blob: a5fa03c0c19b6c8167f2e68a4e4d3b052c0b0bcc [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#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/sched.h>
21#include <linux/string.h>
22#include <linux/mm.h>
23#include <linux/socket.h>
24#include <linux/sockios.h>
25#include <linux/in.h>
26#include <linux/errno.h>
27#include <linux/interrupt.h>
28#include <linux/if_ether.h>
29#include <linux/inet.h>
30#include <linux/netdevice.h>
31#include <linux/etherdevice.h>
32#include <linux/notifier.h>
33#include <net/ip.h>
34#include <net/route.h>
35#include <linux/skbuff.h>
36#include <net/sock.h>
37#include <net/pkt_sched.h>
38
39
40struct 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
49static struct Qdisc *
50prio_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
Jamal Hadi Salim29f1df62006-01-08 22:35:55 -080056 *qerr = NET_XMIT_BYPASS;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 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
84static int
85prio_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) {
Jamal Hadi Salim29f1df62006-01-08 22:35:55 -080093
94 if (ret == NET_XMIT_BYPASS)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 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) {
Jamal Hadi Salim29f1df62006-01-08 22:35:55 -0800121 if (ret == NET_XMIT_BYPASS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 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];
Patrick McHardy6d037a22006-03-20 19:00:49 -0800167 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 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
Amnon Aaronsohndd914b42006-01-17 02:24:26 -0800230 for (i=0; i<q->bands; i++) {
231 if (q->queues[i] == &noop_qdisc) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 struct Qdisc *child;
233 child = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops);
234 if (child) {
235 sch_tree_lock(sch);
Amnon Aaronsohndd914b42006-01-17 02:24:26 -0800236 child = xchg(&q->queues[i], child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
238 if (child != &noop_qdisc)
239 qdisc_destroy(child);
240 sch_tree_unlock(sch);
241 }
242 }
243 }
244 return 0;
245}
246
247static int prio_init(struct Qdisc *sch, struct rtattr *opt)
248{
249 struct prio_sched_data *q = qdisc_priv(sch);
250 int i;
251
252 for (i=0; i<TCQ_PRIO_BANDS; i++)
253 q->queues[i] = &noop_qdisc;
254
255 if (opt == NULL) {
256 return -EINVAL;
257 } else {
258 int err;
259
260 if ((err= prio_tune(sch, opt)) != 0)
261 return err;
262 }
263 return 0;
264}
265
266static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
267{
268 struct prio_sched_data *q = qdisc_priv(sch);
269 unsigned char *b = skb->tail;
270 struct tc_prio_qopt opt;
271
272 opt.bands = q->bands;
273 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
274 RTA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
275 return skb->len;
276
277rtattr_failure:
278 skb_trim(skb, b - skb->data);
279 return -1;
280}
281
282static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
283 struct Qdisc **old)
284{
285 struct prio_sched_data *q = qdisc_priv(sch);
286 unsigned long band = arg - 1;
287
288 if (band >= q->bands)
289 return -EINVAL;
290
291 if (new == NULL)
292 new = &noop_qdisc;
293
294 sch_tree_lock(sch);
295 *old = q->queues[band];
296 q->queues[band] = new;
297 sch->q.qlen -= (*old)->q.qlen;
298 qdisc_reset(*old);
299 sch_tree_unlock(sch);
300
301 return 0;
302}
303
304static struct Qdisc *
305prio_leaf(struct Qdisc *sch, unsigned long arg)
306{
307 struct prio_sched_data *q = qdisc_priv(sch);
308 unsigned long band = arg - 1;
309
310 if (band >= q->bands)
311 return NULL;
312
313 return q->queues[band];
314}
315
316static unsigned long prio_get(struct Qdisc *sch, u32 classid)
317{
318 struct prio_sched_data *q = qdisc_priv(sch);
319 unsigned long band = TC_H_MIN(classid);
320
321 if (band - 1 >= q->bands)
322 return 0;
323 return band;
324}
325
326static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
327{
328 return prio_get(sch, classid);
329}
330
331
332static void prio_put(struct Qdisc *q, unsigned long cl)
333{
334 return;
335}
336
337static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct rtattr **tca, unsigned long *arg)
338{
339 unsigned long cl = *arg;
340 struct prio_sched_data *q = qdisc_priv(sch);
341
342 if (cl - 1 > q->bands)
343 return -ENOENT;
344 return 0;
345}
346
347static int prio_delete(struct Qdisc *sch, unsigned long cl)
348{
349 struct prio_sched_data *q = qdisc_priv(sch);
350 if (cl - 1 > q->bands)
351 return -ENOENT;
352 return 0;
353}
354
355
356static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
357 struct tcmsg *tcm)
358{
359 struct prio_sched_data *q = qdisc_priv(sch);
360
361 if (cl - 1 > q->bands)
362 return -ENOENT;
363 tcm->tcm_handle |= TC_H_MIN(cl);
364 if (q->queues[cl-1])
365 tcm->tcm_info = q->queues[cl-1]->handle;
366 return 0;
367}
368
369static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
370{
371 struct prio_sched_data *q = qdisc_priv(sch);
372 int prio;
373
374 if (arg->stop)
375 return;
376
377 for (prio = 0; prio < q->bands; prio++) {
378 if (arg->count < arg->skip) {
379 arg->count++;
380 continue;
381 }
382 if (arg->fn(sch, prio+1, arg) < 0) {
383 arg->stop = 1;
384 break;
385 }
386 arg->count++;
387 }
388}
389
390static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
391{
392 struct prio_sched_data *q = qdisc_priv(sch);
393
394 if (cl)
395 return NULL;
396 return &q->filter_list;
397}
398
399static struct Qdisc_class_ops prio_class_ops = {
400 .graft = prio_graft,
401 .leaf = prio_leaf,
402 .get = prio_get,
403 .put = prio_put,
404 .change = prio_change,
405 .delete = prio_delete,
406 .walk = prio_walk,
407 .tcf_chain = prio_find_tcf,
408 .bind_tcf = prio_bind,
409 .unbind_tcf = prio_put,
410 .dump = prio_dump_class,
411};
412
413static struct Qdisc_ops prio_qdisc_ops = {
414 .next = NULL,
415 .cl_ops = &prio_class_ops,
416 .id = "prio",
417 .priv_size = sizeof(struct prio_sched_data),
418 .enqueue = prio_enqueue,
419 .dequeue = prio_dequeue,
420 .requeue = prio_requeue,
421 .drop = prio_drop,
422 .init = prio_init,
423 .reset = prio_reset,
424 .destroy = prio_destroy,
425 .change = prio_tune,
426 .dump = prio_dump,
427 .owner = THIS_MODULE,
428};
429
430static int __init prio_module_init(void)
431{
432 return register_qdisc(&prio_qdisc_ops);
433}
434
435static void __exit prio_module_exit(void)
436{
437 unregister_qdisc(&prio_qdisc_ops);
438}
439
440module_init(prio_module_init)
441module_exit(prio_module_exit)
442
443MODULE_LICENSE("GPL");