sch_codel.c

  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
 28struct 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 */
 39static 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
 51static void drop_func(struct sk_buff *skb, void *ctx)
 52{
 53	struct Qdisc *sch = ctx;
 54
 55	qdisc_dequeue_drop(sch, skb, SKB_DROP_REASON_QDISC_CONGESTED);
 56	qdisc_qstats_drop(sch);
 57}
 58
 59static 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
 78static 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
 93static 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
101static int codel_change(struct Qdisc *sch, struct nlattr *opt,
102			struct netlink_ext_ack *extack)
103{
104	unsigned int dropped_pkts = 0, dropped_bytes = 0;
105	struct codel_sched_data *q = qdisc_priv(sch);
106	struct nlattr *tb[TCA_CODEL_MAX + 1];
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	while (sch->q.qlen > sch->limit) {
146		struct sk_buff *skb = qdisc_dequeue_internal(sch, true);
147
148		if (!skb)
149			break;
150
151		dropped_pkts++;
152		dropped_bytes += qdisc_pkt_len(skb);
153		rtnl_qdisc_drop(skb, sch);
154	}
155	qdisc_tree_reduce_backlog(sch, dropped_pkts, dropped_bytes);
156
157	sch_tree_unlock(sch);
158	return 0;
159}
160
161static int codel_init(struct Qdisc *sch, struct nlattr *opt,
162		      struct netlink_ext_ack *extack)
163{
164	struct codel_sched_data *q = qdisc_priv(sch);
165
166	sch->limit = DEFAULT_CODEL_LIMIT;
167
168	codel_params_init(&q->params);
169	codel_vars_init(&q->vars);
170	codel_stats_init(&q->stats);
171	q->params.mtu = psched_mtu(qdisc_dev(sch));
172
173	if (opt) {
174		int err = codel_change(sch, opt, extack);
175
176		if (err)
177			return err;
178	}
179
180	if (sch->limit >= 1)
181		sch->flags |= TCQ_F_CAN_BYPASS;
182	else
183		sch->flags &= ~TCQ_F_CAN_BYPASS;
184
185	sch->flags |= TCQ_F_DEQUEUE_DROPS;
186
187	return 0;
188}
189
190static int codel_dump(struct Qdisc *sch, struct sk_buff *skb)
191{
192	struct codel_sched_data *q = qdisc_priv(sch);
193	codel_time_t ce_threshold;
194	struct nlattr *opts;
195
196	opts = nla_nest_start_noflag(skb, TCA_OPTIONS);
197	if (opts == NULL)
198		goto nla_put_failure;
199
200	if (nla_put_u32(skb, TCA_CODEL_TARGET,
201			codel_time_to_us(READ_ONCE(q->params.target))) ||
202	    nla_put_u32(skb, TCA_CODEL_LIMIT,
203			READ_ONCE(sch->limit)) ||
204	    nla_put_u32(skb, TCA_CODEL_INTERVAL,
205			codel_time_to_us(READ_ONCE(q->params.interval))) ||
206	    nla_put_u32(skb, TCA_CODEL_ECN,
207			READ_ONCE(q->params.ecn)))
208		goto nla_put_failure;
209	ce_threshold = READ_ONCE(q->params.ce_threshold);
210	if (ce_threshold != CODEL_DISABLED_THRESHOLD &&
211	    nla_put_u32(skb, TCA_CODEL_CE_THRESHOLD,
212			codel_time_to_us(ce_threshold)))
213		goto nla_put_failure;
214	return nla_nest_end(skb, opts);
215
216nla_put_failure:
217	nla_nest_cancel(skb, opts);
218	return -1;
219}
220
221static int codel_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
222{
223	const struct codel_sched_data *q = qdisc_priv(sch);
224	struct tc_codel_xstats st = {
225		.maxpacket	= q->stats.maxpacket,
226		.count		= q->vars.count,
227		.lastcount	= q->vars.lastcount,
228		.drop_overlimit = q->drop_overlimit,
229		.ldelay		= codel_time_to_us(q->vars.ldelay),
230		.dropping	= q->vars.dropping,
231		.ecn_mark	= q->stats.ecn_mark,
232		.ce_mark	= q->stats.ce_mark,
233	};
234
235	if (q->vars.dropping) {
236		codel_tdiff_t delta = q->vars.drop_next - codel_get_time();
237
238		if (delta >= 0)
239			st.drop_next = codel_time_to_us(delta);
240		else
241			st.drop_next = -codel_time_to_us(-delta);
242	}
243
244	return gnet_stats_copy_app(d, &st, sizeof(st));
245}
246
247static void codel_reset(struct Qdisc *sch)
248{
249	struct codel_sched_data *q = qdisc_priv(sch);
250
251	qdisc_reset_queue(sch);
252	codel_vars_init(&q->vars);
253}
254
255static struct Qdisc_ops codel_qdisc_ops __read_mostly = {
256	.id		=	"codel",
257	.priv_size	=	sizeof(struct codel_sched_data),
258
259	.enqueue	=	codel_qdisc_enqueue,
260	.dequeue	=	codel_qdisc_dequeue,
261	.peek		=	qdisc_peek_dequeued,
262	.init		=	codel_init,
263	.reset		=	codel_reset,
264	.change 	=	codel_change,
265	.dump		=	codel_dump,
266	.dump_stats	=	codel_dump_stats,
267	.owner		=	THIS_MODULE,
268};
269MODULE_ALIAS_NET_SCH("codel");
270
271static int __init codel_module_init(void)
272{
273	return register_qdisc(&codel_qdisc_ops);
274}
275
276static void __exit codel_module_exit(void)
277{
278	unregister_qdisc(&codel_qdisc_ops);
279}
280
281module_init(codel_module_init)
282module_exit(codel_module_exit)
283
284MODULE_DESCRIPTION("Controlled Delay queue discipline");
285MODULE_AUTHOR("Dave Taht");
286MODULE_AUTHOR("Eric Dumazet");
287MODULE_LICENSE("Dual BSD/GPL");