xref: /linux/net/sched/sch_gred.c (revision f3d9478b2ce468c3115b02ecae7e975990697f15) 
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/config.h>
22 #include <linux/module.h>
23 #include <linux/types.h>
24 #include <linux/kernel.h>
25 #include <linux/netdevice.h>
26 #include <linux/skbuff.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 struct gred_sched_data;
34 struct gred_sched;
35 
36 struct gred_sched_data
37 {
38 	u32		limit;		/* HARD maximal queue length	*/
39 	u32      	DP;		/* the drop pramaters */
40 	u32		bytesin;	/* bytes seen on virtualQ so far*/
41 	u32		packetsin;	/* packets seen on virtualQ so far*/
42 	u32		backlog;	/* bytes on the virtualQ */
43 	u8		prio;		/* the prio of this vq */
44 
45 	struct red_parms parms;
46 	struct red_stats stats;
47 };
48 
49 enum {
50 	GRED_WRED_MODE = 1,
51 	GRED_RIO_MODE,
52 };
53 
54 struct gred_sched
55 {
56 	struct gred_sched_data *tab[MAX_DPs];
57 	unsigned long	flags;
58 	u32		red_flags;
59 	u32 		DPs;
60 	u32 		def;
61 	struct red_parms wred_set;
62 };
63 
64 static inline int gred_wred_mode(struct gred_sched *table)
65 {
66 	return test_bit(GRED_WRED_MODE, &table->flags);
67 }
68 
69 static inline void gred_enable_wred_mode(struct gred_sched *table)
70 {
71 	__set_bit(GRED_WRED_MODE, &table->flags);
72 }
73 
74 static inline void gred_disable_wred_mode(struct gred_sched *table)
75 {
76 	__clear_bit(GRED_WRED_MODE, &table->flags);
77 }
78 
79 static inline int gred_rio_mode(struct gred_sched *table)
80 {
81 	return test_bit(GRED_RIO_MODE, &table->flags);
82 }
83 
84 static inline void gred_enable_rio_mode(struct gred_sched *table)
85 {
86 	__set_bit(GRED_RIO_MODE, &table->flags);
87 }
88 
89 static inline void gred_disable_rio_mode(struct gred_sched *table)
90 {
91 	__clear_bit(GRED_RIO_MODE, &table->flags);
92 }
93 
94 static inline int gred_wred_mode_check(struct Qdisc *sch)
95 {
96 	struct gred_sched *table = qdisc_priv(sch);
97 	int i;
98 
99 	/* Really ugly O(n^2) but shouldn't be necessary too frequent. */
100 	for (i = 0; i < table->DPs; i++) {
101 		struct gred_sched_data *q = table->tab[i];
102 		int n;
103 
104 		if (q == NULL)
105 			continue;
106 
107 		for (n = 0; n < table->DPs; n++)
108 			if (table->tab[n] && table->tab[n] != q &&
109 			    table->tab[n]->prio == q->prio)
110 				return 1;
111 	}
112 
113 	return 0;
114 }
115 
116 static inline unsigned int gred_backlog(struct gred_sched *table,
117 					struct gred_sched_data *q,
118 					struct Qdisc *sch)
119 {
120 	if (gred_wred_mode(table))
121 		return sch->qstats.backlog;
122 	else
123 		return q->backlog;
124 }
125 
126 static inline u16 tc_index_to_dp(struct sk_buff *skb)
127 {
128 	return skb->tc_index & GRED_VQ_MASK;
129 }
130 
131 static inline void gred_load_wred_set(struct gred_sched *table,
132 				      struct gred_sched_data *q)
133 {
134 	q->parms.qavg = table->wred_set.qavg;
135 	q->parms.qidlestart = table->wred_set.qidlestart;
136 }
137 
138 static inline void gred_store_wred_set(struct gred_sched *table,
139 				       struct gred_sched_data *q)
140 {
141 	table->wred_set.qavg = q->parms.qavg;
142 }
143 
144 static inline int gred_use_ecn(struct gred_sched *t)
145 {
146 	return t->red_flags & TC_RED_ECN;
147 }
148 
149 static inline int gred_use_harddrop(struct gred_sched *t)
150 {
151 	return t->red_flags & TC_RED_HARDDROP;
152 }
153 
154 static int gred_enqueue(struct sk_buff *skb, struct Qdisc* sch)
155 {
156 	struct gred_sched_data *q=NULL;
157 	struct gred_sched *t= qdisc_priv(sch);
158 	unsigned long qavg = 0;
159 	u16 dp = tc_index_to_dp(skb);
160 
161 	if (dp >= t->DPs  || (q = t->tab[dp]) == NULL) {
162 		dp = t->def;
163 
164 		if ((q = t->tab[dp]) == NULL) {
165 			/* Pass through packets not assigned to a DP
166 			 * if no default DP has been configured. This
167 			 * allows for DP flows to be left untouched.
168 			 */
169 			if (skb_queue_len(&sch->q) < sch->dev->tx_queue_len)
170 				return qdisc_enqueue_tail(skb, sch);
171 			else
172 				goto drop;
173 		}
174 
175 		/* fix tc_index? --could be controvesial but needed for
176 		   requeueing */
177 		skb->tc_index = (skb->tc_index & ~GRED_VQ_MASK) | dp;
178 	}
179 
180 	/* sum up all the qaves of prios <= to ours to get the new qave */
181 	if (!gred_wred_mode(t) && gred_rio_mode(t)) {
182 		int i;
183 
184 		for (i = 0; i < t->DPs; i++) {
185 			if (t->tab[i] && t->tab[i]->prio < q->prio &&
186 			    !red_is_idling(&t->tab[i]->parms))
187 				qavg +=t->tab[i]->parms.qavg;
188 		}
189 
190 	}
191 
192 	q->packetsin++;
193 	q->bytesin += skb->len;
194 
195 	if (gred_wred_mode(t))
196 		gred_load_wred_set(t, q);
197 
198 	q->parms.qavg = red_calc_qavg(&q->parms, gred_backlog(t, q, sch));
199 
200 	if (red_is_idling(&q->parms))
201 		red_end_of_idle_period(&q->parms);
202 
203 	if (gred_wred_mode(t))
204 		gred_store_wred_set(t, q);
205 
206 	switch (red_action(&q->parms, q->parms.qavg + qavg)) {
207 		case RED_DONT_MARK:
208 			break;
209 
210 		case RED_PROB_MARK:
211 			sch->qstats.overlimits++;
212 			if (!gred_use_ecn(t) || !INET_ECN_set_ce(skb)) {
213 				q->stats.prob_drop++;
214 				goto congestion_drop;
215 			}
216 
217 			q->stats.prob_mark++;
218 			break;
219 
220 		case RED_HARD_MARK:
221 			sch->qstats.overlimits++;
222 			if (gred_use_harddrop(t) || !gred_use_ecn(t) ||
223 			    !INET_ECN_set_ce(skb)) {
224 				q->stats.forced_drop++;
225 				goto congestion_drop;
226 			}
227 			q->stats.forced_mark++;
228 			break;
229 	}
230 
231 	if (q->backlog + skb->len <= q->limit) {
232 		q->backlog += skb->len;
233 		return qdisc_enqueue_tail(skb, sch);
234 	}
235 
236 	q->stats.pdrop++;
237 drop:
238 	return qdisc_drop(skb, sch);
239 
240 congestion_drop:
241 	qdisc_drop(skb, sch);
242 	return NET_XMIT_CN;
243 }
244 
245 static int gred_requeue(struct sk_buff *skb, struct Qdisc* sch)
246 {
247 	struct gred_sched *t = qdisc_priv(sch);
248 	struct gred_sched_data *q;
249 	u16 dp = tc_index_to_dp(skb);
250 
251 	if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
252 		if (net_ratelimit())
253 			printk(KERN_WARNING "GRED: Unable to relocate VQ 0x%x "
254 			       "for requeue, screwing up backlog.\n",
255 			       tc_index_to_dp(skb));
256 	} else {
257 		if (red_is_idling(&q->parms))
258 			red_end_of_idle_period(&q->parms);
259 		q->backlog += skb->len;
260 	}
261 
262 	return qdisc_requeue(skb, sch);
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 			if (net_ratelimit())
278 				printk(KERN_WARNING "GRED: Unable to relocate "
279 				       "VQ 0x%x after dequeue, screwing up "
280 				       "backlog.\n", tc_index_to_dp(skb));
281 		} else {
282 			q->backlog -= skb->len;
283 
284 			if (!q->backlog && !gred_wred_mode(t))
285 				red_start_of_idle_period(&q->parms);
286 		}
287 
288 		return skb;
289 	}
290 
291 	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
292 		red_start_of_idle_period(&t->wred_set);
293 
294 	return NULL;
295 }
296 
297 static unsigned int gred_drop(struct Qdisc* sch)
298 {
299 	struct sk_buff *skb;
300 	struct gred_sched *t = qdisc_priv(sch);
301 
302 	skb = qdisc_dequeue_tail(sch);
303 	if (skb) {
304 		unsigned int len = skb->len;
305 		struct gred_sched_data *q;
306 		u16 dp = tc_index_to_dp(skb);
307 
308 		if (dp >= t->DPs || (q = t->tab[dp]) == NULL) {
309 			if (net_ratelimit())
310 				printk(KERN_WARNING "GRED: Unable to relocate "
311 				       "VQ 0x%x while dropping, screwing up "
312 				       "backlog.\n", tc_index_to_dp(skb));
313 		} else {
314 			q->backlog -= len;
315 			q->stats.other++;
316 
317 			if (!q->backlog && !gred_wred_mode(t))
318 				red_start_of_idle_period(&q->parms);
319 		}
320 
321 		qdisc_drop(skb, sch);
322 		return len;
323 	}
324 
325 	if (gred_wred_mode(t) && !red_is_idling(&t->wred_set))
326 		red_start_of_idle_period(&t->wred_set);
327 
328 	return 0;
329 
330 }
331 
332 static void gred_reset(struct Qdisc* sch)
333 {
334 	int i;
335 	struct gred_sched *t = qdisc_priv(sch);
336 
337 	qdisc_reset_queue(sch);
338 
339         for (i = 0; i < t->DPs; i++) {
340 		struct gred_sched_data *q = t->tab[i];
341 
342 		if (!q)
343 			continue;
344 
345 		red_restart(&q->parms);
346 		q->backlog = 0;
347 	}
348 }
349 
350 static inline void gred_destroy_vq(struct gred_sched_data *q)
351 {
352 	kfree(q);
353 }
354 
355 static inline int gred_change_table_def(struct Qdisc *sch, struct rtattr *dps)
356 {
357 	struct gred_sched *table = qdisc_priv(sch);
358 	struct tc_gred_sopt *sopt;
359 	int i;
360 
361 	if (dps == NULL || RTA_PAYLOAD(dps) < sizeof(*sopt))
362 		return -EINVAL;
363 
364 	sopt = RTA_DATA(dps);
365 
366 	if (sopt->DPs > MAX_DPs || sopt->DPs == 0 || sopt->def_DP >= sopt->DPs)
367 		return -EINVAL;
368 
369 	sch_tree_lock(sch);
370 	table->DPs = sopt->DPs;
371 	table->def = sopt->def_DP;
372 	table->red_flags = sopt->flags;
373 
374 	/*
375 	 * Every entry point to GRED is synchronized with the above code
376 	 * and the DP is checked against DPs, i.e. shadowed VQs can no
377 	 * longer be found so we can unlock right here.
378 	 */
379 	sch_tree_unlock(sch);
380 
381 	if (sopt->grio) {
382 		gred_enable_rio_mode(table);
383 		gred_disable_wred_mode(table);
384 		if (gred_wred_mode_check(sch))
385 			gred_enable_wred_mode(table);
386 	} else {
387 		gred_disable_rio_mode(table);
388 		gred_disable_wred_mode(table);
389 	}
390 
391 	for (i = table->DPs; i < MAX_DPs; i++) {
392 		if (table->tab[i]) {
393 			printk(KERN_WARNING "GRED: Warning: Destroying "
394 			       "shadowed VQ 0x%x\n", i);
395 			gred_destroy_vq(table->tab[i]);
396 			table->tab[i] = NULL;
397   		}
398 	}
399 
400 	return 0;
401 }
402 
403 static inline int gred_change_vq(struct Qdisc *sch, int dp,
404 				 struct tc_gred_qopt *ctl, int prio, u8 *stab)
405 {
406 	struct gred_sched *table = qdisc_priv(sch);
407 	struct gred_sched_data *q;
408 
409 	if (table->tab[dp] == NULL) {
410 		table->tab[dp] = kmalloc(sizeof(*q), GFP_KERNEL);
411 		if (table->tab[dp] == NULL)
412 			return -ENOMEM;
413 		memset(table->tab[dp], 0, sizeof(*q));
414 	}
415 
416 	q = table->tab[dp];
417 	q->DP = dp;
418 	q->prio = prio;
419 	q->limit = ctl->limit;
420 
421 	if (q->backlog == 0)
422 		red_end_of_idle_period(&q->parms);
423 
424 	red_set_parms(&q->parms,
425 		      ctl->qth_min, ctl->qth_max, ctl->Wlog, ctl->Plog,
426 		      ctl->Scell_log, stab);
427 
428 	return 0;
429 }
430 
431 static int gred_change(struct Qdisc *sch, struct rtattr *opt)
432 {
433 	struct gred_sched *table = qdisc_priv(sch);
434 	struct tc_gred_qopt *ctl;
435 	struct rtattr *tb[TCA_GRED_MAX];
436 	int err = -EINVAL, prio = GRED_DEF_PRIO;
437 	u8 *stab;
438 
439 	if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
440 		return -EINVAL;
441 
442 	if (tb[TCA_GRED_PARMS-1] == NULL && tb[TCA_GRED_STAB-1] == NULL)
443 		return gred_change_table_def(sch, opt);
444 
445 	if (tb[TCA_GRED_PARMS-1] == NULL ||
446 	    RTA_PAYLOAD(tb[TCA_GRED_PARMS-1]) < sizeof(*ctl) ||
447 	    tb[TCA_GRED_STAB-1] == NULL ||
448 	    RTA_PAYLOAD(tb[TCA_GRED_STAB-1]) < 256)
449 		return -EINVAL;
450 
451 	ctl = RTA_DATA(tb[TCA_GRED_PARMS-1]);
452 	stab = RTA_DATA(tb[TCA_GRED_STAB-1]);
453 
454 	if (ctl->DP >= table->DPs)
455 		goto errout;
456 
457 	if (gred_rio_mode(table)) {
458 		if (ctl->prio == 0) {
459 			int def_prio = GRED_DEF_PRIO;
460 
461 			if (table->tab[table->def])
462 				def_prio = table->tab[table->def]->prio;
463 
464 			printk(KERN_DEBUG "GRED: DP %u does not have a prio "
465 			       "setting default to %d\n", ctl->DP, def_prio);
466 
467 			prio = def_prio;
468 		} else
469 			prio = ctl->prio;
470 	}
471 
472 	sch_tree_lock(sch);
473 
474 	err = gred_change_vq(sch, ctl->DP, ctl, prio, stab);
475 	if (err < 0)
476 		goto errout_locked;
477 
478 	if (gred_rio_mode(table)) {
479 		gred_disable_wred_mode(table);
480 		if (gred_wred_mode_check(sch))
481 			gred_enable_wred_mode(table);
482 	}
483 
484 	err = 0;
485 
486 errout_locked:
487 	sch_tree_unlock(sch);
488 errout:
489 	return err;
490 }
491 
492 static int gred_init(struct Qdisc *sch, struct rtattr *opt)
493 {
494 	struct rtattr *tb[TCA_GRED_MAX];
495 
496 	if (opt == NULL || rtattr_parse_nested(tb, TCA_GRED_MAX, opt))
497 		return -EINVAL;
498 
499 	if (tb[TCA_GRED_PARMS-1] || tb[TCA_GRED_STAB-1])
500 		return -EINVAL;
501 
502 	return gred_change_table_def(sch, tb[TCA_GRED_DPS-1]);
503 }
504 
505 static int gred_dump(struct Qdisc *sch, struct sk_buff *skb)
506 {
507 	struct gred_sched *table = qdisc_priv(sch);
508 	struct rtattr *parms, *opts = NULL;
509 	int i;
510 	struct tc_gred_sopt sopt = {
511 		.DPs	= table->DPs,
512 		.def_DP	= table->def,
513 		.grio	= gred_rio_mode(table),
514 		.flags	= table->red_flags,
515 	};
516 
517 	opts = RTA_NEST(skb, TCA_OPTIONS);
518 	RTA_PUT(skb, TCA_GRED_DPS, sizeof(sopt), &sopt);
519 	parms = RTA_NEST(skb, TCA_GRED_PARMS);
520 
521 	for (i = 0; i < MAX_DPs; i++) {
522 		struct gred_sched_data *q = table->tab[i];
523 		struct tc_gred_qopt opt;
524 
525 		memset(&opt, 0, sizeof(opt));
526 
527 		if (!q) {
528 			/* hack -- fix at some point with proper message
529 			   This is how we indicate to tc that there is no VQ
530 			   at this DP */
531 
532 			opt.DP = MAX_DPs + i;
533 			goto append_opt;
534 		}
535 
536 		opt.limit	= q->limit;
537 		opt.DP		= q->DP;
538 		opt.backlog	= q->backlog;
539 		opt.prio	= q->prio;
540 		opt.qth_min	= q->parms.qth_min >> q->parms.Wlog;
541 		opt.qth_max	= q->parms.qth_max >> q->parms.Wlog;
542 		opt.Wlog	= q->parms.Wlog;
543 		opt.Plog	= q->parms.Plog;
544 		opt.Scell_log	= q->parms.Scell_log;
545 		opt.other	= q->stats.other;
546 		opt.early	= q->stats.prob_drop;
547 		opt.forced	= q->stats.forced_drop;
548 		opt.pdrop	= q->stats.pdrop;
549 		opt.packets	= q->packetsin;
550 		opt.bytesin	= q->bytesin;
551 
552 		if (gred_wred_mode(table)) {
553 			q->parms.qidlestart =
554 				table->tab[table->def]->parms.qidlestart;
555 			q->parms.qavg = table->tab[table->def]->parms.qavg;
556 		}
557 
558 		opt.qave = red_calc_qavg(&q->parms, q->parms.qavg);
559 
560 append_opt:
561 		RTA_APPEND(skb, sizeof(opt), &opt);
562 	}
563 
564 	RTA_NEST_END(skb, parms);
565 
566 	return RTA_NEST_END(skb, opts);
567 
568 rtattr_failure:
569 	return RTA_NEST_CANCEL(skb, opts);
570 }
571 
572 static void gred_destroy(struct Qdisc *sch)
573 {
574 	struct gred_sched *table = qdisc_priv(sch);
575 	int i;
576 
577 	for (i = 0; i < table->DPs; i++) {
578 		if (table->tab[i])
579 			gred_destroy_vq(table->tab[i]);
580 	}
581 }
582 
583 static struct Qdisc_ops gred_qdisc_ops = {
584 	.id		=	"gred",
585 	.priv_size	=	sizeof(struct gred_sched),
586 	.enqueue	=	gred_enqueue,
587 	.dequeue	=	gred_dequeue,
588 	.requeue	=	gred_requeue,
589 	.drop		=	gred_drop,
590 	.init		=	gred_init,
591 	.reset		=	gred_reset,
592 	.destroy	=	gred_destroy,
593 	.change		=	gred_change,
594 	.dump		=	gred_dump,
595 	.owner		=	THIS_MODULE,
596 };
597 
598 static int __init gred_module_init(void)
599 {
600 	return register_qdisc(&gred_qdisc_ops);
601 }
602 
603 static void __exit gred_module_exit(void)
604 {
605 	unregister_qdisc(&gred_qdisc_ops);
606 }
607 
608 module_init(gred_module_init)
609 module_exit(gred_module_exit)
610 
611 MODULE_LICENSE("GPL");
612