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 void free_fib_info(struct fib_info *fi)
153 if (fi->fib_dead == 0) {
154 pr_warning("Freeing alive fib_info %p\n", fi);
157 change_nexthops(fi) {
158 if (nexthop_nh->nh_dev)
159 dev_put(nexthop_nh->nh_dev);
160 nexthop_nh->nh_dev = NULL;
161 } endfor_nexthops(fi);
163 release_net(fi->fib_net);
167 void fib_release_info(struct fib_info *fi)
169 spin_lock_bh(&fib_info_lock);
170 if (fi && --fi->fib_treeref == 0) {
171 hlist_del(&fi->fib_hash);
173 hlist_del(&fi->fib_lhash);
174 change_nexthops(fi) {
175 if (!nexthop_nh->nh_dev)
177 hlist_del(&nexthop_nh->nh_hash);
178 } endfor_nexthops(fi)
182 spin_unlock_bh(&fib_info_lock);
185 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
187 const struct fib_nh *onh = ofi->fib_nh;
190 if (nh->nh_oif != onh->nh_oif ||
191 nh->nh_gw != onh->nh_gw ||
192 nh->nh_scope != onh->nh_scope ||
193 #ifdef CONFIG_IP_ROUTE_MULTIPATH
194 nh->nh_weight != onh->nh_weight ||
196 #ifdef CONFIG_NET_CLS_ROUTE
197 nh->nh_tclassid != onh->nh_tclassid ||
199 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_F_DEAD))
202 } endfor_nexthops(fi);
206 static inline unsigned int fib_devindex_hashfn(unsigned int val)
208 unsigned int mask = DEVINDEX_HASHSIZE - 1;
211 (val >> DEVINDEX_HASHBITS) ^
212 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
215 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
217 unsigned int mask = (fib_hash_size - 1);
218 unsigned int val = fi->fib_nhs;
220 val ^= fi->fib_protocol;
221 val ^= (__force u32)fi->fib_prefsrc;
222 val ^= fi->fib_priority;
224 val ^= fib_devindex_hashfn(nh->nh_oif);
225 } endfor_nexthops(fi)
227 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
230 static struct fib_info *fib_find_info(const struct fib_info *nfi)
232 struct hlist_head *head;
233 struct hlist_node *node;
237 hash = fib_info_hashfn(nfi);
238 head = &fib_info_hash[hash];
240 hlist_for_each_entry(fi, node, head, fib_hash) {
241 if (!net_eq(fi->fib_net, nfi->fib_net))
243 if (fi->fib_nhs != nfi->fib_nhs)
245 if (nfi->fib_protocol == fi->fib_protocol &&
246 nfi->fib_prefsrc == fi->fib_prefsrc &&
247 nfi->fib_priority == fi->fib_priority &&
248 memcmp(nfi->fib_metrics, fi->fib_metrics,
249 sizeof(fi->fib_metrics)) == 0 &&
250 ((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_F_DEAD) == 0 &&
251 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
258 /* Check, that the gateway is already configured.
259 * Used only by redirect accept routine.
261 int ip_fib_check_default(__be32 gw, struct net_device *dev)
263 struct hlist_head *head;
264 struct hlist_node *node;
268 spin_lock(&fib_info_lock);
270 hash = fib_devindex_hashfn(dev->ifindex);
271 head = &fib_info_devhash[hash];
272 hlist_for_each_entry(nh, node, head, nh_hash) {
273 if (nh->nh_dev == dev &&
275 !(nh->nh_flags & RTNH_F_DEAD)) {
276 spin_unlock(&fib_info_lock);
281 spin_unlock(&fib_info_lock);
286 static inline size_t fib_nlmsg_size(struct fib_info *fi)
288 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
289 + nla_total_size(4) /* RTA_TABLE */
290 + nla_total_size(4) /* RTA_DST */
291 + nla_total_size(4) /* RTA_PRIORITY */
292 + nla_total_size(4); /* RTA_PREFSRC */
294 /* space for nested metrics */
295 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
298 /* Also handles the special case fib_nhs == 1 */
300 /* each nexthop is packed in an attribute */
301 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
303 /* may contain flow and gateway attribute */
304 nhsize += 2 * nla_total_size(4);
306 /* all nexthops are packed in a nested attribute */
307 payload += nla_total_size(fi->fib_nhs * nhsize);
313 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
314 int dst_len, u32 tb_id, struct nl_info *info,
315 unsigned int nlm_flags)
318 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
321 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
325 err = fib_dump_info(skb, info->pid, seq, event, tb_id,
326 fa->fa_type, fa->fa_scope, key, dst_len,
327 fa->fa_tos, fa->fa_info, nlm_flags);
329 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
330 WARN_ON(err == -EMSGSIZE);
334 rtnl_notify(skb, info->nl_net, info->pid, RTNLGRP_IPV4_ROUTE,
335 info->nlh, GFP_KERNEL);
339 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
342 /* Return the first fib alias matching TOS with
343 * priority less than or equal to PRIO.
345 struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio)
348 struct fib_alias *fa;
349 list_for_each_entry(fa, fah, fa_list) {
350 if (fa->fa_tos > tos)
352 if (fa->fa_info->fib_priority >= prio ||
360 int fib_detect_death(struct fib_info *fi, int order,
361 struct fib_info **last_resort, int *last_idx, int dflt)
364 int state = NUD_NONE;
366 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
368 state = n->nud_state;
371 if (state == NUD_REACHABLE)
373 if ((state & NUD_VALID) && order != dflt)
375 if ((state & NUD_VALID) ||
376 (*last_idx < 0 && order > dflt)) {
383 #ifdef CONFIG_IP_ROUTE_MULTIPATH
385 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
389 while (rtnh_ok(rtnh, remaining)) {
391 rtnh = rtnh_next(rtnh, &remaining);
394 /* leftover implies invalid nexthop configuration, discard it */
395 return remaining > 0 ? 0 : nhs;
398 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
399 int remaining, struct fib_config *cfg)
401 change_nexthops(fi) {
404 if (!rtnh_ok(rtnh, remaining))
407 nexthop_nh->nh_flags =
408 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
409 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
410 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
412 attrlen = rtnh_attrlen(rtnh);
414 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
416 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
417 nexthop_nh->nh_gw = nla ? nla_get_be32(nla) : 0;
418 #ifdef CONFIG_NET_CLS_ROUTE
419 nla = nla_find(attrs, attrlen, RTA_FLOW);
420 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
424 rtnh = rtnh_next(rtnh, &remaining);
425 } endfor_nexthops(fi);
432 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
434 #ifdef CONFIG_IP_ROUTE_MULTIPATH
435 struct rtnexthop *rtnh;
439 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
442 if (cfg->fc_oif || cfg->fc_gw) {
443 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
444 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
449 #ifdef CONFIG_IP_ROUTE_MULTIPATH
450 if (cfg->fc_mp == NULL)
454 remaining = cfg->fc_mp_len;
459 if (!rtnh_ok(rtnh, remaining))
462 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
465 attrlen = rtnh_attrlen(rtnh);
467 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
469 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
470 if (nla && nla_get_be32(nla) != nh->nh_gw)
472 #ifdef CONFIG_NET_CLS_ROUTE
473 nla = nla_find(attrs, attrlen, RTA_FLOW);
474 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
479 rtnh = rtnh_next(rtnh, &remaining);
480 } endfor_nexthops(fi);
490 * Semantics of nexthop is very messy by historical reasons.
491 * We have to take into account, that:
492 * a) gateway can be actually local interface address,
493 * so that gatewayed route is direct.
494 * b) gateway must be on-link address, possibly
495 * described not by an ifaddr, but also by a direct route.
496 * c) If both gateway and interface are specified, they should not
498 * d) If we use tunnel routes, gateway could be not on-link.
500 * Attempt to reconcile all of these (alas, self-contradictory) conditions
501 * results in pretty ugly and hairy code with obscure logic.
503 * I chose to generalized it instead, so that the size
504 * of code does not increase practically, but it becomes
506 * Every prefix is assigned a "scope" value: "host" is local address,
507 * "link" is direct route,
508 * [ ... "site" ... "interior" ... ]
509 * and "universe" is true gateway route with global meaning.
511 * Every prefix refers to a set of "nexthop"s (gw, oif),
512 * where gw must have narrower scope. This recursion stops
513 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
514 * which means that gw is forced to be on link.
516 * Code is still hairy, but now it is apparently logically
517 * consistent and very flexible. F.e. as by-product it allows
518 * to co-exists in peace independent exterior and interior
521 * Normally it looks as following.
523 * {universe prefix} -> (gw, oif) [scope link]
525 * |-> {link prefix} -> (gw, oif) [scope local]
527 * |-> {local prefix} (terminal node)
529 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
534 struct net_device *dev;
536 net = cfg->fc_nlinfo.nl_net;
538 struct fib_result res;
540 if (nh->nh_flags & RTNH_F_ONLINK) {
542 if (cfg->fc_scope >= RT_SCOPE_LINK)
544 if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
546 dev = __dev_get_by_index(net, nh->nh_oif);
549 if (!(dev->flags & IFF_UP))
553 nh->nh_scope = RT_SCOPE_LINK;
561 .scope = cfg->fc_scope + 1,
567 /* It is not necessary, but requires a bit of thinking */
568 if (fl.fl4_scope < RT_SCOPE_LINK)
569 fl.fl4_scope = RT_SCOPE_LINK;
570 err = fib_lookup(net, &fl, &res);
575 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
577 nh->nh_scope = res.scope;
578 nh->nh_oif = FIB_RES_OIF(res);
579 nh->nh_dev = dev = FIB_RES_DEV(res);
584 if (!(dev->flags & IFF_UP))
591 struct in_device *in_dev;
593 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
596 in_dev = inetdev_by_index(net, nh->nh_oif);
599 if (!(in_dev->dev->flags & IFF_UP)) {
603 nh->nh_dev = in_dev->dev;
604 dev_hold(nh->nh_dev);
605 nh->nh_scope = RT_SCOPE_HOST;
611 static inline unsigned int fib_laddr_hashfn(__be32 val)
613 unsigned int mask = (fib_hash_size - 1);
615 return ((__force u32)val ^
616 ((__force u32)val >> 7) ^
617 ((__force u32)val >> 14)) & mask;
620 static struct hlist_head *fib_hash_alloc(int bytes)
622 if (bytes <= PAGE_SIZE)
623 return kzalloc(bytes, GFP_KERNEL);
625 return (struct hlist_head *)
626 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
630 static void fib_hash_free(struct hlist_head *hash, int bytes)
635 if (bytes <= PAGE_SIZE)
638 free_pages((unsigned long) hash, get_order(bytes));
641 static void fib_hash_move(struct hlist_head *new_info_hash,
642 struct hlist_head *new_laddrhash,
643 unsigned int new_size)
645 struct hlist_head *old_info_hash, *old_laddrhash;
646 unsigned int old_size = fib_hash_size;
647 unsigned int i, bytes;
649 spin_lock_bh(&fib_info_lock);
650 old_info_hash = fib_info_hash;
651 old_laddrhash = fib_info_laddrhash;
652 fib_hash_size = new_size;
654 for (i = 0; i < old_size; i++) {
655 struct hlist_head *head = &fib_info_hash[i];
656 struct hlist_node *node, *n;
659 hlist_for_each_entry_safe(fi, node, n, head, fib_hash) {
660 struct hlist_head *dest;
661 unsigned int new_hash;
663 hlist_del(&fi->fib_hash);
665 new_hash = fib_info_hashfn(fi);
666 dest = &new_info_hash[new_hash];
667 hlist_add_head(&fi->fib_hash, dest);
670 fib_info_hash = new_info_hash;
672 for (i = 0; i < old_size; i++) {
673 struct hlist_head *lhead = &fib_info_laddrhash[i];
674 struct hlist_node *node, *n;
677 hlist_for_each_entry_safe(fi, node, n, lhead, fib_lhash) {
678 struct hlist_head *ldest;
679 unsigned int new_hash;
681 hlist_del(&fi->fib_lhash);
683 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
684 ldest = &new_laddrhash[new_hash];
685 hlist_add_head(&fi->fib_lhash, ldest);
688 fib_info_laddrhash = new_laddrhash;
690 spin_unlock_bh(&fib_info_lock);
692 bytes = old_size * sizeof(struct hlist_head *);
693 fib_hash_free(old_info_hash, bytes);
694 fib_hash_free(old_laddrhash, bytes);
697 struct fib_info *fib_create_info(struct fib_config *cfg)
700 struct fib_info *fi = NULL;
701 struct fib_info *ofi;
703 struct net *net = cfg->fc_nlinfo.nl_net;
705 /* Fast check to catch the most weird cases */
706 if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
709 #ifdef CONFIG_IP_ROUTE_MULTIPATH
711 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
718 if (fib_info_cnt >= fib_hash_size) {
719 unsigned int new_size = fib_hash_size << 1;
720 struct hlist_head *new_info_hash;
721 struct hlist_head *new_laddrhash;
726 bytes = new_size * sizeof(struct hlist_head *);
727 new_info_hash = fib_hash_alloc(bytes);
728 new_laddrhash = fib_hash_alloc(bytes);
729 if (!new_info_hash || !new_laddrhash) {
730 fib_hash_free(new_info_hash, bytes);
731 fib_hash_free(new_laddrhash, bytes);
733 fib_hash_move(new_info_hash, new_laddrhash, new_size);
739 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
744 fi->fib_net = hold_net(net);
745 fi->fib_protocol = cfg->fc_protocol;
746 fi->fib_flags = cfg->fc_flags;
747 fi->fib_priority = cfg->fc_priority;
748 fi->fib_prefsrc = cfg->fc_prefsrc;
751 change_nexthops(fi) {
752 nexthop_nh->nh_parent = fi;
753 } endfor_nexthops(fi)
759 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
760 int type = nla_type(nla);
765 fi->fib_metrics[type - 1] = nla_get_u32(nla);
771 #ifdef CONFIG_IP_ROUTE_MULTIPATH
772 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
775 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
777 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
779 #ifdef CONFIG_NET_CLS_ROUTE
780 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
787 struct fib_nh *nh = fi->fib_nh;
789 nh->nh_oif = cfg->fc_oif;
790 nh->nh_gw = cfg->fc_gw;
791 nh->nh_flags = cfg->fc_flags;
792 #ifdef CONFIG_NET_CLS_ROUTE
793 nh->nh_tclassid = cfg->fc_flow;
795 #ifdef CONFIG_IP_ROUTE_MULTIPATH
800 if (fib_props[cfg->fc_type].error) {
801 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
806 if (cfg->fc_scope > RT_SCOPE_HOST)
809 if (cfg->fc_scope == RT_SCOPE_HOST) {
810 struct fib_nh *nh = fi->fib_nh;
812 /* Local address is added. */
813 if (nhs != 1 || nh->nh_gw)
815 nh->nh_scope = RT_SCOPE_NOWHERE;
816 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
818 if (nh->nh_dev == NULL)
821 change_nexthops(fi) {
822 err = fib_check_nh(cfg, fi, nexthop_nh);
825 } endfor_nexthops(fi)
828 if (fi->fib_prefsrc) {
829 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
830 fi->fib_prefsrc != cfg->fc_dst)
831 if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
836 ofi = fib_find_info(fi);
845 atomic_inc(&fi->fib_clntref);
846 spin_lock_bh(&fib_info_lock);
847 hlist_add_head(&fi->fib_hash,
848 &fib_info_hash[fib_info_hashfn(fi)]);
849 if (fi->fib_prefsrc) {
850 struct hlist_head *head;
852 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
853 hlist_add_head(&fi->fib_lhash, head);
855 change_nexthops(fi) {
856 struct hlist_head *head;
859 if (!nexthop_nh->nh_dev)
861 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
862 head = &fib_info_devhash[hash];
863 hlist_add_head(&nexthop_nh->nh_hash, head);
864 } endfor_nexthops(fi)
865 spin_unlock_bh(&fib_info_lock);
880 /* Note! fib_semantic_match intentionally uses RCU list functions. */
881 int fib_semantic_match(struct list_head *head, const struct flowi *flp,
882 struct fib_result *res, int prefixlen)
884 struct fib_alias *fa;
887 list_for_each_entry_rcu(fa, head, fa_list) {
891 fa->fa_tos != flp->fl4_tos)
894 if (fa->fa_scope < flp->fl4_scope)
897 fa->fa_state |= FA_S_ACCESSED;
899 err = fib_props[fa->fa_type].error;
901 struct fib_info *fi = fa->fa_info;
903 if (fi->fib_flags & RTNH_F_DEAD)
906 switch (fa->fa_type) {
913 if (nh->nh_flags & RTNH_F_DEAD)
915 if (!flp->oif || flp->oif == nh->nh_oif)
918 #ifdef CONFIG_IP_ROUTE_MULTIPATH
919 if (nhsel < fi->fib_nhs) {
931 pr_warning("fib_semantic_match bad type %#x\n",
941 res->prefixlen = prefixlen;
942 res->nh_sel = nh_sel;
943 res->type = fa->fa_type;
944 res->scope = fa->fa_scope;
945 res->fi = fa->fa_info;
946 atomic_inc(&res->fi->fib_clntref);
950 /* Find appropriate source address to this destination */
952 __be32 __fib_res_prefsrc(struct fib_result *res)
954 return inet_select_addr(FIB_RES_DEV(*res), FIB_RES_GW(*res), res->scope);
957 int fib_dump_info(struct sk_buff *skb, u32 pid, u32 seq, int event,
958 u32 tb_id, u8 type, u8 scope, __be32 dst, int dst_len, u8 tos,
959 struct fib_info *fi, unsigned int flags)
961 struct nlmsghdr *nlh;
964 nlh = nlmsg_put(skb, pid, seq, event, sizeof(*rtm), flags);
968 rtm = nlmsg_data(nlh);
969 rtm->rtm_family = AF_INET;
970 rtm->rtm_dst_len = dst_len;
971 rtm->rtm_src_len = 0;
974 rtm->rtm_table = tb_id;
976 rtm->rtm_table = RT_TABLE_COMPAT;
977 NLA_PUT_U32(skb, RTA_TABLE, tb_id);
978 rtm->rtm_type = type;
979 rtm->rtm_flags = fi->fib_flags;
980 rtm->rtm_scope = scope;
981 rtm->rtm_protocol = fi->fib_protocol;
983 if (rtm->rtm_dst_len)
984 NLA_PUT_BE32(skb, RTA_DST, dst);
986 if (fi->fib_priority)
987 NLA_PUT_U32(skb, RTA_PRIORITY, fi->fib_priority);
989 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
990 goto nla_put_failure;
993 NLA_PUT_BE32(skb, RTA_PREFSRC, fi->fib_prefsrc);
995 if (fi->fib_nhs == 1) {
996 if (fi->fib_nh->nh_gw)
997 NLA_PUT_BE32(skb, RTA_GATEWAY, fi->fib_nh->nh_gw);
999 if (fi->fib_nh->nh_oif)
1000 NLA_PUT_U32(skb, RTA_OIF, fi->fib_nh->nh_oif);
1001 #ifdef CONFIG_NET_CLS_ROUTE
1002 if (fi->fib_nh[0].nh_tclassid)
1003 NLA_PUT_U32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid);
1006 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1007 if (fi->fib_nhs > 1) {
1008 struct rtnexthop *rtnh;
1011 mp = nla_nest_start(skb, RTA_MULTIPATH);
1013 goto nla_put_failure;
1016 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1018 goto nla_put_failure;
1020 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1021 rtnh->rtnh_hops = nh->nh_weight - 1;
1022 rtnh->rtnh_ifindex = nh->nh_oif;
1025 NLA_PUT_BE32(skb, RTA_GATEWAY, nh->nh_gw);
1026 #ifdef CONFIG_NET_CLS_ROUTE
1027 if (nh->nh_tclassid)
1028 NLA_PUT_U32(skb, RTA_FLOW, nh->nh_tclassid);
1030 /* length of rtnetlink header + attributes */
1031 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1032 } endfor_nexthops(fi);
1034 nla_nest_end(skb, mp);
1037 return nlmsg_end(skb, nlh);
1040 nlmsg_cancel(skb, nlh);
1046 * - local address disappeared -> we must delete all the entries
1048 * - device went down -> we must shutdown all nexthops going via it.
1050 int fib_sync_down_addr(struct net *net, __be32 local)
1053 unsigned int hash = fib_laddr_hashfn(local);
1054 struct hlist_head *head = &fib_info_laddrhash[hash];
1055 struct hlist_node *node;
1056 struct fib_info *fi;
1058 if (fib_info_laddrhash == NULL || local == 0)
1061 hlist_for_each_entry(fi, node, head, fib_lhash) {
1062 if (!net_eq(fi->fib_net, net))
1064 if (fi->fib_prefsrc == local) {
1065 fi->fib_flags |= RTNH_F_DEAD;
1072 int fib_sync_down_dev(struct net_device *dev, int force)
1075 int scope = RT_SCOPE_NOWHERE;
1076 struct fib_info *prev_fi = NULL;
1077 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1078 struct hlist_head *head = &fib_info_devhash[hash];
1079 struct hlist_node *node;
1085 hlist_for_each_entry(nh, node, head, nh_hash) {
1086 struct fib_info *fi = nh->nh_parent;
1089 BUG_ON(!fi->fib_nhs);
1090 if (nh->nh_dev != dev || fi == prev_fi)
1094 change_nexthops(fi) {
1095 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1097 else if (nexthop_nh->nh_dev == dev &&
1098 nexthop_nh->nh_scope != scope) {
1099 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1100 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1101 spin_lock_bh(&fib_multipath_lock);
1102 fi->fib_power -= nexthop_nh->nh_power;
1103 nexthop_nh->nh_power = 0;
1104 spin_unlock_bh(&fib_multipath_lock);
1108 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1109 if (force > 1 && nexthop_nh->nh_dev == dev) {
1114 } endfor_nexthops(fi)
1115 if (dead == fi->fib_nhs) {
1116 fi->fib_flags |= RTNH_F_DEAD;
1124 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1127 * Dead device goes up. We wake up dead nexthops.
1128 * It takes sense only on multipath routes.
1130 int fib_sync_up(struct net_device *dev)
1132 struct fib_info *prev_fi;
1134 struct hlist_head *head;
1135 struct hlist_node *node;
1139 if (!(dev->flags & IFF_UP))
1143 hash = fib_devindex_hashfn(dev->ifindex);
1144 head = &fib_info_devhash[hash];
1147 hlist_for_each_entry(nh, node, head, nh_hash) {
1148 struct fib_info *fi = nh->nh_parent;
1151 BUG_ON(!fi->fib_nhs);
1152 if (nh->nh_dev != dev || fi == prev_fi)
1157 change_nexthops(fi) {
1158 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1162 if (nexthop_nh->nh_dev == NULL ||
1163 !(nexthop_nh->nh_dev->flags & IFF_UP))
1165 if (nexthop_nh->nh_dev != dev ||
1166 !__in_dev_get_rtnl(dev))
1169 spin_lock_bh(&fib_multipath_lock);
1170 nexthop_nh->nh_power = 0;
1171 nexthop_nh->nh_flags &= ~RTNH_F_DEAD;
1172 spin_unlock_bh(&fib_multipath_lock);
1173 } endfor_nexthops(fi)
1176 fi->fib_flags &= ~RTNH_F_DEAD;
1185 * The algorithm is suboptimal, but it provides really
1186 * fair weighted route distribution.
1188 void fib_select_multipath(const struct flowi *flp, struct fib_result *res)
1190 struct fib_info *fi = res->fi;
1193 spin_lock_bh(&fib_multipath_lock);
1194 if (fi->fib_power <= 0) {
1196 change_nexthops(fi) {
1197 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD)) {
1198 power += nexthop_nh->nh_weight;
1199 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1201 } endfor_nexthops(fi);
1202 fi->fib_power = power;
1204 spin_unlock_bh(&fib_multipath_lock);
1205 /* Race condition: route has just become dead. */
1212 /* w should be random number [0..fi->fib_power-1],
1213 * it is pretty bad approximation.
1216 w = jiffies % fi->fib_power;
1218 change_nexthops(fi) {
1219 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1220 nexthop_nh->nh_power) {
1221 w -= nexthop_nh->nh_power;
1223 nexthop_nh->nh_power--;
1225 res->nh_sel = nhsel;
1226 spin_unlock_bh(&fib_multipath_lock);
1230 } endfor_nexthops(fi);
1232 /* Race condition: route has just become dead. */
1234 spin_unlock_bh(&fib_multipath_lock);