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/unregister 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 /* Get default qdisc if not otherwise specified */
204 void qdisc_get_default(char *name
, size_t len
)
206 read_lock(&qdisc_mod_lock
);
207 strlcpy(name
, default_qdisc_ops
->id
, len
);
208 read_unlock(&qdisc_mod_lock
);
211 static struct Qdisc_ops
*qdisc_lookup_default(const char *name
)
213 struct Qdisc_ops
*q
= NULL
;
215 for (q
= qdisc_base
; q
; q
= q
->next
) {
216 if (!strcmp(name
, q
->id
)) {
217 if (!try_module_get(q
->owner
))
226 /* Set new default qdisc to use */
227 int qdisc_set_default(const char *name
)
229 const struct Qdisc_ops
*ops
;
231 if (!capable(CAP_NET_ADMIN
))
234 write_lock(&qdisc_mod_lock
);
235 ops
= qdisc_lookup_default(name
);
237 /* Not found, drop lock and try to load module */
238 write_unlock(&qdisc_mod_lock
);
239 request_module("sch_%s", name
);
240 write_lock(&qdisc_mod_lock
);
242 ops
= qdisc_lookup_default(name
);
246 /* Set new default */
247 module_put(default_qdisc_ops
->owner
);
248 default_qdisc_ops
= ops
;
250 write_unlock(&qdisc_mod_lock
);
252 return ops
? 0 : -ENOENT
;
255 /* We know handle. Find qdisc among all qdisc's attached to device
256 * (root qdisc, all its children, children of children etc.)
257 * Note: caller either uses rtnl or rcu_read_lock()
260 static struct Qdisc
*qdisc_match_from_root(struct Qdisc
*root
, u32 handle
)
264 if (!(root
->flags
& TCQ_F_BUILTIN
) &&
265 root
->handle
== handle
)
268 list_for_each_entry_rcu(q
, &root
->list
, list
) {
269 if (q
->handle
== handle
)
275 void qdisc_list_add(struct Qdisc
*q
)
277 if ((q
->parent
!= TC_H_ROOT
) && !(q
->flags
& TCQ_F_INGRESS
)) {
278 struct Qdisc
*root
= qdisc_dev(q
)->qdisc
;
280 WARN_ON_ONCE(root
== &noop_qdisc
);
282 list_add_tail_rcu(&q
->list
, &root
->list
);
285 EXPORT_SYMBOL(qdisc_list_add
);
287 void qdisc_list_del(struct Qdisc
*q
)
289 if ((q
->parent
!= TC_H_ROOT
) && !(q
->flags
& TCQ_F_INGRESS
)) {
291 list_del_rcu(&q
->list
);
294 EXPORT_SYMBOL(qdisc_list_del
);
296 struct Qdisc
*qdisc_lookup(struct net_device
*dev
, u32 handle
)
300 q
= qdisc_match_from_root(dev
->qdisc
, handle
);
304 if (dev_ingress_queue(dev
))
305 q
= qdisc_match_from_root(
306 dev_ingress_queue(dev
)->qdisc_sleeping
,
312 static struct Qdisc
*qdisc_leaf(struct Qdisc
*p
, u32 classid
)
316 const struct Qdisc_class_ops
*cops
= p
->ops
->cl_ops
;
320 cl
= cops
->get(p
, classid
);
324 leaf
= cops
->leaf(p
, cl
);
329 /* Find queueing discipline by name */
331 static struct Qdisc_ops
*qdisc_lookup_ops(struct nlattr
*kind
)
333 struct Qdisc_ops
*q
= NULL
;
336 read_lock(&qdisc_mod_lock
);
337 for (q
= qdisc_base
; q
; q
= q
->next
) {
338 if (nla_strcmp(kind
, q
->id
) == 0) {
339 if (!try_module_get(q
->owner
))
344 read_unlock(&qdisc_mod_lock
);
349 /* The linklayer setting were not transferred from iproute2, in older
350 * versions, and the rate tables lookup systems have been dropped in
351 * the kernel. To keep backward compatible with older iproute2 tc
352 * utils, we detect the linklayer setting by detecting if the rate
353 * table were modified.
355 * For linklayer ATM table entries, the rate table will be aligned to
356 * 48 bytes, thus some table entries will contain the same value. The
357 * mpu (min packet unit) is also encoded into the old rate table, thus
358 * starting from the mpu, we find low and high table entries for
359 * mapping this cell. If these entries contain the same value, when
360 * the rate tables have been modified for linklayer ATM.
362 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
363 * and then roundup to the next cell, calc the table entry one below,
366 static __u8
__detect_linklayer(struct tc_ratespec
*r
, __u32
*rtab
)
368 int low
= roundup(r
->mpu
, 48);
369 int high
= roundup(low
+1, 48);
370 int cell_low
= low
>> r
->cell_log
;
371 int cell_high
= (high
>> r
->cell_log
) - 1;
373 /* rtab is too inaccurate at rates > 100Mbit/s */
374 if ((r
->rate
> (100000000/8)) || (rtab
[0] == 0)) {
375 pr_debug("TC linklayer: Giving up ATM detection\n");
376 return TC_LINKLAYER_ETHERNET
;
379 if ((cell_high
> cell_low
) && (cell_high
< 256)
380 && (rtab
[cell_low
] == rtab
[cell_high
])) {
381 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
382 cell_low
, cell_high
, rtab
[cell_high
]);
383 return TC_LINKLAYER_ATM
;
385 return TC_LINKLAYER_ETHERNET
;
388 static struct qdisc_rate_table
*qdisc_rtab_list
;
390 struct qdisc_rate_table
*qdisc_get_rtab(struct tc_ratespec
*r
, struct nlattr
*tab
)
392 struct qdisc_rate_table
*rtab
;
394 if (tab
== NULL
|| r
->rate
== 0 || r
->cell_log
== 0 ||
395 nla_len(tab
) != TC_RTAB_SIZE
)
398 for (rtab
= qdisc_rtab_list
; rtab
; rtab
= rtab
->next
) {
399 if (!memcmp(&rtab
->rate
, r
, sizeof(struct tc_ratespec
)) &&
400 !memcmp(&rtab
->data
, nla_data(tab
), 1024)) {
406 rtab
= kmalloc(sizeof(*rtab
), GFP_KERNEL
);
410 memcpy(rtab
->data
, nla_data(tab
), 1024);
411 if (r
->linklayer
== TC_LINKLAYER_UNAWARE
)
412 r
->linklayer
= __detect_linklayer(r
, rtab
->data
);
413 rtab
->next
= qdisc_rtab_list
;
414 qdisc_rtab_list
= rtab
;
418 EXPORT_SYMBOL(qdisc_get_rtab
);
420 void qdisc_put_rtab(struct qdisc_rate_table
*tab
)
422 struct qdisc_rate_table
*rtab
, **rtabp
;
424 if (!tab
|| --tab
->refcnt
)
427 for (rtabp
= &qdisc_rtab_list
;
428 (rtab
= *rtabp
) != NULL
;
429 rtabp
= &rtab
->next
) {
437 EXPORT_SYMBOL(qdisc_put_rtab
);
439 static LIST_HEAD(qdisc_stab_list
);
440 static DEFINE_SPINLOCK(qdisc_stab_lock
);
442 static const struct nla_policy stab_policy
[TCA_STAB_MAX
+ 1] = {
443 [TCA_STAB_BASE
] = { .len
= sizeof(struct tc_sizespec
) },
444 [TCA_STAB_DATA
] = { .type
= NLA_BINARY
},
447 static struct qdisc_size_table
*qdisc_get_stab(struct nlattr
*opt
)
449 struct nlattr
*tb
[TCA_STAB_MAX
+ 1];
450 struct qdisc_size_table
*stab
;
451 struct tc_sizespec
*s
;
452 unsigned int tsize
= 0;
456 err
= nla_parse_nested(tb
, TCA_STAB_MAX
, opt
, stab_policy
);
459 if (!tb
[TCA_STAB_BASE
])
460 return ERR_PTR(-EINVAL
);
462 s
= nla_data(tb
[TCA_STAB_BASE
]);
465 if (!tb
[TCA_STAB_DATA
])
466 return ERR_PTR(-EINVAL
);
467 tab
= nla_data(tb
[TCA_STAB_DATA
]);
468 tsize
= nla_len(tb
[TCA_STAB_DATA
]) / sizeof(u16
);
471 if (tsize
!= s
->tsize
|| (!tab
&& tsize
> 0))
472 return ERR_PTR(-EINVAL
);
474 spin_lock(&qdisc_stab_lock
);
476 list_for_each_entry(stab
, &qdisc_stab_list
, list
) {
477 if (memcmp(&stab
->szopts
, s
, sizeof(*s
)))
479 if (tsize
> 0 && memcmp(stab
->data
, tab
, tsize
* sizeof(u16
)))
482 spin_unlock(&qdisc_stab_lock
);
486 spin_unlock(&qdisc_stab_lock
);
488 stab
= kmalloc(sizeof(*stab
) + tsize
* sizeof(u16
), GFP_KERNEL
);
490 return ERR_PTR(-ENOMEM
);
495 memcpy(stab
->data
, tab
, tsize
* sizeof(u16
));
497 spin_lock(&qdisc_stab_lock
);
498 list_add_tail(&stab
->list
, &qdisc_stab_list
);
499 spin_unlock(&qdisc_stab_lock
);
504 static void stab_kfree_rcu(struct rcu_head
*head
)
506 kfree(container_of(head
, struct qdisc_size_table
, rcu
));
509 void qdisc_put_stab(struct qdisc_size_table
*tab
)
514 spin_lock(&qdisc_stab_lock
);
516 if (--tab
->refcnt
== 0) {
517 list_del(&tab
->list
);
518 call_rcu_bh(&tab
->rcu
, stab_kfree_rcu
);
521 spin_unlock(&qdisc_stab_lock
);
523 EXPORT_SYMBOL(qdisc_put_stab
);
525 static int qdisc_dump_stab(struct sk_buff
*skb
, struct qdisc_size_table
*stab
)
529 nest
= nla_nest_start(skb
, TCA_STAB
);
531 goto nla_put_failure
;
532 if (nla_put(skb
, TCA_STAB_BASE
, sizeof(stab
->szopts
), &stab
->szopts
))
533 goto nla_put_failure
;
534 nla_nest_end(skb
, nest
);
542 void __qdisc_calculate_pkt_len(struct sk_buff
*skb
, const struct qdisc_size_table
*stab
)
546 pkt_len
= skb
->len
+ stab
->szopts
.overhead
;
547 if (unlikely(!stab
->szopts
.tsize
))
550 slot
= pkt_len
+ stab
->szopts
.cell_align
;
551 if (unlikely(slot
< 0))
554 slot
>>= stab
->szopts
.cell_log
;
555 if (likely(slot
< stab
->szopts
.tsize
))
556 pkt_len
= stab
->data
[slot
];
558 pkt_len
= stab
->data
[stab
->szopts
.tsize
- 1] *
559 (slot
/ stab
->szopts
.tsize
) +
560 stab
->data
[slot
% stab
->szopts
.tsize
];
562 pkt_len
<<= stab
->szopts
.size_log
;
564 if (unlikely(pkt_len
< 1))
566 qdisc_skb_cb(skb
)->pkt_len
= pkt_len
;
568 EXPORT_SYMBOL(__qdisc_calculate_pkt_len
);
570 void qdisc_warn_nonwc(const char *txt
, struct Qdisc
*qdisc
)
572 if (!(qdisc
->flags
& TCQ_F_WARN_NONWC
)) {
573 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
574 txt
, qdisc
->ops
->id
, qdisc
->handle
>> 16);
575 qdisc
->flags
|= TCQ_F_WARN_NONWC
;
578 EXPORT_SYMBOL(qdisc_warn_nonwc
);
580 static enum hrtimer_restart
qdisc_watchdog(struct hrtimer
*timer
)
582 struct qdisc_watchdog
*wd
= container_of(timer
, struct qdisc_watchdog
,
586 qdisc_unthrottled(wd
->qdisc
);
587 __netif_schedule(qdisc_root(wd
->qdisc
));
590 return HRTIMER_NORESTART
;
593 void qdisc_watchdog_init(struct qdisc_watchdog
*wd
, struct Qdisc
*qdisc
)
595 hrtimer_init(&wd
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS_PINNED
);
596 wd
->timer
.function
= qdisc_watchdog
;
599 EXPORT_SYMBOL(qdisc_watchdog_init
);
601 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog
*wd
, u64 expires
, bool throttle
)
603 if (test_bit(__QDISC_STATE_DEACTIVATED
,
604 &qdisc_root_sleeping(wd
->qdisc
)->state
))
608 qdisc_throttled(wd
->qdisc
);
610 hrtimer_start(&wd
->timer
,
611 ns_to_ktime(expires
),
612 HRTIMER_MODE_ABS_PINNED
);
614 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns
);
616 void qdisc_watchdog_cancel(struct qdisc_watchdog
*wd
)
618 hrtimer_cancel(&wd
->timer
);
619 qdisc_unthrottled(wd
->qdisc
);
621 EXPORT_SYMBOL(qdisc_watchdog_cancel
);
623 static struct hlist_head
*qdisc_class_hash_alloc(unsigned int n
)
625 unsigned int size
= n
* sizeof(struct hlist_head
), i
;
626 struct hlist_head
*h
;
628 if (size
<= PAGE_SIZE
)
629 h
= kmalloc(size
, GFP_KERNEL
);
631 h
= (struct hlist_head
*)
632 __get_free_pages(GFP_KERNEL
, get_order(size
));
635 for (i
= 0; i
< n
; i
++)
636 INIT_HLIST_HEAD(&h
[i
]);
641 static void qdisc_class_hash_free(struct hlist_head
*h
, unsigned int n
)
643 unsigned int size
= n
* sizeof(struct hlist_head
);
645 if (size
<= PAGE_SIZE
)
648 free_pages((unsigned long)h
, get_order(size
));
651 void qdisc_class_hash_grow(struct Qdisc
*sch
, struct Qdisc_class_hash
*clhash
)
653 struct Qdisc_class_common
*cl
;
654 struct hlist_node
*next
;
655 struct hlist_head
*nhash
, *ohash
;
656 unsigned int nsize
, nmask
, osize
;
659 /* Rehash when load factor exceeds 0.75 */
660 if (clhash
->hashelems
* 4 <= clhash
->hashsize
* 3)
662 nsize
= clhash
->hashsize
* 2;
664 nhash
= qdisc_class_hash_alloc(nsize
);
668 ohash
= clhash
->hash
;
669 osize
= clhash
->hashsize
;
672 for (i
= 0; i
< osize
; i
++) {
673 hlist_for_each_entry_safe(cl
, next
, &ohash
[i
], hnode
) {
674 h
= qdisc_class_hash(cl
->classid
, nmask
);
675 hlist_add_head(&cl
->hnode
, &nhash
[h
]);
678 clhash
->hash
= nhash
;
679 clhash
->hashsize
= nsize
;
680 clhash
->hashmask
= nmask
;
681 sch_tree_unlock(sch
);
683 qdisc_class_hash_free(ohash
, osize
);
685 EXPORT_SYMBOL(qdisc_class_hash_grow
);
687 int qdisc_class_hash_init(struct Qdisc_class_hash
*clhash
)
689 unsigned int size
= 4;
691 clhash
->hash
= qdisc_class_hash_alloc(size
);
692 if (clhash
->hash
== NULL
)
694 clhash
->hashsize
= size
;
695 clhash
->hashmask
= size
- 1;
696 clhash
->hashelems
= 0;
699 EXPORT_SYMBOL(qdisc_class_hash_init
);
701 void qdisc_class_hash_destroy(struct Qdisc_class_hash
*clhash
)
703 qdisc_class_hash_free(clhash
->hash
, clhash
->hashsize
);
705 EXPORT_SYMBOL(qdisc_class_hash_destroy
);
707 void qdisc_class_hash_insert(struct Qdisc_class_hash
*clhash
,
708 struct Qdisc_class_common
*cl
)
712 INIT_HLIST_NODE(&cl
->hnode
);
713 h
= qdisc_class_hash(cl
->classid
, clhash
->hashmask
);
714 hlist_add_head(&cl
->hnode
, &clhash
->hash
[h
]);
717 EXPORT_SYMBOL(qdisc_class_hash_insert
);
719 void qdisc_class_hash_remove(struct Qdisc_class_hash
*clhash
,
720 struct Qdisc_class_common
*cl
)
722 hlist_del(&cl
->hnode
);
725 EXPORT_SYMBOL(qdisc_class_hash_remove
);
727 /* Allocate an unique handle from space managed by kernel
728 * Possible range is [8000-FFFF]:0000 (0x8000 values)
730 static u32
qdisc_alloc_handle(struct net_device
*dev
)
733 static u32 autohandle
= TC_H_MAKE(0x80000000U
, 0);
736 autohandle
+= TC_H_MAKE(0x10000U
, 0);
737 if (autohandle
== TC_H_MAKE(TC_H_ROOT
, 0))
738 autohandle
= TC_H_MAKE(0x80000000U
, 0);
739 if (!qdisc_lookup(dev
, autohandle
))
747 void qdisc_tree_reduce_backlog(struct Qdisc
*sch
, unsigned int n
,
750 const struct Qdisc_class_ops
*cops
;
755 if (n
== 0 && len
== 0)
757 drops
= max_t(int, n
, 0);
759 while ((parentid
= sch
->parent
)) {
760 if (TC_H_MAJ(parentid
) == TC_H_MAJ(TC_H_INGRESS
))
763 if (sch
->flags
& TCQ_F_NOPARENT
)
765 /* TODO: perform the search on a per txq basis */
766 sch
= qdisc_lookup(qdisc_dev(sch
), TC_H_MAJ(parentid
));
768 WARN_ON_ONCE(parentid
!= TC_H_ROOT
);
771 cops
= sch
->ops
->cl_ops
;
772 if (cops
->qlen_notify
) {
773 cl
= cops
->get(sch
, parentid
);
774 cops
->qlen_notify(sch
, cl
);
778 sch
->qstats
.backlog
-= len
;
779 __qdisc_qstats_drop(sch
, drops
);
783 EXPORT_SYMBOL(qdisc_tree_reduce_backlog
);
785 static void notify_and_destroy(struct net
*net
, struct sk_buff
*skb
,
786 struct nlmsghdr
*n
, u32 clid
,
787 struct Qdisc
*old
, struct Qdisc
*new)
790 qdisc_notify(net
, skb
, n
, clid
, old
, new);
796 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
799 * When appropriate send a netlink notification using 'skb'
802 * On success, destroy old qdisc.
805 static int qdisc_graft(struct net_device
*dev
, struct Qdisc
*parent
,
806 struct sk_buff
*skb
, struct nlmsghdr
*n
, u32 classid
,
807 struct Qdisc
*new, struct Qdisc
*old
)
809 struct Qdisc
*q
= old
;
810 struct net
*net
= dev_net(dev
);
813 if (parent
== NULL
) {
814 unsigned int i
, num_q
, ingress
;
817 num_q
= dev
->num_tx_queues
;
818 if ((q
&& q
->flags
& TCQ_F_INGRESS
) ||
819 (new && new->flags
& TCQ_F_INGRESS
)) {
822 if (!dev_ingress_queue(dev
))
826 if (dev
->flags
& IFF_UP
)
829 if (new && new->ops
->attach
)
832 for (i
= 0; i
< num_q
; i
++) {
833 struct netdev_queue
*dev_queue
= dev_ingress_queue(dev
);
836 dev_queue
= netdev_get_tx_queue(dev
, i
);
838 old
= dev_graft_qdisc(dev_queue
, new);
840 atomic_inc(&new->refcnt
);
848 notify_and_destroy(net
, skb
, n
, classid
,
850 if (new && !new->ops
->attach
)
851 atomic_inc(&new->refcnt
);
852 dev
->qdisc
= new ? : &noop_qdisc
;
854 if (new && new->ops
->attach
)
855 new->ops
->attach(new);
857 notify_and_destroy(net
, skb
, n
, classid
, old
, new);
860 if (dev
->flags
& IFF_UP
)
863 const struct Qdisc_class_ops
*cops
= parent
->ops
->cl_ops
;
866 if (cops
&& cops
->graft
) {
867 unsigned long cl
= cops
->get(parent
, classid
);
869 err
= cops
->graft(parent
, cl
, new, &old
);
870 cops
->put(parent
, cl
);
875 notify_and_destroy(net
, skb
, n
, classid
, old
, new);
880 /* lockdep annotation is needed for ingress; egress gets it only for name */
881 static struct lock_class_key qdisc_tx_lock
;
882 static struct lock_class_key qdisc_rx_lock
;
885 Allocate and initialize new qdisc.
887 Parameters are passed via opt.
890 static struct Qdisc
*
891 qdisc_create(struct net_device
*dev
, struct netdev_queue
*dev_queue
,
892 struct Qdisc
*p
, u32 parent
, u32 handle
,
893 struct nlattr
**tca
, int *errp
)
896 struct nlattr
*kind
= tca
[TCA_KIND
];
898 struct Qdisc_ops
*ops
;
899 struct qdisc_size_table
*stab
;
901 ops
= qdisc_lookup_ops(kind
);
902 #ifdef CONFIG_MODULES
903 if (ops
== NULL
&& kind
!= NULL
) {
905 if (nla_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
906 /* We dropped the RTNL semaphore in order to
907 * perform the module load. So, even if we
908 * succeeded in loading the module we have to
909 * tell the caller to replay the request. We
910 * indicate this using -EAGAIN.
911 * We replay the request because the device may
912 * go away in the mean time.
915 request_module("sch_%s", name
);
917 ops
= qdisc_lookup_ops(kind
);
919 /* We will try again qdisc_lookup_ops,
920 * so don't keep a reference.
922 module_put(ops
->owner
);
934 sch
= qdisc_alloc(dev_queue
, ops
);
940 sch
->parent
= parent
;
942 if (handle
== TC_H_INGRESS
) {
943 sch
->flags
|= TCQ_F_INGRESS
;
944 handle
= TC_H_MAKE(TC_H_INGRESS
, 0);
945 lockdep_set_class(qdisc_lock(sch
), &qdisc_rx_lock
);
948 handle
= qdisc_alloc_handle(dev
);
953 lockdep_set_class(qdisc_lock(sch
), &qdisc_tx_lock
);
954 if (!netif_is_multiqueue(dev
))
955 sch
->flags
|= TCQ_F_ONETXQUEUE
;
958 sch
->handle
= handle
;
960 if (!ops
->init
|| (err
= ops
->init(sch
, tca
[TCA_OPTIONS
])) == 0) {
961 if (qdisc_is_percpu_stats(sch
)) {
963 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu
);
964 if (!sch
->cpu_bstats
)
967 sch
->cpu_qstats
= alloc_percpu(struct gnet_stats_queue
);
968 if (!sch
->cpu_qstats
)
973 stab
= qdisc_get_stab(tca
[TCA_STAB
]);
978 rcu_assign_pointer(sch
->stab
, stab
);
981 spinlock_t
*root_lock
;
984 if (sch
->flags
& TCQ_F_MQROOT
)
987 if ((sch
->parent
!= TC_H_ROOT
) &&
988 !(sch
->flags
& TCQ_F_INGRESS
) &&
989 (!p
|| !(p
->flags
& TCQ_F_MQROOT
)))
990 root_lock
= qdisc_root_sleeping_lock(sch
);
992 root_lock
= qdisc_lock(sch
);
994 err
= gen_new_estimator(&sch
->bstats
,
1003 qdisc_list_add(sch
);
1009 kfree((char *) sch
- sch
->padded
);
1011 module_put(ops
->owner
);
1017 free_percpu(sch
->cpu_bstats
);
1018 free_percpu(sch
->cpu_qstats
);
1020 * Any broken qdiscs that would require a ops->reset() here?
1021 * The qdisc was never in action so it shouldn't be necessary.
1023 qdisc_put_stab(rtnl_dereference(sch
->stab
));
1029 static int qdisc_change(struct Qdisc
*sch
, struct nlattr
**tca
)
1031 struct qdisc_size_table
*ostab
, *stab
= NULL
;
1034 if (tca
[TCA_OPTIONS
]) {
1035 if (sch
->ops
->change
== NULL
)
1037 err
= sch
->ops
->change(sch
, tca
[TCA_OPTIONS
]);
1042 if (tca
[TCA_STAB
]) {
1043 stab
= qdisc_get_stab(tca
[TCA_STAB
]);
1045 return PTR_ERR(stab
);
1048 ostab
= rtnl_dereference(sch
->stab
);
1049 rcu_assign_pointer(sch
->stab
, stab
);
1050 qdisc_put_stab(ostab
);
1052 if (tca
[TCA_RATE
]) {
1053 /* NB: ignores errors from replace_estimator
1054 because change can't be undone. */
1055 if (sch
->flags
& TCQ_F_MQROOT
)
1057 gen_replace_estimator(&sch
->bstats
,
1060 qdisc_root_sleeping_lock(sch
),
1067 struct check_loop_arg
{
1068 struct qdisc_walker w
;
1073 static int check_loop_fn(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*w
);
1075 static int check_loop(struct Qdisc
*q
, struct Qdisc
*p
, int depth
)
1077 struct check_loop_arg arg
;
1079 if (q
->ops
->cl_ops
== NULL
)
1082 arg
.w
.stop
= arg
.w
.skip
= arg
.w
.count
= 0;
1083 arg
.w
.fn
= check_loop_fn
;
1086 q
->ops
->cl_ops
->walk(q
, &arg
.w
);
1087 return arg
.w
.stop
? -ELOOP
: 0;
1091 check_loop_fn(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*w
)
1094 const struct Qdisc_class_ops
*cops
= q
->ops
->cl_ops
;
1095 struct check_loop_arg
*arg
= (struct check_loop_arg
*)w
;
1097 leaf
= cops
->leaf(q
, cl
);
1099 if (leaf
== arg
->p
|| arg
->depth
> 7)
1101 return check_loop(leaf
, arg
->p
, arg
->depth
+ 1);
1110 static int tc_get_qdisc(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1112 struct net
*net
= sock_net(skb
->sk
);
1113 struct tcmsg
*tcm
= nlmsg_data(n
);
1114 struct nlattr
*tca
[TCA_MAX
+ 1];
1115 struct net_device
*dev
;
1117 struct Qdisc
*q
= NULL
;
1118 struct Qdisc
*p
= NULL
;
1121 if ((n
->nlmsg_type
!= RTM_GETQDISC
) &&
1122 !netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1125 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1129 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1133 clid
= tcm
->tcm_parent
;
1135 if (clid
!= TC_H_ROOT
) {
1136 if (TC_H_MAJ(clid
) != TC_H_MAJ(TC_H_INGRESS
)) {
1137 p
= qdisc_lookup(dev
, TC_H_MAJ(clid
));
1140 q
= qdisc_leaf(p
, clid
);
1141 } else if (dev_ingress_queue(dev
)) {
1142 q
= dev_ingress_queue(dev
)->qdisc_sleeping
;
1150 if (tcm
->tcm_handle
&& q
->handle
!= tcm
->tcm_handle
)
1153 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1158 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1161 if (n
->nlmsg_type
== RTM_DELQDISC
) {
1166 err
= qdisc_graft(dev
, p
, skb
, n
, clid
, NULL
, q
);
1170 qdisc_notify(net
, skb
, n
, clid
, NULL
, q
);
1176 * Create/change qdisc.
1179 static int tc_modify_qdisc(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1181 struct net
*net
= sock_net(skb
->sk
);
1183 struct nlattr
*tca
[TCA_MAX
+ 1];
1184 struct net_device
*dev
;
1186 struct Qdisc
*q
, *p
;
1189 if (!netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1193 /* Reinit, just in case something touches this. */
1194 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1198 tcm
= nlmsg_data(n
);
1199 clid
= tcm
->tcm_parent
;
1202 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1208 if (clid
!= TC_H_ROOT
) {
1209 if (clid
!= TC_H_INGRESS
) {
1210 p
= qdisc_lookup(dev
, TC_H_MAJ(clid
));
1213 q
= qdisc_leaf(p
, clid
);
1214 } else if (dev_ingress_queue_create(dev
)) {
1215 q
= dev_ingress_queue(dev
)->qdisc_sleeping
;
1221 /* It may be default qdisc, ignore it */
1222 if (q
&& q
->handle
== 0)
1225 if (!q
|| !tcm
->tcm_handle
|| q
->handle
!= tcm
->tcm_handle
) {
1226 if (tcm
->tcm_handle
) {
1227 if (q
&& !(n
->nlmsg_flags
& NLM_F_REPLACE
))
1229 if (TC_H_MIN(tcm
->tcm_handle
))
1231 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1233 goto create_n_graft
;
1234 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1236 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1239 (p
&& check_loop(q
, p
, 0)))
1241 atomic_inc(&q
->refcnt
);
1245 goto create_n_graft
;
1247 /* This magic test requires explanation.
1249 * We know, that some child q is already
1250 * attached to this parent and have choice:
1251 * either to change it or to create/graft new one.
1253 * 1. We are allowed to create/graft only
1254 * if CREATE and REPLACE flags are set.
1256 * 2. If EXCL is set, requestor wanted to say,
1257 * that qdisc tcm_handle is not expected
1258 * to exist, so that we choose create/graft too.
1260 * 3. The last case is when no flags are set.
1261 * Alas, it is sort of hole in API, we
1262 * cannot decide what to do unambiguously.
1263 * For now we select create/graft, if
1264 * user gave KIND, which does not match existing.
1266 if ((n
->nlmsg_flags
& NLM_F_CREATE
) &&
1267 (n
->nlmsg_flags
& NLM_F_REPLACE
) &&
1268 ((n
->nlmsg_flags
& NLM_F_EXCL
) ||
1270 nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))))
1271 goto create_n_graft
;
1275 if (!tcm
->tcm_handle
)
1277 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1280 /* Change qdisc parameters */
1283 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1285 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1287 err
= qdisc_change(q
, tca
);
1289 qdisc_notify(net
, skb
, n
, clid
, NULL
, q
);
1293 if (!(n
->nlmsg_flags
& NLM_F_CREATE
))
1295 if (clid
== TC_H_INGRESS
) {
1296 if (dev_ingress_queue(dev
))
1297 q
= qdisc_create(dev
, dev_ingress_queue(dev
), p
,
1298 tcm
->tcm_parent
, tcm
->tcm_parent
,
1303 struct netdev_queue
*dev_queue
;
1305 if (p
&& p
->ops
->cl_ops
&& p
->ops
->cl_ops
->select_queue
)
1306 dev_queue
= p
->ops
->cl_ops
->select_queue(p
, tcm
);
1308 dev_queue
= p
->dev_queue
;
1310 dev_queue
= netdev_get_tx_queue(dev
, 0);
1312 q
= qdisc_create(dev
, dev_queue
, p
,
1313 tcm
->tcm_parent
, tcm
->tcm_handle
,
1323 err
= qdisc_graft(dev
, p
, skb
, n
, clid
, q
, NULL
);
1333 static int tc_fill_qdisc(struct sk_buff
*skb
, struct Qdisc
*q
, u32 clid
,
1334 u32 portid
, u32 seq
, u16 flags
, int event
)
1336 struct gnet_stats_basic_cpu __percpu
*cpu_bstats
= NULL
;
1337 struct gnet_stats_queue __percpu
*cpu_qstats
= NULL
;
1339 struct nlmsghdr
*nlh
;
1340 unsigned char *b
= skb_tail_pointer(skb
);
1342 struct qdisc_size_table
*stab
;
1346 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
1348 goto out_nlmsg_trim
;
1349 tcm
= nlmsg_data(nlh
);
1350 tcm
->tcm_family
= AF_UNSPEC
;
1353 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
1354 tcm
->tcm_parent
= clid
;
1355 tcm
->tcm_handle
= q
->handle
;
1356 tcm
->tcm_info
= atomic_read(&q
->refcnt
);
1357 if (nla_put_string(skb
, TCA_KIND
, q
->ops
->id
))
1358 goto nla_put_failure
;
1359 if (q
->ops
->dump
&& q
->ops
->dump(q
, skb
) < 0)
1360 goto nla_put_failure
;
1363 stab
= rtnl_dereference(q
->stab
);
1364 if (stab
&& qdisc_dump_stab(skb
, stab
) < 0)
1365 goto nla_put_failure
;
1367 if (gnet_stats_start_copy_compat(skb
, TCA_STATS2
, TCA_STATS
, TCA_XSTATS
,
1368 qdisc_root_sleeping_lock(q
), &d
) < 0)
1369 goto nla_put_failure
;
1371 if (q
->ops
->dump_stats
&& q
->ops
->dump_stats(q
, &d
) < 0)
1372 goto nla_put_failure
;
1374 if (qdisc_is_percpu_stats(q
)) {
1375 cpu_bstats
= q
->cpu_bstats
;
1376 cpu_qstats
= q
->cpu_qstats
;
1379 if (gnet_stats_copy_basic(&d
, cpu_bstats
, &q
->bstats
) < 0 ||
1380 gnet_stats_copy_rate_est(&d
, &q
->bstats
, &q
->rate_est
) < 0 ||
1381 gnet_stats_copy_queue(&d
, cpu_qstats
, &q
->qstats
, qlen
) < 0)
1382 goto nla_put_failure
;
1384 if (gnet_stats_finish_copy(&d
) < 0)
1385 goto nla_put_failure
;
1387 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
1396 static bool tc_qdisc_dump_ignore(struct Qdisc
*q
)
1398 return (q
->flags
& TCQ_F_BUILTIN
) ? true : false;
1401 static int qdisc_notify(struct net
*net
, struct sk_buff
*oskb
,
1402 struct nlmsghdr
*n
, u32 clid
,
1403 struct Qdisc
*old
, struct Qdisc
*new)
1405 struct sk_buff
*skb
;
1406 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1408 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1412 if (old
&& !tc_qdisc_dump_ignore(old
)) {
1413 if (tc_fill_qdisc(skb
, old
, clid
, portid
, n
->nlmsg_seq
,
1414 0, RTM_DELQDISC
) < 0)
1417 if (new && !tc_qdisc_dump_ignore(new)) {
1418 if (tc_fill_qdisc(skb
, new, clid
, portid
, n
->nlmsg_seq
,
1419 old
? NLM_F_REPLACE
: 0, RTM_NEWQDISC
) < 0)
1424 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1425 n
->nlmsg_flags
& NLM_F_ECHO
);
1432 static int tc_dump_qdisc_root(struct Qdisc
*root
, struct sk_buff
*skb
,
1433 struct netlink_callback
*cb
,
1434 int *q_idx_p
, int s_q_idx
)
1436 int ret
= 0, q_idx
= *q_idx_p
;
1443 if (q_idx
< s_q_idx
) {
1446 if (!tc_qdisc_dump_ignore(q
) &&
1447 tc_fill_qdisc(skb
, q
, q
->parent
, NETLINK_CB(cb
->skb
).portid
,
1448 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWQDISC
) <= 0)
1452 list_for_each_entry(q
, &root
->list
, list
) {
1453 if (q_idx
< s_q_idx
) {
1457 if (!tc_qdisc_dump_ignore(q
) &&
1458 tc_fill_qdisc(skb
, q
, q
->parent
, NETLINK_CB(cb
->skb
).portid
,
1459 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWQDISC
) <= 0)
1472 static int tc_dump_qdisc(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1474 struct net
*net
= sock_net(skb
->sk
);
1477 struct net_device
*dev
;
1479 s_idx
= cb
->args
[0];
1480 s_q_idx
= q_idx
= cb
->args
[1];
1484 for_each_netdev(net
, dev
) {
1485 struct netdev_queue
*dev_queue
;
1493 if (tc_dump_qdisc_root(dev
->qdisc
, skb
, cb
, &q_idx
, s_q_idx
) < 0)
1496 dev_queue
= dev_ingress_queue(dev
);
1498 tc_dump_qdisc_root(dev_queue
->qdisc_sleeping
, skb
, cb
,
1499 &q_idx
, s_q_idx
) < 0)
1508 cb
->args
[1] = q_idx
;
1515 /************************************************
1516 * Traffic classes manipulation. *
1517 ************************************************/
1521 static int tc_ctl_tclass(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1523 struct net
*net
= sock_net(skb
->sk
);
1524 struct tcmsg
*tcm
= nlmsg_data(n
);
1525 struct nlattr
*tca
[TCA_MAX
+ 1];
1526 struct net_device
*dev
;
1527 struct Qdisc
*q
= NULL
;
1528 const struct Qdisc_class_ops
*cops
;
1529 unsigned long cl
= 0;
1530 unsigned long new_cl
;
1536 if ((n
->nlmsg_type
!= RTM_GETTCLASS
) &&
1537 !netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1540 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1544 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1549 parent == TC_H_UNSPEC - unspecified parent.
1550 parent == TC_H_ROOT - class is root, which has no parent.
1551 parent == X:0 - parent is root class.
1552 parent == X:Y - parent is a node in hierarchy.
1553 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1555 handle == 0:0 - generate handle from kernel pool.
1556 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1557 handle == X:Y - clear.
1558 handle == X:0 - root class.
1561 /* Step 1. Determine qdisc handle X:0 */
1563 portid
= tcm
->tcm_parent
;
1564 clid
= tcm
->tcm_handle
;
1565 qid
= TC_H_MAJ(clid
);
1567 if (portid
!= TC_H_ROOT
) {
1568 u32 qid1
= TC_H_MAJ(portid
);
1571 /* If both majors are known, they must be identical. */
1576 } else if (qid
== 0)
1577 qid
= dev
->qdisc
->handle
;
1579 /* Now qid is genuine qdisc handle consistent
1580 * both with parent and child.
1582 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1585 portid
= TC_H_MAKE(qid
, portid
);
1588 qid
= dev
->qdisc
->handle
;
1591 /* OK. Locate qdisc */
1592 q
= qdisc_lookup(dev
, qid
);
1596 /* An check that it supports classes */
1597 cops
= q
->ops
->cl_ops
;
1601 /* Now try to get class */
1603 if (portid
== TC_H_ROOT
)
1606 clid
= TC_H_MAKE(qid
, clid
);
1609 cl
= cops
->get(q
, clid
);
1613 if (n
->nlmsg_type
!= RTM_NEWTCLASS
||
1614 !(n
->nlmsg_flags
& NLM_F_CREATE
))
1617 switch (n
->nlmsg_type
) {
1620 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1626 err
= cops
->delete(q
, cl
);
1628 tclass_notify(net
, skb
, n
, q
, cl
, RTM_DELTCLASS
);
1631 err
= tclass_notify(net
, skb
, n
, q
, cl
, RTM_NEWTCLASS
);
1642 err
= cops
->change(q
, clid
, portid
, tca
, &new_cl
);
1644 tclass_notify(net
, skb
, n
, q
, new_cl
, RTM_NEWTCLASS
);
1654 static int tc_fill_tclass(struct sk_buff
*skb
, struct Qdisc
*q
,
1656 u32 portid
, u32 seq
, u16 flags
, int event
)
1659 struct nlmsghdr
*nlh
;
1660 unsigned char *b
= skb_tail_pointer(skb
);
1662 const struct Qdisc_class_ops
*cl_ops
= q
->ops
->cl_ops
;
1665 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
1667 goto out_nlmsg_trim
;
1668 tcm
= nlmsg_data(nlh
);
1669 tcm
->tcm_family
= AF_UNSPEC
;
1672 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
1673 tcm
->tcm_parent
= q
->handle
;
1674 tcm
->tcm_handle
= q
->handle
;
1676 if (nla_put_string(skb
, TCA_KIND
, q
->ops
->id
))
1677 goto nla_put_failure
;
1678 if (cl_ops
->dump
&& cl_ops
->dump(q
, cl
, skb
, tcm
) < 0)
1679 goto nla_put_failure
;
1681 if (gnet_stats_start_copy_compat(skb
, TCA_STATS2
, TCA_STATS
, TCA_XSTATS
,
1682 qdisc_root_sleeping_lock(q
), &d
) < 0)
1683 goto nla_put_failure
;
1685 if (cl_ops
->dump_stats
&& cl_ops
->dump_stats(q
, cl
, &d
) < 0)
1686 goto nla_put_failure
;
1688 if (gnet_stats_finish_copy(&d
) < 0)
1689 goto nla_put_failure
;
1691 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
1700 static int tclass_notify(struct net
*net
, struct sk_buff
*oskb
,
1701 struct nlmsghdr
*n
, struct Qdisc
*q
,
1702 unsigned long cl
, int event
)
1704 struct sk_buff
*skb
;
1705 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1707 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1711 if (tc_fill_tclass(skb
, q
, cl
, portid
, n
->nlmsg_seq
, 0, event
) < 0) {
1716 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1717 n
->nlmsg_flags
& NLM_F_ECHO
);
1720 struct qdisc_dump_args
{
1721 struct qdisc_walker w
;
1722 struct sk_buff
*skb
;
1723 struct netlink_callback
*cb
;
1726 static int qdisc_class_dump(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*arg
)
1728 struct qdisc_dump_args
*a
= (struct qdisc_dump_args
*)arg
;
1730 return tc_fill_tclass(a
->skb
, q
, cl
, NETLINK_CB(a
->cb
->skb
).portid
,
1731 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTCLASS
);
1734 static int tc_dump_tclass_qdisc(struct Qdisc
*q
, struct sk_buff
*skb
,
1735 struct tcmsg
*tcm
, struct netlink_callback
*cb
,
1738 struct qdisc_dump_args arg
;
1740 if (tc_qdisc_dump_ignore(q
) ||
1741 *t_p
< s_t
|| !q
->ops
->cl_ops
||
1743 TC_H_MAJ(tcm
->tcm_parent
) != q
->handle
)) {
1748 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
1749 arg
.w
.fn
= qdisc_class_dump
;
1753 arg
.w
.skip
= cb
->args
[1];
1755 q
->ops
->cl_ops
->walk(q
, &arg
.w
);
1756 cb
->args
[1] = arg
.w
.count
;
1763 static int tc_dump_tclass_root(struct Qdisc
*root
, struct sk_buff
*skb
,
1764 struct tcmsg
*tcm
, struct netlink_callback
*cb
,
1772 if (tc_dump_tclass_qdisc(root
, skb
, tcm
, cb
, t_p
, s_t
) < 0)
1775 list_for_each_entry(q
, &root
->list
, list
) {
1776 if (tc_dump_tclass_qdisc(q
, skb
, tcm
, cb
, t_p
, s_t
) < 0)
1783 static int tc_dump_tclass(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1785 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
1786 struct net
*net
= sock_net(skb
->sk
);
1787 struct netdev_queue
*dev_queue
;
1788 struct net_device
*dev
;
1791 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
1793 dev
= dev_get_by_index(net
, tcm
->tcm_ifindex
);
1800 if (tc_dump_tclass_root(dev
->qdisc
, skb
, tcm
, cb
, &t
, s_t
) < 0)
1803 dev_queue
= dev_ingress_queue(dev
);
1805 tc_dump_tclass_root(dev_queue
->qdisc_sleeping
, skb
, tcm
, cb
,
1816 /* Main classifier routine: scans classifier chain attached
1817 * to this qdisc, (optionally) tests for protocol and asks
1818 * specific classifiers.
1820 int tc_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
1821 struct tcf_result
*res
, bool compat_mode
)
1823 __be16 protocol
= tc_skb_protocol(skb
);
1824 #ifdef CONFIG_NET_CLS_ACT
1825 const struct tcf_proto
*old_tp
= tp
;
1830 for (; tp
; tp
= rcu_dereference_bh(tp
->next
)) {
1833 if (tp
->protocol
!= protocol
&&
1834 tp
->protocol
!= htons(ETH_P_ALL
))
1837 err
= tp
->classify(skb
, tp
, res
);
1838 #ifdef CONFIG_NET_CLS_ACT
1839 if (unlikely(err
== TC_ACT_RECLASSIFY
&& !compat_mode
))
1846 return TC_ACT_UNSPEC
; /* signal: continue lookup */
1847 #ifdef CONFIG_NET_CLS_ACT
1849 if (unlikely(limit
++ >= MAX_REC_LOOP
)) {
1850 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
1851 tp
->q
->ops
->id
, tp
->prio
& 0xffff,
1852 ntohs(tp
->protocol
));
1857 protocol
= tc_skb_protocol(skb
);
1861 EXPORT_SYMBOL(tc_classify
);
1863 bool tcf_destroy(struct tcf_proto
*tp
, bool force
)
1865 if (tp
->ops
->destroy(tp
, force
)) {
1866 module_put(tp
->ops
->owner
);
1874 void tcf_destroy_chain(struct tcf_proto __rcu
**fl
)
1876 struct tcf_proto
*tp
;
1878 while ((tp
= rtnl_dereference(*fl
)) != NULL
) {
1879 RCU_INIT_POINTER(*fl
, tp
->next
);
1880 tcf_destroy(tp
, true);
1883 EXPORT_SYMBOL(tcf_destroy_chain
);
1885 #ifdef CONFIG_PROC_FS
1886 static int psched_show(struct seq_file
*seq
, void *v
)
1888 seq_printf(seq
, "%08x %08x %08x %08x\n",
1889 (u32
)NSEC_PER_USEC
, (u32
)PSCHED_TICKS2NS(1),
1891 (u32
)NSEC_PER_SEC
/ hrtimer_resolution
);
1896 static int psched_open(struct inode
*inode
, struct file
*file
)
1898 return single_open(file
, psched_show
, NULL
);
1901 static const struct file_operations psched_fops
= {
1902 .owner
= THIS_MODULE
,
1903 .open
= psched_open
,
1905 .llseek
= seq_lseek
,
1906 .release
= single_release
,
1909 static int __net_init
psched_net_init(struct net
*net
)
1911 struct proc_dir_entry
*e
;
1913 e
= proc_create("psched", 0, net
->proc_net
, &psched_fops
);
1920 static void __net_exit
psched_net_exit(struct net
*net
)
1922 remove_proc_entry("psched", net
->proc_net
);
1925 static int __net_init
psched_net_init(struct net
*net
)
1930 static void __net_exit
psched_net_exit(struct net
*net
)
1935 static struct pernet_operations psched_net_ops
= {
1936 .init
= psched_net_init
,
1937 .exit
= psched_net_exit
,
1940 static int __init
pktsched_init(void)
1944 err
= register_pernet_subsys(&psched_net_ops
);
1946 pr_err("pktsched_init: "
1947 "cannot initialize per netns operations\n");
1951 register_qdisc(&pfifo_fast_ops
);
1952 register_qdisc(&pfifo_qdisc_ops
);
1953 register_qdisc(&bfifo_qdisc_ops
);
1954 register_qdisc(&pfifo_head_drop_qdisc_ops
);
1955 register_qdisc(&mq_qdisc_ops
);
1956 register_qdisc(&noqueue_qdisc_ops
);
1958 rtnl_register(PF_UNSPEC
, RTM_NEWQDISC
, tc_modify_qdisc
, NULL
, NULL
);
1959 rtnl_register(PF_UNSPEC
, RTM_DELQDISC
, tc_get_qdisc
, NULL
, NULL
);
1960 rtnl_register(PF_UNSPEC
, RTM_GETQDISC
, tc_get_qdisc
, tc_dump_qdisc
, NULL
);
1961 rtnl_register(PF_UNSPEC
, RTM_NEWTCLASS
, tc_ctl_tclass
, NULL
, NULL
);
1962 rtnl_register(PF_UNSPEC
, RTM_DELTCLASS
, tc_ctl_tclass
, NULL
, NULL
);
1963 rtnl_register(PF_UNSPEC
, RTM_GETTCLASS
, tc_ctl_tclass
, tc_dump_tclass
, NULL
);
1968 subsys_initcall(pktsched_init
);