projects / GitHub / LineageOS / android_kernel_samsung_universal7580.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
BACKPORT: binder: Set end of SG buffer area properly.
[GitHub/LineageOS/android_kernel_samsung_universal7580.git] / net / sched / sch_api.c
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
38 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
39 struct nlmsghdr *n, u32 clid,
40 struct Qdisc *old, struct Qdisc *new);
41 static 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. */
126 static DEFINE_RWLOCK(qdisc_mod_lock);
127
128
129 /************************************************
130 * Queueing disciplines manipulation. *
131 ************************************************/
132
133
134 /* The list of all installed queueing disciplines. */
135
136 static struct Qdisc_ops *qdisc_base;
137
138 /* Register/uregister queueing discipline */
139
140 int 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;
174 out:
175 write_unlock(&qdisc_mod_lock);
176 return rc;
177
178 out_einval:
179 rc = -EINVAL;
180 goto out;
181 }
182 EXPORT_SYMBOL(register_qdisc);
183
184 int 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 }
201 EXPORT_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
207 static 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
222 static 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
228 void 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 }
233 EXPORT_SYMBOL(qdisc_list_del);
234
235 struct 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);
247 out:
248 return q;
249 }
250
251 static 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
270 static 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
288 /* The linklayer setting were not transferred from iproute2, in older
289 * versions, and the rate tables lookup systems have been dropped in
290 * the kernel. To keep backward compatible with older iproute2 tc
291 * utils, we detect the linklayer setting by detecting if the rate
292 * table were modified.
293 *
294 * For linklayer ATM table entries, the rate table will be aligned to
295 * 48 bytes, thus some table entries will contain the same value. The
296 * mpu (min packet unit) is also encoded into the old rate table, thus
297 * starting from the mpu, we find low and high table entries for
298 * mapping this cell. If these entries contain the same value, when
299 * the rate tables have been modified for linklayer ATM.
300 *
301 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
302 * and then roundup to the next cell, calc the table entry one below,
303 * and compare.
304 */
305 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
306 {
307 int low = roundup(r->mpu, 48);
308 int high = roundup(low+1, 48);
309 int cell_low = low >> r->cell_log;
310 int cell_high = (high >> r->cell_log) - 1;
311
312 /* rtab is too inaccurate at rates > 100Mbit/s */
313 if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
314 pr_debug("TC linklayer: Giving up ATM detection\n");
315 return TC_LINKLAYER_ETHERNET;
316 }
317
318 if ((cell_high > cell_low) && (cell_high < 256)
319 && (rtab[cell_low] == rtab[cell_high])) {
320 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
321 cell_low, cell_high, rtab[cell_high]);
322 return TC_LINKLAYER_ATM;
323 }
324 return TC_LINKLAYER_ETHERNET;
325 }
326
327 static struct qdisc_rate_table *qdisc_rtab_list;
328
329 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
330 {
331 struct qdisc_rate_table *rtab;
332
333 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
334 nla_len(tab) != TC_RTAB_SIZE)
335 return NULL;
336
337 for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
338 if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
339 !memcmp(&rtab->data, nla_data(tab), 1024)) {
340 rtab->refcnt++;
341 return rtab;
342 }
343 }
344
345 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
346 if (rtab) {
347 rtab->rate = *r;
348 rtab->refcnt = 1;
349 memcpy(rtab->data, nla_data(tab), 1024);
350 if (r->linklayer == TC_LINKLAYER_UNAWARE)
351 r->linklayer = __detect_linklayer(r, rtab->data);
352 rtab->next = qdisc_rtab_list;
353 qdisc_rtab_list = rtab;
354 }
355 return rtab;
356 }
357 EXPORT_SYMBOL(qdisc_get_rtab);
358
359 void qdisc_put_rtab(struct qdisc_rate_table *tab)
360 {
361 struct qdisc_rate_table *rtab, **rtabp;
362
363 if (!tab || --tab->refcnt)
364 return;
365
366 for (rtabp = &qdisc_rtab_list;
367 (rtab = *rtabp) != NULL;
368 rtabp = &rtab->next) {
369 if (rtab == tab) {
370 *rtabp = rtab->next;
371 kfree(rtab);
372 return;
373 }
374 }
375 }
376 EXPORT_SYMBOL(qdisc_put_rtab);
377
378 static LIST_HEAD(qdisc_stab_list);
379 static DEFINE_SPINLOCK(qdisc_stab_lock);
380
381 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
382 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
383 [TCA_STAB_DATA] = { .type = NLA_BINARY },
384 };
385
386 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
387 {
388 struct nlattr *tb[TCA_STAB_MAX + 1];
389 struct qdisc_size_table *stab;
390 struct tc_sizespec *s;
391 unsigned int tsize = 0;
392 u16 *tab = NULL;
393 int err;
394
395 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
396 if (err < 0)
397 return ERR_PTR(err);
398 if (!tb[TCA_STAB_BASE])
399 return ERR_PTR(-EINVAL);
400
401 s = nla_data(tb[TCA_STAB_BASE]);
402
403 if (s->tsize > 0) {
404 if (!tb[TCA_STAB_DATA])
405 return ERR_PTR(-EINVAL);
406 tab = nla_data(tb[TCA_STAB_DATA]);
407 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
408 }
409
410 if (tsize != s->tsize || (!tab && tsize > 0))
411 return ERR_PTR(-EINVAL);
412
413 spin_lock(&qdisc_stab_lock);
414
415 list_for_each_entry(stab, &qdisc_stab_list, list) {
416 if (memcmp(&stab->szopts, s, sizeof(*s)))
417 continue;
418 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
419 continue;
420 stab->refcnt++;
421 spin_unlock(&qdisc_stab_lock);
422 return stab;
423 }
424
425 spin_unlock(&qdisc_stab_lock);
426
427 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
428 if (!stab)
429 return ERR_PTR(-ENOMEM);
430
431 stab->refcnt = 1;
432 stab->szopts = *s;
433 if (tsize > 0)
434 memcpy(stab->data, tab, tsize * sizeof(u16));
435
436 spin_lock(&qdisc_stab_lock);
437 list_add_tail(&stab->list, &qdisc_stab_list);
438 spin_unlock(&qdisc_stab_lock);
439
440 return stab;
441 }
442
443 static void stab_kfree_rcu(struct rcu_head *head)
444 {
445 kfree(container_of(head, struct qdisc_size_table, rcu));
446 }
447
448 void qdisc_put_stab(struct qdisc_size_table *tab)
449 {
450 if (!tab)
451 return;
452
453 spin_lock(&qdisc_stab_lock);
454
455 if (--tab->refcnt == 0) {
456 list_del(&tab->list);
457 call_rcu_bh(&tab->rcu, stab_kfree_rcu);
458 }
459
460 spin_unlock(&qdisc_stab_lock);
461 }
462 EXPORT_SYMBOL(qdisc_put_stab);
463
464 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
465 {
466 struct nlattr *nest;
467
468 nest = nla_nest_start(skb, TCA_STAB);
469 if (nest == NULL)
470 goto nla_put_failure;
471 if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
472 goto nla_put_failure;
473 nla_nest_end(skb, nest);
474
475 return skb->len;
476
477 nla_put_failure:
478 return -1;
479 }
480
481 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
482 {
483 int pkt_len, slot;
484
485 pkt_len = skb->len + stab->szopts.overhead;
486 if (unlikely(!stab->szopts.tsize))
487 goto out;
488
489 slot = pkt_len + stab->szopts.cell_align;
490 if (unlikely(slot < 0))
491 slot = 0;
492
493 slot >>= stab->szopts.cell_log;
494 if (likely(slot < stab->szopts.tsize))
495 pkt_len = stab->data[slot];
496 else
497 pkt_len = stab->data[stab->szopts.tsize - 1] *
498 (slot / stab->szopts.tsize) +
499 stab->data[slot % stab->szopts.tsize];
500
501 pkt_len <<= stab->szopts.size_log;
502 out:
503 if (unlikely(pkt_len < 1))
504 pkt_len = 1;
505 qdisc_skb_cb(skb)->pkt_len = pkt_len;
506 }
507 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
508
509 void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
510 {
511 if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
512 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
513 txt, qdisc->ops->id, qdisc->handle >> 16);
514 qdisc->flags |= TCQ_F_WARN_NONWC;
515 }
516 }
517 EXPORT_SYMBOL(qdisc_warn_nonwc);
518
519 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
520 {
521 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
522 timer);
523
524 qdisc_unthrottled(wd->qdisc);
525 __netif_schedule(qdisc_root(wd->qdisc));
526
527 return HRTIMER_NORESTART;
528 }
529
530 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
531 {
532 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
533 wd->timer.function = qdisc_watchdog;
534 wd->qdisc = qdisc;
535 }
536 EXPORT_SYMBOL(qdisc_watchdog_init);
537
538 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
539 {
540 if (test_bit(__QDISC_STATE_DEACTIVATED,
541 &qdisc_root_sleeping(wd->qdisc)->state))
542 return;
543
544 qdisc_throttled(wd->qdisc);
545
546 hrtimer_start(&wd->timer,
547 ns_to_ktime(expires),
548 HRTIMER_MODE_ABS);
549 }
550 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
551
552 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
553 {
554 hrtimer_cancel(&wd->timer);
555 qdisc_unthrottled(wd->qdisc);
556 }
557 EXPORT_SYMBOL(qdisc_watchdog_cancel);
558
559 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
560 {
561 unsigned int size = n * sizeof(struct hlist_head), i;
562 struct hlist_head *h;
563
564 if (size <= PAGE_SIZE)
565 h = kmalloc(size, GFP_KERNEL);
566 else
567 h = (struct hlist_head *)
568 __get_free_pages(GFP_KERNEL, get_order(size));
569
570 if (h != NULL) {
571 for (i = 0; i < n; i++)
572 INIT_HLIST_HEAD(&h[i]);
573 }
574 return h;
575 }
576
577 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
578 {
579 unsigned int size = n * sizeof(struct hlist_head);
580
581 if (size <= PAGE_SIZE)
582 kfree(h);
583 else
584 free_pages((unsigned long)h, get_order(size));
585 }
586
587 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
588 {
589 struct Qdisc_class_common *cl;
590 struct hlist_node *next;
591 struct hlist_head *nhash, *ohash;
592 unsigned int nsize, nmask, osize;
593 unsigned int i, h;
594
595 /* Rehash when load factor exceeds 0.75 */
596 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
597 return;
598 nsize = clhash->hashsize * 2;
599 nmask = nsize - 1;
600 nhash = qdisc_class_hash_alloc(nsize);
601 if (nhash == NULL)
602 return;
603
604 ohash = clhash->hash;
605 osize = clhash->hashsize;
606
607 sch_tree_lock(sch);
608 for (i = 0; i < osize; i++) {
609 hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
610 h = qdisc_class_hash(cl->classid, nmask);
611 hlist_add_head(&cl->hnode, &nhash[h]);
612 }
613 }
614 clhash->hash = nhash;
615 clhash->hashsize = nsize;
616 clhash->hashmask = nmask;
617 sch_tree_unlock(sch);
618
619 qdisc_class_hash_free(ohash, osize);
620 }
621 EXPORT_SYMBOL(qdisc_class_hash_grow);
622
623 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
624 {
625 unsigned int size = 4;
626
627 clhash->hash = qdisc_class_hash_alloc(size);
628 if (clhash->hash == NULL)
629 return -ENOMEM;
630 clhash->hashsize = size;
631 clhash->hashmask = size - 1;
632 clhash->hashelems = 0;
633 return 0;
634 }
635 EXPORT_SYMBOL(qdisc_class_hash_init);
636
637 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
638 {
639 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
640 }
641 EXPORT_SYMBOL(qdisc_class_hash_destroy);
642
643 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
644 struct Qdisc_class_common *cl)
645 {
646 unsigned int h;
647
648 INIT_HLIST_NODE(&cl->hnode);
649 h = qdisc_class_hash(cl->classid, clhash->hashmask);
650 hlist_add_head(&cl->hnode, &clhash->hash[h]);
651 clhash->hashelems++;
652 }
653 EXPORT_SYMBOL(qdisc_class_hash_insert);
654
655 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
656 struct Qdisc_class_common *cl)
657 {
658 hlist_del(&cl->hnode);
659 clhash->hashelems--;
660 }
661 EXPORT_SYMBOL(qdisc_class_hash_remove);
662
663 /* Allocate an unique handle from space managed by kernel
664 * Possible range is [8000-FFFF]:0000 (0x8000 values)
665 */
666 static u32 qdisc_alloc_handle(struct net_device *dev)
667 {
668 int i = 0x8000;
669 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
670
671 do {
672 autohandle += TC_H_MAKE(0x10000U, 0);
673 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
674 autohandle = TC_H_MAKE(0x80000000U, 0);
675 if (!qdisc_lookup(dev, autohandle))
676 return autohandle;
677 cond_resched();
678 } while (--i > 0);
679
680 return 0;
681 }
682
683 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
684 {
685 const struct Qdisc_class_ops *cops;
686 unsigned long cl;
687 u32 parentid;
688
689 if (n == 0)
690 return;
691 while ((parentid = sch->parent)) {
692 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
693 return;
694
695 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
696 if (sch == NULL) {
697 WARN_ON(parentid != TC_H_ROOT);
698 return;
699 }
700 cops = sch->ops->cl_ops;
701 if (cops->qlen_notify) {
702 cl = cops->get(sch, parentid);
703 cops->qlen_notify(sch, cl);
704 cops->put(sch, cl);
705 }
706 sch->q.qlen -= n;
707 }
708 }
709 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
710
711 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
712 struct nlmsghdr *n, u32 clid,
713 struct Qdisc *old, struct Qdisc *new)
714 {
715 if (new || old)
716 qdisc_notify(net, skb, n, clid, old, new);
717
718 if (old)
719 qdisc_destroy(old);
720 }
721
722 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
723 * to device "dev".
724 *
725 * When appropriate send a netlink notification using 'skb'
726 * and "n".
727 *
728 * On success, destroy old qdisc.
729 */
730
731 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
732 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
733 struct Qdisc *new, struct Qdisc *old)
734 {
735 struct Qdisc *q = old;
736 struct net *net = dev_net(dev);
737 int err = 0;
738
739 if (parent == NULL) {
740 unsigned int i, num_q, ingress;
741
742 ingress = 0;
743 num_q = dev->num_tx_queues;
744 if ((q && q->flags & TCQ_F_INGRESS) ||
745 (new && new->flags & TCQ_F_INGRESS)) {
746 num_q = 1;
747 ingress = 1;
748 if (!dev_ingress_queue(dev))
749 return -ENOENT;
750 }
751
752 if (dev->flags & IFF_UP)
753 dev_deactivate(dev);
754
755 if (new && new->ops->attach)
756 goto skip;
757
758 for (i = 0; i < num_q; i++) {
759 struct netdev_queue *dev_queue = dev_ingress_queue(dev);
760
761 if (!ingress)
762 dev_queue = netdev_get_tx_queue(dev, i);
763
764 old = dev_graft_qdisc(dev_queue, new);
765 if (new && i > 0)
766 atomic_inc(&new->refcnt);
767
768 if (!ingress)
769 qdisc_destroy(old);
770 }
771
772 skip:
773 if (!ingress) {
774 notify_and_destroy(net, skb, n, classid,
775 dev->qdisc, new);
776 if (new && !new->ops->attach)
777 atomic_inc(&new->refcnt);
778 dev->qdisc = new ? : &noop_qdisc;
779
780 if (new && new->ops->attach)
781 new->ops->attach(new);
782 } else {
783 notify_and_destroy(net, skb, n, classid, old, new);
784 }
785
786 if (dev->flags & IFF_UP)
787 dev_activate(dev);
788 } else {
789 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
790
791 err = -EOPNOTSUPP;
792 if (cops && cops->graft) {
793 unsigned long cl = cops->get(parent, classid);
794 if (cl) {
795 err = cops->graft(parent, cl, new, &old);
796 cops->put(parent, cl);
797 } else
798 err = -ENOENT;
799 }
800 if (!err)
801 notify_and_destroy(net, skb, n, classid, old, new);
802 }
803 return err;
804 }
805
806 /* lockdep annotation is needed for ingress; egress gets it only for name */
807 static struct lock_class_key qdisc_tx_lock;
808 static struct lock_class_key qdisc_rx_lock;
809
810 /*
811 Allocate and initialize new qdisc.
812
813 Parameters are passed via opt.
814 */
815
816 static struct Qdisc *
817 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
818 struct Qdisc *p, u32 parent, u32 handle,
819 struct nlattr **tca, int *errp)
820 {
821 int err;
822 struct nlattr *kind = tca[TCA_KIND];
823 struct Qdisc *sch;
824 struct Qdisc_ops *ops;
825 struct qdisc_size_table *stab;
826
827 ops = qdisc_lookup_ops(kind);
828 #ifdef CONFIG_MODULES
829 if (ops == NULL && kind != NULL) {
830 char name[IFNAMSIZ];
831 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
832 /* We dropped the RTNL semaphore in order to
833 * perform the module load. So, even if we
834 * succeeded in loading the module we have to
835 * tell the caller to replay the request. We
836 * indicate this using -EAGAIN.
837 * We replay the request because the device may
838 * go away in the mean time.
839 */
840 rtnl_unlock();
841 request_module("sch_%s", name);
842 rtnl_lock();
843 ops = qdisc_lookup_ops(kind);
844 if (ops != NULL) {
845 /* We will try again qdisc_lookup_ops,
846 * so don't keep a reference.
847 */
848 module_put(ops->owner);
849 err = -EAGAIN;
850 goto err_out;
851 }
852 }
853 }
854 #endif
855
856 err = -ENOENT;
857 if (ops == NULL)
858 goto err_out;
859
860 sch = qdisc_alloc(dev_queue, ops);
861 if (IS_ERR(sch)) {
862 err = PTR_ERR(sch);
863 goto err_out2;
864 }
865
866 sch->parent = parent;
867
868 if (handle == TC_H_INGRESS) {
869 sch->flags |= TCQ_F_INGRESS;
870 handle = TC_H_MAKE(TC_H_INGRESS, 0);
871 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
872 } else {
873 if (handle == 0) {
874 handle = qdisc_alloc_handle(dev);
875 err = -ENOMEM;
876 if (handle == 0)
877 goto err_out3;
878 }
879 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
880 if (!netif_is_multiqueue(dev))
881 sch->flags |= TCQ_F_ONETXQUEUE;
882 }
883
884 sch->handle = handle;
885
886 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
887 if (tca[TCA_STAB]) {
888 stab = qdisc_get_stab(tca[TCA_STAB]);
889 if (IS_ERR(stab)) {
890 err = PTR_ERR(stab);
891 goto err_out4;
892 }
893 rcu_assign_pointer(sch->stab, stab);
894 }
895 if (tca[TCA_RATE]) {
896 spinlock_t *root_lock;
897
898 err = -EOPNOTSUPP;
899 if (sch->flags & TCQ_F_MQROOT)
900 goto err_out4;
901
902 if ((sch->parent != TC_H_ROOT) &&
903 !(sch->flags & TCQ_F_INGRESS) &&
904 (!p || !(p->flags & TCQ_F_MQROOT)))
905 root_lock = qdisc_root_sleeping_lock(sch);
906 else
907 root_lock = qdisc_lock(sch);
908
909 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
910 root_lock, tca[TCA_RATE]);
911 if (err)
912 goto err_out4;
913 }
914
915 qdisc_list_add(sch);
916
917 return sch;
918 }
919 err_out3:
920 dev_put(dev);
921 kfree((char *) sch - sch->padded);
922 err_out2:
923 module_put(ops->owner);
924 err_out:
925 *errp = err;
926 return NULL;
927
928 err_out4:
929 /*
930 * Any broken qdiscs that would require a ops->reset() here?
931 * The qdisc was never in action so it shouldn't be necessary.
932 */
933 qdisc_put_stab(rtnl_dereference(sch->stab));
934 if (ops->destroy)
935 ops->destroy(sch);
936 goto err_out3;
937 }
938
939 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
940 {
941 struct qdisc_size_table *ostab, *stab = NULL;
942 int err = 0;
943
944 if (tca[TCA_OPTIONS]) {
945 if (sch->ops->change == NULL)
946 return -EINVAL;
947 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
948 if (err)
949 return err;
950 }
951
952 if (tca[TCA_STAB]) {
953 stab = qdisc_get_stab(tca[TCA_STAB]);
954 if (IS_ERR(stab))
955 return PTR_ERR(stab);
956 }
957
958 ostab = rtnl_dereference(sch->stab);
959 rcu_assign_pointer(sch->stab, stab);
960 qdisc_put_stab(ostab);
961
962 if (tca[TCA_RATE]) {
963 /* NB: ignores errors from replace_estimator
964 because change can't be undone. */
965 if (sch->flags & TCQ_F_MQROOT)
966 goto out;
967 gen_replace_estimator(&sch->bstats, &sch->rate_est,
968 qdisc_root_sleeping_lock(sch),
969 tca[TCA_RATE]);
970 }
971 out:
972 return 0;
973 }
974
975 struct check_loop_arg {
976 struct qdisc_walker w;
977 struct Qdisc *p;
978 int depth;
979 };
980
981 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
982
983 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
984 {
985 struct check_loop_arg arg;
986
987 if (q->ops->cl_ops == NULL)
988 return 0;
989
990 arg.w.stop = arg.w.skip = arg.w.count = 0;
991 arg.w.fn = check_loop_fn;
992 arg.depth = depth;
993 arg.p = p;
994 q->ops->cl_ops->walk(q, &arg.w);
995 return arg.w.stop ? -ELOOP : 0;
996 }
997
998 static int
999 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1000 {
1001 struct Qdisc *leaf;
1002 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1003 struct check_loop_arg *arg = (struct check_loop_arg *)w;
1004
1005 leaf = cops->leaf(q, cl);
1006 if (leaf) {
1007 if (leaf == arg->p || arg->depth > 7)
1008 return -ELOOP;
1009 return check_loop(leaf, arg->p, arg->depth + 1);
1010 }
1011 return 0;
1012 }
1013
1014 /*
1015 * Delete/get qdisc.
1016 */
1017
1018 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1019 {
1020 struct net *net = sock_net(skb->sk);
1021 struct tcmsg *tcm = nlmsg_data(n);
1022 struct nlattr *tca[TCA_MAX + 1];
1023 struct net_device *dev;
1024 u32 clid;
1025 struct Qdisc *q = NULL;
1026 struct Qdisc *p = NULL;
1027 int err;
1028
1029 if ((n->nlmsg_type != RTM_GETQDISC) && !netlink_capable(skb, CAP_NET_ADMIN))
1030 return -EPERM;
1031
1032 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1033 if (err < 0)
1034 return err;
1035
1036 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1037 if (!dev)
1038 return -ENODEV;
1039
1040 clid = tcm->tcm_parent;
1041 if (clid) {
1042 if (clid != TC_H_ROOT) {
1043 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1044 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1045 if (!p)
1046 return -ENOENT;
1047 q = qdisc_leaf(p, clid);
1048 } else if (dev_ingress_queue(dev)) {
1049 q = dev_ingress_queue(dev)->qdisc_sleeping;
1050 }
1051 } else {
1052 q = dev->qdisc;
1053 }
1054 if (!q)
1055 return -ENOENT;
1056
1057 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1058 return -EINVAL;
1059 } else {
1060 q = qdisc_lookup(dev, tcm->tcm_handle);
1061 if (!q)
1062 return -ENOENT;
1063 }
1064
1065 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1066 return -EINVAL;
1067
1068 if (n->nlmsg_type == RTM_DELQDISC) {
1069 if (!clid)
1070 return -EINVAL;
1071 if (q->handle == 0)
1072 return -ENOENT;
1073 err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1074 if (err != 0)
1075 return err;
1076 } else {
1077 qdisc_notify(net, skb, n, clid, NULL, q);
1078 }
1079 return 0;
1080 }
1081
1082 /*
1083 * enable/disable flow on qdisc.
1084 */
1085 void
1086 tc_qdisc_flow_control(struct net_device *dev, u32 tcm_handle, int enable_flow)
1087 {
1088 struct Qdisc *q;
1089 struct __qdisc_change_req {
1090 struct nlattr attr;
1091 struct tc_prio_qopt data;
1092 } req = {
1093 .attr = {sizeof(struct __qdisc_change_req), TCA_OPTIONS},
1094 .data = {3, {1, 2, 2, 2, 1, 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1}, 1}
1095 };
1096
1097 /* override flow bit */
1098 req.data.enable_flow = enable_flow;
1099
1100 /* look up using tcm handle */
1101 q = qdisc_lookup(dev, tcm_handle);
1102
1103 /* call registered change function */
1104 if (q) {
1105 if (q->ops->change(q, &(req.attr)) != 0)
1106 pr_err("tc_qdisc_flow_control: qdisc change failed");
1107 }
1108 }
1109 EXPORT_SYMBOL(tc_qdisc_flow_control);
1110
1111
1112
1113 /*
1114 * Create/change qdisc.
1115 */
1116
1117 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1118 {
1119 struct net *net = sock_net(skb->sk);
1120 struct tcmsg *tcm;
1121 struct nlattr *tca[TCA_MAX + 1];
1122 struct net_device *dev;
1123 u32 clid;
1124 struct Qdisc *q, *p;
1125 int err;
1126
1127 if (!netlink_capable(skb, CAP_NET_ADMIN))
1128 return -EPERM;
1129
1130 replay:
1131 /* Reinit, just in case something touches this. */
1132 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1133 if (err < 0)
1134 return err;
1135
1136 tcm = nlmsg_data(n);
1137 clid = tcm->tcm_parent;
1138 q = p = NULL;
1139
1140 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1141 if (!dev)
1142 return -ENODEV;
1143
1144
1145 if (clid) {
1146 if (clid != TC_H_ROOT) {
1147 if (clid != TC_H_INGRESS) {
1148 p = qdisc_lookup(dev, TC_H_MAJ(clid));
1149 if (!p)
1150 return -ENOENT;
1151 q = qdisc_leaf(p, clid);
1152 } else if (dev_ingress_queue_create(dev)) {
1153 q = dev_ingress_queue(dev)->qdisc_sleeping;
1154 }
1155 } else {
1156 q = dev->qdisc;
1157 }
1158
1159 /* It may be default qdisc, ignore it */
1160 if (q && q->handle == 0)
1161 q = NULL;
1162
1163 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1164 if (tcm->tcm_handle) {
1165 if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1166 return -EEXIST;
1167 if (TC_H_MIN(tcm->tcm_handle))
1168 return -EINVAL;
1169 q = qdisc_lookup(dev, tcm->tcm_handle);
1170 if (!q)
1171 goto create_n_graft;
1172 if (n->nlmsg_flags & NLM_F_EXCL)
1173 return -EEXIST;
1174 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1175 return -EINVAL;
1176 if (q == p ||
1177 (p && check_loop(q, p, 0)))
1178 return -ELOOP;
1179 atomic_inc(&q->refcnt);
1180 goto graft;
1181 } else {
1182 if (!q)
1183 goto create_n_graft;
1184
1185 /* This magic test requires explanation.
1186 *
1187 * We know, that some child q is already
1188 * attached to this parent and have choice:
1189 * either to change it or to create/graft new one.
1190 *
1191 * 1. We are allowed to create/graft only
1192 * if CREATE and REPLACE flags are set.
1193 *
1194 * 2. If EXCL is set, requestor wanted to say,
1195 * that qdisc tcm_handle is not expected
1196 * to exist, so that we choose create/graft too.
1197 *
1198 * 3. The last case is when no flags are set.
1199 * Alas, it is sort of hole in API, we
1200 * cannot decide what to do unambiguously.
1201 * For now we select create/graft, if
1202 * user gave KIND, which does not match existing.
1203 */
1204 if ((n->nlmsg_flags & NLM_F_CREATE) &&
1205 (n->nlmsg_flags & NLM_F_REPLACE) &&
1206 ((n->nlmsg_flags & NLM_F_EXCL) ||
1207 (tca[TCA_KIND] &&
1208 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1209 goto create_n_graft;
1210 }
1211 }
1212 } else {
1213 if (!tcm->tcm_handle)
1214 return -EINVAL;
1215 q = qdisc_lookup(dev, tcm->tcm_handle);
1216 }
1217
1218 /* Change qdisc parameters */
1219 if (q == NULL)
1220 return -ENOENT;
1221 if (n->nlmsg_flags & NLM_F_EXCL)
1222 return -EEXIST;
1223 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1224 return -EINVAL;
1225 err = qdisc_change(q, tca);
1226 if (err == 0)
1227 qdisc_notify(net, skb, n, clid, NULL, q);
1228 return err;
1229
1230 create_n_graft:
1231 if (!(n->nlmsg_flags & NLM_F_CREATE))
1232 return -ENOENT;
1233 if (clid == TC_H_INGRESS) {
1234 if (dev_ingress_queue(dev))
1235 q = qdisc_create(dev, dev_ingress_queue(dev), p,
1236 tcm->tcm_parent, tcm->tcm_parent,
1237 tca, &err);
1238 else
1239 err = -ENOENT;
1240 } else {
1241 struct netdev_queue *dev_queue;
1242
1243 if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1244 dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1245 else if (p)
1246 dev_queue = p->dev_queue;
1247 else
1248 dev_queue = netdev_get_tx_queue(dev, 0);
1249
1250 q = qdisc_create(dev, dev_queue, p,
1251 tcm->tcm_parent, tcm->tcm_handle,
1252 tca, &err);
1253 }
1254 if (q == NULL) {
1255 if (err == -EAGAIN)
1256 goto replay;
1257 return err;
1258 }
1259
1260 graft:
1261 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1262 if (err) {
1263 if (q)
1264 qdisc_destroy(q);
1265 return err;
1266 }
1267
1268 return 0;
1269 }
1270
1271 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1272 u32 portid, u32 seq, u16 flags, int event)
1273 {
1274 struct tcmsg *tcm;
1275 struct nlmsghdr *nlh;
1276 unsigned char *b = skb_tail_pointer(skb);
1277 struct gnet_dump d;
1278 struct qdisc_size_table *stab;
1279
1280 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1281 if (!nlh)
1282 goto out_nlmsg_trim;
1283 tcm = nlmsg_data(nlh);
1284 tcm->tcm_family = AF_UNSPEC;
1285 tcm->tcm__pad1 = 0;
1286 tcm->tcm__pad2 = 0;
1287 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1288 tcm->tcm_parent = clid;
1289 tcm->tcm_handle = q->handle;
1290 tcm->tcm_info = atomic_read(&q->refcnt);
1291 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1292 goto nla_put_failure;
1293 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1294 goto nla_put_failure;
1295 q->qstats.qlen = q->q.qlen;
1296
1297 stab = rtnl_dereference(q->stab);
1298 if (stab && qdisc_dump_stab(skb, stab) < 0)
1299 goto nla_put_failure;
1300
1301 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1302 qdisc_root_sleeping_lock(q), &d) < 0)
1303 goto nla_put_failure;
1304
1305 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1306 goto nla_put_failure;
1307
1308 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1309 gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1310 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1311 goto nla_put_failure;
1312
1313 if (gnet_stats_finish_copy(&d) < 0)
1314 goto nla_put_failure;
1315
1316 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1317 return skb->len;
1318
1319 out_nlmsg_trim:
1320 nla_put_failure:
1321 nlmsg_trim(skb, b);
1322 return -1;
1323 }
1324
1325 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1326 {
1327 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1328 }
1329
1330 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1331 struct nlmsghdr *n, u32 clid,
1332 struct Qdisc *old, struct Qdisc *new)
1333 {
1334 struct sk_buff *skb;
1335 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1336
1337 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1338 if (!skb)
1339 return -ENOBUFS;
1340
1341 if (old && !tc_qdisc_dump_ignore(old)) {
1342 if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1343 0, RTM_DELQDISC) < 0)
1344 goto err_out;
1345 }
1346 if (new && !tc_qdisc_dump_ignore(new)) {
1347 if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1348 old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1349 goto err_out;
1350 }
1351
1352 if (skb->len)
1353 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1354 n->nlmsg_flags & NLM_F_ECHO);
1355
1356 err_out:
1357 kfree_skb(skb);
1358 return -EINVAL;
1359 }
1360
1361 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1362 struct netlink_callback *cb,
1363 int *q_idx_p, int s_q_idx)
1364 {
1365 int ret = 0, q_idx = *q_idx_p;
1366 struct Qdisc *q;
1367
1368 if (!root)
1369 return 0;
1370
1371 q = root;
1372 if (q_idx < s_q_idx) {
1373 q_idx++;
1374 } else {
1375 if (!tc_qdisc_dump_ignore(q) &&
1376 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1377 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1378 goto done;
1379 q_idx++;
1380 }
1381 list_for_each_entry(q, &root->list, list) {
1382 if (q_idx < s_q_idx) {
1383 q_idx++;
1384 continue;
1385 }
1386 if (!tc_qdisc_dump_ignore(q) &&
1387 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1388 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1389 goto done;
1390 q_idx++;
1391 }
1392
1393 out:
1394 *q_idx_p = q_idx;
1395 return ret;
1396 done:
1397 ret = -1;
1398 goto out;
1399 }
1400
1401 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1402 {
1403 struct net *net = sock_net(skb->sk);
1404 int idx, q_idx;
1405 int s_idx, s_q_idx;
1406 struct net_device *dev;
1407
1408 s_idx = cb->args[0];
1409 s_q_idx = q_idx = cb->args[1];
1410
1411 rcu_read_lock();
1412 idx = 0;
1413 for_each_netdev_rcu(net, dev) {
1414 struct netdev_queue *dev_queue;
1415
1416 if (idx < s_idx)
1417 goto cont;
1418 if (idx > s_idx)
1419 s_q_idx = 0;
1420 q_idx = 0;
1421
1422 if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1423 goto done;
1424
1425 dev_queue = dev_ingress_queue(dev);
1426 if (dev_queue &&
1427 tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1428 &q_idx, s_q_idx) < 0)
1429 goto done;
1430
1431 cont:
1432 idx++;
1433 }
1434
1435 done:
1436 rcu_read_unlock();
1437
1438 cb->args[0] = idx;
1439 cb->args[1] = q_idx;
1440
1441 return skb->len;
1442 }
1443
1444
1445
1446 /************************************************
1447 * Traffic classes manipulation. *
1448 ************************************************/
1449
1450
1451
1452 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1453 {
1454 struct net *net = sock_net(skb->sk);
1455 struct tcmsg *tcm = nlmsg_data(n);
1456 struct nlattr *tca[TCA_MAX + 1];
1457 struct net_device *dev;
1458 struct Qdisc *q = NULL;
1459 const struct Qdisc_class_ops *cops;
1460 unsigned long cl = 0;
1461 unsigned long new_cl;
1462 u32 portid;
1463 u32 clid;
1464 u32 qid;
1465 int err;
1466
1467 if ((n->nlmsg_type != RTM_GETTCLASS) && !netlink_capable(skb, CAP_NET_ADMIN))
1468 return -EPERM;
1469
1470 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1471 if (err < 0)
1472 return err;
1473
1474 dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1475 if (!dev)
1476 return -ENODEV;
1477
1478 /*
1479 parent == TC_H_UNSPEC - unspecified parent.
1480 parent == TC_H_ROOT - class is root, which has no parent.
1481 parent == X:0 - parent is root class.
1482 parent == X:Y - parent is a node in hierarchy.
1483 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1484
1485 handle == 0:0 - generate handle from kernel pool.
1486 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1487 handle == X:Y - clear.
1488 handle == X:0 - root class.
1489 */
1490
1491 /* Step 1. Determine qdisc handle X:0 */
1492
1493 portid = tcm->tcm_parent;
1494 clid = tcm->tcm_handle;
1495 qid = TC_H_MAJ(clid);
1496
1497 if (portid != TC_H_ROOT) {
1498 u32 qid1 = TC_H_MAJ(portid);
1499
1500 if (qid && qid1) {
1501 /* If both majors are known, they must be identical. */
1502 if (qid != qid1)
1503 return -EINVAL;
1504 } else if (qid1) {
1505 qid = qid1;
1506 } else if (qid == 0)
1507 qid = dev->qdisc->handle;
1508
1509 /* Now qid is genuine qdisc handle consistent
1510 * both with parent and child.
1511 *
1512 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1513 */
1514 if (portid)
1515 portid = TC_H_MAKE(qid, portid);
1516 } else {
1517 if (qid == 0)
1518 qid = dev->qdisc->handle;
1519 }
1520
1521 /* OK. Locate qdisc */
1522 q = qdisc_lookup(dev, qid);
1523 if (!q)
1524 return -ENOENT;
1525
1526 /* An check that it supports classes */
1527 cops = q->ops->cl_ops;
1528 if (cops == NULL)
1529 return -EINVAL;
1530
1531 /* Now try to get class */
1532 if (clid == 0) {
1533 if (portid == TC_H_ROOT)
1534 clid = qid;
1535 } else
1536 clid = TC_H_MAKE(qid, clid);
1537
1538 if (clid)
1539 cl = cops->get(q, clid);
1540
1541 if (cl == 0) {
1542 err = -ENOENT;
1543 if (n->nlmsg_type != RTM_NEWTCLASS ||
1544 !(n->nlmsg_flags & NLM_F_CREATE))
1545 goto out;
1546 } else {
1547 switch (n->nlmsg_type) {
1548 case RTM_NEWTCLASS:
1549 err = -EEXIST;
1550 if (n->nlmsg_flags & NLM_F_EXCL)
1551 goto out;
1552 break;
1553 case RTM_DELTCLASS:
1554 err = -EOPNOTSUPP;
1555 if (cops->delete)
1556 err = cops->delete(q, cl);
1557 if (err == 0)
1558 tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1559 goto out;
1560 case RTM_GETTCLASS:
1561 err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1562 goto out;
1563 default:
1564 err = -EINVAL;
1565 goto out;
1566 }
1567 }
1568
1569 new_cl = cl;
1570 err = -EOPNOTSUPP;
1571 if (cops->change)
1572 err = cops->change(q, clid, portid, tca, &new_cl);
1573 if (err == 0)
1574 tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1575
1576 out:
1577 if (cl)
1578 cops->put(q, cl);
1579
1580 return err;
1581 }
1582
1583
1584 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1585 unsigned long cl,
1586 u32 portid, u32 seq, u16 flags, int event)
1587 {
1588 struct tcmsg *tcm;
1589 struct nlmsghdr *nlh;
1590 unsigned char *b = skb_tail_pointer(skb);
1591 struct gnet_dump d;
1592 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1593
1594 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1595 if (!nlh)
1596 goto out_nlmsg_trim;
1597 tcm = nlmsg_data(nlh);
1598 tcm->tcm_family = AF_UNSPEC;
1599 tcm->tcm__pad1 = 0;
1600 tcm->tcm__pad2 = 0;
1601 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1602 tcm->tcm_parent = q->handle;
1603 tcm->tcm_handle = q->handle;
1604 tcm->tcm_info = 0;
1605 if (nla_put_string(skb, TCA_KIND, q->ops->id))
1606 goto nla_put_failure;
1607 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1608 goto nla_put_failure;
1609
1610 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1611 qdisc_root_sleeping_lock(q), &d) < 0)
1612 goto nla_put_failure;
1613
1614 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1615 goto nla_put_failure;
1616
1617 if (gnet_stats_finish_copy(&d) < 0)
1618 goto nla_put_failure;
1619
1620 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1621 return skb->len;
1622
1623 out_nlmsg_trim:
1624 nla_put_failure:
1625 nlmsg_trim(skb, b);
1626 return -1;
1627 }
1628
1629 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1630 struct nlmsghdr *n, struct Qdisc *q,
1631 unsigned long cl, int event)
1632 {
1633 struct sk_buff *skb;
1634 u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1635
1636 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1637 if (!skb)
1638 return -ENOBUFS;
1639
1640 if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1641 kfree_skb(skb);
1642 return -EINVAL;
1643 }
1644
1645 return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1646 n->nlmsg_flags & NLM_F_ECHO);
1647 }
1648
1649 struct qdisc_dump_args {
1650 struct qdisc_walker w;
1651 struct sk_buff *skb;
1652 struct netlink_callback *cb;
1653 };
1654
1655 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1656 {
1657 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1658
1659 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1660 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1661 }
1662
1663 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1664 struct tcmsg *tcm, struct netlink_callback *cb,
1665 int *t_p, int s_t)
1666 {
1667 struct qdisc_dump_args arg;
1668
1669 if (tc_qdisc_dump_ignore(q) ||
1670 *t_p < s_t || !q->ops->cl_ops ||
1671 (tcm->tcm_parent &&
1672 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1673 (*t_p)++;
1674 return 0;
1675 }
1676 if (*t_p > s_t)
1677 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1678 arg.w.fn = qdisc_class_dump;
1679 arg.skb = skb;
1680 arg.cb = cb;
1681 arg.w.stop = 0;
1682 arg.w.skip = cb->args[1];
1683 arg.w.count = 0;
1684 q->ops->cl_ops->walk(q, &arg.w);
1685 cb->args[1] = arg.w.count;
1686 if (arg.w.stop)
1687 return -1;
1688 (*t_p)++;
1689 return 0;
1690 }
1691
1692 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1693 struct tcmsg *tcm, struct netlink_callback *cb,
1694 int *t_p, int s_t)
1695 {
1696 struct Qdisc *q;
1697
1698 if (!root)
1699 return 0;
1700
1701 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1702 return -1;
1703
1704 list_for_each_entry(q, &root->list, list) {
1705 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1706 return -1;
1707 }
1708
1709 return 0;
1710 }
1711
1712 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1713 {
1714 struct tcmsg *tcm = nlmsg_data(cb->nlh);
1715 struct net *net = sock_net(skb->sk);
1716 struct netdev_queue *dev_queue;
1717 struct net_device *dev;
1718 int t, s_t;
1719
1720 if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1721 return 0;
1722 dev = dev_get_by_index(net, tcm->tcm_ifindex);
1723 if (!dev)
1724 return 0;
1725
1726 s_t = cb->args[0];
1727 t = 0;
1728
1729 if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1730 goto done;
1731
1732 dev_queue = dev_ingress_queue(dev);
1733 if (dev_queue &&
1734 tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1735 &t, s_t) < 0)
1736 goto done;
1737
1738 done:
1739 cb->args[0] = t;
1740
1741 dev_put(dev);
1742 return skb->len;
1743 }
1744
1745 /* Main classifier routine: scans classifier chain attached
1746 * to this qdisc, (optionally) tests for protocol and asks
1747 * specific classifiers.
1748 */
1749 int tc_classify_compat(struct sk_buff *skb, const struct tcf_proto *tp,
1750 struct tcf_result *res)
1751 {
1752 __be16 protocol = skb->protocol;
1753 int err;
1754
1755 for (; tp; tp = tp->next) {
1756 if (tp->protocol != protocol &&
1757 tp->protocol != htons(ETH_P_ALL))
1758 continue;
1759 err = tp->classify(skb, tp, res);
1760
1761 if (err >= 0) {
1762 #ifdef CONFIG_NET_CLS_ACT
1763 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1764 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1765 #endif
1766 return err;
1767 }
1768 }
1769 return -1;
1770 }
1771 EXPORT_SYMBOL(tc_classify_compat);
1772
1773 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1774 struct tcf_result *res)
1775 {
1776 int err = 0;
1777 #ifdef CONFIG_NET_CLS_ACT
1778 const struct tcf_proto *otp = tp;
1779 reclassify:
1780 #endif
1781
1782 err = tc_classify_compat(skb, tp, res);
1783 #ifdef CONFIG_NET_CLS_ACT
1784 if (err == TC_ACT_RECLASSIFY) {
1785 u32 verd = G_TC_VERD(skb->tc_verd);
1786 tp = otp;
1787
1788 if (verd++ >= MAX_REC_LOOP) {
1789 net_notice_ratelimited("%s: packet reclassify loop rule prio %u protocol %02x\n",
1790 tp->q->ops->id,
1791 tp->prio & 0xffff,
1792 ntohs(tp->protocol));
1793 return TC_ACT_SHOT;
1794 }
1795 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1796 goto reclassify;
1797 }
1798 #endif
1799 return err;
1800 }
1801 EXPORT_SYMBOL(tc_classify);
1802
1803 void tcf_destroy(struct tcf_proto *tp)
1804 {
1805 tp->ops->destroy(tp);
1806 module_put(tp->ops->owner);
1807 kfree(tp);
1808 }
1809
1810 void tcf_destroy_chain(struct tcf_proto **fl)
1811 {
1812 struct tcf_proto *tp;
1813
1814 while ((tp = *fl) != NULL) {
1815 *fl = tp->next;
1816 tcf_destroy(tp);
1817 }
1818 }
1819 EXPORT_SYMBOL(tcf_destroy_chain);
1820
1821 #ifdef CONFIG_PROC_FS
1822 static int psched_show(struct seq_file *seq, void *v)
1823 {
1824 struct timespec ts;
1825
1826 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1827 seq_printf(seq, "%08x %08x %08x %08x\n",
1828 (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1829 1000000,
1830 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1831
1832 return 0;
1833 }
1834
1835 static int psched_open(struct inode *inode, struct file *file)
1836 {
1837 return single_open(file, psched_show, NULL);
1838 }
1839
1840 static const struct file_operations psched_fops = {
1841 .owner = THIS_MODULE,
1842 .open = psched_open,
1843 .read = seq_read,
1844 .llseek = seq_lseek,
1845 .release = single_release,
1846 };
1847
1848 static int __net_init psched_net_init(struct net *net)
1849 {
1850 struct proc_dir_entry *e;
1851
1852 e = proc_create("psched", 0, net->proc_net, &psched_fops);
1853 if (e == NULL)
1854 return -ENOMEM;
1855
1856 return 0;
1857 }
1858
1859 static void __net_exit psched_net_exit(struct net *net)
1860 {
1861 remove_proc_entry("psched", net->proc_net);
1862 }
1863 #else
1864 static int __net_init psched_net_init(struct net *net)
1865 {
1866 return 0;
1867 }
1868
1869 static void __net_exit psched_net_exit(struct net *net)
1870 {
1871 }
1872 #endif
1873
1874 static struct pernet_operations psched_net_ops = {
1875 .init = psched_net_init,
1876 .exit = psched_net_exit,
1877 };
1878
1879 static int __init pktsched_init(void)
1880 {
1881 int err;
1882
1883 err = register_pernet_subsys(&psched_net_ops);
1884 if (err) {
1885 pr_err("pktsched_init: "
1886 "cannot initialize per netns operations\n");
1887 return err;
1888 }
1889
1890 register_qdisc(&pfifo_qdisc_ops);
1891 register_qdisc(&bfifo_qdisc_ops);
1892 register_qdisc(&pfifo_head_drop_qdisc_ops);
1893 register_qdisc(&mq_qdisc_ops);
1894
1895 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1896 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1897 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1898 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1899 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1900 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1901
1902 return 0;
1903 }
1904
1905 subsys_initcall(pktsched_init);