xref: /openbmc/linux/net/sched/sch_api.c (revision c699ce1a)
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 #include <net/tc_wrapper.h>
35 
36 #include <trace/events/qdisc.h>
37 
38 /*
39 
40    Short review.
41    -------------
42 
43    This file consists of two interrelated parts:
44 
45    1. queueing disciplines manager frontend.
46    2. traffic classes manager frontend.
47 
48    Generally, queueing discipline ("qdisc") is a black box,
49    which is able to enqueue packets and to dequeue them (when
50    device is ready to send something) in order and at times
51    determined by algorithm hidden in it.
52 
53    qdisc's are divided to two categories:
54    - "queues", which have no internal structure visible from outside.
55    - "schedulers", which split all the packets to "traffic classes",
56      using "packet classifiers" (look at cls_api.c)
57 
58    In turn, classes may have child qdiscs (as rule, queues)
59    attached to them etc. etc. etc.
60 
61    The goal of the routines in this file is to translate
62    information supplied by user in the form of handles
63    to more intelligible for kernel form, to make some sanity
64    checks and part of work, which is common to all qdiscs
65    and to provide rtnetlink notifications.
66 
67    All real intelligent work is done inside qdisc modules.
68 
69 
70 
71    Every discipline has two major routines: enqueue and dequeue.
72 
73    ---dequeue
74 
75    dequeue usually returns a skb to send. It is allowed to return NULL,
76    but it does not mean that queue is empty, it just means that
77    discipline does not want to send anything this time.
78    Queue is really empty if q->q.qlen == 0.
79    For complicated disciplines with multiple queues q->q is not
80    real packet queue, but however q->q.qlen must be valid.
81 
82    ---enqueue
83 
84    enqueue returns 0, if packet was enqueued successfully.
85    If packet (this one or another one) was dropped, it returns
86    not zero error code.
87    NET_XMIT_DROP 	- this packet dropped
88      Expected action: do not backoff, but wait until queue will clear.
89    NET_XMIT_CN	 	- probably this packet enqueued, but another one dropped.
90      Expected action: backoff or ignore
91 
92    Auxiliary routines:
93 
94    ---peek
95 
96    like dequeue but without removing a packet from the queue
97 
98    ---reset
99 
100    returns qdisc to initial state: purge all buffers, clear all
101    timers, counters (except for statistics) etc.
102 
103    ---init
104 
105    initializes newly created qdisc.
106 
107    ---destroy
108 
109    destroys resources allocated by init and during lifetime of qdisc.
110 
111    ---change
112 
113    changes qdisc parameters.
114  */
115 
116 /* Protects list of registered TC modules. It is pure SMP lock. */
117 static DEFINE_RWLOCK(qdisc_mod_lock);
118 
119 
120 /************************************************
121  *	Queueing disciplines manipulation.	*
122  ************************************************/
123 
124 
125 /* The list of all installed queueing disciplines. */
126 
127 static struct Qdisc_ops *qdisc_base;
128 
129 /* Register/unregister queueing discipline */
130 
131 int register_qdisc(struct Qdisc_ops *qops)
132 {
133 	struct Qdisc_ops *q, **qp;
134 	int rc = -EEXIST;
135 
136 	write_lock(&qdisc_mod_lock);
137 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
138 		if (!strcmp(qops->id, q->id))
139 			goto out;
140 
141 	if (qops->enqueue == NULL)
142 		qops->enqueue = noop_qdisc_ops.enqueue;
143 	if (qops->peek == NULL) {
144 		if (qops->dequeue == NULL)
145 			qops->peek = noop_qdisc_ops.peek;
146 		else
147 			goto out_einval;
148 	}
149 	if (qops->dequeue == NULL)
150 		qops->dequeue = noop_qdisc_ops.dequeue;
151 
152 	if (qops->cl_ops) {
153 		const struct Qdisc_class_ops *cops = qops->cl_ops;
154 
155 		if (!(cops->find && cops->walk && cops->leaf))
156 			goto out_einval;
157 
158 		if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
159 			goto out_einval;
160 	}
161 
162 	qops->next = NULL;
163 	*qp = qops;
164 	rc = 0;
165 out:
166 	write_unlock(&qdisc_mod_lock);
167 	return rc;
168 
169 out_einval:
170 	rc = -EINVAL;
171 	goto out;
172 }
173 EXPORT_SYMBOL(register_qdisc);
174 
175 void unregister_qdisc(struct Qdisc_ops *qops)
176 {
177 	struct Qdisc_ops *q, **qp;
178 	int err = -ENOENT;
179 
180 	write_lock(&qdisc_mod_lock);
181 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
182 		if (q == qops)
183 			break;
184 	if (q) {
185 		*qp = q->next;
186 		q->next = NULL;
187 		err = 0;
188 	}
189 	write_unlock(&qdisc_mod_lock);
190 
191 	WARN(err, "unregister qdisc(%s) failed\n", qops->id);
192 }
193 EXPORT_SYMBOL(unregister_qdisc);
194 
195 /* Get default qdisc if not otherwise specified */
196 void qdisc_get_default(char *name, size_t len)
197 {
198 	read_lock(&qdisc_mod_lock);
199 	strscpy(name, default_qdisc_ops->id, len);
200 	read_unlock(&qdisc_mod_lock);
201 }
202 
203 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
204 {
205 	struct Qdisc_ops *q = NULL;
206 
207 	for (q = qdisc_base; q; q = q->next) {
208 		if (!strcmp(name, q->id)) {
209 			if (!try_module_get(q->owner))
210 				q = NULL;
211 			break;
212 		}
213 	}
214 
215 	return q;
216 }
217 
218 /* Set new default qdisc to use */
219 int qdisc_set_default(const char *name)
220 {
221 	const struct Qdisc_ops *ops;
222 
223 	if (!capable(CAP_NET_ADMIN))
224 		return -EPERM;
225 
226 	write_lock(&qdisc_mod_lock);
227 	ops = qdisc_lookup_default(name);
228 	if (!ops) {
229 		/* Not found, drop lock and try to load module */
230 		write_unlock(&qdisc_mod_lock);
231 		request_module("sch_%s", name);
232 		write_lock(&qdisc_mod_lock);
233 
234 		ops = qdisc_lookup_default(name);
235 	}
236 
237 	if (ops) {
238 		/* Set new default */
239 		module_put(default_qdisc_ops->owner);
240 		default_qdisc_ops = ops;
241 	}
242 	write_unlock(&qdisc_mod_lock);
243 
244 	return ops ? 0 : -ENOENT;
245 }
246 
247 #ifdef CONFIG_NET_SCH_DEFAULT
248 /* Set default value from kernel config */
249 static int __init sch_default_qdisc(void)
250 {
251 	return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
252 }
253 late_initcall(sch_default_qdisc);
254 #endif
255 
256 /* We know handle. Find qdisc among all qdisc's attached to device
257  * (root qdisc, all its children, children of children etc.)
258  * Note: caller either uses rtnl or rcu_read_lock()
259  */
260 
261 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
262 {
263 	struct Qdisc *q;
264 
265 	if (!qdisc_dev(root))
266 		return (root->handle == handle ? root : NULL);
267 
268 	if (!(root->flags & TCQ_F_BUILTIN) &&
269 	    root->handle == handle)
270 		return root;
271 
272 	hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle,
273 				   lockdep_rtnl_is_held()) {
274 		if (q->handle == handle)
275 			return q;
276 	}
277 	return NULL;
278 }
279 
280 void qdisc_hash_add(struct Qdisc *q, bool invisible)
281 {
282 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
283 		ASSERT_RTNL();
284 		hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
285 		if (invisible)
286 			q->flags |= TCQ_F_INVISIBLE;
287 	}
288 }
289 EXPORT_SYMBOL(qdisc_hash_add);
290 
291 void qdisc_hash_del(struct Qdisc *q)
292 {
293 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
294 		ASSERT_RTNL();
295 		hash_del_rcu(&q->hash);
296 	}
297 }
298 EXPORT_SYMBOL(qdisc_hash_del);
299 
300 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
301 {
302 	struct Qdisc *q;
303 
304 	if (!handle)
305 		return NULL;
306 	q = qdisc_match_from_root(rtnl_dereference(dev->qdisc), handle);
307 	if (q)
308 		goto out;
309 
310 	if (dev_ingress_queue(dev))
311 		q = qdisc_match_from_root(
312 			dev_ingress_queue(dev)->qdisc_sleeping,
313 			handle);
314 out:
315 	return q;
316 }
317 
318 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
319 {
320 	struct netdev_queue *nq;
321 	struct Qdisc *q;
322 
323 	if (!handle)
324 		return NULL;
325 	q = qdisc_match_from_root(rcu_dereference(dev->qdisc), handle);
326 	if (q)
327 		goto out;
328 
329 	nq = dev_ingress_queue_rcu(dev);
330 	if (nq)
331 		q = qdisc_match_from_root(nq->qdisc_sleeping, handle);
332 out:
333 	return q;
334 }
335 
336 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
337 {
338 	unsigned long cl;
339 	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
340 
341 	if (cops == NULL)
342 		return NULL;
343 	cl = cops->find(p, classid);
344 
345 	if (cl == 0)
346 		return NULL;
347 	return cops->leaf(p, cl);
348 }
349 
350 /* Find queueing discipline by name */
351 
352 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
353 {
354 	struct Qdisc_ops *q = NULL;
355 
356 	if (kind) {
357 		read_lock(&qdisc_mod_lock);
358 		for (q = qdisc_base; q; q = q->next) {
359 			if (nla_strcmp(kind, q->id) == 0) {
360 				if (!try_module_get(q->owner))
361 					q = NULL;
362 				break;
363 			}
364 		}
365 		read_unlock(&qdisc_mod_lock);
366 	}
367 	return q;
368 }
369 
370 /* The linklayer setting were not transferred from iproute2, in older
371  * versions, and the rate tables lookup systems have been dropped in
372  * the kernel. To keep backward compatible with older iproute2 tc
373  * utils, we detect the linklayer setting by detecting if the rate
374  * table were modified.
375  *
376  * For linklayer ATM table entries, the rate table will be aligned to
377  * 48 bytes, thus some table entries will contain the same value.  The
378  * mpu (min packet unit) is also encoded into the old rate table, thus
379  * starting from the mpu, we find low and high table entries for
380  * mapping this cell.  If these entries contain the same value, when
381  * the rate tables have been modified for linklayer ATM.
382  *
383  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
384  * and then roundup to the next cell, calc the table entry one below,
385  * and compare.
386  */
387 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
388 {
389 	int low       = roundup(r->mpu, 48);
390 	int high      = roundup(low+1, 48);
391 	int cell_low  = low >> r->cell_log;
392 	int cell_high = (high >> r->cell_log) - 1;
393 
394 	/* rtab is too inaccurate at rates > 100Mbit/s */
395 	if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
396 		pr_debug("TC linklayer: Giving up ATM detection\n");
397 		return TC_LINKLAYER_ETHERNET;
398 	}
399 
400 	if ((cell_high > cell_low) && (cell_high < 256)
401 	    && (rtab[cell_low] == rtab[cell_high])) {
402 		pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
403 			 cell_low, cell_high, rtab[cell_high]);
404 		return TC_LINKLAYER_ATM;
405 	}
406 	return TC_LINKLAYER_ETHERNET;
407 }
408 
409 static struct qdisc_rate_table *qdisc_rtab_list;
410 
411 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
412 					struct nlattr *tab,
413 					struct netlink_ext_ack *extack)
414 {
415 	struct qdisc_rate_table *rtab;
416 
417 	if (tab == NULL || r->rate == 0 ||
418 	    r->cell_log == 0 || r->cell_log >= 32 ||
419 	    nla_len(tab) != TC_RTAB_SIZE) {
420 		NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
421 		return NULL;
422 	}
423 
424 	for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
425 		if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
426 		    !memcmp(&rtab->data, nla_data(tab), 1024)) {
427 			rtab->refcnt++;
428 			return rtab;
429 		}
430 	}
431 
432 	rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
433 	if (rtab) {
434 		rtab->rate = *r;
435 		rtab->refcnt = 1;
436 		memcpy(rtab->data, nla_data(tab), 1024);
437 		if (r->linklayer == TC_LINKLAYER_UNAWARE)
438 			r->linklayer = __detect_linklayer(r, rtab->data);
439 		rtab->next = qdisc_rtab_list;
440 		qdisc_rtab_list = rtab;
441 	} else {
442 		NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
443 	}
444 	return rtab;
445 }
446 EXPORT_SYMBOL(qdisc_get_rtab);
447 
448 void qdisc_put_rtab(struct qdisc_rate_table *tab)
449 {
450 	struct qdisc_rate_table *rtab, **rtabp;
451 
452 	if (!tab || --tab->refcnt)
453 		return;
454 
455 	for (rtabp = &qdisc_rtab_list;
456 	     (rtab = *rtabp) != NULL;
457 	     rtabp = &rtab->next) {
458 		if (rtab == tab) {
459 			*rtabp = rtab->next;
460 			kfree(rtab);
461 			return;
462 		}
463 	}
464 }
465 EXPORT_SYMBOL(qdisc_put_rtab);
466 
467 static LIST_HEAD(qdisc_stab_list);
468 
469 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
470 	[TCA_STAB_BASE]	= { .len = sizeof(struct tc_sizespec) },
471 	[TCA_STAB_DATA] = { .type = NLA_BINARY },
472 };
473 
474 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
475 					       struct netlink_ext_ack *extack)
476 {
477 	struct nlattr *tb[TCA_STAB_MAX + 1];
478 	struct qdisc_size_table *stab;
479 	struct tc_sizespec *s;
480 	unsigned int tsize = 0;
481 	u16 *tab = NULL;
482 	int err;
483 
484 	err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
485 					  extack);
486 	if (err < 0)
487 		return ERR_PTR(err);
488 	if (!tb[TCA_STAB_BASE]) {
489 		NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
490 		return ERR_PTR(-EINVAL);
491 	}
492 
493 	s = nla_data(tb[TCA_STAB_BASE]);
494 
495 	if (s->tsize > 0) {
496 		if (!tb[TCA_STAB_DATA]) {
497 			NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
498 			return ERR_PTR(-EINVAL);
499 		}
500 		tab = nla_data(tb[TCA_STAB_DATA]);
501 		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
502 	}
503 
504 	if (tsize != s->tsize || (!tab && tsize > 0)) {
505 		NL_SET_ERR_MSG(extack, "Invalid size of size table");
506 		return ERR_PTR(-EINVAL);
507 	}
508 
509 	list_for_each_entry(stab, &qdisc_stab_list, list) {
510 		if (memcmp(&stab->szopts, s, sizeof(*s)))
511 			continue;
512 		if (tsize > 0 &&
513 		    memcmp(stab->data, tab, flex_array_size(stab, data, tsize)))
514 			continue;
515 		stab->refcnt++;
516 		return stab;
517 	}
518 
519 	if (s->size_log > STAB_SIZE_LOG_MAX ||
520 	    s->cell_log > STAB_SIZE_LOG_MAX) {
521 		NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
522 		return ERR_PTR(-EINVAL);
523 	}
524 
525 	stab = kmalloc(struct_size(stab, data, tsize), GFP_KERNEL);
526 	if (!stab)
527 		return ERR_PTR(-ENOMEM);
528 
529 	stab->refcnt = 1;
530 	stab->szopts = *s;
531 	if (tsize > 0)
532 		memcpy(stab->data, tab, flex_array_size(stab, data, tsize));
533 
534 	list_add_tail(&stab->list, &qdisc_stab_list);
535 
536 	return stab;
537 }
538 
539 void qdisc_put_stab(struct qdisc_size_table *tab)
540 {
541 	if (!tab)
542 		return;
543 
544 	if (--tab->refcnt == 0) {
545 		list_del(&tab->list);
546 		kfree_rcu(tab, rcu);
547 	}
548 }
549 EXPORT_SYMBOL(qdisc_put_stab);
550 
551 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
552 {
553 	struct nlattr *nest;
554 
555 	nest = nla_nest_start_noflag(skb, TCA_STAB);
556 	if (nest == NULL)
557 		goto nla_put_failure;
558 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
559 		goto nla_put_failure;
560 	nla_nest_end(skb, nest);
561 
562 	return skb->len;
563 
564 nla_put_failure:
565 	return -1;
566 }
567 
568 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
569 			       const struct qdisc_size_table *stab)
570 {
571 	int pkt_len, slot;
572 
573 	pkt_len = skb->len + stab->szopts.overhead;
574 	if (unlikely(!stab->szopts.tsize))
575 		goto out;
576 
577 	slot = pkt_len + stab->szopts.cell_align;
578 	if (unlikely(slot < 0))
579 		slot = 0;
580 
581 	slot >>= stab->szopts.cell_log;
582 	if (likely(slot < stab->szopts.tsize))
583 		pkt_len = stab->data[slot];
584 	else
585 		pkt_len = stab->data[stab->szopts.tsize - 1] *
586 				(slot / stab->szopts.tsize) +
587 				stab->data[slot % stab->szopts.tsize];
588 
589 	pkt_len <<= stab->szopts.size_log;
590 out:
591 	if (unlikely(pkt_len < 1))
592 		pkt_len = 1;
593 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
594 }
595 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
596 
597 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
598 {
599 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
600 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
601 			txt, qdisc->ops->id, qdisc->handle >> 16);
602 		qdisc->flags |= TCQ_F_WARN_NONWC;
603 	}
604 }
605 EXPORT_SYMBOL(qdisc_warn_nonwc);
606 
607 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
608 {
609 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
610 						 timer);
611 
612 	rcu_read_lock();
613 	__netif_schedule(qdisc_root(wd->qdisc));
614 	rcu_read_unlock();
615 
616 	return HRTIMER_NORESTART;
617 }
618 
619 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
620 				 clockid_t clockid)
621 {
622 	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
623 	wd->timer.function = qdisc_watchdog;
624 	wd->qdisc = qdisc;
625 }
626 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
627 
628 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
629 {
630 	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
631 }
632 EXPORT_SYMBOL(qdisc_watchdog_init);
633 
634 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
635 				      u64 delta_ns)
636 {
637 	if (test_bit(__QDISC_STATE_DEACTIVATED,
638 		     &qdisc_root_sleeping(wd->qdisc)->state))
639 		return;
640 
641 	if (hrtimer_is_queued(&wd->timer)) {
642 		/* If timer is already set in [expires, expires + delta_ns],
643 		 * do not reprogram it.
644 		 */
645 		if (wd->last_expires - expires <= delta_ns)
646 			return;
647 	}
648 
649 	wd->last_expires = expires;
650 	hrtimer_start_range_ns(&wd->timer,
651 			       ns_to_ktime(expires),
652 			       delta_ns,
653 			       HRTIMER_MODE_ABS_PINNED);
654 }
655 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
656 
657 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
658 {
659 	hrtimer_cancel(&wd->timer);
660 }
661 EXPORT_SYMBOL(qdisc_watchdog_cancel);
662 
663 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
664 {
665 	struct hlist_head *h;
666 	unsigned int i;
667 
668 	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
669 
670 	if (h != NULL) {
671 		for (i = 0; i < n; i++)
672 			INIT_HLIST_HEAD(&h[i]);
673 	}
674 	return h;
675 }
676 
677 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
678 {
679 	struct Qdisc_class_common *cl;
680 	struct hlist_node *next;
681 	struct hlist_head *nhash, *ohash;
682 	unsigned int nsize, nmask, osize;
683 	unsigned int i, h;
684 
685 	/* Rehash when load factor exceeds 0.75 */
686 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
687 		return;
688 	nsize = clhash->hashsize * 2;
689 	nmask = nsize - 1;
690 	nhash = qdisc_class_hash_alloc(nsize);
691 	if (nhash == NULL)
692 		return;
693 
694 	ohash = clhash->hash;
695 	osize = clhash->hashsize;
696 
697 	sch_tree_lock(sch);
698 	for (i = 0; i < osize; i++) {
699 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
700 			h = qdisc_class_hash(cl->classid, nmask);
701 			hlist_add_head(&cl->hnode, &nhash[h]);
702 		}
703 	}
704 	clhash->hash     = nhash;
705 	clhash->hashsize = nsize;
706 	clhash->hashmask = nmask;
707 	sch_tree_unlock(sch);
708 
709 	kvfree(ohash);
710 }
711 EXPORT_SYMBOL(qdisc_class_hash_grow);
712 
713 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
714 {
715 	unsigned int size = 4;
716 
717 	clhash->hash = qdisc_class_hash_alloc(size);
718 	if (!clhash->hash)
719 		return -ENOMEM;
720 	clhash->hashsize  = size;
721 	clhash->hashmask  = size - 1;
722 	clhash->hashelems = 0;
723 	return 0;
724 }
725 EXPORT_SYMBOL(qdisc_class_hash_init);
726 
727 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
728 {
729 	kvfree(clhash->hash);
730 }
731 EXPORT_SYMBOL(qdisc_class_hash_destroy);
732 
733 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
734 			     struct Qdisc_class_common *cl)
735 {
736 	unsigned int h;
737 
738 	INIT_HLIST_NODE(&cl->hnode);
739 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
740 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
741 	clhash->hashelems++;
742 }
743 EXPORT_SYMBOL(qdisc_class_hash_insert);
744 
745 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
746 			     struct Qdisc_class_common *cl)
747 {
748 	hlist_del(&cl->hnode);
749 	clhash->hashelems--;
750 }
751 EXPORT_SYMBOL(qdisc_class_hash_remove);
752 
753 /* Allocate an unique handle from space managed by kernel
754  * Possible range is [8000-FFFF]:0000 (0x8000 values)
755  */
756 static u32 qdisc_alloc_handle(struct net_device *dev)
757 {
758 	int i = 0x8000;
759 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
760 
761 	do {
762 		autohandle += TC_H_MAKE(0x10000U, 0);
763 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
764 			autohandle = TC_H_MAKE(0x80000000U, 0);
765 		if (!qdisc_lookup(dev, autohandle))
766 			return autohandle;
767 		cond_resched();
768 	} while	(--i > 0);
769 
770 	return 0;
771 }
772 
773 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
774 {
775 	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
776 	const struct Qdisc_class_ops *cops;
777 	unsigned long cl;
778 	u32 parentid;
779 	bool notify;
780 	int drops;
781 
782 	if (n == 0 && len == 0)
783 		return;
784 	drops = max_t(int, n, 0);
785 	rcu_read_lock();
786 	while ((parentid = sch->parent)) {
787 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
788 			break;
789 
790 		if (sch->flags & TCQ_F_NOPARENT)
791 			break;
792 		/* Notify parent qdisc only if child qdisc becomes empty.
793 		 *
794 		 * If child was empty even before update then backlog
795 		 * counter is screwed and we skip notification because
796 		 * parent class is already passive.
797 		 *
798 		 * If the original child was offloaded then it is allowed
799 		 * to be seem as empty, so the parent is notified anyway.
800 		 */
801 		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
802 						       !qdisc_is_offloaded);
803 		/* TODO: perform the search on a per txq basis */
804 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
805 		if (sch == NULL) {
806 			WARN_ON_ONCE(parentid != TC_H_ROOT);
807 			break;
808 		}
809 		cops = sch->ops->cl_ops;
810 		if (notify && cops->qlen_notify) {
811 			cl = cops->find(sch, parentid);
812 			cops->qlen_notify(sch, cl);
813 		}
814 		sch->q.qlen -= n;
815 		sch->qstats.backlog -= len;
816 		__qdisc_qstats_drop(sch, drops);
817 	}
818 	rcu_read_unlock();
819 }
820 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
821 
822 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
823 			      void *type_data)
824 {
825 	struct net_device *dev = qdisc_dev(sch);
826 	int err;
827 
828 	sch->flags &= ~TCQ_F_OFFLOADED;
829 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
830 		return 0;
831 
832 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
833 	if (err == -EOPNOTSUPP)
834 		return 0;
835 
836 	if (!err)
837 		sch->flags |= TCQ_F_OFFLOADED;
838 
839 	return err;
840 }
841 EXPORT_SYMBOL(qdisc_offload_dump_helper);
842 
843 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
844 				struct Qdisc *new, struct Qdisc *old,
845 				enum tc_setup_type type, void *type_data,
846 				struct netlink_ext_ack *extack)
847 {
848 	bool any_qdisc_is_offloaded;
849 	int err;
850 
851 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
852 		return;
853 
854 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
855 
856 	/* Don't report error if the graft is part of destroy operation. */
857 	if (!err || !new || new == &noop_qdisc)
858 		return;
859 
860 	/* Don't report error if the parent, the old child and the new
861 	 * one are not offloaded.
862 	 */
863 	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
864 	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
865 	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
866 
867 	if (any_qdisc_is_offloaded)
868 		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
869 }
870 EXPORT_SYMBOL(qdisc_offload_graft_helper);
871 
872 void qdisc_offload_query_caps(struct net_device *dev,
873 			      enum tc_setup_type type,
874 			      void *caps, size_t caps_len)
875 {
876 	const struct net_device_ops *ops = dev->netdev_ops;
877 	struct tc_query_caps_base base = {
878 		.type = type,
879 		.caps = caps,
880 	};
881 
882 	memset(caps, 0, caps_len);
883 
884 	if (ops->ndo_setup_tc)
885 		ops->ndo_setup_tc(dev, TC_QUERY_CAPS, &base);
886 }
887 EXPORT_SYMBOL(qdisc_offload_query_caps);
888 
889 static void qdisc_offload_graft_root(struct net_device *dev,
890 				     struct Qdisc *new, struct Qdisc *old,
891 				     struct netlink_ext_ack *extack)
892 {
893 	struct tc_root_qopt_offload graft_offload = {
894 		.command	= TC_ROOT_GRAFT,
895 		.handle		= new ? new->handle : 0,
896 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
897 				  (old && old->flags & TCQ_F_INGRESS),
898 	};
899 
900 	qdisc_offload_graft_helper(dev, NULL, new, old,
901 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
902 }
903 
904 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
905 			 u32 portid, u32 seq, u16 flags, int event)
906 {
907 	struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
908 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
909 	struct tcmsg *tcm;
910 	struct nlmsghdr  *nlh;
911 	unsigned char *b = skb_tail_pointer(skb);
912 	struct gnet_dump d;
913 	struct qdisc_size_table *stab;
914 	u32 block_index;
915 	__u32 qlen;
916 
917 	cond_resched();
918 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
919 	if (!nlh)
920 		goto out_nlmsg_trim;
921 	tcm = nlmsg_data(nlh);
922 	tcm->tcm_family = AF_UNSPEC;
923 	tcm->tcm__pad1 = 0;
924 	tcm->tcm__pad2 = 0;
925 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
926 	tcm->tcm_parent = clid;
927 	tcm->tcm_handle = q->handle;
928 	tcm->tcm_info = refcount_read(&q->refcnt);
929 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
930 		goto nla_put_failure;
931 	if (q->ops->ingress_block_get) {
932 		block_index = q->ops->ingress_block_get(q);
933 		if (block_index &&
934 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
935 			goto nla_put_failure;
936 	}
937 	if (q->ops->egress_block_get) {
938 		block_index = q->ops->egress_block_get(q);
939 		if (block_index &&
940 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
941 			goto nla_put_failure;
942 	}
943 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
944 		goto nla_put_failure;
945 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
946 		goto nla_put_failure;
947 	qlen = qdisc_qlen_sum(q);
948 
949 	stab = rtnl_dereference(q->stab);
950 	if (stab && qdisc_dump_stab(skb, stab) < 0)
951 		goto nla_put_failure;
952 
953 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
954 					 NULL, &d, TCA_PAD) < 0)
955 		goto nla_put_failure;
956 
957 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
958 		goto nla_put_failure;
959 
960 	if (qdisc_is_percpu_stats(q)) {
961 		cpu_bstats = q->cpu_bstats;
962 		cpu_qstats = q->cpu_qstats;
963 	}
964 
965 	if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
966 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
967 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
968 		goto nla_put_failure;
969 
970 	if (gnet_stats_finish_copy(&d) < 0)
971 		goto nla_put_failure;
972 
973 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
974 	return skb->len;
975 
976 out_nlmsg_trim:
977 nla_put_failure:
978 	nlmsg_trim(skb, b);
979 	return -1;
980 }
981 
982 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
983 {
984 	if (q->flags & TCQ_F_BUILTIN)
985 		return true;
986 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
987 		return true;
988 
989 	return false;
990 }
991 
992 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
993 			struct nlmsghdr *n, u32 clid,
994 			struct Qdisc *old, struct Qdisc *new)
995 {
996 	struct sk_buff *skb;
997 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
998 
999 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1000 	if (!skb)
1001 		return -ENOBUFS;
1002 
1003 	if (old && !tc_qdisc_dump_ignore(old, false)) {
1004 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1005 				  0, RTM_DELQDISC) < 0)
1006 			goto err_out;
1007 	}
1008 	if (new && !tc_qdisc_dump_ignore(new, false)) {
1009 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1010 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC) < 0)
1011 			goto err_out;
1012 	}
1013 
1014 	if (skb->len)
1015 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1016 				      n->nlmsg_flags & NLM_F_ECHO);
1017 
1018 err_out:
1019 	kfree_skb(skb);
1020 	return -EINVAL;
1021 }
1022 
1023 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1024 			       struct nlmsghdr *n, u32 clid,
1025 			       struct Qdisc *old, struct Qdisc *new)
1026 {
1027 	if (new || old)
1028 		qdisc_notify(net, skb, n, clid, old, new);
1029 
1030 	if (old)
1031 		qdisc_put(old);
1032 }
1033 
1034 static void qdisc_clear_nolock(struct Qdisc *sch)
1035 {
1036 	sch->flags &= ~TCQ_F_NOLOCK;
1037 	if (!(sch->flags & TCQ_F_CPUSTATS))
1038 		return;
1039 
1040 	free_percpu(sch->cpu_bstats);
1041 	free_percpu(sch->cpu_qstats);
1042 	sch->cpu_bstats = NULL;
1043 	sch->cpu_qstats = NULL;
1044 	sch->flags &= ~TCQ_F_CPUSTATS;
1045 }
1046 
1047 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1048  * to device "dev".
1049  *
1050  * When appropriate send a netlink notification using 'skb'
1051  * and "n".
1052  *
1053  * On success, destroy old qdisc.
1054  */
1055 
1056 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1057 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1058 		       struct Qdisc *new, struct Qdisc *old,
1059 		       struct netlink_ext_ack *extack)
1060 {
1061 	struct Qdisc *q = old;
1062 	struct net *net = dev_net(dev);
1063 
1064 	if (parent == NULL) {
1065 		unsigned int i, num_q, ingress;
1066 
1067 		ingress = 0;
1068 		num_q = dev->num_tx_queues;
1069 		if ((q && q->flags & TCQ_F_INGRESS) ||
1070 		    (new && new->flags & TCQ_F_INGRESS)) {
1071 			num_q = 1;
1072 			ingress = 1;
1073 			if (!dev_ingress_queue(dev)) {
1074 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1075 				return -ENOENT;
1076 			}
1077 		}
1078 
1079 		if (dev->flags & IFF_UP)
1080 			dev_deactivate(dev);
1081 
1082 		qdisc_offload_graft_root(dev, new, old, extack);
1083 
1084 		if (new && new->ops->attach && !ingress)
1085 			goto skip;
1086 
1087 		for (i = 0; i < num_q; i++) {
1088 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1089 
1090 			if (!ingress)
1091 				dev_queue = netdev_get_tx_queue(dev, i);
1092 
1093 			old = dev_graft_qdisc(dev_queue, new);
1094 			if (new && i > 0)
1095 				qdisc_refcount_inc(new);
1096 
1097 			if (!ingress)
1098 				qdisc_put(old);
1099 		}
1100 
1101 skip:
1102 		if (!ingress) {
1103 			old = rtnl_dereference(dev->qdisc);
1104 			if (new && !new->ops->attach)
1105 				qdisc_refcount_inc(new);
1106 			rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1107 
1108 			notify_and_destroy(net, skb, n, classid, old, new);
1109 
1110 			if (new && new->ops->attach)
1111 				new->ops->attach(new);
1112 		} else {
1113 			notify_and_destroy(net, skb, n, classid, old, new);
1114 		}
1115 
1116 		if (dev->flags & IFF_UP)
1117 			dev_activate(dev);
1118 	} else {
1119 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1120 		unsigned long cl;
1121 		int err;
1122 
1123 		/* Only support running class lockless if parent is lockless */
1124 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1125 			qdisc_clear_nolock(new);
1126 
1127 		if (!cops || !cops->graft)
1128 			return -EOPNOTSUPP;
1129 
1130 		cl = cops->find(parent, classid);
1131 		if (!cl) {
1132 			NL_SET_ERR_MSG(extack, "Specified class not found");
1133 			return -ENOENT;
1134 		}
1135 
1136 		err = cops->graft(parent, cl, new, &old, extack);
1137 		if (err)
1138 			return err;
1139 		notify_and_destroy(net, skb, n, classid, old, new);
1140 	}
1141 	return 0;
1142 }
1143 
1144 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1145 				   struct netlink_ext_ack *extack)
1146 {
1147 	u32 block_index;
1148 
1149 	if (tca[TCA_INGRESS_BLOCK]) {
1150 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1151 
1152 		if (!block_index) {
1153 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1154 			return -EINVAL;
1155 		}
1156 		if (!sch->ops->ingress_block_set) {
1157 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1158 			return -EOPNOTSUPP;
1159 		}
1160 		sch->ops->ingress_block_set(sch, block_index);
1161 	}
1162 	if (tca[TCA_EGRESS_BLOCK]) {
1163 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1164 
1165 		if (!block_index) {
1166 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1167 			return -EINVAL;
1168 		}
1169 		if (!sch->ops->egress_block_set) {
1170 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1171 			return -EOPNOTSUPP;
1172 		}
1173 		sch->ops->egress_block_set(sch, block_index);
1174 	}
1175 	return 0;
1176 }
1177 
1178 /*
1179    Allocate and initialize new qdisc.
1180 
1181    Parameters are passed via opt.
1182  */
1183 
1184 static struct Qdisc *qdisc_create(struct net_device *dev,
1185 				  struct netdev_queue *dev_queue,
1186 				  u32 parent, u32 handle,
1187 				  struct nlattr **tca, int *errp,
1188 				  struct netlink_ext_ack *extack)
1189 {
1190 	int err;
1191 	struct nlattr *kind = tca[TCA_KIND];
1192 	struct Qdisc *sch;
1193 	struct Qdisc_ops *ops;
1194 	struct qdisc_size_table *stab;
1195 
1196 	ops = qdisc_lookup_ops(kind);
1197 #ifdef CONFIG_MODULES
1198 	if (ops == NULL && kind != NULL) {
1199 		char name[IFNAMSIZ];
1200 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1201 			/* We dropped the RTNL semaphore in order to
1202 			 * perform the module load.  So, even if we
1203 			 * succeeded in loading the module we have to
1204 			 * tell the caller to replay the request.  We
1205 			 * indicate this using -EAGAIN.
1206 			 * We replay the request because the device may
1207 			 * go away in the mean time.
1208 			 */
1209 			rtnl_unlock();
1210 			request_module("sch_%s", name);
1211 			rtnl_lock();
1212 			ops = qdisc_lookup_ops(kind);
1213 			if (ops != NULL) {
1214 				/* We will try again qdisc_lookup_ops,
1215 				 * so don't keep a reference.
1216 				 */
1217 				module_put(ops->owner);
1218 				err = -EAGAIN;
1219 				goto err_out;
1220 			}
1221 		}
1222 	}
1223 #endif
1224 
1225 	err = -ENOENT;
1226 	if (!ops) {
1227 		NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1228 		goto err_out;
1229 	}
1230 
1231 	sch = qdisc_alloc(dev_queue, ops, extack);
1232 	if (IS_ERR(sch)) {
1233 		err = PTR_ERR(sch);
1234 		goto err_out2;
1235 	}
1236 
1237 	sch->parent = parent;
1238 
1239 	if (handle == TC_H_INGRESS) {
1240 		sch->flags |= TCQ_F_INGRESS;
1241 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1242 	} else {
1243 		if (handle == 0) {
1244 			handle = qdisc_alloc_handle(dev);
1245 			if (handle == 0) {
1246 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1247 				err = -ENOSPC;
1248 				goto err_out3;
1249 			}
1250 		}
1251 		if (!netif_is_multiqueue(dev))
1252 			sch->flags |= TCQ_F_ONETXQUEUE;
1253 	}
1254 
1255 	sch->handle = handle;
1256 
1257 	/* This exist to keep backward compatible with a userspace
1258 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1259 	 * facility on older kernels by setting tx_queue_len=0 (prior
1260 	 * to qdisc init), and then forgot to reinit tx_queue_len
1261 	 * before again attaching a qdisc.
1262 	 */
1263 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1264 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1265 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1266 	}
1267 
1268 	err = qdisc_block_indexes_set(sch, tca, extack);
1269 	if (err)
1270 		goto err_out3;
1271 
1272 	if (ops->init) {
1273 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1274 		if (err != 0)
1275 			goto err_out5;
1276 	}
1277 
1278 	if (tca[TCA_STAB]) {
1279 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1280 		if (IS_ERR(stab)) {
1281 			err = PTR_ERR(stab);
1282 			goto err_out4;
1283 		}
1284 		rcu_assign_pointer(sch->stab, stab);
1285 	}
1286 	if (tca[TCA_RATE]) {
1287 		err = -EOPNOTSUPP;
1288 		if (sch->flags & TCQ_F_MQROOT) {
1289 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1290 			goto err_out4;
1291 		}
1292 
1293 		err = gen_new_estimator(&sch->bstats,
1294 					sch->cpu_bstats,
1295 					&sch->rate_est,
1296 					NULL,
1297 					true,
1298 					tca[TCA_RATE]);
1299 		if (err) {
1300 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1301 			goto err_out4;
1302 		}
1303 	}
1304 
1305 	qdisc_hash_add(sch, false);
1306 	trace_qdisc_create(ops, dev, parent);
1307 
1308 	return sch;
1309 
1310 err_out5:
1311 	/* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1312 	if (ops->destroy)
1313 		ops->destroy(sch);
1314 err_out3:
1315 	netdev_put(dev, &sch->dev_tracker);
1316 	qdisc_free(sch);
1317 err_out2:
1318 	module_put(ops->owner);
1319 err_out:
1320 	*errp = err;
1321 	return NULL;
1322 
1323 err_out4:
1324 	/*
1325 	 * Any broken qdiscs that would require a ops->reset() here?
1326 	 * The qdisc was never in action so it shouldn't be necessary.
1327 	 */
1328 	qdisc_put_stab(rtnl_dereference(sch->stab));
1329 	if (ops->destroy)
1330 		ops->destroy(sch);
1331 	goto err_out3;
1332 }
1333 
1334 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1335 			struct netlink_ext_ack *extack)
1336 {
1337 	struct qdisc_size_table *ostab, *stab = NULL;
1338 	int err = 0;
1339 
1340 	if (tca[TCA_OPTIONS]) {
1341 		if (!sch->ops->change) {
1342 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1343 			return -EINVAL;
1344 		}
1345 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1346 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1347 			return -EOPNOTSUPP;
1348 		}
1349 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1350 		if (err)
1351 			return err;
1352 	}
1353 
1354 	if (tca[TCA_STAB]) {
1355 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1356 		if (IS_ERR(stab))
1357 			return PTR_ERR(stab);
1358 	}
1359 
1360 	ostab = rtnl_dereference(sch->stab);
1361 	rcu_assign_pointer(sch->stab, stab);
1362 	qdisc_put_stab(ostab);
1363 
1364 	if (tca[TCA_RATE]) {
1365 		/* NB: ignores errors from replace_estimator
1366 		   because change can't be undone. */
1367 		if (sch->flags & TCQ_F_MQROOT)
1368 			goto out;
1369 		gen_replace_estimator(&sch->bstats,
1370 				      sch->cpu_bstats,
1371 				      &sch->rate_est,
1372 				      NULL,
1373 				      true,
1374 				      tca[TCA_RATE]);
1375 	}
1376 out:
1377 	return 0;
1378 }
1379 
1380 struct check_loop_arg {
1381 	struct qdisc_walker	w;
1382 	struct Qdisc		*p;
1383 	int			depth;
1384 };
1385 
1386 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1387 			 struct qdisc_walker *w);
1388 
1389 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1390 {
1391 	struct check_loop_arg	arg;
1392 
1393 	if (q->ops->cl_ops == NULL)
1394 		return 0;
1395 
1396 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1397 	arg.w.fn = check_loop_fn;
1398 	arg.depth = depth;
1399 	arg.p = p;
1400 	q->ops->cl_ops->walk(q, &arg.w);
1401 	return arg.w.stop ? -ELOOP : 0;
1402 }
1403 
1404 static int
1405 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1406 {
1407 	struct Qdisc *leaf;
1408 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1409 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1410 
1411 	leaf = cops->leaf(q, cl);
1412 	if (leaf) {
1413 		if (leaf == arg->p || arg->depth > 7)
1414 			return -ELOOP;
1415 		return check_loop(leaf, arg->p, arg->depth + 1);
1416 	}
1417 	return 0;
1418 }
1419 
1420 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1421 	[TCA_KIND]		= { .type = NLA_STRING },
1422 	[TCA_RATE]		= { .type = NLA_BINARY,
1423 				    .len = sizeof(struct tc_estimator) },
1424 	[TCA_STAB]		= { .type = NLA_NESTED },
1425 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1426 	[TCA_CHAIN]		= { .type = NLA_U32 },
1427 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1428 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1429 };
1430 
1431 /*
1432  * Delete/get qdisc.
1433  */
1434 
1435 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1436 			struct netlink_ext_ack *extack)
1437 {
1438 	struct net *net = sock_net(skb->sk);
1439 	struct tcmsg *tcm = nlmsg_data(n);
1440 	struct nlattr *tca[TCA_MAX + 1];
1441 	struct net_device *dev;
1442 	u32 clid;
1443 	struct Qdisc *q = NULL;
1444 	struct Qdisc *p = NULL;
1445 	int err;
1446 
1447 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1448 				     rtm_tca_policy, extack);
1449 	if (err < 0)
1450 		return err;
1451 
1452 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1453 	if (!dev)
1454 		return -ENODEV;
1455 
1456 	clid = tcm->tcm_parent;
1457 	if (clid) {
1458 		if (clid != TC_H_ROOT) {
1459 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1460 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1461 				if (!p) {
1462 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1463 					return -ENOENT;
1464 				}
1465 				q = qdisc_leaf(p, clid);
1466 			} else if (dev_ingress_queue(dev)) {
1467 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1468 			}
1469 		} else {
1470 			q = rtnl_dereference(dev->qdisc);
1471 		}
1472 		if (!q) {
1473 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1474 			return -ENOENT;
1475 		}
1476 
1477 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1478 			NL_SET_ERR_MSG(extack, "Invalid handle");
1479 			return -EINVAL;
1480 		}
1481 	} else {
1482 		q = qdisc_lookup(dev, tcm->tcm_handle);
1483 		if (!q) {
1484 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1485 			return -ENOENT;
1486 		}
1487 	}
1488 
1489 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1490 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1491 		return -EINVAL;
1492 	}
1493 
1494 	if (n->nlmsg_type == RTM_DELQDISC) {
1495 		if (!clid) {
1496 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1497 			return -EINVAL;
1498 		}
1499 		if (q->handle == 0) {
1500 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1501 			return -ENOENT;
1502 		}
1503 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1504 		if (err != 0)
1505 			return err;
1506 	} else {
1507 		qdisc_notify(net, skb, n, clid, NULL, q);
1508 	}
1509 	return 0;
1510 }
1511 
1512 /*
1513  * Create/change qdisc.
1514  */
1515 
1516 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1517 			   struct netlink_ext_ack *extack)
1518 {
1519 	struct net *net = sock_net(skb->sk);
1520 	struct tcmsg *tcm;
1521 	struct nlattr *tca[TCA_MAX + 1];
1522 	struct net_device *dev;
1523 	u32 clid;
1524 	struct Qdisc *q, *p;
1525 	int err;
1526 
1527 replay:
1528 	/* Reinit, just in case something touches this. */
1529 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1530 				     rtm_tca_policy, extack);
1531 	if (err < 0)
1532 		return err;
1533 
1534 	tcm = nlmsg_data(n);
1535 	clid = tcm->tcm_parent;
1536 	q = p = NULL;
1537 
1538 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1539 	if (!dev)
1540 		return -ENODEV;
1541 
1542 
1543 	if (clid) {
1544 		if (clid != TC_H_ROOT) {
1545 			if (clid != TC_H_INGRESS) {
1546 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1547 				if (!p) {
1548 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1549 					return -ENOENT;
1550 				}
1551 				q = qdisc_leaf(p, clid);
1552 			} else if (dev_ingress_queue_create(dev)) {
1553 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1554 			}
1555 		} else {
1556 			q = rtnl_dereference(dev->qdisc);
1557 		}
1558 
1559 		/* It may be default qdisc, ignore it */
1560 		if (q && q->handle == 0)
1561 			q = NULL;
1562 
1563 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1564 			if (tcm->tcm_handle) {
1565 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1566 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1567 					return -EEXIST;
1568 				}
1569 				if (TC_H_MIN(tcm->tcm_handle)) {
1570 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1571 					return -EINVAL;
1572 				}
1573 				q = qdisc_lookup(dev, tcm->tcm_handle);
1574 				if (!q)
1575 					goto create_n_graft;
1576 				if (n->nlmsg_flags & NLM_F_EXCL) {
1577 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1578 					return -EEXIST;
1579 				}
1580 				if (tca[TCA_KIND] &&
1581 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1582 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1583 					return -EINVAL;
1584 				}
1585 				if (q == p ||
1586 				    (p && check_loop(q, p, 0))) {
1587 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1588 					return -ELOOP;
1589 				}
1590 				qdisc_refcount_inc(q);
1591 				goto graft;
1592 			} else {
1593 				if (!q)
1594 					goto create_n_graft;
1595 
1596 				/* This magic test requires explanation.
1597 				 *
1598 				 *   We know, that some child q is already
1599 				 *   attached to this parent and have choice:
1600 				 *   either to change it or to create/graft new one.
1601 				 *
1602 				 *   1. We are allowed to create/graft only
1603 				 *   if CREATE and REPLACE flags are set.
1604 				 *
1605 				 *   2. If EXCL is set, requestor wanted to say,
1606 				 *   that qdisc tcm_handle is not expected
1607 				 *   to exist, so that we choose create/graft too.
1608 				 *
1609 				 *   3. The last case is when no flags are set.
1610 				 *   Alas, it is sort of hole in API, we
1611 				 *   cannot decide what to do unambiguously.
1612 				 *   For now we select create/graft, if
1613 				 *   user gave KIND, which does not match existing.
1614 				 */
1615 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1616 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1617 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1618 				     (tca[TCA_KIND] &&
1619 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1620 					goto create_n_graft;
1621 			}
1622 		}
1623 	} else {
1624 		if (!tcm->tcm_handle) {
1625 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1626 			return -EINVAL;
1627 		}
1628 		q = qdisc_lookup(dev, tcm->tcm_handle);
1629 	}
1630 
1631 	/* Change qdisc parameters */
1632 	if (!q) {
1633 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1634 		return -ENOENT;
1635 	}
1636 	if (n->nlmsg_flags & NLM_F_EXCL) {
1637 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1638 		return -EEXIST;
1639 	}
1640 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1641 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1642 		return -EINVAL;
1643 	}
1644 	err = qdisc_change(q, tca, extack);
1645 	if (err == 0)
1646 		qdisc_notify(net, skb, n, clid, NULL, q);
1647 	return err;
1648 
1649 create_n_graft:
1650 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1651 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1652 		return -ENOENT;
1653 	}
1654 	if (clid == TC_H_INGRESS) {
1655 		if (dev_ingress_queue(dev)) {
1656 			q = qdisc_create(dev, dev_ingress_queue(dev),
1657 					 tcm->tcm_parent, tcm->tcm_parent,
1658 					 tca, &err, extack);
1659 		} else {
1660 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1661 			err = -ENOENT;
1662 		}
1663 	} else {
1664 		struct netdev_queue *dev_queue;
1665 
1666 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1667 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1668 		else if (p)
1669 			dev_queue = p->dev_queue;
1670 		else
1671 			dev_queue = netdev_get_tx_queue(dev, 0);
1672 
1673 		q = qdisc_create(dev, dev_queue,
1674 				 tcm->tcm_parent, tcm->tcm_handle,
1675 				 tca, &err, extack);
1676 	}
1677 	if (q == NULL) {
1678 		if (err == -EAGAIN)
1679 			goto replay;
1680 		return err;
1681 	}
1682 
1683 graft:
1684 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1685 	if (err) {
1686 		if (q)
1687 			qdisc_put(q);
1688 		return err;
1689 	}
1690 
1691 	return 0;
1692 }
1693 
1694 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1695 			      struct netlink_callback *cb,
1696 			      int *q_idx_p, int s_q_idx, bool recur,
1697 			      bool dump_invisible)
1698 {
1699 	int ret = 0, q_idx = *q_idx_p;
1700 	struct Qdisc *q;
1701 	int b;
1702 
1703 	if (!root)
1704 		return 0;
1705 
1706 	q = root;
1707 	if (q_idx < s_q_idx) {
1708 		q_idx++;
1709 	} else {
1710 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1711 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1712 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1713 				  RTM_NEWQDISC) <= 0)
1714 			goto done;
1715 		q_idx++;
1716 	}
1717 
1718 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1719 	 * itself has already been dumped.
1720 	 *
1721 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1722 	 * qdisc hashtable, we don't want to hit it again
1723 	 */
1724 	if (!qdisc_dev(root) || !recur)
1725 		goto out;
1726 
1727 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1728 		if (q_idx < s_q_idx) {
1729 			q_idx++;
1730 			continue;
1731 		}
1732 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1733 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1734 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1735 				  RTM_NEWQDISC) <= 0)
1736 			goto done;
1737 		q_idx++;
1738 	}
1739 
1740 out:
1741 	*q_idx_p = q_idx;
1742 	return ret;
1743 done:
1744 	ret = -1;
1745 	goto out;
1746 }
1747 
1748 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1749 {
1750 	struct net *net = sock_net(skb->sk);
1751 	int idx, q_idx;
1752 	int s_idx, s_q_idx;
1753 	struct net_device *dev;
1754 	const struct nlmsghdr *nlh = cb->nlh;
1755 	struct nlattr *tca[TCA_MAX + 1];
1756 	int err;
1757 
1758 	s_idx = cb->args[0];
1759 	s_q_idx = q_idx = cb->args[1];
1760 
1761 	idx = 0;
1762 	ASSERT_RTNL();
1763 
1764 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1765 				     rtm_tca_policy, cb->extack);
1766 	if (err < 0)
1767 		return err;
1768 
1769 	for_each_netdev(net, dev) {
1770 		struct netdev_queue *dev_queue;
1771 
1772 		if (idx < s_idx)
1773 			goto cont;
1774 		if (idx > s_idx)
1775 			s_q_idx = 0;
1776 		q_idx = 0;
1777 
1778 		if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1779 				       skb, cb, &q_idx, s_q_idx,
1780 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1781 			goto done;
1782 
1783 		dev_queue = dev_ingress_queue(dev);
1784 		if (dev_queue &&
1785 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1786 				       &q_idx, s_q_idx, false,
1787 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1788 			goto done;
1789 
1790 cont:
1791 		idx++;
1792 	}
1793 
1794 done:
1795 	cb->args[0] = idx;
1796 	cb->args[1] = q_idx;
1797 
1798 	return skb->len;
1799 }
1800 
1801 
1802 
1803 /************************************************
1804  *	Traffic classes manipulation.		*
1805  ************************************************/
1806 
1807 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1808 			  unsigned long cl,
1809 			  u32 portid, u32 seq, u16 flags, int event)
1810 {
1811 	struct tcmsg *tcm;
1812 	struct nlmsghdr  *nlh;
1813 	unsigned char *b = skb_tail_pointer(skb);
1814 	struct gnet_dump d;
1815 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1816 
1817 	cond_resched();
1818 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1819 	if (!nlh)
1820 		goto out_nlmsg_trim;
1821 	tcm = nlmsg_data(nlh);
1822 	tcm->tcm_family = AF_UNSPEC;
1823 	tcm->tcm__pad1 = 0;
1824 	tcm->tcm__pad2 = 0;
1825 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1826 	tcm->tcm_parent = q->handle;
1827 	tcm->tcm_handle = q->handle;
1828 	tcm->tcm_info = 0;
1829 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1830 		goto nla_put_failure;
1831 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1832 		goto nla_put_failure;
1833 
1834 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1835 					 NULL, &d, TCA_PAD) < 0)
1836 		goto nla_put_failure;
1837 
1838 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1839 		goto nla_put_failure;
1840 
1841 	if (gnet_stats_finish_copy(&d) < 0)
1842 		goto nla_put_failure;
1843 
1844 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1845 	return skb->len;
1846 
1847 out_nlmsg_trim:
1848 nla_put_failure:
1849 	nlmsg_trim(skb, b);
1850 	return -1;
1851 }
1852 
1853 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1854 			 struct nlmsghdr *n, struct Qdisc *q,
1855 			 unsigned long cl, int event)
1856 {
1857 	struct sk_buff *skb;
1858 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1859 
1860 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1861 	if (!skb)
1862 		return -ENOBUFS;
1863 
1864 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event) < 0) {
1865 		kfree_skb(skb);
1866 		return -EINVAL;
1867 	}
1868 
1869 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1870 			      n->nlmsg_flags & NLM_F_ECHO);
1871 }
1872 
1873 static int tclass_del_notify(struct net *net,
1874 			     const struct Qdisc_class_ops *cops,
1875 			     struct sk_buff *oskb, struct nlmsghdr *n,
1876 			     struct Qdisc *q, unsigned long cl,
1877 			     struct netlink_ext_ack *extack)
1878 {
1879 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1880 	struct sk_buff *skb;
1881 	int err = 0;
1882 
1883 	if (!cops->delete)
1884 		return -EOPNOTSUPP;
1885 
1886 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1887 	if (!skb)
1888 		return -ENOBUFS;
1889 
1890 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1891 			   RTM_DELTCLASS) < 0) {
1892 		kfree_skb(skb);
1893 		return -EINVAL;
1894 	}
1895 
1896 	err = cops->delete(q, cl, extack);
1897 	if (err) {
1898 		kfree_skb(skb);
1899 		return err;
1900 	}
1901 
1902 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1903 			     n->nlmsg_flags & NLM_F_ECHO);
1904 	return err;
1905 }
1906 
1907 #ifdef CONFIG_NET_CLS
1908 
1909 struct tcf_bind_args {
1910 	struct tcf_walker w;
1911 	unsigned long base;
1912 	unsigned long cl;
1913 	u32 classid;
1914 };
1915 
1916 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1917 {
1918 	struct tcf_bind_args *a = (void *)arg;
1919 
1920 	if (n && tp->ops->bind_class) {
1921 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1922 
1923 		sch_tree_lock(q);
1924 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1925 		sch_tree_unlock(q);
1926 	}
1927 	return 0;
1928 }
1929 
1930 struct tc_bind_class_args {
1931 	struct qdisc_walker w;
1932 	unsigned long new_cl;
1933 	u32 portid;
1934 	u32 clid;
1935 };
1936 
1937 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1938 				struct qdisc_walker *w)
1939 {
1940 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1941 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1942 	struct tcf_block *block;
1943 	struct tcf_chain *chain;
1944 
1945 	block = cops->tcf_block(q, cl, NULL);
1946 	if (!block)
1947 		return 0;
1948 	for (chain = tcf_get_next_chain(block, NULL);
1949 	     chain;
1950 	     chain = tcf_get_next_chain(block, chain)) {
1951 		struct tcf_proto *tp;
1952 
1953 		for (tp = tcf_get_next_proto(chain, NULL);
1954 		     tp; tp = tcf_get_next_proto(chain, tp)) {
1955 			struct tcf_bind_args arg = {};
1956 
1957 			arg.w.fn = tcf_node_bind;
1958 			arg.classid = a->clid;
1959 			arg.base = cl;
1960 			arg.cl = a->new_cl;
1961 			tp->ops->walk(tp, &arg.w, true);
1962 		}
1963 	}
1964 
1965 	return 0;
1966 }
1967 
1968 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1969 			   unsigned long new_cl)
1970 {
1971 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1972 	struct tc_bind_class_args args = {};
1973 
1974 	if (!cops->tcf_block)
1975 		return;
1976 	args.portid = portid;
1977 	args.clid = clid;
1978 	args.new_cl = new_cl;
1979 	args.w.fn = tc_bind_class_walker;
1980 	q->ops->cl_ops->walk(q, &args.w);
1981 }
1982 
1983 #else
1984 
1985 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1986 			   unsigned long new_cl)
1987 {
1988 }
1989 
1990 #endif
1991 
1992 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
1993 			 struct netlink_ext_ack *extack)
1994 {
1995 	struct net *net = sock_net(skb->sk);
1996 	struct tcmsg *tcm = nlmsg_data(n);
1997 	struct nlattr *tca[TCA_MAX + 1];
1998 	struct net_device *dev;
1999 	struct Qdisc *q = NULL;
2000 	const struct Qdisc_class_ops *cops;
2001 	unsigned long cl = 0;
2002 	unsigned long new_cl;
2003 	u32 portid;
2004 	u32 clid;
2005 	u32 qid;
2006 	int err;
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 = rtnl_dereference(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 = rtnl_dereference(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(rtnl_dereference(dev->qdisc),
2219 				skb, tcm, cb, &t, s_t, true) < 0)
2220 		goto done;
2221 
2222 	dev_queue = dev_ingress_queue(dev);
2223 	if (dev_queue &&
2224 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2225 				&t, s_t, false) < 0)
2226 		goto done;
2227 
2228 done:
2229 	cb->args[0] = t;
2230 
2231 	dev_put(dev);
2232 	return skb->len;
2233 }
2234 
2235 #ifdef CONFIG_PROC_FS
2236 static int psched_show(struct seq_file *seq, void *v)
2237 {
2238 	seq_printf(seq, "%08x %08x %08x %08x\n",
2239 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2240 		   1000000,
2241 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2242 
2243 	return 0;
2244 }
2245 
2246 static int __net_init psched_net_init(struct net *net)
2247 {
2248 	struct proc_dir_entry *e;
2249 
2250 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2251 	if (e == NULL)
2252 		return -ENOMEM;
2253 
2254 	return 0;
2255 }
2256 
2257 static void __net_exit psched_net_exit(struct net *net)
2258 {
2259 	remove_proc_entry("psched", net->proc_net);
2260 }
2261 #else
2262 static int __net_init psched_net_init(struct net *net)
2263 {
2264 	return 0;
2265 }
2266 
2267 static void __net_exit psched_net_exit(struct net *net)
2268 {
2269 }
2270 #endif
2271 
2272 static struct pernet_operations psched_net_ops = {
2273 	.init = psched_net_init,
2274 	.exit = psched_net_exit,
2275 };
2276 
2277 DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper);
2278 
2279 static int __init pktsched_init(void)
2280 {
2281 	int err;
2282 
2283 	err = register_pernet_subsys(&psched_net_ops);
2284 	if (err) {
2285 		pr_err("pktsched_init: "
2286 		       "cannot initialize per netns operations\n");
2287 		return err;
2288 	}
2289 
2290 	register_qdisc(&pfifo_fast_ops);
2291 	register_qdisc(&pfifo_qdisc_ops);
2292 	register_qdisc(&bfifo_qdisc_ops);
2293 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2294 	register_qdisc(&mq_qdisc_ops);
2295 	register_qdisc(&noqueue_qdisc_ops);
2296 
2297 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2298 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2299 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2300 		      0);
2301 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2302 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2303 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2304 		      0);
2305 
2306 	tc_wrapper_init();
2307 
2308 	return 0;
2309 }
2310 
2311 subsys_initcall(pktsched_init);
2312