xref: /openbmc/linux/net/sched/sch_api.c (revision e330fb14)
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_cpu __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(qdisc_root_sleeping_running(q),
947 				  &d, cpu_bstats, &q->bstats) < 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)
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 					   dev->qdisc, new);
1087 			if (new && !new->ops->attach)
1088 				qdisc_refcount_inc(new);
1089 			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 				  struct Qdisc *p, 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 not found");
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 		seqcount_t *running;
1269 
1270 		err = -EOPNOTSUPP;
1271 		if (sch->flags & TCQ_F_MQROOT) {
1272 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1273 			goto err_out4;
1274 		}
1275 
1276 		if (sch->parent != TC_H_ROOT &&
1277 		    !(sch->flags & TCQ_F_INGRESS) &&
1278 		    (!p || !(p->flags & TCQ_F_MQROOT)))
1279 			running = qdisc_root_sleeping_running(sch);
1280 		else
1281 			running = &sch->running;
1282 
1283 		err = gen_new_estimator(&sch->bstats,
1284 					sch->cpu_bstats,
1285 					&sch->rate_est,
1286 					NULL,
1287 					running,
1288 					tca[TCA_RATE]);
1289 		if (err) {
1290 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1291 			goto err_out4;
1292 		}
1293 	}
1294 
1295 	qdisc_hash_add(sch, false);
1296 	trace_qdisc_create(ops, dev, parent);
1297 
1298 	return sch;
1299 
1300 err_out5:
1301 	/* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1302 	if (ops->destroy)
1303 		ops->destroy(sch);
1304 err_out3:
1305 	dev_put(dev);
1306 	qdisc_free(sch);
1307 err_out2:
1308 	module_put(ops->owner);
1309 err_out:
1310 	*errp = err;
1311 	return NULL;
1312 
1313 err_out4:
1314 	/*
1315 	 * Any broken qdiscs that would require a ops->reset() here?
1316 	 * The qdisc was never in action so it shouldn't be necessary.
1317 	 */
1318 	qdisc_put_stab(rtnl_dereference(sch->stab));
1319 	if (ops->destroy)
1320 		ops->destroy(sch);
1321 	goto err_out3;
1322 }
1323 
1324 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1325 			struct netlink_ext_ack *extack)
1326 {
1327 	struct qdisc_size_table *ostab, *stab = NULL;
1328 	int err = 0;
1329 
1330 	if (tca[TCA_OPTIONS]) {
1331 		if (!sch->ops->change) {
1332 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1333 			return -EINVAL;
1334 		}
1335 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1336 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1337 			return -EOPNOTSUPP;
1338 		}
1339 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1340 		if (err)
1341 			return err;
1342 	}
1343 
1344 	if (tca[TCA_STAB]) {
1345 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1346 		if (IS_ERR(stab))
1347 			return PTR_ERR(stab);
1348 	}
1349 
1350 	ostab = rtnl_dereference(sch->stab);
1351 	rcu_assign_pointer(sch->stab, stab);
1352 	qdisc_put_stab(ostab);
1353 
1354 	if (tca[TCA_RATE]) {
1355 		/* NB: ignores errors from replace_estimator
1356 		   because change can't be undone. */
1357 		if (sch->flags & TCQ_F_MQROOT)
1358 			goto out;
1359 		gen_replace_estimator(&sch->bstats,
1360 				      sch->cpu_bstats,
1361 				      &sch->rate_est,
1362 				      NULL,
1363 				      qdisc_root_sleeping_running(sch),
1364 				      tca[TCA_RATE]);
1365 	}
1366 out:
1367 	return 0;
1368 }
1369 
1370 struct check_loop_arg {
1371 	struct qdisc_walker	w;
1372 	struct Qdisc		*p;
1373 	int			depth;
1374 };
1375 
1376 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1377 			 struct qdisc_walker *w);
1378 
1379 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1380 {
1381 	struct check_loop_arg	arg;
1382 
1383 	if (q->ops->cl_ops == NULL)
1384 		return 0;
1385 
1386 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1387 	arg.w.fn = check_loop_fn;
1388 	arg.depth = depth;
1389 	arg.p = p;
1390 	q->ops->cl_ops->walk(q, &arg.w);
1391 	return arg.w.stop ? -ELOOP : 0;
1392 }
1393 
1394 static int
1395 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1396 {
1397 	struct Qdisc *leaf;
1398 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1399 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1400 
1401 	leaf = cops->leaf(q, cl);
1402 	if (leaf) {
1403 		if (leaf == arg->p || arg->depth > 7)
1404 			return -ELOOP;
1405 		return check_loop(leaf, arg->p, arg->depth + 1);
1406 	}
1407 	return 0;
1408 }
1409 
1410 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1411 	[TCA_KIND]		= { .type = NLA_STRING },
1412 	[TCA_RATE]		= { .type = NLA_BINARY,
1413 				    .len = sizeof(struct tc_estimator) },
1414 	[TCA_STAB]		= { .type = NLA_NESTED },
1415 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1416 	[TCA_CHAIN]		= { .type = NLA_U32 },
1417 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1418 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1419 };
1420 
1421 /*
1422  * Delete/get qdisc.
1423  */
1424 
1425 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1426 			struct netlink_ext_ack *extack)
1427 {
1428 	struct net *net = sock_net(skb->sk);
1429 	struct tcmsg *tcm = nlmsg_data(n);
1430 	struct nlattr *tca[TCA_MAX + 1];
1431 	struct net_device *dev;
1432 	u32 clid;
1433 	struct Qdisc *q = NULL;
1434 	struct Qdisc *p = NULL;
1435 	int err;
1436 
1437 	if ((n->nlmsg_type != RTM_GETQDISC) &&
1438 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1439 		return -EPERM;
1440 
1441 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1442 				     rtm_tca_policy, extack);
1443 	if (err < 0)
1444 		return err;
1445 
1446 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1447 	if (!dev)
1448 		return -ENODEV;
1449 
1450 	clid = tcm->tcm_parent;
1451 	if (clid) {
1452 		if (clid != TC_H_ROOT) {
1453 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1454 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1455 				if (!p) {
1456 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1457 					return -ENOENT;
1458 				}
1459 				q = qdisc_leaf(p, clid);
1460 			} else if (dev_ingress_queue(dev)) {
1461 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1462 			}
1463 		} else {
1464 			q = dev->qdisc;
1465 		}
1466 		if (!q) {
1467 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1468 			return -ENOENT;
1469 		}
1470 
1471 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1472 			NL_SET_ERR_MSG(extack, "Invalid handle");
1473 			return -EINVAL;
1474 		}
1475 	} else {
1476 		q = qdisc_lookup(dev, tcm->tcm_handle);
1477 		if (!q) {
1478 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1479 			return -ENOENT;
1480 		}
1481 	}
1482 
1483 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1484 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1485 		return -EINVAL;
1486 	}
1487 
1488 	if (n->nlmsg_type == RTM_DELQDISC) {
1489 		if (!clid) {
1490 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1491 			return -EINVAL;
1492 		}
1493 		if (q->handle == 0) {
1494 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1495 			return -ENOENT;
1496 		}
1497 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1498 		if (err != 0)
1499 			return err;
1500 	} else {
1501 		qdisc_notify(net, skb, n, clid, NULL, q);
1502 	}
1503 	return 0;
1504 }
1505 
1506 /*
1507  * Create/change qdisc.
1508  */
1509 
1510 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1511 			   struct netlink_ext_ack *extack)
1512 {
1513 	struct net *net = sock_net(skb->sk);
1514 	struct tcmsg *tcm;
1515 	struct nlattr *tca[TCA_MAX + 1];
1516 	struct net_device *dev;
1517 	u32 clid;
1518 	struct Qdisc *q, *p;
1519 	int err;
1520 
1521 	if (!netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
1522 		return -EPERM;
1523 
1524 replay:
1525 	/* Reinit, just in case something touches this. */
1526 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1527 				     rtm_tca_policy, extack);
1528 	if (err < 0)
1529 		return err;
1530 
1531 	tcm = nlmsg_data(n);
1532 	clid = tcm->tcm_parent;
1533 	q = p = NULL;
1534 
1535 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1536 	if (!dev)
1537 		return -ENODEV;
1538 
1539 
1540 	if (clid) {
1541 		if (clid != TC_H_ROOT) {
1542 			if (clid != TC_H_INGRESS) {
1543 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1544 				if (!p) {
1545 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1546 					return -ENOENT;
1547 				}
1548 				q = qdisc_leaf(p, clid);
1549 			} else if (dev_ingress_queue_create(dev)) {
1550 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1551 			}
1552 		} else {
1553 			q = dev->qdisc;
1554 		}
1555 
1556 		/* It may be default qdisc, ignore it */
1557 		if (q && q->handle == 0)
1558 			q = NULL;
1559 
1560 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1561 			if (tcm->tcm_handle) {
1562 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1563 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1564 					return -EEXIST;
1565 				}
1566 				if (TC_H_MIN(tcm->tcm_handle)) {
1567 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1568 					return -EINVAL;
1569 				}
1570 				q = qdisc_lookup(dev, tcm->tcm_handle);
1571 				if (!q)
1572 					goto create_n_graft;
1573 				if (n->nlmsg_flags & NLM_F_EXCL) {
1574 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1575 					return -EEXIST;
1576 				}
1577 				if (tca[TCA_KIND] &&
1578 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1579 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1580 					return -EINVAL;
1581 				}
1582 				if (q == p ||
1583 				    (p && check_loop(q, p, 0))) {
1584 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1585 					return -ELOOP;
1586 				}
1587 				qdisc_refcount_inc(q);
1588 				goto graft;
1589 			} else {
1590 				if (!q)
1591 					goto create_n_graft;
1592 
1593 				/* This magic test requires explanation.
1594 				 *
1595 				 *   We know, that some child q is already
1596 				 *   attached to this parent and have choice:
1597 				 *   either to change it or to create/graft new one.
1598 				 *
1599 				 *   1. We are allowed to create/graft only
1600 				 *   if CREATE and REPLACE flags are set.
1601 				 *
1602 				 *   2. If EXCL is set, requestor wanted to say,
1603 				 *   that qdisc tcm_handle is not expected
1604 				 *   to exist, so that we choose create/graft too.
1605 				 *
1606 				 *   3. The last case is when no flags are set.
1607 				 *   Alas, it is sort of hole in API, we
1608 				 *   cannot decide what to do unambiguously.
1609 				 *   For now we select create/graft, if
1610 				 *   user gave KIND, which does not match existing.
1611 				 */
1612 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1613 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1614 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1615 				     (tca[TCA_KIND] &&
1616 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1617 					goto create_n_graft;
1618 			}
1619 		}
1620 	} else {
1621 		if (!tcm->tcm_handle) {
1622 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1623 			return -EINVAL;
1624 		}
1625 		q = qdisc_lookup(dev, tcm->tcm_handle);
1626 	}
1627 
1628 	/* Change qdisc parameters */
1629 	if (!q) {
1630 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1631 		return -ENOENT;
1632 	}
1633 	if (n->nlmsg_flags & NLM_F_EXCL) {
1634 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1635 		return -EEXIST;
1636 	}
1637 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1638 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1639 		return -EINVAL;
1640 	}
1641 	err = qdisc_change(q, tca, extack);
1642 	if (err == 0)
1643 		qdisc_notify(net, skb, n, clid, NULL, q);
1644 	return err;
1645 
1646 create_n_graft:
1647 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1648 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1649 		return -ENOENT;
1650 	}
1651 	if (clid == TC_H_INGRESS) {
1652 		if (dev_ingress_queue(dev)) {
1653 			q = qdisc_create(dev, dev_ingress_queue(dev), p,
1654 					 tcm->tcm_parent, tcm->tcm_parent,
1655 					 tca, &err, extack);
1656 		} else {
1657 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1658 			err = -ENOENT;
1659 		}
1660 	} else {
1661 		struct netdev_queue *dev_queue;
1662 
1663 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1664 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1665 		else if (p)
1666 			dev_queue = p->dev_queue;
1667 		else
1668 			dev_queue = netdev_get_tx_queue(dev, 0);
1669 
1670 		q = qdisc_create(dev, dev_queue, p,
1671 				 tcm->tcm_parent, tcm->tcm_handle,
1672 				 tca, &err, extack);
1673 	}
1674 	if (q == NULL) {
1675 		if (err == -EAGAIN)
1676 			goto replay;
1677 		return err;
1678 	}
1679 
1680 graft:
1681 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1682 	if (err) {
1683 		if (q)
1684 			qdisc_put(q);
1685 		return err;
1686 	}
1687 
1688 	return 0;
1689 }
1690 
1691 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1692 			      struct netlink_callback *cb,
1693 			      int *q_idx_p, int s_q_idx, bool recur,
1694 			      bool dump_invisible)
1695 {
1696 	int ret = 0, q_idx = *q_idx_p;
1697 	struct Qdisc *q;
1698 	int b;
1699 
1700 	if (!root)
1701 		return 0;
1702 
1703 	q = root;
1704 	if (q_idx < s_q_idx) {
1705 		q_idx++;
1706 	} else {
1707 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1708 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1709 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1710 				  RTM_NEWQDISC) <= 0)
1711 			goto done;
1712 		q_idx++;
1713 	}
1714 
1715 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1716 	 * itself has already been dumped.
1717 	 *
1718 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1719 	 * qdisc hashtable, we don't want to hit it again
1720 	 */
1721 	if (!qdisc_dev(root) || !recur)
1722 		goto out;
1723 
1724 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1725 		if (q_idx < s_q_idx) {
1726 			q_idx++;
1727 			continue;
1728 		}
1729 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1730 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1731 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1732 				  RTM_NEWQDISC) <= 0)
1733 			goto done;
1734 		q_idx++;
1735 	}
1736 
1737 out:
1738 	*q_idx_p = q_idx;
1739 	return ret;
1740 done:
1741 	ret = -1;
1742 	goto out;
1743 }
1744 
1745 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1746 {
1747 	struct net *net = sock_net(skb->sk);
1748 	int idx, q_idx;
1749 	int s_idx, s_q_idx;
1750 	struct net_device *dev;
1751 	const struct nlmsghdr *nlh = cb->nlh;
1752 	struct nlattr *tca[TCA_MAX + 1];
1753 	int err;
1754 
1755 	s_idx = cb->args[0];
1756 	s_q_idx = q_idx = cb->args[1];
1757 
1758 	idx = 0;
1759 	ASSERT_RTNL();
1760 
1761 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1762 				     rtm_tca_policy, cb->extack);
1763 	if (err < 0)
1764 		return err;
1765 
1766 	for_each_netdev(net, dev) {
1767 		struct netdev_queue *dev_queue;
1768 
1769 		if (idx < s_idx)
1770 			goto cont;
1771 		if (idx > s_idx)
1772 			s_q_idx = 0;
1773 		q_idx = 0;
1774 
1775 		if (tc_dump_qdisc_root(dev->qdisc, skb, cb, &q_idx, s_q_idx,
1776 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1777 			goto done;
1778 
1779 		dev_queue = dev_ingress_queue(dev);
1780 		if (dev_queue &&
1781 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1782 				       &q_idx, s_q_idx, false,
1783 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1784 			goto done;
1785 
1786 cont:
1787 		idx++;
1788 	}
1789 
1790 done:
1791 	cb->args[0] = idx;
1792 	cb->args[1] = q_idx;
1793 
1794 	return skb->len;
1795 }
1796 
1797 
1798 
1799 /************************************************
1800  *	Traffic classes manipulation.		*
1801  ************************************************/
1802 
1803 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1804 			  unsigned long cl,
1805 			  u32 portid, u32 seq, u16 flags, int event)
1806 {
1807 	struct tcmsg *tcm;
1808 	struct nlmsghdr  *nlh;
1809 	unsigned char *b = skb_tail_pointer(skb);
1810 	struct gnet_dump d;
1811 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1812 
1813 	cond_resched();
1814 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1815 	if (!nlh)
1816 		goto out_nlmsg_trim;
1817 	tcm = nlmsg_data(nlh);
1818 	tcm->tcm_family = AF_UNSPEC;
1819 	tcm->tcm__pad1 = 0;
1820 	tcm->tcm__pad2 = 0;
1821 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1822 	tcm->tcm_parent = q->handle;
1823 	tcm->tcm_handle = q->handle;
1824 	tcm->tcm_info = 0;
1825 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1826 		goto nla_put_failure;
1827 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1828 		goto nla_put_failure;
1829 
1830 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1831 					 NULL, &d, TCA_PAD) < 0)
1832 		goto nla_put_failure;
1833 
1834 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1835 		goto nla_put_failure;
1836 
1837 	if (gnet_stats_finish_copy(&d) < 0)
1838 		goto nla_put_failure;
1839 
1840 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1841 	return skb->len;
1842 
1843 out_nlmsg_trim:
1844 nla_put_failure:
1845 	nlmsg_trim(skb, b);
1846 	return -1;
1847 }
1848 
1849 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1850 			 struct nlmsghdr *n, struct Qdisc *q,
1851 			 unsigned long cl, int event)
1852 {
1853 	struct sk_buff *skb;
1854 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1855 
1856 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1857 	if (!skb)
1858 		return -ENOBUFS;
1859 
1860 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1861 		kfree_skb(skb);
1862 		return -EINVAL;
1863 	}
1864 
1865 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1866 			      n->nlmsg_flags & NLM_F_ECHO);
1867 }
1868 
1869 static int tclass_del_notify(struct net *net,
1870 			     const struct Qdisc_class_ops *cops,
1871 			     struct sk_buff *oskb, struct nlmsghdr *n,
1872 			     struct Qdisc *q, unsigned long cl,
1873 			     struct netlink_ext_ack *extack)
1874 {
1875 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1876 	struct sk_buff *skb;
1877 	int err = 0;
1878 
1879 	if (!cops->delete)
1880 		return -EOPNOTSUPP;
1881 
1882 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1883 	if (!skb)
1884 		return -ENOBUFS;
1885 
1886 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1887 			   RTM_DELTCLASS) < 0) {
1888 		kfree_skb(skb);
1889 		return -EINVAL;
1890 	}
1891 
1892 	err = cops->delete(q, cl, extack);
1893 	if (err) {
1894 		kfree_skb(skb);
1895 		return err;
1896 	}
1897 
1898 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1899 			     n->nlmsg_flags & NLM_F_ECHO);
1900 	return err;
1901 }
1902 
1903 #ifdef CONFIG_NET_CLS
1904 
1905 struct tcf_bind_args {
1906 	struct tcf_walker w;
1907 	unsigned long base;
1908 	unsigned long cl;
1909 	u32 classid;
1910 };
1911 
1912 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1913 {
1914 	struct tcf_bind_args *a = (void *)arg;
1915 
1916 	if (tp->ops->bind_class) {
1917 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1918 
1919 		sch_tree_lock(q);
1920 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1921 		sch_tree_unlock(q);
1922 	}
1923 	return 0;
1924 }
1925 
1926 struct tc_bind_class_args {
1927 	struct qdisc_walker w;
1928 	unsigned long new_cl;
1929 	u32 portid;
1930 	u32 clid;
1931 };
1932 
1933 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1934 				struct qdisc_walker *w)
1935 {
1936 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1937 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1938 	struct tcf_block *block;
1939 	struct tcf_chain *chain;
1940 
1941 	block = cops->tcf_block(q, cl, NULL);
1942 	if (!block)
1943 		return 0;
1944 	for (chain = tcf_get_next_chain(block, NULL);
1945 	     chain;
1946 	     chain = tcf_get_next_chain(block, chain)) {
1947 		struct tcf_proto *tp;
1948 
1949 		for (tp = tcf_get_next_proto(chain, NULL);
1950 		     tp; tp = tcf_get_next_proto(chain, tp)) {
1951 			struct tcf_bind_args arg = {};
1952 
1953 			arg.w.fn = tcf_node_bind;
1954 			arg.classid = a->clid;
1955 			arg.base = cl;
1956 			arg.cl = a->new_cl;
1957 			tp->ops->walk(tp, &arg.w, true);
1958 		}
1959 	}
1960 
1961 	return 0;
1962 }
1963 
1964 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1965 			   unsigned long new_cl)
1966 {
1967 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1968 	struct tc_bind_class_args args = {};
1969 
1970 	if (!cops->tcf_block)
1971 		return;
1972 	args.portid = portid;
1973 	args.clid = clid;
1974 	args.new_cl = new_cl;
1975 	args.w.fn = tc_bind_class_walker;
1976 	q->ops->cl_ops->walk(q, &args.w);
1977 }
1978 
1979 #else
1980 
1981 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1982 			   unsigned long new_cl)
1983 {
1984 }
1985 
1986 #endif
1987 
1988 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1989 			 struct netlink_ext_ack *extack)
1990 {
1991 	struct net *net = sock_net(skb->sk);
1992 	struct tcmsg *tcm = nlmsg_data(n);
1993 	struct nlattr *tca[TCA_MAX + 1];
1994 	struct net_device *dev;
1995 	struct Qdisc *q = NULL;
1996 	const struct Qdisc_class_ops *cops;
1997 	unsigned long cl = 0;
1998 	unsigned long new_cl;
1999 	u32 portid;
2000 	u32 clid;
2001 	u32 qid;
2002 	int err;
2003 
2004 	if ((n->nlmsg_type != RTM_GETTCLASS) &&
2005 	    !netlink_ns_capable(skb, net->user_ns, CAP_NET_ADMIN))
2006 		return -EPERM;
2007 
2008 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2009 				     rtm_tca_policy, extack);
2010 	if (err < 0)
2011 		return err;
2012 
2013 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2014 	if (!dev)
2015 		return -ENODEV;
2016 
2017 	/*
2018 	   parent == TC_H_UNSPEC - unspecified parent.
2019 	   parent == TC_H_ROOT   - class is root, which has no parent.
2020 	   parent == X:0	 - parent is root class.
2021 	   parent == X:Y	 - parent is a node in hierarchy.
2022 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2023 
2024 	   handle == 0:0	 - generate handle from kernel pool.
2025 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2026 	   handle == X:Y	 - clear.
2027 	   handle == X:0	 - root class.
2028 	 */
2029 
2030 	/* Step 1. Determine qdisc handle X:0 */
2031 
2032 	portid = tcm->tcm_parent;
2033 	clid = tcm->tcm_handle;
2034 	qid = TC_H_MAJ(clid);
2035 
2036 	if (portid != TC_H_ROOT) {
2037 		u32 qid1 = TC_H_MAJ(portid);
2038 
2039 		if (qid && qid1) {
2040 			/* If both majors are known, they must be identical. */
2041 			if (qid != qid1)
2042 				return -EINVAL;
2043 		} else if (qid1) {
2044 			qid = qid1;
2045 		} else if (qid == 0)
2046 			qid = dev->qdisc->handle;
2047 
2048 		/* Now qid is genuine qdisc handle consistent
2049 		 * both with parent and child.
2050 		 *
2051 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2052 		 */
2053 		if (portid)
2054 			portid = TC_H_MAKE(qid, portid);
2055 	} else {
2056 		if (qid == 0)
2057 			qid = dev->qdisc->handle;
2058 	}
2059 
2060 	/* OK. Locate qdisc */
2061 	q = qdisc_lookup(dev, qid);
2062 	if (!q)
2063 		return -ENOENT;
2064 
2065 	/* An check that it supports classes */
2066 	cops = q->ops->cl_ops;
2067 	if (cops == NULL)
2068 		return -EINVAL;
2069 
2070 	/* Now try to get class */
2071 	if (clid == 0) {
2072 		if (portid == TC_H_ROOT)
2073 			clid = qid;
2074 	} else
2075 		clid = TC_H_MAKE(qid, clid);
2076 
2077 	if (clid)
2078 		cl = cops->find(q, clid);
2079 
2080 	if (cl == 0) {
2081 		err = -ENOENT;
2082 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2083 		    !(n->nlmsg_flags & NLM_F_CREATE))
2084 			goto out;
2085 	} else {
2086 		switch (n->nlmsg_type) {
2087 		case RTM_NEWTCLASS:
2088 			err = -EEXIST;
2089 			if (n->nlmsg_flags & NLM_F_EXCL)
2090 				goto out;
2091 			break;
2092 		case RTM_DELTCLASS:
2093 			err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2094 			/* Unbind the class with flilters with 0 */
2095 			tc_bind_tclass(q, portid, clid, 0);
2096 			goto out;
2097 		case RTM_GETTCLASS:
2098 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS);
2099 			goto out;
2100 		default:
2101 			err = -EINVAL;
2102 			goto out;
2103 		}
2104 	}
2105 
2106 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2107 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2108 		return -EOPNOTSUPP;
2109 	}
2110 
2111 	new_cl = cl;
2112 	err = -EOPNOTSUPP;
2113 	if (cops->change)
2114 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2115 	if (err == 0) {
2116 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS);
2117 		/* We just create a new class, need to do reverse binding. */
2118 		if (cl != new_cl)
2119 			tc_bind_tclass(q, portid, clid, new_cl);
2120 	}
2121 out:
2122 	return err;
2123 }
2124 
2125 struct qdisc_dump_args {
2126 	struct qdisc_walker	w;
2127 	struct sk_buff		*skb;
2128 	struct netlink_callback	*cb;
2129 };
2130 
2131 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2132 			    struct qdisc_walker *arg)
2133 {
2134 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2135 
2136 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2137 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2138 			      RTM_NEWTCLASS);
2139 }
2140 
2141 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2142 				struct tcmsg *tcm, struct netlink_callback *cb,
2143 				int *t_p, int s_t)
2144 {
2145 	struct qdisc_dump_args arg;
2146 
2147 	if (tc_qdisc_dump_ignore(q, false) ||
2148 	    *t_p < s_t || !q->ops->cl_ops ||
2149 	    (tcm->tcm_parent &&
2150 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2151 		(*t_p)++;
2152 		return 0;
2153 	}
2154 	if (*t_p > s_t)
2155 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2156 	arg.w.fn = qdisc_class_dump;
2157 	arg.skb = skb;
2158 	arg.cb = cb;
2159 	arg.w.stop  = 0;
2160 	arg.w.skip = cb->args[1];
2161 	arg.w.count = 0;
2162 	q->ops->cl_ops->walk(q, &arg.w);
2163 	cb->args[1] = arg.w.count;
2164 	if (arg.w.stop)
2165 		return -1;
2166 	(*t_p)++;
2167 	return 0;
2168 }
2169 
2170 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2171 			       struct tcmsg *tcm, struct netlink_callback *cb,
2172 			       int *t_p, int s_t, bool recur)
2173 {
2174 	struct Qdisc *q;
2175 	int b;
2176 
2177 	if (!root)
2178 		return 0;
2179 
2180 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2181 		return -1;
2182 
2183 	if (!qdisc_dev(root) || !recur)
2184 		return 0;
2185 
2186 	if (tcm->tcm_parent) {
2187 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2188 		if (q && q != root &&
2189 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2190 			return -1;
2191 		return 0;
2192 	}
2193 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2194 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2195 			return -1;
2196 	}
2197 
2198 	return 0;
2199 }
2200 
2201 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2202 {
2203 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2204 	struct net *net = sock_net(skb->sk);
2205 	struct netdev_queue *dev_queue;
2206 	struct net_device *dev;
2207 	int t, s_t;
2208 
2209 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2210 		return 0;
2211 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2212 	if (!dev)
2213 		return 0;
2214 
2215 	s_t = cb->args[0];
2216 	t = 0;
2217 
2218 	if (tc_dump_tclass_root(dev->qdisc, skb, tcm, cb, &t, s_t, true) < 0)
2219 		goto done;
2220 
2221 	dev_queue = dev_ingress_queue(dev);
2222 	if (dev_queue &&
2223 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2224 				&t, s_t, false) < 0)
2225 		goto done;
2226 
2227 done:
2228 	cb->args[0] = t;
2229 
2230 	dev_put(dev);
2231 	return skb->len;
2232 }
2233 
2234 #ifdef CONFIG_PROC_FS
2235 static int psched_show(struct seq_file *seq, void *v)
2236 {
2237 	seq_printf(seq, "%08x %08x %08x %08x\n",
2238 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2239 		   1000000,
2240 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2241 
2242 	return 0;
2243 }
2244 
2245 static int __net_init psched_net_init(struct net *net)
2246 {
2247 	struct proc_dir_entry *e;
2248 
2249 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2250 	if (e == NULL)
2251 		return -ENOMEM;
2252 
2253 	return 0;
2254 }
2255 
2256 static void __net_exit psched_net_exit(struct net *net)
2257 {
2258 	remove_proc_entry("psched", net->proc_net);
2259 }
2260 #else
2261 static int __net_init psched_net_init(struct net *net)
2262 {
2263 	return 0;
2264 }
2265 
2266 static void __net_exit psched_net_exit(struct net *net)
2267 {
2268 }
2269 #endif
2270 
2271 static struct pernet_operations psched_net_ops = {
2272 	.init = psched_net_init,
2273 	.exit = psched_net_exit,
2274 };
2275 
2276 static int __init pktsched_init(void)
2277 {
2278 	int err;
2279 
2280 	err = register_pernet_subsys(&psched_net_ops);
2281 	if (err) {
2282 		pr_err("pktsched_init: "
2283 		       "cannot initialize per netns operations\n");
2284 		return err;
2285 	}
2286 
2287 	register_qdisc(&pfifo_fast_ops);
2288 	register_qdisc(&pfifo_qdisc_ops);
2289 	register_qdisc(&bfifo_qdisc_ops);
2290 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2291 	register_qdisc(&mq_qdisc_ops);
2292 	register_qdisc(&noqueue_qdisc_ops);
2293 
2294 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2295 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2296 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2297 		      0);
2298 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2299 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2300 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2301 		      0);
2302 
2303 	return 0;
2304 }
2305 
2306 subsys_initcall(pktsched_init);
2307