xref: /openbmc/linux/net/sched/sch_gred.c (revision f9a82c48)
1 /*
2  * net/sched/sch_gred.c	Generic Random Early Detection queue.
3  *
4  *
5  *              This program is free software; you can redistribute it and/or
6  *              modify it under the terms of the GNU General Public License
7  *              as published by the Free Software Foundation; either version
8  *              2 of the License, or (at your option) any later version.
9  *
10  * Authors:    J Hadi Salim (hadi@cyberus.ca) 1998-2002
11  *
12  *             991129: -  Bug fix with grio mode
13  *		       - a better sing. AvgQ mode with Grio(WRED)
14  *		       - A finer grained VQ dequeue based on sugestion
15  *		         from Ren Liu
16  *		       - More error checks
17  *
18  *  For all the glorious comments look at include/net/red.h
19  */
20 
21 #include <linux/slab.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
24 #include <linux/kernel.h>
25 #include <linux/skbuff.h>
26 #include <net/pkt_cls.h>
27 #include <net/pkt_sched.h>
28 #include <net/red.h>
29 
30 #define GRED_DEF_PRIO (MAX_DPs / 2)
31 #define GRED_VQ_MASK (MAX_DPs - 1)
32 
33 #define GRED_VQ_RED_FLAGS	(TC_RED_ECN | TC_RED_HARDDROP)
34 
35 struct gred_sched_data;
36 struct gred_sched;
37 
38 struct gred_sched_data {
39 	u32		limit;		/* HARD maximal queue length	*/
40 	u32		DP;		/* the drop parameters */
41 	u32		red_flags;	/* virtualQ version of red_flags */
42 	u64		bytesin;	/* bytes seen on virtualQ so far*/
43 	u32		packetsin;	/* packets seen on virtualQ so far*/
44 	u32		backlog;	/* bytes on the virtualQ */
45 	u8		prio;		/* the prio of this vq */
46 
47 	struct red_parms parms;
48 	struct red_vars  vars;
49 	struct red_stats stats;
50 };
51 
52 enum {
53 	GRED_WRED_MODE = 1,
54 	GRED_RIO_MODE,
55 };
56 
57 struct gred_sched {
58 	struct gred_sched_data *tab[MAX_DPs];
59 	unsigned long	flags;
60 	u32		red_flags;
61 	u32 		DPs;
62 	u32 		def;
63 	struct red_vars wred_set;
64 };
65 
66 static inline int gred_wred_mode(struct gred_sched *table)
67 {
68 	return test_bit(GRED_WRED_MODE, &table->flags);
69 }
70 
71 static inline void gred_enable_wred_mode(struct gred_sched *table)
72 {
73 	__set_bit(GRED_WRED_MODE, &table->flags);
74 }
75 
76 static inline void gred_disable_wred_mode(struct gred_sched *table)
77 {
78 	__clear_bit(GRED_WRED_MODE, &table->flags);
79 }
80 
81 static inline int gred_rio_mode(struct gred_sched *table)
82 {
83 	return test_bit(GRED_RIO_MODE, &table->flags);
84 }
85 
86 static inline void gred_enable_rio_mode(struct gred_sched *table)
87 {
88 	__set_bit(GRED_RIO_MODE, &table->flags);
89 }
90 
91 static inline void gred_disable_rio_mode(struct gred_sched *table)
92 {
93 	__clear_bit(GRED_RIO_MODE, &table->flags);
94 }
95 
96 static inline int gred_wred_mode_check(struct Qdisc *sch)
97 {
98 	struct gred_sched *table = qdisc_priv(sch);
99 	int i;
100 
101 	/* Really ugly O(n^2) but shouldn't be necessary too frequent. */
102 	for (i = 0; i < table->DPs; i++) {
103 		struct gred_sched_data *q = table->tab[i];
104 		int n;
105 
106 		if (q == NULL)
107 			continue;
108 
109 		for (n = i + 1; n < table->DPs; n++)
110 			if (table->tab[n] && table->tab[n]->prio == q->prio)
111 				return 1;
112 	}
113 
114 	return 0;
115 }
116 
117 static inline unsigned int gred_backlog(struct gred_sched *table,
118 					struct gred_sched_data *q,
119 					struct Qdisc *sch)
120 {
121 	if (gred_wred_mode(table))
122 		return sch->qstats.backlog;
123 	else
124 		return q->backlog;
125 }
126 
127 static inline u16 tc_index_to_dp(struct sk_buff *skb)
128 {
129 	return skb->tc_index & GRED_VQ_MASK;
130 }
131 
132 static inline void gred_load_wred_set(const struct gred_sched *table,
133 				      struct gred_sched_data *q)
134 {
135 	q->vars.qavg = table->wred_set.qavg;
136 	q->vars.qidlestart = table->wred_set.qidlestart;
137 }
138 
139 static inline void gred_store_wred_set(struct gred_sched *table,
140 				       struct gred_sched_data *q)
141 {
142 	table->wred_set.qavg = q->vars.qavg;
143 	table->wred_set.qidlestart = q->vars.qidlestart;
144 }
145 
146 static int gred_use_ecn(struct gred_sched_data *q)
147 {
148 	return q->red_flags & TC_RED_ECN;
149 }
150 
151 static int gred_use_harddrop(struct gred_sched_data *q)
152 {
153 	return q->red_flags & TC_RED_HARDDROP;
154 }
155 
156 static bool gred_per_vq_red_flags_used(struct gred_sched *table)
157 {
158 	unsigned int i;
159 
160 	/* Local per-vq flags couldn't have been set unless global are 0 */
161 	if (table->red_flags)
162 		return false;
163 	for (i = 0; i < MAX_DPs; i++)
164 		if (table->tab[i] && table->tab[i]->red_flags)
165 			return true;
166 	return false;
167 }
168 
169 static int gred_enqueue(struct sk_buff *skb, struct Qdisc *sch,
170 			struct sk_buff **to_free)
171 {
172 	struct gred_sched_data *q = NULL;
173 	struct gred_sched *t = qdisc_priv(sch);
174 	unsigned long qavg = 0;
175 	u16 dp = tc_index_to_dp(skb);
176 
177 	if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
178 		dp = t->def;
179 
180 		q = t->tab[dp];
181 		if (!q) {
182 			/* Pass through packets not assigned to a DP
183 			 * if no default DP has been configured. This
184 			 * allows for DP flows to be left untouched.
185 			 */
186 			if (likely(sch->qstats.backlog + qdisc_pkt_len(skb) <=
187 					sch->limit))
188 				return qdisc_enqueue_tail(skb, sch);
189 			else
190 				goto drop;
191 		}
192 
193 		/* fix tc_index? --could be controversial but needed for
194 		   requeueing */
195 		skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
196 	}
197 
198 	/* sum up all the qaves of prios < ours to get the new qave */
199 	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
200 		int i;
201 
202 		for (i = 0; i < t->DPs; i++) {
203 			if (t->tab[i] && t->tab[i]->prio < q->prio &&
204 			    !red_is_idling(&t->tab[i]->vars))
205 				qavg += t->tab[i]->vars.qavg;
206 		}
207 
208 	}
209 
210 	q->packetsin++;
211 	q->bytesin += qdisc_pkt_len(skb);
212 
213 	if (gred_wred_mode(t))
214 		gred_load_wred_set(t, q);
215 
216 	q->vars.qavg = red_calc_qavg(&q->parms,
217 				     &q->vars,
218 				     gred_backlog(t, q, sch));
219 
220 	if (red_is_idling(&q->vars))
221 		red_end_of_idle_period(&q->vars);
222 
223 	if (gred_wred_mode(t))
224 		gred_store_wred_set(t, q);
225 
226 	switch (red_action(&q->parms, &q->vars, q->vars.qavg + qavg)) {
227 	case RED_DONT_MARK:
228 		break;
229 
230 	case RED_PROB_MARK:
231 		qdisc_qstats_overlimit(sch);
232 		if (!gred_use_ecn(q) || !INET_ECN_set_ce(skb)) {
233 			q->stats.prob_drop++;
234 			goto congestion_drop;
235 		}
236 
237 		q->stats.prob_mark++;
238 		break;
239 
240 	case RED_HARD_MARK:
241 		qdisc_qstats_overlimit(sch);
242 		if (gred_use_harddrop(q) || !gred_use_ecn(q) ||
243 		    !INET_ECN_set_ce(skb)) {
244 			q->stats.forced_drop++;
245 			goto congestion_drop;
246 		}
247 		q->stats.forced_mark++;
248 		break;
249 	}
250 
251 	if (gred_backlog(t, q, sch) + qdisc_pkt_len(skb) <= q->limit) {
252 		q->backlog += qdisc_pkt_len(skb);
253 		return qdisc_enqueue_tail(skb, sch);
254 	}
255 
256 	q->stats.pdrop++;
257 drop:
258 	return qdisc_drop(skb, sch, to_free);
259 
260 congestion_drop:
261 	qdisc_drop(skb, sch, to_free);
262 	return NET_XMIT_CN;
263 }
264 
265 static struct sk_buff *gred_dequeue(struct Qdisc *sch)
266 {
267 	struct sk_buff *skb;
268 	struct gred_sched *t = qdisc_priv(sch);
269 
270 	skb = qdisc_dequeue_head(sch);
271 
272 	if (skb) {
273 		struct gred_sched_data *q;
274 		u16 dp = tc_index_to_dp(skb);
275 
276 		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
277 			net_warn_ratelimited("GRED: Unable to relocate VQ 0x%x after dequeue, screwing up backlog\n",
278 					     tc_index_to_dp(skb));
279 		} else {
280 			q->backlog -= qdisc_pkt_len(skb);
281 
282 			if (gred_wred_mode(t)) {
283 				if (!sch->qstats.backlog)
284 					red_start_of_idle_period(&t->wred_set);
285 			} else {
286 				if (!q->backlog)
287 					red_start_of_idle_period(&q->vars);
288 			}
289 		}
290 
291 		return skb;
292 	}
293 
294 	return NULL;
295 }
296 
297 static void gred_reset(struct Qdisc *sch)
298 {
299 	int i;
300 	struct gred_sched *t = qdisc_priv(sch);
301 
302 	qdisc_reset_queue(sch);
303 
304 	for (i = 0; i < t->DPs; i++) {
305 		struct gred_sched_data *q = t->tab[i];
306 
307 		if (!q)
308 			continue;
309 
310 		red_restart(&q->vars);
311 		q->backlog = 0;
312 	}
313 }
314 
315 static void gred_offload(struct Qdisc *sch, enum tc_gred_command command)
316 {
317 	struct gred_sched *table = qdisc_priv(sch);
318 	struct net_device *dev = qdisc_dev(sch);
319 	struct tc_gred_qopt_offload opt = {
320 		.command	= command,
321 		.handle		= sch->handle,
322 		.parent		= sch->parent,
323 	};
324 
325 	if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
326 		return;
327 
328 	if (command == TC_GRED_REPLACE) {
329 		unsigned int i;
330 
331 		opt.set.grio_on = gred_rio_mode(table);
332 		opt.set.wred_on = gred_wred_mode(table);
333 		opt.set.dp_cnt = table->DPs;
334 		opt.set.dp_def = table->def;
335 
336 		for (i = 0; i < table->DPs; i++) {
337 			struct gred_sched_data *q = table->tab[i];
338 
339 			if (!q)
340 				continue;
341 			opt.set.tab[i].present = true;
342 			opt.set.tab[i].limit = q->limit;
343 			opt.set.tab[i].prio = q->prio;
344 			opt.set.tab[i].min = q->parms.qth_min >> q->parms.Wlog;
345 			opt.set.tab[i].max = q->parms.qth_max >> q->parms.Wlog;
346 			opt.set.tab[i].is_ecn = gred_use_ecn(q);
347 			opt.set.tab[i].is_harddrop = gred_use_harddrop(q);
348 			opt.set.tab[i].probability = q->parms.max_P;
349 			opt.set.tab[i].backlog = &q->backlog;
350 		}
351 		opt.set.qstats = &sch->qstats;
352 	}
353 
354 	dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_GRED, &opt);
355 }
356 
357 static int gred_offload_dump_stats(struct Qdisc *sch)
358 {
359 	struct gred_sched *table = qdisc_priv(sch);
360 	struct tc_gred_qopt_offload *hw_stats;
361 	unsigned int i;
362 	int ret;
363 
364 	hw_stats = kzalloc(sizeof(*hw_stats), GFP_KERNEL);
365 	if (!hw_stats)
366 		return -ENOMEM;
367 
368 	hw_stats->command = TC_GRED_STATS;
369 	hw_stats->handle = sch->handle;
370 	hw_stats->parent = sch->parent;
371 
372 	for (i = 0; i < MAX_DPs; i++)
373 		if (table->tab[i])
374 			hw_stats->stats.xstats[i] = &table->tab[i]->stats;
375 
376 	ret = qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_GRED, hw_stats);
377 	/* Even if driver returns failure adjust the stats - in case offload
378 	 * ended but driver still wants to adjust the values.
379 	 */
380 	for (i = 0; i < MAX_DPs; i++) {
381 		if (!table->tab[i])
382 			continue;
383 		table->tab[i]->packetsin += hw_stats->stats.bstats[i].packets;
384 		table->tab[i]->bytesin += hw_stats->stats.bstats[i].bytes;
385 		table->tab[i]->backlog += hw_stats->stats.qstats[i].backlog;
386 
387 		_bstats_update(&sch->bstats,
388 			       hw_stats->stats.bstats[i].bytes,
389 			       hw_stats->stats.bstats[i].packets);
390 		sch->qstats.qlen += hw_stats->stats.qstats[i].qlen;
391 		sch->qstats.backlog += hw_stats->stats.qstats[i].backlog;
392 		sch->qstats.drops += hw_stats->stats.qstats[i].drops;
393 		sch->qstats.requeues += hw_stats->stats.qstats[i].requeues;
394 		sch->qstats.overlimits += hw_stats->stats.qstats[i].overlimits;
395 	}
396 
397 	kfree(hw_stats);
398 	return ret;
399 }
400 
401 static inline void gred_destroy_vq(struct gred_sched_data *q)
402 {
403 	kfree(q);
404 }
405 
406 static int gred_change_table_def(struct Qdisc *sch, struct nlattr *dps,
407 				 struct netlink_ext_ack *extack)
408 {
409 	struct gred_sched *table = qdisc_priv(sch);
410 	struct tc_gred_sopt *sopt;
411 	bool red_flags_changed;
412 	int i;
413 
414 	if (!dps)
415 		return -EINVAL;
416 
417 	sopt = nla_data(dps);
418 
419 	if (sopt->DPs > MAX_DPs) {
420 		NL_SET_ERR_MSG_MOD(extack, "number of virtual queues too high");
421 		return -EINVAL;
422 	}
423 	if (sopt->DPs == 0) {
424 		NL_SET_ERR_MSG_MOD(extack,
425 				   "number of virtual queues can't be 0");
426 		return -EINVAL;
427 	}
428 	if (sopt->def_DP >= sopt->DPs) {
429 		NL_SET_ERR_MSG_MOD(extack, "default virtual queue above virtual queue count");
430 		return -EINVAL;
431 	}
432 	if (sopt->flags && gred_per_vq_red_flags_used(table)) {
433 		NL_SET_ERR_MSG_MOD(extack, "can't set per-Qdisc RED flags when per-virtual queue flags are used");
434 		return -EINVAL;
435 	}
436 
437 	sch_tree_lock(sch);
438 	table->DPs = sopt->DPs;
439 	table->def = sopt->def_DP;
440 	red_flags_changed = table->red_flags != sopt->flags;
441 	table->red_flags = sopt->flags;
442 
443 	/*
444 	 * Every entry point to GRED is synchronized with the above code
445 	 * and the DP is checked against DPs, i.e. shadowed VQs can no
446 	 * longer be found so we can unlock right here.
447 	 */
448 	sch_tree_unlock(sch);
449 
450 	if (sopt->grio) {
451 		gred_enable_rio_mode(table);
452 		gred_disable_wred_mode(table);
453 		if (gred_wred_mode_check(sch))
454 			gred_enable_wred_mode(table);
455 	} else {
456 		gred_disable_rio_mode(table);
457 		gred_disable_wred_mode(table);
458 	}
459 
460 	if (red_flags_changed)
461 		for (i = 0; i < table->DPs; i++)
462 			if (table->tab[i])
463 				table->tab[i]->red_flags =
464 					table->red_flags & GRED_VQ_RED_FLAGS;
465 
466 	for (i = table->DPs; i < MAX_DPs; i++) {
467 		if (table->tab[i]) {
468 			pr_warn("GRED: Warning: Destroying shadowed VQ 0x%x\n",
469 				i);
470 			gred_destroy_vq(table->tab[i]);
471 			table->tab[i] = NULL;
472 		}
473 	}
474 
475 	gred_offload(sch, TC_GRED_REPLACE);
476 	return 0;
477 }
478 
479 static inline int gred_change_vq(struct Qdisc *sch, int dp,
480 				 struct tc_gred_qopt *ctl, int prio,
481 				 u8 *stab, u32 max_P,
482 				 struct gred_sched_data **prealloc,
483 				 struct netlink_ext_ack *extack)
484 {
485 	struct gred_sched *table = qdisc_priv(sch);
486 	struct gred_sched_data *q = table->tab[dp];
487 
488 	if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog)) {
489 		NL_SET_ERR_MSG_MOD(extack, "invalid RED parameters");
490 		return -EINVAL;
491 	}
492 
493 	if (!q) {
494 		table->tab[dp] = q = *prealloc;
495 		*prealloc = NULL;
496 		if (!q)
497 			return -ENOMEM;
498 		q->red_flags = table->red_flags & GRED_VQ_RED_FLAGS;
499 	}
500 
501 	q->DP = dp;
502 	q->prio = prio;
503 	if (ctl->limit > sch->limit)
504 		q->limit = sch->limit;
505 	else
506 		q->limit = ctl->limit;
507 
508 	if (q->backlog == 0)
509 		red_end_of_idle_period(&q->vars);
510 
511 	red_set_parms(&q->parms,
512 		      ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
513 		      ctl->Scell_log, stab, max_P);
514 	red_set_vars(&q->vars);
515 	return 0;
516 }
517 
518 static const struct nla_policy gred_vq_policy[TCA_GRED_VQ_MAX + 1] = {
519 	[TCA_GRED_VQ_DP]	= { .type = NLA_U32 },
520 	[TCA_GRED_VQ_FLAGS]	= { .type = NLA_U32 },
521 };
522 
523 static const struct nla_policy gred_vqe_policy[TCA_GRED_VQ_ENTRY_MAX + 1] = {
524 	[TCA_GRED_VQ_ENTRY]	= { .type = NLA_NESTED },
525 };
526 
527 static const struct nla_policy gred_policy[TCA_GRED_MAX + 1] = {
528 	[TCA_GRED_PARMS]	= { .len = sizeof(struct tc_gred_qopt) },
529 	[TCA_GRED_STAB]		= { .len = 256 },
530 	[TCA_GRED_DPS]		= { .len = sizeof(struct tc_gred_sopt) },
531 	[TCA_GRED_MAX_P]	= { .type = NLA_U32 },
532 	[TCA_GRED_LIMIT]	= { .type = NLA_U32 },
533 	[TCA_GRED_VQ_LIST]	= { .type = NLA_NESTED },
534 };
535 
536 static void gred_vq_apply(struct gred_sched *table, const struct nlattr *entry)
537 {
538 	struct nlattr *tb[TCA_GRED_VQ_MAX + 1];
539 	u32 dp;
540 
541 	nla_parse_nested(tb, TCA_GRED_VQ_MAX, entry, gred_vq_policy, NULL);
542 
543 	dp = nla_get_u32(tb[TCA_GRED_VQ_DP]);
544 
545 	if (tb[TCA_GRED_VQ_FLAGS])
546 		table->tab[dp]->red_flags = nla_get_u32(tb[TCA_GRED_VQ_FLAGS]);
547 }
548 
549 static void gred_vqs_apply(struct gred_sched *table, struct nlattr *vqs)
550 {
551 	const struct nlattr *attr;
552 	int rem;
553 
554 	nla_for_each_nested(attr, vqs, rem) {
555 		switch (nla_type(attr)) {
556 		case TCA_GRED_VQ_ENTRY:
557 			gred_vq_apply(table, attr);
558 			break;
559 		}
560 	}
561 }
562 
563 static int gred_vq_validate(struct gred_sched *table, u32 cdp,
564 			    const struct nlattr *entry,
565 			    struct netlink_ext_ack *extack)
566 {
567 	struct nlattr *tb[TCA_GRED_VQ_MAX + 1];
568 	int err;
569 	u32 dp;
570 
571 	err = nla_parse_nested(tb, TCA_GRED_VQ_MAX, entry, gred_vq_policy,
572 			       extack);
573 	if (err < 0)
574 		return err;
575 
576 	if (!tb[TCA_GRED_VQ_DP]) {
577 		NL_SET_ERR_MSG_MOD(extack, "Virtual queue with no index specified");
578 		return -EINVAL;
579 	}
580 	dp = nla_get_u32(tb[TCA_GRED_VQ_DP]);
581 	if (dp >= table->DPs) {
582 		NL_SET_ERR_MSG_MOD(extack, "Virtual queue with index out of bounds");
583 		return -EINVAL;
584 	}
585 	if (dp != cdp && !table->tab[dp]) {
586 		NL_SET_ERR_MSG_MOD(extack, "Virtual queue not yet instantiated");
587 		return -EINVAL;
588 	}
589 
590 	if (tb[TCA_GRED_VQ_FLAGS]) {
591 		u32 red_flags = nla_get_u32(tb[TCA_GRED_VQ_FLAGS]);
592 
593 		if (table->red_flags && table->red_flags != red_flags) {
594 			NL_SET_ERR_MSG_MOD(extack, "can't change per-virtual queue RED flags when per-Qdisc flags are used");
595 			return -EINVAL;
596 		}
597 		if (red_flags & ~GRED_VQ_RED_FLAGS) {
598 			NL_SET_ERR_MSG_MOD(extack,
599 					   "invalid RED flags specified");
600 			return -EINVAL;
601 		}
602 	}
603 
604 	return 0;
605 }
606 
607 static int gred_vqs_validate(struct gred_sched *table, u32 cdp,
608 			     struct nlattr *vqs, struct netlink_ext_ack *extack)
609 {
610 	const struct nlattr *attr;
611 	int rem, err;
612 
613 	err = nla_validate_nested(vqs, TCA_GRED_VQ_ENTRY_MAX,
614 				  gred_vqe_policy, extack);
615 	if (err < 0)
616 		return err;
617 
618 	nla_for_each_nested(attr, vqs, rem) {
619 		switch (nla_type(attr)) {
620 		case TCA_GRED_VQ_ENTRY:
621 			err = gred_vq_validate(table, cdp, attr, extack);
622 			if (err)
623 				return err;
624 			break;
625 		default:
626 			NL_SET_ERR_MSG_MOD(extack, "GRED_VQ_LIST can contain only entry attributes");
627 			return -EINVAL;
628 		}
629 	}
630 
631 	if (rem > 0) {
632 		NL_SET_ERR_MSG_MOD(extack, "Trailing data after parsing virtual queue list");
633 		return -EINVAL;
634 	}
635 
636 	return 0;
637 }
638 
639 static int gred_change(struct Qdisc *sch, struct nlattr *opt,
640 		       struct netlink_ext_ack *extack)
641 {
642 	struct gred_sched *table = qdisc_priv(sch);
643 	struct tc_gred_qopt *ctl;
644 	struct nlattr *tb[TCA_GRED_MAX + 1];
645 	int err, prio = GRED_DEF_PRIO;
646 	u8 *stab;
647 	u32 max_P;
648 	struct gred_sched_data *prealloc;
649 
650 	if (opt == NULL)
651 		return -EINVAL;
652 
653 	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy, extack);
654 	if (err < 0)
655 		return err;
656 
657 	if (tb[TCA_GRED_PARMS] == NULL && tb[TCA_GRED_STAB] == NULL) {
658 		if (tb[TCA_GRED_LIMIT] != NULL)
659 			sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
660 		return gred_change_table_def(sch, tb[TCA_GRED_DPS], extack);
661 	}
662 
663 	if (tb[TCA_GRED_PARMS] == NULL ||
664 	    tb[TCA_GRED_STAB] == NULL ||
665 	    tb[TCA_GRED_LIMIT] != NULL) {
666 		NL_SET_ERR_MSG_MOD(extack, "can't configure Qdisc and virtual queue at the same time");
667 		return -EINVAL;
668 	}
669 
670 	max_P = tb[TCA_GRED_MAX_P] ? nla_get_u32(tb[TCA_GRED_MAX_P]) : 0;
671 
672 	ctl = nla_data(tb[TCA_GRED_PARMS]);
673 	stab = nla_data(tb[TCA_GRED_STAB]);
674 
675 	if (ctl->DP >= table->DPs) {
676 		NL_SET_ERR_MSG_MOD(extack, "virtual queue index above virtual queue count");
677 		return -EINVAL;
678 	}
679 
680 	if (tb[TCA_GRED_VQ_LIST]) {
681 		err = gred_vqs_validate(table, ctl->DP, tb[TCA_GRED_VQ_LIST],
682 					extack);
683 		if (err)
684 			return err;
685 	}
686 
687 	if (gred_rio_mode(table)) {
688 		if (ctl->prio == 0) {
689 			int def_prio = GRED_DEF_PRIO;
690 
691 			if (table->tab[table->def])
692 				def_prio = table->tab[table->def]->prio;
693 
694 			printk(KERN_DEBUG "GRED: DP %u does not have a prio "
695 			       "setting default to %d\n", ctl->DP, def_prio);
696 
697 			prio = def_prio;
698 		} else
699 			prio = ctl->prio;
700 	}
701 
702 	prealloc = kzalloc(sizeof(*prealloc), GFP_KERNEL);
703 	sch_tree_lock(sch);
704 
705 	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab, max_P, &prealloc,
706 			     extack);
707 	if (err < 0)
708 		goto err_unlock_free;
709 
710 	if (tb[TCA_GRED_VQ_LIST])
711 		gred_vqs_apply(table, tb[TCA_GRED_VQ_LIST]);
712 
713 	if (gred_rio_mode(table)) {
714 		gred_disable_wred_mode(table);
715 		if (gred_wred_mode_check(sch))
716 			gred_enable_wred_mode(table);
717 	}
718 
719 	sch_tree_unlock(sch);
720 	kfree(prealloc);
721 
722 	gred_offload(sch, TC_GRED_REPLACE);
723 	return 0;
724 
725 err_unlock_free:
726 	sch_tree_unlock(sch);
727 	kfree(prealloc);
728 	return err;
729 }
730 
731 static int gred_init(struct Qdisc *sch, struct nlattr *opt,
732 		     struct netlink_ext_ack *extack)
733 {
734 	struct nlattr *tb[TCA_GRED_MAX + 1];
735 	int err;
736 
737 	if (!opt)
738 		return -EINVAL;
739 
740 	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy, extack);
741 	if (err < 0)
742 		return err;
743 
744 	if (tb[TCA_GRED_PARMS] || tb[TCA_GRED_STAB]) {
745 		NL_SET_ERR_MSG_MOD(extack,
746 				   "virtual queue configuration can't be specified at initialization time");
747 		return -EINVAL;
748 	}
749 
750 	if (tb[TCA_GRED_LIMIT])
751 		sch->limit = nla_get_u32(tb[TCA_GRED_LIMIT]);
752 	else
753 		sch->limit = qdisc_dev(sch)->tx_queue_len
754 		             * psched_mtu(qdisc_dev(sch));
755 
756 	return gred_change_table_def(sch, tb[TCA_GRED_DPS], extack);
757 }
758 
759 static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
760 {
761 	struct gred_sched *table = qdisc_priv(sch);
762 	struct nlattr *parms, *vqs, *opts = NULL;
763 	int i;
764 	u32 max_p[MAX_DPs];
765 	struct tc_gred_sopt sopt = {
766 		.DPs	= table->DPs,
767 		.def_DP	= table->def,
768 		.grio	= gred_rio_mode(table),
769 		.flags	= table->red_flags,
770 	};
771 
772 	if (gred_offload_dump_stats(sch))
773 		goto nla_put_failure;
774 
775 	opts = nla_nest_start(skb, TCA_OPTIONS);
776 	if (opts == NULL)
777 		goto nla_put_failure;
778 	if (nla_put(skb, TCA_GRED_DPS, sizeof(sopt), &sopt))
779 		goto nla_put_failure;
780 
781 	for (i = 0; i < MAX_DPs; i++) {
782 		struct gred_sched_data *q = table->tab[i];
783 
784 		max_p[i] = q ? q->parms.max_P : 0;
785 	}
786 	if (nla_put(skb, TCA_GRED_MAX_P, sizeof(max_p), max_p))
787 		goto nla_put_failure;
788 
789 	if (nla_put_u32(skb, TCA_GRED_LIMIT, sch->limit))
790 		goto nla_put_failure;
791 
792 	/* Old style all-in-one dump of VQs */
793 	parms = nla_nest_start(skb, TCA_GRED_PARMS);
794 	if (parms == NULL)
795 		goto nla_put_failure;
796 
797 	for (i = 0; i < MAX_DPs; i++) {
798 		struct gred_sched_data *q = table->tab[i];
799 		struct tc_gred_qopt opt;
800 		unsigned long qavg;
801 
802 		memset(&opt, 0, sizeof(opt));
803 
804 		if (!q) {
805 			/* hack -- fix at some point with proper message
806 			   This is how we indicate to tc that there is no VQ
807 			   at this DP */
808 
809 			opt.DP = MAX_DPs + i;
810 			goto append_opt;
811 		}
812 
813 		opt.limit	= q->limit;
814 		opt.DP		= q->DP;
815 		opt.backlog	= gred_backlog(table, q, sch);
816 		opt.prio	= q->prio;
817 		opt.qth_min	= q->parms.qth_min >> q->parms.Wlog;
818 		opt.qth_max	= q->parms.qth_max >> q->parms.Wlog;
819 		opt.Wlog	= q->parms.Wlog;
820 		opt.Plog	= q->parms.Plog;
821 		opt.Scell_log	= q->parms.Scell_log;
822 		opt.other	= q->stats.other;
823 		opt.early	= q->stats.prob_drop;
824 		opt.forced	= q->stats.forced_drop;
825 		opt.pdrop	= q->stats.pdrop;
826 		opt.packets	= q->packetsin;
827 		opt.bytesin	= q->bytesin;
828 
829 		if (gred_wred_mode(table))
830 			gred_load_wred_set(table, q);
831 
832 		qavg = red_calc_qavg(&q->parms, &q->vars,
833 				     q->vars.qavg >> q->parms.Wlog);
834 		opt.qave = qavg >> q->parms.Wlog;
835 
836 append_opt:
837 		if (nla_append(skb, sizeof(opt), &opt) < 0)
838 			goto nla_put_failure;
839 	}
840 
841 	nla_nest_end(skb, parms);
842 
843 	/* Dump the VQs again, in more structured way */
844 	vqs = nla_nest_start(skb, TCA_GRED_VQ_LIST);
845 	if (!vqs)
846 		goto nla_put_failure;
847 
848 	for (i = 0; i < MAX_DPs; i++) {
849 		struct gred_sched_data *q = table->tab[i];
850 		struct nlattr *vq;
851 
852 		if (!q)
853 			continue;
854 
855 		vq = nla_nest_start(skb, TCA_GRED_VQ_ENTRY);
856 		if (!vq)
857 			goto nla_put_failure;
858 
859 		if (nla_put_u32(skb, TCA_GRED_VQ_DP, q->DP))
860 			goto nla_put_failure;
861 
862 		if (nla_put_u32(skb, TCA_GRED_VQ_FLAGS, q->red_flags))
863 			goto nla_put_failure;
864 
865 		/* Stats */
866 		if (nla_put_u64_64bit(skb, TCA_GRED_VQ_STAT_BYTES, q->bytesin,
867 				      TCA_GRED_VQ_PAD))
868 			goto nla_put_failure;
869 		if (nla_put_u32(skb, TCA_GRED_VQ_STAT_PACKETS, q->packetsin))
870 			goto nla_put_failure;
871 		if (nla_put_u32(skb, TCA_GRED_VQ_STAT_BACKLOG,
872 				gred_backlog(table, q, sch)))
873 			goto nla_put_failure;
874 		if (nla_put_u32(skb, TCA_GRED_VQ_STAT_PROB_DROP,
875 				q->stats.prob_drop))
876 			goto nla_put_failure;
877 		if (nla_put_u32(skb, TCA_GRED_VQ_STAT_PROB_MARK,
878 				q->stats.prob_mark))
879 			goto nla_put_failure;
880 		if (nla_put_u32(skb, TCA_GRED_VQ_STAT_FORCED_DROP,
881 				q->stats.forced_drop))
882 			goto nla_put_failure;
883 		if (nla_put_u32(skb, TCA_GRED_VQ_STAT_FORCED_MARK,
884 				q->stats.forced_mark))
885 			goto nla_put_failure;
886 		if (nla_put_u32(skb, TCA_GRED_VQ_STAT_PDROP, q->stats.pdrop))
887 			goto nla_put_failure;
888 		if (nla_put_u32(skb, TCA_GRED_VQ_STAT_OTHER, q->stats.other))
889 			goto nla_put_failure;
890 
891 		nla_nest_end(skb, vq);
892 	}
893 	nla_nest_end(skb, vqs);
894 
895 	return nla_nest_end(skb, opts);
896 
897 nla_put_failure:
898 	nla_nest_cancel(skb, opts);
899 	return -EMSGSIZE;
900 }
901 
902 static void gred_destroy(struct Qdisc *sch)
903 {
904 	struct gred_sched *table = qdisc_priv(sch);
905 	int i;
906 
907 	for (i = 0; i < table->DPs; i++) {
908 		if (table->tab[i])
909 			gred_destroy_vq(table->tab[i]);
910 	}
911 	gred_offload(sch, TC_GRED_DESTROY);
912 }
913 
914 static struct Qdisc_ops gred_qdisc_ops __read_mostly = {
915 	.id		=	"gred",
916 	.priv_size	=	sizeof(struct gred_sched),
917 	.enqueue	=	gred_enqueue,
918 	.dequeue	=	gred_dequeue,
919 	.peek		=	qdisc_peek_head,
920 	.init		=	gred_init,
921 	.reset		=	gred_reset,
922 	.destroy	=	gred_destroy,
923 	.change		=	gred_change,
924 	.dump		=	gred_dump,
925 	.owner		=	THIS_MODULE,
926 };
927 
928 static int __init gred_module_init(void)
929 {
930 	return register_qdisc(&gred_qdisc_ops);
931 }
932 
933 static void __exit gred_module_exit(void)
934 {
935 	unregister_qdisc(&gred_qdisc_ops);
936 }
937 
938 module_init(gred_module_init)
939 module_exit(gred_module_exit)
940 
941 MODULE_LICENSE("GPL");
942