xref: /linux/net/sched/sch_codel.c (revision e814f3fd16acfb7f9966773953de8f740a1e3202)
1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /*
3  * Codel - The Controlled-Delay Active Queue Management algorithm
4  *
5  *  Copyright (C) 2011-2012 Kathleen Nichols <nichols@pollere.com>
6  *  Copyright (C) 2011-2012 Van Jacobson <van@pollere.net>
7  *
8  *  Implemented on linux by :
9  *  Copyright (C) 2012 Michael D. Taht <dave.taht@bufferbloat.net>
10  *  Copyright (C) 2012,2015 Eric Dumazet <edumazet@google.com>
11  */
12 
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/errno.h>
18 #include <linux/skbuff.h>
19 #include <linux/prefetch.h>
20 #include <net/pkt_sched.h>
21 #include <net/codel.h>
22 #include <net/codel_impl.h>
23 #include <net/codel_qdisc.h>
24 
25 
26 #define DEFAULT_CODEL_LIMIT 1000
27 
28 struct codel_sched_data {
29 	struct codel_params	params;
30 	struct codel_vars	vars;
31 	struct codel_stats	stats;
32 	u32			drop_overlimit;
33 };
34 
35 /* This is the specific function called from codel_dequeue()
36  * to dequeue a packet from queue. Note: backlog is handled in
37  * codel, we dont need to reduce it here.
38  */
39 static struct sk_buff *dequeue_func(struct codel_vars *vars, void *ctx)
40 {
41 	struct Qdisc *sch = ctx;
42 	struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
43 
44 	if (skb) {
45 		sch->qstats.backlog -= qdisc_pkt_len(skb);
46 		prefetch(&skb->end); /* we'll need skb_shinfo() */
47 	}
48 	return skb;
49 }
50 
51 static void drop_func(struct sk_buff *skb, void *ctx)
52 {
53 	struct Qdisc *sch = ctx;
54 
55 	kfree_skb_reason(skb, SKB_DROP_REASON_QDISC_CONGESTED);
56 	qdisc_qstats_drop(sch);
57 }
58 
59 static struct sk_buff *codel_qdisc_dequeue(struct Qdisc *sch)
60 {
61 	struct codel_sched_data *q = qdisc_priv(sch);
62 	struct sk_buff *skb;
63 
64 	skb = codel_dequeue(sch, &sch->qstats.backlog, &q->params, &q->vars,
65 			    &q->stats, qdisc_pkt_len, codel_get_enqueue_time,
66 			    drop_func, dequeue_func);
67 
68 	/* We cant call qdisc_tree_reduce_backlog() if our qlen is 0,
69 	 * or HTB crashes. Defer it for next round.
70 	 */
71 	if (q->stats.drop_count && sch->q.qlen) {
72 		qdisc_tree_reduce_backlog(sch, q->stats.drop_count, q->stats.drop_len);
73 		q->stats.drop_count = 0;
74 		q->stats.drop_len = 0;
75 	}
76 	if (skb)
77 		qdisc_bstats_update(sch, skb);
78 	return skb;
79 }
80 
81 static int codel_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
82 			       struct sk_buff **to_free)
83 {
84 	struct codel_sched_data *q;
85 
86 	if (likely(qdisc_qlen(sch) < sch->limit)) {
87 		codel_set_enqueue_time(skb);
88 		return qdisc_enqueue_tail(skb, sch);
89 	}
90 	q = qdisc_priv(sch);
91 	q->drop_overlimit++;
92 	return qdisc_drop_reason(skb, sch, to_free,
93 				 SKB_DROP_REASON_QDISC_OVERLIMIT);
94 }
95 
96 static const struct nla_policy codel_policy[TCA_CODEL_MAX + 1] = {
97 	[TCA_CODEL_TARGET]	= { .type = NLA_U32 },
98 	[TCA_CODEL_LIMIT]	= { .type = NLA_U32 },
99 	[TCA_CODEL_INTERVAL]	= { .type = NLA_U32 },
100 	[TCA_CODEL_ECN]		= { .type = NLA_U32 },
101 	[TCA_CODEL_CE_THRESHOLD]= { .type = NLA_U32 },
102 };
103 
104 static int codel_change(struct Qdisc *sch, struct nlattr *opt,
105 			struct netlink_ext_ack *extack)
106 {
107 	struct codel_sched_data *q = qdisc_priv(sch);
108 	struct nlattr *tb[TCA_CODEL_MAX + 1];
109 	unsigned int qlen, dropped = 0;
110 	int err;
111 
112 	err = nla_parse_nested_deprecated(tb, TCA_CODEL_MAX, opt,
113 					  codel_policy, NULL);
114 	if (err < 0)
115 		return err;
116 
117 	sch_tree_lock(sch);
118 
119 	if (tb[TCA_CODEL_TARGET]) {
120 		u32 target = nla_get_u32(tb[TCA_CODEL_TARGET]);
121 
122 		WRITE_ONCE(q->params.target,
123 			   ((u64)target * NSEC_PER_USEC) >> CODEL_SHIFT);
124 	}
125 
126 	if (tb[TCA_CODEL_CE_THRESHOLD]) {
127 		u64 val = nla_get_u32(tb[TCA_CODEL_CE_THRESHOLD]);
128 
129 		WRITE_ONCE(q->params.ce_threshold,
130 			   (val * NSEC_PER_USEC) >> CODEL_SHIFT);
131 	}
132 
133 	if (tb[TCA_CODEL_INTERVAL]) {
134 		u32 interval = nla_get_u32(tb[TCA_CODEL_INTERVAL]);
135 
136 		WRITE_ONCE(q->params.interval,
137 			   ((u64)interval * NSEC_PER_USEC) >> CODEL_SHIFT);
138 	}
139 
140 	if (tb[TCA_CODEL_LIMIT])
141 		WRITE_ONCE(sch->limit,
142 			   nla_get_u32(tb[TCA_CODEL_LIMIT]));
143 
144 	if (tb[TCA_CODEL_ECN])
145 		WRITE_ONCE(q->params.ecn,
146 			   !!nla_get_u32(tb[TCA_CODEL_ECN]));
147 
148 	qlen = sch->q.qlen;
149 	while (sch->q.qlen > sch->limit) {
150 		struct sk_buff *skb = __qdisc_dequeue_head(&sch->q);
151 
152 		dropped += qdisc_pkt_len(skb);
153 		qdisc_qstats_backlog_dec(sch, skb);
154 		rtnl_qdisc_drop(skb, sch);
155 	}
156 	qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);
157 
158 	sch_tree_unlock(sch);
159 	return 0;
160 }
161 
162 static int codel_init(struct Qdisc *sch, struct nlattr *opt,
163 		      struct netlink_ext_ack *extack)
164 {
165 	struct codel_sched_data *q = qdisc_priv(sch);
166 
167 	sch->limit = DEFAULT_CODEL_LIMIT;
168 
169 	codel_params_init(&q->params);
170 	codel_vars_init(&q->vars);
171 	codel_stats_init(&q->stats);
172 	q->params.mtu = psched_mtu(qdisc_dev(sch));
173 
174 	if (opt) {
175 		int err = codel_change(sch, opt, extack);
176 
177 		if (err)
178 			return err;
179 	}
180 
181 	if (sch->limit >= 1)
182 		sch->flags |= TCQ_F_CAN_BYPASS;
183 	else
184 		sch->flags &= ~TCQ_F_CAN_BYPASS;
185 
186 	return 0;
187 }
188 
189 static int codel_dump(struct Qdisc *sch, struct sk_buff *skb)
190 {
191 	struct codel_sched_data *q = qdisc_priv(sch);
192 	codel_time_t ce_threshold;
193 	struct nlattr *opts;
194 
195 	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
196 	if (opts == NULL)
197 		goto nla_put_failure;
198 
199 	if (nla_put_u32(skb, TCA_CODEL_TARGET,
200 			codel_time_to_us(READ_ONCE(q->params.target))) ||
201 	    nla_put_u32(skb, TCA_CODEL_LIMIT,
202 			READ_ONCE(sch->limit)) ||
203 	    nla_put_u32(skb, TCA_CODEL_INTERVAL,
204 			codel_time_to_us(READ_ONCE(q->params.interval))) ||
205 	    nla_put_u32(skb, TCA_CODEL_ECN,
206 			READ_ONCE(q->params.ecn)))
207 		goto nla_put_failure;
208 	ce_threshold = READ_ONCE(q->params.ce_threshold);
209 	if (ce_threshold != CODEL_DISABLED_THRESHOLD &&
210 	    nla_put_u32(skb, TCA_CODEL_CE_THRESHOLD,
211 			codel_time_to_us(ce_threshold)))
212 		goto nla_put_failure;
213 	return nla_nest_end(skb, opts);
214 
215 nla_put_failure:
216 	nla_nest_cancel(skb, opts);
217 	return -1;
218 }
219 
220 static int codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
221 {
222 	const struct codel_sched_data *q = qdisc_priv(sch);
223 	struct tc_codel_xstats st = {
224 		.maxpacket	= q->stats.maxpacket,
225 		.count		= q->vars.count,
226 		.lastcount	= q->vars.lastcount,
227 		.drop_overlimit = q->drop_overlimit,
228 		.ldelay		= codel_time_to_us(q->vars.ldelay),
229 		.dropping	= q->vars.dropping,
230 		.ecn_mark	= q->stats.ecn_mark,
231 		.ce_mark	= q->stats.ce_mark,
232 	};
233 
234 	if (q->vars.dropping) {
235 		codel_tdiff_t delta = q->vars.drop_next - codel_get_time();
236 
237 		if (delta >= 0)
238 			st.drop_next = codel_time_to_us(delta);
239 		else
240 			st.drop_next = -codel_time_to_us(-delta);
241 	}
242 
243 	return gnet_stats_copy_app(d, &st, sizeof(st));
244 }
245 
246 static void codel_reset(struct Qdisc *sch)
247 {
248 	struct codel_sched_data *q = qdisc_priv(sch);
249 
250 	qdisc_reset_queue(sch);
251 	codel_vars_init(&q->vars);
252 }
253 
254 static struct Qdisc_ops codel_qdisc_ops __read_mostly = {
255 	.id		=	"codel",
256 	.priv_size	=	sizeof(struct codel_sched_data),
257 
258 	.enqueue	=	codel_qdisc_enqueue,
259 	.dequeue	=	codel_qdisc_dequeue,
260 	.peek		=	qdisc_peek_dequeued,
261 	.init		=	codel_init,
262 	.reset		=	codel_reset,
263 	.change 	=	codel_change,
264 	.dump		=	codel_dump,
265 	.dump_stats	=	codel_dump_stats,
266 	.owner		=	THIS_MODULE,
267 };
268 MODULE_ALIAS_NET_SCH("codel");
269 
270 static int __init codel_module_init(void)
271 {
272 	return register_qdisc(&codel_qdisc_ops);
273 }
274 
275 static void __exit codel_module_exit(void)
276 {
277 	unregister_qdisc(&codel_qdisc_ops);
278 }
279 
280 module_init(codel_module_init)
281 module_exit(codel_module_exit)
282 
283 MODULE_DESCRIPTION("Controlled Delay queue discipline");
284 MODULE_AUTHOR("Dave Taht");
285 MODULE_AUTHOR("Eric Dumazet");
286 MODULE_LICENSE("Dual BSD/GPL");
287