2 * net/sched/sch_drr.c Deficit Round Robin scheduler
4 * Copyright (c) 2008 Patrick McHardy <kaber@trash.net>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * version 2 as published by the Free Software Foundation.
11 #include <linux/module.h>
12 #include <linux/init.h>
13 #include <linux/errno.h>
14 #include <linux/netdevice.h>
15 #include <linux/pkt_sched.h>
16 #include <net/sch_generic.h>
17 #include <net/pkt_sched.h>
18 #include <net/pkt_cls.h>
21 struct Qdisc_class_common common;
23 unsigned int filter_cnt;
25 struct gnet_stats_basic bstats;
26 struct gnet_stats_queue qstats;
27 struct gnet_stats_rate_est rate_est;
28 struct list_head alist;
36 struct list_head active;
37 struct tcf_proto *filter_list;
38 struct Qdisc_class_hash clhash;
41 static struct drr_class *drr_find_class(struct Qdisc *sch, u32 classid)
43 struct drr_sched *q = qdisc_priv(sch);
44 struct Qdisc_class_common *clc;
46 clc = qdisc_class_find(&q->clhash, classid);
49 return container_of(clc, struct drr_class, common);
52 static void drr_purge_queue(struct drr_class *cl)
54 unsigned int len = cl->qdisc->q.qlen;
56 qdisc_reset(cl->qdisc);
57 qdisc_tree_decrease_qlen(cl->qdisc, len);
60 static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = {
61 [TCA_DRR_QUANTUM] = { .type = NLA_U32 },
64 static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
65 struct nlattr **tca, unsigned long *arg)
67 struct drr_sched *q = qdisc_priv(sch);
68 struct drr_class *cl = (struct drr_class *)*arg;
69 struct nlattr *opt = tca[TCA_OPTIONS];
70 struct nlattr *tb[TCA_DRR_MAX + 1];
77 err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy);
81 if (tb[TCA_DRR_QUANTUM]) {
82 quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]);
86 quantum = psched_mtu(qdisc_dev(sch));
90 err = gen_replace_estimator(&cl->bstats, &cl->rate_est,
91 qdisc_root_sleeping_lock(sch),
98 if (tb[TCA_DRR_QUANTUM])
99 cl->quantum = quantum;
100 sch_tree_unlock(sch);
105 cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL);
110 cl->common.classid = classid;
111 cl->quantum = quantum;
112 cl->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
113 &pfifo_qdisc_ops, classid);
114 if (cl->qdisc == NULL)
115 cl->qdisc = &noop_qdisc;
118 err = gen_replace_estimator(&cl->bstats, &cl->rate_est,
119 qdisc_root_sleeping_lock(sch),
122 qdisc_destroy(cl->qdisc);
129 qdisc_class_hash_insert(&q->clhash, &cl->common);
130 sch_tree_unlock(sch);
132 qdisc_class_hash_grow(sch, &q->clhash);
134 *arg = (unsigned long)cl;
138 static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl)
140 gen_kill_estimator(&cl->bstats, &cl->rate_est);
141 qdisc_destroy(cl->qdisc);
145 static int drr_delete_class(struct Qdisc *sch, unsigned long arg)
147 struct drr_sched *q = qdisc_priv(sch);
148 struct drr_class *cl = (struct drr_class *)arg;
150 if (cl->filter_cnt > 0)
156 qdisc_class_hash_remove(&q->clhash, &cl->common);
158 BUG_ON(--cl->refcnt == 0);
160 * This shouldn't happen: we "hold" one cops->get() when called
161 * from tc_ctl_tclass; the destroy method is done from cops->put().
164 sch_tree_unlock(sch);
168 static unsigned long drr_get_class(struct Qdisc *sch, u32 classid)
170 struct drr_class *cl = drr_find_class(sch, classid);
175 return (unsigned long)cl;
178 static void drr_put_class(struct Qdisc *sch, unsigned long arg)
180 struct drr_class *cl = (struct drr_class *)arg;
182 if (--cl->refcnt == 0)
183 drr_destroy_class(sch, cl);
186 static struct tcf_proto **drr_tcf_chain(struct Qdisc *sch, unsigned long cl)
188 struct drr_sched *q = qdisc_priv(sch);
193 return &q->filter_list;
196 static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent,
199 struct drr_class *cl = drr_find_class(sch, classid);
204 return (unsigned long)cl;
207 static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg)
209 struct drr_class *cl = (struct drr_class *)arg;
214 static int drr_graft_class(struct Qdisc *sch, unsigned long arg,
215 struct Qdisc *new, struct Qdisc **old)
217 struct drr_class *cl = (struct drr_class *)arg;
220 new = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
221 &pfifo_qdisc_ops, cl->common.classid);
230 sch_tree_unlock(sch);
234 static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg)
236 struct drr_class *cl = (struct drr_class *)arg;
241 static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg)
243 struct drr_class *cl = (struct drr_class *)arg;
245 if (cl->qdisc->q.qlen == 0)
246 list_del(&cl->alist);
249 static int drr_dump_class(struct Qdisc *sch, unsigned long arg,
250 struct sk_buff *skb, struct tcmsg *tcm)
252 struct drr_class *cl = (struct drr_class *)arg;
255 tcm->tcm_parent = TC_H_ROOT;
256 tcm->tcm_handle = cl->common.classid;
257 tcm->tcm_info = cl->qdisc->handle;
259 nest = nla_nest_start(skb, TCA_OPTIONS);
261 goto nla_put_failure;
262 NLA_PUT_U32(skb, TCA_DRR_QUANTUM, cl->quantum);
263 return nla_nest_end(skb, nest);
266 nla_nest_cancel(skb, nest);
270 static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg,
273 struct drr_class *cl = (struct drr_class *)arg;
274 struct tc_drr_stats xstats;
276 memset(&xstats, 0, sizeof(xstats));
277 if (cl->qdisc->q.qlen)
278 xstats.deficit = cl->deficit;
280 if (gnet_stats_copy_basic(d, &cl->bstats) < 0 ||
281 gnet_stats_copy_rate_est(d, &cl->rate_est) < 0 ||
282 gnet_stats_copy_queue(d, &cl->qdisc->qstats) < 0)
285 return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
288 static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg)
290 struct drr_sched *q = qdisc_priv(sch);
291 struct drr_class *cl;
292 struct hlist_node *n;
298 for (i = 0; i < q->clhash.hashsize; i++) {
299 hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) {
300 if (arg->count < arg->skip) {
304 if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
313 static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch,
316 struct drr_sched *q = qdisc_priv(sch);
317 struct drr_class *cl;
318 struct tcf_result res;
321 if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) {
322 cl = drr_find_class(sch, skb->priority);
327 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
328 result = tc_classify(skb, q->filter_list, &res);
330 #ifdef CONFIG_NET_CLS_ACT
334 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
339 cl = (struct drr_class *)res.class;
341 cl = drr_find_class(sch, res.classid);
347 static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch)
349 struct drr_sched *q = qdisc_priv(sch);
350 struct drr_class *cl;
354 cl = drr_classify(skb, sch, &err);
356 if (err & __NET_XMIT_BYPASS)
362 len = qdisc_pkt_len(skb);
363 err = qdisc_enqueue(skb, cl->qdisc);
364 if (unlikely(err != NET_XMIT_SUCCESS)) {
365 if (net_xmit_drop_count(err)) {
372 if (cl->qdisc->q.qlen == 1) {
373 list_add_tail(&cl->alist, &q->active);
374 cl->deficit = cl->quantum;
377 cl->bstats.packets++;
378 cl->bstats.bytes += len;
379 sch->bstats.packets++;
380 sch->bstats.bytes += len;
386 static struct sk_buff *drr_dequeue(struct Qdisc *sch)
388 struct drr_sched *q = qdisc_priv(sch);
389 struct drr_class *cl;
393 if (list_empty(&q->active))
396 cl = list_first_entry(&q->active, struct drr_class, alist);
397 skb = cl->qdisc->ops->peek(cl->qdisc);
401 len = qdisc_pkt_len(skb);
402 if (len <= cl->deficit) {
404 skb = qdisc_dequeue_peeked(cl->qdisc);
405 if (cl->qdisc->q.qlen == 0)
406 list_del(&cl->alist);
411 cl->deficit += cl->quantum;
412 list_move_tail(&cl->alist, &q->active);
418 static unsigned int drr_drop(struct Qdisc *sch)
420 struct drr_sched *q = qdisc_priv(sch);
421 struct drr_class *cl;
424 list_for_each_entry(cl, &q->active, alist) {
425 if (cl->qdisc->ops->drop) {
426 len = cl->qdisc->ops->drop(cl->qdisc);
429 if (cl->qdisc->q.qlen == 0)
430 list_del(&cl->alist);
438 static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
440 struct drr_sched *q = qdisc_priv(sch);
443 err = qdisc_class_hash_init(&q->clhash);
446 INIT_LIST_HEAD(&q->active);
450 static void drr_reset_qdisc(struct Qdisc *sch)
452 struct drr_sched *q = qdisc_priv(sch);
453 struct drr_class *cl;
454 struct hlist_node *n;
457 for (i = 0; i < q->clhash.hashsize; i++) {
458 hlist_for_each_entry(cl, n, &q->clhash.hash[i], common.hnode) {
459 if (cl->qdisc->q.qlen)
460 list_del(&cl->alist);
461 qdisc_reset(cl->qdisc);
467 static void drr_destroy_qdisc(struct Qdisc *sch)
469 struct drr_sched *q = qdisc_priv(sch);
470 struct drr_class *cl;
471 struct hlist_node *n, *next;
474 tcf_destroy_chain(&q->filter_list);
476 for (i = 0; i < q->clhash.hashsize; i++) {
477 hlist_for_each_entry_safe(cl, n, next, &q->clhash.hash[i],
479 drr_destroy_class(sch, cl);
481 qdisc_class_hash_destroy(&q->clhash);
484 static const struct Qdisc_class_ops drr_class_ops = {
485 .change = drr_change_class,
486 .delete = drr_delete_class,
487 .get = drr_get_class,
488 .put = drr_put_class,
489 .tcf_chain = drr_tcf_chain,
490 .bind_tcf = drr_bind_tcf,
491 .unbind_tcf = drr_unbind_tcf,
492 .graft = drr_graft_class,
493 .leaf = drr_class_leaf,
494 .qlen_notify = drr_qlen_notify,
495 .dump = drr_dump_class,
496 .dump_stats = drr_dump_class_stats,
500 static struct Qdisc_ops drr_qdisc_ops __read_mostly = {
501 .cl_ops = &drr_class_ops,
503 .priv_size = sizeof(struct drr_sched),
504 .enqueue = drr_enqueue,
505 .dequeue = drr_dequeue,
506 .peek = qdisc_peek_dequeued,
508 .init = drr_init_qdisc,
509 .reset = drr_reset_qdisc,
510 .destroy = drr_destroy_qdisc,
511 .owner = THIS_MODULE,
514 static int __init drr_init(void)
516 return register_qdisc(&drr_qdisc_ops);
519 static void __exit drr_exit(void)
521 unregister_qdisc(&drr_qdisc_ops);
524 module_init(drr_init);
525 module_exit(drr_exit);
526 MODULE_LICENSE("GPL");