Loading...
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.qlen += qdisc->qstats.qlen;
116 sch->qstats.backlog += qdisc->qstats.backlog;
117 sch->qstats.drops += qdisc->qstats.drops;
118 sch->qstats.requeues += qdisc->qstats.requeues;
119 sch->qstats.overlimits += qdisc->qstats.overlimits;
120 spin_unlock_bh(qdisc_lock(qdisc));
121 }
122 return 0;
123}
124
125static struct netdev_queue *mq_queue_get(struct Qdisc *sch, unsigned long cl)
126{
127 struct net_device *dev = qdisc_dev(sch);
128 unsigned long ntx = cl - 1;
129
130 if (ntx >= dev->num_tx_queues)
131 return NULL;
132 return netdev_get_tx_queue(dev, ntx);
133}
134
135static struct netdev_queue *mq_select_queue(struct Qdisc *sch,
136 struct tcmsg *tcm)
137{
138 unsigned int ntx = TC_H_MIN(tcm->tcm_parent);
139 struct netdev_queue *dev_queue = mq_queue_get(sch, ntx);
140
141 if (!dev_queue) {
142 struct net_device *dev = qdisc_dev(sch);
143
144 return netdev_get_tx_queue(dev, 0);
145 }
146 return dev_queue;
147}
148
149static int mq_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
150 struct Qdisc **old)
151{
152 struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
153 struct net_device *dev = qdisc_dev(sch);
154
155 if (dev->flags & IFF_UP)
156 dev_deactivate(dev);
157
158 *old = dev_graft_qdisc(dev_queue, new);
159 if (new)
160 new->flags |= TCQ_F_ONETXQUEUE;
161 if (dev->flags & IFF_UP)
162 dev_activate(dev);
163 return 0;
164}
165
166static struct Qdisc *mq_leaf(struct Qdisc *sch, unsigned long cl)
167{
168 struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
169
170 return dev_queue->qdisc_sleeping;
171}
172
173static unsigned long mq_get(struct Qdisc *sch, u32 classid)
174{
175 unsigned int ntx = TC_H_MIN(classid);
176
177 if (!mq_queue_get(sch, ntx))
178 return 0;
179 return ntx;
180}
181
182static void mq_put(struct Qdisc *sch, unsigned long cl)
183{
184}
185
186static int mq_dump_class(struct Qdisc *sch, unsigned long cl,
187 struct sk_buff *skb, struct tcmsg *tcm)
188{
189 struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
190
191 tcm->tcm_parent = TC_H_ROOT;
192 tcm->tcm_handle |= TC_H_MIN(cl);
193 tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
194 return 0;
195}
196
197static int mq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
198 struct gnet_dump *d)
199{
200 struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
201
202 sch = dev_queue->qdisc_sleeping;
203 sch->qstats.qlen = sch->q.qlen;
204 if (gnet_stats_copy_basic(d, &sch->bstats) < 0 ||
205 gnet_stats_copy_queue(d, &sch->qstats) < 0)
206 return -1;
207 return 0;
208}
209
210static void mq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
211{
212 struct net_device *dev = qdisc_dev(sch);
213 unsigned int ntx;
214
215 if (arg->stop)
216 return;
217
218 arg->count = arg->skip;
219 for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
220 if (arg->fn(sch, ntx + 1, arg) < 0) {
221 arg->stop = 1;
222 break;
223 }
224 arg->count++;
225 }
226}
227
228static const struct Qdisc_class_ops mq_class_ops = {
229 .select_queue = mq_select_queue,
230 .graft = mq_graft,
231 .leaf = mq_leaf,
232 .get = mq_get,
233 .put = mq_put,
234 .walk = mq_walk,
235 .dump = mq_dump_class,
236 .dump_stats = mq_dump_class_stats,
237};
238
239struct Qdisc_ops mq_qdisc_ops __read_mostly = {
240 .cl_ops = &mq_class_ops,
241 .id = "mq",
242 .priv_size = sizeof(struct mq_sched),
243 .init = mq_init,
244 .destroy = mq_destroy,
245 .attach = mq_attach,
246 .dump = mq_dump,
247 .owner = THIS_MODULE,
248};
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * net/sched/sch_mq.c Classful multiqueue dummy scheduler
4 *
5 * Copyright (c) 2009 Patrick McHardy <kaber@trash.net>
6 */
7
8#include <linux/types.h>
9#include <linux/slab.h>
10#include <linux/kernel.h>
11#include <linux/export.h>
12#include <linux/string.h>
13#include <linux/errno.h>
14#include <linux/skbuff.h>
15#include <net/netlink.h>
16#include <net/pkt_cls.h>
17#include <net/pkt_sched.h>
18#include <net/sch_generic.h>
19
20struct mq_sched {
21 struct Qdisc **qdiscs;
22};
23
24static int mq_offload(struct Qdisc *sch, enum tc_mq_command cmd)
25{
26 struct net_device *dev = qdisc_dev(sch);
27 struct tc_mq_qopt_offload opt = {
28 .command = cmd,
29 .handle = sch->handle,
30 };
31
32 if (!tc_can_offload(dev) || !dev->netdev_ops->ndo_setup_tc)
33 return -EOPNOTSUPP;
34
35 return dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_QDISC_MQ, &opt);
36}
37
38static int mq_offload_stats(struct Qdisc *sch)
39{
40 struct tc_mq_qopt_offload opt = {
41 .command = TC_MQ_STATS,
42 .handle = sch->handle,
43 .stats = {
44 .bstats = &sch->bstats,
45 .qstats = &sch->qstats,
46 },
47 };
48
49 return qdisc_offload_dump_helper(sch, TC_SETUP_QDISC_MQ, &opt);
50}
51
52static void mq_destroy(struct Qdisc *sch)
53{
54 struct net_device *dev = qdisc_dev(sch);
55 struct mq_sched *priv = qdisc_priv(sch);
56 unsigned int ntx;
57
58 mq_offload(sch, TC_MQ_DESTROY);
59
60 if (!priv->qdiscs)
61 return;
62 for (ntx = 0; ntx < dev->num_tx_queues && priv->qdiscs[ntx]; ntx++)
63 qdisc_put(priv->qdiscs[ntx]);
64 kfree(priv->qdiscs);
65}
66
67static int mq_init(struct Qdisc *sch, struct nlattr *opt,
68 struct netlink_ext_ack *extack)
69{
70 struct net_device *dev = qdisc_dev(sch);
71 struct mq_sched *priv = qdisc_priv(sch);
72 struct netdev_queue *dev_queue;
73 struct Qdisc *qdisc;
74 unsigned int ntx;
75
76 if (sch->parent != TC_H_ROOT)
77 return -EOPNOTSUPP;
78
79 if (!netif_is_multiqueue(dev))
80 return -EOPNOTSUPP;
81
82 /* pre-allocate qdiscs, attachment can't fail */
83 priv->qdiscs = kcalloc(dev->num_tx_queues, sizeof(priv->qdiscs[0]),
84 GFP_KERNEL);
85 if (!priv->qdiscs)
86 return -ENOMEM;
87
88 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
89 dev_queue = netdev_get_tx_queue(dev, ntx);
90 qdisc = qdisc_create_dflt(dev_queue, get_default_qdisc_ops(dev, ntx),
91 TC_H_MAKE(TC_H_MAJ(sch->handle),
92 TC_H_MIN(ntx + 1)),
93 extack);
94 if (!qdisc)
95 return -ENOMEM;
96 priv->qdiscs[ntx] = qdisc;
97 qdisc->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
98 }
99
100 sch->flags |= TCQ_F_MQROOT;
101
102 mq_offload(sch, TC_MQ_CREATE);
103 return 0;
104}
105
106static void mq_attach(struct Qdisc *sch)
107{
108 struct net_device *dev = qdisc_dev(sch);
109 struct mq_sched *priv = qdisc_priv(sch);
110 struct Qdisc *qdisc, *old;
111 unsigned int ntx;
112
113 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
114 qdisc = priv->qdiscs[ntx];
115 old = dev_graft_qdisc(qdisc->dev_queue, qdisc);
116 if (old)
117 qdisc_put(old);
118#ifdef CONFIG_NET_SCHED
119 if (ntx < dev->real_num_tx_queues)
120 qdisc_hash_add(qdisc, false);
121#endif
122
123 }
124 kfree(priv->qdiscs);
125 priv->qdiscs = NULL;
126}
127
128static int mq_dump(struct Qdisc *sch, struct sk_buff *skb)
129{
130 struct net_device *dev = qdisc_dev(sch);
131 struct Qdisc *qdisc;
132 unsigned int ntx;
133 __u32 qlen = 0;
134
135 sch->q.qlen = 0;
136 memset(&sch->bstats, 0, sizeof(sch->bstats));
137 memset(&sch->qstats, 0, sizeof(sch->qstats));
138
139 /* MQ supports lockless qdiscs. However, statistics accounting needs
140 * to account for all, none, or a mix of locked and unlocked child
141 * qdiscs. Percpu stats are added to counters in-band and locking
142 * qdisc totals are added at end.
143 */
144 for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
145 qdisc = netdev_get_tx_queue(dev, ntx)->qdisc_sleeping;
146 spin_lock_bh(qdisc_lock(qdisc));
147
148 if (qdisc_is_percpu_stats(qdisc)) {
149 qlen = qdisc_qlen_sum(qdisc);
150 __gnet_stats_copy_basic(NULL, &sch->bstats,
151 qdisc->cpu_bstats,
152 &qdisc->bstats);
153 __gnet_stats_copy_queue(&sch->qstats,
154 qdisc->cpu_qstats,
155 &qdisc->qstats, qlen);
156 sch->q.qlen += qlen;
157 } else {
158 sch->q.qlen += qdisc->q.qlen;
159 sch->bstats.bytes += qdisc->bstats.bytes;
160 sch->bstats.packets += qdisc->bstats.packets;
161 sch->qstats.qlen += qdisc->qstats.qlen;
162 sch->qstats.backlog += qdisc->qstats.backlog;
163 sch->qstats.drops += qdisc->qstats.drops;
164 sch->qstats.requeues += qdisc->qstats.requeues;
165 sch->qstats.overlimits += qdisc->qstats.overlimits;
166 }
167
168 spin_unlock_bh(qdisc_lock(qdisc));
169 }
170
171 return mq_offload_stats(sch);
172}
173
174static struct netdev_queue *mq_queue_get(struct Qdisc *sch, unsigned long cl)
175{
176 struct net_device *dev = qdisc_dev(sch);
177 unsigned long ntx = cl - 1;
178
179 if (ntx >= dev->num_tx_queues)
180 return NULL;
181 return netdev_get_tx_queue(dev, ntx);
182}
183
184static struct netdev_queue *mq_select_queue(struct Qdisc *sch,
185 struct tcmsg *tcm)
186{
187 return mq_queue_get(sch, TC_H_MIN(tcm->tcm_parent));
188}
189
190static int mq_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
191 struct Qdisc **old, struct netlink_ext_ack *extack)
192{
193 struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
194 struct tc_mq_qopt_offload graft_offload;
195 struct net_device *dev = qdisc_dev(sch);
196
197 if (dev->flags & IFF_UP)
198 dev_deactivate(dev);
199
200 *old = dev_graft_qdisc(dev_queue, new);
201 if (new)
202 new->flags |= TCQ_F_ONETXQUEUE | TCQ_F_NOPARENT;
203 if (dev->flags & IFF_UP)
204 dev_activate(dev);
205
206 graft_offload.handle = sch->handle;
207 graft_offload.graft_params.queue = cl - 1;
208 graft_offload.graft_params.child_handle = new ? new->handle : 0;
209 graft_offload.command = TC_MQ_GRAFT;
210
211 qdisc_offload_graft_helper(qdisc_dev(sch), sch, new, *old,
212 TC_SETUP_QDISC_MQ, &graft_offload, extack);
213 return 0;
214}
215
216static struct Qdisc *mq_leaf(struct Qdisc *sch, unsigned long cl)
217{
218 struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
219
220 return dev_queue->qdisc_sleeping;
221}
222
223static unsigned long mq_find(struct Qdisc *sch, u32 classid)
224{
225 unsigned int ntx = TC_H_MIN(classid);
226
227 if (!mq_queue_get(sch, ntx))
228 return 0;
229 return ntx;
230}
231
232static int mq_dump_class(struct Qdisc *sch, unsigned long cl,
233 struct sk_buff *skb, struct tcmsg *tcm)
234{
235 struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
236
237 tcm->tcm_parent = TC_H_ROOT;
238 tcm->tcm_handle |= TC_H_MIN(cl);
239 tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
240 return 0;
241}
242
243static int mq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
244 struct gnet_dump *d)
245{
246 struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
247
248 sch = dev_queue->qdisc_sleeping;
249 if (gnet_stats_copy_basic(&sch->running, d, sch->cpu_bstats,
250 &sch->bstats) < 0 ||
251 qdisc_qstats_copy(d, sch) < 0)
252 return -1;
253 return 0;
254}
255
256static void mq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
257{
258 struct net_device *dev = qdisc_dev(sch);
259 unsigned int ntx;
260
261 if (arg->stop)
262 return;
263
264 arg->count = arg->skip;
265 for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
266 if (arg->fn(sch, ntx + 1, arg) < 0) {
267 arg->stop = 1;
268 break;
269 }
270 arg->count++;
271 }
272}
273
274static const struct Qdisc_class_ops mq_class_ops = {
275 .select_queue = mq_select_queue,
276 .graft = mq_graft,
277 .leaf = mq_leaf,
278 .find = mq_find,
279 .walk = mq_walk,
280 .dump = mq_dump_class,
281 .dump_stats = mq_dump_class_stats,
282};
283
284struct Qdisc_ops mq_qdisc_ops __read_mostly = {
285 .cl_ops = &mq_class_ops,
286 .id = "mq",
287 .priv_size = sizeof(struct mq_sched),
288 .init = mq_init,
289 .destroy = mq_destroy,
290 .attach = mq_attach,
291 .dump = mq_dump,
292 .owner = THIS_MODULE,
293};