3651da3e280211fddfa167dbe595283af399ac29
[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
97 static int
98 prio_requeue(struct sk_buff *skb, struct Qdisc* sch)
99 {
100         struct Qdisc *qdisc;
101         int ret;
102
103         qdisc = prio_classify(skb, sch, &ret);
104 #ifdef CONFIG_NET_CLS_ACT
105         if (qdisc == NULL) {
106                 if (ret & __NET_XMIT_BYPASS)
107                         sch->qstats.drops++;
108                 kfree_skb(skb);
109                 return ret;
110         }
111 #endif
112
113         if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) {
114                 sch->q.qlen++;
115                 sch->qstats.requeues++;
116                 return NET_XMIT_SUCCESS;
117         }
118         if (net_xmit_drop_count(ret))
119                 sch->qstats.drops++;
120         return ret;
121 }
122
123 static struct sk_buff *prio_peek(struct Qdisc *sch)
124 {
125         struct prio_sched_data *q = qdisc_priv(sch);
126         int prio;
127
128         for (prio = 0; prio < q->bands; prio++) {
129                 struct Qdisc *qdisc = q->queues[prio];
130                 struct sk_buff *skb = qdisc->ops->peek(qdisc);
131                 if (skb)
132                         return skb;
133         }
134         return NULL;
135 }
136
137 static struct sk_buff *prio_dequeue(struct Qdisc* sch)
138 {
139         struct prio_sched_data *q = qdisc_priv(sch);
140         int prio;
141
142         for (prio = 0; prio < q->bands; prio++) {
143                 struct Qdisc *qdisc = q->queues[prio];
144                 struct sk_buff *skb = qdisc->dequeue(qdisc);
145                 if (skb) {
146                         sch->q.qlen--;
147                         return skb;
148                 }
149         }
150         return NULL;
151
152 }
153
154 static unsigned int prio_drop(struct Qdisc* sch)
155 {
156         struct prio_sched_data *q = qdisc_priv(sch);
157         int prio;
158         unsigned int len;
159         struct Qdisc *qdisc;
160
161         for (prio = q->bands-1; prio >= 0; prio--) {
162                 qdisc = q->queues[prio];
163                 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
164                         sch->q.qlen--;
165                         return len;
166                 }
167         }
168         return 0;
169 }
170
171
172 static void
173 prio_reset(struct Qdisc* sch)
174 {
175         int prio;
176         struct prio_sched_data *q = qdisc_priv(sch);
177
178         for (prio=0; prio<q->bands; prio++)
179                 qdisc_reset(q->queues[prio]);
180         sch->q.qlen = 0;
181 }
182
183 static void
184 prio_destroy(struct Qdisc* sch)
185 {
186         int prio;
187         struct prio_sched_data *q = qdisc_priv(sch);
188
189         tcf_destroy_chain(&q->filter_list);
190         for (prio=0; prio<q->bands; prio++)
191                 qdisc_destroy(q->queues[prio]);
192 }
193
194 static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
195 {
196         struct prio_sched_data *q = qdisc_priv(sch);
197         struct tc_prio_qopt *qopt;
198         int i;
199
200         if (nla_len(opt) < sizeof(*qopt))
201                 return -EINVAL;
202         qopt = nla_data(opt);
203
204         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
205                 return -EINVAL;
206
207         for (i=0; i<=TC_PRIO_MAX; i++) {
208                 if (qopt->priomap[i] >= qopt->bands)
209                         return -EINVAL;
210         }
211
212         sch_tree_lock(sch);
213         q->bands = qopt->bands;
214         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
215
216         for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
217                 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
218                 if (child != &noop_qdisc) {
219                         qdisc_tree_decrease_qlen(child, child->q.qlen);
220                         qdisc_destroy(child);
221                 }
222         }
223         sch_tree_unlock(sch);
224
225         for (i=0; i<q->bands; i++) {
226                 if (q->queues[i] == &noop_qdisc) {
227                         struct Qdisc *child;
228                         child = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
229                                                   &pfifo_qdisc_ops,
230                                                   TC_H_MAKE(sch->handle, i + 1));
231                         if (child) {
232                                 sch_tree_lock(sch);
233                                 child = xchg(&q->queues[i], child);
234
235                                 if (child != &noop_qdisc) {
236                                         qdisc_tree_decrease_qlen(child,
237                                                                  child->q.qlen);
238                                         qdisc_destroy(child);
239                                 }
240                                 sch_tree_unlock(sch);
241                         }
242                 }
243         }
244         return 0;
245 }
246
247 static int prio_init(struct Qdisc *sch, struct nlattr *opt)
248 {
249         struct prio_sched_data *q = qdisc_priv(sch);
250         int i;
251
252         for (i=0; i<TCQ_PRIO_BANDS; i++)
253                 q->queues[i] = &noop_qdisc;
254
255         if (opt == NULL) {
256                 return -EINVAL;
257         } else {
258                 int err;
259
260                 if ((err= prio_tune(sch, opt)) != 0)
261                         return err;
262         }
263         return 0;
264 }
265
266 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
267 {
268         struct prio_sched_data *q = qdisc_priv(sch);
269         unsigned char *b = skb_tail_pointer(skb);
270         struct tc_prio_qopt opt;
271
272         opt.bands = q->bands;
273         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
274
275         NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
276
277         return skb->len;
278
279 nla_put_failure:
280         nlmsg_trim(skb, b);
281         return -1;
282 }
283
284 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
285                       struct Qdisc **old)
286 {
287         struct prio_sched_data *q = qdisc_priv(sch);
288         unsigned long band = arg - 1;
289
290         if (band >= q->bands)
291                 return -EINVAL;
292
293         if (new == NULL)
294                 new = &noop_qdisc;
295
296         sch_tree_lock(sch);
297         *old = q->queues[band];
298         q->queues[band] = new;
299         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
300         qdisc_reset(*old);
301         sch_tree_unlock(sch);
302
303         return 0;
304 }
305
306 static struct Qdisc *
307 prio_leaf(struct Qdisc *sch, unsigned long arg)
308 {
309         struct prio_sched_data *q = qdisc_priv(sch);
310         unsigned long band = arg - 1;
311
312         if (band >= q->bands)
313                 return NULL;
314
315         return q->queues[band];
316 }
317
318 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
319 {
320         struct prio_sched_data *q = qdisc_priv(sch);
321         unsigned long band = TC_H_MIN(classid);
322
323         if (band - 1 >= q->bands)
324                 return 0;
325         return band;
326 }
327
328 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
329 {
330         return prio_get(sch, classid);
331 }
332
333
334 static void prio_put(struct Qdisc *q, unsigned long cl)
335 {
336         return;
337 }
338
339 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct nlattr **tca, unsigned long *arg)
340 {
341         unsigned long cl = *arg;
342         struct prio_sched_data *q = qdisc_priv(sch);
343
344         if (cl - 1 > q->bands)
345                 return -ENOENT;
346         return 0;
347 }
348
349 static int prio_delete(struct Qdisc *sch, unsigned long cl)
350 {
351         struct prio_sched_data *q = qdisc_priv(sch);
352         if (cl - 1 > q->bands)
353                 return -ENOENT;
354         return 0;
355 }
356
357
358 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
359                            struct tcmsg *tcm)
360 {
361         struct prio_sched_data *q = qdisc_priv(sch);
362
363         if (cl - 1 > q->bands)
364                 return -ENOENT;
365         tcm->tcm_handle |= TC_H_MIN(cl);
366         if (q->queues[cl-1])
367                 tcm->tcm_info = q->queues[cl-1]->handle;
368         return 0;
369 }
370
371 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
372                                  struct gnet_dump *d)
373 {
374         struct prio_sched_data *q = qdisc_priv(sch);
375         struct Qdisc *cl_q;
376
377         cl_q = q->queues[cl - 1];
378         if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
379             gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
380                 return -1;
381
382         return 0;
383 }
384
385 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
386 {
387         struct prio_sched_data *q = qdisc_priv(sch);
388         int prio;
389
390         if (arg->stop)
391                 return;
392
393         for (prio = 0; prio < q->bands; prio++) {
394                 if (arg->count < arg->skip) {
395                         arg->count++;
396                         continue;
397                 }
398                 if (arg->fn(sch, prio+1, arg) < 0) {
399                         arg->stop = 1;
400                         break;
401                 }
402                 arg->count++;
403         }
404 }
405
406 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
407 {
408         struct prio_sched_data *q = qdisc_priv(sch);
409
410         if (cl)
411                 return NULL;
412         return &q->filter_list;
413 }
414
415 static const struct Qdisc_class_ops prio_class_ops = {
416         .graft          =       prio_graft,
417         .leaf           =       prio_leaf,
418         .get            =       prio_get,
419         .put            =       prio_put,
420         .change         =       prio_change,
421         .delete         =       prio_delete,
422         .walk           =       prio_walk,
423         .tcf_chain      =       prio_find_tcf,
424         .bind_tcf       =       prio_bind,
425         .unbind_tcf     =       prio_put,
426         .dump           =       prio_dump_class,
427         .dump_stats     =       prio_dump_class_stats,
428 };
429
430 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
431         .next           =       NULL,
432         .cl_ops         =       &prio_class_ops,
433         .id             =       "prio",
434         .priv_size      =       sizeof(struct prio_sched_data),
435         .enqueue        =       prio_enqueue,
436         .dequeue        =       prio_dequeue,
437         .peek           =       prio_peek,
438         .requeue        =       prio_requeue,
439         .drop           =       prio_drop,
440         .init           =       prio_init,
441         .reset          =       prio_reset,
442         .destroy        =       prio_destroy,
443         .change         =       prio_tune,
444         .dump           =       prio_dump,
445         .owner          =       THIS_MODULE,
446 };
447
448 static int __init prio_module_init(void)
449 {
450         return register_qdisc(&prio_qdisc_ops);
451 }
452
453 static void __exit prio_module_exit(void)
454 {
455         unregister_qdisc(&prio_qdisc_ops);
456 }
457
458 module_init(prio_module_init)
459 module_exit(prio_module_exit)
460
461 MODULE_LICENSE("GPL");