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: FIB frontend.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/bitops.h>
15 #include <linux/capability.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
19 #include <linux/string.h>
20 #include <linux/socket.h>
21 #include <linux/sockios.h>
22 #include <linux/errno.h>
24 #include <linux/inet.h>
25 #include <linux/inetdevice.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_addr.h>
28 #include <linux/if_arp.h>
29 #include <linux/skbuff.h>
30 #include <linux/cache.h>
31 #include <linux/init.h>
32 #include <linux/list.h>
33 #include <linux/slab.h>
36 #include <net/protocol.h>
37 #include <net/route.h>
41 #include <net/ip_fib.h>
42 #include <net/nexthop.h>
43 #include <net/rtnetlink.h>
45 #include <net/l3mdev.h>
46 #include <net/lwtunnel.h>
47 #include <trace/events/fib.h>
49 #ifndef CONFIG_IP_MULTIPLE_TABLES
51 static int __net_init fib4_rules_init(struct net *net)
53 struct fib_table *local_table, *main_table;
55 main_table = fib_trie_table(RT_TABLE_MAIN, NULL);
59 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
63 hlist_add_head_rcu(&local_table->tb_hlist,
64 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
65 hlist_add_head_rcu(&main_table->tb_hlist,
66 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
70 fib_free_table(main_table);
75 struct fib_table *fib_new_table(struct net *net, u32 id)
77 struct fib_table *tb, *alias = NULL;
82 tb = fib_get_table(net, id);
86 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
87 alias = fib_new_table(net, RT_TABLE_MAIN);
89 tb = fib_trie_table(id, alias);
95 rcu_assign_pointer(net->ipv4.fib_main, tb);
97 case RT_TABLE_DEFAULT:
98 rcu_assign_pointer(net->ipv4.fib_default, tb);
104 h = id & (FIB_TABLE_HASHSZ - 1);
105 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
108 EXPORT_SYMBOL_GPL(fib_new_table);
110 /* caller must hold either rtnl or rcu read lock */
111 struct fib_table *fib_get_table(struct net *net, u32 id)
113 struct fib_table *tb;
114 struct hlist_head *head;
119 h = id & (FIB_TABLE_HASHSZ - 1);
121 head = &net->ipv4.fib_table_hash[h];
122 hlist_for_each_entry_rcu(tb, head, tb_hlist,
123 lockdep_rtnl_is_held()) {
129 #endif /* CONFIG_IP_MULTIPLE_TABLES */
131 static void fib_replace_table(struct net *net, struct fib_table *old,
132 struct fib_table *new)
134 #ifdef CONFIG_IP_MULTIPLE_TABLES
135 switch (new->tb_id) {
137 rcu_assign_pointer(net->ipv4.fib_main, new);
139 case RT_TABLE_DEFAULT:
140 rcu_assign_pointer(net->ipv4.fib_default, new);
147 /* replace the old table in the hlist */
148 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
151 int fib_unmerge(struct net *net)
153 struct fib_table *old, *new, *main_table;
155 /* attempt to fetch local table if it has been allocated */
156 old = fib_get_table(net, RT_TABLE_LOCAL);
160 new = fib_trie_unmerge(old);
164 /* table is already unmerged */
168 /* replace merged table with clean table */
169 fib_replace_table(net, old, new);
172 /* attempt to fetch main table if it has been allocated */
173 main_table = fib_get_table(net, RT_TABLE_MAIN);
177 /* flush local entries from main table */
178 fib_table_flush_external(main_table);
183 void fib_flush(struct net *net)
188 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
189 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
190 struct hlist_node *tmp;
191 struct fib_table *tb;
193 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
194 flushed += fib_table_flush(net, tb, false);
202 * Find address type as if only "dev" was present in the system. If
203 * on_dev is NULL then all interfaces are taken into consideration.
205 static inline unsigned int __inet_dev_addr_type(struct net *net,
206 const struct net_device *dev,
207 __be32 addr, u32 tb_id)
209 struct flowi4 fl4 = { .daddr = addr };
210 struct fib_result res;
211 unsigned int ret = RTN_BROADCAST;
212 struct fib_table *table;
214 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
215 return RTN_BROADCAST;
216 if (ipv4_is_multicast(addr))
217 return RTN_MULTICAST;
221 table = fib_get_table(net, tb_id);
224 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
225 struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
227 if (!dev || dev == nhc->nhc_dev)
236 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
238 return __inet_dev_addr_type(net, NULL, addr, tb_id);
240 EXPORT_SYMBOL(inet_addr_type_table);
242 unsigned int inet_addr_type(struct net *net, __be32 addr)
244 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
246 EXPORT_SYMBOL(inet_addr_type);
248 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
251 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
253 return __inet_dev_addr_type(net, dev, addr, rt_table);
255 EXPORT_SYMBOL(inet_dev_addr_type);
257 /* inet_addr_type with dev == NULL but using the table from a dev
258 * if one is associated
260 unsigned int inet_addr_type_dev_table(struct net *net,
261 const struct net_device *dev,
264 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
266 return __inet_dev_addr_type(net, NULL, addr, rt_table);
268 EXPORT_SYMBOL(inet_addr_type_dev_table);
270 __be32 fib_compute_spec_dst(struct sk_buff *skb)
272 struct net_device *dev = skb->dev;
273 struct in_device *in_dev;
274 struct fib_result res;
279 rt = skb_rtable(skb);
280 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
282 return ip_hdr(skb)->daddr;
284 in_dev = __in_dev_get_rcu(dev);
288 scope = RT_SCOPE_UNIVERSE;
289 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
290 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
291 struct flowi4 fl4 = {
292 .flowi4_iif = LOOPBACK_IFINDEX,
293 .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
294 .daddr = ip_hdr(skb)->saddr,
295 .flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
296 .flowi4_scope = scope,
297 .flowi4_mark = vmark ? skb->mark : 0,
299 if (!fib_lookup(net, &fl4, &res, 0))
300 return fib_result_prefsrc(net, &res);
302 scope = RT_SCOPE_LINK;
305 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
308 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
310 bool dev_match = false;
311 #ifdef CONFIG_IP_ROUTE_MULTIPATH
312 if (unlikely(fi->nh)) {
313 dev_match = nexthop_uses_dev(fi->nh, dev);
317 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
318 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
320 if (nhc_l3mdev_matches_dev(nhc, dev)) {
327 if (fib_info_nhc(fi, 0)->nhc_dev == dev)
333 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
335 /* Given (packet source, input interface) and optional (dst, oif, tos):
336 * - (main) check, that source is valid i.e. not broadcast or our local
338 * - figure out what "logical" interface this packet arrived
339 * and calculate "specific destination" address.
340 * - check, that packet arrived from expected physical interface.
341 * called with rcu_read_lock()
343 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
344 u8 tos, int oif, struct net_device *dev,
345 int rpf, struct in_device *idev, u32 *itag)
347 struct net *net = dev_net(dev);
348 struct flow_keys flkeys;
350 struct fib_result res;
355 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
357 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
360 fl4.flowi4_tos = tos;
361 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
362 fl4.flowi4_tun_key.tun_id = 0;
363 fl4.flowi4_flags = 0;
364 fl4.flowi4_uid = sock_net_uid(net, NULL);
365 fl4.flowi4_multipath_hash = 0;
367 no_addr = idev->ifa_list == NULL;
369 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
370 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
371 fl4.flowi4_proto = 0;
375 swap(fl4.fl4_sport, fl4.fl4_dport);
378 if (fib_lookup(net, &fl4, &res, 0))
380 if (res.type != RTN_UNICAST &&
381 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
383 fib_combine_itag(itag, &res);
385 dev_match = fib_info_nh_uses_dev(res.fi, dev);
386 /* This is not common, loopback packets retain skb_dst so normally they
387 * would not even hit this slow path.
389 dev_match = dev_match || (res.type == RTN_LOCAL &&
390 dev == net->loopback_dev);
392 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_LINK;
399 fl4.flowi4_oif = dev->ifindex;
402 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
403 if (res.type == RTN_UNICAST)
404 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_LINK;
420 /* Ignore rp_filter for packets protected by IPsec. */
421 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
422 u8 tos, int oif, struct net_device *dev,
423 struct in_device *idev, u32 *itag)
425 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
426 struct net *net = dev_net(dev);
428 if (!r && !fib_num_tclassid_users(net) &&
429 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
430 if (IN_DEV_ACCEPT_LOCAL(idev))
432 /* with custom local routes in place, checking local addresses
433 * only will be too optimistic, with custom rules, checking
434 * local addresses only can be too strict, e.g. due to vrf
436 if (net->ipv4.fib_has_custom_local_routes ||
437 fib4_has_custom_rules(net))
439 if (inet_lookup_ifaddr_rcu(net, src))
448 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
451 static inline __be32 sk_extract_addr(struct sockaddr *addr)
453 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
456 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
460 nla = (struct nlattr *) ((char *) mx + len);
461 nla->nla_type = type;
462 nla->nla_len = nla_attr_size(4);
463 *(u32 *) nla_data(nla) = value;
465 return len + nla_total_size(4);
468 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
469 struct fib_config *cfg)
474 memset(cfg, 0, sizeof(*cfg));
475 cfg->fc_nlinfo.nl_net = net;
477 if (rt->rt_dst.sa_family != AF_INET)
478 return -EAFNOSUPPORT;
481 * Check mask for validity:
482 * a) it must be contiguous.
483 * b) destination must have all host bits clear.
484 * c) if application forgot to set correct family (AF_INET),
485 * reject request unless it is absolutely clear i.e.
486 * both family and mask are zero.
489 addr = sk_extract_addr(&rt->rt_dst);
490 if (!(rt->rt_flags & RTF_HOST)) {
491 __be32 mask = sk_extract_addr(&rt->rt_genmask);
493 if (rt->rt_genmask.sa_family != AF_INET) {
494 if (mask || rt->rt_genmask.sa_family)
495 return -EAFNOSUPPORT;
498 if (bad_mask(mask, addr))
501 plen = inet_mask_len(mask);
504 cfg->fc_dst_len = plen;
507 if (cmd != SIOCDELRT) {
508 cfg->fc_nlflags = NLM_F_CREATE;
509 cfg->fc_protocol = RTPROT_BOOT;
513 cfg->fc_priority = rt->rt_metric - 1;
515 if (rt->rt_flags & RTF_REJECT) {
516 cfg->fc_scope = RT_SCOPE_HOST;
517 cfg->fc_type = RTN_UNREACHABLE;
521 cfg->fc_scope = RT_SCOPE_NOWHERE;
522 cfg->fc_type = RTN_UNICAST;
526 struct net_device *dev;
527 char devname[IFNAMSIZ];
529 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
532 devname[IFNAMSIZ-1] = 0;
533 colon = strchr(devname, ':');
536 dev = __dev_get_by_name(net, devname);
539 cfg->fc_oif = dev->ifindex;
540 cfg->fc_table = l3mdev_fib_table(dev);
542 const struct in_ifaddr *ifa;
543 struct in_device *in_dev;
545 in_dev = __in_dev_get_rtnl(dev);
552 in_dev_for_each_ifa_rcu(ifa, in_dev) {
553 if (strcmp(ifa->ifa_label, devname) == 0)
560 cfg->fc_prefsrc = ifa->ifa_local;
564 addr = sk_extract_addr(&rt->rt_gateway);
565 if (rt->rt_gateway.sa_family == AF_INET && addr) {
566 unsigned int addr_type;
569 cfg->fc_gw_family = AF_INET;
570 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
571 if (rt->rt_flags & RTF_GATEWAY &&
572 addr_type == RTN_UNICAST)
573 cfg->fc_scope = RT_SCOPE_UNIVERSE;
577 cfg->fc_table = RT_TABLE_MAIN;
579 if (cmd == SIOCDELRT)
582 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
585 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
586 cfg->fc_scope = RT_SCOPE_LINK;
588 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
592 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
596 if (rt->rt_flags & RTF_MTU)
597 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
599 if (rt->rt_flags & RTF_WINDOW)
600 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
602 if (rt->rt_flags & RTF_IRTT)
603 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
606 cfg->fc_mx_len = len;
613 * Handle IP routing ioctl calls.
614 * These are used to manipulate the routing tables
616 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
618 struct fib_config cfg;
622 case SIOCADDRT: /* Add a route */
623 case SIOCDELRT: /* Delete a route */
624 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
628 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
630 struct fib_table *tb;
632 if (cmd == SIOCDELRT) {
633 tb = fib_get_table(net, cfg.fc_table);
635 err = fib_table_delete(net, tb, &cfg,
640 tb = fib_new_table(net, cfg.fc_table);
642 err = fib_table_insert(net, tb,
648 /* allocated by rtentry_to_fib_config() */
657 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
658 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 },
659 [RTA_DST] = { .type = NLA_U32 },
660 [RTA_SRC] = { .type = NLA_U32 },
661 [RTA_IIF] = { .type = NLA_U32 },
662 [RTA_OIF] = { .type = NLA_U32 },
663 [RTA_GATEWAY] = { .type = NLA_U32 },
664 [RTA_PRIORITY] = { .type = NLA_U32 },
665 [RTA_PREFSRC] = { .type = NLA_U32 },
666 [RTA_METRICS] = { .type = NLA_NESTED },
667 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
668 [RTA_FLOW] = { .type = NLA_U32 },
669 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
670 [RTA_ENCAP] = { .type = NLA_NESTED },
671 [RTA_UID] = { .type = NLA_U32 },
672 [RTA_MARK] = { .type = NLA_U32 },
673 [RTA_TABLE] = { .type = NLA_U32 },
674 [RTA_IP_PROTO] = { .type = NLA_U8 },
675 [RTA_SPORT] = { .type = NLA_U16 },
676 [RTA_DPORT] = { .type = NLA_U16 },
677 [RTA_NH_ID] = { .type = NLA_U32 },
680 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
681 struct netlink_ext_ack *extack)
686 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
687 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
692 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
694 switch (via->rtvia_family) {
696 if (alen != sizeof(__be32)) {
697 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
700 cfg->fc_gw_family = AF_INET;
701 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
704 #if IS_ENABLED(CONFIG_IPV6)
705 if (alen != sizeof(struct in6_addr)) {
706 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
709 cfg->fc_gw_family = AF_INET6;
710 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
712 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
717 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
724 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
725 struct nlmsghdr *nlh, struct fib_config *cfg,
726 struct netlink_ext_ack *extack)
728 bool has_gw = false, has_via = false;
733 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
734 rtm_ipv4_policy, extack);
738 memset(cfg, 0, sizeof(*cfg));
740 rtm = nlmsg_data(nlh);
741 cfg->fc_dst_len = rtm->rtm_dst_len;
742 cfg->fc_tos = rtm->rtm_tos;
743 cfg->fc_table = rtm->rtm_table;
744 cfg->fc_protocol = rtm->rtm_protocol;
745 cfg->fc_scope = rtm->rtm_scope;
746 cfg->fc_type = rtm->rtm_type;
747 cfg->fc_flags = rtm->rtm_flags;
748 cfg->fc_nlflags = nlh->nlmsg_flags;
750 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
751 cfg->fc_nlinfo.nlh = nlh;
752 cfg->fc_nlinfo.nl_net = net;
754 if (cfg->fc_type > RTN_MAX) {
755 NL_SET_ERR_MSG(extack, "Invalid route type");
760 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
761 switch (nla_type(attr)) {
763 cfg->fc_dst = nla_get_be32(attr);
766 cfg->fc_oif = nla_get_u32(attr);
770 cfg->fc_gw4 = nla_get_be32(attr);
772 cfg->fc_gw_family = AF_INET;
776 err = fib_gw_from_via(cfg, attr, extack);
781 cfg->fc_priority = nla_get_u32(attr);
784 cfg->fc_prefsrc = nla_get_be32(attr);
787 cfg->fc_mx = nla_data(attr);
788 cfg->fc_mx_len = nla_len(attr);
791 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
796 cfg->fc_mp = nla_data(attr);
797 cfg->fc_mp_len = nla_len(attr);
800 cfg->fc_flow = nla_get_u32(attr);
803 cfg->fc_table = nla_get_u32(attr);
806 cfg->fc_encap = attr;
809 cfg->fc_encap_type = nla_get_u16(attr);
810 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
816 cfg->fc_nh_id = nla_get_u32(attr);
822 if (cfg->fc_oif || cfg->fc_gw_family ||
823 cfg->fc_encap || cfg->fc_mp) {
824 NL_SET_ERR_MSG(extack,
825 "Nexthop specification and nexthop id are mutually exclusive");
830 if (has_gw && has_via) {
831 NL_SET_ERR_MSG(extack,
832 "Nexthop configuration can not contain both GATEWAY and VIA");
837 cfg->fc_table = RT_TABLE_MAIN;
844 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
845 struct netlink_ext_ack *extack)
847 struct net *net = sock_net(skb->sk);
848 struct fib_config cfg;
849 struct fib_table *tb;
852 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
856 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
857 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
862 tb = fib_get_table(net, cfg.fc_table);
864 NL_SET_ERR_MSG(extack, "FIB table does not exist");
869 err = fib_table_delete(net, tb, &cfg, extack);
874 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
875 struct netlink_ext_ack *extack)
877 struct net *net = sock_net(skb->sk);
878 struct fib_config cfg;
879 struct fib_table *tb;
882 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
886 tb = fib_new_table(net, cfg.fc_table);
892 err = fib_table_insert(net, tb, &cfg, extack);
893 if (!err && cfg.fc_type == RTN_LOCAL)
894 net->ipv4.fib_has_custom_local_routes = true;
899 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
900 struct fib_dump_filter *filter,
901 struct netlink_callback *cb)
903 struct netlink_ext_ack *extack = cb->extack;
904 struct nlattr *tb[RTA_MAX + 1];
910 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
911 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
915 rtm = nlmsg_data(nlh);
916 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
918 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
922 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
923 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
926 if (rtm->rtm_flags & RTM_F_CLONED)
927 filter->dump_routes = false;
929 filter->dump_exceptions = false;
931 filter->flags = rtm->rtm_flags;
932 filter->protocol = rtm->rtm_protocol;
933 filter->rt_type = rtm->rtm_type;
934 filter->table_id = rtm->rtm_table;
936 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
937 rtm_ipv4_policy, extack);
941 for (i = 0; i <= RTA_MAX; ++i) {
949 filter->table_id = nla_get_u32(tb[i]);
952 ifindex = nla_get_u32(tb[i]);
953 filter->dev = __dev_get_by_index(net, ifindex);
958 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
963 if (filter->flags || filter->protocol || filter->rt_type ||
964 filter->table_id || filter->dev) {
965 filter->filter_set = 1;
966 cb->answer_flags = NLM_F_DUMP_FILTERED;
971 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
973 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
975 struct fib_dump_filter filter = { .dump_routes = true,
976 .dump_exceptions = true };
977 const struct nlmsghdr *nlh = cb->nlh;
978 struct net *net = sock_net(skb->sk);
980 unsigned int e = 0, s_e;
981 struct fib_table *tb;
982 struct hlist_head *head;
985 if (cb->strict_check) {
986 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
989 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
990 struct rtmsg *rtm = nlmsg_data(nlh);
992 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
995 /* ipv4 does not use prefix flag */
996 if (filter.flags & RTM_F_PREFIX)
999 if (filter.table_id) {
1000 tb = fib_get_table(net, filter.table_id);
1002 if (rtnl_msg_family(cb->nlh) != PF_INET)
1005 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1010 err = fib_table_dump(tb, skb, cb, &filter);
1012 return skb->len ? : err;
1020 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1022 head = &net->ipv4.fib_table_hash[h];
1023 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1027 memset(&cb->args[2], 0, sizeof(cb->args) -
1028 2 * sizeof(cb->args[0]));
1029 err = fib_table_dump(tb, skb, cb, &filter);
1031 if (likely(skb->len))
1052 /* Prepare and feed intra-kernel routing request.
1053 * Really, it should be netlink message, but :-( netlink
1054 * can be not configured, so that we feed it directly
1055 * to fib engine. It is legal, because all events occur
1056 * only when netlink is already locked.
1058 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1059 struct in_ifaddr *ifa, u32 rt_priority)
1061 struct net *net = dev_net(ifa->ifa_dev->dev);
1062 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1063 struct fib_table *tb;
1064 struct fib_config cfg = {
1065 .fc_protocol = RTPROT_KERNEL,
1068 .fc_dst_len = dst_len,
1069 .fc_priority = rt_priority,
1070 .fc_prefsrc = ifa->ifa_local,
1071 .fc_oif = ifa->ifa_dev->dev->ifindex,
1072 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1079 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1081 tb = fib_new_table(net, tb_id);
1085 cfg.fc_table = tb->tb_id;
1087 if (type != RTN_LOCAL)
1088 cfg.fc_scope = RT_SCOPE_LINK;
1090 cfg.fc_scope = RT_SCOPE_HOST;
1092 if (cmd == RTM_NEWROUTE)
1093 fib_table_insert(net, tb, &cfg, NULL);
1095 fib_table_delete(net, tb, &cfg, NULL);
1098 void fib_add_ifaddr(struct in_ifaddr *ifa)
1100 struct in_device *in_dev = ifa->ifa_dev;
1101 struct net_device *dev = in_dev->dev;
1102 struct in_ifaddr *prim = ifa;
1103 __be32 mask = ifa->ifa_mask;
1104 __be32 addr = ifa->ifa_local;
1105 __be32 prefix = ifa->ifa_address & mask;
1107 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1108 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1110 pr_warn("%s: bug: prim == NULL\n", __func__);
1115 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1117 if (!(dev->flags & IFF_UP))
1120 /* Add broadcast address, if it is explicitly assigned. */
1121 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1122 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1124 arp_invalidate(dev, ifa->ifa_broadcast, false);
1127 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1128 (prefix != addr || ifa->ifa_prefixlen < 32)) {
1129 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1130 fib_magic(RTM_NEWROUTE,
1131 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1132 prefix, ifa->ifa_prefixlen, prim,
1133 ifa->ifa_rt_priority);
1135 /* Add the network broadcast address, when it makes sense */
1136 if (ifa->ifa_prefixlen < 31) {
1137 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1139 arp_invalidate(dev, prefix | ~mask, false);
1144 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1146 __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1147 struct in_device *in_dev = ifa->ifa_dev;
1148 struct net_device *dev = in_dev->dev;
1150 if (!(dev->flags & IFF_UP) ||
1151 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1152 ipv4_is_zeronet(prefix) ||
1153 (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1157 fib_magic(RTM_NEWROUTE,
1158 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1159 prefix, ifa->ifa_prefixlen, ifa, new_metric);
1161 /* delete the old */
1162 fib_magic(RTM_DELROUTE,
1163 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1164 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1167 /* Delete primary or secondary address.
1168 * Optionally, on secondary address promotion consider the addresses
1169 * from subnet iprim as deleted, even if they are in device list.
1170 * In this case the secondary ifa can be in device list.
1172 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1174 struct in_device *in_dev = ifa->ifa_dev;
1175 struct net_device *dev = in_dev->dev;
1176 struct in_ifaddr *ifa1;
1177 struct in_ifaddr *prim = ifa, *prim1 = NULL;
1178 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1179 __be32 any = ifa->ifa_address & ifa->ifa_mask;
1184 unsigned int ok = 0;
1185 int subnet = 0; /* Primary network */
1186 int gone = 1; /* Address is missing */
1187 int same_prefsrc = 0; /* Another primary with same IP */
1189 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1190 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1192 /* if the device has been deleted, we don't perform
1196 pr_warn("%s: bug: prim == NULL\n", __func__);
1199 if (iprim && iprim != prim) {
1200 pr_warn("%s: bug: iprim != prim\n", __func__);
1203 } else if (!ipv4_is_zeronet(any) &&
1204 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1205 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1206 fib_magic(RTM_DELROUTE,
1207 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1208 any, ifa->ifa_prefixlen, prim, 0);
1215 /* Deletion is more complicated than add.
1216 * We should take care of not to delete too much :-)
1218 * Scan address list to be sure that addresses are really gone.
1221 in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1223 /* promotion, keep the IP */
1227 /* Ignore IFAs from our subnet */
1228 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1229 inet_ifa_match(ifa1->ifa_address, iprim))
1232 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1233 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1234 /* Another address from our subnet? */
1235 if (ifa1->ifa_mask == prim->ifa_mask &&
1236 inet_ifa_match(ifa1->ifa_address, prim))
1239 /* We reached the secondaries, so
1240 * same_prefsrc should be determined.
1244 /* Search new prim1 if ifa1 is not
1245 * using the current prim1
1248 ifa1->ifa_mask != prim1->ifa_mask ||
1249 !inet_ifa_match(ifa1->ifa_address, prim1))
1250 prim1 = inet_ifa_byprefix(in_dev,
1255 if (prim1->ifa_local != prim->ifa_local)
1259 if (prim->ifa_local != ifa1->ifa_local)
1265 if (ifa->ifa_local == ifa1->ifa_local)
1267 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1269 if (brd == ifa1->ifa_broadcast)
1271 if (any == ifa1->ifa_broadcast)
1273 /* primary has network specific broadcasts */
1274 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1275 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1276 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1278 if (!ipv4_is_zeronet(any1)) {
1279 if (ifa->ifa_broadcast == brd1 ||
1280 ifa->ifa_broadcast == any1)
1282 if (brd == brd1 || brd == any1)
1284 if (any == brd1 || any == any1)
1293 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1295 if (subnet && ifa->ifa_prefixlen < 31) {
1296 if (!(ok & BRD1_OK))
1297 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1299 if (!(ok & BRD0_OK))
1300 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1303 if (!(ok & LOCAL_OK)) {
1304 unsigned int addr_type;
1306 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1308 /* Check, that this local address finally disappeared. */
1309 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1311 if (gone && addr_type != RTN_LOCAL) {
1312 /* And the last, but not the least thing.
1313 * We must flush stray FIB entries.
1315 * First of all, we scan fib_info list searching
1316 * for stray nexthop entries, then ignite fib_flush.
1318 if (fib_sync_down_addr(dev, ifa->ifa_local))
1319 fib_flush(dev_net(dev));
1328 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1331 struct fib_result res;
1332 struct flowi4 fl4 = {
1333 .flowi4_mark = frn->fl_mark,
1334 .daddr = frn->fl_addr,
1335 .flowi4_tos = frn->fl_tos,
1336 .flowi4_scope = frn->fl_scope,
1338 struct fib_table *tb;
1342 tb = fib_get_table(net, frn->tb_id_in);
1348 frn->tb_id = tb->tb_id;
1349 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1352 frn->prefixlen = res.prefixlen;
1353 frn->nh_sel = res.nh_sel;
1354 frn->type = res.type;
1355 frn->scope = res.scope;
1363 static void nl_fib_input(struct sk_buff *skb)
1366 struct fib_result_nl *frn;
1367 struct nlmsghdr *nlh;
1370 net = sock_net(skb->sk);
1371 nlh = nlmsg_hdr(skb);
1372 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1373 skb->len < nlh->nlmsg_len ||
1374 nlmsg_len(nlh) < sizeof(*frn))
1377 skb = netlink_skb_clone(skb, GFP_KERNEL);
1380 nlh = nlmsg_hdr(skb);
1382 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1383 nl_fib_lookup(net, frn);
1385 portid = NETLINK_CB(skb).portid; /* netlink portid */
1386 NETLINK_CB(skb).portid = 0; /* from kernel */
1387 NETLINK_CB(skb).dst_group = 0; /* unicast */
1388 nlmsg_unicast(net->ipv4.fibnl, skb, portid);
1391 static int __net_init nl_fib_lookup_init(struct net *net)
1394 struct netlink_kernel_cfg cfg = {
1395 .input = nl_fib_input,
1398 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1400 return -EAFNOSUPPORT;
1401 net->ipv4.fibnl = sk;
1405 static void nl_fib_lookup_exit(struct net *net)
1407 netlink_kernel_release(net->ipv4.fibnl);
1408 net->ipv4.fibnl = NULL;
1411 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1414 if (fib_sync_down_dev(dev, event, force))
1415 fib_flush(dev_net(dev));
1417 rt_cache_flush(dev_net(dev));
1421 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1423 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1424 struct net_device *dev = ifa->ifa_dev->dev;
1425 struct net *net = dev_net(dev);
1429 fib_add_ifaddr(ifa);
1430 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1431 fib_sync_up(dev, RTNH_F_DEAD);
1433 atomic_inc(&net->ipv4.dev_addr_genid);
1434 rt_cache_flush(dev_net(dev));
1437 fib_del_ifaddr(ifa, NULL);
1438 atomic_inc(&net->ipv4.dev_addr_genid);
1439 if (!ifa->ifa_dev->ifa_list) {
1440 /* Last address was deleted from this interface.
1443 fib_disable_ip(dev, event, true);
1445 rt_cache_flush(dev_net(dev));
1452 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1454 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1455 struct netdev_notifier_changeupper_info *upper_info = ptr;
1456 struct netdev_notifier_info_ext *info_ext = ptr;
1457 struct in_device *in_dev;
1458 struct net *net = dev_net(dev);
1459 struct in_ifaddr *ifa;
1462 if (event == NETDEV_UNREGISTER) {
1463 fib_disable_ip(dev, event, true);
1468 in_dev = __in_dev_get_rtnl(dev);
1474 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1475 fib_add_ifaddr(ifa);
1477 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1478 fib_sync_up(dev, RTNH_F_DEAD);
1480 atomic_inc(&net->ipv4.dev_addr_genid);
1481 rt_cache_flush(net);
1484 fib_disable_ip(dev, event, false);
1487 flags = dev_get_flags(dev);
1488 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1489 fib_sync_up(dev, RTNH_F_LINKDOWN);
1491 fib_sync_down_dev(dev, event, false);
1492 rt_cache_flush(net);
1494 case NETDEV_CHANGEMTU:
1495 fib_sync_mtu(dev, info_ext->ext.mtu);
1496 rt_cache_flush(net);
1498 case NETDEV_CHANGEUPPER:
1500 /* flush all routes if dev is linked to or unlinked from
1501 * an L3 master device (e.g., VRF)
1503 if (upper_info->upper_dev &&
1504 netif_is_l3_master(upper_info->upper_dev))
1505 fib_disable_ip(dev, NETDEV_DOWN, true);
1511 static struct notifier_block fib_inetaddr_notifier = {
1512 .notifier_call = fib_inetaddr_event,
1515 static struct notifier_block fib_netdev_notifier = {
1516 .notifier_call = fib_netdev_event,
1519 static int __net_init ip_fib_net_init(struct net *net)
1522 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1524 err = fib4_notifier_init(net);
1528 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1529 /* Default to 3-tuple */
1530 net->ipv4.sysctl_fib_multipath_hash_fields =
1531 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
1534 /* Avoid false sharing : Use at least a full cache line */
1535 size = max_t(size_t, size, L1_CACHE_BYTES);
1537 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1538 if (!net->ipv4.fib_table_hash) {
1540 goto err_table_hash_alloc;
1543 err = fib4_rules_init(net);
1545 goto err_rules_init;
1549 kfree(net->ipv4.fib_table_hash);
1550 err_table_hash_alloc:
1551 fib4_notifier_exit(net);
1555 static void ip_fib_net_exit(struct net *net)
1560 #ifdef CONFIG_IP_MULTIPLE_TABLES
1561 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1562 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1564 /* Destroy the tables in reverse order to guarantee that the
1565 * local table, ID 255, is destroyed before the main table, ID
1566 * 254. This is necessary as the local table may contain
1567 * references to data contained in the main table.
1569 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1570 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1571 struct hlist_node *tmp;
1572 struct fib_table *tb;
1574 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1575 hlist_del(&tb->tb_hlist);
1576 fib_table_flush(net, tb, true);
1581 #ifdef CONFIG_IP_MULTIPLE_TABLES
1582 fib4_rules_exit(net);
1585 kfree(net->ipv4.fib_table_hash);
1586 fib4_notifier_exit(net);
1589 static int __net_init fib_net_init(struct net *net)
1593 #ifdef CONFIG_IP_ROUTE_CLASSID
1594 atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1596 error = ip_fib_net_init(net);
1599 error = nl_fib_lookup_init(net);
1602 error = fib_proc_init(net);
1609 nl_fib_lookup_exit(net);
1611 ip_fib_net_exit(net);
1615 static void __net_exit fib_net_exit(struct net *net)
1618 nl_fib_lookup_exit(net);
1619 ip_fib_net_exit(net);
1622 static struct pernet_operations fib_net_ops = {
1623 .init = fib_net_init,
1624 .exit = fib_net_exit,
1627 void __init ip_fib_init(void)
1631 register_pernet_subsys(&fib_net_ops);
1633 register_netdevice_notifier(&fib_netdev_notifier);
1634 register_inetaddr_notifier(&fib_inetaddr_notifier);
1636 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1637 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1638 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);