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>
33 #include <linux/nospec.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_cnt;
58 #define DEVINDEX_HASHBITS 8
59 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
60 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
62 /* for_nexthops and change_nexthops only used when nexthop object
63 * is not set in a fib_info. The logic within can reference fib_nh.
65 #ifdef CONFIG_IP_ROUTE_MULTIPATH
67 #define for_nexthops(fi) { \
68 int nhsel; const struct fib_nh *nh; \
69 for (nhsel = 0, nh = (fi)->fib_nh; \
70 nhsel < fib_info_num_path((fi)); \
73 #define change_nexthops(fi) { \
74 int nhsel; struct fib_nh *nexthop_nh; \
75 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
76 nhsel < fib_info_num_path((fi)); \
77 nexthop_nh++, nhsel++)
79 #else /* CONFIG_IP_ROUTE_MULTIPATH */
81 /* Hope, that gcc will optimize it to get rid of dummy loop */
83 #define for_nexthops(fi) { \
84 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
85 for (nhsel = 0; nhsel < 1; nhsel++)
87 #define change_nexthops(fi) { \
89 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
90 for (nhsel = 0; nhsel < 1; nhsel++)
92 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
94 #define endfor_nexthops(fi) }
97 const struct fib_prop fib_props[RTN_MAX + 1] = {
100 .scope = RT_SCOPE_NOWHERE,
104 .scope = RT_SCOPE_UNIVERSE,
108 .scope = RT_SCOPE_HOST,
112 .scope = RT_SCOPE_LINK,
116 .scope = RT_SCOPE_LINK,
120 .scope = RT_SCOPE_UNIVERSE,
124 .scope = RT_SCOPE_UNIVERSE,
126 [RTN_UNREACHABLE] = {
127 .error = -EHOSTUNREACH,
128 .scope = RT_SCOPE_UNIVERSE,
132 .scope = RT_SCOPE_UNIVERSE,
136 .scope = RT_SCOPE_UNIVERSE,
140 .scope = RT_SCOPE_NOWHERE,
144 .scope = RT_SCOPE_NOWHERE,
148 static void rt_fibinfo_free(struct rtable __rcu **rtp)
150 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
155 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
156 * because we waited an RCU grace period before calling
157 * free_fib_info_rcu()
160 dst_dev_put(&rt->dst);
161 dst_release_immediate(&rt->dst);
164 static void free_nh_exceptions(struct fib_nh_common *nhc)
166 struct fnhe_hash_bucket *hash;
169 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
172 for (i = 0; i < FNHE_HASH_SIZE; i++) {
173 struct fib_nh_exception *fnhe;
175 fnhe = rcu_dereference_protected(hash[i].chain, 1);
177 struct fib_nh_exception *next;
179 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
181 rt_fibinfo_free(&fnhe->fnhe_rth_input);
182 rt_fibinfo_free(&fnhe->fnhe_rth_output);
192 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
199 for_each_possible_cpu(cpu) {
202 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
204 dst_dev_put(&rt->dst);
205 dst_release_immediate(&rt->dst);
211 void fib_nh_common_release(struct fib_nh_common *nhc)
214 dev_put(nhc->nhc_dev);
216 lwtstate_put(nhc->nhc_lwtstate);
217 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
218 rt_fibinfo_free(&nhc->nhc_rth_input);
219 free_nh_exceptions(nhc);
221 EXPORT_SYMBOL_GPL(fib_nh_common_release);
223 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
225 #ifdef CONFIG_IP_ROUTE_CLASSID
226 if (fib_nh->nh_tclassid)
227 atomic_dec(&net->ipv4.fib_num_tclassid_users);
229 fib_nh_common_release(&fib_nh->nh_common);
232 /* Release a nexthop info record */
233 static void free_fib_info_rcu(struct rcu_head *head)
235 struct fib_info *fi = container_of(head, struct fib_info, rcu);
240 change_nexthops(fi) {
241 fib_nh_release(fi->fib_net, nexthop_nh);
242 } endfor_nexthops(fi);
245 ip_fib_metrics_put(fi->fib_metrics);
250 void free_fib_info(struct fib_info *fi)
252 if (fi->fib_dead == 0) {
253 pr_warn("Freeing alive fib_info %p\n", fi);
258 call_rcu(&fi->rcu, free_fib_info_rcu);
260 EXPORT_SYMBOL_GPL(free_fib_info);
262 void fib_release_info(struct fib_info *fi)
264 spin_lock_bh(&fib_info_lock);
265 if (fi && --fi->fib_treeref == 0) {
266 hlist_del(&fi->fib_hash);
268 hlist_del(&fi->fib_lhash);
270 list_del(&fi->nh_list);
272 change_nexthops(fi) {
273 if (!nexthop_nh->fib_nh_dev)
275 hlist_del(&nexthop_nh->nh_hash);
276 } endfor_nexthops(fi)
278 /* Paired with READ_ONCE() from fib_table_lookup() */
279 WRITE_ONCE(fi->fib_dead, 1);
282 spin_unlock_bh(&fib_info_lock);
285 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
287 const struct fib_nh *onh;
289 if (fi->nh || ofi->nh)
290 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
292 if (ofi->fib_nhs == 0)
296 onh = fib_info_nh(ofi, nhsel);
298 if (nh->fib_nh_oif != onh->fib_nh_oif ||
299 nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
300 nh->fib_nh_scope != onh->fib_nh_scope ||
301 #ifdef CONFIG_IP_ROUTE_MULTIPATH
302 nh->fib_nh_weight != onh->fib_nh_weight ||
304 #ifdef CONFIG_IP_ROUTE_CLASSID
305 nh->nh_tclassid != onh->nh_tclassid ||
307 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
308 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
311 if (nh->fib_nh_gw_family == AF_INET &&
312 nh->fib_nh_gw4 != onh->fib_nh_gw4)
315 if (nh->fib_nh_gw_family == AF_INET6 &&
316 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
318 } endfor_nexthops(fi);
322 static inline unsigned int fib_devindex_hashfn(unsigned int val)
324 return hash_32(val, DEVINDEX_HASHBITS);
327 static struct hlist_head *
328 fib_info_devhash_bucket(const struct net_device *dev)
330 u32 val = net_hash_mix(dev_net(dev)) ^ dev->ifindex;
332 return &fib_info_devhash[fib_devindex_hashfn(val)];
335 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
336 u32 prefsrc, u32 priority)
338 unsigned int val = init_val;
340 val ^= (protocol << 8) | scope;
347 static unsigned int fib_info_hashfn_result(unsigned int val)
349 unsigned int mask = (fib_info_hash_size - 1);
351 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
354 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
358 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
359 fi->fib_scope, (__force u32)fi->fib_prefsrc,
363 val ^= fib_devindex_hashfn(fi->nh->id);
366 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
367 } endfor_nexthops(fi)
370 return fib_info_hashfn_result(val);
373 /* no metrics, only nexthop id */
374 static struct fib_info *fib_find_info_nh(struct net *net,
375 const struct fib_config *cfg)
377 struct hlist_head *head;
381 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
382 cfg->fc_protocol, cfg->fc_scope,
383 (__force u32)cfg->fc_prefsrc,
385 hash = fib_info_hashfn_result(hash);
386 head = &fib_info_hash[hash];
388 hlist_for_each_entry(fi, head, fib_hash) {
389 if (!net_eq(fi->fib_net, net))
391 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
393 if (cfg->fc_protocol == fi->fib_protocol &&
394 cfg->fc_scope == fi->fib_scope &&
395 cfg->fc_prefsrc == fi->fib_prefsrc &&
396 cfg->fc_priority == fi->fib_priority &&
397 cfg->fc_type == fi->fib_type &&
398 cfg->fc_table == fi->fib_tb_id &&
399 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
406 static struct fib_info *fib_find_info(struct fib_info *nfi)
408 struct hlist_head *head;
412 hash = fib_info_hashfn(nfi);
413 head = &fib_info_hash[hash];
415 hlist_for_each_entry(fi, head, fib_hash) {
416 if (!net_eq(fi->fib_net, nfi->fib_net))
418 if (fi->fib_nhs != nfi->fib_nhs)
420 if (nfi->fib_protocol == fi->fib_protocol &&
421 nfi->fib_scope == fi->fib_scope &&
422 nfi->fib_prefsrc == fi->fib_prefsrc &&
423 nfi->fib_priority == fi->fib_priority &&
424 nfi->fib_type == fi->fib_type &&
425 nfi->fib_tb_id == fi->fib_tb_id &&
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 static inline 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)
515 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
518 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
522 err = fib_dump_info(skb, info->portid, seq, event, tb_id,
523 fa->fa_type, key, dst_len,
524 fa->fa_tos, fa->fa_info, nlm_flags);
526 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
527 WARN_ON(err == -EMSGSIZE);
531 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
532 info->nlh, GFP_KERNEL);
536 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
539 static int fib_detect_death(struct fib_info *fi, int order,
540 struct fib_info **last_resort, int *last_idx,
543 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
545 int state = NUD_NONE;
547 if (likely(nhc->nhc_gw_family == AF_INET))
548 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
549 else if (nhc->nhc_gw_family == AF_INET6)
550 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
556 state = n->nud_state;
561 if (state == NUD_REACHABLE)
563 if ((state & NUD_VALID) && order != dflt)
565 if ((state & NUD_VALID) ||
566 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
573 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap,
574 u16 encap_type, 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(encap_type, encap, nhc->nhc_family,
593 cfg, &lwtstate, extack);
597 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
603 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
604 nhc->nhc_pcpu_rth_output = NULL;
607 EXPORT_SYMBOL_GPL(fib_nh_common_init);
609 int fib_nh_init(struct net *net, struct fib_nh *nh,
610 struct fib_config *cfg, int nh_weight,
611 struct netlink_ext_ack *extack)
615 nh->fib_nh_family = AF_INET;
617 err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
618 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
622 nh->fib_nh_oif = cfg->fc_oif;
623 nh->fib_nh_gw_family = cfg->fc_gw_family;
624 if (cfg->fc_gw_family == AF_INET)
625 nh->fib_nh_gw4 = cfg->fc_gw4;
626 else if (cfg->fc_gw_family == AF_INET6)
627 nh->fib_nh_gw6 = cfg->fc_gw6;
629 nh->fib_nh_flags = cfg->fc_flags;
631 #ifdef CONFIG_IP_ROUTE_CLASSID
632 nh->nh_tclassid = cfg->fc_flow;
634 atomic_inc(&net->ipv4.fib_num_tclassid_users);
636 #ifdef CONFIG_IP_ROUTE_MULTIPATH
637 nh->fib_nh_weight = nh_weight;
642 #ifdef CONFIG_IP_ROUTE_MULTIPATH
644 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
645 struct netlink_ext_ack *extack)
649 while (rtnh_ok(rtnh, remaining)) {
651 rtnh = rtnh_next(rtnh, &remaining);
654 /* leftover implies invalid nexthop configuration, discard it */
656 NL_SET_ERR_MSG(extack,
657 "Invalid nexthop configuration - extra data after nexthops");
664 static int fib_gw_from_attr(__be32 *gw, struct nlattr *nla,
665 struct netlink_ext_ack *extack)
667 if (nla_len(nla) < sizeof(*gw)) {
668 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_GATEWAY");
672 *gw = nla_get_in_addr(nla);
677 /* only called when fib_nh is integrated into fib_info */
678 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
679 int remaining, struct fib_config *cfg,
680 struct netlink_ext_ack *extack)
682 struct net *net = fi->fib_net;
683 struct fib_config fib_cfg;
687 change_nexthops(fi) {
690 memset(&fib_cfg, 0, sizeof(fib_cfg));
692 if (!rtnh_ok(rtnh, remaining)) {
693 NL_SET_ERR_MSG(extack,
694 "Invalid nexthop configuration - extra data after nexthop");
698 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
699 NL_SET_ERR_MSG(extack,
700 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
704 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
705 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
707 attrlen = rtnh_attrlen(rtnh);
709 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
711 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
712 nlav = nla_find(attrs, attrlen, RTA_VIA);
714 NL_SET_ERR_MSG(extack,
715 "Nexthop configuration can not contain both GATEWAY and VIA");
719 ret = fib_gw_from_attr(&fib_cfg.fc_gw4, nla,
725 fib_cfg.fc_gw_family = AF_INET;
727 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
732 nla = nla_find(attrs, attrlen, RTA_FLOW);
734 if (nla_len(nla) < sizeof(u32)) {
735 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
738 fib_cfg.fc_flow = nla_get_u32(nla);
741 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
742 /* RTA_ENCAP_TYPE length checked in
743 * lwtunnel_valid_encap_type_attr
745 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
747 fib_cfg.fc_encap_type = nla_get_u16(nla);
750 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
751 rtnh->rtnh_hops + 1, extack);
755 rtnh = rtnh_next(rtnh, &remaining);
756 } endfor_nexthops(fi);
759 nh = fib_info_nh(fi, 0);
760 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
761 NL_SET_ERR_MSG(extack,
762 "Nexthop device index does not match RTA_OIF");
765 if (cfg->fc_gw_family) {
766 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
767 (cfg->fc_gw_family == AF_INET &&
768 nh->fib_nh_gw4 != cfg->fc_gw4) ||
769 (cfg->fc_gw_family == AF_INET6 &&
770 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
771 NL_SET_ERR_MSG(extack,
772 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
776 #ifdef CONFIG_IP_ROUTE_CLASSID
777 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
778 NL_SET_ERR_MSG(extack,
779 "Nexthop class id does not match RTA_FLOW");
788 /* only called when fib_nh is integrated into fib_info */
789 static void fib_rebalance(struct fib_info *fi)
794 if (fib_info_num_path(fi) < 2)
799 if (nh->fib_nh_flags & RTNH_F_DEAD)
802 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
803 nh->fib_nh_flags & RTNH_F_LINKDOWN)
806 total += nh->fib_nh_weight;
807 } endfor_nexthops(fi);
810 change_nexthops(fi) {
813 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
815 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
816 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
819 w += nexthop_nh->fib_nh_weight;
820 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
824 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
825 } endfor_nexthops(fi);
827 #else /* CONFIG_IP_ROUTE_MULTIPATH */
829 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
830 int remaining, struct fib_config *cfg,
831 struct netlink_ext_ack *extack)
833 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
838 #define fib_rebalance(fi) do { } while (0)
840 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
842 static int fib_encap_match(u16 encap_type,
843 struct nlattr *encap,
844 const struct fib_nh *nh,
845 const struct fib_config *cfg,
846 struct netlink_ext_ack *extack)
848 struct lwtunnel_state *lwtstate;
851 if (encap_type == LWTUNNEL_ENCAP_NONE)
854 ret = lwtunnel_build_state(encap_type, encap, AF_INET,
855 cfg, &lwtstate, extack);
857 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
858 lwtstate_free(lwtstate);
864 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
865 struct netlink_ext_ack *extack)
867 #ifdef CONFIG_IP_ROUTE_MULTIPATH
868 struct rtnexthop *rtnh;
872 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
876 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
882 if (cfg->fc_oif || cfg->fc_gw_family || cfg->fc_mp)
887 if (cfg->fc_oif || cfg->fc_gw_family) {
890 nh = fib_info_nh(fi, 0);
892 if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
896 #ifdef CONFIG_IP_ROUTE_CLASSID
898 cfg->fc_flow != nh->nh_tclassid)
901 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
902 (cfg->fc_gw_family &&
903 cfg->fc_gw_family != nh->fib_nh_gw_family))
906 if (cfg->fc_gw_family == AF_INET &&
907 cfg->fc_gw4 != nh->fib_nh_gw4)
910 if (cfg->fc_gw_family == AF_INET6 &&
911 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
917 #ifdef CONFIG_IP_ROUTE_MULTIPATH
922 remaining = cfg->fc_mp_len;
927 if (!rtnh_ok(rtnh, remaining))
930 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
933 attrlen = rtnh_attrlen(rtnh);
935 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
938 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
939 nlav = nla_find(attrs, attrlen, RTA_VIA);
941 NL_SET_ERR_MSG(extack,
942 "Nexthop configuration can not contain both GATEWAY and VIA");
949 err = fib_gw_from_attr(&gw, nla, extack);
953 if (nh->fib_nh_gw_family != AF_INET ||
954 gw != nh->fib_nh_gw4)
957 struct fib_config cfg2;
959 err = fib_gw_from_via(&cfg2, nlav, extack);
963 switch (nh->fib_nh_gw_family) {
965 if (cfg2.fc_gw_family != AF_INET ||
966 cfg2.fc_gw4 != nh->fib_nh_gw4)
970 if (cfg2.fc_gw_family != AF_INET6 ||
971 ipv6_addr_cmp(&cfg2.fc_gw6,
978 #ifdef CONFIG_IP_ROUTE_CLASSID
979 nla = nla_find(attrs, attrlen, RTA_FLOW);
981 if (nla_len(nla) < sizeof(u32)) {
982 NL_SET_ERR_MSG(extack, "Invalid RTA_FLOW");
985 if (nla_get_u32(nla) != nh->nh_tclassid)
991 rtnh = rtnh_next(rtnh, &remaining);
992 } endfor_nexthops(fi);
997 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
1005 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
1006 int type = nla_type(nla);
1011 if (type > RTAX_MAX)
1014 type = array_index_nospec(type, RTAX_MAX + 1);
1015 if (type == RTAX_CC_ALGO) {
1016 char tmp[TCP_CA_NAME_MAX];
1017 bool ecn_ca = false;
1019 nla_strlcpy(tmp, nla, sizeof(tmp));
1020 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
1022 if (nla_len(nla) != sizeof(u32))
1024 val = nla_get_u32(nla);
1027 fi_val = fi->fib_metrics->metrics[type - 1];
1028 if (type == RTAX_FEATURES)
1029 fi_val &= ~DST_FEATURE_ECN_CA;
1038 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
1039 u32 table, struct netlink_ext_ack *extack)
1041 struct fib6_config cfg = {
1043 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
1044 .fc_ifindex = nh->fib_nh_oif,
1045 .fc_gateway = nh->fib_nh_gw6,
1047 struct fib6_nh fib6_nh = {};
1050 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1052 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1053 dev_hold(nh->fib_nh_dev);
1054 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1055 nh->fib_nh_scope = RT_SCOPE_LINK;
1057 ipv6_stub->fib6_nh_release(&fib6_nh);
1067 * Semantics of nexthop is very messy by historical reasons.
1068 * We have to take into account, that:
1069 * a) gateway can be actually local interface address,
1070 * so that gatewayed route is direct.
1071 * b) gateway must be on-link address, possibly
1072 * described not by an ifaddr, but also by a direct route.
1073 * c) If both gateway and interface are specified, they should not
1075 * d) If we use tunnel routes, gateway could be not on-link.
1077 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1078 * results in pretty ugly and hairy code with obscure logic.
1080 * I chose to generalized it instead, so that the size
1081 * of code does not increase practically, but it becomes
1082 * much more general.
1083 * Every prefix is assigned a "scope" value: "host" is local address,
1084 * "link" is direct route,
1085 * [ ... "site" ... "interior" ... ]
1086 * and "universe" is true gateway route with global meaning.
1088 * Every prefix refers to a set of "nexthop"s (gw, oif),
1089 * where gw must have narrower scope. This recursion stops
1090 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1091 * which means that gw is forced to be on link.
1093 * Code is still hairy, but now it is apparently logically
1094 * consistent and very flexible. F.e. as by-product it allows
1095 * to co-exists in peace independent exterior and interior
1096 * routing processes.
1098 * Normally it looks as following.
1100 * {universe prefix} -> (gw, oif) [scope link]
1102 * |-> {link prefix} -> (gw, oif) [scope local]
1104 * |-> {local prefix} (terminal node)
1106 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1107 u8 scope, struct netlink_ext_ack *extack)
1109 struct net_device *dev;
1110 struct fib_result res;
1113 if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1114 unsigned int addr_type;
1116 if (scope >= RT_SCOPE_LINK) {
1117 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1120 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1122 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1125 if (!(dev->flags & IFF_UP)) {
1126 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1129 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1130 if (addr_type != RTN_UNICAST) {
1131 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1134 if (!netif_carrier_ok(dev))
1135 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1136 nh->fib_nh_dev = dev;
1138 nh->fib_nh_scope = RT_SCOPE_LINK;
1143 struct fib_table *tbl = NULL;
1144 struct flowi4 fl4 = {
1145 .daddr = nh->fib_nh_gw4,
1146 .flowi4_scope = scope + 1,
1147 .flowi4_oif = nh->fib_nh_oif,
1148 .flowi4_iif = LOOPBACK_IFINDEX,
1151 /* It is not necessary, but requires a bit of thinking */
1152 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1153 fl4.flowi4_scope = RT_SCOPE_LINK;
1155 if (table && table != RT_TABLE_MAIN)
1156 tbl = fib_get_table(net, table);
1159 err = fib_table_lookup(tbl, &fl4, &res,
1160 FIB_LOOKUP_IGNORE_LINKSTATE |
1163 /* on error or if no table given do full lookup. This
1164 * is needed for example when nexthops are in the local
1165 * table rather than the given table
1168 err = fib_lookup(net, &fl4, &res,
1169 FIB_LOOKUP_IGNORE_LINKSTATE);
1173 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1179 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1180 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1183 nh->fib_nh_scope = res.scope;
1184 nh->fib_nh_oif = FIB_RES_OIF(res);
1185 nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1187 NL_SET_ERR_MSG(extack,
1188 "No egress device for nexthop gateway");
1192 if (!netif_carrier_ok(dev))
1193 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1194 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1200 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1201 struct netlink_ext_ack *extack)
1203 struct in_device *in_dev;
1206 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1207 NL_SET_ERR_MSG(extack,
1208 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1215 in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1219 if (!(in_dev->dev->flags & IFF_UP)) {
1220 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1224 nh->fib_nh_dev = in_dev->dev;
1225 dev_hold(nh->fib_nh_dev);
1226 nh->fib_nh_scope = RT_SCOPE_LINK;
1227 if (!netif_carrier_ok(nh->fib_nh_dev))
1228 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1235 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1236 struct netlink_ext_ack *extack)
1240 if (nh->fib_nh_gw_family == AF_INET)
1241 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1242 else if (nh->fib_nh_gw_family == AF_INET6)
1243 err = fib_check_nh_v6_gw(net, nh, table, extack);
1245 err = fib_check_nh_nongw(net, nh, extack);
1250 static inline unsigned int fib_laddr_hashfn(__be32 val)
1252 unsigned int mask = (fib_info_hash_size - 1);
1254 return ((__force u32)val ^
1255 ((__force u32)val >> 7) ^
1256 ((__force u32)val >> 14)) & mask;
1259 static struct hlist_head *fib_info_hash_alloc(int bytes)
1261 if (bytes <= PAGE_SIZE)
1262 return kzalloc(bytes, GFP_KERNEL);
1264 return (struct hlist_head *)
1265 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1269 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1274 if (bytes <= PAGE_SIZE)
1277 free_pages((unsigned long) hash, get_order(bytes));
1280 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1281 struct hlist_head *new_laddrhash,
1282 unsigned int new_size)
1284 struct hlist_head *old_info_hash, *old_laddrhash;
1285 unsigned int old_size = fib_info_hash_size;
1286 unsigned int i, bytes;
1288 spin_lock_bh(&fib_info_lock);
1289 old_info_hash = fib_info_hash;
1290 old_laddrhash = fib_info_laddrhash;
1291 fib_info_hash_size = new_size;
1293 for (i = 0; i < old_size; i++) {
1294 struct hlist_head *head = &fib_info_hash[i];
1295 struct hlist_node *n;
1296 struct fib_info *fi;
1298 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1299 struct hlist_head *dest;
1300 unsigned int new_hash;
1302 new_hash = fib_info_hashfn(fi);
1303 dest = &new_info_hash[new_hash];
1304 hlist_add_head(&fi->fib_hash, dest);
1307 fib_info_hash = new_info_hash;
1309 for (i = 0; i < old_size; i++) {
1310 struct hlist_head *lhead = &fib_info_laddrhash[i];
1311 struct hlist_node *n;
1312 struct fib_info *fi;
1314 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1315 struct hlist_head *ldest;
1316 unsigned int new_hash;
1318 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1319 ldest = &new_laddrhash[new_hash];
1320 hlist_add_head(&fi->fib_lhash, ldest);
1323 fib_info_laddrhash = new_laddrhash;
1325 spin_unlock_bh(&fib_info_lock);
1327 bytes = old_size * sizeof(struct hlist_head *);
1328 fib_info_hash_free(old_info_hash, bytes);
1329 fib_info_hash_free(old_laddrhash, bytes);
1332 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1333 unsigned char scope)
1337 if (nhc->nhc_family != AF_INET)
1338 return inet_select_addr(nhc->nhc_dev, 0, scope);
1340 nh = container_of(nhc, struct fib_nh, nh_common);
1341 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1342 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1344 return nh->nh_saddr;
1347 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1349 struct fib_nh_common *nhc = res->nhc;
1351 if (res->fi->fib_prefsrc)
1352 return res->fi->fib_prefsrc;
1354 if (nhc->nhc_family == AF_INET) {
1357 nh = container_of(nhc, struct fib_nh, nh_common);
1358 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1359 return nh->nh_saddr;
1362 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1365 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1367 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1368 fib_prefsrc != cfg->fc_dst) {
1369 u32 tb_id = cfg->fc_table;
1372 if (tb_id == RT_TABLE_MAIN)
1373 tb_id = RT_TABLE_LOCAL;
1375 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1376 fib_prefsrc, tb_id);
1378 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1379 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1380 fib_prefsrc, RT_TABLE_LOCAL);
1383 if (rc != RTN_LOCAL)
1389 struct fib_info *fib_create_info(struct fib_config *cfg,
1390 struct netlink_ext_ack *extack)
1393 struct fib_info *fi = NULL;
1394 struct nexthop *nh = NULL;
1395 struct fib_info *ofi;
1397 struct net *net = cfg->fc_nlinfo.nl_net;
1399 if (cfg->fc_type > RTN_MAX)
1402 /* Fast check to catch the most weird cases */
1403 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1404 NL_SET_ERR_MSG(extack, "Invalid scope");
1408 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1409 NL_SET_ERR_MSG(extack,
1410 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1414 if (cfg->fc_nh_id) {
1416 fi = fib_find_info_nh(net, cfg);
1423 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1425 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1431 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1433 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1440 if (fib_info_cnt >= fib_info_hash_size) {
1441 unsigned int new_size = fib_info_hash_size << 1;
1442 struct hlist_head *new_info_hash;
1443 struct hlist_head *new_laddrhash;
1448 bytes = new_size * sizeof(struct hlist_head *);
1449 new_info_hash = fib_info_hash_alloc(bytes);
1450 new_laddrhash = fib_info_hash_alloc(bytes);
1451 if (!new_info_hash || !new_laddrhash) {
1452 fib_info_hash_free(new_info_hash, bytes);
1453 fib_info_hash_free(new_laddrhash, bytes);
1455 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1457 if (!fib_info_hash_size)
1461 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1464 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1465 cfg->fc_mx_len, extack);
1466 if (IS_ERR(fi->fib_metrics)) {
1467 err = PTR_ERR(fi->fib_metrics);
1469 return ERR_PTR(err);
1474 fi->fib_protocol = cfg->fc_protocol;
1475 fi->fib_scope = cfg->fc_scope;
1476 fi->fib_flags = cfg->fc_flags;
1477 fi->fib_priority = cfg->fc_priority;
1478 fi->fib_prefsrc = cfg->fc_prefsrc;
1479 fi->fib_type = cfg->fc_type;
1480 fi->fib_tb_id = cfg->fc_table;
1484 if (!nexthop_get(nh)) {
1485 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1492 change_nexthops(fi) {
1493 nexthop_nh->nh_parent = fi;
1494 } endfor_nexthops(fi)
1497 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1500 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1506 if (fib_props[cfg->fc_type].error) {
1507 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1508 NL_SET_ERR_MSG(extack,
1509 "Gateway, device and multipath can not be specified for this route type");
1514 switch (cfg->fc_type) {
1522 NL_SET_ERR_MSG(extack, "Invalid route type");
1527 if (cfg->fc_scope > RT_SCOPE_HOST) {
1528 NL_SET_ERR_MSG(extack, "Invalid scope");
1533 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1536 } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1537 struct fib_nh *nh = fi->fib_nh;
1539 /* Local address is added. */
1541 NL_SET_ERR_MSG(extack,
1542 "Route with host scope can not have multiple nexthops");
1545 if (nh->fib_nh_gw_family) {
1546 NL_SET_ERR_MSG(extack,
1547 "Route with host scope can not have a gateway");
1550 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1551 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1553 if (!nh->fib_nh_dev)
1558 change_nexthops(fi) {
1559 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1560 cfg->fc_table, cfg->fc_scope,
1564 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1566 } endfor_nexthops(fi)
1567 if (linkdown == fi->fib_nhs)
1568 fi->fib_flags |= RTNH_F_LINKDOWN;
1571 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1572 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1577 change_nexthops(fi) {
1578 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1580 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1581 fi->fib_nh_is_v6 = true;
1582 } endfor_nexthops(fi)
1588 ofi = fib_find_info(fi);
1590 /* fib_table_lookup() should not see @fi yet. */
1598 refcount_set(&fi->fib_clntref, 1);
1599 spin_lock_bh(&fib_info_lock);
1600 hlist_add_head(&fi->fib_hash,
1601 &fib_info_hash[fib_info_hashfn(fi)]);
1602 if (fi->fib_prefsrc) {
1603 struct hlist_head *head;
1605 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1606 hlist_add_head(&fi->fib_lhash, head);
1609 list_add(&fi->nh_list, &nh->fi_list);
1611 change_nexthops(fi) {
1612 struct hlist_head *head;
1614 if (!nexthop_nh->fib_nh_dev)
1616 head = fib_info_devhash_bucket(nexthop_nh->fib_nh_dev);
1617 hlist_add_head(&nexthop_nh->nh_hash, head);
1618 } endfor_nexthops(fi)
1620 spin_unlock_bh(&fib_info_lock);
1628 /* fib_table_lookup() should not see @fi yet. */
1633 return ERR_PTR(err);
1636 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1637 u8 rt_family, unsigned char *flags, bool skip_oif)
1639 if (nhc->nhc_flags & RTNH_F_DEAD)
1640 *flags |= RTNH_F_DEAD;
1642 if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1643 *flags |= RTNH_F_LINKDOWN;
1646 switch (nhc->nhc_family) {
1648 if (ip_ignore_linkdown(nhc->nhc_dev))
1649 *flags |= RTNH_F_DEAD;
1652 if (ip6_ignore_linkdown(nhc->nhc_dev))
1653 *flags |= RTNH_F_DEAD;
1659 switch (nhc->nhc_gw_family) {
1661 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1662 goto nla_put_failure;
1665 /* if gateway family does not match nexthop family
1666 * gateway is encoded as RTA_VIA
1668 if (rt_family != nhc->nhc_gw_family) {
1669 int alen = sizeof(struct in6_addr);
1673 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1675 goto nla_put_failure;
1677 via = nla_data(nla);
1678 via->rtvia_family = AF_INET6;
1679 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1680 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1681 &nhc->nhc_gw.ipv6) < 0) {
1682 goto nla_put_failure;
1687 *flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1688 if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1689 *flags |= RTNH_F_OFFLOAD;
1691 if (!skip_oif && nhc->nhc_dev &&
1692 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1693 goto nla_put_failure;
1695 if (nhc->nhc_lwtstate &&
1696 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1697 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1698 goto nla_put_failure;
1705 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1707 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1708 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1709 int nh_weight, u8 rt_family, u32 nh_tclassid)
1711 const struct net_device *dev = nhc->nhc_dev;
1712 struct rtnexthop *rtnh;
1713 unsigned char flags = 0;
1715 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1717 goto nla_put_failure;
1719 rtnh->rtnh_hops = nh_weight - 1;
1720 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1722 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1723 goto nla_put_failure;
1725 rtnh->rtnh_flags = flags;
1727 if (nh_tclassid && nla_put_u32(skb, RTA_FLOW, nh_tclassid))
1728 goto nla_put_failure;
1730 /* length of rtnetlink header + attributes */
1731 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1738 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1741 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1742 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1746 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1748 goto nla_put_failure;
1750 if (unlikely(fi->nh)) {
1751 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1752 goto nla_put_failure;
1757 u32 nh_tclassid = 0;
1758 #ifdef CONFIG_IP_ROUTE_CLASSID
1759 nh_tclassid = nh->nh_tclassid;
1761 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1762 AF_INET, nh_tclassid) < 0)
1763 goto nla_put_failure;
1764 } endfor_nexthops(fi);
1767 nla_nest_end(skb, mp);
1775 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1781 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1782 u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos,
1783 struct fib_info *fi, unsigned int flags)
1785 unsigned int nhs = fib_info_num_path(fi);
1786 struct nlmsghdr *nlh;
1789 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1793 rtm = nlmsg_data(nlh);
1794 rtm->rtm_family = AF_INET;
1795 rtm->rtm_dst_len = dst_len;
1796 rtm->rtm_src_len = 0;
1799 rtm->rtm_table = tb_id;
1801 rtm->rtm_table = RT_TABLE_COMPAT;
1802 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1803 goto nla_put_failure;
1804 rtm->rtm_type = type;
1805 rtm->rtm_flags = fi->fib_flags;
1806 rtm->rtm_scope = fi->fib_scope;
1807 rtm->rtm_protocol = fi->fib_protocol;
1809 if (rtm->rtm_dst_len &&
1810 nla_put_in_addr(skb, RTA_DST, dst))
1811 goto nla_put_failure;
1812 if (fi->fib_priority &&
1813 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1814 goto nla_put_failure;
1815 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1816 goto nla_put_failure;
1818 if (fi->fib_prefsrc &&
1819 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1820 goto nla_put_failure;
1823 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1824 goto nla_put_failure;
1825 if (nexthop_is_blackhole(fi->nh))
1826 rtm->rtm_type = RTN_BLACKHOLE;
1830 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1831 unsigned char flags = 0;
1833 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1834 goto nla_put_failure;
1836 rtm->rtm_flags = flags;
1837 #ifdef CONFIG_IP_ROUTE_CLASSID
1838 if (nhc->nhc_family == AF_INET) {
1841 nh = container_of(nhc, struct fib_nh, nh_common);
1842 if (nh->nh_tclassid &&
1843 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1844 goto nla_put_failure;
1848 if (fib_add_multipath(skb, fi) < 0)
1849 goto nla_put_failure;
1852 nlmsg_end(skb, nlh);
1856 nlmsg_cancel(skb, nlh);
1862 * - local address disappeared -> we must delete all the entries
1864 * - device went down -> we must shutdown all nexthops going via it.
1866 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1869 unsigned int hash = fib_laddr_hashfn(local);
1870 struct hlist_head *head = &fib_info_laddrhash[hash];
1871 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1872 struct net *net = dev_net(dev);
1873 struct fib_info *fi;
1875 if (!fib_info_laddrhash || local == 0)
1878 hlist_for_each_entry(fi, head, fib_lhash) {
1879 if (!net_eq(fi->fib_net, net) ||
1880 fi->fib_tb_id != tb_id)
1882 if (fi->fib_prefsrc == local) {
1883 fi->fib_flags |= RTNH_F_DEAD;
1890 static int call_fib_nh_notifiers(struct fib_nh *nh,
1891 enum fib_event_type event_type)
1893 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1894 struct fib_nh_notifier_info info = {
1898 switch (event_type) {
1899 case FIB_EVENT_NH_ADD:
1900 if (nh->fib_nh_flags & RTNH_F_DEAD)
1902 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1904 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1906 case FIB_EVENT_NH_DEL:
1907 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1908 (nh->fib_nh_flags & RTNH_F_DEAD))
1909 return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1910 event_type, &info.info);
1918 /* Update the PMTU of exceptions when:
1919 * - the new MTU of the first hop becomes smaller than the PMTU
1920 * - the old MTU was the same as the PMTU, and it limited discovery of
1921 * larger MTUs on the path. With that limit raised, we can now
1922 * discover larger MTUs
1923 * A special case is locked exceptions, for which the PMTU is smaller
1924 * than the minimal accepted PMTU:
1925 * - if the new MTU is greater than the PMTU, don't make any change
1926 * - otherwise, unlock and set PMTU
1928 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1930 struct fnhe_hash_bucket *bucket;
1933 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1937 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1938 struct fib_nh_exception *fnhe;
1940 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1942 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1943 if (fnhe->fnhe_mtu_locked) {
1944 if (new <= fnhe->fnhe_pmtu) {
1945 fnhe->fnhe_pmtu = new;
1946 fnhe->fnhe_mtu_locked = false;
1948 } else if (new < fnhe->fnhe_pmtu ||
1949 orig == fnhe->fnhe_pmtu) {
1950 fnhe->fnhe_pmtu = new;
1956 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1958 struct hlist_head *head = fib_info_devhash_bucket(dev);
1961 hlist_for_each_entry(nh, head, nh_hash) {
1962 if (nh->fib_nh_dev == dev)
1963 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1967 /* Event force Flags Description
1968 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1969 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1970 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1971 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1973 * only used when fib_nh is built into fib_info
1975 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1977 struct hlist_head *head = fib_info_devhash_bucket(dev);
1978 struct fib_info *prev_fi = NULL;
1979 int scope = RT_SCOPE_NOWHERE;
1986 hlist_for_each_entry(nh, head, nh_hash) {
1987 struct fib_info *fi = nh->nh_parent;
1990 BUG_ON(!fi->fib_nhs);
1991 if (nh->fib_nh_dev != dev || fi == prev_fi)
1995 change_nexthops(fi) {
1996 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1998 else if (nexthop_nh->fib_nh_dev == dev &&
1999 nexthop_nh->fib_nh_scope != scope) {
2002 case NETDEV_UNREGISTER:
2003 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
2006 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
2009 call_fib_nh_notifiers(nexthop_nh,
2013 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2014 if (event == NETDEV_UNREGISTER &&
2015 nexthop_nh->fib_nh_dev == dev) {
2020 } endfor_nexthops(fi)
2021 if (dead == fi->fib_nhs) {
2024 case NETDEV_UNREGISTER:
2025 fi->fib_flags |= RTNH_F_DEAD;
2028 fi->fib_flags |= RTNH_F_LINKDOWN;
2040 /* Must be invoked inside of an RCU protected region. */
2041 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
2043 struct fib_info *fi = NULL, *last_resort = NULL;
2044 struct hlist_head *fa_head = res->fa_head;
2045 struct fib_table *tb = res->table;
2046 u8 slen = 32 - res->prefixlen;
2047 int order = -1, last_idx = -1;
2048 struct fib_alias *fa, *fa1 = NULL;
2049 u32 last_prio = res->fi->fib_priority;
2052 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2053 struct fib_info *next_fi = fa->fa_info;
2054 struct fib_nh_common *nhc;
2056 if (fa->fa_slen != slen)
2058 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2060 if (fa->tb_id != tb->tb_id)
2062 if (next_fi->fib_priority > last_prio &&
2063 fa->fa_tos == last_tos) {
2068 if (next_fi->fib_flags & RTNH_F_DEAD)
2070 last_tos = fa->fa_tos;
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 (READ_ONCE(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 && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_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);