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>
32 #include <linux/hash.h>
35 #include <net/inet_dscp.h>
37 #include <net/protocol.h>
38 #include <net/route.h>
41 #include <net/ip_fib.h>
42 #include <net/ip6_fib.h>
43 #include <net/nexthop.h>
44 #include <net/netlink.h>
46 #include <net/lwtunnel.h>
47 #include <net/fib_notifier.h>
48 #include <net/addrconf.h>
50 #include "fib_lookup.h"
52 static DEFINE_SPINLOCK(fib_info_lock);
53 static struct hlist_head *fib_info_hash;
54 static struct hlist_head *fib_info_laddrhash;
55 static unsigned int fib_info_hash_size;
56 static unsigned int fib_info_hash_bits;
57 static unsigned int fib_info_cnt;
59 #define DEVINDEX_HASHBITS 8
60 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
61 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
63 /* for_nexthops and change_nexthops only used when nexthop object
64 * is not set in a fib_info. The logic within can reference fib_nh.
66 #ifdef CONFIG_IP_ROUTE_MULTIPATH
68 #define for_nexthops(fi) { \
69 int nhsel; const struct fib_nh *nh; \
70 for (nhsel = 0, nh = (fi)->fib_nh; \
71 nhsel < fib_info_num_path((fi)); \
74 #define change_nexthops(fi) { \
75 int nhsel; struct fib_nh *nexthop_nh; \
76 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
77 nhsel < fib_info_num_path((fi)); \
78 nexthop_nh++, nhsel++)
80 #else /* CONFIG_IP_ROUTE_MULTIPATH */
82 /* Hope, that gcc will optimize it to get rid of dummy loop */
84 #define for_nexthops(fi) { \
85 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
86 for (nhsel = 0; nhsel < 1; nhsel++)
88 #define change_nexthops(fi) { \
90 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
91 for (nhsel = 0; nhsel < 1; nhsel++)
93 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
95 #define endfor_nexthops(fi) }
98 const struct fib_prop fib_props[RTN_MAX + 1] = {
101 .scope = RT_SCOPE_NOWHERE,
105 .scope = RT_SCOPE_UNIVERSE,
109 .scope = RT_SCOPE_HOST,
113 .scope = RT_SCOPE_LINK,
117 .scope = RT_SCOPE_LINK,
121 .scope = RT_SCOPE_UNIVERSE,
125 .scope = RT_SCOPE_UNIVERSE,
127 [RTN_UNREACHABLE] = {
128 .error = -EHOSTUNREACH,
129 .scope = RT_SCOPE_UNIVERSE,
133 .scope = RT_SCOPE_UNIVERSE,
137 .scope = RT_SCOPE_UNIVERSE,
141 .scope = RT_SCOPE_NOWHERE,
145 .scope = RT_SCOPE_NOWHERE,
149 static void rt_fibinfo_free(struct rtable __rcu **rtp)
151 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
156 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
157 * because we waited an RCU grace period before calling
158 * free_fib_info_rcu()
161 dst_dev_put(&rt->dst);
162 dst_release_immediate(&rt->dst);
165 static void free_nh_exceptions(struct fib_nh_common *nhc)
167 struct fnhe_hash_bucket *hash;
170 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
173 for (i = 0; i < FNHE_HASH_SIZE; i++) {
174 struct fib_nh_exception *fnhe;
176 fnhe = rcu_dereference_protected(hash[i].chain, 1);
178 struct fib_nh_exception *next;
180 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
182 rt_fibinfo_free(&fnhe->fnhe_rth_input);
183 rt_fibinfo_free(&fnhe->fnhe_rth_output);
193 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
200 for_each_possible_cpu(cpu) {
203 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
205 dst_dev_put(&rt->dst);
206 dst_release_immediate(&rt->dst);
212 void fib_nh_common_release(struct fib_nh_common *nhc)
214 netdev_put(nhc->nhc_dev, &nhc->nhc_dev_tracker);
215 lwtstate_put(nhc->nhc_lwtstate);
216 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
217 rt_fibinfo_free(&nhc->nhc_rth_input);
218 free_nh_exceptions(nhc);
220 EXPORT_SYMBOL_GPL(fib_nh_common_release);
222 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
224 #ifdef CONFIG_IP_ROUTE_CLASSID
225 if (fib_nh->nh_tclassid)
226 atomic_dec(&net->ipv4.fib_num_tclassid_users);
228 fib_nh_common_release(&fib_nh->nh_common);
231 /* Release a nexthop info record */
232 static void free_fib_info_rcu(struct rcu_head *head)
234 struct fib_info *fi = container_of(head, struct fib_info, rcu);
239 change_nexthops(fi) {
240 fib_nh_release(fi->fib_net, nexthop_nh);
241 } endfor_nexthops(fi);
244 ip_fib_metrics_put(fi->fib_metrics);
249 void free_fib_info(struct fib_info *fi)
251 if (fi->fib_dead == 0) {
252 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 && refcount_dec_and_test(&fi->fib_treeref)) {
264 hlist_del(&fi->fib_hash);
266 /* Paired with READ_ONCE() in fib_create_info(). */
267 WRITE_ONCE(fib_info_cnt, fib_info_cnt - 1);
270 hlist_del(&fi->fib_lhash);
272 list_del(&fi->nh_list);
274 change_nexthops(fi) {
275 if (!nexthop_nh->fib_nh_dev)
277 hlist_del(&nexthop_nh->nh_hash);
278 } endfor_nexthops(fi)
283 spin_unlock_bh(&fib_info_lock);
286 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
288 const struct fib_nh *onh;
290 if (fi->nh || ofi->nh)
291 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
293 if (ofi->fib_nhs == 0)
297 onh = fib_info_nh(ofi, nhsel);
299 if (nh->fib_nh_oif != onh->fib_nh_oif ||
300 nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
301 nh->fib_nh_scope != onh->fib_nh_scope ||
302 #ifdef CONFIG_IP_ROUTE_MULTIPATH
303 nh->fib_nh_weight != onh->fib_nh_weight ||
305 #ifdef CONFIG_IP_ROUTE_CLASSID
306 nh->nh_tclassid != onh->nh_tclassid ||
308 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
309 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
312 if (nh->fib_nh_gw_family == AF_INET &&
313 nh->fib_nh_gw4 != onh->fib_nh_gw4)
316 if (nh->fib_nh_gw_family == AF_INET6 &&
317 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
319 } endfor_nexthops(fi);
323 static inline unsigned int fib_devindex_hashfn(unsigned int val)
325 return hash_32(val, DEVINDEX_HASHBITS);
328 static struct hlist_head *
329 fib_info_devhash_bucket(const struct net_device *dev)
331 u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
333 return &fib_info_devhash[fib_devindex_hashfn(val)];
336 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
337 u32 prefsrc, u32 priority)
339 unsigned int val = init_val;
341 val ^= (protocol << 8) | scope;
348 static unsigned int fib_info_hashfn_result(unsigned int val)
350 unsigned int mask = (fib_info_hash_size - 1);
352 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
355 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
359 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
360 fi->fib_scope, (__force u32)fi->fib_prefsrc,
364 val ^= fib_devindex_hashfn(fi->nh->id);
367 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
368 } endfor_nexthops(fi)
371 return fib_info_hashfn_result(val);
374 /* no metrics, only nexthop id */
375 static struct fib_info *fib_find_info_nh(struct net *net,
376 const struct fib_config *cfg)
378 struct hlist_head *head;
382 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
383 cfg->fc_protocol, cfg->fc_scope,
384 (__force u32)cfg->fc_prefsrc,
386 hash = fib_info_hashfn_result(hash);
387 head = &fib_info_hash[hash];
389 hlist_for_each_entry(fi, head, fib_hash) {
390 if (!net_eq(fi->fib_net, net))
392 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
394 if (cfg->fc_protocol == fi->fib_protocol &&
395 cfg->fc_scope == fi->fib_scope &&
396 cfg->fc_prefsrc == fi->fib_prefsrc &&
397 cfg->fc_priority == fi->fib_priority &&
398 cfg->fc_type == fi->fib_type &&
399 cfg->fc_table == fi->fib_tb_id &&
400 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
407 static struct fib_info *fib_find_info(struct fib_info *nfi)
409 struct hlist_head *head;
413 hash = fib_info_hashfn(nfi);
414 head = &fib_info_hash[hash];
416 hlist_for_each_entry(fi, head, fib_hash) {
417 if (!net_eq(fi->fib_net, nfi->fib_net))
419 if (fi->fib_nhs != nfi->fib_nhs)
421 if (nfi->fib_protocol == fi->fib_protocol &&
422 nfi->fib_scope == fi->fib_scope &&
423 nfi->fib_prefsrc == fi->fib_prefsrc &&
424 nfi->fib_priority == fi->fib_priority &&
425 nfi->fib_type == fi->fib_type &&
426 memcmp(nfi->fib_metrics, fi->fib_metrics,
427 sizeof(u32) * RTAX_MAX) == 0 &&
428 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
429 nh_comp(fi, nfi) == 0)
436 /* Check, that the gateway is already configured.
437 * Used only by redirect accept routine.
439 int ip_fib_check_default(__be32 gw, struct net_device *dev)
441 struct hlist_head *head;
444 spin_lock(&fib_info_lock);
446 head = fib_info_devhash_bucket(dev);
448 hlist_for_each_entry(nh, head, nh_hash) {
449 if (nh->fib_nh_dev == dev &&
450 nh->fib_nh_gw4 == gw &&
451 !(nh->fib_nh_flags & RTNH_F_DEAD)) {
452 spin_unlock(&fib_info_lock);
457 spin_unlock(&fib_info_lock);
462 size_t fib_nlmsg_size(struct fib_info *fi)
464 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
465 + nla_total_size(4) /* RTA_TABLE */
466 + nla_total_size(4) /* RTA_DST */
467 + nla_total_size(4) /* RTA_PRIORITY */
468 + nla_total_size(4) /* RTA_PREFSRC */
469 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
470 unsigned int nhs = fib_info_num_path(fi);
472 /* space for nested metrics */
473 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
476 payload += nla_total_size(4); /* RTA_NH_ID */
479 size_t nh_encapsize = 0;
480 /* Also handles the special case nhs == 1 */
482 /* each nexthop is packed in an attribute */
483 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
486 /* may contain flow and gateway attribute */
487 nhsize += 2 * nla_total_size(4);
489 /* grab encap info */
490 for (i = 0; i < fib_info_num_path(fi); i++) {
491 struct fib_nh_common *nhc = fib_info_nhc(fi, i);
493 if (nhc->nhc_lwtstate) {
495 nh_encapsize += lwtunnel_get_encap_size(
498 nh_encapsize += nla_total_size(2);
502 /* all nexthops are packed in a nested attribute */
503 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
510 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
511 int dst_len, u32 tb_id, const struct nl_info *info,
512 unsigned int nlm_flags)
514 struct fib_rt_info fri;
516 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
519 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
523 fri.fi = fa->fa_info;
526 fri.dst_len = dst_len;
527 fri.dscp = fa->fa_dscp;
528 fri.type = fa->fa_type;
529 fri.offload = READ_ONCE(fa->offload);
530 fri.trap = READ_ONCE(fa->trap);
531 fri.offload_failed = READ_ONCE(fa->offload_failed);
532 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
534 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
535 WARN_ON(err == -EMSGSIZE);
539 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
540 info->nlh, GFP_KERNEL);
544 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
547 static int fib_detect_death(struct fib_info *fi, int order,
548 struct fib_info **last_resort, int *last_idx,
551 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
553 int state = NUD_NONE;
555 if (likely(nhc->nhc_gw_family == AF_INET))
556 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
557 else if (nhc->nhc_gw_family == AF_INET6)
558 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
564 state = n->nud_state;
569 if (state == NUD_REACHABLE)
571 if ((state & NUD_VALID) && order != dflt)
573 if ((state & NUD_VALID) ||
574 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
581 int fib_nh_common_init(struct net *net, struct fib_nh_common *nhc,
582 struct nlattr *encap, u16 encap_type,
583 void *cfg, gfp_t gfp_flags,
584 struct netlink_ext_ack *extack)
588 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
590 if (!nhc->nhc_pcpu_rth_output)
594 struct lwtunnel_state *lwtstate;
596 if (encap_type == LWTUNNEL_ENCAP_NONE) {
597 NL_SET_ERR_MSG(extack, "LWT encap type not specified");
601 err = lwtunnel_build_state(net, encap_type, encap,
602 nhc->nhc_family, cfg, &lwtstate,
607 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
613 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
614 nhc->nhc_pcpu_rth_output = NULL;
617 EXPORT_SYMBOL_GPL(fib_nh_common_init);
619 int fib_nh_init(struct net *net, struct fib_nh *nh,
620 struct fib_config *cfg, int nh_weight,
621 struct netlink_ext_ack *extack)
625 nh->fib_nh_family = AF_INET;
627 err = fib_nh_common_init(net, &nh->nh_common, cfg->fc_encap,
628 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
632 nh->fib_nh_oif = cfg->fc_oif;
633 nh->fib_nh_gw_family = cfg->fc_gw_family;
634 if (cfg->fc_gw_family == AF_INET)
635 nh->fib_nh_gw4 = cfg->fc_gw4;
636 else if (cfg->fc_gw_family == AF_INET6)
637 nh->fib_nh_gw6 = cfg->fc_gw6;
639 nh->fib_nh_flags = cfg->fc_flags;
641 #ifdef CONFIG_IP_ROUTE_CLASSID
642 nh->nh_tclassid = cfg->fc_flow;
644 atomic_inc(&net->ipv4.fib_num_tclassid_users);
646 #ifdef CONFIG_IP_ROUTE_MULTIPATH
647 nh->fib_nh_weight = nh_weight;
652 #ifdef CONFIG_IP_ROUTE_MULTIPATH
654 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
655 struct netlink_ext_ack *extack)
659 while (rtnh_ok(rtnh, remaining)) {
661 rtnh = rtnh_next(rtnh, &remaining);
664 /* leftover implies invalid nexthop configuration, discard it */
666 NL_SET_ERR_MSG(extack,
667 "Invalid nexthop configuration - extra data after nexthops");
674 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
675 struct netlink_ext_ack *extack)
677 if (nla_len(nla) < sizeof(*gw)) {
678 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
682 *gw = nla_get_in_addr(nla);
687 /* only called when fib_nh is integrated into fib_info */
688 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
689 int remaining, struct fib_config *cfg,
690 struct netlink_ext_ack *extack)
692 struct net *net = fi->fib_net;
693 struct fib_config fib_cfg;
697 change_nexthops(fi) {
700 memset(&fib_cfg, 0, sizeof(fib_cfg));
702 if (!rtnh_ok(rtnh, remaining)) {
703 NL_SET_ERR_MSG(extack,
704 "Invalid nexthop configuration - extra data after nexthop");
708 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
709 NL_SET_ERR_MSG(extack,
710 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
714 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
715 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
717 attrlen = rtnh_attrlen(rtnh);
719 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
721 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
722 nlav = nla_find(attrs, attrlen, RTA_VIA);
724 NL_SET_ERR_MSG(extack,
725 "Nexthop configuration can not contain both GATEWAY and VIA");
729 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
735 fib_cfg.fc_gw_family = AF_INET;
737 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
742 nla = nla_find(attrs, attrlen, RTA_FLOW);
744 if (nla_len(nla) < sizeof(u32)) {
745 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
748 fib_cfg.fc_flow = nla_get_u32(nla);
751 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
752 /* RTA_ENCAP_TYPE length checked in
753 * lwtunnel_valid_encap_type_attr
755 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
757 fib_cfg.fc_encap_type = nla_get_u16(nla);
760 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
761 rtnh->rtnh_hops + 1, extack);
765 rtnh = rtnh_next(rtnh, &remaining);
766 } endfor_nexthops(fi);
769 nh = fib_info_nh(fi, 0);
770 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
771 NL_SET_ERR_MSG(extack,
772 "Nexthop device index does not match RTA_OIF");
775 if (cfg->fc_gw_family) {
776 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
777 (cfg->fc_gw_family == AF_INET &&
778 nh->fib_nh_gw4 != cfg->fc_gw4) ||
779 (cfg->fc_gw_family == AF_INET6 &&
780 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
781 NL_SET_ERR_MSG(extack,
782 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
786 #ifdef CONFIG_IP_ROUTE_CLASSID
787 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
788 NL_SET_ERR_MSG(extack,
789 "Nexthop class id does not match RTA_FLOW");
798 /* only called when fib_nh is integrated into fib_info */
799 static void fib_rebalance(struct fib_info *fi)
804 if (fib_info_num_path(fi) < 2)
809 if (nh->fib_nh_flags & RTNH_F_DEAD)
812 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
813 nh->fib_nh_flags & RTNH_F_LINKDOWN)
816 total += nh->fib_nh_weight;
817 } endfor_nexthops(fi);
820 change_nexthops(fi) {
823 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
825 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
826 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
829 w += nexthop_nh->fib_nh_weight;
830 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
834 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
835 } endfor_nexthops(fi);
837 #else /* CONFIG_IP_ROUTE_MULTIPATH */
839 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
840 int remaining, struct fib_config *cfg,
841 struct netlink_ext_ack *extack)
843 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
848 #define fib_rebalance(fi) do { } while (0)
850 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
852 static int fib_encap_match(struct net *net, u16 encap_type,
853 struct nlattr *encap,
854 const struct fib_nh *nh,
855 const struct fib_config *cfg,
856 struct netlink_ext_ack *extack)
858 struct lwtunnel_state *lwtstate;
861 if (encap_type == LWTUNNEL_ENCAP_NONE)
864 ret = lwtunnel_build_state(net, encap_type, encap, AF_INET,
865 cfg, &lwtstate, extack);
867 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
868 lwtstate_free(lwtstate);
874 int fib_nh_match(struct net *net, struct fib_config *cfg, struct fib_info *fi,
875 struct netlink_ext_ack *extack)
877 #ifdef CONFIG_IP_ROUTE_MULTIPATH
878 struct rtnexthop *rtnh;
882 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
886 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
891 if (cfg->fc_oif || cfg->fc_gw_family) {
894 /* cannot match on nexthop object attributes */
898 nh = fib_info_nh(fi, 0);
900 if (fib_encap_match(net, cfg->fc_encap_type,
901 cfg->fc_encap, nh, cfg, extack))
904 #ifdef CONFIG_IP_ROUTE_CLASSID
906 cfg->fc_flow != nh->nh_tclassid)
909 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
910 (cfg->fc_gw_family &&
911 cfg->fc_gw_family != nh->fib_nh_gw_family))
914 if (cfg->fc_gw_family == AF_INET &&
915 cfg->fc_gw4 != nh->fib_nh_gw4)
918 if (cfg->fc_gw_family == AF_INET6 &&
919 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
925 #ifdef CONFIG_IP_ROUTE_MULTIPATH
930 remaining = cfg->fc_mp_len;
935 if (!rtnh_ok(rtnh, remaining))
938 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
941 attrlen = rtnh_attrlen(rtnh);
943 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
946 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
947 nlav = nla_find(attrs, attrlen, RTA_VIA);
949 NL_SET_ERR_MSG(extack,
950 "Nexthop configuration can not contain both GATEWAY and VIA");
957 err = fib_gw_from_attr(&gw, nla, extack);
961 if (nh->fib_nh_gw_family != AF_INET ||
962 gw != nh->fib_nh_gw4)
965 struct fib_config cfg2;
967 err = fib_gw_from_via(&cfg2, nlav, extack);
971 switch (nh->fib_nh_gw_family) {
973 if (cfg2.fc_gw_family != AF_INET ||
974 cfg2.fc_gw4 != nh->fib_nh_gw4)
978 if (cfg2.fc_gw_family != AF_INET6 ||
979 ipv6_addr_cmp(&cfg2.fc_gw6,
986 #ifdef CONFIG_IP_ROUTE_CLASSID
987 nla = nla_find(attrs, attrlen, RTA_FLOW);
989 if (nla_len(nla) < sizeof(u32)) {
990 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
993 if (nla_get_u32(nla) != nh->nh_tclassid)
999 rtnh = rtnh_next(rtnh, &remaining);
1000 } endfor_nexthops(fi);
1005 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1013 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1014 int type = nla_type(nla);
1019 if (type > RTAX_MAX)
1022 if (type == RTAX_CC_ALGO) {
1023 char tmp[TCP_CA_NAME_MAX];
1024 bool ecn_ca = false;
1026 nla_strscpy(tmp, nla, sizeof(tmp));
1027 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1029 if (nla_len(nla) != sizeof(u32))
1031 val = nla_get_u32(nla);
1034 fi_val = fi->fib_metrics->metrics[type - 1];
1035 if (type == RTAX_FEATURES)
1036 fi_val &= ~DST_FEATURE_ECN_CA;
1045 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1046 u32 table, struct netlink_ext_ack *extack)
1048 struct fib6_config cfg = {
1050 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1051 .fc_ifindex = nh->fib_nh_oif,
1052 .fc_gateway = nh->fib_nh_gw6,
1054 struct fib6_nh fib6_nh = {};
1057 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1059 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1060 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1062 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1063 nh->fib_nh_scope = RT_SCOPE_LINK;
1065 ipv6_stub->fib6_nh_release(&fib6_nh);
1075 * Semantics of nexthop is very messy by historical reasons.
1076 * We have to take into account, that:
1077 * a) gateway can be actually local interface address,
1078 * so that gatewayed route is direct.
1079 * b) gateway must be on-link address, possibly
1080 * described not by an ifaddr, but also by a direct route.
1081 * c) If both gateway and interface are specified, they should not
1083 * d) If we use tunnel routes, gateway could be not on-link.
1085 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1086 * results in pretty ugly and hairy code with obscure logic.
1088 * I chose to generalized it instead, so that the size
1089 * of code does not increase practically, but it becomes
1090 * much more general.
1091 * Every prefix is assigned a "scope" value: "host" is local address,
1092 * "link" is direct route,
1093 * [ ... "site" ... "interior" ... ]
1094 * and "universe" is true gateway route with global meaning.
1096 * Every prefix refers to a set of "nexthop"s (gw, oif),
1097 * where gw must have narrower scope. This recursion stops
1098 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1099 * which means that gw is forced to be on link.
1101 * Code is still hairy, but now it is apparently logically
1102 * consistent and very flexible. F.e. as by-product it allows
1103 * to co-exists in peace independent exterior and interior
1104 * routing processes.
1106 * Normally it looks as following.
1108 * {universe prefix} -> (gw, oif) [scope link]
1110 * |-> {link prefix} -> (gw, oif) [scope local]
1112 * |-> {local prefix} (terminal node)
1114 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1115 u8 scope, struct netlink_ext_ack *extack)
1117 struct net_device *dev;
1118 struct fib_result res;
1121 if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1122 unsigned int addr_type;
1124 if (scope >= RT_SCOPE_LINK) {
1125 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1128 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1130 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1133 if (!(dev->flags & IFF_UP)) {
1134 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1137 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1138 if (addr_type != RTN_UNICAST) {
1139 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1142 if (!netif_carrier_ok(dev))
1143 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1144 nh->fib_nh_dev = dev;
1145 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1146 nh->fib_nh_scope = RT_SCOPE_LINK;
1151 struct fib_table *tbl = NULL;
1152 struct flowi4 fl4 = {
1153 .daddr = nh->fib_nh_gw4,
1154 .flowi4_scope = scope + 1,
1155 .flowi4_oif = nh->fib_nh_oif,
1156 .flowi4_iif = LOOPBACK_IFINDEX,
1159 /* It is not necessary, but requires a bit of thinking */
1160 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1161 fl4.flowi4_scope = RT_SCOPE_LINK;
1163 if (table && table != RT_TABLE_MAIN)
1164 tbl = fib_get_table(net, table);
1167 err = fib_table_lookup(tbl, &fl4, &res,
1168 FIB_LOOKUP_IGNORE_LINKSTATE |
1171 /* on error or if no table given do full lookup. This
1172 * is needed for example when nexthops are in the local
1173 * table rather than the given table
1176 err = fib_lookup(net, &fl4, &res,
1177 FIB_LOOKUP_IGNORE_LINKSTATE);
1181 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1187 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1188 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1191 nh->fib_nh_scope = res.scope;
1192 nh->fib_nh_oif = FIB_RES_OIF(res);
1193 nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1195 NL_SET_ERR_MSG(extack,
1196 "No egress device for nexthop gateway");
1199 netdev_hold(dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1200 if (!netif_carrier_ok(dev))
1201 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1202 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1208 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1209 struct netlink_ext_ack *extack)
1211 struct in_device *in_dev;
1214 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1215 NL_SET_ERR_MSG(extack,
1216 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1223 in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1227 if (!(in_dev->dev->flags & IFF_UP)) {
1228 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1232 nh->fib_nh_dev = in_dev->dev;
1233 netdev_hold(nh->fib_nh_dev, &nh->fib_nh_dev_tracker, GFP_ATOMIC);
1234 nh->fib_nh_scope = RT_SCOPE_HOST;
1235 if (!netif_carrier_ok(nh->fib_nh_dev))
1236 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1243 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1244 struct netlink_ext_ack *extack)
1248 if (nh->fib_nh_gw_family == AF_INET)
1249 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1250 else if (nh->fib_nh_gw_family == AF_INET6)
1251 err = fib_check_nh_v6_gw(net, nh, table, extack);
1253 err = fib_check_nh_nongw(net, nh, extack);
1258 static struct hlist_head *
1259 fib_info_laddrhash_bucket(const struct net *net, __be32 val)
1261 u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
1262 fib_info_hash_bits);
1264 return &fib_info_laddrhash[slot];
1267 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1268 struct hlist_head *new_laddrhash,
1269 unsigned int new_size)
1271 struct hlist_head *old_info_hash, *old_laddrhash;
1272 unsigned int old_size = fib_info_hash_size;
1275 spin_lock_bh(&fib_info_lock);
1276 old_info_hash = fib_info_hash;
1277 old_laddrhash = fib_info_laddrhash;
1278 fib_info_hash_size = new_size;
1279 fib_info_hash_bits = ilog2(new_size);
1281 for (i = 0; i < old_size; i++) {
1282 struct hlist_head *head = &fib_info_hash[i];
1283 struct hlist_node *n;
1284 struct fib_info *fi;
1286 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1287 struct hlist_head *dest;
1288 unsigned int new_hash;
1290 new_hash = fib_info_hashfn(fi);
1291 dest = &new_info_hash[new_hash];
1292 hlist_add_head(&fi->fib_hash, dest);
1295 fib_info_hash = new_info_hash;
1297 fib_info_laddrhash = new_laddrhash;
1298 for (i = 0; i < old_size; i++) {
1299 struct hlist_head *lhead = &old_laddrhash[i];
1300 struct hlist_node *n;
1301 struct fib_info *fi;
1303 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1304 struct hlist_head *ldest;
1306 ldest = fib_info_laddrhash_bucket(fi->fib_net,
1308 hlist_add_head(&fi->fib_lhash, ldest);
1312 spin_unlock_bh(&fib_info_lock);
1314 kvfree(old_info_hash);
1315 kvfree(old_laddrhash);
1318 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1319 unsigned char scope)
1323 if (nhc->nhc_family != AF_INET)
1324 return inet_select_addr(nhc->nhc_dev, 0, scope);
1326 nh = container_of(nhc, struct fib_nh, nh_common);
1327 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1328 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1330 return nh->nh_saddr;
1333 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1335 struct fib_nh_common *nhc = res->nhc;
1337 if (res->fi->fib_prefsrc)
1338 return res->fi->fib_prefsrc;
1340 if (nhc->nhc_family == AF_INET) {
1343 nh = container_of(nhc, struct fib_nh, nh_common);
1344 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1345 return nh->nh_saddr;
1348 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1351 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1353 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1354 fib_prefsrc != cfg->fc_dst) {
1355 u32 tb_id = cfg->fc_table;
1358 if (tb_id == RT_TABLE_MAIN)
1359 tb_id = RT_TABLE_LOCAL;
1361 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1362 fib_prefsrc, tb_id);
1364 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1365 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1366 fib_prefsrc, RT_TABLE_LOCAL);
1369 if (rc != RTN_LOCAL)
1375 struct fib_info *fib_create_info(struct fib_config *cfg,
1376 struct netlink_ext_ack *extack)
1379 struct fib_info *fi = NULL;
1380 struct nexthop *nh = NULL;
1381 struct fib_info *ofi;
1383 struct net *net = cfg->fc_nlinfo.nl_net;
1385 if (cfg->fc_type > RTN_MAX)
1388 /* Fast check to catch the most weird cases */
1389 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1390 NL_SET_ERR_MSG(extack, "Invalid scope");
1394 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1395 NL_SET_ERR_MSG(extack,
1396 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1400 if (cfg->fc_nh_id) {
1402 fi = fib_find_info_nh(net, cfg);
1404 refcount_inc(&fi->fib_treeref);
1409 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1411 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1417 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1419 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1427 /* Paired with WRITE_ONCE() in fib_release_info() */
1428 if (READ_ONCE(fib_info_cnt) >= fib_info_hash_size) {
1429 unsigned int new_size = fib_info_hash_size << 1;
1430 struct hlist_head *new_info_hash;
1431 struct hlist_head *new_laddrhash;
1436 bytes = (size_t)new_size * sizeof(struct hlist_head *);
1437 new_info_hash = kvzalloc(bytes, GFP_KERNEL);
1438 new_laddrhash = kvzalloc(bytes, GFP_KERNEL);
1439 if (!new_info_hash || !new_laddrhash) {
1440 kvfree(new_info_hash);
1441 kvfree(new_laddrhash);
1443 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1445 if (!fib_info_hash_size)
1449 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1452 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1453 cfg->fc_mx_len, extack);
1454 if (IS_ERR(fi->fib_metrics)) {
1455 err = PTR_ERR(fi->fib_metrics);
1457 return ERR_PTR(err);
1461 fi->fib_protocol = cfg->fc_protocol;
1462 fi->fib_scope = cfg->fc_scope;
1463 fi->fib_flags = cfg->fc_flags;
1464 fi->fib_priority = cfg->fc_priority;
1465 fi->fib_prefsrc = cfg->fc_prefsrc;
1466 fi->fib_type = cfg->fc_type;
1467 fi->fib_tb_id = cfg->fc_table;
1471 if (!nexthop_get(nh)) {
1472 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1479 change_nexthops(fi) {
1480 nexthop_nh->nh_parent = fi;
1481 } endfor_nexthops(fi)
1484 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1487 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1493 if (fib_props[cfg->fc_type].error) {
1494 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1495 NL_SET_ERR_MSG(extack,
1496 "Gateway, device and multipath can not be specified for this route type");
1501 switch (cfg->fc_type) {
1509 NL_SET_ERR_MSG(extack, "Invalid route type");
1514 if (cfg->fc_scope > RT_SCOPE_HOST) {
1515 NL_SET_ERR_MSG(extack, "Invalid scope");
1520 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1523 } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1524 struct fib_nh *nh = fi->fib_nh;
1526 /* Local address is added. */
1528 NL_SET_ERR_MSG(extack,
1529 "Route with host scope can not have multiple nexthops");
1532 if (nh->fib_nh_gw_family) {
1533 NL_SET_ERR_MSG(extack,
1534 "Route with host scope can not have a gateway");
1537 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1538 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1540 if (!nh->fib_nh_dev)
1542 netdev_tracker_alloc(nh->fib_nh_dev, &nh->fib_nh_dev_tracker,
1547 change_nexthops(fi) {
1548 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1549 cfg->fc_table, cfg->fc_scope,
1553 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1555 } endfor_nexthops(fi)
1556 if (linkdown == fi->fib_nhs)
1557 fi->fib_flags |= RTNH_F_LINKDOWN;
1560 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1561 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1566 change_nexthops(fi) {
1567 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1569 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1570 fi->fib_nh_is_v6 = true;
1571 } endfor_nexthops(fi)
1577 ofi = fib_find_info(fi);
1581 refcount_inc(&ofi->fib_treeref);
1585 refcount_set(&fi->fib_treeref, 1);
1586 refcount_set(&fi->fib_clntref, 1);
1587 spin_lock_bh(&fib_info_lock);
1589 hlist_add_head(&fi->fib_hash,
1590 &fib_info_hash[fib_info_hashfn(fi)]);
1591 if (fi->fib_prefsrc) {
1592 struct hlist_head *head;
1594 head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
1595 hlist_add_head(&fi->fib_lhash, head);
1598 list_add(&fi->nh_list, &nh->fi_list);
1600 change_nexthops(fi) {
1601 struct hlist_head *head;
1603 if (!nexthop_nh->fib_nh_dev)
1605 head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1606 hlist_add_head(&nexthop_nh->nh_hash, head);
1607 } endfor_nexthops(fi)
1609 spin_unlock_bh(&fib_info_lock);
1621 return ERR_PTR(err);
1624 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1625 u8 rt_family, unsigned char *flags, bool skip_oif)
1627 if (nhc->nhc_flags & RTNH_F_DEAD)
1628 *flags |= RTNH_F_DEAD;
1630 if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1631 *flags |= RTNH_F_LINKDOWN;
1634 switch (nhc->nhc_family) {
1636 if (ip_ignore_linkdown(nhc->nhc_dev))
1637 *flags |= RTNH_F_DEAD;
1640 if (ip6_ignore_linkdown(nhc->nhc_dev))
1641 *flags |= RTNH_F_DEAD;
1647 switch (nhc->nhc_gw_family) {
1649 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1650 goto nla_put_failure;
1653 /* if gateway family does not match nexthop family
1654 * gateway is encoded as RTA_VIA
1656 if (rt_family != nhc->nhc_gw_family) {
1657 int alen = sizeof(struct in6_addr);
1661 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1663 goto nla_put_failure;
1665 via = nla_data(nla);
1666 via->rtvia_family = AF_INET6;
1667 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1668 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1669 &nhc->nhc_gw.ipv6) < 0) {
1670 goto nla_put_failure;
1675 *flags |= (nhc->nhc_flags &
1676 (RTNH_F_ONLINK | RTNH_F_OFFLOAD | RTNH_F_TRAP));
1678 if (!skip_oif && nhc->nhc_dev &&
1679 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1680 goto nla_put_failure;
1682 if (nhc->nhc_lwtstate &&
1683 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1684 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1685 goto nla_put_failure;
1692 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1694 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1695 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1696 int nh_weight, u8 rt_family, u32 nh_tclassid)
1698 const struct net_device *dev = nhc->nhc_dev;
1699 struct rtnexthop *rtnh;
1700 unsigned char flags = 0;
1702 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1704 goto nla_put_failure;
1706 rtnh->rtnh_hops = nh_weight - 1;
1707 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1709 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1710 goto nla_put_failure;
1712 rtnh->rtnh_flags = flags;
1714 if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1715 goto nla_put_failure;
1717 /* length of rtnetlink header + attributes */
1718 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1725 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1728 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1729 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1733 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1735 goto nla_put_failure;
1737 if (unlikely(fi->nh)) {
1738 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1739 goto nla_put_failure;
1744 u32 nh_tclassid = 0;
1745 #ifdef CONFIG_IP_ROUTE_CLASSID
1746 nh_tclassid = nh->nh_tclassid;
1748 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1749 AF_INET, nh_tclassid) < 0)
1750 goto nla_put_failure;
1751 } endfor_nexthops(fi);
1754 nla_nest_end(skb, mp);
1762 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1768 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1769 const struct fib_rt_info *fri, unsigned int flags)
1771 unsigned int nhs = fib_info_num_path(fri->fi);
1772 struct fib_info *fi = fri->fi;
1773 u32 tb_id = fri->tb_id;
1774 struct nlmsghdr *nlh;
1777 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1781 rtm = nlmsg_data(nlh);
1782 rtm->rtm_family = AF_INET;
1783 rtm->rtm_dst_len = fri->dst_len;
1784 rtm->rtm_src_len = 0;
1785 rtm->rtm_tos = inet_dscp_to_dsfield(fri->dscp);
1787 rtm->rtm_table = tb_id;
1789 rtm->rtm_table = RT_TABLE_COMPAT;
1790 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1791 goto nla_put_failure;
1792 rtm->rtm_type = fri->type;
1793 rtm->rtm_flags = fi->fib_flags;
1794 rtm->rtm_scope = fi->fib_scope;
1795 rtm->rtm_protocol = fi->fib_protocol;
1797 if (rtm->rtm_dst_len &&
1798 nla_put_in_addr(skb, RTA_DST, fri->dst))
1799 goto nla_put_failure;
1800 if (fi->fib_priority &&
1801 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1802 goto nla_put_failure;
1803 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1804 goto nla_put_failure;
1806 if (fi->fib_prefsrc &&
1807 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1808 goto nla_put_failure;
1811 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1812 goto nla_put_failure;
1813 if (nexthop_is_blackhole(fi->nh))
1814 rtm->rtm_type = RTN_BLACKHOLE;
1815 if (!fi->fib_net->ipv4.sysctl_nexthop_compat_mode)
1820 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1821 unsigned char flags = 0;
1823 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1824 goto nla_put_failure;
1826 rtm->rtm_flags = flags;
1827 #ifdef CONFIG_IP_ROUTE_CLASSID
1828 if (nhc->nhc_family == AF_INET) {
1831 nh = container_of(nhc, struct fib_nh, nh_common);
1832 if (nh->nh_tclassid &&
1833 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1834 goto nla_put_failure;
1838 if (fib_add_multipath(skb, fi) < 0)
1839 goto nla_put_failure;
1844 rtm->rtm_flags |= RTM_F_OFFLOAD;
1846 rtm->rtm_flags |= RTM_F_TRAP;
1847 if (fri->offload_failed)
1848 rtm->rtm_flags |= RTM_F_OFFLOAD_FAILED;
1850 nlmsg_end(skb, nlh);
1854 nlmsg_cancel(skb, nlh);
1860 * - local address disappeared -> we must delete all the entries
1862 * - device went down -> we must shutdown all nexthops going via it.
1864 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1866 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1867 struct net *net = dev_net(dev);
1868 struct hlist_head *head;
1869 struct fib_info *fi;
1872 if (!fib_info_laddrhash || local == 0)
1875 head = fib_info_laddrhash_bucket(net, local);
1876 hlist_for_each_entry(fi, head, fib_lhash) {
1877 if (!net_eq(fi->fib_net, net) ||
1878 fi->fib_tb_id != tb_id)
1880 if (fi->fib_prefsrc == local) {
1881 fi->fib_flags |= RTNH_F_DEAD;
1888 static int call_fib_nh_notifiers(struct fib_nh *nh,
1889 enum fib_event_type event_type)
1891 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1892 struct fib_nh_notifier_info info = {
1896 switch (event_type) {
1897 case FIB_EVENT_NH_ADD:
1898 if (nh->fib_nh_flags & RTNH_F_DEAD)
1900 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1902 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1904 case FIB_EVENT_NH_DEL:
1905 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1906 (nh->fib_nh_flags & RTNH_F_DEAD))
1907 return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1908 event_type, &info.info);
1917 /* Update the PMTU of exceptions when:
1918 * - the new MTU of the first hop becomes smaller than the PMTU
1919 * - the old MTU was the same as the PMTU, and it limited discovery of
1920 * larger MTUs on the path. With that limit raised, we can now
1921 * discover larger MTUs
1922 * A special case is locked exceptions, for which the PMTU is smaller
1923 * than the minimal accepted PMTU:
1924 * - if the new MTU is greater than the PMTU, don't make any change
1925 * - otherwise, unlock and set PMTU
1927 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1929 struct fnhe_hash_bucket *bucket;
1932 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1936 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1937 struct fib_nh_exception *fnhe;
1939 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1941 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1942 if (fnhe->fnhe_mtu_locked) {
1943 if (new <= fnhe->fnhe_pmtu) {
1944 fnhe->fnhe_pmtu = new;
1945 fnhe->fnhe_mtu_locked = false;
1947 } else if (new < fnhe->fnhe_pmtu ||
1948 orig == fnhe->fnhe_pmtu) {
1949 fnhe->fnhe_pmtu = new;
1955 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1957 struct hlist_head *head = fib_info_devhash_bucket(dev);
1960 hlist_for_each_entry(nh, head, nh_hash) {
1961 if (nh->fib_nh_dev == dev)
1962 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1966 /* Event force Flags Description
1967 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1968 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1969 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1970 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1972 * only used when fib_nh is built into fib_info
1974 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1976 struct hlist_head *head = fib_info_devhash_bucket(dev);
1977 struct fib_info *prev_fi = NULL;
1978 int scope = RT_SCOPE_NOWHERE;
1985 hlist_for_each_entry(nh, head, nh_hash) {
1986 struct fib_info *fi = nh->nh_parent;
1989 BUG_ON(!fi->fib_nhs);
1990 if (nh->fib_nh_dev != dev || fi == prev_fi)
1994 change_nexthops(fi) {
1995 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1997 else if (nexthop_nh->fib_nh_dev == dev &&
1998 nexthop_nh->fib_nh_scope != scope) {
2001 case NETDEV_UNREGISTER:
2002 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2005 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2008 call_fib_nh_notifiers(nexthop_nh,
2012 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2013 if (event == NETDEV_UNREGISTER &&
2014 nexthop_nh->fib_nh_dev == dev) {
2019 } endfor_nexthops(fi)
2020 if (dead == fi->fib_nhs) {
2023 case NETDEV_UNREGISTER:
2024 fi->fib_flags |= RTNH_F_DEAD;
2027 fi->fib_flags |= RTNH_F_LINKDOWN;
2039 /* Must be invoked inside of an RCU protected region. */
2040 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2042 struct fib_info *fi = NULL, *last_resort = NULL;
2043 struct hlist_head *fa_head = res->fa_head;
2044 struct fib_table *tb = res->table;
2045 u8 slen = 32 - res->prefixlen;
2046 int order = -1, last_idx = -1;
2047 struct fib_alias *fa, *fa1 = NULL;
2048 u32 last_prio = res->fi->fib_priority;
2049 dscp_t last_dscp = 0;
2051 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2052 struct fib_info *next_fi = fa->fa_info;
2053 struct fib_nh_common *nhc;
2055 if (fa->fa_slen != slen)
2058 fa->fa_dscp != inet_dsfield_to_dscp(flp->flowi4_tos))
2060 if (fa->tb_id != tb->tb_id)
2062 if (next_fi->fib_priority > last_prio &&
2063 fa->fa_dscp == last_dscp) {
2068 if (next_fi->fib_flags & RTNH_F_DEAD)
2070 last_dscp = fa->fa_dscp;
2071 last_prio = next_fi->fib_priority;
2073 if (next_fi->fib_scope != res->scope ||
2074 fa->fa_type != RTN_UNICAST)
2077 nhc = fib_info_nhc(next_fi, 0);
2078 if (!nhc->nhc_gw_family || nhc->nhc_scope != RT_SCOPE_LINK)
2081 fib_alias_accessed(fa);
2084 if (next_fi != res->fi)
2087 } else if (!fib_detect_death(fi, order, &last_resort,
2088 &last_idx, fa1->fa_default)) {
2089 fib_result_assign(res, fi);
2090 fa1->fa_default = order;
2097 if (order <= 0 || !fi) {
2099 fa1->fa_default = -1;
2103 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2105 fib_result_assign(res, fi);
2106 fa1->fa_default = order;
2111 fib_result_assign(res, last_resort);
2112 fa1->fa_default = last_idx;
2118 * Dead device goes up. We wake up dead nexthops.
2119 * It takes sense only on multipath routes.
2121 * only used when fib_nh is built into fib_info
2123 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2125 struct fib_info *prev_fi;
2126 struct hlist_head *head;
2130 if (!(dev->flags & IFF_UP))
2133 if (nh_flags & RTNH_F_DEAD) {
2134 unsigned int flags = dev_get_flags(dev);
2136 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2137 nh_flags |= RTNH_F_LINKDOWN;
2141 head = fib_info_devhash_bucket(dev);
2144 hlist_for_each_entry(nh, head, nh_hash) {
2145 struct fib_info *fi = nh->nh_parent;
2148 BUG_ON(!fi->fib_nhs);
2149 if (nh->fib_nh_dev != dev || fi == prev_fi)
2154 change_nexthops(fi) {
2155 if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2159 if (!nexthop_nh->fib_nh_dev ||
2160 !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2162 if (nexthop_nh->fib_nh_dev != dev ||
2163 !__in_dev_get_rtnl(dev))
2166 nexthop_nh->fib_nh_flags &= ~nh_flags;
2167 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2168 } endfor_nexthops(fi)
2171 fi->fib_flags &= ~nh_flags;
2181 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2182 static bool fib_good_nh(const struct fib_nh *nh)
2184 int state = NUD_REACHABLE;
2186 if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2187 struct neighbour *n;
2191 if (likely(nh->fib_nh_gw_family == AF_INET))
2192 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2193 (__force u32)nh->fib_nh_gw4);
2194 else if (nh->fib_nh_gw_family == AF_INET6)
2195 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2200 state = n->nud_state;
2202 rcu_read_unlock_bh();
2205 return !!(state & NUD_VALID);
2208 void fib_select_multipath(struct fib_result *res, int hash)
2210 struct fib_info *fi = res->fi;
2211 struct net *net = fi->fib_net;
2214 if (unlikely(res->fi->nh)) {
2215 nexthop_path_fib_result(res, hash);
2219 change_nexthops(fi) {
2220 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2221 if (!fib_good_nh(nexthop_nh))
2224 res->nh_sel = nhsel;
2225 res->nhc = &nexthop_nh->nh_common;
2230 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2233 res->nh_sel = nhsel;
2234 res->nhc = &nexthop_nh->nh_common;
2236 } endfor_nexthops(fi);
2240 void fib_select_path(struct net *net, struct fib_result *res,
2241 struct flowi4 *fl4, const struct sk_buff *skb)
2243 if (fl4->flowi4_oif)
2246 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2247 if (fib_info_num_path(res->fi) > 1) {
2248 int h = fib_multipath_hash(net, fl4, skb, NULL);
2250 fib_select_multipath(res, h);
2254 if (!res->prefixlen &&
2255 res->table->tb_num_default > 1 &&
2256 res->type == RTN_UNICAST)
2257 fib_select_default(fl4, res);
2261 fl4->saddr = fib_result_prefsrc(net, res);