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;
576 if (cmd == SIOCDELRT)
579 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
582 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
583 cfg->fc_scope = RT_SCOPE_LINK;
585 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
589 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
593 if (rt->rt_flags & RTF_MTU)
594 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
596 if (rt->rt_flags & RTF_WINDOW)
597 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
599 if (rt->rt_flags & RTF_IRTT)
600 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
603 cfg->fc_mx_len = len;
610 * Handle IP routing ioctl calls.
611 * These are used to manipulate the routing tables
613 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
615 struct fib_config cfg;
619 case SIOCADDRT: /* Add a route */
620 case SIOCDELRT: /* Delete a route */
621 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
625 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
627 struct fib_table *tb;
629 if (cmd == SIOCDELRT) {
630 tb = fib_get_table(net, cfg.fc_table);
632 err = fib_table_delete(net, tb, &cfg,
637 tb = fib_new_table(net, cfg.fc_table);
639 err = fib_table_insert(net, tb,
645 /* allocated by rtentry_to_fib_config() */
654 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
655 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 },
656 [RTA_DST] = { .type = NLA_U32 },
657 [RTA_SRC] = { .type = NLA_U32 },
658 [RTA_IIF] = { .type = NLA_U32 },
659 [RTA_OIF] = { .type = NLA_U32 },
660 [RTA_GATEWAY] = { .type = NLA_U32 },
661 [RTA_PRIORITY] = { .type = NLA_U32 },
662 [RTA_PREFSRC] = { .type = NLA_U32 },
663 [RTA_METRICS] = { .type = NLA_NESTED },
664 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
665 [RTA_FLOW] = { .type = NLA_U32 },
666 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
667 [RTA_ENCAP] = { .type = NLA_NESTED },
668 [RTA_UID] = { .type = NLA_U32 },
669 [RTA_MARK] = { .type = NLA_U32 },
670 [RTA_TABLE] = { .type = NLA_U32 },
671 [RTA_IP_PROTO] = { .type = NLA_U8 },
672 [RTA_SPORT] = { .type = NLA_U16 },
673 [RTA_DPORT] = { .type = NLA_U16 },
674 [RTA_NH_ID] = { .type = NLA_U32 },
677 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
678 struct netlink_ext_ack *extack)
683 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
684 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
689 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
691 switch (via->rtvia_family) {
693 if (alen != sizeof(__be32)) {
694 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
697 cfg->fc_gw_family = AF_INET;
698 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
701 #if IS_ENABLED(CONFIG_IPV6)
702 if (alen != sizeof(struct in6_addr)) {
703 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
706 cfg->fc_gw_family = AF_INET6;
707 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
709 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
714 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
721 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
722 struct nlmsghdr *nlh, struct fib_config *cfg,
723 struct netlink_ext_ack *extack)
725 bool has_gw = false, has_via = false;
730 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
731 rtm_ipv4_policy, extack);
735 memset(cfg, 0, sizeof(*cfg));
737 rtm = nlmsg_data(nlh);
738 cfg->fc_dst_len = rtm->rtm_dst_len;
739 cfg->fc_tos = rtm->rtm_tos;
740 cfg->fc_table = rtm->rtm_table;
741 cfg->fc_protocol = rtm->rtm_protocol;
742 cfg->fc_scope = rtm->rtm_scope;
743 cfg->fc_type = rtm->rtm_type;
744 cfg->fc_flags = rtm->rtm_flags;
745 cfg->fc_nlflags = nlh->nlmsg_flags;
747 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
748 cfg->fc_nlinfo.nlh = nlh;
749 cfg->fc_nlinfo.nl_net = net;
751 if (cfg->fc_type > RTN_MAX) {
752 NL_SET_ERR_MSG(extack, "Invalid route type");
757 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
758 switch (nla_type(attr)) {
760 cfg->fc_dst = nla_get_be32(attr);
763 cfg->fc_oif = nla_get_u32(attr);
767 cfg->fc_gw4 = nla_get_be32(attr);
769 cfg->fc_gw_family = AF_INET;
773 err = fib_gw_from_via(cfg, attr, extack);
778 cfg->fc_priority = nla_get_u32(attr);
781 cfg->fc_prefsrc = nla_get_be32(attr);
784 cfg->fc_mx = nla_data(attr);
785 cfg->fc_mx_len = nla_len(attr);
788 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
793 cfg->fc_mp = nla_data(attr);
794 cfg->fc_mp_len = nla_len(attr);
797 cfg->fc_flow = nla_get_u32(attr);
800 cfg->fc_table = nla_get_u32(attr);
803 cfg->fc_encap = attr;
806 cfg->fc_encap_type = nla_get_u16(attr);
807 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
813 cfg->fc_nh_id = nla_get_u32(attr);
819 if (cfg->fc_oif || cfg->fc_gw_family ||
820 cfg->fc_encap || cfg->fc_mp) {
821 NL_SET_ERR_MSG(extack,
822 "Nexthop specification and nexthop id are mutually exclusive");
827 if (has_gw && has_via) {
828 NL_SET_ERR_MSG(extack,
829 "Nexthop configuration can not contain both GATEWAY and VIA");
838 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
839 struct netlink_ext_ack *extack)
841 struct net *net = sock_net(skb->sk);
842 struct fib_config cfg;
843 struct fib_table *tb;
846 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
850 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
851 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
856 tb = fib_get_table(net, cfg.fc_table);
858 NL_SET_ERR_MSG(extack, "FIB table does not exist");
863 err = fib_table_delete(net, tb, &cfg, extack);
868 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
869 struct netlink_ext_ack *extack)
871 struct net *net = sock_net(skb->sk);
872 struct fib_config cfg;
873 struct fib_table *tb;
876 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
880 tb = fib_new_table(net, cfg.fc_table);
886 err = fib_table_insert(net, tb, &cfg, extack);
887 if (!err && cfg.fc_type == RTN_LOCAL)
888 net->ipv4.fib_has_custom_local_routes = true;
893 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
894 struct fib_dump_filter *filter,
895 struct netlink_callback *cb)
897 struct netlink_ext_ack *extack = cb->extack;
898 struct nlattr *tb[RTA_MAX + 1];
904 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
905 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
909 rtm = nlmsg_data(nlh);
910 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
912 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
916 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
917 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
920 if (rtm->rtm_flags & RTM_F_CLONED)
921 filter->dump_routes = false;
923 filter->dump_exceptions = false;
925 filter->flags = rtm->rtm_flags;
926 filter->protocol = rtm->rtm_protocol;
927 filter->rt_type = rtm->rtm_type;
928 filter->table_id = rtm->rtm_table;
930 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
931 rtm_ipv4_policy, extack);
935 for (i = 0; i <= RTA_MAX; ++i) {
943 filter->table_id = nla_get_u32(tb[i]);
946 ifindex = nla_get_u32(tb[i]);
947 filter->dev = __dev_get_by_index(net, ifindex);
952 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
957 if (filter->flags || filter->protocol || filter->rt_type ||
958 filter->table_id || filter->dev) {
959 filter->filter_set = 1;
960 cb->answer_flags = NLM_F_DUMP_FILTERED;
965 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
967 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
969 struct fib_dump_filter filter = { .dump_routes = true,
970 .dump_exceptions = true };
971 const struct nlmsghdr *nlh = cb->nlh;
972 struct net *net = sock_net(skb->sk);
974 unsigned int e = 0, s_e;
975 struct fib_table *tb;
976 struct hlist_head *head;
979 if (cb->strict_check) {
980 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
983 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
984 struct rtmsg *rtm = nlmsg_data(nlh);
986 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
989 /* ipv4 does not use prefix flag */
990 if (filter.flags & RTM_F_PREFIX)
993 if (filter.table_id) {
994 tb = fib_get_table(net, filter.table_id);
996 if (rtnl_msg_family(cb->nlh) != PF_INET)
999 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1004 err = fib_table_dump(tb, skb, cb, &filter);
1006 return skb->len ? : err;
1014 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1016 head = &net->ipv4.fib_table_hash[h];
1017 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1021 memset(&cb->args[2], 0, sizeof(cb->args) -
1022 2 * sizeof(cb->args[0]));
1023 err = fib_table_dump(tb, skb, cb, &filter);
1025 if (likely(skb->len))
1046 /* Prepare and feed intra-kernel routing request.
1047 * Really, it should be netlink message, but :-( netlink
1048 * can be not configured, so that we feed it directly
1049 * to fib engine. It is legal, because all events occur
1050 * only when netlink is already locked.
1052 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1053 struct in_ifaddr *ifa, u32 rt_priority)
1055 struct net *net = dev_net(ifa->ifa_dev->dev);
1056 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1057 struct fib_table *tb;
1058 struct fib_config cfg = {
1059 .fc_protocol = RTPROT_KERNEL,
1062 .fc_dst_len = dst_len,
1063 .fc_priority = rt_priority,
1064 .fc_prefsrc = ifa->ifa_local,
1065 .fc_oif = ifa->ifa_dev->dev->ifindex,
1066 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1073 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1075 tb = fib_new_table(net, tb_id);
1079 cfg.fc_table = tb->tb_id;
1081 if (type != RTN_LOCAL)
1082 cfg.fc_scope = RT_SCOPE_LINK;
1084 cfg.fc_scope = RT_SCOPE_HOST;
1086 if (cmd == RTM_NEWROUTE)
1087 fib_table_insert(net, tb, &cfg, NULL);
1089 fib_table_delete(net, tb, &cfg, NULL);
1092 void fib_add_ifaddr(struct in_ifaddr *ifa)
1094 struct in_device *in_dev = ifa->ifa_dev;
1095 struct net_device *dev = in_dev->dev;
1096 struct in_ifaddr *prim = ifa;
1097 __be32 mask = ifa->ifa_mask;
1098 __be32 addr = ifa->ifa_local;
1099 __be32 prefix = ifa->ifa_address & mask;
1101 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1102 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1104 pr_warn("%s: bug: prim == NULL\n", __func__);
1109 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1111 if (!(dev->flags & IFF_UP))
1114 /* Add broadcast address, if it is explicitly assigned. */
1115 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1116 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1118 arp_invalidate(dev, ifa->ifa_broadcast, false);
1121 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1122 (prefix != addr || ifa->ifa_prefixlen < 32)) {
1123 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1124 fib_magic(RTM_NEWROUTE,
1125 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1126 prefix, ifa->ifa_prefixlen, prim,
1127 ifa->ifa_rt_priority);
1129 /* Add network specific broadcasts, when it takes a sense */
1130 if (ifa->ifa_prefixlen < 31) {
1131 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1133 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1135 arp_invalidate(dev, prefix | ~mask, false);
1140 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1142 __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1143 struct in_device *in_dev = ifa->ifa_dev;
1144 struct net_device *dev = in_dev->dev;
1146 if (!(dev->flags & IFF_UP) ||
1147 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1148 ipv4_is_zeronet(prefix) ||
1149 (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1153 fib_magic(RTM_NEWROUTE,
1154 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1155 prefix, ifa->ifa_prefixlen, ifa, new_metric);
1157 /* delete the old */
1158 fib_magic(RTM_DELROUTE,
1159 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1160 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1163 /* Delete primary or secondary address.
1164 * Optionally, on secondary address promotion consider the addresses
1165 * from subnet iprim as deleted, even if they are in device list.
1166 * In this case the secondary ifa can be in device list.
1168 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1170 struct in_device *in_dev = ifa->ifa_dev;
1171 struct net_device *dev = in_dev->dev;
1172 struct in_ifaddr *ifa1;
1173 struct in_ifaddr *prim = ifa, *prim1 = NULL;
1174 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1175 __be32 any = ifa->ifa_address & ifa->ifa_mask;
1180 unsigned int ok = 0;
1181 int subnet = 0; /* Primary network */
1182 int gone = 1; /* Address is missing */
1183 int same_prefsrc = 0; /* Another primary with same IP */
1185 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1186 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1188 /* if the device has been deleted, we don't perform
1192 pr_warn("%s: bug: prim == NULL\n", __func__);
1195 if (iprim && iprim != prim) {
1196 pr_warn("%s: bug: iprim != prim\n", __func__);
1199 } else if (!ipv4_is_zeronet(any) &&
1200 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1201 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1202 fib_magic(RTM_DELROUTE,
1203 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1204 any, ifa->ifa_prefixlen, prim, 0);
1211 /* Deletion is more complicated than add.
1212 * We should take care of not to delete too much :-)
1214 * Scan address list to be sure that addresses are really gone.
1217 in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1219 /* promotion, keep the IP */
1223 /* Ignore IFAs from our subnet */
1224 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1225 inet_ifa_match(ifa1->ifa_address, iprim))
1228 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1229 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1230 /* Another address from our subnet? */
1231 if (ifa1->ifa_mask == prim->ifa_mask &&
1232 inet_ifa_match(ifa1->ifa_address, prim))
1235 /* We reached the secondaries, so
1236 * same_prefsrc should be determined.
1240 /* Search new prim1 if ifa1 is not
1241 * using the current prim1
1244 ifa1->ifa_mask != prim1->ifa_mask ||
1245 !inet_ifa_match(ifa1->ifa_address, prim1))
1246 prim1 = inet_ifa_byprefix(in_dev,
1251 if (prim1->ifa_local != prim->ifa_local)
1255 if (prim->ifa_local != ifa1->ifa_local)
1261 if (ifa->ifa_local == ifa1->ifa_local)
1263 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1265 if (brd == ifa1->ifa_broadcast)
1267 if (any == ifa1->ifa_broadcast)
1269 /* primary has network specific broadcasts */
1270 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1271 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1272 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1274 if (!ipv4_is_zeronet(any1)) {
1275 if (ifa->ifa_broadcast == brd1 ||
1276 ifa->ifa_broadcast == any1)
1278 if (brd == brd1 || brd == any1)
1280 if (any == brd1 || any == any1)
1289 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1291 if (subnet && ifa->ifa_prefixlen < 31) {
1292 if (!(ok & BRD1_OK))
1293 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1295 if (!(ok & BRD0_OK))
1296 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1299 if (!(ok & LOCAL_OK)) {
1300 unsigned int addr_type;
1302 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1304 /* Check, that this local address finally disappeared. */
1305 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1307 if (gone && addr_type != RTN_LOCAL) {
1308 /* And the last, but not the least thing.
1309 * We must flush stray FIB entries.
1311 * First of all, we scan fib_info list searching
1312 * for stray nexthop entries, then ignite fib_flush.
1314 if (fib_sync_down_addr(dev, ifa->ifa_local))
1315 fib_flush(dev_net(dev));
1324 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1327 struct fib_result res;
1328 struct flowi4 fl4 = {
1329 .flowi4_mark = frn->fl_mark,
1330 .daddr = frn->fl_addr,
1331 .flowi4_tos = frn->fl_tos,
1332 .flowi4_scope = frn->fl_scope,
1334 struct fib_table *tb;
1338 tb = fib_get_table(net, frn->tb_id_in);
1344 frn->tb_id = tb->tb_id;
1345 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1348 frn->prefixlen = res.prefixlen;
1349 frn->nh_sel = res.nh_sel;
1350 frn->type = res.type;
1351 frn->scope = res.scope;
1359 static void nl_fib_input(struct sk_buff *skb)
1362 struct fib_result_nl *frn;
1363 struct nlmsghdr *nlh;
1366 net = sock_net(skb->sk);
1367 nlh = nlmsg_hdr(skb);
1368 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1369 skb->len < nlh->nlmsg_len ||
1370 nlmsg_len(nlh) < sizeof(*frn))
1373 skb = netlink_skb_clone(skb, GFP_KERNEL);
1376 nlh = nlmsg_hdr(skb);
1378 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1379 nl_fib_lookup(net, frn);
1381 portid = NETLINK_CB(skb).portid; /* netlink portid */
1382 NETLINK_CB(skb).portid = 0; /* from kernel */
1383 NETLINK_CB(skb).dst_group = 0; /* unicast */
1384 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1387 static int __net_init nl_fib_lookup_init(struct net *net)
1390 struct netlink_kernel_cfg cfg = {
1391 .input = nl_fib_input,
1394 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1396 return -EAFNOSUPPORT;
1397 net->ipv4.fibnl = sk;
1401 static void nl_fib_lookup_exit(struct net *net)
1403 netlink_kernel_release(net->ipv4.fibnl);
1404 net->ipv4.fibnl = NULL;
1407 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1410 if (fib_sync_down_dev(dev, event, force))
1411 fib_flush(dev_net(dev));
1413 rt_cache_flush(dev_net(dev));
1417 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1419 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1420 struct net_device *dev = ifa->ifa_dev->dev;
1421 struct net *net = dev_net(dev);
1425 fib_add_ifaddr(ifa);
1426 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1427 fib_sync_up(dev, RTNH_F_DEAD);
1429 atomic_inc(&net->ipv4.dev_addr_genid);
1430 rt_cache_flush(dev_net(dev));
1433 fib_del_ifaddr(ifa, NULL);
1434 atomic_inc(&net->ipv4.dev_addr_genid);
1435 if (!ifa->ifa_dev->ifa_list) {
1436 /* Last address was deleted from this interface.
1439 fib_disable_ip(dev, event, true);
1441 rt_cache_flush(dev_net(dev));
1448 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1450 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1451 struct netdev_notifier_changeupper_info *upper_info = ptr;
1452 struct netdev_notifier_info_ext *info_ext = ptr;
1453 struct in_device *in_dev;
1454 struct net *net = dev_net(dev);
1455 struct in_ifaddr *ifa;
1458 if (event == NETDEV_UNREGISTER) {
1459 fib_disable_ip(dev, event, true);
1464 in_dev = __in_dev_get_rtnl(dev);
1470 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1471 fib_add_ifaddr(ifa);
1473 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1474 fib_sync_up(dev, RTNH_F_DEAD);
1476 atomic_inc(&net->ipv4.dev_addr_genid);
1477 rt_cache_flush(net);
1480 fib_disable_ip(dev, event, false);
1483 flags = dev_get_flags(dev);
1484 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1485 fib_sync_up(dev, RTNH_F_LINKDOWN);
1487 fib_sync_down_dev(dev, event, false);
1488 rt_cache_flush(net);
1490 case NETDEV_CHANGEMTU:
1491 fib_sync_mtu(dev, info_ext->ext.mtu);
1492 rt_cache_flush(net);
1494 case NETDEV_CHANGEUPPER:
1496 /* flush all routes if dev is linked to or unlinked from
1497 * an L3 master device (e.g., VRF)
1499 if (upper_info->upper_dev &&
1500 netif_is_l3_master(upper_info->upper_dev))
1501 fib_disable_ip(dev, NETDEV_DOWN, true);
1507 static struct notifier_block fib_inetaddr_notifier = {
1508 .notifier_call = fib_inetaddr_event,
1511 static struct notifier_block fib_netdev_notifier = {
1512 .notifier_call = fib_netdev_event,
1515 static int __net_init ip_fib_net_init(struct net *net)
1518 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1520 err = fib4_notifier_init(net);
1524 /* Avoid false sharing : Use at least a full cache line */
1525 size = max_t(size_t, size, L1_CACHE_BYTES);
1527 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1528 if (!net->ipv4.fib_table_hash) {
1530 goto err_table_hash_alloc;
1533 err = fib4_rules_init(net);
1535 goto err_rules_init;
1539 kfree(net->ipv4.fib_table_hash);
1540 err_table_hash_alloc:
1541 fib4_notifier_exit(net);
1545 static void ip_fib_net_exit(struct net *net)
1550 #ifdef CONFIG_IP_MULTIPLE_TABLES
1551 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1552 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1554 /* Destroy the tables in reverse order to guarantee that the
1555 * local table, ID 255, is destroyed before the main table, ID
1556 * 254. This is necessary as the local table may contain
1557 * references to data contained in the main table.
1559 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1560 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1561 struct hlist_node *tmp;
1562 struct fib_table *tb;
1564 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1565 hlist_del(&tb->tb_hlist);
1566 fib_table_flush(net, tb, true);
1571 #ifdef CONFIG_IP_MULTIPLE_TABLES
1572 fib4_rules_exit(net);
1575 kfree(net->ipv4.fib_table_hash);
1576 fib4_notifier_exit(net);
1579 static int __net_init fib_net_init(struct net *net)
1583 #ifdef CONFIG_IP_ROUTE_CLASSID
1584 atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1586 error = ip_fib_net_init(net);
1589 error = nl_fib_lookup_init(net);
1592 error = fib_proc_init(net);
1599 nl_fib_lookup_exit(net);
1601 ip_fib_net_exit(net);
1605 static void __net_exit fib_net_exit(struct net *net)
1608 nl_fib_lookup_exit(net);
1609 ip_fib_net_exit(net);
1612 static struct pernet_operations fib_net_ops = {
1613 .init = fib_net_init,
1614 .exit = fib_net_exit,
1617 void __init ip_fib_init(void)
1621 register_pernet_subsys(&fib_net_ops);
1623 register_netdevice_notifier(&fib_netdev_notifier);
1624 register_inetaddr_notifier(&fib_inetaddr_notifier);
1626 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1627 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1628 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);