Merge master.kernel.org:/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
[linux-2.6] / 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 <asm/uaccess.h>
16 #include <asm/system.h>
17 #include <linux/bitops.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/sched.h>
21 #include <linux/string.h>
22 #include <linux/mm.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/in.h>
26 #include <linux/errno.h>
27 #include <linux/interrupt.h>
28 #include <linux/if_ether.h>
29 #include <linux/inet.h>
30 #include <linux/netdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/notifier.h>
33 #include <net/ip.h>
34 #include <net/route.h>
35 #include <linux/skbuff.h>
36 #include <net/sock.h>
37 #include <net/pkt_sched.h>
38
39
40 struct prio_sched_data
41 {
42         int bands;
43         struct tcf_proto *filter_list;
44         u8  prio2band[TC_PRIO_MAX+1];
45         struct Qdisc *queues[TCQ_PRIO_BANDS];
46 };
47
48
49 static struct Qdisc *
50 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
51 {
52         struct prio_sched_data *q = qdisc_priv(sch);
53         u32 band = skb->priority;
54         struct tcf_result res;
55
56         *qerr = NET_XMIT_BYPASS;
57         if (TC_H_MAJ(skb->priority) != sch->handle) {
58 #ifdef CONFIG_NET_CLS_ACT
59                 switch (tc_classify(skb, q->filter_list, &res)) {
60                 case TC_ACT_STOLEN:
61                 case TC_ACT_QUEUED:
62                         *qerr = NET_XMIT_SUCCESS;
63                 case TC_ACT_SHOT:
64                         return NULL;
65                 };
66
67                 if (!q->filter_list ) {
68 #else
69                 if (!q->filter_list || tc_classify(skb, q->filter_list, &res)) {
70 #endif
71                         if (TC_H_MAJ(band))
72                                 band = 0;
73                         return q->queues[q->prio2band[band&TC_PRIO_MAX]];
74                 }
75                 band = res.classid;
76         }
77         band = TC_H_MIN(band) - 1;
78         if (band > q->bands)
79                 return q->queues[q->prio2band[0]];
80
81         return q->queues[band];
82 }
83
84 static int
85 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
86 {
87         struct Qdisc *qdisc;
88         int ret;
89
90         qdisc = prio_classify(skb, sch, &ret);
91 #ifdef CONFIG_NET_CLS_ACT
92         if (qdisc == NULL) {
93
94                 if (ret == NET_XMIT_BYPASS)
95                         sch->qstats.drops++;
96                 kfree_skb(skb);
97                 return ret;
98         }
99 #endif
100
101         if ((ret = qdisc->enqueue(skb, qdisc)) == NET_XMIT_SUCCESS) {
102                 sch->bstats.bytes += skb->len;
103                 sch->bstats.packets++;
104                 sch->q.qlen++;
105                 return NET_XMIT_SUCCESS;
106         }
107         sch->qstats.drops++;
108         return ret;
109 }
110
111
112 static int
113 prio_requeue(struct sk_buff *skb, struct Qdisc* sch)
114 {
115         struct Qdisc *qdisc;
116         int ret;
117
118         qdisc = prio_classify(skb, sch, &ret);
119 #ifdef CONFIG_NET_CLS_ACT
120         if (qdisc == NULL) {
121                 if (ret == NET_XMIT_BYPASS)
122                         sch->qstats.drops++;
123                 kfree_skb(skb);
124                 return ret;
125         }
126 #endif
127
128         if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) {
129                 sch->q.qlen++;
130                 sch->qstats.requeues++;
131                 return 0;
132         }
133         sch->qstats.drops++;
134         return NET_XMIT_DROP;
135 }
136
137
138 static struct sk_buff *
139 prio_dequeue(struct Qdisc* sch)
140 {
141         struct sk_buff *skb;
142         struct prio_sched_data *q = qdisc_priv(sch);
143         int prio;
144         struct Qdisc *qdisc;
145
146         for (prio = 0; prio < q->bands; prio++) {
147                 qdisc = q->queues[prio];
148                 skb = qdisc->dequeue(qdisc);
149                 if (skb) {
150                         sch->q.qlen--;
151                         return skb;
152                 }
153         }
154         return NULL;
155
156 }
157
158 static unsigned int prio_drop(struct Qdisc* sch)
159 {
160         struct prio_sched_data *q = qdisc_priv(sch);
161         int prio;
162         unsigned int len;
163         struct Qdisc *qdisc;
164
165         for (prio = q->bands-1; prio >= 0; prio--) {
166                 qdisc = q->queues[prio];
167                 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
168                         sch->q.qlen--;
169                         return len;
170                 }
171         }
172         return 0;
173 }
174
175
176 static void
177 prio_reset(struct Qdisc* sch)
178 {
179         int prio;
180         struct prio_sched_data *q = qdisc_priv(sch);
181
182         for (prio=0; prio<q->bands; prio++)
183                 qdisc_reset(q->queues[prio]);
184         sch->q.qlen = 0;
185 }
186
187 static void
188 prio_destroy(struct Qdisc* sch)
189 {
190         int prio;
191         struct prio_sched_data *q = qdisc_priv(sch);
192         struct tcf_proto *tp;
193
194         while ((tp = q->filter_list) != NULL) {
195                 q->filter_list = tp->next;
196                 tcf_destroy(tp);
197         }
198
199         for (prio=0; prio<q->bands; prio++)
200                 qdisc_destroy(q->queues[prio]);
201 }
202
203 static int prio_tune(struct Qdisc *sch, struct rtattr *opt)
204 {
205         struct prio_sched_data *q = qdisc_priv(sch);
206         struct tc_prio_qopt *qopt = RTA_DATA(opt);
207         int i;
208
209         if (opt->rta_len < RTA_LENGTH(sizeof(*qopt)))
210                 return -EINVAL;
211         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
212                 return -EINVAL;
213
214         for (i=0; i<=TC_PRIO_MAX; i++) {
215                 if (qopt->priomap[i] >= qopt->bands)
216                         return -EINVAL;
217         }
218
219         sch_tree_lock(sch);
220         q->bands = qopt->bands;
221         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
222
223         for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
224                 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
225                 if (child != &noop_qdisc) {
226                         qdisc_tree_decrease_qlen(child, child->q.qlen);
227                         qdisc_destroy(child);
228                 }
229         }
230         sch_tree_unlock(sch);
231
232         for (i=0; i<q->bands; i++) {
233                 if (q->queues[i] == &noop_qdisc) {
234                         struct Qdisc *child;
235                         child = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
236                                                   TC_H_MAKE(sch->handle, i + 1));
237                         if (child) {
238                                 sch_tree_lock(sch);
239                                 child = xchg(&q->queues[i], child);
240
241                                 if (child != &noop_qdisc) {
242                                         qdisc_tree_decrease_qlen(child,
243                                                                  child->q.qlen);
244                                         qdisc_destroy(child);
245                                 }
246                                 sch_tree_unlock(sch);
247                         }
248                 }
249         }
250         return 0;
251 }
252
253 static int prio_init(struct Qdisc *sch, struct rtattr *opt)
254 {
255         struct prio_sched_data *q = qdisc_priv(sch);
256         int i;
257
258         for (i=0; i<TCQ_PRIO_BANDS; i++)
259                 q->queues[i] = &noop_qdisc;
260
261         if (opt == NULL) {
262                 return -EINVAL;
263         } else {
264                 int err;
265
266                 if ((err= prio_tune(sch, opt)) != 0)
267                         return err;
268         }
269         return 0;
270 }
271
272 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
273 {
274         struct prio_sched_data *q = qdisc_priv(sch);
275         unsigned char    *b = skb->tail;
276         struct tc_prio_qopt opt;
277
278         opt.bands = q->bands;
279         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
280         RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
281         return skb->len;
282
283 rtattr_failure:
284         skb_trim(skb, b - skb->data);
285         return -1;
286 }
287
288 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
289                       struct Qdisc **old)
290 {
291         struct prio_sched_data *q = qdisc_priv(sch);
292         unsigned long band = arg - 1;
293
294         if (band >= q->bands)
295                 return -EINVAL;
296
297         if (new == NULL)
298                 new = &noop_qdisc;
299
300         sch_tree_lock(sch);
301         *old = q->queues[band];
302         q->queues[band] = new;
303         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
304         qdisc_reset(*old);
305         sch_tree_unlock(sch);
306
307         return 0;
308 }
309
310 static struct Qdisc *
311 prio_leaf(struct Qdisc *sch, unsigned long arg)
312 {
313         struct prio_sched_data *q = qdisc_priv(sch);
314         unsigned long band = arg - 1;
315
316         if (band >= q->bands)
317                 return NULL;
318
319         return q->queues[band];
320 }
321
322 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
323 {
324         struct prio_sched_data *q = qdisc_priv(sch);
325         unsigned long band = TC_H_MIN(classid);
326
327         if (band - 1 >= q->bands)
328                 return 0;
329         return band;
330 }
331
332 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
333 {
334         return prio_get(sch, classid);
335 }
336
337
338 static void prio_put(struct Qdisc *q, unsigned long cl)
339 {
340         return;
341 }
342
343 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct rtattr **tca, unsigned long *arg)
344 {
345         unsigned long cl = *arg;
346         struct prio_sched_data *q = qdisc_priv(sch);
347
348         if (cl - 1 > q->bands)
349                 return -ENOENT;
350         return 0;
351 }
352
353 static int prio_delete(struct Qdisc *sch, unsigned long cl)
354 {
355         struct prio_sched_data *q = qdisc_priv(sch);
356         if (cl - 1 > q->bands)
357                 return -ENOENT;
358         return 0;
359 }
360
361
362 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
363                            struct tcmsg *tcm)
364 {
365         struct prio_sched_data *q = qdisc_priv(sch);
366
367         if (cl - 1 > q->bands)
368                 return -ENOENT;
369         tcm->tcm_handle |= TC_H_MIN(cl);
370         if (q->queues[cl-1])
371                 tcm->tcm_info = q->queues[cl-1]->handle;
372         return 0;
373 }
374
375 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
376                                  struct gnet_dump *d)
377 {
378         struct prio_sched_data *q = qdisc_priv(sch);
379         struct Qdisc *cl_q;
380
381         cl_q = q->queues[cl - 1];
382         if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
383             gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
384                 return -1;
385
386         return 0;
387 }
388
389 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
390 {
391         struct prio_sched_data *q = qdisc_priv(sch);
392         int prio;
393
394         if (arg->stop)
395                 return;
396
397         for (prio = 0; prio < q->bands; prio++) {
398                 if (arg->count < arg->skip) {
399                         arg->count++;
400                         continue;
401                 }
402                 if (arg->fn(sch, prio+1, arg) < 0) {
403                         arg->stop = 1;
404                         break;
405                 }
406                 arg->count++;
407         }
408 }
409
410 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
411 {
412         struct prio_sched_data *q = qdisc_priv(sch);
413
414         if (cl)
415                 return NULL;
416         return &q->filter_list;
417 }
418
419 static struct Qdisc_class_ops prio_class_ops = {
420         .graft          =       prio_graft,
421         .leaf           =       prio_leaf,
422         .get            =       prio_get,
423         .put            =       prio_put,
424         .change         =       prio_change,
425         .delete         =       prio_delete,
426         .walk           =       prio_walk,
427         .tcf_chain      =       prio_find_tcf,
428         .bind_tcf       =       prio_bind,
429         .unbind_tcf     =       prio_put,
430         .dump           =       prio_dump_class,
431         .dump_stats     =       prio_dump_class_stats,
432 };
433
434 static struct Qdisc_ops prio_qdisc_ops = {
435         .next           =       NULL,
436         .cl_ops         =       &prio_class_ops,
437         .id             =       "prio",
438         .priv_size      =       sizeof(struct prio_sched_data),
439         .enqueue        =       prio_enqueue,
440         .dequeue        =       prio_dequeue,
441         .requeue        =       prio_requeue,
442         .drop           =       prio_drop,
443         .init           =       prio_init,
444         .reset          =       prio_reset,
445         .destroy        =       prio_destroy,
446         .change         =       prio_tune,
447         .dump           =       prio_dump,
448         .owner          =       THIS_MODULE,
449 };
450
451 static int __init prio_module_init(void)
452 {
453         return register_qdisc(&prio_qdisc_ops);
454 }
455
456 static void __exit prio_module_exit(void)
457 {
458         unregister_qdisc(&prio_qdisc_ops);
459 }
460
461 module_init(prio_module_init)
462 module_exit(prio_module_exit)
463
464 MODULE_LICENSE("GPL");