00db53eb8159ecadc156fe1eab669cf179d4008d
[linux-2.6.git] / net / sched / sch_red.c
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:
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
15  */
16
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/netdevice.h>
21 #include <linux/skbuff.h>
22 #include <net/pkt_sched.h>
23 #include <net/inet_ecn.h>
24 #include <net/red.h>
25
26
27 /*      Parameters, settable by user:
28         -----------------------------
29
30         limit           - bytes (must be > qth_max + burst)
31
32         Hard limit on queue length, should be chosen >qth_max
33         to allow packet bursts. This parameter does not
34         affect the algorithms behaviour and can be chosen
35         arbitrarily high (well, less than ram size)
36         Really, this limit will never be reached
37         if RED works correctly.
38  */
39
40 struct red_sched_data
41 {
42         u32                     limit;          /* HARD maximal queue length */
43         unsigned char           flags;
44         struct red_parms        parms;
45         struct red_stats        stats;
46         struct Qdisc            *qdisc;
47 };
48
49 static inline int red_use_ecn(struct red_sched_data *q)
50 {
51         return q->flags & TC_RED_ECN;
52 }
53
54 static inline int red_use_harddrop(struct red_sched_data *q)
55 {
56         return q->flags & TC_RED_HARDDROP;
57 }
58
59 static int red_enqueue(struct sk_buff *skb, struct Qdisc* sch)
60 {
61         struct red_sched_data *q = qdisc_priv(sch);
62         struct Qdisc *child = q->qdisc;
63         int ret;
64
65         q->parms.qavg = red_calc_qavg(&q->parms, child->qstats.backlog);
66
67         if (red_is_idling(&q->parms))
68                 red_end_of_idle_period(&q->parms);
69
70         switch (red_action(&q->parms, q->parms.qavg)) {
71                 case RED_DONT_MARK:
72                         break;
73
74                 case RED_PROB_MARK:
75                         sch->qstats.overlimits++;
76                         if (!red_use_ecn(q) || !INET_ECN_set_ce(skb)) {
77                                 q->stats.prob_drop++;
78                                 goto congestion_drop;
79                         }
80
81                         q->stats.prob_mark++;
82                         break;
83
84                 case RED_HARD_MARK:
85                         sch->qstats.overlimits++;
86                         if (red_use_harddrop(q) || !red_use_ecn(q) ||
87                             !INET_ECN_set_ce(skb)) {
88                                 q->stats.forced_drop++;
89                                 goto congestion_drop;
90                         }
91
92                         q->stats.forced_mark++;
93                         break;
94         }
95
96         ret = child->enqueue(skb, child);
97         if (likely(ret == NET_XMIT_SUCCESS)) {
98                 sch->bstats.bytes += skb->len;
99                 sch->bstats.packets++;
100                 sch->q.qlen++;
101         } else {
102                 q->stats.pdrop++;
103                 sch->qstats.drops++;
104         }
105         return ret;
106
107 congestion_drop:
108         qdisc_drop(skb, sch);
109         return NET_XMIT_CN;
110 }
111
112 static int red_requeue(struct sk_buff *skb, struct Qdisc* sch)
113 {
114         struct red_sched_data *q = qdisc_priv(sch);
115         struct Qdisc *child = q->qdisc;
116         int ret;
117
118         if (red_is_idling(&q->parms))
119                 red_end_of_idle_period(&q->parms);
120
121         ret = child->ops->requeue(skb, child);
122         if (likely(ret == NET_XMIT_SUCCESS)) {
123                 sch->qstats.requeues++;
124                 sch->q.qlen++;
125         }
126         return ret;
127 }
128
129 static struct sk_buff * red_dequeue(struct Qdisc* sch)
130 {
131         struct sk_buff *skb;
132         struct red_sched_data *q = qdisc_priv(sch);
133         struct Qdisc *child = q->qdisc;
134
135         skb = child->dequeue(child);
136         if (skb)
137                 sch->q.qlen--;
138         else if (!red_is_idling(&q->parms))
139                 red_start_of_idle_period(&q->parms);
140
141         return skb;
142 }
143
144 static unsigned int red_drop(struct Qdisc* sch)
145 {
146         struct red_sched_data *q = qdisc_priv(sch);
147         struct Qdisc *child = q->qdisc;
148         unsigned int len;
149
150         if (child->ops->drop && (len = child->ops->drop(child)) > 0) {
151                 q->stats.other++;
152                 sch->qstats.drops++;
153                 sch->q.qlen--;
154                 return len;
155         }
156
157         if (!red_is_idling(&q->parms))
158                 red_start_of_idle_period(&q->parms);
159
160         return 0;
161 }
162
163 static void red_reset(struct Qdisc* sch)
164 {
165         struct red_sched_data *q = qdisc_priv(sch);
166
167         qdisc_reset(q->qdisc);
168         sch->q.qlen = 0;
169         red_restart(&q->parms);
170 }
171
172 static void red_destroy(struct Qdisc *sch)
173 {
174         struct red_sched_data *q = qdisc_priv(sch);
175         qdisc_destroy(q->qdisc);
176 }
177
178 static struct Qdisc *red_create_dflt(struct Qdisc *sch, u32 limit)
179 {
180         struct Qdisc *q;
181         struct rtattr *rta;
182         int ret;
183
184         q = qdisc_create_dflt(sch->dev, &bfifo_qdisc_ops,
185                               TC_H_MAKE(sch->handle, 1));
186         if (q) {
187                 rta = kmalloc(RTA_LENGTH(sizeof(struct tc_fifo_qopt)),
188                               GFP_KERNEL);
189                 if (rta) {
190                         rta->rta_type = RTM_NEWQDISC;
191                         rta->rta_len = RTA_LENGTH(sizeof(struct tc_fifo_qopt));
192                         ((struct tc_fifo_qopt *)RTA_DATA(rta))->limit = limit;
193
194                         ret = q->ops->change(q, rta);
195                         kfree(rta);
196
197                         if (ret == 0)
198                                 return q;
199                 }
200                 qdisc_destroy(q);
201         }
202         return NULL;
203 }
204
205 static int red_change(struct Qdisc *sch, struct rtattr *opt)
206 {
207         struct red_sched_data *q = qdisc_priv(sch);
208         struct rtattr *tb[TCA_RED_MAX];
209         struct tc_red_qopt *ctl;
210         struct Qdisc *child = NULL;
211
212         if (opt == NULL || rtattr_parse_nested(tb, TCA_RED_MAX, opt))
213                 return -EINVAL;
214
215         if (tb[TCA_RED_PARMS-1] == NULL ||
216             RTA_PAYLOAD(tb[TCA_RED_PARMS-1]) < sizeof(*ctl) ||
217             tb[TCA_RED_STAB-1] == NULL ||
218             RTA_PAYLOAD(tb[TCA_RED_STAB-1]) < RED_STAB_SIZE)
219                 return -EINVAL;
220
221         ctl = RTA_DATA(tb[TCA_RED_PARMS-1]);
222
223         if (ctl->limit > 0) {
224                 child = red_create_dflt(sch, ctl->limit);
225                 if (child == NULL)
226                         return -ENOMEM;
227         }
228
229         sch_tree_lock(sch);
230         q->flags = ctl->flags;
231         q->limit = ctl->limit;
232         if (child) {
233                 qdisc_tree_decrease_qlen(q->qdisc, q->qdisc->q.qlen);
234                 qdisc_destroy(xchg(&q->qdisc, child));
235         }
236
237         red_set_parms(&q->parms, ctl->qth_min, ctl->qth_max, ctl->Wlog,
238                                  ctl->Plog, ctl->Scell_log,
239                                  RTA_DATA(tb[TCA_RED_STAB-1]));
240
241         if (skb_queue_empty(&sch->q))
242                 red_end_of_idle_period(&q->parms);
243
244         sch_tree_unlock(sch);
245         return 0;
246 }
247
248 static int red_init(struct Qdisc* sch, struct rtattr *opt)
249 {
250         struct red_sched_data *q = qdisc_priv(sch);
251
252         q->qdisc = &noop_qdisc;
253         return red_change(sch, opt);
254 }
255
256 static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
257 {
258         struct red_sched_data *q = qdisc_priv(sch);
259         struct rtattr *opts = NULL;
260         struct tc_red_qopt opt = {
261                 .limit          = q->limit,
262                 .flags          = q->flags,
263                 .qth_min        = q->parms.qth_min >> q->parms.Wlog,
264                 .qth_max        = q->parms.qth_max >> q->parms.Wlog,
265                 .Wlog           = q->parms.Wlog,
266                 .Plog           = q->parms.Plog,
267                 .Scell_log      = q->parms.Scell_log,
268         };
269
270         opts = RTA_NEST(skb, TCA_OPTIONS);
271         RTA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt);
272         return RTA_NEST_END(skb, opts);
273
274 rtattr_failure:
275         return RTA_NEST_CANCEL(skb, opts);
276 }
277
278 static int red_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
279 {
280         struct red_sched_data *q = qdisc_priv(sch);
281         struct tc_red_xstats st = {
282                 .early  = q->stats.prob_drop + q->stats.forced_drop,
283                 .pdrop  = q->stats.pdrop,
284                 .other  = q->stats.other,
285                 .marked = q->stats.prob_mark + q->stats.forced_mark,
286         };
287
288         return gnet_stats_copy_app(d, &st, sizeof(st));
289 }
290
291 static int red_dump_class(struct Qdisc *sch, unsigned long cl,
292                           struct sk_buff *skb, struct tcmsg *tcm)
293 {
294         struct red_sched_data *q = qdisc_priv(sch);
295
296         if (cl != 1)
297                 return -ENOENT;
298         tcm->tcm_handle |= TC_H_MIN(1);
299         tcm->tcm_info = q->qdisc->handle;
300         return 0;
301 }
302
303 static int red_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
304                      struct Qdisc **old)
305 {
306         struct red_sched_data *q = qdisc_priv(sch);
307
308         if (new == NULL)
309                 new = &noop_qdisc;
310
311         sch_tree_lock(sch);
312         *old = xchg(&q->qdisc, new);
313         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
314         qdisc_reset(*old);
315         sch_tree_unlock(sch);
316         return 0;
317 }
318
319 static struct Qdisc *red_leaf(struct Qdisc *sch, unsigned long arg)
320 {
321         struct red_sched_data *q = qdisc_priv(sch);
322         return q->qdisc;
323 }
324
325 static unsigned long red_get(struct Qdisc *sch, u32 classid)
326 {
327         return 1;
328 }
329
330 static void red_put(struct Qdisc *sch, unsigned long arg)
331 {
332         return;
333 }
334
335 static int red_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
336                             struct rtattr **tca, unsigned long *arg)
337 {
338         return -ENOSYS;
339 }
340
341 static int red_delete(struct Qdisc *sch, unsigned long cl)
342 {
343         return -ENOSYS;
344 }
345
346 static void red_walk(struct Qdisc *sch, struct qdisc_walker *walker)
347 {
348         if (!walker->stop) {
349                 if (walker->count >= walker->skip)
350                         if (walker->fn(sch, 1, walker) < 0) {
351                                 walker->stop = 1;
352                                 return;
353                         }
354                 walker->count++;
355         }
356 }
357
358 static struct tcf_proto **red_find_tcf(struct Qdisc *sch, unsigned long cl)
359 {
360         return NULL;
361 }
362
363 static struct Qdisc_class_ops red_class_ops = {
364         .graft          =       red_graft,
365         .leaf           =       red_leaf,
366         .get            =       red_get,
367         .put            =       red_put,
368         .change         =       red_change_class,
369         .delete         =       red_delete,
370         .walk           =       red_walk,
371         .tcf_chain      =       red_find_tcf,
372         .dump           =       red_dump_class,
373 };
374
375 static struct Qdisc_ops red_qdisc_ops = {
376         .id             =       "red",
377         .priv_size      =       sizeof(struct red_sched_data),
378         .cl_ops         =       &red_class_ops,
379         .enqueue        =       red_enqueue,
380         .dequeue        =       red_dequeue,
381         .requeue        =       red_requeue,
382         .drop           =       red_drop,
383         .init           =       red_init,
384         .reset          =       red_reset,
385         .destroy        =       red_destroy,
386         .change         =       red_change,
387         .dump           =       red_dump,
388         .dump_stats     =       red_dump_stats,
389         .owner          =       THIS_MODULE,
390 };
391
392 static int __init red_module_init(void)
393 {
394         return register_qdisc(&red_qdisc_ops);
395 }
396
397 static void __exit red_module_exit(void)
398 {
399         unregister_qdisc(&red_qdisc_ops);
400 }
401
402 module_init(red_module_init)
403 module_exit(red_module_exit)
404
405 MODULE_LICENSE("GPL");