[WATCHDOG] VFS clean-up
[linux-2.6/mini2440.git] / net / sched / sch_prio.c
blob2d8c08493d6e9275683161418c0fd26cf4ec5a37
1 /*
2 * net/sched/sch_prio.c Simple 3-band priority "scheduler".
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.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * Fixes: 19990609: J Hadi Salim <hadi@nortelnetworks.com>:
11 * Init -- EINVAL when opt undefined
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <net/netlink.h>
21 #include <net/pkt_sched.h>
24 struct prio_sched_data
26 int bands;
27 int curband; /* for round-robin */
28 struct tcf_proto *filter_list;
29 u8 prio2band[TC_PRIO_MAX+1];
30 struct Qdisc *queues[TCQ_PRIO_BANDS];
31 int mq;
35 static struct Qdisc *
36 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
38 struct prio_sched_data *q = qdisc_priv(sch);
39 u32 band = skb->priority;
40 struct tcf_result res;
42 *qerr = NET_XMIT_BYPASS;
43 if (TC_H_MAJ(skb->priority) != sch->handle) {
44 #ifdef CONFIG_NET_CLS_ACT
45 switch (tc_classify(skb, q->filter_list, &res)) {
46 case TC_ACT_STOLEN:
47 case TC_ACT_QUEUED:
48 *qerr = NET_XMIT_SUCCESS;
49 case TC_ACT_SHOT:
50 return NULL;
53 if (!q->filter_list ) {
54 #else
55 if (!q->filter_list || tc_classify(skb, q->filter_list, &res)) {
56 #endif
57 if (TC_H_MAJ(band))
58 band = 0;
59 band = q->prio2band[band&TC_PRIO_MAX];
60 goto out;
62 band = res.classid;
64 band = TC_H_MIN(band) - 1;
65 if (band >= q->bands)
66 band = q->prio2band[0];
67 out:
68 if (q->mq)
69 skb_set_queue_mapping(skb, band);
70 return q->queues[band];
73 static int
74 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
76 struct Qdisc *qdisc;
77 int ret;
79 qdisc = prio_classify(skb, sch, &ret);
80 #ifdef CONFIG_NET_CLS_ACT
81 if (qdisc == NULL) {
83 if (ret == NET_XMIT_BYPASS)
84 sch->qstats.drops++;
85 kfree_skb(skb);
86 return ret;
88 #endif
90 if ((ret = qdisc->enqueue(skb, qdisc)) == NET_XMIT_SUCCESS) {
91 sch->bstats.bytes += skb->len;
92 sch->bstats.packets++;
93 sch->q.qlen++;
94 return NET_XMIT_SUCCESS;
96 sch->qstats.drops++;
97 return ret;
101 static int
102 prio_requeue(struct sk_buff *skb, struct Qdisc* sch)
104 struct Qdisc *qdisc;
105 int ret;
107 qdisc = prio_classify(skb, sch, &ret);
108 #ifdef CONFIG_NET_CLS_ACT
109 if (qdisc == NULL) {
110 if (ret == NET_XMIT_BYPASS)
111 sch->qstats.drops++;
112 kfree_skb(skb);
113 return ret;
115 #endif
117 if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) {
118 sch->q.qlen++;
119 sch->qstats.requeues++;
120 return 0;
122 sch->qstats.drops++;
123 return NET_XMIT_DROP;
127 static struct sk_buff *
128 prio_dequeue(struct Qdisc* sch)
130 struct sk_buff *skb;
131 struct prio_sched_data *q = qdisc_priv(sch);
132 int prio;
133 struct Qdisc *qdisc;
135 for (prio = 0; prio < q->bands; prio++) {
136 /* Check if the target subqueue is available before
137 * pulling an skb. This way we avoid excessive requeues
138 * for slower queues.
140 if (!netif_subqueue_stopped(sch->dev, (q->mq ? prio : 0))) {
141 qdisc = q->queues[prio];
142 skb = qdisc->dequeue(qdisc);
143 if (skb) {
144 sch->q.qlen--;
145 return skb;
149 return NULL;
153 static struct sk_buff *rr_dequeue(struct Qdisc* sch)
155 struct sk_buff *skb;
156 struct prio_sched_data *q = qdisc_priv(sch);
157 struct Qdisc *qdisc;
158 int bandcount;
160 /* Only take one pass through the queues. If nothing is available,
161 * return nothing.
163 for (bandcount = 0; bandcount < q->bands; bandcount++) {
164 /* Check if the target subqueue is available before
165 * pulling an skb. This way we avoid excessive requeues
166 * for slower queues. If the queue is stopped, try the
167 * next queue.
169 if (!netif_subqueue_stopped(sch->dev,
170 (q->mq ? q->curband : 0))) {
171 qdisc = q->queues[q->curband];
172 skb = qdisc->dequeue(qdisc);
173 if (skb) {
174 sch->q.qlen--;
175 q->curband++;
176 if (q->curband >= q->bands)
177 q->curband = 0;
178 return skb;
181 q->curband++;
182 if (q->curband >= q->bands)
183 q->curband = 0;
185 return NULL;
188 static unsigned int prio_drop(struct Qdisc* sch)
190 struct prio_sched_data *q = qdisc_priv(sch);
191 int prio;
192 unsigned int len;
193 struct Qdisc *qdisc;
195 for (prio = q->bands-1; prio >= 0; prio--) {
196 qdisc = q->queues[prio];
197 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
198 sch->q.qlen--;
199 return len;
202 return 0;
206 static void
207 prio_reset(struct Qdisc* sch)
209 int prio;
210 struct prio_sched_data *q = qdisc_priv(sch);
212 for (prio=0; prio<q->bands; prio++)
213 qdisc_reset(q->queues[prio]);
214 sch->q.qlen = 0;
217 static void
218 prio_destroy(struct Qdisc* sch)
220 int prio;
221 struct prio_sched_data *q = qdisc_priv(sch);
223 tcf_destroy_chain(q->filter_list);
224 for (prio=0; prio<q->bands; prio++)
225 qdisc_destroy(q->queues[prio]);
228 static int prio_tune(struct Qdisc *sch, struct rtattr *opt)
230 struct prio_sched_data *q = qdisc_priv(sch);
231 struct tc_prio_qopt *qopt;
232 struct rtattr *tb[TCA_PRIO_MAX];
233 int i;
235 if (rtattr_parse_nested_compat(tb, TCA_PRIO_MAX, opt, qopt,
236 sizeof(*qopt)))
237 return -EINVAL;
238 q->bands = qopt->bands;
239 /* If we're multiqueue, make sure the number of incoming bands
240 * matches the number of queues on the device we're associating with.
241 * If the number of bands requested is zero, then set q->bands to
242 * dev->egress_subqueue_count.
244 q->mq = RTA_GET_FLAG(tb[TCA_PRIO_MQ - 1]);
245 if (q->mq) {
246 if (sch->handle != TC_H_ROOT)
247 return -EINVAL;
248 if (netif_is_multiqueue(sch->dev)) {
249 if (q->bands == 0)
250 q->bands = sch->dev->egress_subqueue_count;
251 else if (q->bands != sch->dev->egress_subqueue_count)
252 return -EINVAL;
253 } else
254 return -EOPNOTSUPP;
257 if (q->bands > TCQ_PRIO_BANDS || q->bands < 2)
258 return -EINVAL;
260 for (i=0; i<=TC_PRIO_MAX; i++) {
261 if (qopt->priomap[i] >= q->bands)
262 return -EINVAL;
265 sch_tree_lock(sch);
266 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
268 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
269 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
270 if (child != &noop_qdisc) {
271 qdisc_tree_decrease_qlen(child, child->q.qlen);
272 qdisc_destroy(child);
275 sch_tree_unlock(sch);
277 for (i=0; i<q->bands; i++) {
278 if (q->queues[i] == &noop_qdisc) {
279 struct Qdisc *child;
280 child = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
281 TC_H_MAKE(sch->handle, i + 1));
282 if (child) {
283 sch_tree_lock(sch);
284 child = xchg(&q->queues[i], child);
286 if (child != &noop_qdisc) {
287 qdisc_tree_decrease_qlen(child,
288 child->q.qlen);
289 qdisc_destroy(child);
291 sch_tree_unlock(sch);
295 return 0;
298 static int prio_init(struct Qdisc *sch, struct rtattr *opt)
300 struct prio_sched_data *q = qdisc_priv(sch);
301 int i;
303 for (i=0; i<TCQ_PRIO_BANDS; i++)
304 q->queues[i] = &noop_qdisc;
306 if (opt == NULL) {
307 return -EINVAL;
308 } else {
309 int err;
311 if ((err= prio_tune(sch, opt)) != 0)
312 return err;
314 return 0;
317 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
319 struct prio_sched_data *q = qdisc_priv(sch);
320 unsigned char *b = skb_tail_pointer(skb);
321 struct rtattr *nest;
322 struct tc_prio_qopt opt;
324 opt.bands = q->bands;
325 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
327 nest = RTA_NEST_COMPAT(skb, TCA_OPTIONS, sizeof(opt), &opt);
328 if (q->mq)
329 RTA_PUT_FLAG(skb, TCA_PRIO_MQ);
330 RTA_NEST_COMPAT_END(skb, nest);
332 return skb->len;
334 rtattr_failure:
335 nlmsg_trim(skb, b);
336 return -1;
339 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
340 struct Qdisc **old)
342 struct prio_sched_data *q = qdisc_priv(sch);
343 unsigned long band = arg - 1;
345 if (band >= q->bands)
346 return -EINVAL;
348 if (new == NULL)
349 new = &noop_qdisc;
351 sch_tree_lock(sch);
352 *old = q->queues[band];
353 q->queues[band] = new;
354 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
355 qdisc_reset(*old);
356 sch_tree_unlock(sch);
358 return 0;
361 static struct Qdisc *
362 prio_leaf(struct Qdisc *sch, unsigned long arg)
364 struct prio_sched_data *q = qdisc_priv(sch);
365 unsigned long band = arg - 1;
367 if (band >= q->bands)
368 return NULL;
370 return q->queues[band];
373 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
375 struct prio_sched_data *q = qdisc_priv(sch);
376 unsigned long band = TC_H_MIN(classid);
378 if (band - 1 >= q->bands)
379 return 0;
380 return band;
383 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
385 return prio_get(sch, classid);
389 static void prio_put(struct Qdisc *q, unsigned long cl)
391 return;
394 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct rtattr **tca, unsigned long *arg)
396 unsigned long cl = *arg;
397 struct prio_sched_data *q = qdisc_priv(sch);
399 if (cl - 1 > q->bands)
400 return -ENOENT;
401 return 0;
404 static int prio_delete(struct Qdisc *sch, unsigned long cl)
406 struct prio_sched_data *q = qdisc_priv(sch);
407 if (cl - 1 > q->bands)
408 return -ENOENT;
409 return 0;
413 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
414 struct tcmsg *tcm)
416 struct prio_sched_data *q = qdisc_priv(sch);
418 if (cl - 1 > q->bands)
419 return -ENOENT;
420 tcm->tcm_handle |= TC_H_MIN(cl);
421 if (q->queues[cl-1])
422 tcm->tcm_info = q->queues[cl-1]->handle;
423 return 0;
426 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
427 struct gnet_dump *d)
429 struct prio_sched_data *q = qdisc_priv(sch);
430 struct Qdisc *cl_q;
432 cl_q = q->queues[cl - 1];
433 if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
434 gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
435 return -1;
437 return 0;
440 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
442 struct prio_sched_data *q = qdisc_priv(sch);
443 int prio;
445 if (arg->stop)
446 return;
448 for (prio = 0; prio < q->bands; prio++) {
449 if (arg->count < arg->skip) {
450 arg->count++;
451 continue;
453 if (arg->fn(sch, prio+1, arg) < 0) {
454 arg->stop = 1;
455 break;
457 arg->count++;
461 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
463 struct prio_sched_data *q = qdisc_priv(sch);
465 if (cl)
466 return NULL;
467 return &q->filter_list;
470 static struct Qdisc_class_ops prio_class_ops = {
471 .graft = prio_graft,
472 .leaf = prio_leaf,
473 .get = prio_get,
474 .put = prio_put,
475 .change = prio_change,
476 .delete = prio_delete,
477 .walk = prio_walk,
478 .tcf_chain = prio_find_tcf,
479 .bind_tcf = prio_bind,
480 .unbind_tcf = prio_put,
481 .dump = prio_dump_class,
482 .dump_stats = prio_dump_class_stats,
485 static struct Qdisc_ops prio_qdisc_ops = {
486 .next = NULL,
487 .cl_ops = &prio_class_ops,
488 .id = "prio",
489 .priv_size = sizeof(struct prio_sched_data),
490 .enqueue = prio_enqueue,
491 .dequeue = prio_dequeue,
492 .requeue = prio_requeue,
493 .drop = prio_drop,
494 .init = prio_init,
495 .reset = prio_reset,
496 .destroy = prio_destroy,
497 .change = prio_tune,
498 .dump = prio_dump,
499 .owner = THIS_MODULE,
502 static struct Qdisc_ops rr_qdisc_ops = {
503 .next = NULL,
504 .cl_ops = &prio_class_ops,
505 .id = "rr",
506 .priv_size = sizeof(struct prio_sched_data),
507 .enqueue = prio_enqueue,
508 .dequeue = rr_dequeue,
509 .requeue = prio_requeue,
510 .drop = prio_drop,
511 .init = prio_init,
512 .reset = prio_reset,
513 .destroy = prio_destroy,
514 .change = prio_tune,
515 .dump = prio_dump,
516 .owner = THIS_MODULE,
519 static int __init prio_module_init(void)
521 int err;
523 err = register_qdisc(&prio_qdisc_ops);
524 if (err < 0)
525 return err;
526 err = register_qdisc(&rr_qdisc_ops);
527 if (err < 0)
528 unregister_qdisc(&prio_qdisc_ops);
529 return err;
532 static void __exit prio_module_exit(void)
534 unregister_qdisc(&prio_qdisc_ops);
535 unregister_qdisc(&rr_qdisc_ops);
538 module_init(prio_module_init)
539 module_exit(prio_module_exit)
541 MODULE_LICENSE("GPL");
542 MODULE_ALIAS("sch_rr");