2 * net/sched/cls_api.c Packet classifier 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 * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
17 #include <asm/uaccess.h>
18 #include <asm/system.h>
19 #include <linux/bitops.h>
20 #include <linux/module.h>
21 #include <linux/types.h>
22 #include <linux/kernel.h>
23 #include <linux/string.h>
25 #include <linux/socket.h>
26 #include <linux/sockios.h>
28 #include <linux/errno.h>
29 #include <linux/interrupt.h>
30 #include <linux/netdevice.h>
31 #include <linux/skbuff.h>
32 #include <linux/init.h>
33 #include <linux/kmod.h>
34 #include <linux/netlink.h>
35 #include <net/netlink.h>
37 #include <net/pkt_sched.h>
38 #include <net/pkt_cls.h>
41 #define DPRINTK(format,args...) printk(KERN_DEBUG format,##args)
43 #define DPRINTK(format,args...)
46 /* The list of all installed classifier types */
48 static struct tcf_proto_ops
*tcf_proto_base
;
50 /* Protects list of registered TC modules. It is pure SMP lock. */
51 static DEFINE_RWLOCK(cls_mod_lock
);
53 /* Find classifier type by string name */
55 static struct tcf_proto_ops
* tcf_proto_lookup_ops(struct rtattr
*kind
)
57 struct tcf_proto_ops
*t
= NULL
;
60 read_lock(&cls_mod_lock
);
61 for (t
= tcf_proto_base
; t
; t
= t
->next
) {
62 if (rtattr_strcmp(kind
, t
->kind
) == 0) {
63 if (!try_module_get(t
->owner
))
68 read_unlock(&cls_mod_lock
);
73 /* Register(unregister) new classifier type */
75 int register_tcf_proto_ops(struct tcf_proto_ops
*ops
)
77 struct tcf_proto_ops
*t
, **tp
;
80 write_lock(&cls_mod_lock
);
81 for (tp
= &tcf_proto_base
; (t
= *tp
) != NULL
; tp
= &t
->next
)
82 if (!strcmp(ops
->kind
, t
->kind
))
89 write_unlock(&cls_mod_lock
);
93 int unregister_tcf_proto_ops(struct tcf_proto_ops
*ops
)
95 struct tcf_proto_ops
*t
, **tp
;
98 write_lock(&cls_mod_lock
);
99 for (tp
= &tcf_proto_base
; (t
=*tp
) != NULL
; tp
= &t
->next
)
108 write_unlock(&cls_mod_lock
);
112 static int tfilter_notify(struct sk_buff
*oskb
, struct nlmsghdr
*n
,
113 struct tcf_proto
*tp
, unsigned long fh
, int event
);
116 /* Select new prio value from the range, managed by kernel. */
118 static __inline__ u32
tcf_auto_prio(struct tcf_proto
*tp
)
120 u32 first
= TC_H_MAKE(0xC0000000U
,0U);
128 /* Add/change/delete/get a filter node */
130 static int tc_ctl_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
, void *arg
)
138 struct net_device
*dev
;
140 struct tcf_proto
**back
, **chain
;
141 struct tcf_proto
*tp
;
142 struct tcf_proto_ops
*tp_ops
;
143 struct Qdisc_class_ops
*cops
;
151 protocol
= TC_H_MIN(t
->tcm_info
);
152 prio
= TC_H_MAJ(t
->tcm_info
);
154 parent
= t
->tcm_parent
;
158 /* If no priority is given, user wants we allocated it. */
159 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
161 prio
= TC_H_MAKE(0x80000000U
,0U);
164 /* Find head of filter chain. */
167 if ((dev
= __dev_get_by_index(t
->tcm_ifindex
)) == NULL
)
172 q
= dev
->qdisc_sleeping
;
174 } else if ((q
= qdisc_lookup(dev
, TC_H_MAJ(t
->tcm_parent
))) == NULL
)
177 /* Is it classful? */
178 if ((cops
= q
->ops
->cl_ops
) == NULL
)
181 /* Do we search for filter, attached to class? */
182 if (TC_H_MIN(parent
)) {
183 cl
= cops
->get(q
, parent
);
188 /* And the last stroke */
189 chain
= cops
->tcf_chain(q
, cl
);
194 /* Check the chain for existence of proto-tcf with this priority */
195 for (back
= chain
; (tp
=*back
) != NULL
; back
= &tp
->next
) {
196 if (tp
->prio
>= prio
) {
197 if (tp
->prio
== prio
) {
198 if (!nprio
|| (tp
->protocol
!= protocol
&& protocol
))
207 /* Proto-tcf does not exist, create new one */
209 if (tca
[TCA_KIND
-1] == NULL
|| !protocol
)
213 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
217 /* Create new proto tcf */
220 if ((tp
= kzalloc(sizeof(*tp
), GFP_KERNEL
)) == NULL
)
223 tp_ops
= tcf_proto_lookup_ops(tca
[TCA_KIND
-1]);
224 if (tp_ops
== NULL
) {
226 struct rtattr
*kind
= tca
[TCA_KIND
-1];
230 rtattr_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
232 request_module("cls_%s", name
);
234 tp_ops
= tcf_proto_lookup_ops(kind
);
235 /* We dropped the RTNL semaphore in order to
236 * perform the module load. So, even if we
237 * succeeded in loading the module we have to
238 * replay the request. We indicate this using
241 if (tp_ops
!= NULL
) {
242 module_put(tp_ops
->owner
);
251 tp
->protocol
= protocol
;
252 tp
->prio
= nprio
? : tcf_auto_prio(*back
);
254 tp
->classify
= tp_ops
->classify
;
255 tp
->classid
= parent
;
256 if ((err
= tp_ops
->init(tp
)) != 0) {
257 module_put(tp_ops
->owner
);
262 qdisc_lock_tree(dev
);
265 qdisc_unlock_tree(dev
);
267 } else if (tca
[TCA_KIND
-1] && rtattr_strcmp(tca
[TCA_KIND
-1], tp
->ops
->kind
))
270 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
273 if (n
->nlmsg_type
== RTM_DELTFILTER
&& t
->tcm_handle
== 0) {
274 qdisc_lock_tree(dev
);
276 qdisc_unlock_tree(dev
);
278 tfilter_notify(skb
, n
, tp
, fh
, RTM_DELTFILTER
);
285 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
288 switch (n
->nlmsg_type
) {
291 if (n
->nlmsg_flags
&NLM_F_EXCL
)
295 err
= tp
->ops
->delete(tp
, fh
);
297 tfilter_notify(skb
, n
, tp
, fh
, RTM_DELTFILTER
);
300 err
= tfilter_notify(skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
308 err
= tp
->ops
->change(tp
, cl
, t
->tcm_handle
, tca
, &fh
);
310 tfilter_notify(skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
316 /* Replay the request. */
322 tcf_fill_node(struct sk_buff
*skb
, struct tcf_proto
*tp
, unsigned long fh
,
323 u32 pid
, u32 seq
, u16 flags
, int event
)
326 struct nlmsghdr
*nlh
;
327 unsigned char *b
= skb_tail_pointer(skb
);
329 nlh
= NLMSG_NEW(skb
, pid
, seq
, event
, sizeof(*tcm
), flags
);
330 tcm
= NLMSG_DATA(nlh
);
331 tcm
->tcm_family
= AF_UNSPEC
;
334 tcm
->tcm_ifindex
= tp
->q
->dev
->ifindex
;
335 tcm
->tcm_parent
= tp
->classid
;
336 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
337 RTA_PUT(skb
, TCA_KIND
, IFNAMSIZ
, tp
->ops
->kind
);
338 tcm
->tcm_handle
= fh
;
339 if (RTM_DELTFILTER
!= event
) {
341 if (tp
->ops
->dump
&& tp
->ops
->dump(tp
, fh
, skb
, tcm
) < 0)
344 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
353 static int tfilter_notify(struct sk_buff
*oskb
, struct nlmsghdr
*n
,
354 struct tcf_proto
*tp
, unsigned long fh
, int event
)
357 u32 pid
= oskb
? NETLINK_CB(oskb
).pid
: 0;
359 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
363 if (tcf_fill_node(skb
, tp
, fh
, pid
, n
->nlmsg_seq
, 0, event
) <= 0) {
368 return rtnetlink_send(skb
, pid
, RTNLGRP_TC
, n
->nlmsg_flags
&NLM_F_ECHO
);
375 struct netlink_callback
*cb
;
378 static int tcf_node_dump(struct tcf_proto
*tp
, unsigned long n
, struct tcf_walker
*arg
)
380 struct tcf_dump_args
*a
= (void*)arg
;
382 return tcf_fill_node(a
->skb
, tp
, n
, NETLINK_CB(a
->cb
->skb
).pid
,
383 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTFILTER
);
386 static int tc_dump_tfilter(struct sk_buff
*skb
, struct netlink_callback
*cb
)
390 struct net_device
*dev
;
392 struct tcf_proto
*tp
, **chain
;
393 struct tcmsg
*tcm
= (struct tcmsg
*)NLMSG_DATA(cb
->nlh
);
394 unsigned long cl
= 0;
395 struct Qdisc_class_ops
*cops
;
396 struct tcf_dump_args arg
;
398 if (cb
->nlh
->nlmsg_len
< NLMSG_LENGTH(sizeof(*tcm
)))
400 if ((dev
= dev_get_by_index(tcm
->tcm_ifindex
)) == NULL
)
403 read_lock(&qdisc_tree_lock
);
404 if (!tcm
->tcm_parent
)
405 q
= dev
->qdisc_sleeping
;
407 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
410 if ((cops
= q
->ops
->cl_ops
) == NULL
)
412 if (TC_H_MIN(tcm
->tcm_parent
)) {
413 cl
= cops
->get(q
, tcm
->tcm_parent
);
417 chain
= cops
->tcf_chain(q
, cl
);
423 for (tp
=*chain
, t
=0; tp
; tp
= tp
->next
, t
++) {
424 if (t
< s_t
) continue;
425 if (TC_H_MAJ(tcm
->tcm_info
) &&
426 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
428 if (TC_H_MIN(tcm
->tcm_info
) &&
429 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
432 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
433 if (cb
->args
[1] == 0) {
434 if (tcf_fill_node(skb
, tp
, 0, NETLINK_CB(cb
->skb
).pid
,
435 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTFILTER
) <= 0) {
440 if (tp
->ops
->walk
== NULL
)
442 arg
.w
.fn
= tcf_node_dump
;
446 arg
.w
.skip
= cb
->args
[1]-1;
448 tp
->ops
->walk(tp
, &arg
.w
);
449 cb
->args
[1] = arg
.w
.count
+1;
460 read_unlock(&qdisc_tree_lock
);
466 tcf_exts_destroy(struct tcf_proto
*tp
, struct tcf_exts
*exts
)
468 #ifdef CONFIG_NET_CLS_ACT
470 tcf_action_destroy(exts
->action
, TCA_ACT_UNBIND
);
473 #elif defined CONFIG_NET_CLS_POLICE
475 tcf_police_release(exts
->police
, TCA_ACT_UNBIND
);
483 tcf_exts_validate(struct tcf_proto
*tp
, struct rtattr
**tb
,
484 struct rtattr
*rate_tlv
, struct tcf_exts
*exts
,
485 struct tcf_ext_map
*map
)
487 memset(exts
, 0, sizeof(*exts
));
489 #ifdef CONFIG_NET_CLS_ACT
492 struct tc_action
*act
;
494 if (map
->police
&& tb
[map
->police
-1]) {
495 act
= tcf_action_init_1(tb
[map
->police
-1], rate_tlv
, "police",
496 TCA_ACT_NOREPLACE
, TCA_ACT_BIND
, &err
);
500 act
->type
= TCA_OLD_COMPAT
;
502 } else if (map
->action
&& tb
[map
->action
-1]) {
503 act
= tcf_action_init(tb
[map
->action
-1], rate_tlv
, NULL
,
504 TCA_ACT_NOREPLACE
, TCA_ACT_BIND
, &err
);
511 #elif defined CONFIG_NET_CLS_POLICE
512 if (map
->police
&& tb
[map
->police
-1]) {
513 struct tcf_police
*p
;
515 p
= tcf_police_locate(tb
[map
->police
-1], rate_tlv
);
520 } else if (map
->action
&& tb
[map
->action
-1])
523 if ((map
->action
&& tb
[map
->action
-1]) ||
524 (map
->police
&& tb
[map
->police
-1]))
532 tcf_exts_change(struct tcf_proto
*tp
, struct tcf_exts
*dst
,
533 struct tcf_exts
*src
)
535 #ifdef CONFIG_NET_CLS_ACT
537 struct tc_action
*act
;
539 act
= xchg(&dst
->action
, src
->action
);
542 tcf_action_destroy(act
, TCA_ACT_UNBIND
);
544 #elif defined CONFIG_NET_CLS_POLICE
546 struct tcf_police
*p
;
548 p
= xchg(&dst
->police
, src
->police
);
551 tcf_police_release(p
, TCA_ACT_UNBIND
);
557 tcf_exts_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
,
558 struct tcf_ext_map
*map
)
560 #ifdef CONFIG_NET_CLS_ACT
561 if (map
->action
&& exts
->action
) {
563 * again for backward compatible mode - we want
564 * to work with both old and new modes of entering
565 * tc data even if iproute2 was newer - jhs
567 struct rtattr
*p_rta
= (struct rtattr
*)skb_tail_pointer(skb
);
569 if (exts
->action
->type
!= TCA_OLD_COMPAT
) {
570 RTA_PUT(skb
, map
->action
, 0, NULL
);
571 if (tcf_action_dump(skb
, exts
->action
, 0, 0) < 0)
573 p_rta
->rta_len
= skb_tail_pointer(skb
) - (u8
*)p_rta
;
574 } else if (map
->police
) {
575 RTA_PUT(skb
, map
->police
, 0, NULL
);
576 if (tcf_action_dump_old(skb
, exts
->action
, 0, 0) < 0)
578 p_rta
->rta_len
= skb_tail_pointer(skb
) - (u8
*)p_rta
;
581 #elif defined CONFIG_NET_CLS_POLICE
582 if (map
->police
&& exts
->police
) {
583 struct rtattr
*p_rta
= (struct rtattr
*)skb_tail_pointer(skb
);
585 RTA_PUT(skb
, map
->police
, 0, NULL
);
587 if (tcf_police_dump(skb
, exts
->police
) < 0)
590 p_rta
->rta_len
= skb_tail_pointer(skb
) - (u8
*)p_rta
;
594 rtattr_failure
: __attribute__ ((unused
))
599 tcf_exts_dump_stats(struct sk_buff
*skb
, struct tcf_exts
*exts
,
600 struct tcf_ext_map
*map
)
602 #ifdef CONFIG_NET_CLS_ACT
604 if (tcf_action_copy_stats(skb
, exts
->action
, 1) < 0)
606 #elif defined CONFIG_NET_CLS_POLICE
608 if (tcf_police_dump_stats(skb
, exts
->police
) < 0)
612 rtattr_failure
: __attribute__ ((unused
))
616 static int __init
tc_filter_init(void)
618 rtnl_register(PF_UNSPEC
, RTM_NEWTFILTER
, tc_ctl_tfilter
, NULL
);
619 rtnl_register(PF_UNSPEC
, RTM_DELTFILTER
, tc_ctl_tfilter
, NULL
);
620 rtnl_register(PF_UNSPEC
, RTM_GETTFILTER
, tc_ctl_tfilter
,
626 subsys_initcall(tc_filter_init
);
628 EXPORT_SYMBOL(register_tcf_proto_ops
);
629 EXPORT_SYMBOL(unregister_tcf_proto_ops
);
630 EXPORT_SYMBOL(tcf_exts_validate
);
631 EXPORT_SYMBOL(tcf_exts_destroy
);
632 EXPORT_SYMBOL(tcf_exts_change
);
633 EXPORT_SYMBOL(tcf_exts_dump
);
634 EXPORT_SYMBOL(tcf_exts_dump_stats
);