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 <asm/system.h>
18 #include <linux/bitops.h>
19 #include <linux/types.h>
20 #include <linux/kernel.h>
21 #include <linux/jiffies.h>
23 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/errno.h>
28 #include <linux/inet.h>
29 #include <linux/inetdevice.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/proc_fs.h>
33 #include <linux/skbuff.h>
34 #include <linux/init.h>
35 #include <linux/slab.h>
39 #include <net/protocol.h>
40 #include <net/route.h>
43 #include <net/ip_fib.h>
44 #include <net/netlink.h>
45 #include <net/nexthop.h>
47 #include "fib_lookup.h"
49 static DEFINE_SPINLOCK(fib_info_lock);
50 static struct hlist_head *fib_info_hash;
51 static struct hlist_head *fib_info_laddrhash;
52 static unsigned int fib_hash_size;
53 static unsigned int fib_info_cnt;
55 #define DEVINDEX_HASHBITS 8
56 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
57 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
59 #ifdef CONFIG_IP_ROUTE_MULTIPATH
61 static DEFINE_SPINLOCK(fib_multipath_lock);
63 #define for_nexthops(fi) { \
64 int nhsel; const struct fib_nh *nh; \
65 for (nhsel = 0, nh = (fi)->fib_nh; \
66 nhsel < (fi)->fib_nhs; \
69 #define change_nexthops(fi) { \
70 int nhsel; struct fib_nh *nexthop_nh; \
71 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
72 nhsel < (fi)->fib_nhs; \
73 nexthop_nh++, nhsel++)
75 #else /* CONFIG_IP_ROUTE_MULTIPATH */
77 /* Hope, that gcc will optimize it to get rid of dummy loop */
79 #define for_nexthops(fi) { \
80 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
81 for (nhsel = 0; nhsel < 1; nhsel++)
83 #define change_nexthops(fi) { \
85 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
86 for (nhsel = 0; nhsel < 1; nhsel++)
88 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
90 #define endfor_nexthops(fi) }
97 } fib_props[RTN_MAX + 1] = {
100 .scope = RT_SCOPE_NOWHERE,
104 .scope = RT_SCOPE_UNIVERSE,
108 .scope = RT_SCOPE_HOST,
112 .scope = RT_SCOPE_LINK,
116 .scope = RT_SCOPE_LINK,
120 .scope = RT_SCOPE_UNIVERSE,
124 .scope = RT_SCOPE_UNIVERSE,
126 [RTN_UNREACHABLE] = {
127 .error = -EHOSTUNREACH,
128 .scope = RT_SCOPE_UNIVERSE,
132 .scope = RT_SCOPE_UNIVERSE,
136 .scope = RT_SCOPE_UNIVERSE,
140 .scope = RT_SCOPE_NOWHERE,
144 .scope = RT_SCOPE_NOWHERE,
149 /* Release a nexthop info record */
151 static void free_fib_info_rcu(struct rcu_head *head)
153 struct fib_info *fi = container_of(head, struct fib_info, rcu);
155 kfree(fi->fib_metrics);
159 void free_fib_info(struct fib_info *fi)
161 if (fi->fib_dead == 0) {
162 pr_warning("Freeing alive fib_info %p\n", fi);
165 change_nexthops(fi) {
166 if (nexthop_nh->nh_dev)
167 dev_put(nexthop_nh->nh_dev);
168 nexthop_nh->nh_dev = NULL;
169 } endfor_nexthops(fi);
171 release_net(fi->fib_net);
172 call_rcu(&fi->rcu, free_fib_info_rcu);
175 void fib_release_info(struct fib_info *fi)
177 spin_lock_bh(&fib_info_lock);
178 if (fi && --fi->fib_treeref == 0) {
179 hlist_del(&fi->fib_hash);
181 hlist_del(&fi->fib_lhash);
182 change_nexthops(fi) {
183 if (!nexthop_nh->nh_dev)
185 hlist_del(&nexthop_nh->nh_hash);
186 } endfor_nexthops(fi)
190 spin_unlock_bh(&fib_info_lock);
193 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
195 const struct fib_nh *onh = ofi->fib_nh;
198 if (nh->nh_oif != onh->nh_oif ||
199 nh->nh_gw != onh->nh_gw ||
200 nh->nh_scope != onh->nh_scope ||
201 #ifdef CONFIG_IP_ROUTE_MULTIPATH
202 nh->nh_weight != onh->nh_weight ||
204 #ifdef CONFIG_IP_ROUTE_CLASSID
205 nh->nh_tclassid != onh->nh_tclassid ||
207 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
210 } endfor_nexthops(fi);
214 static inline unsigned int fib_devindex_hashfn(unsigned int val)
216 unsigned int mask = DEVINDEX_HASHSIZE - 1;
219 (val >> DEVINDEX_HASHBITS) ^
220 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
223 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
225 unsigned int mask = (fib_hash_size - 1);
226 unsigned int val = fi->fib_nhs;
228 val ^= fi->fib_protocol;
229 val ^= (__force u32)fi->fib_prefsrc;
230 val ^= fi->fib_priority;
232 val ^= fib_devindex_hashfn(nh->nh_oif);
233 } endfor_nexthops(fi)
235 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
238 static struct fib_info *fib_find_info(const struct fib_info *nfi)
240 struct hlist_head *head;
241 struct hlist_node *node;
245 hash = fib_info_hashfn(nfi);
246 head = &fib_info_hash[hash];
248 hlist_for_each_entry(fi, node, head, fib_hash) {
249 if (!net_eq(fi->fib_net, nfi->fib_net))
251 if (fi->fib_nhs != nfi->fib_nhs)
253 if (nfi->fib_protocol == fi->fib_protocol &&
254 nfi->fib_prefsrc == fi->fib_prefsrc &&
255 nfi->fib_priority == fi->fib_priority &&
256 memcmp(nfi->fib_metrics, fi->fib_metrics,
257 sizeof(fi->fib_metrics)) == 0 &&
258 ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
259 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
266 /* Check, that the gateway is already configured.
267 * Used only by redirect accept routine.
269 int ip_fib_check_default(__be32 gw, struct net_device *dev)
271 struct hlist_head *head;
272 struct hlist_node *node;
276 spin_lock(&fib_info_lock);
278 hash = fib_devindex_hashfn(dev->ifindex);
279 head = &fib_info_devhash[hash];
280 hlist_for_each_entry(nh, node, head, nh_hash) {
281 if (nh->nh_dev == dev &&
283 !(nh->nh_flags & RTNH_F_DEAD)) {
284 spin_unlock(&fib_info_lock);
289 spin_unlock(&fib_info_lock);
294 static inline size_t fib_nlmsg_size(struct fib_info *fi)
296 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
297 + nla_total_size(4) /* RTA_TABLE */
298 + nla_total_size(4) /* RTA_DST */
299 + nla_total_size(4) /* RTA_PRIORITY */
300 + nla_total_size(4); /* RTA_PREFSRC */
302 /* space for nested metrics */
303 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
306 /* Also handles the special case fib_nhs == 1 */
308 /* each nexthop is packed in an attribute */
309 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
311 /* may contain flow and gateway attribute */
312 nhsize += 2 * nla_total_size(4);
314 /* all nexthops are packed in a nested attribute */
315 payload += nla_total_size(fi->fib_nhs * nhsize);
321 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
322 int dst_len, u32 tb_id, struct nl_info *info,
323 unsigned int nlm_flags)
326 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
329 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
333 err = fib_dump_info(skb, info->pid, seq, event, tb_id,
334 fa->fa_type, fa->fa_scope, key, dst_len,
335 fa->fa_tos, fa->fa_info, nlm_flags);
337 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
338 WARN_ON(err == -EMSGSIZE);
342 rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
343 info->nlh, GFP_KERNEL);
347 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
350 /* Return the first fib alias matching TOS with
351 * priority less than or equal to PRIO.
353 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
356 struct fib_alias *fa;
357 list_for_each_entry(fa, fah, fa_list) {
358 if (fa->fa_tos > tos)
360 if (fa->fa_info->fib_priority >= prio ||
368 int fib_detect_death(struct fib_info *fi, int order,
369 struct fib_info **last_resort, int *last_idx, int dflt)
372 int state = NUD_NONE;
374 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
376 state = n->nud_state;
379 if (state == NUD_REACHABLE)
381 if ((state & NUD_VALID) && order != dflt)
383 if ((state & NUD_VALID) ||
384 (*last_idx < 0 && order > dflt)) {
391 #ifdef CONFIG_IP_ROUTE_MULTIPATH
393 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
397 while (rtnh_ok(rtnh, remaining)) {
399 rtnh = rtnh_next(rtnh, &remaining);
402 /* leftover implies invalid nexthop configuration, discard it */
403 return remaining > 0 ? 0 : nhs;
406 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
407 int remaining, struct fib_config *cfg)
409 change_nexthops(fi) {
412 if (!rtnh_ok(rtnh, remaining))
415 nexthop_nh->nh_flags =
416 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
417 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
418 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
420 attrlen = rtnh_attrlen(rtnh);
422 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
424 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
425 nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
426 #ifdef CONFIG_IP_ROUTE_CLASSID
427 nla = nla_find(attrs, attrlen, RTA_FLOW);
428 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
432 rtnh = rtnh_next(rtnh, &remaining);
433 } endfor_nexthops(fi);
440 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
442 #ifdef CONFIG_IP_ROUTE_MULTIPATH
443 struct rtnexthop *rtnh;
447 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
450 if (cfg->fc_oif || cfg->fc_gw) {
451 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
452 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
457 #ifdef CONFIG_IP_ROUTE_MULTIPATH
458 if (cfg->fc_mp == NULL)
462 remaining = cfg->fc_mp_len;
467 if (!rtnh_ok(rtnh, remaining))
470 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
473 attrlen = rtnh_attrlen(rtnh);
475 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
477 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
478 if (nla && nla_get_be32(nla) != nh->nh_gw)
480 #ifdef CONFIG_IP_ROUTE_CLASSID
481 nla = nla_find(attrs, attrlen, RTA_FLOW);
482 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
487 rtnh = rtnh_next(rtnh, &remaining);
488 } endfor_nexthops(fi);
498 * Semantics of nexthop is very messy by historical reasons.
499 * We have to take into account, that:
500 * a) gateway can be actually local interface address,
501 * so that gatewayed route is direct.
502 * b) gateway must be on-link address, possibly
503 * described not by an ifaddr, but also by a direct route.
504 * c) If both gateway and interface are specified, they should not
506 * d) If we use tunnel routes, gateway could be not on-link.
508 * Attempt to reconcile all of these (alas, self-contradictory) conditions
509 * results in pretty ugly and hairy code with obscure logic.
511 * I chose to generalized it instead, so that the size
512 * of code does not increase practically, but it becomes
514 * Every prefix is assigned a "scope" value: "host" is local address,
515 * "link" is direct route,
516 * [ ... "site" ... "interior" ... ]
517 * and "universe" is true gateway route with global meaning.
519 * Every prefix refers to a set of "nexthop"s (gw, oif),
520 * where gw must have narrower scope. This recursion stops
521 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
522 * which means that gw is forced to be on link.
524 * Code is still hairy, but now it is apparently logically
525 * consistent and very flexible. F.e. as by-product it allows
526 * to co-exists in peace independent exterior and interior
529 * Normally it looks as following.
531 * {universe prefix} -> (gw, oif) [scope link]
533 * |-> {link prefix} -> (gw, oif) [scope local]
535 * |-> {local prefix} (terminal node)
537 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
542 struct net_device *dev;
544 net = cfg->fc_nlinfo.nl_net;
546 struct fib_result res;
548 if (nh->nh_flags & RTNH_F_ONLINK) {
550 if (cfg->fc_scope >= RT_SCOPE_LINK)
552 if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
554 dev = __dev_get_by_index(net, nh->nh_oif);
557 if (!(dev->flags & IFF_UP))
561 nh->nh_scope = RT_SCOPE_LINK;
567 .fl4_dst = nh->nh_gw,
568 .fl4_scope = cfg->fc_scope + 1,
572 /* It is not necessary, but requires a bit of thinking */
573 if (fl.fl4_scope < RT_SCOPE_LINK)
574 fl.fl4_scope = RT_SCOPE_LINK;
575 err = fib_lookup(net, &fl, &res);
582 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
584 nh->nh_scope = res.scope;
585 nh->nh_oif = FIB_RES_OIF(res);
586 nh->nh_dev = dev = FIB_RES_DEV(res);
590 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
592 struct in_device *in_dev;
594 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
599 in_dev = inetdev_by_index(net, nh->nh_oif);
603 if (!(in_dev->dev->flags & IFF_UP))
605 nh->nh_dev = in_dev->dev;
606 dev_hold(nh->nh_dev);
607 nh->nh_scope = RT_SCOPE_HOST;
615 static inline unsigned int fib_laddr_hashfn(__be32 val)
617 unsigned int mask = (fib_hash_size - 1);
619 return ((__force u32)val ^
620 ((__force u32)val >> 7) ^
621 ((__force u32)val >> 14)) & mask;
624 static struct hlist_head *fib_hash_alloc(int bytes)
626 if (bytes <= PAGE_SIZE)
627 return kzalloc(bytes, GFP_KERNEL);
629 return (struct hlist_head *)
630 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
634 static void fib_hash_free(struct hlist_head *hash, int bytes)
639 if (bytes <= PAGE_SIZE)
642 free_pages((unsigned long) hash, get_order(bytes));
645 static void fib_hash_move(struct hlist_head *new_info_hash,
646 struct hlist_head *new_laddrhash,
647 unsigned int new_size)
649 struct hlist_head *old_info_hash, *old_laddrhash;
650 unsigned int old_size = fib_hash_size;
651 unsigned int i, bytes;
653 spin_lock_bh(&fib_info_lock);
654 old_info_hash = fib_info_hash;
655 old_laddrhash = fib_info_laddrhash;
656 fib_hash_size = new_size;
658 for (i = 0; i < old_size; i++) {
659 struct hlist_head *head = &fib_info_hash[i];
660 struct hlist_node *node, *n;
663 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
664 struct hlist_head *dest;
665 unsigned int new_hash;
667 hlist_del(&fi->fib_hash);
669 new_hash = fib_info_hashfn(fi);
670 dest = &new_info_hash[new_hash];
671 hlist_add_head(&fi->fib_hash, dest);
674 fib_info_hash = new_info_hash;
676 for (i = 0; i < old_size; i++) {
677 struct hlist_head *lhead = &fib_info_laddrhash[i];
678 struct hlist_node *node, *n;
681 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
682 struct hlist_head *ldest;
683 unsigned int new_hash;
685 hlist_del(&fi->fib_lhash);
687 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
688 ldest = &new_laddrhash[new_hash];
689 hlist_add_head(&fi->fib_lhash, ldest);
692 fib_info_laddrhash = new_laddrhash;
694 spin_unlock_bh(&fib_info_lock);
696 bytes = old_size * sizeof(struct hlist_head *);
697 fib_hash_free(old_info_hash, bytes);
698 fib_hash_free(old_laddrhash, bytes);
701 struct fib_info *fib_create_info(struct fib_config *cfg)
704 struct fib_info *fi = NULL;
705 struct fib_info *ofi;
707 struct net *net = cfg->fc_nlinfo.nl_net;
709 /* Fast check to catch the most weird cases */
710 if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
713 #ifdef CONFIG_IP_ROUTE_MULTIPATH
715 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
722 if (fib_info_cnt >= fib_hash_size) {
723 unsigned int new_size = fib_hash_size << 1;
724 struct hlist_head *new_info_hash;
725 struct hlist_head *new_laddrhash;
730 bytes = new_size * sizeof(struct hlist_head *);
731 new_info_hash = fib_hash_alloc(bytes);
732 new_laddrhash = fib_hash_alloc(bytes);
733 if (!new_info_hash || !new_laddrhash) {
734 fib_hash_free(new_info_hash, bytes);
735 fib_hash_free(new_laddrhash, bytes);
737 fib_hash_move(new_info_hash, new_laddrhash, new_size);
743 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
746 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
747 if (!fi->fib_metrics)
751 fi->fib_net = hold_net(net);
752 fi->fib_protocol = cfg->fc_protocol;
753 fi->fib_flags = cfg->fc_flags;
754 fi->fib_priority = cfg->fc_priority;
755 fi->fib_prefsrc = cfg->fc_prefsrc;
758 change_nexthops(fi) {
759 nexthop_nh->nh_parent = fi;
760 } endfor_nexthops(fi)
766 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
767 int type = nla_type(nla);
772 fi->fib_metrics[type - 1] = nla_get_u32(nla);
778 #ifdef CONFIG_IP_ROUTE_MULTIPATH
779 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
782 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
784 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
786 #ifdef CONFIG_IP_ROUTE_CLASSID
787 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
794 struct fib_nh *nh = fi->fib_nh;
796 nh->nh_oif = cfg->fc_oif;
797 nh->nh_gw = cfg->fc_gw;
798 nh->nh_flags = cfg->fc_flags;
799 #ifdef CONFIG_IP_ROUTE_CLASSID
800 nh->nh_tclassid = cfg->fc_flow;
802 #ifdef CONFIG_IP_ROUTE_MULTIPATH
807 if (fib_props[cfg->fc_type].error) {
808 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
813 if (cfg->fc_scope > RT_SCOPE_HOST)
816 if (cfg->fc_scope == RT_SCOPE_HOST) {
817 struct fib_nh *nh = fi->fib_nh;
819 /* Local address is added. */
820 if (nhs != 1 || nh->nh_gw)
822 nh->nh_scope = RT_SCOPE_NOWHERE;
823 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
825 if (nh->nh_dev == NULL)
828 change_nexthops(fi) {
829 err = fib_check_nh(cfg, fi, nexthop_nh);
832 } endfor_nexthops(fi)
835 if (fi->fib_prefsrc) {
836 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
837 fi->fib_prefsrc != cfg->fc_dst)
838 if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
843 ofi = fib_find_info(fi);
852 atomic_inc(&fi->fib_clntref);
853 spin_lock_bh(&fib_info_lock);
854 hlist_add_head(&fi->fib_hash,
855 &fib_info_hash[fib_info_hashfn(fi)]);
856 if (fi->fib_prefsrc) {
857 struct hlist_head *head;
859 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
860 hlist_add_head(&fi->fib_lhash, head);
862 change_nexthops(fi) {
863 struct hlist_head *head;
866 if (!nexthop_nh->nh_dev)
868 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
869 head = &fib_info_devhash[hash];
870 hlist_add_head(&nexthop_nh->nh_hash, head);
871 } endfor_nexthops(fi)
872 spin_unlock_bh(&fib_info_lock);
887 /* Note! fib_semantic_match intentionally uses RCU list functions. */
888 int fib_semantic_match(struct list_head *head, const struct flowi *flp,
889 struct fib_result *res, int prefixlen, int fib_flags)
891 struct fib_alias *fa;
894 list_for_each_entry_rcu(fa, head, fa_list) {
898 fa->fa_tos != flp->fl4_tos)
901 if (fa->fa_scope < flp->fl4_scope)
904 fib_alias_accessed(fa);
906 err = fib_props[fa->fa_type].error;
908 struct fib_info *fi = fa->fa_info;
910 if (fi->fib_flags & RTNH_F_DEAD)
913 switch (fa->fa_type) {
920 if (nh->nh_flags & RTNH_F_DEAD)
922 if (!flp->oif || flp->oif == nh->nh_oif)
925 #ifdef CONFIG_IP_ROUTE_MULTIPATH
926 if (nhsel < fi->fib_nhs) {
938 pr_warning("fib_semantic_match bad type %#x\n",
948 res->prefixlen = prefixlen;
949 res->nh_sel = nh_sel;
950 res->type = fa->fa_type;
951 res->scope = fa->fa_scope;
952 res->fi = fa->fa_info;
953 if (!(fib_flags & FIB_LOOKUP_NOREF))
954 atomic_inc(&res->fi->fib_clntref);
958 /* Find appropriate source address to this destination */
960 __be32 __fib_res_prefsrc(struct fib_result *res)
962 return inet_select_addr(FIB_RES_DEV(*res), FIB_RES_GW(*res), res->scope);
965 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
966 u32 tb_id, u8 type, u8 scope, __be32 dst, int dst_len, u8 tos,
967 struct fib_info *fi, unsigned int flags)
969 struct nlmsghdr *nlh;
972 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
976 rtm = nlmsg_data(nlh);
977 rtm->rtm_family = AF_INET;
978 rtm->rtm_dst_len = dst_len;
979 rtm->rtm_src_len = 0;
982 rtm->rtm_table = tb_id;
984 rtm->rtm_table = RT_TABLE_COMPAT;
985 NLA_PUT_U32(skb, RTA_TABLE, tb_id);
986 rtm->rtm_type = type;
987 rtm->rtm_flags = fi->fib_flags;
988 rtm->rtm_scope = scope;
989 rtm->rtm_protocol = fi->fib_protocol;
991 if (rtm->rtm_dst_len)
992 NLA_PUT_BE32(skb, RTA_DST, dst);
994 if (fi->fib_priority)
995 NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
997 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
998 goto nla_put_failure;
1000 if (fi->fib_prefsrc)
1001 NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
1003 if (fi->fib_nhs == 1) {
1004 if (fi->fib_nh->nh_gw)
1005 NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
1007 if (fi->fib_nh->nh_oif)
1008 NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
1009 #ifdef CONFIG_IP_ROUTE_CLASSID
1010 if (fi->fib_nh[0].nh_tclassid)
1011 NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
1014 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1015 if (fi->fib_nhs > 1) {
1016 struct rtnexthop *rtnh;
1019 mp = nla_nest_start(skb, RTA_MULTIPATH);
1021 goto nla_put_failure;
1024 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1026 goto nla_put_failure;
1028 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1029 rtnh->rtnh_hops = nh->nh_weight - 1;
1030 rtnh->rtnh_ifindex = nh->nh_oif;
1033 NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
1034 #ifdef CONFIG_IP_ROUTE_CLASSID
1035 if (nh->nh_tclassid)
1036 NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
1038 /* length of rtnetlink header + attributes */
1039 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1040 } endfor_nexthops(fi);
1042 nla_nest_end(skb, mp);
1045 return nlmsg_end(skb, nlh);
1048 nlmsg_cancel(skb, nlh);
1054 * - local address disappeared -> we must delete all the entries
1056 * - device went down -> we must shutdown all nexthops going via it.
1058 int fib_sync_down_addr(struct net *net, __be32 local)
1061 unsigned int hash = fib_laddr_hashfn(local);
1062 struct hlist_head *head = &fib_info_laddrhash[hash];
1063 struct hlist_node *node;
1064 struct fib_info *fi;
1066 if (fib_info_laddrhash == NULL || local == 0)
1069 hlist_for_each_entry(fi, node, head, fib_lhash) {
1070 if (!net_eq(fi->fib_net, net))
1072 if (fi->fib_prefsrc == local) {
1073 fi->fib_flags |= RTNH_F_DEAD;
1080 int fib_sync_down_dev(struct net_device *dev, int force)
1083 int scope = RT_SCOPE_NOWHERE;
1084 struct fib_info *prev_fi = NULL;
1085 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1086 struct hlist_head *head = &fib_info_devhash[hash];
1087 struct hlist_node *node;
1093 hlist_for_each_entry(nh, node, head, nh_hash) {
1094 struct fib_info *fi = nh->nh_parent;
1097 BUG_ON(!fi->fib_nhs);
1098 if (nh->nh_dev != dev || fi == prev_fi)
1102 change_nexthops(fi) {
1103 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1105 else if (nexthop_nh->nh_dev == dev &&
1106 nexthop_nh->nh_scope != scope) {
1107 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1108 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1109 spin_lock_bh(&fib_multipath_lock);
1110 fi->fib_power -= nexthop_nh->nh_power;
1111 nexthop_nh->nh_power = 0;
1112 spin_unlock_bh(&fib_multipath_lock);
1116 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1117 if (force > 1 && nexthop_nh->nh_dev == dev) {
1122 } endfor_nexthops(fi)
1123 if (dead == fi->fib_nhs) {
1124 fi->fib_flags |= RTNH_F_DEAD;
1132 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1135 * Dead device goes up. We wake up dead nexthops.
1136 * It takes sense only on multipath routes.
1138 int fib_sync_up(struct net_device *dev)
1140 struct fib_info *prev_fi;
1142 struct hlist_head *head;
1143 struct hlist_node *node;
1147 if (!(dev->flags & IFF_UP))
1151 hash = fib_devindex_hashfn(dev->ifindex);
1152 head = &fib_info_devhash[hash];
1155 hlist_for_each_entry(nh, node, head, nh_hash) {
1156 struct fib_info *fi = nh->nh_parent;
1159 BUG_ON(!fi->fib_nhs);
1160 if (nh->nh_dev != dev || fi == prev_fi)
1165 change_nexthops(fi) {
1166 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1170 if (nexthop_nh->nh_dev == NULL ||
1171 !(nexthop_nh->nh_dev->flags & IFF_UP))
1173 if (nexthop_nh->nh_dev != dev ||
1174 !__in_dev_get_rtnl(dev))
1177 spin_lock_bh(&fib_multipath_lock);
1178 nexthop_nh->nh_power = 0;
1179 nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1180 spin_unlock_bh(&fib_multipath_lock);
1181 } endfor_nexthops(fi)
1184 fi->fib_flags &= ~RTNH_F_DEAD;
1193 * The algorithm is suboptimal, but it provides really
1194 * fair weighted route distribution.
1196 void fib_select_multipath(const struct flowi *flp, struct fib_result *res)
1198 struct fib_info *fi = res->fi;
1201 spin_lock_bh(&fib_multipath_lock);
1202 if (fi->fib_power <= 0) {
1204 change_nexthops(fi) {
1205 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1206 power += nexthop_nh->nh_weight;
1207 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1209 } endfor_nexthops(fi);
1210 fi->fib_power = power;
1212 spin_unlock_bh(&fib_multipath_lock);
1213 /* Race condition: route has just become dead. */
1220 /* w should be random number [0..fi->fib_power-1],
1221 * it is pretty bad approximation.
1224 w = jiffies % fi->fib_power;
1226 change_nexthops(fi) {
1227 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1228 nexthop_nh->nh_power) {
1229 w -= nexthop_nh->nh_power;
1231 nexthop_nh->nh_power--;
1233 res->nh_sel = nhsel;
1234 spin_unlock_bh(&fib_multipath_lock);
1238 } endfor_nexthops(fi);
1240 /* Race condition: route has just become dead. */
1242 spin_unlock_bh(&fib_multipath_lock);