xref: /openbmc/linux/net/sched/sch_api.c (revision ca637c0e)
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 void 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 
190 	WARN(err, "unregister qdisc(%s) failed\n", qops->id);
191 }
192 EXPORT_SYMBOL(unregister_qdisc);
193 
194 /* Get default qdisc if not otherwise specified */
195 void qdisc_get_default(char *name, size_t len)
196 {
197 	read_lock(&qdisc_mod_lock);
198 	strscpy(name, default_qdisc_ops->id, len);
199 	read_unlock(&qdisc_mod_lock);
200 }
201 
202 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
203 {
204 	struct Qdisc_ops *q = NULL;
205 
206 	for (q = qdisc_base; q; q = q->next) {
207 		if (!strcmp(name, q->id)) {
208 			if (!try_module_get(q->owner))
209 				q = NULL;
210 			break;
211 		}
212 	}
213 
214 	return q;
215 }
216 
217 /* Set new default qdisc to use */
218 int qdisc_set_default(const char *name)
219 {
220 	const struct Qdisc_ops *ops;
221 
222 	if (!capable(CAP_NET_ADMIN))
223 		return -EPERM;
224 
225 	write_lock(&qdisc_mod_lock);
226 	ops = qdisc_lookup_default(name);
227 	if (!ops) {
228 		/* Not found, drop lock and try to load module */
229 		write_unlock(&qdisc_mod_lock);
230 		request_module("sch_%s", name);
231 		write_lock(&qdisc_mod_lock);
232 
233 		ops = qdisc_lookup_default(name);
234 	}
235 
236 	if (ops) {
237 		/* Set new default */
238 		module_put(default_qdisc_ops->owner);
239 		default_qdisc_ops = ops;
240 	}
241 	write_unlock(&qdisc_mod_lock);
242 
243 	return ops ? 0 : -ENOENT;
244 }
245 
246 #ifdef CONFIG_NET_SCH_DEFAULT
247 /* Set default value from kernel config */
248 static int __init sch_default_qdisc(void)
249 {
250 	return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
251 }
252 late_initcall(sch_default_qdisc);
253 #endif
254 
255 /* We know handle. Find qdisc among all qdisc's attached to device
256  * (root qdisc, all its children, children of children etc.)
257  * Note: caller either uses rtnl or rcu_read_lock()
258  */
259 
260 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
261 {
262 	struct Qdisc *q;
263 
264 	if (!qdisc_dev(root))
265 		return (root->handle == handle ? root : NULL);
266 
267 	if (!(root->flags & TCQ_F_BUILTIN) &&
268 	    root->handle == handle)
269 		return root;
270 
271 	hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle,
272 				   lockdep_rtnl_is_held()) {
273 		if (q->handle == handle)
274 			return q;
275 	}
276 	return NULL;
277 }
278 
279 void qdisc_hash_add(struct Qdisc *q, bool invisible)
280 {
281 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
282 		ASSERT_RTNL();
283 		hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
284 		if (invisible)
285 			q->flags |= TCQ_F_INVISIBLE;
286 	}
287 }
288 EXPORT_SYMBOL(qdisc_hash_add);
289 
290 void qdisc_hash_del(struct Qdisc *q)
291 {
292 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
293 		ASSERT_RTNL();
294 		hash_del_rcu(&q->hash);
295 	}
296 }
297 EXPORT_SYMBOL(qdisc_hash_del);
298 
299 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
300 {
301 	struct Qdisc *q;
302 
303 	if (!handle)
304 		return NULL;
305 	q = qdisc_match_from_root(rtnl_dereference(dev->qdisc), handle);
306 	if (q)
307 		goto out;
308 
309 	if (dev_ingress_queue(dev))
310 		q = qdisc_match_from_root(
311 			dev_ingress_queue(dev)->qdisc_sleeping,
312 			handle);
313 out:
314 	return q;
315 }
316 
317 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
318 {
319 	struct netdev_queue *nq;
320 	struct Qdisc *q;
321 
322 	if (!handle)
323 		return NULL;
324 	q = qdisc_match_from_root(rcu_dereference(dev->qdisc), handle);
325 	if (q)
326 		goto out;
327 
328 	nq = dev_ingress_queue_rcu(dev);
329 	if (nq)
330 		q = qdisc_match_from_root(nq->qdisc_sleeping, handle);
331 out:
332 	return q;
333 }
334 
335 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
336 {
337 	unsigned long cl;
338 	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
339 
340 	if (cops == NULL)
341 		return NULL;
342 	cl = cops->find(p, classid);
343 
344 	if (cl == 0)
345 		return NULL;
346 	return cops->leaf(p, cl);
347 }
348 
349 /* Find queueing discipline by name */
350 
351 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
352 {
353 	struct Qdisc_ops *q = NULL;
354 
355 	if (kind) {
356 		read_lock(&qdisc_mod_lock);
357 		for (q = qdisc_base; q; q = q->next) {
358 			if (nla_strcmp(kind, q->id) == 0) {
359 				if (!try_module_get(q->owner))
360 					q = NULL;
361 				break;
362 			}
363 		}
364 		read_unlock(&qdisc_mod_lock);
365 	}
366 	return q;
367 }
368 
369 /* The linklayer setting were not transferred from iproute2, in older
370  * versions, and the rate tables lookup systems have been dropped in
371  * the kernel. To keep backward compatible with older iproute2 tc
372  * utils, we detect the linklayer setting by detecting if the rate
373  * table were modified.
374  *
375  * For linklayer ATM table entries, the rate table will be aligned to
376  * 48 bytes, thus some table entries will contain the same value.  The
377  * mpu (min packet unit) is also encoded into the old rate table, thus
378  * starting from the mpu, we find low and high table entries for
379  * mapping this cell.  If these entries contain the same value, when
380  * the rate tables have been modified for linklayer ATM.
381  *
382  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
383  * and then roundup to the next cell, calc the table entry one below,
384  * and compare.
385  */
386 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
387 {
388 	int low       = roundup(r->mpu, 48);
389 	int high      = roundup(low+1, 48);
390 	int cell_low  = low >> r->cell_log;
391 	int cell_high = (high >> r->cell_log) - 1;
392 
393 	/* rtab is too inaccurate at rates > 100Mbit/s */
394 	if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
395 		pr_debug("TC linklayer: Giving up ATM detection\n");
396 		return TC_LINKLAYER_ETHERNET;
397 	}
398 
399 	if ((cell_high > cell_low) && (cell_high < 256)
400 	    && (rtab[cell_low] == rtab[cell_high])) {
401 		pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
402 			 cell_low, cell_high, rtab[cell_high]);
403 		return TC_LINKLAYER_ATM;
404 	}
405 	return TC_LINKLAYER_ETHERNET;
406 }
407 
408 static struct qdisc_rate_table *qdisc_rtab_list;
409 
410 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
411 					struct nlattr *tab,
412 					struct netlink_ext_ack *extack)
413 {
414 	struct qdisc_rate_table *rtab;
415 
416 	if (tab == NULL || r->rate == 0 ||
417 	    r->cell_log == 0 || r->cell_log >= 32 ||
418 	    nla_len(tab) != TC_RTAB_SIZE) {
419 		NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
420 		return NULL;
421 	}
422 
423 	for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
424 		if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
425 		    !memcmp(&rtab->data, nla_data(tab), 1024)) {
426 			rtab->refcnt++;
427 			return rtab;
428 		}
429 	}
430 
431 	rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
432 	if (rtab) {
433 		rtab->rate = *r;
434 		rtab->refcnt = 1;
435 		memcpy(rtab->data, nla_data(tab), 1024);
436 		if (r->linklayer == TC_LINKLAYER_UNAWARE)
437 			r->linklayer = __detect_linklayer(r, rtab->data);
438 		rtab->next = qdisc_rtab_list;
439 		qdisc_rtab_list = rtab;
440 	} else {
441 		NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
442 	}
443 	return rtab;
444 }
445 EXPORT_SYMBOL(qdisc_get_rtab);
446 
447 void qdisc_put_rtab(struct qdisc_rate_table *tab)
448 {
449 	struct qdisc_rate_table *rtab, **rtabp;
450 
451 	if (!tab || --tab->refcnt)
452 		return;
453 
454 	for (rtabp = &qdisc_rtab_list;
455 	     (rtab = *rtabp) != NULL;
456 	     rtabp = &rtab->next) {
457 		if (rtab == tab) {
458 			*rtabp = rtab->next;
459 			kfree(rtab);
460 			return;
461 		}
462 	}
463 }
464 EXPORT_SYMBOL(qdisc_put_rtab);
465 
466 static LIST_HEAD(qdisc_stab_list);
467 
468 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
469 	[TCA_STAB_BASE]	= { .len = sizeof(struct tc_sizespec) },
470 	[TCA_STAB_DATA] = { .type = NLA_BINARY },
471 };
472 
473 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
474 					       struct netlink_ext_ack *extack)
475 {
476 	struct nlattr *tb[TCA_STAB_MAX + 1];
477 	struct qdisc_size_table *stab;
478 	struct tc_sizespec *s;
479 	unsigned int tsize = 0;
480 	u16 *tab = NULL;
481 	int err;
482 
483 	err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
484 					  extack);
485 	if (err < 0)
486 		return ERR_PTR(err);
487 	if (!tb[TCA_STAB_BASE]) {
488 		NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
489 		return ERR_PTR(-EINVAL);
490 	}
491 
492 	s = nla_data(tb[TCA_STAB_BASE]);
493 
494 	if (s->tsize > 0) {
495 		if (!tb[TCA_STAB_DATA]) {
496 			NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
497 			return ERR_PTR(-EINVAL);
498 		}
499 		tab = nla_data(tb[TCA_STAB_DATA]);
500 		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
501 	}
502 
503 	if (tsize != s->tsize || (!tab && tsize > 0)) {
504 		NL_SET_ERR_MSG(extack, "Invalid size of size table");
505 		return ERR_PTR(-EINVAL);
506 	}
507 
508 	list_for_each_entry(stab, &qdisc_stab_list, list) {
509 		if (memcmp(&stab->szopts, s, sizeof(*s)))
510 			continue;
511 		if (tsize > 0 &&
512 		    memcmp(stab->data, tab, flex_array_size(stab, data, tsize)))
513 			continue;
514 		stab->refcnt++;
515 		return stab;
516 	}
517 
518 	if (s->size_log > STAB_SIZE_LOG_MAX ||
519 	    s->cell_log > STAB_SIZE_LOG_MAX) {
520 		NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
521 		return ERR_PTR(-EINVAL);
522 	}
523 
524 	stab = kmalloc(struct_size(stab, data, tsize), GFP_KERNEL);
525 	if (!stab)
526 		return ERR_PTR(-ENOMEM);
527 
528 	stab->refcnt = 1;
529 	stab->szopts = *s;
530 	if (tsize > 0)
531 		memcpy(stab->data, tab, flex_array_size(stab, data, tsize));
532 
533 	list_add_tail(&stab->list, &qdisc_stab_list);
534 
535 	return stab;
536 }
537 
538 void qdisc_put_stab(struct qdisc_size_table *tab)
539 {
540 	if (!tab)
541 		return;
542 
543 	if (--tab->refcnt == 0) {
544 		list_del(&tab->list);
545 		kfree_rcu(tab, rcu);
546 	}
547 }
548 EXPORT_SYMBOL(qdisc_put_stab);
549 
550 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
551 {
552 	struct nlattr *nest;
553 
554 	nest = nla_nest_start_noflag(skb, TCA_STAB);
555 	if (nest == NULL)
556 		goto nla_put_failure;
557 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
558 		goto nla_put_failure;
559 	nla_nest_end(skb, nest);
560 
561 	return skb->len;
562 
563 nla_put_failure:
564 	return -1;
565 }
566 
567 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
568 			       const struct qdisc_size_table *stab)
569 {
570 	int pkt_len, slot;
571 
572 	pkt_len = skb->len + stab->szopts.overhead;
573 	if (unlikely(!stab->szopts.tsize))
574 		goto out;
575 
576 	slot = pkt_len + stab->szopts.cell_align;
577 	if (unlikely(slot < 0))
578 		slot = 0;
579 
580 	slot >>= stab->szopts.cell_log;
581 	if (likely(slot < stab->szopts.tsize))
582 		pkt_len = stab->data[slot];
583 	else
584 		pkt_len = stab->data[stab->szopts.tsize - 1] *
585 				(slot / stab->szopts.tsize) +
586 				stab->data[slot % stab->szopts.tsize];
587 
588 	pkt_len <<= stab->szopts.size_log;
589 out:
590 	if (unlikely(pkt_len < 1))
591 		pkt_len = 1;
592 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
593 }
594 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
595 
596 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
597 {
598 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
599 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
600 			txt, qdisc->ops->id, qdisc->handle >> 16);
601 		qdisc->flags |= TCQ_F_WARN_NONWC;
602 	}
603 }
604 EXPORT_SYMBOL(qdisc_warn_nonwc);
605 
606 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
607 {
608 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
609 						 timer);
610 
611 	rcu_read_lock();
612 	__netif_schedule(qdisc_root(wd->qdisc));
613 	rcu_read_unlock();
614 
615 	return HRTIMER_NORESTART;
616 }
617 
618 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
619 				 clockid_t clockid)
620 {
621 	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
622 	wd->timer.function = qdisc_watchdog;
623 	wd->qdisc = qdisc;
624 }
625 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
626 
627 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
628 {
629 	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
630 }
631 EXPORT_SYMBOL(qdisc_watchdog_init);
632 
633 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
634 				      u64 delta_ns)
635 {
636 	if (test_bit(__QDISC_STATE_DEACTIVATED,
637 		     &qdisc_root_sleeping(wd->qdisc)->state))
638 		return;
639 
640 	if (hrtimer_is_queued(&wd->timer)) {
641 		/* If timer is already set in [expires, expires + delta_ns],
642 		 * do not reprogram it.
643 		 */
644 		if (wd->last_expires - expires <= delta_ns)
645 			return;
646 	}
647 
648 	wd->last_expires = expires;
649 	hrtimer_start_range_ns(&wd->timer,
650 			       ns_to_ktime(expires),
651 			       delta_ns,
652 			       HRTIMER_MODE_ABS_PINNED);
653 }
654 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
655 
656 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
657 {
658 	hrtimer_cancel(&wd->timer);
659 }
660 EXPORT_SYMBOL(qdisc_watchdog_cancel);
661 
662 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
663 {
664 	struct hlist_head *h;
665 	unsigned int i;
666 
667 	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
668 
669 	if (h != NULL) {
670 		for (i = 0; i < n; i++)
671 			INIT_HLIST_HEAD(&h[i]);
672 	}
673 	return h;
674 }
675 
676 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
677 {
678 	struct Qdisc_class_common *cl;
679 	struct hlist_node *next;
680 	struct hlist_head *nhash, *ohash;
681 	unsigned int nsize, nmask, osize;
682 	unsigned int i, h;
683 
684 	/* Rehash when load factor exceeds 0.75 */
685 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
686 		return;
687 	nsize = clhash->hashsize * 2;
688 	nmask = nsize - 1;
689 	nhash = qdisc_class_hash_alloc(nsize);
690 	if (nhash == NULL)
691 		return;
692 
693 	ohash = clhash->hash;
694 	osize = clhash->hashsize;
695 
696 	sch_tree_lock(sch);
697 	for (i = 0; i < osize; i++) {
698 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
699 			h = qdisc_class_hash(cl->classid, nmask);
700 			hlist_add_head(&cl->hnode, &nhash[h]);
701 		}
702 	}
703 	clhash->hash     = nhash;
704 	clhash->hashsize = nsize;
705 	clhash->hashmask = nmask;
706 	sch_tree_unlock(sch);
707 
708 	kvfree(ohash);
709 }
710 EXPORT_SYMBOL(qdisc_class_hash_grow);
711 
712 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
713 {
714 	unsigned int size = 4;
715 
716 	clhash->hash = qdisc_class_hash_alloc(size);
717 	if (!clhash->hash)
718 		return -ENOMEM;
719 	clhash->hashsize  = size;
720 	clhash->hashmask  = size - 1;
721 	clhash->hashelems = 0;
722 	return 0;
723 }
724 EXPORT_SYMBOL(qdisc_class_hash_init);
725 
726 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
727 {
728 	kvfree(clhash->hash);
729 }
730 EXPORT_SYMBOL(qdisc_class_hash_destroy);
731 
732 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
733 			     struct Qdisc_class_common *cl)
734 {
735 	unsigned int h;
736 
737 	INIT_HLIST_NODE(&cl->hnode);
738 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
739 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
740 	clhash->hashelems++;
741 }
742 EXPORT_SYMBOL(qdisc_class_hash_insert);
743 
744 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
745 			     struct Qdisc_class_common *cl)
746 {
747 	hlist_del(&cl->hnode);
748 	clhash->hashelems--;
749 }
750 EXPORT_SYMBOL(qdisc_class_hash_remove);
751 
752 /* Allocate an unique handle from space managed by kernel
753  * Possible range is [8000-FFFF]:0000 (0x8000 values)
754  */
755 static u32 qdisc_alloc_handle(struct net_device *dev)
756 {
757 	int i = 0x8000;
758 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
759 
760 	do {
761 		autohandle += TC_H_MAKE(0x10000U, 0);
762 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
763 			autohandle = TC_H_MAKE(0x80000000U, 0);
764 		if (!qdisc_lookup(dev, autohandle))
765 			return autohandle;
766 		cond_resched();
767 	} while	(--i > 0);
768 
769 	return 0;
770 }
771 
772 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
773 {
774 	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
775 	const struct Qdisc_class_ops *cops;
776 	unsigned long cl;
777 	u32 parentid;
778 	bool notify;
779 	int drops;
780 
781 	if (n == 0 && len == 0)
782 		return;
783 	drops = max_t(int, n, 0);
784 	rcu_read_lock();
785 	while ((parentid = sch->parent)) {
786 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
787 			break;
788 
789 		if (sch->flags & TCQ_F_NOPARENT)
790 			break;
791 		/* Notify parent qdisc only if child qdisc becomes empty.
792 		 *
793 		 * If child was empty even before update then backlog
794 		 * counter is screwed and we skip notification because
795 		 * parent class is already passive.
796 		 *
797 		 * If the original child was offloaded then it is allowed
798 		 * to be seem as empty, so the parent is notified anyway.
799 		 */
800 		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
801 						       !qdisc_is_offloaded);
802 		/* TODO: perform the search on a per txq basis */
803 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
804 		if (sch == NULL) {
805 			WARN_ON_ONCE(parentid != TC_H_ROOT);
806 			break;
807 		}
808 		cops = sch->ops->cl_ops;
809 		if (notify && cops->qlen_notify) {
810 			cl = cops->find(sch, parentid);
811 			cops->qlen_notify(sch, cl);
812 		}
813 		sch->q.qlen -= n;
814 		sch->qstats.backlog -= len;
815 		__qdisc_qstats_drop(sch, drops);
816 	}
817 	rcu_read_unlock();
818 }
819 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
820 
821 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
822 			      void *type_data)
823 {
824 	struct net_device *dev = qdisc_dev(sch);
825 	int err;
826 
827 	sch->flags &= ~TCQ_F_OFFLOADED;
828 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
829 		return 0;
830 
831 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
832 	if (err == -EOPNOTSUPP)
833 		return 0;
834 
835 	if (!err)
836 		sch->flags |= TCQ_F_OFFLOADED;
837 
838 	return err;
839 }
840 EXPORT_SYMBOL(qdisc_offload_dump_helper);
841 
842 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
843 				struct Qdisc *new, struct Qdisc *old,
844 				enum tc_setup_type type, void *type_data,
845 				struct netlink_ext_ack *extack)
846 {
847 	bool any_qdisc_is_offloaded;
848 	int err;
849 
850 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
851 		return;
852 
853 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
854 
855 	/* Don't report error if the graft is part of destroy operation. */
856 	if (!err || !new || new == &noop_qdisc)
857 		return;
858 
859 	/* Don't report error if the parent, the old child and the new
860 	 * one are not offloaded.
861 	 */
862 	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
863 	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
864 	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
865 
866 	if (any_qdisc_is_offloaded)
867 		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
868 }
869 EXPORT_SYMBOL(qdisc_offload_graft_helper);
870 
871 void qdisc_offload_query_caps(struct net_device *dev,
872 			      enum tc_setup_type type,
873 			      void *caps, size_t caps_len)
874 {
875 	const struct net_device_ops *ops = dev->netdev_ops;
876 	struct tc_query_caps_base base = {
877 		.type = type,
878 		.caps = caps,
879 	};
880 
881 	memset(caps, 0, caps_len);
882 
883 	if (ops->ndo_setup_tc)
884 		ops->ndo_setup_tc(dev, TC_QUERY_CAPS, &base);
885 }
886 EXPORT_SYMBOL(qdisc_offload_query_caps);
887 
888 static void qdisc_offload_graft_root(struct net_device *dev,
889 				     struct Qdisc *new, struct Qdisc *old,
890 				     struct netlink_ext_ack *extack)
891 {
892 	struct tc_root_qopt_offload graft_offload = {
893 		.command	= TC_ROOT_GRAFT,
894 		.handle		= new ? new->handle : 0,
895 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
896 				  (old && old->flags & TCQ_F_INGRESS),
897 	};
898 
899 	qdisc_offload_graft_helper(dev, NULL, new, old,
900 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
901 }
902 
903 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
904 			 u32 portid, u32 seq, u16 flags, int event)
905 {
906 	struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
907 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
908 	struct tcmsg *tcm;
909 	struct nlmsghdr  *nlh;
910 	unsigned char *b = skb_tail_pointer(skb);
911 	struct gnet_dump d;
912 	struct qdisc_size_table *stab;
913 	u32 block_index;
914 	__u32 qlen;
915 
916 	cond_resched();
917 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
918 	if (!nlh)
919 		goto out_nlmsg_trim;
920 	tcm = nlmsg_data(nlh);
921 	tcm->tcm_family = AF_UNSPEC;
922 	tcm->tcm__pad1 = 0;
923 	tcm->tcm__pad2 = 0;
924 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
925 	tcm->tcm_parent = clid;
926 	tcm->tcm_handle = q->handle;
927 	tcm->tcm_info = refcount_read(&q->refcnt);
928 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
929 		goto nla_put_failure;
930 	if (q->ops->ingress_block_get) {
931 		block_index = q->ops->ingress_block_get(q);
932 		if (block_index &&
933 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
934 			goto nla_put_failure;
935 	}
936 	if (q->ops->egress_block_get) {
937 		block_index = q->ops->egress_block_get(q);
938 		if (block_index &&
939 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
940 			goto nla_put_failure;
941 	}
942 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
943 		goto nla_put_failure;
944 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
945 		goto nla_put_failure;
946 	qlen = qdisc_qlen_sum(q);
947 
948 	stab = rtnl_dereference(q->stab);
949 	if (stab && qdisc_dump_stab(skb, stab) < 0)
950 		goto nla_put_failure;
951 
952 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
953 					 NULL, &d, TCA_PAD) < 0)
954 		goto nla_put_failure;
955 
956 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
957 		goto nla_put_failure;
958 
959 	if (qdisc_is_percpu_stats(q)) {
960 		cpu_bstats = q->cpu_bstats;
961 		cpu_qstats = q->cpu_qstats;
962 	}
963 
964 	if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
965 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
966 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
967 		goto nla_put_failure;
968 
969 	if (gnet_stats_finish_copy(&d) < 0)
970 		goto nla_put_failure;
971 
972 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
973 	return skb->len;
974 
975 out_nlmsg_trim:
976 nla_put_failure:
977 	nlmsg_trim(skb, b);
978 	return -1;
979 }
980 
981 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
982 {
983 	if (q->flags & TCQ_F_BUILTIN)
984 		return true;
985 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
986 		return true;
987 
988 	return false;
989 }
990 
991 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
992 			struct nlmsghdr *n, u32 clid,
993 			struct Qdisc *old, struct Qdisc *new)
994 {
995 	struct sk_buff *skb;
996 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
997 
998 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
999 	if (!skb)
1000 		return -ENOBUFS;
1001 
1002 	if (old && !tc_qdisc_dump_ignore(old, false)) {
1003 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1004 				  0, RTM_DELQDISC) < 0)
1005 			goto err_out;
1006 	}
1007 	if (new && !tc_qdisc_dump_ignore(new, false)) {
1008 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1009 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1010 			goto err_out;
1011 	}
1012 
1013 	if (skb->len)
1014 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1015 				      n->nlmsg_flags & NLM_F_ECHO);
1016 
1017 err_out:
1018 	kfree_skb(skb);
1019 	return -EINVAL;
1020 }
1021 
1022 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1023 			       struct nlmsghdr *n, u32 clid,
1024 			       struct Qdisc *old, struct Qdisc *new)
1025 {
1026 	if (new || old)
1027 		qdisc_notify(net, skb, n, clid, old, new);
1028 
1029 	if (old)
1030 		qdisc_put(old);
1031 }
1032 
1033 static void qdisc_clear_nolock(struct Qdisc *sch)
1034 {
1035 	sch->flags &= ~TCQ_F_NOLOCK;
1036 	if (!(sch->flags & TCQ_F_CPUSTATS))
1037 		return;
1038 
1039 	free_percpu(sch->cpu_bstats);
1040 	free_percpu(sch->cpu_qstats);
1041 	sch->cpu_bstats = NULL;
1042 	sch->cpu_qstats = NULL;
1043 	sch->flags &= ~TCQ_F_CPUSTATS;
1044 }
1045 
1046 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1047  * to device "dev".
1048  *
1049  * When appropriate send a netlink notification using 'skb'
1050  * and "n".
1051  *
1052  * On success, destroy old qdisc.
1053  */
1054 
1055 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1056 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1057 		       struct Qdisc *new, struct Qdisc *old,
1058 		       struct netlink_ext_ack *extack)
1059 {
1060 	struct Qdisc *q = old;
1061 	struct net *net = dev_net(dev);
1062 
1063 	if (parent == NULL) {
1064 		unsigned int i, num_q, ingress;
1065 
1066 		ingress = 0;
1067 		num_q = dev->num_tx_queues;
1068 		if ((q && q->flags & TCQ_F_INGRESS) ||
1069 		    (new && new->flags & TCQ_F_INGRESS)) {
1070 			num_q = 1;
1071 			ingress = 1;
1072 			if (!dev_ingress_queue(dev)) {
1073 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1074 				return -ENOENT;
1075 			}
1076 		}
1077 
1078 		if (dev->flags & IFF_UP)
1079 			dev_deactivate(dev);
1080 
1081 		qdisc_offload_graft_root(dev, new, old, extack);
1082 
1083 		if (new && new->ops->attach && !ingress)
1084 			goto skip;
1085 
1086 		for (i = 0; i < num_q; i++) {
1087 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1088 
1089 			if (!ingress)
1090 				dev_queue = netdev_get_tx_queue(dev, i);
1091 
1092 			old = dev_graft_qdisc(dev_queue, new);
1093 			if (new && i > 0)
1094 				qdisc_refcount_inc(new);
1095 
1096 			if (!ingress)
1097 				qdisc_put(old);
1098 		}
1099 
1100 skip:
1101 		if (!ingress) {
1102 			old = rtnl_dereference(dev->qdisc);
1103 			if (new && !new->ops->attach)
1104 				qdisc_refcount_inc(new);
1105 			rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1106 
1107 			notify_and_destroy(net, skb, n, classid, old, new);
1108 
1109 			if (new && new->ops->attach)
1110 				new->ops->attach(new);
1111 		} else {
1112 			notify_and_destroy(net, skb, n, classid, old, new);
1113 		}
1114 
1115 		if (dev->flags & IFF_UP)
1116 			dev_activate(dev);
1117 	} else {
1118 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1119 		unsigned long cl;
1120 		int err;
1121 
1122 		/* Only support running class lockless if parent is lockless */
1123 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1124 			qdisc_clear_nolock(new);
1125 
1126 		if (!cops || !cops->graft)
1127 			return -EOPNOTSUPP;
1128 
1129 		cl = cops->find(parent, classid);
1130 		if (!cl) {
1131 			NL_SET_ERR_MSG(extack, "Specified class not found");
1132 			return -ENOENT;
1133 		}
1134 
1135 		err = cops->graft(parent, cl, new, &old, extack);
1136 		if (err)
1137 			return err;
1138 		notify_and_destroy(net, skb, n, classid, old, new);
1139 	}
1140 	return 0;
1141 }
1142 
1143 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1144 				   struct netlink_ext_ack *extack)
1145 {
1146 	u32 block_index;
1147 
1148 	if (tca[TCA_INGRESS_BLOCK]) {
1149 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1150 
1151 		if (!block_index) {
1152 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1153 			return -EINVAL;
1154 		}
1155 		if (!sch->ops->ingress_block_set) {
1156 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1157 			return -EOPNOTSUPP;
1158 		}
1159 		sch->ops->ingress_block_set(sch, block_index);
1160 	}
1161 	if (tca[TCA_EGRESS_BLOCK]) {
1162 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1163 
1164 		if (!block_index) {
1165 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1166 			return -EINVAL;
1167 		}
1168 		if (!sch->ops->egress_block_set) {
1169 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1170 			return -EOPNOTSUPP;
1171 		}
1172 		sch->ops->egress_block_set(sch, block_index);
1173 	}
1174 	return 0;
1175 }
1176 
1177 /*
1178    Allocate and initialize new qdisc.
1179 
1180    Parameters are passed via opt.
1181  */
1182 
1183 static struct Qdisc *qdisc_create(struct net_device *dev,
1184 				  struct netdev_queue *dev_queue,
1185 				  u32 parent, u32 handle,
1186 				  struct nlattr **tca, int *errp,
1187 				  struct netlink_ext_ack *extack)
1188 {
1189 	int err;
1190 	struct nlattr *kind = tca[TCA_KIND];
1191 	struct Qdisc *sch;
1192 	struct Qdisc_ops *ops;
1193 	struct qdisc_size_table *stab;
1194 
1195 	ops = qdisc_lookup_ops(kind);
1196 #ifdef CONFIG_MODULES
1197 	if (ops == NULL && kind != NULL) {
1198 		char name[IFNAMSIZ];
1199 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1200 			/* We dropped the RTNL semaphore in order to
1201 			 * perform the module load.  So, even if we
1202 			 * succeeded in loading the module we have to
1203 			 * tell the caller to replay the request.  We
1204 			 * indicate this using -EAGAIN.
1205 			 * We replay the request because the device may
1206 			 * go away in the mean time.
1207 			 */
1208 			rtnl_unlock();
1209 			request_module("sch_%s", name);
1210 			rtnl_lock();
1211 			ops = qdisc_lookup_ops(kind);
1212 			if (ops != NULL) {
1213 				/* We will try again qdisc_lookup_ops,
1214 				 * so don't keep a reference.
1215 				 */
1216 				module_put(ops->owner);
1217 				err = -EAGAIN;
1218 				goto err_out;
1219 			}
1220 		}
1221 	}
1222 #endif
1223 
1224 	err = -ENOENT;
1225 	if (!ops) {
1226 		NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1227 		goto err_out;
1228 	}
1229 
1230 	sch = qdisc_alloc(dev_queue, ops, extack);
1231 	if (IS_ERR(sch)) {
1232 		err = PTR_ERR(sch);
1233 		goto err_out2;
1234 	}
1235 
1236 	sch->parent = parent;
1237 
1238 	if (handle == TC_H_INGRESS) {
1239 		sch->flags |= TCQ_F_INGRESS;
1240 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1241 	} else {
1242 		if (handle == 0) {
1243 			handle = qdisc_alloc_handle(dev);
1244 			if (handle == 0) {
1245 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1246 				err = -ENOSPC;
1247 				goto err_out3;
1248 			}
1249 		}
1250 		if (!netif_is_multiqueue(dev))
1251 			sch->flags |= TCQ_F_ONETXQUEUE;
1252 	}
1253 
1254 	sch->handle = handle;
1255 
1256 	/* This exist to keep backward compatible with a userspace
1257 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1258 	 * facility on older kernels by setting tx_queue_len=0 (prior
1259 	 * to qdisc init), and then forgot to reinit tx_queue_len
1260 	 * before again attaching a qdisc.
1261 	 */
1262 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1263 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1264 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1265 	}
1266 
1267 	err = qdisc_block_indexes_set(sch, tca, extack);
1268 	if (err)
1269 		goto err_out3;
1270 
1271 	if (ops->init) {
1272 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1273 		if (err != 0)
1274 			goto err_out5;
1275 	}
1276 
1277 	if (tca[TCA_STAB]) {
1278 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1279 		if (IS_ERR(stab)) {
1280 			err = PTR_ERR(stab);
1281 			goto err_out4;
1282 		}
1283 		rcu_assign_pointer(sch->stab, stab);
1284 	}
1285 	if (tca[TCA_RATE]) {
1286 		err = -EOPNOTSUPP;
1287 		if (sch->flags & TCQ_F_MQROOT) {
1288 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1289 			goto err_out4;
1290 		}
1291 
1292 		err = gen_new_estimator(&sch->bstats,
1293 					sch->cpu_bstats,
1294 					&sch->rate_est,
1295 					NULL,
1296 					true,
1297 					tca[TCA_RATE]);
1298 		if (err) {
1299 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1300 			goto err_out4;
1301 		}
1302 	}
1303 
1304 	qdisc_hash_add(sch, false);
1305 	trace_qdisc_create(ops, dev, parent);
1306 
1307 	return sch;
1308 
1309 err_out5:
1310 	/* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1311 	if (ops->destroy)
1312 		ops->destroy(sch);
1313 err_out3:
1314 	netdev_put(dev, &sch->dev_tracker);
1315 	qdisc_free(sch);
1316 err_out2:
1317 	module_put(ops->owner);
1318 err_out:
1319 	*errp = err;
1320 	return NULL;
1321 
1322 err_out4:
1323 	/*
1324 	 * Any broken qdiscs that would require a ops->reset() here?
1325 	 * The qdisc was never in action so it shouldn't be necessary.
1326 	 */
1327 	qdisc_put_stab(rtnl_dereference(sch->stab));
1328 	if (ops->destroy)
1329 		ops->destroy(sch);
1330 	goto err_out3;
1331 }
1332 
1333 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1334 			struct netlink_ext_ack *extack)
1335 {
1336 	struct qdisc_size_table *ostab, *stab = NULL;
1337 	int err = 0;
1338 
1339 	if (tca[TCA_OPTIONS]) {
1340 		if (!sch->ops->change) {
1341 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1342 			return -EINVAL;
1343 		}
1344 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1345 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1346 			return -EOPNOTSUPP;
1347 		}
1348 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1349 		if (err)
1350 			return err;
1351 	}
1352 
1353 	if (tca[TCA_STAB]) {
1354 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1355 		if (IS_ERR(stab))
1356 			return PTR_ERR(stab);
1357 	}
1358 
1359 	ostab = rtnl_dereference(sch->stab);
1360 	rcu_assign_pointer(sch->stab, stab);
1361 	qdisc_put_stab(ostab);
1362 
1363 	if (tca[TCA_RATE]) {
1364 		/* NB: ignores errors from replace_estimator
1365 		   because change can't be undone. */
1366 		if (sch->flags & TCQ_F_MQROOT)
1367 			goto out;
1368 		gen_replace_estimator(&sch->bstats,
1369 				      sch->cpu_bstats,
1370 				      &sch->rate_est,
1371 				      NULL,
1372 				      true,
1373 				      tca[TCA_RATE]);
1374 	}
1375 out:
1376 	return 0;
1377 }
1378 
1379 struct check_loop_arg {
1380 	struct qdisc_walker	w;
1381 	struct Qdisc		*p;
1382 	int			depth;
1383 };
1384 
1385 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1386 			 struct qdisc_walker *w);
1387 
1388 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1389 {
1390 	struct check_loop_arg	arg;
1391 
1392 	if (q->ops->cl_ops == NULL)
1393 		return 0;
1394 
1395 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1396 	arg.w.fn = check_loop_fn;
1397 	arg.depth = depth;
1398 	arg.p = p;
1399 	q->ops->cl_ops->walk(q, &arg.w);
1400 	return arg.w.stop ? -ELOOP : 0;
1401 }
1402 
1403 static int
1404 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1405 {
1406 	struct Qdisc *leaf;
1407 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1408 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1409 
1410 	leaf = cops->leaf(q, cl);
1411 	if (leaf) {
1412 		if (leaf == arg->p || arg->depth > 7)
1413 			return -ELOOP;
1414 		return check_loop(leaf, arg->p, arg->depth + 1);
1415 	}
1416 	return 0;
1417 }
1418 
1419 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1420 	[TCA_KIND]		= { .type = NLA_STRING },
1421 	[TCA_RATE]		= { .type = NLA_BINARY,
1422 				    .len = sizeof(struct tc_estimator) },
1423 	[TCA_STAB]		= { .type = NLA_NESTED },
1424 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1425 	[TCA_CHAIN]		= { .type = NLA_U32 },
1426 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1427 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1428 };
1429 
1430 /*
1431  * Delete/get qdisc.
1432  */
1433 
1434 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1435 			struct netlink_ext_ack *extack)
1436 {
1437 	struct net *net = sock_net(skb->sk);
1438 	struct tcmsg *tcm = nlmsg_data(n);
1439 	struct nlattr *tca[TCA_MAX + 1];
1440 	struct net_device *dev;
1441 	u32 clid;
1442 	struct Qdisc *q = NULL;
1443 	struct Qdisc *p = NULL;
1444 	int err;
1445 
1446 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1447 				     rtm_tca_policy, extack);
1448 	if (err < 0)
1449 		return err;
1450 
1451 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1452 	if (!dev)
1453 		return -ENODEV;
1454 
1455 	clid = tcm->tcm_parent;
1456 	if (clid) {
1457 		if (clid != TC_H_ROOT) {
1458 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1459 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1460 				if (!p) {
1461 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1462 					return -ENOENT;
1463 				}
1464 				q = qdisc_leaf(p, clid);
1465 			} else if (dev_ingress_queue(dev)) {
1466 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1467 			}
1468 		} else {
1469 			q = rtnl_dereference(dev->qdisc);
1470 		}
1471 		if (!q) {
1472 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1473 			return -ENOENT;
1474 		}
1475 
1476 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1477 			NL_SET_ERR_MSG(extack, "Invalid handle");
1478 			return -EINVAL;
1479 		}
1480 	} else {
1481 		q = qdisc_lookup(dev, tcm->tcm_handle);
1482 		if (!q) {
1483 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1484 			return -ENOENT;
1485 		}
1486 	}
1487 
1488 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1489 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1490 		return -EINVAL;
1491 	}
1492 
1493 	if (n->nlmsg_type == RTM_DELQDISC) {
1494 		if (!clid) {
1495 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1496 			return -EINVAL;
1497 		}
1498 		if (q->handle == 0) {
1499 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1500 			return -ENOENT;
1501 		}
1502 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1503 		if (err != 0)
1504 			return err;
1505 	} else {
1506 		qdisc_notify(net, skb, n, clid, NULL, q);
1507 	}
1508 	return 0;
1509 }
1510 
1511 /*
1512  * Create/change qdisc.
1513  */
1514 
1515 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1516 			   struct netlink_ext_ack *extack)
1517 {
1518 	struct net *net = sock_net(skb->sk);
1519 	struct tcmsg *tcm;
1520 	struct nlattr *tca[TCA_MAX + 1];
1521 	struct net_device *dev;
1522 	u32 clid;
1523 	struct Qdisc *q, *p;
1524 	int err;
1525 
1526 replay:
1527 	/* Reinit, just in case something touches this. */
1528 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1529 				     rtm_tca_policy, extack);
1530 	if (err < 0)
1531 		return err;
1532 
1533 	tcm = nlmsg_data(n);
1534 	clid = tcm->tcm_parent;
1535 	q = p = NULL;
1536 
1537 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1538 	if (!dev)
1539 		return -ENODEV;
1540 
1541 
1542 	if (clid) {
1543 		if (clid != TC_H_ROOT) {
1544 			if (clid != TC_H_INGRESS) {
1545 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1546 				if (!p) {
1547 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1548 					return -ENOENT;
1549 				}
1550 				q = qdisc_leaf(p, clid);
1551 			} else if (dev_ingress_queue_create(dev)) {
1552 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1553 			}
1554 		} else {
1555 			q = rtnl_dereference(dev->qdisc);
1556 		}
1557 
1558 		/* It may be default qdisc, ignore it */
1559 		if (q && q->handle == 0)
1560 			q = NULL;
1561 
1562 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1563 			if (tcm->tcm_handle) {
1564 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1565 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1566 					return -EEXIST;
1567 				}
1568 				if (TC_H_MIN(tcm->tcm_handle)) {
1569 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1570 					return -EINVAL;
1571 				}
1572 				q = qdisc_lookup(dev, tcm->tcm_handle);
1573 				if (!q)
1574 					goto create_n_graft;
1575 				if (n->nlmsg_flags & NLM_F_EXCL) {
1576 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1577 					return -EEXIST;
1578 				}
1579 				if (tca[TCA_KIND] &&
1580 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1581 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1582 					return -EINVAL;
1583 				}
1584 				if (q == p ||
1585 				    (p && check_loop(q, p, 0))) {
1586 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1587 					return -ELOOP;
1588 				}
1589 				qdisc_refcount_inc(q);
1590 				goto graft;
1591 			} else {
1592 				if (!q)
1593 					goto create_n_graft;
1594 
1595 				/* This magic test requires explanation.
1596 				 *
1597 				 *   We know, that some child q is already
1598 				 *   attached to this parent and have choice:
1599 				 *   either to change it or to create/graft new one.
1600 				 *
1601 				 *   1. We are allowed to create/graft only
1602 				 *   if CREATE and REPLACE flags are set.
1603 				 *
1604 				 *   2. If EXCL is set, requestor wanted to say,
1605 				 *   that qdisc tcm_handle is not expected
1606 				 *   to exist, so that we choose create/graft too.
1607 				 *
1608 				 *   3. The last case is when no flags are set.
1609 				 *   Alas, it is sort of hole in API, we
1610 				 *   cannot decide what to do unambiguously.
1611 				 *   For now we select create/graft, if
1612 				 *   user gave KIND, which does not match existing.
1613 				 */
1614 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1615 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1616 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1617 				     (tca[TCA_KIND] &&
1618 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1619 					goto create_n_graft;
1620 			}
1621 		}
1622 	} else {
1623 		if (!tcm->tcm_handle) {
1624 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1625 			return -EINVAL;
1626 		}
1627 		q = qdisc_lookup(dev, tcm->tcm_handle);
1628 	}
1629 
1630 	/* Change qdisc parameters */
1631 	if (!q) {
1632 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1633 		return -ENOENT;
1634 	}
1635 	if (n->nlmsg_flags & NLM_F_EXCL) {
1636 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1637 		return -EEXIST;
1638 	}
1639 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1640 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1641 		return -EINVAL;
1642 	}
1643 	err = qdisc_change(q, tca, extack);
1644 	if (err == 0)
1645 		qdisc_notify(net, skb, n, clid, NULL, q);
1646 	return err;
1647 
1648 create_n_graft:
1649 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1650 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1651 		return -ENOENT;
1652 	}
1653 	if (clid == TC_H_INGRESS) {
1654 		if (dev_ingress_queue(dev)) {
1655 			q = qdisc_create(dev, dev_ingress_queue(dev),
1656 					 tcm->tcm_parent, tcm->tcm_parent,
1657 					 tca, &err, extack);
1658 		} else {
1659 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1660 			err = -ENOENT;
1661 		}
1662 	} else {
1663 		struct netdev_queue *dev_queue;
1664 
1665 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1666 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1667 		else if (p)
1668 			dev_queue = p->dev_queue;
1669 		else
1670 			dev_queue = netdev_get_tx_queue(dev, 0);
1671 
1672 		q = qdisc_create(dev, dev_queue,
1673 				 tcm->tcm_parent, tcm->tcm_handle,
1674 				 tca, &err, extack);
1675 	}
1676 	if (q == NULL) {
1677 		if (err == -EAGAIN)
1678 			goto replay;
1679 		return err;
1680 	}
1681 
1682 graft:
1683 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1684 	if (err) {
1685 		if (q)
1686 			qdisc_put(q);
1687 		return err;
1688 	}
1689 
1690 	return 0;
1691 }
1692 
1693 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1694 			      struct netlink_callback *cb,
1695 			      int *q_idx_p, int s_q_idx, bool recur,
1696 			      bool dump_invisible)
1697 {
1698 	int ret = 0, q_idx = *q_idx_p;
1699 	struct Qdisc *q;
1700 	int b;
1701 
1702 	if (!root)
1703 		return 0;
1704 
1705 	q = root;
1706 	if (q_idx < s_q_idx) {
1707 		q_idx++;
1708 	} else {
1709 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1710 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1711 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1712 				  RTM_NEWQDISC) <= 0)
1713 			goto done;
1714 		q_idx++;
1715 	}
1716 
1717 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1718 	 * itself has already been dumped.
1719 	 *
1720 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1721 	 * qdisc hashtable, we don't want to hit it again
1722 	 */
1723 	if (!qdisc_dev(root) || !recur)
1724 		goto out;
1725 
1726 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1727 		if (q_idx < s_q_idx) {
1728 			q_idx++;
1729 			continue;
1730 		}
1731 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1732 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1733 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1734 				  RTM_NEWQDISC) <= 0)
1735 			goto done;
1736 		q_idx++;
1737 	}
1738 
1739 out:
1740 	*q_idx_p = q_idx;
1741 	return ret;
1742 done:
1743 	ret = -1;
1744 	goto out;
1745 }
1746 
1747 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1748 {
1749 	struct net *net = sock_net(skb->sk);
1750 	int idx, q_idx;
1751 	int s_idx, s_q_idx;
1752 	struct net_device *dev;
1753 	const struct nlmsghdr *nlh = cb->nlh;
1754 	struct nlattr *tca[TCA_MAX + 1];
1755 	int err;
1756 
1757 	s_idx = cb->args[0];
1758 	s_q_idx = q_idx = cb->args[1];
1759 
1760 	idx = 0;
1761 	ASSERT_RTNL();
1762 
1763 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1764 				     rtm_tca_policy, cb->extack);
1765 	if (err < 0)
1766 		return err;
1767 
1768 	for_each_netdev(net, dev) {
1769 		struct netdev_queue *dev_queue;
1770 
1771 		if (idx < s_idx)
1772 			goto cont;
1773 		if (idx > s_idx)
1774 			s_q_idx = 0;
1775 		q_idx = 0;
1776 
1777 		if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1778 				       skb, cb, &q_idx, s_q_idx,
1779 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1780 			goto done;
1781 
1782 		dev_queue = dev_ingress_queue(dev);
1783 		if (dev_queue &&
1784 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1785 				       &q_idx, s_q_idx, false,
1786 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1787 			goto done;
1788 
1789 cont:
1790 		idx++;
1791 	}
1792 
1793 done:
1794 	cb->args[0] = idx;
1795 	cb->args[1] = q_idx;
1796 
1797 	return skb->len;
1798 }
1799 
1800 
1801 
1802 /************************************************
1803  *	Traffic classes manipulation.		*
1804  ************************************************/
1805 
1806 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1807 			  unsigned long cl,
1808 			  u32 portid, u32 seq, u16 flags, int event)
1809 {
1810 	struct tcmsg *tcm;
1811 	struct nlmsghdr  *nlh;
1812 	unsigned char *b = skb_tail_pointer(skb);
1813 	struct gnet_dump d;
1814 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1815 
1816 	cond_resched();
1817 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1818 	if (!nlh)
1819 		goto out_nlmsg_trim;
1820 	tcm = nlmsg_data(nlh);
1821 	tcm->tcm_family = AF_UNSPEC;
1822 	tcm->tcm__pad1 = 0;
1823 	tcm->tcm__pad2 = 0;
1824 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1825 	tcm->tcm_parent = q->handle;
1826 	tcm->tcm_handle = q->handle;
1827 	tcm->tcm_info = 0;
1828 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1829 		goto nla_put_failure;
1830 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1831 		goto nla_put_failure;
1832 
1833 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1834 					 NULL, &d, TCA_PAD) < 0)
1835 		goto nla_put_failure;
1836 
1837 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1838 		goto nla_put_failure;
1839 
1840 	if (gnet_stats_finish_copy(&d) < 0)
1841 		goto nla_put_failure;
1842 
1843 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1844 	return skb->len;
1845 
1846 out_nlmsg_trim:
1847 nla_put_failure:
1848 	nlmsg_trim(skb, b);
1849 	return -1;
1850 }
1851 
1852 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1853 			 struct nlmsghdr *n, struct Qdisc *q,
1854 			 unsigned long cl, int event)
1855 {
1856 	struct sk_buff *skb;
1857 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1858 
1859 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1860 	if (!skb)
1861 		return -ENOBUFS;
1862 
1863 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1864 		kfree_skb(skb);
1865 		return -EINVAL;
1866 	}
1867 
1868 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1869 			      n->nlmsg_flags & NLM_F_ECHO);
1870 }
1871 
1872 static int tclass_del_notify(struct net *net,
1873 			     const struct Qdisc_class_ops *cops,
1874 			     struct sk_buff *oskb, struct nlmsghdr *n,
1875 			     struct Qdisc *q, unsigned long cl,
1876 			     struct netlink_ext_ack *extack)
1877 {
1878 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1879 	struct sk_buff *skb;
1880 	int err = 0;
1881 
1882 	if (!cops->delete)
1883 		return -EOPNOTSUPP;
1884 
1885 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1886 	if (!skb)
1887 		return -ENOBUFS;
1888 
1889 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1890 			   RTM_DELTCLASS) < 0) {
1891 		kfree_skb(skb);
1892 		return -EINVAL;
1893 	}
1894 
1895 	err = cops->delete(q, cl, extack);
1896 	if (err) {
1897 		kfree_skb(skb);
1898 		return err;
1899 	}
1900 
1901 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1902 			     n->nlmsg_flags & NLM_F_ECHO);
1903 	return err;
1904 }
1905 
1906 #ifdef CONFIG_NET_CLS
1907 
1908 struct tcf_bind_args {
1909 	struct tcf_walker w;
1910 	unsigned long base;
1911 	unsigned long cl;
1912 	u32 classid;
1913 };
1914 
1915 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1916 {
1917 	struct tcf_bind_args *a = (void *)arg;
1918 
1919 	if (n && tp->ops->bind_class) {
1920 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1921 
1922 		sch_tree_lock(q);
1923 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1924 		sch_tree_unlock(q);
1925 	}
1926 	return 0;
1927 }
1928 
1929 struct tc_bind_class_args {
1930 	struct qdisc_walker w;
1931 	unsigned long new_cl;
1932 	u32 portid;
1933 	u32 clid;
1934 };
1935 
1936 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1937 				struct qdisc_walker *w)
1938 {
1939 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1940 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1941 	struct tcf_block *block;
1942 	struct tcf_chain *chain;
1943 
1944 	block = cops->tcf_block(q, cl, NULL);
1945 	if (!block)
1946 		return 0;
1947 	for (chain = tcf_get_next_chain(block, NULL);
1948 	     chain;
1949 	     chain = tcf_get_next_chain(block, chain)) {
1950 		struct tcf_proto *tp;
1951 
1952 		for (tp = tcf_get_next_proto(chain, NULL);
1953 		     tp; tp = tcf_get_next_proto(chain, tp)) {
1954 			struct tcf_bind_args arg = {};
1955 
1956 			arg.w.fn = tcf_node_bind;
1957 			arg.classid = a->clid;
1958 			arg.base = cl;
1959 			arg.cl = a->new_cl;
1960 			tp->ops->walk(tp, &arg.w, true);
1961 		}
1962 	}
1963 
1964 	return 0;
1965 }
1966 
1967 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1968 			   unsigned long new_cl)
1969 {
1970 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1971 	struct tc_bind_class_args args = {};
1972 
1973 	if (!cops->tcf_block)
1974 		return;
1975 	args.portid = portid;
1976 	args.clid = clid;
1977 	args.new_cl = new_cl;
1978 	args.w.fn = tc_bind_class_walker;
1979 	q->ops->cl_ops->walk(q, &args.w);
1980 }
1981 
1982 #else
1983 
1984 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1985 			   unsigned long new_cl)
1986 {
1987 }
1988 
1989 #endif
1990 
1991 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1992 			 struct netlink_ext_ack *extack)
1993 {
1994 	struct net *net = sock_net(skb->sk);
1995 	struct tcmsg *tcm = nlmsg_data(n);
1996 	struct nlattr *tca[TCA_MAX + 1];
1997 	struct net_device *dev;
1998 	struct Qdisc *q = NULL;
1999 	const struct Qdisc_class_ops *cops;
2000 	unsigned long cl = 0;
2001 	unsigned long new_cl;
2002 	u32 portid;
2003 	u32 clid;
2004 	u32 qid;
2005 	int err;
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 = rtnl_dereference(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 = rtnl_dereference(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(rtnl_dereference(dev->qdisc),
2218 				skb, tcm, cb, &t, s_t, true) < 0)
2219 		goto done;
2220 
2221 	dev_queue = dev_ingress_queue(dev);
2222 	if (dev_queue &&
2223 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2224 				&t, s_t, false) < 0)
2225 		goto done;
2226 
2227 done:
2228 	cb->args[0] = t;
2229 
2230 	dev_put(dev);
2231 	return skb->len;
2232 }
2233 
2234 #ifdef CONFIG_PROC_FS
2235 static int psched_show(struct seq_file *seq, void *v)
2236 {
2237 	seq_printf(seq, "%08x %08x %08x %08x\n",
2238 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2239 		   1000000,
2240 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2241 
2242 	return 0;
2243 }
2244 
2245 static int __net_init psched_net_init(struct net *net)
2246 {
2247 	struct proc_dir_entry *e;
2248 
2249 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2250 	if (e == NULL)
2251 		return -ENOMEM;
2252 
2253 	return 0;
2254 }
2255 
2256 static void __net_exit psched_net_exit(struct net *net)
2257 {
2258 	remove_proc_entry("psched", net->proc_net);
2259 }
2260 #else
2261 static int __net_init psched_net_init(struct net *net)
2262 {
2263 	return 0;
2264 }
2265 
2266 static void __net_exit psched_net_exit(struct net *net)
2267 {
2268 }
2269 #endif
2270 
2271 static struct pernet_operations psched_net_ops = {
2272 	.init = psched_net_init,
2273 	.exit = psched_net_exit,
2274 };
2275 
2276 static int __init pktsched_init(void)
2277 {
2278 	int err;
2279 
2280 	err = register_pernet_subsys(&psched_net_ops);
2281 	if (err) {
2282 		pr_err("pktsched_init: "
2283 		       "cannot initialize per netns operations\n");
2284 		return err;
2285 	}
2286 
2287 	register_qdisc(&pfifo_fast_ops);
2288 	register_qdisc(&pfifo_qdisc_ops);
2289 	register_qdisc(&bfifo_qdisc_ops);
2290 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2291 	register_qdisc(&mq_qdisc_ops);
2292 	register_qdisc(&noqueue_qdisc_ops);
2293 
2294 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2295 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2296 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2297 		      0);
2298 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2299 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2300 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2301 		      0);
2302 
2303 	return 0;
2304 }
2305 
2306 subsys_initcall(pktsched_init);
2307