xref: /openbmc/linux/net/sched/sch_api.c (revision df687341)
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/slab.h>
31 #include <linux/hashtable.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 #include <net/pkt_cls.h>
38 
39 /*
40 
41    Short review.
42    -------------
43 
44    This file consists of two interrelated parts:
45 
46    1. queueing disciplines manager frontend.
47    2. traffic classes manager frontend.
48 
49    Generally, queueing discipline ("qdisc") is a black box,
50    which is able to enqueue packets and to dequeue them (when
51    device is ready to send something) in order and at times
52    determined by algorithm hidden in it.
53 
54    qdisc's are divided to two categories:
55    - "queues", which have no internal structure visible from outside.
56    - "schedulers", which split all the packets to "traffic classes",
57      using "packet classifiers" (look at cls_api.c)
58 
59    In turn, classes may have child qdiscs (as rule, queues)
60    attached to them etc. etc. etc.
61 
62    The goal of the routines in this file is to translate
63    information supplied by user in the form of handles
64    to more intelligible for kernel form, to make some sanity
65    checks and part of work, which is common to all qdiscs
66    and to provide rtnetlink notifications.
67 
68    All real intelligent work is done inside qdisc modules.
69 
70 
71 
72    Every discipline has two major routines: enqueue and dequeue.
73 
74    ---dequeue
75 
76    dequeue usually returns a skb to send. It is allowed to return NULL,
77    but it does not mean that queue is empty, it just means that
78    discipline does not want to send anything this time.
79    Queue is really empty if q->q.qlen == 0.
80    For complicated disciplines with multiple queues q->q is not
81    real packet queue, but however q->q.qlen must be valid.
82 
83    ---enqueue
84 
85    enqueue returns 0, if packet was enqueued successfully.
86    If packet (this one or another one) was dropped, it returns
87    not zero error code.
88    NET_XMIT_DROP 	- this packet dropped
89      Expected action: do not backoff, but wait until queue will clear.
90    NET_XMIT_CN	 	- probably this packet enqueued, but another one dropped.
91      Expected action: backoff or ignore
92 
93    Auxiliary routines:
94 
95    ---peek
96 
97    like dequeue but without removing a packet from the queue
98 
99    ---reset
100 
101    returns qdisc to initial state: purge all buffers, clear all
102    timers, counters (except for statistics) etc.
103 
104    ---init
105 
106    initializes newly created qdisc.
107 
108    ---destroy
109 
110    destroys resources allocated by init and during lifetime of qdisc.
111 
112    ---change
113 
114    changes qdisc parameters.
115  */
116 
117 /* Protects list of registered TC modules. It is pure SMP lock. */
118 static DEFINE_RWLOCK(qdisc_mod_lock);
119 
120 
121 /************************************************
122  *	Queueing disciplines manipulation.	*
123  ************************************************/
124 
125 
126 /* The list of all installed queueing disciplines. */
127 
128 static struct Qdisc_ops *qdisc_base;
129 
130 /* Register/unregister queueing discipline */
131 
132 int register_qdisc(struct Qdisc_ops *qops)
133 {
134 	struct Qdisc_ops *q, **qp;
135 	int rc = -EEXIST;
136 
137 	write_lock(&qdisc_mod_lock);
138 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
139 		if (!strcmp(qops->id, q->id))
140 			goto out;
141 
142 	if (qops->enqueue == NULL)
143 		qops->enqueue = noop_qdisc_ops.enqueue;
144 	if (qops->peek == NULL) {
145 		if (qops->dequeue == NULL)
146 			qops->peek = noop_qdisc_ops.peek;
147 		else
148 			goto out_einval;
149 	}
150 	if (qops->dequeue == NULL)
151 		qops->dequeue = noop_qdisc_ops.dequeue;
152 
153 	if (qops->cl_ops) {
154 		const struct Qdisc_class_ops *cops = qops->cl_ops;
155 
156 		if (!(cops->find && cops->walk && cops->leaf))
157 			goto out_einval;
158 
159 		if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
160 			goto out_einval;
161 	}
162 
163 	qops->next = NULL;
164 	*qp = qops;
165 	rc = 0;
166 out:
167 	write_unlock(&qdisc_mod_lock);
168 	return rc;
169 
170 out_einval:
171 	rc = -EINVAL;
172 	goto out;
173 }
174 EXPORT_SYMBOL(register_qdisc);
175 
176 int unregister_qdisc(struct Qdisc_ops *qops)
177 {
178 	struct Qdisc_ops *q, **qp;
179 	int err = -ENOENT;
180 
181 	write_lock(&qdisc_mod_lock);
182 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
183 		if (q == qops)
184 			break;
185 	if (q) {
186 		*qp = q->next;
187 		q->next = NULL;
188 		err = 0;
189 	}
190 	write_unlock(&qdisc_mod_lock);
191 	return err;
192 }
193 EXPORT_SYMBOL(unregister_qdisc);
194 
195 /* Get default qdisc if not otherwise specified */
196 void qdisc_get_default(char *name, size_t len)
197 {
198 	read_lock(&qdisc_mod_lock);
199 	strlcpy(name, default_qdisc_ops->id, len);
200 	read_unlock(&qdisc_mod_lock);
201 }
202 
203 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
204 {
205 	struct Qdisc_ops *q = NULL;
206 
207 	for (q = qdisc_base; q; q = q->next) {
208 		if (!strcmp(name, q->id)) {
209 			if (!try_module_get(q->owner))
210 				q = NULL;
211 			break;
212 		}
213 	}
214 
215 	return q;
216 }
217 
218 /* Set new default qdisc to use */
219 int qdisc_set_default(const char *name)
220 {
221 	const struct Qdisc_ops *ops;
222 
223 	if (!capable(CAP_NET_ADMIN))
224 		return -EPERM;
225 
226 	write_lock(&qdisc_mod_lock);
227 	ops = qdisc_lookup_default(name);
228 	if (!ops) {
229 		/* Not found, drop lock and try to load module */
230 		write_unlock(&qdisc_mod_lock);
231 		request_module("sch_%s", name);
232 		write_lock(&qdisc_mod_lock);
233 
234 		ops = qdisc_lookup_default(name);
235 	}
236 
237 	if (ops) {
238 		/* Set new default */
239 		module_put(default_qdisc_ops->owner);
240 		default_qdisc_ops = ops;
241 	}
242 	write_unlock(&qdisc_mod_lock);
243 
244 	return ops ? 0 : -ENOENT;
245 }
246 
247 #ifdef CONFIG_NET_SCH_DEFAULT
248 /* Set default value from kernel config */
249 static int __init sch_default_qdisc(void)
250 {
251 	return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
252 }
253 late_initcall(sch_default_qdisc);
254 #endif
255 
256 /* We know handle. Find qdisc among all qdisc's attached to device
257  * (root qdisc, all its children, children of children etc.)
258  * Note: caller either uses rtnl or rcu_read_lock()
259  */
260 
261 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
262 {
263 	struct Qdisc *q;
264 
265 	if (!qdisc_dev(root))
266 		return (root->handle == handle ? root : NULL);
267 
268 	if (!(root->flags & TCQ_F_BUILTIN) &&
269 	    root->handle == handle)
270 		return root;
271 
272 	hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle) {
273 		if (q->handle == handle)
274 			return q;
275 	}
276 	return NULL;
277 }
278 
279 void qdisc_hash_add(struct Qdisc *q, bool invisible)
280 {
281 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
282 		ASSERT_RTNL();
283 		hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
284 		if (invisible)
285 			q->flags |= TCQ_F_INVISIBLE;
286 	}
287 }
288 EXPORT_SYMBOL(qdisc_hash_add);
289 
290 void qdisc_hash_del(struct Qdisc *q)
291 {
292 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
293 		ASSERT_RTNL();
294 		hash_del_rcu(&q->hash);
295 	}
296 }
297 EXPORT_SYMBOL(qdisc_hash_del);
298 
299 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
300 {
301 	struct Qdisc *q;
302 
303 	if (!handle)
304 		return NULL;
305 	q = qdisc_match_from_root(dev->qdisc, handle);
306 	if (q)
307 		goto out;
308 
309 	if (dev_ingress_queue(dev))
310 		q = qdisc_match_from_root(
311 			dev_ingress_queue(dev)->qdisc_sleeping,
312 			handle);
313 out:
314 	return q;
315 }
316 
317 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
318 {
319 	struct netdev_queue *nq;
320 	struct Qdisc *q;
321 
322 	if (!handle)
323 		return NULL;
324 	q = qdisc_match_from_root(dev->qdisc, handle);
325 	if (q)
326 		goto out;
327 
328 	nq = dev_ingress_queue_rcu(dev);
329 	if (nq)
330 		q = qdisc_match_from_root(nq->qdisc_sleeping, handle);
331 out:
332 	return q;
333 }
334 
335 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
336 {
337 	unsigned long cl;
338 	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
339 
340 	if (cops == NULL)
341 		return NULL;
342 	cl = cops->find(p, classid);
343 
344 	if (cl == 0)
345 		return NULL;
346 	return cops->leaf(p, cl);
347 }
348 
349 /* Find queueing discipline by name */
350 
351 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
352 {
353 	struct Qdisc_ops *q = NULL;
354 
355 	if (kind) {
356 		read_lock(&qdisc_mod_lock);
357 		for (q = qdisc_base; q; q = q->next) {
358 			if (nla_strcmp(kind, q->id) == 0) {
359 				if (!try_module_get(q->owner))
360 					q = NULL;
361 				break;
362 			}
363 		}
364 		read_unlock(&qdisc_mod_lock);
365 	}
366 	return q;
367 }
368 
369 /* The linklayer setting were not transferred from iproute2, in older
370  * versions, and the rate tables lookup systems have been dropped in
371  * the kernel. To keep backward compatible with older iproute2 tc
372  * utils, we detect the linklayer setting by detecting if the rate
373  * table were modified.
374  *
375  * For linklayer ATM table entries, the rate table will be aligned to
376  * 48 bytes, thus some table entries will contain the same value.  The
377  * mpu (min packet unit) is also encoded into the old rate table, thus
378  * starting from the mpu, we find low and high table entries for
379  * mapping this cell.  If these entries contain the same value, when
380  * the rate tables have been modified for linklayer ATM.
381  *
382  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
383  * and then roundup to the next cell, calc the table entry one below,
384  * and compare.
385  */
386 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
387 {
388 	int low       = roundup(r->mpu, 48);
389 	int high      = roundup(low+1, 48);
390 	int cell_low  = low >> r->cell_log;
391 	int cell_high = (high >> r->cell_log) - 1;
392 
393 	/* rtab is too inaccurate at rates > 100Mbit/s */
394 	if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
395 		pr_debug("TC linklayer: Giving up ATM detection\n");
396 		return TC_LINKLAYER_ETHERNET;
397 	}
398 
399 	if ((cell_high > cell_low) && (cell_high < 256)
400 	    && (rtab[cell_low] == rtab[cell_high])) {
401 		pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
402 			 cell_low, cell_high, rtab[cell_high]);
403 		return TC_LINKLAYER_ATM;
404 	}
405 	return TC_LINKLAYER_ETHERNET;
406 }
407 
408 static struct qdisc_rate_table *qdisc_rtab_list;
409 
410 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
411 					struct nlattr *tab,
412 					struct netlink_ext_ack *extack)
413 {
414 	struct qdisc_rate_table *rtab;
415 
416 	if (tab == NULL || r->rate == 0 || r->cell_log == 0 ||
417 	    nla_len(tab) != TC_RTAB_SIZE) {
418 		NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
419 		return NULL;
420 	}
421 
422 	for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
423 		if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
424 		    !memcmp(&rtab->data, nla_data(tab), 1024)) {
425 			rtab->refcnt++;
426 			return rtab;
427 		}
428 	}
429 
430 	rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
431 	if (rtab) {
432 		rtab->rate = *r;
433 		rtab->refcnt = 1;
434 		memcpy(rtab->data, nla_data(tab), 1024);
435 		if (r->linklayer == TC_LINKLAYER_UNAWARE)
436 			r->linklayer = __detect_linklayer(r, rtab->data);
437 		rtab->next = qdisc_rtab_list;
438 		qdisc_rtab_list = rtab;
439 	} else {
440 		NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
441 	}
442 	return rtab;
443 }
444 EXPORT_SYMBOL(qdisc_get_rtab);
445 
446 void qdisc_put_rtab(struct qdisc_rate_table *tab)
447 {
448 	struct qdisc_rate_table *rtab, **rtabp;
449 
450 	if (!tab || --tab->refcnt)
451 		return;
452 
453 	for (rtabp = &qdisc_rtab_list;
454 	     (rtab = *rtabp) != NULL;
455 	     rtabp = &rtab->next) {
456 		if (rtab == tab) {
457 			*rtabp = rtab->next;
458 			kfree(rtab);
459 			return;
460 		}
461 	}
462 }
463 EXPORT_SYMBOL(qdisc_put_rtab);
464 
465 static LIST_HEAD(qdisc_stab_list);
466 
467 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
468 	[TCA_STAB_BASE]	= { .len = sizeof(struct tc_sizespec) },
469 	[TCA_STAB_DATA] = { .type = NLA_BINARY },
470 };
471 
472 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
473 					       struct netlink_ext_ack *extack)
474 {
475 	struct nlattr *tb[TCA_STAB_MAX + 1];
476 	struct qdisc_size_table *stab;
477 	struct tc_sizespec *s;
478 	unsigned int tsize = 0;
479 	u16 *tab = NULL;
480 	int err;
481 
482 	err = nla_parse_nested(tb, TCA_STAB_MAX, opt, stab_policy, extack);
483 	if (err < 0)
484 		return ERR_PTR(err);
485 	if (!tb[TCA_STAB_BASE]) {
486 		NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
487 		return ERR_PTR(-EINVAL);
488 	}
489 
490 	s = nla_data(tb[TCA_STAB_BASE]);
491 
492 	if (s->tsize > 0) {
493 		if (!tb[TCA_STAB_DATA]) {
494 			NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
495 			return ERR_PTR(-EINVAL);
496 		}
497 		tab = nla_data(tb[TCA_STAB_DATA]);
498 		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
499 	}
500 
501 	if (tsize != s->tsize || (!tab && tsize > 0)) {
502 		NL_SET_ERR_MSG(extack, "Invalid size of size table");
503 		return ERR_PTR(-EINVAL);
504 	}
505 
506 	list_for_each_entry(stab, &qdisc_stab_list, list) {
507 		if (memcmp(&stab->szopts, s, sizeof(*s)))
508 			continue;
509 		if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
510 			continue;
511 		stab->refcnt++;
512 		return stab;
513 	}
514 
515 	stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
516 	if (!stab)
517 		return ERR_PTR(-ENOMEM);
518 
519 	stab->refcnt = 1;
520 	stab->szopts = *s;
521 	if (tsize > 0)
522 		memcpy(stab->data, tab, tsize * sizeof(u16));
523 
524 	list_add_tail(&stab->list, &qdisc_stab_list);
525 
526 	return stab;
527 }
528 
529 static void stab_kfree_rcu(struct rcu_head *head)
530 {
531 	kfree(container_of(head, struct qdisc_size_table, rcu));
532 }
533 
534 void qdisc_put_stab(struct qdisc_size_table *tab)
535 {
536 	if (!tab)
537 		return;
538 
539 	if (--tab->refcnt == 0) {
540 		list_del(&tab->list);
541 		call_rcu(&tab->rcu, stab_kfree_rcu);
542 	}
543 }
544 EXPORT_SYMBOL(qdisc_put_stab);
545 
546 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
547 {
548 	struct nlattr *nest;
549 
550 	nest = nla_nest_start(skb, TCA_STAB);
551 	if (nest == NULL)
552 		goto nla_put_failure;
553 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
554 		goto nla_put_failure;
555 	nla_nest_end(skb, nest);
556 
557 	return skb->len;
558 
559 nla_put_failure:
560 	return -1;
561 }
562 
563 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
564 			       const struct qdisc_size_table *stab)
565 {
566 	int pkt_len, slot;
567 
568 	pkt_len = skb->len + stab->szopts.overhead;
569 	if (unlikely(!stab->szopts.tsize))
570 		goto out;
571 
572 	slot = pkt_len + stab->szopts.cell_align;
573 	if (unlikely(slot < 0))
574 		slot = 0;
575 
576 	slot >>= stab->szopts.cell_log;
577 	if (likely(slot < stab->szopts.tsize))
578 		pkt_len = stab->data[slot];
579 	else
580 		pkt_len = stab->data[stab->szopts.tsize - 1] *
581 				(slot / stab->szopts.tsize) +
582 				stab->data[slot % stab->szopts.tsize];
583 
584 	pkt_len <<= stab->szopts.size_log;
585 out:
586 	if (unlikely(pkt_len < 1))
587 		pkt_len = 1;
588 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
589 }
590 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
591 
592 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
593 {
594 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
595 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
596 			txt, qdisc->ops->id, qdisc->handle >> 16);
597 		qdisc->flags |= TCQ_F_WARN_NONWC;
598 	}
599 }
600 EXPORT_SYMBOL(qdisc_warn_nonwc);
601 
602 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
603 {
604 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
605 						 timer);
606 
607 	rcu_read_lock();
608 	__netif_schedule(qdisc_root(wd->qdisc));
609 	rcu_read_unlock();
610 
611 	return HRTIMER_NORESTART;
612 }
613 
614 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
615 				 clockid_t clockid)
616 {
617 	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
618 	wd->timer.function = qdisc_watchdog;
619 	wd->qdisc = qdisc;
620 }
621 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
622 
623 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
624 {
625 	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
626 }
627 EXPORT_SYMBOL(qdisc_watchdog_init);
628 
629 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog *wd, u64 expires)
630 {
631 	if (test_bit(__QDISC_STATE_DEACTIVATED,
632 		     &qdisc_root_sleeping(wd->qdisc)->state))
633 		return;
634 
635 	if (wd->last_expires == expires)
636 		return;
637 
638 	wd->last_expires = expires;
639 	hrtimer_start(&wd->timer,
640 		      ns_to_ktime(expires),
641 		      HRTIMER_MODE_ABS_PINNED);
642 }
643 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns);
644 
645 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
646 {
647 	hrtimer_cancel(&wd->timer);
648 }
649 EXPORT_SYMBOL(qdisc_watchdog_cancel);
650 
651 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
652 {
653 	struct hlist_head *h;
654 	unsigned int i;
655 
656 	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
657 
658 	if (h != NULL) {
659 		for (i = 0; i < n; i++)
660 			INIT_HLIST_HEAD(&h[i]);
661 	}
662 	return h;
663 }
664 
665 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
666 {
667 	struct Qdisc_class_common *cl;
668 	struct hlist_node *next;
669 	struct hlist_head *nhash, *ohash;
670 	unsigned int nsize, nmask, osize;
671 	unsigned int i, h;
672 
673 	/* Rehash when load factor exceeds 0.75 */
674 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
675 		return;
676 	nsize = clhash->hashsize * 2;
677 	nmask = nsize - 1;
678 	nhash = qdisc_class_hash_alloc(nsize);
679 	if (nhash == NULL)
680 		return;
681 
682 	ohash = clhash->hash;
683 	osize = clhash->hashsize;
684 
685 	sch_tree_lock(sch);
686 	for (i = 0; i < osize; i++) {
687 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
688 			h = qdisc_class_hash(cl->classid, nmask);
689 			hlist_add_head(&cl->hnode, &nhash[h]);
690 		}
691 	}
692 	clhash->hash     = nhash;
693 	clhash->hashsize = nsize;
694 	clhash->hashmask = nmask;
695 	sch_tree_unlock(sch);
696 
697 	kvfree(ohash);
698 }
699 EXPORT_SYMBOL(qdisc_class_hash_grow);
700 
701 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
702 {
703 	unsigned int size = 4;
704 
705 	clhash->hash = qdisc_class_hash_alloc(size);
706 	if (!clhash->hash)
707 		return -ENOMEM;
708 	clhash->hashsize  = size;
709 	clhash->hashmask  = size - 1;
710 	clhash->hashelems = 0;
711 	return 0;
712 }
713 EXPORT_SYMBOL(qdisc_class_hash_init);
714 
715 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
716 {
717 	kvfree(clhash->hash);
718 }
719 EXPORT_SYMBOL(qdisc_class_hash_destroy);
720 
721 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
722 			     struct Qdisc_class_common *cl)
723 {
724 	unsigned int h;
725 
726 	INIT_HLIST_NODE(&cl->hnode);
727 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
728 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
729 	clhash->hashelems++;
730 }
731 EXPORT_SYMBOL(qdisc_class_hash_insert);
732 
733 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
734 			     struct Qdisc_class_common *cl)
735 {
736 	hlist_del(&cl->hnode);
737 	clhash->hashelems--;
738 }
739 EXPORT_SYMBOL(qdisc_class_hash_remove);
740 
741 /* Allocate an unique handle from space managed by kernel
742  * Possible range is [8000-FFFF]:0000 (0x8000 values)
743  */
744 static u32 qdisc_alloc_handle(struct net_device *dev)
745 {
746 	int i = 0x8000;
747 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
748 
749 	do {
750 		autohandle += TC_H_MAKE(0x10000U, 0);
751 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
752 			autohandle = TC_H_MAKE(0x80000000U, 0);
753 		if (!qdisc_lookup(dev, autohandle))
754 			return autohandle;
755 		cond_resched();
756 	} while	(--i > 0);
757 
758 	return 0;
759 }
760 
761 void qdisc_tree_reduce_backlog(struct Qdisc *sch, unsigned int n,
762 			       unsigned int len)
763 {
764 	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
765 	const struct Qdisc_class_ops *cops;
766 	unsigned long cl;
767 	u32 parentid;
768 	bool notify;
769 	int drops;
770 
771 	if (n == 0 && len == 0)
772 		return;
773 	drops = max_t(int, n, 0);
774 	rcu_read_lock();
775 	while ((parentid = sch->parent)) {
776 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
777 			break;
778 
779 		if (sch->flags & TCQ_F_NOPARENT)
780 			break;
781 		/* Notify parent qdisc only if child qdisc becomes empty.
782 		 *
783 		 * If child was empty even before update then backlog
784 		 * counter is screwed and we skip notification because
785 		 * parent class is already passive.
786 		 *
787 		 * If the original child was offloaded then it is allowed
788 		 * to be seem as empty, so the parent is notified anyway.
789 		 */
790 		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
791 						       !qdisc_is_offloaded);
792 		/* TODO: perform the search on a per txq basis */
793 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
794 		if (sch == NULL) {
795 			WARN_ON_ONCE(parentid != TC_H_ROOT);
796 			break;
797 		}
798 		cops = sch->ops->cl_ops;
799 		if (notify && cops->qlen_notify) {
800 			cl = cops->find(sch, parentid);
801 			cops->qlen_notify(sch, cl);
802 		}
803 		sch->q.qlen -= n;
804 		sch->qstats.backlog -= len;
805 		__qdisc_qstats_drop(sch, drops);
806 	}
807 	rcu_read_unlock();
808 }
809 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
810 
811 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
812 			      void *type_data)
813 {
814 	struct net_device *dev = qdisc_dev(sch);
815 	int err;
816 
817 	sch->flags &= ~TCQ_F_OFFLOADED;
818 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
819 		return 0;
820 
821 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
822 	if (err == -EOPNOTSUPP)
823 		return 0;
824 
825 	if (!err)
826 		sch->flags |= TCQ_F_OFFLOADED;
827 
828 	return err;
829 }
830 EXPORT_SYMBOL(qdisc_offload_dump_helper);
831 
832 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
833 				struct Qdisc *new, struct Qdisc *old,
834 				enum tc_setup_type type, void *type_data,
835 				struct netlink_ext_ack *extack)
836 {
837 	bool any_qdisc_is_offloaded;
838 	int err;
839 
840 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
841 		return;
842 
843 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
844 
845 	/* Don't report error if the graft is part of destroy operation. */
846 	if (!err || !new || new == &noop_qdisc)
847 		return;
848 
849 	/* Don't report error if the parent, the old child and the new
850 	 * one are not offloaded.
851 	 */
852 	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
853 	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
854 	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
855 
856 	if (any_qdisc_is_offloaded)
857 		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
858 }
859 EXPORT_SYMBOL(qdisc_offload_graft_helper);
860 
861 static void qdisc_offload_graft_root(struct net_device *dev,
862 				     struct Qdisc *new, struct Qdisc *old,
863 				     struct netlink_ext_ack *extack)
864 {
865 	struct tc_root_qopt_offload graft_offload = {
866 		.command	= TC_ROOT_GRAFT,
867 		.handle		= new ? new->handle : 0,
868 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
869 				  (old && old->flags & TCQ_F_INGRESS),
870 	};
871 
872 	qdisc_offload_graft_helper(dev, NULL, new, old,
873 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
874 }
875 
876 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
877 			 u32 portid, u32 seq, u16 flags, int event)
878 {
879 	struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
880 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
881 	struct tcmsg *tcm;
882 	struct nlmsghdr  *nlh;
883 	unsigned char *b = skb_tail_pointer(skb);
884 	struct gnet_dump d;
885 	struct qdisc_size_table *stab;
886 	u32 block_index;
887 	__u32 qlen;
888 
889 	cond_resched();
890 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
891 	if (!nlh)
892 		goto out_nlmsg_trim;
893 	tcm = nlmsg_data(nlh);
894 	tcm->tcm_family = AF_UNSPEC;
895 	tcm->tcm__pad1 = 0;
896 	tcm->tcm__pad2 = 0;
897 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
898 	tcm->tcm_parent = clid;
899 	tcm->tcm_handle = q->handle;
900 	tcm->tcm_info = refcount_read(&q->refcnt);
901 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
902 		goto nla_put_failure;
903 	if (q->ops->ingress_block_get) {
904 		block_index = q->ops->ingress_block_get(q);
905 		if (block_index &&
906 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
907 			goto nla_put_failure;
908 	}
909 	if (q->ops->egress_block_get) {
910 		block_index = q->ops->egress_block_get(q);
911 		if (block_index &&
912 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
913 			goto nla_put_failure;
914 	}
915 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
916 		goto nla_put_failure;
917 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
918 		goto nla_put_failure;
919 	qlen = qdisc_qlen_sum(q);
920 
921 	stab = rtnl_dereference(q->stab);
922 	if (stab && qdisc_dump_stab(skb, stab) < 0)
923 		goto nla_put_failure;
924 
925 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
926 					 NULL, &d, TCA_PAD) < 0)
927 		goto nla_put_failure;
928 
929 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
930 		goto nla_put_failure;
931 
932 	if (qdisc_is_percpu_stats(q)) {
933 		cpu_bstats = q->cpu_bstats;
934 		cpu_qstats = q->cpu_qstats;
935 	}
936 
937 	if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
938 				  &d, cpu_bstats, &q->bstats) < 0 ||
939 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
940 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
941 		goto nla_put_failure;
942 
943 	if (gnet_stats_finish_copy(&d) < 0)
944 		goto nla_put_failure;
945 
946 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
947 	return skb->len;
948 
949 out_nlmsg_trim:
950 nla_put_failure:
951 	nlmsg_trim(skb, b);
952 	return -1;
953 }
954 
955 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
956 {
957 	if (q->flags & TCQ_F_BUILTIN)
958 		return true;
959 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
960 		return true;
961 
962 	return false;
963 }
964 
965 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
966 			struct nlmsghdr *n, u32 clid,
967 			struct Qdisc *old, struct Qdisc *new)
968 {
969 	struct sk_buff *skb;
970 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
971 
972 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
973 	if (!skb)
974 		return -ENOBUFS;
975 
976 	if (old && !tc_qdisc_dump_ignore(old, false)) {
977 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
978 				  0, RTM_DELQDISC) < 0)
979 			goto err_out;
980 	}
981 	if (new && !tc_qdisc_dump_ignore(new, false)) {
982 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
983 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
984 			goto err_out;
985 	}
986 
987 	if (skb->len)
988 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
989 				      n->nlmsg_flags & NLM_F_ECHO);
990 
991 err_out:
992 	kfree_skb(skb);
993 	return -EINVAL;
994 }
995 
996 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
997 			       struct nlmsghdr *n, u32 clid,
998 			       struct Qdisc *old, struct Qdisc *new)
999 {
1000 	if (new || old)
1001 		qdisc_notify(net, skb, n, clid, old, new);
1002 
1003 	if (old)
1004 		qdisc_put(old);
1005 }
1006 
1007 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1008  * to device "dev".
1009  *
1010  * When appropriate send a netlink notification using 'skb'
1011  * and "n".
1012  *
1013  * On success, destroy old qdisc.
1014  */
1015 
1016 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1017 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1018 		       struct Qdisc *new, struct Qdisc *old,
1019 		       struct netlink_ext_ack *extack)
1020 {
1021 	struct Qdisc *q = old;
1022 	struct net *net = dev_net(dev);
1023 
1024 	if (parent == NULL) {
1025 		unsigned int i, num_q, ingress;
1026 
1027 		ingress = 0;
1028 		num_q = dev->num_tx_queues;
1029 		if ((q && q->flags & TCQ_F_INGRESS) ||
1030 		    (new && new->flags & TCQ_F_INGRESS)) {
1031 			num_q = 1;
1032 			ingress = 1;
1033 			if (!dev_ingress_queue(dev)) {
1034 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1035 				return -ENOENT;
1036 			}
1037 		}
1038 
1039 		if (dev->flags & IFF_UP)
1040 			dev_deactivate(dev);
1041 
1042 		qdisc_offload_graft_root(dev, new, old, extack);
1043 
1044 		if (new && new->ops->attach)
1045 			goto skip;
1046 
1047 		for (i = 0; i < num_q; i++) {
1048 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1049 
1050 			if (!ingress)
1051 				dev_queue = netdev_get_tx_queue(dev, i);
1052 
1053 			old = dev_graft_qdisc(dev_queue, new);
1054 			if (new && i > 0)
1055 				qdisc_refcount_inc(new);
1056 
1057 			if (!ingress)
1058 				qdisc_put(old);
1059 		}
1060 
1061 skip:
1062 		if (!ingress) {
1063 			notify_and_destroy(net, skb, n, classid,
1064 					   dev->qdisc, new);
1065 			if (new && !new->ops->attach)
1066 				qdisc_refcount_inc(new);
1067 			dev->qdisc = new ? : &noop_qdisc;
1068 
1069 			if (new && new->ops->attach)
1070 				new->ops->attach(new);
1071 		} else {
1072 			notify_and_destroy(net, skb, n, classid, old, new);
1073 		}
1074 
1075 		if (dev->flags & IFF_UP)
1076 			dev_activate(dev);
1077 	} else {
1078 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1079 		unsigned long cl;
1080 		int err;
1081 
1082 		/* Only support running class lockless if parent is lockless */
1083 		if (new && (new->flags & TCQ_F_NOLOCK) &&
1084 		    parent && !(parent->flags & TCQ_F_NOLOCK))
1085 			new->flags &= ~TCQ_F_NOLOCK;
1086 
1087 		if (!cops || !cops->graft)
1088 			return -EOPNOTSUPP;
1089 
1090 		cl = cops->find(parent, classid);
1091 		if (!cl) {
1092 			NL_SET_ERR_MSG(extack, "Specified class not found");
1093 			return -ENOENT;
1094 		}
1095 
1096 		err = cops->graft(parent, cl, new, &old, extack);
1097 		if (err)
1098 			return err;
1099 		notify_and_destroy(net, skb, n, classid, old, new);
1100 	}
1101 	return 0;
1102 }
1103 
1104 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1105 				   struct netlink_ext_ack *extack)
1106 {
1107 	u32 block_index;
1108 
1109 	if (tca[TCA_INGRESS_BLOCK]) {
1110 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1111 
1112 		if (!block_index) {
1113 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1114 			return -EINVAL;
1115 		}
1116 		if (!sch->ops->ingress_block_set) {
1117 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1118 			return -EOPNOTSUPP;
1119 		}
1120 		sch->ops->ingress_block_set(sch, block_index);
1121 	}
1122 	if (tca[TCA_EGRESS_BLOCK]) {
1123 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1124 
1125 		if (!block_index) {
1126 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1127 			return -EINVAL;
1128 		}
1129 		if (!sch->ops->egress_block_set) {
1130 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1131 			return -EOPNOTSUPP;
1132 		}
1133 		sch->ops->egress_block_set(sch, block_index);
1134 	}
1135 	return 0;
1136 }
1137 
1138 /*
1139    Allocate and initialize new qdisc.
1140 
1141    Parameters are passed via opt.
1142  */
1143 
1144 static struct Qdisc *qdisc_create(struct net_device *dev,
1145 				  struct netdev_queue *dev_queue,
1146 				  struct Qdisc *p, u32 parent, u32 handle,
1147 				  struct nlattr **tca, int *errp,
1148 				  struct netlink_ext_ack *extack)
1149 {
1150 	int err;
1151 	struct nlattr *kind = tca[TCA_KIND];
1152 	struct Qdisc *sch;
1153 	struct Qdisc_ops *ops;
1154 	struct qdisc_size_table *stab;
1155 
1156 	ops = qdisc_lookup_ops(kind);
1157 #ifdef CONFIG_MODULES
1158 	if (ops == NULL && kind != NULL) {
1159 		char name[IFNAMSIZ];
1160 		if (nla_strlcpy(name, kind, IFNAMSIZ) < IFNAMSIZ) {
1161 			/* We dropped the RTNL semaphore in order to
1162 			 * perform the module load.  So, even if we
1163 			 * succeeded in loading the module we have to
1164 			 * tell the caller to replay the request.  We
1165 			 * indicate this using -EAGAIN.
1166 			 * We replay the request because the device may
1167 			 * go away in the mean time.
1168 			 */
1169 			rtnl_unlock();
1170 			request_module("sch_%s", name);
1171 			rtnl_lock();
1172 			ops = qdisc_lookup_ops(kind);
1173 			if (ops != NULL) {
1174 				/* We will try again qdisc_lookup_ops,
1175 				 * so don't keep a reference.
1176 				 */
1177 				module_put(ops->owner);
1178 				err = -EAGAIN;
1179 				goto err_out;
1180 			}
1181 		}
1182 	}
1183 #endif
1184 
1185 	err = -ENOENT;
1186 	if (!ops) {
1187 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1188 		goto err_out;
1189 	}
1190 
1191 	sch = qdisc_alloc(dev_queue, ops, extack);
1192 	if (IS_ERR(sch)) {
1193 		err = PTR_ERR(sch);
1194 		goto err_out2;
1195 	}
1196 
1197 	sch->parent = parent;
1198 
1199 	if (handle == TC_H_INGRESS) {
1200 		sch->flags |= TCQ_F_INGRESS;
1201 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1202 	} else {
1203 		if (handle == 0) {
1204 			handle = qdisc_alloc_handle(dev);
1205 			err = -ENOMEM;
1206 			if (handle == 0)
1207 				goto err_out3;
1208 		}
1209 		if (!netif_is_multiqueue(dev))
1210 			sch->flags |= TCQ_F_ONETXQUEUE;
1211 	}
1212 
1213 	sch->handle = handle;
1214 
1215 	/* This exist to keep backward compatible with a userspace
1216 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1217 	 * facility on older kernels by setting tx_queue_len=0 (prior
1218 	 * to qdisc init), and then forgot to reinit tx_queue_len
1219 	 * before again attaching a qdisc.
1220 	 */
1221 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1222 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1223 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1224 	}
1225 
1226 	err = qdisc_block_indexes_set(sch, tca, extack);
1227 	if (err)
1228 		goto err_out3;
1229 
1230 	if (ops->init) {
1231 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1232 		if (err != 0)
1233 			goto err_out5;
1234 	}
1235 
1236 	if (tca[TCA_STAB]) {
1237 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1238 		if (IS_ERR(stab)) {
1239 			err = PTR_ERR(stab);
1240 			goto err_out4;
1241 		}
1242 		rcu_assign_pointer(sch->stab, stab);
1243 	}
1244 	if (tca[TCA_RATE]) {
1245 		seqcount_t *running;
1246 
1247 		err = -EOPNOTSUPP;
1248 		if (sch->flags & TCQ_F_MQROOT) {
1249 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1250 			goto err_out4;
1251 		}
1252 
1253 		if (sch->parent != TC_H_ROOT &&
1254 		    !(sch->flags & TCQ_F_INGRESS) &&
1255 		    (!p || !(p->flags & TCQ_F_MQROOT)))
1256 			running = qdisc_root_sleeping_running(sch);
1257 		else
1258 			running = &sch->running;
1259 
1260 		err = gen_new_estimator(&sch->bstats,
1261 					sch->cpu_bstats,
1262 					&sch->rate_est,
1263 					NULL,
1264 					running,
1265 					tca[TCA_RATE]);
1266 		if (err) {
1267 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1268 			goto err_out4;
1269 		}
1270 	}
1271 
1272 	qdisc_hash_add(sch, false);
1273 
1274 	return sch;
1275 
1276 err_out5:
1277 	/* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1278 	if (ops->destroy)
1279 		ops->destroy(sch);
1280 err_out3:
1281 	dev_put(dev);
1282 	qdisc_free(sch);
1283 err_out2:
1284 	module_put(ops->owner);
1285 err_out:
1286 	*errp = err;
1287 	return NULL;
1288 
1289 err_out4:
1290 	/*
1291 	 * Any broken qdiscs that would require a ops->reset() here?
1292 	 * The qdisc was never in action so it shouldn't be necessary.
1293 	 */
1294 	qdisc_put_stab(rtnl_dereference(sch->stab));
1295 	if (ops->destroy)
1296 		ops->destroy(sch);
1297 	goto err_out3;
1298 }
1299 
1300 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1301 			struct netlink_ext_ack *extack)
1302 {
1303 	struct qdisc_size_table *ostab, *stab = NULL;
1304 	int err = 0;
1305 
1306 	if (tca[TCA_OPTIONS]) {
1307 		if (!sch->ops->change) {
1308 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1309 			return -EINVAL;
1310 		}
1311 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1312 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1313 			return -EOPNOTSUPP;
1314 		}
1315 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1316 		if (err)
1317 			return err;
1318 	}
1319 
1320 	if (tca[TCA_STAB]) {
1321 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1322 		if (IS_ERR(stab))
1323 			return PTR_ERR(stab);
1324 	}
1325 
1326 	ostab = rtnl_dereference(sch->stab);
1327 	rcu_assign_pointer(sch->stab, stab);
1328 	qdisc_put_stab(ostab);
1329 
1330 	if (tca[TCA_RATE]) {
1331 		/* NB: ignores errors from replace_estimator
1332 		   because change can't be undone. */
1333 		if (sch->flags & TCQ_F_MQROOT)
1334 			goto out;
1335 		gen_replace_estimator(&sch->bstats,
1336 				      sch->cpu_bstats,
1337 				      &sch->rate_est,
1338 				      NULL,
1339 				      qdisc_root_sleeping_running(sch),
1340 				      tca[TCA_RATE]);
1341 	}
1342 out:
1343 	return 0;
1344 }
1345 
1346 struct check_loop_arg {
1347 	struct qdisc_walker	w;
1348 	struct Qdisc		*p;
1349 	int			depth;
1350 };
1351 
1352 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1353 			 struct qdisc_walker *w);
1354 
1355 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1356 {
1357 	struct check_loop_arg	arg;
1358 
1359 	if (q->ops->cl_ops == NULL)
1360 		return 0;
1361 
1362 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1363 	arg.w.fn = check_loop_fn;
1364 	arg.depth = depth;
1365 	arg.p = p;
1366 	q->ops->cl_ops->walk(q, &arg.w);
1367 	return arg.w.stop ? -ELOOP : 0;
1368 }
1369 
1370 static int
1371 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1372 {
1373 	struct Qdisc *leaf;
1374 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1375 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1376 
1377 	leaf = cops->leaf(q, cl);
1378 	if (leaf) {
1379 		if (leaf == arg->p || arg->depth > 7)
1380 			return -ELOOP;
1381 		return check_loop(leaf, arg->p, arg->depth + 1);
1382 	}
1383 	return 0;
1384 }
1385 
1386 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1387 	[TCA_KIND]		= { .type = NLA_STRING },
1388 	[TCA_RATE]		= { .type = NLA_BINARY,
1389 				    .len = sizeof(struct tc_estimator) },
1390 	[TCA_STAB]		= { .type = NLA_NESTED },
1391 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1392 	[TCA_CHAIN]		= { .type = NLA_U32 },
1393 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1394 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1395 };
1396 
1397 /*
1398  * Delete/get qdisc.
1399  */
1400 
1401 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1402 			struct netlink_ext_ack *extack)
1403 {
1404 	struct net *net = sock_net(skb->sk);
1405 	struct tcmsg *tcm = nlmsg_data(n);
1406 	struct nlattr *tca[TCA_MAX + 1];
1407 	struct net_device *dev;
1408 	u32 clid;
1409 	struct Qdisc *q = NULL;
1410 	struct Qdisc *p = NULL;
1411 	int err;
1412 
1413 	if ((n->nlmsg_type != RTM_GETQDISC) &&
1414 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1415 		return -EPERM;
1416 
1417 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
1418 			  extack);
1419 	if (err < 0)
1420 		return err;
1421 
1422 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1423 	if (!dev)
1424 		return -ENODEV;
1425 
1426 	clid = tcm->tcm_parent;
1427 	if (clid) {
1428 		if (clid != TC_H_ROOT) {
1429 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1430 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1431 				if (!p) {
1432 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1433 					return -ENOENT;
1434 				}
1435 				q = qdisc_leaf(p, clid);
1436 			} else if (dev_ingress_queue(dev)) {
1437 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1438 			}
1439 		} else {
1440 			q = dev->qdisc;
1441 		}
1442 		if (!q) {
1443 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1444 			return -ENOENT;
1445 		}
1446 
1447 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1448 			NL_SET_ERR_MSG(extack, "Invalid handle");
1449 			return -EINVAL;
1450 		}
1451 	} else {
1452 		q = qdisc_lookup(dev, tcm->tcm_handle);
1453 		if (!q) {
1454 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1455 			return -ENOENT;
1456 		}
1457 	}
1458 
1459 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1460 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1461 		return -EINVAL;
1462 	}
1463 
1464 	if (n->nlmsg_type == RTM_DELQDISC) {
1465 		if (!clid) {
1466 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1467 			return -EINVAL;
1468 		}
1469 		if (q->handle == 0) {
1470 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1471 			return -ENOENT;
1472 		}
1473 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1474 		if (err != 0)
1475 			return err;
1476 	} else {
1477 		qdisc_notify(net, skb, n, clid, NULL, q);
1478 	}
1479 	return 0;
1480 }
1481 
1482 /*
1483  * Create/change qdisc.
1484  */
1485 
1486 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1487 			   struct netlink_ext_ack *extack)
1488 {
1489 	struct net *net = sock_net(skb->sk);
1490 	struct tcmsg *tcm;
1491 	struct nlattr *tca[TCA_MAX + 1];
1492 	struct net_device *dev;
1493 	u32 clid;
1494 	struct Qdisc *q, *p;
1495 	int err;
1496 
1497 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1498 		return -EPERM;
1499 
1500 replay:
1501 	/* Reinit, just in case something touches this. */
1502 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
1503 			  extack);
1504 	if (err < 0)
1505 		return err;
1506 
1507 	tcm = nlmsg_data(n);
1508 	clid = tcm->tcm_parent;
1509 	q = p = NULL;
1510 
1511 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1512 	if (!dev)
1513 		return -ENODEV;
1514 
1515 
1516 	if (clid) {
1517 		if (clid != TC_H_ROOT) {
1518 			if (clid != TC_H_INGRESS) {
1519 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1520 				if (!p) {
1521 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1522 					return -ENOENT;
1523 				}
1524 				q = qdisc_leaf(p, clid);
1525 			} else if (dev_ingress_queue_create(dev)) {
1526 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1527 			}
1528 		} else {
1529 			q = dev->qdisc;
1530 		}
1531 
1532 		/* It may be default qdisc, ignore it */
1533 		if (q && q->handle == 0)
1534 			q = NULL;
1535 
1536 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1537 			if (tcm->tcm_handle) {
1538 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1539 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1540 					return -EEXIST;
1541 				}
1542 				if (TC_H_MIN(tcm->tcm_handle)) {
1543 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1544 					return -EINVAL;
1545 				}
1546 				q = qdisc_lookup(dev, tcm->tcm_handle);
1547 				if (!q)
1548 					goto create_n_graft;
1549 				if (n->nlmsg_flags & NLM_F_EXCL) {
1550 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1551 					return -EEXIST;
1552 				}
1553 				if (tca[TCA_KIND] &&
1554 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1555 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1556 					return -EINVAL;
1557 				}
1558 				if (q == p ||
1559 				    (p && check_loop(q, p, 0))) {
1560 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1561 					return -ELOOP;
1562 				}
1563 				qdisc_refcount_inc(q);
1564 				goto graft;
1565 			} else {
1566 				if (!q)
1567 					goto create_n_graft;
1568 
1569 				/* This magic test requires explanation.
1570 				 *
1571 				 *   We know, that some child q is already
1572 				 *   attached to this parent and have choice:
1573 				 *   either to change it or to create/graft new one.
1574 				 *
1575 				 *   1. We are allowed to create/graft only
1576 				 *   if CREATE and REPLACE flags are set.
1577 				 *
1578 				 *   2. If EXCL is set, requestor wanted to say,
1579 				 *   that qdisc tcm_handle is not expected
1580 				 *   to exist, so that we choose create/graft too.
1581 				 *
1582 				 *   3. The last case is when no flags are set.
1583 				 *   Alas, it is sort of hole in API, we
1584 				 *   cannot decide what to do unambiguously.
1585 				 *   For now we select create/graft, if
1586 				 *   user gave KIND, which does not match existing.
1587 				 */
1588 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1589 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1590 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1591 				     (tca[TCA_KIND] &&
1592 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1593 					goto create_n_graft;
1594 			}
1595 		}
1596 	} else {
1597 		if (!tcm->tcm_handle) {
1598 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1599 			return -EINVAL;
1600 		}
1601 		q = qdisc_lookup(dev, tcm->tcm_handle);
1602 	}
1603 
1604 	/* Change qdisc parameters */
1605 	if (!q) {
1606 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1607 		return -ENOENT;
1608 	}
1609 	if (n->nlmsg_flags & NLM_F_EXCL) {
1610 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1611 		return -EEXIST;
1612 	}
1613 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1614 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1615 		return -EINVAL;
1616 	}
1617 	err = qdisc_change(q, tca, extack);
1618 	if (err == 0)
1619 		qdisc_notify(net, skb, n, clid, NULL, q);
1620 	return err;
1621 
1622 create_n_graft:
1623 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1624 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1625 		return -ENOENT;
1626 	}
1627 	if (clid == TC_H_INGRESS) {
1628 		if (dev_ingress_queue(dev)) {
1629 			q = qdisc_create(dev, dev_ingress_queue(dev), p,
1630 					 tcm->tcm_parent, tcm->tcm_parent,
1631 					 tca, &err, extack);
1632 		} else {
1633 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1634 			err = -ENOENT;
1635 		}
1636 	} else {
1637 		struct netdev_queue *dev_queue;
1638 
1639 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1640 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1641 		else if (p)
1642 			dev_queue = p->dev_queue;
1643 		else
1644 			dev_queue = netdev_get_tx_queue(dev, 0);
1645 
1646 		q = qdisc_create(dev, dev_queue, p,
1647 				 tcm->tcm_parent, tcm->tcm_handle,
1648 				 tca, &err, extack);
1649 	}
1650 	if (q == NULL) {
1651 		if (err == -EAGAIN)
1652 			goto replay;
1653 		return err;
1654 	}
1655 
1656 graft:
1657 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1658 	if (err) {
1659 		if (q)
1660 			qdisc_put(q);
1661 		return err;
1662 	}
1663 
1664 	return 0;
1665 }
1666 
1667 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1668 			      struct netlink_callback *cb,
1669 			      int *q_idx_p, int s_q_idx, bool recur,
1670 			      bool dump_invisible)
1671 {
1672 	int ret = 0, q_idx = *q_idx_p;
1673 	struct Qdisc *q;
1674 	int b;
1675 
1676 	if (!root)
1677 		return 0;
1678 
1679 	q = root;
1680 	if (q_idx < s_q_idx) {
1681 		q_idx++;
1682 	} else {
1683 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1684 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1685 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1686 				  RTM_NEWQDISC) <= 0)
1687 			goto done;
1688 		q_idx++;
1689 	}
1690 
1691 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1692 	 * itself has already been dumped.
1693 	 *
1694 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1695 	 * qdisc hashtable, we don't want to hit it again
1696 	 */
1697 	if (!qdisc_dev(root) || !recur)
1698 		goto out;
1699 
1700 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1701 		if (q_idx < s_q_idx) {
1702 			q_idx++;
1703 			continue;
1704 		}
1705 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1706 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1707 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1708 				  RTM_NEWQDISC) <= 0)
1709 			goto done;
1710 		q_idx++;
1711 	}
1712 
1713 out:
1714 	*q_idx_p = q_idx;
1715 	return ret;
1716 done:
1717 	ret = -1;
1718 	goto out;
1719 }
1720 
1721 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1722 {
1723 	struct net *net = sock_net(skb->sk);
1724 	int idx, q_idx;
1725 	int s_idx, s_q_idx;
1726 	struct net_device *dev;
1727 	const struct nlmsghdr *nlh = cb->nlh;
1728 	struct nlattr *tca[TCA_MAX + 1];
1729 	int err;
1730 
1731 	s_idx = cb->args[0];
1732 	s_q_idx = q_idx = cb->args[1];
1733 
1734 	idx = 0;
1735 	ASSERT_RTNL();
1736 
1737 	err = nlmsg_parse(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1738 			  rtm_tca_policy, cb->extack);
1739 	if (err < 0)
1740 		return err;
1741 
1742 	for_each_netdev(net, dev) {
1743 		struct netdev_queue *dev_queue;
1744 
1745 		if (idx < s_idx)
1746 			goto cont;
1747 		if (idx > s_idx)
1748 			s_q_idx = 0;
1749 		q_idx = 0;
1750 
1751 		if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1752 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1753 			goto done;
1754 
1755 		dev_queue = dev_ingress_queue(dev);
1756 		if (dev_queue &&
1757 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1758 				       &q_idx, s_q_idx, false,
1759 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1760 			goto done;
1761 
1762 cont:
1763 		idx++;
1764 	}
1765 
1766 done:
1767 	cb->args[0] = idx;
1768 	cb->args[1] = q_idx;
1769 
1770 	return skb->len;
1771 }
1772 
1773 
1774 
1775 /************************************************
1776  *	Traffic classes manipulation.		*
1777  ************************************************/
1778 
1779 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1780 			  unsigned long cl,
1781 			  u32 portid, u32 seq, u16 flags, int event)
1782 {
1783 	struct tcmsg *tcm;
1784 	struct nlmsghdr  *nlh;
1785 	unsigned char *b = skb_tail_pointer(skb);
1786 	struct gnet_dump d;
1787 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1788 
1789 	cond_resched();
1790 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1791 	if (!nlh)
1792 		goto out_nlmsg_trim;
1793 	tcm = nlmsg_data(nlh);
1794 	tcm->tcm_family = AF_UNSPEC;
1795 	tcm->tcm__pad1 = 0;
1796 	tcm->tcm__pad2 = 0;
1797 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1798 	tcm->tcm_parent = q->handle;
1799 	tcm->tcm_handle = q->handle;
1800 	tcm->tcm_info = 0;
1801 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1802 		goto nla_put_failure;
1803 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1804 		goto nla_put_failure;
1805 
1806 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1807 					 NULL, &d, TCA_PAD) < 0)
1808 		goto nla_put_failure;
1809 
1810 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1811 		goto nla_put_failure;
1812 
1813 	if (gnet_stats_finish_copy(&d) < 0)
1814 		goto nla_put_failure;
1815 
1816 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1817 	return skb->len;
1818 
1819 out_nlmsg_trim:
1820 nla_put_failure:
1821 	nlmsg_trim(skb, b);
1822 	return -1;
1823 }
1824 
1825 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1826 			 struct nlmsghdr *n, struct Qdisc *q,
1827 			 unsigned long cl, int event)
1828 {
1829 	struct sk_buff *skb;
1830 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1831 
1832 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1833 	if (!skb)
1834 		return -ENOBUFS;
1835 
1836 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1837 		kfree_skb(skb);
1838 		return -EINVAL;
1839 	}
1840 
1841 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1842 			      n->nlmsg_flags & NLM_F_ECHO);
1843 }
1844 
1845 static int tclass_del_notify(struct net *net,
1846 			     const struct Qdisc_class_ops *cops,
1847 			     struct sk_buff *oskb, struct nlmsghdr *n,
1848 			     struct Qdisc *q, unsigned long cl)
1849 {
1850 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1851 	struct sk_buff *skb;
1852 	int err = 0;
1853 
1854 	if (!cops->delete)
1855 		return -EOPNOTSUPP;
1856 
1857 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1858 	if (!skb)
1859 		return -ENOBUFS;
1860 
1861 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1862 			   RTM_DELTCLASS) < 0) {
1863 		kfree_skb(skb);
1864 		return -EINVAL;
1865 	}
1866 
1867 	err = cops->delete(q, cl);
1868 	if (err) {
1869 		kfree_skb(skb);
1870 		return err;
1871 	}
1872 
1873 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1874 			      n->nlmsg_flags & NLM_F_ECHO);
1875 }
1876 
1877 #ifdef CONFIG_NET_CLS
1878 
1879 struct tcf_bind_args {
1880 	struct tcf_walker w;
1881 	u32 classid;
1882 	unsigned long cl;
1883 };
1884 
1885 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1886 {
1887 	struct tcf_bind_args *a = (void *)arg;
1888 
1889 	if (tp->ops->bind_class) {
1890 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1891 
1892 		sch_tree_lock(q);
1893 		tp->ops->bind_class(n, a->classid, a->cl);
1894 		sch_tree_unlock(q);
1895 	}
1896 	return 0;
1897 }
1898 
1899 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1900 			   unsigned long new_cl)
1901 {
1902 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1903 	struct tcf_block *block;
1904 	struct tcf_chain *chain;
1905 	unsigned long cl;
1906 
1907 	cl = cops->find(q, portid);
1908 	if (!cl)
1909 		return;
1910 	block = cops->tcf_block(q, cl, NULL);
1911 	if (!block)
1912 		return;
1913 	list_for_each_entry(chain, &block->chain_list, list) {
1914 		struct tcf_proto *tp;
1915 
1916 		for (tp = rtnl_dereference(chain->filter_chain);
1917 		     tp; tp = rtnl_dereference(tp->next)) {
1918 			struct tcf_bind_args arg = {};
1919 
1920 			arg.w.fn = tcf_node_bind;
1921 			arg.classid = clid;
1922 			arg.cl = new_cl;
1923 			tp->ops->walk(tp, &arg.w);
1924 		}
1925 	}
1926 }
1927 
1928 #else
1929 
1930 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1931 			   unsigned long new_cl)
1932 {
1933 }
1934 
1935 #endif
1936 
1937 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1938 			 struct netlink_ext_ack *extack)
1939 {
1940 	struct net *net = sock_net(skb->sk);
1941 	struct tcmsg *tcm = nlmsg_data(n);
1942 	struct nlattr *tca[TCA_MAX + 1];
1943 	struct net_device *dev;
1944 	struct Qdisc *q = NULL;
1945 	const struct Qdisc_class_ops *cops;
1946 	unsigned long cl = 0;
1947 	unsigned long new_cl;
1948 	u32 portid;
1949 	u32 clid;
1950 	u32 qid;
1951 	int err;
1952 
1953 	if ((n->nlmsg_type != RTM_GETTCLASS) &&
1954 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1955 		return -EPERM;
1956 
1957 	err = nlmsg_parse(n, sizeof(*tcm), tca, TCA_MAX, rtm_tca_policy,
1958 			  extack);
1959 	if (err < 0)
1960 		return err;
1961 
1962 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1963 	if (!dev)
1964 		return -ENODEV;
1965 
1966 	/*
1967 	   parent == TC_H_UNSPEC - unspecified parent.
1968 	   parent == TC_H_ROOT   - class is root, which has no parent.
1969 	   parent == X:0	 - parent is root class.
1970 	   parent == X:Y	 - parent is a node in hierarchy.
1971 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
1972 
1973 	   handle == 0:0	 - generate handle from kernel pool.
1974 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
1975 	   handle == X:Y	 - clear.
1976 	   handle == X:0	 - root class.
1977 	 */
1978 
1979 	/* Step 1. Determine qdisc handle X:0 */
1980 
1981 	portid = tcm->tcm_parent;
1982 	clid = tcm->tcm_handle;
1983 	qid = TC_H_MAJ(clid);
1984 
1985 	if (portid != TC_H_ROOT) {
1986 		u32 qid1 = TC_H_MAJ(portid);
1987 
1988 		if (qid && qid1) {
1989 			/* If both majors are known, they must be identical. */
1990 			if (qid != qid1)
1991 				return -EINVAL;
1992 		} else if (qid1) {
1993 			qid = qid1;
1994 		} else if (qid == 0)
1995 			qid = dev->qdisc->handle;
1996 
1997 		/* Now qid is genuine qdisc handle consistent
1998 		 * both with parent and child.
1999 		 *
2000 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2001 		 */
2002 		if (portid)
2003 			portid = TC_H_MAKE(qid, portid);
2004 	} else {
2005 		if (qid == 0)
2006 			qid = dev->qdisc->handle;
2007 	}
2008 
2009 	/* OK. Locate qdisc */
2010 	q = qdisc_lookup(dev, qid);
2011 	if (!q)
2012 		return -ENOENT;
2013 
2014 	/* An check that it supports classes */
2015 	cops = q->ops->cl_ops;
2016 	if (cops == NULL)
2017 		return -EINVAL;
2018 
2019 	/* Now try to get class */
2020 	if (clid == 0) {
2021 		if (portid == TC_H_ROOT)
2022 			clid = qid;
2023 	} else
2024 		clid = TC_H_MAKE(qid, clid);
2025 
2026 	if (clid)
2027 		cl = cops->find(q, clid);
2028 
2029 	if (cl == 0) {
2030 		err = -ENOENT;
2031 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2032 		    !(n->nlmsg_flags & NLM_F_CREATE))
2033 			goto out;
2034 	} else {
2035 		switch (n->nlmsg_type) {
2036 		case RTM_NEWTCLASS:
2037 			err = -EEXIST;
2038 			if (n->nlmsg_flags & NLM_F_EXCL)
2039 				goto out;
2040 			break;
2041 		case RTM_DELTCLASS:
2042 			err = tclass_del_notify(net, cops, skb, n, q, cl);
2043 			/* Unbind the class with flilters with 0 */
2044 			tc_bind_tclass(q, portid, clid, 0);
2045 			goto out;
2046 		case RTM_GETTCLASS:
2047 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
2048 			goto out;
2049 		default:
2050 			err = -EINVAL;
2051 			goto out;
2052 		}
2053 	}
2054 
2055 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2056 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2057 		return -EOPNOTSUPP;
2058 	}
2059 
2060 	new_cl = cl;
2061 	err = -EOPNOTSUPP;
2062 	if (cops->change)
2063 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2064 	if (err == 0) {
2065 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
2066 		/* We just create a new class, need to do reverse binding. */
2067 		if (cl != new_cl)
2068 			tc_bind_tclass(q, portid, clid, new_cl);
2069 	}
2070 out:
2071 	return err;
2072 }
2073 
2074 struct qdisc_dump_args {
2075 	struct qdisc_walker	w;
2076 	struct sk_buff		*skb;
2077 	struct netlink_callback	*cb;
2078 };
2079 
2080 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2081 			    struct qdisc_walker *arg)
2082 {
2083 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2084 
2085 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2086 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2087 			      RTM_NEWTCLASS);
2088 }
2089 
2090 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2091 				struct tcmsg *tcm, struct netlink_callback *cb,
2092 				int *t_p, int s_t)
2093 {
2094 	struct qdisc_dump_args arg;
2095 
2096 	if (tc_qdisc_dump_ignore(q, false) ||
2097 	    *t_p < s_t || !q->ops->cl_ops ||
2098 	    (tcm->tcm_parent &&
2099 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2100 		(*t_p)++;
2101 		return 0;
2102 	}
2103 	if (*t_p > s_t)
2104 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2105 	arg.w.fn = qdisc_class_dump;
2106 	arg.skb = skb;
2107 	arg.cb = cb;
2108 	arg.w.stop  = 0;
2109 	arg.w.skip = cb->args[1];
2110 	arg.w.count = 0;
2111 	q->ops->cl_ops->walk(q, &arg.w);
2112 	cb->args[1] = arg.w.count;
2113 	if (arg.w.stop)
2114 		return -1;
2115 	(*t_p)++;
2116 	return 0;
2117 }
2118 
2119 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2120 			       struct tcmsg *tcm, struct netlink_callback *cb,
2121 			       int *t_p, int s_t)
2122 {
2123 	struct Qdisc *q;
2124 	int b;
2125 
2126 	if (!root)
2127 		return 0;
2128 
2129 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2130 		return -1;
2131 
2132 	if (!qdisc_dev(root))
2133 		return 0;
2134 
2135 	if (tcm->tcm_parent) {
2136 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2137 		if (q && q != root &&
2138 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2139 			return -1;
2140 		return 0;
2141 	}
2142 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2143 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2144 			return -1;
2145 	}
2146 
2147 	return 0;
2148 }
2149 
2150 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2151 {
2152 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2153 	struct net *net = sock_net(skb->sk);
2154 	struct netdev_queue *dev_queue;
2155 	struct net_device *dev;
2156 	int t, s_t;
2157 
2158 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2159 		return 0;
2160 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2161 	if (!dev)
2162 		return 0;
2163 
2164 	s_t = cb->args[0];
2165 	t = 0;
2166 
2167 	if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
2168 		goto done;
2169 
2170 	dev_queue = dev_ingress_queue(dev);
2171 	if (dev_queue &&
2172 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2173 				&t, s_t) < 0)
2174 		goto done;
2175 
2176 done:
2177 	cb->args[0] = t;
2178 
2179 	dev_put(dev);
2180 	return skb->len;
2181 }
2182 
2183 #ifdef CONFIG_PROC_FS
2184 static int psched_show(struct seq_file *seq, void *v)
2185 {
2186 	seq_printf(seq, "%08x %08x %08x %08x\n",
2187 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2188 		   1000000,
2189 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2190 
2191 	return 0;
2192 }
2193 
2194 static int __net_init psched_net_init(struct net *net)
2195 {
2196 	struct proc_dir_entry *e;
2197 
2198 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2199 	if (e == NULL)
2200 		return -ENOMEM;
2201 
2202 	return 0;
2203 }
2204 
2205 static void __net_exit psched_net_exit(struct net *net)
2206 {
2207 	remove_proc_entry("psched", net->proc_net);
2208 }
2209 #else
2210 static int __net_init psched_net_init(struct net *net)
2211 {
2212 	return 0;
2213 }
2214 
2215 static void __net_exit psched_net_exit(struct net *net)
2216 {
2217 }
2218 #endif
2219 
2220 static struct pernet_operations psched_net_ops = {
2221 	.init = psched_net_init,
2222 	.exit = psched_net_exit,
2223 };
2224 
2225 static int __init pktsched_init(void)
2226 {
2227 	int err;
2228 
2229 	err = register_pernet_subsys(&psched_net_ops);
2230 	if (err) {
2231 		pr_err("pktsched_init: "
2232 		       "cannot initialize per netns operations\n");
2233 		return err;
2234 	}
2235 
2236 	register_qdisc(&pfifo_fast_ops);
2237 	register_qdisc(&pfifo_qdisc_ops);
2238 	register_qdisc(&bfifo_qdisc_ops);
2239 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2240 	register_qdisc(&mq_qdisc_ops);
2241 	register_qdisc(&noqueue_qdisc_ops);
2242 
2243 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2244 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2245 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2246 		      0);
2247 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2248 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2249 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2250 		      0);
2251 
2252 	return 0;
2253 }
2254 
2255 subsys_initcall(pktsched_init);
2256