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#include <net/tc_wrapper.h>
35
36#include <trace/events/qdisc.h>
37
38/*
39
40 Short review.
41 -------------
42
43 This file consists of two interrelated parts:
44
45 1. queueing disciplines manager frontend.
46 2. traffic classes manager frontend.
47
48 Generally, queueing discipline ("qdisc") is a black box,
49 which is able to enqueue packets and to dequeue them (when
50 device is ready to send something) in order and at times
51 determined by algorithm hidden in it.
52
53 qdisc's are divided to two categories:
54 - "queues", which have no internal structure visible from outside.
55 - "schedulers", which split all the packets to "traffic classes",
56 using "packet classifiers" (look at cls_api.c)
57
58 In turn, classes may have child qdiscs (as rule, queues)
59 attached to them etc. etc. etc.
60
61 The goal of the routines in this file is to translate
62 information supplied by user in the form of handles
63 to more intelligible for kernel form, to make some sanity
64 checks and part of work, which is common to all qdiscs
65 and to provide rtnetlink notifications.
66
67 All real intelligent work is done inside qdisc modules.
68
69
70
71 Every discipline has two major routines: enqueue and dequeue.
72
73 ---dequeue
74
75 dequeue usually returns a skb to send. It is allowed to return NULL,
76 but it does not mean that queue is empty, it just means that
77 discipline does not want to send anything this time.
78 Queue is really empty if q->q.qlen == 0.
79 For complicated disciplines with multiple queues q->q is not
80 real packet queue, but however q->q.qlen must be valid.
81
82 ---enqueue
83
84 enqueue returns 0, if packet was enqueued successfully.
85 If packet (this one or another one) was dropped, it returns
86 not zero error code.
87 NET_XMIT_DROP - this packet dropped
88 Expected action: do not backoff, but wait until queue will clear.
89 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
90 Expected action: backoff or ignore
91
92 Auxiliary routines:
93
94 ---peek
95
96 like dequeue but without removing a packet from the queue
97
98 ---reset
99
100 returns qdisc to initial state: purge all buffers, clear all
101 timers, counters (except for statistics) etc.
102
103 ---init
104
105 initializes newly created qdisc.
106
107 ---destroy
108
109 destroys resources allocated by init and during lifetime of qdisc.
110
111 ---change
112
113 changes qdisc parameters.
114 */
115
116/* Protects list of registered TC modules. It is pure SMP lock. */
117static DEFINE_RWLOCK(qdisc_mod_lock);
118
119
120/************************************************
121 * Queueing disciplines manipulation. *
122 ************************************************/
123
124
125/* The list of all installed queueing disciplines. */
126
127static struct Qdisc_ops *qdisc_base;
128
129/* Register/unregister queueing discipline */
130
131int register_qdisc(struct Qdisc_ops *qops)
132{
133 struct Qdisc_ops *q, **qp;
134 int rc = -EEXIST;
135
136 write_lock(&qdisc_mod_lock);
137 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
138 if (!strcmp(qops->id, q->id))
139 goto out;
140
141 if (qops->enqueue == NULL)
142 qops->enqueue = noop_qdisc_ops.enqueue;
143 if (qops->peek == NULL) {
144 if (qops->dequeue == NULL)
145 qops->peek = noop_qdisc_ops.peek;
146 else
147 goto out_einval;
148 }
149 if (qops->dequeue == NULL)
150 qops->dequeue = noop_qdisc_ops.dequeue;
151
152 if (qops->cl_ops) {
153 const struct Qdisc_class_ops *cops = qops->cl_ops;
154
155 if (!(cops->find && cops->walk && cops->leaf))
156 goto out_einval;
157
158 if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
159 goto out_einval;
160 }
161
162 qops->next = NULL;
163 *qp = qops;
164 rc = 0;
165out:
166 write_unlock(&qdisc_mod_lock);
167 return rc;
168
169out_einval:
170 rc = -EINVAL;
171 goto out;
172}
173EXPORT_SYMBOL(register_qdisc);
174
175void unregister_qdisc(struct Qdisc_ops *qops)
176{
177 struct Qdisc_ops *q, **qp;
178 int err = -ENOENT;
179
180 write_lock(&qdisc_mod_lock);
181 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
182 if (q == qops)
183 break;
184 if (q) {
185 *qp = q->next;
186 q->next = NULL;
187 err = 0;
188 }
189 write_unlock(&qdisc_mod_lock);
190
191 WARN(err, "unregister qdisc(%s) failed\n", qops->id);
192}
193EXPORT_SYMBOL(unregister_qdisc);
194
195/* Get default qdisc if not otherwise specified */
196void qdisc_get_default(char *name, size_t len)
197{
198 read_lock(&qdisc_mod_lock);
199 strscpy(name, default_qdisc_ops->id, len);
200 read_unlock(&qdisc_mod_lock);
201}
202
203static struct Qdisc_ops *qdisc_lookup_default(const char *name)
204{
205 struct Qdisc_ops *q = NULL;
206
207 for (q = qdisc_base; q; q = q->next) {
208 if (!strcmp(name, q->id)) {
209 if (!try_module_get(q->owner))
210 q = NULL;
211 break;
212 }
213 }
214
215 return q;
216}
217
218/* Set new default qdisc to use */
219int qdisc_set_default(const char *name)
220{
221 const struct Qdisc_ops *ops;
222
223 if (!capable(CAP_NET_ADMIN))
224 return -EPERM;
225
226 write_lock(&qdisc_mod_lock);
227 ops = qdisc_lookup_default(name);
228 if (!ops) {
229 /* Not found, drop lock and try to load module */
230 write_unlock(&qdisc_mod_lock);
231 request_module("sch_%s", name);
232 write_lock(&qdisc_mod_lock);
233
234 ops = qdisc_lookup_default(name);
235 }
236
237 if (ops) {
238 /* Set new default */
239 module_put(default_qdisc_ops->owner);
240 default_qdisc_ops = ops;
241 }
242 write_unlock(&qdisc_mod_lock);
243
244 return ops ? 0 : -ENOENT;
245}
246
247#ifdef CONFIG_NET_SCH_DEFAULT
248/* Set default value from kernel config */
249static int __init sch_default_qdisc(void)
250{
251 return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
252}
253late_initcall(sch_default_qdisc);
254#endif
255
256/* We know handle. Find qdisc among all qdisc's attached to device
257 * (root qdisc, all its children, children of children etc.)
258 * Note: caller either uses rtnl or rcu_read_lock()
259 */
260
261static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
262{
263 struct Qdisc *q;
264
265 if (!qdisc_dev(root))
266 return (root->handle == handle ? root : NULL);
267
268 if (!(root->flags & TCQ_F_BUILTIN) &&
269 root->handle == handle)
270 return root;
271
272 hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle,
273 lockdep_rtnl_is_held()) {
274 if (q->handle == handle)
275 return q;
276 }
277 return NULL;
278}
279
280void qdisc_hash_add(struct Qdisc *q, bool invisible)
281{
282 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
283 ASSERT_RTNL();
284 hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
285 if (invisible)
286 q->flags |= TCQ_F_INVISIBLE;
287 }
288}
289EXPORT_SYMBOL(qdisc_hash_add);
290
291void qdisc_hash_del(struct Qdisc *q)
292{
293 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
294 ASSERT_RTNL();
295 hash_del_rcu(&q->hash);
296 }
297}
298EXPORT_SYMBOL(qdisc_hash_del);
299
300struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
301{
302 struct Qdisc *q;
303
304 if (!handle)
305 return NULL;
306 q = qdisc_match_from_root(rtnl_dereference(dev->qdisc), handle);
307 if (q)
308 goto out;
309
310 if (dev_ingress_queue(dev))
311 q = qdisc_match_from_root(
312 dev_ingress_queue(dev)->qdisc_sleeping,
313 handle);
314out:
315 return q;
316}
317
318struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
319{
320 struct netdev_queue *nq;
321 struct Qdisc *q;
322
323 if (!handle)
324 return NULL;
325 q = qdisc_match_from_root(rcu_dereference(dev->qdisc), handle);
326 if (q)
327 goto out;
328
329 nq = dev_ingress_queue_rcu(dev);
330 if (nq)
331 q = qdisc_match_from_root(nq->qdisc_sleeping, handle);
332out:
333 return q;
334}
335
336static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
337{
338 unsigned long cl;
339 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
340
341 if (cops == NULL)
342 return NULL;
343 cl = cops->find(p, classid);
344
345 if (cl == 0)
346 return NULL;
347 return cops->leaf(p, cl);
348}
349
350/* Find queueing discipline by name */
351
352static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
353{
354 struct Qdisc_ops *q = NULL;
355
356 if (kind) {
357 read_lock(&qdisc_mod_lock);
358 for (q = qdisc_base; q; q = q->next) {
359 if (nla_strcmp(kind, q->id) == 0) {
360 if (!try_module_get(q->owner))
361 q = NULL;
362 break;
363 }
364 }
365 read_unlock(&qdisc_mod_lock);
366 }
367 return q;
368}
369
370/* The linklayer setting were not transferred from iproute2, in older
371 * versions, and the rate tables lookup systems have been dropped in
372 * the kernel. To keep backward compatible with older iproute2 tc
373 * utils, we detect the linklayer setting by detecting if the rate
374 * table were modified.
375 *
376 * For linklayer ATM table entries, the rate table will be aligned to
377 * 48 bytes, thus some table entries will contain the same value. The
378 * mpu (min packet unit) is also encoded into the old rate table, thus
379 * starting from the mpu, we find low and high table entries for
380 * mapping this cell. If these entries contain the same value, when
381 * the rate tables have been modified for linklayer ATM.
382 *
383 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
384 * and then roundup to the next cell, calc the table entry one below,
385 * and compare.
386 */
387static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
388{
389 int low = roundup(r->mpu, 48);
390 int high = roundup(low+1, 48);
391 int cell_low = low >> r->cell_log;
392 int cell_high = (high >> r->cell_log) - 1;
393
394 /* rtab is too inaccurate at rates > 100Mbit/s */
395 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
396 pr_debug("TC linklayer: Giving up ATM detection\n");
397 return TC_LINKLAYER_ETHERNET;
398 }
399
400 if ((cell_high > cell_low) && (cell_high < 256)
401 && (rtab[cell_low] == rtab[cell_high])) {
402 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
403 cell_low, cell_high, rtab[cell_high]);
404 return TC_LINKLAYER_ATM;
405 }
406 return TC_LINKLAYER_ETHERNET;
407}
408
409static struct qdisc_rate_table *qdisc_rtab_list;
410
411struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
412 struct nlattr *tab,
413 struct netlink_ext_ack *extack)
414{
415 struct qdisc_rate_table *rtab;
416
417 if (tab == NULL || r->rate == 0 ||
418 r->cell_log == 0 || r->cell_log >= 32 ||
419 nla_len(tab) != TC_RTAB_SIZE) {
420 NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
421 return NULL;
422 }
423
424 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
425 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
426 !memcmp(&rtab->data, nla_data(tab), 1024)) {
427 rtab->refcnt++;
428 return rtab;
429 }
430 }
431
432 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
433 if (rtab) {
434 rtab->rate = *r;
435 rtab->refcnt = 1;
436 memcpy(rtab->data, nla_data(tab), 1024);
437 if (r->linklayer == TC_LINKLAYER_UNAWARE)
438 r->linklayer = __detect_linklayer(r, rtab->data);
439 rtab->next = qdisc_rtab_list;
440 qdisc_rtab_list = rtab;
441 } else {
442 NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
443 }
444 return rtab;
445}
446EXPORT_SYMBOL(qdisc_get_rtab);
447
448void qdisc_put_rtab(struct qdisc_rate_table *tab)
449{
450 struct qdisc_rate_table *rtab, **rtabp;
451
452 if (!tab || --tab->refcnt)
453 return;
454
455 for (rtabp = &qdisc_rtab_list;
456 (rtab = *rtabp) != NULL;
457 rtabp = &rtab->next) {
458 if (rtab == tab) {
459 *rtabp = rtab->next;
460 kfree(rtab);
461 return;
462 }
463 }
464}
465EXPORT_SYMBOL(qdisc_put_rtab);
466
467static LIST_HEAD(qdisc_stab_list);
468
469static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
470 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
471 [TCA_STAB_DATA] = { .type = NLA_BINARY },
472};
473
474static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
475 struct netlink_ext_ack *extack)
476{
477 struct nlattr *tb[TCA_STAB_MAX + 1];
478 struct qdisc_size_table *stab;
479 struct tc_sizespec *s;
480 unsigned int tsize = 0;
481 u16 *tab = NULL;
482 int err;
483
484 err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
485 extack);
486 if (err < 0)
487 return ERR_PTR(err);
488 if (!tb[TCA_STAB_BASE]) {
489 NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
490 return ERR_PTR(-EINVAL);
491 }
492
493 s = nla_data(tb[TCA_STAB_BASE]);
494
495 if (s->tsize > 0) {
496 if (!tb[TCA_STAB_DATA]) {
497 NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
498 return ERR_PTR(-EINVAL);
499 }
500 tab = nla_data(tb[TCA_STAB_DATA]);
501 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
502 }
503
504 if (tsize != s->tsize || (!tab && tsize > 0)) {
505 NL_SET_ERR_MSG(extack, "Invalid size of size table");
506 return ERR_PTR(-EINVAL);
507 }
508
509 list_for_each_entry(stab, &qdisc_stab_list, list) {
510 if (memcmp(&stab->szopts, s, sizeof(*s)))
511 continue;
512 if (tsize > 0 &&
513 memcmp(stab->data, tab, flex_array_size(stab, data, tsize)))
514 continue;
515 stab->refcnt++;
516 return stab;
517 }
518
519 if (s->size_log > STAB_SIZE_LOG_MAX ||
520 s->cell_log > STAB_SIZE_LOG_MAX) {
521 NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
522 return ERR_PTR(-EINVAL);
523 }
524
525 stab = kmalloc(struct_size(stab, data, tsize), GFP_KERNEL);
526 if (!stab)
527 return ERR_PTR(-ENOMEM);
528
529 stab->refcnt = 1;
530 stab->szopts = *s;
531 if (tsize > 0)
532 memcpy(stab->data, tab, flex_array_size(stab, data, tsize));
533
534 list_add_tail(&stab->list, &qdisc_stab_list);
535
536 return stab;
537}
538
539void qdisc_put_stab(struct qdisc_size_table *tab)
540{
541 if (!tab)
542 return;
543
544 if (--tab->refcnt == 0) {
545 list_del(&tab->list);
546 kfree_rcu(tab, rcu);
547 }
548}
549EXPORT_SYMBOL(qdisc_put_stab);
550
551static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
552{
553 struct nlattr *nest;
554
555 nest = nla_nest_start_noflag(skb, TCA_STAB);
556 if (nest == NULL)
557 goto nla_put_failure;
558 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
559 goto nla_put_failure;
560 nla_nest_end(skb, nest);
561
562 return skb->len;
563
564nla_put_failure:
565 return -1;
566}
567
568void __qdisc_calculate_pkt_len(struct sk_buff *skb,
569 const struct qdisc_size_table *stab)
570{
571 int pkt_len, slot;
572
573 pkt_len = skb->len + stab->szopts.overhead;
574 if (unlikely(!stab->szopts.tsize))
575 goto out;
576
577 slot = pkt_len + stab->szopts.cell_align;
578 if (unlikely(slot < 0))
579 slot = 0;
580
581 slot >>= stab->szopts.cell_log;
582 if (likely(slot < stab->szopts.tsize))
583 pkt_len = stab->data[slot];
584 else
585 pkt_len = stab->data[stab->szopts.tsize - 1] *
586 (slot / stab->szopts.tsize) +
587 stab->data[slot % stab->szopts.tsize];
588
589 pkt_len <<= stab->szopts.size_log;
590out:
591 if (unlikely(pkt_len < 1))
592 pkt_len = 1;
593 qdisc_skb_cb(skb)->pkt_len = pkt_len;
594}
595EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
596
597void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
598{
599 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
600 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
601 txt, qdisc->ops->id, qdisc->handle >> 16);
602 qdisc->flags |= TCQ_F_WARN_NONWC;
603 }
604}
605EXPORT_SYMBOL(qdisc_warn_nonwc);
606
607static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
608{
609 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
610 timer);
611
612 rcu_read_lock();
613 __netif_schedule(qdisc_root(wd->qdisc));
614 rcu_read_unlock();
615
616 return HRTIMER_NORESTART;
617}
618
619void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
620 clockid_t clockid)
621{
622 hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
623 wd->timer.function = qdisc_watchdog;
624 wd->qdisc = qdisc;
625}
626EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
627
628void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
629{
630 qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
631}
632EXPORT_SYMBOL(qdisc_watchdog_init);
633
634void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
635 u64 delta_ns)
636{
637 if (test_bit(__QDISC_STATE_DEACTIVATED,
638 &qdisc_root_sleeping(wd->qdisc)->state))
639 return;
640
641 if (hrtimer_is_queued(&wd->timer)) {
642 /* If timer is already set in [expires, expires + delta_ns],
643 * do not reprogram it.
644 */
645 if (wd->last_expires - expires <= delta_ns)
646 return;
647 }
648
649 wd->last_expires = expires;
650 hrtimer_start_range_ns(&wd->timer,
651 ns_to_ktime(expires),
652 delta_ns,
653 HRTIMER_MODE_ABS_PINNED);
654}
655EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
656
657void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
658{
659 hrtimer_cancel(&wd->timer);
660}
661EXPORT_SYMBOL(qdisc_watchdog_cancel);
662
663static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
664{
665 struct hlist_head *h;
666 unsigned int i;
667
668 h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
669
670 if (h != NULL) {
671 for (i = 0; i < n; i++)
672 INIT_HLIST_HEAD(&h[i]);
673 }
674 return h;
675}
676
677void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
678{
679 struct Qdisc_class_common *cl;
680 struct hlist_node *next;
681 struct hlist_head *nhash, *ohash;
682 unsigned int nsize, nmask, osize;
683 unsigned int i, h;
684
685 /* Rehash when load factor exceeds 0.75 */
686 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
687 return;
688 nsize = clhash->hashsize * 2;
689 nmask = nsize - 1;
690 nhash = qdisc_class_hash_alloc(nsize);
691 if (nhash == NULL)
692 return;
693
694 ohash = clhash->hash;
695 osize = clhash->hashsize;
696
697 sch_tree_lock(sch);
698 for (i = 0; i < osize; i++) {
699 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
700 h = qdisc_class_hash(cl->classid, nmask);
701 hlist_add_head(&cl->hnode, &nhash[h]);
702 }
703 }
704 clhash->hash = nhash;
705 clhash->hashsize = nsize;
706 clhash->hashmask = nmask;
707 sch_tree_unlock(sch);
708
709 kvfree(ohash);
710}
711EXPORT_SYMBOL(qdisc_class_hash_grow);
712
713int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
714{
715 unsigned int size = 4;
716
717 clhash->hash = qdisc_class_hash_alloc(size);
718 if (!clhash->hash)
719 return -ENOMEM;
720 clhash->hashsize = size;
721 clhash->hashmask = size - 1;
722 clhash->hashelems = 0;
723 return 0;
724}
725EXPORT_SYMBOL(qdisc_class_hash_init);
726
727void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
728{
729 kvfree(clhash->hash);
730}
731EXPORT_SYMBOL(qdisc_class_hash_destroy);
732
733void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
734 struct Qdisc_class_common *cl)
735{
736 unsigned int h;
737
738 INIT_HLIST_NODE(&cl->hnode);
739 h = qdisc_class_hash(cl->classid, clhash->hashmask);
740 hlist_add_head(&cl->hnode, &clhash->hash[h]);
741 clhash->hashelems++;
742}
743EXPORT_SYMBOL(qdisc_class_hash_insert);
744
745void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
746 struct Qdisc_class_common *cl)
747{
748 hlist_del(&cl->hnode);
749 clhash->hashelems--;
750}
751EXPORT_SYMBOL(qdisc_class_hash_remove);
752
753/* Allocate an unique handle from space managed by kernel
754 * Possible range is [8000-FFFF]:0000 (0x8000 values)
755 */
756static u32 qdisc_alloc_handle(struct net_device *dev)
757{
758 int i = 0x8000;
759 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
760
761 do {
762 autohandle += TC_H_MAKE(0x10000U, 0);
763 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
764 autohandle = TC_H_MAKE(0x80000000U, 0);
765 if (!qdisc_lookup(dev, autohandle))
766 return autohandle;
767 cond_resched();
768 } while (--i > 0);
769
770 return 0;
771}
772
773void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
774{
775 bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
776 const struct Qdisc_class_ops *cops;
777 unsigned long cl;
778 u32 parentid;
779 bool notify;
780 int drops;
781
782 if (n == 0 && len == 0)
783 return;
784 drops = max_t(int, n, 0);
785 rcu_read_lock();
786 while ((parentid = sch->parent)) {
787 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
788 break;
789
790 if (sch->flags & TCQ_F_NOPARENT)
791 break;
792 /* Notify parent qdisc only if child qdisc becomes empty.
793 *
794 * If child was empty even before update then backlog
795 * counter is screwed and we skip notification because
796 * parent class is already passive.
797 *
798 * If the original child was offloaded then it is allowed
799 * to be seem as empty, so the parent is notified anyway.
800 */
801 notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
802 !qdisc_is_offloaded);
803 /* TODO: perform the search on a per txq basis */
804 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
805 if (sch == NULL) {
806 WARN_ON_ONCE(parentid != TC_H_ROOT);
807 break;
808 }
809 cops = sch->ops->cl_ops;
810 if (notify && cops->qlen_notify) {
811 cl = cops->find(sch, parentid);
812 cops->qlen_notify(sch, cl);
813 }
814 sch->q.qlen -= n;
815 sch->qstats.backlog -= len;
816 __qdisc_qstats_drop(sch, drops);
817 }
818 rcu_read_unlock();
819}
820EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
821
822int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
823 void *type_data)
824{
825 struct net_device *dev = qdisc_dev(sch);
826 int err;
827
828 sch->flags &= ~TCQ_F_OFFLOADED;
829 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
830 return 0;
831
832 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
833 if (err == -EOPNOTSUPP)
834 return 0;
835
836 if (!err)
837 sch->flags |= TCQ_F_OFFLOADED;
838
839 return err;
840}
841EXPORT_SYMBOL(qdisc_offload_dump_helper);
842
843void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
844 struct Qdisc *new, struct Qdisc *old,
845 enum tc_setup_type type, void *type_data,
846 struct netlink_ext_ack *extack)
847{
848 bool any_qdisc_is_offloaded;
849 int err;
850
851 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
852 return;
853
854 err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
855
856 /* Don't report error if the graft is part of destroy operation. */
857 if (!err || !new || new == &noop_qdisc)
858 return;
859
860 /* Don't report error if the parent, the old child and the new
861 * one are not offloaded.
862 */
863 any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
864 any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
865 any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
866
867 if (any_qdisc_is_offloaded)
868 NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
869}
870EXPORT_SYMBOL(qdisc_offload_graft_helper);
871
872void qdisc_offload_query_caps(struct net_device *dev,
873 enum tc_setup_type type,
874 void *caps, size_t caps_len)
875{
876 const struct net_device_ops *ops = dev->netdev_ops;
877 struct tc_query_caps_base base = {
878 .type = type,
879 .caps = caps,
880 };
881
882 memset(caps, 0, caps_len);
883
884 if (ops->ndo_setup_tc)
885 ops->ndo_setup_tc(dev, TC_QUERY_CAPS, &base);
886}
887EXPORT_SYMBOL(qdisc_offload_query_caps);
888
889static void qdisc_offload_graft_root(struct net_device *dev,
890 struct Qdisc *new, struct Qdisc *old,
891 struct netlink_ext_ack *extack)
892{
893 struct tc_root_qopt_offload graft_offload = {
894 .command = TC_ROOT_GRAFT,
895 .handle = new ? new->handle : 0,
896 .ingress = (new && new->flags & TCQ_F_INGRESS) ||
897 (old && old->flags & TCQ_F_INGRESS),
898 };
899
900 qdisc_offload_graft_helper(dev, NULL, new, old,
901 TC_SETUP_ROOT_QDISC, &graft_offload, extack);
902}
903
904static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
905 u32 portid, u32 seq, u16 flags, int event)
906{
907 struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
908 struct gnet_stats_queue __percpu *cpu_qstats = NULL;
909 struct tcmsg *tcm;
910 struct nlmsghdr *nlh;
911 unsigned char *b = skb_tail_pointer(skb);
912 struct gnet_dump d;
913 struct qdisc_size_table *stab;
914 u32 block_index;
915 __u32 qlen;
916
917 cond_resched();
918 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
919 if (!nlh)
920 goto out_nlmsg_trim;
921 tcm = nlmsg_data(nlh);
922 tcm->tcm_family = AF_UNSPEC;
923 tcm->tcm__pad1 = 0;
924 tcm->tcm__pad2 = 0;
925 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
926 tcm->tcm_parent = clid;
927 tcm->tcm_handle = q->handle;
928 tcm->tcm_info = refcount_read(&q->refcnt);
929 if (nla_put_string(skb, TCA_KIND, q->ops->id))
930 goto nla_put_failure;
931 if (q->ops->ingress_block_get) {
932 block_index = q->ops->ingress_block_get(q);
933 if (block_index &&
934 nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
935 goto nla_put_failure;
936 }
937 if (q->ops->egress_block_get) {
938 block_index = q->ops->egress_block_get(q);
939 if (block_index &&
940 nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
941 goto nla_put_failure;
942 }
943 if (q->ops->dump && q->ops->dump(q, skb) < 0)
944 goto nla_put_failure;
945 if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
946 goto nla_put_failure;
947 qlen = qdisc_qlen_sum(q);
948
949 stab = rtnl_dereference(q->stab);
950 if (stab && qdisc_dump_stab(skb, stab) < 0)
951 goto nla_put_failure;
952
953 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
954 NULL, &d, TCA_PAD) < 0)
955 goto nla_put_failure;
956
957 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
958 goto nla_put_failure;
959
960 if (qdisc_is_percpu_stats(q)) {
961 cpu_bstats = q->cpu_bstats;
962 cpu_qstats = q->cpu_qstats;
963 }
964
965 if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
966 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
967 gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
968 goto nla_put_failure;
969
970 if (gnet_stats_finish_copy(&d) < 0)
971 goto nla_put_failure;
972
973 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
974 return skb->len;
975
976out_nlmsg_trim:
977nla_put_failure:
978 nlmsg_trim(skb, b);
979 return -1;
980}
981
982static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
983{
984 if (q->flags & TCQ_F_BUILTIN)
985 return true;
986 if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
987 return true;
988
989 return false;
990}
991
992static int qdisc_notify(struct net *net, struct sk_buff *oskb,
993 struct nlmsghdr *n, u32 clid,
994 struct Qdisc *old, struct Qdisc *new)
995{
996 struct sk_buff *skb;
997 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
998
999 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1000 if (!skb)
1001 return -ENOBUFS;
1002
1003 if (old && !tc_qdisc_dump_ignore(old, false)) {
1004 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1005 0, RTM_DELQDISC) < 0)
1006 goto err_out;
1007 }
1008 if (new && !tc_qdisc_dump_ignore(new, false)) {
1009 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1010 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1011 goto err_out;
1012 }
1013
1014 if (skb->len)
1015 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1016 n->nlmsg_flags & NLM_F_ECHO);
1017
1018err_out:
1019 kfree_skb(skb);
1020 return -EINVAL;
1021}
1022
1023static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1024 struct nlmsghdr *n, u32 clid,
1025 struct Qdisc *old, struct Qdisc *new)
1026{
1027 if (new || old)
1028 qdisc_notify(net, skb, n, clid, old, new);
1029
1030 if (old)
1031 qdisc_put(old);
1032}
1033
1034static void qdisc_clear_nolock(struct Qdisc *sch)
1035{
1036 sch->flags &= ~TCQ_F_NOLOCK;
1037 if (!(sch->flags & TCQ_F_CPUSTATS))
1038 return;
1039
1040 free_percpu(sch->cpu_bstats);
1041 free_percpu(sch->cpu_qstats);
1042 sch->cpu_bstats = NULL;
1043 sch->cpu_qstats = NULL;
1044 sch->flags &= ~TCQ_F_CPUSTATS;
1045}
1046
1047/* Graft qdisc "new" to class "classid" of qdisc "parent" or
1048 * to device "dev".
1049 *
1050 * When appropriate send a netlink notification using 'skb'
1051 * and "n".
1052 *
1053 * On success, destroy old qdisc.
1054 */
1055
1056static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1057 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1058 struct Qdisc *new, struct Qdisc *old,
1059 struct netlink_ext_ack *extack)
1060{
1061 struct Qdisc *q = old;
1062 struct net *net = dev_net(dev);
1063
1064 if (parent == NULL) {
1065 unsigned int i, num_q, ingress;
1066
1067 ingress = 0;
1068 num_q = dev->num_tx_queues;
1069 if ((q && q->flags & TCQ_F_INGRESS) ||
1070 (new && new->flags & TCQ_F_INGRESS)) {
1071 num_q = 1;
1072 ingress = 1;
1073 if (!dev_ingress_queue(dev)) {
1074 NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1075 return -ENOENT;
1076 }
1077 }
1078
1079 if (dev->flags & IFF_UP)
1080 dev_deactivate(dev);
1081
1082 qdisc_offload_graft_root(dev, new, old, extack);
1083
1084 if (new && new->ops->attach && !ingress)
1085 goto skip;
1086
1087 for (i = 0; i < num_q; i++) {
1088 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1089
1090 if (!ingress)
1091 dev_queue = netdev_get_tx_queue(dev, i);
1092
1093 old = dev_graft_qdisc(dev_queue, new);
1094 if (new && i > 0)
1095 qdisc_refcount_inc(new);
1096
1097 if (!ingress)
1098 qdisc_put(old);
1099 }
1100
1101skip:
1102 if (!ingress) {
1103 old = rtnl_dereference(dev->qdisc);
1104 if (new && !new->ops->attach)
1105 qdisc_refcount_inc(new);
1106 rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1107
1108 notify_and_destroy(net, skb, n, classid, old, new);
1109
1110 if (new && new->ops->attach)
1111 new->ops->attach(new);
1112 } else {
1113 notify_and_destroy(net, skb, n, classid, old, new);
1114 }
1115
1116 if (dev->flags & IFF_UP)
1117 dev_activate(dev);
1118 } else {
1119 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1120 unsigned long cl;
1121 int err;
1122
1123 /* Only support running class lockless if parent is lockless */
1124 if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1125 qdisc_clear_nolock(new);
1126
1127 if (!cops || !cops->graft)
1128 return -EOPNOTSUPP;
1129
1130 cl = cops->find(parent, classid);
1131 if (!cl) {
1132 NL_SET_ERR_MSG(extack, "Specified class not found");
1133 return -ENOENT;
1134 }
1135
1136 if (new && new->ops == &noqueue_qdisc_ops) {
1137 NL_SET_ERR_MSG(extack, "Cannot assign noqueue to a class");
1138 return -EINVAL;
1139 }
1140
1141 err = cops->graft(parent, cl, new, &old, extack);
1142 if (err)
1143 return err;
1144 notify_and_destroy(net, skb, n, classid, old, new);
1145 }
1146 return 0;
1147}
1148
1149static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1150 struct netlink_ext_ack *extack)
1151{
1152 u32 block_index;
1153
1154 if (tca[TCA_INGRESS_BLOCK]) {
1155 block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1156
1157 if (!block_index) {
1158 NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1159 return -EINVAL;
1160 }
1161 if (!sch->ops->ingress_block_set) {
1162 NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1163 return -EOPNOTSUPP;
1164 }
1165 sch->ops->ingress_block_set(sch, block_index);
1166 }
1167 if (tca[TCA_EGRESS_BLOCK]) {
1168 block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1169
1170 if (!block_index) {
1171 NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1172 return -EINVAL;
1173 }
1174 if (!sch->ops->egress_block_set) {
1175 NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1176 return -EOPNOTSUPP;
1177 }
1178 sch->ops->egress_block_set(sch, block_index);
1179 }
1180 return 0;
1181}
1182
1183/*
1184 Allocate and initialize new qdisc.
1185
1186 Parameters are passed via opt.
1187 */
1188
1189static struct Qdisc *qdisc_create(struct net_device *dev,
1190 struct netdev_queue *dev_queue,
1191 u32 parent, u32 handle,
1192 struct nlattr **tca, int *errp,
1193 struct netlink_ext_ack *extack)
1194{
1195 int err;
1196 struct nlattr *kind = tca[TCA_KIND];
1197 struct Qdisc *sch;
1198 struct Qdisc_ops *ops;
1199 struct qdisc_size_table *stab;
1200
1201 ops = qdisc_lookup_ops(kind);
1202#ifdef CONFIG_MODULES
1203 if (ops == NULL && kind != NULL) {
1204 char name[IFNAMSIZ];
1205 if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1206 /* We dropped the RTNL semaphore in order to
1207 * perform the module load. So, even if we
1208 * succeeded in loading the module we have to
1209 * tell the caller to replay the request. We
1210 * indicate this using -EAGAIN.
1211 * We replay the request because the device may
1212 * go away in the mean time.
1213 */
1214 rtnl_unlock();
1215 request_module("sch_%s", name);
1216 rtnl_lock();
1217 ops = qdisc_lookup_ops(kind);
1218 if (ops != NULL) {
1219 /* We will try again qdisc_lookup_ops,
1220 * so don't keep a reference.
1221 */
1222 module_put(ops->owner);
1223 err = -EAGAIN;
1224 goto err_out;
1225 }
1226 }
1227 }
1228#endif
1229
1230 err = -ENOENT;
1231 if (!ops) {
1232 NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1233 goto err_out;
1234 }
1235
1236 sch = qdisc_alloc(dev_queue, ops, extack);
1237 if (IS_ERR(sch)) {
1238 err = PTR_ERR(sch);
1239 goto err_out2;
1240 }
1241
1242 sch->parent = parent;
1243
1244 if (handle == TC_H_INGRESS) {
1245 sch->flags |= TCQ_F_INGRESS;
1246 handle = TC_H_MAKE(TC_H_INGRESS, 0);
1247 } else {
1248 if (handle == 0) {
1249 handle = qdisc_alloc_handle(dev);
1250 if (handle == 0) {
1251 NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1252 err = -ENOSPC;
1253 goto err_out3;
1254 }
1255 }
1256 if (!netif_is_multiqueue(dev))
1257 sch->flags |= TCQ_F_ONETXQUEUE;
1258 }
1259
1260 sch->handle = handle;
1261
1262 /* This exist to keep backward compatible with a userspace
1263 * loophole, what allowed userspace to get IFF_NO_QUEUE
1264 * facility on older kernels by setting tx_queue_len=0 (prior
1265 * to qdisc init), and then forgot to reinit tx_queue_len
1266 * before again attaching a qdisc.
1267 */
1268 if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1269 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1270 netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1271 }
1272
1273 err = qdisc_block_indexes_set(sch, tca, extack);
1274 if (err)
1275 goto err_out3;
1276
1277 if (ops->init) {
1278 err = ops->init(sch, tca[TCA_OPTIONS], extack);
1279 if (err != 0)
1280 goto err_out5;
1281 }
1282
1283 if (tca[TCA_STAB]) {
1284 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1285 if (IS_ERR(stab)) {
1286 err = PTR_ERR(stab);
1287 goto err_out4;
1288 }
1289 rcu_assign_pointer(sch->stab, stab);
1290 }
1291 if (tca[TCA_RATE]) {
1292 err = -EOPNOTSUPP;
1293 if (sch->flags & TCQ_F_MQROOT) {
1294 NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1295 goto err_out4;
1296 }
1297
1298 err = gen_new_estimator(&sch->bstats,
1299 sch->cpu_bstats,
1300 &sch->rate_est,
1301 NULL,
1302 true,
1303 tca[TCA_RATE]);
1304 if (err) {
1305 NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1306 goto err_out4;
1307 }
1308 }
1309
1310 qdisc_hash_add(sch, false);
1311 trace_qdisc_create(ops, dev, parent);
1312
1313 return sch;
1314
1315err_out5:
1316 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1317 if (ops->destroy)
1318 ops->destroy(sch);
1319err_out3:
1320 netdev_put(dev, &sch->dev_tracker);
1321 qdisc_free(sch);
1322err_out2:
1323 module_put(ops->owner);
1324err_out:
1325 *errp = err;
1326 return NULL;
1327
1328err_out4:
1329 /*
1330 * Any broken qdiscs that would require a ops->reset() here?
1331 * The qdisc was never in action so it shouldn't be necessary.
1332 */
1333 qdisc_put_stab(rtnl_dereference(sch->stab));
1334 if (ops->destroy)
1335 ops->destroy(sch);
1336 goto err_out3;
1337}
1338
1339static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1340 struct netlink_ext_ack *extack)
1341{
1342 struct qdisc_size_table *ostab, *stab = NULL;
1343 int err = 0;
1344
1345 if (tca[TCA_OPTIONS]) {
1346 if (!sch->ops->change) {
1347 NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1348 return -EINVAL;
1349 }
1350 if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1351 NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1352 return -EOPNOTSUPP;
1353 }
1354 err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1355 if (err)
1356 return err;
1357 }
1358
1359 if (tca[TCA_STAB]) {
1360 stab = qdisc_get_stab(tca[TCA_STAB], extack);
1361 if (IS_ERR(stab))
1362 return PTR_ERR(stab);
1363 }
1364
1365 ostab = rtnl_dereference(sch->stab);
1366 rcu_assign_pointer(sch->stab, stab);
1367 qdisc_put_stab(ostab);
1368
1369 if (tca[TCA_RATE]) {
1370 /* NB: ignores errors from replace_estimator
1371 because change can't be undone. */
1372 if (sch->flags & TCQ_F_MQROOT)
1373 goto out;
1374 gen_replace_estimator(&sch->bstats,
1375 sch->cpu_bstats,
1376 &sch->rate_est,
1377 NULL,
1378 true,
1379 tca[TCA_RATE]);
1380 }
1381out:
1382 return 0;
1383}
1384
1385struct check_loop_arg {
1386 struct qdisc_walker w;
1387 struct Qdisc *p;
1388 int depth;
1389};
1390
1391static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1392 struct qdisc_walker *w);
1393
1394static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1395{
1396 struct check_loop_arg arg;
1397
1398 if (q->ops->cl_ops == NULL)
1399 return 0;
1400
1401 arg.w.stop = arg.w.skip = arg.w.count = 0;
1402 arg.w.fn = check_loop_fn;
1403 arg.depth = depth;
1404 arg.p = p;
1405 q->ops->cl_ops->walk(q, &arg.w);
1406 return arg.w.stop ? -ELOOP : 0;
1407}
1408
1409static int
1410check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1411{
1412 struct Qdisc *leaf;
1413 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1414 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1415
1416 leaf = cops->leaf(q, cl);
1417 if (leaf) {
1418 if (leaf == arg->p || arg->depth > 7)
1419 return -ELOOP;
1420 return check_loop(leaf, arg->p, arg->depth + 1);
1421 }
1422 return 0;
1423}
1424
1425const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1426 [TCA_KIND] = { .type = NLA_STRING },
1427 [TCA_RATE] = { .type = NLA_BINARY,
1428 .len = sizeof(struct tc_estimator) },
1429 [TCA_STAB] = { .type = NLA_NESTED },
1430 [TCA_DUMP_INVISIBLE] = { .type = NLA_FLAG },
1431 [TCA_CHAIN] = { .type = NLA_U32 },
1432 [TCA_INGRESS_BLOCK] = { .type = NLA_U32 },
1433 [TCA_EGRESS_BLOCK] = { .type = NLA_U32 },
1434};
1435
1436/*
1437 * Delete/get qdisc.
1438 */
1439
1440static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1441 struct netlink_ext_ack *extack)
1442{
1443 struct net *net = sock_net(skb->sk);
1444 struct tcmsg *tcm = nlmsg_data(n);
1445 struct nlattr *tca[TCA_MAX + 1];
1446 struct net_device *dev;
1447 u32 clid;
1448 struct Qdisc *q = NULL;
1449 struct Qdisc *p = NULL;
1450 int err;
1451
1452 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1453 rtm_tca_policy, extack);
1454 if (err < 0)
1455 return err;
1456
1457 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1458 if (!dev)
1459 return -ENODEV;
1460
1461 clid = tcm->tcm_parent;
1462 if (clid) {
1463 if (clid != TC_H_ROOT) {
1464 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1465 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1466 if (!p) {
1467 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1468 return -ENOENT;
1469 }
1470 q = qdisc_leaf(p, clid);
1471 } else if (dev_ingress_queue(dev)) {
1472 q = dev_ingress_queue(dev)->qdisc_sleeping;
1473 }
1474 } else {
1475 q = rtnl_dereference(dev->qdisc);
1476 }
1477 if (!q) {
1478 NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1479 return -ENOENT;
1480 }
1481
1482 if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1483 NL_SET_ERR_MSG(extack, "Invalid handle");
1484 return -EINVAL;
1485 }
1486 } else {
1487 q = qdisc_lookup(dev, tcm->tcm_handle);
1488 if (!q) {
1489 NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1490 return -ENOENT;
1491 }
1492 }
1493
1494 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1495 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1496 return -EINVAL;
1497 }
1498
1499 if (n->nlmsg_type == RTM_DELQDISC) {
1500 if (!clid) {
1501 NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1502 return -EINVAL;
1503 }
1504 if (q->handle == 0) {
1505 NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1506 return -ENOENT;
1507 }
1508 err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1509 if (err != 0)
1510 return err;
1511 } else {
1512 qdisc_notify(net, skb, n, clid, NULL, q);
1513 }
1514 return 0;
1515}
1516
1517/*
1518 * Create/change qdisc.
1519 */
1520
1521static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1522 struct netlink_ext_ack *extack)
1523{
1524 struct net *net = sock_net(skb->sk);
1525 struct tcmsg *tcm;
1526 struct nlattr *tca[TCA_MAX + 1];
1527 struct net_device *dev;
1528 u32 clid;
1529 struct Qdisc *q, *p;
1530 int err;
1531
1532replay:
1533 /* Reinit, just in case something touches this. */
1534 err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1535 rtm_tca_policy, extack);
1536 if (err < 0)
1537 return err;
1538
1539 tcm = nlmsg_data(n);
1540 clid = tcm->tcm_parent;
1541 q = p = NULL;
1542
1543 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1544 if (!dev)
1545 return -ENODEV;
1546
1547
1548 if (clid) {
1549 if (clid != TC_H_ROOT) {
1550 if (clid != TC_H_INGRESS) {
1551 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1552 if (!p) {
1553 NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1554 return -ENOENT;
1555 }
1556 q = qdisc_leaf(p, clid);
1557 } else if (dev_ingress_queue_create(dev)) {
1558 q = dev_ingress_queue(dev)->qdisc_sleeping;
1559 }
1560 } else {
1561 q = rtnl_dereference(dev->qdisc);
1562 }
1563
1564 /* It may be default qdisc, ignore it */
1565 if (q && q->handle == 0)
1566 q = NULL;
1567
1568 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1569 if (tcm->tcm_handle) {
1570 if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1571 NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1572 return -EEXIST;
1573 }
1574 if (TC_H_MIN(tcm->tcm_handle)) {
1575 NL_SET_ERR_MSG(extack, "Invalid minor handle");
1576 return -EINVAL;
1577 }
1578 q = qdisc_lookup(dev, tcm->tcm_handle);
1579 if (!q)
1580 goto create_n_graft;
1581 if (n->nlmsg_flags & NLM_F_EXCL) {
1582 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1583 return -EEXIST;
1584 }
1585 if (tca[TCA_KIND] &&
1586 nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1587 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1588 return -EINVAL;
1589 }
1590 if (q == p ||
1591 (p && check_loop(q, p, 0))) {
1592 NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1593 return -ELOOP;
1594 }
1595 qdisc_refcount_inc(q);
1596 goto graft;
1597 } else {
1598 if (!q)
1599 goto create_n_graft;
1600
1601 /* This magic test requires explanation.
1602 *
1603 * We know, that some child q is already
1604 * attached to this parent and have choice:
1605 * either to change it or to create/graft new one.
1606 *
1607 * 1. We are allowed to create/graft only
1608 * if CREATE and REPLACE flags are set.
1609 *
1610 * 2. If EXCL is set, requestor wanted to say,
1611 * that qdisc tcm_handle is not expected
1612 * to exist, so that we choose create/graft too.
1613 *
1614 * 3. The last case is when no flags are set.
1615 * Alas, it is sort of hole in API, we
1616 * cannot decide what to do unambiguously.
1617 * For now we select create/graft, if
1618 * user gave KIND, which does not match existing.
1619 */
1620 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1621 (n->nlmsg_flags & NLM_F_REPLACE) &&
1622 ((n->nlmsg_flags & NLM_F_EXCL) ||
1623 (tca[TCA_KIND] &&
1624 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1625 goto create_n_graft;
1626 }
1627 }
1628 } else {
1629 if (!tcm->tcm_handle) {
1630 NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1631 return -EINVAL;
1632 }
1633 q = qdisc_lookup(dev, tcm->tcm_handle);
1634 }
1635
1636 /* Change qdisc parameters */
1637 if (!q) {
1638 NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1639 return -ENOENT;
1640 }
1641 if (n->nlmsg_flags & NLM_F_EXCL) {
1642 NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1643 return -EEXIST;
1644 }
1645 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1646 NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1647 return -EINVAL;
1648 }
1649 err = qdisc_change(q, tca, extack);
1650 if (err == 0)
1651 qdisc_notify(net, skb, n, clid, NULL, q);
1652 return err;
1653
1654create_n_graft:
1655 if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1656 NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1657 return -ENOENT;
1658 }
1659 if (clid == TC_H_INGRESS) {
1660 if (dev_ingress_queue(dev)) {
1661 q = qdisc_create(dev, dev_ingress_queue(dev),
1662 tcm->tcm_parent, tcm->tcm_parent,
1663 tca, &err, extack);
1664 } else {
1665 NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1666 err = -ENOENT;
1667 }
1668 } else {
1669 struct netdev_queue *dev_queue;
1670
1671 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1672 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1673 else if (p)
1674 dev_queue = p->dev_queue;
1675 else
1676 dev_queue = netdev_get_tx_queue(dev, 0);
1677
1678 q = qdisc_create(dev, dev_queue,
1679 tcm->tcm_parent, tcm->tcm_handle,
1680 tca, &err, extack);
1681 }
1682 if (q == NULL) {
1683 if (err == -EAGAIN)
1684 goto replay;
1685 return err;
1686 }
1687
1688graft:
1689 err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1690 if (err) {
1691 if (q)
1692 qdisc_put(q);
1693 return err;
1694 }
1695
1696 return 0;
1697}
1698
1699static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1700 struct netlink_callback *cb,
1701 int *q_idx_p, int s_q_idx, bool recur,
1702 bool dump_invisible)
1703{
1704 int ret = 0, q_idx = *q_idx_p;
1705 struct Qdisc *q;
1706 int b;
1707
1708 if (!root)
1709 return 0;
1710
1711 q = root;
1712 if (q_idx < s_q_idx) {
1713 q_idx++;
1714 } else {
1715 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1716 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1717 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1718 RTM_NEWQDISC) <= 0)
1719 goto done;
1720 q_idx++;
1721 }
1722
1723 /* If dumping singletons, there is no qdisc_dev(root) and the singleton
1724 * itself has already been dumped.
1725 *
1726 * If we've already dumped the top-level (ingress) qdisc above and the global
1727 * qdisc hashtable, we don't want to hit it again
1728 */
1729 if (!qdisc_dev(root) || !recur)
1730 goto out;
1731
1732 hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1733 if (q_idx < s_q_idx) {
1734 q_idx++;
1735 continue;
1736 }
1737 if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1738 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1739 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1740 RTM_NEWQDISC) <= 0)
1741 goto done;
1742 q_idx++;
1743 }
1744
1745out:
1746 *q_idx_p = q_idx;
1747 return ret;
1748done:
1749 ret = -1;
1750 goto out;
1751}
1752
1753static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1754{
1755 struct net *net = sock_net(skb->sk);
1756 int idx, q_idx;
1757 int s_idx, s_q_idx;
1758 struct net_device *dev;
1759 const struct nlmsghdr *nlh = cb->nlh;
1760 struct nlattr *tca[TCA_MAX + 1];
1761 int err;
1762
1763 s_idx = cb->args[0];
1764 s_q_idx = q_idx = cb->args[1];
1765
1766 idx = 0;
1767 ASSERT_RTNL();
1768
1769 err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1770 rtm_tca_policy, cb->extack);
1771 if (err < 0)
1772 return err;
1773
1774 for_each_netdev(net, dev) {
1775 struct netdev_queue *dev_queue;
1776
1777 if (idx < s_idx)
1778 goto cont;
1779 if (idx > s_idx)
1780 s_q_idx = 0;
1781 q_idx = 0;
1782
1783 if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1784 skb, cb, &q_idx, s_q_idx,
1785 true, tca[TCA_DUMP_INVISIBLE]) < 0)
1786 goto done;
1787
1788 dev_queue = dev_ingress_queue(dev);
1789 if (dev_queue &&
1790 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1791 &q_idx, s_q_idx, false,
1792 tca[TCA_DUMP_INVISIBLE]) < 0)
1793 goto done;
1794
1795cont:
1796 idx++;
1797 }
1798
1799done:
1800 cb->args[0] = idx;
1801 cb->args[1] = q_idx;
1802
1803 return skb->len;
1804}
1805
1806
1807
1808/************************************************
1809 * Traffic classes manipulation. *
1810 ************************************************/
1811
1812static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1813 unsigned long cl,
1814 u32 portid, u32 seq, u16 flags, int event)
1815{
1816 struct tcmsg *tcm;
1817 struct nlmsghdr *nlh;
1818 unsigned char *b = skb_tail_pointer(skb);
1819 struct gnet_dump d;
1820 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1821
1822 cond_resched();
1823 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1824 if (!nlh)
1825 goto out_nlmsg_trim;
1826 tcm = nlmsg_data(nlh);
1827 tcm->tcm_family = AF_UNSPEC;
1828 tcm->tcm__pad1 = 0;
1829 tcm->tcm__pad2 = 0;
1830 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1831 tcm->tcm_parent = q->handle;
1832 tcm->tcm_handle = q->handle;
1833 tcm->tcm_info = 0;
1834 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1835 goto nla_put_failure;
1836 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1837 goto nla_put_failure;
1838
1839 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1840 NULL, &d, TCA_PAD) < 0)
1841 goto nla_put_failure;
1842
1843 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1844 goto nla_put_failure;
1845
1846 if (gnet_stats_finish_copy(&d) < 0)
1847 goto nla_put_failure;
1848
1849 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1850 return skb->len;
1851
1852out_nlmsg_trim:
1853nla_put_failure:
1854 nlmsg_trim(skb, b);
1855 return -1;
1856}
1857
1858static int tclass_notify(struct net *net, struct sk_buff *oskb,
1859 struct nlmsghdr *n, struct Qdisc *q,
1860 unsigned long cl, int event)
1861{
1862 struct sk_buff *skb;
1863 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1864
1865 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1866 if (!skb)
1867 return -ENOBUFS;
1868
1869 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1870 kfree_skb(skb);
1871 return -EINVAL;
1872 }
1873
1874 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1875 n->nlmsg_flags & NLM_F_ECHO);
1876}
1877
1878static int tclass_del_notify(struct net *net,
1879 const struct Qdisc_class_ops *cops,
1880 struct sk_buff *oskb, struct nlmsghdr *n,
1881 struct Qdisc *q, unsigned long cl,
1882 struct netlink_ext_ack *extack)
1883{
1884 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1885 struct sk_buff *skb;
1886 int err = 0;
1887
1888 if (!cops->delete)
1889 return -EOPNOTSUPP;
1890
1891 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1892 if (!skb)
1893 return -ENOBUFS;
1894
1895 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1896 RTM_DELTCLASS) < 0) {
1897 kfree_skb(skb);
1898 return -EINVAL;
1899 }
1900
1901 err = cops->delete(q, cl, extack);
1902 if (err) {
1903 kfree_skb(skb);
1904 return err;
1905 }
1906
1907 err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1908 n->nlmsg_flags & NLM_F_ECHO);
1909 return err;
1910}
1911
1912#ifdef CONFIG_NET_CLS
1913
1914struct tcf_bind_args {
1915 struct tcf_walker w;
1916 unsigned long base;
1917 unsigned long cl;
1918 u32 classid;
1919};
1920
1921static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1922{
1923 struct tcf_bind_args *a = (void *)arg;
1924
1925 if (n && tp->ops->bind_class) {
1926 struct Qdisc *q = tcf_block_q(tp->chain->block);
1927
1928 sch_tree_lock(q);
1929 tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1930 sch_tree_unlock(q);
1931 }
1932 return 0;
1933}
1934
1935struct tc_bind_class_args {
1936 struct qdisc_walker w;
1937 unsigned long new_cl;
1938 u32 portid;
1939 u32 clid;
1940};
1941
1942static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1943 struct qdisc_walker *w)
1944{
1945 struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1946 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1947 struct tcf_block *block;
1948 struct tcf_chain *chain;
1949
1950 block = cops->tcf_block(q, cl, NULL);
1951 if (!block)
1952 return 0;
1953 for (chain = tcf_get_next_chain(block, NULL);
1954 chain;
1955 chain = tcf_get_next_chain(block, chain)) {
1956 struct tcf_proto *tp;
1957
1958 for (tp = tcf_get_next_proto(chain, NULL);
1959 tp; tp = tcf_get_next_proto(chain, tp)) {
1960 struct tcf_bind_args arg = {};
1961
1962 arg.w.fn = tcf_node_bind;
1963 arg.classid = a->clid;
1964 arg.base = cl;
1965 arg.cl = a->new_cl;
1966 tp->ops->walk(tp, &arg.w, true);
1967 }
1968 }
1969
1970 return 0;
1971}
1972
1973static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1974 unsigned long new_cl)
1975{
1976 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1977 struct tc_bind_class_args args = {};
1978
1979 if (!cops->tcf_block)
1980 return;
1981 args.portid = portid;
1982 args.clid = clid;
1983 args.new_cl = new_cl;
1984 args.w.fn = tc_bind_class_walker;
1985 q->ops->cl_ops->walk(q, &args.w);
1986}
1987
1988#else
1989
1990static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1991 unsigned long new_cl)
1992{
1993}
1994
1995#endif
1996
1997static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1998 struct netlink_ext_ack *extack)
1999{
2000 struct net *net = sock_net(skb->sk);
2001 struct tcmsg *tcm = nlmsg_data(n);
2002 struct nlattr *tca[TCA_MAX + 1];
2003 struct net_device *dev;
2004 struct Qdisc *q = NULL;
2005 const struct Qdisc_class_ops *cops;
2006 unsigned long cl = 0;
2007 unsigned long new_cl;
2008 u32 portid;
2009 u32 clid;
2010 u32 qid;
2011 int err;
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 = rtnl_dereference(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 = rtnl_dereference(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(rtnl_dereference(dev->qdisc),
2224 skb, tcm, cb, &t, s_t, true) < 0)
2225 goto done;
2226
2227 dev_queue = dev_ingress_queue(dev);
2228 if (dev_queue &&
2229 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2230 &t, s_t, false) < 0)
2231 goto done;
2232
2233done:
2234 cb->args[0] = t;
2235
2236 dev_put(dev);
2237 return skb->len;
2238}
2239
2240#ifdef CONFIG_PROC_FS
2241static int psched_show(struct seq_file *seq, void *v)
2242{
2243 seq_printf(seq, "%08x %08x %08x %08x\n",
2244 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2245 1000000,
2246 (u32)NSEC_PER_SEC / hrtimer_resolution);
2247
2248 return 0;
2249}
2250
2251static int __net_init psched_net_init(struct net *net)
2252{
2253 struct proc_dir_entry *e;
2254
2255 e = proc_create_single("psched", 0, net->proc_net, psched_show);
2256 if (e == NULL)
2257 return -ENOMEM;
2258
2259 return 0;
2260}
2261
2262static void __net_exit psched_net_exit(struct net *net)
2263{
2264 remove_proc_entry("psched", net->proc_net);
2265}
2266#else
2267static int __net_init psched_net_init(struct net *net)
2268{
2269 return 0;
2270}
2271
2272static void __net_exit psched_net_exit(struct net *net)
2273{
2274}
2275#endif
2276
2277static struct pernet_operations psched_net_ops = {
2278 .init = psched_net_init,
2279 .exit = psched_net_exit,
2280};
2281
2282DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper);
2283
2284static int __init pktsched_init(void)
2285{
2286 int err;
2287
2288 err = register_pernet_subsys(&psched_net_ops);
2289 if (err) {
2290 pr_err("pktsched_init: "
2291 "cannot initialize per netns operations\n");
2292 return err;
2293 }
2294
2295 register_qdisc(&pfifo_fast_ops);
2296 register_qdisc(&pfifo_qdisc_ops);
2297 register_qdisc(&bfifo_qdisc_ops);
2298 register_qdisc(&pfifo_head_drop_qdisc_ops);
2299 register_qdisc(&mq_qdisc_ops);
2300 register_qdisc(&noqueue_qdisc_ops);
2301
2302 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2303 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2304 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2305 0);
2306 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2307 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2308 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2309 0);
2310
2311 tc_wrapper_init();
2312
2313 return 0;
2314}
2315
2316subsys_initcall(pktsched_init);