xref: /linux/net/sched/sch_api.c (revision 7265706c8fd57722f622f336ec110cb35f83e739)
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 
31 #include <net/net_namespace.h>
32 #include <net/sock.h>
33 #include <net/netlink.h>
34 #include <net/pkt_sched.h>
35 
36 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n, u32 clid,
37 			struct Qdisc *old, struct Qdisc *new);
38 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
39 			 struct Qdisc *q, unsigned long cl, int event);
40 
41 /*
42 
43    Short review.
44    -------------
45 
46    This file consists of two interrelated parts:
47 
48    1. queueing disciplines manager frontend.
49    2. traffic classes manager frontend.
50 
51    Generally, queueing discipline ("qdisc") is a black box,
52    which is able to enqueue packets and to dequeue them (when
53    device is ready to send something) in order and at times
54    determined by algorithm hidden in it.
55 
56    qdisc's are divided to two categories:
57    - "queues", which have no internal structure visible from outside.
58    - "schedulers", which split all the packets to "traffic classes",
59      using "packet classifiers" (look at cls_api.c)
60 
61    In turn, classes may have child qdiscs (as rule, queues)
62    attached to them etc. etc. etc.
63 
64    The goal of the routines in this file is to translate
65    information supplied by user in the form of handles
66    to more intelligible for kernel form, to make some sanity
67    checks and part of work, which is common to all qdiscs
68    and to provide rtnetlink notifications.
69 
70    All real intelligent work is done inside qdisc modules.
71 
72 
73 
74    Every discipline has two major routines: enqueue and dequeue.
75 
76    ---dequeue
77 
78    dequeue usually returns a skb to send. It is allowed to return NULL,
79    but it does not mean that queue is empty, it just means that
80    discipline does not want to send anything this time.
81    Queue is really empty if q->q.qlen == 0.
82    For complicated disciplines with multiple queues q->q is not
83    real packet queue, but however q->q.qlen must be valid.
84 
85    ---enqueue
86 
87    enqueue returns 0, if packet was enqueued successfully.
88    If packet (this one or another one) was dropped, it returns
89    not zero error code.
90    NET_XMIT_DROP 	- this packet dropped
91      Expected action: do not backoff, but wait until queue will clear.
92    NET_XMIT_CN	 	- probably this packet enqueued, but another one dropped.
93      Expected action: backoff or ignore
94    NET_XMIT_POLICED	- dropped by police.
95      Expected action: backoff or error to real-time apps.
96 
97    Auxiliary routines:
98 
99    ---requeue
100 
101    requeues once dequeued packet. It is used for non-standard or
102    just buggy devices, which can defer output even if netif_queue_stopped()=0.
103 
104    ---reset
105 
106    returns qdisc to initial state: purge all buffers, clear all
107    timers, counters (except for statistics) etc.
108 
109    ---init
110 
111    initializes newly created qdisc.
112 
113    ---destroy
114 
115    destroys resources allocated by init and during lifetime of qdisc.
116 
117    ---change
118 
119    changes qdisc parameters.
120  */
121 
122 /* Protects list of registered TC modules. It is pure SMP lock. */
123 static DEFINE_RWLOCK(qdisc_mod_lock);
124 
125 
126 /************************************************
127  *	Queueing disciplines manipulation.	*
128  ************************************************/
129 
130 
131 /* The list of all installed queueing disciplines. */
132 
133 static struct Qdisc_ops *qdisc_base;
134 
135 /* Register/uregister queueing discipline */
136 
137 int register_qdisc(struct Qdisc_ops *qops)
138 {
139 	struct Qdisc_ops *q, **qp;
140 	int rc = -EEXIST;
141 
142 	write_lock(&qdisc_mod_lock);
143 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
144 		if (!strcmp(qops->id, q->id))
145 			goto out;
146 
147 	if (qops->enqueue == NULL)
148 		qops->enqueue = noop_qdisc_ops.enqueue;
149 	if (qops->requeue == NULL)
150 		qops->requeue = noop_qdisc_ops.requeue;
151 	if (qops->dequeue == NULL)
152 		qops->dequeue = noop_qdisc_ops.dequeue;
153 
154 	qops->next = NULL;
155 	*qp = qops;
156 	rc = 0;
157 out:
158 	write_unlock(&qdisc_mod_lock);
159 	return rc;
160 }
161 EXPORT_SYMBOL(register_qdisc);
162 
163 int unregister_qdisc(struct Qdisc_ops *qops)
164 {
165 	struct Qdisc_ops *q, **qp;
166 	int err = -ENOENT;
167 
168 	write_lock(&qdisc_mod_lock);
169 	for (qp = &qdisc_base; (q=*qp)!=NULL; qp = &q->next)
170 		if (q == qops)
171 			break;
172 	if (q) {
173 		*qp = q->next;
174 		q->next = NULL;
175 		err = 0;
176 	}
177 	write_unlock(&qdisc_mod_lock);
178 	return err;
179 }
180 EXPORT_SYMBOL(unregister_qdisc);
181 
182 /* We know handle. Find qdisc among all qdisc's attached to device
183    (root qdisc, all its children, children of children etc.)
184  */
185 
186 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
187 {
188 	unsigned int i;
189 
190 	for (i = 0; i < dev->num_tx_queues; i++) {
191 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
192 		struct Qdisc *q, *txq_root = txq->qdisc_sleeping;
193 
194 		if (!(txq_root->flags & TCQ_F_BUILTIN) &&
195 		    txq_root->handle == handle)
196 			return txq_root;
197 
198 		list_for_each_entry(q, &txq_root->list, list) {
199 			if (q->handle == handle)
200 				return q;
201 		}
202 	}
203 	return NULL;
204 }
205 
206 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
207 {
208 	unsigned long cl;
209 	struct Qdisc *leaf;
210 	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
211 
212 	if (cops == NULL)
213 		return NULL;
214 	cl = cops->get(p, classid);
215 
216 	if (cl == 0)
217 		return NULL;
218 	leaf = cops->leaf(p, cl);
219 	cops->put(p, cl);
220 	return leaf;
221 }
222 
223 /* Find queueing discipline by name */
224 
225 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
226 {
227 	struct Qdisc_ops *q = NULL;
228 
229 	if (kind) {
230 		read_lock(&qdisc_mod_lock);
231 		for (q = qdisc_base; q; q = q->next) {
232 			if (nla_strcmp(kind, q->id) == 0) {
233 				if (!try_module_get(q->owner))
234 					q = NULL;
235 				break;
236 			}
237 		}
238 		read_unlock(&qdisc_mod_lock);
239 	}
240 	return q;
241 }
242 
243 static struct qdisc_rate_table *qdisc_rtab_list;
244 
245 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
246 {
247 	struct qdisc_rate_table *rtab;
248 
249 	for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
250 		if (memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) == 0) {
251 			rtab->refcnt++;
252 			return rtab;
253 		}
254 	}
255 
256 	if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
257 	    nla_len(tab) != TC_RTAB_SIZE)
258 		return NULL;
259 
260 	rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
261 	if (rtab) {
262 		rtab->rate = *r;
263 		rtab->refcnt = 1;
264 		memcpy(rtab->data, nla_data(tab), 1024);
265 		rtab->next = qdisc_rtab_list;
266 		qdisc_rtab_list = rtab;
267 	}
268 	return rtab;
269 }
270 EXPORT_SYMBOL(qdisc_get_rtab);
271 
272 void qdisc_put_rtab(struct qdisc_rate_table *tab)
273 {
274 	struct qdisc_rate_table *rtab, **rtabp;
275 
276 	if (!tab || --tab->refcnt)
277 		return;
278 
279 	for (rtabp = &qdisc_rtab_list; (rtab=*rtabp) != NULL; rtabp = &rtab->next) {
280 		if (rtab == tab) {
281 			*rtabp = rtab->next;
282 			kfree(rtab);
283 			return;
284 		}
285 	}
286 }
287 EXPORT_SYMBOL(qdisc_put_rtab);
288 
289 static LIST_HEAD(qdisc_stab_list);
290 static DEFINE_SPINLOCK(qdisc_stab_lock);
291 
292 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
293 	[TCA_STAB_BASE]	= { .len = sizeof(struct tc_sizespec) },
294 	[TCA_STAB_DATA] = { .type = NLA_BINARY },
295 };
296 
297 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
298 {
299 	struct nlattr *tb[TCA_STAB_MAX + 1];
300 	struct qdisc_size_table *stab;
301 	struct tc_sizespec *s;
302 	unsigned int tsize = 0;
303 	u16 *tab = NULL;
304 	int err;
305 
306 	err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
307 	if (err < 0)
308 		return ERR_PTR(err);
309 	if (!tb[TCA_STAB_BASE])
310 		return ERR_PTR(-EINVAL);
311 
312 	s = nla_data(tb[TCA_STAB_BASE]);
313 
314 	if (s->tsize > 0) {
315 		if (!tb[TCA_STAB_DATA])
316 			return ERR_PTR(-EINVAL);
317 		tab = nla_data(tb[TCA_STAB_DATA]);
318 		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
319 	}
320 
321 	if (!s || tsize != s->tsize || (!tab && tsize > 0))
322 		return ERR_PTR(-EINVAL);
323 
324 	spin_lock(&qdisc_stab_lock);
325 
326 	list_for_each_entry(stab, &qdisc_stab_list, list) {
327 		if (memcmp(&stab->szopts, s, sizeof(*s)))
328 			continue;
329 		if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
330 			continue;
331 		stab->refcnt++;
332 		spin_unlock(&qdisc_stab_lock);
333 		return stab;
334 	}
335 
336 	spin_unlock(&qdisc_stab_lock);
337 
338 	stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
339 	if (!stab)
340 		return ERR_PTR(-ENOMEM);
341 
342 	stab->refcnt = 1;
343 	stab->szopts = *s;
344 	if (tsize > 0)
345 		memcpy(stab->data, tab, tsize * sizeof(u16));
346 
347 	spin_lock(&qdisc_stab_lock);
348 	list_add_tail(&stab->list, &qdisc_stab_list);
349 	spin_unlock(&qdisc_stab_lock);
350 
351 	return stab;
352 }
353 
354 void qdisc_put_stab(struct qdisc_size_table *tab)
355 {
356 	if (!tab)
357 		return;
358 
359 	spin_lock(&qdisc_stab_lock);
360 
361 	if (--tab->refcnt == 0) {
362 		list_del(&tab->list);
363 		kfree(tab);
364 	}
365 
366 	spin_unlock(&qdisc_stab_lock);
367 }
368 EXPORT_SYMBOL(qdisc_put_stab);
369 
370 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
371 {
372 	struct nlattr *nest;
373 
374 	nest = nla_nest_start(skb, TCA_STAB);
375 	NLA_PUT(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts);
376 	nla_nest_end(skb, nest);
377 
378 	return skb->len;
379 
380 nla_put_failure:
381 	return -1;
382 }
383 
384 void qdisc_calculate_pkt_len(struct sk_buff *skb, struct qdisc_size_table *stab)
385 {
386 	int pkt_len, slot;
387 
388 	pkt_len = skb->len + stab->szopts.overhead;
389 	if (unlikely(!stab->szopts.tsize))
390 		goto out;
391 
392 	slot = pkt_len + stab->szopts.cell_align;
393 	if (unlikely(slot < 0))
394 		slot = 0;
395 
396 	slot >>= stab->szopts.cell_log;
397 	if (likely(slot < stab->szopts.tsize))
398 		pkt_len = stab->data[slot];
399 	else
400 		pkt_len = stab->data[stab->szopts.tsize - 1] *
401 				(slot / stab->szopts.tsize) +
402 				stab->data[slot % stab->szopts.tsize];
403 
404 	pkt_len <<= stab->szopts.size_log;
405 out:
406 	if (unlikely(pkt_len < 1))
407 		pkt_len = 1;
408 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
409 }
410 EXPORT_SYMBOL(qdisc_calculate_pkt_len);
411 
412 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
413 {
414 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
415 						 timer);
416 
417 	wd->qdisc->flags &= ~TCQ_F_THROTTLED;
418 	smp_wmb();
419 	__netif_schedule(wd->qdisc);
420 
421 	return HRTIMER_NORESTART;
422 }
423 
424 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
425 {
426 	hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
427 	wd->timer.function = qdisc_watchdog;
428 	wd->qdisc = qdisc;
429 }
430 EXPORT_SYMBOL(qdisc_watchdog_init);
431 
432 void qdisc_watchdog_schedule(struct qdisc_watchdog *wd, psched_time_t expires)
433 {
434 	ktime_t time;
435 
436 	wd->qdisc->flags |= TCQ_F_THROTTLED;
437 	time = ktime_set(0, 0);
438 	time = ktime_add_ns(time, PSCHED_US2NS(expires));
439 	hrtimer_start(&wd->timer, time, HRTIMER_MODE_ABS);
440 }
441 EXPORT_SYMBOL(qdisc_watchdog_schedule);
442 
443 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
444 {
445 	hrtimer_cancel(&wd->timer);
446 	wd->qdisc->flags &= ~TCQ_F_THROTTLED;
447 }
448 EXPORT_SYMBOL(qdisc_watchdog_cancel);
449 
450 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
451 {
452 	unsigned int size = n * sizeof(struct hlist_head), i;
453 	struct hlist_head *h;
454 
455 	if (size <= PAGE_SIZE)
456 		h = kmalloc(size, GFP_KERNEL);
457 	else
458 		h = (struct hlist_head *)
459 			__get_free_pages(GFP_KERNEL, get_order(size));
460 
461 	if (h != NULL) {
462 		for (i = 0; i < n; i++)
463 			INIT_HLIST_HEAD(&h[i]);
464 	}
465 	return h;
466 }
467 
468 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
469 {
470 	unsigned int size = n * sizeof(struct hlist_head);
471 
472 	if (size <= PAGE_SIZE)
473 		kfree(h);
474 	else
475 		free_pages((unsigned long)h, get_order(size));
476 }
477 
478 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
479 {
480 	struct Qdisc_class_common *cl;
481 	struct hlist_node *n, *next;
482 	struct hlist_head *nhash, *ohash;
483 	unsigned int nsize, nmask, osize;
484 	unsigned int i, h;
485 
486 	/* Rehash when load factor exceeds 0.75 */
487 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
488 		return;
489 	nsize = clhash->hashsize * 2;
490 	nmask = nsize - 1;
491 	nhash = qdisc_class_hash_alloc(nsize);
492 	if (nhash == NULL)
493 		return;
494 
495 	ohash = clhash->hash;
496 	osize = clhash->hashsize;
497 
498 	sch_tree_lock(sch);
499 	for (i = 0; i < osize; i++) {
500 		hlist_for_each_entry_safe(cl, n, next, &ohash[i], hnode) {
501 			h = qdisc_class_hash(cl->classid, nmask);
502 			hlist_add_head(&cl->hnode, &nhash[h]);
503 		}
504 	}
505 	clhash->hash     = nhash;
506 	clhash->hashsize = nsize;
507 	clhash->hashmask = nmask;
508 	sch_tree_unlock(sch);
509 
510 	qdisc_class_hash_free(ohash, osize);
511 }
512 EXPORT_SYMBOL(qdisc_class_hash_grow);
513 
514 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
515 {
516 	unsigned int size = 4;
517 
518 	clhash->hash = qdisc_class_hash_alloc(size);
519 	if (clhash->hash == NULL)
520 		return -ENOMEM;
521 	clhash->hashsize  = size;
522 	clhash->hashmask  = size - 1;
523 	clhash->hashelems = 0;
524 	return 0;
525 }
526 EXPORT_SYMBOL(qdisc_class_hash_init);
527 
528 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
529 {
530 	qdisc_class_hash_free(clhash->hash, clhash->hashsize);
531 }
532 EXPORT_SYMBOL(qdisc_class_hash_destroy);
533 
534 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
535 			     struct Qdisc_class_common *cl)
536 {
537 	unsigned int h;
538 
539 	INIT_HLIST_NODE(&cl->hnode);
540 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
541 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
542 	clhash->hashelems++;
543 }
544 EXPORT_SYMBOL(qdisc_class_hash_insert);
545 
546 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
547 			     struct Qdisc_class_common *cl)
548 {
549 	hlist_del(&cl->hnode);
550 	clhash->hashelems--;
551 }
552 EXPORT_SYMBOL(qdisc_class_hash_remove);
553 
554 /* Allocate an unique handle from space managed by kernel */
555 
556 static u32 qdisc_alloc_handle(struct net_device *dev)
557 {
558 	int i = 0x10000;
559 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
560 
561 	do {
562 		autohandle += TC_H_MAKE(0x10000U, 0);
563 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
564 			autohandle = TC_H_MAKE(0x80000000U, 0);
565 	} while	(qdisc_lookup(dev, autohandle) && --i > 0);
566 
567 	return i>0 ? autohandle : 0;
568 }
569 
570 /* Attach toplevel qdisc to device queue. */
571 
572 static struct Qdisc *dev_graft_qdisc(struct netdev_queue *dev_queue,
573 				     struct Qdisc *qdisc)
574 {
575 	struct Qdisc *oqdisc = dev_queue->qdisc_sleeping;
576 	spinlock_t *root_lock;
577 
578 	root_lock = qdisc_root_lock(oqdisc);
579 	spin_lock_bh(root_lock);
580 
581 	/* Prune old scheduler */
582 	if (oqdisc && atomic_read(&oqdisc->refcnt) <= 1)
583 		qdisc_reset(oqdisc);
584 
585 	/* ... and graft new one */
586 	if (qdisc == NULL)
587 		qdisc = &noop_qdisc;
588 	dev_queue->qdisc_sleeping = qdisc;
589 	dev_queue->qdisc = &noop_qdisc;
590 
591 	spin_unlock_bh(root_lock);
592 
593 	return oqdisc;
594 }
595 
596 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
597 {
598 	const struct Qdisc_class_ops *cops;
599 	unsigned long cl;
600 	u32 parentid;
601 
602 	if (n == 0)
603 		return;
604 	while ((parentid = sch->parent)) {
605 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
606 			return;
607 
608 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
609 		if (sch == NULL) {
610 			WARN_ON(parentid != TC_H_ROOT);
611 			return;
612 		}
613 		cops = sch->ops->cl_ops;
614 		if (cops->qlen_notify) {
615 			cl = cops->get(sch, parentid);
616 			cops->qlen_notify(sch, cl);
617 			cops->put(sch, cl);
618 		}
619 		sch->q.qlen -= n;
620 	}
621 }
622 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
623 
624 static void notify_and_destroy(struct sk_buff *skb, struct nlmsghdr *n, u32 clid,
625 			       struct Qdisc *old, struct Qdisc *new)
626 {
627 	if (new || old)
628 		qdisc_notify(skb, n, clid, old, new);
629 
630 	if (old) {
631 		spin_lock_bh(&old->q.lock);
632 		qdisc_destroy(old);
633 		spin_unlock_bh(&old->q.lock);
634 	}
635 }
636 
637 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
638  * to device "dev".
639  *
640  * When appropriate send a netlink notification using 'skb'
641  * and "n".
642  *
643  * On success, destroy old qdisc.
644  */
645 
646 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
647 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
648 		       struct Qdisc *new, struct Qdisc *old)
649 {
650 	struct Qdisc *q = old;
651 	int err = 0;
652 
653 	if (parent == NULL) {
654 		unsigned int i, num_q, ingress;
655 
656 		ingress = 0;
657 		num_q = dev->num_tx_queues;
658 		if ((q && q->flags & TCQ_F_INGRESS) ||
659 		    (new && new->flags & TCQ_F_INGRESS)) {
660 			num_q = 1;
661 			ingress = 1;
662 		}
663 
664 		if (dev->flags & IFF_UP)
665 			dev_deactivate(dev);
666 
667 		for (i = 0; i < num_q; i++) {
668 			struct netdev_queue *dev_queue = &dev->rx_queue;
669 
670 			if (!ingress)
671 				dev_queue = netdev_get_tx_queue(dev, i);
672 
673 			old = dev_graft_qdisc(dev_queue, new);
674 			if (new && i > 0)
675 				atomic_inc(&new->refcnt);
676 
677 			notify_and_destroy(skb, n, classid, old, new);
678 		}
679 
680 		if (dev->flags & IFF_UP)
681 			dev_activate(dev);
682 	} else {
683 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
684 
685 		err = -EINVAL;
686 
687 		if (cops) {
688 			unsigned long cl = cops->get(parent, classid);
689 			if (cl) {
690 				err = cops->graft(parent, cl, new, &old);
691 				cops->put(parent, cl);
692 			}
693 		}
694 		if (!err)
695 			notify_and_destroy(skb, n, classid, old, new);
696 	}
697 	return err;
698 }
699 
700 /*
701    Allocate and initialize new qdisc.
702 
703    Parameters are passed via opt.
704  */
705 
706 static struct Qdisc *
707 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
708 	     u32 parent, u32 handle, struct nlattr **tca, int *errp)
709 {
710 	int err;
711 	struct nlattr *kind = tca[TCA_KIND];
712 	struct Qdisc *sch;
713 	struct Qdisc_ops *ops;
714 	struct qdisc_size_table *stab;
715 
716 	ops = qdisc_lookup_ops(kind);
717 #ifdef CONFIG_KMOD
718 	if (ops == NULL && kind != NULL) {
719 		char name[IFNAMSIZ];
720 		if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
721 			/* We dropped the RTNL semaphore in order to
722 			 * perform the module load.  So, even if we
723 			 * succeeded in loading the module we have to
724 			 * tell the caller to replay the request.  We
725 			 * indicate this using -EAGAIN.
726 			 * We replay the request because the device may
727 			 * go away in the mean time.
728 			 */
729 			rtnl_unlock();
730 			request_module("sch_%s", name);
731 			rtnl_lock();
732 			ops = qdisc_lookup_ops(kind);
733 			if (ops != NULL) {
734 				/* We will try again qdisc_lookup_ops,
735 				 * so don't keep a reference.
736 				 */
737 				module_put(ops->owner);
738 				err = -EAGAIN;
739 				goto err_out;
740 			}
741 		}
742 	}
743 #endif
744 
745 	err = -ENOENT;
746 	if (ops == NULL)
747 		goto err_out;
748 
749 	sch = qdisc_alloc(dev_queue, ops);
750 	if (IS_ERR(sch)) {
751 		err = PTR_ERR(sch);
752 		goto err_out2;
753 	}
754 
755 	sch->parent = parent;
756 
757 	if (handle == TC_H_INGRESS) {
758 		sch->flags |= TCQ_F_INGRESS;
759 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
760 	} else {
761 		if (handle == 0) {
762 			handle = qdisc_alloc_handle(dev);
763 			err = -ENOMEM;
764 			if (handle == 0)
765 				goto err_out3;
766 		}
767 	}
768 
769 	sch->handle = handle;
770 
771 	if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
772 		if (tca[TCA_STAB]) {
773 			stab = qdisc_get_stab(tca[TCA_STAB]);
774 			if (IS_ERR(stab)) {
775 				err = PTR_ERR(stab);
776 				goto err_out3;
777 			}
778 			sch->stab = stab;
779 		}
780 		if (tca[TCA_RATE]) {
781 			err = gen_new_estimator(&sch->bstats, &sch->rate_est,
782 						qdisc_root_lock(sch),
783 						tca[TCA_RATE]);
784 			if (err) {
785 				/*
786 				 * Any broken qdiscs that would require
787 				 * a ops->reset() here? The qdisc was never
788 				 * in action so it shouldn't be necessary.
789 				 */
790 				if (ops->destroy)
791 					ops->destroy(sch);
792 				goto err_out3;
793 			}
794 		}
795 		if ((parent != TC_H_ROOT) && !(sch->flags & TCQ_F_INGRESS))
796 			list_add_tail(&sch->list, &dev_queue->qdisc_sleeping->list);
797 
798 		return sch;
799 	}
800 err_out3:
801 	qdisc_put_stab(sch->stab);
802 	dev_put(dev);
803 	kfree((char *) sch - sch->padded);
804 err_out2:
805 	module_put(ops->owner);
806 err_out:
807 	*errp = err;
808 	return NULL;
809 }
810 
811 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
812 {
813 	struct qdisc_size_table *stab = NULL;
814 	int err = 0;
815 
816 	if (tca[TCA_OPTIONS]) {
817 		if (sch->ops->change == NULL)
818 			return -EINVAL;
819 		err = sch->ops->change(sch, tca[TCA_OPTIONS]);
820 		if (err)
821 			return err;
822 	}
823 
824 	if (tca[TCA_STAB]) {
825 		stab = qdisc_get_stab(tca[TCA_STAB]);
826 		if (IS_ERR(stab))
827 			return PTR_ERR(stab);
828 	}
829 
830 	qdisc_put_stab(sch->stab);
831 	sch->stab = stab;
832 
833 	if (tca[TCA_RATE])
834 		gen_replace_estimator(&sch->bstats, &sch->rate_est,
835 				      qdisc_root_lock(sch), tca[TCA_RATE]);
836 	return 0;
837 }
838 
839 struct check_loop_arg
840 {
841 	struct qdisc_walker 	w;
842 	struct Qdisc		*p;
843 	int			depth;
844 };
845 
846 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
847 
848 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
849 {
850 	struct check_loop_arg	arg;
851 
852 	if (q->ops->cl_ops == NULL)
853 		return 0;
854 
855 	arg.w.stop = arg.w.skip = arg.w.count = 0;
856 	arg.w.fn = check_loop_fn;
857 	arg.depth = depth;
858 	arg.p = p;
859 	q->ops->cl_ops->walk(q, &arg.w);
860 	return arg.w.stop ? -ELOOP : 0;
861 }
862 
863 static int
864 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
865 {
866 	struct Qdisc *leaf;
867 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
868 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
869 
870 	leaf = cops->leaf(q, cl);
871 	if (leaf) {
872 		if (leaf == arg->p || arg->depth > 7)
873 			return -ELOOP;
874 		return check_loop(leaf, arg->p, arg->depth + 1);
875 	}
876 	return 0;
877 }
878 
879 /*
880  * Delete/get qdisc.
881  */
882 
883 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
884 {
885 	struct net *net = sock_net(skb->sk);
886 	struct tcmsg *tcm = NLMSG_DATA(n);
887 	struct nlattr *tca[TCA_MAX + 1];
888 	struct net_device *dev;
889 	u32 clid = tcm->tcm_parent;
890 	struct Qdisc *q = NULL;
891 	struct Qdisc *p = NULL;
892 	int err;
893 
894 	if (net != &init_net)
895 		return -EINVAL;
896 
897 	if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
898 		return -ENODEV;
899 
900 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
901 	if (err < 0)
902 		return err;
903 
904 	if (clid) {
905 		if (clid != TC_H_ROOT) {
906 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
907 				if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
908 					return -ENOENT;
909 				q = qdisc_leaf(p, clid);
910 			} else { /* ingress */
911 				q = dev->rx_queue.qdisc;
912 			}
913 		} else {
914 			struct netdev_queue *dev_queue;
915 			dev_queue = netdev_get_tx_queue(dev, 0);
916 			q = dev_queue->qdisc_sleeping;
917 		}
918 		if (!q)
919 			return -ENOENT;
920 
921 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
922 			return -EINVAL;
923 	} else {
924 		if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
925 			return -ENOENT;
926 	}
927 
928 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
929 		return -EINVAL;
930 
931 	if (n->nlmsg_type == RTM_DELQDISC) {
932 		if (!clid)
933 			return -EINVAL;
934 		if (q->handle == 0)
935 			return -ENOENT;
936 		if ((err = qdisc_graft(dev, p, skb, n, clid, NULL, q)) != 0)
937 			return err;
938 	} else {
939 		qdisc_notify(skb, n, clid, NULL, q);
940 	}
941 	return 0;
942 }
943 
944 /*
945    Create/change qdisc.
946  */
947 
948 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
949 {
950 	struct net *net = sock_net(skb->sk);
951 	struct tcmsg *tcm;
952 	struct nlattr *tca[TCA_MAX + 1];
953 	struct net_device *dev;
954 	u32 clid;
955 	struct Qdisc *q, *p;
956 	int err;
957 
958 	if (net != &init_net)
959 		return -EINVAL;
960 
961 replay:
962 	/* Reinit, just in case something touches this. */
963 	tcm = NLMSG_DATA(n);
964 	clid = tcm->tcm_parent;
965 	q = p = NULL;
966 
967 	if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
968 		return -ENODEV;
969 
970 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
971 	if (err < 0)
972 		return err;
973 
974 	if (clid) {
975 		if (clid != TC_H_ROOT) {
976 			if (clid != TC_H_INGRESS) {
977 				if ((p = qdisc_lookup(dev, TC_H_MAJ(clid))) == NULL)
978 					return -ENOENT;
979 				q = qdisc_leaf(p, clid);
980 			} else { /*ingress */
981 				q = dev->rx_queue.qdisc;
982 			}
983 		} else {
984 			struct netdev_queue *dev_queue;
985 			dev_queue = netdev_get_tx_queue(dev, 0);
986 			q = dev_queue->qdisc_sleeping;
987 		}
988 
989 		/* It may be default qdisc, ignore it */
990 		if (q && q->handle == 0)
991 			q = NULL;
992 
993 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
994 			if (tcm->tcm_handle) {
995 				if (q && !(n->nlmsg_flags&NLM_F_REPLACE))
996 					return -EEXIST;
997 				if (TC_H_MIN(tcm->tcm_handle))
998 					return -EINVAL;
999 				if ((q = qdisc_lookup(dev, tcm->tcm_handle)) == NULL)
1000 					goto create_n_graft;
1001 				if (n->nlmsg_flags&NLM_F_EXCL)
1002 					return -EEXIST;
1003 				if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1004 					return -EINVAL;
1005 				if (q == p ||
1006 				    (p && check_loop(q, p, 0)))
1007 					return -ELOOP;
1008 				atomic_inc(&q->refcnt);
1009 				goto graft;
1010 			} else {
1011 				if (q == NULL)
1012 					goto create_n_graft;
1013 
1014 				/* This magic test requires explanation.
1015 				 *
1016 				 *   We know, that some child q is already
1017 				 *   attached to this parent and have choice:
1018 				 *   either to change it or to create/graft new one.
1019 				 *
1020 				 *   1. We are allowed to create/graft only
1021 				 *   if CREATE and REPLACE flags are set.
1022 				 *
1023 				 *   2. If EXCL is set, requestor wanted to say,
1024 				 *   that qdisc tcm_handle is not expected
1025 				 *   to exist, so that we choose create/graft too.
1026 				 *
1027 				 *   3. The last case is when no flags are set.
1028 				 *   Alas, it is sort of hole in API, we
1029 				 *   cannot decide what to do unambiguously.
1030 				 *   For now we select create/graft, if
1031 				 *   user gave KIND, which does not match existing.
1032 				 */
1033 				if ((n->nlmsg_flags&NLM_F_CREATE) &&
1034 				    (n->nlmsg_flags&NLM_F_REPLACE) &&
1035 				    ((n->nlmsg_flags&NLM_F_EXCL) ||
1036 				     (tca[TCA_KIND] &&
1037 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1038 					goto create_n_graft;
1039 			}
1040 		}
1041 	} else {
1042 		if (!tcm->tcm_handle)
1043 			return -EINVAL;
1044 		q = qdisc_lookup(dev, tcm->tcm_handle);
1045 	}
1046 
1047 	/* Change qdisc parameters */
1048 	if (q == NULL)
1049 		return -ENOENT;
1050 	if (n->nlmsg_flags&NLM_F_EXCL)
1051 		return -EEXIST;
1052 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1053 		return -EINVAL;
1054 	err = qdisc_change(q, tca);
1055 	if (err == 0)
1056 		qdisc_notify(skb, n, clid, NULL, q);
1057 	return err;
1058 
1059 create_n_graft:
1060 	if (!(n->nlmsg_flags&NLM_F_CREATE))
1061 		return -ENOENT;
1062 	if (clid == TC_H_INGRESS)
1063 		q = qdisc_create(dev, &dev->rx_queue,
1064 				 tcm->tcm_parent, tcm->tcm_parent,
1065 				 tca, &err);
1066 	else
1067 		q = qdisc_create(dev, netdev_get_tx_queue(dev, 0),
1068 				 tcm->tcm_parent, tcm->tcm_handle,
1069 				 tca, &err);
1070 	if (q == NULL) {
1071 		if (err == -EAGAIN)
1072 			goto replay;
1073 		return err;
1074 	}
1075 
1076 graft:
1077 	if (1) {
1078 		spinlock_t *root_lock;
1079 
1080 		err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1081 		if (err) {
1082 			if (q) {
1083 				root_lock = qdisc_root_lock(q);
1084 				spin_lock_bh(root_lock);
1085 				qdisc_destroy(q);
1086 				spin_unlock_bh(root_lock);
1087 			}
1088 			return err;
1089 		}
1090 	}
1091 	return 0;
1092 }
1093 
1094 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1095 			 u32 pid, u32 seq, u16 flags, int event)
1096 {
1097 	struct tcmsg *tcm;
1098 	struct nlmsghdr  *nlh;
1099 	unsigned char *b = skb_tail_pointer(skb);
1100 	struct gnet_dump d;
1101 
1102 	nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1103 	tcm = NLMSG_DATA(nlh);
1104 	tcm->tcm_family = AF_UNSPEC;
1105 	tcm->tcm__pad1 = 0;
1106 	tcm->tcm__pad2 = 0;
1107 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1108 	tcm->tcm_parent = clid;
1109 	tcm->tcm_handle = q->handle;
1110 	tcm->tcm_info = atomic_read(&q->refcnt);
1111 	NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1112 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
1113 		goto nla_put_failure;
1114 	q->qstats.qlen = q->q.qlen;
1115 
1116 	if (q->stab && qdisc_dump_stab(skb, q->stab) < 0)
1117 		goto nla_put_failure;
1118 
1119 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1120 					 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1121 		goto nla_put_failure;
1122 
1123 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1124 		goto nla_put_failure;
1125 
1126 	if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1127 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
1128 	    gnet_stats_copy_queue(&d, &q->qstats) < 0)
1129 		goto nla_put_failure;
1130 
1131 	if (gnet_stats_finish_copy(&d) < 0)
1132 		goto nla_put_failure;
1133 
1134 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1135 	return skb->len;
1136 
1137 nlmsg_failure:
1138 nla_put_failure:
1139 	nlmsg_trim(skb, b);
1140 	return -1;
1141 }
1142 
1143 static int qdisc_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1144 			u32 clid, struct Qdisc *old, struct Qdisc *new)
1145 {
1146 	struct sk_buff *skb;
1147 	u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1148 
1149 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1150 	if (!skb)
1151 		return -ENOBUFS;
1152 
1153 	if (old && old->handle) {
1154 		if (tc_fill_qdisc(skb, old, clid, pid, n->nlmsg_seq, 0, RTM_DELQDISC) < 0)
1155 			goto err_out;
1156 	}
1157 	if (new) {
1158 		if (tc_fill_qdisc(skb, new, clid, pid, n->nlmsg_seq, old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1159 			goto err_out;
1160 	}
1161 
1162 	if (skb->len)
1163 		return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1164 
1165 err_out:
1166 	kfree_skb(skb);
1167 	return -EINVAL;
1168 }
1169 
1170 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1171 {
1172 	return (q->flags & TCQ_F_BUILTIN) ? true : false;
1173 }
1174 
1175 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1176 			      struct netlink_callback *cb,
1177 			      int *q_idx_p, int s_q_idx)
1178 {
1179 	int ret = 0, q_idx = *q_idx_p;
1180 	struct Qdisc *q;
1181 
1182 	if (!root)
1183 		return 0;
1184 
1185 	q = root;
1186 	if (q_idx < s_q_idx) {
1187 		q_idx++;
1188 	} else {
1189 		if (!tc_qdisc_dump_ignore(q) &&
1190 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1191 				  cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1192 			goto done;
1193 		q_idx++;
1194 	}
1195 	list_for_each_entry(q, &root->list, list) {
1196 		if (q_idx < s_q_idx) {
1197 			q_idx++;
1198 			continue;
1199 		}
1200 		if (!tc_qdisc_dump_ignore(q) &&
1201 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).pid,
1202 				  cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1203 			goto done;
1204 		q_idx++;
1205 	}
1206 
1207 out:
1208 	*q_idx_p = q_idx;
1209 	return ret;
1210 done:
1211 	ret = -1;
1212 	goto out;
1213 }
1214 
1215 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1216 {
1217 	struct net *net = sock_net(skb->sk);
1218 	int idx, q_idx;
1219 	int s_idx, s_q_idx;
1220 	struct net_device *dev;
1221 
1222 	if (net != &init_net)
1223 		return 0;
1224 
1225 	s_idx = cb->args[0];
1226 	s_q_idx = q_idx = cb->args[1];
1227 	read_lock(&dev_base_lock);
1228 	idx = 0;
1229 	for_each_netdev(&init_net, dev) {
1230 		struct netdev_queue *dev_queue;
1231 
1232 		if (idx < s_idx)
1233 			goto cont;
1234 		if (idx > s_idx)
1235 			s_q_idx = 0;
1236 		q_idx = 0;
1237 
1238 		dev_queue = netdev_get_tx_queue(dev, 0);
1239 		if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1240 			goto done;
1241 
1242 		dev_queue = &dev->rx_queue;
1243 		if (tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb, &q_idx, s_q_idx) < 0)
1244 			goto done;
1245 
1246 cont:
1247 		idx++;
1248 	}
1249 
1250 done:
1251 	read_unlock(&dev_base_lock);
1252 
1253 	cb->args[0] = idx;
1254 	cb->args[1] = q_idx;
1255 
1256 	return skb->len;
1257 }
1258 
1259 
1260 
1261 /************************************************
1262  *	Traffic classes manipulation.		*
1263  ************************************************/
1264 
1265 
1266 
1267 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n, void *arg)
1268 {
1269 	struct net *net = sock_net(skb->sk);
1270 	struct netdev_queue *dev_queue;
1271 	struct tcmsg *tcm = NLMSG_DATA(n);
1272 	struct nlattr *tca[TCA_MAX + 1];
1273 	struct net_device *dev;
1274 	struct Qdisc *q = NULL;
1275 	const struct Qdisc_class_ops *cops;
1276 	unsigned long cl = 0;
1277 	unsigned long new_cl;
1278 	u32 pid = tcm->tcm_parent;
1279 	u32 clid = tcm->tcm_handle;
1280 	u32 qid = TC_H_MAJ(clid);
1281 	int err;
1282 
1283 	if (net != &init_net)
1284 		return -EINVAL;
1285 
1286 	if ((dev = __dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1287 		return -ENODEV;
1288 
1289 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1290 	if (err < 0)
1291 		return err;
1292 
1293 	/*
1294 	   parent == TC_H_UNSPEC - unspecified parent.
1295 	   parent == TC_H_ROOT   - class is root, which has no parent.
1296 	   parent == X:0	 - parent is root class.
1297 	   parent == X:Y	 - parent is a node in hierarchy.
1298 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
1299 
1300 	   handle == 0:0	 - generate handle from kernel pool.
1301 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
1302 	   handle == X:Y	 - clear.
1303 	   handle == X:0	 - root class.
1304 	 */
1305 
1306 	/* Step 1. Determine qdisc handle X:0 */
1307 
1308 	dev_queue = netdev_get_tx_queue(dev, 0);
1309 	if (pid != TC_H_ROOT) {
1310 		u32 qid1 = TC_H_MAJ(pid);
1311 
1312 		if (qid && qid1) {
1313 			/* If both majors are known, they must be identical. */
1314 			if (qid != qid1)
1315 				return -EINVAL;
1316 		} else if (qid1) {
1317 			qid = qid1;
1318 		} else if (qid == 0)
1319 			qid = dev_queue->qdisc_sleeping->handle;
1320 
1321 		/* Now qid is genuine qdisc handle consistent
1322 		   both with parent and child.
1323 
1324 		   TC_H_MAJ(pid) still may be unspecified, complete it now.
1325 		 */
1326 		if (pid)
1327 			pid = TC_H_MAKE(qid, pid);
1328 	} else {
1329 		if (qid == 0)
1330 			qid = dev_queue->qdisc_sleeping->handle;
1331 	}
1332 
1333 	/* OK. Locate qdisc */
1334 	if ((q = qdisc_lookup(dev, qid)) == NULL)
1335 		return -ENOENT;
1336 
1337 	/* An check that it supports classes */
1338 	cops = q->ops->cl_ops;
1339 	if (cops == NULL)
1340 		return -EINVAL;
1341 
1342 	/* Now try to get class */
1343 	if (clid == 0) {
1344 		if (pid == TC_H_ROOT)
1345 			clid = qid;
1346 	} else
1347 		clid = TC_H_MAKE(qid, clid);
1348 
1349 	if (clid)
1350 		cl = cops->get(q, clid);
1351 
1352 	if (cl == 0) {
1353 		err = -ENOENT;
1354 		if (n->nlmsg_type != RTM_NEWTCLASS || !(n->nlmsg_flags&NLM_F_CREATE))
1355 			goto out;
1356 	} else {
1357 		switch (n->nlmsg_type) {
1358 		case RTM_NEWTCLASS:
1359 			err = -EEXIST;
1360 			if (n->nlmsg_flags&NLM_F_EXCL)
1361 				goto out;
1362 			break;
1363 		case RTM_DELTCLASS:
1364 			err = cops->delete(q, cl);
1365 			if (err == 0)
1366 				tclass_notify(skb, n, q, cl, RTM_DELTCLASS);
1367 			goto out;
1368 		case RTM_GETTCLASS:
1369 			err = tclass_notify(skb, n, q, cl, RTM_NEWTCLASS);
1370 			goto out;
1371 		default:
1372 			err = -EINVAL;
1373 			goto out;
1374 		}
1375 	}
1376 
1377 	new_cl = cl;
1378 	err = cops->change(q, clid, pid, tca, &new_cl);
1379 	if (err == 0)
1380 		tclass_notify(skb, n, q, new_cl, RTM_NEWTCLASS);
1381 
1382 out:
1383 	if (cl)
1384 		cops->put(q, cl);
1385 
1386 	return err;
1387 }
1388 
1389 
1390 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1391 			  unsigned long cl,
1392 			  u32 pid, u32 seq, u16 flags, int event)
1393 {
1394 	struct tcmsg *tcm;
1395 	struct nlmsghdr  *nlh;
1396 	unsigned char *b = skb_tail_pointer(skb);
1397 	struct gnet_dump d;
1398 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1399 
1400 	nlh = NLMSG_NEW(skb, pid, seq, event, sizeof(*tcm), flags);
1401 	tcm = NLMSG_DATA(nlh);
1402 	tcm->tcm_family = AF_UNSPEC;
1403 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1404 	tcm->tcm_parent = q->handle;
1405 	tcm->tcm_handle = q->handle;
1406 	tcm->tcm_info = 0;
1407 	NLA_PUT_STRING(skb, TCA_KIND, q->ops->id);
1408 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1409 		goto nla_put_failure;
1410 
1411 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS,
1412 					 TCA_XSTATS, qdisc_root_lock(q), &d) < 0)
1413 		goto nla_put_failure;
1414 
1415 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1416 		goto nla_put_failure;
1417 
1418 	if (gnet_stats_finish_copy(&d) < 0)
1419 		goto nla_put_failure;
1420 
1421 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1422 	return skb->len;
1423 
1424 nlmsg_failure:
1425 nla_put_failure:
1426 	nlmsg_trim(skb, b);
1427 	return -1;
1428 }
1429 
1430 static int tclass_notify(struct sk_buff *oskb, struct nlmsghdr *n,
1431 			  struct Qdisc *q, unsigned long cl, int event)
1432 {
1433 	struct sk_buff *skb;
1434 	u32 pid = oskb ? NETLINK_CB(oskb).pid : 0;
1435 
1436 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1437 	if (!skb)
1438 		return -ENOBUFS;
1439 
1440 	if (tc_fill_tclass(skb, q, cl, pid, n->nlmsg_seq, 0, event) < 0) {
1441 		kfree_skb(skb);
1442 		return -EINVAL;
1443 	}
1444 
1445 	return rtnetlink_send(skb, &init_net, pid, RTNLGRP_TC, n->nlmsg_flags&NLM_F_ECHO);
1446 }
1447 
1448 struct qdisc_dump_args
1449 {
1450 	struct qdisc_walker w;
1451 	struct sk_buff *skb;
1452 	struct netlink_callback *cb;
1453 };
1454 
1455 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1456 {
1457 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1458 
1459 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).pid,
1460 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1461 }
1462 
1463 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1464 				struct tcmsg *tcm, struct netlink_callback *cb,
1465 				int *t_p, int s_t)
1466 {
1467 	struct qdisc_dump_args arg;
1468 
1469 	if (tc_qdisc_dump_ignore(q) ||
1470 	    *t_p < s_t || !q->ops->cl_ops ||
1471 	    (tcm->tcm_parent &&
1472 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1473 		(*t_p)++;
1474 		return 0;
1475 	}
1476 	if (*t_p > s_t)
1477 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1478 	arg.w.fn = qdisc_class_dump;
1479 	arg.skb = skb;
1480 	arg.cb = cb;
1481 	arg.w.stop  = 0;
1482 	arg.w.skip = cb->args[1];
1483 	arg.w.count = 0;
1484 	q->ops->cl_ops->walk(q, &arg.w);
1485 	cb->args[1] = arg.w.count;
1486 	if (arg.w.stop)
1487 		return -1;
1488 	(*t_p)++;
1489 	return 0;
1490 }
1491 
1492 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1493 			       struct tcmsg *tcm, struct netlink_callback *cb,
1494 			       int *t_p, int s_t)
1495 {
1496 	struct Qdisc *q;
1497 
1498 	if (!root)
1499 		return 0;
1500 
1501 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1502 		return -1;
1503 
1504 	list_for_each_entry(q, &root->list, list) {
1505 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1506 			return -1;
1507 	}
1508 
1509 	return 0;
1510 }
1511 
1512 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1513 {
1514 	struct tcmsg *tcm = (struct tcmsg*)NLMSG_DATA(cb->nlh);
1515 	struct net *net = sock_net(skb->sk);
1516 	struct netdev_queue *dev_queue;
1517 	struct net_device *dev;
1518 	int t, s_t;
1519 
1520 	if (net != &init_net)
1521 		return 0;
1522 
1523 	if (cb->nlh->nlmsg_len < NLMSG_LENGTH(sizeof(*tcm)))
1524 		return 0;
1525 	if ((dev = dev_get_by_index(&init_net, tcm->tcm_ifindex)) == NULL)
1526 		return 0;
1527 
1528 	s_t = cb->args[0];
1529 	t = 0;
1530 
1531 	dev_queue = netdev_get_tx_queue(dev, 0);
1532 	if (tc_dump_tclass_root(dev_queue->qdisc, skb, tcm, cb, &t, s_t) < 0)
1533 		goto done;
1534 
1535 	dev_queue = &dev->rx_queue;
1536 	if (tc_dump_tclass_root(dev_queue->qdisc, skb, tcm, cb, &t, s_t) < 0)
1537 		goto done;
1538 
1539 done:
1540 	cb->args[0] = t;
1541 
1542 	dev_put(dev);
1543 	return skb->len;
1544 }
1545 
1546 /* Main classifier routine: scans classifier chain attached
1547    to this qdisc, (optionally) tests for protocol and asks
1548    specific classifiers.
1549  */
1550 int tc_classify_compat(struct sk_buff *skb, struct tcf_proto *tp,
1551 		       struct tcf_result *res)
1552 {
1553 	__be16 protocol = skb->protocol;
1554 	int err = 0;
1555 
1556 	for (; tp; tp = tp->next) {
1557 		if ((tp->protocol == protocol ||
1558 		     tp->protocol == htons(ETH_P_ALL)) &&
1559 		    (err = tp->classify(skb, tp, res)) >= 0) {
1560 #ifdef CONFIG_NET_CLS_ACT
1561 			if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1562 				skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1563 #endif
1564 			return err;
1565 		}
1566 	}
1567 	return -1;
1568 }
1569 EXPORT_SYMBOL(tc_classify_compat);
1570 
1571 int tc_classify(struct sk_buff *skb, struct tcf_proto *tp,
1572 		struct tcf_result *res)
1573 {
1574 	int err = 0;
1575 	__be16 protocol;
1576 #ifdef CONFIG_NET_CLS_ACT
1577 	struct tcf_proto *otp = tp;
1578 reclassify:
1579 #endif
1580 	protocol = skb->protocol;
1581 
1582 	err = tc_classify_compat(skb, tp, res);
1583 #ifdef CONFIG_NET_CLS_ACT
1584 	if (err == TC_ACT_RECLASSIFY) {
1585 		u32 verd = G_TC_VERD(skb->tc_verd);
1586 		tp = otp;
1587 
1588 		if (verd++ >= MAX_REC_LOOP) {
1589 			printk("rule prio %u protocol %02x reclassify loop, "
1590 			       "packet dropped\n",
1591 			       tp->prio&0xffff, ntohs(tp->protocol));
1592 			return TC_ACT_SHOT;
1593 		}
1594 		skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1595 		goto reclassify;
1596 	}
1597 #endif
1598 	return err;
1599 }
1600 EXPORT_SYMBOL(tc_classify);
1601 
1602 void tcf_destroy(struct tcf_proto *tp)
1603 {
1604 	tp->ops->destroy(tp);
1605 	module_put(tp->ops->owner);
1606 	kfree(tp);
1607 }
1608 
1609 void tcf_destroy_chain(struct tcf_proto **fl)
1610 {
1611 	struct tcf_proto *tp;
1612 
1613 	while ((tp = *fl) != NULL) {
1614 		*fl = tp->next;
1615 		tcf_destroy(tp);
1616 	}
1617 }
1618 EXPORT_SYMBOL(tcf_destroy_chain);
1619 
1620 #ifdef CONFIG_PROC_FS
1621 static int psched_show(struct seq_file *seq, void *v)
1622 {
1623 	struct timespec ts;
1624 
1625 	hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1626 	seq_printf(seq, "%08x %08x %08x %08x\n",
1627 		   (u32)NSEC_PER_USEC, (u32)PSCHED_US2NS(1),
1628 		   1000000,
1629 		   (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1630 
1631 	return 0;
1632 }
1633 
1634 static int psched_open(struct inode *inode, struct file *file)
1635 {
1636 	return single_open(file, psched_show, PDE(inode)->data);
1637 }
1638 
1639 static const struct file_operations psched_fops = {
1640 	.owner = THIS_MODULE,
1641 	.open = psched_open,
1642 	.read  = seq_read,
1643 	.llseek = seq_lseek,
1644 	.release = single_release,
1645 };
1646 #endif
1647 
1648 static int __init pktsched_init(void)
1649 {
1650 	register_qdisc(&pfifo_qdisc_ops);
1651 	register_qdisc(&bfifo_qdisc_ops);
1652 	proc_net_fops_create(&init_net, "psched", 0, &psched_fops);
1653 
1654 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL);
1655 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL);
1656 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc);
1657 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL);
1658 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL);
1659 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass);
1660 
1661 	return 0;
1662 }
1663 
1664 subsys_initcall(pktsched_init);
1665