * On success, destroy old qdisc. | |
658 | */ | |
659 | ||
660 | static int qdisc_graft(struct net_device *dev, struct Qdisc *parent, | |
661 | struct sk_buff *skb, struct nlmsghdr *n, u32 classid, | |
662 | struct Qdisc *new, struct Qdisc *old) | |
663 | { | |
664 | struct Qdisc *q = old; | |
665 | int err = 0; | |
666 | ||
667 | if (parent == NULL) { | |
668 | unsigned int i, num_q, ingress; | |
669 | ||
670 | ingress = 0; | |
671 | num_q = dev->num_tx_queues; | |
672 | if ((q && q->flags & TCQ_F_INGRESS) || | |
673 | (new && new->flags & TCQ_F_INGRESS)) { | |
674 | num_q = 1; | |
675 | ingress = 1; | |
676 | } | |
677 | ||
678 | if (dev->flags & IFF_UP) | |
679 | dev_deactivate(dev); | |
680 | ||
681 | for (i = 0; i < num_q; i++) { | |
682 | struct netdev_queue *dev_queue = &dev->rx_queue; | |
683 | ||
684 | if (!ingress) | |
685 | dev_queue = netdev_get_tx_queue(dev, i); | |
686 | ||
687 | old = dev_graft_qdisc(dev_queue, new); | |
688 | if (new && i > 0) | |
689 | atomic_inc(&new->refcnt); | |
690 | ||
691 | qdisc_destroy(old); | |
692 | } | |
693 | ||
694 | notify_and_destroy(skb, n, classid, dev->qdisc, new); | |
695 | if (new) | |
696 | atomic_inc(&new->refcnt); | |
697 | dev->qdisc = new ? : &noop_qdisc; | |
698 | ||
699 | if (dev->flags & IFF_UP) | |
700 | dev_activate(dev); | |
701 | } else { | |
702 | const struct Qdisc_class_ops *cops = parent->ops->cl_ops; | |
703 | ||
704 | err = -EOPNOTSUPP; | |
705 | if (cops && cops->graft) { | |
706 | unsigned long cl = cops->get(parent, classid); | |
707 | if (cl) { | |
708 | err = cops->graft(parent, cl, new, &old); | |
709 | cops->put(parent, cl); | |
710 | } else | |
711 | err = -ENOENT; | |
712 | } | |
713 | if (!err) | |
714 | notify_and_destroy(skb, n, classid, old, new); | |
715 | } | |
716 | return err; | |
717 | } | |
718 | ||
719 | /* lockdep annotation is needed for ingress; egress gets it only for name */ | |
720 | static struct lock_class_key qdisc_tx_lock; | |
721 | static struct lock_class_key qdisc_rx_lock; | |
722 | ||
723 | /* | |
724 | Allocate and initialize new qdisc. | |
725 | ||
726 | Parameters are passed via opt. | |
727 | */ | |
728 | ||
729 | static struct Qdisc * | |
730 | qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue, | |
731 | u32 parent, u32 handle, struct nlattr **tca, int *errp) | |
732 | { | |
733 | int err; | |
734 | struct nlattr *kind = tca[TCA_KIND]; | |
735 | struct Qdisc *sch; | |
736 | struct Qdisc_ops *ops; | |
737 | struct qdisc_size_table *stab; | |
738 | ||
739 | ops = qdisc_lookup_ops(kind); | |
740 | #ifdef CONFIG_MODULES | |
741 | if (ops == NULL && kind != NULL) { | |
742 | char name[IFNAMSIZ]; | |
743 | if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) { | |
744 | /* We dropped the RTNL semaphore in order to | |
745 | * perform the module load. So, even if we | |
746 | * succeeded in loading the module we have to | |
747 | * tell the caller to replay the request. We | |
748 | * indicate this using -EAGAIN. | |
749 | * We replay the request because the device may | |
750 | * go away in the mean time. | |
751 | */ | |
752 | rtnl_unlock(); | |
753 | request_module("sch_%s", name); | |
754 | rtnl_lock(); | |
755 | ops = qdisc_lookup_ops(kind); | |
756 | if (ops != NULL) { | |
757 | /* We will try again qdisc_lookup_ops, | |
758 | * so don't keep a reference. | |
759 | */ | |
760 | module_put(ops->owner); | |
761 | err = -EAGAIN; | |
762 | goto err_out; | |
763 | } | |
764 | } | |
765 | } | |
766 | #endif | |
767 | ||
768 | err = -ENOENT; | |
769 | if (ops == NULL) | |
770 | goto err_out; | |
771 | ||
772 | sch = qdisc_alloc(dev_queue, ops); | |
773 | if (IS_ERR(sch)) { | |
774 | err = PTR_ERR(sch); | |
775 | goto err_out2; | |
776 | } | |
777 | ||
778 | sch->parent = parent; | |
779 | ||
780 | if (handle == TC_H_INGRESS) { | |
781 | sch->flags |= TCQ_F_INGRESS; | |
782 | handle = TC_H_MAKE(TC_H_INGRESS, 0); | |
783 | lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock); | |
784 | } else { | |
785 | if (handle == 0) { | |
786 | handle = qdisc_alloc_handle(dev); | |
787 | err = -ENOMEM; | |
788 | if (handle == 0) | |
789 | goto err_out3; | |
790 | } | |
791 | lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock); | |
792 | } | |
793 | ||
794 | sch->handle = handle; | |
795 | ||
796 | if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) { | |
797 | if (tca[TCA_STAB]) { | |
798 | stab = qdisc_get_stab(tca[TCA_STAB]); | |
799 | if (IS_ERR(stab)) { | |
800 | err = PTR_ERR(stab); | |
801 | goto err_out3; | |
802 | } | |
803 | sch->stab = stab; | |
804 | } | |
805 | if (tca[TCA_RATE]) { | |
806 | spinlock_t *root_lock; | |
807 | ||
808 | if ((sch->parent != TC_H_ROOT) && | |
809 | !(sch->flags & TCQ_F_INGRESS)) | |
810 | root_lock = qdisc_root_sleeping_lock(sch); | |
811 | else | |
812 | root_lock = qdisc_lock(sch); | |
813 | ||
814 | err = gen_new_estimator(&sch->bstats, &sch->rate_est, | |
815 | root_lock, tca[TCA_RATE]); | |
816 | if (err) { | |
817 | /* | |
818 | * Any broken qdiscs that would require | |
819 | * a ops->reset() here? The qdisc was never | |
820 | * in action so it shouldn't be necessary. | |
821 | */ | |
822 | if (ops->destroy) | |
823 | ops->destroy(sch); | |
824 | goto err_out3; | |
825 | } | |
826 | } | |
827 | ||
828 | qdisc_list_add(sch); | |
829 | ||
830 | return sch; | |
831 | } | |
832 | err_out3: | |
833 | qdisc_put_stab(sch->stab); | |
834 | dev_put(dev); | |
835 | kfree((char *) sch - sch->padded); | |
836 | err_out2: | |
837 | module_put(ops->owner); | |
838 | err_out: | |
839 | *errp = err; | |
840 | return NULL; | |
841 | } | |
842 | ||
843 | static int qdisc_change(struct Qdisc *sch, struct nlattr **tca) | |
844 | { | |
845 | struct qdisc_size_table *stab = NULL; | |
846 | int err = 0; | |
847 | ||
848 | if (tca[TCA_OPTIONS]) { | |
849 | if (sch->ops->change == NULL) | |
850 | return -EINVAL; | |
851 | err = sch->ops->change(sch, tca[TCA_OPTIONS]); | |
852 | if (err) | |
853 | return err; | |
854 | } | |
855 | ||
856 | if (tca[TCA_STAB]) { | |
857 | stab = qdisc_get_stab(tca[TCA_STAB]); | |
858 | if (IS_ERR(stab)) | |
859 | return PTR_ERR(stab); | |
860 | } | |
861 | ||
862 | qdisc_put_stab(sch->stab); | |
863 | sch->stab = stab; | |
864 | ||
865 | if (tca[TCA_RATE]) | |
866 | /* NB: ignores errors from replace_estimator | |
867 | because change can't be undone. */ | |
868 | gen_replace_estimator(&sch->bstats, &sch->rate_est, | |
869 | qdisc_root_sleeping_lock(sch), | |
870 | tca[TCA_RATE]); | |
871 | ||
872 | return 0; | |
873 | } | |
874 | ||
875 | struct check_loop_arg | |
876 | { | |
877 | struct qdisc_walker w; | |
878 | struct Qdisc *p; | |
879 | int depth; | |
880 | }; | |
881 | ||
882 | static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w); | |
883 | ||
884 | static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth) | |
885 | { | |
886 | struct check_loop_arg arg; | |
887 | ||
888 | if (q->ops->cl_ops == NULL) | |
889 | return 0; | |
890 | ||
891 | arg.w.stop = arg.w.skip = arg.w.count = 0; | |
892 | arg.w.fn = check_loop_fn; | |
893 | arg.depth = depth; | |
894 | arg.p = p; | |
895 | q->ops->cl_ops->walk(q, &arg.w); | |
896 | return arg.w.stop ? -ELOOP : 0; | |
897 | } | |
898 | ||
899 | static int | |
900 | check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w) | |
901 | { | |
902 | struct Qdisc *leaf; | |
903 | const struct Qdisc_class_ops *cops = q->ops->cl_ops; | |
904 | struct check_loop_arg *arg = (struct check_loop_arg *)w; | |
905 | ||
906 | leaf = cops->leaf(q, cl); | |
907 | if (leaf) { | |
908 | if (leaf == arg->p || arg->depth > 7) | |
909 | return -ELOOP; | |
910 | return check_loop(leaf, arg->p, arg->depth + 1); | |
911 | } | |
912 | return 0; | |
913 | } | |
914 | ||
915 | /* | |
916 | * Delete/get qdisc. | |
917 | */ | |
918 | ||
919 | static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
920 | { | |
921 | struct net *net = sock_net(skb->sk); | |
922 | struct tcmsg *tcm = NLMSG_DATA(n); | |
923 | struct nlattr *tca[TCA_MAX + 1]; | |
924 | struct net_device *dev; | |
925 | u32 clid = tcm->tcm_parent; | |
926 | struct Qdisc *q = NULL; | |
927 | struct Qdisc *p = NULL; | |
928 | int err; | |
929 | ||
930 | if (net != &init_net) | |
931 | return -EINVAL; | |
932 | ||
933 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) | |
934 | return -ENODEV; | |
935 | ||
936 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); | |
937 | if (err < 0) | |
938 | return err; | |
939 | ||
940 | if (clid) { | |
941 | if (clid != TC_H_ROOT) { | |
942 | if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) { | |
943 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
944 | return -ENOENT; | |
945 | q = qdisc_leaf(p, clid); | |
946 | } else { /* ingress */ | |
947 | q = dev->rx_queue.qdisc_sleeping; | |
948 | } | |
949 | } else { | |
950 | q = dev->qdisc; | |
951 | } | |
952 | if (!q) | |
953 | return -ENOENT; | |
954 | ||
955 | if (tcm->tcm_handle && q->handle != tcm->tcm_handle) | |
956 | return -EINVAL; | |
957 | } else { | |
958 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
959 | return -ENOENT; | |
960 | } | |
961 | ||
962 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) | |
963 | return -EINVAL; | |
964 | ||
965 | if (n->nlmsg_type == RTM_DELQDISC) { | |
966 | if (!clid) | |
967 | return -EINVAL; | |
968 | if (q->handle == 0) | |
969 | return -ENOENT; | |
970 | if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0) | |
971 | return err; | |
972 | } else { | |
973 | qdisc_notify(skb, n, clid, NULL, q); | |
974 | } | |
975 | return 0; | |
976 | } | |
977 | ||
978 | /* | |
979 | Create/change qdisc. | |
980 | */ | |
981 | ||
982 | static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
983 | { | |
984 | struct net *net = sock_net(skb->sk); | |
985 | struct tcmsg *tcm; | |
986 | struct nlattr *tca[TCA_MAX + 1]; | |
987 | struct net_device *dev; | |
988 | u32 clid; | |
989 | struct Qdisc *q, *p; | |
990 | int err; | |
991 | ||
992 | if (net != &init_net) | |
993 | return -EINVAL; | |
994 | ||
995 | replay: | |
996 | /* Reinit, just in case something touches this. */ | |
997 | tcm = NLMSG_DATA(n); | |
998 | clid = tcm->tcm_parent; | |
999 | q = p = NULL; | |
1000 | ||
1001 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) | |
1002 | return -ENODEV; | |
1003 | ||
1004 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); | |
1005 | if (err < 0) | |
1006 | return err; | |
1007 | ||
1008 | if (clid) { | |
1009 | if (clid != TC_H_ROOT) { | |
1010 | if (clid != TC_H_INGRESS) { | |
1011 | if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL) | |
1012 | return -ENOENT; | |
1013 | q = qdisc_leaf(p, clid); | |
1014 | } else { /*ingress */ | |
1015 | q = dev->rx_queue.qdisc_sleeping; | |
1016 | } | |
1017 | } else { | |
1018 | q = dev->qdisc; | |
1019 | } | |
1020 | ||
1021 | /* It may be default qdisc, ignore it */ | |
1022 | if (q && q->handle == 0) | |
1023 | q = NULL; | |
1024 | ||
1025 | if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) { | |
1026 | if (tcm->tcm_handle) { | |
1027 | if (q && !(n->nlmsg_flags&NLM_F_REPLACE)) | |
1028 | return -EEXIST; | |
1029 | if (TC_H_MIN(tcm->tcm_handle)) | |
1030 | return -EINVAL; | |
1031 | if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL) | |
1032 | goto create_n_graft; | |
1033 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1034 | return -EEXIST; | |
1035 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) | |
1036 | return -EINVAL; | |
1037 | if (q == p || | |
1038 | (p && check_loop(q, p, 0))) | |
1039 | return -ELOOP; | |
1040 | atomic_inc(&q->refcnt); | |
1041 | goto graft; | |
1042 | } else { | |
1043 | if (q == NULL) | |
1044 | goto create_n_graft; | |
1045 | ||
1046 | /* This magic test requires explanation. | |
1047 | * | |
1048 | * We know, that some child q is already | |
1049 | * attached to this parent and have choice: | |
1050 | * either to change it or to create/graft new one. | |
1051 | * | |
1052 | * 1. We are allowed to create/graft only | |
1053 | * if CREATE and REPLACE flags are set. | |
1054 | * | |
1055 | * 2. If EXCL is set, requestor wanted to say, | |
1056 | * that qdisc tcm_handle is not expected | |
1057 | * to exist, so that we choose create/graft too. | |
1058 | * | |
1059 | * 3. The last case is when no flags are set. | |
1060 | * Alas, it is sort of hole in API, we | |
1061 | * cannot decide what to do unambiguously. | |
1062 | * For now we select create/graft, if | |
1063 | * user gave KIND, which does not match existing. | |
1064 | */ | |
1065 | if ((n->nlmsg_flags&NLM_F_CREATE) && | |
1066 | (n->nlmsg_flags&NLM_F_REPLACE) && | |
1067 | ((n->nlmsg_flags&NLM_F_EXCL) || | |
1068 | (tca[TCA_KIND] && | |
1069 | nla_strcmp(tca[TCA_KIND], q->ops->id)))) | |
1070 | goto create_n_graft; | |
1071 | } | |
1072 | } | |
1073 | } else { | |
1074 | if (!tcm->tcm_handle) | |
1075 | return -EINVAL; | |
1076 | q = qdisc_lookup(dev, tcm->tcm_handle); | |
1077 | } | |
1078 | ||
1079 | /* Change qdisc parameters */ | |
1080 | if (q == NULL) | |
1081 | return -ENOENT; | |
1082 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1083 | return -EEXIST; | |
1084 | if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) | |
1085 | return -EINVAL; | |
1086 | err = qdisc_change(q, tca); | |
1087 | if (err == 0) | |
1088 | qdisc_notify(skb, n, clid, NULL, q); | |
1089 | return err; | |
1090 | ||
1091 | create_n_graft: | |
1092 | if (!(n->nlmsg_flags&NLM_F_CREATE)) | |
1093 | return -ENOENT; | |
1094 | if (clid == TC_H_INGRESS) | |
1095 | q = qdisc_create(dev, &dev->rx_queue, | |
1096 | tcm->tcm_parent, tcm->tcm_parent, | |
1097 | tca, &err); | |
1098 | else | |
1099 | q = qdisc_create(dev, netdev_get_tx_queue(dev, 0), | |
1100 | tcm->tcm_parent, tcm->tcm_handle, | |
1101 | tca, &err); | |
1102 | if (q == NULL) { | |
1103 | if (err == -EAGAIN) | |
1104 | goto replay; | |
1105 | return err; | |
1106 | } | |
1107 | ||
1108 | graft: | |
1109 | err = qdisc_graft(dev, p, skb, n, clid, q, NULL); | |
1110 | if (err) { | |
1111 | if (q) | |
1112 | qdisc_destroy(q); | |
1113 | return err; | |
1114 | } | |
1115 | ||
1116 | return 0; | |
1117 | } | |
1118 | ||
1119 | static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid, | |
1120 | u32 pid, u32 seq, u16 flags, int event) | |
1121 | { | |
1122 | struct tcmsg *tcm; | |
1123 | struct nlmsghdr *nlh; | |
1124 | unsigned char *b = skb_tail_pointer(skb); | |
1125 | struct gnet_dump d; | |
1126 | ||
1127 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); | |
1128 | tcm = NLMSG_DATA(nlh); | |
1129 | tcm->tcm_family = AF_UNSPEC; | |
1130 | tcm->tcm__pad1 = 0; | |
1131 | tcm->tcm__pad2 = 0; | |
1132 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; | |
1133 | tcm->tcm_parent = clid; | |
1134 | tcm->tcm_handle = q->handle; | |
1135 | tcm->tcm_info = atomic_read(&q->refcnt); | |
1136 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); | |
1137 | if (q->ops->dump && q->ops->dump(q, skb) < 0) | |
1138 | goto nla_put_failure; | |
1139 | q->qstats.qlen = q->q.qlen; | |
1140 | ||
1141 | if (q->stab && qdisc_dump_stab(skb, q->stab) < 0) | |
1142 | goto nla_put_failure; | |
1143 | ||
1144 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, | |
1145 | qdisc_root_sleeping_lock(q), &d) < 0) | |
1146 | goto nla_put_failure; | |
1147 | ||
1148 | if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0) | |
1149 | goto nla_put_failure; | |
1150 | ||
1151 | if (gnet_stats_copy_basic(&d, &q->bstats) < 0 || | |
1152 | gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 || | |
1153 | gnet_stats_copy_queue(&d, &q->qstats) < 0) | |
1154 | goto nla_put_failure; | |
1155 | ||
1156 | if (gnet_stats_finish_copy(&d) < 0) | |
1157 | goto nla_put_failure; | |
1158 | ||
1159 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; | |
1160 | return skb->len; | |
1161 | ||
1162 | nlmsg_failure: | |
1163 | nla_put_failure: | |
1164 | nlmsg_trim(skb, b); | |
1165 | return -1; | |
1166 | } | |
1167 | ||
1168 | static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1169 | u32 clid, struct Qdisc *old, struct Qdisc *new) | |
1170 | { | |
1171 | struct sk_buff *skb; | |
1172 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1173 | ||
1174 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1175 | if (!skb) | |
1176 | return -ENOBUFS; | |
1177 | ||
1178 | if (old && old->handle) { | |
1179 | if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0) | |
1180 | goto err_out; | |
1181 | } | |
1182 | if (new) { | |
1183 | if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0) | |
1184 | goto err_out; | |
1185 | } | |
1186 | ||
1187 | if (skb->len) | |
1188 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); | |
1189 | ||
1190 | err_out: | |
1191 | kfree_skb(skb); | |
1192 | return -EINVAL; | |
1193 | } | |
1194 | ||
1195 | static bool tc_qdisc_dump_ignore(struct Qdisc *q) | |
1196 | { | |
1197 | return (q->flags & TCQ_F_BUILTIN) ? true : false; | |
1198 | } | |
1199 | ||
1200 | static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb, | |
1201 | struct netlink_callback *cb, | |
1202 | int *q_idx_p, int s_q_idx) | |
1203 | { | |
1204 | int ret = 0, q_idx = *q_idx_p; | |
1205 | struct Qdisc *q; | |
1206 | ||
1207 | if (!root) | |
1208 | return 0; | |
1209 | ||
1210 | q = root; | |
1211 | if (q_idx < s_q_idx) { | |
1212 | q_idx++; | |
1213 | } else { | |
1214 | if (!tc_qdisc_dump_ignore(q) && | |
1215 | tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
1216 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) | |
1217 | goto done; | |
1218 | q_idx++; | |
1219 | } | |
1220 | list_for_each_entry(q, &root->list, list) { | |
1221 | if (q_idx < s_q_idx) { | |
1222 | q_idx++; | |
1223 | continue; | |
1224 | } | |
1225 | if (!tc_qdisc_dump_ignore(q) && | |
1226 | tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid, | |
1227 | cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0) | |
1228 | goto done; | |
1229 | q_idx++; | |
1230 | } | |
1231 | ||
1232 | out: | |
1233 | *q_idx_p = q_idx; | |
1234 | return ret; | |
1235 | done: | |
1236 | ret = -1; | |
1237 | goto out; | |
1238 | } | |
1239 | ||
1240 | static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb) | |
1241 | { | |
1242 | struct net *net = sock_net(skb->sk); | |
1243 | int idx, q_idx; | |
1244 | int s_idx, s_q_idx; | |
1245 | struct net_device *dev; | |
1246 | ||
1247 | if (net != &init_net) | |
1248 | return 0; | |
1249 | ||
1250 | s_idx = cb->args[0]; | |
1251 | s_q_idx = q_idx = cb->args[1]; | |
1252 | read_lock(&dev_base_lock); | |
1253 | idx = 0; | |
1254 | for_each_netdev(&init_net, dev) { | |
1255 | struct netdev_queue *dev_queue; | |
1256 | ||
1257 | if (idx < s_idx) | |
1258 | goto cont; | |
1259 | if (idx > s_idx) | |
1260 | s_q_idx = 0; | |
1261 | q_idx = 0; | |
1262 | ||
1263 | if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0) | |
1264 | goto done; | |
1265 | ||
1266 | dev_queue = &dev->rx_queue; | |
1267 | if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0) | |
1268 | goto done; | |
1269 | ||
1270 | cont: | |
1271 | idx++; | |
1272 | } | |
1273 | ||
1274 | done: | |
1275 | read_unlock(&dev_base_lock); | |
1276 | ||
1277 | cb->args[0] = idx; | |
1278 | cb->args[1] = q_idx; | |
1279 | ||
1280 | return skb->len; | |
1281 | } | |
1282 | ||
1283 | ||
1284 | ||
1285 | /************************************************ | |
1286 | * Traffic classes manipulation. * | |
1287 | ************************************************/ | |
1288 | ||
1289 | ||
1290 | ||
1291 | static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg) | |
1292 | { | |
1293 | struct net *net = sock_net(skb->sk); | |
1294 | struct tcmsg *tcm = NLMSG_DATA(n); | |
1295 | struct nlattr *tca[TCA_MAX + 1]; | |
1296 | struct net_device *dev; | |
1297 | struct Qdisc *q = NULL; | |
1298 | const struct Qdisc_class_ops *cops; | |
1299 | unsigned long cl = 0; | |
1300 | unsigned long new_cl; | |
1301 | u32 pid = tcm->tcm_parent; | |
1302 | u32 clid = tcm->tcm_handle; | |
1303 | u32 qid = TC_H_MAJ(clid); | |
1304 | int err; | |
1305 | ||
1306 | if (net != &init_net) | |
1307 | return -EINVAL; | |
1308 | ||
1309 | if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) | |
1310 | return -ENODEV; | |
1311 | ||
1312 | err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL); | |
1313 | if (err < 0) | |
1314 | return err; | |
1315 | ||
1316 | /* | |
1317 | parent == TC_H_UNSPEC - unspecified parent. | |
1318 | parent == TC_H_ROOT - class is root, which has no parent. | |
1319 | parent == X:0 - parent is root class. | |
1320 | parent == X:Y - parent is a node in hierarchy. | |
1321 | parent == 0:Y - parent is X:Y, where X:0 is qdisc. | |
1322 | ||
1323 | handle == 0:0 - generate handle from kernel pool. | |
1324 | handle == 0:Y - class is X:Y, where X:0 is qdisc. | |
1325 | handle == X:Y - clear. | |
1326 | handle == X:0 - root class. | |
1327 | */ | |
1328 | ||
1329 | /* Step 1. Determine qdisc handle X:0 */ | |
1330 | ||
1331 | if (pid != TC_H_ROOT) { | |
1332 | u32 qid1 = TC_H_MAJ(pid); | |
1333 | ||
1334 | if (qid && qid1) { | |
1335 | /* If both majors are known, they must be identical. */ | |
1336 | if (qid != qid1) | |
1337 | return -EINVAL; | |
1338 | } else if (qid1) { | |
1339 | qid = qid1; | |
1340 | } else if (qid == 0) | |
1341 | qid = dev->qdisc->handle; | |
1342 | ||
1343 | /* Now qid is genuine qdisc handle consistent | |
1344 | both with parent and child. | |
1345 | ||
1346 | TC_H_MAJ(pid) still may be unspecified, complete it now. | |
1347 | */ | |
1348 | if (pid) | |
1349 | pid = TC_H_MAKE(qid, pid); | |
1350 | } else { | |
1351 | if (qid == 0) | |
1352 | qid = dev->qdisc->handle; | |
1353 | } | |
1354 | ||
1355 | /* OK. Locate qdisc */ | |
1356 | if ((q = qdisc_lookup(dev, qid)) == NULL) | |
1357 | return -ENOENT; | |
1358 | ||
1359 | /* An check that it supports classes */ | |
1360 | cops = q->ops->cl_ops; | |
1361 | if (cops == NULL) | |
1362 | return -EINVAL; | |
1363 | ||
1364 | /* Now try to get class */ | |
1365 | if (clid == 0) { | |
1366 | if (pid == TC_H_ROOT) | |
1367 | clid = qid; | |
1368 | } else | |
1369 | clid = TC_H_MAKE(qid, clid); | |
1370 | ||
1371 | if (clid) | |
1372 | cl = cops->get(q, clid); | |
1373 | ||
1374 | if (cl == 0) { | |
1375 | err = -ENOENT; | |
1376 | if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE)) | |
1377 | goto out; | |
1378 | } else { | |
1379 | switch (n->nlmsg_type) { | |
1380 | case RTM_NEWTCLASS: | |
1381 | err = -EEXIST; | |
1382 | if (n->nlmsg_flags&NLM_F_EXCL) | |
1383 | goto out; | |
1384 | break; | |
1385 | case RTM_DELTCLASS: | |
1386 | err = -EOPNOTSUPP; | |
1387 | if (cops->delete) | |
1388 | err = cops->delete(q, cl); | |
1389 | if (err == 0) | |
1390 | tclass_notify(skb, n, q, cl, RTM_DELTCLASS); | |
1391 | goto out; | |
1392 | case RTM_GETTCLASS: | |
1393 | err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS); | |
1394 | goto out; | |
1395 | default: | |
1396 | err = -EINVAL; | |
1397 | goto out; | |
1398 | } | |
1399 | } | |
1400 | ||
1401 | new_cl = cl; | |
1402 | err = -EOPNOTSUPP; | |
1403 | if (cops->change) | |
1404 | err = cops->change(q, clid, pid, tca, &new_cl); | |
1405 | if (err == 0) | |
1406 | tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS); | |
1407 | ||
1408 | out: | |
1409 | if (cl) | |
1410 | cops->put(q, cl); | |
1411 | ||
1412 | return err; | |
1413 | } | |
1414 | ||
1415 | ||
1416 | static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q, | |
1417 | unsigned long cl, | |
1418 | u32 pid, u32 seq, u16 flags, int event) | |
1419 | { | |
1420 | struct tcmsg *tcm; | |
1421 | struct nlmsghdr *nlh; | |
1422 | unsigned char *b = skb_tail_pointer(skb); | |
1423 | struct gnet_dump d; | |
1424 | const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops; | |
1425 | ||
1426 | nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags); | |
1427 | tcm = NLMSG_DATA(nlh); | |
1428 | tcm->tcm_family = AF_UNSPEC; | |
1429 | tcm->tcm_ifindex = qdisc_dev(q)->ifindex; | |
1430 | tcm->tcm_parent = q->handle; | |
1431 | tcm->tcm_handle = q->handle; | |
1432 | tcm->tcm_info = 0; | |
1433 | NLA_PUT_STRING(skb, TCA_KIND, q->ops->id); | |
1434 | if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0) | |
1435 | goto nla_put_failure; | |
1436 | ||
1437 | if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS, | |
1438 | qdisc_root_sleeping_lock(q), &d) < 0) | |
1439 | goto nla_put_failure; | |
1440 | ||
1441 | if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0) | |
1442 | goto nla_put_failure; | |
1443 | ||
1444 | if (gnet_stats_finish_copy(&d) < 0) | |
1445 | goto nla_put_failure; | |
1446 | ||
1447 | nlh->nlmsg_len = skb_tail_pointer(skb) - b; | |
1448 | return skb->len; | |
1449 | ||
1450 | nlmsg_failure: | |
1451 | nla_put_failure: | |
1452 | nlmsg_trim(skb, b); | |
1453 | return -1; | |
1454 | } | |
1455 | ||
1456 | static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n, | |
1457 | struct Qdisc *q, unsigned long cl, int event) | |
1458 | { | |
1459 | struct sk_buff *skb; | |
1460 | u32 pid = oskb ? NETLINK_CB(oskb).pid : 0; | |
1461 | ||
1462 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); | |
1463 | if (!skb) | |
1464 | return -ENOBUFS; | |
1465 | ||
1466 | if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) { | |
1467 | kfree_skb(skb); | |
1468 | return -EINVAL; | |
1469 | } | |
1470 | ||
1471 | return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO); | |
1472 | } | |
1473 | ||
1474 | struct qdisc_dump_args | |
1475 | { | |
1476 | struct qdisc_walker w; | |
1477 | struct sk_buff *skb; | |
1478 | struct netlink_callback *cb; | |
1479 | }; | |
1480 | ||
1481 | static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg) | |
1482 | { | |
1483 | struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg; | |
1484 | ||
1485 | return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid, | |
1486 | a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS); | |
1487 | } | |
1488 | ||
1489 | static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb, | |
1490 | struct tcmsg *tcm, struct netlink_callback *cb, | |
1491 | int *t_p, int s_t) | |
1492 | { | |
1493 | struct qdisc_dump_args arg; | |
1494 | ||
1495 | if (tc_qdisc_dump_ignore(q) || | |
1496 | *t_p < s_t || !q->ops->cl_ops || | |
1497 | (tcm->tcm_parent && | |
1498 | TC_H_MAJ(tcm->tcm_parent) != q->handle)) { | |
1499 | (*t_p)++; | |
1500 | return 0; | |
1501 | } | |
1502 | if (*t_p > s_t) | |
1503 | memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0])); | |
1504 | arg.w.fn = qdisc_class_dump; | |
1505 | arg.skb = skb; | |
1506 | arg.cb = cb; | |
1507 | arg.w.stop = 0; | |
1508 | arg.w.skip = cb->args[1]; | |
1509 | arg.w.count = 0; | |
1510 | q->ops->cl_ops->walk(q, &arg.w); | |
1511 | cb->args[1] = arg.w.count; | |
1512 | if (arg.w.stop) | |
1513 | return -1; | |
1514 | (*t_p)++; | |
1515 | return 0; | |
1516 | } | |
1517 | ||
1518 | static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb, | |
1519 | struct tcmsg *tcm, struct netlink_callback *cb, | |
1520 | int *t_p, int s_t) | |
1521 | { | |
1522 | struct Qdisc *q; | |
1523 | ||
1524 | if (!root) | |
1525 | return 0; | |
1526 | ||
1527 | if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0) | |
1528 | return -1; | |
1529 | ||
1530 | list_for_each_entry(q, &root->list, list) { | |
1531 | if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0) | |
1532 | return -1; | |
1533 | } | |
1534 | ||
1535 | return 0; | |
1536 | } | |
1537 | ||
1538 | static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb) | |
1539 | { | |
1540 | struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh); | |
1541 | struct net *net = sock_net(skb->sk); | |
1542 | struct netdev_queue *dev_queue; | |
1543 | struct net_device *dev; | |
1544 | int t, s_t; | |
1545 | ||
1546 | if (net != &init_net) | |
1547 | return 0; | |
1548 | ||
1549 | if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm))) | |
1550 | return 0; | |
1551 | if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL) | |
1552 | return 0; | |
1553 | ||
1554 | s_t = cb->args[0]; | |
1555 | t = 0; | |
1556 | ||
1557 | if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0) | |
1558 | goto done; | |
1559 | ||
1560 | dev_queue = &dev->rx_queue; | |
1561 | if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0) | |
1562 | goto done; | |
1563 | ||
1564 | done: | |
1565 | cb->args[0] = t; | |
1566 | ||
1567 | dev_put(dev); | |
1568 | return skb->len; | |
1569 | } | |
1570 | ||
1571 | /* Main classifier routine: scans classifier chain attached | |
1572 | to this qdisc, (optionally) tests for protocol and asks | |
1573 | specific classifiers. | |
1574 | */ | |
1575 | int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp, | |
1576 | struct tcf_result *res) | |
1577 | { | |
1578 | __be16 protocol = skb->protocol; | |
1579 | int err = 0; | |
1580 | ||
1581 | for (; tp; tp = tp->next) { | |
1582 | if ((tp->protocol == protocol || | |
1583 | tp->protocol == htons(ETH_P_ALL)) && | |
1584 | (err = tp->classify(skb, tp, res)) >= 0) { | |
1585 | #ifdef CONFIG_NET_CLS_ACT | |
1586 | if (err != TC_ACT_RECLASSIFY && skb->tc_verd) | |
1587 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0); | |
1588 | #endif | |
1589 | return err; | |
1590 | } | |
1591 | } | |
1592 | return -1; | |
1593 | } | |
1594 | EXPORT_SYMBOL(tc_classify_compat); | |
1595 | ||
1596 | int tc_classify(struct sk_buff *skb, struct tcf_proto *tp, | |
1597 | struct tcf_result *res) | |
1598 | { | |
1599 | int err = 0; | |
1600 | __be16 protocol; | |
1601 | #ifdef CONFIG_NET_CLS_ACT | |
1602 | struct tcf_proto *otp = tp; | |
1603 | reclassify: | |
1604 | #endif | |
1605 | protocol = skb->protocol; | |
1606 | ||
1607 | err = tc_classify_compat(skb, tp, res); | |
1608 | #ifdef CONFIG_NET_CLS_ACT | |
1609 | if (err == TC_ACT_RECLASSIFY) { | |
1610 | u32 verd = G_TC_VERD(skb->tc_verd); | |
1611 | tp = otp; | |
1612 | ||
1613 | if (verd++ >= MAX_REC_LOOP) { | |
1614 | printk("rule prio %u protocol %02x reclassify loop, " | |
1615 | "packet dropped\n", | |
1616 | tp->prio&0xffff, ntohs(tp->protocol)); | |
1617 | return TC_ACT_SHOT; | |
1618 | } | |
1619 | skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd); | |
1620 | goto reclassify; | |
1621 | } | |
1622 | #endif | |
1623 | return err; | |
1624 | } | |
1625 | EXPORT_SYMBOL(tc_classify); | |
1626 | ||
1627 | void tcf_destroy(struct tcf_proto *tp) | |
1628 | { | |
1629 | tp->ops->destroy(tp); | |
1630 | module_put(tp->ops->owner); | |
1631 | kfree(tp); | |
1632 | } | |
1633 | ||
1634 | void tcf_destroy_chain(struct tcf_proto **fl) | |
1635 | { | |
1636 | struct tcf_proto *tp; | |
1637 | ||
1638 | while ((tp = *fl) != NULL) { | |
1639 | *fl = tp->next; | |
1640 | tcf_destroy(tp); | |
1641 | } | |
1642 | } | |
1643 | EXPORT_SYMBOL(tcf_destroy_chain); | |
1644 | ||
1645 | #ifdef CONFIG_PROC_FS | |
1646 | static int psched_show(struct seq_file *seq, void *v) | |
1647 | { | |
1648 | struct timespec ts; | |
1649 | ||
1650 | hrtimer_get_res(CLOCK_MONOTONIC, &ts); | |
1651 | seq_printf(seq, "%08x %08x %08x %08x\n", | |
1652 | (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1), | |
1653 | 1000000, | |
1654 | (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts))); | |
1655 | ||
1656 | return 0; | |
1657 | } | |
1658 | ||
1659 | static int psched_open(struct inode *inode, struct file *file) | |
1660 | { | |
1661 | return single_open(file, psched_show, PDE(inode)->data); | |
1662 | } | |
1663 | ||
1664 | static const struct file_operations psched_fops = { | |
1665 | .owner = THIS_MODULE, | |
1666 | .open = psched_open, | |
1667 | .read = seq_read, | |
1668 | .llseek = seq_lseek, | |
1669 | .release = single_release, | |
1670 | }; | |
1671 | #endif | |
1672 | ||
1673 | static int __init pktsched_init(void) | |
1674 | { | |
1675 | register_qdisc(&pfifo_qdisc_ops); | |
1676 | register_qdisc(&bfifo_qdisc_ops); | |
1677 | proc_net_fops_create(&init_net, "psched", 0, &psched_fops); | |
1678 | ||
1679 | rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL); | |
1680 | rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL); | |
1681 | rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc); | |
1682 | rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL); | |
1683 | rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL); | |
1684 | rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass); | |
1685 | ||
1686 | return 0; | |
1687 | } | |
1688 | ||
1689 | subsys_initcall(pktsched_init); |