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