xref: /openbmc/linux/net/sched/sch_api.c (revision 901181b7)
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 &&
511 		    memcmp(stab->data, tab, flex_array_size(stab, data, tsize)))
512 			continue;
513 		stab->refcnt++;
514 		return stab;
515 	}
516 
517 	if (s->size_log > STAB_SIZE_LOG_MAX ||
518 	    s->cell_log > STAB_SIZE_LOG_MAX) {
519 		NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
520 		return ERR_PTR(-EINVAL);
521 	}
522 
523 	stab = kmalloc(struct_size(stab, data, tsize), GFP_KERNEL);
524 	if (!stab)
525 		return ERR_PTR(-ENOMEM);
526 
527 	stab->refcnt = 1;
528 	stab->szopts = *s;
529 	if (tsize > 0)
530 		memcpy(stab->data, tab, flex_array_size(stab, data, tsize));
531 
532 	list_add_tail(&stab->list, &qdisc_stab_list);
533 
534 	return stab;
535 }
536 
537 void qdisc_put_stab(struct qdisc_size_table *tab)
538 {
539 	if (!tab)
540 		return;
541 
542 	if (--tab->refcnt == 0) {
543 		list_del(&tab->list);
544 		kfree_rcu(tab, rcu);
545 	}
546 }
547 EXPORT_SYMBOL(qdisc_put_stab);
548 
549 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
550 {
551 	struct nlattr *nest;
552 
553 	nest = nla_nest_start_noflag(skb, TCA_STAB);
554 	if (nest == NULL)
555 		goto nla_put_failure;
556 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
557 		goto nla_put_failure;
558 	nla_nest_end(skb, nest);
559 
560 	return skb->len;
561 
562 nla_put_failure:
563 	return -1;
564 }
565 
566 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
567 			       const struct qdisc_size_table *stab)
568 {
569 	int pkt_len, slot;
570 
571 	pkt_len = skb->len + stab->szopts.overhead;
572 	if (unlikely(!stab->szopts.tsize))
573 		goto out;
574 
575 	slot = pkt_len + stab->szopts.cell_align;
576 	if (unlikely(slot < 0))
577 		slot = 0;
578 
579 	slot >>= stab->szopts.cell_log;
580 	if (likely(slot < stab->szopts.tsize))
581 		pkt_len = stab->data[slot];
582 	else
583 		pkt_len = stab->data[stab->szopts.tsize - 1] *
584 				(slot / stab->szopts.tsize) +
585 				stab->data[slot % stab->szopts.tsize];
586 
587 	pkt_len <<= stab->szopts.size_log;
588 out:
589 	if (unlikely(pkt_len < 1))
590 		pkt_len = 1;
591 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
592 }
593 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
594 
595 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
596 {
597 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
598 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
599 			txt, qdisc->ops->id, qdisc->handle >> 16);
600 		qdisc->flags |= TCQ_F_WARN_NONWC;
601 	}
602 }
603 EXPORT_SYMBOL(qdisc_warn_nonwc);
604 
605 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
606 {
607 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
608 						 timer);
609 
610 	rcu_read_lock();
611 	__netif_schedule(qdisc_root(wd->qdisc));
612 	rcu_read_unlock();
613 
614 	return HRTIMER_NORESTART;
615 }
616 
617 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
618 				 clockid_t clockid)
619 {
620 	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
621 	wd->timer.function = qdisc_watchdog;
622 	wd->qdisc = qdisc;
623 }
624 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
625 
626 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
627 {
628 	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
629 }
630 EXPORT_SYMBOL(qdisc_watchdog_init);
631 
632 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
633 				      u64 delta_ns)
634 {
635 	if (test_bit(__QDISC_STATE_DEACTIVATED,
636 		     &qdisc_root_sleeping(wd->qdisc)->state))
637 		return;
638 
639 	if (hrtimer_is_queued(&wd->timer)) {
640 		/* If timer is already set in [expires, expires + delta_ns],
641 		 * do not reprogram it.
642 		 */
643 		if (wd->last_expires - expires <= delta_ns)
644 			return;
645 	}
646 
647 	wd->last_expires = expires;
648 	hrtimer_start_range_ns(&wd->timer,
649 			       ns_to_ktime(expires),
650 			       delta_ns,
651 			       HRTIMER_MODE_ABS_PINNED);
652 }
653 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
654 
655 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
656 {
657 	hrtimer_cancel(&wd->timer);
658 }
659 EXPORT_SYMBOL(qdisc_watchdog_cancel);
660 
661 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
662 {
663 	struct hlist_head *h;
664 	unsigned int i;
665 
666 	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
667 
668 	if (h != NULL) {
669 		for (i = 0; i < n; i++)
670 			INIT_HLIST_HEAD(&h[i]);
671 	}
672 	return h;
673 }
674 
675 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
676 {
677 	struct Qdisc_class_common *cl;
678 	struct hlist_node *next;
679 	struct hlist_head *nhash, *ohash;
680 	unsigned int nsize, nmask, osize;
681 	unsigned int i, h;
682 
683 	/* Rehash when load factor exceeds 0.75 */
684 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
685 		return;
686 	nsize = clhash->hashsize * 2;
687 	nmask = nsize - 1;
688 	nhash = qdisc_class_hash_alloc(nsize);
689 	if (nhash == NULL)
690 		return;
691 
692 	ohash = clhash->hash;
693 	osize = clhash->hashsize;
694 
695 	sch_tree_lock(sch);
696 	for (i = 0; i < osize; i++) {
697 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
698 			h = qdisc_class_hash(cl->classid, nmask);
699 			hlist_add_head(&cl->hnode, &nhash[h]);
700 		}
701 	}
702 	clhash->hash     = nhash;
703 	clhash->hashsize = nsize;
704 	clhash->hashmask = nmask;
705 	sch_tree_unlock(sch);
706 
707 	kvfree(ohash);
708 }
709 EXPORT_SYMBOL(qdisc_class_hash_grow);
710 
711 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
712 {
713 	unsigned int size = 4;
714 
715 	clhash->hash = qdisc_class_hash_alloc(size);
716 	if (!clhash->hash)
717 		return -ENOMEM;
718 	clhash->hashsize  = size;
719 	clhash->hashmask  = size - 1;
720 	clhash->hashelems = 0;
721 	return 0;
722 }
723 EXPORT_SYMBOL(qdisc_class_hash_init);
724 
725 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
726 {
727 	kvfree(clhash->hash);
728 }
729 EXPORT_SYMBOL(qdisc_class_hash_destroy);
730 
731 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
732 			     struct Qdisc_class_common *cl)
733 {
734 	unsigned int h;
735 
736 	INIT_HLIST_NODE(&cl->hnode);
737 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
738 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
739 	clhash->hashelems++;
740 }
741 EXPORT_SYMBOL(qdisc_class_hash_insert);
742 
743 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
744 			     struct Qdisc_class_common *cl)
745 {
746 	hlist_del(&cl->hnode);
747 	clhash->hashelems--;
748 }
749 EXPORT_SYMBOL(qdisc_class_hash_remove);
750 
751 /* Allocate an unique handle from space managed by kernel
752  * Possible range is [8000-FFFF]:0000 (0x8000 values)
753  */
754 static u32 qdisc_alloc_handle(struct net_device *dev)
755 {
756 	int i = 0x8000;
757 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
758 
759 	do {
760 		autohandle += TC_H_MAKE(0x10000U, 0);
761 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
762 			autohandle = TC_H_MAKE(0x80000000U, 0);
763 		if (!qdisc_lookup(dev, autohandle))
764 			return autohandle;
765 		cond_resched();
766 	} while	(--i > 0);
767 
768 	return 0;
769 }
770 
771 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
772 {
773 	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
774 	const struct Qdisc_class_ops *cops;
775 	unsigned long cl;
776 	u32 parentid;
777 	bool notify;
778 	int drops;
779 
780 	if (n == 0 && len == 0)
781 		return;
782 	drops = max_t(int, n, 0);
783 	rcu_read_lock();
784 	while ((parentid = sch->parent)) {
785 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
786 			break;
787 
788 		if (sch->flags & TCQ_F_NOPARENT)
789 			break;
790 		/* Notify parent qdisc only if child qdisc becomes empty.
791 		 *
792 		 * If child was empty even before update then backlog
793 		 * counter is screwed and we skip notification because
794 		 * parent class is already passive.
795 		 *
796 		 * If the original child was offloaded then it is allowed
797 		 * to be seem as empty, so the parent is notified anyway.
798 		 */
799 		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
800 						       !qdisc_is_offloaded);
801 		/* TODO: perform the search on a per txq basis */
802 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
803 		if (sch == NULL) {
804 			WARN_ON_ONCE(parentid != TC_H_ROOT);
805 			break;
806 		}
807 		cops = sch->ops->cl_ops;
808 		if (notify && cops->qlen_notify) {
809 			cl = cops->find(sch, parentid);
810 			cops->qlen_notify(sch, cl);
811 		}
812 		sch->q.qlen -= n;
813 		sch->qstats.backlog -= len;
814 		__qdisc_qstats_drop(sch, drops);
815 	}
816 	rcu_read_unlock();
817 }
818 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
819 
820 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
821 			      void *type_data)
822 {
823 	struct net_device *dev = qdisc_dev(sch);
824 	int err;
825 
826 	sch->flags &= ~TCQ_F_OFFLOADED;
827 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
828 		return 0;
829 
830 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
831 	if (err == -EOPNOTSUPP)
832 		return 0;
833 
834 	if (!err)
835 		sch->flags |= TCQ_F_OFFLOADED;
836 
837 	return err;
838 }
839 EXPORT_SYMBOL(qdisc_offload_dump_helper);
840 
841 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
842 				struct Qdisc *new, struct Qdisc *old,
843 				enum tc_setup_type type, void *type_data,
844 				struct netlink_ext_ack *extack)
845 {
846 	bool any_qdisc_is_offloaded;
847 	int err;
848 
849 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
850 		return;
851 
852 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
853 
854 	/* Don't report error if the graft is part of destroy operation. */
855 	if (!err || !new || new == &noop_qdisc)
856 		return;
857 
858 	/* Don't report error if the parent, the old child and the new
859 	 * one are not offloaded.
860 	 */
861 	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
862 	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
863 	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
864 
865 	if (any_qdisc_is_offloaded)
866 		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
867 }
868 EXPORT_SYMBOL(qdisc_offload_graft_helper);
869 
870 static void qdisc_offload_graft_root(struct net_device *dev,
871 				     struct Qdisc *new, struct Qdisc *old,
872 				     struct netlink_ext_ack *extack)
873 {
874 	struct tc_root_qopt_offload graft_offload = {
875 		.command	= TC_ROOT_GRAFT,
876 		.handle		= new ? new->handle : 0,
877 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
878 				  (old && old->flags & TCQ_F_INGRESS),
879 	};
880 
881 	qdisc_offload_graft_helper(dev, NULL, new, old,
882 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
883 }
884 
885 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
886 			 u32 portid, u32 seq, u16 flags, int event)
887 {
888 	struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
889 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
890 	struct tcmsg *tcm;
891 	struct nlmsghdr  *nlh;
892 	unsigned char *b = skb_tail_pointer(skb);
893 	struct gnet_dump d;
894 	struct qdisc_size_table *stab;
895 	u32 block_index;
896 	__u32 qlen;
897 
898 	cond_resched();
899 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
900 	if (!nlh)
901 		goto out_nlmsg_trim;
902 	tcm = nlmsg_data(nlh);
903 	tcm->tcm_family = AF_UNSPEC;
904 	tcm->tcm__pad1 = 0;
905 	tcm->tcm__pad2 = 0;
906 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
907 	tcm->tcm_parent = clid;
908 	tcm->tcm_handle = q->handle;
909 	tcm->tcm_info = refcount_read(&q->refcnt);
910 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
911 		goto nla_put_failure;
912 	if (q->ops->ingress_block_get) {
913 		block_index = q->ops->ingress_block_get(q);
914 		if (block_index &&
915 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
916 			goto nla_put_failure;
917 	}
918 	if (q->ops->egress_block_get) {
919 		block_index = q->ops->egress_block_get(q);
920 		if (block_index &&
921 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
922 			goto nla_put_failure;
923 	}
924 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
925 		goto nla_put_failure;
926 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
927 		goto nla_put_failure;
928 	qlen = qdisc_qlen_sum(q);
929 
930 	stab = rtnl_dereference(q->stab);
931 	if (stab && qdisc_dump_stab(skb, stab) < 0)
932 		goto nla_put_failure;
933 
934 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
935 					 NULL, &d, TCA_PAD) < 0)
936 		goto nla_put_failure;
937 
938 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
939 		goto nla_put_failure;
940 
941 	if (qdisc_is_percpu_stats(q)) {
942 		cpu_bstats = q->cpu_bstats;
943 		cpu_qstats = q->cpu_qstats;
944 	}
945 
946 	if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 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 		err = -EOPNOTSUPP;
1268 		if (sch->flags & TCQ_F_MQROOT) {
1269 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1270 			goto err_out4;
1271 		}
1272 
1273 		err = gen_new_estimator(&sch->bstats,
1274 					sch->cpu_bstats,
1275 					&sch->rate_est,
1276 					NULL,
1277 					true,
1278 					tca[TCA_RATE]);
1279 		if (err) {
1280 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1281 			goto err_out4;
1282 		}
1283 	}
1284 
1285 	qdisc_hash_add(sch, false);
1286 	trace_qdisc_create(ops, dev, parent);
1287 
1288 	return sch;
1289 
1290 err_out5:
1291 	/* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1292 	if (ops->destroy)
1293 		ops->destroy(sch);
1294 err_out3:
1295 	dev_put(dev);
1296 	qdisc_free(sch);
1297 err_out2:
1298 	module_put(ops->owner);
1299 err_out:
1300 	*errp = err;
1301 	return NULL;
1302 
1303 err_out4:
1304 	/*
1305 	 * Any broken qdiscs that would require a ops->reset() here?
1306 	 * The qdisc was never in action so it shouldn't be necessary.
1307 	 */
1308 	qdisc_put_stab(rtnl_dereference(sch->stab));
1309 	if (ops->destroy)
1310 		ops->destroy(sch);
1311 	goto err_out3;
1312 }
1313 
1314 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1315 			struct netlink_ext_ack *extack)
1316 {
1317 	struct qdisc_size_table *ostab, *stab = NULL;
1318 	int err = 0;
1319 
1320 	if (tca[TCA_OPTIONS]) {
1321 		if (!sch->ops->change) {
1322 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1323 			return -EINVAL;
1324 		}
1325 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1326 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1327 			return -EOPNOTSUPP;
1328 		}
1329 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1330 		if (err)
1331 			return err;
1332 	}
1333 
1334 	if (tca[TCA_STAB]) {
1335 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1336 		if (IS_ERR(stab))
1337 			return PTR_ERR(stab);
1338 	}
1339 
1340 	ostab = rtnl_dereference(sch->stab);
1341 	rcu_assign_pointer(sch->stab, stab);
1342 	qdisc_put_stab(ostab);
1343 
1344 	if (tca[TCA_RATE]) {
1345 		/* NB: ignores errors from replace_estimator
1346 		   because change can't be undone. */
1347 		if (sch->flags & TCQ_F_MQROOT)
1348 			goto out;
1349 		gen_replace_estimator(&sch->bstats,
1350 				      sch->cpu_bstats,
1351 				      &sch->rate_est,
1352 				      NULL,
1353 				      true,
1354 				      tca[TCA_RATE]);
1355 	}
1356 out:
1357 	return 0;
1358 }
1359 
1360 struct check_loop_arg {
1361 	struct qdisc_walker	w;
1362 	struct Qdisc		*p;
1363 	int			depth;
1364 };
1365 
1366 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1367 			 struct qdisc_walker *w);
1368 
1369 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1370 {
1371 	struct check_loop_arg	arg;
1372 
1373 	if (q->ops->cl_ops == NULL)
1374 		return 0;
1375 
1376 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1377 	arg.w.fn = check_loop_fn;
1378 	arg.depth = depth;
1379 	arg.p = p;
1380 	q->ops->cl_ops->walk(q, &arg.w);
1381 	return arg.w.stop ? -ELOOP : 0;
1382 }
1383 
1384 static int
1385 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1386 {
1387 	struct Qdisc *leaf;
1388 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1389 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1390 
1391 	leaf = cops->leaf(q, cl);
1392 	if (leaf) {
1393 		if (leaf == arg->p || arg->depth > 7)
1394 			return -ELOOP;
1395 		return check_loop(leaf, arg->p, arg->depth + 1);
1396 	}
1397 	return 0;
1398 }
1399 
1400 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1401 	[TCA_KIND]		= { .type = NLA_STRING },
1402 	[TCA_RATE]		= { .type = NLA_BINARY,
1403 				    .len = sizeof(struct tc_estimator) },
1404 	[TCA_STAB]		= { .type = NLA_NESTED },
1405 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1406 	[TCA_CHAIN]		= { .type = NLA_U32 },
1407 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1408 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1409 };
1410 
1411 /*
1412  * Delete/get qdisc.
1413  */
1414 
1415 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1416 			struct netlink_ext_ack *extack)
1417 {
1418 	struct net *net = sock_net(skb->sk);
1419 	struct tcmsg *tcm = nlmsg_data(n);
1420 	struct nlattr *tca[TCA_MAX + 1];
1421 	struct net_device *dev;
1422 	u32 clid;
1423 	struct Qdisc *q = NULL;
1424 	struct Qdisc *p = NULL;
1425 	int err;
1426 
1427 	if ((n->nlmsg_type != RTM_GETQDISC) &&
1428 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1429 		return -EPERM;
1430 
1431 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1432 				     rtm_tca_policy, extack);
1433 	if (err < 0)
1434 		return err;
1435 
1436 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1437 	if (!dev)
1438 		return -ENODEV;
1439 
1440 	clid = tcm->tcm_parent;
1441 	if (clid) {
1442 		if (clid != TC_H_ROOT) {
1443 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1444 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1445 				if (!p) {
1446 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1447 					return -ENOENT;
1448 				}
1449 				q = qdisc_leaf(p, clid);
1450 			} else if (dev_ingress_queue(dev)) {
1451 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1452 			}
1453 		} else {
1454 			q = dev->qdisc;
1455 		}
1456 		if (!q) {
1457 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1458 			return -ENOENT;
1459 		}
1460 
1461 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1462 			NL_SET_ERR_MSG(extack, "Invalid handle");
1463 			return -EINVAL;
1464 		}
1465 	} else {
1466 		q = qdisc_lookup(dev, tcm->tcm_handle);
1467 		if (!q) {
1468 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1469 			return -ENOENT;
1470 		}
1471 	}
1472 
1473 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1474 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1475 		return -EINVAL;
1476 	}
1477 
1478 	if (n->nlmsg_type == RTM_DELQDISC) {
1479 		if (!clid) {
1480 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1481 			return -EINVAL;
1482 		}
1483 		if (q->handle == 0) {
1484 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1485 			return -ENOENT;
1486 		}
1487 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1488 		if (err != 0)
1489 			return err;
1490 	} else {
1491 		qdisc_notify(net, skb, n, clid, NULL, q);
1492 	}
1493 	return 0;
1494 }
1495 
1496 /*
1497  * Create/change qdisc.
1498  */
1499 
1500 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1501 			   struct netlink_ext_ack *extack)
1502 {
1503 	struct net *net = sock_net(skb->sk);
1504 	struct tcmsg *tcm;
1505 	struct nlattr *tca[TCA_MAX + 1];
1506 	struct net_device *dev;
1507 	u32 clid;
1508 	struct Qdisc *q, *p;
1509 	int err;
1510 
1511 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1512 		return -EPERM;
1513 
1514 replay:
1515 	/* Reinit, just in case something touches this. */
1516 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1517 				     rtm_tca_policy, extack);
1518 	if (err < 0)
1519 		return err;
1520 
1521 	tcm = nlmsg_data(n);
1522 	clid = tcm->tcm_parent;
1523 	q = p = NULL;
1524 
1525 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1526 	if (!dev)
1527 		return -ENODEV;
1528 
1529 
1530 	if (clid) {
1531 		if (clid != TC_H_ROOT) {
1532 			if (clid != TC_H_INGRESS) {
1533 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1534 				if (!p) {
1535 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1536 					return -ENOENT;
1537 				}
1538 				q = qdisc_leaf(p, clid);
1539 			} else if (dev_ingress_queue_create(dev)) {
1540 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1541 			}
1542 		} else {
1543 			q = dev->qdisc;
1544 		}
1545 
1546 		/* It may be default qdisc, ignore it */
1547 		if (q && q->handle == 0)
1548 			q = NULL;
1549 
1550 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1551 			if (tcm->tcm_handle) {
1552 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1553 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1554 					return -EEXIST;
1555 				}
1556 				if (TC_H_MIN(tcm->tcm_handle)) {
1557 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1558 					return -EINVAL;
1559 				}
1560 				q = qdisc_lookup(dev, tcm->tcm_handle);
1561 				if (!q)
1562 					goto create_n_graft;
1563 				if (n->nlmsg_flags & NLM_F_EXCL) {
1564 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1565 					return -EEXIST;
1566 				}
1567 				if (tca[TCA_KIND] &&
1568 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1569 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1570 					return -EINVAL;
1571 				}
1572 				if (q == p ||
1573 				    (p && check_loop(q, p, 0))) {
1574 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1575 					return -ELOOP;
1576 				}
1577 				qdisc_refcount_inc(q);
1578 				goto graft;
1579 			} else {
1580 				if (!q)
1581 					goto create_n_graft;
1582 
1583 				/* This magic test requires explanation.
1584 				 *
1585 				 *   We know, that some child q is already
1586 				 *   attached to this parent and have choice:
1587 				 *   either to change it or to create/graft new one.
1588 				 *
1589 				 *   1. We are allowed to create/graft only
1590 				 *   if CREATE and REPLACE flags are set.
1591 				 *
1592 				 *   2. If EXCL is set, requestor wanted to say,
1593 				 *   that qdisc tcm_handle is not expected
1594 				 *   to exist, so that we choose create/graft too.
1595 				 *
1596 				 *   3. The last case is when no flags are set.
1597 				 *   Alas, it is sort of hole in API, we
1598 				 *   cannot decide what to do unambiguously.
1599 				 *   For now we select create/graft, if
1600 				 *   user gave KIND, which does not match existing.
1601 				 */
1602 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1603 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1604 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1605 				     (tca[TCA_KIND] &&
1606 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1607 					goto create_n_graft;
1608 			}
1609 		}
1610 	} else {
1611 		if (!tcm->tcm_handle) {
1612 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1613 			return -EINVAL;
1614 		}
1615 		q = qdisc_lookup(dev, tcm->tcm_handle);
1616 	}
1617 
1618 	/* Change qdisc parameters */
1619 	if (!q) {
1620 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1621 		return -ENOENT;
1622 	}
1623 	if (n->nlmsg_flags & NLM_F_EXCL) {
1624 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1625 		return -EEXIST;
1626 	}
1627 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1628 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1629 		return -EINVAL;
1630 	}
1631 	err = qdisc_change(q, tca, extack);
1632 	if (err == 0)
1633 		qdisc_notify(net, skb, n, clid, NULL, q);
1634 	return err;
1635 
1636 create_n_graft:
1637 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1638 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1639 		return -ENOENT;
1640 	}
1641 	if (clid == TC_H_INGRESS) {
1642 		if (dev_ingress_queue(dev)) {
1643 			q = qdisc_create(dev, dev_ingress_queue(dev), p,
1644 					 tcm->tcm_parent, tcm->tcm_parent,
1645 					 tca, &err, extack);
1646 		} else {
1647 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1648 			err = -ENOENT;
1649 		}
1650 	} else {
1651 		struct netdev_queue *dev_queue;
1652 
1653 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1654 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1655 		else if (p)
1656 			dev_queue = p->dev_queue;
1657 		else
1658 			dev_queue = netdev_get_tx_queue(dev, 0);
1659 
1660 		q = qdisc_create(dev, dev_queue, p,
1661 				 tcm->tcm_parent, tcm->tcm_handle,
1662 				 tca, &err, extack);
1663 	}
1664 	if (q == NULL) {
1665 		if (err == -EAGAIN)
1666 			goto replay;
1667 		return err;
1668 	}
1669 
1670 graft:
1671 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1672 	if (err) {
1673 		if (q)
1674 			qdisc_put(q);
1675 		return err;
1676 	}
1677 
1678 	return 0;
1679 }
1680 
1681 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1682 			      struct netlink_callback *cb,
1683 			      int *q_idx_p, int s_q_idx, bool recur,
1684 			      bool dump_invisible)
1685 {
1686 	int ret = 0, q_idx = *q_idx_p;
1687 	struct Qdisc *q;
1688 	int b;
1689 
1690 	if (!root)
1691 		return 0;
1692 
1693 	q = root;
1694 	if (q_idx < s_q_idx) {
1695 		q_idx++;
1696 	} else {
1697 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1698 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1699 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1700 				  RTM_NEWQDISC) <= 0)
1701 			goto done;
1702 		q_idx++;
1703 	}
1704 
1705 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1706 	 * itself has already been dumped.
1707 	 *
1708 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1709 	 * qdisc hashtable, we don't want to hit it again
1710 	 */
1711 	if (!qdisc_dev(root) || !recur)
1712 		goto out;
1713 
1714 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1715 		if (q_idx < s_q_idx) {
1716 			q_idx++;
1717 			continue;
1718 		}
1719 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1720 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1721 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1722 				  RTM_NEWQDISC) <= 0)
1723 			goto done;
1724 		q_idx++;
1725 	}
1726 
1727 out:
1728 	*q_idx_p = q_idx;
1729 	return ret;
1730 done:
1731 	ret = -1;
1732 	goto out;
1733 }
1734 
1735 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1736 {
1737 	struct net *net = sock_net(skb->sk);
1738 	int idx, q_idx;
1739 	int s_idx, s_q_idx;
1740 	struct net_device *dev;
1741 	const struct nlmsghdr *nlh = cb->nlh;
1742 	struct nlattr *tca[TCA_MAX + 1];
1743 	int err;
1744 
1745 	s_idx = cb->args[0];
1746 	s_q_idx = q_idx = cb->args[1];
1747 
1748 	idx = 0;
1749 	ASSERT_RTNL();
1750 
1751 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1752 				     rtm_tca_policy, cb->extack);
1753 	if (err < 0)
1754 		return err;
1755 
1756 	for_each_netdev(net, dev) {
1757 		struct netdev_queue *dev_queue;
1758 
1759 		if (idx < s_idx)
1760 			goto cont;
1761 		if (idx > s_idx)
1762 			s_q_idx = 0;
1763 		q_idx = 0;
1764 
1765 		if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1766 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1767 			goto done;
1768 
1769 		dev_queue = dev_ingress_queue(dev);
1770 		if (dev_queue &&
1771 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1772 				       &q_idx, s_q_idx, false,
1773 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1774 			goto done;
1775 
1776 cont:
1777 		idx++;
1778 	}
1779 
1780 done:
1781 	cb->args[0] = idx;
1782 	cb->args[1] = q_idx;
1783 
1784 	return skb->len;
1785 }
1786 
1787 
1788 
1789 /************************************************
1790  *	Traffic classes manipulation.		*
1791  ************************************************/
1792 
1793 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1794 			  unsigned long cl,
1795 			  u32 portid, u32 seq, u16 flags, int event)
1796 {
1797 	struct tcmsg *tcm;
1798 	struct nlmsghdr  *nlh;
1799 	unsigned char *b = skb_tail_pointer(skb);
1800 	struct gnet_dump d;
1801 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1802 
1803 	cond_resched();
1804 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1805 	if (!nlh)
1806 		goto out_nlmsg_trim;
1807 	tcm = nlmsg_data(nlh);
1808 	tcm->tcm_family = AF_UNSPEC;
1809 	tcm->tcm__pad1 = 0;
1810 	tcm->tcm__pad2 = 0;
1811 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1812 	tcm->tcm_parent = q->handle;
1813 	tcm->tcm_handle = q->handle;
1814 	tcm->tcm_info = 0;
1815 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1816 		goto nla_put_failure;
1817 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1818 		goto nla_put_failure;
1819 
1820 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1821 					 NULL, &d, TCA_PAD) < 0)
1822 		goto nla_put_failure;
1823 
1824 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1825 		goto nla_put_failure;
1826 
1827 	if (gnet_stats_finish_copy(&d) < 0)
1828 		goto nla_put_failure;
1829 
1830 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1831 	return skb->len;
1832 
1833 out_nlmsg_trim:
1834 nla_put_failure:
1835 	nlmsg_trim(skb, b);
1836 	return -1;
1837 }
1838 
1839 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1840 			 struct nlmsghdr *n, struct Qdisc *q,
1841 			 unsigned long cl, int event)
1842 {
1843 	struct sk_buff *skb;
1844 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1845 
1846 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1847 	if (!skb)
1848 		return -ENOBUFS;
1849 
1850 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1851 		kfree_skb(skb);
1852 		return -EINVAL;
1853 	}
1854 
1855 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1856 			      n->nlmsg_flags & NLM_F_ECHO);
1857 }
1858 
1859 static int tclass_del_notify(struct net *net,
1860 			     const struct Qdisc_class_ops *cops,
1861 			     struct sk_buff *oskb, struct nlmsghdr *n,
1862 			     struct Qdisc *q, unsigned long cl,
1863 			     struct netlink_ext_ack *extack)
1864 {
1865 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1866 	struct sk_buff *skb;
1867 	int err = 0;
1868 
1869 	if (!cops->delete)
1870 		return -EOPNOTSUPP;
1871 
1872 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1873 	if (!skb)
1874 		return -ENOBUFS;
1875 
1876 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1877 			   RTM_DELTCLASS) < 0) {
1878 		kfree_skb(skb);
1879 		return -EINVAL;
1880 	}
1881 
1882 	err = cops->delete(q, cl, extack);
1883 	if (err) {
1884 		kfree_skb(skb);
1885 		return err;
1886 	}
1887 
1888 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1889 			     n->nlmsg_flags & NLM_F_ECHO);
1890 	return err;
1891 }
1892 
1893 #ifdef CONFIG_NET_CLS
1894 
1895 struct tcf_bind_args {
1896 	struct tcf_walker w;
1897 	unsigned long base;
1898 	unsigned long cl;
1899 	u32 classid;
1900 };
1901 
1902 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1903 {
1904 	struct tcf_bind_args *a = (void *)arg;
1905 
1906 	if (tp->ops->bind_class) {
1907 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1908 
1909 		sch_tree_lock(q);
1910 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1911 		sch_tree_unlock(q);
1912 	}
1913 	return 0;
1914 }
1915 
1916 struct tc_bind_class_args {
1917 	struct qdisc_walker w;
1918 	unsigned long new_cl;
1919 	u32 portid;
1920 	u32 clid;
1921 };
1922 
1923 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1924 				struct qdisc_walker *w)
1925 {
1926 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1927 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1928 	struct tcf_block *block;
1929 	struct tcf_chain *chain;
1930 
1931 	block = cops->tcf_block(q, cl, NULL);
1932 	if (!block)
1933 		return 0;
1934 	for (chain = tcf_get_next_chain(block, NULL);
1935 	     chain;
1936 	     chain = tcf_get_next_chain(block, chain)) {
1937 		struct tcf_proto *tp;
1938 
1939 		for (tp = tcf_get_next_proto(chain, NULL);
1940 		     tp; tp = tcf_get_next_proto(chain, tp)) {
1941 			struct tcf_bind_args arg = {};
1942 
1943 			arg.w.fn = tcf_node_bind;
1944 			arg.classid = a->clid;
1945 			arg.base = cl;
1946 			arg.cl = a->new_cl;
1947 			tp->ops->walk(tp, &arg.w, true);
1948 		}
1949 	}
1950 
1951 	return 0;
1952 }
1953 
1954 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1955 			   unsigned long new_cl)
1956 {
1957 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1958 	struct tc_bind_class_args args = {};
1959 
1960 	if (!cops->tcf_block)
1961 		return;
1962 	args.portid = portid;
1963 	args.clid = clid;
1964 	args.new_cl = new_cl;
1965 	args.w.fn = tc_bind_class_walker;
1966 	q->ops->cl_ops->walk(q, &args.w);
1967 }
1968 
1969 #else
1970 
1971 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1972 			   unsigned long new_cl)
1973 {
1974 }
1975 
1976 #endif
1977 
1978 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1979 			 struct netlink_ext_ack *extack)
1980 {
1981 	struct net *net = sock_net(skb->sk);
1982 	struct tcmsg *tcm = nlmsg_data(n);
1983 	struct nlattr *tca[TCA_MAX + 1];
1984 	struct net_device *dev;
1985 	struct Qdisc *q = NULL;
1986 	const struct Qdisc_class_ops *cops;
1987 	unsigned long cl = 0;
1988 	unsigned long new_cl;
1989 	u32 portid;
1990 	u32 clid;
1991 	u32 qid;
1992 	int err;
1993 
1994 	if ((n->nlmsg_type != RTM_GETTCLASS) &&
1995 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1996 		return -EPERM;
1997 
1998 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1999 				     rtm_tca_policy, extack);
2000 	if (err < 0)
2001 		return err;
2002 
2003 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2004 	if (!dev)
2005 		return -ENODEV;
2006 
2007 	/*
2008 	   parent == TC_H_UNSPEC - unspecified parent.
2009 	   parent == TC_H_ROOT   - class is root, which has no parent.
2010 	   parent == X:0	 - parent is root class.
2011 	   parent == X:Y	 - parent is a node in hierarchy.
2012 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2013 
2014 	   handle == 0:0	 - generate handle from kernel pool.
2015 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2016 	   handle == X:Y	 - clear.
2017 	   handle == X:0	 - root class.
2018 	 */
2019 
2020 	/* Step 1. Determine qdisc handle X:0 */
2021 
2022 	portid = tcm->tcm_parent;
2023 	clid = tcm->tcm_handle;
2024 	qid = TC_H_MAJ(clid);
2025 
2026 	if (portid != TC_H_ROOT) {
2027 		u32 qid1 = TC_H_MAJ(portid);
2028 
2029 		if (qid && qid1) {
2030 			/* If both majors are known, they must be identical. */
2031 			if (qid != qid1)
2032 				return -EINVAL;
2033 		} else if (qid1) {
2034 			qid = qid1;
2035 		} else if (qid == 0)
2036 			qid = dev->qdisc->handle;
2037 
2038 		/* Now qid is genuine qdisc handle consistent
2039 		 * both with parent and child.
2040 		 *
2041 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2042 		 */
2043 		if (portid)
2044 			portid = TC_H_MAKE(qid, portid);
2045 	} else {
2046 		if (qid == 0)
2047 			qid = dev->qdisc->handle;
2048 	}
2049 
2050 	/* OK. Locate qdisc */
2051 	q = qdisc_lookup(dev, qid);
2052 	if (!q)
2053 		return -ENOENT;
2054 
2055 	/* An check that it supports classes */
2056 	cops = q->ops->cl_ops;
2057 	if (cops == NULL)
2058 		return -EINVAL;
2059 
2060 	/* Now try to get class */
2061 	if (clid == 0) {
2062 		if (portid == TC_H_ROOT)
2063 			clid = qid;
2064 	} else
2065 		clid = TC_H_MAKE(qid, clid);
2066 
2067 	if (clid)
2068 		cl = cops->find(q, clid);
2069 
2070 	if (cl == 0) {
2071 		err = -ENOENT;
2072 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2073 		    !(n->nlmsg_flags & NLM_F_CREATE))
2074 			goto out;
2075 	} else {
2076 		switch (n->nlmsg_type) {
2077 		case RTM_NEWTCLASS:
2078 			err = -EEXIST;
2079 			if (n->nlmsg_flags & NLM_F_EXCL)
2080 				goto out;
2081 			break;
2082 		case RTM_DELTCLASS:
2083 			err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2084 			/* Unbind the class with flilters with 0 */
2085 			tc_bind_tclass(q, portid, clid, 0);
2086 			goto out;
2087 		case RTM_GETTCLASS:
2088 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
2089 			goto out;
2090 		default:
2091 			err = -EINVAL;
2092 			goto out;
2093 		}
2094 	}
2095 
2096 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2097 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2098 		return -EOPNOTSUPP;
2099 	}
2100 
2101 	new_cl = cl;
2102 	err = -EOPNOTSUPP;
2103 	if (cops->change)
2104 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2105 	if (err == 0) {
2106 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
2107 		/* We just create a new class, need to do reverse binding. */
2108 		if (cl != new_cl)
2109 			tc_bind_tclass(q, portid, clid, new_cl);
2110 	}
2111 out:
2112 	return err;
2113 }
2114 
2115 struct qdisc_dump_args {
2116 	struct qdisc_walker	w;
2117 	struct sk_buff		*skb;
2118 	struct netlink_callback	*cb;
2119 };
2120 
2121 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2122 			    struct qdisc_walker *arg)
2123 {
2124 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2125 
2126 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2127 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2128 			      RTM_NEWTCLASS);
2129 }
2130 
2131 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2132 				struct tcmsg *tcm, struct netlink_callback *cb,
2133 				int *t_p, int s_t)
2134 {
2135 	struct qdisc_dump_args arg;
2136 
2137 	if (tc_qdisc_dump_ignore(q, false) ||
2138 	    *t_p < s_t || !q->ops->cl_ops ||
2139 	    (tcm->tcm_parent &&
2140 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2141 		(*t_p)++;
2142 		return 0;
2143 	}
2144 	if (*t_p > s_t)
2145 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2146 	arg.w.fn = qdisc_class_dump;
2147 	arg.skb = skb;
2148 	arg.cb = cb;
2149 	arg.w.stop  = 0;
2150 	arg.w.skip = cb->args[1];
2151 	arg.w.count = 0;
2152 	q->ops->cl_ops->walk(q, &arg.w);
2153 	cb->args[1] = arg.w.count;
2154 	if (arg.w.stop)
2155 		return -1;
2156 	(*t_p)++;
2157 	return 0;
2158 }
2159 
2160 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2161 			       struct tcmsg *tcm, struct netlink_callback *cb,
2162 			       int *t_p, int s_t, bool recur)
2163 {
2164 	struct Qdisc *q;
2165 	int b;
2166 
2167 	if (!root)
2168 		return 0;
2169 
2170 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2171 		return -1;
2172 
2173 	if (!qdisc_dev(root) || !recur)
2174 		return 0;
2175 
2176 	if (tcm->tcm_parent) {
2177 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2178 		if (q && q != root &&
2179 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2180 			return -1;
2181 		return 0;
2182 	}
2183 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2184 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2185 			return -1;
2186 	}
2187 
2188 	return 0;
2189 }
2190 
2191 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2192 {
2193 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2194 	struct net *net = sock_net(skb->sk);
2195 	struct netdev_queue *dev_queue;
2196 	struct net_device *dev;
2197 	int t, s_t;
2198 
2199 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2200 		return 0;
2201 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2202 	if (!dev)
2203 		return 0;
2204 
2205 	s_t = cb->args[0];
2206 	t = 0;
2207 
2208 	if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t, true) < 0)
2209 		goto done;
2210 
2211 	dev_queue = dev_ingress_queue(dev);
2212 	if (dev_queue &&
2213 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2214 				&t, s_t, false) < 0)
2215 		goto done;
2216 
2217 done:
2218 	cb->args[0] = t;
2219 
2220 	dev_put(dev);
2221 	return skb->len;
2222 }
2223 
2224 #ifdef CONFIG_PROC_FS
2225 static int psched_show(struct seq_file *seq, void *v)
2226 {
2227 	seq_printf(seq, "%08x %08x %08x %08x\n",
2228 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2229 		   1000000,
2230 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2231 
2232 	return 0;
2233 }
2234 
2235 static int __net_init psched_net_init(struct net *net)
2236 {
2237 	struct proc_dir_entry *e;
2238 
2239 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2240 	if (e == NULL)
2241 		return -ENOMEM;
2242 
2243 	return 0;
2244 }
2245 
2246 static void __net_exit psched_net_exit(struct net *net)
2247 {
2248 	remove_proc_entry("psched", net->proc_net);
2249 }
2250 #else
2251 static int __net_init psched_net_init(struct net *net)
2252 {
2253 	return 0;
2254 }
2255 
2256 static void __net_exit psched_net_exit(struct net *net)
2257 {
2258 }
2259 #endif
2260 
2261 static struct pernet_operations psched_net_ops = {
2262 	.init = psched_net_init,
2263 	.exit = psched_net_exit,
2264 };
2265 
2266 static int __init pktsched_init(void)
2267 {
2268 	int err;
2269 
2270 	err = register_pernet_subsys(&psched_net_ops);
2271 	if (err) {
2272 		pr_err("pktsched_init: "
2273 		       "cannot initialize per netns operations\n");
2274 		return err;
2275 	}
2276 
2277 	register_qdisc(&pfifo_fast_ops);
2278 	register_qdisc(&pfifo_qdisc_ops);
2279 	register_qdisc(&bfifo_qdisc_ops);
2280 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2281 	register_qdisc(&mq_qdisc_ops);
2282 	register_qdisc(&noqueue_qdisc_ops);
2283 
2284 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2285 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2286 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2287 		      0);
2288 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2289 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2290 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2291 		      0);
2292 
2293 	return 0;
2294 }
2295 
2296 subsys_initcall(pktsched_init);
2297