xref: /linux/net/sched/sch_api.c (revision 9052e9c95d908d6c3d7570aadc8898e1d871c8bb)
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 	if (s->size_log > STAB_SIZE_LOG_MAX ||
517 	    s->cell_log > STAB_SIZE_LOG_MAX) {
518 		NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
519 		return ERR_PTR(-EINVAL);
520 	}
521 
522 	stab = kmalloc(sizeof(*stab) + tsize * sizeof(u16), GFP_KERNEL);
523 	if (!stab)
524 		return ERR_PTR(-ENOMEM);
525 
526 	stab->refcnt = 1;
527 	stab->szopts = *s;
528 	if (tsize > 0)
529 		memcpy(stab->data, tab, tsize * sizeof(u16));
530 
531 	list_add_tail(&stab->list, &qdisc_stab_list);
532 
533 	return stab;
534 }
535 
536 void qdisc_put_stab(struct qdisc_size_table *tab)
537 {
538 	if (!tab)
539 		return;
540 
541 	if (--tab->refcnt == 0) {
542 		list_del(&tab->list);
543 		kfree_rcu(tab, rcu);
544 	}
545 }
546 EXPORT_SYMBOL(qdisc_put_stab);
547 
548 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
549 {
550 	struct nlattr *nest;
551 
552 	nest = nla_nest_start_noflag(skb, TCA_STAB);
553 	if (nest == NULL)
554 		goto nla_put_failure;
555 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
556 		goto nla_put_failure;
557 	nla_nest_end(skb, nest);
558 
559 	return skb->len;
560 
561 nla_put_failure:
562 	return -1;
563 }
564 
565 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
566 			       const struct qdisc_size_table *stab)
567 {
568 	int pkt_len, slot;
569 
570 	pkt_len = skb->len + stab->szopts.overhead;
571 	if (unlikely(!stab->szopts.tsize))
572 		goto out;
573 
574 	slot = pkt_len + stab->szopts.cell_align;
575 	if (unlikely(slot < 0))
576 		slot = 0;
577 
578 	slot >>= stab->szopts.cell_log;
579 	if (likely(slot < stab->szopts.tsize))
580 		pkt_len = stab->data[slot];
581 	else
582 		pkt_len = stab->data[stab->szopts.tsize - 1] *
583 				(slot / stab->szopts.tsize) +
584 				stab->data[slot % stab->szopts.tsize];
585 
586 	pkt_len <<= stab->szopts.size_log;
587 out:
588 	if (unlikely(pkt_len < 1))
589 		pkt_len = 1;
590 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
591 }
592 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
593 
594 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
595 {
596 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
597 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
598 			txt, qdisc->ops->id, qdisc->handle >> 16);
599 		qdisc->flags |= TCQ_F_WARN_NONWC;
600 	}
601 }
602 EXPORT_SYMBOL(qdisc_warn_nonwc);
603 
604 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
605 {
606 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
607 						 timer);
608 
609 	rcu_read_lock();
610 	__netif_schedule(qdisc_root(wd->qdisc));
611 	rcu_read_unlock();
612 
613 	return HRTIMER_NORESTART;
614 }
615 
616 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
617 				 clockid_t clockid)
618 {
619 	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
620 	wd->timer.function = qdisc_watchdog;
621 	wd->qdisc = qdisc;
622 }
623 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
624 
625 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
626 {
627 	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
628 }
629 EXPORT_SYMBOL(qdisc_watchdog_init);
630 
631 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
632 				      u64 delta_ns)
633 {
634 	if (test_bit(__QDISC_STATE_DEACTIVATED,
635 		     &qdisc_root_sleeping(wd->qdisc)->state))
636 		return;
637 
638 	if (hrtimer_is_queued(&wd->timer)) {
639 		/* If timer is already set in [expires, expires + delta_ns],
640 		 * do not reprogram it.
641 		 */
642 		if (wd->last_expires - expires <= delta_ns)
643 			return;
644 	}
645 
646 	wd->last_expires = expires;
647 	hrtimer_start_range_ns(&wd->timer,
648 			       ns_to_ktime(expires),
649 			       delta_ns,
650 			       HRTIMER_MODE_ABS_PINNED);
651 }
652 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
653 
654 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
655 {
656 	hrtimer_cancel(&wd->timer);
657 }
658 EXPORT_SYMBOL(qdisc_watchdog_cancel);
659 
660 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
661 {
662 	struct hlist_head *h;
663 	unsigned int i;
664 
665 	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
666 
667 	if (h != NULL) {
668 		for (i = 0; i < n; i++)
669 			INIT_HLIST_HEAD(&h[i]);
670 	}
671 	return h;
672 }
673 
674 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
675 {
676 	struct Qdisc_class_common *cl;
677 	struct hlist_node *next;
678 	struct hlist_head *nhash, *ohash;
679 	unsigned int nsize, nmask, osize;
680 	unsigned int i, h;
681 
682 	/* Rehash when load factor exceeds 0.75 */
683 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
684 		return;
685 	nsize = clhash->hashsize * 2;
686 	nmask = nsize - 1;
687 	nhash = qdisc_class_hash_alloc(nsize);
688 	if (nhash == NULL)
689 		return;
690 
691 	ohash = clhash->hash;
692 	osize = clhash->hashsize;
693 
694 	sch_tree_lock(sch);
695 	for (i = 0; i < osize; i++) {
696 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
697 			h = qdisc_class_hash(cl->classid, nmask);
698 			hlist_add_head(&cl->hnode, &nhash[h]);
699 		}
700 	}
701 	clhash->hash     = nhash;
702 	clhash->hashsize = nsize;
703 	clhash->hashmask = nmask;
704 	sch_tree_unlock(sch);
705 
706 	kvfree(ohash);
707 }
708 EXPORT_SYMBOL(qdisc_class_hash_grow);
709 
710 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
711 {
712 	unsigned int size = 4;
713 
714 	clhash->hash = qdisc_class_hash_alloc(size);
715 	if (!clhash->hash)
716 		return -ENOMEM;
717 	clhash->hashsize  = size;
718 	clhash->hashmask  = size - 1;
719 	clhash->hashelems = 0;
720 	return 0;
721 }
722 EXPORT_SYMBOL(qdisc_class_hash_init);
723 
724 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
725 {
726 	kvfree(clhash->hash);
727 }
728 EXPORT_SYMBOL(qdisc_class_hash_destroy);
729 
730 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
731 			     struct Qdisc_class_common *cl)
732 {
733 	unsigned int h;
734 
735 	INIT_HLIST_NODE(&cl->hnode);
736 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
737 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
738 	clhash->hashelems++;
739 }
740 EXPORT_SYMBOL(qdisc_class_hash_insert);
741 
742 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
743 			     struct Qdisc_class_common *cl)
744 {
745 	hlist_del(&cl->hnode);
746 	clhash->hashelems--;
747 }
748 EXPORT_SYMBOL(qdisc_class_hash_remove);
749 
750 /* Allocate an unique handle from space managed by kernel
751  * Possible range is [8000-FFFF]:0000 (0x8000 values)
752  */
753 static u32 qdisc_alloc_handle(struct net_device *dev)
754 {
755 	int i = 0x8000;
756 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
757 
758 	do {
759 		autohandle += TC_H_MAKE(0x10000U, 0);
760 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
761 			autohandle = TC_H_MAKE(0x80000000U, 0);
762 		if (!qdisc_lookup(dev, autohandle))
763 			return autohandle;
764 		cond_resched();
765 	} while	(--i > 0);
766 
767 	return 0;
768 }
769 
770 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
771 {
772 	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
773 	const struct Qdisc_class_ops *cops;
774 	unsigned long cl;
775 	u32 parentid;
776 	bool notify;
777 	int drops;
778 
779 	if (n == 0 && len == 0)
780 		return;
781 	drops = max_t(int, n, 0);
782 	rcu_read_lock();
783 	while ((parentid = sch->parent)) {
784 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
785 			break;
786 
787 		if (sch->flags & TCQ_F_NOPARENT)
788 			break;
789 		/* Notify parent qdisc only if child qdisc becomes empty.
790 		 *
791 		 * If child was empty even before update then backlog
792 		 * counter is screwed and we skip notification because
793 		 * parent class is already passive.
794 		 *
795 		 * If the original child was offloaded then it is allowed
796 		 * to be seem as empty, so the parent is notified anyway.
797 		 */
798 		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
799 						       !qdisc_is_offloaded);
800 		/* TODO: perform the search on a per txq basis */
801 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
802 		if (sch == NULL) {
803 			WARN_ON_ONCE(parentid != TC_H_ROOT);
804 			break;
805 		}
806 		cops = sch->ops->cl_ops;
807 		if (notify && cops->qlen_notify) {
808 			cl = cops->find(sch, parentid);
809 			cops->qlen_notify(sch, cl);
810 		}
811 		sch->q.qlen -= n;
812 		sch->qstats.backlog -= len;
813 		__qdisc_qstats_drop(sch, drops);
814 	}
815 	rcu_read_unlock();
816 }
817 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
818 
819 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
820 			      void *type_data)
821 {
822 	struct net_device *dev = qdisc_dev(sch);
823 	int err;
824 
825 	sch->flags &= ~TCQ_F_OFFLOADED;
826 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
827 		return 0;
828 
829 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
830 	if (err == -EOPNOTSUPP)
831 		return 0;
832 
833 	if (!err)
834 		sch->flags |= TCQ_F_OFFLOADED;
835 
836 	return err;
837 }
838 EXPORT_SYMBOL(qdisc_offload_dump_helper);
839 
840 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
841 				struct Qdisc *new, struct Qdisc *old,
842 				enum tc_setup_type type, void *type_data,
843 				struct netlink_ext_ack *extack)
844 {
845 	bool any_qdisc_is_offloaded;
846 	int err;
847 
848 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
849 		return;
850 
851 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
852 
853 	/* Don't report error if the graft is part of destroy operation. */
854 	if (!err || !new || new == &noop_qdisc)
855 		return;
856 
857 	/* Don't report error if the parent, the old child and the new
858 	 * one are not offloaded.
859 	 */
860 	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
861 	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
862 	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
863 
864 	if (any_qdisc_is_offloaded)
865 		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
866 }
867 EXPORT_SYMBOL(qdisc_offload_graft_helper);
868 
869 static void qdisc_offload_graft_root(struct net_device *dev,
870 				     struct Qdisc *new, struct Qdisc *old,
871 				     struct netlink_ext_ack *extack)
872 {
873 	struct tc_root_qopt_offload graft_offload = {
874 		.command	= TC_ROOT_GRAFT,
875 		.handle		= new ? new->handle : 0,
876 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
877 				  (old && old->flags & TCQ_F_INGRESS),
878 	};
879 
880 	qdisc_offload_graft_helper(dev, NULL, new, old,
881 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
882 }
883 
884 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
885 			 u32 portid, u32 seq, u16 flags, int event)
886 {
887 	struct gnet_stats_basic_cpu __percpu *cpu_bstats = NULL;
888 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
889 	struct tcmsg *tcm;
890 	struct nlmsghdr  *nlh;
891 	unsigned char *b = skb_tail_pointer(skb);
892 	struct gnet_dump d;
893 	struct qdisc_size_table *stab;
894 	u32 block_index;
895 	__u32 qlen;
896 
897 	cond_resched();
898 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
899 	if (!nlh)
900 		goto out_nlmsg_trim;
901 	tcm = nlmsg_data(nlh);
902 	tcm->tcm_family = AF_UNSPEC;
903 	tcm->tcm__pad1 = 0;
904 	tcm->tcm__pad2 = 0;
905 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
906 	tcm->tcm_parent = clid;
907 	tcm->tcm_handle = q->handle;
908 	tcm->tcm_info = refcount_read(&q->refcnt);
909 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
910 		goto nla_put_failure;
911 	if (q->ops->ingress_block_get) {
912 		block_index = q->ops->ingress_block_get(q);
913 		if (block_index &&
914 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
915 			goto nla_put_failure;
916 	}
917 	if (q->ops->egress_block_get) {
918 		block_index = q->ops->egress_block_get(q);
919 		if (block_index &&
920 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
921 			goto nla_put_failure;
922 	}
923 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
924 		goto nla_put_failure;
925 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
926 		goto nla_put_failure;
927 	qlen = qdisc_qlen_sum(q);
928 
929 	stab = rtnl_dereference(q->stab);
930 	if (stab && qdisc_dump_stab(skb, stab) < 0)
931 		goto nla_put_failure;
932 
933 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
934 					 NULL, &d, TCA_PAD) < 0)
935 		goto nla_put_failure;
936 
937 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
938 		goto nla_put_failure;
939 
940 	if (qdisc_is_percpu_stats(q)) {
941 		cpu_bstats = q->cpu_bstats;
942 		cpu_qstats = q->cpu_qstats;
943 	}
944 
945 	if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q),
946 				  &d, cpu_bstats, &q->bstats) < 0 ||
947 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
948 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
949 		goto nla_put_failure;
950 
951 	if (gnet_stats_finish_copy(&d) < 0)
952 		goto nla_put_failure;
953 
954 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
955 	return skb->len;
956 
957 out_nlmsg_trim:
958 nla_put_failure:
959 	nlmsg_trim(skb, b);
960 	return -1;
961 }
962 
963 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
964 {
965 	if (q->flags & TCQ_F_BUILTIN)
966 		return true;
967 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
968 		return true;
969 
970 	return false;
971 }
972 
973 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
974 			struct nlmsghdr *n, u32 clid,
975 			struct Qdisc *old, struct Qdisc *new)
976 {
977 	struct sk_buff *skb;
978 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
979 
980 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
981 	if (!skb)
982 		return -ENOBUFS;
983 
984 	if (old && !tc_qdisc_dump_ignore(old, false)) {
985 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
986 				  0, RTM_DELQDISC) < 0)
987 			goto err_out;
988 	}
989 	if (new && !tc_qdisc_dump_ignore(new, false)) {
990 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
991 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
992 			goto err_out;
993 	}
994 
995 	if (skb->len)
996 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
997 				      n->nlmsg_flags & NLM_F_ECHO);
998 
999 err_out:
1000 	kfree_skb(skb);
1001 	return -EINVAL;
1002 }
1003 
1004 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1005 			       struct nlmsghdr *n, u32 clid,
1006 			       struct Qdisc *old, struct Qdisc *new)
1007 {
1008 	if (new || old)
1009 		qdisc_notify(net, skb, n, clid, old, new);
1010 
1011 	if (old)
1012 		qdisc_put(old);
1013 }
1014 
1015 static void qdisc_clear_nolock(struct Qdisc *sch)
1016 {
1017 	sch->flags &= ~TCQ_F_NOLOCK;
1018 	if (!(sch->flags & TCQ_F_CPUSTATS))
1019 		return;
1020 
1021 	free_percpu(sch->cpu_bstats);
1022 	free_percpu(sch->cpu_qstats);
1023 	sch->cpu_bstats = NULL;
1024 	sch->cpu_qstats = NULL;
1025 	sch->flags &= ~TCQ_F_CPUSTATS;
1026 }
1027 
1028 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1029  * to device "dev".
1030  *
1031  * When appropriate send a netlink notification using 'skb'
1032  * and "n".
1033  *
1034  * On success, destroy old qdisc.
1035  */
1036 
1037 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1038 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1039 		       struct Qdisc *new, struct Qdisc *old,
1040 		       struct netlink_ext_ack *extack)
1041 {
1042 	struct Qdisc *q = old;
1043 	struct net *net = dev_net(dev);
1044 
1045 	if (parent == NULL) {
1046 		unsigned int i, num_q, ingress;
1047 
1048 		ingress = 0;
1049 		num_q = dev->num_tx_queues;
1050 		if ((q && q->flags & TCQ_F_INGRESS) ||
1051 		    (new && new->flags & TCQ_F_INGRESS)) {
1052 			num_q = 1;
1053 			ingress = 1;
1054 			if (!dev_ingress_queue(dev)) {
1055 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1056 				return -ENOENT;
1057 			}
1058 		}
1059 
1060 		if (dev->flags & IFF_UP)
1061 			dev_deactivate(dev);
1062 
1063 		qdisc_offload_graft_root(dev, new, old, extack);
1064 
1065 		if (new && new->ops->attach)
1066 			goto skip;
1067 
1068 		for (i = 0; i < num_q; i++) {
1069 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1070 
1071 			if (!ingress)
1072 				dev_queue = netdev_get_tx_queue(dev, i);
1073 
1074 			old = dev_graft_qdisc(dev_queue, new);
1075 			if (new && i > 0)
1076 				qdisc_refcount_inc(new);
1077 
1078 			if (!ingress)
1079 				qdisc_put(old);
1080 		}
1081 
1082 skip:
1083 		if (!ingress) {
1084 			notify_and_destroy(net, skb, n, classid,
1085 					   dev->qdisc, new);
1086 			if (new && !new->ops->attach)
1087 				qdisc_refcount_inc(new);
1088 			dev->qdisc = new ? : &noop_qdisc;
1089 
1090 			if (new && new->ops->attach)
1091 				new->ops->attach(new);
1092 		} else {
1093 			notify_and_destroy(net, skb, n, classid, old, new);
1094 		}
1095 
1096 		if (dev->flags & IFF_UP)
1097 			dev_activate(dev);
1098 	} else {
1099 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1100 		unsigned long cl;
1101 		int err;
1102 
1103 		/* Only support running class lockless if parent is lockless */
1104 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1105 			qdisc_clear_nolock(new);
1106 
1107 		if (!cops || !cops->graft)
1108 			return -EOPNOTSUPP;
1109 
1110 		cl = cops->find(parent, classid);
1111 		if (!cl) {
1112 			NL_SET_ERR_MSG(extack, "Specified class not found");
1113 			return -ENOENT;
1114 		}
1115 
1116 		err = cops->graft(parent, cl, new, &old, extack);
1117 		if (err)
1118 			return err;
1119 		notify_and_destroy(net, skb, n, classid, old, new);
1120 	}
1121 	return 0;
1122 }
1123 
1124 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1125 				   struct netlink_ext_ack *extack)
1126 {
1127 	u32 block_index;
1128 
1129 	if (tca[TCA_INGRESS_BLOCK]) {
1130 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1131 
1132 		if (!block_index) {
1133 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1134 			return -EINVAL;
1135 		}
1136 		if (!sch->ops->ingress_block_set) {
1137 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1138 			return -EOPNOTSUPP;
1139 		}
1140 		sch->ops->ingress_block_set(sch, block_index);
1141 	}
1142 	if (tca[TCA_EGRESS_BLOCK]) {
1143 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1144 
1145 		if (!block_index) {
1146 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1147 			return -EINVAL;
1148 		}
1149 		if (!sch->ops->egress_block_set) {
1150 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1151 			return -EOPNOTSUPP;
1152 		}
1153 		sch->ops->egress_block_set(sch, block_index);
1154 	}
1155 	return 0;
1156 }
1157 
1158 /*
1159    Allocate and initialize new qdisc.
1160 
1161    Parameters are passed via opt.
1162  */
1163 
1164 static struct Qdisc *qdisc_create(struct net_device *dev,
1165 				  struct netdev_queue *dev_queue,
1166 				  struct Qdisc *p, u32 parent, u32 handle,
1167 				  struct nlattr **tca, int *errp,
1168 				  struct netlink_ext_ack *extack)
1169 {
1170 	int err;
1171 	struct nlattr *kind = tca[TCA_KIND];
1172 	struct Qdisc *sch;
1173 	struct Qdisc_ops *ops;
1174 	struct qdisc_size_table *stab;
1175 
1176 	ops = qdisc_lookup_ops(kind);
1177 #ifdef CONFIG_MODULES
1178 	if (ops == NULL && kind != NULL) {
1179 		char name[IFNAMSIZ];
1180 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1181 			/* We dropped the RTNL semaphore in order to
1182 			 * perform the module load.  So, even if we
1183 			 * succeeded in loading the module we have to
1184 			 * tell the caller to replay the request.  We
1185 			 * indicate this using -EAGAIN.
1186 			 * We replay the request because the device may
1187 			 * go away in the mean time.
1188 			 */
1189 			rtnl_unlock();
1190 			request_module("sch_%s", name);
1191 			rtnl_lock();
1192 			ops = qdisc_lookup_ops(kind);
1193 			if (ops != NULL) {
1194 				/* We will try again qdisc_lookup_ops,
1195 				 * so don't keep a reference.
1196 				 */
1197 				module_put(ops->owner);
1198 				err = -EAGAIN;
1199 				goto err_out;
1200 			}
1201 		}
1202 	}
1203 #endif
1204 
1205 	err = -ENOENT;
1206 	if (!ops) {
1207 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1208 		goto err_out;
1209 	}
1210 
1211 	sch = qdisc_alloc(dev_queue, ops, extack);
1212 	if (IS_ERR(sch)) {
1213 		err = PTR_ERR(sch);
1214 		goto err_out2;
1215 	}
1216 
1217 	sch->parent = parent;
1218 
1219 	if (handle == TC_H_INGRESS) {
1220 		sch->flags |= TCQ_F_INGRESS;
1221 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1222 	} else {
1223 		if (handle == 0) {
1224 			handle = qdisc_alloc_handle(dev);
1225 			if (handle == 0) {
1226 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1227 				err = -ENOSPC;
1228 				goto err_out3;
1229 			}
1230 		}
1231 		if (!netif_is_multiqueue(dev))
1232 			sch->flags |= TCQ_F_ONETXQUEUE;
1233 	}
1234 
1235 	sch->handle = handle;
1236 
1237 	/* This exist to keep backward compatible with a userspace
1238 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1239 	 * facility on older kernels by setting tx_queue_len=0 (prior
1240 	 * to qdisc init), and then forgot to reinit tx_queue_len
1241 	 * before again attaching a qdisc.
1242 	 */
1243 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1244 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1245 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1246 	}
1247 
1248 	err = qdisc_block_indexes_set(sch, tca, extack);
1249 	if (err)
1250 		goto err_out3;
1251 
1252 	if (ops->init) {
1253 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1254 		if (err != 0)
1255 			goto err_out5;
1256 	}
1257 
1258 	if (tca[TCA_STAB]) {
1259 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1260 		if (IS_ERR(stab)) {
1261 			err = PTR_ERR(stab);
1262 			goto err_out4;
1263 		}
1264 		rcu_assign_pointer(sch->stab, stab);
1265 	}
1266 	if (tca[TCA_RATE]) {
1267 		seqcount_t *running;
1268 
1269 		err = -EOPNOTSUPP;
1270 		if (sch->flags & TCQ_F_MQROOT) {
1271 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1272 			goto err_out4;
1273 		}
1274 
1275 		if (sch->parent != TC_H_ROOT &&
1276 		    !(sch->flags & TCQ_F_INGRESS) &&
1277 		    (!p || !(p->flags & TCQ_F_MQROOT)))
1278 			running = qdisc_root_sleeping_running(sch);
1279 		else
1280 			running = &sch->running;
1281 
1282 		err = gen_new_estimator(&sch->bstats,
1283 					sch->cpu_bstats,
1284 					&sch->rate_est,
1285 					NULL,
1286 					running,
1287 					tca[TCA_RATE]);
1288 		if (err) {
1289 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1290 			goto err_out4;
1291 		}
1292 	}
1293 
1294 	qdisc_hash_add(sch, false);
1295 	trace_qdisc_create(ops, dev, parent);
1296 
1297 	return sch;
1298 
1299 err_out5:
1300 	/* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1301 	if (ops->destroy)
1302 		ops->destroy(sch);
1303 err_out3:
1304 	dev_put(dev);
1305 	qdisc_free(sch);
1306 err_out2:
1307 	module_put(ops->owner);
1308 err_out:
1309 	*errp = err;
1310 	return NULL;
1311 
1312 err_out4:
1313 	/*
1314 	 * Any broken qdiscs that would require a ops->reset() here?
1315 	 * The qdisc was never in action so it shouldn't be necessary.
1316 	 */
1317 	qdisc_put_stab(rtnl_dereference(sch->stab));
1318 	if (ops->destroy)
1319 		ops->destroy(sch);
1320 	goto err_out3;
1321 }
1322 
1323 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1324 			struct netlink_ext_ack *extack)
1325 {
1326 	struct qdisc_size_table *ostab, *stab = NULL;
1327 	int err = 0;
1328 
1329 	if (tca[TCA_OPTIONS]) {
1330 		if (!sch->ops->change) {
1331 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1332 			return -EINVAL;
1333 		}
1334 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1335 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1336 			return -EOPNOTSUPP;
1337 		}
1338 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1339 		if (err)
1340 			return err;
1341 	}
1342 
1343 	if (tca[TCA_STAB]) {
1344 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1345 		if (IS_ERR(stab))
1346 			return PTR_ERR(stab);
1347 	}
1348 
1349 	ostab = rtnl_dereference(sch->stab);
1350 	rcu_assign_pointer(sch->stab, stab);
1351 	qdisc_put_stab(ostab);
1352 
1353 	if (tca[TCA_RATE]) {
1354 		/* NB: ignores errors from replace_estimator
1355 		   because change can't be undone. */
1356 		if (sch->flags & TCQ_F_MQROOT)
1357 			goto out;
1358 		gen_replace_estimator(&sch->bstats,
1359 				      sch->cpu_bstats,
1360 				      &sch->rate_est,
1361 				      NULL,
1362 				      qdisc_root_sleeping_running(sch),
1363 				      tca[TCA_RATE]);
1364 	}
1365 out:
1366 	return 0;
1367 }
1368 
1369 struct check_loop_arg {
1370 	struct qdisc_walker	w;
1371 	struct Qdisc		*p;
1372 	int			depth;
1373 };
1374 
1375 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1376 			 struct qdisc_walker *w);
1377 
1378 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1379 {
1380 	struct check_loop_arg	arg;
1381 
1382 	if (q->ops->cl_ops == NULL)
1383 		return 0;
1384 
1385 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1386 	arg.w.fn = check_loop_fn;
1387 	arg.depth = depth;
1388 	arg.p = p;
1389 	q->ops->cl_ops->walk(q, &arg.w);
1390 	return arg.w.stop ? -ELOOP : 0;
1391 }
1392 
1393 static int
1394 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1395 {
1396 	struct Qdisc *leaf;
1397 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1398 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1399 
1400 	leaf = cops->leaf(q, cl);
1401 	if (leaf) {
1402 		if (leaf == arg->p || arg->depth > 7)
1403 			return -ELOOP;
1404 		return check_loop(leaf, arg->p, arg->depth + 1);
1405 	}
1406 	return 0;
1407 }
1408 
1409 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1410 	[TCA_KIND]		= { .type = NLA_STRING },
1411 	[TCA_RATE]		= { .type = NLA_BINARY,
1412 				    .len = sizeof(struct tc_estimator) },
1413 	[TCA_STAB]		= { .type = NLA_NESTED },
1414 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1415 	[TCA_CHAIN]		= { .type = NLA_U32 },
1416 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1417 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1418 };
1419 
1420 /*
1421  * Delete/get qdisc.
1422  */
1423 
1424 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1425 			struct netlink_ext_ack *extack)
1426 {
1427 	struct net *net = sock_net(skb->sk);
1428 	struct tcmsg *tcm = nlmsg_data(n);
1429 	struct nlattr *tca[TCA_MAX + 1];
1430 	struct net_device *dev;
1431 	u32 clid;
1432 	struct Qdisc *q = NULL;
1433 	struct Qdisc *p = NULL;
1434 	int err;
1435 
1436 	if ((n->nlmsg_type != RTM_GETQDISC) &&
1437 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1438 		return -EPERM;
1439 
1440 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1441 				     rtm_tca_policy, extack);
1442 	if (err < 0)
1443 		return err;
1444 
1445 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1446 	if (!dev)
1447 		return -ENODEV;
1448 
1449 	clid = tcm->tcm_parent;
1450 	if (clid) {
1451 		if (clid != TC_H_ROOT) {
1452 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1453 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1454 				if (!p) {
1455 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1456 					return -ENOENT;
1457 				}
1458 				q = qdisc_leaf(p, clid);
1459 			} else if (dev_ingress_queue(dev)) {
1460 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1461 			}
1462 		} else {
1463 			q = dev->qdisc;
1464 		}
1465 		if (!q) {
1466 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1467 			return -ENOENT;
1468 		}
1469 
1470 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1471 			NL_SET_ERR_MSG(extack, "Invalid handle");
1472 			return -EINVAL;
1473 		}
1474 	} else {
1475 		q = qdisc_lookup(dev, tcm->tcm_handle);
1476 		if (!q) {
1477 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1478 			return -ENOENT;
1479 		}
1480 	}
1481 
1482 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1483 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1484 		return -EINVAL;
1485 	}
1486 
1487 	if (n->nlmsg_type == RTM_DELQDISC) {
1488 		if (!clid) {
1489 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1490 			return -EINVAL;
1491 		}
1492 		if (q->handle == 0) {
1493 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1494 			return -ENOENT;
1495 		}
1496 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1497 		if (err != 0)
1498 			return err;
1499 	} else {
1500 		qdisc_notify(net, skb, n, clid, NULL, q);
1501 	}
1502 	return 0;
1503 }
1504 
1505 /*
1506  * Create/change qdisc.
1507  */
1508 
1509 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1510 			   struct netlink_ext_ack *extack)
1511 {
1512 	struct net *net = sock_net(skb->sk);
1513 	struct tcmsg *tcm;
1514 	struct nlattr *tca[TCA_MAX + 1];
1515 	struct net_device *dev;
1516 	u32 clid;
1517 	struct Qdisc *q, *p;
1518 	int err;
1519 
1520 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1521 		return -EPERM;
1522 
1523 replay:
1524 	/* Reinit, just in case something touches this. */
1525 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1526 				     rtm_tca_policy, extack);
1527 	if (err < 0)
1528 		return err;
1529 
1530 	tcm = nlmsg_data(n);
1531 	clid = tcm->tcm_parent;
1532 	q = p = NULL;
1533 
1534 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1535 	if (!dev)
1536 		return -ENODEV;
1537 
1538 
1539 	if (clid) {
1540 		if (clid != TC_H_ROOT) {
1541 			if (clid != TC_H_INGRESS) {
1542 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1543 				if (!p) {
1544 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1545 					return -ENOENT;
1546 				}
1547 				q = qdisc_leaf(p, clid);
1548 			} else if (dev_ingress_queue_create(dev)) {
1549 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1550 			}
1551 		} else {
1552 			q = dev->qdisc;
1553 		}
1554 
1555 		/* It may be default qdisc, ignore it */
1556 		if (q && q->handle == 0)
1557 			q = NULL;
1558 
1559 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1560 			if (tcm->tcm_handle) {
1561 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1562 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1563 					return -EEXIST;
1564 				}
1565 				if (TC_H_MIN(tcm->tcm_handle)) {
1566 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1567 					return -EINVAL;
1568 				}
1569 				q = qdisc_lookup(dev, tcm->tcm_handle);
1570 				if (!q)
1571 					goto create_n_graft;
1572 				if (n->nlmsg_flags & NLM_F_EXCL) {
1573 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1574 					return -EEXIST;
1575 				}
1576 				if (tca[TCA_KIND] &&
1577 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1578 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1579 					return -EINVAL;
1580 				}
1581 				if (q == p ||
1582 				    (p && check_loop(q, p, 0))) {
1583 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1584 					return -ELOOP;
1585 				}
1586 				qdisc_refcount_inc(q);
1587 				goto graft;
1588 			} else {
1589 				if (!q)
1590 					goto create_n_graft;
1591 
1592 				/* This magic test requires explanation.
1593 				 *
1594 				 *   We know, that some child q is already
1595 				 *   attached to this parent and have choice:
1596 				 *   either to change it or to create/graft new one.
1597 				 *
1598 				 *   1. We are allowed to create/graft only
1599 				 *   if CREATE and REPLACE flags are set.
1600 				 *
1601 				 *   2. If EXCL is set, requestor wanted to say,
1602 				 *   that qdisc tcm_handle is not expected
1603 				 *   to exist, so that we choose create/graft too.
1604 				 *
1605 				 *   3. The last case is when no flags are set.
1606 				 *   Alas, it is sort of hole in API, we
1607 				 *   cannot decide what to do unambiguously.
1608 				 *   For now we select create/graft, if
1609 				 *   user gave KIND, which does not match existing.
1610 				 */
1611 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1612 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1613 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1614 				     (tca[TCA_KIND] &&
1615 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1616 					goto create_n_graft;
1617 			}
1618 		}
1619 	} else {
1620 		if (!tcm->tcm_handle) {
1621 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1622 			return -EINVAL;
1623 		}
1624 		q = qdisc_lookup(dev, tcm->tcm_handle);
1625 	}
1626 
1627 	/* Change qdisc parameters */
1628 	if (!q) {
1629 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1630 		return -ENOENT;
1631 	}
1632 	if (n->nlmsg_flags & NLM_F_EXCL) {
1633 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1634 		return -EEXIST;
1635 	}
1636 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1637 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1638 		return -EINVAL;
1639 	}
1640 	err = qdisc_change(q, tca, extack);
1641 	if (err == 0)
1642 		qdisc_notify(net, skb, n, clid, NULL, q);
1643 	return err;
1644 
1645 create_n_graft:
1646 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1647 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1648 		return -ENOENT;
1649 	}
1650 	if (clid == TC_H_INGRESS) {
1651 		if (dev_ingress_queue(dev)) {
1652 			q = qdisc_create(dev, dev_ingress_queue(dev), p,
1653 					 tcm->tcm_parent, tcm->tcm_parent,
1654 					 tca, &err, extack);
1655 		} else {
1656 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1657 			err = -ENOENT;
1658 		}
1659 	} else {
1660 		struct netdev_queue *dev_queue;
1661 
1662 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1663 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1664 		else if (p)
1665 			dev_queue = p->dev_queue;
1666 		else
1667 			dev_queue = netdev_get_tx_queue(dev, 0);
1668 
1669 		q = qdisc_create(dev, dev_queue, p,
1670 				 tcm->tcm_parent, tcm->tcm_handle,
1671 				 tca, &err, extack);
1672 	}
1673 	if (q == NULL) {
1674 		if (err == -EAGAIN)
1675 			goto replay;
1676 		return err;
1677 	}
1678 
1679 graft:
1680 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1681 	if (err) {
1682 		if (q)
1683 			qdisc_put(q);
1684 		return err;
1685 	}
1686 
1687 	return 0;
1688 }
1689 
1690 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1691 			      struct netlink_callback *cb,
1692 			      int *q_idx_p, int s_q_idx, bool recur,
1693 			      bool dump_invisible)
1694 {
1695 	int ret = 0, q_idx = *q_idx_p;
1696 	struct Qdisc *q;
1697 	int b;
1698 
1699 	if (!root)
1700 		return 0;
1701 
1702 	q = root;
1703 	if (q_idx < s_q_idx) {
1704 		q_idx++;
1705 	} else {
1706 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1707 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1708 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1709 				  RTM_NEWQDISC) <= 0)
1710 			goto done;
1711 		q_idx++;
1712 	}
1713 
1714 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1715 	 * itself has already been dumped.
1716 	 *
1717 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1718 	 * qdisc hashtable, we don't want to hit it again
1719 	 */
1720 	if (!qdisc_dev(root) || !recur)
1721 		goto out;
1722 
1723 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1724 		if (q_idx < s_q_idx) {
1725 			q_idx++;
1726 			continue;
1727 		}
1728 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1729 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1730 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1731 				  RTM_NEWQDISC) <= 0)
1732 			goto done;
1733 		q_idx++;
1734 	}
1735 
1736 out:
1737 	*q_idx_p = q_idx;
1738 	return ret;
1739 done:
1740 	ret = -1;
1741 	goto out;
1742 }
1743 
1744 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1745 {
1746 	struct net *net = sock_net(skb->sk);
1747 	int idx, q_idx;
1748 	int s_idx, s_q_idx;
1749 	struct net_device *dev;
1750 	const struct nlmsghdr *nlh = cb->nlh;
1751 	struct nlattr *tca[TCA_MAX + 1];
1752 	int err;
1753 
1754 	s_idx = cb->args[0];
1755 	s_q_idx = q_idx = cb->args[1];
1756 
1757 	idx = 0;
1758 	ASSERT_RTNL();
1759 
1760 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1761 				     rtm_tca_policy, cb->extack);
1762 	if (err < 0)
1763 		return err;
1764 
1765 	for_each_netdev(net, dev) {
1766 		struct netdev_queue *dev_queue;
1767 
1768 		if (idx < s_idx)
1769 			goto cont;
1770 		if (idx > s_idx)
1771 			s_q_idx = 0;
1772 		q_idx = 0;
1773 
1774 		if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1775 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1776 			goto done;
1777 
1778 		dev_queue = dev_ingress_queue(dev);
1779 		if (dev_queue &&
1780 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1781 				       &q_idx, s_q_idx, false,
1782 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1783 			goto done;
1784 
1785 cont:
1786 		idx++;
1787 	}
1788 
1789 done:
1790 	cb->args[0] = idx;
1791 	cb->args[1] = q_idx;
1792 
1793 	return skb->len;
1794 }
1795 
1796 
1797 
1798 /************************************************
1799  *	Traffic classes manipulation.		*
1800  ************************************************/
1801 
1802 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1803 			  unsigned long cl,
1804 			  u32 portid, u32 seq, u16 flags, int event)
1805 {
1806 	struct tcmsg *tcm;
1807 	struct nlmsghdr  *nlh;
1808 	unsigned char *b = skb_tail_pointer(skb);
1809 	struct gnet_dump d;
1810 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1811 
1812 	cond_resched();
1813 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1814 	if (!nlh)
1815 		goto out_nlmsg_trim;
1816 	tcm = nlmsg_data(nlh);
1817 	tcm->tcm_family = AF_UNSPEC;
1818 	tcm->tcm__pad1 = 0;
1819 	tcm->tcm__pad2 = 0;
1820 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1821 	tcm->tcm_parent = q->handle;
1822 	tcm->tcm_handle = q->handle;
1823 	tcm->tcm_info = 0;
1824 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1825 		goto nla_put_failure;
1826 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1827 		goto nla_put_failure;
1828 
1829 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1830 					 NULL, &d, TCA_PAD) < 0)
1831 		goto nla_put_failure;
1832 
1833 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1834 		goto nla_put_failure;
1835 
1836 	if (gnet_stats_finish_copy(&d) < 0)
1837 		goto nla_put_failure;
1838 
1839 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1840 	return skb->len;
1841 
1842 out_nlmsg_trim:
1843 nla_put_failure:
1844 	nlmsg_trim(skb, b);
1845 	return -1;
1846 }
1847 
1848 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1849 			 struct nlmsghdr *n, struct Qdisc *q,
1850 			 unsigned long cl, int event)
1851 {
1852 	struct sk_buff *skb;
1853 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1854 
1855 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1856 	if (!skb)
1857 		return -ENOBUFS;
1858 
1859 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1860 		kfree_skb(skb);
1861 		return -EINVAL;
1862 	}
1863 
1864 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1865 			      n->nlmsg_flags & NLM_F_ECHO);
1866 }
1867 
1868 static int tclass_del_notify(struct net *net,
1869 			     const struct Qdisc_class_ops *cops,
1870 			     struct sk_buff *oskb, struct nlmsghdr *n,
1871 			     struct Qdisc *q, unsigned long cl,
1872 			     struct netlink_ext_ack *extack)
1873 {
1874 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1875 	struct sk_buff *skb;
1876 	int err = 0;
1877 
1878 	if (!cops->delete)
1879 		return -EOPNOTSUPP;
1880 
1881 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1882 	if (!skb)
1883 		return -ENOBUFS;
1884 
1885 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1886 			   RTM_DELTCLASS) < 0) {
1887 		kfree_skb(skb);
1888 		return -EINVAL;
1889 	}
1890 
1891 	err = cops->delete(q, cl, extack);
1892 	if (err) {
1893 		kfree_skb(skb);
1894 		return err;
1895 	}
1896 
1897 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1898 			     n->nlmsg_flags & NLM_F_ECHO);
1899 	return err;
1900 }
1901 
1902 #ifdef CONFIG_NET_CLS
1903 
1904 struct tcf_bind_args {
1905 	struct tcf_walker w;
1906 	unsigned long base;
1907 	unsigned long cl;
1908 	u32 classid;
1909 };
1910 
1911 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1912 {
1913 	struct tcf_bind_args *a = (void *)arg;
1914 
1915 	if (tp->ops->bind_class) {
1916 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1917 
1918 		sch_tree_lock(q);
1919 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1920 		sch_tree_unlock(q);
1921 	}
1922 	return 0;
1923 }
1924 
1925 struct tc_bind_class_args {
1926 	struct qdisc_walker w;
1927 	unsigned long new_cl;
1928 	u32 portid;
1929 	u32 clid;
1930 };
1931 
1932 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1933 				struct qdisc_walker *w)
1934 {
1935 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1936 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1937 	struct tcf_block *block;
1938 	struct tcf_chain *chain;
1939 
1940 	block = cops->tcf_block(q, cl, NULL);
1941 	if (!block)
1942 		return 0;
1943 	for (chain = tcf_get_next_chain(block, NULL);
1944 	     chain;
1945 	     chain = tcf_get_next_chain(block, chain)) {
1946 		struct tcf_proto *tp;
1947 
1948 		for (tp = tcf_get_next_proto(chain, NULL);
1949 		     tp; tp = tcf_get_next_proto(chain, tp)) {
1950 			struct tcf_bind_args arg = {};
1951 
1952 			arg.w.fn = tcf_node_bind;
1953 			arg.classid = a->clid;
1954 			arg.base = cl;
1955 			arg.cl = a->new_cl;
1956 			tp->ops->walk(tp, &arg.w, true);
1957 		}
1958 	}
1959 
1960 	return 0;
1961 }
1962 
1963 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1964 			   unsigned long new_cl)
1965 {
1966 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1967 	struct tc_bind_class_args args = {};
1968 
1969 	if (!cops->tcf_block)
1970 		return;
1971 	args.portid = portid;
1972 	args.clid = clid;
1973 	args.new_cl = new_cl;
1974 	args.w.fn = tc_bind_class_walker;
1975 	q->ops->cl_ops->walk(q, &args.w);
1976 }
1977 
1978 #else
1979 
1980 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1981 			   unsigned long new_cl)
1982 {
1983 }
1984 
1985 #endif
1986 
1987 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1988 			 struct netlink_ext_ack *extack)
1989 {
1990 	struct net *net = sock_net(skb->sk);
1991 	struct tcmsg *tcm = nlmsg_data(n);
1992 	struct nlattr *tca[TCA_MAX + 1];
1993 	struct net_device *dev;
1994 	struct Qdisc *q = NULL;
1995 	const struct Qdisc_class_ops *cops;
1996 	unsigned long cl = 0;
1997 	unsigned long new_cl;
1998 	u32 portid;
1999 	u32 clid;
2000 	u32 qid;
2001 	int err;
2002 
2003 	if ((n->nlmsg_type != RTM_GETTCLASS) &&
2004 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2005 		return -EPERM;
2006 
2007 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2008 				     rtm_tca_policy, extack);
2009 	if (err < 0)
2010 		return err;
2011 
2012 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2013 	if (!dev)
2014 		return -ENODEV;
2015 
2016 	/*
2017 	   parent == TC_H_UNSPEC - unspecified parent.
2018 	   parent == TC_H_ROOT   - class is root, which has no parent.
2019 	   parent == X:0	 - parent is root class.
2020 	   parent == X:Y	 - parent is a node in hierarchy.
2021 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2022 
2023 	   handle == 0:0	 - generate handle from kernel pool.
2024 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2025 	   handle == X:Y	 - clear.
2026 	   handle == X:0	 - root class.
2027 	 */
2028 
2029 	/* Step 1. Determine qdisc handle X:0 */
2030 
2031 	portid = tcm->tcm_parent;
2032 	clid = tcm->tcm_handle;
2033 	qid = TC_H_MAJ(clid);
2034 
2035 	if (portid != TC_H_ROOT) {
2036 		u32 qid1 = TC_H_MAJ(portid);
2037 
2038 		if (qid && qid1) {
2039 			/* If both majors are known, they must be identical. */
2040 			if (qid != qid1)
2041 				return -EINVAL;
2042 		} else if (qid1) {
2043 			qid = qid1;
2044 		} else if (qid == 0)
2045 			qid = dev->qdisc->handle;
2046 
2047 		/* Now qid is genuine qdisc handle consistent
2048 		 * both with parent and child.
2049 		 *
2050 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2051 		 */
2052 		if (portid)
2053 			portid = TC_H_MAKE(qid, portid);
2054 	} else {
2055 		if (qid == 0)
2056 			qid = dev->qdisc->handle;
2057 	}
2058 
2059 	/* OK. Locate qdisc */
2060 	q = qdisc_lookup(dev, qid);
2061 	if (!q)
2062 		return -ENOENT;
2063 
2064 	/* An check that it supports classes */
2065 	cops = q->ops->cl_ops;
2066 	if (cops == NULL)
2067 		return -EINVAL;
2068 
2069 	/* Now try to get class */
2070 	if (clid == 0) {
2071 		if (portid == TC_H_ROOT)
2072 			clid = qid;
2073 	} else
2074 		clid = TC_H_MAKE(qid, clid);
2075 
2076 	if (clid)
2077 		cl = cops->find(q, clid);
2078 
2079 	if (cl == 0) {
2080 		err = -ENOENT;
2081 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2082 		    !(n->nlmsg_flags & NLM_F_CREATE))
2083 			goto out;
2084 	} else {
2085 		switch (n->nlmsg_type) {
2086 		case RTM_NEWTCLASS:
2087 			err = -EEXIST;
2088 			if (n->nlmsg_flags & NLM_F_EXCL)
2089 				goto out;
2090 			break;
2091 		case RTM_DELTCLASS:
2092 			err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2093 			/* Unbind the class with flilters with 0 */
2094 			tc_bind_tclass(q, portid, clid, 0);
2095 			goto out;
2096 		case RTM_GETTCLASS:
2097 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
2098 			goto out;
2099 		default:
2100 			err = -EINVAL;
2101 			goto out;
2102 		}
2103 	}
2104 
2105 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2106 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2107 		return -EOPNOTSUPP;
2108 	}
2109 
2110 	new_cl = cl;
2111 	err = -EOPNOTSUPP;
2112 	if (cops->change)
2113 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2114 	if (err == 0) {
2115 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
2116 		/* We just create a new class, need to do reverse binding. */
2117 		if (cl != new_cl)
2118 			tc_bind_tclass(q, portid, clid, new_cl);
2119 	}
2120 out:
2121 	return err;
2122 }
2123 
2124 struct qdisc_dump_args {
2125 	struct qdisc_walker	w;
2126 	struct sk_buff		*skb;
2127 	struct netlink_callback	*cb;
2128 };
2129 
2130 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2131 			    struct qdisc_walker *arg)
2132 {
2133 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2134 
2135 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2136 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2137 			      RTM_NEWTCLASS);
2138 }
2139 
2140 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2141 				struct tcmsg *tcm, struct netlink_callback *cb,
2142 				int *t_p, int s_t)
2143 {
2144 	struct qdisc_dump_args arg;
2145 
2146 	if (tc_qdisc_dump_ignore(q, false) ||
2147 	    *t_p < s_t || !q->ops->cl_ops ||
2148 	    (tcm->tcm_parent &&
2149 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2150 		(*t_p)++;
2151 		return 0;
2152 	}
2153 	if (*t_p > s_t)
2154 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2155 	arg.w.fn = qdisc_class_dump;
2156 	arg.skb = skb;
2157 	arg.cb = cb;
2158 	arg.w.stop  = 0;
2159 	arg.w.skip = cb->args[1];
2160 	arg.w.count = 0;
2161 	q->ops->cl_ops->walk(q, &arg.w);
2162 	cb->args[1] = arg.w.count;
2163 	if (arg.w.stop)
2164 		return -1;
2165 	(*t_p)++;
2166 	return 0;
2167 }
2168 
2169 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2170 			       struct tcmsg *tcm, struct netlink_callback *cb,
2171 			       int *t_p, int s_t, bool recur)
2172 {
2173 	struct Qdisc *q;
2174 	int b;
2175 
2176 	if (!root)
2177 		return 0;
2178 
2179 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2180 		return -1;
2181 
2182 	if (!qdisc_dev(root) || !recur)
2183 		return 0;
2184 
2185 	if (tcm->tcm_parent) {
2186 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2187 		if (q && q != root &&
2188 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2189 			return -1;
2190 		return 0;
2191 	}
2192 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2193 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2194 			return -1;
2195 	}
2196 
2197 	return 0;
2198 }
2199 
2200 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2201 {
2202 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2203 	struct net *net = sock_net(skb->sk);
2204 	struct netdev_queue *dev_queue;
2205 	struct net_device *dev;
2206 	int t, s_t;
2207 
2208 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2209 		return 0;
2210 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2211 	if (!dev)
2212 		return 0;
2213 
2214 	s_t = cb->args[0];
2215 	t = 0;
2216 
2217 	if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t, true) < 0)
2218 		goto done;
2219 
2220 	dev_queue = dev_ingress_queue(dev);
2221 	if (dev_queue &&
2222 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2223 				&t, s_t, false) < 0)
2224 		goto done;
2225 
2226 done:
2227 	cb->args[0] = t;
2228 
2229 	dev_put(dev);
2230 	return skb->len;
2231 }
2232 
2233 #ifdef CONFIG_PROC_FS
2234 static int psched_show(struct seq_file *seq, void *v)
2235 {
2236 	seq_printf(seq, "%08x %08x %08x %08x\n",
2237 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2238 		   1000000,
2239 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2240 
2241 	return 0;
2242 }
2243 
2244 static int __net_init psched_net_init(struct net *net)
2245 {
2246 	struct proc_dir_entry *e;
2247 
2248 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2249 	if (e == NULL)
2250 		return -ENOMEM;
2251 
2252 	return 0;
2253 }
2254 
2255 static void __net_exit psched_net_exit(struct net *net)
2256 {
2257 	remove_proc_entry("psched", net->proc_net);
2258 }
2259 #else
2260 static int __net_init psched_net_init(struct net *net)
2261 {
2262 	return 0;
2263 }
2264 
2265 static void __net_exit psched_net_exit(struct net *net)
2266 {
2267 }
2268 #endif
2269 
2270 static struct pernet_operations psched_net_ops = {
2271 	.init = psched_net_init,
2272 	.exit = psched_net_exit,
2273 };
2274 
2275 static int __init pktsched_init(void)
2276 {
2277 	int err;
2278 
2279 	err = register_pernet_subsys(&psched_net_ops);
2280 	if (err) {
2281 		pr_err("pktsched_init: "
2282 		       "cannot initialize per netns operations\n");
2283 		return err;
2284 	}
2285 
2286 	register_qdisc(&pfifo_fast_ops);
2287 	register_qdisc(&pfifo_qdisc_ops);
2288 	register_qdisc(&bfifo_qdisc_ops);
2289 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2290 	register_qdisc(&mq_qdisc_ops);
2291 	register_qdisc(&noqueue_qdisc_ops);
2292 
2293 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2294 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2295 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2296 		      0);
2297 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2298 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2299 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2300 		      0);
2301 
2302 	return 0;
2303 }
2304 
2305 subsys_initcall(pktsched_init);
2306