2 * net/core/fib_rules.c Generic Routing Rules
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License as
6 * published by the Free Software Foundation, version 2.
8 * Authors: Thomas Graf <tgraf@suug.ch>
11 #include <linux/types.h>
12 #include <linux/kernel.h>
13 #include <linux/slab.h>
14 #include <linux/list.h>
15 #include <linux/module.h>
16 #include <net/net_namespace.h>
18 #include <net/fib_rules.h>
20 int fib_default_rule_add(struct fib_rules_ops
*ops
,
21 u32 pref
, u32 table
, u32 flags
)
25 r
= kzalloc(ops
->rule_size
, GFP_KERNEL
);
29 atomic_set(&r
->refcnt
, 1);
30 r
->action
= FR_ACT_TO_TBL
;
34 r
->fr_net
= ops
->fro_net
;
36 r
->suppress_prefixlen
= -1;
37 r
->suppress_ifgroup
= -1;
39 /* The lock is not required here, the list in unreacheable
40 * at the moment this function is called */
41 list_add_tail(&r
->list
, &ops
->rules_list
);
44 EXPORT_SYMBOL(fib_default_rule_add
);
46 u32
fib_default_rule_pref(struct fib_rules_ops
*ops
)
48 struct list_head
*pos
;
49 struct fib_rule
*rule
;
51 if (!list_empty(&ops
->rules_list
)) {
52 pos
= ops
->rules_list
.next
;
53 if (pos
->next
!= &ops
->rules_list
) {
54 rule
= list_entry(pos
->next
, struct fib_rule
, list
);
56 return rule
->pref
- 1;
62 EXPORT_SYMBOL(fib_default_rule_pref
);
64 static void notify_rule_change(int event
, struct fib_rule
*rule
,
65 struct fib_rules_ops
*ops
, struct nlmsghdr
*nlh
,
68 static struct fib_rules_ops
*lookup_rules_ops(struct net
*net
, int family
)
70 struct fib_rules_ops
*ops
;
73 list_for_each_entry_rcu(ops
, &net
->rules_ops
, list
) {
74 if (ops
->family
== family
) {
75 if (!try_module_get(ops
->owner
))
86 static void rules_ops_put(struct fib_rules_ops
*ops
)
89 module_put(ops
->owner
);
92 static void flush_route_cache(struct fib_rules_ops
*ops
)
95 ops
->flush_cache(ops
);
98 static int __fib_rules_register(struct fib_rules_ops
*ops
)
101 struct fib_rules_ops
*o
;
106 if (ops
->rule_size
< sizeof(struct fib_rule
))
109 if (ops
->match
== NULL
|| ops
->configure
== NULL
||
110 ops
->compare
== NULL
|| ops
->fill
== NULL
||
114 spin_lock(&net
->rules_mod_lock
);
115 list_for_each_entry(o
, &net
->rules_ops
, list
)
116 if (ops
->family
== o
->family
)
119 list_add_tail_rcu(&ops
->list
, &net
->rules_ops
);
122 spin_unlock(&net
->rules_mod_lock
);
127 struct fib_rules_ops
*
128 fib_rules_register(const struct fib_rules_ops
*tmpl
, struct net
*net
)
130 struct fib_rules_ops
*ops
;
133 ops
= kmemdup(tmpl
, sizeof(*ops
), GFP_KERNEL
);
135 return ERR_PTR(-ENOMEM
);
137 INIT_LIST_HEAD(&ops
->rules_list
);
140 err
= __fib_rules_register(ops
);
148 EXPORT_SYMBOL_GPL(fib_rules_register
);
150 static void fib_rules_cleanup_ops(struct fib_rules_ops
*ops
)
152 struct fib_rule
*rule
, *tmp
;
154 list_for_each_entry_safe(rule
, tmp
, &ops
->rules_list
, list
) {
155 list_del_rcu(&rule
->list
);
162 void fib_rules_unregister(struct fib_rules_ops
*ops
)
164 struct net
*net
= ops
->fro_net
;
166 spin_lock(&net
->rules_mod_lock
);
167 list_del_rcu(&ops
->list
);
168 spin_unlock(&net
->rules_mod_lock
);
170 fib_rules_cleanup_ops(ops
);
173 EXPORT_SYMBOL_GPL(fib_rules_unregister
);
175 static int fib_rule_match(struct fib_rule
*rule
, struct fib_rules_ops
*ops
,
176 struct flowi
*fl
, int flags
)
180 if (rule
->iifindex
&& (rule
->iifindex
!= fl
->flowi_iif
))
183 if (rule
->oifindex
&& (rule
->oifindex
!= fl
->flowi_oif
))
186 if ((rule
->mark
^ fl
->flowi_mark
) & rule
->mark_mask
)
189 ret
= ops
->match(rule
, fl
, flags
);
191 return (rule
->flags
& FIB_RULE_INVERT
) ? !ret
: ret
;
194 int fib_rules_lookup(struct fib_rules_ops
*ops
, struct flowi
*fl
,
195 int flags
, struct fib_lookup_arg
*arg
)
197 struct fib_rule
*rule
;
202 list_for_each_entry_rcu(rule
, &ops
->rules_list
, list
) {
204 if (!fib_rule_match(rule
, ops
, fl
, flags
))
207 if (rule
->action
== FR_ACT_GOTO
) {
208 struct fib_rule
*target
;
210 target
= rcu_dereference(rule
->ctarget
);
211 if (target
== NULL
) {
217 } else if (rule
->action
== FR_ACT_NOP
)
220 err
= ops
->action(rule
, fl
, flags
, arg
);
222 if (!err
&& ops
->suppress
&& ops
->suppress(rule
, arg
))
225 if (err
!= -EAGAIN
) {
226 if ((arg
->flags
& FIB_LOOKUP_NOREF
) ||
227 likely(atomic_inc_not_zero(&rule
->refcnt
))) {
241 EXPORT_SYMBOL_GPL(fib_rules_lookup
);
243 static int validate_rulemsg(struct fib_rule_hdr
*frh
, struct nlattr
**tb
,
244 struct fib_rules_ops
*ops
)
249 if (tb
[FRA_SRC
] == NULL
||
250 frh
->src_len
> (ops
->addr_size
* 8) ||
251 nla_len(tb
[FRA_SRC
]) != ops
->addr_size
)
255 if (tb
[FRA_DST
] == NULL
||
256 frh
->dst_len
> (ops
->addr_size
* 8) ||
257 nla_len(tb
[FRA_DST
]) != ops
->addr_size
)
265 static int fib_nl_newrule(struct sk_buff
*skb
, struct nlmsghdr
* nlh
)
267 struct net
*net
= sock_net(skb
->sk
);
268 struct fib_rule_hdr
*frh
= nlmsg_data(nlh
);
269 struct fib_rules_ops
*ops
= NULL
;
270 struct fib_rule
*rule
, *r
, *last
= NULL
;
271 struct nlattr
*tb
[FRA_MAX
+1];
272 int err
= -EINVAL
, unresolved
= 0;
274 if (nlh
->nlmsg_len
< nlmsg_msg_size(sizeof(*frh
)))
277 ops
= lookup_rules_ops(net
, frh
->family
);
283 err
= nlmsg_parse(nlh
, sizeof(*frh
), tb
, FRA_MAX
, ops
->policy
);
287 err
= validate_rulemsg(frh
, tb
, ops
);
291 rule
= kzalloc(ops
->rule_size
, GFP_KERNEL
);
298 if (tb
[FRA_PRIORITY
])
299 rule
->pref
= nla_get_u32(tb
[FRA_PRIORITY
]);
301 if (tb
[FRA_IIFNAME
]) {
302 struct net_device
*dev
;
305 nla_strlcpy(rule
->iifname
, tb
[FRA_IIFNAME
], IFNAMSIZ
);
306 dev
= __dev_get_by_name(net
, rule
->iifname
);
308 rule
->iifindex
= dev
->ifindex
;
311 if (tb
[FRA_OIFNAME
]) {
312 struct net_device
*dev
;
315 nla_strlcpy(rule
->oifname
, tb
[FRA_OIFNAME
], IFNAMSIZ
);
316 dev
= __dev_get_by_name(net
, rule
->oifname
);
318 rule
->oifindex
= dev
->ifindex
;
321 if (tb
[FRA_FWMARK
]) {
322 rule
->mark
= nla_get_u32(tb
[FRA_FWMARK
]);
324 /* compatibility: if the mark value is non-zero all bits
325 * are compared unless a mask is explicitly specified.
327 rule
->mark_mask
= 0xFFFFFFFF;
331 rule
->mark_mask
= nla_get_u32(tb
[FRA_FWMASK
]);
333 rule
->action
= frh
->action
;
334 rule
->flags
= frh
->flags
;
335 rule
->table
= frh_get_table(frh
, tb
);
336 if (tb
[FRA_SUPPRESS_PREFIXLEN
])
337 rule
->suppress_prefixlen
= nla_get_u32(tb
[FRA_SUPPRESS_PREFIXLEN
]);
339 rule
->suppress_prefixlen
= -1;
341 if (tb
[FRA_SUPPRESS_IFGROUP
])
342 rule
->suppress_ifgroup
= nla_get_u32(tb
[FRA_SUPPRESS_IFGROUP
]);
344 rule
->suppress_ifgroup
= -1;
346 if (!tb
[FRA_PRIORITY
] && ops
->default_pref
)
347 rule
->pref
= ops
->default_pref(ops
);
351 if (rule
->action
!= FR_ACT_GOTO
)
354 rule
->target
= nla_get_u32(tb
[FRA_GOTO
]);
355 /* Backward jumps are prohibited to avoid endless loops */
356 if (rule
->target
<= rule
->pref
)
359 list_for_each_entry(r
, &ops
->rules_list
, list
) {
360 if (r
->pref
== rule
->target
) {
361 RCU_INIT_POINTER(rule
->ctarget
, r
);
366 if (rcu_dereference_protected(rule
->ctarget
, 1) == NULL
)
368 } else if (rule
->action
== FR_ACT_GOTO
)
371 err
= ops
->configure(rule
, skb
, frh
, tb
);
375 list_for_each_entry(r
, &ops
->rules_list
, list
) {
376 if (r
->pref
> rule
->pref
)
384 list_add_rcu(&rule
->list
, &last
->list
);
386 list_add_rcu(&rule
->list
, &ops
->rules_list
);
388 if (ops
->unresolved_rules
) {
390 * There are unresolved goto rules in the list, check if
391 * any of them are pointing to this new rule.
393 list_for_each_entry(r
, &ops
->rules_list
, list
) {
394 if (r
->action
== FR_ACT_GOTO
&&
395 r
->target
== rule
->pref
&&
396 rtnl_dereference(r
->ctarget
) == NULL
) {
397 rcu_assign_pointer(r
->ctarget
, rule
);
398 if (--ops
->unresolved_rules
== 0)
404 if (rule
->action
== FR_ACT_GOTO
)
405 ops
->nr_goto_rules
++;
408 ops
->unresolved_rules
++;
410 notify_rule_change(RTM_NEWRULE
, rule
, ops
, nlh
, NETLINK_CB(skb
).portid
);
411 flush_route_cache(ops
);
422 static int fib_nl_delrule(struct sk_buff
*skb
, struct nlmsghdr
* nlh
)
424 struct net
*net
= sock_net(skb
->sk
);
425 struct fib_rule_hdr
*frh
= nlmsg_data(nlh
);
426 struct fib_rules_ops
*ops
= NULL
;
427 struct fib_rule
*rule
, *tmp
;
428 struct nlattr
*tb
[FRA_MAX
+1];
431 if (nlh
->nlmsg_len
< nlmsg_msg_size(sizeof(*frh
)))
434 ops
= lookup_rules_ops(net
, frh
->family
);
440 err
= nlmsg_parse(nlh
, sizeof(*frh
), tb
, FRA_MAX
, ops
->policy
);
444 err
= validate_rulemsg(frh
, tb
, ops
);
448 list_for_each_entry(rule
, &ops
->rules_list
, list
) {
449 if (frh
->action
&& (frh
->action
!= rule
->action
))
452 if (frh_get_table(frh
, tb
) &&
453 (frh_get_table(frh
, tb
) != rule
->table
))
456 if (tb
[FRA_PRIORITY
] &&
457 (rule
->pref
!= nla_get_u32(tb
[FRA_PRIORITY
])))
460 if (tb
[FRA_IIFNAME
] &&
461 nla_strcmp(tb
[FRA_IIFNAME
], rule
->iifname
))
464 if (tb
[FRA_OIFNAME
] &&
465 nla_strcmp(tb
[FRA_OIFNAME
], rule
->oifname
))
468 if (tb
[FRA_FWMARK
] &&
469 (rule
->mark
!= nla_get_u32(tb
[FRA_FWMARK
])))
472 if (tb
[FRA_FWMASK
] &&
473 (rule
->mark_mask
!= nla_get_u32(tb
[FRA_FWMASK
])))
476 if (!ops
->compare(rule
, frh
, tb
))
479 if (rule
->flags
& FIB_RULE_PERMANENT
) {
485 err
= ops
->delete(rule
);
490 list_del_rcu(&rule
->list
);
492 if (rule
->action
== FR_ACT_GOTO
) {
493 ops
->nr_goto_rules
--;
494 if (rtnl_dereference(rule
->ctarget
) == NULL
)
495 ops
->unresolved_rules
--;
499 * Check if this rule is a target to any of them. If so,
500 * disable them. As this operation is eventually very
501 * expensive, it is only performed if goto rules have
502 * actually been added.
504 if (ops
->nr_goto_rules
> 0) {
505 list_for_each_entry(tmp
, &ops
->rules_list
, list
) {
506 if (rtnl_dereference(tmp
->ctarget
) == rule
) {
507 RCU_INIT_POINTER(tmp
->ctarget
, NULL
);
508 ops
->unresolved_rules
++;
513 notify_rule_change(RTM_DELRULE
, rule
, ops
, nlh
,
514 NETLINK_CB(skb
).portid
);
516 flush_route_cache(ops
);
527 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops
*ops
,
528 struct fib_rule
*rule
)
530 size_t payload
= NLMSG_ALIGN(sizeof(struct fib_rule_hdr
))
531 + nla_total_size(IFNAMSIZ
) /* FRA_IIFNAME */
532 + nla_total_size(IFNAMSIZ
) /* FRA_OIFNAME */
533 + nla_total_size(4) /* FRA_PRIORITY */
534 + nla_total_size(4) /* FRA_TABLE */
535 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
536 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
537 + nla_total_size(4) /* FRA_FWMARK */
538 + nla_total_size(4); /* FRA_FWMASK */
540 if (ops
->nlmsg_payload
)
541 payload
+= ops
->nlmsg_payload(rule
);
546 static int fib_nl_fill_rule(struct sk_buff
*skb
, struct fib_rule
*rule
,
547 u32 pid
, u32 seq
, int type
, int flags
,
548 struct fib_rules_ops
*ops
)
550 struct nlmsghdr
*nlh
;
551 struct fib_rule_hdr
*frh
;
553 nlh
= nlmsg_put(skb
, pid
, seq
, type
, sizeof(*frh
), flags
);
557 frh
= nlmsg_data(nlh
);
558 frh
->family
= ops
->family
;
559 frh
->table
= rule
->table
;
560 if (nla_put_u32(skb
, FRA_TABLE
, rule
->table
))
561 goto nla_put_failure
;
562 if (nla_put_u32(skb
, FRA_SUPPRESS_PREFIXLEN
, rule
->suppress_prefixlen
))
563 goto nla_put_failure
;
566 frh
->action
= rule
->action
;
567 frh
->flags
= rule
->flags
;
569 if (rule
->action
== FR_ACT_GOTO
&&
570 rcu_access_pointer(rule
->ctarget
) == NULL
)
571 frh
->flags
|= FIB_RULE_UNRESOLVED
;
573 if (rule
->iifname
[0]) {
574 if (nla_put_string(skb
, FRA_IIFNAME
, rule
->iifname
))
575 goto nla_put_failure
;
576 if (rule
->iifindex
== -1)
577 frh
->flags
|= FIB_RULE_IIF_DETACHED
;
580 if (rule
->oifname
[0]) {
581 if (nla_put_string(skb
, FRA_OIFNAME
, rule
->oifname
))
582 goto nla_put_failure
;
583 if (rule
->oifindex
== -1)
584 frh
->flags
|= FIB_RULE_OIF_DETACHED
;
588 nla_put_u32(skb
, FRA_PRIORITY
, rule
->pref
)) ||
590 nla_put_u32(skb
, FRA_FWMARK
, rule
->mark
)) ||
591 ((rule
->mark_mask
|| rule
->mark
) &&
592 nla_put_u32(skb
, FRA_FWMASK
, rule
->mark_mask
)) ||
594 nla_put_u32(skb
, FRA_GOTO
, rule
->target
)))
595 goto nla_put_failure
;
597 if (rule
->suppress_ifgroup
!= -1) {
598 if (nla_put_u32(skb
, FRA_SUPPRESS_IFGROUP
, rule
->suppress_ifgroup
))
599 goto nla_put_failure
;
602 if (ops
->fill(rule
, skb
, frh
) < 0)
603 goto nla_put_failure
;
609 nlmsg_cancel(skb
, nlh
);
613 static int dump_rules(struct sk_buff
*skb
, struct netlink_callback
*cb
,
614 struct fib_rules_ops
*ops
)
617 struct fib_rule
*rule
;
621 list_for_each_entry_rcu(rule
, &ops
->rules_list
, list
) {
622 if (idx
< cb
->args
[1])
625 err
= fib_nl_fill_rule(skb
, rule
, NETLINK_CB(cb
->skb
).portid
,
626 cb
->nlh
->nlmsg_seq
, RTM_NEWRULE
,
640 static int fib_nl_dumprule(struct sk_buff
*skb
, struct netlink_callback
*cb
)
642 struct net
*net
= sock_net(skb
->sk
);
643 struct fib_rules_ops
*ops
;
646 family
= rtnl_msg_family(cb
->nlh
);
647 if (family
!= AF_UNSPEC
) {
648 /* Protocol specific dump request */
649 ops
= lookup_rules_ops(net
, family
);
651 return -EAFNOSUPPORT
;
653 dump_rules(skb
, cb
, ops
);
659 list_for_each_entry_rcu(ops
, &net
->rules_ops
, list
) {
660 if (idx
< cb
->args
[0] || !try_module_get(ops
->owner
))
663 if (dump_rules(skb
, cb
, ops
) < 0)
676 static void notify_rule_change(int event
, struct fib_rule
*rule
,
677 struct fib_rules_ops
*ops
, struct nlmsghdr
*nlh
,
685 skb
= nlmsg_new(fib_rule_nlmsg_size(ops
, rule
), GFP_KERNEL
);
689 err
= fib_nl_fill_rule(skb
, rule
, pid
, nlh
->nlmsg_seq
, event
, 0, ops
);
691 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
692 WARN_ON(err
== -EMSGSIZE
);
697 rtnl_notify(skb
, net
, pid
, ops
->nlgroup
, nlh
, GFP_KERNEL
);
701 rtnl_set_sk_err(net
, ops
->nlgroup
, err
);
704 static void attach_rules(struct list_head
*rules
, struct net_device
*dev
)
706 struct fib_rule
*rule
;
708 list_for_each_entry(rule
, rules
, list
) {
709 if (rule
->iifindex
== -1 &&
710 strcmp(dev
->name
, rule
->iifname
) == 0)
711 rule
->iifindex
= dev
->ifindex
;
712 if (rule
->oifindex
== -1 &&
713 strcmp(dev
->name
, rule
->oifname
) == 0)
714 rule
->oifindex
= dev
->ifindex
;
718 static void detach_rules(struct list_head
*rules
, struct net_device
*dev
)
720 struct fib_rule
*rule
;
722 list_for_each_entry(rule
, rules
, list
) {
723 if (rule
->iifindex
== dev
->ifindex
)
725 if (rule
->oifindex
== dev
->ifindex
)
731 static int fib_rules_event(struct notifier_block
*this, unsigned long event
,
734 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
735 struct net
*net
= dev_net(dev
);
736 struct fib_rules_ops
*ops
;
741 case NETDEV_REGISTER
:
742 list_for_each_entry(ops
, &net
->rules_ops
, list
)
743 attach_rules(&ops
->rules_list
, dev
);
746 case NETDEV_CHANGENAME
:
747 list_for_each_entry(ops
, &net
->rules_ops
, list
) {
748 detach_rules(&ops
->rules_list
, dev
);
749 attach_rules(&ops
->rules_list
, dev
);
753 case NETDEV_UNREGISTER
:
754 list_for_each_entry(ops
, &net
->rules_ops
, list
)
755 detach_rules(&ops
->rules_list
, dev
);
762 static struct notifier_block fib_rules_notifier
= {
763 .notifier_call
= fib_rules_event
,
766 static int __net_init
fib_rules_net_init(struct net
*net
)
768 INIT_LIST_HEAD(&net
->rules_ops
);
769 spin_lock_init(&net
->rules_mod_lock
);
773 static struct pernet_operations fib_rules_net_ops
= {
774 .init
= fib_rules_net_init
,
777 static int __init
fib_rules_init(void)
780 rtnl_register(PF_UNSPEC
, RTM_NEWRULE
, fib_nl_newrule
, NULL
, NULL
);
781 rtnl_register(PF_UNSPEC
, RTM_DELRULE
, fib_nl_delrule
, NULL
, NULL
);
782 rtnl_register(PF_UNSPEC
, RTM_GETRULE
, NULL
, fib_nl_dumprule
, NULL
);
784 err
= register_pernet_subsys(&fib_rules_net_ops
);
788 err
= register_netdevice_notifier(&fib_rules_notifier
);
790 goto fail_unregister
;
795 unregister_pernet_subsys(&fib_rules_net_ops
);
797 rtnl_unregister(PF_UNSPEC
, RTM_NEWRULE
);
798 rtnl_unregister(PF_UNSPEC
, RTM_DELRULE
);
799 rtnl_unregister(PF_UNSPEC
, RTM_GETRULE
);
803 subsys_initcall(fib_rules_init
);