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>
32 #include <linux/hashtable.h>
34 #include <net/net_namespace.h>
36 #include <net/netlink.h>
37 #include <net/pkt_sched.h>
39 static int qdisc_notify(struct net
*net
, struct sk_buff
*oskb
,
40 struct nlmsghdr
*n
, u32 clid
,
41 struct Qdisc
*old
, struct Qdisc
*new);
42 static int tclass_notify(struct net
*net
, struct sk_buff
*oskb
,
43 struct nlmsghdr
*n
, struct Qdisc
*q
,
44 unsigned long cl
, int event
);
51 This file consists of two interrelated parts:
53 1. queueing disciplines manager frontend.
54 2. traffic classes manager frontend.
56 Generally, queueing discipline ("qdisc") is a black box,
57 which is able to enqueue packets and to dequeue them (when
58 device is ready to send something) in order and at times
59 determined by algorithm hidden in it.
61 qdisc's are divided to two categories:
62 - "queues", which have no internal structure visible from outside.
63 - "schedulers", which split all the packets to "traffic classes",
64 using "packet classifiers" (look at cls_api.c)
66 In turn, classes may have child qdiscs (as rule, queues)
67 attached to them etc. etc. etc.
69 The goal of the routines in this file is to translate
70 information supplied by user in the form of handles
71 to more intelligible for kernel form, to make some sanity
72 checks and part of work, which is common to all qdiscs
73 and to provide rtnetlink notifications.
75 All real intelligent work is done inside qdisc modules.
79 Every discipline has two major routines: enqueue and dequeue.
83 dequeue usually returns a skb to send. It is allowed to return NULL,
84 but it does not mean that queue is empty, it just means that
85 discipline does not want to send anything this time.
86 Queue is really empty if q->q.qlen == 0.
87 For complicated disciplines with multiple queues q->q is not
88 real packet queue, but however q->q.qlen must be valid.
92 enqueue returns 0, if packet was enqueued successfully.
93 If packet (this one or another one) was dropped, it returns
95 NET_XMIT_DROP - this packet dropped
96 Expected action: do not backoff, but wait until queue will clear.
97 NET_XMIT_CN - probably this packet enqueued, but another one dropped.
98 Expected action: backoff or ignore
104 like dequeue but without removing a packet from the queue
108 returns qdisc to initial state: purge all buffers, clear all
109 timers, counters (except for statistics) etc.
113 initializes newly created qdisc.
117 destroys resources allocated by init and during lifetime of qdisc.
121 changes qdisc parameters.
124 /* Protects list of registered TC modules. It is pure SMP lock. */
125 static DEFINE_RWLOCK(qdisc_mod_lock
);
128 /************************************************
129 * Queueing disciplines manipulation. *
130 ************************************************/
133 /* The list of all installed queueing disciplines. */
135 static struct Qdisc_ops
*qdisc_base
;
137 /* Register/unregister queueing discipline */
139 int register_qdisc(struct Qdisc_ops
*qops
)
141 struct Qdisc_ops
*q
, **qp
;
144 write_lock(&qdisc_mod_lock
);
145 for (qp
= &qdisc_base
; (q
= *qp
) != NULL
; qp
= &q
->next
)
146 if (!strcmp(qops
->id
, q
->id
))
149 if (qops
->enqueue
== NULL
)
150 qops
->enqueue
= noop_qdisc_ops
.enqueue
;
151 if (qops
->peek
== NULL
) {
152 if (qops
->dequeue
== NULL
)
153 qops
->peek
= noop_qdisc_ops
.peek
;
157 if (qops
->dequeue
== NULL
)
158 qops
->dequeue
= noop_qdisc_ops
.dequeue
;
161 const struct Qdisc_class_ops
*cops
= qops
->cl_ops
;
163 if (!(cops
->get
&& cops
->put
&& cops
->walk
&& cops
->leaf
))
166 if (cops
->tcf_chain
&& !(cops
->bind_tcf
&& cops
->unbind_tcf
))
174 write_unlock(&qdisc_mod_lock
);
181 EXPORT_SYMBOL(register_qdisc
);
183 int unregister_qdisc(struct Qdisc_ops
*qops
)
185 struct Qdisc_ops
*q
, **qp
;
188 write_lock(&qdisc_mod_lock
);
189 for (qp
= &qdisc_base
; (q
= *qp
) != NULL
; qp
= &q
->next
)
197 write_unlock(&qdisc_mod_lock
);
200 EXPORT_SYMBOL(unregister_qdisc
);
202 /* Get default qdisc if not otherwise specified */
203 void qdisc_get_default(char *name
, size_t len
)
205 read_lock(&qdisc_mod_lock
);
206 strlcpy(name
, default_qdisc_ops
->id
, len
);
207 read_unlock(&qdisc_mod_lock
);
210 static struct Qdisc_ops
*qdisc_lookup_default(const char *name
)
212 struct Qdisc_ops
*q
= NULL
;
214 for (q
= qdisc_base
; q
; q
= q
->next
) {
215 if (!strcmp(name
, q
->id
)) {
216 if (!try_module_get(q
->owner
))
225 /* Set new default qdisc to use */
226 int qdisc_set_default(const char *name
)
228 const struct Qdisc_ops
*ops
;
230 if (!capable(CAP_NET_ADMIN
))
233 write_lock(&qdisc_mod_lock
);
234 ops
= qdisc_lookup_default(name
);
236 /* Not found, drop lock and try to load module */
237 write_unlock(&qdisc_mod_lock
);
238 request_module("sch_%s", name
);
239 write_lock(&qdisc_mod_lock
);
241 ops
= qdisc_lookup_default(name
);
245 /* Set new default */
246 module_put(default_qdisc_ops
->owner
);
247 default_qdisc_ops
= ops
;
249 write_unlock(&qdisc_mod_lock
);
251 return ops
? 0 : -ENOENT
;
254 /* We know handle. Find qdisc among all qdisc's attached to device
255 * (root qdisc, all its children, children of children etc.)
256 * Note: caller either uses rtnl or rcu_read_lock()
259 static struct Qdisc
*qdisc_match_from_root(struct Qdisc
*root
, u32 handle
)
263 if (!qdisc_dev(root
))
264 return (root
->handle
== handle
? root
: NULL
);
266 if (!(root
->flags
& TCQ_F_BUILTIN
) &&
267 root
->handle
== handle
)
270 hash_for_each_possible_rcu(qdisc_dev(root
)->qdisc_hash
, q
, hash
, handle
) {
271 if (q
->handle
== handle
)
277 void qdisc_hash_add(struct Qdisc
*q
)
279 if ((q
->parent
!= TC_H_ROOT
) && !(q
->flags
& TCQ_F_INGRESS
)) {
280 struct Qdisc
*root
= qdisc_dev(q
)->qdisc
;
282 WARN_ON_ONCE(root
== &noop_qdisc
);
284 hash_add_rcu(qdisc_dev(q
)->qdisc_hash
, &q
->hash
, q
->handle
);
287 EXPORT_SYMBOL(qdisc_hash_add
);
289 void qdisc_hash_del(struct Qdisc
*q
)
291 if ((q
->parent
!= TC_H_ROOT
) && !(q
->flags
& TCQ_F_INGRESS
)) {
293 hash_del_rcu(&q
->hash
);
296 EXPORT_SYMBOL(qdisc_hash_del
);
298 struct Qdisc
*qdisc_lookup(struct net_device
*dev
, u32 handle
)
302 q
= qdisc_match_from_root(dev
->qdisc
, handle
);
306 if (dev_ingress_queue(dev
))
307 q
= qdisc_match_from_root(
308 dev_ingress_queue(dev
)->qdisc_sleeping
,
314 static struct Qdisc
*qdisc_leaf(struct Qdisc
*p
, u32 classid
)
318 const struct Qdisc_class_ops
*cops
= p
->ops
->cl_ops
;
322 cl
= cops
->get(p
, classid
);
326 leaf
= cops
->leaf(p
, cl
);
331 /* Find queueing discipline by name */
333 static struct Qdisc_ops
*qdisc_lookup_ops(struct nlattr
*kind
)
335 struct Qdisc_ops
*q
= NULL
;
338 read_lock(&qdisc_mod_lock
);
339 for (q
= qdisc_base
; q
; q
= q
->next
) {
340 if (nla_strcmp(kind
, q
->id
) == 0) {
341 if (!try_module_get(q
->owner
))
346 read_unlock(&qdisc_mod_lock
);
351 /* The linklayer setting were not transferred from iproute2, in older
352 * versions, and the rate tables lookup systems have been dropped in
353 * the kernel. To keep backward compatible with older iproute2 tc
354 * utils, we detect the linklayer setting by detecting if the rate
355 * table were modified.
357 * For linklayer ATM table entries, the rate table will be aligned to
358 * 48 bytes, thus some table entries will contain the same value. The
359 * mpu (min packet unit) is also encoded into the old rate table, thus
360 * starting from the mpu, we find low and high table entries for
361 * mapping this cell. If these entries contain the same value, when
362 * the rate tables have been modified for linklayer ATM.
364 * This is done by rounding mpu to the nearest 48 bytes cell/entry,
365 * and then roundup to the next cell, calc the table entry one below,
368 static __u8
__detect_linklayer(struct tc_ratespec
*r
, __u32
*rtab
)
370 int low
= roundup(r
->mpu
, 48);
371 int high
= roundup(low
+1, 48);
372 int cell_low
= low
>> r
->cell_log
;
373 int cell_high
= (high
>> r
->cell_log
) - 1;
375 /* rtab is too inaccurate at rates > 100Mbit/s */
376 if ((r
->rate
> (100000000/8)) || (rtab
[0] == 0)) {
377 pr_debug("TC linklayer: Giving up ATM detection\n");
378 return TC_LINKLAYER_ETHERNET
;
381 if ((cell_high
> cell_low
) && (cell_high
< 256)
382 && (rtab
[cell_low
] == rtab
[cell_high
])) {
383 pr_debug("TC linklayer: Detected ATM, low(%d)=high(%d)=%u\n",
384 cell_low
, cell_high
, rtab
[cell_high
]);
385 return TC_LINKLAYER_ATM
;
387 return TC_LINKLAYER_ETHERNET
;
390 static struct qdisc_rate_table
*qdisc_rtab_list
;
392 struct qdisc_rate_table
*qdisc_get_rtab(struct tc_ratespec
*r
,
395 struct qdisc_rate_table
*rtab
;
397 if (tab
== NULL
|| r
->rate
== 0 || r
->cell_log
== 0 ||
398 nla_len(tab
) != TC_RTAB_SIZE
)
401 for (rtab
= qdisc_rtab_list
; rtab
; rtab
= rtab
->next
) {
402 if (!memcmp(&rtab
->rate
, r
, sizeof(struct tc_ratespec
)) &&
403 !memcmp(&rtab
->data
, nla_data(tab
), 1024)) {
409 rtab
= kmalloc(sizeof(*rtab
), GFP_KERNEL
);
413 memcpy(rtab
->data
, nla_data(tab
), 1024);
414 if (r
->linklayer
== TC_LINKLAYER_UNAWARE
)
415 r
->linklayer
= __detect_linklayer(r
, rtab
->data
);
416 rtab
->next
= qdisc_rtab_list
;
417 qdisc_rtab_list
= rtab
;
421 EXPORT_SYMBOL(qdisc_get_rtab
);
423 void qdisc_put_rtab(struct qdisc_rate_table
*tab
)
425 struct qdisc_rate_table
*rtab
, **rtabp
;
427 if (!tab
|| --tab
->refcnt
)
430 for (rtabp
= &qdisc_rtab_list
;
431 (rtab
= *rtabp
) != NULL
;
432 rtabp
= &rtab
->next
) {
440 EXPORT_SYMBOL(qdisc_put_rtab
);
442 static LIST_HEAD(qdisc_stab_list
);
443 static DEFINE_SPINLOCK(qdisc_stab_lock
);
445 static const struct nla_policy stab_policy
[TCA_STAB_MAX
+ 1] = {
446 [TCA_STAB_BASE
] = { .len
= sizeof(struct tc_sizespec
) },
447 [TCA_STAB_DATA
] = { .type
= NLA_BINARY
},
450 static struct qdisc_size_table
*qdisc_get_stab(struct nlattr
*opt
)
452 struct nlattr
*tb
[TCA_STAB_MAX
+ 1];
453 struct qdisc_size_table
*stab
;
454 struct tc_sizespec
*s
;
455 unsigned int tsize
= 0;
459 err
= nla_parse_nested(tb
, TCA_STAB_MAX
, opt
, stab_policy
);
462 if (!tb
[TCA_STAB_BASE
])
463 return ERR_PTR(-EINVAL
);
465 s
= nla_data(tb
[TCA_STAB_BASE
]);
468 if (!tb
[TCA_STAB_DATA
])
469 return ERR_PTR(-EINVAL
);
470 tab
= nla_data(tb
[TCA_STAB_DATA
]);
471 tsize
= nla_len(tb
[TCA_STAB_DATA
]) / sizeof(u16
);
474 if (tsize
!= s
->tsize
|| (!tab
&& tsize
> 0))
475 return ERR_PTR(-EINVAL
);
477 spin_lock(&qdisc_stab_lock
);
479 list_for_each_entry(stab
, &qdisc_stab_list
, list
) {
480 if (memcmp(&stab
->szopts
, s
, sizeof(*s
)))
482 if (tsize
> 0 && memcmp(stab
->data
, tab
, tsize
* sizeof(u16
)))
485 spin_unlock(&qdisc_stab_lock
);
489 spin_unlock(&qdisc_stab_lock
);
491 stab
= kmalloc(sizeof(*stab
) + tsize
* sizeof(u16
), GFP_KERNEL
);
493 return ERR_PTR(-ENOMEM
);
498 memcpy(stab
->data
, tab
, tsize
* sizeof(u16
));
500 spin_lock(&qdisc_stab_lock
);
501 list_add_tail(&stab
->list
, &qdisc_stab_list
);
502 spin_unlock(&qdisc_stab_lock
);
507 static void stab_kfree_rcu(struct rcu_head
*head
)
509 kfree(container_of(head
, struct qdisc_size_table
, rcu
));
512 void qdisc_put_stab(struct qdisc_size_table
*tab
)
517 spin_lock(&qdisc_stab_lock
);
519 if (--tab
->refcnt
== 0) {
520 list_del(&tab
->list
);
521 call_rcu_bh(&tab
->rcu
, stab_kfree_rcu
);
524 spin_unlock(&qdisc_stab_lock
);
526 EXPORT_SYMBOL(qdisc_put_stab
);
528 static int qdisc_dump_stab(struct sk_buff
*skb
, struct qdisc_size_table
*stab
)
532 nest
= nla_nest_start(skb
, TCA_STAB
);
534 goto nla_put_failure
;
535 if (nla_put(skb
, TCA_STAB_BASE
, sizeof(stab
->szopts
), &stab
->szopts
))
536 goto nla_put_failure
;
537 nla_nest_end(skb
, nest
);
545 void __qdisc_calculate_pkt_len(struct sk_buff
*skb
,
546 const struct qdisc_size_table
*stab
)
550 pkt_len
= skb
->len
+ stab
->szopts
.overhead
;
551 if (unlikely(!stab
->szopts
.tsize
))
554 slot
= pkt_len
+ stab
->szopts
.cell_align
;
555 if (unlikely(slot
< 0))
558 slot
>>= stab
->szopts
.cell_log
;
559 if (likely(slot
< stab
->szopts
.tsize
))
560 pkt_len
= stab
->data
[slot
];
562 pkt_len
= stab
->data
[stab
->szopts
.tsize
- 1] *
563 (slot
/ stab
->szopts
.tsize
) +
564 stab
->data
[slot
% stab
->szopts
.tsize
];
566 pkt_len
<<= stab
->szopts
.size_log
;
568 if (unlikely(pkt_len
< 1))
570 qdisc_skb_cb(skb
)->pkt_len
= pkt_len
;
572 EXPORT_SYMBOL(__qdisc_calculate_pkt_len
);
574 void qdisc_warn_nonwc(const char *txt
, struct Qdisc
*qdisc
)
576 if (!(qdisc
->flags
& TCQ_F_WARN_NONWC
)) {
577 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
578 txt
, qdisc
->ops
->id
, qdisc
->handle
>> 16);
579 qdisc
->flags
|= TCQ_F_WARN_NONWC
;
582 EXPORT_SYMBOL(qdisc_warn_nonwc
);
584 static enum hrtimer_restart
qdisc_watchdog(struct hrtimer
*timer
)
586 struct qdisc_watchdog
*wd
= container_of(timer
, struct qdisc_watchdog
,
590 __netif_schedule(qdisc_root(wd
->qdisc
));
593 return HRTIMER_NORESTART
;
596 void qdisc_watchdog_init(struct qdisc_watchdog
*wd
, struct Qdisc
*qdisc
)
598 hrtimer_init(&wd
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS_PINNED
);
599 wd
->timer
.function
= qdisc_watchdog
;
602 EXPORT_SYMBOL(qdisc_watchdog_init
);
604 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog
*wd
, u64 expires
)
606 if (test_bit(__QDISC_STATE_DEACTIVATED
,
607 &qdisc_root_sleeping(wd
->qdisc
)->state
))
610 if (wd
->last_expires
== expires
)
613 wd
->last_expires
= expires
;
614 hrtimer_start(&wd
->timer
,
615 ns_to_ktime(expires
),
616 HRTIMER_MODE_ABS_PINNED
);
618 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns
);
620 void qdisc_watchdog_cancel(struct qdisc_watchdog
*wd
)
622 hrtimer_cancel(&wd
->timer
);
624 EXPORT_SYMBOL(qdisc_watchdog_cancel
);
626 static struct hlist_head
*qdisc_class_hash_alloc(unsigned int n
)
628 unsigned int size
= n
* sizeof(struct hlist_head
), i
;
629 struct hlist_head
*h
;
631 if (size
<= PAGE_SIZE
)
632 h
= kmalloc(size
, GFP_KERNEL
);
634 h
= (struct hlist_head
*)
635 __get_free_pages(GFP_KERNEL
, get_order(size
));
638 for (i
= 0; i
< n
; i
++)
639 INIT_HLIST_HEAD(&h
[i
]);
644 static void qdisc_class_hash_free(struct hlist_head
*h
, unsigned int n
)
646 unsigned int size
= n
* sizeof(struct hlist_head
);
648 if (size
<= PAGE_SIZE
)
651 free_pages((unsigned long)h
, get_order(size
));
654 void qdisc_class_hash_grow(struct Qdisc
*sch
, struct Qdisc_class_hash
*clhash
)
656 struct Qdisc_class_common
*cl
;
657 struct hlist_node
*next
;
658 struct hlist_head
*nhash
, *ohash
;
659 unsigned int nsize
, nmask
, osize
;
662 /* Rehash when load factor exceeds 0.75 */
663 if (clhash
->hashelems
* 4 <= clhash
->hashsize
* 3)
665 nsize
= clhash
->hashsize
* 2;
667 nhash
= qdisc_class_hash_alloc(nsize
);
671 ohash
= clhash
->hash
;
672 osize
= clhash
->hashsize
;
675 for (i
= 0; i
< osize
; i
++) {
676 hlist_for_each_entry_safe(cl
, next
, &ohash
[i
], hnode
) {
677 h
= qdisc_class_hash(cl
->classid
, nmask
);
678 hlist_add_head(&cl
->hnode
, &nhash
[h
]);
681 clhash
->hash
= nhash
;
682 clhash
->hashsize
= nsize
;
683 clhash
->hashmask
= nmask
;
684 sch_tree_unlock(sch
);
686 qdisc_class_hash_free(ohash
, osize
);
688 EXPORT_SYMBOL(qdisc_class_hash_grow
);
690 int qdisc_class_hash_init(struct Qdisc_class_hash
*clhash
)
692 unsigned int size
= 4;
694 clhash
->hash
= qdisc_class_hash_alloc(size
);
695 if (clhash
->hash
== NULL
)
697 clhash
->hashsize
= size
;
698 clhash
->hashmask
= size
- 1;
699 clhash
->hashelems
= 0;
702 EXPORT_SYMBOL(qdisc_class_hash_init
);
704 void qdisc_class_hash_destroy(struct Qdisc_class_hash
*clhash
)
706 qdisc_class_hash_free(clhash
->hash
, clhash
->hashsize
);
708 EXPORT_SYMBOL(qdisc_class_hash_destroy
);
710 void qdisc_class_hash_insert(struct Qdisc_class_hash
*clhash
,
711 struct Qdisc_class_common
*cl
)
715 INIT_HLIST_NODE(&cl
->hnode
);
716 h
= qdisc_class_hash(cl
->classid
, clhash
->hashmask
);
717 hlist_add_head(&cl
->hnode
, &clhash
->hash
[h
]);
720 EXPORT_SYMBOL(qdisc_class_hash_insert
);
722 void qdisc_class_hash_remove(struct Qdisc_class_hash
*clhash
,
723 struct Qdisc_class_common
*cl
)
725 hlist_del(&cl
->hnode
);
728 EXPORT_SYMBOL(qdisc_class_hash_remove
);
730 /* Allocate an unique handle from space managed by kernel
731 * Possible range is [8000-FFFF]:0000 (0x8000 values)
733 static u32
qdisc_alloc_handle(struct net_device
*dev
)
736 static u32 autohandle
= TC_H_MAKE(0x80000000U
, 0);
739 autohandle
+= TC_H_MAKE(0x10000U
, 0);
740 if (autohandle
== TC_H_MAKE(TC_H_ROOT
, 0))
741 autohandle
= TC_H_MAKE(0x80000000U
, 0);
742 if (!qdisc_lookup(dev
, autohandle
))
750 void qdisc_tree_reduce_backlog(struct Qdisc
*sch
, unsigned int n
,
753 const struct Qdisc_class_ops
*cops
;
758 if (n
== 0 && len
== 0)
760 drops
= max_t(int, n
, 0);
762 while ((parentid
= sch
->parent
)) {
763 if (TC_H_MAJ(parentid
) == TC_H_MAJ(TC_H_INGRESS
))
766 if (sch
->flags
& TCQ_F_NOPARENT
)
768 /* TODO: perform the search on a per txq basis */
769 sch
= qdisc_lookup(qdisc_dev(sch
), TC_H_MAJ(parentid
));
771 WARN_ON_ONCE(parentid
!= TC_H_ROOT
);
774 cops
= sch
->ops
->cl_ops
;
775 if (cops
->qlen_notify
) {
776 cl
= cops
->get(sch
, parentid
);
777 cops
->qlen_notify(sch
, cl
);
781 sch
->qstats
.backlog
-= len
;
782 __qdisc_qstats_drop(sch
, drops
);
786 EXPORT_SYMBOL(qdisc_tree_reduce_backlog
);
788 static void notify_and_destroy(struct net
*net
, struct sk_buff
*skb
,
789 struct nlmsghdr
*n
, u32 clid
,
790 struct Qdisc
*old
, struct Qdisc
*new)
793 qdisc_notify(net
, skb
, n
, clid
, old
, new);
799 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
802 * When appropriate send a netlink notification using 'skb'
805 * On success, destroy old qdisc.
808 static int qdisc_graft(struct net_device
*dev
, struct Qdisc
*parent
,
809 struct sk_buff
*skb
, struct nlmsghdr
*n
, u32 classid
,
810 struct Qdisc
*new, struct Qdisc
*old
)
812 struct Qdisc
*q
= old
;
813 struct net
*net
= dev_net(dev
);
816 if (parent
== NULL
) {
817 unsigned int i
, num_q
, ingress
;
820 num_q
= dev
->num_tx_queues
;
821 if ((q
&& q
->flags
& TCQ_F_INGRESS
) ||
822 (new && new->flags
& TCQ_F_INGRESS
)) {
825 if (!dev_ingress_queue(dev
))
829 if (dev
->flags
& IFF_UP
)
832 if (new && new->ops
->attach
)
835 for (i
= 0; i
< num_q
; i
++) {
836 struct netdev_queue
*dev_queue
= dev_ingress_queue(dev
);
839 dev_queue
= netdev_get_tx_queue(dev
, i
);
841 old
= dev_graft_qdisc(dev_queue
, new);
843 atomic_inc(&new->refcnt
);
851 notify_and_destroy(net
, skb
, n
, classid
,
853 if (new && !new->ops
->attach
)
854 atomic_inc(&new->refcnt
);
855 dev
->qdisc
= new ? : &noop_qdisc
;
857 if (new && new->ops
->attach
)
858 new->ops
->attach(new);
860 notify_and_destroy(net
, skb
, n
, classid
, old
, new);
863 if (dev
->flags
& IFF_UP
)
866 const struct Qdisc_class_ops
*cops
= parent
->ops
->cl_ops
;
869 if (cops
&& cops
->graft
) {
870 unsigned long cl
= cops
->get(parent
, classid
);
872 err
= cops
->graft(parent
, cl
, new, &old
);
873 cops
->put(parent
, cl
);
878 notify_and_destroy(net
, skb
, n
, classid
, old
, new);
883 /* lockdep annotation is needed for ingress; egress gets it only for name */
884 static struct lock_class_key qdisc_tx_lock
;
885 static struct lock_class_key qdisc_rx_lock
;
888 Allocate and initialize new qdisc.
890 Parameters are passed via opt.
893 static struct Qdisc
*qdisc_create(struct net_device
*dev
,
894 struct netdev_queue
*dev_queue
,
895 struct Qdisc
*p
, u32 parent
, u32 handle
,
896 struct nlattr
**tca
, int *errp
)
899 struct nlattr
*kind
= tca
[TCA_KIND
];
901 struct Qdisc_ops
*ops
;
902 struct qdisc_size_table
*stab
;
904 ops
= qdisc_lookup_ops(kind
);
905 #ifdef CONFIG_MODULES
906 if (ops
== NULL
&& kind
!= NULL
) {
908 if (nla_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
909 /* We dropped the RTNL semaphore in order to
910 * perform the module load. So, even if we
911 * succeeded in loading the module we have to
912 * tell the caller to replay the request. We
913 * indicate this using -EAGAIN.
914 * We replay the request because the device may
915 * go away in the mean time.
918 request_module("sch_%s", name
);
920 ops
= qdisc_lookup_ops(kind
);
922 /* We will try again qdisc_lookup_ops,
923 * so don't keep a reference.
925 module_put(ops
->owner
);
937 sch
= qdisc_alloc(dev_queue
, ops
);
943 sch
->parent
= parent
;
945 if (handle
== TC_H_INGRESS
) {
946 sch
->flags
|= TCQ_F_INGRESS
;
947 handle
= TC_H_MAKE(TC_H_INGRESS
, 0);
948 lockdep_set_class(qdisc_lock(sch
), &qdisc_rx_lock
);
951 handle
= qdisc_alloc_handle(dev
);
956 lockdep_set_class(qdisc_lock(sch
), &qdisc_tx_lock
);
957 if (!netif_is_multiqueue(dev
))
958 sch
->flags
|= TCQ_F_ONETXQUEUE
;
961 sch
->handle
= handle
;
963 /* This exist to keep backward compatible with a userspace
964 * loophole, what allowed userspace to get IFF_NO_QUEUE
965 * facility on older kernels by setting tx_queue_len=0 (prior
966 * to qdisc init), and then forgot to reinit tx_queue_len
967 * before again attaching a qdisc.
969 if ((dev
->priv_flags
& IFF_NO_QUEUE
) && (dev
->tx_queue_len
== 0)) {
970 dev
->tx_queue_len
= DEFAULT_TX_QUEUE_LEN
;
971 netdev_info(dev
, "Caught tx_queue_len zero misconfig\n");
974 if (!ops
->init
|| (err
= ops
->init(sch
, tca
[TCA_OPTIONS
])) == 0) {
975 if (qdisc_is_percpu_stats(sch
)) {
977 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu
);
978 if (!sch
->cpu_bstats
)
981 sch
->cpu_qstats
= alloc_percpu(struct gnet_stats_queue
);
982 if (!sch
->cpu_qstats
)
987 stab
= qdisc_get_stab(tca
[TCA_STAB
]);
992 rcu_assign_pointer(sch
->stab
, stab
);
998 if (sch
->flags
& TCQ_F_MQROOT
)
1001 if ((sch
->parent
!= TC_H_ROOT
) &&
1002 !(sch
->flags
& TCQ_F_INGRESS
) &&
1003 (!p
|| !(p
->flags
& TCQ_F_MQROOT
)))
1004 running
= qdisc_root_sleeping_running(sch
);
1006 running
= &sch
->running
;
1008 err
= gen_new_estimator(&sch
->bstats
,
1018 qdisc_hash_add(sch
);
1022 /* ops->init() failed, we call ->destroy() like qdisc_create_dflt() */
1026 kfree((char *) sch
- sch
->padded
);
1028 module_put(ops
->owner
);
1034 free_percpu(sch
->cpu_bstats
);
1035 free_percpu(sch
->cpu_qstats
);
1037 * Any broken qdiscs that would require a ops->reset() here?
1038 * The qdisc was never in action so it shouldn't be necessary.
1040 qdisc_put_stab(rtnl_dereference(sch
->stab
));
1046 static int qdisc_change(struct Qdisc
*sch
, struct nlattr
**tca
)
1048 struct qdisc_size_table
*ostab
, *stab
= NULL
;
1051 if (tca
[TCA_OPTIONS
]) {
1052 if (sch
->ops
->change
== NULL
)
1054 err
= sch
->ops
->change(sch
, tca
[TCA_OPTIONS
]);
1059 if (tca
[TCA_STAB
]) {
1060 stab
= qdisc_get_stab(tca
[TCA_STAB
]);
1062 return PTR_ERR(stab
);
1065 ostab
= rtnl_dereference(sch
->stab
);
1066 rcu_assign_pointer(sch
->stab
, stab
);
1067 qdisc_put_stab(ostab
);
1069 if (tca
[TCA_RATE
]) {
1070 /* NB: ignores errors from replace_estimator
1071 because change can't be undone. */
1072 if (sch
->flags
& TCQ_F_MQROOT
)
1074 gen_replace_estimator(&sch
->bstats
,
1078 qdisc_root_sleeping_running(sch
),
1085 struct check_loop_arg
{
1086 struct qdisc_walker w
;
1091 static int check_loop_fn(struct Qdisc
*q
, unsigned long cl
,
1092 struct qdisc_walker
*w
);
1094 static int check_loop(struct Qdisc
*q
, struct Qdisc
*p
, int depth
)
1096 struct check_loop_arg arg
;
1098 if (q
->ops
->cl_ops
== NULL
)
1101 arg
.w
.stop
= arg
.w
.skip
= arg
.w
.count
= 0;
1102 arg
.w
.fn
= check_loop_fn
;
1105 q
->ops
->cl_ops
->walk(q
, &arg
.w
);
1106 return arg
.w
.stop
? -ELOOP
: 0;
1110 check_loop_fn(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*w
)
1113 const struct Qdisc_class_ops
*cops
= q
->ops
->cl_ops
;
1114 struct check_loop_arg
*arg
= (struct check_loop_arg
*)w
;
1116 leaf
= cops
->leaf(q
, cl
);
1118 if (leaf
== arg
->p
|| arg
->depth
> 7)
1120 return check_loop(leaf
, arg
->p
, arg
->depth
+ 1);
1129 static int tc_get_qdisc(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1131 struct net
*net
= sock_net(skb
->sk
);
1132 struct tcmsg
*tcm
= nlmsg_data(n
);
1133 struct nlattr
*tca
[TCA_MAX
+ 1];
1134 struct net_device
*dev
;
1136 struct Qdisc
*q
= NULL
;
1137 struct Qdisc
*p
= NULL
;
1140 if ((n
->nlmsg_type
!= RTM_GETQDISC
) &&
1141 !netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1144 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1148 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1152 clid
= tcm
->tcm_parent
;
1154 if (clid
!= TC_H_ROOT
) {
1155 if (TC_H_MAJ(clid
) != TC_H_MAJ(TC_H_INGRESS
)) {
1156 p
= qdisc_lookup(dev
, TC_H_MAJ(clid
));
1159 q
= qdisc_leaf(p
, clid
);
1160 } else if (dev_ingress_queue(dev
)) {
1161 q
= dev_ingress_queue(dev
)->qdisc_sleeping
;
1169 if (tcm
->tcm_handle
&& q
->handle
!= tcm
->tcm_handle
)
1172 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1177 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1180 if (n
->nlmsg_type
== RTM_DELQDISC
) {
1185 err
= qdisc_graft(dev
, p
, skb
, n
, clid
, NULL
, q
);
1189 qdisc_notify(net
, skb
, n
, clid
, NULL
, q
);
1195 * Create/change qdisc.
1198 static int tc_modify_qdisc(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1200 struct net
*net
= sock_net(skb
->sk
);
1202 struct nlattr
*tca
[TCA_MAX
+ 1];
1203 struct net_device
*dev
;
1205 struct Qdisc
*q
, *p
;
1208 if (!netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1212 /* Reinit, just in case something touches this. */
1213 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1217 tcm
= nlmsg_data(n
);
1218 clid
= tcm
->tcm_parent
;
1221 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1227 if (clid
!= TC_H_ROOT
) {
1228 if (clid
!= TC_H_INGRESS
) {
1229 p
= qdisc_lookup(dev
, TC_H_MAJ(clid
));
1232 q
= qdisc_leaf(p
, clid
);
1233 } else if (dev_ingress_queue_create(dev
)) {
1234 q
= dev_ingress_queue(dev
)->qdisc_sleeping
;
1240 /* It may be default qdisc, ignore it */
1241 if (q
&& q
->handle
== 0)
1244 if (!q
|| !tcm
->tcm_handle
|| q
->handle
!= tcm
->tcm_handle
) {
1245 if (tcm
->tcm_handle
) {
1246 if (q
&& !(n
->nlmsg_flags
& NLM_F_REPLACE
))
1248 if (TC_H_MIN(tcm
->tcm_handle
))
1250 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1252 goto create_n_graft
;
1253 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1255 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1258 (p
&& check_loop(q
, p
, 0)))
1260 atomic_inc(&q
->refcnt
);
1264 goto create_n_graft
;
1266 /* This magic test requires explanation.
1268 * We know, that some child q is already
1269 * attached to this parent and have choice:
1270 * either to change it or to create/graft new one.
1272 * 1. We are allowed to create/graft only
1273 * if CREATE and REPLACE flags are set.
1275 * 2. If EXCL is set, requestor wanted to say,
1276 * that qdisc tcm_handle is not expected
1277 * to exist, so that we choose create/graft too.
1279 * 3. The last case is when no flags are set.
1280 * Alas, it is sort of hole in API, we
1281 * cannot decide what to do unambiguously.
1282 * For now we select create/graft, if
1283 * user gave KIND, which does not match existing.
1285 if ((n
->nlmsg_flags
& NLM_F_CREATE
) &&
1286 (n
->nlmsg_flags
& NLM_F_REPLACE
) &&
1287 ((n
->nlmsg_flags
& NLM_F_EXCL
) ||
1289 nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))))
1290 goto create_n_graft
;
1294 if (!tcm
->tcm_handle
)
1296 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1299 /* Change qdisc parameters */
1302 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1304 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1306 err
= qdisc_change(q
, tca
);
1308 qdisc_notify(net
, skb
, n
, clid
, NULL
, q
);
1312 if (!(n
->nlmsg_flags
& NLM_F_CREATE
))
1314 if (clid
== TC_H_INGRESS
) {
1315 if (dev_ingress_queue(dev
))
1316 q
= qdisc_create(dev
, dev_ingress_queue(dev
), p
,
1317 tcm
->tcm_parent
, tcm
->tcm_parent
,
1322 struct netdev_queue
*dev_queue
;
1324 if (p
&& p
->ops
->cl_ops
&& p
->ops
->cl_ops
->select_queue
)
1325 dev_queue
= p
->ops
->cl_ops
->select_queue(p
, tcm
);
1327 dev_queue
= p
->dev_queue
;
1329 dev_queue
= netdev_get_tx_queue(dev
, 0);
1331 q
= qdisc_create(dev
, dev_queue
, p
,
1332 tcm
->tcm_parent
, tcm
->tcm_handle
,
1342 err
= qdisc_graft(dev
, p
, skb
, n
, clid
, q
, NULL
);
1352 static int tc_fill_qdisc(struct sk_buff
*skb
, struct Qdisc
*q
, u32 clid
,
1353 u32 portid
, u32 seq
, u16 flags
, int event
)
1355 struct gnet_stats_basic_cpu __percpu
*cpu_bstats
= NULL
;
1356 struct gnet_stats_queue __percpu
*cpu_qstats
= NULL
;
1358 struct nlmsghdr
*nlh
;
1359 unsigned char *b
= skb_tail_pointer(skb
);
1361 struct qdisc_size_table
*stab
;
1365 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
1367 goto out_nlmsg_trim
;
1368 tcm
= nlmsg_data(nlh
);
1369 tcm
->tcm_family
= AF_UNSPEC
;
1372 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
1373 tcm
->tcm_parent
= clid
;
1374 tcm
->tcm_handle
= q
->handle
;
1375 tcm
->tcm_info
= atomic_read(&q
->refcnt
);
1376 if (nla_put_string(skb
, TCA_KIND
, q
->ops
->id
))
1377 goto nla_put_failure
;
1378 if (q
->ops
->dump
&& q
->ops
->dump(q
, skb
) < 0)
1379 goto nla_put_failure
;
1382 stab
= rtnl_dereference(q
->stab
);
1383 if (stab
&& qdisc_dump_stab(skb
, stab
) < 0)
1384 goto nla_put_failure
;
1386 if (gnet_stats_start_copy_compat(skb
, TCA_STATS2
, TCA_STATS
, TCA_XSTATS
,
1387 NULL
, &d
, TCA_PAD
) < 0)
1388 goto nla_put_failure
;
1390 if (q
->ops
->dump_stats
&& q
->ops
->dump_stats(q
, &d
) < 0)
1391 goto nla_put_failure
;
1393 if (qdisc_is_percpu_stats(q
)) {
1394 cpu_bstats
= q
->cpu_bstats
;
1395 cpu_qstats
= q
->cpu_qstats
;
1398 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q
),
1399 &d
, cpu_bstats
, &q
->bstats
) < 0 ||
1400 gnet_stats_copy_rate_est(&d
, &q
->rate_est
) < 0 ||
1401 gnet_stats_copy_queue(&d
, cpu_qstats
, &q
->qstats
, qlen
) < 0)
1402 goto nla_put_failure
;
1404 if (gnet_stats_finish_copy(&d
) < 0)
1405 goto nla_put_failure
;
1407 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
1416 static bool tc_qdisc_dump_ignore(struct Qdisc
*q
)
1418 return (q
->flags
& TCQ_F_BUILTIN
) ? true : false;
1421 static int qdisc_notify(struct net
*net
, struct sk_buff
*oskb
,
1422 struct nlmsghdr
*n
, u32 clid
,
1423 struct Qdisc
*old
, struct Qdisc
*new)
1425 struct sk_buff
*skb
;
1426 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1428 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1432 if (old
&& !tc_qdisc_dump_ignore(old
)) {
1433 if (tc_fill_qdisc(skb
, old
, clid
, portid
, n
->nlmsg_seq
,
1434 0, RTM_DELQDISC
) < 0)
1437 if (new && !tc_qdisc_dump_ignore(new)) {
1438 if (tc_fill_qdisc(skb
, new, clid
, portid
, n
->nlmsg_seq
,
1439 old
? NLM_F_REPLACE
: 0, RTM_NEWQDISC
) < 0)
1444 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1445 n
->nlmsg_flags
& NLM_F_ECHO
);
1452 static int tc_dump_qdisc_root(struct Qdisc
*root
, struct sk_buff
*skb
,
1453 struct netlink_callback
*cb
,
1454 int *q_idx_p
, int s_q_idx
, bool recur
)
1456 int ret
= 0, q_idx
= *q_idx_p
;
1464 if (q_idx
< s_q_idx
) {
1467 if (!tc_qdisc_dump_ignore(q
) &&
1468 tc_fill_qdisc(skb
, q
, q
->parent
, NETLINK_CB(cb
->skb
).portid
,
1469 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1475 /* If dumping singletons, there is no qdisc_dev(root) and the singleton
1476 * itself has already been dumped.
1478 * If we've already dumped the top-level (ingress) qdisc above and the global
1479 * qdisc hashtable, we don't want to hit it again
1481 if (!qdisc_dev(root
) || !recur
)
1484 hash_for_each(qdisc_dev(root
)->qdisc_hash
, b
, q
, hash
) {
1485 if (q_idx
< s_q_idx
) {
1489 if (!tc_qdisc_dump_ignore(q
) &&
1490 tc_fill_qdisc(skb
, q
, q
->parent
, NETLINK_CB(cb
->skb
).portid
,
1491 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1505 static int tc_dump_qdisc(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1507 struct net
*net
= sock_net(skb
->sk
);
1510 struct net_device
*dev
;
1512 s_idx
= cb
->args
[0];
1513 s_q_idx
= q_idx
= cb
->args
[1];
1517 for_each_netdev(net
, dev
) {
1518 struct netdev_queue
*dev_queue
;
1526 if (tc_dump_qdisc_root(dev
->qdisc
, skb
, cb
, &q_idx
, s_q_idx
,
1530 dev_queue
= dev_ingress_queue(dev
);
1532 tc_dump_qdisc_root(dev_queue
->qdisc_sleeping
, skb
, cb
,
1533 &q_idx
, s_q_idx
, false) < 0)
1542 cb
->args
[1] = q_idx
;
1549 /************************************************
1550 * Traffic classes manipulation. *
1551 ************************************************/
1555 static int tc_ctl_tclass(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1557 struct net
*net
= sock_net(skb
->sk
);
1558 struct tcmsg
*tcm
= nlmsg_data(n
);
1559 struct nlattr
*tca
[TCA_MAX
+ 1];
1560 struct net_device
*dev
;
1561 struct Qdisc
*q
= NULL
;
1562 const struct Qdisc_class_ops
*cops
;
1563 unsigned long cl
= 0;
1564 unsigned long new_cl
;
1570 if ((n
->nlmsg_type
!= RTM_GETTCLASS
) &&
1571 !netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1574 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1578 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1583 parent == TC_H_UNSPEC - unspecified parent.
1584 parent == TC_H_ROOT - class is root, which has no parent.
1585 parent == X:0 - parent is root class.
1586 parent == X:Y - parent is a node in hierarchy.
1587 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1589 handle == 0:0 - generate handle from kernel pool.
1590 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1591 handle == X:Y - clear.
1592 handle == X:0 - root class.
1595 /* Step 1. Determine qdisc handle X:0 */
1597 portid
= tcm
->tcm_parent
;
1598 clid
= tcm
->tcm_handle
;
1599 qid
= TC_H_MAJ(clid
);
1601 if (portid
!= TC_H_ROOT
) {
1602 u32 qid1
= TC_H_MAJ(portid
);
1605 /* If both majors are known, they must be identical. */
1610 } else if (qid
== 0)
1611 qid
= dev
->qdisc
->handle
;
1613 /* Now qid is genuine qdisc handle consistent
1614 * both with parent and child.
1616 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1619 portid
= TC_H_MAKE(qid
, portid
);
1622 qid
= dev
->qdisc
->handle
;
1625 /* OK. Locate qdisc */
1626 q
= qdisc_lookup(dev
, qid
);
1630 /* An check that it supports classes */
1631 cops
= q
->ops
->cl_ops
;
1635 /* Now try to get class */
1637 if (portid
== TC_H_ROOT
)
1640 clid
= TC_H_MAKE(qid
, clid
);
1643 cl
= cops
->get(q
, clid
);
1647 if (n
->nlmsg_type
!= RTM_NEWTCLASS
||
1648 !(n
->nlmsg_flags
& NLM_F_CREATE
))
1651 switch (n
->nlmsg_type
) {
1654 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1660 err
= cops
->delete(q
, cl
);
1662 tclass_notify(net
, skb
, n
, q
, cl
,
1666 err
= tclass_notify(net
, skb
, n
, q
, cl
, RTM_NEWTCLASS
);
1677 err
= cops
->change(q
, clid
, portid
, tca
, &new_cl
);
1679 tclass_notify(net
, skb
, n
, q
, new_cl
, RTM_NEWTCLASS
);
1689 static int tc_fill_tclass(struct sk_buff
*skb
, struct Qdisc
*q
,
1691 u32 portid
, u32 seq
, u16 flags
, int event
)
1694 struct nlmsghdr
*nlh
;
1695 unsigned char *b
= skb_tail_pointer(skb
);
1697 const struct Qdisc_class_ops
*cl_ops
= q
->ops
->cl_ops
;
1700 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
1702 goto out_nlmsg_trim
;
1703 tcm
= nlmsg_data(nlh
);
1704 tcm
->tcm_family
= AF_UNSPEC
;
1707 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
1708 tcm
->tcm_parent
= q
->handle
;
1709 tcm
->tcm_handle
= q
->handle
;
1711 if (nla_put_string(skb
, TCA_KIND
, q
->ops
->id
))
1712 goto nla_put_failure
;
1713 if (cl_ops
->dump
&& cl_ops
->dump(q
, cl
, skb
, tcm
) < 0)
1714 goto nla_put_failure
;
1716 if (gnet_stats_start_copy_compat(skb
, TCA_STATS2
, TCA_STATS
, TCA_XSTATS
,
1717 NULL
, &d
, TCA_PAD
) < 0)
1718 goto nla_put_failure
;
1720 if (cl_ops
->dump_stats
&& cl_ops
->dump_stats(q
, cl
, &d
) < 0)
1721 goto nla_put_failure
;
1723 if (gnet_stats_finish_copy(&d
) < 0)
1724 goto nla_put_failure
;
1726 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
1735 static int tclass_notify(struct net
*net
, struct sk_buff
*oskb
,
1736 struct nlmsghdr
*n
, struct Qdisc
*q
,
1737 unsigned long cl
, int event
)
1739 struct sk_buff
*skb
;
1740 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1742 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1746 if (tc_fill_tclass(skb
, q
, cl
, portid
, n
->nlmsg_seq
, 0, event
) < 0) {
1751 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1752 n
->nlmsg_flags
& NLM_F_ECHO
);
1755 struct qdisc_dump_args
{
1756 struct qdisc_walker w
;
1757 struct sk_buff
*skb
;
1758 struct netlink_callback
*cb
;
1761 static int qdisc_class_dump(struct Qdisc
*q
, unsigned long cl
,
1762 struct qdisc_walker
*arg
)
1764 struct qdisc_dump_args
*a
= (struct qdisc_dump_args
*)arg
;
1766 return tc_fill_tclass(a
->skb
, q
, cl
, NETLINK_CB(a
->cb
->skb
).portid
,
1767 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1771 static int tc_dump_tclass_qdisc(struct Qdisc
*q
, struct sk_buff
*skb
,
1772 struct tcmsg
*tcm
, struct netlink_callback
*cb
,
1775 struct qdisc_dump_args arg
;
1777 if (tc_qdisc_dump_ignore(q
) ||
1778 *t_p
< s_t
|| !q
->ops
->cl_ops
||
1780 TC_H_MAJ(tcm
->tcm_parent
) != q
->handle
)) {
1785 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
1786 arg
.w
.fn
= qdisc_class_dump
;
1790 arg
.w
.skip
= cb
->args
[1];
1792 q
->ops
->cl_ops
->walk(q
, &arg
.w
);
1793 cb
->args
[1] = arg
.w
.count
;
1800 static int tc_dump_tclass_root(struct Qdisc
*root
, struct sk_buff
*skb
,
1801 struct tcmsg
*tcm
, struct netlink_callback
*cb
,
1810 if (tc_dump_tclass_qdisc(root
, skb
, tcm
, cb
, t_p
, s_t
) < 0)
1813 if (!qdisc_dev(root
))
1816 hash_for_each(qdisc_dev(root
)->qdisc_hash
, b
, q
, hash
) {
1817 if (tc_dump_tclass_qdisc(q
, skb
, tcm
, cb
, t_p
, s_t
) < 0)
1824 static int tc_dump_tclass(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1826 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
1827 struct net
*net
= sock_net(skb
->sk
);
1828 struct netdev_queue
*dev_queue
;
1829 struct net_device
*dev
;
1832 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
1834 dev
= dev_get_by_index(net
, tcm
->tcm_ifindex
);
1841 if (tc_dump_tclass_root(dev
->qdisc
, skb
, tcm
, cb
, &t
, s_t
) < 0)
1844 dev_queue
= dev_ingress_queue(dev
);
1846 tc_dump_tclass_root(dev_queue
->qdisc_sleeping
, skb
, tcm
, cb
,
1857 /* Main classifier routine: scans classifier chain attached
1858 * to this qdisc, (optionally) tests for protocol and asks
1859 * specific classifiers.
1861 int tc_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
1862 struct tcf_result
*res
, bool compat_mode
)
1864 __be16 protocol
= tc_skb_protocol(skb
);
1865 #ifdef CONFIG_NET_CLS_ACT
1866 const int max_reclassify_loop
= 4;
1867 const struct tcf_proto
*old_tp
= tp
;
1872 for (; tp
; tp
= rcu_dereference_bh(tp
->next
)) {
1875 if (tp
->protocol
!= protocol
&&
1876 tp
->protocol
!= htons(ETH_P_ALL
))
1879 err
= tp
->classify(skb
, tp
, res
);
1880 #ifdef CONFIG_NET_CLS_ACT
1881 if (unlikely(err
== TC_ACT_RECLASSIFY
&& !compat_mode
))
1888 return TC_ACT_UNSPEC
; /* signal: continue lookup */
1889 #ifdef CONFIG_NET_CLS_ACT
1891 if (unlikely(limit
++ >= max_reclassify_loop
)) {
1892 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
1893 tp
->q
->ops
->id
, tp
->prio
& 0xffff,
1894 ntohs(tp
->protocol
));
1899 protocol
= tc_skb_protocol(skb
);
1903 EXPORT_SYMBOL(tc_classify
);
1905 #ifdef CONFIG_PROC_FS
1906 static int psched_show(struct seq_file
*seq
, void *v
)
1908 seq_printf(seq
, "%08x %08x %08x %08x\n",
1909 (u32
)NSEC_PER_USEC
, (u32
)PSCHED_TICKS2NS(1),
1911 (u32
)NSEC_PER_SEC
/ hrtimer_resolution
);
1916 static int psched_open(struct inode
*inode
, struct file
*file
)
1918 return single_open(file
, psched_show
, NULL
);
1921 static const struct file_operations psched_fops
= {
1922 .owner
= THIS_MODULE
,
1923 .open
= psched_open
,
1925 .llseek
= seq_lseek
,
1926 .release
= single_release
,
1929 static int __net_init
psched_net_init(struct net
*net
)
1931 struct proc_dir_entry
*e
;
1933 e
= proc_create("psched", 0, net
->proc_net
, &psched_fops
);
1940 static void __net_exit
psched_net_exit(struct net
*net
)
1942 remove_proc_entry("psched", net
->proc_net
);
1945 static int __net_init
psched_net_init(struct net
*net
)
1950 static void __net_exit
psched_net_exit(struct net
*net
)
1955 static struct pernet_operations psched_net_ops
= {
1956 .init
= psched_net_init
,
1957 .exit
= psched_net_exit
,
1960 static int __init
pktsched_init(void)
1964 err
= register_pernet_subsys(&psched_net_ops
);
1966 pr_err("pktsched_init: "
1967 "cannot initialize per netns operations\n");
1971 register_qdisc(&pfifo_fast_ops
);
1972 register_qdisc(&pfifo_qdisc_ops
);
1973 register_qdisc(&bfifo_qdisc_ops
);
1974 register_qdisc(&pfifo_head_drop_qdisc_ops
);
1975 register_qdisc(&mq_qdisc_ops
);
1976 register_qdisc(&noqueue_qdisc_ops
);
1978 rtnl_register(PF_UNSPEC
, RTM_NEWQDISC
, tc_modify_qdisc
, NULL
, NULL
);
1979 rtnl_register(PF_UNSPEC
, RTM_DELQDISC
, tc_get_qdisc
, NULL
, NULL
);
1980 rtnl_register(PF_UNSPEC
, RTM_GETQDISC
, tc_get_qdisc
, tc_dump_qdisc
,
1982 rtnl_register(PF_UNSPEC
, RTM_NEWTCLASS
, tc_ctl_tclass
, NULL
, NULL
);
1983 rtnl_register(PF_UNSPEC
, RTM_DELTCLASS
, tc_ctl_tclass
, NULL
, NULL
);
1984 rtnl_register(PF_UNSPEC
, RTM_GETTCLASS
, tc_ctl_tclass
, tc_dump_tclass
,
1990 subsys_initcall(pktsched_init
);