Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net/sched/sch_api.c Packet scheduler API.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 *
7 * Fixes:
8 *
9 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired.
10 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
11 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support
12 */
13
14#include <linux/module.h>
15#include <linux/types.h>
16#include <linux/kernel.h>
17#include <linux/string.h>
18#include <linux/errno.h>
19#include <linux/skbuff.h>
20#include <linux/init.h>
21#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/kmod.h>
24#include <linux/list.h>
25#include <linux/hrtimer.h>
26#include <linux/slab.h>
27#include <linux/hashtable.h>
28
29#include <net/net_namespace.h>
30#include <net/sock.h>
31#include <net/netlink.h>
32#include <net/pkt_sched.h>
33#include <net/pkt_cls.h>
34
35/*
36
37 Short review.
38 -------------
39
40 This file consists of two interrelated parts:
41
42 1. queueing disciplines manager frontend.
43 2. traffic classes manager frontend.
44
45 Generally, queueing discipline ("qdisc") is a black box,
46 which is able to enqueue packets and to dequeue them (when
47 device is ready to send something) in order and at times
48 determined by algorithm hidden in it.
49
50 qdisc's are divided to two categories:
51 - "queues", which have no internal structure visible from outside.
52 - "schedulers", which split all the packets to "traffic classes",
53 using "packet classifiers" (look at cls_api.c)
54
55 In turn, classes may have child qdiscs (as rule, queues)
56 attached to them etc. etc. etc.
57
58 The goal of the routines in this file is to translate
59 information supplied by user in the form of handles
60 to more intelligible for kernel form, to make some sanity
61 checks and part of work, which is common to all qdiscs
62 and to provide rtnetlink notifications.
63
64 All real intelligent work is done inside qdisc modules.
65
66
67
68 Every discipline has two major routines: enqueue and dequeue.
69
70 ---dequeue
71
72 dequeue usually returns a skb to send. It is allowed to return NULL,
73 but it does not mean that queue is empty, it just means that
74 discipline does not want to send anything this time.
75 Queue is really empty if q->q.qlen == 0.
76 For complicated disciplines with multiple queues q->q is not
77 real packet queue, but however q->q.qlen must be valid.
78
79 ---enqueue
80
81 enqueue returns 0, if packet was enqueued successfully.
82 If packet (this one or another one) was dropped, it returns
83 not zero error code.
84 NET_XMIT_DROP - this packet dropped
85 Expected action: do not backoff, but wait until queue will clear.
86 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
87 Expected action: backoff or ignore
88
89 Auxiliary routines:
90
91 ---peek
92
93 like dequeue but without removing a packet from the queue
94
95 ---reset
96
97 returns qdisc to initial state: purge all buffers, clear all
98 timers, counters (except for statistics) etc.
99
100 ---init
101
102 initializes newly created qdisc.
103
104 ---destroy
105
106 destroys resources allocated by init and during lifetime of qdisc.
107
108 ---change
109
110 changes qdisc parameters.
111 */
112
113/* Protects list of registered TC modules. It is pure SMP lock. */
114static DEFINE_RWLOCK(qdisc_mod_lock);
115
116
117/************************************************
118 * Queueing disciplines manipulation. *
119 ************************************************/
120
121
122/* The list of all installed queueing disciplines. */
123
124static struct Qdisc_ops *qdisc_base;
125
126/* Register/unregister queueing discipline */
127
128int register_qdisc(struct Qdisc_ops *qops)
129{
130 struct Qdisc_ops *q, **qp;
131 int rc = -EEXIST;
132
133 write_lock(&qdisc_mod_lock);
134 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
135 if (!strcmp(qops->id, q->id))
136 goto out;
137
138 if (qops->enqueue == NULL)
139 qops->enqueue = noop_qdisc_ops.enqueue;
140 if (qops->peek == NULL) {
141 if (qops->dequeue == NULL)
142 qops->peek = noop_qdisc_ops.peek;
143 else
144 goto out_einval;
145 }
146 if (qops->dequeue == NULL)
147 qops->dequeue = noop_qdisc_ops.dequeue;
148
149 if (qops->cl_ops) {
150 const struct Qdisc_class_ops *cops = qops->cl_ops;
151
152 if (!(cops->find && cops->walk && cops->leaf))
153 goto out_einval;
154
155 if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
156 goto out_einval;
157 }
158
159 qops->next = NULL;
160 *qp = qops;
161 rc = 0;
162out:
163 write_unlock(&qdisc_mod_lock);
164 return rc;
165
166out_einval:
167 rc = -EINVAL;
168 goto out;
169}
170EXPORT_SYMBOL(register_qdisc);
171
172int unregister_qdisc(struct Qdisc_ops *qops)
173{
174 struct Qdisc_ops *q, **qp;
175 int err = -ENOENT;
176
177 write_lock(&qdisc_mod_lock);
178 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
179 if (q == qops)
180 break;
181 if (q) {
182 *qp = q->next;
183 q->next = NULL;
184 err = 0;
185 }
186 write_unlock(&qdisc_mod_lock);
187 return err;
188}
189EXPORT_SYMBOL(unregister_qdisc);
190
191/* Get default qdisc if not otherwise specified */
192void qdisc_get_default(char *name, size_t len)
193{
194 read_lock(&qdisc_mod_lock);
195 strlcpy(name, default_qdisc_ops->id, len);
196 read_unlock(&qdisc_mod_lock);
197}
198
199static struct Qdisc_ops *qdisc_lookup_default(const char *name)
200{
201 struct Qdisc_ops *q = NULL;
202
203 for (q = qdisc_base; q; q = q->next) {
204 if (!strcmp(name, q->id)) {
205 if (!try_module_get(q->owner))
206 q = NULL;
207 break;
208 }
209 }
210
211 return q;
212}
213
214/* Set new default qdisc to use */
215int qdisc_set_default(const char *name)
216{
217 const struct Qdisc_ops *ops;
218
219 if (!capable(CAP_NET_ADMIN))
220 return -EPERM;
221
222 write_lock(&qdisc_mod_lock);
223 ops = qdisc_lookup_default(name);
224 if (!ops) {
225 /* Not found, drop lock and try to load module */
226 write_unlock(&qdisc_mod_lock);
227 request_module("sch_%s", name);
228 write_lock(&qdisc_mod_lock);
229
230 ops = qdisc_lookup_default(name);
231 }
232
233 if (ops) {
234 /* Set new default */
235 module_put(default_qdisc_ops->owner);
236 default_qdisc_ops = ops;
237 }
238 write_unlock(&qdisc_mod_lock);
239
240 return ops ? 0 : -ENOENT;
241}
242
243#ifdef CONFIG_NET_SCH_DEFAULT
244/* Set default value from kernel config */
245static int __init sch_default_qdisc(void)
246{
247 return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
248}
249late_initcall(sch_default_qdisc);
250#endif
251
252/* We know handle. Find qdisc among all qdisc's attached to device
253 * (root qdisc, all its children, children of children etc.)
254 * Note: caller either uses rtnl or rcu_read_lock()
255 */
256
257static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
258{
259 struct Qdisc *q;
260
261 if (!qdisc_dev(root))
262 return (root->handle == handle ? root : NULL);
263
264 if (!(root->flags & TCQ_F_BUILTIN) &&
265 root->handle == handle)
266 return root;
267
268 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) {
269 if (q->handle == handle)
270 return q;
271 }
272 return NULL;
273}
274
275void qdisc_hash_add(struct Qdisc *q, bool invisible)
276{
277 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
278 ASSERT_RTNL();
279 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
280 if (invisible)
281 q->flags |= TCQ_F_INVISIBLE;
282 }
283}
284EXPORT_SYMBOL(qdisc_hash_add);
285
286void qdisc_hash_del(struct Qdisc *q)
287{
288 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
289 ASSERT_RTNL();
290 hash_del_rcu(&q->hash);
291 }
292}
293EXPORT_SYMBOL(qdisc_hash_del);
294
295struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
296{
297 struct Qdisc *q;
298
299 if (!handle)
300 return NULL;
301 q = qdisc_match_from_root(dev->qdisc, handle);
302 if (q)
303 goto out;
304
305 if (dev_ingress_queue(dev))
306 q = qdisc_match_from_root(
307 dev_ingress_queue(dev)->qdisc_sleeping,
308 handle);
309out:
310 return q;
311}
312
313struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
314{
315 struct netdev_queue *nq;
316 struct Qdisc *q;
317
318 if (!handle)
319 return NULL;
320 q = qdisc_match_from_root(dev->qdisc, handle);
321 if (q)
322 goto out;
323
324 nq = dev_ingress_queue_rcu(dev);
325 if (nq)
326 q = qdisc_match_from_root(nq->qdisc_sleeping, handle);
327out:
328 return q;
329}
330
331static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
332{
333 unsigned long cl;
334 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
335
336 if (cops == NULL)
337 return NULL;
338 cl = cops->find(p, classid);
339
340 if (cl == 0)
341 return NULL;
342 return cops->leaf(p, cl);
343}
344
345/* Find queueing discipline by name */
346
347static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
348{
349 struct Qdisc_ops *q = NULL;
350
351 if (kind) {
352 read_lock(&qdisc_mod_lock);
353 for (q = qdisc_base; q; q = q->next) {
354 if (nla_strcmp(kind, q->id) == 0) {
355 if (!try_module_get(q->owner))
356 q = NULL;
357 break;
358 }
359 }
360 read_unlock(&qdisc_mod_lock);
361 }
362 return q;
363}
364
365/* The linklayer setting were not transferred from iproute2, in older
366 * versions, and the rate tables lookup systems have been dropped in
367 * the kernel. To keep backward compatible with older iproute2 tc
368 * utils, we detect the linklayer setting by detecting if the rate
369 * table were modified.
370 *
371 * For linklayer ATM table entries, the rate table will be aligned to
372 * 48 bytes, thus some table entries will contain the same value. The
373 * mpu (min packet unit) is also encoded into the old rate table, thus
374 * starting from the mpu, we find low and high table entries for
375 * mapping this cell. If these entries contain the same value, when
376 * the rate tables have been modified for linklayer ATM.
377 *
378 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
379 * and then roundup to the next cell, calc the table entry one below,
380 * and compare.
381 */
382static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
383{
384 int low = roundup(r->mpu, 48);
385 int high = roundup(low+1, 48);
386 int cell_low = low >> r->cell_log;
387 int cell_high = (high >> r->cell_log) - 1;
388
389 /* rtab is too inaccurate at rates > 100Mbit/s */
390 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
391 pr_debug("TC linklayer: Giving up ATM detection\n");
392 return TC_LINKLAYER_ETHERNET;
393 }
394
395 if ((cell_high > cell_low) && (cell_high < 256)
396 && (rtab[cell_low] == rtab[cell_high])) {
397 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
398 cell_low, cell_high, rtab[cell_high]);
399 return TC_LINKLAYER_ATM;
400 }
401 return TC_LINKLAYER_ETHERNET;
402}
403
404static struct qdisc_rate_table *qdisc_rtab_list;
405
406struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
407 struct nlattr *tab,
408 struct netlink_ext_ack *extack)
409{
410 struct qdisc_rate_table *rtab;
411
412 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
413 nla_len(tab) != TC_RTAB_SIZE) {
414 NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
415 return NULL;
416 }
417
418 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
419 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
420 !memcmp(&rtab->data, nla_data(tab), 1024)) {
421 rtab->refcnt++;
422 return rtab;
423 }
424 }
425
426 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
427 if (rtab) {
428 rtab->rate = *r;
429 rtab->refcnt = 1;
430 memcpy(rtab->data, nla_data(tab), 1024);
431 if (r->linklayer == TC_LINKLAYER_UNAWARE)
432 r->linklayer = __detect_linklayer(r, rtab->data);
433 rtab->next = qdisc_rtab_list;
434 qdisc_rtab_list = rtab;
435 } else {
436 NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
437 }
438 return rtab;
439}
440EXPORT_SYMBOL(qdisc_get_rtab);
441
442void qdisc_put_rtab(struct qdisc_rate_table *tab)
443{
444 struct qdisc_rate_table *rtab, **rtabp;
445
446 if (!tab || --tab->refcnt)
447 return;
448
449 for (rtabp = &qdisc_rtab_list;
450 (rtab = *rtabp) != NULL;
451 rtabp = &rtab->next) {
452 if (rtab == tab) {
453 *rtabp = rtab->next;
454 kfree(rtab);
455 return;
456 }
457 }
458}
459EXPORT_SYMBOL(qdisc_put_rtab);
460
461static LIST_HEAD(qdisc_stab_list);
462
463static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
464 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
465 [TCA_STAB_DATA] = { .type = NLA_BINARY },
466};
467
468static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
469 struct netlink_ext_ack *extack)
470{
471 struct nlattr *tb[TCA_STAB_MAX + 1];
472 struct qdisc_size_table *stab;
473 struct tc_sizespec *s;
474 unsigned int tsize = 0;
475 u16 *tab = NULL;
476 int err;
477
478 err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
479 extack);
480 if (err < 0)
481 return ERR_PTR(err);
482 if (!tb[TCA_STAB_BASE]) {
483 NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
484 return ERR_PTR(-EINVAL);
485 }
486
487 s = nla_data(tb[TCA_STAB_BASE]);
488
489 if (s->tsize > 0) {
490 if (!tb[TCA_STAB_DATA]) {
491 NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
492 return ERR_PTR(-EINVAL);
493 }
494 tab = nla_data(tb[TCA_STAB_DATA]);
495 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
496 }
497
498 if (tsize != s->tsize || (!tab && tsize > 0)) {
499 NL_SET_ERR_MSG(extack, "Invalid size of size table");
500 return ERR_PTR(-EINVAL);
501 }
502
503 list_for_each_entry(stab, &qdisc_stab_list, list) {
504 if (memcmp(&stab->szopts, s, sizeof(*s)))
505 continue;
506 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
507 continue;
508 stab->refcnt++;
509 return stab;
510 }
511
512 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
513 if (!stab)
514 return ERR_PTR(-ENOMEM);
515
516 stab->refcnt = 1;
517 stab->szopts = *s;
518 if (tsize > 0)
519 memcpy(stab->data, tab, tsize * sizeof(u16));
520
521 list_add_tail(&stab->list, &qdisc_stab_list);
522
523 return stab;
524}
525
526void qdisc_put_stab(struct qdisc_size_table *tab)
527{
528 if (!tab)
529 return;
530
531 if (--tab->refcnt == 0) {
532 list_del(&tab->list);
533 kfree_rcu(tab, rcu);
534 }
535}
536EXPORT_SYMBOL(qdisc_put_stab);
537
538static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
539{
540 struct nlattr *nest;
541
542 nest = nla_nest_start_noflag(skb, TCA_STAB);
543 if (nest == NULL)
544 goto nla_put_failure;
545 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
546 goto nla_put_failure;
547 nla_nest_end(skb, nest);
548
549 return skb->len;
550
551nla_put_failure:
552 return -1;
553}
554
555void __qdisc_calculate_pkt_len(struct sk_buff *skb,
556 const struct qdisc_size_table *stab)
557{
558 int pkt_len, slot;
559
560 pkt_len = skb->len + stab->szopts.overhead;
561 if (unlikely(!stab->szopts.tsize))
562 goto out;
563
564 slot = pkt_len + stab->szopts.cell_align;
565 if (unlikely(slot < 0))
566 slot = 0;
567
568 slot >>= stab->szopts.cell_log;
569 if (likely(slot < stab->szopts.tsize))
570 pkt_len = stab->data[slot];
571 else
572 pkt_len = stab->data[stab->szopts.tsize - 1] *
573 (slot / stab->szopts.tsize) +
574 stab->data[slot % stab->szopts.tsize];
575
576 pkt_len <<= stab->szopts.size_log;
577out:
578 if (unlikely(pkt_len < 1))
579 pkt_len = 1;
580 qdisc_skb_cb(skb)->pkt_len = pkt_len;
581}
582EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
583
584void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
585{
586 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
587 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
588 txt, qdisc->ops->id, qdisc->handle >> 16);
589 qdisc->flags |= TCQ_F_WARN_NONWC;
590 }
591}
592EXPORT_SYMBOL(qdisc_warn_nonwc);
593
594static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
595{
596 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
597 timer);
598
599 rcu_read_lock();
600 __netif_schedule(qdisc_root(wd->qdisc));
601 rcu_read_unlock();
602
603 return HRTIMER_NORESTART;
604}
605
606void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
607 clockid_t clockid)
608{
609 hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
610 wd->timer.function = qdisc_watchdog;
611 wd->qdisc = qdisc;
612}
613EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
614
615void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
616{
617 qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
618}
619EXPORT_SYMBOL(qdisc_watchdog_init);
620
621void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
622{
623 if (test_bit(__QDISC_STATE_DEACTIVATED,
624 &qdisc_root_sleeping(wd->qdisc)->state))
625 return;
626
627 if (wd->last_expires == expires)
628 return;
629
630 wd->last_expires = expires;
631 hrtimer_start(&wd->timer,
632 ns_to_ktime(expires),
633 HRTIMER_MODE_ABS_PINNED);
634}
635EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
636
637void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
638{
639 hrtimer_cancel(&wd->timer);
640}
641EXPORT_SYMBOL(qdisc_watchdog_cancel);
642
643static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
644{
645 struct hlist_head *h;
646 unsigned int i;
647
648 h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
649
650 if (h != NULL) {
651 for (i = 0; i < n; i++)
652 INIT_HLIST_HEAD(&h[i]);
653 }
654 return h;
655}
656
657void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
658{
659 struct Qdisc_class_common *cl;
660 struct hlist_node *next;
661 struct hlist_head *nhash, *ohash;
662 unsigned int nsize, nmask, osize;
663 unsigned int i, h;
664
665 /* Rehash when load factor exceeds 0.75 */
666 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
667 return;
668 nsize = clhash->hashsize * 2;
669 nmask = nsize - 1;
670 nhash = qdisc_class_hash_alloc(nsize);
671 if (nhash == NULL)
672 return;
673
674 ohash = clhash->hash;
675 osize = clhash->hashsize;
676
677 sch_tree_lock(sch);
678 for (i = 0; i < osize; i++) {
679 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
680 h = qdisc_class_hash(cl->classid, nmask);
681 hlist_add_head(&cl->hnode, &nhash[h]);
682 }
683 }
684 clhash->hash = nhash;
685 clhash->hashsize = nsize;
686 clhash->hashmask = nmask;
687 sch_tree_unlock(sch);
688
689 kvfree(ohash);
690}
691EXPORT_SYMBOL(qdisc_class_hash_grow);
692
693int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
694{
695 unsigned int size = 4;
696
697 clhash->hash = qdisc_class_hash_alloc(size);
698 if (!clhash->hash)
699 return -ENOMEM;
700 clhash->hashsize = size;
701 clhash->hashmask = size - 1;
702 clhash->hashelems = 0;
703 return 0;
704}
705EXPORT_SYMBOL(qdisc_class_hash_init);
706
707void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
708{
709 kvfree(clhash->hash);
710}
711EXPORT_SYMBOL(qdisc_class_hash_destroy);
712
713void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
714 struct Qdisc_class_common *cl)
715{
716 unsigned int h;
717
718 INIT_HLIST_NODE(&cl->hnode);
719 h = qdisc_class_hash(cl->classid, clhash->hashmask);
720 hlist_add_head(&cl->hnode, &clhash->hash[h]);
721 clhash->hashelems++;
722}
723EXPORT_SYMBOL(qdisc_class_hash_insert);
724
725void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
726 struct Qdisc_class_common *cl)
727{
728 hlist_del(&cl->hnode);
729 clhash->hashelems--;
730}
731EXPORT_SYMBOL(qdisc_class_hash_remove);
732
733/* Allocate an unique handle from space managed by kernel
734 * Possible range is [8000-FFFF]:0000 (0x8000 values)
735 */
736static u32 qdisc_alloc_handle(struct net_device *dev)
737{
738 int i = 0x8000;
739 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
740
741 do {
742 autohandle += TC_H_MAKE(0x10000U, 0);
743 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
744 autohandle = TC_H_MAKE(0x80000000U, 0);
745 if (!qdisc_lookup(dev, autohandle))
746 return autohandle;
747 cond_resched();
748 } while (--i > 0);
749
750 return 0;
751}
752
753void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
754{
755 bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
756 const struct Qdisc_class_ops *cops;
757 unsigned long cl;
758 u32 parentid;
759 bool notify;
760 int drops;
761
762 if (n == 0 && len == 0)
763 return;
764 drops = max_t(int, n, 0);
765 rcu_read_lock();
766 while ((parentid = sch->parent)) {
767 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
768 break;
769
770 if (sch->flags & TCQ_F_NOPARENT)
771 break;
772 /* Notify parent qdisc only if child qdisc becomes empty.
773 *
774 * If child was empty even before update then backlog
775 * counter is screwed and we skip notification because
776 * parent class is already passive.
777 *
778 * If the original child was offloaded then it is allowed
779 * to be seem as empty, so the parent is notified anyway.
780 */
781 notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
782 !qdisc_is_offloaded);
783 /* TODO: perform the search on a per txq basis */
784 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
785 if (sch == NULL) {
786 WARN_ON_ONCE(parentid != TC_H_ROOT);
787 break;
788 }
789 cops = sch->ops->cl_ops;
790 if (notify && cops->qlen_notify) {
791 cl = cops->find(sch, parentid);
792 cops->qlen_notify(sch, cl);
793 }
794 sch->q.qlen -= n;
795 sch->qstats.backlog -= len;
796 __qdisc_qstats_drop(sch, drops);
797 }
798 rcu_read_unlock();
799}
800EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
801
802int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
803 void *type_data)
804{
805 struct net_device *dev = qdisc_dev(sch);
806 int err;
807
808 sch->flags &= ~TCQ_F_OFFLOADED;
809 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
810 return 0;
811
812 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
813 if (err == -EOPNOTSUPP)
814 return 0;
815
816 if (!err)
817 sch->flags |= TCQ_F_OFFLOADED;
818
819 return err;
820}
821EXPORT_SYMBOL(qdisc_offload_dump_helper);
822
823void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
824 struct Qdisc *new, struct Qdisc *old,
825 enum tc_setup_type type, void *type_data,
826 struct netlink_ext_ack *extack)
827{
828 bool any_qdisc_is_offloaded;
829 int err;
830
831 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
832 return;
833
834 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
835
836 /* Don't report error if the graft is part of destroy operation. */
837 if (!err || !new || new == &noop_qdisc)
838 return;
839
840 /* Don't report error if the parent, the old child and the new
841 * one are not offloaded.
842 */
843 any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
844 any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
845 any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
846
847 if (any_qdisc_is_offloaded)
848 NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
849}
850EXPORT_SYMBOL(qdisc_offload_graft_helper);
851
852static void qdisc_offload_graft_root(struct net_device *dev,
853 struct Qdisc *new, struct Qdisc *old,
854 struct netlink_ext_ack *extack)
855{
856 struct tc_root_qopt_offload graft_offload = {
857 .command = TC_ROOT_GRAFT,
858 .handle = new ? new->handle : 0,
859 .ingress = (new && new->flags & TCQ_F_INGRESS) ||
860 (old && old->flags & TCQ_F_INGRESS),
861 };
862
863 qdisc_offload_graft_helper(dev, NULL, new, old,
864 TC_SETUP_ROOT_QDISC, &graft_offload, extack);
865}
866
867static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
868 u32 portid, u32 seq, u16 flags, int event)
869{
870 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
871 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
872 struct tcmsg *tcm;
873 struct nlmsghdr *nlh;
874 unsigned char *b = skb_tail_pointer(skb);
875 struct gnet_dump d;
876 struct qdisc_size_table *stab;
877 u32 block_index;
878 __u32 qlen;
879
880 cond_resched();
881 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
882 if (!nlh)
883 goto out_nlmsg_trim;
884 tcm = nlmsg_data(nlh);
885 tcm->tcm_family = AF_UNSPEC;
886 tcm->tcm__pad1 = 0;
887 tcm->tcm__pad2 = 0;
888 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
889 tcm->tcm_parent = clid;
890 tcm->tcm_handle = q->handle;
891 tcm->tcm_info = refcount_read(&q->refcnt);
892 if (nla_put_string(skb, TCA_KIND, q->ops->id))
893 goto nla_put_failure;
894 if (q->ops->ingress_block_get) {
895 block_index = q->ops->ingress_block_get(q);
896 if (block_index &&
897 nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
898 goto nla_put_failure;
899 }
900 if (q->ops->egress_block_get) {
901 block_index = q->ops->egress_block_get(q);
902 if (block_index &&
903 nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
904 goto nla_put_failure;
905 }
906 if (q->ops->dump && q->ops->dump(q, skb) < 0)
907 goto nla_put_failure;
908 if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
909 goto nla_put_failure;
910 qlen = qdisc_qlen_sum(q);
911
912 stab = rtnl_dereference(q->stab);
913 if (stab && qdisc_dump_stab(skb, stab) < 0)
914 goto nla_put_failure;
915
916 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
917 NULL, &d, TCA_PAD) < 0)
918 goto nla_put_failure;
919
920 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
921 goto nla_put_failure;
922
923 if (qdisc_is_percpu_stats(q)) {
924 cpu_bstats = q->cpu_bstats;
925 cpu_qstats = q->cpu_qstats;
926 }
927
928 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
929 &d, cpu_bstats, &q->bstats) < 0 ||
930 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
931 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
932 goto nla_put_failure;
933
934 if (gnet_stats_finish_copy(&d) < 0)
935 goto nla_put_failure;
936
937 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
938 return skb->len;
939
940out_nlmsg_trim:
941nla_put_failure:
942 nlmsg_trim(skb, b);
943 return -1;
944}
945
946static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
947{
948 if (q->flags & TCQ_F_BUILTIN)
949 return true;
950 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
951 return true;
952
953 return false;
954}
955
956static int qdisc_notify(struct net *net, struct sk_buff *oskb,
957 struct nlmsghdr *n, u32 clid,
958 struct Qdisc *old, struct Qdisc *new)
959{
960 struct sk_buff *skb;
961 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
962
963 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
964 if (!skb)
965 return -ENOBUFS;
966
967 if (old && !tc_qdisc_dump_ignore(old, false)) {
968 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
969 0, RTM_DELQDISC) < 0)
970 goto err_out;
971 }
972 if (new && !tc_qdisc_dump_ignore(new, false)) {
973 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
974 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
975 goto err_out;
976 }
977
978 if (skb->len)
979 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
980 n->nlmsg_flags & NLM_F_ECHO);
981
982err_out:
983 kfree_skb(skb);
984 return -EINVAL;
985}
986
987static void notify_and_destroy(struct net *net, struct sk_buff *skb,
988 struct nlmsghdr *n, u32 clid,
989 struct Qdisc *old, struct Qdisc *new)
990{
991 if (new || old)
992 qdisc_notify(net, skb, n, clid, old, new);
993
994 if (old)
995 qdisc_put(old);
996}
997
998static void qdisc_clear_nolock(struct Qdisc *sch)
999{
1000 sch->flags &= ~TCQ_F_NOLOCK;
1001 if (!(sch->flags & TCQ_F_CPUSTATS))
1002 return;
1003
1004 free_percpu(sch->cpu_bstats);
1005 free_percpu(sch->cpu_qstats);
1006 sch->cpu_bstats = NULL;
1007 sch->cpu_qstats = NULL;
1008 sch->flags &= ~TCQ_F_CPUSTATS;
1009}
1010
1011/* Graft qdisc "new" to class "classid" of qdisc "parent" or
1012 * to device "dev".
1013 *
1014 * When appropriate send a netlink notification using 'skb'
1015 * and "n".
1016 *
1017 * On success, destroy old qdisc.
1018 */
1019
1020static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1021 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1022 struct Qdisc *new, struct Qdisc *old,
1023 struct netlink_ext_ack *extack)
1024{
1025 struct Qdisc *q = old;
1026 struct net *net = dev_net(dev);
1027
1028 if (parent == NULL) {
1029 unsigned int i, num_q, ingress;
1030
1031 ingress = 0;
1032 num_q = dev->num_tx_queues;
1033 if ((q && q->flags & TCQ_F_INGRESS) ||
1034 (new && new->flags & TCQ_F_INGRESS)) {
1035 num_q = 1;
1036 ingress = 1;
1037 if (!dev_ingress_queue(dev)) {
1038 NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1039 return -ENOENT;
1040 }
1041 }
1042
1043 if (dev->flags & IFF_UP)
1044 dev_deactivate(dev);
1045
1046 qdisc_offload_graft_root(dev, new, old, extack);
1047
1048 if (new && new->ops->attach)
1049 goto skip;
1050
1051 for (i = 0; i < num_q; i++) {
1052 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1053
1054 if (!ingress)
1055 dev_queue = netdev_get_tx_queue(dev, i);
1056
1057 old = dev_graft_qdisc(dev_queue, new);
1058 if (new && i > 0)
1059 qdisc_refcount_inc(new);
1060
1061 if (!ingress)
1062 qdisc_put(old);
1063 }
1064
1065skip:
1066 if (!ingress) {
1067 notify_and_destroy(net, skb, n, classid,
1068 dev->qdisc, new);
1069 if (new && !new->ops->attach)
1070 qdisc_refcount_inc(new);
1071 dev->qdisc = new ? : &noop_qdisc;
1072
1073 if (new && new->ops->attach)
1074 new->ops->attach(new);
1075 } else {
1076 notify_and_destroy(net, skb, n, classid, old, new);
1077 }
1078
1079 if (dev->flags & IFF_UP)
1080 dev_activate(dev);
1081 } else {
1082 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1083 unsigned long cl;
1084 int err;
1085
1086 /* Only support running class lockless if parent is lockless */
1087 if (new && (new->flags & TCQ_F_NOLOCK) &&
1088 parent && !(parent->flags & TCQ_F_NOLOCK))
1089 qdisc_clear_nolock(new);
1090
1091 if (!cops || !cops->graft)
1092 return -EOPNOTSUPP;
1093
1094 cl = cops->find(parent, classid);
1095 if (!cl) {
1096 NL_SET_ERR_MSG(extack, "Specified class not found");
1097 return -ENOENT;
1098 }
1099
1100 err = cops->graft(parent, cl, new, &old, extack);
1101 if (err)
1102 return err;
1103 notify_and_destroy(net, skb, n, classid, old, new);
1104 }
1105 return 0;
1106}
1107
1108static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1109 struct netlink_ext_ack *extack)
1110{
1111 u32 block_index;
1112
1113 if (tca[TCA_INGRESS_BLOCK]) {
1114 block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1115
1116 if (!block_index) {
1117 NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1118 return -EINVAL;
1119 }
1120 if (!sch->ops->ingress_block_set) {
1121 NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1122 return -EOPNOTSUPP;
1123 }
1124 sch->ops->ingress_block_set(sch, block_index);
1125 }
1126 if (tca[TCA_EGRESS_BLOCK]) {
1127 block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1128
1129 if (!block_index) {
1130 NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1131 return -EINVAL;
1132 }
1133 if (!sch->ops->egress_block_set) {
1134 NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1135 return -EOPNOTSUPP;
1136 }
1137 sch->ops->egress_block_set(sch, block_index);
1138 }
1139 return 0;
1140}
1141
1142/*
1143 Allocate and initialize new qdisc.
1144
1145 Parameters are passed via opt.
1146 */
1147
1148static struct Qdisc *qdisc_create(struct net_device *dev,
1149 struct netdev_queue *dev_queue,
1150 struct Qdisc *p, u32 parent, u32 handle,
1151 struct nlattr **tca, int *errp,
1152 struct netlink_ext_ack *extack)
1153{
1154 int err;
1155 struct nlattr *kind = tca[TCA_KIND];
1156 struct Qdisc *sch;
1157 struct Qdisc_ops *ops;
1158 struct qdisc_size_table *stab;
1159
1160 ops = qdisc_lookup_ops(kind);
1161#ifdef CONFIG_MODULES
1162 if (ops == NULL && kind != NULL) {
1163 char name[IFNAMSIZ];
1164 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
1165 /* We dropped the RTNL semaphore in order to
1166 * perform the module load. So, even if we
1167 * succeeded in loading the module we have to
1168 * tell the caller to replay the request. We
1169 * indicate this using -EAGAIN.
1170 * We replay the request because the device may
1171 * go away in the mean time.
1172 */
1173 rtnl_unlock();
1174 request_module("sch_%s", name);
1175 rtnl_lock();
1176 ops = qdisc_lookup_ops(kind);
1177 if (ops != NULL) {
1178 /* We will try again qdisc_lookup_ops,
1179 * so don't keep a reference.
1180 */
1181 module_put(ops->owner);
1182 err = -EAGAIN;
1183 goto err_out;
1184 }
1185 }
1186 }
1187#endif
1188
1189 err = -ENOENT;
1190 if (!ops) {
1191 NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1192 goto err_out;
1193 }
1194
1195 sch = qdisc_alloc(dev_queue, ops, extack);
1196 if (IS_ERR(sch)) {
1197 err = PTR_ERR(sch);
1198 goto err_out2;
1199 }
1200
1201 sch->parent = parent;
1202
1203 if (handle == TC_H_INGRESS) {
1204 sch->flags |= TCQ_F_INGRESS;
1205 handle = TC_H_MAKE(TC_H_INGRESS, 0);
1206 } else {
1207 if (handle == 0) {
1208 handle = qdisc_alloc_handle(dev);
1209 if (handle == 0) {
1210 NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1211 err = -ENOSPC;
1212 goto err_out3;
1213 }
1214 }
1215 if (!netif_is_multiqueue(dev))
1216 sch->flags |= TCQ_F_ONETXQUEUE;
1217 }
1218
1219 sch->handle = handle;
1220
1221 /* This exist to keep backward compatible with a userspace
1222 * loophole, what allowed userspace to get IFF_NO_QUEUE
1223 * facility on older kernels by setting tx_queue_len=0 (prior
1224 * to qdisc init), and then forgot to reinit tx_queue_len
1225 * before again attaching a qdisc.
1226 */
1227 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1228 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1229 netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1230 }
1231
1232 err = qdisc_block_indexes_set(sch, tca, extack);
1233 if (err)
1234 goto err_out3;
1235
1236 if (ops->init) {
1237 err = ops->init(sch, tca[TCA_OPTIONS], extack);
1238 if (err != 0)
1239 goto err_out5;
1240 }
1241
1242 if (tca[TCA_STAB]) {
1243 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1244 if (IS_ERR(stab)) {
1245 err = PTR_ERR(stab);
1246 goto err_out4;
1247 }
1248 rcu_assign_pointer(sch->stab, stab);
1249 }
1250 if (tca[TCA_RATE]) {
1251 seqcount_t *running;
1252
1253 err = -EOPNOTSUPP;
1254 if (sch->flags & TCQ_F_MQROOT) {
1255 NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1256 goto err_out4;
1257 }
1258
1259 if (sch->parent != TC_H_ROOT &&
1260 !(sch->flags & TCQ_F_INGRESS) &&
1261 (!p || !(p->flags & TCQ_F_MQROOT)))
1262 running = qdisc_root_sleeping_running(sch);
1263 else
1264 running = &sch->running;
1265
1266 err = gen_new_estimator(&sch->bstats,
1267 sch->cpu_bstats,
1268 &sch->rate_est,
1269 NULL,
1270 running,
1271 tca[TCA_RATE]);
1272 if (err) {
1273 NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1274 goto err_out4;
1275 }
1276 }
1277
1278 qdisc_hash_add(sch, false);
1279
1280 return sch;
1281
1282err_out5:
1283 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1284 if (ops->destroy)
1285 ops->destroy(sch);
1286err_out3:
1287 dev_put(dev);
1288 qdisc_free(sch);
1289err_out2:
1290 module_put(ops->owner);
1291err_out:
1292 *errp = err;
1293 return NULL;
1294
1295err_out4:
1296 /*
1297 * Any broken qdiscs that would require a ops->reset() here?
1298 * The qdisc was never in action so it shouldn't be necessary.
1299 */
1300 qdisc_put_stab(rtnl_dereference(sch->stab));
1301 if (ops->destroy)
1302 ops->destroy(sch);
1303 goto err_out3;
1304}
1305
1306static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1307 struct netlink_ext_ack *extack)
1308{
1309 struct qdisc_size_table *ostab, *stab = NULL;
1310 int err = 0;
1311
1312 if (tca[TCA_OPTIONS]) {
1313 if (!sch->ops->change) {
1314 NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1315 return -EINVAL;
1316 }
1317 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1318 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1319 return -EOPNOTSUPP;
1320 }
1321 err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1322 if (err)
1323 return err;
1324 }
1325
1326 if (tca[TCA_STAB]) {
1327 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1328 if (IS_ERR(stab))
1329 return PTR_ERR(stab);
1330 }
1331
1332 ostab = rtnl_dereference(sch->stab);
1333 rcu_assign_pointer(sch->stab, stab);
1334 qdisc_put_stab(ostab);
1335
1336 if (tca[TCA_RATE]) {
1337 /* NB: ignores errors from replace_estimator
1338 because change can't be undone. */
1339 if (sch->flags & TCQ_F_MQROOT)
1340 goto out;
1341 gen_replace_estimator(&sch->bstats,
1342 sch->cpu_bstats,
1343 &sch->rate_est,
1344 NULL,
1345 qdisc_root_sleeping_running(sch),
1346 tca[TCA_RATE]);
1347 }
1348out:
1349 return 0;
1350}
1351
1352struct check_loop_arg {
1353 struct qdisc_walker w;
1354 struct Qdisc *p;
1355 int depth;
1356};
1357
1358static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1359 struct qdisc_walker *w);
1360
1361static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1362{
1363 struct check_loop_arg arg;
1364
1365 if (q->ops->cl_ops == NULL)
1366 return 0;
1367
1368 arg.w.stop = arg.w.skip = arg.w.count = 0;
1369 arg.w.fn = check_loop_fn;
1370 arg.depth = depth;
1371 arg.p = p;
1372 q->ops->cl_ops->walk(q, &arg.w);
1373 return arg.w.stop ? -ELOOP : 0;
1374}
1375
1376static int
1377check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1378{
1379 struct Qdisc *leaf;
1380 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1381 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1382
1383 leaf = cops->leaf(q, cl);
1384 if (leaf) {
1385 if (leaf == arg->p || arg->depth > 7)
1386 return -ELOOP;
1387 return check_loop(leaf, arg->p, arg->depth + 1);
1388 }
1389 return 0;
1390}
1391
1392const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1393 [TCA_KIND] = { .type = NLA_STRING },
1394 [TCA_RATE] = { .type = NLA_BINARY,
1395 .len = sizeof(struct tc_estimator) },
1396 [TCA_STAB] = { .type = NLA_NESTED },
1397 [TCA_DUMP_INVISIBLE] = { .type = NLA_FLAG },
1398 [TCA_CHAIN] = { .type = NLA_U32 },
1399 [TCA_INGRESS_BLOCK] = { .type = NLA_U32 },
1400 [TCA_EGRESS_BLOCK] = { .type = NLA_U32 },
1401};
1402
1403/*
1404 * Delete/get qdisc.
1405 */
1406
1407static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1408 struct netlink_ext_ack *extack)
1409{
1410 struct net *net = sock_net(skb->sk);
1411 struct tcmsg *tcm = nlmsg_data(n);
1412 struct nlattr *tca[TCA_MAX + 1];
1413 struct net_device *dev;
1414 u32 clid;
1415 struct Qdisc *q = NULL;
1416 struct Qdisc *p = NULL;
1417 int err;
1418
1419 if ((n->nlmsg_type != RTM_GETQDISC) &&
1420 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1421 return -EPERM;
1422
1423 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1424 rtm_tca_policy, extack);
1425 if (err < 0)
1426 return err;
1427
1428 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1429 if (!dev)
1430 return -ENODEV;
1431
1432 clid = tcm->tcm_parent;
1433 if (clid) {
1434 if (clid != TC_H_ROOT) {
1435 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1436 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1437 if (!p) {
1438 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1439 return -ENOENT;
1440 }
1441 q = qdisc_leaf(p, clid);
1442 } else if (dev_ingress_queue(dev)) {
1443 q = dev_ingress_queue(dev)->qdisc_sleeping;
1444 }
1445 } else {
1446 q = dev->qdisc;
1447 }
1448 if (!q) {
1449 NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1450 return -ENOENT;
1451 }
1452
1453 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1454 NL_SET_ERR_MSG(extack, "Invalid handle");
1455 return -EINVAL;
1456 }
1457 } else {
1458 q = qdisc_lookup(dev, tcm->tcm_handle);
1459 if (!q) {
1460 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1461 return -ENOENT;
1462 }
1463 }
1464
1465 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1466 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1467 return -EINVAL;
1468 }
1469
1470 if (n->nlmsg_type == RTM_DELQDISC) {
1471 if (!clid) {
1472 NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1473 return -EINVAL;
1474 }
1475 if (q->handle == 0) {
1476 NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1477 return -ENOENT;
1478 }
1479 err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1480 if (err != 0)
1481 return err;
1482 } else {
1483 qdisc_notify(net, skb, n, clid, NULL, q);
1484 }
1485 return 0;
1486}
1487
1488/*
1489 * Create/change qdisc.
1490 */
1491
1492static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1493 struct netlink_ext_ack *extack)
1494{
1495 struct net *net = sock_net(skb->sk);
1496 struct tcmsg *tcm;
1497 struct nlattr *tca[TCA_MAX + 1];
1498 struct net_device *dev;
1499 u32 clid;
1500 struct Qdisc *q, *p;
1501 int err;
1502
1503 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1504 return -EPERM;
1505
1506replay:
1507 /* Reinit, just in case something touches this. */
1508 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1509 rtm_tca_policy, extack);
1510 if (err < 0)
1511 return err;
1512
1513 tcm = nlmsg_data(n);
1514 clid = tcm->tcm_parent;
1515 q = p = NULL;
1516
1517 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1518 if (!dev)
1519 return -ENODEV;
1520
1521
1522 if (clid) {
1523 if (clid != TC_H_ROOT) {
1524 if (clid != TC_H_INGRESS) {
1525 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1526 if (!p) {
1527 NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1528 return -ENOENT;
1529 }
1530 q = qdisc_leaf(p, clid);
1531 } else if (dev_ingress_queue_create(dev)) {
1532 q = dev_ingress_queue(dev)->qdisc_sleeping;
1533 }
1534 } else {
1535 q = dev->qdisc;
1536 }
1537
1538 /* It may be default qdisc, ignore it */
1539 if (q && q->handle == 0)
1540 q = NULL;
1541
1542 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1543 if (tcm->tcm_handle) {
1544 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1545 NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1546 return -EEXIST;
1547 }
1548 if (TC_H_MIN(tcm->tcm_handle)) {
1549 NL_SET_ERR_MSG(extack, "Invalid minor handle");
1550 return -EINVAL;
1551 }
1552 q = qdisc_lookup(dev, tcm->tcm_handle);
1553 if (!q)
1554 goto create_n_graft;
1555 if (n->nlmsg_flags & NLM_F_EXCL) {
1556 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1557 return -EEXIST;
1558 }
1559 if (tca[TCA_KIND] &&
1560 nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1561 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1562 return -EINVAL;
1563 }
1564 if (q == p ||
1565 (p && check_loop(q, p, 0))) {
1566 NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1567 return -ELOOP;
1568 }
1569 qdisc_refcount_inc(q);
1570 goto graft;
1571 } else {
1572 if (!q)
1573 goto create_n_graft;
1574
1575 /* This magic test requires explanation.
1576 *
1577 * We know, that some child q is already
1578 * attached to this parent and have choice:
1579 * either to change it or to create/graft new one.
1580 *
1581 * 1. We are allowed to create/graft only
1582 * if CREATE and REPLACE flags are set.
1583 *
1584 * 2. If EXCL is set, requestor wanted to say,
1585 * that qdisc tcm_handle is not expected
1586 * to exist, so that we choose create/graft too.
1587 *
1588 * 3. The last case is when no flags are set.
1589 * Alas, it is sort of hole in API, we
1590 * cannot decide what to do unambiguously.
1591 * For now we select create/graft, if
1592 * user gave KIND, which does not match existing.
1593 */
1594 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1595 (n->nlmsg_flags & NLM_F_REPLACE) &&
1596 ((n->nlmsg_flags & NLM_F_EXCL) ||
1597 (tca[TCA_KIND] &&
1598 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1599 goto create_n_graft;
1600 }
1601 }
1602 } else {
1603 if (!tcm->tcm_handle) {
1604 NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1605 return -EINVAL;
1606 }
1607 q = qdisc_lookup(dev, tcm->tcm_handle);
1608 }
1609
1610 /* Change qdisc parameters */
1611 if (!q) {
1612 NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1613 return -ENOENT;
1614 }
1615 if (n->nlmsg_flags & NLM_F_EXCL) {
1616 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1617 return -EEXIST;
1618 }
1619 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1620 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1621 return -EINVAL;
1622 }
1623 err = qdisc_change(q, tca, extack);
1624 if (err == 0)
1625 qdisc_notify(net, skb, n, clid, NULL, q);
1626 return err;
1627
1628create_n_graft:
1629 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1630 NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1631 return -ENOENT;
1632 }
1633 if (clid == TC_H_INGRESS) {
1634 if (dev_ingress_queue(dev)) {
1635 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1636 tcm->tcm_parent, tcm->tcm_parent,
1637 tca, &err, extack);
1638 } else {
1639 NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1640 err = -ENOENT;
1641 }
1642 } else {
1643 struct netdev_queue *dev_queue;
1644
1645 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1646 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1647 else if (p)
1648 dev_queue = p->dev_queue;
1649 else
1650 dev_queue = netdev_get_tx_queue(dev, 0);
1651
1652 q = qdisc_create(dev, dev_queue, p,
1653 tcm->tcm_parent, tcm->tcm_handle,
1654 tca, &err, extack);
1655 }
1656 if (q == NULL) {
1657 if (err == -EAGAIN)
1658 goto replay;
1659 return err;
1660 }
1661
1662graft:
1663 err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1664 if (err) {
1665 if (q)
1666 qdisc_put(q);
1667 return err;
1668 }
1669
1670 return 0;
1671}
1672
1673static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1674 struct netlink_callback *cb,
1675 int *q_idx_p, int s_q_idx, bool recur,
1676 bool dump_invisible)
1677{
1678 int ret = 0, q_idx = *q_idx_p;
1679 struct Qdisc *q;
1680 int b;
1681
1682 if (!root)
1683 return 0;
1684
1685 q = root;
1686 if (q_idx < s_q_idx) {
1687 q_idx++;
1688 } else {
1689 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1690 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1691 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1692 RTM_NEWQDISC) <= 0)
1693 goto done;
1694 q_idx++;
1695 }
1696
1697 /* If dumping singletons, there is no qdisc_dev(root) and the singleton
1698 * itself has already been dumped.
1699 *
1700 * If we've already dumped the top-level (ingress) qdisc above and the global
1701 * qdisc hashtable, we don't want to hit it again
1702 */
1703 if (!qdisc_dev(root) || !recur)
1704 goto out;
1705
1706 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1707 if (q_idx < s_q_idx) {
1708 q_idx++;
1709 continue;
1710 }
1711 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1712 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1713 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1714 RTM_NEWQDISC) <= 0)
1715 goto done;
1716 q_idx++;
1717 }
1718
1719out:
1720 *q_idx_p = q_idx;
1721 return ret;
1722done:
1723 ret = -1;
1724 goto out;
1725}
1726
1727static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1728{
1729 struct net *net = sock_net(skb->sk);
1730 int idx, q_idx;
1731 int s_idx, s_q_idx;
1732 struct net_device *dev;
1733 const struct nlmsghdr *nlh = cb->nlh;
1734 struct nlattr *tca[TCA_MAX + 1];
1735 int err;
1736
1737 s_idx = cb->args[0];
1738 s_q_idx = q_idx = cb->args[1];
1739
1740 idx = 0;
1741 ASSERT_RTNL();
1742
1743 err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1744 rtm_tca_policy, cb->extack);
1745 if (err < 0)
1746 return err;
1747
1748 for_each_netdev(net, dev) {
1749 struct netdev_queue *dev_queue;
1750
1751 if (idx < s_idx)
1752 goto cont;
1753 if (idx > s_idx)
1754 s_q_idx = 0;
1755 q_idx = 0;
1756
1757 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1758 true, tca[TCA_DUMP_INVISIBLE]) < 0)
1759 goto done;
1760
1761 dev_queue = dev_ingress_queue(dev);
1762 if (dev_queue &&
1763 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1764 &q_idx, s_q_idx, false,
1765 tca[TCA_DUMP_INVISIBLE]) < 0)
1766 goto done;
1767
1768cont:
1769 idx++;
1770 }
1771
1772done:
1773 cb->args[0] = idx;
1774 cb->args[1] = q_idx;
1775
1776 return skb->len;
1777}
1778
1779
1780
1781/************************************************
1782 * Traffic classes manipulation. *
1783 ************************************************/
1784
1785static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1786 unsigned long cl,
1787 u32 portid, u32 seq, u16 flags, int event)
1788{
1789 struct tcmsg *tcm;
1790 struct nlmsghdr *nlh;
1791 unsigned char *b = skb_tail_pointer(skb);
1792 struct gnet_dump d;
1793 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1794
1795 cond_resched();
1796 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1797 if (!nlh)
1798 goto out_nlmsg_trim;
1799 tcm = nlmsg_data(nlh);
1800 tcm->tcm_family = AF_UNSPEC;
1801 tcm->tcm__pad1 = 0;
1802 tcm->tcm__pad2 = 0;
1803 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1804 tcm->tcm_parent = q->handle;
1805 tcm->tcm_handle = q->handle;
1806 tcm->tcm_info = 0;
1807 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1808 goto nla_put_failure;
1809 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1810 goto nla_put_failure;
1811
1812 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1813 NULL, &d, TCA_PAD) < 0)
1814 goto nla_put_failure;
1815
1816 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1817 goto nla_put_failure;
1818
1819 if (gnet_stats_finish_copy(&d) < 0)
1820 goto nla_put_failure;
1821
1822 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1823 return skb->len;
1824
1825out_nlmsg_trim:
1826nla_put_failure:
1827 nlmsg_trim(skb, b);
1828 return -1;
1829}
1830
1831static int tclass_notify(struct net *net, struct sk_buff *oskb,
1832 struct nlmsghdr *n, struct Qdisc *q,
1833 unsigned long cl, int event)
1834{
1835 struct sk_buff *skb;
1836 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1837 int err = 0;
1838
1839 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1840 if (!skb)
1841 return -ENOBUFS;
1842
1843 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1844 kfree_skb(skb);
1845 return -EINVAL;
1846 }
1847
1848 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1849 n->nlmsg_flags & NLM_F_ECHO);
1850 if (err > 0)
1851 err = 0;
1852 return err;
1853}
1854
1855static int tclass_del_notify(struct net *net,
1856 const struct Qdisc_class_ops *cops,
1857 struct sk_buff *oskb, struct nlmsghdr *n,
1858 struct Qdisc *q, unsigned long cl)
1859{
1860 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1861 struct sk_buff *skb;
1862 int err = 0;
1863
1864 if (!cops->delete)
1865 return -EOPNOTSUPP;
1866
1867 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1868 if (!skb)
1869 return -ENOBUFS;
1870
1871 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1872 RTM_DELTCLASS) < 0) {
1873 kfree_skb(skb);
1874 return -EINVAL;
1875 }
1876
1877 err = cops->delete(q, cl);
1878 if (err) {
1879 kfree_skb(skb);
1880 return err;
1881 }
1882
1883 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1884 n->nlmsg_flags & NLM_F_ECHO);
1885 if (err > 0)
1886 err = 0;
1887 return err;
1888}
1889
1890#ifdef CONFIG_NET_CLS
1891
1892struct tcf_bind_args {
1893 struct tcf_walker w;
1894 u32 classid;
1895 unsigned long cl;
1896};
1897
1898static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1899{
1900 struct tcf_bind_args *a = (void *)arg;
1901
1902 if (tp->ops->bind_class) {
1903 struct Qdisc *q = tcf_block_q(tp->chain->block);
1904
1905 sch_tree_lock(q);
1906 tp->ops->bind_class(n, a->classid, a->cl);
1907 sch_tree_unlock(q);
1908 }
1909 return 0;
1910}
1911
1912static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1913 unsigned long new_cl)
1914{
1915 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1916 struct tcf_block *block;
1917 struct tcf_chain *chain;
1918 unsigned long cl;
1919
1920 cl = cops->find(q, portid);
1921 if (!cl)
1922 return;
1923 if (!cops->tcf_block)
1924 return;
1925 block = cops->tcf_block(q, cl, NULL);
1926 if (!block)
1927 return;
1928 for (chain = tcf_get_next_chain(block, NULL);
1929 chain;
1930 chain = tcf_get_next_chain(block, chain)) {
1931 struct tcf_proto *tp;
1932
1933 for (tp = tcf_get_next_proto(chain, NULL, true);
1934 tp; tp = tcf_get_next_proto(chain, tp, true)) {
1935 struct tcf_bind_args arg = {};
1936
1937 arg.w.fn = tcf_node_bind;
1938 arg.classid = clid;
1939 arg.cl = new_cl;
1940 tp->ops->walk(tp, &arg.w, true);
1941 }
1942 }
1943}
1944
1945#else
1946
1947static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1948 unsigned long new_cl)
1949{
1950}
1951
1952#endif
1953
1954static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1955 struct netlink_ext_ack *extack)
1956{
1957 struct net *net = sock_net(skb->sk);
1958 struct tcmsg *tcm = nlmsg_data(n);
1959 struct nlattr *tca[TCA_MAX + 1];
1960 struct net_device *dev;
1961 struct Qdisc *q = NULL;
1962 const struct Qdisc_class_ops *cops;
1963 unsigned long cl = 0;
1964 unsigned long new_cl;
1965 u32 portid;
1966 u32 clid;
1967 u32 qid;
1968 int err;
1969
1970 if ((n->nlmsg_type != RTM_GETTCLASS) &&
1971 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1972 return -EPERM;
1973
1974 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1975 rtm_tca_policy, extack);
1976 if (err < 0)
1977 return err;
1978
1979 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1980 if (!dev)
1981 return -ENODEV;
1982
1983 /*
1984 parent == TC_H_UNSPEC - unspecified parent.
1985 parent == TC_H_ROOT - class is root, which has no parent.
1986 parent == X:0 - parent is root class.
1987 parent == X:Y - parent is a node in hierarchy.
1988 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1989
1990 handle == 0:0 - generate handle from kernel pool.
1991 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1992 handle == X:Y - clear.
1993 handle == X:0 - root class.
1994 */
1995
1996 /* Step 1. Determine qdisc handle X:0 */
1997
1998 portid = tcm->tcm_parent;
1999 clid = tcm->tcm_handle;
2000 qid = TC_H_MAJ(clid);
2001
2002 if (portid != TC_H_ROOT) {
2003 u32 qid1 = TC_H_MAJ(portid);
2004
2005 if (qid && qid1) {
2006 /* If both majors are known, they must be identical. */
2007 if (qid != qid1)
2008 return -EINVAL;
2009 } else if (qid1) {
2010 qid = qid1;
2011 } else if (qid == 0)
2012 qid = dev->qdisc->handle;
2013
2014 /* Now qid is genuine qdisc handle consistent
2015 * both with parent and child.
2016 *
2017 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2018 */
2019 if (portid)
2020 portid = TC_H_MAKE(qid, portid);
2021 } else {
2022 if (qid == 0)
2023 qid = dev->qdisc->handle;
2024 }
2025
2026 /* OK. Locate qdisc */
2027 q = qdisc_lookup(dev, qid);
2028 if (!q)
2029 return -ENOENT;
2030
2031 /* An check that it supports classes */
2032 cops = q->ops->cl_ops;
2033 if (cops == NULL)
2034 return -EINVAL;
2035
2036 /* Now try to get class */
2037 if (clid == 0) {
2038 if (portid == TC_H_ROOT)
2039 clid = qid;
2040 } else
2041 clid = TC_H_MAKE(qid, clid);
2042
2043 if (clid)
2044 cl = cops->find(q, clid);
2045
2046 if (cl == 0) {
2047 err = -ENOENT;
2048 if (n->nlmsg_type != RTM_NEWTCLASS ||
2049 !(n->nlmsg_flags & NLM_F_CREATE))
2050 goto out;
2051 } else {
2052 switch (n->nlmsg_type) {
2053 case RTM_NEWTCLASS:
2054 err = -EEXIST;
2055 if (n->nlmsg_flags & NLM_F_EXCL)
2056 goto out;
2057 break;
2058 case RTM_DELTCLASS:
2059 err = tclass_del_notify(net, cops, skb, n, q, cl);
2060 /* Unbind the class with flilters with 0 */
2061 tc_bind_tclass(q, portid, clid, 0);
2062 goto out;
2063 case RTM_GETTCLASS:
2064 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
2065 goto out;
2066 default:
2067 err = -EINVAL;
2068 goto out;
2069 }
2070 }
2071
2072 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2073 NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2074 return -EOPNOTSUPP;
2075 }
2076
2077 new_cl = cl;
2078 err = -EOPNOTSUPP;
2079 if (cops->change)
2080 err = cops->change(q, clid, portid, tca, &new_cl, extack);
2081 if (err == 0) {
2082 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
2083 /* We just create a new class, need to do reverse binding. */
2084 if (cl != new_cl)
2085 tc_bind_tclass(q, portid, clid, new_cl);
2086 }
2087out:
2088 return err;
2089}
2090
2091struct qdisc_dump_args {
2092 struct qdisc_walker w;
2093 struct sk_buff *skb;
2094 struct netlink_callback *cb;
2095};
2096
2097static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2098 struct qdisc_walker *arg)
2099{
2100 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2101
2102 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2103 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2104 RTM_NEWTCLASS);
2105}
2106
2107static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2108 struct tcmsg *tcm, struct netlink_callback *cb,
2109 int *t_p, int s_t)
2110{
2111 struct qdisc_dump_args arg;
2112
2113 if (tc_qdisc_dump_ignore(q, false) ||
2114 *t_p < s_t || !q->ops->cl_ops ||
2115 (tcm->tcm_parent &&
2116 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2117 (*t_p)++;
2118 return 0;
2119 }
2120 if (*t_p > s_t)
2121 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2122 arg.w.fn = qdisc_class_dump;
2123 arg.skb = skb;
2124 arg.cb = cb;
2125 arg.w.stop = 0;
2126 arg.w.skip = cb->args[1];
2127 arg.w.count = 0;
2128 q->ops->cl_ops->walk(q, &arg.w);
2129 cb->args[1] = arg.w.count;
2130 if (arg.w.stop)
2131 return -1;
2132 (*t_p)++;
2133 return 0;
2134}
2135
2136static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2137 struct tcmsg *tcm, struct netlink_callback *cb,
2138 int *t_p, int s_t)
2139{
2140 struct Qdisc *q;
2141 int b;
2142
2143 if (!root)
2144 return 0;
2145
2146 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2147 return -1;
2148
2149 if (!qdisc_dev(root))
2150 return 0;
2151
2152 if (tcm->tcm_parent) {
2153 q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2154 if (q && q != root &&
2155 tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2156 return -1;
2157 return 0;
2158 }
2159 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2160 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2161 return -1;
2162 }
2163
2164 return 0;
2165}
2166
2167static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2168{
2169 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2170 struct net *net = sock_net(skb->sk);
2171 struct netdev_queue *dev_queue;
2172 struct net_device *dev;
2173 int t, s_t;
2174
2175 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2176 return 0;
2177 dev = dev_get_by_index(net, tcm->tcm_ifindex);
2178 if (!dev)
2179 return 0;
2180
2181 s_t = cb->args[0];
2182 t = 0;
2183
2184 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
2185 goto done;
2186
2187 dev_queue = dev_ingress_queue(dev);
2188 if (dev_queue &&
2189 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2190 &t, s_t) < 0)
2191 goto done;
2192
2193done:
2194 cb->args[0] = t;
2195
2196 dev_put(dev);
2197 return skb->len;
2198}
2199
2200#ifdef CONFIG_PROC_FS
2201static int psched_show(struct seq_file *seq, void *v)
2202{
2203 seq_printf(seq, "%08x %08x %08x %08x\n",
2204 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2205 1000000,
2206 (u32)NSEC_PER_SEC / hrtimer_resolution);
2207
2208 return 0;
2209}
2210
2211static int __net_init psched_net_init(struct net *net)
2212{
2213 struct proc_dir_entry *e;
2214
2215 e = proc_create_single("psched", 0, net->proc_net, psched_show);
2216 if (e == NULL)
2217 return -ENOMEM;
2218
2219 return 0;
2220}
2221
2222static void __net_exit psched_net_exit(struct net *net)
2223{
2224 remove_proc_entry("psched", net->proc_net);
2225}
2226#else
2227static int __net_init psched_net_init(struct net *net)
2228{
2229 return 0;
2230}
2231
2232static void __net_exit psched_net_exit(struct net *net)
2233{
2234}
2235#endif
2236
2237static struct pernet_operations psched_net_ops = {
2238 .init = psched_net_init,
2239 .exit = psched_net_exit,
2240};
2241
2242static int __init pktsched_init(void)
2243{
2244 int err;
2245
2246 err = register_pernet_subsys(&psched_net_ops);
2247 if (err) {
2248 pr_err("pktsched_init: "
2249 "cannot initialize per netns operations\n");
2250 return err;
2251 }
2252
2253 register_qdisc(&pfifo_fast_ops);
2254 register_qdisc(&pfifo_qdisc_ops);
2255 register_qdisc(&bfifo_qdisc_ops);
2256 register_qdisc(&pfifo_head_drop_qdisc_ops);
2257 register_qdisc(&mq_qdisc_ops);
2258 register_qdisc(&noqueue_qdisc_ops);
2259
2260 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2261 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2262 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2263 0);
2264 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2265 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2266 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2267 0);
2268
2269 return 0;
2270}
2271
2272subsys_initcall(pktsched_init);
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
38static int qdisc_notify(struct net *net, struct sk_buff *oskb,
39 struct nlmsghdr *n, u32 clid,
40 struct Qdisc *old, struct Qdisc *new);
41static 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. */
126static DEFINE_RWLOCK(qdisc_mod_lock);
127
128
129/************************************************
130 * Queueing disciplines manipulation. *
131 ************************************************/
132
133
134/* The list of all installed queueing disciplines. */
135
136static struct Qdisc_ops *qdisc_base;
137
138/* Register/uregister queueing discipline */
139
140int 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;
174out:
175 write_unlock(&qdisc_mod_lock);
176 return rc;
177
178out_einval:
179 rc = -EINVAL;
180 goto out;
181}
182EXPORT_SYMBOL(register_qdisc);
183
184int 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}
201EXPORT_SYMBOL(unregister_qdisc);
202
203/* We know handle. Find qdisc among all qdisc's attached to device
204 (root qdisc, all its children, children of children etc.)
205 */
206
207static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
208{
209 struct Qdisc *q;
210
211 if (!(root->flags & TCQ_F_BUILTIN) &&
212 root->handle == handle)
213 return root;
214
215 list_for_each_entry(q, &root->list, list) {
216 if (q->handle == handle)
217 return q;
218 }
219 return NULL;
220}
221
222static void qdisc_list_add(struct Qdisc *q)
223{
224 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
225 list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list);
226}
227
228void qdisc_list_del(struct Qdisc *q)
229{
230 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
231 list_del(&q->list);
232}
233EXPORT_SYMBOL(qdisc_list_del);
234
235struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
236{
237 struct Qdisc *q;
238
239 q = qdisc_match_from_root(dev->qdisc, handle);
240 if (q)
241 goto out;
242
243 if (dev_ingress_queue(dev))
244 q = qdisc_match_from_root(
245 dev_ingress_queue(dev)->qdisc_sleeping,
246 handle);
247out:
248 return q;
249}
250
251static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
252{
253 unsigned long cl;
254 struct Qdisc *leaf;
255 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
256
257 if (cops == NULL)
258 return NULL;
259 cl = cops->get(p, classid);
260
261 if (cl == 0)
262 return NULL;
263 leaf = cops->leaf(p, cl);
264 cops->put(p, cl);
265 return leaf;
266}
267
268/* Find queueing discipline by name */
269
270static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
271{
272 struct Qdisc_ops *q = NULL;
273
274 if (kind) {
275 read_lock(&qdisc_mod_lock);
276 for (q = qdisc_base; q; q = q->next) {
277 if (nla_strcmp(kind, q->id) == 0) {
278 if (!try_module_get(q->owner))
279 q = NULL;
280 break;
281 }
282 }
283 read_unlock(&qdisc_mod_lock);
284 }
285 return q;
286}
287
288static struct qdisc_rate_table *qdisc_rtab_list;
289
290struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
291{
292 struct qdisc_rate_table *rtab;
293
294 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
295 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
296 rtab->refcnt++;
297 return rtab;
298 }
299 }
300
301 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
302 nla_len(tab) != TC_RTAB_SIZE)
303 return NULL;
304
305 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
306 if (rtab) {
307 rtab->rate = *r;
308 rtab->refcnt = 1;
309 memcpy(rtab->data, nla_data(tab), 1024);
310 rtab->next = qdisc_rtab_list;
311 qdisc_rtab_list = rtab;
312 }
313 return rtab;
314}
315EXPORT_SYMBOL(qdisc_get_rtab);
316
317void qdisc_put_rtab(struct qdisc_rate_table *tab)
318{
319 struct qdisc_rate_table *rtab, **rtabp;
320
321 if (!tab || --tab->refcnt)
322 return;
323
324 for (rtabp = &qdisc_rtab_list;
325 (rtab = *rtabp) != NULL;
326 rtabp = &rtab->next) {
327 if (rtab == tab) {
328 *rtabp = rtab->next;
329 kfree(rtab);
330 return;
331 }
332 }
333}
334EXPORT_SYMBOL(qdisc_put_rtab);
335
336static LIST_HEAD(qdisc_stab_list);
337static DEFINE_SPINLOCK(qdisc_stab_lock);
338
339static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
340 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
341 [TCA_STAB_DATA] = { .type = NLA_BINARY },
342};
343
344static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
345{
346 struct nlattr *tb[TCA_STAB_MAX + 1];
347 struct qdisc_size_table *stab;
348 struct tc_sizespec *s;
349 unsigned int tsize = 0;
350 u16 *tab = NULL;
351 int err;
352
353 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
354 if (err < 0)
355 return ERR_PTR(err);
356 if (!tb[TCA_STAB_BASE])
357 return ERR_PTR(-EINVAL);
358
359 s = nla_data(tb[TCA_STAB_BASE]);
360
361 if (s->tsize > 0) {
362 if (!tb[TCA_STAB_DATA])
363 return ERR_PTR(-EINVAL);
364 tab = nla_data(tb[TCA_STAB_DATA]);
365 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
366 }
367
368 if (tsize != s->tsize || (!tab && tsize > 0))
369 return ERR_PTR(-EINVAL);
370
371 spin_lock(&qdisc_stab_lock);
372
373 list_for_each_entry(stab, &qdisc_stab_list, list) {
374 if (memcmp(&stab->szopts, s, sizeof(*s)))
375 continue;
376 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
377 continue;
378 stab->refcnt++;
379 spin_unlock(&qdisc_stab_lock);
380 return stab;
381 }
382
383 spin_unlock(&qdisc_stab_lock);
384
385 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
386 if (!stab)
387 return ERR_PTR(-ENOMEM);
388
389 stab->refcnt = 1;
390 stab->szopts = *s;
391 if (tsize > 0)
392 memcpy(stab->data, tab, tsize * sizeof(u16));
393
394 spin_lock(&qdisc_stab_lock);
395 list_add_tail(&stab->list, &qdisc_stab_list);
396 spin_unlock(&qdisc_stab_lock);
397
398 return stab;
399}
400
401static void stab_kfree_rcu(struct rcu_head *head)
402{
403 kfree(container_of(head, struct qdisc_size_table, rcu));
404}
405
406void qdisc_put_stab(struct qdisc_size_table *tab)
407{
408 if (!tab)
409 return;
410
411 spin_lock(&qdisc_stab_lock);
412
413 if (--tab->refcnt == 0) {
414 list_del(&tab->list);
415 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
416 }
417
418 spin_unlock(&qdisc_stab_lock);
419}
420EXPORT_SYMBOL(qdisc_put_stab);
421
422static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
423{
424 struct nlattr *nest;
425
426 nest = nla_nest_start(skb, TCA_STAB);
427 if (nest == NULL)
428 goto nla_put_failure;
429 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
430 nla_nest_end(skb, nest);
431
432 return skb->len;
433
434nla_put_failure:
435 return -1;
436}
437
438void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
439{
440 int pkt_len, slot;
441
442 pkt_len = skb->len + stab->szopts.overhead;
443 if (unlikely(!stab->szopts.tsize))
444 goto out;
445
446 slot = pkt_len + stab->szopts.cell_align;
447 if (unlikely(slot < 0))
448 slot = 0;
449
450 slot >>= stab->szopts.cell_log;
451 if (likely(slot < stab->szopts.tsize))
452 pkt_len = stab->data[slot];
453 else
454 pkt_len = stab->data[stab->szopts.tsize - 1] *
455 (slot / stab->szopts.tsize) +
456 stab->data[slot % stab->szopts.tsize];
457
458 pkt_len <<= stab->szopts.size_log;
459out:
460 if (unlikely(pkt_len < 1))
461 pkt_len = 1;
462 qdisc_skb_cb(skb)->pkt_len = pkt_len;
463}
464EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
465
466void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
467{
468 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
469 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
470 txt, qdisc->ops->id, qdisc->handle >> 16);
471 qdisc->flags |= TCQ_F_WARN_NONWC;
472 }
473}
474EXPORT_SYMBOL(qdisc_warn_nonwc);
475
476static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
477{
478 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
479 timer);
480
481 qdisc_unthrottled(wd->qdisc);
482 __netif_schedule(qdisc_root(wd->qdisc));
483
484 return HRTIMER_NORESTART;
485}
486
487void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
488{
489 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
490 wd->timer.function = qdisc_watchdog;
491 wd->qdisc = qdisc;
492}
493EXPORT_SYMBOL(qdisc_watchdog_init);
494
495void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
496{
497 ktime_t time;
498
499 if (test_bit(__QDISC_STATE_DEACTIVATED,
500 &qdisc_root_sleeping(wd->qdisc)->state))
501 return;
502
503 qdisc_throttled(wd->qdisc);
504 time = ktime_set(0, 0);
505 time = ktime_add_ns(time, PSCHED_TICKS2NS(expires));
506 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
507}
508EXPORT_SYMBOL(qdisc_watchdog_schedule);
509
510void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
511{
512 hrtimer_cancel(&wd->timer);
513 qdisc_unthrottled(wd->qdisc);
514}
515EXPORT_SYMBOL(qdisc_watchdog_cancel);
516
517static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
518{
519 unsigned int size = n * sizeof(struct hlist_head), i;
520 struct hlist_head *h;
521
522 if (size <= PAGE_SIZE)
523 h = kmalloc(size, GFP_KERNEL);
524 else
525 h = (struct hlist_head *)
526 __get_free_pages(GFP_KERNEL, get_order(size));
527
528 if (h != NULL) {
529 for (i = 0; i < n; i++)
530 INIT_HLIST_HEAD(&h[i]);
531 }
532 return h;
533}
534
535static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
536{
537 unsigned int size = n * sizeof(struct hlist_head);
538
539 if (size <= PAGE_SIZE)
540 kfree(h);
541 else
542 free_pages((unsigned long)h, get_order(size));
543}
544
545void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
546{
547 struct Qdisc_class_common *cl;
548 struct hlist_node *n, *next;
549 struct hlist_head *nhash, *ohash;
550 unsigned int nsize, nmask, osize;
551 unsigned int i, h;
552
553 /* Rehash when load factor exceeds 0.75 */
554 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
555 return;
556 nsize = clhash->hashsize * 2;
557 nmask = nsize - 1;
558 nhash = qdisc_class_hash_alloc(nsize);
559 if (nhash == NULL)
560 return;
561
562 ohash = clhash->hash;
563 osize = clhash->hashsize;
564
565 sch_tree_lock(sch);
566 for (i = 0; i < osize; i++) {
567 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
568 h = qdisc_class_hash(cl->classid, nmask);
569 hlist_add_head(&cl->hnode, &nhash[h]);
570 }
571 }
572 clhash->hash = nhash;
573 clhash->hashsize = nsize;
574 clhash->hashmask = nmask;
575 sch_tree_unlock(sch);
576
577 qdisc_class_hash_free(ohash, osize);
578}
579EXPORT_SYMBOL(qdisc_class_hash_grow);
580
581int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
582{
583 unsigned int size = 4;
584
585 clhash->hash = qdisc_class_hash_alloc(size);
586 if (clhash->hash == NULL)
587 return -ENOMEM;
588 clhash->hashsize = size;
589 clhash->hashmask = size - 1;
590 clhash->hashelems = 0;
591 return 0;
592}
593EXPORT_SYMBOL(qdisc_class_hash_init);
594
595void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
596{
597 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
598}
599EXPORT_SYMBOL(qdisc_class_hash_destroy);
600
601void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
602 struct Qdisc_class_common *cl)
603{
604 unsigned int h;
605
606 INIT_HLIST_NODE(&cl->hnode);
607 h = qdisc_class_hash(cl->classid, clhash->hashmask);
608 hlist_add_head(&cl->hnode, &clhash->hash[h]);
609 clhash->hashelems++;
610}
611EXPORT_SYMBOL(qdisc_class_hash_insert);
612
613void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
614 struct Qdisc_class_common *cl)
615{
616 hlist_del(&cl->hnode);
617 clhash->hashelems--;
618}
619EXPORT_SYMBOL(qdisc_class_hash_remove);
620
621/* Allocate an unique handle from space managed by kernel */
622
623static u32 qdisc_alloc_handle(struct net_device *dev)
624{
625 int i = 0x10000;
626 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
627
628 do {
629 autohandle += TC_H_MAKE(0x10000U, 0);
630 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
631 autohandle = TC_H_MAKE(0x80000000U, 0);
632 } while (qdisc_lookup(dev, autohandle) && --i > 0);
633
634 return i > 0 ? autohandle : 0;
635}
636
637void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
638{
639 const struct Qdisc_class_ops *cops;
640 unsigned long cl;
641 u32 parentid;
642
643 if (n == 0)
644 return;
645 while ((parentid = sch->parent)) {
646 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
647 return;
648
649 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
650 if (sch == NULL) {
651 WARN_ON(parentid != TC_H_ROOT);
652 return;
653 }
654 cops = sch->ops->cl_ops;
655 if (cops->qlen_notify) {
656 cl = cops->get(sch, parentid);
657 cops->qlen_notify(sch, cl);
658 cops->put(sch, cl);
659 }
660 sch->q.qlen -= n;
661 }
662}
663EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
664
665static void notify_and_destroy(struct net *net, struct sk_buff *skb,
666 struct nlmsghdr *n, u32 clid,
667 struct Qdisc *old, struct Qdisc *new)
668{
669 if (new || old)
670 qdisc_notify(net, skb, n, clid, old, new);
671
672 if (old)
673 qdisc_destroy(old);
674}
675
676/* Graft qdisc "new" to class "classid" of qdisc "parent" or
677 * to device "dev".
678 *
679 * When appropriate send a netlink notification using 'skb'
680 * and "n".
681 *
682 * On success, destroy old qdisc.
683 */
684
685static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
686 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
687 struct Qdisc *new, struct Qdisc *old)
688{
689 struct Qdisc *q = old;
690 struct net *net = dev_net(dev);
691 int err = 0;
692
693 if (parent == NULL) {
694 unsigned int i, num_q, ingress;
695
696 ingress = 0;
697 num_q = dev->num_tx_queues;
698 if ((q && q->flags & TCQ_F_INGRESS) ||
699 (new && new->flags & TCQ_F_INGRESS)) {
700 num_q = 1;
701 ingress = 1;
702 if (!dev_ingress_queue(dev))
703 return -ENOENT;
704 }
705
706 if (dev->flags & IFF_UP)
707 dev_deactivate(dev);
708
709 if (new && new->ops->attach) {
710 new->ops->attach(new);
711 num_q = 0;
712 }
713
714 for (i = 0; i < num_q; i++) {
715 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
716
717 if (!ingress)
718 dev_queue = netdev_get_tx_queue(dev, i);
719
720 old = dev_graft_qdisc(dev_queue, new);
721 if (new && i > 0)
722 atomic_inc(&new->refcnt);
723
724 if (!ingress)
725 qdisc_destroy(old);
726 }
727
728 if (!ingress) {
729 notify_and_destroy(net, skb, n, classid,
730 dev->qdisc, new);
731 if (new && !new->ops->attach)
732 atomic_inc(&new->refcnt);
733 dev->qdisc = new ? : &noop_qdisc;
734 } else {
735 notify_and_destroy(net, skb, n, classid, old, new);
736 }
737
738 if (dev->flags & IFF_UP)
739 dev_activate(dev);
740 } else {
741 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
742
743 err = -EOPNOTSUPP;
744 if (cops && cops->graft) {
745 unsigned long cl = cops->get(parent, classid);
746 if (cl) {
747 err = cops->graft(parent, cl, new, &old);
748 cops->put(parent, cl);
749 } else
750 err = -ENOENT;
751 }
752 if (!err)
753 notify_and_destroy(net, skb, n, classid, old, new);
754 }
755 return err;
756}
757
758/* lockdep annotation is needed for ingress; egress gets it only for name */
759static struct lock_class_key qdisc_tx_lock;
760static struct lock_class_key qdisc_rx_lock;
761
762/*
763 Allocate and initialize new qdisc.
764
765 Parameters are passed via opt.
766 */
767
768static struct Qdisc *
769qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
770 struct Qdisc *p, u32 parent, u32 handle,
771 struct nlattr **tca, int *errp)
772{
773 int err;
774 struct nlattr *kind = tca[TCA_KIND];
775 struct Qdisc *sch;
776 struct Qdisc_ops *ops;
777 struct qdisc_size_table *stab;
778
779 ops = qdisc_lookup_ops(kind);
780#ifdef CONFIG_MODULES
781 if (ops == NULL && kind != NULL) {
782 char name[IFNAMSIZ];
783 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
784 /* We dropped the RTNL semaphore in order to
785 * perform the module load. So, even if we
786 * succeeded in loading the module we have to
787 * tell the caller to replay the request. We
788 * indicate this using -EAGAIN.
789 * We replay the request because the device may
790 * go away in the mean time.
791 */
792 rtnl_unlock();
793 request_module("sch_%s", name);
794 rtnl_lock();
795 ops = qdisc_lookup_ops(kind);
796 if (ops != NULL) {
797 /* We will try again qdisc_lookup_ops,
798 * so don't keep a reference.
799 */
800 module_put(ops->owner);
801 err = -EAGAIN;
802 goto err_out;
803 }
804 }
805 }
806#endif
807
808 err = -ENOENT;
809 if (ops == NULL)
810 goto err_out;
811
812 sch = qdisc_alloc(dev_queue, ops);
813 if (IS_ERR(sch)) {
814 err = PTR_ERR(sch);
815 goto err_out2;
816 }
817
818 sch->parent = parent;
819
820 if (handle == TC_H_INGRESS) {
821 sch->flags |= TCQ_F_INGRESS;
822 handle = TC_H_MAKE(TC_H_INGRESS, 0);
823 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
824 } else {
825 if (handle == 0) {
826 handle = qdisc_alloc_handle(dev);
827 err = -ENOMEM;
828 if (handle == 0)
829 goto err_out3;
830 }
831 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
832 }
833
834 sch->handle = handle;
835
836 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
837 if (tca[TCA_STAB]) {
838 stab = qdisc_get_stab(tca[TCA_STAB]);
839 if (IS_ERR(stab)) {
840 err = PTR_ERR(stab);
841 goto err_out4;
842 }
843 rcu_assign_pointer(sch->stab, stab);
844 }
845 if (tca[TCA_RATE]) {
846 spinlock_t *root_lock;
847
848 err = -EOPNOTSUPP;
849 if (sch->flags & TCQ_F_MQROOT)
850 goto err_out4;
851
852 if ((sch->parent != TC_H_ROOT) &&
853 !(sch->flags & TCQ_F_INGRESS) &&
854 (!p || !(p->flags & TCQ_F_MQROOT)))
855 root_lock = qdisc_root_sleeping_lock(sch);
856 else
857 root_lock = qdisc_lock(sch);
858
859 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
860 root_lock, tca[TCA_RATE]);
861 if (err)
862 goto err_out4;
863 }
864
865 qdisc_list_add(sch);
866
867 return sch;
868 }
869err_out3:
870 dev_put(dev);
871 kfree((char *) sch - sch->padded);
872err_out2:
873 module_put(ops->owner);
874err_out:
875 *errp = err;
876 return NULL;
877
878err_out4:
879 /*
880 * Any broken qdiscs that would require a ops->reset() here?
881 * The qdisc was never in action so it shouldn't be necessary.
882 */
883 qdisc_put_stab(rtnl_dereference(sch->stab));
884 if (ops->destroy)
885 ops->destroy(sch);
886 goto err_out3;
887}
888
889static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
890{
891 struct qdisc_size_table *ostab, *stab = NULL;
892 int err = 0;
893
894 if (tca[TCA_OPTIONS]) {
895 if (sch->ops->change == NULL)
896 return -EINVAL;
897 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
898 if (err)
899 return err;
900 }
901
902 if (tca[TCA_STAB]) {
903 stab = qdisc_get_stab(tca[TCA_STAB]);
904 if (IS_ERR(stab))
905 return PTR_ERR(stab);
906 }
907
908 ostab = rtnl_dereference(sch->stab);
909 rcu_assign_pointer(sch->stab, stab);
910 qdisc_put_stab(ostab);
911
912 if (tca[TCA_RATE]) {
913 /* NB: ignores errors from replace_estimator
914 because change can't be undone. */
915 if (sch->flags & TCQ_F_MQROOT)
916 goto out;
917 gen_replace_estimator(&sch->bstats, &sch->rate_est,
918 qdisc_root_sleeping_lock(sch),
919 tca[TCA_RATE]);
920 }
921out:
922 return 0;
923}
924
925struct check_loop_arg {
926 struct qdisc_walker w;
927 struct Qdisc *p;
928 int depth;
929};
930
931static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
932
933static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
934{
935 struct check_loop_arg arg;
936
937 if (q->ops->cl_ops == NULL)
938 return 0;
939
940 arg.w.stop = arg.w.skip = arg.w.count = 0;
941 arg.w.fn = check_loop_fn;
942 arg.depth = depth;
943 arg.p = p;
944 q->ops->cl_ops->walk(q, &arg.w);
945 return arg.w.stop ? -ELOOP : 0;
946}
947
948static int
949check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
950{
951 struct Qdisc *leaf;
952 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
953 struct check_loop_arg *arg = (struct check_loop_arg *)w;
954
955 leaf = cops->leaf(q, cl);
956 if (leaf) {
957 if (leaf == arg->p || arg->depth > 7)
958 return -ELOOP;
959 return check_loop(leaf, arg->p, arg->depth + 1);
960 }
961 return 0;
962}
963
964/*
965 * Delete/get qdisc.
966 */
967
968static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
969{
970 struct net *net = sock_net(skb->sk);
971 struct tcmsg *tcm = NLMSG_DATA(n);
972 struct nlattr *tca[TCA_MAX + 1];
973 struct net_device *dev;
974 u32 clid = tcm->tcm_parent;
975 struct Qdisc *q = NULL;
976 struct Qdisc *p = NULL;
977 int err;
978
979 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
980 if (!dev)
981 return -ENODEV;
982
983 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
984 if (err < 0)
985 return err;
986
987 if (clid) {
988 if (clid != TC_H_ROOT) {
989 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
990 p = qdisc_lookup(dev, TC_H_MAJ(clid));
991 if (!p)
992 return -ENOENT;
993 q = qdisc_leaf(p, clid);
994 } else if (dev_ingress_queue(dev)) {
995 q = dev_ingress_queue(dev)->qdisc_sleeping;
996 }
997 } else {
998 q = dev->qdisc;
999 }
1000 if (!q)
1001 return -ENOENT;
1002
1003 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1004 return -EINVAL;
1005 } else {
1006 q = qdisc_lookup(dev, tcm->tcm_handle);
1007 if (!q)
1008 return -ENOENT;
1009 }
1010
1011 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1012 return -EINVAL;
1013
1014 if (n->nlmsg_type == RTM_DELQDISC) {
1015 if (!clid)
1016 return -EINVAL;
1017 if (q->handle == 0)
1018 return -ENOENT;
1019 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1020 if (err != 0)
1021 return err;
1022 } else {
1023 qdisc_notify(net, skb, n, clid, NULL, q);
1024 }
1025 return 0;
1026}
1027
1028/*
1029 * Create/change qdisc.
1030 */
1031
1032static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1033{
1034 struct net *net = sock_net(skb->sk);
1035 struct tcmsg *tcm;
1036 struct nlattr *tca[TCA_MAX + 1];
1037 struct net_device *dev;
1038 u32 clid;
1039 struct Qdisc *q, *p;
1040 int err;
1041
1042replay:
1043 /* Reinit, just in case something touches this. */
1044 tcm = NLMSG_DATA(n);
1045 clid = tcm->tcm_parent;
1046 q = p = NULL;
1047
1048 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1049 if (!dev)
1050 return -ENODEV;
1051
1052 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1053 if (err < 0)
1054 return err;
1055
1056 if (clid) {
1057 if (clid != TC_H_ROOT) {
1058 if (clid != TC_H_INGRESS) {
1059 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1060 if (!p)
1061 return -ENOENT;
1062 q = qdisc_leaf(p, clid);
1063 } else if (dev_ingress_queue_create(dev)) {
1064 q = dev_ingress_queue(dev)->qdisc_sleeping;
1065 }
1066 } else {
1067 q = dev->qdisc;
1068 }
1069
1070 /* It may be default qdisc, ignore it */
1071 if (q && q->handle == 0)
1072 q = NULL;
1073
1074 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1075 if (tcm->tcm_handle) {
1076 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1077 return -EEXIST;
1078 if (TC_H_MIN(tcm->tcm_handle))
1079 return -EINVAL;
1080 q = qdisc_lookup(dev, tcm->tcm_handle);
1081 if (!q)
1082 goto create_n_graft;
1083 if (n->nlmsg_flags & NLM_F_EXCL)
1084 return -EEXIST;
1085 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1086 return -EINVAL;
1087 if (q == p ||
1088 (p && check_loop(q, p, 0)))
1089 return -ELOOP;
1090 atomic_inc(&q->refcnt);
1091 goto graft;
1092 } else {
1093 if (!q)
1094 goto create_n_graft;
1095
1096 /* This magic test requires explanation.
1097 *
1098 * We know, that some child q is already
1099 * attached to this parent and have choice:
1100 * either to change it or to create/graft new one.
1101 *
1102 * 1. We are allowed to create/graft only
1103 * if CREATE and REPLACE flags are set.
1104 *
1105 * 2. If EXCL is set, requestor wanted to say,
1106 * that qdisc tcm_handle is not expected
1107 * to exist, so that we choose create/graft too.
1108 *
1109 * 3. The last case is when no flags are set.
1110 * Alas, it is sort of hole in API, we
1111 * cannot decide what to do unambiguously.
1112 * For now we select create/graft, if
1113 * user gave KIND, which does not match existing.
1114 */
1115 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1116 (n->nlmsg_flags & NLM_F_REPLACE) &&
1117 ((n->nlmsg_flags & NLM_F_EXCL) ||
1118 (tca[TCA_KIND] &&
1119 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1120 goto create_n_graft;
1121 }
1122 }
1123 } else {
1124 if (!tcm->tcm_handle)
1125 return -EINVAL;
1126 q = qdisc_lookup(dev, tcm->tcm_handle);
1127 }
1128
1129 /* Change qdisc parameters */
1130 if (q == NULL)
1131 return -ENOENT;
1132 if (n->nlmsg_flags & NLM_F_EXCL)
1133 return -EEXIST;
1134 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1135 return -EINVAL;
1136 err = qdisc_change(q, tca);
1137 if (err == 0)
1138 qdisc_notify(net, skb, n, clid, NULL, q);
1139 return err;
1140
1141create_n_graft:
1142 if (!(n->nlmsg_flags & NLM_F_CREATE))
1143 return -ENOENT;
1144 if (clid == TC_H_INGRESS) {
1145 if (dev_ingress_queue(dev))
1146 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1147 tcm->tcm_parent, tcm->tcm_parent,
1148 tca, &err);
1149 else
1150 err = -ENOENT;
1151 } else {
1152 struct netdev_queue *dev_queue;
1153
1154 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1155 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1156 else if (p)
1157 dev_queue = p->dev_queue;
1158 else
1159 dev_queue = netdev_get_tx_queue(dev, 0);
1160
1161 q = qdisc_create(dev, dev_queue, p,
1162 tcm->tcm_parent, tcm->tcm_handle,
1163 tca, &err);
1164 }
1165 if (q == NULL) {
1166 if (err == -EAGAIN)
1167 goto replay;
1168 return err;
1169 }
1170
1171graft:
1172 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1173 if (err) {
1174 if (q)
1175 qdisc_destroy(q);
1176 return err;
1177 }
1178
1179 return 0;
1180}
1181
1182static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1183 u32 pid, u32 seq, u16 flags, int event)
1184{
1185 struct tcmsg *tcm;
1186 struct nlmsghdr *nlh;
1187 unsigned char *b = skb_tail_pointer(skb);
1188 struct gnet_dump d;
1189 struct qdisc_size_table *stab;
1190
1191 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1192 tcm = NLMSG_DATA(nlh);
1193 tcm->tcm_family = AF_UNSPEC;
1194 tcm->tcm__pad1 = 0;
1195 tcm->tcm__pad2 = 0;
1196 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1197 tcm->tcm_parent = clid;
1198 tcm->tcm_handle = q->handle;
1199 tcm->tcm_info = atomic_read(&q->refcnt);
1200 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1201 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1202 goto nla_put_failure;
1203 q->qstats.qlen = q->q.qlen;
1204
1205 stab = rtnl_dereference(q->stab);
1206 if (stab && qdisc_dump_stab(skb, stab) < 0)
1207 goto nla_put_failure;
1208
1209 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1210 qdisc_root_sleeping_lock(q), &d) < 0)
1211 goto nla_put_failure;
1212
1213 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1214 goto nla_put_failure;
1215
1216 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1217 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1218 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1219 goto nla_put_failure;
1220
1221 if (gnet_stats_finish_copy(&d) < 0)
1222 goto nla_put_failure;
1223
1224 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1225 return skb->len;
1226
1227nlmsg_failure:
1228nla_put_failure:
1229 nlmsg_trim(skb, b);
1230 return -1;
1231}
1232
1233static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1234{
1235 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1236}
1237
1238static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1239 struct nlmsghdr *n, u32 clid,
1240 struct Qdisc *old, struct Qdisc *new)
1241{
1242 struct sk_buff *skb;
1243 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1244
1245 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1246 if (!skb)
1247 return -ENOBUFS;
1248
1249 if (old && !tc_qdisc_dump_ignore(old)) {
1250 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq,
1251 0, RTM_DELQDISC) < 0)
1252 goto err_out;
1253 }
1254 if (new && !tc_qdisc_dump_ignore(new)) {
1255 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq,
1256 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1257 goto err_out;
1258 }
1259
1260 if (skb->len)
1261 return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
1262 n->nlmsg_flags & NLM_F_ECHO);
1263
1264err_out:
1265 kfree_skb(skb);
1266 return -EINVAL;
1267}
1268
1269static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1270 struct netlink_callback *cb,
1271 int *q_idx_p, int s_q_idx)
1272{
1273 int ret = 0, q_idx = *q_idx_p;
1274 struct Qdisc *q;
1275
1276 if (!root)
1277 return 0;
1278
1279 q = root;
1280 if (q_idx < s_q_idx) {
1281 q_idx++;
1282 } else {
1283 if (!tc_qdisc_dump_ignore(q) &&
1284 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1285 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1286 goto done;
1287 q_idx++;
1288 }
1289 list_for_each_entry(q, &root->list, list) {
1290 if (q_idx < s_q_idx) {
1291 q_idx++;
1292 continue;
1293 }
1294 if (!tc_qdisc_dump_ignore(q) &&
1295 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1296 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1297 goto done;
1298 q_idx++;
1299 }
1300
1301out:
1302 *q_idx_p = q_idx;
1303 return ret;
1304done:
1305 ret = -1;
1306 goto out;
1307}
1308
1309static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1310{
1311 struct net *net = sock_net(skb->sk);
1312 int idx, q_idx;
1313 int s_idx, s_q_idx;
1314 struct net_device *dev;
1315
1316 s_idx = cb->args[0];
1317 s_q_idx = q_idx = cb->args[1];
1318
1319 rcu_read_lock();
1320 idx = 0;
1321 for_each_netdev_rcu(net, dev) {
1322 struct netdev_queue *dev_queue;
1323
1324 if (idx < s_idx)
1325 goto cont;
1326 if (idx > s_idx)
1327 s_q_idx = 0;
1328 q_idx = 0;
1329
1330 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1331 goto done;
1332
1333 dev_queue = dev_ingress_queue(dev);
1334 if (dev_queue &&
1335 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1336 &q_idx, s_q_idx) < 0)
1337 goto done;
1338
1339cont:
1340 idx++;
1341 }
1342
1343done:
1344 rcu_read_unlock();
1345
1346 cb->args[0] = idx;
1347 cb->args[1] = q_idx;
1348
1349 return skb->len;
1350}
1351
1352
1353
1354/************************************************
1355 * Traffic classes manipulation. *
1356 ************************************************/
1357
1358
1359
1360static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1361{
1362 struct net *net = sock_net(skb->sk);
1363 struct tcmsg *tcm = NLMSG_DATA(n);
1364 struct nlattr *tca[TCA_MAX + 1];
1365 struct net_device *dev;
1366 struct Qdisc *q = NULL;
1367 const struct Qdisc_class_ops *cops;
1368 unsigned long cl = 0;
1369 unsigned long new_cl;
1370 u32 pid = tcm->tcm_parent;
1371 u32 clid = tcm->tcm_handle;
1372 u32 qid = TC_H_MAJ(clid);
1373 int err;
1374
1375 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1376 if (!dev)
1377 return -ENODEV;
1378
1379 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1380 if (err < 0)
1381 return err;
1382
1383 /*
1384 parent == TC_H_UNSPEC - unspecified parent.
1385 parent == TC_H_ROOT - class is root, which has no parent.
1386 parent == X:0 - parent is root class.
1387 parent == X:Y - parent is a node in hierarchy.
1388 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1389
1390 handle == 0:0 - generate handle from kernel pool.
1391 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1392 handle == X:Y - clear.
1393 handle == X:0 - root class.
1394 */
1395
1396 /* Step 1. Determine qdisc handle X:0 */
1397
1398 if (pid != TC_H_ROOT) {
1399 u32 qid1 = TC_H_MAJ(pid);
1400
1401 if (qid && qid1) {
1402 /* If both majors are known, they must be identical. */
1403 if (qid != qid1)
1404 return -EINVAL;
1405 } else if (qid1) {
1406 qid = qid1;
1407 } else if (qid == 0)
1408 qid = dev->qdisc->handle;
1409
1410 /* Now qid is genuine qdisc handle consistent
1411 * both with parent and child.
1412 *
1413 * TC_H_MAJ(pid) still may be unspecified, complete it now.
1414 */
1415 if (pid)
1416 pid = TC_H_MAKE(qid, pid);
1417 } else {
1418 if (qid == 0)
1419 qid = dev->qdisc->handle;
1420 }
1421
1422 /* OK. Locate qdisc */
1423 q = qdisc_lookup(dev, qid);
1424 if (!q)
1425 return -ENOENT;
1426
1427 /* An check that it supports classes */
1428 cops = q->ops->cl_ops;
1429 if (cops == NULL)
1430 return -EINVAL;
1431
1432 /* Now try to get class */
1433 if (clid == 0) {
1434 if (pid == TC_H_ROOT)
1435 clid = qid;
1436 } else
1437 clid = TC_H_MAKE(qid, clid);
1438
1439 if (clid)
1440 cl = cops->get(q, clid);
1441
1442 if (cl == 0) {
1443 err = -ENOENT;
1444 if (n->nlmsg_type != RTM_NEWTCLASS ||
1445 !(n->nlmsg_flags & NLM_F_CREATE))
1446 goto out;
1447 } else {
1448 switch (n->nlmsg_type) {
1449 case RTM_NEWTCLASS:
1450 err = -EEXIST;
1451 if (n->nlmsg_flags & NLM_F_EXCL)
1452 goto out;
1453 break;
1454 case RTM_DELTCLASS:
1455 err = -EOPNOTSUPP;
1456 if (cops->delete)
1457 err = cops->delete(q, cl);
1458 if (err == 0)
1459 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1460 goto out;
1461 case RTM_GETTCLASS:
1462 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1463 goto out;
1464 default:
1465 err = -EINVAL;
1466 goto out;
1467 }
1468 }
1469
1470 new_cl = cl;
1471 err = -EOPNOTSUPP;
1472 if (cops->change)
1473 err = cops->change(q, clid, pid, tca, &new_cl);
1474 if (err == 0)
1475 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1476
1477out:
1478 if (cl)
1479 cops->put(q, cl);
1480
1481 return err;
1482}
1483
1484
1485static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1486 unsigned long cl,
1487 u32 pid, u32 seq, u16 flags, int event)
1488{
1489 struct tcmsg *tcm;
1490 struct nlmsghdr *nlh;
1491 unsigned char *b = skb_tail_pointer(skb);
1492 struct gnet_dump d;
1493 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1494
1495 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1496 tcm = NLMSG_DATA(nlh);
1497 tcm->tcm_family = AF_UNSPEC;
1498 tcm->tcm__pad1 = 0;
1499 tcm->tcm__pad2 = 0;
1500 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1501 tcm->tcm_parent = q->handle;
1502 tcm->tcm_handle = q->handle;
1503 tcm->tcm_info = 0;
1504 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1505 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1506 goto nla_put_failure;
1507
1508 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1509 qdisc_root_sleeping_lock(q), &d) < 0)
1510 goto nla_put_failure;
1511
1512 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1513 goto nla_put_failure;
1514
1515 if (gnet_stats_finish_copy(&d) < 0)
1516 goto nla_put_failure;
1517
1518 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1519 return skb->len;
1520
1521nlmsg_failure:
1522nla_put_failure:
1523 nlmsg_trim(skb, b);
1524 return -1;
1525}
1526
1527static int tclass_notify(struct net *net, struct sk_buff *oskb,
1528 struct nlmsghdr *n, struct Qdisc *q,
1529 unsigned long cl, int event)
1530{
1531 struct sk_buff *skb;
1532 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1533
1534 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1535 if (!skb)
1536 return -ENOBUFS;
1537
1538 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1539 kfree_skb(skb);
1540 return -EINVAL;
1541 }
1542
1543 return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
1544 n->nlmsg_flags & NLM_F_ECHO);
1545}
1546
1547struct qdisc_dump_args {
1548 struct qdisc_walker w;
1549 struct sk_buff *skb;
1550 struct netlink_callback *cb;
1551};
1552
1553static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1554{
1555 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1556
1557 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1558 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1559}
1560
1561static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1562 struct tcmsg *tcm, struct netlink_callback *cb,
1563 int *t_p, int s_t)
1564{
1565 struct qdisc_dump_args arg;
1566
1567 if (tc_qdisc_dump_ignore(q) ||
1568 *t_p < s_t || !q->ops->cl_ops ||
1569 (tcm->tcm_parent &&
1570 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1571 (*t_p)++;
1572 return 0;
1573 }
1574 if (*t_p > s_t)
1575 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1576 arg.w.fn = qdisc_class_dump;
1577 arg.skb = skb;
1578 arg.cb = cb;
1579 arg.w.stop = 0;
1580 arg.w.skip = cb->args[1];
1581 arg.w.count = 0;
1582 q->ops->cl_ops->walk(q, &arg.w);
1583 cb->args[1] = arg.w.count;
1584 if (arg.w.stop)
1585 return -1;
1586 (*t_p)++;
1587 return 0;
1588}
1589
1590static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1591 struct tcmsg *tcm, struct netlink_callback *cb,
1592 int *t_p, int s_t)
1593{
1594 struct Qdisc *q;
1595
1596 if (!root)
1597 return 0;
1598
1599 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1600 return -1;
1601
1602 list_for_each_entry(q, &root->list, list) {
1603 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1604 return -1;
1605 }
1606
1607 return 0;
1608}
1609
1610static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1611{
1612 struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
1613 struct net *net = sock_net(skb->sk);
1614 struct netdev_queue *dev_queue;
1615 struct net_device *dev;
1616 int t, s_t;
1617
1618 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1619 return 0;
1620 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1621 if (!dev)
1622 return 0;
1623
1624 s_t = cb->args[0];
1625 t = 0;
1626
1627 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1628 goto done;
1629
1630 dev_queue = dev_ingress_queue(dev);
1631 if (dev_queue &&
1632 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1633 &t, s_t) < 0)
1634 goto done;
1635
1636done:
1637 cb->args[0] = t;
1638
1639 dev_put(dev);
1640 return skb->len;
1641}
1642
1643/* Main classifier routine: scans classifier chain attached
1644 * to this qdisc, (optionally) tests for protocol and asks
1645 * specific classifiers.
1646 */
1647int tc_classify_compat(struct sk_buff *skb, const struct tcf_proto *tp,
1648 struct tcf_result *res)
1649{
1650 __be16 protocol = skb->protocol;
1651 int err;
1652
1653 for (; tp; tp = tp->next) {
1654 if (tp->protocol != protocol &&
1655 tp->protocol != htons(ETH_P_ALL))
1656 continue;
1657 err = tp->classify(skb, tp, res);
1658
1659 if (err >= 0) {
1660#ifdef CONFIG_NET_CLS_ACT
1661 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1662 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1663#endif
1664 return err;
1665 }
1666 }
1667 return -1;
1668}
1669EXPORT_SYMBOL(tc_classify_compat);
1670
1671int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1672 struct tcf_result *res)
1673{
1674 int err = 0;
1675#ifdef CONFIG_NET_CLS_ACT
1676 const struct tcf_proto *otp = tp;
1677reclassify:
1678#endif
1679
1680 err = tc_classify_compat(skb, tp, res);
1681#ifdef CONFIG_NET_CLS_ACT
1682 if (err == TC_ACT_RECLASSIFY) {
1683 u32 verd = G_TC_VERD(skb->tc_verd);
1684 tp = otp;
1685
1686 if (verd++ >= MAX_REC_LOOP) {
1687 if (net_ratelimit())
1688 pr_notice("%s: packet reclassify loop"
1689 " rule prio %u protocol %02x\n",
1690 tp->q->ops->id,
1691 tp->prio & 0xffff,
1692 ntohs(tp->protocol));
1693 return TC_ACT_SHOT;
1694 }
1695 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1696 goto reclassify;
1697 }
1698#endif
1699 return err;
1700}
1701EXPORT_SYMBOL(tc_classify);
1702
1703void tcf_destroy(struct tcf_proto *tp)
1704{
1705 tp->ops->destroy(tp);
1706 module_put(tp->ops->owner);
1707 kfree(tp);
1708}
1709
1710void tcf_destroy_chain(struct tcf_proto **fl)
1711{
1712 struct tcf_proto *tp;
1713
1714 while ((tp = *fl) != NULL) {
1715 *fl = tp->next;
1716 tcf_destroy(tp);
1717 }
1718}
1719EXPORT_SYMBOL(tcf_destroy_chain);
1720
1721#ifdef CONFIG_PROC_FS
1722static int psched_show(struct seq_file *seq, void *v)
1723{
1724 struct timespec ts;
1725
1726 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1727 seq_printf(seq, "%08x %08x %08x %08x\n",
1728 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1729 1000000,
1730 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1731
1732 return 0;
1733}
1734
1735static int psched_open(struct inode *inode, struct file *file)
1736{
1737 return single_open(file, psched_show, NULL);
1738}
1739
1740static const struct file_operations psched_fops = {
1741 .owner = THIS_MODULE,
1742 .open = psched_open,
1743 .read = seq_read,
1744 .llseek = seq_lseek,
1745 .release = single_release,
1746};
1747
1748static int __net_init psched_net_init(struct net *net)
1749{
1750 struct proc_dir_entry *e;
1751
1752 e = proc_net_fops_create(net, "psched", 0, &psched_fops);
1753 if (e == NULL)
1754 return -ENOMEM;
1755
1756 return 0;
1757}
1758
1759static void __net_exit psched_net_exit(struct net *net)
1760{
1761 proc_net_remove(net, "psched");
1762}
1763#else
1764static int __net_init psched_net_init(struct net *net)
1765{
1766 return 0;
1767}
1768
1769static void __net_exit psched_net_exit(struct net *net)
1770{
1771}
1772#endif
1773
1774static struct pernet_operations psched_net_ops = {
1775 .init = psched_net_init,
1776 .exit = psched_net_exit,
1777};
1778
1779static int __init pktsched_init(void)
1780{
1781 int err;
1782
1783 err = register_pernet_subsys(&psched_net_ops);
1784 if (err) {
1785 pr_err("pktsched_init: "
1786 "cannot initialize per netns operations\n");
1787 return err;
1788 }
1789
1790 register_qdisc(&pfifo_qdisc_ops);
1791 register_qdisc(&bfifo_qdisc_ops);
1792 register_qdisc(&pfifo_head_drop_qdisc_ops);
1793 register_qdisc(&mq_qdisc_ops);
1794
1795 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1796 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1797 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1798 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1799 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1800 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1801
1802 return 0;
1803}
1804
1805subsys_initcall(pktsched_init);