xref: /linux/net/sched/sch_codel.c (revision 95db3b255fde4e830e5f8cc011eb404023f669d4)
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 = __skb_dequeue(&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 	qdisc_drop(skb, sch);
86 }
87 
88 static struct sk_buff *codel_qdisc_dequeue(struct Qdisc *sch)
89 {
90 	struct codel_sched_data *q = qdisc_priv(sch);
91 	struct sk_buff *skb;
92 
93 	skb = codel_dequeue(sch, &sch->qstats.backlog, &q->params, &q->vars,
94 			    &q->stats, qdisc_pkt_len, codel_get_enqueue_time,
95 			    drop_func, dequeue_func);
96 
97 	/* We cant call qdisc_tree_reduce_backlog() if our qlen is 0,
98 	 * or HTB crashes. Defer it for next round.
99 	 */
100 	if (q->stats.drop_count && sch->q.qlen) {
101 		qdisc_tree_reduce_backlog(sch, q->stats.drop_count, q->stats.drop_len);
102 		q->stats.drop_count = 0;
103 		q->stats.drop_len = 0;
104 	}
105 	if (skb)
106 		qdisc_bstats_update(sch, skb);
107 	return skb;
108 }
109 
110 static int codel_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
111 {
112 	struct codel_sched_data *q;
113 
114 	if (likely(qdisc_qlen(sch) < sch->limit)) {
115 		codel_set_enqueue_time(skb);
116 		return qdisc_enqueue_tail(skb, sch);
117 	}
118 	q = qdisc_priv(sch);
119 	q->drop_overlimit++;
120 	return qdisc_drop(skb, sch);
121 }
122 
123 static const struct nla_policy codel_policy[TCA_CODEL_MAX + 1] = {
124 	[TCA_CODEL_TARGET]	= { .type = NLA_U32 },
125 	[TCA_CODEL_LIMIT]	= { .type = NLA_U32 },
126 	[TCA_CODEL_INTERVAL]	= { .type = NLA_U32 },
127 	[TCA_CODEL_ECN]		= { .type = NLA_U32 },
128 	[TCA_CODEL_CE_THRESHOLD]= { .type = NLA_U32 },
129 };
130 
131 static int codel_change(struct Qdisc *sch, struct nlattr *opt)
132 {
133 	struct codel_sched_data *q = qdisc_priv(sch);
134 	struct nlattr *tb[TCA_CODEL_MAX + 1];
135 	unsigned int qlen, dropped = 0;
136 	int err;
137 
138 	if (!opt)
139 		return -EINVAL;
140 
141 	err = nla_parse_nested(tb, TCA_CODEL_MAX, opt, codel_policy);
142 	if (err < 0)
143 		return err;
144 
145 	sch_tree_lock(sch);
146 
147 	if (tb[TCA_CODEL_TARGET]) {
148 		u32 target = nla_get_u32(tb[TCA_CODEL_TARGET]);
149 
150 		q->params.target = ((u64)target * NSEC_PER_USEC) >> CODEL_SHIFT;
151 	}
152 
153 	if (tb[TCA_CODEL_CE_THRESHOLD]) {
154 		u64 val = nla_get_u32(tb[TCA_CODEL_CE_THRESHOLD]);
155 
156 		q->params.ce_threshold = (val * NSEC_PER_USEC) >> CODEL_SHIFT;
157 	}
158 
159 	if (tb[TCA_CODEL_INTERVAL]) {
160 		u32 interval = nla_get_u32(tb[TCA_CODEL_INTERVAL]);
161 
162 		q->params.interval = ((u64)interval * NSEC_PER_USEC) >> CODEL_SHIFT;
163 	}
164 
165 	if (tb[TCA_CODEL_LIMIT])
166 		sch->limit = nla_get_u32(tb[TCA_CODEL_LIMIT]);
167 
168 	if (tb[TCA_CODEL_ECN])
169 		q->params.ecn = !!nla_get_u32(tb[TCA_CODEL_ECN]);
170 
171 	qlen = sch->q.qlen;
172 	while (sch->q.qlen > sch->limit) {
173 		struct sk_buff *skb = __skb_dequeue(&sch->q);
174 
175 		dropped += qdisc_pkt_len(skb);
176 		qdisc_qstats_backlog_dec(sch, skb);
177 		qdisc_drop(skb, sch);
178 	}
179 	qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);
180 
181 	sch_tree_unlock(sch);
182 	return 0;
183 }
184 
185 static int codel_init(struct Qdisc *sch, struct nlattr *opt)
186 {
187 	struct codel_sched_data *q = qdisc_priv(sch);
188 
189 	sch->limit = DEFAULT_CODEL_LIMIT;
190 
191 	codel_params_init(&q->params);
192 	codel_vars_init(&q->vars);
193 	codel_stats_init(&q->stats);
194 	q->params.mtu = psched_mtu(qdisc_dev(sch));
195 
196 	if (opt) {
197 		int err = codel_change(sch, opt);
198 
199 		if (err)
200 			return err;
201 	}
202 
203 	if (sch->limit >= 1)
204 		sch->flags |= TCQ_F_CAN_BYPASS;
205 	else
206 		sch->flags &= ~TCQ_F_CAN_BYPASS;
207 
208 	return 0;
209 }
210 
211 static int codel_dump(struct Qdisc *sch, struct sk_buff *skb)
212 {
213 	struct codel_sched_data *q = qdisc_priv(sch);
214 	struct nlattr *opts;
215 
216 	opts = nla_nest_start(skb, TCA_OPTIONS);
217 	if (opts == NULL)
218 		goto nla_put_failure;
219 
220 	if (nla_put_u32(skb, TCA_CODEL_TARGET,
221 			codel_time_to_us(q->params.target)) ||
222 	    nla_put_u32(skb, TCA_CODEL_LIMIT,
223 			sch->limit) ||
224 	    nla_put_u32(skb, TCA_CODEL_INTERVAL,
225 			codel_time_to_us(q->params.interval)) ||
226 	    nla_put_u32(skb, TCA_CODEL_ECN,
227 			q->params.ecn))
228 		goto nla_put_failure;
229 	if (q->params.ce_threshold != CODEL_DISABLED_THRESHOLD &&
230 	    nla_put_u32(skb, TCA_CODEL_CE_THRESHOLD,
231 			codel_time_to_us(q->params.ce_threshold)))
232 		goto nla_put_failure;
233 	return nla_nest_end(skb, opts);
234 
235 nla_put_failure:
236 	nla_nest_cancel(skb, opts);
237 	return -1;
238 }
239 
240 static int codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
241 {
242 	const struct codel_sched_data *q = qdisc_priv(sch);
243 	struct tc_codel_xstats st = {
244 		.maxpacket	= q->stats.maxpacket,
245 		.count		= q->vars.count,
246 		.lastcount	= q->vars.lastcount,
247 		.drop_overlimit = q->drop_overlimit,
248 		.ldelay		= codel_time_to_us(q->vars.ldelay),
249 		.dropping	= q->vars.dropping,
250 		.ecn_mark	= q->stats.ecn_mark,
251 		.ce_mark	= q->stats.ce_mark,
252 	};
253 
254 	if (q->vars.dropping) {
255 		codel_tdiff_t delta = q->vars.drop_next - codel_get_time();
256 
257 		if (delta >= 0)
258 			st.drop_next = codel_time_to_us(delta);
259 		else
260 			st.drop_next = -codel_time_to_us(-delta);
261 	}
262 
263 	return gnet_stats_copy_app(d, &st, sizeof(st));
264 }
265 
266 static void codel_reset(struct Qdisc *sch)
267 {
268 	struct codel_sched_data *q = qdisc_priv(sch);
269 
270 	qdisc_reset_queue(sch);
271 	codel_vars_init(&q->vars);
272 }
273 
274 static struct Qdisc_ops codel_qdisc_ops __read_mostly = {
275 	.id		=	"codel",
276 	.priv_size	=	sizeof(struct codel_sched_data),
277 
278 	.enqueue	=	codel_qdisc_enqueue,
279 	.dequeue	=	codel_qdisc_dequeue,
280 	.peek		=	qdisc_peek_dequeued,
281 	.init		=	codel_init,
282 	.reset		=	codel_reset,
283 	.change 	=	codel_change,
284 	.dump		=	codel_dump,
285 	.dump_stats	=	codel_dump_stats,
286 	.owner		=	THIS_MODULE,
287 };
288 
289 static int __init codel_module_init(void)
290 {
291 	return register_qdisc(&codel_qdisc_ops);
292 }
293 
294 static void __exit codel_module_exit(void)
295 {
296 	unregister_qdisc(&codel_qdisc_ops);
297 }
298 
299 module_init(codel_module_init)
300 module_exit(codel_module_exit)
301 
302 MODULE_DESCRIPTION("Controlled Delay queue discipline");
303 MODULE_AUTHOR("Dave Taht");
304 MODULE_AUTHOR("Eric Dumazet");
305 MODULE_LICENSE("Dual BSD/GPL");
306