blob: 3ac0f495bad09e110285b913e023eff8bd1d8b2e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
41struct 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
50static struct Qdisc *
51prio_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_DROP;
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
85static int
86prio_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 if (ret == NET_XMIT_DROP)
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
112static int
113prio_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_DROP)
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
138static struct sk_buff *
139prio_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
158static 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 ((len = qdisc->ops->drop(qdisc)) != 0) {
168 sch->q.qlen--;
169 return len;
170 }
171 }
172 return 0;
173}
174
175
176static void
177prio_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
187static void
188prio_destroy(struct Qdisc* sch)
189{
190 int prio;
191 struct prio_sched_data *q = qdisc_priv(sch);
192 struct tcf_proto *tp;
193
194 while ((tp = q->filter_list) != NULL) {
195 q->filter_list = tp->next;
196 tcf_destroy(tp);
197 }
198
199 for (prio=0; prio<q->bands; prio++)
200 qdisc_destroy(q->queues[prio]);
201}
202
203static int prio_tune(struct Qdisc *sch, struct rtattr *opt)
204{
205 struct prio_sched_data *q = qdisc_priv(sch);
206 struct tc_prio_qopt *qopt = RTA_DATA(opt);
207 int i;
208
209 if (opt->rta_len < RTA_LENGTH(sizeof(*qopt)))
210 return -EINVAL;
211 if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
212 return -EINVAL;
213
214 for (i=0; i<=TC_PRIO_MAX; i++) {
215 if (qopt->priomap[i] >= qopt->bands)
216 return -EINVAL;
217 }
218
219 sch_tree_lock(sch);
220 q->bands = qopt->bands;
221 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
222
223 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
224 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
225 if (child != &noop_qdisc)
226 qdisc_destroy(child);
227 }
228 sch_tree_unlock(sch);
229
230 for (i=0; i<=TC_PRIO_MAX; i++) {
231 int band = q->prio2band[i];
232 if (q->queues[band] == &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[band], child);
238
239 if (child != &noop_qdisc)
240 qdisc_destroy(child);
241 sch_tree_unlock(sch);
242 }
243 }
244 }
245 return 0;
246}
247
248static 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
267static 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
278rtattr_failure:
279 skb_trim(skb, b - skb->data);
280 return -1;
281}
282
283static 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
305static struct Qdisc *
306prio_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
317static 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
327static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
328{
329 return prio_get(sch, classid);
330}
331
332
333static void prio_put(struct Qdisc *q, unsigned long cl)
334{
335 return;
336}
337
338static 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
348static 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
357static 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
370static 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
391static 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
400static 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
414static 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
431static int __init prio_module_init(void)
432{
433 return register_qdisc(&prio_qdisc_ops);
434}
435
436static void __exit prio_module_exit(void)
437{
438 unregister_qdisc(&prio_qdisc_ops);
439}
440
441module_init(prio_module_init)
442module_exit(prio_module_exit)
443
444MODULE_LICENSE("GPL");