xref: /openbmc/linux/net/sched/sch_gred.c (revision 54525552) 
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_sched.h>
27 #include <net/red.h>
28 
29 #define GRED_DEF_PRIO (MAX_DPs / 2)
30 #define GRED_VQ_MASK (MAX_DPs - 1)
31 
32 struct gred_sched_data;
33 struct gred_sched;
34 
35 struct gred_sched_data {
36 	u32		limit;		/* HARD maximal queue length	*/
37 	u32      	DP;		/* the drop pramaters */
38 	u32		bytesin;	/* bytes seen on virtualQ so far*/
39 	u32		packetsin;	/* packets seen on virtualQ so far*/
40 	u32		backlog;	/* bytes on the virtualQ */
41 	u8		prio;		/* the prio of this vq */
42 
43 	struct red_parms parms;
44 	struct red_stats stats;
45 };
46 
47 enum {
48 	GRED_WRED_MODE = 1,
49 	GRED_RIO_MODE,
50 };
51 
52 struct gred_sched {
53 	struct gred_sched_data *tab[MAX_DPs];
54 	unsigned long	flags;
55 	u32		red_flags;
56 	u32 		DPs;
57 	u32 		def;
58 	struct red_parms wred_set;
59 };
60 
61 static inline int gred_wred_mode(struct gred_sched *table)
62 {
63 	return test_bit(GRED_WRED_MODE, &table->flags);
64 }
65 
66 static inline void gred_enable_wred_mode(struct gred_sched *table)
67 {
68 	__set_bit(GRED_WRED_MODE, &table->flags);
69 }
70 
71 static inline void gred_disable_wred_mode(struct gred_sched *table)
72 {
73 	__clear_bit(GRED_WRED_MODE, &table->flags);
74 }
75 
76 static inline int gred_rio_mode(struct gred_sched *table)
77 {
78 	return test_bit(GRED_RIO_MODE, &table->flags);
79 }
80 
81 static inline void gred_enable_rio_mode(struct gred_sched *table)
82 {
83 	__set_bit(GRED_RIO_MODE, &table->flags);
84 }
85 
86 static inline void gred_disable_rio_mode(struct gred_sched *table)
87 {
88 	__clear_bit(GRED_RIO_MODE, &table->flags);
89 }
90 
91 static inline int gred_wred_mode_check(struct Qdisc *sch)
92 {
93 	struct gred_sched *table = qdisc_priv(sch);
94 	int i;
95 
96 	/* Really ugly O(n^2) but shouldn't be necessary too frequent. */
97 	for (i = 0; i < table->DPs; i++) {
98 		struct gred_sched_data *q = table->tab[i];
99 		int n;
100 
101 		if (q == NULL)
102 			continue;
103 
104 		for (n = 0; n < table->DPs; n++)
105 			if (table->tab[n] && table->tab[n] != q &&
106 			    table->tab[n]->prio == q->prio)
107 				return 1;
108 	}
109 
110 	return 0;
111 }
112 
113 static inline unsigned int gred_backlog(struct gred_sched *table,
114 					struct gred_sched_data *q,
115 					struct Qdisc *sch)
116 {
117 	if (gred_wred_mode(table))
118 		return sch->qstats.backlog;
119 	else
120 		return q->backlog;
121 }
122 
123 static inline u16 tc_index_to_dp(struct sk_buff *skb)
124 {
125 	return skb->tc_index & GRED_VQ_MASK;
126 }
127 
128 static inline void gred_load_wred_set(struct gred_sched *table,
129 				      struct gred_sched_data *q)
130 {
131 	q->parms.qavg = table->wred_set.qavg;
132 	q->parms.qidlestart = table->wred_set.qidlestart;
133 }
134 
135 static inline void gred_store_wred_set(struct gred_sched *table,
136 				       struct gred_sched_data *q)
137 {
138 	table->wred_set.qavg = q->parms.qavg;
139 }
140 
141 static inline int gred_use_ecn(struct gred_sched *t)
142 {
143 	return t->red_flags & TC_RED_ECN;
144 }
145 
146 static inline int gred_use_harddrop(struct gred_sched *t)
147 {
148 	return t->red_flags & TC_RED_HARDDROP;
149 }
150 
151 static int gred_enqueue(struct sk_buff *skb, struct Qdisc *sch)
152 {
153 	struct gred_sched_data *q = NULL;
154 	struct gred_sched *t = qdisc_priv(sch);
155 	unsigned long qavg = 0;
156 	u16 dp = tc_index_to_dp(skb);
157 
158 	if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
159 		dp = t->def;
160 
161 		q = t->tab[dp];
162 		if (!q) {
163 			/* Pass through packets not assigned to a DP
164 			 * if no default DP has been configured. This
165 			 * allows for DP flows to be left untouched.
166 			 */
167 			if (skb_queue_len(&sch->q) < qdisc_dev(sch)->tx_queue_len)
168 				return qdisc_enqueue_tail(skb, sch);
169 			else
170 				goto drop;
171 		}
172 
173 		/* fix tc_index? --could be controvesial but needed for
174 		   requeueing */
175 		skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
176 	}
177 
178 	/* sum up all the qaves of prios <= to ours to get the new qave */
179 	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
180 		int i;
181 
182 		for (i = 0; i < t->DPs; i++) {
183 			if (t->tab[i] && t->tab[i]->prio < q->prio &&
184 			    !red_is_idling(&t->tab[i]->parms))
185 				qavg += t->tab[i]->parms.qavg;
186 		}
187 
188 	}
189 
190 	q->packetsin++;
191 	q->bytesin += qdisc_pkt_len(skb);
192 
193 	if (gred_wred_mode(t))
194 		gred_load_wred_set(t, q);
195 
196 	q->parms.qavg = red_calc_qavg(&q->parms, gred_backlog(t, q, sch));
197 
198 	if (red_is_idling(&q->parms))
199 		red_end_of_idle_period(&q->parms);
200 
201 	if (gred_wred_mode(t))
202 		gred_store_wred_set(t, q);
203 
204 	switch (red_action(&q->parms, q->parms.qavg + qavg)) {
205 	case RED_DONT_MARK:
206 		break;
207 
208 	case RED_PROB_MARK:
209 		sch->qstats.overlimits++;
210 		if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) {
211 			q->stats.prob_drop++;
212 			goto congestion_drop;
213 		}
214 
215 		q->stats.prob_mark++;
216 		break;
217 
218 	case RED_HARD_MARK:
219 		sch->qstats.overlimits++;
220 		if (gred_use_harddrop(t) || !gred_use_ecn(t) ||
221 		    !INET_ECN_set_ce(skb)) {
222 			q->stats.forced_drop++;
223 			goto congestion_drop;
224 		}
225 		q->stats.forced_mark++;
226 		break;
227 	}
228 
229 	if (q->backlog + qdisc_pkt_len(skb) <= q->limit) {
230 		q->backlog += qdisc_pkt_len(skb);
231 		return qdisc_enqueue_tail(skb, sch);
232 	}
233 
234 	q->stats.pdrop++;
235 drop:
236 	return qdisc_drop(skb, sch);
237 
238 congestion_drop:
239 	qdisc_drop(skb, sch);
240 	return NET_XMIT_CN;
241 }
242 
243 static struct sk_buff *gred_dequeue(struct Qdisc *sch)
244 {
245 	struct sk_buff *skb;
246 	struct gred_sched *t = qdisc_priv(sch);
247 
248 	skb = qdisc_dequeue_head(sch);
249 
250 	if (skb) {
251 		struct gred_sched_data *q;
252 		u16 dp = tc_index_to_dp(skb);
253 
254 		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
255 			if (net_ratelimit())
256 				pr_warning("GRED: Unable to relocate VQ 0x%x "
257 					   "after dequeue, screwing up "
258 					   "backlog.\n", tc_index_to_dp(skb));
259 		} else {
260 			q->backlog -= qdisc_pkt_len(skb);
261 
262 			if (!q->backlog && !gred_wred_mode(t))
263 				red_start_of_idle_period(&q->parms);
264 		}
265 
266 		return skb;
267 	}
268 
269 	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
270 		red_start_of_idle_period(&t->wred_set);
271 
272 	return NULL;
273 }
274 
275 static unsigned int gred_drop(struct Qdisc *sch)
276 {
277 	struct sk_buff *skb;
278 	struct gred_sched *t = qdisc_priv(sch);
279 
280 	skb = qdisc_dequeue_tail(sch);
281 	if (skb) {
282 		unsigned int len = qdisc_pkt_len(skb);
283 		struct gred_sched_data *q;
284 		u16 dp = tc_index_to_dp(skb);
285 
286 		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
287 			if (net_ratelimit())
288 				pr_warning("GRED: Unable to relocate VQ 0x%x "
289 					   "while dropping, screwing up "
290 					   "backlog.\n", tc_index_to_dp(skb));
291 		} else {
292 			q->backlog -= len;
293 			q->stats.other++;
294 
295 			if (!q->backlog && !gred_wred_mode(t))
296 				red_start_of_idle_period(&q->parms);
297 		}
298 
299 		qdisc_drop(skb, sch);
300 		return len;
301 	}
302 
303 	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
304 		red_start_of_idle_period(&t->wred_set);
305 
306 	return 0;
307 
308 }
309 
310 static void gred_reset(struct Qdisc *sch)
311 {
312 	int i;
313 	struct gred_sched *t = qdisc_priv(sch);
314 
315 	qdisc_reset_queue(sch);
316 
317 	for (i = 0; i < t->DPs; i++) {
318 		struct gred_sched_data *q = t->tab[i];
319 
320 		if (!q)
321 			continue;
322 
323 		red_restart(&q->parms);
324 		q->backlog = 0;
325 	}
326 }
327 
328 static inline void gred_destroy_vq(struct gred_sched_data *q)
329 {
330 	kfree(q);
331 }
332 
333 static inline int gred_change_table_def(struct Qdisc *sch, struct nlattr *dps)
334 {
335 	struct gred_sched *table = qdisc_priv(sch);
336 	struct tc_gred_sopt *sopt;
337 	int i;
338 
339 	if (dps == NULL)
340 		return -EINVAL;
341 
342 	sopt = nla_data(dps);
343 
344 	if (sopt->DPs > MAX_DPs || sopt->DPs == 0 || sopt->def_DP >= sopt->DPs)
345 		return -EINVAL;
346 
347 	sch_tree_lock(sch);
348 	table->DPs = sopt->DPs;
349 	table->def = sopt->def_DP;
350 	table->red_flags = sopt->flags;
351 
352 	/*
353 	 * Every entry point to GRED is synchronized with the above code
354 	 * and the DP is checked against DPs, i.e. shadowed VQs can no
355 	 * longer be found so we can unlock right here.
356 	 */
357 	sch_tree_unlock(sch);
358 
359 	if (sopt->grio) {
360 		gred_enable_rio_mode(table);
361 		gred_disable_wred_mode(table);
362 		if (gred_wred_mode_check(sch))
363 			gred_enable_wred_mode(table);
364 	} else {
365 		gred_disable_rio_mode(table);
366 		gred_disable_wred_mode(table);
367 	}
368 
369 	for (i = table->DPs; i < MAX_DPs; i++) {
370 		if (table->tab[i]) {
371 			pr_warning("GRED: Warning: Destroying "
372 				   "shadowed VQ 0x%x\n", i);
373 			gred_destroy_vq(table->tab[i]);
374 			table->tab[i] = NULL;
375 		}
376 	}
377 
378 	return 0;
379 }
380 
381 static inline int gred_change_vq(struct Qdisc *sch, int dp,
382 				 struct tc_gred_qopt *ctl, int prio, u8 *stab)
383 {
384 	struct gred_sched *table = qdisc_priv(sch);
385 	struct gred_sched_data *q;
386 
387 	if (table->tab[dp] == NULL) {
388 		table->tab[dp] = kzalloc(sizeof(*q), GFP_KERNEL);
389 		if (table->tab[dp] == NULL)
390 			return -ENOMEM;
391 	}
392 
393 	q = table->tab[dp];
394 	q->DP = dp;
395 	q->prio = prio;
396 	q->limit = ctl->limit;
397 
398 	if (q->backlog == 0)
399 		red_end_of_idle_period(&q->parms);
400 
401 	red_set_parms(&q->parms,
402 		      ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
403 		      ctl->Scell_log, stab);
404 
405 	return 0;
406 }
407 
408 static const struct nla_policy gred_policy[TCA_GRED_MAX + 1] = {
409 	[TCA_GRED_PARMS]	= { .len = sizeof(struct tc_gred_qopt) },
410 	[TCA_GRED_STAB]		= { .len = 256 },
411 	[TCA_GRED_DPS]		= { .len = sizeof(struct tc_gred_sopt) },
412 };
413 
414 static int gred_change(struct Qdisc *sch, struct nlattr *opt)
415 {
416 	struct gred_sched *table = qdisc_priv(sch);
417 	struct tc_gred_qopt *ctl;
418 	struct nlattr *tb[TCA_GRED_MAX + 1];
419 	int err, prio = GRED_DEF_PRIO;
420 	u8 *stab;
421 
422 	if (opt == NULL)
423 		return -EINVAL;
424 
425 	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy);
426 	if (err < 0)
427 		return err;
428 
429 	if (tb[TCA_GRED_PARMS] == NULL && tb[TCA_GRED_STAB] == NULL)
430 		return gred_change_table_def(sch, opt);
431 
432 	if (tb[TCA_GRED_PARMS] == NULL ||
433 	    tb[TCA_GRED_STAB] == NULL)
434 		return -EINVAL;
435 
436 	err = -EINVAL;
437 	ctl = nla_data(tb[TCA_GRED_PARMS]);
438 	stab = nla_data(tb[TCA_GRED_STAB]);
439 
440 	if (ctl->DP >= table->DPs)
441 		goto errout;
442 
443 	if (gred_rio_mode(table)) {
444 		if (ctl->prio == 0) {
445 			int def_prio = GRED_DEF_PRIO;
446 
447 			if (table->tab[table->def])
448 				def_prio = table->tab[table->def]->prio;
449 
450 			printk(KERN_DEBUG "GRED: DP %u does not have a prio "
451 			       "setting default to %d\n", ctl->DP, def_prio);
452 
453 			prio = def_prio;
454 		} else
455 			prio = ctl->prio;
456 	}
457 
458 	sch_tree_lock(sch);
459 
460 	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab);
461 	if (err < 0)
462 		goto errout_locked;
463 
464 	if (gred_rio_mode(table)) {
465 		gred_disable_wred_mode(table);
466 		if (gred_wred_mode_check(sch))
467 			gred_enable_wred_mode(table);
468 	}
469 
470 	err = 0;
471 
472 errout_locked:
473 	sch_tree_unlock(sch);
474 errout:
475 	return err;
476 }
477 
478 static int gred_init(struct Qdisc *sch, struct nlattr *opt)
479 {
480 	struct nlattr *tb[TCA_GRED_MAX + 1];
481 	int err;
482 
483 	if (opt == NULL)
484 		return -EINVAL;
485 
486 	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy);
487 	if (err < 0)
488 		return err;
489 
490 	if (tb[TCA_GRED_PARMS] || tb[TCA_GRED_STAB])
491 		return -EINVAL;
492 
493 	return gred_change_table_def(sch, tb[TCA_GRED_DPS]);
494 }
495 
496 static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
497 {
498 	struct gred_sched *table = qdisc_priv(sch);
499 	struct nlattr *parms, *opts = NULL;
500 	int i;
501 	struct tc_gred_sopt sopt = {
502 		.DPs	= table->DPs,
503 		.def_DP	= table->def,
504 		.grio	= gred_rio_mode(table),
505 		.flags	= table->red_flags,
506 	};
507 
508 	opts = nla_nest_start(skb, TCA_OPTIONS);
509 	if (opts == NULL)
510 		goto nla_put_failure;
511 	NLA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
512 	parms = nla_nest_start(skb, TCA_GRED_PARMS);
513 	if (parms == NULL)
514 		goto nla_put_failure;
515 
516 	for (i = 0; i < MAX_DPs; i++) {
517 		struct gred_sched_data *q = table->tab[i];
518 		struct tc_gred_qopt opt;
519 
520 		memset(&opt, 0, sizeof(opt));
521 
522 		if (!q) {
523 			/* hack -- fix at some point with proper message
524 			   This is how we indicate to tc that there is no VQ
525 			   at this DP */
526 
527 			opt.DP = MAX_DPs + i;
528 			goto append_opt;
529 		}
530 
531 		opt.limit	= q->limit;
532 		opt.DP		= q->DP;
533 		opt.backlog	= q->backlog;
534 		opt.prio	= q->prio;
535 		opt.qth_min	= q->parms.qth_min >> q->parms.Wlog;
536 		opt.qth_max	= q->parms.qth_max >> q->parms.Wlog;
537 		opt.Wlog	= q->parms.Wlog;
538 		opt.Plog	= q->parms.Plog;
539 		opt.Scell_log	= q->parms.Scell_log;
540 		opt.other	= q->stats.other;
541 		opt.early	= q->stats.prob_drop;
542 		opt.forced	= q->stats.forced_drop;
543 		opt.pdrop	= q->stats.pdrop;
544 		opt.packets	= q->packetsin;
545 		opt.bytesin	= q->bytesin;
546 
547 		if (gred_wred_mode(table)) {
548 			q->parms.qidlestart =
549 				table->tab[table->def]->parms.qidlestart;
550 			q->parms.qavg = table->tab[table->def]->parms.qavg;
551 		}
552 
553 		opt.qave = red_calc_qavg(&q->parms, q->parms.qavg);
554 
555 append_opt:
556 		if (nla_append(skb, sizeof(opt), &opt) < 0)
557 			goto nla_put_failure;
558 	}
559 
560 	nla_nest_end(skb, parms);
561 
562 	return nla_nest_end(skb, opts);
563 
564 nla_put_failure:
565 	nla_nest_cancel(skb, opts);
566 	return -EMSGSIZE;
567 }
568 
569 static void gred_destroy(struct Qdisc *sch)
570 {
571 	struct gred_sched *table = qdisc_priv(sch);
572 	int i;
573 
574 	for (i = 0; i < table->DPs; i++) {
575 		if (table->tab[i])
576 			gred_destroy_vq(table->tab[i]);
577 	}
578 }
579 
580 static struct Qdisc_ops gred_qdisc_ops __read_mostly = {
581 	.id		=	"gred",
582 	.priv_size	=	sizeof(struct gred_sched),
583 	.enqueue	=	gred_enqueue,
584 	.dequeue	=	gred_dequeue,
585 	.peek		=	qdisc_peek_head,
586 	.drop		=	gred_drop,
587 	.init		=	gred_init,
588 	.reset		=	gred_reset,
589 	.destroy	=	gred_destroy,
590 	.change		=	gred_change,
591 	.dump		=	gred_dump,
592 	.owner		=	THIS_MODULE,
593 };
594 
595 static int __init gred_module_init(void)
596 {
597 	return register_qdisc(&gred_qdisc_ops);
598 }
599 
600 static void __exit gred_module_exit(void)
601 {
602 	unregister_qdisc(&gred_qdisc_ops);
603 }
604 
605 module_init(gred_module_init)
606 module_exit(gred_module_exit)
607 
608 MODULE_LICENSE("GPL");
609