b1c95bce33ce6f2f8a360e93ab234f8db1ab42c9
[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/slab.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/string.h>
19 #include <linux/errno.h>
20 #include <linux/skbuff.h>
21 #include <net/netlink.h>
22 #include <net/pkt_sched.h>
23
24
25 struct prio_sched_data
26 {
27         int bands;
28         struct tcf_proto *filter_list;
29         u8  prio2band[TC_PRIO_MAX+1];
30         struct Qdisc *queues[TCQ_PRIO_BANDS];
31 };
32
33
34 static struct Qdisc *
35 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
36 {
37         struct prio_sched_data *q = qdisc_priv(sch);
38         u32 band = skb->priority;
39         struct tcf_result res;
40         int err;
41
42         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
43         if (TC_H_MAJ(skb->priority) != sch->handle) {
44                 err = tc_classify(skb, q->filter_list, &res);
45 #ifdef CONFIG_NET_CLS_ACT
46                 switch (err) {
47                 case TC_ACT_STOLEN:
48                 case TC_ACT_QUEUED:
49                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
50                 case TC_ACT_SHOT:
51                         return NULL;
52                 }
53 #endif
54                 if (!q->filter_list || err < 0) {
55                         if (TC_H_MAJ(band))
56                                 band = 0;
57                         return q->queues[q->prio2band[band&TC_PRIO_MAX]];
58                 }
59                 band = res.classid;
60         }
61         band = TC_H_MIN(band) - 1;
62         if (band >= q->bands)
63                 return q->queues[q->prio2band[0]];
64
65         return q->queues[band];
66 }
67
68 static int
69 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
70 {
71         struct Qdisc *qdisc;
72         int ret;
73
74         qdisc = prio_classify(skb, sch, &ret);
75 #ifdef CONFIG_NET_CLS_ACT
76         if (qdisc == NULL) {
77
78                 if (ret & __NET_XMIT_BYPASS)
79                         sch->qstats.drops++;
80                 kfree_skb(skb);
81                 return ret;
82         }
83 #endif
84
85         ret = qdisc_enqueue(skb, qdisc);
86         if (ret == NET_XMIT_SUCCESS) {
87                 sch->bstats.bytes += qdisc_pkt_len(skb);
88                 sch->bstats.packets++;
89                 sch->q.qlen++;
90                 return NET_XMIT_SUCCESS;
91         }
92         if (net_xmit_drop_count(ret))
93                 sch->qstats.drops++;
94         return ret;
95 }
96
97 static struct sk_buff *prio_peek(struct Qdisc *sch)
98 {
99         struct prio_sched_data *q = qdisc_priv(sch);
100         int prio;
101
102         for (prio = 0; prio < q->bands; prio++) {
103                 struct Qdisc *qdisc = q->queues[prio];
104                 struct sk_buff *skb = qdisc->ops->peek(qdisc);
105                 if (skb)
106                         return skb;
107         }
108         return NULL;
109 }
110
111 static struct sk_buff *prio_dequeue(struct Qdisc* sch)
112 {
113         struct prio_sched_data *q = qdisc_priv(sch);
114         int prio;
115
116         for (prio = 0; prio < q->bands; prio++) {
117                 struct Qdisc *qdisc = q->queues[prio];
118                 struct sk_buff *skb = qdisc->dequeue(qdisc);
119                 if (skb) {
120                         sch->q.qlen--;
121                         return skb;
122                 }
123         }
124         return NULL;
125
126 }
127
128 static unsigned int prio_drop(struct Qdisc* sch)
129 {
130         struct prio_sched_data *q = qdisc_priv(sch);
131         int prio;
132         unsigned int len;
133         struct Qdisc *qdisc;
134
135         for (prio = q->bands-1; prio >= 0; prio--) {
136                 qdisc = q->queues[prio];
137                 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
138                         sch->q.qlen--;
139                         return len;
140                 }
141         }
142         return 0;
143 }
144
145
146 static void
147 prio_reset(struct Qdisc* sch)
148 {
149         int prio;
150         struct prio_sched_data *q = qdisc_priv(sch);
151
152         for (prio=0; prio<q->bands; prio++)
153                 qdisc_reset(q->queues[prio]);
154         sch->q.qlen = 0;
155 }
156
157 static void
158 prio_destroy(struct Qdisc* sch)
159 {
160         int prio;
161         struct prio_sched_data *q = qdisc_priv(sch);
162
163         tcf_destroy_chain(&q->filter_list);
164         for (prio=0; prio<q->bands; prio++)
165                 qdisc_destroy(q->queues[prio]);
166 }
167
168 static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
169 {
170         struct prio_sched_data *q = qdisc_priv(sch);
171         struct tc_prio_qopt *qopt;
172         int i;
173
174         if (nla_len(opt) < sizeof(*qopt))
175                 return -EINVAL;
176         qopt = nla_data(opt);
177
178         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
179                 return -EINVAL;
180
181         for (i=0; i<=TC_PRIO_MAX; i++) {
182                 if (qopt->priomap[i] >= qopt->bands)
183                         return -EINVAL;
184         }
185
186         sch_tree_lock(sch);
187         q->bands = qopt->bands;
188         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
189
190         for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
191                 struct Qdisc *child = q->queues[i];
192                 q->queues[i] = &noop_qdisc;
193                 if (child != &noop_qdisc) {
194                         qdisc_tree_decrease_qlen(child, child->q.qlen);
195                         qdisc_destroy(child);
196                 }
197         }
198         sch_tree_unlock(sch);
199
200         for (i=0; i<q->bands; i++) {
201                 if (q->queues[i] == &noop_qdisc) {
202                         struct Qdisc *child, *old;
203                         child = qdisc_create_dflt(sch->dev_queue,
204                                                   &pfifo_qdisc_ops,
205                                                   TC_H_MAKE(sch->handle, i + 1));
206                         if (child) {
207                                 sch_tree_lock(sch);
208                                 old = q->queues[i];
209                                 q->queues[i] = child;
210
211                                 if (old != &noop_qdisc) {
212                                         qdisc_tree_decrease_qlen(old,
213                                                                  old->q.qlen);
214                                         qdisc_destroy(old);
215                                 }
216                                 sch_tree_unlock(sch);
217                         }
218                 }
219         }
220         return 0;
221 }
222
223 static int prio_init(struct Qdisc *sch, struct nlattr *opt)
224 {
225         struct prio_sched_data *q = qdisc_priv(sch);
226         int i;
227
228         for (i=0; i<TCQ_PRIO_BANDS; i++)
229                 q->queues[i] = &noop_qdisc;
230
231         if (opt == NULL) {
232                 return -EINVAL;
233         } else {
234                 int err;
235
236                 if ((err= prio_tune(sch, opt)) != 0)
237                         return err;
238         }
239         return 0;
240 }
241
242 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
243 {
244         struct prio_sched_data *q = qdisc_priv(sch);
245         unsigned char *b = skb_tail_pointer(skb);
246         struct tc_prio_qopt opt;
247
248         opt.bands = q->bands;
249         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
250
251         NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
252
253         return skb->len;
254
255 nla_put_failure:
256         nlmsg_trim(skb, b);
257         return -1;
258 }
259
260 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
261                       struct Qdisc **old)
262 {
263         struct prio_sched_data *q = qdisc_priv(sch);
264         unsigned long band = arg - 1;
265
266         if (new == NULL)
267                 new = &noop_qdisc;
268
269         sch_tree_lock(sch);
270         *old = q->queues[band];
271         q->queues[band] = new;
272         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
273         qdisc_reset(*old);
274         sch_tree_unlock(sch);
275
276         return 0;
277 }
278
279 static struct Qdisc *
280 prio_leaf(struct Qdisc *sch, unsigned long arg)
281 {
282         struct prio_sched_data *q = qdisc_priv(sch);
283         unsigned long band = arg - 1;
284
285         return q->queues[band];
286 }
287
288 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
289 {
290         struct prio_sched_data *q = qdisc_priv(sch);
291         unsigned long band = TC_H_MIN(classid);
292
293         if (band - 1 >= q->bands)
294                 return 0;
295         return band;
296 }
297
298 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
299 {
300         return prio_get(sch, classid);
301 }
302
303
304 static void prio_put(struct Qdisc *q, unsigned long cl)
305 {
306 }
307
308 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
309                            struct tcmsg *tcm)
310 {
311         struct prio_sched_data *q = qdisc_priv(sch);
312
313         tcm->tcm_handle |= TC_H_MIN(cl);
314         tcm->tcm_info = q->queues[cl-1]->handle;
315         return 0;
316 }
317
318 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
319                                  struct gnet_dump *d)
320 {
321         struct prio_sched_data *q = qdisc_priv(sch);
322         struct Qdisc *cl_q;
323
324         cl_q = q->queues[cl - 1];
325         cl_q->qstats.qlen = cl_q->q.qlen;
326         if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
327             gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
328                 return -1;
329
330         return 0;
331 }
332
333 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
334 {
335         struct prio_sched_data *q = qdisc_priv(sch);
336         int prio;
337
338         if (arg->stop)
339                 return;
340
341         for (prio = 0; prio < q->bands; prio++) {
342                 if (arg->count < arg->skip) {
343                         arg->count++;
344                         continue;
345                 }
346                 if (arg->fn(sch, prio+1, arg) < 0) {
347                         arg->stop = 1;
348                         break;
349                 }
350                 arg->count++;
351         }
352 }
353
354 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
355 {
356         struct prio_sched_data *q = qdisc_priv(sch);
357
358         if (cl)
359                 return NULL;
360         return &q->filter_list;
361 }
362
363 static const struct Qdisc_class_ops prio_class_ops = {
364         .graft          =       prio_graft,
365         .leaf           =       prio_leaf,
366         .get            =       prio_get,
367         .put            =       prio_put,
368         .walk           =       prio_walk,
369         .tcf_chain      =       prio_find_tcf,
370         .bind_tcf       =       prio_bind,
371         .unbind_tcf     =       prio_put,
372         .dump           =       prio_dump_class,
373         .dump_stats     =       prio_dump_class_stats,
374 };
375
376 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
377         .next           =       NULL,
378         .cl_ops         =       &prio_class_ops,
379         .id             =       "prio",
380         .priv_size      =       sizeof(struct prio_sched_data),
381         .enqueue        =       prio_enqueue,
382         .dequeue        =       prio_dequeue,
383         .peek           =       prio_peek,
384         .drop           =       prio_drop,
385         .init           =       prio_init,
386         .reset          =       prio_reset,
387         .destroy        =       prio_destroy,
388         .change         =       prio_tune,
389         .dump           =       prio_dump,
390         .owner          =       THIS_MODULE,
391 };
392
393 static int __init prio_module_init(void)
394 {
395         return register_qdisc(&prio_qdisc_ops);
396 }
397
398 static void __exit prio_module_exit(void)
399 {
400         unregister_qdisc(&prio_qdisc_ops);
401 }
402
403 module_init(prio_module_init)
404 module_exit(prio_module_exit)
405
406 MODULE_LICENSE("GPL");