?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
perf tools: Document --children option in more detail
[linux-2.6/btrfs-unstable.git] / net / sched / sch_api.c
blob243b7d169d6183f662ab7f30d0e93492b29e79e3
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 #include <linux/slab.h>
32
33 #include <net/net_namespace.h>
34 #include <net/sock.h>
35 #include <net/netlink.h>
36 #include <net/pkt_sched.h>
37
38 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
39 struct nlmsghdr *n, u32 clid,
40 struct Qdisc *old, struct Qdisc *new);
41 static int tclass_notify(struct net *net, struct sk_buff *oskb,
42 struct nlmsghdr *n, struct Qdisc *q,
43 unsigned long cl, int event);
44
45 /*
46
47 Short review.
48 -------------
49
50 This file consists of two interrelated parts:
51
52 1. queueing disciplines manager frontend.
53 2. traffic classes manager frontend.
54
55 Generally, queueing discipline ("qdisc") is a black box,
56 which is able to enqueue packets and to dequeue them (when
57 device is ready to send something) in order and at times
58 determined by algorithm hidden in it.
59
60 qdisc's are divided to two categories:
61 - "queues", which have no internal structure visible from outside.
62 - "schedulers", which split all the packets to "traffic classes",
63 using "packet classifiers" (look at cls_api.c)
64
65 In turn, classes may have child qdiscs (as rule, queues)
66 attached to them etc. etc. etc.
67
68 The goal of the routines in this file is to translate
69 information supplied by user in the form of handles
70 to more intelligible for kernel form, to make some sanity
71 checks and part of work, which is common to all qdiscs
72 and to provide rtnetlink notifications.
73
74 All real intelligent work is done inside qdisc modules.
75
76
77
78 Every discipline has two major routines: enqueue and dequeue.
79
80 ---dequeue
81
82 dequeue usually returns a skb to send. It is allowed to return NULL,
83 but it does not mean that queue is empty, it just means that
84 discipline does not want to send anything this time.
85 Queue is really empty if q->q.qlen == 0.
86 For complicated disciplines with multiple queues q->q is not
87 real packet queue, but however q->q.qlen must be valid.
88
89 ---enqueue
90
91 enqueue returns 0, if packet was enqueued successfully.
92 If packet (this one or another one) was dropped, it returns
93 not zero error code.
94 NET_XMIT_DROP - this packet dropped
95 Expected action: do not backoff, but wait until queue will clear.
96 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
97 Expected action: backoff or ignore
98 NET_XMIT_POLICED - dropped by police.
99 Expected action: backoff or error to real-time apps.
100
101 Auxiliary routines:
102
103 ---peek
104
105 like dequeue but without removing a packet from the queue
106
107 ---reset
108
109 returns qdisc to initial state: purge all buffers, clear all
110 timers, counters (except for statistics) etc.
111
112 ---init
113
114 initializes newly created qdisc.
115
116 ---destroy
117
118 destroys resources allocated by init and during lifetime of qdisc.
119
120 ---change
121
122 changes qdisc parameters.
123 */
124
125 /* Protects list of registered TC modules. It is pure SMP lock. */
126 static DEFINE_RWLOCK(qdisc_mod_lock);
127
128
129 /************************************************
130 * Queueing disciplines manipulation. *
131 ************************************************/
132
133
134 /* The list of all installed queueing disciplines. */
135
136 static struct Qdisc_ops *qdisc_base;
137
138 /* Register/unregister queueing discipline */
139
140 int register_qdisc(struct Qdisc_ops *qops)
141 {
142 struct Qdisc_ops *q, **qp;
143 int rc = -EEXIST;
144
145 write_lock(&qdisc_mod_lock);
146 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
147 if (!strcmp(qops->id, q->id))
148 goto out;
149
150 if (qops->enqueue == NULL)
151 qops->enqueue = noop_qdisc_ops.enqueue;
152 if (qops->peek == NULL) {
153 if (qops->dequeue == NULL)
154 qops->peek = noop_qdisc_ops.peek;
155 else
156 goto out_einval;
157 }
158 if (qops->dequeue == NULL)
159 qops->dequeue = noop_qdisc_ops.dequeue;
160
161 if (qops->cl_ops) {
162 const struct Qdisc_class_ops *cops = qops->cl_ops;
163
164 if (!(cops->get && cops->put && cops->walk && cops->leaf))
165 goto out_einval;
166
167 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
168 goto out_einval;
169 }
170
171 qops->next = NULL;
172 *qp = qops;
173 rc = 0;
174 out:
175 write_unlock(&qdisc_mod_lock);
176 return rc;
177
178 out_einval:
179 rc = -EINVAL;
180 goto out;
181 }
182 EXPORT_SYMBOL(register_qdisc);
183
184 int unregister_qdisc(struct Qdisc_ops *qops)
185 {
186 struct Qdisc_ops *q, **qp;
187 int err = -ENOENT;
188
189 write_lock(&qdisc_mod_lock);
190 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
191 if (q == qops)
192 break;
193 if (q) {
194 *qp = q->next;
195 q->next = NULL;
196 err = 0;
197 }
198 write_unlock(&qdisc_mod_lock);
199 return err;
200 }
201 EXPORT_SYMBOL(unregister_qdisc);
202
203 /* Get default qdisc if not otherwise specified */
204 void qdisc_get_default(char *name, size_t len)
205 {
206 read_lock(&qdisc_mod_lock);
207 strlcpy(name, default_qdisc_ops->id, len);
208 read_unlock(&qdisc_mod_lock);
209 }
210
211 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
212 {
213 struct Qdisc_ops *q = NULL;
214
215 for (q = qdisc_base; q; q = q->next) {
216 if (!strcmp(name, q->id)) {
217 if (!try_module_get(q->owner))
218 q = NULL;
219 break;
220 }
221 }
222
223 return q;
224 }
225
226 /* Set new default qdisc to use */
227 int qdisc_set_default(const char *name)
228 {
229 const struct Qdisc_ops *ops;
230
231 if (!capable(CAP_NET_ADMIN))
232 return -EPERM;
233
234 write_lock(&qdisc_mod_lock);
235 ops = qdisc_lookup_default(name);
236 if (!ops) {
237 /* Not found, drop lock and try to load module */
238 write_unlock(&qdisc_mod_lock);
239 request_module("sch_%s", name);
240 write_lock(&qdisc_mod_lock);
241
242 ops = qdisc_lookup_default(name);
243 }
244
245 if (ops) {
246 /* Set new default */
247 module_put(default_qdisc_ops->owner);
248 default_qdisc_ops = ops;
249 }
250 write_unlock(&qdisc_mod_lock);
251
252 return ops ? 0 : -ENOENT;
253 }
254
255 /* We know handle. Find qdisc among all qdisc's attached to device
256 (root qdisc, all its children, children of children etc.)
257 */
258
259 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
260 {
261 struct Qdisc *q;
262
263 if (!(root->flags & TCQ_F_BUILTIN) &&
264 root->handle == handle)
265 return root;
266
267 list_for_each_entry(q, &root->list, list) {
268 if (q->handle == handle)
269 return q;
270 }
271 return NULL;
272 }
273
274 void qdisc_list_add(struct Qdisc *q)
275 {
276 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
277 struct Qdisc *root = qdisc_dev(q)->qdisc;
278
279 WARN_ON_ONCE(root == &noop_qdisc);
280 list_add_tail(&q->list, &root->list);
281 }
282 }
283 EXPORT_SYMBOL(qdisc_list_add);
284
285 void qdisc_list_del(struct Qdisc *q)
286 {
287 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
288 list_del(&q->list);
289 }
290 EXPORT_SYMBOL(qdisc_list_del);
291
292 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
293 {
294 struct Qdisc *q;
295
296 q = qdisc_match_from_root(dev->qdisc, handle);
297 if (q)
298 goto out;
299
300 if (dev_ingress_queue(dev))
301 q = qdisc_match_from_root(
302 dev_ingress_queue(dev)->qdisc_sleeping,
303 handle);
304 out:
305 return q;
306 }
307
308 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
309 {
310 unsigned long cl;
311 struct Qdisc *leaf;
312 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
313
314 if (cops == NULL)
315 return NULL;
316 cl = cops->get(p, classid);
317
318 if (cl == 0)
319 return NULL;
320 leaf = cops->leaf(p, cl);
321 cops->put(p, cl);
322 return leaf;
323 }
324
325 /* Find queueing discipline by name */
326
327 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
328 {
329 struct Qdisc_ops *q = NULL;
330
331 if (kind) {
332 read_lock(&qdisc_mod_lock);
333 for (q = qdisc_base; q; q = q->next) {
334 if (nla_strcmp(kind, q->id) == 0) {
335 if (!try_module_get(q->owner))
336 q = NULL;
337 break;
338 }
339 }
340 read_unlock(&qdisc_mod_lock);
341 }
342 return q;
343 }
344
345 /* The linklayer setting were not transferred from iproute2, in older
346 * versions, and the rate tables lookup systems have been dropped in
347 * the kernel. To keep backward compatible with older iproute2 tc
348 * utils, we detect the linklayer setting by detecting if the rate
349 * table were modified.
350 *
351 * For linklayer ATM table entries, the rate table will be aligned to
352 * 48 bytes, thus some table entries will contain the same value. The
353 * mpu (min packet unit) is also encoded into the old rate table, thus
354 * starting from the mpu, we find low and high table entries for
355 * mapping this cell. If these entries contain the same value, when
356 * the rate tables have been modified for linklayer ATM.
357 *
358 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
359 * and then roundup to the next cell, calc the table entry one below,
360 * and compare.
361 */
362 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
363 {
364 int low = roundup(r->mpu, 48);
365 int high = roundup(low+1, 48);
366 int cell_low = low >> r->cell_log;
367 int cell_high = (high >> r->cell_log) - 1;
368
369 /* rtab is too inaccurate at rates > 100Mbit/s */
370 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
371 pr_debug("TC linklayer: Giving up ATM detection\n");
372 return TC_LINKLAYER_ETHERNET;
373 }
374
375 if ((cell_high > cell_low) && (cell_high < 256)
376 && (rtab[cell_low] == rtab[cell_high])) {
377 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
378 cell_low, cell_high, rtab[cell_high]);
379 return TC_LINKLAYER_ATM;
380 }
381 return TC_LINKLAYER_ETHERNET;
382 }
383
384 static struct qdisc_rate_table *qdisc_rtab_list;
385
386 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
387 {
388 struct qdisc_rate_table *rtab;
389
390 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
391 nla_len(tab) != TC_RTAB_SIZE)
392 return NULL;
393
394 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
395 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
396 !memcmp(&rtab->data, nla_data(tab), 1024)) {
397 rtab->refcnt++;
398 return rtab;
399 }
400 }
401
402 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
403 if (rtab) {
404 rtab->rate = *r;
405 rtab->refcnt = 1;
406 memcpy(rtab->data, nla_data(tab), 1024);
407 if (r->linklayer == TC_LINKLAYER_UNAWARE)
408 r->linklayer = __detect_linklayer(r, rtab->data);
409 rtab->next = qdisc_rtab_list;
410 qdisc_rtab_list = rtab;
411 }
412 return rtab;
413 }
414 EXPORT_SYMBOL(qdisc_get_rtab);
415
416 void qdisc_put_rtab(struct qdisc_rate_table *tab)
417 {
418 struct qdisc_rate_table *rtab, **rtabp;
419
420 if (!tab || --tab->refcnt)
421 return;
422
423 for (rtabp = &qdisc_rtab_list;
424 (rtab = *rtabp) != NULL;
425 rtabp = &rtab->next) {
426 if (rtab == tab) {
427 *rtabp = rtab->next;
428 kfree(rtab);
429 return;
430 }
431 }
432 }
433 EXPORT_SYMBOL(qdisc_put_rtab);
434
435 static LIST_HEAD(qdisc_stab_list);
436 static DEFINE_SPINLOCK(qdisc_stab_lock);
437
438 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
439 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
440 [TCA_STAB_DATA] = { .type = NLA_BINARY },
441 };
442
443 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
444 {
445 struct nlattr *tb[TCA_STAB_MAX + 1];
446 struct qdisc_size_table *stab;
447 struct tc_sizespec *s;
448 unsigned int tsize = 0;
449 u16 *tab = NULL;
450 int err;
451
452 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
453 if (err < 0)
454 return ERR_PTR(err);
455 if (!tb[TCA_STAB_BASE])
456 return ERR_PTR(-EINVAL);
457
458 s = nla_data(tb[TCA_STAB_BASE]);
459
460 if (s->tsize > 0) {
461 if (!tb[TCA_STAB_DATA])
462 return ERR_PTR(-EINVAL);
463 tab = nla_data(tb[TCA_STAB_DATA]);
464 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
465 }
466
467 if (tsize != s->tsize || (!tab && tsize > 0))
468 return ERR_PTR(-EINVAL);
469
470 spin_lock(&qdisc_stab_lock);
471
472 list_for_each_entry(stab, &qdisc_stab_list, list) {
473 if (memcmp(&stab->szopts, s, sizeof(*s)))
474 continue;
475 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
476 continue;
477 stab->refcnt++;
478 spin_unlock(&qdisc_stab_lock);
479 return stab;
480 }
481
482 spin_unlock(&qdisc_stab_lock);
483
484 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
485 if (!stab)
486 return ERR_PTR(-ENOMEM);
487
488 stab->refcnt = 1;
489 stab->szopts = *s;
490 if (tsize > 0)
491 memcpy(stab->data, tab, tsize * sizeof(u16));
492
493 spin_lock(&qdisc_stab_lock);
494 list_add_tail(&stab->list, &qdisc_stab_list);
495 spin_unlock(&qdisc_stab_lock);
496
497 return stab;
498 }
499
500 static void stab_kfree_rcu(struct rcu_head *head)
501 {
502 kfree(container_of(head, struct qdisc_size_table, rcu));
503 }
504
505 void qdisc_put_stab(struct qdisc_size_table *tab)
506 {
507 if (!tab)
508 return;
509
510 spin_lock(&qdisc_stab_lock);
511
512 if (--tab->refcnt == 0) {
513 list_del(&tab->list);
514 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
515 }
516
517 spin_unlock(&qdisc_stab_lock);
518 }
519 EXPORT_SYMBOL(qdisc_put_stab);
520
521 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
522 {
523 struct nlattr *nest;
524
525 nest = nla_nest_start(skb, TCA_STAB);
526 if (nest == NULL)
527 goto nla_put_failure;
528 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
529 goto nla_put_failure;
530 nla_nest_end(skb, nest);
531
532 return skb->len;
533
534 nla_put_failure:
535 return -1;
536 }
537
538 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
539 {
540 int pkt_len, slot;
541
542 pkt_len = skb->len + stab->szopts.overhead;
543 if (unlikely(!stab->szopts.tsize))
544 goto out;
545
546 slot = pkt_len + stab->szopts.cell_align;
547 if (unlikely(slot < 0))
548 slot = 0;
549
550 slot >>= stab->szopts.cell_log;
551 if (likely(slot < stab->szopts.tsize))
552 pkt_len = stab->data[slot];
553 else
554 pkt_len = stab->data[stab->szopts.tsize - 1] *
555 (slot / stab->szopts.tsize) +
556 stab->data[slot % stab->szopts.tsize];
557
558 pkt_len <<= stab->szopts.size_log;
559 out:
560 if (unlikely(pkt_len < 1))
561 pkt_len = 1;
562 qdisc_skb_cb(skb)->pkt_len = pkt_len;
563 }
564 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
565
566 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
567 {
568 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
569 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
570 txt, qdisc->ops->id, qdisc->handle >> 16);
571 qdisc->flags |= TCQ_F_WARN_NONWC;
572 }
573 }
574 EXPORT_SYMBOL(qdisc_warn_nonwc);
575
576 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
577 {
578 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
579 timer);
580
581 rcu_read_lock();
582 qdisc_unthrottled(wd->qdisc);
583 __netif_schedule(qdisc_root(wd->qdisc));
584 rcu_read_unlock();
585
586 return HRTIMER_NORESTART;
587 }
588
589 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
590 {
591 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
592 wd->timer.function = qdisc_watchdog;
593 wd->qdisc = qdisc;
594 }
595 EXPORT_SYMBOL(qdisc_watchdog_init);
596
597 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires, bool throttle)
598 {
599 if (test_bit(__QDISC_STATE_DEACTIVATED,
600 &qdisc_root_sleeping(wd->qdisc)->state))
601 return;
602
603 if (throttle)
604 qdisc_throttled(wd->qdisc);
605
606 hrtimer_start(&wd->timer,
607 ns_to_ktime(expires),
608 HRTIMER_MODE_ABS_PINNED);
609 }
610 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
611
612 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
613 {
614 hrtimer_cancel(&wd->timer);
615 qdisc_unthrottled(wd->qdisc);
616 }
617 EXPORT_SYMBOL(qdisc_watchdog_cancel);
618
619 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
620 {
621 unsigned int size = n * sizeof(struct hlist_head), i;
622 struct hlist_head *h;
623
624 if (size <= PAGE_SIZE)
625 h = kmalloc(size, GFP_KERNEL);
626 else
627 h = (struct hlist_head *)
628 __get_free_pages(GFP_KERNEL, get_order(size));
629
630 if (h != NULL) {
631 for (i = 0; i < n; i++)
632 INIT_HLIST_HEAD(&h[i]);
633 }
634 return h;
635 }
636
637 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
638 {
639 unsigned int size = n * sizeof(struct hlist_head);
640
641 if (size <= PAGE_SIZE)
642 kfree(h);
643 else
644 free_pages((unsigned long)h, get_order(size));
645 }
646
647 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
648 {
649 struct Qdisc_class_common *cl;
650 struct hlist_node *next;
651 struct hlist_head *nhash, *ohash;
652 unsigned int nsize, nmask, osize;
653 unsigned int i, h;
654
655 /* Rehash when load factor exceeds 0.75 */
656 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
657 return;
658 nsize = clhash->hashsize * 2;
659 nmask = nsize - 1;
660 nhash = qdisc_class_hash_alloc(nsize);
661 if (nhash == NULL)
662 return;
663
664 ohash = clhash->hash;
665 osize = clhash->hashsize;
666
667 sch_tree_lock(sch);
668 for (i = 0; i < osize; i++) {
669 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
670 h = qdisc_class_hash(cl->classid, nmask);
671 hlist_add_head(&cl->hnode, &nhash[h]);
672 }
673 }
674 clhash->hash = nhash;
675 clhash->hashsize = nsize;
676 clhash->hashmask = nmask;
677 sch_tree_unlock(sch);
678
679 qdisc_class_hash_free(ohash, osize);
680 }
681 EXPORT_SYMBOL(qdisc_class_hash_grow);
682
683 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
684 {
685 unsigned int size = 4;
686
687 clhash->hash = qdisc_class_hash_alloc(size);
688 if (clhash->hash == NULL)
689 return -ENOMEM;
690 clhash->hashsize = size;
691 clhash->hashmask = size - 1;
692 clhash->hashelems = 0;
693 return 0;
694 }
695 EXPORT_SYMBOL(qdisc_class_hash_init);
696
697 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
698 {
699 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
700 }
701 EXPORT_SYMBOL(qdisc_class_hash_destroy);
702
703 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
704 struct Qdisc_class_common *cl)
705 {
706 unsigned int h;
707
708 INIT_HLIST_NODE(&cl->hnode);
709 h = qdisc_class_hash(cl->classid, clhash->hashmask);
710 hlist_add_head(&cl->hnode, &clhash->hash[h]);
711 clhash->hashelems++;
712 }
713 EXPORT_SYMBOL(qdisc_class_hash_insert);
714
715 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
716 struct Qdisc_class_common *cl)
717 {
718 hlist_del(&cl->hnode);
719 clhash->hashelems--;
720 }
721 EXPORT_SYMBOL(qdisc_class_hash_remove);
722
723 /* Allocate an unique handle from space managed by kernel
724 * Possible range is [8000-FFFF]:0000 (0x8000 values)
725 */
726 static u32 qdisc_alloc_handle(struct net_device *dev)
727 {
728 int i = 0x8000;
729 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
730
731 do {
732 autohandle += TC_H_MAKE(0x10000U, 0);
733 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
734 autohandle = TC_H_MAKE(0x80000000U, 0);
735 if (!qdisc_lookup(dev, autohandle))
736 return autohandle;
737 cond_resched();
738 } while (--i > 0);
739
740 return 0;
741 }
742
743 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
744 {
745 const struct Qdisc_class_ops *cops;
746 unsigned long cl;
747 u32 parentid;
748 int drops;
749
750 if (n == 0)
751 return;
752 drops = max_t(int, n, 0);
753 while ((parentid = sch->parent)) {
754 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
755 return;
756
757 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
758 if (sch == NULL) {
759 WARN_ON(parentid != TC_H_ROOT);
760 return;
761 }
762 cops = sch->ops->cl_ops;
763 if (cops->qlen_notify) {
764 cl = cops->get(sch, parentid);
765 cops->qlen_notify(sch, cl);
766 cops->put(sch, cl);
767 }
768 sch->q.qlen -= n;
769 __qdisc_qstats_drop(sch, drops);
770 }
771 }
772 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
773
774 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
775 struct nlmsghdr *n, u32 clid,
776 struct Qdisc *old, struct Qdisc *new)
777 {
778 if (new || old)
779 qdisc_notify(net, skb, n, clid, old, new);
780
781 if (old)
782 qdisc_destroy(old);
783 }
784
785 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
786 * to device "dev".
787 *
788 * When appropriate send a netlink notification using 'skb'
789 * and "n".
790 *
791 * On success, destroy old qdisc.
792 */
793
794 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
795 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
796 struct Qdisc *new, struct Qdisc *old)
797 {
798 struct Qdisc *q = old;
799 struct net *net = dev_net(dev);
800 int err = 0;
801
802 if (parent == NULL) {
803 unsigned int i, num_q, ingress;
804
805 ingress = 0;
806 num_q = dev->num_tx_queues;
807 if ((q && q->flags & TCQ_F_INGRESS) ||
808 (new && new->flags & TCQ_F_INGRESS)) {
809 num_q = 1;
810 ingress = 1;
811 if (!dev_ingress_queue(dev))
812 return -ENOENT;
813 }
814
815 if (dev->flags & IFF_UP)
816 dev_deactivate(dev);
817
818 if (new && new->ops->attach) {
819 new->ops->attach(new);
820 num_q = 0;
821 }
822
823 for (i = 0; i < num_q; i++) {
824 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
825
826 if (!ingress)
827 dev_queue = netdev_get_tx_queue(dev, i);
828
829 old = dev_graft_qdisc(dev_queue, new);
830 if (new && i > 0)
831 atomic_inc(&new->refcnt);
832
833 if (!ingress)
834 qdisc_destroy(old);
835 }
836
837 if (!ingress) {
838 notify_and_destroy(net, skb, n, classid,
839 dev->qdisc, new);
840 if (new && !new->ops->attach)
841 atomic_inc(&new->refcnt);
842 dev->qdisc = new ? : &noop_qdisc;
843 } else {
844 notify_and_destroy(net, skb, n, classid, old, new);
845 }
846
847 if (dev->flags & IFF_UP)
848 dev_activate(dev);
849 } else {
850 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
851
852 err = -EOPNOTSUPP;
853 if (cops && cops->graft) {
854 unsigned long cl = cops->get(parent, classid);
855 if (cl) {
856 err = cops->graft(parent, cl, new, &old);
857 cops->put(parent, cl);
858 } else
859 err = -ENOENT;
860 }
861 if (!err)
862 notify_and_destroy(net, skb, n, classid, old, new);
863 }
864 return err;
865 }
866
867 /* lockdep annotation is needed for ingress; egress gets it only for name */
868 static struct lock_class_key qdisc_tx_lock;
869 static struct lock_class_key qdisc_rx_lock;
870
871 /*
872 Allocate and initialize new qdisc.
873
874 Parameters are passed via opt.
875 */
876
877 static struct Qdisc *
878 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
879 struct Qdisc *p, u32 parent, u32 handle,
880 struct nlattr **tca, int *errp)
881 {
882 int err;
883 struct nlattr *kind = tca[TCA_KIND];
884 struct Qdisc *sch;
885 struct Qdisc_ops *ops;
886 struct qdisc_size_table *stab;
887
888 ops = qdisc_lookup_ops(kind);
889 #ifdef CONFIG_MODULES
890 if (ops == NULL && kind != NULL) {
891 char name[IFNAMSIZ];
892 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
893 /* We dropped the RTNL semaphore in order to
894 * perform the module load. So, even if we
895 * succeeded in loading the module we have to
896 * tell the caller to replay the request. We
897 * indicate this using -EAGAIN.
898 * We replay the request because the device may
899 * go away in the mean time.
900 */
901 rtnl_unlock();
902 request_module("sch_%s", name);
903 rtnl_lock();
904 ops = qdisc_lookup_ops(kind);
905 if (ops != NULL) {
906 /* We will try again qdisc_lookup_ops,
907 * so don't keep a reference.
908 */
909 module_put(ops->owner);
910 err = -EAGAIN;
911 goto err_out;
912 }
913 }
914 }
915 #endif
916
917 err = -ENOENT;
918 if (ops == NULL)
919 goto err_out;
920
921 sch = qdisc_alloc(dev_queue, ops);
922 if (IS_ERR(sch)) {
923 err = PTR_ERR(sch);
924 goto err_out2;
925 }
926
927 sch->parent = parent;
928
929 if (handle == TC_H_INGRESS) {
930 sch->flags |= TCQ_F_INGRESS;
931 handle = TC_H_MAKE(TC_H_INGRESS, 0);
932 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
933 } else {
934 if (handle == 0) {
935 handle = qdisc_alloc_handle(dev);
936 err = -ENOMEM;
937 if (handle == 0)
938 goto err_out3;
939 }
940 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
941 if (!netif_is_multiqueue(dev))
942 sch->flags |= TCQ_F_ONETXQUEUE;
943 }
944
945 sch->handle = handle;
946
947 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
948 if (qdisc_is_percpu_stats(sch)) {
949 sch->cpu_bstats =
950 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
951 if (!sch->cpu_bstats)
952 goto err_out4;
953
954 sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
955 if (!sch->cpu_qstats)
956 goto err_out4;
957 }
958
959 if (tca[TCA_STAB]) {
960 stab = qdisc_get_stab(tca[TCA_STAB]);
961 if (IS_ERR(stab)) {
962 err = PTR_ERR(stab);
963 goto err_out4;
964 }
965 rcu_assign_pointer(sch->stab, stab);
966 }
967 if (tca[TCA_RATE]) {
968 spinlock_t *root_lock;
969
970 err = -EOPNOTSUPP;
971 if (sch->flags & TCQ_F_MQROOT)
972 goto err_out4;
973
974 if ((sch->parent != TC_H_ROOT) &&
975 !(sch->flags & TCQ_F_INGRESS) &&
976 (!p || !(p->flags & TCQ_F_MQROOT)))
977 root_lock = qdisc_root_sleeping_lock(sch);
978 else
979 root_lock = qdisc_lock(sch);
980
981 err = gen_new_estimator(&sch->bstats,
982 sch->cpu_bstats,
983 &sch->rate_est,
984 root_lock,
985 tca[TCA_RATE]);
986 if (err)
987 goto err_out4;
988 }
989
990 qdisc_list_add(sch);
991
992 return sch;
993 }
994 err_out3:
995 dev_put(dev);
996 kfree((char *) sch - sch->padded);
997 err_out2:
998 module_put(ops->owner);
999 err_out:
1000 *errp = err;
1001 return NULL;
1002
1003 err_out4:
1004 free_percpu(sch->cpu_bstats);
1005 free_percpu(sch->cpu_qstats);
1006 /*
1007 * Any broken qdiscs that would require a ops->reset() here?
1008 * The qdisc was never in action so it shouldn't be necessary.
1009 */
1010 qdisc_put_stab(rtnl_dereference(sch->stab));
1011 if (ops->destroy)
1012 ops->destroy(sch);
1013 goto err_out3;
1014 }
1015
1016 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
1017 {
1018 struct qdisc_size_table *ostab, *stab = NULL;
1019 int err = 0;
1020
1021 if (tca[TCA_OPTIONS]) {
1022 if (sch->ops->change == NULL)
1023 return -EINVAL;
1024 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1025 if (err)
1026 return err;
1027 }
1028
1029 if (tca[TCA_STAB]) {
1030 stab = qdisc_get_stab(tca[TCA_STAB]);
1031 if (IS_ERR(stab))
1032 return PTR_ERR(stab);
1033 }
1034
1035 ostab = rtnl_dereference(sch->stab);
1036 rcu_assign_pointer(sch->stab, stab);
1037 qdisc_put_stab(ostab);
1038
1039 if (tca[TCA_RATE]) {
1040 /* NB: ignores errors from replace_estimator
1041 because change can't be undone. */
1042 if (sch->flags & TCQ_F_MQROOT)
1043 goto out;
1044 gen_replace_estimator(&sch->bstats,
1045 sch->cpu_bstats,
1046 &sch->rate_est,
1047 qdisc_root_sleeping_lock(sch),
1048 tca[TCA_RATE]);
1049 }
1050 out:
1051 return 0;
1052 }
1053
1054 struct check_loop_arg {
1055 struct qdisc_walker w;
1056 struct Qdisc *p;
1057 int depth;
1058 };
1059
1060 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
1061
1062 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1063 {
1064 struct check_loop_arg arg;
1065
1066 if (q->ops->cl_ops == NULL)
1067 return 0;
1068
1069 arg.w.stop = arg.w.skip = arg.w.count = 0;
1070 arg.w.fn = check_loop_fn;
1071 arg.depth = depth;
1072 arg.p = p;
1073 q->ops->cl_ops->walk(q, &arg.w);
1074 return arg.w.stop ? -ELOOP : 0;
1075 }
1076
1077 static int
1078 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1079 {
1080 struct Qdisc *leaf;
1081 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1082 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1083
1084 leaf = cops->leaf(q, cl);
1085 if (leaf) {
1086 if (leaf == arg->p || arg->depth > 7)
1087 return -ELOOP;
1088 return check_loop(leaf, arg->p, arg->depth + 1);
1089 }
1090 return 0;
1091 }
1092
1093 /*
1094 * Delete/get qdisc.
1095 */
1096
1097 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1098 {
1099 struct net *net = sock_net(skb->sk);
1100 struct tcmsg *tcm = nlmsg_data(n);
1101 struct nlattr *tca[TCA_MAX + 1];
1102 struct net_device *dev;
1103 u32 clid;
1104 struct Qdisc *q = NULL;
1105 struct Qdisc *p = NULL;
1106 int err;
1107
1108 if ((n->nlmsg_type != RTM_GETQDISC) &&
1109 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1110 return -EPERM;
1111
1112 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1113 if (err < 0)
1114 return err;
1115
1116 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1117 if (!dev)
1118 return -ENODEV;
1119
1120 clid = tcm->tcm_parent;
1121 if (clid) {
1122 if (clid != TC_H_ROOT) {
1123 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1124 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1125 if (!p)
1126 return -ENOENT;
1127 q = qdisc_leaf(p, clid);
1128 } else if (dev_ingress_queue(dev)) {
1129 q = dev_ingress_queue(dev)->qdisc_sleeping;
1130 }
1131 } else {
1132 q = dev->qdisc;
1133 }
1134 if (!q)
1135 return -ENOENT;
1136
1137 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1138 return -EINVAL;
1139 } else {
1140 q = qdisc_lookup(dev, tcm->tcm_handle);
1141 if (!q)
1142 return -ENOENT;
1143 }
1144
1145 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1146 return -EINVAL;
1147
1148 if (n->nlmsg_type == RTM_DELQDISC) {
1149 if (!clid)
1150 return -EINVAL;
1151 if (q->handle == 0)
1152 return -ENOENT;
1153 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1154 if (err != 0)
1155 return err;
1156 } else {
1157 qdisc_notify(net, skb, n, clid, NULL, q);
1158 }
1159 return 0;
1160 }
1161
1162 /*
1163 * Create/change qdisc.
1164 */
1165
1166 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1167 {
1168 struct net *net = sock_net(skb->sk);
1169 struct tcmsg *tcm;
1170 struct nlattr *tca[TCA_MAX + 1];
1171 struct net_device *dev;
1172 u32 clid;
1173 struct Qdisc *q, *p;
1174 int err;
1175
1176 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1177 return -EPERM;
1178
1179 replay:
1180 /* Reinit, just in case something touches this. */
1181 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1182 if (err < 0)
1183 return err;
1184
1185 tcm = nlmsg_data(n);
1186 clid = tcm->tcm_parent;
1187 q = p = NULL;
1188
1189 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1190 if (!dev)
1191 return -ENODEV;
1192
1193
1194 if (clid) {
1195 if (clid != TC_H_ROOT) {
1196 if (clid != TC_H_INGRESS) {
1197 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1198 if (!p)
1199 return -ENOENT;
1200 q = qdisc_leaf(p, clid);
1201 } else if (dev_ingress_queue_create(dev)) {
1202 q = dev_ingress_queue(dev)->qdisc_sleeping;
1203 }
1204 } else {
1205 q = dev->qdisc;
1206 }
1207
1208 /* It may be default qdisc, ignore it */
1209 if (q && q->handle == 0)
1210 q = NULL;
1211
1212 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1213 if (tcm->tcm_handle) {
1214 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1215 return -EEXIST;
1216 if (TC_H_MIN(tcm->tcm_handle))
1217 return -EINVAL;
1218 q = qdisc_lookup(dev, tcm->tcm_handle);
1219 if (!q)
1220 goto create_n_graft;
1221 if (n->nlmsg_flags & NLM_F_EXCL)
1222 return -EEXIST;
1223 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1224 return -EINVAL;
1225 if (q == p ||
1226 (p && check_loop(q, p, 0)))
1227 return -ELOOP;
1228 atomic_inc(&q->refcnt);
1229 goto graft;
1230 } else {
1231 if (!q)
1232 goto create_n_graft;
1233
1234 /* This magic test requires explanation.
1235 *
1236 * We know, that some child q is already
1237 * attached to this parent and have choice:
1238 * either to change it or to create/graft new one.
1239 *
1240 * 1. We are allowed to create/graft only
1241 * if CREATE and REPLACE flags are set.
1242 *
1243 * 2. If EXCL is set, requestor wanted to say,
1244 * that qdisc tcm_handle is not expected
1245 * to exist, so that we choose create/graft too.
1246 *
1247 * 3. The last case is when no flags are set.
1248 * Alas, it is sort of hole in API, we
1249 * cannot decide what to do unambiguously.
1250 * For now we select create/graft, if
1251 * user gave KIND, which does not match existing.
1252 */
1253 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1254 (n->nlmsg_flags & NLM_F_REPLACE) &&
1255 ((n->nlmsg_flags & NLM_F_EXCL) ||
1256 (tca[TCA_KIND] &&
1257 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1258 goto create_n_graft;
1259 }
1260 }
1261 } else {
1262 if (!tcm->tcm_handle)
1263 return -EINVAL;
1264 q = qdisc_lookup(dev, tcm->tcm_handle);
1265 }
1266
1267 /* Change qdisc parameters */
1268 if (q == NULL)
1269 return -ENOENT;
1270 if (n->nlmsg_flags & NLM_F_EXCL)
1271 return -EEXIST;
1272 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1273 return -EINVAL;
1274 err = qdisc_change(q, tca);
1275 if (err == 0)
1276 qdisc_notify(net, skb, n, clid, NULL, q);
1277 return err;
1278
1279 create_n_graft:
1280 if (!(n->nlmsg_flags & NLM_F_CREATE))
1281 return -ENOENT;
1282 if (clid == TC_H_INGRESS) {
1283 if (dev_ingress_queue(dev))
1284 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1285 tcm->tcm_parent, tcm->tcm_parent,
1286 tca, &err);
1287 else
1288 err = -ENOENT;
1289 } else {
1290 struct netdev_queue *dev_queue;
1291
1292 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1293 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1294 else if (p)
1295 dev_queue = p->dev_queue;
1296 else
1297 dev_queue = netdev_get_tx_queue(dev, 0);
1298
1299 q = qdisc_create(dev, dev_queue, p,
1300 tcm->tcm_parent, tcm->tcm_handle,
1301 tca, &err);
1302 }
1303 if (q == NULL) {
1304 if (err == -EAGAIN)
1305 goto replay;
1306 return err;
1307 }
1308
1309 graft:
1310 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1311 if (err) {
1312 if (q)
1313 qdisc_destroy(q);
1314 return err;
1315 }
1316
1317 return 0;
1318 }
1319
1320 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1321 u32 portid, u32 seq, u16 flags, int event)
1322 {
1323 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
1324 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
1325 struct tcmsg *tcm;
1326 struct nlmsghdr *nlh;
1327 unsigned char *b = skb_tail_pointer(skb);
1328 struct gnet_dump d;
1329 struct qdisc_size_table *stab;
1330 __u32 qlen;
1331
1332 cond_resched();
1333 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1334 if (!nlh)
1335 goto out_nlmsg_trim;
1336 tcm = nlmsg_data(nlh);
1337 tcm->tcm_family = AF_UNSPEC;
1338 tcm->tcm__pad1 = 0;
1339 tcm->tcm__pad2 = 0;
1340 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1341 tcm->tcm_parent = clid;
1342 tcm->tcm_handle = q->handle;
1343 tcm->tcm_info = atomic_read(&q->refcnt);
1344 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1345 goto nla_put_failure;
1346 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1347 goto nla_put_failure;
1348 qlen = q->q.qlen;
1349
1350 stab = rtnl_dereference(q->stab);
1351 if (stab && qdisc_dump_stab(skb, stab) < 0)
1352 goto nla_put_failure;
1353
1354 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1355 qdisc_root_sleeping_lock(q), &d) < 0)
1356 goto nla_put_failure;
1357
1358 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1359 goto nla_put_failure;
1360
1361 if (qdisc_is_percpu_stats(q)) {
1362 cpu_bstats = q->cpu_bstats;
1363 cpu_qstats = q->cpu_qstats;
1364 }
1365
1366 if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats) < 0 ||
1367 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1368 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
1369 goto nla_put_failure;
1370
1371 if (gnet_stats_finish_copy(&d) < 0)
1372 goto nla_put_failure;
1373
1374 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1375 return skb->len;
1376
1377 out_nlmsg_trim:
1378 nla_put_failure:
1379 nlmsg_trim(skb, b);
1380 return -1;
1381 }
1382
1383 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1384 {
1385 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1386 }
1387
1388 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1389 struct nlmsghdr *n, u32 clid,
1390 struct Qdisc *old, struct Qdisc *new)
1391 {
1392 struct sk_buff *skb;
1393 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1394
1395 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1396 if (!skb)
1397 return -ENOBUFS;
1398
1399 if (old && !tc_qdisc_dump_ignore(old)) {
1400 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1401 0, RTM_DELQDISC) < 0)
1402 goto err_out;
1403 }
1404 if (new && !tc_qdisc_dump_ignore(new)) {
1405 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1406 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1407 goto err_out;
1408 }
1409
1410 if (skb->len)
1411 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1412 n->nlmsg_flags & NLM_F_ECHO);
1413
1414 err_out:
1415 kfree_skb(skb);
1416 return -EINVAL;
1417 }
1418
1419 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1420 struct netlink_callback *cb,
1421 int *q_idx_p, int s_q_idx)
1422 {
1423 int ret = 0, q_idx = *q_idx_p;
1424 struct Qdisc *q;
1425
1426 if (!root)
1427 return 0;
1428
1429 q = root;
1430 if (q_idx < s_q_idx) {
1431 q_idx++;
1432 } else {
1433 if (!tc_qdisc_dump_ignore(q) &&
1434 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1435 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1436 goto done;
1437 q_idx++;
1438 }
1439 list_for_each_entry(q, &root->list, list) {
1440 if (q_idx < s_q_idx) {
1441 q_idx++;
1442 continue;
1443 }
1444 if (!tc_qdisc_dump_ignore(q) &&
1445 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1446 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1447 goto done;
1448 q_idx++;
1449 }
1450
1451 out:
1452 *q_idx_p = q_idx;
1453 return ret;
1454 done:
1455 ret = -1;
1456 goto out;
1457 }
1458
1459 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1460 {
1461 struct net *net = sock_net(skb->sk);
1462 int idx, q_idx;
1463 int s_idx, s_q_idx;
1464 struct net_device *dev;
1465
1466 s_idx = cb->args[0];
1467 s_q_idx = q_idx = cb->args[1];
1468
1469 idx = 0;
1470 ASSERT_RTNL();
1471 for_each_netdev(net, dev) {
1472 struct netdev_queue *dev_queue;
1473
1474 if (idx < s_idx)
1475 goto cont;
1476 if (idx > s_idx)
1477 s_q_idx = 0;
1478 q_idx = 0;
1479
1480 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1481 goto done;
1482
1483 dev_queue = dev_ingress_queue(dev);
1484 if (dev_queue &&
1485 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1486 &q_idx, s_q_idx) < 0)
1487 goto done;
1488
1489 cont:
1490 idx++;
1491 }
1492
1493 done:
1494 cb->args[0] = idx;
1495 cb->args[1] = q_idx;
1496
1497 return skb->len;
1498 }
1499
1500
1501
1502 /************************************************
1503 * Traffic classes manipulation. *
1504 ************************************************/
1505
1506
1507
1508 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1509 {
1510 struct net *net = sock_net(skb->sk);
1511 struct tcmsg *tcm = nlmsg_data(n);
1512 struct nlattr *tca[TCA_MAX + 1];
1513 struct net_device *dev;
1514 struct Qdisc *q = NULL;
1515 const struct Qdisc_class_ops *cops;
1516 unsigned long cl = 0;
1517 unsigned long new_cl;
1518 u32 portid;
1519 u32 clid;
1520 u32 qid;
1521 int err;
1522
1523 if ((n->nlmsg_type != RTM_GETTCLASS) &&
1524 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1525 return -EPERM;
1526
1527 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1528 if (err < 0)
1529 return err;
1530
1531 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1532 if (!dev)
1533 return -ENODEV;
1534
1535 /*
1536 parent == TC_H_UNSPEC - unspecified parent.
1537 parent == TC_H_ROOT - class is root, which has no parent.
1538 parent == X:0 - parent is root class.
1539 parent == X:Y - parent is a node in hierarchy.
1540 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1541
1542 handle == 0:0 - generate handle from kernel pool.
1543 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1544 handle == X:Y - clear.
1545 handle == X:0 - root class.
1546 */
1547
1548 /* Step 1. Determine qdisc handle X:0 */
1549
1550 portid = tcm->tcm_parent;
1551 clid = tcm->tcm_handle;
1552 qid = TC_H_MAJ(clid);
1553
1554 if (portid != TC_H_ROOT) {
1555 u32 qid1 = TC_H_MAJ(portid);
1556
1557 if (qid && qid1) {
1558 /* If both majors are known, they must be identical. */
1559 if (qid != qid1)
1560 return -EINVAL;
1561 } else if (qid1) {
1562 qid = qid1;
1563 } else if (qid == 0)
1564 qid = dev->qdisc->handle;
1565
1566 /* Now qid is genuine qdisc handle consistent
1567 * both with parent and child.
1568 *
1569 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1570 */
1571 if (portid)
1572 portid = TC_H_MAKE(qid, portid);
1573 } else {
1574 if (qid == 0)
1575 qid = dev->qdisc->handle;
1576 }
1577
1578 /* OK. Locate qdisc */
1579 q = qdisc_lookup(dev, qid);
1580 if (!q)
1581 return -ENOENT;
1582
1583 /* An check that it supports classes */
1584 cops = q->ops->cl_ops;
1585 if (cops == NULL)
1586 return -EINVAL;
1587
1588 /* Now try to get class */
1589 if (clid == 0) {
1590 if (portid == TC_H_ROOT)
1591 clid = qid;
1592 } else
1593 clid = TC_H_MAKE(qid, clid);
1594
1595 if (clid)
1596 cl = cops->get(q, clid);
1597
1598 if (cl == 0) {
1599 err = -ENOENT;
1600 if (n->nlmsg_type != RTM_NEWTCLASS ||
1601 !(n->nlmsg_flags & NLM_F_CREATE))
1602 goto out;
1603 } else {
1604 switch (n->nlmsg_type) {
1605 case RTM_NEWTCLASS:
1606 err = -EEXIST;
1607 if (n->nlmsg_flags & NLM_F_EXCL)
1608 goto out;
1609 break;
1610 case RTM_DELTCLASS:
1611 err = -EOPNOTSUPP;
1612 if (cops->delete)
1613 err = cops->delete(q, cl);
1614 if (err == 0)
1615 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1616 goto out;
1617 case RTM_GETTCLASS:
1618 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1619 goto out;
1620 default:
1621 err = -EINVAL;
1622 goto out;
1623 }
1624 }
1625
1626 new_cl = cl;
1627 err = -EOPNOTSUPP;
1628 if (cops->change)
1629 err = cops->change(q, clid, portid, tca, &new_cl);
1630 if (err == 0)
1631 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1632
1633 out:
1634 if (cl)
1635 cops->put(q, cl);
1636
1637 return err;
1638 }
1639
1640
1641 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1642 unsigned long cl,
1643 u32 portid, u32 seq, u16 flags, int event)
1644 {
1645 struct tcmsg *tcm;
1646 struct nlmsghdr *nlh;
1647 unsigned char *b = skb_tail_pointer(skb);
1648 struct gnet_dump d;
1649 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1650
1651 cond_resched();
1652 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1653 if (!nlh)
1654 goto out_nlmsg_trim;
1655 tcm = nlmsg_data(nlh);
1656 tcm->tcm_family = AF_UNSPEC;
1657 tcm->tcm__pad1 = 0;
1658 tcm->tcm__pad2 = 0;
1659 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1660 tcm->tcm_parent = q->handle;
1661 tcm->tcm_handle = q->handle;
1662 tcm->tcm_info = 0;
1663 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1664 goto nla_put_failure;
1665 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1666 goto nla_put_failure;
1667
1668 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1669 qdisc_root_sleeping_lock(q), &d) < 0)
1670 goto nla_put_failure;
1671
1672 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1673 goto nla_put_failure;
1674
1675 if (gnet_stats_finish_copy(&d) < 0)
1676 goto nla_put_failure;
1677
1678 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1679 return skb->len;
1680
1681 out_nlmsg_trim:
1682 nla_put_failure:
1683 nlmsg_trim(skb, b);
1684 return -1;
1685 }
1686
1687 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1688 struct nlmsghdr *n, struct Qdisc *q,
1689 unsigned long cl, int event)
1690 {
1691 struct sk_buff *skb;
1692 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1693
1694 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1695 if (!skb)
1696 return -ENOBUFS;
1697
1698 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1699 kfree_skb(skb);
1700 return -EINVAL;
1701 }
1702
1703 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1704 n->nlmsg_flags & NLM_F_ECHO);
1705 }
1706
1707 struct qdisc_dump_args {
1708 struct qdisc_walker w;
1709 struct sk_buff *skb;
1710 struct netlink_callback *cb;
1711 };
1712
1713 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1714 {
1715 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1716
1717 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1718 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1719 }
1720
1721 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1722 struct tcmsg *tcm, struct netlink_callback *cb,
1723 int *t_p, int s_t)
1724 {
1725 struct qdisc_dump_args arg;
1726
1727 if (tc_qdisc_dump_ignore(q) ||
1728 *t_p < s_t || !q->ops->cl_ops ||
1729 (tcm->tcm_parent &&
1730 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1731 (*t_p)++;
1732 return 0;
1733 }
1734 if (*t_p > s_t)
1735 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1736 arg.w.fn = qdisc_class_dump;
1737 arg.skb = skb;
1738 arg.cb = cb;
1739 arg.w.stop = 0;
1740 arg.w.skip = cb->args[1];
1741 arg.w.count = 0;
1742 q->ops->cl_ops->walk(q, &arg.w);
1743 cb->args[1] = arg.w.count;
1744 if (arg.w.stop)
1745 return -1;
1746 (*t_p)++;
1747 return 0;
1748 }
1749
1750 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1751 struct tcmsg *tcm, struct netlink_callback *cb,
1752 int *t_p, int s_t)
1753 {
1754 struct Qdisc *q;
1755
1756 if (!root)
1757 return 0;
1758
1759 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1760 return -1;
1761
1762 list_for_each_entry(q, &root->list, list) {
1763 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1764 return -1;
1765 }
1766
1767 return 0;
1768 }
1769
1770 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1771 {
1772 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1773 struct net *net = sock_net(skb->sk);
1774 struct netdev_queue *dev_queue;
1775 struct net_device *dev;
1776 int t, s_t;
1777
1778 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1779 return 0;
1780 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1781 if (!dev)
1782 return 0;
1783
1784 s_t = cb->args[0];
1785 t = 0;
1786
1787 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1788 goto done;
1789
1790 dev_queue = dev_ingress_queue(dev);
1791 if (dev_queue &&
1792 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1793 &t, s_t) < 0)
1794 goto done;
1795
1796 done:
1797 cb->args[0] = t;
1798
1799 dev_put(dev);
1800 return skb->len;
1801 }
1802
1803 /* Main classifier routine: scans classifier chain attached
1804 * to this qdisc, (optionally) tests for protocol and asks
1805 * specific classifiers.
1806 */
1807 int tc_classify_compat(struct sk_buff *skb, const struct tcf_proto *tp,
1808 struct tcf_result *res)
1809 {
1810 __be16 protocol = tc_skb_protocol(skb);
1811 int err;
1812
1813 for (; tp; tp = rcu_dereference_bh(tp->next)) {
1814 if (tp->protocol != protocol &&
1815 tp->protocol != htons(ETH_P_ALL))
1816 continue;
1817 err = tp->classify(skb, tp, res);
1818
1819 if (err >= 0) {
1820 #ifdef CONFIG_NET_CLS_ACT
1821 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1822 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1823 #endif
1824 return err;
1825 }
1826 }
1827 return -1;
1828 }
1829 EXPORT_SYMBOL(tc_classify_compat);
1830
1831 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1832 struct tcf_result *res)
1833 {
1834 int err = 0;
1835 #ifdef CONFIG_NET_CLS_ACT
1836 const struct tcf_proto *otp = tp;
1837 reclassify:
1838 #endif
1839
1840 err = tc_classify_compat(skb, tp, res);
1841 #ifdef CONFIG_NET_CLS_ACT
1842 if (err == TC_ACT_RECLASSIFY) {
1843 u32 verd = G_TC_VERD(skb->tc_verd);
1844 tp = otp;
1845
1846 if (verd++ >= MAX_REC_LOOP) {
1847 net_notice_ratelimited("%s: packet reclassify loop rule prio %u protocol %02x\n",
1848 tp->q->ops->id,
1849 tp->prio & 0xffff,
1850 ntohs(tp->protocol));
1851 return TC_ACT_SHOT;
1852 }
1853 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1854 goto reclassify;
1855 }
1856 #endif
1857 return err;
1858 }
1859 EXPORT_SYMBOL(tc_classify);
1860
1861 void tcf_destroy(struct tcf_proto *tp)
1862 {
1863 tp->ops->destroy(tp);
1864 module_put(tp->ops->owner);
1865 kfree_rcu(tp, rcu);
1866 }
1867
1868 void tcf_destroy_chain(struct tcf_proto __rcu **fl)
1869 {
1870 struct tcf_proto *tp;
1871
1872 while ((tp = rtnl_dereference(*fl)) != NULL) {
1873 RCU_INIT_POINTER(*fl, tp->next);
1874 tcf_destroy(tp);
1875 }
1876 }
1877 EXPORT_SYMBOL(tcf_destroy_chain);
1878
1879 #ifdef CONFIG_PROC_FS
1880 static int psched_show(struct seq_file *seq, void *v)
1881 {
1882 struct timespec ts;
1883
1884 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1885 seq_printf(seq, "%08x %08x %08x %08x\n",
1886 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1887 1000000,
1888 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1889
1890 return 0;
1891 }
1892
1893 static int psched_open(struct inode *inode, struct file *file)
1894 {
1895 return single_open(file, psched_show, NULL);
1896 }
1897
1898 static const struct file_operations psched_fops = {
1899 .owner = THIS_MODULE,
1900 .open = psched_open,
1901 .read = seq_read,
1902 .llseek = seq_lseek,
1903 .release = single_release,
1904 };
1905
1906 static int __net_init psched_net_init(struct net *net)
1907 {
1908 struct proc_dir_entry *e;
1909
1910 e = proc_create("psched", 0, net->proc_net, &psched_fops);
1911 if (e == NULL)
1912 return -ENOMEM;
1913
1914 return 0;
1915 }
1916
1917 static void __net_exit psched_net_exit(struct net *net)
1918 {
1919 remove_proc_entry("psched", net->proc_net);
1920 }
1921 #else
1922 static int __net_init psched_net_init(struct net *net)
1923 {
1924 return 0;
1925 }
1926
1927 static void __net_exit psched_net_exit(struct net *net)
1928 {
1929 }
1930 #endif
1931
1932 static struct pernet_operations psched_net_ops = {
1933 .init = psched_net_init,
1934 .exit = psched_net_exit,
1935 };
1936
1937 static int __init pktsched_init(void)
1938 {
1939 int err;
1940
1941 err = register_pernet_subsys(&psched_net_ops);
1942 if (err) {
1943 pr_err("pktsched_init: "
1944 "cannot initialize per netns operations\n");
1945 return err;
1946 }
1947
1948 register_qdisc(&pfifo_fast_ops);
1949 register_qdisc(&pfifo_qdisc_ops);
1950 register_qdisc(&bfifo_qdisc_ops);
1951 register_qdisc(&pfifo_head_drop_qdisc_ops);
1952 register_qdisc(&mq_qdisc_ops);
1953
1954 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1955 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1956 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1957 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1958 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1959 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1960
1961 return 0;
1962 }
1963
1964 subsys_initcall(pktsched_init);
(deprecated) Btrfs devel tree