Merge git://oss.sgi.com:8090/oss/git/xfs-2.6
[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/config.h>
15 #include <linux/module.h>
16 #include <asm/uaccess.h>
17 #include <asm/system.h>
18 #include <linux/bitops.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/sched.h>
22 #include <linux/string.h>
23 #include <linux/mm.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/in.h>
27 #include <linux/errno.h>
28 #include <linux/interrupt.h>
29 #include <linux/if_ether.h>
30 #include <linux/inet.h>
31 #include <linux/netdevice.h>
32 #include <linux/etherdevice.h>
33 #include <linux/notifier.h>
34 #include <net/ip.h>
35 #include <net/route.h>
36 #include <linux/skbuff.h>
37 #include <net/sock.h>
38 #include <net/pkt_sched.h>
39
40
41 struct prio_sched_data
42 {
43         int bands;
44         struct tcf_proto *filter_list;
45         u8  prio2band[TC_PRIO_MAX+1];
46         struct Qdisc *queues[TCQ_PRIO_BANDS];
47 };
48
49
50 static struct Qdisc *
51 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
52 {
53         struct prio_sched_data *q = qdisc_priv(sch);
54         u32 band = skb->priority;
55         struct tcf_result res;
56
57         *qerr = NET_XMIT_BYPASS;
58         if (TC_H_MAJ(skb->priority) != sch->handle) {
59 #ifdef CONFIG_NET_CLS_ACT
60                 switch (tc_classify(skb, q->filter_list, &res)) {
61                 case TC_ACT_STOLEN:
62                 case TC_ACT_QUEUED:
63                         *qerr = NET_XMIT_SUCCESS;
64                 case TC_ACT_SHOT:
65                         return NULL;
66                 };
67
68                 if (!q->filter_list ) {
69 #else
70                 if (!q->filter_list || tc_classify(skb, q->filter_list, &res)) {
71 #endif
72                         if (TC_H_MAJ(band))
73                                 band = 0;
74                         return q->queues[q->prio2band[band&TC_PRIO_MAX]];
75                 }
76                 band = res.classid;
77         }
78         band = TC_H_MIN(band) - 1;
79         if (band > q->bands)
80                 return q->queues[q->prio2band[0]];
81
82         return q->queues[band];
83 }
84
85 static int
86 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
87 {
88         struct Qdisc *qdisc;
89         int ret;
90
91         qdisc = prio_classify(skb, sch, &ret);
92 #ifdef CONFIG_NET_CLS_ACT
93         if (qdisc == NULL) {
94
95                 if (ret == NET_XMIT_BYPASS)
96                         sch->qstats.drops++;
97                 kfree_skb(skb);
98                 return ret;
99         }
100 #endif
101
102         if ((ret = qdisc->enqueue(skb, qdisc)) == NET_XMIT_SUCCESS) {
103                 sch->bstats.bytes += skb->len;
104                 sch->bstats.packets++;
105                 sch->q.qlen++;
106                 return NET_XMIT_SUCCESS;
107         }
108         sch->qstats.drops++;
109         return ret; 
110 }
111
112
113 static int
114 prio_requeue(struct sk_buff *skb, struct Qdisc* sch)
115 {
116         struct Qdisc *qdisc;
117         int ret;
118
119         qdisc = prio_classify(skb, sch, &ret);
120 #ifdef CONFIG_NET_CLS_ACT
121         if (qdisc == NULL) {
122                 if (ret == NET_XMIT_BYPASS)
123                         sch->qstats.drops++;
124                 kfree_skb(skb);
125                 return ret;
126         }
127 #endif
128
129         if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) {
130                 sch->q.qlen++;
131                 sch->qstats.requeues++;
132                 return 0;
133         }
134         sch->qstats.drops++;
135         return NET_XMIT_DROP;
136 }
137
138
139 static struct sk_buff *
140 prio_dequeue(struct Qdisc* sch)
141 {
142         struct sk_buff *skb;
143         struct prio_sched_data *q = qdisc_priv(sch);
144         int prio;
145         struct Qdisc *qdisc;
146
147         for (prio = 0; prio < q->bands; prio++) {
148                 qdisc = q->queues[prio];
149                 skb = qdisc->dequeue(qdisc);
150                 if (skb) {
151                         sch->q.qlen--;
152                         return skb;
153                 }
154         }
155         return NULL;
156
157 }
158
159 static unsigned int prio_drop(struct Qdisc* sch)
160 {
161         struct prio_sched_data *q = qdisc_priv(sch);
162         int prio;
163         unsigned int len;
164         struct Qdisc *qdisc;
165
166         for (prio = q->bands-1; prio >= 0; prio--) {
167                 qdisc = q->queues[prio];
168                 if ((len = qdisc->ops->drop(qdisc)) != 0) {
169                         sch->q.qlen--;
170                         return len;
171                 }
172         }
173         return 0;
174 }
175
176
177 static void
178 prio_reset(struct Qdisc* sch)
179 {
180         int prio;
181         struct prio_sched_data *q = qdisc_priv(sch);
182
183         for (prio=0; prio<q->bands; prio++)
184                 qdisc_reset(q->queues[prio]);
185         sch->q.qlen = 0;
186 }
187
188 static void
189 prio_destroy(struct Qdisc* sch)
190 {
191         int prio;
192         struct prio_sched_data *q = qdisc_priv(sch);
193         struct tcf_proto *tp;
194
195         while ((tp = q->filter_list) != NULL) {
196                 q->filter_list = tp->next;
197                 tcf_destroy(tp);
198         }
199
200         for (prio=0; prio<q->bands; prio++)
201                 qdisc_destroy(q->queues[prio]);
202 }
203
204 static int prio_tune(struct Qdisc *sch, struct rtattr *opt)
205 {
206         struct prio_sched_data *q = qdisc_priv(sch);
207         struct tc_prio_qopt *qopt = RTA_DATA(opt);
208         int i;
209
210         if (opt->rta_len < RTA_LENGTH(sizeof(*qopt)))
211                 return -EINVAL;
212         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
213                 return -EINVAL;
214
215         for (i=0; i<=TC_PRIO_MAX; i++) {
216                 if (qopt->priomap[i] >= qopt->bands)
217                         return -EINVAL;
218         }
219
220         sch_tree_lock(sch);
221         q->bands = qopt->bands;
222         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
223
224         for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
225                 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
226                 if (child != &noop_qdisc)
227                         qdisc_destroy(child);
228         }
229         sch_tree_unlock(sch);
230
231         for (i=0; i<q->bands; i++) {
232                 if (q->queues[i] == &noop_qdisc) {
233                         struct Qdisc *child;
234                         child = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
235                         if (child) {
236                                 sch_tree_lock(sch);
237                                 child = xchg(&q->queues[i], child);
238
239                                 if (child != &noop_qdisc)
240                                         qdisc_destroy(child);
241                                 sch_tree_unlock(sch);
242                         }
243                 }
244         }
245         return 0;
246 }
247
248 static int prio_init(struct Qdisc *sch, struct rtattr *opt)
249 {
250         struct prio_sched_data *q = qdisc_priv(sch);
251         int i;
252
253         for (i=0; i<TCQ_PRIO_BANDS; i++)
254                 q->queues[i] = &noop_qdisc;
255
256         if (opt == NULL) {
257                 return -EINVAL;
258         } else {
259                 int err;
260
261                 if ((err= prio_tune(sch, opt)) != 0)
262                         return err;
263         }
264         return 0;
265 }
266
267 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
268 {
269         struct prio_sched_data *q = qdisc_priv(sch);
270         unsigned char    *b = skb->tail;
271         struct tc_prio_qopt opt;
272
273         opt.bands = q->bands;
274         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
275         RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
276         return skb->len;
277
278 rtattr_failure:
279         skb_trim(skb, b - skb->data);
280         return -1;
281 }
282
283 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
284                       struct Qdisc **old)
285 {
286         struct prio_sched_data *q = qdisc_priv(sch);
287         unsigned long band = arg - 1;
288
289         if (band >= q->bands)
290                 return -EINVAL;
291
292         if (new == NULL)
293                 new = &noop_qdisc;
294
295         sch_tree_lock(sch);
296         *old = q->queues[band];
297         q->queues[band] = new;
298         sch->q.qlen -= (*old)->q.qlen;
299         qdisc_reset(*old);
300         sch_tree_unlock(sch);
301
302         return 0;
303 }
304
305 static struct Qdisc *
306 prio_leaf(struct Qdisc *sch, unsigned long arg)
307 {
308         struct prio_sched_data *q = qdisc_priv(sch);
309         unsigned long band = arg - 1;
310
311         if (band >= q->bands)
312                 return NULL;
313
314         return q->queues[band];
315 }
316
317 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
318 {
319         struct prio_sched_data *q = qdisc_priv(sch);
320         unsigned long band = TC_H_MIN(classid);
321
322         if (band - 1 >= q->bands)
323                 return 0;
324         return band;
325 }
326
327 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
328 {
329         return prio_get(sch, classid);
330 }
331
332
333 static void prio_put(struct Qdisc *q, unsigned long cl)
334 {
335         return;
336 }
337
338 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct rtattr **tca, unsigned long *arg)
339 {
340         unsigned long cl = *arg;
341         struct prio_sched_data *q = qdisc_priv(sch);
342
343         if (cl - 1 > q->bands)
344                 return -ENOENT;
345         return 0;
346 }
347
348 static int prio_delete(struct Qdisc *sch, unsigned long cl)
349 {
350         struct prio_sched_data *q = qdisc_priv(sch);
351         if (cl - 1 > q->bands)
352                 return -ENOENT;
353         return 0;
354 }
355
356
357 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
358                            struct tcmsg *tcm)
359 {
360         struct prio_sched_data *q = qdisc_priv(sch);
361
362         if (cl - 1 > q->bands)
363                 return -ENOENT;
364         tcm->tcm_handle |= TC_H_MIN(cl);
365         if (q->queues[cl-1])
366                 tcm->tcm_info = q->queues[cl-1]->handle;
367         return 0;
368 }
369
370 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
371 {
372         struct prio_sched_data *q = qdisc_priv(sch);
373         int prio;
374
375         if (arg->stop)
376                 return;
377
378         for (prio = 0; prio < q->bands; prio++) {
379                 if (arg->count < arg->skip) {
380                         arg->count++;
381                         continue;
382                 }
383                 if (arg->fn(sch, prio+1, arg) < 0) {
384                         arg->stop = 1;
385                         break;
386                 }
387                 arg->count++;
388         }
389 }
390
391 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
392 {
393         struct prio_sched_data *q = qdisc_priv(sch);
394
395         if (cl)
396                 return NULL;
397         return &q->filter_list;
398 }
399
400 static struct Qdisc_class_ops prio_class_ops = {
401         .graft          =       prio_graft,
402         .leaf           =       prio_leaf,
403         .get            =       prio_get,
404         .put            =       prio_put,
405         .change         =       prio_change,
406         .delete         =       prio_delete,
407         .walk           =       prio_walk,
408         .tcf_chain      =       prio_find_tcf,
409         .bind_tcf       =       prio_bind,
410         .unbind_tcf     =       prio_put,
411         .dump           =       prio_dump_class,
412 };
413
414 static struct Qdisc_ops prio_qdisc_ops = {
415         .next           =       NULL,
416         .cl_ops         =       &prio_class_ops,
417         .id             =       "prio",
418         .priv_size      =       sizeof(struct prio_sched_data),
419         .enqueue        =       prio_enqueue,
420         .dequeue        =       prio_dequeue,
421         .requeue        =       prio_requeue,
422         .drop           =       prio_drop,
423         .init           =       prio_init,
424         .reset          =       prio_reset,
425         .destroy        =       prio_destroy,
426         .change         =       prio_tune,
427         .dump           =       prio_dump,
428         .owner          =       THIS_MODULE,
429 };
430
431 static int __init prio_module_init(void)
432 {
433         return register_qdisc(&prio_qdisc_ops);
434 }
435
436 static void __exit prio_module_exit(void) 
437 {
438         unregister_qdisc(&prio_qdisc_ops);
439 }
440
441 module_init(prio_module_init)
442 module_exit(prio_module_exit)
443
444 MODULE_LICENSE("GPL");