xref: /openbmc/linux/net/sched/sch_api.c (revision 9cfc5c90)
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(const 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 	rcu_read_lock();
582 	qdisc_unthrottled(wd->qdisc);
583 	__netif_schedule(qdisc_root(wd->qdisc));
584 	rcu_read_unlock();
585 
586 	return HRTIMER_NORESTART;
587 }
588 
589 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
590 {
591 	hrtimer_init(&wd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
592 	wd->timer.function = qdisc_watchdog;
593 	wd->qdisc = qdisc;
594 }
595 EXPORT_SYMBOL(qdisc_watchdog_init);
596 
597 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires, bool throttle)
598 {
599 	if (test_bit(__QDISC_STATE_DEACTIVATED,
600 		     &qdisc_root_sleeping(wd->qdisc)->state))
601 		return;
602 
603 	if (throttle)
604 		qdisc_throttled(wd->qdisc);
605 
606 	hrtimer_start(&wd->timer,
607 		      ns_to_ktime(expires),
608 		      HRTIMER_MODE_ABS_PINNED);
609 }
610 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
611 
612 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
613 {
614 	hrtimer_cancel(&wd->timer);
615 	qdisc_unthrottled(wd->qdisc);
616 }
617 EXPORT_SYMBOL(qdisc_watchdog_cancel);
618 
619 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
620 {
621 	unsigned int size = n * sizeof(struct hlist_head), i;
622 	struct hlist_head *h;
623 
624 	if (size <= PAGE_SIZE)
625 		h = kmalloc(size, GFP_KERNEL);
626 	else
627 		h = (struct hlist_head *)
628 			__get_free_pages(GFP_KERNEL, get_order(size));
629 
630 	if (h != NULL) {
631 		for (i = 0; i < n; i++)
632 			INIT_HLIST_HEAD(&h[i]);
633 	}
634 	return h;
635 }
636 
637 static void qdisc_class_hash_free(struct hlist_head *h, unsigned int n)
638 {
639 	unsigned int size = n * sizeof(struct hlist_head);
640 
641 	if (size <= PAGE_SIZE)
642 		kfree(h);
643 	else
644 		free_pages((unsigned long)h, get_order(size));
645 }
646 
647 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
648 {
649 	struct Qdisc_class_common *cl;
650 	struct hlist_node *next;
651 	struct hlist_head *nhash, *ohash;
652 	unsigned int nsize, nmask, osize;
653 	unsigned int i, h;
654 
655 	/* Rehash when load factor exceeds 0.75 */
656 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
657 		return;
658 	nsize = clhash->hashsize * 2;
659 	nmask = nsize - 1;
660 	nhash = qdisc_class_hash_alloc(nsize);
661 	if (nhash == NULL)
662 		return;
663 
664 	ohash = clhash->hash;
665 	osize = clhash->hashsize;
666 
667 	sch_tree_lock(sch);
668 	for (i = 0; i < osize; i++) {
669 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
670 			h = qdisc_class_hash(cl->classid, nmask);
671 			hlist_add_head(&cl->hnode, &nhash[h]);
672 		}
673 	}
674 	clhash->hash     = nhash;
675 	clhash->hashsize = nsize;
676 	clhash->hashmask = nmask;
677 	sch_tree_unlock(sch);
678 
679 	qdisc_class_hash_free(ohash, osize);
680 }
681 EXPORT_SYMBOL(qdisc_class_hash_grow);
682 
683 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
684 {
685 	unsigned int size = 4;
686 
687 	clhash->hash = qdisc_class_hash_alloc(size);
688 	if (clhash->hash == NULL)
689 		return -ENOMEM;
690 	clhash->hashsize  = size;
691 	clhash->hashmask  = size - 1;
692 	clhash->hashelems = 0;
693 	return 0;
694 }
695 EXPORT_SYMBOL(qdisc_class_hash_init);
696 
697 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
698 {
699 	qdisc_class_hash_free(clhash->hash, clhash->hashsize);
700 }
701 EXPORT_SYMBOL(qdisc_class_hash_destroy);
702 
703 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
704 			     struct Qdisc_class_common *cl)
705 {
706 	unsigned int h;
707 
708 	INIT_HLIST_NODE(&cl->hnode);
709 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
710 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
711 	clhash->hashelems++;
712 }
713 EXPORT_SYMBOL(qdisc_class_hash_insert);
714 
715 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
716 			     struct Qdisc_class_common *cl)
717 {
718 	hlist_del(&cl->hnode);
719 	clhash->hashelems--;
720 }
721 EXPORT_SYMBOL(qdisc_class_hash_remove);
722 
723 /* Allocate an unique handle from space managed by kernel
724  * Possible range is [8000-FFFF]:0000 (0x8000 values)
725  */
726 static u32 qdisc_alloc_handle(struct net_device *dev)
727 {
728 	int i = 0x8000;
729 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
730 
731 	do {
732 		autohandle += TC_H_MAKE(0x10000U, 0);
733 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
734 			autohandle = TC_H_MAKE(0x80000000U, 0);
735 		if (!qdisc_lookup(dev, autohandle))
736 			return autohandle;
737 		cond_resched();
738 	} while	(--i > 0);
739 
740 	return 0;
741 }
742 
743 void qdisc_tree_decrease_qlen(struct Qdisc *sch, unsigned int n)
744 {
745 	const struct Qdisc_class_ops *cops;
746 	unsigned long cl;
747 	u32 parentid;
748 	int drops;
749 
750 	if (n == 0)
751 		return;
752 	drops = max_t(int, n, 0);
753 	while ((parentid = sch->parent)) {
754 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
755 			return;
756 
757 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
758 		if (sch == NULL) {
759 			WARN_ON(parentid != TC_H_ROOT);
760 			return;
761 		}
762 		cops = sch->ops->cl_ops;
763 		if (cops->qlen_notify) {
764 			cl = cops->get(sch, parentid);
765 			cops->qlen_notify(sch, cl);
766 			cops->put(sch, cl);
767 		}
768 		sch->q.qlen -= n;
769 		__qdisc_qstats_drop(sch, drops);
770 	}
771 }
772 EXPORT_SYMBOL(qdisc_tree_decrease_qlen);
773 
774 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
775 			       struct nlmsghdr *n, u32 clid,
776 			       struct Qdisc *old, struct Qdisc *new)
777 {
778 	if (new || old)
779 		qdisc_notify(net, skb, n, clid, old, new);
780 
781 	if (old)
782 		qdisc_destroy(old);
783 }
784 
785 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
786  * to device "dev".
787  *
788  * When appropriate send a netlink notification using 'skb'
789  * and "n".
790  *
791  * On success, destroy old qdisc.
792  */
793 
794 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
795 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
796 		       struct Qdisc *new, struct Qdisc *old)
797 {
798 	struct Qdisc *q = old;
799 	struct net *net = dev_net(dev);
800 	int err = 0;
801 
802 	if (parent == NULL) {
803 		unsigned int i, num_q, ingress;
804 
805 		ingress = 0;
806 		num_q = dev->num_tx_queues;
807 		if ((q && q->flags & TCQ_F_INGRESS) ||
808 		    (new && new->flags & TCQ_F_INGRESS)) {
809 			num_q = 1;
810 			ingress = 1;
811 			if (!dev_ingress_queue(dev))
812 				return -ENOENT;
813 		}
814 
815 		if (dev->flags & IFF_UP)
816 			dev_deactivate(dev);
817 
818 		if (new && new->ops->attach)
819 			goto skip;
820 
821 		for (i = 0; i < num_q; i++) {
822 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
823 
824 			if (!ingress)
825 				dev_queue = netdev_get_tx_queue(dev, i);
826 
827 			old = dev_graft_qdisc(dev_queue, new);
828 			if (new && i > 0)
829 				atomic_inc(&new->refcnt);
830 
831 			if (!ingress)
832 				qdisc_destroy(old);
833 		}
834 
835 skip:
836 		if (!ingress) {
837 			notify_and_destroy(net, skb, n, classid,
838 					   dev->qdisc, new);
839 			if (new && !new->ops->attach)
840 				atomic_inc(&new->refcnt);
841 			dev->qdisc = new ? : &noop_qdisc;
842 
843 			if (new && new->ops->attach)
844 				new->ops->attach(new);
845 		} else {
846 			notify_and_destroy(net, skb, n, classid, old, new);
847 		}
848 
849 		if (dev->flags & IFF_UP)
850 			dev_activate(dev);
851 	} else {
852 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
853 
854 		err = -EOPNOTSUPP;
855 		if (cops && cops->graft) {
856 			unsigned long cl = cops->get(parent, classid);
857 			if (cl) {
858 				err = cops->graft(parent, cl, new, &old);
859 				cops->put(parent, cl);
860 			} else
861 				err = -ENOENT;
862 		}
863 		if (!err)
864 			notify_and_destroy(net, skb, n, classid, old, new);
865 	}
866 	return err;
867 }
868 
869 /* lockdep annotation is needed for ingress; egress gets it only for name */
870 static struct lock_class_key qdisc_tx_lock;
871 static struct lock_class_key qdisc_rx_lock;
872 
873 /*
874    Allocate and initialize new qdisc.
875 
876    Parameters are passed via opt.
877  */
878 
879 static struct Qdisc *
880 qdisc_create(struct net_device *dev, struct netdev_queue *dev_queue,
881 	     struct Qdisc *p, u32 parent, u32 handle,
882 	     struct nlattr **tca, int *errp)
883 {
884 	int err;
885 	struct nlattr *kind = tca[TCA_KIND];
886 	struct Qdisc *sch;
887 	struct Qdisc_ops *ops;
888 	struct qdisc_size_table *stab;
889 
890 	ops = qdisc_lookup_ops(kind);
891 #ifdef CONFIG_MODULES
892 	if (ops == NULL && kind != NULL) {
893 		char name[IFNAMSIZ];
894 		if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
895 			/* We dropped the RTNL semaphore in order to
896 			 * perform the module load.  So, even if we
897 			 * succeeded in loading the module we have to
898 			 * tell the caller to replay the request.  We
899 			 * indicate this using -EAGAIN.
900 			 * We replay the request because the device may
901 			 * go away in the mean time.
902 			 */
903 			rtnl_unlock();
904 			request_module("sch_%s", name);
905 			rtnl_lock();
906 			ops = qdisc_lookup_ops(kind);
907 			if (ops != NULL) {
908 				/* We will try again qdisc_lookup_ops,
909 				 * so don't keep a reference.
910 				 */
911 				module_put(ops->owner);
912 				err = -EAGAIN;
913 				goto err_out;
914 			}
915 		}
916 	}
917 #endif
918 
919 	err = -ENOENT;
920 	if (ops == NULL)
921 		goto err_out;
922 
923 	sch = qdisc_alloc(dev_queue, ops);
924 	if (IS_ERR(sch)) {
925 		err = PTR_ERR(sch);
926 		goto err_out2;
927 	}
928 
929 	sch->parent = parent;
930 
931 	if (handle == TC_H_INGRESS) {
932 		sch->flags |= TCQ_F_INGRESS;
933 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
934 		lockdep_set_class(qdisc_lock(sch), &qdisc_rx_lock);
935 	} else {
936 		if (handle == 0) {
937 			handle = qdisc_alloc_handle(dev);
938 			err = -ENOMEM;
939 			if (handle == 0)
940 				goto err_out3;
941 		}
942 		lockdep_set_class(qdisc_lock(sch), &qdisc_tx_lock);
943 		if (!netif_is_multiqueue(dev))
944 			sch->flags |= TCQ_F_ONETXQUEUE;
945 	}
946 
947 	sch->handle = handle;
948 
949 	if (!ops->init || (err = ops->init(sch, tca[TCA_OPTIONS])) == 0) {
950 		if (qdisc_is_percpu_stats(sch)) {
951 			sch->cpu_bstats =
952 				netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu);
953 			if (!sch->cpu_bstats)
954 				goto err_out4;
955 
956 			sch->cpu_qstats = alloc_percpu(struct gnet_stats_queue);
957 			if (!sch->cpu_qstats)
958 				goto err_out4;
959 		}
960 
961 		if (tca[TCA_STAB]) {
962 			stab = qdisc_get_stab(tca[TCA_STAB]);
963 			if (IS_ERR(stab)) {
964 				err = PTR_ERR(stab);
965 				goto err_out4;
966 			}
967 			rcu_assign_pointer(sch->stab, stab);
968 		}
969 		if (tca[TCA_RATE]) {
970 			spinlock_t *root_lock;
971 
972 			err = -EOPNOTSUPP;
973 			if (sch->flags & TCQ_F_MQROOT)
974 				goto err_out4;
975 
976 			if ((sch->parent != TC_H_ROOT) &&
977 			    !(sch->flags & TCQ_F_INGRESS) &&
978 			    (!p || !(p->flags & TCQ_F_MQROOT)))
979 				root_lock = qdisc_root_sleeping_lock(sch);
980 			else
981 				root_lock = qdisc_lock(sch);
982 
983 			err = gen_new_estimator(&sch->bstats,
984 						sch->cpu_bstats,
985 						&sch->rate_est,
986 						root_lock,
987 						tca[TCA_RATE]);
988 			if (err)
989 				goto err_out4;
990 		}
991 
992 		qdisc_list_add(sch);
993 
994 		return sch;
995 	}
996 err_out3:
997 	dev_put(dev);
998 	kfree((char *) sch - sch->padded);
999 err_out2:
1000 	module_put(ops->owner);
1001 err_out:
1002 	*errp = err;
1003 	return NULL;
1004 
1005 err_out4:
1006 	free_percpu(sch->cpu_bstats);
1007 	free_percpu(sch->cpu_qstats);
1008 	/*
1009 	 * Any broken qdiscs that would require a ops->reset() here?
1010 	 * The qdisc was never in action so it shouldn't be necessary.
1011 	 */
1012 	qdisc_put_stab(rtnl_dereference(sch->stab));
1013 	if (ops->destroy)
1014 		ops->destroy(sch);
1015 	goto err_out3;
1016 }
1017 
1018 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca)
1019 {
1020 	struct qdisc_size_table *ostab, *stab = NULL;
1021 	int err = 0;
1022 
1023 	if (tca[TCA_OPTIONS]) {
1024 		if (sch->ops->change == NULL)
1025 			return -EINVAL;
1026 		err = sch->ops->change(sch, tca[TCA_OPTIONS]);
1027 		if (err)
1028 			return err;
1029 	}
1030 
1031 	if (tca[TCA_STAB]) {
1032 		stab = qdisc_get_stab(tca[TCA_STAB]);
1033 		if (IS_ERR(stab))
1034 			return PTR_ERR(stab);
1035 	}
1036 
1037 	ostab = rtnl_dereference(sch->stab);
1038 	rcu_assign_pointer(sch->stab, stab);
1039 	qdisc_put_stab(ostab);
1040 
1041 	if (tca[TCA_RATE]) {
1042 		/* NB: ignores errors from replace_estimator
1043 		   because change can't be undone. */
1044 		if (sch->flags & TCQ_F_MQROOT)
1045 			goto out;
1046 		gen_replace_estimator(&sch->bstats,
1047 				      sch->cpu_bstats,
1048 				      &sch->rate_est,
1049 				      qdisc_root_sleeping_lock(sch),
1050 				      tca[TCA_RATE]);
1051 	}
1052 out:
1053 	return 0;
1054 }
1055 
1056 struct check_loop_arg {
1057 	struct qdisc_walker	w;
1058 	struct Qdisc		*p;
1059 	int			depth;
1060 };
1061 
1062 static int check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w);
1063 
1064 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1065 {
1066 	struct check_loop_arg	arg;
1067 
1068 	if (q->ops->cl_ops == NULL)
1069 		return 0;
1070 
1071 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1072 	arg.w.fn = check_loop_fn;
1073 	arg.depth = depth;
1074 	arg.p = p;
1075 	q->ops->cl_ops->walk(q, &arg.w);
1076 	return arg.w.stop ? -ELOOP : 0;
1077 }
1078 
1079 static int
1080 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1081 {
1082 	struct Qdisc *leaf;
1083 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1084 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1085 
1086 	leaf = cops->leaf(q, cl);
1087 	if (leaf) {
1088 		if (leaf == arg->p || arg->depth > 7)
1089 			return -ELOOP;
1090 		return check_loop(leaf, arg->p, arg->depth + 1);
1091 	}
1092 	return 0;
1093 }
1094 
1095 /*
1096  * Delete/get qdisc.
1097  */
1098 
1099 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1100 {
1101 	struct net *net = sock_net(skb->sk);
1102 	struct tcmsg *tcm = nlmsg_data(n);
1103 	struct nlattr *tca[TCA_MAX + 1];
1104 	struct net_device *dev;
1105 	u32 clid;
1106 	struct Qdisc *q = NULL;
1107 	struct Qdisc *p = NULL;
1108 	int err;
1109 
1110 	if ((n->nlmsg_type != RTM_GETQDISC) &&
1111 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1112 		return -EPERM;
1113 
1114 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1115 	if (err < 0)
1116 		return err;
1117 
1118 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1119 	if (!dev)
1120 		return -ENODEV;
1121 
1122 	clid = tcm->tcm_parent;
1123 	if (clid) {
1124 		if (clid != TC_H_ROOT) {
1125 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1126 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1127 				if (!p)
1128 					return -ENOENT;
1129 				q = qdisc_leaf(p, clid);
1130 			} else if (dev_ingress_queue(dev)) {
1131 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1132 			}
1133 		} else {
1134 			q = dev->qdisc;
1135 		}
1136 		if (!q)
1137 			return -ENOENT;
1138 
1139 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle)
1140 			return -EINVAL;
1141 	} else {
1142 		q = qdisc_lookup(dev, tcm->tcm_handle);
1143 		if (!q)
1144 			return -ENOENT;
1145 	}
1146 
1147 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1148 		return -EINVAL;
1149 
1150 	if (n->nlmsg_type == RTM_DELQDISC) {
1151 		if (!clid)
1152 			return -EINVAL;
1153 		if (q->handle == 0)
1154 			return -ENOENT;
1155 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q);
1156 		if (err != 0)
1157 			return err;
1158 	} else {
1159 		qdisc_notify(net, skb, n, clid, NULL, q);
1160 	}
1161 	return 0;
1162 }
1163 
1164 /*
1165  * Create/change qdisc.
1166  */
1167 
1168 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n)
1169 {
1170 	struct net *net = sock_net(skb->sk);
1171 	struct tcmsg *tcm;
1172 	struct nlattr *tca[TCA_MAX + 1];
1173 	struct net_device *dev;
1174 	u32 clid;
1175 	struct Qdisc *q, *p;
1176 	int err;
1177 
1178 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1179 		return -EPERM;
1180 
1181 replay:
1182 	/* Reinit, just in case something touches this. */
1183 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1184 	if (err < 0)
1185 		return err;
1186 
1187 	tcm = nlmsg_data(n);
1188 	clid = tcm->tcm_parent;
1189 	q = p = NULL;
1190 
1191 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1192 	if (!dev)
1193 		return -ENODEV;
1194 
1195 
1196 	if (clid) {
1197 		if (clid != TC_H_ROOT) {
1198 			if (clid != TC_H_INGRESS) {
1199 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1200 				if (!p)
1201 					return -ENOENT;
1202 				q = qdisc_leaf(p, clid);
1203 			} else if (dev_ingress_queue_create(dev)) {
1204 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1205 			}
1206 		} else {
1207 			q = dev->qdisc;
1208 		}
1209 
1210 		/* It may be default qdisc, ignore it */
1211 		if (q && q->handle == 0)
1212 			q = NULL;
1213 
1214 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1215 			if (tcm->tcm_handle) {
1216 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE))
1217 					return -EEXIST;
1218 				if (TC_H_MIN(tcm->tcm_handle))
1219 					return -EINVAL;
1220 				q = qdisc_lookup(dev, tcm->tcm_handle);
1221 				if (!q)
1222 					goto create_n_graft;
1223 				if (n->nlmsg_flags & NLM_F_EXCL)
1224 					return -EEXIST;
1225 				if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1226 					return -EINVAL;
1227 				if (q == p ||
1228 				    (p && check_loop(q, p, 0)))
1229 					return -ELOOP;
1230 				atomic_inc(&q->refcnt);
1231 				goto graft;
1232 			} else {
1233 				if (!q)
1234 					goto create_n_graft;
1235 
1236 				/* This magic test requires explanation.
1237 				 *
1238 				 *   We know, that some child q is already
1239 				 *   attached to this parent and have choice:
1240 				 *   either to change it or to create/graft new one.
1241 				 *
1242 				 *   1. We are allowed to create/graft only
1243 				 *   if CREATE and REPLACE flags are set.
1244 				 *
1245 				 *   2. If EXCL is set, requestor wanted to say,
1246 				 *   that qdisc tcm_handle is not expected
1247 				 *   to exist, so that we choose create/graft too.
1248 				 *
1249 				 *   3. The last case is when no flags are set.
1250 				 *   Alas, it is sort of hole in API, we
1251 				 *   cannot decide what to do unambiguously.
1252 				 *   For now we select create/graft, if
1253 				 *   user gave KIND, which does not match existing.
1254 				 */
1255 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1256 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1257 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1258 				     (tca[TCA_KIND] &&
1259 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1260 					goto create_n_graft;
1261 			}
1262 		}
1263 	} else {
1264 		if (!tcm->tcm_handle)
1265 			return -EINVAL;
1266 		q = qdisc_lookup(dev, tcm->tcm_handle);
1267 	}
1268 
1269 	/* Change qdisc parameters */
1270 	if (q == NULL)
1271 		return -ENOENT;
1272 	if (n->nlmsg_flags & NLM_F_EXCL)
1273 		return -EEXIST;
1274 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id))
1275 		return -EINVAL;
1276 	err = qdisc_change(q, tca);
1277 	if (err == 0)
1278 		qdisc_notify(net, skb, n, clid, NULL, q);
1279 	return err;
1280 
1281 create_n_graft:
1282 	if (!(n->nlmsg_flags & NLM_F_CREATE))
1283 		return -ENOENT;
1284 	if (clid == TC_H_INGRESS) {
1285 		if (dev_ingress_queue(dev))
1286 			q = qdisc_create(dev, dev_ingress_queue(dev), p,
1287 					 tcm->tcm_parent, tcm->tcm_parent,
1288 					 tca, &err);
1289 		else
1290 			err = -ENOENT;
1291 	} else {
1292 		struct netdev_queue *dev_queue;
1293 
1294 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1295 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1296 		else if (p)
1297 			dev_queue = p->dev_queue;
1298 		else
1299 			dev_queue = netdev_get_tx_queue(dev, 0);
1300 
1301 		q = qdisc_create(dev, dev_queue, p,
1302 				 tcm->tcm_parent, tcm->tcm_handle,
1303 				 tca, &err);
1304 	}
1305 	if (q == NULL) {
1306 		if (err == -EAGAIN)
1307 			goto replay;
1308 		return err;
1309 	}
1310 
1311 graft:
1312 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL);
1313 	if (err) {
1314 		if (q)
1315 			qdisc_destroy(q);
1316 		return err;
1317 	}
1318 
1319 	return 0;
1320 }
1321 
1322 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
1323 			 u32 portid, u32 seq, u16 flags, int event)
1324 {
1325 	struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
1326 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
1327 	struct tcmsg *tcm;
1328 	struct nlmsghdr  *nlh;
1329 	unsigned char *b = skb_tail_pointer(skb);
1330 	struct gnet_dump d;
1331 	struct qdisc_size_table *stab;
1332 	__u32 qlen;
1333 
1334 	cond_resched();
1335 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1336 	if (!nlh)
1337 		goto out_nlmsg_trim;
1338 	tcm = nlmsg_data(nlh);
1339 	tcm->tcm_family = AF_UNSPEC;
1340 	tcm->tcm__pad1 = 0;
1341 	tcm->tcm__pad2 = 0;
1342 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1343 	tcm->tcm_parent = clid;
1344 	tcm->tcm_handle = q->handle;
1345 	tcm->tcm_info = atomic_read(&q->refcnt);
1346 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1347 		goto nla_put_failure;
1348 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
1349 		goto nla_put_failure;
1350 	qlen = q->q.qlen;
1351 
1352 	stab = rtnl_dereference(q->stab);
1353 	if (stab && qdisc_dump_stab(skb, stab) < 0)
1354 		goto nla_put_failure;
1355 
1356 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1357 					 qdisc_root_sleeping_lock(q), &d) < 0)
1358 		goto nla_put_failure;
1359 
1360 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
1361 		goto nla_put_failure;
1362 
1363 	if (qdisc_is_percpu_stats(q)) {
1364 		cpu_bstats = q->cpu_bstats;
1365 		cpu_qstats = q->cpu_qstats;
1366 	}
1367 
1368 	if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats) < 0 ||
1369 	    gnet_stats_copy_rate_est(&d, &q->bstats, &q->rate_est) < 0 ||
1370 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
1371 		goto nla_put_failure;
1372 
1373 	if (gnet_stats_finish_copy(&d) < 0)
1374 		goto nla_put_failure;
1375 
1376 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1377 	return skb->len;
1378 
1379 out_nlmsg_trim:
1380 nla_put_failure:
1381 	nlmsg_trim(skb, b);
1382 	return -1;
1383 }
1384 
1385 static bool tc_qdisc_dump_ignore(struct Qdisc *q)
1386 {
1387 	return (q->flags & TCQ_F_BUILTIN) ? true : false;
1388 }
1389 
1390 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1391 			struct nlmsghdr *n, u32 clid,
1392 			struct Qdisc *old, struct Qdisc *new)
1393 {
1394 	struct sk_buff *skb;
1395 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1396 
1397 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1398 	if (!skb)
1399 		return -ENOBUFS;
1400 
1401 	if (old && !tc_qdisc_dump_ignore(old)) {
1402 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1403 				  0, RTM_DELQDISC) < 0)
1404 			goto err_out;
1405 	}
1406 	if (new && !tc_qdisc_dump_ignore(new)) {
1407 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1408 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1409 			goto err_out;
1410 	}
1411 
1412 	if (skb->len)
1413 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1414 				      n->nlmsg_flags & NLM_F_ECHO);
1415 
1416 err_out:
1417 	kfree_skb(skb);
1418 	return -EINVAL;
1419 }
1420 
1421 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1422 			      struct netlink_callback *cb,
1423 			      int *q_idx_p, int s_q_idx)
1424 {
1425 	int ret = 0, q_idx = *q_idx_p;
1426 	struct Qdisc *q;
1427 
1428 	if (!root)
1429 		return 0;
1430 
1431 	q = root;
1432 	if (q_idx < s_q_idx) {
1433 		q_idx++;
1434 	} else {
1435 		if (!tc_qdisc_dump_ignore(q) &&
1436 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1437 				  cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1438 			goto done;
1439 		q_idx++;
1440 	}
1441 	list_for_each_entry(q, &root->list, list) {
1442 		if (q_idx < s_q_idx) {
1443 			q_idx++;
1444 			continue;
1445 		}
1446 		if (!tc_qdisc_dump_ignore(q) &&
1447 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1448 				  cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWQDISC) <= 0)
1449 			goto done;
1450 		q_idx++;
1451 	}
1452 
1453 out:
1454 	*q_idx_p = q_idx;
1455 	return ret;
1456 done:
1457 	ret = -1;
1458 	goto out;
1459 }
1460 
1461 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1462 {
1463 	struct net *net = sock_net(skb->sk);
1464 	int idx, q_idx;
1465 	int s_idx, s_q_idx;
1466 	struct net_device *dev;
1467 
1468 	s_idx = cb->args[0];
1469 	s_q_idx = q_idx = cb->args[1];
1470 
1471 	idx = 0;
1472 	ASSERT_RTNL();
1473 	for_each_netdev(net, dev) {
1474 		struct netdev_queue *dev_queue;
1475 
1476 		if (idx < s_idx)
1477 			goto cont;
1478 		if (idx > s_idx)
1479 			s_q_idx = 0;
1480 		q_idx = 0;
1481 
1482 		if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx) < 0)
1483 			goto done;
1484 
1485 		dev_queue = dev_ingress_queue(dev);
1486 		if (dev_queue &&
1487 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1488 				       &q_idx, s_q_idx) < 0)
1489 			goto done;
1490 
1491 cont:
1492 		idx++;
1493 	}
1494 
1495 done:
1496 	cb->args[0] = idx;
1497 	cb->args[1] = q_idx;
1498 
1499 	return skb->len;
1500 }
1501 
1502 
1503 
1504 /************************************************
1505  *	Traffic classes manipulation.		*
1506  ************************************************/
1507 
1508 
1509 
1510 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n)
1511 {
1512 	struct net *net = sock_net(skb->sk);
1513 	struct tcmsg *tcm = nlmsg_data(n);
1514 	struct nlattr *tca[TCA_MAX + 1];
1515 	struct net_device *dev;
1516 	struct Qdisc *q = NULL;
1517 	const struct Qdisc_class_ops *cops;
1518 	unsigned long cl = 0;
1519 	unsigned long new_cl;
1520 	u32 portid;
1521 	u32 clid;
1522 	u32 qid;
1523 	int err;
1524 
1525 	if ((n->nlmsg_type != RTM_GETTCLASS) &&
1526 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1527 		return -EPERM;
1528 
1529 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, NULL);
1530 	if (err < 0)
1531 		return err;
1532 
1533 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1534 	if (!dev)
1535 		return -ENODEV;
1536 
1537 	/*
1538 	   parent == TC_H_UNSPEC - unspecified parent.
1539 	   parent == TC_H_ROOT   - class is root, which has no parent.
1540 	   parent == X:0	 - parent is root class.
1541 	   parent == X:Y	 - parent is a node in hierarchy.
1542 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
1543 
1544 	   handle == 0:0	 - generate handle from kernel pool.
1545 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
1546 	   handle == X:Y	 - clear.
1547 	   handle == X:0	 - root class.
1548 	 */
1549 
1550 	/* Step 1. Determine qdisc handle X:0 */
1551 
1552 	portid = tcm->tcm_parent;
1553 	clid = tcm->tcm_handle;
1554 	qid = TC_H_MAJ(clid);
1555 
1556 	if (portid != TC_H_ROOT) {
1557 		u32 qid1 = TC_H_MAJ(portid);
1558 
1559 		if (qid && qid1) {
1560 			/* If both majors are known, they must be identical. */
1561 			if (qid != qid1)
1562 				return -EINVAL;
1563 		} else if (qid1) {
1564 			qid = qid1;
1565 		} else if (qid == 0)
1566 			qid = dev->qdisc->handle;
1567 
1568 		/* Now qid is genuine qdisc handle consistent
1569 		 * both with parent and child.
1570 		 *
1571 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1572 		 */
1573 		if (portid)
1574 			portid = TC_H_MAKE(qid, portid);
1575 	} else {
1576 		if (qid == 0)
1577 			qid = dev->qdisc->handle;
1578 	}
1579 
1580 	/* OK. Locate qdisc */
1581 	q = qdisc_lookup(dev, qid);
1582 	if (!q)
1583 		return -ENOENT;
1584 
1585 	/* An check that it supports classes */
1586 	cops = q->ops->cl_ops;
1587 	if (cops == NULL)
1588 		return -EINVAL;
1589 
1590 	/* Now try to get class */
1591 	if (clid == 0) {
1592 		if (portid == TC_H_ROOT)
1593 			clid = qid;
1594 	} else
1595 		clid = TC_H_MAKE(qid, clid);
1596 
1597 	if (clid)
1598 		cl = cops->get(q, clid);
1599 
1600 	if (cl == 0) {
1601 		err = -ENOENT;
1602 		if (n->nlmsg_type != RTM_NEWTCLASS ||
1603 		    !(n->nlmsg_flags & NLM_F_CREATE))
1604 			goto out;
1605 	} else {
1606 		switch (n->nlmsg_type) {
1607 		case RTM_NEWTCLASS:
1608 			err = -EEXIST;
1609 			if (n->nlmsg_flags & NLM_F_EXCL)
1610 				goto out;
1611 			break;
1612 		case RTM_DELTCLASS:
1613 			err = -EOPNOTSUPP;
1614 			if (cops->delete)
1615 				err = cops->delete(q, cl);
1616 			if (err == 0)
1617 				tclass_notify(net, skb, n, q, cl, RTM_DELTCLASS);
1618 			goto out;
1619 		case RTM_GETTCLASS:
1620 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
1621 			goto out;
1622 		default:
1623 			err = -EINVAL;
1624 			goto out;
1625 		}
1626 	}
1627 
1628 	new_cl = cl;
1629 	err = -EOPNOTSUPP;
1630 	if (cops->change)
1631 		err = cops->change(q, clid, portid, tca, &new_cl);
1632 	if (err == 0)
1633 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
1634 
1635 out:
1636 	if (cl)
1637 		cops->put(q, cl);
1638 
1639 	return err;
1640 }
1641 
1642 
1643 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1644 			  unsigned long cl,
1645 			  u32 portid, u32 seq, u16 flags, int event)
1646 {
1647 	struct tcmsg *tcm;
1648 	struct nlmsghdr  *nlh;
1649 	unsigned char *b = skb_tail_pointer(skb);
1650 	struct gnet_dump d;
1651 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1652 
1653 	cond_resched();
1654 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1655 	if (!nlh)
1656 		goto out_nlmsg_trim;
1657 	tcm = nlmsg_data(nlh);
1658 	tcm->tcm_family = AF_UNSPEC;
1659 	tcm->tcm__pad1 = 0;
1660 	tcm->tcm__pad2 = 0;
1661 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1662 	tcm->tcm_parent = q->handle;
1663 	tcm->tcm_handle = q->handle;
1664 	tcm->tcm_info = 0;
1665 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1666 		goto nla_put_failure;
1667 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1668 		goto nla_put_failure;
1669 
1670 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1671 					 qdisc_root_sleeping_lock(q), &d) < 0)
1672 		goto nla_put_failure;
1673 
1674 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1675 		goto nla_put_failure;
1676 
1677 	if (gnet_stats_finish_copy(&d) < 0)
1678 		goto nla_put_failure;
1679 
1680 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1681 	return skb->len;
1682 
1683 out_nlmsg_trim:
1684 nla_put_failure:
1685 	nlmsg_trim(skb, b);
1686 	return -1;
1687 }
1688 
1689 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1690 			 struct nlmsghdr *n, struct Qdisc *q,
1691 			 unsigned long cl, int event)
1692 {
1693 	struct sk_buff *skb;
1694 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1695 
1696 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1697 	if (!skb)
1698 		return -ENOBUFS;
1699 
1700 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1701 		kfree_skb(skb);
1702 		return -EINVAL;
1703 	}
1704 
1705 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1706 			      n->nlmsg_flags & NLM_F_ECHO);
1707 }
1708 
1709 struct qdisc_dump_args {
1710 	struct qdisc_walker	w;
1711 	struct sk_buff		*skb;
1712 	struct netlink_callback	*cb;
1713 };
1714 
1715 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl, struct qdisc_walker *arg)
1716 {
1717 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
1718 
1719 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
1720 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI, RTM_NEWTCLASS);
1721 }
1722 
1723 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
1724 				struct tcmsg *tcm, struct netlink_callback *cb,
1725 				int *t_p, int s_t)
1726 {
1727 	struct qdisc_dump_args arg;
1728 
1729 	if (tc_qdisc_dump_ignore(q) ||
1730 	    *t_p < s_t || !q->ops->cl_ops ||
1731 	    (tcm->tcm_parent &&
1732 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
1733 		(*t_p)++;
1734 		return 0;
1735 	}
1736 	if (*t_p > s_t)
1737 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
1738 	arg.w.fn = qdisc_class_dump;
1739 	arg.skb = skb;
1740 	arg.cb = cb;
1741 	arg.w.stop  = 0;
1742 	arg.w.skip = cb->args[1];
1743 	arg.w.count = 0;
1744 	q->ops->cl_ops->walk(q, &arg.w);
1745 	cb->args[1] = arg.w.count;
1746 	if (arg.w.stop)
1747 		return -1;
1748 	(*t_p)++;
1749 	return 0;
1750 }
1751 
1752 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
1753 			       struct tcmsg *tcm, struct netlink_callback *cb,
1754 			       int *t_p, int s_t)
1755 {
1756 	struct Qdisc *q;
1757 
1758 	if (!root)
1759 		return 0;
1760 
1761 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
1762 		return -1;
1763 
1764 	list_for_each_entry(q, &root->list, list) {
1765 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
1766 			return -1;
1767 	}
1768 
1769 	return 0;
1770 }
1771 
1772 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
1773 {
1774 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
1775 	struct net *net = sock_net(skb->sk);
1776 	struct netdev_queue *dev_queue;
1777 	struct net_device *dev;
1778 	int t, s_t;
1779 
1780 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
1781 		return 0;
1782 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
1783 	if (!dev)
1784 		return 0;
1785 
1786 	s_t = cb->args[0];
1787 	t = 0;
1788 
1789 	if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
1790 		goto done;
1791 
1792 	dev_queue = dev_ingress_queue(dev);
1793 	if (dev_queue &&
1794 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
1795 				&t, s_t) < 0)
1796 		goto done;
1797 
1798 done:
1799 	cb->args[0] = t;
1800 
1801 	dev_put(dev);
1802 	return skb->len;
1803 }
1804 
1805 /* Main classifier routine: scans classifier chain attached
1806  * to this qdisc, (optionally) tests for protocol and asks
1807  * specific classifiers.
1808  */
1809 int tc_classify(struct sk_buff *skb, const struct tcf_proto *tp,
1810 		struct tcf_result *res, bool compat_mode)
1811 {
1812 	__be16 protocol = tc_skb_protocol(skb);
1813 #ifdef CONFIG_NET_CLS_ACT
1814 	const struct tcf_proto *old_tp = tp;
1815 	int limit = 0;
1816 
1817 reclassify:
1818 #endif
1819 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
1820 		int err;
1821 
1822 		if (tp->protocol != protocol &&
1823 		    tp->protocol != htons(ETH_P_ALL))
1824 			continue;
1825 
1826 		err = tp->classify(skb, tp, res);
1827 #ifdef CONFIG_NET_CLS_ACT
1828 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode))
1829 			goto reset;
1830 #endif
1831 		if (err >= 0)
1832 			return err;
1833 	}
1834 
1835 	return -1;
1836 #ifdef CONFIG_NET_CLS_ACT
1837 reset:
1838 	if (unlikely(limit++ >= MAX_REC_LOOP)) {
1839 		net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
1840 				       tp->q->ops->id, tp->prio & 0xffff,
1841 				       ntohs(tp->protocol));
1842 		return TC_ACT_SHOT;
1843 	}
1844 
1845 	tp = old_tp;
1846 	goto reclassify;
1847 #endif
1848 }
1849 EXPORT_SYMBOL(tc_classify);
1850 
1851 bool tcf_destroy(struct tcf_proto *tp, bool force)
1852 {
1853 	if (tp->ops->destroy(tp, force)) {
1854 		module_put(tp->ops->owner);
1855 		kfree_rcu(tp, rcu);
1856 		return true;
1857 	}
1858 
1859 	return false;
1860 }
1861 
1862 void tcf_destroy_chain(struct tcf_proto __rcu **fl)
1863 {
1864 	struct tcf_proto *tp;
1865 
1866 	while ((tp = rtnl_dereference(*fl)) != NULL) {
1867 		RCU_INIT_POINTER(*fl, tp->next);
1868 		tcf_destroy(tp, true);
1869 	}
1870 }
1871 EXPORT_SYMBOL(tcf_destroy_chain);
1872 
1873 #ifdef CONFIG_PROC_FS
1874 static int psched_show(struct seq_file *seq, void *v)
1875 {
1876 	seq_printf(seq, "%08x %08x %08x %08x\n",
1877 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
1878 		   1000000,
1879 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
1880 
1881 	return 0;
1882 }
1883 
1884 static int psched_open(struct inode *inode, struct file *file)
1885 {
1886 	return single_open(file, psched_show, NULL);
1887 }
1888 
1889 static const struct file_operations psched_fops = {
1890 	.owner = THIS_MODULE,
1891 	.open = psched_open,
1892 	.read  = seq_read,
1893 	.llseek = seq_lseek,
1894 	.release = single_release,
1895 };
1896 
1897 static int __net_init psched_net_init(struct net *net)
1898 {
1899 	struct proc_dir_entry *e;
1900 
1901 	e = proc_create("psched", 0, net->proc_net, &psched_fops);
1902 	if (e == NULL)
1903 		return -ENOMEM;
1904 
1905 	return 0;
1906 }
1907 
1908 static void __net_exit psched_net_exit(struct net *net)
1909 {
1910 	remove_proc_entry("psched", net->proc_net);
1911 }
1912 #else
1913 static int __net_init psched_net_init(struct net *net)
1914 {
1915 	return 0;
1916 }
1917 
1918 static void __net_exit psched_net_exit(struct net *net)
1919 {
1920 }
1921 #endif
1922 
1923 static struct pernet_operations psched_net_ops = {
1924 	.init = psched_net_init,
1925 	.exit = psched_net_exit,
1926 };
1927 
1928 static int __init pktsched_init(void)
1929 {
1930 	int err;
1931 
1932 	err = register_pernet_subsys(&psched_net_ops);
1933 	if (err) {
1934 		pr_err("pktsched_init: "
1935 		       "cannot initialize per netns operations\n");
1936 		return err;
1937 	}
1938 
1939 	register_qdisc(&pfifo_fast_ops);
1940 	register_qdisc(&pfifo_qdisc_ops);
1941 	register_qdisc(&bfifo_qdisc_ops);
1942 	register_qdisc(&pfifo_head_drop_qdisc_ops);
1943 	register_qdisc(&mq_qdisc_ops);
1944 	register_qdisc(&noqueue_qdisc_ops);
1945 
1946 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, NULL);
1947 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, NULL);
1948 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc, NULL);
1949 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, NULL);
1950 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, NULL);
1951 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass, NULL);
1952 
1953 	return 0;
1954 }
1955 
1956 subsys_initcall(pktsched_init);
1957