1/*
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
14#include <linux/module.h>
15#include <linux/slab.h>
16#include <linux/types.h>
17#include <linux/kernel.h>
18#include <linux/string.h>
19#include <linux/errno.h>
20#include <linux/skbuff.h>
21#include <net/netlink.h>
22#include <net/pkt_sched.h>
23
24
25struct prio_sched_data {
26	int bands;
27	struct tcf_proto __rcu *filter_list;
28	u8  prio2band[TC_PRIO_MAX+1];
29	struct Qdisc *queues[TCQ_PRIO_BANDS];
30};
31
32
33static struct Qdisc *
34prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
35{
36	struct prio_sched_data *q = qdisc_priv(sch);
37	u32 band = skb->priority;
38	struct tcf_result res;
39	struct tcf_proto *fl;
40	int err;
41
42	*qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
43	if (TC_H_MAJ(skb->priority) != sch->handle) {
44		fl = rcu_dereference_bh(q->filter_list);
45		err = tc_classify(skb, fl, &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 | __NET_XMIT_STOLEN;
51		case TC_ACT_SHOT:
52			return NULL;
53		}
54#endif
55		if (!fl || err < 0) {
56			if (TC_H_MAJ(band))
57				band = 0;
58			return q->queues[q->prio2band[band & TC_PRIO_MAX]];
59		}
60		band = res.classid;
61	}
62	band = TC_H_MIN(band) - 1;
63	if (band >= q->bands)
64		return q->queues[q->prio2band[0]];
65
66	return q->queues[band];
67}
68
69static int
70prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
71{
72	struct Qdisc *qdisc;
73	int ret;
74
75	qdisc = prio_classify(skb, sch, &ret);
76#ifdef CONFIG_NET_CLS_ACT
77	if (qdisc == NULL) {
78
79		if (ret & __NET_XMIT_BYPASS)
80			qdisc_qstats_drop(sch);
81		kfree_skb(skb);
82		return ret;
83	}
84#endif
85
86	ret = qdisc_enqueue(skb, qdisc);
87	if (ret == NET_XMIT_SUCCESS) {
88		sch->q.qlen++;
89		return NET_XMIT_SUCCESS;
90	}
91	if (net_xmit_drop_count(ret))
92		qdisc_qstats_drop(sch);
93	return ret;
94}
95
96static struct sk_buff *prio_peek(struct Qdisc *sch)
97{
98	struct prio_sched_data *q = qdisc_priv(sch);
99	int prio;
100
101	for (prio = 0; prio < q->bands; prio++) {
102		struct Qdisc *qdisc = q->queues[prio];
103		struct sk_buff *skb = qdisc->ops->peek(qdisc);
104		if (skb)
105			return skb;
106	}
107	return NULL;
108}
109
110static struct sk_buff *prio_dequeue(struct Qdisc *sch)
111{
112	struct prio_sched_data *q = qdisc_priv(sch);
113	int prio;
114
115	for (prio = 0; prio < q->bands; prio++) {
116		struct Qdisc *qdisc = q->queues[prio];
117		struct sk_buff *skb = qdisc_dequeue_peeked(qdisc);
118		if (skb) {
119			qdisc_bstats_update(sch, skb);
120			sch->q.qlen--;
121			return skb;
122		}
123	}
124	return NULL;
125
126}
127
128static unsigned int prio_drop(struct Qdisc *sch)
129{
130	struct prio_sched_data *q = qdisc_priv(sch);
131	int prio;
132	unsigned int len;
133	struct Qdisc *qdisc;
134
135	for (prio = q->bands-1; prio >= 0; prio--) {
136		qdisc = q->queues[prio];
137		if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
138			sch->q.qlen--;
139			return len;
140		}
141	}
142	return 0;
143}
144
145
146static void
147prio_reset(struct Qdisc *sch)
148{
149	int prio;
150	struct prio_sched_data *q = qdisc_priv(sch);
151
152	for (prio = 0; prio < q->bands; prio++)
153		qdisc_reset(q->queues[prio]);
154	sch->q.qlen = 0;
155}
156
157static void
158prio_destroy(struct Qdisc *sch)
159{
160	int prio;
161	struct prio_sched_data *q = qdisc_priv(sch);
162
163	tcf_destroy_chain(&q->filter_list);
164	for (prio = 0; prio < q->bands; prio++)
165		qdisc_destroy(q->queues[prio]);
166}
167
168static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
169{
170	struct prio_sched_data *q = qdisc_priv(sch);
171	struct tc_prio_qopt *qopt;
172	int i;
173
174	if (nla_len(opt) < sizeof(*qopt))
175		return -EINVAL;
176	qopt = nla_data(opt);
177
178	if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
179		return -EINVAL;
180
181	for (i = 0; i <= TC_PRIO_MAX; i++) {
182		if (qopt->priomap[i] >= qopt->bands)
183			return -EINVAL;
184	}
185
186	sch_tree_lock(sch);
187	q->bands = qopt->bands;
188	memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
189
190	for (i = q->bands; i < TCQ_PRIO_BANDS; i++) {
191		struct Qdisc *child = q->queues[i];
192		q->queues[i] = &noop_qdisc;
193		if (child != &noop_qdisc) {
194			qdisc_tree_decrease_qlen(child, child->q.qlen);
195			qdisc_destroy(child);
196		}
197	}
198	sch_tree_unlock(sch);
199
200	for (i = 0; i < q->bands; i++) {
201		if (q->queues[i] == &noop_qdisc) {
202			struct Qdisc *child, *old;
203
204			child = qdisc_create_dflt(sch->dev_queue,
205						  &pfifo_qdisc_ops,
206						  TC_H_MAKE(sch->handle, i + 1));
207			if (child) {
208				sch_tree_lock(sch);
209				old = q->queues[i];
210				q->queues[i] = child;
211
212				if (old != &noop_qdisc) {
213					qdisc_tree_decrease_qlen(old,
214								 old->q.qlen);
215					qdisc_destroy(old);
216				}
217				sch_tree_unlock(sch);
218			}
219		}
220	}
221	return 0;
222}
223
224static int prio_init(struct Qdisc *sch, struct nlattr *opt)
225{
226	struct prio_sched_data *q = qdisc_priv(sch);
227	int i;
228
229	for (i = 0; i < TCQ_PRIO_BANDS; i++)
230		q->queues[i] = &noop_qdisc;
231
232	if (opt == NULL) {
233		return -EINVAL;
234	} else {
235		int err;
236
237		if ((err = prio_tune(sch, opt)) != 0)
238			return err;
239	}
240	return 0;
241}
242
243static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
244{
245	struct prio_sched_data *q = qdisc_priv(sch);
246	unsigned char *b = skb_tail_pointer(skb);
247	struct tc_prio_qopt opt;
248
249	opt.bands = q->bands;
250	memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX + 1);
251
252	if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
253		goto nla_put_failure;
254
255	return skb->len;
256
257nla_put_failure:
258	nlmsg_trim(skb, b);
259	return -1;
260}
261
262static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
263		      struct Qdisc **old)
264{
265	struct prio_sched_data *q = qdisc_priv(sch);
266	unsigned long band = arg - 1;
267
268	if (new == NULL)
269		new = &noop_qdisc;
270
271	sch_tree_lock(sch);
272	*old = q->queues[band];
273	q->queues[band] = new;
274	qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
275	qdisc_reset(*old);
276	sch_tree_unlock(sch);
277
278	return 0;
279}
280
281static struct Qdisc *
282prio_leaf(struct Qdisc *sch, unsigned long arg)
283{
284	struct prio_sched_data *q = qdisc_priv(sch);
285	unsigned long band = arg - 1;
286
287	return q->queues[band];
288}
289
290static unsigned long prio_get(struct Qdisc *sch, u32 classid)
291{
292	struct prio_sched_data *q = qdisc_priv(sch);
293	unsigned long band = TC_H_MIN(classid);
294
295	if (band - 1 >= q->bands)
296		return 0;
297	return band;
298}
299
300static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
301{
302	return prio_get(sch, classid);
303}
304
305
306static void prio_put(struct Qdisc *q, unsigned long cl)
307{
308}
309
310static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
311			   struct tcmsg *tcm)
312{
313	struct prio_sched_data *q = qdisc_priv(sch);
314
315	tcm->tcm_handle |= TC_H_MIN(cl);
316	tcm->tcm_info = q->queues[cl-1]->handle;
317	return 0;
318}
319
320static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
321				 struct gnet_dump *d)
322{
323	struct prio_sched_data *q = qdisc_priv(sch);
324	struct Qdisc *cl_q;
325
326	cl_q = q->queues[cl - 1];
327	if (gnet_stats_copy_basic(d, NULL, &cl_q->bstats) < 0 ||
328	    gnet_stats_copy_queue(d, NULL, &cl_q->qstats, cl_q->q.qlen) < 0)
329		return -1;
330
331	return 0;
332}
333
334static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
335{
336	struct prio_sched_data *q = qdisc_priv(sch);
337	int prio;
338
339	if (arg->stop)
340		return;
341
342	for (prio = 0; prio < q->bands; prio++) {
343		if (arg->count < arg->skip) {
344			arg->count++;
345			continue;
346		}
347		if (arg->fn(sch, prio + 1, arg) < 0) {
348			arg->stop = 1;
349			break;
350		}
351		arg->count++;
352	}
353}
354
355static struct tcf_proto __rcu **prio_find_tcf(struct Qdisc *sch,
356					      unsigned long cl)
357{
358	struct prio_sched_data *q = qdisc_priv(sch);
359
360	if (cl)
361		return NULL;
362	return &q->filter_list;
363}
364
365static const struct Qdisc_class_ops prio_class_ops = {
366	.graft		=	prio_graft,
367	.leaf		=	prio_leaf,
368	.get		=	prio_get,
369	.put		=	prio_put,
370	.walk		=	prio_walk,
371	.tcf_chain	=	prio_find_tcf,
372	.bind_tcf	=	prio_bind,
373	.unbind_tcf	=	prio_put,
374	.dump		=	prio_dump_class,
375	.dump_stats	=	prio_dump_class_stats,
376};
377
378static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
379	.next		=	NULL,
380	.cl_ops		=	&prio_class_ops,
381	.id		=	"prio",
382	.priv_size	=	sizeof(struct prio_sched_data),
383	.enqueue	=	prio_enqueue,
384	.dequeue	=	prio_dequeue,
385	.peek		=	prio_peek,
386	.drop		=	prio_drop,
387	.init		=	prio_init,
388	.reset		=	prio_reset,
389	.destroy	=	prio_destroy,
390	.change		=	prio_tune,
391	.dump		=	prio_dump,
392	.owner		=	THIS_MODULE,
393};
394
395static int __init prio_module_init(void)
396{
397	return register_qdisc(&prio_qdisc_ops);
398}
399
400static void __exit prio_module_exit(void)
401{
402	unregister_qdisc(&prio_qdisc_ops);
403}
404
405module_init(prio_module_init)
406module_exit(prio_module_exit)
407
408MODULE_LICENSE("GPL");
409