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>
35 #include <linux/netlink.h>
39 #include <net/protocol.h>
40 #include <net/route.h>
43 #include <net/ip_fib.h>
44 #include <net/netlink.h>
45 #include <net/nexthop.h>
46 #include <net/lwtunnel.h>
47 #include <net/fib_notifier.h>
49 #include "fib_lookup.h"
51 static DEFINE_SPINLOCK(fib_info_lock);
52 static struct hlist_head *fib_info_hash;
53 static struct hlist_head *fib_info_laddrhash;
54 static unsigned int fib_info_hash_size;
55 static unsigned int fib_info_cnt;
57 #define DEVINDEX_HASHBITS 8
58 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
59 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
61 #ifdef CONFIG_IP_ROUTE_MULTIPATH
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()
156 dst_dev_put(&rt->dst);
157 dst_release_immediate(&rt->dst);
160 static void free_nh_exceptions(struct fib_nh *nh)
162 struct fnhe_hash_bucket *hash;
165 hash = rcu_dereference_protected(nh->nh_exceptions, 1);
168 for (i = 0; i < FNHE_HASH_SIZE; i++) {
169 struct fib_nh_exception *fnhe;
171 fnhe = rcu_dereference_protected(hash[i].chain, 1);
173 struct fib_nh_exception *next;
175 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
177 rt_fibinfo_free(&fnhe->fnhe_rth_input);
178 rt_fibinfo_free(&fnhe->fnhe_rth_output);
188 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
195 for_each_possible_cpu(cpu) {
198 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
200 dst_dev_put(&rt->dst);
201 dst_release_immediate(&rt->dst);
207 /* Release a nexthop info record */
208 static void free_fib_info_rcu(struct rcu_head *head)
210 struct fib_info *fi = container_of(head, struct fib_info, rcu);
211 struct dst_metrics *m;
213 change_nexthops(fi) {
214 if (nexthop_nh->nh_dev)
215 dev_put(nexthop_nh->nh_dev);
216 lwtstate_put(nexthop_nh->nh_lwtstate);
217 free_nh_exceptions(nexthop_nh);
218 rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output);
219 rt_fibinfo_free(&nexthop_nh->nh_rth_input);
220 } endfor_nexthops(fi);
223 if (m != &dst_default_metrics && refcount_dec_and_test(&m->refcnt))
228 void free_fib_info(struct fib_info *fi)
230 if (fi->fib_dead == 0) {
231 pr_warn("Freeing alive fib_info %p\n", fi);
235 #ifdef CONFIG_IP_ROUTE_CLASSID
236 change_nexthops(fi) {
237 if (nexthop_nh->nh_tclassid)
238 fi->fib_net->ipv4.fib_num_tclassid_users--;
239 } endfor_nexthops(fi);
241 call_rcu(&fi->rcu, free_fib_info_rcu);
243 EXPORT_SYMBOL_GPL(free_fib_info);
245 void fib_release_info(struct fib_info *fi)
247 spin_lock_bh(&fib_info_lock);
248 if (fi && --fi->fib_treeref == 0) {
249 hlist_del(&fi->fib_hash);
251 hlist_del(&fi->fib_lhash);
252 change_nexthops(fi) {
253 if (!nexthop_nh->nh_dev)
255 hlist_del(&nexthop_nh->nh_hash);
256 } endfor_nexthops(fi)
260 spin_unlock_bh(&fib_info_lock);
263 static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi)
265 const struct fib_nh *onh = ofi->fib_nh;
268 if (nh->nh_oif != onh->nh_oif ||
269 nh->nh_gw != onh->nh_gw ||
270 nh->nh_scope != onh->nh_scope ||
271 #ifdef CONFIG_IP_ROUTE_MULTIPATH
272 nh->nh_weight != onh->nh_weight ||
274 #ifdef CONFIG_IP_ROUTE_CLASSID
275 nh->nh_tclassid != onh->nh_tclassid ||
277 lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) ||
278 ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK))
281 } endfor_nexthops(fi);
285 static inline unsigned int fib_devindex_hashfn(unsigned int val)
287 unsigned int mask = DEVINDEX_HASHSIZE - 1;
290 (val >> DEVINDEX_HASHBITS) ^
291 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
294 static inline unsigned int fib_info_hashfn(const struct fib_info *fi)
296 unsigned int mask = (fib_info_hash_size - 1);
297 unsigned int val = fi->fib_nhs;
299 val ^= (fi->fib_protocol << 8) | fi->fib_scope;
300 val ^= (__force u32)fi->fib_prefsrc;
301 val ^= fi->fib_priority;
303 val ^= fib_devindex_hashfn(nh->nh_oif);
304 } endfor_nexthops(fi)
306 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
309 static struct fib_info *fib_find_info(const struct fib_info *nfi)
311 struct hlist_head *head;
315 hash = fib_info_hashfn(nfi);
316 head = &fib_info_hash[hash];
318 hlist_for_each_entry(fi, head, fib_hash) {
319 if (!net_eq(fi->fib_net, nfi->fib_net))
321 if (fi->fib_nhs != nfi->fib_nhs)
323 if (nfi->fib_protocol == fi->fib_protocol &&
324 nfi->fib_scope == fi->fib_scope &&
325 nfi->fib_prefsrc == fi->fib_prefsrc &&
326 nfi->fib_priority == fi->fib_priority &&
327 nfi->fib_type == fi->fib_type &&
328 memcmp(nfi->fib_metrics, fi->fib_metrics,
329 sizeof(u32) * RTAX_MAX) == 0 &&
330 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
331 (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0))
338 /* Check, that the gateway is already configured.
339 * Used only by redirect accept routine.
341 int ip_fib_check_default(__be32 gw, struct net_device *dev)
343 struct hlist_head *head;
347 spin_lock(&fib_info_lock);
349 hash = fib_devindex_hashfn(dev->ifindex);
350 head = &fib_info_devhash[hash];
351 hlist_for_each_entry(nh, head, nh_hash) {
352 if (nh->nh_dev == dev &&
354 !(nh->nh_flags & RTNH_F_DEAD)) {
355 spin_unlock(&fib_info_lock);
360 spin_unlock(&fib_info_lock);
365 static inline size_t fib_nlmsg_size(struct fib_info *fi)
367 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
368 + nla_total_size(4) /* RTA_TABLE */
369 + nla_total_size(4) /* RTA_DST */
370 + nla_total_size(4) /* RTA_PRIORITY */
371 + nla_total_size(4) /* RTA_PREFSRC */
372 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
374 /* space for nested metrics */
375 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
378 size_t nh_encapsize = 0;
379 /* Also handles the special case fib_nhs == 1 */
381 /* each nexthop is packed in an attribute */
382 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
384 /* may contain flow and gateway attribute */
385 nhsize += 2 * nla_total_size(4);
387 /* grab encap info */
389 if (nh->nh_lwtstate) {
391 nh_encapsize += lwtunnel_get_encap_size(
394 nh_encapsize += nla_total_size(2);
396 } endfor_nexthops(fi);
398 /* all nexthops are packed in a nested attribute */
399 payload += nla_total_size((fi->fib_nhs * nhsize) +
407 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
408 int dst_len, u32 tb_id, const struct nl_info *info,
409 unsigned int nlm_flags)
412 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
415 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
419 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
420 fa->fa_type, key, dst_len,
421 fa->fa_tos, fa->fa_info, nlm_flags);
423 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
424 WARN_ON(err == -EMSGSIZE);
428 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
429 info->nlh, GFP_KERNEL);
433 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
436 static int fib_detect_death(struct fib_info *fi, int order,
437 struct fib_info **last_resort, int *last_idx,
441 int state = NUD_NONE;
443 n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev);
445 state = n->nud_state;
450 if (state == NUD_REACHABLE)
452 if ((state & NUD_VALID) && order != dflt)
454 if ((state & NUD_VALID) ||
455 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
462 #ifdef CONFIG_IP_ROUTE_MULTIPATH
464 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
465 struct netlink_ext_ack *extack)
469 while (rtnh_ok(rtnh, remaining)) {
471 rtnh = rtnh_next(rtnh, &remaining);
474 /* leftover implies invalid nexthop configuration, discard it */
476 NL_SET_ERR_MSG(extack,
477 "Invalid nexthop configuration - extra data after nexthops");
484 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
485 int remaining, struct fib_config *cfg,
486 struct netlink_ext_ack *extack)
490 change_nexthops(fi) {
493 if (!rtnh_ok(rtnh, remaining)) {
494 NL_SET_ERR_MSG(extack,
495 "Invalid nexthop configuration - extra data after nexthop");
499 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
500 NL_SET_ERR_MSG(extack,
501 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
505 nexthop_nh->nh_flags =
506 (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
507 nexthop_nh->nh_oif = rtnh->rtnh_ifindex;
508 nexthop_nh->nh_weight = rtnh->rtnh_hops + 1;
510 attrlen = rtnh_attrlen(rtnh);
512 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
514 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
515 nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0;
516 #ifdef CONFIG_IP_ROUTE_CLASSID
517 nla = nla_find(attrs, attrlen, RTA_FLOW);
518 nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0;
519 if (nexthop_nh->nh_tclassid)
520 fi->fib_net->ipv4.fib_num_tclassid_users++;
522 nla = nla_find(attrs, attrlen, RTA_ENCAP);
524 struct lwtunnel_state *lwtstate;
525 struct nlattr *nla_entype;
527 nla_entype = nla_find(attrs, attrlen,
530 NL_SET_BAD_ATTR(extack, nla);
531 NL_SET_ERR_MSG(extack,
532 "Encap type is missing");
536 ret = lwtunnel_build_state(nla_get_u16(
542 nexthop_nh->nh_lwtstate =
543 lwtstate_get(lwtstate);
547 rtnh = rtnh_next(rtnh, &remaining);
548 } endfor_nexthops(fi);
559 static void fib_rebalance(struct fib_info *fi)
563 struct in_device *in_dev;
570 if (nh->nh_flags & RTNH_F_DEAD)
573 in_dev = __in_dev_get_rtnl(nh->nh_dev);
576 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
577 nh->nh_flags & RTNH_F_LINKDOWN)
580 total += nh->nh_weight;
581 } endfor_nexthops(fi);
584 change_nexthops(fi) {
587 in_dev = __in_dev_get_rtnl(nexthop_nh->nh_dev);
589 if (nexthop_nh->nh_flags & RTNH_F_DEAD) {
592 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
593 nexthop_nh->nh_flags & RTNH_F_LINKDOWN) {
596 w += nexthop_nh->nh_weight;
597 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
601 atomic_set(&nexthop_nh->nh_upper_bound, upper_bound);
602 } endfor_nexthops(fi);
605 static inline void fib_add_weight(struct fib_info *fi,
606 const struct fib_nh *nh)
608 fi->fib_weight += nh->nh_weight;
611 #else /* CONFIG_IP_ROUTE_MULTIPATH */
613 #define fib_rebalance(fi) do { } while (0)
614 #define fib_add_weight(fi, nh) do { } while (0)
616 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
618 static int fib_encap_match(u16 encap_type,
619 struct nlattr *encap,
620 const struct fib_nh *nh,
621 const struct fib_config *cfg,
622 struct netlink_ext_ack *extack)
624 struct lwtunnel_state *lwtstate;
627 if (encap_type == LWTUNNEL_ENCAP_NONE)
630 ret = lwtunnel_build_state(encap_type, encap, AF_INET,
631 cfg, &lwtstate, extack);
633 result = lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate);
634 lwtstate_free(lwtstate);
640 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
641 struct netlink_ext_ack *extack)
643 #ifdef CONFIG_IP_ROUTE_MULTIPATH
644 struct rtnexthop *rtnh;
648 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
651 if (cfg->fc_oif || cfg->fc_gw) {
653 if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
654 fi->fib_nh, cfg, extack))
657 #ifdef CONFIG_IP_ROUTE_CLASSID
659 cfg->fc_flow != fi->fib_nh->nh_tclassid)
662 if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) &&
663 (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw))
668 #ifdef CONFIG_IP_ROUTE_MULTIPATH
673 remaining = cfg->fc_mp_len;
678 if (!rtnh_ok(rtnh, remaining))
681 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif)
684 attrlen = rtnh_attrlen(rtnh);
686 struct nlattr *nla, *attrs = rtnh_attrs(rtnh);
688 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
689 if (nla && nla_get_in_addr(nla) != nh->nh_gw)
691 #ifdef CONFIG_IP_ROUTE_CLASSID
692 nla = nla_find(attrs, attrlen, RTA_FLOW);
693 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
698 rtnh = rtnh_next(rtnh, &remaining);
699 } endfor_nexthops(fi);
704 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
712 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
713 int type = nla_type(nla);
721 if (type == RTAX_CC_ALGO) {
722 char tmp[TCP_CA_NAME_MAX];
725 nla_strlcpy(tmp, nla, sizeof(tmp));
726 val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
728 if (nla_len(nla) != sizeof(u32))
730 val = nla_get_u32(nla);
733 fi_val = fi->fib_metrics->metrics[type - 1];
734 if (type == RTAX_FEATURES)
735 fi_val &= ~DST_FEATURE_ECN_CA;
749 * Semantics of nexthop is very messy by historical reasons.
750 * We have to take into account, that:
751 * a) gateway can be actually local interface address,
752 * so that gatewayed route is direct.
753 * b) gateway must be on-link address, possibly
754 * described not by an ifaddr, but also by a direct route.
755 * c) If both gateway and interface are specified, they should not
757 * d) If we use tunnel routes, gateway could be not on-link.
759 * Attempt to reconcile all of these (alas, self-contradictory) conditions
760 * results in pretty ugly and hairy code with obscure logic.
762 * I chose to generalized it instead, so that the size
763 * of code does not increase practically, but it becomes
765 * Every prefix is assigned a "scope" value: "host" is local address,
766 * "link" is direct route,
767 * [ ... "site" ... "interior" ... ]
768 * and "universe" is true gateway route with global meaning.
770 * Every prefix refers to a set of "nexthop"s (gw, oif),
771 * where gw must have narrower scope. This recursion stops
772 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
773 * which means that gw is forced to be on link.
775 * Code is still hairy, but now it is apparently logically
776 * consistent and very flexible. F.e. as by-product it allows
777 * to co-exists in peace independent exterior and interior
780 * Normally it looks as following.
782 * {universe prefix} -> (gw, oif) [scope link]
784 * |-> {link prefix} -> (gw, oif) [scope local]
786 * |-> {local prefix} (terminal node)
788 static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
789 struct fib_nh *nh, struct netlink_ext_ack *extack)
793 struct net_device *dev;
795 net = cfg->fc_nlinfo.nl_net;
797 struct fib_result res;
799 if (nh->nh_flags & RTNH_F_ONLINK) {
800 unsigned int addr_type;
802 if (cfg->fc_scope >= RT_SCOPE_LINK) {
803 NL_SET_ERR_MSG(extack,
804 "Nexthop has invalid scope");
807 dev = __dev_get_by_index(net, nh->nh_oif);
810 if (!(dev->flags & IFF_UP)) {
811 NL_SET_ERR_MSG(extack,
812 "Nexthop device is not up");
815 addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw);
816 if (addr_type != RTN_UNICAST) {
817 NL_SET_ERR_MSG(extack,
818 "Nexthop has invalid gateway");
821 if (!netif_carrier_ok(dev))
822 nh->nh_flags |= RTNH_F_LINKDOWN;
825 nh->nh_scope = RT_SCOPE_LINK;
830 struct fib_table *tbl = NULL;
831 struct flowi4 fl4 = {
833 .flowi4_scope = cfg->fc_scope + 1,
834 .flowi4_oif = nh->nh_oif,
835 .flowi4_iif = LOOPBACK_IFINDEX,
838 /* It is not necessary, but requires a bit of thinking */
839 if (fl4.flowi4_scope < RT_SCOPE_LINK)
840 fl4.flowi4_scope = RT_SCOPE_LINK;
842 if (cfg->fc_table && cfg->fc_table != RT_TABLE_MAIN)
843 tbl = fib_get_table(net, cfg->fc_table);
846 err = fib_table_lookup(tbl, &fl4, &res,
847 FIB_LOOKUP_IGNORE_LINKSTATE |
850 /* on error or if no table given do full lookup. This
851 * is needed for example when nexthops are in the local
852 * table rather than the given table
855 err = fib_lookup(net, &fl4, &res,
856 FIB_LOOKUP_IGNORE_LINKSTATE);
860 NL_SET_ERR_MSG(extack,
861 "Nexthop has invalid gateway");
867 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
868 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
871 nh->nh_scope = res.scope;
872 nh->nh_oif = FIB_RES_OIF(res);
873 nh->nh_dev = dev = FIB_RES_DEV(res);
875 NL_SET_ERR_MSG(extack,
876 "No egress device for nexthop gateway");
880 if (!netif_carrier_ok(dev))
881 nh->nh_flags |= RTNH_F_LINKDOWN;
882 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
884 struct in_device *in_dev;
886 if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
887 NL_SET_ERR_MSG(extack,
888 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
893 in_dev = inetdev_by_index(net, nh->nh_oif);
897 if (!(in_dev->dev->flags & IFF_UP)) {
898 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
901 nh->nh_dev = in_dev->dev;
902 dev_hold(nh->nh_dev);
903 nh->nh_scope = RT_SCOPE_HOST;
904 if (!netif_carrier_ok(nh->nh_dev))
905 nh->nh_flags |= RTNH_F_LINKDOWN;
913 static inline unsigned int fib_laddr_hashfn(__be32 val)
915 unsigned int mask = (fib_info_hash_size - 1);
917 return ((__force u32)val ^
918 ((__force u32)val >> 7) ^
919 ((__force u32)val >> 14)) & mask;
922 static struct hlist_head *fib_info_hash_alloc(int bytes)
924 if (bytes <= PAGE_SIZE)
925 return kzalloc(bytes, GFP_KERNEL);
927 return (struct hlist_head *)
928 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
932 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
937 if (bytes <= PAGE_SIZE)
940 free_pages((unsigned long) hash, get_order(bytes));
943 static void fib_info_hash_move(struct hlist_head *new_info_hash,
944 struct hlist_head *new_laddrhash,
945 unsigned int new_size)
947 struct hlist_head *old_info_hash, *old_laddrhash;
948 unsigned int old_size = fib_info_hash_size;
949 unsigned int i, bytes;
951 spin_lock_bh(&fib_info_lock);
952 old_info_hash = fib_info_hash;
953 old_laddrhash = fib_info_laddrhash;
954 fib_info_hash_size = new_size;
956 for (i = 0; i < old_size; i++) {
957 struct hlist_head *head = &fib_info_hash[i];
958 struct hlist_node *n;
961 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
962 struct hlist_head *dest;
963 unsigned int new_hash;
965 new_hash = fib_info_hashfn(fi);
966 dest = &new_info_hash[new_hash];
967 hlist_add_head(&fi->fib_hash, dest);
970 fib_info_hash = new_info_hash;
972 for (i = 0; i < old_size; i++) {
973 struct hlist_head *lhead = &fib_info_laddrhash[i];
974 struct hlist_node *n;
977 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
978 struct hlist_head *ldest;
979 unsigned int new_hash;
981 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
982 ldest = &new_laddrhash[new_hash];
983 hlist_add_head(&fi->fib_lhash, ldest);
986 fib_info_laddrhash = new_laddrhash;
988 spin_unlock_bh(&fib_info_lock);
990 bytes = old_size * sizeof(struct hlist_head *);
991 fib_info_hash_free(old_info_hash, bytes);
992 fib_info_hash_free(old_laddrhash, bytes);
995 __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh)
997 nh->nh_saddr = inet_select_addr(nh->nh_dev,
999 nh->nh_parent->fib_scope);
1000 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1002 return nh->nh_saddr;
1005 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1007 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1008 fib_prefsrc != cfg->fc_dst) {
1009 u32 tb_id = cfg->fc_table;
1012 if (tb_id == RT_TABLE_MAIN)
1013 tb_id = RT_TABLE_LOCAL;
1015 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1016 fib_prefsrc, tb_id);
1018 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1019 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1020 fib_prefsrc, RT_TABLE_LOCAL);
1023 if (rc != RTN_LOCAL)
1030 fib_convert_metrics(struct fib_info *fi, const struct fib_config *cfg)
1032 bool ecn_ca = false;
1039 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1040 int type = nla_type(nla);
1045 if (type > RTAX_MAX)
1048 if (type == RTAX_CC_ALGO) {
1049 char tmp[TCP_CA_NAME_MAX];
1051 nla_strlcpy(tmp, nla, sizeof(tmp));
1052 val = tcp_ca_get_key_by_name(tmp, &ecn_ca);
1053 if (val == TCP_CA_UNSPEC)
1056 if (nla_len(nla) != sizeof(u32))
1058 val = nla_get_u32(nla);
1060 if (type == RTAX_ADVMSS && val > 65535 - 40)
1062 if (type == RTAX_MTU && val > 65535 - 15)
1064 if (type == RTAX_HOPLIMIT && val > 255)
1066 if (type == RTAX_FEATURES && (val & ~RTAX_FEATURE_MASK))
1068 fi->fib_metrics->metrics[type - 1] = val;
1072 fi->fib_metrics->metrics[RTAX_FEATURES - 1] |= DST_FEATURE_ECN_CA;
1077 struct fib_info *fib_create_info(struct fib_config *cfg,
1078 struct netlink_ext_ack *extack)
1081 struct fib_info *fi = NULL;
1082 struct fib_info *ofi;
1084 struct net *net = cfg->fc_nlinfo.nl_net;
1086 if (cfg->fc_type > RTN_MAX)
1089 /* Fast check to catch the most weird cases */
1090 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1091 NL_SET_ERR_MSG(extack, "Invalid scope");
1095 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1096 NL_SET_ERR_MSG(extack,
1097 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1101 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1103 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1110 if (fib_info_cnt >= fib_info_hash_size) {
1111 unsigned int new_size = fib_info_hash_size << 1;
1112 struct hlist_head *new_info_hash;
1113 struct hlist_head *new_laddrhash;
1118 bytes = new_size * sizeof(struct hlist_head *);
1119 new_info_hash = fib_info_hash_alloc(bytes);
1120 new_laddrhash = fib_info_hash_alloc(bytes);
1121 if (!new_info_hash || !new_laddrhash) {
1122 fib_info_hash_free(new_info_hash, bytes);
1123 fib_info_hash_free(new_laddrhash, bytes);
1125 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1127 if (!fib_info_hash_size)
1131 fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL);
1135 fi->fib_metrics = kzalloc(sizeof(*fi->fib_metrics), GFP_KERNEL);
1136 if (unlikely(!fi->fib_metrics)) {
1138 return ERR_PTR(err);
1140 refcount_set(&fi->fib_metrics->refcnt, 1);
1142 fi->fib_metrics = (struct dst_metrics *)&dst_default_metrics;
1146 fi->fib_protocol = cfg->fc_protocol;
1147 fi->fib_scope = cfg->fc_scope;
1148 fi->fib_flags = cfg->fc_flags;
1149 fi->fib_priority = cfg->fc_priority;
1150 fi->fib_prefsrc = cfg->fc_prefsrc;
1151 fi->fib_type = cfg->fc_type;
1152 fi->fib_tb_id = cfg->fc_table;
1155 change_nexthops(fi) {
1156 nexthop_nh->nh_parent = fi;
1157 nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *);
1158 if (!nexthop_nh->nh_pcpu_rth_output)
1160 } endfor_nexthops(fi)
1162 err = fib_convert_metrics(fi, cfg);
1167 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1168 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg, extack);
1171 if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) {
1172 NL_SET_ERR_MSG(extack,
1173 "Nexthop device index does not match RTA_OIF");
1176 if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) {
1177 NL_SET_ERR_MSG(extack,
1178 "Nexthop gateway does not match RTA_GATEWAY");
1181 #ifdef CONFIG_IP_ROUTE_CLASSID
1182 if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) {
1183 NL_SET_ERR_MSG(extack,
1184 "Nexthop class id does not match RTA_FLOW");
1189 NL_SET_ERR_MSG(extack,
1190 "Multipath support not enabled in kernel");
1194 struct fib_nh *nh = fi->fib_nh;
1196 if (cfg->fc_encap) {
1197 struct lwtunnel_state *lwtstate;
1199 if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) {
1200 NL_SET_ERR_MSG(extack,
1201 "LWT encap type not specified");
1204 err = lwtunnel_build_state(cfg->fc_encap_type,
1205 cfg->fc_encap, AF_INET, cfg,
1210 nh->nh_lwtstate = lwtstate_get(lwtstate);
1212 nh->nh_oif = cfg->fc_oif;
1213 nh->nh_gw = cfg->fc_gw;
1214 nh->nh_flags = cfg->fc_flags;
1215 #ifdef CONFIG_IP_ROUTE_CLASSID
1216 nh->nh_tclassid = cfg->fc_flow;
1217 if (nh->nh_tclassid)
1218 fi->fib_net->ipv4.fib_num_tclassid_users++;
1220 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1225 if (fib_props[cfg->fc_type].error) {
1226 if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) {
1227 NL_SET_ERR_MSG(extack,
1228 "Gateway, device and multipath can not be specified for this route type");
1233 switch (cfg->fc_type) {
1241 NL_SET_ERR_MSG(extack, "Invalid route type");
1246 if (cfg->fc_scope > RT_SCOPE_HOST) {
1247 NL_SET_ERR_MSG(extack, "Invalid scope");
1251 if (cfg->fc_scope == RT_SCOPE_HOST) {
1252 struct fib_nh *nh = fi->fib_nh;
1254 /* Local address is added. */
1256 NL_SET_ERR_MSG(extack,
1257 "Route with host scope can not have multiple nexthops");
1261 NL_SET_ERR_MSG(extack,
1262 "Route with host scope can not have a gateway");
1265 nh->nh_scope = RT_SCOPE_NOWHERE;
1266 nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif);
1273 change_nexthops(fi) {
1274 err = fib_check_nh(cfg, fi, nexthop_nh, extack);
1277 if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN)
1279 } endfor_nexthops(fi)
1280 if (linkdown == fi->fib_nhs)
1281 fi->fib_flags |= RTNH_F_LINKDOWN;
1284 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1285 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1289 change_nexthops(fi) {
1290 fib_info_update_nh_saddr(net, nexthop_nh);
1291 fib_add_weight(fi, nexthop_nh);
1292 } endfor_nexthops(fi)
1297 ofi = fib_find_info(fi);
1306 refcount_set(&fi->fib_clntref, 1);
1307 spin_lock_bh(&fib_info_lock);
1308 hlist_add_head(&fi->fib_hash,
1309 &fib_info_hash[fib_info_hashfn(fi)]);
1310 if (fi->fib_prefsrc) {
1311 struct hlist_head *head;
1313 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1314 hlist_add_head(&fi->fib_lhash, head);
1316 change_nexthops(fi) {
1317 struct hlist_head *head;
1320 if (!nexthop_nh->nh_dev)
1322 hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex);
1323 head = &fib_info_devhash[hash];
1324 hlist_add_head(&nexthop_nh->nh_hash, head);
1325 } endfor_nexthops(fi)
1326 spin_unlock_bh(&fib_info_lock);
1338 return ERR_PTR(err);
1341 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1342 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1343 struct fib_info *fi, unsigned int flags)
1345 struct nlmsghdr *nlh;
1348 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1352 rtm = nlmsg_data(nlh);
1353 rtm->rtm_family = AF_INET;
1354 rtm->rtm_dst_len = dst_len;
1355 rtm->rtm_src_len = 0;
1358 rtm->rtm_table = tb_id;
1360 rtm->rtm_table = RT_TABLE_COMPAT;
1361 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1362 goto nla_put_failure;
1363 rtm->rtm_type = type;
1364 rtm->rtm_flags = fi->fib_flags;
1365 rtm->rtm_scope = fi->fib_scope;
1366 rtm->rtm_protocol = fi->fib_protocol;
1368 if (rtm->rtm_dst_len &&
1369 nla_put_in_addr(skb, RTA_DST, dst))
1370 goto nla_put_failure;
1371 if (fi->fib_priority &&
1372 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1373 goto nla_put_failure;
1374 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1375 goto nla_put_failure;
1377 if (fi->fib_prefsrc &&
1378 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1379 goto nla_put_failure;
1380 if (fi->fib_nhs == 1) {
1381 if (fi->fib_nh->nh_gw &&
1382 nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw))
1383 goto nla_put_failure;
1384 if (fi->fib_nh->nh_oif &&
1385 nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif))
1386 goto nla_put_failure;
1387 if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) {
1388 struct in_device *in_dev;
1391 in_dev = __in_dev_get_rcu(fi->fib_nh->nh_dev);
1393 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1394 rtm->rtm_flags |= RTNH_F_DEAD;
1397 if (fi->fib_nh->nh_flags & RTNH_F_OFFLOAD)
1398 rtm->rtm_flags |= RTNH_F_OFFLOAD;
1399 #ifdef CONFIG_IP_ROUTE_CLASSID
1400 if (fi->fib_nh[0].nh_tclassid &&
1401 nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid))
1402 goto nla_put_failure;
1404 if (fi->fib_nh->nh_lwtstate &&
1405 lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate) < 0)
1406 goto nla_put_failure;
1408 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1409 if (fi->fib_nhs > 1) {
1410 struct rtnexthop *rtnh;
1413 mp = nla_nest_start(skb, RTA_MULTIPATH);
1415 goto nla_put_failure;
1418 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1420 goto nla_put_failure;
1422 rtnh->rtnh_flags = nh->nh_flags & 0xFF;
1423 if (nh->nh_flags & RTNH_F_LINKDOWN) {
1424 struct in_device *in_dev;
1427 in_dev = __in_dev_get_rcu(nh->nh_dev);
1429 IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev))
1430 rtnh->rtnh_flags |= RTNH_F_DEAD;
1433 rtnh->rtnh_hops = nh->nh_weight - 1;
1434 rtnh->rtnh_ifindex = nh->nh_oif;
1437 nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw))
1438 goto nla_put_failure;
1439 #ifdef CONFIG_IP_ROUTE_CLASSID
1440 if (nh->nh_tclassid &&
1441 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1442 goto nla_put_failure;
1444 if (nh->nh_lwtstate &&
1445 lwtunnel_fill_encap(skb, nh->nh_lwtstate) < 0)
1446 goto nla_put_failure;
1448 /* length of rtnetlink header + attributes */
1449 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh;
1450 } endfor_nexthops(fi);
1452 nla_nest_end(skb, mp);
1455 nlmsg_end(skb, nlh);
1459 nlmsg_cancel(skb, nlh);
1465 * - local address disappeared -> we must delete all the entries
1467 * - device went down -> we must shutdown all nexthops going via it.
1469 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1472 unsigned int hash = fib_laddr_hashfn(local);
1473 struct hlist_head *head = &fib_info_laddrhash[hash];
1474 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1475 struct net *net = dev_net(dev);
1476 struct fib_info *fi;
1478 if (!fib_info_laddrhash || local == 0)
1481 hlist_for_each_entry(fi, head, fib_lhash) {
1482 if (!net_eq(fi->fib_net, net) ||
1483 fi->fib_tb_id != tb_id)
1485 if (fi->fib_prefsrc == local) {
1486 fi->fib_flags |= RTNH_F_DEAD;
1493 static int call_fib_nh_notifiers(struct fib_nh *fib_nh,
1494 enum fib_event_type event_type)
1496 struct in_device *in_dev = __in_dev_get_rtnl(fib_nh->nh_dev);
1497 struct fib_nh_notifier_info info = {
1501 switch (event_type) {
1502 case FIB_EVENT_NH_ADD:
1503 if (fib_nh->nh_flags & RTNH_F_DEAD)
1505 if (IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1506 fib_nh->nh_flags & RTNH_F_LINKDOWN)
1508 return call_fib4_notifiers(dev_net(fib_nh->nh_dev), event_type,
1510 case FIB_EVENT_NH_DEL:
1511 if ((in_dev && IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) &&
1512 fib_nh->nh_flags & RTNH_F_LINKDOWN) ||
1513 (fib_nh->nh_flags & RTNH_F_DEAD))
1514 return call_fib4_notifiers(dev_net(fib_nh->nh_dev),
1515 event_type, &info.info);
1523 /* Update the PMTU of exceptions when:
1524 * - the new MTU of the first hop becomes smaller than the PMTU
1525 * - the old MTU was the same as the PMTU, and it limited discovery of
1526 * larger MTUs on the path. With that limit raised, we can now
1527 * discover larger MTUs
1528 * A special case is locked exceptions, for which the PMTU is smaller
1529 * than the minimal accepted PMTU:
1530 * - if the new MTU is greater than the PMTU, don't make any change
1531 * - otherwise, unlock and set PMTU
1533 static void nh_update_mtu(struct fib_nh *nh, u32 new, u32 orig)
1535 struct fnhe_hash_bucket *bucket;
1538 bucket = rcu_dereference_protected(nh->nh_exceptions, 1);
1542 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1543 struct fib_nh_exception *fnhe;
1545 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1547 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1548 if (fnhe->fnhe_mtu_locked) {
1549 if (new <= fnhe->fnhe_pmtu) {
1550 fnhe->fnhe_pmtu = new;
1551 fnhe->fnhe_mtu_locked = false;
1553 } else if (new < fnhe->fnhe_pmtu ||
1554 orig == fnhe->fnhe_pmtu) {
1555 fnhe->fnhe_pmtu = new;
1561 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1563 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1564 struct hlist_head *head = &fib_info_devhash[hash];
1567 hlist_for_each_entry(nh, head, nh_hash) {
1568 if (nh->nh_dev == dev)
1569 nh_update_mtu(nh, dev->mtu, orig_mtu);
1573 /* Event force Flags Description
1574 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1575 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1576 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1577 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1579 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1582 int scope = RT_SCOPE_NOWHERE;
1583 struct fib_info *prev_fi = NULL;
1584 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1585 struct hlist_head *head = &fib_info_devhash[hash];
1591 hlist_for_each_entry(nh, head, nh_hash) {
1592 struct fib_info *fi = nh->nh_parent;
1595 BUG_ON(!fi->fib_nhs);
1596 if (nh->nh_dev != dev || fi == prev_fi)
1600 change_nexthops(fi) {
1601 if (nexthop_nh->nh_flags & RTNH_F_DEAD)
1603 else if (nexthop_nh->nh_dev == dev &&
1604 nexthop_nh->nh_scope != scope) {
1607 case NETDEV_UNREGISTER:
1608 nexthop_nh->nh_flags |= RTNH_F_DEAD;
1611 nexthop_nh->nh_flags |= RTNH_F_LINKDOWN;
1614 call_fib_nh_notifiers(nexthop_nh,
1618 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1619 if (event == NETDEV_UNREGISTER &&
1620 nexthop_nh->nh_dev == dev) {
1625 } endfor_nexthops(fi)
1626 if (dead == fi->fib_nhs) {
1629 case NETDEV_UNREGISTER:
1630 fi->fib_flags |= RTNH_F_DEAD;
1633 fi->fib_flags |= RTNH_F_LINKDOWN;
1645 /* Must be invoked inside of an RCU protected region. */
1646 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1648 struct fib_info *fi = NULL, *last_resort = NULL;
1649 struct hlist_head *fa_head = res->fa_head;
1650 struct fib_table *tb = res->table;
1651 u8 slen = 32 - res->prefixlen;
1652 int order = -1, last_idx = -1;
1653 struct fib_alias *fa, *fa1 = NULL;
1654 u32 last_prio = res->fi->fib_priority;
1657 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
1658 struct fib_info *next_fi = fa->fa_info;
1660 if (fa->fa_slen != slen)
1662 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
1664 if (fa->tb_id != tb->tb_id)
1666 if (next_fi->fib_priority > last_prio &&
1667 fa->fa_tos == last_tos) {
1672 if (next_fi->fib_flags & RTNH_F_DEAD)
1674 last_tos = fa->fa_tos;
1675 last_prio = next_fi->fib_priority;
1677 if (next_fi->fib_scope != res->scope ||
1678 fa->fa_type != RTN_UNICAST)
1680 if (!next_fi->fib_nh[0].nh_gw ||
1681 next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK)
1684 fib_alias_accessed(fa);
1687 if (next_fi != res->fi)
1690 } else if (!fib_detect_death(fi, order, &last_resort,
1691 &last_idx, fa1->fa_default)) {
1692 fib_result_assign(res, fi);
1693 fa1->fa_default = order;
1700 if (order <= 0 || !fi) {
1702 fa1->fa_default = -1;
1706 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
1708 fib_result_assign(res, fi);
1709 fa1->fa_default = order;
1714 fib_result_assign(res, last_resort);
1715 fa1->fa_default = last_idx;
1721 * Dead device goes up. We wake up dead nexthops.
1722 * It takes sense only on multipath routes.
1724 int fib_sync_up(struct net_device *dev, unsigned int nh_flags)
1726 struct fib_info *prev_fi;
1728 struct hlist_head *head;
1732 if (!(dev->flags & IFF_UP))
1735 if (nh_flags & RTNH_F_DEAD) {
1736 unsigned int flags = dev_get_flags(dev);
1738 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1739 nh_flags |= RTNH_F_LINKDOWN;
1743 hash = fib_devindex_hashfn(dev->ifindex);
1744 head = &fib_info_devhash[hash];
1747 hlist_for_each_entry(nh, head, nh_hash) {
1748 struct fib_info *fi = nh->nh_parent;
1751 BUG_ON(!fi->fib_nhs);
1752 if (nh->nh_dev != dev || fi == prev_fi)
1757 change_nexthops(fi) {
1758 if (!(nexthop_nh->nh_flags & nh_flags)) {
1762 if (!nexthop_nh->nh_dev ||
1763 !(nexthop_nh->nh_dev->flags & IFF_UP))
1765 if (nexthop_nh->nh_dev != dev ||
1766 !__in_dev_get_rtnl(dev))
1769 nexthop_nh->nh_flags &= ~nh_flags;
1770 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
1771 } endfor_nexthops(fi)
1774 fi->fib_flags &= ~nh_flags;
1784 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1785 static bool fib_good_nh(const struct fib_nh *nh)
1787 int state = NUD_REACHABLE;
1789 if (nh->nh_scope == RT_SCOPE_LINK) {
1790 struct neighbour *n;
1794 n = __ipv4_neigh_lookup_noref(nh->nh_dev,
1795 (__force u32)nh->nh_gw);
1797 state = n->nud_state;
1799 rcu_read_unlock_bh();
1802 return !!(state & NUD_VALID);
1805 void fib_select_multipath(struct fib_result *res, int hash)
1807 struct fib_info *fi = res->fi;
1808 struct net *net = fi->fib_net;
1812 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
1813 if (!fib_good_nh(nh))
1816 res->nh_sel = nhsel;
1821 if (hash > atomic_read(&nh->nh_upper_bound))
1824 res->nh_sel = nhsel;
1826 } endfor_nexthops(fi);
1830 void fib_select_path(struct net *net, struct fib_result *res,
1831 struct flowi4 *fl4, const struct sk_buff *skb)
1835 oif_check = (fl4->flowi4_oif == 0 ||
1836 fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF);
1838 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1839 if (res->fi->fib_nhs > 1 && oif_check) {
1840 int h = fib_multipath_hash(res->fi, fl4, skb);
1842 fib_select_multipath(res, h);
1846 if (!res->prefixlen &&
1847 res->table->tb_num_default > 1 &&
1848 res->type == RTN_UNICAST && oif_check)
1849 fib_select_default(fl4, res);
1852 fl4->saddr = FIB_RES_PREFSRC(net, *res);
1854 EXPORT_SYMBOL_GPL(fib_select_path);