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 	if (q->stats.drop_count) {
69 		qdisc_tree_reduce_backlog(sch, q->stats.drop_count, q->stats.drop_len);
70 		q->stats.drop_count = 0;
71 		q->stats.drop_len = 0;
72 	}
73 	if (skb)
74 		qdisc_bstats_update(sch, skb);
75 	return skb;
76 }
77 
78 static int codel_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
79 			       struct sk_buff **to_free)
80 {
81 	struct codel_sched_data *q;
82 
83 	if (likely(qdisc_qlen(sch) < sch->limit)) {
84 		codel_set_enqueue_time(skb);
85 		return qdisc_enqueue_tail(skb, sch);
86 	}
87 	q = qdisc_priv(sch);
88 	q->drop_overlimit++;
89 	return qdisc_drop_reason(skb, sch, to_free,
90 				 SKB_DROP_REASON_QDISC_OVERLIMIT);
91 }
92 
93 static const struct nla_policy codel_policy[TCA_CODEL_MAX + 1] = {
94 	[TCA_CODEL_TARGET]	= { .type = NLA_U32 },
95 	[TCA_CODEL_LIMIT]	= { .type = NLA_U32 },
96 	[TCA_CODEL_INTERVAL]	= { .type = NLA_U32 },
97 	[TCA_CODEL_ECN]		= { .type = NLA_U32 },
98 	[TCA_CODEL_CE_THRESHOLD]= { .type = NLA_U32 },
99 };
100 
101 static int codel_change(struct Qdisc *sch, struct nlattr *opt,
102 			struct netlink_ext_ack *extack)
103 {
104 	struct codel_sched_data *q = qdisc_priv(sch);
105 	struct nlattr *tb[TCA_CODEL_MAX + 1];
106 	unsigned int qlen, dropped = 0;
107 	int err;
108 
109 	err = nla_parse_nested_deprecated(tb, TCA_CODEL_MAX, opt,
110 					  codel_policy, NULL);
111 	if (err < 0)
112 		return err;
113 
114 	sch_tree_lock(sch);
115 
116 	if (tb[TCA_CODEL_TARGET]) {
117 		u32 target = nla_get_u32(tb[TCA_CODEL_TARGET]);
118 
119 		WRITE_ONCE(q->params.target,
120 			   ((u64)target * NSEC_PER_USEC) >> CODEL_SHIFT);
121 	}
122 
123 	if (tb[TCA_CODEL_CE_THRESHOLD]) {
124 		u64 val = nla_get_u32(tb[TCA_CODEL_CE_THRESHOLD]);
125 
126 		WRITE_ONCE(q->params.ce_threshold,
127 			   (val * NSEC_PER_USEC) >> CODEL_SHIFT);
128 	}
129 
130 	if (tb[TCA_CODEL_INTERVAL]) {
131 		u32 interval = nla_get_u32(tb[TCA_CODEL_INTERVAL]);
132 
133 		WRITE_ONCE(q->params.interval,
134 			   ((u64)interval * NSEC_PER_USEC) >> CODEL_SHIFT);
135 	}
136 
137 	if (tb[TCA_CODEL_LIMIT])
138 		WRITE_ONCE(sch->limit,
139 			   nla_get_u32(tb[TCA_CODEL_LIMIT]));
140 
141 	if (tb[TCA_CODEL_ECN])
142 		WRITE_ONCE(q->params.ecn,
143 			   !!nla_get_u32(tb[TCA_CODEL_ECN]));
144 
145 	qlen = sch->q.qlen;
146 	while (sch->q.qlen > sch->limit) {
147 		struct sk_buff *skb = qdisc_dequeue_internal(sch, true);
148 
149 		dropped += qdisc_pkt_len(skb);
150 		qdisc_qstats_backlog_dec(sch, skb);
151 		rtnl_qdisc_drop(skb, sch);
152 	}
153 	qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);
154 
155 	sch_tree_unlock(sch);
156 	return 0;
157 }
158 
159 static int codel_init(struct Qdisc *sch, struct nlattr *opt,
160 		      struct netlink_ext_ack *extack)
161 {
162 	struct codel_sched_data *q = qdisc_priv(sch);
163 
164 	sch->limit = DEFAULT_CODEL_LIMIT;
165 
166 	codel_params_init(&q->params);
167 	codel_vars_init(&q->vars);
168 	codel_stats_init(&q->stats);
169 	q->params.mtu = psched_mtu(qdisc_dev(sch));
170 
171 	if (opt) {
172 		int err = codel_change(sch, opt, extack);
173 
174 		if (err)
175 			return err;
176 	}
177 
178 	if (sch->limit >= 1)
179 		sch->flags |= TCQ_F_CAN_BYPASS;
180 	else
181 		sch->flags &= ~TCQ_F_CAN_BYPASS;
182 
183 	return 0;
184 }
185 
186 static int codel_dump(struct Qdisc *sch, struct sk_buff *skb)
187 {
188 	struct codel_sched_data *q = qdisc_priv(sch);
189 	codel_time_t ce_threshold;
190 	struct nlattr *opts;
191 
192 	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
193 	if (opts == NULL)
194 		goto nla_put_failure;
195 
196 	if (nla_put_u32(skb, TCA_CODEL_TARGET,
197 			codel_time_to_us(READ_ONCE(q->params.target))) ||
198 	    nla_put_u32(skb, TCA_CODEL_LIMIT,
199 			READ_ONCE(sch->limit)) ||
200 	    nla_put_u32(skb, TCA_CODEL_INTERVAL,
201 			codel_time_to_us(READ_ONCE(q->params.interval))) ||
202 	    nla_put_u32(skb, TCA_CODEL_ECN,
203 			READ_ONCE(q->params.ecn)))
204 		goto nla_put_failure;
205 	ce_threshold = READ_ONCE(q->params.ce_threshold);
206 	if (ce_threshold != CODEL_DISABLED_THRESHOLD &&
207 	    nla_put_u32(skb, TCA_CODEL_CE_THRESHOLD,
208 			codel_time_to_us(ce_threshold)))
209 		goto nla_put_failure;
210 	return nla_nest_end(skb, opts);
211 
212 nla_put_failure:
213 	nla_nest_cancel(skb, opts);
214 	return -1;
215 }
216 
217 static int codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
218 {
219 	const struct codel_sched_data *q = qdisc_priv(sch);
220 	struct tc_codel_xstats st = {
221 		.maxpacket	= q->stats.maxpacket,
222 		.count		= q->vars.count,
223 		.lastcount	= q->vars.lastcount,
224 		.drop_overlimit = q->drop_overlimit,
225 		.ldelay		= codel_time_to_us(q->vars.ldelay),
226 		.dropping	= q->vars.dropping,
227 		.ecn_mark	= q->stats.ecn_mark,
228 		.ce_mark	= q->stats.ce_mark,
229 	};
230 
231 	if (q->vars.dropping) {
232 		codel_tdiff_t delta = q->vars.drop_next - codel_get_time();
233 
234 		if (delta >= 0)
235 			st.drop_next = codel_time_to_us(delta);
236 		else
237 			st.drop_next = -codel_time_to_us(-delta);
238 	}
239 
240 	return gnet_stats_copy_app(d, &st, sizeof(st));
241 }
242 
243 static void codel_reset(struct Qdisc *sch)
244 {
245 	struct codel_sched_data *q = qdisc_priv(sch);
246 
247 	qdisc_reset_queue(sch);
248 	codel_vars_init(&q->vars);
249 }
250 
251 static struct Qdisc_ops codel_qdisc_ops __read_mostly = {
252 	.id		=	"codel",
253 	.priv_size	=	sizeof(struct codel_sched_data),
254 
255 	.enqueue	=	codel_qdisc_enqueue,
256 	.dequeue	=	codel_qdisc_dequeue,
257 	.peek		=	qdisc_peek_dequeued,
258 	.init		=	codel_init,
259 	.reset		=	codel_reset,
260 	.change 	=	codel_change,
261 	.dump		=	codel_dump,
262 	.dump_stats	=	codel_dump_stats,
263 	.owner		=	THIS_MODULE,
264 };
265 MODULE_ALIAS_NET_SCH("codel");
266 
267 static int __init codel_module_init(void)
268 {
269 	return register_qdisc(&codel_qdisc_ops);
270 }
271 
272 static void __exit codel_module_exit(void)
273 {
274 	unregister_qdisc(&codel_qdisc_ops);
275 }
276 
277 module_init(codel_module_init)
278 module_exit(codel_module_exit)
279 
280 MODULE_DESCRIPTION("Controlled Delay queue discipline");
281 MODULE_AUTHOR("Dave Taht");
282 MODULE_AUTHOR("Eric Dumazet");
283 MODULE_LICENSE("Dual BSD/GPL");
284