xref: /linux/net/sched/sch_api.c (revision b15a4cfe100b9acd097d3ae7052448bd1cdc2a3b)
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_get_notify(struct net *net, struct sk_buff *oskb,
1007 			    struct nlmsghdr *n, u32 clid, struct Qdisc *q,
1008 			    struct netlink_ext_ack *extack)
1009 {
1010 	struct sk_buff *skb;
1011 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1012 
1013 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1014 	if (!skb)
1015 		return -ENOBUFS;
1016 
1017 	if (!tc_qdisc_dump_ignore(q, false)) {
1018 		if (tc_fill_qdisc(skb, q, clid, portid, n->nlmsg_seq, 0,
1019 				  RTM_NEWQDISC, extack) < 0)
1020 			goto err_out;
1021 	}
1022 
1023 	if (skb->len)
1024 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1025 				      n->nlmsg_flags & NLM_F_ECHO);
1026 
1027 err_out:
1028 	kfree_skb(skb);
1029 	return -EINVAL;
1030 }
1031 
1032 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1033 			struct nlmsghdr *n, u32 clid,
1034 			struct Qdisc *old, struct Qdisc *new,
1035 			struct netlink_ext_ack *extack)
1036 {
1037 	struct sk_buff *skb;
1038 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1039 
1040 	if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
1041 		return 0;
1042 
1043 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1044 	if (!skb)
1045 		return -ENOBUFS;
1046 
1047 	if (old && !tc_qdisc_dump_ignore(old, false)) {
1048 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1049 				  0, RTM_DELQDISC, extack) < 0)
1050 			goto err_out;
1051 	}
1052 	if (new && !tc_qdisc_dump_ignore(new, false)) {
1053 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1054 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC, extack) < 0)
1055 			goto err_out;
1056 	}
1057 
1058 	if (skb->len)
1059 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1060 				      n->nlmsg_flags & NLM_F_ECHO);
1061 
1062 err_out:
1063 	kfree_skb(skb);
1064 	return -EINVAL;
1065 }
1066 
1067 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1068 			       struct nlmsghdr *n, u32 clid,
1069 			       struct Qdisc *old, struct Qdisc *new,
1070 			       struct netlink_ext_ack *extack)
1071 {
1072 	if (new || old)
1073 		qdisc_notify(net, skb, n, clid, old, new, extack);
1074 
1075 	if (old)
1076 		qdisc_put(old);
1077 }
1078 
1079 static void qdisc_clear_nolock(struct Qdisc *sch)
1080 {
1081 	sch->flags &= ~TCQ_F_NOLOCK;
1082 	if (!(sch->flags & TCQ_F_CPUSTATS))
1083 		return;
1084 
1085 	free_percpu(sch->cpu_bstats);
1086 	free_percpu(sch->cpu_qstats);
1087 	sch->cpu_bstats = NULL;
1088 	sch->cpu_qstats = NULL;
1089 	sch->flags &= ~TCQ_F_CPUSTATS;
1090 }
1091 
1092 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1093  * to device "dev".
1094  *
1095  * When appropriate send a netlink notification using 'skb'
1096  * and "n".
1097  *
1098  * On success, destroy old qdisc.
1099  */
1100 
1101 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1102 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1103 		       struct Qdisc *new, struct Qdisc *old,
1104 		       struct netlink_ext_ack *extack)
1105 {
1106 	struct Qdisc *q = old;
1107 	struct net *net = dev_net(dev);
1108 
1109 	if (parent == NULL) {
1110 		unsigned int i, num_q, ingress;
1111 		struct netdev_queue *dev_queue;
1112 
1113 		ingress = 0;
1114 		num_q = dev->num_tx_queues;
1115 		if ((q && q->flags & TCQ_F_INGRESS) ||
1116 		    (new && new->flags & TCQ_F_INGRESS)) {
1117 			ingress = 1;
1118 			dev_queue = dev_ingress_queue(dev);
1119 			if (!dev_queue) {
1120 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1121 				return -ENOENT;
1122 			}
1123 
1124 			q = rtnl_dereference(dev_queue->qdisc_sleeping);
1125 
1126 			/* This is the counterpart of that qdisc_refcount_inc_nz() call in
1127 			 * __tcf_qdisc_find() for filter requests.
1128 			 */
1129 			if (!qdisc_refcount_dec_if_one(q)) {
1130 				NL_SET_ERR_MSG(extack,
1131 					       "Current ingress or clsact Qdisc has ongoing filter requests");
1132 				return -EBUSY;
1133 			}
1134 		}
1135 
1136 		if (dev->flags & IFF_UP)
1137 			dev_deactivate(dev);
1138 
1139 		qdisc_offload_graft_root(dev, new, old, extack);
1140 
1141 		if (new && new->ops->attach && !ingress)
1142 			goto skip;
1143 
1144 		if (!ingress) {
1145 			for (i = 0; i < num_q; i++) {
1146 				dev_queue = netdev_get_tx_queue(dev, i);
1147 				old = dev_graft_qdisc(dev_queue, new);
1148 
1149 				if (new && i > 0)
1150 					qdisc_refcount_inc(new);
1151 				qdisc_put(old);
1152 			}
1153 		} else {
1154 			old = dev_graft_qdisc(dev_queue, NULL);
1155 
1156 			/* {ingress,clsact}_destroy() @old before grafting @new to avoid
1157 			 * unprotected concurrent accesses to net_device::miniq_{in,e}gress
1158 			 * pointer(s) in mini_qdisc_pair_swap().
1159 			 */
1160 			qdisc_notify(net, skb, n, classid, old, new, extack);
1161 			qdisc_destroy(old);
1162 
1163 			dev_graft_qdisc(dev_queue, new);
1164 		}
1165 
1166 skip:
1167 		if (!ingress) {
1168 			old = rtnl_dereference(dev->qdisc);
1169 			if (new && !new->ops->attach)
1170 				qdisc_refcount_inc(new);
1171 			rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1172 
1173 			notify_and_destroy(net, skb, n, classid, old, new, extack);
1174 
1175 			if (new && new->ops->attach)
1176 				new->ops->attach(new);
1177 		}
1178 
1179 		if (dev->flags & IFF_UP)
1180 			dev_activate(dev);
1181 	} else {
1182 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1183 		unsigned long cl;
1184 		int err;
1185 
1186 		/* Only support running class lockless if parent is lockless */
1187 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1188 			qdisc_clear_nolock(new);
1189 
1190 		if (!cops || !cops->graft)
1191 			return -EOPNOTSUPP;
1192 
1193 		cl = cops->find(parent, classid);
1194 		if (!cl) {
1195 			NL_SET_ERR_MSG(extack, "Specified class not found");
1196 			return -ENOENT;
1197 		}
1198 
1199 		if (new && new->ops == &noqueue_qdisc_ops) {
1200 			NL_SET_ERR_MSG(extack, "Cannot assign noqueue to a class");
1201 			return -EINVAL;
1202 		}
1203 
1204 		err = cops->graft(parent, cl, new, &old, extack);
1205 		if (err)
1206 			return err;
1207 		notify_and_destroy(net, skb, n, classid, old, new, extack);
1208 	}
1209 	return 0;
1210 }
1211 
1212 static int qdisc_block_add_dev(struct Qdisc *sch, struct net_device *dev,
1213 			       struct netlink_ext_ack *extack)
1214 {
1215 	const struct Qdisc_class_ops *cl_ops = sch->ops->cl_ops;
1216 	struct tcf_block *block;
1217 	int err;
1218 
1219 	block = cl_ops->tcf_block(sch, TC_H_MIN_INGRESS, NULL);
1220 	if (block) {
1221 		err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1222 		if (err) {
1223 			NL_SET_ERR_MSG(extack,
1224 				       "ingress block dev insert failed");
1225 			return err;
1226 		}
1227 	}
1228 
1229 	block = cl_ops->tcf_block(sch, TC_H_MIN_EGRESS, NULL);
1230 	if (block) {
1231 		err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1232 		if (err) {
1233 			NL_SET_ERR_MSG(extack,
1234 				       "Egress block dev insert failed");
1235 			goto err_out;
1236 		}
1237 	}
1238 
1239 	return 0;
1240 
1241 err_out:
1242 	block = cl_ops->tcf_block(sch, TC_H_MIN_INGRESS, NULL);
1243 	if (block)
1244 		xa_erase(&block->ports, dev->ifindex);
1245 
1246 	return err;
1247 }
1248 
1249 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1250 				   struct netlink_ext_ack *extack)
1251 {
1252 	u32 block_index;
1253 
1254 	if (tca[TCA_INGRESS_BLOCK]) {
1255 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1256 
1257 		if (!block_index) {
1258 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1259 			return -EINVAL;
1260 		}
1261 		if (!sch->ops->ingress_block_set) {
1262 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1263 			return -EOPNOTSUPP;
1264 		}
1265 		sch->ops->ingress_block_set(sch, block_index);
1266 	}
1267 	if (tca[TCA_EGRESS_BLOCK]) {
1268 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1269 
1270 		if (!block_index) {
1271 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1272 			return -EINVAL;
1273 		}
1274 		if (!sch->ops->egress_block_set) {
1275 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1276 			return -EOPNOTSUPP;
1277 		}
1278 		sch->ops->egress_block_set(sch, block_index);
1279 	}
1280 	return 0;
1281 }
1282 
1283 /*
1284    Allocate and initialize new qdisc.
1285 
1286    Parameters are passed via opt.
1287  */
1288 
1289 static struct Qdisc *qdisc_create(struct net_device *dev,
1290 				  struct netdev_queue *dev_queue,
1291 				  u32 parent, u32 handle,
1292 				  struct nlattr **tca, int *errp,
1293 				  struct netlink_ext_ack *extack)
1294 {
1295 	int err;
1296 	struct nlattr *kind = tca[TCA_KIND];
1297 	struct Qdisc *sch;
1298 	struct Qdisc_ops *ops;
1299 	struct qdisc_size_table *stab;
1300 
1301 	ops = qdisc_lookup_ops(kind);
1302 #ifdef CONFIG_MODULES
1303 	if (ops == NULL && kind != NULL) {
1304 		char name[IFNAMSIZ];
1305 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1306 			/* We dropped the RTNL semaphore in order to
1307 			 * perform the module load.  So, even if we
1308 			 * succeeded in loading the module we have to
1309 			 * tell the caller to replay the request.  We
1310 			 * indicate this using -EAGAIN.
1311 			 * We replay the request because the device may
1312 			 * go away in the mean time.
1313 			 */
1314 			rtnl_unlock();
1315 			request_module("sch_%s", name);
1316 			rtnl_lock();
1317 			ops = qdisc_lookup_ops(kind);
1318 			if (ops != NULL) {
1319 				/* We will try again qdisc_lookup_ops,
1320 				 * so don't keep a reference.
1321 				 */
1322 				module_put(ops->owner);
1323 				err = -EAGAIN;
1324 				goto err_out;
1325 			}
1326 		}
1327 	}
1328 #endif
1329 
1330 	err = -ENOENT;
1331 	if (!ops) {
1332 		NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1333 		goto err_out;
1334 	}
1335 
1336 	sch = qdisc_alloc(dev_queue, ops, extack);
1337 	if (IS_ERR(sch)) {
1338 		err = PTR_ERR(sch);
1339 		goto err_out2;
1340 	}
1341 
1342 	sch->parent = parent;
1343 
1344 	if (handle == TC_H_INGRESS) {
1345 		if (!(sch->flags & TCQ_F_INGRESS)) {
1346 			NL_SET_ERR_MSG(extack,
1347 				       "Specified parent ID is reserved for ingress and clsact Qdiscs");
1348 			err = -EINVAL;
1349 			goto err_out3;
1350 		}
1351 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1352 	} else {
1353 		if (handle == 0) {
1354 			handle = qdisc_alloc_handle(dev);
1355 			if (handle == 0) {
1356 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1357 				err = -ENOSPC;
1358 				goto err_out3;
1359 			}
1360 		}
1361 		if (!netif_is_multiqueue(dev))
1362 			sch->flags |= TCQ_F_ONETXQUEUE;
1363 	}
1364 
1365 	sch->handle = handle;
1366 
1367 	/* This exist to keep backward compatible with a userspace
1368 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1369 	 * facility on older kernels by setting tx_queue_len=0 (prior
1370 	 * to qdisc init), and then forgot to reinit tx_queue_len
1371 	 * before again attaching a qdisc.
1372 	 */
1373 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1374 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1375 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1376 	}
1377 
1378 	err = qdisc_block_indexes_set(sch, tca, extack);
1379 	if (err)
1380 		goto err_out3;
1381 
1382 	if (tca[TCA_STAB]) {
1383 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1384 		if (IS_ERR(stab)) {
1385 			err = PTR_ERR(stab);
1386 			goto err_out3;
1387 		}
1388 		rcu_assign_pointer(sch->stab, stab);
1389 	}
1390 
1391 	if (ops->init) {
1392 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1393 		if (err != 0)
1394 			goto err_out4;
1395 	}
1396 
1397 	if (tca[TCA_RATE]) {
1398 		err = -EOPNOTSUPP;
1399 		if (sch->flags & TCQ_F_MQROOT) {
1400 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1401 			goto err_out4;
1402 		}
1403 
1404 		err = gen_new_estimator(&sch->bstats,
1405 					sch->cpu_bstats,
1406 					&sch->rate_est,
1407 					NULL,
1408 					true,
1409 					tca[TCA_RATE]);
1410 		if (err) {
1411 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1412 			goto err_out4;
1413 		}
1414 	}
1415 
1416 	qdisc_hash_add(sch, false);
1417 	trace_qdisc_create(ops, dev, parent);
1418 
1419 	err = qdisc_block_add_dev(sch, dev, extack);
1420 	if (err)
1421 		goto err_out4;
1422 
1423 	return sch;
1424 
1425 err_out4:
1426 	/* Even if ops->init() failed, we call ops->destroy()
1427 	 * like qdisc_create_dflt().
1428 	 */
1429 	if (ops->destroy)
1430 		ops->destroy(sch);
1431 	qdisc_put_stab(rtnl_dereference(sch->stab));
1432 err_out3:
1433 	netdev_put(dev, &sch->dev_tracker);
1434 	qdisc_free(sch);
1435 err_out2:
1436 	module_put(ops->owner);
1437 err_out:
1438 	*errp = err;
1439 	return NULL;
1440 }
1441 
1442 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1443 			struct netlink_ext_ack *extack)
1444 {
1445 	struct qdisc_size_table *ostab, *stab = NULL;
1446 	int err = 0;
1447 
1448 	if (tca[TCA_OPTIONS]) {
1449 		if (!sch->ops->change) {
1450 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1451 			return -EINVAL;
1452 		}
1453 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1454 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1455 			return -EOPNOTSUPP;
1456 		}
1457 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1458 		if (err)
1459 			return err;
1460 	}
1461 
1462 	if (tca[TCA_STAB]) {
1463 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1464 		if (IS_ERR(stab))
1465 			return PTR_ERR(stab);
1466 	}
1467 
1468 	ostab = rtnl_dereference(sch->stab);
1469 	rcu_assign_pointer(sch->stab, stab);
1470 	qdisc_put_stab(ostab);
1471 
1472 	if (tca[TCA_RATE]) {
1473 		/* NB: ignores errors from replace_estimator
1474 		   because change can't be undone. */
1475 		if (sch->flags & TCQ_F_MQROOT)
1476 			goto out;
1477 		gen_replace_estimator(&sch->bstats,
1478 				      sch->cpu_bstats,
1479 				      &sch->rate_est,
1480 				      NULL,
1481 				      true,
1482 				      tca[TCA_RATE]);
1483 	}
1484 out:
1485 	return 0;
1486 }
1487 
1488 struct check_loop_arg {
1489 	struct qdisc_walker	w;
1490 	struct Qdisc		*p;
1491 	int			depth;
1492 };
1493 
1494 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1495 			 struct qdisc_walker *w);
1496 
1497 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1498 {
1499 	struct check_loop_arg	arg;
1500 
1501 	if (q->ops->cl_ops == NULL)
1502 		return 0;
1503 
1504 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1505 	arg.w.fn = check_loop_fn;
1506 	arg.depth = depth;
1507 	arg.p = p;
1508 	q->ops->cl_ops->walk(q, &arg.w);
1509 	return arg.w.stop ? -ELOOP : 0;
1510 }
1511 
1512 static int
1513 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1514 {
1515 	struct Qdisc *leaf;
1516 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1517 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1518 
1519 	leaf = cops->leaf(q, cl);
1520 	if (leaf) {
1521 		if (leaf == arg->p || arg->depth > 7)
1522 			return -ELOOP;
1523 		return check_loop(leaf, arg->p, arg->depth + 1);
1524 	}
1525 	return 0;
1526 }
1527 
1528 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1529 	[TCA_KIND]		= { .type = NLA_STRING },
1530 	[TCA_RATE]		= { .type = NLA_BINARY,
1531 				    .len = sizeof(struct tc_estimator) },
1532 	[TCA_STAB]		= { .type = NLA_NESTED },
1533 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1534 	[TCA_CHAIN]		= { .type = NLA_U32 },
1535 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1536 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1537 };
1538 
1539 /*
1540  * Delete/get qdisc.
1541  */
1542 
1543 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1544 			struct netlink_ext_ack *extack)
1545 {
1546 	struct net *net = sock_net(skb->sk);
1547 	struct tcmsg *tcm = nlmsg_data(n);
1548 	struct nlattr *tca[TCA_MAX + 1];
1549 	struct net_device *dev;
1550 	u32 clid;
1551 	struct Qdisc *q = NULL;
1552 	struct Qdisc *p = NULL;
1553 	int err;
1554 
1555 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1556 				     rtm_tca_policy, extack);
1557 	if (err < 0)
1558 		return err;
1559 
1560 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1561 	if (!dev)
1562 		return -ENODEV;
1563 
1564 	clid = tcm->tcm_parent;
1565 	if (clid) {
1566 		if (clid != TC_H_ROOT) {
1567 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1568 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1569 				if (!p) {
1570 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1571 					return -ENOENT;
1572 				}
1573 				q = qdisc_leaf(p, clid);
1574 			} else if (dev_ingress_queue(dev)) {
1575 				q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1576 			}
1577 		} else {
1578 			q = rtnl_dereference(dev->qdisc);
1579 		}
1580 		if (!q) {
1581 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1582 			return -ENOENT;
1583 		}
1584 
1585 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1586 			NL_SET_ERR_MSG(extack, "Invalid handle");
1587 			return -EINVAL;
1588 		}
1589 	} else {
1590 		q = qdisc_lookup(dev, tcm->tcm_handle);
1591 		if (!q) {
1592 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1593 			return -ENOENT;
1594 		}
1595 	}
1596 
1597 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1598 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1599 		return -EINVAL;
1600 	}
1601 
1602 	if (n->nlmsg_type == RTM_DELQDISC) {
1603 		if (!clid) {
1604 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1605 			return -EINVAL;
1606 		}
1607 		if (q->handle == 0) {
1608 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1609 			return -ENOENT;
1610 		}
1611 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1612 		if (err != 0)
1613 			return err;
1614 	} else {
1615 		qdisc_get_notify(net, skb, n, clid, q, NULL);
1616 	}
1617 	return 0;
1618 }
1619 
1620 static bool req_create_or_replace(struct nlmsghdr *n)
1621 {
1622 	return (n->nlmsg_flags & NLM_F_CREATE &&
1623 		n->nlmsg_flags & NLM_F_REPLACE);
1624 }
1625 
1626 static bool req_create_exclusive(struct nlmsghdr *n)
1627 {
1628 	return (n->nlmsg_flags & NLM_F_CREATE &&
1629 		n->nlmsg_flags & NLM_F_EXCL);
1630 }
1631 
1632 static bool req_change(struct nlmsghdr *n)
1633 {
1634 	return (!(n->nlmsg_flags & NLM_F_CREATE) &&
1635 		!(n->nlmsg_flags & NLM_F_REPLACE) &&
1636 		!(n->nlmsg_flags & NLM_F_EXCL));
1637 }
1638 
1639 /*
1640  * Create/change qdisc.
1641  */
1642 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1643 			   struct netlink_ext_ack *extack)
1644 {
1645 	struct net *net = sock_net(skb->sk);
1646 	struct tcmsg *tcm;
1647 	struct nlattr *tca[TCA_MAX + 1];
1648 	struct net_device *dev;
1649 	u32 clid;
1650 	struct Qdisc *q, *p;
1651 	int err;
1652 
1653 replay:
1654 	/* Reinit, just in case something touches this. */
1655 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1656 				     rtm_tca_policy, extack);
1657 	if (err < 0)
1658 		return err;
1659 
1660 	tcm = nlmsg_data(n);
1661 	clid = tcm->tcm_parent;
1662 	q = p = NULL;
1663 
1664 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1665 	if (!dev)
1666 		return -ENODEV;
1667 
1668 
1669 	if (clid) {
1670 		if (clid != TC_H_ROOT) {
1671 			if (clid != TC_H_INGRESS) {
1672 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1673 				if (!p) {
1674 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1675 					return -ENOENT;
1676 				}
1677 				q = qdisc_leaf(p, clid);
1678 			} else if (dev_ingress_queue_create(dev)) {
1679 				q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1680 			}
1681 		} else {
1682 			q = rtnl_dereference(dev->qdisc);
1683 		}
1684 
1685 		/* It may be default qdisc, ignore it */
1686 		if (q && q->handle == 0)
1687 			q = NULL;
1688 
1689 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1690 			if (tcm->tcm_handle) {
1691 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1692 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1693 					return -EEXIST;
1694 				}
1695 				if (TC_H_MIN(tcm->tcm_handle)) {
1696 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1697 					return -EINVAL;
1698 				}
1699 				q = qdisc_lookup(dev, tcm->tcm_handle);
1700 				if (!q)
1701 					goto create_n_graft;
1702 				if (n->nlmsg_flags & NLM_F_EXCL) {
1703 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1704 					return -EEXIST;
1705 				}
1706 				if (tca[TCA_KIND] &&
1707 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1708 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1709 					return -EINVAL;
1710 				}
1711 				if (q->flags & TCQ_F_INGRESS) {
1712 					NL_SET_ERR_MSG(extack,
1713 						       "Cannot regraft ingress or clsact Qdiscs");
1714 					return -EINVAL;
1715 				}
1716 				if (q == p ||
1717 				    (p && check_loop(q, p, 0))) {
1718 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1719 					return -ELOOP;
1720 				}
1721 				if (clid == TC_H_INGRESS) {
1722 					NL_SET_ERR_MSG(extack, "Ingress cannot graft directly");
1723 					return -EINVAL;
1724 				}
1725 				qdisc_refcount_inc(q);
1726 				goto graft;
1727 			} else {
1728 				if (!q)
1729 					goto create_n_graft;
1730 
1731 				/* This magic test requires explanation.
1732 				 *
1733 				 *   We know, that some child q is already
1734 				 *   attached to this parent and have choice:
1735 				 *   1) change it or 2) create/graft new one.
1736 				 *   If the requested qdisc kind is different
1737 				 *   than the existing one, then we choose graft.
1738 				 *   If they are the same then this is "change"
1739 				 *   operation - just let it fallthrough..
1740 				 *
1741 				 *   1. We are allowed to create/graft only
1742 				 *   if the request is explicitly stating
1743 				 *   "please create if it doesn't exist".
1744 				 *
1745 				 *   2. If the request is to exclusive create
1746 				 *   then the qdisc tcm_handle is not expected
1747 				 *   to exist, so that we choose create/graft too.
1748 				 *
1749 				 *   3. The last case is when no flags are set.
1750 				 *   This will happen when for example tc
1751 				 *   utility issues a "change" command.
1752 				 *   Alas, it is sort of hole in API, we
1753 				 *   cannot decide what to do unambiguously.
1754 				 *   For now we select create/graft.
1755 				 */
1756 				if (tca[TCA_KIND] &&
1757 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1758 					if (req_create_or_replace(n) ||
1759 					    req_create_exclusive(n))
1760 						goto create_n_graft;
1761 					else if (req_change(n))
1762 						goto create_n_graft2;
1763 				}
1764 			}
1765 		}
1766 	} else {
1767 		if (!tcm->tcm_handle) {
1768 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1769 			return -EINVAL;
1770 		}
1771 		q = qdisc_lookup(dev, tcm->tcm_handle);
1772 	}
1773 
1774 	/* Change qdisc parameters */
1775 	if (!q) {
1776 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1777 		return -ENOENT;
1778 	}
1779 	if (n->nlmsg_flags & NLM_F_EXCL) {
1780 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1781 		return -EEXIST;
1782 	}
1783 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1784 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1785 		return -EINVAL;
1786 	}
1787 	err = qdisc_change(q, tca, extack);
1788 	if (err == 0)
1789 		qdisc_notify(net, skb, n, clid, NULL, q, extack);
1790 	return err;
1791 
1792 create_n_graft:
1793 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1794 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1795 		return -ENOENT;
1796 	}
1797 create_n_graft2:
1798 	if (clid == TC_H_INGRESS) {
1799 		if (dev_ingress_queue(dev)) {
1800 			q = qdisc_create(dev, dev_ingress_queue(dev),
1801 					 tcm->tcm_parent, tcm->tcm_parent,
1802 					 tca, &err, extack);
1803 		} else {
1804 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1805 			err = -ENOENT;
1806 		}
1807 	} else {
1808 		struct netdev_queue *dev_queue;
1809 
1810 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1811 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1812 		else if (p)
1813 			dev_queue = p->dev_queue;
1814 		else
1815 			dev_queue = netdev_get_tx_queue(dev, 0);
1816 
1817 		q = qdisc_create(dev, dev_queue,
1818 				 tcm->tcm_parent, tcm->tcm_handle,
1819 				 tca, &err, extack);
1820 	}
1821 	if (q == NULL) {
1822 		if (err == -EAGAIN)
1823 			goto replay;
1824 		return err;
1825 	}
1826 
1827 graft:
1828 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1829 	if (err) {
1830 		if (q)
1831 			qdisc_put(q);
1832 		return err;
1833 	}
1834 
1835 	return 0;
1836 }
1837 
1838 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1839 			      struct netlink_callback *cb,
1840 			      int *q_idx_p, int s_q_idx, bool recur,
1841 			      bool dump_invisible)
1842 {
1843 	int ret = 0, q_idx = *q_idx_p;
1844 	struct Qdisc *q;
1845 	int b;
1846 
1847 	if (!root)
1848 		return 0;
1849 
1850 	q = root;
1851 	if (q_idx < s_q_idx) {
1852 		q_idx++;
1853 	} else {
1854 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1855 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1856 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1857 				  RTM_NEWQDISC, NULL) <= 0)
1858 			goto done;
1859 		q_idx++;
1860 	}
1861 
1862 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1863 	 * itself has already been dumped.
1864 	 *
1865 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1866 	 * qdisc hashtable, we don't want to hit it again
1867 	 */
1868 	if (!qdisc_dev(root) || !recur)
1869 		goto out;
1870 
1871 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1872 		if (q_idx < s_q_idx) {
1873 			q_idx++;
1874 			continue;
1875 		}
1876 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1877 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1878 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1879 				  RTM_NEWQDISC, NULL) <= 0)
1880 			goto done;
1881 		q_idx++;
1882 	}
1883 
1884 out:
1885 	*q_idx_p = q_idx;
1886 	return ret;
1887 done:
1888 	ret = -1;
1889 	goto out;
1890 }
1891 
1892 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1893 {
1894 	struct net *net = sock_net(skb->sk);
1895 	int idx, q_idx;
1896 	int s_idx, s_q_idx;
1897 	struct net_device *dev;
1898 	const struct nlmsghdr *nlh = cb->nlh;
1899 	struct nlattr *tca[TCA_MAX + 1];
1900 	int err;
1901 
1902 	s_idx = cb->args[0];
1903 	s_q_idx = q_idx = cb->args[1];
1904 
1905 	idx = 0;
1906 	ASSERT_RTNL();
1907 
1908 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1909 				     rtm_tca_policy, cb->extack);
1910 	if (err < 0)
1911 		return err;
1912 
1913 	for_each_netdev(net, dev) {
1914 		struct netdev_queue *dev_queue;
1915 
1916 		if (idx < s_idx)
1917 			goto cont;
1918 		if (idx > s_idx)
1919 			s_q_idx = 0;
1920 		q_idx = 0;
1921 
1922 		if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1923 				       skb, cb, &q_idx, s_q_idx,
1924 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1925 			goto done;
1926 
1927 		dev_queue = dev_ingress_queue(dev);
1928 		if (dev_queue &&
1929 		    tc_dump_qdisc_root(rtnl_dereference(dev_queue->qdisc_sleeping),
1930 				       skb, cb, &q_idx, s_q_idx, false,
1931 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1932 			goto done;
1933 
1934 cont:
1935 		idx++;
1936 	}
1937 
1938 done:
1939 	cb->args[0] = idx;
1940 	cb->args[1] = q_idx;
1941 
1942 	return skb->len;
1943 }
1944 
1945 
1946 
1947 /************************************************
1948  *	Traffic classes manipulation.		*
1949  ************************************************/
1950 
1951 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1952 			  unsigned long cl, u32 portid, u32 seq, u16 flags,
1953 			  int event, struct netlink_ext_ack *extack)
1954 {
1955 	struct tcmsg *tcm;
1956 	struct nlmsghdr  *nlh;
1957 	unsigned char *b = skb_tail_pointer(skb);
1958 	struct gnet_dump d;
1959 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1960 
1961 	cond_resched();
1962 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1963 	if (!nlh)
1964 		goto out_nlmsg_trim;
1965 	tcm = nlmsg_data(nlh);
1966 	tcm->tcm_family = AF_UNSPEC;
1967 	tcm->tcm__pad1 = 0;
1968 	tcm->tcm__pad2 = 0;
1969 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1970 	tcm->tcm_parent = q->handle;
1971 	tcm->tcm_handle = q->handle;
1972 	tcm->tcm_info = 0;
1973 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1974 		goto nla_put_failure;
1975 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1976 		goto nla_put_failure;
1977 
1978 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1979 					 NULL, &d, TCA_PAD) < 0)
1980 		goto nla_put_failure;
1981 
1982 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1983 		goto nla_put_failure;
1984 
1985 	if (gnet_stats_finish_copy(&d) < 0)
1986 		goto nla_put_failure;
1987 
1988 	if (extack && extack->_msg &&
1989 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
1990 		goto out_nlmsg_trim;
1991 
1992 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1993 
1994 	return skb->len;
1995 
1996 out_nlmsg_trim:
1997 nla_put_failure:
1998 	nlmsg_trim(skb, b);
1999 	return -1;
2000 }
2001 
2002 static int tclass_notify(struct net *net, struct sk_buff *oskb,
2003 			 struct nlmsghdr *n, struct Qdisc *q,
2004 			 unsigned long cl, int event, struct netlink_ext_ack *extack)
2005 {
2006 	struct sk_buff *skb;
2007 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2008 
2009 	if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
2010 		return 0;
2011 
2012 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2013 	if (!skb)
2014 		return -ENOBUFS;
2015 
2016 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event, extack) < 0) {
2017 		kfree_skb(skb);
2018 		return -EINVAL;
2019 	}
2020 
2021 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2022 			      n->nlmsg_flags & NLM_F_ECHO);
2023 }
2024 
2025 static int tclass_get_notify(struct net *net, struct sk_buff *oskb,
2026 			     struct nlmsghdr *n, struct Qdisc *q,
2027 			     unsigned long cl, struct netlink_ext_ack *extack)
2028 {
2029 	struct sk_buff *skb;
2030 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2031 
2032 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2033 	if (!skb)
2034 		return -ENOBUFS;
2035 
2036 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, RTM_NEWTCLASS,
2037 			   extack) < 0) {
2038 		kfree_skb(skb);
2039 		return -EINVAL;
2040 	}
2041 
2042 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2043 			      n->nlmsg_flags & NLM_F_ECHO);
2044 }
2045 
2046 static int tclass_del_notify(struct net *net,
2047 			     const struct Qdisc_class_ops *cops,
2048 			     struct sk_buff *oskb, struct nlmsghdr *n,
2049 			     struct Qdisc *q, unsigned long cl,
2050 			     struct netlink_ext_ack *extack)
2051 {
2052 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2053 	struct sk_buff *skb;
2054 	int err = 0;
2055 
2056 	if (!cops->delete)
2057 		return -EOPNOTSUPP;
2058 
2059 	if (rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC)) {
2060 		skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2061 		if (!skb)
2062 			return -ENOBUFS;
2063 
2064 		if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
2065 				   RTM_DELTCLASS, extack) < 0) {
2066 			kfree_skb(skb);
2067 			return -EINVAL;
2068 		}
2069 	} else {
2070 		skb = NULL;
2071 	}
2072 
2073 	err = cops->delete(q, cl, extack);
2074 	if (err) {
2075 		kfree_skb(skb);
2076 		return err;
2077 	}
2078 
2079 	err = rtnetlink_maybe_send(skb, net, portid, RTNLGRP_TC,
2080 				   n->nlmsg_flags & NLM_F_ECHO);
2081 	return err;
2082 }
2083 
2084 #ifdef CONFIG_NET_CLS
2085 
2086 struct tcf_bind_args {
2087 	struct tcf_walker w;
2088 	unsigned long base;
2089 	unsigned long cl;
2090 	u32 classid;
2091 };
2092 
2093 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2094 {
2095 	struct tcf_bind_args *a = (void *)arg;
2096 
2097 	if (n && tp->ops->bind_class) {
2098 		struct Qdisc *q = tcf_block_q(tp->chain->block);
2099 
2100 		sch_tree_lock(q);
2101 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
2102 		sch_tree_unlock(q);
2103 	}
2104 	return 0;
2105 }
2106 
2107 struct tc_bind_class_args {
2108 	struct qdisc_walker w;
2109 	unsigned long new_cl;
2110 	u32 portid;
2111 	u32 clid;
2112 };
2113 
2114 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
2115 				struct qdisc_walker *w)
2116 {
2117 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
2118 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
2119 	struct tcf_block *block;
2120 	struct tcf_chain *chain;
2121 
2122 	block = cops->tcf_block(q, cl, NULL);
2123 	if (!block)
2124 		return 0;
2125 	for (chain = tcf_get_next_chain(block, NULL);
2126 	     chain;
2127 	     chain = tcf_get_next_chain(block, chain)) {
2128 		struct tcf_proto *tp;
2129 
2130 		for (tp = tcf_get_next_proto(chain, NULL);
2131 		     tp; tp = tcf_get_next_proto(chain, tp)) {
2132 			struct tcf_bind_args arg = {};
2133 
2134 			arg.w.fn = tcf_node_bind;
2135 			arg.classid = a->clid;
2136 			arg.base = cl;
2137 			arg.cl = a->new_cl;
2138 			tp->ops->walk(tp, &arg.w, true);
2139 		}
2140 	}
2141 
2142 	return 0;
2143 }
2144 
2145 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2146 			   unsigned long new_cl)
2147 {
2148 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
2149 	struct tc_bind_class_args args = {};
2150 
2151 	if (!cops->tcf_block)
2152 		return;
2153 	args.portid = portid;
2154 	args.clid = clid;
2155 	args.new_cl = new_cl;
2156 	args.w.fn = tc_bind_class_walker;
2157 	q->ops->cl_ops->walk(q, &args.w);
2158 }
2159 
2160 #else
2161 
2162 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2163 			   unsigned long new_cl)
2164 {
2165 }
2166 
2167 #endif
2168 
2169 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
2170 			 struct netlink_ext_ack *extack)
2171 {
2172 	struct net *net = sock_net(skb->sk);
2173 	struct tcmsg *tcm = nlmsg_data(n);
2174 	struct nlattr *tca[TCA_MAX + 1];
2175 	struct net_device *dev;
2176 	struct Qdisc *q = NULL;
2177 	const struct Qdisc_class_ops *cops;
2178 	unsigned long cl = 0;
2179 	unsigned long new_cl;
2180 	u32 portid;
2181 	u32 clid;
2182 	u32 qid;
2183 	int err;
2184 
2185 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2186 				     rtm_tca_policy, extack);
2187 	if (err < 0)
2188 		return err;
2189 
2190 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2191 	if (!dev)
2192 		return -ENODEV;
2193 
2194 	/*
2195 	   parent == TC_H_UNSPEC - unspecified parent.
2196 	   parent == TC_H_ROOT   - class is root, which has no parent.
2197 	   parent == X:0	 - parent is root class.
2198 	   parent == X:Y	 - parent is a node in hierarchy.
2199 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2200 
2201 	   handle == 0:0	 - generate handle from kernel pool.
2202 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2203 	   handle == X:Y	 - clear.
2204 	   handle == X:0	 - root class.
2205 	 */
2206 
2207 	/* Step 1. Determine qdisc handle X:0 */
2208 
2209 	portid = tcm->tcm_parent;
2210 	clid = tcm->tcm_handle;
2211 	qid = TC_H_MAJ(clid);
2212 
2213 	if (portid != TC_H_ROOT) {
2214 		u32 qid1 = TC_H_MAJ(portid);
2215 
2216 		if (qid && qid1) {
2217 			/* If both majors are known, they must be identical. */
2218 			if (qid != qid1)
2219 				return -EINVAL;
2220 		} else if (qid1) {
2221 			qid = qid1;
2222 		} else if (qid == 0)
2223 			qid = rtnl_dereference(dev->qdisc)->handle;
2224 
2225 		/* Now qid is genuine qdisc handle consistent
2226 		 * both with parent and child.
2227 		 *
2228 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2229 		 */
2230 		if (portid)
2231 			portid = TC_H_MAKE(qid, portid);
2232 	} else {
2233 		if (qid == 0)
2234 			qid = rtnl_dereference(dev->qdisc)->handle;
2235 	}
2236 
2237 	/* OK. Locate qdisc */
2238 	q = qdisc_lookup(dev, qid);
2239 	if (!q)
2240 		return -ENOENT;
2241 
2242 	/* An check that it supports classes */
2243 	cops = q->ops->cl_ops;
2244 	if (cops == NULL)
2245 		return -EINVAL;
2246 
2247 	/* Now try to get class */
2248 	if (clid == 0) {
2249 		if (portid == TC_H_ROOT)
2250 			clid = qid;
2251 	} else
2252 		clid = TC_H_MAKE(qid, clid);
2253 
2254 	if (clid)
2255 		cl = cops->find(q, clid);
2256 
2257 	if (cl == 0) {
2258 		err = -ENOENT;
2259 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2260 		    !(n->nlmsg_flags & NLM_F_CREATE))
2261 			goto out;
2262 	} else {
2263 		switch (n->nlmsg_type) {
2264 		case RTM_NEWTCLASS:
2265 			err = -EEXIST;
2266 			if (n->nlmsg_flags & NLM_F_EXCL)
2267 				goto out;
2268 			break;
2269 		case RTM_DELTCLASS:
2270 			err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2271 			/* Unbind the class with flilters with 0 */
2272 			tc_bind_tclass(q, portid, clid, 0);
2273 			goto out;
2274 		case RTM_GETTCLASS:
2275 			err = tclass_get_notify(net, skb, n, q, cl, extack);
2276 			goto out;
2277 		default:
2278 			err = -EINVAL;
2279 			goto out;
2280 		}
2281 	}
2282 
2283 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2284 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2285 		return -EOPNOTSUPP;
2286 	}
2287 
2288 	new_cl = cl;
2289 	err = -EOPNOTSUPP;
2290 	if (cops->change)
2291 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2292 	if (err == 0) {
2293 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS, extack);
2294 		/* We just create a new class, need to do reverse binding. */
2295 		if (cl != new_cl)
2296 			tc_bind_tclass(q, portid, clid, new_cl);
2297 	}
2298 out:
2299 	return err;
2300 }
2301 
2302 struct qdisc_dump_args {
2303 	struct qdisc_walker	w;
2304 	struct sk_buff		*skb;
2305 	struct netlink_callback	*cb;
2306 };
2307 
2308 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2309 			    struct qdisc_walker *arg)
2310 {
2311 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2312 
2313 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2314 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2315 			      RTM_NEWTCLASS, NULL);
2316 }
2317 
2318 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2319 				struct tcmsg *tcm, struct netlink_callback *cb,
2320 				int *t_p, int s_t)
2321 {
2322 	struct qdisc_dump_args arg;
2323 
2324 	if (tc_qdisc_dump_ignore(q, false) ||
2325 	    *t_p < s_t || !q->ops->cl_ops ||
2326 	    (tcm->tcm_parent &&
2327 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2328 		(*t_p)++;
2329 		return 0;
2330 	}
2331 	if (*t_p > s_t)
2332 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2333 	arg.w.fn = qdisc_class_dump;
2334 	arg.skb = skb;
2335 	arg.cb = cb;
2336 	arg.w.stop  = 0;
2337 	arg.w.skip = cb->args[1];
2338 	arg.w.count = 0;
2339 	q->ops->cl_ops->walk(q, &arg.w);
2340 	cb->args[1] = arg.w.count;
2341 	if (arg.w.stop)
2342 		return -1;
2343 	(*t_p)++;
2344 	return 0;
2345 }
2346 
2347 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2348 			       struct tcmsg *tcm, struct netlink_callback *cb,
2349 			       int *t_p, int s_t, bool recur)
2350 {
2351 	struct Qdisc *q;
2352 	int b;
2353 
2354 	if (!root)
2355 		return 0;
2356 
2357 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2358 		return -1;
2359 
2360 	if (!qdisc_dev(root) || !recur)
2361 		return 0;
2362 
2363 	if (tcm->tcm_parent) {
2364 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2365 		if (q && q != root &&
2366 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2367 			return -1;
2368 		return 0;
2369 	}
2370 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2371 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2372 			return -1;
2373 	}
2374 
2375 	return 0;
2376 }
2377 
2378 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2379 {
2380 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2381 	struct net *net = sock_net(skb->sk);
2382 	struct netdev_queue *dev_queue;
2383 	struct net_device *dev;
2384 	int t, s_t;
2385 
2386 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2387 		return 0;
2388 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2389 	if (!dev)
2390 		return 0;
2391 
2392 	s_t = cb->args[0];
2393 	t = 0;
2394 
2395 	if (tc_dump_tclass_root(rtnl_dereference(dev->qdisc),
2396 				skb, tcm, cb, &t, s_t, true) < 0)
2397 		goto done;
2398 
2399 	dev_queue = dev_ingress_queue(dev);
2400 	if (dev_queue &&
2401 	    tc_dump_tclass_root(rtnl_dereference(dev_queue->qdisc_sleeping),
2402 				skb, tcm, cb, &t, s_t, false) < 0)
2403 		goto done;
2404 
2405 done:
2406 	cb->args[0] = t;
2407 
2408 	dev_put(dev);
2409 	return skb->len;
2410 }
2411 
2412 #ifdef CONFIG_PROC_FS
2413 static int psched_show(struct seq_file *seq, void *v)
2414 {
2415 	seq_printf(seq, "%08x %08x %08x %08x\n",
2416 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2417 		   1000000,
2418 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2419 
2420 	return 0;
2421 }
2422 
2423 static int __net_init psched_net_init(struct net *net)
2424 {
2425 	struct proc_dir_entry *e;
2426 
2427 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2428 	if (e == NULL)
2429 		return -ENOMEM;
2430 
2431 	return 0;
2432 }
2433 
2434 static void __net_exit psched_net_exit(struct net *net)
2435 {
2436 	remove_proc_entry("psched", net->proc_net);
2437 }
2438 #else
2439 static int __net_init psched_net_init(struct net *net)
2440 {
2441 	return 0;
2442 }
2443 
2444 static void __net_exit psched_net_exit(struct net *net)
2445 {
2446 }
2447 #endif
2448 
2449 static struct pernet_operations psched_net_ops = {
2450 	.init = psched_net_init,
2451 	.exit = psched_net_exit,
2452 };
2453 
2454 #if IS_ENABLED(CONFIG_RETPOLINE)
2455 DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper);
2456 #endif
2457 
2458 static int __init pktsched_init(void)
2459 {
2460 	int err;
2461 
2462 	err = register_pernet_subsys(&psched_net_ops);
2463 	if (err) {
2464 		pr_err("pktsched_init: "
2465 		       "cannot initialize per netns operations\n");
2466 		return err;
2467 	}
2468 
2469 	register_qdisc(&pfifo_fast_ops);
2470 	register_qdisc(&pfifo_qdisc_ops);
2471 	register_qdisc(&bfifo_qdisc_ops);
2472 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2473 	register_qdisc(&mq_qdisc_ops);
2474 	register_qdisc(&noqueue_qdisc_ops);
2475 
2476 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2477 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2478 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2479 		      0);
2480 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2481 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2482 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2483 		      0);
2484 
2485 	tc_wrapper_init();
2486 
2487 	return 0;
2488 }
2489 
2490 subsys_initcall(pktsched_init);
2491