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_input
);
173 rt_fibinfo_free(&fnhe
->fnhe_rth_output
);
183 static void rt_fibinfo_free_cpus(struct rtable __rcu
* __percpu
*rtp
)
190 for_each_possible_cpu(cpu
) {
193 rt
= rcu_dereference_protected(*per_cpu_ptr(rtp
, cpu
), 1);
200 /* Release a nexthop info record */
201 static void free_fib_info_rcu(struct rcu_head
*head
)
203 struct fib_info
*fi
= container_of(head
, struct fib_info
, rcu
);
205 change_nexthops(fi
) {
206 if (nexthop_nh
->nh_dev
)
207 dev_put(nexthop_nh
->nh_dev
);
208 if (nexthop_nh
->nh_exceptions
)
209 free_nh_exceptions(nexthop_nh
);
210 rt_fibinfo_free_cpus(nexthop_nh
->nh_pcpu_rth_output
);
211 rt_fibinfo_free(&nexthop_nh
->nh_rth_input
);
212 } endfor_nexthops(fi
);
214 release_net(fi
->fib_net
);
215 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
)
216 kfree(fi
->fib_metrics
);
220 void free_fib_info(struct fib_info
*fi
)
222 if (fi
->fib_dead
== 0) {
223 pr_warn("Freeing alive fib_info %p\n", fi
);
227 #ifdef CONFIG_IP_ROUTE_CLASSID
228 change_nexthops(fi
) {
229 if (nexthop_nh
->nh_tclassid
)
230 fi
->fib_net
->ipv4
.fib_num_tclassid_users
--;
231 } endfor_nexthops(fi
);
233 call_rcu(&fi
->rcu
, free_fib_info_rcu
);
236 void fib_release_info(struct fib_info
*fi
)
238 spin_lock_bh(&fib_info_lock
);
239 if (fi
&& --fi
->fib_treeref
== 0) {
240 hlist_del(&fi
->fib_hash
);
242 hlist_del(&fi
->fib_lhash
);
243 change_nexthops(fi
) {
244 if (!nexthop_nh
->nh_dev
)
246 hlist_del(&nexthop_nh
->nh_hash
);
247 } endfor_nexthops(fi
)
251 spin_unlock_bh(&fib_info_lock
);
254 static inline int nh_comp(const struct fib_info
*fi
, const struct fib_info
*ofi
)
256 const struct fib_nh
*onh
= ofi
->fib_nh
;
259 if (nh
->nh_oif
!= onh
->nh_oif
||
260 nh
->nh_gw
!= onh
->nh_gw
||
261 nh
->nh_scope
!= onh
->nh_scope
||
262 #ifdef CONFIG_IP_ROUTE_MULTIPATH
263 nh
->nh_weight
!= onh
->nh_weight
||
265 #ifdef CONFIG_IP_ROUTE_CLASSID
266 nh
->nh_tclassid
!= onh
->nh_tclassid
||
268 ((nh
->nh_flags
^ onh
->nh_flags
) & ~RTNH_F_DEAD
))
271 } endfor_nexthops(fi
);
275 static inline unsigned int fib_devindex_hashfn(unsigned int val
)
277 unsigned int mask
= DEVINDEX_HASHSIZE
- 1;
280 (val
>> DEVINDEX_HASHBITS
) ^
281 (val
>> (DEVINDEX_HASHBITS
* 2))) & mask
;
284 static inline unsigned int fib_info_hashfn(const struct fib_info
*fi
)
286 unsigned int mask
= (fib_info_hash_size
- 1);
287 unsigned int val
= fi
->fib_nhs
;
289 val
^= (fi
->fib_protocol
<< 8) | fi
->fib_scope
;
290 val
^= (__force u32
)fi
->fib_prefsrc
;
291 val
^= fi
->fib_priority
;
293 val
^= fib_devindex_hashfn(nh
->nh_oif
);
294 } endfor_nexthops(fi
)
296 return (val
^ (val
>> 7) ^ (val
>> 12)) & mask
;
299 static struct fib_info
*fib_find_info(const struct fib_info
*nfi
)
301 struct hlist_head
*head
;
305 hash
= fib_info_hashfn(nfi
);
306 head
= &fib_info_hash
[hash
];
308 hlist_for_each_entry(fi
, 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 nfi
->fib_type
== fi
->fib_type
&&
318 memcmp(nfi
->fib_metrics
, fi
->fib_metrics
,
319 sizeof(u32
) * RTAX_MAX
) == 0 &&
320 ((nfi
->fib_flags
^ fi
->fib_flags
) & ~RTNH_F_DEAD
) == 0 &&
321 (nfi
->fib_nhs
== 0 || nh_comp(fi
, nfi
) == 0))
328 /* Check, that the gateway is already configured.
329 * Used only by redirect accept routine.
331 int ip_fib_check_default(__be32 gw
, struct net_device
*dev
)
333 struct hlist_head
*head
;
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
, 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
, const 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
->portid
, 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
->portid
, 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 static int fib_detect_death(struct fib_info
*fi
, int order
,
430 struct fib_info
**last_resort
, int *last_idx
,
434 int state
= NUD_NONE
;
436 n
= neigh_lookup(&arp_tbl
, &fi
->fib_nh
[0].nh_gw
, fi
->fib_dev
);
438 state
= n
->nud_state
;
441 if (state
== NUD_REACHABLE
)
443 if ((state
& NUD_VALID
) && order
!= dflt
)
445 if ((state
& NUD_VALID
) ||
446 (*last_idx
< 0 && order
> dflt
)) {
453 #ifdef CONFIG_IP_ROUTE_MULTIPATH
455 static int fib_count_nexthops(struct rtnexthop
*rtnh
, int remaining
)
459 while (rtnh_ok(rtnh
, remaining
)) {
461 rtnh
= rtnh_next(rtnh
, &remaining
);
464 /* leftover implies invalid nexthop configuration, discard it */
465 return remaining
> 0 ? 0 : nhs
;
468 static int fib_get_nhs(struct fib_info
*fi
, struct rtnexthop
*rtnh
,
469 int remaining
, struct fib_config
*cfg
)
471 change_nexthops(fi
) {
474 if (!rtnh_ok(rtnh
, remaining
))
477 nexthop_nh
->nh_flags
=
478 (cfg
->fc_flags
& ~0xFF) | rtnh
->rtnh_flags
;
479 nexthop_nh
->nh_oif
= rtnh
->rtnh_ifindex
;
480 nexthop_nh
->nh_weight
= rtnh
->rtnh_hops
+ 1;
482 attrlen
= rtnh_attrlen(rtnh
);
484 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
486 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
487 nexthop_nh
->nh_gw
= nla
? nla_get_be32(nla
) : 0;
488 #ifdef CONFIG_IP_ROUTE_CLASSID
489 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
490 nexthop_nh
->nh_tclassid
= nla
? nla_get_u32(nla
) : 0;
491 if (nexthop_nh
->nh_tclassid
)
492 fi
->fib_net
->ipv4
.fib_num_tclassid_users
++;
496 rtnh
= rtnh_next(rtnh
, &remaining
);
497 } endfor_nexthops(fi
);
504 int fib_nh_match(struct fib_config
*cfg
, struct fib_info
*fi
)
506 #ifdef CONFIG_IP_ROUTE_MULTIPATH
507 struct rtnexthop
*rtnh
;
511 if (cfg
->fc_priority
&& cfg
->fc_priority
!= fi
->fib_priority
)
514 if (cfg
->fc_oif
|| cfg
->fc_gw
) {
515 if ((!cfg
->fc_oif
|| cfg
->fc_oif
== fi
->fib_nh
->nh_oif
) &&
516 (!cfg
->fc_gw
|| cfg
->fc_gw
== fi
->fib_nh
->nh_gw
))
521 #ifdef CONFIG_IP_ROUTE_MULTIPATH
522 if (cfg
->fc_mp
== NULL
)
526 remaining
= cfg
->fc_mp_len
;
531 if (!rtnh_ok(rtnh
, remaining
))
534 if (rtnh
->rtnh_ifindex
&& rtnh
->rtnh_ifindex
!= nh
->nh_oif
)
537 attrlen
= rtnh_attrlen(rtnh
);
539 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
541 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
542 if (nla
&& nla_get_be32(nla
) != nh
->nh_gw
)
544 #ifdef CONFIG_IP_ROUTE_CLASSID
545 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
546 if (nla
&& nla_get_u32(nla
) != nh
->nh_tclassid
)
551 rtnh
= rtnh_next(rtnh
, &remaining
);
552 } endfor_nexthops(fi
);
562 * Semantics of nexthop is very messy by historical reasons.
563 * We have to take into account, that:
564 * a) gateway can be actually local interface address,
565 * so that gatewayed route is direct.
566 * b) gateway must be on-link address, possibly
567 * described not by an ifaddr, but also by a direct route.
568 * c) If both gateway and interface are specified, they should not
570 * d) If we use tunnel routes, gateway could be not on-link.
572 * Attempt to reconcile all of these (alas, self-contradictory) conditions
573 * results in pretty ugly and hairy code with obscure logic.
575 * I chose to generalized it instead, so that the size
576 * of code does not increase practically, but it becomes
578 * Every prefix is assigned a "scope" value: "host" is local address,
579 * "link" is direct route,
580 * [ ... "site" ... "interior" ... ]
581 * and "universe" is true gateway route with global meaning.
583 * Every prefix refers to a set of "nexthop"s (gw, oif),
584 * where gw must have narrower scope. This recursion stops
585 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
586 * which means that gw is forced to be on link.
588 * Code is still hairy, but now it is apparently logically
589 * consistent and very flexible. F.e. as by-product it allows
590 * to co-exists in peace independent exterior and interior
593 * Normally it looks as following.
595 * {universe prefix} -> (gw, oif) [scope link]
597 * |-> {link prefix} -> (gw, oif) [scope local]
599 * |-> {local prefix} (terminal node)
601 static int fib_check_nh(struct fib_config
*cfg
, struct fib_info
*fi
,
606 struct net_device
*dev
;
608 net
= cfg
->fc_nlinfo
.nl_net
;
610 struct fib_result res
;
612 if (nh
->nh_flags
& RTNH_F_ONLINK
) {
614 if (cfg
->fc_scope
>= RT_SCOPE_LINK
)
616 if (inet_addr_type(net
, nh
->nh_gw
) != RTN_UNICAST
)
618 dev
= __dev_get_by_index(net
, nh
->nh_oif
);
621 if (!(dev
->flags
& IFF_UP
))
625 nh
->nh_scope
= RT_SCOPE_LINK
;
630 struct flowi4 fl4
= {
632 .flowi4_scope
= cfg
->fc_scope
+ 1,
633 .flowi4_oif
= nh
->nh_oif
,
634 .flowi4_iif
= LOOPBACK_IFINDEX
,
637 /* It is not necessary, but requires a bit of thinking */
638 if (fl4
.flowi4_scope
< RT_SCOPE_LINK
)
639 fl4
.flowi4_scope
= RT_SCOPE_LINK
;
640 err
= fib_lookup(net
, &fl4
, &res
);
647 if (res
.type
!= RTN_UNICAST
&& res
.type
!= RTN_LOCAL
)
649 nh
->nh_scope
= res
.scope
;
650 nh
->nh_oif
= FIB_RES_OIF(res
);
651 nh
->nh_dev
= dev
= FIB_RES_DEV(res
);
655 err
= (dev
->flags
& IFF_UP
) ? 0 : -ENETDOWN
;
657 struct in_device
*in_dev
;
659 if (nh
->nh_flags
& (RTNH_F_PERVASIVE
| RTNH_F_ONLINK
))
664 in_dev
= inetdev_by_index(net
, nh
->nh_oif
);
668 if (!(in_dev
->dev
->flags
& IFF_UP
))
670 nh
->nh_dev
= in_dev
->dev
;
671 dev_hold(nh
->nh_dev
);
672 nh
->nh_scope
= RT_SCOPE_HOST
;
680 static inline unsigned int fib_laddr_hashfn(__be32 val
)
682 unsigned int mask
= (fib_info_hash_size
- 1);
684 return ((__force u32
)val
^
685 ((__force u32
)val
>> 7) ^
686 ((__force u32
)val
>> 14)) & mask
;
689 static struct hlist_head
*fib_info_hash_alloc(int bytes
)
691 if (bytes
<= PAGE_SIZE
)
692 return kzalloc(bytes
, GFP_KERNEL
);
694 return (struct hlist_head
*)
695 __get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
699 static void fib_info_hash_free(struct hlist_head
*hash
, int bytes
)
704 if (bytes
<= PAGE_SIZE
)
707 free_pages((unsigned long) hash
, get_order(bytes
));
710 static void fib_info_hash_move(struct hlist_head
*new_info_hash
,
711 struct hlist_head
*new_laddrhash
,
712 unsigned int new_size
)
714 struct hlist_head
*old_info_hash
, *old_laddrhash
;
715 unsigned int old_size
= fib_info_hash_size
;
716 unsigned int i
, bytes
;
718 spin_lock_bh(&fib_info_lock
);
719 old_info_hash
= fib_info_hash
;
720 old_laddrhash
= fib_info_laddrhash
;
721 fib_info_hash_size
= new_size
;
723 for (i
= 0; i
< old_size
; i
++) {
724 struct hlist_head
*head
= &fib_info_hash
[i
];
725 struct hlist_node
*n
;
728 hlist_for_each_entry_safe(fi
, n
, head
, fib_hash
) {
729 struct hlist_head
*dest
;
730 unsigned int new_hash
;
732 hlist_del(&fi
->fib_hash
);
734 new_hash
= fib_info_hashfn(fi
);
735 dest
= &new_info_hash
[new_hash
];
736 hlist_add_head(&fi
->fib_hash
, dest
);
739 fib_info_hash
= new_info_hash
;
741 for (i
= 0; i
< old_size
; i
++) {
742 struct hlist_head
*lhead
= &fib_info_laddrhash
[i
];
743 struct hlist_node
*n
;
746 hlist_for_each_entry_safe(fi
, n
, lhead
, fib_lhash
) {
747 struct hlist_head
*ldest
;
748 unsigned int new_hash
;
750 hlist_del(&fi
->fib_lhash
);
752 new_hash
= fib_laddr_hashfn(fi
->fib_prefsrc
);
753 ldest
= &new_laddrhash
[new_hash
];
754 hlist_add_head(&fi
->fib_lhash
, ldest
);
757 fib_info_laddrhash
= new_laddrhash
;
759 spin_unlock_bh(&fib_info_lock
);
761 bytes
= old_size
* sizeof(struct hlist_head
*);
762 fib_info_hash_free(old_info_hash
, bytes
);
763 fib_info_hash_free(old_laddrhash
, bytes
);
766 __be32
fib_info_update_nh_saddr(struct net
*net
, struct fib_nh
*nh
)
768 nh
->nh_saddr
= inet_select_addr(nh
->nh_dev
,
770 nh
->nh_parent
->fib_scope
);
771 nh
->nh_saddr_genid
= atomic_read(&net
->ipv4
.dev_addr_genid
);
776 struct fib_info
*fib_create_info(struct fib_config
*cfg
)
779 struct fib_info
*fi
= NULL
;
780 struct fib_info
*ofi
;
782 struct net
*net
= cfg
->fc_nlinfo
.nl_net
;
784 if (cfg
->fc_type
> RTN_MAX
)
787 /* Fast check to catch the most weird cases */
788 if (fib_props
[cfg
->fc_type
].scope
> cfg
->fc_scope
)
791 #ifdef CONFIG_IP_ROUTE_MULTIPATH
793 nhs
= fib_count_nexthops(cfg
->fc_mp
, cfg
->fc_mp_len
);
800 if (fib_info_cnt
>= fib_info_hash_size
) {
801 unsigned int new_size
= fib_info_hash_size
<< 1;
802 struct hlist_head
*new_info_hash
;
803 struct hlist_head
*new_laddrhash
;
808 bytes
= new_size
* sizeof(struct hlist_head
*);
809 new_info_hash
= fib_info_hash_alloc(bytes
);
810 new_laddrhash
= fib_info_hash_alloc(bytes
);
811 if (!new_info_hash
|| !new_laddrhash
) {
812 fib_info_hash_free(new_info_hash
, bytes
);
813 fib_info_hash_free(new_laddrhash
, bytes
);
815 fib_info_hash_move(new_info_hash
, new_laddrhash
, new_size
);
817 if (!fib_info_hash_size
)
821 fi
= kzalloc(sizeof(*fi
)+nhs
*sizeof(struct fib_nh
), GFP_KERNEL
);
826 fi
->fib_metrics
= kzalloc(sizeof(u32
) * RTAX_MAX
, GFP_KERNEL
);
827 if (!fi
->fib_metrics
)
830 fi
->fib_metrics
= (u32
*) dst_default_metrics
;
832 fi
->fib_net
= hold_net(net
);
833 fi
->fib_protocol
= cfg
->fc_protocol
;
834 fi
->fib_scope
= cfg
->fc_scope
;
835 fi
->fib_flags
= cfg
->fc_flags
;
836 fi
->fib_priority
= cfg
->fc_priority
;
837 fi
->fib_prefsrc
= cfg
->fc_prefsrc
;
838 fi
->fib_type
= cfg
->fc_type
;
841 change_nexthops(fi
) {
842 nexthop_nh
->nh_parent
= fi
;
843 nexthop_nh
->nh_pcpu_rth_output
= alloc_percpu(struct rtable __rcu
*);
844 if (!nexthop_nh
->nh_pcpu_rth_output
)
846 } endfor_nexthops(fi
)
852 nla_for_each_attr(nla
, cfg
->fc_mx
, cfg
->fc_mx_len
, remaining
) {
853 int type
= nla_type(nla
);
860 val
= nla_get_u32(nla
);
861 if (type
== RTAX_ADVMSS
&& val
> 65535 - 40)
863 if (type
== RTAX_MTU
&& val
> 65535 - 15)
865 fi
->fib_metrics
[type
- 1] = val
;
871 #ifdef CONFIG_IP_ROUTE_MULTIPATH
872 err
= fib_get_nhs(fi
, cfg
->fc_mp
, cfg
->fc_mp_len
, cfg
);
875 if (cfg
->fc_oif
&& fi
->fib_nh
->nh_oif
!= cfg
->fc_oif
)
877 if (cfg
->fc_gw
&& fi
->fib_nh
->nh_gw
!= cfg
->fc_gw
)
879 #ifdef CONFIG_IP_ROUTE_CLASSID
880 if (cfg
->fc_flow
&& fi
->fib_nh
->nh_tclassid
!= cfg
->fc_flow
)
887 struct fib_nh
*nh
= fi
->fib_nh
;
889 nh
->nh_oif
= cfg
->fc_oif
;
890 nh
->nh_gw
= cfg
->fc_gw
;
891 nh
->nh_flags
= cfg
->fc_flags
;
892 #ifdef CONFIG_IP_ROUTE_CLASSID
893 nh
->nh_tclassid
= cfg
->fc_flow
;
895 fi
->fib_net
->ipv4
.fib_num_tclassid_users
++;
897 #ifdef CONFIG_IP_ROUTE_MULTIPATH
902 if (fib_props
[cfg
->fc_type
].error
) {
903 if (cfg
->fc_gw
|| cfg
->fc_oif
|| cfg
->fc_mp
)
907 switch (cfg
->fc_type
) {
919 if (cfg
->fc_scope
> RT_SCOPE_HOST
)
922 if (cfg
->fc_scope
== RT_SCOPE_HOST
) {
923 struct fib_nh
*nh
= fi
->fib_nh
;
925 /* Local address is added. */
926 if (nhs
!= 1 || nh
->nh_gw
)
928 nh
->nh_scope
= RT_SCOPE_NOWHERE
;
929 nh
->nh_dev
= dev_get_by_index(net
, fi
->fib_nh
->nh_oif
);
931 if (nh
->nh_dev
== NULL
)
934 change_nexthops(fi
) {
935 err
= fib_check_nh(cfg
, fi
, nexthop_nh
);
938 } endfor_nexthops(fi
)
941 if (fi
->fib_prefsrc
) {
942 if (cfg
->fc_type
!= RTN_LOCAL
|| !cfg
->fc_dst
||
943 fi
->fib_prefsrc
!= cfg
->fc_dst
)
944 if (inet_addr_type(net
, fi
->fib_prefsrc
) != RTN_LOCAL
)
948 change_nexthops(fi
) {
949 fib_info_update_nh_saddr(net
, nexthop_nh
);
950 } endfor_nexthops(fi
)
953 ofi
= fib_find_info(fi
);
962 atomic_inc(&fi
->fib_clntref
);
963 spin_lock_bh(&fib_info_lock
);
964 hlist_add_head(&fi
->fib_hash
,
965 &fib_info_hash
[fib_info_hashfn(fi
)]);
966 if (fi
->fib_prefsrc
) {
967 struct hlist_head
*head
;
969 head
= &fib_info_laddrhash
[fib_laddr_hashfn(fi
->fib_prefsrc
)];
970 hlist_add_head(&fi
->fib_lhash
, head
);
972 change_nexthops(fi
) {
973 struct hlist_head
*head
;
976 if (!nexthop_nh
->nh_dev
)
978 hash
= fib_devindex_hashfn(nexthop_nh
->nh_dev
->ifindex
);
979 head
= &fib_info_devhash
[hash
];
980 hlist_add_head(&nexthop_nh
->nh_hash
, head
);
981 } endfor_nexthops(fi
)
982 spin_unlock_bh(&fib_info_lock
);
997 int fib_dump_info(struct sk_buff
*skb
, u32 portid
, u32 seq
, int event
,
998 u32 tb_id
, u8 type
, __be32 dst
, int dst_len
, u8 tos
,
999 struct fib_info
*fi
, unsigned int flags
)
1001 struct nlmsghdr
*nlh
;
1004 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*rtm
), flags
);
1008 rtm
= nlmsg_data(nlh
);
1009 rtm
->rtm_family
= AF_INET
;
1010 rtm
->rtm_dst_len
= dst_len
;
1011 rtm
->rtm_src_len
= 0;
1014 rtm
->rtm_table
= tb_id
;
1016 rtm
->rtm_table
= RT_TABLE_COMPAT
;
1017 if (nla_put_u32(skb
, RTA_TABLE
, tb_id
))
1018 goto nla_put_failure
;
1019 rtm
->rtm_type
= type
;
1020 rtm
->rtm_flags
= fi
->fib_flags
;
1021 rtm
->rtm_scope
= fi
->fib_scope
;
1022 rtm
->rtm_protocol
= fi
->fib_protocol
;
1024 if (rtm
->rtm_dst_len
&&
1025 nla_put_be32(skb
, RTA_DST
, dst
))
1026 goto nla_put_failure
;
1027 if (fi
->fib_priority
&&
1028 nla_put_u32(skb
, RTA_PRIORITY
, fi
->fib_priority
))
1029 goto nla_put_failure
;
1030 if (rtnetlink_put_metrics(skb
, fi
->fib_metrics
) < 0)
1031 goto nla_put_failure
;
1033 if (fi
->fib_prefsrc
&&
1034 nla_put_be32(skb
, RTA_PREFSRC
, fi
->fib_prefsrc
))
1035 goto nla_put_failure
;
1036 if (fi
->fib_nhs
== 1) {
1037 if (fi
->fib_nh
->nh_gw
&&
1038 nla_put_be32(skb
, RTA_GATEWAY
, fi
->fib_nh
->nh_gw
))
1039 goto nla_put_failure
;
1040 if (fi
->fib_nh
->nh_oif
&&
1041 nla_put_u32(skb
, RTA_OIF
, fi
->fib_nh
->nh_oif
))
1042 goto nla_put_failure
;
1043 #ifdef CONFIG_IP_ROUTE_CLASSID
1044 if (fi
->fib_nh
[0].nh_tclassid
&&
1045 nla_put_u32(skb
, RTA_FLOW
, fi
->fib_nh
[0].nh_tclassid
))
1046 goto nla_put_failure
;
1049 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1050 if (fi
->fib_nhs
> 1) {
1051 struct rtnexthop
*rtnh
;
1054 mp
= nla_nest_start(skb
, RTA_MULTIPATH
);
1056 goto nla_put_failure
;
1059 rtnh
= nla_reserve_nohdr(skb
, sizeof(*rtnh
));
1061 goto nla_put_failure
;
1063 rtnh
->rtnh_flags
= nh
->nh_flags
& 0xFF;
1064 rtnh
->rtnh_hops
= nh
->nh_weight
- 1;
1065 rtnh
->rtnh_ifindex
= nh
->nh_oif
;
1068 nla_put_be32(skb
, RTA_GATEWAY
, nh
->nh_gw
))
1069 goto nla_put_failure
;
1070 #ifdef CONFIG_IP_ROUTE_CLASSID
1071 if (nh
->nh_tclassid
&&
1072 nla_put_u32(skb
, RTA_FLOW
, nh
->nh_tclassid
))
1073 goto nla_put_failure
;
1075 /* length of rtnetlink header + attributes */
1076 rtnh
->rtnh_len
= nlmsg_get_pos(skb
) - (void *) rtnh
;
1077 } endfor_nexthops(fi
);
1079 nla_nest_end(skb
, mp
);
1082 return nlmsg_end(skb
, nlh
);
1085 nlmsg_cancel(skb
, nlh
);
1091 * - local address disappeared -> we must delete all the entries
1093 * - device went down -> we must shutdown all nexthops going via it.
1095 int fib_sync_down_addr(struct net
*net
, __be32 local
)
1098 unsigned int hash
= fib_laddr_hashfn(local
);
1099 struct hlist_head
*head
= &fib_info_laddrhash
[hash
];
1100 struct fib_info
*fi
;
1102 if (fib_info_laddrhash
== NULL
|| local
== 0)
1105 hlist_for_each_entry(fi
, head
, fib_lhash
) {
1106 if (!net_eq(fi
->fib_net
, net
))
1108 if (fi
->fib_prefsrc
== local
) {
1109 fi
->fib_flags
|= RTNH_F_DEAD
;
1116 int fib_sync_down_dev(struct net_device
*dev
, int force
)
1119 int scope
= RT_SCOPE_NOWHERE
;
1120 struct fib_info
*prev_fi
= NULL
;
1121 unsigned int hash
= fib_devindex_hashfn(dev
->ifindex
);
1122 struct hlist_head
*head
= &fib_info_devhash
[hash
];
1128 hlist_for_each_entry(nh
, head
, nh_hash
) {
1129 struct fib_info
*fi
= nh
->nh_parent
;
1132 BUG_ON(!fi
->fib_nhs
);
1133 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1137 change_nexthops(fi
) {
1138 if (nexthop_nh
->nh_flags
& RTNH_F_DEAD
)
1140 else if (nexthop_nh
->nh_dev
== dev
&&
1141 nexthop_nh
->nh_scope
!= scope
) {
1142 nexthop_nh
->nh_flags
|= RTNH_F_DEAD
;
1143 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1144 spin_lock_bh(&fib_multipath_lock
);
1145 fi
->fib_power
-= nexthop_nh
->nh_power
;
1146 nexthop_nh
->nh_power
= 0;
1147 spin_unlock_bh(&fib_multipath_lock
);
1151 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1152 if (force
> 1 && nexthop_nh
->nh_dev
== dev
) {
1157 } endfor_nexthops(fi
)
1158 if (dead
== fi
->fib_nhs
) {
1159 fi
->fib_flags
|= RTNH_F_DEAD
;
1167 /* Must be invoked inside of an RCU protected region. */
1168 void fib_select_default(struct fib_result
*res
)
1170 struct fib_info
*fi
= NULL
, *last_resort
= NULL
;
1171 struct list_head
*fa_head
= res
->fa_head
;
1172 struct fib_table
*tb
= res
->table
;
1173 int order
= -1, last_idx
= -1;
1174 struct fib_alias
*fa
;
1176 list_for_each_entry_rcu(fa
, fa_head
, fa_list
) {
1177 struct fib_info
*next_fi
= fa
->fa_info
;
1179 if (next_fi
->fib_scope
!= res
->scope
||
1180 fa
->fa_type
!= RTN_UNICAST
)
1183 if (next_fi
->fib_priority
> res
->fi
->fib_priority
)
1185 if (!next_fi
->fib_nh
[0].nh_gw
||
1186 next_fi
->fib_nh
[0].nh_scope
!= RT_SCOPE_LINK
)
1189 fib_alias_accessed(fa
);
1192 if (next_fi
!= res
->fi
)
1194 } else if (!fib_detect_death(fi
, order
, &last_resort
,
1195 &last_idx
, tb
->tb_default
)) {
1196 fib_result_assign(res
, fi
);
1197 tb
->tb_default
= order
;
1204 if (order
<= 0 || fi
== NULL
) {
1205 tb
->tb_default
= -1;
1209 if (!fib_detect_death(fi
, order
, &last_resort
, &last_idx
,
1211 fib_result_assign(res
, fi
);
1212 tb
->tb_default
= order
;
1217 fib_result_assign(res
, last_resort
);
1218 tb
->tb_default
= last_idx
;
1223 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1226 * Dead device goes up. We wake up dead nexthops.
1227 * It takes sense only on multipath routes.
1229 int fib_sync_up(struct net_device
*dev
)
1231 struct fib_info
*prev_fi
;
1233 struct hlist_head
*head
;
1237 if (!(dev
->flags
& IFF_UP
))
1241 hash
= fib_devindex_hashfn(dev
->ifindex
);
1242 head
= &fib_info_devhash
[hash
];
1245 hlist_for_each_entry(nh
, head
, nh_hash
) {
1246 struct fib_info
*fi
= nh
->nh_parent
;
1249 BUG_ON(!fi
->fib_nhs
);
1250 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1255 change_nexthops(fi
) {
1256 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1260 if (nexthop_nh
->nh_dev
== NULL
||
1261 !(nexthop_nh
->nh_dev
->flags
& IFF_UP
))
1263 if (nexthop_nh
->nh_dev
!= dev
||
1264 !__in_dev_get_rtnl(dev
))
1267 spin_lock_bh(&fib_multipath_lock
);
1268 nexthop_nh
->nh_power
= 0;
1269 nexthop_nh
->nh_flags
&= ~RTNH_F_DEAD
;
1270 spin_unlock_bh(&fib_multipath_lock
);
1271 } endfor_nexthops(fi
)
1274 fi
->fib_flags
&= ~RTNH_F_DEAD
;
1283 * The algorithm is suboptimal, but it provides really
1284 * fair weighted route distribution.
1286 void fib_select_multipath(struct fib_result
*res
)
1288 struct fib_info
*fi
= res
->fi
;
1291 spin_lock_bh(&fib_multipath_lock
);
1292 if (fi
->fib_power
<= 0) {
1294 change_nexthops(fi
) {
1295 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
)) {
1296 power
+= nexthop_nh
->nh_weight
;
1297 nexthop_nh
->nh_power
= nexthop_nh
->nh_weight
;
1299 } endfor_nexthops(fi
);
1300 fi
->fib_power
= power
;
1302 spin_unlock_bh(&fib_multipath_lock
);
1303 /* Race condition: route has just become dead. */
1310 /* w should be random number [0..fi->fib_power-1],
1311 * it is pretty bad approximation.
1314 w
= jiffies
% fi
->fib_power
;
1316 change_nexthops(fi
) {
1317 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
) &&
1318 nexthop_nh
->nh_power
) {
1319 w
-= nexthop_nh
->nh_power
;
1321 nexthop_nh
->nh_power
--;
1323 res
->nh_sel
= nhsel
;
1324 spin_unlock_bh(&fib_multipath_lock
);
1328 } endfor_nexthops(fi
);
1330 /* Race condition: route has just become dead. */
1332 spin_unlock_bh(&fib_multipath_lock
);