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// 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#include <trace/events/qdisc.h>
36
37/*
38
39 Short review.
40 -------------
41
42 This file consists of two interrelated parts:
43
44 1. queueing disciplines manager frontend.
45 2. traffic classes manager frontend.
46
47 Generally, queueing discipline ("qdisc") is a black box,
48 which is able to enqueue packets and to dequeue them (when
49 device is ready to send something) in order and at times
50 determined by algorithm hidden in it.
51
52 qdisc's are divided to two categories:
53 - "queues", which have no internal structure visible from outside.
54 - "schedulers", which split all the packets to "traffic classes",
55 using "packet classifiers" (look at cls_api.c)
56
57 In turn, classes may have child qdiscs (as rule, queues)
58 attached to them etc. etc. etc.
59
60 The goal of the routines in this file is to translate
61 information supplied by user in the form of handles
62 to more intelligible for kernel form, to make some sanity
63 checks and part of work, which is common to all qdiscs
64 and to provide rtnetlink notifications.
65
66 All real intelligent work is done inside qdisc modules.
67
68
69
70 Every discipline has two major routines: enqueue and dequeue.
71
72 ---dequeue
73
74 dequeue usually returns a skb to send. It is allowed to return NULL,
75 but it does not mean that queue is empty, it just means that
76 discipline does not want to send anything this time.
77 Queue is really empty if q->q.qlen == 0.
78 For complicated disciplines with multiple queues q->q is not
79 real packet queue, but however q->q.qlen must be valid.
80
81 ---enqueue
82
83 enqueue returns 0, if packet was enqueued successfully.
84 If packet (this one or another one) was dropped, it returns
85 not zero error code.
86 NET_XMIT_DROP - this packet dropped
87 Expected action: do not backoff, but wait until queue will clear.
88 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
89 Expected action: backoff or ignore
90
91 Auxiliary routines:
92
93 ---peek
94
95 like dequeue but without removing a packet from the queue
96
97 ---reset
98
99 returns qdisc to initial state: purge all buffers, clear all
100 timers, counters (except for statistics) etc.
101
102 ---init
103
104 initializes newly created qdisc.
105
106 ---destroy
107
108 destroys resources allocated by init and during lifetime of qdisc.
109
110 ---change
111
112 changes qdisc parameters.
113 */
114
115/* Protects list of registered TC modules. It is pure SMP lock. */
116static DEFINE_RWLOCK(qdisc_mod_lock);
117
118
119/************************************************
120 * Queueing disciplines manipulation. *
121 ************************************************/
122
123
124/* The list of all installed queueing disciplines. */
125
126static struct Qdisc_ops *qdisc_base;
127
128/* Register/unregister queueing discipline */
129
130int register_qdisc(struct Qdisc_ops *qops)
131{
132 struct Qdisc_ops *q, **qp;
133 int rc = -EEXIST;
134
135 write_lock(&qdisc_mod_lock);
136 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
137 if (!strcmp(qops->id, q->id))
138 goto out;
139
140 if (qops->enqueue == NULL)
141 qops->enqueue = noop_qdisc_ops.enqueue;
142 if (qops->peek == NULL) {
143 if (qops->dequeue == NULL)
144 qops->peek = noop_qdisc_ops.peek;
145 else
146 goto out_einval;
147 }
148 if (qops->dequeue == NULL)
149 qops->dequeue = noop_qdisc_ops.dequeue;
150
151 if (qops->cl_ops) {
152 const struct Qdisc_class_ops *cops = qops->cl_ops;
153
154 if (!(cops->find && cops->walk && cops->leaf))
155 goto out_einval;
156
157 if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
158 goto out_einval;
159 }
160
161 qops->next = NULL;
162 *qp = qops;
163 rc = 0;
164out:
165 write_unlock(&qdisc_mod_lock);
166 return rc;
167
168out_einval:
169 rc = -EINVAL;
170 goto out;
171}
172EXPORT_SYMBOL(register_qdisc);
173
174int unregister_qdisc(struct Qdisc_ops *qops)
175{
176 struct Qdisc_ops *q, **qp;
177 int err = -ENOENT;
178
179 write_lock(&qdisc_mod_lock);
180 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
181 if (q == qops)
182 break;
183 if (q) {
184 *qp = q->next;
185 q->next = NULL;
186 err = 0;
187 }
188 write_unlock(&qdisc_mod_lock);
189 return err;
190}
191EXPORT_SYMBOL(unregister_qdisc);
192
193/* Get default qdisc if not otherwise specified */
194void qdisc_get_default(char *name, size_t len)
195{
196 read_lock(&qdisc_mod_lock);
197 strlcpy(name, default_qdisc_ops->id, len);
198 read_unlock(&qdisc_mod_lock);
199}
200
201static struct Qdisc_ops *qdisc_lookup_default(const char *name)
202{
203 struct Qdisc_ops *q = NULL;
204
205 for (q = qdisc_base; q; q = q->next) {
206 if (!strcmp(name, q->id)) {
207 if (!try_module_get(q->owner))
208 q = NULL;
209 break;
210 }
211 }
212
213 return q;
214}
215
216/* Set new default qdisc to use */
217int qdisc_set_default(const char *name)
218{
219 const struct Qdisc_ops *ops;
220
221 if (!capable(CAP_NET_ADMIN))
222 return -EPERM;
223
224 write_lock(&qdisc_mod_lock);
225 ops = qdisc_lookup_default(name);
226 if (!ops) {
227 /* Not found, drop lock and try to load module */
228 write_unlock(&qdisc_mod_lock);
229 request_module("sch_%s", name);
230 write_lock(&qdisc_mod_lock);
231
232 ops = qdisc_lookup_default(name);
233 }
234
235 if (ops) {
236 /* Set new default */
237 module_put(default_qdisc_ops->owner);
238 default_qdisc_ops = ops;
239 }
240 write_unlock(&qdisc_mod_lock);
241
242 return ops ? 0 : -ENOENT;
243}
244
245#ifdef CONFIG_NET_SCH_DEFAULT
246/* Set default value from kernel config */
247static int __init sch_default_qdisc(void)
248{
249 return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
250}
251late_initcall(sch_default_qdisc);
252#endif
253
254/* We know handle. Find qdisc among all qdisc's attached to device
255 * (root qdisc, all its children, children of children etc.)
256 * Note: caller either uses rtnl or rcu_read_lock()
257 */
258
259static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
260{
261 struct Qdisc *q;
262
263 if (!qdisc_dev(root))
264 return (root->handle == handle ? root : NULL);
265
266 if (!(root->flags & TCQ_F_BUILTIN) &&
267 root->handle == handle)
268 return root;
269
270 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle,
271 lockdep_rtnl_is_held()) {
272 if (q->handle == handle)
273 return q;
274 }
275 return NULL;
276}
277
278void qdisc_hash_add(struct Qdisc *q, bool invisible)
279{
280 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
281 ASSERT_RTNL();
282 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
283 if (invisible)
284 q->flags |= TCQ_F_INVISIBLE;
285 }
286}
287EXPORT_SYMBOL(qdisc_hash_add);
288
289void qdisc_hash_del(struct Qdisc *q)
290{
291 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
292 ASSERT_RTNL();
293 hash_del_rcu(&q->hash);
294 }
295}
296EXPORT_SYMBOL(qdisc_hash_del);
297
298struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
299{
300 struct Qdisc *q;
301
302 if (!handle)
303 return NULL;
304 q = qdisc_match_from_root(dev->qdisc, handle);
305 if (q)
306 goto out;
307
308 if (dev_ingress_queue(dev))
309 q = qdisc_match_from_root(
310 dev_ingress_queue(dev)->qdisc_sleeping,
311 handle);
312out:
313 return q;
314}
315
316struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
317{
318 struct netdev_queue *nq;
319 struct Qdisc *q;
320
321 if (!handle)
322 return NULL;
323 q = qdisc_match_from_root(dev->qdisc, handle);
324 if (q)
325 goto out;
326
327 nq = dev_ingress_queue_rcu(dev);
328 if (nq)
329 q = qdisc_match_from_root(nq->qdisc_sleeping, handle);
330out:
331 return q;
332}
333
334static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
335{
336 unsigned long cl;
337 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
338
339 if (cops == NULL)
340 return NULL;
341 cl = cops->find(p, classid);
342
343 if (cl == 0)
344 return NULL;
345 return cops->leaf(p, cl);
346}
347
348/* Find queueing discipline by name */
349
350static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
351{
352 struct Qdisc_ops *q = NULL;
353
354 if (kind) {
355 read_lock(&qdisc_mod_lock);
356 for (q = qdisc_base; q; q = q->next) {
357 if (nla_strcmp(kind, q->id) == 0) {
358 if (!try_module_get(q->owner))
359 q = NULL;
360 break;
361 }
362 }
363 read_unlock(&qdisc_mod_lock);
364 }
365 return q;
366}
367
368/* The linklayer setting were not transferred from iproute2, in older
369 * versions, and the rate tables lookup systems have been dropped in
370 * the kernel. To keep backward compatible with older iproute2 tc
371 * utils, we detect the linklayer setting by detecting if the rate
372 * table were modified.
373 *
374 * For linklayer ATM table entries, the rate table will be aligned to
375 * 48 bytes, thus some table entries will contain the same value. The
376 * mpu (min packet unit) is also encoded into the old rate table, thus
377 * starting from the mpu, we find low and high table entries for
378 * mapping this cell. If these entries contain the same value, when
379 * the rate tables have been modified for linklayer ATM.
380 *
381 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
382 * and then roundup to the next cell, calc the table entry one below,
383 * and compare.
384 */
385static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
386{
387 int low = roundup(r->mpu, 48);
388 int high = roundup(low+1, 48);
389 int cell_low = low >> r->cell_log;
390 int cell_high = (high >> r->cell_log) - 1;
391
392 /* rtab is too inaccurate at rates > 100Mbit/s */
393 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
394 pr_debug("TC linklayer: Giving up ATM detection\n");
395 return TC_LINKLAYER_ETHERNET;
396 }
397
398 if ((cell_high > cell_low) && (cell_high < 256)
399 && (rtab[cell_low] == rtab[cell_high])) {
400 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
401 cell_low, cell_high, rtab[cell_high]);
402 return TC_LINKLAYER_ATM;
403 }
404 return TC_LINKLAYER_ETHERNET;
405}
406
407static struct qdisc_rate_table *qdisc_rtab_list;
408
409struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
410 struct nlattr *tab,
411 struct netlink_ext_ack *extack)
412{
413 struct qdisc_rate_table *rtab;
414
415 if (tab == NULL || r->rate == 0 ||
416 r->cell_log == 0 || r->cell_log >= 32 ||
417 nla_len(tab) != TC_RTAB_SIZE) {
418 NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
419 return NULL;
420 }
421
422 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
423 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
424 !memcmp(&rtab->data, nla_data(tab), 1024)) {
425 rtab->refcnt++;
426 return rtab;
427 }
428 }
429
430 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
431 if (rtab) {
432 rtab->rate = *r;
433 rtab->refcnt = 1;
434 memcpy(rtab->data, nla_data(tab), 1024);
435 if (r->linklayer == TC_LINKLAYER_UNAWARE)
436 r->linklayer = __detect_linklayer(r, rtab->data);
437 rtab->next = qdisc_rtab_list;
438 qdisc_rtab_list = rtab;
439 } else {
440 NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
441 }
442 return rtab;
443}
444EXPORT_SYMBOL(qdisc_get_rtab);
445
446void qdisc_put_rtab(struct qdisc_rate_table *tab)
447{
448 struct qdisc_rate_table *rtab, **rtabp;
449
450 if (!tab || --tab->refcnt)
451 return;
452
453 for (rtabp = &qdisc_rtab_list;
454 (rtab = *rtabp) != NULL;
455 rtabp = &rtab->next) {
456 if (rtab == tab) {
457 *rtabp = rtab->next;
458 kfree(rtab);
459 return;
460 }
461 }
462}
463EXPORT_SYMBOL(qdisc_put_rtab);
464
465static LIST_HEAD(qdisc_stab_list);
466
467static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
468 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
469 [TCA_STAB_DATA] = { .type = NLA_BINARY },
470};
471
472static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
473 struct netlink_ext_ack *extack)
474{
475 struct nlattr *tb[TCA_STAB_MAX + 1];
476 struct qdisc_size_table *stab;
477 struct tc_sizespec *s;
478 unsigned int tsize = 0;
479 u16 *tab = NULL;
480 int err;
481
482 err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
483 extack);
484 if (err < 0)
485 return ERR_PTR(err);
486 if (!tb[TCA_STAB_BASE]) {
487 NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
488 return ERR_PTR(-EINVAL);
489 }
490
491 s = nla_data(tb[TCA_STAB_BASE]);
492
493 if (s->tsize > 0) {
494 if (!tb[TCA_STAB_DATA]) {
495 NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
496 return ERR_PTR(-EINVAL);
497 }
498 tab = nla_data(tb[TCA_STAB_DATA]);
499 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
500 }
501
502 if (tsize != s->tsize || (!tab && tsize > 0)) {
503 NL_SET_ERR_MSG(extack, "Invalid size of size table");
504 return ERR_PTR(-EINVAL);
505 }
506
507 list_for_each_entry(stab, &qdisc_stab_list, list) {
508 if (memcmp(&stab->szopts, s, sizeof(*s)))
509 continue;
510 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
511 continue;
512 stab->refcnt++;
513 return stab;
514 }
515
516 if (s->size_log > STAB_SIZE_LOG_MAX ||
517 s->cell_log > STAB_SIZE_LOG_MAX) {
518 NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
519 return ERR_PTR(-EINVAL);
520 }
521
522 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
523 if (!stab)
524 return ERR_PTR(-ENOMEM);
525
526 stab->refcnt = 1;
527 stab->szopts = *s;
528 if (tsize > 0)
529 memcpy(stab->data, tab, tsize * sizeof(u16));
530
531 list_add_tail(&stab->list, &qdisc_stab_list);
532
533 return stab;
534}
535
536void qdisc_put_stab(struct qdisc_size_table *tab)
537{
538 if (!tab)
539 return;
540
541 if (--tab->refcnt == 0) {
542 list_del(&tab->list);
543 kfree_rcu(tab, rcu);
544 }
545}
546EXPORT_SYMBOL(qdisc_put_stab);
547
548static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
549{
550 struct nlattr *nest;
551
552 nest = nla_nest_start_noflag(skb, TCA_STAB);
553 if (nest == NULL)
554 goto nla_put_failure;
555 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
556 goto nla_put_failure;
557 nla_nest_end(skb, nest);
558
559 return skb->len;
560
561nla_put_failure:
562 return -1;
563}
564
565void __qdisc_calculate_pkt_len(struct sk_buff *skb,
566 const struct qdisc_size_table *stab)
567{
568 int pkt_len, slot;
569
570 pkt_len = skb->len + stab->szopts.overhead;
571 if (unlikely(!stab->szopts.tsize))
572 goto out;
573
574 slot = pkt_len + stab->szopts.cell_align;
575 if (unlikely(slot < 0))
576 slot = 0;
577
578 slot >>= stab->szopts.cell_log;
579 if (likely(slot < stab->szopts.tsize))
580 pkt_len = stab->data[slot];
581 else
582 pkt_len = stab->data[stab->szopts.tsize - 1] *
583 (slot / stab->szopts.tsize) +
584 stab->data[slot % stab->szopts.tsize];
585
586 pkt_len <<= stab->szopts.size_log;
587out:
588 if (unlikely(pkt_len < 1))
589 pkt_len = 1;
590 qdisc_skb_cb(skb)->pkt_len = pkt_len;
591}
592EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
593
594void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
595{
596 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
597 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
598 txt, qdisc->ops->id, qdisc->handle >> 16);
599 qdisc->flags |= TCQ_F_WARN_NONWC;
600 }
601}
602EXPORT_SYMBOL(qdisc_warn_nonwc);
603
604static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
605{
606 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
607 timer);
608
609 rcu_read_lock();
610 __netif_schedule(qdisc_root(wd->qdisc));
611 rcu_read_unlock();
612
613 return HRTIMER_NORESTART;
614}
615
616void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
617 clockid_t clockid)
618{
619 hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
620 wd->timer.function = qdisc_watchdog;
621 wd->qdisc = qdisc;
622}
623EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
624
625void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
626{
627 qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
628}
629EXPORT_SYMBOL(qdisc_watchdog_init);
630
631void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
632 u64 delta_ns)
633{
634 if (test_bit(__QDISC_STATE_DEACTIVATED,
635 &qdisc_root_sleeping(wd->qdisc)->state))
636 return;
637
638 if (hrtimer_is_queued(&wd->timer)) {
639 /* If timer is already set in [expires, expires + delta_ns],
640 * do not reprogram it.
641 */
642 if (wd->last_expires - expires <= delta_ns)
643 return;
644 }
645
646 wd->last_expires = expires;
647 hrtimer_start_range_ns(&wd->timer,
648 ns_to_ktime(expires),
649 delta_ns,
650 HRTIMER_MODE_ABS_PINNED);
651}
652EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
653
654void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
655{
656 hrtimer_cancel(&wd->timer);
657}
658EXPORT_SYMBOL(qdisc_watchdog_cancel);
659
660static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
661{
662 struct hlist_head *h;
663 unsigned int i;
664
665 h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
666
667 if (h != NULL) {
668 for (i = 0; i < n; i++)
669 INIT_HLIST_HEAD(&h[i]);
670 }
671 return h;
672}
673
674void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
675{
676 struct Qdisc_class_common *cl;
677 struct hlist_node *next;
678 struct hlist_head *nhash, *ohash;
679 unsigned int nsize, nmask, osize;
680 unsigned int i, h;
681
682 /* Rehash when load factor exceeds 0.75 */
683 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
684 return;
685 nsize = clhash->hashsize * 2;
686 nmask = nsize - 1;
687 nhash = qdisc_class_hash_alloc(nsize);
688 if (nhash == NULL)
689 return;
690
691 ohash = clhash->hash;
692 osize = clhash->hashsize;
693
694 sch_tree_lock(sch);
695 for (i = 0; i < osize; i++) {
696 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
697 h = qdisc_class_hash(cl->classid, nmask);
698 hlist_add_head(&cl->hnode, &nhash[h]);
699 }
700 }
701 clhash->hash = nhash;
702 clhash->hashsize = nsize;
703 clhash->hashmask = nmask;
704 sch_tree_unlock(sch);
705
706 kvfree(ohash);
707}
708EXPORT_SYMBOL(qdisc_class_hash_grow);
709
710int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
711{
712 unsigned int size = 4;
713
714 clhash->hash = qdisc_class_hash_alloc(size);
715 if (!clhash->hash)
716 return -ENOMEM;
717 clhash->hashsize = size;
718 clhash->hashmask = size - 1;
719 clhash->hashelems = 0;
720 return 0;
721}
722EXPORT_SYMBOL(qdisc_class_hash_init);
723
724void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
725{
726 kvfree(clhash->hash);
727}
728EXPORT_SYMBOL(qdisc_class_hash_destroy);
729
730void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
731 struct Qdisc_class_common *cl)
732{
733 unsigned int h;
734
735 INIT_HLIST_NODE(&cl->hnode);
736 h = qdisc_class_hash(cl->classid, clhash->hashmask);
737 hlist_add_head(&cl->hnode, &clhash->hash[h]);
738 clhash->hashelems++;
739}
740EXPORT_SYMBOL(qdisc_class_hash_insert);
741
742void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
743 struct Qdisc_class_common *cl)
744{
745 hlist_del(&cl->hnode);
746 clhash->hashelems--;
747}
748EXPORT_SYMBOL(qdisc_class_hash_remove);
749
750/* Allocate an unique handle from space managed by kernel
751 * Possible range is [8000-FFFF]:0000 (0x8000 values)
752 */
753static u32 qdisc_alloc_handle(struct net_device *dev)
754{
755 int i = 0x8000;
756 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
757
758 do {
759 autohandle += TC_H_MAKE(0x10000U, 0);
760 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
761 autohandle = TC_H_MAKE(0x80000000U, 0);
762 if (!qdisc_lookup(dev, autohandle))
763 return autohandle;
764 cond_resched();
765 } while (--i > 0);
766
767 return 0;
768}
769
770void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
771{
772 bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
773 const struct Qdisc_class_ops *cops;
774 unsigned long cl;
775 u32 parentid;
776 bool notify;
777 int drops;
778
779 if (n == 0 && len == 0)
780 return;
781 drops = max_t(int, n, 0);
782 rcu_read_lock();
783 while ((parentid = sch->parent)) {
784 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
785 break;
786
787 if (sch->flags & TCQ_F_NOPARENT)
788 break;
789 /* Notify parent qdisc only if child qdisc becomes empty.
790 *
791 * If child was empty even before update then backlog
792 * counter is screwed and we skip notification because
793 * parent class is already passive.
794 *
795 * If the original child was offloaded then it is allowed
796 * to be seem as empty, so the parent is notified anyway.
797 */
798 notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
799 !qdisc_is_offloaded);
800 /* TODO: perform the search on a per txq basis */
801 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
802 if (sch == NULL) {
803 WARN_ON_ONCE(parentid != TC_H_ROOT);
804 break;
805 }
806 cops = sch->ops->cl_ops;
807 if (notify && cops->qlen_notify) {
808 cl = cops->find(sch, parentid);
809 cops->qlen_notify(sch, cl);
810 }
811 sch->q.qlen -= n;
812 sch->qstats.backlog -= len;
813 __qdisc_qstats_drop(sch, drops);
814 }
815 rcu_read_unlock();
816}
817EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
818
819int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
820 void *type_data)
821{
822 struct net_device *dev = qdisc_dev(sch);
823 int err;
824
825 sch->flags &= ~TCQ_F_OFFLOADED;
826 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
827 return 0;
828
829 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
830 if (err == -EOPNOTSUPP)
831 return 0;
832
833 if (!err)
834 sch->flags |= TCQ_F_OFFLOADED;
835
836 return err;
837}
838EXPORT_SYMBOL(qdisc_offload_dump_helper);
839
840void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
841 struct Qdisc *new, struct Qdisc *old,
842 enum tc_setup_type type, void *type_data,
843 struct netlink_ext_ack *extack)
844{
845 bool any_qdisc_is_offloaded;
846 int err;
847
848 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
849 return;
850
851 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
852
853 /* Don't report error if the graft is part of destroy operation. */
854 if (!err || !new || new == &noop_qdisc)
855 return;
856
857 /* Don't report error if the parent, the old child and the new
858 * one are not offloaded.
859 */
860 any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
861 any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
862 any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
863
864 if (any_qdisc_is_offloaded)
865 NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
866}
867EXPORT_SYMBOL(qdisc_offload_graft_helper);
868
869static void qdisc_offload_graft_root(struct net_device *dev,
870 struct Qdisc *new, struct Qdisc *old,
871 struct netlink_ext_ack *extack)
872{
873 struct tc_root_qopt_offload graft_offload = {
874 .command = TC_ROOT_GRAFT,
875 .handle = new ? new->handle : 0,
876 .ingress = (new && new->flags & TCQ_F_INGRESS) ||
877 (old && old->flags & TCQ_F_INGRESS),
878 };
879
880 qdisc_offload_graft_helper(dev, NULL, new, old,
881 TC_SETUP_ROOT_QDISC, &graft_offload, extack);
882}
883
884static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
885 u32 portid, u32 seq, u16 flags, int event)
886{
887 struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
888 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
889 struct tcmsg *tcm;
890 struct nlmsghdr *nlh;
891 unsigned char *b = skb_tail_pointer(skb);
892 struct gnet_dump d;
893 struct qdisc_size_table *stab;
894 u32 block_index;
895 __u32 qlen;
896
897 cond_resched();
898 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
899 if (!nlh)
900 goto out_nlmsg_trim;
901 tcm = nlmsg_data(nlh);
902 tcm->tcm_family = AF_UNSPEC;
903 tcm->tcm__pad1 = 0;
904 tcm->tcm__pad2 = 0;
905 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
906 tcm->tcm_parent = clid;
907 tcm->tcm_handle = q->handle;
908 tcm->tcm_info = refcount_read(&q->refcnt);
909 if (nla_put_string(skb, TCA_KIND, q->ops->id))
910 goto nla_put_failure;
911 if (q->ops->ingress_block_get) {
912 block_index = q->ops->ingress_block_get(q);
913 if (block_index &&
914 nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
915 goto nla_put_failure;
916 }
917 if (q->ops->egress_block_get) {
918 block_index = q->ops->egress_block_get(q);
919 if (block_index &&
920 nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
921 goto nla_put_failure;
922 }
923 if (q->ops->dump && q->ops->dump(q, skb) < 0)
924 goto nla_put_failure;
925 if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
926 goto nla_put_failure;
927 qlen = qdisc_qlen_sum(q);
928
929 stab = rtnl_dereference(q->stab);
930 if (stab && qdisc_dump_stab(skb, stab) < 0)
931 goto nla_put_failure;
932
933 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
934 NULL, &d, TCA_PAD) < 0)
935 goto nla_put_failure;
936
937 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
938 goto nla_put_failure;
939
940 if (qdisc_is_percpu_stats(q)) {
941 cpu_bstats = q->cpu_bstats;
942 cpu_qstats = q->cpu_qstats;
943 }
944
945 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
946 &d, cpu_bstats, &q->bstats) < 0 ||
947 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
948 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
949 goto nla_put_failure;
950
951 if (gnet_stats_finish_copy(&d) < 0)
952 goto nla_put_failure;
953
954 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
955 return skb->len;
956
957out_nlmsg_trim:
958nla_put_failure:
959 nlmsg_trim(skb, b);
960 return -1;
961}
962
963static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
964{
965 if (q->flags & TCQ_F_BUILTIN)
966 return true;
967 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
968 return true;
969
970 return false;
971}
972
973static int qdisc_notify(struct net *net, struct sk_buff *oskb,
974 struct nlmsghdr *n, u32 clid,
975 struct Qdisc *old, struct Qdisc *new)
976{
977 struct sk_buff *skb;
978 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
979
980 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
981 if (!skb)
982 return -ENOBUFS;
983
984 if (old && !tc_qdisc_dump_ignore(old, false)) {
985 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
986 0, RTM_DELQDISC) < 0)
987 goto err_out;
988 }
989 if (new && !tc_qdisc_dump_ignore(new, false)) {
990 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
991 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
992 goto err_out;
993 }
994
995 if (skb->len)
996 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
997 n->nlmsg_flags & NLM_F_ECHO);
998
999err_out:
1000 kfree_skb(skb);
1001 return -EINVAL;
1002}
1003
1004static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1005 struct nlmsghdr *n, u32 clid,
1006 struct Qdisc *old, struct Qdisc *new)
1007{
1008 if (new || old)
1009 qdisc_notify(net, skb, n, clid, old, new);
1010
1011 if (old)
1012 qdisc_put(old);
1013}
1014
1015static void qdisc_clear_nolock(struct Qdisc *sch)
1016{
1017 sch->flags &= ~TCQ_F_NOLOCK;
1018 if (!(sch->flags & TCQ_F_CPUSTATS))
1019 return;
1020
1021 free_percpu(sch->cpu_bstats);
1022 free_percpu(sch->cpu_qstats);
1023 sch->cpu_bstats = NULL;
1024 sch->cpu_qstats = NULL;
1025 sch->flags &= ~TCQ_F_CPUSTATS;
1026}
1027
1028/* Graft qdisc "new" to class "classid" of qdisc "parent" or
1029 * to device "dev".
1030 *
1031 * When appropriate send a netlink notification using 'skb'
1032 * and "n".
1033 *
1034 * On success, destroy old qdisc.
1035 */
1036
1037static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1038 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1039 struct Qdisc *new, struct Qdisc *old,
1040 struct netlink_ext_ack *extack)
1041{
1042 struct Qdisc *q = old;
1043 struct net *net = dev_net(dev);
1044
1045 if (parent == NULL) {
1046 unsigned int i, num_q, ingress;
1047
1048 ingress = 0;
1049 num_q = dev->num_tx_queues;
1050 if ((q && q->flags & TCQ_F_INGRESS) ||
1051 (new && new->flags & TCQ_F_INGRESS)) {
1052 num_q = 1;
1053 ingress = 1;
1054 if (!dev_ingress_queue(dev)) {
1055 NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1056 return -ENOENT;
1057 }
1058 }
1059
1060 if (dev->flags & IFF_UP)
1061 dev_deactivate(dev);
1062
1063 qdisc_offload_graft_root(dev, new, old, extack);
1064
1065 if (new && new->ops->attach)
1066 goto skip;
1067
1068 for (i = 0; i < num_q; i++) {
1069 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1070
1071 if (!ingress)
1072 dev_queue = netdev_get_tx_queue(dev, i);
1073
1074 old = dev_graft_qdisc(dev_queue, new);
1075 if (new && i > 0)
1076 qdisc_refcount_inc(new);
1077
1078 if (!ingress)
1079 qdisc_put(old);
1080 }
1081
1082skip:
1083 if (!ingress) {
1084 notify_and_destroy(net, skb, n, classid,
1085 dev->qdisc, new);
1086 if (new && !new->ops->attach)
1087 qdisc_refcount_inc(new);
1088 dev->qdisc = new ? : &noop_qdisc;
1089
1090 if (new && new->ops->attach)
1091 new->ops->attach(new);
1092 } else {
1093 notify_and_destroy(net, skb, n, classid, old, new);
1094 }
1095
1096 if (dev->flags & IFF_UP)
1097 dev_activate(dev);
1098 } else {
1099 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1100 unsigned long cl;
1101 int err;
1102
1103 /* Only support running class lockless if parent is lockless */
1104 if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1105 qdisc_clear_nolock(new);
1106
1107 if (!cops || !cops->graft)
1108 return -EOPNOTSUPP;
1109
1110 cl = cops->find(parent, classid);
1111 if (!cl) {
1112 NL_SET_ERR_MSG(extack, "Specified class not found");
1113 return -ENOENT;
1114 }
1115
1116 err = cops->graft(parent, cl, new, &old, extack);
1117 if (err)
1118 return err;
1119 notify_and_destroy(net, skb, n, classid, old, new);
1120 }
1121 return 0;
1122}
1123
1124static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1125 struct netlink_ext_ack *extack)
1126{
1127 u32 block_index;
1128
1129 if (tca[TCA_INGRESS_BLOCK]) {
1130 block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1131
1132 if (!block_index) {
1133 NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1134 return -EINVAL;
1135 }
1136 if (!sch->ops->ingress_block_set) {
1137 NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1138 return -EOPNOTSUPP;
1139 }
1140 sch->ops->ingress_block_set(sch, block_index);
1141 }
1142 if (tca[TCA_EGRESS_BLOCK]) {
1143 block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1144
1145 if (!block_index) {
1146 NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1147 return -EINVAL;
1148 }
1149 if (!sch->ops->egress_block_set) {
1150 NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1151 return -EOPNOTSUPP;
1152 }
1153 sch->ops->egress_block_set(sch, block_index);
1154 }
1155 return 0;
1156}
1157
1158/*
1159 Allocate and initialize new qdisc.
1160
1161 Parameters are passed via opt.
1162 */
1163
1164static struct Qdisc *qdisc_create(struct net_device *dev,
1165 struct netdev_queue *dev_queue,
1166 struct Qdisc *p, u32 parent, u32 handle,
1167 struct nlattr **tca, int *errp,
1168 struct netlink_ext_ack *extack)
1169{
1170 int err;
1171 struct nlattr *kind = tca[TCA_KIND];
1172 struct Qdisc *sch;
1173 struct Qdisc_ops *ops;
1174 struct qdisc_size_table *stab;
1175
1176 ops = qdisc_lookup_ops(kind);
1177#ifdef CONFIG_MODULES
1178 if (ops == NULL && kind != NULL) {
1179 char name[IFNAMSIZ];
1180 if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1181 /* We dropped the RTNL semaphore in order to
1182 * perform the module load. So, even if we
1183 * succeeded in loading the module we have to
1184 * tell the caller to replay the request. We
1185 * indicate this using -EAGAIN.
1186 * We replay the request because the device may
1187 * go away in the mean time.
1188 */
1189 rtnl_unlock();
1190 request_module("sch_%s", name);
1191 rtnl_lock();
1192 ops = qdisc_lookup_ops(kind);
1193 if (ops != NULL) {
1194 /* We will try again qdisc_lookup_ops,
1195 * so don't keep a reference.
1196 */
1197 module_put(ops->owner);
1198 err = -EAGAIN;
1199 goto err_out;
1200 }
1201 }
1202 }
1203#endif
1204
1205 err = -ENOENT;
1206 if (!ops) {
1207 NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1208 goto err_out;
1209 }
1210
1211 sch = qdisc_alloc(dev_queue, ops, extack);
1212 if (IS_ERR(sch)) {
1213 err = PTR_ERR(sch);
1214 goto err_out2;
1215 }
1216
1217 sch->parent = parent;
1218
1219 if (handle == TC_H_INGRESS) {
1220 sch->flags |= TCQ_F_INGRESS;
1221 handle = TC_H_MAKE(TC_H_INGRESS, 0);
1222 } else {
1223 if (handle == 0) {
1224 handle = qdisc_alloc_handle(dev);
1225 if (handle == 0) {
1226 NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1227 err = -ENOSPC;
1228 goto err_out3;
1229 }
1230 }
1231 if (!netif_is_multiqueue(dev))
1232 sch->flags |= TCQ_F_ONETXQUEUE;
1233 }
1234
1235 sch->handle = handle;
1236
1237 /* This exist to keep backward compatible with a userspace
1238 * loophole, what allowed userspace to get IFF_NO_QUEUE
1239 * facility on older kernels by setting tx_queue_len=0 (prior
1240 * to qdisc init), and then forgot to reinit tx_queue_len
1241 * before again attaching a qdisc.
1242 */
1243 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1244 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1245 netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1246 }
1247
1248 err = qdisc_block_indexes_set(sch, tca, extack);
1249 if (err)
1250 goto err_out3;
1251
1252 if (ops->init) {
1253 err = ops->init(sch, tca[TCA_OPTIONS], extack);
1254 if (err != 0)
1255 goto err_out5;
1256 }
1257
1258 if (tca[TCA_STAB]) {
1259 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1260 if (IS_ERR(stab)) {
1261 err = PTR_ERR(stab);
1262 goto err_out4;
1263 }
1264 rcu_assign_pointer(sch->stab, stab);
1265 }
1266 if (tca[TCA_RATE]) {
1267 seqcount_t *running;
1268
1269 err = -EOPNOTSUPP;
1270 if (sch->flags & TCQ_F_MQROOT) {
1271 NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1272 goto err_out4;
1273 }
1274
1275 if (sch->parent != TC_H_ROOT &&
1276 !(sch->flags & TCQ_F_INGRESS) &&
1277 (!p || !(p->flags & TCQ_F_MQROOT)))
1278 running = qdisc_root_sleeping_running(sch);
1279 else
1280 running = &sch->running;
1281
1282 err = gen_new_estimator(&sch->bstats,
1283 sch->cpu_bstats,
1284 &sch->rate_est,
1285 NULL,
1286 running,
1287 tca[TCA_RATE]);
1288 if (err) {
1289 NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1290 goto err_out4;
1291 }
1292 }
1293
1294 qdisc_hash_add(sch, false);
1295 trace_qdisc_create(ops, dev, parent);
1296
1297 return sch;
1298
1299err_out5:
1300 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1301 if (ops->destroy)
1302 ops->destroy(sch);
1303err_out3:
1304 dev_put(dev);
1305 qdisc_free(sch);
1306err_out2:
1307 module_put(ops->owner);
1308err_out:
1309 *errp = err;
1310 return NULL;
1311
1312err_out4:
1313 /*
1314 * Any broken qdiscs that would require a ops->reset() here?
1315 * The qdisc was never in action so it shouldn't be necessary.
1316 */
1317 qdisc_put_stab(rtnl_dereference(sch->stab));
1318 if (ops->destroy)
1319 ops->destroy(sch);
1320 goto err_out3;
1321}
1322
1323static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1324 struct netlink_ext_ack *extack)
1325{
1326 struct qdisc_size_table *ostab, *stab = NULL;
1327 int err = 0;
1328
1329 if (tca[TCA_OPTIONS]) {
1330 if (!sch->ops->change) {
1331 NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1332 return -EINVAL;
1333 }
1334 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1335 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1336 return -EOPNOTSUPP;
1337 }
1338 err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1339 if (err)
1340 return err;
1341 }
1342
1343 if (tca[TCA_STAB]) {
1344 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1345 if (IS_ERR(stab))
1346 return PTR_ERR(stab);
1347 }
1348
1349 ostab = rtnl_dereference(sch->stab);
1350 rcu_assign_pointer(sch->stab, stab);
1351 qdisc_put_stab(ostab);
1352
1353 if (tca[TCA_RATE]) {
1354 /* NB: ignores errors from replace_estimator
1355 because change can't be undone. */
1356 if (sch->flags & TCQ_F_MQROOT)
1357 goto out;
1358 gen_replace_estimator(&sch->bstats,
1359 sch->cpu_bstats,
1360 &sch->rate_est,
1361 NULL,
1362 qdisc_root_sleeping_running(sch),
1363 tca[TCA_RATE]);
1364 }
1365out:
1366 return 0;
1367}
1368
1369struct check_loop_arg {
1370 struct qdisc_walker w;
1371 struct Qdisc *p;
1372 int depth;
1373};
1374
1375static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1376 struct qdisc_walker *w);
1377
1378static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1379{
1380 struct check_loop_arg arg;
1381
1382 if (q->ops->cl_ops == NULL)
1383 return 0;
1384
1385 arg.w.stop = arg.w.skip = arg.w.count = 0;
1386 arg.w.fn = check_loop_fn;
1387 arg.depth = depth;
1388 arg.p = p;
1389 q->ops->cl_ops->walk(q, &arg.w);
1390 return arg.w.stop ? -ELOOP : 0;
1391}
1392
1393static int
1394check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1395{
1396 struct Qdisc *leaf;
1397 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1398 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1399
1400 leaf = cops->leaf(q, cl);
1401 if (leaf) {
1402 if (leaf == arg->p || arg->depth > 7)
1403 return -ELOOP;
1404 return check_loop(leaf, arg->p, arg->depth + 1);
1405 }
1406 return 0;
1407}
1408
1409const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1410 [TCA_KIND] = { .type = NLA_STRING },
1411 [TCA_RATE] = { .type = NLA_BINARY,
1412 .len = sizeof(struct tc_estimator) },
1413 [TCA_STAB] = { .type = NLA_NESTED },
1414 [TCA_DUMP_INVISIBLE] = { .type = NLA_FLAG },
1415 [TCA_CHAIN] = { .type = NLA_U32 },
1416 [TCA_INGRESS_BLOCK] = { .type = NLA_U32 },
1417 [TCA_EGRESS_BLOCK] = { .type = NLA_U32 },
1418};
1419
1420/*
1421 * Delete/get qdisc.
1422 */
1423
1424static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1425 struct netlink_ext_ack *extack)
1426{
1427 struct net *net = sock_net(skb->sk);
1428 struct tcmsg *tcm = nlmsg_data(n);
1429 struct nlattr *tca[TCA_MAX + 1];
1430 struct net_device *dev;
1431 u32 clid;
1432 struct Qdisc *q = NULL;
1433 struct Qdisc *p = NULL;
1434 int err;
1435
1436 if ((n->nlmsg_type != RTM_GETQDISC) &&
1437 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1438 return -EPERM;
1439
1440 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1441 rtm_tca_policy, extack);
1442 if (err < 0)
1443 return err;
1444
1445 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1446 if (!dev)
1447 return -ENODEV;
1448
1449 clid = tcm->tcm_parent;
1450 if (clid) {
1451 if (clid != TC_H_ROOT) {
1452 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1453 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1454 if (!p) {
1455 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1456 return -ENOENT;
1457 }
1458 q = qdisc_leaf(p, clid);
1459 } else if (dev_ingress_queue(dev)) {
1460 q = dev_ingress_queue(dev)->qdisc_sleeping;
1461 }
1462 } else {
1463 q = dev->qdisc;
1464 }
1465 if (!q) {
1466 NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1467 return -ENOENT;
1468 }
1469
1470 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1471 NL_SET_ERR_MSG(extack, "Invalid handle");
1472 return -EINVAL;
1473 }
1474 } else {
1475 q = qdisc_lookup(dev, tcm->tcm_handle);
1476 if (!q) {
1477 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1478 return -ENOENT;
1479 }
1480 }
1481
1482 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1483 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1484 return -EINVAL;
1485 }
1486
1487 if (n->nlmsg_type == RTM_DELQDISC) {
1488 if (!clid) {
1489 NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1490 return -EINVAL;
1491 }
1492 if (q->handle == 0) {
1493 NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1494 return -ENOENT;
1495 }
1496 err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1497 if (err != 0)
1498 return err;
1499 } else {
1500 qdisc_notify(net, skb, n, clid, NULL, q);
1501 }
1502 return 0;
1503}
1504
1505/*
1506 * Create/change qdisc.
1507 */
1508
1509static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1510 struct netlink_ext_ack *extack)
1511{
1512 struct net *net = sock_net(skb->sk);
1513 struct tcmsg *tcm;
1514 struct nlattr *tca[TCA_MAX + 1];
1515 struct net_device *dev;
1516 u32 clid;
1517 struct Qdisc *q, *p;
1518 int err;
1519
1520 if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1521 return -EPERM;
1522
1523replay:
1524 /* Reinit, just in case something touches this. */
1525 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1526 rtm_tca_policy, extack);
1527 if (err < 0)
1528 return err;
1529
1530 tcm = nlmsg_data(n);
1531 clid = tcm->tcm_parent;
1532 q = p = NULL;
1533
1534 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1535 if (!dev)
1536 return -ENODEV;
1537
1538
1539 if (clid) {
1540 if (clid != TC_H_ROOT) {
1541 if (clid != TC_H_INGRESS) {
1542 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1543 if (!p) {
1544 NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1545 return -ENOENT;
1546 }
1547 q = qdisc_leaf(p, clid);
1548 } else if (dev_ingress_queue_create(dev)) {
1549 q = dev_ingress_queue(dev)->qdisc_sleeping;
1550 }
1551 } else {
1552 q = dev->qdisc;
1553 }
1554
1555 /* It may be default qdisc, ignore it */
1556 if (q && q->handle == 0)
1557 q = NULL;
1558
1559 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1560 if (tcm->tcm_handle) {
1561 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1562 NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1563 return -EEXIST;
1564 }
1565 if (TC_H_MIN(tcm->tcm_handle)) {
1566 NL_SET_ERR_MSG(extack, "Invalid minor handle");
1567 return -EINVAL;
1568 }
1569 q = qdisc_lookup(dev, tcm->tcm_handle);
1570 if (!q)
1571 goto create_n_graft;
1572 if (n->nlmsg_flags & NLM_F_EXCL) {
1573 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1574 return -EEXIST;
1575 }
1576 if (tca[TCA_KIND] &&
1577 nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1578 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1579 return -EINVAL;
1580 }
1581 if (q == p ||
1582 (p && check_loop(q, p, 0))) {
1583 NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1584 return -ELOOP;
1585 }
1586 qdisc_refcount_inc(q);
1587 goto graft;
1588 } else {
1589 if (!q)
1590 goto create_n_graft;
1591
1592 /* This magic test requires explanation.
1593 *
1594 * We know, that some child q is already
1595 * attached to this parent and have choice:
1596 * either to change it or to create/graft new one.
1597 *
1598 * 1. We are allowed to create/graft only
1599 * if CREATE and REPLACE flags are set.
1600 *
1601 * 2. If EXCL is set, requestor wanted to say,
1602 * that qdisc tcm_handle is not expected
1603 * to exist, so that we choose create/graft too.
1604 *
1605 * 3. The last case is when no flags are set.
1606 * Alas, it is sort of hole in API, we
1607 * cannot decide what to do unambiguously.
1608 * For now we select create/graft, if
1609 * user gave KIND, which does not match existing.
1610 */
1611 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1612 (n->nlmsg_flags & NLM_F_REPLACE) &&
1613 ((n->nlmsg_flags & NLM_F_EXCL) ||
1614 (tca[TCA_KIND] &&
1615 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1616 goto create_n_graft;
1617 }
1618 }
1619 } else {
1620 if (!tcm->tcm_handle) {
1621 NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1622 return -EINVAL;
1623 }
1624 q = qdisc_lookup(dev, tcm->tcm_handle);
1625 }
1626
1627 /* Change qdisc parameters */
1628 if (!q) {
1629 NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1630 return -ENOENT;
1631 }
1632 if (n->nlmsg_flags & NLM_F_EXCL) {
1633 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1634 return -EEXIST;
1635 }
1636 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1637 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1638 return -EINVAL;
1639 }
1640 err = qdisc_change(q, tca, extack);
1641 if (err == 0)
1642 qdisc_notify(net, skb, n, clid, NULL, q);
1643 return err;
1644
1645create_n_graft:
1646 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1647 NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1648 return -ENOENT;
1649 }
1650 if (clid == TC_H_INGRESS) {
1651 if (dev_ingress_queue(dev)) {
1652 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1653 tcm->tcm_parent, tcm->tcm_parent,
1654 tca, &err, extack);
1655 } else {
1656 NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1657 err = -ENOENT;
1658 }
1659 } else {
1660 struct netdev_queue *dev_queue;
1661
1662 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1663 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1664 else if (p)
1665 dev_queue = p->dev_queue;
1666 else
1667 dev_queue = netdev_get_tx_queue(dev, 0);
1668
1669 q = qdisc_create(dev, dev_queue, p,
1670 tcm->tcm_parent, tcm->tcm_handle,
1671 tca, &err, extack);
1672 }
1673 if (q == NULL) {
1674 if (err == -EAGAIN)
1675 goto replay;
1676 return err;
1677 }
1678
1679graft:
1680 err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1681 if (err) {
1682 if (q)
1683 qdisc_put(q);
1684 return err;
1685 }
1686
1687 return 0;
1688}
1689
1690static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1691 struct netlink_callback *cb,
1692 int *q_idx_p, int s_q_idx, bool recur,
1693 bool dump_invisible)
1694{
1695 int ret = 0, q_idx = *q_idx_p;
1696 struct Qdisc *q;
1697 int b;
1698
1699 if (!root)
1700 return 0;
1701
1702 q = root;
1703 if (q_idx < s_q_idx) {
1704 q_idx++;
1705 } else {
1706 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1707 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1708 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1709 RTM_NEWQDISC) <= 0)
1710 goto done;
1711 q_idx++;
1712 }
1713
1714 /* If dumping singletons, there is no qdisc_dev(root) and the singleton
1715 * itself has already been dumped.
1716 *
1717 * If we've already dumped the top-level (ingress) qdisc above and the global
1718 * qdisc hashtable, we don't want to hit it again
1719 */
1720 if (!qdisc_dev(root) || !recur)
1721 goto out;
1722
1723 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1724 if (q_idx < s_q_idx) {
1725 q_idx++;
1726 continue;
1727 }
1728 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1729 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1730 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1731 RTM_NEWQDISC) <= 0)
1732 goto done;
1733 q_idx++;
1734 }
1735
1736out:
1737 *q_idx_p = q_idx;
1738 return ret;
1739done:
1740 ret = -1;
1741 goto out;
1742}
1743
1744static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1745{
1746 struct net *net = sock_net(skb->sk);
1747 int idx, q_idx;
1748 int s_idx, s_q_idx;
1749 struct net_device *dev;
1750 const struct nlmsghdr *nlh = cb->nlh;
1751 struct nlattr *tca[TCA_MAX + 1];
1752 int err;
1753
1754 s_idx = cb->args[0];
1755 s_q_idx = q_idx = cb->args[1];
1756
1757 idx = 0;
1758 ASSERT_RTNL();
1759
1760 err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1761 rtm_tca_policy, cb->extack);
1762 if (err < 0)
1763 return err;
1764
1765 for_each_netdev(net, dev) {
1766 struct netdev_queue *dev_queue;
1767
1768 if (idx < s_idx)
1769 goto cont;
1770 if (idx > s_idx)
1771 s_q_idx = 0;
1772 q_idx = 0;
1773
1774 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1775 true, tca[TCA_DUMP_INVISIBLE]) < 0)
1776 goto done;
1777
1778 dev_queue = dev_ingress_queue(dev);
1779 if (dev_queue &&
1780 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1781 &q_idx, s_q_idx, false,
1782 tca[TCA_DUMP_INVISIBLE]) < 0)
1783 goto done;
1784
1785cont:
1786 idx++;
1787 }
1788
1789done:
1790 cb->args[0] = idx;
1791 cb->args[1] = q_idx;
1792
1793 return skb->len;
1794}
1795
1796
1797
1798/************************************************
1799 * Traffic classes manipulation. *
1800 ************************************************/
1801
1802static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1803 unsigned long cl,
1804 u32 portid, u32 seq, u16 flags, int event)
1805{
1806 struct tcmsg *tcm;
1807 struct nlmsghdr *nlh;
1808 unsigned char *b = skb_tail_pointer(skb);
1809 struct gnet_dump d;
1810 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1811
1812 cond_resched();
1813 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1814 if (!nlh)
1815 goto out_nlmsg_trim;
1816 tcm = nlmsg_data(nlh);
1817 tcm->tcm_family = AF_UNSPEC;
1818 tcm->tcm__pad1 = 0;
1819 tcm->tcm__pad2 = 0;
1820 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1821 tcm->tcm_parent = q->handle;
1822 tcm->tcm_handle = q->handle;
1823 tcm->tcm_info = 0;
1824 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1825 goto nla_put_failure;
1826 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1827 goto nla_put_failure;
1828
1829 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1830 NULL, &d, TCA_PAD) < 0)
1831 goto nla_put_failure;
1832
1833 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1834 goto nla_put_failure;
1835
1836 if (gnet_stats_finish_copy(&d) < 0)
1837 goto nla_put_failure;
1838
1839 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1840 return skb->len;
1841
1842out_nlmsg_trim:
1843nla_put_failure:
1844 nlmsg_trim(skb, b);
1845 return -1;
1846}
1847
1848static int tclass_notify(struct net *net, struct sk_buff *oskb,
1849 struct nlmsghdr *n, struct Qdisc *q,
1850 unsigned long cl, int event)
1851{
1852 struct sk_buff *skb;
1853 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1854 int err = 0;
1855
1856 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1857 if (!skb)
1858 return -ENOBUFS;
1859
1860 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1861 kfree_skb(skb);
1862 return -EINVAL;
1863 }
1864
1865 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1866 n->nlmsg_flags & NLM_F_ECHO);
1867 if (err > 0)
1868 err = 0;
1869 return err;
1870}
1871
1872static int tclass_del_notify(struct net *net,
1873 const struct Qdisc_class_ops *cops,
1874 struct sk_buff *oskb, struct nlmsghdr *n,
1875 struct Qdisc *q, unsigned long cl,
1876 struct netlink_ext_ack *extack)
1877{
1878 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1879 struct sk_buff *skb;
1880 int err = 0;
1881
1882 if (!cops->delete)
1883 return -EOPNOTSUPP;
1884
1885 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1886 if (!skb)
1887 return -ENOBUFS;
1888
1889 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1890 RTM_DELTCLASS) < 0) {
1891 kfree_skb(skb);
1892 return -EINVAL;
1893 }
1894
1895 err = cops->delete(q, cl, extack);
1896 if (err) {
1897 kfree_skb(skb);
1898 return err;
1899 }
1900
1901 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1902 n->nlmsg_flags & NLM_F_ECHO);
1903 if (err > 0)
1904 err = 0;
1905 return err;
1906}
1907
1908#ifdef CONFIG_NET_CLS
1909
1910struct tcf_bind_args {
1911 struct tcf_walker w;
1912 unsigned long base;
1913 unsigned long cl;
1914 u32 classid;
1915};
1916
1917static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1918{
1919 struct tcf_bind_args *a = (void *)arg;
1920
1921 if (tp->ops->bind_class) {
1922 struct Qdisc *q = tcf_block_q(tp->chain->block);
1923
1924 sch_tree_lock(q);
1925 tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1926 sch_tree_unlock(q);
1927 }
1928 return 0;
1929}
1930
1931struct tc_bind_class_args {
1932 struct qdisc_walker w;
1933 unsigned long new_cl;
1934 u32 portid;
1935 u32 clid;
1936};
1937
1938static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1939 struct qdisc_walker *w)
1940{
1941 struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1942 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1943 struct tcf_block *block;
1944 struct tcf_chain *chain;
1945
1946 block = cops->tcf_block(q, cl, NULL);
1947 if (!block)
1948 return 0;
1949 for (chain = tcf_get_next_chain(block, NULL);
1950 chain;
1951 chain = tcf_get_next_chain(block, chain)) {
1952 struct tcf_proto *tp;
1953
1954 for (tp = tcf_get_next_proto(chain, NULL);
1955 tp; tp = tcf_get_next_proto(chain, tp)) {
1956 struct tcf_bind_args arg = {};
1957
1958 arg.w.fn = tcf_node_bind;
1959 arg.classid = a->clid;
1960 arg.base = cl;
1961 arg.cl = a->new_cl;
1962 tp->ops->walk(tp, &arg.w, true);
1963 }
1964 }
1965
1966 return 0;
1967}
1968
1969static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1970 unsigned long new_cl)
1971{
1972 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1973 struct tc_bind_class_args args = {};
1974
1975 if (!cops->tcf_block)
1976 return;
1977 args.portid = portid;
1978 args.clid = clid;
1979 args.new_cl = new_cl;
1980 args.w.fn = tc_bind_class_walker;
1981 q->ops->cl_ops->walk(q, &args.w);
1982}
1983
1984#else
1985
1986static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1987 unsigned long new_cl)
1988{
1989}
1990
1991#endif
1992
1993static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1994 struct netlink_ext_ack *extack)
1995{
1996 struct net *net = sock_net(skb->sk);
1997 struct tcmsg *tcm = nlmsg_data(n);
1998 struct nlattr *tca[TCA_MAX + 1];
1999 struct net_device *dev;
2000 struct Qdisc *q = NULL;
2001 const struct Qdisc_class_ops *cops;
2002 unsigned long cl = 0;
2003 unsigned long new_cl;
2004 u32 portid;
2005 u32 clid;
2006 u32 qid;
2007 int err;
2008
2009 if ((n->nlmsg_type != RTM_GETTCLASS) &&
2010 !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2011 return -EPERM;
2012
2013 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2014 rtm_tca_policy, extack);
2015 if (err < 0)
2016 return err;
2017
2018 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2019 if (!dev)
2020 return -ENODEV;
2021
2022 /*
2023 parent == TC_H_UNSPEC - unspecified parent.
2024 parent == TC_H_ROOT - class is root, which has no parent.
2025 parent == X:0 - parent is root class.
2026 parent == X:Y - parent is a node in hierarchy.
2027 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
2028
2029 handle == 0:0 - generate handle from kernel pool.
2030 handle == 0:Y - class is X:Y, where X:0 is qdisc.
2031 handle == X:Y - clear.
2032 handle == X:0 - root class.
2033 */
2034
2035 /* Step 1. Determine qdisc handle X:0 */
2036
2037 portid = tcm->tcm_parent;
2038 clid = tcm->tcm_handle;
2039 qid = TC_H_MAJ(clid);
2040
2041 if (portid != TC_H_ROOT) {
2042 u32 qid1 = TC_H_MAJ(portid);
2043
2044 if (qid && qid1) {
2045 /* If both majors are known, they must be identical. */
2046 if (qid != qid1)
2047 return -EINVAL;
2048 } else if (qid1) {
2049 qid = qid1;
2050 } else if (qid == 0)
2051 qid = dev->qdisc->handle;
2052
2053 /* Now qid is genuine qdisc handle consistent
2054 * both with parent and child.
2055 *
2056 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2057 */
2058 if (portid)
2059 portid = TC_H_MAKE(qid, portid);
2060 } else {
2061 if (qid == 0)
2062 qid = dev->qdisc->handle;
2063 }
2064
2065 /* OK. Locate qdisc */
2066 q = qdisc_lookup(dev, qid);
2067 if (!q)
2068 return -ENOENT;
2069
2070 /* An check that it supports classes */
2071 cops = q->ops->cl_ops;
2072 if (cops == NULL)
2073 return -EINVAL;
2074
2075 /* Now try to get class */
2076 if (clid == 0) {
2077 if (portid == TC_H_ROOT)
2078 clid = qid;
2079 } else
2080 clid = TC_H_MAKE(qid, clid);
2081
2082 if (clid)
2083 cl = cops->find(q, clid);
2084
2085 if (cl == 0) {
2086 err = -ENOENT;
2087 if (n->nlmsg_type != RTM_NEWTCLASS ||
2088 !(n->nlmsg_flags & NLM_F_CREATE))
2089 goto out;
2090 } else {
2091 switch (n->nlmsg_type) {
2092 case RTM_NEWTCLASS:
2093 err = -EEXIST;
2094 if (n->nlmsg_flags & NLM_F_EXCL)
2095 goto out;
2096 break;
2097 case RTM_DELTCLASS:
2098 err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2099 /* Unbind the class with flilters with 0 */
2100 tc_bind_tclass(q, portid, clid, 0);
2101 goto out;
2102 case RTM_GETTCLASS:
2103 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
2104 goto out;
2105 default:
2106 err = -EINVAL;
2107 goto out;
2108 }
2109 }
2110
2111 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2112 NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2113 return -EOPNOTSUPP;
2114 }
2115
2116 new_cl = cl;
2117 err = -EOPNOTSUPP;
2118 if (cops->change)
2119 err = cops->change(q, clid, portid, tca, &new_cl, extack);
2120 if (err == 0) {
2121 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
2122 /* We just create a new class, need to do reverse binding. */
2123 if (cl != new_cl)
2124 tc_bind_tclass(q, portid, clid, new_cl);
2125 }
2126out:
2127 return err;
2128}
2129
2130struct qdisc_dump_args {
2131 struct qdisc_walker w;
2132 struct sk_buff *skb;
2133 struct netlink_callback *cb;
2134};
2135
2136static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2137 struct qdisc_walker *arg)
2138{
2139 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2140
2141 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2142 a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2143 RTM_NEWTCLASS);
2144}
2145
2146static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2147 struct tcmsg *tcm, struct netlink_callback *cb,
2148 int *t_p, int s_t)
2149{
2150 struct qdisc_dump_args arg;
2151
2152 if (tc_qdisc_dump_ignore(q, false) ||
2153 *t_p < s_t || !q->ops->cl_ops ||
2154 (tcm->tcm_parent &&
2155 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2156 (*t_p)++;
2157 return 0;
2158 }
2159 if (*t_p > s_t)
2160 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2161 arg.w.fn = qdisc_class_dump;
2162 arg.skb = skb;
2163 arg.cb = cb;
2164 arg.w.stop = 0;
2165 arg.w.skip = cb->args[1];
2166 arg.w.count = 0;
2167 q->ops->cl_ops->walk(q, &arg.w);
2168 cb->args[1] = arg.w.count;
2169 if (arg.w.stop)
2170 return -1;
2171 (*t_p)++;
2172 return 0;
2173}
2174
2175static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2176 struct tcmsg *tcm, struct netlink_callback *cb,
2177 int *t_p, int s_t, bool recur)
2178{
2179 struct Qdisc *q;
2180 int b;
2181
2182 if (!root)
2183 return 0;
2184
2185 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2186 return -1;
2187
2188 if (!qdisc_dev(root) || !recur)
2189 return 0;
2190
2191 if (tcm->tcm_parent) {
2192 q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2193 if (q && q != root &&
2194 tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2195 return -1;
2196 return 0;
2197 }
2198 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2199 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2200 return -1;
2201 }
2202
2203 return 0;
2204}
2205
2206static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2207{
2208 struct tcmsg *tcm = nlmsg_data(cb->nlh);
2209 struct net *net = sock_net(skb->sk);
2210 struct netdev_queue *dev_queue;
2211 struct net_device *dev;
2212 int t, s_t;
2213
2214 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2215 return 0;
2216 dev = dev_get_by_index(net, tcm->tcm_ifindex);
2217 if (!dev)
2218 return 0;
2219
2220 s_t = cb->args[0];
2221 t = 0;
2222
2223 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t, true) < 0)
2224 goto done;
2225
2226 dev_queue = dev_ingress_queue(dev);
2227 if (dev_queue &&
2228 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2229 &t, s_t, false) < 0)
2230 goto done;
2231
2232done:
2233 cb->args[0] = t;
2234
2235 dev_put(dev);
2236 return skb->len;
2237}
2238
2239#ifdef CONFIG_PROC_FS
2240static int psched_show(struct seq_file *seq, void *v)
2241{
2242 seq_printf(seq, "%08x %08x %08x %08x\n",
2243 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2244 1000000,
2245 (u32)NSEC_PER_SEC / hrtimer_resolution);
2246
2247 return 0;
2248}
2249
2250static int __net_init psched_net_init(struct net *net)
2251{
2252 struct proc_dir_entry *e;
2253
2254 e = proc_create_single("psched", 0, net->proc_net, psched_show);
2255 if (e == NULL)
2256 return -ENOMEM;
2257
2258 return 0;
2259}
2260
2261static void __net_exit psched_net_exit(struct net *net)
2262{
2263 remove_proc_entry("psched", net->proc_net);
2264}
2265#else
2266static int __net_init psched_net_init(struct net *net)
2267{
2268 return 0;
2269}
2270
2271static void __net_exit psched_net_exit(struct net *net)
2272{
2273}
2274#endif
2275
2276static struct pernet_operations psched_net_ops = {
2277 .init = psched_net_init,
2278 .exit = psched_net_exit,
2279};
2280
2281static int __init pktsched_init(void)
2282{
2283 int err;
2284
2285 err = register_pernet_subsys(&psched_net_ops);
2286 if (err) {
2287 pr_err("pktsched_init: "
2288 "cannot initialize per netns operations\n");
2289 return err;
2290 }
2291
2292 register_qdisc(&pfifo_fast_ops);
2293 register_qdisc(&pfifo_qdisc_ops);
2294 register_qdisc(&bfifo_qdisc_ops);
2295 register_qdisc(&pfifo_head_drop_qdisc_ops);
2296 register_qdisc(&mq_qdisc_ops);
2297 register_qdisc(&noqueue_qdisc_ops);
2298
2299 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2300 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2301 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2302 0);
2303 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2304 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2305 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2306 0);
2307
2308 return 0;
2309}
2310
2311subsys_initcall(pktsched_init);