2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * IPv4 Forwarding Information Base: semantics.
8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 #include <asm/uaccess.h>
17 #include <asm/system.h>
18 #include <linux/bitops.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/jiffies.h>
23 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/errno.h>
28 #include <linux/inet.h>
29 #include <linux/inetdevice.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/proc_fs.h>
33 #include <linux/skbuff.h>
34 #include <linux/init.h>
35 #include <linux/slab.h>
39 #include <net/protocol.h>
40 #include <net/route.h>
43 #include <net/ip_fib.h>
44 #include <net/netlink.h>
45 #include <net/nexthop.h>
47 #include "fib_lookup.h"
49 static DEFINE_SPINLOCK(fib_info_lock
);
50 static struct hlist_head
*fib_info_hash
;
51 static struct hlist_head
*fib_info_laddrhash
;
52 static unsigned int fib_info_hash_size
;
53 static unsigned int fib_info_cnt
;
55 #define DEVINDEX_HASHBITS 8
56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
57 static struct hlist_head fib_info_devhash
[DEVINDEX_HASHSIZE
];
59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
61 static DEFINE_SPINLOCK(fib_multipath_lock
);
63 #define for_nexthops(fi) { \
64 int nhsel; const struct fib_nh *nh; \
65 for (nhsel = 0, nh = (fi)->fib_nh; \
66 nhsel < (fi)->fib_nhs; \
69 #define change_nexthops(fi) { \
70 int nhsel; struct fib_nh *nexthop_nh; \
71 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
72 nhsel < (fi)->fib_nhs; \
73 nexthop_nh++, nhsel++)
75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
77 /* Hope, that gcc will optimize it to get rid of dummy loop */
79 #define for_nexthops(fi) { \
80 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
81 for (nhsel = 0; nhsel < 1; nhsel++)
83 #define change_nexthops(fi) { \
85 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
86 for (nhsel = 0; nhsel < 1; nhsel++)
88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
90 #define endfor_nexthops(fi) }
97 } fib_props
[RTN_MAX
+ 1] = {
100 .scope
= RT_SCOPE_NOWHERE
,
104 .scope
= RT_SCOPE_UNIVERSE
,
108 .scope
= RT_SCOPE_HOST
,
112 .scope
= RT_SCOPE_LINK
,
116 .scope
= RT_SCOPE_LINK
,
120 .scope
= RT_SCOPE_UNIVERSE
,
124 .scope
= RT_SCOPE_UNIVERSE
,
126 [RTN_UNREACHABLE
] = {
127 .error
= -EHOSTUNREACH
,
128 .scope
= RT_SCOPE_UNIVERSE
,
132 .scope
= RT_SCOPE_UNIVERSE
,
136 .scope
= RT_SCOPE_UNIVERSE
,
140 .scope
= RT_SCOPE_NOWHERE
,
144 .scope
= RT_SCOPE_NOWHERE
,
149 /* Release a nexthop info record */
151 static void free_fib_info_rcu(struct rcu_head
*head
)
153 struct fib_info
*fi
= container_of(head
, struct fib_info
, rcu
);
155 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
)
156 kfree(fi
->fib_metrics
);
160 void free_fib_info(struct fib_info
*fi
)
162 if (fi
->fib_dead
== 0) {
163 pr_warning("Freeing alive fib_info %p\n", fi
);
166 change_nexthops(fi
) {
167 if (nexthop_nh
->nh_dev
)
168 dev_put(nexthop_nh
->nh_dev
);
169 nexthop_nh
->nh_dev
= NULL
;
170 } endfor_nexthops(fi
);
172 release_net(fi
->fib_net
);
173 call_rcu(&fi
->rcu
, free_fib_info_rcu
);
176 void fib_release_info(struct fib_info
*fi
)
178 spin_lock_bh(&fib_info_lock
);
179 if (fi
&& --fi
->fib_treeref
== 0) {
180 hlist_del(&fi
->fib_hash
);
182 hlist_del(&fi
->fib_lhash
);
183 change_nexthops(fi
) {
184 if (!nexthop_nh
->nh_dev
)
186 hlist_del(&nexthop_nh
->nh_hash
);
187 } endfor_nexthops(fi
)
191 spin_unlock_bh(&fib_info_lock
);
194 static inline int nh_comp(const struct fib_info
*fi
, const struct fib_info
*ofi
)
196 const struct fib_nh
*onh
= ofi
->fib_nh
;
199 if (nh
->nh_oif
!= onh
->nh_oif
||
200 nh
->nh_gw
!= onh
->nh_gw
||
201 nh
->nh_scope
!= onh
->nh_scope
||
202 #ifdef CONFIG_IP_ROUTE_MULTIPATH
203 nh
->nh_weight
!= onh
->nh_weight
||
205 #ifdef CONFIG_IP_ROUTE_CLASSID
206 nh
->nh_tclassid
!= onh
->nh_tclassid
||
208 ((nh
->nh_flags
^ onh
->nh_flags
) & ~RTNH_F_DEAD
))
211 } endfor_nexthops(fi
);
215 static inline unsigned int fib_devindex_hashfn(unsigned int val
)
217 unsigned int mask
= DEVINDEX_HASHSIZE
- 1;
220 (val
>> DEVINDEX_HASHBITS
) ^
221 (val
>> (DEVINDEX_HASHBITS
* 2))) & mask
;
224 static inline unsigned int fib_info_hashfn(const struct fib_info
*fi
)
226 unsigned int mask
= (fib_info_hash_size
- 1);
227 unsigned int val
= fi
->fib_nhs
;
229 val
^= fi
->fib_protocol
;
230 val
^= (__force u32
)fi
->fib_prefsrc
;
231 val
^= fi
->fib_priority
;
233 val
^= fib_devindex_hashfn(nh
->nh_oif
);
234 } endfor_nexthops(fi
)
236 return (val
^ (val
>> 7) ^ (val
>> 12)) & mask
;
239 static struct fib_info
*fib_find_info(const struct fib_info
*nfi
)
241 struct hlist_head
*head
;
242 struct hlist_node
*node
;
246 hash
= fib_info_hashfn(nfi
);
247 head
= &fib_info_hash
[hash
];
249 hlist_for_each_entry(fi
, node
, head
, fib_hash
) {
250 if (!net_eq(fi
->fib_net
, nfi
->fib_net
))
252 if (fi
->fib_nhs
!= nfi
->fib_nhs
)
254 if (nfi
->fib_protocol
== fi
->fib_protocol
&&
255 nfi
->fib_prefsrc
== fi
->fib_prefsrc
&&
256 nfi
->fib_priority
== fi
->fib_priority
&&
257 memcmp(nfi
->fib_metrics
, fi
->fib_metrics
,
258 sizeof(fi
->fib_metrics
)) == 0 &&
259 ((nfi
->fib_flags
^ fi
->fib_flags
) & ~RTNH_F_DEAD
) == 0 &&
260 (nfi
->fib_nhs
== 0 || nh_comp(fi
, nfi
) == 0))
267 /* Check, that the gateway is already configured.
268 * Used only by redirect accept routine.
270 int ip_fib_check_default(__be32 gw
, struct net_device
*dev
)
272 struct hlist_head
*head
;
273 struct hlist_node
*node
;
277 spin_lock(&fib_info_lock
);
279 hash
= fib_devindex_hashfn(dev
->ifindex
);
280 head
= &fib_info_devhash
[hash
];
281 hlist_for_each_entry(nh
, node
, head
, nh_hash
) {
282 if (nh
->nh_dev
== dev
&&
284 !(nh
->nh_flags
& RTNH_F_DEAD
)) {
285 spin_unlock(&fib_info_lock
);
290 spin_unlock(&fib_info_lock
);
295 static inline size_t fib_nlmsg_size(struct fib_info
*fi
)
297 size_t payload
= NLMSG_ALIGN(sizeof(struct rtmsg
))
298 + nla_total_size(4) /* RTA_TABLE */
299 + nla_total_size(4) /* RTA_DST */
300 + nla_total_size(4) /* RTA_PRIORITY */
301 + nla_total_size(4); /* RTA_PREFSRC */
303 /* space for nested metrics */
304 payload
+= nla_total_size((RTAX_MAX
* nla_total_size(4)));
307 /* Also handles the special case fib_nhs == 1 */
309 /* each nexthop is packed in an attribute */
310 size_t nhsize
= nla_total_size(sizeof(struct rtnexthop
));
312 /* may contain flow and gateway attribute */
313 nhsize
+= 2 * nla_total_size(4);
315 /* all nexthops are packed in a nested attribute */
316 payload
+= nla_total_size(fi
->fib_nhs
* nhsize
);
322 void rtmsg_fib(int event
, __be32 key
, struct fib_alias
*fa
,
323 int dst_len
, u32 tb_id
, struct nl_info
*info
,
324 unsigned int nlm_flags
)
327 u32 seq
= info
->nlh
? info
->nlh
->nlmsg_seq
: 0;
330 skb
= nlmsg_new(fib_nlmsg_size(fa
->fa_info
), GFP_KERNEL
);
334 err
= fib_dump_info(skb
, info
->pid
, seq
, event
, tb_id
,
335 fa
->fa_type
, fa
->fa_scope
, key
, dst_len
,
336 fa
->fa_tos
, fa
->fa_info
, nlm_flags
);
338 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
339 WARN_ON(err
== -EMSGSIZE
);
343 rtnl_notify(skb
, info
->nl_net
, info
->pid
, RTNLGRP_IPV4_ROUTE
,
344 info
->nlh
, GFP_KERNEL
);
348 rtnl_set_sk_err(info
->nl_net
, RTNLGRP_IPV4_ROUTE
, err
);
351 /* Return the first fib alias matching TOS with
352 * priority less than or equal to PRIO.
354 struct fib_alias
*fib_find_alias(struct list_head
*fah
, u8 tos
, u32 prio
)
357 struct fib_alias
*fa
;
358 list_for_each_entry(fa
, fah
, fa_list
) {
359 if (fa
->fa_tos
> tos
)
361 if (fa
->fa_info
->fib_priority
>= prio
||
369 int fib_detect_death(struct fib_info
*fi
, int order
,
370 struct fib_info
**last_resort
, int *last_idx
, int dflt
)
373 int state
= NUD_NONE
;
375 n
= neigh_lookup(&arp_tbl
, &fi
->fib_nh
[0].nh_gw
, fi
->fib_dev
);
377 state
= n
->nud_state
;
380 if (state
== NUD_REACHABLE
)
382 if ((state
& NUD_VALID
) && order
!= dflt
)
384 if ((state
& NUD_VALID
) ||
385 (*last_idx
< 0 && order
> dflt
)) {
392 #ifdef CONFIG_IP_ROUTE_MULTIPATH
394 static int fib_count_nexthops(struct rtnexthop
*rtnh
, int remaining
)
398 while (rtnh_ok(rtnh
, remaining
)) {
400 rtnh
= rtnh_next(rtnh
, &remaining
);
403 /* leftover implies invalid nexthop configuration, discard it */
404 return remaining
> 0 ? 0 : nhs
;
407 static int fib_get_nhs(struct fib_info
*fi
, struct rtnexthop
*rtnh
,
408 int remaining
, struct fib_config
*cfg
)
410 change_nexthops(fi
) {
413 if (!rtnh_ok(rtnh
, remaining
))
416 nexthop_nh
->nh_flags
=
417 (cfg
->fc_flags
& ~0xFF) | rtnh
->rtnh_flags
;
418 nexthop_nh
->nh_oif
= rtnh
->rtnh_ifindex
;
419 nexthop_nh
->nh_weight
= rtnh
->rtnh_hops
+ 1;
421 attrlen
= rtnh_attrlen(rtnh
);
423 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
425 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
426 nexthop_nh
->nh_gw
= nla
? nla_get_be32(nla
) : 0;
427 #ifdef CONFIG_IP_ROUTE_CLASSID
428 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
429 nexthop_nh
->nh_tclassid
= nla
? nla_get_u32(nla
) : 0;
433 rtnh
= rtnh_next(rtnh
, &remaining
);
434 } endfor_nexthops(fi
);
441 int fib_nh_match(struct fib_config
*cfg
, struct fib_info
*fi
)
443 #ifdef CONFIG_IP_ROUTE_MULTIPATH
444 struct rtnexthop
*rtnh
;
448 if (cfg
->fc_priority
&& cfg
->fc_priority
!= fi
->fib_priority
)
451 if (cfg
->fc_oif
|| cfg
->fc_gw
) {
452 if ((!cfg
->fc_oif
|| cfg
->fc_oif
== fi
->fib_nh
->nh_oif
) &&
453 (!cfg
->fc_gw
|| cfg
->fc_gw
== fi
->fib_nh
->nh_gw
))
458 #ifdef CONFIG_IP_ROUTE_MULTIPATH
459 if (cfg
->fc_mp
== NULL
)
463 remaining
= cfg
->fc_mp_len
;
468 if (!rtnh_ok(rtnh
, remaining
))
471 if (rtnh
->rtnh_ifindex
&& rtnh
->rtnh_ifindex
!= nh
->nh_oif
)
474 attrlen
= rtnh_attrlen(rtnh
);
476 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
478 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
479 if (nla
&& nla_get_be32(nla
) != nh
->nh_gw
)
481 #ifdef CONFIG_IP_ROUTE_CLASSID
482 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
483 if (nla
&& nla_get_u32(nla
) != nh
->nh_tclassid
)
488 rtnh
= rtnh_next(rtnh
, &remaining
);
489 } endfor_nexthops(fi
);
499 * Semantics of nexthop is very messy by historical reasons.
500 * We have to take into account, that:
501 * a) gateway can be actually local interface address,
502 * so that gatewayed route is direct.
503 * b) gateway must be on-link address, possibly
504 * described not by an ifaddr, but also by a direct route.
505 * c) If both gateway and interface are specified, they should not
507 * d) If we use tunnel routes, gateway could be not on-link.
509 * Attempt to reconcile all of these (alas, self-contradictory) conditions
510 * results in pretty ugly and hairy code with obscure logic.
512 * I chose to generalized it instead, so that the size
513 * of code does not increase practically, but it becomes
515 * Every prefix is assigned a "scope" value: "host" is local address,
516 * "link" is direct route,
517 * [ ... "site" ... "interior" ... ]
518 * and "universe" is true gateway route with global meaning.
520 * Every prefix refers to a set of "nexthop"s (gw, oif),
521 * where gw must have narrower scope. This recursion stops
522 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
523 * which means that gw is forced to be on link.
525 * Code is still hairy, but now it is apparently logically
526 * consistent and very flexible. F.e. as by-product it allows
527 * to co-exists in peace independent exterior and interior
530 * Normally it looks as following.
532 * {universe prefix} -> (gw, oif) [scope link]
534 * |-> {link prefix} -> (gw, oif) [scope local]
536 * |-> {local prefix} (terminal node)
538 static int fib_check_nh(struct fib_config
*cfg
, struct fib_info
*fi
,
543 struct net_device
*dev
;
545 net
= cfg
->fc_nlinfo
.nl_net
;
547 struct fib_result res
;
549 if (nh
->nh_flags
& RTNH_F_ONLINK
) {
551 if (cfg
->fc_scope
>= RT_SCOPE_LINK
)
553 if (inet_addr_type(net
, nh
->nh_gw
) != RTN_UNICAST
)
555 dev
= __dev_get_by_index(net
, nh
->nh_oif
);
558 if (!(dev
->flags
& IFF_UP
))
562 nh
->nh_scope
= RT_SCOPE_LINK
;
568 .fl4_dst
= nh
->nh_gw
,
569 .fl4_scope
= cfg
->fc_scope
+ 1,
573 /* It is not necessary, but requires a bit of thinking */
574 if (fl
.fl4_scope
< RT_SCOPE_LINK
)
575 fl
.fl4_scope
= RT_SCOPE_LINK
;
576 err
= fib_lookup(net
, &fl
, &res
);
583 if (res
.type
!= RTN_UNICAST
&& res
.type
!= RTN_LOCAL
)
585 nh
->nh_scope
= res
.scope
;
586 nh
->nh_oif
= FIB_RES_OIF(res
);
587 nh
->nh_dev
= dev
= FIB_RES_DEV(res
);
591 err
= (dev
->flags
& IFF_UP
) ? 0 : -ENETDOWN
;
593 struct in_device
*in_dev
;
595 if (nh
->nh_flags
& (RTNH_F_PERVASIVE
| RTNH_F_ONLINK
))
600 in_dev
= inetdev_by_index(net
, nh
->nh_oif
);
604 if (!(in_dev
->dev
->flags
& IFF_UP
))
606 nh
->nh_dev
= in_dev
->dev
;
607 dev_hold(nh
->nh_dev
);
608 nh
->nh_scope
= RT_SCOPE_HOST
;
616 static inline unsigned int fib_laddr_hashfn(__be32 val
)
618 unsigned int mask
= (fib_info_hash_size
- 1);
620 return ((__force u32
)val
^
621 ((__force u32
)val
>> 7) ^
622 ((__force u32
)val
>> 14)) & mask
;
625 static struct hlist_head
*fib_info_hash_alloc(int bytes
)
627 if (bytes
<= PAGE_SIZE
)
628 return kzalloc(bytes
, GFP_KERNEL
);
630 return (struct hlist_head
*)
631 __get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
635 static void fib_info_hash_free(struct hlist_head
*hash
, int bytes
)
640 if (bytes
<= PAGE_SIZE
)
643 free_pages((unsigned long) hash
, get_order(bytes
));
646 static void fib_info_hash_move(struct hlist_head
*new_info_hash
,
647 struct hlist_head
*new_laddrhash
,
648 unsigned int new_size
)
650 struct hlist_head
*old_info_hash
, *old_laddrhash
;
651 unsigned int old_size
= fib_info_hash_size
;
652 unsigned int i
, bytes
;
654 spin_lock_bh(&fib_info_lock
);
655 old_info_hash
= fib_info_hash
;
656 old_laddrhash
= fib_info_laddrhash
;
657 fib_info_hash_size
= new_size
;
659 for (i
= 0; i
< old_size
; i
++) {
660 struct hlist_head
*head
= &fib_info_hash
[i
];
661 struct hlist_node
*node
, *n
;
664 hlist_for_each_entry_safe(fi
, node
, n
, head
, fib_hash
) {
665 struct hlist_head
*dest
;
666 unsigned int new_hash
;
668 hlist_del(&fi
->fib_hash
);
670 new_hash
= fib_info_hashfn(fi
);
671 dest
= &new_info_hash
[new_hash
];
672 hlist_add_head(&fi
->fib_hash
, dest
);
675 fib_info_hash
= new_info_hash
;
677 for (i
= 0; i
< old_size
; i
++) {
678 struct hlist_head
*lhead
= &fib_info_laddrhash
[i
];
679 struct hlist_node
*node
, *n
;
682 hlist_for_each_entry_safe(fi
, node
, n
, lhead
, fib_lhash
) {
683 struct hlist_head
*ldest
;
684 unsigned int new_hash
;
686 hlist_del(&fi
->fib_lhash
);
688 new_hash
= fib_laddr_hashfn(fi
->fib_prefsrc
);
689 ldest
= &new_laddrhash
[new_hash
];
690 hlist_add_head(&fi
->fib_lhash
, ldest
);
693 fib_info_laddrhash
= new_laddrhash
;
695 spin_unlock_bh(&fib_info_lock
);
697 bytes
= old_size
* sizeof(struct hlist_head
*);
698 fib_info_hash_free(old_info_hash
, bytes
);
699 fib_info_hash_free(old_laddrhash
, bytes
);
702 struct fib_info
*fib_create_info(struct fib_config
*cfg
)
705 struct fib_info
*fi
= NULL
;
706 struct fib_info
*ofi
;
708 struct net
*net
= cfg
->fc_nlinfo
.nl_net
;
710 /* Fast check to catch the most weird cases */
711 if (fib_props
[cfg
->fc_type
].scope
> cfg
->fc_scope
)
714 #ifdef CONFIG_IP_ROUTE_MULTIPATH
716 nhs
= fib_count_nexthops(cfg
->fc_mp
, cfg
->fc_mp_len
);
723 if (fib_info_cnt
>= fib_info_hash_size
) {
724 unsigned int new_size
= fib_info_hash_size
<< 1;
725 struct hlist_head
*new_info_hash
;
726 struct hlist_head
*new_laddrhash
;
731 bytes
= new_size
* sizeof(struct hlist_head
*);
732 new_info_hash
= fib_info_hash_alloc(bytes
);
733 new_laddrhash
= fib_info_hash_alloc(bytes
);
734 if (!new_info_hash
|| !new_laddrhash
) {
735 fib_info_hash_free(new_info_hash
, bytes
);
736 fib_info_hash_free(new_laddrhash
, bytes
);
738 fib_info_hash_move(new_info_hash
, new_laddrhash
, new_size
);
740 if (!fib_info_hash_size
)
744 fi
= kzalloc(sizeof(*fi
)+nhs
*sizeof(struct fib_nh
), GFP_KERNEL
);
748 fi
->fib_metrics
= kzalloc(sizeof(u32
) * RTAX_MAX
, GFP_KERNEL
);
749 if (!fi
->fib_metrics
)
752 fi
->fib_metrics
= (u32
*) dst_default_metrics
;
755 fi
->fib_net
= hold_net(net
);
756 fi
->fib_protocol
= cfg
->fc_protocol
;
757 fi
->fib_flags
= cfg
->fc_flags
;
758 fi
->fib_priority
= cfg
->fc_priority
;
759 fi
->fib_prefsrc
= cfg
->fc_prefsrc
;
762 change_nexthops(fi
) {
763 nexthop_nh
->nh_parent
= fi
;
764 } endfor_nexthops(fi
)
770 nla_for_each_attr(nla
, cfg
->fc_mx
, cfg
->fc_mx_len
, remaining
) {
771 int type
= nla_type(nla
);
776 fi
->fib_metrics
[type
- 1] = nla_get_u32(nla
);
782 #ifdef CONFIG_IP_ROUTE_MULTIPATH
783 err
= fib_get_nhs(fi
, cfg
->fc_mp
, cfg
->fc_mp_len
, cfg
);
786 if (cfg
->fc_oif
&& fi
->fib_nh
->nh_oif
!= cfg
->fc_oif
)
788 if (cfg
->fc_gw
&& fi
->fib_nh
->nh_gw
!= cfg
->fc_gw
)
790 #ifdef CONFIG_IP_ROUTE_CLASSID
791 if (cfg
->fc_flow
&& fi
->fib_nh
->nh_tclassid
!= cfg
->fc_flow
)
798 struct fib_nh
*nh
= fi
->fib_nh
;
800 nh
->nh_oif
= cfg
->fc_oif
;
801 nh
->nh_gw
= cfg
->fc_gw
;
802 nh
->nh_flags
= cfg
->fc_flags
;
803 #ifdef CONFIG_IP_ROUTE_CLASSID
804 nh
->nh_tclassid
= cfg
->fc_flow
;
806 #ifdef CONFIG_IP_ROUTE_MULTIPATH
811 if (fib_props
[cfg
->fc_type
].error
) {
812 if (cfg
->fc_gw
|| cfg
->fc_oif
|| cfg
->fc_mp
)
817 if (cfg
->fc_scope
> RT_SCOPE_HOST
)
820 if (cfg
->fc_scope
== RT_SCOPE_HOST
) {
821 struct fib_nh
*nh
= fi
->fib_nh
;
823 /* Local address is added. */
824 if (nhs
!= 1 || nh
->nh_gw
)
826 nh
->nh_scope
= RT_SCOPE_NOWHERE
;
827 nh
->nh_dev
= dev_get_by_index(net
, fi
->fib_nh
->nh_oif
);
829 if (nh
->nh_dev
== NULL
)
832 change_nexthops(fi
) {
833 err
= fib_check_nh(cfg
, fi
, nexthop_nh
);
836 } endfor_nexthops(fi
)
839 if (fi
->fib_prefsrc
) {
840 if (cfg
->fc_type
!= RTN_LOCAL
|| !cfg
->fc_dst
||
841 fi
->fib_prefsrc
!= cfg
->fc_dst
)
842 if (inet_addr_type(net
, fi
->fib_prefsrc
) != RTN_LOCAL
)
847 ofi
= fib_find_info(fi
);
856 atomic_inc(&fi
->fib_clntref
);
857 spin_lock_bh(&fib_info_lock
);
858 hlist_add_head(&fi
->fib_hash
,
859 &fib_info_hash
[fib_info_hashfn(fi
)]);
860 if (fi
->fib_prefsrc
) {
861 struct hlist_head
*head
;
863 head
= &fib_info_laddrhash
[fib_laddr_hashfn(fi
->fib_prefsrc
)];
864 hlist_add_head(&fi
->fib_lhash
, head
);
866 change_nexthops(fi
) {
867 struct hlist_head
*head
;
870 if (!nexthop_nh
->nh_dev
)
872 hash
= fib_devindex_hashfn(nexthop_nh
->nh_dev
->ifindex
);
873 head
= &fib_info_devhash
[hash
];
874 hlist_add_head(&nexthop_nh
->nh_hash
, head
);
875 } endfor_nexthops(fi
)
876 spin_unlock_bh(&fib_info_lock
);
891 /* Note! fib_semantic_match intentionally uses RCU list functions. */
892 int fib_semantic_match(struct fib_table
*tb
, struct list_head
*head
,
893 const struct flowi
*flp
, struct fib_result
*res
,
894 int prefixlen
, int fib_flags
)
896 struct fib_alias
*fa
;
899 list_for_each_entry_rcu(fa
, head
, fa_list
) {
903 fa
->fa_tos
!= flp
->fl4_tos
)
906 if (fa
->fa_scope
< flp
->fl4_scope
)
909 fib_alias_accessed(fa
);
911 err
= fib_props
[fa
->fa_type
].error
;
913 struct fib_info
*fi
= fa
->fa_info
;
915 if (fi
->fib_flags
& RTNH_F_DEAD
)
918 switch (fa
->fa_type
) {
925 if (nh
->nh_flags
& RTNH_F_DEAD
)
927 if (!flp
->oif
|| flp
->oif
== nh
->nh_oif
)
930 #ifdef CONFIG_IP_ROUTE_MULTIPATH
931 if (nhsel
< fi
->fib_nhs
) {
943 pr_warning("fib_semantic_match bad type %#x\n",
953 res
->prefixlen
= prefixlen
;
954 res
->nh_sel
= nh_sel
;
955 res
->type
= fa
->fa_type
;
956 res
->scope
= fa
->fa_scope
;
957 res
->fi
= fa
->fa_info
;
960 if (!(fib_flags
& FIB_LOOKUP_NOREF
))
961 atomic_inc(&res
->fi
->fib_clntref
);
965 /* Find appropriate source address to this destination */
967 __be32
__fib_res_prefsrc(struct fib_result
*res
)
969 return inet_select_addr(FIB_RES_DEV(*res
), FIB_RES_GW(*res
), res
->scope
);
972 int fib_dump_info(struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
973 u32 tb_id
, u8 type
, u8 scope
, __be32 dst
, int dst_len
, u8 tos
,
974 struct fib_info
*fi
, unsigned int flags
)
976 struct nlmsghdr
*nlh
;
979 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*rtm
), flags
);
983 rtm
= nlmsg_data(nlh
);
984 rtm
->rtm_family
= AF_INET
;
985 rtm
->rtm_dst_len
= dst_len
;
986 rtm
->rtm_src_len
= 0;
989 rtm
->rtm_table
= tb_id
;
991 rtm
->rtm_table
= RT_TABLE_COMPAT
;
992 NLA_PUT_U32(skb
, RTA_TABLE
, tb_id
);
993 rtm
->rtm_type
= type
;
994 rtm
->rtm_flags
= fi
->fib_flags
;
995 rtm
->rtm_scope
= scope
;
996 rtm
->rtm_protocol
= fi
->fib_protocol
;
998 if (rtm
->rtm_dst_len
)
999 NLA_PUT_BE32(skb
, RTA_DST
, dst
);
1001 if (fi
->fib_priority
)
1002 NLA_PUT_U32(skb
, RTA_PRIORITY
, fi
->fib_priority
);
1004 if (rtnetlink_put_metrics(skb
, fi
->fib_metrics
) < 0)
1005 goto nla_put_failure
;
1007 if (fi
->fib_prefsrc
)
1008 NLA_PUT_BE32(skb
, RTA_PREFSRC
, fi
->fib_prefsrc
);
1010 if (fi
->fib_nhs
== 1) {
1011 if (fi
->fib_nh
->nh_gw
)
1012 NLA_PUT_BE32(skb
, RTA_GATEWAY
, fi
->fib_nh
->nh_gw
);
1014 if (fi
->fib_nh
->nh_oif
)
1015 NLA_PUT_U32(skb
, RTA_OIF
, fi
->fib_nh
->nh_oif
);
1016 #ifdef CONFIG_IP_ROUTE_CLASSID
1017 if (fi
->fib_nh
[0].nh_tclassid
)
1018 NLA_PUT_U32(skb
, RTA_FLOW
, fi
->fib_nh
[0].nh_tclassid
);
1021 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1022 if (fi
->fib_nhs
> 1) {
1023 struct rtnexthop
*rtnh
;
1026 mp
= nla_nest_start(skb
, RTA_MULTIPATH
);
1028 goto nla_put_failure
;
1031 rtnh
= nla_reserve_nohdr(skb
, sizeof(*rtnh
));
1033 goto nla_put_failure
;
1035 rtnh
->rtnh_flags
= nh
->nh_flags
& 0xFF;
1036 rtnh
->rtnh_hops
= nh
->nh_weight
- 1;
1037 rtnh
->rtnh_ifindex
= nh
->nh_oif
;
1040 NLA_PUT_BE32(skb
, RTA_GATEWAY
, nh
->nh_gw
);
1041 #ifdef CONFIG_IP_ROUTE_CLASSID
1042 if (nh
->nh_tclassid
)
1043 NLA_PUT_U32(skb
, RTA_FLOW
, nh
->nh_tclassid
);
1045 /* length of rtnetlink header + attributes */
1046 rtnh
->rtnh_len
= nlmsg_get_pos(skb
) - (void *) rtnh
;
1047 } endfor_nexthops(fi
);
1049 nla_nest_end(skb
, mp
);
1052 return nlmsg_end(skb
, nlh
);
1055 nlmsg_cancel(skb
, nlh
);
1061 * - local address disappeared -> we must delete all the entries
1063 * - device went down -> we must shutdown all nexthops going via it.
1065 int fib_sync_down_addr(struct net
*net
, __be32 local
)
1068 unsigned int hash
= fib_laddr_hashfn(local
);
1069 struct hlist_head
*head
= &fib_info_laddrhash
[hash
];
1070 struct hlist_node
*node
;
1071 struct fib_info
*fi
;
1073 if (fib_info_laddrhash
== NULL
|| local
== 0)
1076 hlist_for_each_entry(fi
, node
, head
, fib_lhash
) {
1077 if (!net_eq(fi
->fib_net
, net
))
1079 if (fi
->fib_prefsrc
== local
) {
1080 fi
->fib_flags
|= RTNH_F_DEAD
;
1087 int fib_sync_down_dev(struct net_device
*dev
, int force
)
1090 int scope
= RT_SCOPE_NOWHERE
;
1091 struct fib_info
*prev_fi
= NULL
;
1092 unsigned int hash
= fib_devindex_hashfn(dev
->ifindex
);
1093 struct hlist_head
*head
= &fib_info_devhash
[hash
];
1094 struct hlist_node
*node
;
1100 hlist_for_each_entry(nh
, node
, head
, nh_hash
) {
1101 struct fib_info
*fi
= nh
->nh_parent
;
1104 BUG_ON(!fi
->fib_nhs
);
1105 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1109 change_nexthops(fi
) {
1110 if (nexthop_nh
->nh_flags
& RTNH_F_DEAD
)
1112 else if (nexthop_nh
->nh_dev
== dev
&&
1113 nexthop_nh
->nh_scope
!= scope
) {
1114 nexthop_nh
->nh_flags
|= RTNH_F_DEAD
;
1115 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1116 spin_lock_bh(&fib_multipath_lock
);
1117 fi
->fib_power
-= nexthop_nh
->nh_power
;
1118 nexthop_nh
->nh_power
= 0;
1119 spin_unlock_bh(&fib_multipath_lock
);
1123 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1124 if (force
> 1 && nexthop_nh
->nh_dev
== dev
) {
1129 } endfor_nexthops(fi
)
1130 if (dead
== fi
->fib_nhs
) {
1131 fi
->fib_flags
|= RTNH_F_DEAD
;
1139 /* Must be invoked inside of an RCU protected region. */
1140 void fib_select_default(struct fib_result
*res
)
1142 struct fib_info
*fi
= NULL
, *last_resort
= NULL
;
1143 struct list_head
*fa_head
= res
->fa_head
;
1144 struct fib_table
*tb
= res
->table
;
1145 int order
= -1, last_idx
= -1;
1146 struct fib_alias
*fa
;
1148 list_for_each_entry_rcu(fa
, fa_head
, fa_list
) {
1149 struct fib_info
*next_fi
= fa
->fa_info
;
1151 if (fa
->fa_scope
!= res
->scope
||
1152 fa
->fa_type
!= RTN_UNICAST
)
1155 if (next_fi
->fib_priority
> res
->fi
->fib_priority
)
1157 if (!next_fi
->fib_nh
[0].nh_gw
||
1158 next_fi
->fib_nh
[0].nh_scope
!= RT_SCOPE_LINK
)
1161 fib_alias_accessed(fa
);
1164 if (next_fi
!= res
->fi
)
1166 } else if (!fib_detect_death(fi
, order
, &last_resort
,
1167 &last_idx
, tb
->tb_default
)) {
1168 fib_result_assign(res
, fi
);
1169 tb
->tb_default
= order
;
1176 if (order
<= 0 || fi
== NULL
) {
1177 tb
->tb_default
= -1;
1181 if (!fib_detect_death(fi
, order
, &last_resort
, &last_idx
,
1183 fib_result_assign(res
, fi
);
1184 tb
->tb_default
= order
;
1189 fib_result_assign(res
, last_resort
);
1190 tb
->tb_default
= last_idx
;
1195 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1198 * Dead device goes up. We wake up dead nexthops.
1199 * It takes sense only on multipath routes.
1201 int fib_sync_up(struct net_device
*dev
)
1203 struct fib_info
*prev_fi
;
1205 struct hlist_head
*head
;
1206 struct hlist_node
*node
;
1210 if (!(dev
->flags
& IFF_UP
))
1214 hash
= fib_devindex_hashfn(dev
->ifindex
);
1215 head
= &fib_info_devhash
[hash
];
1218 hlist_for_each_entry(nh
, node
, head
, nh_hash
) {
1219 struct fib_info
*fi
= nh
->nh_parent
;
1222 BUG_ON(!fi
->fib_nhs
);
1223 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1228 change_nexthops(fi
) {
1229 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1233 if (nexthop_nh
->nh_dev
== NULL
||
1234 !(nexthop_nh
->nh_dev
->flags
& IFF_UP
))
1236 if (nexthop_nh
->nh_dev
!= dev
||
1237 !__in_dev_get_rtnl(dev
))
1240 spin_lock_bh(&fib_multipath_lock
);
1241 nexthop_nh
->nh_power
= 0;
1242 nexthop_nh
->nh_flags
&= ~RTNH_F_DEAD
;
1243 spin_unlock_bh(&fib_multipath_lock
);
1244 } endfor_nexthops(fi
)
1247 fi
->fib_flags
&= ~RTNH_F_DEAD
;
1256 * The algorithm is suboptimal, but it provides really
1257 * fair weighted route distribution.
1259 void fib_select_multipath(const struct flowi
*flp
, struct fib_result
*res
)
1261 struct fib_info
*fi
= res
->fi
;
1264 spin_lock_bh(&fib_multipath_lock
);
1265 if (fi
->fib_power
<= 0) {
1267 change_nexthops(fi
) {
1268 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1269 power
+= nexthop_nh
->nh_weight
;
1270 nexthop_nh
->nh_power
= nexthop_nh
->nh_weight
;
1272 } endfor_nexthops(fi
);
1273 fi
->fib_power
= power
;
1275 spin_unlock_bh(&fib_multipath_lock
);
1276 /* Race condition: route has just become dead. */
1283 /* w should be random number [0..fi->fib_power-1],
1284 * it is pretty bad approximation.
1287 w
= jiffies
% fi
->fib_power
;
1289 change_nexthops(fi
) {
1290 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
) &&
1291 nexthop_nh
->nh_power
) {
1292 w
-= nexthop_nh
->nh_power
;
1294 nexthop_nh
->nh_power
--;
1296 res
->nh_sel
= nhsel
;
1297 spin_unlock_bh(&fib_multipath_lock
);
1301 } endfor_nexthops(fi
);
1303 /* Race condition: route has just become dead. */
1305 spin_unlock_bh(&fib_multipath_lock
);