xref: /linux/net/sched/sch_ets.c (revision 37744feebc086908fd89760650f458ab19071750)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * net/sched/sch_ets.c         Enhanced Transmission Selection scheduler
4  *
5  * Description
6  * -----------
7  *
8  * The Enhanced Transmission Selection scheduler is a classful queuing
9  * discipline that merges functionality of PRIO and DRR qdiscs in one scheduler.
10  * ETS makes it easy to configure a set of strict and bandwidth-sharing bands to
11  * implement the transmission selection described in 802.1Qaz.
12  *
13  * Although ETS is technically classful, it's not possible to add and remove
14  * classes at will. Instead one specifies number of classes, how many are
15  * PRIO-like and how many DRR-like, and quanta for the latter.
16  *
17  * Algorithm
18  * ---------
19  *
20  * The strict classes, if any, are tried for traffic first: first band 0, if it
21  * has no traffic then band 1, etc.
22  *
23  * When there is no traffic in any of the strict queues, the bandwidth-sharing
24  * ones are tried next. Each band is assigned a deficit counter, initialized to
25  * "quantum" of that band. ETS maintains a list of active bandwidth-sharing
26  * bands whose qdiscs are non-empty. A packet is dequeued from the band at the
27  * head of the list if the packet size is smaller or equal to the deficit
28  * counter. If the counter is too small, it is increased by "quantum" and the
29  * scheduler moves on to the next band in the active list.
30  */
31 
32 #include <linux/module.h>
33 #include <net/gen_stats.h>
34 #include <net/netlink.h>
35 #include <net/pkt_cls.h>
36 #include <net/pkt_sched.h>
37 #include <net/sch_generic.h>
38 
39 struct ets_class {
40 	struct list_head alist; /* In struct ets_sched.active. */
41 	struct Qdisc *qdisc;
42 	u32 quantum;
43 	u32 deficit;
44 	struct gnet_stats_basic_packed bstats;
45 	struct gnet_stats_queue qstats;
46 };
47 
48 struct ets_sched {
49 	struct list_head active;
50 	struct tcf_proto __rcu *filter_list;
51 	struct tcf_block *block;
52 	unsigned int nbands;
53 	unsigned int nstrict;
54 	u8 prio2band[TC_PRIO_MAX + 1];
55 	struct ets_class classes[TCQ_ETS_MAX_BANDS];
56 };
57 
58 static const struct nla_policy ets_policy[TCA_ETS_MAX + 1] = {
59 	[TCA_ETS_NBANDS] = { .type = NLA_U8 },
60 	[TCA_ETS_NSTRICT] = { .type = NLA_U8 },
61 	[TCA_ETS_QUANTA] = { .type = NLA_NESTED },
62 	[TCA_ETS_PRIOMAP] = { .type = NLA_NESTED },
63 };
64 
65 static const struct nla_policy ets_priomap_policy[TCA_ETS_MAX + 1] = {
66 	[TCA_ETS_PRIOMAP_BAND] = { .type = NLA_U8 },
67 };
68 
69 static const struct nla_policy ets_quanta_policy[TCA_ETS_MAX + 1] = {
70 	[TCA_ETS_QUANTA_BAND] = { .type = NLA_U32 },
71 };
72 
73 static const struct nla_policy ets_class_policy[TCA_ETS_MAX + 1] = {
74 	[TCA_ETS_QUANTA_BAND] = { .type = NLA_U32 },
75 };
76 
77 static int ets_quantum_parse(struct Qdisc *sch, const struct nlattr *attr,
78 			     unsigned int *quantum,
79 			     struct netlink_ext_ack *extack)
80 {
81 	*quantum = nla_get_u32(attr);
82 	if (!*quantum) {
83 		NL_SET_ERR_MSG(extack, "ETS quantum cannot be zero");
84 		return -EINVAL;
85 	}
86 	return 0;
87 }
88 
89 static struct ets_class *
90 ets_class_from_arg(struct Qdisc *sch, unsigned long arg)
91 {
92 	struct ets_sched *q = qdisc_priv(sch);
93 
94 	return &q->classes[arg - 1];
95 }
96 
97 static u32 ets_class_id(struct Qdisc *sch, const struct ets_class *cl)
98 {
99 	struct ets_sched *q = qdisc_priv(sch);
100 	int band = cl - q->classes;
101 
102 	return TC_H_MAKE(sch->handle, band + 1);
103 }
104 
105 static void ets_offload_change(struct Qdisc *sch)
106 {
107 	struct net_device *dev = qdisc_dev(sch);
108 	struct ets_sched *q = qdisc_priv(sch);
109 	struct tc_ets_qopt_offload qopt;
110 	unsigned int w_psum_prev = 0;
111 	unsigned int q_psum = 0;
112 	unsigned int q_sum = 0;
113 	unsigned int quantum;
114 	unsigned int w_psum;
115 	unsigned int weight;
116 	unsigned int i;
117 
118 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
119 		return;
120 
121 	qopt.command = TC_ETS_REPLACE;
122 	qopt.handle = sch->handle;
123 	qopt.parent = sch->parent;
124 	qopt.replace_params.bands = q->nbands;
125 	qopt.replace_params.qstats = &sch->qstats;
126 	memcpy(&qopt.replace_params.priomap,
127 	       q->prio2band, sizeof(q->prio2band));
128 
129 	for (i = 0; i < q->nbands; i++)
130 		q_sum += q->classes[i].quantum;
131 
132 	for (i = 0; i < q->nbands; i++) {
133 		quantum = q->classes[i].quantum;
134 		q_psum += quantum;
135 		w_psum = quantum ? q_psum * 100 / q_sum : 0;
136 		weight = w_psum - w_psum_prev;
137 		w_psum_prev = w_psum;
138 
139 		qopt.replace_params.quanta[i] = quantum;
140 		qopt.replace_params.weights[i] = weight;
141 	}
142 
143 	dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_ETS, &qopt);
144 }
145 
146 static void ets_offload_destroy(struct Qdisc *sch)
147 {
148 	struct net_device *dev = qdisc_dev(sch);
149 	struct tc_ets_qopt_offload qopt;
150 
151 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
152 		return;
153 
154 	qopt.command = TC_ETS_DESTROY;
155 	qopt.handle = sch->handle;
156 	qopt.parent = sch->parent;
157 	dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_ETS, &qopt);
158 }
159 
160 static void ets_offload_graft(struct Qdisc *sch, struct Qdisc *new,
161 			      struct Qdisc *old, unsigned long arg,
162 			      struct netlink_ext_ack *extack)
163 {
164 	struct net_device *dev = qdisc_dev(sch);
165 	struct tc_ets_qopt_offload qopt;
166 
167 	qopt.command = TC_ETS_GRAFT;
168 	qopt.handle = sch->handle;
169 	qopt.parent = sch->parent;
170 	qopt.graft_params.band = arg - 1;
171 	qopt.graft_params.child_handle = new->handle;
172 
173 	qdisc_offload_graft_helper(dev, sch, new, old, TC_SETUP_QDISC_ETS,
174 				   &qopt, extack);
175 }
176 
177 static int ets_offload_dump(struct Qdisc *sch)
178 {
179 	struct tc_ets_qopt_offload qopt;
180 
181 	qopt.command = TC_ETS_STATS;
182 	qopt.handle = sch->handle;
183 	qopt.parent = sch->parent;
184 	qopt.stats.bstats = &sch->bstats;
185 	qopt.stats.qstats = &sch->qstats;
186 
187 	return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_ETS, &qopt);
188 }
189 
190 static bool ets_class_is_strict(struct ets_sched *q, const struct ets_class *cl)
191 {
192 	unsigned int band = cl - q->classes;
193 
194 	return band < q->nstrict;
195 }
196 
197 static int ets_class_change(struct Qdisc *sch, u32 classid, u32 parentid,
198 			    struct nlattr **tca, unsigned long *arg,
199 			    struct netlink_ext_ack *extack)
200 {
201 	struct ets_class *cl = ets_class_from_arg(sch, *arg);
202 	struct ets_sched *q = qdisc_priv(sch);
203 	struct nlattr *opt = tca[TCA_OPTIONS];
204 	struct nlattr *tb[TCA_ETS_MAX + 1];
205 	unsigned int quantum;
206 	int err;
207 
208 	/* Classes can be added and removed only through Qdisc_ops.change
209 	 * interface.
210 	 */
211 	if (!cl) {
212 		NL_SET_ERR_MSG(extack, "Fine-grained class addition and removal is not supported");
213 		return -EOPNOTSUPP;
214 	}
215 
216 	if (!opt) {
217 		NL_SET_ERR_MSG(extack, "ETS options are required for this operation");
218 		return -EINVAL;
219 	}
220 
221 	err = nla_parse_nested(tb, TCA_ETS_MAX, opt, ets_class_policy, extack);
222 	if (err < 0)
223 		return err;
224 
225 	if (!tb[TCA_ETS_QUANTA_BAND])
226 		/* Nothing to configure. */
227 		return 0;
228 
229 	if (ets_class_is_strict(q, cl)) {
230 		NL_SET_ERR_MSG(extack, "Strict bands do not have a configurable quantum");
231 		return -EINVAL;
232 	}
233 
234 	err = ets_quantum_parse(sch, tb[TCA_ETS_QUANTA_BAND], &quantum,
235 				extack);
236 	if (err)
237 		return err;
238 
239 	sch_tree_lock(sch);
240 	cl->quantum = quantum;
241 	sch_tree_unlock(sch);
242 
243 	ets_offload_change(sch);
244 	return 0;
245 }
246 
247 static int ets_class_graft(struct Qdisc *sch, unsigned long arg,
248 			   struct Qdisc *new, struct Qdisc **old,
249 			   struct netlink_ext_ack *extack)
250 {
251 	struct ets_class *cl = ets_class_from_arg(sch, arg);
252 
253 	if (!new) {
254 		new = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
255 					ets_class_id(sch, cl), NULL);
256 		if (!new)
257 			new = &noop_qdisc;
258 		else
259 			qdisc_hash_add(new, true);
260 	}
261 
262 	*old = qdisc_replace(sch, new, &cl->qdisc);
263 	ets_offload_graft(sch, new, *old, arg, extack);
264 	return 0;
265 }
266 
267 static struct Qdisc *ets_class_leaf(struct Qdisc *sch, unsigned long arg)
268 {
269 	struct ets_class *cl = ets_class_from_arg(sch, arg);
270 
271 	return cl->qdisc;
272 }
273 
274 static unsigned long ets_class_find(struct Qdisc *sch, u32 classid)
275 {
276 	unsigned long band = TC_H_MIN(classid);
277 	struct ets_sched *q = qdisc_priv(sch);
278 
279 	if (band - 1 >= q->nbands)
280 		return 0;
281 	return band;
282 }
283 
284 static void ets_class_qlen_notify(struct Qdisc *sch, unsigned long arg)
285 {
286 	struct ets_class *cl = ets_class_from_arg(sch, arg);
287 	struct ets_sched *q = qdisc_priv(sch);
288 
289 	/* We get notified about zero-length child Qdiscs as well if they are
290 	 * offloaded. Those aren't on the active list though, so don't attempt
291 	 * to remove them.
292 	 */
293 	if (!ets_class_is_strict(q, cl) && sch->q.qlen)
294 		list_del(&cl->alist);
295 }
296 
297 static int ets_class_dump(struct Qdisc *sch, unsigned long arg,
298 			  struct sk_buff *skb, struct tcmsg *tcm)
299 {
300 	struct ets_class *cl = ets_class_from_arg(sch, arg);
301 	struct ets_sched *q = qdisc_priv(sch);
302 	struct nlattr *nest;
303 
304 	tcm->tcm_parent = TC_H_ROOT;
305 	tcm->tcm_handle = ets_class_id(sch, cl);
306 	tcm->tcm_info = cl->qdisc->handle;
307 
308 	nest = nla_nest_start_noflag(skb, TCA_OPTIONS);
309 	if (!nest)
310 		goto nla_put_failure;
311 	if (!ets_class_is_strict(q, cl)) {
312 		if (nla_put_u32(skb, TCA_ETS_QUANTA_BAND, cl->quantum))
313 			goto nla_put_failure;
314 	}
315 	return nla_nest_end(skb, nest);
316 
317 nla_put_failure:
318 	nla_nest_cancel(skb, nest);
319 	return -EMSGSIZE;
320 }
321 
322 static int ets_class_dump_stats(struct Qdisc *sch, unsigned long arg,
323 				struct gnet_dump *d)
324 {
325 	struct ets_class *cl = ets_class_from_arg(sch, arg);
326 	struct Qdisc *cl_q = cl->qdisc;
327 
328 	if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch),
329 				  d, NULL, &cl_q->bstats) < 0 ||
330 	    qdisc_qstats_copy(d, cl_q) < 0)
331 		return -1;
332 
333 	return 0;
334 }
335 
336 static void ets_qdisc_walk(struct Qdisc *sch, struct qdisc_walker *arg)
337 {
338 	struct ets_sched *q = qdisc_priv(sch);
339 	int i;
340 
341 	if (arg->stop)
342 		return;
343 
344 	for (i = 0; i < q->nbands; i++) {
345 		if (arg->count < arg->skip) {
346 			arg->count++;
347 			continue;
348 		}
349 		if (arg->fn(sch, i + 1, arg) < 0) {
350 			arg->stop = 1;
351 			break;
352 		}
353 		arg->count++;
354 	}
355 }
356 
357 static struct tcf_block *
358 ets_qdisc_tcf_block(struct Qdisc *sch, unsigned long cl,
359 		    struct netlink_ext_ack *extack)
360 {
361 	struct ets_sched *q = qdisc_priv(sch);
362 
363 	if (cl) {
364 		NL_SET_ERR_MSG(extack, "ETS classid must be zero");
365 		return NULL;
366 	}
367 
368 	return q->block;
369 }
370 
371 static unsigned long ets_qdisc_bind_tcf(struct Qdisc *sch, unsigned long parent,
372 					u32 classid)
373 {
374 	return ets_class_find(sch, classid);
375 }
376 
377 static void ets_qdisc_unbind_tcf(struct Qdisc *sch, unsigned long arg)
378 {
379 }
380 
381 static struct ets_class *ets_classify(struct sk_buff *skb, struct Qdisc *sch,
382 				      int *qerr)
383 {
384 	struct ets_sched *q = qdisc_priv(sch);
385 	u32 band = skb->priority;
386 	struct tcf_result res;
387 	struct tcf_proto *fl;
388 	int err;
389 
390 	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
391 	if (TC_H_MAJ(skb->priority) != sch->handle) {
392 		fl = rcu_dereference_bh(q->filter_list);
393 		err = tcf_classify(skb, fl, &res, false);
394 #ifdef CONFIG_NET_CLS_ACT
395 		switch (err) {
396 		case TC_ACT_STOLEN:
397 		case TC_ACT_QUEUED:
398 		case TC_ACT_TRAP:
399 			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
400 			/* fall through */
401 		case TC_ACT_SHOT:
402 			return NULL;
403 		}
404 #endif
405 		if (!fl || err < 0) {
406 			if (TC_H_MAJ(band))
407 				band = 0;
408 			return &q->classes[q->prio2band[band & TC_PRIO_MAX]];
409 		}
410 		band = res.classid;
411 	}
412 	band = TC_H_MIN(band) - 1;
413 	if (band >= q->nbands)
414 		return &q->classes[q->prio2band[0]];
415 	return &q->classes[band];
416 }
417 
418 static int ets_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
419 			     struct sk_buff **to_free)
420 {
421 	unsigned int len = qdisc_pkt_len(skb);
422 	struct ets_sched *q = qdisc_priv(sch);
423 	struct ets_class *cl;
424 	int err = 0;
425 	bool first;
426 
427 	cl = ets_classify(skb, sch, &err);
428 	if (!cl) {
429 		if (err & __NET_XMIT_BYPASS)
430 			qdisc_qstats_drop(sch);
431 		__qdisc_drop(skb, to_free);
432 		return err;
433 	}
434 
435 	first = !cl->qdisc->q.qlen;
436 	err = qdisc_enqueue(skb, cl->qdisc, to_free);
437 	if (unlikely(err != NET_XMIT_SUCCESS)) {
438 		if (net_xmit_drop_count(err)) {
439 			cl->qstats.drops++;
440 			qdisc_qstats_drop(sch);
441 		}
442 		return err;
443 	}
444 
445 	if (first && !ets_class_is_strict(q, cl)) {
446 		list_add_tail(&cl->alist, &q->active);
447 		cl->deficit = cl->quantum;
448 	}
449 
450 	sch->qstats.backlog += len;
451 	sch->q.qlen++;
452 	return err;
453 }
454 
455 static struct sk_buff *
456 ets_qdisc_dequeue_skb(struct Qdisc *sch, struct sk_buff *skb)
457 {
458 	qdisc_bstats_update(sch, skb);
459 	qdisc_qstats_backlog_dec(sch, skb);
460 	sch->q.qlen--;
461 	return skb;
462 }
463 
464 static struct sk_buff *ets_qdisc_dequeue(struct Qdisc *sch)
465 {
466 	struct ets_sched *q = qdisc_priv(sch);
467 	struct ets_class *cl;
468 	struct sk_buff *skb;
469 	unsigned int band;
470 	unsigned int len;
471 
472 	while (1) {
473 		for (band = 0; band < q->nstrict; band++) {
474 			cl = &q->classes[band];
475 			skb = qdisc_dequeue_peeked(cl->qdisc);
476 			if (skb)
477 				return ets_qdisc_dequeue_skb(sch, skb);
478 		}
479 
480 		if (list_empty(&q->active))
481 			goto out;
482 
483 		cl = list_first_entry(&q->active, struct ets_class, alist);
484 		skb = cl->qdisc->ops->peek(cl->qdisc);
485 		if (!skb) {
486 			qdisc_warn_nonwc(__func__, cl->qdisc);
487 			goto out;
488 		}
489 
490 		len = qdisc_pkt_len(skb);
491 		if (len <= cl->deficit) {
492 			cl->deficit -= len;
493 			skb = qdisc_dequeue_peeked(cl->qdisc);
494 			if (unlikely(!skb))
495 				goto out;
496 			if (cl->qdisc->q.qlen == 0)
497 				list_del(&cl->alist);
498 			return ets_qdisc_dequeue_skb(sch, skb);
499 		}
500 
501 		cl->deficit += cl->quantum;
502 		list_move_tail(&cl->alist, &q->active);
503 	}
504 out:
505 	return NULL;
506 }
507 
508 static int ets_qdisc_priomap_parse(struct nlattr *priomap_attr,
509 				   unsigned int nbands, u8 *priomap,
510 				   struct netlink_ext_ack *extack)
511 {
512 	const struct nlattr *attr;
513 	int prio = 0;
514 	u8 band;
515 	int rem;
516 	int err;
517 
518 	err = __nla_validate_nested(priomap_attr, TCA_ETS_MAX,
519 				    ets_priomap_policy, NL_VALIDATE_STRICT,
520 				    extack);
521 	if (err)
522 		return err;
523 
524 	nla_for_each_nested(attr, priomap_attr, rem) {
525 		switch (nla_type(attr)) {
526 		case TCA_ETS_PRIOMAP_BAND:
527 			if (prio > TC_PRIO_MAX) {
528 				NL_SET_ERR_MSG_MOD(extack, "Too many priorities in ETS priomap");
529 				return -EINVAL;
530 			}
531 			band = nla_get_u8(attr);
532 			if (band >= nbands) {
533 				NL_SET_ERR_MSG_MOD(extack, "Invalid band number in ETS priomap");
534 				return -EINVAL;
535 			}
536 			priomap[prio++] = band;
537 			break;
538 		default:
539 			WARN_ON_ONCE(1); /* Validate should have caught this. */
540 			return -EINVAL;
541 		}
542 	}
543 
544 	return 0;
545 }
546 
547 static int ets_qdisc_quanta_parse(struct Qdisc *sch, struct nlattr *quanta_attr,
548 				  unsigned int nbands, unsigned int nstrict,
549 				  unsigned int *quanta,
550 				  struct netlink_ext_ack *extack)
551 {
552 	const struct nlattr *attr;
553 	int band = nstrict;
554 	int rem;
555 	int err;
556 
557 	err = __nla_validate_nested(quanta_attr, TCA_ETS_MAX,
558 				    ets_quanta_policy, NL_VALIDATE_STRICT,
559 				    extack);
560 	if (err < 0)
561 		return err;
562 
563 	nla_for_each_nested(attr, quanta_attr, rem) {
564 		switch (nla_type(attr)) {
565 		case TCA_ETS_QUANTA_BAND:
566 			if (band >= nbands) {
567 				NL_SET_ERR_MSG_MOD(extack, "ETS quanta has more values than bands");
568 				return -EINVAL;
569 			}
570 			err = ets_quantum_parse(sch, attr, &quanta[band++],
571 						extack);
572 			if (err)
573 				return err;
574 			break;
575 		default:
576 			WARN_ON_ONCE(1); /* Validate should have caught this. */
577 			return -EINVAL;
578 		}
579 	}
580 
581 	return 0;
582 }
583 
584 static int ets_qdisc_change(struct Qdisc *sch, struct nlattr *opt,
585 			    struct netlink_ext_ack *extack)
586 {
587 	unsigned int quanta[TCQ_ETS_MAX_BANDS] = {0};
588 	struct Qdisc *queues[TCQ_ETS_MAX_BANDS];
589 	struct ets_sched *q = qdisc_priv(sch);
590 	struct nlattr *tb[TCA_ETS_MAX + 1];
591 	unsigned int oldbands = q->nbands;
592 	u8 priomap[TC_PRIO_MAX + 1];
593 	unsigned int nstrict = 0;
594 	unsigned int nbands;
595 	unsigned int i;
596 	int err;
597 
598 	if (!opt) {
599 		NL_SET_ERR_MSG(extack, "ETS options are required for this operation");
600 		return -EINVAL;
601 	}
602 
603 	err = nla_parse_nested(tb, TCA_ETS_MAX, opt, ets_policy, extack);
604 	if (err < 0)
605 		return err;
606 
607 	if (!tb[TCA_ETS_NBANDS]) {
608 		NL_SET_ERR_MSG_MOD(extack, "Number of bands is a required argument");
609 		return -EINVAL;
610 	}
611 	nbands = nla_get_u8(tb[TCA_ETS_NBANDS]);
612 	if (nbands < 1 || nbands > TCQ_ETS_MAX_BANDS) {
613 		NL_SET_ERR_MSG_MOD(extack, "Invalid number of bands");
614 		return -EINVAL;
615 	}
616 	/* Unless overridden, traffic goes to the last band. */
617 	memset(priomap, nbands - 1, sizeof(priomap));
618 
619 	if (tb[TCA_ETS_NSTRICT]) {
620 		nstrict = nla_get_u8(tb[TCA_ETS_NSTRICT]);
621 		if (nstrict > nbands) {
622 			NL_SET_ERR_MSG_MOD(extack, "Invalid number of strict bands");
623 			return -EINVAL;
624 		}
625 	}
626 
627 	if (tb[TCA_ETS_PRIOMAP]) {
628 		err = ets_qdisc_priomap_parse(tb[TCA_ETS_PRIOMAP],
629 					      nbands, priomap, extack);
630 		if (err)
631 			return err;
632 	}
633 
634 	if (tb[TCA_ETS_QUANTA]) {
635 		err = ets_qdisc_quanta_parse(sch, tb[TCA_ETS_QUANTA],
636 					     nbands, nstrict, quanta, extack);
637 		if (err)
638 			return err;
639 	}
640 	/* If there are more bands than strict + quanta provided, the remaining
641 	 * ones are ETS with quantum of MTU. Initialize the missing values here.
642 	 */
643 	for (i = nstrict; i < nbands; i++) {
644 		if (!quanta[i])
645 			quanta[i] = psched_mtu(qdisc_dev(sch));
646 	}
647 
648 	/* Before commit, make sure we can allocate all new qdiscs */
649 	for (i = oldbands; i < nbands; i++) {
650 		queues[i] = qdisc_create_dflt(sch->dev_queue, &pfifo_qdisc_ops,
651 					      ets_class_id(sch, &q->classes[i]),
652 					      extack);
653 		if (!queues[i]) {
654 			while (i > oldbands)
655 				qdisc_put(queues[--i]);
656 			return -ENOMEM;
657 		}
658 	}
659 
660 	sch_tree_lock(sch);
661 
662 	q->nbands = nbands;
663 	q->nstrict = nstrict;
664 	memcpy(q->prio2band, priomap, sizeof(priomap));
665 
666 	for (i = q->nbands; i < oldbands; i++)
667 		qdisc_tree_flush_backlog(q->classes[i].qdisc);
668 
669 	for (i = 0; i < q->nbands; i++)
670 		q->classes[i].quantum = quanta[i];
671 
672 	for (i = oldbands; i < q->nbands; i++) {
673 		q->classes[i].qdisc = queues[i];
674 		if (q->classes[i].qdisc != &noop_qdisc)
675 			qdisc_hash_add(q->classes[i].qdisc, true);
676 	}
677 
678 	sch_tree_unlock(sch);
679 
680 	ets_offload_change(sch);
681 	for (i = q->nbands; i < oldbands; i++) {
682 		qdisc_put(q->classes[i].qdisc);
683 		memset(&q->classes[i], 0, sizeof(q->classes[i]));
684 	}
685 	return 0;
686 }
687 
688 static int ets_qdisc_init(struct Qdisc *sch, struct nlattr *opt,
689 			  struct netlink_ext_ack *extack)
690 {
691 	struct ets_sched *q = qdisc_priv(sch);
692 	int err;
693 
694 	if (!opt)
695 		return -EINVAL;
696 
697 	err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
698 	if (err)
699 		return err;
700 
701 	INIT_LIST_HEAD(&q->active);
702 	return ets_qdisc_change(sch, opt, extack);
703 }
704 
705 static void ets_qdisc_reset(struct Qdisc *sch)
706 {
707 	struct ets_sched *q = qdisc_priv(sch);
708 	int band;
709 
710 	for (band = q->nstrict; band < q->nbands; band++) {
711 		if (q->classes[band].qdisc->q.qlen)
712 			list_del(&q->classes[band].alist);
713 	}
714 	for (band = 0; band < q->nbands; band++)
715 		qdisc_reset(q->classes[band].qdisc);
716 	sch->qstats.backlog = 0;
717 	sch->q.qlen = 0;
718 }
719 
720 static void ets_qdisc_destroy(struct Qdisc *sch)
721 {
722 	struct ets_sched *q = qdisc_priv(sch);
723 	int band;
724 
725 	ets_offload_destroy(sch);
726 	tcf_block_put(q->block);
727 	for (band = 0; band < q->nbands; band++)
728 		qdisc_put(q->classes[band].qdisc);
729 }
730 
731 static int ets_qdisc_dump(struct Qdisc *sch, struct sk_buff *skb)
732 {
733 	struct ets_sched *q = qdisc_priv(sch);
734 	struct nlattr *opts;
735 	struct nlattr *nest;
736 	int band;
737 	int prio;
738 	int err;
739 
740 	err = ets_offload_dump(sch);
741 	if (err)
742 		return err;
743 
744 	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
745 	if (!opts)
746 		goto nla_err;
747 
748 	if (nla_put_u8(skb, TCA_ETS_NBANDS, q->nbands))
749 		goto nla_err;
750 
751 	if (q->nstrict &&
752 	    nla_put_u8(skb, TCA_ETS_NSTRICT, q->nstrict))
753 		goto nla_err;
754 
755 	if (q->nbands > q->nstrict) {
756 		nest = nla_nest_start(skb, TCA_ETS_QUANTA);
757 		if (!nest)
758 			goto nla_err;
759 
760 		for (band = q->nstrict; band < q->nbands; band++) {
761 			if (nla_put_u32(skb, TCA_ETS_QUANTA_BAND,
762 					q->classes[band].quantum))
763 				goto nla_err;
764 		}
765 
766 		nla_nest_end(skb, nest);
767 	}
768 
769 	nest = nla_nest_start(skb, TCA_ETS_PRIOMAP);
770 	if (!nest)
771 		goto nla_err;
772 
773 	for (prio = 0; prio <= TC_PRIO_MAX; prio++) {
774 		if (nla_put_u8(skb, TCA_ETS_PRIOMAP_BAND, q->prio2band[prio]))
775 			goto nla_err;
776 	}
777 
778 	nla_nest_end(skb, nest);
779 
780 	return nla_nest_end(skb, opts);
781 
782 nla_err:
783 	nla_nest_cancel(skb, opts);
784 	return -EMSGSIZE;
785 }
786 
787 static const struct Qdisc_class_ops ets_class_ops = {
788 	.change		= ets_class_change,
789 	.graft		= ets_class_graft,
790 	.leaf		= ets_class_leaf,
791 	.find		= ets_class_find,
792 	.qlen_notify	= ets_class_qlen_notify,
793 	.dump		= ets_class_dump,
794 	.dump_stats	= ets_class_dump_stats,
795 	.walk		= ets_qdisc_walk,
796 	.tcf_block	= ets_qdisc_tcf_block,
797 	.bind_tcf	= ets_qdisc_bind_tcf,
798 	.unbind_tcf	= ets_qdisc_unbind_tcf,
799 };
800 
801 static struct Qdisc_ops ets_qdisc_ops __read_mostly = {
802 	.cl_ops		= &ets_class_ops,
803 	.id		= "ets",
804 	.priv_size	= sizeof(struct ets_sched),
805 	.enqueue	= ets_qdisc_enqueue,
806 	.dequeue	= ets_qdisc_dequeue,
807 	.peek		= qdisc_peek_dequeued,
808 	.change		= ets_qdisc_change,
809 	.init		= ets_qdisc_init,
810 	.reset		= ets_qdisc_reset,
811 	.destroy	= ets_qdisc_destroy,
812 	.dump		= ets_qdisc_dump,
813 	.owner		= THIS_MODULE,
814 };
815 
816 static int __init ets_init(void)
817 {
818 	return register_qdisc(&ets_qdisc_ops);
819 }
820 
821 static void __exit ets_exit(void)
822 {
823 	unregister_qdisc(&ets_qdisc_ops);
824 }
825 
826 module_init(ets_init);
827 module_exit(ets_exit);
828 MODULE_LICENSE("GPL");
829