xref: /openbmc/linux/net/sched/sch_red.c (revision 8ee90c5c) 
1 /*
2  * net/sched/sch_red.c	Random Early Detection queue.
3  *
4  *		This program is free software; you can redistribute it and/or
5  *		modify it under the terms of the GNU General Public License
6  *		as published by the Free Software Foundation; either version
7  *		2 of the License, or (at your option) any later version.
8  *
9  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  *
11  * Changes:
12  * J Hadi Salim 980914:	computation fixes
13  * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
14  * J Hadi Salim 980816:  ECN support
15  */
16 
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/skbuff.h>
21 #include <net/pkt_sched.h>
22 #include <net/inet_ecn.h>
23 #include <net/red.h>
24 
25 
26 /*	Parameters, settable by user:
27 	-----------------------------
28 
29 	limit		- bytes (must be > qth_max + burst)
30 
31 	Hard limit on queue length, should be chosen >qth_max
32 	to allow packet bursts. This parameter does not
33 	affect the algorithms behaviour and can be chosen
34 	arbitrarily high (well, less than ram size)
35 	Really, this limit will never be reached
36 	if RED works correctly.
37  */
38 
39 struct red_sched_data {
40 	u32			limit;		/* HARD maximal queue length */
41 	unsigned char		flags;
42 	struct timer_list	adapt_timer;
43 	struct red_parms	parms;
44 	struct red_vars		vars;
45 	struct red_stats	stats;
46 	struct Qdisc		*qdisc;
47 };
48 
49 static inline int red_use_ecn(struct red_sched_data *q)
50 {
51 	return q->flags & TC_RED_ECN;
52 }
53 
54 static inline int red_use_harddrop(struct red_sched_data *q)
55 {
56 	return q->flags & TC_RED_HARDDROP;
57 }
58 
59 static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch,
60 		       struct sk_buff **to_free)
61 {
62 	struct red_sched_data *q = qdisc_priv(sch);
63 	struct Qdisc *child = q->qdisc;
64 	int ret;
65 
66 	q->vars.qavg = red_calc_qavg(&q->parms,
67 				     &q->vars,
68 				     child->qstats.backlog);
69 
70 	if (red_is_idling(&q->vars))
71 		red_end_of_idle_period(&q->vars);
72 
73 	switch (red_action(&q->parms, &q->vars, q->vars.qavg)) {
74 	case RED_DONT_MARK:
75 		break;
76 
77 	case RED_PROB_MARK:
78 		qdisc_qstats_overlimit(sch);
79 		if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) {
80 			q->stats.prob_drop++;
81 			goto congestion_drop;
82 		}
83 
84 		q->stats.prob_mark++;
85 		break;
86 
87 	case RED_HARD_MARK:
88 		qdisc_qstats_overlimit(sch);
89 		if (red_use_harddrop(q) || !red_use_ecn(q) ||
90 		    !INET_ECN_set_ce(skb)) {
91 			q->stats.forced_drop++;
92 			goto congestion_drop;
93 		}
94 
95 		q->stats.forced_mark++;
96 		break;
97 	}
98 
99 	ret = qdisc_enqueue(skb, child, to_free);
100 	if (likely(ret == NET_XMIT_SUCCESS)) {
101 		qdisc_qstats_backlog_inc(sch, skb);
102 		sch->q.qlen++;
103 	} else if (net_xmit_drop_count(ret)) {
104 		q->stats.pdrop++;
105 		qdisc_qstats_drop(sch);
106 	}
107 	return ret;
108 
109 congestion_drop:
110 	qdisc_drop(skb, sch, to_free);
111 	return NET_XMIT_CN;
112 }
113 
114 static struct sk_buff *red_dequeue(struct Qdisc *sch)
115 {
116 	struct sk_buff *skb;
117 	struct red_sched_data *q = qdisc_priv(sch);
118 	struct Qdisc *child = q->qdisc;
119 
120 	skb = child->dequeue(child);
121 	if (skb) {
122 		qdisc_bstats_update(sch, skb);
123 		qdisc_qstats_backlog_dec(sch, skb);
124 		sch->q.qlen--;
125 	} else {
126 		if (!red_is_idling(&q->vars))
127 			red_start_of_idle_period(&q->vars);
128 	}
129 	return skb;
130 }
131 
132 static struct sk_buff *red_peek(struct Qdisc *sch)
133 {
134 	struct red_sched_data *q = qdisc_priv(sch);
135 	struct Qdisc *child = q->qdisc;
136 
137 	return child->ops->peek(child);
138 }
139 
140 static void red_reset(struct Qdisc *sch)
141 {
142 	struct red_sched_data *q = qdisc_priv(sch);
143 
144 	qdisc_reset(q->qdisc);
145 	sch->qstats.backlog = 0;
146 	sch->q.qlen = 0;
147 	red_restart(&q->vars);
148 }
149 
150 static void red_destroy(struct Qdisc *sch)
151 {
152 	struct red_sched_data *q = qdisc_priv(sch);
153 
154 	del_timer_sync(&q->adapt_timer);
155 	qdisc_destroy(q->qdisc);
156 }
157 
158 static const struct nla_policy red_policy[TCA_RED_MAX + 1] = {
159 	[TCA_RED_PARMS]	= { .len = sizeof(struct tc_red_qopt) },
160 	[TCA_RED_STAB]	= { .len = RED_STAB_SIZE },
161 	[TCA_RED_MAX_P] = { .type = NLA_U32 },
162 };
163 
164 static int red_change(struct Qdisc *sch, struct nlattr *opt)
165 {
166 	struct red_sched_data *q = qdisc_priv(sch);
167 	struct nlattr *tb[TCA_RED_MAX + 1];
168 	struct tc_red_qopt *ctl;
169 	struct Qdisc *child = NULL;
170 	int err;
171 	u32 max_P;
172 
173 	if (opt == NULL)
174 		return -EINVAL;
175 
176 	err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy, NULL);
177 	if (err < 0)
178 		return err;
179 
180 	if (tb[TCA_RED_PARMS] == NULL ||
181 	    tb[TCA_RED_STAB] == NULL)
182 		return -EINVAL;
183 
184 	max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0;
185 
186 	ctl = nla_data(tb[TCA_RED_PARMS]);
187 
188 	if (ctl->limit > 0) {
189 		child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit);
190 		if (IS_ERR(child))
191 			return PTR_ERR(child);
192 	}
193 
194 	if (child != &noop_qdisc)
195 		qdisc_hash_add(child, true);
196 	sch_tree_lock(sch);
197 	q->flags = ctl->flags;
198 	q->limit = ctl->limit;
199 	if (child) {
200 		qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen,
201 					  q->qdisc->qstats.backlog);
202 		qdisc_destroy(q->qdisc);
203 		q->qdisc = child;
204 	}
205 
206 	red_set_parms(&q->parms,
207 		      ctl->qth_min, ctl->qth_max, ctl->Wlog,
208 		      ctl->Plog, ctl->Scell_log,
209 		      nla_data(tb[TCA_RED_STAB]),
210 		      max_P);
211 	red_set_vars(&q->vars);
212 
213 	del_timer(&q->adapt_timer);
214 	if (ctl->flags & TC_RED_ADAPTATIVE)
215 		mod_timer(&q->adapt_timer, jiffies + HZ/2);
216 
217 	if (!q->qdisc->q.qlen)
218 		red_start_of_idle_period(&q->vars);
219 
220 	sch_tree_unlock(sch);
221 	return 0;
222 }
223 
224 static inline void red_adaptative_timer(unsigned long arg)
225 {
226 	struct Qdisc *sch = (struct Qdisc *)arg;
227 	struct red_sched_data *q = qdisc_priv(sch);
228 	spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
229 
230 	spin_lock(root_lock);
231 	red_adaptative_algo(&q->parms, &q->vars);
232 	mod_timer(&q->adapt_timer, jiffies + HZ/2);
233 	spin_unlock(root_lock);
234 }
235 
236 static int red_init(struct Qdisc *sch, struct nlattr *opt)
237 {
238 	struct red_sched_data *q = qdisc_priv(sch);
239 
240 	q->qdisc = &noop_qdisc;
241 	setup_timer(&q->adapt_timer, red_adaptative_timer, (unsigned long)sch);
242 	return red_change(sch, opt);
243 }
244 
245 static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
246 {
247 	struct red_sched_data *q = qdisc_priv(sch);
248 	struct nlattr *opts = NULL;
249 	struct tc_red_qopt opt = {
250 		.limit		= q->limit,
251 		.flags		= q->flags,
252 		.qth_min	= q->parms.qth_min >> q->parms.Wlog,
253 		.qth_max	= q->parms.qth_max >> q->parms.Wlog,
254 		.Wlog		= q->parms.Wlog,
255 		.Plog		= q->parms.Plog,
256 		.Scell_log	= q->parms.Scell_log,
257 	};
258 
259 	sch->qstats.backlog = q->qdisc->qstats.backlog;
260 	opts = nla_nest_start(skb, TCA_OPTIONS);
261 	if (opts == NULL)
262 		goto nla_put_failure;
263 	if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) ||
264 	    nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P))
265 		goto nla_put_failure;
266 	return nla_nest_end(skb, opts);
267 
268 nla_put_failure:
269 	nla_nest_cancel(skb, opts);
270 	return -EMSGSIZE;
271 }
272 
273 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
274 {
275 	struct red_sched_data *q = qdisc_priv(sch);
276 	struct tc_red_xstats st = {
277 		.early	= q->stats.prob_drop + q->stats.forced_drop,
278 		.pdrop	= q->stats.pdrop,
279 		.other	= q->stats.other,
280 		.marked	= q->stats.prob_mark + q->stats.forced_mark,
281 	};
282 
283 	return gnet_stats_copy_app(d, &st, sizeof(st));
284 }
285 
286 static int red_dump_class(struct Qdisc *sch, unsigned long cl,
287 			  struct sk_buff *skb, struct tcmsg *tcm)
288 {
289 	struct red_sched_data *q = qdisc_priv(sch);
290 
291 	tcm->tcm_handle |= TC_H_MIN(1);
292 	tcm->tcm_info = q->qdisc->handle;
293 	return 0;
294 }
295 
296 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
297 		     struct Qdisc **old)
298 {
299 	struct red_sched_data *q = qdisc_priv(sch);
300 
301 	if (new == NULL)
302 		new = &noop_qdisc;
303 
304 	*old = qdisc_replace(sch, new, &q->qdisc);
305 	return 0;
306 }
307 
308 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg)
309 {
310 	struct red_sched_data *q = qdisc_priv(sch);
311 	return q->qdisc;
312 }
313 
314 static unsigned long red_find(struct Qdisc *sch, u32 classid)
315 {
316 	return 1;
317 }
318 
319 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker)
320 {
321 	if (!walker->stop) {
322 		if (walker->count >= walker->skip)
323 			if (walker->fn(sch, 1, walker) < 0) {
324 				walker->stop = 1;
325 				return;
326 			}
327 		walker->count++;
328 	}
329 }
330 
331 static const struct Qdisc_class_ops red_class_ops = {
332 	.graft		=	red_graft,
333 	.leaf		=	red_leaf,
334 	.find		=	red_find,
335 	.walk		=	red_walk,
336 	.dump		=	red_dump_class,
337 };
338 
339 static struct Qdisc_ops red_qdisc_ops __read_mostly = {
340 	.id		=	"red",
341 	.priv_size	=	sizeof(struct red_sched_data),
342 	.cl_ops		=	&red_class_ops,
343 	.enqueue	=	red_enqueue,
344 	.dequeue	=	red_dequeue,
345 	.peek		=	red_peek,
346 	.init		=	red_init,
347 	.reset		=	red_reset,
348 	.destroy	=	red_destroy,
349 	.change		=	red_change,
350 	.dump		=	red_dump,
351 	.dump_stats	=	red_dump_stats,
352 	.owner		=	THIS_MODULE,
353 };
354 
355 static int __init red_module_init(void)
356 {
357 	return register_qdisc(&red_qdisc_ops);
358 }
359 
360 static void __exit red_module_exit(void)
361 {
362 	unregister_qdisc(&red_qdisc_ops);
363 }
364 
365 module_init(red_module_init)
366 module_exit(red_module_exit)
367 
368 MODULE_LICENSE("GPL");
369