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);
74 static bool fib4_has_custom_rules(struct net *net)
80 struct fib_table *fib_new_table(struct net *net, u32 id)
82 struct fib_table *tb, *alias = NULL;
87 tb = fib_get_table(net, id);
91 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
92 alias = fib_new_table(net, RT_TABLE_MAIN);
94 tb = fib_trie_table(id, alias);
100 rcu_assign_pointer(net->ipv4.fib_main, tb);
102 case RT_TABLE_DEFAULT:
103 rcu_assign_pointer(net->ipv4.fib_default, tb);
109 h = id & (FIB_TABLE_HASHSZ - 1);
110 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
113 EXPORT_SYMBOL_GPL(fib_new_table);
115 /* caller must hold either rtnl or rcu read lock */
116 struct fib_table *fib_get_table(struct net *net, u32 id)
118 struct fib_table *tb;
119 struct hlist_head *head;
124 h = id & (FIB_TABLE_HASHSZ - 1);
126 head = &net->ipv4.fib_table_hash[h];
127 hlist_for_each_entry_rcu(tb, head, tb_hlist,
128 lockdep_rtnl_is_held()) {
135 static bool fib4_has_custom_rules(struct net *net)
137 return net->ipv4.fib_has_custom_rules;
139 #endif /* CONFIG_IP_MULTIPLE_TABLES */
141 static void fib_replace_table(struct net *net, struct fib_table *old,
142 struct fib_table *new)
144 #ifdef CONFIG_IP_MULTIPLE_TABLES
145 switch (new->tb_id) {
147 rcu_assign_pointer(net->ipv4.fib_main, new);
149 case RT_TABLE_DEFAULT:
150 rcu_assign_pointer(net->ipv4.fib_default, new);
157 /* replace the old table in the hlist */
158 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
161 int fib_unmerge(struct net *net)
163 struct fib_table *old, *new, *main_table;
165 /* attempt to fetch local table if it has been allocated */
166 old = fib_get_table(net, RT_TABLE_LOCAL);
170 new = fib_trie_unmerge(old);
174 /* table is already unmerged */
178 /* replace merged table with clean table */
179 fib_replace_table(net, old, new);
182 /* attempt to fetch main table if it has been allocated */
183 main_table = fib_get_table(net, RT_TABLE_MAIN);
187 /* flush local entries from main table */
188 fib_table_flush_external(main_table);
193 void fib_flush(struct net *net)
198 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
199 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
200 struct hlist_node *tmp;
201 struct fib_table *tb;
203 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
204 flushed += fib_table_flush(net, tb, false);
212 * Find address type as if only "dev" was present in the system. If
213 * on_dev is NULL then all interfaces are taken into consideration.
215 static inline unsigned int __inet_dev_addr_type(struct net *net,
216 const struct net_device *dev,
217 __be32 addr, u32 tb_id)
219 struct flowi4 fl4 = { .daddr = addr };
220 struct fib_result res;
221 unsigned int ret = RTN_BROADCAST;
222 struct fib_table *table;
224 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
225 return RTN_BROADCAST;
226 if (ipv4_is_multicast(addr))
227 return RTN_MULTICAST;
231 table = fib_get_table(net, tb_id);
234 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
235 struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
237 if (!dev || dev == nhc->nhc_dev)
246 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
248 return __inet_dev_addr_type(net, NULL, addr, tb_id);
250 EXPORT_SYMBOL(inet_addr_type_table);
252 unsigned int inet_addr_type(struct net *net, __be32 addr)
254 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
256 EXPORT_SYMBOL(inet_addr_type);
258 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
261 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
263 return __inet_dev_addr_type(net, dev, addr, rt_table);
265 EXPORT_SYMBOL(inet_dev_addr_type);
267 /* inet_addr_type with dev == NULL but using the table from a dev
268 * if one is associated
270 unsigned int inet_addr_type_dev_table(struct net *net,
271 const struct net_device *dev,
274 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
276 return __inet_dev_addr_type(net, NULL, addr, rt_table);
278 EXPORT_SYMBOL(inet_addr_type_dev_table);
280 __be32 fib_compute_spec_dst(struct sk_buff *skb)
282 struct net_device *dev = skb->dev;
283 struct in_device *in_dev;
284 struct fib_result res;
289 rt = skb_rtable(skb);
290 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
292 return ip_hdr(skb)->daddr;
294 in_dev = __in_dev_get_rcu(dev);
298 scope = RT_SCOPE_UNIVERSE;
299 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
300 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
301 struct flowi4 fl4 = {
302 .flowi4_iif = LOOPBACK_IFINDEX,
303 .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
304 .daddr = ip_hdr(skb)->saddr,
305 .flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
306 .flowi4_scope = scope,
307 .flowi4_mark = vmark ? skb->mark : 0,
309 if (!fib_lookup(net, &fl4, &res, 0))
310 return fib_result_prefsrc(net, &res);
312 scope = RT_SCOPE_LINK;
315 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
318 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
320 bool dev_match = false;
321 #ifdef CONFIG_IP_ROUTE_MULTIPATH
322 if (unlikely(fi->nh)) {
323 dev_match = nexthop_uses_dev(fi->nh, dev);
327 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
328 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
330 if (nhc_l3mdev_matches_dev(nhc, dev)) {
337 if (fib_info_nhc(fi, 0)->nhc_dev == dev)
343 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
345 /* Given (packet source, input interface) and optional (dst, oif, tos):
346 * - (main) check, that source is valid i.e. not broadcast or our local
348 * - figure out what "logical" interface this packet arrived
349 * and calculate "specific destination" address.
350 * - check, that packet arrived from expected physical interface.
351 * called with rcu_read_lock()
353 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
354 u8 tos, int oif, struct net_device *dev,
355 int rpf, struct in_device *idev, u32 *itag)
357 struct net *net = dev_net(dev);
358 struct flow_keys flkeys;
360 struct fib_result res;
365 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
367 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
370 fl4.flowi4_tos = tos;
371 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
372 fl4.flowi4_tun_key.tun_id = 0;
373 fl4.flowi4_flags = 0;
374 fl4.flowi4_uid = sock_net_uid(net, NULL);
375 fl4.flowi4_multipath_hash = 0;
377 no_addr = idev->ifa_list == NULL;
379 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
380 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
381 fl4.flowi4_proto = 0;
385 swap(fl4.fl4_sport, fl4.fl4_dport);
388 if (fib_lookup(net, &fl4, &res, 0))
390 if (res.type != RTN_UNICAST &&
391 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
393 fib_combine_itag(itag, &res);
395 dev_match = fib_info_nh_uses_dev(res.fi, dev);
396 /* This is not common, loopback packets retain skb_dst so normally they
397 * would not even hit this slow path.
399 dev_match = dev_match || (res.type == RTN_LOCAL &&
400 dev == net->loopback_dev);
402 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_LINK;
409 fl4.flowi4_oif = dev->ifindex;
412 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
413 if (res.type == RTN_UNICAST)
414 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_LINK;
430 /* Ignore rp_filter for packets protected by IPsec. */
431 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
432 u8 tos, int oif, struct net_device *dev,
433 struct in_device *idev, u32 *itag)
435 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
436 struct net *net = dev_net(dev);
438 if (!r && !fib_num_tclassid_users(net) &&
439 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
440 if (IN_DEV_ACCEPT_LOCAL(idev))
442 /* with custom local routes in place, checking local addresses
443 * only will be too optimistic, with custom rules, checking
444 * local addresses only can be too strict, e.g. due to vrf
446 if (net->ipv4.fib_has_custom_local_routes ||
447 fib4_has_custom_rules(net))
449 if (inet_lookup_ifaddr_rcu(net, src))
458 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
461 static inline __be32 sk_extract_addr(struct sockaddr *addr)
463 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
466 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
470 nla = (struct nlattr *) ((char *) mx + len);
471 nla->nla_type = type;
472 nla->nla_len = nla_attr_size(4);
473 *(u32 *) nla_data(nla) = value;
475 return len + nla_total_size(4);
478 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
479 struct fib_config *cfg)
484 memset(cfg, 0, sizeof(*cfg));
485 cfg->fc_nlinfo.nl_net = net;
487 if (rt->rt_dst.sa_family != AF_INET)
488 return -EAFNOSUPPORT;
491 * Check mask for validity:
492 * a) it must be contiguous.
493 * b) destination must have all host bits clear.
494 * c) if application forgot to set correct family (AF_INET),
495 * reject request unless it is absolutely clear i.e.
496 * both family and mask are zero.
499 addr = sk_extract_addr(&rt->rt_dst);
500 if (!(rt->rt_flags & RTF_HOST)) {
501 __be32 mask = sk_extract_addr(&rt->rt_genmask);
503 if (rt->rt_genmask.sa_family != AF_INET) {
504 if (mask || rt->rt_genmask.sa_family)
505 return -EAFNOSUPPORT;
508 if (bad_mask(mask, addr))
511 plen = inet_mask_len(mask);
514 cfg->fc_dst_len = plen;
517 if (cmd != SIOCDELRT) {
518 cfg->fc_nlflags = NLM_F_CREATE;
519 cfg->fc_protocol = RTPROT_BOOT;
523 cfg->fc_priority = rt->rt_metric - 1;
525 if (rt->rt_flags & RTF_REJECT) {
526 cfg->fc_scope = RT_SCOPE_HOST;
527 cfg->fc_type = RTN_UNREACHABLE;
531 cfg->fc_scope = RT_SCOPE_NOWHERE;
532 cfg->fc_type = RTN_UNICAST;
536 struct net_device *dev;
537 char devname[IFNAMSIZ];
539 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
542 devname[IFNAMSIZ-1] = 0;
543 colon = strchr(devname, ':');
546 dev = __dev_get_by_name(net, devname);
549 cfg->fc_oif = dev->ifindex;
550 cfg->fc_table = l3mdev_fib_table(dev);
552 const struct in_ifaddr *ifa;
553 struct in_device *in_dev;
555 in_dev = __in_dev_get_rtnl(dev);
562 in_dev_for_each_ifa_rcu(ifa, in_dev) {
563 if (strcmp(ifa->ifa_label, devname) == 0)
570 cfg->fc_prefsrc = ifa->ifa_local;
574 addr = sk_extract_addr(&rt->rt_gateway);
575 if (rt->rt_gateway.sa_family == AF_INET && addr) {
576 unsigned int addr_type;
579 cfg->fc_gw_family = AF_INET;
580 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
581 if (rt->rt_flags & RTF_GATEWAY &&
582 addr_type == RTN_UNICAST)
583 cfg->fc_scope = RT_SCOPE_UNIVERSE;
587 cfg->fc_table = RT_TABLE_MAIN;
589 if (cmd == SIOCDELRT)
592 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
595 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
596 cfg->fc_scope = RT_SCOPE_LINK;
598 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
602 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
606 if (rt->rt_flags & RTF_MTU)
607 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
609 if (rt->rt_flags & RTF_WINDOW)
610 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
612 if (rt->rt_flags & RTF_IRTT)
613 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
616 cfg->fc_mx_len = len;
623 * Handle IP routing ioctl calls.
624 * These are used to manipulate the routing tables
626 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
628 struct fib_config cfg;
632 case SIOCADDRT: /* Add a route */
633 case SIOCDELRT: /* Delete a route */
634 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
638 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
640 struct fib_table *tb;
642 if (cmd == SIOCDELRT) {
643 tb = fib_get_table(net, cfg.fc_table);
645 err = fib_table_delete(net, tb, &cfg,
650 tb = fib_new_table(net, cfg.fc_table);
652 err = fib_table_insert(net, tb,
658 /* allocated by rtentry_to_fib_config() */
667 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
668 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 },
669 [RTA_DST] = { .type = NLA_U32 },
670 [RTA_SRC] = { .type = NLA_U32 },
671 [RTA_IIF] = { .type = NLA_U32 },
672 [RTA_OIF] = { .type = NLA_U32 },
673 [RTA_GATEWAY] = { .type = NLA_U32 },
674 [RTA_PRIORITY] = { .type = NLA_U32 },
675 [RTA_PREFSRC] = { .type = NLA_U32 },
676 [RTA_METRICS] = { .type = NLA_NESTED },
677 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
678 [RTA_FLOW] = { .type = NLA_U32 },
679 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
680 [RTA_ENCAP] = { .type = NLA_NESTED },
681 [RTA_UID] = { .type = NLA_U32 },
682 [RTA_MARK] = { .type = NLA_U32 },
683 [RTA_TABLE] = { .type = NLA_U32 },
684 [RTA_IP_PROTO] = { .type = NLA_U8 },
685 [RTA_SPORT] = { .type = NLA_U16 },
686 [RTA_DPORT] = { .type = NLA_U16 },
687 [RTA_NH_ID] = { .type = NLA_U32 },
690 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
691 struct netlink_ext_ack *extack)
696 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
697 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
702 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
704 switch (via->rtvia_family) {
706 if (alen != sizeof(__be32)) {
707 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
710 cfg->fc_gw_family = AF_INET;
711 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
714 #if IS_ENABLED(CONFIG_IPV6)
715 if (alen != sizeof(struct in6_addr)) {
716 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
719 cfg->fc_gw_family = AF_INET6;
720 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
722 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
727 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
734 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
735 struct nlmsghdr *nlh, struct fib_config *cfg,
736 struct netlink_ext_ack *extack)
738 bool has_gw = false, has_via = false;
743 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
744 rtm_ipv4_policy, extack);
748 memset(cfg, 0, sizeof(*cfg));
750 rtm = nlmsg_data(nlh);
751 cfg->fc_dst_len = rtm->rtm_dst_len;
752 cfg->fc_tos = rtm->rtm_tos;
753 cfg->fc_table = rtm->rtm_table;
754 cfg->fc_protocol = rtm->rtm_protocol;
755 cfg->fc_scope = rtm->rtm_scope;
756 cfg->fc_type = rtm->rtm_type;
757 cfg->fc_flags = rtm->rtm_flags;
758 cfg->fc_nlflags = nlh->nlmsg_flags;
760 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
761 cfg->fc_nlinfo.nlh = nlh;
762 cfg->fc_nlinfo.nl_net = net;
764 if (cfg->fc_type > RTN_MAX) {
765 NL_SET_ERR_MSG(extack, "Invalid route type");
770 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
771 switch (nla_type(attr)) {
773 cfg->fc_dst = nla_get_be32(attr);
776 cfg->fc_oif = nla_get_u32(attr);
780 cfg->fc_gw4 = nla_get_be32(attr);
782 cfg->fc_gw_family = AF_INET;
786 err = fib_gw_from_via(cfg, attr, extack);
791 cfg->fc_priority = nla_get_u32(attr);
794 cfg->fc_prefsrc = nla_get_be32(attr);
797 cfg->fc_mx = nla_data(attr);
798 cfg->fc_mx_len = nla_len(attr);
801 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
806 cfg->fc_mp = nla_data(attr);
807 cfg->fc_mp_len = nla_len(attr);
810 cfg->fc_flow = nla_get_u32(attr);
813 cfg->fc_table = nla_get_u32(attr);
816 cfg->fc_encap = attr;
819 cfg->fc_encap_type = nla_get_u16(attr);
820 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
826 cfg->fc_nh_id = nla_get_u32(attr);
832 if (cfg->fc_oif || cfg->fc_gw_family ||
833 cfg->fc_encap || cfg->fc_mp) {
834 NL_SET_ERR_MSG(extack,
835 "Nexthop specification and nexthop id are mutually exclusive");
840 if (has_gw && has_via) {
841 NL_SET_ERR_MSG(extack,
842 "Nexthop configuration can not contain both GATEWAY and VIA");
847 cfg->fc_table = RT_TABLE_MAIN;
854 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
855 struct netlink_ext_ack *extack)
857 struct net *net = sock_net(skb->sk);
858 struct fib_config cfg;
859 struct fib_table *tb;
862 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
866 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
867 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
872 tb = fib_get_table(net, cfg.fc_table);
874 NL_SET_ERR_MSG(extack, "FIB table does not exist");
879 err = fib_table_delete(net, tb, &cfg, extack);
884 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
885 struct netlink_ext_ack *extack)
887 struct net *net = sock_net(skb->sk);
888 struct fib_config cfg;
889 struct fib_table *tb;
892 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
896 tb = fib_new_table(net, cfg.fc_table);
902 err = fib_table_insert(net, tb, &cfg, extack);
903 if (!err && cfg.fc_type == RTN_LOCAL)
904 net->ipv4.fib_has_custom_local_routes = true;
909 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
910 struct fib_dump_filter *filter,
911 struct netlink_callback *cb)
913 struct netlink_ext_ack *extack = cb->extack;
914 struct nlattr *tb[RTA_MAX + 1];
920 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
921 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
925 rtm = nlmsg_data(nlh);
926 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
928 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
932 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
933 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
936 if (rtm->rtm_flags & RTM_F_CLONED)
937 filter->dump_routes = false;
939 filter->dump_exceptions = false;
941 filter->flags = rtm->rtm_flags;
942 filter->protocol = rtm->rtm_protocol;
943 filter->rt_type = rtm->rtm_type;
944 filter->table_id = rtm->rtm_table;
946 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
947 rtm_ipv4_policy, extack);
951 for (i = 0; i <= RTA_MAX; ++i) {
959 filter->table_id = nla_get_u32(tb[i]);
962 ifindex = nla_get_u32(tb[i]);
963 filter->dev = __dev_get_by_index(net, ifindex);
968 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
973 if (filter->flags || filter->protocol || filter->rt_type ||
974 filter->table_id || filter->dev) {
975 filter->filter_set = 1;
976 cb->answer_flags = NLM_F_DUMP_FILTERED;
981 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
983 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
985 struct fib_dump_filter filter = { .dump_routes = true,
986 .dump_exceptions = true };
987 const struct nlmsghdr *nlh = cb->nlh;
988 struct net *net = sock_net(skb->sk);
990 unsigned int e = 0, s_e;
991 struct fib_table *tb;
992 struct hlist_head *head;
995 if (cb->strict_check) {
996 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
999 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
1000 struct rtmsg *rtm = nlmsg_data(nlh);
1002 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
1005 /* ipv4 does not use prefix flag */
1006 if (filter.flags & RTM_F_PREFIX)
1009 if (filter.table_id) {
1010 tb = fib_get_table(net, filter.table_id);
1012 if (rtnl_msg_family(cb->nlh) != PF_INET)
1015 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1020 err = fib_table_dump(tb, skb, cb, &filter);
1022 return skb->len ? : err;
1030 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1032 head = &net->ipv4.fib_table_hash[h];
1033 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1037 memset(&cb->args[2], 0, sizeof(cb->args) -
1038 2 * sizeof(cb->args[0]));
1039 err = fib_table_dump(tb, skb, cb, &filter);
1041 if (likely(skb->len))
1062 /* Prepare and feed intra-kernel routing request.
1063 * Really, it should be netlink message, but :-( netlink
1064 * can be not configured, so that we feed it directly
1065 * to fib engine. It is legal, because all events occur
1066 * only when netlink is already locked.
1068 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1069 struct in_ifaddr *ifa, u32 rt_priority)
1071 struct net *net = dev_net(ifa->ifa_dev->dev);
1072 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1073 struct fib_table *tb;
1074 struct fib_config cfg = {
1075 .fc_protocol = RTPROT_KERNEL,
1078 .fc_dst_len = dst_len,
1079 .fc_priority = rt_priority,
1080 .fc_prefsrc = ifa->ifa_local,
1081 .fc_oif = ifa->ifa_dev->dev->ifindex,
1082 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1089 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1091 tb = fib_new_table(net, tb_id);
1095 cfg.fc_table = tb->tb_id;
1097 if (type != RTN_LOCAL)
1098 cfg.fc_scope = RT_SCOPE_LINK;
1100 cfg.fc_scope = RT_SCOPE_HOST;
1102 if (cmd == RTM_NEWROUTE)
1103 fib_table_insert(net, tb, &cfg, NULL);
1105 fib_table_delete(net, tb, &cfg, NULL);
1108 void fib_add_ifaddr(struct in_ifaddr *ifa)
1110 struct in_device *in_dev = ifa->ifa_dev;
1111 struct net_device *dev = in_dev->dev;
1112 struct in_ifaddr *prim = ifa;
1113 __be32 mask = ifa->ifa_mask;
1114 __be32 addr = ifa->ifa_local;
1115 __be32 prefix = ifa->ifa_address & mask;
1117 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1118 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1120 pr_warn("%s: bug: prim == NULL\n", __func__);
1125 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1127 if (!(dev->flags & IFF_UP))
1130 /* Add broadcast address, if it is explicitly assigned. */
1131 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1132 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1134 arp_invalidate(dev, ifa->ifa_broadcast, false);
1137 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1138 (prefix != addr || ifa->ifa_prefixlen < 32)) {
1139 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1140 fib_magic(RTM_NEWROUTE,
1141 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1142 prefix, ifa->ifa_prefixlen, prim,
1143 ifa->ifa_rt_priority);
1145 /* Add network specific broadcasts, when it takes a sense */
1146 if (ifa->ifa_prefixlen < 31) {
1147 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1149 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1151 arp_invalidate(dev, prefix | ~mask, false);
1156 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1158 __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1159 struct in_device *in_dev = ifa->ifa_dev;
1160 struct net_device *dev = in_dev->dev;
1162 if (!(dev->flags & IFF_UP) ||
1163 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1164 ipv4_is_zeronet(prefix) ||
1165 (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1169 fib_magic(RTM_NEWROUTE,
1170 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1171 prefix, ifa->ifa_prefixlen, ifa, new_metric);
1173 /* delete the old */
1174 fib_magic(RTM_DELROUTE,
1175 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1176 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1179 /* Delete primary or secondary address.
1180 * Optionally, on secondary address promotion consider the addresses
1181 * from subnet iprim as deleted, even if they are in device list.
1182 * In this case the secondary ifa can be in device list.
1184 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1186 struct in_device *in_dev = ifa->ifa_dev;
1187 struct net_device *dev = in_dev->dev;
1188 struct in_ifaddr *ifa1;
1189 struct in_ifaddr *prim = ifa, *prim1 = NULL;
1190 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1191 __be32 any = ifa->ifa_address & ifa->ifa_mask;
1196 unsigned int ok = 0;
1197 int subnet = 0; /* Primary network */
1198 int gone = 1; /* Address is missing */
1199 int same_prefsrc = 0; /* Another primary with same IP */
1201 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1202 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1204 /* if the device has been deleted, we don't perform
1208 pr_warn("%s: bug: prim == NULL\n", __func__);
1211 if (iprim && iprim != prim) {
1212 pr_warn("%s: bug: iprim != prim\n", __func__);
1215 } else if (!ipv4_is_zeronet(any) &&
1216 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1217 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1218 fib_magic(RTM_DELROUTE,
1219 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1220 any, ifa->ifa_prefixlen, prim, 0);
1227 /* Deletion is more complicated than add.
1228 * We should take care of not to delete too much :-)
1230 * Scan address list to be sure that addresses are really gone.
1233 in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1235 /* promotion, keep the IP */
1239 /* Ignore IFAs from our subnet */
1240 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1241 inet_ifa_match(ifa1->ifa_address, iprim))
1244 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1245 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1246 /* Another address from our subnet? */
1247 if (ifa1->ifa_mask == prim->ifa_mask &&
1248 inet_ifa_match(ifa1->ifa_address, prim))
1251 /* We reached the secondaries, so
1252 * same_prefsrc should be determined.
1256 /* Search new prim1 if ifa1 is not
1257 * using the current prim1
1260 ifa1->ifa_mask != prim1->ifa_mask ||
1261 !inet_ifa_match(ifa1->ifa_address, prim1))
1262 prim1 = inet_ifa_byprefix(in_dev,
1267 if (prim1->ifa_local != prim->ifa_local)
1271 if (prim->ifa_local != ifa1->ifa_local)
1277 if (ifa->ifa_local == ifa1->ifa_local)
1279 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1281 if (brd == ifa1->ifa_broadcast)
1283 if (any == ifa1->ifa_broadcast)
1285 /* primary has network specific broadcasts */
1286 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1287 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1288 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1290 if (!ipv4_is_zeronet(any1)) {
1291 if (ifa->ifa_broadcast == brd1 ||
1292 ifa->ifa_broadcast == any1)
1294 if (brd == brd1 || brd == any1)
1296 if (any == brd1 || any == any1)
1305 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1307 if (subnet && ifa->ifa_prefixlen < 31) {
1308 if (!(ok & BRD1_OK))
1309 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1311 if (!(ok & BRD0_OK))
1312 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1315 if (!(ok & LOCAL_OK)) {
1316 unsigned int addr_type;
1318 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1320 /* Check, that this local address finally disappeared. */
1321 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1323 if (gone && addr_type != RTN_LOCAL) {
1324 /* And the last, but not the least thing.
1325 * We must flush stray FIB entries.
1327 * First of all, we scan fib_info list searching
1328 * for stray nexthop entries, then ignite fib_flush.
1330 if (fib_sync_down_addr(dev, ifa->ifa_local))
1331 fib_flush(dev_net(dev));
1340 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1343 struct fib_result res;
1344 struct flowi4 fl4 = {
1345 .flowi4_mark = frn->fl_mark,
1346 .daddr = frn->fl_addr,
1347 .flowi4_tos = frn->fl_tos,
1348 .flowi4_scope = frn->fl_scope,
1350 struct fib_table *tb;
1354 tb = fib_get_table(net, frn->tb_id_in);
1360 frn->tb_id = tb->tb_id;
1361 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1364 frn->prefixlen = res.prefixlen;
1365 frn->nh_sel = res.nh_sel;
1366 frn->type = res.type;
1367 frn->scope = res.scope;
1375 static void nl_fib_input(struct sk_buff *skb)
1378 struct fib_result_nl *frn;
1379 struct nlmsghdr *nlh;
1382 net = sock_net(skb->sk);
1383 nlh = nlmsg_hdr(skb);
1384 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1385 skb->len < nlh->nlmsg_len ||
1386 nlmsg_len(nlh) < sizeof(*frn))
1389 skb = netlink_skb_clone(skb, GFP_KERNEL);
1392 nlh = nlmsg_hdr(skb);
1394 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1395 nl_fib_lookup(net, frn);
1397 portid = NETLINK_CB(skb).portid; /* netlink portid */
1398 NETLINK_CB(skb).portid = 0; /* from kernel */
1399 NETLINK_CB(skb).dst_group = 0; /* unicast */
1400 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1403 static int __net_init nl_fib_lookup_init(struct net *net)
1406 struct netlink_kernel_cfg cfg = {
1407 .input = nl_fib_input,
1410 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1412 return -EAFNOSUPPORT;
1413 net->ipv4.fibnl = sk;
1417 static void nl_fib_lookup_exit(struct net *net)
1419 netlink_kernel_release(net->ipv4.fibnl);
1420 net->ipv4.fibnl = NULL;
1423 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1426 if (fib_sync_down_dev(dev, event, force))
1427 fib_flush(dev_net(dev));
1429 rt_cache_flush(dev_net(dev));
1433 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1435 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1436 struct net_device *dev = ifa->ifa_dev->dev;
1437 struct net *net = dev_net(dev);
1441 fib_add_ifaddr(ifa);
1442 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1443 fib_sync_up(dev, RTNH_F_DEAD);
1445 atomic_inc(&net->ipv4.dev_addr_genid);
1446 rt_cache_flush(dev_net(dev));
1449 fib_del_ifaddr(ifa, NULL);
1450 atomic_inc(&net->ipv4.dev_addr_genid);
1451 if (!ifa->ifa_dev->ifa_list) {
1452 /* Last address was deleted from this interface.
1455 fib_disable_ip(dev, event, true);
1457 rt_cache_flush(dev_net(dev));
1464 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1466 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1467 struct netdev_notifier_changeupper_info *upper_info = ptr;
1468 struct netdev_notifier_info_ext *info_ext = ptr;
1469 struct in_device *in_dev;
1470 struct net *net = dev_net(dev);
1471 struct in_ifaddr *ifa;
1474 if (event == NETDEV_UNREGISTER) {
1475 fib_disable_ip(dev, event, true);
1480 in_dev = __in_dev_get_rtnl(dev);
1486 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1487 fib_add_ifaddr(ifa);
1489 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1490 fib_sync_up(dev, RTNH_F_DEAD);
1492 atomic_inc(&net->ipv4.dev_addr_genid);
1493 rt_cache_flush(net);
1496 fib_disable_ip(dev, event, false);
1499 flags = dev_get_flags(dev);
1500 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1501 fib_sync_up(dev, RTNH_F_LINKDOWN);
1503 fib_sync_down_dev(dev, event, false);
1504 rt_cache_flush(net);
1506 case NETDEV_CHANGEMTU:
1507 fib_sync_mtu(dev, info_ext->ext.mtu);
1508 rt_cache_flush(net);
1510 case NETDEV_CHANGEUPPER:
1512 /* flush all routes if dev is linked to or unlinked from
1513 * an L3 master device (e.g., VRF)
1515 if (upper_info->upper_dev &&
1516 netif_is_l3_master(upper_info->upper_dev))
1517 fib_disable_ip(dev, NETDEV_DOWN, true);
1523 static struct notifier_block fib_inetaddr_notifier = {
1524 .notifier_call = fib_inetaddr_event,
1527 static struct notifier_block fib_netdev_notifier = {
1528 .notifier_call = fib_netdev_event,
1531 static int __net_init ip_fib_net_init(struct net *net)
1534 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1536 err = fib4_notifier_init(net);
1540 /* Avoid false sharing : Use at least a full cache line */
1541 size = max_t(size_t, size, L1_CACHE_BYTES);
1543 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1544 if (!net->ipv4.fib_table_hash) {
1546 goto err_table_hash_alloc;
1549 err = fib4_rules_init(net);
1551 goto err_rules_init;
1555 kfree(net->ipv4.fib_table_hash);
1556 err_table_hash_alloc:
1557 fib4_notifier_exit(net);
1561 static void ip_fib_net_exit(struct net *net)
1566 #ifdef CONFIG_IP_MULTIPLE_TABLES
1567 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1568 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1570 /* Destroy the tables in reverse order to guarantee that the
1571 * local table, ID 255, is destroyed before the main table, ID
1572 * 254. This is necessary as the local table may contain
1573 * references to data contained in the main table.
1575 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1576 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1577 struct hlist_node *tmp;
1578 struct fib_table *tb;
1580 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1581 hlist_del(&tb->tb_hlist);
1582 fib_table_flush(net, tb, true);
1587 #ifdef CONFIG_IP_MULTIPLE_TABLES
1588 fib4_rules_exit(net);
1591 kfree(net->ipv4.fib_table_hash);
1592 fib4_notifier_exit(net);
1595 static int __net_init fib_net_init(struct net *net)
1599 #ifdef CONFIG_IP_ROUTE_CLASSID
1600 atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1602 error = ip_fib_net_init(net);
1605 error = nl_fib_lookup_init(net);
1608 error = fib_proc_init(net);
1615 nl_fib_lookup_exit(net);
1617 ip_fib_net_exit(net);
1621 static void __net_exit fib_net_exit(struct net *net)
1624 nl_fib_lookup_exit(net);
1625 ip_fib_net_exit(net);
1628 static struct pernet_operations fib_net_ops = {
1629 .init = fib_net_init,
1630 .exit = fib_net_exit,
1633 void __init ip_fib_init(void)
1637 register_pernet_subsys(&fib_net_ops);
1639 register_netdevice_notifier(&fib_netdev_notifier);
1640 register_inetaddr_notifier(&fib_inetaddr_notifier);
1642 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1643 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1644 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);