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
= nh
->nh_exceptions
;
163 for (i
= 0; i
< FNHE_HASH_SIZE
; i
++) {
164 struct fib_nh_exception
*fnhe
;
166 fnhe
= rcu_dereference_protected(hash
[i
].chain
, 1);
168 struct fib_nh_exception
*next
;
170 next
= rcu_dereference_protected(fnhe
->fnhe_next
, 1);
172 rt_fibinfo_free(&fnhe
->fnhe_rth
);
182 static void rt_fibinfo_free_cpus(struct rtable __rcu
* __percpu
*rtp
)
189 for_each_possible_cpu(cpu
) {
192 rt
= rcu_dereference_protected(*per_cpu_ptr(rtp
, cpu
), 1);
199 /* Release a nexthop info record */
200 static void free_fib_info_rcu(struct rcu_head
*head
)
202 struct fib_info
*fi
= container_of(head
, struct fib_info
, rcu
);
204 change_nexthops(fi
) {
205 if (nexthop_nh
->nh_dev
)
206 dev_put(nexthop_nh
->nh_dev
);
207 if (nexthop_nh
->nh_exceptions
)
208 free_nh_exceptions(nexthop_nh
);
209 rt_fibinfo_free_cpus(nexthop_nh
->nh_pcpu_rth_output
);
210 rt_fibinfo_free(&nexthop_nh
->nh_rth_input
);
211 } endfor_nexthops(fi
);
213 release_net(fi
->fib_net
);
214 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
)
215 kfree(fi
->fib_metrics
);
219 void free_fib_info(struct fib_info
*fi
)
221 if (fi
->fib_dead
== 0) {
222 pr_warn("Freeing alive fib_info %p\n", fi
);
226 #ifdef CONFIG_IP_ROUTE_CLASSID
227 change_nexthops(fi
) {
228 if (nexthop_nh
->nh_tclassid
)
229 fi
->fib_net
->ipv4
.fib_num_tclassid_users
--;
230 } endfor_nexthops(fi
);
232 call_rcu(&fi
->rcu
, free_fib_info_rcu
);
235 void fib_release_info(struct fib_info
*fi
)
237 spin_lock_bh(&fib_info_lock
);
238 if (fi
&& --fi
->fib_treeref
== 0) {
239 hlist_del(&fi
->fib_hash
);
241 hlist_del(&fi
->fib_lhash
);
242 change_nexthops(fi
) {
243 if (!nexthop_nh
->nh_dev
)
245 hlist_del(&nexthop_nh
->nh_hash
);
246 } endfor_nexthops(fi
)
250 spin_unlock_bh(&fib_info_lock
);
253 static inline int nh_comp(const struct fib_info
*fi
, const struct fib_info
*ofi
)
255 const struct fib_nh
*onh
= ofi
->fib_nh
;
258 if (nh
->nh_oif
!= onh
->nh_oif
||
259 nh
->nh_gw
!= onh
->nh_gw
||
260 nh
->nh_scope
!= onh
->nh_scope
||
261 #ifdef CONFIG_IP_ROUTE_MULTIPATH
262 nh
->nh_weight
!= onh
->nh_weight
||
264 #ifdef CONFIG_IP_ROUTE_CLASSID
265 nh
->nh_tclassid
!= onh
->nh_tclassid
||
267 ((nh
->nh_flags
^ onh
->nh_flags
) & ~RTNH_F_DEAD
))
270 } endfor_nexthops(fi
);
274 static inline unsigned int fib_devindex_hashfn(unsigned int val
)
276 unsigned int mask
= DEVINDEX_HASHSIZE
- 1;
279 (val
>> DEVINDEX_HASHBITS
) ^
280 (val
>> (DEVINDEX_HASHBITS
* 2))) & mask
;
283 static inline unsigned int fib_info_hashfn(const struct fib_info
*fi
)
285 unsigned int mask
= (fib_info_hash_size
- 1);
286 unsigned int val
= fi
->fib_nhs
;
288 val
^= (fi
->fib_protocol
<< 8) | fi
->fib_scope
;
289 val
^= (__force u32
)fi
->fib_prefsrc
;
290 val
^= fi
->fib_priority
;
292 val
^= fib_devindex_hashfn(nh
->nh_oif
);
293 } endfor_nexthops(fi
)
295 return (val
^ (val
>> 7) ^ (val
>> 12)) & mask
;
298 static struct fib_info
*fib_find_info(const struct fib_info
*nfi
)
300 struct hlist_head
*head
;
301 struct hlist_node
*node
;
305 hash
= fib_info_hashfn(nfi
);
306 head
= &fib_info_hash
[hash
];
308 hlist_for_each_entry(fi
, node
, head
, fib_hash
) {
309 if (!net_eq(fi
->fib_net
, nfi
->fib_net
))
311 if (fi
->fib_nhs
!= nfi
->fib_nhs
)
313 if (nfi
->fib_protocol
== fi
->fib_protocol
&&
314 nfi
->fib_scope
== fi
->fib_scope
&&
315 nfi
->fib_prefsrc
== fi
->fib_prefsrc
&&
316 nfi
->fib_priority
== fi
->fib_priority
&&
317 memcmp(nfi
->fib_metrics
, fi
->fib_metrics
,
318 sizeof(u32
) * RTAX_MAX
) == 0 &&
319 ((nfi
->fib_flags
^ fi
->fib_flags
) & ~RTNH_F_DEAD
) == 0 &&
320 (nfi
->fib_nhs
== 0 || nh_comp(fi
, nfi
) == 0))
327 /* Check, that the gateway is already configured.
328 * Used only by redirect accept routine.
330 int ip_fib_check_default(__be32 gw
, struct net_device
*dev
)
332 struct hlist_head
*head
;
333 struct hlist_node
*node
;
337 spin_lock(&fib_info_lock
);
339 hash
= fib_devindex_hashfn(dev
->ifindex
);
340 head
= &fib_info_devhash
[hash
];
341 hlist_for_each_entry(nh
, node
, head
, nh_hash
) {
342 if (nh
->nh_dev
== dev
&&
344 !(nh
->nh_flags
& RTNH_F_DEAD
)) {
345 spin_unlock(&fib_info_lock
);
350 spin_unlock(&fib_info_lock
);
355 static inline size_t fib_nlmsg_size(struct fib_info
*fi
)
357 size_t payload
= NLMSG_ALIGN(sizeof(struct rtmsg
))
358 + nla_total_size(4) /* RTA_TABLE */
359 + nla_total_size(4) /* RTA_DST */
360 + nla_total_size(4) /* RTA_PRIORITY */
361 + nla_total_size(4); /* RTA_PREFSRC */
363 /* space for nested metrics */
364 payload
+= nla_total_size((RTAX_MAX
* nla_total_size(4)));
367 /* Also handles the special case fib_nhs == 1 */
369 /* each nexthop is packed in an attribute */
370 size_t nhsize
= nla_total_size(sizeof(struct rtnexthop
));
372 /* may contain flow and gateway attribute */
373 nhsize
+= 2 * nla_total_size(4);
375 /* all nexthops are packed in a nested attribute */
376 payload
+= nla_total_size(fi
->fib_nhs
* nhsize
);
382 void rtmsg_fib(int event
, __be32 key
, struct fib_alias
*fa
,
383 int dst_len
, u32 tb_id
, struct nl_info
*info
,
384 unsigned int nlm_flags
)
387 u32 seq
= info
->nlh
? info
->nlh
->nlmsg_seq
: 0;
390 skb
= nlmsg_new(fib_nlmsg_size(fa
->fa_info
), GFP_KERNEL
);
394 err
= fib_dump_info(skb
, info
->pid
, seq
, event
, tb_id
,
395 fa
->fa_type
, key
, dst_len
,
396 fa
->fa_tos
, fa
->fa_info
, nlm_flags
);
398 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
399 WARN_ON(err
== -EMSGSIZE
);
403 rtnl_notify(skb
, info
->nl_net
, info
->pid
, RTNLGRP_IPV4_ROUTE
,
404 info
->nlh
, GFP_KERNEL
);
408 rtnl_set_sk_err(info
->nl_net
, RTNLGRP_IPV4_ROUTE
, err
);
411 /* Return the first fib alias matching TOS with
412 * priority less than or equal to PRIO.
414 struct fib_alias
*fib_find_alias(struct list_head
*fah
, u8 tos
, u32 prio
)
417 struct fib_alias
*fa
;
418 list_for_each_entry(fa
, fah
, fa_list
) {
419 if (fa
->fa_tos
> tos
)
421 if (fa
->fa_info
->fib_priority
>= prio
||
429 int fib_detect_death(struct fib_info
*fi
, int order
,
430 struct fib_info
**last_resort
, int *last_idx
, int dflt
)
433 int state
= NUD_NONE
;
435 n
= neigh_lookup(&arp_tbl
, &fi
->fib_nh
[0].nh_gw
, fi
->fib_dev
);
437 state
= n
->nud_state
;
440 if (state
== NUD_REACHABLE
)
442 if ((state
& NUD_VALID
) && order
!= dflt
)
444 if ((state
& NUD_VALID
) ||
445 (*last_idx
< 0 && order
> dflt
)) {
452 #ifdef CONFIG_IP_ROUTE_MULTIPATH
454 static int fib_count_nexthops(struct rtnexthop
*rtnh
, int remaining
)
458 while (rtnh_ok(rtnh
, remaining
)) {
460 rtnh
= rtnh_next(rtnh
, &remaining
);
463 /* leftover implies invalid nexthop configuration, discard it */
464 return remaining
> 0 ? 0 : nhs
;
467 static int fib_get_nhs(struct fib_info
*fi
, struct rtnexthop
*rtnh
,
468 int remaining
, struct fib_config
*cfg
)
470 change_nexthops(fi
) {
473 if (!rtnh_ok(rtnh
, remaining
))
476 nexthop_nh
->nh_flags
=
477 (cfg
->fc_flags
& ~0xFF) | rtnh
->rtnh_flags
;
478 nexthop_nh
->nh_oif
= rtnh
->rtnh_ifindex
;
479 nexthop_nh
->nh_weight
= rtnh
->rtnh_hops
+ 1;
481 attrlen
= rtnh_attrlen(rtnh
);
483 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
485 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
486 nexthop_nh
->nh_gw
= nla
? nla_get_be32(nla
) : 0;
487 #ifdef CONFIG_IP_ROUTE_CLASSID
488 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
489 nexthop_nh
->nh_tclassid
= nla
? nla_get_u32(nla
) : 0;
490 if (nexthop_nh
->nh_tclassid
)
491 fi
->fib_net
->ipv4
.fib_num_tclassid_users
++;
495 rtnh
= rtnh_next(rtnh
, &remaining
);
496 } endfor_nexthops(fi
);
503 int fib_nh_match(struct fib_config
*cfg
, struct fib_info
*fi
)
505 #ifdef CONFIG_IP_ROUTE_MULTIPATH
506 struct rtnexthop
*rtnh
;
510 if (cfg
->fc_priority
&& cfg
->fc_priority
!= fi
->fib_priority
)
513 if (cfg
->fc_oif
|| cfg
->fc_gw
) {
514 if ((!cfg
->fc_oif
|| cfg
->fc_oif
== fi
->fib_nh
->nh_oif
) &&
515 (!cfg
->fc_gw
|| cfg
->fc_gw
== fi
->fib_nh
->nh_gw
))
520 #ifdef CONFIG_IP_ROUTE_MULTIPATH
521 if (cfg
->fc_mp
== NULL
)
525 remaining
= cfg
->fc_mp_len
;
530 if (!rtnh_ok(rtnh
, remaining
))
533 if (rtnh
->rtnh_ifindex
&& rtnh
->rtnh_ifindex
!= nh
->nh_oif
)
536 attrlen
= rtnh_attrlen(rtnh
);
538 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
540 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
541 if (nla
&& nla_get_be32(nla
) != nh
->nh_gw
)
543 #ifdef CONFIG_IP_ROUTE_CLASSID
544 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
545 if (nla
&& nla_get_u32(nla
) != nh
->nh_tclassid
)
550 rtnh
= rtnh_next(rtnh
, &remaining
);
551 } endfor_nexthops(fi
);
561 * Semantics of nexthop is very messy by historical reasons.
562 * We have to take into account, that:
563 * a) gateway can be actually local interface address,
564 * so that gatewayed route is direct.
565 * b) gateway must be on-link address, possibly
566 * described not by an ifaddr, but also by a direct route.
567 * c) If both gateway and interface are specified, they should not
569 * d) If we use tunnel routes, gateway could be not on-link.
571 * Attempt to reconcile all of these (alas, self-contradictory) conditions
572 * results in pretty ugly and hairy code with obscure logic.
574 * I chose to generalized it instead, so that the size
575 * of code does not increase practically, but it becomes
577 * Every prefix is assigned a "scope" value: "host" is local address,
578 * "link" is direct route,
579 * [ ... "site" ... "interior" ... ]
580 * and "universe" is true gateway route with global meaning.
582 * Every prefix refers to a set of "nexthop"s (gw, oif),
583 * where gw must have narrower scope. This recursion stops
584 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
585 * which means that gw is forced to be on link.
587 * Code is still hairy, but now it is apparently logically
588 * consistent and very flexible. F.e. as by-product it allows
589 * to co-exists in peace independent exterior and interior
592 * Normally it looks as following.
594 * {universe prefix} -> (gw, oif) [scope link]
596 * |-> {link prefix} -> (gw, oif) [scope local]
598 * |-> {local prefix} (terminal node)
600 static int fib_check_nh(struct fib_config
*cfg
, struct fib_info
*fi
,
605 struct net_device
*dev
;
607 net
= cfg
->fc_nlinfo
.nl_net
;
609 struct fib_result res
;
611 if (nh
->nh_flags
& RTNH_F_ONLINK
) {
613 if (cfg
->fc_scope
>= RT_SCOPE_LINK
)
615 if (inet_addr_type(net
, nh
->nh_gw
) != RTN_UNICAST
)
617 dev
= __dev_get_by_index(net
, nh
->nh_oif
);
620 if (!(dev
->flags
& IFF_UP
))
624 nh
->nh_scope
= RT_SCOPE_LINK
;
629 struct flowi4 fl4
= {
631 .flowi4_scope
= cfg
->fc_scope
+ 1,
632 .flowi4_oif
= nh
->nh_oif
,
635 /* It is not necessary, but requires a bit of thinking */
636 if (fl4
.flowi4_scope
< RT_SCOPE_LINK
)
637 fl4
.flowi4_scope
= RT_SCOPE_LINK
;
638 err
= fib_lookup(net
, &fl4
, &res
);
645 if (res
.type
!= RTN_UNICAST
&& res
.type
!= RTN_LOCAL
)
647 nh
->nh_scope
= res
.scope
;
648 nh
->nh_oif
= FIB_RES_OIF(res
);
649 nh
->nh_dev
= dev
= FIB_RES_DEV(res
);
653 err
= (dev
->flags
& IFF_UP
) ? 0 : -ENETDOWN
;
655 struct in_device
*in_dev
;
657 if (nh
->nh_flags
& (RTNH_F_PERVASIVE
| RTNH_F_ONLINK
))
662 in_dev
= inetdev_by_index(net
, nh
->nh_oif
);
666 if (!(in_dev
->dev
->flags
& IFF_UP
))
668 nh
->nh_dev
= in_dev
->dev
;
669 dev_hold(nh
->nh_dev
);
670 nh
->nh_scope
= RT_SCOPE_HOST
;
678 static inline unsigned int fib_laddr_hashfn(__be32 val
)
680 unsigned int mask
= (fib_info_hash_size
- 1);
682 return ((__force u32
)val
^
683 ((__force u32
)val
>> 7) ^
684 ((__force u32
)val
>> 14)) & mask
;
687 static struct hlist_head
*fib_info_hash_alloc(int bytes
)
689 if (bytes
<= PAGE_SIZE
)
690 return kzalloc(bytes
, GFP_KERNEL
);
692 return (struct hlist_head
*)
693 __get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
697 static void fib_info_hash_free(struct hlist_head
*hash
, int bytes
)
702 if (bytes
<= PAGE_SIZE
)
705 free_pages((unsigned long) hash
, get_order(bytes
));
708 static void fib_info_hash_move(struct hlist_head
*new_info_hash
,
709 struct hlist_head
*new_laddrhash
,
710 unsigned int new_size
)
712 struct hlist_head
*old_info_hash
, *old_laddrhash
;
713 unsigned int old_size
= fib_info_hash_size
;
714 unsigned int i
, bytes
;
716 spin_lock_bh(&fib_info_lock
);
717 old_info_hash
= fib_info_hash
;
718 old_laddrhash
= fib_info_laddrhash
;
719 fib_info_hash_size
= new_size
;
721 for (i
= 0; i
< old_size
; i
++) {
722 struct hlist_head
*head
= &fib_info_hash
[i
];
723 struct hlist_node
*node
, *n
;
726 hlist_for_each_entry_safe(fi
, node
, n
, head
, fib_hash
) {
727 struct hlist_head
*dest
;
728 unsigned int new_hash
;
730 hlist_del(&fi
->fib_hash
);
732 new_hash
= fib_info_hashfn(fi
);
733 dest
= &new_info_hash
[new_hash
];
734 hlist_add_head(&fi
->fib_hash
, dest
);
737 fib_info_hash
= new_info_hash
;
739 for (i
= 0; i
< old_size
; i
++) {
740 struct hlist_head
*lhead
= &fib_info_laddrhash
[i
];
741 struct hlist_node
*node
, *n
;
744 hlist_for_each_entry_safe(fi
, node
, n
, lhead
, fib_lhash
) {
745 struct hlist_head
*ldest
;
746 unsigned int new_hash
;
748 hlist_del(&fi
->fib_lhash
);
750 new_hash
= fib_laddr_hashfn(fi
->fib_prefsrc
);
751 ldest
= &new_laddrhash
[new_hash
];
752 hlist_add_head(&fi
->fib_lhash
, ldest
);
755 fib_info_laddrhash
= new_laddrhash
;
757 spin_unlock_bh(&fib_info_lock
);
759 bytes
= old_size
* sizeof(struct hlist_head
*);
760 fib_info_hash_free(old_info_hash
, bytes
);
761 fib_info_hash_free(old_laddrhash
, bytes
);
764 __be32
fib_info_update_nh_saddr(struct net
*net
, struct fib_nh
*nh
)
766 nh
->nh_saddr
= inet_select_addr(nh
->nh_dev
,
768 nh
->nh_parent
->fib_scope
);
769 nh
->nh_saddr_genid
= atomic_read(&net
->ipv4
.dev_addr_genid
);
774 struct fib_info
*fib_create_info(struct fib_config
*cfg
)
777 struct fib_info
*fi
= NULL
;
778 struct fib_info
*ofi
;
780 struct net
*net
= cfg
->fc_nlinfo
.nl_net
;
782 if (cfg
->fc_type
> RTN_MAX
)
785 /* Fast check to catch the most weird cases */
786 if (fib_props
[cfg
->fc_type
].scope
> cfg
->fc_scope
)
789 #ifdef CONFIG_IP_ROUTE_MULTIPATH
791 nhs
= fib_count_nexthops(cfg
->fc_mp
, cfg
->fc_mp_len
);
798 if (fib_info_cnt
>= fib_info_hash_size
) {
799 unsigned int new_size
= fib_info_hash_size
<< 1;
800 struct hlist_head
*new_info_hash
;
801 struct hlist_head
*new_laddrhash
;
806 bytes
= new_size
* sizeof(struct hlist_head
*);
807 new_info_hash
= fib_info_hash_alloc(bytes
);
808 new_laddrhash
= fib_info_hash_alloc(bytes
);
809 if (!new_info_hash
|| !new_laddrhash
) {
810 fib_info_hash_free(new_info_hash
, bytes
);
811 fib_info_hash_free(new_laddrhash
, bytes
);
813 fib_info_hash_move(new_info_hash
, new_laddrhash
, new_size
);
815 if (!fib_info_hash_size
)
819 fi
= kzalloc(sizeof(*fi
)+nhs
*sizeof(struct fib_nh
), GFP_KERNEL
);
823 fi
->fib_metrics
= kzalloc(sizeof(u32
) * RTAX_MAX
, GFP_KERNEL
);
824 if (!fi
->fib_metrics
)
827 fi
->fib_metrics
= (u32
*) dst_default_metrics
;
830 fi
->fib_net
= hold_net(net
);
831 fi
->fib_protocol
= cfg
->fc_protocol
;
832 fi
->fib_scope
= cfg
->fc_scope
;
833 fi
->fib_flags
= cfg
->fc_flags
;
834 fi
->fib_priority
= cfg
->fc_priority
;
835 fi
->fib_prefsrc
= cfg
->fc_prefsrc
;
838 change_nexthops(fi
) {
839 nexthop_nh
->nh_parent
= fi
;
840 nexthop_nh
->nh_pcpu_rth_output
= alloc_percpu(struct rtable __rcu
*);
841 } endfor_nexthops(fi
)
847 nla_for_each_attr(nla
, cfg
->fc_mx
, cfg
->fc_mx_len
, remaining
) {
848 int type
= nla_type(nla
);
855 val
= nla_get_u32(nla
);
856 if (type
== RTAX_ADVMSS
&& val
> 65535 - 40)
858 if (type
== RTAX_MTU
&& val
> 65535 - 15)
860 fi
->fib_metrics
[type
- 1] = val
;
866 #ifdef CONFIG_IP_ROUTE_MULTIPATH
867 err
= fib_get_nhs(fi
, cfg
->fc_mp
, cfg
->fc_mp_len
, cfg
);
870 if (cfg
->fc_oif
&& fi
->fib_nh
->nh_oif
!= cfg
->fc_oif
)
872 if (cfg
->fc_gw
&& fi
->fib_nh
->nh_gw
!= cfg
->fc_gw
)
874 #ifdef CONFIG_IP_ROUTE_CLASSID
875 if (cfg
->fc_flow
&& fi
->fib_nh
->nh_tclassid
!= cfg
->fc_flow
)
882 struct fib_nh
*nh
= fi
->fib_nh
;
884 nh
->nh_oif
= cfg
->fc_oif
;
885 nh
->nh_gw
= cfg
->fc_gw
;
886 nh
->nh_flags
= cfg
->fc_flags
;
887 #ifdef CONFIG_IP_ROUTE_CLASSID
888 nh
->nh_tclassid
= cfg
->fc_flow
;
890 fi
->fib_net
->ipv4
.fib_num_tclassid_users
++;
892 #ifdef CONFIG_IP_ROUTE_MULTIPATH
897 if (fib_props
[cfg
->fc_type
].error
) {
898 if (cfg
->fc_gw
|| cfg
->fc_oif
|| cfg
->fc_mp
)
902 switch (cfg
->fc_type
) {
914 if (cfg
->fc_scope
> RT_SCOPE_HOST
)
917 if (cfg
->fc_scope
== RT_SCOPE_HOST
) {
918 struct fib_nh
*nh
= fi
->fib_nh
;
920 /* Local address is added. */
921 if (nhs
!= 1 || nh
->nh_gw
)
923 nh
->nh_scope
= RT_SCOPE_NOWHERE
;
924 nh
->nh_dev
= dev_get_by_index(net
, fi
->fib_nh
->nh_oif
);
926 if (nh
->nh_dev
== NULL
)
929 change_nexthops(fi
) {
930 err
= fib_check_nh(cfg
, fi
, nexthop_nh
);
933 } endfor_nexthops(fi
)
936 if (fi
->fib_prefsrc
) {
937 if (cfg
->fc_type
!= RTN_LOCAL
|| !cfg
->fc_dst
||
938 fi
->fib_prefsrc
!= cfg
->fc_dst
)
939 if (inet_addr_type(net
, fi
->fib_prefsrc
) != RTN_LOCAL
)
943 change_nexthops(fi
) {
944 fib_info_update_nh_saddr(net
, nexthop_nh
);
945 } endfor_nexthops(fi
)
948 ofi
= fib_find_info(fi
);
957 atomic_inc(&fi
->fib_clntref
);
958 spin_lock_bh(&fib_info_lock
);
959 hlist_add_head(&fi
->fib_hash
,
960 &fib_info_hash
[fib_info_hashfn(fi
)]);
961 if (fi
->fib_prefsrc
) {
962 struct hlist_head
*head
;
964 head
= &fib_info_laddrhash
[fib_laddr_hashfn(fi
->fib_prefsrc
)];
965 hlist_add_head(&fi
->fib_lhash
, head
);
967 change_nexthops(fi
) {
968 struct hlist_head
*head
;
971 if (!nexthop_nh
->nh_dev
)
973 hash
= fib_devindex_hashfn(nexthop_nh
->nh_dev
->ifindex
);
974 head
= &fib_info_devhash
[hash
];
975 hlist_add_head(&nexthop_nh
->nh_hash
, head
);
976 } endfor_nexthops(fi
)
977 spin_unlock_bh(&fib_info_lock
);
992 int fib_dump_info(struct sk_buff
*skb
, u32 pid
, u32 seq
, int event
,
993 u32 tb_id
, u8 type
, __be32 dst
, int dst_len
, u8 tos
,
994 struct fib_info
*fi
, unsigned int flags
)
996 struct nlmsghdr
*nlh
;
999 nlh
= nlmsg_put(skb
, pid
, seq
, event
, sizeof(*rtm
), flags
);
1003 rtm
= nlmsg_data(nlh
);
1004 rtm
->rtm_family
= AF_INET
;
1005 rtm
->rtm_dst_len
= dst_len
;
1006 rtm
->rtm_src_len
= 0;
1009 rtm
->rtm_table
= tb_id
;
1011 rtm
->rtm_table
= RT_TABLE_COMPAT
;
1012 if (nla_put_u32(skb
, RTA_TABLE
, tb_id
))
1013 goto nla_put_failure
;
1014 rtm
->rtm_type
= type
;
1015 rtm
->rtm_flags
= fi
->fib_flags
;
1016 rtm
->rtm_scope
= fi
->fib_scope
;
1017 rtm
->rtm_protocol
= fi
->fib_protocol
;
1019 if (rtm
->rtm_dst_len
&&
1020 nla_put_be32(skb
, RTA_DST
, dst
))
1021 goto nla_put_failure
;
1022 if (fi
->fib_priority
&&
1023 nla_put_u32(skb
, RTA_PRIORITY
, fi
->fib_priority
))
1024 goto nla_put_failure
;
1025 if (rtnetlink_put_metrics(skb
, fi
->fib_metrics
) < 0)
1026 goto nla_put_failure
;
1028 if (fi
->fib_prefsrc
&&
1029 nla_put_be32(skb
, RTA_PREFSRC
, fi
->fib_prefsrc
))
1030 goto nla_put_failure
;
1031 if (fi
->fib_nhs
== 1) {
1032 if (fi
->fib_nh
->nh_gw
&&
1033 nla_put_be32(skb
, RTA_GATEWAY
, fi
->fib_nh
->nh_gw
))
1034 goto nla_put_failure
;
1035 if (fi
->fib_nh
->nh_oif
&&
1036 nla_put_u32(skb
, RTA_OIF
, fi
->fib_nh
->nh_oif
))
1037 goto nla_put_failure
;
1038 #ifdef CONFIG_IP_ROUTE_CLASSID
1039 if (fi
->fib_nh
[0].nh_tclassid
&&
1040 nla_put_u32(skb
, RTA_FLOW
, fi
->fib_nh
[0].nh_tclassid
))
1041 goto nla_put_failure
;
1044 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1045 if (fi
->fib_nhs
> 1) {
1046 struct rtnexthop
*rtnh
;
1049 mp
= nla_nest_start(skb
, RTA_MULTIPATH
);
1051 goto nla_put_failure
;
1054 rtnh
= nla_reserve_nohdr(skb
, sizeof(*rtnh
));
1056 goto nla_put_failure
;
1058 rtnh
->rtnh_flags
= nh
->nh_flags
& 0xFF;
1059 rtnh
->rtnh_hops
= nh
->nh_weight
- 1;
1060 rtnh
->rtnh_ifindex
= nh
->nh_oif
;
1063 nla_put_be32(skb
, RTA_GATEWAY
, nh
->nh_gw
))
1064 goto nla_put_failure
;
1065 #ifdef CONFIG_IP_ROUTE_CLASSID
1066 if (nh
->nh_tclassid
&&
1067 nla_put_u32(skb
, RTA_FLOW
, nh
->nh_tclassid
))
1068 goto nla_put_failure
;
1070 /* length of rtnetlink header + attributes */
1071 rtnh
->rtnh_len
= nlmsg_get_pos(skb
) - (void *) rtnh
;
1072 } endfor_nexthops(fi
);
1074 nla_nest_end(skb
, mp
);
1077 return 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 hlist_node
*node
;
1096 struct fib_info
*fi
;
1098 if (fib_info_laddrhash
== NULL
|| local
== 0)
1101 hlist_for_each_entry(fi
, node
, head
, fib_lhash
) {
1102 if (!net_eq(fi
->fib_net
, net
))
1104 if (fi
->fib_prefsrc
== local
) {
1105 fi
->fib_flags
|= RTNH_F_DEAD
;
1112 int fib_sync_down_dev(struct net_device
*dev
, int force
)
1115 int scope
= RT_SCOPE_NOWHERE
;
1116 struct fib_info
*prev_fi
= NULL
;
1117 unsigned int hash
= fib_devindex_hashfn(dev
->ifindex
);
1118 struct hlist_head
*head
= &fib_info_devhash
[hash
];
1119 struct hlist_node
*node
;
1125 hlist_for_each_entry(nh
, node
, head
, nh_hash
) {
1126 struct fib_info
*fi
= nh
->nh_parent
;
1129 BUG_ON(!fi
->fib_nhs
);
1130 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1134 change_nexthops(fi
) {
1135 if (nexthop_nh
->nh_flags
& RTNH_F_DEAD
)
1137 else if (nexthop_nh
->nh_dev
== dev
&&
1138 nexthop_nh
->nh_scope
!= scope
) {
1139 nexthop_nh
->nh_flags
|= RTNH_F_DEAD
;
1140 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1141 spin_lock_bh(&fib_multipath_lock
);
1142 fi
->fib_power
-= nexthop_nh
->nh_power
;
1143 nexthop_nh
->nh_power
= 0;
1144 spin_unlock_bh(&fib_multipath_lock
);
1148 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1149 if (force
> 1 && nexthop_nh
->nh_dev
== dev
) {
1154 } endfor_nexthops(fi
)
1155 if (dead
== fi
->fib_nhs
) {
1156 fi
->fib_flags
|= RTNH_F_DEAD
;
1164 /* Must be invoked inside of an RCU protected region. */
1165 void fib_select_default(struct fib_result
*res
)
1167 struct fib_info
*fi
= NULL
, *last_resort
= NULL
;
1168 struct list_head
*fa_head
= res
->fa_head
;
1169 struct fib_table
*tb
= res
->table
;
1170 int order
= -1, last_idx
= -1;
1171 struct fib_alias
*fa
;
1173 list_for_each_entry_rcu(fa
, fa_head
, fa_list
) {
1174 struct fib_info
*next_fi
= fa
->fa_info
;
1176 if (next_fi
->fib_scope
!= res
->scope
||
1177 fa
->fa_type
!= RTN_UNICAST
)
1180 if (next_fi
->fib_priority
> res
->fi
->fib_priority
)
1182 if (!next_fi
->fib_nh
[0].nh_gw
||
1183 next_fi
->fib_nh
[0].nh_scope
!= RT_SCOPE_LINK
)
1186 fib_alias_accessed(fa
);
1189 if (next_fi
!= res
->fi
)
1191 } else if (!fib_detect_death(fi
, order
, &last_resort
,
1192 &last_idx
, tb
->tb_default
)) {
1193 fib_result_assign(res
, fi
);
1194 tb
->tb_default
= order
;
1201 if (order
<= 0 || fi
== NULL
) {
1202 tb
->tb_default
= -1;
1206 if (!fib_detect_death(fi
, order
, &last_resort
, &last_idx
,
1208 fib_result_assign(res
, fi
);
1209 tb
->tb_default
= order
;
1214 fib_result_assign(res
, last_resort
);
1215 tb
->tb_default
= last_idx
;
1220 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1223 * Dead device goes up. We wake up dead nexthops.
1224 * It takes sense only on multipath routes.
1226 int fib_sync_up(struct net_device
*dev
)
1228 struct fib_info
*prev_fi
;
1230 struct hlist_head
*head
;
1231 struct hlist_node
*node
;
1235 if (!(dev
->flags
& IFF_UP
))
1239 hash
= fib_devindex_hashfn(dev
->ifindex
);
1240 head
= &fib_info_devhash
[hash
];
1243 hlist_for_each_entry(nh
, node
, head
, nh_hash
) {
1244 struct fib_info
*fi
= nh
->nh_parent
;
1247 BUG_ON(!fi
->fib_nhs
);
1248 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1253 change_nexthops(fi
) {
1254 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1258 if (nexthop_nh
->nh_dev
== NULL
||
1259 !(nexthop_nh
->nh_dev
->flags
& IFF_UP
))
1261 if (nexthop_nh
->nh_dev
!= dev
||
1262 !__in_dev_get_rtnl(dev
))
1265 spin_lock_bh(&fib_multipath_lock
);
1266 nexthop_nh
->nh_power
= 0;
1267 nexthop_nh
->nh_flags
&= ~RTNH_F_DEAD
;
1268 spin_unlock_bh(&fib_multipath_lock
);
1269 } endfor_nexthops(fi
)
1272 fi
->fib_flags
&= ~RTNH_F_DEAD
;
1281 * The algorithm is suboptimal, but it provides really
1282 * fair weighted route distribution.
1284 void fib_select_multipath(struct fib_result
*res
)
1286 struct fib_info
*fi
= res
->fi
;
1289 spin_lock_bh(&fib_multipath_lock
);
1290 if (fi
->fib_power
<= 0) {
1292 change_nexthops(fi
) {
1293 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1294 power
+= nexthop_nh
->nh_weight
;
1295 nexthop_nh
->nh_power
= nexthop_nh
->nh_weight
;
1297 } endfor_nexthops(fi
);
1298 fi
->fib_power
= power
;
1300 spin_unlock_bh(&fib_multipath_lock
);
1301 /* Race condition: route has just become dead. */
1308 /* w should be random number [0..fi->fib_power-1],
1309 * it is pretty bad approximation.
1312 w
= jiffies
% fi
->fib_power
;
1314 change_nexthops(fi
) {
1315 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
) &&
1316 nexthop_nh
->nh_power
) {
1317 w
-= nexthop_nh
->nh_power
;
1319 nexthop_nh
->nh_power
--;
1321 res
->nh_sel
= nhsel
;
1322 spin_unlock_bh(&fib_multipath_lock
);
1326 } endfor_nexthops(fi
);
1328 /* Race condition: route has just become dead. */
1330 spin_unlock_bh(&fib_multipath_lock
);