Merge branch 'for-linus-4.1' of git://git.kernel.org/pub/scm/linux/kernel/git/mason...
[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>
5a0e3ad6 15#include <linux/slab.h>
1da177e4
LT
16#include <linux/types.h>
17#include <linux/kernel.h>
1da177e4 18#include <linux/string.h>
1da177e4 19#include <linux/errno.h>
1da177e4 20#include <linux/skbuff.h>
dc5fc579 21#include <net/netlink.h>
1da177e4
LT
22#include <net/pkt_sched.h>
23
24
cc7ec456 25struct prio_sched_data {
1da177e4 26 int bands;
25d8c0d5 27 struct tcf_proto __rcu *filter_list;
1da177e4
LT
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;
25d8c0d5 39 struct tcf_proto *fl;
bdba91ec 40 int err;
1da177e4 41
c27f339a 42 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
1da177e4 43 if (TC_H_MAJ(skb->priority) != sch->handle) {
25d8c0d5
JF
44 fl = rcu_dereference_bh(q->filter_list);
45 err = tc_classify(skb, fl, &res);
1da177e4 46#ifdef CONFIG_NET_CLS_ACT
dbaaa07a 47 switch (err) {
1da177e4
LT
48 case TC_ACT_STOLEN:
49 case TC_ACT_QUEUED:
378a2f09 50 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
1da177e4
LT
51 case TC_ACT_SHOT:
52 return NULL;
3ff50b79 53 }
1da177e4 54#endif
25d8c0d5 55 if (!fl || err < 0) {
1da177e4
LT
56 if (TC_H_MAJ(band))
57 band = 0;
cc7ec456 58 return q->queues[q->prio2band[band & TC_PRIO_MAX]];
1da177e4
LT
59 }
60 band = res.classid;
61 }
62 band = TC_H_MIN(band) - 1;
3e5c2d3b 63 if (band >= q->bands)
1d8ae3fd
DM
64 return q->queues[q->prio2band[0]];
65
1da177e4
LT
66 return q->queues[band];
67}
68
69static int
70prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
71{
72 struct Qdisc *qdisc;
73 int ret;
74
75 qdisc = prio_classify(skb, sch, &ret);
76#ifdef CONFIG_NET_CLS_ACT
77 if (qdisc == NULL) {
29f1df6c 78
c27f339a 79 if (ret & __NET_XMIT_BYPASS)
25331d6c 80 qdisc_qstats_drop(sch);
1da177e4
LT
81 kfree_skb(skb);
82 return ret;
83 }
84#endif
85
5f86173b
JK
86 ret = qdisc_enqueue(skb, qdisc);
87 if (ret == NET_XMIT_SUCCESS) {
1da177e4
LT
88 sch->q.qlen++;
89 return NET_XMIT_SUCCESS;
90 }
378a2f09 91 if (net_xmit_drop_count(ret))
25331d6c 92 qdisc_qstats_drop(sch);
10297b99 93 return ret;
1da177e4
LT
94}
95
48a8f519
PM
96static struct sk_buff *prio_peek(struct Qdisc *sch)
97{
98 struct prio_sched_data *q = qdisc_priv(sch);
99 int prio;
100
101 for (prio = 0; prio < q->bands; prio++) {
102 struct Qdisc *qdisc = q->queues[prio];
103 struct sk_buff *skb = qdisc->ops->peek(qdisc);
104 if (skb)
105 return skb;
106 }
107 return NULL;
108}
1da177e4 109
cc7ec456 110static struct sk_buff *prio_dequeue(struct Qdisc *sch)
1da177e4 111{
1da177e4
LT
112 struct prio_sched_data *q = qdisc_priv(sch);
113 int prio;
1da177e4
LT
114
115 for (prio = 0; prio < q->bands; prio++) {
1d8ae3fd 116 struct Qdisc *qdisc = q->queues[prio];
3557619f 117 struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
1d8ae3fd 118 if (skb) {
9190b3b3 119 qdisc_bstats_update(sch, skb);
1d8ae3fd
DM
120 sch->q.qlen--;
121 return skb;
1da177e4
LT
122 }
123 }
124 return NULL;
125
126}
127
cc7ec456 128static unsigned int prio_drop(struct Qdisc *sch)
1da177e4
LT
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];
6d037a26 137 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
1da177e4
LT
138 sch->q.qlen--;
139 return len;
140 }
141 }
142 return 0;
143}
144
145
146static void
cc7ec456 147prio_reset(struct Qdisc *sch)
1da177e4
LT
148{
149 int prio;
150 struct prio_sched_data *q = qdisc_priv(sch);
151
cc7ec456 152 for (prio = 0; prio < q->bands; prio++)
1da177e4
LT
153 qdisc_reset(q->queues[prio]);
154 sch->q.qlen = 0;
155}
156
157static void
cc7ec456 158prio_destroy(struct Qdisc *sch)
1da177e4
LT
159{
160 int prio;
161 struct prio_sched_data *q = qdisc_priv(sch);
1da177e4 162
ff31ab56 163 tcf_destroy_chain(&q->filter_list);
cc7ec456 164 for (prio = 0; prio < q->bands; prio++)
1da177e4
LT
165 qdisc_destroy(q->queues[prio]);
166}
167
1e90474c 168static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
1da177e4
LT
169{
170 struct prio_sched_data *q = qdisc_priv(sch);
d62733c8 171 struct tc_prio_qopt *qopt;
1da177e4
LT
172 int i;
173
1d8ae3fd
DM
174 if (nla_len(opt) < sizeof(*qopt))
175 return -EINVAL;
176 qopt = nla_data(opt);
d62733c8 177
1d8ae3fd 178 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
1da177e4
LT
179 return -EINVAL;
180
cc7ec456 181 for (i = 0; i <= TC_PRIO_MAX; i++) {
1d8ae3fd 182 if (qopt->priomap[i] >= qopt->bands)
1da177e4
LT
183 return -EINVAL;
184 }
185
186 sch_tree_lock(sch);
1d8ae3fd 187 q->bands = qopt->bands;
1da177e4
LT
188 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
189
cc7ec456 190 for (i = q->bands; i < TCQ_PRIO_BANDS; i++) {
b94c8afc
PM
191 struct Qdisc *child = q->queues[i];
192 q->queues[i] = &noop_qdisc;
5e50da01
PM
193 if (child != &noop_qdisc) {
194 qdisc_tree_decrease_qlen(child, child->q.qlen);
1da177e4 195 qdisc_destroy(child);
5e50da01 196 }
1da177e4
LT
197 }
198 sch_tree_unlock(sch);
199
cc7ec456 200 for (i = 0; i < q->bands; i++) {
dd914b40 201 if (q->queues[i] == &noop_qdisc) {
b94c8afc 202 struct Qdisc *child, *old;
cc7ec456 203
3511c913 204 child = qdisc_create_dflt(sch->dev_queue,
bb949fbd 205 &pfifo_qdisc_ops,
9f9afec4 206 TC_H_MAKE(sch->handle, i + 1));
1da177e4
LT
207 if (child) {
208 sch_tree_lock(sch);
b94c8afc
PM
209 old = q->queues[i];
210 q->queues[i] = child;
1da177e4 211
b94c8afc
PM
212 if (old != &noop_qdisc) {
213 qdisc_tree_decrease_qlen(old,
214 old->q.qlen);
215 qdisc_destroy(old);
5e50da01 216 }
1da177e4
LT
217 sch_tree_unlock(sch);
218 }
219 }
220 }
221 return 0;
222}
223
1e90474c 224static int prio_init(struct Qdisc *sch, struct nlattr *opt)
1da177e4
LT
225{
226 struct prio_sched_data *q = qdisc_priv(sch);
227 int i;
228
cc7ec456 229 for (i = 0; i < TCQ_PRIO_BANDS; i++)
1da177e4
LT
230 q->queues[i] = &noop_qdisc;
231
232 if (opt == NULL) {
233 return -EINVAL;
234 } else {
235 int err;
236
cc7ec456 237 if ((err = prio_tune(sch, opt)) != 0)
1da177e4
LT
238 return err;
239 }
240 return 0;
241}
242
243static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
244{
245 struct prio_sched_data *q = qdisc_priv(sch);
27a884dc 246 unsigned char *b = skb_tail_pointer(skb);
1da177e4
LT
247 struct tc_prio_qopt opt;
248
249 opt.bands = q->bands;
cc7ec456 250 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
d62733c8 251
1b34ec43
DM
252 if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
253 goto nla_put_failure;
d62733c8 254
1da177e4
LT
255 return skb->len;
256
1e90474c 257nla_put_failure:
dc5fc579 258 nlmsg_trim(skb, b);
1da177e4
LT
259 return -1;
260}
261
262static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
263 struct Qdisc **old)
264{
265 struct prio_sched_data *q = qdisc_priv(sch);
266 unsigned long band = arg - 1;
267
1da177e4
LT
268 if (new == NULL)
269 new = &noop_qdisc;
270
271 sch_tree_lock(sch);
272 *old = q->queues[band];
273 q->queues[band] = new;
5e50da01 274 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
1da177e4
LT
275 qdisc_reset(*old);
276 sch_tree_unlock(sch);
277
278 return 0;
279}
280
281static struct Qdisc *
282prio_leaf(struct Qdisc *sch, unsigned long arg)
283{
284 struct prio_sched_data *q = qdisc_priv(sch);
285 unsigned long band = arg - 1;
286
1da177e4
LT
287 return q->queues[band];
288}
289
290static unsigned long prio_get(struct Qdisc *sch, u32 classid)
291{
292 struct prio_sched_data *q = qdisc_priv(sch);
293 unsigned long band = TC_H_MIN(classid);
294
295 if (band - 1 >= q->bands)
296 return 0;
297 return band;
298}
299
300static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
301{
302 return prio_get(sch, classid);
303}
304
305
306static void prio_put(struct Qdisc *q, unsigned long cl)
307{
1da177e4
LT
308}
309
1da177e4
LT
310static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
311 struct tcmsg *tcm)
312{
313 struct prio_sched_data *q = qdisc_priv(sch);
314
1da177e4 315 tcm->tcm_handle |= TC_H_MIN(cl);
5b9a9ccf 316 tcm->tcm_info = q->queues[cl-1]->handle;
1da177e4
LT
317 return 0;
318}
319
2cf6c36c
JP
320static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
321 struct gnet_dump *d)
322{
323 struct prio_sched_data *q = qdisc_priv(sch);
324 struct Qdisc *cl_q;
325
326 cl_q = q->queues[cl - 1];
22e0f8b9 327 if (gnet_stats_copy_basic(d, NULL, &cl_q->bstats) < 0 ||
b0ab6f92 328 gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
2cf6c36c
JP
329 return -1;
330
331 return 0;
332}
333
1da177e4
LT
334static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
335{
336 struct prio_sched_data *q = qdisc_priv(sch);
337 int prio;
338
339 if (arg->stop)
340 return;
341
342 for (prio = 0; prio < q->bands; prio++) {
343 if (arg->count < arg->skip) {
344 arg->count++;
345 continue;
346 }
cc7ec456 347 if (arg->fn(sch, prio + 1, arg) < 0) {
1da177e4
LT
348 arg->stop = 1;
349 break;
350 }
351 arg->count++;
352 }
353}
354
25d8c0d5
JF
355static struct tcf_proto __rcu **prio_find_tcf(struct Qdisc *sch,
356 unsigned long cl)
1da177e4
LT
357{
358 struct prio_sched_data *q = qdisc_priv(sch);
359
360 if (cl)
361 return NULL;
362 return &q->filter_list;
363}
364
20fea08b 365static const struct Qdisc_class_ops prio_class_ops = {
1da177e4
LT
366 .graft = prio_graft,
367 .leaf = prio_leaf,
368 .get = prio_get,
369 .put = prio_put,
1da177e4
LT
370 .walk = prio_walk,
371 .tcf_chain = prio_find_tcf,
372 .bind_tcf = prio_bind,
373 .unbind_tcf = prio_put,
374 .dump = prio_dump_class,
2cf6c36c 375 .dump_stats = prio_dump_class_stats,
1da177e4
LT
376};
377
20fea08b 378static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
1da177e4
LT
379 .next = NULL,
380 .cl_ops = &prio_class_ops,
381 .id = "prio",
382 .priv_size = sizeof(struct prio_sched_data),
383 .enqueue = prio_enqueue,
384 .dequeue = prio_dequeue,
48a8f519 385 .peek = prio_peek,
1da177e4
LT
386 .drop = prio_drop,
387 .init = prio_init,
388 .reset = prio_reset,
389 .destroy = prio_destroy,
390 .change = prio_tune,
391 .dump = prio_dump,
392 .owner = THIS_MODULE,
393};
394
395static int __init prio_module_init(void)
396{
1d8ae3fd 397 return register_qdisc(&prio_qdisc_ops);
1da177e4
LT
398}
399
10297b99 400static void __exit prio_module_exit(void)
1da177e4
LT
401{
402 unregister_qdisc(&prio_qdisc_ops);
403}
404
405module_init(prio_module_init)
406module_exit(prio_module_exit)
407
408MODULE_LICENSE("GPL");