2 * net/sched/sch_red.c Random Early Detection queue.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
12 * J Hadi Salim 980914: computation fixes
13 * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly.
14 * J Hadi Salim 980816: ECN support
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/skbuff.h>
21 #include <net/pkt_sched.h>
22 #include <net/inet_ecn.h>
26 /* Parameters, settable by user:
27 -----------------------------
29 limit - bytes (must be > qth_max + burst)
31 Hard limit on queue length, should be chosen >qth_max
32 to allow packet bursts. This parameter does not
33 affect the algorithms behaviour and can be chosen
34 arbitrarily high (well, less than ram size)
35 Really, this limit will never be reached
36 if RED works correctly.
39 struct red_sched_data {
40 u32 limit; /* HARD maximal queue length */
42 struct timer_list adapt_timer;
43 struct red_parms parms;
45 struct red_stats stats;
49 static inline int red_use_ecn(struct red_sched_data *q)
51 return q->flags & TC_RED_ECN;
54 static inline int red_use_harddrop(struct red_sched_data *q)
56 return q->flags & TC_RED_HARDDROP;
59 static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch,
60 struct sk_buff **to_free)
62 struct red_sched_data *q = qdisc_priv(sch);
63 struct Qdisc *child = q->qdisc;
66 q->vars.qavg = red_calc_qavg(&q->parms,
68 child->qstats.backlog);
70 if (red_is_idling(&q->vars))
71 red_end_of_idle_period(&q->vars);
73 switch (red_action(&q->parms, &q->vars, q->vars.qavg)) {
78 qdisc_qstats_overlimit(sch);
79 if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) {
88 qdisc_qstats_overlimit(sch);
89 if (red_use_harddrop(q) || !red_use_ecn(q) ||
90 !INET_ECN_set_ce(skb)) {
91 q->stats.forced_drop++;
95 q->stats.forced_mark++;
99 ret = qdisc_enqueue(skb, child, to_free);
100 if (likely(ret == NET_XMIT_SUCCESS)) {
101 qdisc_qstats_backlog_inc(sch, skb);
103 } else if (net_xmit_drop_count(ret)) {
105 qdisc_qstats_drop(sch);
110 qdisc_drop(skb, sch, to_free);
114 static struct sk_buff *red_dequeue(struct Qdisc *sch)
117 struct red_sched_data *q = qdisc_priv(sch);
118 struct Qdisc *child = q->qdisc;
120 skb = child->dequeue(child);
122 qdisc_bstats_update(sch, skb);
123 qdisc_qstats_backlog_dec(sch, skb);
126 if (!red_is_idling(&q->vars))
127 red_start_of_idle_period(&q->vars);
132 static struct sk_buff *red_peek(struct Qdisc *sch)
134 struct red_sched_data *q = qdisc_priv(sch);
135 struct Qdisc *child = q->qdisc;
137 return child->ops->peek(child);
140 static void red_reset(struct Qdisc *sch)
142 struct red_sched_data *q = qdisc_priv(sch);
144 qdisc_reset(q->qdisc);
145 sch->qstats.backlog = 0;
147 red_restart(&q->vars);
150 static void red_destroy(struct Qdisc *sch)
152 struct red_sched_data *q = qdisc_priv(sch);
154 del_timer_sync(&q->adapt_timer);
155 qdisc_destroy(q->qdisc);
158 static const struct nla_policy red_policy[TCA_RED_MAX + 1] = {
159 [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) },
160 [TCA_RED_STAB] = { .len = RED_STAB_SIZE },
161 [TCA_RED_MAX_P] = { .type = NLA_U32 },
164 static int red_change(struct Qdisc *sch, struct nlattr *opt)
166 struct red_sched_data *q = qdisc_priv(sch);
167 struct nlattr *tb[TCA_RED_MAX + 1];
168 struct tc_red_qopt *ctl;
169 struct Qdisc *child = NULL;
177 err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy, NULL);
181 if (tb[TCA_RED_PARMS] == NULL ||
182 tb[TCA_RED_STAB] == NULL)
185 max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0;
187 ctl = nla_data(tb[TCA_RED_PARMS]);
188 stab = nla_data(tb[TCA_RED_STAB]);
189 if (!red_check_params(ctl->qth_min, ctl->qth_max, ctl->Wlog,
190 ctl->Scell_log, stab))
193 if (ctl->limit > 0) {
194 child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit);
196 return PTR_ERR(child);
198 /* child is fifo, no need to check for noop_qdisc */
199 qdisc_hash_add(child, true);
203 q->flags = ctl->flags;
204 q->limit = ctl->limit;
206 qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen,
207 q->qdisc->qstats.backlog);
208 qdisc_destroy(q->qdisc);
212 red_set_parms(&q->parms,
213 ctl->qth_min, ctl->qth_max, ctl->Wlog,
214 ctl->Plog, ctl->Scell_log,
217 red_set_vars(&q->vars);
219 del_timer(&q->adapt_timer);
220 if (ctl->flags & TC_RED_ADAPTATIVE)
221 mod_timer(&q->adapt_timer, jiffies + HZ/2);
223 if (!q->qdisc->q.qlen)
224 red_start_of_idle_period(&q->vars);
226 sch_tree_unlock(sch);
230 static inline void red_adaptative_timer(unsigned long arg)
232 struct Qdisc *sch = (struct Qdisc *)arg;
233 struct red_sched_data *q = qdisc_priv(sch);
234 spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
236 spin_lock(root_lock);
237 red_adaptative_algo(&q->parms, &q->vars);
238 mod_timer(&q->adapt_timer, jiffies + HZ/2);
239 spin_unlock(root_lock);
242 static int red_init(struct Qdisc *sch, struct nlattr *opt)
244 struct red_sched_data *q = qdisc_priv(sch);
246 q->qdisc = &noop_qdisc;
247 setup_timer(&q->adapt_timer, red_adaptative_timer, (unsigned long)sch);
248 return red_change(sch, opt);
251 static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
253 struct red_sched_data *q = qdisc_priv(sch);
254 struct nlattr *opts = NULL;
255 struct tc_red_qopt opt = {
258 .qth_min = q->parms.qth_min >> q->parms.Wlog,
259 .qth_max = q->parms.qth_max >> q->parms.Wlog,
260 .Wlog = q->parms.Wlog,
261 .Plog = q->parms.Plog,
262 .Scell_log = q->parms.Scell_log,
265 sch->qstats.backlog = q->qdisc->qstats.backlog;
266 opts = nla_nest_start(skb, TCA_OPTIONS);
268 goto nla_put_failure;
269 if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) ||
270 nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P))
271 goto nla_put_failure;
272 return nla_nest_end(skb, opts);
275 nla_nest_cancel(skb, opts);
279 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
281 struct red_sched_data *q = qdisc_priv(sch);
282 struct tc_red_xstats st = {
283 .early = q->stats.prob_drop + q->stats.forced_drop,
284 .pdrop = q->stats.pdrop,
285 .other = q->stats.other,
286 .marked = q->stats.prob_mark + q->stats.forced_mark,
289 return gnet_stats_copy_app(d, &st, sizeof(st));
292 static int red_dump_class(struct Qdisc *sch, unsigned long cl,
293 struct sk_buff *skb, struct tcmsg *tcm)
295 struct red_sched_data *q = qdisc_priv(sch);
297 tcm->tcm_handle |= TC_H_MIN(1);
298 tcm->tcm_info = q->qdisc->handle;
302 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
305 struct red_sched_data *q = qdisc_priv(sch);
310 *old = qdisc_replace(sch, new, &q->qdisc);
314 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg)
316 struct red_sched_data *q = qdisc_priv(sch);
320 static unsigned long red_find(struct Qdisc *sch, u32 classid)
325 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker)
328 if (walker->count >= walker->skip)
329 if (walker->fn(sch, 1, walker) < 0) {
337 static const struct Qdisc_class_ops red_class_ops = {
342 .dump = red_dump_class,
345 static struct Qdisc_ops red_qdisc_ops __read_mostly = {
347 .priv_size = sizeof(struct red_sched_data),
348 .cl_ops = &red_class_ops,
349 .enqueue = red_enqueue,
350 .dequeue = red_dequeue,
354 .destroy = red_destroy,
355 .change = red_change,
357 .dump_stats = red_dump_stats,
358 .owner = THIS_MODULE,
361 static int __init red_module_init(void)
363 return register_qdisc(&red_qdisc_ops);
366 static void __exit red_module_exit(void)
368 unregister_qdisc(&red_qdisc_ops);
371 module_init(red_module_init)
372 module_exit(red_module_exit)
374 MODULE_LICENSE("GPL");