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)
211 dev_put(nhc->nhc_dev);
212 lwtstate_put(nhc->nhc_lwtstate);
213 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
214 rt_fibinfo_free(&nhc->nhc_rth_input);
215 free_nh_exceptions(nhc);
217 EXPORT_SYMBOL_GPL(fib_nh_common_release);
219 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
221 #ifdef CONFIG_IP_ROUTE_CLASSID
222 if (fib_nh->nh_tclassid)
223 net->ipv4.fib_num_tclassid_users--;
225 fib_nh_common_release(&fib_nh->nh_common);
228 /* Release a nexthop info record */
229 static void free_fib_info_rcu(struct rcu_head *head)
231 struct fib_info *fi = container_of(head, struct fib_info, rcu);
236 change_nexthops(fi) {
237 fib_nh_release(fi->fib_net, nexthop_nh);
238 } endfor_nexthops(fi);
241 ip_fib_metrics_put(fi->fib_metrics);
246 void free_fib_info(struct fib_info *fi)
248 if (fi->fib_dead == 0) {
249 pr_warn("Freeing alive fib_info %p\n", fi);
254 call_rcu(&fi->rcu, free_fib_info_rcu);
256 EXPORT_SYMBOL_GPL(free_fib_info);
258 void fib_release_info(struct fib_info *fi)
260 spin_lock_bh(&fib_info_lock);
261 if (fi && refcount_dec_and_test(&fi->fib_treeref)) {
262 hlist_del(&fi->fib_hash);
264 hlist_del(&fi->fib_lhash);
266 list_del(&fi->nh_list);
268 change_nexthops(fi) {
269 if (!nexthop_nh->fib_nh_dev)
271 hlist_del(&nexthop_nh->nh_hash);
272 } endfor_nexthops(fi)
277 spin_unlock_bh(&fib_info_lock);
280 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
282 const struct fib_nh *onh;
284 if (fi->nh || ofi->nh)
285 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
287 if (ofi->fib_nhs == 0)
291 onh = fib_info_nh(ofi, nhsel);
293 if (nh->fib_nh_oif != onh->fib_nh_oif ||
294 nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
295 nh->fib_nh_scope != onh->fib_nh_scope ||
296 #ifdef CONFIG_IP_ROUTE_MULTIPATH
297 nh->fib_nh_weight != onh->fib_nh_weight ||
299 #ifdef CONFIG_IP_ROUTE_CLASSID
300 nh->nh_tclassid != onh->nh_tclassid ||
302 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
303 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
306 if (nh->fib_nh_gw_family == AF_INET &&
307 nh->fib_nh_gw4 != onh->fib_nh_gw4)
310 if (nh->fib_nh_gw_family == AF_INET6 &&
311 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
313 } endfor_nexthops(fi);
317 static inline unsigned int fib_devindex_hashfn(unsigned int val)
319 unsigned int mask = DEVINDEX_HASHSIZE - 1;
322 (val >> DEVINDEX_HASHBITS) ^
323 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
326 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
327 u32 prefsrc, u32 priority)
329 unsigned int val = init_val;
331 val ^= (protocol << 8) | scope;
338 static unsigned int fib_info_hashfn_result(unsigned int val)
340 unsigned int mask = (fib_info_hash_size - 1);
342 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
345 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
349 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
350 fi->fib_scope, (__force u32)fi->fib_prefsrc,
354 val ^= fib_devindex_hashfn(fi->nh->id);
357 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
358 } endfor_nexthops(fi)
361 return fib_info_hashfn_result(val);
364 /* no metrics, only nexthop id */
365 static struct fib_info *fib_find_info_nh(struct net *net,
366 const struct fib_config *cfg)
368 struct hlist_head *head;
372 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
373 cfg->fc_protocol, cfg->fc_scope,
374 (__force u32)cfg->fc_prefsrc,
376 hash = fib_info_hashfn_result(hash);
377 head = &fib_info_hash[hash];
379 hlist_for_each_entry(fi, head, fib_hash) {
380 if (!net_eq(fi->fib_net, net))
382 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
384 if (cfg->fc_protocol == fi->fib_protocol &&
385 cfg->fc_scope == fi->fib_scope &&
386 cfg->fc_prefsrc == fi->fib_prefsrc &&
387 cfg->fc_priority == fi->fib_priority &&
388 cfg->fc_type == fi->fib_type &&
389 cfg->fc_table == fi->fib_tb_id &&
390 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
397 static struct fib_info *fib_find_info(struct fib_info *nfi)
399 struct hlist_head *head;
403 hash = fib_info_hashfn(nfi);
404 head = &fib_info_hash[hash];
406 hlist_for_each_entry(fi, head, fib_hash) {
407 if (!net_eq(fi->fib_net, nfi->fib_net))
409 if (fi->fib_nhs != nfi->fib_nhs)
411 if (nfi->fib_protocol == fi->fib_protocol &&
412 nfi->fib_scope == fi->fib_scope &&
413 nfi->fib_prefsrc == fi->fib_prefsrc &&
414 nfi->fib_priority == fi->fib_priority &&
415 nfi->fib_type == fi->fib_type &&
416 memcmp(nfi->fib_metrics, fi->fib_metrics,
417 sizeof(u32) * RTAX_MAX) == 0 &&
418 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
419 nh_comp(fi, nfi) == 0)
426 /* Check, that the gateway is already configured.
427 * Used only by redirect accept routine.
429 int ip_fib_check_default(__be32 gw, struct net_device *dev)
431 struct hlist_head *head;
435 spin_lock(&fib_info_lock);
437 hash = fib_devindex_hashfn(dev->ifindex);
438 head = &fib_info_devhash[hash];
439 hlist_for_each_entry(nh, head, nh_hash) {
440 if (nh->fib_nh_dev == dev &&
441 nh->fib_nh_gw4 == gw &&
442 !(nh->fib_nh_flags & RTNH_F_DEAD)) {
443 spin_unlock(&fib_info_lock);
448 spin_unlock(&fib_info_lock);
453 size_t fib_nlmsg_size(struct fib_info *fi)
455 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
456 + nla_total_size(4) /* RTA_TABLE */
457 + nla_total_size(4) /* RTA_DST */
458 + nla_total_size(4) /* RTA_PRIORITY */
459 + nla_total_size(4) /* RTA_PREFSRC */
460 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
461 unsigned int nhs = fib_info_num_path(fi);
463 /* space for nested metrics */
464 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
467 payload += nla_total_size(4); /* RTA_NH_ID */
470 size_t nh_encapsize = 0;
471 /* Also handles the special case nhs == 1 */
473 /* each nexthop is packed in an attribute */
474 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
477 /* may contain flow and gateway attribute */
478 nhsize += 2 * nla_total_size(4);
480 /* grab encap info */
481 for (i = 0; i < fib_info_num_path(fi); i++) {
482 struct fib_nh_common *nhc = fib_info_nhc(fi, i);
484 if (nhc->nhc_lwtstate) {
486 nh_encapsize += lwtunnel_get_encap_size(
489 nh_encapsize += nla_total_size(2);
493 /* all nexthops are packed in a nested attribute */
494 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
501 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
502 int dst_len, u32 tb_id, const struct nl_info *info,
503 unsigned int nlm_flags)
505 struct fib_rt_info fri;
507 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
510 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
514 fri.fi = fa->fa_info;
517 fri.dst_len = dst_len;
518 fri.tos = fa->fa_tos;
519 fri.type = fa->fa_type;
520 fri.offload = fa->offload;
522 fri.offload_failed = fa->offload_failed;
523 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
525 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
526 WARN_ON(err == -EMSGSIZE);
530 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
531 info->nlh, GFP_KERNEL);
535 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
538 static int fib_detect_death(struct fib_info *fi, int order,
539 struct fib_info **last_resort, int *last_idx,
542 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
544 int state = NUD_NONE;
546 if (likely(nhc->nhc_gw_family == AF_INET))
547 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
548 else if (nhc->nhc_gw_family == AF_INET6)
549 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
555 state = n->nud_state;
560 if (state == NUD_REACHABLE)
562 if ((state & NUD_VALID) && order != dflt)
564 if ((state & NUD_VALID) ||
565 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
572 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
573 struct nlattr *encap, u16 encap_type,
574 void *cfg, gfp_t gfp_flags,
575 struct netlink_ext_ack *extack)
579 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
581 if (!nhc->nhc_pcpu_rth_output)
585 struct lwtunnel_state *lwtstate;
587 if (encap_type == LWTUNNEL_ENCAP_NONE) {
588 NL_SET_ERR_MSG(extack, "LWT encap type not specified");
592 err = lwtunnel_build_state(net, encap_type, encap,
593 nhc->nhc_family, cfg, &lwtstate,
598 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
604 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
605 nhc->nhc_pcpu_rth_output = NULL;
608 EXPORT_SYMBOL_GPL(fib_nh_common_init);
610 int fib_nh_init(struct net *net, struct fib_nh *nh,
611 struct fib_config *cfg, int nh_weight,
612 struct netlink_ext_ack *extack)
616 nh->fib_nh_family = AF_INET;
618 err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
619 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
623 nh->fib_nh_oif = cfg->fc_oif;
624 nh->fib_nh_gw_family = cfg->fc_gw_family;
625 if (cfg->fc_gw_family == AF_INET)
626 nh->fib_nh_gw4 = cfg->fc_gw4;
627 else if (cfg->fc_gw_family == AF_INET6)
628 nh->fib_nh_gw6 = cfg->fc_gw6;
630 nh->fib_nh_flags = cfg->fc_flags;
632 #ifdef CONFIG_IP_ROUTE_CLASSID
633 nh->nh_tclassid = cfg->fc_flow;
635 net->ipv4.fib_num_tclassid_users++;
637 #ifdef CONFIG_IP_ROUTE_MULTIPATH
638 nh->fib_nh_weight = nh_weight;
643 #ifdef CONFIG_IP_ROUTE_MULTIPATH
645 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
646 struct netlink_ext_ack *extack)
650 while (rtnh_ok(rtnh, remaining)) {
652 rtnh = rtnh_next(rtnh, &remaining);
655 /* leftover implies invalid nexthop configuration, discard it */
657 NL_SET_ERR_MSG(extack,
658 "Invalid nexthop configuration - extra data after nexthops");
665 /* only called when fib_nh is integrated into fib_info */
666 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
667 int remaining, struct fib_config *cfg,
668 struct netlink_ext_ack *extack)
670 struct net *net = fi->fib_net;
671 struct fib_config fib_cfg;
675 change_nexthops(fi) {
678 memset(&fib_cfg, 0, sizeof(fib_cfg));
680 if (!rtnh_ok(rtnh, remaining)) {
681 NL_SET_ERR_MSG(extack,
682 "Invalid nexthop configuration - extra data after nexthop");
686 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
687 NL_SET_ERR_MSG(extack,
688 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
692 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
693 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
695 attrlen = rtnh_attrlen(rtnh);
697 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
699 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
700 nlav = nla_find(attrs, attrlen, RTA_VIA);
702 NL_SET_ERR_MSG(extack,
703 "Nexthop configuration can not contain both GATEWAY and VIA");
707 fib_cfg.fc_gw4 = nla_get_in_addr(nla);
709 fib_cfg.fc_gw_family = AF_INET;
711 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
716 nla = nla_find(attrs, attrlen, RTA_FLOW);
718 fib_cfg.fc_flow = nla_get_u32(nla);
720 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
721 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
723 fib_cfg.fc_encap_type = nla_get_u16(nla);
726 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
727 rtnh->rtnh_hops + 1, extack);
731 rtnh = rtnh_next(rtnh, &remaining);
732 } endfor_nexthops(fi);
735 nh = fib_info_nh(fi, 0);
736 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
737 NL_SET_ERR_MSG(extack,
738 "Nexthop device index does not match RTA_OIF");
741 if (cfg->fc_gw_family) {
742 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
743 (cfg->fc_gw_family == AF_INET &&
744 nh->fib_nh_gw4 != cfg->fc_gw4) ||
745 (cfg->fc_gw_family == AF_INET6 &&
746 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
747 NL_SET_ERR_MSG(extack,
748 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
752 #ifdef CONFIG_IP_ROUTE_CLASSID
753 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
754 NL_SET_ERR_MSG(extack,
755 "Nexthop class id does not match RTA_FLOW");
764 /* only called when fib_nh is integrated into fib_info */
765 static void fib_rebalance(struct fib_info *fi)
770 if (fib_info_num_path(fi) < 2)
775 if (nh->fib_nh_flags & RTNH_F_DEAD)
778 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
779 nh->fib_nh_flags & RTNH_F_LINKDOWN)
782 total += nh->fib_nh_weight;
783 } endfor_nexthops(fi);
786 change_nexthops(fi) {
789 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
791 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
792 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
795 w += nexthop_nh->fib_nh_weight;
796 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
800 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
801 } endfor_nexthops(fi);
803 #else /* CONFIG_IP_ROUTE_MULTIPATH */
805 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
806 int remaining, struct fib_config *cfg,
807 struct netlink_ext_ack *extack)
809 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
814 #define fib_rebalance(fi) do { } while (0)
816 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
818 static int fib_encap_match(struct net *net, u16 encap_type,
819 struct nlattr *encap,
820 const struct fib_nh *nh,
821 const struct fib_config *cfg,
822 struct netlink_ext_ack *extack)
824 struct lwtunnel_state *lwtstate;
827 if (encap_type == LWTUNNEL_ENCAP_NONE)
830 ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
831 cfg, &lwtstate, extack);
833 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
834 lwtstate_free(lwtstate);
840 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
841 struct netlink_ext_ack *extack)
843 #ifdef CONFIG_IP_ROUTE_MULTIPATH
844 struct rtnexthop *rtnh;
848 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
852 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
857 if (cfg->fc_oif || cfg->fc_gw_family) {
858 struct fib_nh *nh = fib_info_nh(fi, 0);
861 if (fib_encap_match(net, cfg->fc_encap_type,
862 cfg->fc_encap, nh, cfg, extack))
865 #ifdef CONFIG_IP_ROUTE_CLASSID
867 cfg->fc_flow != nh->nh_tclassid)
870 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
871 (cfg->fc_gw_family &&
872 cfg->fc_gw_family != nh->fib_nh_gw_family))
875 if (cfg->fc_gw_family == AF_INET &&
876 cfg->fc_gw4 != nh->fib_nh_gw4)
879 if (cfg->fc_gw_family == AF_INET6 &&
880 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
886 #ifdef CONFIG_IP_ROUTE_MULTIPATH
891 remaining = cfg->fc_mp_len;
896 if (!rtnh_ok(rtnh, remaining))
899 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
902 attrlen = rtnh_attrlen(rtnh);
904 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
906 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
907 nlav = nla_find(attrs, attrlen, RTA_VIA);
909 NL_SET_ERR_MSG(extack,
910 "Nexthop configuration can not contain both GATEWAY and VIA");
915 if (nh->fib_nh_gw_family != AF_INET ||
916 nla_get_in_addr(nla) != nh->fib_nh_gw4)
919 struct fib_config cfg2;
922 err = fib_gw_from_via(&cfg2, nlav, extack);
926 switch (nh->fib_nh_gw_family) {
928 if (cfg2.fc_gw_family != AF_INET ||
929 cfg2.fc_gw4 != nh->fib_nh_gw4)
933 if (cfg2.fc_gw_family != AF_INET6 ||
934 ipv6_addr_cmp(&cfg2.fc_gw6,
941 #ifdef CONFIG_IP_ROUTE_CLASSID
942 nla = nla_find(attrs, attrlen, RTA_FLOW);
943 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
948 rtnh = rtnh_next(rtnh, &remaining);
949 } endfor_nexthops(fi);
954 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
962 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
963 int type = nla_type(nla);
971 if (type == RTAX_CC_ALGO) {
972 char tmp[TCP_CA_NAME_MAX];
975 nla_strscpy(tmp, nla, sizeof(tmp));
976 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
978 if (nla_len(nla) != sizeof(u32))
980 val = nla_get_u32(nla);
983 fi_val = fi->fib_metrics->metrics[type - 1];
984 if (type == RTAX_FEATURES)
985 fi_val &= ~DST_FEATURE_ECN_CA;
994 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
995 u32 table, struct netlink_ext_ack *extack)
997 struct fib6_config cfg = {
999 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1000 .fc_ifindex = nh->fib_nh_oif,
1001 .fc_gateway = nh->fib_nh_gw6,
1003 struct fib6_nh fib6_nh = {};
1006 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1008 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1009 dev_hold(nh->fib_nh_dev);
1010 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1011 nh->fib_nh_scope = RT_SCOPE_LINK;
1013 ipv6_stub->fib6_nh_release(&fib6_nh);
1023 * Semantics of nexthop is very messy by historical reasons.
1024 * We have to take into account, that:
1025 * a) gateway can be actually local interface address,
1026 * so that gatewayed route is direct.
1027 * b) gateway must be on-link address, possibly
1028 * described not by an ifaddr, but also by a direct route.
1029 * c) If both gateway and interface are specified, they should not
1031 * d) If we use tunnel routes, gateway could be not on-link.
1033 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1034 * results in pretty ugly and hairy code with obscure logic.
1036 * I chose to generalized it instead, so that the size
1037 * of code does not increase practically, but it becomes
1038 * much more general.
1039 * Every prefix is assigned a "scope" value: "host" is local address,
1040 * "link" is direct route,
1041 * [ ... "site" ... "interior" ... ]
1042 * and "universe" is true gateway route with global meaning.
1044 * Every prefix refers to a set of "nexthop"s (gw, oif),
1045 * where gw must have narrower scope. This recursion stops
1046 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1047 * which means that gw is forced to be on link.
1049 * Code is still hairy, but now it is apparently logically
1050 * consistent and very flexible. F.e. as by-product it allows
1051 * to co-exists in peace independent exterior and interior
1052 * routing processes.
1054 * Normally it looks as following.
1056 * {universe prefix} -> (gw, oif) [scope link]
1058 * |-> {link prefix} -> (gw, oif) [scope local]
1060 * |-> {local prefix} (terminal node)
1062 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1063 u8 scope, struct netlink_ext_ack *extack)
1065 struct net_device *dev;
1066 struct fib_result res;
1069 if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1070 unsigned int addr_type;
1072 if (scope >= RT_SCOPE_LINK) {
1073 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1076 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1078 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1081 if (!(dev->flags & IFF_UP)) {
1082 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1085 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1086 if (addr_type != RTN_UNICAST) {
1087 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1090 if (!netif_carrier_ok(dev))
1091 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1092 nh->fib_nh_dev = dev;
1094 nh->fib_nh_scope = RT_SCOPE_LINK;
1099 struct fib_table *tbl = NULL;
1100 struct flowi4 fl4 = {
1101 .daddr = nh->fib_nh_gw4,
1102 .flowi4_scope = scope + 1,
1103 .flowi4_oif = nh->fib_nh_oif,
1104 .flowi4_iif = LOOPBACK_IFINDEX,
1107 /* It is not necessary, but requires a bit of thinking */
1108 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1109 fl4.flowi4_scope = RT_SCOPE_LINK;
1111 if (table && table != RT_TABLE_MAIN)
1112 tbl = fib_get_table(net, table);
1115 err = fib_table_lookup(tbl, &fl4, &res,
1116 FIB_LOOKUP_IGNORE_LINKSTATE |
1119 /* on error or if no table given do full lookup. This
1120 * is needed for example when nexthops are in the local
1121 * table rather than the given table
1124 err = fib_lookup(net, &fl4, &res,
1125 FIB_LOOKUP_IGNORE_LINKSTATE);
1129 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1135 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1136 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1139 nh->fib_nh_scope = res.scope;
1140 nh->fib_nh_oif = FIB_RES_OIF(res);
1141 nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1143 NL_SET_ERR_MSG(extack,
1144 "No egress device for nexthop gateway");
1148 if (!netif_carrier_ok(dev))
1149 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1150 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1156 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1157 struct netlink_ext_ack *extack)
1159 struct in_device *in_dev;
1162 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1163 NL_SET_ERR_MSG(extack,
1164 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1171 in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1175 if (!(in_dev->dev->flags & IFF_UP)) {
1176 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1180 nh->fib_nh_dev = in_dev->dev;
1181 dev_hold(nh->fib_nh_dev);
1182 nh->fib_nh_scope = RT_SCOPE_HOST;
1183 if (!netif_carrier_ok(nh->fib_nh_dev))
1184 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1191 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1192 struct netlink_ext_ack *extack)
1196 if (nh->fib_nh_gw_family == AF_INET)
1197 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1198 else if (nh->fib_nh_gw_family == AF_INET6)
1199 err = fib_check_nh_v6_gw(net, nh, table, extack);
1201 err = fib_check_nh_nongw(net, nh, extack);
1206 static inline unsigned int fib_laddr_hashfn(__be32 val)
1208 unsigned int mask = (fib_info_hash_size - 1);
1210 return ((__force u32)val ^
1211 ((__force u32)val >> 7) ^
1212 ((__force u32)val >> 14)) & mask;
1215 static struct hlist_head *fib_info_hash_alloc(int bytes)
1217 if (bytes <= PAGE_SIZE)
1218 return kzalloc(bytes, GFP_KERNEL);
1220 return (struct hlist_head *)
1221 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1225 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1230 if (bytes <= PAGE_SIZE)
1233 free_pages((unsigned long) hash, get_order(bytes));
1236 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1237 struct hlist_head *new_laddrhash,
1238 unsigned int new_size)
1240 struct hlist_head *old_info_hash, *old_laddrhash;
1241 unsigned int old_size = fib_info_hash_size;
1242 unsigned int i, bytes;
1244 spin_lock_bh(&fib_info_lock);
1245 old_info_hash = fib_info_hash;
1246 old_laddrhash = fib_info_laddrhash;
1247 fib_info_hash_size = new_size;
1249 for (i = 0; i < old_size; i++) {
1250 struct hlist_head *head = &fib_info_hash[i];
1251 struct hlist_node *n;
1252 struct fib_info *fi;
1254 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1255 struct hlist_head *dest;
1256 unsigned int new_hash;
1258 new_hash = fib_info_hashfn(fi);
1259 dest = &new_info_hash[new_hash];
1260 hlist_add_head(&fi->fib_hash, dest);
1263 fib_info_hash = new_info_hash;
1265 for (i = 0; i < old_size; i++) {
1266 struct hlist_head *lhead = &fib_info_laddrhash[i];
1267 struct hlist_node *n;
1268 struct fib_info *fi;
1270 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1271 struct hlist_head *ldest;
1272 unsigned int new_hash;
1274 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1275 ldest = &new_laddrhash[new_hash];
1276 hlist_add_head(&fi->fib_lhash, ldest);
1279 fib_info_laddrhash = new_laddrhash;
1281 spin_unlock_bh(&fib_info_lock);
1283 bytes = old_size * sizeof(struct hlist_head *);
1284 fib_info_hash_free(old_info_hash, bytes);
1285 fib_info_hash_free(old_laddrhash, bytes);
1288 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1289 unsigned char scope)
1293 if (nhc->nhc_family != AF_INET)
1294 return inet_select_addr(nhc->nhc_dev, 0, scope);
1296 nh = container_of(nhc, struct fib_nh, nh_common);
1297 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1298 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1300 return nh->nh_saddr;
1303 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1305 struct fib_nh_common *nhc = res->nhc;
1307 if (res->fi->fib_prefsrc)
1308 return res->fi->fib_prefsrc;
1310 if (nhc->nhc_family == AF_INET) {
1313 nh = container_of(nhc, struct fib_nh, nh_common);
1314 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1315 return nh->nh_saddr;
1318 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1321 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1323 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1324 fib_prefsrc != cfg->fc_dst) {
1325 u32 tb_id = cfg->fc_table;
1328 if (tb_id == RT_TABLE_MAIN)
1329 tb_id = RT_TABLE_LOCAL;
1331 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1332 fib_prefsrc, tb_id);
1334 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1335 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1336 fib_prefsrc, RT_TABLE_LOCAL);
1339 if (rc != RTN_LOCAL)
1345 struct fib_info *fib_create_info(struct fib_config *cfg,
1346 struct netlink_ext_ack *extack)
1349 struct fib_info *fi = NULL;
1350 struct nexthop *nh = NULL;
1351 struct fib_info *ofi;
1353 struct net *net = cfg->fc_nlinfo.nl_net;
1355 if (cfg->fc_type > RTN_MAX)
1358 /* Fast check to catch the most weird cases */
1359 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1360 NL_SET_ERR_MSG(extack, "Invalid scope");
1364 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1365 NL_SET_ERR_MSG(extack,
1366 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1370 if (cfg->fc_nh_id) {
1372 fi = fib_find_info_nh(net, cfg);
1374 refcount_inc(&fi->fib_treeref);
1379 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1381 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1387 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1389 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1396 if (fib_info_cnt >= fib_info_hash_size) {
1397 unsigned int new_size = fib_info_hash_size << 1;
1398 struct hlist_head *new_info_hash;
1399 struct hlist_head *new_laddrhash;
1404 bytes = new_size * sizeof(struct hlist_head *);
1405 new_info_hash = fib_info_hash_alloc(bytes);
1406 new_laddrhash = fib_info_hash_alloc(bytes);
1407 if (!new_info_hash || !new_laddrhash) {
1408 fib_info_hash_free(new_info_hash, bytes);
1409 fib_info_hash_free(new_laddrhash, bytes);
1411 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1413 if (!fib_info_hash_size)
1417 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1420 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1421 cfg->fc_mx_len, extack);
1422 if (IS_ERR(fi->fib_metrics)) {
1423 err = PTR_ERR(fi->fib_metrics);
1425 return ERR_PTR(err);
1430 fi->fib_protocol = cfg->fc_protocol;
1431 fi->fib_scope = cfg->fc_scope;
1432 fi->fib_flags = cfg->fc_flags;
1433 fi->fib_priority = cfg->fc_priority;
1434 fi->fib_prefsrc = cfg->fc_prefsrc;
1435 fi->fib_type = cfg->fc_type;
1436 fi->fib_tb_id = cfg->fc_table;
1440 if (!nexthop_get(nh)) {
1441 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1448 change_nexthops(fi) {
1449 nexthop_nh->nh_parent = fi;
1450 } endfor_nexthops(fi)
1453 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1456 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1462 if (fib_props[cfg->fc_type].error) {
1463 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1464 NL_SET_ERR_MSG(extack,
1465 "Gateway, device and multipath can not be specified for this route type");
1470 switch (cfg->fc_type) {
1478 NL_SET_ERR_MSG(extack, "Invalid route type");
1483 if (cfg->fc_scope > RT_SCOPE_HOST) {
1484 NL_SET_ERR_MSG(extack, "Invalid scope");
1489 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1492 } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1493 struct fib_nh *nh = fi->fib_nh;
1495 /* Local address is added. */
1497 NL_SET_ERR_MSG(extack,
1498 "Route with host scope can not have multiple nexthops");
1501 if (nh->fib_nh_gw_family) {
1502 NL_SET_ERR_MSG(extack,
1503 "Route with host scope can not have a gateway");
1506 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1507 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1509 if (!nh->fib_nh_dev)
1514 change_nexthops(fi) {
1515 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1516 cfg->fc_table, cfg->fc_scope,
1520 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1522 } endfor_nexthops(fi)
1523 if (linkdown == fi->fib_nhs)
1524 fi->fib_flags |= RTNH_F_LINKDOWN;
1527 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1528 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1533 change_nexthops(fi) {
1534 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1536 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1537 fi->fib_nh_is_v6 = true;
1538 } endfor_nexthops(fi)
1544 ofi = fib_find_info(fi);
1548 refcount_inc(&ofi->fib_treeref);
1552 refcount_set(&fi->fib_treeref, 1);
1553 refcount_set(&fi->fib_clntref, 1);
1554 spin_lock_bh(&fib_info_lock);
1555 hlist_add_head(&fi->fib_hash,
1556 &fib_info_hash[fib_info_hashfn(fi)]);
1557 if (fi->fib_prefsrc) {
1558 struct hlist_head *head;
1560 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1561 hlist_add_head(&fi->fib_lhash, head);
1564 list_add(&fi->nh_list, &nh->fi_list);
1566 change_nexthops(fi) {
1567 struct hlist_head *head;
1570 if (!nexthop_nh->fib_nh_dev)
1572 hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1573 head = &fib_info_devhash[hash];
1574 hlist_add_head(&nexthop_nh->nh_hash, head);
1575 } endfor_nexthops(fi)
1577 spin_unlock_bh(&fib_info_lock);
1589 return ERR_PTR(err);
1592 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1593 u8 rt_family, unsigned char *flags, bool skip_oif)
1595 if (nhc->nhc_flags & RTNH_F_DEAD)
1596 *flags |= RTNH_F_DEAD;
1598 if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1599 *flags |= RTNH_F_LINKDOWN;
1602 switch (nhc->nhc_family) {
1604 if (ip_ignore_linkdown(nhc->nhc_dev))
1605 *flags |= RTNH_F_DEAD;
1608 if (ip6_ignore_linkdown(nhc->nhc_dev))
1609 *flags |= RTNH_F_DEAD;
1615 switch (nhc->nhc_gw_family) {
1617 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1618 goto nla_put_failure;
1621 /* if gateway family does not match nexthop family
1622 * gateway is encoded as RTA_VIA
1624 if (rt_family != nhc->nhc_gw_family) {
1625 int alen = sizeof(struct in6_addr);
1629 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1631 goto nla_put_failure;
1633 via = nla_data(nla);
1634 via->rtvia_family = AF_INET6;
1635 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1636 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1637 &nhc->nhc_gw.ipv6) < 0) {
1638 goto nla_put_failure;
1643 *flags |= (nhc->nhc_flags &
1644 (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1646 if (!skip_oif && nhc->nhc_dev &&
1647 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1648 goto nla_put_failure;
1650 if (nhc->nhc_lwtstate &&
1651 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1652 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1653 goto nla_put_failure;
1660 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1662 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1663 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1664 int nh_weight, u8 rt_family, u32 nh_tclassid)
1666 const struct net_device *dev = nhc->nhc_dev;
1667 struct rtnexthop *rtnh;
1668 unsigned char flags = 0;
1670 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1672 goto nla_put_failure;
1674 rtnh->rtnh_hops = nh_weight - 1;
1675 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1677 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1678 goto nla_put_failure;
1680 rtnh->rtnh_flags = flags;
1682 if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1683 goto nla_put_failure;
1685 /* length of rtnetlink header + attributes */
1686 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1693 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1696 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1697 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1701 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1703 goto nla_put_failure;
1705 if (unlikely(fi->nh)) {
1706 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1707 goto nla_put_failure;
1712 u32 nh_tclassid = 0;
1713 #ifdef CONFIG_IP_ROUTE_CLASSID
1714 nh_tclassid = nh->nh_tclassid;
1716 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1717 AF_INET, nh_tclassid) < 0)
1718 goto nla_put_failure;
1719 } endfor_nexthops(fi);
1722 nla_nest_end(skb, mp);
1730 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1736 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1737 const struct fib_rt_info *fri, unsigned int flags)
1739 unsigned int nhs = fib_info_num_path(fri->fi);
1740 struct fib_info *fi = fri->fi;
1741 u32 tb_id = fri->tb_id;
1742 struct nlmsghdr *nlh;
1745 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1749 rtm = nlmsg_data(nlh);
1750 rtm->rtm_family = AF_INET;
1751 rtm->rtm_dst_len = fri->dst_len;
1752 rtm->rtm_src_len = 0;
1753 rtm->rtm_tos = fri->tos;
1755 rtm->rtm_table = tb_id;
1757 rtm->rtm_table = RT_TABLE_COMPAT;
1758 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1759 goto nla_put_failure;
1760 rtm->rtm_type = fri->type;
1761 rtm->rtm_flags = fi->fib_flags;
1762 rtm->rtm_scope = fi->fib_scope;
1763 rtm->rtm_protocol = fi->fib_protocol;
1765 if (rtm->rtm_dst_len &&
1766 nla_put_in_addr(skb, RTA_DST, fri->dst))
1767 goto nla_put_failure;
1768 if (fi->fib_priority &&
1769 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1770 goto nla_put_failure;
1771 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1772 goto nla_put_failure;
1774 if (fi->fib_prefsrc &&
1775 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1776 goto nla_put_failure;
1779 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1780 goto nla_put_failure;
1781 if (nexthop_is_blackhole(fi->nh))
1782 rtm->rtm_type = RTN_BLACKHOLE;
1783 if (!fi->fib_net->ipv4.sysctl_nexthop_compat_mode)
1788 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1789 unsigned char flags = 0;
1791 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1792 goto nla_put_failure;
1794 rtm->rtm_flags = flags;
1795 #ifdef CONFIG_IP_ROUTE_CLASSID
1796 if (nhc->nhc_family == AF_INET) {
1799 nh = container_of(nhc, struct fib_nh, nh_common);
1800 if (nh->nh_tclassid &&
1801 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1802 goto nla_put_failure;
1806 if (fib_add_multipath(skb, fi) < 0)
1807 goto nla_put_failure;
1812 rtm->rtm_flags |= RTM_F_OFFLOAD;
1814 rtm->rtm_flags |= RTM_F_TRAP;
1815 if (fri->offload_failed)
1816 rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1818 nlmsg_end(skb, nlh);
1822 nlmsg_cancel(skb, nlh);
1828 * - local address disappeared -> we must delete all the entries
1830 * - device went down -> we must shutdown all nexthops going via it.
1832 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1835 unsigned int hash = fib_laddr_hashfn(local);
1836 struct hlist_head *head = &fib_info_laddrhash[hash];
1837 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1838 struct net *net = dev_net(dev);
1839 struct fib_info *fi;
1841 if (!fib_info_laddrhash || local == 0)
1844 hlist_for_each_entry(fi, head, fib_lhash) {
1845 if (!net_eq(fi->fib_net, net) ||
1846 fi->fib_tb_id != tb_id)
1848 if (fi->fib_prefsrc == local) {
1849 fi->fib_flags |= RTNH_F_DEAD;
1856 static int call_fib_nh_notifiers(struct fib_nh *nh,
1857 enum fib_event_type event_type)
1859 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1860 struct fib_nh_notifier_info info = {
1864 switch (event_type) {
1865 case FIB_EVENT_NH_ADD:
1866 if (nh->fib_nh_flags & RTNH_F_DEAD)
1868 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1870 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1872 case FIB_EVENT_NH_DEL:
1873 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1874 (nh->fib_nh_flags & RTNH_F_DEAD))
1875 return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1876 event_type, &info.info);
1885 /* Update the PMTU of exceptions when:
1886 * - the new MTU of the first hop becomes smaller than the PMTU
1887 * - the old MTU was the same as the PMTU, and it limited discovery of
1888 * larger MTUs on the path. With that limit raised, we can now
1889 * discover larger MTUs
1890 * A special case is locked exceptions, for which the PMTU is smaller
1891 * than the minimal accepted PMTU:
1892 * - if the new MTU is greater than the PMTU, don't make any change
1893 * - otherwise, unlock and set PMTU
1895 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1897 struct fnhe_hash_bucket *bucket;
1900 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1904 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1905 struct fib_nh_exception *fnhe;
1907 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1909 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1910 if (fnhe->fnhe_mtu_locked) {
1911 if (new <= fnhe->fnhe_pmtu) {
1912 fnhe->fnhe_pmtu = new;
1913 fnhe->fnhe_mtu_locked = false;
1915 } else if (new < fnhe->fnhe_pmtu ||
1916 orig == fnhe->fnhe_pmtu) {
1917 fnhe->fnhe_pmtu = new;
1923 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1925 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1926 struct hlist_head *head = &fib_info_devhash[hash];
1929 hlist_for_each_entry(nh, head, nh_hash) {
1930 if (nh->fib_nh_dev == dev)
1931 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1935 /* Event force Flags Description
1936 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1937 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1938 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1939 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1941 * only used when fib_nh is built into fib_info
1943 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1946 int scope = RT_SCOPE_NOWHERE;
1947 struct fib_info *prev_fi = NULL;
1948 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1949 struct hlist_head *head = &fib_info_devhash[hash];
1955 hlist_for_each_entry(nh, head, nh_hash) {
1956 struct fib_info *fi = nh->nh_parent;
1959 BUG_ON(!fi->fib_nhs);
1960 if (nh->fib_nh_dev != dev || fi == prev_fi)
1964 change_nexthops(fi) {
1965 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1967 else if (nexthop_nh->fib_nh_dev == dev &&
1968 nexthop_nh->fib_nh_scope != scope) {
1971 case NETDEV_UNREGISTER:
1972 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1975 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1978 call_fib_nh_notifiers(nexthop_nh,
1982 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1983 if (event == NETDEV_UNREGISTER &&
1984 nexthop_nh->fib_nh_dev == dev) {
1989 } endfor_nexthops(fi)
1990 if (dead == fi->fib_nhs) {
1993 case NETDEV_UNREGISTER:
1994 fi->fib_flags |= RTNH_F_DEAD;
1997 fi->fib_flags |= RTNH_F_LINKDOWN;
2009 /* Must be invoked inside of an RCU protected region. */
2010 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2012 struct fib_info *fi = NULL, *last_resort = NULL;
2013 struct hlist_head *fa_head = res->fa_head;
2014 struct fib_table *tb = res->table;
2015 u8 slen = 32 - res->prefixlen;
2016 int order = -1, last_idx = -1;
2017 struct fib_alias *fa, *fa1 = NULL;
2018 u32 last_prio = res->fi->fib_priority;
2021 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2022 struct fib_info *next_fi = fa->fa_info;
2023 struct fib_nh_common *nhc;
2025 if (fa->fa_slen != slen)
2027 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2029 if (fa->tb_id != tb->tb_id)
2031 if (next_fi->fib_priority > last_prio &&
2032 fa->fa_tos == last_tos) {
2037 if (next_fi->fib_flags & RTNH_F_DEAD)
2039 last_tos = fa->fa_tos;
2040 last_prio = next_fi->fib_priority;
2042 if (next_fi->fib_scope != res->scope ||
2043 fa->fa_type != RTN_UNICAST)
2046 nhc = fib_info_nhc(next_fi, 0);
2047 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2050 fib_alias_accessed(fa);
2053 if (next_fi != res->fi)
2056 } else if (!fib_detect_death(fi, order, &last_resort,
2057 &last_idx, fa1->fa_default)) {
2058 fib_result_assign(res, fi);
2059 fa1->fa_default = order;
2066 if (order <= 0 || !fi) {
2068 fa1->fa_default = -1;
2072 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2074 fib_result_assign(res, fi);
2075 fa1->fa_default = order;
2080 fib_result_assign(res, last_resort);
2081 fa1->fa_default = last_idx;
2087 * Dead device goes up. We wake up dead nexthops.
2088 * It takes sense only on multipath routes.
2090 * only used when fib_nh is built into fib_info
2092 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2094 struct fib_info *prev_fi;
2096 struct hlist_head *head;
2100 if (!(dev->flags & IFF_UP))
2103 if (nh_flags & RTNH_F_DEAD) {
2104 unsigned int flags = dev_get_flags(dev);
2106 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2107 nh_flags |= RTNH_F_LINKDOWN;
2111 hash = fib_devindex_hashfn(dev->ifindex);
2112 head = &fib_info_devhash[hash];
2115 hlist_for_each_entry(nh, head, nh_hash) {
2116 struct fib_info *fi = nh->nh_parent;
2119 BUG_ON(!fi->fib_nhs);
2120 if (nh->fib_nh_dev != dev || fi == prev_fi)
2125 change_nexthops(fi) {
2126 if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2130 if (!nexthop_nh->fib_nh_dev ||
2131 !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2133 if (nexthop_nh->fib_nh_dev != dev ||
2134 !__in_dev_get_rtnl(dev))
2137 nexthop_nh->fib_nh_flags &= ~nh_flags;
2138 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2139 } endfor_nexthops(fi)
2142 fi->fib_flags &= ~nh_flags;
2152 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2153 static bool fib_good_nh(const struct fib_nh *nh)
2155 int state = NUD_REACHABLE;
2157 if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2158 struct neighbour *n;
2162 if (likely(nh->fib_nh_gw_family == AF_INET))
2163 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2164 (__force u32)nh->fib_nh_gw4);
2165 else if (nh->fib_nh_gw_family == AF_INET6)
2166 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2171 state = n->nud_state;
2173 rcu_read_unlock_bh();
2176 return !!(state & NUD_VALID);
2179 void fib_select_multipath(struct fib_result *res, int hash)
2181 struct fib_info *fi = res->fi;
2182 struct net *net = fi->fib_net;
2185 if (unlikely(res->fi->nh)) {
2186 nexthop_path_fib_result(res, hash);
2190 change_nexthops(fi) {
2191 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2192 if (!fib_good_nh(nexthop_nh))
2195 res->nh_sel = nhsel;
2196 res->nhc = &nexthop_nh->nh_common;
2201 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2204 res->nh_sel = nhsel;
2205 res->nhc = &nexthop_nh->nh_common;
2207 } endfor_nexthops(fi);
2211 void fib_select_path(struct net *net, struct fib_result *res,
2212 struct flowi4 *fl4, const struct sk_buff *skb)
2214 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2217 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2218 if (fib_info_num_path(res->fi) > 1) {
2219 int h = fib_multipath_hash(net, fl4, skb, NULL);
2221 fib_select_multipath(res, h);
2225 if (!res->prefixlen &&
2226 res->table->tb_num_default > 1 &&
2227 res->type == RTN_UNICAST)
2228 fib_select_default(fl4, res);
2232 fl4->saddr = fib_result_prefsrc(net, res);