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 <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/string.h>
21 #include <linux/errno.h>
22 #include <linux/skbuff.h>
23 #include <linux/init.h>
24 #include <linux/kmod.h>
25 #include <linux/netlink.h>
26 #include <net/net_namespace.h>
28 #include <net/netlink.h>
29 #include <net/pkt_sched.h>
30 #include <net/pkt_cls.h>
32 /* The list of all installed classifier types */
34 static struct tcf_proto_ops
*tcf_proto_base __read_mostly
;
36 /* Protects list of registered TC modules. It is pure SMP lock. */
37 static DEFINE_RWLOCK(cls_mod_lock
);
39 /* Find classifier type by string name */
41 static struct tcf_proto_ops
*tcf_proto_lookup_ops(struct rtattr
*kind
)
43 struct tcf_proto_ops
*t
= NULL
;
46 read_lock(&cls_mod_lock
);
47 for (t
= tcf_proto_base
; t
; t
= t
->next
) {
48 if (rtattr_strcmp(kind
, t
->kind
) == 0) {
49 if (!try_module_get(t
->owner
))
54 read_unlock(&cls_mod_lock
);
59 /* Register(unregister) new classifier type */
61 int register_tcf_proto_ops(struct tcf_proto_ops
*ops
)
63 struct tcf_proto_ops
*t
, **tp
;
66 write_lock(&cls_mod_lock
);
67 for (tp
= &tcf_proto_base
; (t
= *tp
) != NULL
; tp
= &t
->next
)
68 if (!strcmp(ops
->kind
, t
->kind
))
75 write_unlock(&cls_mod_lock
);
78 EXPORT_SYMBOL(register_tcf_proto_ops
);
80 int unregister_tcf_proto_ops(struct tcf_proto_ops
*ops
)
82 struct tcf_proto_ops
*t
, **tp
;
85 write_lock(&cls_mod_lock
);
86 for (tp
= &tcf_proto_base
; (t
=*tp
) != NULL
; tp
= &t
->next
)
95 write_unlock(&cls_mod_lock
);
98 EXPORT_SYMBOL(unregister_tcf_proto_ops
);
100 static int tfilter_notify(struct sk_buff
*oskb
, struct nlmsghdr
*n
,
101 struct tcf_proto
*tp
, unsigned long fh
, int event
);
104 /* Select new prio value from the range, managed by kernel. */
106 static inline u32
tcf_auto_prio(struct tcf_proto
*tp
)
108 u32 first
= TC_H_MAKE(0xC0000000U
, 0U);
116 /* Add/change/delete/get a filter node */
118 static int tc_ctl_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
, void *arg
)
120 struct net
*net
= skb
->sk
->sk_net
;
127 struct net_device
*dev
;
129 struct tcf_proto
**back
, **chain
;
130 struct tcf_proto
*tp
;
131 struct tcf_proto_ops
*tp_ops
;
132 const struct Qdisc_class_ops
*cops
;
137 if (net
!= &init_net
)
143 protocol
= TC_H_MIN(t
->tcm_info
);
144 prio
= TC_H_MAJ(t
->tcm_info
);
146 parent
= t
->tcm_parent
;
150 /* If no priority is given, user wants we allocated it. */
151 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
153 prio
= TC_H_MAKE(0x80000000U
, 0U);
156 /* Find head of filter chain. */
159 dev
= __dev_get_by_index(&init_net
, t
->tcm_ifindex
);
165 q
= dev
->qdisc_sleeping
;
168 q
= qdisc_lookup(dev
, TC_H_MAJ(t
->tcm_parent
));
173 /* Is it classful? */
174 if ((cops
= q
->ops
->cl_ops
) == NULL
)
177 /* Do we search for filter, attached to class? */
178 if (TC_H_MIN(parent
)) {
179 cl
= cops
->get(q
, parent
);
184 /* And the last stroke */
185 chain
= cops
->tcf_chain(q
, cl
);
190 /* Check the chain for existence of proto-tcf with this priority */
191 for (back
= chain
; (tp
=*back
) != NULL
; back
= &tp
->next
) {
192 if (tp
->prio
>= prio
) {
193 if (tp
->prio
== prio
) {
194 if (!nprio
|| (tp
->protocol
!= protocol
&& protocol
))
203 /* Proto-tcf does not exist, create new one */
205 if (tca
[TCA_KIND
-1] == NULL
|| !protocol
)
209 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
213 /* Create new proto tcf */
216 tp
= kzalloc(sizeof(*tp
), GFP_KERNEL
);
220 tp_ops
= tcf_proto_lookup_ops(tca
[TCA_KIND
-1]);
221 if (tp_ops
== NULL
) {
223 struct rtattr
*kind
= tca
[TCA_KIND
-1];
227 rtattr_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
229 request_module("cls_%s", name
);
231 tp_ops
= tcf_proto_lookup_ops(kind
);
232 /* We dropped the RTNL semaphore in order to
233 * perform the module load. So, even if we
234 * succeeded in loading the module we have to
235 * replay the request. We indicate this using
238 if (tp_ops
!= NULL
) {
239 module_put(tp_ops
->owner
);
248 tp
->protocol
= protocol
;
249 tp
->prio
= nprio
? : tcf_auto_prio(*back
);
251 tp
->classify
= tp_ops
->classify
;
252 tp
->classid
= parent
;
254 err
= tp_ops
->init(tp
);
256 module_put(tp_ops
->owner
);
261 qdisc_lock_tree(dev
);
264 qdisc_unlock_tree(dev
);
266 } else if (tca
[TCA_KIND
-1] && rtattr_strcmp(tca
[TCA_KIND
-1], tp
->ops
->kind
))
269 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
272 if (n
->nlmsg_type
== RTM_DELTFILTER
&& t
->tcm_handle
== 0) {
273 qdisc_lock_tree(dev
);
275 qdisc_unlock_tree(dev
);
277 tfilter_notify(skb
, n
, tp
, fh
, RTM_DELTFILTER
);
284 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
285 !(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. */
321 static int tcf_fill_node(struct sk_buff
*skb
, struct tcf_proto
*tp
,
322 unsigned long fh
, u32 pid
, u32 seq
, u16 flags
, int event
)
325 struct nlmsghdr
*nlh
;
326 unsigned char *b
= skb_tail_pointer(skb
);
328 nlh
= NLMSG_NEW(skb
, pid
, seq
, event
, sizeof(*tcm
), flags
);
329 tcm
= NLMSG_DATA(nlh
);
330 tcm
->tcm_family
= AF_UNSPEC
;
333 tcm
->tcm_ifindex
= tp
->q
->dev
->ifindex
;
334 tcm
->tcm_parent
= tp
->classid
;
335 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
336 RTA_PUT(skb
, TCA_KIND
, IFNAMSIZ
, tp
->ops
->kind
);
337 tcm
->tcm_handle
= fh
;
338 if (RTM_DELTFILTER
!= event
) {
340 if (tp
->ops
->dump
&& tp
->ops
->dump(tp
, fh
, skb
, tcm
) < 0)
343 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
352 static int tfilter_notify(struct sk_buff
*oskb
, struct nlmsghdr
*n
,
353 struct tcf_proto
*tp
, unsigned long fh
, int event
)
356 u32 pid
= oskb
? NETLINK_CB(oskb
).pid
: 0;
358 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
362 if (tcf_fill_node(skb
, tp
, fh
, pid
, n
->nlmsg_seq
, 0, event
) <= 0) {
367 return rtnetlink_send(skb
, &init_net
, pid
, RTNLGRP_TC
,
368 n
->nlmsg_flags
& NLM_F_ECHO
);
371 struct tcf_dump_args
{
374 struct netlink_callback
*cb
;
377 static int tcf_node_dump(struct tcf_proto
*tp
, unsigned long n
,
378 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
)
388 struct net
*net
= skb
->sk
->sk_net
;
391 struct net_device
*dev
;
393 struct tcf_proto
*tp
, **chain
;
394 struct tcmsg
*tcm
= (struct tcmsg
*)NLMSG_DATA(cb
->nlh
);
395 unsigned long cl
= 0;
396 const struct Qdisc_class_ops
*cops
;
397 struct tcf_dump_args arg
;
399 if (net
!= &init_net
)
402 if (cb
->nlh
->nlmsg_len
< NLMSG_LENGTH(sizeof(*tcm
)))
404 if ((dev
= dev_get_by_index(&init_net
, tcm
->tcm_ifindex
)) == NULL
)
407 if (!tcm
->tcm_parent
)
408 q
= dev
->qdisc_sleeping
;
410 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
413 if ((cops
= q
->ops
->cl_ops
) == NULL
)
415 if (TC_H_MIN(tcm
->tcm_parent
)) {
416 cl
= cops
->get(q
, tcm
->tcm_parent
);
420 chain
= cops
->tcf_chain(q
, cl
);
426 for (tp
=*chain
, t
=0; tp
; tp
= tp
->next
, t
++) {
427 if (t
< s_t
) continue;
428 if (TC_H_MAJ(tcm
->tcm_info
) &&
429 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
431 if (TC_H_MIN(tcm
->tcm_info
) &&
432 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
435 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
436 if (cb
->args
[1] == 0) {
437 if (tcf_fill_node(skb
, tp
, 0, NETLINK_CB(cb
->skb
).pid
,
438 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
439 RTM_NEWTFILTER
) <= 0)
444 if (tp
->ops
->walk
== NULL
)
446 arg
.w
.fn
= tcf_node_dump
;
450 arg
.w
.skip
= cb
->args
[1]-1;
452 tp
->ops
->walk(tp
, &arg
.w
);
453 cb
->args
[1] = arg
.w
.count
+1;
468 void tcf_exts_destroy(struct tcf_proto
*tp
, struct tcf_exts
*exts
)
470 #ifdef CONFIG_NET_CLS_ACT
472 tcf_action_destroy(exts
->action
, TCA_ACT_UNBIND
);
477 EXPORT_SYMBOL(tcf_exts_destroy
);
479 int tcf_exts_validate(struct tcf_proto
*tp
, struct rtattr
**tb
,
480 struct rtattr
*rate_tlv
, struct tcf_exts
*exts
,
481 struct tcf_ext_map
*map
)
483 memset(exts
, 0, sizeof(*exts
));
485 #ifdef CONFIG_NET_CLS_ACT
488 struct tc_action
*act
;
490 if (map
->police
&& tb
[map
->police
-1]) {
491 act
= tcf_action_init_1(tb
[map
->police
-1], rate_tlv
,
492 "police", TCA_ACT_NOREPLACE
,
497 act
->type
= TCA_OLD_COMPAT
;
499 } else if (map
->action
&& tb
[map
->action
-1]) {
500 act
= tcf_action_init(tb
[map
->action
-1], rate_tlv
, NULL
,
501 TCA_ACT_NOREPLACE
, TCA_ACT_BIND
, &err
);
509 if ((map
->action
&& tb
[map
->action
-1]) ||
510 (map
->police
&& tb
[map
->police
-1]))
516 EXPORT_SYMBOL(tcf_exts_validate
);
518 void tcf_exts_change(struct tcf_proto
*tp
, struct tcf_exts
*dst
,
519 struct tcf_exts
*src
)
521 #ifdef CONFIG_NET_CLS_ACT
523 struct tc_action
*act
;
525 act
= xchg(&dst
->action
, src
->action
);
528 tcf_action_destroy(act
, TCA_ACT_UNBIND
);
532 EXPORT_SYMBOL(tcf_exts_change
);
534 int tcf_exts_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
,
535 struct tcf_ext_map
*map
)
537 #ifdef CONFIG_NET_CLS_ACT
538 if (map
->action
&& exts
->action
) {
540 * again for backward compatible mode - we want
541 * to work with both old and new modes of entering
542 * tc data even if iproute2 was newer - jhs
544 struct rtattr
*p_rta
= (struct rtattr
*)skb_tail_pointer(skb
);
546 if (exts
->action
->type
!= TCA_OLD_COMPAT
) {
547 RTA_PUT(skb
, map
->action
, 0, NULL
);
548 if (tcf_action_dump(skb
, exts
->action
, 0, 0) < 0)
550 p_rta
->rta_len
= skb_tail_pointer(skb
) - (u8
*)p_rta
;
551 } else if (map
->police
) {
552 RTA_PUT(skb
, map
->police
, 0, NULL
);
553 if (tcf_action_dump_old(skb
, exts
->action
, 0, 0) < 0)
555 p_rta
->rta_len
= skb_tail_pointer(skb
) - (u8
*)p_rta
;
560 rtattr_failure
: __attribute__ ((unused
))
563 EXPORT_SYMBOL(tcf_exts_dump
);
566 int tcf_exts_dump_stats(struct sk_buff
*skb
, struct tcf_exts
*exts
,
567 struct tcf_ext_map
*map
)
569 #ifdef CONFIG_NET_CLS_ACT
571 if (tcf_action_copy_stats(skb
, exts
->action
, 1) < 0)
575 rtattr_failure
: __attribute__ ((unused
))
578 EXPORT_SYMBOL(tcf_exts_dump_stats
);
580 static int __init
tc_filter_init(void)
582 rtnl_register(PF_UNSPEC
, RTM_NEWTFILTER
, tc_ctl_tfilter
, NULL
);
583 rtnl_register(PF_UNSPEC
, RTM_DELTFILTER
, tc_ctl_tfilter
, NULL
);
584 rtnl_register(PF_UNSPEC
, RTM_GETTFILTER
, tc_ctl_tfilter
,
590 subsys_initcall(tc_filter_init
);