xref: /linux/net/sched/sch_api.c (revision 067012974c8ae31a8886046df082aeba93592972)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/sch_api.c	Packet scheduler API.
4  *
5  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6  *
7  * Fixes:
8  *
9  * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired.
10  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
11  * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support
12  */
13 
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <linux/init.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/kmod.h>
24 #include <linux/list.h>
25 #include <linux/hrtimer.h>
26 #include <linux/slab.h>
27 #include <linux/hashtable.h>
28 
29 #include <net/net_namespace.h>
30 #include <net/sock.h>
31 #include <net/netlink.h>
32 #include <net/pkt_sched.h>
33 #include <net/pkt_cls.h>
34 
35 #include <trace/events/qdisc.h>
36 
37 /*
38 
39    Short review.
40    -------------
41 
42    This file consists of two interrelated parts:
43 
44    1. queueing disciplines manager frontend.
45    2. traffic classes manager frontend.
46 
47    Generally, queueing discipline ("qdisc") is a black box,
48    which is able to enqueue packets and to dequeue them (when
49    device is ready to send something) in order and at times
50    determined by algorithm hidden in it.
51 
52    qdisc's are divided to two categories:
53    - "queues", which have no internal structure visible from outside.
54    - "schedulers", which split all the packets to "traffic classes",
55      using "packet classifiers" (look at cls_api.c)
56 
57    In turn, classes may have child qdiscs (as rule, queues)
58    attached to them etc. etc. etc.
59 
60    The goal of the routines in this file is to translate
61    information supplied by user in the form of handles
62    to more intelligible for kernel form, to make some sanity
63    checks and part of work, which is common to all qdiscs
64    and to provide rtnetlink notifications.
65 
66    All real intelligent work is done inside qdisc modules.
67 
68 
69 
70    Every discipline has two major routines: enqueue and dequeue.
71 
72    ---dequeue
73 
74    dequeue usually returns a skb to send. It is allowed to return NULL,
75    but it does not mean that queue is empty, it just means that
76    discipline does not want to send anything this time.
77    Queue is really empty if q->q.qlen == 0.
78    For complicated disciplines with multiple queues q->q is not
79    real packet queue, but however q->q.qlen must be valid.
80 
81    ---enqueue
82 
83    enqueue returns 0, if packet was enqueued successfully.
84    If packet (this one or another one) was dropped, it returns
85    not zero error code.
86    NET_XMIT_DROP 	- this packet dropped
87      Expected action: do not backoff, but wait until queue will clear.
88    NET_XMIT_CN	 	- probably this packet enqueued, but another one dropped.
89      Expected action: backoff or ignore
90 
91    Auxiliary routines:
92 
93    ---peek
94 
95    like dequeue but without removing a packet from the queue
96 
97    ---reset
98 
99    returns qdisc to initial state: purge all buffers, clear all
100    timers, counters (except for statistics) etc.
101 
102    ---init
103 
104    initializes newly created qdisc.
105 
106    ---destroy
107 
108    destroys resources allocated by init and during lifetime of qdisc.
109 
110    ---change
111 
112    changes qdisc parameters.
113  */
114 
115 /* Protects list of registered TC modules. It is pure SMP lock. */
116 static DEFINE_RWLOCK(qdisc_mod_lock);
117 
118 
119 /************************************************
120  *	Queueing disciplines manipulation.	*
121  ************************************************/
122 
123 
124 /* The list of all installed queueing disciplines. */
125 
126 static struct Qdisc_ops *qdisc_base;
127 
128 /* Register/unregister queueing discipline */
129 
130 int register_qdisc(struct Qdisc_ops *qops)
131 {
132 	struct Qdisc_ops *q, **qp;
133 	int rc = -EEXIST;
134 
135 	write_lock(&qdisc_mod_lock);
136 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
137 		if (!strcmp(qops->id, q->id))
138 			goto out;
139 
140 	if (qops->enqueue == NULL)
141 		qops->enqueue = noop_qdisc_ops.enqueue;
142 	if (qops->peek == NULL) {
143 		if (qops->dequeue == NULL)
144 			qops->peek = noop_qdisc_ops.peek;
145 		else
146 			goto out_einval;
147 	}
148 	if (qops->dequeue == NULL)
149 		qops->dequeue = noop_qdisc_ops.dequeue;
150 
151 	if (qops->cl_ops) {
152 		const struct Qdisc_class_ops *cops = qops->cl_ops;
153 
154 		if (!(cops->find && cops->walk && cops->leaf))
155 			goto out_einval;
156 
157 		if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
158 			goto out_einval;
159 	}
160 
161 	qops->next = NULL;
162 	*qp = qops;
163 	rc = 0;
164 out:
165 	write_unlock(&qdisc_mod_lock);
166 	return rc;
167 
168 out_einval:
169 	rc = -EINVAL;
170 	goto out;
171 }
172 EXPORT_SYMBOL(register_qdisc);
173 
174 int unregister_qdisc(struct Qdisc_ops *qops)
175 {
176 	struct Qdisc_ops *q, **qp;
177 	int err = -ENOENT;
178 
179 	write_lock(&qdisc_mod_lock);
180 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
181 		if (q == qops)
182 			break;
183 	if (q) {
184 		*qp = q->next;
185 		q->next = NULL;
186 		err = 0;
187 	}
188 	write_unlock(&qdisc_mod_lock);
189 	return err;
190 }
191 EXPORT_SYMBOL(unregister_qdisc);
192 
193 /* Get default qdisc if not otherwise specified */
194 void qdisc_get_default(char *name, size_t len)
195 {
196 	read_lock(&qdisc_mod_lock);
197 	strlcpy(name, default_qdisc_ops->id, len);
198 	read_unlock(&qdisc_mod_lock);
199 }
200 
201 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
202 {
203 	struct Qdisc_ops *q = NULL;
204 
205 	for (q = qdisc_base; q; q = q->next) {
206 		if (!strcmp(name, q->id)) {
207 			if (!try_module_get(q->owner))
208 				q = NULL;
209 			break;
210 		}
211 	}
212 
213 	return q;
214 }
215 
216 /* Set new default qdisc to use */
217 int qdisc_set_default(const char *name)
218 {
219 	const struct Qdisc_ops *ops;
220 
221 	if (!capable(CAP_NET_ADMIN))
222 		return -EPERM;
223 
224 	write_lock(&qdisc_mod_lock);
225 	ops = qdisc_lookup_default(name);
226 	if (!ops) {
227 		/* Not found, drop lock and try to load module */
228 		write_unlock(&qdisc_mod_lock);
229 		request_module("sch_%s", name);
230 		write_lock(&qdisc_mod_lock);
231 
232 		ops = qdisc_lookup_default(name);
233 	}
234 
235 	if (ops) {
236 		/* Set new default */
237 		module_put(default_qdisc_ops->owner);
238 		default_qdisc_ops = ops;
239 	}
240 	write_unlock(&qdisc_mod_lock);
241 
242 	return ops ? 0 : -ENOENT;
243 }
244 
245 #ifdef CONFIG_NET_SCH_DEFAULT
246 /* Set default value from kernel config */
247 static int __init sch_default_qdisc(void)
248 {
249 	return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
250 }
251 late_initcall(sch_default_qdisc);
252 #endif
253 
254 /* We know handle. Find qdisc among all qdisc's attached to device
255  * (root qdisc, all its children, children of children etc.)
256  * Note: caller either uses rtnl or rcu_read_lock()
257  */
258 
259 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
260 {
261 	struct Qdisc *q;
262 
263 	if (!qdisc_dev(root))
264 		return (root->handle == handle ? root : NULL);
265 
266 	if (!(root->flags & TCQ_F_BUILTIN) &&
267 	    root->handle == handle)
268 		return root;
269 
270 	hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle,
271 				   lockdep_rtnl_is_held()) {
272 		if (q->handle == handle)
273 			return q;
274 	}
275 	return NULL;
276 }
277 
278 void qdisc_hash_add(struct Qdisc *q, bool invisible)
279 {
280 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
281 		ASSERT_RTNL();
282 		hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
283 		if (invisible)
284 			q->flags |= TCQ_F_INVISIBLE;
285 	}
286 }
287 EXPORT_SYMBOL(qdisc_hash_add);
288 
289 void qdisc_hash_del(struct Qdisc *q)
290 {
291 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
292 		ASSERT_RTNL();
293 		hash_del_rcu(&q->hash);
294 	}
295 }
296 EXPORT_SYMBOL(qdisc_hash_del);
297 
298 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
299 {
300 	struct Qdisc *q;
301 
302 	if (!handle)
303 		return NULL;
304 	q = qdisc_match_from_root(dev->qdisc, handle);
305 	if (q)
306 		goto out;
307 
308 	if (dev_ingress_queue(dev))
309 		q = qdisc_match_from_root(
310 			dev_ingress_queue(dev)->qdisc_sleeping,
311 			handle);
312 out:
313 	return q;
314 }
315 
316 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
317 {
318 	struct netdev_queue *nq;
319 	struct Qdisc *q;
320 
321 	if (!handle)
322 		return NULL;
323 	q = qdisc_match_from_root(dev->qdisc, handle);
324 	if (q)
325 		goto out;
326 
327 	nq = dev_ingress_queue_rcu(dev);
328 	if (nq)
329 		q = qdisc_match_from_root(nq->qdisc_sleeping, handle);
330 out:
331 	return q;
332 }
333 
334 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
335 {
336 	unsigned long cl;
337 	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
338 
339 	if (cops == NULL)
340 		return NULL;
341 	cl = cops->find(p, classid);
342 
343 	if (cl == 0)
344 		return NULL;
345 	return cops->leaf(p, cl);
346 }
347 
348 /* Find queueing discipline by name */
349 
350 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
351 {
352 	struct Qdisc_ops *q = NULL;
353 
354 	if (kind) {
355 		read_lock(&qdisc_mod_lock);
356 		for (q = qdisc_base; q; q = q->next) {
357 			if (nla_strcmp(kind, q->id) == 0) {
358 				if (!try_module_get(q->owner))
359 					q = NULL;
360 				break;
361 			}
362 		}
363 		read_unlock(&qdisc_mod_lock);
364 	}
365 	return q;
366 }
367 
368 /* The linklayer setting were not transferred from iproute2, in older
369  * versions, and the rate tables lookup systems have been dropped in
370  * the kernel. To keep backward compatible with older iproute2 tc
371  * utils, we detect the linklayer setting by detecting if the rate
372  * table were modified.
373  *
374  * For linklayer ATM table entries, the rate table will be aligned to
375  * 48 bytes, thus some table entries will contain the same value.  The
376  * mpu (min packet unit) is also encoded into the old rate table, thus
377  * starting from the mpu, we find low and high table entries for
378  * mapping this cell.  If these entries contain the same value, when
379  * the rate tables have been modified for linklayer ATM.
380  *
381  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
382  * and then roundup to the next cell, calc the table entry one below,
383  * and compare.
384  */
385 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
386 {
387 	int low       = roundup(r->mpu, 48);
388 	int high      = roundup(low+1, 48);
389 	int cell_low  = low >> r->cell_log;
390 	int cell_high = (high >> r->cell_log) - 1;
391 
392 	/* rtab is too inaccurate at rates > 100Mbit/s */
393 	if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
394 		pr_debug("TC linklayer: Giving up ATM detection\n");
395 		return TC_LINKLAYER_ETHERNET;
396 	}
397 
398 	if ((cell_high > cell_low) && (cell_high < 256)
399 	    && (rtab[cell_low] == rtab[cell_high])) {
400 		pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
401 			 cell_low, cell_high, rtab[cell_high]);
402 		return TC_LINKLAYER_ATM;
403 	}
404 	return TC_LINKLAYER_ETHERNET;
405 }
406 
407 static struct qdisc_rate_table *qdisc_rtab_list;
408 
409 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
410 					struct nlattr *tab,
411 					struct netlink_ext_ack *extack)
412 {
413 	struct qdisc_rate_table *rtab;
414 
415 	if (tab == NULL || r->rate == 0 ||
416 	    r->cell_log == 0 || r->cell_log >= 32 ||
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_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
483 					  extack);
484 	if (err < 0)
485 		return ERR_PTR(err);
486 	if (!tb[TCA_STAB_BASE]) {
487 		NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
488 		return ERR_PTR(-EINVAL);
489 	}
490 
491 	s = nla_data(tb[TCA_STAB_BASE]);
492 
493 	if (s->tsize > 0) {
494 		if (!tb[TCA_STAB_DATA]) {
495 			NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
496 			return ERR_PTR(-EINVAL);
497 		}
498 		tab = nla_data(tb[TCA_STAB_DATA]);
499 		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
500 	}
501 
502 	if (tsize != s->tsize || (!tab && tsize > 0)) {
503 		NL_SET_ERR_MSG(extack, "Invalid size of size table");
504 		return ERR_PTR(-EINVAL);
505 	}
506 
507 	list_for_each_entry(stab, &qdisc_stab_list, list) {
508 		if (memcmp(&stab->szopts, s, sizeof(*s)))
509 			continue;
510 		if (tsize > 0 && memcmp(stab->data, tab, tsize * sizeof(u16)))
511 			continue;
512 		stab->refcnt++;
513 		return stab;
514 	}
515 
516 	stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
517 	if (!stab)
518 		return ERR_PTR(-ENOMEM);
519 
520 	stab->refcnt = 1;
521 	stab->szopts = *s;
522 	if (tsize > 0)
523 		memcpy(stab->data, tab, tsize * sizeof(u16));
524 
525 	list_add_tail(&stab->list, &qdisc_stab_list);
526 
527 	return stab;
528 }
529 
530 void qdisc_put_stab(struct qdisc_size_table *tab)
531 {
532 	if (!tab)
533 		return;
534 
535 	if (--tab->refcnt == 0) {
536 		list_del(&tab->list);
537 		kfree_rcu(tab, rcu);
538 	}
539 }
540 EXPORT_SYMBOL(qdisc_put_stab);
541 
542 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
543 {
544 	struct nlattr *nest;
545 
546 	nest = nla_nest_start_noflag(skb, TCA_STAB);
547 	if (nest == NULL)
548 		goto nla_put_failure;
549 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
550 		goto nla_put_failure;
551 	nla_nest_end(skb, nest);
552 
553 	return skb->len;
554 
555 nla_put_failure:
556 	return -1;
557 }
558 
559 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
560 			       const struct qdisc_size_table *stab)
561 {
562 	int pkt_len, slot;
563 
564 	pkt_len = skb->len + stab->szopts.overhead;
565 	if (unlikely(!stab->szopts.tsize))
566 		goto out;
567 
568 	slot = pkt_len + stab->szopts.cell_align;
569 	if (unlikely(slot < 0))
570 		slot = 0;
571 
572 	slot >>= stab->szopts.cell_log;
573 	if (likely(slot < stab->szopts.tsize))
574 		pkt_len = stab->data[slot];
575 	else
576 		pkt_len = stab->data[stab->szopts.tsize - 1] *
577 				(slot / stab->szopts.tsize) +
578 				stab->data[slot % stab->szopts.tsize];
579 
580 	pkt_len <<= stab->szopts.size_log;
581 out:
582 	if (unlikely(pkt_len < 1))
583 		pkt_len = 1;
584 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
585 }
586 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
587 
588 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
589 {
590 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
591 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
592 			txt, qdisc->ops->id, qdisc->handle >> 16);
593 		qdisc->flags |= TCQ_F_WARN_NONWC;
594 	}
595 }
596 EXPORT_SYMBOL(qdisc_warn_nonwc);
597 
598 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
599 {
600 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
601 						 timer);
602 
603 	rcu_read_lock();
604 	__netif_schedule(qdisc_root(wd->qdisc));
605 	rcu_read_unlock();
606 
607 	return HRTIMER_NORESTART;
608 }
609 
610 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
611 				 clockid_t clockid)
612 {
613 	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
614 	wd->timer.function = qdisc_watchdog;
615 	wd->qdisc = qdisc;
616 }
617 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
618 
619 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
620 {
621 	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
622 }
623 EXPORT_SYMBOL(qdisc_watchdog_init);
624 
625 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
626 				      u64 delta_ns)
627 {
628 	if (test_bit(__QDISC_STATE_DEACTIVATED,
629 		     &qdisc_root_sleeping(wd->qdisc)->state))
630 		return;
631 
632 	if (hrtimer_is_queued(&wd->timer)) {
633 		/* If timer is already set in [expires, expires + delta_ns],
634 		 * do not reprogram it.
635 		 */
636 		if (wd->last_expires - expires <= delta_ns)
637 			return;
638 	}
639 
640 	wd->last_expires = expires;
641 	hrtimer_start_range_ns(&wd->timer,
642 			       ns_to_ktime(expires),
643 			       delta_ns,
644 			       HRTIMER_MODE_ABS_PINNED);
645 }
646 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
647 
648 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
649 {
650 	hrtimer_cancel(&wd->timer);
651 }
652 EXPORT_SYMBOL(qdisc_watchdog_cancel);
653 
654 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
655 {
656 	struct hlist_head *h;
657 	unsigned int i;
658 
659 	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
660 
661 	if (h != NULL) {
662 		for (i = 0; i < n; i++)
663 			INIT_HLIST_HEAD(&h[i]);
664 	}
665 	return h;
666 }
667 
668 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
669 {
670 	struct Qdisc_class_common *cl;
671 	struct hlist_node *next;
672 	struct hlist_head *nhash, *ohash;
673 	unsigned int nsize, nmask, osize;
674 	unsigned int i, h;
675 
676 	/* Rehash when load factor exceeds 0.75 */
677 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
678 		return;
679 	nsize = clhash->hashsize * 2;
680 	nmask = nsize - 1;
681 	nhash = qdisc_class_hash_alloc(nsize);
682 	if (nhash == NULL)
683 		return;
684 
685 	ohash = clhash->hash;
686 	osize = clhash->hashsize;
687 
688 	sch_tree_lock(sch);
689 	for (i = 0; i < osize; i++) {
690 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
691 			h = qdisc_class_hash(cl->classid, nmask);
692 			hlist_add_head(&cl->hnode, &nhash[h]);
693 		}
694 	}
695 	clhash->hash     = nhash;
696 	clhash->hashsize = nsize;
697 	clhash->hashmask = nmask;
698 	sch_tree_unlock(sch);
699 
700 	kvfree(ohash);
701 }
702 EXPORT_SYMBOL(qdisc_class_hash_grow);
703 
704 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
705 {
706 	unsigned int size = 4;
707 
708 	clhash->hash = qdisc_class_hash_alloc(size);
709 	if (!clhash->hash)
710 		return -ENOMEM;
711 	clhash->hashsize  = size;
712 	clhash->hashmask  = size - 1;
713 	clhash->hashelems = 0;
714 	return 0;
715 }
716 EXPORT_SYMBOL(qdisc_class_hash_init);
717 
718 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
719 {
720 	kvfree(clhash->hash);
721 }
722 EXPORT_SYMBOL(qdisc_class_hash_destroy);
723 
724 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
725 			     struct Qdisc_class_common *cl)
726 {
727 	unsigned int h;
728 
729 	INIT_HLIST_NODE(&cl->hnode);
730 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
731 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
732 	clhash->hashelems++;
733 }
734 EXPORT_SYMBOL(qdisc_class_hash_insert);
735 
736 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
737 			     struct Qdisc_class_common *cl)
738 {
739 	hlist_del(&cl->hnode);
740 	clhash->hashelems--;
741 }
742 EXPORT_SYMBOL(qdisc_class_hash_remove);
743 
744 /* Allocate an unique handle from space managed by kernel
745  * Possible range is [8000-FFFF]:0000 (0x8000 values)
746  */
747 static u32 qdisc_alloc_handle(struct net_device *dev)
748 {
749 	int i = 0x8000;
750 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
751 
752 	do {
753 		autohandle += TC_H_MAKE(0x10000U, 0);
754 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
755 			autohandle = TC_H_MAKE(0x80000000U, 0);
756 		if (!qdisc_lookup(dev, autohandle))
757 			return autohandle;
758 		cond_resched();
759 	} while	(--i > 0);
760 
761 	return 0;
762 }
763 
764 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
765 {
766 	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
767 	const struct Qdisc_class_ops *cops;
768 	unsigned long cl;
769 	u32 parentid;
770 	bool notify;
771 	int drops;
772 
773 	if (n == 0 && len == 0)
774 		return;
775 	drops = max_t(int, n, 0);
776 	rcu_read_lock();
777 	while ((parentid = sch->parent)) {
778 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
779 			break;
780 
781 		if (sch->flags & TCQ_F_NOPARENT)
782 			break;
783 		/* Notify parent qdisc only if child qdisc becomes empty.
784 		 *
785 		 * If child was empty even before update then backlog
786 		 * counter is screwed and we skip notification because
787 		 * parent class is already passive.
788 		 *
789 		 * If the original child was offloaded then it is allowed
790 		 * to be seem as empty, so the parent is notified anyway.
791 		 */
792 		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
793 						       !qdisc_is_offloaded);
794 		/* TODO: perform the search on a per txq basis */
795 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
796 		if (sch == NULL) {
797 			WARN_ON_ONCE(parentid != TC_H_ROOT);
798 			break;
799 		}
800 		cops = sch->ops->cl_ops;
801 		if (notify && cops->qlen_notify) {
802 			cl = cops->find(sch, parentid);
803 			cops->qlen_notify(sch, cl);
804 		}
805 		sch->q.qlen -= n;
806 		sch->qstats.backlog -= len;
807 		__qdisc_qstats_drop(sch, drops);
808 	}
809 	rcu_read_unlock();
810 }
811 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
812 
813 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
814 			      void *type_data)
815 {
816 	struct net_device *dev = qdisc_dev(sch);
817 	int err;
818 
819 	sch->flags &= ~TCQ_F_OFFLOADED;
820 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
821 		return 0;
822 
823 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
824 	if (err == -EOPNOTSUPP)
825 		return 0;
826 
827 	if (!err)
828 		sch->flags |= TCQ_F_OFFLOADED;
829 
830 	return err;
831 }
832 EXPORT_SYMBOL(qdisc_offload_dump_helper);
833 
834 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
835 				struct Qdisc *new, struct Qdisc *old,
836 				enum tc_setup_type type, void *type_data,
837 				struct netlink_ext_ack *extack)
838 {
839 	bool any_qdisc_is_offloaded;
840 	int err;
841 
842 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
843 		return;
844 
845 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
846 
847 	/* Don't report error if the graft is part of destroy operation. */
848 	if (!err || !new || new == &noop_qdisc)
849 		return;
850 
851 	/* Don't report error if the parent, the old child and the new
852 	 * one are not offloaded.
853 	 */
854 	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
855 	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
856 	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
857 
858 	if (any_qdisc_is_offloaded)
859 		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
860 }
861 EXPORT_SYMBOL(qdisc_offload_graft_helper);
862 
863 static void qdisc_offload_graft_root(struct net_device *dev,
864 				     struct Qdisc *new, struct Qdisc *old,
865 				     struct netlink_ext_ack *extack)
866 {
867 	struct tc_root_qopt_offload graft_offload = {
868 		.command	= TC_ROOT_GRAFT,
869 		.handle		= new ? new->handle : 0,
870 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
871 				  (old && old->flags & TCQ_F_INGRESS),
872 	};
873 
874 	qdisc_offload_graft_helper(dev, NULL, new, old,
875 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
876 }
877 
878 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
879 			 u32 portid, u32 seq, u16 flags, int event)
880 {
881 	struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
882 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
883 	struct tcmsg *tcm;
884 	struct nlmsghdr  *nlh;
885 	unsigned char *b = skb_tail_pointer(skb);
886 	struct gnet_dump d;
887 	struct qdisc_size_table *stab;
888 	u32 block_index;
889 	__u32 qlen;
890 
891 	cond_resched();
892 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
893 	if (!nlh)
894 		goto out_nlmsg_trim;
895 	tcm = nlmsg_data(nlh);
896 	tcm->tcm_family = AF_UNSPEC;
897 	tcm->tcm__pad1 = 0;
898 	tcm->tcm__pad2 = 0;
899 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
900 	tcm->tcm_parent = clid;
901 	tcm->tcm_handle = q->handle;
902 	tcm->tcm_info = refcount_read(&q->refcnt);
903 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
904 		goto nla_put_failure;
905 	if (q->ops->ingress_block_get) {
906 		block_index = q->ops->ingress_block_get(q);
907 		if (block_index &&
908 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
909 			goto nla_put_failure;
910 	}
911 	if (q->ops->egress_block_get) {
912 		block_index = q->ops->egress_block_get(q);
913 		if (block_index &&
914 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
915 			goto nla_put_failure;
916 	}
917 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
918 		goto nla_put_failure;
919 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
920 		goto nla_put_failure;
921 	qlen = qdisc_qlen_sum(q);
922 
923 	stab = rtnl_dereference(q->stab);
924 	if (stab && qdisc_dump_stab(skb, stab) < 0)
925 		goto nla_put_failure;
926 
927 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
928 					 NULL, &d, TCA_PAD) < 0)
929 		goto nla_put_failure;
930 
931 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
932 		goto nla_put_failure;
933 
934 	if (qdisc_is_percpu_stats(q)) {
935 		cpu_bstats = q->cpu_bstats;
936 		cpu_qstats = q->cpu_qstats;
937 	}
938 
939 	if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
940 				  &d, cpu_bstats, &q->bstats) < 0 ||
941 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
942 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
943 		goto nla_put_failure;
944 
945 	if (gnet_stats_finish_copy(&d) < 0)
946 		goto nla_put_failure;
947 
948 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
949 	return skb->len;
950 
951 out_nlmsg_trim:
952 nla_put_failure:
953 	nlmsg_trim(skb, b);
954 	return -1;
955 }
956 
957 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
958 {
959 	if (q->flags & TCQ_F_BUILTIN)
960 		return true;
961 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
962 		return true;
963 
964 	return false;
965 }
966 
967 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
968 			struct nlmsghdr *n, u32 clid,
969 			struct Qdisc *old, struct Qdisc *new)
970 {
971 	struct sk_buff *skb;
972 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
973 
974 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
975 	if (!skb)
976 		return -ENOBUFS;
977 
978 	if (old && !tc_qdisc_dump_ignore(old, false)) {
979 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
980 				  0, RTM_DELQDISC) < 0)
981 			goto err_out;
982 	}
983 	if (new && !tc_qdisc_dump_ignore(new, false)) {
984 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
985 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
986 			goto err_out;
987 	}
988 
989 	if (skb->len)
990 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
991 				      n->nlmsg_flags & NLM_F_ECHO);
992 
993 err_out:
994 	kfree_skb(skb);
995 	return -EINVAL;
996 }
997 
998 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
999 			       struct nlmsghdr *n, u32 clid,
1000 			       struct Qdisc *old, struct Qdisc *new)
1001 {
1002 	if (new || old)
1003 		qdisc_notify(net, skb, n, clid, old, new);
1004 
1005 	if (old)
1006 		qdisc_put(old);
1007 }
1008 
1009 static void qdisc_clear_nolock(struct Qdisc *sch)
1010 {
1011 	sch->flags &= ~TCQ_F_NOLOCK;
1012 	if (!(sch->flags & TCQ_F_CPUSTATS))
1013 		return;
1014 
1015 	free_percpu(sch->cpu_bstats);
1016 	free_percpu(sch->cpu_qstats);
1017 	sch->cpu_bstats = NULL;
1018 	sch->cpu_qstats = NULL;
1019 	sch->flags &= ~TCQ_F_CPUSTATS;
1020 }
1021 
1022 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1023  * to device "dev".
1024  *
1025  * When appropriate send a netlink notification using 'skb'
1026  * and "n".
1027  *
1028  * On success, destroy old qdisc.
1029  */
1030 
1031 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1032 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1033 		       struct Qdisc *new, struct Qdisc *old,
1034 		       struct netlink_ext_ack *extack)
1035 {
1036 	struct Qdisc *q = old;
1037 	struct net *net = dev_net(dev);
1038 
1039 	if (parent == NULL) {
1040 		unsigned int i, num_q, ingress;
1041 
1042 		ingress = 0;
1043 		num_q = dev->num_tx_queues;
1044 		if ((q && q->flags & TCQ_F_INGRESS) ||
1045 		    (new && new->flags & TCQ_F_INGRESS)) {
1046 			num_q = 1;
1047 			ingress = 1;
1048 			if (!dev_ingress_queue(dev)) {
1049 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1050 				return -ENOENT;
1051 			}
1052 		}
1053 
1054 		if (dev->flags & IFF_UP)
1055 			dev_deactivate(dev);
1056 
1057 		qdisc_offload_graft_root(dev, new, old, extack);
1058 
1059 		if (new && new->ops->attach)
1060 			goto skip;
1061 
1062 		for (i = 0; i < num_q; i++) {
1063 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1064 
1065 			if (!ingress)
1066 				dev_queue = netdev_get_tx_queue(dev, i);
1067 
1068 			old = dev_graft_qdisc(dev_queue, new);
1069 			if (new && i > 0)
1070 				qdisc_refcount_inc(new);
1071 
1072 			if (!ingress)
1073 				qdisc_put(old);
1074 		}
1075 
1076 skip:
1077 		if (!ingress) {
1078 			notify_and_destroy(net, skb, n, classid,
1079 					   dev->qdisc, new);
1080 			if (new && !new->ops->attach)
1081 				qdisc_refcount_inc(new);
1082 			dev->qdisc = new ? : &noop_qdisc;
1083 
1084 			if (new && new->ops->attach)
1085 				new->ops->attach(new);
1086 		} else {
1087 			notify_and_destroy(net, skb, n, classid, old, new);
1088 		}
1089 
1090 		if (dev->flags & IFF_UP)
1091 			dev_activate(dev);
1092 	} else {
1093 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1094 		unsigned long cl;
1095 		int err;
1096 
1097 		/* Only support running class lockless if parent is lockless */
1098 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1099 			qdisc_clear_nolock(new);
1100 
1101 		if (!cops || !cops->graft)
1102 			return -EOPNOTSUPP;
1103 
1104 		cl = cops->find(parent, classid);
1105 		if (!cl) {
1106 			NL_SET_ERR_MSG(extack, "Specified class not found");
1107 			return -ENOENT;
1108 		}
1109 
1110 		err = cops->graft(parent, cl, new, &old, extack);
1111 		if (err)
1112 			return err;
1113 		notify_and_destroy(net, skb, n, classid, old, new);
1114 	}
1115 	return 0;
1116 }
1117 
1118 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1119 				   struct netlink_ext_ack *extack)
1120 {
1121 	u32 block_index;
1122 
1123 	if (tca[TCA_INGRESS_BLOCK]) {
1124 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1125 
1126 		if (!block_index) {
1127 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1128 			return -EINVAL;
1129 		}
1130 		if (!sch->ops->ingress_block_set) {
1131 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1132 			return -EOPNOTSUPP;
1133 		}
1134 		sch->ops->ingress_block_set(sch, block_index);
1135 	}
1136 	if (tca[TCA_EGRESS_BLOCK]) {
1137 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1138 
1139 		if (!block_index) {
1140 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1141 			return -EINVAL;
1142 		}
1143 		if (!sch->ops->egress_block_set) {
1144 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1145 			return -EOPNOTSUPP;
1146 		}
1147 		sch->ops->egress_block_set(sch, block_index);
1148 	}
1149 	return 0;
1150 }
1151 
1152 /*
1153    Allocate and initialize new qdisc.
1154 
1155    Parameters are passed via opt.
1156  */
1157 
1158 static struct Qdisc *qdisc_create(struct net_device *dev,
1159 				  struct netdev_queue *dev_queue,
1160 				  struct Qdisc *p, u32 parent, u32 handle,
1161 				  struct nlattr **tca, int *errp,
1162 				  struct netlink_ext_ack *extack)
1163 {
1164 	int err;
1165 	struct nlattr *kind = tca[TCA_KIND];
1166 	struct Qdisc *sch;
1167 	struct Qdisc_ops *ops;
1168 	struct qdisc_size_table *stab;
1169 
1170 	ops = qdisc_lookup_ops(kind);
1171 #ifdef CONFIG_MODULES
1172 	if (ops == NULL && kind != NULL) {
1173 		char name[IFNAMSIZ];
1174 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1175 			/* We dropped the RTNL semaphore in order to
1176 			 * perform the module load.  So, even if we
1177 			 * succeeded in loading the module we have to
1178 			 * tell the caller to replay the request.  We
1179 			 * indicate this using -EAGAIN.
1180 			 * We replay the request because the device may
1181 			 * go away in the mean time.
1182 			 */
1183 			rtnl_unlock();
1184 			request_module("sch_%s", name);
1185 			rtnl_lock();
1186 			ops = qdisc_lookup_ops(kind);
1187 			if (ops != NULL) {
1188 				/* We will try again qdisc_lookup_ops,
1189 				 * so don't keep a reference.
1190 				 */
1191 				module_put(ops->owner);
1192 				err = -EAGAIN;
1193 				goto err_out;
1194 			}
1195 		}
1196 	}
1197 #endif
1198 
1199 	err = -ENOENT;
1200 	if (!ops) {
1201 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1202 		goto err_out;
1203 	}
1204 
1205 	sch = qdisc_alloc(dev_queue, ops, extack);
1206 	if (IS_ERR(sch)) {
1207 		err = PTR_ERR(sch);
1208 		goto err_out2;
1209 	}
1210 
1211 	sch->parent = parent;
1212 
1213 	if (handle == TC_H_INGRESS) {
1214 		sch->flags |= TCQ_F_INGRESS;
1215 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1216 	} else {
1217 		if (handle == 0) {
1218 			handle = qdisc_alloc_handle(dev);
1219 			if (handle == 0) {
1220 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1221 				err = -ENOSPC;
1222 				goto err_out3;
1223 			}
1224 		}
1225 		if (!netif_is_multiqueue(dev))
1226 			sch->flags |= TCQ_F_ONETXQUEUE;
1227 	}
1228 
1229 	sch->handle = handle;
1230 
1231 	/* This exist to keep backward compatible with a userspace
1232 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1233 	 * facility on older kernels by setting tx_queue_len=0 (prior
1234 	 * to qdisc init), and then forgot to reinit tx_queue_len
1235 	 * before again attaching a qdisc.
1236 	 */
1237 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1238 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1239 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1240 	}
1241 
1242 	err = qdisc_block_indexes_set(sch, tca, extack);
1243 	if (err)
1244 		goto err_out3;
1245 
1246 	if (ops->init) {
1247 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1248 		if (err != 0)
1249 			goto err_out5;
1250 	}
1251 
1252 	if (tca[TCA_STAB]) {
1253 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1254 		if (IS_ERR(stab)) {
1255 			err = PTR_ERR(stab);
1256 			goto err_out4;
1257 		}
1258 		rcu_assign_pointer(sch->stab, stab);
1259 	}
1260 	if (tca[TCA_RATE]) {
1261 		seqcount_t *running;
1262 
1263 		err = -EOPNOTSUPP;
1264 		if (sch->flags & TCQ_F_MQROOT) {
1265 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1266 			goto err_out4;
1267 		}
1268 
1269 		if (sch->parent != TC_H_ROOT &&
1270 		    !(sch->flags & TCQ_F_INGRESS) &&
1271 		    (!p || !(p->flags & TCQ_F_MQROOT)))
1272 			running = qdisc_root_sleeping_running(sch);
1273 		else
1274 			running = &sch->running;
1275 
1276 		err = gen_new_estimator(&sch->bstats,
1277 					sch->cpu_bstats,
1278 					&sch->rate_est,
1279 					NULL,
1280 					running,
1281 					tca[TCA_RATE]);
1282 		if (err) {
1283 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1284 			goto err_out4;
1285 		}
1286 	}
1287 
1288 	qdisc_hash_add(sch, false);
1289 	trace_qdisc_create(ops, dev, parent);
1290 
1291 	return sch;
1292 
1293 err_out5:
1294 	/* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1295 	if (ops->destroy)
1296 		ops->destroy(sch);
1297 err_out3:
1298 	dev_put(dev);
1299 	qdisc_free(sch);
1300 err_out2:
1301 	module_put(ops->owner);
1302 err_out:
1303 	*errp = err;
1304 	return NULL;
1305 
1306 err_out4:
1307 	/*
1308 	 * Any broken qdiscs that would require a ops->reset() here?
1309 	 * The qdisc was never in action so it shouldn't be necessary.
1310 	 */
1311 	qdisc_put_stab(rtnl_dereference(sch->stab));
1312 	if (ops->destroy)
1313 		ops->destroy(sch);
1314 	goto err_out3;
1315 }
1316 
1317 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1318 			struct netlink_ext_ack *extack)
1319 {
1320 	struct qdisc_size_table *ostab, *stab = NULL;
1321 	int err = 0;
1322 
1323 	if (tca[TCA_OPTIONS]) {
1324 		if (!sch->ops->change) {
1325 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1326 			return -EINVAL;
1327 		}
1328 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1329 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1330 			return -EOPNOTSUPP;
1331 		}
1332 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1333 		if (err)
1334 			return err;
1335 	}
1336 
1337 	if (tca[TCA_STAB]) {
1338 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1339 		if (IS_ERR(stab))
1340 			return PTR_ERR(stab);
1341 	}
1342 
1343 	ostab = rtnl_dereference(sch->stab);
1344 	rcu_assign_pointer(sch->stab, stab);
1345 	qdisc_put_stab(ostab);
1346 
1347 	if (tca[TCA_RATE]) {
1348 		/* NB: ignores errors from replace_estimator
1349 		   because change can't be undone. */
1350 		if (sch->flags & TCQ_F_MQROOT)
1351 			goto out;
1352 		gen_replace_estimator(&sch->bstats,
1353 				      sch->cpu_bstats,
1354 				      &sch->rate_est,
1355 				      NULL,
1356 				      qdisc_root_sleeping_running(sch),
1357 				      tca[TCA_RATE]);
1358 	}
1359 out:
1360 	return 0;
1361 }
1362 
1363 struct check_loop_arg {
1364 	struct qdisc_walker	w;
1365 	struct Qdisc		*p;
1366 	int			depth;
1367 };
1368 
1369 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1370 			 struct qdisc_walker *w);
1371 
1372 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1373 {
1374 	struct check_loop_arg	arg;
1375 
1376 	if (q->ops->cl_ops == NULL)
1377 		return 0;
1378 
1379 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1380 	arg.w.fn = check_loop_fn;
1381 	arg.depth = depth;
1382 	arg.p = p;
1383 	q->ops->cl_ops->walk(q, &arg.w);
1384 	return arg.w.stop ? -ELOOP : 0;
1385 }
1386 
1387 static int
1388 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1389 {
1390 	struct Qdisc *leaf;
1391 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1392 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1393 
1394 	leaf = cops->leaf(q, cl);
1395 	if (leaf) {
1396 		if (leaf == arg->p || arg->depth > 7)
1397 			return -ELOOP;
1398 		return check_loop(leaf, arg->p, arg->depth + 1);
1399 	}
1400 	return 0;
1401 }
1402 
1403 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1404 	[TCA_KIND]		= { .type = NLA_STRING },
1405 	[TCA_RATE]		= { .type = NLA_BINARY,
1406 				    .len = sizeof(struct tc_estimator) },
1407 	[TCA_STAB]		= { .type = NLA_NESTED },
1408 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1409 	[TCA_CHAIN]		= { .type = NLA_U32 },
1410 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1411 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1412 };
1413 
1414 /*
1415  * Delete/get qdisc.
1416  */
1417 
1418 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1419 			struct netlink_ext_ack *extack)
1420 {
1421 	struct net *net = sock_net(skb->sk);
1422 	struct tcmsg *tcm = nlmsg_data(n);
1423 	struct nlattr *tca[TCA_MAX + 1];
1424 	struct net_device *dev;
1425 	u32 clid;
1426 	struct Qdisc *q = NULL;
1427 	struct Qdisc *p = NULL;
1428 	int err;
1429 
1430 	if ((n->nlmsg_type != RTM_GETQDISC) &&
1431 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1432 		return -EPERM;
1433 
1434 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1435 				     rtm_tca_policy, extack);
1436 	if (err < 0)
1437 		return err;
1438 
1439 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1440 	if (!dev)
1441 		return -ENODEV;
1442 
1443 	clid = tcm->tcm_parent;
1444 	if (clid) {
1445 		if (clid != TC_H_ROOT) {
1446 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1447 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1448 				if (!p) {
1449 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1450 					return -ENOENT;
1451 				}
1452 				q = qdisc_leaf(p, clid);
1453 			} else if (dev_ingress_queue(dev)) {
1454 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1455 			}
1456 		} else {
1457 			q = dev->qdisc;
1458 		}
1459 		if (!q) {
1460 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1461 			return -ENOENT;
1462 		}
1463 
1464 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1465 			NL_SET_ERR_MSG(extack, "Invalid handle");
1466 			return -EINVAL;
1467 		}
1468 	} else {
1469 		q = qdisc_lookup(dev, tcm->tcm_handle);
1470 		if (!q) {
1471 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1472 			return -ENOENT;
1473 		}
1474 	}
1475 
1476 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1477 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1478 		return -EINVAL;
1479 	}
1480 
1481 	if (n->nlmsg_type == RTM_DELQDISC) {
1482 		if (!clid) {
1483 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1484 			return -EINVAL;
1485 		}
1486 		if (q->handle == 0) {
1487 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1488 			return -ENOENT;
1489 		}
1490 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1491 		if (err != 0)
1492 			return err;
1493 	} else {
1494 		qdisc_notify(net, skb, n, clid, NULL, q);
1495 	}
1496 	return 0;
1497 }
1498 
1499 /*
1500  * Create/change qdisc.
1501  */
1502 
1503 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1504 			   struct netlink_ext_ack *extack)
1505 {
1506 	struct net *net = sock_net(skb->sk);
1507 	struct tcmsg *tcm;
1508 	struct nlattr *tca[TCA_MAX + 1];
1509 	struct net_device *dev;
1510 	u32 clid;
1511 	struct Qdisc *q, *p;
1512 	int err;
1513 
1514 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1515 		return -EPERM;
1516 
1517 replay:
1518 	/* Reinit, just in case something touches this. */
1519 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1520 				     rtm_tca_policy, extack);
1521 	if (err < 0)
1522 		return err;
1523 
1524 	tcm = nlmsg_data(n);
1525 	clid = tcm->tcm_parent;
1526 	q = p = NULL;
1527 
1528 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1529 	if (!dev)
1530 		return -ENODEV;
1531 
1532 
1533 	if (clid) {
1534 		if (clid != TC_H_ROOT) {
1535 			if (clid != TC_H_INGRESS) {
1536 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1537 				if (!p) {
1538 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1539 					return -ENOENT;
1540 				}
1541 				q = qdisc_leaf(p, clid);
1542 			} else if (dev_ingress_queue_create(dev)) {
1543 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1544 			}
1545 		} else {
1546 			q = dev->qdisc;
1547 		}
1548 
1549 		/* It may be default qdisc, ignore it */
1550 		if (q && q->handle == 0)
1551 			q = NULL;
1552 
1553 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1554 			if (tcm->tcm_handle) {
1555 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1556 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1557 					return -EEXIST;
1558 				}
1559 				if (TC_H_MIN(tcm->tcm_handle)) {
1560 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1561 					return -EINVAL;
1562 				}
1563 				q = qdisc_lookup(dev, tcm->tcm_handle);
1564 				if (!q)
1565 					goto create_n_graft;
1566 				if (n->nlmsg_flags & NLM_F_EXCL) {
1567 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1568 					return -EEXIST;
1569 				}
1570 				if (tca[TCA_KIND] &&
1571 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1572 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1573 					return -EINVAL;
1574 				}
1575 				if (q == p ||
1576 				    (p && check_loop(q, p, 0))) {
1577 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1578 					return -ELOOP;
1579 				}
1580 				qdisc_refcount_inc(q);
1581 				goto graft;
1582 			} else {
1583 				if (!q)
1584 					goto create_n_graft;
1585 
1586 				/* This magic test requires explanation.
1587 				 *
1588 				 *   We know, that some child q is already
1589 				 *   attached to this parent and have choice:
1590 				 *   either to change it or to create/graft new one.
1591 				 *
1592 				 *   1. We are allowed to create/graft only
1593 				 *   if CREATE and REPLACE flags are set.
1594 				 *
1595 				 *   2. If EXCL is set, requestor wanted to say,
1596 				 *   that qdisc tcm_handle is not expected
1597 				 *   to exist, so that we choose create/graft too.
1598 				 *
1599 				 *   3. The last case is when no flags are set.
1600 				 *   Alas, it is sort of hole in API, we
1601 				 *   cannot decide what to do unambiguously.
1602 				 *   For now we select create/graft, if
1603 				 *   user gave KIND, which does not match existing.
1604 				 */
1605 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1606 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1607 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1608 				     (tca[TCA_KIND] &&
1609 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1610 					goto create_n_graft;
1611 			}
1612 		}
1613 	} else {
1614 		if (!tcm->tcm_handle) {
1615 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1616 			return -EINVAL;
1617 		}
1618 		q = qdisc_lookup(dev, tcm->tcm_handle);
1619 	}
1620 
1621 	/* Change qdisc parameters */
1622 	if (!q) {
1623 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1624 		return -ENOENT;
1625 	}
1626 	if (n->nlmsg_flags & NLM_F_EXCL) {
1627 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1628 		return -EEXIST;
1629 	}
1630 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1631 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1632 		return -EINVAL;
1633 	}
1634 	err = qdisc_change(q, tca, extack);
1635 	if (err == 0)
1636 		qdisc_notify(net, skb, n, clid, NULL, q);
1637 	return err;
1638 
1639 create_n_graft:
1640 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1641 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1642 		return -ENOENT;
1643 	}
1644 	if (clid == TC_H_INGRESS) {
1645 		if (dev_ingress_queue(dev)) {
1646 			q = qdisc_create(dev, dev_ingress_queue(dev), p,
1647 					 tcm->tcm_parent, tcm->tcm_parent,
1648 					 tca, &err, extack);
1649 		} else {
1650 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1651 			err = -ENOENT;
1652 		}
1653 	} else {
1654 		struct netdev_queue *dev_queue;
1655 
1656 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1657 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1658 		else if (p)
1659 			dev_queue = p->dev_queue;
1660 		else
1661 			dev_queue = netdev_get_tx_queue(dev, 0);
1662 
1663 		q = qdisc_create(dev, dev_queue, p,
1664 				 tcm->tcm_parent, tcm->tcm_handle,
1665 				 tca, &err, extack);
1666 	}
1667 	if (q == NULL) {
1668 		if (err == -EAGAIN)
1669 			goto replay;
1670 		return err;
1671 	}
1672 
1673 graft:
1674 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1675 	if (err) {
1676 		if (q)
1677 			qdisc_put(q);
1678 		return err;
1679 	}
1680 
1681 	return 0;
1682 }
1683 
1684 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1685 			      struct netlink_callback *cb,
1686 			      int *q_idx_p, int s_q_idx, bool recur,
1687 			      bool dump_invisible)
1688 {
1689 	int ret = 0, q_idx = *q_idx_p;
1690 	struct Qdisc *q;
1691 	int b;
1692 
1693 	if (!root)
1694 		return 0;
1695 
1696 	q = root;
1697 	if (q_idx < s_q_idx) {
1698 		q_idx++;
1699 	} else {
1700 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1701 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1702 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1703 				  RTM_NEWQDISC) <= 0)
1704 			goto done;
1705 		q_idx++;
1706 	}
1707 
1708 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1709 	 * itself has already been dumped.
1710 	 *
1711 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1712 	 * qdisc hashtable, we don't want to hit it again
1713 	 */
1714 	if (!qdisc_dev(root) || !recur)
1715 		goto out;
1716 
1717 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1718 		if (q_idx < s_q_idx) {
1719 			q_idx++;
1720 			continue;
1721 		}
1722 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1723 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1724 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1725 				  RTM_NEWQDISC) <= 0)
1726 			goto done;
1727 		q_idx++;
1728 	}
1729 
1730 out:
1731 	*q_idx_p = q_idx;
1732 	return ret;
1733 done:
1734 	ret = -1;
1735 	goto out;
1736 }
1737 
1738 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1739 {
1740 	struct net *net = sock_net(skb->sk);
1741 	int idx, q_idx;
1742 	int s_idx, s_q_idx;
1743 	struct net_device *dev;
1744 	const struct nlmsghdr *nlh = cb->nlh;
1745 	struct nlattr *tca[TCA_MAX + 1];
1746 	int err;
1747 
1748 	s_idx = cb->args[0];
1749 	s_q_idx = q_idx = cb->args[1];
1750 
1751 	idx = 0;
1752 	ASSERT_RTNL();
1753 
1754 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1755 				     rtm_tca_policy, cb->extack);
1756 	if (err < 0)
1757 		return err;
1758 
1759 	for_each_netdev(net, dev) {
1760 		struct netdev_queue *dev_queue;
1761 
1762 		if (idx < s_idx)
1763 			goto cont;
1764 		if (idx > s_idx)
1765 			s_q_idx = 0;
1766 		q_idx = 0;
1767 
1768 		if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1769 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1770 			goto done;
1771 
1772 		dev_queue = dev_ingress_queue(dev);
1773 		if (dev_queue &&
1774 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1775 				       &q_idx, s_q_idx, false,
1776 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1777 			goto done;
1778 
1779 cont:
1780 		idx++;
1781 	}
1782 
1783 done:
1784 	cb->args[0] = idx;
1785 	cb->args[1] = q_idx;
1786 
1787 	return skb->len;
1788 }
1789 
1790 
1791 
1792 /************************************************
1793  *	Traffic classes manipulation.		*
1794  ************************************************/
1795 
1796 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1797 			  unsigned long cl,
1798 			  u32 portid, u32 seq, u16 flags, int event)
1799 {
1800 	struct tcmsg *tcm;
1801 	struct nlmsghdr  *nlh;
1802 	unsigned char *b = skb_tail_pointer(skb);
1803 	struct gnet_dump d;
1804 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1805 
1806 	cond_resched();
1807 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1808 	if (!nlh)
1809 		goto out_nlmsg_trim;
1810 	tcm = nlmsg_data(nlh);
1811 	tcm->tcm_family = AF_UNSPEC;
1812 	tcm->tcm__pad1 = 0;
1813 	tcm->tcm__pad2 = 0;
1814 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1815 	tcm->tcm_parent = q->handle;
1816 	tcm->tcm_handle = q->handle;
1817 	tcm->tcm_info = 0;
1818 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1819 		goto nla_put_failure;
1820 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1821 		goto nla_put_failure;
1822 
1823 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1824 					 NULL, &d, TCA_PAD) < 0)
1825 		goto nla_put_failure;
1826 
1827 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1828 		goto nla_put_failure;
1829 
1830 	if (gnet_stats_finish_copy(&d) < 0)
1831 		goto nla_put_failure;
1832 
1833 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1834 	return skb->len;
1835 
1836 out_nlmsg_trim:
1837 nla_put_failure:
1838 	nlmsg_trim(skb, b);
1839 	return -1;
1840 }
1841 
1842 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1843 			 struct nlmsghdr *n, struct Qdisc *q,
1844 			 unsigned long cl, int event)
1845 {
1846 	struct sk_buff *skb;
1847 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1848 	int err = 0;
1849 
1850 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1851 	if (!skb)
1852 		return -ENOBUFS;
1853 
1854 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1855 		kfree_skb(skb);
1856 		return -EINVAL;
1857 	}
1858 
1859 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1860 			     n->nlmsg_flags & NLM_F_ECHO);
1861 	if (err > 0)
1862 		err = 0;
1863 	return err;
1864 }
1865 
1866 static int tclass_del_notify(struct net *net,
1867 			     const struct Qdisc_class_ops *cops,
1868 			     struct sk_buff *oskb, struct nlmsghdr *n,
1869 			     struct Qdisc *q, unsigned long cl)
1870 {
1871 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1872 	struct sk_buff *skb;
1873 	int err = 0;
1874 
1875 	if (!cops->delete)
1876 		return -EOPNOTSUPP;
1877 
1878 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1879 	if (!skb)
1880 		return -ENOBUFS;
1881 
1882 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1883 			   RTM_DELTCLASS) < 0) {
1884 		kfree_skb(skb);
1885 		return -EINVAL;
1886 	}
1887 
1888 	err = cops->delete(q, cl);
1889 	if (err) {
1890 		kfree_skb(skb);
1891 		return err;
1892 	}
1893 
1894 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1895 			     n->nlmsg_flags & NLM_F_ECHO);
1896 	if (err > 0)
1897 		err = 0;
1898 	return err;
1899 }
1900 
1901 #ifdef CONFIG_NET_CLS
1902 
1903 struct tcf_bind_args {
1904 	struct tcf_walker w;
1905 	unsigned long base;
1906 	unsigned long cl;
1907 	u32 classid;
1908 };
1909 
1910 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1911 {
1912 	struct tcf_bind_args *a = (void *)arg;
1913 
1914 	if (tp->ops->bind_class) {
1915 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1916 
1917 		sch_tree_lock(q);
1918 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1919 		sch_tree_unlock(q);
1920 	}
1921 	return 0;
1922 }
1923 
1924 struct tc_bind_class_args {
1925 	struct qdisc_walker w;
1926 	unsigned long new_cl;
1927 	u32 portid;
1928 	u32 clid;
1929 };
1930 
1931 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1932 				struct qdisc_walker *w)
1933 {
1934 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1935 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1936 	struct tcf_block *block;
1937 	struct tcf_chain *chain;
1938 
1939 	block = cops->tcf_block(q, cl, NULL);
1940 	if (!block)
1941 		return 0;
1942 	for (chain = tcf_get_next_chain(block, NULL);
1943 	     chain;
1944 	     chain = tcf_get_next_chain(block, chain)) {
1945 		struct tcf_proto *tp;
1946 
1947 		for (tp = tcf_get_next_proto(chain, NULL);
1948 		     tp; tp = tcf_get_next_proto(chain, tp)) {
1949 			struct tcf_bind_args arg = {};
1950 
1951 			arg.w.fn = tcf_node_bind;
1952 			arg.classid = a->clid;
1953 			arg.base = cl;
1954 			arg.cl = a->new_cl;
1955 			tp->ops->walk(tp, &arg.w, true);
1956 		}
1957 	}
1958 
1959 	return 0;
1960 }
1961 
1962 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1963 			   unsigned long new_cl)
1964 {
1965 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1966 	struct tc_bind_class_args args = {};
1967 
1968 	if (!cops->tcf_block)
1969 		return;
1970 	args.portid = portid;
1971 	args.clid = clid;
1972 	args.new_cl = new_cl;
1973 	args.w.fn = tc_bind_class_walker;
1974 	q->ops->cl_ops->walk(q, &args.w);
1975 }
1976 
1977 #else
1978 
1979 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1980 			   unsigned long new_cl)
1981 {
1982 }
1983 
1984 #endif
1985 
1986 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1987 			 struct netlink_ext_ack *extack)
1988 {
1989 	struct net *net = sock_net(skb->sk);
1990 	struct tcmsg *tcm = nlmsg_data(n);
1991 	struct nlattr *tca[TCA_MAX + 1];
1992 	struct net_device *dev;
1993 	struct Qdisc *q = NULL;
1994 	const struct Qdisc_class_ops *cops;
1995 	unsigned long cl = 0;
1996 	unsigned long new_cl;
1997 	u32 portid;
1998 	u32 clid;
1999 	u32 qid;
2000 	int err;
2001 
2002 	if ((n->nlmsg_type != RTM_GETTCLASS) &&
2003 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2004 		return -EPERM;
2005 
2006 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2007 				     rtm_tca_policy, extack);
2008 	if (err < 0)
2009 		return err;
2010 
2011 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2012 	if (!dev)
2013 		return -ENODEV;
2014 
2015 	/*
2016 	   parent == TC_H_UNSPEC - unspecified parent.
2017 	   parent == TC_H_ROOT   - class is root, which has no parent.
2018 	   parent == X:0	 - parent is root class.
2019 	   parent == X:Y	 - parent is a node in hierarchy.
2020 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2021 
2022 	   handle == 0:0	 - generate handle from kernel pool.
2023 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2024 	   handle == X:Y	 - clear.
2025 	   handle == X:0	 - root class.
2026 	 */
2027 
2028 	/* Step 1. Determine qdisc handle X:0 */
2029 
2030 	portid = tcm->tcm_parent;
2031 	clid = tcm->tcm_handle;
2032 	qid = TC_H_MAJ(clid);
2033 
2034 	if (portid != TC_H_ROOT) {
2035 		u32 qid1 = TC_H_MAJ(portid);
2036 
2037 		if (qid && qid1) {
2038 			/* If both majors are known, they must be identical. */
2039 			if (qid != qid1)
2040 				return -EINVAL;
2041 		} else if (qid1) {
2042 			qid = qid1;
2043 		} else if (qid == 0)
2044 			qid = dev->qdisc->handle;
2045 
2046 		/* Now qid is genuine qdisc handle consistent
2047 		 * both with parent and child.
2048 		 *
2049 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2050 		 */
2051 		if (portid)
2052 			portid = TC_H_MAKE(qid, portid);
2053 	} else {
2054 		if (qid == 0)
2055 			qid = dev->qdisc->handle;
2056 	}
2057 
2058 	/* OK. Locate qdisc */
2059 	q = qdisc_lookup(dev, qid);
2060 	if (!q)
2061 		return -ENOENT;
2062 
2063 	/* An check that it supports classes */
2064 	cops = q->ops->cl_ops;
2065 	if (cops == NULL)
2066 		return -EINVAL;
2067 
2068 	/* Now try to get class */
2069 	if (clid == 0) {
2070 		if (portid == TC_H_ROOT)
2071 			clid = qid;
2072 	} else
2073 		clid = TC_H_MAKE(qid, clid);
2074 
2075 	if (clid)
2076 		cl = cops->find(q, clid);
2077 
2078 	if (cl == 0) {
2079 		err = -ENOENT;
2080 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2081 		    !(n->nlmsg_flags & NLM_F_CREATE))
2082 			goto out;
2083 	} else {
2084 		switch (n->nlmsg_type) {
2085 		case RTM_NEWTCLASS:
2086 			err = -EEXIST;
2087 			if (n->nlmsg_flags & NLM_F_EXCL)
2088 				goto out;
2089 			break;
2090 		case RTM_DELTCLASS:
2091 			err = tclass_del_notify(net, cops, skb, n, q, cl);
2092 			/* Unbind the class with flilters with 0 */
2093 			tc_bind_tclass(q, portid, clid, 0);
2094 			goto out;
2095 		case RTM_GETTCLASS:
2096 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
2097 			goto out;
2098 		default:
2099 			err = -EINVAL;
2100 			goto out;
2101 		}
2102 	}
2103 
2104 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2105 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2106 		return -EOPNOTSUPP;
2107 	}
2108 
2109 	new_cl = cl;
2110 	err = -EOPNOTSUPP;
2111 	if (cops->change)
2112 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2113 	if (err == 0) {
2114 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
2115 		/* We just create a new class, need to do reverse binding. */
2116 		if (cl != new_cl)
2117 			tc_bind_tclass(q, portid, clid, new_cl);
2118 	}
2119 out:
2120 	return err;
2121 }
2122 
2123 struct qdisc_dump_args {
2124 	struct qdisc_walker	w;
2125 	struct sk_buff		*skb;
2126 	struct netlink_callback	*cb;
2127 };
2128 
2129 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2130 			    struct qdisc_walker *arg)
2131 {
2132 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2133 
2134 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2135 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2136 			      RTM_NEWTCLASS);
2137 }
2138 
2139 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2140 				struct tcmsg *tcm, struct netlink_callback *cb,
2141 				int *t_p, int s_t)
2142 {
2143 	struct qdisc_dump_args arg;
2144 
2145 	if (tc_qdisc_dump_ignore(q, false) ||
2146 	    *t_p < s_t || !q->ops->cl_ops ||
2147 	    (tcm->tcm_parent &&
2148 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2149 		(*t_p)++;
2150 		return 0;
2151 	}
2152 	if (*t_p > s_t)
2153 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2154 	arg.w.fn = qdisc_class_dump;
2155 	arg.skb = skb;
2156 	arg.cb = cb;
2157 	arg.w.stop  = 0;
2158 	arg.w.skip = cb->args[1];
2159 	arg.w.count = 0;
2160 	q->ops->cl_ops->walk(q, &arg.w);
2161 	cb->args[1] = arg.w.count;
2162 	if (arg.w.stop)
2163 		return -1;
2164 	(*t_p)++;
2165 	return 0;
2166 }
2167 
2168 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2169 			       struct tcmsg *tcm, struct netlink_callback *cb,
2170 			       int *t_p, int s_t)
2171 {
2172 	struct Qdisc *q;
2173 	int b;
2174 
2175 	if (!root)
2176 		return 0;
2177 
2178 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2179 		return -1;
2180 
2181 	if (!qdisc_dev(root))
2182 		return 0;
2183 
2184 	if (tcm->tcm_parent) {
2185 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2186 		if (q && q != root &&
2187 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2188 			return -1;
2189 		return 0;
2190 	}
2191 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2192 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2193 			return -1;
2194 	}
2195 
2196 	return 0;
2197 }
2198 
2199 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2200 {
2201 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2202 	struct net *net = sock_net(skb->sk);
2203 	struct netdev_queue *dev_queue;
2204 	struct net_device *dev;
2205 	int t, s_t;
2206 
2207 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2208 		return 0;
2209 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2210 	if (!dev)
2211 		return 0;
2212 
2213 	s_t = cb->args[0];
2214 	t = 0;
2215 
2216 	if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t) < 0)
2217 		goto done;
2218 
2219 	dev_queue = dev_ingress_queue(dev);
2220 	if (dev_queue &&
2221 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2222 				&t, s_t) < 0)
2223 		goto done;
2224 
2225 done:
2226 	cb->args[0] = t;
2227 
2228 	dev_put(dev);
2229 	return skb->len;
2230 }
2231 
2232 #ifdef CONFIG_PROC_FS
2233 static int psched_show(struct seq_file *seq, void *v)
2234 {
2235 	seq_printf(seq, "%08x %08x %08x %08x\n",
2236 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2237 		   1000000,
2238 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2239 
2240 	return 0;
2241 }
2242 
2243 static int __net_init psched_net_init(struct net *net)
2244 {
2245 	struct proc_dir_entry *e;
2246 
2247 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2248 	if (e == NULL)
2249 		return -ENOMEM;
2250 
2251 	return 0;
2252 }
2253 
2254 static void __net_exit psched_net_exit(struct net *net)
2255 {
2256 	remove_proc_entry("psched", net->proc_net);
2257 }
2258 #else
2259 static int __net_init psched_net_init(struct net *net)
2260 {
2261 	return 0;
2262 }
2263 
2264 static void __net_exit psched_net_exit(struct net *net)
2265 {
2266 }
2267 #endif
2268 
2269 static struct pernet_operations psched_net_ops = {
2270 	.init = psched_net_init,
2271 	.exit = psched_net_exit,
2272 };
2273 
2274 static int __init pktsched_init(void)
2275 {
2276 	int err;
2277 
2278 	err = register_pernet_subsys(&psched_net_ops);
2279 	if (err) {
2280 		pr_err("pktsched_init: "
2281 		       "cannot initialize per netns operations\n");
2282 		return err;
2283 	}
2284 
2285 	register_qdisc(&pfifo_fast_ops);
2286 	register_qdisc(&pfifo_qdisc_ops);
2287 	register_qdisc(&bfifo_qdisc_ops);
2288 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2289 	register_qdisc(&mq_qdisc_ops);
2290 	register_qdisc(&noqueue_qdisc_ops);
2291 
2292 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2293 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2294 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2295 		      0);
2296 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2297 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2298 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2299 		      0);
2300 
2301 	return 0;
2302 }
2303 
2304 subsys_initcall(pktsched_init);
2305