xref: /openbmc/linux/net/sched/sch_api.c (revision 7e9f2839)
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 			rtnl_dereference(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(rcu_dereference(nq->qdisc_sleeping),
332 					  handle);
333 out:
334 	return q;
335 }
336 
337 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
338 {
339 	unsigned long cl;
340 	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
341 
342 	if (cops == NULL)
343 		return NULL;
344 	cl = cops->find(p, classid);
345 
346 	if (cl == 0)
347 		return NULL;
348 	return cops->leaf(p, cl);
349 }
350 
351 /* Find queueing discipline by name */
352 
353 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
354 {
355 	struct Qdisc_ops *q = NULL;
356 
357 	if (kind) {
358 		read_lock(&qdisc_mod_lock);
359 		for (q = qdisc_base; q; q = q->next) {
360 			if (nla_strcmp(kind, q->id) == 0) {
361 				if (!try_module_get(q->owner))
362 					q = NULL;
363 				break;
364 			}
365 		}
366 		read_unlock(&qdisc_mod_lock);
367 	}
368 	return q;
369 }
370 
371 /* The linklayer setting were not transferred from iproute2, in older
372  * versions, and the rate tables lookup systems have been dropped in
373  * the kernel. To keep backward compatible with older iproute2 tc
374  * utils, we detect the linklayer setting by detecting if the rate
375  * table were modified.
376  *
377  * For linklayer ATM table entries, the rate table will be aligned to
378  * 48 bytes, thus some table entries will contain the same value.  The
379  * mpu (min packet unit) is also encoded into the old rate table, thus
380  * starting from the mpu, we find low and high table entries for
381  * mapping this cell.  If these entries contain the same value, when
382  * the rate tables have been modified for linklayer ATM.
383  *
384  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
385  * and then roundup to the next cell, calc the table entry one below,
386  * and compare.
387  */
388 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
389 {
390 	int low       = roundup(r->mpu, 48);
391 	int high      = roundup(low+1, 48);
392 	int cell_low  = low >> r->cell_log;
393 	int cell_high = (high >> r->cell_log) - 1;
394 
395 	/* rtab is too inaccurate at rates > 100Mbit/s */
396 	if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
397 		pr_debug("TC linklayer: Giving up ATM detection\n");
398 		return TC_LINKLAYER_ETHERNET;
399 	}
400 
401 	if ((cell_high > cell_low) && (cell_high < 256)
402 	    && (rtab[cell_low] == rtab[cell_high])) {
403 		pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
404 			 cell_low, cell_high, rtab[cell_high]);
405 		return TC_LINKLAYER_ATM;
406 	}
407 	return TC_LINKLAYER_ETHERNET;
408 }
409 
410 static struct qdisc_rate_table *qdisc_rtab_list;
411 
412 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
413 					struct nlattr *tab,
414 					struct netlink_ext_ack *extack)
415 {
416 	struct qdisc_rate_table *rtab;
417 
418 	if (tab == NULL || r->rate == 0 ||
419 	    r->cell_log == 0 || r->cell_log >= 32 ||
420 	    nla_len(tab) != TC_RTAB_SIZE) {
421 		NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
422 		return NULL;
423 	}
424 
425 	for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
426 		if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
427 		    !memcmp(&rtab->data, nla_data(tab), 1024)) {
428 			rtab->refcnt++;
429 			return rtab;
430 		}
431 	}
432 
433 	rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
434 	if (rtab) {
435 		rtab->rate = *r;
436 		rtab->refcnt = 1;
437 		memcpy(rtab->data, nla_data(tab), 1024);
438 		if (r->linklayer == TC_LINKLAYER_UNAWARE)
439 			r->linklayer = __detect_linklayer(r, rtab->data);
440 		rtab->next = qdisc_rtab_list;
441 		qdisc_rtab_list = rtab;
442 	} else {
443 		NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
444 	}
445 	return rtab;
446 }
447 EXPORT_SYMBOL(qdisc_get_rtab);
448 
449 void qdisc_put_rtab(struct qdisc_rate_table *tab)
450 {
451 	struct qdisc_rate_table *rtab, **rtabp;
452 
453 	if (!tab || --tab->refcnt)
454 		return;
455 
456 	for (rtabp = &qdisc_rtab_list;
457 	     (rtab = *rtabp) != NULL;
458 	     rtabp = &rtab->next) {
459 		if (rtab == tab) {
460 			*rtabp = rtab->next;
461 			kfree(rtab);
462 			return;
463 		}
464 	}
465 }
466 EXPORT_SYMBOL(qdisc_put_rtab);
467 
468 static LIST_HEAD(qdisc_stab_list);
469 
470 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
471 	[TCA_STAB_BASE]	= { .len = sizeof(struct tc_sizespec) },
472 	[TCA_STAB_DATA] = { .type = NLA_BINARY },
473 };
474 
475 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
476 					       struct netlink_ext_ack *extack)
477 {
478 	struct nlattr *tb[TCA_STAB_MAX + 1];
479 	struct qdisc_size_table *stab;
480 	struct tc_sizespec *s;
481 	unsigned int tsize = 0;
482 	u16 *tab = NULL;
483 	int err;
484 
485 	err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
486 					  extack);
487 	if (err < 0)
488 		return ERR_PTR(err);
489 	if (!tb[TCA_STAB_BASE]) {
490 		NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
491 		return ERR_PTR(-EINVAL);
492 	}
493 
494 	s = nla_data(tb[TCA_STAB_BASE]);
495 
496 	if (s->tsize > 0) {
497 		if (!tb[TCA_STAB_DATA]) {
498 			NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
499 			return ERR_PTR(-EINVAL);
500 		}
501 		tab = nla_data(tb[TCA_STAB_DATA]);
502 		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
503 	}
504 
505 	if (tsize != s->tsize || (!tab && tsize > 0)) {
506 		NL_SET_ERR_MSG(extack, "Invalid size of size table");
507 		return ERR_PTR(-EINVAL);
508 	}
509 
510 	list_for_each_entry(stab, &qdisc_stab_list, list) {
511 		if (memcmp(&stab->szopts, s, sizeof(*s)))
512 			continue;
513 		if (tsize > 0 &&
514 		    memcmp(stab->data, tab, flex_array_size(stab, data, tsize)))
515 			continue;
516 		stab->refcnt++;
517 		return stab;
518 	}
519 
520 	if (s->size_log > STAB_SIZE_LOG_MAX ||
521 	    s->cell_log > STAB_SIZE_LOG_MAX) {
522 		NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
523 		return ERR_PTR(-EINVAL);
524 	}
525 
526 	stab = kmalloc(struct_size(stab, data, tsize), GFP_KERNEL);
527 	if (!stab)
528 		return ERR_PTR(-ENOMEM);
529 
530 	stab->refcnt = 1;
531 	stab->szopts = *s;
532 	if (tsize > 0)
533 		memcpy(stab->data, tab, flex_array_size(stab, data, tsize));
534 
535 	list_add_tail(&stab->list, &qdisc_stab_list);
536 
537 	return stab;
538 }
539 
540 void qdisc_put_stab(struct qdisc_size_table *tab)
541 {
542 	if (!tab)
543 		return;
544 
545 	if (--tab->refcnt == 0) {
546 		list_del(&tab->list);
547 		kfree_rcu(tab, rcu);
548 	}
549 }
550 EXPORT_SYMBOL(qdisc_put_stab);
551 
552 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
553 {
554 	struct nlattr *nest;
555 
556 	nest = nla_nest_start_noflag(skb, TCA_STAB);
557 	if (nest == NULL)
558 		goto nla_put_failure;
559 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
560 		goto nla_put_failure;
561 	nla_nest_end(skb, nest);
562 
563 	return skb->len;
564 
565 nla_put_failure:
566 	return -1;
567 }
568 
569 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
570 			       const struct qdisc_size_table *stab)
571 {
572 	int pkt_len, slot;
573 
574 	pkt_len = skb->len + stab->szopts.overhead;
575 	if (unlikely(!stab->szopts.tsize))
576 		goto out;
577 
578 	slot = pkt_len + stab->szopts.cell_align;
579 	if (unlikely(slot < 0))
580 		slot = 0;
581 
582 	slot >>= stab->szopts.cell_log;
583 	if (likely(slot < stab->szopts.tsize))
584 		pkt_len = stab->data[slot];
585 	else
586 		pkt_len = stab->data[stab->szopts.tsize - 1] *
587 				(slot / stab->szopts.tsize) +
588 				stab->data[slot % stab->szopts.tsize];
589 
590 	pkt_len <<= stab->szopts.size_log;
591 out:
592 	if (unlikely(pkt_len < 1))
593 		pkt_len = 1;
594 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
595 }
596 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
597 
598 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
599 {
600 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
601 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
602 			txt, qdisc->ops->id, qdisc->handle >> 16);
603 		qdisc->flags |= TCQ_F_WARN_NONWC;
604 	}
605 }
606 EXPORT_SYMBOL(qdisc_warn_nonwc);
607 
608 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
609 {
610 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
611 						 timer);
612 
613 	rcu_read_lock();
614 	__netif_schedule(qdisc_root(wd->qdisc));
615 	rcu_read_unlock();
616 
617 	return HRTIMER_NORESTART;
618 }
619 
620 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
621 				 clockid_t clockid)
622 {
623 	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
624 	wd->timer.function = qdisc_watchdog;
625 	wd->qdisc = qdisc;
626 }
627 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
628 
629 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
630 {
631 	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
632 }
633 EXPORT_SYMBOL(qdisc_watchdog_init);
634 
635 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
636 				      u64 delta_ns)
637 {
638 	bool deactivated;
639 
640 	rcu_read_lock();
641 	deactivated = test_bit(__QDISC_STATE_DEACTIVATED,
642 			       &qdisc_root_sleeping(wd->qdisc)->state);
643 	rcu_read_unlock();
644 	if (deactivated)
645 		return;
646 
647 	if (hrtimer_is_queued(&wd->timer)) {
648 		u64 softexpires;
649 
650 		softexpires = ktime_to_ns(hrtimer_get_softexpires(&wd->timer));
651 		/* If timer is already set in [expires, expires + delta_ns],
652 		 * do not reprogram it.
653 		 */
654 		if (softexpires - expires <= delta_ns)
655 			return;
656 	}
657 
658 	hrtimer_start_range_ns(&wd->timer,
659 			       ns_to_ktime(expires),
660 			       delta_ns,
661 			       HRTIMER_MODE_ABS_PINNED);
662 }
663 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
664 
665 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
666 {
667 	hrtimer_cancel(&wd->timer);
668 }
669 EXPORT_SYMBOL(qdisc_watchdog_cancel);
670 
671 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
672 {
673 	struct hlist_head *h;
674 	unsigned int i;
675 
676 	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
677 
678 	if (h != NULL) {
679 		for (i = 0; i < n; i++)
680 			INIT_HLIST_HEAD(&h[i]);
681 	}
682 	return h;
683 }
684 
685 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
686 {
687 	struct Qdisc_class_common *cl;
688 	struct hlist_node *next;
689 	struct hlist_head *nhash, *ohash;
690 	unsigned int nsize, nmask, osize;
691 	unsigned int i, h;
692 
693 	/* Rehash when load factor exceeds 0.75 */
694 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
695 		return;
696 	nsize = clhash->hashsize * 2;
697 	nmask = nsize - 1;
698 	nhash = qdisc_class_hash_alloc(nsize);
699 	if (nhash == NULL)
700 		return;
701 
702 	ohash = clhash->hash;
703 	osize = clhash->hashsize;
704 
705 	sch_tree_lock(sch);
706 	for (i = 0; i < osize; i++) {
707 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
708 			h = qdisc_class_hash(cl->classid, nmask);
709 			hlist_add_head(&cl->hnode, &nhash[h]);
710 		}
711 	}
712 	clhash->hash     = nhash;
713 	clhash->hashsize = nsize;
714 	clhash->hashmask = nmask;
715 	sch_tree_unlock(sch);
716 
717 	kvfree(ohash);
718 }
719 EXPORT_SYMBOL(qdisc_class_hash_grow);
720 
721 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
722 {
723 	unsigned int size = 4;
724 
725 	clhash->hash = qdisc_class_hash_alloc(size);
726 	if (!clhash->hash)
727 		return -ENOMEM;
728 	clhash->hashsize  = size;
729 	clhash->hashmask  = size - 1;
730 	clhash->hashelems = 0;
731 	return 0;
732 }
733 EXPORT_SYMBOL(qdisc_class_hash_init);
734 
735 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
736 {
737 	kvfree(clhash->hash);
738 }
739 EXPORT_SYMBOL(qdisc_class_hash_destroy);
740 
741 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
742 			     struct Qdisc_class_common *cl)
743 {
744 	unsigned int h;
745 
746 	INIT_HLIST_NODE(&cl->hnode);
747 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
748 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
749 	clhash->hashelems++;
750 }
751 EXPORT_SYMBOL(qdisc_class_hash_insert);
752 
753 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
754 			     struct Qdisc_class_common *cl)
755 {
756 	hlist_del(&cl->hnode);
757 	clhash->hashelems--;
758 }
759 EXPORT_SYMBOL(qdisc_class_hash_remove);
760 
761 /* Allocate an unique handle from space managed by kernel
762  * Possible range is [8000-FFFF]:0000 (0x8000 values)
763  */
764 static u32 qdisc_alloc_handle(struct net_device *dev)
765 {
766 	int i = 0x8000;
767 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
768 
769 	do {
770 		autohandle += TC_H_MAKE(0x10000U, 0);
771 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
772 			autohandle = TC_H_MAKE(0x80000000U, 0);
773 		if (!qdisc_lookup(dev, autohandle))
774 			return autohandle;
775 		cond_resched();
776 	} while	(--i > 0);
777 
778 	return 0;
779 }
780 
781 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
782 {
783 	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
784 	const struct Qdisc_class_ops *cops;
785 	unsigned long cl;
786 	u32 parentid;
787 	bool notify;
788 	int drops;
789 
790 	if (n == 0 && len == 0)
791 		return;
792 	drops = max_t(int, n, 0);
793 	rcu_read_lock();
794 	while ((parentid = sch->parent)) {
795 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
796 			break;
797 
798 		if (sch->flags & TCQ_F_NOPARENT)
799 			break;
800 		/* Notify parent qdisc only if child qdisc becomes empty.
801 		 *
802 		 * If child was empty even before update then backlog
803 		 * counter is screwed and we skip notification because
804 		 * parent class is already passive.
805 		 *
806 		 * If the original child was offloaded then it is allowed
807 		 * to be seem as empty, so the parent is notified anyway.
808 		 */
809 		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
810 						       !qdisc_is_offloaded);
811 		/* TODO: perform the search on a per txq basis */
812 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
813 		if (sch == NULL) {
814 			WARN_ON_ONCE(parentid != TC_H_ROOT);
815 			break;
816 		}
817 		cops = sch->ops->cl_ops;
818 		if (notify && cops->qlen_notify) {
819 			cl = cops->find(sch, parentid);
820 			cops->qlen_notify(sch, cl);
821 		}
822 		sch->q.qlen -= n;
823 		sch->qstats.backlog -= len;
824 		__qdisc_qstats_drop(sch, drops);
825 	}
826 	rcu_read_unlock();
827 }
828 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
829 
830 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
831 			      void *type_data)
832 {
833 	struct net_device *dev = qdisc_dev(sch);
834 	int err;
835 
836 	sch->flags &= ~TCQ_F_OFFLOADED;
837 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
838 		return 0;
839 
840 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
841 	if (err == -EOPNOTSUPP)
842 		return 0;
843 
844 	if (!err)
845 		sch->flags |= TCQ_F_OFFLOADED;
846 
847 	return err;
848 }
849 EXPORT_SYMBOL(qdisc_offload_dump_helper);
850 
851 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
852 				struct Qdisc *new, struct Qdisc *old,
853 				enum tc_setup_type type, void *type_data,
854 				struct netlink_ext_ack *extack)
855 {
856 	bool any_qdisc_is_offloaded;
857 	int err;
858 
859 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
860 		return;
861 
862 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
863 
864 	/* Don't report error if the graft is part of destroy operation. */
865 	if (!err || !new || new == &noop_qdisc)
866 		return;
867 
868 	/* Don't report error if the parent, the old child and the new
869 	 * one are not offloaded.
870 	 */
871 	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
872 	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
873 	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
874 
875 	if (any_qdisc_is_offloaded)
876 		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
877 }
878 EXPORT_SYMBOL(qdisc_offload_graft_helper);
879 
880 void qdisc_offload_query_caps(struct net_device *dev,
881 			      enum tc_setup_type type,
882 			      void *caps, size_t caps_len)
883 {
884 	const struct net_device_ops *ops = dev->netdev_ops;
885 	struct tc_query_caps_base base = {
886 		.type = type,
887 		.caps = caps,
888 	};
889 
890 	memset(caps, 0, caps_len);
891 
892 	if (ops->ndo_setup_tc)
893 		ops->ndo_setup_tc(dev, TC_QUERY_CAPS, &base);
894 }
895 EXPORT_SYMBOL(qdisc_offload_query_caps);
896 
897 static void qdisc_offload_graft_root(struct net_device *dev,
898 				     struct Qdisc *new, struct Qdisc *old,
899 				     struct netlink_ext_ack *extack)
900 {
901 	struct tc_root_qopt_offload graft_offload = {
902 		.command	= TC_ROOT_GRAFT,
903 		.handle		= new ? new->handle : 0,
904 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
905 				  (old && old->flags & TCQ_F_INGRESS),
906 	};
907 
908 	qdisc_offload_graft_helper(dev, NULL, new, old,
909 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
910 }
911 
912 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
913 			 u32 portid, u32 seq, u16 flags, int event,
914 			 struct netlink_ext_ack *extack)
915 {
916 	struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
917 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
918 	struct tcmsg *tcm;
919 	struct nlmsghdr  *nlh;
920 	unsigned char *b = skb_tail_pointer(skb);
921 	struct gnet_dump d;
922 	struct qdisc_size_table *stab;
923 	u32 block_index;
924 	__u32 qlen;
925 
926 	cond_resched();
927 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
928 	if (!nlh)
929 		goto out_nlmsg_trim;
930 	tcm = nlmsg_data(nlh);
931 	tcm->tcm_family = AF_UNSPEC;
932 	tcm->tcm__pad1 = 0;
933 	tcm->tcm__pad2 = 0;
934 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
935 	tcm->tcm_parent = clid;
936 	tcm->tcm_handle = q->handle;
937 	tcm->tcm_info = refcount_read(&q->refcnt);
938 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
939 		goto nla_put_failure;
940 	if (q->ops->ingress_block_get) {
941 		block_index = q->ops->ingress_block_get(q);
942 		if (block_index &&
943 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
944 			goto nla_put_failure;
945 	}
946 	if (q->ops->egress_block_get) {
947 		block_index = q->ops->egress_block_get(q);
948 		if (block_index &&
949 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
950 			goto nla_put_failure;
951 	}
952 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
953 		goto nla_put_failure;
954 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
955 		goto nla_put_failure;
956 	qlen = qdisc_qlen_sum(q);
957 
958 	stab = rtnl_dereference(q->stab);
959 	if (stab && qdisc_dump_stab(skb, stab) < 0)
960 		goto nla_put_failure;
961 
962 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
963 					 NULL, &d, TCA_PAD) < 0)
964 		goto nla_put_failure;
965 
966 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
967 		goto nla_put_failure;
968 
969 	if (qdisc_is_percpu_stats(q)) {
970 		cpu_bstats = q->cpu_bstats;
971 		cpu_qstats = q->cpu_qstats;
972 	}
973 
974 	if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
975 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
976 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
977 		goto nla_put_failure;
978 
979 	if (gnet_stats_finish_copy(&d) < 0)
980 		goto nla_put_failure;
981 
982 	if (extack && extack->_msg &&
983 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
984 		goto out_nlmsg_trim;
985 
986 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
987 
988 	return skb->len;
989 
990 out_nlmsg_trim:
991 nla_put_failure:
992 	nlmsg_trim(skb, b);
993 	return -1;
994 }
995 
996 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
997 {
998 	if (q->flags & TCQ_F_BUILTIN)
999 		return true;
1000 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
1001 		return true;
1002 
1003 	return false;
1004 }
1005 
1006 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1007 			struct nlmsghdr *n, u32 clid,
1008 			struct Qdisc *old, struct Qdisc *new,
1009 			struct netlink_ext_ack *extack)
1010 {
1011 	struct sk_buff *skb;
1012 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1013 
1014 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1015 	if (!skb)
1016 		return -ENOBUFS;
1017 
1018 	if (old && !tc_qdisc_dump_ignore(old, false)) {
1019 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1020 				  0, RTM_DELQDISC, extack) < 0)
1021 			goto err_out;
1022 	}
1023 	if (new && !tc_qdisc_dump_ignore(new, false)) {
1024 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1025 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC, extack) < 0)
1026 			goto err_out;
1027 	}
1028 
1029 	if (skb->len)
1030 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1031 				      n->nlmsg_flags & NLM_F_ECHO);
1032 
1033 err_out:
1034 	kfree_skb(skb);
1035 	return -EINVAL;
1036 }
1037 
1038 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1039 			       struct nlmsghdr *n, u32 clid,
1040 			       struct Qdisc *old, struct Qdisc *new,
1041 			       struct netlink_ext_ack *extack)
1042 {
1043 	if (new || old)
1044 		qdisc_notify(net, skb, n, clid, old, new, extack);
1045 
1046 	if (old)
1047 		qdisc_put(old);
1048 }
1049 
1050 static void qdisc_clear_nolock(struct Qdisc *sch)
1051 {
1052 	sch->flags &= ~TCQ_F_NOLOCK;
1053 	if (!(sch->flags & TCQ_F_CPUSTATS))
1054 		return;
1055 
1056 	free_percpu(sch->cpu_bstats);
1057 	free_percpu(sch->cpu_qstats);
1058 	sch->cpu_bstats = NULL;
1059 	sch->cpu_qstats = NULL;
1060 	sch->flags &= ~TCQ_F_CPUSTATS;
1061 }
1062 
1063 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1064  * to device "dev".
1065  *
1066  * When appropriate send a netlink notification using 'skb'
1067  * and "n".
1068  *
1069  * On success, destroy old qdisc.
1070  */
1071 
1072 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1073 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1074 		       struct Qdisc *new, struct Qdisc *old,
1075 		       struct netlink_ext_ack *extack)
1076 {
1077 	struct Qdisc *q = old;
1078 	struct net *net = dev_net(dev);
1079 
1080 	if (parent == NULL) {
1081 		unsigned int i, num_q, ingress;
1082 		struct netdev_queue *dev_queue;
1083 
1084 		ingress = 0;
1085 		num_q = dev->num_tx_queues;
1086 		if ((q && q->flags & TCQ_F_INGRESS) ||
1087 		    (new && new->flags & TCQ_F_INGRESS)) {
1088 			ingress = 1;
1089 			dev_queue = dev_ingress_queue(dev);
1090 			if (!dev_queue) {
1091 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1092 				return -ENOENT;
1093 			}
1094 
1095 			q = rtnl_dereference(dev_queue->qdisc_sleeping);
1096 
1097 			/* This is the counterpart of that qdisc_refcount_inc_nz() call in
1098 			 * __tcf_qdisc_find() for filter requests.
1099 			 */
1100 			if (!qdisc_refcount_dec_if_one(q)) {
1101 				NL_SET_ERR_MSG(extack,
1102 					       "Current ingress or clsact Qdisc has ongoing filter requests");
1103 				return -EBUSY;
1104 			}
1105 		}
1106 
1107 		if (dev->flags & IFF_UP)
1108 			dev_deactivate(dev);
1109 
1110 		qdisc_offload_graft_root(dev, new, old, extack);
1111 
1112 		if (new && new->ops->attach && !ingress)
1113 			goto skip;
1114 
1115 		if (!ingress) {
1116 			for (i = 0; i < num_q; i++) {
1117 				dev_queue = netdev_get_tx_queue(dev, i);
1118 				old = dev_graft_qdisc(dev_queue, new);
1119 
1120 				if (new && i > 0)
1121 					qdisc_refcount_inc(new);
1122 				qdisc_put(old);
1123 			}
1124 		} else {
1125 			old = dev_graft_qdisc(dev_queue, NULL);
1126 
1127 			/* {ingress,clsact}_destroy() @old before grafting @new to avoid
1128 			 * unprotected concurrent accesses to net_device::miniq_{in,e}gress
1129 			 * pointer(s) in mini_qdisc_pair_swap().
1130 			 */
1131 			qdisc_notify(net, skb, n, classid, old, new, extack);
1132 			qdisc_destroy(old);
1133 
1134 			dev_graft_qdisc(dev_queue, new);
1135 		}
1136 
1137 skip:
1138 		if (!ingress) {
1139 			old = rtnl_dereference(dev->qdisc);
1140 			if (new && !new->ops->attach)
1141 				qdisc_refcount_inc(new);
1142 			rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1143 
1144 			notify_and_destroy(net, skb, n, classid, old, new, extack);
1145 
1146 			if (new && new->ops->attach)
1147 				new->ops->attach(new);
1148 		}
1149 
1150 		if (dev->flags & IFF_UP)
1151 			dev_activate(dev);
1152 	} else {
1153 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1154 		unsigned long cl;
1155 		int err;
1156 
1157 		/* Only support running class lockless if parent is lockless */
1158 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1159 			qdisc_clear_nolock(new);
1160 
1161 		if (!cops || !cops->graft)
1162 			return -EOPNOTSUPP;
1163 
1164 		cl = cops->find(parent, classid);
1165 		if (!cl) {
1166 			NL_SET_ERR_MSG(extack, "Specified class not found");
1167 			return -ENOENT;
1168 		}
1169 
1170 		if (new && new->ops == &noqueue_qdisc_ops) {
1171 			NL_SET_ERR_MSG(extack, "Cannot assign noqueue to a class");
1172 			return -EINVAL;
1173 		}
1174 
1175 		err = cops->graft(parent, cl, new, &old, extack);
1176 		if (err)
1177 			return err;
1178 		notify_and_destroy(net, skb, n, classid, old, new, extack);
1179 	}
1180 	return 0;
1181 }
1182 
1183 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1184 				   struct netlink_ext_ack *extack)
1185 {
1186 	u32 block_index;
1187 
1188 	if (tca[TCA_INGRESS_BLOCK]) {
1189 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1190 
1191 		if (!block_index) {
1192 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1193 			return -EINVAL;
1194 		}
1195 		if (!sch->ops->ingress_block_set) {
1196 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1197 			return -EOPNOTSUPP;
1198 		}
1199 		sch->ops->ingress_block_set(sch, block_index);
1200 	}
1201 	if (tca[TCA_EGRESS_BLOCK]) {
1202 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1203 
1204 		if (!block_index) {
1205 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1206 			return -EINVAL;
1207 		}
1208 		if (!sch->ops->egress_block_set) {
1209 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1210 			return -EOPNOTSUPP;
1211 		}
1212 		sch->ops->egress_block_set(sch, block_index);
1213 	}
1214 	return 0;
1215 }
1216 
1217 /*
1218    Allocate and initialize new qdisc.
1219 
1220    Parameters are passed via opt.
1221  */
1222 
1223 static struct Qdisc *qdisc_create(struct net_device *dev,
1224 				  struct netdev_queue *dev_queue,
1225 				  u32 parent, u32 handle,
1226 				  struct nlattr **tca, int *errp,
1227 				  struct netlink_ext_ack *extack)
1228 {
1229 	int err;
1230 	struct nlattr *kind = tca[TCA_KIND];
1231 	struct Qdisc *sch;
1232 	struct Qdisc_ops *ops;
1233 	struct qdisc_size_table *stab;
1234 
1235 	ops = qdisc_lookup_ops(kind);
1236 #ifdef CONFIG_MODULES
1237 	if (ops == NULL && kind != NULL) {
1238 		char name[IFNAMSIZ];
1239 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1240 			/* We dropped the RTNL semaphore in order to
1241 			 * perform the module load.  So, even if we
1242 			 * succeeded in loading the module we have to
1243 			 * tell the caller to replay the request.  We
1244 			 * indicate this using -EAGAIN.
1245 			 * We replay the request because the device may
1246 			 * go away in the mean time.
1247 			 */
1248 			rtnl_unlock();
1249 			request_module("sch_%s", name);
1250 			rtnl_lock();
1251 			ops = qdisc_lookup_ops(kind);
1252 			if (ops != NULL) {
1253 				/* We will try again qdisc_lookup_ops,
1254 				 * so don't keep a reference.
1255 				 */
1256 				module_put(ops->owner);
1257 				err = -EAGAIN;
1258 				goto err_out;
1259 			}
1260 		}
1261 	}
1262 #endif
1263 
1264 	err = -ENOENT;
1265 	if (!ops) {
1266 		NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1267 		goto err_out;
1268 	}
1269 
1270 	sch = qdisc_alloc(dev_queue, ops, extack);
1271 	if (IS_ERR(sch)) {
1272 		err = PTR_ERR(sch);
1273 		goto err_out2;
1274 	}
1275 
1276 	sch->parent = parent;
1277 
1278 	if (handle == TC_H_INGRESS) {
1279 		if (!(sch->flags & TCQ_F_INGRESS)) {
1280 			NL_SET_ERR_MSG(extack,
1281 				       "Specified parent ID is reserved for ingress and clsact Qdiscs");
1282 			err = -EINVAL;
1283 			goto err_out3;
1284 		}
1285 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1286 	} else {
1287 		if (handle == 0) {
1288 			handle = qdisc_alloc_handle(dev);
1289 			if (handle == 0) {
1290 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1291 				err = -ENOSPC;
1292 				goto err_out3;
1293 			}
1294 		}
1295 		if (!netif_is_multiqueue(dev))
1296 			sch->flags |= TCQ_F_ONETXQUEUE;
1297 	}
1298 
1299 	sch->handle = handle;
1300 
1301 	/* This exist to keep backward compatible with a userspace
1302 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1303 	 * facility on older kernels by setting tx_queue_len=0 (prior
1304 	 * to qdisc init), and then forgot to reinit tx_queue_len
1305 	 * before again attaching a qdisc.
1306 	 */
1307 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1308 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1309 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1310 	}
1311 
1312 	err = qdisc_block_indexes_set(sch, tca, extack);
1313 	if (err)
1314 		goto err_out3;
1315 
1316 	if (tca[TCA_STAB]) {
1317 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1318 		if (IS_ERR(stab)) {
1319 			err = PTR_ERR(stab);
1320 			goto err_out3;
1321 		}
1322 		rcu_assign_pointer(sch->stab, stab);
1323 	}
1324 
1325 	if (ops->init) {
1326 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1327 		if (err != 0)
1328 			goto err_out4;
1329 	}
1330 
1331 	if (tca[TCA_RATE]) {
1332 		err = -EOPNOTSUPP;
1333 		if (sch->flags & TCQ_F_MQROOT) {
1334 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1335 			goto err_out4;
1336 		}
1337 
1338 		err = gen_new_estimator(&sch->bstats,
1339 					sch->cpu_bstats,
1340 					&sch->rate_est,
1341 					NULL,
1342 					true,
1343 					tca[TCA_RATE]);
1344 		if (err) {
1345 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1346 			goto err_out4;
1347 		}
1348 	}
1349 
1350 	qdisc_hash_add(sch, false);
1351 	trace_qdisc_create(ops, dev, parent);
1352 
1353 	return sch;
1354 
1355 err_out4:
1356 	/* Even if ops->init() failed, we call ops->destroy()
1357 	 * like qdisc_create_dflt().
1358 	 */
1359 	if (ops->destroy)
1360 		ops->destroy(sch);
1361 	qdisc_put_stab(rtnl_dereference(sch->stab));
1362 err_out3:
1363 	netdev_put(dev, &sch->dev_tracker);
1364 	qdisc_free(sch);
1365 err_out2:
1366 	module_put(ops->owner);
1367 err_out:
1368 	*errp = err;
1369 	return NULL;
1370 }
1371 
1372 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1373 			struct netlink_ext_ack *extack)
1374 {
1375 	struct qdisc_size_table *ostab, *stab = NULL;
1376 	int err = 0;
1377 
1378 	if (tca[TCA_OPTIONS]) {
1379 		if (!sch->ops->change) {
1380 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1381 			return -EINVAL;
1382 		}
1383 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1384 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1385 			return -EOPNOTSUPP;
1386 		}
1387 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1388 		if (err)
1389 			return err;
1390 	}
1391 
1392 	if (tca[TCA_STAB]) {
1393 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1394 		if (IS_ERR(stab))
1395 			return PTR_ERR(stab);
1396 	}
1397 
1398 	ostab = rtnl_dereference(sch->stab);
1399 	rcu_assign_pointer(sch->stab, stab);
1400 	qdisc_put_stab(ostab);
1401 
1402 	if (tca[TCA_RATE]) {
1403 		/* NB: ignores errors from replace_estimator
1404 		   because change can't be undone. */
1405 		if (sch->flags & TCQ_F_MQROOT)
1406 			goto out;
1407 		gen_replace_estimator(&sch->bstats,
1408 				      sch->cpu_bstats,
1409 				      &sch->rate_est,
1410 				      NULL,
1411 				      true,
1412 				      tca[TCA_RATE]);
1413 	}
1414 out:
1415 	return 0;
1416 }
1417 
1418 struct check_loop_arg {
1419 	struct qdisc_walker	w;
1420 	struct Qdisc		*p;
1421 	int			depth;
1422 };
1423 
1424 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1425 			 struct qdisc_walker *w);
1426 
1427 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1428 {
1429 	struct check_loop_arg	arg;
1430 
1431 	if (q->ops->cl_ops == NULL)
1432 		return 0;
1433 
1434 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1435 	arg.w.fn = check_loop_fn;
1436 	arg.depth = depth;
1437 	arg.p = p;
1438 	q->ops->cl_ops->walk(q, &arg.w);
1439 	return arg.w.stop ? -ELOOP : 0;
1440 }
1441 
1442 static int
1443 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1444 {
1445 	struct Qdisc *leaf;
1446 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1447 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1448 
1449 	leaf = cops->leaf(q, cl);
1450 	if (leaf) {
1451 		if (leaf == arg->p || arg->depth > 7)
1452 			return -ELOOP;
1453 		return check_loop(leaf, arg->p, arg->depth + 1);
1454 	}
1455 	return 0;
1456 }
1457 
1458 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1459 	[TCA_KIND]		= { .type = NLA_STRING },
1460 	[TCA_RATE]		= { .type = NLA_BINARY,
1461 				    .len = sizeof(struct tc_estimator) },
1462 	[TCA_STAB]		= { .type = NLA_NESTED },
1463 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1464 	[TCA_CHAIN]		= { .type = NLA_U32 },
1465 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1466 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1467 };
1468 
1469 /*
1470  * Delete/get qdisc.
1471  */
1472 
1473 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1474 			struct netlink_ext_ack *extack)
1475 {
1476 	struct net *net = sock_net(skb->sk);
1477 	struct tcmsg *tcm = nlmsg_data(n);
1478 	struct nlattr *tca[TCA_MAX + 1];
1479 	struct net_device *dev;
1480 	u32 clid;
1481 	struct Qdisc *q = NULL;
1482 	struct Qdisc *p = NULL;
1483 	int err;
1484 
1485 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1486 				     rtm_tca_policy, extack);
1487 	if (err < 0)
1488 		return err;
1489 
1490 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1491 	if (!dev)
1492 		return -ENODEV;
1493 
1494 	clid = tcm->tcm_parent;
1495 	if (clid) {
1496 		if (clid != TC_H_ROOT) {
1497 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1498 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1499 				if (!p) {
1500 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1501 					return -ENOENT;
1502 				}
1503 				q = qdisc_leaf(p, clid);
1504 			} else if (dev_ingress_queue(dev)) {
1505 				q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1506 			}
1507 		} else {
1508 			q = rtnl_dereference(dev->qdisc);
1509 		}
1510 		if (!q) {
1511 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1512 			return -ENOENT;
1513 		}
1514 
1515 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1516 			NL_SET_ERR_MSG(extack, "Invalid handle");
1517 			return -EINVAL;
1518 		}
1519 	} else {
1520 		q = qdisc_lookup(dev, tcm->tcm_handle);
1521 		if (!q) {
1522 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1523 			return -ENOENT;
1524 		}
1525 	}
1526 
1527 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1528 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1529 		return -EINVAL;
1530 	}
1531 
1532 	if (n->nlmsg_type == RTM_DELQDISC) {
1533 		if (!clid) {
1534 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1535 			return -EINVAL;
1536 		}
1537 		if (q->handle == 0) {
1538 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1539 			return -ENOENT;
1540 		}
1541 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1542 		if (err != 0)
1543 			return err;
1544 	} else {
1545 		qdisc_notify(net, skb, n, clid, NULL, q, NULL);
1546 	}
1547 	return 0;
1548 }
1549 
1550 /*
1551  * Create/change qdisc.
1552  */
1553 
1554 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1555 			   struct netlink_ext_ack *extack)
1556 {
1557 	struct net *net = sock_net(skb->sk);
1558 	struct tcmsg *tcm;
1559 	struct nlattr *tca[TCA_MAX + 1];
1560 	struct net_device *dev;
1561 	u32 clid;
1562 	struct Qdisc *q, *p;
1563 	int err;
1564 
1565 replay:
1566 	/* Reinit, just in case something touches this. */
1567 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1568 				     rtm_tca_policy, extack);
1569 	if (err < 0)
1570 		return err;
1571 
1572 	tcm = nlmsg_data(n);
1573 	clid = tcm->tcm_parent;
1574 	q = p = NULL;
1575 
1576 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1577 	if (!dev)
1578 		return -ENODEV;
1579 
1580 
1581 	if (clid) {
1582 		if (clid != TC_H_ROOT) {
1583 			if (clid != TC_H_INGRESS) {
1584 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1585 				if (!p) {
1586 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1587 					return -ENOENT;
1588 				}
1589 				q = qdisc_leaf(p, clid);
1590 			} else if (dev_ingress_queue_create(dev)) {
1591 				q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1592 			}
1593 		} else {
1594 			q = rtnl_dereference(dev->qdisc);
1595 		}
1596 
1597 		/* It may be default qdisc, ignore it */
1598 		if (q && q->handle == 0)
1599 			q = NULL;
1600 
1601 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1602 			if (tcm->tcm_handle) {
1603 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1604 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1605 					return -EEXIST;
1606 				}
1607 				if (TC_H_MIN(tcm->tcm_handle)) {
1608 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1609 					return -EINVAL;
1610 				}
1611 				q = qdisc_lookup(dev, tcm->tcm_handle);
1612 				if (!q)
1613 					goto create_n_graft;
1614 				if (n->nlmsg_flags & NLM_F_EXCL) {
1615 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1616 					return -EEXIST;
1617 				}
1618 				if (tca[TCA_KIND] &&
1619 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1620 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1621 					return -EINVAL;
1622 				}
1623 				if (q->flags & TCQ_F_INGRESS) {
1624 					NL_SET_ERR_MSG(extack,
1625 						       "Cannot regraft ingress or clsact Qdiscs");
1626 					return -EINVAL;
1627 				}
1628 				if (q == p ||
1629 				    (p && check_loop(q, p, 0))) {
1630 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1631 					return -ELOOP;
1632 				}
1633 				if (clid == TC_H_INGRESS) {
1634 					NL_SET_ERR_MSG(extack, "Ingress cannot graft directly");
1635 					return -EINVAL;
1636 				}
1637 				qdisc_refcount_inc(q);
1638 				goto graft;
1639 			} else {
1640 				if (!q)
1641 					goto create_n_graft;
1642 
1643 				/* This magic test requires explanation.
1644 				 *
1645 				 *   We know, that some child q is already
1646 				 *   attached to this parent and have choice:
1647 				 *   either to change it or to create/graft new one.
1648 				 *
1649 				 *   1. We are allowed to create/graft only
1650 				 *   if CREATE and REPLACE flags are set.
1651 				 *
1652 				 *   2. If EXCL is set, requestor wanted to say,
1653 				 *   that qdisc tcm_handle is not expected
1654 				 *   to exist, so that we choose create/graft too.
1655 				 *
1656 				 *   3. The last case is when no flags are set.
1657 				 *   Alas, it is sort of hole in API, we
1658 				 *   cannot decide what to do unambiguously.
1659 				 *   For now we select create/graft, if
1660 				 *   user gave KIND, which does not match existing.
1661 				 */
1662 				if ((n->nlmsg_flags & NLM_F_CREATE) &&
1663 				    (n->nlmsg_flags & NLM_F_REPLACE) &&
1664 				    ((n->nlmsg_flags & NLM_F_EXCL) ||
1665 				     (tca[TCA_KIND] &&
1666 				      nla_strcmp(tca[TCA_KIND], q->ops->id))))
1667 					goto create_n_graft;
1668 			}
1669 		}
1670 	} else {
1671 		if (!tcm->tcm_handle) {
1672 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1673 			return -EINVAL;
1674 		}
1675 		q = qdisc_lookup(dev, tcm->tcm_handle);
1676 	}
1677 
1678 	/* Change qdisc parameters */
1679 	if (!q) {
1680 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1681 		return -ENOENT;
1682 	}
1683 	if (n->nlmsg_flags & NLM_F_EXCL) {
1684 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1685 		return -EEXIST;
1686 	}
1687 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1688 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1689 		return -EINVAL;
1690 	}
1691 	err = qdisc_change(q, tca, extack);
1692 	if (err == 0)
1693 		qdisc_notify(net, skb, n, clid, NULL, q, extack);
1694 	return err;
1695 
1696 create_n_graft:
1697 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1698 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1699 		return -ENOENT;
1700 	}
1701 	if (clid == TC_H_INGRESS) {
1702 		if (dev_ingress_queue(dev)) {
1703 			q = qdisc_create(dev, dev_ingress_queue(dev),
1704 					 tcm->tcm_parent, tcm->tcm_parent,
1705 					 tca, &err, extack);
1706 		} else {
1707 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1708 			err = -ENOENT;
1709 		}
1710 	} else {
1711 		struct netdev_queue *dev_queue;
1712 
1713 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1714 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1715 		else if (p)
1716 			dev_queue = p->dev_queue;
1717 		else
1718 			dev_queue = netdev_get_tx_queue(dev, 0);
1719 
1720 		q = qdisc_create(dev, dev_queue,
1721 				 tcm->tcm_parent, tcm->tcm_handle,
1722 				 tca, &err, extack);
1723 	}
1724 	if (q == NULL) {
1725 		if (err == -EAGAIN)
1726 			goto replay;
1727 		return err;
1728 	}
1729 
1730 graft:
1731 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1732 	if (err) {
1733 		if (q)
1734 			qdisc_put(q);
1735 		return err;
1736 	}
1737 
1738 	return 0;
1739 }
1740 
1741 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1742 			      struct netlink_callback *cb,
1743 			      int *q_idx_p, int s_q_idx, bool recur,
1744 			      bool dump_invisible)
1745 {
1746 	int ret = 0, q_idx = *q_idx_p;
1747 	struct Qdisc *q;
1748 	int b;
1749 
1750 	if (!root)
1751 		return 0;
1752 
1753 	q = root;
1754 	if (q_idx < s_q_idx) {
1755 		q_idx++;
1756 	} else {
1757 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1758 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1759 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1760 				  RTM_NEWQDISC, NULL) <= 0)
1761 			goto done;
1762 		q_idx++;
1763 	}
1764 
1765 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1766 	 * itself has already been dumped.
1767 	 *
1768 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1769 	 * qdisc hashtable, we don't want to hit it again
1770 	 */
1771 	if (!qdisc_dev(root) || !recur)
1772 		goto out;
1773 
1774 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1775 		if (q_idx < s_q_idx) {
1776 			q_idx++;
1777 			continue;
1778 		}
1779 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1780 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1781 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1782 				  RTM_NEWQDISC, NULL) <= 0)
1783 			goto done;
1784 		q_idx++;
1785 	}
1786 
1787 out:
1788 	*q_idx_p = q_idx;
1789 	return ret;
1790 done:
1791 	ret = -1;
1792 	goto out;
1793 }
1794 
1795 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1796 {
1797 	struct net *net = sock_net(skb->sk);
1798 	int idx, q_idx;
1799 	int s_idx, s_q_idx;
1800 	struct net_device *dev;
1801 	const struct nlmsghdr *nlh = cb->nlh;
1802 	struct nlattr *tca[TCA_MAX + 1];
1803 	int err;
1804 
1805 	s_idx = cb->args[0];
1806 	s_q_idx = q_idx = cb->args[1];
1807 
1808 	idx = 0;
1809 	ASSERT_RTNL();
1810 
1811 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1812 				     rtm_tca_policy, cb->extack);
1813 	if (err < 0)
1814 		return err;
1815 
1816 	for_each_netdev(net, dev) {
1817 		struct netdev_queue *dev_queue;
1818 
1819 		if (idx < s_idx)
1820 			goto cont;
1821 		if (idx > s_idx)
1822 			s_q_idx = 0;
1823 		q_idx = 0;
1824 
1825 		if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1826 				       skb, cb, &q_idx, s_q_idx,
1827 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1828 			goto done;
1829 
1830 		dev_queue = dev_ingress_queue(dev);
1831 		if (dev_queue &&
1832 		    tc_dump_qdisc_root(rtnl_dereference(dev_queue->qdisc_sleeping),
1833 				       skb, cb, &q_idx, s_q_idx, false,
1834 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1835 			goto done;
1836 
1837 cont:
1838 		idx++;
1839 	}
1840 
1841 done:
1842 	cb->args[0] = idx;
1843 	cb->args[1] = q_idx;
1844 
1845 	return skb->len;
1846 }
1847 
1848 
1849 
1850 /************************************************
1851  *	Traffic classes manipulation.		*
1852  ************************************************/
1853 
1854 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1855 			  unsigned long cl, u32 portid, u32 seq, u16 flags,
1856 			  int event, struct netlink_ext_ack *extack)
1857 {
1858 	struct tcmsg *tcm;
1859 	struct nlmsghdr  *nlh;
1860 	unsigned char *b = skb_tail_pointer(skb);
1861 	struct gnet_dump d;
1862 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1863 
1864 	cond_resched();
1865 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1866 	if (!nlh)
1867 		goto out_nlmsg_trim;
1868 	tcm = nlmsg_data(nlh);
1869 	tcm->tcm_family = AF_UNSPEC;
1870 	tcm->tcm__pad1 = 0;
1871 	tcm->tcm__pad2 = 0;
1872 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1873 	tcm->tcm_parent = q->handle;
1874 	tcm->tcm_handle = q->handle;
1875 	tcm->tcm_info = 0;
1876 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1877 		goto nla_put_failure;
1878 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1879 		goto nla_put_failure;
1880 
1881 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1882 					 NULL, &d, TCA_PAD) < 0)
1883 		goto nla_put_failure;
1884 
1885 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1886 		goto nla_put_failure;
1887 
1888 	if (gnet_stats_finish_copy(&d) < 0)
1889 		goto nla_put_failure;
1890 
1891 	if (extack && extack->_msg &&
1892 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
1893 		goto out_nlmsg_trim;
1894 
1895 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1896 
1897 	return skb->len;
1898 
1899 out_nlmsg_trim:
1900 nla_put_failure:
1901 	nlmsg_trim(skb, b);
1902 	return -1;
1903 }
1904 
1905 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1906 			 struct nlmsghdr *n, struct Qdisc *q,
1907 			 unsigned long cl, int event, struct netlink_ext_ack *extack)
1908 {
1909 	struct sk_buff *skb;
1910 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1911 
1912 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1913 	if (!skb)
1914 		return -ENOBUFS;
1915 
1916 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event, extack) < 0) {
1917 		kfree_skb(skb);
1918 		return -EINVAL;
1919 	}
1920 
1921 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1922 			      n->nlmsg_flags & NLM_F_ECHO);
1923 }
1924 
1925 static int tclass_del_notify(struct net *net,
1926 			     const struct Qdisc_class_ops *cops,
1927 			     struct sk_buff *oskb, struct nlmsghdr *n,
1928 			     struct Qdisc *q, unsigned long cl,
1929 			     struct netlink_ext_ack *extack)
1930 {
1931 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1932 	struct sk_buff *skb;
1933 	int err = 0;
1934 
1935 	if (!cops->delete)
1936 		return -EOPNOTSUPP;
1937 
1938 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1939 	if (!skb)
1940 		return -ENOBUFS;
1941 
1942 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
1943 			   RTM_DELTCLASS, extack) < 0) {
1944 		kfree_skb(skb);
1945 		return -EINVAL;
1946 	}
1947 
1948 	err = cops->delete(q, cl, extack);
1949 	if (err) {
1950 		kfree_skb(skb);
1951 		return err;
1952 	}
1953 
1954 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1955 			     n->nlmsg_flags & NLM_F_ECHO);
1956 	return err;
1957 }
1958 
1959 #ifdef CONFIG_NET_CLS
1960 
1961 struct tcf_bind_args {
1962 	struct tcf_walker w;
1963 	unsigned long base;
1964 	unsigned long cl;
1965 	u32 classid;
1966 };
1967 
1968 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
1969 {
1970 	struct tcf_bind_args *a = (void *)arg;
1971 
1972 	if (n && tp->ops->bind_class) {
1973 		struct Qdisc *q = tcf_block_q(tp->chain->block);
1974 
1975 		sch_tree_lock(q);
1976 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
1977 		sch_tree_unlock(q);
1978 	}
1979 	return 0;
1980 }
1981 
1982 struct tc_bind_class_args {
1983 	struct qdisc_walker w;
1984 	unsigned long new_cl;
1985 	u32 portid;
1986 	u32 clid;
1987 };
1988 
1989 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
1990 				struct qdisc_walker *w)
1991 {
1992 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
1993 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1994 	struct tcf_block *block;
1995 	struct tcf_chain *chain;
1996 
1997 	block = cops->tcf_block(q, cl, NULL);
1998 	if (!block)
1999 		return 0;
2000 	for (chain = tcf_get_next_chain(block, NULL);
2001 	     chain;
2002 	     chain = tcf_get_next_chain(block, chain)) {
2003 		struct tcf_proto *tp;
2004 
2005 		for (tp = tcf_get_next_proto(chain, NULL);
2006 		     tp; tp = tcf_get_next_proto(chain, tp)) {
2007 			struct tcf_bind_args arg = {};
2008 
2009 			arg.w.fn = tcf_node_bind;
2010 			arg.classid = a->clid;
2011 			arg.base = cl;
2012 			arg.cl = a->new_cl;
2013 			tp->ops->walk(tp, &arg.w, true);
2014 		}
2015 	}
2016 
2017 	return 0;
2018 }
2019 
2020 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2021 			   unsigned long new_cl)
2022 {
2023 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
2024 	struct tc_bind_class_args args = {};
2025 
2026 	if (!cops->tcf_block)
2027 		return;
2028 	args.portid = portid;
2029 	args.clid = clid;
2030 	args.new_cl = new_cl;
2031 	args.w.fn = tc_bind_class_walker;
2032 	q->ops->cl_ops->walk(q, &args.w);
2033 }
2034 
2035 #else
2036 
2037 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2038 			   unsigned long new_cl)
2039 {
2040 }
2041 
2042 #endif
2043 
2044 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
2045 			 struct netlink_ext_ack *extack)
2046 {
2047 	struct net *net = sock_net(skb->sk);
2048 	struct tcmsg *tcm = nlmsg_data(n);
2049 	struct nlattr *tca[TCA_MAX + 1];
2050 	struct net_device *dev;
2051 	struct Qdisc *q = NULL;
2052 	const struct Qdisc_class_ops *cops;
2053 	unsigned long cl = 0;
2054 	unsigned long new_cl;
2055 	u32 portid;
2056 	u32 clid;
2057 	u32 qid;
2058 	int err;
2059 
2060 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2061 				     rtm_tca_policy, extack);
2062 	if (err < 0)
2063 		return err;
2064 
2065 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2066 	if (!dev)
2067 		return -ENODEV;
2068 
2069 	/*
2070 	   parent == TC_H_UNSPEC - unspecified parent.
2071 	   parent == TC_H_ROOT   - class is root, which has no parent.
2072 	   parent == X:0	 - parent is root class.
2073 	   parent == X:Y	 - parent is a node in hierarchy.
2074 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2075 
2076 	   handle == 0:0	 - generate handle from kernel pool.
2077 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2078 	   handle == X:Y	 - clear.
2079 	   handle == X:0	 - root class.
2080 	 */
2081 
2082 	/* Step 1. Determine qdisc handle X:0 */
2083 
2084 	portid = tcm->tcm_parent;
2085 	clid = tcm->tcm_handle;
2086 	qid = TC_H_MAJ(clid);
2087 
2088 	if (portid != TC_H_ROOT) {
2089 		u32 qid1 = TC_H_MAJ(portid);
2090 
2091 		if (qid && qid1) {
2092 			/* If both majors are known, they must be identical. */
2093 			if (qid != qid1)
2094 				return -EINVAL;
2095 		} else if (qid1) {
2096 			qid = qid1;
2097 		} else if (qid == 0)
2098 			qid = rtnl_dereference(dev->qdisc)->handle;
2099 
2100 		/* Now qid is genuine qdisc handle consistent
2101 		 * both with parent and child.
2102 		 *
2103 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2104 		 */
2105 		if (portid)
2106 			portid = TC_H_MAKE(qid, portid);
2107 	} else {
2108 		if (qid == 0)
2109 			qid = rtnl_dereference(dev->qdisc)->handle;
2110 	}
2111 
2112 	/* OK. Locate qdisc */
2113 	q = qdisc_lookup(dev, qid);
2114 	if (!q)
2115 		return -ENOENT;
2116 
2117 	/* An check that it supports classes */
2118 	cops = q->ops->cl_ops;
2119 	if (cops == NULL)
2120 		return -EINVAL;
2121 
2122 	/* Now try to get class */
2123 	if (clid == 0) {
2124 		if (portid == TC_H_ROOT)
2125 			clid = qid;
2126 	} else
2127 		clid = TC_H_MAKE(qid, clid);
2128 
2129 	if (clid)
2130 		cl = cops->find(q, clid);
2131 
2132 	if (cl == 0) {
2133 		err = -ENOENT;
2134 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2135 		    !(n->nlmsg_flags & NLM_F_CREATE))
2136 			goto out;
2137 	} else {
2138 		switch (n->nlmsg_type) {
2139 		case RTM_NEWTCLASS:
2140 			err = -EEXIST;
2141 			if (n->nlmsg_flags & NLM_F_EXCL)
2142 				goto out;
2143 			break;
2144 		case RTM_DELTCLASS:
2145 			err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2146 			/* Unbind the class with flilters with 0 */
2147 			tc_bind_tclass(q, portid, clid, 0);
2148 			goto out;
2149 		case RTM_GETTCLASS:
2150 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS, extack);
2151 			goto out;
2152 		default:
2153 			err = -EINVAL;
2154 			goto out;
2155 		}
2156 	}
2157 
2158 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2159 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2160 		return -EOPNOTSUPP;
2161 	}
2162 
2163 	new_cl = cl;
2164 	err = -EOPNOTSUPP;
2165 	if (cops->change)
2166 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2167 	if (err == 0) {
2168 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS, extack);
2169 		/* We just create a new class, need to do reverse binding. */
2170 		if (cl != new_cl)
2171 			tc_bind_tclass(q, portid, clid, new_cl);
2172 	}
2173 out:
2174 	return err;
2175 }
2176 
2177 struct qdisc_dump_args {
2178 	struct qdisc_walker	w;
2179 	struct sk_buff		*skb;
2180 	struct netlink_callback	*cb;
2181 };
2182 
2183 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2184 			    struct qdisc_walker *arg)
2185 {
2186 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2187 
2188 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2189 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2190 			      RTM_NEWTCLASS, NULL);
2191 }
2192 
2193 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2194 				struct tcmsg *tcm, struct netlink_callback *cb,
2195 				int *t_p, int s_t)
2196 {
2197 	struct qdisc_dump_args arg;
2198 
2199 	if (tc_qdisc_dump_ignore(q, false) ||
2200 	    *t_p < s_t || !q->ops->cl_ops ||
2201 	    (tcm->tcm_parent &&
2202 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2203 		(*t_p)++;
2204 		return 0;
2205 	}
2206 	if (*t_p > s_t)
2207 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2208 	arg.w.fn = qdisc_class_dump;
2209 	arg.skb = skb;
2210 	arg.cb = cb;
2211 	arg.w.stop  = 0;
2212 	arg.w.skip = cb->args[1];
2213 	arg.w.count = 0;
2214 	q->ops->cl_ops->walk(q, &arg.w);
2215 	cb->args[1] = arg.w.count;
2216 	if (arg.w.stop)
2217 		return -1;
2218 	(*t_p)++;
2219 	return 0;
2220 }
2221 
2222 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2223 			       struct tcmsg *tcm, struct netlink_callback *cb,
2224 			       int *t_p, int s_t, bool recur)
2225 {
2226 	struct Qdisc *q;
2227 	int b;
2228 
2229 	if (!root)
2230 		return 0;
2231 
2232 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2233 		return -1;
2234 
2235 	if (!qdisc_dev(root) || !recur)
2236 		return 0;
2237 
2238 	if (tcm->tcm_parent) {
2239 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2240 		if (q && q != root &&
2241 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2242 			return -1;
2243 		return 0;
2244 	}
2245 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2246 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2247 			return -1;
2248 	}
2249 
2250 	return 0;
2251 }
2252 
2253 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2254 {
2255 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2256 	struct net *net = sock_net(skb->sk);
2257 	struct netdev_queue *dev_queue;
2258 	struct net_device *dev;
2259 	int t, s_t;
2260 
2261 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2262 		return 0;
2263 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2264 	if (!dev)
2265 		return 0;
2266 
2267 	s_t = cb->args[0];
2268 	t = 0;
2269 
2270 	if (tc_dump_tclass_root(rtnl_dereference(dev->qdisc),
2271 				skb, tcm, cb, &t, s_t, true) < 0)
2272 		goto done;
2273 
2274 	dev_queue = dev_ingress_queue(dev);
2275 	if (dev_queue &&
2276 	    tc_dump_tclass_root(rtnl_dereference(dev_queue->qdisc_sleeping),
2277 				skb, tcm, cb, &t, s_t, false) < 0)
2278 		goto done;
2279 
2280 done:
2281 	cb->args[0] = t;
2282 
2283 	dev_put(dev);
2284 	return skb->len;
2285 }
2286 
2287 #ifdef CONFIG_PROC_FS
2288 static int psched_show(struct seq_file *seq, void *v)
2289 {
2290 	seq_printf(seq, "%08x %08x %08x %08x\n",
2291 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2292 		   1000000,
2293 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2294 
2295 	return 0;
2296 }
2297 
2298 static int __net_init psched_net_init(struct net *net)
2299 {
2300 	struct proc_dir_entry *e;
2301 
2302 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2303 	if (e == NULL)
2304 		return -ENOMEM;
2305 
2306 	return 0;
2307 }
2308 
2309 static void __net_exit psched_net_exit(struct net *net)
2310 {
2311 	remove_proc_entry("psched", net->proc_net);
2312 }
2313 #else
2314 static int __net_init psched_net_init(struct net *net)
2315 {
2316 	return 0;
2317 }
2318 
2319 static void __net_exit psched_net_exit(struct net *net)
2320 {
2321 }
2322 #endif
2323 
2324 static struct pernet_operations psched_net_ops = {
2325 	.init = psched_net_init,
2326 	.exit = psched_net_exit,
2327 };
2328 
2329 #if IS_ENABLED(CONFIG_RETPOLINE)
2330 DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper);
2331 #endif
2332 
2333 static int __init pktsched_init(void)
2334 {
2335 	int err;
2336 
2337 	err = register_pernet_subsys(&psched_net_ops);
2338 	if (err) {
2339 		pr_err("pktsched_init: "
2340 		       "cannot initialize per netns operations\n");
2341 		return err;
2342 	}
2343 
2344 	register_qdisc(&pfifo_fast_ops);
2345 	register_qdisc(&pfifo_qdisc_ops);
2346 	register_qdisc(&bfifo_qdisc_ops);
2347 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2348 	register_qdisc(&mq_qdisc_ops);
2349 	register_qdisc(&noqueue_qdisc_ops);
2350 
2351 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2352 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2353 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2354 		      0);
2355 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2356 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2357 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2358 		      0);
2359 
2360 	tc_wrapper_init();
2361 
2362 	return 0;
2363 }
2364 
2365 subsys_initcall(pktsched_init);
2366