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 | ||
39 | struct red_sched_data | |
40 | { | |
6b31b28a TG |
41 | u32 limit; /* HARD maximal queue length */ |
42 | unsigned char flags; | |
43 | struct red_parms parms; | |
44 | struct red_stats stats; | |
f38c39d6 | 45 | struct Qdisc *qdisc; |
1da177e4 LT |
46 | }; |
47 | ||
6b31b28a | 48 | static inline int red_use_ecn(struct red_sched_data *q) |
1da177e4 | 49 | { |
6b31b28a | 50 | return q->flags & TC_RED_ECN; |
1da177e4 LT |
51 | } |
52 | ||
bdc450a0 TG |
53 | static inline int red_use_harddrop(struct red_sched_data *q) |
54 | { | |
55 | return q->flags & TC_RED_HARDDROP; | |
56 | } | |
57 | ||
dba051f3 | 58 | static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch) |
1da177e4 LT |
59 | { |
60 | struct red_sched_data *q = qdisc_priv(sch); | |
f38c39d6 PM |
61 | struct Qdisc *child = q->qdisc; |
62 | int ret; | |
1da177e4 | 63 | |
f38c39d6 | 64 | q->parms.qavg = red_calc_qavg(&q->parms, child->qstats.backlog); |
1da177e4 | 65 | |
6b31b28a TG |
66 | if (red_is_idling(&q->parms)) |
67 | red_end_of_idle_period(&q->parms); | |
1da177e4 | 68 | |
6b31b28a TG |
69 | switch (red_action(&q->parms, q->parms.qavg)) { |
70 | case RED_DONT_MARK: | |
71 | break; | |
1da177e4 | 72 | |
6b31b28a TG |
73 | case RED_PROB_MARK: |
74 | sch->qstats.overlimits++; | |
75 | if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) { | |
76 | q->stats.prob_drop++; | |
77 | goto congestion_drop; | |
78 | } | |
1da177e4 | 79 | |
6b31b28a TG |
80 | q->stats.prob_mark++; |
81 | break; | |
82 | ||
83 | case RED_HARD_MARK: | |
84 | sch->qstats.overlimits++; | |
bdc450a0 TG |
85 | if (red_use_harddrop(q) || !red_use_ecn(q) || |
86 | !INET_ECN_set_ce(skb)) { | |
6b31b28a TG |
87 | q->stats.forced_drop++; |
88 | goto congestion_drop; | |
89 | } | |
90 | ||
91 | q->stats.forced_mark++; | |
92 | break; | |
1da177e4 LT |
93 | } |
94 | ||
f38c39d6 PM |
95 | ret = child->enqueue(skb, child); |
96 | if (likely(ret == NET_XMIT_SUCCESS)) { | |
97 | sch->bstats.bytes += skb->len; | |
98 | sch->bstats.packets++; | |
99 | sch->q.qlen++; | |
100 | } else { | |
101 | q->stats.pdrop++; | |
102 | sch->qstats.drops++; | |
103 | } | |
104 | return ret; | |
6b31b28a TG |
105 | |
106 | congestion_drop: | |
9e178ff2 | 107 | qdisc_drop(skb, sch); |
1da177e4 LT |
108 | return NET_XMIT_CN; |
109 | } | |
110 | ||
dba051f3 | 111 | static int red_requeue(struct sk_buff *skb, struct Qdisc* sch) |
1da177e4 LT |
112 | { |
113 | struct red_sched_data *q = qdisc_priv(sch); | |
f38c39d6 PM |
114 | struct Qdisc *child = q->qdisc; |
115 | int ret; | |
1da177e4 | 116 | |
6b31b28a TG |
117 | if (red_is_idling(&q->parms)) |
118 | red_end_of_idle_period(&q->parms); | |
1da177e4 | 119 | |
f38c39d6 PM |
120 | ret = child->ops->requeue(skb, child); |
121 | if (likely(ret == NET_XMIT_SUCCESS)) { | |
122 | sch->qstats.requeues++; | |
123 | sch->q.qlen++; | |
124 | } | |
125 | return ret; | |
1da177e4 LT |
126 | } |
127 | ||
dba051f3 | 128 | static struct sk_buff * red_dequeue(struct Qdisc* sch) |
1da177e4 LT |
129 | { |
130 | struct sk_buff *skb; | |
131 | struct red_sched_data *q = qdisc_priv(sch); | |
f38c39d6 | 132 | struct Qdisc *child = q->qdisc; |
1da177e4 | 133 | |
f38c39d6 PM |
134 | skb = child->dequeue(child); |
135 | if (skb) | |
136 | sch->q.qlen--; | |
137 | else if (!red_is_idling(&q->parms)) | |
9e178ff2 TG |
138 | red_start_of_idle_period(&q->parms); |
139 | ||
140 | return skb; | |
1da177e4 LT |
141 | } |
142 | ||
143 | static unsigned int red_drop(struct Qdisc* sch) | |
144 | { | |
1da177e4 | 145 | struct red_sched_data *q = qdisc_priv(sch); |
f38c39d6 PM |
146 | struct Qdisc *child = q->qdisc; |
147 | unsigned int len; | |
1da177e4 | 148 | |
f38c39d6 | 149 | if (child->ops->drop && (len = child->ops->drop(child)) > 0) { |
6b31b28a | 150 | q->stats.other++; |
f38c39d6 PM |
151 | sch->qstats.drops++; |
152 | sch->q.qlen--; | |
1da177e4 LT |
153 | return len; |
154 | } | |
6b31b28a | 155 | |
6a1b63d4 TG |
156 | if (!red_is_idling(&q->parms)) |
157 | red_start_of_idle_period(&q->parms); | |
158 | ||
1da177e4 LT |
159 | return 0; |
160 | } | |
161 | ||
162 | static void red_reset(struct Qdisc* sch) | |
163 | { | |
164 | struct red_sched_data *q = qdisc_priv(sch); | |
165 | ||
f38c39d6 PM |
166 | qdisc_reset(q->qdisc); |
167 | sch->q.qlen = 0; | |
6b31b28a | 168 | red_restart(&q->parms); |
1da177e4 LT |
169 | } |
170 | ||
f38c39d6 PM |
171 | static void red_destroy(struct Qdisc *sch) |
172 | { | |
173 | struct red_sched_data *q = qdisc_priv(sch); | |
174 | qdisc_destroy(q->qdisc); | |
175 | } | |
176 | ||
9f9afec4 | 177 | static struct Qdisc *red_create_dflt(struct Qdisc *sch, u32 limit) |
f38c39d6 | 178 | { |
9f9afec4 | 179 | struct Qdisc *q; |
1e90474c | 180 | struct nlattr *nla; |
f38c39d6 PM |
181 | int ret; |
182 | ||
9f9afec4 PM |
183 | q = qdisc_create_dflt(sch->dev, &bfifo_qdisc_ops, |
184 | TC_H_MAKE(sch->handle, 1)); | |
f38c39d6 | 185 | if (q) { |
1e90474c | 186 | nla = kmalloc(nla_attr_size(sizeof(struct tc_fifo_qopt)), |
10297b99 | 187 | GFP_KERNEL); |
1e90474c PM |
188 | if (nla) { |
189 | nla->nla_type = RTM_NEWQDISC; | |
190 | nla->nla_len = nla_attr_size(sizeof(struct tc_fifo_qopt)); | |
191 | ((struct tc_fifo_qopt *)nla_data(nla))->limit = limit; | |
f38c39d6 | 192 | |
1e90474c PM |
193 | ret = q->ops->change(q, nla); |
194 | kfree(nla); | |
f38c39d6 PM |
195 | |
196 | if (ret == 0) | |
197 | return q; | |
198 | } | |
199 | qdisc_destroy(q); | |
200 | } | |
201 | return NULL; | |
202 | } | |
203 | ||
27a3421e PM |
204 | static const struct nla_policy red_policy[TCA_RED_MAX + 1] = { |
205 | [TCA_RED_PARMS] = { .len = sizeof(struct tc_red_qopt) }, | |
206 | [TCA_RED_STAB] = { .len = RED_STAB_SIZE }, | |
207 | }; | |
208 | ||
1e90474c | 209 | static int red_change(struct Qdisc *sch, struct nlattr *opt) |
1da177e4 LT |
210 | { |
211 | struct red_sched_data *q = qdisc_priv(sch); | |
1e90474c | 212 | struct nlattr *tb[TCA_RED_MAX + 1]; |
1da177e4 | 213 | struct tc_red_qopt *ctl; |
f38c39d6 | 214 | struct Qdisc *child = NULL; |
cee63723 | 215 | int err; |
1da177e4 | 216 | |
cee63723 | 217 | if (opt == NULL) |
dba051f3 TG |
218 | return -EINVAL; |
219 | ||
27a3421e | 220 | err = nla_parse_nested(tb, TCA_RED_MAX, opt, red_policy); |
cee63723 PM |
221 | if (err < 0) |
222 | return err; | |
223 | ||
1e90474c | 224 | if (tb[TCA_RED_PARMS] == NULL || |
27a3421e | 225 | tb[TCA_RED_STAB] == NULL) |
1da177e4 LT |
226 | return -EINVAL; |
227 | ||
1e90474c | 228 | ctl = nla_data(tb[TCA_RED_PARMS]); |
1da177e4 | 229 | |
f38c39d6 | 230 | if (ctl->limit > 0) { |
9f9afec4 | 231 | child = red_create_dflt(sch, ctl->limit); |
f38c39d6 PM |
232 | if (child == NULL) |
233 | return -ENOMEM; | |
234 | } | |
235 | ||
1da177e4 LT |
236 | sch_tree_lock(sch); |
237 | q->flags = ctl->flags; | |
1da177e4 | 238 | q->limit = ctl->limit; |
5e50da01 PM |
239 | if (child) { |
240 | qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen); | |
f38c39d6 | 241 | qdisc_destroy(xchg(&q->qdisc, child)); |
5e50da01 | 242 | } |
1da177e4 | 243 | |
6b31b28a TG |
244 | red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog, |
245 | ctl->Plog, ctl->Scell_log, | |
1e90474c | 246 | nla_data(tb[TCA_RED_STAB])); |
6b31b28a | 247 | |
b03efcfb | 248 | if (skb_queue_empty(&sch->q)) |
6b31b28a | 249 | red_end_of_idle_period(&q->parms); |
dba051f3 | 250 | |
1da177e4 LT |
251 | sch_tree_unlock(sch); |
252 | return 0; | |
253 | } | |
254 | ||
1e90474c | 255 | static int red_init(struct Qdisc* sch, struct nlattr *opt) |
1da177e4 | 256 | { |
f38c39d6 PM |
257 | struct red_sched_data *q = qdisc_priv(sch); |
258 | ||
259 | q->qdisc = &noop_qdisc; | |
1da177e4 LT |
260 | return red_change(sch, opt); |
261 | } | |
262 | ||
263 | static int red_dump(struct Qdisc *sch, struct sk_buff *skb) | |
264 | { | |
265 | struct red_sched_data *q = qdisc_priv(sch); | |
1e90474c | 266 | struct nlattr *opts = NULL; |
6b31b28a TG |
267 | struct tc_red_qopt opt = { |
268 | .limit = q->limit, | |
269 | .flags = q->flags, | |
270 | .qth_min = q->parms.qth_min >> q->parms.Wlog, | |
271 | .qth_max = q->parms.qth_max >> q->parms.Wlog, | |
272 | .Wlog = q->parms.Wlog, | |
273 | .Plog = q->parms.Plog, | |
274 | .Scell_log = q->parms.Scell_log, | |
275 | }; | |
1da177e4 | 276 | |
1e90474c PM |
277 | opts = nla_nest_start(skb, TCA_OPTIONS); |
278 | if (opts == NULL) | |
279 | goto nla_put_failure; | |
280 | NLA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt); | |
281 | return nla_nest_end(skb, opts); | |
1da177e4 | 282 | |
1e90474c | 283 | nla_put_failure: |
bc3ed28c TG |
284 | nla_nest_cancel(skb, opts); |
285 | return -EMSGSIZE; | |
1da177e4 LT |
286 | } |
287 | ||
288 | static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d) | |
289 | { | |
290 | struct red_sched_data *q = qdisc_priv(sch); | |
6b31b28a TG |
291 | struct tc_red_xstats st = { |
292 | .early = q->stats.prob_drop + q->stats.forced_drop, | |
293 | .pdrop = q->stats.pdrop, | |
294 | .other = q->stats.other, | |
295 | .marked = q->stats.prob_mark + q->stats.forced_mark, | |
296 | }; | |
297 | ||
298 | return gnet_stats_copy_app(d, &st, sizeof(st)); | |
1da177e4 LT |
299 | } |
300 | ||
f38c39d6 PM |
301 | static int red_dump_class(struct Qdisc *sch, unsigned long cl, |
302 | struct sk_buff *skb, struct tcmsg *tcm) | |
303 | { | |
304 | struct red_sched_data *q = qdisc_priv(sch); | |
305 | ||
306 | if (cl != 1) | |
307 | return -ENOENT; | |
308 | tcm->tcm_handle |= TC_H_MIN(1); | |
309 | tcm->tcm_info = q->qdisc->handle; | |
310 | return 0; | |
311 | } | |
312 | ||
313 | static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, | |
314 | struct Qdisc **old) | |
315 | { | |
316 | struct red_sched_data *q = qdisc_priv(sch); | |
317 | ||
318 | if (new == NULL) | |
319 | new = &noop_qdisc; | |
320 | ||
321 | sch_tree_lock(sch); | |
322 | *old = xchg(&q->qdisc, new); | |
5e50da01 | 323 | qdisc_tree_decrease_qlen(*old, (*old)->q.qlen); |
f38c39d6 | 324 | qdisc_reset(*old); |
f38c39d6 PM |
325 | sch_tree_unlock(sch); |
326 | return 0; | |
327 | } | |
328 | ||
329 | static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg) | |
330 | { | |
331 | struct red_sched_data *q = qdisc_priv(sch); | |
332 | return q->qdisc; | |
333 | } | |
334 | ||
335 | static unsigned long red_get(struct Qdisc *sch, u32 classid) | |
336 | { | |
337 | return 1; | |
338 | } | |
339 | ||
340 | static void red_put(struct Qdisc *sch, unsigned long arg) | |
341 | { | |
342 | return; | |
343 | } | |
344 | ||
345 | static int red_change_class(struct Qdisc *sch, u32 classid, u32 parentid, | |
1e90474c | 346 | struct nlattr **tca, unsigned long *arg) |
f38c39d6 PM |
347 | { |
348 | return -ENOSYS; | |
349 | } | |
350 | ||
351 | static int red_delete(struct Qdisc *sch, unsigned long cl) | |
352 | { | |
353 | return -ENOSYS; | |
354 | } | |
355 | ||
356 | static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker) | |
357 | { | |
358 | if (!walker->stop) { | |
359 | if (walker->count >= walker->skip) | |
360 | if (walker->fn(sch, 1, walker) < 0) { | |
361 | walker->stop = 1; | |
362 | return; | |
363 | } | |
364 | walker->count++; | |
365 | } | |
366 | } | |
367 | ||
368 | static struct tcf_proto **red_find_tcf(struct Qdisc *sch, unsigned long cl) | |
369 | { | |
370 | return NULL; | |
371 | } | |
372 | ||
20fea08b | 373 | static const struct Qdisc_class_ops red_class_ops = { |
f38c39d6 PM |
374 | .graft = red_graft, |
375 | .leaf = red_leaf, | |
376 | .get = red_get, | |
377 | .put = red_put, | |
378 | .change = red_change_class, | |
379 | .delete = red_delete, | |
380 | .walk = red_walk, | |
381 | .tcf_chain = red_find_tcf, | |
382 | .dump = red_dump_class, | |
383 | }; | |
384 | ||
20fea08b | 385 | static struct Qdisc_ops red_qdisc_ops __read_mostly = { |
1da177e4 LT |
386 | .id = "red", |
387 | .priv_size = sizeof(struct red_sched_data), | |
f38c39d6 | 388 | .cl_ops = &red_class_ops, |
1da177e4 LT |
389 | .enqueue = red_enqueue, |
390 | .dequeue = red_dequeue, | |
391 | .requeue = red_requeue, | |
392 | .drop = red_drop, | |
393 | .init = red_init, | |
394 | .reset = red_reset, | |
f38c39d6 | 395 | .destroy = red_destroy, |
1da177e4 LT |
396 | .change = red_change, |
397 | .dump = red_dump, | |
398 | .dump_stats = red_dump_stats, | |
399 | .owner = THIS_MODULE, | |
400 | }; | |
401 | ||
402 | static int __init red_module_init(void) | |
403 | { | |
404 | return register_qdisc(&red_qdisc_ops); | |
405 | } | |
dba051f3 TG |
406 | |
407 | static void __exit red_module_exit(void) | |
1da177e4 LT |
408 | { |
409 | unregister_qdisc(&red_qdisc_ops); | |
410 | } | |
dba051f3 | 411 | |
1da177e4 LT |
412 | module_init(red_module_init) |
413 | module_exit(red_module_exit) | |
dba051f3 | 414 | |
1da177e4 | 415 | MODULE_LICENSE("GPL"); |