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
16 #include <asm/uaccess.h>
17 #include <asm/system.h>
18 #include <asm/bitops.h>
19 #include <linux/config.h>
20 #include <linux/types.h>
21 #include <linux/kernel.h>
22 #include <linux/sched.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/rtnetlink.h>
33 #include <linux/init.h>
34 #include <linux/kmod.h>
36 #include <net/pkt_sched.h>
38 /* The list of all installed classifier types */
40 static struct tcf_proto_ops
*tcf_proto_base
;
42 /* Protects list of registered TC modules. It is pure SMP lock. */
43 static rwlock_t cls_mod_lock
= RW_LOCK_UNLOCKED
;
45 /* Find classifier type by string name */
47 struct tcf_proto_ops
* tcf_proto_lookup_ops(struct rtattr
*kind
)
49 struct tcf_proto_ops
*t
= NULL
;
52 read_lock(&cls_mod_lock
);
53 for (t
= tcf_proto_base
; t
; t
= t
->next
) {
54 if (rtattr_strcmp(kind
, t
->kind
) == 0)
57 read_unlock(&cls_mod_lock
);
62 /* Register(unregister) new classifier type */
64 int register_tcf_proto_ops(struct tcf_proto_ops
*ops
)
66 struct tcf_proto_ops
*t
, **tp
;
68 write_lock(&cls_mod_lock
);
69 for (tp
= &tcf_proto_base
; (t
=*tp
) != NULL
; tp
= &t
->next
) {
70 if (strcmp(ops
->kind
, t
->kind
) == 0) {
71 write_unlock(&cls_mod_lock
);
78 write_unlock(&cls_mod_lock
);
82 int unregister_tcf_proto_ops(struct tcf_proto_ops
*ops
)
84 struct tcf_proto_ops
*t
, **tp
;
86 write_lock(&cls_mod_lock
);
87 for (tp
= &tcf_proto_base
; (t
=*tp
) != NULL
; tp
= &t
->next
)
92 write_unlock(&cls_mod_lock
);
96 write_unlock(&cls_mod_lock
);
100 #ifdef CONFIG_RTNETLINK
102 static int tfilter_notify(struct sk_buff
*oskb
, struct nlmsghdr
*n
,
103 struct tcf_proto
*tp
, unsigned long fh
, int event
);
106 /* Select new prio value from the range, managed by kernel. */
108 static __inline__ u32
tcf_auto_prio(struct tcf_proto
*tp
)
110 u32 first
= TC_H_MAKE(0xC0000000U
,0U);
118 /* Add/change/delete/get a filter node */
120 static int tc_ctl_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
, void *arg
)
122 struct rtattr
**tca
= arg
;
123 struct tcmsg
*t
= NLMSG_DATA(n
);
124 u32 protocol
= TC_H_MIN(t
->tcm_info
);
125 u32 prio
= TC_H_MAJ(t
->tcm_info
);
127 u32 parent
= t
->tcm_parent
;
130 struct tcf_proto
**back
, **chain
;
131 struct tcf_proto
*tp
= NULL
;
132 struct tcf_proto_ops
*tp_ops
;
133 struct Qdisc_class_ops
*cops
;
134 unsigned long cl
= 0;
139 /* If no priority is given, user wants we allocated it. */
140 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
142 prio
= TC_H_MAKE(0x80000000U
,0U);
145 /* Find head of filter chain. */
148 if ((dev
= dev_get_by_index(t
->tcm_ifindex
)) == NULL
)
153 q
= dev
->qdisc_sleeping
;
155 } else if ((q
= qdisc_lookup(dev
, TC_H_MAJ(t
->tcm_parent
))) == NULL
)
158 /* Is it classful? */
159 if ((cops
= q
->ops
->cl_ops
) == NULL
)
162 /* Do we search for filter, attached to class? */
163 if (TC_H_MIN(parent
)) {
164 cl
= cops
->get(q
, parent
);
169 /* And the last stroke */
170 chain
= cops
->tcf_chain(q
, cl
);
175 /* Check the chain for existence of proto-tcf with this priority */
176 for (back
= chain
; (tp
=*back
) != NULL
; back
= &tp
->next
) {
177 if (tp
->prio
>= prio
) {
178 if (tp
->prio
== prio
) {
179 if (!nprio
|| (tp
->protocol
!= protocol
&& protocol
))
188 /* Proto-tcf does not exist, create new one */
190 if (tca
[TCA_KIND
-1] == NULL
|| !protocol
)
194 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
198 /* Create new proto tcf */
201 if ((tp
= kmalloc(sizeof(*tp
), GFP_KERNEL
)) == NULL
)
203 tp_ops
= tcf_proto_lookup_ops(tca
[TCA_KIND
-1]);
205 if (tp_ops
==NULL
&& tca
[TCA_KIND
-1] != NULL
) {
206 struct rtattr
*kind
= tca
[TCA_KIND
-1];
207 char module_name
[4 + IFNAMSIZ
+ 1];
209 if (RTA_PAYLOAD(kind
) <= IFNAMSIZ
) {
210 sprintf(module_name
, "cls_%s", (char*)RTA_DATA(kind
));
211 request_module (module_name
);
212 tp_ops
= tcf_proto_lookup_ops(kind
);
216 if (tp_ops
== NULL
) {
221 memset(tp
, 0, sizeof(*tp
));
223 tp
->protocol
= protocol
;
224 tp
->prio
= nprio
? : tcf_auto_prio(*back
);
226 tp
->classify
= tp_ops
->classify
;
227 tp
->classid
= parent
;
228 err
= tp_ops
->init(tp
);
233 write_lock(&qdisc_tree_lock
);
234 spin_lock_bh(&dev
->queue_lock
);
237 spin_unlock_bh(&dev
->queue_lock
);
238 write_unlock(&qdisc_tree_lock
);
239 } else if (tca
[TCA_KIND
-1] && rtattr_strcmp(tca
[TCA_KIND
-1], tp
->ops
->kind
))
242 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
245 if (n
->nlmsg_type
== RTM_DELTFILTER
&& t
->tcm_handle
== 0) {
246 write_lock(&qdisc_tree_lock
);
247 spin_lock_bh(&dev
->queue_lock
);
249 spin_unlock_bh(&dev
->queue_lock
);
250 write_unlock(&qdisc_tree_lock
);
252 tp
->ops
->destroy(tp
);
259 if (n
->nlmsg_type
!= RTM_NEWTFILTER
|| !(n
->nlmsg_flags
&NLM_F_CREATE
))
262 switch (n
->nlmsg_type
) {
265 if (n
->nlmsg_flags
&NLM_F_EXCL
)
269 err
= tp
->ops
->delete(tp
, fh
);
272 err
= tfilter_notify(skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
280 err
= tp
->ops
->change(tp
, cl
, t
->tcm_handle
, tca
, &fh
);
282 tfilter_notify(skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
291 tcf_fill_node(struct sk_buff
*skb
, struct tcf_proto
*tp
, unsigned long fh
,
292 u32 pid
, u32 seq
, unsigned flags
, int event
)
295 struct nlmsghdr
*nlh
;
296 unsigned char *b
= skb
->tail
;
298 nlh
= NLMSG_PUT(skb
, pid
, seq
, event
, sizeof(*tcm
));
299 nlh
->nlmsg_flags
= flags
;
300 tcm
= NLMSG_DATA(nlh
);
301 tcm
->tcm_family
= AF_UNSPEC
;
302 tcm
->tcm_ifindex
= tp
->q
->dev
->ifindex
;
303 tcm
->tcm_parent
= tp
->classid
;
305 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
306 RTA_PUT(skb
, TCA_KIND
, IFNAMSIZ
, tp
->ops
->kind
);
307 if (tp
->ops
->dump
&& tp
->ops
->dump(tp
, fh
, skb
, tcm
) < 0)
309 nlh
->nlmsg_len
= skb
->tail
- b
;
314 skb_trim(skb
, b
- skb
->data
);
318 static int tfilter_notify(struct sk_buff
*oskb
, struct nlmsghdr
*n
,
319 struct tcf_proto
*tp
, unsigned long fh
, int event
)
322 u32 pid
= oskb
? NETLINK_CB(oskb
).pid
: 0;
324 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
328 if (tcf_fill_node(skb
, tp
, fh
, pid
, n
->nlmsg_seq
, 0, event
) <= 0) {
333 return rtnetlink_send(skb
, pid
, RTMGRP_TC
, n
->nlmsg_flags
&NLM_F_ECHO
);
340 struct netlink_callback
*cb
;
343 static int tcf_node_dump(struct tcf_proto
*tp
, unsigned long n
, struct tcf_walker
*arg
)
345 struct tcf_dump_args
*a
= (void*)arg
;
347 return tcf_fill_node(a
->skb
, tp
, n
, NETLINK_CB(a
->cb
->skb
).pid
,
348 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTFILTER
);
351 static int tc_dump_tfilter(struct sk_buff
*skb
, struct netlink_callback
*cb
)
357 struct tcf_proto
*tp
, **chain
;
358 struct tcmsg
*tcm
= (struct tcmsg
*)NLMSG_DATA(cb
->nlh
);
359 unsigned long cl
= 0;
360 struct Qdisc_class_ops
*cops
;
361 struct tcf_dump_args arg
;
363 if (cb
->nlh
->nlmsg_len
< NLMSG_LENGTH(sizeof(*tcm
)))
365 if ((dev
= dev_get_by_index(tcm
->tcm_ifindex
)) == NULL
)
368 read_lock(&qdisc_tree_lock
);
369 if (!tcm
->tcm_parent
)
370 q
= dev
->qdisc_sleeping
;
372 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
374 read_unlock(&qdisc_tree_lock
);
377 if ((cops
= q
->ops
->cl_ops
) == NULL
)
379 if (TC_H_MIN(tcm
->tcm_parent
)) {
380 cl
= cops
->get(q
, tcm
->tcm_parent
);
384 chain
= cops
->tcf_chain(q
, cl
);
390 for (tp
=*chain
, t
=0; tp
; tp
= tp
->next
, t
++) {
391 if (t
< s_t
) continue;
392 if (TC_H_MAJ(tcm
->tcm_info
) &&
393 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
395 if (TC_H_MIN(tcm
->tcm_info
) &&
396 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
399 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
400 if (cb
->args
[1] == 0) {
401 if (tcf_fill_node(skb
, tp
, 0, NETLINK_CB(cb
->skb
).pid
,
402 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTFILTER
) <= 0) {
407 if (tp
->ops
->walk
== NULL
)
409 arg
.w
.fn
= tcf_node_dump
;
413 arg
.w
.skip
= cb
->args
[1]-1;
415 tp
->ops
->walk(tp
, &arg
.w
);
416 cb
->args
[1] = arg
.w
.count
+1;
427 read_unlock(&qdisc_tree_lock
);
434 __initfunc(int tc_filter_init(void))
436 #ifdef CONFIG_RTNETLINK
437 struct rtnetlink_link
*link_p
= rtnetlink_links
[PF_UNSPEC
];
439 /* Setup rtnetlink links. It is made here to avoid
440 exporting large number of public symbols.
444 link_p
[RTM_NEWTFILTER
-RTM_BASE
].doit
= tc_ctl_tfilter
;
445 link_p
[RTM_DELTFILTER
-RTM_BASE
].doit
= tc_ctl_tfilter
;
446 link_p
[RTM_GETTFILTER
-RTM_BASE
].doit
= tc_ctl_tfilter
;
447 link_p
[RTM_GETTFILTER
-RTM_BASE
].dumpit
= tc_dump_tfilter
;
450 #define INIT_TC_FILTER(name) { \
451 extern struct tcf_proto_ops cls_##name##_ops; \
452 register_tcf_proto_ops(&cls_##name##_ops); \
455 #ifdef CONFIG_NET_CLS_U32
458 #ifdef CONFIG_NET_CLS_ROUTE4
459 INIT_TC_FILTER(route4
);
461 #ifdef CONFIG_NET_CLS_FW
464 #ifdef CONFIG_NET_CLS_RSVP
465 INIT_TC_FILTER(rsvp
);
467 #ifdef CONFIG_NET_CLS_RSVP6
468 INIT_TC_FILTER(rsvp6
);