xref: /openbmc/linux/net/sched/sch_api.c (revision 1edd0337)
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 			 struct netlink_ext_ack *extack)
907 {
908 	struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
909 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
910 	struct tcmsg *tcm;
911 	struct nlmsghdr  *nlh;
912 	unsigned char *b = skb_tail_pointer(skb);
913 	struct gnet_dump d;
914 	struct qdisc_size_table *stab;
915 	u32 block_index;
916 	__u32 qlen;
917 
918 	cond_resched();
919 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
920 	if (!nlh)
921 		goto out_nlmsg_trim;
922 	tcm = nlmsg_data(nlh);
923 	tcm->tcm_family = AF_UNSPEC;
924 	tcm->tcm__pad1 = 0;
925 	tcm->tcm__pad2 = 0;
926 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
927 	tcm->tcm_parent = clid;
928 	tcm->tcm_handle = q->handle;
929 	tcm->tcm_info = refcount_read(&q->refcnt);
930 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
931 		goto nla_put_failure;
932 	if (q->ops->ingress_block_get) {
933 		block_index = q->ops->ingress_block_get(q);
934 		if (block_index &&
935 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
936 			goto nla_put_failure;
937 	}
938 	if (q->ops->egress_block_get) {
939 		block_index = q->ops->egress_block_get(q);
940 		if (block_index &&
941 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
942 			goto nla_put_failure;
943 	}
944 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
945 		goto nla_put_failure;
946 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
947 		goto nla_put_failure;
948 	qlen = qdisc_qlen_sum(q);
949 
950 	stab = rtnl_dereference(q->stab);
951 	if (stab && qdisc_dump_stab(skb, stab) < 0)
952 		goto nla_put_failure;
953 
954 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
955 					 NULL, &d, TCA_PAD) < 0)
956 		goto nla_put_failure;
957 
958 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
959 		goto nla_put_failure;
960 
961 	if (qdisc_is_percpu_stats(q)) {
962 		cpu_bstats = q->cpu_bstats;
963 		cpu_qstats = q->cpu_qstats;
964 	}
965 
966 	if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
967 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
968 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
969 		goto nla_put_failure;
970 
971 	if (gnet_stats_finish_copy(&d) < 0)
972 		goto nla_put_failure;
973 
974 	if (extack && extack->_msg &&
975 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
976 		goto out_nlmsg_trim;
977 
978 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
979 
980 	return skb->len;
981 
982 out_nlmsg_trim:
983 nla_put_failure:
984 	nlmsg_trim(skb, b);
985 	return -1;
986 }
987 
988 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
989 {
990 	if (q->flags & TCQ_F_BUILTIN)
991 		return true;
992 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
993 		return true;
994 
995 	return false;
996 }
997 
998 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
999 			struct nlmsghdr *n, u32 clid,
1000 			struct Qdisc *old, struct Qdisc *new,
1001 			struct netlink_ext_ack *extack)
1002 {
1003 	struct sk_buff *skb;
1004 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1005 
1006 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1007 	if (!skb)
1008 		return -ENOBUFS;
1009 
1010 	if (old && !tc_qdisc_dump_ignore(old, false)) {
1011 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1012 				  0, RTM_DELQDISC, extack) < 0)
1013 			goto err_out;
1014 	}
1015 	if (new && !tc_qdisc_dump_ignore(new, false)) {
1016 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1017 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC, extack) < 0)
1018 			goto err_out;
1019 	}
1020 
1021 	if (skb->len)
1022 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1023 				      n->nlmsg_flags & NLM_F_ECHO);
1024 
1025 err_out:
1026 	kfree_skb(skb);
1027 	return -EINVAL;
1028 }
1029 
1030 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1031 			       struct nlmsghdr *n, u32 clid,
1032 			       struct Qdisc *old, struct Qdisc *new,
1033 			       struct netlink_ext_ack *extack)
1034 {
1035 	if (new || old)
1036 		qdisc_notify(net, skb, n, clid, old, new, extack);
1037 
1038 	if (old)
1039 		qdisc_put(old);
1040 }
1041 
1042 static void qdisc_clear_nolock(struct Qdisc *sch)
1043 {
1044 	sch->flags &= ~TCQ_F_NOLOCK;
1045 	if (!(sch->flags & TCQ_F_CPUSTATS))
1046 		return;
1047 
1048 	free_percpu(sch->cpu_bstats);
1049 	free_percpu(sch->cpu_qstats);
1050 	sch->cpu_bstats = NULL;
1051 	sch->cpu_qstats = NULL;
1052 	sch->flags &= ~TCQ_F_CPUSTATS;
1053 }
1054 
1055 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1056  * to device "dev".
1057  *
1058  * When appropriate send a netlink notification using 'skb'
1059  * and "n".
1060  *
1061  * On success, destroy old qdisc.
1062  */
1063 
1064 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1065 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1066 		       struct Qdisc *new, struct Qdisc *old,
1067 		       struct netlink_ext_ack *extack)
1068 {
1069 	struct Qdisc *q = old;
1070 	struct net *net = dev_net(dev);
1071 
1072 	if (parent == NULL) {
1073 		unsigned int i, num_q, ingress;
1074 
1075 		ingress = 0;
1076 		num_q = dev->num_tx_queues;
1077 		if ((q && q->flags & TCQ_F_INGRESS) ||
1078 		    (new && new->flags & TCQ_F_INGRESS)) {
1079 			num_q = 1;
1080 			ingress = 1;
1081 			if (!dev_ingress_queue(dev)) {
1082 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1083 				return -ENOENT;
1084 			}
1085 		}
1086 
1087 		if (dev->flags & IFF_UP)
1088 			dev_deactivate(dev);
1089 
1090 		qdisc_offload_graft_root(dev, new, old, extack);
1091 
1092 		if (new && new->ops->attach && !ingress)
1093 			goto skip;
1094 
1095 		for (i = 0; i < num_q; i++) {
1096 			struct netdev_queue *dev_queue = dev_ingress_queue(dev);
1097 
1098 			if (!ingress)
1099 				dev_queue = netdev_get_tx_queue(dev, i);
1100 
1101 			old = dev_graft_qdisc(dev_queue, new);
1102 			if (new && i > 0)
1103 				qdisc_refcount_inc(new);
1104 
1105 			if (!ingress)
1106 				qdisc_put(old);
1107 		}
1108 
1109 skip:
1110 		if (!ingress) {
1111 			old = rtnl_dereference(dev->qdisc);
1112 			if (new && !new->ops->attach)
1113 				qdisc_refcount_inc(new);
1114 			rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1115 
1116 			notify_and_destroy(net, skb, n, classid, old, new, extack);
1117 
1118 			if (new && new->ops->attach)
1119 				new->ops->attach(new);
1120 		} else {
1121 			notify_and_destroy(net, skb, n, classid, old, new, extack);
1122 		}
1123 
1124 		if (dev->flags & IFF_UP)
1125 			dev_activate(dev);
1126 	} else {
1127 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1128 		unsigned long cl;
1129 		int err;
1130 
1131 		/* Only support running class lockless if parent is lockless */
1132 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1133 			qdisc_clear_nolock(new);
1134 
1135 		if (!cops || !cops->graft)
1136 			return -EOPNOTSUPP;
1137 
1138 		cl = cops->find(parent, classid);
1139 		if (!cl) {
1140 			NL_SET_ERR_MSG(extack, "Specified class not found");
1141 			return -ENOENT;
1142 		}
1143 
1144 		if (new && new->ops == &noqueue_qdisc_ops) {
1145 			NL_SET_ERR_MSG(extack, "Cannot assign noqueue to a class");
1146 			return -EINVAL;
1147 		}
1148 
1149 		err = cops->graft(parent, cl, new, &old, extack);
1150 		if (err)
1151 			return err;
1152 		notify_and_destroy(net, skb, n, classid, old, new, extack);
1153 	}
1154 	return 0;
1155 }
1156 
1157 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1158 				   struct netlink_ext_ack *extack)
1159 {
1160 	u32 block_index;
1161 
1162 	if (tca[TCA_INGRESS_BLOCK]) {
1163 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1164 
1165 		if (!block_index) {
1166 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1167 			return -EINVAL;
1168 		}
1169 		if (!sch->ops->ingress_block_set) {
1170 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1171 			return -EOPNOTSUPP;
1172 		}
1173 		sch->ops->ingress_block_set(sch, block_index);
1174 	}
1175 	if (tca[TCA_EGRESS_BLOCK]) {
1176 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1177 
1178 		if (!block_index) {
1179 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1180 			return -EINVAL;
1181 		}
1182 		if (!sch->ops->egress_block_set) {
1183 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1184 			return -EOPNOTSUPP;
1185 		}
1186 		sch->ops->egress_block_set(sch, block_index);
1187 	}
1188 	return 0;
1189 }
1190 
1191 /*
1192    Allocate and initialize new qdisc.
1193 
1194    Parameters are passed via opt.
1195  */
1196 
1197 static struct Qdisc *qdisc_create(struct net_device *dev,
1198 				  struct netdev_queue *dev_queue,
1199 				  u32 parent, u32 handle,
1200 				  struct nlattr **tca, int *errp,
1201 				  struct netlink_ext_ack *extack)
1202 {
1203 	int err;
1204 	struct nlattr *kind = tca[TCA_KIND];
1205 	struct Qdisc *sch;
1206 	struct Qdisc_ops *ops;
1207 	struct qdisc_size_table *stab;
1208 
1209 	ops = qdisc_lookup_ops(kind);
1210 #ifdef CONFIG_MODULES
1211 	if (ops == NULL && kind != NULL) {
1212 		char name[IFNAMSIZ];
1213 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1214 			/* We dropped the RTNL semaphore in order to
1215 			 * perform the module load.  So, even if we
1216 			 * succeeded in loading the module we have to
1217 			 * tell the caller to replay the request.  We
1218 			 * indicate this using -EAGAIN.
1219 			 * We replay the request because the device may
1220 			 * go away in the mean time.
1221 			 */
1222 			rtnl_unlock();
1223 			request_module("sch_%s", name);
1224 			rtnl_lock();
1225 			ops = qdisc_lookup_ops(kind);
1226 			if (ops != NULL) {
1227 				/* We will try again qdisc_lookup_ops,
1228 				 * so don't keep a reference.
1229 				 */
1230 				module_put(ops->owner);
1231 				err = -EAGAIN;
1232 				goto err_out;
1233 			}
1234 		}
1235 	}
1236 #endif
1237 
1238 	err = -ENOENT;
1239 	if (!ops) {
1240 		NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1241 		goto err_out;
1242 	}
1243 
1244 	sch = qdisc_alloc(dev_queue, ops, extack);
1245 	if (IS_ERR(sch)) {
1246 		err = PTR_ERR(sch);
1247 		goto err_out2;
1248 	}
1249 
1250 	sch->parent = parent;
1251 
1252 	if (handle == TC_H_INGRESS) {
1253 		sch->flags |= TCQ_F_INGRESS;
1254 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1255 	} else {
1256 		if (handle == 0) {
1257 			handle = qdisc_alloc_handle(dev);
1258 			if (handle == 0) {
1259 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1260 				err = -ENOSPC;
1261 				goto err_out3;
1262 			}
1263 		}
1264 		if (!netif_is_multiqueue(dev))
1265 			sch->flags |= TCQ_F_ONETXQUEUE;
1266 	}
1267 
1268 	sch->handle = handle;
1269 
1270 	/* This exist to keep backward compatible with a userspace
1271 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1272 	 * facility on older kernels by setting tx_queue_len=0 (prior
1273 	 * to qdisc init), and then forgot to reinit tx_queue_len
1274 	 * before again attaching a qdisc.
1275 	 */
1276 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1277 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1278 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1279 	}
1280 
1281 	err = qdisc_block_indexes_set(sch, tca, extack);
1282 	if (err)
1283 		goto err_out3;
1284 
1285 	if (tca[TCA_STAB]) {
1286 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1287 		if (IS_ERR(stab)) {
1288 			err = PTR_ERR(stab);
1289 			goto err_out3;
1290 		}
1291 		rcu_assign_pointer(sch->stab, stab);
1292 	}
1293 
1294 	if (ops->init) {
1295 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1296 		if (err != 0)
1297 			goto err_out4;
1298 	}
1299 
1300 	if (tca[TCA_RATE]) {
1301 		err = -EOPNOTSUPP;
1302 		if (sch->flags & TCQ_F_MQROOT) {
1303 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1304 			goto err_out4;
1305 		}
1306 
1307 		err = gen_new_estimator(&sch->bstats,
1308 					sch->cpu_bstats,
1309 					&sch->rate_est,
1310 					NULL,
1311 					true,
1312 					tca[TCA_RATE]);
1313 		if (err) {
1314 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1315 			goto err_out4;
1316 		}
1317 	}
1318 
1319 	qdisc_hash_add(sch, false);
1320 	trace_qdisc_create(ops, dev, parent);
1321 
1322 	return sch;
1323 
1324 err_out4:
1325 	/* Even if ops->init() failed, we call ops->destroy()
1326 	 * like qdisc_create_dflt().
1327 	 */
1328 	if (ops->destroy)
1329 		ops->destroy(sch);
1330 	qdisc_put_stab(rtnl_dereference(sch->stab));
1331 err_out3:
1332 	netdev_put(dev, &sch->dev_tracker);
1333 	qdisc_free(sch);
1334 err_out2:
1335 	module_put(ops->owner);
1336 err_out:
1337 	*errp = err;
1338 	return NULL;
1339 }
1340 
1341 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1342 			struct netlink_ext_ack *extack)
1343 {
1344 	struct qdisc_size_table *ostab, *stab = NULL;
1345 	int err = 0;
1346 
1347 	if (tca[TCA_OPTIONS]) {
1348 		if (!sch->ops->change) {
1349 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1350 			return -EINVAL;
1351 		}
1352 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1353 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1354 			return -EOPNOTSUPP;
1355 		}
1356 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1357 		if (err)
1358 			return err;
1359 	}
1360 
1361 	if (tca[TCA_STAB]) {
1362 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1363 		if (IS_ERR(stab))
1364 			return PTR_ERR(stab);
1365 	}
1366 
1367 	ostab = rtnl_dereference(sch->stab);
1368 	rcu_assign_pointer(sch->stab, stab);
1369 	qdisc_put_stab(ostab);
1370 
1371 	if (tca[TCA_RATE]) {
1372 		/* NB: ignores errors from replace_estimator
1373 		   because change can't be undone. */
1374 		if (sch->flags & TCQ_F_MQROOT)
1375 			goto out;
1376 		gen_replace_estimator(&sch->bstats,
1377 				      sch->cpu_bstats,
1378 				      &sch->rate_est,
1379 				      NULL,
1380 				      true,
1381 				      tca[TCA_RATE]);
1382 	}
1383 out:
1384 	return 0;
1385 }
1386 
1387 struct check_loop_arg {
1388 	struct qdisc_walker	w;
1389 	struct Qdisc		*p;
1390 	int			depth;
1391 };
1392 
1393 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1394 			 struct qdisc_walker *w);
1395 
1396 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1397 {
1398 	struct check_loop_arg	arg;
1399 
1400 	if (q->ops->cl_ops == NULL)
1401 		return 0;
1402 
1403 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1404 	arg.w.fn = check_loop_fn;
1405 	arg.depth = depth;
1406 	arg.p = p;
1407 	q->ops->cl_ops->walk(q, &arg.w);
1408 	return arg.w.stop ? -ELOOP : 0;
1409 }
1410 
1411 static int
1412 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1413 {
1414 	struct Qdisc *leaf;
1415 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1416 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1417 
1418 	leaf = cops->leaf(q, cl);
1419 	if (leaf) {
1420 		if (leaf == arg->p || arg->depth > 7)
1421 			return -ELOOP;
1422 		return check_loop(leaf, arg->p, arg->depth + 1);
1423 	}
1424 	return 0;
1425 }
1426 
1427 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1428 	[TCA_KIND]		= { .type = NLA_STRING },
1429 	[TCA_RATE]		= { .type = NLA_BINARY,
1430 				    .len = sizeof(struct tc_estimator) },
1431 	[TCA_STAB]		= { .type = NLA_NESTED },
1432 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1433 	[TCA_CHAIN]		= { .type = NLA_U32 },
1434 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1435 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1436 };
1437 
1438 /*
1439  * Delete/get qdisc.
1440  */
1441 
1442 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1443 			struct netlink_ext_ack *extack)
1444 {
1445 	struct net *net = sock_net(skb->sk);
1446 	struct tcmsg *tcm = nlmsg_data(n);
1447 	struct nlattr *tca[TCA_MAX + 1];
1448 	struct net_device *dev;
1449 	u32 clid;
1450 	struct Qdisc *q = NULL;
1451 	struct Qdisc *p = NULL;
1452 	int err;
1453 
1454 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1455 				     rtm_tca_policy, extack);
1456 	if (err < 0)
1457 		return err;
1458 
1459 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1460 	if (!dev)
1461 		return -ENODEV;
1462 
1463 	clid = tcm->tcm_parent;
1464 	if (clid) {
1465 		if (clid != TC_H_ROOT) {
1466 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1467 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1468 				if (!p) {
1469 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1470 					return -ENOENT;
1471 				}
1472 				q = qdisc_leaf(p, clid);
1473 			} else if (dev_ingress_queue(dev)) {
1474 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1475 			}
1476 		} else {
1477 			q = rtnl_dereference(dev->qdisc);
1478 		}
1479 		if (!q) {
1480 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1481 			return -ENOENT;
1482 		}
1483 
1484 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1485 			NL_SET_ERR_MSG(extack, "Invalid handle");
1486 			return -EINVAL;
1487 		}
1488 	} else {
1489 		q = qdisc_lookup(dev, tcm->tcm_handle);
1490 		if (!q) {
1491 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1492 			return -ENOENT;
1493 		}
1494 	}
1495 
1496 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1497 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1498 		return -EINVAL;
1499 	}
1500 
1501 	if (n->nlmsg_type == RTM_DELQDISC) {
1502 		if (!clid) {
1503 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1504 			return -EINVAL;
1505 		}
1506 		if (q->handle == 0) {
1507 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1508 			return -ENOENT;
1509 		}
1510 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1511 		if (err != 0)
1512 			return err;
1513 	} else {
1514 		qdisc_notify(net, skb, n, clid, NULL, q, NULL);
1515 	}
1516 	return 0;
1517 }
1518 
1519 /*
1520  * Create/change qdisc.
1521  */
1522 
1523 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1524 			   struct netlink_ext_ack *extack)
1525 {
1526 	struct net *net = sock_net(skb->sk);
1527 	struct tcmsg *tcm;
1528 	struct nlattr *tca[TCA_MAX + 1];
1529 	struct net_device *dev;
1530 	u32 clid;
1531 	struct Qdisc *q, *p;
1532 	int err;
1533 
1534 replay:
1535 	/* Reinit, just in case something touches this. */
1536 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1537 				     rtm_tca_policy, extack);
1538 	if (err < 0)
1539 		return err;
1540 
1541 	tcm = nlmsg_data(n);
1542 	clid = tcm->tcm_parent;
1543 	q = p = NULL;
1544 
1545 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1546 	if (!dev)
1547 		return -ENODEV;
1548 
1549 
1550 	if (clid) {
1551 		if (clid != TC_H_ROOT) {
1552 			if (clid != TC_H_INGRESS) {
1553 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1554 				if (!p) {
1555 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1556 					return -ENOENT;
1557 				}
1558 				q = qdisc_leaf(p, clid);
1559 			} else if (dev_ingress_queue_create(dev)) {
1560 				q = dev_ingress_queue(dev)->qdisc_sleeping;
1561 			}
1562 		} else {
1563 			q = rtnl_dereference(dev->qdisc);
1564 		}
1565 
1566 		/* It may be default qdisc, ignore it */
1567 		if (q && q->handle == 0)
1568 			q = NULL;
1569 
1570 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1571 			if (tcm->tcm_handle) {
1572 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1573 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1574 					return -EEXIST;
1575 				}
1576 				if (TC_H_MIN(tcm->tcm_handle)) {
1577 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1578 					return -EINVAL;
1579 				}
1580 				q = qdisc_lookup(dev, tcm->tcm_handle);
1581 				if (!q)
1582 					goto create_n_graft;
1583 				if (n->nlmsg_flags & NLM_F_EXCL) {
1584 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1585 					return -EEXIST;
1586 				}
1587 				if (tca[TCA_KIND] &&
1588 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1589 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1590 					return -EINVAL;
1591 				}
1592 				if (q == p ||
1593 				    (p && check_loop(q, p, 0))) {
1594 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1595 					return -ELOOP;
1596 				}
1597 				qdisc_refcount_inc(q);
1598 				goto graft;
1599 			} else {
1600 				if (!q)
1601 					goto create_n_graft;
1602 
1603 				/* This magic test requires explanation.
1604 				 *
1605 				 *   We know, that some child q is already
1606 				 *   attached to this parent and have choice:
1607 				 *   either to change it or to create/graft new one.
1608 				 *
1609 				 *   1. We are allowed to create/graft only
1610 				 *   if CREATE and REPLACE flags are set.
1611 				 *
1612 				 *   2. If EXCL is set, requestor wanted to say,
1613 				 *   that qdisc tcm_handle is not expected
1614 				 *   to exist, so that we choose create/graft too.
1615 				 *
1616 				 *   3. The last case is when no flags are set.
1617 				 *   Alas, it is sort of hole in API, we
1618 				 *   cannot decide what to do unambiguously.
1619 				 *   For now we select create/graft, if
1620 				 *   user gave KIND, which does not match existing.
1621 				 */
1622 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1623 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1624 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1625 				     (tca[TCA_KIND] &&
1626 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1627 					goto create_n_graft;
1628 			}
1629 		}
1630 	} else {
1631 		if (!tcm->tcm_handle) {
1632 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1633 			return -EINVAL;
1634 		}
1635 		q = qdisc_lookup(dev, tcm->tcm_handle);
1636 	}
1637 
1638 	/* Change qdisc parameters */
1639 	if (!q) {
1640 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1641 		return -ENOENT;
1642 	}
1643 	if (n->nlmsg_flags & NLM_F_EXCL) {
1644 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1645 		return -EEXIST;
1646 	}
1647 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1648 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1649 		return -EINVAL;
1650 	}
1651 	err = qdisc_change(q, tca, extack);
1652 	if (err == 0)
1653 		qdisc_notify(net, skb, n, clid, NULL, q, extack);
1654 	return err;
1655 
1656 create_n_graft:
1657 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1658 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1659 		return -ENOENT;
1660 	}
1661 	if (clid == TC_H_INGRESS) {
1662 		if (dev_ingress_queue(dev)) {
1663 			q = qdisc_create(dev, dev_ingress_queue(dev),
1664 					 tcm->tcm_parent, tcm->tcm_parent,
1665 					 tca, &err, extack);
1666 		} else {
1667 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1668 			err = -ENOENT;
1669 		}
1670 	} else {
1671 		struct netdev_queue *dev_queue;
1672 
1673 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1674 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1675 		else if (p)
1676 			dev_queue = p->dev_queue;
1677 		else
1678 			dev_queue = netdev_get_tx_queue(dev, 0);
1679 
1680 		q = qdisc_create(dev, dev_queue,
1681 				 tcm->tcm_parent, tcm->tcm_handle,
1682 				 tca, &err, extack);
1683 	}
1684 	if (q == NULL) {
1685 		if (err == -EAGAIN)
1686 			goto replay;
1687 		return err;
1688 	}
1689 
1690 graft:
1691 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1692 	if (err) {
1693 		if (q)
1694 			qdisc_put(q);
1695 		return err;
1696 	}
1697 
1698 	return 0;
1699 }
1700 
1701 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1702 			      struct netlink_callback *cb,
1703 			      int *q_idx_p, int s_q_idx, bool recur,
1704 			      bool dump_invisible)
1705 {
1706 	int ret = 0, q_idx = *q_idx_p;
1707 	struct Qdisc *q;
1708 	int b;
1709 
1710 	if (!root)
1711 		return 0;
1712 
1713 	q = root;
1714 	if (q_idx < s_q_idx) {
1715 		q_idx++;
1716 	} else {
1717 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1718 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1719 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1720 				  RTM_NEWQDISC, NULL) <= 0)
1721 			goto done;
1722 		q_idx++;
1723 	}
1724 
1725 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1726 	 * itself has already been dumped.
1727 	 *
1728 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1729 	 * qdisc hashtable, we don't want to hit it again
1730 	 */
1731 	if (!qdisc_dev(root) || !recur)
1732 		goto out;
1733 
1734 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1735 		if (q_idx < s_q_idx) {
1736 			q_idx++;
1737 			continue;
1738 		}
1739 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1740 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1741 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1742 				  RTM_NEWQDISC, NULL) <= 0)
1743 			goto done;
1744 		q_idx++;
1745 	}
1746 
1747 out:
1748 	*q_idx_p = q_idx;
1749 	return ret;
1750 done:
1751 	ret = -1;
1752 	goto out;
1753 }
1754 
1755 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1756 {
1757 	struct net *net = sock_net(skb->sk);
1758 	int idx, q_idx;
1759 	int s_idx, s_q_idx;
1760 	struct net_device *dev;
1761 	const struct nlmsghdr *nlh = cb->nlh;
1762 	struct nlattr *tca[TCA_MAX + 1];
1763 	int err;
1764 
1765 	s_idx = cb->args[0];
1766 	s_q_idx = q_idx = cb->args[1];
1767 
1768 	idx = 0;
1769 	ASSERT_RTNL();
1770 
1771 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1772 				     rtm_tca_policy, cb->extack);
1773 	if (err < 0)
1774 		return err;
1775 
1776 	for_each_netdev(net, dev) {
1777 		struct netdev_queue *dev_queue;
1778 
1779 		if (idx < s_idx)
1780 			goto cont;
1781 		if (idx > s_idx)
1782 			s_q_idx = 0;
1783 		q_idx = 0;
1784 
1785 		if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1786 				       skb, cb, &q_idx, s_q_idx,
1787 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1788 			goto done;
1789 
1790 		dev_queue = dev_ingress_queue(dev);
1791 		if (dev_queue &&
1792 		    tc_dump_qdisc_root(dev_queue->qdisc_sleeping, skb, cb,
1793 				       &q_idx, s_q_idx, false,
1794 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1795 			goto done;
1796 
1797 cont:
1798 		idx++;
1799 	}
1800 
1801 done:
1802 	cb->args[0] = idx;
1803 	cb->args[1] = q_idx;
1804 
1805 	return skb->len;
1806 }
1807 
1808 
1809 
1810 /************************************************
1811  *	Traffic classes manipulation.		*
1812  ************************************************/
1813 
1814 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1815 			  unsigned long cl, u32 portid, u32 seq, u16 flags,
1816 			  int event, struct netlink_ext_ack *extack)
1817 {
1818 	struct tcmsg *tcm;
1819 	struct nlmsghdr  *nlh;
1820 	unsigned char *b = skb_tail_pointer(skb);
1821 	struct gnet_dump d;
1822 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1823 
1824 	cond_resched();
1825 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1826 	if (!nlh)
1827 		goto out_nlmsg_trim;
1828 	tcm = nlmsg_data(nlh);
1829 	tcm->tcm_family = AF_UNSPEC;
1830 	tcm->tcm__pad1 = 0;
1831 	tcm->tcm__pad2 = 0;
1832 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1833 	tcm->tcm_parent = q->handle;
1834 	tcm->tcm_handle = q->handle;
1835 	tcm->tcm_info = 0;
1836 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1837 		goto nla_put_failure;
1838 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1839 		goto nla_put_failure;
1840 
1841 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1842 					 NULL, &d, TCA_PAD) < 0)
1843 		goto nla_put_failure;
1844 
1845 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1846 		goto nla_put_failure;
1847 
1848 	if (gnet_stats_finish_copy(&d) < 0)
1849 		goto nla_put_failure;
1850 
1851 	if (extack && extack->_msg &&
1852 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
1853 		goto out_nlmsg_trim;
1854 
1855 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1856 
1857 	return skb->len;
1858 
1859 out_nlmsg_trim:
1860 nla_put_failure:
1861 	nlmsg_trim(skb, b);
1862 	return -1;
1863 }
1864 
1865 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1866 			 struct nlmsghdr *n, struct Qdisc *q,
1867 			 unsigned long cl, int event, struct netlink_ext_ack *extack)
1868 {
1869 	struct sk_buff *skb;
1870 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1871 
1872 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1873 	if (!skb)
1874 		return -ENOBUFS;
1875 
1876 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event, extack) < 0) {
1877 		kfree_skb(skb);
1878 		return -EINVAL;
1879 	}
1880 
1881 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1882 			      n->nlmsg_flags & NLM_F_ECHO);
1883 }
1884 
1885 static int tclass_del_notify(struct net *net,
1886 			     const struct Qdisc_class_ops *cops,
1887 			     struct sk_buff *oskb, struct nlmsghdr *n,
1888 			     struct Qdisc *q, unsigned long cl,
1889 			     struct netlink_ext_ack *extack)
1890 {
1891 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1892 	struct sk_buff *skb;
1893 	int err = 0;
1894 
1895 	if (!cops->delete)
1896 		return -EOPNOTSUPP;
1897 
1898 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1899 	if (!skb)
1900 		return -ENOBUFS;
1901 
1902 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1903 			   RTM_DELTCLASS, extack) < 0) {
1904 		kfree_skb(skb);
1905 		return -EINVAL;
1906 	}
1907 
1908 	err = cops->delete(q, cl, extack);
1909 	if (err) {
1910 		kfree_skb(skb);
1911 		return err;
1912 	}
1913 
1914 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1915 			     n->nlmsg_flags & NLM_F_ECHO);
1916 	return err;
1917 }
1918 
1919 #ifdef CONFIG_NET_CLS
1920 
1921 struct tcf_bind_args {
1922 	struct tcf_walker w;
1923 	unsigned long base;
1924 	unsigned long cl;
1925 	u32 classid;
1926 };
1927 
1928 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1929 {
1930 	struct tcf_bind_args *a = (void *)arg;
1931 
1932 	if (n && tp->ops->bind_class) {
1933 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1934 
1935 		sch_tree_lock(q);
1936 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1937 		sch_tree_unlock(q);
1938 	}
1939 	return 0;
1940 }
1941 
1942 struct tc_bind_class_args {
1943 	struct qdisc_walker w;
1944 	unsigned long new_cl;
1945 	u32 portid;
1946 	u32 clid;
1947 };
1948 
1949 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1950 				struct qdisc_walker *w)
1951 {
1952 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1953 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1954 	struct tcf_block *block;
1955 	struct tcf_chain *chain;
1956 
1957 	block = cops->tcf_block(q, cl, NULL);
1958 	if (!block)
1959 		return 0;
1960 	for (chain = tcf_get_next_chain(block, NULL);
1961 	     chain;
1962 	     chain = tcf_get_next_chain(block, chain)) {
1963 		struct tcf_proto *tp;
1964 
1965 		for (tp = tcf_get_next_proto(chain, NULL);
1966 		     tp; tp = tcf_get_next_proto(chain, tp)) {
1967 			struct tcf_bind_args arg = {};
1968 
1969 			arg.w.fn = tcf_node_bind;
1970 			arg.classid = a->clid;
1971 			arg.base = cl;
1972 			arg.cl = a->new_cl;
1973 			tp->ops->walk(tp, &arg.w, true);
1974 		}
1975 	}
1976 
1977 	return 0;
1978 }
1979 
1980 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1981 			   unsigned long new_cl)
1982 {
1983 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1984 	struct tc_bind_class_args args = {};
1985 
1986 	if (!cops->tcf_block)
1987 		return;
1988 	args.portid = portid;
1989 	args.clid = clid;
1990 	args.new_cl = new_cl;
1991 	args.w.fn = tc_bind_class_walker;
1992 	q->ops->cl_ops->walk(q, &args.w);
1993 }
1994 
1995 #else
1996 
1997 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
1998 			   unsigned long new_cl)
1999 {
2000 }
2001 
2002 #endif
2003 
2004 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
2005 			 struct netlink_ext_ack *extack)
2006 {
2007 	struct net *net = sock_net(skb->sk);
2008 	struct tcmsg *tcm = nlmsg_data(n);
2009 	struct nlattr *tca[TCA_MAX + 1];
2010 	struct net_device *dev;
2011 	struct Qdisc *q = NULL;
2012 	const struct Qdisc_class_ops *cops;
2013 	unsigned long cl = 0;
2014 	unsigned long new_cl;
2015 	u32 portid;
2016 	u32 clid;
2017 	u32 qid;
2018 	int err;
2019 
2020 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2021 				     rtm_tca_policy, extack);
2022 	if (err < 0)
2023 		return err;
2024 
2025 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2026 	if (!dev)
2027 		return -ENODEV;
2028 
2029 	/*
2030 	   parent == TC_H_UNSPEC - unspecified parent.
2031 	   parent == TC_H_ROOT   - class is root, which has no parent.
2032 	   parent == X:0	 - parent is root class.
2033 	   parent == X:Y	 - parent is a node in hierarchy.
2034 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2035 
2036 	   handle == 0:0	 - generate handle from kernel pool.
2037 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2038 	   handle == X:Y	 - clear.
2039 	   handle == X:0	 - root class.
2040 	 */
2041 
2042 	/* Step 1. Determine qdisc handle X:0 */
2043 
2044 	portid = tcm->tcm_parent;
2045 	clid = tcm->tcm_handle;
2046 	qid = TC_H_MAJ(clid);
2047 
2048 	if (portid != TC_H_ROOT) {
2049 		u32 qid1 = TC_H_MAJ(portid);
2050 
2051 		if (qid && qid1) {
2052 			/* If both majors are known, they must be identical. */
2053 			if (qid != qid1)
2054 				return -EINVAL;
2055 		} else if (qid1) {
2056 			qid = qid1;
2057 		} else if (qid == 0)
2058 			qid = rtnl_dereference(dev->qdisc)->handle;
2059 
2060 		/* Now qid is genuine qdisc handle consistent
2061 		 * both with parent and child.
2062 		 *
2063 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2064 		 */
2065 		if (portid)
2066 			portid = TC_H_MAKE(qid, portid);
2067 	} else {
2068 		if (qid == 0)
2069 			qid = rtnl_dereference(dev->qdisc)->handle;
2070 	}
2071 
2072 	/* OK. Locate qdisc */
2073 	q = qdisc_lookup(dev, qid);
2074 	if (!q)
2075 		return -ENOENT;
2076 
2077 	/* An check that it supports classes */
2078 	cops = q->ops->cl_ops;
2079 	if (cops == NULL)
2080 		return -EINVAL;
2081 
2082 	/* Now try to get class */
2083 	if (clid == 0) {
2084 		if (portid == TC_H_ROOT)
2085 			clid = qid;
2086 	} else
2087 		clid = TC_H_MAKE(qid, clid);
2088 
2089 	if (clid)
2090 		cl = cops->find(q, clid);
2091 
2092 	if (cl == 0) {
2093 		err = -ENOENT;
2094 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2095 		    !(n->nlmsg_flags & NLM_F_CREATE))
2096 			goto out;
2097 	} else {
2098 		switch (n->nlmsg_type) {
2099 		case RTM_NEWTCLASS:
2100 			err = -EEXIST;
2101 			if (n->nlmsg_flags & NLM_F_EXCL)
2102 				goto out;
2103 			break;
2104 		case RTM_DELTCLASS:
2105 			err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2106 			/* Unbind the class with flilters with 0 */
2107 			tc_bind_tclass(q, portid, clid, 0);
2108 			goto out;
2109 		case RTM_GETTCLASS:
2110 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS, extack);
2111 			goto out;
2112 		default:
2113 			err = -EINVAL;
2114 			goto out;
2115 		}
2116 	}
2117 
2118 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2119 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2120 		return -EOPNOTSUPP;
2121 	}
2122 
2123 	new_cl = cl;
2124 	err = -EOPNOTSUPP;
2125 	if (cops->change)
2126 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2127 	if (err == 0) {
2128 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS, extack);
2129 		/* We just create a new class, need to do reverse binding. */
2130 		if (cl != new_cl)
2131 			tc_bind_tclass(q, portid, clid, new_cl);
2132 	}
2133 out:
2134 	return err;
2135 }
2136 
2137 struct qdisc_dump_args {
2138 	struct qdisc_walker	w;
2139 	struct sk_buff		*skb;
2140 	struct netlink_callback	*cb;
2141 };
2142 
2143 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2144 			    struct qdisc_walker *arg)
2145 {
2146 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2147 
2148 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2149 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2150 			      RTM_NEWTCLASS, NULL);
2151 }
2152 
2153 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2154 				struct tcmsg *tcm, struct netlink_callback *cb,
2155 				int *t_p, int s_t)
2156 {
2157 	struct qdisc_dump_args arg;
2158 
2159 	if (tc_qdisc_dump_ignore(q, false) ||
2160 	    *t_p < s_t || !q->ops->cl_ops ||
2161 	    (tcm->tcm_parent &&
2162 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2163 		(*t_p)++;
2164 		return 0;
2165 	}
2166 	if (*t_p > s_t)
2167 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2168 	arg.w.fn = qdisc_class_dump;
2169 	arg.skb = skb;
2170 	arg.cb = cb;
2171 	arg.w.stop  = 0;
2172 	arg.w.skip = cb->args[1];
2173 	arg.w.count = 0;
2174 	q->ops->cl_ops->walk(q, &arg.w);
2175 	cb->args[1] = arg.w.count;
2176 	if (arg.w.stop)
2177 		return -1;
2178 	(*t_p)++;
2179 	return 0;
2180 }
2181 
2182 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2183 			       struct tcmsg *tcm, struct netlink_callback *cb,
2184 			       int *t_p, int s_t, bool recur)
2185 {
2186 	struct Qdisc *q;
2187 	int b;
2188 
2189 	if (!root)
2190 		return 0;
2191 
2192 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2193 		return -1;
2194 
2195 	if (!qdisc_dev(root) || !recur)
2196 		return 0;
2197 
2198 	if (tcm->tcm_parent) {
2199 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2200 		if (q && q != root &&
2201 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2202 			return -1;
2203 		return 0;
2204 	}
2205 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2206 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2207 			return -1;
2208 	}
2209 
2210 	return 0;
2211 }
2212 
2213 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2214 {
2215 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2216 	struct net *net = sock_net(skb->sk);
2217 	struct netdev_queue *dev_queue;
2218 	struct net_device *dev;
2219 	int t, s_t;
2220 
2221 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2222 		return 0;
2223 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2224 	if (!dev)
2225 		return 0;
2226 
2227 	s_t = cb->args[0];
2228 	t = 0;
2229 
2230 	if (tc_dump_tclass_root(rtnl_dereference(dev->qdisc),
2231 				skb, tcm, cb, &t, s_t, true) < 0)
2232 		goto done;
2233 
2234 	dev_queue = dev_ingress_queue(dev);
2235 	if (dev_queue &&
2236 	    tc_dump_tclass_root(dev_queue->qdisc_sleeping, skb, tcm, cb,
2237 				&t, s_t, false) < 0)
2238 		goto done;
2239 
2240 done:
2241 	cb->args[0] = t;
2242 
2243 	dev_put(dev);
2244 	return skb->len;
2245 }
2246 
2247 #ifdef CONFIG_PROC_FS
2248 static int psched_show(struct seq_file *seq, void *v)
2249 {
2250 	seq_printf(seq, "%08x %08x %08x %08x\n",
2251 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2252 		   1000000,
2253 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2254 
2255 	return 0;
2256 }
2257 
2258 static int __net_init psched_net_init(struct net *net)
2259 {
2260 	struct proc_dir_entry *e;
2261 
2262 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2263 	if (e == NULL)
2264 		return -ENOMEM;
2265 
2266 	return 0;
2267 }
2268 
2269 static void __net_exit psched_net_exit(struct net *net)
2270 {
2271 	remove_proc_entry("psched", net->proc_net);
2272 }
2273 #else
2274 static int __net_init psched_net_init(struct net *net)
2275 {
2276 	return 0;
2277 }
2278 
2279 static void __net_exit psched_net_exit(struct net *net)
2280 {
2281 }
2282 #endif
2283 
2284 static struct pernet_operations psched_net_ops = {
2285 	.init = psched_net_init,
2286 	.exit = psched_net_exit,
2287 };
2288 
2289 DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper);
2290 
2291 static int __init pktsched_init(void)
2292 {
2293 	int err;
2294 
2295 	err = register_pernet_subsys(&psched_net_ops);
2296 	if (err) {
2297 		pr_err("pktsched_init: "
2298 		       "cannot initialize per netns operations\n");
2299 		return err;
2300 	}
2301 
2302 	register_qdisc(&pfifo_fast_ops);
2303 	register_qdisc(&pfifo_qdisc_ops);
2304 	register_qdisc(&bfifo_qdisc_ops);
2305 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2306 	register_qdisc(&mq_qdisc_ops);
2307 	register_qdisc(&noqueue_qdisc_ops);
2308 
2309 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2310 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2311 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2312 		      0);
2313 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2314 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2315 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2316 		      0);
2317 
2318 	tc_wrapper_init();
2319 
2320 	return 0;
2321 }
2322 
2323 subsys_initcall(pktsched_init);
2324