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);
207 change_nexthops(fi) {
208 if (nexthop_nh->nh_dev)
209 dev_put(nexthop_nh->nh_dev);
210 lwtstate_put(nexthop_nh->nh_lwtstate);
211 free_nh_exceptions(nexthop_nh);
212 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
213 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
214 } endfor_nexthops(fi);
216 if (fi->fib_metrics != (u32 *) dst_default_metrics)
217 kfree(fi->fib_metrics);
221 void free_fib_info(struct fib_info *fi)
223 if (fi->fib_dead == 0) {
224 pr_warn("Freeing alive fib_info %p\n", fi);
228 #ifdef CONFIG_IP_ROUTE_CLASSID
229 change_nexthops(fi) {
230 if (nexthop_nh->nh_tclassid)
231 fi->fib_net->ipv4.fib_num_tclassid_users--;
232 } endfor_nexthops(fi);
234 call_rcu(&fi->rcu, free_fib_info_rcu);
236 EXPORT_SYMBOL_GPL(free_fib_info);
238 void fib_release_info(struct fib_info *fi)
240 spin_lock_bh(&fib_info_lock);
241 if (fi && --fi->fib_treeref == 0) {
242 hlist_del(&fi->fib_hash);
244 hlist_del(&fi->fib_lhash);
245 change_nexthops(fi) {
246 if (!nexthop_nh->nh_dev)
248 hlist_del(&nexthop_nh->nh_hash);
249 } endfor_nexthops(fi)
253 spin_unlock_bh(&fib_info_lock);
256 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
258 const struct fib_nh *onh = ofi->fib_nh;
261 if (nh->nh_oif != onh->nh_oif ||
262 nh->nh_gw != onh->nh_gw ||
263 nh->nh_scope != onh->nh_scope ||
264 #ifdef CONFIG_IP_ROUTE_MULTIPATH
265 nh->nh_weight != onh->nh_weight ||
267 #ifdef CONFIG_IP_ROUTE_CLASSID
268 nh->nh_tclassid != onh->nh_tclassid ||
270 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
271 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
274 } endfor_nexthops(fi);
278 static inline unsigned int fib_devindex_hashfn(unsigned int val)
280 unsigned int mask = DEVINDEX_HASHSIZE - 1;
283 (val >> DEVINDEX_HASHBITS) ^
284 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
287 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
289 unsigned int mask = (fib_info_hash_size - 1);
290 unsigned int val = fi->fib_nhs;
292 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
293 val ^= (__force u32)fi->fib_prefsrc;
294 val ^= fi->fib_priority;
296 val ^= fib_devindex_hashfn(nh->nh_oif);
297 } endfor_nexthops(fi)
299 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
302 static struct fib_info *fib_find_info(const struct fib_info *nfi)
304 struct hlist_head *head;
308 hash = fib_info_hashfn(nfi);
309 head = &fib_info_hash[hash];
311 hlist_for_each_entry(fi, head, fib_hash) {
312 if (!net_eq(fi->fib_net, nfi->fib_net))
314 if (fi->fib_nhs != nfi->fib_nhs)
316 if (nfi->fib_protocol == fi->fib_protocol &&
317 nfi->fib_scope == fi->fib_scope &&
318 nfi->fib_prefsrc == fi->fib_prefsrc &&
319 nfi->fib_priority == fi->fib_priority &&
320 nfi->fib_type == fi->fib_type &&
321 memcmp(nfi->fib_metrics, fi->fib_metrics,
322 sizeof(u32) * RTAX_MAX) == 0 &&
323 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
324 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
331 /* Check, that the gateway is already configured.
332 * Used only by redirect accept routine.
334 int ip_fib_check_default(__be32 gw, struct net_device *dev)
336 struct hlist_head *head;
340 spin_lock(&fib_info_lock);
342 hash = fib_devindex_hashfn(dev->ifindex);
343 head = &fib_info_devhash[hash];
344 hlist_for_each_entry(nh, head, nh_hash) {
345 if (nh->nh_dev == dev &&
347 !(nh->nh_flags & RTNH_F_DEAD)) {
348 spin_unlock(&fib_info_lock);
353 spin_unlock(&fib_info_lock);
358 static inline size_t fib_nlmsg_size(struct fib_info *fi)
360 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
361 + nla_total_size(4) /* RTA_TABLE */
362 + nla_total_size(4) /* RTA_DST */
363 + nla_total_size(4) /* RTA_PRIORITY */
364 + nla_total_size(4) /* RTA_PREFSRC */
365 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
367 /* space for nested metrics */
368 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
371 size_t nh_encapsize = 0;
372 /* Also handles the special case fib_nhs == 1 */
374 /* each nexthop is packed in an attribute */
375 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
377 /* may contain flow and gateway attribute */
378 nhsize += 2 * nla_total_size(4);
380 /* grab encap info */
382 if (nh->nh_lwtstate) {
384 nh_encapsize += lwtunnel_get_encap_size(
387 nh_encapsize += nla_total_size(2);
389 } endfor_nexthops(fi);
391 /* all nexthops are packed in a nested attribute */
392 payload += nla_total_size((fi->fib_nhs * nhsize) +
400 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
401 int dst_len, u32 tb_id, const struct nl_info *info,
402 unsigned int nlm_flags)
405 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
408 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
412 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
413 fa->fa_type, key, dst_len,
414 fa->fa_tos, fa->fa_info, nlm_flags);
416 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
417 WARN_ON(err == -EMSGSIZE);
421 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
422 info->nlh, GFP_KERNEL);
426 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
429 static int fib_detect_death(struct fib_info *fi, int order,
430 struct fib_info **last_resort, int *last_idx,
434 int state = NUD_NONE;
436 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
438 state = n->nud_state;
443 if (state == NUD_REACHABLE)
445 if ((state & NUD_VALID) && order != dflt)
447 if ((state & NUD_VALID) ||
448 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
455 #ifdef CONFIG_IP_ROUTE_MULTIPATH
457 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
458 struct netlink_ext_ack *extack)
462 while (rtnh_ok(rtnh, remaining)) {
464 rtnh = rtnh_next(rtnh, &remaining);
467 /* leftover implies invalid nexthop configuration, discard it */
468 return remaining > 0 ? 0 : nhs;
471 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
472 int remaining, struct fib_config *cfg,
473 struct netlink_ext_ack *extack)
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,
719 struct fib_nh *nh, struct netlink_ext_ack *extack)
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))
804 in_dev = inetdev_by_index(net, nh->nh_oif);
808 if (!(in_dev->dev->flags & IFF_UP))
810 nh->nh_dev = in_dev->dev;
811 dev_hold(nh->nh_dev);
812 nh->nh_scope = RT_SCOPE_HOST;
813 if (!netif_carrier_ok(nh->nh_dev))
814 nh->nh_flags |= RTNH_F_LINKDOWN;
822 static inline unsigned int fib_laddr_hashfn(__be32 val)
824 unsigned int mask = (fib_info_hash_size - 1);
826 return ((__force u32)val ^
827 ((__force u32)val >> 7) ^
828 ((__force u32)val >> 14)) & mask;
831 static struct hlist_head *fib_info_hash_alloc(int bytes)
833 if (bytes <= PAGE_SIZE)
834 return kzalloc(bytes, GFP_KERNEL);
836 return (struct hlist_head *)
837 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
841 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
846 if (bytes <= PAGE_SIZE)
849 free_pages((unsigned long) hash, get_order(bytes));
852 static void fib_info_hash_move(struct hlist_head *new_info_hash,
853 struct hlist_head *new_laddrhash,
854 unsigned int new_size)
856 struct hlist_head *old_info_hash, *old_laddrhash;
857 unsigned int old_size = fib_info_hash_size;
858 unsigned int i, bytes;
860 spin_lock_bh(&fib_info_lock);
861 old_info_hash = fib_info_hash;
862 old_laddrhash = fib_info_laddrhash;
863 fib_info_hash_size = new_size;
865 for (i = 0; i < old_size; i++) {
866 struct hlist_head *head = &fib_info_hash[i];
867 struct hlist_node *n;
870 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
871 struct hlist_head *dest;
872 unsigned int new_hash;
874 new_hash = fib_info_hashfn(fi);
875 dest = &new_info_hash[new_hash];
876 hlist_add_head(&fi->fib_hash, dest);
879 fib_info_hash = new_info_hash;
881 for (i = 0; i < old_size; i++) {
882 struct hlist_head *lhead = &fib_info_laddrhash[i];
883 struct hlist_node *n;
886 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
887 struct hlist_head *ldest;
888 unsigned int new_hash;
890 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
891 ldest = &new_laddrhash[new_hash];
892 hlist_add_head(&fi->fib_lhash, ldest);
895 fib_info_laddrhash = new_laddrhash;
897 spin_unlock_bh(&fib_info_lock);
899 bytes = old_size * sizeof(struct hlist_head *);
900 fib_info_hash_free(old_info_hash, bytes);
901 fib_info_hash_free(old_laddrhash, bytes);
904 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
906 nh->nh_saddr = inet_select_addr(nh->nh_dev,
908 nh->nh_parent->fib_scope);
909 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
914 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
916 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
917 fib_prefsrc != cfg->fc_dst) {
918 u32 tb_id = cfg->fc_table;
921 if (tb_id == RT_TABLE_MAIN)
922 tb_id = RT_TABLE_LOCAL;
924 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
927 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
928 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
929 fib_prefsrc, RT_TABLE_LOCAL);
939 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
948 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
949 int type = nla_type(nla);
957 if (type == RTAX_CC_ALGO) {
958 char tmp[TCP_CA_NAME_MAX];
960 nla_strlcpy(tmp, nla, sizeof(tmp));
961 val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
962 if (val == TCP_CA_UNSPEC)
965 val = nla_get_u32(nla);
967 if (type == RTAX_ADVMSS && val > 65535 - 40)
969 if (type == RTAX_MTU && val > 65535 - 15)
971 if (type == RTAX_HOPLIMIT && val > 255)
973 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
975 fi->fib_metrics[type - 1] = val;
979 fi->fib_metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
984 struct fib_info *fib_create_info(struct fib_config *cfg,
985 struct netlink_ext_ack *extack)
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, extack);
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(u32) * RTAX_MAX, GFP_KERNEL);
1039 if (!fi->fib_metrics)
1042 fi->fib_metrics = (u32 *) dst_default_metrics;
1045 fi->fib_protocol = cfg->fc_protocol;
1046 fi->fib_scope = cfg->fc_scope;
1047 fi->fib_flags = cfg->fc_flags;
1048 fi->fib_priority = cfg->fc_priority;
1049 fi->fib_prefsrc = cfg->fc_prefsrc;
1050 fi->fib_type = cfg->fc_type;
1051 fi->fib_tb_id = cfg->fc_table;
1054 change_nexthops(fi) {
1055 nexthop_nh->nh_parent = fi;
1056 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1057 if (!nexthop_nh->nh_pcpu_rth_output)
1059 } endfor_nexthops(fi)
1061 err = fib_convert_metrics(fi, cfg);
1066 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1067 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1070 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif)
1072 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw)
1074 #ifdef CONFIG_IP_ROUTE_CLASSID
1075 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow)
1082 struct fib_nh *nh = fi->fib_nh;
1084 if (cfg->fc_encap) {
1085 struct lwtunnel_state *lwtstate;
1087 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE)
1089 err = lwtunnel_build_state(cfg->fc_encap_type,
1090 cfg->fc_encap, AF_INET, cfg,
1095 nh->nh_lwtstate = lwtstate_get(lwtstate);
1097 nh->nh_oif = cfg->fc_oif;
1098 nh->nh_gw = cfg->fc_gw;
1099 nh->nh_flags = cfg->fc_flags;
1100 #ifdef CONFIG_IP_ROUTE_CLASSID
1101 nh->nh_tclassid = cfg->fc_flow;
1102 if (nh->nh_tclassid)
1103 fi->fib_net->ipv4.fib_num_tclassid_users++;
1105 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1110 if (fib_props[cfg->fc_type].error) {
1111 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp)
1115 switch (cfg->fc_type) {
1127 if (cfg->fc_scope > RT_SCOPE_HOST)
1130 if (cfg->fc_scope == RT_SCOPE_HOST) {
1131 struct fib_nh *nh = fi->fib_nh;
1133 /* Local address is added. */
1138 nh->nh_scope = RT_SCOPE_NOWHERE;
1139 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1146 change_nexthops(fi) {
1147 err = fib_check_nh(cfg, fi, nexthop_nh, extack);
1150 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1152 } endfor_nexthops(fi)
1153 if (linkdown == fi->fib_nhs)
1154 fi->fib_flags |= RTNH_F_LINKDOWN;
1157 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc))
1160 change_nexthops(fi) {
1161 fib_info_update_nh_saddr(net, nexthop_nh);
1162 fib_add_weight(fi, nexthop_nh);
1163 } endfor_nexthops(fi)
1168 ofi = fib_find_info(fi);
1177 atomic_inc(&fi->fib_clntref);
1178 spin_lock_bh(&fib_info_lock);
1179 hlist_add_head(&fi->fib_hash,
1180 &fib_info_hash[fib_info_hashfn(fi)]);
1181 if (fi->fib_prefsrc) {
1182 struct hlist_head *head;
1184 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1185 hlist_add_head(&fi->fib_lhash, head);
1187 change_nexthops(fi) {
1188 struct hlist_head *head;
1191 if (!nexthop_nh->nh_dev)
1193 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1194 head = &fib_info_devhash[hash];
1195 hlist_add_head(&nexthop_nh->nh_hash, head);
1196 } endfor_nexthops(fi)
1197 spin_unlock_bh(&fib_info_lock);
1209 return ERR_PTR(err);
1212 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1213 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1214 struct fib_info *fi, unsigned int flags)
1216 struct nlmsghdr *nlh;
1219 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1223 rtm = nlmsg_data(nlh);
1224 rtm->rtm_family = AF_INET;
1225 rtm->rtm_dst_len = dst_len;
1226 rtm->rtm_src_len = 0;
1229 rtm->rtm_table = tb_id;
1231 rtm->rtm_table = RT_TABLE_COMPAT;
1232 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1233 goto nla_put_failure;
1234 rtm->rtm_type = type;
1235 rtm->rtm_flags = fi->fib_flags;
1236 rtm->rtm_scope = fi->fib_scope;
1237 rtm->rtm_protocol = fi->fib_protocol;
1239 if (rtm->rtm_dst_len &&
1240 nla_put_in_addr(skb, RTA_DST, dst))
1241 goto nla_put_failure;
1242 if (fi->fib_priority &&
1243 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1244 goto nla_put_failure;
1245 if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0)
1246 goto nla_put_failure;
1248 if (fi->fib_prefsrc &&
1249 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1250 goto nla_put_failure;
1251 if (fi->fib_nhs == 1) {
1252 struct in_device *in_dev;
1254 if (fi->fib_nh->nh_gw &&
1255 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1256 goto nla_put_failure;
1257 if (fi->fib_nh->nh_oif &&
1258 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1259 goto nla_put_failure;
1260 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1261 in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev);
1263 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1264 rtm->rtm_flags |= RTNH_F_DEAD;
1266 #ifdef CONFIG_IP_ROUTE_CLASSID
1267 if (fi->fib_nh[0].nh_tclassid &&
1268 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1269 goto nla_put_failure;
1271 if (fi->fib_nh->nh_lwtstate &&
1272 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1273 goto nla_put_failure;
1275 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1276 if (fi->fib_nhs > 1) {
1277 struct rtnexthop *rtnh;
1280 mp = nla_nest_start(skb, RTA_MULTIPATH);
1282 goto nla_put_failure;
1285 struct in_device *in_dev;
1287 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1289 goto nla_put_failure;
1291 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1292 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1293 in_dev = __in_dev_get_rtnl(nh->nh_dev);
1295 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1296 rtnh->rtnh_flags |= RTNH_F_DEAD;
1298 rtnh->rtnh_hops = nh->nh_weight - 1;
1299 rtnh->rtnh_ifindex = nh->nh_oif;
1302 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1303 goto nla_put_failure;
1304 #ifdef CONFIG_IP_ROUTE_CLASSID
1305 if (nh->nh_tclassid &&
1306 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1307 goto nla_put_failure;
1309 if (nh->nh_lwtstate &&
1310 lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1311 goto nla_put_failure;
1313 /* length of rtnetlink header + attributes */
1314 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1315 } endfor_nexthops(fi);
1317 nla_nest_end(skb, mp);
1320 nlmsg_end(skb, nlh);
1324 nlmsg_cancel(skb, nlh);
1330 * - local address disappeared -> we must delete all the entries
1332 * - device went down -> we must shutdown all nexthops going via it.
1334 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1337 unsigned int hash = fib_laddr_hashfn(local);
1338 struct hlist_head *head = &fib_info_laddrhash[hash];
1339 struct net *net = dev_net(dev);
1340 int tb_id = l3mdev_fib_table(dev);
1341 struct fib_info *fi;
1343 if (!fib_info_laddrhash || local == 0)
1346 hlist_for_each_entry(fi, head, fib_lhash) {
1347 if (!net_eq(fi->fib_net, net) ||
1348 fi->fib_tb_id != tb_id)
1350 if (fi->fib_prefsrc == local) {
1351 fi->fib_flags |= RTNH_F_DEAD;
1358 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1359 enum fib_event_type event_type)
1361 struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1362 struct fib_nh_notifier_info info = {
1366 switch (event_type) {
1367 case FIB_EVENT_NH_ADD:
1368 if (fib_nh->nh_flags & RTNH_F_DEAD)
1370 if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1371 fib_nh->nh_flags & RTNH_F_LINKDOWN)
1373 return call_fib_notifiers(dev_net(fib_nh->nh_dev), event_type,
1375 case FIB_EVENT_NH_DEL:
1376 if ((IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1377 fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1378 (fib_nh->nh_flags & RTNH_F_DEAD))
1379 return call_fib_notifiers(dev_net(fib_nh->nh_dev),
1380 event_type, &info.info);
1388 /* Event force Flags Description
1389 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1390 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1391 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1392 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1394 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1397 int scope = RT_SCOPE_NOWHERE;
1398 struct fib_info *prev_fi = NULL;
1399 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1400 struct hlist_head *head = &fib_info_devhash[hash];
1406 hlist_for_each_entry(nh, head, nh_hash) {
1407 struct fib_info *fi = nh->nh_parent;
1410 BUG_ON(!fi->fib_nhs);
1411 if (nh->nh_dev != dev || fi == prev_fi)
1415 change_nexthops(fi) {
1416 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1418 else if (nexthop_nh->nh_dev == dev &&
1419 nexthop_nh->nh_scope != scope) {
1422 case NETDEV_UNREGISTER:
1423 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1426 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1429 call_fib_nh_notifiers(nexthop_nh,
1433 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1434 if (event == NETDEV_UNREGISTER &&
1435 nexthop_nh->nh_dev == dev) {
1440 } endfor_nexthops(fi)
1441 if (dead == fi->fib_nhs) {
1444 case NETDEV_UNREGISTER:
1445 fi->fib_flags |= RTNH_F_DEAD;
1448 fi->fib_flags |= RTNH_F_LINKDOWN;
1460 /* Must be invoked inside of an RCU protected region. */
1461 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1463 struct fib_info *fi = NULL, *last_resort = NULL;
1464 struct hlist_head *fa_head = res->fa_head;
1465 struct fib_table *tb = res->table;
1466 u8 slen = 32 - res->prefixlen;
1467 int order = -1, last_idx = -1;
1468 struct fib_alias *fa, *fa1 = NULL;
1469 u32 last_prio = res->fi->fib_priority;
1472 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1473 struct fib_info *next_fi = fa->fa_info;
1475 if (fa->fa_slen != slen)
1477 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1479 if (fa->tb_id != tb->tb_id)
1481 if (next_fi->fib_priority > last_prio &&
1482 fa->fa_tos == last_tos) {
1487 if (next_fi->fib_flags & RTNH_F_DEAD)
1489 last_tos = fa->fa_tos;
1490 last_prio = next_fi->fib_priority;
1492 if (next_fi->fib_scope != res->scope ||
1493 fa->fa_type != RTN_UNICAST)
1495 if (!next_fi->fib_nh[0].nh_gw ||
1496 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1499 fib_alias_accessed(fa);
1502 if (next_fi != res->fi)
1505 } else if (!fib_detect_death(fi, order, &last_resort,
1506 &last_idx, fa1->fa_default)) {
1507 fib_result_assign(res, fi);
1508 fa1->fa_default = order;
1515 if (order <= 0 || !fi) {
1517 fa1->fa_default = -1;
1521 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1523 fib_result_assign(res, fi);
1524 fa1->fa_default = order;
1529 fib_result_assign(res, last_resort);
1530 fa1->fa_default = last_idx;
1536 * Dead device goes up. We wake up dead nexthops.
1537 * It takes sense only on multipath routes.
1539 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1541 struct fib_info *prev_fi;
1543 struct hlist_head *head;
1547 if (!(dev->flags & IFF_UP))
1550 if (nh_flags & RTNH_F_DEAD) {
1551 unsigned int flags = dev_get_flags(dev);
1553 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1554 nh_flags |= RTNH_F_LINKDOWN;
1558 hash = fib_devindex_hashfn(dev->ifindex);
1559 head = &fib_info_devhash[hash];
1562 hlist_for_each_entry(nh, head, nh_hash) {
1563 struct fib_info *fi = nh->nh_parent;
1566 BUG_ON(!fi->fib_nhs);
1567 if (nh->nh_dev != dev || fi == prev_fi)
1572 change_nexthops(fi) {
1573 if (!(nexthop_nh->nh_flags & nh_flags)) {
1577 if (!nexthop_nh->nh_dev ||
1578 !(nexthop_nh->nh_dev->flags & IFF_UP))
1580 if (nexthop_nh->nh_dev != dev ||
1581 !__in_dev_get_rtnl(dev))
1584 nexthop_nh->nh_flags &= ~nh_flags;
1585 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1586 } endfor_nexthops(fi)
1589 fi->fib_flags &= ~nh_flags;
1599 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1600 static bool fib_good_nh(const struct fib_nh *nh)
1602 int state = NUD_REACHABLE;
1604 if (nh->nh_scope == RT_SCOPE_LINK) {
1605 struct neighbour *n;
1609 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1610 (__force u32)nh->nh_gw);
1612 state = n->nud_state;
1614 rcu_read_unlock_bh();
1617 return !!(state & NUD_VALID);
1620 void fib_select_multipath(struct fib_result *res, int hash)
1622 struct fib_info *fi = res->fi;
1623 struct net *net = fi->fib_net;
1627 if (hash > atomic_read(&nh->nh_upper_bound))
1630 if (!net->ipv4.sysctl_fib_multipath_use_neigh ||
1632 res->nh_sel = nhsel;
1636 res->nh_sel = nhsel;
1639 } endfor_nexthops(fi);
1643 void fib_select_path(struct net *net, struct fib_result *res,
1644 struct flowi4 *fl4, const struct sk_buff *skb)
1648 oif_check = (fl4->flowi4_oif == 0 ||
1649 fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1651 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1652 if (res->fi->fib_nhs > 1 && oif_check) {
1653 int h = fib_multipath_hash(res->fi, fl4, skb);
1655 fib_select_multipath(res, h);
1659 if (!res->prefixlen &&
1660 res->table->tb_num_default > 1 &&
1661 res->type == RTN_UNICAST && oif_check)
1662 fib_select_default(fl4, res);
1665 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1667 EXPORT_SYMBOL_GPL(fib_select_path);