1/*
2 * net/sched/sch_mq.c		Classful multiqueue dummy scheduler
3 *
4 * Copyright (c) 2009 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/types.h>
12#include <linux/slab.h>
13#include <linux/kernel.h>
14#include <linux/export.h>
15#include <linux/string.h>
16#include <linux/errno.h>
17#include <linux/skbuff.h>
18#include <net/netlink.h>
19#include <net/pkt_sched.h>
20
21struct mq_sched {
22	struct Qdisc		**qdiscs;
23};
24
25static void mq_destroy(struct Qdisc *sch)
26{
27	struct net_device *dev = qdisc_dev(sch);
28	struct mq_sched *priv = qdisc_priv(sch);
29	unsigned int ntx;
30
31	if (!priv->qdiscs)
32		return;
33	for (ntx = 0; ntx < dev->num_tx_queues && priv->qdiscs[ntx]; ntx++)
34		qdisc_destroy(priv->qdiscs[ntx]);
35	kfree(priv->qdiscs);
36}
37
38static int mq_init(struct Qdisc *sch, struct nlattr *opt)
39{
40	struct net_device *dev = qdisc_dev(sch);
41	struct mq_sched *priv = qdisc_priv(sch);
42	struct netdev_queue *dev_queue;
43	struct Qdisc *qdisc;
44	unsigned int ntx;
45
46	if (sch->parent != TC_H_ROOT)
47		return -EOPNOTSUPP;
48
49	if (!netif_is_multiqueue(dev))
50		return -EOPNOTSUPP;
51
52	/* pre-allocate qdiscs, attachment can't fail */
53	priv->qdiscs = kcalloc(dev->num_tx_queues, sizeof(priv->qdiscs[0]),
54			       GFP_KERNEL);
55	if (priv->qdiscs == NULL)
56		return -ENOMEM;
57
58	for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
59		dev_queue = netdev_get_tx_queue(dev, ntx);
60		qdisc = qdisc_create_dflt(dev_queue, default_qdisc_ops,
61					  TC_H_MAKE(TC_H_MAJ(sch->handle),
62						    TC_H_MIN(ntx + 1)));
63		if (qdisc == NULL)
64			goto err;
65		priv->qdiscs[ntx] = qdisc;
66		qdisc->flags |= TCQ_F_ONETXQUEUE;
67	}
68
69	sch->flags |= TCQ_F_MQROOT;
70	return 0;
71
72err:
73	mq_destroy(sch);
74	return -ENOMEM;
75}
76
77static void mq_attach(struct Qdisc *sch)
78{
79	struct net_device *dev = qdisc_dev(sch);
80	struct mq_sched *priv = qdisc_priv(sch);
81	struct Qdisc *qdisc, *old;
82	unsigned int ntx;
83
84	for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
85		qdisc = priv->qdiscs[ntx];
86		old = dev_graft_qdisc(qdisc->dev_queue, qdisc);
87		if (old)
88			qdisc_destroy(old);
89#ifdef CONFIG_NET_SCHED
90		if (ntx < dev->real_num_tx_queues)
91			qdisc_list_add(qdisc);
92#endif
93
94	}
95	kfree(priv->qdiscs);
96	priv->qdiscs = NULL;
97}
98
99static int mq_dump(struct Qdisc *sch, struct sk_buff *skb)
100{
101	struct net_device *dev = qdisc_dev(sch);
102	struct Qdisc *qdisc;
103	unsigned int ntx;
104
105	sch->q.qlen = 0;
106	memset(&sch->bstats, 0, sizeof(sch->bstats));
107	memset(&sch->qstats, 0, sizeof(sch->qstats));
108
109	for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
110		qdisc = netdev_get_tx_queue(dev, ntx)->qdisc_sleeping;
111		spin_lock_bh(qdisc_lock(qdisc));
112		sch->q.qlen		+= qdisc->q.qlen;
113		sch->bstats.bytes	+= qdisc->bstats.bytes;
114		sch->bstats.packets	+= qdisc->bstats.packets;
115		sch->qstats.backlog	+= qdisc->qstats.backlog;
116		sch->qstats.drops	+= qdisc->qstats.drops;
117		sch->qstats.requeues	+= qdisc->qstats.requeues;
118		sch->qstats.overlimits	+= qdisc->qstats.overlimits;
119		spin_unlock_bh(qdisc_lock(qdisc));
120	}
121	return 0;
122}
123
124static struct netdev_queue *mq_queue_get(struct Qdisc *sch, unsigned long cl)
125{
126	struct net_device *dev = qdisc_dev(sch);
127	unsigned long ntx = cl - 1;
128
129	if (ntx >= dev->num_tx_queues)
130		return NULL;
131	return netdev_get_tx_queue(dev, ntx);
132}
133
134static struct netdev_queue *mq_select_queue(struct Qdisc *sch,
135					    struct tcmsg *tcm)
136{
137	unsigned int ntx = TC_H_MIN(tcm->tcm_parent);
138	struct netdev_queue *dev_queue = mq_queue_get(sch, ntx);
139
140	if (!dev_queue) {
141		struct net_device *dev = qdisc_dev(sch);
142
143		return netdev_get_tx_queue(dev, 0);
144	}
145	return dev_queue;
146}
147
148static int mq_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
149		    struct Qdisc **old)
150{
151	struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
152	struct net_device *dev = qdisc_dev(sch);
153
154	if (dev->flags & IFF_UP)
155		dev_deactivate(dev);
156
157	*old = dev_graft_qdisc(dev_queue, new);
158	if (new)
159		new->flags |= TCQ_F_ONETXQUEUE;
160	if (dev->flags & IFF_UP)
161		dev_activate(dev);
162	return 0;
163}
164
165static struct Qdisc *mq_leaf(struct Qdisc *sch, unsigned long cl)
166{
167	struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
168
169	return dev_queue->qdisc_sleeping;
170}
171
172static unsigned long mq_get(struct Qdisc *sch, u32 classid)
173{
174	unsigned int ntx = TC_H_MIN(classid);
175
176	if (!mq_queue_get(sch, ntx))
177		return 0;
178	return ntx;
179}
180
181static void mq_put(struct Qdisc *sch, unsigned long cl)
182{
183}
184
185static int mq_dump_class(struct Qdisc *sch, unsigned long cl,
186			 struct sk_buff *skb, struct tcmsg *tcm)
187{
188	struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
189
190	tcm->tcm_parent = TC_H_ROOT;
191	tcm->tcm_handle |= TC_H_MIN(cl);
192	tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
193	return 0;
194}
195
196static int mq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
197			       struct gnet_dump *d)
198{
199	struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
200
201	sch = dev_queue->qdisc_sleeping;
202	if (gnet_stats_copy_basic(d, NULL, &sch->bstats) < 0 ||
203	    gnet_stats_copy_queue(d, NULL, &sch->qstats, sch->q.qlen) < 0)
204		return -1;
205	return 0;
206}
207
208static void mq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
209{
210	struct net_device *dev = qdisc_dev(sch);
211	unsigned int ntx;
212
213	if (arg->stop)
214		return;
215
216	arg->count = arg->skip;
217	for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
218		if (arg->fn(sch, ntx + 1, arg) < 0) {
219			arg->stop = 1;
220			break;
221		}
222		arg->count++;
223	}
224}
225
226static const struct Qdisc_class_ops mq_class_ops = {
227	.select_queue	= mq_select_queue,
228	.graft		= mq_graft,
229	.leaf		= mq_leaf,
230	.get		= mq_get,
231	.put		= mq_put,
232	.walk		= mq_walk,
233	.dump		= mq_dump_class,
234	.dump_stats	= mq_dump_class_stats,
235};
236
237struct Qdisc_ops mq_qdisc_ops __read_mostly = {
238	.cl_ops		= &mq_class_ops,
239	.id		= "mq",
240	.priv_size	= sizeof(struct mq_sched),
241	.init		= mq_init,
242	.destroy	= mq_destroy,
243	.attach		= mq_attach,
244	.dump		= mq_dump,
245	.owner		= THIS_MODULE,
246};
247