2 * Copyright (c) 2008, Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
16 * Author: Alexander Duyck <alexander.h.duyck@intel.com>
19 #include <linux/module.h>
20 #include <linux/slab.h>
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/errno.h>
25 #include <linux/skbuff.h>
26 #include <net/netlink.h>
27 #include <net/pkt_sched.h>
28 #include <net/pkt_cls.h>
30 struct multiq_sched_data
{
34 struct tcf_proto __rcu
*filter_list
;
35 struct tcf_block
*block
;
36 struct Qdisc
**queues
;
41 multiq_classify(struct sk_buff
*skb
, struct Qdisc
*sch
, int *qerr
)
43 struct multiq_sched_data
*q
= qdisc_priv(sch
);
45 struct tcf_result res
;
46 struct tcf_proto
*fl
= rcu_dereference_bh(q
->filter_list
);
49 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
50 err
= tcf_classify(skb
, fl
, &res
, false);
51 #ifdef CONFIG_NET_CLS_ACT
56 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
62 band
= skb_get_queue_mapping(skb
);
67 return q
->queues
[band
];
71 multiq_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
,
72 struct sk_buff
**to_free
)
77 qdisc
= multiq_classify(skb
, sch
, &ret
);
78 #ifdef CONFIG_NET_CLS_ACT
81 if (ret
& __NET_XMIT_BYPASS
)
82 qdisc_qstats_drop(sch
);
83 __qdisc_drop(skb
, to_free
);
88 ret
= qdisc_enqueue(skb
, qdisc
, to_free
);
89 if (ret
== NET_XMIT_SUCCESS
) {
91 return NET_XMIT_SUCCESS
;
93 if (net_xmit_drop_count(ret
))
94 qdisc_qstats_drop(sch
);
98 static struct sk_buff
*multiq_dequeue(struct Qdisc
*sch
)
100 struct multiq_sched_data
*q
= qdisc_priv(sch
);
105 for (band
= 0; band
< q
->bands
; band
++) {
106 /* cycle through bands to ensure fairness */
108 if (q
->curband
>= q
->bands
)
111 /* Check that target subqueue is available before
112 * pulling an skb to avoid head-of-line blocking.
114 if (!netif_xmit_stopped(
115 netdev_get_tx_queue(qdisc_dev(sch
), q
->curband
))) {
116 qdisc
= q
->queues
[q
->curband
];
117 skb
= qdisc
->dequeue(qdisc
);
119 qdisc_bstats_update(sch
, skb
);
129 static struct sk_buff
*multiq_peek(struct Qdisc
*sch
)
131 struct multiq_sched_data
*q
= qdisc_priv(sch
);
132 unsigned int curband
= q
->curband
;
137 for (band
= 0; band
< q
->bands
; band
++) {
138 /* cycle through bands to ensure fairness */
140 if (curband
>= q
->bands
)
143 /* Check that target subqueue is available before
144 * pulling an skb to avoid head-of-line blocking.
146 if (!netif_xmit_stopped(
147 netdev_get_tx_queue(qdisc_dev(sch
), curband
))) {
148 qdisc
= q
->queues
[curband
];
149 skb
= qdisc
->ops
->peek(qdisc
);
159 multiq_reset(struct Qdisc
*sch
)
162 struct multiq_sched_data
*q
= qdisc_priv(sch
);
164 for (band
= 0; band
< q
->bands
; band
++)
165 qdisc_reset(q
->queues
[band
]);
171 multiq_destroy(struct Qdisc
*sch
)
174 struct multiq_sched_data
*q
= qdisc_priv(sch
);
176 tcf_block_put(q
->block
);
177 for (band
= 0; band
< q
->bands
; band
++)
178 qdisc_destroy(q
->queues
[band
]);
183 static int multiq_tune(struct Qdisc
*sch
, struct nlattr
*opt
)
185 struct multiq_sched_data
*q
= qdisc_priv(sch
);
186 struct tc_multiq_qopt
*qopt
;
189 if (!netif_is_multiqueue(qdisc_dev(sch
)))
191 if (nla_len(opt
) < sizeof(*qopt
))
194 qopt
= nla_data(opt
);
196 qopt
->bands
= qdisc_dev(sch
)->real_num_tx_queues
;
199 q
->bands
= qopt
->bands
;
200 for (i
= q
->bands
; i
< q
->max_bands
; i
++) {
201 if (q
->queues
[i
] != &noop_qdisc
) {
202 struct Qdisc
*child
= q
->queues
[i
];
203 q
->queues
[i
] = &noop_qdisc
;
204 qdisc_tree_reduce_backlog(child
, child
->q
.qlen
,
205 child
->qstats
.backlog
);
206 qdisc_destroy(child
);
210 sch_tree_unlock(sch
);
212 for (i
= 0; i
< q
->bands
; i
++) {
213 if (q
->queues
[i
] == &noop_qdisc
) {
214 struct Qdisc
*child
, *old
;
215 child
= qdisc_create_dflt(sch
->dev_queue
,
217 TC_H_MAKE(sch
->handle
,
222 q
->queues
[i
] = child
;
223 if (child
!= &noop_qdisc
)
224 qdisc_hash_add(child
, true);
226 if (old
!= &noop_qdisc
) {
227 qdisc_tree_reduce_backlog(old
,
229 old
->qstats
.backlog
);
232 sch_tree_unlock(sch
);
239 static int multiq_init(struct Qdisc
*sch
, struct nlattr
*opt
)
241 struct multiq_sched_data
*q
= qdisc_priv(sch
);
249 err
= tcf_block_get(&q
->block
, &q
->filter_list
, sch
);
253 q
->max_bands
= qdisc_dev(sch
)->num_tx_queues
;
255 q
->queues
= kcalloc(q
->max_bands
, sizeof(struct Qdisc
*), GFP_KERNEL
);
258 for (i
= 0; i
< q
->max_bands
; i
++)
259 q
->queues
[i
] = &noop_qdisc
;
261 return multiq_tune(sch
, opt
);
264 static int multiq_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
266 struct multiq_sched_data
*q
= qdisc_priv(sch
);
267 unsigned char *b
= skb_tail_pointer(skb
);
268 struct tc_multiq_qopt opt
;
270 opt
.bands
= q
->bands
;
271 opt
.max_bands
= q
->max_bands
;
273 if (nla_put(skb
, TCA_OPTIONS
, sizeof(opt
), &opt
))
274 goto nla_put_failure
;
283 static int multiq_graft(struct Qdisc
*sch
, unsigned long arg
, struct Qdisc
*new,
286 struct multiq_sched_data
*q
= qdisc_priv(sch
);
287 unsigned long band
= arg
- 1;
292 *old
= qdisc_replace(sch
, new, &q
->queues
[band
]);
296 static struct Qdisc
*
297 multiq_leaf(struct Qdisc
*sch
, unsigned long arg
)
299 struct multiq_sched_data
*q
= qdisc_priv(sch
);
300 unsigned long band
= arg
- 1;
302 return q
->queues
[band
];
305 static unsigned long multiq_find(struct Qdisc
*sch
, u32 classid
)
307 struct multiq_sched_data
*q
= qdisc_priv(sch
);
308 unsigned long band
= TC_H_MIN(classid
);
310 if (band
- 1 >= q
->bands
)
315 static unsigned long multiq_bind(struct Qdisc
*sch
, unsigned long parent
,
318 return multiq_find(sch
, classid
);
322 static void multiq_unbind(struct Qdisc
*q
, unsigned long cl
)
326 static int multiq_dump_class(struct Qdisc
*sch
, unsigned long cl
,
327 struct sk_buff
*skb
, struct tcmsg
*tcm
)
329 struct multiq_sched_data
*q
= qdisc_priv(sch
);
331 tcm
->tcm_handle
|= TC_H_MIN(cl
);
332 tcm
->tcm_info
= q
->queues
[cl
- 1]->handle
;
336 static int multiq_dump_class_stats(struct Qdisc
*sch
, unsigned long cl
,
339 struct multiq_sched_data
*q
= qdisc_priv(sch
);
342 cl_q
= q
->queues
[cl
- 1];
343 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch
),
344 d
, NULL
, &cl_q
->bstats
) < 0 ||
345 gnet_stats_copy_queue(d
, NULL
, &cl_q
->qstats
, cl_q
->q
.qlen
) < 0)
351 static void multiq_walk(struct Qdisc
*sch
, struct qdisc_walker
*arg
)
353 struct multiq_sched_data
*q
= qdisc_priv(sch
);
359 for (band
= 0; band
< q
->bands
; band
++) {
360 if (arg
->count
< arg
->skip
) {
364 if (arg
->fn(sch
, band
+ 1, arg
) < 0) {
372 static struct tcf_block
*multiq_tcf_block(struct Qdisc
*sch
, unsigned long cl
)
374 struct multiq_sched_data
*q
= qdisc_priv(sch
);
381 static const struct Qdisc_class_ops multiq_class_ops
= {
382 .graft
= multiq_graft
,
386 .tcf_block
= multiq_tcf_block
,
387 .bind_tcf
= multiq_bind
,
388 .unbind_tcf
= multiq_unbind
,
389 .dump
= multiq_dump_class
,
390 .dump_stats
= multiq_dump_class_stats
,
393 static struct Qdisc_ops multiq_qdisc_ops __read_mostly
= {
395 .cl_ops
= &multiq_class_ops
,
397 .priv_size
= sizeof(struct multiq_sched_data
),
398 .enqueue
= multiq_enqueue
,
399 .dequeue
= multiq_dequeue
,
402 .reset
= multiq_reset
,
403 .destroy
= multiq_destroy
,
404 .change
= multiq_tune
,
406 .owner
= THIS_MODULE
,
409 static int __init
multiq_module_init(void)
411 return register_qdisc(&multiq_qdisc_ops
);
414 static void __exit
multiq_module_exit(void)
416 unregister_qdisc(&multiq_qdisc_ops
);
419 module_init(multiq_module_init
)
420 module_exit(multiq_module_exit
)
422 MODULE_LICENSE("GPL");