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 static void rt_fibinfo_free(struct rtable __rcu
**rtp
)
145 struct rtable
*rt
= rcu_dereference_protected(*rtp
, 1);
150 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
151 * because we waited an RCU grace period before calling
152 * free_fib_info_rcu()
158 static void free_nh_exceptions(struct fib_nh
*nh
)
160 struct fnhe_hash_bucket
*hash
;
163 hash
= rcu_dereference_protected(nh
->nh_exceptions
, 1);
166 for (i
= 0; i
< FNHE_HASH_SIZE
; i
++) {
167 struct fib_nh_exception
*fnhe
;
169 fnhe
= rcu_dereference_protected(hash
[i
].chain
, 1);
171 struct fib_nh_exception
*next
;
173 next
= rcu_dereference_protected(fnhe
->fnhe_next
, 1);
175 rt_fibinfo_free(&fnhe
->fnhe_rth_input
);
176 rt_fibinfo_free(&fnhe
->fnhe_rth_output
);
186 static void rt_fibinfo_free_cpus(struct rtable __rcu
* __percpu
*rtp
)
193 for_each_possible_cpu(cpu
) {
196 rt
= rcu_dereference_protected(*per_cpu_ptr(rtp
, cpu
), 1);
203 /* Release a nexthop info record */
204 static void free_fib_info_rcu(struct rcu_head
*head
)
206 struct fib_info
*fi
= container_of(head
, struct fib_info
, rcu
);
208 change_nexthops(fi
) {
209 if (nexthop_nh
->nh_dev
)
210 dev_put(nexthop_nh
->nh_dev
);
211 free_nh_exceptions(nexthop_nh
);
212 rt_fibinfo_free_cpus(nexthop_nh
->nh_pcpu_rth_output
);
213 rt_fibinfo_free(&nexthop_nh
->nh_rth_input
);
214 } endfor_nexthops(fi
);
216 release_net(fi
->fib_net
);
217 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
)
218 kfree(fi
->fib_metrics
);
222 void free_fib_info(struct fib_info
*fi
)
224 if (fi
->fib_dead
== 0) {
225 pr_warn("Freeing alive fib_info %p\n", fi
);
229 #ifdef CONFIG_IP_ROUTE_CLASSID
230 change_nexthops(fi
) {
231 if (nexthop_nh
->nh_tclassid
)
232 fi
->fib_net
->ipv4
.fib_num_tclassid_users
--;
233 } endfor_nexthops(fi
);
235 call_rcu(&fi
->rcu
, free_fib_info_rcu
);
238 void fib_release_info(struct fib_info
*fi
)
240 spin_lock_bh(&fib_info_lock
);
241 if (fi
&& --fi
->fib_treeref
== 0) {
242 hlist_del(&fi
->fib_hash
);
244 hlist_del(&fi
->fib_lhash
);
245 change_nexthops(fi
) {
246 if (!nexthop_nh
->nh_dev
)
248 hlist_del(&nexthop_nh
->nh_hash
);
249 } endfor_nexthops(fi
)
253 spin_unlock_bh(&fib_info_lock
);
256 static inline int nh_comp(const struct fib_info
*fi
, const struct fib_info
*ofi
)
258 const struct fib_nh
*onh
= ofi
->fib_nh
;
261 if (nh
->nh_oif
!= onh
->nh_oif
||
262 nh
->nh_gw
!= onh
->nh_gw
||
263 nh
->nh_scope
!= onh
->nh_scope
||
264 #ifdef CONFIG_IP_ROUTE_MULTIPATH
265 nh
->nh_weight
!= onh
->nh_weight
||
267 #ifdef CONFIG_IP_ROUTE_CLASSID
268 nh
->nh_tclassid
!= onh
->nh_tclassid
||
270 ((nh
->nh_flags
^ onh
->nh_flags
) & ~RTNH_F_DEAD
))
273 } endfor_nexthops(fi
);
277 static inline unsigned int fib_devindex_hashfn(unsigned int val
)
279 unsigned int mask
= DEVINDEX_HASHSIZE
- 1;
282 (val
>> DEVINDEX_HASHBITS
) ^
283 (val
>> (DEVINDEX_HASHBITS
* 2))) & mask
;
286 static inline unsigned int fib_info_hashfn(const struct fib_info
*fi
)
288 unsigned int mask
= (fib_info_hash_size
- 1);
289 unsigned int val
= fi
->fib_nhs
;
291 val
^= (fi
->fib_protocol
<< 8) | fi
->fib_scope
;
292 val
^= (__force u32
)fi
->fib_prefsrc
;
293 val
^= fi
->fib_priority
;
295 val
^= fib_devindex_hashfn(nh
->nh_oif
);
296 } endfor_nexthops(fi
)
298 return (val
^ (val
>> 7) ^ (val
>> 12)) & mask
;
301 static struct fib_info
*fib_find_info(const struct fib_info
*nfi
)
303 struct hlist_head
*head
;
307 hash
= fib_info_hashfn(nfi
);
308 head
= &fib_info_hash
[hash
];
310 hlist_for_each_entry(fi
, head
, fib_hash
) {
311 if (!net_eq(fi
->fib_net
, nfi
->fib_net
))
313 if (fi
->fib_nhs
!= nfi
->fib_nhs
)
315 if (nfi
->fib_protocol
== fi
->fib_protocol
&&
316 nfi
->fib_scope
== fi
->fib_scope
&&
317 nfi
->fib_prefsrc
== fi
->fib_prefsrc
&&
318 nfi
->fib_priority
== fi
->fib_priority
&&
319 nfi
->fib_type
== fi
->fib_type
&&
320 memcmp(nfi
->fib_metrics
, fi
->fib_metrics
,
321 sizeof(u32
) * RTAX_MAX
) == 0 &&
322 ((nfi
->fib_flags
^ fi
->fib_flags
) & ~RTNH_F_DEAD
) == 0 &&
323 (nfi
->fib_nhs
== 0 || nh_comp(fi
, nfi
) == 0))
330 /* Check, that the gateway is already configured.
331 * Used only by redirect accept routine.
333 int ip_fib_check_default(__be32 gw
, struct net_device
*dev
)
335 struct hlist_head
*head
;
339 spin_lock(&fib_info_lock
);
341 hash
= fib_devindex_hashfn(dev
->ifindex
);
342 head
= &fib_info_devhash
[hash
];
343 hlist_for_each_entry(nh
, head
, nh_hash
) {
344 if (nh
->nh_dev
== dev
&&
346 !(nh
->nh_flags
& RTNH_F_DEAD
)) {
347 spin_unlock(&fib_info_lock
);
352 spin_unlock(&fib_info_lock
);
357 static inline size_t fib_nlmsg_size(struct fib_info
*fi
)
359 size_t payload
= NLMSG_ALIGN(sizeof(struct rtmsg
))
360 + nla_total_size(4) /* RTA_TABLE */
361 + nla_total_size(4) /* RTA_DST */
362 + nla_total_size(4) /* RTA_PRIORITY */
363 + nla_total_size(4) /* RTA_PREFSRC */
364 + nla_total_size(TCP_CA_NAME_MAX
); /* RTAX_CC_ALGO */
366 /* space for nested metrics */
367 payload
+= nla_total_size((RTAX_MAX
* nla_total_size(4)));
370 /* Also handles the special case fib_nhs == 1 */
372 /* each nexthop is packed in an attribute */
373 size_t nhsize
= nla_total_size(sizeof(struct rtnexthop
));
375 /* may contain flow and gateway attribute */
376 nhsize
+= 2 * nla_total_size(4);
378 /* all nexthops are packed in a nested attribute */
379 payload
+= nla_total_size(fi
->fib_nhs
* nhsize
);
385 void rtmsg_fib(int event
, __be32 key
, struct fib_alias
*fa
,
386 int dst_len
, u32 tb_id
, const struct nl_info
*info
,
387 unsigned int nlm_flags
)
390 u32 seq
= info
->nlh
? info
->nlh
->nlmsg_seq
: 0;
393 skb
= nlmsg_new(fib_nlmsg_size(fa
->fa_info
), GFP_KERNEL
);
397 err
= fib_dump_info(skb
, info
->portid
, seq
, event
, tb_id
,
398 fa
->fa_type
, key
, dst_len
,
399 fa
->fa_tos
, fa
->fa_info
, nlm_flags
);
401 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
402 WARN_ON(err
== -EMSGSIZE
);
406 rtnl_notify(skb
, info
->nl_net
, info
->portid
, RTNLGRP_IPV4_ROUTE
,
407 info
->nlh
, GFP_KERNEL
);
411 rtnl_set_sk_err(info
->nl_net
, RTNLGRP_IPV4_ROUTE
, err
);
414 static int fib_detect_death(struct fib_info
*fi
, int order
,
415 struct fib_info
**last_resort
, int *last_idx
,
419 int state
= NUD_NONE
;
421 n
= neigh_lookup(&arp_tbl
, &fi
->fib_nh
[0].nh_gw
, fi
->fib_dev
);
423 state
= n
->nud_state
;
426 if (state
== NUD_REACHABLE
)
428 if ((state
& NUD_VALID
) && order
!= dflt
)
430 if ((state
& NUD_VALID
) ||
431 (*last_idx
< 0 && order
> dflt
)) {
438 #ifdef CONFIG_IP_ROUTE_MULTIPATH
440 static int fib_count_nexthops(struct rtnexthop
*rtnh
, int remaining
)
444 while (rtnh_ok(rtnh
, remaining
)) {
446 rtnh
= rtnh_next(rtnh
, &remaining
);
449 /* leftover implies invalid nexthop configuration, discard it */
450 return remaining
> 0 ? 0 : nhs
;
453 static int fib_get_nhs(struct fib_info
*fi
, struct rtnexthop
*rtnh
,
454 int remaining
, struct fib_config
*cfg
)
456 change_nexthops(fi
) {
459 if (!rtnh_ok(rtnh
, remaining
))
462 nexthop_nh
->nh_flags
=
463 (cfg
->fc_flags
& ~0xFF) | rtnh
->rtnh_flags
;
464 nexthop_nh
->nh_oif
= rtnh
->rtnh_ifindex
;
465 nexthop_nh
->nh_weight
= rtnh
->rtnh_hops
+ 1;
467 attrlen
= rtnh_attrlen(rtnh
);
469 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
471 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
472 nexthop_nh
->nh_gw
= nla
? nla_get_be32(nla
) : 0;
473 #ifdef CONFIG_IP_ROUTE_CLASSID
474 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
475 nexthop_nh
->nh_tclassid
= nla
? nla_get_u32(nla
) : 0;
476 if (nexthop_nh
->nh_tclassid
)
477 fi
->fib_net
->ipv4
.fib_num_tclassid_users
++;
481 rtnh
= rtnh_next(rtnh
, &remaining
);
482 } endfor_nexthops(fi
);
489 int fib_nh_match(struct fib_config
*cfg
, struct fib_info
*fi
)
491 #ifdef CONFIG_IP_ROUTE_MULTIPATH
492 struct rtnexthop
*rtnh
;
496 if (cfg
->fc_priority
&& cfg
->fc_priority
!= fi
->fib_priority
)
499 if (cfg
->fc_oif
|| cfg
->fc_gw
) {
500 if ((!cfg
->fc_oif
|| cfg
->fc_oif
== fi
->fib_nh
->nh_oif
) &&
501 (!cfg
->fc_gw
|| cfg
->fc_gw
== fi
->fib_nh
->nh_gw
))
506 #ifdef CONFIG_IP_ROUTE_MULTIPATH
507 if (cfg
->fc_mp
== NULL
)
511 remaining
= cfg
->fc_mp_len
;
516 if (!rtnh_ok(rtnh
, remaining
))
519 if (rtnh
->rtnh_ifindex
&& rtnh
->rtnh_ifindex
!= nh
->nh_oif
)
522 attrlen
= rtnh_attrlen(rtnh
);
524 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
526 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
527 if (nla
&& nla_get_be32(nla
) != nh
->nh_gw
)
529 #ifdef CONFIG_IP_ROUTE_CLASSID
530 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
531 if (nla
&& nla_get_u32(nla
) != nh
->nh_tclassid
)
536 rtnh
= rtnh_next(rtnh
, &remaining
);
537 } endfor_nexthops(fi
);
547 * Semantics of nexthop is very messy by historical reasons.
548 * We have to take into account, that:
549 * a) gateway can be actually local interface address,
550 * so that gatewayed route is direct.
551 * b) gateway must be on-link address, possibly
552 * described not by an ifaddr, but also by a direct route.
553 * c) If both gateway and interface are specified, they should not
555 * d) If we use tunnel routes, gateway could be not on-link.
557 * Attempt to reconcile all of these (alas, self-contradictory) conditions
558 * results in pretty ugly and hairy code with obscure logic.
560 * I chose to generalized it instead, so that the size
561 * of code does not increase practically, but it becomes
563 * Every prefix is assigned a "scope" value: "host" is local address,
564 * "link" is direct route,
565 * [ ... "site" ... "interior" ... ]
566 * and "universe" is true gateway route with global meaning.
568 * Every prefix refers to a set of "nexthop"s (gw, oif),
569 * where gw must have narrower scope. This recursion stops
570 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
571 * which means that gw is forced to be on link.
573 * Code is still hairy, but now it is apparently logically
574 * consistent and very flexible. F.e. as by-product it allows
575 * to co-exists in peace independent exterior and interior
578 * Normally it looks as following.
580 * {universe prefix} -> (gw, oif) [scope link]
582 * |-> {link prefix} -> (gw, oif) [scope local]
584 * |-> {local prefix} (terminal node)
586 static int fib_check_nh(struct fib_config
*cfg
, struct fib_info
*fi
,
591 struct net_device
*dev
;
593 net
= cfg
->fc_nlinfo
.nl_net
;
595 struct fib_result res
;
597 if (nh
->nh_flags
& RTNH_F_ONLINK
) {
599 if (cfg
->fc_scope
>= RT_SCOPE_LINK
)
601 if (inet_addr_type(net
, nh
->nh_gw
) != RTN_UNICAST
)
603 dev
= __dev_get_by_index(net
, nh
->nh_oif
);
606 if (!(dev
->flags
& IFF_UP
))
610 nh
->nh_scope
= RT_SCOPE_LINK
;
615 struct flowi4 fl4
= {
617 .flowi4_scope
= cfg
->fc_scope
+ 1,
618 .flowi4_oif
= nh
->nh_oif
,
619 .flowi4_iif
= LOOPBACK_IFINDEX
,
622 /* It is not necessary, but requires a bit of thinking */
623 if (fl4
.flowi4_scope
< RT_SCOPE_LINK
)
624 fl4
.flowi4_scope
= RT_SCOPE_LINK
;
625 err
= fib_lookup(net
, &fl4
, &res
);
632 if (res
.type
!= RTN_UNICAST
&& res
.type
!= RTN_LOCAL
)
634 nh
->nh_scope
= res
.scope
;
635 nh
->nh_oif
= FIB_RES_OIF(res
);
636 nh
->nh_dev
= dev
= FIB_RES_DEV(res
);
640 err
= (dev
->flags
& IFF_UP
) ? 0 : -ENETDOWN
;
642 struct in_device
*in_dev
;
644 if (nh
->nh_flags
& (RTNH_F_PERVASIVE
| RTNH_F_ONLINK
))
649 in_dev
= inetdev_by_index(net
, nh
->nh_oif
);
653 if (!(in_dev
->dev
->flags
& IFF_UP
))
655 nh
->nh_dev
= in_dev
->dev
;
656 dev_hold(nh
->nh_dev
);
657 nh
->nh_scope
= RT_SCOPE_HOST
;
665 static inline unsigned int fib_laddr_hashfn(__be32 val
)
667 unsigned int mask
= (fib_info_hash_size
- 1);
669 return ((__force u32
)val
^
670 ((__force u32
)val
>> 7) ^
671 ((__force u32
)val
>> 14)) & mask
;
674 static struct hlist_head
*fib_info_hash_alloc(int bytes
)
676 if (bytes
<= PAGE_SIZE
)
677 return kzalloc(bytes
, GFP_KERNEL
);
679 return (struct hlist_head
*)
680 __get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
684 static void fib_info_hash_free(struct hlist_head
*hash
, int bytes
)
689 if (bytes
<= PAGE_SIZE
)
692 free_pages((unsigned long) hash
, get_order(bytes
));
695 static void fib_info_hash_move(struct hlist_head
*new_info_hash
,
696 struct hlist_head
*new_laddrhash
,
697 unsigned int new_size
)
699 struct hlist_head
*old_info_hash
, *old_laddrhash
;
700 unsigned int old_size
= fib_info_hash_size
;
701 unsigned int i
, bytes
;
703 spin_lock_bh(&fib_info_lock
);
704 old_info_hash
= fib_info_hash
;
705 old_laddrhash
= fib_info_laddrhash
;
706 fib_info_hash_size
= new_size
;
708 for (i
= 0; i
< old_size
; i
++) {
709 struct hlist_head
*head
= &fib_info_hash
[i
];
710 struct hlist_node
*n
;
713 hlist_for_each_entry_safe(fi
, n
, head
, fib_hash
) {
714 struct hlist_head
*dest
;
715 unsigned int new_hash
;
717 hlist_del(&fi
->fib_hash
);
719 new_hash
= fib_info_hashfn(fi
);
720 dest
= &new_info_hash
[new_hash
];
721 hlist_add_head(&fi
->fib_hash
, dest
);
724 fib_info_hash
= new_info_hash
;
726 for (i
= 0; i
< old_size
; i
++) {
727 struct hlist_head
*lhead
= &fib_info_laddrhash
[i
];
728 struct hlist_node
*n
;
731 hlist_for_each_entry_safe(fi
, n
, lhead
, fib_lhash
) {
732 struct hlist_head
*ldest
;
733 unsigned int new_hash
;
735 hlist_del(&fi
->fib_lhash
);
737 new_hash
= fib_laddr_hashfn(fi
->fib_prefsrc
);
738 ldest
= &new_laddrhash
[new_hash
];
739 hlist_add_head(&fi
->fib_lhash
, ldest
);
742 fib_info_laddrhash
= new_laddrhash
;
744 spin_unlock_bh(&fib_info_lock
);
746 bytes
= old_size
* sizeof(struct hlist_head
*);
747 fib_info_hash_free(old_info_hash
, bytes
);
748 fib_info_hash_free(old_laddrhash
, bytes
);
751 __be32
fib_info_update_nh_saddr(struct net
*net
, struct fib_nh
*nh
)
753 nh
->nh_saddr
= inet_select_addr(nh
->nh_dev
,
755 nh
->nh_parent
->fib_scope
);
756 nh
->nh_saddr_genid
= atomic_read(&net
->ipv4
.dev_addr_genid
);
761 struct fib_info
*fib_create_info(struct fib_config
*cfg
)
764 struct fib_info
*fi
= NULL
;
765 struct fib_info
*ofi
;
767 struct net
*net
= cfg
->fc_nlinfo
.nl_net
;
769 if (cfg
->fc_type
> RTN_MAX
)
772 /* Fast check to catch the most weird cases */
773 if (fib_props
[cfg
->fc_type
].scope
> cfg
->fc_scope
)
776 #ifdef CONFIG_IP_ROUTE_MULTIPATH
778 nhs
= fib_count_nexthops(cfg
->fc_mp
, cfg
->fc_mp_len
);
785 if (fib_info_cnt
>= fib_info_hash_size
) {
786 unsigned int new_size
= fib_info_hash_size
<< 1;
787 struct hlist_head
*new_info_hash
;
788 struct hlist_head
*new_laddrhash
;
793 bytes
= new_size
* sizeof(struct hlist_head
*);
794 new_info_hash
= fib_info_hash_alloc(bytes
);
795 new_laddrhash
= fib_info_hash_alloc(bytes
);
796 if (!new_info_hash
|| !new_laddrhash
) {
797 fib_info_hash_free(new_info_hash
, bytes
);
798 fib_info_hash_free(new_laddrhash
, bytes
);
800 fib_info_hash_move(new_info_hash
, new_laddrhash
, new_size
);
802 if (!fib_info_hash_size
)
806 fi
= kzalloc(sizeof(*fi
)+nhs
*sizeof(struct fib_nh
), GFP_KERNEL
);
811 fi
->fib_metrics
= kzalloc(sizeof(u32
) * RTAX_MAX
, GFP_KERNEL
);
812 if (!fi
->fib_metrics
)
815 fi
->fib_metrics
= (u32
*) dst_default_metrics
;
817 fi
->fib_net
= hold_net(net
);
818 fi
->fib_protocol
= cfg
->fc_protocol
;
819 fi
->fib_scope
= cfg
->fc_scope
;
820 fi
->fib_flags
= cfg
->fc_flags
;
821 fi
->fib_priority
= cfg
->fc_priority
;
822 fi
->fib_prefsrc
= cfg
->fc_prefsrc
;
823 fi
->fib_type
= cfg
->fc_type
;
826 change_nexthops(fi
) {
827 nexthop_nh
->nh_parent
= fi
;
828 nexthop_nh
->nh_pcpu_rth_output
= alloc_percpu(struct rtable __rcu
*);
829 if (!nexthop_nh
->nh_pcpu_rth_output
)
831 } endfor_nexthops(fi
)
837 nla_for_each_attr(nla
, cfg
->fc_mx
, cfg
->fc_mx_len
, remaining
) {
838 int type
= nla_type(nla
);
845 if (type
== RTAX_CC_ALGO
) {
846 char tmp
[TCP_CA_NAME_MAX
];
848 nla_strlcpy(tmp
, nla
, sizeof(tmp
));
849 val
= tcp_ca_get_key_by_name(tmp
);
850 if (val
== TCP_CA_UNSPEC
)
853 val
= nla_get_u32(nla
);
855 if (type
== RTAX_ADVMSS
&& val
> 65535 - 40)
857 if (type
== RTAX_MTU
&& val
> 65535 - 15)
859 fi
->fib_metrics
[type
- 1] = val
;
865 #ifdef CONFIG_IP_ROUTE_MULTIPATH
866 err
= fib_get_nhs(fi
, cfg
->fc_mp
, cfg
->fc_mp_len
, cfg
);
869 if (cfg
->fc_oif
&& fi
->fib_nh
->nh_oif
!= cfg
->fc_oif
)
871 if (cfg
->fc_gw
&& fi
->fib_nh
->nh_gw
!= cfg
->fc_gw
)
873 #ifdef CONFIG_IP_ROUTE_CLASSID
874 if (cfg
->fc_flow
&& fi
->fib_nh
->nh_tclassid
!= cfg
->fc_flow
)
881 struct fib_nh
*nh
= fi
->fib_nh
;
883 nh
->nh_oif
= cfg
->fc_oif
;
884 nh
->nh_gw
= cfg
->fc_gw
;
885 nh
->nh_flags
= cfg
->fc_flags
;
886 #ifdef CONFIG_IP_ROUTE_CLASSID
887 nh
->nh_tclassid
= cfg
->fc_flow
;
889 fi
->fib_net
->ipv4
.fib_num_tclassid_users
++;
891 #ifdef CONFIG_IP_ROUTE_MULTIPATH
896 if (fib_props
[cfg
->fc_type
].error
) {
897 if (cfg
->fc_gw
|| cfg
->fc_oif
|| cfg
->fc_mp
)
901 switch (cfg
->fc_type
) {
913 if (cfg
->fc_scope
> RT_SCOPE_HOST
)
916 if (cfg
->fc_scope
== RT_SCOPE_HOST
) {
917 struct fib_nh
*nh
= fi
->fib_nh
;
919 /* Local address is added. */
920 if (nhs
!= 1 || nh
->nh_gw
)
922 nh
->nh_scope
= RT_SCOPE_NOWHERE
;
923 nh
->nh_dev
= dev_get_by_index(net
, fi
->fib_nh
->nh_oif
);
925 if (nh
->nh_dev
== NULL
)
928 change_nexthops(fi
) {
929 err
= fib_check_nh(cfg
, fi
, nexthop_nh
);
932 } endfor_nexthops(fi
)
935 if (fi
->fib_prefsrc
) {
936 if (cfg
->fc_type
!= RTN_LOCAL
|| !cfg
->fc_dst
||
937 fi
->fib_prefsrc
!= cfg
->fc_dst
)
938 if (inet_addr_type(net
, fi
->fib_prefsrc
) != RTN_LOCAL
)
942 change_nexthops(fi
) {
943 fib_info_update_nh_saddr(net
, nexthop_nh
);
944 } endfor_nexthops(fi
)
947 ofi
= fib_find_info(fi
);
956 atomic_inc(&fi
->fib_clntref
);
957 spin_lock_bh(&fib_info_lock
);
958 hlist_add_head(&fi
->fib_hash
,
959 &fib_info_hash
[fib_info_hashfn(fi
)]);
960 if (fi
->fib_prefsrc
) {
961 struct hlist_head
*head
;
963 head
= &fib_info_laddrhash
[fib_laddr_hashfn(fi
->fib_prefsrc
)];
964 hlist_add_head(&fi
->fib_lhash
, head
);
966 change_nexthops(fi
) {
967 struct hlist_head
*head
;
970 if (!nexthop_nh
->nh_dev
)
972 hash
= fib_devindex_hashfn(nexthop_nh
->nh_dev
->ifindex
);
973 head
= &fib_info_devhash
[hash
];
974 hlist_add_head(&nexthop_nh
->nh_hash
, head
);
975 } endfor_nexthops(fi
)
976 spin_unlock_bh(&fib_info_lock
);
991 int fib_dump_info(struct sk_buff
*skb
, u32 portid
, u32 seq
, int event
,
992 u32 tb_id
, u8 type
, __be32 dst
, int dst_len
, u8 tos
,
993 struct fib_info
*fi
, unsigned int flags
)
995 struct nlmsghdr
*nlh
;
998 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*rtm
), flags
);
1002 rtm
= nlmsg_data(nlh
);
1003 rtm
->rtm_family
= AF_INET
;
1004 rtm
->rtm_dst_len
= dst_len
;
1005 rtm
->rtm_src_len
= 0;
1008 rtm
->rtm_table
= tb_id
;
1010 rtm
->rtm_table
= RT_TABLE_COMPAT
;
1011 if (nla_put_u32(skb
, RTA_TABLE
, tb_id
))
1012 goto nla_put_failure
;
1013 rtm
->rtm_type
= type
;
1014 rtm
->rtm_flags
= fi
->fib_flags
;
1015 rtm
->rtm_scope
= fi
->fib_scope
;
1016 rtm
->rtm_protocol
= fi
->fib_protocol
;
1018 if (rtm
->rtm_dst_len
&&
1019 nla_put_be32(skb
, RTA_DST
, dst
))
1020 goto nla_put_failure
;
1021 if (fi
->fib_priority
&&
1022 nla_put_u32(skb
, RTA_PRIORITY
, fi
->fib_priority
))
1023 goto nla_put_failure
;
1024 if (rtnetlink_put_metrics(skb
, fi
->fib_metrics
) < 0)
1025 goto nla_put_failure
;
1027 if (fi
->fib_prefsrc
&&
1028 nla_put_be32(skb
, RTA_PREFSRC
, fi
->fib_prefsrc
))
1029 goto nla_put_failure
;
1030 if (fi
->fib_nhs
== 1) {
1031 if (fi
->fib_nh
->nh_gw
&&
1032 nla_put_be32(skb
, RTA_GATEWAY
, fi
->fib_nh
->nh_gw
))
1033 goto nla_put_failure
;
1034 if (fi
->fib_nh
->nh_oif
&&
1035 nla_put_u32(skb
, RTA_OIF
, fi
->fib_nh
->nh_oif
))
1036 goto nla_put_failure
;
1037 #ifdef CONFIG_IP_ROUTE_CLASSID
1038 if (fi
->fib_nh
[0].nh_tclassid
&&
1039 nla_put_u32(skb
, RTA_FLOW
, fi
->fib_nh
[0].nh_tclassid
))
1040 goto nla_put_failure
;
1043 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1044 if (fi
->fib_nhs
> 1) {
1045 struct rtnexthop
*rtnh
;
1048 mp
= nla_nest_start(skb
, RTA_MULTIPATH
);
1050 goto nla_put_failure
;
1053 rtnh
= nla_reserve_nohdr(skb
, sizeof(*rtnh
));
1055 goto nla_put_failure
;
1057 rtnh
->rtnh_flags
= nh
->nh_flags
& 0xFF;
1058 rtnh
->rtnh_hops
= nh
->nh_weight
- 1;
1059 rtnh
->rtnh_ifindex
= nh
->nh_oif
;
1062 nla_put_be32(skb
, RTA_GATEWAY
, nh
->nh_gw
))
1063 goto nla_put_failure
;
1064 #ifdef CONFIG_IP_ROUTE_CLASSID
1065 if (nh
->nh_tclassid
&&
1066 nla_put_u32(skb
, RTA_FLOW
, nh
->nh_tclassid
))
1067 goto nla_put_failure
;
1069 /* length of rtnetlink header + attributes */
1070 rtnh
->rtnh_len
= nlmsg_get_pos(skb
) - (void *) rtnh
;
1071 } endfor_nexthops(fi
);
1073 nla_nest_end(skb
, mp
);
1076 nlmsg_end(skb
, nlh
);
1080 nlmsg_cancel(skb
, nlh
);
1086 * - local address disappeared -> we must delete all the entries
1088 * - device went down -> we must shutdown all nexthops going via it.
1090 int fib_sync_down_addr(struct net
*net
, __be32 local
)
1093 unsigned int hash
= fib_laddr_hashfn(local
);
1094 struct hlist_head
*head
= &fib_info_laddrhash
[hash
];
1095 struct fib_info
*fi
;
1097 if (fib_info_laddrhash
== NULL
|| local
== 0)
1100 hlist_for_each_entry(fi
, head
, fib_lhash
) {
1101 if (!net_eq(fi
->fib_net
, net
))
1103 if (fi
->fib_prefsrc
== local
) {
1104 fi
->fib_flags
|= RTNH_F_DEAD
;
1111 int fib_sync_down_dev(struct net_device
*dev
, int force
)
1114 int scope
= RT_SCOPE_NOWHERE
;
1115 struct fib_info
*prev_fi
= NULL
;
1116 unsigned int hash
= fib_devindex_hashfn(dev
->ifindex
);
1117 struct hlist_head
*head
= &fib_info_devhash
[hash
];
1123 hlist_for_each_entry(nh
, head
, nh_hash
) {
1124 struct fib_info
*fi
= nh
->nh_parent
;
1127 BUG_ON(!fi
->fib_nhs
);
1128 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1132 change_nexthops(fi
) {
1133 if (nexthop_nh
->nh_flags
& RTNH_F_DEAD
)
1135 else if (nexthop_nh
->nh_dev
== dev
&&
1136 nexthop_nh
->nh_scope
!= scope
) {
1137 nexthop_nh
->nh_flags
|= RTNH_F_DEAD
;
1138 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1139 spin_lock_bh(&fib_multipath_lock
);
1140 fi
->fib_power
-= nexthop_nh
->nh_power
;
1141 nexthop_nh
->nh_power
= 0;
1142 spin_unlock_bh(&fib_multipath_lock
);
1146 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1147 if (force
> 1 && nexthop_nh
->nh_dev
== dev
) {
1152 } endfor_nexthops(fi
)
1153 if (dead
== fi
->fib_nhs
) {
1154 fi
->fib_flags
|= RTNH_F_DEAD
;
1162 /* Must be invoked inside of an RCU protected region. */
1163 void fib_select_default(struct fib_result
*res
)
1165 struct fib_info
*fi
= NULL
, *last_resort
= NULL
;
1166 struct hlist_head
*fa_head
= res
->fa_head
;
1167 struct fib_table
*tb
= res
->table
;
1168 int order
= -1, last_idx
= -1;
1169 struct fib_alias
*fa
;
1171 hlist_for_each_entry_rcu(fa
, fa_head
, fa_list
) {
1172 struct fib_info
*next_fi
= fa
->fa_info
;
1174 if (next_fi
->fib_scope
!= res
->scope
||
1175 fa
->fa_type
!= RTN_UNICAST
)
1178 if (next_fi
->fib_priority
> res
->fi
->fib_priority
)
1180 if (!next_fi
->fib_nh
[0].nh_gw
||
1181 next_fi
->fib_nh
[0].nh_scope
!= RT_SCOPE_LINK
)
1184 fib_alias_accessed(fa
);
1187 if (next_fi
!= res
->fi
)
1189 } else if (!fib_detect_death(fi
, order
, &last_resort
,
1190 &last_idx
, tb
->tb_default
)) {
1191 fib_result_assign(res
, fi
);
1192 tb
->tb_default
= order
;
1199 if (order
<= 0 || fi
== NULL
) {
1200 tb
->tb_default
= -1;
1204 if (!fib_detect_death(fi
, order
, &last_resort
, &last_idx
,
1206 fib_result_assign(res
, fi
);
1207 tb
->tb_default
= order
;
1212 fib_result_assign(res
, last_resort
);
1213 tb
->tb_default
= last_idx
;
1218 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1221 * Dead device goes up. We wake up dead nexthops.
1222 * It takes sense only on multipath routes.
1224 int fib_sync_up(struct net_device
*dev
)
1226 struct fib_info
*prev_fi
;
1228 struct hlist_head
*head
;
1232 if (!(dev
->flags
& IFF_UP
))
1236 hash
= fib_devindex_hashfn(dev
->ifindex
);
1237 head
= &fib_info_devhash
[hash
];
1240 hlist_for_each_entry(nh
, head
, nh_hash
) {
1241 struct fib_info
*fi
= nh
->nh_parent
;
1244 BUG_ON(!fi
->fib_nhs
);
1245 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1250 change_nexthops(fi
) {
1251 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1255 if (nexthop_nh
->nh_dev
== NULL
||
1256 !(nexthop_nh
->nh_dev
->flags
& IFF_UP
))
1258 if (nexthop_nh
->nh_dev
!= dev
||
1259 !__in_dev_get_rtnl(dev
))
1262 spin_lock_bh(&fib_multipath_lock
);
1263 nexthop_nh
->nh_power
= 0;
1264 nexthop_nh
->nh_flags
&= ~RTNH_F_DEAD
;
1265 spin_unlock_bh(&fib_multipath_lock
);
1266 } endfor_nexthops(fi
)
1269 fi
->fib_flags
&= ~RTNH_F_DEAD
;
1278 * The algorithm is suboptimal, but it provides really
1279 * fair weighted route distribution.
1281 void fib_select_multipath(struct fib_result
*res
)
1283 struct fib_info
*fi
= res
->fi
;
1286 spin_lock_bh(&fib_multipath_lock
);
1287 if (fi
->fib_power
<= 0) {
1289 change_nexthops(fi
) {
1290 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1291 power
+= nexthop_nh
->nh_weight
;
1292 nexthop_nh
->nh_power
= nexthop_nh
->nh_weight
;
1294 } endfor_nexthops(fi
);
1295 fi
->fib_power
= power
;
1297 spin_unlock_bh(&fib_multipath_lock
);
1298 /* Race condition: route has just become dead. */
1305 /* w should be random number [0..fi->fib_power-1],
1306 * it is pretty bad approximation.
1309 w
= jiffies
% fi
->fib_power
;
1311 change_nexthops(fi
) {
1312 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
) &&
1313 nexthop_nh
->nh_power
) {
1314 w
-= nexthop_nh
->nh_power
;
1316 nexthop_nh
->nh_power
--;
1318 res
->nh_sel
= nhsel
;
1319 spin_unlock_bh(&fib_multipath_lock
);
1323 } endfor_nexthops(fi
);
1325 /* Race condition: route has just become dead. */
1327 spin_unlock_bh(&fib_multipath_lock
);