Loading...
1/*
2 * net/sched/sch_api.c Packet scheduler API.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 *
11 * Fixes:
12 *
13 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired.
14 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
15 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support
16 */
17
18#include <linux/module.h>
19#include <linux/types.h>
20#include <linux/kernel.h>
21#include <linux/string.h>
22#include <linux/errno.h>
23#include <linux/skbuff.h>
24#include <linux/init.h>
25#include <linux/proc_fs.h>
26#include <linux/seq_file.h>
27#include <linux/kmod.h>
28#include <linux/list.h>
29#include <linux/hrtimer.h>
30#include <linux/lockdep.h>
31#include <linux/slab.h>
32
33#include <net/net_namespace.h>
34#include <net/sock.h>
35#include <net/netlink.h>
36#include <net/pkt_sched.h>
37
38static int qdisc_notify(struct net *net, struct sk_buff *oskb,
39 struct nlmsghdr *n, u32 clid,
40 struct Qdisc *old, struct Qdisc *new);
41static int tclass_notify(struct net *net, struct sk_buff *oskb,
42 struct nlmsghdr *n, struct Qdisc *q,
43 unsigned long cl, int event);
44
45/*
46
47 Short review.
48 -------------
49
50 This file consists of two interrelated parts:
51
52 1. queueing disciplines manager frontend.
53 2. traffic classes manager frontend.
54
55 Generally, queueing discipline ("qdisc") is a black box,
56 which is able to enqueue packets and to dequeue them (when
57 device is ready to send something) in order and at times
58 determined by algorithm hidden in it.
59
60 qdisc's are divided to two categories:
61 - "queues", which have no internal structure visible from outside.
62 - "schedulers", which split all the packets to "traffic classes",
63 using "packet classifiers" (look at cls_api.c)
64
65 In turn, classes may have child qdiscs (as rule, queues)
66 attached to them etc. etc. etc.
67
68 The goal of the routines in this file is to translate
69 information supplied by user in the form of handles
70 to more intelligible for kernel form, to make some sanity
71 checks and part of work, which is common to all qdiscs
72 and to provide rtnetlink notifications.
73
74 All real intelligent work is done inside qdisc modules.
75
76
77
78 Every discipline has two major routines: enqueue and dequeue.
79
80 ---dequeue
81
82 dequeue usually returns a skb to send. It is allowed to return NULL,
83 but it does not mean that queue is empty, it just means that
84 discipline does not want to send anything this time.
85 Queue is really empty if q->q.qlen == 0.
86 For complicated disciplines with multiple queues q->q is not
87 real packet queue, but however q->q.qlen must be valid.
88
89 ---enqueue
90
91 enqueue returns 0, if packet was enqueued successfully.
92 If packet (this one or another one) was dropped, it returns
93 not zero error code.
94 NET_XMIT_DROP - this packet dropped
95 Expected action: do not backoff, but wait until queue will clear.
96 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
97 Expected action: backoff or ignore
98 NET_XMIT_POLICED - dropped by police.
99 Expected action: backoff or error to real-time apps.
100
101 Auxiliary routines:
102
103 ---peek
104
105 like dequeue but without removing a packet from the queue
106
107 ---reset
108
109 returns qdisc to initial state: purge all buffers, clear all
110 timers, counters (except for statistics) etc.
111
112 ---init
113
114 initializes newly created qdisc.
115
116 ---destroy
117
118 destroys resources allocated by init and during lifetime of qdisc.
119
120 ---change
121
122 changes qdisc parameters.
123 */
124
125/* Protects list of registered TC modules. It is pure SMP lock. */
126static DEFINE_RWLOCK(qdisc_mod_lock);
127
128
129/************************************************
130 * Queueing disciplines manipulation. *
131 ************************************************/
132
133
134/* The list of all installed queueing disciplines. */
135
136static struct Qdisc_ops *qdisc_base;
137
138/* Register/uregister queueing discipline */
139
140int register_qdisc(struct Qdisc_ops *qops)
141{
142 struct Qdisc_ops *q, **qp;
143 int rc = -EEXIST;
144
145 write_lock(&qdisc_mod_lock);
146 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
147 if (!strcmp(qops->id, q->id))
148 goto out;
149
150 if (qops->enqueue == NULL)
151 qops->enqueue = noop_qdisc_ops.enqueue;
152 if (qops->peek == NULL) {
153 if (qops->dequeue == NULL)
154 qops->peek = noop_qdisc_ops.peek;
155 else
156 goto out_einval;
157 }
158 if (qops->dequeue == NULL)
159 qops->dequeue = noop_qdisc_ops.dequeue;
160
161 if (qops->cl_ops) {
162 const struct Qdisc_class_ops *cops = qops->cl_ops;
163
164 if (!(cops->get && cops->put && cops->walk && cops->leaf))
165 goto out_einval;
166
167 if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
168 goto out_einval;
169 }
170
171 qops->next = NULL;
172 *qp = qops;
173 rc = 0;
174out:
175 write_unlock(&qdisc_mod_lock);
176 return rc;
177
178out_einval:
179 rc = -EINVAL;
180 goto out;
181}
182EXPORT_SYMBOL(register_qdisc);
183
184int unregister_qdisc(struct Qdisc_ops *qops)
185{
186 struct Qdisc_ops *q, **qp;
187 int err = -ENOENT;
188
189 write_lock(&qdisc_mod_lock);
190 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
191 if (q == qops)
192 break;
193 if (q) {
194 *qp = q->next;
195 q->next = NULL;
196 err = 0;
197 }
198 write_unlock(&qdisc_mod_lock);
199 return err;
200}
201EXPORT_SYMBOL(unregister_qdisc);
202
203/* We know handle. Find qdisc among all qdisc's attached to device
204 (root qdisc, all its children, children of children etc.)
205 */
206
207static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
208{
209 struct Qdisc *q;
210
211 if (!(root->flags & TCQ_F_BUILTIN) &&
212 root->handle == handle)
213 return root;
214
215 list_for_each_entry(q, &root->list, list) {
216 if (q->handle == handle)
217 return q;
218 }
219 return NULL;
220}
221
222static void qdisc_list_add(struct Qdisc *q)
223{
224 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
225 list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list);
226}
227
228void qdisc_list_del(struct Qdisc *q)
229{
230 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
231 list_del(&q->list);
232}
233EXPORT_SYMBOL(qdisc_list_del);
234
235struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
236{
237 struct Qdisc *q;
238
239 q = qdisc_match_from_root(dev->qdisc, handle);
240 if (q)
241 goto out;
242
243 if (dev_ingress_queue(dev))
244 q = qdisc_match_from_root(
245 dev_ingress_queue(dev)->qdisc_sleeping,
246 handle);
247out:
248 return q;
249}
250
251static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
252{
253 unsigned long cl;
254 struct Qdisc *leaf;
255 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
256
257 if (cops == NULL)
258 return NULL;
259 cl = cops->get(p, classid);
260
261 if (cl == 0)
262 return NULL;
263 leaf = cops->leaf(p, cl);
264 cops->put(p, cl);
265 return leaf;
266}
267
268/* Find queueing discipline by name */
269
270static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
271{
272 struct Qdisc_ops *q = NULL;
273
274 if (kind) {
275 read_lock(&qdisc_mod_lock);
276 for (q = qdisc_base; q; q = q->next) {
277 if (nla_strcmp(kind, q->id) == 0) {
278 if (!try_module_get(q->owner))
279 q = NULL;
280 break;
281 }
282 }
283 read_unlock(&qdisc_mod_lock);
284 }
285 return q;
286}
287
288static struct qdisc_rate_table *qdisc_rtab_list;
289
290struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
291{
292 struct qdisc_rate_table *rtab;
293
294 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
295 if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
296 rtab->refcnt++;
297 return rtab;
298 }
299 }
300
301 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
302 nla_len(tab) != TC_RTAB_SIZE)
303 return NULL;
304
305 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
306 if (rtab) {
307 rtab->rate = *r;
308 rtab->refcnt = 1;
309 memcpy(rtab->data, nla_data(tab), 1024);
310 rtab->next = qdisc_rtab_list;
311 qdisc_rtab_list = rtab;
312 }
313 return rtab;
314}
315EXPORT_SYMBOL(qdisc_get_rtab);
316
317void qdisc_put_rtab(struct qdisc_rate_table *tab)
318{
319 struct qdisc_rate_table *rtab, **rtabp;
320
321 if (!tab || --tab->refcnt)
322 return;
323
324 for (rtabp = &qdisc_rtab_list;
325 (rtab = *rtabp) != NULL;
326 rtabp = &rtab->next) {
327 if (rtab == tab) {
328 *rtabp = rtab->next;
329 kfree(rtab);
330 return;
331 }
332 }
333}
334EXPORT_SYMBOL(qdisc_put_rtab);
335
336static LIST_HEAD(qdisc_stab_list);
337static DEFINE_SPINLOCK(qdisc_stab_lock);
338
339static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
340 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
341 [TCA_STAB_DATA] = { .type = NLA_BINARY },
342};
343
344static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
345{
346 struct nlattr *tb[TCA_STAB_MAX + 1];
347 struct qdisc_size_table *stab;
348 struct tc_sizespec *s;
349 unsigned int tsize = 0;
350 u16 *tab = NULL;
351 int err;
352
353 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
354 if (err < 0)
355 return ERR_PTR(err);
356 if (!tb[TCA_STAB_BASE])
357 return ERR_PTR(-EINVAL);
358
359 s = nla_data(tb[TCA_STAB_BASE]);
360
361 if (s->tsize > 0) {
362 if (!tb[TCA_STAB_DATA])
363 return ERR_PTR(-EINVAL);
364 tab = nla_data(tb[TCA_STAB_DATA]);
365 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
366 }
367
368 if (tsize != s->tsize || (!tab && tsize > 0))
369 return ERR_PTR(-EINVAL);
370
371 spin_lock(&qdisc_stab_lock);
372
373 list_for_each_entry(stab, &qdisc_stab_list, list) {
374 if (memcmp(&stab->szopts, s, sizeof(*s)))
375 continue;
376 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
377 continue;
378 stab->refcnt++;
379 spin_unlock(&qdisc_stab_lock);
380 return stab;
381 }
382
383 spin_unlock(&qdisc_stab_lock);
384
385 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
386 if (!stab)
387 return ERR_PTR(-ENOMEM);
388
389 stab->refcnt = 1;
390 stab->szopts = *s;
391 if (tsize > 0)
392 memcpy(stab->data, tab, tsize * sizeof(u16));
393
394 spin_lock(&qdisc_stab_lock);
395 list_add_tail(&stab->list, &qdisc_stab_list);
396 spin_unlock(&qdisc_stab_lock);
397
398 return stab;
399}
400
401static void stab_kfree_rcu(struct rcu_head *head)
402{
403 kfree(container_of(head, struct qdisc_size_table, rcu));
404}
405
406void qdisc_put_stab(struct qdisc_size_table *tab)
407{
408 if (!tab)
409 return;
410
411 spin_lock(&qdisc_stab_lock);
412
413 if (--tab->refcnt == 0) {
414 list_del(&tab->list);
415 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
416 }
417
418 spin_unlock(&qdisc_stab_lock);
419}
420EXPORT_SYMBOL(qdisc_put_stab);
421
422static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
423{
424 struct nlattr *nest;
425
426 nest = nla_nest_start(skb, TCA_STAB);
427 if (nest == NULL)
428 goto nla_put_failure;
429 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
430 goto nla_put_failure;
431 nla_nest_end(skb, nest);
432
433 return skb->len;
434
435nla_put_failure:
436 return -1;
437}
438
439void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
440{
441 int pkt_len, slot;
442
443 pkt_len = skb->len + stab->szopts.overhead;
444 if (unlikely(!stab->szopts.tsize))
445 goto out;
446
447 slot = pkt_len + stab->szopts.cell_align;
448 if (unlikely(slot < 0))
449 slot = 0;
450
451 slot >>= stab->szopts.cell_log;
452 if (likely(slot < stab->szopts.tsize))
453 pkt_len = stab->data[slot];
454 else
455 pkt_len = stab->data[stab->szopts.tsize - 1] *
456 (slot / stab->szopts.tsize) +
457 stab->data[slot % stab->szopts.tsize];
458
459 pkt_len <<= stab->szopts.size_log;
460out:
461 if (unlikely(pkt_len < 1))
462 pkt_len = 1;
463 qdisc_skb_cb(skb)->pkt_len = pkt_len;
464}
465EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
466
467void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
468{
469 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
470 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
471 txt, qdisc->ops->id, qdisc->handle >> 16);
472 qdisc->flags |= TCQ_F_WARN_NONWC;
473 }
474}
475EXPORT_SYMBOL(qdisc_warn_nonwc);
476
477static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
478{
479 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
480 timer);
481
482 qdisc_unthrottled(wd->qdisc);
483 __netif_schedule(qdisc_root(wd->qdisc));
484
485 return HRTIMER_NORESTART;
486}
487
488void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
489{
490 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
491 wd->timer.function = qdisc_watchdog;
492 wd->qdisc = qdisc;
493}
494EXPORT_SYMBOL(qdisc_watchdog_init);
495
496void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
497{
498 ktime_t time;
499
500 if (test_bit(__QDISC_STATE_DEACTIVATED,
501 &qdisc_root_sleeping(wd->qdisc)->state))
502 return;
503
504 qdisc_throttled(wd->qdisc);
505 time = ktime_set(0, 0);
506 time = ktime_add_ns(time, PSCHED_TICKS2NS(expires));
507 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
508}
509EXPORT_SYMBOL(qdisc_watchdog_schedule);
510
511void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
512{
513 hrtimer_cancel(&wd->timer);
514 qdisc_unthrottled(wd->qdisc);
515}
516EXPORT_SYMBOL(qdisc_watchdog_cancel);
517
518static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
519{
520 unsigned int size = n * sizeof(struct hlist_head), i;
521 struct hlist_head *h;
522
523 if (size <= PAGE_SIZE)
524 h = kmalloc(size, GFP_KERNEL);
525 else
526 h = (struct hlist_head *)
527 __get_free_pages(GFP_KERNEL, get_order(size));
528
529 if (h != NULL) {
530 for (i = 0; i < n; i++)
531 INIT_HLIST_HEAD(&h[i]);
532 }
533 return h;
534}
535
536static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
537{
538 unsigned int size = n * sizeof(struct hlist_head);
539
540 if (size <= PAGE_SIZE)
541 kfree(h);
542 else
543 free_pages((unsigned long)h, get_order(size));
544}
545
546void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
547{
548 struct Qdisc_class_common *cl;
549 struct hlist_node *n, *next;
550 struct hlist_head *nhash, *ohash;
551 unsigned int nsize, nmask, osize;
552 unsigned int i, h;
553
554 /* Rehash when load factor exceeds 0.75 */
555 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
556 return;
557 nsize = clhash->hashsize * 2;
558 nmask = nsize - 1;
559 nhash = qdisc_class_hash_alloc(nsize);
560 if (nhash == NULL)
561 return;
562
563 ohash = clhash->hash;
564 osize = clhash->hashsize;
565
566 sch_tree_lock(sch);
567 for (i = 0; i < osize; i++) {
568 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
569 h = qdisc_class_hash(cl->classid, nmask);
570 hlist_add_head(&cl->hnode, &nhash[h]);
571 }
572 }
573 clhash->hash = nhash;
574 clhash->hashsize = nsize;
575 clhash->hashmask = nmask;
576 sch_tree_unlock(sch);
577
578 qdisc_class_hash_free(ohash, osize);
579}
580EXPORT_SYMBOL(qdisc_class_hash_grow);
581
582int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
583{
584 unsigned int size = 4;
585
586 clhash->hash = qdisc_class_hash_alloc(size);
587 if (clhash->hash == NULL)
588 return -ENOMEM;
589 clhash->hashsize = size;
590 clhash->hashmask = size - 1;
591 clhash->hashelems = 0;
592 return 0;
593}
594EXPORT_SYMBOL(qdisc_class_hash_init);
595
596void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
597{
598 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
599}
600EXPORT_SYMBOL(qdisc_class_hash_destroy);
601
602void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
603 struct Qdisc_class_common *cl)
604{
605 unsigned int h;
606
607 INIT_HLIST_NODE(&cl->hnode);
608 h = qdisc_class_hash(cl->classid, clhash->hashmask);
609 hlist_add_head(&cl->hnode, &clhash->hash[h]);
610 clhash->hashelems++;
611}
612EXPORT_SYMBOL(qdisc_class_hash_insert);
613
614void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
615 struct Qdisc_class_common *cl)
616{
617 hlist_del(&cl->hnode);
618 clhash->hashelems--;
619}
620EXPORT_SYMBOL(qdisc_class_hash_remove);
621
622/* Allocate an unique handle from space managed by kernel
623 * Possible range is [8000-FFFF]:0000 (0x8000 values)
624 */
625static u32 qdisc_alloc_handle(struct net_device *dev)
626{
627 int i = 0x8000;
628 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
629
630 do {
631 autohandle += TC_H_MAKE(0x10000U, 0);
632 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
633 autohandle = TC_H_MAKE(0x80000000U, 0);
634 if (!qdisc_lookup(dev, autohandle))
635 return autohandle;
636 cond_resched();
637 } while (--i > 0);
638
639 return 0;
640}
641
642void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
643{
644 const struct Qdisc_class_ops *cops;
645 unsigned long cl;
646 u32 parentid;
647
648 if (n == 0)
649 return;
650 while ((parentid = sch->parent)) {
651 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
652 return;
653
654 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
655 if (sch == NULL) {
656 WARN_ON(parentid != TC_H_ROOT);
657 return;
658 }
659 cops = sch->ops->cl_ops;
660 if (cops->qlen_notify) {
661 cl = cops->get(sch, parentid);
662 cops->qlen_notify(sch, cl);
663 cops->put(sch, cl);
664 }
665 sch->q.qlen -= n;
666 }
667}
668EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
669
670static void notify_and_destroy(struct net *net, struct sk_buff *skb,
671 struct nlmsghdr *n, u32 clid,
672 struct Qdisc *old, struct Qdisc *new)
673{
674 if (new || old)
675 qdisc_notify(net, skb, n, clid, old, new);
676
677 if (old)
678 qdisc_destroy(old);
679}
680
681/* Graft qdisc "new" to class "classid" of qdisc "parent" or
682 * to device "dev".
683 *
684 * When appropriate send a netlink notification using 'skb'
685 * and "n".
686 *
687 * On success, destroy old qdisc.
688 */
689
690static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
691 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
692 struct Qdisc *new, struct Qdisc *old)
693{
694 struct Qdisc *q = old;
695 struct net *net = dev_net(dev);
696 int err = 0;
697
698 if (parent == NULL) {
699 unsigned int i, num_q, ingress;
700
701 ingress = 0;
702 num_q = dev->num_tx_queues;
703 if ((q && q->flags & TCQ_F_INGRESS) ||
704 (new && new->flags & TCQ_F_INGRESS)) {
705 num_q = 1;
706 ingress = 1;
707 if (!dev_ingress_queue(dev))
708 return -ENOENT;
709 }
710
711 if (dev->flags & IFF_UP)
712 dev_deactivate(dev);
713
714 if (new && new->ops->attach) {
715 new->ops->attach(new);
716 num_q = 0;
717 }
718
719 for (i = 0; i < num_q; i++) {
720 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
721
722 if (!ingress)
723 dev_queue = netdev_get_tx_queue(dev, i);
724
725 old = dev_graft_qdisc(dev_queue, new);
726 if (new && i > 0)
727 atomic_inc(&new->refcnt);
728
729 if (!ingress)
730 qdisc_destroy(old);
731 }
732
733 if (!ingress) {
734 notify_and_destroy(net, skb, n, classid,
735 dev->qdisc, new);
736 if (new && !new->ops->attach)
737 atomic_inc(&new->refcnt);
738 dev->qdisc = new ? : &noop_qdisc;
739 } else {
740 notify_and_destroy(net, skb, n, classid, old, new);
741 }
742
743 if (dev->flags & IFF_UP)
744 dev_activate(dev);
745 } else {
746 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
747
748 err = -EOPNOTSUPP;
749 if (cops && cops->graft) {
750 unsigned long cl = cops->get(parent, classid);
751 if (cl) {
752 err = cops->graft(parent, cl, new, &old);
753 cops->put(parent, cl);
754 } else
755 err = -ENOENT;
756 }
757 if (!err)
758 notify_and_destroy(net, skb, n, classid, old, new);
759 }
760 return err;
761}
762
763/* lockdep annotation is needed for ingress; egress gets it only for name */
764static struct lock_class_key qdisc_tx_lock;
765static struct lock_class_key qdisc_rx_lock;
766
767/*
768 Allocate and initialize new qdisc.
769
770 Parameters are passed via opt.
771 */
772
773static struct Qdisc *
774qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
775 struct Qdisc *p, u32 parent, u32 handle,
776 struct nlattr **tca, int *errp)
777{
778 int err;
779 struct nlattr *kind = tca[TCA_KIND];
780 struct Qdisc *sch;
781 struct Qdisc_ops *ops;
782 struct qdisc_size_table *stab;
783
784 ops = qdisc_lookup_ops(kind);
785#ifdef CONFIG_MODULES
786 if (ops == NULL && kind != NULL) {
787 char name[IFNAMSIZ];
788 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
789 /* We dropped the RTNL semaphore in order to
790 * perform the module load. So, even if we
791 * succeeded in loading the module we have to
792 * tell the caller to replay the request. We
793 * indicate this using -EAGAIN.
794 * We replay the request because the device may
795 * go away in the mean time.
796 */
797 rtnl_unlock();
798 request_module("sch_%s", name);
799 rtnl_lock();
800 ops = qdisc_lookup_ops(kind);
801 if (ops != NULL) {
802 /* We will try again qdisc_lookup_ops,
803 * so don't keep a reference.
804 */
805 module_put(ops->owner);
806 err = -EAGAIN;
807 goto err_out;
808 }
809 }
810 }
811#endif
812
813 err = -ENOENT;
814 if (ops == NULL)
815 goto err_out;
816
817 sch = qdisc_alloc(dev_queue, ops);
818 if (IS_ERR(sch)) {
819 err = PTR_ERR(sch);
820 goto err_out2;
821 }
822
823 sch->parent = parent;
824
825 if (handle == TC_H_INGRESS) {
826 sch->flags |= TCQ_F_INGRESS;
827 handle = TC_H_MAKE(TC_H_INGRESS, 0);
828 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
829 } else {
830 if (handle == 0) {
831 handle = qdisc_alloc_handle(dev);
832 err = -ENOMEM;
833 if (handle == 0)
834 goto err_out3;
835 }
836 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
837 }
838
839 sch->handle = handle;
840
841 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
842 if (tca[TCA_STAB]) {
843 stab = qdisc_get_stab(tca[TCA_STAB]);
844 if (IS_ERR(stab)) {
845 err = PTR_ERR(stab);
846 goto err_out4;
847 }
848 rcu_assign_pointer(sch->stab, stab);
849 }
850 if (tca[TCA_RATE]) {
851 spinlock_t *root_lock;
852
853 err = -EOPNOTSUPP;
854 if (sch->flags & TCQ_F_MQROOT)
855 goto err_out4;
856
857 if ((sch->parent != TC_H_ROOT) &&
858 !(sch->flags & TCQ_F_INGRESS) &&
859 (!p || !(p->flags & TCQ_F_MQROOT)))
860 root_lock = qdisc_root_sleeping_lock(sch);
861 else
862 root_lock = qdisc_lock(sch);
863
864 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
865 root_lock, tca[TCA_RATE]);
866 if (err)
867 goto err_out4;
868 }
869
870 qdisc_list_add(sch);
871
872 return sch;
873 }
874err_out3:
875 dev_put(dev);
876 kfree((char *) sch - sch->padded);
877err_out2:
878 module_put(ops->owner);
879err_out:
880 *errp = err;
881 return NULL;
882
883err_out4:
884 /*
885 * Any broken qdiscs that would require a ops->reset() here?
886 * The qdisc was never in action so it shouldn't be necessary.
887 */
888 qdisc_put_stab(rtnl_dereference(sch->stab));
889 if (ops->destroy)
890 ops->destroy(sch);
891 goto err_out3;
892}
893
894static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
895{
896 struct qdisc_size_table *ostab, *stab = NULL;
897 int err = 0;
898
899 if (tca[TCA_OPTIONS]) {
900 if (sch->ops->change == NULL)
901 return -EINVAL;
902 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
903 if (err)
904 return err;
905 }
906
907 if (tca[TCA_STAB]) {
908 stab = qdisc_get_stab(tca[TCA_STAB]);
909 if (IS_ERR(stab))
910 return PTR_ERR(stab);
911 }
912
913 ostab = rtnl_dereference(sch->stab);
914 rcu_assign_pointer(sch->stab, stab);
915 qdisc_put_stab(ostab);
916
917 if (tca[TCA_RATE]) {
918 /* NB: ignores errors from replace_estimator
919 because change can't be undone. */
920 if (sch->flags & TCQ_F_MQROOT)
921 goto out;
922 gen_replace_estimator(&sch->bstats, &sch->rate_est,
923 qdisc_root_sleeping_lock(sch),
924 tca[TCA_RATE]);
925 }
926out:
927 return 0;
928}
929
930struct check_loop_arg {
931 struct qdisc_walker w;
932 struct Qdisc *p;
933 int depth;
934};
935
936static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
937
938static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
939{
940 struct check_loop_arg arg;
941
942 if (q->ops->cl_ops == NULL)
943 return 0;
944
945 arg.w.stop = arg.w.skip = arg.w.count = 0;
946 arg.w.fn = check_loop_fn;
947 arg.depth = depth;
948 arg.p = p;
949 q->ops->cl_ops->walk(q, &arg.w);
950 return arg.w.stop ? -ELOOP : 0;
951}
952
953static int
954check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
955{
956 struct Qdisc *leaf;
957 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
958 struct check_loop_arg *arg = (struct check_loop_arg *)w;
959
960 leaf = cops->leaf(q, cl);
961 if (leaf) {
962 if (leaf == arg->p || arg->depth > 7)
963 return -ELOOP;
964 return check_loop(leaf, arg->p, arg->depth + 1);
965 }
966 return 0;
967}
968
969/*
970 * Delete/get qdisc.
971 */
972
973static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
974{
975 struct net *net = sock_net(skb->sk);
976 struct tcmsg *tcm = NLMSG_DATA(n);
977 struct nlattr *tca[TCA_MAX + 1];
978 struct net_device *dev;
979 u32 clid = tcm->tcm_parent;
980 struct Qdisc *q = NULL;
981 struct Qdisc *p = NULL;
982 int err;
983
984 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
985 if (!dev)
986 return -ENODEV;
987
988 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
989 if (err < 0)
990 return err;
991
992 if (clid) {
993 if (clid != TC_H_ROOT) {
994 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
995 p = qdisc_lookup(dev, TC_H_MAJ(clid));
996 if (!p)
997 return -ENOENT;
998 q = qdisc_leaf(p, clid);
999 } else if (dev_ingress_queue(dev)) {
1000 q = dev_ingress_queue(dev)->qdisc_sleeping;
1001 }
1002 } else {
1003 q = dev->qdisc;
1004 }
1005 if (!q)
1006 return -ENOENT;
1007
1008 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1009 return -EINVAL;
1010 } else {
1011 q = qdisc_lookup(dev, tcm->tcm_handle);
1012 if (!q)
1013 return -ENOENT;
1014 }
1015
1016 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1017 return -EINVAL;
1018
1019 if (n->nlmsg_type == RTM_DELQDISC) {
1020 if (!clid)
1021 return -EINVAL;
1022 if (q->handle == 0)
1023 return -ENOENT;
1024 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1025 if (err != 0)
1026 return err;
1027 } else {
1028 qdisc_notify(net, skb, n, clid, NULL, q);
1029 }
1030 return 0;
1031}
1032
1033/*
1034 * Create/change qdisc.
1035 */
1036
1037static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1038{
1039 struct net *net = sock_net(skb->sk);
1040 struct tcmsg *tcm;
1041 struct nlattr *tca[TCA_MAX + 1];
1042 struct net_device *dev;
1043 u32 clid;
1044 struct Qdisc *q, *p;
1045 int err;
1046
1047replay:
1048 /* Reinit, just in case something touches this. */
1049 tcm = NLMSG_DATA(n);
1050 clid = tcm->tcm_parent;
1051 q = p = NULL;
1052
1053 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1054 if (!dev)
1055 return -ENODEV;
1056
1057 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1058 if (err < 0)
1059 return err;
1060
1061 if (clid) {
1062 if (clid != TC_H_ROOT) {
1063 if (clid != TC_H_INGRESS) {
1064 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1065 if (!p)
1066 return -ENOENT;
1067 q = qdisc_leaf(p, clid);
1068 } else if (dev_ingress_queue_create(dev)) {
1069 q = dev_ingress_queue(dev)->qdisc_sleeping;
1070 }
1071 } else {
1072 q = dev->qdisc;
1073 }
1074
1075 /* It may be default qdisc, ignore it */
1076 if (q && q->handle == 0)
1077 q = NULL;
1078
1079 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1080 if (tcm->tcm_handle) {
1081 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1082 return -EEXIST;
1083 if (TC_H_MIN(tcm->tcm_handle))
1084 return -EINVAL;
1085 q = qdisc_lookup(dev, tcm->tcm_handle);
1086 if (!q)
1087 goto create_n_graft;
1088 if (n->nlmsg_flags & NLM_F_EXCL)
1089 return -EEXIST;
1090 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1091 return -EINVAL;
1092 if (q == p ||
1093 (p && check_loop(q, p, 0)))
1094 return -ELOOP;
1095 atomic_inc(&q->refcnt);
1096 goto graft;
1097 } else {
1098 if (!q)
1099 goto create_n_graft;
1100
1101 /* This magic test requires explanation.
1102 *
1103 * We know, that some child q is already
1104 * attached to this parent and have choice:
1105 * either to change it or to create/graft new one.
1106 *
1107 * 1. We are allowed to create/graft only
1108 * if CREATE and REPLACE flags are set.
1109 *
1110 * 2. If EXCL is set, requestor wanted to say,
1111 * that qdisc tcm_handle is not expected
1112 * to exist, so that we choose create/graft too.
1113 *
1114 * 3. The last case is when no flags are set.
1115 * Alas, it is sort of hole in API, we
1116 * cannot decide what to do unambiguously.
1117 * For now we select create/graft, if
1118 * user gave KIND, which does not match existing.
1119 */
1120 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1121 (n->nlmsg_flags & NLM_F_REPLACE) &&
1122 ((n->nlmsg_flags & NLM_F_EXCL) ||
1123 (tca[TCA_KIND] &&
1124 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1125 goto create_n_graft;
1126 }
1127 }
1128 } else {
1129 if (!tcm->tcm_handle)
1130 return -EINVAL;
1131 q = qdisc_lookup(dev, tcm->tcm_handle);
1132 }
1133
1134 /* Change qdisc parameters */
1135 if (q == NULL)
1136 return -ENOENT;
1137 if (n->nlmsg_flags & NLM_F_EXCL)
1138 return -EEXIST;
1139 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1140 return -EINVAL;
1141 err = qdisc_change(q, tca);
1142 if (err == 0)
1143 qdisc_notify(net, skb, n, clid, NULL, q);
1144 return err;
1145
1146create_n_graft:
1147 if (!(n->nlmsg_flags & NLM_F_CREATE))
1148 return -ENOENT;
1149 if (clid == TC_H_INGRESS) {
1150 if (dev_ingress_queue(dev))
1151 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1152 tcm->tcm_parent, tcm->tcm_parent,
1153 tca, &err);
1154 else
1155 err = -ENOENT;
1156 } else {
1157 struct netdev_queue *dev_queue;
1158
1159 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1160 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1161 else if (p)
1162 dev_queue = p->dev_queue;
1163 else
1164 dev_queue = netdev_get_tx_queue(dev, 0);
1165
1166 q = qdisc_create(dev, dev_queue, p,
1167 tcm->tcm_parent, tcm->tcm_handle,
1168 tca, &err);
1169 }
1170 if (q == NULL) {
1171 if (err == -EAGAIN)
1172 goto replay;
1173 return err;
1174 }
1175
1176graft:
1177 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1178 if (err) {
1179 if (q)
1180 qdisc_destroy(q);
1181 return err;
1182 }
1183
1184 return 0;
1185}
1186
1187static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1188 u32 pid, u32 seq, u16 flags, int event)
1189{
1190 struct tcmsg *tcm;
1191 struct nlmsghdr *nlh;
1192 unsigned char *b = skb_tail_pointer(skb);
1193 struct gnet_dump d;
1194 struct qdisc_size_table *stab;
1195
1196 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1197 tcm = NLMSG_DATA(nlh);
1198 tcm->tcm_family = AF_UNSPEC;
1199 tcm->tcm__pad1 = 0;
1200 tcm->tcm__pad2 = 0;
1201 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1202 tcm->tcm_parent = clid;
1203 tcm->tcm_handle = q->handle;
1204 tcm->tcm_info = atomic_read(&q->refcnt);
1205 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1206 goto nla_put_failure;
1207 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1208 goto nla_put_failure;
1209 q->qstats.qlen = q->q.qlen;
1210
1211 stab = rtnl_dereference(q->stab);
1212 if (stab && qdisc_dump_stab(skb, stab) < 0)
1213 goto nla_put_failure;
1214
1215 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1216 qdisc_root_sleeping_lock(q), &d) < 0)
1217 goto nla_put_failure;
1218
1219 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1220 goto nla_put_failure;
1221
1222 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1223 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1224 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1225 goto nla_put_failure;
1226
1227 if (gnet_stats_finish_copy(&d) < 0)
1228 goto nla_put_failure;
1229
1230 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1231 return skb->len;
1232
1233nlmsg_failure:
1234nla_put_failure:
1235 nlmsg_trim(skb, b);
1236 return -1;
1237}
1238
1239static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1240{
1241 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1242}
1243
1244static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1245 struct nlmsghdr *n, u32 clid,
1246 struct Qdisc *old, struct Qdisc *new)
1247{
1248 struct sk_buff *skb;
1249 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1250
1251 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1252 if (!skb)
1253 return -ENOBUFS;
1254
1255 if (old && !tc_qdisc_dump_ignore(old)) {
1256 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq,
1257 0, RTM_DELQDISC) < 0)
1258 goto err_out;
1259 }
1260 if (new && !tc_qdisc_dump_ignore(new)) {
1261 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq,
1262 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1263 goto err_out;
1264 }
1265
1266 if (skb->len)
1267 return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
1268 n->nlmsg_flags & NLM_F_ECHO);
1269
1270err_out:
1271 kfree_skb(skb);
1272 return -EINVAL;
1273}
1274
1275static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1276 struct netlink_callback *cb,
1277 int *q_idx_p, int s_q_idx)
1278{
1279 int ret = 0, q_idx = *q_idx_p;
1280 struct Qdisc *q;
1281
1282 if (!root)
1283 return 0;
1284
1285 q = root;
1286 if (q_idx < s_q_idx) {
1287 q_idx++;
1288 } else {
1289 if (!tc_qdisc_dump_ignore(q) &&
1290 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1291 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1292 goto done;
1293 q_idx++;
1294 }
1295 list_for_each_entry(q, &root->list, list) {
1296 if (q_idx < s_q_idx) {
1297 q_idx++;
1298 continue;
1299 }
1300 if (!tc_qdisc_dump_ignore(q) &&
1301 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1302 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1303 goto done;
1304 q_idx++;
1305 }
1306
1307out:
1308 *q_idx_p = q_idx;
1309 return ret;
1310done:
1311 ret = -1;
1312 goto out;
1313}
1314
1315static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1316{
1317 struct net *net = sock_net(skb->sk);
1318 int idx, q_idx;
1319 int s_idx, s_q_idx;
1320 struct net_device *dev;
1321
1322 s_idx = cb->args[0];
1323 s_q_idx = q_idx = cb->args[1];
1324
1325 rcu_read_lock();
1326 idx = 0;
1327 for_each_netdev_rcu(net, dev) {
1328 struct netdev_queue *dev_queue;
1329
1330 if (idx < s_idx)
1331 goto cont;
1332 if (idx > s_idx)
1333 s_q_idx = 0;
1334 q_idx = 0;
1335
1336 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1337 goto done;
1338
1339 dev_queue = dev_ingress_queue(dev);
1340 if (dev_queue &&
1341 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1342 &q_idx, s_q_idx) < 0)
1343 goto done;
1344
1345cont:
1346 idx++;
1347 }
1348
1349done:
1350 rcu_read_unlock();
1351
1352 cb->args[0] = idx;
1353 cb->args[1] = q_idx;
1354
1355 return skb->len;
1356}
1357
1358
1359
1360/************************************************
1361 * Traffic classes manipulation. *
1362 ************************************************/
1363
1364
1365
1366static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1367{
1368 struct net *net = sock_net(skb->sk);
1369 struct tcmsg *tcm = NLMSG_DATA(n);
1370 struct nlattr *tca[TCA_MAX + 1];
1371 struct net_device *dev;
1372 struct Qdisc *q = NULL;
1373 const struct Qdisc_class_ops *cops;
1374 unsigned long cl = 0;
1375 unsigned long new_cl;
1376 u32 pid = tcm->tcm_parent;
1377 u32 clid = tcm->tcm_handle;
1378 u32 qid = TC_H_MAJ(clid);
1379 int err;
1380
1381 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1382 if (!dev)
1383 return -ENODEV;
1384
1385 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1386 if (err < 0)
1387 return err;
1388
1389 /*
1390 parent == TC_H_UNSPEC - unspecified parent.
1391 parent == TC_H_ROOT - class is root, which has no parent.
1392 parent == X:0 - parent is root class.
1393 parent == X:Y - parent is a node in hierarchy.
1394 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1395
1396 handle == 0:0 - generate handle from kernel pool.
1397 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1398 handle == X:Y - clear.
1399 handle == X:0 - root class.
1400 */
1401
1402 /* Step 1. Determine qdisc handle X:0 */
1403
1404 if (pid != TC_H_ROOT) {
1405 u32 qid1 = TC_H_MAJ(pid);
1406
1407 if (qid && qid1) {
1408 /* If both majors are known, they must be identical. */
1409 if (qid != qid1)
1410 return -EINVAL;
1411 } else if (qid1) {
1412 qid = qid1;
1413 } else if (qid == 0)
1414 qid = dev->qdisc->handle;
1415
1416 /* Now qid is genuine qdisc handle consistent
1417 * both with parent and child.
1418 *
1419 * TC_H_MAJ(pid) still may be unspecified, complete it now.
1420 */
1421 if (pid)
1422 pid = TC_H_MAKE(qid, pid);
1423 } else {
1424 if (qid == 0)
1425 qid = dev->qdisc->handle;
1426 }
1427
1428 /* OK. Locate qdisc */
1429 q = qdisc_lookup(dev, qid);
1430 if (!q)
1431 return -ENOENT;
1432
1433 /* An check that it supports classes */
1434 cops = q->ops->cl_ops;
1435 if (cops == NULL)
1436 return -EINVAL;
1437
1438 /* Now try to get class */
1439 if (clid == 0) {
1440 if (pid == TC_H_ROOT)
1441 clid = qid;
1442 } else
1443 clid = TC_H_MAKE(qid, clid);
1444
1445 if (clid)
1446 cl = cops->get(q, clid);
1447
1448 if (cl == 0) {
1449 err = -ENOENT;
1450 if (n->nlmsg_type != RTM_NEWTCLASS ||
1451 !(n->nlmsg_flags & NLM_F_CREATE))
1452 goto out;
1453 } else {
1454 switch (n->nlmsg_type) {
1455 case RTM_NEWTCLASS:
1456 err = -EEXIST;
1457 if (n->nlmsg_flags & NLM_F_EXCL)
1458 goto out;
1459 break;
1460 case RTM_DELTCLASS:
1461 err = -EOPNOTSUPP;
1462 if (cops->delete)
1463 err = cops->delete(q, cl);
1464 if (err == 0)
1465 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1466 goto out;
1467 case RTM_GETTCLASS:
1468 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1469 goto out;
1470 default:
1471 err = -EINVAL;
1472 goto out;
1473 }
1474 }
1475
1476 new_cl = cl;
1477 err = -EOPNOTSUPP;
1478 if (cops->change)
1479 err = cops->change(q, clid, pid, tca, &new_cl);
1480 if (err == 0)
1481 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1482
1483out:
1484 if (cl)
1485 cops->put(q, cl);
1486
1487 return err;
1488}
1489
1490
1491static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1492 unsigned long cl,
1493 u32 pid, u32 seq, u16 flags, int event)
1494{
1495 struct tcmsg *tcm;
1496 struct nlmsghdr *nlh;
1497 unsigned char *b = skb_tail_pointer(skb);
1498 struct gnet_dump d;
1499 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1500
1501 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1502 tcm = NLMSG_DATA(nlh);
1503 tcm->tcm_family = AF_UNSPEC;
1504 tcm->tcm__pad1 = 0;
1505 tcm->tcm__pad2 = 0;
1506 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1507 tcm->tcm_parent = q->handle;
1508 tcm->tcm_handle = q->handle;
1509 tcm->tcm_info = 0;
1510 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1511 goto nla_put_failure;
1512 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1513 goto nla_put_failure;
1514
1515 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1516 qdisc_root_sleeping_lock(q), &d) < 0)
1517 goto nla_put_failure;
1518
1519 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1520 goto nla_put_failure;
1521
1522 if (gnet_stats_finish_copy(&d) < 0)
1523 goto nla_put_failure;
1524
1525 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1526 return skb->len;
1527
1528nlmsg_failure:
1529nla_put_failure:
1530 nlmsg_trim(skb, b);
1531 return -1;
1532}
1533
1534static int tclass_notify(struct net *net, struct sk_buff *oskb,
1535 struct nlmsghdr *n, struct Qdisc *q,
1536 unsigned long cl, int event)
1537{
1538 struct sk_buff *skb;
1539 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1540
1541 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1542 if (!skb)
1543 return -ENOBUFS;
1544
1545 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1546 kfree_skb(skb);
1547 return -EINVAL;
1548 }
1549
1550 return rtnetlink_send(skb, net, pid, RTNLGRP_TC,
1551 n->nlmsg_flags & NLM_F_ECHO);
1552}
1553
1554struct qdisc_dump_args {
1555 struct qdisc_walker w;
1556 struct sk_buff *skb;
1557 struct netlink_callback *cb;
1558};
1559
1560static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1561{
1562 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1563
1564 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1565 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1566}
1567
1568static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1569 struct tcmsg *tcm, struct netlink_callback *cb,
1570 int *t_p, int s_t)
1571{
1572 struct qdisc_dump_args arg;
1573
1574 if (tc_qdisc_dump_ignore(q) ||
1575 *t_p < s_t || !q->ops->cl_ops ||
1576 (tcm->tcm_parent &&
1577 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1578 (*t_p)++;
1579 return 0;
1580 }
1581 if (*t_p > s_t)
1582 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1583 arg.w.fn = qdisc_class_dump;
1584 arg.skb = skb;
1585 arg.cb = cb;
1586 arg.w.stop = 0;
1587 arg.w.skip = cb->args[1];
1588 arg.w.count = 0;
1589 q->ops->cl_ops->walk(q, &arg.w);
1590 cb->args[1] = arg.w.count;
1591 if (arg.w.stop)
1592 return -1;
1593 (*t_p)++;
1594 return 0;
1595}
1596
1597static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1598 struct tcmsg *tcm, struct netlink_callback *cb,
1599 int *t_p, int s_t)
1600{
1601 struct Qdisc *q;
1602
1603 if (!root)
1604 return 0;
1605
1606 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1607 return -1;
1608
1609 list_for_each_entry(q, &root->list, list) {
1610 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1611 return -1;
1612 }
1613
1614 return 0;
1615}
1616
1617static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1618{
1619 struct tcmsg *tcm = (struct tcmsg *)NLMSG_DATA(cb->nlh);
1620 struct net *net = sock_net(skb->sk);
1621 struct netdev_queue *dev_queue;
1622 struct net_device *dev;
1623 int t, s_t;
1624
1625 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1626 return 0;
1627 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1628 if (!dev)
1629 return 0;
1630
1631 s_t = cb->args[0];
1632 t = 0;
1633
1634 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1635 goto done;
1636
1637 dev_queue = dev_ingress_queue(dev);
1638 if (dev_queue &&
1639 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1640 &t, s_t) < 0)
1641 goto done;
1642
1643done:
1644 cb->args[0] = t;
1645
1646 dev_put(dev);
1647 return skb->len;
1648}
1649
1650/* Main classifier routine: scans classifier chain attached
1651 * to this qdisc, (optionally) tests for protocol and asks
1652 * specific classifiers.
1653 */
1654int tc_classify_compat(struct sk_buff *skb, const struct tcf_proto *tp,
1655 struct tcf_result *res)
1656{
1657 __be16 protocol = skb->protocol;
1658 int err;
1659
1660 for (; tp; tp = tp->next) {
1661 if (tp->protocol != protocol &&
1662 tp->protocol != htons(ETH_P_ALL))
1663 continue;
1664 err = tp->classify(skb, tp, res);
1665
1666 if (err >= 0) {
1667#ifdef CONFIG_NET_CLS_ACT
1668 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1669 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1670#endif
1671 return err;
1672 }
1673 }
1674 return -1;
1675}
1676EXPORT_SYMBOL(tc_classify_compat);
1677
1678int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1679 struct tcf_result *res)
1680{
1681 int err = 0;
1682#ifdef CONFIG_NET_CLS_ACT
1683 const struct tcf_proto *otp = tp;
1684reclassify:
1685#endif
1686
1687 err = tc_classify_compat(skb, tp, res);
1688#ifdef CONFIG_NET_CLS_ACT
1689 if (err == TC_ACT_RECLASSIFY) {
1690 u32 verd = G_TC_VERD(skb->tc_verd);
1691 tp = otp;
1692
1693 if (verd++ >= MAX_REC_LOOP) {
1694 net_notice_ratelimited("%s: packet reclassify loop rule prio %u protocol %02x\n",
1695 tp->q->ops->id,
1696 tp->prio & 0xffff,
1697 ntohs(tp->protocol));
1698 return TC_ACT_SHOT;
1699 }
1700 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1701 goto reclassify;
1702 }
1703#endif
1704 return err;
1705}
1706EXPORT_SYMBOL(tc_classify);
1707
1708void tcf_destroy(struct tcf_proto *tp)
1709{
1710 tp->ops->destroy(tp);
1711 module_put(tp->ops->owner);
1712 kfree(tp);
1713}
1714
1715void tcf_destroy_chain(struct tcf_proto **fl)
1716{
1717 struct tcf_proto *tp;
1718
1719 while ((tp = *fl) != NULL) {
1720 *fl = tp->next;
1721 tcf_destroy(tp);
1722 }
1723}
1724EXPORT_SYMBOL(tcf_destroy_chain);
1725
1726#ifdef CONFIG_PROC_FS
1727static int psched_show(struct seq_file *seq, void *v)
1728{
1729 struct timespec ts;
1730
1731 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1732 seq_printf(seq, "%08x %08x %08x %08x\n",
1733 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1734 1000000,
1735 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1736
1737 return 0;
1738}
1739
1740static int psched_open(struct inode *inode, struct file *file)
1741{
1742 return single_open(file, psched_show, NULL);
1743}
1744
1745static const struct file_operations psched_fops = {
1746 .owner = THIS_MODULE,
1747 .open = psched_open,
1748 .read = seq_read,
1749 .llseek = seq_lseek,
1750 .release = single_release,
1751};
1752
1753static int __net_init psched_net_init(struct net *net)
1754{
1755 struct proc_dir_entry *e;
1756
1757 e = proc_net_fops_create(net, "psched", 0, &psched_fops);
1758 if (e == NULL)
1759 return -ENOMEM;
1760
1761 return 0;
1762}
1763
1764static void __net_exit psched_net_exit(struct net *net)
1765{
1766 proc_net_remove(net, "psched");
1767}
1768#else
1769static int __net_init psched_net_init(struct net *net)
1770{
1771 return 0;
1772}
1773
1774static void __net_exit psched_net_exit(struct net *net)
1775{
1776}
1777#endif
1778
1779static struct pernet_operations psched_net_ops = {
1780 .init = psched_net_init,
1781 .exit = psched_net_exit,
1782};
1783
1784static int __init pktsched_init(void)
1785{
1786 int err;
1787
1788 err = register_pernet_subsys(&psched_net_ops);
1789 if (err) {
1790 pr_err("pktsched_init: "
1791 "cannot initialize per netns operations\n");
1792 return err;
1793 }
1794
1795 register_qdisc(&pfifo_qdisc_ops);
1796 register_qdisc(&bfifo_qdisc_ops);
1797 register_qdisc(&pfifo_head_drop_qdisc_ops);
1798 register_qdisc(&mq_qdisc_ops);
1799
1800 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1801 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1802 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1803 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1804 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1805 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1806
1807 return 0;
1808}
1809
1810subsys_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);