xref: /linux/net/sched/sch_api.c (revision e814f3fd16acfb7f9966773953de8f740a1e3202)
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(NET_SCH_ALIAS_PREFIX "%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 
597 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
598 {
599 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
600 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
601 			txt, qdisc->ops->id, qdisc->handle >> 16);
602 		qdisc->flags |= TCQ_F_WARN_NONWC;
603 	}
604 }
605 EXPORT_SYMBOL(qdisc_warn_nonwc);
606 
607 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
608 {
609 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
610 						 timer);
611 
612 	rcu_read_lock();
613 	__netif_schedule(qdisc_root(wd->qdisc));
614 	rcu_read_unlock();
615 
616 	return HRTIMER_NORESTART;
617 }
618 
619 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
620 				 clockid_t clockid)
621 {
622 	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
623 	wd->timer.function = qdisc_watchdog;
624 	wd->qdisc = qdisc;
625 }
626 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
627 
628 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
629 {
630 	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
631 }
632 EXPORT_SYMBOL(qdisc_watchdog_init);
633 
634 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
635 				      u64 delta_ns)
636 {
637 	bool deactivated;
638 
639 	rcu_read_lock();
640 	deactivated = test_bit(__QDISC_STATE_DEACTIVATED,
641 			       &qdisc_root_sleeping(wd->qdisc)->state);
642 	rcu_read_unlock();
643 	if (deactivated)
644 		return;
645 
646 	if (hrtimer_is_queued(&wd->timer)) {
647 		u64 softexpires;
648 
649 		softexpires = ktime_to_ns(hrtimer_get_softexpires(&wd->timer));
650 		/* If timer is already set in [expires, expires + delta_ns],
651 		 * do not reprogram it.
652 		 */
653 		if (softexpires - expires <= delta_ns)
654 			return;
655 	}
656 
657 	hrtimer_start_range_ns(&wd->timer,
658 			       ns_to_ktime(expires),
659 			       delta_ns,
660 			       HRTIMER_MODE_ABS_PINNED);
661 }
662 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
663 
664 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
665 {
666 	hrtimer_cancel(&wd->timer);
667 }
668 EXPORT_SYMBOL(qdisc_watchdog_cancel);
669 
670 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
671 {
672 	struct hlist_head *h;
673 	unsigned int i;
674 
675 	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
676 
677 	if (h != NULL) {
678 		for (i = 0; i < n; i++)
679 			INIT_HLIST_HEAD(&h[i]);
680 	}
681 	return h;
682 }
683 
684 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
685 {
686 	struct Qdisc_class_common *cl;
687 	struct hlist_node *next;
688 	struct hlist_head *nhash, *ohash;
689 	unsigned int nsize, nmask, osize;
690 	unsigned int i, h;
691 
692 	/* Rehash when load factor exceeds 0.75 */
693 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
694 		return;
695 	nsize = clhash->hashsize * 2;
696 	nmask = nsize - 1;
697 	nhash = qdisc_class_hash_alloc(nsize);
698 	if (nhash == NULL)
699 		return;
700 
701 	ohash = clhash->hash;
702 	osize = clhash->hashsize;
703 
704 	sch_tree_lock(sch);
705 	for (i = 0; i < osize; i++) {
706 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
707 			h = qdisc_class_hash(cl->classid, nmask);
708 			hlist_add_head(&cl->hnode, &nhash[h]);
709 		}
710 	}
711 	clhash->hash     = nhash;
712 	clhash->hashsize = nsize;
713 	clhash->hashmask = nmask;
714 	sch_tree_unlock(sch);
715 
716 	kvfree(ohash);
717 }
718 EXPORT_SYMBOL(qdisc_class_hash_grow);
719 
720 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
721 {
722 	unsigned int size = 4;
723 
724 	clhash->hash = qdisc_class_hash_alloc(size);
725 	if (!clhash->hash)
726 		return -ENOMEM;
727 	clhash->hashsize  = size;
728 	clhash->hashmask  = size - 1;
729 	clhash->hashelems = 0;
730 	return 0;
731 }
732 EXPORT_SYMBOL(qdisc_class_hash_init);
733 
734 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
735 {
736 	kvfree(clhash->hash);
737 }
738 EXPORT_SYMBOL(qdisc_class_hash_destroy);
739 
740 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
741 			     struct Qdisc_class_common *cl)
742 {
743 	unsigned int h;
744 
745 	INIT_HLIST_NODE(&cl->hnode);
746 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
747 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
748 	clhash->hashelems++;
749 }
750 EXPORT_SYMBOL(qdisc_class_hash_insert);
751 
752 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
753 			     struct Qdisc_class_common *cl)
754 {
755 	hlist_del(&cl->hnode);
756 	clhash->hashelems--;
757 }
758 EXPORT_SYMBOL(qdisc_class_hash_remove);
759 
760 /* Allocate an unique handle from space managed by kernel
761  * Possible range is [8000-FFFF]:0000 (0x8000 values)
762  */
763 static u32 qdisc_alloc_handle(struct net_device *dev)
764 {
765 	int i = 0x8000;
766 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
767 
768 	do {
769 		autohandle += TC_H_MAKE(0x10000U, 0);
770 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
771 			autohandle = TC_H_MAKE(0x80000000U, 0);
772 		if (!qdisc_lookup(dev, autohandle))
773 			return autohandle;
774 		cond_resched();
775 	} while	(--i > 0);
776 
777 	return 0;
778 }
779 
780 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
781 {
782 	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
783 	const struct Qdisc_class_ops *cops;
784 	unsigned long cl;
785 	u32 parentid;
786 	bool notify;
787 	int drops;
788 
789 	if (n == 0 && len == 0)
790 		return;
791 	drops = max_t(int, n, 0);
792 	rcu_read_lock();
793 	while ((parentid = sch->parent)) {
794 		if (parentid == TC_H_ROOT)
795 			break;
796 
797 		if (sch->flags & TCQ_F_NOPARENT)
798 			break;
799 		/* Notify parent qdisc only if child qdisc becomes empty.
800 		 *
801 		 * If child was empty even before update then backlog
802 		 * counter is screwed and we skip notification because
803 		 * parent class is already passive.
804 		 *
805 		 * If the original child was offloaded then it is allowed
806 		 * to be seem as empty, so the parent is notified anyway.
807 		 */
808 		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
809 						       !qdisc_is_offloaded);
810 		/* TODO: perform the search on a per txq basis */
811 		sch = qdisc_lookup_rcu(qdisc_dev(sch), TC_H_MAJ(parentid));
812 		if (sch == NULL) {
813 			WARN_ON_ONCE(parentid != TC_H_ROOT);
814 			break;
815 		}
816 		cops = sch->ops->cl_ops;
817 		if (notify && cops->qlen_notify) {
818 			cl = cops->find(sch, parentid);
819 			cops->qlen_notify(sch, cl);
820 		}
821 		sch->q.qlen -= n;
822 		sch->qstats.backlog -= len;
823 		__qdisc_qstats_drop(sch, drops);
824 	}
825 	rcu_read_unlock();
826 }
827 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
828 
829 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
830 			      void *type_data)
831 {
832 	struct net_device *dev = qdisc_dev(sch);
833 	int err;
834 
835 	sch->flags &= ~TCQ_F_OFFLOADED;
836 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
837 		return 0;
838 
839 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
840 	if (err == -EOPNOTSUPP)
841 		return 0;
842 
843 	if (!err)
844 		sch->flags |= TCQ_F_OFFLOADED;
845 
846 	return err;
847 }
848 EXPORT_SYMBOL(qdisc_offload_dump_helper);
849 
850 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
851 				struct Qdisc *new, struct Qdisc *old,
852 				enum tc_setup_type type, void *type_data,
853 				struct netlink_ext_ack *extack)
854 {
855 	bool any_qdisc_is_offloaded;
856 	int err;
857 
858 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
859 		return;
860 
861 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
862 
863 	/* Don't report error if the graft is part of destroy operation. */
864 	if (!err || !new || new == &noop_qdisc)
865 		return;
866 
867 	/* Don't report error if the parent, the old child and the new
868 	 * one are not offloaded.
869 	 */
870 	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
871 	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
872 	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
873 
874 	if (any_qdisc_is_offloaded)
875 		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
876 }
877 EXPORT_SYMBOL(qdisc_offload_graft_helper);
878 
879 void qdisc_offload_query_caps(struct net_device *dev,
880 			      enum tc_setup_type type,
881 			      void *caps, size_t caps_len)
882 {
883 	const struct net_device_ops *ops = dev->netdev_ops;
884 	struct tc_query_caps_base base = {
885 		.type = type,
886 		.caps = caps,
887 	};
888 
889 	memset(caps, 0, caps_len);
890 
891 	if (ops->ndo_setup_tc)
892 		ops->ndo_setup_tc(dev, TC_QUERY_CAPS, &base);
893 }
894 EXPORT_SYMBOL(qdisc_offload_query_caps);
895 
896 static void qdisc_offload_graft_root(struct net_device *dev,
897 				     struct Qdisc *new, struct Qdisc *old,
898 				     struct netlink_ext_ack *extack)
899 {
900 	struct tc_root_qopt_offload graft_offload = {
901 		.command	= TC_ROOT_GRAFT,
902 		.handle		= new ? new->handle : 0,
903 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
904 				  (old && old->flags & TCQ_F_INGRESS),
905 	};
906 
907 	qdisc_offload_graft_helper(dev, NULL, new, old,
908 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
909 }
910 
911 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
912 			 u32 portid, u32 seq, u16 flags, int event,
913 			 struct netlink_ext_ack *extack)
914 {
915 	struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
916 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
917 	struct tcmsg *tcm;
918 	struct nlmsghdr  *nlh;
919 	unsigned char *b = skb_tail_pointer(skb);
920 	struct gnet_dump d;
921 	struct qdisc_size_table *stab;
922 	u32 block_index;
923 	__u32 qlen;
924 
925 	cond_resched();
926 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
927 	if (!nlh)
928 		goto out_nlmsg_trim;
929 	tcm = nlmsg_data(nlh);
930 	tcm->tcm_family = AF_UNSPEC;
931 	tcm->tcm__pad1 = 0;
932 	tcm->tcm__pad2 = 0;
933 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
934 	tcm->tcm_parent = clid;
935 	tcm->tcm_handle = q->handle;
936 	tcm->tcm_info = refcount_read(&q->refcnt);
937 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
938 		goto nla_put_failure;
939 	if (q->ops->ingress_block_get) {
940 		block_index = q->ops->ingress_block_get(q);
941 		if (block_index &&
942 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
943 			goto nla_put_failure;
944 	}
945 	if (q->ops->egress_block_get) {
946 		block_index = q->ops->egress_block_get(q);
947 		if (block_index &&
948 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
949 			goto nla_put_failure;
950 	}
951 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
952 		goto nla_put_failure;
953 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
954 		goto nla_put_failure;
955 	qlen = qdisc_qlen_sum(q);
956 
957 	stab = rtnl_dereference(q->stab);
958 	if (stab && qdisc_dump_stab(skb, stab) < 0)
959 		goto nla_put_failure;
960 
961 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
962 					 NULL, &d, TCA_PAD) < 0)
963 		goto nla_put_failure;
964 
965 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
966 		goto nla_put_failure;
967 
968 	if (qdisc_is_percpu_stats(q)) {
969 		cpu_bstats = q->cpu_bstats;
970 		cpu_qstats = q->cpu_qstats;
971 	}
972 
973 	if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
974 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
975 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
976 		goto nla_put_failure;
977 
978 	if (gnet_stats_finish_copy(&d) < 0)
979 		goto nla_put_failure;
980 
981 	if (extack && extack->_msg &&
982 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
983 		goto out_nlmsg_trim;
984 
985 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
986 
987 	return skb->len;
988 
989 out_nlmsg_trim:
990 nla_put_failure:
991 	nlmsg_trim(skb, b);
992 	return -1;
993 }
994 
995 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
996 {
997 	if (q->flags & TCQ_F_BUILTIN)
998 		return true;
999 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
1000 		return true;
1001 
1002 	return false;
1003 }
1004 
1005 static int qdisc_get_notify(struct net *net, struct sk_buff *oskb,
1006 			    struct nlmsghdr *n, u32 clid, struct Qdisc *q,
1007 			    struct netlink_ext_ack *extack)
1008 {
1009 	struct sk_buff *skb;
1010 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1011 
1012 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1013 	if (!skb)
1014 		return -ENOBUFS;
1015 
1016 	if (!tc_qdisc_dump_ignore(q, false)) {
1017 		if (tc_fill_qdisc(skb, q, clid, portid, n->nlmsg_seq, 0,
1018 				  RTM_NEWQDISC, extack) < 0)
1019 			goto err_out;
1020 	}
1021 
1022 	if (skb->len)
1023 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1024 				      n->nlmsg_flags & NLM_F_ECHO);
1025 
1026 err_out:
1027 	kfree_skb(skb);
1028 	return -EINVAL;
1029 }
1030 
1031 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1032 			struct nlmsghdr *n, u32 clid,
1033 			struct Qdisc *old, struct Qdisc *new,
1034 			struct netlink_ext_ack *extack)
1035 {
1036 	struct sk_buff *skb;
1037 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1038 
1039 	if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
1040 		return 0;
1041 
1042 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1043 	if (!skb)
1044 		return -ENOBUFS;
1045 
1046 	if (old && !tc_qdisc_dump_ignore(old, false)) {
1047 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1048 				  0, RTM_DELQDISC, extack) < 0)
1049 			goto err_out;
1050 	}
1051 	if (new && !tc_qdisc_dump_ignore(new, false)) {
1052 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1053 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC, extack) < 0)
1054 			goto err_out;
1055 	}
1056 
1057 	if (skb->len)
1058 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1059 				      n->nlmsg_flags & NLM_F_ECHO);
1060 
1061 err_out:
1062 	kfree_skb(skb);
1063 	return -EINVAL;
1064 }
1065 
1066 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1067 			       struct nlmsghdr *n, u32 clid,
1068 			       struct Qdisc *old, struct Qdisc *new,
1069 			       struct netlink_ext_ack *extack)
1070 {
1071 	if (new || old)
1072 		qdisc_notify(net, skb, n, clid, old, new, extack);
1073 
1074 	if (old)
1075 		qdisc_put(old);
1076 }
1077 
1078 static void qdisc_clear_nolock(struct Qdisc *sch)
1079 {
1080 	sch->flags &= ~TCQ_F_NOLOCK;
1081 	if (!(sch->flags & TCQ_F_CPUSTATS))
1082 		return;
1083 
1084 	free_percpu(sch->cpu_bstats);
1085 	free_percpu(sch->cpu_qstats);
1086 	sch->cpu_bstats = NULL;
1087 	sch->cpu_qstats = NULL;
1088 	sch->flags &= ~TCQ_F_CPUSTATS;
1089 }
1090 
1091 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1092  * to device "dev".
1093  *
1094  * When appropriate send a netlink notification using 'skb'
1095  * and "n".
1096  *
1097  * On success, destroy old qdisc.
1098  */
1099 
1100 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1101 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1102 		       struct Qdisc *new, struct Qdisc *old,
1103 		       struct netlink_ext_ack *extack)
1104 {
1105 	struct Qdisc *q = old;
1106 	struct net *net = dev_net(dev);
1107 
1108 	if (parent == NULL) {
1109 		unsigned int i, num_q, ingress;
1110 		struct netdev_queue *dev_queue;
1111 
1112 		ingress = 0;
1113 		num_q = dev->num_tx_queues;
1114 		if ((q && q->flags & TCQ_F_INGRESS) ||
1115 		    (new && new->flags & TCQ_F_INGRESS)) {
1116 			ingress = 1;
1117 			dev_queue = dev_ingress_queue(dev);
1118 			if (!dev_queue) {
1119 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1120 				return -ENOENT;
1121 			}
1122 
1123 			q = rtnl_dereference(dev_queue->qdisc_sleeping);
1124 
1125 			/* This is the counterpart of that qdisc_refcount_inc_nz() call in
1126 			 * __tcf_qdisc_find() for filter requests.
1127 			 */
1128 			if (!qdisc_refcount_dec_if_one(q)) {
1129 				NL_SET_ERR_MSG(extack,
1130 					       "Current ingress or clsact Qdisc has ongoing filter requests");
1131 				return -EBUSY;
1132 			}
1133 		}
1134 
1135 		if (dev->flags & IFF_UP)
1136 			dev_deactivate(dev);
1137 
1138 		qdisc_offload_graft_root(dev, new, old, extack);
1139 
1140 		if (new && new->ops->attach && !ingress)
1141 			goto skip;
1142 
1143 		if (!ingress) {
1144 			for (i = 0; i < num_q; i++) {
1145 				dev_queue = netdev_get_tx_queue(dev, i);
1146 				old = dev_graft_qdisc(dev_queue, new);
1147 
1148 				if (new && i > 0)
1149 					qdisc_refcount_inc(new);
1150 				qdisc_put(old);
1151 			}
1152 		} else {
1153 			old = dev_graft_qdisc(dev_queue, NULL);
1154 
1155 			/* {ingress,clsact}_destroy() @old before grafting @new to avoid
1156 			 * unprotected concurrent accesses to net_device::miniq_{in,e}gress
1157 			 * pointer(s) in mini_qdisc_pair_swap().
1158 			 */
1159 			qdisc_notify(net, skb, n, classid, old, new, extack);
1160 			qdisc_destroy(old);
1161 
1162 			dev_graft_qdisc(dev_queue, new);
1163 		}
1164 
1165 skip:
1166 		if (!ingress) {
1167 			old = rtnl_dereference(dev->qdisc);
1168 			if (new && !new->ops->attach)
1169 				qdisc_refcount_inc(new);
1170 			rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1171 
1172 			notify_and_destroy(net, skb, n, classid, old, new, extack);
1173 
1174 			if (new && new->ops->attach)
1175 				new->ops->attach(new);
1176 		}
1177 
1178 		if (dev->flags & IFF_UP)
1179 			dev_activate(dev);
1180 	} else {
1181 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1182 		unsigned long cl;
1183 		int err;
1184 
1185 		/* Only support running class lockless if parent is lockless */
1186 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1187 			qdisc_clear_nolock(new);
1188 
1189 		if (!cops || !cops->graft)
1190 			return -EOPNOTSUPP;
1191 
1192 		cl = cops->find(parent, classid);
1193 		if (!cl) {
1194 			NL_SET_ERR_MSG(extack, "Specified class not found");
1195 			return -ENOENT;
1196 		}
1197 
1198 		if (new && new->ops == &noqueue_qdisc_ops) {
1199 			NL_SET_ERR_MSG(extack, "Cannot assign noqueue to a class");
1200 			return -EINVAL;
1201 		}
1202 
1203 		if (new &&
1204 		    !(parent->flags & TCQ_F_MQROOT) &&
1205 		    rcu_access_pointer(new->stab)) {
1206 			NL_SET_ERR_MSG(extack, "STAB not supported on a non root");
1207 			return -EINVAL;
1208 		}
1209 		err = cops->graft(parent, cl, new, &old, extack);
1210 		if (err)
1211 			return err;
1212 		notify_and_destroy(net, skb, n, classid, old, new, extack);
1213 	}
1214 	return 0;
1215 }
1216 
1217 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1218 				   struct netlink_ext_ack *extack)
1219 {
1220 	u32 block_index;
1221 
1222 	if (tca[TCA_INGRESS_BLOCK]) {
1223 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1224 
1225 		if (!block_index) {
1226 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1227 			return -EINVAL;
1228 		}
1229 		if (!sch->ops->ingress_block_set) {
1230 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1231 			return -EOPNOTSUPP;
1232 		}
1233 		sch->ops->ingress_block_set(sch, block_index);
1234 	}
1235 	if (tca[TCA_EGRESS_BLOCK]) {
1236 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1237 
1238 		if (!block_index) {
1239 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1240 			return -EINVAL;
1241 		}
1242 		if (!sch->ops->egress_block_set) {
1243 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1244 			return -EOPNOTSUPP;
1245 		}
1246 		sch->ops->egress_block_set(sch, block_index);
1247 	}
1248 	return 0;
1249 }
1250 
1251 /*
1252    Allocate and initialize new qdisc.
1253 
1254    Parameters are passed via opt.
1255  */
1256 
1257 static struct Qdisc *qdisc_create(struct net_device *dev,
1258 				  struct netdev_queue *dev_queue,
1259 				  u32 parent, u32 handle,
1260 				  struct nlattr **tca, int *errp,
1261 				  struct netlink_ext_ack *extack)
1262 {
1263 	int err;
1264 	struct nlattr *kind = tca[TCA_KIND];
1265 	struct Qdisc *sch;
1266 	struct Qdisc_ops *ops;
1267 	struct qdisc_size_table *stab;
1268 
1269 	ops = qdisc_lookup_ops(kind);
1270 #ifdef CONFIG_MODULES
1271 	if (ops == NULL && kind != NULL) {
1272 		char name[IFNAMSIZ];
1273 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1274 			/* We dropped the RTNL semaphore in order to
1275 			 * perform the module load.  So, even if we
1276 			 * succeeded in loading the module we have to
1277 			 * tell the caller to replay the request.  We
1278 			 * indicate this using -EAGAIN.
1279 			 * We replay the request because the device may
1280 			 * go away in the mean time.
1281 			 */
1282 			rtnl_unlock();
1283 			request_module(NET_SCH_ALIAS_PREFIX "%s", name);
1284 			rtnl_lock();
1285 			ops = qdisc_lookup_ops(kind);
1286 			if (ops != NULL) {
1287 				/* We will try again qdisc_lookup_ops,
1288 				 * so don't keep a reference.
1289 				 */
1290 				module_put(ops->owner);
1291 				err = -EAGAIN;
1292 				goto err_out;
1293 			}
1294 		}
1295 	}
1296 #endif
1297 
1298 	err = -ENOENT;
1299 	if (!ops) {
1300 		NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1301 		goto err_out;
1302 	}
1303 
1304 	sch = qdisc_alloc(dev_queue, ops, extack);
1305 	if (IS_ERR(sch)) {
1306 		err = PTR_ERR(sch);
1307 		goto err_out2;
1308 	}
1309 
1310 	sch->parent = parent;
1311 
1312 	if (handle == TC_H_INGRESS) {
1313 		if (!(sch->flags & TCQ_F_INGRESS)) {
1314 			NL_SET_ERR_MSG(extack,
1315 				       "Specified parent ID is reserved for ingress and clsact Qdiscs");
1316 			err = -EINVAL;
1317 			goto err_out3;
1318 		}
1319 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1320 	} else {
1321 		if (handle == 0) {
1322 			handle = qdisc_alloc_handle(dev);
1323 			if (handle == 0) {
1324 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1325 				err = -ENOSPC;
1326 				goto err_out3;
1327 			}
1328 		}
1329 		if (!netif_is_multiqueue(dev))
1330 			sch->flags |= TCQ_F_ONETXQUEUE;
1331 	}
1332 
1333 	sch->handle = handle;
1334 
1335 	/* This exist to keep backward compatible with a userspace
1336 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1337 	 * facility on older kernels by setting tx_queue_len=0 (prior
1338 	 * to qdisc init), and then forgot to reinit tx_queue_len
1339 	 * before again attaching a qdisc.
1340 	 */
1341 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1342 		WRITE_ONCE(dev->tx_queue_len, DEFAULT_TX_QUEUE_LEN);
1343 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1344 	}
1345 
1346 	err = qdisc_block_indexes_set(sch, tca, extack);
1347 	if (err)
1348 		goto err_out3;
1349 
1350 	if (tca[TCA_STAB]) {
1351 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1352 		if (IS_ERR(stab)) {
1353 			err = PTR_ERR(stab);
1354 			goto err_out3;
1355 		}
1356 		rcu_assign_pointer(sch->stab, stab);
1357 	}
1358 
1359 	if (ops->init) {
1360 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1361 		if (err != 0)
1362 			goto err_out4;
1363 	}
1364 
1365 	if (tca[TCA_RATE]) {
1366 		err = -EOPNOTSUPP;
1367 		if (sch->flags & TCQ_F_MQROOT) {
1368 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1369 			goto err_out4;
1370 		}
1371 
1372 		err = gen_new_estimator(&sch->bstats,
1373 					sch->cpu_bstats,
1374 					&sch->rate_est,
1375 					NULL,
1376 					true,
1377 					tca[TCA_RATE]);
1378 		if (err) {
1379 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1380 			goto err_out4;
1381 		}
1382 	}
1383 
1384 	qdisc_hash_add(sch, false);
1385 	trace_qdisc_create(ops, dev, parent);
1386 
1387 	return sch;
1388 
1389 err_out4:
1390 	/* Even if ops->init() failed, we call ops->destroy()
1391 	 * like qdisc_create_dflt().
1392 	 */
1393 	if (ops->destroy)
1394 		ops->destroy(sch);
1395 	qdisc_put_stab(rtnl_dereference(sch->stab));
1396 err_out3:
1397 	lockdep_unregister_key(&sch->root_lock_key);
1398 	netdev_put(dev, &sch->dev_tracker);
1399 	qdisc_free(sch);
1400 err_out2:
1401 	module_put(ops->owner);
1402 err_out:
1403 	*errp = err;
1404 	return NULL;
1405 }
1406 
1407 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1408 			struct netlink_ext_ack *extack)
1409 {
1410 	struct qdisc_size_table *ostab, *stab = NULL;
1411 	int err = 0;
1412 
1413 	if (tca[TCA_OPTIONS]) {
1414 		if (!sch->ops->change) {
1415 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1416 			return -EINVAL;
1417 		}
1418 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1419 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1420 			return -EOPNOTSUPP;
1421 		}
1422 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1423 		if (err)
1424 			return err;
1425 	}
1426 
1427 	if (tca[TCA_STAB]) {
1428 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1429 		if (IS_ERR(stab))
1430 			return PTR_ERR(stab);
1431 	}
1432 
1433 	ostab = rtnl_dereference(sch->stab);
1434 	rcu_assign_pointer(sch->stab, stab);
1435 	qdisc_put_stab(ostab);
1436 
1437 	if (tca[TCA_RATE]) {
1438 		/* NB: ignores errors from replace_estimator
1439 		   because change can't be undone. */
1440 		if (sch->flags & TCQ_F_MQROOT)
1441 			goto out;
1442 		gen_replace_estimator(&sch->bstats,
1443 				      sch->cpu_bstats,
1444 				      &sch->rate_est,
1445 				      NULL,
1446 				      true,
1447 				      tca[TCA_RATE]);
1448 	}
1449 out:
1450 	return 0;
1451 }
1452 
1453 struct check_loop_arg {
1454 	struct qdisc_walker	w;
1455 	struct Qdisc		*p;
1456 	int			depth;
1457 };
1458 
1459 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1460 			 struct qdisc_walker *w);
1461 
1462 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1463 {
1464 	struct check_loop_arg	arg;
1465 
1466 	if (q->ops->cl_ops == NULL)
1467 		return 0;
1468 
1469 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1470 	arg.w.fn = check_loop_fn;
1471 	arg.depth = depth;
1472 	arg.p = p;
1473 	q->ops->cl_ops->walk(q, &arg.w);
1474 	return arg.w.stop ? -ELOOP : 0;
1475 }
1476 
1477 static int
1478 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1479 {
1480 	struct Qdisc *leaf;
1481 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1482 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1483 
1484 	leaf = cops->leaf(q, cl);
1485 	if (leaf) {
1486 		if (leaf == arg->p || arg->depth > 7)
1487 			return -ELOOP;
1488 		return check_loop(leaf, arg->p, arg->depth + 1);
1489 	}
1490 	return 0;
1491 }
1492 
1493 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1494 	[TCA_KIND]		= { .type = NLA_STRING },
1495 	[TCA_RATE]		= { .type = NLA_BINARY,
1496 				    .len = sizeof(struct tc_estimator) },
1497 	[TCA_STAB]		= { .type = NLA_NESTED },
1498 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1499 	[TCA_CHAIN]		= { .type = NLA_U32 },
1500 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1501 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1502 };
1503 
1504 /*
1505  * Delete/get qdisc.
1506  */
1507 
1508 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1509 			struct netlink_ext_ack *extack)
1510 {
1511 	struct net *net = sock_net(skb->sk);
1512 	struct tcmsg *tcm = nlmsg_data(n);
1513 	struct nlattr *tca[TCA_MAX + 1];
1514 	struct net_device *dev;
1515 	u32 clid;
1516 	struct Qdisc *q = NULL;
1517 	struct Qdisc *p = NULL;
1518 	int err;
1519 
1520 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1521 				     rtm_tca_policy, extack);
1522 	if (err < 0)
1523 		return err;
1524 
1525 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1526 	if (!dev)
1527 		return -ENODEV;
1528 
1529 	clid = tcm->tcm_parent;
1530 	if (clid) {
1531 		if (clid != TC_H_ROOT) {
1532 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1533 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1534 				if (!p) {
1535 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1536 					return -ENOENT;
1537 				}
1538 				q = qdisc_leaf(p, clid);
1539 			} else if (dev_ingress_queue(dev)) {
1540 				q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1541 			}
1542 		} else {
1543 			q = rtnl_dereference(dev->qdisc);
1544 		}
1545 		if (!q) {
1546 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1547 			return -ENOENT;
1548 		}
1549 
1550 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1551 			NL_SET_ERR_MSG(extack, "Invalid handle");
1552 			return -EINVAL;
1553 		}
1554 	} else {
1555 		q = qdisc_lookup(dev, tcm->tcm_handle);
1556 		if (!q) {
1557 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1558 			return -ENOENT;
1559 		}
1560 	}
1561 
1562 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1563 		NL_SET_ERR_MSG(extack, "Invalid qdisc name: must match existing qdisc");
1564 		return -EINVAL;
1565 	}
1566 
1567 	if (n->nlmsg_type == RTM_DELQDISC) {
1568 		if (!clid) {
1569 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1570 			return -EINVAL;
1571 		}
1572 		if (q->handle == 0) {
1573 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1574 			return -ENOENT;
1575 		}
1576 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1577 		if (err != 0)
1578 			return err;
1579 	} else {
1580 		qdisc_get_notify(net, skb, n, clid, q, NULL);
1581 	}
1582 	return 0;
1583 }
1584 
1585 static bool req_create_or_replace(struct nlmsghdr *n)
1586 {
1587 	return (n->nlmsg_flags & NLM_F_CREATE &&
1588 		n->nlmsg_flags & NLM_F_REPLACE);
1589 }
1590 
1591 static bool req_create_exclusive(struct nlmsghdr *n)
1592 {
1593 	return (n->nlmsg_flags & NLM_F_CREATE &&
1594 		n->nlmsg_flags & NLM_F_EXCL);
1595 }
1596 
1597 static bool req_change(struct nlmsghdr *n)
1598 {
1599 	return (!(n->nlmsg_flags & NLM_F_CREATE) &&
1600 		!(n->nlmsg_flags & NLM_F_REPLACE) &&
1601 		!(n->nlmsg_flags & NLM_F_EXCL));
1602 }
1603 
1604 /*
1605  * Create/change qdisc.
1606  */
1607 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1608 			   struct netlink_ext_ack *extack)
1609 {
1610 	struct net *net = sock_net(skb->sk);
1611 	struct tcmsg *tcm;
1612 	struct nlattr *tca[TCA_MAX + 1];
1613 	struct net_device *dev;
1614 	u32 clid;
1615 	struct Qdisc *q, *p;
1616 	int err;
1617 
1618 replay:
1619 	/* Reinit, just in case something touches this. */
1620 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1621 				     rtm_tca_policy, extack);
1622 	if (err < 0)
1623 		return err;
1624 
1625 	tcm = nlmsg_data(n);
1626 	clid = tcm->tcm_parent;
1627 	q = p = NULL;
1628 
1629 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1630 	if (!dev)
1631 		return -ENODEV;
1632 
1633 
1634 	if (clid) {
1635 		if (clid != TC_H_ROOT) {
1636 			if (clid != TC_H_INGRESS) {
1637 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1638 				if (!p) {
1639 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1640 					return -ENOENT;
1641 				}
1642 				q = qdisc_leaf(p, clid);
1643 			} else if (dev_ingress_queue_create(dev)) {
1644 				q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1645 			}
1646 		} else {
1647 			q = rtnl_dereference(dev->qdisc);
1648 		}
1649 
1650 		/* It may be default qdisc, ignore it */
1651 		if (q && q->handle == 0)
1652 			q = NULL;
1653 
1654 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1655 			if (tcm->tcm_handle) {
1656 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1657 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1658 					return -EEXIST;
1659 				}
1660 				if (TC_H_MIN(tcm->tcm_handle)) {
1661 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1662 					return -EINVAL;
1663 				}
1664 				q = qdisc_lookup(dev, tcm->tcm_handle);
1665 				if (!q)
1666 					goto create_n_graft;
1667 				if (q->parent != tcm->tcm_parent) {
1668 					NL_SET_ERR_MSG(extack, "Cannot move an existing qdisc to a different parent");
1669 					return -EINVAL;
1670 				}
1671 				if (n->nlmsg_flags & NLM_F_EXCL) {
1672 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1673 					return -EEXIST;
1674 				}
1675 				if (tca[TCA_KIND] &&
1676 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1677 					NL_SET_ERR_MSG(extack, "Invalid qdisc name: must match existing qdisc");
1678 					return -EINVAL;
1679 				}
1680 				if (q->flags & TCQ_F_INGRESS) {
1681 					NL_SET_ERR_MSG(extack,
1682 						       "Cannot regraft ingress or clsact Qdiscs");
1683 					return -EINVAL;
1684 				}
1685 				if (q == p ||
1686 				    (p && check_loop(q, p, 0))) {
1687 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1688 					return -ELOOP;
1689 				}
1690 				if (clid == TC_H_INGRESS) {
1691 					NL_SET_ERR_MSG(extack, "Ingress cannot graft directly");
1692 					return -EINVAL;
1693 				}
1694 				qdisc_refcount_inc(q);
1695 				goto graft;
1696 			} else {
1697 				if (!q)
1698 					goto create_n_graft;
1699 
1700 				/* This magic test requires explanation.
1701 				 *
1702 				 *   We know, that some child q is already
1703 				 *   attached to this parent and have choice:
1704 				 *   1) change it or 2) create/graft new one.
1705 				 *   If the requested qdisc kind is different
1706 				 *   than the existing one, then we choose graft.
1707 				 *   If they are the same then this is "change"
1708 				 *   operation - just let it fallthrough..
1709 				 *
1710 				 *   1. We are allowed to create/graft only
1711 				 *   if the request is explicitly stating
1712 				 *   "please create if it doesn't exist".
1713 				 *
1714 				 *   2. If the request is to exclusive create
1715 				 *   then the qdisc tcm_handle is not expected
1716 				 *   to exist, so that we choose create/graft too.
1717 				 *
1718 				 *   3. The last case is when no flags are set.
1719 				 *   This will happen when for example tc
1720 				 *   utility issues a "change" command.
1721 				 *   Alas, it is sort of hole in API, we
1722 				 *   cannot decide what to do unambiguously.
1723 				 *   For now we select create/graft.
1724 				 */
1725 				if (tca[TCA_KIND] &&
1726 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1727 					if (req_create_or_replace(n) ||
1728 					    req_create_exclusive(n))
1729 						goto create_n_graft;
1730 					else if (req_change(n))
1731 						goto create_n_graft2;
1732 				}
1733 			}
1734 		}
1735 	} else {
1736 		if (!tcm->tcm_handle) {
1737 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1738 			return -EINVAL;
1739 		}
1740 		q = qdisc_lookup(dev, tcm->tcm_handle);
1741 	}
1742 
1743 	/* Change qdisc parameters */
1744 	if (!q) {
1745 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1746 		return -ENOENT;
1747 	}
1748 	if (n->nlmsg_flags & NLM_F_EXCL) {
1749 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1750 		return -EEXIST;
1751 	}
1752 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1753 		NL_SET_ERR_MSG(extack, "Invalid qdisc name: must match existing qdisc");
1754 		return -EINVAL;
1755 	}
1756 	err = qdisc_change(q, tca, extack);
1757 	if (err == 0)
1758 		qdisc_notify(net, skb, n, clid, NULL, q, extack);
1759 	return err;
1760 
1761 create_n_graft:
1762 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1763 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1764 		return -ENOENT;
1765 	}
1766 create_n_graft2:
1767 	if (clid == TC_H_INGRESS) {
1768 		if (dev_ingress_queue(dev)) {
1769 			q = qdisc_create(dev, dev_ingress_queue(dev),
1770 					 tcm->tcm_parent, tcm->tcm_parent,
1771 					 tca, &err, extack);
1772 		} else {
1773 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1774 			err = -ENOENT;
1775 		}
1776 	} else {
1777 		struct netdev_queue *dev_queue;
1778 
1779 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1780 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1781 		else if (p)
1782 			dev_queue = p->dev_queue;
1783 		else
1784 			dev_queue = netdev_get_tx_queue(dev, 0);
1785 
1786 		q = qdisc_create(dev, dev_queue,
1787 				 tcm->tcm_parent, tcm->tcm_handle,
1788 				 tca, &err, extack);
1789 	}
1790 	if (q == NULL) {
1791 		if (err == -EAGAIN)
1792 			goto replay;
1793 		return err;
1794 	}
1795 
1796 graft:
1797 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1798 	if (err) {
1799 		if (q)
1800 			qdisc_put(q);
1801 		return err;
1802 	}
1803 
1804 	return 0;
1805 }
1806 
1807 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1808 			      struct netlink_callback *cb,
1809 			      int *q_idx_p, int s_q_idx, bool recur,
1810 			      bool dump_invisible)
1811 {
1812 	int ret = 0, q_idx = *q_idx_p;
1813 	struct Qdisc *q;
1814 	int b;
1815 
1816 	if (!root)
1817 		return 0;
1818 
1819 	q = root;
1820 	if (q_idx < s_q_idx) {
1821 		q_idx++;
1822 	} else {
1823 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1824 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1825 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1826 				  RTM_NEWQDISC, NULL) <= 0)
1827 			goto done;
1828 		q_idx++;
1829 	}
1830 
1831 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1832 	 * itself has already been dumped.
1833 	 *
1834 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1835 	 * qdisc hashtable, we don't want to hit it again
1836 	 */
1837 	if (!qdisc_dev(root) || !recur)
1838 		goto out;
1839 
1840 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1841 		if (q_idx < s_q_idx) {
1842 			q_idx++;
1843 			continue;
1844 		}
1845 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1846 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1847 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1848 				  RTM_NEWQDISC, NULL) <= 0)
1849 			goto done;
1850 		q_idx++;
1851 	}
1852 
1853 out:
1854 	*q_idx_p = q_idx;
1855 	return ret;
1856 done:
1857 	ret = -1;
1858 	goto out;
1859 }
1860 
1861 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1862 {
1863 	struct net *net = sock_net(skb->sk);
1864 	int idx, q_idx;
1865 	int s_idx, s_q_idx;
1866 	struct net_device *dev;
1867 	const struct nlmsghdr *nlh = cb->nlh;
1868 	struct nlattr *tca[TCA_MAX + 1];
1869 	int err;
1870 
1871 	s_idx = cb->args[0];
1872 	s_q_idx = q_idx = cb->args[1];
1873 
1874 	idx = 0;
1875 	ASSERT_RTNL();
1876 
1877 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1878 				     rtm_tca_policy, cb->extack);
1879 	if (err < 0)
1880 		return err;
1881 
1882 	for_each_netdev(net, dev) {
1883 		struct netdev_queue *dev_queue;
1884 
1885 		if (idx < s_idx)
1886 			goto cont;
1887 		if (idx > s_idx)
1888 			s_q_idx = 0;
1889 		q_idx = 0;
1890 
1891 		if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1892 				       skb, cb, &q_idx, s_q_idx,
1893 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1894 			goto done;
1895 
1896 		dev_queue = dev_ingress_queue(dev);
1897 		if (dev_queue &&
1898 		    tc_dump_qdisc_root(rtnl_dereference(dev_queue->qdisc_sleeping),
1899 				       skb, cb, &q_idx, s_q_idx, false,
1900 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1901 			goto done;
1902 
1903 cont:
1904 		idx++;
1905 	}
1906 
1907 done:
1908 	cb->args[0] = idx;
1909 	cb->args[1] = q_idx;
1910 
1911 	return skb->len;
1912 }
1913 
1914 
1915 
1916 /************************************************
1917  *	Traffic classes manipulation.		*
1918  ************************************************/
1919 
1920 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1921 			  unsigned long cl, u32 portid, u32 seq, u16 flags,
1922 			  int event, struct netlink_ext_ack *extack)
1923 {
1924 	struct tcmsg *tcm;
1925 	struct nlmsghdr  *nlh;
1926 	unsigned char *b = skb_tail_pointer(skb);
1927 	struct gnet_dump d;
1928 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1929 
1930 	cond_resched();
1931 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1932 	if (!nlh)
1933 		goto out_nlmsg_trim;
1934 	tcm = nlmsg_data(nlh);
1935 	tcm->tcm_family = AF_UNSPEC;
1936 	tcm->tcm__pad1 = 0;
1937 	tcm->tcm__pad2 = 0;
1938 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1939 	tcm->tcm_parent = q->handle;
1940 	tcm->tcm_handle = q->handle;
1941 	tcm->tcm_info = 0;
1942 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1943 		goto nla_put_failure;
1944 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1945 		goto nla_put_failure;
1946 
1947 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1948 					 NULL, &d, TCA_PAD) < 0)
1949 		goto nla_put_failure;
1950 
1951 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1952 		goto nla_put_failure;
1953 
1954 	if (gnet_stats_finish_copy(&d) < 0)
1955 		goto nla_put_failure;
1956 
1957 	if (extack && extack->_msg &&
1958 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
1959 		goto out_nlmsg_trim;
1960 
1961 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1962 
1963 	return skb->len;
1964 
1965 out_nlmsg_trim:
1966 nla_put_failure:
1967 	nlmsg_trim(skb, b);
1968 	return -1;
1969 }
1970 
1971 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1972 			 struct nlmsghdr *n, struct Qdisc *q,
1973 			 unsigned long cl, int event, struct netlink_ext_ack *extack)
1974 {
1975 	struct sk_buff *skb;
1976 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1977 
1978 	if (!rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC))
1979 		return 0;
1980 
1981 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1982 	if (!skb)
1983 		return -ENOBUFS;
1984 
1985 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event, extack) < 0) {
1986 		kfree_skb(skb);
1987 		return -EINVAL;
1988 	}
1989 
1990 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1991 			      n->nlmsg_flags & NLM_F_ECHO);
1992 }
1993 
1994 static int tclass_get_notify(struct net *net, struct sk_buff *oskb,
1995 			     struct nlmsghdr *n, struct Qdisc *q,
1996 			     unsigned long cl, struct netlink_ext_ack *extack)
1997 {
1998 	struct sk_buff *skb;
1999 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2000 
2001 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2002 	if (!skb)
2003 		return -ENOBUFS;
2004 
2005 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, RTM_NEWTCLASS,
2006 			   extack) < 0) {
2007 		kfree_skb(skb);
2008 		return -EINVAL;
2009 	}
2010 
2011 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2012 			      n->nlmsg_flags & NLM_F_ECHO);
2013 }
2014 
2015 static int tclass_del_notify(struct net *net,
2016 			     const struct Qdisc_class_ops *cops,
2017 			     struct sk_buff *oskb, struct nlmsghdr *n,
2018 			     struct Qdisc *q, unsigned long cl,
2019 			     struct netlink_ext_ack *extack)
2020 {
2021 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2022 	struct sk_buff *skb;
2023 	int err = 0;
2024 
2025 	if (!cops->delete)
2026 		return -EOPNOTSUPP;
2027 
2028 	if (rtnl_notify_needed(net, n->nlmsg_flags, RTNLGRP_TC)) {
2029 		skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2030 		if (!skb)
2031 			return -ENOBUFS;
2032 
2033 		if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
2034 				   RTM_DELTCLASS, extack) < 0) {
2035 			kfree_skb(skb);
2036 			return -EINVAL;
2037 		}
2038 	} else {
2039 		skb = NULL;
2040 	}
2041 
2042 	err = cops->delete(q, cl, extack);
2043 	if (err) {
2044 		kfree_skb(skb);
2045 		return err;
2046 	}
2047 
2048 	err = rtnetlink_maybe_send(skb, net, portid, RTNLGRP_TC,
2049 				   n->nlmsg_flags & NLM_F_ECHO);
2050 	return err;
2051 }
2052 
2053 #ifdef CONFIG_NET_CLS
2054 
2055 struct tcf_bind_args {
2056 	struct tcf_walker w;
2057 	unsigned long base;
2058 	unsigned long cl;
2059 	u32 classid;
2060 };
2061 
2062 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2063 {
2064 	struct tcf_bind_args *a = (void *)arg;
2065 
2066 	if (n && tp->ops->bind_class) {
2067 		struct Qdisc *q = tcf_block_q(tp->chain->block);
2068 
2069 		sch_tree_lock(q);
2070 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
2071 		sch_tree_unlock(q);
2072 	}
2073 	return 0;
2074 }
2075 
2076 struct tc_bind_class_args {
2077 	struct qdisc_walker w;
2078 	unsigned long new_cl;
2079 	u32 portid;
2080 	u32 clid;
2081 };
2082 
2083 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
2084 				struct qdisc_walker *w)
2085 {
2086 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
2087 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
2088 	struct tcf_block *block;
2089 	struct tcf_chain *chain;
2090 
2091 	block = cops->tcf_block(q, cl, NULL);
2092 	if (!block)
2093 		return 0;
2094 	for (chain = tcf_get_next_chain(block, NULL);
2095 	     chain;
2096 	     chain = tcf_get_next_chain(block, chain)) {
2097 		struct tcf_proto *tp;
2098 
2099 		for (tp = tcf_get_next_proto(chain, NULL);
2100 		     tp; tp = tcf_get_next_proto(chain, tp)) {
2101 			struct tcf_bind_args arg = {};
2102 
2103 			arg.w.fn = tcf_node_bind;
2104 			arg.classid = a->clid;
2105 			arg.base = cl;
2106 			arg.cl = a->new_cl;
2107 			tp->ops->walk(tp, &arg.w, true);
2108 		}
2109 	}
2110 
2111 	return 0;
2112 }
2113 
2114 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2115 			   unsigned long new_cl)
2116 {
2117 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
2118 	struct tc_bind_class_args args = {};
2119 
2120 	if (!cops->tcf_block)
2121 		return;
2122 	args.portid = portid;
2123 	args.clid = clid;
2124 	args.new_cl = new_cl;
2125 	args.w.fn = tc_bind_class_walker;
2126 	q->ops->cl_ops->walk(q, &args.w);
2127 }
2128 
2129 #else
2130 
2131 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2132 			   unsigned long new_cl)
2133 {
2134 }
2135 
2136 #endif
2137 
2138 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
2139 			 struct netlink_ext_ack *extack)
2140 {
2141 	struct net *net = sock_net(skb->sk);
2142 	struct tcmsg *tcm = nlmsg_data(n);
2143 	struct nlattr *tca[TCA_MAX + 1];
2144 	struct net_device *dev;
2145 	struct Qdisc *q = NULL;
2146 	const struct Qdisc_class_ops *cops;
2147 	unsigned long cl = 0;
2148 	unsigned long new_cl;
2149 	u32 portid;
2150 	u32 clid;
2151 	u32 qid;
2152 	int err;
2153 
2154 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2155 				     rtm_tca_policy, extack);
2156 	if (err < 0)
2157 		return err;
2158 
2159 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2160 	if (!dev)
2161 		return -ENODEV;
2162 
2163 	/*
2164 	   parent == TC_H_UNSPEC - unspecified parent.
2165 	   parent == TC_H_ROOT   - class is root, which has no parent.
2166 	   parent == X:0	 - parent is root class.
2167 	   parent == X:Y	 - parent is a node in hierarchy.
2168 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2169 
2170 	   handle == 0:0	 - generate handle from kernel pool.
2171 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2172 	   handle == X:Y	 - clear.
2173 	   handle == X:0	 - root class.
2174 	 */
2175 
2176 	/* Step 1. Determine qdisc handle X:0 */
2177 
2178 	portid = tcm->tcm_parent;
2179 	clid = tcm->tcm_handle;
2180 	qid = TC_H_MAJ(clid);
2181 
2182 	if (portid != TC_H_ROOT) {
2183 		u32 qid1 = TC_H_MAJ(portid);
2184 
2185 		if (qid && qid1) {
2186 			/* If both majors are known, they must be identical. */
2187 			if (qid != qid1)
2188 				return -EINVAL;
2189 		} else if (qid1) {
2190 			qid = qid1;
2191 		} else if (qid == 0)
2192 			qid = rtnl_dereference(dev->qdisc)->handle;
2193 
2194 		/* Now qid is genuine qdisc handle consistent
2195 		 * both with parent and child.
2196 		 *
2197 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2198 		 */
2199 		if (portid)
2200 			portid = TC_H_MAKE(qid, portid);
2201 	} else {
2202 		if (qid == 0)
2203 			qid = rtnl_dereference(dev->qdisc)->handle;
2204 	}
2205 
2206 	/* OK. Locate qdisc */
2207 	q = qdisc_lookup(dev, qid);
2208 	if (!q)
2209 		return -ENOENT;
2210 
2211 	/* An check that it supports classes */
2212 	cops = q->ops->cl_ops;
2213 	if (cops == NULL)
2214 		return -EINVAL;
2215 
2216 	/* Now try to get class */
2217 	if (clid == 0) {
2218 		if (portid == TC_H_ROOT)
2219 			clid = qid;
2220 	} else
2221 		clid = TC_H_MAKE(qid, clid);
2222 
2223 	if (clid)
2224 		cl = cops->find(q, clid);
2225 
2226 	if (cl == 0) {
2227 		err = -ENOENT;
2228 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2229 		    !(n->nlmsg_flags & NLM_F_CREATE))
2230 			goto out;
2231 	} else {
2232 		switch (n->nlmsg_type) {
2233 		case RTM_NEWTCLASS:
2234 			err = -EEXIST;
2235 			if (n->nlmsg_flags & NLM_F_EXCL)
2236 				goto out;
2237 			break;
2238 		case RTM_DELTCLASS:
2239 			err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2240 			/* Unbind the class with flilters with 0 */
2241 			tc_bind_tclass(q, portid, clid, 0);
2242 			goto out;
2243 		case RTM_GETTCLASS:
2244 			err = tclass_get_notify(net, skb, n, q, cl, extack);
2245 			goto out;
2246 		default:
2247 			err = -EINVAL;
2248 			goto out;
2249 		}
2250 	}
2251 
2252 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2253 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2254 		return -EOPNOTSUPP;
2255 	}
2256 
2257 	new_cl = cl;
2258 	err = -EOPNOTSUPP;
2259 	if (cops->change)
2260 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2261 	if (err == 0) {
2262 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS, extack);
2263 		/* We just create a new class, need to do reverse binding. */
2264 		if (cl != new_cl)
2265 			tc_bind_tclass(q, portid, clid, new_cl);
2266 	}
2267 out:
2268 	return err;
2269 }
2270 
2271 struct qdisc_dump_args {
2272 	struct qdisc_walker	w;
2273 	struct sk_buff		*skb;
2274 	struct netlink_callback	*cb;
2275 };
2276 
2277 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2278 			    struct qdisc_walker *arg)
2279 {
2280 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2281 
2282 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2283 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2284 			      RTM_NEWTCLASS, NULL);
2285 }
2286 
2287 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2288 				struct tcmsg *tcm, struct netlink_callback *cb,
2289 				int *t_p, int s_t)
2290 {
2291 	struct qdisc_dump_args arg;
2292 
2293 	if (tc_qdisc_dump_ignore(q, false) ||
2294 	    *t_p < s_t || !q->ops->cl_ops ||
2295 	    (tcm->tcm_parent &&
2296 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2297 		(*t_p)++;
2298 		return 0;
2299 	}
2300 	if (*t_p > s_t)
2301 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2302 	arg.w.fn = qdisc_class_dump;
2303 	arg.skb = skb;
2304 	arg.cb = cb;
2305 	arg.w.stop  = 0;
2306 	arg.w.skip = cb->args[1];
2307 	arg.w.count = 0;
2308 	q->ops->cl_ops->walk(q, &arg.w);
2309 	cb->args[1] = arg.w.count;
2310 	if (arg.w.stop)
2311 		return -1;
2312 	(*t_p)++;
2313 	return 0;
2314 }
2315 
2316 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2317 			       struct tcmsg *tcm, struct netlink_callback *cb,
2318 			       int *t_p, int s_t, bool recur)
2319 {
2320 	struct Qdisc *q;
2321 	int b;
2322 
2323 	if (!root)
2324 		return 0;
2325 
2326 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2327 		return -1;
2328 
2329 	if (!qdisc_dev(root) || !recur)
2330 		return 0;
2331 
2332 	if (tcm->tcm_parent) {
2333 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2334 		if (q && q != root &&
2335 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2336 			return -1;
2337 		return 0;
2338 	}
2339 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2340 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2341 			return -1;
2342 	}
2343 
2344 	return 0;
2345 }
2346 
2347 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2348 {
2349 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2350 	struct net *net = sock_net(skb->sk);
2351 	struct netdev_queue *dev_queue;
2352 	struct net_device *dev;
2353 	int t, s_t;
2354 
2355 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2356 		return 0;
2357 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2358 	if (!dev)
2359 		return 0;
2360 
2361 	s_t = cb->args[0];
2362 	t = 0;
2363 
2364 	if (tc_dump_tclass_root(rtnl_dereference(dev->qdisc),
2365 				skb, tcm, cb, &t, s_t, true) < 0)
2366 		goto done;
2367 
2368 	dev_queue = dev_ingress_queue(dev);
2369 	if (dev_queue &&
2370 	    tc_dump_tclass_root(rtnl_dereference(dev_queue->qdisc_sleeping),
2371 				skb, tcm, cb, &t, s_t, false) < 0)
2372 		goto done;
2373 
2374 done:
2375 	cb->args[0] = t;
2376 
2377 	dev_put(dev);
2378 	return skb->len;
2379 }
2380 
2381 #ifdef CONFIG_PROC_FS
2382 static int psched_show(struct seq_file *seq, void *v)
2383 {
2384 	seq_printf(seq, "%08x %08x %08x %08x\n",
2385 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2386 		   1000000,
2387 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2388 
2389 	return 0;
2390 }
2391 
2392 static int __net_init psched_net_init(struct net *net)
2393 {
2394 	struct proc_dir_entry *e;
2395 
2396 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2397 	if (e == NULL)
2398 		return -ENOMEM;
2399 
2400 	return 0;
2401 }
2402 
2403 static void __net_exit psched_net_exit(struct net *net)
2404 {
2405 	remove_proc_entry("psched", net->proc_net);
2406 }
2407 #else
2408 static int __net_init psched_net_init(struct net *net)
2409 {
2410 	return 0;
2411 }
2412 
2413 static void __net_exit psched_net_exit(struct net *net)
2414 {
2415 }
2416 #endif
2417 
2418 static struct pernet_operations psched_net_ops = {
2419 	.init = psched_net_init,
2420 	.exit = psched_net_exit,
2421 };
2422 
2423 #if IS_ENABLED(CONFIG_MITIGATION_RETPOLINE)
2424 DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper);
2425 #endif
2426 
2427 static const struct rtnl_msg_handler psched_rtnl_msg_handlers[] __initconst = {
2428 	{.msgtype = RTM_NEWQDISC, .doit = tc_modify_qdisc},
2429 	{.msgtype = RTM_DELQDISC, .doit = tc_get_qdisc},
2430 	{.msgtype = RTM_GETQDISC, .doit = tc_get_qdisc,
2431 	 .dumpit = tc_dump_qdisc},
2432 	{.msgtype = RTM_NEWTCLASS, .doit = tc_ctl_tclass},
2433 	{.msgtype = RTM_DELTCLASS, .doit = tc_ctl_tclass},
2434 	{.msgtype = RTM_GETTCLASS, .doit = tc_ctl_tclass,
2435 	 .dumpit = tc_dump_tclass},
2436 };
2437 
2438 static int __init pktsched_init(void)
2439 {
2440 	int err;
2441 
2442 	err = register_pernet_subsys(&psched_net_ops);
2443 	if (err) {
2444 		pr_err("pktsched_init: "
2445 		       "cannot initialize per netns operations\n");
2446 		return err;
2447 	}
2448 
2449 	register_qdisc(&pfifo_fast_ops);
2450 	register_qdisc(&pfifo_qdisc_ops);
2451 	register_qdisc(&bfifo_qdisc_ops);
2452 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2453 	register_qdisc(&mq_qdisc_ops);
2454 	register_qdisc(&noqueue_qdisc_ops);
2455 
2456 	rtnl_register_many(psched_rtnl_msg_handlers);
2457 
2458 	tc_wrapper_init();
2459 
2460 	return 0;
2461 }
2462 
2463 subsys_initcall(pktsched_init);
2464