xref: /openbmc/linux/net/sched/sch_api.c (revision afb46f79)
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/unregister 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 void qdisc_list_add(struct Qdisc *q)
275 {
276 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
277 		struct Qdisc *root = qdisc_dev(q)->qdisc;
278 
279 		WARN_ON_ONCE(root == &noop_qdisc);
280 		list_add_tail(&q->list, &root->list);
281 	}
282 }
283 EXPORT_SYMBOL(qdisc_list_add);
284 
285 void qdisc_list_del(struct Qdisc *q)
286 {
287 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS))
288 		list_del(&q->list);
289 }
290 EXPORT_SYMBOL(qdisc_list_del);
291 
292 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
293 {
294 	struct Qdisc *q;
295 
296 	q = qdisc_match_from_root(dev->qdisc, handle);
297 	if (q)
298 		goto out;
299 
300 	if (dev_ingress_queue(dev))
301 		q = qdisc_match_from_root(
302 			dev_ingress_queue(dev)->qdisc_sleeping,
303 			handle);
304 out:
305 	return q;
306 }
307 
308 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
309 {
310 	unsigned long cl;
311 	struct Qdisc *leaf;
312 	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
313 
314 	if (cops == NULL)
315 		return NULL;
316 	cl = cops->get(p, classid);
317 
318 	if (cl == 0)
319 		return NULL;
320 	leaf = cops->leaf(p, cl);
321 	cops->put(p, cl);
322 	return leaf;
323 }
324 
325 /* Find queueing discipline by name */
326 
327 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
328 {
329 	struct Qdisc_ops *q = NULL;
330 
331 	if (kind) {
332 		read_lock(&qdisc_mod_lock);
333 		for (q = qdisc_base; q; q = q->next) {
334 			if (nla_strcmp(kind, q->id) == 0) {
335 				if (!try_module_get(q->owner))
336 					q = NULL;
337 				break;
338 			}
339 		}
340 		read_unlock(&qdisc_mod_lock);
341 	}
342 	return q;
343 }
344 
345 /* The linklayer setting were not transferred from iproute2, in older
346  * versions, and the rate tables lookup systems have been dropped in
347  * the kernel. To keep backward compatible with older iproute2 tc
348  * utils, we detect the linklayer setting by detecting if the rate
349  * table were modified.
350  *
351  * For linklayer ATM table entries, the rate table will be aligned to
352  * 48 bytes, thus some table entries will contain the same value.  The
353  * mpu (min packet unit) is also encoded into the old rate table, thus
354  * starting from the mpu, we find low and high table entries for
355  * mapping this cell.  If these entries contain the same value, when
356  * the rate tables have been modified for linklayer ATM.
357  *
358  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
359  * and then roundup to the next cell, calc the table entry one below,
360  * and compare.
361  */
362 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
363 {
364 	int low       = roundup(r->mpu, 48);
365 	int high      = roundup(low+1, 48);
366 	int cell_low  = low >> r->cell_log;
367 	int cell_high = (high >> r->cell_log) - 1;
368 
369 	/* rtab is too inaccurate at rates > 100Mbit/s */
370 	if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
371 		pr_debug("TC linklayer: Giving up ATM detection\n");
372 		return TC_LINKLAYER_ETHERNET;
373 	}
374 
375 	if ((cell_high > cell_low) && (cell_high < 256)
376 	    && (rtab[cell_low] == rtab[cell_high])) {
377 		pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
378 			 cell_low, cell_high, rtab[cell_high]);
379 		return TC_LINKLAYER_ATM;
380 	}
381 	return TC_LINKLAYER_ETHERNET;
382 }
383 
384 static struct qdisc_rate_table *qdisc_rtab_list;
385 
386 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r, struct nlattr *tab)
387 {
388 	struct qdisc_rate_table *rtab;
389 
390 	if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
391 	    nla_len(tab) != TC_RTAB_SIZE)
392 		return NULL;
393 
394 	for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
395 		if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
396 		    !memcmp(&rtab->data, nla_data(tab), 1024)) {
397 			rtab->refcnt++;
398 			return rtab;
399 		}
400 	}
401 
402 	rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
403 	if (rtab) {
404 		rtab->rate = *r;
405 		rtab->refcnt = 1;
406 		memcpy(rtab->data, nla_data(tab), 1024);
407 		if (r->linklayer == TC_LINKLAYER_UNAWARE)
408 			r->linklayer = __detect_linklayer(r, rtab->data);
409 		rtab->next = qdisc_rtab_list;
410 		qdisc_rtab_list = rtab;
411 	}
412 	return rtab;
413 }
414 EXPORT_SYMBOL(qdisc_get_rtab);
415 
416 void qdisc_put_rtab(struct qdisc_rate_table *tab)
417 {
418 	struct qdisc_rate_table *rtab, **rtabp;
419 
420 	if (!tab || --tab->refcnt)
421 		return;
422 
423 	for (rtabp = &qdisc_rtab_list;
424 	     (rtab = *rtabp) != NULL;
425 	     rtabp = &rtab->next) {
426 		if (rtab == tab) {
427 			*rtabp = rtab->next;
428 			kfree(rtab);
429 			return;
430 		}
431 	}
432 }
433 EXPORT_SYMBOL(qdisc_put_rtab);
434 
435 static LIST_HEAD(qdisc_stab_list);
436 static DEFINE_SPINLOCK(qdisc_stab_lock);
437 
438 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
439 	[TCA_STAB_BASE]	= { .len = sizeof(struct tc_sizespec) },
440 	[TCA_STAB_DATA] = { .type = NLA_BINARY },
441 };
442 
443 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt)
444 {
445 	struct nlattr *tb[TCA_STAB_MAX + 1];
446 	struct qdisc_size_table *stab;
447 	struct tc_sizespec *s;
448 	unsigned int tsize = 0;
449 	u16 *tab = NULL;
450 	int err;
451 
452 	err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy);
453 	if (err < 0)
454 		return ERR_PTR(err);
455 	if (!tb[TCA_STAB_BASE])
456 		return ERR_PTR(-EINVAL);
457 
458 	s = nla_data(tb[TCA_STAB_BASE]);
459 
460 	if (s->tsize > 0) {
461 		if (!tb[TCA_STAB_DATA])
462 			return ERR_PTR(-EINVAL);
463 		tab = nla_data(tb[TCA_STAB_DATA]);
464 		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
465 	}
466 
467 	if (tsize != s->tsize || (!tab && tsize > 0))
468 		return ERR_PTR(-EINVAL);
469 
470 	spin_lock(&qdisc_stab_lock);
471 
472 	list_for_each_entry(stab, &qdisc_stab_list, list) {
473 		if (memcmp(&stab->szopts, s, sizeof(*s)))
474 			continue;
475 		if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
476 			continue;
477 		stab->refcnt++;
478 		spin_unlock(&qdisc_stab_lock);
479 		return stab;
480 	}
481 
482 	spin_unlock(&qdisc_stab_lock);
483 
484 	stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
485 	if (!stab)
486 		return ERR_PTR(-ENOMEM);
487 
488 	stab->refcnt = 1;
489 	stab->szopts = *s;
490 	if (tsize > 0)
491 		memcpy(stab->data, tab, tsize * sizeof(u16));
492 
493 	spin_lock(&qdisc_stab_lock);
494 	list_add_tail(&stab->list, &qdisc_stab_list);
495 	spin_unlock(&qdisc_stab_lock);
496 
497 	return stab;
498 }
499 
500 static void stab_kfree_rcu(struct rcu_head *head)
501 {
502 	kfree(container_of(head, struct qdisc_size_table, rcu));
503 }
504 
505 void qdisc_put_stab(struct qdisc_size_table *tab)
506 {
507 	if (!tab)
508 		return;
509 
510 	spin_lock(&qdisc_stab_lock);
511 
512 	if (--tab->refcnt == 0) {
513 		list_del(&tab->list);
514 		call_rcu_bh(&tab->rcu, stab_kfree_rcu);
515 	}
516 
517 	spin_unlock(&qdisc_stab_lock);
518 }
519 EXPORT_SYMBOL(qdisc_put_stab);
520 
521 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
522 {
523 	struct nlattr *nest;
524 
525 	nest = nla_nest_start(skb, TCA_STAB);
526 	if (nest == NULL)
527 		goto nla_put_failure;
528 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
529 		goto nla_put_failure;
530 	nla_nest_end(skb, nest);
531 
532 	return skb->len;
533 
534 nla_put_failure:
535 	return -1;
536 }
537 
538 void __qdisc_calculate_pkt_len(struct sk_buff *skb, const struct qdisc_size_table *stab)
539 {
540 	int pkt_len, slot;
541 
542 	pkt_len = skb->len + stab->szopts.overhead;
543 	if (unlikely(!stab->szopts.tsize))
544 		goto out;
545 
546 	slot = pkt_len + stab->szopts.cell_align;
547 	if (unlikely(slot < 0))
548 		slot = 0;
549 
550 	slot >>= stab->szopts.cell_log;
551 	if (likely(slot < stab->szopts.tsize))
552 		pkt_len = stab->data[slot];
553 	else
554 		pkt_len = stab->data[stab->szopts.tsize - 1] *
555 				(slot / stab->szopts.tsize) +
556 				stab->data[slot % stab->szopts.tsize];
557 
558 	pkt_len <<= stab->szopts.size_log;
559 out:
560 	if (unlikely(pkt_len < 1))
561 		pkt_len = 1;
562 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
563 }
564 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
565 
566 void qdisc_warn_nonwc(char *txt, struct Qdisc *qdisc)
567 {
568 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
569 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
570 			txt, qdisc->ops->id, qdisc->handle >> 16);
571 		qdisc->flags |= TCQ_F_WARN_NONWC;
572 	}
573 }
574 EXPORT_SYMBOL(qdisc_warn_nonwc);
575 
576 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
577 {
578 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
579 						 timer);
580 
581 	qdisc_unthrottled(wd->qdisc);
582 	__netif_schedule(qdisc_root(wd->qdisc));
583 
584 	return HRTIMER_NORESTART;
585 }
586 
587 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
588 {
589 	hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
590 	wd->timer.function = qdisc_watchdog;
591 	wd->qdisc = qdisc;
592 }
593 EXPORT_SYMBOL(qdisc_watchdog_init);
594 
595 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
596 {
597 	if (test_bit(__QDISC_STATE_DEACTIVATED,
598 		     &qdisc_root_sleeping(wd->qdisc)->state))
599 		return;
600 
601 	qdisc_throttled(wd->qdisc);
602 
603 	hrtimer_start(&wd->timer,
604 		      ns_to_ktime(expires),
605 		      HRTIMER_MODE_ABS);
606 }
607 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
608 
609 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
610 {
611 	hrtimer_cancel(&wd->timer);
612 	qdisc_unthrottled(wd->qdisc);
613 }
614 EXPORT_SYMBOL(qdisc_watchdog_cancel);
615 
616 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
617 {
618 	unsigned int size = n * sizeof(struct hlist_head), i;
619 	struct hlist_head *h;
620 
621 	if (size <= PAGE_SIZE)
622 		h = kmalloc(size, GFP_KERNEL);
623 	else
624 		h = (struct hlist_head *)
625 			__get_free_pages(GFP_KERNEL, get_order(size));
626 
627 	if (h != NULL) {
628 		for (i = 0; i < n; i++)
629 			INIT_HLIST_HEAD(&h[i]);
630 	}
631 	return h;
632 }
633 
634 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
635 {
636 	unsigned int size = n * sizeof(struct hlist_head);
637 
638 	if (size <= PAGE_SIZE)
639 		kfree(h);
640 	else
641 		free_pages((unsigned long)h, get_order(size));
642 }
643 
644 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
645 {
646 	struct Qdisc_class_common *cl;
647 	struct hlist_node *next;
648 	struct hlist_head *nhash, *ohash;
649 	unsigned int nsize, nmask, osize;
650 	unsigned int i, h;
651 
652 	/* Rehash when load factor exceeds 0.75 */
653 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
654 		return;
655 	nsize = clhash->hashsize * 2;
656 	nmask = nsize - 1;
657 	nhash = qdisc_class_hash_alloc(nsize);
658 	if (nhash == NULL)
659 		return;
660 
661 	ohash = clhash->hash;
662 	osize = clhash->hashsize;
663 
664 	sch_tree_lock(sch);
665 	for (i = 0; i < osize; i++) {
666 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
667 			h = qdisc_class_hash(cl->classid, nmask);
668 			hlist_add_head(&cl->hnode, &nhash[h]);
669 		}
670 	}
671 	clhash->hash     = nhash;
672 	clhash->hashsize = nsize;
673 	clhash->hashmask = nmask;
674 	sch_tree_unlock(sch);
675 
676 	qdisc_class_hash_free(ohash, osize);
677 }
678 EXPORT_SYMBOL(qdisc_class_hash_grow);
679 
680 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
681 {
682 	unsigned int size = 4;
683 
684 	clhash->hash = qdisc_class_hash_alloc(size);
685 	if (clhash->hash == NULL)
686 		return -ENOMEM;
687 	clhash->hashsize  = size;
688 	clhash->hashmask  = size - 1;
689 	clhash->hashelems = 0;
690 	return 0;
691 }
692 EXPORT_SYMBOL(qdisc_class_hash_init);
693 
694 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
695 {
696 	qdisc_class_hash_free(clhash->hash, clhash->hashsize);
697 }
698 EXPORT_SYMBOL(qdisc_class_hash_destroy);
699 
700 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
701 			     struct Qdisc_class_common *cl)
702 {
703 	unsigned int h;
704 
705 	INIT_HLIST_NODE(&cl->hnode);
706 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
707 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
708 	clhash->hashelems++;
709 }
710 EXPORT_SYMBOL(qdisc_class_hash_insert);
711 
712 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
713 			     struct Qdisc_class_common *cl)
714 {
715 	hlist_del(&cl->hnode);
716 	clhash->hashelems--;
717 }
718 EXPORT_SYMBOL(qdisc_class_hash_remove);
719 
720 /* Allocate an unique handle from space managed by kernel
721  * Possible range is [8000-FFFF]:0000 (0x8000 values)
722  */
723 static u32 qdisc_alloc_handle(struct net_device *dev)
724 {
725 	int i = 0x8000;
726 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
727 
728 	do {
729 		autohandle += TC_H_MAKE(0x10000U, 0);
730 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
731 			autohandle = TC_H_MAKE(0x80000000U, 0);
732 		if (!qdisc_lookup(dev, autohandle))
733 			return autohandle;
734 		cond_resched();
735 	} while	(--i > 0);
736 
737 	return 0;
738 }
739 
740 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
741 {
742 	const struct Qdisc_class_ops *cops;
743 	unsigned long cl;
744 	u32 parentid;
745 	int drops;
746 
747 	if (n == 0)
748 		return;
749 	drops = max_t(int, n, 0);
750 	while ((parentid = sch->parent)) {
751 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
752 			return;
753 
754 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
755 		if (sch == NULL) {
756 			WARN_ON(parentid != TC_H_ROOT);
757 			return;
758 		}
759 		cops = sch->ops->cl_ops;
760 		if (cops->qlen_notify) {
761 			cl = cops->get(sch, parentid);
762 			cops->qlen_notify(sch, cl);
763 			cops->put(sch, cl);
764 		}
765 		sch->q.qlen -= n;
766 		sch->qstats.drops += drops;
767 	}
768 }
769 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
770 
771 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
772 			       struct nlmsghdr *n, u32 clid,
773 			       struct Qdisc *old, struct Qdisc *new)
774 {
775 	if (new || old)
776 		qdisc_notify(net, skb, n, clid, old, new);
777 
778 	if (old)
779 		qdisc_destroy(old);
780 }
781 
782 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
783  * to device "dev".
784  *
785  * When appropriate send a netlink notification using 'skb'
786  * and "n".
787  *
788  * On success, destroy old qdisc.
789  */
790 
791 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
792 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
793 		       struct Qdisc *new, struct Qdisc *old)
794 {
795 	struct Qdisc *q = old;
796 	struct net *net = dev_net(dev);
797 	int err = 0;
798 
799 	if (parent == NULL) {
800 		unsigned int i, num_q, ingress;
801 
802 		ingress = 0;
803 		num_q = dev->num_tx_queues;
804 		if ((q && q->flags & TCQ_F_INGRESS) ||
805 		    (new && new->flags & TCQ_F_INGRESS)) {
806 			num_q = 1;
807 			ingress = 1;
808 			if (!dev_ingress_queue(dev))
809 				return -ENOENT;
810 		}
811 
812 		if (dev->flags & IFF_UP)
813 			dev_deactivate(dev);
814 
815 		if (new && new->ops->attach) {
816 			new->ops->attach(new);
817 			num_q = 0;
818 		}
819 
820 		for (i = 0; i < num_q; i++) {
821 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
822 
823 			if (!ingress)
824 				dev_queue = netdev_get_tx_queue(dev, i);
825 
826 			old = dev_graft_qdisc(dev_queue, new);
827 			if (new && i > 0)
828 				atomic_inc(&new->refcnt);
829 
830 			if (!ingress)
831 				qdisc_destroy(old);
832 		}
833 
834 		if (!ingress) {
835 			notify_and_destroy(net, skb, n, classid,
836 					   dev->qdisc, new);
837 			if (new && !new->ops->attach)
838 				atomic_inc(&new->refcnt);
839 			dev->qdisc = new ? : &noop_qdisc;
840 		} else {
841 			notify_and_destroy(net, skb, n, classid, old, new);
842 		}
843 
844 		if (dev->flags & IFF_UP)
845 			dev_activate(dev);
846 	} else {
847 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
848 
849 		err = -EOPNOTSUPP;
850 		if (cops && cops->graft) {
851 			unsigned long cl = cops->get(parent, classid);
852 			if (cl) {
853 				err = cops->graft(parent, cl, new, &old);
854 				cops->put(parent, cl);
855 			} else
856 				err = -ENOENT;
857 		}
858 		if (!err)
859 			notify_and_destroy(net, skb, n, classid, old, new);
860 	}
861 	return err;
862 }
863 
864 /* lockdep annotation is needed for ingress; egress gets it only for name */
865 static struct lock_class_key qdisc_tx_lock;
866 static struct lock_class_key qdisc_rx_lock;
867 
868 /*
869    Allocate and initialize new qdisc.
870 
871    Parameters are passed via opt.
872  */
873 
874 static struct Qdisc *
875 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
876 	     struct Qdisc *p, u32 parent, u32 handle,
877 	     struct nlattr **tca, int *errp)
878 {
879 	int err;
880 	struct nlattr *kind = tca[TCA_KIND];
881 	struct Qdisc *sch;
882 	struct Qdisc_ops *ops;
883 	struct qdisc_size_table *stab;
884 
885 	ops = qdisc_lookup_ops(kind);
886 #ifdef CONFIG_MODULES
887 	if (ops == NULL && kind != NULL) {
888 		char name[IFNAMSIZ];
889 		if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
890 			/* We dropped the RTNL semaphore in order to
891 			 * perform the module load.  So, even if we
892 			 * succeeded in loading the module we have to
893 			 * tell the caller to replay the request.  We
894 			 * indicate this using -EAGAIN.
895 			 * We replay the request because the device may
896 			 * go away in the mean time.
897 			 */
898 			rtnl_unlock();
899 			request_module("sch_%s", name);
900 			rtnl_lock();
901 			ops = qdisc_lookup_ops(kind);
902 			if (ops != NULL) {
903 				/* We will try again qdisc_lookup_ops,
904 				 * so don't keep a reference.
905 				 */
906 				module_put(ops->owner);
907 				err = -EAGAIN;
908 				goto err_out;
909 			}
910 		}
911 	}
912 #endif
913 
914 	err = -ENOENT;
915 	if (ops == NULL)
916 		goto err_out;
917 
918 	sch = qdisc_alloc(dev_queue, ops);
919 	if (IS_ERR(sch)) {
920 		err = PTR_ERR(sch);
921 		goto err_out2;
922 	}
923 
924 	sch->parent = parent;
925 
926 	if (handle == TC_H_INGRESS) {
927 		sch->flags |= TCQ_F_INGRESS;
928 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
929 		lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
930 	} else {
931 		if (handle == 0) {
932 			handle = qdisc_alloc_handle(dev);
933 			err = -ENOMEM;
934 			if (handle == 0)
935 				goto err_out3;
936 		}
937 		lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
938 		if (!netif_is_multiqueue(dev))
939 			sch->flags |= TCQ_F_ONETXQUEUE;
940 	}
941 
942 	sch->handle = handle;
943 
944 	if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
945 		if (tca[TCA_STAB]) {
946 			stab = qdisc_get_stab(tca[TCA_STAB]);
947 			if (IS_ERR(stab)) {
948 				err = PTR_ERR(stab);
949 				goto err_out4;
950 			}
951 			rcu_assign_pointer(sch->stab, stab);
952 		}
953 		if (tca[TCA_RATE]) {
954 			spinlock_t *root_lock;
955 
956 			err = -EOPNOTSUPP;
957 			if (sch->flags & TCQ_F_MQROOT)
958 				goto err_out4;
959 
960 			if ((sch->parent != TC_H_ROOT) &&
961 			    !(sch->flags & TCQ_F_INGRESS) &&
962 			    (!p || !(p->flags & TCQ_F_MQROOT)))
963 				root_lock = qdisc_root_sleeping_lock(sch);
964 			else
965 				root_lock = qdisc_lock(sch);
966 
967 			err = gen_new_estimator(&sch->bstats, &sch->rate_est,
968 						root_lock, tca[TCA_RATE]);
969 			if (err)
970 				goto err_out4;
971 		}
972 
973 		qdisc_list_add(sch);
974 
975 		return sch;
976 	}
977 err_out3:
978 	dev_put(dev);
979 	kfree((char *) sch - sch->padded);
980 err_out2:
981 	module_put(ops->owner);
982 err_out:
983 	*errp = err;
984 	return NULL;
985 
986 err_out4:
987 	/*
988 	 * Any broken qdiscs that would require a ops->reset() here?
989 	 * The qdisc was never in action so it shouldn't be necessary.
990 	 */
991 	qdisc_put_stab(rtnl_dereference(sch->stab));
992 	if (ops->destroy)
993 		ops->destroy(sch);
994 	goto err_out3;
995 }
996 
997 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
998 {
999 	struct qdisc_size_table *ostab, *stab = NULL;
1000 	int err = 0;
1001 
1002 	if (tca[TCA_OPTIONS]) {
1003 		if (sch->ops->change == NULL)
1004 			return -EINVAL;
1005 		err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1006 		if (err)
1007 			return err;
1008 	}
1009 
1010 	if (tca[TCA_STAB]) {
1011 		stab = qdisc_get_stab(tca[TCA_STAB]);
1012 		if (IS_ERR(stab))
1013 			return PTR_ERR(stab);
1014 	}
1015 
1016 	ostab = rtnl_dereference(sch->stab);
1017 	rcu_assign_pointer(sch->stab, stab);
1018 	qdisc_put_stab(ostab);
1019 
1020 	if (tca[TCA_RATE]) {
1021 		/* NB: ignores errors from replace_estimator
1022 		   because change can't be undone. */
1023 		if (sch->flags & TCQ_F_MQROOT)
1024 			goto out;
1025 		gen_replace_estimator(&sch->bstats, &sch->rate_est,
1026 					    qdisc_root_sleeping_lock(sch),
1027 					    tca[TCA_RATE]);
1028 	}
1029 out:
1030 	return 0;
1031 }
1032 
1033 struct check_loop_arg {
1034 	struct qdisc_walker	w;
1035 	struct Qdisc		*p;
1036 	int			depth;
1037 };
1038 
1039 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
1040 
1041 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1042 {
1043 	struct check_loop_arg	arg;
1044 
1045 	if (q->ops->cl_ops == NULL)
1046 		return 0;
1047 
1048 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1049 	arg.w.fn = check_loop_fn;
1050 	arg.depth = depth;
1051 	arg.p = p;
1052 	q->ops->cl_ops->walk(q, &arg.w);
1053 	return arg.w.stop ? -ELOOP : 0;
1054 }
1055 
1056 static int
1057 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1058 {
1059 	struct Qdisc *leaf;
1060 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1061 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1062 
1063 	leaf = cops->leaf(q, cl);
1064 	if (leaf) {
1065 		if (leaf == arg->p || arg->depth > 7)
1066 			return -ELOOP;
1067 		return check_loop(leaf, arg->p, arg->depth + 1);
1068 	}
1069 	return 0;
1070 }
1071 
1072 /*
1073  * Delete/get qdisc.
1074  */
1075 
1076 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1077 {
1078 	struct net *net = sock_net(skb->sk);
1079 	struct tcmsg *tcm = nlmsg_data(n);
1080 	struct nlattr *tca[TCA_MAX + 1];
1081 	struct net_device *dev;
1082 	u32 clid;
1083 	struct Qdisc *q = NULL;
1084 	struct Qdisc *p = NULL;
1085 	int err;
1086 
1087 	if ((n->nlmsg_type != RTM_GETQDISC) && !capable(CAP_NET_ADMIN))
1088 		return -EPERM;
1089 
1090 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1091 	if (err < 0)
1092 		return err;
1093 
1094 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1095 	if (!dev)
1096 		return -ENODEV;
1097 
1098 	clid = tcm->tcm_parent;
1099 	if (clid) {
1100 		if (clid != TC_H_ROOT) {
1101 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1102 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1103 				if (!p)
1104 					return -ENOENT;
1105 				q = qdisc_leaf(p, clid);
1106 			} else if (dev_ingress_queue(dev)) {
1107 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1108 			}
1109 		} else {
1110 			q = dev->qdisc;
1111 		}
1112 		if (!q)
1113 			return -ENOENT;
1114 
1115 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1116 			return -EINVAL;
1117 	} else {
1118 		q = qdisc_lookup(dev, tcm->tcm_handle);
1119 		if (!q)
1120 			return -ENOENT;
1121 	}
1122 
1123 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1124 		return -EINVAL;
1125 
1126 	if (n->nlmsg_type == RTM_DELQDISC) {
1127 		if (!clid)
1128 			return -EINVAL;
1129 		if (q->handle == 0)
1130 			return -ENOENT;
1131 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1132 		if (err != 0)
1133 			return err;
1134 	} else {
1135 		qdisc_notify(net, skb, n, clid, NULL, q);
1136 	}
1137 	return 0;
1138 }
1139 
1140 /*
1141  * Create/change qdisc.
1142  */
1143 
1144 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1145 {
1146 	struct net *net = sock_net(skb->sk);
1147 	struct tcmsg *tcm;
1148 	struct nlattr *tca[TCA_MAX + 1];
1149 	struct net_device *dev;
1150 	u32 clid;
1151 	struct Qdisc *q, *p;
1152 	int err;
1153 
1154 	if (!capable(CAP_NET_ADMIN))
1155 		return -EPERM;
1156 
1157 replay:
1158 	/* Reinit, just in case something touches this. */
1159 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1160 	if (err < 0)
1161 		return err;
1162 
1163 	tcm = nlmsg_data(n);
1164 	clid = tcm->tcm_parent;
1165 	q = p = NULL;
1166 
1167 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1168 	if (!dev)
1169 		return -ENODEV;
1170 
1171 
1172 	if (clid) {
1173 		if (clid != TC_H_ROOT) {
1174 			if (clid != TC_H_INGRESS) {
1175 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1176 				if (!p)
1177 					return -ENOENT;
1178 				q = qdisc_leaf(p, clid);
1179 			} else if (dev_ingress_queue_create(dev)) {
1180 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1181 			}
1182 		} else {
1183 			q = dev->qdisc;
1184 		}
1185 
1186 		/* It may be default qdisc, ignore it */
1187 		if (q && q->handle == 0)
1188 			q = NULL;
1189 
1190 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1191 			if (tcm->tcm_handle) {
1192 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1193 					return -EEXIST;
1194 				if (TC_H_MIN(tcm->tcm_handle))
1195 					return -EINVAL;
1196 				q = qdisc_lookup(dev, tcm->tcm_handle);
1197 				if (!q)
1198 					goto create_n_graft;
1199 				if (n->nlmsg_flags & NLM_F_EXCL)
1200 					return -EEXIST;
1201 				if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1202 					return -EINVAL;
1203 				if (q == p ||
1204 				    (p && check_loop(q, p, 0)))
1205 					return -ELOOP;
1206 				atomic_inc(&q->refcnt);
1207 				goto graft;
1208 			} else {
1209 				if (!q)
1210 					goto create_n_graft;
1211 
1212 				/* This magic test requires explanation.
1213 				 *
1214 				 *   We know, that some child q is already
1215 				 *   attached to this parent and have choice:
1216 				 *   either to change it or to create/graft new one.
1217 				 *
1218 				 *   1. We are allowed to create/graft only
1219 				 *   if CREATE and REPLACE flags are set.
1220 				 *
1221 				 *   2. If EXCL is set, requestor wanted to say,
1222 				 *   that qdisc tcm_handle is not expected
1223 				 *   to exist, so that we choose create/graft too.
1224 				 *
1225 				 *   3. The last case is when no flags are set.
1226 				 *   Alas, it is sort of hole in API, we
1227 				 *   cannot decide what to do unambiguously.
1228 				 *   For now we select create/graft, if
1229 				 *   user gave KIND, which does not match existing.
1230 				 */
1231 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1232 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1233 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1234 				     (tca[TCA_KIND] &&
1235 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1236 					goto create_n_graft;
1237 			}
1238 		}
1239 	} else {
1240 		if (!tcm->tcm_handle)
1241 			return -EINVAL;
1242 		q = qdisc_lookup(dev, tcm->tcm_handle);
1243 	}
1244 
1245 	/* Change qdisc parameters */
1246 	if (q == NULL)
1247 		return -ENOENT;
1248 	if (n->nlmsg_flags & NLM_F_EXCL)
1249 		return -EEXIST;
1250 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1251 		return -EINVAL;
1252 	err = qdisc_change(q, tca);
1253 	if (err == 0)
1254 		qdisc_notify(net, skb, n, clid, NULL, q);
1255 	return err;
1256 
1257 create_n_graft:
1258 	if (!(n->nlmsg_flags & NLM_F_CREATE))
1259 		return -ENOENT;
1260 	if (clid == TC_H_INGRESS) {
1261 		if (dev_ingress_queue(dev))
1262 			q = qdisc_create(dev, dev_ingress_queue(dev), p,
1263 					 tcm->tcm_parent, tcm->tcm_parent,
1264 					 tca, &err);
1265 		else
1266 			err = -ENOENT;
1267 	} else {
1268 		struct netdev_queue *dev_queue;
1269 
1270 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1271 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1272 		else if (p)
1273 			dev_queue = p->dev_queue;
1274 		else
1275 			dev_queue = netdev_get_tx_queue(dev, 0);
1276 
1277 		q = qdisc_create(dev, dev_queue, p,
1278 				 tcm->tcm_parent, tcm->tcm_handle,
1279 				 tca, &err);
1280 	}
1281 	if (q == NULL) {
1282 		if (err == -EAGAIN)
1283 			goto replay;
1284 		return err;
1285 	}
1286 
1287 graft:
1288 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1289 	if (err) {
1290 		if (q)
1291 			qdisc_destroy(q);
1292 		return err;
1293 	}
1294 
1295 	return 0;
1296 }
1297 
1298 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1299 			 u32 portid, u32 seq, u16 flags, int event)
1300 {
1301 	struct tcmsg *tcm;
1302 	struct nlmsghdr  *nlh;
1303 	unsigned char *b = skb_tail_pointer(skb);
1304 	struct gnet_dump d;
1305 	struct qdisc_size_table *stab;
1306 
1307 	cond_resched();
1308 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1309 	if (!nlh)
1310 		goto out_nlmsg_trim;
1311 	tcm = nlmsg_data(nlh);
1312 	tcm->tcm_family = AF_UNSPEC;
1313 	tcm->tcm__pad1 = 0;
1314 	tcm->tcm__pad2 = 0;
1315 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1316 	tcm->tcm_parent = clid;
1317 	tcm->tcm_handle = q->handle;
1318 	tcm->tcm_info = atomic_read(&q->refcnt);
1319 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1320 		goto nla_put_failure;
1321 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
1322 		goto nla_put_failure;
1323 	q->qstats.qlen = q->q.qlen;
1324 
1325 	stab = rtnl_dereference(q->stab);
1326 	if (stab && qdisc_dump_stab(skb, stab) < 0)
1327 		goto nla_put_failure;
1328 
1329 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1330 					 qdisc_root_sleeping_lock(q), &d) < 0)
1331 		goto nla_put_failure;
1332 
1333 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1334 		goto nla_put_failure;
1335 
1336 	if (gnet_stats_copy_basic(&d, &q->bstats) < 0 ||
1337 	    gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1338 	    gnet_stats_copy_queue(&d, &q->qstats) < 0)
1339 		goto nla_put_failure;
1340 
1341 	if (gnet_stats_finish_copy(&d) < 0)
1342 		goto nla_put_failure;
1343 
1344 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1345 	return skb->len;
1346 
1347 out_nlmsg_trim:
1348 nla_put_failure:
1349 	nlmsg_trim(skb, b);
1350 	return -1;
1351 }
1352 
1353 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1354 {
1355 	return (q->flags & TCQ_F_BUILTIN) ? true : false;
1356 }
1357 
1358 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1359 			struct nlmsghdr *n, u32 clid,
1360 			struct Qdisc *old, struct Qdisc *new)
1361 {
1362 	struct sk_buff *skb;
1363 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1364 
1365 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1366 	if (!skb)
1367 		return -ENOBUFS;
1368 
1369 	if (old && !tc_qdisc_dump_ignore(old)) {
1370 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1371 				  0, RTM_DELQDISC) < 0)
1372 			goto err_out;
1373 	}
1374 	if (new && !tc_qdisc_dump_ignore(new)) {
1375 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1376 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1377 			goto err_out;
1378 	}
1379 
1380 	if (skb->len)
1381 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1382 				      n->nlmsg_flags & NLM_F_ECHO);
1383 
1384 err_out:
1385 	kfree_skb(skb);
1386 	return -EINVAL;
1387 }
1388 
1389 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1390 			      struct netlink_callback *cb,
1391 			      int *q_idx_p, int s_q_idx)
1392 {
1393 	int ret = 0, q_idx = *q_idx_p;
1394 	struct Qdisc *q;
1395 
1396 	if (!root)
1397 		return 0;
1398 
1399 	q = root;
1400 	if (q_idx < s_q_idx) {
1401 		q_idx++;
1402 	} else {
1403 		if (!tc_qdisc_dump_ignore(q) &&
1404 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1405 				  cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1406 			goto done;
1407 		q_idx++;
1408 	}
1409 	list_for_each_entry(q, &root->list, list) {
1410 		if (q_idx < s_q_idx) {
1411 			q_idx++;
1412 			continue;
1413 		}
1414 		if (!tc_qdisc_dump_ignore(q) &&
1415 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1416 				  cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1417 			goto done;
1418 		q_idx++;
1419 	}
1420 
1421 out:
1422 	*q_idx_p = q_idx;
1423 	return ret;
1424 done:
1425 	ret = -1;
1426 	goto out;
1427 }
1428 
1429 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1430 {
1431 	struct net *net = sock_net(skb->sk);
1432 	int idx, q_idx;
1433 	int s_idx, s_q_idx;
1434 	struct net_device *dev;
1435 
1436 	s_idx = cb->args[0];
1437 	s_q_idx = q_idx = cb->args[1];
1438 
1439 	idx = 0;
1440 	ASSERT_RTNL();
1441 	for_each_netdev(net, dev) {
1442 		struct netdev_queue *dev_queue;
1443 
1444 		if (idx < s_idx)
1445 			goto cont;
1446 		if (idx > s_idx)
1447 			s_q_idx = 0;
1448 		q_idx = 0;
1449 
1450 		if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1451 			goto done;
1452 
1453 		dev_queue = dev_ingress_queue(dev);
1454 		if (dev_queue &&
1455 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1456 				       &q_idx, s_q_idx) < 0)
1457 			goto done;
1458 
1459 cont:
1460 		idx++;
1461 	}
1462 
1463 done:
1464 	cb->args[0] = idx;
1465 	cb->args[1] = q_idx;
1466 
1467 	return skb->len;
1468 }
1469 
1470 
1471 
1472 /************************************************
1473  *	Traffic classes manipulation.		*
1474  ************************************************/
1475 
1476 
1477 
1478 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1479 {
1480 	struct net *net = sock_net(skb->sk);
1481 	struct tcmsg *tcm = nlmsg_data(n);
1482 	struct nlattr *tca[TCA_MAX + 1];
1483 	struct net_device *dev;
1484 	struct Qdisc *q = NULL;
1485 	const struct Qdisc_class_ops *cops;
1486 	unsigned long cl = 0;
1487 	unsigned long new_cl;
1488 	u32 portid;
1489 	u32 clid;
1490 	u32 qid;
1491 	int err;
1492 
1493 	if ((n->nlmsg_type != RTM_GETTCLASS) && !capable(CAP_NET_ADMIN))
1494 		return -EPERM;
1495 
1496 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1497 	if (err < 0)
1498 		return err;
1499 
1500 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1501 	if (!dev)
1502 		return -ENODEV;
1503 
1504 	/*
1505 	   parent == TC_H_UNSPEC - unspecified parent.
1506 	   parent == TC_H_ROOT   - class is root, which has no parent.
1507 	   parent == X:0	 - parent is root class.
1508 	   parent == X:Y	 - parent is a node in hierarchy.
1509 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
1510 
1511 	   handle == 0:0	 - generate handle from kernel pool.
1512 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
1513 	   handle == X:Y	 - clear.
1514 	   handle == X:0	 - root class.
1515 	 */
1516 
1517 	/* Step 1. Determine qdisc handle X:0 */
1518 
1519 	portid = tcm->tcm_parent;
1520 	clid = tcm->tcm_handle;
1521 	qid = TC_H_MAJ(clid);
1522 
1523 	if (portid != TC_H_ROOT) {
1524 		u32 qid1 = TC_H_MAJ(portid);
1525 
1526 		if (qid && qid1) {
1527 			/* If both majors are known, they must be identical. */
1528 			if (qid != qid1)
1529 				return -EINVAL;
1530 		} else if (qid1) {
1531 			qid = qid1;
1532 		} else if (qid == 0)
1533 			qid = dev->qdisc->handle;
1534 
1535 		/* Now qid is genuine qdisc handle consistent
1536 		 * both with parent and child.
1537 		 *
1538 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1539 		 */
1540 		if (portid)
1541 			portid = TC_H_MAKE(qid, portid);
1542 	} else {
1543 		if (qid == 0)
1544 			qid = dev->qdisc->handle;
1545 	}
1546 
1547 	/* OK. Locate qdisc */
1548 	q = qdisc_lookup(dev, qid);
1549 	if (!q)
1550 		return -ENOENT;
1551 
1552 	/* An check that it supports classes */
1553 	cops = q->ops->cl_ops;
1554 	if (cops == NULL)
1555 		return -EINVAL;
1556 
1557 	/* Now try to get class */
1558 	if (clid == 0) {
1559 		if (portid == TC_H_ROOT)
1560 			clid = qid;
1561 	} else
1562 		clid = TC_H_MAKE(qid, clid);
1563 
1564 	if (clid)
1565 		cl = cops->get(q, clid);
1566 
1567 	if (cl == 0) {
1568 		err = -ENOENT;
1569 		if (n->nlmsg_type != RTM_NEWTCLASS ||
1570 		    !(n->nlmsg_flags & NLM_F_CREATE))
1571 			goto out;
1572 	} else {
1573 		switch (n->nlmsg_type) {
1574 		case RTM_NEWTCLASS:
1575 			err = -EEXIST;
1576 			if (n->nlmsg_flags & NLM_F_EXCL)
1577 				goto out;
1578 			break;
1579 		case RTM_DELTCLASS:
1580 			err = -EOPNOTSUPP;
1581 			if (cops->delete)
1582 				err = cops->delete(q, cl);
1583 			if (err == 0)
1584 				tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1585 			goto out;
1586 		case RTM_GETTCLASS:
1587 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1588 			goto out;
1589 		default:
1590 			err = -EINVAL;
1591 			goto out;
1592 		}
1593 	}
1594 
1595 	new_cl = cl;
1596 	err = -EOPNOTSUPP;
1597 	if (cops->change)
1598 		err = cops->change(q, clid, portid, tca, &new_cl);
1599 	if (err == 0)
1600 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1601 
1602 out:
1603 	if (cl)
1604 		cops->put(q, cl);
1605 
1606 	return err;
1607 }
1608 
1609 
1610 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1611 			  unsigned long cl,
1612 			  u32 portid, u32 seq, u16 flags, int event)
1613 {
1614 	struct tcmsg *tcm;
1615 	struct nlmsghdr  *nlh;
1616 	unsigned char *b = skb_tail_pointer(skb);
1617 	struct gnet_dump d;
1618 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1619 
1620 	cond_resched();
1621 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1622 	if (!nlh)
1623 		goto out_nlmsg_trim;
1624 	tcm = nlmsg_data(nlh);
1625 	tcm->tcm_family = AF_UNSPEC;
1626 	tcm->tcm__pad1 = 0;
1627 	tcm->tcm__pad2 = 0;
1628 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1629 	tcm->tcm_parent = q->handle;
1630 	tcm->tcm_handle = q->handle;
1631 	tcm->tcm_info = 0;
1632 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1633 		goto nla_put_failure;
1634 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1635 		goto nla_put_failure;
1636 
1637 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1638 					 qdisc_root_sleeping_lock(q), &d) < 0)
1639 		goto nla_put_failure;
1640 
1641 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1642 		goto nla_put_failure;
1643 
1644 	if (gnet_stats_finish_copy(&d) < 0)
1645 		goto nla_put_failure;
1646 
1647 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1648 	return skb->len;
1649 
1650 out_nlmsg_trim:
1651 nla_put_failure:
1652 	nlmsg_trim(skb, b);
1653 	return -1;
1654 }
1655 
1656 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1657 			 struct nlmsghdr *n, struct Qdisc *q,
1658 			 unsigned long cl, int event)
1659 {
1660 	struct sk_buff *skb;
1661 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1662 
1663 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1664 	if (!skb)
1665 		return -ENOBUFS;
1666 
1667 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1668 		kfree_skb(skb);
1669 		return -EINVAL;
1670 	}
1671 
1672 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1673 			      n->nlmsg_flags & NLM_F_ECHO);
1674 }
1675 
1676 struct qdisc_dump_args {
1677 	struct qdisc_walker	w;
1678 	struct sk_buff		*skb;
1679 	struct netlink_callback	*cb;
1680 };
1681 
1682 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1683 {
1684 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1685 
1686 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1687 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1688 }
1689 
1690 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1691 				struct tcmsg *tcm, struct netlink_callback *cb,
1692 				int *t_p, int s_t)
1693 {
1694 	struct qdisc_dump_args arg;
1695 
1696 	if (tc_qdisc_dump_ignore(q) ||
1697 	    *t_p < s_t || !q->ops->cl_ops ||
1698 	    (tcm->tcm_parent &&
1699 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1700 		(*t_p)++;
1701 		return 0;
1702 	}
1703 	if (*t_p > s_t)
1704 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1705 	arg.w.fn = qdisc_class_dump;
1706 	arg.skb = skb;
1707 	arg.cb = cb;
1708 	arg.w.stop  = 0;
1709 	arg.w.skip = cb->args[1];
1710 	arg.w.count = 0;
1711 	q->ops->cl_ops->walk(q, &arg.w);
1712 	cb->args[1] = arg.w.count;
1713 	if (arg.w.stop)
1714 		return -1;
1715 	(*t_p)++;
1716 	return 0;
1717 }
1718 
1719 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1720 			       struct tcmsg *tcm, struct netlink_callback *cb,
1721 			       int *t_p, int s_t)
1722 {
1723 	struct Qdisc *q;
1724 
1725 	if (!root)
1726 		return 0;
1727 
1728 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1729 		return -1;
1730 
1731 	list_for_each_entry(q, &root->list, list) {
1732 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1733 			return -1;
1734 	}
1735 
1736 	return 0;
1737 }
1738 
1739 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1740 {
1741 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
1742 	struct net *net = sock_net(skb->sk);
1743 	struct netdev_queue *dev_queue;
1744 	struct net_device *dev;
1745 	int t, s_t;
1746 
1747 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1748 		return 0;
1749 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
1750 	if (!dev)
1751 		return 0;
1752 
1753 	s_t = cb->args[0];
1754 	t = 0;
1755 
1756 	if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1757 		goto done;
1758 
1759 	dev_queue = dev_ingress_queue(dev);
1760 	if (dev_queue &&
1761 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1762 				&t, s_t) < 0)
1763 		goto done;
1764 
1765 done:
1766 	cb->args[0] = t;
1767 
1768 	dev_put(dev);
1769 	return skb->len;
1770 }
1771 
1772 /* Main classifier routine: scans classifier chain attached
1773  * to this qdisc, (optionally) tests for protocol and asks
1774  * specific classifiers.
1775  */
1776 int tc_classify_compat(struct sk_buff *skb, const struct tcf_proto *tp,
1777 		       struct tcf_result *res)
1778 {
1779 	__be16 protocol = skb->protocol;
1780 	int err;
1781 
1782 	for (; tp; tp = tp->next) {
1783 		if (tp->protocol != protocol &&
1784 		    tp->protocol != htons(ETH_P_ALL))
1785 			continue;
1786 		err = tp->classify(skb, tp, res);
1787 
1788 		if (err >= 0) {
1789 #ifdef CONFIG_NET_CLS_ACT
1790 			if (err != TC_ACT_RECLASSIFY && skb->tc_verd)
1791 				skb->tc_verd = SET_TC_VERD(skb->tc_verd, 0);
1792 #endif
1793 			return err;
1794 		}
1795 	}
1796 	return -1;
1797 }
1798 EXPORT_SYMBOL(tc_classify_compat);
1799 
1800 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1801 		struct tcf_result *res)
1802 {
1803 	int err = 0;
1804 #ifdef CONFIG_NET_CLS_ACT
1805 	const struct tcf_proto *otp = tp;
1806 reclassify:
1807 #endif
1808 
1809 	err = tc_classify_compat(skb, tp, res);
1810 #ifdef CONFIG_NET_CLS_ACT
1811 	if (err == TC_ACT_RECLASSIFY) {
1812 		u32 verd = G_TC_VERD(skb->tc_verd);
1813 		tp = otp;
1814 
1815 		if (verd++ >= MAX_REC_LOOP) {
1816 			net_notice_ratelimited("%s: packet reclassify loop rule prio %u protocol %02x\n",
1817 					       tp->q->ops->id,
1818 					       tp->prio & 0xffff,
1819 					       ntohs(tp->protocol));
1820 			return TC_ACT_SHOT;
1821 		}
1822 		skb->tc_verd = SET_TC_VERD(skb->tc_verd, verd);
1823 		goto reclassify;
1824 	}
1825 #endif
1826 	return err;
1827 }
1828 EXPORT_SYMBOL(tc_classify);
1829 
1830 void tcf_destroy(struct tcf_proto *tp)
1831 {
1832 	tp->ops->destroy(tp);
1833 	module_put(tp->ops->owner);
1834 	kfree(tp);
1835 }
1836 
1837 void tcf_destroy_chain(struct tcf_proto **fl)
1838 {
1839 	struct tcf_proto *tp;
1840 
1841 	while ((tp = *fl) != NULL) {
1842 		*fl = tp->next;
1843 		tcf_destroy(tp);
1844 	}
1845 }
1846 EXPORT_SYMBOL(tcf_destroy_chain);
1847 
1848 #ifdef CONFIG_PROC_FS
1849 static int psched_show(struct seq_file *seq, void *v)
1850 {
1851 	struct timespec ts;
1852 
1853 	hrtimer_get_res(CLOCK_MONOTONIC, &ts);
1854 	seq_printf(seq, "%08x %08x %08x %08x\n",
1855 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1856 		   1000000,
1857 		   (u32)NSEC_PER_SEC/(u32)ktime_to_ns(timespec_to_ktime(ts)));
1858 
1859 	return 0;
1860 }
1861 
1862 static int psched_open(struct inode *inode, struct file *file)
1863 {
1864 	return single_open(file, psched_show, NULL);
1865 }
1866 
1867 static const struct file_operations psched_fops = {
1868 	.owner = THIS_MODULE,
1869 	.open = psched_open,
1870 	.read  = seq_read,
1871 	.llseek = seq_lseek,
1872 	.release = single_release,
1873 };
1874 
1875 static int __net_init psched_net_init(struct net *net)
1876 {
1877 	struct proc_dir_entry *e;
1878 
1879 	e = proc_create("psched", 0, net->proc_net, &psched_fops);
1880 	if (e == NULL)
1881 		return -ENOMEM;
1882 
1883 	return 0;
1884 }
1885 
1886 static void __net_exit psched_net_exit(struct net *net)
1887 {
1888 	remove_proc_entry("psched", net->proc_net);
1889 }
1890 #else
1891 static int __net_init psched_net_init(struct net *net)
1892 {
1893 	return 0;
1894 }
1895 
1896 static void __net_exit psched_net_exit(struct net *net)
1897 {
1898 }
1899 #endif
1900 
1901 static struct pernet_operations psched_net_ops = {
1902 	.init = psched_net_init,
1903 	.exit = psched_net_exit,
1904 };
1905 
1906 static int __init pktsched_init(void)
1907 {
1908 	int err;
1909 
1910 	err = register_pernet_subsys(&psched_net_ops);
1911 	if (err) {
1912 		pr_err("pktsched_init: "
1913 		       "cannot initialize per netns operations\n");
1914 		return err;
1915 	}
1916 
1917 	register_qdisc(&pfifo_fast_ops);
1918 	register_qdisc(&pfifo_qdisc_ops);
1919 	register_qdisc(&bfifo_qdisc_ops);
1920 	register_qdisc(&pfifo_head_drop_qdisc_ops);
1921 	register_qdisc(&mq_qdisc_ops);
1922 
1923 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1924 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1925 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1926 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1927 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1928 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1929 
1930 	return 0;
1931 }
1932 
1933 subsys_initcall(pktsched_init);
1934