]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - net/sched/sch_api.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wirel...
[mirror_ubuntu-artful-kernel.git] / net / sched / sch_api.c
CommitLineData
1da177e4
LT		 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
1da177e4
LT		 18#include <linux/module.h>
19#include <linux/types.h>
20#include <linux/kernel.h>
1da177e4 21#include <linux/string.h>
1da177e4 22#include <linux/errno.h>
1da177e4 23#include <linux/skbuff.h>
1da177e4
LT		 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>
4179477f 29#include <linux/hrtimer.h>
25bfcd5a 30#include <linux/lockdep.h>
1da177e4 31
457c4cbc 32#include <net/net_namespace.h>
b854272b 33#include <net/sock.h>
dc5fc579 34#include <net/netlink.h>
1da177e4
LT		 35#include <net/pkt_sched.h>
36
1da177e4
LT		 37static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid,
38 struct Qdisc *old, struct Qdisc *new);
39static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
40 struct Qdisc *q, unsigned long cl, int event);
41
42/*
43
44 Short review.
45 -------------
46
47 This file consists of two interrelated parts:
48
49 1. queueing disciplines manager frontend.
50 2. traffic classes manager frontend.
51
52 Generally, queueing discipline ("qdisc") is a black box,
53 which is able to enqueue packets and to dequeue them (when
54 device is ready to send something) in order and at times
55 determined by algorithm hidden in it.
56
57 qdisc's are divided to two categories:
58 - "queues", which have no internal structure visible from outside.
59 - "schedulers", which split all the packets to "traffic classes",
60 using "packet classifiers" (look at cls_api.c)
61
62 In turn, classes may have child qdiscs (as rule, queues)
63 attached to them etc. etc. etc.
64
65 The goal of the routines in this file is to translate
66 information supplied by user in the form of handles
67 to more intelligible for kernel form, to make some sanity
68 checks and part of work, which is common to all qdiscs
69 and to provide rtnetlink notifications.
70
71 All real intelligent work is done inside qdisc modules.
72
73
74
75 Every discipline has two major routines: enqueue and dequeue.
76
77 ---dequeue
78
79 dequeue usually returns a skb to send. It is allowed to return NULL,
80 but it does not mean that queue is empty, it just means that
81 discipline does not want to send anything this time.
82 Queue is really empty if q->q.qlen == 0.
83 For complicated disciplines with multiple queues q->q is not
84 real packet queue, but however q->q.qlen must be valid.
85
86 ---enqueue
87
88 enqueue returns 0, if packet was enqueued successfully.
89 If packet (this one or another one) was dropped, it returns
90 not zero error code.
91 NET_XMIT_DROP - this packet dropped
92 Expected action: do not backoff, but wait until queue will clear.
93 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
94 Expected action: backoff or ignore
95 NET_XMIT_POLICED - dropped by police.
96 Expected action: backoff or error to real-time apps.
97
98 Auxiliary routines:
99
100 ---requeue
101
102 requeues once dequeued packet. It is used for non-standard or
e65d22e1 103 just buggy devices, which can defer output even if netif_queue_stopped()=0.
1da177e4
LT		 104
105 ---reset
106
107 returns qdisc to initial state: purge all buffers, clear all
108 timers, counters (except for statistics) etc.
109
110 ---init
111
112 initializes newly created qdisc.
113
114 ---destroy
115
116 destroys resources allocated by init and during lifetime of qdisc.
117
118 ---change
119
120 changes qdisc parameters.
121 */
122
123/* Protects list of registered TC modules. It is pure SMP lock. */
124static DEFINE_RWLOCK(qdisc_mod_lock);
125
126
127/************************************************
128 * Queueing disciplines manipulation. *
129 ************************************************/
130
131
132/* The list of all installed queueing disciplines. */
133
134static struct Qdisc_ops *qdisc_base;
135
136/* Register/uregister queueing discipline */
137
138int register_qdisc(struct Qdisc_ops *qops)
139{
140 struct Qdisc_ops *q, **qp;
141 int rc = -EEXIST;
142
143 write_lock(&qdisc_mod_lock);
144 for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
145 if (!strcmp(qops->id, q->id))
146 goto out;
147
148 if (qops->enqueue == NULL)
149 qops->enqueue = noop_qdisc_ops.enqueue;
150 if (qops->requeue == NULL)
151 qops->requeue = noop_qdisc_ops.requeue;
152 if (qops->dequeue == NULL)
153 qops->dequeue = noop_qdisc_ops.dequeue;
154
155 qops->next = NULL;
156 *qp = qops;
157 rc = 0;
158out:
159 write_unlock(&qdisc_mod_lock);
160 return rc;
161}
62e3ba1b 162EXPORT_SYMBOL(register_qdisc);
1da177e4
LT		 163
164int unregister_qdisc(struct Qdisc_ops *qops)
165{
166 struct Qdisc_ops *q, **qp;
167 int err = -ENOENT;
168
169 write_lock(&qdisc_mod_lock);
170 for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
171 if (q == qops)
172 break;
173 if (q) {
174 *qp = q->next;
175 q->next = NULL;
176 err = 0;
177 }
178 write_unlock(&qdisc_mod_lock);
179 return err;
180}
62e3ba1b 181EXPORT_SYMBOL(unregister_qdisc);
1da177e4
LT		 182
183/* We know handle. Find qdisc among all qdisc's attached to device
184 (root qdisc, all its children, children of children etc.)
185 */
186
8123b421
DM		 187struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
188{
189 struct Qdisc *q;
190
191 if (!(root->flags & TCQ_F_BUILTIN) &&
192 root->handle == handle)
193 return root;
194
195 list_for_each_entry(q, &root->list, list) {
196 if (q->handle == handle)
197 return q;
198 }
199 return NULL;
200}
201
f6e0b239
JP		 202/*
203 * This lock is needed until some qdiscs stop calling qdisc_tree_decrease_qlen()
204 * without rtnl_lock(); currently hfsc_dequeue(), netem_dequeue(), tbf_dequeue()
205 */
206static DEFINE_SPINLOCK(qdisc_list_lock);
207
208static void qdisc_list_add(struct Qdisc *q)
209{
210 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
211 spin_lock_bh(&qdisc_list_lock);
212 list_add_tail(&q->list, &qdisc_root_sleeping(q)->list);
213 spin_unlock_bh(&qdisc_list_lock);
214 }
215}
216
217void qdisc_list_del(struct Qdisc *q)
218{
219 if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
220 spin_lock_bh(&qdisc_list_lock);
221 list_del(&q->list);
222 spin_unlock_bh(&qdisc_list_lock);
223 }
224}
225EXPORT_SYMBOL(qdisc_list_del);
226
ead81cc5 227struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
1da177e4 228{
30723673 229 unsigned int i;
f6e0b239
JP		 230 struct Qdisc *q;
231
232 spin_lock_bh(&qdisc_list_lock);
30723673
DM		 233
234 for (i = 0; i < dev->num_tx_queues; i++) {
235 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
f6e0b239 236 struct Qdisc *txq_root = txq->qdisc_sleeping;
1da177e4 237
8123b421
DM		 238 q = qdisc_match_from_root(txq_root, handle);
239 if (q)
f6e0b239 240 goto unlock;
1da177e4 241 }
f6e0b239
JP		 242
243 q = qdisc_match_from_root(dev->rx_queue.qdisc_sleeping, handle);
244
245unlock:
246 spin_unlock_bh(&qdisc_list_lock);
247
248 return q;
1da177e4
LT		 249}
250
251static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
252{
253 unsigned long cl;
254 struct Qdisc *leaf;
20fea08b 255 const struct Qdisc_class_ops *cops = p->ops->cl_ops;
1da177e4
LT		 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
1e90474c 270static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
1da177e4
LT		 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) {
1e90474c 277 if (nla_strcmp(kind, q->id) == 0) {
1da177e4
LT		 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
1e90474c 290struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
1da177e4
LT		 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
5feb5e1a
PM		 301 if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
302 nla_len(tab) != TC_RTAB_SIZE)
1da177e4
LT		 303 return NULL;
304
305 rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
306 if (rtab) {
307 rtab->rate = *r;
308 rtab->refcnt = 1;
1e90474c 309 memcpy(rtab->data, nla_data(tab), 1024);
1da177e4
LT		 310 rtab->next = qdisc_rtab_list;
311 qdisc_rtab_list = rtab;
312 }
313 return rtab;
314}
62e3ba1b 315EXPORT_SYMBOL(qdisc_get_rtab);
1da177e4
LT		 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; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
325 if (rtab == tab) {
326 *rtabp = rtab->next;
327 kfree(rtab);
328 return;
329 }
330 }
331}
62e3ba1b 332EXPORT_SYMBOL(qdisc_put_rtab);
1da177e4 333
175f9c1b
JK		 334static LIST_HEAD(qdisc_stab_list);
335static DEFINE_SPINLOCK(qdisc_stab_lock);
336
337static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
338 [TCA_STAB_BASE] = { .len = sizeof(struct tc_sizespec) },
339 [TCA_STAB_DATA] = { .type = NLA_BINARY },
340};
341
342static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
343{
344 struct nlattr *tb[TCA_STAB_MAX + 1];
345 struct qdisc_size_table *stab;
346 struct tc_sizespec *s;
347 unsigned int tsize = 0;
348 u16 *tab = NULL;
349 int err;
350
351 err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
352 if (err < 0)
353 return ERR_PTR(err);
354 if (!tb[TCA_STAB_BASE])
355 return ERR_PTR(-EINVAL);
356
357 s = nla_data(tb[TCA_STAB_BASE]);
358
359 if (s->tsize > 0) {
360 if (!tb[TCA_STAB_DATA])
361 return ERR_PTR(-EINVAL);
362 tab = nla_data(tb[TCA_STAB_DATA]);
363 tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
364 }
365
366 if (!s || tsize != s->tsize || (!tab && tsize > 0))
367 return ERR_PTR(-EINVAL);
368
f3b9605d 369 spin_lock(&qdisc_stab_lock);
175f9c1b
JK		 370
371 list_for_each_entry(stab, &qdisc_stab_list, list) {
372 if (memcmp(&stab->szopts, s, sizeof(*s)))
373 continue;
374 if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
375 continue;
376 stab->refcnt++;
f3b9605d 377 spin_unlock(&qdisc_stab_lock);
175f9c1b
JK		 378 return stab;
379 }
380
f3b9605d 381 spin_unlock(&qdisc_stab_lock);
175f9c1b
JK		 382
383 stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
384 if (!stab)
385 return ERR_PTR(-ENOMEM);
386
387 stab->refcnt = 1;
388 stab->szopts = *s;
389 if (tsize > 0)
390 memcpy(stab->data, tab, tsize * sizeof(u16));
391
f3b9605d 392 spin_lock(&qdisc_stab_lock);
175f9c1b 393 list_add_tail(&stab->list, &qdisc_stab_list);
f3b9605d 394 spin_unlock(&qdisc_stab_lock);
175f9c1b
JK		 395
396 return stab;
397}
398
399void qdisc_put_stab(struct qdisc_size_table *tab)
400{
401 if (!tab)
402 return;
403
f3b9605d 404 spin_lock(&qdisc_stab_lock);
175f9c1b
JK		 405
406 if (--tab->refcnt == 0) {
407 list_del(&tab->list);
408 kfree(tab);
409 }
410
f3b9605d 411 spin_unlock(&qdisc_stab_lock);
175f9c1b
JK		 412}
413EXPORT_SYMBOL(qdisc_put_stab);
414
415static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
416{
417 struct nlattr *nest;
418
419 nest = nla_nest_start(skb, TCA_STAB);
420 NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
421 nla_nest_end(skb, nest);
422
423 return skb->len;
424
425nla_put_failure:
426 return -1;
427}
428
429void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
430{
431 int pkt_len, slot;
432
433 pkt_len = skb->len + stab->szopts.overhead;
434 if (unlikely(!stab->szopts.tsize))
435 goto out;
436
437 slot = pkt_len + stab->szopts.cell_align;
438 if (unlikely(slot < 0))
439 slot = 0;
440
441 slot >>= stab->szopts.cell_log;
442 if (likely(slot < stab->szopts.tsize))
443 pkt_len = stab->data[slot];
444 else
445 pkt_len = stab->data[stab->szopts.tsize - 1] *
446 (slot / stab->szopts.tsize) +
447 stab->data[slot % stab->szopts.tsize];
448
449 pkt_len <<= stab->szopts.size_log;
450out:
451 if (unlikely(pkt_len < 1))
452 pkt_len = 1;
453 qdisc_skb_cb(skb)->pkt_len = pkt_len;
454}
455EXPORT_SYMBOL(qdisc_calculate_pkt_len);
456
4179477f
PM		 457static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
458{
459 struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
460 timer);
461
462 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
11274e5a 463 smp_wmb();
8608db03 464 __netif_schedule(qdisc_root(wd->qdisc));
1936502d 465
4179477f
PM		 466 return HRTIMER_NORESTART;
467}
468
469void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
470{
471 hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
472 wd->timer.function = qdisc_watchdog;
473 wd->qdisc = qdisc;
474}
475EXPORT_SYMBOL(qdisc_watchdog_init);
476
477void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
478{
479 ktime_t time;
480
2540e051
JP		 481 if (test_bit(__QDISC_STATE_DEACTIVATED,
482 &qdisc_root_sleeping(wd->qdisc)->state))
483 return;
484
4179477f
PM		 485 wd->qdisc->flags |= TCQ_F_THROTTLED;
486 time = ktime_set(0, 0);
487 time = ktime_add_ns(time, PSCHED_US2NS(expires));
488 hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
489}
490EXPORT_SYMBOL(qdisc_watchdog_schedule);
491
492void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
493{
494 hrtimer_cancel(&wd->timer);
495 wd->qdisc->flags &= ~TCQ_F_THROTTLED;
496}
497EXPORT_SYMBOL(qdisc_watchdog_cancel);
1da177e4 498
a94f779f 499static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
6fe1c7a5
PM		 500{
501 unsigned int size = n * sizeof(struct hlist_head), i;
502 struct hlist_head *h;
503
504 if (size <= PAGE_SIZE)
505 h = kmalloc(size, GFP_KERNEL);
506 else
507 h = (struct hlist_head *)
508 __get_free_pages(GFP_KERNEL, get_order(size));
509
510 if (h != NULL) {
511 for (i = 0; i < n; i++)
512 INIT_HLIST_HEAD(&h[i]);
513 }
514 return h;
515}
516
517static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
518{
519 unsigned int size = n * sizeof(struct hlist_head);
520
521 if (size <= PAGE_SIZE)
522 kfree(h);
523 else
524 free_pages((unsigned long)h, get_order(size));
525}
526
527void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
528{
529 struct Qdisc_class_common *cl;
530 struct hlist_node *n, *next;
531 struct hlist_head *nhash, *ohash;
532 unsigned int nsize, nmask, osize;
533 unsigned int i, h;
534
535 /* Rehash when load factor exceeds 0.75 */
536 if (clhash->hashelems * 4 <= clhash->hashsize * 3)
537 return;
538 nsize = clhash->hashsize * 2;
539 nmask = nsize - 1;
540 nhash = qdisc_class_hash_alloc(nsize);
541 if (nhash == NULL)
542 return;
543
544 ohash = clhash->hash;
545 osize = clhash->hashsize;
546
547 sch_tree_lock(sch);
548 for (i = 0; i < osize; i++) {
549 hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
550 h = qdisc_class_hash(cl->classid, nmask);
551 hlist_add_head(&cl->hnode, &nhash[h]);
552 }
553 }
554 clhash->hash = nhash;
555 clhash->hashsize = nsize;
556 clhash->hashmask = nmask;
557 sch_tree_unlock(sch);
558
559 qdisc_class_hash_free(ohash, osize);
560}
561EXPORT_SYMBOL(qdisc_class_hash_grow);
562
563int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
564{
565 unsigned int size = 4;
566
567 clhash->hash = qdisc_class_hash_alloc(size);
568 if (clhash->hash == NULL)
569 return -ENOMEM;
570 clhash->hashsize = size;
571 clhash->hashmask = size - 1;
572 clhash->hashelems = 0;
573 return 0;
574}
575EXPORT_SYMBOL(qdisc_class_hash_init);
576
577void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
578{
579 qdisc_class_hash_free(clhash->hash, clhash->hashsize);
580}
581EXPORT_SYMBOL(qdisc_class_hash_destroy);
582
583void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
584 struct Qdisc_class_common *cl)
585{
586 unsigned int h;
587
588 INIT_HLIST_NODE(&cl->hnode);
589 h = qdisc_class_hash(cl->classid, clhash->hashmask);
590 hlist_add_head(&cl->hnode, &clhash->hash[h]);
591 clhash->hashelems++;
592}
593EXPORT_SYMBOL(qdisc_class_hash_insert);
594
595void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
596 struct Qdisc_class_common *cl)
597{
598 hlist_del(&cl->hnode);
599 clhash->hashelems--;
600}
601EXPORT_SYMBOL(qdisc_class_hash_remove);
602
1da177e4
LT		 603/* Allocate an unique handle from space managed by kernel */
604
605static u32 qdisc_alloc_handle(struct net_device *dev)
606{
607 int i = 0x10000;
608 static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
609
610 do {
611 autohandle += TC_H_MAKE(0x10000U, 0);
612 if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
613 autohandle = TC_H_MAKE(0x80000000U, 0);
614 } while (qdisc_lookup(dev, autohandle) && --i > 0);
615
616 return i>0 ? autohandle : 0;
617}
618
99194cff 619/* Attach toplevel qdisc to device queue. */
1da177e4 620
99194cff
DM		 621static struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
622 struct Qdisc *qdisc)
1da177e4 623{
8d50b53d 624 struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
53049978 625 spinlock_t *root_lock;
53049978 626
666d9bbe 627 root_lock = qdisc_lock(oqdisc);
53049978
DM		 628 spin_lock_bh(root_lock);
629
8d50b53d
DM		 630 /* Prune old scheduler */
631 if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
632 qdisc_reset(oqdisc);
1da177e4 633
8d50b53d
DM		 634 /* ... and graft new one */
635 if (qdisc == NULL)
636 qdisc = &noop_qdisc;
637 dev_queue->qdisc_sleeping = qdisc;
f7a54c13 638 rcu_assign_pointer(dev_queue->qdisc, &noop_qdisc);
1da177e4 639
53049978 640 spin_unlock_bh(root_lock);
1da177e4 641
1da177e4
LT		 642 return oqdisc;
643}
644
43effa1e
PM		 645void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
646{
20fea08b 647 const struct Qdisc_class_ops *cops;
43effa1e
PM		 648 unsigned long cl;
649 u32 parentid;
650
651 if (n == 0)
652 return;
653 while ((parentid = sch->parent)) {
066a3b5b
JP		 654 if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
655 return;
656
5ce2d488 657 sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
ffc8fefa
PM		 658 if (sch == NULL) {
659 WARN_ON(parentid != TC_H_ROOT);
660 return;
661 }
43effa1e
PM		 662 cops = sch->ops->cl_ops;
663 if (cops->qlen_notify) {
664 cl = cops->get(sch, parentid);
665 cops->qlen_notify(sch, cl);
666 cops->put(sch, cl);
667 }
668 sch->q.qlen -= n;
669 }
670}
671EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
1da177e4 672
99194cff
DM		 673static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid,
674 struct Qdisc *old, struct Qdisc *new)
675{
676 if (new || old)
677 qdisc_notify(skb, n, clid, old, new);
1da177e4 678
4d8863a2 679 if (old)
99194cff 680 qdisc_destroy(old);
99194cff
DM		 681}
682
683/* Graft qdisc "new" to class "classid" of qdisc "parent" or
684 * to device "dev".
685 *
686 * When appropriate send a netlink notification using 'skb'
687 * and "n".
688 *
689 * On success, destroy old qdisc.
1da177e4
LT		 690 */
691
692static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
99194cff
DM		 693 struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
694 struct Qdisc *new, struct Qdisc *old)
1da177e4 695{
99194cff 696 struct Qdisc *q = old;
1da177e4 697 int err = 0;
1da177e4 698
10297b99 699 if (parent == NULL) {
99194cff
DM		 700 unsigned int i, num_q, ingress;
701
702 ingress = 0;
703 num_q = dev->num_tx_queues;
8d50b53d
DM		 704 if ((q && q->flags & TCQ_F_INGRESS) ||
705 (new && new->flags & TCQ_F_INGRESS)) {
99194cff
DM		 706 num_q = 1;
707 ingress = 1;
708 }
709
710 if (dev->flags & IFF_UP)
711 dev_deactivate(dev);
712
713 for (i = 0; i < num_q; i++) {
714 struct netdev_queue *dev_queue = &dev->rx_queue;
715
716 if (!ingress)
717 dev_queue = netdev_get_tx_queue(dev, i);
718
8d50b53d
DM		 719 old = dev_graft_qdisc(dev_queue, new);
720 if (new && i > 0)
721 atomic_inc(&new->refcnt);
722
99194cff 723 notify_and_destroy(skb, n, classid, old, new);
1da177e4 724 }
99194cff
DM		 725
726 if (dev->flags & IFF_UP)
727 dev_activate(dev);
1da177e4 728 } else {
20fea08b 729 const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1da177e4
LT		 730
731 err = -EINVAL;
732
733 if (cops) {
734 unsigned long cl = cops->get(parent, classid);
735 if (cl) {
99194cff 736 err = cops->graft(parent, cl, new, &old);
1da177e4
LT		 737 cops->put(parent, cl);
738 }
739 }
99194cff
DM		 740 if (!err)
741 notify_and_destroy(skb, n, classid, old, new);
1da177e4
LT		 742 }
743 return err;
744}
745
25bfcd5a
JP		 746/* lockdep annotation is needed for ingress; egress gets it only for name */
747static struct lock_class_key qdisc_tx_lock;
748static struct lock_class_key qdisc_rx_lock;
749
1da177e4
LT		 750/*
751 Allocate and initialize new qdisc.
752
753 Parameters are passed via opt.
754 */
755
756static struct Qdisc *
bb949fbd
DM		 757qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
758 u32 parent, u32 handle, struct nlattr **tca, int *errp)
1da177e4
LT		 759{
760 int err;
1e90474c 761 struct nlattr *kind = tca[TCA_KIND];
1da177e4
LT		 762 struct Qdisc *sch;
763 struct Qdisc_ops *ops;
175f9c1b 764 struct qdisc_size_table *stab;
1da177e4
LT		 765
766 ops = qdisc_lookup_ops(kind);
95a5afca 767#ifdef CONFIG_MODULES
1da177e4
LT		 768 if (ops == NULL && kind != NULL) {
769 char name[IFNAMSIZ];
1e90474c 770 if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
1da177e4
LT		 771 /* We dropped the RTNL semaphore in order to
772 * perform the module load. So, even if we
773 * succeeded in loading the module we have to
774 * tell the caller to replay the request. We
775 * indicate this using -EAGAIN.
776 * We replay the request because the device may
777 * go away in the mean time.
778 */
779 rtnl_unlock();
780 request_module("sch_%s", name);
781 rtnl_lock();
782 ops = qdisc_lookup_ops(kind);
783 if (ops != NULL) {
784 /* We will try again qdisc_lookup_ops,
785 * so don't keep a reference.
786 */
787 module_put(ops->owner);
788 err = -EAGAIN;
789 goto err_out;
790 }
791 }
792 }
793#endif
794
b9e2cc0f 795 err = -ENOENT;
1da177e4
LT		 796 if (ops == NULL)
797 goto err_out;
798
5ce2d488 799 sch = qdisc_alloc(dev_queue, ops);
3d54b82f
TG		 800 if (IS_ERR(sch)) {
801 err = PTR_ERR(sch);
1da177e4 802 goto err_out2;
3d54b82f 803 }
1da177e4 804
ffc8fefa
PM		 805 sch->parent = parent;
806
3d54b82f 807 if (handle == TC_H_INGRESS) {
1da177e4 808 sch->flags |= TCQ_F_INGRESS;
3d54b82f 809 handle = TC_H_MAKE(TC_H_INGRESS, 0);
25bfcd5a 810 lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
fd44de7c 811 } else {
fd44de7c
PM		 812 if (handle == 0) {
813 handle = qdisc_alloc_handle(dev);
814 err = -ENOMEM;
815 if (handle == 0)
816 goto err_out3;
817 }
25bfcd5a 818 lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
1da177e4
LT		 819 }
820
3d54b82f 821 sch->handle = handle;
1da177e4 822
1e90474c 823 if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
175f9c1b
JK		 824 if (tca[TCA_STAB]) {
825 stab = qdisc_get_stab(tca[TCA_STAB]);
826 if (IS_ERR(stab)) {
827 err = PTR_ERR(stab);
828 goto err_out3;
829 }
830 sch->stab = stab;
831 }
1e90474c 832 if (tca[TCA_RATE]) {
f6f9b93f
JP		 833 spinlock_t *root_lock;
834
835 if ((sch->parent != TC_H_ROOT) &&
836 !(sch->flags & TCQ_F_INGRESS))
837 root_lock = qdisc_root_sleeping_lock(sch);
838 else
839 root_lock = qdisc_lock(sch);
840
023e09a7 841 err = gen_new_estimator(&sch->bstats, &sch->rate_est,
f6f9b93f 842 root_lock, tca[TCA_RATE]);
023e09a7
TG		 843 if (err) {
844 /*
845 * Any broken qdiscs that would require
846 * a ops->reset() here? The qdisc was never
847 * in action so it shouldn't be necessary.
848 */
849 if (ops->destroy)
850 ops->destroy(sch);
851 goto err_out3;
852 }
853 }
f6e0b239
JP		 854
855 qdisc_list_add(sch);
1da177e4 856
1da177e4
LT		 857 return sch;
858 }
859err_out3:
175f9c1b 860 qdisc_put_stab(sch->stab);
1da177e4 861 dev_put(dev);
3d54b82f 862 kfree((char *) sch - sch->padded);
1da177e4
LT		 863err_out2:
864 module_put(ops->owner);
865err_out:
866 *errp = err;
1da177e4
LT		 867 return NULL;
868}
869
1e90474c 870static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
1da177e4 871{
175f9c1b
JK		 872 struct qdisc_size_table *stab = NULL;
873 int err = 0;
1da177e4 874
175f9c1b 875 if (tca[TCA_OPTIONS]) {
1da177e4
LT		 876 if (sch->ops->change == NULL)
877 return -EINVAL;
1e90474c 878 err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1da177e4
LT		 879 if (err)
880 return err;
881 }
175f9c1b
JK		 882
883 if (tca[TCA_STAB]) {
884 stab = qdisc_get_stab(tca[TCA_STAB]);
885 if (IS_ERR(stab))
886 return PTR_ERR(stab);
887 }
888
889 qdisc_put_stab(sch->stab);
890 sch->stab = stab;
891
1e90474c 892 if (tca[TCA_RATE])
1da177e4 893 gen_replace_estimator(&sch->bstats, &sch->rate_est,
f6f9b93f
JP		 894 qdisc_root_sleeping_lock(sch),
895 tca[TCA_RATE]);
1da177e4
LT		 896 return 0;
897}
898
899struct check_loop_arg
900{
901 struct qdisc_walker w;
902 struct Qdisc *p;
903 int depth;
904};
905
906static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
907
908static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
909{
910 struct check_loop_arg arg;
911
912 if (q->ops->cl_ops == NULL)
913 return 0;
914
915 arg.w.stop = arg.w.skip = arg.w.count = 0;
916 arg.w.fn = check_loop_fn;
917 arg.depth = depth;
918 arg.p = p;
919 q->ops->cl_ops->walk(q, &arg.w);
920 return arg.w.stop ? -ELOOP : 0;
921}
922
923static int
924check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
925{
926 struct Qdisc *leaf;
20fea08b 927 const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1da177e4
LT		 928 struct check_loop_arg *arg = (struct check_loop_arg *)w;
929
930 leaf = cops->leaf(q, cl);
931 if (leaf) {
932 if (leaf == arg->p || arg->depth > 7)
933 return -ELOOP;
934 return check_loop(leaf, arg->p, arg->depth + 1);
935 }
936 return 0;
937}
938
939/*
940 * Delete/get qdisc.
941 */
942
943static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
944{
3b1e0a65 945 struct net *net = sock_net(skb->sk);
1da177e4 946 struct tcmsg *tcm = NLMSG_DATA(n);
1e90474c 947 struct nlattr *tca[TCA_MAX + 1];
1da177e4
LT		 948 struct net_device *dev;
949 u32 clid = tcm->tcm_parent;
950 struct Qdisc *q = NULL;
951 struct Qdisc *p = NULL;
952 int err;
953
b854272b
DL		 954 if (net != &init_net)
955 return -EINVAL;
956
881d966b 957 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1da177e4
LT		 958 return -ENODEV;
959
1e90474c
PM		 960 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
961 if (err < 0)
962 return err;
963
1da177e4
LT		 964 if (clid) {
965 if (clid != TC_H_ROOT) {
966 if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
967 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
968 return -ENOENT;
969 q = qdisc_leaf(p, clid);
970 } else { /* ingress */
8123b421 971 q = dev->rx_queue.qdisc_sleeping;
10297b99 972 }
1da177e4 973 } else {
e8a0464c
DM		 974 struct netdev_queue *dev_queue;
975 dev_queue = netdev_get_tx_queue(dev, 0);
b0e1e646 976 q = dev_queue->qdisc_sleeping;
1da177e4
LT		 977 }
978 if (!q)
979 return -ENOENT;
980
981 if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
982 return -EINVAL;
983 } else {
984 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
985 return -ENOENT;
986 }
987
1e90474c 988 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1da177e4
LT		 989 return -EINVAL;
990
991 if (n->nlmsg_type == RTM_DELQDISC) {
992 if (!clid)
993 return -EINVAL;
994 if (q->handle == 0)
995 return -ENOENT;
99194cff 996 if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0)
1da177e4 997 return err;
1da177e4
LT		 998 } else {
999 qdisc_notify(skb, n, clid, NULL, q);
1000 }
1001 return 0;
1002}
1003
1004/*
1005 Create/change qdisc.
1006 */
1007
1008static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1009{
3b1e0a65 1010 struct net *net = sock_net(skb->sk);
1da177e4 1011 struct tcmsg *tcm;
1e90474c 1012 struct nlattr *tca[TCA_MAX + 1];
1da177e4
LT		 1013 struct net_device *dev;
1014 u32 clid;
1015 struct Qdisc *q, *p;
1016 int err;
1017
b854272b
DL		 1018 if (net != &init_net)
1019 return -EINVAL;
1020
1da177e4
LT		 1021replay:
1022 /* Reinit, just in case something touches this. */
1023 tcm = NLMSG_DATA(n);
1da177e4
LT		 1024 clid = tcm->tcm_parent;
1025 q = p = NULL;
1026
881d966b 1027 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1da177e4
LT		 1028 return -ENODEV;
1029
1e90474c
PM		 1030 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1031 if (err < 0)
1032 return err;
1033
1da177e4
LT		 1034 if (clid) {
1035 if (clid != TC_H_ROOT) {
1036 if (clid != TC_H_INGRESS) {
1037 if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
1038 return -ENOENT;
1039 q = qdisc_leaf(p, clid);
1040 } else { /*ingress */
8123b421 1041 q = dev->rx_queue.qdisc_sleeping;
1da177e4
LT		 1042 }
1043 } else {
e8a0464c
DM		 1044 struct netdev_queue *dev_queue;
1045 dev_queue = netdev_get_tx_queue(dev, 0);
b0e1e646 1046 q = dev_queue->qdisc_sleeping;
1da177e4
LT		 1047 }
1048
1049 /* It may be default qdisc, ignore it */
1050 if (q && q->handle == 0)
1051 q = NULL;
1052
1053 if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1054 if (tcm->tcm_handle) {
1055 if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
1056 return -EEXIST;
1057 if (TC_H_MIN(tcm->tcm_handle))
1058 return -EINVAL;
1059 if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
1060 goto create_n_graft;
1061 if (n->nlmsg_flags&NLM_F_EXCL)
1062 return -EEXIST;
1e90474c 1063 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1da177e4
LT		 1064 return -EINVAL;
1065 if (q == p ||
1066 (p && check_loop(q, p, 0)))
1067 return -ELOOP;
1068 atomic_inc(&q->refcnt);
1069 goto graft;
1070 } else {
1071 if (q == NULL)
1072 goto create_n_graft;
1073
1074 /* This magic test requires explanation.
1075 *
1076 * We know, that some child q is already
1077 * attached to this parent and have choice:
1078 * either to change it or to create/graft new one.
1079 *
1080 * 1. We are allowed to create/graft only
1081 * if CREATE and REPLACE flags are set.
1082 *
1083 * 2. If EXCL is set, requestor wanted to say,
1084 * that qdisc tcm_handle is not expected
1085 * to exist, so that we choose create/graft too.
1086 *
1087 * 3. The last case is when no flags are set.
1088 * Alas, it is sort of hole in API, we
1089 * cannot decide what to do unambiguously.
1090 * For now we select create/graft, if
1091 * user gave KIND, which does not match existing.
1092 */
1093 if ((n->nlmsg_flags&NLM_F_CREATE) &&
1094 (n->nlmsg_flags&NLM_F_REPLACE) &&
1095 ((n->nlmsg_flags&NLM_F_EXCL) ||
1e90474c
PM		 1096 (tca[TCA_KIND] &&
1097 nla_strcmp(tca[TCA_KIND], q->ops->id))))
1da177e4
LT		 1098 goto create_n_graft;
1099 }
1100 }
1101 } else {
1102 if (!tcm->tcm_handle)
1103 return -EINVAL;
1104 q = qdisc_lookup(dev, tcm->tcm_handle);
1105 }
1106
1107 /* Change qdisc parameters */
1108 if (q == NULL)
1109 return -ENOENT;
1110 if (n->nlmsg_flags&NLM_F_EXCL)
1111 return -EEXIST;
1e90474c 1112 if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1da177e4
LT		 1113 return -EINVAL;
1114 err = qdisc_change(q, tca);
1115 if (err == 0)
1116 qdisc_notify(skb, n, clid, NULL, q);
1117 return err;
1118
1119create_n_graft:
1120 if (!(n->nlmsg_flags&NLM_F_CREATE))
1121 return -ENOENT;
1122 if (clid == TC_H_INGRESS)
bb949fbd
DM		 1123 q = qdisc_create(dev, &dev->rx_queue,
1124 tcm->tcm_parent, tcm->tcm_parent,
ffc8fefa 1125 tca, &err);
10297b99 1126 else
e8a0464c 1127 q = qdisc_create(dev, netdev_get_tx_queue(dev, 0),
bb949fbd 1128 tcm->tcm_parent, tcm->tcm_handle,
ffc8fefa 1129 tca, &err);
1da177e4
LT		 1130 if (q == NULL) {
1131 if (err == -EAGAIN)
1132 goto replay;
1133 return err;
1134 }
1135
1136graft:
e5befbd9
IJ		 1137 err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1138 if (err) {
1139 if (q)
1140 qdisc_destroy(q);
1141 return err;
1da177e4 1142 }
e5befbd9 1143
1da177e4
LT		 1144 return 0;
1145}
1146
1147static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
e431b8c0 1148 u32 pid, u32 seq, u16 flags, int event)
1da177e4
LT		 1149{
1150 struct tcmsg *tcm;
1151 struct nlmsghdr *nlh;
27a884dc 1152 unsigned char *b = skb_tail_pointer(skb);
1da177e4
LT		 1153 struct gnet_dump d;
1154
e431b8c0 1155 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1da177e4
LT		 1156 tcm = NLMSG_DATA(nlh);
1157 tcm->tcm_family = AF_UNSPEC;
9ef1d4c7
PM		 1158 tcm->tcm__pad1 = 0;
1159 tcm->tcm__pad2 = 0;
5ce2d488 1160 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1da177e4
LT		 1161 tcm->tcm_parent = clid;
1162 tcm->tcm_handle = q->handle;
1163 tcm->tcm_info = atomic_read(&q->refcnt);
57e1c487 1164 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1da177e4 1165 if (q->ops->dump && q->ops->dump(q, skb) < 0)
1e90474c 1166 goto nla_put_failure;
1da177e4
LT		 1167 q->qstats.qlen = q->q.qlen;
1168
175f9c1b
JK		 1169 if (q->stab && qdisc_dump_stab(skb, q->stab) < 0)
1170 goto nla_put_failure;
1171
102396ae
JP		 1172 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1173 qdisc_root_sleeping_lock(q), &d) < 0)
1e90474c 1174 goto nla_put_failure;
1da177e4
LT		 1175
1176 if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1e90474c 1177 goto nla_put_failure;
1da177e4
LT		 1178
1179 if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1da177e4 1180 gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1da177e4 1181 gnet_stats_copy_queue(&d, &q->qstats) < 0)
1e90474c 1182 goto nla_put_failure;
10297b99 1183
1da177e4 1184 if (gnet_stats_finish_copy(&d) < 0)
1e90474c 1185 goto nla_put_failure;
10297b99 1186
27a884dc 1187 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1da177e4
LT		 1188 return skb->len;
1189
1190nlmsg_failure:
1e90474c 1191nla_put_failure:
dc5fc579 1192 nlmsg_trim(skb, b);
1da177e4
LT		 1193 return -1;
1194}
1195
1196static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1197 u32 clid, struct Qdisc *old, struct Qdisc *new)
1198{
1199 struct sk_buff *skb;
1200 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1201
1202 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1203 if (!skb)
1204 return -ENOBUFS;
1205
1206 if (old && old->handle) {
1207 if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
1208 goto err_out;
1209 }
1210 if (new) {
1211 if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1212 goto err_out;
1213 }
1214
1215 if (skb->len)
97c53cac 1216 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1da177e4
LT		 1217
1218err_out:
1219 kfree_skb(skb);
1220 return -EINVAL;
1221}
1222
30723673
DM		 1223static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1224{
1225 return (q->flags & TCQ_F_BUILTIN) ? true : false;
1226}
1227
1228static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1229 struct netlink_callback *cb,
1230 int *q_idx_p, int s_q_idx)
1231{
1232 int ret = 0, q_idx = *q_idx_p;
1233 struct Qdisc *q;
1234
1235 if (!root)
1236 return 0;
1237
1238 q = root;
1239 if (q_idx < s_q_idx) {
1240 q_idx++;
1241 } else {
1242 if (!tc_qdisc_dump_ignore(q) &&
1243 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1244 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1245 goto done;
1246 q_idx++;
1247 }
1248 list_for_each_entry(q, &root->list, list) {
1249 if (q_idx < s_q_idx) {
1250 q_idx++;
1251 continue;
1252 }
1253 if (!tc_qdisc_dump_ignore(q) &&
1254 tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1255 cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1256 goto done;
1257 q_idx++;
1258 }
1259
1260out:
1261 *q_idx_p = q_idx;
1262 return ret;
1263done:
1264 ret = -1;
1265 goto out;
1266}
1267
1da177e4
LT		 1268static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1269{
3b1e0a65 1270 struct net *net = sock_net(skb->sk);
1da177e4
LT		 1271 int idx, q_idx;
1272 int s_idx, s_q_idx;
1273 struct net_device *dev;
1da177e4 1274
b854272b
DL		 1275 if (net != &init_net)
1276 return 0;
1277
1da177e4
LT		 1278 s_idx = cb->args[0];
1279 s_q_idx = q_idx = cb->args[1];
1280 read_lock(&dev_base_lock);
7562f876 1281 idx = 0;
881d966b 1282 for_each_netdev(&init_net, dev) {
30723673
DM		 1283 struct netdev_queue *dev_queue;
1284
1da177e4 1285 if (idx < s_idx)
7562f876 1286 goto cont;
1da177e4
LT		 1287 if (idx > s_idx)
1288 s_q_idx = 0;
1da177e4 1289 q_idx = 0;
30723673
DM		 1290
1291 dev_queue = netdev_get_tx_queue(dev, 0);
827ebd64 1292 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
30723673
DM		 1293 goto done;
1294
1295 dev_queue = &dev->rx_queue;
827ebd64 1296 if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
30723673
DM		 1297 goto done;
1298
7562f876
PE		 1299cont:
1300 idx++;
1da177e4
LT		 1301 }
1302
1303done:
1304 read_unlock(&dev_base_lock);
1305
1306 cb->args[0] = idx;
1307 cb->args[1] = q_idx;
1308
1309 return skb->len;
1310}
1311
1312
1313
1314/************************************************
1315 * Traffic classes manipulation. *
1316 ************************************************/
1317
1318
1319
1320static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1321{
3b1e0a65 1322 struct net *net = sock_net(skb->sk);
b0e1e646 1323 struct netdev_queue *dev_queue;
1da177e4 1324 struct tcmsg *tcm = NLMSG_DATA(n);
1e90474c 1325 struct nlattr *tca[TCA_MAX + 1];
1da177e4
LT		 1326 struct net_device *dev;
1327 struct Qdisc *q = NULL;
20fea08b 1328 const struct Qdisc_class_ops *cops;
1da177e4
LT		 1329 unsigned long cl = 0;
1330 unsigned long new_cl;
1331 u32 pid = tcm->tcm_parent;
1332 u32 clid = tcm->tcm_handle;
1333 u32 qid = TC_H_MAJ(clid);
1334 int err;
1335
b854272b
DL		 1336 if (net != &init_net)
1337 return -EINVAL;
1338
881d966b 1339 if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1da177e4
LT		 1340 return -ENODEV;
1341
1e90474c
PM		 1342 err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1343 if (err < 0)
1344 return err;
1345
1da177e4
LT		 1346 /*
1347 parent == TC_H_UNSPEC - unspecified parent.
1348 parent == TC_H_ROOT - class is root, which has no parent.
1349 parent == X:0 - parent is root class.
1350 parent == X:Y - parent is a node in hierarchy.
1351 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1352
1353 handle == 0:0 - generate handle from kernel pool.
1354 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1355 handle == X:Y - clear.
1356 handle == X:0 - root class.
1357 */
1358
1359 /* Step 1. Determine qdisc handle X:0 */
1360
e8a0464c 1361 dev_queue = netdev_get_tx_queue(dev, 0);
1da177e4
LT		 1362 if (pid != TC_H_ROOT) {
1363 u32 qid1 = TC_H_MAJ(pid);
1364
1365 if (qid && qid1) {
1366 /* If both majors are known, they must be identical. */
1367 if (qid != qid1)
1368 return -EINVAL;
1369 } else if (qid1) {
1370 qid = qid1;
1371 } else if (qid == 0)
b0e1e646 1372 qid = dev_queue->qdisc_sleeping->handle;
1da177e4
LT		 1373
1374 /* Now qid is genuine qdisc handle consistent
1375 both with parent and child.
1376
1377 TC_H_MAJ(pid) still may be unspecified, complete it now.
1378 */
1379 if (pid)
1380 pid = TC_H_MAKE(qid, pid);
1381 } else {
1382 if (qid == 0)
b0e1e646 1383 qid = dev_queue->qdisc_sleeping->handle;
1da177e4
LT		 1384 }
1385
1386 /* OK. Locate qdisc */
10297b99 1387 if ((q = qdisc_lookup(dev, qid)) == NULL)
1da177e4
LT		 1388 return -ENOENT;
1389
1390 /* An check that it supports classes */
1391 cops = q->ops->cl_ops;
1392 if (cops == NULL)
1393 return -EINVAL;
1394
1395 /* Now try to get class */
1396 if (clid == 0) {
1397 if (pid == TC_H_ROOT)
1398 clid = qid;
1399 } else
1400 clid = TC_H_MAKE(qid, clid);
1401
1402 if (clid)
1403 cl = cops->get(q, clid);
1404
1405 if (cl == 0) {
1406 err = -ENOENT;
1407 if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1408 goto out;
1409 } else {
1410 switch (n->nlmsg_type) {
10297b99 1411 case RTM_NEWTCLASS:
1da177e4
LT		 1412 err = -EEXIST;
1413 if (n->nlmsg_flags&NLM_F_EXCL)
1414 goto out;
1415 break;
1416 case RTM_DELTCLASS:
1417 err = cops->delete(q, cl);
1418 if (err == 0)
1419 tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1420 goto out;
1421 case RTM_GETTCLASS:
1422 err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1423 goto out;
1424 default:
1425 err = -EINVAL;
1426 goto out;
1427 }
1428 }
1429
1430 new_cl = cl;
1431 err = cops->change(q, clid, pid, tca, &new_cl);
1432 if (err == 0)
1433 tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1434
1435out:
1436 if (cl)
1437 cops->put(q, cl);
1438
1439 return err;
1440}
1441
1442
1443static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1444 unsigned long cl,
e431b8c0 1445 u32 pid, u32 seq, u16 flags, int event)
1da177e4
LT		 1446{
1447 struct tcmsg *tcm;
1448 struct nlmsghdr *nlh;
27a884dc 1449 unsigned char *b = skb_tail_pointer(skb);
1da177e4 1450 struct gnet_dump d;
20fea08b 1451 const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1da177e4 1452
e431b8c0 1453 nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1da177e4
LT		 1454 tcm = NLMSG_DATA(nlh);
1455 tcm->tcm_family = AF_UNSPEC;
5ce2d488 1456 tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1da177e4
LT		 1457 tcm->tcm_parent = q->handle;
1458 tcm->tcm_handle = q->handle;
1459 tcm->tcm_info = 0;
57e1c487 1460 NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1da177e4 1461 if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1e90474c 1462 goto nla_put_failure;
1da177e4 1463
102396ae
JP		 1464 if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1465 qdisc_root_sleeping_lock(q), &d) < 0)
1e90474c 1466 goto nla_put_failure;
1da177e4
LT		 1467
1468 if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1e90474c 1469 goto nla_put_failure;
1da177e4
LT		 1470
1471 if (gnet_stats_finish_copy(&d) < 0)
1e90474c 1472 goto nla_put_failure;
1da177e4 1473
27a884dc 1474 nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1da177e4
LT		 1475 return skb->len;
1476
1477nlmsg_failure:
1e90474c 1478nla_put_failure:
dc5fc579 1479 nlmsg_trim(skb, b);
1da177e4
LT		 1480 return -1;
1481}
1482
1483static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1484 struct Qdisc *q, unsigned long cl, int event)
1485{
1486 struct sk_buff *skb;
1487 u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1488
1489 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1490 if (!skb)
1491 return -ENOBUFS;
1492
1493 if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1494 kfree_skb(skb);
1495 return -EINVAL;
1496 }
1497
97c53cac 1498 return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1da177e4
LT		 1499}
1500
1501struct qdisc_dump_args
1502{
1503 struct qdisc_walker w;
1504 struct sk_buff *skb;
1505 struct netlink_callback *cb;
1506};
1507
1508static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1509{
1510 struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1511
1512 return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1513 a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1514}
1515
30723673
DM		 1516static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1517 struct tcmsg *tcm, struct netlink_callback *cb,
1518 int *t_p, int s_t)
1519{
1520 struct qdisc_dump_args arg;
1521
1522 if (tc_qdisc_dump_ignore(q) ||
1523 *t_p < s_t || !q->ops->cl_ops ||
1524 (tcm->tcm_parent &&
1525 TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1526 (*t_p)++;
1527 return 0;
1528 }
1529 if (*t_p > s_t)
1530 memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1531 arg.w.fn = qdisc_class_dump;
1532 arg.skb = skb;
1533 arg.cb = cb;
1534 arg.w.stop = 0;
1535 arg.w.skip = cb->args[1];
1536 arg.w.count = 0;
1537 q->ops->cl_ops->walk(q, &arg.w);
1538 cb->args[1] = arg.w.count;
1539 if (arg.w.stop)
1540 return -1;
1541 (*t_p)++;
1542 return 0;
1543}
1544
1545static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1546 struct tcmsg *tcm, struct netlink_callback *cb,
1547 int *t_p, int s_t)
1548{
1549 struct Qdisc *q;
1550
1551 if (!root)
1552 return 0;
1553
1554 if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1555 return -1;
1556
1557 list_for_each_entry(q, &root->list, list) {
1558 if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1559 return -1;
1560 }
1561
1562 return 0;
1563}
1564
1da177e4
LT		 1565static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1566{
30723673 1567 struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
3b1e0a65 1568 struct net *net = sock_net(skb->sk);
30723673 1569 struct netdev_queue *dev_queue;
1da177e4 1570 struct net_device *dev;
30723673 1571 int t, s_t;
1da177e4 1572
b854272b
DL		 1573 if (net != &init_net)
1574 return 0;
1575
1da177e4
LT		 1576 if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1577 return 0;
881d966b 1578 if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1da177e4
LT		 1579 return 0;
1580
1581 s_t = cb->args[0];
1582 t = 0;
1583
30723673 1584 dev_queue = netdev_get_tx_queue(dev, 0);
8123b421 1585 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
30723673
DM		 1586 goto done;
1587
1588 dev_queue = &dev->rx_queue;
8123b421 1589 if (tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb, &t, s_t) < 0)
30723673 1590 goto done;
1da177e4 1591
30723673 1592done:
1da177e4
LT		 1593 cb->args[0] = t;
1594
1595 dev_put(dev);
1596 return skb->len;
1597}
1598
1599/* Main classifier routine: scans classifier chain attached
1600 to this qdisc, (optionally) tests for protocol and asks
1601 specific classifiers.
1602 */
73ca4918
PM		 1603int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1604 struct tcf_result *res)
1605{
1606 __be16 protocol = skb->protocol;
1607 int err = 0;
1608
1609 for (; tp; tp = tp->next) {
1610 if ((tp->protocol == protocol ||
1611 tp->protocol == htons(ETH_P_ALL)) &&
1612 (err = tp->classify(skb, tp, res)) >= 0) {
1613#ifdef CONFIG_NET_CLS_ACT
1614 if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1615 skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1616#endif
1617 return err;
1618 }
1619 }
1620 return -1;
1621}
1622EXPORT_SYMBOL(tc_classify_compat);
1623
1da177e4 1624int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
73ca4918 1625 struct tcf_result *res)
1da177e4
LT		 1626{
1627 int err = 0;
73ca4918 1628 __be16 protocol;
1da177e4
LT		 1629#ifdef CONFIG_NET_CLS_ACT
1630 struct tcf_proto *otp = tp;
1631reclassify:
1632#endif
1633 protocol = skb->protocol;
1634
73ca4918 1635 err = tc_classify_compat(skb, tp, res);
1da177e4 1636#ifdef CONFIG_NET_CLS_ACT
73ca4918
PM		 1637 if (err == TC_ACT_RECLASSIFY) {
1638 u32 verd = G_TC_VERD(skb->tc_verd);
1639 tp = otp;
1640
1641 if (verd++ >= MAX_REC_LOOP) {
1642 printk("rule prio %u protocol %02x reclassify loop, "
1643 "packet dropped\n",
1644 tp->prio&0xffff, ntohs(tp->protocol));
1645 return TC_ACT_SHOT;
1da177e4 1646 }
73ca4918
PM		 1647 skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1648 goto reclassify;
1da177e4 1649 }
73ca4918
PM		 1650#endif
1651 return err;
1da177e4 1652}
73ca4918 1653EXPORT_SYMBOL(tc_classify);
1da177e4 1654
a48b5a61
PM		 1655void tcf_destroy(struct tcf_proto *tp)
1656{
1657 tp->ops->destroy(tp);
1658 module_put(tp->ops->owner);
1659 kfree(tp);
1660}
1661
ff31ab56 1662void tcf_destroy_chain(struct tcf_proto **fl)
a48b5a61
PM		 1663{
1664 struct tcf_proto *tp;
1665
ff31ab56
PM		 1666 while ((tp = *fl) != NULL) {
1667 *fl = tp->next;
a48b5a61
PM		 1668 tcf_destroy(tp);
1669 }
1670}
1671EXPORT_SYMBOL(tcf_destroy_chain);
1672
1da177e4
LT		 1673#ifdef CONFIG_PROC_FS
1674static int psched_show(struct seq_file *seq, void *v)
1675{
3c0cfc13
PM		 1676 struct timespec ts;
1677
1678 hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1da177e4 1679 seq_printf(seq, "%08x %08x %08x %08x\n",
641b9e0e 1680 (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
514bca32 1681 1000000,
3c0cfc13 1682 (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1da177e4
LT		 1683
1684 return 0;
1685}
1686
1687static int psched_open(struct inode *inode, struct file *file)
1688{
1689 return single_open(file, psched_show, PDE(inode)->data);
1690}
1691
da7071d7 1692static const struct file_operations psched_fops = {
1da177e4
LT		 1693 .owner = THIS_MODULE,
1694 .open = psched_open,
1695 .read = seq_read,
1696 .llseek = seq_lseek,
1697 .release = single_release,
10297b99 1698};
1da177e4
LT		 1699#endif
1700
1da177e4
LT		 1701static int __init pktsched_init(void)
1702{
1da177e4
LT		 1703 register_qdisc(&pfifo_qdisc_ops);
1704 register_qdisc(&bfifo_qdisc_ops);
457c4cbc 1705 proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1da177e4 1706
be577ddc
TG		 1707 rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1708 rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1709 rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1710 rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1711 rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1712 rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1713
1da177e4
LT		 1714 return 0;
1715}
1716
1717subsys_initcall(pktsched_init);
Ubuntu Artful kernel mirror