Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6-block.git] / net / sched / sch_prio.c
CommitLineData
1da177e4
LT
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>
10297b99 10 * Fixes: 19990609: J Hadi Salim <hadi@nortelnetworks.com>:
1da177e4
LT
11 * Init -- EINVAL when opt undefined
12 */
13
1da177e4 14#include <linux/module.h>
1da177e4
LT
15#include <linux/types.h>
16#include <linux/kernel.h>
1da177e4 17#include <linux/string.h>
1da177e4 18#include <linux/errno.h>
1da177e4 19#include <linux/skbuff.h>
dc5fc579 20#include <net/netlink.h>
1da177e4
LT
21#include <net/pkt_sched.h>
22
23
24struct 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
33static struct Qdisc *
34prio_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;
bdba91ec 39 int err;
1da177e4 40
c27f339a 41 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
1da177e4 42 if (TC_H_MAJ(skb->priority) != sch->handle) {
bdba91ec 43 err = tc_classify(skb, q->filter_list, &res);
1da177e4 44#ifdef CONFIG_NET_CLS_ACT
dbaaa07a 45 switch (err) {
1da177e4
LT
46 case TC_ACT_STOLEN:
47 case TC_ACT_QUEUED:
378a2f09 48 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
1da177e4
LT
49 case TC_ACT_SHOT:
50 return NULL;
3ff50b79 51 }
1da177e4 52#endif
bdba91ec 53 if (!q->filter_list || err < 0) {
1da177e4
LT
54 if (TC_H_MAJ(band))
55 band = 0;
1d8ae3fd 56 return q->queues[q->prio2band[band&TC_PRIO_MAX]];
1da177e4
LT
57 }
58 band = res.classid;
59 }
60 band = TC_H_MIN(band) - 1;
3e5c2d3b 61 if (band >= q->bands)
1d8ae3fd
DM
62 return q->queues[q->prio2band[0]];
63
1da177e4
LT
64 return q->queues[band];
65}
66
67static int
68prio_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) {
29f1df6c 76
c27f339a 77 if (ret & __NET_XMIT_BYPASS)
1da177e4
LT
78 sch->qstats.drops++;
79 kfree_skb(skb);
80 return ret;
81 }
82#endif
83
5f86173b
JK
84 ret = qdisc_enqueue(skb, qdisc);
85 if (ret == NET_XMIT_SUCCESS) {
0abf77e5 86 sch->bstats.bytes += qdisc_pkt_len(skb);
1da177e4
LT
87 sch->bstats.packets++;
88 sch->q.qlen++;
89 return NET_XMIT_SUCCESS;
90 }
378a2f09
JP
91 if (net_xmit_drop_count(ret))
92 sch->qstats.drops++;
10297b99 93 return ret;
1da177e4
LT
94}
95
96
97static int
98prio_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) {
c27f339a 106 if (ret & __NET_XMIT_BYPASS)
1da177e4
LT
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++;
4cf7cb28 116 return NET_XMIT_SUCCESS;
1da177e4 117 }
378a2f09
JP
118 if (net_xmit_drop_count(ret))
119 sch->qstats.drops++;
0d40b6e5 120 return ret;
1da177e4
LT
121}
122
123
1d8ae3fd 124static struct sk_buff *prio_dequeue(struct Qdisc* sch)
1da177e4 125{
1da177e4
LT
126 struct prio_sched_data *q = qdisc_priv(sch);
127 int prio;
1da177e4
LT
128
129 for (prio = 0; prio < q->bands; prio++) {
1d8ae3fd
DM
130 struct Qdisc *qdisc = q->queues[prio];
131 struct sk_buff *skb = qdisc->dequeue(qdisc);
132 if (skb) {
133 sch->q.qlen--;
134 return skb;
1da177e4
LT
135 }
136 }
137 return NULL;
138
139}
140
141static unsigned int prio_drop(struct Qdisc* sch)
142{
143 struct prio_sched_data *q = qdisc_priv(sch);
144 int prio;
145 unsigned int len;
146 struct Qdisc *qdisc;
147
148 for (prio = q->bands-1; prio >= 0; prio--) {
149 qdisc = q->queues[prio];
6d037a26 150 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
1da177e4
LT
151 sch->q.qlen--;
152 return len;
153 }
154 }
155 return 0;
156}
157
158
159static void
160prio_reset(struct Qdisc* sch)
161{
162 int prio;
163 struct prio_sched_data *q = qdisc_priv(sch);
164
165 for (prio=0; prio<q->bands; prio++)
166 qdisc_reset(q->queues[prio]);
167 sch->q.qlen = 0;
168}
169
170static void
171prio_destroy(struct Qdisc* sch)
172{
173 int prio;
174 struct prio_sched_data *q = qdisc_priv(sch);
1da177e4 175
ff31ab56 176 tcf_destroy_chain(&q->filter_list);
1da177e4
LT
177 for (prio=0; prio<q->bands; prio++)
178 qdisc_destroy(q->queues[prio]);
179}
180
1e90474c 181static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
1da177e4
LT
182{
183 struct prio_sched_data *q = qdisc_priv(sch);
d62733c8 184 struct tc_prio_qopt *qopt;
1da177e4
LT
185 int i;
186
1d8ae3fd
DM
187 if (nla_len(opt) < sizeof(*qopt))
188 return -EINVAL;
189 qopt = nla_data(opt);
d62733c8 190
1d8ae3fd 191 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
1da177e4
LT
192 return -EINVAL;
193
194 for (i=0; i<=TC_PRIO_MAX; i++) {
1d8ae3fd 195 if (qopt->priomap[i] >= qopt->bands)
1da177e4
LT
196 return -EINVAL;
197 }
198
199 sch_tree_lock(sch);
1d8ae3fd 200 q->bands = qopt->bands;
1da177e4
LT
201 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
202
203 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
204 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
5e50da01
PM
205 if (child != &noop_qdisc) {
206 qdisc_tree_decrease_qlen(child, child->q.qlen);
1da177e4 207 qdisc_destroy(child);
5e50da01 208 }
1da177e4
LT
209 }
210 sch_tree_unlock(sch);
211
dd914b40
AA
212 for (i=0; i<q->bands; i++) {
213 if (q->queues[i] == &noop_qdisc) {
1da177e4 214 struct Qdisc *child;
5ce2d488 215 child = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
bb949fbd 216 &pfifo_qdisc_ops,
9f9afec4 217 TC_H_MAKE(sch->handle, i + 1));
1da177e4
LT
218 if (child) {
219 sch_tree_lock(sch);
dd914b40 220 child = xchg(&q->queues[i], child);
1da177e4 221
5e50da01
PM
222 if (child != &noop_qdisc) {
223 qdisc_tree_decrease_qlen(child,
224 child->q.qlen);
1da177e4 225 qdisc_destroy(child);
5e50da01 226 }
1da177e4
LT
227 sch_tree_unlock(sch);
228 }
229 }
230 }
231 return 0;
232}
233
1e90474c 234static int prio_init(struct Qdisc *sch, struct nlattr *opt)
1da177e4
LT
235{
236 struct prio_sched_data *q = qdisc_priv(sch);
237 int i;
238
239 for (i=0; i<TCQ_PRIO_BANDS; i++)
240 q->queues[i] = &noop_qdisc;
241
242 if (opt == NULL) {
243 return -EINVAL;
244 } else {
245 int err;
246
247 if ((err= prio_tune(sch, opt)) != 0)
248 return err;
249 }
250 return 0;
251}
252
253static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
254{
255 struct prio_sched_data *q = qdisc_priv(sch);
27a884dc 256 unsigned char *b = skb_tail_pointer(skb);
1e90474c 257 struct nlattr *nest;
1da177e4
LT
258 struct tc_prio_qopt opt;
259
260 opt.bands = q->bands;
261 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
d62733c8 262
1e90474c
PM
263 nest = nla_nest_compat_start(skb, TCA_OPTIONS, sizeof(opt), &opt);
264 if (nest == NULL)
265 goto nla_put_failure;
1e90474c 266 nla_nest_compat_end(skb, nest);
d62733c8 267
1da177e4
LT
268 return skb->len;
269
1e90474c 270nla_put_failure:
dc5fc579 271 nlmsg_trim(skb, b);
1da177e4
LT
272 return -1;
273}
274
275static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
276 struct Qdisc **old)
277{
278 struct prio_sched_data *q = qdisc_priv(sch);
279 unsigned long band = arg - 1;
280
281 if (band >= q->bands)
282 return -EINVAL;
283
284 if (new == NULL)
285 new = &noop_qdisc;
286
287 sch_tree_lock(sch);
288 *old = q->queues[band];
289 q->queues[band] = new;
5e50da01 290 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1da177e4
LT
291 qdisc_reset(*old);
292 sch_tree_unlock(sch);
293
294 return 0;
295}
296
297static struct Qdisc *
298prio_leaf(struct Qdisc *sch, unsigned long arg)
299{
300 struct prio_sched_data *q = qdisc_priv(sch);
301 unsigned long band = arg - 1;
302
303 if (band >= q->bands)
304 return NULL;
305
306 return q->queues[band];
307}
308
309static unsigned long prio_get(struct Qdisc *sch, u32 classid)
310{
311 struct prio_sched_data *q = qdisc_priv(sch);
312 unsigned long band = TC_H_MIN(classid);
313
314 if (band - 1 >= q->bands)
315 return 0;
316 return band;
317}
318
319static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
320{
321 return prio_get(sch, classid);
322}
323
324
325static void prio_put(struct Qdisc *q, unsigned long cl)
326{
327 return;
328}
329
1e90474c 330static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct nlattr **tca, unsigned long *arg)
1da177e4
LT
331{
332 unsigned long cl = *arg;
333 struct prio_sched_data *q = qdisc_priv(sch);
334
335 if (cl - 1 > q->bands)
336 return -ENOENT;
337 return 0;
338}
339
340static int prio_delete(struct Qdisc *sch, unsigned long cl)
341{
342 struct prio_sched_data *q = qdisc_priv(sch);
343 if (cl - 1 > q->bands)
344 return -ENOENT;
345 return 0;
346}
347
348
349static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
350 struct tcmsg *tcm)
351{
352 struct prio_sched_data *q = qdisc_priv(sch);
353
354 if (cl - 1 > q->bands)
355 return -ENOENT;
356 tcm->tcm_handle |= TC_H_MIN(cl);
357 if (q->queues[cl-1])
358 tcm->tcm_info = q->queues[cl-1]->handle;
359 return 0;
360}
361
2cf6c36c
JP
362static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
363 struct gnet_dump *d)
364{
365 struct prio_sched_data *q = qdisc_priv(sch);
366 struct Qdisc *cl_q;
367
368 cl_q = q->queues[cl - 1];
369 if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
370 gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
371 return -1;
372
373 return 0;
374}
375
1da177e4
LT
376static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
377{
378 struct prio_sched_data *q = qdisc_priv(sch);
379 int prio;
380
381 if (arg->stop)
382 return;
383
384 for (prio = 0; prio < q->bands; prio++) {
385 if (arg->count < arg->skip) {
386 arg->count++;
387 continue;
388 }
389 if (arg->fn(sch, prio+1, arg) < 0) {
390 arg->stop = 1;
391 break;
392 }
393 arg->count++;
394 }
395}
396
397static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
398{
399 struct prio_sched_data *q = qdisc_priv(sch);
400
401 if (cl)
402 return NULL;
403 return &q->filter_list;
404}
405
20fea08b 406static const struct Qdisc_class_ops prio_class_ops = {
1da177e4
LT
407 .graft = prio_graft,
408 .leaf = prio_leaf,
409 .get = prio_get,
410 .put = prio_put,
411 .change = prio_change,
412 .delete = prio_delete,
413 .walk = prio_walk,
414 .tcf_chain = prio_find_tcf,
415 .bind_tcf = prio_bind,
416 .unbind_tcf = prio_put,
417 .dump = prio_dump_class,
2cf6c36c 418 .dump_stats = prio_dump_class_stats,
1da177e4
LT
419};
420
20fea08b 421static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
1da177e4
LT
422 .next = NULL,
423 .cl_ops = &prio_class_ops,
424 .id = "prio",
425 .priv_size = sizeof(struct prio_sched_data),
426 .enqueue = prio_enqueue,
427 .dequeue = prio_dequeue,
428 .requeue = prio_requeue,
429 .drop = prio_drop,
430 .init = prio_init,
431 .reset = prio_reset,
432 .destroy = prio_destroy,
433 .change = prio_tune,
434 .dump = prio_dump,
435 .owner = THIS_MODULE,
436};
437
438static int __init prio_module_init(void)
439{
1d8ae3fd 440 return register_qdisc(&prio_qdisc_ops);
1da177e4
LT
441}
442
10297b99 443static void __exit prio_module_exit(void)
1da177e4
LT
444{
445 unregister_qdisc(&prio_qdisc_ops);
446}
447
448module_init(prio_module_init)
449module_exit(prio_module_exit)
450
451MODULE_LICENSE("GPL");