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 */
34 static LIST_HEAD(tcf_proto_base
);
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 const struct tcf_proto_ops
*tcf_proto_lookup_ops(struct nlattr
*kind
)
43 const struct tcf_proto_ops
*t
, *res
= NULL
;
46 read_lock(&cls_mod_lock
);
47 list_for_each_entry(t
, &tcf_proto_base
, head
) {
48 if (nla_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
;
66 write_lock(&cls_mod_lock
);
67 list_for_each_entry(t
, &tcf_proto_base
, head
)
68 if (!strcmp(ops
->kind
, t
->kind
))
71 list_add_tail(&ops
->head
, &tcf_proto_base
);
74 write_unlock(&cls_mod_lock
);
77 EXPORT_SYMBOL(register_tcf_proto_ops
);
79 int unregister_tcf_proto_ops(struct tcf_proto_ops
*ops
)
81 struct tcf_proto_ops
*t
;
84 /* Wait for outstanding call_rcu()s, if any, from a
85 * tcf_proto_ops's destroy() handler.
89 write_lock(&cls_mod_lock
);
90 list_for_each_entry(t
, &tcf_proto_base
, head
) {
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
)
123 struct net
*net
= sock_net(skb
->sk
);
124 struct nlattr
*tca
[TCA_MAX
+ 1];
130 struct net_device
*dev
;
132 struct tcf_proto __rcu
**back
;
133 struct tcf_proto __rcu
**chain
;
134 struct tcf_proto
*tp
;
135 const struct tcf_proto_ops
*tp_ops
;
136 const struct Qdisc_class_ops
*cops
;
142 if ((n
->nlmsg_type
!= RTM_GETTFILTER
) &&
143 !netlink_ns_capable(skb
, net
->user_ns
, 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 */
206 (tp
= rtnl_dereference(*back
)) != NULL
;
208 if (tp
->prio
>= prio
) {
209 if (tp
->prio
== prio
) {
211 (tp
->protocol
!= protocol
&& protocol
))
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
;
268 TC_H_MAJ(tcf_auto_prio(rtnl_dereference(*back
)));
270 tp
->classify
= tp_ops
->classify
;
271 tp
->classid
= parent
;
273 err
= tp_ops
->init(tp
);
275 module_put(tp_ops
->owner
);
282 } else if (tca
[TCA_KIND
] && nla_strcmp(tca
[TCA_KIND
], tp
->ops
->kind
))
285 fh
= tp
->ops
->get(tp
, t
->tcm_handle
);
288 if (n
->nlmsg_type
== RTM_DELTFILTER
&& t
->tcm_handle
== 0) {
289 struct tcf_proto
*next
= rtnl_dereference(tp
->next
);
291 RCU_INIT_POINTER(*back
, next
);
293 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_DELTFILTER
);
294 tcf_destroy(tp
, true);
300 if (n
->nlmsg_type
!= RTM_NEWTFILTER
||
301 !(n
->nlmsg_flags
& NLM_F_CREATE
))
304 switch (n
->nlmsg_type
) {
307 if (n
->nlmsg_flags
& NLM_F_EXCL
) {
309 tcf_destroy(tp
, true);
314 err
= tp
->ops
->delete(tp
, fh
);
316 struct tcf_proto
*next
= rtnl_dereference(tp
->next
);
318 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_DELTFILTER
);
319 if (tcf_destroy(tp
, false))
320 RCU_INIT_POINTER(*back
, next
);
324 err
= tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
332 err
= tp
->ops
->change(net
, skb
, tp
, cl
, t
->tcm_handle
, tca
, &fh
,
333 n
->nlmsg_flags
& NLM_F_CREATE
? TCA_ACT_NOREPLACE
: TCA_ACT_REPLACE
);
336 RCU_INIT_POINTER(tp
->next
, rtnl_dereference(*back
));
337 rcu_assign_pointer(*back
, tp
);
339 tfilter_notify(net
, skb
, n
, tp
, fh
, RTM_NEWTFILTER
);
342 tcf_destroy(tp
, true);
349 /* Replay the request. */
354 static int tcf_fill_node(struct net
*net
, struct sk_buff
*skb
, struct tcf_proto
*tp
,
355 unsigned long fh
, u32 portid
, u32 seq
, u16 flags
, int event
)
358 struct nlmsghdr
*nlh
;
359 unsigned char *b
= skb_tail_pointer(skb
);
361 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*tcm
), flags
);
364 tcm
= nlmsg_data(nlh
);
365 tcm
->tcm_family
= AF_UNSPEC
;
368 tcm
->tcm_ifindex
= qdisc_dev(tp
->q
)->ifindex
;
369 tcm
->tcm_parent
= tp
->classid
;
370 tcm
->tcm_info
= TC_H_MAKE(tp
->prio
, tp
->protocol
);
371 if (nla_put_string(skb
, TCA_KIND
, tp
->ops
->kind
))
372 goto nla_put_failure
;
373 tcm
->tcm_handle
= fh
;
374 if (RTM_DELTFILTER
!= event
) {
376 if (tp
->ops
->dump
&& tp
->ops
->dump(net
, tp
, fh
, skb
, tcm
) < 0)
377 goto nla_put_failure
;
379 nlh
->nlmsg_len
= skb_tail_pointer(skb
) - b
;
388 static int tfilter_notify(struct net
*net
, struct sk_buff
*oskb
,
389 struct nlmsghdr
*n
, struct tcf_proto
*tp
,
390 unsigned long fh
, int event
)
393 u32 portid
= oskb
? NETLINK_CB(oskb
).portid
: 0;
395 skb
= alloc_skb(NLMSG_GOODSIZE
, GFP_KERNEL
);
399 if (tcf_fill_node(net
, skb
, tp
, fh
, portid
, n
->nlmsg_seq
, 0, event
) <= 0) {
404 return rtnetlink_send(skb
, net
, portid
, RTNLGRP_TC
,
405 n
->nlmsg_flags
& NLM_F_ECHO
);
408 struct tcf_dump_args
{
411 struct netlink_callback
*cb
;
414 static int tcf_node_dump(struct tcf_proto
*tp
, unsigned long n
,
415 struct tcf_walker
*arg
)
417 struct tcf_dump_args
*a
= (void *)arg
;
418 struct net
*net
= sock_net(a
->skb
->sk
);
420 return tcf_fill_node(net
, a
->skb
, tp
, n
, NETLINK_CB(a
->cb
->skb
).portid
,
421 a
->cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
, RTM_NEWTFILTER
);
424 /* called with RTNL */
425 static int tc_dump_tfilter(struct sk_buff
*skb
, struct netlink_callback
*cb
)
427 struct net
*net
= sock_net(skb
->sk
);
430 struct net_device
*dev
;
432 struct tcf_proto
*tp
, __rcu
**chain
;
433 struct tcmsg
*tcm
= nlmsg_data(cb
->nlh
);
434 unsigned long cl
= 0;
435 const struct Qdisc_class_ops
*cops
;
436 struct tcf_dump_args arg
;
438 if (nlmsg_len(cb
->nlh
) < sizeof(*tcm
))
440 dev
= __dev_get_by_index(net
, tcm
->tcm_ifindex
);
444 if (!tcm
->tcm_parent
)
447 q
= qdisc_lookup(dev
, TC_H_MAJ(tcm
->tcm_parent
));
450 cops
= q
->ops
->cl_ops
;
453 if (cops
->tcf_chain
== NULL
)
455 if (TC_H_MIN(tcm
->tcm_parent
)) {
456 cl
= cops
->get(q
, tcm
->tcm_parent
);
460 chain
= cops
->tcf_chain(q
, cl
);
466 for (tp
= rtnl_dereference(*chain
), t
= 0;
467 tp
; tp
= rtnl_dereference(tp
->next
), t
++) {
470 if (TC_H_MAJ(tcm
->tcm_info
) &&
471 TC_H_MAJ(tcm
->tcm_info
) != tp
->prio
)
473 if (TC_H_MIN(tcm
->tcm_info
) &&
474 TC_H_MIN(tcm
->tcm_info
) != tp
->protocol
)
477 memset(&cb
->args
[1], 0, sizeof(cb
->args
)-sizeof(cb
->args
[0]));
478 if (cb
->args
[1] == 0) {
479 if (tcf_fill_node(net
, skb
, tp
, 0, NETLINK_CB(cb
->skb
).portid
,
480 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
481 RTM_NEWTFILTER
) <= 0)
486 if (tp
->ops
->walk
== NULL
)
488 arg
.w
.fn
= tcf_node_dump
;
492 arg
.w
.skip
= cb
->args
[1] - 1;
494 tp
->ops
->walk(tp
, &arg
.w
);
495 cb
->args
[1] = arg
.w
.count
+ 1;
509 void tcf_exts_destroy(struct tcf_exts
*exts
)
511 #ifdef CONFIG_NET_CLS_ACT
512 tcf_action_destroy(&exts
->actions
, TCA_ACT_UNBIND
);
513 INIT_LIST_HEAD(&exts
->actions
);
516 EXPORT_SYMBOL(tcf_exts_destroy
);
518 int tcf_exts_validate(struct net
*net
, struct tcf_proto
*tp
, struct nlattr
**tb
,
519 struct nlattr
*rate_tlv
, struct tcf_exts
*exts
, bool ovr
)
521 #ifdef CONFIG_NET_CLS_ACT
523 struct tc_action
*act
;
525 INIT_LIST_HEAD(&exts
->actions
);
526 if (exts
->police
&& tb
[exts
->police
]) {
527 act
= tcf_action_init_1(net
, tb
[exts
->police
], rate_tlv
,
533 act
->type
= exts
->type
= TCA_OLD_COMPAT
;
534 list_add(&act
->list
, &exts
->actions
);
535 } else if (exts
->action
&& tb
[exts
->action
]) {
537 err
= tcf_action_init(net
, tb
[exts
->action
], rate_tlv
,
539 TCA_ACT_BIND
, &exts
->actions
);
545 if ((exts
->action
&& tb
[exts
->action
]) ||
546 (exts
->police
&& tb
[exts
->police
]))
552 EXPORT_SYMBOL(tcf_exts_validate
);
554 void tcf_exts_change(struct tcf_proto
*tp
, struct tcf_exts
*dst
,
555 struct tcf_exts
*src
)
557 #ifdef CONFIG_NET_CLS_ACT
560 list_splice_init(&dst
->actions
, &tmp
);
561 list_splice(&src
->actions
, &dst
->actions
);
562 dst
->type
= src
->type
;
564 tcf_action_destroy(&tmp
, TCA_ACT_UNBIND
);
567 EXPORT_SYMBOL(tcf_exts_change
);
569 #define tcf_exts_first_act(ext) \
570 list_first_entry_or_null(&(exts)->actions, \
571 struct tc_action, list)
573 int tcf_exts_dump(struct sk_buff
*skb
, struct tcf_exts
*exts
)
575 #ifdef CONFIG_NET_CLS_ACT
578 if (exts
->action
&& !list_empty(&exts
->actions
)) {
580 * again for backward compatible mode - we want
581 * to work with both old and new modes of entering
582 * tc data even if iproute2 was newer - jhs
584 if (exts
->type
!= TCA_OLD_COMPAT
) {
585 nest
= nla_nest_start(skb
, exts
->action
);
587 goto nla_put_failure
;
588 if (tcf_action_dump(skb
, &exts
->actions
, 0, 0) < 0)
589 goto nla_put_failure
;
590 nla_nest_end(skb
, nest
);
591 } else if (exts
->police
) {
592 struct tc_action
*act
= tcf_exts_first_act(exts
);
593 nest
= nla_nest_start(skb
, exts
->police
);
594 if (nest
== NULL
|| !act
)
595 goto nla_put_failure
;
596 if (tcf_action_dump_old(skb
, act
, 0, 0) < 0)
597 goto nla_put_failure
;
598 nla_nest_end(skb
, nest
);
604 nla_nest_cancel(skb
, nest
);
610 EXPORT_SYMBOL(tcf_exts_dump
);
613 int tcf_exts_dump_stats(struct sk_buff
*skb
, struct tcf_exts
*exts
)
615 #ifdef CONFIG_NET_CLS_ACT
616 struct tc_action
*a
= tcf_exts_first_act(exts
);
617 if (a
!= NULL
&& tcf_action_copy_stats(skb
, a
, 1) < 0)
622 EXPORT_SYMBOL(tcf_exts_dump_stats
);
624 static int __init
tc_filter_init(void)
626 rtnl_register(PF_UNSPEC
, RTM_NEWTFILTER
, tc_ctl_tfilter
, NULL
, NULL
);
627 rtnl_register(PF_UNSPEC
, RTM_DELTFILTER
, tc_ctl_tfilter
, NULL
, NULL
);
628 rtnl_register(PF_UNSPEC
, RTM_GETTFILTER
, tc_ctl_tfilter
,
629 tc_dump_tfilter
, NULL
);
634 subsys_initcall(tc_filter_init
);