2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * IPv4 Forwarding Information Base: FIB frontend.
8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 #include <linux/module.h>
17 #include <linux/uaccess.h>
18 #include <linux/bitops.h>
19 #include <linux/capability.h>
20 #include <linux/types.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/errno.h>
28 #include <linux/inet.h>
29 #include <linux/inetdevice.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_addr.h>
32 #include <linux/if_arp.h>
33 #include <linux/skbuff.h>
34 #include <linux/cache.h>
35 #include <linux/init.h>
36 #include <linux/list.h>
37 #include <linux/slab.h>
40 #include <net/protocol.h>
41 #include <net/route.h>
45 #include <net/ip_fib.h>
46 #include <net/nexthop.h>
47 #include <net/rtnetlink.h>
49 #include <net/l3mdev.h>
50 #include <net/lwtunnel.h>
51 #include <trace/events/fib.h>
53 #ifndef CONFIG_IP_MULTIPLE_TABLES
55 static int __net_init fib4_rules_init(struct net *net)
57 struct fib_table *local_table, *main_table;
59 main_table = fib_trie_table(RT_TABLE_MAIN, NULL);
63 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
67 hlist_add_head_rcu(&local_table->tb_hlist,
68 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
69 hlist_add_head_rcu(&main_table->tb_hlist,
70 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
74 fib_free_table(main_table);
78 static bool fib4_has_custom_rules(struct net *net)
84 struct fib_table *fib_new_table(struct net *net, u32 id)
86 struct fib_table *tb, *alias = NULL;
91 tb = fib_get_table(net, id);
95 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
96 alias = fib_new_table(net, RT_TABLE_MAIN);
98 tb = fib_trie_table(id, alias);
104 rcu_assign_pointer(net->ipv4.fib_main, tb);
106 case RT_TABLE_DEFAULT:
107 rcu_assign_pointer(net->ipv4.fib_default, tb);
113 h = id & (FIB_TABLE_HASHSZ - 1);
114 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
117 EXPORT_SYMBOL_GPL(fib_new_table);
119 /* caller must hold either rtnl or rcu read lock */
120 struct fib_table *fib_get_table(struct net *net, u32 id)
122 struct fib_table *tb;
123 struct hlist_head *head;
128 h = id & (FIB_TABLE_HASHSZ - 1);
130 head = &net->ipv4.fib_table_hash[h];
131 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
138 static bool fib4_has_custom_rules(struct net *net)
140 return net->ipv4.fib_has_custom_rules;
142 #endif /* CONFIG_IP_MULTIPLE_TABLES */
144 static void fib_replace_table(struct net *net, struct fib_table *old,
145 struct fib_table *new)
147 #ifdef CONFIG_IP_MULTIPLE_TABLES
148 switch (new->tb_id) {
150 rcu_assign_pointer(net->ipv4.fib_main, new);
152 case RT_TABLE_DEFAULT:
153 rcu_assign_pointer(net->ipv4.fib_default, new);
160 /* replace the old table in the hlist */
161 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
164 int fib_unmerge(struct net *net)
166 struct fib_table *old, *new, *main_table;
168 /* attempt to fetch local table if it has been allocated */
169 old = fib_get_table(net, RT_TABLE_LOCAL);
173 new = fib_trie_unmerge(old);
177 /* table is already unmerged */
181 /* replace merged table with clean table */
182 fib_replace_table(net, old, new);
185 /* attempt to fetch main table if it has been allocated */
186 main_table = fib_get_table(net, RT_TABLE_MAIN);
190 /* flush local entries from main table */
191 fib_table_flush_external(main_table);
196 void fib_flush(struct net *net)
201 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
202 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
203 struct hlist_node *tmp;
204 struct fib_table *tb;
206 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
207 flushed += fib_table_flush(net, tb, false);
215 * Find address type as if only "dev" was present in the system. If
216 * on_dev is NULL then all interfaces are taken into consideration.
218 static inline unsigned int __inet_dev_addr_type(struct net *net,
219 const struct net_device *dev,
220 __be32 addr, u32 tb_id)
222 struct flowi4 fl4 = { .daddr = addr };
223 struct fib_result res;
224 unsigned int ret = RTN_BROADCAST;
225 struct fib_table *table;
227 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
228 return RTN_BROADCAST;
229 if (ipv4_is_multicast(addr))
230 return RTN_MULTICAST;
234 table = fib_get_table(net, tb_id);
237 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
238 struct fib_nh *nh = fib_info_nh(res.fi, 0);
240 if (!dev || dev == nh->fib_nh_dev)
249 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
251 return __inet_dev_addr_type(net, NULL, addr, tb_id);
253 EXPORT_SYMBOL(inet_addr_type_table);
255 unsigned int inet_addr_type(struct net *net, __be32 addr)
257 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
259 EXPORT_SYMBOL(inet_addr_type);
261 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
264 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
266 return __inet_dev_addr_type(net, dev, addr, rt_table);
268 EXPORT_SYMBOL(inet_dev_addr_type);
270 /* inet_addr_type with dev == NULL but using the table from a dev
271 * if one is associated
273 unsigned int inet_addr_type_dev_table(struct net *net,
274 const struct net_device *dev,
277 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
279 return __inet_dev_addr_type(net, NULL, addr, rt_table);
281 EXPORT_SYMBOL(inet_addr_type_dev_table);
283 __be32 fib_compute_spec_dst(struct sk_buff *skb)
285 struct net_device *dev = skb->dev;
286 struct in_device *in_dev;
287 struct fib_result res;
292 rt = skb_rtable(skb);
293 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
295 return ip_hdr(skb)->daddr;
297 in_dev = __in_dev_get_rcu(dev);
301 scope = RT_SCOPE_UNIVERSE;
302 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
303 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
304 struct flowi4 fl4 = {
305 .flowi4_iif = LOOPBACK_IFINDEX,
306 .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
307 .daddr = ip_hdr(skb)->saddr,
308 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
309 .flowi4_scope = scope,
310 .flowi4_mark = vmark ? skb->mark : 0,
312 if (!fib_lookup(net, &fl4, &res, 0))
313 return fib_result_prefsrc(net, &res);
315 scope = RT_SCOPE_LINK;
318 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
321 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
323 bool dev_match = false;
324 #ifdef CONFIG_IP_ROUTE_MULTIPATH
327 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
328 const struct fib_nh *nh = fib_info_nh(fi, ret);
330 if (nh->fib_nh_dev == dev) {
333 } else if (l3mdev_master_ifindex_rcu(nh->fib_nh_dev) == dev->ifindex) {
339 if (fib_info_nh(fi, 0)->fib_nh_dev == dev)
345 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
347 /* Given (packet source, input interface) and optional (dst, oif, tos):
348 * - (main) check, that source is valid i.e. not broadcast or our local
350 * - figure out what "logical" interface this packet arrived
351 * and calculate "specific destination" address.
352 * - check, that packet arrived from expected physical interface.
353 * called with rcu_read_lock()
355 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
356 u8 tos, int oif, struct net_device *dev,
357 int rpf, struct in_device *idev, u32 *itag)
359 struct net *net = dev_net(dev);
360 struct flow_keys flkeys;
362 struct fib_result res;
367 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
369 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
372 fl4.flowi4_tos = tos;
373 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
374 fl4.flowi4_tun_key.tun_id = 0;
375 fl4.flowi4_flags = 0;
376 fl4.flowi4_uid = sock_net_uid(net, NULL);
378 no_addr = idev->ifa_list == NULL;
380 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
381 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
382 fl4.flowi4_proto = 0;
387 if (fib_lookup(net, &fl4, &res, 0))
389 if (res.type != RTN_UNICAST &&
390 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
392 fib_combine_itag(itag, &res);
394 dev_match = fib_info_nh_uses_dev(res.fi, dev);
396 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
403 fl4.flowi4_oif = dev->ifindex;
406 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
407 if (res.type == RTN_UNICAST)
408 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
424 /* Ignore rp_filter for packets protected by IPsec. */
425 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
426 u8 tos, int oif, struct net_device *dev,
427 struct in_device *idev, u32 *itag)
429 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
430 struct net *net = dev_net(dev);
432 if (!r && !fib_num_tclassid_users(net) &&
433 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
434 if (IN_DEV_ACCEPT_LOCAL(idev))
436 /* with custom local routes in place, checking local addresses
437 * only will be too optimistic, with custom rules, checking
438 * local addresses only can be too strict, e.g. due to vrf
440 if (net->ipv4.fib_has_custom_local_routes ||
441 fib4_has_custom_rules(net))
443 if (inet_lookup_ifaddr_rcu(net, src))
452 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
455 static inline __be32 sk_extract_addr(struct sockaddr *addr)
457 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
460 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
464 nla = (struct nlattr *) ((char *) mx + len);
465 nla->nla_type = type;
466 nla->nla_len = nla_attr_size(4);
467 *(u32 *) nla_data(nla) = value;
469 return len + nla_total_size(4);
472 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
473 struct fib_config *cfg)
478 memset(cfg, 0, sizeof(*cfg));
479 cfg->fc_nlinfo.nl_net = net;
481 if (rt->rt_dst.sa_family != AF_INET)
482 return -EAFNOSUPPORT;
485 * Check mask for validity:
486 * a) it must be contiguous.
487 * b) destination must have all host bits clear.
488 * c) if application forgot to set correct family (AF_INET),
489 * reject request unless it is absolutely clear i.e.
490 * both family and mask are zero.
493 addr = sk_extract_addr(&rt->rt_dst);
494 if (!(rt->rt_flags & RTF_HOST)) {
495 __be32 mask = sk_extract_addr(&rt->rt_genmask);
497 if (rt->rt_genmask.sa_family != AF_INET) {
498 if (mask || rt->rt_genmask.sa_family)
499 return -EAFNOSUPPORT;
502 if (bad_mask(mask, addr))
505 plen = inet_mask_len(mask);
508 cfg->fc_dst_len = plen;
511 if (cmd != SIOCDELRT) {
512 cfg->fc_nlflags = NLM_F_CREATE;
513 cfg->fc_protocol = RTPROT_BOOT;
517 cfg->fc_priority = rt->rt_metric - 1;
519 if (rt->rt_flags & RTF_REJECT) {
520 cfg->fc_scope = RT_SCOPE_HOST;
521 cfg->fc_type = RTN_UNREACHABLE;
525 cfg->fc_scope = RT_SCOPE_NOWHERE;
526 cfg->fc_type = RTN_UNICAST;
530 struct net_device *dev;
531 char devname[IFNAMSIZ];
533 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
536 devname[IFNAMSIZ-1] = 0;
537 colon = strchr(devname, ':');
540 dev = __dev_get_by_name(net, devname);
543 cfg->fc_oif = dev->ifindex;
544 cfg->fc_table = l3mdev_fib_table(dev);
546 const struct in_ifaddr *ifa;
547 struct in_device *in_dev;
549 in_dev = __in_dev_get_rtnl(dev);
556 in_dev_for_each_ifa_rcu(ifa, in_dev) {
557 if (strcmp(ifa->ifa_label, devname) == 0)
564 cfg->fc_prefsrc = ifa->ifa_local;
568 addr = sk_extract_addr(&rt->rt_gateway);
569 if (rt->rt_gateway.sa_family == AF_INET && addr) {
570 unsigned int addr_type;
573 cfg->fc_gw_family = AF_INET;
574 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
575 if (rt->rt_flags & RTF_GATEWAY &&
576 addr_type == RTN_UNICAST)
577 cfg->fc_scope = RT_SCOPE_UNIVERSE;
580 if (cmd == SIOCDELRT)
583 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
586 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
587 cfg->fc_scope = RT_SCOPE_LINK;
589 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
593 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
597 if (rt->rt_flags & RTF_MTU)
598 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
600 if (rt->rt_flags & RTF_WINDOW)
601 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
603 if (rt->rt_flags & RTF_IRTT)
604 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
607 cfg->fc_mx_len = len;
614 * Handle IP routing ioctl calls.
615 * These are used to manipulate the routing tables
617 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
619 struct fib_config cfg;
623 case SIOCADDRT: /* Add a route */
624 case SIOCDELRT: /* Delete a route */
625 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
629 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
631 struct fib_table *tb;
633 if (cmd == SIOCDELRT) {
634 tb = fib_get_table(net, cfg.fc_table);
636 err = fib_table_delete(net, tb, &cfg,
641 tb = fib_new_table(net, cfg.fc_table);
643 err = fib_table_insert(net, tb,
649 /* allocated by rtentry_to_fib_config() */
658 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
659 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 },
660 [RTA_DST] = { .type = NLA_U32 },
661 [RTA_SRC] = { .type = NLA_U32 },
662 [RTA_IIF] = { .type = NLA_U32 },
663 [RTA_OIF] = { .type = NLA_U32 },
664 [RTA_GATEWAY] = { .type = NLA_U32 },
665 [RTA_PRIORITY] = { .type = NLA_U32 },
666 [RTA_PREFSRC] = { .type = NLA_U32 },
667 [RTA_METRICS] = { .type = NLA_NESTED },
668 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
669 [RTA_FLOW] = { .type = NLA_U32 },
670 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
671 [RTA_ENCAP] = { .type = NLA_NESTED },
672 [RTA_UID] = { .type = NLA_U32 },
673 [RTA_MARK] = { .type = NLA_U32 },
674 [RTA_TABLE] = { .type = NLA_U32 },
675 [RTA_IP_PROTO] = { .type = NLA_U8 },
676 [RTA_SPORT] = { .type = NLA_U16 },
677 [RTA_DPORT] = { .type = NLA_U16 },
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);
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,
818 if (has_gw && has_via) {
819 NL_SET_ERR_MSG(extack,
820 "Nexthop configuration can not contain both GATEWAY and VIA");
829 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
830 struct netlink_ext_ack *extack)
832 struct net *net = sock_net(skb->sk);
833 struct fib_config cfg;
834 struct fib_table *tb;
837 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
841 tb = fib_get_table(net, cfg.fc_table);
843 NL_SET_ERR_MSG(extack, "FIB table does not exist");
848 err = fib_table_delete(net, tb, &cfg, extack);
853 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
854 struct netlink_ext_ack *extack)
856 struct net *net = sock_net(skb->sk);
857 struct fib_config cfg;
858 struct fib_table *tb;
861 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
865 tb = fib_new_table(net, cfg.fc_table);
871 err = fib_table_insert(net, tb, &cfg, extack);
872 if (!err && cfg.fc_type == RTN_LOCAL)
873 net->ipv4.fib_has_custom_local_routes = true;
878 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
879 struct fib_dump_filter *filter,
880 struct netlink_callback *cb)
882 struct netlink_ext_ack *extack = cb->extack;
883 struct nlattr *tb[RTA_MAX + 1];
889 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
890 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
894 rtm = nlmsg_data(nlh);
895 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
897 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
900 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
901 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
905 filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
906 filter->flags = rtm->rtm_flags;
907 filter->protocol = rtm->rtm_protocol;
908 filter->rt_type = rtm->rtm_type;
909 filter->table_id = rtm->rtm_table;
911 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
912 rtm_ipv4_policy, extack);
916 for (i = 0; i <= RTA_MAX; ++i) {
924 filter->table_id = nla_get_u32(tb[i]);
927 ifindex = nla_get_u32(tb[i]);
928 filter->dev = __dev_get_by_index(net, ifindex);
933 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
938 if (filter->flags || filter->protocol || filter->rt_type ||
939 filter->table_id || filter->dev) {
940 filter->filter_set = 1;
941 cb->answer_flags = NLM_F_DUMP_FILTERED;
946 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
948 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
950 const struct nlmsghdr *nlh = cb->nlh;
951 struct net *net = sock_net(skb->sk);
952 struct fib_dump_filter filter = {};
954 unsigned int e = 0, s_e;
955 struct fib_table *tb;
956 struct hlist_head *head;
959 if (cb->strict_check) {
960 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
963 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
964 struct rtmsg *rtm = nlmsg_data(nlh);
966 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
969 /* fib entries are never clones and ipv4 does not use prefix flag */
970 if (filter.flags & (RTM_F_PREFIX | RTM_F_CLONED))
973 if (filter.table_id) {
974 tb = fib_get_table(net, filter.table_id);
976 if (filter.dump_all_families)
979 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
983 err = fib_table_dump(tb, skb, cb, &filter);
984 return skb->len ? : err;
992 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
994 head = &net->ipv4.fib_table_hash[h];
995 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
999 memset(&cb->args[2], 0, sizeof(cb->args) -
1000 2 * sizeof(cb->args[0]));
1001 err = fib_table_dump(tb, skb, cb, &filter);
1003 if (likely(skb->len))
1024 /* Prepare and feed intra-kernel routing request.
1025 * Really, it should be netlink message, but :-( netlink
1026 * can be not configured, so that we feed it directly
1027 * to fib engine. It is legal, because all events occur
1028 * only when netlink is already locked.
1030 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1031 struct in_ifaddr *ifa, u32 rt_priority)
1033 struct net *net = dev_net(ifa->ifa_dev->dev);
1034 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1035 struct fib_table *tb;
1036 struct fib_config cfg = {
1037 .fc_protocol = RTPROT_KERNEL,
1040 .fc_dst_len = dst_len,
1041 .fc_priority = rt_priority,
1042 .fc_prefsrc = ifa->ifa_local,
1043 .fc_oif = ifa->ifa_dev->dev->ifindex,
1044 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1051 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1053 tb = fib_new_table(net, tb_id);
1057 cfg.fc_table = tb->tb_id;
1059 if (type != RTN_LOCAL)
1060 cfg.fc_scope = RT_SCOPE_LINK;
1062 cfg.fc_scope = RT_SCOPE_HOST;
1064 if (cmd == RTM_NEWROUTE)
1065 fib_table_insert(net, tb, &cfg, NULL);
1067 fib_table_delete(net, tb, &cfg, NULL);
1070 void fib_add_ifaddr(struct in_ifaddr *ifa)
1072 struct in_device *in_dev = ifa->ifa_dev;
1073 struct net_device *dev = in_dev->dev;
1074 struct in_ifaddr *prim = ifa;
1075 __be32 mask = ifa->ifa_mask;
1076 __be32 addr = ifa->ifa_local;
1077 __be32 prefix = ifa->ifa_address & mask;
1079 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1080 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1082 pr_warn("%s: bug: prim == NULL\n", __func__);
1087 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1089 if (!(dev->flags & IFF_UP))
1092 /* Add broadcast address, if it is explicitly assigned. */
1093 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1094 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1097 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1098 (prefix != addr || ifa->ifa_prefixlen < 32)) {
1099 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1100 fib_magic(RTM_NEWROUTE,
1101 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1102 prefix, ifa->ifa_prefixlen, prim,
1103 ifa->ifa_rt_priority);
1105 /* Add network specific broadcasts, when it takes a sense */
1106 if (ifa->ifa_prefixlen < 31) {
1107 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1109 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1115 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1117 __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1118 struct in_device *in_dev = ifa->ifa_dev;
1119 struct net_device *dev = in_dev->dev;
1121 if (!(dev->flags & IFF_UP) ||
1122 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1123 ipv4_is_zeronet(prefix) ||
1124 prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1128 fib_magic(RTM_NEWROUTE,
1129 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1130 prefix, ifa->ifa_prefixlen, ifa, new_metric);
1132 /* delete the old */
1133 fib_magic(RTM_DELROUTE,
1134 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1135 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1138 /* Delete primary or secondary address.
1139 * Optionally, on secondary address promotion consider the addresses
1140 * from subnet iprim as deleted, even if they are in device list.
1141 * In this case the secondary ifa can be in device list.
1143 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1145 struct in_device *in_dev = ifa->ifa_dev;
1146 struct net_device *dev = in_dev->dev;
1147 struct in_ifaddr *ifa1;
1148 struct in_ifaddr *prim = ifa, *prim1 = NULL;
1149 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1150 __be32 any = ifa->ifa_address & ifa->ifa_mask;
1155 unsigned int ok = 0;
1156 int subnet = 0; /* Primary network */
1157 int gone = 1; /* Address is missing */
1158 int same_prefsrc = 0; /* Another primary with same IP */
1160 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1161 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1163 /* if the device has been deleted, we don't perform
1167 pr_warn("%s: bug: prim == NULL\n", __func__);
1170 if (iprim && iprim != prim) {
1171 pr_warn("%s: bug: iprim != prim\n", __func__);
1174 } else if (!ipv4_is_zeronet(any) &&
1175 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1176 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1177 fib_magic(RTM_DELROUTE,
1178 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1179 any, ifa->ifa_prefixlen, prim, 0);
1186 /* Deletion is more complicated than add.
1187 * We should take care of not to delete too much :-)
1189 * Scan address list to be sure that addresses are really gone.
1192 in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1194 /* promotion, keep the IP */
1198 /* Ignore IFAs from our subnet */
1199 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1200 inet_ifa_match(ifa1->ifa_address, iprim))
1203 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1204 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1205 /* Another address from our subnet? */
1206 if (ifa1->ifa_mask == prim->ifa_mask &&
1207 inet_ifa_match(ifa1->ifa_address, prim))
1210 /* We reached the secondaries, so
1211 * same_prefsrc should be determined.
1215 /* Search new prim1 if ifa1 is not
1216 * using the current prim1
1219 ifa1->ifa_mask != prim1->ifa_mask ||
1220 !inet_ifa_match(ifa1->ifa_address, prim1))
1221 prim1 = inet_ifa_byprefix(in_dev,
1226 if (prim1->ifa_local != prim->ifa_local)
1230 if (prim->ifa_local != ifa1->ifa_local)
1236 if (ifa->ifa_local == ifa1->ifa_local)
1238 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1240 if (brd == ifa1->ifa_broadcast)
1242 if (any == ifa1->ifa_broadcast)
1244 /* primary has network specific broadcasts */
1245 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1246 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1247 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1249 if (!ipv4_is_zeronet(any1)) {
1250 if (ifa->ifa_broadcast == brd1 ||
1251 ifa->ifa_broadcast == any1)
1253 if (brd == brd1 || brd == any1)
1255 if (any == brd1 || any == any1)
1264 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1266 if (subnet && ifa->ifa_prefixlen < 31) {
1267 if (!(ok & BRD1_OK))
1268 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1270 if (!(ok & BRD0_OK))
1271 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1274 if (!(ok & LOCAL_OK)) {
1275 unsigned int addr_type;
1277 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1279 /* Check, that this local address finally disappeared. */
1280 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1282 if (gone && addr_type != RTN_LOCAL) {
1283 /* And the last, but not the least thing.
1284 * We must flush stray FIB entries.
1286 * First of all, we scan fib_info list searching
1287 * for stray nexthop entries, then ignite fib_flush.
1289 if (fib_sync_down_addr(dev, ifa->ifa_local))
1290 fib_flush(dev_net(dev));
1299 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1302 struct fib_result res;
1303 struct flowi4 fl4 = {
1304 .flowi4_mark = frn->fl_mark,
1305 .daddr = frn->fl_addr,
1306 .flowi4_tos = frn->fl_tos,
1307 .flowi4_scope = frn->fl_scope,
1309 struct fib_table *tb;
1313 tb = fib_get_table(net, frn->tb_id_in);
1319 frn->tb_id = tb->tb_id;
1320 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1323 frn->prefixlen = res.prefixlen;
1324 frn->nh_sel = res.nh_sel;
1325 frn->type = res.type;
1326 frn->scope = res.scope;
1334 static void nl_fib_input(struct sk_buff *skb)
1337 struct fib_result_nl *frn;
1338 struct nlmsghdr *nlh;
1341 net = sock_net(skb->sk);
1342 nlh = nlmsg_hdr(skb);
1343 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1344 skb->len < nlh->nlmsg_len ||
1345 nlmsg_len(nlh) < sizeof(*frn))
1348 skb = netlink_skb_clone(skb, GFP_KERNEL);
1351 nlh = nlmsg_hdr(skb);
1353 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1354 nl_fib_lookup(net, frn);
1356 portid = NETLINK_CB(skb).portid; /* netlink portid */
1357 NETLINK_CB(skb).portid = 0; /* from kernel */
1358 NETLINK_CB(skb).dst_group = 0; /* unicast */
1359 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1362 static int __net_init nl_fib_lookup_init(struct net *net)
1365 struct netlink_kernel_cfg cfg = {
1366 .input = nl_fib_input,
1369 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1371 return -EAFNOSUPPORT;
1372 net->ipv4.fibnl = sk;
1376 static void nl_fib_lookup_exit(struct net *net)
1378 netlink_kernel_release(net->ipv4.fibnl);
1379 net->ipv4.fibnl = NULL;
1382 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1385 if (fib_sync_down_dev(dev, event, force))
1386 fib_flush(dev_net(dev));
1388 rt_cache_flush(dev_net(dev));
1392 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1394 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1395 struct net_device *dev = ifa->ifa_dev->dev;
1396 struct net *net = dev_net(dev);
1400 fib_add_ifaddr(ifa);
1401 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1402 fib_sync_up(dev, RTNH_F_DEAD);
1404 atomic_inc(&net->ipv4.dev_addr_genid);
1405 rt_cache_flush(dev_net(dev));
1408 fib_del_ifaddr(ifa, NULL);
1409 atomic_inc(&net->ipv4.dev_addr_genid);
1410 if (!ifa->ifa_dev->ifa_list) {
1411 /* Last address was deleted from this interface.
1414 fib_disable_ip(dev, event, true);
1416 rt_cache_flush(dev_net(dev));
1423 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1425 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1426 struct netdev_notifier_changeupper_info *upper_info = ptr;
1427 struct netdev_notifier_info_ext *info_ext = ptr;
1428 struct in_device *in_dev;
1429 struct net *net = dev_net(dev);
1430 struct in_ifaddr *ifa;
1433 if (event == NETDEV_UNREGISTER) {
1434 fib_disable_ip(dev, event, true);
1439 in_dev = __in_dev_get_rtnl(dev);
1445 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1446 fib_add_ifaddr(ifa);
1448 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1449 fib_sync_up(dev, RTNH_F_DEAD);
1451 atomic_inc(&net->ipv4.dev_addr_genid);
1452 rt_cache_flush(net);
1455 fib_disable_ip(dev, event, false);
1458 flags = dev_get_flags(dev);
1459 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1460 fib_sync_up(dev, RTNH_F_LINKDOWN);
1462 fib_sync_down_dev(dev, event, false);
1463 rt_cache_flush(net);
1465 case NETDEV_CHANGEMTU:
1466 fib_sync_mtu(dev, info_ext->ext.mtu);
1467 rt_cache_flush(net);
1469 case NETDEV_CHANGEUPPER:
1471 /* flush all routes if dev is linked to or unlinked from
1472 * an L3 master device (e.g., VRF)
1474 if (upper_info->upper_dev &&
1475 netif_is_l3_master(upper_info->upper_dev))
1476 fib_disable_ip(dev, NETDEV_DOWN, true);
1482 static struct notifier_block fib_inetaddr_notifier = {
1483 .notifier_call = fib_inetaddr_event,
1486 static struct notifier_block fib_netdev_notifier = {
1487 .notifier_call = fib_netdev_event,
1490 static int __net_init ip_fib_net_init(struct net *net)
1493 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1495 err = fib4_notifier_init(net);
1499 /* Avoid false sharing : Use at least a full cache line */
1500 size = max_t(size_t, size, L1_CACHE_BYTES);
1502 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1503 if (!net->ipv4.fib_table_hash) {
1505 goto err_table_hash_alloc;
1508 err = fib4_rules_init(net);
1510 goto err_rules_init;
1514 kfree(net->ipv4.fib_table_hash);
1515 err_table_hash_alloc:
1516 fib4_notifier_exit(net);
1520 static void ip_fib_net_exit(struct net *net)
1525 #ifdef CONFIG_IP_MULTIPLE_TABLES
1526 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1527 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1529 /* Destroy the tables in reverse order to guarantee that the
1530 * local table, ID 255, is destroyed before the main table, ID
1531 * 254. This is necessary as the local table may contain
1532 * references to data contained in the main table.
1534 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1535 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1536 struct hlist_node *tmp;
1537 struct fib_table *tb;
1539 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1540 hlist_del(&tb->tb_hlist);
1541 fib_table_flush(net, tb, true);
1546 #ifdef CONFIG_IP_MULTIPLE_TABLES
1547 fib4_rules_exit(net);
1550 kfree(net->ipv4.fib_table_hash);
1551 fib4_notifier_exit(net);
1554 static int __net_init fib_net_init(struct net *net)
1558 #ifdef CONFIG_IP_ROUTE_CLASSID
1559 net->ipv4.fib_num_tclassid_users = 0;
1561 error = ip_fib_net_init(net);
1564 error = nl_fib_lookup_init(net);
1567 error = fib_proc_init(net);
1574 nl_fib_lookup_exit(net);
1576 ip_fib_net_exit(net);
1580 static void __net_exit fib_net_exit(struct net *net)
1583 nl_fib_lookup_exit(net);
1584 ip_fib_net_exit(net);
1587 static struct pernet_operations fib_net_ops = {
1588 .init = fib_net_init,
1589 .exit = fib_net_exit,
1592 void __init ip_fib_init(void)
1596 register_pernet_subsys(&fib_net_ops);
1598 register_netdevice_notifier(&fib_netdev_notifier);
1599 register_inetaddr_notifier(&fib_inetaddr_notifier);
1601 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1602 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1603 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);