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>
30 struct multiq_sched_data
{
34 struct tcf_proto __rcu
*filter_list
;
35 struct Qdisc
**queues
;
40 multiq_classify(struct sk_buff
*skb
, struct Qdisc
*sch
, int *qerr
)
42 struct multiq_sched_data
*q
= qdisc_priv(sch
);
44 struct tcf_result res
;
45 struct tcf_proto
*fl
= rcu_dereference_bh(q
->filter_list
);
48 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_BYPASS
;
49 err
= tc_classify(skb
, fl
, &res
, false);
50 #ifdef CONFIG_NET_CLS_ACT
54 *qerr
= NET_XMIT_SUCCESS
| __NET_XMIT_STOLEN
;
59 band
= skb_get_queue_mapping(skb
);
64 return q
->queues
[band
];
68 multiq_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
,
69 struct sk_buff
**to_free
)
74 qdisc
= multiq_classify(skb
, sch
, &ret
);
75 #ifdef CONFIG_NET_CLS_ACT
78 if (ret
& __NET_XMIT_BYPASS
)
79 qdisc_qstats_drop(sch
);
80 __qdisc_drop(skb
, to_free
);
85 ret
= qdisc_enqueue(skb
, qdisc
, to_free
);
86 if (ret
== NET_XMIT_SUCCESS
) {
88 return NET_XMIT_SUCCESS
;
90 if (net_xmit_drop_count(ret
))
91 qdisc_qstats_drop(sch
);
95 static struct sk_buff
*multiq_dequeue(struct Qdisc
*sch
)
97 struct multiq_sched_data
*q
= qdisc_priv(sch
);
102 for (band
= 0; band
< q
->bands
; band
++) {
103 /* cycle through bands to ensure fairness */
105 if (q
->curband
>= q
->bands
)
108 /* Check that target subqueue is available before
109 * pulling an skb to avoid head-of-line blocking.
111 if (!netif_xmit_stopped(
112 netdev_get_tx_queue(qdisc_dev(sch
), q
->curband
))) {
113 qdisc
= q
->queues
[q
->curband
];
114 skb
= qdisc
->dequeue(qdisc
);
116 qdisc_bstats_update(sch
, skb
);
126 static struct sk_buff
*multiq_peek(struct Qdisc
*sch
)
128 struct multiq_sched_data
*q
= qdisc_priv(sch
);
129 unsigned int curband
= q
->curband
;
134 for (band
= 0; band
< q
->bands
; band
++) {
135 /* cycle through bands to ensure fairness */
137 if (curband
>= q
->bands
)
140 /* Check that target subqueue is available before
141 * pulling an skb to avoid head-of-line blocking.
143 if (!netif_xmit_stopped(
144 netdev_get_tx_queue(qdisc_dev(sch
), curband
))) {
145 qdisc
= q
->queues
[curband
];
146 skb
= qdisc
->ops
->peek(qdisc
);
156 multiq_reset(struct Qdisc
*sch
)
159 struct multiq_sched_data
*q
= qdisc_priv(sch
);
161 for (band
= 0; band
< q
->bands
; band
++)
162 qdisc_reset(q
->queues
[band
]);
168 multiq_destroy(struct Qdisc
*sch
)
171 struct multiq_sched_data
*q
= qdisc_priv(sch
);
173 tcf_destroy_chain(&q
->filter_list
);
174 for (band
= 0; band
< q
->bands
; band
++)
175 qdisc_destroy(q
->queues
[band
]);
180 static int multiq_tune(struct Qdisc
*sch
, struct nlattr
*opt
)
182 struct multiq_sched_data
*q
= qdisc_priv(sch
);
183 struct tc_multiq_qopt
*qopt
;
186 if (!netif_is_multiqueue(qdisc_dev(sch
)))
188 if (nla_len(opt
) < sizeof(*qopt
))
191 qopt
= nla_data(opt
);
193 qopt
->bands
= qdisc_dev(sch
)->real_num_tx_queues
;
196 q
->bands
= qopt
->bands
;
197 for (i
= q
->bands
; i
< q
->max_bands
; i
++) {
198 if (q
->queues
[i
] != &noop_qdisc
) {
199 struct Qdisc
*child
= q
->queues
[i
];
200 q
->queues
[i
] = &noop_qdisc
;
201 qdisc_tree_reduce_backlog(child
, child
->q
.qlen
,
202 child
->qstats
.backlog
);
203 qdisc_destroy(child
);
207 sch_tree_unlock(sch
);
209 for (i
= 0; i
< q
->bands
; i
++) {
210 if (q
->queues
[i
] == &noop_qdisc
) {
211 struct Qdisc
*child
, *old
;
212 child
= qdisc_create_dflt(sch
->dev_queue
,
214 TC_H_MAKE(sch
->handle
,
219 q
->queues
[i
] = child
;
221 if (old
!= &noop_qdisc
) {
222 qdisc_tree_reduce_backlog(old
,
224 old
->qstats
.backlog
);
227 sch_tree_unlock(sch
);
234 static int multiq_init(struct Qdisc
*sch
, struct nlattr
*opt
)
236 struct multiq_sched_data
*q
= qdisc_priv(sch
);
244 q
->max_bands
= qdisc_dev(sch
)->num_tx_queues
;
246 q
->queues
= kcalloc(q
->max_bands
, sizeof(struct Qdisc
*), GFP_KERNEL
);
249 for (i
= 0; i
< q
->max_bands
; i
++)
250 q
->queues
[i
] = &noop_qdisc
;
252 err
= multiq_tune(sch
, opt
);
260 static int multiq_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
262 struct multiq_sched_data
*q
= qdisc_priv(sch
);
263 unsigned char *b
= skb_tail_pointer(skb
);
264 struct tc_multiq_qopt opt
;
266 opt
.bands
= q
->bands
;
267 opt
.max_bands
= q
->max_bands
;
269 if (nla_put(skb
, TCA_OPTIONS
, sizeof(opt
), &opt
))
270 goto nla_put_failure
;
279 static int multiq_graft(struct Qdisc
*sch
, unsigned long arg
, struct Qdisc
*new,
282 struct multiq_sched_data
*q
= qdisc_priv(sch
);
283 unsigned long band
= arg
- 1;
288 *old
= qdisc_replace(sch
, new, &q
->queues
[band
]);
292 static struct Qdisc
*
293 multiq_leaf(struct Qdisc
*sch
, unsigned long arg
)
295 struct multiq_sched_data
*q
= qdisc_priv(sch
);
296 unsigned long band
= arg
- 1;
298 return q
->queues
[band
];
301 static unsigned long multiq_get(struct Qdisc
*sch
, u32 classid
)
303 struct multiq_sched_data
*q
= qdisc_priv(sch
);
304 unsigned long band
= TC_H_MIN(classid
);
306 if (band
- 1 >= q
->bands
)
311 static unsigned long multiq_bind(struct Qdisc
*sch
, unsigned long parent
,
314 return multiq_get(sch
, classid
);
318 static void multiq_put(struct Qdisc
*q
, unsigned long cl
)
322 static int multiq_dump_class(struct Qdisc
*sch
, unsigned long cl
,
323 struct sk_buff
*skb
, struct tcmsg
*tcm
)
325 struct multiq_sched_data
*q
= qdisc_priv(sch
);
327 tcm
->tcm_handle
|= TC_H_MIN(cl
);
328 tcm
->tcm_info
= q
->queues
[cl
- 1]->handle
;
332 static int multiq_dump_class_stats(struct Qdisc
*sch
, unsigned long cl
,
335 struct multiq_sched_data
*q
= qdisc_priv(sch
);
338 cl_q
= q
->queues
[cl
- 1];
339 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(sch
),
340 d
, NULL
, &cl_q
->bstats
) < 0 ||
341 gnet_stats_copy_queue(d
, NULL
, &cl_q
->qstats
, cl_q
->q
.qlen
) < 0)
347 static void multiq_walk(struct Qdisc
*sch
, struct qdisc_walker
*arg
)
349 struct multiq_sched_data
*q
= qdisc_priv(sch
);
355 for (band
= 0; band
< q
->bands
; band
++) {
356 if (arg
->count
< arg
->skip
) {
360 if (arg
->fn(sch
, band
+ 1, arg
) < 0) {
368 static struct tcf_proto __rcu
**multiq_find_tcf(struct Qdisc
*sch
,
371 struct multiq_sched_data
*q
= qdisc_priv(sch
);
375 return &q
->filter_list
;
378 static const struct Qdisc_class_ops multiq_class_ops
= {
379 .graft
= multiq_graft
,
384 .tcf_chain
= multiq_find_tcf
,
385 .bind_tcf
= multiq_bind
,
386 .unbind_tcf
= multiq_put
,
387 .dump
= multiq_dump_class
,
388 .dump_stats
= multiq_dump_class_stats
,
391 static struct Qdisc_ops multiq_qdisc_ops __read_mostly
= {
393 .cl_ops
= &multiq_class_ops
,
395 .priv_size
= sizeof(struct multiq_sched_data
),
396 .enqueue
= multiq_enqueue
,
397 .dequeue
= multiq_dequeue
,
400 .reset
= multiq_reset
,
401 .destroy
= multiq_destroy
,
402 .change
= multiq_tune
,
404 .owner
= THIS_MODULE
,
407 static int __init
multiq_module_init(void)
409 return register_qdisc(&multiq_qdisc_ops
);
412 static void __exit
multiq_module_exit(void)
414 unregister_qdisc(&multiq_qdisc_ops
);
417 module_init(multiq_module_init
)
418 module_exit(multiq_module_exit
)
420 MODULE_LICENSE("GPL");