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 <linux/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 #define for_nexthops(fi) { \
62 int nhsel; const struct fib_nh *nh; \
63 for (nhsel = 0, nh = (fi)->fib_nh; \
64 nhsel < (fi)->fib_nhs; \
67 #define change_nexthops(fi) { \
68 int nhsel; struct fib_nh *nexthop_nh; \
69 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
70 nhsel < (fi)->fib_nhs; \
71 nexthop_nh++, nhsel++)
73 #else /* CONFIG_IP_ROUTE_MULTIPATH */
75 /* Hope, that gcc will optimize it to get rid of dummy loop */
77 #define for_nexthops(fi) { \
78 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
79 for (nhsel = 0; nhsel < 1; nhsel++)
81 #define change_nexthops(fi) { \
83 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
84 for (nhsel = 0; nhsel < 1; nhsel++)
86 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
88 #define endfor_nexthops(fi) }
91 const struct fib_prop fib_props[RTN_MAX + 1] = {
94 .scope = RT_SCOPE_NOWHERE,
98 .scope = RT_SCOPE_UNIVERSE,
102 .scope = RT_SCOPE_HOST,
106 .scope = RT_SCOPE_LINK,
110 .scope = RT_SCOPE_LINK,
114 .scope = RT_SCOPE_UNIVERSE,
118 .scope = RT_SCOPE_UNIVERSE,
120 [RTN_UNREACHABLE] = {
121 .error = -EHOSTUNREACH,
122 .scope = RT_SCOPE_UNIVERSE,
126 .scope = RT_SCOPE_UNIVERSE,
130 .scope = RT_SCOPE_UNIVERSE,
134 .scope = RT_SCOPE_NOWHERE,
138 .scope = RT_SCOPE_NOWHERE,
142 static void rt_fibinfo_free(struct rtable __rcu **rtp)
144 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
149 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
150 * because we waited an RCU grace period before calling
151 * free_fib_info_rcu()
157 static void free_nh_exceptions(struct fib_nh *nh)
159 struct fnhe_hash_bucket *hash;
162 hash = rcu_dereference_protected(nh->nh_exceptions, 1);
165 for (i = 0; i < FNHE_HASH_SIZE; i++) {
166 struct fib_nh_exception *fnhe;
168 fnhe = rcu_dereference_protected(hash[i].chain, 1);
170 struct fib_nh_exception *next;
172 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
174 rt_fibinfo_free(&fnhe->fnhe_rth_input);
175 rt_fibinfo_free(&fnhe->fnhe_rth_output);
185 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
192 for_each_possible_cpu(cpu) {
195 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
202 /* Release a nexthop info record */
203 static void free_fib_info_rcu(struct rcu_head *head)
205 struct fib_info *fi = container_of(head, struct fib_info, rcu);
206 struct dst_metrics *m;
208 change_nexthops(fi) {
209 if (nexthop_nh->nh_dev)
210 dev_put(nexthop_nh->nh_dev);
211 lwtstate_put(nexthop_nh->nh_lwtstate);
212 free_nh_exceptions(nexthop_nh);
213 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
214 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
215 } endfor_nexthops(fi);
218 if (m != &dst_default_metrics && atomic_dec_and_test(&m->refcnt))
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);
238 EXPORT_SYMBOL_GPL(free_fib_info);
240 void fib_release_info(struct fib_info *fi)
242 spin_lock_bh(&fib_info_lock);
243 if (fi && --fi->fib_treeref == 0) {
244 hlist_del(&fi->fib_hash);
246 hlist_del(&fi->fib_lhash);
247 change_nexthops(fi) {
248 if (!nexthop_nh->nh_dev)
250 hlist_del(&nexthop_nh->nh_hash);
251 } endfor_nexthops(fi)
255 spin_unlock_bh(&fib_info_lock);
258 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
260 const struct fib_nh *onh = ofi->fib_nh;
263 if (nh->nh_oif != onh->nh_oif ||
264 nh->nh_gw != onh->nh_gw ||
265 nh->nh_scope != onh->nh_scope ||
266 #ifdef CONFIG_IP_ROUTE_MULTIPATH
267 nh->nh_weight != onh->nh_weight ||
269 #ifdef CONFIG_IP_ROUTE_CLASSID
270 nh->nh_tclassid != onh->nh_tclassid ||
272 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
273 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
276 } endfor_nexthops(fi);
280 static inline unsigned int fib_devindex_hashfn(unsigned int val)
282 unsigned int mask = DEVINDEX_HASHSIZE - 1;
285 (val >> DEVINDEX_HASHBITS) ^
286 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
289 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
291 unsigned int mask = (fib_info_hash_size - 1);
292 unsigned int val = fi->fib_nhs;
294 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
295 val ^= (__force u32)fi->fib_prefsrc;
296 val ^= fi->fib_priority;
298 val ^= fib_devindex_hashfn(nh->nh_oif);
299 } endfor_nexthops(fi)
301 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
304 static struct fib_info *fib_find_info(const struct fib_info *nfi)
306 struct hlist_head *head;
310 hash = fib_info_hashfn(nfi);
311 head = &fib_info_hash[hash];
313 hlist_for_each_entry(fi, head, fib_hash) {
314 if (!net_eq(fi->fib_net, nfi->fib_net))
316 if (fi->fib_nhs != nfi->fib_nhs)
318 if (nfi->fib_protocol == fi->fib_protocol &&
319 nfi->fib_scope == fi->fib_scope &&
320 nfi->fib_prefsrc == fi->fib_prefsrc &&
321 nfi->fib_priority == fi->fib_priority &&
322 nfi->fib_type == fi->fib_type &&
323 memcmp(nfi->fib_metrics, fi->fib_metrics,
324 sizeof(u32) * RTAX_MAX) == 0 &&
325 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
326 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
333 /* Check, that the gateway is already configured.
334 * Used only by redirect accept routine.
336 int ip_fib_check_default(__be32 gw, struct net_device *dev)
338 struct hlist_head *head;
342 spin_lock(&fib_info_lock);
344 hash = fib_devindex_hashfn(dev->ifindex);
345 head = &fib_info_devhash[hash];
346 hlist_for_each_entry(nh, head, nh_hash) {
347 if (nh->nh_dev == dev &&
349 !(nh->nh_flags & RTNH_F_DEAD)) {
350 spin_unlock(&fib_info_lock);
355 spin_unlock(&fib_info_lock);
360 static inline size_t fib_nlmsg_size(struct fib_info *fi)
362 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
363 + nla_total_size(4) /* RTA_TABLE */
364 + nla_total_size(4) /* RTA_DST */
365 + nla_total_size(4) /* RTA_PRIORITY */
366 + nla_total_size(4) /* RTA_PREFSRC */
367 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
369 /* space for nested metrics */
370 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
373 size_t nh_encapsize = 0;
374 /* Also handles the special case fib_nhs == 1 */
376 /* each nexthop is packed in an attribute */
377 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
379 /* may contain flow and gateway attribute */
380 nhsize += 2 * nla_total_size(4);
382 /* grab encap info */
384 if (nh->nh_lwtstate) {
386 nh_encapsize += lwtunnel_get_encap_size(
389 nh_encapsize += nla_total_size(2);
391 } endfor_nexthops(fi);
393 /* all nexthops are packed in a nested attribute */
394 payload += nla_total_size((fi->fib_nhs * nhsize) +
402 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
403 int dst_len, u32 tb_id, const struct nl_info *info,
404 unsigned int nlm_flags)
407 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
410 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
414 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
415 fa->fa_type, key, dst_len,
416 fa->fa_tos, fa->fa_info, nlm_flags);
418 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
419 WARN_ON(err == -EMSGSIZE);
423 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
424 info->nlh, GFP_KERNEL);
428 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
431 static int fib_detect_death(struct fib_info *fi, int order,
432 struct fib_info **last_resort, int *last_idx,
436 int state = NUD_NONE;
438 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
440 state = n->nud_state;
445 if (state == NUD_REACHABLE)
447 if ((state & NUD_VALID) && order != dflt)
449 if ((state & NUD_VALID) ||
450 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
457 #ifdef CONFIG_IP_ROUTE_MULTIPATH
459 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining)
463 while (rtnh_ok(rtnh, remaining)) {
465 rtnh = rtnh_next(rtnh, &remaining);
468 /* leftover implies invalid nexthop configuration, discard it */
469 return remaining > 0 ? 0 : nhs;
472 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
473 int remaining, struct fib_config *cfg)
477 change_nexthops(fi) {
480 if (!rtnh_ok(rtnh, remaining))
483 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
486 nexthop_nh->nh_flags =
487 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
488 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
489 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
491 attrlen = rtnh_attrlen(rtnh);
493 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
495 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
496 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
497 #ifdef CONFIG_IP_ROUTE_CLASSID
498 nla = nla_find(attrs, attrlen, RTA_FLOW);
499 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
500 if (nexthop_nh->nh_tclassid)
501 fi->fib_net->ipv4.fib_num_tclassid_users++;
503 nla = nla_find(attrs, attrlen, RTA_ENCAP);
505 struct lwtunnel_state *lwtstate;
506 struct nlattr *nla_entype;
508 nla_entype = nla_find(attrs, attrlen,
513 ret = lwtunnel_build_state(nla_get_u16(
519 nexthop_nh->nh_lwtstate =
520 lwtstate_get(lwtstate);
524 rtnh = rtnh_next(rtnh, &remaining);
525 } endfor_nexthops(fi);
536 static void fib_rebalance(struct fib_info *fi)
540 struct in_device *in_dev;
547 if (nh->nh_flags & RTNH_F_DEAD)
550 in_dev = __in_dev_get_rtnl(nh->nh_dev);
553 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
554 nh->nh_flags & RTNH_F_LINKDOWN)
557 total += nh->nh_weight;
558 } endfor_nexthops(fi);
561 change_nexthops(fi) {
564 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
566 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
569 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
570 nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
573 w += nexthop_nh->nh_weight;
574 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
578 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
579 } endfor_nexthops(fi);
582 static inline void fib_add_weight(struct fib_info *fi,
583 const struct fib_nh *nh)
585 fi->fib_weight += nh->nh_weight;
588 #else /* CONFIG_IP_ROUTE_MULTIPATH */
590 #define fib_rebalance(fi) do { } while (0)
591 #define fib_add_weight(fi, nh) do { } while (0)
593 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
595 static int fib_encap_match(u16 encap_type,
596 struct nlattr *encap,
597 const struct fib_nh *nh,
598 const struct fib_config *cfg)
600 struct lwtunnel_state *lwtstate;
603 if (encap_type == LWTUNNEL_ENCAP_NONE)
606 ret = lwtunnel_build_state(encap_type, encap,
607 AF_INET, cfg, &lwtstate);
609 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
610 lwtstate_free(lwtstate);
616 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi)
618 #ifdef CONFIG_IP_ROUTE_MULTIPATH
619 struct rtnexthop *rtnh;
623 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
626 if (cfg->fc_oif || cfg->fc_gw) {
628 if (fib_encap_match(cfg->fc_encap_type,
629 cfg->fc_encap, fi->fib_nh, cfg))
632 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
633 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
638 #ifdef CONFIG_IP_ROUTE_MULTIPATH
643 remaining = cfg->fc_mp_len;
648 if (!rtnh_ok(rtnh, remaining))
651 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
654 attrlen = rtnh_attrlen(rtnh);
656 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
658 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
659 if (nla && nla_get_in_addr(nla) != nh->nh_gw)
661 #ifdef CONFIG_IP_ROUTE_CLASSID
662 nla = nla_find(attrs, attrlen, RTA_FLOW);
663 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
668 rtnh = rtnh_next(rtnh, &remaining);
669 } endfor_nexthops(fi);
679 * Semantics of nexthop is very messy by historical reasons.
680 * We have to take into account, that:
681 * a) gateway can be actually local interface address,
682 * so that gatewayed route is direct.
683 * b) gateway must be on-link address, possibly
684 * described not by an ifaddr, but also by a direct route.
685 * c) If both gateway and interface are specified, they should not
687 * d) If we use tunnel routes, gateway could be not on-link.
689 * Attempt to reconcile all of these (alas, self-contradictory) conditions
690 * results in pretty ugly and hairy code with obscure logic.
692 * I chose to generalized it instead, so that the size
693 * of code does not increase practically, but it becomes
695 * Every prefix is assigned a "scope" value: "host" is local address,
696 * "link" is direct route,
697 * [ ... "site" ... "interior" ... ]
698 * and "universe" is true gateway route with global meaning.
700 * Every prefix refers to a set of "nexthop"s (gw, oif),
701 * where gw must have narrower scope. This recursion stops
702 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
703 * which means that gw is forced to be on link.
705 * Code is still hairy, but now it is apparently logically
706 * consistent and very flexible. F.e. as by-product it allows
707 * to co-exists in peace independent exterior and interior
710 * Normally it looks as following.
712 * {universe prefix} -> (gw, oif) [scope link]
714 * |-> {link prefix} -> (gw, oif) [scope local]
716 * |-> {local prefix} (terminal node)
718 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
723 struct net_device *dev;
725 net = cfg->fc_nlinfo.nl_net;
727 struct fib_result res;
729 if (nh->nh_flags & RTNH_F_ONLINK) {
730 unsigned int addr_type;
732 if (cfg->fc_scope >= RT_SCOPE_LINK)
734 dev = __dev_get_by_index(net, nh->nh_oif);
737 if (!(dev->flags & IFF_UP))
739 addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
740 if (addr_type != RTN_UNICAST)
742 if (!netif_carrier_ok(dev))
743 nh->nh_flags |= RTNH_F_LINKDOWN;
746 nh->nh_scope = RT_SCOPE_LINK;
751 struct fib_table *tbl = NULL;
752 struct flowi4 fl4 = {
754 .flowi4_scope = cfg->fc_scope + 1,
755 .flowi4_oif = nh->nh_oif,
756 .flowi4_iif = LOOPBACK_IFINDEX,
759 /* It is not necessary, but requires a bit of thinking */
760 if (fl4.flowi4_scope < RT_SCOPE_LINK)
761 fl4.flowi4_scope = RT_SCOPE_LINK;
764 tbl = fib_get_table(net, cfg->fc_table);
767 err = fib_table_lookup(tbl, &fl4, &res,
768 FIB_LOOKUP_IGNORE_LINKSTATE |
771 /* on error or if no table given do full lookup. This
772 * is needed for example when nexthops are in the local
773 * table rather than the given table
776 err = fib_lookup(net, &fl4, &res,
777 FIB_LOOKUP_IGNORE_LINKSTATE);
786 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL)
788 nh->nh_scope = res.scope;
789 nh->nh_oif = FIB_RES_OIF(res);
790 nh->nh_dev = dev = FIB_RES_DEV(res);
794 if (!netif_carrier_ok(dev))
795 nh->nh_flags |= RTNH_F_LINKDOWN;
796 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
798 struct in_device *in_dev;
800 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK))
805 in_dev = inetdev_by_index(net, nh->nh_oif);
809 if (!(in_dev->dev->flags & IFF_UP))
811 nh->nh_dev = in_dev->dev;
812 dev_hold(nh->nh_dev);
813 nh->nh_scope = RT_SCOPE_HOST;
814 if (!netif_carrier_ok(nh->nh_dev))
815 nh->nh_flags |= RTNH_F_LINKDOWN;
823 static inline unsigned int fib_laddr_hashfn(__be32 val)
825 unsigned int mask = (fib_info_hash_size - 1);
827 return ((__force u32)val ^
828 ((__force u32)val >> 7) ^
829 ((__force u32)val >> 14)) & mask;
832 static struct hlist_head *fib_info_hash_alloc(int bytes)
834 if (bytes <= PAGE_SIZE)
835 return kzalloc(bytes, GFP_KERNEL);
837 return (struct hlist_head *)
838 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
842 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
847 if (bytes <= PAGE_SIZE)
850 free_pages((unsigned long) hash, get_order(bytes));
853 static void fib_info_hash_move(struct hlist_head *new_info_hash,
854 struct hlist_head *new_laddrhash,
855 unsigned int new_size)
857 struct hlist_head *old_info_hash, *old_laddrhash;
858 unsigned int old_size = fib_info_hash_size;
859 unsigned int i, bytes;
861 spin_lock_bh(&fib_info_lock);
862 old_info_hash = fib_info_hash;
863 old_laddrhash = fib_info_laddrhash;
864 fib_info_hash_size = new_size;
866 for (i = 0; i < old_size; i++) {
867 struct hlist_head *head = &fib_info_hash[i];
868 struct hlist_node *n;
871 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
872 struct hlist_head *dest;
873 unsigned int new_hash;
875 new_hash = fib_info_hashfn(fi);
876 dest = &new_info_hash[new_hash];
877 hlist_add_head(&fi->fib_hash, dest);
880 fib_info_hash = new_info_hash;
882 for (i = 0; i < old_size; i++) {
883 struct hlist_head *lhead = &fib_info_laddrhash[i];
884 struct hlist_node *n;
887 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
888 struct hlist_head *ldest;
889 unsigned int new_hash;
891 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
892 ldest = &new_laddrhash[new_hash];
893 hlist_add_head(&fi->fib_lhash, ldest);
896 fib_info_laddrhash = new_laddrhash;
898 spin_unlock_bh(&fib_info_lock);
900 bytes = old_size * sizeof(struct hlist_head *);
901 fib_info_hash_free(old_info_hash, bytes);
902 fib_info_hash_free(old_laddrhash, bytes);
905 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
907 nh->nh_saddr = inet_select_addr(nh->nh_dev,
909 nh->nh_parent->fib_scope);
910 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
915 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
917 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
918 fib_prefsrc != cfg->fc_dst) {
919 u32 tb_id = cfg->fc_table;
922 if (tb_id == RT_TABLE_MAIN)
923 tb_id = RT_TABLE_LOCAL;
925 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
928 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
929 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
930 fib_prefsrc, RT_TABLE_LOCAL);
940 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
949 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
950 int type = nla_type(nla);
958 if (type == RTAX_CC_ALGO) {
959 char tmp[TCP_CA_NAME_MAX];
961 nla_strlcpy(tmp, nla, sizeof(tmp));
962 val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
963 if (val == TCP_CA_UNSPEC)
966 val = nla_get_u32(nla);
968 if (type == RTAX_ADVMSS && val > 65535 - 40)
970 if (type == RTAX_MTU && val > 65535 - 15)
972 if (type == RTAX_HOPLIMIT && val > 255)
974 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
976 fi->fib_metrics->metrics[type - 1] = val;
980 fi->fib_metrics->metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
985 struct fib_info *fib_create_info(struct fib_config *cfg)
988 struct fib_info *fi = NULL;
989 struct fib_info *ofi;
991 struct net *net = cfg->fc_nlinfo.nl_net;
993 if (cfg->fc_type > RTN_MAX)
996 /* Fast check to catch the most weird cases */
997 if (fib_props[cfg->fc_type].scope > cfg->fc_scope)
1000 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN))
1003 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1005 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len);
1012 if (fib_info_cnt >= fib_info_hash_size) {
1013 unsigned int new_size = fib_info_hash_size << 1;
1014 struct hlist_head *new_info_hash;
1015 struct hlist_head *new_laddrhash;
1020 bytes = new_size * sizeof(struct hlist_head *);
1021 new_info_hash = fib_info_hash_alloc(bytes);
1022 new_laddrhash = fib_info_hash_alloc(bytes);
1023 if (!new_info_hash || !new_laddrhash) {
1024 fib_info_hash_free(new_info_hash, bytes);
1025 fib_info_hash_free(new_laddrhash, bytes);
1027 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1029 if (!fib_info_hash_size)
1033 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1038 fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1039 if (!fi->fib_metrics)
1041 atomic_set(&fi->fib_metrics->refcnt, 1);
1043 fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1046 fi->fib_protocol = cfg->fc_protocol;
1047 fi->fib_scope = cfg->fc_scope;
1048 fi->fib_flags = cfg->fc_flags;
1049 fi->fib_priority = cfg->fc_priority;
1050 fi->fib_prefsrc = cfg->fc_prefsrc;
1051 fi->fib_type = cfg->fc_type;
1052 fi->fib_tb_id = cfg->fc_table;
1055 change_nexthops(fi) {
1056 nexthop_nh->nh_parent = fi;
1057 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1058 if (!nexthop_nh->nh_pcpu_rth_output)
1060 } endfor_nexthops(fi)
1062 err = fib_convert_metrics(fi, cfg);
1067 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1068 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg);
1071 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1073 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1075 #ifdef CONFIG_IP_ROUTE_CLASSID
1076 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1083 struct fib_nh *nh = fi->fib_nh;
1085 if (cfg->fc_encap) {
1086 struct lwtunnel_state *lwtstate;
1088 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1090 err = lwtunnel_build_state(cfg->fc_encap_type,
1091 cfg->fc_encap, AF_INET, cfg,
1096 nh->nh_lwtstate = lwtstate_get(lwtstate);
1098 nh->nh_oif = cfg->fc_oif;
1099 nh->nh_gw = cfg->fc_gw;
1100 nh->nh_flags = cfg->fc_flags;
1101 #ifdef CONFIG_IP_ROUTE_CLASSID
1102 nh->nh_tclassid = cfg->fc_flow;
1103 if (nh->nh_tclassid)
1104 fi->fib_net->ipv4.fib_num_tclassid_users++;
1106 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1111 if (fib_props[cfg->fc_type].error) {
1112 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1116 switch (cfg->fc_type) {
1128 if (cfg->fc_scope > RT_SCOPE_HOST)
1131 if (cfg->fc_scope == RT_SCOPE_HOST) {
1132 struct fib_nh *nh = fi->fib_nh;
1134 /* Local address is added. */
1135 if (nhs != 1 || nh->nh_gw)
1137 nh->nh_scope = RT_SCOPE_NOWHERE;
1138 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1145 change_nexthops(fi) {
1146 err = fib_check_nh(cfg, fi, nexthop_nh);
1149 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1151 } endfor_nexthops(fi)
1152 if (linkdown == fi->fib_nhs)
1153 fi->fib_flags |= RTNH_F_LINKDOWN;
1156 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1159 change_nexthops(fi) {
1160 fib_info_update_nh_saddr(net, nexthop_nh);
1161 fib_add_weight(fi, nexthop_nh);
1162 } endfor_nexthops(fi)
1167 ofi = fib_find_info(fi);
1176 atomic_inc(&fi->fib_clntref);
1177 spin_lock_bh(&fib_info_lock);
1178 hlist_add_head(&fi->fib_hash,
1179 &fib_info_hash[fib_info_hashfn(fi)]);
1180 if (fi->fib_prefsrc) {
1181 struct hlist_head *head;
1183 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1184 hlist_add_head(&fi->fib_lhash, head);
1186 change_nexthops(fi) {
1187 struct hlist_head *head;
1190 if (!nexthop_nh->nh_dev)
1192 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1193 head = &fib_info_devhash[hash];
1194 hlist_add_head(&nexthop_nh->nh_hash, head);
1195 } endfor_nexthops(fi)
1196 spin_unlock_bh(&fib_info_lock);
1208 return ERR_PTR(err);
1211 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1212 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1213 struct fib_info *fi, unsigned int flags)
1215 struct nlmsghdr *nlh;
1218 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1222 rtm = nlmsg_data(nlh);
1223 rtm->rtm_family = AF_INET;
1224 rtm->rtm_dst_len = dst_len;
1225 rtm->rtm_src_len = 0;
1228 rtm->rtm_table = tb_id;
1230 rtm->rtm_table = RT_TABLE_COMPAT;
1231 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1232 goto nla_put_failure;
1233 rtm->rtm_type = type;
1234 rtm->rtm_flags = fi->fib_flags;
1235 rtm->rtm_scope = fi->fib_scope;
1236 rtm->rtm_protocol = fi->fib_protocol;
1238 if (rtm->rtm_dst_len &&
1239 nla_put_in_addr(skb, RTA_DST, dst))
1240 goto nla_put_failure;
1241 if (fi->fib_priority &&
1242 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1243 goto nla_put_failure;
1244 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1245 goto nla_put_failure;
1247 if (fi->fib_prefsrc &&
1248 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1249 goto nla_put_failure;
1250 if (fi->fib_nhs == 1) {
1251 struct in_device *in_dev;
1253 if (fi->fib_nh->nh_gw &&
1254 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1255 goto nla_put_failure;
1256 if (fi->fib_nh->nh_oif &&
1257 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1258 goto nla_put_failure;
1259 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1260 in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1262 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1263 rtm->rtm_flags |= RTNH_F_DEAD;
1265 #ifdef CONFIG_IP_ROUTE_CLASSID
1266 if (fi->fib_nh[0].nh_tclassid &&
1267 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1268 goto nla_put_failure;
1270 if (fi->fib_nh->nh_lwtstate &&
1271 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1272 goto nla_put_failure;
1274 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1275 if (fi->fib_nhs > 1) {
1276 struct rtnexthop *rtnh;
1279 mp = nla_nest_start(skb, RTA_MULTIPATH);
1281 goto nla_put_failure;
1284 struct in_device *in_dev;
1286 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1288 goto nla_put_failure;
1290 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1291 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1292 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1294 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1295 rtnh->rtnh_flags |= RTNH_F_DEAD;
1297 rtnh->rtnh_hops = nh->nh_weight - 1;
1298 rtnh->rtnh_ifindex = nh->nh_oif;
1301 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1302 goto nla_put_failure;
1303 #ifdef CONFIG_IP_ROUTE_CLASSID
1304 if (nh->nh_tclassid &&
1305 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1306 goto nla_put_failure;
1308 if (nh->nh_lwtstate &&
1309 lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1310 goto nla_put_failure;
1312 /* length of rtnetlink header + attributes */
1313 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1314 } endfor_nexthops(fi);
1316 nla_nest_end(skb, mp);
1319 nlmsg_end(skb, nlh);
1323 nlmsg_cancel(skb, nlh);
1329 * - local address disappeared -> we must delete all the entries
1331 * - device went down -> we must shutdown all nexthops going via it.
1333 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1336 unsigned int hash = fib_laddr_hashfn(local);
1337 struct hlist_head *head = &fib_info_laddrhash[hash];
1338 struct net *net = dev_net(dev);
1339 int tb_id = l3mdev_fib_table(dev);
1340 struct fib_info *fi;
1342 if (!fib_info_laddrhash || local == 0)
1345 hlist_for_each_entry(fi, head, fib_lhash) {
1346 if (!net_eq(fi->fib_net, net) ||
1347 fi->fib_tb_id != tb_id)
1349 if (fi->fib_prefsrc == local) {
1350 fi->fib_flags |= RTNH_F_DEAD;
1357 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1358 enum fib_event_type event_type)
1360 struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1361 struct fib_nh_notifier_info info = {
1365 switch (event_type) {
1366 case FIB_EVENT_NH_ADD:
1367 if (fib_nh->nh_flags & RTNH_F_DEAD)
1369 if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1370 fib_nh->nh_flags & RTNH_F_LINKDOWN)
1372 return call_fib_notifiers(dev_net(fib_nh->nh_dev), event_type,
1374 case FIB_EVENT_NH_DEL:
1375 if ((IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1376 fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1377 (fib_nh->nh_flags & RTNH_F_DEAD))
1378 return call_fib_notifiers(dev_net(fib_nh->nh_dev),
1379 event_type, &info.info);
1387 /* Event force Flags Description
1388 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1389 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1390 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1391 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1393 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1396 int scope = RT_SCOPE_NOWHERE;
1397 struct fib_info *prev_fi = NULL;
1398 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1399 struct hlist_head *head = &fib_info_devhash[hash];
1405 hlist_for_each_entry(nh, head, nh_hash) {
1406 struct fib_info *fi = nh->nh_parent;
1409 BUG_ON(!fi->fib_nhs);
1410 if (nh->nh_dev != dev || fi == prev_fi)
1414 change_nexthops(fi) {
1415 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1417 else if (nexthop_nh->nh_dev == dev &&
1418 nexthop_nh->nh_scope != scope) {
1421 case NETDEV_UNREGISTER:
1422 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1425 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1428 call_fib_nh_notifiers(nexthop_nh,
1432 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1433 if (event == NETDEV_UNREGISTER &&
1434 nexthop_nh->nh_dev == dev) {
1439 } endfor_nexthops(fi)
1440 if (dead == fi->fib_nhs) {
1443 case NETDEV_UNREGISTER:
1444 fi->fib_flags |= RTNH_F_DEAD;
1447 fi->fib_flags |= RTNH_F_LINKDOWN;
1459 /* Must be invoked inside of an RCU protected region. */
1460 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1462 struct fib_info *fi = NULL, *last_resort = NULL;
1463 struct hlist_head *fa_head = res->fa_head;
1464 struct fib_table *tb = res->table;
1465 u8 slen = 32 - res->prefixlen;
1466 int order = -1, last_idx = -1;
1467 struct fib_alias *fa, *fa1 = NULL;
1468 u32 last_prio = res->fi->fib_priority;
1471 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1472 struct fib_info *next_fi = fa->fa_info;
1474 if (fa->fa_slen != slen)
1476 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1478 if (fa->tb_id != tb->tb_id)
1480 if (next_fi->fib_priority > last_prio &&
1481 fa->fa_tos == last_tos) {
1486 if (next_fi->fib_flags & RTNH_F_DEAD)
1488 last_tos = fa->fa_tos;
1489 last_prio = next_fi->fib_priority;
1491 if (next_fi->fib_scope != res->scope ||
1492 fa->fa_type != RTN_UNICAST)
1494 if (!next_fi->fib_nh[0].nh_gw ||
1495 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1498 fib_alias_accessed(fa);
1501 if (next_fi != res->fi)
1504 } else if (!fib_detect_death(fi, order, &last_resort,
1505 &last_idx, fa1->fa_default)) {
1506 fib_result_assign(res, fi);
1507 fa1->fa_default = order;
1514 if (order <= 0 || !fi) {
1516 fa1->fa_default = -1;
1520 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1522 fib_result_assign(res, fi);
1523 fa1->fa_default = order;
1528 fib_result_assign(res, last_resort);
1529 fa1->fa_default = last_idx;
1535 * Dead device goes up. We wake up dead nexthops.
1536 * It takes sense only on multipath routes.
1538 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1540 struct fib_info *prev_fi;
1542 struct hlist_head *head;
1546 if (!(dev->flags & IFF_UP))
1549 if (nh_flags & RTNH_F_DEAD) {
1550 unsigned int flags = dev_get_flags(dev);
1552 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1553 nh_flags |= RTNH_F_LINKDOWN;
1557 hash = fib_devindex_hashfn(dev->ifindex);
1558 head = &fib_info_devhash[hash];
1561 hlist_for_each_entry(nh, head, nh_hash) {
1562 struct fib_info *fi = nh->nh_parent;
1565 BUG_ON(!fi->fib_nhs);
1566 if (nh->nh_dev != dev || fi == prev_fi)
1571 change_nexthops(fi) {
1572 if (!(nexthop_nh->nh_flags & nh_flags)) {
1576 if (!nexthop_nh->nh_dev ||
1577 !(nexthop_nh->nh_dev->flags & IFF_UP))
1579 if (nexthop_nh->nh_dev != dev ||
1580 !__in_dev_get_rtnl(dev))
1583 nexthop_nh->nh_flags &= ~nh_flags;
1584 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1585 } endfor_nexthops(fi)
1588 fi->fib_flags &= ~nh_flags;
1598 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1599 static bool fib_good_nh(const struct fib_nh *nh)
1601 int state = NUD_REACHABLE;
1603 if (nh->nh_scope == RT_SCOPE_LINK) {
1604 struct neighbour *n;
1608 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1609 (__force u32)nh->nh_gw);
1611 state = n->nud_state;
1613 rcu_read_unlock_bh();
1616 return !!(state & NUD_VALID);
1619 void fib_select_multipath(struct fib_result *res, int hash)
1621 struct fib_info *fi = res->fi;
1622 struct net *net = fi->fib_net;
1626 if (hash > atomic_read(&nh->nh_upper_bound))
1629 if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1631 res->nh_sel = nhsel;
1635 res->nh_sel = nhsel;
1638 } endfor_nexthops(fi);
1642 void fib_select_path(struct net *net, struct fib_result *res,
1643 struct flowi4 *fl4, const struct sk_buff *skb)
1647 oif_check = (fl4->flowi4_oif == 0 ||
1648 fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1650 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1651 if (res->fi->fib_nhs > 1 && oif_check) {
1652 int h = fib_multipath_hash(res->fi, fl4, skb);
1654 fib_select_multipath(res, h);
1658 if (!res->prefixlen &&
1659 res->table->tb_num_default > 1 &&
1660 res->type == RTN_UNICAST && oif_check)
1661 fib_select_default(fl4, res);
1664 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1666 EXPORT_SYMBOL_GPL(fib_select_path);