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 <linux/err.h>
27 #include <linux/slab.h>
28 #include <net/net_namespace.h>
30 #include <net/netlink.h>
31 #include <net/pkt_sched.h>
32 #include <net/pkt_cls.h>
34 /* The list of all installed classifier types */
36 static struct tcf_proto_ops
*tcf_proto_base __read_mostly
;
38 /* Protects list of registered TC modules. It is pure SMP lock. */
39 static DEFINE_RWLOCK(cls_mod_lock
);
41 /* Find classifier type by string name */
43 static struct tcf_proto_ops
*tcf_proto_lookup_ops(struct nlattr
*kind
)
45 struct tcf_proto_ops
*t
= NULL
;
48 read_lock(&cls_mod_lock
);
49 for (t
= tcf_proto_base
; t
; t
= t
->next
) {
50 if (nla_strcmp(kind
, t
->kind
) == 0) {
51 if (!try_module_get(t
->owner
))
56 read_unlock(&cls_mod_lock
);
61 /* Register(unregister) new classifier type */
63 int register_tcf_proto_ops(struct tcf_proto_ops
*ops
)
65 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
))
77 write_unlock(&cls_mod_lock
);
80 EXPORT_SYMBOL(register_tcf_proto_ops
);
82 int unregister_tcf_proto_ops(struct tcf_proto_ops
*ops
)
84 struct tcf_proto_ops
*t
, **tp
;
87 write_lock(&cls_mod_lock
);
88 for (tp
= &tcf_proto_base
; (t
= *tp
) != NULL
; tp
= &t
->next
)
97 write_unlock(&cls_mod_lock
);
100 EXPORT_SYMBOL(unregister_tcf_proto_ops
);
102 static int tfilter_notify(struct net
*net
, struct sk_buff
*oskb
,
103 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
104 unsigned long fh
, int event
);
107 /* Select new prio value from the range, managed by kernel. */
109 static inline u32
tcf_auto_prio(struct tcf_proto
*tp
)
111 u32 first
= TC_H_MAKE(0xC0000000U
, 0U);
114 first
= tp
->prio
- 1;
119 /* Add/change/delete/get a filter node */
121 static int tc_ctl_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
, void *arg
)
123 struct net
*net
= sock_net(skb
->sk
);
124 struct nlattr
*tca
[TCA_MAX
+ 1];
125 spinlock_t
*root_lock
;
131 struct net_device
*dev
;
133 struct tcf_proto
**back
, **chain
;
134 struct tcf_proto
*tp
;
135 struct tcf_proto_ops
*tp_ops
;
136 const struct Qdisc_class_ops
*cops
;
144 protocol
= TC_H_MIN(t
->tcm_info
);
145 prio
= TC_H_MAJ(t
->tcm_info
);
147 parent
= t
->tcm_parent
;
151 /* If no priority is given, user wants we allocated it. */
152 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
153 !(n
->nlmsg_flags
& NLM_F_CREATE
))
155 prio
= TC_H_MAKE(0x80000000U
, 0U);
158 /* Find head of filter chain. */
161 dev
= __dev_get_by_index(net
, t
->tcm_ifindex
);
165 err
= nlmsg_parse(n
, sizeof(*t
), tca
, TCA_MAX
, NULL
);
174 q
= qdisc_lookup(dev
, TC_H_MAJ(t
->tcm_parent
));
179 /* Is it classful? */
180 cops
= q
->ops
->cl_ops
;
184 if (cops
->tcf_chain
== NULL
)
187 /* Do we search for filter, attached to class? */
188 if (TC_H_MIN(parent
)) {
189 cl
= cops
->get(q
, parent
);
194 /* And the last stroke */
195 chain
= cops
->tcf_chain(q
, cl
);
200 /* Check the chain for existence of proto-tcf with this priority */
201 for (back
= chain
; (tp
= *back
) != NULL
; back
= &tp
->next
) {
202 if (tp
->prio
>= prio
) {
203 if (tp
->prio
== prio
) {
205 (tp
->protocol
!= protocol
&& protocol
))
213 root_lock
= qdisc_root_sleeping_lock(q
);
216 /* Proto-tcf does not exist, create new one */
218 if (tca
[TCA_KIND
] == NULL
|| !protocol
)
222 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
223 !(n
->nlmsg_flags
& NLM_F_CREATE
))
227 /* Create new proto tcf */
230 tp
= kzalloc(sizeof(*tp
), GFP_KERNEL
);
234 tp_ops
= tcf_proto_lookup_ops(tca
[TCA_KIND
]);
235 if (tp_ops
== NULL
) {
236 #ifdef CONFIG_MODULES
237 struct nlattr
*kind
= tca
[TCA_KIND
];
241 nla_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
243 request_module("cls_%s", name
);
245 tp_ops
= tcf_proto_lookup_ops(kind
);
246 /* We dropped the RTNL semaphore in order to
247 * perform the module load. So, even if we
248 * succeeded in loading the module we have to
249 * replay the request. We indicate this using
252 if (tp_ops
!= NULL
) {
253 module_put(tp_ops
->owner
);
262 tp
->protocol
= protocol
;
263 tp
->prio
= nprio
? : TC_H_MAJ(tcf_auto_prio(*back
));
265 tp
->classify
= tp_ops
->classify
;
266 tp
->classid
= parent
;
268 err
= tp_ops
->init(tp
);
270 module_put(tp_ops
->owner
);
277 } else if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
))
280 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
283 if (n
->nlmsg_type
== RTM_DELTFILTER
&& t
->tcm_handle
== 0) {
284 spin_lock_bh(root_lock
);
286 spin_unlock_bh(root_lock
);
288 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_DELTFILTER
);
295 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
296 !(n
->nlmsg_flags
& NLM_F_CREATE
))
299 switch (n
->nlmsg_type
) {
302 if (n
->nlmsg_flags
& NLM_F_EXCL
) {
309 err
= tp
->ops
->delete(tp
, fh
);
311 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_DELTFILTER
);
314 err
= tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
322 err
= tp
->ops
->change(tp
, cl
, t
->tcm_handle
, tca
, &fh
);
325 spin_lock_bh(root_lock
);
328 spin_unlock_bh(root_lock
);
330 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
340 /* Replay the request. */
345 static int tcf_fill_node(struct sk_buff
*skb
, struct tcf_proto
*tp
,
346 unsigned long fh
, u32 pid
, u32 seq
, u16 flags
, int event
)
349 struct nlmsghdr
*nlh
;
350 unsigned char *b
= skb_tail_pointer(skb
);
352 nlh
= NLMSG_NEW(skb
, pid
, seq
, event
, sizeof(*tcm
), flags
);
353 tcm
= NLMSG_DATA(nlh
);
354 tcm
->tcm_family
= AF_UNSPEC
;
357 tcm
->tcm_ifindex
= qdisc_dev(tp
->q
)->ifindex
;
358 tcm
->tcm_parent
= tp
->classid
;
359 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
360 NLA_PUT_STRING(skb
, TCA_KIND
, tp
->ops
->kind
);
361 tcm
->tcm_handle
= fh
;
362 if (RTM_DELTFILTER
!= event
) {
364 if (tp
->ops
->dump
&& tp
->ops
->dump(tp
, fh
, skb
, tcm
) < 0)
365 goto nla_put_failure
;
367 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
376 static int tfilter_notify(struct net
*net
, struct sk_buff
*oskb
,
377 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
378 unsigned long fh
, int event
)
381 u32 pid
= oskb
? NETLINK_CB(oskb
).pid
: 0;
383 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
387 if (tcf_fill_node(skb
, tp
, fh
, pid
, n
->nlmsg_seq
, 0, event
) <= 0) {
392 return rtnetlink_send(skb
, net
, pid
, RTNLGRP_TC
,
393 n
->nlmsg_flags
& NLM_F_ECHO
);
396 struct tcf_dump_args
{
399 struct netlink_callback
*cb
;
402 static int tcf_node_dump(struct tcf_proto
*tp
, unsigned long n
,
403 struct tcf_walker
*arg
)
405 struct tcf_dump_args
*a
= (void *)arg
;
407 return tcf_fill_node(a
->skb
, tp
, n
, NETLINK_CB(a
->cb
->skb
).pid
,
408 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTFILTER
);
411 /* called with RTNL */
412 static int tc_dump_tfilter(struct sk_buff
*skb
, struct netlink_callback
*cb
)
414 struct net
*net
= sock_net(skb
->sk
);
417 struct net_device
*dev
;
419 struct tcf_proto
*tp
, **chain
;
420 struct tcmsg
*tcm
= (struct tcmsg
*)NLMSG_DATA(cb
->nlh
);
421 unsigned long cl
= 0;
422 const struct Qdisc_class_ops
*cops
;
423 struct tcf_dump_args arg
;
425 if (cb
->nlh
->nlmsg_len
< NLMSG_LENGTH(sizeof(*tcm
)))
427 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
431 if (!tcm
->tcm_parent
)
434 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
437 cops
= q
->ops
->cl_ops
;
440 if (cops
->tcf_chain
== NULL
)
442 if (TC_H_MIN(tcm
->tcm_parent
)) {
443 cl
= cops
->get(q
, tcm
->tcm_parent
);
447 chain
= cops
->tcf_chain(q
, cl
);
453 for (tp
= *chain
, t
= 0; tp
; tp
= tp
->next
, t
++) {
456 if (TC_H_MAJ(tcm
->tcm_info
) &&
457 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
459 if (TC_H_MIN(tcm
->tcm_info
) &&
460 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
463 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
464 if (cb
->args
[1] == 0) {
465 if (tcf_fill_node(skb
, tp
, 0, NETLINK_CB(cb
->skb
).pid
,
466 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
467 RTM_NEWTFILTER
) <= 0)
472 if (tp
->ops
->walk
== NULL
)
474 arg
.w
.fn
= tcf_node_dump
;
478 arg
.w
.skip
= cb
->args
[1] - 1;
480 tp
->ops
->walk(tp
, &arg
.w
);
481 cb
->args
[1] = arg
.w
.count
+ 1;
495 void tcf_exts_destroy(struct tcf_proto
*tp
, struct tcf_exts
*exts
)
497 #ifdef CONFIG_NET_CLS_ACT
499 tcf_action_destroy(exts
->action
, TCA_ACT_UNBIND
);
504 EXPORT_SYMBOL(tcf_exts_destroy
);
506 int tcf_exts_validate(struct tcf_proto
*tp
, struct nlattr
**tb
,
507 struct nlattr
*rate_tlv
, struct tcf_exts
*exts
,
508 const struct tcf_ext_map
*map
)
510 memset(exts
, 0, sizeof(*exts
));
512 #ifdef CONFIG_NET_CLS_ACT
514 struct tc_action
*act
;
516 if (map
->police
&& tb
[map
->police
]) {
517 act
= tcf_action_init_1(tb
[map
->police
], rate_tlv
,
518 "police", TCA_ACT_NOREPLACE
,
523 act
->type
= TCA_OLD_COMPAT
;
525 } else if (map
->action
&& tb
[map
->action
]) {
526 act
= tcf_action_init(tb
[map
->action
], rate_tlv
, NULL
,
527 TCA_ACT_NOREPLACE
, TCA_ACT_BIND
);
535 if ((map
->action
&& tb
[map
->action
]) ||
536 (map
->police
&& tb
[map
->police
]))
542 EXPORT_SYMBOL(tcf_exts_validate
);
544 void tcf_exts_change(struct tcf_proto
*tp
, struct tcf_exts
*dst
,
545 struct tcf_exts
*src
)
547 #ifdef CONFIG_NET_CLS_ACT
549 struct tc_action
*act
;
552 dst
->action
= src
->action
;
555 tcf_action_destroy(act
, TCA_ACT_UNBIND
);
559 EXPORT_SYMBOL(tcf_exts_change
);
561 int tcf_exts_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
,
562 const struct tcf_ext_map
*map
)
564 #ifdef CONFIG_NET_CLS_ACT
565 if (map
->action
&& exts
->action
) {
567 * again for backward compatible mode - we want
568 * to work with both old and new modes of entering
569 * tc data even if iproute2 was newer - jhs
573 if (exts
->action
->type
!= TCA_OLD_COMPAT
) {
574 nest
= nla_nest_start(skb
, map
->action
);
576 goto nla_put_failure
;
577 if (tcf_action_dump(skb
, exts
->action
, 0, 0) < 0)
578 goto nla_put_failure
;
579 nla_nest_end(skb
, nest
);
580 } else if (map
->police
) {
581 nest
= nla_nest_start(skb
, map
->police
);
583 goto nla_put_failure
;
584 if (tcf_action_dump_old(skb
, exts
->action
, 0, 0) < 0)
585 goto nla_put_failure
;
586 nla_nest_end(skb
, nest
);
591 nla_put_failure
: __attribute__ ((unused
))
594 EXPORT_SYMBOL(tcf_exts_dump
);
597 int tcf_exts_dump_stats(struct sk_buff
*skb
, struct tcf_exts
*exts
,
598 const struct tcf_ext_map
*map
)
600 #ifdef CONFIG_NET_CLS_ACT
602 if (tcf_action_copy_stats(skb
, exts
->action
, 1) < 0)
603 goto nla_put_failure
;
606 nla_put_failure
: __attribute__ ((unused
))
609 EXPORT_SYMBOL(tcf_exts_dump_stats
);
611 static int __init
tc_filter_init(void)
613 rtnl_register(PF_UNSPEC
, RTM_NEWTFILTER
, tc_ctl_tfilter
, NULL
, NULL
);
614 rtnl_register(PF_UNSPEC
, RTM_DELTFILTER
, tc_ctl_tfilter
, NULL
, NULL
);
615 rtnl_register(PF_UNSPEC
, RTM_GETTFILTER
, tc_ctl_tfilter
,
616 tc_dump_tfilter
, NULL
);
621 subsys_initcall(tc_filter_init
);