xref: /linux/net/sched/sch_api.c (revision 993498e537af9260e697219ce41b41b22b6199cc)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/sch_api.c	Packet scheduler API.
4  *
5  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6  *
7  * Fixes:
8  *
9  * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired.
10  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
11  * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support
12  */
13 
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <linux/init.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/kmod.h>
24 #include <linux/list.h>
25 #include <linux/hrtimer.h>
26 #include <linux/slab.h>
27 #include <linux/hashtable.h>
28 
29 #include <net/net_namespace.h>
30 #include <net/sock.h>
31 #include <net/netlink.h>
32 #include <net/pkt_sched.h>
33 #include <net/pkt_cls.h>
34 #include <net/tc_wrapper.h>
35 
36 #include <trace/events/qdisc.h>
37 
38 /*
39 
40    Short review.
41    -------------
42 
43    This file consists of two interrelated parts:
44 
45    1. queueing disciplines manager frontend.
46    2. traffic classes manager frontend.
47 
48    Generally, queueing discipline ("qdisc") is a black box,
49    which is able to enqueue packets and to dequeue them (when
50    device is ready to send something) in order and at times
51    determined by algorithm hidden in it.
52 
53    qdisc's are divided to two categories:
54    - "queues", which have no internal structure visible from outside.
55    - "schedulers", which split all the packets to "traffic classes",
56      using "packet classifiers" (look at cls_api.c)
57 
58    In turn, classes may have child qdiscs (as rule, queues)
59    attached to them etc. etc. etc.
60 
61    The goal of the routines in this file is to translate
62    information supplied by user in the form of handles
63    to more intelligible for kernel form, to make some sanity
64    checks and part of work, which is common to all qdiscs
65    and to provide rtnetlink notifications.
66 
67    All real intelligent work is done inside qdisc modules.
68 
69 
70 
71    Every discipline has two major routines: enqueue and dequeue.
72 
73    ---dequeue
74 
75    dequeue usually returns a skb to send. It is allowed to return NULL,
76    but it does not mean that queue is empty, it just means that
77    discipline does not want to send anything this time.
78    Queue is really empty if q->q.qlen == 0.
79    For complicated disciplines with multiple queues q->q is not
80    real packet queue, but however q->q.qlen must be valid.
81 
82    ---enqueue
83 
84    enqueue returns 0, if packet was enqueued successfully.
85    If packet (this one or another one) was dropped, it returns
86    not zero error code.
87    NET_XMIT_DROP 	- this packet dropped
88      Expected action: do not backoff, but wait until queue will clear.
89    NET_XMIT_CN	 	- probably this packet enqueued, but another one dropped.
90      Expected action: backoff or ignore
91 
92    Auxiliary routines:
93 
94    ---peek
95 
96    like dequeue but without removing a packet from the queue
97 
98    ---reset
99 
100    returns qdisc to initial state: purge all buffers, clear all
101    timers, counters (except for statistics) etc.
102 
103    ---init
104 
105    initializes newly created qdisc.
106 
107    ---destroy
108 
109    destroys resources allocated by init and during lifetime of qdisc.
110 
111    ---change
112 
113    changes qdisc parameters.
114  */
115 
116 /* Protects list of registered TC modules. It is pure SMP lock. */
117 static DEFINE_RWLOCK(qdisc_mod_lock);
118 
119 
120 /************************************************
121  *	Queueing disciplines manipulation.	*
122  ************************************************/
123 
124 
125 /* The list of all installed queueing disciplines. */
126 
127 static struct Qdisc_ops *qdisc_base;
128 
129 /* Register/unregister queueing discipline */
130 
131 int register_qdisc(struct Qdisc_ops *qops)
132 {
133 	struct Qdisc_ops *q, **qp;
134 	int rc = -EEXIST;
135 
136 	write_lock(&qdisc_mod_lock);
137 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
138 		if (!strcmp(qops->id, q->id))
139 			goto out;
140 
141 	if (qops->enqueue == NULL)
142 		qops->enqueue = noop_qdisc_ops.enqueue;
143 	if (qops->peek == NULL) {
144 		if (qops->dequeue == NULL)
145 			qops->peek = noop_qdisc_ops.peek;
146 		else
147 			goto out_einval;
148 	}
149 	if (qops->dequeue == NULL)
150 		qops->dequeue = noop_qdisc_ops.dequeue;
151 
152 	if (qops->cl_ops) {
153 		const struct Qdisc_class_ops *cops = qops->cl_ops;
154 
155 		if (!(cops->find && cops->walk && cops->leaf))
156 			goto out_einval;
157 
158 		if (cops->tcf_block && !(cops->bind_tcf && cops->unbind_tcf))
159 			goto out_einval;
160 	}
161 
162 	qops->next = NULL;
163 	*qp = qops;
164 	rc = 0;
165 out:
166 	write_unlock(&qdisc_mod_lock);
167 	return rc;
168 
169 out_einval:
170 	rc = -EINVAL;
171 	goto out;
172 }
173 EXPORT_SYMBOL(register_qdisc);
174 
175 void unregister_qdisc(struct Qdisc_ops *qops)
176 {
177 	struct Qdisc_ops *q, **qp;
178 	int err = -ENOENT;
179 
180 	write_lock(&qdisc_mod_lock);
181 	for (qp = &qdisc_base; (q = *qp) != NULL; qp = &q->next)
182 		if (q == qops)
183 			break;
184 	if (q) {
185 		*qp = q->next;
186 		q->next = NULL;
187 		err = 0;
188 	}
189 	write_unlock(&qdisc_mod_lock);
190 
191 	WARN(err, "unregister qdisc(%s) failed\n", qops->id);
192 }
193 EXPORT_SYMBOL(unregister_qdisc);
194 
195 /* Get default qdisc if not otherwise specified */
196 void qdisc_get_default(char *name, size_t len)
197 {
198 	read_lock(&qdisc_mod_lock);
199 	strscpy(name, default_qdisc_ops->id, len);
200 	read_unlock(&qdisc_mod_lock);
201 }
202 
203 static struct Qdisc_ops *qdisc_lookup_default(const char *name)
204 {
205 	struct Qdisc_ops *q = NULL;
206 
207 	for (q = qdisc_base; q; q = q->next) {
208 		if (!strcmp(name, q->id)) {
209 			if (!try_module_get(q->owner))
210 				q = NULL;
211 			break;
212 		}
213 	}
214 
215 	return q;
216 }
217 
218 /* Set new default qdisc to use */
219 int qdisc_set_default(const char *name)
220 {
221 	const struct Qdisc_ops *ops;
222 
223 	if (!capable(CAP_NET_ADMIN))
224 		return -EPERM;
225 
226 	write_lock(&qdisc_mod_lock);
227 	ops = qdisc_lookup_default(name);
228 	if (!ops) {
229 		/* Not found, drop lock and try to load module */
230 		write_unlock(&qdisc_mod_lock);
231 		request_module("sch_%s", name);
232 		write_lock(&qdisc_mod_lock);
233 
234 		ops = qdisc_lookup_default(name);
235 	}
236 
237 	if (ops) {
238 		/* Set new default */
239 		module_put(default_qdisc_ops->owner);
240 		default_qdisc_ops = ops;
241 	}
242 	write_unlock(&qdisc_mod_lock);
243 
244 	return ops ? 0 : -ENOENT;
245 }
246 
247 #ifdef CONFIG_NET_SCH_DEFAULT
248 /* Set default value from kernel config */
249 static int __init sch_default_qdisc(void)
250 {
251 	return qdisc_set_default(CONFIG_DEFAULT_NET_SCH);
252 }
253 late_initcall(sch_default_qdisc);
254 #endif
255 
256 /* We know handle. Find qdisc among all qdisc's attached to device
257  * (root qdisc, all its children, children of children etc.)
258  * Note: caller either uses rtnl or rcu_read_lock()
259  */
260 
261 static struct Qdisc *qdisc_match_from_root(struct Qdisc *root, u32 handle)
262 {
263 	struct Qdisc *q;
264 
265 	if (!qdisc_dev(root))
266 		return (root->handle == handle ? root : NULL);
267 
268 	if (!(root->flags & TCQ_F_BUILTIN) &&
269 	    root->handle == handle)
270 		return root;
271 
272 	hash_for_each_possible_rcu(qdisc_dev(root)->qdisc_hash, q, hash, handle,
273 				   lockdep_rtnl_is_held()) {
274 		if (q->handle == handle)
275 			return q;
276 	}
277 	return NULL;
278 }
279 
280 void qdisc_hash_add(struct Qdisc *q, bool invisible)
281 {
282 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
283 		ASSERT_RTNL();
284 		hash_add_rcu(qdisc_dev(q)->qdisc_hash, &q->hash, q->handle);
285 		if (invisible)
286 			q->flags |= TCQ_F_INVISIBLE;
287 	}
288 }
289 EXPORT_SYMBOL(qdisc_hash_add);
290 
291 void qdisc_hash_del(struct Qdisc *q)
292 {
293 	if ((q->parent != TC_H_ROOT) && !(q->flags & TCQ_F_INGRESS)) {
294 		ASSERT_RTNL();
295 		hash_del_rcu(&q->hash);
296 	}
297 }
298 EXPORT_SYMBOL(qdisc_hash_del);
299 
300 struct Qdisc *qdisc_lookup(struct net_device *dev, u32 handle)
301 {
302 	struct Qdisc *q;
303 
304 	if (!handle)
305 		return NULL;
306 	q = qdisc_match_from_root(rtnl_dereference(dev->qdisc), handle);
307 	if (q)
308 		goto out;
309 
310 	if (dev_ingress_queue(dev))
311 		q = qdisc_match_from_root(
312 			rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping),
313 			handle);
314 out:
315 	return q;
316 }
317 
318 struct Qdisc *qdisc_lookup_rcu(struct net_device *dev, u32 handle)
319 {
320 	struct netdev_queue *nq;
321 	struct Qdisc *q;
322 
323 	if (!handle)
324 		return NULL;
325 	q = qdisc_match_from_root(rcu_dereference(dev->qdisc), handle);
326 	if (q)
327 		goto out;
328 
329 	nq = dev_ingress_queue_rcu(dev);
330 	if (nq)
331 		q = qdisc_match_from_root(rcu_dereference(nq->qdisc_sleeping),
332 					  handle);
333 out:
334 	return q;
335 }
336 
337 static struct Qdisc *qdisc_leaf(struct Qdisc *p, u32 classid)
338 {
339 	unsigned long cl;
340 	const struct Qdisc_class_ops *cops = p->ops->cl_ops;
341 
342 	if (cops == NULL)
343 		return NULL;
344 	cl = cops->find(p, classid);
345 
346 	if (cl == 0)
347 		return NULL;
348 	return cops->leaf(p, cl);
349 }
350 
351 /* Find queueing discipline by name */
352 
353 static struct Qdisc_ops *qdisc_lookup_ops(struct nlattr *kind)
354 {
355 	struct Qdisc_ops *q = NULL;
356 
357 	if (kind) {
358 		read_lock(&qdisc_mod_lock);
359 		for (q = qdisc_base; q; q = q->next) {
360 			if (nla_strcmp(kind, q->id) == 0) {
361 				if (!try_module_get(q->owner))
362 					q = NULL;
363 				break;
364 			}
365 		}
366 		read_unlock(&qdisc_mod_lock);
367 	}
368 	return q;
369 }
370 
371 /* The linklayer setting were not transferred from iproute2, in older
372  * versions, and the rate tables lookup systems have been dropped in
373  * the kernel. To keep backward compatible with older iproute2 tc
374  * utils, we detect the linklayer setting by detecting if the rate
375  * table were modified.
376  *
377  * For linklayer ATM table entries, the rate table will be aligned to
378  * 48 bytes, thus some table entries will contain the same value.  The
379  * mpu (min packet unit) is also encoded into the old rate table, thus
380  * starting from the mpu, we find low and high table entries for
381  * mapping this cell.  If these entries contain the same value, when
382  * the rate tables have been modified for linklayer ATM.
383  *
384  * This is done by rounding mpu to the nearest 48 bytes cell/entry,
385  * and then roundup to the next cell, calc the table entry one below,
386  * and compare.
387  */
388 static __u8 __detect_linklayer(struct tc_ratespec *r, __u32 *rtab)
389 {
390 	int low       = roundup(r->mpu, 48);
391 	int high      = roundup(low+1, 48);
392 	int cell_low  = low >> r->cell_log;
393 	int cell_high = (high >> r->cell_log) - 1;
394 
395 	/* rtab is too inaccurate at rates > 100Mbit/s */
396 	if ((r->rate > (100000000/8)) || (rtab[0] == 0)) {
397 		pr_debug("TC linklayer: Giving up ATM detection\n");
398 		return TC_LINKLAYER_ETHERNET;
399 	}
400 
401 	if ((cell_high > cell_low) && (cell_high < 256)
402 	    && (rtab[cell_low] == rtab[cell_high])) {
403 		pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
404 			 cell_low, cell_high, rtab[cell_high]);
405 		return TC_LINKLAYER_ATM;
406 	}
407 	return TC_LINKLAYER_ETHERNET;
408 }
409 
410 static struct qdisc_rate_table *qdisc_rtab_list;
411 
412 struct qdisc_rate_table *qdisc_get_rtab(struct tc_ratespec *r,
413 					struct nlattr *tab,
414 					struct netlink_ext_ack *extack)
415 {
416 	struct qdisc_rate_table *rtab;
417 
418 	if (tab == NULL || r->rate == 0 ||
419 	    r->cell_log == 0 || r->cell_log >= 32 ||
420 	    nla_len(tab) != TC_RTAB_SIZE) {
421 		NL_SET_ERR_MSG(extack, "Invalid rate table parameters for searching");
422 		return NULL;
423 	}
424 
425 	for (rtab = qdisc_rtab_list; rtab; rtab = rtab->next) {
426 		if (!memcmp(&rtab->rate, r, sizeof(struct tc_ratespec)) &&
427 		    !memcmp(&rtab->data, nla_data(tab), 1024)) {
428 			rtab->refcnt++;
429 			return rtab;
430 		}
431 	}
432 
433 	rtab = kmalloc(sizeof(*rtab), GFP_KERNEL);
434 	if (rtab) {
435 		rtab->rate = *r;
436 		rtab->refcnt = 1;
437 		memcpy(rtab->data, nla_data(tab), 1024);
438 		if (r->linklayer == TC_LINKLAYER_UNAWARE)
439 			r->linklayer = __detect_linklayer(r, rtab->data);
440 		rtab->next = qdisc_rtab_list;
441 		qdisc_rtab_list = rtab;
442 	} else {
443 		NL_SET_ERR_MSG(extack, "Failed to allocate new qdisc rate table");
444 	}
445 	return rtab;
446 }
447 EXPORT_SYMBOL(qdisc_get_rtab);
448 
449 void qdisc_put_rtab(struct qdisc_rate_table *tab)
450 {
451 	struct qdisc_rate_table *rtab, **rtabp;
452 
453 	if (!tab || --tab->refcnt)
454 		return;
455 
456 	for (rtabp = &qdisc_rtab_list;
457 	     (rtab = *rtabp) != NULL;
458 	     rtabp = &rtab->next) {
459 		if (rtab == tab) {
460 			*rtabp = rtab->next;
461 			kfree(rtab);
462 			return;
463 		}
464 	}
465 }
466 EXPORT_SYMBOL(qdisc_put_rtab);
467 
468 static LIST_HEAD(qdisc_stab_list);
469 
470 static const struct nla_policy stab_policy[TCA_STAB_MAX + 1] = {
471 	[TCA_STAB_BASE]	= { .len = sizeof(struct tc_sizespec) },
472 	[TCA_STAB_DATA] = { .type = NLA_BINARY },
473 };
474 
475 static struct qdisc_size_table *qdisc_get_stab(struct nlattr *opt,
476 					       struct netlink_ext_ack *extack)
477 {
478 	struct nlattr *tb[TCA_STAB_MAX + 1];
479 	struct qdisc_size_table *stab;
480 	struct tc_sizespec *s;
481 	unsigned int tsize = 0;
482 	u16 *tab = NULL;
483 	int err;
484 
485 	err = nla_parse_nested_deprecated(tb, TCA_STAB_MAX, opt, stab_policy,
486 					  extack);
487 	if (err < 0)
488 		return ERR_PTR(err);
489 	if (!tb[TCA_STAB_BASE]) {
490 		NL_SET_ERR_MSG(extack, "Size table base attribute is missing");
491 		return ERR_PTR(-EINVAL);
492 	}
493 
494 	s = nla_data(tb[TCA_STAB_BASE]);
495 
496 	if (s->tsize > 0) {
497 		if (!tb[TCA_STAB_DATA]) {
498 			NL_SET_ERR_MSG(extack, "Size table data attribute is missing");
499 			return ERR_PTR(-EINVAL);
500 		}
501 		tab = nla_data(tb[TCA_STAB_DATA]);
502 		tsize = nla_len(tb[TCA_STAB_DATA]) / sizeof(u16);
503 	}
504 
505 	if (tsize != s->tsize || (!tab && tsize > 0)) {
506 		NL_SET_ERR_MSG(extack, "Invalid size of size table");
507 		return ERR_PTR(-EINVAL);
508 	}
509 
510 	list_for_each_entry(stab, &qdisc_stab_list, list) {
511 		if (memcmp(&stab->szopts, s, sizeof(*s)))
512 			continue;
513 		if (tsize > 0 &&
514 		    memcmp(stab->data, tab, flex_array_size(stab, data, tsize)))
515 			continue;
516 		stab->refcnt++;
517 		return stab;
518 	}
519 
520 	if (s->size_log > STAB_SIZE_LOG_MAX ||
521 	    s->cell_log > STAB_SIZE_LOG_MAX) {
522 		NL_SET_ERR_MSG(extack, "Invalid logarithmic size of size table");
523 		return ERR_PTR(-EINVAL);
524 	}
525 
526 	stab = kmalloc(struct_size(stab, data, tsize), GFP_KERNEL);
527 	if (!stab)
528 		return ERR_PTR(-ENOMEM);
529 
530 	stab->refcnt = 1;
531 	stab->szopts = *s;
532 	if (tsize > 0)
533 		memcpy(stab->data, tab, flex_array_size(stab, data, tsize));
534 
535 	list_add_tail(&stab->list, &qdisc_stab_list);
536 
537 	return stab;
538 }
539 
540 void qdisc_put_stab(struct qdisc_size_table *tab)
541 {
542 	if (!tab)
543 		return;
544 
545 	if (--tab->refcnt == 0) {
546 		list_del(&tab->list);
547 		kfree_rcu(tab, rcu);
548 	}
549 }
550 EXPORT_SYMBOL(qdisc_put_stab);
551 
552 static int qdisc_dump_stab(struct sk_buff *skb, struct qdisc_size_table *stab)
553 {
554 	struct nlattr *nest;
555 
556 	nest = nla_nest_start_noflag(skb, TCA_STAB);
557 	if (nest == NULL)
558 		goto nla_put_failure;
559 	if (nla_put(skb, TCA_STAB_BASE, sizeof(stab->szopts), &stab->szopts))
560 		goto nla_put_failure;
561 	nla_nest_end(skb, nest);
562 
563 	return skb->len;
564 
565 nla_put_failure:
566 	return -1;
567 }
568 
569 void __qdisc_calculate_pkt_len(struct sk_buff *skb,
570 			       const struct qdisc_size_table *stab)
571 {
572 	int pkt_len, slot;
573 
574 	pkt_len = skb->len + stab->szopts.overhead;
575 	if (unlikely(!stab->szopts.tsize))
576 		goto out;
577 
578 	slot = pkt_len + stab->szopts.cell_align;
579 	if (unlikely(slot < 0))
580 		slot = 0;
581 
582 	slot >>= stab->szopts.cell_log;
583 	if (likely(slot < stab->szopts.tsize))
584 		pkt_len = stab->data[slot];
585 	else
586 		pkt_len = stab->data[stab->szopts.tsize - 1] *
587 				(slot / stab->szopts.tsize) +
588 				stab->data[slot % stab->szopts.tsize];
589 
590 	pkt_len <<= stab->szopts.size_log;
591 out:
592 	if (unlikely(pkt_len < 1))
593 		pkt_len = 1;
594 	qdisc_skb_cb(skb)->pkt_len = pkt_len;
595 }
596 EXPORT_SYMBOL(__qdisc_calculate_pkt_len);
597 
598 void qdisc_warn_nonwc(const char *txt, struct Qdisc *qdisc)
599 {
600 	if (!(qdisc->flags & TCQ_F_WARN_NONWC)) {
601 		pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
602 			txt, qdisc->ops->id, qdisc->handle >> 16);
603 		qdisc->flags |= TCQ_F_WARN_NONWC;
604 	}
605 }
606 EXPORT_SYMBOL(qdisc_warn_nonwc);
607 
608 static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
609 {
610 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
611 						 timer);
612 
613 	rcu_read_lock();
614 	__netif_schedule(qdisc_root(wd->qdisc));
615 	rcu_read_unlock();
616 
617 	return HRTIMER_NORESTART;
618 }
619 
620 void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
621 				 clockid_t clockid)
622 {
623 	hrtimer_init(&wd->timer, clockid, HRTIMER_MODE_ABS_PINNED);
624 	wd->timer.function = qdisc_watchdog;
625 	wd->qdisc = qdisc;
626 }
627 EXPORT_SYMBOL(qdisc_watchdog_init_clockid);
628 
629 void qdisc_watchdog_init(struct qdisc_watchdog *wd, struct Qdisc *qdisc)
630 {
631 	qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC);
632 }
633 EXPORT_SYMBOL(qdisc_watchdog_init);
634 
635 void qdisc_watchdog_schedule_range_ns(struct qdisc_watchdog *wd, u64 expires,
636 				      u64 delta_ns)
637 {
638 	bool deactivated;
639 
640 	rcu_read_lock();
641 	deactivated = test_bit(__QDISC_STATE_DEACTIVATED,
642 			       &qdisc_root_sleeping(wd->qdisc)->state);
643 	rcu_read_unlock();
644 	if (deactivated)
645 		return;
646 
647 	if (hrtimer_is_queued(&wd->timer)) {
648 		u64 softexpires;
649 
650 		softexpires = ktime_to_ns(hrtimer_get_softexpires(&wd->timer));
651 		/* If timer is already set in [expires, expires + delta_ns],
652 		 * do not reprogram it.
653 		 */
654 		if (softexpires - expires <= delta_ns)
655 			return;
656 	}
657 
658 	hrtimer_start_range_ns(&wd->timer,
659 			       ns_to_ktime(expires),
660 			       delta_ns,
661 			       HRTIMER_MODE_ABS_PINNED);
662 }
663 EXPORT_SYMBOL(qdisc_watchdog_schedule_range_ns);
664 
665 void qdisc_watchdog_cancel(struct qdisc_watchdog *wd)
666 {
667 	hrtimer_cancel(&wd->timer);
668 }
669 EXPORT_SYMBOL(qdisc_watchdog_cancel);
670 
671 static struct hlist_head *qdisc_class_hash_alloc(unsigned int n)
672 {
673 	struct hlist_head *h;
674 	unsigned int i;
675 
676 	h = kvmalloc_array(n, sizeof(struct hlist_head), GFP_KERNEL);
677 
678 	if (h != NULL) {
679 		for (i = 0; i < n; i++)
680 			INIT_HLIST_HEAD(&h[i]);
681 	}
682 	return h;
683 }
684 
685 void qdisc_class_hash_grow(struct Qdisc *sch, struct Qdisc_class_hash *clhash)
686 {
687 	struct Qdisc_class_common *cl;
688 	struct hlist_node *next;
689 	struct hlist_head *nhash, *ohash;
690 	unsigned int nsize, nmask, osize;
691 	unsigned int i, h;
692 
693 	/* Rehash when load factor exceeds 0.75 */
694 	if (clhash->hashelems * 4 <= clhash->hashsize * 3)
695 		return;
696 	nsize = clhash->hashsize * 2;
697 	nmask = nsize - 1;
698 	nhash = qdisc_class_hash_alloc(nsize);
699 	if (nhash == NULL)
700 		return;
701 
702 	ohash = clhash->hash;
703 	osize = clhash->hashsize;
704 
705 	sch_tree_lock(sch);
706 	for (i = 0; i < osize; i++) {
707 		hlist_for_each_entry_safe(cl, next, &ohash[i], hnode) {
708 			h = qdisc_class_hash(cl->classid, nmask);
709 			hlist_add_head(&cl->hnode, &nhash[h]);
710 		}
711 	}
712 	clhash->hash     = nhash;
713 	clhash->hashsize = nsize;
714 	clhash->hashmask = nmask;
715 	sch_tree_unlock(sch);
716 
717 	kvfree(ohash);
718 }
719 EXPORT_SYMBOL(qdisc_class_hash_grow);
720 
721 int qdisc_class_hash_init(struct Qdisc_class_hash *clhash)
722 {
723 	unsigned int size = 4;
724 
725 	clhash->hash = qdisc_class_hash_alloc(size);
726 	if (!clhash->hash)
727 		return -ENOMEM;
728 	clhash->hashsize  = size;
729 	clhash->hashmask  = size - 1;
730 	clhash->hashelems = 0;
731 	return 0;
732 }
733 EXPORT_SYMBOL(qdisc_class_hash_init);
734 
735 void qdisc_class_hash_destroy(struct Qdisc_class_hash *clhash)
736 {
737 	kvfree(clhash->hash);
738 }
739 EXPORT_SYMBOL(qdisc_class_hash_destroy);
740 
741 void qdisc_class_hash_insert(struct Qdisc_class_hash *clhash,
742 			     struct Qdisc_class_common *cl)
743 {
744 	unsigned int h;
745 
746 	INIT_HLIST_NODE(&cl->hnode);
747 	h = qdisc_class_hash(cl->classid, clhash->hashmask);
748 	hlist_add_head(&cl->hnode, &clhash->hash[h]);
749 	clhash->hashelems++;
750 }
751 EXPORT_SYMBOL(qdisc_class_hash_insert);
752 
753 void qdisc_class_hash_remove(struct Qdisc_class_hash *clhash,
754 			     struct Qdisc_class_common *cl)
755 {
756 	hlist_del(&cl->hnode);
757 	clhash->hashelems--;
758 }
759 EXPORT_SYMBOL(qdisc_class_hash_remove);
760 
761 /* Allocate an unique handle from space managed by kernel
762  * Possible range is [8000-FFFF]:0000 (0x8000 values)
763  */
764 static u32 qdisc_alloc_handle(struct net_device *dev)
765 {
766 	int i = 0x8000;
767 	static u32 autohandle = TC_H_MAKE(0x80000000U, 0);
768 
769 	do {
770 		autohandle += TC_H_MAKE(0x10000U, 0);
771 		if (autohandle == TC_H_MAKE(TC_H_ROOT, 0))
772 			autohandle = TC_H_MAKE(0x80000000U, 0);
773 		if (!qdisc_lookup(dev, autohandle))
774 			return autohandle;
775 		cond_resched();
776 	} while	(--i > 0);
777 
778 	return 0;
779 }
780 
781 void qdisc_tree_reduce_backlog(struct Qdisc *sch, int n, int len)
782 {
783 	bool qdisc_is_offloaded = sch->flags & TCQ_F_OFFLOADED;
784 	const struct Qdisc_class_ops *cops;
785 	unsigned long cl;
786 	u32 parentid;
787 	bool notify;
788 	int drops;
789 
790 	if (n == 0 && len == 0)
791 		return;
792 	drops = max_t(int, n, 0);
793 	rcu_read_lock();
794 	while ((parentid = sch->parent)) {
795 		if (TC_H_MAJ(parentid) == TC_H_MAJ(TC_H_INGRESS))
796 			break;
797 
798 		if (sch->flags & TCQ_F_NOPARENT)
799 			break;
800 		/* Notify parent qdisc only if child qdisc becomes empty.
801 		 *
802 		 * If child was empty even before update then backlog
803 		 * counter is screwed and we skip notification because
804 		 * parent class is already passive.
805 		 *
806 		 * If the original child was offloaded then it is allowed
807 		 * to be seem as empty, so the parent is notified anyway.
808 		 */
809 		notify = !sch->q.qlen && !WARN_ON_ONCE(!n &&
810 						       !qdisc_is_offloaded);
811 		/* TODO: perform the search on a per txq basis */
812 		sch = qdisc_lookup(qdisc_dev(sch), TC_H_MAJ(parentid));
813 		if (sch == NULL) {
814 			WARN_ON_ONCE(parentid != TC_H_ROOT);
815 			break;
816 		}
817 		cops = sch->ops->cl_ops;
818 		if (notify && cops->qlen_notify) {
819 			cl = cops->find(sch, parentid);
820 			cops->qlen_notify(sch, cl);
821 		}
822 		sch->q.qlen -= n;
823 		sch->qstats.backlog -= len;
824 		__qdisc_qstats_drop(sch, drops);
825 	}
826 	rcu_read_unlock();
827 }
828 EXPORT_SYMBOL(qdisc_tree_reduce_backlog);
829 
830 int qdisc_offload_dump_helper(struct Qdisc *sch, enum tc_setup_type type,
831 			      void *type_data)
832 {
833 	struct net_device *dev = qdisc_dev(sch);
834 	int err;
835 
836 	sch->flags &= ~TCQ_F_OFFLOADED;
837 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
838 		return 0;
839 
840 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
841 	if (err == -EOPNOTSUPP)
842 		return 0;
843 
844 	if (!err)
845 		sch->flags |= TCQ_F_OFFLOADED;
846 
847 	return err;
848 }
849 EXPORT_SYMBOL(qdisc_offload_dump_helper);
850 
851 void qdisc_offload_graft_helper(struct net_device *dev, struct Qdisc *sch,
852 				struct Qdisc *new, struct Qdisc *old,
853 				enum tc_setup_type type, void *type_data,
854 				struct netlink_ext_ack *extack)
855 {
856 	bool any_qdisc_is_offloaded;
857 	int err;
858 
859 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
860 		return;
861 
862 	err = dev->netdev_ops->ndo_setup_tc(dev, type, type_data);
863 
864 	/* Don't report error if the graft is part of destroy operation. */
865 	if (!err || !new || new == &noop_qdisc)
866 		return;
867 
868 	/* Don't report error if the parent, the old child and the new
869 	 * one are not offloaded.
870 	 */
871 	any_qdisc_is_offloaded = new->flags & TCQ_F_OFFLOADED;
872 	any_qdisc_is_offloaded |= sch && sch->flags & TCQ_F_OFFLOADED;
873 	any_qdisc_is_offloaded |= old && old->flags & TCQ_F_OFFLOADED;
874 
875 	if (any_qdisc_is_offloaded)
876 		NL_SET_ERR_MSG(extack, "Offloading graft operation failed.");
877 }
878 EXPORT_SYMBOL(qdisc_offload_graft_helper);
879 
880 void qdisc_offload_query_caps(struct net_device *dev,
881 			      enum tc_setup_type type,
882 			      void *caps, size_t caps_len)
883 {
884 	const struct net_device_ops *ops = dev->netdev_ops;
885 	struct tc_query_caps_base base = {
886 		.type = type,
887 		.caps = caps,
888 	};
889 
890 	memset(caps, 0, caps_len);
891 
892 	if (ops->ndo_setup_tc)
893 		ops->ndo_setup_tc(dev, TC_QUERY_CAPS, &base);
894 }
895 EXPORT_SYMBOL(qdisc_offload_query_caps);
896 
897 static void qdisc_offload_graft_root(struct net_device *dev,
898 				     struct Qdisc *new, struct Qdisc *old,
899 				     struct netlink_ext_ack *extack)
900 {
901 	struct tc_root_qopt_offload graft_offload = {
902 		.command	= TC_ROOT_GRAFT,
903 		.handle		= new ? new->handle : 0,
904 		.ingress	= (new && new->flags & TCQ_F_INGRESS) ||
905 				  (old && old->flags & TCQ_F_INGRESS),
906 	};
907 
908 	qdisc_offload_graft_helper(dev, NULL, new, old,
909 				   TC_SETUP_ROOT_QDISC, &graft_offload, extack);
910 }
911 
912 static int tc_fill_qdisc(struct sk_buff *skb, struct Qdisc *q, u32 clid,
913 			 u32 portid, u32 seq, u16 flags, int event,
914 			 struct netlink_ext_ack *extack)
915 {
916 	struct gnet_stats_basic_sync __percpu *cpu_bstats = NULL;
917 	struct gnet_stats_queue __percpu *cpu_qstats = NULL;
918 	struct tcmsg *tcm;
919 	struct nlmsghdr  *nlh;
920 	unsigned char *b = skb_tail_pointer(skb);
921 	struct gnet_dump d;
922 	struct qdisc_size_table *stab;
923 	u32 block_index;
924 	__u32 qlen;
925 
926 	cond_resched();
927 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
928 	if (!nlh)
929 		goto out_nlmsg_trim;
930 	tcm = nlmsg_data(nlh);
931 	tcm->tcm_family = AF_UNSPEC;
932 	tcm->tcm__pad1 = 0;
933 	tcm->tcm__pad2 = 0;
934 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
935 	tcm->tcm_parent = clid;
936 	tcm->tcm_handle = q->handle;
937 	tcm->tcm_info = refcount_read(&q->refcnt);
938 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
939 		goto nla_put_failure;
940 	if (q->ops->ingress_block_get) {
941 		block_index = q->ops->ingress_block_get(q);
942 		if (block_index &&
943 		    nla_put_u32(skb, TCA_INGRESS_BLOCK, block_index))
944 			goto nla_put_failure;
945 	}
946 	if (q->ops->egress_block_get) {
947 		block_index = q->ops->egress_block_get(q);
948 		if (block_index &&
949 		    nla_put_u32(skb, TCA_EGRESS_BLOCK, block_index))
950 			goto nla_put_failure;
951 	}
952 	if (q->ops->dump && q->ops->dump(q, skb) < 0)
953 		goto nla_put_failure;
954 	if (nla_put_u8(skb, TCA_HW_OFFLOAD, !!(q->flags & TCQ_F_OFFLOADED)))
955 		goto nla_put_failure;
956 	qlen = qdisc_qlen_sum(q);
957 
958 	stab = rtnl_dereference(q->stab);
959 	if (stab && qdisc_dump_stab(skb, stab) < 0)
960 		goto nla_put_failure;
961 
962 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
963 					 NULL, &d, TCA_PAD) < 0)
964 		goto nla_put_failure;
965 
966 	if (q->ops->dump_stats && q->ops->dump_stats(q, &d) < 0)
967 		goto nla_put_failure;
968 
969 	if (qdisc_is_percpu_stats(q)) {
970 		cpu_bstats = q->cpu_bstats;
971 		cpu_qstats = q->cpu_qstats;
972 	}
973 
974 	if (gnet_stats_copy_basic(&d, cpu_bstats, &q->bstats, true) < 0 ||
975 	    gnet_stats_copy_rate_est(&d, &q->rate_est) < 0 ||
976 	    gnet_stats_copy_queue(&d, cpu_qstats, &q->qstats, qlen) < 0)
977 		goto nla_put_failure;
978 
979 	if (gnet_stats_finish_copy(&d) < 0)
980 		goto nla_put_failure;
981 
982 	if (extack && extack->_msg &&
983 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
984 		goto out_nlmsg_trim;
985 
986 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
987 
988 	return skb->len;
989 
990 out_nlmsg_trim:
991 nla_put_failure:
992 	nlmsg_trim(skb, b);
993 	return -1;
994 }
995 
996 static bool tc_qdisc_dump_ignore(struct Qdisc *q, bool dump_invisible)
997 {
998 	if (q->flags & TCQ_F_BUILTIN)
999 		return true;
1000 	if ((q->flags & TCQ_F_INVISIBLE) && !dump_invisible)
1001 		return true;
1002 
1003 	return false;
1004 }
1005 
1006 static int qdisc_notify(struct net *net, struct sk_buff *oskb,
1007 			struct nlmsghdr *n, u32 clid,
1008 			struct Qdisc *old, struct Qdisc *new,
1009 			struct netlink_ext_ack *extack)
1010 {
1011 	struct sk_buff *skb;
1012 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1013 
1014 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1015 	if (!skb)
1016 		return -ENOBUFS;
1017 
1018 	if (old && !tc_qdisc_dump_ignore(old, false)) {
1019 		if (tc_fill_qdisc(skb, old, clid, portid, n->nlmsg_seq,
1020 				  0, RTM_DELQDISC, extack) < 0)
1021 			goto err_out;
1022 	}
1023 	if (new && !tc_qdisc_dump_ignore(new, false)) {
1024 		if (tc_fill_qdisc(skb, new, clid, portid, n->nlmsg_seq,
1025 				  old ? NLM_F_REPLACE : 0, RTM_NEWQDISC, extack) < 0)
1026 			goto err_out;
1027 	}
1028 
1029 	if (skb->len)
1030 		return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1031 				      n->nlmsg_flags & NLM_F_ECHO);
1032 
1033 err_out:
1034 	kfree_skb(skb);
1035 	return -EINVAL;
1036 }
1037 
1038 static void notify_and_destroy(struct net *net, struct sk_buff *skb,
1039 			       struct nlmsghdr *n, u32 clid,
1040 			       struct Qdisc *old, struct Qdisc *new,
1041 			       struct netlink_ext_ack *extack)
1042 {
1043 	if (new || old)
1044 		qdisc_notify(net, skb, n, clid, old, new, extack);
1045 
1046 	if (old)
1047 		qdisc_put(old);
1048 }
1049 
1050 static void qdisc_clear_nolock(struct Qdisc *sch)
1051 {
1052 	sch->flags &= ~TCQ_F_NOLOCK;
1053 	if (!(sch->flags & TCQ_F_CPUSTATS))
1054 		return;
1055 
1056 	free_percpu(sch->cpu_bstats);
1057 	free_percpu(sch->cpu_qstats);
1058 	sch->cpu_bstats = NULL;
1059 	sch->cpu_qstats = NULL;
1060 	sch->flags &= ~TCQ_F_CPUSTATS;
1061 }
1062 
1063 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
1064  * to device "dev".
1065  *
1066  * When appropriate send a netlink notification using 'skb'
1067  * and "n".
1068  *
1069  * On success, destroy old qdisc.
1070  */
1071 
1072 static int qdisc_graft(struct net_device *dev, struct Qdisc *parent,
1073 		       struct sk_buff *skb, struct nlmsghdr *n, u32 classid,
1074 		       struct Qdisc *new, struct Qdisc *old,
1075 		       struct netlink_ext_ack *extack)
1076 {
1077 	struct Qdisc *q = old;
1078 	struct net *net = dev_net(dev);
1079 
1080 	if (parent == NULL) {
1081 		unsigned int i, num_q, ingress;
1082 		struct netdev_queue *dev_queue;
1083 
1084 		ingress = 0;
1085 		num_q = dev->num_tx_queues;
1086 		if ((q && q->flags & TCQ_F_INGRESS) ||
1087 		    (new && new->flags & TCQ_F_INGRESS)) {
1088 			ingress = 1;
1089 			dev_queue = dev_ingress_queue(dev);
1090 			if (!dev_queue) {
1091 				NL_SET_ERR_MSG(extack, "Device does not have an ingress queue");
1092 				return -ENOENT;
1093 			}
1094 
1095 			q = rtnl_dereference(dev_queue->qdisc_sleeping);
1096 
1097 			/* This is the counterpart of that qdisc_refcount_inc_nz() call in
1098 			 * __tcf_qdisc_find() for filter requests.
1099 			 */
1100 			if (!qdisc_refcount_dec_if_one(q)) {
1101 				NL_SET_ERR_MSG(extack,
1102 					       "Current ingress or clsact Qdisc has ongoing filter requests");
1103 				return -EBUSY;
1104 			}
1105 		}
1106 
1107 		if (dev->flags & IFF_UP)
1108 			dev_deactivate(dev);
1109 
1110 		qdisc_offload_graft_root(dev, new, old, extack);
1111 
1112 		if (new && new->ops->attach && !ingress)
1113 			goto skip;
1114 
1115 		if (!ingress) {
1116 			for (i = 0; i < num_q; i++) {
1117 				dev_queue = netdev_get_tx_queue(dev, i);
1118 				old = dev_graft_qdisc(dev_queue, new);
1119 
1120 				if (new && i > 0)
1121 					qdisc_refcount_inc(new);
1122 				qdisc_put(old);
1123 			}
1124 		} else {
1125 			old = dev_graft_qdisc(dev_queue, NULL);
1126 
1127 			/* {ingress,clsact}_destroy() @old before grafting @new to avoid
1128 			 * unprotected concurrent accesses to net_device::miniq_{in,e}gress
1129 			 * pointer(s) in mini_qdisc_pair_swap().
1130 			 */
1131 			qdisc_notify(net, skb, n, classid, old, new, extack);
1132 			qdisc_destroy(old);
1133 
1134 			dev_graft_qdisc(dev_queue, new);
1135 		}
1136 
1137 skip:
1138 		if (!ingress) {
1139 			old = rtnl_dereference(dev->qdisc);
1140 			if (new && !new->ops->attach)
1141 				qdisc_refcount_inc(new);
1142 			rcu_assign_pointer(dev->qdisc, new ? : &noop_qdisc);
1143 
1144 			notify_and_destroy(net, skb, n, classid, old, new, extack);
1145 
1146 			if (new && new->ops->attach)
1147 				new->ops->attach(new);
1148 		}
1149 
1150 		if (dev->flags & IFF_UP)
1151 			dev_activate(dev);
1152 	} else {
1153 		const struct Qdisc_class_ops *cops = parent->ops->cl_ops;
1154 		unsigned long cl;
1155 		int err;
1156 
1157 		/* Only support running class lockless if parent is lockless */
1158 		if (new && (new->flags & TCQ_F_NOLOCK) && !(parent->flags & TCQ_F_NOLOCK))
1159 			qdisc_clear_nolock(new);
1160 
1161 		if (!cops || !cops->graft)
1162 			return -EOPNOTSUPP;
1163 
1164 		cl = cops->find(parent, classid);
1165 		if (!cl) {
1166 			NL_SET_ERR_MSG(extack, "Specified class not found");
1167 			return -ENOENT;
1168 		}
1169 
1170 		if (new && new->ops == &noqueue_qdisc_ops) {
1171 			NL_SET_ERR_MSG(extack, "Cannot assign noqueue to a class");
1172 			return -EINVAL;
1173 		}
1174 
1175 		err = cops->graft(parent, cl, new, &old, extack);
1176 		if (err)
1177 			return err;
1178 		notify_and_destroy(net, skb, n, classid, old, new, extack);
1179 	}
1180 	return 0;
1181 }
1182 
1183 static int qdisc_block_add_dev(struct Qdisc *sch, struct net_device *dev,
1184 			       struct netlink_ext_ack *extack)
1185 {
1186 	const struct Qdisc_class_ops *cl_ops = sch->ops->cl_ops;
1187 	struct tcf_block *block;
1188 	int err;
1189 
1190 	block = cl_ops->tcf_block(sch, TC_H_MIN_INGRESS, NULL);
1191 	if (block) {
1192 		err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1193 		if (err) {
1194 			NL_SET_ERR_MSG(extack,
1195 				       "ingress block dev insert failed");
1196 			return err;
1197 		}
1198 	}
1199 
1200 	block = cl_ops->tcf_block(sch, TC_H_MIN_EGRESS, NULL);
1201 	if (block) {
1202 		err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
1203 		if (err) {
1204 			NL_SET_ERR_MSG(extack,
1205 				       "Egress block dev insert failed");
1206 			goto err_out;
1207 		}
1208 	}
1209 
1210 	return 0;
1211 
1212 err_out:
1213 	block = cl_ops->tcf_block(sch, TC_H_MIN_INGRESS, NULL);
1214 	if (block)
1215 		xa_erase(&block->ports, dev->ifindex);
1216 
1217 	return err;
1218 }
1219 
1220 static int qdisc_block_indexes_set(struct Qdisc *sch, struct nlattr **tca,
1221 				   struct netlink_ext_ack *extack)
1222 {
1223 	u32 block_index;
1224 
1225 	if (tca[TCA_INGRESS_BLOCK]) {
1226 		block_index = nla_get_u32(tca[TCA_INGRESS_BLOCK]);
1227 
1228 		if (!block_index) {
1229 			NL_SET_ERR_MSG(extack, "Ingress block index cannot be 0");
1230 			return -EINVAL;
1231 		}
1232 		if (!sch->ops->ingress_block_set) {
1233 			NL_SET_ERR_MSG(extack, "Ingress block sharing is not supported");
1234 			return -EOPNOTSUPP;
1235 		}
1236 		sch->ops->ingress_block_set(sch, block_index);
1237 	}
1238 	if (tca[TCA_EGRESS_BLOCK]) {
1239 		block_index = nla_get_u32(tca[TCA_EGRESS_BLOCK]);
1240 
1241 		if (!block_index) {
1242 			NL_SET_ERR_MSG(extack, "Egress block index cannot be 0");
1243 			return -EINVAL;
1244 		}
1245 		if (!sch->ops->egress_block_set) {
1246 			NL_SET_ERR_MSG(extack, "Egress block sharing is not supported");
1247 			return -EOPNOTSUPP;
1248 		}
1249 		sch->ops->egress_block_set(sch, block_index);
1250 	}
1251 	return 0;
1252 }
1253 
1254 /*
1255    Allocate and initialize new qdisc.
1256 
1257    Parameters are passed via opt.
1258  */
1259 
1260 static struct Qdisc *qdisc_create(struct net_device *dev,
1261 				  struct netdev_queue *dev_queue,
1262 				  u32 parent, u32 handle,
1263 				  struct nlattr **tca, int *errp,
1264 				  struct netlink_ext_ack *extack)
1265 {
1266 	int err;
1267 	struct nlattr *kind = tca[TCA_KIND];
1268 	struct Qdisc *sch;
1269 	struct Qdisc_ops *ops;
1270 	struct qdisc_size_table *stab;
1271 
1272 	ops = qdisc_lookup_ops(kind);
1273 #ifdef CONFIG_MODULES
1274 	if (ops == NULL && kind != NULL) {
1275 		char name[IFNAMSIZ];
1276 		if (nla_strscpy(name, kind, IFNAMSIZ) >= 0) {
1277 			/* We dropped the RTNL semaphore in order to
1278 			 * perform the module load.  So, even if we
1279 			 * succeeded in loading the module we have to
1280 			 * tell the caller to replay the request.  We
1281 			 * indicate this using -EAGAIN.
1282 			 * We replay the request because the device may
1283 			 * go away in the mean time.
1284 			 */
1285 			rtnl_unlock();
1286 			request_module("sch_%s", name);
1287 			rtnl_lock();
1288 			ops = qdisc_lookup_ops(kind);
1289 			if (ops != NULL) {
1290 				/* We will try again qdisc_lookup_ops,
1291 				 * so don't keep a reference.
1292 				 */
1293 				module_put(ops->owner);
1294 				err = -EAGAIN;
1295 				goto err_out;
1296 			}
1297 		}
1298 	}
1299 #endif
1300 
1301 	err = -ENOENT;
1302 	if (!ops) {
1303 		NL_SET_ERR_MSG(extack, "Specified qdisc kind is unknown");
1304 		goto err_out;
1305 	}
1306 
1307 	sch = qdisc_alloc(dev_queue, ops, extack);
1308 	if (IS_ERR(sch)) {
1309 		err = PTR_ERR(sch);
1310 		goto err_out2;
1311 	}
1312 
1313 	sch->parent = parent;
1314 
1315 	if (handle == TC_H_INGRESS) {
1316 		if (!(sch->flags & TCQ_F_INGRESS)) {
1317 			NL_SET_ERR_MSG(extack,
1318 				       "Specified parent ID is reserved for ingress and clsact Qdiscs");
1319 			err = -EINVAL;
1320 			goto err_out3;
1321 		}
1322 		handle = TC_H_MAKE(TC_H_INGRESS, 0);
1323 	} else {
1324 		if (handle == 0) {
1325 			handle = qdisc_alloc_handle(dev);
1326 			if (handle == 0) {
1327 				NL_SET_ERR_MSG(extack, "Maximum number of qdisc handles was exceeded");
1328 				err = -ENOSPC;
1329 				goto err_out3;
1330 			}
1331 		}
1332 		if (!netif_is_multiqueue(dev))
1333 			sch->flags |= TCQ_F_ONETXQUEUE;
1334 	}
1335 
1336 	sch->handle = handle;
1337 
1338 	/* This exist to keep backward compatible with a userspace
1339 	 * loophole, what allowed userspace to get IFF_NO_QUEUE
1340 	 * facility on older kernels by setting tx_queue_len=0 (prior
1341 	 * to qdisc init), and then forgot to reinit tx_queue_len
1342 	 * before again attaching a qdisc.
1343 	 */
1344 	if ((dev->priv_flags & IFF_NO_QUEUE) && (dev->tx_queue_len == 0)) {
1345 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
1346 		netdev_info(dev, "Caught tx_queue_len zero misconfig\n");
1347 	}
1348 
1349 	err = qdisc_block_indexes_set(sch, tca, extack);
1350 	if (err)
1351 		goto err_out3;
1352 
1353 	if (tca[TCA_STAB]) {
1354 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1355 		if (IS_ERR(stab)) {
1356 			err = PTR_ERR(stab);
1357 			goto err_out3;
1358 		}
1359 		rcu_assign_pointer(sch->stab, stab);
1360 	}
1361 
1362 	if (ops->init) {
1363 		err = ops->init(sch, tca[TCA_OPTIONS], extack);
1364 		if (err != 0)
1365 			goto err_out4;
1366 	}
1367 
1368 	if (tca[TCA_RATE]) {
1369 		err = -EOPNOTSUPP;
1370 		if (sch->flags & TCQ_F_MQROOT) {
1371 			NL_SET_ERR_MSG(extack, "Cannot attach rate estimator to a multi-queue root qdisc");
1372 			goto err_out4;
1373 		}
1374 
1375 		err = gen_new_estimator(&sch->bstats,
1376 					sch->cpu_bstats,
1377 					&sch->rate_est,
1378 					NULL,
1379 					true,
1380 					tca[TCA_RATE]);
1381 		if (err) {
1382 			NL_SET_ERR_MSG(extack, "Failed to generate new estimator");
1383 			goto err_out4;
1384 		}
1385 	}
1386 
1387 	qdisc_hash_add(sch, false);
1388 	trace_qdisc_create(ops, dev, parent);
1389 
1390 	err = qdisc_block_add_dev(sch, dev, extack);
1391 	if (err)
1392 		goto err_out4;
1393 
1394 	return sch;
1395 
1396 err_out4:
1397 	/* Even if ops->init() failed, we call ops->destroy()
1398 	 * like qdisc_create_dflt().
1399 	 */
1400 	if (ops->destroy)
1401 		ops->destroy(sch);
1402 	qdisc_put_stab(rtnl_dereference(sch->stab));
1403 err_out3:
1404 	netdev_put(dev, &sch->dev_tracker);
1405 	qdisc_free(sch);
1406 err_out2:
1407 	module_put(ops->owner);
1408 err_out:
1409 	*errp = err;
1410 	return NULL;
1411 }
1412 
1413 static int qdisc_change(struct Qdisc *sch, struct nlattr **tca,
1414 			struct netlink_ext_ack *extack)
1415 {
1416 	struct qdisc_size_table *ostab, *stab = NULL;
1417 	int err = 0;
1418 
1419 	if (tca[TCA_OPTIONS]) {
1420 		if (!sch->ops->change) {
1421 			NL_SET_ERR_MSG(extack, "Change operation not supported by specified qdisc");
1422 			return -EINVAL;
1423 		}
1424 		if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
1425 			NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
1426 			return -EOPNOTSUPP;
1427 		}
1428 		err = sch->ops->change(sch, tca[TCA_OPTIONS], extack);
1429 		if (err)
1430 			return err;
1431 	}
1432 
1433 	if (tca[TCA_STAB]) {
1434 		stab = qdisc_get_stab(tca[TCA_STAB], extack);
1435 		if (IS_ERR(stab))
1436 			return PTR_ERR(stab);
1437 	}
1438 
1439 	ostab = rtnl_dereference(sch->stab);
1440 	rcu_assign_pointer(sch->stab, stab);
1441 	qdisc_put_stab(ostab);
1442 
1443 	if (tca[TCA_RATE]) {
1444 		/* NB: ignores errors from replace_estimator
1445 		   because change can't be undone. */
1446 		if (sch->flags & TCQ_F_MQROOT)
1447 			goto out;
1448 		gen_replace_estimator(&sch->bstats,
1449 				      sch->cpu_bstats,
1450 				      &sch->rate_est,
1451 				      NULL,
1452 				      true,
1453 				      tca[TCA_RATE]);
1454 	}
1455 out:
1456 	return 0;
1457 }
1458 
1459 struct check_loop_arg {
1460 	struct qdisc_walker	w;
1461 	struct Qdisc		*p;
1462 	int			depth;
1463 };
1464 
1465 static int check_loop_fn(struct Qdisc *q, unsigned long cl,
1466 			 struct qdisc_walker *w);
1467 
1468 static int check_loop(struct Qdisc *q, struct Qdisc *p, int depth)
1469 {
1470 	struct check_loop_arg	arg;
1471 
1472 	if (q->ops->cl_ops == NULL)
1473 		return 0;
1474 
1475 	arg.w.stop = arg.w.skip = arg.w.count = 0;
1476 	arg.w.fn = check_loop_fn;
1477 	arg.depth = depth;
1478 	arg.p = p;
1479 	q->ops->cl_ops->walk(q, &arg.w);
1480 	return arg.w.stop ? -ELOOP : 0;
1481 }
1482 
1483 static int
1484 check_loop_fn(struct Qdisc *q, unsigned long cl, struct qdisc_walker *w)
1485 {
1486 	struct Qdisc *leaf;
1487 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1488 	struct check_loop_arg *arg = (struct check_loop_arg *)w;
1489 
1490 	leaf = cops->leaf(q, cl);
1491 	if (leaf) {
1492 		if (leaf == arg->p || arg->depth > 7)
1493 			return -ELOOP;
1494 		return check_loop(leaf, arg->p, arg->depth + 1);
1495 	}
1496 	return 0;
1497 }
1498 
1499 const struct nla_policy rtm_tca_policy[TCA_MAX + 1] = {
1500 	[TCA_KIND]		= { .type = NLA_STRING },
1501 	[TCA_RATE]		= { .type = NLA_BINARY,
1502 				    .len = sizeof(struct tc_estimator) },
1503 	[TCA_STAB]		= { .type = NLA_NESTED },
1504 	[TCA_DUMP_INVISIBLE]	= { .type = NLA_FLAG },
1505 	[TCA_CHAIN]		= { .type = NLA_U32 },
1506 	[TCA_INGRESS_BLOCK]	= { .type = NLA_U32 },
1507 	[TCA_EGRESS_BLOCK]	= { .type = NLA_U32 },
1508 };
1509 
1510 /*
1511  * Delete/get qdisc.
1512  */
1513 
1514 static int tc_get_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1515 			struct netlink_ext_ack *extack)
1516 {
1517 	struct net *net = sock_net(skb->sk);
1518 	struct tcmsg *tcm = nlmsg_data(n);
1519 	struct nlattr *tca[TCA_MAX + 1];
1520 	struct net_device *dev;
1521 	u32 clid;
1522 	struct Qdisc *q = NULL;
1523 	struct Qdisc *p = NULL;
1524 	int err;
1525 
1526 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1527 				     rtm_tca_policy, extack);
1528 	if (err < 0)
1529 		return err;
1530 
1531 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1532 	if (!dev)
1533 		return -ENODEV;
1534 
1535 	clid = tcm->tcm_parent;
1536 	if (clid) {
1537 		if (clid != TC_H_ROOT) {
1538 			if (TC_H_MAJ(clid) != TC_H_MAJ(TC_H_INGRESS)) {
1539 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1540 				if (!p) {
1541 					NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified classid");
1542 					return -ENOENT;
1543 				}
1544 				q = qdisc_leaf(p, clid);
1545 			} else if (dev_ingress_queue(dev)) {
1546 				q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1547 			}
1548 		} else {
1549 			q = rtnl_dereference(dev->qdisc);
1550 		}
1551 		if (!q) {
1552 			NL_SET_ERR_MSG(extack, "Cannot find specified qdisc on specified device");
1553 			return -ENOENT;
1554 		}
1555 
1556 		if (tcm->tcm_handle && q->handle != tcm->tcm_handle) {
1557 			NL_SET_ERR_MSG(extack, "Invalid handle");
1558 			return -EINVAL;
1559 		}
1560 	} else {
1561 		q = qdisc_lookup(dev, tcm->tcm_handle);
1562 		if (!q) {
1563 			NL_SET_ERR_MSG(extack, "Failed to find qdisc with specified handle");
1564 			return -ENOENT;
1565 		}
1566 	}
1567 
1568 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1569 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1570 		return -EINVAL;
1571 	}
1572 
1573 	if (n->nlmsg_type == RTM_DELQDISC) {
1574 		if (!clid) {
1575 			NL_SET_ERR_MSG(extack, "Classid cannot be zero");
1576 			return -EINVAL;
1577 		}
1578 		if (q->handle == 0) {
1579 			NL_SET_ERR_MSG(extack, "Cannot delete qdisc with handle of zero");
1580 			return -ENOENT;
1581 		}
1582 		err = qdisc_graft(dev, p, skb, n, clid, NULL, q, extack);
1583 		if (err != 0)
1584 			return err;
1585 	} else {
1586 		qdisc_notify(net, skb, n, clid, NULL, q, NULL);
1587 	}
1588 	return 0;
1589 }
1590 
1591 static bool req_create_or_replace(struct nlmsghdr *n)
1592 {
1593 	return (n->nlmsg_flags & NLM_F_CREATE &&
1594 		n->nlmsg_flags & NLM_F_REPLACE);
1595 }
1596 
1597 static bool req_create_exclusive(struct nlmsghdr *n)
1598 {
1599 	return (n->nlmsg_flags & NLM_F_CREATE &&
1600 		n->nlmsg_flags & NLM_F_EXCL);
1601 }
1602 
1603 static bool req_change(struct nlmsghdr *n)
1604 {
1605 	return (!(n->nlmsg_flags & NLM_F_CREATE) &&
1606 		!(n->nlmsg_flags & NLM_F_REPLACE) &&
1607 		!(n->nlmsg_flags & NLM_F_EXCL));
1608 }
1609 
1610 /*
1611  * Create/change qdisc.
1612  */
1613 static int tc_modify_qdisc(struct sk_buff *skb, struct nlmsghdr *n,
1614 			   struct netlink_ext_ack *extack)
1615 {
1616 	struct net *net = sock_net(skb->sk);
1617 	struct tcmsg *tcm;
1618 	struct nlattr *tca[TCA_MAX + 1];
1619 	struct net_device *dev;
1620 	u32 clid;
1621 	struct Qdisc *q, *p;
1622 	int err;
1623 
1624 replay:
1625 	/* Reinit, just in case something touches this. */
1626 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
1627 				     rtm_tca_policy, extack);
1628 	if (err < 0)
1629 		return err;
1630 
1631 	tcm = nlmsg_data(n);
1632 	clid = tcm->tcm_parent;
1633 	q = p = NULL;
1634 
1635 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
1636 	if (!dev)
1637 		return -ENODEV;
1638 
1639 
1640 	if (clid) {
1641 		if (clid != TC_H_ROOT) {
1642 			if (clid != TC_H_INGRESS) {
1643 				p = qdisc_lookup(dev, TC_H_MAJ(clid));
1644 				if (!p) {
1645 					NL_SET_ERR_MSG(extack, "Failed to find specified qdisc");
1646 					return -ENOENT;
1647 				}
1648 				q = qdisc_leaf(p, clid);
1649 			} else if (dev_ingress_queue_create(dev)) {
1650 				q = rtnl_dereference(dev_ingress_queue(dev)->qdisc_sleeping);
1651 			}
1652 		} else {
1653 			q = rtnl_dereference(dev->qdisc);
1654 		}
1655 
1656 		/* It may be default qdisc, ignore it */
1657 		if (q && q->handle == 0)
1658 			q = NULL;
1659 
1660 		if (!q || !tcm->tcm_handle || q->handle != tcm->tcm_handle) {
1661 			if (tcm->tcm_handle) {
1662 				if (q && !(n->nlmsg_flags & NLM_F_REPLACE)) {
1663 					NL_SET_ERR_MSG(extack, "NLM_F_REPLACE needed to override");
1664 					return -EEXIST;
1665 				}
1666 				if (TC_H_MIN(tcm->tcm_handle)) {
1667 					NL_SET_ERR_MSG(extack, "Invalid minor handle");
1668 					return -EINVAL;
1669 				}
1670 				q = qdisc_lookup(dev, tcm->tcm_handle);
1671 				if (!q)
1672 					goto create_n_graft;
1673 				if (n->nlmsg_flags & NLM_F_EXCL) {
1674 					NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot override");
1675 					return -EEXIST;
1676 				}
1677 				if (tca[TCA_KIND] &&
1678 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1679 					NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1680 					return -EINVAL;
1681 				}
1682 				if (q->flags & TCQ_F_INGRESS) {
1683 					NL_SET_ERR_MSG(extack,
1684 						       "Cannot regraft ingress or clsact Qdiscs");
1685 					return -EINVAL;
1686 				}
1687 				if (q == p ||
1688 				    (p && check_loop(q, p, 0))) {
1689 					NL_SET_ERR_MSG(extack, "Qdisc parent/child loop detected");
1690 					return -ELOOP;
1691 				}
1692 				if (clid == TC_H_INGRESS) {
1693 					NL_SET_ERR_MSG(extack, "Ingress cannot graft directly");
1694 					return -EINVAL;
1695 				}
1696 				qdisc_refcount_inc(q);
1697 				goto graft;
1698 			} else {
1699 				if (!q)
1700 					goto create_n_graft;
1701 
1702 				/* This magic test requires explanation.
1703 				 *
1704 				 *   We know, that some child q is already
1705 				 *   attached to this parent and have choice:
1706 				 *   1) change it or 2) create/graft new one.
1707 				 *   If the requested qdisc kind is different
1708 				 *   than the existing one, then we choose graft.
1709 				 *   If they are the same then this is "change"
1710 				 *   operation - just let it fallthrough..
1711 				 *
1712 				 *   1. We are allowed to create/graft only
1713 				 *   if the request is explicitly stating
1714 				 *   "please create if it doesn't exist".
1715 				 *
1716 				 *   2. If the request is to exclusive create
1717 				 *   then the qdisc tcm_handle is not expected
1718 				 *   to exist, so that we choose create/graft too.
1719 				 *
1720 				 *   3. The last case is when no flags are set.
1721 				 *   This will happen when for example tc
1722 				 *   utility issues a "change" command.
1723 				 *   Alas, it is sort of hole in API, we
1724 				 *   cannot decide what to do unambiguously.
1725 				 *   For now we select create/graft.
1726 				 */
1727 				if (tca[TCA_KIND] &&
1728 				    nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1729 					if (req_create_or_replace(n) ||
1730 					    req_create_exclusive(n))
1731 						goto create_n_graft;
1732 					else if (req_change(n))
1733 						goto create_n_graft2;
1734 				}
1735 			}
1736 		}
1737 	} else {
1738 		if (!tcm->tcm_handle) {
1739 			NL_SET_ERR_MSG(extack, "Handle cannot be zero");
1740 			return -EINVAL;
1741 		}
1742 		q = qdisc_lookup(dev, tcm->tcm_handle);
1743 	}
1744 
1745 	/* Change qdisc parameters */
1746 	if (!q) {
1747 		NL_SET_ERR_MSG(extack, "Specified qdisc not found");
1748 		return -ENOENT;
1749 	}
1750 	if (n->nlmsg_flags & NLM_F_EXCL) {
1751 		NL_SET_ERR_MSG(extack, "Exclusivity flag on, cannot modify");
1752 		return -EEXIST;
1753 	}
1754 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], q->ops->id)) {
1755 		NL_SET_ERR_MSG(extack, "Invalid qdisc name");
1756 		return -EINVAL;
1757 	}
1758 	err = qdisc_change(q, tca, extack);
1759 	if (err == 0)
1760 		qdisc_notify(net, skb, n, clid, NULL, q, extack);
1761 	return err;
1762 
1763 create_n_graft:
1764 	if (!(n->nlmsg_flags & NLM_F_CREATE)) {
1765 		NL_SET_ERR_MSG(extack, "Qdisc not found. To create specify NLM_F_CREATE flag");
1766 		return -ENOENT;
1767 	}
1768 create_n_graft2:
1769 	if (clid == TC_H_INGRESS) {
1770 		if (dev_ingress_queue(dev)) {
1771 			q = qdisc_create(dev, dev_ingress_queue(dev),
1772 					 tcm->tcm_parent, tcm->tcm_parent,
1773 					 tca, &err, extack);
1774 		} else {
1775 			NL_SET_ERR_MSG(extack, "Cannot find ingress queue for specified device");
1776 			err = -ENOENT;
1777 		}
1778 	} else {
1779 		struct netdev_queue *dev_queue;
1780 
1781 		if (p && p->ops->cl_ops && p->ops->cl_ops->select_queue)
1782 			dev_queue = p->ops->cl_ops->select_queue(p, tcm);
1783 		else if (p)
1784 			dev_queue = p->dev_queue;
1785 		else
1786 			dev_queue = netdev_get_tx_queue(dev, 0);
1787 
1788 		q = qdisc_create(dev, dev_queue,
1789 				 tcm->tcm_parent, tcm->tcm_handle,
1790 				 tca, &err, extack);
1791 	}
1792 	if (q == NULL) {
1793 		if (err == -EAGAIN)
1794 			goto replay;
1795 		return err;
1796 	}
1797 
1798 graft:
1799 	err = qdisc_graft(dev, p, skb, n, clid, q, NULL, extack);
1800 	if (err) {
1801 		if (q)
1802 			qdisc_put(q);
1803 		return err;
1804 	}
1805 
1806 	return 0;
1807 }
1808 
1809 static int tc_dump_qdisc_root(struct Qdisc *root, struct sk_buff *skb,
1810 			      struct netlink_callback *cb,
1811 			      int *q_idx_p, int s_q_idx, bool recur,
1812 			      bool dump_invisible)
1813 {
1814 	int ret = 0, q_idx = *q_idx_p;
1815 	struct Qdisc *q;
1816 	int b;
1817 
1818 	if (!root)
1819 		return 0;
1820 
1821 	q = root;
1822 	if (q_idx < s_q_idx) {
1823 		q_idx++;
1824 	} else {
1825 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1826 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1827 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1828 				  RTM_NEWQDISC, NULL) <= 0)
1829 			goto done;
1830 		q_idx++;
1831 	}
1832 
1833 	/* If dumping singletons, there is no qdisc_dev(root) and the singleton
1834 	 * itself has already been dumped.
1835 	 *
1836 	 * If we've already dumped the top-level (ingress) qdisc above and the global
1837 	 * qdisc hashtable, we don't want to hit it again
1838 	 */
1839 	if (!qdisc_dev(root) || !recur)
1840 		goto out;
1841 
1842 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
1843 		if (q_idx < s_q_idx) {
1844 			q_idx++;
1845 			continue;
1846 		}
1847 		if (!tc_qdisc_dump_ignore(q, dump_invisible) &&
1848 		    tc_fill_qdisc(skb, q, q->parent, NETLINK_CB(cb->skb).portid,
1849 				  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1850 				  RTM_NEWQDISC, NULL) <= 0)
1851 			goto done;
1852 		q_idx++;
1853 	}
1854 
1855 out:
1856 	*q_idx_p = q_idx;
1857 	return ret;
1858 done:
1859 	ret = -1;
1860 	goto out;
1861 }
1862 
1863 static int tc_dump_qdisc(struct sk_buff *skb, struct netlink_callback *cb)
1864 {
1865 	struct net *net = sock_net(skb->sk);
1866 	int idx, q_idx;
1867 	int s_idx, s_q_idx;
1868 	struct net_device *dev;
1869 	const struct nlmsghdr *nlh = cb->nlh;
1870 	struct nlattr *tca[TCA_MAX + 1];
1871 	int err;
1872 
1873 	s_idx = cb->args[0];
1874 	s_q_idx = q_idx = cb->args[1];
1875 
1876 	idx = 0;
1877 	ASSERT_RTNL();
1878 
1879 	err = nlmsg_parse_deprecated(nlh, sizeof(struct tcmsg), tca, TCA_MAX,
1880 				     rtm_tca_policy, cb->extack);
1881 	if (err < 0)
1882 		return err;
1883 
1884 	for_each_netdev(net, dev) {
1885 		struct netdev_queue *dev_queue;
1886 
1887 		if (idx < s_idx)
1888 			goto cont;
1889 		if (idx > s_idx)
1890 			s_q_idx = 0;
1891 		q_idx = 0;
1892 
1893 		if (tc_dump_qdisc_root(rtnl_dereference(dev->qdisc),
1894 				       skb, cb, &q_idx, s_q_idx,
1895 				       true, tca[TCA_DUMP_INVISIBLE]) < 0)
1896 			goto done;
1897 
1898 		dev_queue = dev_ingress_queue(dev);
1899 		if (dev_queue &&
1900 		    tc_dump_qdisc_root(rtnl_dereference(dev_queue->qdisc_sleeping),
1901 				       skb, cb, &q_idx, s_q_idx, false,
1902 				       tca[TCA_DUMP_INVISIBLE]) < 0)
1903 			goto done;
1904 
1905 cont:
1906 		idx++;
1907 	}
1908 
1909 done:
1910 	cb->args[0] = idx;
1911 	cb->args[1] = q_idx;
1912 
1913 	return skb->len;
1914 }
1915 
1916 
1917 
1918 /************************************************
1919  *	Traffic classes manipulation.		*
1920  ************************************************/
1921 
1922 static int tc_fill_tclass(struct sk_buff *skb, struct Qdisc *q,
1923 			  unsigned long cl, u32 portid, u32 seq, u16 flags,
1924 			  int event, struct netlink_ext_ack *extack)
1925 {
1926 	struct tcmsg *tcm;
1927 	struct nlmsghdr  *nlh;
1928 	unsigned char *b = skb_tail_pointer(skb);
1929 	struct gnet_dump d;
1930 	const struct Qdisc_class_ops *cl_ops = q->ops->cl_ops;
1931 
1932 	cond_resched();
1933 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1934 	if (!nlh)
1935 		goto out_nlmsg_trim;
1936 	tcm = nlmsg_data(nlh);
1937 	tcm->tcm_family = AF_UNSPEC;
1938 	tcm->tcm__pad1 = 0;
1939 	tcm->tcm__pad2 = 0;
1940 	tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1941 	tcm->tcm_parent = q->handle;
1942 	tcm->tcm_handle = q->handle;
1943 	tcm->tcm_info = 0;
1944 	if (nla_put_string(skb, TCA_KIND, q->ops->id))
1945 		goto nla_put_failure;
1946 	if (cl_ops->dump && cl_ops->dump(q, cl, skb, tcm) < 0)
1947 		goto nla_put_failure;
1948 
1949 	if (gnet_stats_start_copy_compat(skb, TCA_STATS2, TCA_STATS, TCA_XSTATS,
1950 					 NULL, &d, TCA_PAD) < 0)
1951 		goto nla_put_failure;
1952 
1953 	if (cl_ops->dump_stats && cl_ops->dump_stats(q, cl, &d) < 0)
1954 		goto nla_put_failure;
1955 
1956 	if (gnet_stats_finish_copy(&d) < 0)
1957 		goto nla_put_failure;
1958 
1959 	if (extack && extack->_msg &&
1960 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
1961 		goto out_nlmsg_trim;
1962 
1963 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
1964 
1965 	return skb->len;
1966 
1967 out_nlmsg_trim:
1968 nla_put_failure:
1969 	nlmsg_trim(skb, b);
1970 	return -1;
1971 }
1972 
1973 static int tclass_notify(struct net *net, struct sk_buff *oskb,
1974 			 struct nlmsghdr *n, struct Qdisc *q,
1975 			 unsigned long cl, int event, struct netlink_ext_ack *extack)
1976 {
1977 	struct sk_buff *skb;
1978 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
1979 
1980 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1981 	if (!skb)
1982 		return -ENOBUFS;
1983 
1984 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0, event, extack) < 0) {
1985 		kfree_skb(skb);
1986 		return -EINVAL;
1987 	}
1988 
1989 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC,
1990 			      n->nlmsg_flags & NLM_F_ECHO);
1991 }
1992 
1993 static int tclass_del_notify(struct net *net,
1994 			     const struct Qdisc_class_ops *cops,
1995 			     struct sk_buff *oskb, struct nlmsghdr *n,
1996 			     struct Qdisc *q, unsigned long cl,
1997 			     struct netlink_ext_ack *extack)
1998 {
1999 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2000 	struct sk_buff *skb;
2001 	int err = 0;
2002 
2003 	if (!cops->delete)
2004 		return -EOPNOTSUPP;
2005 
2006 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2007 	if (!skb)
2008 		return -ENOBUFS;
2009 
2010 	if (tc_fill_tclass(skb, q, cl, portid, n->nlmsg_seq, 0,
2011 			   RTM_DELTCLASS, extack) < 0) {
2012 		kfree_skb(skb);
2013 		return -EINVAL;
2014 	}
2015 
2016 	err = cops->delete(q, cl, extack);
2017 	if (err) {
2018 		kfree_skb(skb);
2019 		return err;
2020 	}
2021 
2022 	err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2023 			     n->nlmsg_flags & NLM_F_ECHO);
2024 	return err;
2025 }
2026 
2027 #ifdef CONFIG_NET_CLS
2028 
2029 struct tcf_bind_args {
2030 	struct tcf_walker w;
2031 	unsigned long base;
2032 	unsigned long cl;
2033 	u32 classid;
2034 };
2035 
2036 static int tcf_node_bind(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2037 {
2038 	struct tcf_bind_args *a = (void *)arg;
2039 
2040 	if (n && tp->ops->bind_class) {
2041 		struct Qdisc *q = tcf_block_q(tp->chain->block);
2042 
2043 		sch_tree_lock(q);
2044 		tp->ops->bind_class(n, a->classid, a->cl, q, a->base);
2045 		sch_tree_unlock(q);
2046 	}
2047 	return 0;
2048 }
2049 
2050 struct tc_bind_class_args {
2051 	struct qdisc_walker w;
2052 	unsigned long new_cl;
2053 	u32 portid;
2054 	u32 clid;
2055 };
2056 
2057 static int tc_bind_class_walker(struct Qdisc *q, unsigned long cl,
2058 				struct qdisc_walker *w)
2059 {
2060 	struct tc_bind_class_args *a = (struct tc_bind_class_args *)w;
2061 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
2062 	struct tcf_block *block;
2063 	struct tcf_chain *chain;
2064 
2065 	block = cops->tcf_block(q, cl, NULL);
2066 	if (!block)
2067 		return 0;
2068 	for (chain = tcf_get_next_chain(block, NULL);
2069 	     chain;
2070 	     chain = tcf_get_next_chain(block, chain)) {
2071 		struct tcf_proto *tp;
2072 
2073 		for (tp = tcf_get_next_proto(chain, NULL);
2074 		     tp; tp = tcf_get_next_proto(chain, tp)) {
2075 			struct tcf_bind_args arg = {};
2076 
2077 			arg.w.fn = tcf_node_bind;
2078 			arg.classid = a->clid;
2079 			arg.base = cl;
2080 			arg.cl = a->new_cl;
2081 			tp->ops->walk(tp, &arg.w, true);
2082 		}
2083 	}
2084 
2085 	return 0;
2086 }
2087 
2088 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2089 			   unsigned long new_cl)
2090 {
2091 	const struct Qdisc_class_ops *cops = q->ops->cl_ops;
2092 	struct tc_bind_class_args args = {};
2093 
2094 	if (!cops->tcf_block)
2095 		return;
2096 	args.portid = portid;
2097 	args.clid = clid;
2098 	args.new_cl = new_cl;
2099 	args.w.fn = tc_bind_class_walker;
2100 	q->ops->cl_ops->walk(q, &args.w);
2101 }
2102 
2103 #else
2104 
2105 static void tc_bind_tclass(struct Qdisc *q, u32 portid, u32 clid,
2106 			   unsigned long new_cl)
2107 {
2108 }
2109 
2110 #endif
2111 
2112 static int tc_ctl_tclass(struct sk_buff *skb, struct nlmsghdr *n,
2113 			 struct netlink_ext_ack *extack)
2114 {
2115 	struct net *net = sock_net(skb->sk);
2116 	struct tcmsg *tcm = nlmsg_data(n);
2117 	struct nlattr *tca[TCA_MAX + 1];
2118 	struct net_device *dev;
2119 	struct Qdisc *q = NULL;
2120 	const struct Qdisc_class_ops *cops;
2121 	unsigned long cl = 0;
2122 	unsigned long new_cl;
2123 	u32 portid;
2124 	u32 clid;
2125 	u32 qid;
2126 	int err;
2127 
2128 	err = nlmsg_parse_deprecated(n, sizeof(*tcm), tca, TCA_MAX,
2129 				     rtm_tca_policy, extack);
2130 	if (err < 0)
2131 		return err;
2132 
2133 	dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2134 	if (!dev)
2135 		return -ENODEV;
2136 
2137 	/*
2138 	   parent == TC_H_UNSPEC - unspecified parent.
2139 	   parent == TC_H_ROOT   - class is root, which has no parent.
2140 	   parent == X:0	 - parent is root class.
2141 	   parent == X:Y	 - parent is a node in hierarchy.
2142 	   parent == 0:Y	 - parent is X:Y, where X:0 is qdisc.
2143 
2144 	   handle == 0:0	 - generate handle from kernel pool.
2145 	   handle == 0:Y	 - class is X:Y, where X:0 is qdisc.
2146 	   handle == X:Y	 - clear.
2147 	   handle == X:0	 - root class.
2148 	 */
2149 
2150 	/* Step 1. Determine qdisc handle X:0 */
2151 
2152 	portid = tcm->tcm_parent;
2153 	clid = tcm->tcm_handle;
2154 	qid = TC_H_MAJ(clid);
2155 
2156 	if (portid != TC_H_ROOT) {
2157 		u32 qid1 = TC_H_MAJ(portid);
2158 
2159 		if (qid && qid1) {
2160 			/* If both majors are known, they must be identical. */
2161 			if (qid != qid1)
2162 				return -EINVAL;
2163 		} else if (qid1) {
2164 			qid = qid1;
2165 		} else if (qid == 0)
2166 			qid = rtnl_dereference(dev->qdisc)->handle;
2167 
2168 		/* Now qid is genuine qdisc handle consistent
2169 		 * both with parent and child.
2170 		 *
2171 		 * TC_H_MAJ(portid) still may be unspecified, complete it now.
2172 		 */
2173 		if (portid)
2174 			portid = TC_H_MAKE(qid, portid);
2175 	} else {
2176 		if (qid == 0)
2177 			qid = rtnl_dereference(dev->qdisc)->handle;
2178 	}
2179 
2180 	/* OK. Locate qdisc */
2181 	q = qdisc_lookup(dev, qid);
2182 	if (!q)
2183 		return -ENOENT;
2184 
2185 	/* An check that it supports classes */
2186 	cops = q->ops->cl_ops;
2187 	if (cops == NULL)
2188 		return -EINVAL;
2189 
2190 	/* Now try to get class */
2191 	if (clid == 0) {
2192 		if (portid == TC_H_ROOT)
2193 			clid = qid;
2194 	} else
2195 		clid = TC_H_MAKE(qid, clid);
2196 
2197 	if (clid)
2198 		cl = cops->find(q, clid);
2199 
2200 	if (cl == 0) {
2201 		err = -ENOENT;
2202 		if (n->nlmsg_type != RTM_NEWTCLASS ||
2203 		    !(n->nlmsg_flags & NLM_F_CREATE))
2204 			goto out;
2205 	} else {
2206 		switch (n->nlmsg_type) {
2207 		case RTM_NEWTCLASS:
2208 			err = -EEXIST;
2209 			if (n->nlmsg_flags & NLM_F_EXCL)
2210 				goto out;
2211 			break;
2212 		case RTM_DELTCLASS:
2213 			err = tclass_del_notify(net, cops, skb, n, q, cl, extack);
2214 			/* Unbind the class with flilters with 0 */
2215 			tc_bind_tclass(q, portid, clid, 0);
2216 			goto out;
2217 		case RTM_GETTCLASS:
2218 			err = tclass_notify(net, skb, n, q, cl, RTM_NEWTCLASS, extack);
2219 			goto out;
2220 		default:
2221 			err = -EINVAL;
2222 			goto out;
2223 		}
2224 	}
2225 
2226 	if (tca[TCA_INGRESS_BLOCK] || tca[TCA_EGRESS_BLOCK]) {
2227 		NL_SET_ERR_MSG(extack, "Shared blocks are not supported for classes");
2228 		return -EOPNOTSUPP;
2229 	}
2230 
2231 	new_cl = cl;
2232 	err = -EOPNOTSUPP;
2233 	if (cops->change)
2234 		err = cops->change(q, clid, portid, tca, &new_cl, extack);
2235 	if (err == 0) {
2236 		tclass_notify(net, skb, n, q, new_cl, RTM_NEWTCLASS, extack);
2237 		/* We just create a new class, need to do reverse binding. */
2238 		if (cl != new_cl)
2239 			tc_bind_tclass(q, portid, clid, new_cl);
2240 	}
2241 out:
2242 	return err;
2243 }
2244 
2245 struct qdisc_dump_args {
2246 	struct qdisc_walker	w;
2247 	struct sk_buff		*skb;
2248 	struct netlink_callback	*cb;
2249 };
2250 
2251 static int qdisc_class_dump(struct Qdisc *q, unsigned long cl,
2252 			    struct qdisc_walker *arg)
2253 {
2254 	struct qdisc_dump_args *a = (struct qdisc_dump_args *)arg;
2255 
2256 	return tc_fill_tclass(a->skb, q, cl, NETLINK_CB(a->cb->skb).portid,
2257 			      a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2258 			      RTM_NEWTCLASS, NULL);
2259 }
2260 
2261 static int tc_dump_tclass_qdisc(struct Qdisc *q, struct sk_buff *skb,
2262 				struct tcmsg *tcm, struct netlink_callback *cb,
2263 				int *t_p, int s_t)
2264 {
2265 	struct qdisc_dump_args arg;
2266 
2267 	if (tc_qdisc_dump_ignore(q, false) ||
2268 	    *t_p < s_t || !q->ops->cl_ops ||
2269 	    (tcm->tcm_parent &&
2270 	     TC_H_MAJ(tcm->tcm_parent) != q->handle)) {
2271 		(*t_p)++;
2272 		return 0;
2273 	}
2274 	if (*t_p > s_t)
2275 		memset(&cb->args[1], 0, sizeof(cb->args)-sizeof(cb->args[0]));
2276 	arg.w.fn = qdisc_class_dump;
2277 	arg.skb = skb;
2278 	arg.cb = cb;
2279 	arg.w.stop  = 0;
2280 	arg.w.skip = cb->args[1];
2281 	arg.w.count = 0;
2282 	q->ops->cl_ops->walk(q, &arg.w);
2283 	cb->args[1] = arg.w.count;
2284 	if (arg.w.stop)
2285 		return -1;
2286 	(*t_p)++;
2287 	return 0;
2288 }
2289 
2290 static int tc_dump_tclass_root(struct Qdisc *root, struct sk_buff *skb,
2291 			       struct tcmsg *tcm, struct netlink_callback *cb,
2292 			       int *t_p, int s_t, bool recur)
2293 {
2294 	struct Qdisc *q;
2295 	int b;
2296 
2297 	if (!root)
2298 		return 0;
2299 
2300 	if (tc_dump_tclass_qdisc(root, skb, tcm, cb, t_p, s_t) < 0)
2301 		return -1;
2302 
2303 	if (!qdisc_dev(root) || !recur)
2304 		return 0;
2305 
2306 	if (tcm->tcm_parent) {
2307 		q = qdisc_match_from_root(root, TC_H_MAJ(tcm->tcm_parent));
2308 		if (q && q != root &&
2309 		    tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2310 			return -1;
2311 		return 0;
2312 	}
2313 	hash_for_each(qdisc_dev(root)->qdisc_hash, b, q, hash) {
2314 		if (tc_dump_tclass_qdisc(q, skb, tcm, cb, t_p, s_t) < 0)
2315 			return -1;
2316 	}
2317 
2318 	return 0;
2319 }
2320 
2321 static int tc_dump_tclass(struct sk_buff *skb, struct netlink_callback *cb)
2322 {
2323 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2324 	struct net *net = sock_net(skb->sk);
2325 	struct netdev_queue *dev_queue;
2326 	struct net_device *dev;
2327 	int t, s_t;
2328 
2329 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2330 		return 0;
2331 	dev = dev_get_by_index(net, tcm->tcm_ifindex);
2332 	if (!dev)
2333 		return 0;
2334 
2335 	s_t = cb->args[0];
2336 	t = 0;
2337 
2338 	if (tc_dump_tclass_root(rtnl_dereference(dev->qdisc),
2339 				skb, tcm, cb, &t, s_t, true) < 0)
2340 		goto done;
2341 
2342 	dev_queue = dev_ingress_queue(dev);
2343 	if (dev_queue &&
2344 	    tc_dump_tclass_root(rtnl_dereference(dev_queue->qdisc_sleeping),
2345 				skb, tcm, cb, &t, s_t, false) < 0)
2346 		goto done;
2347 
2348 done:
2349 	cb->args[0] = t;
2350 
2351 	dev_put(dev);
2352 	return skb->len;
2353 }
2354 
2355 #ifdef CONFIG_PROC_FS
2356 static int psched_show(struct seq_file *seq, void *v)
2357 {
2358 	seq_printf(seq, "%08x %08x %08x %08x\n",
2359 		   (u32)NSEC_PER_USEC, (u32)PSCHED_TICKS2NS(1),
2360 		   1000000,
2361 		   (u32)NSEC_PER_SEC / hrtimer_resolution);
2362 
2363 	return 0;
2364 }
2365 
2366 static int __net_init psched_net_init(struct net *net)
2367 {
2368 	struct proc_dir_entry *e;
2369 
2370 	e = proc_create_single("psched", 0, net->proc_net, psched_show);
2371 	if (e == NULL)
2372 		return -ENOMEM;
2373 
2374 	return 0;
2375 }
2376 
2377 static void __net_exit psched_net_exit(struct net *net)
2378 {
2379 	remove_proc_entry("psched", net->proc_net);
2380 }
2381 #else
2382 static int __net_init psched_net_init(struct net *net)
2383 {
2384 	return 0;
2385 }
2386 
2387 static void __net_exit psched_net_exit(struct net *net)
2388 {
2389 }
2390 #endif
2391 
2392 static struct pernet_operations psched_net_ops = {
2393 	.init = psched_net_init,
2394 	.exit = psched_net_exit,
2395 };
2396 
2397 #if IS_ENABLED(CONFIG_RETPOLINE)
2398 DEFINE_STATIC_KEY_FALSE(tc_skip_wrapper);
2399 #endif
2400 
2401 static int __init pktsched_init(void)
2402 {
2403 	int err;
2404 
2405 	err = register_pernet_subsys(&psched_net_ops);
2406 	if (err) {
2407 		pr_err("pktsched_init: "
2408 		       "cannot initialize per netns operations\n");
2409 		return err;
2410 	}
2411 
2412 	register_qdisc(&pfifo_fast_ops);
2413 	register_qdisc(&pfifo_qdisc_ops);
2414 	register_qdisc(&bfifo_qdisc_ops);
2415 	register_qdisc(&pfifo_head_drop_qdisc_ops);
2416 	register_qdisc(&mq_qdisc_ops);
2417 	register_qdisc(&noqueue_qdisc_ops);
2418 
2419 	rtnl_register(PF_UNSPEC, RTM_NEWQDISC, tc_modify_qdisc, NULL, 0);
2420 	rtnl_register(PF_UNSPEC, RTM_DELQDISC, tc_get_qdisc, NULL, 0);
2421 	rtnl_register(PF_UNSPEC, RTM_GETQDISC, tc_get_qdisc, tc_dump_qdisc,
2422 		      0);
2423 	rtnl_register(PF_UNSPEC, RTM_NEWTCLASS, tc_ctl_tclass, NULL, 0);
2424 	rtnl_register(PF_UNSPEC, RTM_DELTCLASS, tc_ctl_tclass, NULL, 0);
2425 	rtnl_register(PF_UNSPEC, RTM_GETTCLASS, tc_ctl_tclass, tc_dump_tclass,
2426 		      0);
2427 
2428 	tc_wrapper_init();
2429 
2430 	return 0;
2431 }
2432 
2433 subsys_initcall(pktsched_init);
2434