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>
19 #include <net/ip_tunnels.h>
21 int fib_default_rule_add(struct fib_rules_ops
*ops
,
22 u32 pref
, u32 table
, u32 flags
)
26 r
= kzalloc(ops
->rule_size
, GFP_KERNEL
);
30 atomic_set(&r
->refcnt
, 1);
31 r
->action
= FR_ACT_TO_TBL
;
35 r
->fr_net
= ops
->fro_net
;
37 r
->suppress_prefixlen
= -1;
38 r
->suppress_ifgroup
= -1;
40 /* The lock is not required here, the list in unreacheable
41 * at the moment this function is called */
42 list_add_tail(&r
->list
, &ops
->rules_list
);
45 EXPORT_SYMBOL(fib_default_rule_add
);
47 static u32
fib_default_rule_pref(struct fib_rules_ops
*ops
)
49 struct list_head
*pos
;
50 struct fib_rule
*rule
;
52 if (!list_empty(&ops
->rules_list
)) {
53 pos
= ops
->rules_list
.next
;
54 if (pos
->next
!= &ops
->rules_list
) {
55 rule
= list_entry(pos
->next
, struct fib_rule
, list
);
57 return rule
->pref
- 1;
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 if (rule
->tun_id
&& (rule
->tun_id
!= fl
->flowi_tun_key
.tun_id
))
192 ret
= ops
->match(rule
, fl
, flags
);
194 return (rule
->flags
& FIB_RULE_INVERT
) ? !ret
: ret
;
197 int fib_rules_lookup(struct fib_rules_ops
*ops
, struct flowi
*fl
,
198 int flags
, struct fib_lookup_arg
*arg
)
200 struct fib_rule
*rule
;
205 list_for_each_entry_rcu(rule
, &ops
->rules_list
, list
) {
207 if (!fib_rule_match(rule
, ops
, fl
, flags
))
210 if (rule
->action
== FR_ACT_GOTO
) {
211 struct fib_rule
*target
;
213 target
= rcu_dereference(rule
->ctarget
);
214 if (target
== NULL
) {
220 } else if (rule
->action
== FR_ACT_NOP
)
223 err
= ops
->action(rule
, fl
, flags
, arg
);
225 if (!err
&& ops
->suppress
&& ops
->suppress(rule
, arg
))
228 if (err
!= -EAGAIN
) {
229 if ((arg
->flags
& FIB_LOOKUP_NOREF
) ||
230 likely(atomic_inc_not_zero(&rule
->refcnt
))) {
244 EXPORT_SYMBOL_GPL(fib_rules_lookup
);
246 static int validate_rulemsg(struct fib_rule_hdr
*frh
, struct nlattr
**tb
,
247 struct fib_rules_ops
*ops
)
252 if (tb
[FRA_SRC
] == NULL
||
253 frh
->src_len
> (ops
->addr_size
* 8) ||
254 nla_len(tb
[FRA_SRC
]) != ops
->addr_size
)
258 if (tb
[FRA_DST
] == NULL
||
259 frh
->dst_len
> (ops
->addr_size
* 8) ||
260 nla_len(tb
[FRA_DST
]) != ops
->addr_size
)
268 static int fib_nl_newrule(struct sk_buff
*skb
, struct nlmsghdr
* nlh
)
270 struct net
*net
= sock_net(skb
->sk
);
271 struct fib_rule_hdr
*frh
= nlmsg_data(nlh
);
272 struct fib_rules_ops
*ops
= NULL
;
273 struct fib_rule
*rule
, *r
, *last
= NULL
;
274 struct nlattr
*tb
[FRA_MAX
+1];
275 int err
= -EINVAL
, unresolved
= 0;
277 if (nlh
->nlmsg_len
< nlmsg_msg_size(sizeof(*frh
)))
280 ops
= lookup_rules_ops(net
, frh
->family
);
286 err
= nlmsg_parse(nlh
, sizeof(*frh
), tb
, FRA_MAX
, ops
->policy
);
290 err
= validate_rulemsg(frh
, tb
, ops
);
294 rule
= kzalloc(ops
->rule_size
, GFP_KERNEL
);
301 rule
->pref
= tb
[FRA_PRIORITY
] ? nla_get_u32(tb
[FRA_PRIORITY
])
302 : fib_default_rule_pref(ops
);
304 if (tb
[FRA_IIFNAME
]) {
305 struct net_device
*dev
;
308 nla_strlcpy(rule
->iifname
, tb
[FRA_IIFNAME
], IFNAMSIZ
);
309 dev
= __dev_get_by_name(net
, rule
->iifname
);
311 rule
->iifindex
= dev
->ifindex
;
314 if (tb
[FRA_OIFNAME
]) {
315 struct net_device
*dev
;
318 nla_strlcpy(rule
->oifname
, tb
[FRA_OIFNAME
], IFNAMSIZ
);
319 dev
= __dev_get_by_name(net
, rule
->oifname
);
321 rule
->oifindex
= dev
->ifindex
;
324 if (tb
[FRA_FWMARK
]) {
325 rule
->mark
= nla_get_u32(tb
[FRA_FWMARK
]);
327 /* compatibility: if the mark value is non-zero all bits
328 * are compared unless a mask is explicitly specified.
330 rule
->mark_mask
= 0xFFFFFFFF;
334 rule
->mark_mask
= nla_get_u32(tb
[FRA_FWMASK
]);
337 rule
->tun_id
= nla_get_be64(tb
[FRA_TUN_ID
]);
339 rule
->action
= frh
->action
;
340 rule
->flags
= frh
->flags
;
341 rule
->table
= frh_get_table(frh
, tb
);
342 if (tb
[FRA_SUPPRESS_PREFIXLEN
])
343 rule
->suppress_prefixlen
= nla_get_u32(tb
[FRA_SUPPRESS_PREFIXLEN
]);
345 rule
->suppress_prefixlen
= -1;
347 if (tb
[FRA_SUPPRESS_IFGROUP
])
348 rule
->suppress_ifgroup
= nla_get_u32(tb
[FRA_SUPPRESS_IFGROUP
]);
350 rule
->suppress_ifgroup
= -1;
354 if (rule
->action
!= FR_ACT_GOTO
)
357 rule
->target
= nla_get_u32(tb
[FRA_GOTO
]);
358 /* Backward jumps are prohibited to avoid endless loops */
359 if (rule
->target
<= rule
->pref
)
362 list_for_each_entry(r
, &ops
->rules_list
, list
) {
363 if (r
->pref
== rule
->target
) {
364 RCU_INIT_POINTER(rule
->ctarget
, r
);
369 if (rcu_dereference_protected(rule
->ctarget
, 1) == NULL
)
371 } else if (rule
->action
== FR_ACT_GOTO
)
374 err
= ops
->configure(rule
, skb
, frh
, tb
);
378 list_for_each_entry(r
, &ops
->rules_list
, list
) {
379 if (r
->pref
> rule
->pref
)
387 list_add_rcu(&rule
->list
, &last
->list
);
389 list_add_rcu(&rule
->list
, &ops
->rules_list
);
391 if (ops
->unresolved_rules
) {
393 * There are unresolved goto rules in the list, check if
394 * any of them are pointing to this new rule.
396 list_for_each_entry(r
, &ops
->rules_list
, list
) {
397 if (r
->action
== FR_ACT_GOTO
&&
398 r
->target
== rule
->pref
&&
399 rtnl_dereference(r
->ctarget
) == NULL
) {
400 rcu_assign_pointer(r
->ctarget
, rule
);
401 if (--ops
->unresolved_rules
== 0)
407 if (rule
->action
== FR_ACT_GOTO
)
408 ops
->nr_goto_rules
++;
411 ops
->unresolved_rules
++;
414 ip_tunnel_need_metadata();
416 notify_rule_change(RTM_NEWRULE
, rule
, ops
, nlh
, NETLINK_CB(skb
).portid
);
417 flush_route_cache(ops
);
428 static int fib_nl_delrule(struct sk_buff
*skb
, struct nlmsghdr
* nlh
)
430 struct net
*net
= sock_net(skb
->sk
);
431 struct fib_rule_hdr
*frh
= nlmsg_data(nlh
);
432 struct fib_rules_ops
*ops
= NULL
;
433 struct fib_rule
*rule
, *tmp
;
434 struct nlattr
*tb
[FRA_MAX
+1];
437 if (nlh
->nlmsg_len
< nlmsg_msg_size(sizeof(*frh
)))
440 ops
= lookup_rules_ops(net
, frh
->family
);
446 err
= nlmsg_parse(nlh
, sizeof(*frh
), tb
, FRA_MAX
, ops
->policy
);
450 err
= validate_rulemsg(frh
, tb
, ops
);
454 list_for_each_entry(rule
, &ops
->rules_list
, list
) {
455 if (frh
->action
&& (frh
->action
!= rule
->action
))
458 if (frh_get_table(frh
, tb
) &&
459 (frh_get_table(frh
, tb
) != rule
->table
))
462 if (tb
[FRA_PRIORITY
] &&
463 (rule
->pref
!= nla_get_u32(tb
[FRA_PRIORITY
])))
466 if (tb
[FRA_IIFNAME
] &&
467 nla_strcmp(tb
[FRA_IIFNAME
], rule
->iifname
))
470 if (tb
[FRA_OIFNAME
] &&
471 nla_strcmp(tb
[FRA_OIFNAME
], rule
->oifname
))
474 if (tb
[FRA_FWMARK
] &&
475 (rule
->mark
!= nla_get_u32(tb
[FRA_FWMARK
])))
478 if (tb
[FRA_FWMASK
] &&
479 (rule
->mark_mask
!= nla_get_u32(tb
[FRA_FWMASK
])))
482 if (tb
[FRA_TUN_ID
] &&
483 (rule
->tun_id
!= nla_get_be64(tb
[FRA_TUN_ID
])))
486 if (!ops
->compare(rule
, frh
, tb
))
489 if (rule
->flags
& FIB_RULE_PERMANENT
) {
495 err
= ops
->delete(rule
);
501 ip_tunnel_unneed_metadata();
503 list_del_rcu(&rule
->list
);
505 if (rule
->action
== FR_ACT_GOTO
) {
506 ops
->nr_goto_rules
--;
507 if (rtnl_dereference(rule
->ctarget
) == NULL
)
508 ops
->unresolved_rules
--;
512 * Check if this rule is a target to any of them. If so,
513 * disable them. As this operation is eventually very
514 * expensive, it is only performed if goto rules have
515 * actually been added.
517 if (ops
->nr_goto_rules
> 0) {
518 list_for_each_entry(tmp
, &ops
->rules_list
, list
) {
519 if (rtnl_dereference(tmp
->ctarget
) == rule
) {
520 RCU_INIT_POINTER(tmp
->ctarget
, NULL
);
521 ops
->unresolved_rules
++;
526 notify_rule_change(RTM_DELRULE
, rule
, ops
, nlh
,
527 NETLINK_CB(skb
).portid
);
529 flush_route_cache(ops
);
540 static inline size_t fib_rule_nlmsg_size(struct fib_rules_ops
*ops
,
541 struct fib_rule
*rule
)
543 size_t payload
= NLMSG_ALIGN(sizeof(struct fib_rule_hdr
))
544 + nla_total_size(IFNAMSIZ
) /* FRA_IIFNAME */
545 + nla_total_size(IFNAMSIZ
) /* FRA_OIFNAME */
546 + nla_total_size(4) /* FRA_PRIORITY */
547 + nla_total_size(4) /* FRA_TABLE */
548 + nla_total_size(4) /* FRA_SUPPRESS_PREFIXLEN */
549 + nla_total_size(4) /* FRA_SUPPRESS_IFGROUP */
550 + nla_total_size(4) /* FRA_FWMARK */
551 + nla_total_size(4) /* FRA_FWMASK */
552 + nla_total_size(8); /* FRA_TUN_ID */
554 if (ops
->nlmsg_payload
)
555 payload
+= ops
->nlmsg_payload(rule
);
560 static int fib_nl_fill_rule(struct sk_buff
*skb
, struct fib_rule
*rule
,
561 u32 pid
, u32 seq
, int type
, int flags
,
562 struct fib_rules_ops
*ops
)
564 struct nlmsghdr
*nlh
;
565 struct fib_rule_hdr
*frh
;
567 nlh
= nlmsg_put(skb
, pid
, seq
, type
, sizeof(*frh
), flags
);
571 frh
= nlmsg_data(nlh
);
572 frh
->family
= ops
->family
;
573 frh
->table
= rule
->table
;
574 if (nla_put_u32(skb
, FRA_TABLE
, rule
->table
))
575 goto nla_put_failure
;
576 if (nla_put_u32(skb
, FRA_SUPPRESS_PREFIXLEN
, rule
->suppress_prefixlen
))
577 goto nla_put_failure
;
580 frh
->action
= rule
->action
;
581 frh
->flags
= rule
->flags
;
583 if (rule
->action
== FR_ACT_GOTO
&&
584 rcu_access_pointer(rule
->ctarget
) == NULL
)
585 frh
->flags
|= FIB_RULE_UNRESOLVED
;
587 if (rule
->iifname
[0]) {
588 if (nla_put_string(skb
, FRA_IIFNAME
, rule
->iifname
))
589 goto nla_put_failure
;
590 if (rule
->iifindex
== -1)
591 frh
->flags
|= FIB_RULE_IIF_DETACHED
;
594 if (rule
->oifname
[0]) {
595 if (nla_put_string(skb
, FRA_OIFNAME
, rule
->oifname
))
596 goto nla_put_failure
;
597 if (rule
->oifindex
== -1)
598 frh
->flags
|= FIB_RULE_OIF_DETACHED
;
602 nla_put_u32(skb
, FRA_PRIORITY
, rule
->pref
)) ||
604 nla_put_u32(skb
, FRA_FWMARK
, rule
->mark
)) ||
605 ((rule
->mark_mask
|| rule
->mark
) &&
606 nla_put_u32(skb
, FRA_FWMASK
, rule
->mark_mask
)) ||
608 nla_put_u32(skb
, FRA_GOTO
, rule
->target
)) ||
610 nla_put_be64(skb
, FRA_TUN_ID
, rule
->tun_id
)))
611 goto nla_put_failure
;
613 if (rule
->suppress_ifgroup
!= -1) {
614 if (nla_put_u32(skb
, FRA_SUPPRESS_IFGROUP
, rule
->suppress_ifgroup
))
615 goto nla_put_failure
;
618 if (ops
->fill(rule
, skb
, frh
) < 0)
619 goto nla_put_failure
;
625 nlmsg_cancel(skb
, nlh
);
629 static int dump_rules(struct sk_buff
*skb
, struct netlink_callback
*cb
,
630 struct fib_rules_ops
*ops
)
633 struct fib_rule
*rule
;
637 list_for_each_entry_rcu(rule
, &ops
->rules_list
, list
) {
638 if (idx
< cb
->args
[1])
641 err
= fib_nl_fill_rule(skb
, rule
, NETLINK_CB(cb
->skb
).portid
,
642 cb
->nlh
->nlmsg_seq
, RTM_NEWRULE
,
656 static int fib_nl_dumprule(struct sk_buff
*skb
, struct netlink_callback
*cb
)
658 struct net
*net
= sock_net(skb
->sk
);
659 struct fib_rules_ops
*ops
;
662 family
= rtnl_msg_family(cb
->nlh
);
663 if (family
!= AF_UNSPEC
) {
664 /* Protocol specific dump request */
665 ops
= lookup_rules_ops(net
, family
);
667 return -EAFNOSUPPORT
;
669 dump_rules(skb
, cb
, ops
);
675 list_for_each_entry_rcu(ops
, &net
->rules_ops
, list
) {
676 if (idx
< cb
->args
[0] || !try_module_get(ops
->owner
))
679 if (dump_rules(skb
, cb
, ops
) < 0)
692 static void notify_rule_change(int event
, struct fib_rule
*rule
,
693 struct fib_rules_ops
*ops
, struct nlmsghdr
*nlh
,
701 skb
= nlmsg_new(fib_rule_nlmsg_size(ops
, rule
), GFP_KERNEL
);
705 err
= fib_nl_fill_rule(skb
, rule
, pid
, nlh
->nlmsg_seq
, event
, 0, ops
);
707 /* -EMSGSIZE implies BUG in fib_rule_nlmsg_size() */
708 WARN_ON(err
== -EMSGSIZE
);
713 rtnl_notify(skb
, net
, pid
, ops
->nlgroup
, nlh
, GFP_KERNEL
);
717 rtnl_set_sk_err(net
, ops
->nlgroup
, err
);
720 static void attach_rules(struct list_head
*rules
, struct net_device
*dev
)
722 struct fib_rule
*rule
;
724 list_for_each_entry(rule
, rules
, list
) {
725 if (rule
->iifindex
== -1 &&
726 strcmp(dev
->name
, rule
->iifname
) == 0)
727 rule
->iifindex
= dev
->ifindex
;
728 if (rule
->oifindex
== -1 &&
729 strcmp(dev
->name
, rule
->oifname
) == 0)
730 rule
->oifindex
= dev
->ifindex
;
734 static void detach_rules(struct list_head
*rules
, struct net_device
*dev
)
736 struct fib_rule
*rule
;
738 list_for_each_entry(rule
, rules
, list
) {
739 if (rule
->iifindex
== dev
->ifindex
)
741 if (rule
->oifindex
== dev
->ifindex
)
747 static int fib_rules_event(struct notifier_block
*this, unsigned long event
,
750 struct net_device
*dev
= netdev_notifier_info_to_dev(ptr
);
751 struct net
*net
= dev_net(dev
);
752 struct fib_rules_ops
*ops
;
757 case NETDEV_REGISTER
:
758 list_for_each_entry(ops
, &net
->rules_ops
, list
)
759 attach_rules(&ops
->rules_list
, dev
);
762 case NETDEV_CHANGENAME
:
763 list_for_each_entry(ops
, &net
->rules_ops
, list
) {
764 detach_rules(&ops
->rules_list
, dev
);
765 attach_rules(&ops
->rules_list
, dev
);
769 case NETDEV_UNREGISTER
:
770 list_for_each_entry(ops
, &net
->rules_ops
, list
)
771 detach_rules(&ops
->rules_list
, dev
);
778 static struct notifier_block fib_rules_notifier
= {
779 .notifier_call
= fib_rules_event
,
782 static int __net_init
fib_rules_net_init(struct net
*net
)
784 INIT_LIST_HEAD(&net
->rules_ops
);
785 spin_lock_init(&net
->rules_mod_lock
);
789 static struct pernet_operations fib_rules_net_ops
= {
790 .init
= fib_rules_net_init
,
793 static int __init
fib_rules_init(void)
796 rtnl_register(PF_UNSPEC
, RTM_NEWRULE
, fib_nl_newrule
, NULL
, NULL
);
797 rtnl_register(PF_UNSPEC
, RTM_DELRULE
, fib_nl_delrule
, NULL
, NULL
);
798 rtnl_register(PF_UNSPEC
, RTM_GETRULE
, NULL
, fib_nl_dumprule
, NULL
);
800 err
= register_pernet_subsys(&fib_rules_net_ops
);
804 err
= register_netdevice_notifier(&fib_rules_notifier
);
806 goto fail_unregister
;
811 unregister_pernet_subsys(&fib_rules_net_ops
);
813 rtnl_unregister(PF_UNSPEC
, RTM_NEWRULE
);
814 rtnl_unregister(PF_UNSPEC
, RTM_DELRULE
);
815 rtnl_unregister(PF_UNSPEC
, RTM_GETRULE
);
819 subsys_initcall(fib_rules_init
);