]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * net/sched/sch_red.c Random Early Detection queue. | |
3 | * | |
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. | |
8 | * | |
9 | * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> | |
10 | * | |
11 | * Changes: | |
dba051f3 | 12 | * J Hadi Salim 980914: computation fixes |
1da177e4 | 13 | * Alexey Makarenko <makar@phoenix.kharkov.ua> 990814: qave on idle link was calculated incorrectly. |
dba051f3 | 14 | * J Hadi Salim 980816: ECN support |
1da177e4 LT |
15 | */ |
16 | ||
1da177e4 | 17 | #include <linux/module.h> |
1da177e4 LT |
18 | #include <linux/types.h> |
19 | #include <linux/kernel.h> | |
1da177e4 | 20 | #include <linux/skbuff.h> |
1da177e4 LT |
21 | #include <net/pkt_sched.h> |
22 | #include <net/inet_ecn.h> | |
6b31b28a | 23 | #include <net/red.h> |
1da177e4 LT |
24 | |
25 | ||
6b31b28a | 26 | /* Parameters, settable by user: |
1da177e4 LT |
27 | ----------------------------- |
28 | ||
29 | limit - bytes (must be > qth_max + burst) | |
30 | ||
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. | |
1da177e4 LT |
37 | */ |
38 | ||
cc7ec456 | 39 | struct red_sched_data { |
6b31b28a TG |
40 | u32 limit; /* HARD maximal queue length */ |
41 | unsigned char flags; | |
8af2a218 | 42 | struct timer_list adapt_timer; |
6b31b28a | 43 | struct red_parms parms; |
eeca6688 | 44 | struct red_vars vars; |
6b31b28a | 45 | struct red_stats stats; |
f38c39d6 | 46 | struct Qdisc *qdisc; |
1da177e4 LT |
47 | }; |
48 | ||
6b31b28a | 49 | static inline int red_use_ecn(struct red_sched_data *q) |
1da177e4 | 50 | { |
6b31b28a | 51 | return q->flags & TC_RED_ECN; |
1da177e4 LT |
52 | } |
53 | ||
bdc450a0 TG |
54 | static inline int red_use_harddrop(struct red_sched_data *q) |
55 | { | |
56 | return q->flags & TC_RED_HARDDROP; | |
57 | } | |
58 | ||
520ac30f ED |
59 | static int red_enqueue(struct sk_buff *skb, struct Qdisc *sch, |
60 | struct sk_buff **to_free) | |
1da177e4 LT |
61 | { |
62 | struct red_sched_data *q = qdisc_priv(sch); | |
f38c39d6 PM |
63 | struct Qdisc *child = q->qdisc; |
64 | int ret; | |
1da177e4 | 65 | |
eeca6688 ED |
66 | q->vars.qavg = red_calc_qavg(&q->parms, |
67 | &q->vars, | |
68 | child->qstats.backlog); | |
1da177e4 | 69 | |
eeca6688 ED |
70 | if (red_is_idling(&q->vars)) |
71 | red_end_of_idle_period(&q->vars); | |
1da177e4 | 72 | |
eeca6688 | 73 | switch (red_action(&q->parms, &q->vars, q->vars.qavg)) { |
cc7ec456 ED |
74 | case RED_DONT_MARK: |
75 | break; | |
76 | ||
77 | case RED_PROB_MARK: | |
25331d6c | 78 | qdisc_qstats_overlimit(sch); |
cc7ec456 ED |
79 | if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) { |
80 | q->stats.prob_drop++; | |
81 | goto congestion_drop; | |
82 | } | |
83 | ||
84 | q->stats.prob_mark++; | |
85 | break; | |
86 | ||
87 | case RED_HARD_MARK: | |
25331d6c | 88 | qdisc_qstats_overlimit(sch); |
cc7ec456 ED |
89 | if (red_use_harddrop(q) || !red_use_ecn(q) || |
90 | !INET_ECN_set_ce(skb)) { | |
91 | q->stats.forced_drop++; | |
92 | goto congestion_drop; | |
93 | } | |
94 | ||
95 | q->stats.forced_mark++; | |
96 | break; | |
1da177e4 LT |
97 | } |
98 | ||
520ac30f | 99 | ret = qdisc_enqueue(skb, child, to_free); |
f38c39d6 | 100 | if (likely(ret == NET_XMIT_SUCCESS)) { |
d7f4f332 | 101 | qdisc_qstats_backlog_inc(sch, skb); |
f38c39d6 | 102 | sch->q.qlen++; |
378a2f09 | 103 | } else if (net_xmit_drop_count(ret)) { |
f38c39d6 | 104 | q->stats.pdrop++; |
25331d6c | 105 | qdisc_qstats_drop(sch); |
f38c39d6 PM |
106 | } |
107 | return ret; | |
6b31b28a TG |
108 | |
109 | congestion_drop: | |
520ac30f | 110 | qdisc_drop(skb, sch, to_free); |
1da177e4 LT |
111 | return NET_XMIT_CN; |
112 | } | |
113 | ||
cc7ec456 | 114 | static struct sk_buff *red_dequeue(struct Qdisc *sch) |
1da177e4 LT |
115 | { |
116 | struct sk_buff *skb; | |
117 | struct red_sched_data *q = qdisc_priv(sch); | |
f38c39d6 | 118 | struct Qdisc *child = q->qdisc; |
1da177e4 | 119 | |
f38c39d6 | 120 | skb = child->dequeue(child); |
9190b3b3 ED |
121 | if (skb) { |
122 | qdisc_bstats_update(sch, skb); | |
d7f4f332 | 123 | qdisc_qstats_backlog_dec(sch, skb); |
f38c39d6 | 124 | sch->q.qlen--; |
9190b3b3 | 125 | } else { |
eeca6688 ED |
126 | if (!red_is_idling(&q->vars)) |
127 | red_start_of_idle_period(&q->vars); | |
9190b3b3 | 128 | } |
9e178ff2 | 129 | return skb; |
1da177e4 LT |
130 | } |
131 | ||
cc7ec456 | 132 | static struct sk_buff *red_peek(struct Qdisc *sch) |
8e3af978 JP |
133 | { |
134 | struct red_sched_data *q = qdisc_priv(sch); | |
135 | struct Qdisc *child = q->qdisc; | |
136 | ||
137 | return child->ops->peek(child); | |
138 | } | |
139 | ||
cc7ec456 | 140 | static void red_reset(struct Qdisc *sch) |
1da177e4 LT |
141 | { |
142 | struct red_sched_data *q = qdisc_priv(sch); | |
143 | ||
f38c39d6 | 144 | qdisc_reset(q->qdisc); |
d7f4f332 | 145 | sch->qstats.backlog = 0; |
f38c39d6 | 146 | sch->q.qlen = 0; |
eeca6688 | 147 | red_restart(&q->vars); |
1da177e4 LT |
148 | } |
149 | ||
f38c39d6 PM |
150 | static void red_destroy(struct Qdisc *sch) |
151 | { | |
152 | struct red_sched_data *q = qdisc_priv(sch); | |
8af2a218 ED |
153 | |
154 | del_timer_sync(&q->adapt_timer); | |
f38c39d6 PM |
155 | qdisc_destroy(q->qdisc); |
156 | } | |
157 | ||
27a3421e PM |
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 }, | |
a73ed26b | 161 | [TCA_RED_MAX_P] = { .type = NLA_U32 }, |
27a3421e PM |
162 | }; |
163 | ||
1e90474c | 164 | static int red_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
165 | { |
166 | struct red_sched_data *q = qdisc_priv(sch); | |
1e90474c | 167 | struct nlattr *tb[TCA_RED_MAX + 1]; |
1da177e4 | 168 | struct tc_red_qopt *ctl; |
f38c39d6 | 169 | struct Qdisc *child = NULL; |
cee63723 | 170 | int err; |
a73ed26b | 171 | u32 max_P; |
1da177e4 | 172 | |
cee63723 | 173 | if (opt == NULL) |
dba051f3 TG |
174 | return -EINVAL; |
175 | ||
fceb6435 | 176 | err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy, NULL); |
cee63723 PM |
177 | if (err < 0) |
178 | return err; | |
179 | ||
1e90474c | 180 | if (tb[TCA_RED_PARMS] == NULL || |
27a3421e | 181 | tb[TCA_RED_STAB] == NULL) |
1da177e4 LT |
182 | return -EINVAL; |
183 | ||
a73ed26b ED |
184 | max_P = tb[TCA_RED_MAX_P] ? nla_get_u32(tb[TCA_RED_MAX_P]) : 0; |
185 | ||
1e90474c | 186 | ctl = nla_data(tb[TCA_RED_PARMS]); |
1da177e4 | 187 | |
f38c39d6 | 188 | if (ctl->limit > 0) { |
fb0305ce PM |
189 | child = fifo_create_dflt(sch, &bfifo_qdisc_ops, ctl->limit); |
190 | if (IS_ERR(child)) | |
191 | return PTR_ERR(child); | |
f38c39d6 PM |
192 | } |
193 | ||
49b49971 JK |
194 | if (child != &noop_qdisc) |
195 | qdisc_hash_add(child, true); | |
1da177e4 LT |
196 | sch_tree_lock(sch); |
197 | q->flags = ctl->flags; | |
1da177e4 | 198 | q->limit = ctl->limit; |
5e50da01 | 199 | if (child) { |
2ccccf5f WC |
200 | qdisc_tree_reduce_backlog(q->qdisc, q->qdisc->q.qlen, |
201 | q->qdisc->qstats.backlog); | |
b94c8afc PM |
202 | qdisc_destroy(q->qdisc); |
203 | q->qdisc = child; | |
5e50da01 | 204 | } |
1da177e4 | 205 | |
eeca6688 ED |
206 | red_set_parms(&q->parms, |
207 | ctl->qth_min, ctl->qth_max, ctl->Wlog, | |
a73ed26b ED |
208 | ctl->Plog, ctl->Scell_log, |
209 | nla_data(tb[TCA_RED_STAB]), | |
210 | max_P); | |
eeca6688 | 211 | red_set_vars(&q->vars); |
6b31b28a | 212 | |
8af2a218 ED |
213 | del_timer(&q->adapt_timer); |
214 | if (ctl->flags & TC_RED_ADAPTATIVE) | |
215 | mod_timer(&q->adapt_timer, jiffies + HZ/2); | |
216 | ||
1ee5fa1e | 217 | if (!q->qdisc->q.qlen) |
eeca6688 | 218 | red_start_of_idle_period(&q->vars); |
dba051f3 | 219 | |
1da177e4 LT |
220 | sch_tree_unlock(sch); |
221 | return 0; | |
222 | } | |
223 | ||
8af2a218 ED |
224 | static inline void red_adaptative_timer(unsigned long arg) |
225 | { | |
226 | struct Qdisc *sch = (struct Qdisc *)arg; | |
227 | struct red_sched_data *q = qdisc_priv(sch); | |
228 | spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch)); | |
229 | ||
230 | spin_lock(root_lock); | |
eeca6688 | 231 | red_adaptative_algo(&q->parms, &q->vars); |
8af2a218 ED |
232 | mod_timer(&q->adapt_timer, jiffies + HZ/2); |
233 | spin_unlock(root_lock); | |
234 | } | |
235 | ||
cc7ec456 | 236 | static int red_init(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 | 237 | { |
f38c39d6 PM |
238 | struct red_sched_data *q = qdisc_priv(sch); |
239 | ||
240 | q->qdisc = &noop_qdisc; | |
8af2a218 | 241 | setup_timer(&q->adapt_timer, red_adaptative_timer, (unsigned long)sch); |
1da177e4 LT |
242 | return red_change(sch, opt); |
243 | } | |
244 | ||
245 | static int red_dump(struct Qdisc *sch, struct sk_buff *skb) | |
246 | { | |
247 | struct red_sched_data *q = qdisc_priv(sch); | |
1e90474c | 248 | struct nlattr *opts = NULL; |
6b31b28a TG |
249 | struct tc_red_qopt opt = { |
250 | .limit = q->limit, | |
251 | .flags = q->flags, | |
252 | .qth_min = q->parms.qth_min >> q->parms.Wlog, | |
253 | .qth_max = q->parms.qth_max >> q->parms.Wlog, | |
254 | .Wlog = q->parms.Wlog, | |
255 | .Plog = q->parms.Plog, | |
256 | .Scell_log = q->parms.Scell_log, | |
257 | }; | |
1da177e4 | 258 | |
0dfb33a0 | 259 | sch->qstats.backlog = q->qdisc->qstats.backlog; |
1e90474c PM |
260 | opts = nla_nest_start(skb, TCA_OPTIONS); |
261 | if (opts == NULL) | |
262 | goto nla_put_failure; | |
1b34ec43 DM |
263 | if (nla_put(skb, TCA_RED_PARMS, sizeof(opt), &opt) || |
264 | nla_put_u32(skb, TCA_RED_MAX_P, q->parms.max_P)) | |
265 | goto nla_put_failure; | |
1e90474c | 266 | return nla_nest_end(skb, opts); |
1da177e4 | 267 | |
1e90474c | 268 | nla_put_failure: |
bc3ed28c TG |
269 | nla_nest_cancel(skb, opts); |
270 | return -EMSGSIZE; | |
1da177e4 LT |
271 | } |
272 | ||
273 | static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
274 | { | |
275 | struct red_sched_data *q = qdisc_priv(sch); | |
6b31b28a TG |
276 | struct tc_red_xstats st = { |
277 | .early = q->stats.prob_drop + q->stats.forced_drop, | |
278 | .pdrop = q->stats.pdrop, | |
279 | .other = q->stats.other, | |
280 | .marked = q->stats.prob_mark + q->stats.forced_mark, | |
281 | }; | |
282 | ||
283 | return gnet_stats_copy_app(d, &st, sizeof(st)); | |
1da177e4 LT |
284 | } |
285 | ||
f38c39d6 PM |
286 | static int red_dump_class(struct Qdisc *sch, unsigned long cl, |
287 | struct sk_buff *skb, struct tcmsg *tcm) | |
288 | { | |
289 | struct red_sched_data *q = qdisc_priv(sch); | |
290 | ||
f38c39d6 PM |
291 | tcm->tcm_handle |= TC_H_MIN(1); |
292 | tcm->tcm_info = q->qdisc->handle; | |
293 | return 0; | |
294 | } | |
295 | ||
296 | static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
297 | struct Qdisc **old) | |
298 | { | |
299 | struct red_sched_data *q = qdisc_priv(sch); | |
300 | ||
301 | if (new == NULL) | |
302 | new = &noop_qdisc; | |
303 | ||
86a7996c | 304 | *old = qdisc_replace(sch, new, &q->qdisc); |
f38c39d6 PM |
305 | return 0; |
306 | } | |
307 | ||
308 | static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg) | |
309 | { | |
310 | struct red_sched_data *q = qdisc_priv(sch); | |
311 | return q->qdisc; | |
312 | } | |
313 | ||
314 | static unsigned long red_get(struct Qdisc *sch, u32 classid) | |
315 | { | |
316 | return 1; | |
317 | } | |
318 | ||
319 | static void red_put(struct Qdisc *sch, unsigned long arg) | |
320 | { | |
f38c39d6 PM |
321 | } |
322 | ||
f38c39d6 PM |
323 | static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker) |
324 | { | |
325 | if (!walker->stop) { | |
326 | if (walker->count >= walker->skip) | |
327 | if (walker->fn(sch, 1, walker) < 0) { | |
328 | walker->stop = 1; | |
329 | return; | |
330 | } | |
331 | walker->count++; | |
332 | } | |
333 | } | |
334 | ||
20fea08b | 335 | static const struct Qdisc_class_ops red_class_ops = { |
f38c39d6 PM |
336 | .graft = red_graft, |
337 | .leaf = red_leaf, | |
338 | .get = red_get, | |
339 | .put = red_put, | |
f38c39d6 | 340 | .walk = red_walk, |
f38c39d6 PM |
341 | .dump = red_dump_class, |
342 | }; | |
343 | ||
20fea08b | 344 | static struct Qdisc_ops red_qdisc_ops __read_mostly = { |
1da177e4 LT |
345 | .id = "red", |
346 | .priv_size = sizeof(struct red_sched_data), | |
f38c39d6 | 347 | .cl_ops = &red_class_ops, |
1da177e4 LT |
348 | .enqueue = red_enqueue, |
349 | .dequeue = red_dequeue, | |
8e3af978 | 350 | .peek = red_peek, |
1da177e4 LT |
351 | .init = red_init, |
352 | .reset = red_reset, | |
f38c39d6 | 353 | .destroy = red_destroy, |
1da177e4 LT |
354 | .change = red_change, |
355 | .dump = red_dump, | |
356 | .dump_stats = red_dump_stats, | |
357 | .owner = THIS_MODULE, | |
358 | }; | |
359 | ||
360 | static int __init red_module_init(void) | |
361 | { | |
362 | return register_qdisc(&red_qdisc_ops); | |
363 | } | |
dba051f3 TG |
364 | |
365 | static void __exit red_module_exit(void) | |
1da177e4 LT |
366 | { |
367 | unregister_qdisc(&red_qdisc_ops); | |
368 | } | |
dba051f3 | 369 | |
1da177e4 LT |
370 | module_init(red_module_init) |
371 | module_exit(red_module_exit) | |
dba051f3 | 372 | |
1da177e4 | 373 | MODULE_LICENSE("GPL"); |