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 */
dequeue_func(struct codel_vars * vars,void * ctx)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
drop_func(struct sk_buff * skb,void * ctx)51 static void drop_func(struct sk_buff *skb, void *ctx)
52 {
53 struct Qdisc *sch = ctx;
54
55 kfree_skb(skb);
56 qdisc_qstats_drop(sch);
57 }
58
codel_qdisc_dequeue(struct Qdisc * sch)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
codel_qdisc_enqueue(struct sk_buff * skb,struct Qdisc * sch,struct sk_buff ** to_free)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(skb, sch, to_free);
90 }
91
92 static const struct nla_policy codel_policy[TCA_CODEL_MAX + 1] = {
93 [TCA_CODEL_TARGET] = { .type = NLA_U32 },
94 [TCA_CODEL_LIMIT] = { .type = NLA_U32 },
95 [TCA_CODEL_INTERVAL] = { .type = NLA_U32 },
96 [TCA_CODEL_ECN] = { .type = NLA_U32 },
97 [TCA_CODEL_CE_THRESHOLD]= { .type = NLA_U32 },
98 };
99
codel_change(struct Qdisc * sch,struct nlattr * opt,struct netlink_ext_ack * extack)100 static int codel_change(struct Qdisc *sch, struct nlattr *opt,
101 struct netlink_ext_ack *extack)
102 {
103 struct codel_sched_data *q = qdisc_priv(sch);
104 struct nlattr *tb[TCA_CODEL_MAX + 1];
105 unsigned int qlen, dropped = 0;
106 int err;
107
108 err = nla_parse_nested_deprecated(tb, TCA_CODEL_MAX, opt,
109 codel_policy, NULL);
110 if (err < 0)
111 return err;
112
113 sch_tree_lock(sch);
114
115 if (tb[TCA_CODEL_TARGET]) {
116 u32 target = nla_get_u32(tb[TCA_CODEL_TARGET]);
117
118 WRITE_ONCE(q->params.target,
119 ((u64)target * NSEC_PER_USEC) >> CODEL_SHIFT);
120 }
121
122 if (tb[TCA_CODEL_CE_THRESHOLD]) {
123 u64 val = nla_get_u32(tb[TCA_CODEL_CE_THRESHOLD]);
124
125 WRITE_ONCE(q->params.ce_threshold,
126 (val * NSEC_PER_USEC) >> CODEL_SHIFT);
127 }
128
129 if (tb[TCA_CODEL_INTERVAL]) {
130 u32 interval = nla_get_u32(tb[TCA_CODEL_INTERVAL]);
131
132 WRITE_ONCE(q->params.interval,
133 ((u64)interval * NSEC_PER_USEC) >> CODEL_SHIFT);
134 }
135
136 if (tb[TCA_CODEL_LIMIT])
137 WRITE_ONCE(sch->limit,
138 nla_get_u32(tb[TCA_CODEL_LIMIT]));
139
140 if (tb[TCA_CODEL_ECN])
141 WRITE_ONCE(q->params.ecn,
142 !!nla_get_u32(tb[TCA_CODEL_ECN]));
143
144 qlen = sch->q.qlen;
145 while (sch->q.qlen > sch->limit) {
146 struct sk_buff *skb = qdisc_dequeue_internal(sch, true);
147
148 dropped += qdisc_pkt_len(skb);
149 qdisc_qstats_backlog_dec(sch, skb);
150 rtnl_qdisc_drop(skb, sch);
151 }
152 qdisc_tree_reduce_backlog(sch, qlen - sch->q.qlen, dropped);
153
154 sch_tree_unlock(sch);
155 return 0;
156 }
157
codel_init(struct Qdisc * sch,struct nlattr * opt,struct netlink_ext_ack * extack)158 static int codel_init(struct Qdisc *sch, struct nlattr *opt,
159 struct netlink_ext_ack *extack)
160 {
161 struct codel_sched_data *q = qdisc_priv(sch);
162
163 sch->limit = DEFAULT_CODEL_LIMIT;
164
165 codel_params_init(&q->params);
166 codel_vars_init(&q->vars);
167 codel_stats_init(&q->stats);
168 q->params.mtu = psched_mtu(qdisc_dev(sch));
169
170 if (opt) {
171 int err = codel_change(sch, opt, extack);
172
173 if (err)
174 return err;
175 }
176
177 if (sch->limit >= 1)
178 sch->flags |= TCQ_F_CAN_BYPASS;
179 else
180 sch->flags &= ~TCQ_F_CAN_BYPASS;
181
182 return 0;
183 }
184
codel_dump(struct Qdisc * sch,struct sk_buff * skb)185 static int codel_dump(struct Qdisc *sch, struct sk_buff *skb)
186 {
187 struct codel_sched_data *q = qdisc_priv(sch);
188 codel_time_t ce_threshold;
189 struct nlattr *opts;
190
191 opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
192 if (opts == NULL)
193 goto nla_put_failure;
194
195 if (nla_put_u32(skb, TCA_CODEL_TARGET,
196 codel_time_to_us(READ_ONCE(q->params.target))) ||
197 nla_put_u32(skb, TCA_CODEL_LIMIT,
198 READ_ONCE(sch->limit)) ||
199 nla_put_u32(skb, TCA_CODEL_INTERVAL,
200 codel_time_to_us(READ_ONCE(q->params.interval))) ||
201 nla_put_u32(skb, TCA_CODEL_ECN,
202 READ_ONCE(q->params.ecn)))
203 goto nla_put_failure;
204 ce_threshold = READ_ONCE(q->params.ce_threshold);
205 if (ce_threshold != CODEL_DISABLED_THRESHOLD &&
206 nla_put_u32(skb, TCA_CODEL_CE_THRESHOLD,
207 codel_time_to_us(ce_threshold)))
208 goto nla_put_failure;
209 return nla_nest_end(skb, opts);
210
211 nla_put_failure:
212 nla_nest_cancel(skb, opts);
213 return -1;
214 }
215
codel_dump_stats(struct Qdisc * sch,struct gnet_dump * d)216 static int codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
217 {
218 const struct codel_sched_data *q = qdisc_priv(sch);
219 struct tc_codel_xstats st = {
220 .maxpacket = q->stats.maxpacket,
221 .count = q->vars.count,
222 .lastcount = q->vars.lastcount,
223 .drop_overlimit = q->drop_overlimit,
224 .ldelay = codel_time_to_us(q->vars.ldelay),
225 .dropping = q->vars.dropping,
226 .ecn_mark = q->stats.ecn_mark,
227 .ce_mark = q->stats.ce_mark,
228 };
229
230 if (q->vars.dropping) {
231 codel_tdiff_t delta = q->vars.drop_next - codel_get_time();
232
233 if (delta >= 0)
234 st.drop_next = codel_time_to_us(delta);
235 else
236 st.drop_next = -codel_time_to_us(-delta);
237 }
238
239 return gnet_stats_copy_app(d, &st, sizeof(st));
240 }
241
codel_reset(struct Qdisc * sch)242 static void codel_reset(struct Qdisc *sch)
243 {
244 struct codel_sched_data *q = qdisc_priv(sch);
245
246 qdisc_reset_queue(sch);
247 codel_vars_init(&q->vars);
248 }
249
250 static struct Qdisc_ops codel_qdisc_ops __read_mostly = {
251 .id = "codel",
252 .priv_size = sizeof(struct codel_sched_data),
253
254 .enqueue = codel_qdisc_enqueue,
255 .dequeue = codel_qdisc_dequeue,
256 .peek = qdisc_peek_dequeued,
257 .init = codel_init,
258 .reset = codel_reset,
259 .change = codel_change,
260 .dump = codel_dump,
261 .dump_stats = codel_dump_stats,
262 .owner = THIS_MODULE,
263 };
264 MODULE_ALIAS_NET_SCH("codel");
265
codel_module_init(void)266 static int __init codel_module_init(void)
267 {
268 return register_qdisc(&codel_qdisc_ops);
269 }
270
codel_module_exit(void)271 static void __exit codel_module_exit(void)
272 {
273 unregister_qdisc(&codel_qdisc_ops);
274 }
275
276 module_init(codel_module_init)
277 module_exit(codel_module_exit)
278
279 MODULE_DESCRIPTION("Controlled Delay queue discipline");
280 MODULE_AUTHOR("Dave Taht");
281 MODULE_AUTHOR("Eric Dumazet");
282 MODULE_LICENSE("Dual BSD/GPL");
283