1/*
2 * net/sched/sch_drr.c         Deficit Round Robin scheduler
3 *
4 * Copyright (c) 2008 Patrick McHardy <kaber@trash.net>
5 *
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.
9 */
10
11#include <linux/module.h>
12#include <linux/slab.h>
13#include <linux/init.h>
14#include <linux/errno.h>
15#include <linux/netdevice.h>
16#include <linux/pkt_sched.h>
17#include <net/sch_generic.h>
18#include <net/pkt_sched.h>
19#include <net/pkt_cls.h>
20
21struct drr_class {
22	struct Qdisc_class_common	common;
23	unsigned int			refcnt;
24	unsigned int			filter_cnt;
25
26	struct gnet_stats_basic_packed		bstats;
27	struct gnet_stats_queue		qstats;
28	struct gnet_stats_rate_est64	rate_est;
29	struct list_head		alist;
30	struct Qdisc			*qdisc;
31
32	u32				quantum;
33	u32				deficit;
34};
35
36struct drr_sched {
37	struct list_head		active;
38	struct tcf_proto __rcu		*filter_list;
39	struct Qdisc_class_hash		clhash;
40};
41
42static struct drr_class *drr_find_class(struct Qdisc *sch, u32 classid)
43{
44	struct drr_sched *q = qdisc_priv(sch);
45	struct Qdisc_class_common *clc;
46
47	clc = qdisc_class_find(&q->clhash, classid);
48	if (clc == NULL)
49		return NULL;
50	return container_of(clc, struct drr_class, common);
51}
52
53static void drr_purge_queue(struct drr_class *cl)
54{
55	unsigned int len = cl->qdisc->q.qlen;
56
57	qdisc_reset(cl->qdisc);
58	qdisc_tree_decrease_qlen(cl->qdisc, len);
59}
60
61static const struct nla_policy drr_policy[TCA_DRR_MAX + 1] = {
62	[TCA_DRR_QUANTUM]	= { .type = NLA_U32 },
63};
64
65static int drr_change_class(struct Qdisc *sch, u32 classid, u32 parentid,
66			    struct nlattr **tca, unsigned long *arg)
67{
68	struct drr_sched *q = qdisc_priv(sch);
69	struct drr_class *cl = (struct drr_class *)*arg;
70	struct nlattr *opt = tca[TCA_OPTIONS];
71	struct nlattr *tb[TCA_DRR_MAX + 1];
72	u32 quantum;
73	int err;
74
75	if (!opt)
76		return -EINVAL;
77
78	err = nla_parse_nested(tb, TCA_DRR_MAX, opt, drr_policy);
79	if (err < 0)
80		return err;
81
82	if (tb[TCA_DRR_QUANTUM]) {
83		quantum = nla_get_u32(tb[TCA_DRR_QUANTUM]);
84		if (quantum == 0)
85			return -EINVAL;
86	} else
87		quantum = psched_mtu(qdisc_dev(sch));
88
89	if (cl != NULL) {
90		if (tca[TCA_RATE]) {
91			err = gen_replace_estimator(&cl->bstats, NULL,
92						    &cl->rate_est,
93						    qdisc_root_sleeping_lock(sch),
94						    tca[TCA_RATE]);
95			if (err)
96				return err;
97		}
98
99		sch_tree_lock(sch);
100		if (tb[TCA_DRR_QUANTUM])
101			cl->quantum = quantum;
102		sch_tree_unlock(sch);
103
104		return 0;
105	}
106
107	cl = kzalloc(sizeof(struct drr_class), GFP_KERNEL);
108	if (cl == NULL)
109		return -ENOBUFS;
110
111	cl->refcnt	   = 1;
112	cl->common.classid = classid;
113	cl->quantum	   = quantum;
114	cl->qdisc	   = qdisc_create_dflt(sch->dev_queue,
115					       &pfifo_qdisc_ops, classid);
116	if (cl->qdisc == NULL)
117		cl->qdisc = &noop_qdisc;
118
119	if (tca[TCA_RATE]) {
120		err = gen_replace_estimator(&cl->bstats, NULL, &cl->rate_est,
121					    qdisc_root_sleeping_lock(sch),
122					    tca[TCA_RATE]);
123		if (err) {
124			qdisc_destroy(cl->qdisc);
125			kfree(cl);
126			return err;
127		}
128	}
129
130	sch_tree_lock(sch);
131	qdisc_class_hash_insert(&q->clhash, &cl->common);
132	sch_tree_unlock(sch);
133
134	qdisc_class_hash_grow(sch, &q->clhash);
135
136	*arg = (unsigned long)cl;
137	return 0;
138}
139
140static void drr_destroy_class(struct Qdisc *sch, struct drr_class *cl)
141{
142	gen_kill_estimator(&cl->bstats, &cl->rate_est);
143	qdisc_destroy(cl->qdisc);
144	kfree(cl);
145}
146
147static int drr_delete_class(struct Qdisc *sch, unsigned long arg)
148{
149	struct drr_sched *q = qdisc_priv(sch);
150	struct drr_class *cl = (struct drr_class *)arg;
151
152	if (cl->filter_cnt > 0)
153		return -EBUSY;
154
155	sch_tree_lock(sch);
156
157	drr_purge_queue(cl);
158	qdisc_class_hash_remove(&q->clhash, &cl->common);
159
160	BUG_ON(--cl->refcnt == 0);
161	/*
162	 * This shouldn't happen: we "hold" one cops->get() when called
163	 * from tc_ctl_tclass; the destroy method is done from cops->put().
164	 */
165
166	sch_tree_unlock(sch);
167	return 0;
168}
169
170static unsigned long drr_get_class(struct Qdisc *sch, u32 classid)
171{
172	struct drr_class *cl = drr_find_class(sch, classid);
173
174	if (cl != NULL)
175		cl->refcnt++;
176
177	return (unsigned long)cl;
178}
179
180static void drr_put_class(struct Qdisc *sch, unsigned long arg)
181{
182	struct drr_class *cl = (struct drr_class *)arg;
183
184	if (--cl->refcnt == 0)
185		drr_destroy_class(sch, cl);
186}
187
188static struct tcf_proto __rcu **drr_tcf_chain(struct Qdisc *sch,
189					      unsigned long cl)
190{
191	struct drr_sched *q = qdisc_priv(sch);
192
193	if (cl)
194		return NULL;
195
196	return &q->filter_list;
197}
198
199static unsigned long drr_bind_tcf(struct Qdisc *sch, unsigned long parent,
200				  u32 classid)
201{
202	struct drr_class *cl = drr_find_class(sch, classid);
203
204	if (cl != NULL)
205		cl->filter_cnt++;
206
207	return (unsigned long)cl;
208}
209
210static void drr_unbind_tcf(struct Qdisc *sch, unsigned long arg)
211{
212	struct drr_class *cl = (struct drr_class *)arg;
213
214	cl->filter_cnt--;
215}
216
217static int drr_graft_class(struct Qdisc *sch, unsigned long arg,
218			   struct Qdisc *new, struct Qdisc **old)
219{
220	struct drr_class *cl = (struct drr_class *)arg;
221
222	if (new == NULL) {
223		new = qdisc_create_dflt(sch->dev_queue,
224					&pfifo_qdisc_ops, cl->common.classid);
225		if (new == NULL)
226			new = &noop_qdisc;
227	}
228
229	sch_tree_lock(sch);
230	drr_purge_queue(cl);
231	*old = cl->qdisc;
232	cl->qdisc = new;
233	sch_tree_unlock(sch);
234	return 0;
235}
236
237static struct Qdisc *drr_class_leaf(struct Qdisc *sch, unsigned long arg)
238{
239	struct drr_class *cl = (struct drr_class *)arg;
240
241	return cl->qdisc;
242}
243
244static void drr_qlen_notify(struct Qdisc *csh, unsigned long arg)
245{
246	struct drr_class *cl = (struct drr_class *)arg;
247
248	if (cl->qdisc->q.qlen == 0)
249		list_del(&cl->alist);
250}
251
252static int drr_dump_class(struct Qdisc *sch, unsigned long arg,
253			  struct sk_buff *skb, struct tcmsg *tcm)
254{
255	struct drr_class *cl = (struct drr_class *)arg;
256	struct nlattr *nest;
257
258	tcm->tcm_parent	= TC_H_ROOT;
259	tcm->tcm_handle	= cl->common.classid;
260	tcm->tcm_info	= cl->qdisc->handle;
261
262	nest = nla_nest_start(skb, TCA_OPTIONS);
263	if (nest == NULL)
264		goto nla_put_failure;
265	if (nla_put_u32(skb, TCA_DRR_QUANTUM, cl->quantum))
266		goto nla_put_failure;
267	return nla_nest_end(skb, nest);
268
269nla_put_failure:
270	nla_nest_cancel(skb, nest);
271	return -EMSGSIZE;
272}
273
274static int drr_dump_class_stats(struct Qdisc *sch, unsigned long arg,
275				struct gnet_dump *d)
276{
277	struct drr_class *cl = (struct drr_class *)arg;
278	__u32 qlen = cl->qdisc->q.qlen;
279	struct tc_drr_stats xstats;
280
281	memset(&xstats, 0, sizeof(xstats));
282	if (qlen)
283		xstats.deficit = cl->deficit;
284
285	if (gnet_stats_copy_basic(d, NULL, &cl->bstats) < 0 ||
286	    gnet_stats_copy_rate_est(d, &cl->bstats, &cl->rate_est) < 0 ||
287	    gnet_stats_copy_queue(d, NULL, &cl->qdisc->qstats, qlen) < 0)
288		return -1;
289
290	return gnet_stats_copy_app(d, &xstats, sizeof(xstats));
291}
292
293static void drr_walk(struct Qdisc *sch, struct qdisc_walker *arg)
294{
295	struct drr_sched *q = qdisc_priv(sch);
296	struct drr_class *cl;
297	unsigned int i;
298
299	if (arg->stop)
300		return;
301
302	for (i = 0; i < q->clhash.hashsize; i++) {
303		hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
304			if (arg->count < arg->skip) {
305				arg->count++;
306				continue;
307			}
308			if (arg->fn(sch, (unsigned long)cl, arg) < 0) {
309				arg->stop = 1;
310				return;
311			}
312			arg->count++;
313		}
314	}
315}
316
317static struct drr_class *drr_classify(struct sk_buff *skb, struct Qdisc *sch,
318				      int *qerr)
319{
320	struct drr_sched *q = qdisc_priv(sch);
321	struct drr_class *cl;
322	struct tcf_result res;
323	struct tcf_proto *fl;
324	int result;
325
326	if (TC_H_MAJ(skb->priority ^ sch->handle) == 0) {
327		cl = drr_find_class(sch, skb->priority);
328		if (cl != NULL)
329			return cl;
330	}
331
332	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
333	fl = rcu_dereference_bh(q->filter_list);
334	result = tc_classify(skb, fl, &res);
335	if (result >= 0) {
336#ifdef CONFIG_NET_CLS_ACT
337		switch (result) {
338		case TC_ACT_QUEUED:
339		case TC_ACT_STOLEN:
340			*qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
341		case TC_ACT_SHOT:
342			return NULL;
343		}
344#endif
345		cl = (struct drr_class *)res.class;
346		if (cl == NULL)
347			cl = drr_find_class(sch, res.classid);
348		return cl;
349	}
350	return NULL;
351}
352
353static int drr_enqueue(struct sk_buff *skb, struct Qdisc *sch)
354{
355	struct drr_sched *q = qdisc_priv(sch);
356	struct drr_class *cl;
357	int err = 0;
358
359	cl = drr_classify(skb, sch, &err);
360	if (cl == NULL) {
361		if (err & __NET_XMIT_BYPASS)
362			qdisc_qstats_drop(sch);
363		kfree_skb(skb);
364		return err;
365	}
366
367	err = qdisc_enqueue(skb, cl->qdisc);
368	if (unlikely(err != NET_XMIT_SUCCESS)) {
369		if (net_xmit_drop_count(err)) {
370			cl->qstats.drops++;
371			qdisc_qstats_drop(sch);
372		}
373		return err;
374	}
375
376	if (cl->qdisc->q.qlen == 1) {
377		list_add_tail(&cl->alist, &q->active);
378		cl->deficit = cl->quantum;
379	}
380
381	sch->q.qlen++;
382	return err;
383}
384
385static struct sk_buff *drr_dequeue(struct Qdisc *sch)
386{
387	struct drr_sched *q = qdisc_priv(sch);
388	struct drr_class *cl;
389	struct sk_buff *skb;
390	unsigned int len;
391
392	if (list_empty(&q->active))
393		goto out;
394	while (1) {
395		cl = list_first_entry(&q->active, struct drr_class, alist);
396		skb = cl->qdisc->ops->peek(cl->qdisc);
397		if (skb == NULL) {
398			qdisc_warn_nonwc(__func__, cl->qdisc);
399			goto out;
400		}
401
402		len = qdisc_pkt_len(skb);
403		if (len <= cl->deficit) {
404			cl->deficit -= len;
405			skb = qdisc_dequeue_peeked(cl->qdisc);
406			if (cl->qdisc->q.qlen == 0)
407				list_del(&cl->alist);
408
409			bstats_update(&cl->bstats, skb);
410			qdisc_bstats_update(sch, skb);
411			sch->q.qlen--;
412			return skb;
413		}
414
415		cl->deficit += cl->quantum;
416		list_move_tail(&cl->alist, &q->active);
417	}
418out:
419	return NULL;
420}
421
422static unsigned int drr_drop(struct Qdisc *sch)
423{
424	struct drr_sched *q = qdisc_priv(sch);
425	struct drr_class *cl;
426	unsigned int len;
427
428	list_for_each_entry(cl, &q->active, alist) {
429		if (cl->qdisc->ops->drop) {
430			len = cl->qdisc->ops->drop(cl->qdisc);
431			if (len > 0) {
432				sch->q.qlen--;
433				if (cl->qdisc->q.qlen == 0)
434					list_del(&cl->alist);
435				return len;
436			}
437		}
438	}
439	return 0;
440}
441
442static int drr_init_qdisc(struct Qdisc *sch, struct nlattr *opt)
443{
444	struct drr_sched *q = qdisc_priv(sch);
445	int err;
446
447	err = qdisc_class_hash_init(&q->clhash);
448	if (err < 0)
449		return err;
450	INIT_LIST_HEAD(&q->active);
451	return 0;
452}
453
454static void drr_reset_qdisc(struct Qdisc *sch)
455{
456	struct drr_sched *q = qdisc_priv(sch);
457	struct drr_class *cl;
458	unsigned int i;
459
460	for (i = 0; i < q->clhash.hashsize; i++) {
461		hlist_for_each_entry(cl, &q->clhash.hash[i], common.hnode) {
462			if (cl->qdisc->q.qlen)
463				list_del(&cl->alist);
464			qdisc_reset(cl->qdisc);
465		}
466	}
467	sch->q.qlen = 0;
468}
469
470static void drr_destroy_qdisc(struct Qdisc *sch)
471{
472	struct drr_sched *q = qdisc_priv(sch);
473	struct drr_class *cl;
474	struct hlist_node *next;
475	unsigned int i;
476
477	tcf_destroy_chain(&q->filter_list);
478
479	for (i = 0; i < q->clhash.hashsize; i++) {
480		hlist_for_each_entry_safe(cl, next, &q->clhash.hash[i],
481					  common.hnode)
482			drr_destroy_class(sch, cl);
483	}
484	qdisc_class_hash_destroy(&q->clhash);
485}
486
487static const struct Qdisc_class_ops drr_class_ops = {
488	.change		= drr_change_class,
489	.delete		= drr_delete_class,
490	.get		= drr_get_class,
491	.put		= drr_put_class,
492	.tcf_chain	= drr_tcf_chain,
493	.bind_tcf	= drr_bind_tcf,
494	.unbind_tcf	= drr_unbind_tcf,
495	.graft		= drr_graft_class,
496	.leaf		= drr_class_leaf,
497	.qlen_notify	= drr_qlen_notify,
498	.dump		= drr_dump_class,
499	.dump_stats	= drr_dump_class_stats,
500	.walk		= drr_walk,
501};
502
503static struct Qdisc_ops drr_qdisc_ops __read_mostly = {
504	.cl_ops		= &drr_class_ops,
505	.id		= "drr",
506	.priv_size	= sizeof(struct drr_sched),
507	.enqueue	= drr_enqueue,
508	.dequeue	= drr_dequeue,
509	.peek		= qdisc_peek_dequeued,
510	.drop		= drr_drop,
511	.init		= drr_init_qdisc,
512	.reset		= drr_reset_qdisc,
513	.destroy	= drr_destroy_qdisc,
514	.owner		= THIS_MODULE,
515};
516
517static int __init drr_init(void)
518{
519	return register_qdisc(&drr_qdisc_ops);
520}
521
522static void __exit drr_exit(void)
523{
524	unregister_qdisc(&drr_qdisc_ops);
525}
526
527module_init(drr_init);
528module_exit(drr_exit);
529MODULE_LICENSE("GPL");
530