2 * net/sched/sch_api.c Packet scheduler API.
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>
13 * Rani Assaf <rani@magic.metawire.com> :980802: JIFFIES and CPU clock sources are repaired.
14 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
15 * Jamal Hadi Salim <hadi@nortelnetworks.com>: 990601: ingress support
18 #include <linux/module.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/string.h>
22 #include <linux/errno.h>
23 #include <linux/skbuff.h>
24 #include <linux/init.h>
25 #include <linux/proc_fs.h>
26 #include <linux/seq_file.h>
27 #include <linux/kmod.h>
28 #include <linux/list.h>
29 #include <linux/hrtimer.h>
30 #include <linux/lockdep.h>
31 #include <linux/slab.h>
33 #include <net/net_namespace.h>
35 #include <net/netlink.h>
36 #include <net/pkt_sched.h>
38 static int qdisc_notify(struct net
*net
, struct sk_buff
*oskb
,
39 struct nlmsghdr
*n
, u32 clid
,
40 struct Qdisc
*old
, struct Qdisc
*new);
41 static int tclass_notify(struct net
*net
, struct sk_buff
*oskb
,
42 struct nlmsghdr
*n
, struct Qdisc
*q
,
43 unsigned long cl
, int event
);
50 This file consists of two interrelated parts:
52 1. queueing disciplines manager frontend.
53 2. traffic classes manager frontend.
55 Generally, queueing discipline ("qdisc") is a black box,
56 which is able to enqueue packets and to dequeue them (when
57 device is ready to send something) in order and at times
58 determined by algorithm hidden in it.
60 qdisc's are divided to two categories:
61 - "queues", which have no internal structure visible from outside.
62 - "schedulers", which split all the packets to "traffic classes",
63 using "packet classifiers" (look at cls_api.c)
65 In turn, classes may have child qdiscs (as rule, queues)
66 attached to them etc. etc. etc.
68 The goal of the routines in this file is to translate
69 information supplied by user in the form of handles
70 to more intelligible for kernel form, to make some sanity
71 checks and part of work, which is common to all qdiscs
72 and to provide rtnetlink notifications.
74 All real intelligent work is done inside qdisc modules.
78 Every discipline has two major routines: enqueue and dequeue.
82 dequeue usually returns a skb to send. It is allowed to return NULL,
83 but it does not mean that queue is empty, it just means that
84 discipline does not want to send anything this time.
85 Queue is really empty if q->q.qlen == 0.
86 For complicated disciplines with multiple queues q->q is not
87 real packet queue, but however q->q.qlen must be valid.
91 enqueue returns 0, if packet was enqueued successfully.
92 If packet (this one or another one) was dropped, it returns
94 NET_XMIT_DROP - this packet dropped
95 Expected action: do not backoff, but wait until queue will clear.
96 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
97 Expected action: backoff or ignore
98 NET_XMIT_POLICED - dropped by police.
99 Expected action: backoff or error to real-time apps.
105 like dequeue but without removing a packet from the queue
109 returns qdisc to initial state: purge all buffers, clear all
110 timers, counters (except for statistics) etc.
114 initializes newly created qdisc.
118 destroys resources allocated by init and during lifetime of qdisc.
122 changes qdisc parameters.
125 /* Protects list of registered TC modules. It is pure SMP lock. */
126 static DEFINE_RWLOCK(qdisc_mod_lock
);
129 /************************************************
130 * Queueing disciplines manipulation. *
131 ************************************************/
134 /* The list of all installed queueing disciplines. */
136 static struct Qdisc_ops
*qdisc_base
;
138 /* Register/uregister queueing discipline */
140 int register_qdisc(struct Qdisc_ops
*qops
)
142 struct Qdisc_ops
*q
, **qp
;
145 write_lock(&qdisc_mod_lock
);
146 for (qp
= &qdisc_base
; (q
= *qp
) != NULL
; qp
= &q
->next
)
147 if (!strcmp(qops
->id
, q
->id
))
150 if (qops
->enqueue
== NULL
)
151 qops
->enqueue
= noop_qdisc_ops
.enqueue
;
152 if (qops
->peek
== NULL
) {
153 if (qops
->dequeue
== NULL
)
154 qops
->peek
= noop_qdisc_ops
.peek
;
158 if (qops
->dequeue
== NULL
)
159 qops
->dequeue
= noop_qdisc_ops
.dequeue
;
162 const struct Qdisc_class_ops
*cops
= qops
->cl_ops
;
164 if (!(cops
->get
&& cops
->put
&& cops
->walk
&& cops
->leaf
))
167 if (cops
->tcf_chain
&& !(cops
->bind_tcf
&& cops
->unbind_tcf
))
175 write_unlock(&qdisc_mod_lock
);
182 EXPORT_SYMBOL(register_qdisc
);
184 int unregister_qdisc(struct Qdisc_ops
*qops
)
186 struct Qdisc_ops
*q
, **qp
;
189 write_lock(&qdisc_mod_lock
);
190 for (qp
= &qdisc_base
; (q
= *qp
) != NULL
; qp
= &q
->next
)
198 write_unlock(&qdisc_mod_lock
);
201 EXPORT_SYMBOL(unregister_qdisc
);
203 /* We know handle. Find qdisc among all qdisc's attached to device
204 (root qdisc, all its children, children of children etc.)
207 static struct Qdisc
*qdisc_match_from_root(struct Qdisc
*root
, u32 handle
)
211 if (!(root
->flags
& TCQ_F_BUILTIN
) &&
212 root
->handle
== handle
)
215 list_for_each_entry(q
, &root
->list
, list
) {
216 if (q
->handle
== handle
)
222 static void qdisc_list_add(struct Qdisc
*q
)
224 if ((q
->parent
!= TC_H_ROOT
) && !(q
->flags
& TCQ_F_INGRESS
))
225 list_add_tail(&q
->list
, &qdisc_dev(q
)->qdisc
->list
);
228 void qdisc_list_del(struct Qdisc
*q
)
230 if ((q
->parent
!= TC_H_ROOT
) && !(q
->flags
& TCQ_F_INGRESS
))
233 EXPORT_SYMBOL(qdisc_list_del
);
235 struct Qdisc
*qdisc_lookup(struct net_device
*dev
, u32 handle
)
239 q
= qdisc_match_from_root(dev
->qdisc
, handle
);
243 if (dev_ingress_queue(dev
))
244 q
= qdisc_match_from_root(
245 dev_ingress_queue(dev
)->qdisc_sleeping
,
251 static struct Qdisc
*qdisc_leaf(struct Qdisc
*p
, u32 classid
)
255 const struct Qdisc_class_ops
*cops
= p
->ops
->cl_ops
;
259 cl
= cops
->get(p
, classid
);
263 leaf
= cops
->leaf(p
, cl
);
268 /* Find queueing discipline by name */
270 static struct Qdisc_ops
*qdisc_lookup_ops(struct nlattr
*kind
)
272 struct Qdisc_ops
*q
= NULL
;
275 read_lock(&qdisc_mod_lock
);
276 for (q
= qdisc_base
; q
; q
= q
->next
) {
277 if (nla_strcmp(kind
, q
->id
) == 0) {
278 if (!try_module_get(q
->owner
))
283 read_unlock(&qdisc_mod_lock
);
288 /* The linklayer setting were not transferred from iproute2, in older
289 * versions, and the rate tables lookup systems have been dropped in
290 * the kernel. To keep backward compatible with older iproute2 tc
291 * utils, we detect the linklayer setting by detecting if the rate
292 * table were modified.
294 * For linklayer ATM table entries, the rate table will be aligned to
295 * 48 bytes, thus some table entries will contain the same value. The
296 * mpu (min packet unit) is also encoded into the old rate table, thus
297 * starting from the mpu, we find low and high table entries for
298 * mapping this cell. If these entries contain the same value, when
299 * the rate tables have been modified for linklayer ATM.
301 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
302 * and then roundup to the next cell, calc the table entry one below,
305 static __u8
__detect_linklayer(struct tc_ratespec
*r
, __u32
*rtab
)
307 int low
= roundup(r
->mpu
, 48);
308 int high
= roundup(low
+1, 48);
309 int cell_low
= low
>> r
->cell_log
;
310 int cell_high
= (high
>> r
->cell_log
) - 1;
312 /* rtab is too inaccurate at rates > 100Mbit/s */
313 if ((r
->rate
> (100000000/8)) || (rtab
[0] == 0)) {
314 pr_debug("TC linklayer: Giving up ATM detection\n");
315 return TC_LINKLAYER_ETHERNET
;
318 if ((cell_high
> cell_low
) && (cell_high
< 256)
319 && (rtab
[cell_low
] == rtab
[cell_high
])) {
320 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
321 cell_low
, cell_high
, rtab
[cell_high
]);
322 return TC_LINKLAYER_ATM
;
324 return TC_LINKLAYER_ETHERNET
;
327 static struct qdisc_rate_table
*qdisc_rtab_list
;
329 struct qdisc_rate_table
*qdisc_get_rtab(struct tc_ratespec
*r
, struct nlattr
*tab
)
331 struct qdisc_rate_table
*rtab
;
333 if (tab
== NULL
|| r
->rate
== 0 || r
->cell_log
== 0 ||
334 nla_len(tab
) != TC_RTAB_SIZE
)
337 for (rtab
= qdisc_rtab_list
; rtab
; rtab
= rtab
->next
) {
338 if (!memcmp(&rtab
->rate
, r
, sizeof(struct tc_ratespec
)) &&
339 !memcmp(&rtab
->data
, nla_data(tab
), 1024)) {
345 rtab
= kmalloc(sizeof(*rtab
), GFP_KERNEL
);
349 memcpy(rtab
->data
, nla_data(tab
), 1024);
350 if (r
->linklayer
== TC_LINKLAYER_UNAWARE
)
351 r
->linklayer
= __detect_linklayer(r
, rtab
->data
);
352 rtab
->next
= qdisc_rtab_list
;
353 qdisc_rtab_list
= rtab
;
357 EXPORT_SYMBOL(qdisc_get_rtab
);
359 void qdisc_put_rtab(struct qdisc_rate_table
*tab
)
361 struct qdisc_rate_table
*rtab
, **rtabp
;
363 if (!tab
|| --tab
->refcnt
)
366 for (rtabp
= &qdisc_rtab_list
;
367 (rtab
= *rtabp
) != NULL
;
368 rtabp
= &rtab
->next
) {
376 EXPORT_SYMBOL(qdisc_put_rtab
);
378 static LIST_HEAD(qdisc_stab_list
);
379 static DEFINE_SPINLOCK(qdisc_stab_lock
);
381 static const struct nla_policy stab_policy
[TCA_STAB_MAX
+ 1] = {
382 [TCA_STAB_BASE
] = { .len
= sizeof(struct tc_sizespec
) },
383 [TCA_STAB_DATA
] = { .type
= NLA_BINARY
},
386 static struct qdisc_size_table
*qdisc_get_stab(struct nlattr
*opt
)
388 struct nlattr
*tb
[TCA_STAB_MAX
+ 1];
389 struct qdisc_size_table
*stab
;
390 struct tc_sizespec
*s
;
391 unsigned int tsize
= 0;
395 err
= nla_parse_nested(tb
, TCA_STAB_MAX
, opt
, stab_policy
);
398 if (!tb
[TCA_STAB_BASE
])
399 return ERR_PTR(-EINVAL
);
401 s
= nla_data(tb
[TCA_STAB_BASE
]);
404 if (!tb
[TCA_STAB_DATA
])
405 return ERR_PTR(-EINVAL
);
406 tab
= nla_data(tb
[TCA_STAB_DATA
]);
407 tsize
= nla_len(tb
[TCA_STAB_DATA
]) / sizeof(u16
);
410 if (tsize
!= s
->tsize
|| (!tab
&& tsize
> 0))
411 return ERR_PTR(-EINVAL
);
413 spin_lock(&qdisc_stab_lock
);
415 list_for_each_entry(stab
, &qdisc_stab_list
, list
) {
416 if (memcmp(&stab
->szopts
, s
, sizeof(*s
)))
418 if (tsize
> 0 && memcmp(stab
->data
, tab
, tsize
* sizeof(u16
)))
421 spin_unlock(&qdisc_stab_lock
);
425 spin_unlock(&qdisc_stab_lock
);
427 stab
= kmalloc(sizeof(*stab
) + tsize
* sizeof(u16
), GFP_KERNEL
);
429 return ERR_PTR(-ENOMEM
);
434 memcpy(stab
->data
, tab
, tsize
* sizeof(u16
));
436 spin_lock(&qdisc_stab_lock
);
437 list_add_tail(&stab
->list
, &qdisc_stab_list
);
438 spin_unlock(&qdisc_stab_lock
);
443 static void stab_kfree_rcu(struct rcu_head
*head
)
445 kfree(container_of(head
, struct qdisc_size_table
, rcu
));
448 void qdisc_put_stab(struct qdisc_size_table
*tab
)
453 spin_lock(&qdisc_stab_lock
);
455 if (--tab
->refcnt
== 0) {
456 list_del(&tab
->list
);
457 call_rcu_bh(&tab
->rcu
, stab_kfree_rcu
);
460 spin_unlock(&qdisc_stab_lock
);
462 EXPORT_SYMBOL(qdisc_put_stab
);
464 static int qdisc_dump_stab(struct sk_buff
*skb
, struct qdisc_size_table
*stab
)
468 nest
= nla_nest_start(skb
, TCA_STAB
);
470 goto nla_put_failure
;
471 if (nla_put(skb
, TCA_STAB_BASE
, sizeof(stab
->szopts
), &stab
->szopts
))
472 goto nla_put_failure
;
473 nla_nest_end(skb
, nest
);
481 void __qdisc_calculate_pkt_len(struct sk_buff
*skb
, const struct qdisc_size_table
*stab
)
485 pkt_len
= skb
->len
+ stab
->szopts
.overhead
;
486 if (unlikely(!stab
->szopts
.tsize
))
489 slot
= pkt_len
+ stab
->szopts
.cell_align
;
490 if (unlikely(slot
< 0))
493 slot
>>= stab
->szopts
.cell_log
;
494 if (likely(slot
< stab
->szopts
.tsize
))
495 pkt_len
= stab
->data
[slot
];
497 pkt_len
= stab
->data
[stab
->szopts
.tsize
- 1] *
498 (slot
/ stab
->szopts
.tsize
) +
499 stab
->data
[slot
% stab
->szopts
.tsize
];
501 pkt_len
<<= stab
->szopts
.size_log
;
503 if (unlikely(pkt_len
< 1))
505 qdisc_skb_cb(skb
)->pkt_len
= pkt_len
;
507 EXPORT_SYMBOL(__qdisc_calculate_pkt_len
);
509 void qdisc_warn_nonwc(char *txt
, struct Qdisc
*qdisc
)
511 if (!(qdisc
->flags
& TCQ_F_WARN_NONWC
)) {
512 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
513 txt
, qdisc
->ops
->id
, qdisc
->handle
>> 16);
514 qdisc
->flags
|= TCQ_F_WARN_NONWC
;
517 EXPORT_SYMBOL(qdisc_warn_nonwc
);
519 static enum hrtimer_restart
qdisc_watchdog(struct hrtimer
*timer
)
521 struct qdisc_watchdog
*wd
= container_of(timer
, struct qdisc_watchdog
,
524 qdisc_unthrottled(wd
->qdisc
);
525 __netif_schedule(qdisc_root(wd
->qdisc
));
527 return HRTIMER_NORESTART
;
530 void qdisc_watchdog_init(struct qdisc_watchdog
*wd
, struct Qdisc
*qdisc
)
532 hrtimer_init(&wd
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS
);
533 wd
->timer
.function
= qdisc_watchdog
;
536 EXPORT_SYMBOL(qdisc_watchdog_init
);
538 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog
*wd
, u64 expires
)
540 if (test_bit(__QDISC_STATE_DEACTIVATED
,
541 &qdisc_root_sleeping(wd
->qdisc
)->state
))
544 qdisc_throttled(wd
->qdisc
);
546 hrtimer_start(&wd
->timer
,
547 ns_to_ktime(expires
),
550 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns
);
552 void qdisc_watchdog_cancel(struct qdisc_watchdog
*wd
)
554 hrtimer_cancel(&wd
->timer
);
555 qdisc_unthrottled(wd
->qdisc
);
557 EXPORT_SYMBOL(qdisc_watchdog_cancel
);
559 static struct hlist_head
*qdisc_class_hash_alloc(unsigned int n
)
561 unsigned int size
= n
* sizeof(struct hlist_head
), i
;
562 struct hlist_head
*h
;
564 if (size
<= PAGE_SIZE
)
565 h
= kmalloc(size
, GFP_KERNEL
);
567 h
= (struct hlist_head
*)
568 __get_free_pages(GFP_KERNEL
, get_order(size
));
571 for (i
= 0; i
< n
; i
++)
572 INIT_HLIST_HEAD(&h
[i
]);
577 static void qdisc_class_hash_free(struct hlist_head
*h
, unsigned int n
)
579 unsigned int size
= n
* sizeof(struct hlist_head
);
581 if (size
<= PAGE_SIZE
)
584 free_pages((unsigned long)h
, get_order(size
));
587 void qdisc_class_hash_grow(struct Qdisc
*sch
, struct Qdisc_class_hash
*clhash
)
589 struct Qdisc_class_common
*cl
;
590 struct hlist_node
*next
;
591 struct hlist_head
*nhash
, *ohash
;
592 unsigned int nsize
, nmask
, osize
;
595 /* Rehash when load factor exceeds 0.75 */
596 if (clhash
->hashelems
* 4 <= clhash
->hashsize
* 3)
598 nsize
= clhash
->hashsize
* 2;
600 nhash
= qdisc_class_hash_alloc(nsize
);
604 ohash
= clhash
->hash
;
605 osize
= clhash
->hashsize
;
608 for (i
= 0; i
< osize
; i
++) {
609 hlist_for_each_entry_safe(cl
, next
, &ohash
[i
], hnode
) {
610 h
= qdisc_class_hash(cl
->classid
, nmask
);
611 hlist_add_head(&cl
->hnode
, &nhash
[h
]);
614 clhash
->hash
= nhash
;
615 clhash
->hashsize
= nsize
;
616 clhash
->hashmask
= nmask
;
617 sch_tree_unlock(sch
);
619 qdisc_class_hash_free(ohash
, osize
);
621 EXPORT_SYMBOL(qdisc_class_hash_grow
);
623 int qdisc_class_hash_init(struct Qdisc_class_hash
*clhash
)
625 unsigned int size
= 4;
627 clhash
->hash
= qdisc_class_hash_alloc(size
);
628 if (clhash
->hash
== NULL
)
630 clhash
->hashsize
= size
;
631 clhash
->hashmask
= size
- 1;
632 clhash
->hashelems
= 0;
635 EXPORT_SYMBOL(qdisc_class_hash_init
);
637 void qdisc_class_hash_destroy(struct Qdisc_class_hash
*clhash
)
639 qdisc_class_hash_free(clhash
->hash
, clhash
->hashsize
);
641 EXPORT_SYMBOL(qdisc_class_hash_destroy
);
643 void qdisc_class_hash_insert(struct Qdisc_class_hash
*clhash
,
644 struct Qdisc_class_common
*cl
)
648 INIT_HLIST_NODE(&cl
->hnode
);
649 h
= qdisc_class_hash(cl
->classid
, clhash
->hashmask
);
650 hlist_add_head(&cl
->hnode
, &clhash
->hash
[h
]);
653 EXPORT_SYMBOL(qdisc_class_hash_insert
);
655 void qdisc_class_hash_remove(struct Qdisc_class_hash
*clhash
,
656 struct Qdisc_class_common
*cl
)
658 hlist_del(&cl
->hnode
);
661 EXPORT_SYMBOL(qdisc_class_hash_remove
);
663 /* Allocate an unique handle from space managed by kernel
664 * Possible range is [8000-FFFF]:0000 (0x8000 values)
666 static u32
qdisc_alloc_handle(struct net_device
*dev
)
669 static u32 autohandle
= TC_H_MAKE(0x80000000U
, 0);
672 autohandle
+= TC_H_MAKE(0x10000U
, 0);
673 if (autohandle
== TC_H_MAKE(TC_H_ROOT
, 0))
674 autohandle
= TC_H_MAKE(0x80000000U
, 0);
675 if (!qdisc_lookup(dev
, autohandle
))
683 void qdisc_tree_decrease_qlen(struct Qdisc
*sch
, unsigned int n
)
685 const struct Qdisc_class_ops
*cops
;
691 while ((parentid
= sch
->parent
)) {
692 if (TC_H_MAJ(parentid
) == TC_H_MAJ(TC_H_INGRESS
))
695 sch
= qdisc_lookup(qdisc_dev(sch
), TC_H_MAJ(parentid
));
697 WARN_ON(parentid
!= TC_H_ROOT
);
700 cops
= sch
->ops
->cl_ops
;
701 if (cops
->qlen_notify
) {
702 cl
= cops
->get(sch
, parentid
);
703 cops
->qlen_notify(sch
, cl
);
709 EXPORT_SYMBOL(qdisc_tree_decrease_qlen
);
711 static void notify_and_destroy(struct net
*net
, struct sk_buff
*skb
,
712 struct nlmsghdr
*n
, u32 clid
,
713 struct Qdisc
*old
, struct Qdisc
*new)
716 qdisc_notify(net
, skb
, n
, clid
, old
, new);
722 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
725 * When appropriate send a netlink notification using 'skb'
728 * On success, destroy old qdisc.
731 static int qdisc_graft(struct net_device
*dev
, struct Qdisc
*parent
,
732 struct sk_buff
*skb
, struct nlmsghdr
*n
, u32 classid
,
733 struct Qdisc
*new, struct Qdisc
*old
)
735 struct Qdisc
*q
= old
;
736 struct net
*net
= dev_net(dev
);
739 if (parent
== NULL
) {
740 unsigned int i
, num_q
, ingress
;
743 num_q
= dev
->num_tx_queues
;
744 if ((q
&& q
->flags
& TCQ_F_INGRESS
) ||
745 (new && new->flags
& TCQ_F_INGRESS
)) {
748 if (!dev_ingress_queue(dev
))
752 if (dev
->flags
& IFF_UP
)
755 if (new && new->ops
->attach
)
758 for (i
= 0; i
< num_q
; i
++) {
759 struct netdev_queue
*dev_queue
= dev_ingress_queue(dev
);
762 dev_queue
= netdev_get_tx_queue(dev
, i
);
764 old
= dev_graft_qdisc(dev_queue
, new);
766 atomic_inc(&new->refcnt
);
774 notify_and_destroy(net
, skb
, n
, classid
,
776 if (new && !new->ops
->attach
)
777 atomic_inc(&new->refcnt
);
778 dev
->qdisc
= new ? : &noop_qdisc
;
780 if (new && new->ops
->attach
)
781 new->ops
->attach(new);
783 notify_and_destroy(net
, skb
, n
, classid
, old
, new);
786 if (dev
->flags
& IFF_UP
)
789 const struct Qdisc_class_ops
*cops
= parent
->ops
->cl_ops
;
792 if (cops
&& cops
->graft
) {
793 unsigned long cl
= cops
->get(parent
, classid
);
795 err
= cops
->graft(parent
, cl
, new, &old
);
796 cops
->put(parent
, cl
);
801 notify_and_destroy(net
, skb
, n
, classid
, old
, new);
806 /* lockdep annotation is needed for ingress; egress gets it only for name */
807 static struct lock_class_key qdisc_tx_lock
;
808 static struct lock_class_key qdisc_rx_lock
;
811 Allocate and initialize new qdisc.
813 Parameters are passed via opt.
816 static struct Qdisc
*
817 qdisc_create(struct net_device
*dev
, struct netdev_queue
*dev_queue
,
818 struct Qdisc
*p
, u32 parent
, u32 handle
,
819 struct nlattr
**tca
, int *errp
)
822 struct nlattr
*kind
= tca
[TCA_KIND
];
824 struct Qdisc_ops
*ops
;
825 struct qdisc_size_table
*stab
;
827 ops
= qdisc_lookup_ops(kind
);
828 #ifdef CONFIG_MODULES
829 if (ops
== NULL
&& kind
!= NULL
) {
831 if (nla_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
832 /* We dropped the RTNL semaphore in order to
833 * perform the module load. So, even if we
834 * succeeded in loading the module we have to
835 * tell the caller to replay the request. We
836 * indicate this using -EAGAIN.
837 * We replay the request because the device may
838 * go away in the mean time.
841 request_module("sch_%s", name
);
843 ops
= qdisc_lookup_ops(kind
);
845 /* We will try again qdisc_lookup_ops,
846 * so don't keep a reference.
848 module_put(ops
->owner
);
860 sch
= qdisc_alloc(dev_queue
, ops
);
866 sch
->parent
= parent
;
868 if (handle
== TC_H_INGRESS
) {
869 sch
->flags
|= TCQ_F_INGRESS
;
870 handle
= TC_H_MAKE(TC_H_INGRESS
, 0);
871 lockdep_set_class(qdisc_lock(sch
), &qdisc_rx_lock
);
874 handle
= qdisc_alloc_handle(dev
);
879 lockdep_set_class(qdisc_lock(sch
), &qdisc_tx_lock
);
880 if (!netif_is_multiqueue(dev
))
881 sch
->flags
|= TCQ_F_ONETXQUEUE
;
884 sch
->handle
= handle
;
886 if (!ops
->init
|| (err
= ops
->init(sch
, tca
[TCA_OPTIONS
])) == 0) {
888 stab
= qdisc_get_stab(tca
[TCA_STAB
]);
893 rcu_assign_pointer(sch
->stab
, stab
);
896 spinlock_t
*root_lock
;
899 if (sch
->flags
& TCQ_F_MQROOT
)
902 if ((sch
->parent
!= TC_H_ROOT
) &&
903 !(sch
->flags
& TCQ_F_INGRESS
) &&
904 (!p
|| !(p
->flags
& TCQ_F_MQROOT
)))
905 root_lock
= qdisc_root_sleeping_lock(sch
);
907 root_lock
= qdisc_lock(sch
);
909 err
= gen_new_estimator(&sch
->bstats
, &sch
->rate_est
,
910 root_lock
, tca
[TCA_RATE
]);
921 kfree((char *) sch
- sch
->padded
);
923 module_put(ops
->owner
);
930 * Any broken qdiscs that would require a ops->reset() here?
931 * The qdisc was never in action so it shouldn't be necessary.
933 qdisc_put_stab(rtnl_dereference(sch
->stab
));
939 static int qdisc_change(struct Qdisc
*sch
, struct nlattr
**tca
)
941 struct qdisc_size_table
*ostab
, *stab
= NULL
;
944 if (tca
[TCA_OPTIONS
]) {
945 if (sch
->ops
->change
== NULL
)
947 err
= sch
->ops
->change(sch
, tca
[TCA_OPTIONS
]);
953 stab
= qdisc_get_stab(tca
[TCA_STAB
]);
955 return PTR_ERR(stab
);
958 ostab
= rtnl_dereference(sch
->stab
);
959 rcu_assign_pointer(sch
->stab
, stab
);
960 qdisc_put_stab(ostab
);
963 /* NB: ignores errors from replace_estimator
964 because change can't be undone. */
965 if (sch
->flags
& TCQ_F_MQROOT
)
967 gen_replace_estimator(&sch
->bstats
, &sch
->rate_est
,
968 qdisc_root_sleeping_lock(sch
),
975 struct check_loop_arg
{
976 struct qdisc_walker w
;
981 static int check_loop_fn(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*w
);
983 static int check_loop(struct Qdisc
*q
, struct Qdisc
*p
, int depth
)
985 struct check_loop_arg arg
;
987 if (q
->ops
->cl_ops
== NULL
)
990 arg
.w
.stop
= arg
.w
.skip
= arg
.w
.count
= 0;
991 arg
.w
.fn
= check_loop_fn
;
994 q
->ops
->cl_ops
->walk(q
, &arg
.w
);
995 return arg
.w
.stop
? -ELOOP
: 0;
999 check_loop_fn(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*w
)
1002 const struct Qdisc_class_ops
*cops
= q
->ops
->cl_ops
;
1003 struct check_loop_arg
*arg
= (struct check_loop_arg
*)w
;
1005 leaf
= cops
->leaf(q
, cl
);
1007 if (leaf
== arg
->p
|| arg
->depth
> 7)
1009 return check_loop(leaf
, arg
->p
, arg
->depth
+ 1);
1018 static int tc_get_qdisc(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1020 struct net
*net
= sock_net(skb
->sk
);
1021 struct tcmsg
*tcm
= nlmsg_data(n
);
1022 struct nlattr
*tca
[TCA_MAX
+ 1];
1023 struct net_device
*dev
;
1025 struct Qdisc
*q
= NULL
;
1026 struct Qdisc
*p
= NULL
;
1029 if ((n
->nlmsg_type
!= RTM_GETQDISC
) && !netlink_capable(skb
, CAP_NET_ADMIN
))
1032 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1036 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1040 clid
= tcm
->tcm_parent
;
1042 if (clid
!= TC_H_ROOT
) {
1043 if (TC_H_MAJ(clid
) != TC_H_MAJ(TC_H_INGRESS
)) {
1044 p
= qdisc_lookup(dev
, TC_H_MAJ(clid
));
1047 q
= qdisc_leaf(p
, clid
);
1048 } else if (dev_ingress_queue(dev
)) {
1049 q
= dev_ingress_queue(dev
)->qdisc_sleeping
;
1057 if (tcm
->tcm_handle
&& q
->handle
!= tcm
->tcm_handle
)
1060 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1065 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1068 if (n
->nlmsg_type
== RTM_DELQDISC
) {
1073 err
= qdisc_graft(dev
, p
, skb
, n
, clid
, NULL
, q
);
1077 qdisc_notify(net
, skb
, n
, clid
, NULL
, q
);
1083 * enable/disable flow on qdisc.
1086 tc_qdisc_flow_control(struct net_device
*dev
, u32 tcm_handle
, int enable_flow
)
1089 struct __qdisc_change_req
{
1091 struct tc_prio_qopt data
;
1093 .attr
= {sizeof(struct __qdisc_change_req
), TCA_OPTIONS
},
1094 .data
= {3, {1, 2, 2, 2, 1, 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1}, 1}
1097 /* override flow bit */
1098 req
.data
.enable_flow
= enable_flow
;
1100 /* look up using tcm handle */
1101 q
= qdisc_lookup(dev
, tcm_handle
);
1103 /* call registered change function */
1105 if (q
->ops
->change(q
, &(req
.attr
)) != 0)
1106 pr_err("tc_qdisc_flow_control: qdisc change failed");
1109 EXPORT_SYMBOL(tc_qdisc_flow_control
);
1114 * Create/change qdisc.
1117 static int tc_modify_qdisc(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1119 struct net
*net
= sock_net(skb
->sk
);
1121 struct nlattr
*tca
[TCA_MAX
+ 1];
1122 struct net_device
*dev
;
1124 struct Qdisc
*q
, *p
;
1127 if (!netlink_capable(skb
, CAP_NET_ADMIN
))
1131 /* Reinit, just in case something touches this. */
1132 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1136 tcm
= nlmsg_data(n
);
1137 clid
= tcm
->tcm_parent
;
1140 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1146 if (clid
!= TC_H_ROOT
) {
1147 if (clid
!= TC_H_INGRESS
) {
1148 p
= qdisc_lookup(dev
, TC_H_MAJ(clid
));
1151 q
= qdisc_leaf(p
, clid
);
1152 } else if (dev_ingress_queue_create(dev
)) {
1153 q
= dev_ingress_queue(dev
)->qdisc_sleeping
;
1159 /* It may be default qdisc, ignore it */
1160 if (q
&& q
->handle
== 0)
1163 if (!q
|| !tcm
->tcm_handle
|| q
->handle
!= tcm
->tcm_handle
) {
1164 if (tcm
->tcm_handle
) {
1165 if (q
&& !(n
->nlmsg_flags
& NLM_F_REPLACE
))
1167 if (TC_H_MIN(tcm
->tcm_handle
))
1169 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1171 goto create_n_graft
;
1172 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1174 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1177 (p
&& check_loop(q
, p
, 0)))
1179 atomic_inc(&q
->refcnt
);
1183 goto create_n_graft
;
1185 /* This magic test requires explanation.
1187 * We know, that some child q is already
1188 * attached to this parent and have choice:
1189 * either to change it or to create/graft new one.
1191 * 1. We are allowed to create/graft only
1192 * if CREATE and REPLACE flags are set.
1194 * 2. If EXCL is set, requestor wanted to say,
1195 * that qdisc tcm_handle is not expected
1196 * to exist, so that we choose create/graft too.
1198 * 3. The last case is when no flags are set.
1199 * Alas, it is sort of hole in API, we
1200 * cannot decide what to do unambiguously.
1201 * For now we select create/graft, if
1202 * user gave KIND, which does not match existing.
1204 if ((n
->nlmsg_flags
& NLM_F_CREATE
) &&
1205 (n
->nlmsg_flags
& NLM_F_REPLACE
) &&
1206 ((n
->nlmsg_flags
& NLM_F_EXCL
) ||
1208 nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))))
1209 goto create_n_graft
;
1213 if (!tcm
->tcm_handle
)
1215 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1218 /* Change qdisc parameters */
1221 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1223 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1225 err
= qdisc_change(q
, tca
);
1227 qdisc_notify(net
, skb
, n
, clid
, NULL
, q
);
1231 if (!(n
->nlmsg_flags
& NLM_F_CREATE
))
1233 if (clid
== TC_H_INGRESS
) {
1234 if (dev_ingress_queue(dev
))
1235 q
= qdisc_create(dev
, dev_ingress_queue(dev
), p
,
1236 tcm
->tcm_parent
, tcm
->tcm_parent
,
1241 struct netdev_queue
*dev_queue
;
1243 if (p
&& p
->ops
->cl_ops
&& p
->ops
->cl_ops
->select_queue
)
1244 dev_queue
= p
->ops
->cl_ops
->select_queue(p
, tcm
);
1246 dev_queue
= p
->dev_queue
;
1248 dev_queue
= netdev_get_tx_queue(dev
, 0);
1250 q
= qdisc_create(dev
, dev_queue
, p
,
1251 tcm
->tcm_parent
, tcm
->tcm_handle
,
1261 err
= qdisc_graft(dev
, p
, skb
, n
, clid
, q
, NULL
);
1271 static int tc_fill_qdisc(struct sk_buff
*skb
, struct Qdisc
*q
, u32 clid
,
1272 u32 portid
, u32 seq
, u16 flags
, int event
)
1275 struct nlmsghdr
*nlh
;
1276 unsigned char *b
= skb_tail_pointer(skb
);
1278 struct qdisc_size_table
*stab
;
1280 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
1282 goto out_nlmsg_trim
;
1283 tcm
= nlmsg_data(nlh
);
1284 tcm
->tcm_family
= AF_UNSPEC
;
1287 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
1288 tcm
->tcm_parent
= clid
;
1289 tcm
->tcm_handle
= q
->handle
;
1290 tcm
->tcm_info
= atomic_read(&q
->refcnt
);
1291 if (nla_put_string(skb
, TCA_KIND
, q
->ops
->id
))
1292 goto nla_put_failure
;
1293 if (q
->ops
->dump
&& q
->ops
->dump(q
, skb
) < 0)
1294 goto nla_put_failure
;
1295 q
->qstats
.qlen
= q
->q
.qlen
;
1297 stab
= rtnl_dereference(q
->stab
);
1298 if (stab
&& qdisc_dump_stab(skb
, stab
) < 0)
1299 goto nla_put_failure
;
1301 if (gnet_stats_start_copy_compat(skb
, TCA_STATS2
, TCA_STATS
, TCA_XSTATS
,
1302 qdisc_root_sleeping_lock(q
), &d
) < 0)
1303 goto nla_put_failure
;
1305 if (q
->ops
->dump_stats
&& q
->ops
->dump_stats(q
, &d
) < 0)
1306 goto nla_put_failure
;
1308 if (gnet_stats_copy_basic(&d
, &q
->bstats
) < 0 ||
1309 gnet_stats_copy_rate_est(&d
, &q
->bstats
, &q
->rate_est
) < 0 ||
1310 gnet_stats_copy_queue(&d
, &q
->qstats
) < 0)
1311 goto nla_put_failure
;
1313 if (gnet_stats_finish_copy(&d
) < 0)
1314 goto nla_put_failure
;
1316 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
1325 static bool tc_qdisc_dump_ignore(struct Qdisc
*q
)
1327 return (q
->flags
& TCQ_F_BUILTIN
) ? true : false;
1330 static int qdisc_notify(struct net
*net
, struct sk_buff
*oskb
,
1331 struct nlmsghdr
*n
, u32 clid
,
1332 struct Qdisc
*old
, struct Qdisc
*new)
1334 struct sk_buff
*skb
;
1335 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1337 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1341 if (old
&& !tc_qdisc_dump_ignore(old
)) {
1342 if (tc_fill_qdisc(skb
, old
, clid
, portid
, n
->nlmsg_seq
,
1343 0, RTM_DELQDISC
) < 0)
1346 if (new && !tc_qdisc_dump_ignore(new)) {
1347 if (tc_fill_qdisc(skb
, new, clid
, portid
, n
->nlmsg_seq
,
1348 old
? NLM_F_REPLACE
: 0, RTM_NEWQDISC
) < 0)
1353 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1354 n
->nlmsg_flags
& NLM_F_ECHO
);
1361 static int tc_dump_qdisc_root(struct Qdisc
*root
, struct sk_buff
*skb
,
1362 struct netlink_callback
*cb
,
1363 int *q_idx_p
, int s_q_idx
)
1365 int ret
= 0, q_idx
= *q_idx_p
;
1372 if (q_idx
< s_q_idx
) {
1375 if (!tc_qdisc_dump_ignore(q
) &&
1376 tc_fill_qdisc(skb
, q
, q
->parent
, NETLINK_CB(cb
->skb
).portid
,
1377 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWQDISC
) <= 0)
1381 list_for_each_entry(q
, &root
->list
, list
) {
1382 if (q_idx
< s_q_idx
) {
1386 if (!tc_qdisc_dump_ignore(q
) &&
1387 tc_fill_qdisc(skb
, q
, q
->parent
, NETLINK_CB(cb
->skb
).portid
,
1388 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWQDISC
) <= 0)
1401 static int tc_dump_qdisc(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1403 struct net
*net
= sock_net(skb
->sk
);
1406 struct net_device
*dev
;
1408 s_idx
= cb
->args
[0];
1409 s_q_idx
= q_idx
= cb
->args
[1];
1413 for_each_netdev_rcu(net
, dev
) {
1414 struct netdev_queue
*dev_queue
;
1422 if (tc_dump_qdisc_root(dev
->qdisc
, skb
, cb
, &q_idx
, s_q_idx
) < 0)
1425 dev_queue
= dev_ingress_queue(dev
);
1427 tc_dump_qdisc_root(dev_queue
->qdisc_sleeping
, skb
, cb
,
1428 &q_idx
, s_q_idx
) < 0)
1439 cb
->args
[1] = q_idx
;
1446 /************************************************
1447 * Traffic classes manipulation. *
1448 ************************************************/
1452 static int tc_ctl_tclass(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1454 struct net
*net
= sock_net(skb
->sk
);
1455 struct tcmsg
*tcm
= nlmsg_data(n
);
1456 struct nlattr
*tca
[TCA_MAX
+ 1];
1457 struct net_device
*dev
;
1458 struct Qdisc
*q
= NULL
;
1459 const struct Qdisc_class_ops
*cops
;
1460 unsigned long cl
= 0;
1461 unsigned long new_cl
;
1467 if ((n
->nlmsg_type
!= RTM_GETTCLASS
) && !netlink_capable(skb
, CAP_NET_ADMIN
))
1470 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1474 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1479 parent == TC_H_UNSPEC - unspecified parent.
1480 parent == TC_H_ROOT - class is root, which has no parent.
1481 parent == X:0 - parent is root class.
1482 parent == X:Y - parent is a node in hierarchy.
1483 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1485 handle == 0:0 - generate handle from kernel pool.
1486 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1487 handle == X:Y - clear.
1488 handle == X:0 - root class.
1491 /* Step 1. Determine qdisc handle X:0 */
1493 portid
= tcm
->tcm_parent
;
1494 clid
= tcm
->tcm_handle
;
1495 qid
= TC_H_MAJ(clid
);
1497 if (portid
!= TC_H_ROOT
) {
1498 u32 qid1
= TC_H_MAJ(portid
);
1501 /* If both majors are known, they must be identical. */
1506 } else if (qid
== 0)
1507 qid
= dev
->qdisc
->handle
;
1509 /* Now qid is genuine qdisc handle consistent
1510 * both with parent and child.
1512 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1515 portid
= TC_H_MAKE(qid
, portid
);
1518 qid
= dev
->qdisc
->handle
;
1521 /* OK. Locate qdisc */
1522 q
= qdisc_lookup(dev
, qid
);
1526 /* An check that it supports classes */
1527 cops
= q
->ops
->cl_ops
;
1531 /* Now try to get class */
1533 if (portid
== TC_H_ROOT
)
1536 clid
= TC_H_MAKE(qid
, clid
);
1539 cl
= cops
->get(q
, clid
);
1543 if (n
->nlmsg_type
!= RTM_NEWTCLASS
||
1544 !(n
->nlmsg_flags
& NLM_F_CREATE
))
1547 switch (n
->nlmsg_type
) {
1550 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1556 err
= cops
->delete(q
, cl
);
1558 tclass_notify(net
, skb
, n
, q
, cl
, RTM_DELTCLASS
);
1561 err
= tclass_notify(net
, skb
, n
, q
, cl
, RTM_NEWTCLASS
);
1572 err
= cops
->change(q
, clid
, portid
, tca
, &new_cl
);
1574 tclass_notify(net
, skb
, n
, q
, new_cl
, RTM_NEWTCLASS
);
1584 static int tc_fill_tclass(struct sk_buff
*skb
, struct Qdisc
*q
,
1586 u32 portid
, u32 seq
, u16 flags
, int event
)
1589 struct nlmsghdr
*nlh
;
1590 unsigned char *b
= skb_tail_pointer(skb
);
1592 const struct Qdisc_class_ops
*cl_ops
= q
->ops
->cl_ops
;
1594 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
1596 goto out_nlmsg_trim
;
1597 tcm
= nlmsg_data(nlh
);
1598 tcm
->tcm_family
= AF_UNSPEC
;
1601 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
1602 tcm
->tcm_parent
= q
->handle
;
1603 tcm
->tcm_handle
= q
->handle
;
1605 if (nla_put_string(skb
, TCA_KIND
, q
->ops
->id
))
1606 goto nla_put_failure
;
1607 if (cl_ops
->dump
&& cl_ops
->dump(q
, cl
, skb
, tcm
) < 0)
1608 goto nla_put_failure
;
1610 if (gnet_stats_start_copy_compat(skb
, TCA_STATS2
, TCA_STATS
, TCA_XSTATS
,
1611 qdisc_root_sleeping_lock(q
), &d
) < 0)
1612 goto nla_put_failure
;
1614 if (cl_ops
->dump_stats
&& cl_ops
->dump_stats(q
, cl
, &d
) < 0)
1615 goto nla_put_failure
;
1617 if (gnet_stats_finish_copy(&d
) < 0)
1618 goto nla_put_failure
;
1620 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
1629 static int tclass_notify(struct net
*net
, struct sk_buff
*oskb
,
1630 struct nlmsghdr
*n
, struct Qdisc
*q
,
1631 unsigned long cl
, int event
)
1633 struct sk_buff
*skb
;
1634 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1636 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1640 if (tc_fill_tclass(skb
, q
, cl
, portid
, n
->nlmsg_seq
, 0, event
) < 0) {
1645 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1646 n
->nlmsg_flags
& NLM_F_ECHO
);
1649 struct qdisc_dump_args
{
1650 struct qdisc_walker w
;
1651 struct sk_buff
*skb
;
1652 struct netlink_callback
*cb
;
1655 static int qdisc_class_dump(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*arg
)
1657 struct qdisc_dump_args
*a
= (struct qdisc_dump_args
*)arg
;
1659 return tc_fill_tclass(a
->skb
, q
, cl
, NETLINK_CB(a
->cb
->skb
).portid
,
1660 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTCLASS
);
1663 static int tc_dump_tclass_qdisc(struct Qdisc
*q
, struct sk_buff
*skb
,
1664 struct tcmsg
*tcm
, struct netlink_callback
*cb
,
1667 struct qdisc_dump_args arg
;
1669 if (tc_qdisc_dump_ignore(q
) ||
1670 *t_p
< s_t
|| !q
->ops
->cl_ops
||
1672 TC_H_MAJ(tcm
->tcm_parent
) != q
->handle
)) {
1677 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
1678 arg
.w
.fn
= qdisc_class_dump
;
1682 arg
.w
.skip
= cb
->args
[1];
1684 q
->ops
->cl_ops
->walk(q
, &arg
.w
);
1685 cb
->args
[1] = arg
.w
.count
;
1692 static int tc_dump_tclass_root(struct Qdisc
*root
, struct sk_buff
*skb
,
1693 struct tcmsg
*tcm
, struct netlink_callback
*cb
,
1701 if (tc_dump_tclass_qdisc(root
, skb
, tcm
, cb
, t_p
, s_t
) < 0)
1704 list_for_each_entry(q
, &root
->list
, list
) {
1705 if (tc_dump_tclass_qdisc(q
, skb
, tcm
, cb
, t_p
, s_t
) < 0)
1712 static int tc_dump_tclass(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1714 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
1715 struct net
*net
= sock_net(skb
->sk
);
1716 struct netdev_queue
*dev_queue
;
1717 struct net_device
*dev
;
1720 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
1722 dev
= dev_get_by_index(net
, tcm
->tcm_ifindex
);
1729 if (tc_dump_tclass_root(dev
->qdisc
, skb
, tcm
, cb
, &t
, s_t
) < 0)
1732 dev_queue
= dev_ingress_queue(dev
);
1734 tc_dump_tclass_root(dev_queue
->qdisc_sleeping
, skb
, tcm
, cb
,
1745 /* Main classifier routine: scans classifier chain attached
1746 * to this qdisc, (optionally) tests for protocol and asks
1747 * specific classifiers.
1749 int tc_classify_compat(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
1750 struct tcf_result
*res
)
1752 __be16 protocol
= skb
->protocol
;
1755 for (; tp
; tp
= tp
->next
) {
1756 if (tp
->protocol
!= protocol
&&
1757 tp
->protocol
!= htons(ETH_P_ALL
))
1759 err
= tp
->classify(skb
, tp
, res
);
1762 #ifdef CONFIG_NET_CLS_ACT
1763 if (err
!= TC_ACT_RECLASSIFY
&& skb
->tc_verd
)
1764 skb
->tc_verd
= SET_TC_VERD(skb
->tc_verd
, 0);
1771 EXPORT_SYMBOL(tc_classify_compat
);
1773 int tc_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
1774 struct tcf_result
*res
)
1777 #ifdef CONFIG_NET_CLS_ACT
1778 const struct tcf_proto
*otp
= tp
;
1782 err
= tc_classify_compat(skb
, tp
, res
);
1783 #ifdef CONFIG_NET_CLS_ACT
1784 if (err
== TC_ACT_RECLASSIFY
) {
1785 u32 verd
= G_TC_VERD(skb
->tc_verd
);
1788 if (verd
++ >= MAX_REC_LOOP
) {
1789 net_notice_ratelimited("%s: packet reclassify loop rule prio %u protocol %02x\n",
1792 ntohs(tp
->protocol
));
1795 skb
->tc_verd
= SET_TC_VERD(skb
->tc_verd
, verd
);
1801 EXPORT_SYMBOL(tc_classify
);
1803 void tcf_destroy(struct tcf_proto
*tp
)
1805 tp
->ops
->destroy(tp
);
1806 module_put(tp
->ops
->owner
);
1810 void tcf_destroy_chain(struct tcf_proto
**fl
)
1812 struct tcf_proto
*tp
;
1814 while ((tp
= *fl
) != NULL
) {
1819 EXPORT_SYMBOL(tcf_destroy_chain
);
1821 #ifdef CONFIG_PROC_FS
1822 static int psched_show(struct seq_file
*seq
, void *v
)
1826 hrtimer_get_res(CLOCK_MONOTONIC
, &ts
);
1827 seq_printf(seq
, "%08x %08x %08x %08x\n",
1828 (u32
)NSEC_PER_USEC
, (u32
)PSCHED_TICKS2NS(1),
1830 (u32
)NSEC_PER_SEC
/(u32
)ktime_to_ns(timespec_to_ktime(ts
)));
1835 static int psched_open(struct inode
*inode
, struct file
*file
)
1837 return single_open(file
, psched_show
, NULL
);
1840 static const struct file_operations psched_fops
= {
1841 .owner
= THIS_MODULE
,
1842 .open
= psched_open
,
1844 .llseek
= seq_lseek
,
1845 .release
= single_release
,
1848 static int __net_init
psched_net_init(struct net
*net
)
1850 struct proc_dir_entry
*e
;
1852 e
= proc_create("psched", 0, net
->proc_net
, &psched_fops
);
1859 static void __net_exit
psched_net_exit(struct net
*net
)
1861 remove_proc_entry("psched", net
->proc_net
);
1864 static int __net_init
psched_net_init(struct net
*net
)
1869 static void __net_exit
psched_net_exit(struct net
*net
)
1874 static struct pernet_operations psched_net_ops
= {
1875 .init
= psched_net_init
,
1876 .exit
= psched_net_exit
,
1879 static int __init
pktsched_init(void)
1883 err
= register_pernet_subsys(&psched_net_ops
);
1885 pr_err("pktsched_init: "
1886 "cannot initialize per netns operations\n");
1890 register_qdisc(&pfifo_qdisc_ops
);
1891 register_qdisc(&bfifo_qdisc_ops
);
1892 register_qdisc(&pfifo_head_drop_qdisc_ops
);
1893 register_qdisc(&mq_qdisc_ops
);
1895 rtnl_register(PF_UNSPEC
, RTM_NEWQDISC
, tc_modify_qdisc
, NULL
, NULL
);
1896 rtnl_register(PF_UNSPEC
, RTM_DELQDISC
, tc_get_qdisc
, NULL
, NULL
);
1897 rtnl_register(PF_UNSPEC
, RTM_GETQDISC
, tc_get_qdisc
, tc_dump_qdisc
, NULL
);
1898 rtnl_register(PF_UNSPEC
, RTM_NEWTCLASS
, tc_ctl_tclass
, NULL
, NULL
);
1899 rtnl_register(PF_UNSPEC
, RTM_DELTCLASS
, tc_ctl_tclass
, NULL
, NULL
);
1900 rtnl_register(PF_UNSPEC
, RTM_GETTCLASS
, tc_ctl_tclass
, tc_dump_tclass
, NULL
);
1905 subsys_initcall(pktsched_init
);