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>
45 #include <net/lwtunnel.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_info_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) }
93 const struct fib_prop fib_props[RTN_MAX + 1] = {
96 .scope = RT_SCOPE_NOWHERE,
100 .scope = RT_SCOPE_UNIVERSE,
104 .scope = RT_SCOPE_HOST,
108 .scope = RT_SCOPE_LINK,
112 .scope = RT_SCOPE_LINK,
116 .scope = RT_SCOPE_UNIVERSE,
120 .scope = RT_SCOPE_UNIVERSE,
122 [RTN_UNREACHABLE] = {
123 .error = -EHOSTUNREACH,
124 .scope = RT_SCOPE_UNIVERSE,
128 .scope = RT_SCOPE_UNIVERSE,
132 .scope = RT_SCOPE_UNIVERSE,
136 .scope = RT_SCOPE_NOWHERE,
140 .scope = RT_SCOPE_NOWHERE,
144 static void rt_fibinfo_free(struct rtable __rcu **rtp)
146 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
151 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
152 * because we waited an RCU grace period before calling
153 * free_fib_info_rcu()
159 static void free_nh_exceptions(struct fib_nh *nh)
161 struct fnhe_hash_bucket *hash;
164 hash = rcu_dereference_protected(nh->nh_exceptions, 1);
167 for (i = 0; i < FNHE_HASH_SIZE; i++) {
168 struct fib_nh_exception *fnhe;
170 fnhe = rcu_dereference_protected(hash[i].chain, 1);
172 struct fib_nh_exception *next;
174 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
176 rt_fibinfo_free(&fnhe->fnhe_rth_input);
177 rt_fibinfo_free(&fnhe->fnhe_rth_output);
187 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
194 for_each_possible_cpu(cpu) {
197 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
204 /* Release a nexthop info record */
205 static void free_fib_info_rcu(struct rcu_head *head)
207 struct fib_info *fi = container_of(head, struct fib_info, rcu);
209 change_nexthops(fi) {
210 if (nexthop_nh->nh_dev)
211 dev_put(nexthop_nh->nh_dev);
212 lwtunnel_state_put(nexthop_nh->nh_lwtstate);
213 free_nh_exceptions(nexthop_nh);
214 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
215 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
216 } endfor_nexthops(fi);
218 if (fi->fib_metrics != (u32 *) dst_default_metrics)
219 kfree(fi->fib_metrics);
223 void free_fib_info(struct fib_info *fi)
225 if (fi->fib_dead == 0) {
226 pr_warn("Freeing alive fib_info %p\n", fi);
230 #ifdef CONFIG_IP_ROUTE_CLASSID
231 change_nexthops(fi) {
232 if (nexthop_nh->nh_tclassid)
233 fi->fib_net->ipv4.fib_num_tclassid_users--;
234 } endfor_nexthops(fi);
236 call_rcu(&fi->rcu, free_fib_info_rcu);
239 void fib_release_info(struct fib_info *fi)
241 spin_lock_bh(&fib_info_lock);
242 if (fi && --fi->fib_treeref == 0) {
243 hlist_del(&fi->fib_hash);
245 hlist_del(&fi->fib_lhash);
246 change_nexthops(fi) {
247 if (!nexthop_nh->nh_dev)
249 hlist_del(&nexthop_nh->nh_hash);
250 } endfor_nexthops(fi)
254 spin_unlock_bh(&fib_info_lock);
257 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
259 const struct fib_nh *onh = ofi->fib_nh;
262 if (nh->nh_oif != onh->nh_oif ||
263 nh->nh_gw != onh->nh_gw ||
264 nh->nh_scope != onh->nh_scope ||
265 #ifdef CONFIG_IP_ROUTE_MULTIPATH
266 nh->nh_weight != onh->nh_weight ||
268 #ifdef CONFIG_IP_ROUTE_CLASSID
269 nh->nh_tclassid != onh->nh_tclassid ||
271 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
272 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
275 } endfor_nexthops(fi);
279 static inline unsigned int fib_devindex_hashfn(unsigned int val)
281 unsigned int mask = DEVINDEX_HASHSIZE - 1;
284 (val >> DEVINDEX_HASHBITS) ^
285 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
288 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
290 unsigned int mask = (fib_info_hash_size - 1);
291 unsigned int val = fi->fib_nhs;
293 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
294 val ^= (__force u32)fi->fib_prefsrc;
295 val ^= fi->fib_priority;
297 val ^= fib_devindex_hashfn(nh->nh_oif);
298 } endfor_nexthops(fi)
300 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
303 static struct fib_info *fib_find_info(const struct fib_info *nfi)
305 struct hlist_head *head;
309 hash = fib_info_hashfn(nfi);
310 head = &fib_info_hash[hash];
312 hlist_for_each_entry(fi, head, fib_hash) {
313 if (!net_eq(fi->fib_net, nfi->fib_net))
315 if (fi->fib_nhs != nfi->fib_nhs)
317 if (nfi->fib_protocol == fi->fib_protocol &&
318 nfi->fib_scope == fi->fib_scope &&
319 nfi->fib_prefsrc == fi->fib_prefsrc &&
320 nfi->fib_priority == fi->fib_priority &&
321 nfi->fib_type == fi->fib_type &&
322 memcmp(nfi->fib_metrics, fi->fib_metrics,
323 sizeof(u32) * RTAX_MAX) == 0 &&
324 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
325 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
332 /* Check, that the gateway is already configured.
333 * Used only by redirect accept routine.
335 int ip_fib_check_default(__be32 gw, struct net_device *dev)
337 struct hlist_head *head;
341 spin_lock(&fib_info_lock);
343 hash = fib_devindex_hashfn(dev->ifindex);
344 head = &fib_info_devhash[hash];
345 hlist_for_each_entry(nh, head, nh_hash) {
346 if (nh->nh_dev == dev &&
348 !(nh->nh_flags & RTNH_F_DEAD)) {
349 spin_unlock(&fib_info_lock);
354 spin_unlock(&fib_info_lock);
359 static inline size_t fib_nlmsg_size(struct fib_info *fi)
361 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
362 + nla_total_size(4) /* RTA_TABLE */
363 + nla_total_size(4) /* RTA_DST */
364 + nla_total_size(4) /* RTA_PRIORITY */
365 + nla_total_size(4) /* RTA_PREFSRC */
366 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
368 /* space for nested metrics */
369 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
372 size_t nh_encapsize = 0;
373 /* Also handles the special case fib_nhs == 1 */
375 /* each nexthop is packed in an attribute */
376 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
378 /* may contain flow and gateway attribute */
379 nhsize += 2 * nla_total_size(4);
381 /* grab encap info */
383 if (nh->nh_lwtstate) {
385 nh_encapsize += lwtunnel_get_encap_size(
388 nh_encapsize += nla_total_size(2);
390 } endfor_nexthops(fi);
392 /* all nexthops are packed in a nested attribute */
393 payload += nla_total_size((fi->fib_nhs * nhsize) +
401 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
402 int dst_len, u32 tb_id, const struct nl_info *info,
403 unsigned int nlm_flags)
406 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
409 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
413 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
414 fa->fa_type, key, dst_len,
415 fa->fa_tos, fa->fa_info, nlm_flags);
417 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
418 WARN_ON(err == -EMSGSIZE);
422 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
423 info->nlh, GFP_KERNEL);
427 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
430 static int fib_detect_death(struct fib_info *fi, int order,
431 struct fib_info **last_resort, int *last_idx,
435 int state = NUD_NONE;
437 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
439 state = n->nud_state;
442 if (state == NUD_REACHABLE)
444 if ((state & NUD_VALID) && order != dflt)
446 if ((state & NUD_VALID) ||
447 (*last_idx < 0 && order > dflt)) {
454 #ifdef CONFIG_IP_ROUTE_MULTIPATH
456 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
460 while (rtnh_ok(rtnh, remaining)) {
462 rtnh = rtnh_next(rtnh, &remaining);
465 /* leftover implies invalid nexthop configuration, discard it */
466 return remaining > 0 ? 0 : nhs;
469 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
470 int remaining, struct fib_config *cfg)
472 struct net *net = cfg->fc_nlinfo.nl_net;
475 change_nexthops(fi) {
478 if (!rtnh_ok(rtnh, remaining))
481 nexthop_nh->nh_flags =
482 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
483 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
484 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
486 attrlen = rtnh_attrlen(rtnh);
488 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
490 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
491 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
492 #ifdef CONFIG_IP_ROUTE_CLASSID
493 nla = nla_find(attrs, attrlen, RTA_FLOW);
494 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
495 if (nexthop_nh->nh_tclassid)
496 fi->fib_net->ipv4.fib_num_tclassid_users++;
498 nla = nla_find(attrs, attrlen, RTA_ENCAP);
500 struct lwtunnel_state *lwtstate;
501 struct net_device *dev = NULL;
502 struct nlattr *nla_entype;
504 nla_entype = nla_find(attrs, attrlen,
509 dev = __dev_get_by_index(net, cfg->fc_oif);
510 ret = lwtunnel_build_state(dev, nla_get_u16(
515 lwtunnel_state_get(lwtstate);
516 nexthop_nh->nh_lwtstate = lwtstate;
520 rtnh = rtnh_next(rtnh, &remaining);
521 } endfor_nexthops(fi);
534 int fib_encap_match(struct net *net, u16 encap_type,
535 struct nlattr *encap,
536 int oif, const struct fib_nh *nh)
538 struct lwtunnel_state *lwtstate;
539 struct net_device *dev = NULL;
542 if (encap_type == LWTUNNEL_ENCAP_NONE)
546 dev = __dev_get_by_index(net, oif);
547 ret = lwtunnel_build_state(dev, encap_type,
550 return lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
555 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
557 struct net *net = cfg->fc_nlinfo.nl_net;
558 #ifdef CONFIG_IP_ROUTE_MULTIPATH
559 struct rtnexthop *rtnh;
563 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
566 if (cfg->fc_oif || cfg->fc_gw) {
568 if (fib_encap_match(net, cfg->fc_encap_type,
569 cfg->fc_encap, cfg->fc_oif,
573 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
574 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
579 #ifdef CONFIG_IP_ROUTE_MULTIPATH
584 remaining = cfg->fc_mp_len;
589 if (!rtnh_ok(rtnh, remaining))
592 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
595 attrlen = rtnh_attrlen(rtnh);
597 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
599 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
600 if (nla && nla_get_in_addr(nla) != nh->nh_gw)
602 #ifdef CONFIG_IP_ROUTE_CLASSID
603 nla = nla_find(attrs, attrlen, RTA_FLOW);
604 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
609 rtnh = rtnh_next(rtnh, &remaining);
610 } endfor_nexthops(fi);
620 * Semantics of nexthop is very messy by historical reasons.
621 * We have to take into account, that:
622 * a) gateway can be actually local interface address,
623 * so that gatewayed route is direct.
624 * b) gateway must be on-link address, possibly
625 * described not by an ifaddr, but also by a direct route.
626 * c) If both gateway and interface are specified, they should not
628 * d) If we use tunnel routes, gateway could be not on-link.
630 * Attempt to reconcile all of these (alas, self-contradictory) conditions
631 * results in pretty ugly and hairy code with obscure logic.
633 * I chose to generalized it instead, so that the size
634 * of code does not increase practically, but it becomes
636 * Every prefix is assigned a "scope" value: "host" is local address,
637 * "link" is direct route,
638 * [ ... "site" ... "interior" ... ]
639 * and "universe" is true gateway route with global meaning.
641 * Every prefix refers to a set of "nexthop"s (gw, oif),
642 * where gw must have narrower scope. This recursion stops
643 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
644 * which means that gw is forced to be on link.
646 * Code is still hairy, but now it is apparently logically
647 * consistent and very flexible. F.e. as by-product it allows
648 * to co-exists in peace independent exterior and interior
651 * Normally it looks as following.
653 * {universe prefix} -> (gw, oif) [scope link]
655 * |-> {link prefix} -> (gw, oif) [scope local]
657 * |-> {local prefix} (terminal node)
659 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
664 struct net_device *dev;
666 net = cfg->fc_nlinfo.nl_net;
668 struct fib_result res;
670 if (nh->nh_flags & RTNH_F_ONLINK) {
672 if (cfg->fc_scope >= RT_SCOPE_LINK)
674 if (inet_addr_type(net, nh->nh_gw) != RTN_UNICAST)
676 dev = __dev_get_by_index(net, nh->nh_oif);
679 if (!(dev->flags & IFF_UP))
681 if (!netif_carrier_ok(dev))
682 nh->nh_flags |= RTNH_F_LINKDOWN;
685 nh->nh_scope = RT_SCOPE_LINK;
690 struct flowi4 fl4 = {
692 .flowi4_scope = cfg->fc_scope + 1,
693 .flowi4_oif = nh->nh_oif,
694 .flowi4_iif = LOOPBACK_IFINDEX,
697 /* It is not necessary, but requires a bit of thinking */
698 if (fl4.flowi4_scope < RT_SCOPE_LINK)
699 fl4.flowi4_scope = RT_SCOPE_LINK;
700 err = fib_lookup(net, &fl4, &res,
701 FIB_LOOKUP_IGNORE_LINKSTATE);
708 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
710 nh->nh_scope = res.scope;
711 nh->nh_oif = FIB_RES_OIF(res);
712 nh->nh_dev = dev = FIB_RES_DEV(res);
716 if (!netif_carrier_ok(dev))
717 nh->nh_flags |= RTNH_F_LINKDOWN;
718 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
720 struct in_device *in_dev;
722 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
727 in_dev = inetdev_by_index(net, nh->nh_oif);
731 if (!(in_dev->dev->flags & IFF_UP))
733 nh->nh_dev = in_dev->dev;
734 dev_hold(nh->nh_dev);
735 nh->nh_scope = RT_SCOPE_HOST;
736 if (!netif_carrier_ok(nh->nh_dev))
737 nh->nh_flags |= RTNH_F_LINKDOWN;
745 static inline unsigned int fib_laddr_hashfn(__be32 val)
747 unsigned int mask = (fib_info_hash_size - 1);
749 return ((__force u32)val ^
750 ((__force u32)val >> 7) ^
751 ((__force u32)val >> 14)) & mask;
754 static struct hlist_head *fib_info_hash_alloc(int bytes)
756 if (bytes <= PAGE_SIZE)
757 return kzalloc(bytes, GFP_KERNEL);
759 return (struct hlist_head *)
760 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
764 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
769 if (bytes <= PAGE_SIZE)
772 free_pages((unsigned long) hash, get_order(bytes));
775 static void fib_info_hash_move(struct hlist_head *new_info_hash,
776 struct hlist_head *new_laddrhash,
777 unsigned int new_size)
779 struct hlist_head *old_info_hash, *old_laddrhash;
780 unsigned int old_size = fib_info_hash_size;
781 unsigned int i, bytes;
783 spin_lock_bh(&fib_info_lock);
784 old_info_hash = fib_info_hash;
785 old_laddrhash = fib_info_laddrhash;
786 fib_info_hash_size = new_size;
788 for (i = 0; i < old_size; i++) {
789 struct hlist_head *head = &fib_info_hash[i];
790 struct hlist_node *n;
793 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
794 struct hlist_head *dest;
795 unsigned int new_hash;
797 new_hash = fib_info_hashfn(fi);
798 dest = &new_info_hash[new_hash];
799 hlist_add_head(&fi->fib_hash, dest);
802 fib_info_hash = new_info_hash;
804 for (i = 0; i < old_size; i++) {
805 struct hlist_head *lhead = &fib_info_laddrhash[i];
806 struct hlist_node *n;
809 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
810 struct hlist_head *ldest;
811 unsigned int new_hash;
813 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
814 ldest = &new_laddrhash[new_hash];
815 hlist_add_head(&fi->fib_lhash, ldest);
818 fib_info_laddrhash = new_laddrhash;
820 spin_unlock_bh(&fib_info_lock);
822 bytes = old_size * sizeof(struct hlist_head *);
823 fib_info_hash_free(old_info_hash, bytes);
824 fib_info_hash_free(old_laddrhash, bytes);
827 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
829 nh->nh_saddr = inet_select_addr(nh->nh_dev,
831 nh->nh_parent->fib_scope);
832 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
837 struct fib_info *fib_create_info(struct fib_config *cfg)
840 struct fib_info *fi = NULL;
841 struct fib_info *ofi;
843 struct net *net = cfg->fc_nlinfo.nl_net;
845 if (cfg->fc_type > RTN_MAX)
848 /* Fast check to catch the most weird cases */
849 if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
852 #ifdef CONFIG_IP_ROUTE_MULTIPATH
854 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
861 if (fib_info_cnt >= fib_info_hash_size) {
862 unsigned int new_size = fib_info_hash_size << 1;
863 struct hlist_head *new_info_hash;
864 struct hlist_head *new_laddrhash;
869 bytes = new_size * sizeof(struct hlist_head *);
870 new_info_hash = fib_info_hash_alloc(bytes);
871 new_laddrhash = fib_info_hash_alloc(bytes);
872 if (!new_info_hash || !new_laddrhash) {
873 fib_info_hash_free(new_info_hash, bytes);
874 fib_info_hash_free(new_laddrhash, bytes);
876 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
878 if (!fib_info_hash_size)
882 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
887 fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL);
888 if (!fi->fib_metrics)
891 fi->fib_metrics = (u32 *) dst_default_metrics;
894 fi->fib_protocol = cfg->fc_protocol;
895 fi->fib_scope = cfg->fc_scope;
896 fi->fib_flags = cfg->fc_flags;
897 fi->fib_priority = cfg->fc_priority;
898 fi->fib_prefsrc = cfg->fc_prefsrc;
899 fi->fib_type = cfg->fc_type;
902 change_nexthops(fi) {
903 nexthop_nh->nh_parent = fi;
904 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
905 if (!nexthop_nh->nh_pcpu_rth_output)
907 } endfor_nexthops(fi)
913 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
914 int type = nla_type(nla);
921 if (type == RTAX_CC_ALGO) {
922 char tmp[TCP_CA_NAME_MAX];
924 nla_strlcpy(tmp, nla, sizeof(tmp));
925 val = tcp_ca_get_key_by_name(tmp);
926 if (val == TCP_CA_UNSPEC)
929 val = nla_get_u32(nla);
931 if (type == RTAX_ADVMSS && val > 65535 - 40)
933 if (type == RTAX_MTU && val > 65535 - 15)
935 fi->fib_metrics[type - 1] = val;
941 #ifdef CONFIG_IP_ROUTE_MULTIPATH
942 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
945 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
947 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
949 #ifdef CONFIG_IP_ROUTE_CLASSID
950 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
957 struct fib_nh *nh = fi->fib_nh;
960 struct lwtunnel_state *lwtstate;
961 struct net_device *dev = NULL;
963 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
966 dev = __dev_get_by_index(net, cfg->fc_oif);
967 err = lwtunnel_build_state(dev, cfg->fc_encap_type,
968 cfg->fc_encap, &lwtstate);
972 lwtunnel_state_get(lwtstate);
973 nh->nh_lwtstate = lwtstate;
975 nh->nh_oif = cfg->fc_oif;
976 nh->nh_gw = cfg->fc_gw;
977 nh->nh_flags = cfg->fc_flags;
978 #ifdef CONFIG_IP_ROUTE_CLASSID
979 nh->nh_tclassid = cfg->fc_flow;
981 fi->fib_net->ipv4.fib_num_tclassid_users++;
983 #ifdef CONFIG_IP_ROUTE_MULTIPATH
988 if (fib_props[cfg->fc_type].error) {
989 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
993 switch (cfg->fc_type) {
1005 if (cfg->fc_scope > RT_SCOPE_HOST)
1008 if (cfg->fc_scope == RT_SCOPE_HOST) {
1009 struct fib_nh *nh = fi->fib_nh;
1011 /* Local address is added. */
1012 if (nhs != 1 || nh->nh_gw)
1014 nh->nh_scope = RT_SCOPE_NOWHERE;
1015 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1022 change_nexthops(fi) {
1023 err = fib_check_nh(cfg, fi, nexthop_nh);
1026 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1028 } endfor_nexthops(fi)
1029 if (linkdown == fi->fib_nhs)
1030 fi->fib_flags |= RTNH_F_LINKDOWN;
1033 if (fi->fib_prefsrc) {
1034 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1035 fi->fib_prefsrc != cfg->fc_dst)
1036 if (inet_addr_type(net, fi->fib_prefsrc) != RTN_LOCAL)
1040 change_nexthops(fi) {
1041 fib_info_update_nh_saddr(net, nexthop_nh);
1042 } endfor_nexthops(fi)
1045 ofi = fib_find_info(fi);
1054 atomic_inc(&fi->fib_clntref);
1055 spin_lock_bh(&fib_info_lock);
1056 hlist_add_head(&fi->fib_hash,
1057 &fib_info_hash[fib_info_hashfn(fi)]);
1058 if (fi->fib_prefsrc) {
1059 struct hlist_head *head;
1061 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1062 hlist_add_head(&fi->fib_lhash, head);
1064 change_nexthops(fi) {
1065 struct hlist_head *head;
1068 if (!nexthop_nh->nh_dev)
1070 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1071 head = &fib_info_devhash[hash];
1072 hlist_add_head(&nexthop_nh->nh_hash, head);
1073 } endfor_nexthops(fi)
1074 spin_unlock_bh(&fib_info_lock);
1086 return ERR_PTR(err);
1089 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1090 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1091 struct fib_info *fi, unsigned int flags)
1093 struct nlmsghdr *nlh;
1096 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1100 rtm = nlmsg_data(nlh);
1101 rtm->rtm_family = AF_INET;
1102 rtm->rtm_dst_len = dst_len;
1103 rtm->rtm_src_len = 0;
1106 rtm->rtm_table = tb_id;
1108 rtm->rtm_table = RT_TABLE_COMPAT;
1109 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1110 goto nla_put_failure;
1111 rtm->rtm_type = type;
1112 rtm->rtm_flags = fi->fib_flags;
1113 rtm->rtm_scope = fi->fib_scope;
1114 rtm->rtm_protocol = fi->fib_protocol;
1116 if (rtm->rtm_dst_len &&
1117 nla_put_in_addr(skb, RTA_DST, dst))
1118 goto nla_put_failure;
1119 if (fi->fib_priority &&
1120 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1121 goto nla_put_failure;
1122 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1123 goto nla_put_failure;
1125 if (fi->fib_prefsrc &&
1126 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1127 goto nla_put_failure;
1128 if (fi->fib_nhs == 1) {
1129 struct in_device *in_dev;
1131 if (fi->fib_nh->nh_gw &&
1132 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1133 goto nla_put_failure;
1134 if (fi->fib_nh->nh_oif &&
1135 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1136 goto nla_put_failure;
1137 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1138 in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1140 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1141 rtm->rtm_flags |= RTNH_F_DEAD;
1143 #ifdef CONFIG_IP_ROUTE_CLASSID
1144 if (fi->fib_nh[0].nh_tclassid &&
1145 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1146 goto nla_put_failure;
1148 if (fi->fib_nh->nh_lwtstate)
1149 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate);
1151 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1152 if (fi->fib_nhs > 1) {
1153 struct rtnexthop *rtnh;
1156 mp = nla_nest_start(skb, RTA_MULTIPATH);
1158 goto nla_put_failure;
1161 struct in_device *in_dev;
1163 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1165 goto nla_put_failure;
1167 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1168 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1169 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1171 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1172 rtnh->rtnh_flags |= RTNH_F_DEAD;
1174 rtnh->rtnh_hops = nh->nh_weight - 1;
1175 rtnh->rtnh_ifindex = nh->nh_oif;
1178 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1179 goto nla_put_failure;
1180 #ifdef CONFIG_IP_ROUTE_CLASSID
1181 if (nh->nh_tclassid &&
1182 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1183 goto nla_put_failure;
1185 if (nh->nh_lwtstate)
1186 lwtunnel_fill_encap(skb, nh->nh_lwtstate);
1187 /* length of rtnetlink header + attributes */
1188 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1189 } endfor_nexthops(fi);
1191 nla_nest_end(skb, mp);
1194 nlmsg_end(skb, nlh);
1198 nlmsg_cancel(skb, nlh);
1204 * - local address disappeared -> we must delete all the entries
1206 * - device went down -> we must shutdown all nexthops going via it.
1208 int fib_sync_down_addr(struct net *net, __be32 local)
1211 unsigned int hash = fib_laddr_hashfn(local);
1212 struct hlist_head *head = &fib_info_laddrhash[hash];
1213 struct fib_info *fi;
1215 if (!fib_info_laddrhash || local == 0)
1218 hlist_for_each_entry(fi, head, fib_lhash) {
1219 if (!net_eq(fi->fib_net, net))
1221 if (fi->fib_prefsrc == local) {
1222 fi->fib_flags |= RTNH_F_DEAD;
1229 int fib_sync_down_dev(struct net_device *dev, unsigned long event)
1232 int scope = RT_SCOPE_NOWHERE;
1233 struct fib_info *prev_fi = NULL;
1234 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1235 struct hlist_head *head = &fib_info_devhash[hash];
1238 if (event == NETDEV_UNREGISTER ||
1239 event == NETDEV_DOWN)
1242 hlist_for_each_entry(nh, head, nh_hash) {
1243 struct fib_info *fi = nh->nh_parent;
1246 BUG_ON(!fi->fib_nhs);
1247 if (nh->nh_dev != dev || fi == prev_fi)
1251 change_nexthops(fi) {
1252 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1254 else if (nexthop_nh->nh_dev == dev &&
1255 nexthop_nh->nh_scope != scope) {
1258 case NETDEV_UNREGISTER:
1259 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1262 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1265 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1266 spin_lock_bh(&fib_multipath_lock);
1267 fi->fib_power -= nexthop_nh->nh_power;
1268 nexthop_nh->nh_power = 0;
1269 spin_unlock_bh(&fib_multipath_lock);
1273 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1274 if (event == NETDEV_UNREGISTER &&
1275 nexthop_nh->nh_dev == dev) {
1280 } endfor_nexthops(fi)
1281 if (dead == fi->fib_nhs) {
1284 case NETDEV_UNREGISTER:
1285 fi->fib_flags |= RTNH_F_DEAD;
1288 fi->fib_flags |= RTNH_F_LINKDOWN;
1298 /* Must be invoked inside of an RCU protected region. */
1299 void fib_select_default(struct fib_result *res)
1301 struct fib_info *fi = NULL, *last_resort = NULL;
1302 struct hlist_head *fa_head = res->fa_head;
1303 struct fib_table *tb = res->table;
1304 int order = -1, last_idx = -1;
1305 struct fib_alias *fa;
1307 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1308 struct fib_info *next_fi = fa->fa_info;
1310 if (next_fi->fib_scope != res->scope ||
1311 fa->fa_type != RTN_UNICAST)
1314 if (next_fi->fib_priority > res->fi->fib_priority)
1316 if (!next_fi->fib_nh[0].nh_gw ||
1317 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1320 fib_alias_accessed(fa);
1323 if (next_fi != res->fi)
1325 } else if (!fib_detect_death(fi, order, &last_resort,
1326 &last_idx, tb->tb_default)) {
1327 fib_result_assign(res, fi);
1328 tb->tb_default = order;
1335 if (order <= 0 || !fi) {
1336 tb->tb_default = -1;
1340 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1342 fib_result_assign(res, fi);
1343 tb->tb_default = order;
1348 fib_result_assign(res, last_resort);
1349 tb->tb_default = last_idx;
1355 * Dead device goes up. We wake up dead nexthops.
1356 * It takes sense only on multipath routes.
1358 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1360 struct fib_info *prev_fi;
1362 struct hlist_head *head;
1366 if (!(dev->flags & IFF_UP))
1370 hash = fib_devindex_hashfn(dev->ifindex);
1371 head = &fib_info_devhash[hash];
1374 hlist_for_each_entry(nh, head, nh_hash) {
1375 struct fib_info *fi = nh->nh_parent;
1378 BUG_ON(!fi->fib_nhs);
1379 if (nh->nh_dev != dev || fi == prev_fi)
1384 change_nexthops(fi) {
1385 if (!(nexthop_nh->nh_flags & nh_flags)) {
1389 if (!nexthop_nh->nh_dev ||
1390 !(nexthop_nh->nh_dev->flags & IFF_UP))
1392 if (nexthop_nh->nh_dev != dev ||
1393 !__in_dev_get_rtnl(dev))
1396 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1397 spin_lock_bh(&fib_multipath_lock);
1398 nexthop_nh->nh_power = 0;
1399 nexthop_nh->nh_flags &= ~nh_flags;
1400 spin_unlock_bh(&fib_multipath_lock);
1402 nexthop_nh->nh_flags &= ~nh_flags;
1404 } endfor_nexthops(fi)
1407 fi->fib_flags &= ~nh_flags;
1415 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1418 * The algorithm is suboptimal, but it provides really
1419 * fair weighted route distribution.
1421 void fib_select_multipath(struct fib_result *res)
1423 struct fib_info *fi = res->fi;
1424 struct in_device *in_dev;
1427 spin_lock_bh(&fib_multipath_lock);
1428 if (fi->fib_power <= 0) {
1430 change_nexthops(fi) {
1431 in_dev = __in_dev_get_rcu(nexthop_nh->nh_dev);
1432 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1435 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1436 nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1438 power += nexthop_nh->nh_weight;
1439 nexthop_nh->nh_power = nexthop_nh->nh_weight;
1440 } endfor_nexthops(fi);
1441 fi->fib_power = power;
1443 spin_unlock_bh(&fib_multipath_lock);
1444 /* Race condition: route has just become dead. */
1451 /* w should be random number [0..fi->fib_power-1],
1452 * it is pretty bad approximation.
1455 w = jiffies % fi->fib_power;
1457 change_nexthops(fi) {
1458 if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) &&
1459 nexthop_nh->nh_power) {
1460 w -= nexthop_nh->nh_power;
1462 nexthop_nh->nh_power--;
1464 res->nh_sel = nhsel;
1465 spin_unlock_bh(&fib_multipath_lock);
1469 } endfor_nexthops(fi);
1471 /* Race condition: route has just become dead. */
1473 spin_unlock_bh(&fib_multipath_lock);