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 if (fi
->fib_metrics
!= (u32
*) dst_default_metrics
)
217 kfree(fi
->fib_metrics
);
221 void free_fib_info(struct fib_info
*fi
)
223 if (fi
->fib_dead
== 0) {
224 pr_warn("Freeing alive fib_info %p\n", fi
);
228 #ifdef CONFIG_IP_ROUTE_CLASSID
229 change_nexthops(fi
) {
230 if (nexthop_nh
->nh_tclassid
)
231 fi
->fib_net
->ipv4
.fib_num_tclassid_users
--;
232 } endfor_nexthops(fi
);
234 call_rcu(&fi
->rcu
, free_fib_info_rcu
);
237 void fib_release_info(struct fib_info
*fi
)
239 spin_lock_bh(&fib_info_lock
);
240 if (fi
&& --fi
->fib_treeref
== 0) {
241 hlist_del(&fi
->fib_hash
);
243 hlist_del(&fi
->fib_lhash
);
244 change_nexthops(fi
) {
245 if (!nexthop_nh
->nh_dev
)
247 hlist_del(&nexthop_nh
->nh_hash
);
248 } endfor_nexthops(fi
)
252 spin_unlock_bh(&fib_info_lock
);
255 static inline int nh_comp(const struct fib_info
*fi
, const struct fib_info
*ofi
)
257 const struct fib_nh
*onh
= ofi
->fib_nh
;
260 if (nh
->nh_oif
!= onh
->nh_oif
||
261 nh
->nh_gw
!= onh
->nh_gw
||
262 nh
->nh_scope
!= onh
->nh_scope
||
263 #ifdef CONFIG_IP_ROUTE_MULTIPATH
264 nh
->nh_weight
!= onh
->nh_weight
||
266 #ifdef CONFIG_IP_ROUTE_CLASSID
267 nh
->nh_tclassid
!= onh
->nh_tclassid
||
269 ((nh
->nh_flags
^ onh
->nh_flags
) & ~RTNH_COMPARE_MASK
))
272 } endfor_nexthops(fi
);
276 static inline unsigned int fib_devindex_hashfn(unsigned int val
)
278 unsigned int mask
= DEVINDEX_HASHSIZE
- 1;
281 (val
>> DEVINDEX_HASHBITS
) ^
282 (val
>> (DEVINDEX_HASHBITS
* 2))) & mask
;
285 static inline unsigned int fib_info_hashfn(const struct fib_info
*fi
)
287 unsigned int mask
= (fib_info_hash_size
- 1);
288 unsigned int val
= fi
->fib_nhs
;
290 val
^= (fi
->fib_protocol
<< 8) | fi
->fib_scope
;
291 val
^= (__force u32
)fi
->fib_prefsrc
;
292 val
^= fi
->fib_priority
;
294 val
^= fib_devindex_hashfn(nh
->nh_oif
);
295 } endfor_nexthops(fi
)
297 return (val
^ (val
>> 7) ^ (val
>> 12)) & mask
;
300 static struct fib_info
*fib_find_info(const struct fib_info
*nfi
)
302 struct hlist_head
*head
;
306 hash
= fib_info_hashfn(nfi
);
307 head
= &fib_info_hash
[hash
];
309 hlist_for_each_entry(fi
, head
, fib_hash
) {
310 if (!net_eq(fi
->fib_net
, nfi
->fib_net
))
312 if (fi
->fib_nhs
!= nfi
->fib_nhs
)
314 if (nfi
->fib_protocol
== fi
->fib_protocol
&&
315 nfi
->fib_scope
== fi
->fib_scope
&&
316 nfi
->fib_prefsrc
== fi
->fib_prefsrc
&&
317 nfi
->fib_priority
== fi
->fib_priority
&&
318 nfi
->fib_type
== fi
->fib_type
&&
319 memcmp(nfi
->fib_metrics
, fi
->fib_metrics
,
320 sizeof(u32
) * RTAX_MAX
) == 0 &&
321 !((nfi
->fib_flags
^ fi
->fib_flags
) & ~RTNH_COMPARE_MASK
) &&
322 (nfi
->fib_nhs
== 0 || nh_comp(fi
, nfi
) == 0))
329 /* Check, that the gateway is already configured.
330 * Used only by redirect accept routine.
332 int ip_fib_check_default(__be32 gw
, struct net_device
*dev
)
334 struct hlist_head
*head
;
338 spin_lock(&fib_info_lock
);
340 hash
= fib_devindex_hashfn(dev
->ifindex
);
341 head
= &fib_info_devhash
[hash
];
342 hlist_for_each_entry(nh
, head
, nh_hash
) {
343 if (nh
->nh_dev
== dev
&&
345 !(nh
->nh_flags
& RTNH_F_DEAD
)) {
346 spin_unlock(&fib_info_lock
);
351 spin_unlock(&fib_info_lock
);
356 static inline size_t fib_nlmsg_size(struct fib_info
*fi
)
358 size_t payload
= NLMSG_ALIGN(sizeof(struct rtmsg
))
359 + nla_total_size(4) /* RTA_TABLE */
360 + nla_total_size(4) /* RTA_DST */
361 + nla_total_size(4) /* RTA_PRIORITY */
362 + nla_total_size(4) /* RTA_PREFSRC */
363 + nla_total_size(TCP_CA_NAME_MAX
); /* RTAX_CC_ALGO */
365 /* space for nested metrics */
366 payload
+= nla_total_size((RTAX_MAX
* nla_total_size(4)));
369 /* Also handles the special case fib_nhs == 1 */
371 /* each nexthop is packed in an attribute */
372 size_t nhsize
= nla_total_size(sizeof(struct rtnexthop
));
374 /* may contain flow and gateway attribute */
375 nhsize
+= 2 * nla_total_size(4);
377 /* all nexthops are packed in a nested attribute */
378 payload
+= nla_total_size(fi
->fib_nhs
* nhsize
);
384 void rtmsg_fib(int event
, __be32 key
, struct fib_alias
*fa
,
385 int dst_len
, u32 tb_id
, const struct nl_info
*info
,
386 unsigned int nlm_flags
)
389 u32 seq
= info
->nlh
? info
->nlh
->nlmsg_seq
: 0;
392 skb
= nlmsg_new(fib_nlmsg_size(fa
->fa_info
), GFP_KERNEL
);
396 err
= fib_dump_info(skb
, info
->portid
, seq
, event
, tb_id
,
397 fa
->fa_type
, key
, dst_len
,
398 fa
->fa_tos
, fa
->fa_info
, nlm_flags
);
400 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
401 WARN_ON(err
== -EMSGSIZE
);
405 rtnl_notify(skb
, info
->nl_net
, info
->portid
, RTNLGRP_IPV4_ROUTE
,
406 info
->nlh
, GFP_KERNEL
);
410 rtnl_set_sk_err(info
->nl_net
, RTNLGRP_IPV4_ROUTE
, err
);
413 static int fib_detect_death(struct fib_info
*fi
, int order
,
414 struct fib_info
**last_resort
, int *last_idx
,
418 int state
= NUD_NONE
;
420 n
= neigh_lookup(&arp_tbl
, &fi
->fib_nh
[0].nh_gw
, fi
->fib_dev
);
422 state
= n
->nud_state
;
425 if (state
== NUD_REACHABLE
)
427 if ((state
& NUD_VALID
) && order
!= dflt
)
429 if ((state
& NUD_VALID
) ||
430 (*last_idx
< 0 && order
> dflt
)) {
437 #ifdef CONFIG_IP_ROUTE_MULTIPATH
439 static int fib_count_nexthops(struct rtnexthop
*rtnh
, int remaining
)
443 while (rtnh_ok(rtnh
, remaining
)) {
445 rtnh
= rtnh_next(rtnh
, &remaining
);
448 /* leftover implies invalid nexthop configuration, discard it */
449 return remaining
> 0 ? 0 : nhs
;
452 static int fib_get_nhs(struct fib_info
*fi
, struct rtnexthop
*rtnh
,
453 int remaining
, struct fib_config
*cfg
)
455 change_nexthops(fi
) {
458 if (!rtnh_ok(rtnh
, remaining
))
461 nexthop_nh
->nh_flags
=
462 (cfg
->fc_flags
& ~0xFF) | rtnh
->rtnh_flags
;
463 nexthop_nh
->nh_oif
= rtnh
->rtnh_ifindex
;
464 nexthop_nh
->nh_weight
= rtnh
->rtnh_hops
+ 1;
466 attrlen
= rtnh_attrlen(rtnh
);
468 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
470 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
471 nexthop_nh
->nh_gw
= nla
? nla_get_in_addr(nla
) : 0;
472 #ifdef CONFIG_IP_ROUTE_CLASSID
473 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
474 nexthop_nh
->nh_tclassid
= nla
? nla_get_u32(nla
) : 0;
475 if (nexthop_nh
->nh_tclassid
)
476 fi
->fib_net
->ipv4
.fib_num_tclassid_users
++;
480 rtnh
= rtnh_next(rtnh
, &remaining
);
481 } endfor_nexthops(fi
);
488 int fib_nh_match(struct fib_config
*cfg
, struct fib_info
*fi
)
490 #ifdef CONFIG_IP_ROUTE_MULTIPATH
491 struct rtnexthop
*rtnh
;
495 if (cfg
->fc_priority
&& cfg
->fc_priority
!= fi
->fib_priority
)
498 if (cfg
->fc_oif
|| cfg
->fc_gw
) {
499 if ((!cfg
->fc_oif
|| cfg
->fc_oif
== fi
->fib_nh
->nh_oif
) &&
500 (!cfg
->fc_gw
|| cfg
->fc_gw
== fi
->fib_nh
->nh_gw
))
505 #ifdef CONFIG_IP_ROUTE_MULTIPATH
510 remaining
= cfg
->fc_mp_len
;
515 if (!rtnh_ok(rtnh
, remaining
))
518 if (rtnh
->rtnh_ifindex
&& rtnh
->rtnh_ifindex
!= nh
->nh_oif
)
521 attrlen
= rtnh_attrlen(rtnh
);
523 struct nlattr
*nla
, *attrs
= rtnh_attrs(rtnh
);
525 nla
= nla_find(attrs
, attrlen
, RTA_GATEWAY
);
526 if (nla
&& nla_get_in_addr(nla
) != nh
->nh_gw
)
528 #ifdef CONFIG_IP_ROUTE_CLASSID
529 nla
= nla_find(attrs
, attrlen
, RTA_FLOW
);
530 if (nla
&& nla_get_u32(nla
) != nh
->nh_tclassid
)
535 rtnh
= rtnh_next(rtnh
, &remaining
);
536 } endfor_nexthops(fi
);
546 * Semantics of nexthop is very messy by historical reasons.
547 * We have to take into account, that:
548 * a) gateway can be actually local interface address,
549 * so that gatewayed route is direct.
550 * b) gateway must be on-link address, possibly
551 * described not by an ifaddr, but also by a direct route.
552 * c) If both gateway and interface are specified, they should not
554 * d) If we use tunnel routes, gateway could be not on-link.
556 * Attempt to reconcile all of these (alas, self-contradictory) conditions
557 * results in pretty ugly and hairy code with obscure logic.
559 * I chose to generalized it instead, so that the size
560 * of code does not increase practically, but it becomes
562 * Every prefix is assigned a "scope" value: "host" is local address,
563 * "link" is direct route,
564 * [ ... "site" ... "interior" ... ]
565 * and "universe" is true gateway route with global meaning.
567 * Every prefix refers to a set of "nexthop"s (gw, oif),
568 * where gw must have narrower scope. This recursion stops
569 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
570 * which means that gw is forced to be on link.
572 * Code is still hairy, but now it is apparently logically
573 * consistent and very flexible. F.e. as by-product it allows
574 * to co-exists in peace independent exterior and interior
577 * Normally it looks as following.
579 * {universe prefix} -> (gw, oif) [scope link]
581 * |-> {link prefix} -> (gw, oif) [scope local]
583 * |-> {local prefix} (terminal node)
585 static int fib_check_nh(struct fib_config
*cfg
, struct fib_info
*fi
,
590 struct net_device
*dev
;
592 net
= cfg
->fc_nlinfo
.nl_net
;
594 struct fib_result res
;
596 if (nh
->nh_flags
& RTNH_F_ONLINK
) {
598 if (cfg
->fc_scope
>= RT_SCOPE_LINK
)
600 if (inet_addr_type(net
, nh
->nh_gw
) != RTN_UNICAST
)
602 dev
= __dev_get_by_index(net
, nh
->nh_oif
);
605 if (!(dev
->flags
& IFF_UP
))
607 if (!netif_carrier_ok(dev
))
608 nh
->nh_flags
|= RTNH_F_LINKDOWN
;
611 nh
->nh_scope
= RT_SCOPE_LINK
;
616 struct flowi4 fl4
= {
618 .flowi4_scope
= cfg
->fc_scope
+ 1,
619 .flowi4_oif
= nh
->nh_oif
,
620 .flowi4_iif
= LOOPBACK_IFINDEX
,
623 /* It is not necessary, but requires a bit of thinking */
624 if (fl4
.flowi4_scope
< RT_SCOPE_LINK
)
625 fl4
.flowi4_scope
= RT_SCOPE_LINK
;
626 err
= fib_lookup(net
, &fl4
, &res
,
627 FIB_LOOKUP_IGNORE_LINKSTATE
);
634 if (res
.type
!= RTN_UNICAST
&& res
.type
!= RTN_LOCAL
)
636 nh
->nh_scope
= res
.scope
;
637 nh
->nh_oif
= FIB_RES_OIF(res
);
638 nh
->nh_dev
= dev
= FIB_RES_DEV(res
);
642 if (!netif_carrier_ok(dev
))
643 nh
->nh_flags
|= RTNH_F_LINKDOWN
;
644 err
= (dev
->flags
& IFF_UP
) ? 0 : -ENETDOWN
;
646 struct in_device
*in_dev
;
648 if (nh
->nh_flags
& (RTNH_F_PERVASIVE
| RTNH_F_ONLINK
))
653 in_dev
= inetdev_by_index(net
, nh
->nh_oif
);
657 if (!(in_dev
->dev
->flags
& IFF_UP
))
659 nh
->nh_dev
= in_dev
->dev
;
660 dev_hold(nh
->nh_dev
);
661 nh
->nh_scope
= RT_SCOPE_HOST
;
662 if (!netif_carrier_ok(nh
->nh_dev
))
663 nh
->nh_flags
|= RTNH_F_LINKDOWN
;
671 static inline unsigned int fib_laddr_hashfn(__be32 val
)
673 unsigned int mask
= (fib_info_hash_size
- 1);
675 return ((__force u32
)val
^
676 ((__force u32
)val
>> 7) ^
677 ((__force u32
)val
>> 14)) & mask
;
680 static struct hlist_head
*fib_info_hash_alloc(int bytes
)
682 if (bytes
<= PAGE_SIZE
)
683 return kzalloc(bytes
, GFP_KERNEL
);
685 return (struct hlist_head
*)
686 __get_free_pages(GFP_KERNEL
| __GFP_ZERO
,
690 static void fib_info_hash_free(struct hlist_head
*hash
, int bytes
)
695 if (bytes
<= PAGE_SIZE
)
698 free_pages((unsigned long) hash
, get_order(bytes
));
701 static void fib_info_hash_move(struct hlist_head
*new_info_hash
,
702 struct hlist_head
*new_laddrhash
,
703 unsigned int new_size
)
705 struct hlist_head
*old_info_hash
, *old_laddrhash
;
706 unsigned int old_size
= fib_info_hash_size
;
707 unsigned int i
, bytes
;
709 spin_lock_bh(&fib_info_lock
);
710 old_info_hash
= fib_info_hash
;
711 old_laddrhash
= fib_info_laddrhash
;
712 fib_info_hash_size
= new_size
;
714 for (i
= 0; i
< old_size
; i
++) {
715 struct hlist_head
*head
= &fib_info_hash
[i
];
716 struct hlist_node
*n
;
719 hlist_for_each_entry_safe(fi
, n
, head
, fib_hash
) {
720 struct hlist_head
*dest
;
721 unsigned int new_hash
;
723 new_hash
= fib_info_hashfn(fi
);
724 dest
= &new_info_hash
[new_hash
];
725 hlist_add_head(&fi
->fib_hash
, dest
);
728 fib_info_hash
= new_info_hash
;
730 for (i
= 0; i
< old_size
; i
++) {
731 struct hlist_head
*lhead
= &fib_info_laddrhash
[i
];
732 struct hlist_node
*n
;
735 hlist_for_each_entry_safe(fi
, n
, lhead
, fib_lhash
) {
736 struct hlist_head
*ldest
;
737 unsigned int new_hash
;
739 new_hash
= fib_laddr_hashfn(fi
->fib_prefsrc
);
740 ldest
= &new_laddrhash
[new_hash
];
741 hlist_add_head(&fi
->fib_lhash
, ldest
);
744 fib_info_laddrhash
= new_laddrhash
;
746 spin_unlock_bh(&fib_info_lock
);
748 bytes
= old_size
* sizeof(struct hlist_head
*);
749 fib_info_hash_free(old_info_hash
, bytes
);
750 fib_info_hash_free(old_laddrhash
, bytes
);
753 __be32
fib_info_update_nh_saddr(struct net
*net
, struct fib_nh
*nh
)
755 nh
->nh_saddr
= inet_select_addr(nh
->nh_dev
,
757 nh
->nh_parent
->fib_scope
);
758 nh
->nh_saddr_genid
= atomic_read(&net
->ipv4
.dev_addr_genid
);
763 struct fib_info
*fib_create_info(struct fib_config
*cfg
)
766 struct fib_info
*fi
= NULL
;
767 struct fib_info
*ofi
;
769 struct net
*net
= cfg
->fc_nlinfo
.nl_net
;
771 if (cfg
->fc_type
> RTN_MAX
)
774 /* Fast check to catch the most weird cases */
775 if (fib_props
[cfg
->fc_type
].scope
> cfg
->fc_scope
)
778 #ifdef CONFIG_IP_ROUTE_MULTIPATH
780 nhs
= fib_count_nexthops(cfg
->fc_mp
, cfg
->fc_mp_len
);
787 if (fib_info_cnt
>= fib_info_hash_size
) {
788 unsigned int new_size
= fib_info_hash_size
<< 1;
789 struct hlist_head
*new_info_hash
;
790 struct hlist_head
*new_laddrhash
;
795 bytes
= new_size
* sizeof(struct hlist_head
*);
796 new_info_hash
= fib_info_hash_alloc(bytes
);
797 new_laddrhash
= fib_info_hash_alloc(bytes
);
798 if (!new_info_hash
|| !new_laddrhash
) {
799 fib_info_hash_free(new_info_hash
, bytes
);
800 fib_info_hash_free(new_laddrhash
, bytes
);
802 fib_info_hash_move(new_info_hash
, new_laddrhash
, new_size
);
804 if (!fib_info_hash_size
)
808 fi
= kzalloc(sizeof(*fi
)+nhs
*sizeof(struct fib_nh
), GFP_KERNEL
);
813 fi
->fib_metrics
= kzalloc(sizeof(u32
) * RTAX_MAX
, GFP_KERNEL
);
814 if (!fi
->fib_metrics
)
817 fi
->fib_metrics
= (u32
*) dst_default_metrics
;
820 fi
->fib_protocol
= cfg
->fc_protocol
;
821 fi
->fib_scope
= cfg
->fc_scope
;
822 fi
->fib_flags
= cfg
->fc_flags
;
823 fi
->fib_priority
= cfg
->fc_priority
;
824 fi
->fib_prefsrc
= cfg
->fc_prefsrc
;
825 fi
->fib_type
= cfg
->fc_type
;
828 change_nexthops(fi
) {
829 nexthop_nh
->nh_parent
= fi
;
830 nexthop_nh
->nh_pcpu_rth_output
= alloc_percpu(struct rtable __rcu
*);
831 if (!nexthop_nh
->nh_pcpu_rth_output
)
833 } endfor_nexthops(fi
)
839 nla_for_each_attr(nla
, cfg
->fc_mx
, cfg
->fc_mx_len
, remaining
) {
840 int type
= nla_type(nla
);
847 if (type
== RTAX_CC_ALGO
) {
848 char tmp
[TCP_CA_NAME_MAX
];
850 nla_strlcpy(tmp
, nla
, sizeof(tmp
));
851 val
= tcp_ca_get_key_by_name(tmp
);
852 if (val
== TCP_CA_UNSPEC
)
855 val
= nla_get_u32(nla
);
857 if (type
== RTAX_ADVMSS
&& val
> 65535 - 40)
859 if (type
== RTAX_MTU
&& val
> 65535 - 15)
861 fi
->fib_metrics
[type
- 1] = val
;
867 #ifdef CONFIG_IP_ROUTE_MULTIPATH
868 err
= fib_get_nhs(fi
, cfg
->fc_mp
, cfg
->fc_mp_len
, cfg
);
871 if (cfg
->fc_oif
&& fi
->fib_nh
->nh_oif
!= cfg
->fc_oif
)
873 if (cfg
->fc_gw
&& fi
->fib_nh
->nh_gw
!= cfg
->fc_gw
)
875 #ifdef CONFIG_IP_ROUTE_CLASSID
876 if (cfg
->fc_flow
&& fi
->fib_nh
->nh_tclassid
!= cfg
->fc_flow
)
883 struct fib_nh
*nh
= fi
->fib_nh
;
885 nh
->nh_oif
= cfg
->fc_oif
;
886 nh
->nh_gw
= cfg
->fc_gw
;
887 nh
->nh_flags
= cfg
->fc_flags
;
888 #ifdef CONFIG_IP_ROUTE_CLASSID
889 nh
->nh_tclassid
= cfg
->fc_flow
;
891 fi
->fib_net
->ipv4
.fib_num_tclassid_users
++;
893 #ifdef CONFIG_IP_ROUTE_MULTIPATH
898 if (fib_props
[cfg
->fc_type
].error
) {
899 if (cfg
->fc_gw
|| cfg
->fc_oif
|| cfg
->fc_mp
)
903 switch (cfg
->fc_type
) {
915 if (cfg
->fc_scope
> RT_SCOPE_HOST
)
918 if (cfg
->fc_scope
== RT_SCOPE_HOST
) {
919 struct fib_nh
*nh
= fi
->fib_nh
;
921 /* Local address is added. */
922 if (nhs
!= 1 || nh
->nh_gw
)
924 nh
->nh_scope
= RT_SCOPE_NOWHERE
;
925 nh
->nh_dev
= dev_get_by_index(net
, fi
->fib_nh
->nh_oif
);
932 change_nexthops(fi
) {
933 err
= fib_check_nh(cfg
, fi
, nexthop_nh
);
936 if (nexthop_nh
->nh_flags
& RTNH_F_LINKDOWN
)
938 } endfor_nexthops(fi
)
939 if (linkdown
== fi
->fib_nhs
)
940 fi
->fib_flags
|= RTNH_F_LINKDOWN
;
943 if (fi
->fib_prefsrc
) {
944 if (cfg
->fc_type
!= RTN_LOCAL
|| !cfg
->fc_dst
||
945 fi
->fib_prefsrc
!= cfg
->fc_dst
)
946 if (inet_addr_type(net
, fi
->fib_prefsrc
) != RTN_LOCAL
)
950 change_nexthops(fi
) {
951 fib_info_update_nh_saddr(net
, nexthop_nh
);
952 } endfor_nexthops(fi
)
955 ofi
= fib_find_info(fi
);
964 atomic_inc(&fi
->fib_clntref
);
965 spin_lock_bh(&fib_info_lock
);
966 hlist_add_head(&fi
->fib_hash
,
967 &fib_info_hash
[fib_info_hashfn(fi
)]);
968 if (fi
->fib_prefsrc
) {
969 struct hlist_head
*head
;
971 head
= &fib_info_laddrhash
[fib_laddr_hashfn(fi
->fib_prefsrc
)];
972 hlist_add_head(&fi
->fib_lhash
, head
);
974 change_nexthops(fi
) {
975 struct hlist_head
*head
;
978 if (!nexthop_nh
->nh_dev
)
980 hash
= fib_devindex_hashfn(nexthop_nh
->nh_dev
->ifindex
);
981 head
= &fib_info_devhash
[hash
];
982 hlist_add_head(&nexthop_nh
->nh_hash
, head
);
983 } endfor_nexthops(fi
)
984 spin_unlock_bh(&fib_info_lock
);
999 int fib_dump_info(struct sk_buff
*skb
, u32 portid
, u32 seq
, int event
,
1000 u32 tb_id
, u8 type
, __be32 dst
, int dst_len
, u8 tos
,
1001 struct fib_info
*fi
, unsigned int flags
)
1003 struct nlmsghdr
*nlh
;
1006 nlh
= nlmsg_put(skb
, portid
, seq
, event
, sizeof(*rtm
), flags
);
1010 rtm
= nlmsg_data(nlh
);
1011 rtm
->rtm_family
= AF_INET
;
1012 rtm
->rtm_dst_len
= dst_len
;
1013 rtm
->rtm_src_len
= 0;
1016 rtm
->rtm_table
= tb_id
;
1018 rtm
->rtm_table
= RT_TABLE_COMPAT
;
1019 if (nla_put_u32(skb
, RTA_TABLE
, tb_id
))
1020 goto nla_put_failure
;
1021 rtm
->rtm_type
= type
;
1022 rtm
->rtm_flags
= fi
->fib_flags
;
1023 rtm
->rtm_scope
= fi
->fib_scope
;
1024 rtm
->rtm_protocol
= fi
->fib_protocol
;
1026 if (rtm
->rtm_dst_len
&&
1027 nla_put_in_addr(skb
, RTA_DST
, dst
))
1028 goto nla_put_failure
;
1029 if (fi
->fib_priority
&&
1030 nla_put_u32(skb
, RTA_PRIORITY
, fi
->fib_priority
))
1031 goto nla_put_failure
;
1032 if (rtnetlink_put_metrics(skb
, fi
->fib_metrics
) < 0)
1033 goto nla_put_failure
;
1035 if (fi
->fib_prefsrc
&&
1036 nla_put_in_addr(skb
, RTA_PREFSRC
, fi
->fib_prefsrc
))
1037 goto nla_put_failure
;
1038 if (fi
->fib_nhs
== 1) {
1039 struct in_device
*in_dev
;
1041 if (fi
->fib_nh
->nh_gw
&&
1042 nla_put_in_addr(skb
, RTA_GATEWAY
, fi
->fib_nh
->nh_gw
))
1043 goto nla_put_failure
;
1044 if (fi
->fib_nh
->nh_oif
&&
1045 nla_put_u32(skb
, RTA_OIF
, fi
->fib_nh
->nh_oif
))
1046 goto nla_put_failure
;
1047 if (fi
->fib_nh
->nh_flags
& RTNH_F_LINKDOWN
) {
1048 in_dev
= __in_dev_get_rcu(fi
->fib_nh
->nh_dev
);
1050 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev
))
1051 rtm
->rtm_flags
|= RTNH_F_DEAD
;
1053 #ifdef CONFIG_IP_ROUTE_CLASSID
1054 if (fi
->fib_nh
[0].nh_tclassid
&&
1055 nla_put_u32(skb
, RTA_FLOW
, fi
->fib_nh
[0].nh_tclassid
))
1056 goto nla_put_failure
;
1059 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1060 if (fi
->fib_nhs
> 1) {
1061 struct rtnexthop
*rtnh
;
1064 mp
= nla_nest_start(skb
, RTA_MULTIPATH
);
1066 goto nla_put_failure
;
1069 struct in_device
*in_dev
;
1071 rtnh
= nla_reserve_nohdr(skb
, sizeof(*rtnh
));
1073 goto nla_put_failure
;
1075 rtnh
->rtnh_flags
= nh
->nh_flags
& 0xFF;
1076 if (nh
->nh_flags
& RTNH_F_LINKDOWN
) {
1077 in_dev
= __in_dev_get_rcu(nh
->nh_dev
);
1079 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev
))
1080 rtnh
->rtnh_flags
|= RTNH_F_DEAD
;
1082 rtnh
->rtnh_hops
= nh
->nh_weight
- 1;
1083 rtnh
->rtnh_ifindex
= nh
->nh_oif
;
1086 nla_put_in_addr(skb
, RTA_GATEWAY
, nh
->nh_gw
))
1087 goto nla_put_failure
;
1088 #ifdef CONFIG_IP_ROUTE_CLASSID
1089 if (nh
->nh_tclassid
&&
1090 nla_put_u32(skb
, RTA_FLOW
, nh
->nh_tclassid
))
1091 goto nla_put_failure
;
1093 /* length of rtnetlink header + attributes */
1094 rtnh
->rtnh_len
= nlmsg_get_pos(skb
) - (void *) rtnh
;
1095 } endfor_nexthops(fi
);
1097 nla_nest_end(skb
, mp
);
1100 nlmsg_end(skb
, nlh
);
1104 nlmsg_cancel(skb
, nlh
);
1110 * - local address disappeared -> we must delete all the entries
1112 * - device went down -> we must shutdown all nexthops going via it.
1114 int fib_sync_down_addr(struct net
*net
, __be32 local
)
1117 unsigned int hash
= fib_laddr_hashfn(local
);
1118 struct hlist_head
*head
= &fib_info_laddrhash
[hash
];
1119 struct fib_info
*fi
;
1121 if (!fib_info_laddrhash
|| local
== 0)
1124 hlist_for_each_entry(fi
, head
, fib_lhash
) {
1125 if (!net_eq(fi
->fib_net
, net
))
1127 if (fi
->fib_prefsrc
== local
) {
1128 fi
->fib_flags
|= RTNH_F_DEAD
;
1135 int fib_sync_down_dev(struct net_device
*dev
, unsigned long event
)
1138 int scope
= RT_SCOPE_NOWHERE
;
1139 struct fib_info
*prev_fi
= NULL
;
1140 unsigned int hash
= fib_devindex_hashfn(dev
->ifindex
);
1141 struct hlist_head
*head
= &fib_info_devhash
[hash
];
1144 if (event
== NETDEV_UNREGISTER
||
1145 event
== NETDEV_DOWN
)
1148 hlist_for_each_entry(nh
, head
, nh_hash
) {
1149 struct fib_info
*fi
= nh
->nh_parent
;
1152 BUG_ON(!fi
->fib_nhs
);
1153 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1157 change_nexthops(fi
) {
1158 if (nexthop_nh
->nh_flags
& RTNH_F_DEAD
)
1160 else if (nexthop_nh
->nh_dev
== dev
&&
1161 nexthop_nh
->nh_scope
!= scope
) {
1164 case NETDEV_UNREGISTER
:
1165 nexthop_nh
->nh_flags
|= RTNH_F_DEAD
;
1168 nexthop_nh
->nh_flags
|= RTNH_F_LINKDOWN
;
1171 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1172 spin_lock_bh(&fib_multipath_lock
);
1173 fi
->fib_power
-= nexthop_nh
->nh_power
;
1174 nexthop_nh
->nh_power
= 0;
1175 spin_unlock_bh(&fib_multipath_lock
);
1179 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1180 if (event
== NETDEV_UNREGISTER
&&
1181 nexthop_nh
->nh_dev
== dev
) {
1186 } endfor_nexthops(fi
)
1187 if (dead
== fi
->fib_nhs
) {
1190 case NETDEV_UNREGISTER
:
1191 fi
->fib_flags
|= RTNH_F_DEAD
;
1194 fi
->fib_flags
|= RTNH_F_LINKDOWN
;
1204 /* Must be invoked inside of an RCU protected region. */
1205 void fib_select_default(struct fib_result
*res
)
1207 struct fib_info
*fi
= NULL
, *last_resort
= NULL
;
1208 struct hlist_head
*fa_head
= res
->fa_head
;
1209 struct fib_table
*tb
= res
->table
;
1210 int order
= -1, last_idx
= -1;
1211 struct fib_alias
*fa
;
1213 hlist_for_each_entry_rcu(fa
, fa_head
, fa_list
) {
1214 struct fib_info
*next_fi
= fa
->fa_info
;
1216 if (next_fi
->fib_scope
!= res
->scope
||
1217 fa
->fa_type
!= RTN_UNICAST
)
1220 if (next_fi
->fib_priority
> res
->fi
->fib_priority
)
1222 if (!next_fi
->fib_nh
[0].nh_gw
||
1223 next_fi
->fib_nh
[0].nh_scope
!= RT_SCOPE_LINK
)
1226 fib_alias_accessed(fa
);
1229 if (next_fi
!= res
->fi
)
1231 } else if (!fib_detect_death(fi
, order
, &last_resort
,
1232 &last_idx
, tb
->tb_default
)) {
1233 fib_result_assign(res
, fi
);
1234 tb
->tb_default
= order
;
1241 if (order
<= 0 || !fi
) {
1242 tb
->tb_default
= -1;
1246 if (!fib_detect_death(fi
, order
, &last_resort
, &last_idx
,
1248 fib_result_assign(res
, fi
);
1249 tb
->tb_default
= order
;
1254 fib_result_assign(res
, last_resort
);
1255 tb
->tb_default
= last_idx
;
1261 * Dead device goes up. We wake up dead nexthops.
1262 * It takes sense only on multipath routes.
1264 int fib_sync_up(struct net_device
*dev
, unsigned int nh_flags
)
1266 struct fib_info
*prev_fi
;
1268 struct hlist_head
*head
;
1272 if (!(dev
->flags
& IFF_UP
))
1276 hash
= fib_devindex_hashfn(dev
->ifindex
);
1277 head
= &fib_info_devhash
[hash
];
1280 hlist_for_each_entry(nh
, head
, nh_hash
) {
1281 struct fib_info
*fi
= nh
->nh_parent
;
1284 BUG_ON(!fi
->fib_nhs
);
1285 if (nh
->nh_dev
!= dev
|| fi
== prev_fi
)
1290 change_nexthops(fi
) {
1291 if (!(nexthop_nh
->nh_flags
& nh_flags
)) {
1295 if (!nexthop_nh
->nh_dev
||
1296 !(nexthop_nh
->nh_dev
->flags
& IFF_UP
))
1298 if (nexthop_nh
->nh_dev
!= dev
||
1299 !__in_dev_get_rtnl(dev
))
1302 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1303 spin_lock_bh(&fib_multipath_lock
);
1304 nexthop_nh
->nh_power
= 0;
1305 nexthop_nh
->nh_flags
&= ~nh_flags
;
1306 spin_unlock_bh(&fib_multipath_lock
);
1308 nexthop_nh
->nh_flags
&= ~nh_flags
;
1310 } endfor_nexthops(fi
)
1313 fi
->fib_flags
&= ~nh_flags
;
1321 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1324 * The algorithm is suboptimal, but it provides really
1325 * fair weighted route distribution.
1327 void fib_select_multipath(struct fib_result
*res
)
1329 struct fib_info
*fi
= res
->fi
;
1330 struct in_device
*in_dev
;
1333 spin_lock_bh(&fib_multipath_lock
);
1334 if (fi
->fib_power
<= 0) {
1336 change_nexthops(fi
) {
1337 in_dev
= __in_dev_get_rcu(nexthop_nh
->nh_dev
);
1338 if (nexthop_nh
->nh_flags
& RTNH_F_DEAD
)
1341 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev
) &&
1342 nexthop_nh
->nh_flags
& RTNH_F_LINKDOWN
)
1344 power
+= nexthop_nh
->nh_weight
;
1345 nexthop_nh
->nh_power
= nexthop_nh
->nh_weight
;
1346 } endfor_nexthops(fi
);
1347 fi
->fib_power
= power
;
1349 spin_unlock_bh(&fib_multipath_lock
);
1350 /* Race condition: route has just become dead. */
1357 /* w should be random number [0..fi->fib_power-1],
1358 * it is pretty bad approximation.
1361 w
= jiffies
% fi
->fib_power
;
1363 change_nexthops(fi
) {
1364 if (!(nexthop_nh
->nh_flags
& RTNH_F_DEAD
) &&
1365 nexthop_nh
->nh_power
) {
1366 w
-= nexthop_nh
->nh_power
;
1368 nexthop_nh
->nh_power
--;
1370 res
->nh_sel
= nhsel
;
1371 spin_unlock_bh(&fib_multipath_lock
);
1375 } endfor_nexthops(fi
);
1377 /* Race condition: route has just become dead. */
1379 spin_unlock_bh(&fib_multipath_lock
);