xref: /openbmc/linux/net/sched/sch_api.c (revision bc000245)
1 /*
2  * net/sched/sch_api.c	Packet scheduler API.
3  *
4  *		This program is free software; you can redistribute it and/or
5  *		modify it under the terms of the GNU General Public License
6  *		as published by the Free Software Foundation; either version
7  *		2 of the License, or (at your option) any later version.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  *
11  * Fixes:
12  *
13  * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired.
14  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
15  * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support
16  */
17 
18 #include <linux/module.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/string.h>
22 #include <linux/errno.h>
23 #include <linux/skbuff.h>
24 #include <linux/init.h>
25 #include <linux/proc_fs.h>
26 #include <linux/seq_file.h>
27 #include <linux/kmod.h>
28 #include <linux/list.h>
29 #include <linux/hrtimer.h>
30 #include <linux/lockdep.h>
31 #include <linux/slab.h>
32 
33 #include <net/net_namespace.h>
34 #include <net/sock.h>
35 #include <net/netlink.h>
36 #include <net/pkt_sched.h>
37 
38 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
39 			struct nlmsghdr *n, u32 clid,
40 			struct Qdisc *old, struct Qdisc *new);
41 static int tclass_notify(struct net *net, struct sk_buff *oskb,
42 			 struct nlmsghdr *n, struct Qdisc *q,
43 			 unsigned long cl, int event);
44 
45 /*
46 
47    Short review.
48    -------------
49 
50    This file consists of two interrelated parts:
51 
52    1. queueing disciplines manager frontend.
53    2. traffic classes manager frontend.
54 
55    Generally, queueing discipline ("qdisc") is a black box,
56    which is able to enqueue packets and to dequeue them (when
57    device is ready to send something) in order and at times
58    determined by algorithm hidden in it.
59 
60    qdisc's are divided to two categories:
61    - "queues", which have no internal structure visible from outside.
62    - "schedulers", which split all the packets to "traffic classes",
63      using "packet classifiers" (look at cls_api.c)
64 
65    In turn, classes may have child qdiscs (as rule, queues)
66    attached to them etc. etc. etc.
67 
68    The goal of the routines in this file is to translate
69    information supplied by user in the form of handles
70    to more intelligible for kernel form, to make some sanity
71    checks and part of work, which is common to all qdiscs
72    and to provide rtnetlink notifications.
73 
74    All real intelligent work is done inside qdisc modules.
75 
76 
77 
78    Every discipline has two major routines: enqueue and dequeue.
79 
80    ---dequeue
81 
82    dequeue usually returns a skb to send. It is allowed to return NULL,
83    but it does not mean that queue is empty, it just means that
84    discipline does not want to send anything this time.
85    Queue is really empty if q->q.qlen == 0.
86    For complicated disciplines with multiple queues q->q is not
87    real packet queue, but however q->q.qlen must be valid.
88 
89    ---enqueue
90 
91    enqueue returns 0, if packet was enqueued successfully.
92    If packet (this one or another one) was dropped, it returns
93    not zero error code.
94    NET_XMIT_DROP 	- this packet dropped
95      Expected action: do not backoff, but wait until queue will clear.
96    NET_XMIT_CN	 	- probably this packet enqueued, but another one dropped.
97      Expected action: backoff or ignore
98    NET_XMIT_POLICED	- dropped by police.
99      Expected action: backoff or error to real-time apps.
100 
101    Auxiliary routines:
102 
103    ---peek
104 
105    like dequeue but without removing a packet from the queue
106 
107    ---reset
108 
109    returns qdisc to initial state: purge all buffers, clear all
110    timers, counters (except for statistics) etc.
111 
112    ---init
113 
114    initializes newly created qdisc.
115 
116    ---destroy
117 
118    destroys resources allocated by init and during lifetime of qdisc.
119 
120    ---change
121 
122    changes qdisc parameters.
123  */
124 
125 /* Protects list of registered TC modules. It is pure SMP lock. */
126 static DEFINE_RWLOCK(qdisc_mod_lock);
127 
128 
129 /************************************************
130  *	Queueing disciplines manipulation.	*
131  ************************************************/
132 
133 
134 /* The list of all installed queueing disciplines. */
135 
136 static struct Qdisc_ops *qdisc_base;
137 
138 /* Register/uregister queueing discipline */
139 
140 int register_qdisc(struct Qdisc_ops *qops)
141 {
142 	struct Qdisc_ops *q, **qp;
143 	int rc = -EEXIST;
144 
145 	write_lock(&qdisc_mod_lock);
146 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
147 		if (!strcmp(qops->id, q->id))
148 			goto out;
149 
150 	if (qops->enqueue == NULL)
151 		qops->enqueue = noop_qdisc_ops.enqueue;
152 	if (qops->peek == NULL) {
153 		if (qops->dequeue == NULL)
154 			qops->peek = noop_qdisc_ops.peek;
155 		else
156 			goto out_einval;
157 	}
158 	if (qops->dequeue == NULL)
159 		qops->dequeue = noop_qdisc_ops.dequeue;
160 
161 	if (qops->cl_ops) {
162 		const struct Qdisc_class_ops *cops = qops->cl_ops;
163 
164 		if (!(cops->get && cops->put && cops->walk && cops->leaf))
165 			goto out_einval;
166 
167 		if (cops->tcf_chain && !(cops->bind_tcf && cops->unbind_tcf))
168 			goto out_einval;
169 	}
170 
171 	qops->next = NULL;
172 	*qp = qops;
173 	rc = 0;
174 out:
175 	write_unlock(&qdisc_mod_lock);
176 	return rc;
177 
178 out_einval:
179 	rc = -EINVAL;
180 	goto out;
181 }
182 EXPORT_SYMBOL(register_qdisc);
183 
184 int unregister_qdisc(struct Qdisc_ops *qops)
185 {
186 	struct Qdisc_ops *q, **qp;
187 	int err = -ENOENT;
188 
189 	write_lock(&qdisc_mod_lock);
190 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
191 		if (q == qops)
192 			break;
193 	if (q) {
194 		*qp = q->next;
195 		q->next = NULL;
196 		err = 0;
197 	}
198 	write_unlock(&qdisc_mod_lock);
199 	return err;
200 }
201 EXPORT_SYMBOL(unregister_qdisc);
202 
203 /* Get default qdisc if not otherwise specified */
204 void qdisc_get_default(char *name, size_t len)
205 {
206 	read_lock(&qdisc_mod_lock);
207 	strlcpy(name, default_qdisc_ops->id, len);
208 	read_unlock(&qdisc_mod_lock);
209 }
210 
211 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
212 {
213 	struct Qdisc_ops *q = NULL;
214 
215 	for (q = qdisc_base; q; q = q->next) {
216 		if (!strcmp(name, q->id)) {
217 			if (!try_module_get(q->owner))
218 				q = NULL;
219 			break;
220 		}
221 	}
222 
223 	return q;
224 }
225 
226 /* Set new default qdisc to use */
227 int qdisc_set_default(const char *name)
228 {
229 	const struct Qdisc_ops *ops;
230 
231 	if (!capable(CAP_NET_ADMIN))
232 		return -EPERM;
233 
234 	write_lock(&qdisc_mod_lock);
235 	ops = qdisc_lookup_default(name);
236 	if (!ops) {
237 		/* Not found, drop lock and try to load module */
238 		write_unlock(&qdisc_mod_lock);
239 		request_module("sch_%s", name);
240 		write_lock(&qdisc_mod_lock);
241 
242 		ops = qdisc_lookup_default(name);
243 	}
244 
245 	if (ops) {
246 		/* Set new default */
247 		module_put(default_qdisc_ops->owner);
248 		default_qdisc_ops = ops;
249 	}
250 	write_unlock(&qdisc_mod_lock);
251 
252 	return ops ? 0 : -ENOENT;
253 }
254 
255 /* We know handle. Find qdisc among all qdisc's attached to device
256    (root qdisc, all its children, children of children etc.)
257  */
258 
259 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
260 {
261 	struct Qdisc *q;
262 
263 	if (!(root->flags & TCQ_F_BUILTIN) &&
264 	    root->handle == handle)
265 		return root;
266 
267 	list_for_each_entry(q, &root->list, list) {
268 		if (q->handle == handle)
269 			return q;
270 	}
271 	return NULL;
272 }
273 
274 static void qdisc_list_add(struct Qdisc *q)
275 {
276 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
277 		list_add_tail(&q->list, &qdisc_dev(q)->qdisc->list);
278 }
279 
280 void qdisc_list_del(struct Qdisc *q)
281 {
282 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
283 		list_del(&q->list);
284 }
285 EXPORT_SYMBOL(qdisc_list_del);
286 
287 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
288 {
289 	struct Qdisc *q;
290 
291 	q = qdisc_match_from_root(dev->qdisc, handle);
292 	if (q)
293 		goto out;
294 
295 	if (dev_ingress_queue(dev))
296 		q = qdisc_match_from_root(
297 			dev_ingress_queue(dev)->qdisc_sleeping,
298 			handle);
299 out:
300 	return q;
301 }
302 
303 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
304 {
305 	unsigned long cl;
306 	struct Qdisc *leaf;
307 	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
308 
309 	if (cops == NULL)
310 		return NULL;
311 	cl = cops->get(p, classid);
312 
313 	if (cl == 0)
314 		return NULL;
315 	leaf = cops->leaf(p, cl);
316 	cops->put(p, cl);
317 	return leaf;
318 }
319 
320 /* Find queueing discipline by name */
321 
322 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
323 {
324 	struct Qdisc_ops *q = NULL;
325 
326 	if (kind) {
327 		read_lock(&qdisc_mod_lock);
328 		for (q = qdisc_base; q; q = q->next) {
329 			if (nla_strcmp(kind, q->id) == 0) {
330 				if (!try_module_get(q->owner))
331 					q = NULL;
332 				break;
333 			}
334 		}
335 		read_unlock(&qdisc_mod_lock);
336 	}
337 	return q;
338 }
339 
340 /* The linklayer setting were not transferred from iproute2, in older
341  * versions, and the rate tables lookup systems have been dropped in
342  * the kernel. To keep backward compatible with older iproute2 tc
343  * utils, we detect the linklayer setting by detecting if the rate
344  * table were modified.
345  *
346  * For linklayer ATM table entries, the rate table will be aligned to
347  * 48 bytes, thus some table entries will contain the same value.  The
348  * mpu (min packet unit) is also encoded into the old rate table, thus
349  * starting from the mpu, we find low and high table entries for
350  * mapping this cell.  If these entries contain the same value, when
351  * the rate tables have been modified for linklayer ATM.
352  *
353  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
354  * and then roundup to the next cell, calc the table entry one below,
355  * and compare.
356  */
357 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
358 {
359 	int low       = roundup(r->mpu, 48);
360 	int high      = roundup(low+1, 48);
361 	int cell_low  = low >> r->cell_log;
362 	int cell_high = (high >> r->cell_log) - 1;
363 
364 	/* rtab is too inaccurate at rates > 100Mbit/s */
365 	if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
366 		pr_debug("TC linklayer: Giving up ATM detection\n");
367 		return TC_LINKLAYER_ETHERNET;
368 	}
369 
370 	if ((cell_high > cell_low) && (cell_high < 256)
371 	    && (rtab[cell_low] == rtab[cell_high])) {
372 		pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
373 			 cell_low, cell_high, rtab[cell_high]);
374 		return TC_LINKLAYER_ATM;
375 	}
376 	return TC_LINKLAYER_ETHERNET;
377 }
378 
379 static struct qdisc_rate_table *qdisc_rtab_list;
380 
381 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
382 {
383 	struct qdisc_rate_table *rtab;
384 
385 	if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
386 	    nla_len(tab) != TC_RTAB_SIZE)
387 		return NULL;
388 
389 	for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
390 		if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
391 		    !memcmp(&rtab->data, nla_data(tab), 1024)) {
392 			rtab->refcnt++;
393 			return rtab;
394 		}
395 	}
396 
397 	rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
398 	if (rtab) {
399 		rtab->rate = *r;
400 		rtab->refcnt = 1;
401 		memcpy(rtab->data, nla_data(tab), 1024);
402 		if (r->linklayer == TC_LINKLAYER_UNAWARE)
403 			r->linklayer = __detect_linklayer(r, rtab->data);
404 		rtab->next = qdisc_rtab_list;
405 		qdisc_rtab_list = rtab;
406 	}
407 	return rtab;
408 }
409 EXPORT_SYMBOL(qdisc_get_rtab);
410 
411 void qdisc_put_rtab(struct qdisc_rate_table *tab)
412 {
413 	struct qdisc_rate_table *rtab, **rtabp;
414 
415 	if (!tab || --tab->refcnt)
416 		return;
417 
418 	for (rtabp = &qdisc_rtab_list;
419 	     (rtab = *rtabp) != NULL;
420 	     rtabp = &rtab->next) {
421 		if (rtab == tab) {
422 			*rtabp = rtab->next;
423 			kfree(rtab);
424 			return;
425 		}
426 	}
427 }
428 EXPORT_SYMBOL(qdisc_put_rtab);
429 
430 static LIST_HEAD(qdisc_stab_list);
431 static DEFINE_SPINLOCK(qdisc_stab_lock);
432 
433 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
434 	[TCA_STAB_BASE]	= { .len = sizeof(struct tc_sizespec) },
435 	[TCA_STAB_DATA] = { .type = NLA_BINARY },
436 };
437 
438 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
439 {
440 	struct nlattr *tb[TCA_STAB_MAX + 1];
441 	struct qdisc_size_table *stab;
442 	struct tc_sizespec *s;
443 	unsigned int tsize = 0;
444 	u16 *tab = NULL;
445 	int err;
446 
447 	err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
448 	if (err < 0)
449 		return ERR_PTR(err);
450 	if (!tb[TCA_STAB_BASE])
451 		return ERR_PTR(-EINVAL);
452 
453 	s = nla_data(tb[TCA_STAB_BASE]);
454 
455 	if (s->tsize > 0) {
456 		if (!tb[TCA_STAB_DATA])
457 			return ERR_PTR(-EINVAL);
458 		tab = nla_data(tb[TCA_STAB_DATA]);
459 		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
460 	}
461 
462 	if (tsize != s->tsize || (!tab && tsize > 0))
463 		return ERR_PTR(-EINVAL);
464 
465 	spin_lock(&qdisc_stab_lock);
466 
467 	list_for_each_entry(stab, &qdisc_stab_list, list) {
468 		if (memcmp(&stab->szopts, s, sizeof(*s)))
469 			continue;
470 		if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
471 			continue;
472 		stab->refcnt++;
473 		spin_unlock(&qdisc_stab_lock);
474 		return stab;
475 	}
476 
477 	spin_unlock(&qdisc_stab_lock);
478 
479 	stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
480 	if (!stab)
481 		return ERR_PTR(-ENOMEM);
482 
483 	stab->refcnt = 1;
484 	stab->szopts = *s;
485 	if (tsize > 0)
486 		memcpy(stab->data, tab, tsize * sizeof(u16));
487 
488 	spin_lock(&qdisc_stab_lock);
489 	list_add_tail(&stab->list, &qdisc_stab_list);
490 	spin_unlock(&qdisc_stab_lock);
491 
492 	return stab;
493 }
494 
495 static void stab_kfree_rcu(struct rcu_head *head)
496 {
497 	kfree(container_of(head, struct qdisc_size_table, rcu));
498 }
499 
500 void qdisc_put_stab(struct qdisc_size_table *tab)
501 {
502 	if (!tab)
503 		return;
504 
505 	spin_lock(&qdisc_stab_lock);
506 
507 	if (--tab->refcnt == 0) {
508 		list_del(&tab->list);
509 		call_rcu_bh(&tab->rcu, stab_kfree_rcu);
510 	}
511 
512 	spin_unlock(&qdisc_stab_lock);
513 }
514 EXPORT_SYMBOL(qdisc_put_stab);
515 
516 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
517 {
518 	struct nlattr *nest;
519 
520 	nest = nla_nest_start(skb, TCA_STAB);
521 	if (nest == NULL)
522 		goto nla_put_failure;
523 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
524 		goto nla_put_failure;
525 	nla_nest_end(skb, nest);
526 
527 	return skb->len;
528 
529 nla_put_failure:
530 	return -1;
531 }
532 
533 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
534 {
535 	int pkt_len, slot;
536 
537 	pkt_len = skb->len + stab->szopts.overhead;
538 	if (unlikely(!stab->szopts.tsize))
539 		goto out;
540 
541 	slot = pkt_len + stab->szopts.cell_align;
542 	if (unlikely(slot < 0))
543 		slot = 0;
544 
545 	slot >>= stab->szopts.cell_log;
546 	if (likely(slot < stab->szopts.tsize))
547 		pkt_len = stab->data[slot];
548 	else
549 		pkt_len = stab->data[stab->szopts.tsize - 1] *
550 				(slot / stab->szopts.tsize) +
551 				stab->data[slot % stab->szopts.tsize];
552 
553 	pkt_len <<= stab->szopts.size_log;
554 out:
555 	if (unlikely(pkt_len < 1))
556 		pkt_len = 1;
557 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
558 }
559 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
560 
561 void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
562 {
563 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
564 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
565 			txt, qdisc->ops->id, qdisc->handle >> 16);
566 		qdisc->flags |= TCQ_F_WARN_NONWC;
567 	}
568 }
569 EXPORT_SYMBOL(qdisc_warn_nonwc);
570 
571 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
572 {
573 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
574 						 timer);
575 
576 	qdisc_unthrottled(wd->qdisc);
577 	__netif_schedule(qdisc_root(wd->qdisc));
578 
579 	return HRTIMER_NORESTART;
580 }
581 
582 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
583 {
584 	hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
585 	wd->timer.function = qdisc_watchdog;
586 	wd->qdisc = qdisc;
587 }
588 EXPORT_SYMBOL(qdisc_watchdog_init);
589 
590 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
591 {
592 	if (test_bit(__QDISC_STATE_DEACTIVATED,
593 		     &qdisc_root_sleeping(wd->qdisc)->state))
594 		return;
595 
596 	qdisc_throttled(wd->qdisc);
597 
598 	hrtimer_start(&wd->timer,
599 		      ns_to_ktime(expires),
600 		      HRTIMER_MODE_ABS);
601 }
602 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
603 
604 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
605 {
606 	hrtimer_cancel(&wd->timer);
607 	qdisc_unthrottled(wd->qdisc);
608 }
609 EXPORT_SYMBOL(qdisc_watchdog_cancel);
610 
611 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
612 {
613 	unsigned int size = n * sizeof(struct hlist_head), i;
614 	struct hlist_head *h;
615 
616 	if (size <= PAGE_SIZE)
617 		h = kmalloc(size, GFP_KERNEL);
618 	else
619 		h = (struct hlist_head *)
620 			__get_free_pages(GFP_KERNEL, get_order(size));
621 
622 	if (h != NULL) {
623 		for (i = 0; i < n; i++)
624 			INIT_HLIST_HEAD(&h[i]);
625 	}
626 	return h;
627 }
628 
629 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
630 {
631 	unsigned int size = n * sizeof(struct hlist_head);
632 
633 	if (size <= PAGE_SIZE)
634 		kfree(h);
635 	else
636 		free_pages((unsigned long)h, get_order(size));
637 }
638 
639 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
640 {
641 	struct Qdisc_class_common *cl;
642 	struct hlist_node *next;
643 	struct hlist_head *nhash, *ohash;
644 	unsigned int nsize, nmask, osize;
645 	unsigned int i, h;
646 
647 	/* Rehash when load factor exceeds 0.75 */
648 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
649 		return;
650 	nsize = clhash->hashsize * 2;
651 	nmask = nsize - 1;
652 	nhash = qdisc_class_hash_alloc(nsize);
653 	if (nhash == NULL)
654 		return;
655 
656 	ohash = clhash->hash;
657 	osize = clhash->hashsize;
658 
659 	sch_tree_lock(sch);
660 	for (i = 0; i < osize; i++) {
661 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
662 			h = qdisc_class_hash(cl->classid, nmask);
663 			hlist_add_head(&cl->hnode, &nhash[h]);
664 		}
665 	}
666 	clhash->hash     = nhash;
667 	clhash->hashsize = nsize;
668 	clhash->hashmask = nmask;
669 	sch_tree_unlock(sch);
670 
671 	qdisc_class_hash_free(ohash, osize);
672 }
673 EXPORT_SYMBOL(qdisc_class_hash_grow);
674 
675 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
676 {
677 	unsigned int size = 4;
678 
679 	clhash->hash = qdisc_class_hash_alloc(size);
680 	if (clhash->hash == NULL)
681 		return -ENOMEM;
682 	clhash->hashsize  = size;
683 	clhash->hashmask  = size - 1;
684 	clhash->hashelems = 0;
685 	return 0;
686 }
687 EXPORT_SYMBOL(qdisc_class_hash_init);
688 
689 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
690 {
691 	qdisc_class_hash_free(clhash->hash, clhash->hashsize);
692 }
693 EXPORT_SYMBOL(qdisc_class_hash_destroy);
694 
695 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
696 			     struct Qdisc_class_common *cl)
697 {
698 	unsigned int h;
699 
700 	INIT_HLIST_NODE(&cl->hnode);
701 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
702 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
703 	clhash->hashelems++;
704 }
705 EXPORT_SYMBOL(qdisc_class_hash_insert);
706 
707 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
708 			     struct Qdisc_class_common *cl)
709 {
710 	hlist_del(&cl->hnode);
711 	clhash->hashelems--;
712 }
713 EXPORT_SYMBOL(qdisc_class_hash_remove);
714 
715 /* Allocate an unique handle from space managed by kernel
716  * Possible range is [8000-FFFF]:0000 (0x8000 values)
717  */
718 static u32 qdisc_alloc_handle(struct net_device *dev)
719 {
720 	int i = 0x8000;
721 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
722 
723 	do {
724 		autohandle += TC_H_MAKE(0x10000U, 0);
725 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
726 			autohandle = TC_H_MAKE(0x80000000U, 0);
727 		if (!qdisc_lookup(dev, autohandle))
728 			return autohandle;
729 		cond_resched();
730 	} while	(--i > 0);
731 
732 	return 0;
733 }
734 
735 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
736 {
737 	const struct Qdisc_class_ops *cops;
738 	unsigned long cl;
739 	u32 parentid;
740 	int drops;
741 
742 	if (n == 0)
743 		return;
744 	drops = max_t(int, n, 0);
745 	while ((parentid = sch->parent)) {
746 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
747 			return;
748 
749 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
750 		if (sch == NULL) {
751 			WARN_ON(parentid != TC_H_ROOT);
752 			return;
753 		}
754 		cops = sch->ops->cl_ops;
755 		if (cops->qlen_notify) {
756 			cl = cops->get(sch, parentid);
757 			cops->qlen_notify(sch, cl);
758 			cops->put(sch, cl);
759 		}
760 		sch->q.qlen -= n;
761 		sch->qstats.drops += drops;
762 	}
763 }
764 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
765 
766 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
767 			       struct nlmsghdr *n, u32 clid,
768 			       struct Qdisc *old, struct Qdisc *new)
769 {
770 	if (new || old)
771 		qdisc_notify(net, skb, n, clid, old, new);
772 
773 	if (old)
774 		qdisc_destroy(old);
775 }
776 
777 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
778  * to device "dev".
779  *
780  * When appropriate send a netlink notification using 'skb'
781  * and "n".
782  *
783  * On success, destroy old qdisc.
784  */
785 
786 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
787 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
788 		       struct Qdisc *new, struct Qdisc *old)
789 {
790 	struct Qdisc *q = old;
791 	struct net *net = dev_net(dev);
792 	int err = 0;
793 
794 	if (parent == NULL) {
795 		unsigned int i, num_q, ingress;
796 
797 		ingress = 0;
798 		num_q = dev->num_tx_queues;
799 		if ((q && q->flags & TCQ_F_INGRESS) ||
800 		    (new && new->flags & TCQ_F_INGRESS)) {
801 			num_q = 1;
802 			ingress = 1;
803 			if (!dev_ingress_queue(dev))
804 				return -ENOENT;
805 		}
806 
807 		if (dev->flags & IFF_UP)
808 			dev_deactivate(dev);
809 
810 		if (new && new->ops->attach) {
811 			new->ops->attach(new);
812 			num_q = 0;
813 		}
814 
815 		for (i = 0; i < num_q; i++) {
816 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
817 
818 			if (!ingress)
819 				dev_queue = netdev_get_tx_queue(dev, i);
820 
821 			old = dev_graft_qdisc(dev_queue, new);
822 			if (new && i > 0)
823 				atomic_inc(&new->refcnt);
824 
825 			if (!ingress)
826 				qdisc_destroy(old);
827 		}
828 
829 		if (!ingress) {
830 			notify_and_destroy(net, skb, n, classid,
831 					   dev->qdisc, new);
832 			if (new && !new->ops->attach)
833 				atomic_inc(&new->refcnt);
834 			dev->qdisc = new ? : &noop_qdisc;
835 		} else {
836 			notify_and_destroy(net, skb, n, classid, old, new);
837 		}
838 
839 		if (dev->flags & IFF_UP)
840 			dev_activate(dev);
841 	} else {
842 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
843 
844 		err = -EOPNOTSUPP;
845 		if (cops && cops->graft) {
846 			unsigned long cl = cops->get(parent, classid);
847 			if (cl) {
848 				err = cops->graft(parent, cl, new, &old);
849 				cops->put(parent, cl);
850 			} else
851 				err = -ENOENT;
852 		}
853 		if (!err)
854 			notify_and_destroy(net, skb, n, classid, old, new);
855 	}
856 	return err;
857 }
858 
859 /* lockdep annotation is needed for ingress; egress gets it only for name */
860 static struct lock_class_key qdisc_tx_lock;
861 static struct lock_class_key qdisc_rx_lock;
862 
863 /*
864    Allocate and initialize new qdisc.
865 
866    Parameters are passed via opt.
867  */
868 
869 static struct Qdisc *
870 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
871 	     struct Qdisc *p, u32 parent, u32 handle,
872 	     struct nlattr **tca, int *errp)
873 {
874 	int err;
875 	struct nlattr *kind = tca[TCA_KIND];
876 	struct Qdisc *sch;
877 	struct Qdisc_ops *ops;
878 	struct qdisc_size_table *stab;
879 
880 	ops = qdisc_lookup_ops(kind);
881 #ifdef CONFIG_MODULES
882 	if (ops == NULL && kind != NULL) {
883 		char name[IFNAMSIZ];
884 		if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
885 			/* We dropped the RTNL semaphore in order to
886 			 * perform the module load.  So, even if we
887 			 * succeeded in loading the module we have to
888 			 * tell the caller to replay the request.  We
889 			 * indicate this using -EAGAIN.
890 			 * We replay the request because the device may
891 			 * go away in the mean time.
892 			 */
893 			rtnl_unlock();
894 			request_module("sch_%s", name);
895 			rtnl_lock();
896 			ops = qdisc_lookup_ops(kind);
897 			if (ops != NULL) {
898 				/* We will try again qdisc_lookup_ops,
899 				 * so don't keep a reference.
900 				 */
901 				module_put(ops->owner);
902 				err = -EAGAIN;
903 				goto err_out;
904 			}
905 		}
906 	}
907 #endif
908 
909 	err = -ENOENT;
910 	if (ops == NULL)
911 		goto err_out;
912 
913 	sch = qdisc_alloc(dev_queue, ops);
914 	if (IS_ERR(sch)) {
915 		err = PTR_ERR(sch);
916 		goto err_out2;
917 	}
918 
919 	sch->parent = parent;
920 
921 	if (handle == TC_H_INGRESS) {
922 		sch->flags |= TCQ_F_INGRESS;
923 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
924 		lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
925 	} else {
926 		if (handle == 0) {
927 			handle = qdisc_alloc_handle(dev);
928 			err = -ENOMEM;
929 			if (handle == 0)
930 				goto err_out3;
931 		}
932 		lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
933 		if (!netif_is_multiqueue(dev))
934 			sch->flags |= TCQ_F_ONETXQUEUE;
935 	}
936 
937 	sch->handle = handle;
938 
939 	if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
940 		if (tca[TCA_STAB]) {
941 			stab = qdisc_get_stab(tca[TCA_STAB]);
942 			if (IS_ERR(stab)) {
943 				err = PTR_ERR(stab);
944 				goto err_out4;
945 			}
946 			rcu_assign_pointer(sch->stab, stab);
947 		}
948 		if (tca[TCA_RATE]) {
949 			spinlock_t *root_lock;
950 
951 			err = -EOPNOTSUPP;
952 			if (sch->flags & TCQ_F_MQROOT)
953 				goto err_out4;
954 
955 			if ((sch->parent != TC_H_ROOT) &&
956 			    !(sch->flags & TCQ_F_INGRESS) &&
957 			    (!p || !(p->flags & TCQ_F_MQROOT)))
958 				root_lock = qdisc_root_sleeping_lock(sch);
959 			else
960 				root_lock = qdisc_lock(sch);
961 
962 			err = gen_new_estimator(&sch->bstats, &sch->rate_est,
963 						root_lock, tca[TCA_RATE]);
964 			if (err)
965 				goto err_out4;
966 		}
967 
968 		qdisc_list_add(sch);
969 
970 		return sch;
971 	}
972 err_out3:
973 	dev_put(dev);
974 	kfree((char *) sch - sch->padded);
975 err_out2:
976 	module_put(ops->owner);
977 err_out:
978 	*errp = err;
979 	return NULL;
980 
981 err_out4:
982 	/*
983 	 * Any broken qdiscs that would require a ops->reset() here?
984 	 * The qdisc was never in action so it shouldn't be necessary.
985 	 */
986 	qdisc_put_stab(rtnl_dereference(sch->stab));
987 	if (ops->destroy)
988 		ops->destroy(sch);
989 	goto err_out3;
990 }
991 
992 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
993 {
994 	struct qdisc_size_table *ostab, *stab = NULL;
995 	int err = 0;
996 
997 	if (tca[TCA_OPTIONS]) {
998 		if (sch->ops->change == NULL)
999 			return -EINVAL;
1000 		err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1001 		if (err)
1002 			return err;
1003 	}
1004 
1005 	if (tca[TCA_STAB]) {
1006 		stab = qdisc_get_stab(tca[TCA_STAB]);
1007 		if (IS_ERR(stab))
1008 			return PTR_ERR(stab);
1009 	}
1010 
1011 	ostab = rtnl_dereference(sch->stab);
1012 	rcu_assign_pointer(sch->stab, stab);
1013 	qdisc_put_stab(ostab);
1014 
1015 	if (tca[TCA_RATE]) {
1016 		/* NB: ignores errors from replace_estimator
1017 		   because change can't be undone. */
1018 		if (sch->flags & TCQ_F_MQROOT)
1019 			goto out;
1020 		gen_replace_estimator(&sch->bstats, &sch->rate_est,
1021 					    qdisc_root_sleeping_lock(sch),
1022 					    tca[TCA_RATE]);
1023 	}
1024 out:
1025 	return 0;
1026 }
1027 
1028 struct check_loop_arg {
1029 	struct qdisc_walker	w;
1030 	struct Qdisc		*p;
1031 	int			depth;
1032 };
1033 
1034 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
1035 
1036 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1037 {
1038 	struct check_loop_arg	arg;
1039 
1040 	if (q->ops->cl_ops == NULL)
1041 		return 0;
1042 
1043 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1044 	arg.w.fn = check_loop_fn;
1045 	arg.depth = depth;
1046 	arg.p = p;
1047 	q->ops->cl_ops->walk(q, &arg.w);
1048 	return arg.w.stop ? -ELOOP : 0;
1049 }
1050 
1051 static int
1052 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1053 {
1054 	struct Qdisc *leaf;
1055 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1056 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1057 
1058 	leaf = cops->leaf(q, cl);
1059 	if (leaf) {
1060 		if (leaf == arg->p || arg->depth > 7)
1061 			return -ELOOP;
1062 		return check_loop(leaf, arg->p, arg->depth + 1);
1063 	}
1064 	return 0;
1065 }
1066 
1067 /*
1068  * Delete/get qdisc.
1069  */
1070 
1071 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1072 {
1073 	struct net *net = sock_net(skb->sk);
1074 	struct tcmsg *tcm = nlmsg_data(n);
1075 	struct nlattr *tca[TCA_MAX + 1];
1076 	struct net_device *dev;
1077 	u32 clid;
1078 	struct Qdisc *q = NULL;
1079 	struct Qdisc *p = NULL;
1080 	int err;
1081 
1082 	if ((n->nlmsg_type != RTM_GETQDISC) && !capable(CAP_NET_ADMIN))
1083 		return -EPERM;
1084 
1085 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1086 	if (err < 0)
1087 		return err;
1088 
1089 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1090 	if (!dev)
1091 		return -ENODEV;
1092 
1093 	clid = tcm->tcm_parent;
1094 	if (clid) {
1095 		if (clid != TC_H_ROOT) {
1096 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1097 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1098 				if (!p)
1099 					return -ENOENT;
1100 				q = qdisc_leaf(p, clid);
1101 			} else if (dev_ingress_queue(dev)) {
1102 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1103 			}
1104 		} else {
1105 			q = dev->qdisc;
1106 		}
1107 		if (!q)
1108 			return -ENOENT;
1109 
1110 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1111 			return -EINVAL;
1112 	} else {
1113 		q = qdisc_lookup(dev, tcm->tcm_handle);
1114 		if (!q)
1115 			return -ENOENT;
1116 	}
1117 
1118 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1119 		return -EINVAL;
1120 
1121 	if (n->nlmsg_type == RTM_DELQDISC) {
1122 		if (!clid)
1123 			return -EINVAL;
1124 		if (q->handle == 0)
1125 			return -ENOENT;
1126 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1127 		if (err != 0)
1128 			return err;
1129 	} else {
1130 		qdisc_notify(net, skb, n, clid, NULL, q);
1131 	}
1132 	return 0;
1133 }
1134 
1135 /*
1136  * Create/change qdisc.
1137  */
1138 
1139 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1140 {
1141 	struct net *net = sock_net(skb->sk);
1142 	struct tcmsg *tcm;
1143 	struct nlattr *tca[TCA_MAX + 1];
1144 	struct net_device *dev;
1145 	u32 clid;
1146 	struct Qdisc *q, *p;
1147 	int err;
1148 
1149 	if (!capable(CAP_NET_ADMIN))
1150 		return -EPERM;
1151 
1152 replay:
1153 	/* Reinit, just in case something touches this. */
1154 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1155 	if (err < 0)
1156 		return err;
1157 
1158 	tcm = nlmsg_data(n);
1159 	clid = tcm->tcm_parent;
1160 	q = p = NULL;
1161 
1162 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1163 	if (!dev)
1164 		return -ENODEV;
1165 
1166 
1167 	if (clid) {
1168 		if (clid != TC_H_ROOT) {
1169 			if (clid != TC_H_INGRESS) {
1170 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1171 				if (!p)
1172 					return -ENOENT;
1173 				q = qdisc_leaf(p, clid);
1174 			} else if (dev_ingress_queue_create(dev)) {
1175 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1176 			}
1177 		} else {
1178 			q = dev->qdisc;
1179 		}
1180 
1181 		/* It may be default qdisc, ignore it */
1182 		if (q && q->handle == 0)
1183 			q = NULL;
1184 
1185 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1186 			if (tcm->tcm_handle) {
1187 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1188 					return -EEXIST;
1189 				if (TC_H_MIN(tcm->tcm_handle))
1190 					return -EINVAL;
1191 				q = qdisc_lookup(dev, tcm->tcm_handle);
1192 				if (!q)
1193 					goto create_n_graft;
1194 				if (n->nlmsg_flags & NLM_F_EXCL)
1195 					return -EEXIST;
1196 				if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1197 					return -EINVAL;
1198 				if (q == p ||
1199 				    (p && check_loop(q, p, 0)))
1200 					return -ELOOP;
1201 				atomic_inc(&q->refcnt);
1202 				goto graft;
1203 			} else {
1204 				if (!q)
1205 					goto create_n_graft;
1206 
1207 				/* This magic test requires explanation.
1208 				 *
1209 				 *   We know, that some child q is already
1210 				 *   attached to this parent and have choice:
1211 				 *   either to change it or to create/graft new one.
1212 				 *
1213 				 *   1. We are allowed to create/graft only
1214 				 *   if CREATE and REPLACE flags are set.
1215 				 *
1216 				 *   2. If EXCL is set, requestor wanted to say,
1217 				 *   that qdisc tcm_handle is not expected
1218 				 *   to exist, so that we choose create/graft too.
1219 				 *
1220 				 *   3. The last case is when no flags are set.
1221 				 *   Alas, it is sort of hole in API, we
1222 				 *   cannot decide what to do unambiguously.
1223 				 *   For now we select create/graft, if
1224 				 *   user gave KIND, which does not match existing.
1225 				 */
1226 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1227 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1228 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1229 				     (tca[TCA_KIND] &&
1230 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1231 					goto create_n_graft;
1232 			}
1233 		}
1234 	} else {
1235 		if (!tcm->tcm_handle)
1236 			return -EINVAL;
1237 		q = qdisc_lookup(dev, tcm->tcm_handle);
1238 	}
1239 
1240 	/* Change qdisc parameters */
1241 	if (q == NULL)
1242 		return -ENOENT;
1243 	if (n->nlmsg_flags & NLM_F_EXCL)
1244 		return -EEXIST;
1245 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1246 		return -EINVAL;
1247 	err = qdisc_change(q, tca);
1248 	if (err == 0)
1249 		qdisc_notify(net, skb, n, clid, NULL, q);
1250 	return err;
1251 
1252 create_n_graft:
1253 	if (!(n->nlmsg_flags & NLM_F_CREATE))
1254 		return -ENOENT;
1255 	if (clid == TC_H_INGRESS) {
1256 		if (dev_ingress_queue(dev))
1257 			q = qdisc_create(dev, dev_ingress_queue(dev), p,
1258 					 tcm->tcm_parent, tcm->tcm_parent,
1259 					 tca, &err);
1260 		else
1261 			err = -ENOENT;
1262 	} else {
1263 		struct netdev_queue *dev_queue;
1264 
1265 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1266 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1267 		else if (p)
1268 			dev_queue = p->dev_queue;
1269 		else
1270 			dev_queue = netdev_get_tx_queue(dev, 0);
1271 
1272 		q = qdisc_create(dev, dev_queue, p,
1273 				 tcm->tcm_parent, tcm->tcm_handle,
1274 				 tca, &err);
1275 	}
1276 	if (q == NULL) {
1277 		if (err == -EAGAIN)
1278 			goto replay;
1279 		return err;
1280 	}
1281 
1282 graft:
1283 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1284 	if (err) {
1285 		if (q)
1286 			qdisc_destroy(q);
1287 		return err;
1288 	}
1289 
1290 	return 0;
1291 }
1292 
1293 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1294 			 u32 portid, u32 seq, u16 flags, int event)
1295 {
1296 	struct tcmsg *tcm;
1297 	struct nlmsghdr  *nlh;
1298 	unsigned char *b = skb_tail_pointer(skb);
1299 	struct gnet_dump d;
1300 	struct qdisc_size_table *stab;
1301 
1302 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1303 	if (!nlh)
1304 		goto out_nlmsg_trim;
1305 	tcm = nlmsg_data(nlh);
1306 	tcm->tcm_family = AF_UNSPEC;
1307 	tcm->tcm__pad1 = 0;
1308 	tcm->tcm__pad2 = 0;
1309 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1310 	tcm->tcm_parent = clid;
1311 	tcm->tcm_handle = q->handle;
1312 	tcm->tcm_info = atomic_read(&q->refcnt);
1313 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1314 		goto nla_put_failure;
1315 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
1316 		goto nla_put_failure;
1317 	q->qstats.qlen = q->q.qlen;
1318 
1319 	stab = rtnl_dereference(q->stab);
1320 	if (stab && qdisc_dump_stab(skb, stab) < 0)
1321 		goto nla_put_failure;
1322 
1323 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1324 					 qdisc_root_sleeping_lock(q), &d) < 0)
1325 		goto nla_put_failure;
1326 
1327 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1328 		goto nla_put_failure;
1329 
1330 	if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1331 	    gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1332 	    gnet_stats_copy_queue(&d, &q->qstats) < 0)
1333 		goto nla_put_failure;
1334 
1335 	if (gnet_stats_finish_copy(&d) < 0)
1336 		goto nla_put_failure;
1337 
1338 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1339 	return skb->len;
1340 
1341 out_nlmsg_trim:
1342 nla_put_failure:
1343 	nlmsg_trim(skb, b);
1344 	return -1;
1345 }
1346 
1347 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1348 {
1349 	return (q->flags & TCQ_F_BUILTIN) ? true : false;
1350 }
1351 
1352 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1353 			struct nlmsghdr *n, u32 clid,
1354 			struct Qdisc *old, struct Qdisc *new)
1355 {
1356 	struct sk_buff *skb;
1357 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1358 
1359 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1360 	if (!skb)
1361 		return -ENOBUFS;
1362 
1363 	if (old && !tc_qdisc_dump_ignore(old)) {
1364 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1365 				  0, RTM_DELQDISC) < 0)
1366 			goto err_out;
1367 	}
1368 	if (new && !tc_qdisc_dump_ignore(new)) {
1369 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1370 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1371 			goto err_out;
1372 	}
1373 
1374 	if (skb->len)
1375 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1376 				      n->nlmsg_flags & NLM_F_ECHO);
1377 
1378 err_out:
1379 	kfree_skb(skb);
1380 	return -EINVAL;
1381 }
1382 
1383 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1384 			      struct netlink_callback *cb,
1385 			      int *q_idx_p, int s_q_idx)
1386 {
1387 	int ret = 0, q_idx = *q_idx_p;
1388 	struct Qdisc *q;
1389 
1390 	if (!root)
1391 		return 0;
1392 
1393 	q = root;
1394 	if (q_idx < s_q_idx) {
1395 		q_idx++;
1396 	} else {
1397 		if (!tc_qdisc_dump_ignore(q) &&
1398 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1399 				  cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1400 			goto done;
1401 		q_idx++;
1402 	}
1403 	list_for_each_entry(q, &root->list, list) {
1404 		if (q_idx < s_q_idx) {
1405 			q_idx++;
1406 			continue;
1407 		}
1408 		if (!tc_qdisc_dump_ignore(q) &&
1409 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1410 				  cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1411 			goto done;
1412 		q_idx++;
1413 	}
1414 
1415 out:
1416 	*q_idx_p = q_idx;
1417 	return ret;
1418 done:
1419 	ret = -1;
1420 	goto out;
1421 }
1422 
1423 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1424 {
1425 	struct net *net = sock_net(skb->sk);
1426 	int idx, q_idx;
1427 	int s_idx, s_q_idx;
1428 	struct net_device *dev;
1429 
1430 	s_idx = cb->args[0];
1431 	s_q_idx = q_idx = cb->args[1];
1432 
1433 	rcu_read_lock();
1434 	idx = 0;
1435 	for_each_netdev_rcu(net, dev) {
1436 		struct netdev_queue *dev_queue;
1437 
1438 		if (idx < s_idx)
1439 			goto cont;
1440 		if (idx > s_idx)
1441 			s_q_idx = 0;
1442 		q_idx = 0;
1443 
1444 		if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1445 			goto done;
1446 
1447 		dev_queue = dev_ingress_queue(dev);
1448 		if (dev_queue &&
1449 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1450 				       &q_idx, s_q_idx) < 0)
1451 			goto done;
1452 
1453 cont:
1454 		idx++;
1455 	}
1456 
1457 done:
1458 	rcu_read_unlock();
1459 
1460 	cb->args[0] = idx;
1461 	cb->args[1] = q_idx;
1462 
1463 	return skb->len;
1464 }
1465 
1466 
1467 
1468 /************************************************
1469  *	Traffic classes manipulation.		*
1470  ************************************************/
1471 
1472 
1473 
1474 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1475 {
1476 	struct net *net = sock_net(skb->sk);
1477 	struct tcmsg *tcm = nlmsg_data(n);
1478 	struct nlattr *tca[TCA_MAX + 1];
1479 	struct net_device *dev;
1480 	struct Qdisc *q = NULL;
1481 	const struct Qdisc_class_ops *cops;
1482 	unsigned long cl = 0;
1483 	unsigned long new_cl;
1484 	u32 portid;
1485 	u32 clid;
1486 	u32 qid;
1487 	int err;
1488 
1489 	if ((n->nlmsg_type != RTM_GETTCLASS) && !capable(CAP_NET_ADMIN))
1490 		return -EPERM;
1491 
1492 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1493 	if (err < 0)
1494 		return err;
1495 
1496 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1497 	if (!dev)
1498 		return -ENODEV;
1499 
1500 	/*
1501 	   parent == TC_H_UNSPEC - unspecified parent.
1502 	   parent == TC_H_ROOT   - class is root, which has no parent.
1503 	   parent == X:0	 - parent is root class.
1504 	   parent == X:Y	 - parent is a node in hierarchy.
1505 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
1506 
1507 	   handle == 0:0	 - generate handle from kernel pool.
1508 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
1509 	   handle == X:Y	 - clear.
1510 	   handle == X:0	 - root class.
1511 	 */
1512 
1513 	/* Step 1. Determine qdisc handle X:0 */
1514 
1515 	portid = tcm->tcm_parent;
1516 	clid = tcm->tcm_handle;
1517 	qid = TC_H_MAJ(clid);
1518 
1519 	if (portid != TC_H_ROOT) {
1520 		u32 qid1 = TC_H_MAJ(portid);
1521 
1522 		if (qid && qid1) {
1523 			/* If both majors are known, they must be identical. */
1524 			if (qid != qid1)
1525 				return -EINVAL;
1526 		} else if (qid1) {
1527 			qid = qid1;
1528 		} else if (qid == 0)
1529 			qid = dev->qdisc->handle;
1530 
1531 		/* Now qid is genuine qdisc handle consistent
1532 		 * both with parent and child.
1533 		 *
1534 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1535 		 */
1536 		if (portid)
1537 			portid = TC_H_MAKE(qid, portid);
1538 	} else {
1539 		if (qid == 0)
1540 			qid = dev->qdisc->handle;
1541 	}
1542 
1543 	/* OK. Locate qdisc */
1544 	q = qdisc_lookup(dev, qid);
1545 	if (!q)
1546 		return -ENOENT;
1547 
1548 	/* An check that it supports classes */
1549 	cops = q->ops->cl_ops;
1550 	if (cops == NULL)
1551 		return -EINVAL;
1552 
1553 	/* Now try to get class */
1554 	if (clid == 0) {
1555 		if (portid == TC_H_ROOT)
1556 			clid = qid;
1557 	} else
1558 		clid = TC_H_MAKE(qid, clid);
1559 
1560 	if (clid)
1561 		cl = cops->get(q, clid);
1562 
1563 	if (cl == 0) {
1564 		err = -ENOENT;
1565 		if (n->nlmsg_type != RTM_NEWTCLASS ||
1566 		    !(n->nlmsg_flags & NLM_F_CREATE))
1567 			goto out;
1568 	} else {
1569 		switch (n->nlmsg_type) {
1570 		case RTM_NEWTCLASS:
1571 			err = -EEXIST;
1572 			if (n->nlmsg_flags & NLM_F_EXCL)
1573 				goto out;
1574 			break;
1575 		case RTM_DELTCLASS:
1576 			err = -EOPNOTSUPP;
1577 			if (cops->delete)
1578 				err = cops->delete(q, cl);
1579 			if (err == 0)
1580 				tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1581 			goto out;
1582 		case RTM_GETTCLASS:
1583 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1584 			goto out;
1585 		default:
1586 			err = -EINVAL;
1587 			goto out;
1588 		}
1589 	}
1590 
1591 	new_cl = cl;
1592 	err = -EOPNOTSUPP;
1593 	if (cops->change)
1594 		err = cops->change(q, clid, portid, tca, &new_cl);
1595 	if (err == 0)
1596 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1597 
1598 out:
1599 	if (cl)
1600 		cops->put(q, cl);
1601 
1602 	return err;
1603 }
1604 
1605 
1606 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1607 			  unsigned long cl,
1608 			  u32 portid, u32 seq, u16 flags, int event)
1609 {
1610 	struct tcmsg *tcm;
1611 	struct nlmsghdr  *nlh;
1612 	unsigned char *b = skb_tail_pointer(skb);
1613 	struct gnet_dump d;
1614 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1615 
1616 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1617 	if (!nlh)
1618 		goto out_nlmsg_trim;
1619 	tcm = nlmsg_data(nlh);
1620 	tcm->tcm_family = AF_UNSPEC;
1621 	tcm->tcm__pad1 = 0;
1622 	tcm->tcm__pad2 = 0;
1623 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1624 	tcm->tcm_parent = q->handle;
1625 	tcm->tcm_handle = q->handle;
1626 	tcm->tcm_info = 0;
1627 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1628 		goto nla_put_failure;
1629 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1630 		goto nla_put_failure;
1631 
1632 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1633 					 qdisc_root_sleeping_lock(q), &d) < 0)
1634 		goto nla_put_failure;
1635 
1636 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1637 		goto nla_put_failure;
1638 
1639 	if (gnet_stats_finish_copy(&d) < 0)
1640 		goto nla_put_failure;
1641 
1642 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1643 	return skb->len;
1644 
1645 out_nlmsg_trim:
1646 nla_put_failure:
1647 	nlmsg_trim(skb, b);
1648 	return -1;
1649 }
1650 
1651 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1652 			 struct nlmsghdr *n, struct Qdisc *q,
1653 			 unsigned long cl, int event)
1654 {
1655 	struct sk_buff *skb;
1656 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1657 
1658 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1659 	if (!skb)
1660 		return -ENOBUFS;
1661 
1662 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1663 		kfree_skb(skb);
1664 		return -EINVAL;
1665 	}
1666 
1667 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1668 			      n->nlmsg_flags & NLM_F_ECHO);
1669 }
1670 
1671 struct qdisc_dump_args {
1672 	struct qdisc_walker	w;
1673 	struct sk_buff		*skb;
1674 	struct netlink_callback	*cb;
1675 };
1676 
1677 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1678 {
1679 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1680 
1681 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1682 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1683 }
1684 
1685 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1686 				struct tcmsg *tcm, struct netlink_callback *cb,
1687 				int *t_p, int s_t)
1688 {
1689 	struct qdisc_dump_args arg;
1690 
1691 	if (tc_qdisc_dump_ignore(q) ||
1692 	    *t_p < s_t || !q->ops->cl_ops ||
1693 	    (tcm->tcm_parent &&
1694 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1695 		(*t_p)++;
1696 		return 0;
1697 	}
1698 	if (*t_p > s_t)
1699 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1700 	arg.w.fn = qdisc_class_dump;
1701 	arg.skb = skb;
1702 	arg.cb = cb;
1703 	arg.w.stop  = 0;
1704 	arg.w.skip = cb->args[1];
1705 	arg.w.count = 0;
1706 	q->ops->cl_ops->walk(q, &arg.w);
1707 	cb->args[1] = arg.w.count;
1708 	if (arg.w.stop)
1709 		return -1;
1710 	(*t_p)++;
1711 	return 0;
1712 }
1713 
1714 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1715 			       struct tcmsg *tcm, struct netlink_callback *cb,
1716 			       int *t_p, int s_t)
1717 {
1718 	struct Qdisc *q;
1719 
1720 	if (!root)
1721 		return 0;
1722 
1723 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1724 		return -1;
1725 
1726 	list_for_each_entry(q, &root->list, list) {
1727 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1728 			return -1;
1729 	}
1730 
1731 	return 0;
1732 }
1733 
1734 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1735 {
1736 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
1737 	struct net *net = sock_net(skb->sk);
1738 	struct netdev_queue *dev_queue;
1739 	struct net_device *dev;
1740 	int t, s_t;
1741 
1742 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1743 		return 0;
1744 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
1745 	if (!dev)
1746 		return 0;
1747 
1748 	s_t = cb->args[0];
1749 	t = 0;
1750 
1751 	if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1752 		goto done;
1753 
1754 	dev_queue = dev_ingress_queue(dev);
1755 	if (dev_queue &&
1756 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1757 				&t, s_t) < 0)
1758 		goto done;
1759 
1760 done:
1761 	cb->args[0] = t;
1762 
1763 	dev_put(dev);
1764 	return skb->len;
1765 }
1766 
1767 /* Main classifier routine: scans classifier chain attached
1768  * to this qdisc, (optionally) tests for protocol and asks
1769  * specific classifiers.
1770  */
1771 int tc_classify_compat(struct sk_buff *skb, const struct tcf_proto *tp,
1772 		       struct tcf_result *res)
1773 {
1774 	__be16 protocol = skb->protocol;
1775 	int err;
1776 
1777 	for (; tp; tp = tp->next) {
1778 		if (tp->protocol != protocol &&
1779 		    tp->protocol != htons(ETH_P_ALL))
1780 			continue;
1781 		err = tp->classify(skb, tp, res);
1782 
1783 		if (err >= 0) {
1784 #ifdef CONFIG_NET_CLS_ACT
1785 			if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1786 				skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1787 #endif
1788 			return err;
1789 		}
1790 	}
1791 	return -1;
1792 }
1793 EXPORT_SYMBOL(tc_classify_compat);
1794 
1795 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1796 		struct tcf_result *res)
1797 {
1798 	int err = 0;
1799 #ifdef CONFIG_NET_CLS_ACT
1800 	const struct tcf_proto *otp = tp;
1801 reclassify:
1802 #endif
1803 
1804 	err = tc_classify_compat(skb, tp, res);
1805 #ifdef CONFIG_NET_CLS_ACT
1806 	if (err == TC_ACT_RECLASSIFY) {
1807 		u32 verd = G_TC_VERD(skb->tc_verd);
1808 		tp = otp;
1809 
1810 		if (verd++ >= MAX_REC_LOOP) {
1811 			net_notice_ratelimited("%s: packet reclassify loop rule prio %u protocol %02x\n",
1812 					       tp->q->ops->id,
1813 					       tp->prio & 0xffff,
1814 					       ntohs(tp->protocol));
1815 			return TC_ACT_SHOT;
1816 		}
1817 		skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1818 		goto reclassify;
1819 	}
1820 #endif
1821 	return err;
1822 }
1823 EXPORT_SYMBOL(tc_classify);
1824 
1825 void tcf_destroy(struct tcf_proto *tp)
1826 {
1827 	tp->ops->destroy(tp);
1828 	module_put(tp->ops->owner);
1829 	kfree(tp);
1830 }
1831 
1832 void tcf_destroy_chain(struct tcf_proto **fl)
1833 {
1834 	struct tcf_proto *tp;
1835 
1836 	while ((tp = *fl) != NULL) {
1837 		*fl = tp->next;
1838 		tcf_destroy(tp);
1839 	}
1840 }
1841 EXPORT_SYMBOL(tcf_destroy_chain);
1842 
1843 #ifdef CONFIG_PROC_FS
1844 static int psched_show(struct seq_file *seq, void *v)
1845 {
1846 	struct timespec ts;
1847 
1848 	hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1849 	seq_printf(seq, "%08x %08x %08x %08x\n",
1850 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1851 		   1000000,
1852 		   (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1853 
1854 	return 0;
1855 }
1856 
1857 static int psched_open(struct inode *inode, struct file *file)
1858 {
1859 	return single_open(file, psched_show, NULL);
1860 }
1861 
1862 static const struct file_operations psched_fops = {
1863 	.owner = THIS_MODULE,
1864 	.open = psched_open,
1865 	.read  = seq_read,
1866 	.llseek = seq_lseek,
1867 	.release = single_release,
1868 };
1869 
1870 static int __net_init psched_net_init(struct net *net)
1871 {
1872 	struct proc_dir_entry *e;
1873 
1874 	e = proc_create("psched", 0, net->proc_net, &psched_fops);
1875 	if (e == NULL)
1876 		return -ENOMEM;
1877 
1878 	return 0;
1879 }
1880 
1881 static void __net_exit psched_net_exit(struct net *net)
1882 {
1883 	remove_proc_entry("psched", net->proc_net);
1884 }
1885 #else
1886 static int __net_init psched_net_init(struct net *net)
1887 {
1888 	return 0;
1889 }
1890 
1891 static void __net_exit psched_net_exit(struct net *net)
1892 {
1893 }
1894 #endif
1895 
1896 static struct pernet_operations psched_net_ops = {
1897 	.init = psched_net_init,
1898 	.exit = psched_net_exit,
1899 };
1900 
1901 static int __init pktsched_init(void)
1902 {
1903 	int err;
1904 
1905 	err = register_pernet_subsys(&psched_net_ops);
1906 	if (err) {
1907 		pr_err("pktsched_init: "
1908 		       "cannot initialize per netns operations\n");
1909 		return err;
1910 	}
1911 
1912 	register_qdisc(&pfifo_fast_ops);
1913 	register_qdisc(&pfifo_qdisc_ops);
1914 	register_qdisc(&bfifo_qdisc_ops);
1915 	register_qdisc(&pfifo_head_drop_qdisc_ops);
1916 	register_qdisc(&mq_qdisc_ops);
1917 
1918 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1919 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1920 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1921 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1922 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1923 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1924 
1925 	return 0;
1926 }
1927 
1928 subsys_initcall(pktsched_init);
1929