xref: /linux/net/sched/sch_codel.c (revision 2fe05e1139a555ae91f00a812cb9520e7d3022ab)
1 /*
2  * Codel - The Controlled-Delay Active Queue Management algorithm
3  *
4  *  Copyright (C) 2011-2012 Kathleen Nichols <nichols@pollere.com>
5  *  Copyright (C) 2011-2012 Van Jacobson <van@pollere.net>
6  *
7  *  Implemented on linux by :
8  *  Copyright (C) 2012 Michael D. Taht <dave.taht@bufferbloat.net>
9  *  Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com>
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions, and the following disclaimer,
16  *    without modification.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. The names of the authors may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * Alternatively, provided that this notice is retained in full, this
24  * software may be distributed under the terms of the GNU General
25  * Public License ("GPL") version 2, in which case the provisions of the
26  * GPL apply INSTEAD OF those given above.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
39  * DAMAGE.
40  *
41  */
42 
43 #include <linux/module.h>
44 #include <linux/slab.h>
45 #include <linux/types.h>
46 #include <linux/kernel.h>
47 #include <linux/errno.h>
48 #include <linux/skbuff.h>
49 #include <linux/prefetch.h>
50 #include <net/pkt_sched.h>
51 #include <net/codel.h>
52 #include <net/codel_impl.h>
53 #include <net/codel_qdisc.h>
54 
55 
56 #define DEFAULT_CODEL_LIMIT 1000
57 
58 struct codel_sched_data {
59 	struct codel_params	params;
60 	struct codel_vars	vars;
61 	struct codel_stats	stats;
62 	u32			drop_overlimit;
63 };
64 
65 /* This is the specific function called from codel_dequeue()
66  * to dequeue a packet from queue. Note: backlog is handled in
67  * codel, we dont need to reduce it here.
68  */
69 static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx)
70 {
71 	struct Qdisc *sch = ctx;
72 	struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
73 
74 	if (skb)
75 		sch->qstats.backlog -= qdisc_pkt_len(skb);
76 
77 	prefetch(&skb->end); /* we'll need skb_shinfo() */
78 	return skb;
79 }
80 
81 static void drop_func(struct sk_buff *skb, void *ctx)
82 {
83 	struct Qdisc *sch = ctx;
84 
85 	kfree_skb(skb);
86 	qdisc_qstats_drop(sch);
87 }
88 
89 static struct sk_buff *codel_qdisc_dequeue(struct Qdisc *sch)
90 {
91 	struct codel_sched_data *q = qdisc_priv(sch);
92 	struct sk_buff *skb;
93 
94 	skb = codel_dequeue(sch, &sch->qstats.backlog, &q->params, &q->vars,
95 			    &q->stats, qdisc_pkt_len, codel_get_enqueue_time,
96 			    drop_func, dequeue_func);
97 
98 	/* We cant call qdisc_tree_reduce_backlog() if our qlen is 0,
99 	 * or HTB crashes. Defer it for next round.
100 	 */
101 	if (q->stats.drop_count && sch->q.qlen) {
102 		qdisc_tree_reduce_backlog(sch, q->stats.drop_count, q->stats.drop_len);
103 		q->stats.drop_count = 0;
104 		q->stats.drop_len = 0;
105 	}
106 	if (skb)
107 		qdisc_bstats_update(sch, skb);
108 	return skb;
109 }
110 
111 static int codel_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
112 			       struct sk_buff **to_free)
113 {
114 	struct codel_sched_data *q;
115 
116 	if (likely(qdisc_qlen(sch) < sch->limit)) {
117 		codel_set_enqueue_time(skb);
118 		return qdisc_enqueue_tail(skb, sch);
119 	}
120 	q = qdisc_priv(sch);
121 	q->drop_overlimit++;
122 	return qdisc_drop(skb, sch, to_free);
123 }
124 
125 static const struct nla_policy codel_policy[TCA_CODEL_MAX + 1] = {
126 	[TCA_CODEL_TARGET]	= { .type = NLA_U32 },
127 	[TCA_CODEL_LIMIT]	= { .type = NLA_U32 },
128 	[TCA_CODEL_INTERVAL]	= { .type = NLA_U32 },
129 	[TCA_CODEL_ECN]		= { .type = NLA_U32 },
130 	[TCA_CODEL_CE_THRESHOLD]= { .type = NLA_U32 },
131 };
132 
133 static int codel_change(struct Qdisc *sch, struct nlattr *opt)
134 {
135 	struct codel_sched_data *q = qdisc_priv(sch);
136 	struct nlattr *tb[TCA_CODEL_MAX + 1];
137 	unsigned int qlen, dropped = 0;
138 	int err;
139 
140 	if (!opt)
141 		return -EINVAL;
142 
143 	err = nla_parse_nested(tb, TCA_CODEL_MAX, opt, codel_policy, NULL);
144 	if (err < 0)
145 		return err;
146 
147 	sch_tree_lock(sch);
148 
149 	if (tb[TCA_CODEL_TARGET]) {
150 		u32 target = nla_get_u32(tb[TCA_CODEL_TARGET]);
151 
152 		q->params.target = ((u64)target * NSEC_PER_USEC) >> CODEL_SHIFT;
153 	}
154 
155 	if (tb[TCA_CODEL_CE_THRESHOLD]) {
156 		u64 val = nla_get_u32(tb[TCA_CODEL_CE_THRESHOLD]);
157 
158 		q->params.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT;
159 	}
160 
161 	if (tb[TCA_CODEL_INTERVAL]) {
162 		u32 interval = nla_get_u32(tb[TCA_CODEL_INTERVAL]);
163 
164 		q->params.interval = ((u64)interval * NSEC_PER_USEC) >> CODEL_SHIFT;
165 	}
166 
167 	if (tb[TCA_CODEL_LIMIT])
168 		sch->limit = nla_get_u32(tb[TCA_CODEL_LIMIT]);
169 
170 	if (tb[TCA_CODEL_ECN])
171 		q->params.ecn = !!nla_get_u32(tb[TCA_CODEL_ECN]);
172 
173 	qlen = sch->q.qlen;
174 	while (sch->q.qlen > sch->limit) {
175 		struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
176 
177 		dropped += qdisc_pkt_len(skb);
178 		qdisc_qstats_backlog_dec(sch, skb);
179 		rtnl_qdisc_drop(skb, sch);
180 	}
181 	qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);
182 
183 	sch_tree_unlock(sch);
184 	return 0;
185 }
186 
187 static int codel_init(struct Qdisc *sch, struct nlattr *opt)
188 {
189 	struct codel_sched_data *q = qdisc_priv(sch);
190 
191 	sch->limit = DEFAULT_CODEL_LIMIT;
192 
193 	codel_params_init(&q->params);
194 	codel_vars_init(&q->vars);
195 	codel_stats_init(&q->stats);
196 	q->params.mtu = psched_mtu(qdisc_dev(sch));
197 
198 	if (opt) {
199 		int err = codel_change(sch, opt);
200 
201 		if (err)
202 			return err;
203 	}
204 
205 	if (sch->limit >= 1)
206 		sch->flags |= TCQ_F_CAN_BYPASS;
207 	else
208 		sch->flags &= ~TCQ_F_CAN_BYPASS;
209 
210 	return 0;
211 }
212 
213 static int codel_dump(struct Qdisc *sch, struct sk_buff *skb)
214 {
215 	struct codel_sched_data *q = qdisc_priv(sch);
216 	struct nlattr *opts;
217 
218 	opts = nla_nest_start(skb, TCA_OPTIONS);
219 	if (opts == NULL)
220 		goto nla_put_failure;
221 
222 	if (nla_put_u32(skb, TCA_CODEL_TARGET,
223 			codel_time_to_us(q->params.target)) ||
224 	    nla_put_u32(skb, TCA_CODEL_LIMIT,
225 			sch->limit) ||
226 	    nla_put_u32(skb, TCA_CODEL_INTERVAL,
227 			codel_time_to_us(q->params.interval)) ||
228 	    nla_put_u32(skb, TCA_CODEL_ECN,
229 			q->params.ecn))
230 		goto nla_put_failure;
231 	if (q->params.ce_threshold != CODEL_DISABLED_THRESHOLD &&
232 	    nla_put_u32(skb, TCA_CODEL_CE_THRESHOLD,
233 			codel_time_to_us(q->params.ce_threshold)))
234 		goto nla_put_failure;
235 	return nla_nest_end(skb, opts);
236 
237 nla_put_failure:
238 	nla_nest_cancel(skb, opts);
239 	return -1;
240 }
241 
242 static int codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
243 {
244 	const struct codel_sched_data *q = qdisc_priv(sch);
245 	struct tc_codel_xstats st = {
246 		.maxpacket	= q->stats.maxpacket,
247 		.count		= q->vars.count,
248 		.lastcount	= q->vars.lastcount,
249 		.drop_overlimit = q->drop_overlimit,
250 		.ldelay		= codel_time_to_us(q->vars.ldelay),
251 		.dropping	= q->vars.dropping,
252 		.ecn_mark	= q->stats.ecn_mark,
253 		.ce_mark	= q->stats.ce_mark,
254 	};
255 
256 	if (q->vars.dropping) {
257 		codel_tdiff_t delta = q->vars.drop_next - codel_get_time();
258 
259 		if (delta >= 0)
260 			st.drop_next = codel_time_to_us(delta);
261 		else
262 			st.drop_next = -codel_time_to_us(-delta);
263 	}
264 
265 	return gnet_stats_copy_app(d, &st, sizeof(st));
266 }
267 
268 static void codel_reset(struct Qdisc *sch)
269 {
270 	struct codel_sched_data *q = qdisc_priv(sch);
271 
272 	qdisc_reset_queue(sch);
273 	codel_vars_init(&q->vars);
274 }
275 
276 static struct Qdisc_ops codel_qdisc_ops __read_mostly = {
277 	.id		=	"codel",
278 	.priv_size	=	sizeof(struct codel_sched_data),
279 
280 	.enqueue	=	codel_qdisc_enqueue,
281 	.dequeue	=	codel_qdisc_dequeue,
282 	.peek		=	qdisc_peek_dequeued,
283 	.init		=	codel_init,
284 	.reset		=	codel_reset,
285 	.change 	=	codel_change,
286 	.dump		=	codel_dump,
287 	.dump_stats	=	codel_dump_stats,
288 	.owner		=	THIS_MODULE,
289 };
290 
291 static int __init codel_module_init(void)
292 {
293 	return register_qdisc(&codel_qdisc_ops);
294 }
295 
296 static void __exit codel_module_exit(void)
297 {
298 	unregister_qdisc(&codel_qdisc_ops);
299 }
300 
301 module_init(codel_module_init)
302 module_exit(codel_module_exit)
303 
304 MODULE_DESCRIPTION("Controlled Delay queue discipline");
305 MODULE_AUTHOR("Dave Taht");
306 MODULE_AUTHOR("Eric Dumazet");
307 MODULE_LICENSE("Dual BSD/GPL");
308