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
, struct nlattr
*tab
)
394 struct qdisc_rate_table
*rtab
;
396 if (tab
== NULL
|| r
->rate
== 0 || r
->cell_log
== 0 ||
397 nla_len(tab
) != TC_RTAB_SIZE
)
400 for (rtab
= qdisc_rtab_list
; rtab
; rtab
= rtab
->next
) {
401 if (!memcmp(&rtab
->rate
, r
, sizeof(struct tc_ratespec
)) &&
402 !memcmp(&rtab
->data
, nla_data(tab
), 1024)) {
408 rtab
= kmalloc(sizeof(*rtab
), GFP_KERNEL
);
412 memcpy(rtab
->data
, nla_data(tab
), 1024);
413 if (r
->linklayer
== TC_LINKLAYER_UNAWARE
)
414 r
->linklayer
= __detect_linklayer(r
, rtab
->data
);
415 rtab
->next
= qdisc_rtab_list
;
416 qdisc_rtab_list
= rtab
;
420 EXPORT_SYMBOL(qdisc_get_rtab
);
422 void qdisc_put_rtab(struct qdisc_rate_table
*tab
)
424 struct qdisc_rate_table
*rtab
, **rtabp
;
426 if (!tab
|| --tab
->refcnt
)
429 for (rtabp
= &qdisc_rtab_list
;
430 (rtab
= *rtabp
) != NULL
;
431 rtabp
= &rtab
->next
) {
439 EXPORT_SYMBOL(qdisc_put_rtab
);
441 static LIST_HEAD(qdisc_stab_list
);
442 static DEFINE_SPINLOCK(qdisc_stab_lock
);
444 static const struct nla_policy stab_policy
[TCA_STAB_MAX
+ 1] = {
445 [TCA_STAB_BASE
] = { .len
= sizeof(struct tc_sizespec
) },
446 [TCA_STAB_DATA
] = { .type
= NLA_BINARY
},
449 static struct qdisc_size_table
*qdisc_get_stab(struct nlattr
*opt
)
451 struct nlattr
*tb
[TCA_STAB_MAX
+ 1];
452 struct qdisc_size_table
*stab
;
453 struct tc_sizespec
*s
;
454 unsigned int tsize
= 0;
458 err
= nla_parse_nested(tb
, TCA_STAB_MAX
, opt
, stab_policy
);
461 if (!tb
[TCA_STAB_BASE
])
462 return ERR_PTR(-EINVAL
);
464 s
= nla_data(tb
[TCA_STAB_BASE
]);
467 if (!tb
[TCA_STAB_DATA
])
468 return ERR_PTR(-EINVAL
);
469 tab
= nla_data(tb
[TCA_STAB_DATA
]);
470 tsize
= nla_len(tb
[TCA_STAB_DATA
]) / sizeof(u16
);
473 if (tsize
!= s
->tsize
|| (!tab
&& tsize
> 0))
474 return ERR_PTR(-EINVAL
);
476 spin_lock(&qdisc_stab_lock
);
478 list_for_each_entry(stab
, &qdisc_stab_list
, list
) {
479 if (memcmp(&stab
->szopts
, s
, sizeof(*s
)))
481 if (tsize
> 0 && memcmp(stab
->data
, tab
, tsize
* sizeof(u16
)))
484 spin_unlock(&qdisc_stab_lock
);
488 spin_unlock(&qdisc_stab_lock
);
490 stab
= kmalloc(sizeof(*stab
) + tsize
* sizeof(u16
), GFP_KERNEL
);
492 return ERR_PTR(-ENOMEM
);
497 memcpy(stab
->data
, tab
, tsize
* sizeof(u16
));
499 spin_lock(&qdisc_stab_lock
);
500 list_add_tail(&stab
->list
, &qdisc_stab_list
);
501 spin_unlock(&qdisc_stab_lock
);
506 static void stab_kfree_rcu(struct rcu_head
*head
)
508 kfree(container_of(head
, struct qdisc_size_table
, rcu
));
511 void qdisc_put_stab(struct qdisc_size_table
*tab
)
516 spin_lock(&qdisc_stab_lock
);
518 if (--tab
->refcnt
== 0) {
519 list_del(&tab
->list
);
520 call_rcu_bh(&tab
->rcu
, stab_kfree_rcu
);
523 spin_unlock(&qdisc_stab_lock
);
525 EXPORT_SYMBOL(qdisc_put_stab
);
527 static int qdisc_dump_stab(struct sk_buff
*skb
, struct qdisc_size_table
*stab
)
531 nest
= nla_nest_start(skb
, TCA_STAB
);
533 goto nla_put_failure
;
534 if (nla_put(skb
, TCA_STAB_BASE
, sizeof(stab
->szopts
), &stab
->szopts
))
535 goto nla_put_failure
;
536 nla_nest_end(skb
, nest
);
544 void __qdisc_calculate_pkt_len(struct sk_buff
*skb
, const struct qdisc_size_table
*stab
)
548 pkt_len
= skb
->len
+ stab
->szopts
.overhead
;
549 if (unlikely(!stab
->szopts
.tsize
))
552 slot
= pkt_len
+ stab
->szopts
.cell_align
;
553 if (unlikely(slot
< 0))
556 slot
>>= stab
->szopts
.cell_log
;
557 if (likely(slot
< stab
->szopts
.tsize
))
558 pkt_len
= stab
->data
[slot
];
560 pkt_len
= stab
->data
[stab
->szopts
.tsize
- 1] *
561 (slot
/ stab
->szopts
.tsize
) +
562 stab
->data
[slot
% stab
->szopts
.tsize
];
564 pkt_len
<<= stab
->szopts
.size_log
;
566 if (unlikely(pkt_len
< 1))
568 qdisc_skb_cb(skb
)->pkt_len
= pkt_len
;
570 EXPORT_SYMBOL(__qdisc_calculate_pkt_len
);
572 void qdisc_warn_nonwc(const char *txt
, struct Qdisc
*qdisc
)
574 if (!(qdisc
->flags
& TCQ_F_WARN_NONWC
)) {
575 pr_warn("%s: %s qdisc %X: is non-work-conserving?\n",
576 txt
, qdisc
->ops
->id
, qdisc
->handle
>> 16);
577 qdisc
->flags
|= TCQ_F_WARN_NONWC
;
580 EXPORT_SYMBOL(qdisc_warn_nonwc
);
582 static enum hrtimer_restart
qdisc_watchdog(struct hrtimer
*timer
)
584 struct qdisc_watchdog
*wd
= container_of(timer
, struct qdisc_watchdog
,
588 __netif_schedule(qdisc_root(wd
->qdisc
));
591 return HRTIMER_NORESTART
;
594 void qdisc_watchdog_init(struct qdisc_watchdog
*wd
, struct Qdisc
*qdisc
)
596 hrtimer_init(&wd
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS_PINNED
);
597 wd
->timer
.function
= qdisc_watchdog
;
600 EXPORT_SYMBOL(qdisc_watchdog_init
);
602 void qdisc_watchdog_schedule_ns(struct qdisc_watchdog
*wd
, u64 expires
)
604 if (test_bit(__QDISC_STATE_DEACTIVATED
,
605 &qdisc_root_sleeping(wd
->qdisc
)->state
))
608 if (wd
->last_expires
== expires
)
611 wd
->last_expires
= expires
;
612 hrtimer_start(&wd
->timer
,
613 ns_to_ktime(expires
),
614 HRTIMER_MODE_ABS_PINNED
);
616 EXPORT_SYMBOL(qdisc_watchdog_schedule_ns
);
618 void qdisc_watchdog_cancel(struct qdisc_watchdog
*wd
)
620 hrtimer_cancel(&wd
->timer
);
622 EXPORT_SYMBOL(qdisc_watchdog_cancel
);
624 static struct hlist_head
*qdisc_class_hash_alloc(unsigned int n
)
626 unsigned int size
= n
* sizeof(struct hlist_head
), i
;
627 struct hlist_head
*h
;
629 if (size
<= PAGE_SIZE
)
630 h
= kmalloc(size
, GFP_KERNEL
);
632 h
= (struct hlist_head
*)
633 __get_free_pages(GFP_KERNEL
, get_order(size
));
636 for (i
= 0; i
< n
; i
++)
637 INIT_HLIST_HEAD(&h
[i
]);
642 static void qdisc_class_hash_free(struct hlist_head
*h
, unsigned int n
)
644 unsigned int size
= n
* sizeof(struct hlist_head
);
646 if (size
<= PAGE_SIZE
)
649 free_pages((unsigned long)h
, get_order(size
));
652 void qdisc_class_hash_grow(struct Qdisc
*sch
, struct Qdisc_class_hash
*clhash
)
654 struct Qdisc_class_common
*cl
;
655 struct hlist_node
*next
;
656 struct hlist_head
*nhash
, *ohash
;
657 unsigned int nsize
, nmask
, osize
;
660 /* Rehash when load factor exceeds 0.75 */
661 if (clhash
->hashelems
* 4 <= clhash
->hashsize
* 3)
663 nsize
= clhash
->hashsize
* 2;
665 nhash
= qdisc_class_hash_alloc(nsize
);
669 ohash
= clhash
->hash
;
670 osize
= clhash
->hashsize
;
673 for (i
= 0; i
< osize
; i
++) {
674 hlist_for_each_entry_safe(cl
, next
, &ohash
[i
], hnode
) {
675 h
= qdisc_class_hash(cl
->classid
, nmask
);
676 hlist_add_head(&cl
->hnode
, &nhash
[h
]);
679 clhash
->hash
= nhash
;
680 clhash
->hashsize
= nsize
;
681 clhash
->hashmask
= nmask
;
682 sch_tree_unlock(sch
);
684 qdisc_class_hash_free(ohash
, osize
);
686 EXPORT_SYMBOL(qdisc_class_hash_grow
);
688 int qdisc_class_hash_init(struct Qdisc_class_hash
*clhash
)
690 unsigned int size
= 4;
692 clhash
->hash
= qdisc_class_hash_alloc(size
);
693 if (clhash
->hash
== NULL
)
695 clhash
->hashsize
= size
;
696 clhash
->hashmask
= size
- 1;
697 clhash
->hashelems
= 0;
700 EXPORT_SYMBOL(qdisc_class_hash_init
);
702 void qdisc_class_hash_destroy(struct Qdisc_class_hash
*clhash
)
704 qdisc_class_hash_free(clhash
->hash
, clhash
->hashsize
);
706 EXPORT_SYMBOL(qdisc_class_hash_destroy
);
708 void qdisc_class_hash_insert(struct Qdisc_class_hash
*clhash
,
709 struct Qdisc_class_common
*cl
)
713 INIT_HLIST_NODE(&cl
->hnode
);
714 h
= qdisc_class_hash(cl
->classid
, clhash
->hashmask
);
715 hlist_add_head(&cl
->hnode
, &clhash
->hash
[h
]);
718 EXPORT_SYMBOL(qdisc_class_hash_insert
);
720 void qdisc_class_hash_remove(struct Qdisc_class_hash
*clhash
,
721 struct Qdisc_class_common
*cl
)
723 hlist_del(&cl
->hnode
);
726 EXPORT_SYMBOL(qdisc_class_hash_remove
);
728 /* Allocate an unique handle from space managed by kernel
729 * Possible range is [8000-FFFF]:0000 (0x8000 values)
731 static u32
qdisc_alloc_handle(struct net_device
*dev
)
734 static u32 autohandle
= TC_H_MAKE(0x80000000U
, 0);
737 autohandle
+= TC_H_MAKE(0x10000U
, 0);
738 if (autohandle
== TC_H_MAKE(TC_H_ROOT
, 0))
739 autohandle
= TC_H_MAKE(0x80000000U
, 0);
740 if (!qdisc_lookup(dev
, autohandle
))
748 void qdisc_tree_reduce_backlog(struct Qdisc
*sch
, unsigned int n
,
751 const struct Qdisc_class_ops
*cops
;
756 if (n
== 0 && len
== 0)
758 drops
= max_t(int, n
, 0);
760 while ((parentid
= sch
->parent
)) {
761 if (TC_H_MAJ(parentid
) == TC_H_MAJ(TC_H_INGRESS
))
764 if (sch
->flags
& TCQ_F_NOPARENT
)
766 /* TODO: perform the search on a per txq basis */
767 sch
= qdisc_lookup(qdisc_dev(sch
), TC_H_MAJ(parentid
));
769 WARN_ON_ONCE(parentid
!= TC_H_ROOT
);
772 cops
= sch
->ops
->cl_ops
;
773 if (cops
->qlen_notify
) {
774 cl
= cops
->get(sch
, parentid
);
775 cops
->qlen_notify(sch
, cl
);
779 sch
->qstats
.backlog
-= len
;
780 __qdisc_qstats_drop(sch
, drops
);
784 EXPORT_SYMBOL(qdisc_tree_reduce_backlog
);
786 static void notify_and_destroy(struct net
*net
, struct sk_buff
*skb
,
787 struct nlmsghdr
*n
, u32 clid
,
788 struct Qdisc
*old
, struct Qdisc
*new)
791 qdisc_notify(net
, skb
, n
, clid
, old
, new);
797 /* Graft qdisc "new" to class "classid" of qdisc "parent" or
800 * When appropriate send a netlink notification using 'skb'
803 * On success, destroy old qdisc.
806 static int qdisc_graft(struct net_device
*dev
, struct Qdisc
*parent
,
807 struct sk_buff
*skb
, struct nlmsghdr
*n
, u32 classid
,
808 struct Qdisc
*new, struct Qdisc
*old
)
810 struct Qdisc
*q
= old
;
811 struct net
*net
= dev_net(dev
);
814 if (parent
== NULL
) {
815 unsigned int i
, num_q
, ingress
;
818 num_q
= dev
->num_tx_queues
;
819 if ((q
&& q
->flags
& TCQ_F_INGRESS
) ||
820 (new && new->flags
& TCQ_F_INGRESS
)) {
823 if (!dev_ingress_queue(dev
))
827 if (dev
->flags
& IFF_UP
)
830 if (new && new->ops
->attach
)
833 for (i
= 0; i
< num_q
; i
++) {
834 struct netdev_queue
*dev_queue
= dev_ingress_queue(dev
);
837 dev_queue
= netdev_get_tx_queue(dev
, i
);
839 old
= dev_graft_qdisc(dev_queue
, new);
841 atomic_inc(&new->refcnt
);
849 notify_and_destroy(net
, skb
, n
, classid
,
851 if (new && !new->ops
->attach
)
852 atomic_inc(&new->refcnt
);
853 dev
->qdisc
= new ? : &noop_qdisc
;
855 if (new && new->ops
->attach
)
856 new->ops
->attach(new);
858 notify_and_destroy(net
, skb
, n
, classid
, old
, new);
861 if (dev
->flags
& IFF_UP
)
864 const struct Qdisc_class_ops
*cops
= parent
->ops
->cl_ops
;
867 if (cops
&& cops
->graft
) {
868 unsigned long cl
= cops
->get(parent
, classid
);
870 err
= cops
->graft(parent
, cl
, new, &old
);
871 cops
->put(parent
, cl
);
876 notify_and_destroy(net
, skb
, n
, classid
, old
, new);
881 /* lockdep annotation is needed for ingress; egress gets it only for name */
882 static struct lock_class_key qdisc_tx_lock
;
883 static struct lock_class_key qdisc_rx_lock
;
886 Allocate and initialize new qdisc.
888 Parameters are passed via opt.
891 static struct Qdisc
*
892 qdisc_create(struct net_device
*dev
, struct netdev_queue
*dev_queue
,
893 struct Qdisc
*p
, u32 parent
, u32 handle
,
894 struct nlattr
**tca
, int *errp
)
897 struct nlattr
*kind
= tca
[TCA_KIND
];
899 struct Qdisc_ops
*ops
;
900 struct qdisc_size_table
*stab
;
902 ops
= qdisc_lookup_ops(kind
);
903 #ifdef CONFIG_MODULES
904 if (ops
== NULL
&& kind
!= NULL
) {
906 if (nla_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
907 /* We dropped the RTNL semaphore in order to
908 * perform the module load. So, even if we
909 * succeeded in loading the module we have to
910 * tell the caller to replay the request. We
911 * indicate this using -EAGAIN.
912 * We replay the request because the device may
913 * go away in the mean time.
916 request_module("sch_%s", name
);
918 ops
= qdisc_lookup_ops(kind
);
920 /* We will try again qdisc_lookup_ops,
921 * so don't keep a reference.
923 module_put(ops
->owner
);
935 sch
= qdisc_alloc(dev_queue
, ops
);
941 sch
->parent
= parent
;
943 if (handle
== TC_H_INGRESS
) {
944 sch
->flags
|= TCQ_F_INGRESS
;
945 handle
= TC_H_MAKE(TC_H_INGRESS
, 0);
946 lockdep_set_class(qdisc_lock(sch
), &qdisc_rx_lock
);
949 handle
= qdisc_alloc_handle(dev
);
954 lockdep_set_class(qdisc_lock(sch
), &qdisc_tx_lock
);
955 if (!netif_is_multiqueue(dev
))
956 sch
->flags
|= TCQ_F_ONETXQUEUE
;
959 sch
->handle
= handle
;
961 if (!ops
->init
|| (err
= ops
->init(sch
, tca
[TCA_OPTIONS
])) == 0) {
962 if (qdisc_is_percpu_stats(sch
)) {
964 netdev_alloc_pcpu_stats(struct gnet_stats_basic_cpu
);
965 if (!sch
->cpu_bstats
)
968 sch
->cpu_qstats
= alloc_percpu(struct gnet_stats_queue
);
969 if (!sch
->cpu_qstats
)
974 stab
= qdisc_get_stab(tca
[TCA_STAB
]);
979 rcu_assign_pointer(sch
->stab
, stab
);
985 if (sch
->flags
& TCQ_F_MQROOT
)
988 if ((sch
->parent
!= TC_H_ROOT
) &&
989 !(sch
->flags
& TCQ_F_INGRESS
) &&
990 (!p
|| !(p
->flags
& TCQ_F_MQROOT
)))
991 running
= qdisc_root_sleeping_running(sch
);
993 running
= &sch
->running
;
995 err
= gen_new_estimator(&sch
->bstats
,
1005 qdisc_hash_add(sch
);
1011 kfree((char *) sch
- sch
->padded
);
1013 module_put(ops
->owner
);
1019 free_percpu(sch
->cpu_bstats
);
1020 free_percpu(sch
->cpu_qstats
);
1022 * Any broken qdiscs that would require a ops->reset() here?
1023 * The qdisc was never in action so it shouldn't be necessary.
1025 qdisc_put_stab(rtnl_dereference(sch
->stab
));
1031 static int qdisc_change(struct Qdisc
*sch
, struct nlattr
**tca
)
1033 struct qdisc_size_table
*ostab
, *stab
= NULL
;
1036 if (tca
[TCA_OPTIONS
]) {
1037 if (sch
->ops
->change
== NULL
)
1039 err
= sch
->ops
->change(sch
, tca
[TCA_OPTIONS
]);
1044 if (tca
[TCA_STAB
]) {
1045 stab
= qdisc_get_stab(tca
[TCA_STAB
]);
1047 return PTR_ERR(stab
);
1050 ostab
= rtnl_dereference(sch
->stab
);
1051 rcu_assign_pointer(sch
->stab
, stab
);
1052 qdisc_put_stab(ostab
);
1054 if (tca
[TCA_RATE
]) {
1055 /* NB: ignores errors from replace_estimator
1056 because change can't be undone. */
1057 if (sch
->flags
& TCQ_F_MQROOT
)
1059 gen_replace_estimator(&sch
->bstats
,
1063 qdisc_root_sleeping_running(sch
),
1070 struct check_loop_arg
{
1071 struct qdisc_walker w
;
1076 static int check_loop_fn(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*w
);
1078 static int check_loop(struct Qdisc
*q
, struct Qdisc
*p
, int depth
)
1080 struct check_loop_arg arg
;
1082 if (q
->ops
->cl_ops
== NULL
)
1085 arg
.w
.stop
= arg
.w
.skip
= arg
.w
.count
= 0;
1086 arg
.w
.fn
= check_loop_fn
;
1089 q
->ops
->cl_ops
->walk(q
, &arg
.w
);
1090 return arg
.w
.stop
? -ELOOP
: 0;
1094 check_loop_fn(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*w
)
1097 const struct Qdisc_class_ops
*cops
= q
->ops
->cl_ops
;
1098 struct check_loop_arg
*arg
= (struct check_loop_arg
*)w
;
1100 leaf
= cops
->leaf(q
, cl
);
1102 if (leaf
== arg
->p
|| arg
->depth
> 7)
1104 return check_loop(leaf
, arg
->p
, arg
->depth
+ 1);
1113 static int tc_get_qdisc(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1115 struct net
*net
= sock_net(skb
->sk
);
1116 struct tcmsg
*tcm
= nlmsg_data(n
);
1117 struct nlattr
*tca
[TCA_MAX
+ 1];
1118 struct net_device
*dev
;
1120 struct Qdisc
*q
= NULL
;
1121 struct Qdisc
*p
= NULL
;
1124 if ((n
->nlmsg_type
!= RTM_GETQDISC
) &&
1125 !netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1128 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1132 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1136 clid
= tcm
->tcm_parent
;
1138 if (clid
!= TC_H_ROOT
) {
1139 if (TC_H_MAJ(clid
) != TC_H_MAJ(TC_H_INGRESS
)) {
1140 p
= qdisc_lookup(dev
, TC_H_MAJ(clid
));
1143 q
= qdisc_leaf(p
, clid
);
1144 } else if (dev_ingress_queue(dev
)) {
1145 q
= dev_ingress_queue(dev
)->qdisc_sleeping
;
1153 if (tcm
->tcm_handle
&& q
->handle
!= tcm
->tcm_handle
)
1156 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1161 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1164 if (n
->nlmsg_type
== RTM_DELQDISC
) {
1169 err
= qdisc_graft(dev
, p
, skb
, n
, clid
, NULL
, q
);
1173 qdisc_notify(net
, skb
, n
, clid
, NULL
, q
);
1179 * Create/change qdisc.
1182 static int tc_modify_qdisc(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1184 struct net
*net
= sock_net(skb
->sk
);
1186 struct nlattr
*tca
[TCA_MAX
+ 1];
1187 struct net_device
*dev
;
1189 struct Qdisc
*q
, *p
;
1192 if (!netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1196 /* Reinit, just in case something touches this. */
1197 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1201 tcm
= nlmsg_data(n
);
1202 clid
= tcm
->tcm_parent
;
1205 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1211 if (clid
!= TC_H_ROOT
) {
1212 if (clid
!= TC_H_INGRESS
) {
1213 p
= qdisc_lookup(dev
, TC_H_MAJ(clid
));
1216 q
= qdisc_leaf(p
, clid
);
1217 } else if (dev_ingress_queue_create(dev
)) {
1218 q
= dev_ingress_queue(dev
)->qdisc_sleeping
;
1224 /* It may be default qdisc, ignore it */
1225 if (q
&& q
->handle
== 0)
1228 if (!q
|| !tcm
->tcm_handle
|| q
->handle
!= tcm
->tcm_handle
) {
1229 if (tcm
->tcm_handle
) {
1230 if (q
&& !(n
->nlmsg_flags
& NLM_F_REPLACE
))
1232 if (TC_H_MIN(tcm
->tcm_handle
))
1234 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1236 goto create_n_graft
;
1237 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1239 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1242 (p
&& check_loop(q
, p
, 0)))
1244 atomic_inc(&q
->refcnt
);
1248 goto create_n_graft
;
1250 /* This magic test requires explanation.
1252 * We know, that some child q is already
1253 * attached to this parent and have choice:
1254 * either to change it or to create/graft new one.
1256 * 1. We are allowed to create/graft only
1257 * if CREATE and REPLACE flags are set.
1259 * 2. If EXCL is set, requestor wanted to say,
1260 * that qdisc tcm_handle is not expected
1261 * to exist, so that we choose create/graft too.
1263 * 3. The last case is when no flags are set.
1264 * Alas, it is sort of hole in API, we
1265 * cannot decide what to do unambiguously.
1266 * For now we select create/graft, if
1267 * user gave KIND, which does not match existing.
1269 if ((n
->nlmsg_flags
& NLM_F_CREATE
) &&
1270 (n
->nlmsg_flags
& NLM_F_REPLACE
) &&
1271 ((n
->nlmsg_flags
& NLM_F_EXCL
) ||
1273 nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))))
1274 goto create_n_graft
;
1278 if (!tcm
->tcm_handle
)
1280 q
= qdisc_lookup(dev
, tcm
->tcm_handle
);
1283 /* Change qdisc parameters */
1286 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1288 if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], q
->ops
->id
))
1290 err
= qdisc_change(q
, tca
);
1292 qdisc_notify(net
, skb
, n
, clid
, NULL
, q
);
1296 if (!(n
->nlmsg_flags
& NLM_F_CREATE
))
1298 if (clid
== TC_H_INGRESS
) {
1299 if (dev_ingress_queue(dev
))
1300 q
= qdisc_create(dev
, dev_ingress_queue(dev
), p
,
1301 tcm
->tcm_parent
, tcm
->tcm_parent
,
1306 struct netdev_queue
*dev_queue
;
1308 if (p
&& p
->ops
->cl_ops
&& p
->ops
->cl_ops
->select_queue
)
1309 dev_queue
= p
->ops
->cl_ops
->select_queue(p
, tcm
);
1311 dev_queue
= p
->dev_queue
;
1313 dev_queue
= netdev_get_tx_queue(dev
, 0);
1315 q
= qdisc_create(dev
, dev_queue
, p
,
1316 tcm
->tcm_parent
, tcm
->tcm_handle
,
1326 err
= qdisc_graft(dev
, p
, skb
, n
, clid
, q
, NULL
);
1336 static int tc_fill_qdisc(struct sk_buff
*skb
, struct Qdisc
*q
, u32 clid
,
1337 u32 portid
, u32 seq
, u16 flags
, int event
)
1339 struct gnet_stats_basic_cpu __percpu
*cpu_bstats
= NULL
;
1340 struct gnet_stats_queue __percpu
*cpu_qstats
= NULL
;
1342 struct nlmsghdr
*nlh
;
1343 unsigned char *b
= skb_tail_pointer(skb
);
1345 struct qdisc_size_table
*stab
;
1349 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
1351 goto out_nlmsg_trim
;
1352 tcm
= nlmsg_data(nlh
);
1353 tcm
->tcm_family
= AF_UNSPEC
;
1356 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
1357 tcm
->tcm_parent
= clid
;
1358 tcm
->tcm_handle
= q
->handle
;
1359 tcm
->tcm_info
= atomic_read(&q
->refcnt
);
1360 if (nla_put_string(skb
, TCA_KIND
, q
->ops
->id
))
1361 goto nla_put_failure
;
1362 if (q
->ops
->dump
&& q
->ops
->dump(q
, skb
) < 0)
1363 goto nla_put_failure
;
1366 stab
= rtnl_dereference(q
->stab
);
1367 if (stab
&& qdisc_dump_stab(skb
, stab
) < 0)
1368 goto nla_put_failure
;
1370 if (gnet_stats_start_copy_compat(skb
, TCA_STATS2
, TCA_STATS
, TCA_XSTATS
,
1371 NULL
, &d
, TCA_PAD
) < 0)
1372 goto nla_put_failure
;
1374 if (q
->ops
->dump_stats
&& q
->ops
->dump_stats(q
, &d
) < 0)
1375 goto nla_put_failure
;
1377 if (qdisc_is_percpu_stats(q
)) {
1378 cpu_bstats
= q
->cpu_bstats
;
1379 cpu_qstats
= q
->cpu_qstats
;
1382 if (gnet_stats_copy_basic(qdisc_root_sleeping_running(q
),
1383 &d
, cpu_bstats
, &q
->bstats
) < 0 ||
1384 gnet_stats_copy_rate_est(&d
, &q
->bstats
, &q
->rate_est
) < 0 ||
1385 gnet_stats_copy_queue(&d
, cpu_qstats
, &q
->qstats
, qlen
) < 0)
1386 goto nla_put_failure
;
1388 if (gnet_stats_finish_copy(&d
) < 0)
1389 goto nla_put_failure
;
1391 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
1400 static bool tc_qdisc_dump_ignore(struct Qdisc
*q
)
1402 return (q
->flags
& TCQ_F_BUILTIN
) ? true : false;
1405 static int qdisc_notify(struct net
*net
, struct sk_buff
*oskb
,
1406 struct nlmsghdr
*n
, u32 clid
,
1407 struct Qdisc
*old
, struct Qdisc
*new)
1409 struct sk_buff
*skb
;
1410 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1412 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1416 if (old
&& !tc_qdisc_dump_ignore(old
)) {
1417 if (tc_fill_qdisc(skb
, old
, clid
, portid
, n
->nlmsg_seq
,
1418 0, RTM_DELQDISC
) < 0)
1421 if (new && !tc_qdisc_dump_ignore(new)) {
1422 if (tc_fill_qdisc(skb
, new, clid
, portid
, n
->nlmsg_seq
,
1423 old
? NLM_F_REPLACE
: 0, RTM_NEWQDISC
) < 0)
1428 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1429 n
->nlmsg_flags
& NLM_F_ECHO
);
1436 static int tc_dump_qdisc_root(struct Qdisc
*root
, struct sk_buff
*skb
,
1437 struct netlink_callback
*cb
,
1438 int *q_idx_p
, int s_q_idx
)
1440 int ret
= 0, q_idx
= *q_idx_p
;
1448 if (q_idx
< s_q_idx
) {
1451 if (!tc_qdisc_dump_ignore(q
) &&
1452 tc_fill_qdisc(skb
, q
, q
->parent
, NETLINK_CB(cb
->skb
).portid
,
1453 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWQDISC
) <= 0)
1458 if (!qdisc_dev(root
))
1461 hash_for_each(qdisc_dev(root
)->qdisc_hash
, b
, q
, hash
) {
1462 if (q_idx
< s_q_idx
) {
1466 if (!tc_qdisc_dump_ignore(q
) &&
1467 tc_fill_qdisc(skb
, q
, q
->parent
, NETLINK_CB(cb
->skb
).portid
,
1468 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWQDISC
) <= 0)
1481 static int tc_dump_qdisc(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1483 struct net
*net
= sock_net(skb
->sk
);
1486 struct net_device
*dev
;
1488 s_idx
= cb
->args
[0];
1489 s_q_idx
= q_idx
= cb
->args
[1];
1493 for_each_netdev(net
, dev
) {
1494 struct netdev_queue
*dev_queue
;
1502 if (tc_dump_qdisc_root(dev
->qdisc
, skb
, cb
, &q_idx
, s_q_idx
) < 0)
1505 dev_queue
= dev_ingress_queue(dev
);
1507 tc_dump_qdisc_root(dev_queue
->qdisc_sleeping
, skb
, cb
,
1508 &q_idx
, s_q_idx
) < 0)
1517 cb
->args
[1] = q_idx
;
1524 /************************************************
1525 * Traffic classes manipulation. *
1526 ************************************************/
1530 static int tc_ctl_tclass(struct sk_buff
*skb
, struct nlmsghdr
*n
)
1532 struct net
*net
= sock_net(skb
->sk
);
1533 struct tcmsg
*tcm
= nlmsg_data(n
);
1534 struct nlattr
*tca
[TCA_MAX
+ 1];
1535 struct net_device
*dev
;
1536 struct Qdisc
*q
= NULL
;
1537 const struct Qdisc_class_ops
*cops
;
1538 unsigned long cl
= 0;
1539 unsigned long new_cl
;
1545 if ((n
->nlmsg_type
!= RTM_GETTCLASS
) &&
1546 !netlink_ns_capable(skb
, net
->user_ns
, CAP_NET_ADMIN
))
1549 err
= nlmsg_parse(n
, sizeof(*tcm
), tca
, TCA_MAX
, NULL
);
1553 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
1558 parent == TC_H_UNSPEC - unspecified parent.
1559 parent == TC_H_ROOT - class is root, which has no parent.
1560 parent == X:0 - parent is root class.
1561 parent == X:Y - parent is a node in hierarchy.
1562 parent == 0:Y - parent is X:Y, where X:0 is qdisc.
1564 handle == 0:0 - generate handle from kernel pool.
1565 handle == 0:Y - class is X:Y, where X:0 is qdisc.
1566 handle == X:Y - clear.
1567 handle == X:0 - root class.
1570 /* Step 1. Determine qdisc handle X:0 */
1572 portid
= tcm
->tcm_parent
;
1573 clid
= tcm
->tcm_handle
;
1574 qid
= TC_H_MAJ(clid
);
1576 if (portid
!= TC_H_ROOT
) {
1577 u32 qid1
= TC_H_MAJ(portid
);
1580 /* If both majors are known, they must be identical. */
1585 } else if (qid
== 0)
1586 qid
= dev
->qdisc
->handle
;
1588 /* Now qid is genuine qdisc handle consistent
1589 * both with parent and child.
1591 * TC_H_MAJ(portid) still may be unspecified, complete it now.
1594 portid
= TC_H_MAKE(qid
, portid
);
1597 qid
= dev
->qdisc
->handle
;
1600 /* OK. Locate qdisc */
1601 q
= qdisc_lookup(dev
, qid
);
1605 /* An check that it supports classes */
1606 cops
= q
->ops
->cl_ops
;
1610 /* Now try to get class */
1612 if (portid
== TC_H_ROOT
)
1615 clid
= TC_H_MAKE(qid
, clid
);
1618 cl
= cops
->get(q
, clid
);
1622 if (n
->nlmsg_type
!= RTM_NEWTCLASS
||
1623 !(n
->nlmsg_flags
& NLM_F_CREATE
))
1626 switch (n
->nlmsg_type
) {
1629 if (n
->nlmsg_flags
& NLM_F_EXCL
)
1635 err
= cops
->delete(q
, cl
);
1637 tclass_notify(net
, skb
, n
, q
, cl
, RTM_DELTCLASS
);
1640 err
= tclass_notify(net
, skb
, n
, q
, cl
, RTM_NEWTCLASS
);
1651 err
= cops
->change(q
, clid
, portid
, tca
, &new_cl
);
1653 tclass_notify(net
, skb
, n
, q
, new_cl
, RTM_NEWTCLASS
);
1663 static int tc_fill_tclass(struct sk_buff
*skb
, struct Qdisc
*q
,
1665 u32 portid
, u32 seq
, u16 flags
, int event
)
1668 struct nlmsghdr
*nlh
;
1669 unsigned char *b
= skb_tail_pointer(skb
);
1671 const struct Qdisc_class_ops
*cl_ops
= q
->ops
->cl_ops
;
1674 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
1676 goto out_nlmsg_trim
;
1677 tcm
= nlmsg_data(nlh
);
1678 tcm
->tcm_family
= AF_UNSPEC
;
1681 tcm
->tcm_ifindex
= qdisc_dev(q
)->ifindex
;
1682 tcm
->tcm_parent
= q
->handle
;
1683 tcm
->tcm_handle
= q
->handle
;
1685 if (nla_put_string(skb
, TCA_KIND
, q
->ops
->id
))
1686 goto nla_put_failure
;
1687 if (cl_ops
->dump
&& cl_ops
->dump(q
, cl
, skb
, tcm
) < 0)
1688 goto nla_put_failure
;
1690 if (gnet_stats_start_copy_compat(skb
, TCA_STATS2
, TCA_STATS
, TCA_XSTATS
,
1691 NULL
, &d
, TCA_PAD
) < 0)
1692 goto nla_put_failure
;
1694 if (cl_ops
->dump_stats
&& cl_ops
->dump_stats(q
, cl
, &d
) < 0)
1695 goto nla_put_failure
;
1697 if (gnet_stats_finish_copy(&d
) < 0)
1698 goto nla_put_failure
;
1700 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
1709 static int tclass_notify(struct net
*net
, struct sk_buff
*oskb
,
1710 struct nlmsghdr
*n
, struct Qdisc
*q
,
1711 unsigned long cl
, int event
)
1713 struct sk_buff
*skb
;
1714 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
1716 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
1720 if (tc_fill_tclass(skb
, q
, cl
, portid
, n
->nlmsg_seq
, 0, event
) < 0) {
1725 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
1726 n
->nlmsg_flags
& NLM_F_ECHO
);
1729 struct qdisc_dump_args
{
1730 struct qdisc_walker w
;
1731 struct sk_buff
*skb
;
1732 struct netlink_callback
*cb
;
1735 static int qdisc_class_dump(struct Qdisc
*q
, unsigned long cl
, struct qdisc_walker
*arg
)
1737 struct qdisc_dump_args
*a
= (struct qdisc_dump_args
*)arg
;
1739 return tc_fill_tclass(a
->skb
, q
, cl
, NETLINK_CB(a
->cb
->skb
).portid
,
1740 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTCLASS
);
1743 static int tc_dump_tclass_qdisc(struct Qdisc
*q
, struct sk_buff
*skb
,
1744 struct tcmsg
*tcm
, struct netlink_callback
*cb
,
1747 struct qdisc_dump_args arg
;
1749 if (tc_qdisc_dump_ignore(q
) ||
1750 *t_p
< s_t
|| !q
->ops
->cl_ops
||
1752 TC_H_MAJ(tcm
->tcm_parent
) != q
->handle
)) {
1757 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
1758 arg
.w
.fn
= qdisc_class_dump
;
1762 arg
.w
.skip
= cb
->args
[1];
1764 q
->ops
->cl_ops
->walk(q
, &arg
.w
);
1765 cb
->args
[1] = arg
.w
.count
;
1772 static int tc_dump_tclass_root(struct Qdisc
*root
, struct sk_buff
*skb
,
1773 struct tcmsg
*tcm
, struct netlink_callback
*cb
,
1782 if (tc_dump_tclass_qdisc(root
, skb
, tcm
, cb
, t_p
, s_t
) < 0)
1785 if (!qdisc_dev(root
))
1788 hash_for_each(qdisc_dev(root
)->qdisc_hash
, b
, q
, hash
) {
1789 if (tc_dump_tclass_qdisc(q
, skb
, tcm
, cb
, t_p
, s_t
) < 0)
1796 static int tc_dump_tclass(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1798 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
1799 struct net
*net
= sock_net(skb
->sk
);
1800 struct netdev_queue
*dev_queue
;
1801 struct net_device
*dev
;
1804 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
1806 dev
= dev_get_by_index(net
, tcm
->tcm_ifindex
);
1813 if (tc_dump_tclass_root(dev
->qdisc
, skb
, tcm
, cb
, &t
, s_t
) < 0)
1816 dev_queue
= dev_ingress_queue(dev
);
1818 tc_dump_tclass_root(dev_queue
->qdisc_sleeping
, skb
, tcm
, cb
,
1829 /* Main classifier routine: scans classifier chain attached
1830 * to this qdisc, (optionally) tests for protocol and asks
1831 * specific classifiers.
1833 int tc_classify(struct sk_buff
*skb
, const struct tcf_proto
*tp
,
1834 struct tcf_result
*res
, bool compat_mode
)
1836 __be16 protocol
= tc_skb_protocol(skb
);
1837 #ifdef CONFIG_NET_CLS_ACT
1838 const struct tcf_proto
*old_tp
= tp
;
1843 for (; tp
; tp
= rcu_dereference_bh(tp
->next
)) {
1846 if (tp
->protocol
!= protocol
&&
1847 tp
->protocol
!= htons(ETH_P_ALL
))
1850 err
= tp
->classify(skb
, tp
, res
);
1851 #ifdef CONFIG_NET_CLS_ACT
1852 if (unlikely(err
== TC_ACT_RECLASSIFY
&& !compat_mode
))
1859 return TC_ACT_UNSPEC
; /* signal: continue lookup */
1860 #ifdef CONFIG_NET_CLS_ACT
1862 if (unlikely(limit
++ >= MAX_REC_LOOP
)) {
1863 net_notice_ratelimited("%s: reclassify loop, rule prio %u, protocol %02x\n",
1864 tp
->q
->ops
->id
, tp
->prio
& 0xffff,
1865 ntohs(tp
->protocol
));
1870 protocol
= tc_skb_protocol(skb
);
1874 EXPORT_SYMBOL(tc_classify
);
1876 bool tcf_destroy(struct tcf_proto
*tp
, bool force
)
1878 if (tp
->ops
->destroy(tp
, force
)) {
1879 module_put(tp
->ops
->owner
);
1887 void tcf_destroy_chain(struct tcf_proto __rcu
**fl
)
1889 struct tcf_proto
*tp
;
1891 while ((tp
= rtnl_dereference(*fl
)) != NULL
) {
1892 RCU_INIT_POINTER(*fl
, tp
->next
);
1893 tcf_destroy(tp
, true);
1896 EXPORT_SYMBOL(tcf_destroy_chain
);
1898 #ifdef CONFIG_PROC_FS
1899 static int psched_show(struct seq_file
*seq
, void *v
)
1901 seq_printf(seq
, "%08x %08x %08x %08x\n",
1902 (u32
)NSEC_PER_USEC
, (u32
)PSCHED_TICKS2NS(1),
1904 (u32
)NSEC_PER_SEC
/ hrtimer_resolution
);
1909 static int psched_open(struct inode
*inode
, struct file
*file
)
1911 return single_open(file
, psched_show
, NULL
);
1914 static const struct file_operations psched_fops
= {
1915 .owner
= THIS_MODULE
,
1916 .open
= psched_open
,
1918 .llseek
= seq_lseek
,
1919 .release
= single_release
,
1922 static int __net_init
psched_net_init(struct net
*net
)
1924 struct proc_dir_entry
*e
;
1926 e
= proc_create("psched", 0, net
->proc_net
, &psched_fops
);
1933 static void __net_exit
psched_net_exit(struct net
*net
)
1935 remove_proc_entry("psched", net
->proc_net
);
1938 static int __net_init
psched_net_init(struct net
*net
)
1943 static void __net_exit
psched_net_exit(struct net
*net
)
1948 static struct pernet_operations psched_net_ops
= {
1949 .init
= psched_net_init
,
1950 .exit
= psched_net_exit
,
1953 static int __init
pktsched_init(void)
1957 err
= register_pernet_subsys(&psched_net_ops
);
1959 pr_err("pktsched_init: "
1960 "cannot initialize per netns operations\n");
1964 register_qdisc(&pfifo_fast_ops
);
1965 register_qdisc(&pfifo_qdisc_ops
);
1966 register_qdisc(&bfifo_qdisc_ops
);
1967 register_qdisc(&pfifo_head_drop_qdisc_ops
);
1968 register_qdisc(&mq_qdisc_ops
);
1969 register_qdisc(&noqueue_qdisc_ops
);
1971 rtnl_register(PF_UNSPEC
, RTM_NEWQDISC
, tc_modify_qdisc
, NULL
, NULL
);
1972 rtnl_register(PF_UNSPEC
, RTM_DELQDISC
, tc_get_qdisc
, NULL
, NULL
);
1973 rtnl_register(PF_UNSPEC
, RTM_GETQDISC
, tc_get_qdisc
, tc_dump_qdisc
, NULL
);
1974 rtnl_register(PF_UNSPEC
, RTM_NEWTCLASS
, tc_ctl_tclass
, NULL
, NULL
);
1975 rtnl_register(PF_UNSPEC
, RTM_DELTCLASS
, tc_ctl_tclass
, NULL
, NULL
);
1976 rtnl_register(PF_UNSPEC
, RTM_GETTCLASS
, tc_ctl_tclass
, tc_dump_tclass
, NULL
);
1981 subsys_initcall(pktsched_init
);