2 * net/sched/sch_prio.c Simple 3-band priority "scheduler".
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.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * Fixes: 19990609: J Hadi Salim <hadi@nortelnetworks.com>:
11 * Init -- EINVAL when opt undefined
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <net/netlink.h>
21 #include <net/pkt_sched.h>
24 struct prio_sched_data
27 int curband
; /* for round-robin */
28 struct tcf_proto
*filter_list
;
29 u8 prio2band
[TC_PRIO_MAX
+1];
30 struct Qdisc
*queues
[TCQ_PRIO_BANDS
];
36 prio_classify(struct sk_buff
*skb
, struct Qdisc
*sch
, int *qerr
)
38 struct prio_sched_data
*q
= qdisc_priv(sch
);
39 u32 band
= skb
->priority
;
40 struct tcf_result res
;
43 *qerr
= NET_XMIT_BYPASS
;
44 if (TC_H_MAJ(skb
->priority
) != sch
->handle
) {
45 err
= tc_classify(skb
, q
->filter_list
, &res
);
46 #ifdef CONFIG_NET_CLS_ACT
50 *qerr
= NET_XMIT_SUCCESS
;
55 if (!q
->filter_list
|| err
< 0) {
58 band
= q
->prio2band
[band
&TC_PRIO_MAX
];
63 band
= TC_H_MIN(band
) - 1;
65 band
= q
->prio2band
[0];
68 skb_set_queue_mapping(skb
, band
);
69 return q
->queues
[band
];
73 prio_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
)
78 qdisc
= prio_classify(skb
, sch
, &ret
);
79 #ifdef CONFIG_NET_CLS_ACT
82 if (ret
== NET_XMIT_BYPASS
)
89 if ((ret
= qdisc
->enqueue(skb
, qdisc
)) == NET_XMIT_SUCCESS
) {
90 sch
->bstats
.bytes
+= skb
->len
;
91 sch
->bstats
.packets
++;
93 return NET_XMIT_SUCCESS
;
101 prio_requeue(struct sk_buff
*skb
, struct Qdisc
* sch
)
106 qdisc
= prio_classify(skb
, sch
, &ret
);
107 #ifdef CONFIG_NET_CLS_ACT
109 if (ret
== NET_XMIT_BYPASS
)
116 if ((ret
= qdisc
->ops
->requeue(skb
, qdisc
)) == NET_XMIT_SUCCESS
) {
118 sch
->qstats
.requeues
++;
122 return NET_XMIT_DROP
;
126 static struct sk_buff
*
127 prio_dequeue(struct Qdisc
* sch
)
130 struct prio_sched_data
*q
= qdisc_priv(sch
);
134 for (prio
= 0; prio
< q
->bands
; prio
++) {
135 /* Check if the target subqueue is available before
136 * pulling an skb. This way we avoid excessive requeues
139 if (!__netif_subqueue_stopped(sch
->dev
, (q
->mq
? prio
: 0))) {
140 qdisc
= q
->queues
[prio
];
141 skb
= qdisc
->dequeue(qdisc
);
152 static struct sk_buff
*rr_dequeue(struct Qdisc
* sch
)
155 struct prio_sched_data
*q
= qdisc_priv(sch
);
159 /* Only take one pass through the queues. If nothing is available,
162 for (bandcount
= 0; bandcount
< q
->bands
; bandcount
++) {
163 /* Check if the target subqueue is available before
164 * pulling an skb. This way we avoid excessive requeues
165 * for slower queues. If the queue is stopped, try the
168 if (!__netif_subqueue_stopped(sch
->dev
,
169 (q
->mq
? q
->curband
: 0))) {
170 qdisc
= q
->queues
[q
->curband
];
171 skb
= qdisc
->dequeue(qdisc
);
175 if (q
->curband
>= q
->bands
)
181 if (q
->curband
>= q
->bands
)
187 static unsigned int prio_drop(struct Qdisc
* sch
)
189 struct prio_sched_data
*q
= qdisc_priv(sch
);
194 for (prio
= q
->bands
-1; prio
>= 0; prio
--) {
195 qdisc
= q
->queues
[prio
];
196 if (qdisc
->ops
->drop
&& (len
= qdisc
->ops
->drop(qdisc
)) != 0) {
206 prio_reset(struct Qdisc
* sch
)
209 struct prio_sched_data
*q
= qdisc_priv(sch
);
211 for (prio
=0; prio
<q
->bands
; prio
++)
212 qdisc_reset(q
->queues
[prio
]);
217 prio_destroy(struct Qdisc
* sch
)
220 struct prio_sched_data
*q
= qdisc_priv(sch
);
222 tcf_destroy_chain(q
->filter_list
);
223 for (prio
=0; prio
<q
->bands
; prio
++)
224 qdisc_destroy(q
->queues
[prio
]);
227 static int prio_tune(struct Qdisc
*sch
, struct rtattr
*opt
)
229 struct prio_sched_data
*q
= qdisc_priv(sch
);
230 struct tc_prio_qopt
*qopt
;
231 struct rtattr
*tb
[TCA_PRIO_MAX
];
234 if (rtattr_parse_nested_compat(tb
, TCA_PRIO_MAX
, opt
, qopt
,
237 q
->bands
= qopt
->bands
;
238 /* If we're multiqueue, make sure the number of incoming bands
239 * matches the number of queues on the device we're associating with.
240 * If the number of bands requested is zero, then set q->bands to
241 * dev->egress_subqueue_count. Also, the root qdisc must be the
242 * only one that is enabled for multiqueue, since it's the only one
243 * that interacts with the underlying device.
245 q
->mq
= RTA_GET_FLAG(tb
[TCA_PRIO_MQ
- 1]);
247 if (sch
->parent
!= TC_H_ROOT
)
249 if (netif_is_multiqueue(sch
->dev
)) {
251 q
->bands
= sch
->dev
->egress_subqueue_count
;
252 else if (q
->bands
!= sch
->dev
->egress_subqueue_count
)
258 if (q
->bands
> TCQ_PRIO_BANDS
|| q
->bands
< 2)
261 for (i
=0; i
<=TC_PRIO_MAX
; i
++) {
262 if (qopt
->priomap
[i
] >= q
->bands
)
267 memcpy(q
->prio2band
, qopt
->priomap
, TC_PRIO_MAX
+1);
269 for (i
=q
->bands
; i
<TCQ_PRIO_BANDS
; i
++) {
270 struct Qdisc
*child
= xchg(&q
->queues
[i
], &noop_qdisc
);
271 if (child
!= &noop_qdisc
) {
272 qdisc_tree_decrease_qlen(child
, child
->q
.qlen
);
273 qdisc_destroy(child
);
276 sch_tree_unlock(sch
);
278 for (i
=0; i
<q
->bands
; i
++) {
279 if (q
->queues
[i
] == &noop_qdisc
) {
281 child
= qdisc_create_dflt(sch
->dev
, &pfifo_qdisc_ops
,
282 TC_H_MAKE(sch
->handle
, i
+ 1));
285 child
= xchg(&q
->queues
[i
], child
);
287 if (child
!= &noop_qdisc
) {
288 qdisc_tree_decrease_qlen(child
,
290 qdisc_destroy(child
);
292 sch_tree_unlock(sch
);
299 static int prio_init(struct Qdisc
*sch
, struct rtattr
*opt
)
301 struct prio_sched_data
*q
= qdisc_priv(sch
);
304 for (i
=0; i
<TCQ_PRIO_BANDS
; i
++)
305 q
->queues
[i
] = &noop_qdisc
;
312 if ((err
= prio_tune(sch
, opt
)) != 0)
318 static int prio_dump(struct Qdisc
*sch
, struct sk_buff
*skb
)
320 struct prio_sched_data
*q
= qdisc_priv(sch
);
321 unsigned char *b
= skb_tail_pointer(skb
);
323 struct tc_prio_qopt opt
;
325 opt
.bands
= q
->bands
;
326 memcpy(&opt
.priomap
, q
->prio2band
, TC_PRIO_MAX
+1);
328 nest
= RTA_NEST_COMPAT(skb
, TCA_OPTIONS
, sizeof(opt
), &opt
);
330 RTA_PUT_FLAG(skb
, TCA_PRIO_MQ
);
331 RTA_NEST_COMPAT_END(skb
, nest
);
340 static int prio_graft(struct Qdisc
*sch
, unsigned long arg
, struct Qdisc
*new,
343 struct prio_sched_data
*q
= qdisc_priv(sch
);
344 unsigned long band
= arg
- 1;
346 if (band
>= q
->bands
)
353 *old
= q
->queues
[band
];
354 q
->queues
[band
] = new;
355 qdisc_tree_decrease_qlen(*old
, (*old
)->q
.qlen
);
357 sch_tree_unlock(sch
);
362 static struct Qdisc
*
363 prio_leaf(struct Qdisc
*sch
, unsigned long arg
)
365 struct prio_sched_data
*q
= qdisc_priv(sch
);
366 unsigned long band
= arg
- 1;
368 if (band
>= q
->bands
)
371 return q
->queues
[band
];
374 static unsigned long prio_get(struct Qdisc
*sch
, u32 classid
)
376 struct prio_sched_data
*q
= qdisc_priv(sch
);
377 unsigned long band
= TC_H_MIN(classid
);
379 if (band
- 1 >= q
->bands
)
384 static unsigned long prio_bind(struct Qdisc
*sch
, unsigned long parent
, u32 classid
)
386 return prio_get(sch
, classid
);
390 static void prio_put(struct Qdisc
*q
, unsigned long cl
)
395 static int prio_change(struct Qdisc
*sch
, u32 handle
, u32 parent
, struct rtattr
**tca
, unsigned long *arg
)
397 unsigned long cl
= *arg
;
398 struct prio_sched_data
*q
= qdisc_priv(sch
);
400 if (cl
- 1 > q
->bands
)
405 static int prio_delete(struct Qdisc
*sch
, unsigned long cl
)
407 struct prio_sched_data
*q
= qdisc_priv(sch
);
408 if (cl
- 1 > q
->bands
)
414 static int prio_dump_class(struct Qdisc
*sch
, unsigned long cl
, struct sk_buff
*skb
,
417 struct prio_sched_data
*q
= qdisc_priv(sch
);
419 if (cl
- 1 > q
->bands
)
421 tcm
->tcm_handle
|= TC_H_MIN(cl
);
423 tcm
->tcm_info
= q
->queues
[cl
-1]->handle
;
427 static int prio_dump_class_stats(struct Qdisc
*sch
, unsigned long cl
,
430 struct prio_sched_data
*q
= qdisc_priv(sch
);
433 cl_q
= q
->queues
[cl
- 1];
434 if (gnet_stats_copy_basic(d
, &cl_q
->bstats
) < 0 ||
435 gnet_stats_copy_queue(d
, &cl_q
->qstats
) < 0)
441 static void prio_walk(struct Qdisc
*sch
, struct qdisc_walker
*arg
)
443 struct prio_sched_data
*q
= qdisc_priv(sch
);
449 for (prio
= 0; prio
< q
->bands
; prio
++) {
450 if (arg
->count
< arg
->skip
) {
454 if (arg
->fn(sch
, prio
+1, arg
) < 0) {
462 static struct tcf_proto
** prio_find_tcf(struct Qdisc
*sch
, unsigned long cl
)
464 struct prio_sched_data
*q
= qdisc_priv(sch
);
468 return &q
->filter_list
;
471 static const struct Qdisc_class_ops prio_class_ops
= {
476 .change
= prio_change
,
477 .delete = prio_delete
,
479 .tcf_chain
= prio_find_tcf
,
480 .bind_tcf
= prio_bind
,
481 .unbind_tcf
= prio_put
,
482 .dump
= prio_dump_class
,
483 .dump_stats
= prio_dump_class_stats
,
486 static struct Qdisc_ops prio_qdisc_ops __read_mostly
= {
488 .cl_ops
= &prio_class_ops
,
490 .priv_size
= sizeof(struct prio_sched_data
),
491 .enqueue
= prio_enqueue
,
492 .dequeue
= prio_dequeue
,
493 .requeue
= prio_requeue
,
497 .destroy
= prio_destroy
,
500 .owner
= THIS_MODULE
,
503 static struct Qdisc_ops rr_qdisc_ops __read_mostly
= {
505 .cl_ops
= &prio_class_ops
,
507 .priv_size
= sizeof(struct prio_sched_data
),
508 .enqueue
= prio_enqueue
,
509 .dequeue
= rr_dequeue
,
510 .requeue
= prio_requeue
,
514 .destroy
= prio_destroy
,
517 .owner
= THIS_MODULE
,
520 static int __init
prio_module_init(void)
524 err
= register_qdisc(&prio_qdisc_ops
);
527 err
= register_qdisc(&rr_qdisc_ops
);
529 unregister_qdisc(&prio_qdisc_ops
);
533 static void __exit
prio_module_exit(void)
535 unregister_qdisc(&prio_qdisc_ops
);
536 unregister_qdisc(&rr_qdisc_ops
);
539 module_init(prio_module_init
)
540 module_exit(prio_module_exit
)
542 MODULE_LICENSE("GPL");
543 MODULE_ALIAS("sch_rr");