Pull percpu-dtc into release branch
[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/string.h>
21 #include <linux/mm.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
24 #include <linux/in.h>
25 #include <linux/errno.h>
26 #include <linux/interrupt.h>
27 #include <linux/if_ether.h>
28 #include <linux/inet.h>
29 #include <linux/netdevice.h>
30 #include <linux/etherdevice.h>
31 #include <linux/notifier.h>
32 #include <net/ip.h>
33 #include <net/route.h>
34 #include <linux/skbuff.h>
35 #include <net/netlink.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
193         tcf_destroy_chain(q->filter_list);
194         for (prio=0; prio<q->bands; prio++)
195                 qdisc_destroy(q->queues[prio]);
196 }
197
198 static int prio_tune(struct Qdisc *sch, struct rtattr *opt)
199 {
200         struct prio_sched_data *q = qdisc_priv(sch);
201         struct tc_prio_qopt *qopt = RTA_DATA(opt);
202         int i;
203
204         if (opt->rta_len < RTA_LENGTH(sizeof(*qopt)))
205                 return -EINVAL;
206         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
207                 return -EINVAL;
208
209         for (i=0; i<=TC_PRIO_MAX; i++) {
210                 if (qopt->priomap[i] >= qopt->bands)
211                         return -EINVAL;
212         }
213
214         sch_tree_lock(sch);
215         q->bands = qopt->bands;
216         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
217
218         for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
219                 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
220                 if (child != &noop_qdisc) {
221                         qdisc_tree_decrease_qlen(child, child->q.qlen);
222                         qdisc_destroy(child);
223                 }
224         }
225         sch_tree_unlock(sch);
226
227         for (i=0; i<q->bands; i++) {
228                 if (q->queues[i] == &noop_qdisc) {
229                         struct Qdisc *child;
230                         child = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
231                                                   TC_H_MAKE(sch->handle, i + 1));
232                         if (child) {
233                                 sch_tree_lock(sch);
234                                 child = xchg(&q->queues[i], child);
235
236                                 if (child != &noop_qdisc) {
237                                         qdisc_tree_decrease_qlen(child,
238                                                                  child->q.qlen);
239                                         qdisc_destroy(child);
240                                 }
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_pointer(skb);
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         nlmsg_trim(skb, b);
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         qdisc_tree_decrease_qlen(*old, (*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 int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
371                                  struct gnet_dump *d)
372 {
373         struct prio_sched_data *q = qdisc_priv(sch);
374         struct Qdisc *cl_q;
375
376         cl_q = q->queues[cl - 1];
377         if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
378             gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
379                 return -1;
380
381         return 0;
382 }
383
384 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
385 {
386         struct prio_sched_data *q = qdisc_priv(sch);
387         int prio;
388
389         if (arg->stop)
390                 return;
391
392         for (prio = 0; prio < q->bands; prio++) {
393                 if (arg->count < arg->skip) {
394                         arg->count++;
395                         continue;
396                 }
397                 if (arg->fn(sch, prio+1, arg) < 0) {
398                         arg->stop = 1;
399                         break;
400                 }
401                 arg->count++;
402         }
403 }
404
405 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
406 {
407         struct prio_sched_data *q = qdisc_priv(sch);
408
409         if (cl)
410                 return NULL;
411         return &q->filter_list;
412 }
413
414 static struct Qdisc_class_ops prio_class_ops = {
415         .graft          =       prio_graft,
416         .leaf           =       prio_leaf,
417         .get            =       prio_get,
418         .put            =       prio_put,
419         .change         =       prio_change,
420         .delete         =       prio_delete,
421         .walk           =       prio_walk,
422         .tcf_chain      =       prio_find_tcf,
423         .bind_tcf       =       prio_bind,
424         .unbind_tcf     =       prio_put,
425         .dump           =       prio_dump_class,
426         .dump_stats     =       prio_dump_class_stats,
427 };
428
429 static struct Qdisc_ops prio_qdisc_ops = {
430         .next           =       NULL,
431         .cl_ops         =       &prio_class_ops,
432         .id             =       "prio",
433         .priv_size      =       sizeof(struct prio_sched_data),
434         .enqueue        =       prio_enqueue,
435         .dequeue        =       prio_dequeue,
436         .requeue        =       prio_requeue,
437         .drop           =       prio_drop,
438         .init           =       prio_init,
439         .reset          =       prio_reset,
440         .destroy        =       prio_destroy,
441         .change         =       prio_tune,
442         .dump           =       prio_dump,
443         .owner          =       THIS_MODULE,
444 };
445
446 static int __init prio_module_init(void)
447 {
448         return register_qdisc(&prio_qdisc_ops);
449 }
450
451 static void __exit prio_module_exit(void)
452 {
453         unregister_qdisc(&prio_qdisc_ops);
454 }
455
456 module_init(prio_module_init)
457 module_exit(prio_module_exit)
458
459 MODULE_LICENSE("GPL");