Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[pv_ops_mirror.git] / net / sched / sch_prio.c
blobde894096e44252e6cc8da2ff548071a82b295685
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;
41 int err;
43 *qerr = NET_XMIT_BYPASS;
44 if (TC_H_MAJ(skb->priority) != sch->handle) {
45 err = tc_classify(skb, q->filter_list, &res);
46 #ifdef CONFIG_NET_CLS_ACT
47 switch (err) {
48 case TC_ACT_STOLEN:
49 case TC_ACT_QUEUED:
50 *qerr = NET_XMIT_SUCCESS;
51 case TC_ACT_SHOT:
52 return NULL;
54 #endif
55 if (!q->filter_list || err < 0) {
56 if (TC_H_MAJ(band))
57 band = 0;
58 band = q->prio2band[band&TC_PRIO_MAX];
59 goto out;
61 band = res.classid;
63 band = TC_H_MIN(band) - 1;
64 if (band >= q->bands)
65 band = q->prio2band[0];
66 out:
67 if (q->mq)
68 skb_set_queue_mapping(skb, band);
69 return q->queues[band];
72 static int
73 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
75 struct Qdisc *qdisc;
76 int ret;
78 qdisc = prio_classify(skb, sch, &ret);
79 #ifdef CONFIG_NET_CLS_ACT
80 if (qdisc == NULL) {
82 if (ret == NET_XMIT_BYPASS)
83 sch->qstats.drops++;
84 kfree_skb(skb);
85 return ret;
87 #endif
89 if ((ret = qdisc->enqueue(skb, qdisc)) == NET_XMIT_SUCCESS) {
90 sch->bstats.bytes += skb->len;
91 sch->bstats.packets++;
92 sch->q.qlen++;
93 return NET_XMIT_SUCCESS;
95 sch->qstats.drops++;
96 return ret;
100 static int
101 prio_requeue(struct sk_buff *skb, struct Qdisc* sch)
103 struct Qdisc *qdisc;
104 int ret;
106 qdisc = prio_classify(skb, sch, &ret);
107 #ifdef CONFIG_NET_CLS_ACT
108 if (qdisc == NULL) {
109 if (ret == NET_XMIT_BYPASS)
110 sch->qstats.drops++;
111 kfree_skb(skb);
112 return ret;
114 #endif
116 if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) {
117 sch->q.qlen++;
118 sch->qstats.requeues++;
119 return 0;
121 sch->qstats.drops++;
122 return NET_XMIT_DROP;
126 static struct sk_buff *
127 prio_dequeue(struct Qdisc* sch)
129 struct sk_buff *skb;
130 struct prio_sched_data *q = qdisc_priv(sch);
131 int prio;
132 struct Qdisc *qdisc;
134 for (prio = 0; prio < q->bands; prio++) {
135 /* Check if the target subqueue is available before
136 * pulling an skb. This way we avoid excessive requeues
137 * for slower queues.
139 if (!__netif_subqueue_stopped(sch->dev, (q->mq ? prio : 0))) {
140 qdisc = q->queues[prio];
141 skb = qdisc->dequeue(qdisc);
142 if (skb) {
143 sch->q.qlen--;
144 return skb;
148 return NULL;
152 static struct sk_buff *rr_dequeue(struct Qdisc* sch)
154 struct sk_buff *skb;
155 struct prio_sched_data *q = qdisc_priv(sch);
156 struct Qdisc *qdisc;
157 int bandcount;
159 /* Only take one pass through the queues. If nothing is available,
160 * return nothing.
162 for (bandcount = 0; bandcount < q->bands; bandcount++) {
163 /* Check if the target subqueue is available before
164 * pulling an skb. This way we avoid excessive requeues
165 * for slower queues. If the queue is stopped, try the
166 * next queue.
168 if (!__netif_subqueue_stopped(sch->dev,
169 (q->mq ? q->curband : 0))) {
170 qdisc = q->queues[q->curband];
171 skb = qdisc->dequeue(qdisc);
172 if (skb) {
173 sch->q.qlen--;
174 q->curband++;
175 if (q->curband >= q->bands)
176 q->curband = 0;
177 return skb;
180 q->curband++;
181 if (q->curband >= q->bands)
182 q->curband = 0;
184 return NULL;
187 static unsigned int prio_drop(struct Qdisc* sch)
189 struct prio_sched_data *q = qdisc_priv(sch);
190 int prio;
191 unsigned int len;
192 struct Qdisc *qdisc;
194 for (prio = q->bands-1; prio >= 0; prio--) {
195 qdisc = q->queues[prio];
196 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
197 sch->q.qlen--;
198 return len;
201 return 0;
205 static void
206 prio_reset(struct Qdisc* sch)
208 int prio;
209 struct prio_sched_data *q = qdisc_priv(sch);
211 for (prio=0; prio<q->bands; prio++)
212 qdisc_reset(q->queues[prio]);
213 sch->q.qlen = 0;
216 static void
217 prio_destroy(struct Qdisc* sch)
219 int prio;
220 struct prio_sched_data *q = qdisc_priv(sch);
222 tcf_destroy_chain(q->filter_list);
223 for (prio=0; prio<q->bands; prio++)
224 qdisc_destroy(q->queues[prio]);
227 static int prio_tune(struct Qdisc *sch, struct rtattr *opt)
229 struct prio_sched_data *q = qdisc_priv(sch);
230 struct tc_prio_qopt *qopt;
231 struct rtattr *tb[TCA_PRIO_MAX];
232 int i;
234 if (rtattr_parse_nested_compat(tb, TCA_PRIO_MAX, opt, qopt,
235 sizeof(*qopt)))
236 return -EINVAL;
237 q->bands = qopt->bands;
238 /* If we're multiqueue, make sure the number of incoming bands
239 * matches the number of queues on the device we're associating with.
240 * If the number of bands requested is zero, then set q->bands to
241 * dev->egress_subqueue_count. Also, the root qdisc must be the
242 * only one that is enabled for multiqueue, since it's the only one
243 * that interacts with the underlying device.
245 q->mq = RTA_GET_FLAG(tb[TCA_PRIO_MQ - 1]);
246 if (q->mq) {
247 if (sch->parent != TC_H_ROOT)
248 return -EINVAL;
249 if (netif_is_multiqueue(sch->dev)) {
250 if (q->bands == 0)
251 q->bands = sch->dev->egress_subqueue_count;
252 else if (q->bands != sch->dev->egress_subqueue_count)
253 return -EINVAL;
254 } else
255 return -EOPNOTSUPP;
258 if (q->bands > TCQ_PRIO_BANDS || q->bands < 2)
259 return -EINVAL;
261 for (i=0; i<=TC_PRIO_MAX; i++) {
262 if (qopt->priomap[i] >= q->bands)
263 return -EINVAL;
266 sch_tree_lock(sch);
267 memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
269 for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
270 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
271 if (child != &noop_qdisc) {
272 qdisc_tree_decrease_qlen(child, child->q.qlen);
273 qdisc_destroy(child);
276 sch_tree_unlock(sch);
278 for (i=0; i<q->bands; i++) {
279 if (q->queues[i] == &noop_qdisc) {
280 struct Qdisc *child;
281 child = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops,
282 TC_H_MAKE(sch->handle, i + 1));
283 if (child) {
284 sch_tree_lock(sch);
285 child = xchg(&q->queues[i], child);
287 if (child != &noop_qdisc) {
288 qdisc_tree_decrease_qlen(child,
289 child->q.qlen);
290 qdisc_destroy(child);
292 sch_tree_unlock(sch);
296 return 0;
299 static int prio_init(struct Qdisc *sch, struct rtattr *opt)
301 struct prio_sched_data *q = qdisc_priv(sch);
302 int i;
304 for (i=0; i<TCQ_PRIO_BANDS; i++)
305 q->queues[i] = &noop_qdisc;
307 if (opt == NULL) {
308 return -EINVAL;
309 } else {
310 int err;
312 if ((err= prio_tune(sch, opt)) != 0)
313 return err;
315 return 0;
318 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
320 struct prio_sched_data *q = qdisc_priv(sch);
321 unsigned char *b = skb_tail_pointer(skb);
322 struct rtattr *nest;
323 struct tc_prio_qopt opt;
325 opt.bands = q->bands;
326 memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
328 nest = RTA_NEST_COMPAT(skb, TCA_OPTIONS, sizeof(opt), &opt);
329 if (q->mq)
330 RTA_PUT_FLAG(skb, TCA_PRIO_MQ);
331 RTA_NEST_COMPAT_END(skb, nest);
333 return skb->len;
335 rtattr_failure:
336 nlmsg_trim(skb, b);
337 return -1;
340 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
341 struct Qdisc **old)
343 struct prio_sched_data *q = qdisc_priv(sch);
344 unsigned long band = arg - 1;
346 if (band >= q->bands)
347 return -EINVAL;
349 if (new == NULL)
350 new = &noop_qdisc;
352 sch_tree_lock(sch);
353 *old = q->queues[band];
354 q->queues[band] = new;
355 qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
356 qdisc_reset(*old);
357 sch_tree_unlock(sch);
359 return 0;
362 static struct Qdisc *
363 prio_leaf(struct Qdisc *sch, unsigned long arg)
365 struct prio_sched_data *q = qdisc_priv(sch);
366 unsigned long band = arg - 1;
368 if (band >= q->bands)
369 return NULL;
371 return q->queues[band];
374 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
376 struct prio_sched_data *q = qdisc_priv(sch);
377 unsigned long band = TC_H_MIN(classid);
379 if (band - 1 >= q->bands)
380 return 0;
381 return band;
384 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
386 return prio_get(sch, classid);
390 static void prio_put(struct Qdisc *q, unsigned long cl)
392 return;
395 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct rtattr **tca, unsigned long *arg)
397 unsigned long cl = *arg;
398 struct prio_sched_data *q = qdisc_priv(sch);
400 if (cl - 1 > q->bands)
401 return -ENOENT;
402 return 0;
405 static int prio_delete(struct Qdisc *sch, unsigned long cl)
407 struct prio_sched_data *q = qdisc_priv(sch);
408 if (cl - 1 > q->bands)
409 return -ENOENT;
410 return 0;
414 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
415 struct tcmsg *tcm)
417 struct prio_sched_data *q = qdisc_priv(sch);
419 if (cl - 1 > q->bands)
420 return -ENOENT;
421 tcm->tcm_handle |= TC_H_MIN(cl);
422 if (q->queues[cl-1])
423 tcm->tcm_info = q->queues[cl-1]->handle;
424 return 0;
427 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
428 struct gnet_dump *d)
430 struct prio_sched_data *q = qdisc_priv(sch);
431 struct Qdisc *cl_q;
433 cl_q = q->queues[cl - 1];
434 if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
435 gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
436 return -1;
438 return 0;
441 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
443 struct prio_sched_data *q = qdisc_priv(sch);
444 int prio;
446 if (arg->stop)
447 return;
449 for (prio = 0; prio < q->bands; prio++) {
450 if (arg->count < arg->skip) {
451 arg->count++;
452 continue;
454 if (arg->fn(sch, prio+1, arg) < 0) {
455 arg->stop = 1;
456 break;
458 arg->count++;
462 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
464 struct prio_sched_data *q = qdisc_priv(sch);
466 if (cl)
467 return NULL;
468 return &q->filter_list;
471 static struct Qdisc_class_ops prio_class_ops = {
472 .graft = prio_graft,
473 .leaf = prio_leaf,
474 .get = prio_get,
475 .put = prio_put,
476 .change = prio_change,
477 .delete = prio_delete,
478 .walk = prio_walk,
479 .tcf_chain = prio_find_tcf,
480 .bind_tcf = prio_bind,
481 .unbind_tcf = prio_put,
482 .dump = prio_dump_class,
483 .dump_stats = prio_dump_class_stats,
486 static struct Qdisc_ops prio_qdisc_ops = {
487 .next = NULL,
488 .cl_ops = &prio_class_ops,
489 .id = "prio",
490 .priv_size = sizeof(struct prio_sched_data),
491 .enqueue = prio_enqueue,
492 .dequeue = prio_dequeue,
493 .requeue = prio_requeue,
494 .drop = prio_drop,
495 .init = prio_init,
496 .reset = prio_reset,
497 .destroy = prio_destroy,
498 .change = prio_tune,
499 .dump = prio_dump,
500 .owner = THIS_MODULE,
503 static struct Qdisc_ops rr_qdisc_ops = {
504 .next = NULL,
505 .cl_ops = &prio_class_ops,
506 .id = "rr",
507 .priv_size = sizeof(struct prio_sched_data),
508 .enqueue = prio_enqueue,
509 .dequeue = rr_dequeue,
510 .requeue = prio_requeue,
511 .drop = prio_drop,
512 .init = prio_init,
513 .reset = prio_reset,
514 .destroy = prio_destroy,
515 .change = prio_tune,
516 .dump = prio_dump,
517 .owner = THIS_MODULE,
520 static int __init prio_module_init(void)
522 int err;
524 err = register_qdisc(&prio_qdisc_ops);
525 if (err < 0)
526 return err;
527 err = register_qdisc(&rr_qdisc_ops);
528 if (err < 0)
529 unregister_qdisc(&prio_qdisc_ops);
530 return err;
533 static void __exit prio_module_exit(void)
535 unregister_qdisc(&prio_qdisc_ops);
536 unregister_qdisc(&rr_qdisc_ops);
539 module_init(prio_module_init)
540 module_exit(prio_module_exit)
542 MODULE_LICENSE("GPL");
543 MODULE_ALIAS("sch_rr");