1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * IPv4 Forwarding Information Base: semantics.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
12 #include <linux/uaccess.h>
13 #include <linux/bitops.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/jiffies.h>
18 #include <linux/string.h>
19 #include <linux/socket.h>
20 #include <linux/sockios.h>
21 #include <linux/errno.h>
23 #include <linux/inet.h>
24 #include <linux/inetdevice.h>
25 #include <linux/netdevice.h>
26 #include <linux/if_arp.h>
27 #include <linux/proc_fs.h>
28 #include <linux/skbuff.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/netlink.h>
35 #include <net/protocol.h>
36 #include <net/route.h>
39 #include <net/ip_fib.h>
40 #include <net/ip6_fib.h>
41 #include <net/nexthop.h>
42 #include <net/netlink.h>
44 #include <net/lwtunnel.h>
45 #include <net/fib_notifier.h>
46 #include <net/addrconf.h>
48 #include "fib_lookup.h"
50 static DEFINE_SPINLOCK(fib_info_lock);
51 static struct hlist_head *fib_info_hash;
52 static struct hlist_head *fib_info_laddrhash;
53 static unsigned int fib_info_hash_size;
54 static unsigned int fib_info_cnt;
56 #define DEVINDEX_HASHBITS 8
57 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
58 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
60 /* for_nexthops and change_nexthops only used when nexthop object
61 * is not set in a fib_info. The logic within can reference fib_nh.
63 #ifdef CONFIG_IP_ROUTE_MULTIPATH
65 #define for_nexthops(fi) { \
66 int nhsel; const struct fib_nh *nh; \
67 for (nhsel = 0, nh = (fi)->fib_nh; \
68 nhsel < fib_info_num_path((fi)); \
71 #define change_nexthops(fi) { \
72 int nhsel; struct fib_nh *nexthop_nh; \
73 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
74 nhsel < fib_info_num_path((fi)); \
75 nexthop_nh++, nhsel++)
77 #else /* CONFIG_IP_ROUTE_MULTIPATH */
79 /* Hope, that gcc will optimize it to get rid of dummy loop */
81 #define for_nexthops(fi) { \
82 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
83 for (nhsel = 0; nhsel < 1; nhsel++)
85 #define change_nexthops(fi) { \
87 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
88 for (nhsel = 0; nhsel < 1; nhsel++)
90 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
92 #define endfor_nexthops(fi) }
95 const struct fib_prop fib_props[RTN_MAX + 1] = {
98 .scope = RT_SCOPE_NOWHERE,
102 .scope = RT_SCOPE_UNIVERSE,
106 .scope = RT_SCOPE_HOST,
110 .scope = RT_SCOPE_LINK,
114 .scope = RT_SCOPE_LINK,
118 .scope = RT_SCOPE_UNIVERSE,
122 .scope = RT_SCOPE_UNIVERSE,
124 [RTN_UNREACHABLE] = {
125 .error = -EHOSTUNREACH,
126 .scope = RT_SCOPE_UNIVERSE,
130 .scope = RT_SCOPE_UNIVERSE,
134 .scope = RT_SCOPE_UNIVERSE,
138 .scope = RT_SCOPE_NOWHERE,
142 .scope = RT_SCOPE_NOWHERE,
146 static void rt_fibinfo_free(struct rtable __rcu **rtp)
148 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
153 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
154 * because we waited an RCU grace period before calling
155 * free_fib_info_rcu()
158 dst_dev_put(&rt->dst);
159 dst_release_immediate(&rt->dst);
162 static void free_nh_exceptions(struct fib_nh_common *nhc)
164 struct fnhe_hash_bucket *hash;
167 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
170 for (i = 0; i < FNHE_HASH_SIZE; i++) {
171 struct fib_nh_exception *fnhe;
173 fnhe = rcu_dereference_protected(hash[i].chain, 1);
175 struct fib_nh_exception *next;
177 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
179 rt_fibinfo_free(&fnhe->fnhe_rth_input);
180 rt_fibinfo_free(&fnhe->fnhe_rth_output);
190 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
197 for_each_possible_cpu(cpu) {
200 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
202 dst_dev_put(&rt->dst);
203 dst_release_immediate(&rt->dst);
209 void fib_nh_common_release(struct fib_nh_common *nhc)
212 dev_put(nhc->nhc_dev);
214 lwtstate_put(nhc->nhc_lwtstate);
215 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
216 rt_fibinfo_free(&nhc->nhc_rth_input);
217 free_nh_exceptions(nhc);
219 EXPORT_SYMBOL_GPL(fib_nh_common_release);
221 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
223 #ifdef CONFIG_IP_ROUTE_CLASSID
224 if (fib_nh->nh_tclassid)
225 net->ipv4.fib_num_tclassid_users--;
227 fib_nh_common_release(&fib_nh->nh_common);
230 /* Release a nexthop info record */
231 static void free_fib_info_rcu(struct rcu_head *head)
233 struct fib_info *fi = container_of(head, struct fib_info, rcu);
238 change_nexthops(fi) {
239 fib_nh_release(fi->fib_net, nexthop_nh);
240 } endfor_nexthops(fi);
243 ip_fib_metrics_put(fi->fib_metrics);
248 void free_fib_info(struct fib_info *fi)
250 if (fi->fib_dead == 0) {
251 pr_warn("Freeing alive fib_info %p\n", fi);
256 call_rcu(&fi->rcu, free_fib_info_rcu);
258 EXPORT_SYMBOL_GPL(free_fib_info);
260 void fib_release_info(struct fib_info *fi)
262 spin_lock_bh(&fib_info_lock);
263 if (fi && --fi->fib_treeref == 0) {
264 hlist_del(&fi->fib_hash);
266 hlist_del(&fi->fib_lhash);
268 list_del(&fi->nh_list);
270 change_nexthops(fi) {
271 if (!nexthop_nh->fib_nh_dev)
273 hlist_del(&nexthop_nh->nh_hash);
274 } endfor_nexthops(fi)
279 spin_unlock_bh(&fib_info_lock);
282 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
284 const struct fib_nh *onh;
286 if (fi->nh || ofi->nh)
287 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
289 if (ofi->fib_nhs == 0)
293 onh = fib_info_nh(ofi, nhsel);
295 if (nh->fib_nh_oif != onh->fib_nh_oif ||
296 nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
297 nh->fib_nh_scope != onh->fib_nh_scope ||
298 #ifdef CONFIG_IP_ROUTE_MULTIPATH
299 nh->fib_nh_weight != onh->fib_nh_weight ||
301 #ifdef CONFIG_IP_ROUTE_CLASSID
302 nh->nh_tclassid != onh->nh_tclassid ||
304 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
305 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
308 if (nh->fib_nh_gw_family == AF_INET &&
309 nh->fib_nh_gw4 != onh->fib_nh_gw4)
312 if (nh->fib_nh_gw_family == AF_INET6 &&
313 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
315 } endfor_nexthops(fi);
319 static inline unsigned int fib_devindex_hashfn(unsigned int val)
321 unsigned int mask = DEVINDEX_HASHSIZE - 1;
324 (val >> DEVINDEX_HASHBITS) ^
325 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
328 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
329 u32 prefsrc, u32 priority)
331 unsigned int val = init_val;
333 val ^= (protocol << 8) | scope;
340 static unsigned int fib_info_hashfn_result(unsigned int val)
342 unsigned int mask = (fib_info_hash_size - 1);
344 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
347 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
351 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
352 fi->fib_scope, (__force u32)fi->fib_prefsrc,
356 val ^= fib_devindex_hashfn(fi->nh->id);
359 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
360 } endfor_nexthops(fi)
363 return fib_info_hashfn_result(val);
366 /* no metrics, only nexthop id */
367 static struct fib_info *fib_find_info_nh(struct net *net,
368 const struct fib_config *cfg)
370 struct hlist_head *head;
374 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
375 cfg->fc_protocol, cfg->fc_scope,
376 (__force u32)cfg->fc_prefsrc,
378 hash = fib_info_hashfn_result(hash);
379 head = &fib_info_hash[hash];
381 hlist_for_each_entry(fi, head, fib_hash) {
382 if (!net_eq(fi->fib_net, net))
384 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
386 if (cfg->fc_protocol == fi->fib_protocol &&
387 cfg->fc_scope == fi->fib_scope &&
388 cfg->fc_prefsrc == fi->fib_prefsrc &&
389 cfg->fc_priority == fi->fib_priority &&
390 cfg->fc_type == fi->fib_type &&
391 cfg->fc_table == fi->fib_tb_id &&
392 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
399 static struct fib_info *fib_find_info(struct fib_info *nfi)
401 struct hlist_head *head;
405 hash = fib_info_hashfn(nfi);
406 head = &fib_info_hash[hash];
408 hlist_for_each_entry(fi, head, fib_hash) {
409 if (!net_eq(fi->fib_net, nfi->fib_net))
411 if (fi->fib_nhs != nfi->fib_nhs)
413 if (nfi->fib_protocol == fi->fib_protocol &&
414 nfi->fib_scope == fi->fib_scope &&
415 nfi->fib_prefsrc == fi->fib_prefsrc &&
416 nfi->fib_priority == fi->fib_priority &&
417 nfi->fib_type == fi->fib_type &&
418 memcmp(nfi->fib_metrics, fi->fib_metrics,
419 sizeof(u32) * RTAX_MAX) == 0 &&
420 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
421 nh_comp(fi, nfi) == 0)
428 /* Check, that the gateway is already configured.
429 * Used only by redirect accept routine.
431 int ip_fib_check_default(__be32 gw, struct net_device *dev)
433 struct hlist_head *head;
437 spin_lock(&fib_info_lock);
439 hash = fib_devindex_hashfn(dev->ifindex);
440 head = &fib_info_devhash[hash];
441 hlist_for_each_entry(nh, head, nh_hash) {
442 if (nh->fib_nh_dev == dev &&
443 nh->fib_nh_gw4 == gw &&
444 !(nh->fib_nh_flags & RTNH_F_DEAD)) {
445 spin_unlock(&fib_info_lock);
450 spin_unlock(&fib_info_lock);
455 static inline size_t fib_nlmsg_size(struct fib_info *fi)
457 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
458 + nla_total_size(4) /* RTA_TABLE */
459 + nla_total_size(4) /* RTA_DST */
460 + nla_total_size(4) /* RTA_PRIORITY */
461 + nla_total_size(4) /* RTA_PREFSRC */
462 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
463 unsigned int nhs = fib_info_num_path(fi);
465 /* space for nested metrics */
466 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
469 payload += nla_total_size(4); /* RTA_NH_ID */
472 size_t nh_encapsize = 0;
473 /* Also handles the special case nhs == 1 */
475 /* each nexthop is packed in an attribute */
476 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
479 /* may contain flow and gateway attribute */
480 nhsize += 2 * nla_total_size(4);
482 /* grab encap info */
483 for (i = 0; i < fib_info_num_path(fi); i++) {
484 struct fib_nh_common *nhc = fib_info_nhc(fi, i);
486 if (nhc->nhc_lwtstate) {
488 nh_encapsize += lwtunnel_get_encap_size(
491 nh_encapsize += nla_total_size(2);
495 /* all nexthops are packed in a nested attribute */
496 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
503 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
504 int dst_len, u32 tb_id, const struct nl_info *info,
505 unsigned int nlm_flags)
508 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
511 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
515 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
516 fa->fa_type, key, dst_len,
517 fa->fa_tos, fa->fa_info, nlm_flags);
519 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
520 WARN_ON(err == -EMSGSIZE);
524 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
525 info->nlh, GFP_KERNEL);
529 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
532 static int fib_detect_death(struct fib_info *fi, int order,
533 struct fib_info **last_resort, int *last_idx,
536 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
538 int state = NUD_NONE;
540 if (likely(nhc->nhc_gw_family == AF_INET))
541 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
542 else if (nhc->nhc_gw_family == AF_INET6)
543 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
549 state = n->nud_state;
554 if (state == NUD_REACHABLE)
556 if ((state & NUD_VALID) && order != dflt)
558 if ((state & NUD_VALID) ||
559 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
566 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap,
567 u16 encap_type, void *cfg, gfp_t gfp_flags,
568 struct netlink_ext_ack *extack)
572 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
574 if (!nhc->nhc_pcpu_rth_output)
578 struct lwtunnel_state *lwtstate;
580 if (encap_type == LWTUNNEL_ENCAP_NONE) {
581 NL_SET_ERR_MSG(extack, "LWT encap type not specified");
585 err = lwtunnel_build_state(encap_type, encap, nhc->nhc_family,
586 cfg, &lwtstate, extack);
590 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
596 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
597 nhc->nhc_pcpu_rth_output = NULL;
600 EXPORT_SYMBOL_GPL(fib_nh_common_init);
602 int fib_nh_init(struct net *net, struct fib_nh *nh,
603 struct fib_config *cfg, int nh_weight,
604 struct netlink_ext_ack *extack)
608 nh->fib_nh_family = AF_INET;
610 err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
611 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
615 nh->fib_nh_oif = cfg->fc_oif;
616 nh->fib_nh_gw_family = cfg->fc_gw_family;
617 if (cfg->fc_gw_family == AF_INET)
618 nh->fib_nh_gw4 = cfg->fc_gw4;
619 else if (cfg->fc_gw_family == AF_INET6)
620 nh->fib_nh_gw6 = cfg->fc_gw6;
622 nh->fib_nh_flags = cfg->fc_flags;
624 #ifdef CONFIG_IP_ROUTE_CLASSID
625 nh->nh_tclassid = cfg->fc_flow;
627 net->ipv4.fib_num_tclassid_users++;
629 #ifdef CONFIG_IP_ROUTE_MULTIPATH
630 nh->fib_nh_weight = nh_weight;
635 #ifdef CONFIG_IP_ROUTE_MULTIPATH
637 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
638 struct netlink_ext_ack *extack)
642 while (rtnh_ok(rtnh, remaining)) {
644 rtnh = rtnh_next(rtnh, &remaining);
647 /* leftover implies invalid nexthop configuration, discard it */
649 NL_SET_ERR_MSG(extack,
650 "Invalid nexthop configuration - extra data after nexthops");
657 /* only called when fib_nh is integrated into fib_info */
658 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
659 int remaining, struct fib_config *cfg,
660 struct netlink_ext_ack *extack)
662 struct net *net = fi->fib_net;
663 struct fib_config fib_cfg;
667 change_nexthops(fi) {
670 memset(&fib_cfg, 0, sizeof(fib_cfg));
672 if (!rtnh_ok(rtnh, remaining)) {
673 NL_SET_ERR_MSG(extack,
674 "Invalid nexthop configuration - extra data after nexthop");
678 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
679 NL_SET_ERR_MSG(extack,
680 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
684 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
685 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
687 attrlen = rtnh_attrlen(rtnh);
689 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
691 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
692 nlav = nla_find(attrs, attrlen, RTA_VIA);
694 NL_SET_ERR_MSG(extack,
695 "Nexthop configuration can not contain both GATEWAY and VIA");
699 fib_cfg.fc_gw4 = nla_get_in_addr(nla);
701 fib_cfg.fc_gw_family = AF_INET;
703 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
708 nla = nla_find(attrs, attrlen, RTA_FLOW);
710 fib_cfg.fc_flow = nla_get_u32(nla);
712 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
713 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
715 fib_cfg.fc_encap_type = nla_get_u16(nla);
718 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
719 rtnh->rtnh_hops + 1, extack);
723 rtnh = rtnh_next(rtnh, &remaining);
724 } endfor_nexthops(fi);
727 nh = fib_info_nh(fi, 0);
728 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
729 NL_SET_ERR_MSG(extack,
730 "Nexthop device index does not match RTA_OIF");
733 if (cfg->fc_gw_family) {
734 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
735 (cfg->fc_gw_family == AF_INET &&
736 nh->fib_nh_gw4 != cfg->fc_gw4) ||
737 (cfg->fc_gw_family == AF_INET6 &&
738 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
739 NL_SET_ERR_MSG(extack,
740 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
744 #ifdef CONFIG_IP_ROUTE_CLASSID
745 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
746 NL_SET_ERR_MSG(extack,
747 "Nexthop class id does not match RTA_FLOW");
756 /* only called when fib_nh is integrated into fib_info */
757 static void fib_rebalance(struct fib_info *fi)
762 if (fib_info_num_path(fi) < 2)
767 if (nh->fib_nh_flags & RTNH_F_DEAD)
770 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
771 nh->fib_nh_flags & RTNH_F_LINKDOWN)
774 total += nh->fib_nh_weight;
775 } endfor_nexthops(fi);
778 change_nexthops(fi) {
781 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
783 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
784 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
787 w += nexthop_nh->fib_nh_weight;
788 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
792 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
793 } endfor_nexthops(fi);
795 #else /* CONFIG_IP_ROUTE_MULTIPATH */
797 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
798 int remaining, struct fib_config *cfg,
799 struct netlink_ext_ack *extack)
801 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
806 #define fib_rebalance(fi) do { } while (0)
808 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
810 static int fib_encap_match(u16 encap_type,
811 struct nlattr *encap,
812 const struct fib_nh *nh,
813 const struct fib_config *cfg,
814 struct netlink_ext_ack *extack)
816 struct lwtunnel_state *lwtstate;
819 if (encap_type == LWTUNNEL_ENCAP_NONE)
822 ret = lwtunnel_build_state(encap_type, encap, AF_INET,
823 cfg, &lwtstate, extack);
825 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
826 lwtstate_free(lwtstate);
832 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
833 struct netlink_ext_ack *extack)
835 #ifdef CONFIG_IP_ROUTE_MULTIPATH
836 struct rtnexthop *rtnh;
840 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
844 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
849 if (cfg->fc_oif || cfg->fc_gw_family) {
850 struct fib_nh *nh = fib_info_nh(fi, 0);
853 if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
857 #ifdef CONFIG_IP_ROUTE_CLASSID
859 cfg->fc_flow != nh->nh_tclassid)
862 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
863 (cfg->fc_gw_family &&
864 cfg->fc_gw_family != nh->fib_nh_gw_family))
867 if (cfg->fc_gw_family == AF_INET &&
868 cfg->fc_gw4 != nh->fib_nh_gw4)
871 if (cfg->fc_gw_family == AF_INET6 &&
872 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
878 #ifdef CONFIG_IP_ROUTE_MULTIPATH
883 remaining = cfg->fc_mp_len;
888 if (!rtnh_ok(rtnh, remaining))
891 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
894 attrlen = rtnh_attrlen(rtnh);
896 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
898 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
899 nlav = nla_find(attrs, attrlen, RTA_VIA);
901 NL_SET_ERR_MSG(extack,
902 "Nexthop configuration can not contain both GATEWAY and VIA");
907 if (nh->fib_nh_gw_family != AF_INET ||
908 nla_get_in_addr(nla) != nh->fib_nh_gw4)
911 struct fib_config cfg2;
914 err = fib_gw_from_via(&cfg2, nlav, extack);
918 switch (nh->fib_nh_gw_family) {
920 if (cfg2.fc_gw_family != AF_INET ||
921 cfg2.fc_gw4 != nh->fib_nh_gw4)
925 if (cfg2.fc_gw_family != AF_INET6 ||
926 ipv6_addr_cmp(&cfg2.fc_gw6,
933 #ifdef CONFIG_IP_ROUTE_CLASSID
934 nla = nla_find(attrs, attrlen, RTA_FLOW);
935 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
940 rtnh = rtnh_next(rtnh, &remaining);
941 } endfor_nexthops(fi);
946 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
954 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
955 int type = nla_type(nla);
963 if (type == RTAX_CC_ALGO) {
964 char tmp[TCP_CA_NAME_MAX];
967 nla_strlcpy(tmp, nla, sizeof(tmp));
968 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
970 if (nla_len(nla) != sizeof(u32))
972 val = nla_get_u32(nla);
975 fi_val = fi->fib_metrics->metrics[type - 1];
976 if (type == RTAX_FEATURES)
977 fi_val &= ~DST_FEATURE_ECN_CA;
986 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
987 u32 table, struct netlink_ext_ack *extack)
989 struct fib6_config cfg = {
991 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
992 .fc_ifindex = nh->fib_nh_oif,
993 .fc_gateway = nh->fib_nh_gw6,
995 struct fib6_nh fib6_nh = {};
998 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1000 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1001 dev_hold(nh->fib_nh_dev);
1002 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1003 nh->fib_nh_scope = RT_SCOPE_LINK;
1005 ipv6_stub->fib6_nh_release(&fib6_nh);
1015 * Semantics of nexthop is very messy by historical reasons.
1016 * We have to take into account, that:
1017 * a) gateway can be actually local interface address,
1018 * so that gatewayed route is direct.
1019 * b) gateway must be on-link address, possibly
1020 * described not by an ifaddr, but also by a direct route.
1021 * c) If both gateway and interface are specified, they should not
1023 * d) If we use tunnel routes, gateway could be not on-link.
1025 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1026 * results in pretty ugly and hairy code with obscure logic.
1028 * I chose to generalized it instead, so that the size
1029 * of code does not increase practically, but it becomes
1030 * much more general.
1031 * Every prefix is assigned a "scope" value: "host" is local address,
1032 * "link" is direct route,
1033 * [ ... "site" ... "interior" ... ]
1034 * and "universe" is true gateway route with global meaning.
1036 * Every prefix refers to a set of "nexthop"s (gw, oif),
1037 * where gw must have narrower scope. This recursion stops
1038 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1039 * which means that gw is forced to be on link.
1041 * Code is still hairy, but now it is apparently logically
1042 * consistent and very flexible. F.e. as by-product it allows
1043 * to co-exists in peace independent exterior and interior
1044 * routing processes.
1046 * Normally it looks as following.
1048 * {universe prefix} -> (gw, oif) [scope link]
1050 * |-> {link prefix} -> (gw, oif) [scope local]
1052 * |-> {local prefix} (terminal node)
1054 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1055 u8 scope, struct netlink_ext_ack *extack)
1057 struct net_device *dev;
1058 struct fib_result res;
1061 if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1062 unsigned int addr_type;
1064 if (scope >= RT_SCOPE_LINK) {
1065 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1068 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1070 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1073 if (!(dev->flags & IFF_UP)) {
1074 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1077 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1078 if (addr_type != RTN_UNICAST) {
1079 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1082 if (!netif_carrier_ok(dev))
1083 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1084 nh->fib_nh_dev = dev;
1086 nh->fib_nh_scope = RT_SCOPE_LINK;
1091 struct fib_table *tbl = NULL;
1092 struct flowi4 fl4 = {
1093 .daddr = nh->fib_nh_gw4,
1094 .flowi4_scope = scope + 1,
1095 .flowi4_oif = nh->fib_nh_oif,
1096 .flowi4_iif = LOOPBACK_IFINDEX,
1099 /* It is not necessary, but requires a bit of thinking */
1100 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1101 fl4.flowi4_scope = RT_SCOPE_LINK;
1104 tbl = fib_get_table(net, table);
1107 err = fib_table_lookup(tbl, &fl4, &res,
1108 FIB_LOOKUP_IGNORE_LINKSTATE |
1111 /* on error or if no table given do full lookup. This
1112 * is needed for example when nexthops are in the local
1113 * table rather than the given table
1116 err = fib_lookup(net, &fl4, &res,
1117 FIB_LOOKUP_IGNORE_LINKSTATE);
1121 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1127 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1128 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1131 nh->fib_nh_scope = res.scope;
1132 nh->fib_nh_oif = FIB_RES_OIF(res);
1133 nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1135 NL_SET_ERR_MSG(extack,
1136 "No egress device for nexthop gateway");
1140 if (!netif_carrier_ok(dev))
1141 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1142 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1148 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1149 struct netlink_ext_ack *extack)
1151 struct in_device *in_dev;
1154 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1155 NL_SET_ERR_MSG(extack,
1156 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1163 in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1167 if (!(in_dev->dev->flags & IFF_UP)) {
1168 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1172 nh->fib_nh_dev = in_dev->dev;
1173 dev_hold(nh->fib_nh_dev);
1174 nh->fib_nh_scope = RT_SCOPE_HOST;
1175 if (!netif_carrier_ok(nh->fib_nh_dev))
1176 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1183 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1184 struct netlink_ext_ack *extack)
1188 if (nh->fib_nh_gw_family == AF_INET)
1189 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1190 else if (nh->fib_nh_gw_family == AF_INET6)
1191 err = fib_check_nh_v6_gw(net, nh, table, extack);
1193 err = fib_check_nh_nongw(net, nh, extack);
1198 static inline unsigned int fib_laddr_hashfn(__be32 val)
1200 unsigned int mask = (fib_info_hash_size - 1);
1202 return ((__force u32)val ^
1203 ((__force u32)val >> 7) ^
1204 ((__force u32)val >> 14)) & mask;
1207 static struct hlist_head *fib_info_hash_alloc(int bytes)
1209 if (bytes <= PAGE_SIZE)
1210 return kzalloc(bytes, GFP_KERNEL);
1212 return (struct hlist_head *)
1213 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1217 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1222 if (bytes <= PAGE_SIZE)
1225 free_pages((unsigned long) hash, get_order(bytes));
1228 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1229 struct hlist_head *new_laddrhash,
1230 unsigned int new_size)
1232 struct hlist_head *old_info_hash, *old_laddrhash;
1233 unsigned int old_size = fib_info_hash_size;
1234 unsigned int i, bytes;
1236 spin_lock_bh(&fib_info_lock);
1237 old_info_hash = fib_info_hash;
1238 old_laddrhash = fib_info_laddrhash;
1239 fib_info_hash_size = new_size;
1241 for (i = 0; i < old_size; i++) {
1242 struct hlist_head *head = &fib_info_hash[i];
1243 struct hlist_node *n;
1244 struct fib_info *fi;
1246 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1247 struct hlist_head *dest;
1248 unsigned int new_hash;
1250 new_hash = fib_info_hashfn(fi);
1251 dest = &new_info_hash[new_hash];
1252 hlist_add_head(&fi->fib_hash, dest);
1255 fib_info_hash = new_info_hash;
1257 for (i = 0; i < old_size; i++) {
1258 struct hlist_head *lhead = &fib_info_laddrhash[i];
1259 struct hlist_node *n;
1260 struct fib_info *fi;
1262 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1263 struct hlist_head *ldest;
1264 unsigned int new_hash;
1266 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1267 ldest = &new_laddrhash[new_hash];
1268 hlist_add_head(&fi->fib_lhash, ldest);
1271 fib_info_laddrhash = new_laddrhash;
1273 spin_unlock_bh(&fib_info_lock);
1275 bytes = old_size * sizeof(struct hlist_head *);
1276 fib_info_hash_free(old_info_hash, bytes);
1277 fib_info_hash_free(old_laddrhash, bytes);
1280 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1281 unsigned char scope)
1285 if (nhc->nhc_family != AF_INET)
1286 return inet_select_addr(nhc->nhc_dev, 0, scope);
1288 nh = container_of(nhc, struct fib_nh, nh_common);
1289 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1290 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1292 return nh->nh_saddr;
1295 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1297 struct fib_nh_common *nhc = res->nhc;
1299 if (res->fi->fib_prefsrc)
1300 return res->fi->fib_prefsrc;
1302 if (nhc->nhc_family == AF_INET) {
1305 nh = container_of(nhc, struct fib_nh, nh_common);
1306 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1307 return nh->nh_saddr;
1310 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1313 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1315 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1316 fib_prefsrc != cfg->fc_dst) {
1317 u32 tb_id = cfg->fc_table;
1320 if (tb_id == RT_TABLE_MAIN)
1321 tb_id = RT_TABLE_LOCAL;
1323 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1324 fib_prefsrc, tb_id);
1326 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1327 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1328 fib_prefsrc, RT_TABLE_LOCAL);
1331 if (rc != RTN_LOCAL)
1337 struct fib_info *fib_create_info(struct fib_config *cfg,
1338 struct netlink_ext_ack *extack)
1341 struct fib_info *fi = NULL;
1342 struct nexthop *nh = NULL;
1343 struct fib_info *ofi;
1345 struct net *net = cfg->fc_nlinfo.nl_net;
1347 if (cfg->fc_type > RTN_MAX)
1350 /* Fast check to catch the most weird cases */
1351 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1352 NL_SET_ERR_MSG(extack, "Invalid scope");
1356 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1357 NL_SET_ERR_MSG(extack,
1358 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1362 if (cfg->fc_nh_id) {
1364 fi = fib_find_info_nh(net, cfg);
1371 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1373 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1379 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1381 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1388 if (fib_info_cnt >= fib_info_hash_size) {
1389 unsigned int new_size = fib_info_hash_size << 1;
1390 struct hlist_head *new_info_hash;
1391 struct hlist_head *new_laddrhash;
1396 bytes = new_size * sizeof(struct hlist_head *);
1397 new_info_hash = fib_info_hash_alloc(bytes);
1398 new_laddrhash = fib_info_hash_alloc(bytes);
1399 if (!new_info_hash || !new_laddrhash) {
1400 fib_info_hash_free(new_info_hash, bytes);
1401 fib_info_hash_free(new_laddrhash, bytes);
1403 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1405 if (!fib_info_hash_size)
1409 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1412 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1413 cfg->fc_mx_len, extack);
1414 if (IS_ERR(fi->fib_metrics)) {
1415 err = PTR_ERR(fi->fib_metrics);
1417 return ERR_PTR(err);
1422 fi->fib_protocol = cfg->fc_protocol;
1423 fi->fib_scope = cfg->fc_scope;
1424 fi->fib_flags = cfg->fc_flags;
1425 fi->fib_priority = cfg->fc_priority;
1426 fi->fib_prefsrc = cfg->fc_prefsrc;
1427 fi->fib_type = cfg->fc_type;
1428 fi->fib_tb_id = cfg->fc_table;
1432 if (!nexthop_get(nh)) {
1433 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1440 change_nexthops(fi) {
1441 nexthop_nh->nh_parent = fi;
1442 } endfor_nexthops(fi)
1445 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1448 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1454 if (fib_props[cfg->fc_type].error) {
1455 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1456 NL_SET_ERR_MSG(extack,
1457 "Gateway, device and multipath can not be specified for this route type");
1462 switch (cfg->fc_type) {
1470 NL_SET_ERR_MSG(extack, "Invalid route type");
1475 if (cfg->fc_scope > RT_SCOPE_HOST) {
1476 NL_SET_ERR_MSG(extack, "Invalid scope");
1481 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1484 } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1485 struct fib_nh *nh = fi->fib_nh;
1487 /* Local address is added. */
1489 NL_SET_ERR_MSG(extack,
1490 "Route with host scope can not have multiple nexthops");
1493 if (nh->fib_nh_gw_family) {
1494 NL_SET_ERR_MSG(extack,
1495 "Route with host scope can not have a gateway");
1498 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1499 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1501 if (!nh->fib_nh_dev)
1506 change_nexthops(fi) {
1507 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1508 cfg->fc_table, cfg->fc_scope,
1512 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1514 } endfor_nexthops(fi)
1515 if (linkdown == fi->fib_nhs)
1516 fi->fib_flags |= RTNH_F_LINKDOWN;
1519 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1520 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1525 change_nexthops(fi) {
1526 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1528 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1529 fi->fib_nh_is_v6 = true;
1530 } endfor_nexthops(fi)
1536 ofi = fib_find_info(fi);
1545 refcount_set(&fi->fib_clntref, 1);
1546 spin_lock_bh(&fib_info_lock);
1547 hlist_add_head(&fi->fib_hash,
1548 &fib_info_hash[fib_info_hashfn(fi)]);
1549 if (fi->fib_prefsrc) {
1550 struct hlist_head *head;
1552 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1553 hlist_add_head(&fi->fib_lhash, head);
1556 list_add(&fi->nh_list, &nh->fi_list);
1558 change_nexthops(fi) {
1559 struct hlist_head *head;
1562 if (!nexthop_nh->fib_nh_dev)
1564 hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1565 head = &fib_info_devhash[hash];
1566 hlist_add_head(&nexthop_nh->nh_hash, head);
1567 } endfor_nexthops(fi)
1569 spin_unlock_bh(&fib_info_lock);
1581 return ERR_PTR(err);
1584 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1585 unsigned char *flags, bool skip_oif)
1587 if (nhc->nhc_flags & RTNH_F_DEAD)
1588 *flags |= RTNH_F_DEAD;
1590 if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1591 *flags |= RTNH_F_LINKDOWN;
1594 switch (nhc->nhc_family) {
1596 if (ip_ignore_linkdown(nhc->nhc_dev))
1597 *flags |= RTNH_F_DEAD;
1600 if (ip6_ignore_linkdown(nhc->nhc_dev))
1601 *flags |= RTNH_F_DEAD;
1607 switch (nhc->nhc_gw_family) {
1609 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1610 goto nla_put_failure;
1613 /* if gateway family does not match nexthop family
1614 * gateway is encoded as RTA_VIA
1616 if (nhc->nhc_gw_family != nhc->nhc_family) {
1617 int alen = sizeof(struct in6_addr);
1621 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1623 goto nla_put_failure;
1625 via = nla_data(nla);
1626 via->rtvia_family = AF_INET6;
1627 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1628 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1629 &nhc->nhc_gw.ipv6) < 0) {
1630 goto nla_put_failure;
1635 *flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1636 if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1637 *flags |= RTNH_F_OFFLOAD;
1639 if (!skip_oif && nhc->nhc_dev &&
1640 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1641 goto nla_put_failure;
1643 if (nhc->nhc_lwtstate &&
1644 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1645 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1646 goto nla_put_failure;
1653 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1655 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1656 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1659 const struct net_device *dev = nhc->nhc_dev;
1660 struct rtnexthop *rtnh;
1661 unsigned char flags = 0;
1663 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1665 goto nla_put_failure;
1667 rtnh->rtnh_hops = nh_weight - 1;
1668 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1670 if (fib_nexthop_info(skb, nhc, &flags, true) < 0)
1671 goto nla_put_failure;
1673 rtnh->rtnh_flags = flags;
1675 /* length of rtnetlink header + attributes */
1676 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1683 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1686 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1687 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1691 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1693 goto nla_put_failure;
1695 if (unlikely(fi->nh)) {
1696 if (nexthop_mpath_fill_node(skb, fi->nh) < 0)
1697 goto nla_put_failure;
1702 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight) < 0)
1703 goto nla_put_failure;
1704 #ifdef CONFIG_IP_ROUTE_CLASSID
1705 if (nh->nh_tclassid &&
1706 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1707 goto nla_put_failure;
1709 } endfor_nexthops(fi);
1712 nla_nest_end(skb, mp);
1720 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1726 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1727 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1728 struct fib_info *fi, unsigned int flags)
1730 unsigned int nhs = fib_info_num_path(fi);
1731 struct nlmsghdr *nlh;
1734 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1738 rtm = nlmsg_data(nlh);
1739 rtm->rtm_family = AF_INET;
1740 rtm->rtm_dst_len = dst_len;
1741 rtm->rtm_src_len = 0;
1744 rtm->rtm_table = tb_id;
1746 rtm->rtm_table = RT_TABLE_COMPAT;
1747 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1748 goto nla_put_failure;
1749 rtm->rtm_type = type;
1750 rtm->rtm_flags = fi->fib_flags;
1751 rtm->rtm_scope = fi->fib_scope;
1752 rtm->rtm_protocol = fi->fib_protocol;
1754 if (rtm->rtm_dst_len &&
1755 nla_put_in_addr(skb, RTA_DST, dst))
1756 goto nla_put_failure;
1757 if (fi->fib_priority &&
1758 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1759 goto nla_put_failure;
1760 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1761 goto nla_put_failure;
1763 if (fi->fib_prefsrc &&
1764 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1765 goto nla_put_failure;
1768 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1769 goto nla_put_failure;
1770 if (nexthop_is_blackhole(fi->nh))
1771 rtm->rtm_type = RTN_BLACKHOLE;
1775 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1776 unsigned char flags = 0;
1778 if (fib_nexthop_info(skb, nhc, &flags, false) < 0)
1779 goto nla_put_failure;
1781 rtm->rtm_flags = flags;
1782 #ifdef CONFIG_IP_ROUTE_CLASSID
1783 if (nhc->nhc_family == AF_INET) {
1786 nh = container_of(nhc, struct fib_nh, nh_common);
1787 if (nh->nh_tclassid &&
1788 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1789 goto nla_put_failure;
1793 if (fib_add_multipath(skb, fi) < 0)
1794 goto nla_put_failure;
1797 nlmsg_end(skb, nlh);
1801 nlmsg_cancel(skb, nlh);
1807 * - local address disappeared -> we must delete all the entries
1809 * - device went down -> we must shutdown all nexthops going via it.
1811 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1814 unsigned int hash = fib_laddr_hashfn(local);
1815 struct hlist_head *head = &fib_info_laddrhash[hash];
1816 struct net *net = dev_net(dev);
1817 int tb_id = l3mdev_fib_table(dev);
1818 struct fib_info *fi;
1820 if (!fib_info_laddrhash || local == 0)
1823 hlist_for_each_entry(fi, head, fib_lhash) {
1824 if (!net_eq(fi->fib_net, net) ||
1825 fi->fib_tb_id != tb_id)
1827 if (fi->fib_prefsrc == local) {
1828 fi->fib_flags |= RTNH_F_DEAD;
1835 static int call_fib_nh_notifiers(struct fib_nh *nh,
1836 enum fib_event_type event_type)
1838 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1839 struct fib_nh_notifier_info info = {
1843 switch (event_type) {
1844 case FIB_EVENT_NH_ADD:
1845 if (nh->fib_nh_flags & RTNH_F_DEAD)
1847 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1849 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1851 case FIB_EVENT_NH_DEL:
1852 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1853 (nh->fib_nh_flags & RTNH_F_DEAD))
1854 return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1855 event_type, &info.info);
1863 /* Update the PMTU of exceptions when:
1864 * - the new MTU of the first hop becomes smaller than the PMTU
1865 * - the old MTU was the same as the PMTU, and it limited discovery of
1866 * larger MTUs on the path. With that limit raised, we can now
1867 * discover larger MTUs
1868 * A special case is locked exceptions, for which the PMTU is smaller
1869 * than the minimal accepted PMTU:
1870 * - if the new MTU is greater than the PMTU, don't make any change
1871 * - otherwise, unlock and set PMTU
1873 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1875 struct fnhe_hash_bucket *bucket;
1878 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1882 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1883 struct fib_nh_exception *fnhe;
1885 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1887 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1888 if (fnhe->fnhe_mtu_locked) {
1889 if (new <= fnhe->fnhe_pmtu) {
1890 fnhe->fnhe_pmtu = new;
1891 fnhe->fnhe_mtu_locked = false;
1893 } else if (new < fnhe->fnhe_pmtu ||
1894 orig == fnhe->fnhe_pmtu) {
1895 fnhe->fnhe_pmtu = new;
1901 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1903 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1904 struct hlist_head *head = &fib_info_devhash[hash];
1907 hlist_for_each_entry(nh, head, nh_hash) {
1908 if (nh->fib_nh_dev == dev)
1909 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1913 /* Event force Flags Description
1914 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1915 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1916 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1917 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1919 * only used when fib_nh is built into fib_info
1921 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1924 int scope = RT_SCOPE_NOWHERE;
1925 struct fib_info *prev_fi = NULL;
1926 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1927 struct hlist_head *head = &fib_info_devhash[hash];
1933 hlist_for_each_entry(nh, head, nh_hash) {
1934 struct fib_info *fi = nh->nh_parent;
1937 BUG_ON(!fi->fib_nhs);
1938 if (nh->fib_nh_dev != dev || fi == prev_fi)
1942 change_nexthops(fi) {
1943 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1945 else if (nexthop_nh->fib_nh_dev == dev &&
1946 nexthop_nh->fib_nh_scope != scope) {
1949 case NETDEV_UNREGISTER:
1950 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1953 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1956 call_fib_nh_notifiers(nexthop_nh,
1960 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1961 if (event == NETDEV_UNREGISTER &&
1962 nexthop_nh->fib_nh_dev == dev) {
1967 } endfor_nexthops(fi)
1968 if (dead == fi->fib_nhs) {
1971 case NETDEV_UNREGISTER:
1972 fi->fib_flags |= RTNH_F_DEAD;
1975 fi->fib_flags |= RTNH_F_LINKDOWN;
1987 /* Must be invoked inside of an RCU protected region. */
1988 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1990 struct fib_info *fi = NULL, *last_resort = NULL;
1991 struct hlist_head *fa_head = res->fa_head;
1992 struct fib_table *tb = res->table;
1993 u8 slen = 32 - res->prefixlen;
1994 int order = -1, last_idx = -1;
1995 struct fib_alias *fa, *fa1 = NULL;
1996 u32 last_prio = res->fi->fib_priority;
1999 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2000 struct fib_info *next_fi = fa->fa_info;
2003 if (fa->fa_slen != slen)
2005 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2007 if (fa->tb_id != tb->tb_id)
2009 if (next_fi->fib_priority > last_prio &&
2010 fa->fa_tos == last_tos) {
2015 if (next_fi->fib_flags & RTNH_F_DEAD)
2017 last_tos = fa->fa_tos;
2018 last_prio = next_fi->fib_priority;
2020 if (next_fi->fib_scope != res->scope ||
2021 fa->fa_type != RTN_UNICAST)
2024 nh = fib_info_nh(next_fi, 0);
2025 if (!nh->fib_nh_gw4 || nh->fib_nh_scope != RT_SCOPE_LINK)
2028 fib_alias_accessed(fa);
2031 if (next_fi != res->fi)
2034 } else if (!fib_detect_death(fi, order, &last_resort,
2035 &last_idx, fa1->fa_default)) {
2036 fib_result_assign(res, fi);
2037 fa1->fa_default = order;
2044 if (order <= 0 || !fi) {
2046 fa1->fa_default = -1;
2050 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2052 fib_result_assign(res, fi);
2053 fa1->fa_default = order;
2058 fib_result_assign(res, last_resort);
2059 fa1->fa_default = last_idx;
2065 * Dead device goes up. We wake up dead nexthops.
2066 * It takes sense only on multipath routes.
2068 * only used when fib_nh is built into fib_info
2070 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2072 struct fib_info *prev_fi;
2074 struct hlist_head *head;
2078 if (!(dev->flags & IFF_UP))
2081 if (nh_flags & RTNH_F_DEAD) {
2082 unsigned int flags = dev_get_flags(dev);
2084 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2085 nh_flags |= RTNH_F_LINKDOWN;
2089 hash = fib_devindex_hashfn(dev->ifindex);
2090 head = &fib_info_devhash[hash];
2093 hlist_for_each_entry(nh, head, nh_hash) {
2094 struct fib_info *fi = nh->nh_parent;
2097 BUG_ON(!fi->fib_nhs);
2098 if (nh->fib_nh_dev != dev || fi == prev_fi)
2103 change_nexthops(fi) {
2104 if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2108 if (!nexthop_nh->fib_nh_dev ||
2109 !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2111 if (nexthop_nh->fib_nh_dev != dev ||
2112 !__in_dev_get_rtnl(dev))
2115 nexthop_nh->fib_nh_flags &= ~nh_flags;
2116 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2117 } endfor_nexthops(fi)
2120 fi->fib_flags &= ~nh_flags;
2130 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2131 static bool fib_good_nh(const struct fib_nh *nh)
2133 int state = NUD_REACHABLE;
2135 if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2136 struct neighbour *n;
2140 if (likely(nh->fib_nh_gw_family == AF_INET))
2141 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2142 (__force u32)nh->fib_nh_gw4);
2143 else if (nh->fib_nh_gw_family == AF_INET6)
2144 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2149 state = n->nud_state;
2151 rcu_read_unlock_bh();
2154 return !!(state & NUD_VALID);
2157 void fib_select_multipath(struct fib_result *res, int hash)
2159 struct fib_info *fi = res->fi;
2160 struct net *net = fi->fib_net;
2163 if (unlikely(res->fi->nh)) {
2164 nexthop_path_fib_result(res, hash);
2168 change_nexthops(fi) {
2169 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2170 if (!fib_good_nh(nexthop_nh))
2173 res->nh_sel = nhsel;
2174 res->nhc = &nexthop_nh->nh_common;
2179 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2182 res->nh_sel = nhsel;
2183 res->nhc = &nexthop_nh->nh_common;
2185 } endfor_nexthops(fi);
2189 void fib_select_path(struct net *net, struct fib_result *res,
2190 struct flowi4 *fl4, const struct sk_buff *skb)
2192 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2195 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2196 if (fib_info_num_path(res->fi) > 1) {
2197 int h = fib_multipath_hash(net, fl4, skb, NULL);
2199 fib_select_multipath(res, h);
2203 if (!res->prefixlen &&
2204 res->table->tb_num_default > 1 &&
2205 res->type == RTN_UNICAST)
2206 fib_select_default(fl4, res);
2210 fl4->saddr = fib_result_prefsrc(net, res);