xref: /openbmc/linux/net/sched/sch_gred.c (revision 5a0e3ad6) 
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 {
37 	u32		limit;		/* HARD maximal queue length	*/
38 	u32      	DP;		/* the drop pramaters */
39 	u32		bytesin;	/* bytes seen on virtualQ so far*/
40 	u32		packetsin;	/* packets seen on virtualQ so far*/
41 	u32		backlog;	/* bytes on the virtualQ */
42 	u8		prio;		/* the prio of this vq */
43 
44 	struct red_parms parms;
45 	struct red_stats stats;
46 };
47 
48 enum {
49 	GRED_WRED_MODE = 1,
50 	GRED_RIO_MODE,
51 };
52 
53 struct gred_sched
54 {
55 	struct gred_sched_data *tab[MAX_DPs];
56 	unsigned long	flags;
57 	u32		red_flags;
58 	u32 		DPs;
59 	u32 		def;
60 	struct red_parms wred_set;
61 };
62 
63 static inline int gred_wred_mode(struct gred_sched *table)
64 {
65 	return test_bit(GRED_WRED_MODE, &table->flags);
66 }
67 
68 static inline void gred_enable_wred_mode(struct gred_sched *table)
69 {
70 	__set_bit(GRED_WRED_MODE, &table->flags);
71 }
72 
73 static inline void gred_disable_wred_mode(struct gred_sched *table)
74 {
75 	__clear_bit(GRED_WRED_MODE, &table->flags);
76 }
77 
78 static inline int gred_rio_mode(struct gred_sched *table)
79 {
80 	return test_bit(GRED_RIO_MODE, &table->flags);
81 }
82 
83 static inline void gred_enable_rio_mode(struct gred_sched *table)
84 {
85 	__set_bit(GRED_RIO_MODE, &table->flags);
86 }
87 
88 static inline void gred_disable_rio_mode(struct gred_sched *table)
89 {
90 	__clear_bit(GRED_RIO_MODE, &table->flags);
91 }
92 
93 static inline int gred_wred_mode_check(struct Qdisc *sch)
94 {
95 	struct gred_sched *table = qdisc_priv(sch);
96 	int i;
97 
98 	/* Really ugly O(n^2) but shouldn't be necessary too frequent. */
99 	for (i = 0; i < table->DPs; i++) {
100 		struct gred_sched_data *q = table->tab[i];
101 		int n;
102 
103 		if (q == NULL)
104 			continue;
105 
106 		for (n = 0; n < table->DPs; n++)
107 			if (table->tab[n] && table->tab[n] != q &&
108 			    table->tab[n]->prio == q->prio)
109 				return 1;
110 	}
111 
112 	return 0;
113 }
114 
115 static inline unsigned int gred_backlog(struct gred_sched *table,
116 					struct gred_sched_data *q,
117 					struct Qdisc *sch)
118 {
119 	if (gred_wred_mode(table))
120 		return sch->qstats.backlog;
121 	else
122 		return q->backlog;
123 }
124 
125 static inline u16 tc_index_to_dp(struct sk_buff *skb)
126 {
127 	return skb->tc_index & GRED_VQ_MASK;
128 }
129 
130 static inline void gred_load_wred_set(struct gred_sched *table,
131 				      struct gred_sched_data *q)
132 {
133 	q->parms.qavg = table->wred_set.qavg;
134 	q->parms.qidlestart = table->wred_set.qidlestart;
135 }
136 
137 static inline void gred_store_wred_set(struct gred_sched *table,
138 				       struct gred_sched_data *q)
139 {
140 	table->wred_set.qavg = q->parms.qavg;
141 }
142 
143 static inline int gred_use_ecn(struct gred_sched *t)
144 {
145 	return t->red_flags & TC_RED_ECN;
146 }
147 
148 static inline int gred_use_harddrop(struct gred_sched *t)
149 {
150 	return t->red_flags & TC_RED_HARDDROP;
151 }
152 
153 static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch)
154 {
155 	struct gred_sched_data *q=NULL;
156 	struct gred_sched *t= qdisc_priv(sch);
157 	unsigned long qavg = 0;
158 	u16 dp = tc_index_to_dp(skb);
159 
160 	if (dp >= t->DPs  || (q = t->tab[dp]) == NULL) {
161 		dp = t->def;
162 
163 		if ((q = t->tab[dp]) == NULL) {
164 			/* Pass through packets not assigned to a DP
165 			 * if no default DP has been configured. This
166 			 * allows for DP flows to be left untouched.
167 			 */
168 			if (skb_queue_len(&sch->q) < qdisc_dev(sch)->tx_queue_len)
169 				return qdisc_enqueue_tail(skb, sch);
170 			else
171 				goto drop;
172 		}
173 
174 		/* fix tc_index? --could be controvesial but needed for
175 		   requeueing */
176 		skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
177 	}
178 
179 	/* sum up all the qaves of prios <= to ours to get the new qave */
180 	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
181 		int i;
182 
183 		for (i = 0; i < t->DPs; i++) {
184 			if (t->tab[i] && t->tab[i]->prio < q->prio &&
185 			    !red_is_idling(&t->tab[i]->parms))
186 				qavg +=t->tab[i]->parms.qavg;
187 		}
188 
189 	}
190 
191 	q->packetsin++;
192 	q->bytesin += qdisc_pkt_len(skb);
193 
194 	if (gred_wred_mode(t))
195 		gred_load_wred_set(t, q);
196 
197 	q->parms.qavg = red_calc_qavg(&q->parms, gred_backlog(t, q, sch));
198 
199 	if (red_is_idling(&q->parms))
200 		red_end_of_idle_period(&q->parms);
201 
202 	if (gred_wred_mode(t))
203 		gred_store_wred_set(t, q);
204 
205 	switch (red_action(&q->parms, q->parms.qavg + qavg)) {
206 		case RED_DONT_MARK:
207 			break;
208 
209 		case RED_PROB_MARK:
210 			sch->qstats.overlimits++;
211 			if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) {
212 				q->stats.prob_drop++;
213 				goto congestion_drop;
214 			}
215 
216 			q->stats.prob_mark++;
217 			break;
218 
219 		case RED_HARD_MARK:
220 			sch->qstats.overlimits++;
221 			if (gred_use_harddrop(t) || !gred_use_ecn(t) ||
222 			    !INET_ECN_set_ce(skb)) {
223 				q->stats.forced_drop++;
224 				goto congestion_drop;
225 			}
226 			q->stats.forced_mark++;
227 			break;
228 	}
229 
230 	if (q->backlog + qdisc_pkt_len(skb) <= q->limit) {
231 		q->backlog += qdisc_pkt_len(skb);
232 		return qdisc_enqueue_tail(skb, sch);
233 	}
234 
235 	q->stats.pdrop++;
236 drop:
237 	return qdisc_drop(skb, sch);
238 
239 congestion_drop:
240 	qdisc_drop(skb, sch);
241 	return NET_XMIT_CN;
242 }
243 
244 static struct sk_buff *gred_dequeue(struct Qdisc* sch)
245 {
246 	struct sk_buff *skb;
247 	struct gred_sched *t = qdisc_priv(sch);
248 
249 	skb = qdisc_dequeue_head(sch);
250 
251 	if (skb) {
252 		struct gred_sched_data *q;
253 		u16 dp = tc_index_to_dp(skb);
254 
255 		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
256 			if (net_ratelimit())
257 				printk(KERN_WARNING "GRED: Unable to relocate "
258 				       "VQ 0x%x after dequeue, screwing up "
259 				       "backlog.\n", tc_index_to_dp(skb));
260 		} else {
261 			q->backlog -= qdisc_pkt_len(skb);
262 
263 			if (!q->backlog && !gred_wred_mode(t))
264 				red_start_of_idle_period(&q->parms);
265 		}
266 
267 		return skb;
268 	}
269 
270 	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
271 		red_start_of_idle_period(&t->wred_set);
272 
273 	return NULL;
274 }
275 
276 static unsigned int gred_drop(struct Qdisc* sch)
277 {
278 	struct sk_buff *skb;
279 	struct gred_sched *t = qdisc_priv(sch);
280 
281 	skb = qdisc_dequeue_tail(sch);
282 	if (skb) {
283 		unsigned int len = qdisc_pkt_len(skb);
284 		struct gred_sched_data *q;
285 		u16 dp = tc_index_to_dp(skb);
286 
287 		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
288 			if (net_ratelimit())
289 				printk(KERN_WARNING "GRED: Unable to relocate "
290 				       "VQ 0x%x while dropping, screwing up "
291 				       "backlog.\n", tc_index_to_dp(skb));
292 		} else {
293 			q->backlog -= len;
294 			q->stats.other++;
295 
296 			if (!q->backlog && !gred_wred_mode(t))
297 				red_start_of_idle_period(&q->parms);
298 		}
299 
300 		qdisc_drop(skb, sch);
301 		return len;
302 	}
303 
304 	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
305 		red_start_of_idle_period(&t->wred_set);
306 
307 	return 0;
308 
309 }
310 
311 static void gred_reset(struct Qdisc* sch)
312 {
313 	int i;
314 	struct gred_sched *t = qdisc_priv(sch);
315 
316 	qdisc_reset_queue(sch);
317 
318 	for (i = 0; i < t->DPs; i++) {
319 		struct gred_sched_data *q = t->tab[i];
320 
321 		if (!q)
322 			continue;
323 
324 		red_restart(&q->parms);
325 		q->backlog = 0;
326 	}
327 }
328 
329 static inline void gred_destroy_vq(struct gred_sched_data *q)
330 {
331 	kfree(q);
332 }
333 
334 static inline int gred_change_table_def(struct Qdisc *sch, struct nlattr *dps)
335 {
336 	struct gred_sched *table = qdisc_priv(sch);
337 	struct tc_gred_sopt *sopt;
338 	int i;
339 
340 	if (dps == NULL)
341 		return -EINVAL;
342 
343 	sopt = nla_data(dps);
344 
345 	if (sopt->DPs > MAX_DPs || sopt->DPs == 0 || sopt->def_DP >= sopt->DPs)
346 		return -EINVAL;
347 
348 	sch_tree_lock(sch);
349 	table->DPs = sopt->DPs;
350 	table->def = sopt->def_DP;
351 	table->red_flags = sopt->flags;
352 
353 	/*
354 	 * Every entry point to GRED is synchronized with the above code
355 	 * and the DP is checked against DPs, i.e. shadowed VQs can no
356 	 * longer be found so we can unlock right here.
357 	 */
358 	sch_tree_unlock(sch);
359 
360 	if (sopt->grio) {
361 		gred_enable_rio_mode(table);
362 		gred_disable_wred_mode(table);
363 		if (gred_wred_mode_check(sch))
364 			gred_enable_wred_mode(table);
365 	} else {
366 		gred_disable_rio_mode(table);
367 		gred_disable_wred_mode(table);
368 	}
369 
370 	for (i = table->DPs; i < MAX_DPs; i++) {
371 		if (table->tab[i]) {
372 			printk(KERN_WARNING "GRED: Warning: Destroying "
373 			       "shadowed VQ 0x%x\n", i);
374 			gred_destroy_vq(table->tab[i]);
375 			table->tab[i] = NULL;
376 		}
377 	}
378 
379 	return 0;
380 }
381 
382 static inline int gred_change_vq(struct Qdisc *sch, int dp,
383 				 struct tc_gred_qopt *ctl, int prio, u8 *stab)
384 {
385 	struct gred_sched *table = qdisc_priv(sch);
386 	struct gred_sched_data *q;
387 
388 	if (table->tab[dp] == NULL) {
389 		table->tab[dp] = kzalloc(sizeof(*q), GFP_KERNEL);
390 		if (table->tab[dp] == NULL)
391 			return -ENOMEM;
392 	}
393 
394 	q = table->tab[dp];
395 	q->DP = dp;
396 	q->prio = prio;
397 	q->limit = ctl->limit;
398 
399 	if (q->backlog == 0)
400 		red_end_of_idle_period(&q->parms);
401 
402 	red_set_parms(&q->parms,
403 		      ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
404 		      ctl->Scell_log, stab);
405 
406 	return 0;
407 }
408 
409 static const struct nla_policy gred_policy[TCA_GRED_MAX + 1] = {
410 	[TCA_GRED_PARMS]	= { .len = sizeof(struct tc_gred_qopt) },
411 	[TCA_GRED_STAB]		= { .len = 256 },
412 	[TCA_GRED_DPS]		= { .len = sizeof(struct tc_gred_sopt) },
413 };
414 
415 static int gred_change(struct Qdisc *sch, struct nlattr *opt)
416 {
417 	struct gred_sched *table = qdisc_priv(sch);
418 	struct tc_gred_qopt *ctl;
419 	struct nlattr *tb[TCA_GRED_MAX + 1];
420 	int err, prio = GRED_DEF_PRIO;
421 	u8 *stab;
422 
423 	if (opt == NULL)
424 		return -EINVAL;
425 
426 	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy);
427 	if (err < 0)
428 		return err;
429 
430 	if (tb[TCA_GRED_PARMS] == NULL && tb[TCA_GRED_STAB] == NULL)
431 		return gred_change_table_def(sch, opt);
432 
433 	if (tb[TCA_GRED_PARMS] == NULL ||
434 	    tb[TCA_GRED_STAB] == NULL)
435 		return -EINVAL;
436 
437 	err = -EINVAL;
438 	ctl = nla_data(tb[TCA_GRED_PARMS]);
439 	stab = nla_data(tb[TCA_GRED_STAB]);
440 
441 	if (ctl->DP >= table->DPs)
442 		goto errout;
443 
444 	if (gred_rio_mode(table)) {
445 		if (ctl->prio == 0) {
446 			int def_prio = GRED_DEF_PRIO;
447 
448 			if (table->tab[table->def])
449 				def_prio = table->tab[table->def]->prio;
450 
451 			printk(KERN_DEBUG "GRED: DP %u does not have a prio "
452 			       "setting default to %d\n", ctl->DP, def_prio);
453 
454 			prio = def_prio;
455 		} else
456 			prio = ctl->prio;
457 	}
458 
459 	sch_tree_lock(sch);
460 
461 	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab);
462 	if (err < 0)
463 		goto errout_locked;
464 
465 	if (gred_rio_mode(table)) {
466 		gred_disable_wred_mode(table);
467 		if (gred_wred_mode_check(sch))
468 			gred_enable_wred_mode(table);
469 	}
470 
471 	err = 0;
472 
473 errout_locked:
474 	sch_tree_unlock(sch);
475 errout:
476 	return err;
477 }
478 
479 static int gred_init(struct Qdisc *sch, struct nlattr *opt)
480 {
481 	struct nlattr *tb[TCA_GRED_MAX + 1];
482 	int err;
483 
484 	if (opt == NULL)
485 		return -EINVAL;
486 
487 	err = nla_parse_nested(tb, TCA_GRED_MAX, opt, gred_policy);
488 	if (err < 0)
489 		return err;
490 
491 	if (tb[TCA_GRED_PARMS] || tb[TCA_GRED_STAB])
492 		return -EINVAL;
493 
494 	return gred_change_table_def(sch, tb[TCA_GRED_DPS]);
495 }
496 
497 static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
498 {
499 	struct gred_sched *table = qdisc_priv(sch);
500 	struct nlattr *parms, *opts = NULL;
501 	int i;
502 	struct tc_gred_sopt sopt = {
503 		.DPs	= table->DPs,
504 		.def_DP	= table->def,
505 		.grio	= gred_rio_mode(table),
506 		.flags	= table->red_flags,
507 	};
508 
509 	opts = nla_nest_start(skb, TCA_OPTIONS);
510 	if (opts == NULL)
511 		goto nla_put_failure;
512 	NLA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
513 	parms = nla_nest_start(skb, TCA_GRED_PARMS);
514 	if (parms == NULL)
515 		goto nla_put_failure;
516 
517 	for (i = 0; i < MAX_DPs; i++) {
518 		struct gred_sched_data *q = table->tab[i];
519 		struct tc_gred_qopt opt;
520 
521 		memset(&opt, 0, sizeof(opt));
522 
523 		if (!q) {
524 			/* hack -- fix at some point with proper message
525 			   This is how we indicate to tc that there is no VQ
526 			   at this DP */
527 
528 			opt.DP = MAX_DPs + i;
529 			goto append_opt;
530 		}
531 
532 		opt.limit	= q->limit;
533 		opt.DP		= q->DP;
534 		opt.backlog	= q->backlog;
535 		opt.prio	= q->prio;
536 		opt.qth_min	= q->parms.qth_min >> q->parms.Wlog;
537 		opt.qth_max	= q->parms.qth_max >> q->parms.Wlog;
538 		opt.Wlog	= q->parms.Wlog;
539 		opt.Plog	= q->parms.Plog;
540 		opt.Scell_log	= q->parms.Scell_log;
541 		opt.other	= q->stats.other;
542 		opt.early	= q->stats.prob_drop;
543 		opt.forced	= q->stats.forced_drop;
544 		opt.pdrop	= q->stats.pdrop;
545 		opt.packets	= q->packetsin;
546 		opt.bytesin	= q->bytesin;
547 
548 		if (gred_wred_mode(table)) {
549 			q->parms.qidlestart =
550 				table->tab[table->def]->parms.qidlestart;
551 			q->parms.qavg = table->tab[table->def]->parms.qavg;
552 		}
553 
554 		opt.qave = red_calc_qavg(&q->parms, q->parms.qavg);
555 
556 append_opt:
557 		if (nla_append(skb, sizeof(opt), &opt) < 0)
558 			goto nla_put_failure;
559 	}
560 
561 	nla_nest_end(skb, parms);
562 
563 	return nla_nest_end(skb, opts);
564 
565 nla_put_failure:
566 	nla_nest_cancel(skb, opts);
567 	return -EMSGSIZE;
568 }
569 
570 static void gred_destroy(struct Qdisc *sch)
571 {
572 	struct gred_sched *table = qdisc_priv(sch);
573 	int i;
574 
575 	for (i = 0; i < table->DPs; i++) {
576 		if (table->tab[i])
577 			gred_destroy_vq(table->tab[i]);
578 	}
579 }
580 
581 static struct Qdisc_ops gred_qdisc_ops __read_mostly = {
582 	.id		=	"gred",
583 	.priv_size	=	sizeof(struct gred_sched),
584 	.enqueue	=	gred_enqueue,
585 	.dequeue	=	gred_dequeue,
586 	.peek		=	qdisc_peek_head,
587 	.drop		=	gred_drop,
588 	.init		=	gred_init,
589 	.reset		=	gred_reset,
590 	.destroy	=	gred_destroy,
591 	.change		=	gred_change,
592 	.dump		=	gred_dump,
593 	.owner		=	THIS_MODULE,
594 };
595 
596 static int __init gred_module_init(void)
597 {
598 	return register_qdisc(&gred_qdisc_ops);
599 }
600 
601 static void __exit gred_module_exit(void)
602 {
603 	unregister_qdisc(&gred_qdisc_ops);
604 }
605 
606 module_init(gred_module_init)
607 module_exit(gred_module_exit)
608 
609 MODULE_LICENSE("GPL");
610