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/err.h>
26 #include <linux/slab.h>
27 #include <net/net_namespace.h>
29 #include <net/netlink.h>
30 #include <net/pkt_sched.h>
31 #include <net/pkt_cls.h>
33 /* The list of all installed classifier types */
35 static struct tcf_proto_ops
*tcf_proto_base __read_mostly
;
37 /* Protects list of registered TC modules. It is pure SMP lock. */
38 static DEFINE_RWLOCK(cls_mod_lock
);
40 /* Find classifier type by string name */
42 static const struct tcf_proto_ops
*tcf_proto_lookup_ops(struct nlattr
*kind
)
44 const struct tcf_proto_ops
*t
= NULL
;
47 read_lock(&cls_mod_lock
);
48 for (t
= tcf_proto_base
; t
; t
= t
->next
) {
49 if (nla_strcmp(kind
, t
->kind
) == 0) {
50 if (!try_module_get(t
->owner
))
55 read_unlock(&cls_mod_lock
);
60 /* Register(unregister) new classifier type */
62 int register_tcf_proto_ops(struct tcf_proto_ops
*ops
)
64 struct tcf_proto_ops
*t
, **tp
;
67 write_lock(&cls_mod_lock
);
68 for (tp
= &tcf_proto_base
; (t
= *tp
) != NULL
; tp
= &t
->next
)
69 if (!strcmp(ops
->kind
, t
->kind
))
76 write_unlock(&cls_mod_lock
);
79 EXPORT_SYMBOL(register_tcf_proto_ops
);
81 int unregister_tcf_proto_ops(struct tcf_proto_ops
*ops
)
83 struct tcf_proto_ops
*t
, **tp
;
86 write_lock(&cls_mod_lock
);
87 for (tp
= &tcf_proto_base
; (t
= *tp
) != NULL
; tp
= &t
->next
)
96 write_unlock(&cls_mod_lock
);
99 EXPORT_SYMBOL(unregister_tcf_proto_ops
);
101 static int tfilter_notify(struct net
*net
, struct sk_buff
*oskb
,
102 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
103 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);
113 first
= tp
->prio
- 1;
118 /* Add/change/delete/get a filter node */
120 static int tc_ctl_tfilter(struct sk_buff
*skb
, struct nlmsghdr
*n
)
122 struct net
*net
= sock_net(skb
->sk
);
123 struct nlattr
*tca
[TCA_MAX
+ 1];
124 spinlock_t
*root_lock
;
130 struct net_device
*dev
;
132 struct tcf_proto
**back
, **chain
;
133 struct tcf_proto
*tp
;
134 const struct tcf_proto_ops
*tp_ops
;
135 const struct Qdisc_class_ops
*cops
;
141 if ((n
->nlmsg_type
!= RTM_GETTFILTER
) && !netlink_capable(skb
, CAP_NET_ADMIN
))
147 err
= nlmsg_parse(n
, sizeof(*t
), tca
, TCA_MAX
, NULL
);
152 protocol
= TC_H_MIN(t
->tcm_info
);
153 prio
= TC_H_MAJ(t
->tcm_info
);
155 parent
= t
->tcm_parent
;
159 /* If no priority is given, user wants we allocated it. */
160 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
161 !(n
->nlmsg_flags
& NLM_F_CREATE
))
163 prio
= TC_H_MAKE(0x80000000U
, 0U);
166 /* Find head of filter chain. */
169 dev
= __dev_get_by_index(net
, t
->tcm_ifindex
);
178 q
= qdisc_lookup(dev
, TC_H_MAJ(t
->tcm_parent
));
183 /* Is it classful? */
184 cops
= q
->ops
->cl_ops
;
188 if (cops
->tcf_chain
== NULL
)
191 /* Do we search for filter, attached to class? */
192 if (TC_H_MIN(parent
)) {
193 cl
= cops
->get(q
, parent
);
198 /* And the last stroke */
199 chain
= cops
->tcf_chain(q
, cl
);
204 /* Check the chain for existence of proto-tcf with this priority */
205 for (back
= chain
; (tp
= *back
) != NULL
; back
= &tp
->next
) {
206 if (tp
->prio
>= prio
) {
207 if (tp
->prio
== prio
) {
209 (tp
->protocol
!= protocol
&& protocol
))
217 root_lock
= qdisc_root_sleeping_lock(q
);
220 /* Proto-tcf does not exist, create new one */
222 if (tca
[TCA_KIND
] == NULL
|| !protocol
)
226 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
227 !(n
->nlmsg_flags
& NLM_F_CREATE
))
231 /* Create new proto tcf */
234 tp
= kzalloc(sizeof(*tp
), GFP_KERNEL
);
238 tp_ops
= tcf_proto_lookup_ops(tca
[TCA_KIND
]);
239 if (tp_ops
== NULL
) {
240 #ifdef CONFIG_MODULES
241 struct nlattr
*kind
= tca
[TCA_KIND
];
245 nla_strlcpy(name
, kind
, IFNAMSIZ
) < IFNAMSIZ
) {
247 request_module("cls_%s", name
);
249 tp_ops
= tcf_proto_lookup_ops(kind
);
250 /* We dropped the RTNL semaphore in order to
251 * perform the module load. So, even if we
252 * succeeded in loading the module we have to
253 * replay the request. We indicate this using
256 if (tp_ops
!= NULL
) {
257 module_put(tp_ops
->owner
);
266 tp
->protocol
= protocol
;
267 tp
->prio
= nprio
? : TC_H_MAJ(tcf_auto_prio(*back
));
269 tp
->classify
= tp_ops
->classify
;
270 tp
->classid
= parent
;
272 err
= tp_ops
->init(tp
);
274 module_put(tp_ops
->owner
);
281 } else if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
))
284 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
287 if (n
->nlmsg_type
== RTM_DELTFILTER
&& t
->tcm_handle
== 0) {
288 spin_lock_bh(root_lock
);
290 spin_unlock_bh(root_lock
);
292 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_DELTFILTER
);
299 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
300 !(n
->nlmsg_flags
& NLM_F_CREATE
))
303 switch (n
->nlmsg_type
) {
306 if (n
->nlmsg_flags
& NLM_F_EXCL
) {
313 err
= tp
->ops
->delete(tp
, fh
);
315 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_DELTFILTER
);
318 err
= tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
326 err
= tp
->ops
->change(net
, skb
, tp
, cl
, t
->tcm_handle
, tca
, &fh
);
329 spin_lock_bh(root_lock
);
332 spin_unlock_bh(root_lock
);
334 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
344 /* Replay the request. */
349 static int tcf_fill_node(struct sk_buff
*skb
, struct tcf_proto
*tp
,
350 unsigned long fh
, u32 portid
, u32 seq
, u16 flags
, int event
)
353 struct nlmsghdr
*nlh
;
354 unsigned char *b
= skb_tail_pointer(skb
);
356 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
359 tcm
= nlmsg_data(nlh
);
360 tcm
->tcm_family
= AF_UNSPEC
;
363 tcm
->tcm_ifindex
= qdisc_dev(tp
->q
)->ifindex
;
364 tcm
->tcm_parent
= tp
->classid
;
365 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
366 if (nla_put_string(skb
, TCA_KIND
, tp
->ops
->kind
))
367 goto nla_put_failure
;
368 tcm
->tcm_handle
= fh
;
369 if (RTM_DELTFILTER
!= event
) {
371 if (tp
->ops
->dump
&& tp
->ops
->dump(tp
, fh
, skb
, tcm
) < 0)
372 goto nla_put_failure
;
374 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
383 static int tfilter_notify(struct net
*net
, struct sk_buff
*oskb
,
384 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
385 unsigned long fh
, int event
)
388 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
390 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
394 if (tcf_fill_node(skb
, tp
, fh
, portid
, n
->nlmsg_seq
, 0, event
) <= 0) {
399 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
400 n
->nlmsg_flags
& NLM_F_ECHO
);
403 struct tcf_dump_args
{
406 struct netlink_callback
*cb
;
409 static int tcf_node_dump(struct tcf_proto
*tp
, unsigned long n
,
410 struct tcf_walker
*arg
)
412 struct tcf_dump_args
*a
= (void *)arg
;
414 return tcf_fill_node(a
->skb
, tp
, n
, NETLINK_CB(a
->cb
->skb
).portid
,
415 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTFILTER
);
418 /* called with RTNL */
419 static int tc_dump_tfilter(struct sk_buff
*skb
, struct netlink_callback
*cb
)
421 struct net
*net
= sock_net(skb
->sk
);
424 struct net_device
*dev
;
426 struct tcf_proto
*tp
, **chain
;
427 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
428 unsigned long cl
= 0;
429 const struct Qdisc_class_ops
*cops
;
430 struct tcf_dump_args arg
;
432 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
434 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
438 if (!tcm
->tcm_parent
)
441 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
444 cops
= q
->ops
->cl_ops
;
447 if (cops
->tcf_chain
== NULL
)
449 if (TC_H_MIN(tcm
->tcm_parent
)) {
450 cl
= cops
->get(q
, tcm
->tcm_parent
);
454 chain
= cops
->tcf_chain(q
, cl
);
460 for (tp
= *chain
, t
= 0; tp
; tp
= tp
->next
, t
++) {
463 if (TC_H_MAJ(tcm
->tcm_info
) &&
464 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
466 if (TC_H_MIN(tcm
->tcm_info
) &&
467 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
470 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
471 if (cb
->args
[1] == 0) {
472 if (tcf_fill_node(skb
, tp
, 0, NETLINK_CB(cb
->skb
).portid
,
473 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
474 RTM_NEWTFILTER
) <= 0)
479 if (tp
->ops
->walk
== NULL
)
481 arg
.w
.fn
= tcf_node_dump
;
485 arg
.w
.skip
= cb
->args
[1] - 1;
487 tp
->ops
->walk(tp
, &arg
.w
);
488 cb
->args
[1] = arg
.w
.count
+ 1;
502 void tcf_exts_destroy(struct tcf_proto
*tp
, struct tcf_exts
*exts
)
504 #ifdef CONFIG_NET_CLS_ACT
506 tcf_action_destroy(exts
->action
, TCA_ACT_UNBIND
);
511 EXPORT_SYMBOL(tcf_exts_destroy
);
513 int tcf_exts_validate(struct net
*net
, struct tcf_proto
*tp
, struct nlattr
**tb
,
514 struct nlattr
*rate_tlv
, struct tcf_exts
*exts
,
515 const struct tcf_ext_map
*map
)
517 memset(exts
, 0, sizeof(*exts
));
519 #ifdef CONFIG_NET_CLS_ACT
521 struct tc_action
*act
;
523 if (map
->police
&& tb
[map
->police
]) {
524 act
= tcf_action_init_1(net
, tb
[map
->police
], rate_tlv
,
525 "police", TCA_ACT_NOREPLACE
,
530 act
->type
= TCA_OLD_COMPAT
;
532 } else if (map
->action
&& tb
[map
->action
]) {
533 act
= tcf_action_init(net
, tb
[map
->action
], rate_tlv
,
534 NULL
, TCA_ACT_NOREPLACE
,
543 if ((map
->action
&& tb
[map
->action
]) ||
544 (map
->police
&& tb
[map
->police
]))
550 EXPORT_SYMBOL(tcf_exts_validate
);
552 void tcf_exts_change(struct tcf_proto
*tp
, struct tcf_exts
*dst
,
553 struct tcf_exts
*src
)
555 #ifdef CONFIG_NET_CLS_ACT
557 struct tc_action
*act
;
560 dst
->action
= src
->action
;
563 tcf_action_destroy(act
, TCA_ACT_UNBIND
);
567 EXPORT_SYMBOL(tcf_exts_change
);
569 int tcf_exts_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
,
570 const struct tcf_ext_map
*map
)
572 #ifdef CONFIG_NET_CLS_ACT
573 if (map
->action
&& exts
->action
) {
575 * again for backward compatible mode - we want
576 * to work with both old and new modes of entering
577 * tc data even if iproute2 was newer - jhs
581 if (exts
->action
->type
!= TCA_OLD_COMPAT
) {
582 nest
= nla_nest_start(skb
, map
->action
);
584 goto nla_put_failure
;
585 if (tcf_action_dump(skb
, exts
->action
, 0, 0) < 0)
586 goto nla_put_failure
;
587 nla_nest_end(skb
, nest
);
588 } else if (map
->police
) {
589 nest
= nla_nest_start(skb
, map
->police
);
591 goto nla_put_failure
;
592 if (tcf_action_dump_old(skb
, exts
->action
, 0, 0) < 0)
593 goto nla_put_failure
;
594 nla_nest_end(skb
, nest
);
599 nla_put_failure
: __attribute__ ((unused
))
602 EXPORT_SYMBOL(tcf_exts_dump
);
605 int tcf_exts_dump_stats(struct sk_buff
*skb
, struct tcf_exts
*exts
,
606 const struct tcf_ext_map
*map
)
608 #ifdef CONFIG_NET_CLS_ACT
610 if (tcf_action_copy_stats(skb
, exts
->action
, 1) < 0)
611 goto nla_put_failure
;
614 nla_put_failure
: __attribute__ ((unused
))
617 EXPORT_SYMBOL(tcf_exts_dump_stats
);
619 static int __init
tc_filter_init(void)
621 rtnl_register(PF_UNSPEC
, RTM_NEWTFILTER
, tc_ctl_tfilter
, NULL
, NULL
);
622 rtnl_register(PF_UNSPEC
, RTM_DELTFILTER
, tc_ctl_tfilter
, NULL
, NULL
);
623 rtnl_register(PF_UNSPEC
, RTM_GETTFILTER
, tc_ctl_tfilter
,
624 tc_dump_tfilter
, NULL
);
629 subsys_initcall(tc_filter_init
);