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 <linux/bitops.h>
18 #include <linux/types.h>
19 #include <linux/kernel.h>
20 #include <linux/jiffies.h>
22 #include <linux/string.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/errno.h>
27 #include <linux/inet.h>
28 #include <linux/inetdevice.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_arp.h>
31 #include <linux/proc_fs.h>
32 #include <linux/skbuff.h>
33 #include <linux/init.h>
34 #include <linux/slab.h>
38 #include <net/protocol.h>
39 #include <net/route.h>
42 #include <net/ip_fib.h>
43 #include <net/netlink.h>
44 #include <net/nexthop.h>
46 #include "fib_lookup.h"
48 static DEFINE_SPINLOCK(fib_info_lock
);
49 static struct hlist_head
*fib_info_hash
;
50 static struct hlist_head
*fib_info_laddrhash
;
51 static unsigned int fib_info_hash_size
;
52 static unsigned int fib_info_cnt
;
54 #define DEVINDEX_HASHBITS 8
55 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
56 static struct hlist_head fib_info_devhash
[DEVINDEX_HASHSIZE
];
58 #ifdef CONFIG_IP_ROUTE_MULTIPATH
60 static DEFINE_SPINLOCK(fib_multipath_lock
);
62 #define for_nexthops(fi) { \
63 int nhsel; const struct fib_nh *nh; \
64 for (nhsel = 0, nh = (fi)->fib_nh; \
65 nhsel < (fi)->fib_nhs; \
68 #define change_nexthops(fi) { \
69 int nhsel; struct fib_nh *nexthop_nh; \
70 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
71 nhsel < (fi)->fib_nhs; \
72 nexthop_nh++, nhsel++)
74 #else /* CONFIG_IP_ROUTE_MULTIPATH */
76 /* Hope, that gcc will optimize it to get rid of dummy loop */
78 #define for_nexthops(fi) { \
79 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
80 for (nhsel = 0; nhsel < 1; nhsel++)
82 #define change_nexthops(fi) { \
84 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
85 for (nhsel = 0; nhsel < 1; nhsel++)
87 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
89 #define endfor_nexthops(fi) }
92 const struct fib_prop fib_props
[RTN_MAX
+ 1] = {
95 .scope
= RT_SCOPE_NOWHERE
,
99 .scope
= RT_SCOPE_UNIVERSE
,
103 .scope
= RT_SCOPE_HOST
,
107 .scope
= RT_SCOPE_LINK
,
111 .scope
= RT_SCOPE_LINK
,
115 .scope
= RT_SCOPE_UNIVERSE
,
119 .scope
= RT_SCOPE_UNIVERSE
,
121 [RTN_UNREACHABLE
] = {
122 .error
= -EHOSTUNREACH
,
123 .scope
= RT_SCOPE_UNIVERSE
,
127 .scope
= RT_SCOPE_UNIVERSE
,
131 .scope
= RT_SCOPE_UNIVERSE
,
135 .scope
= RT_SCOPE_NOWHERE
,
139 .scope
= RT_SCOPE_NOWHERE
,
143 /* Release a nexthop info record */
144 static void free_fib_info_rcu(struct rcu_head
*head
)
146 struct fib_info
*fi
= container_of(head
, struct fib_info
, rcu
);
148 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
)
149 kfree(fi
->fib_metrics
);
153 void free_fib_info(struct fib_info
*fi
)
155 if (fi
->fib_dead
== 0) {
156 pr_warn("Freeing alive fib_info %p\n", fi
);
159 change_nexthops(fi
) {
160 if (nexthop_nh
->nh_dev
)
161 dev_put(nexthop_nh
->nh_dev
);
162 nexthop_nh
->nh_dev
= NULL
;
163 } endfor_nexthops(fi
);
165 release_net(fi
->fib_net
);
166 call_rcu(&fi
->rcu
, free_fib_info_rcu
);
169 void fib_release_info(struct fib_info
*fi
)
171 spin_lock_bh(&fib_info_lock
);
172 if (fi
&& --fi
->fib_treeref
== 0) {
173 hlist_del(&fi
->fib_hash
);
175 hlist_del(&fi
->fib_lhash
);
176 change_nexthops(fi
) {
177 if (!nexthop_nh
->nh_dev
)
179 hlist_del(&nexthop_nh
->nh_hash
);
180 } endfor_nexthops(fi
)
184 spin_unlock_bh(&fib_info_lock
);
187 static inline int nh_comp(const struct fib_info
*fi
, const struct fib_info
*ofi
)
189 const struct fib_nh
*onh
= ofi
->fib_nh
;
192 if (nh
->nh_oif
!= onh
->nh_oif
||
193 nh
->nh_gw
!= onh
->nh_gw
||
194 nh
->nh_scope
!= onh
->nh_scope
||
195 #ifdef CONFIG_IP_ROUTE_MULTIPATH
196 nh
->nh_weight
!= onh
->nh_weight
||
198 #ifdef CONFIG_IP_ROUTE_CLASSID
199 nh
->nh_tclassid
!= onh
->nh_tclassid
||
201 ((nh
->nh_flags
^ onh
->nh_flags
) & ~RTNH_F_DEAD
))
204 } endfor_nexthops(fi
);
208 static inline unsigned int fib_devindex_hashfn(unsigned int val
)
210 unsigned int mask
= DEVINDEX_HASHSIZE
- 1;
213 (val
>> DEVINDEX_HASHBITS
) ^
214 (val
>> (DEVINDEX_HASHBITS
* 2))) & mask
;
217 static inline unsigned int fib_info_hashfn(const struct fib_info
*fi
)
219 unsigned int mask
= (fib_info_hash_size
- 1);
220 unsigned int val
= fi
->fib_nhs
;
222 val
^= (fi
->fib_protocol
<< 8) | fi
->fib_scope
;
223 val
^= (__force u32
)fi
->fib_prefsrc
;
224 val
^= fi
->fib_priority
;
226 val
^= fib_devindex_hashfn(nh
->nh_oif
);
227 } endfor_nexthops(fi
)
229 return (val
^ (val
>> 7) ^ (val
>> 12)) & mask
;
232 static struct fib_info
*fib_find_info(const struct fib_info
*nfi
)
234 struct hlist_head
*head
;
235 struct hlist_node
*node
;
239 hash
= fib_info_hashfn(nfi
);
240 head
= &fib_info_hash
[hash
];
242 hlist_for_each_entry(fi
, node
, head
, fib_hash
) {
243 if (!net_eq(fi
->fib_net
, nfi
->fib_net
))
245 if (fi
->fib_nhs
!= nfi
->fib_nhs
)
247 if (nfi
->fib_protocol
== fi
->fib_protocol
&&
248 nfi
->fib_scope
== fi
->fib_scope
&&
249 nfi
->fib_prefsrc
== fi
->fib_prefsrc
&&
250 nfi
->fib_priority
== fi
->fib_priority
&&
251 memcmp(nfi
->fib_metrics
, fi
->fib_metrics
,
252 sizeof(u32
) * RTAX_MAX
) == 0 &&
253 ((nfi
->fib_flags
^ fi
->fib_flags
) & ~RTNH_F_DEAD
) == 0 &&
254 (nfi
->fib_nhs
== 0 || nh_comp(fi
, nfi
) == 0))
261 /* Check, that the gateway is already configured.
262 * Used only by redirect accept routine.
264 int ip_fib_check_default(__be32 gw
, struct net_device
*dev
)
266 struct hlist_head
*head
;
267 struct hlist_node
*node
;
271 spin_lock(&fib_info_lock
);
273 hash
= fib_devindex_hashfn(dev
->ifindex
);
274 head
= &fib_info_devhash
[hash
];
275 hlist_for_each_entry(nh
, node
, head
, nh_hash
) {
276 if (nh
->nh_dev
== dev
&&
278 !(nh
->nh_flags
& RTNH_F_DEAD
)) {
279 spin_unlock(&fib_info_lock
);
284 spin_unlock(&fib_info_lock
);
289 static inline size_t fib_nlmsg_size(struct fib_info
*fi
)
291 size_t payload
= NLMSG_ALIGN(sizeof(struct rtmsg
))
292 + nla_total_size(4) /* RTA_TABLE */
293 + nla_total_size(4) /* RTA_DST */
294 + nla_total_size(4) /* RTA_PRIORITY */
295 + nla_total_size(4); /* RTA_PREFSRC */
297 /* space for nested metrics */
298 payload
+= nla_total_size((RTAX_MAX
* nla_total_size(4)));
301 /* Also handles the special case fib_nhs == 1 */
303 /* each nexthop is packed in an attribute */
304 size_t nhsize
= nla_total_size(sizeof(struct rtnexthop
));
306 /* may contain flow and gateway attribute */
307 nhsize
+= 2 * nla_total_size(4);
309 /* all nexthops are packed in a nested attribute */
310 payload
+= nla_total_size(fi
->fib_nhs
* nhsize
);
316 void rtmsg_fib(int event
, __be32 key
, struct fib_alias
*fa
,
317 int dst_len
, u32 tb_id
, struct nl_info
*info
,
318 unsigned int nlm_flags
)
321 u32 seq
= info
->nlh
? info
->nlh
->nlmsg_seq
: 0;
324 skb
= nlmsg_new(fib_nlmsg_size(fa
->fa_info
), GFP_KERNEL
);
328 err
= fib_dump_info(skb
, info
->pid
, seq
, event
, tb_id
,
329 fa
->fa_type
, key
, dst_len
,
330 fa
->fa_tos
, fa
->fa_info
, nlm_flags
);
332 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
333 WARN_ON(err
== -EMSGSIZE
);
337 rtnl_notify(skb
, info
->nl_net
, info
->pid
, RTNLGRP_IPV4_ROUTE
,
338 info
->nlh
, GFP_KERNEL
);
342 rtnl_set_sk_err(info
->nl_net
, RTNLGRP_IPV4_ROUTE
, err
);
345 /* Return the first fib alias matching TOS with
346 * priority less than or equal to PRIO.
348 struct fib_alias
*fib_find_alias(struct list_head
*fah
, u8 tos
, u32 prio
)
351 struct fib_alias
*fa
;
352 list_for_each_entry(fa
, fah
, fa_list
) {
353 if (fa
->fa_tos
> tos
)
355 if (fa
->fa_info
->fib_priority
>= prio
||
363 int fib_detect_death(struct fib_info
*fi
, int order
,
364 struct fib_info
**last_resort
, int *last_idx
, int dflt
)
367 int state
= NUD_NONE
;
369 n
= neigh_lookup(&arp_tbl
, &fi
->fib_nh
[0].nh_gw
, fi
->fib_dev
);
371 state
= n
->nud_state
;
374 if (state
== NUD_REACHABLE
)
376 if ((state
& NUD_VALID
) && order
!= dflt
)
378 if ((state
& NUD_VALID
) ||
379 (*last_idx
< 0 && order
> dflt
)) {
386 #ifdef CONFIG_IP_ROUTE_MULTIPATH
388 static int fib_count_nexthops(struct rtnexthop
*rtnh
, int remaining
)
392 while (rtnh_ok(rtnh
, remaining
)) {
394 rtnh
= rtnh_next(rtnh
, &remaining
);
397 /* leftover implies invalid nexthop configuration, discard it */
398 return remaining
> 0 ? 0 : nhs
;
401 static int fib_get_nhs(struct fib_info
*fi
, struct rtnexthop
*rtnh
,
402 int remaining
, struct fib_config
*cfg
)
404 change_nexthops(fi
) {
407 if (!rtnh_ok(rtnh
, remaining
))
410 nexthop_nh
->nh_flags
=
411 (cfg
->fc_flags
& ~0xFF) | rtnh
->rtnh_flags
;
412 nexthop_nh
->nh_oif
= rtnh
->rtnh_ifindex
;
413 nexthop_nh
->nh_weight
= rtnh
->rtnh_hops
+ 1;
415 attrlen
= rtnh_attrlen(rtnh
);
417 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
419 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
420 nexthop_nh
->nh_gw
= nla
? nla_get_be32(nla
) : 0;
421 #ifdef CONFIG_IP_ROUTE_CLASSID
422 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
423 nexthop_nh
->nh_tclassid
= nla
? nla_get_u32(nla
) : 0;
427 rtnh
= rtnh_next(rtnh
, &remaining
);
428 } endfor_nexthops(fi
);
435 int fib_nh_match(struct fib_config
*cfg
, struct fib_info
*fi
)
437 #ifdef CONFIG_IP_ROUTE_MULTIPATH
438 struct rtnexthop
*rtnh
;
442 if (cfg
->fc_priority
&& cfg
->fc_priority
!= fi
->fib_priority
)
445 if (cfg
->fc_oif
|| cfg
->fc_gw
) {
446 if ((!cfg
->fc_oif
|| cfg
->fc_oif
== fi
->fib_nh
->nh_oif
) &&
447 (!cfg
->fc_gw
|| cfg
->fc_gw
== fi
->fib_nh
->nh_gw
))
452 #ifdef CONFIG_IP_ROUTE_MULTIPATH
453 if (cfg
->fc_mp
== NULL
)
457 remaining
= cfg
->fc_mp_len
;
462 if (!rtnh_ok(rtnh
, remaining
))
465 if (rtnh
->rtnh_ifindex
&& rtnh
->rtnh_ifindex
!= nh
->nh_oif
)
468 attrlen
= rtnh_attrlen(rtnh
);
470 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
472 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
473 if (nla
&& nla_get_be32(nla
) != nh
->nh_gw
)
475 #ifdef CONFIG_IP_ROUTE_CLASSID
476 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
477 if (nla
&& nla_get_u32(nla
) != nh
->nh_tclassid
)
482 rtnh
= rtnh_next(rtnh
, &remaining
);
483 } endfor_nexthops(fi
);
493 * Semantics of nexthop is very messy by historical reasons.
494 * We have to take into account, that:
495 * a) gateway can be actually local interface address,
496 * so that gatewayed route is direct.
497 * b) gateway must be on-link address, possibly
498 * described not by an ifaddr, but also by a direct route.
499 * c) If both gateway and interface are specified, they should not
501 * d) If we use tunnel routes, gateway could be not on-link.
503 * Attempt to reconcile all of these (alas, self-contradictory) conditions
504 * results in pretty ugly and hairy code with obscure logic.
506 * I chose to generalized it instead, so that the size
507 * of code does not increase practically, but it becomes
509 * Every prefix is assigned a "scope" value: "host" is local address,
510 * "link" is direct route,
511 * [ ... "site" ... "interior" ... ]
512 * and "universe" is true gateway route with global meaning.
514 * Every prefix refers to a set of "nexthop"s (gw, oif),
515 * where gw must have narrower scope. This recursion stops
516 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
517 * which means that gw is forced to be on link.
519 * Code is still hairy, but now it is apparently logically
520 * consistent and very flexible. F.e. as by-product it allows
521 * to co-exists in peace independent exterior and interior
524 * Normally it looks as following.
526 * {universe prefix} -> (gw, oif) [scope link]
528 * |-> {link prefix} -> (gw, oif) [scope local]
530 * |-> {local prefix} (terminal node)
532 static int fib_check_nh(struct fib_config
*cfg
, struct fib_info
*fi
,
537 struct net_device
*dev
;
539 net
= cfg
->fc_nlinfo
.nl_net
;
541 struct fib_result res
;
543 if (nh
->nh_flags
& RTNH_F_ONLINK
) {
545 if (cfg
->fc_scope
>= RT_SCOPE_LINK
)
547 if (inet_addr_type(net
, nh
->nh_gw
) != RTN_UNICAST
)
549 dev
= __dev_get_by_index(net
, nh
->nh_oif
);
552 if (!(dev
->flags
& IFF_UP
))
556 nh
->nh_scope
= RT_SCOPE_LINK
;
561 struct flowi4 fl4
= {
563 .flowi4_scope
= cfg
->fc_scope
+ 1,
564 .flowi4_oif
= nh
->nh_oif
,
567 /* It is not necessary, but requires a bit of thinking */
568 if (fl4
.flowi4_scope
< RT_SCOPE_LINK
)
569 fl4
.flowi4_scope
= RT_SCOPE_LINK
;
570 err
= fib_lookup(net
, &fl4
, &res
);
577 if (res
.type
!= RTN_UNICAST
&& res
.type
!= RTN_LOCAL
)
579 nh
->nh_scope
= res
.scope
;
580 nh
->nh_oif
= FIB_RES_OIF(res
);
581 nh
->nh_dev
= dev
= FIB_RES_DEV(res
);
585 err
= (dev
->flags
& IFF_UP
) ? 0 : -ENETDOWN
;
587 struct in_device
*in_dev
;
589 if (nh
->nh_flags
& (RTNH_F_PERVASIVE
| RTNH_F_ONLINK
))
594 in_dev
= inetdev_by_index(net
, nh
->nh_oif
);
598 if (!(in_dev
->dev
->flags
& IFF_UP
))
600 nh
->nh_dev
= in_dev
->dev
;
601 dev_hold(nh
->nh_dev
);
602 nh
->nh_scope
= RT_SCOPE_HOST
;
610 static inline unsigned int fib_laddr_hashfn(__be32 val
)
612 unsigned int mask
= (fib_info_hash_size
- 1);
614 return ((__force u32
)val
^
615 ((__force u32
)val
>> 7) ^
616 ((__force u32
)val
>> 14)) & mask
;
619 static struct hlist_head
*fib_info_hash_alloc(int bytes
)
621 if (bytes
<= PAGE_SIZE
)
622 return kzalloc(bytes
, GFP_KERNEL
);
624 return (struct hlist_head
*)
625 __get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
629 static void fib_info_hash_free(struct hlist_head
*hash
, int bytes
)
634 if (bytes
<= PAGE_SIZE
)
637 free_pages((unsigned long) hash
, get_order(bytes
));
640 static void fib_info_hash_move(struct hlist_head
*new_info_hash
,
641 struct hlist_head
*new_laddrhash
,
642 unsigned int new_size
)
644 struct hlist_head
*old_info_hash
, *old_laddrhash
;
645 unsigned int old_size
= fib_info_hash_size
;
646 unsigned int i
, bytes
;
648 spin_lock_bh(&fib_info_lock
);
649 old_info_hash
= fib_info_hash
;
650 old_laddrhash
= fib_info_laddrhash
;
651 fib_info_hash_size
= new_size
;
653 for (i
= 0; i
< old_size
; i
++) {
654 struct hlist_head
*head
= &fib_info_hash
[i
];
655 struct hlist_node
*node
, *n
;
658 hlist_for_each_entry_safe(fi
, node
, n
, head
, fib_hash
) {
659 struct hlist_head
*dest
;
660 unsigned int new_hash
;
662 hlist_del(&fi
->fib_hash
);
664 new_hash
= fib_info_hashfn(fi
);
665 dest
= &new_info_hash
[new_hash
];
666 hlist_add_head(&fi
->fib_hash
, dest
);
669 fib_info_hash
= new_info_hash
;
671 for (i
= 0; i
< old_size
; i
++) {
672 struct hlist_head
*lhead
= &fib_info_laddrhash
[i
];
673 struct hlist_node
*node
, *n
;
676 hlist_for_each_entry_safe(fi
, node
, n
, lhead
, fib_lhash
) {
677 struct hlist_head
*ldest
;
678 unsigned int new_hash
;
680 hlist_del(&fi
->fib_lhash
);
682 new_hash
= fib_laddr_hashfn(fi
->fib_prefsrc
);
683 ldest
= &new_laddrhash
[new_hash
];
684 hlist_add_head(&fi
->fib_lhash
, ldest
);
687 fib_info_laddrhash
= new_laddrhash
;
689 spin_unlock_bh(&fib_info_lock
);
691 bytes
= old_size
* sizeof(struct hlist_head
*);
692 fib_info_hash_free(old_info_hash
, bytes
);
693 fib_info_hash_free(old_laddrhash
, bytes
);
696 __be32
fib_info_update_nh_saddr(struct net
*net
, struct fib_nh
*nh
)
698 nh
->nh_saddr
= inet_select_addr(nh
->nh_dev
,
700 nh
->nh_parent
->fib_scope
);
701 nh
->nh_saddr_genid
= atomic_read(&net
->ipv4
.dev_addr_genid
);
706 struct fib_info
*fib_create_info(struct fib_config
*cfg
)
709 struct fib_info
*fi
= NULL
;
710 struct fib_info
*ofi
;
712 struct net
*net
= cfg
->fc_nlinfo
.nl_net
;
714 if (cfg
->fc_type
> RTN_MAX
)
717 /* Fast check to catch the most weird cases */
718 if (fib_props
[cfg
->fc_type
].scope
> cfg
->fc_scope
)
721 #ifdef CONFIG_IP_ROUTE_MULTIPATH
723 nhs
= fib_count_nexthops(cfg
->fc_mp
, cfg
->fc_mp_len
);
730 if (fib_info_cnt
>= fib_info_hash_size
) {
731 unsigned int new_size
= fib_info_hash_size
<< 1;
732 struct hlist_head
*new_info_hash
;
733 struct hlist_head
*new_laddrhash
;
738 bytes
= new_size
* sizeof(struct hlist_head
*);
739 new_info_hash
= fib_info_hash_alloc(bytes
);
740 new_laddrhash
= fib_info_hash_alloc(bytes
);
741 if (!new_info_hash
|| !new_laddrhash
) {
742 fib_info_hash_free(new_info_hash
, bytes
);
743 fib_info_hash_free(new_laddrhash
, bytes
);
745 fib_info_hash_move(new_info_hash
, new_laddrhash
, new_size
);
747 if (!fib_info_hash_size
)
751 fi
= kzalloc(sizeof(*fi
)+nhs
*sizeof(struct fib_nh
), GFP_KERNEL
);
755 fi
->fib_metrics
= kzalloc(sizeof(u32
) * RTAX_MAX
, GFP_KERNEL
);
756 if (!fi
->fib_metrics
)
759 fi
->fib_metrics
= (u32
*) dst_default_metrics
;
762 fi
->fib_net
= hold_net(net
);
763 fi
->fib_protocol
= cfg
->fc_protocol
;
764 fi
->fib_scope
= cfg
->fc_scope
;
765 fi
->fib_flags
= cfg
->fc_flags
;
766 fi
->fib_priority
= cfg
->fc_priority
;
767 fi
->fib_prefsrc
= cfg
->fc_prefsrc
;
770 change_nexthops(fi
) {
771 nexthop_nh
->nh_parent
= fi
;
772 } endfor_nexthops(fi
)
778 nla_for_each_attr(nla
, cfg
->fc_mx
, cfg
->fc_mx_len
, remaining
) {
779 int type
= nla_type(nla
);
784 fi
->fib_metrics
[type
- 1] = nla_get_u32(nla
);
790 #ifdef CONFIG_IP_ROUTE_MULTIPATH
791 err
= fib_get_nhs(fi
, cfg
->fc_mp
, cfg
->fc_mp_len
, cfg
);
794 if (cfg
->fc_oif
&& fi
->fib_nh
->nh_oif
!= cfg
->fc_oif
)
796 if (cfg
->fc_gw
&& fi
->fib_nh
->nh_gw
!= cfg
->fc_gw
)
798 #ifdef CONFIG_IP_ROUTE_CLASSID
799 if (cfg
->fc_flow
&& fi
->fib_nh
->nh_tclassid
!= cfg
->fc_flow
)
806 struct fib_nh
*nh
= fi
->fib_nh
;
808 nh
->nh_oif
= cfg
->fc_oif
;
809 nh
->nh_gw
= cfg
->fc_gw
;
810 nh
->nh_flags
= cfg
->fc_flags
;
811 #ifdef CONFIG_IP_ROUTE_CLASSID
812 nh
->nh_tclassid
= cfg
->fc_flow
;
814 #ifdef CONFIG_IP_ROUTE_MULTIPATH
819 if (fib_props
[cfg
->fc_type
].error
) {
820 if (cfg
->fc_gw
|| cfg
->fc_oif
|| cfg
->fc_mp
)
824 switch (cfg
->fc_type
) {
836 if (cfg
->fc_scope
> RT_SCOPE_HOST
)
839 if (cfg
->fc_scope
== RT_SCOPE_HOST
) {
840 struct fib_nh
*nh
= fi
->fib_nh
;
842 /* Local address is added. */
843 if (nhs
!= 1 || nh
->nh_gw
)
845 nh
->nh_scope
= RT_SCOPE_NOWHERE
;
846 nh
->nh_dev
= dev_get_by_index(net
, fi
->fib_nh
->nh_oif
);
848 if (nh
->nh_dev
== NULL
)
851 change_nexthops(fi
) {
852 err
= fib_check_nh(cfg
, fi
, nexthop_nh
);
855 } endfor_nexthops(fi
)
858 if (fi
->fib_prefsrc
) {
859 if (cfg
->fc_type
!= RTN_LOCAL
|| !cfg
->fc_dst
||
860 fi
->fib_prefsrc
!= cfg
->fc_dst
)
861 if (inet_addr_type(net
, fi
->fib_prefsrc
) != RTN_LOCAL
)
865 change_nexthops(fi
) {
866 fib_info_update_nh_saddr(net
, nexthop_nh
);
867 } endfor_nexthops(fi
)
870 ofi
= fib_find_info(fi
);
879 atomic_inc(&fi
->fib_clntref
);
880 spin_lock_bh(&fib_info_lock
);
881 hlist_add_head(&fi
->fib_hash
,
882 &fib_info_hash
[fib_info_hashfn(fi
)]);
883 if (fi
->fib_prefsrc
) {
884 struct hlist_head
*head
;
886 head
= &fib_info_laddrhash
[fib_laddr_hashfn(fi
->fib_prefsrc
)];
887 hlist_add_head(&fi
->fib_lhash
, head
);
889 change_nexthops(fi
) {
890 struct hlist_head
*head
;
893 if (!nexthop_nh
->nh_dev
)
895 hash
= fib_devindex_hashfn(nexthop_nh
->nh_dev
->ifindex
);
896 head
= &fib_info_devhash
[hash
];
897 hlist_add_head(&nexthop_nh
->nh_hash
, head
);
898 } endfor_nexthops(fi
)
899 spin_unlock_bh(&fib_info_lock
);
914 int fib_dump_info(struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
915 u32 tb_id
, u8 type
, __be32 dst
, int dst_len
, u8 tos
,
916 struct fib_info
*fi
, unsigned int flags
)
918 struct nlmsghdr
*nlh
;
921 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*rtm
), flags
);
925 rtm
= nlmsg_data(nlh
);
926 rtm
->rtm_family
= AF_INET
;
927 rtm
->rtm_dst_len
= dst_len
;
928 rtm
->rtm_src_len
= 0;
931 rtm
->rtm_table
= tb_id
;
933 rtm
->rtm_table
= RT_TABLE_COMPAT
;
934 if (nla_put_u32(skb
, RTA_TABLE
, tb_id
))
935 goto nla_put_failure
;
936 rtm
->rtm_type
= type
;
937 rtm
->rtm_flags
= fi
->fib_flags
;
938 rtm
->rtm_scope
= fi
->fib_scope
;
939 rtm
->rtm_protocol
= fi
->fib_protocol
;
941 if (rtm
->rtm_dst_len
&&
942 nla_put_be32(skb
, RTA_DST
, dst
))
943 goto nla_put_failure
;
944 if (fi
->fib_priority
&&
945 nla_put_u32(skb
, RTA_PRIORITY
, fi
->fib_priority
))
946 goto nla_put_failure
;
947 if (rtnetlink_put_metrics(skb
, fi
->fib_metrics
) < 0)
948 goto nla_put_failure
;
950 if (fi
->fib_prefsrc
&&
951 nla_put_be32(skb
, RTA_PREFSRC
, fi
->fib_prefsrc
))
952 goto nla_put_failure
;
953 if (fi
->fib_nhs
== 1) {
954 if (fi
->fib_nh
->nh_gw
&&
955 nla_put_be32(skb
, RTA_GATEWAY
, fi
->fib_nh
->nh_gw
))
956 goto nla_put_failure
;
957 if (fi
->fib_nh
->nh_oif
&&
958 nla_put_u32(skb
, RTA_OIF
, fi
->fib_nh
->nh_oif
))
959 goto nla_put_failure
;
960 #ifdef CONFIG_IP_ROUTE_CLASSID
961 if (fi
->fib_nh
[0].nh_tclassid
&&
962 nla_put_u32(skb
, RTA_FLOW
, fi
->fib_nh
[0].nh_tclassid
))
963 goto nla_put_failure
;
966 #ifdef CONFIG_IP_ROUTE_MULTIPATH
967 if (fi
->fib_nhs
> 1) {
968 struct rtnexthop
*rtnh
;
971 mp
= nla_nest_start(skb
, RTA_MULTIPATH
);
973 goto nla_put_failure
;
976 rtnh
= nla_reserve_nohdr(skb
, sizeof(*rtnh
));
978 goto nla_put_failure
;
980 rtnh
->rtnh_flags
= nh
->nh_flags
& 0xFF;
981 rtnh
->rtnh_hops
= nh
->nh_weight
- 1;
982 rtnh
->rtnh_ifindex
= nh
->nh_oif
;
985 nla_put_be32(skb
, RTA_GATEWAY
, nh
->nh_gw
))
986 goto nla_put_failure
;
987 #ifdef CONFIG_IP_ROUTE_CLASSID
988 if (nh
->nh_tclassid
&&
989 nla_put_u32(skb
, RTA_FLOW
, nh
->nh_tclassid
))
990 goto nla_put_failure
;
992 /* length of rtnetlink header + attributes */
993 rtnh
->rtnh_len
= nlmsg_get_pos(skb
) - (void *) rtnh
;
994 } endfor_nexthops(fi
);
996 nla_nest_end(skb
, mp
);
999 return nlmsg_end(skb
, nlh
);
1002 nlmsg_cancel(skb
, nlh
);
1008 * - local address disappeared -> we must delete all the entries
1010 * - device went down -> we must shutdown all nexthops going via it.
1012 int fib_sync_down_addr(struct net
*net
, __be32 local
)
1015 unsigned int hash
= fib_laddr_hashfn(local
);
1016 struct hlist_head
*head
= &fib_info_laddrhash
[hash
];
1017 struct hlist_node
*node
;
1018 struct fib_info
*fi
;
1020 if (fib_info_laddrhash
== NULL
|| local
== 0)
1023 hlist_for_each_entry(fi
, node
, head
, fib_lhash
) {
1024 if (!net_eq(fi
->fib_net
, net
))
1026 if (fi
->fib_prefsrc
== local
) {
1027 fi
->fib_flags
|= RTNH_F_DEAD
;
1034 int fib_sync_down_dev(struct net_device
*dev
, int force
)
1037 int scope
= RT_SCOPE_NOWHERE
;
1038 struct fib_info
*prev_fi
= NULL
;
1039 unsigned int hash
= fib_devindex_hashfn(dev
->ifindex
);
1040 struct hlist_head
*head
= &fib_info_devhash
[hash
];
1041 struct hlist_node
*node
;
1047 hlist_for_each_entry(nh
, node
, head
, nh_hash
) {
1048 struct fib_info
*fi
= nh
->nh_parent
;
1051 BUG_ON(!fi
->fib_nhs
);
1052 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1056 change_nexthops(fi
) {
1057 if (nexthop_nh
->nh_flags
& RTNH_F_DEAD
)
1059 else if (nexthop_nh
->nh_dev
== dev
&&
1060 nexthop_nh
->nh_scope
!= scope
) {
1061 nexthop_nh
->nh_flags
|= RTNH_F_DEAD
;
1062 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1063 spin_lock_bh(&fib_multipath_lock
);
1064 fi
->fib_power
-= nexthop_nh
->nh_power
;
1065 nexthop_nh
->nh_power
= 0;
1066 spin_unlock_bh(&fib_multipath_lock
);
1070 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1071 if (force
> 1 && nexthop_nh
->nh_dev
== dev
) {
1076 } endfor_nexthops(fi
)
1077 if (dead
== fi
->fib_nhs
) {
1078 fi
->fib_flags
|= RTNH_F_DEAD
;
1086 /* Must be invoked inside of an RCU protected region. */
1087 void fib_select_default(struct fib_result
*res
)
1089 struct fib_info
*fi
= NULL
, *last_resort
= NULL
;
1090 struct list_head
*fa_head
= res
->fa_head
;
1091 struct fib_table
*tb
= res
->table
;
1092 int order
= -1, last_idx
= -1;
1093 struct fib_alias
*fa
;
1095 list_for_each_entry_rcu(fa
, fa_head
, fa_list
) {
1096 struct fib_info
*next_fi
= fa
->fa_info
;
1098 if (next_fi
->fib_scope
!= res
->scope
||
1099 fa
->fa_type
!= RTN_UNICAST
)
1102 if (next_fi
->fib_priority
> res
->fi
->fib_priority
)
1104 if (!next_fi
->fib_nh
[0].nh_gw
||
1105 next_fi
->fib_nh
[0].nh_scope
!= RT_SCOPE_LINK
)
1108 fib_alias_accessed(fa
);
1111 if (next_fi
!= res
->fi
)
1113 } else if (!fib_detect_death(fi
, order
, &last_resort
,
1114 &last_idx
, tb
->tb_default
)) {
1115 fib_result_assign(res
, fi
);
1116 tb
->tb_default
= order
;
1123 if (order
<= 0 || fi
== NULL
) {
1124 tb
->tb_default
= -1;
1128 if (!fib_detect_death(fi
, order
, &last_resort
, &last_idx
,
1130 fib_result_assign(res
, fi
);
1131 tb
->tb_default
= order
;
1136 fib_result_assign(res
, last_resort
);
1137 tb
->tb_default
= last_idx
;
1142 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1145 * Dead device goes up. We wake up dead nexthops.
1146 * It takes sense only on multipath routes.
1148 int fib_sync_up(struct net_device
*dev
)
1150 struct fib_info
*prev_fi
;
1152 struct hlist_head
*head
;
1153 struct hlist_node
*node
;
1157 if (!(dev
->flags
& IFF_UP
))
1161 hash
= fib_devindex_hashfn(dev
->ifindex
);
1162 head
= &fib_info_devhash
[hash
];
1165 hlist_for_each_entry(nh
, node
, head
, nh_hash
) {
1166 struct fib_info
*fi
= nh
->nh_parent
;
1169 BUG_ON(!fi
->fib_nhs
);
1170 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1175 change_nexthops(fi
) {
1176 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1180 if (nexthop_nh
->nh_dev
== NULL
||
1181 !(nexthop_nh
->nh_dev
->flags
& IFF_UP
))
1183 if (nexthop_nh
->nh_dev
!= dev
||
1184 !__in_dev_get_rtnl(dev
))
1187 spin_lock_bh(&fib_multipath_lock
);
1188 nexthop_nh
->nh_power
= 0;
1189 nexthop_nh
->nh_flags
&= ~RTNH_F_DEAD
;
1190 spin_unlock_bh(&fib_multipath_lock
);
1191 } endfor_nexthops(fi
)
1194 fi
->fib_flags
&= ~RTNH_F_DEAD
;
1203 * The algorithm is suboptimal, but it provides really
1204 * fair weighted route distribution.
1206 void fib_select_multipath(struct fib_result
*res
)
1208 struct fib_info
*fi
= res
->fi
;
1211 spin_lock_bh(&fib_multipath_lock
);
1212 if (fi
->fib_power
<= 0) {
1214 change_nexthops(fi
) {
1215 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1216 power
+= nexthop_nh
->nh_weight
;
1217 nexthop_nh
->nh_power
= nexthop_nh
->nh_weight
;
1219 } endfor_nexthops(fi
);
1220 fi
->fib_power
= power
;
1222 spin_unlock_bh(&fib_multipath_lock
);
1223 /* Race condition: route has just become dead. */
1230 /* w should be random number [0..fi->fib_power-1],
1231 * it is pretty bad approximation.
1234 w
= jiffies
% fi
->fib_power
;
1236 change_nexthops(fi
) {
1237 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
) &&
1238 nexthop_nh
->nh_power
) {
1239 w
-= nexthop_nh
->nh_power
;
1241 nexthop_nh
->nh_power
--;
1243 res
->nh_sel
= nhsel
;
1244 spin_unlock_bh(&fib_multipath_lock
);
1248 } endfor_nexthops(fi
);
1250 /* Race condition: route has just become dead. */
1252 spin_unlock_bh(&fib_multipath_lock
);