94cecef7014517b5784bdc297c36a8fad644ca28
[linux-2.6.git] / net / sched / sch_prio.c
1 /*
2  * net/sched/sch_prio.c Simple 3-band priority "scheduler".
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  * Fixes:       19990609: J Hadi Salim <hadi@nortelnetworks.com>:
11  *              Init --  EINVAL when opt undefined
12  */
13
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <net/netlink.h>
21 #include <net/pkt_sched.h>
22
23
24 struct prio_sched_data
25 {
26         int bands;
27         struct tcf_proto *filter_list;
28         u8  prio2band[TC_PRIO_MAX+1];
29         struct Qdisc *queues[TCQ_PRIO_BANDS];
30 };
31
32
33 static struct Qdisc *
34 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
35 {
36         struct prio_sched_data *q = qdisc_priv(sch);
37         u32 band = skb->priority;
38         struct tcf_result res;
39         int err;
40
41         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
42         if (TC_H_MAJ(skb->priority) != sch->handle) {
43                 err = tc_classify(skb, q->filter_list, &res);
44 #ifdef CONFIG_NET_CLS_ACT
45                 switch (err) {
46                 case TC_ACT_STOLEN:
47                 case TC_ACT_QUEUED:
48                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
49                 case TC_ACT_SHOT:
50                         return NULL;
51                 }
52 #endif
53                 if (!q->filter_list || err < 0) {
54                         if (TC_H_MAJ(band))
55                                 band = 0;
56                         return q->queues[q->prio2band[band&TC_PRIO_MAX]];
57                 }
58                 band = res.classid;
59         }
60         band = TC_H_MIN(band) - 1;
61         if (band >= q->bands)
62                 return q->queues[q->prio2band[0]];
63
64         return q->queues[band];
65 }
66
67 static int
68 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
69 {
70         struct Qdisc *qdisc;
71         int ret;
72
73         qdisc = prio_classify(skb, sch, &ret);
74 #ifdef CONFIG_NET_CLS_ACT
75         if (qdisc == NULL) {
76
77                 if (ret & __NET_XMIT_BYPASS)
78                         sch->qstats.drops++;
79                 kfree_skb(skb);
80                 return ret;
81         }
82 #endif
83
84         ret = qdisc_enqueue(skb, qdisc);
85         if (ret == NET_XMIT_SUCCESS) {
86                 sch->bstats.bytes += qdisc_pkt_len(skb);
87                 sch->bstats.packets++;
88                 sch->q.qlen++;
89                 return NET_XMIT_SUCCESS;
90         }
91         if (net_xmit_drop_count(ret))
92                 sch->qstats.drops++;
93         return ret;
94 }
95
96 static struct sk_buff *prio_peek(struct Qdisc *sch)
97 {
98         struct prio_sched_data *q = qdisc_priv(sch);
99         int prio;
100
101         for (prio = 0; prio < q->bands; prio++) {
102                 struct Qdisc *qdisc = q->queues[prio];
103                 struct sk_buff *skb = qdisc->ops->peek(qdisc);
104                 if (skb)
105                         return skb;
106         }
107         return NULL;
108 }
109
110 static struct sk_buff *prio_dequeue(struct Qdisc* sch)
111 {
112         struct prio_sched_data *q = qdisc_priv(sch);
113         int prio;
114
115         for (prio = 0; prio < q->bands; prio++) {
116                 struct Qdisc *qdisc = q->queues[prio];
117                 struct sk_buff *skb = qdisc->dequeue(qdisc);
118                 if (skb) {
119                         sch->q.qlen--;
120                         return skb;
121                 }
122         }
123         return NULL;
124
125 }
126
127 static unsigned int prio_drop(struct Qdisc* sch)
128 {
129         struct prio_sched_data *q = qdisc_priv(sch);
130         int prio;
131         unsigned int len;
132         struct Qdisc *qdisc;
133
134         for (prio = q->bands-1; prio >= 0; prio--) {
135                 qdisc = q->queues[prio];
136                 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
137                         sch->q.qlen--;
138                         return len;
139                 }
140         }
141         return 0;
142 }
143
144
145 static void
146 prio_reset(struct Qdisc* sch)
147 {
148         int prio;
149         struct prio_sched_data *q = qdisc_priv(sch);
150
151         for (prio=0; prio<q->bands; prio++)
152                 qdisc_reset(q->queues[prio]);
153         sch->q.qlen = 0;
154 }
155
156 static void
157 prio_destroy(struct Qdisc* sch)
158 {
159         int prio;
160         struct prio_sched_data *q = qdisc_priv(sch);
161
162         tcf_destroy_chain(&q->filter_list);
163         for (prio=0; prio<q->bands; prio++)
164                 qdisc_destroy(q->queues[prio]);
165 }
166
167 static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
168 {
169         struct prio_sched_data *q = qdisc_priv(sch);
170         struct tc_prio_qopt *qopt;
171         int i;
172
173         if (nla_len(opt) < sizeof(*qopt))
174                 return -EINVAL;
175         qopt = nla_data(opt);
176
177         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
178                 return -EINVAL;
179
180         for (i=0; i<=TC_PRIO_MAX; i++) {
181                 if (qopt->priomap[i] >= qopt->bands)
182                         return -EINVAL;
183         }
184
185         sch_tree_lock(sch);
186         q->bands = qopt->bands;
187         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
188
189         for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
190                 struct Qdisc *child = q->queues[i];
191                 q->queues[i] = &noop_qdisc;
192                 if (child != &noop_qdisc) {
193                         qdisc_tree_decrease_qlen(child, child->q.qlen);
194                         qdisc_destroy(child);
195                 }
196         }
197         sch_tree_unlock(sch);
198
199         for (i=0; i<q->bands; i++) {
200                 if (q->queues[i] == &noop_qdisc) {
201                         struct Qdisc *child, *old;
202                         child = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
203                                                   &pfifo_qdisc_ops,
204                                                   TC_H_MAKE(sch->handle, i + 1));
205                         if (child) {
206                                 sch_tree_lock(sch);
207                                 old = q->queues[i];
208                                 q->queues[i] = child;
209
210                                 if (old != &noop_qdisc) {
211                                         qdisc_tree_decrease_qlen(old,
212                                                                  old->q.qlen);
213                                         qdisc_destroy(old);
214                                 }
215                                 sch_tree_unlock(sch);
216                         }
217                 }
218         }
219         return 0;
220 }
221
222 static int prio_init(struct Qdisc *sch, struct nlattr *opt)
223 {
224         struct prio_sched_data *q = qdisc_priv(sch);
225         int i;
226
227         for (i=0; i<TCQ_PRIO_BANDS; i++)
228                 q->queues[i] = &noop_qdisc;
229
230         if (opt == NULL) {
231                 return -EINVAL;
232         } else {
233                 int err;
234
235                 if ((err= prio_tune(sch, opt)) != 0)
236                         return err;
237         }
238         return 0;
239 }
240
241 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
242 {
243         struct prio_sched_data *q = qdisc_priv(sch);
244         unsigned char *b = skb_tail_pointer(skb);
245         struct tc_prio_qopt opt;
246
247         opt.bands = q->bands;
248         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
249
250         NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
251
252         return skb->len;
253
254 nla_put_failure:
255         nlmsg_trim(skb, b);
256         return -1;
257 }
258
259 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
260                       struct Qdisc **old)
261 {
262         struct prio_sched_data *q = qdisc_priv(sch);
263         unsigned long band = arg - 1;
264
265         if (band >= q->bands)
266                 return -EINVAL;
267
268         if (new == NULL)
269                 new = &noop_qdisc;
270
271         sch_tree_lock(sch);
272         *old = q->queues[band];
273         q->queues[band] = new;
274         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
275         qdisc_reset(*old);
276         sch_tree_unlock(sch);
277
278         return 0;
279 }
280
281 static struct Qdisc *
282 prio_leaf(struct Qdisc *sch, unsigned long arg)
283 {
284         struct prio_sched_data *q = qdisc_priv(sch);
285         unsigned long band = arg - 1;
286
287         if (band >= q->bands)
288                 return NULL;
289
290         return q->queues[band];
291 }
292
293 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
294 {
295         struct prio_sched_data *q = qdisc_priv(sch);
296         unsigned long band = TC_H_MIN(classid);
297
298         if (band - 1 >= q->bands)
299                 return 0;
300         return band;
301 }
302
303 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
304 {
305         return prio_get(sch, classid);
306 }
307
308
309 static void prio_put(struct Qdisc *q, unsigned long cl)
310 {
311         return;
312 }
313
314 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct nlattr **tca, unsigned long *arg)
315 {
316         unsigned long cl = *arg;
317         struct prio_sched_data *q = qdisc_priv(sch);
318
319         if (cl - 1 > q->bands)
320                 return -ENOENT;
321         return 0;
322 }
323
324 static int prio_delete(struct Qdisc *sch, unsigned long cl)
325 {
326         struct prio_sched_data *q = qdisc_priv(sch);
327         if (cl - 1 > q->bands)
328                 return -ENOENT;
329         return 0;
330 }
331
332
333 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
334                            struct tcmsg *tcm)
335 {
336         struct prio_sched_data *q = qdisc_priv(sch);
337
338         if (cl - 1 > q->bands)
339                 return -ENOENT;
340         tcm->tcm_handle |= TC_H_MIN(cl);
341         if (q->queues[cl-1])
342                 tcm->tcm_info = q->queues[cl-1]->handle;
343         return 0;
344 }
345
346 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
347                                  struct gnet_dump *d)
348 {
349         struct prio_sched_data *q = qdisc_priv(sch);
350         struct Qdisc *cl_q;
351
352         cl_q = q->queues[cl - 1];
353         if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
354             gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
355                 return -1;
356
357         return 0;
358 }
359
360 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
361 {
362         struct prio_sched_data *q = qdisc_priv(sch);
363         int prio;
364
365         if (arg->stop)
366                 return;
367
368         for (prio = 0; prio < q->bands; prio++) {
369                 if (arg->count < arg->skip) {
370                         arg->count++;
371                         continue;
372                 }
373                 if (arg->fn(sch, prio+1, arg) < 0) {
374                         arg->stop = 1;
375                         break;
376                 }
377                 arg->count++;
378         }
379 }
380
381 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
382 {
383         struct prio_sched_data *q = qdisc_priv(sch);
384
385         if (cl)
386                 return NULL;
387         return &q->filter_list;
388 }
389
390 static const struct Qdisc_class_ops prio_class_ops = {
391         .graft          =       prio_graft,
392         .leaf           =       prio_leaf,
393         .get            =       prio_get,
394         .put            =       prio_put,
395         .change         =       prio_change,
396         .delete         =       prio_delete,
397         .walk           =       prio_walk,
398         .tcf_chain      =       prio_find_tcf,
399         .bind_tcf       =       prio_bind,
400         .unbind_tcf     =       prio_put,
401         .dump           =       prio_dump_class,
402         .dump_stats     =       prio_dump_class_stats,
403 };
404
405 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
406         .next           =       NULL,
407         .cl_ops         =       &prio_class_ops,
408         .id             =       "prio",
409         .priv_size      =       sizeof(struct prio_sched_data),
410         .enqueue        =       prio_enqueue,
411         .dequeue        =       prio_dequeue,
412         .peek           =       prio_peek,
413         .drop           =       prio_drop,
414         .init           =       prio_init,
415         .reset          =       prio_reset,
416         .destroy        =       prio_destroy,
417         .change         =       prio_tune,
418         .dump           =       prio_dump,
419         .owner          =       THIS_MODULE,
420 };
421
422 static int __init prio_module_init(void)
423 {
424         return register_qdisc(&prio_qdisc_ops);
425 }
426
427 static void __exit prio_module_exit(void)
428 {
429         unregister_qdisc(&prio_qdisc_ops);
430 }
431
432 module_init(prio_module_init)
433 module_exit(prio_module_exit)
434
435 MODULE_LICENSE("GPL");