xref: /openbmc/linux/net/sched/sch_api.c (revision 917edfb98c480fec1dce95ece6707779905ca0b5)
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 	strlcpy(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 static void qdisc_offload_graft_root(struct net_device *dev,
872 				     struct Qdisc *new, struct Qdisc *old,
873 				     struct netlink_ext_ack *extack)
874 {
875 	struct tc_root_qopt_offload graft_offload = {
876 		.command	= TC_ROOT_GRAFT,
877 		.handle		= new ? new->handle : 0,
878 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
879 				  (old && old->flags & TCQ_F_INGRESS),
880 	};
881 
882 	qdisc_offload_graft_helper(dev, NULL, new, old,
883 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
884 }
885 
886 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
887 			 u32 portid, u32 seq, u16 flags, int event)
888 {
889 	struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
890 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
891 	struct tcmsg *tcm;
892 	struct nlmsghdr  *nlh;
893 	unsigned char *b = skb_tail_pointer(skb);
894 	struct gnet_dump d;
895 	struct qdisc_size_table *stab;
896 	u32 block_index;
897 	__u32 qlen;
898 
899 	cond_resched();
900 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
901 	if (!nlh)
902 		goto out_nlmsg_trim;
903 	tcm = nlmsg_data(nlh);
904 	tcm->tcm_family = AF_UNSPEC;
905 	tcm->tcm__pad1 = 0;
906 	tcm->tcm__pad2 = 0;
907 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
908 	tcm->tcm_parent = clid;
909 	tcm->tcm_handle = q->handle;
910 	tcm->tcm_info = refcount_read(&q->refcnt);
911 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
912 		goto nla_put_failure;
913 	if (q->ops->ingress_block_get) {
914 		block_index = q->ops->ingress_block_get(q);
915 		if (block_index &&
916 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
917 			goto nla_put_failure;
918 	}
919 	if (q->ops->egress_block_get) {
920 		block_index = q->ops->egress_block_get(q);
921 		if (block_index &&
922 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
923 			goto nla_put_failure;
924 	}
925 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
926 		goto nla_put_failure;
927 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
928 		goto nla_put_failure;
929 	qlen = qdisc_qlen_sum(q);
930 
931 	stab = rtnl_dereference(q->stab);
932 	if (stab && qdisc_dump_stab(skb, stab) < 0)
933 		goto nla_put_failure;
934 
935 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
936 					 NULL, &d, TCA_PAD) < 0)
937 		goto nla_put_failure;
938 
939 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
940 		goto nla_put_failure;
941 
942 	if (qdisc_is_percpu_stats(q)) {
943 		cpu_bstats = q->cpu_bstats;
944 		cpu_qstats = q->cpu_qstats;
945 	}
946 
947 	if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
948 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
949 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
950 		goto nla_put_failure;
951 
952 	if (gnet_stats_finish_copy(&d) < 0)
953 		goto nla_put_failure;
954 
955 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
956 	return skb->len;
957 
958 out_nlmsg_trim:
959 nla_put_failure:
960 	nlmsg_trim(skb, b);
961 	return -1;
962 }
963 
964 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
965 {
966 	if (q->flags & TCQ_F_BUILTIN)
967 		return true;
968 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
969 		return true;
970 
971 	return false;
972 }
973 
974 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
975 			struct nlmsghdr *n, u32 clid,
976 			struct Qdisc *old, struct Qdisc *new)
977 {
978 	struct sk_buff *skb;
979 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
980 
981 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
982 	if (!skb)
983 		return -ENOBUFS;
984 
985 	if (old && !tc_qdisc_dump_ignore(old, false)) {
986 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
987 				  0, RTM_DELQDISC) < 0)
988 			goto err_out;
989 	}
990 	if (new && !tc_qdisc_dump_ignore(new, false)) {
991 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
992 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
993 			goto err_out;
994 	}
995 
996 	if (skb->len)
997 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
998 				      n->nlmsg_flags & NLM_F_ECHO);
999 
1000 err_out:
1001 	kfree_skb(skb);
1002 	return -EINVAL;
1003 }
1004 
1005 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1006 			       struct nlmsghdr *n, u32 clid,
1007 			       struct Qdisc *old, struct Qdisc *new)
1008 {
1009 	if (new || old)
1010 		qdisc_notify(net, skb, n, clid, old, new);
1011 
1012 	if (old)
1013 		qdisc_put(old);
1014 }
1015 
1016 static void qdisc_clear_nolock(struct Qdisc *sch)
1017 {
1018 	sch->flags &= ~TCQ_F_NOLOCK;
1019 	if (!(sch->flags & TCQ_F_CPUSTATS))
1020 		return;
1021 
1022 	free_percpu(sch->cpu_bstats);
1023 	free_percpu(sch->cpu_qstats);
1024 	sch->cpu_bstats = NULL;
1025 	sch->cpu_qstats = NULL;
1026 	sch->flags &= ~TCQ_F_CPUSTATS;
1027 }
1028 
1029 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1030  * to device "dev".
1031  *
1032  * When appropriate send a netlink notification using 'skb'
1033  * and "n".
1034  *
1035  * On success, destroy old qdisc.
1036  */
1037 
1038 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1039 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1040 		       struct Qdisc *new, struct Qdisc *old,
1041 		       struct netlink_ext_ack *extack)
1042 {
1043 	struct Qdisc *q = old;
1044 	struct net *net = dev_net(dev);
1045 
1046 	if (parent == NULL) {
1047 		unsigned int i, num_q, ingress;
1048 
1049 		ingress = 0;
1050 		num_q = dev->num_tx_queues;
1051 		if ((q && q->flags & TCQ_F_INGRESS) ||
1052 		    (new && new->flags & TCQ_F_INGRESS)) {
1053 			num_q = 1;
1054 			ingress = 1;
1055 			if (!dev_ingress_queue(dev)) {
1056 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1057 				return -ENOENT;
1058 			}
1059 		}
1060 
1061 		if (dev->flags & IFF_UP)
1062 			dev_deactivate(dev);
1063 
1064 		qdisc_offload_graft_root(dev, new, old, extack);
1065 
1066 		if (new && new->ops->attach && !ingress)
1067 			goto skip;
1068 
1069 		for (i = 0; i < num_q; i++) {
1070 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1071 
1072 			if (!ingress)
1073 				dev_queue = netdev_get_tx_queue(dev, i);
1074 
1075 			old = dev_graft_qdisc(dev_queue, new);
1076 			if (new && i > 0)
1077 				qdisc_refcount_inc(new);
1078 
1079 			if (!ingress)
1080 				qdisc_put(old);
1081 		}
1082 
1083 skip:
1084 		if (!ingress) {
1085 			notify_and_destroy(net, skb, n, classid,
1086 					   rtnl_dereference(dev->qdisc), new);
1087 			if (new && !new->ops->attach)
1088 				qdisc_refcount_inc(new);
1089 			rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1090 
1091 			if (new && new->ops->attach)
1092 				new->ops->attach(new);
1093 		} else {
1094 			notify_and_destroy(net, skb, n, classid, old, new);
1095 		}
1096 
1097 		if (dev->flags & IFF_UP)
1098 			dev_activate(dev);
1099 	} else {
1100 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1101 		unsigned long cl;
1102 		int err;
1103 
1104 		/* Only support running class lockless if parent is lockless */
1105 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1106 			qdisc_clear_nolock(new);
1107 
1108 		if (!cops || !cops->graft)
1109 			return -EOPNOTSUPP;
1110 
1111 		cl = cops->find(parent, classid);
1112 		if (!cl) {
1113 			NL_SET_ERR_MSG(extack, "Specified class not found");
1114 			return -ENOENT;
1115 		}
1116 
1117 		err = cops->graft(parent, cl, new, &old, extack);
1118 		if (err)
1119 			return err;
1120 		notify_and_destroy(net, skb, n, classid, old, new);
1121 	}
1122 	return 0;
1123 }
1124 
1125 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1126 				   struct netlink_ext_ack *extack)
1127 {
1128 	u32 block_index;
1129 
1130 	if (tca[TCA_INGRESS_BLOCK]) {
1131 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1132 
1133 		if (!block_index) {
1134 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1135 			return -EINVAL;
1136 		}
1137 		if (!sch->ops->ingress_block_set) {
1138 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1139 			return -EOPNOTSUPP;
1140 		}
1141 		sch->ops->ingress_block_set(sch, block_index);
1142 	}
1143 	if (tca[TCA_EGRESS_BLOCK]) {
1144 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1145 
1146 		if (!block_index) {
1147 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1148 			return -EINVAL;
1149 		}
1150 		if (!sch->ops->egress_block_set) {
1151 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1152 			return -EOPNOTSUPP;
1153 		}
1154 		sch->ops->egress_block_set(sch, block_index);
1155 	}
1156 	return 0;
1157 }
1158 
1159 /*
1160    Allocate and initialize new qdisc.
1161 
1162    Parameters are passed via opt.
1163  */
1164 
1165 static struct Qdisc *qdisc_create(struct net_device *dev,
1166 				  struct netdev_queue *dev_queue,
1167 				  u32 parent, u32 handle,
1168 				  struct nlattr **tca, int *errp,
1169 				  struct netlink_ext_ack *extack)
1170 {
1171 	int err;
1172 	struct nlattr *kind = tca[TCA_KIND];
1173 	struct Qdisc *sch;
1174 	struct Qdisc_ops *ops;
1175 	struct qdisc_size_table *stab;
1176 
1177 	ops = qdisc_lookup_ops(kind);
1178 #ifdef CONFIG_MODULES
1179 	if (ops == NULL && kind != NULL) {
1180 		char name[IFNAMSIZ];
1181 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1182 			/* We dropped the RTNL semaphore in order to
1183 			 * perform the module load.  So, even if we
1184 			 * succeeded in loading the module we have to
1185 			 * tell the caller to replay the request.  We
1186 			 * indicate this using -EAGAIN.
1187 			 * We replay the request because the device may
1188 			 * go away in the mean time.
1189 			 */
1190 			rtnl_unlock();
1191 			request_module("sch_%s", name);
1192 			rtnl_lock();
1193 			ops = qdisc_lookup_ops(kind);
1194 			if (ops != NULL) {
1195 				/* We will try again qdisc_lookup_ops,
1196 				 * so don't keep a reference.
1197 				 */
1198 				module_put(ops->owner);
1199 				err = -EAGAIN;
1200 				goto err_out;
1201 			}
1202 		}
1203 	}
1204 #endif
1205 
1206 	err = -ENOENT;
1207 	if (!ops) {
1208 		NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1209 		goto err_out;
1210 	}
1211 
1212 	sch = qdisc_alloc(dev_queue, ops, extack);
1213 	if (IS_ERR(sch)) {
1214 		err = PTR_ERR(sch);
1215 		goto err_out2;
1216 	}
1217 
1218 	sch->parent = parent;
1219 
1220 	if (handle == TC_H_INGRESS) {
1221 		sch->flags |= TCQ_F_INGRESS;
1222 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1223 	} else {
1224 		if (handle == 0) {
1225 			handle = qdisc_alloc_handle(dev);
1226 			if (handle == 0) {
1227 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1228 				err = -ENOSPC;
1229 				goto err_out3;
1230 			}
1231 		}
1232 		if (!netif_is_multiqueue(dev))
1233 			sch->flags |= TCQ_F_ONETXQUEUE;
1234 	}
1235 
1236 	sch->handle = handle;
1237 
1238 	/* This exist to keep backward compatible with a userspace
1239 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1240 	 * facility on older kernels by setting tx_queue_len=0 (prior
1241 	 * to qdisc init), and then forgot to reinit tx_queue_len
1242 	 * before again attaching a qdisc.
1243 	 */
1244 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1245 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1246 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1247 	}
1248 
1249 	err = qdisc_block_indexes_set(sch, tca, extack);
1250 	if (err)
1251 		goto err_out3;
1252 
1253 	if (ops->init) {
1254 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1255 		if (err != 0)
1256 			goto err_out5;
1257 	}
1258 
1259 	if (tca[TCA_STAB]) {
1260 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1261 		if (IS_ERR(stab)) {
1262 			err = PTR_ERR(stab);
1263 			goto err_out4;
1264 		}
1265 		rcu_assign_pointer(sch->stab, stab);
1266 	}
1267 	if (tca[TCA_RATE]) {
1268 		err = -EOPNOTSUPP;
1269 		if (sch->flags & TCQ_F_MQROOT) {
1270 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1271 			goto err_out4;
1272 		}
1273 
1274 		err = gen_new_estimator(&sch->bstats,
1275 					sch->cpu_bstats,
1276 					&sch->rate_est,
1277 					NULL,
1278 					true,
1279 					tca[TCA_RATE]);
1280 		if (err) {
1281 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1282 			goto err_out4;
1283 		}
1284 	}
1285 
1286 	qdisc_hash_add(sch, false);
1287 	trace_qdisc_create(ops, dev, parent);
1288 
1289 	return sch;
1290 
1291 err_out5:
1292 	/* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1293 	if (ops->destroy)
1294 		ops->destroy(sch);
1295 err_out3:
1296 	netdev_put(dev, &sch->dev_tracker);
1297 	qdisc_free(sch);
1298 err_out2:
1299 	module_put(ops->owner);
1300 err_out:
1301 	*errp = err;
1302 	return NULL;
1303 
1304 err_out4:
1305 	/*
1306 	 * Any broken qdiscs that would require a ops->reset() here?
1307 	 * The qdisc was never in action so it shouldn't be necessary.
1308 	 */
1309 	qdisc_put_stab(rtnl_dereference(sch->stab));
1310 	if (ops->destroy)
1311 		ops->destroy(sch);
1312 	goto err_out3;
1313 }
1314 
1315 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1316 			struct netlink_ext_ack *extack)
1317 {
1318 	struct qdisc_size_table *ostab, *stab = NULL;
1319 	int err = 0;
1320 
1321 	if (tca[TCA_OPTIONS]) {
1322 		if (!sch->ops->change) {
1323 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1324 			return -EINVAL;
1325 		}
1326 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1327 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1328 			return -EOPNOTSUPP;
1329 		}
1330 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1331 		if (err)
1332 			return err;
1333 	}
1334 
1335 	if (tca[TCA_STAB]) {
1336 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1337 		if (IS_ERR(stab))
1338 			return PTR_ERR(stab);
1339 	}
1340 
1341 	ostab = rtnl_dereference(sch->stab);
1342 	rcu_assign_pointer(sch->stab, stab);
1343 	qdisc_put_stab(ostab);
1344 
1345 	if (tca[TCA_RATE]) {
1346 		/* NB: ignores errors from replace_estimator
1347 		   because change can't be undone. */
1348 		if (sch->flags & TCQ_F_MQROOT)
1349 			goto out;
1350 		gen_replace_estimator(&sch->bstats,
1351 				      sch->cpu_bstats,
1352 				      &sch->rate_est,
1353 				      NULL,
1354 				      true,
1355 				      tca[TCA_RATE]);
1356 	}
1357 out:
1358 	return 0;
1359 }
1360 
1361 struct check_loop_arg {
1362 	struct qdisc_walker	w;
1363 	struct Qdisc		*p;
1364 	int			depth;
1365 };
1366 
1367 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1368 			 struct qdisc_walker *w);
1369 
1370 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1371 {
1372 	struct check_loop_arg	arg;
1373 
1374 	if (q->ops->cl_ops == NULL)
1375 		return 0;
1376 
1377 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1378 	arg.w.fn = check_loop_fn;
1379 	arg.depth = depth;
1380 	arg.p = p;
1381 	q->ops->cl_ops->walk(q, &arg.w);
1382 	return arg.w.stop ? -ELOOP : 0;
1383 }
1384 
1385 static int
1386 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1387 {
1388 	struct Qdisc *leaf;
1389 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1390 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1391 
1392 	leaf = cops->leaf(q, cl);
1393 	if (leaf) {
1394 		if (leaf == arg->p || arg->depth > 7)
1395 			return -ELOOP;
1396 		return check_loop(leaf, arg->p, arg->depth + 1);
1397 	}
1398 	return 0;
1399 }
1400 
1401 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1402 	[TCA_KIND]		= { .type = NLA_STRING },
1403 	[TCA_RATE]		= { .type = NLA_BINARY,
1404 				    .len = sizeof(struct tc_estimator) },
1405 	[TCA_STAB]		= { .type = NLA_NESTED },
1406 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1407 	[TCA_CHAIN]		= { .type = NLA_U32 },
1408 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1409 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1410 };
1411 
1412 /*
1413  * Delete/get qdisc.
1414  */
1415 
1416 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1417 			struct netlink_ext_ack *extack)
1418 {
1419 	struct net *net = sock_net(skb->sk);
1420 	struct tcmsg *tcm = nlmsg_data(n);
1421 	struct nlattr *tca[TCA_MAX + 1];
1422 	struct net_device *dev;
1423 	u32 clid;
1424 	struct Qdisc *q = NULL;
1425 	struct Qdisc *p = NULL;
1426 	int err;
1427 
1428 	if ((n->nlmsg_type != RTM_GETQDISC) &&
1429 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1430 		return -EPERM;
1431 
1432 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1433 				     rtm_tca_policy, extack);
1434 	if (err < 0)
1435 		return err;
1436 
1437 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1438 	if (!dev)
1439 		return -ENODEV;
1440 
1441 	clid = tcm->tcm_parent;
1442 	if (clid) {
1443 		if (clid != TC_H_ROOT) {
1444 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1445 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1446 				if (!p) {
1447 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1448 					return -ENOENT;
1449 				}
1450 				q = qdisc_leaf(p, clid);
1451 			} else if (dev_ingress_queue(dev)) {
1452 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1453 			}
1454 		} else {
1455 			q = rtnl_dereference(dev->qdisc);
1456 		}
1457 		if (!q) {
1458 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1459 			return -ENOENT;
1460 		}
1461 
1462 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1463 			NL_SET_ERR_MSG(extack, "Invalid handle");
1464 			return -EINVAL;
1465 		}
1466 	} else {
1467 		q = qdisc_lookup(dev, tcm->tcm_handle);
1468 		if (!q) {
1469 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1470 			return -ENOENT;
1471 		}
1472 	}
1473 
1474 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1475 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1476 		return -EINVAL;
1477 	}
1478 
1479 	if (n->nlmsg_type == RTM_DELQDISC) {
1480 		if (!clid) {
1481 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1482 			return -EINVAL;
1483 		}
1484 		if (q->handle == 0) {
1485 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1486 			return -ENOENT;
1487 		}
1488 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1489 		if (err != 0)
1490 			return err;
1491 	} else {
1492 		qdisc_notify(net, skb, n, clid, NULL, q);
1493 	}
1494 	return 0;
1495 }
1496 
1497 /*
1498  * Create/change qdisc.
1499  */
1500 
1501 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1502 			   struct netlink_ext_ack *extack)
1503 {
1504 	struct net *net = sock_net(skb->sk);
1505 	struct tcmsg *tcm;
1506 	struct nlattr *tca[TCA_MAX + 1];
1507 	struct net_device *dev;
1508 	u32 clid;
1509 	struct Qdisc *q, *p;
1510 	int err;
1511 
1512 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1513 		return -EPERM;
1514 
1515 replay:
1516 	/* Reinit, just in case something touches this. */
1517 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1518 				     rtm_tca_policy, extack);
1519 	if (err < 0)
1520 		return err;
1521 
1522 	tcm = nlmsg_data(n);
1523 	clid = tcm->tcm_parent;
1524 	q = p = NULL;
1525 
1526 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1527 	if (!dev)
1528 		return -ENODEV;
1529 
1530 
1531 	if (clid) {
1532 		if (clid != TC_H_ROOT) {
1533 			if (clid != TC_H_INGRESS) {
1534 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1535 				if (!p) {
1536 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1537 					return -ENOENT;
1538 				}
1539 				q = qdisc_leaf(p, clid);
1540 			} else if (dev_ingress_queue_create(dev)) {
1541 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1542 			}
1543 		} else {
1544 			q = rtnl_dereference(dev->qdisc);
1545 		}
1546 
1547 		/* It may be default qdisc, ignore it */
1548 		if (q && q->handle == 0)
1549 			q = NULL;
1550 
1551 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1552 			if (tcm->tcm_handle) {
1553 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1554 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1555 					return -EEXIST;
1556 				}
1557 				if (TC_H_MIN(tcm->tcm_handle)) {
1558 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1559 					return -EINVAL;
1560 				}
1561 				q = qdisc_lookup(dev, tcm->tcm_handle);
1562 				if (!q)
1563 					goto create_n_graft;
1564 				if (n->nlmsg_flags & NLM_F_EXCL) {
1565 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1566 					return -EEXIST;
1567 				}
1568 				if (tca[TCA_KIND] &&
1569 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1570 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1571 					return -EINVAL;
1572 				}
1573 				if (q == p ||
1574 				    (p && check_loop(q, p, 0))) {
1575 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1576 					return -ELOOP;
1577 				}
1578 				qdisc_refcount_inc(q);
1579 				goto graft;
1580 			} else {
1581 				if (!q)
1582 					goto create_n_graft;
1583 
1584 				/* This magic test requires explanation.
1585 				 *
1586 				 *   We know, that some child q is already
1587 				 *   attached to this parent and have choice:
1588 				 *   either to change it or to create/graft new one.
1589 				 *
1590 				 *   1. We are allowed to create/graft only
1591 				 *   if CREATE and REPLACE flags are set.
1592 				 *
1593 				 *   2. If EXCL is set, requestor wanted to say,
1594 				 *   that qdisc tcm_handle is not expected
1595 				 *   to exist, so that we choose create/graft too.
1596 				 *
1597 				 *   3. The last case is when no flags are set.
1598 				 *   Alas, it is sort of hole in API, we
1599 				 *   cannot decide what to do unambiguously.
1600 				 *   For now we select create/graft, if
1601 				 *   user gave KIND, which does not match existing.
1602 				 */
1603 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1604 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1605 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1606 				     (tca[TCA_KIND] &&
1607 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1608 					goto create_n_graft;
1609 			}
1610 		}
1611 	} else {
1612 		if (!tcm->tcm_handle) {
1613 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1614 			return -EINVAL;
1615 		}
1616 		q = qdisc_lookup(dev, tcm->tcm_handle);
1617 	}
1618 
1619 	/* Change qdisc parameters */
1620 	if (!q) {
1621 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1622 		return -ENOENT;
1623 	}
1624 	if (n->nlmsg_flags & NLM_F_EXCL) {
1625 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1626 		return -EEXIST;
1627 	}
1628 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1629 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1630 		return -EINVAL;
1631 	}
1632 	err = qdisc_change(q, tca, extack);
1633 	if (err == 0)
1634 		qdisc_notify(net, skb, n, clid, NULL, q);
1635 	return err;
1636 
1637 create_n_graft:
1638 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1639 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1640 		return -ENOENT;
1641 	}
1642 	if (clid == TC_H_INGRESS) {
1643 		if (dev_ingress_queue(dev)) {
1644 			q = qdisc_create(dev, dev_ingress_queue(dev),
1645 					 tcm->tcm_parent, tcm->tcm_parent,
1646 					 tca, &err, extack);
1647 		} else {
1648 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1649 			err = -ENOENT;
1650 		}
1651 	} else {
1652 		struct netdev_queue *dev_queue;
1653 
1654 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1655 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1656 		else if (p)
1657 			dev_queue = p->dev_queue;
1658 		else
1659 			dev_queue = netdev_get_tx_queue(dev, 0);
1660 
1661 		q = qdisc_create(dev, dev_queue,
1662 				 tcm->tcm_parent, tcm->tcm_handle,
1663 				 tca, &err, extack);
1664 	}
1665 	if (q == NULL) {
1666 		if (err == -EAGAIN)
1667 			goto replay;
1668 		return err;
1669 	}
1670 
1671 graft:
1672 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1673 	if (err) {
1674 		if (q)
1675 			qdisc_put(q);
1676 		return err;
1677 	}
1678 
1679 	return 0;
1680 }
1681 
1682 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1683 			      struct netlink_callback *cb,
1684 			      int *q_idx_p, int s_q_idx, bool recur,
1685 			      bool dump_invisible)
1686 {
1687 	int ret = 0, q_idx = *q_idx_p;
1688 	struct Qdisc *q;
1689 	int b;
1690 
1691 	if (!root)
1692 		return 0;
1693 
1694 	q = root;
1695 	if (q_idx < s_q_idx) {
1696 		q_idx++;
1697 	} else {
1698 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1699 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1700 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1701 				  RTM_NEWQDISC) <= 0)
1702 			goto done;
1703 		q_idx++;
1704 	}
1705 
1706 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1707 	 * itself has already been dumped.
1708 	 *
1709 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1710 	 * qdisc hashtable, we don't want to hit it again
1711 	 */
1712 	if (!qdisc_dev(root) || !recur)
1713 		goto out;
1714 
1715 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1716 		if (q_idx < s_q_idx) {
1717 			q_idx++;
1718 			continue;
1719 		}
1720 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1721 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1722 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1723 				  RTM_NEWQDISC) <= 0)
1724 			goto done;
1725 		q_idx++;
1726 	}
1727 
1728 out:
1729 	*q_idx_p = q_idx;
1730 	return ret;
1731 done:
1732 	ret = -1;
1733 	goto out;
1734 }
1735 
1736 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1737 {
1738 	struct net *net = sock_net(skb->sk);
1739 	int idx, q_idx;
1740 	int s_idx, s_q_idx;
1741 	struct net_device *dev;
1742 	const struct nlmsghdr *nlh = cb->nlh;
1743 	struct nlattr *tca[TCA_MAX + 1];
1744 	int err;
1745 
1746 	s_idx = cb->args[0];
1747 	s_q_idx = q_idx = cb->args[1];
1748 
1749 	idx = 0;
1750 	ASSERT_RTNL();
1751 
1752 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1753 				     rtm_tca_policy, cb->extack);
1754 	if (err < 0)
1755 		return err;
1756 
1757 	for_each_netdev(net, dev) {
1758 		struct netdev_queue *dev_queue;
1759 
1760 		if (idx < s_idx)
1761 			goto cont;
1762 		if (idx > s_idx)
1763 			s_q_idx = 0;
1764 		q_idx = 0;
1765 
1766 		if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1767 				       skb, cb, &q_idx, s_q_idx,
1768 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1769 			goto done;
1770 
1771 		dev_queue = dev_ingress_queue(dev);
1772 		if (dev_queue &&
1773 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1774 				       &q_idx, s_q_idx, false,
1775 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1776 			goto done;
1777 
1778 cont:
1779 		idx++;
1780 	}
1781 
1782 done:
1783 	cb->args[0] = idx;
1784 	cb->args[1] = q_idx;
1785 
1786 	return skb->len;
1787 }
1788 
1789 
1790 
1791 /************************************************
1792  *	Traffic classes manipulation.		*
1793  ************************************************/
1794 
1795 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1796 			  unsigned long cl,
1797 			  u32 portid, u32 seq, u16 flags, int event)
1798 {
1799 	struct tcmsg *tcm;
1800 	struct nlmsghdr  *nlh;
1801 	unsigned char *b = skb_tail_pointer(skb);
1802 	struct gnet_dump d;
1803 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1804 
1805 	cond_resched();
1806 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1807 	if (!nlh)
1808 		goto out_nlmsg_trim;
1809 	tcm = nlmsg_data(nlh);
1810 	tcm->tcm_family = AF_UNSPEC;
1811 	tcm->tcm__pad1 = 0;
1812 	tcm->tcm__pad2 = 0;
1813 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1814 	tcm->tcm_parent = q->handle;
1815 	tcm->tcm_handle = q->handle;
1816 	tcm->tcm_info = 0;
1817 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1818 		goto nla_put_failure;
1819 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1820 		goto nla_put_failure;
1821 
1822 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1823 					 NULL, &d, TCA_PAD) < 0)
1824 		goto nla_put_failure;
1825 
1826 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1827 		goto nla_put_failure;
1828 
1829 	if (gnet_stats_finish_copy(&d) < 0)
1830 		goto nla_put_failure;
1831 
1832 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1833 	return skb->len;
1834 
1835 out_nlmsg_trim:
1836 nla_put_failure:
1837 	nlmsg_trim(skb, b);
1838 	return -1;
1839 }
1840 
1841 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1842 			 struct nlmsghdr *n, struct Qdisc *q,
1843 			 unsigned long cl, int event)
1844 {
1845 	struct sk_buff *skb;
1846 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1847 
1848 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1849 	if (!skb)
1850 		return -ENOBUFS;
1851 
1852 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1853 		kfree_skb(skb);
1854 		return -EINVAL;
1855 	}
1856 
1857 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1858 			      n->nlmsg_flags & NLM_F_ECHO);
1859 }
1860 
1861 static int tclass_del_notify(struct net *net,
1862 			     const struct Qdisc_class_ops *cops,
1863 			     struct sk_buff *oskb, struct nlmsghdr *n,
1864 			     struct Qdisc *q, unsigned long cl,
1865 			     struct netlink_ext_ack *extack)
1866 {
1867 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1868 	struct sk_buff *skb;
1869 	int err = 0;
1870 
1871 	if (!cops->delete)
1872 		return -EOPNOTSUPP;
1873 
1874 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1875 	if (!skb)
1876 		return -ENOBUFS;
1877 
1878 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1879 			   RTM_DELTCLASS) < 0) {
1880 		kfree_skb(skb);
1881 		return -EINVAL;
1882 	}
1883 
1884 	err = cops->delete(q, cl, extack);
1885 	if (err) {
1886 		kfree_skb(skb);
1887 		return err;
1888 	}
1889 
1890 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1891 			     n->nlmsg_flags & NLM_F_ECHO);
1892 	return err;
1893 }
1894 
1895 #ifdef CONFIG_NET_CLS
1896 
1897 struct tcf_bind_args {
1898 	struct tcf_walker w;
1899 	unsigned long base;
1900 	unsigned long cl;
1901 	u32 classid;
1902 };
1903 
1904 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1905 {
1906 	struct tcf_bind_args *a = (void *)arg;
1907 
1908 	if (tp->ops->bind_class) {
1909 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1910 
1911 		sch_tree_lock(q);
1912 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1913 		sch_tree_unlock(q);
1914 	}
1915 	return 0;
1916 }
1917 
1918 struct tc_bind_class_args {
1919 	struct qdisc_walker w;
1920 	unsigned long new_cl;
1921 	u32 portid;
1922 	u32 clid;
1923 };
1924 
1925 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1926 				struct qdisc_walker *w)
1927 {
1928 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1929 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1930 	struct tcf_block *block;
1931 	struct tcf_chain *chain;
1932 
1933 	block = cops->tcf_block(q, cl, NULL);
1934 	if (!block)
1935 		return 0;
1936 	for (chain = tcf_get_next_chain(block, NULL);
1937 	     chain;
1938 	     chain = tcf_get_next_chain(block, chain)) {
1939 		struct tcf_proto *tp;
1940 
1941 		for (tp = tcf_get_next_proto(chain, NULL);
1942 		     tp; tp = tcf_get_next_proto(chain, tp)) {
1943 			struct tcf_bind_args arg = {};
1944 
1945 			arg.w.fn = tcf_node_bind;
1946 			arg.classid = a->clid;
1947 			arg.base = cl;
1948 			arg.cl = a->new_cl;
1949 			tp->ops->walk(tp, &arg.w, true);
1950 		}
1951 	}
1952 
1953 	return 0;
1954 }
1955 
1956 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1957 			   unsigned long new_cl)
1958 {
1959 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1960 	struct tc_bind_class_args args = {};
1961 
1962 	if (!cops->tcf_block)
1963 		return;
1964 	args.portid = portid;
1965 	args.clid = clid;
1966 	args.new_cl = new_cl;
1967 	args.w.fn = tc_bind_class_walker;
1968 	q->ops->cl_ops->walk(q, &args.w);
1969 }
1970 
1971 #else
1972 
1973 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1974 			   unsigned long new_cl)
1975 {
1976 }
1977 
1978 #endif
1979 
1980 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1981 			 struct netlink_ext_ack *extack)
1982 {
1983 	struct net *net = sock_net(skb->sk);
1984 	struct tcmsg *tcm = nlmsg_data(n);
1985 	struct nlattr *tca[TCA_MAX + 1];
1986 	struct net_device *dev;
1987 	struct Qdisc *q = NULL;
1988 	const struct Qdisc_class_ops *cops;
1989 	unsigned long cl = 0;
1990 	unsigned long new_cl;
1991 	u32 portid;
1992 	u32 clid;
1993 	u32 qid;
1994 	int err;
1995 
1996 	if ((n->nlmsg_type != RTM_GETTCLASS) &&
1997 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1998 		return -EPERM;
1999 
2000 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2001 				     rtm_tca_policy, extack);
2002 	if (err < 0)
2003 		return err;
2004 
2005 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2006 	if (!dev)
2007 		return -ENODEV;
2008 
2009 	/*
2010 	   parent == TC_H_UNSPEC - unspecified parent.
2011 	   parent == TC_H_ROOT   - class is root, which has no parent.
2012 	   parent == X:0	 - parent is root class.
2013 	   parent == X:Y	 - parent is a node in hierarchy.
2014 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2015 
2016 	   handle == 0:0	 - generate handle from kernel pool.
2017 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2018 	   handle == X:Y	 - clear.
2019 	   handle == X:0	 - root class.
2020 	 */
2021 
2022 	/* Step 1. Determine qdisc handle X:0 */
2023 
2024 	portid = tcm->tcm_parent;
2025 	clid = tcm->tcm_handle;
2026 	qid = TC_H_MAJ(clid);
2027 
2028 	if (portid != TC_H_ROOT) {
2029 		u32 qid1 = TC_H_MAJ(portid);
2030 
2031 		if (qid && qid1) {
2032 			/* If both majors are known, they must be identical. */
2033 			if (qid != qid1)
2034 				return -EINVAL;
2035 		} else if (qid1) {
2036 			qid = qid1;
2037 		} else if (qid == 0)
2038 			qid = rtnl_dereference(dev->qdisc)->handle;
2039 
2040 		/* Now qid is genuine qdisc handle consistent
2041 		 * both with parent and child.
2042 		 *
2043 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2044 		 */
2045 		if (portid)
2046 			portid = TC_H_MAKE(qid, portid);
2047 	} else {
2048 		if (qid == 0)
2049 			qid = rtnl_dereference(dev->qdisc)->handle;
2050 	}
2051 
2052 	/* OK. Locate qdisc */
2053 	q = qdisc_lookup(dev, qid);
2054 	if (!q)
2055 		return -ENOENT;
2056 
2057 	/* An check that it supports classes */
2058 	cops = q->ops->cl_ops;
2059 	if (cops == NULL)
2060 		return -EINVAL;
2061 
2062 	/* Now try to get class */
2063 	if (clid == 0) {
2064 		if (portid == TC_H_ROOT)
2065 			clid = qid;
2066 	} else
2067 		clid = TC_H_MAKE(qid, clid);
2068 
2069 	if (clid)
2070 		cl = cops->find(q, clid);
2071 
2072 	if (cl == 0) {
2073 		err = -ENOENT;
2074 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2075 		    !(n->nlmsg_flags & NLM_F_CREATE))
2076 			goto out;
2077 	} else {
2078 		switch (n->nlmsg_type) {
2079 		case RTM_NEWTCLASS:
2080 			err = -EEXIST;
2081 			if (n->nlmsg_flags & NLM_F_EXCL)
2082 				goto out;
2083 			break;
2084 		case RTM_DELTCLASS:
2085 			err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2086 			/* Unbind the class with flilters with 0 */
2087 			tc_bind_tclass(q, portid, clid, 0);
2088 			goto out;
2089 		case RTM_GETTCLASS:
2090 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
2091 			goto out;
2092 		default:
2093 			err = -EINVAL;
2094 			goto out;
2095 		}
2096 	}
2097 
2098 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2099 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2100 		return -EOPNOTSUPP;
2101 	}
2102 
2103 	new_cl = cl;
2104 	err = -EOPNOTSUPP;
2105 	if (cops->change)
2106 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2107 	if (err == 0) {
2108 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
2109 		/* We just create a new class, need to do reverse binding. */
2110 		if (cl != new_cl)
2111 			tc_bind_tclass(q, portid, clid, new_cl);
2112 	}
2113 out:
2114 	return err;
2115 }
2116 
2117 struct qdisc_dump_args {
2118 	struct qdisc_walker	w;
2119 	struct sk_buff		*skb;
2120 	struct netlink_callback	*cb;
2121 };
2122 
2123 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2124 			    struct qdisc_walker *arg)
2125 {
2126 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2127 
2128 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2129 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2130 			      RTM_NEWTCLASS);
2131 }
2132 
2133 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2134 				struct tcmsg *tcm, struct netlink_callback *cb,
2135 				int *t_p, int s_t)
2136 {
2137 	struct qdisc_dump_args arg;
2138 
2139 	if (tc_qdisc_dump_ignore(q, false) ||
2140 	    *t_p < s_t || !q->ops->cl_ops ||
2141 	    (tcm->tcm_parent &&
2142 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2143 		(*t_p)++;
2144 		return 0;
2145 	}
2146 	if (*t_p > s_t)
2147 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2148 	arg.w.fn = qdisc_class_dump;
2149 	arg.skb = skb;
2150 	arg.cb = cb;
2151 	arg.w.stop  = 0;
2152 	arg.w.skip = cb->args[1];
2153 	arg.w.count = 0;
2154 	q->ops->cl_ops->walk(q, &arg.w);
2155 	cb->args[1] = arg.w.count;
2156 	if (arg.w.stop)
2157 		return -1;
2158 	(*t_p)++;
2159 	return 0;
2160 }
2161 
2162 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2163 			       struct tcmsg *tcm, struct netlink_callback *cb,
2164 			       int *t_p, int s_t, bool recur)
2165 {
2166 	struct Qdisc *q;
2167 	int b;
2168 
2169 	if (!root)
2170 		return 0;
2171 
2172 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2173 		return -1;
2174 
2175 	if (!qdisc_dev(root) || !recur)
2176 		return 0;
2177 
2178 	if (tcm->tcm_parent) {
2179 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2180 		if (q && q != root &&
2181 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2182 			return -1;
2183 		return 0;
2184 	}
2185 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2186 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2187 			return -1;
2188 	}
2189 
2190 	return 0;
2191 }
2192 
2193 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2194 {
2195 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2196 	struct net *net = sock_net(skb->sk);
2197 	struct netdev_queue *dev_queue;
2198 	struct net_device *dev;
2199 	int t, s_t;
2200 
2201 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2202 		return 0;
2203 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2204 	if (!dev)
2205 		return 0;
2206 
2207 	s_t = cb->args[0];
2208 	t = 0;
2209 
2210 	if (tc_dump_tclass_root(rtnl_dereference(dev->qdisc),
2211 				skb, tcm, cb, &t, s_t, true) < 0)
2212 		goto done;
2213 
2214 	dev_queue = dev_ingress_queue(dev);
2215 	if (dev_queue &&
2216 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2217 				&t, s_t, false) < 0)
2218 		goto done;
2219 
2220 done:
2221 	cb->args[0] = t;
2222 
2223 	dev_put(dev);
2224 	return skb->len;
2225 }
2226 
2227 #ifdef CONFIG_PROC_FS
2228 static int psched_show(struct seq_file *seq, void *v)
2229 {
2230 	seq_printf(seq, "%08x %08x %08x %08x\n",
2231 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2232 		   1000000,
2233 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2234 
2235 	return 0;
2236 }
2237 
2238 static int __net_init psched_net_init(struct net *net)
2239 {
2240 	struct proc_dir_entry *e;
2241 
2242 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2243 	if (e == NULL)
2244 		return -ENOMEM;
2245 
2246 	return 0;
2247 }
2248 
2249 static void __net_exit psched_net_exit(struct net *net)
2250 {
2251 	remove_proc_entry("psched", net->proc_net);
2252 }
2253 #else
2254 static int __net_init psched_net_init(struct net *net)
2255 {
2256 	return 0;
2257 }
2258 
2259 static void __net_exit psched_net_exit(struct net *net)
2260 {
2261 }
2262 #endif
2263 
2264 static struct pernet_operations psched_net_ops = {
2265 	.init = psched_net_init,
2266 	.exit = psched_net_exit,
2267 };
2268 
2269 static int __init pktsched_init(void)
2270 {
2271 	int err;
2272 
2273 	err = register_pernet_subsys(&psched_net_ops);
2274 	if (err) {
2275 		pr_err("pktsched_init: "
2276 		       "cannot initialize per netns operations\n");
2277 		return err;
2278 	}
2279 
2280 	register_qdisc(&pfifo_fast_ops);
2281 	register_qdisc(&pfifo_qdisc_ops);
2282 	register_qdisc(&bfifo_qdisc_ops);
2283 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2284 	register_qdisc(&mq_qdisc_ops);
2285 	register_qdisc(&noqueue_qdisc_ops);
2286 
2287 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2288 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2289 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2290 		      0);
2291 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2292 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2293 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2294 		      0);
2295 
2296 	return 0;
2297 }
2298 
2299 subsys_initcall(pktsched_init);
2300