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>
35 #include <net/inet_dscp.h>
37 #include <net/protocol.h>
38 #include <net/route.h>
42 #include <net/ip_fib.h>
43 #include <net/nexthop.h>
44 #include <net/rtnetlink.h>
46 #include <net/l3mdev.h>
47 #include <net/lwtunnel.h>
48 #include <trace/events/fib.h>
50 #ifndef CONFIG_IP_MULTIPLE_TABLES
52 static int __net_init fib4_rules_init(struct net *net)
54 struct fib_table *local_table, *main_table;
56 main_table = fib_trie_table(RT_TABLE_MAIN, NULL);
60 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
64 hlist_add_head_rcu(&local_table->tb_hlist,
65 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
66 hlist_add_head_rcu(&main_table->tb_hlist,
67 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
71 fib_free_table(main_table);
76 struct fib_table *fib_new_table(struct net *net, u32 id)
78 struct fib_table *tb, *alias = NULL;
83 tb = fib_get_table(net, id);
87 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
88 alias = fib_new_table(net, RT_TABLE_MAIN);
90 tb = fib_trie_table(id, alias);
96 rcu_assign_pointer(net->ipv4.fib_main, tb);
98 case RT_TABLE_DEFAULT:
99 rcu_assign_pointer(net->ipv4.fib_default, tb);
105 h = id & (FIB_TABLE_HASHSZ - 1);
106 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
109 EXPORT_SYMBOL_GPL(fib_new_table);
111 /* caller must hold either rtnl or rcu read lock */
112 struct fib_table *fib_get_table(struct net *net, u32 id)
114 struct fib_table *tb;
115 struct hlist_head *head;
120 h = id & (FIB_TABLE_HASHSZ - 1);
122 head = &net->ipv4.fib_table_hash[h];
123 hlist_for_each_entry_rcu(tb, head, tb_hlist,
124 lockdep_rtnl_is_held()) {
130 #endif /* CONFIG_IP_MULTIPLE_TABLES */
132 static void fib_replace_table(struct net *net, struct fib_table *old,
133 struct fib_table *new)
135 #ifdef CONFIG_IP_MULTIPLE_TABLES
136 switch (new->tb_id) {
138 rcu_assign_pointer(net->ipv4.fib_main, new);
140 case RT_TABLE_DEFAULT:
141 rcu_assign_pointer(net->ipv4.fib_default, new);
148 /* replace the old table in the hlist */
149 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
152 int fib_unmerge(struct net *net)
154 struct fib_table *old, *new, *main_table;
156 /* attempt to fetch local table if it has been allocated */
157 old = fib_get_table(net, RT_TABLE_LOCAL);
161 new = fib_trie_unmerge(old);
165 /* table is already unmerged */
169 /* replace merged table with clean table */
170 fib_replace_table(net, old, new);
173 /* attempt to fetch main table if it has been allocated */
174 main_table = fib_get_table(net, RT_TABLE_MAIN);
178 /* flush local entries from main table */
179 fib_table_flush_external(main_table);
184 void fib_flush(struct net *net)
189 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
190 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
191 struct hlist_node *tmp;
192 struct fib_table *tb;
194 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
195 flushed += fib_table_flush(net, tb, false);
203 * Find address type as if only "dev" was present in the system. If
204 * on_dev is NULL then all interfaces are taken into consideration.
206 static inline unsigned int __inet_dev_addr_type(struct net *net,
207 const struct net_device *dev,
208 __be32 addr, u32 tb_id)
210 struct flowi4 fl4 = { .daddr = addr };
211 struct fib_result res;
212 unsigned int ret = RTN_BROADCAST;
213 struct fib_table *table;
215 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
216 return RTN_BROADCAST;
217 if (ipv4_is_multicast(addr))
218 return RTN_MULTICAST;
222 table = fib_get_table(net, tb_id);
225 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
226 struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
228 if (!dev || dev == nhc->nhc_dev)
237 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
239 return __inet_dev_addr_type(net, NULL, addr, tb_id);
241 EXPORT_SYMBOL(inet_addr_type_table);
243 unsigned int inet_addr_type(struct net *net, __be32 addr)
245 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
247 EXPORT_SYMBOL(inet_addr_type);
249 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
252 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
254 return __inet_dev_addr_type(net, dev, addr, rt_table);
256 EXPORT_SYMBOL(inet_dev_addr_type);
258 /* inet_addr_type with dev == NULL but using the table from a dev
259 * if one is associated
261 unsigned int inet_addr_type_dev_table(struct net *net,
262 const struct net_device *dev,
265 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
267 return __inet_dev_addr_type(net, NULL, addr, rt_table);
269 EXPORT_SYMBOL(inet_addr_type_dev_table);
271 __be32 fib_compute_spec_dst(struct sk_buff *skb)
273 struct net_device *dev = skb->dev;
274 struct in_device *in_dev;
275 struct fib_result res;
280 rt = skb_rtable(skb);
281 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
283 return ip_hdr(skb)->daddr;
285 in_dev = __in_dev_get_rcu(dev);
289 scope = RT_SCOPE_UNIVERSE;
290 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
291 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
292 struct flowi4 fl4 = {
293 .flowi4_iif = LOOPBACK_IFINDEX,
294 .flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev),
295 .daddr = ip_hdr(skb)->saddr,
296 .flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
297 .flowi4_scope = scope,
298 .flowi4_mark = vmark ? skb->mark : 0,
300 if (!fib_lookup(net, &fl4, &res, 0))
301 return fib_result_prefsrc(net, &res);
303 scope = RT_SCOPE_LINK;
306 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
309 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
311 bool dev_match = false;
312 #ifdef CONFIG_IP_ROUTE_MULTIPATH
313 if (unlikely(fi->nh)) {
314 dev_match = nexthop_uses_dev(fi->nh, dev);
318 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
319 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
321 if (nhc_l3mdev_matches_dev(nhc, dev)) {
328 if (fib_info_nhc(fi, 0)->nhc_dev == dev)
334 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
336 /* Given (packet source, input interface) and optional (dst, oif, tos):
337 * - (main) check, that source is valid i.e. not broadcast or our local
339 * - figure out what "logical" interface this packet arrived
340 * and calculate "specific destination" address.
341 * - check, that packet arrived from expected physical interface.
342 * called with rcu_read_lock()
344 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
345 u8 tos, int oif, struct net_device *dev,
346 int rpf, struct in_device *idev, u32 *itag)
348 struct net *net = dev_net(dev);
349 struct flow_keys flkeys;
351 struct fib_result res;
356 fl4.flowi4_l3mdev = 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_HOST;
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_HOST;
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 /* Within the same container, it is regarded as a martian source,
440 * and the same host but different containers are not.
442 if (inet_lookup_ifaddr_rcu(net, src))
451 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
454 static inline __be32 sk_extract_addr(struct sockaddr *addr)
456 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
459 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
463 nla = (struct nlattr *) ((char *) mx + len);
464 nla->nla_type = type;
465 nla->nla_len = nla_attr_size(4);
466 *(u32 *) nla_data(nla) = value;
468 return len + nla_total_size(4);
471 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
472 struct fib_config *cfg)
477 memset(cfg, 0, sizeof(*cfg));
478 cfg->fc_nlinfo.nl_net = net;
480 if (rt->rt_dst.sa_family != AF_INET)
481 return -EAFNOSUPPORT;
484 * Check mask for validity:
485 * a) it must be contiguous.
486 * b) destination must have all host bits clear.
487 * c) if application forgot to set correct family (AF_INET),
488 * reject request unless it is absolutely clear i.e.
489 * both family and mask are zero.
492 addr = sk_extract_addr(&rt->rt_dst);
493 if (!(rt->rt_flags & RTF_HOST)) {
494 __be32 mask = sk_extract_addr(&rt->rt_genmask);
496 if (rt->rt_genmask.sa_family != AF_INET) {
497 if (mask || rt->rt_genmask.sa_family)
498 return -EAFNOSUPPORT;
501 if (bad_mask(mask, addr))
504 plen = inet_mask_len(mask);
507 cfg->fc_dst_len = plen;
510 if (cmd != SIOCDELRT) {
511 cfg->fc_nlflags = NLM_F_CREATE;
512 cfg->fc_protocol = RTPROT_BOOT;
516 cfg->fc_priority = rt->rt_metric - 1;
518 if (rt->rt_flags & RTF_REJECT) {
519 cfg->fc_scope = RT_SCOPE_HOST;
520 cfg->fc_type = RTN_UNREACHABLE;
524 cfg->fc_scope = RT_SCOPE_NOWHERE;
525 cfg->fc_type = RTN_UNICAST;
529 struct net_device *dev;
530 char devname[IFNAMSIZ];
532 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
535 devname[IFNAMSIZ-1] = 0;
536 colon = strchr(devname, ':');
539 dev = __dev_get_by_name(net, devname);
542 cfg->fc_oif = dev->ifindex;
543 cfg->fc_table = l3mdev_fib_table(dev);
545 const struct in_ifaddr *ifa;
546 struct in_device *in_dev;
548 in_dev = __in_dev_get_rtnl(dev);
555 in_dev_for_each_ifa_rcu(ifa, in_dev) {
556 if (strcmp(ifa->ifa_label, devname) == 0)
563 cfg->fc_prefsrc = ifa->ifa_local;
567 addr = sk_extract_addr(&rt->rt_gateway);
568 if (rt->rt_gateway.sa_family == AF_INET && addr) {
569 unsigned int addr_type;
572 cfg->fc_gw_family = AF_INET;
573 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
574 if (rt->rt_flags & RTF_GATEWAY &&
575 addr_type == RTN_UNICAST)
576 cfg->fc_scope = RT_SCOPE_UNIVERSE;
579 if (cmd == SIOCDELRT)
582 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
585 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
586 cfg->fc_scope = RT_SCOPE_LINK;
588 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
592 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
596 if (rt->rt_flags & RTF_MTU)
597 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
599 if (rt->rt_flags & RTF_WINDOW)
600 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
602 if (rt->rt_flags & RTF_IRTT)
603 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
606 cfg->fc_mx_len = len;
613 * Handle IP routing ioctl calls.
614 * These are used to manipulate the routing tables
616 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
618 struct fib_config cfg;
622 case SIOCADDRT: /* Add a route */
623 case SIOCDELRT: /* Delete a route */
624 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
628 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
630 struct fib_table *tb;
632 if (cmd == SIOCDELRT) {
633 tb = fib_get_table(net, cfg.fc_table);
635 err = fib_table_delete(net, tb, &cfg,
640 tb = fib_new_table(net, cfg.fc_table);
642 err = fib_table_insert(net, tb,
648 /* allocated by rtentry_to_fib_config() */
657 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
658 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 },
659 [RTA_DST] = { .type = NLA_U32 },
660 [RTA_SRC] = { .type = NLA_U32 },
661 [RTA_IIF] = { .type = NLA_U32 },
662 [RTA_OIF] = { .type = NLA_U32 },
663 [RTA_GATEWAY] = { .type = NLA_U32 },
664 [RTA_PRIORITY] = { .type = NLA_U32 },
665 [RTA_PREFSRC] = { .type = NLA_U32 },
666 [RTA_METRICS] = { .type = NLA_NESTED },
667 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
668 [RTA_FLOW] = { .type = NLA_U32 },
669 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
670 [RTA_ENCAP] = { .type = NLA_NESTED },
671 [RTA_UID] = { .type = NLA_U32 },
672 [RTA_MARK] = { .type = NLA_U32 },
673 [RTA_TABLE] = { .type = NLA_U32 },
674 [RTA_IP_PROTO] = { .type = NLA_U8 },
675 [RTA_SPORT] = { .type = NLA_U16 },
676 [RTA_DPORT] = { .type = NLA_U16 },
677 [RTA_NH_ID] = { .type = NLA_U32 },
680 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
681 struct netlink_ext_ack *extack)
686 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
687 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
692 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
694 switch (via->rtvia_family) {
696 if (alen != sizeof(__be32)) {
697 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
700 cfg->fc_gw_family = AF_INET;
701 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
704 #if IS_ENABLED(CONFIG_IPV6)
705 if (alen != sizeof(struct in6_addr)) {
706 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
709 cfg->fc_gw_family = AF_INET6;
710 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
712 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
717 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
724 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
725 struct nlmsghdr *nlh, struct fib_config *cfg,
726 struct netlink_ext_ack *extack)
728 bool has_gw = false, has_via = false;
733 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
734 rtm_ipv4_policy, extack);
738 memset(cfg, 0, sizeof(*cfg));
740 rtm = nlmsg_data(nlh);
742 if (!inet_validate_dscp(rtm->rtm_tos)) {
743 NL_SET_ERR_MSG(extack,
744 "Invalid dsfield (tos): ECN bits must be 0");
748 cfg->fc_dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
750 cfg->fc_dst_len = rtm->rtm_dst_len;
751 cfg->fc_table = rtm->rtm_table;
752 cfg->fc_protocol = rtm->rtm_protocol;
753 cfg->fc_scope = rtm->rtm_scope;
754 cfg->fc_type = rtm->rtm_type;
755 cfg->fc_flags = rtm->rtm_flags;
756 cfg->fc_nlflags = nlh->nlmsg_flags;
758 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
759 cfg->fc_nlinfo.nlh = nlh;
760 cfg->fc_nlinfo.nl_net = net;
762 if (cfg->fc_type > RTN_MAX) {
763 NL_SET_ERR_MSG(extack, "Invalid route type");
768 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
769 switch (nla_type(attr)) {
771 cfg->fc_dst = nla_get_be32(attr);
774 cfg->fc_oif = nla_get_u32(attr);
778 cfg->fc_gw4 = nla_get_be32(attr);
780 cfg->fc_gw_family = AF_INET;
784 err = fib_gw_from_via(cfg, attr, extack);
789 cfg->fc_priority = nla_get_u32(attr);
792 cfg->fc_prefsrc = nla_get_be32(attr);
795 cfg->fc_mx = nla_data(attr);
796 cfg->fc_mx_len = nla_len(attr);
799 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
804 cfg->fc_mp = nla_data(attr);
805 cfg->fc_mp_len = nla_len(attr);
808 cfg->fc_flow = nla_get_u32(attr);
811 cfg->fc_table = nla_get_u32(attr);
814 cfg->fc_encap = attr;
817 cfg->fc_encap_type = nla_get_u16(attr);
818 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
824 cfg->fc_nh_id = nla_get_u32(attr);
830 if (cfg->fc_oif || cfg->fc_gw_family ||
831 cfg->fc_encap || cfg->fc_mp) {
832 NL_SET_ERR_MSG(extack,
833 "Nexthop specification and nexthop id are mutually exclusive");
838 if (has_gw && has_via) {
839 NL_SET_ERR_MSG(extack,
840 "Nexthop configuration can not contain both GATEWAY and VIA");
849 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
850 struct netlink_ext_ack *extack)
852 struct net *net = sock_net(skb->sk);
853 struct fib_config cfg;
854 struct fib_table *tb;
857 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
861 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
862 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
867 tb = fib_get_table(net, cfg.fc_table);
869 NL_SET_ERR_MSG(extack, "FIB table does not exist");
874 err = fib_table_delete(net, tb, &cfg, extack);
879 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
880 struct netlink_ext_ack *extack)
882 struct net *net = sock_net(skb->sk);
883 struct fib_config cfg;
884 struct fib_table *tb;
887 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
891 tb = fib_new_table(net, cfg.fc_table);
897 err = fib_table_insert(net, tb, &cfg, extack);
898 if (!err && cfg.fc_type == RTN_LOCAL)
899 net->ipv4.fib_has_custom_local_routes = true;
904 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
905 struct fib_dump_filter *filter,
906 struct netlink_callback *cb)
908 struct netlink_ext_ack *extack = cb->extack;
909 struct nlattr *tb[RTA_MAX + 1];
915 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
916 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
920 rtm = nlmsg_data(nlh);
921 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
923 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
927 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
928 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
931 if (rtm->rtm_flags & RTM_F_CLONED)
932 filter->dump_routes = false;
934 filter->dump_exceptions = false;
936 filter->flags = rtm->rtm_flags;
937 filter->protocol = rtm->rtm_protocol;
938 filter->rt_type = rtm->rtm_type;
939 filter->table_id = rtm->rtm_table;
941 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
942 rtm_ipv4_policy, extack);
946 for (i = 0; i <= RTA_MAX; ++i) {
954 filter->table_id = nla_get_u32(tb[i]);
957 ifindex = nla_get_u32(tb[i]);
958 filter->dev = __dev_get_by_index(net, ifindex);
963 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
968 if (filter->flags || filter->protocol || filter->rt_type ||
969 filter->table_id || filter->dev) {
970 filter->filter_set = 1;
971 cb->answer_flags = NLM_F_DUMP_FILTERED;
976 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
978 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
980 struct fib_dump_filter filter = { .dump_routes = true,
981 .dump_exceptions = true };
982 const struct nlmsghdr *nlh = cb->nlh;
983 struct net *net = sock_net(skb->sk);
985 unsigned int e = 0, s_e;
986 struct fib_table *tb;
987 struct hlist_head *head;
990 if (cb->strict_check) {
991 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
994 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
995 struct rtmsg *rtm = nlmsg_data(nlh);
997 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
1000 /* ipv4 does not use prefix flag */
1001 if (filter.flags & RTM_F_PREFIX)
1004 if (filter.table_id) {
1005 tb = fib_get_table(net, filter.table_id);
1007 if (rtnl_msg_family(cb->nlh) != PF_INET)
1010 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1015 err = fib_table_dump(tb, skb, cb, &filter);
1017 return skb->len ? : err;
1025 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1027 head = &net->ipv4.fib_table_hash[h];
1028 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1032 memset(&cb->args[2], 0, sizeof(cb->args) -
1033 2 * sizeof(cb->args[0]));
1034 err = fib_table_dump(tb, skb, cb, &filter);
1036 if (likely(skb->len))
1057 /* Prepare and feed intra-kernel routing request.
1058 * Really, it should be netlink message, but :-( netlink
1059 * can be not configured, so that we feed it directly
1060 * to fib engine. It is legal, because all events occur
1061 * only when netlink is already locked.
1063 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1064 struct in_ifaddr *ifa, u32 rt_priority)
1066 struct net *net = dev_net(ifa->ifa_dev->dev);
1067 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1068 struct fib_table *tb;
1069 struct fib_config cfg = {
1070 .fc_protocol = RTPROT_KERNEL,
1073 .fc_dst_len = dst_len,
1074 .fc_priority = rt_priority,
1075 .fc_prefsrc = ifa->ifa_local,
1076 .fc_oif = ifa->ifa_dev->dev->ifindex,
1077 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1084 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1086 tb = fib_new_table(net, tb_id);
1090 cfg.fc_table = tb->tb_id;
1092 if (type != RTN_LOCAL)
1093 cfg.fc_scope = RT_SCOPE_LINK;
1095 cfg.fc_scope = RT_SCOPE_HOST;
1097 if (cmd == RTM_NEWROUTE)
1098 fib_table_insert(net, tb, &cfg, NULL);
1100 fib_table_delete(net, tb, &cfg, NULL);
1103 void fib_add_ifaddr(struct in_ifaddr *ifa)
1105 struct in_device *in_dev = ifa->ifa_dev;
1106 struct net_device *dev = in_dev->dev;
1107 struct in_ifaddr *prim = ifa;
1108 __be32 mask = ifa->ifa_mask;
1109 __be32 addr = ifa->ifa_local;
1110 __be32 prefix = ifa->ifa_address & mask;
1112 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1113 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1115 pr_warn("%s: bug: prim == NULL\n", __func__);
1120 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1122 if (!(dev->flags & IFF_UP))
1125 /* Add broadcast address, if it is explicitly assigned. */
1126 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1127 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1129 arp_invalidate(dev, ifa->ifa_broadcast, false);
1132 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1133 (prefix != addr || ifa->ifa_prefixlen < 32)) {
1134 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1135 fib_magic(RTM_NEWROUTE,
1136 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1137 prefix, ifa->ifa_prefixlen, prim,
1138 ifa->ifa_rt_priority);
1140 /* Add the network broadcast address, when it makes sense */
1141 if (ifa->ifa_prefixlen < 31) {
1142 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1144 arp_invalidate(dev, prefix | ~mask, false);
1149 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1151 __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1152 struct in_device *in_dev = ifa->ifa_dev;
1153 struct net_device *dev = in_dev->dev;
1155 if (!(dev->flags & IFF_UP) ||
1156 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1157 ipv4_is_zeronet(prefix) ||
1158 (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1162 fib_magic(RTM_NEWROUTE,
1163 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1164 prefix, ifa->ifa_prefixlen, ifa, new_metric);
1166 /* delete the old */
1167 fib_magic(RTM_DELROUTE,
1168 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1169 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1172 /* Delete primary or secondary address.
1173 * Optionally, on secondary address promotion consider the addresses
1174 * from subnet iprim as deleted, even if they are in device list.
1175 * In this case the secondary ifa can be in device list.
1177 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1179 struct in_device *in_dev = ifa->ifa_dev;
1180 struct net_device *dev = in_dev->dev;
1181 struct in_ifaddr *ifa1;
1182 struct in_ifaddr *prim = ifa, *prim1 = NULL;
1183 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1184 __be32 any = ifa->ifa_address & ifa->ifa_mask;
1189 unsigned int ok = 0;
1190 int subnet = 0; /* Primary network */
1191 int gone = 1; /* Address is missing */
1192 int same_prefsrc = 0; /* Another primary with same IP */
1194 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1195 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1197 /* if the device has been deleted, we don't perform
1201 pr_warn("%s: bug: prim == NULL\n", __func__);
1204 if (iprim && iprim != prim) {
1205 pr_warn("%s: bug: iprim != prim\n", __func__);
1208 } else if (!ipv4_is_zeronet(any) &&
1209 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1210 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1211 fib_magic(RTM_DELROUTE,
1212 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1213 any, ifa->ifa_prefixlen, prim, 0);
1220 /* Deletion is more complicated than add.
1221 * We should take care of not to delete too much :-)
1223 * Scan address list to be sure that addresses are really gone.
1226 in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1228 /* promotion, keep the IP */
1232 /* Ignore IFAs from our subnet */
1233 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1234 inet_ifa_match(ifa1->ifa_address, iprim))
1237 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1238 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1239 /* Another address from our subnet? */
1240 if (ifa1->ifa_mask == prim->ifa_mask &&
1241 inet_ifa_match(ifa1->ifa_address, prim))
1244 /* We reached the secondaries, so
1245 * same_prefsrc should be determined.
1249 /* Search new prim1 if ifa1 is not
1250 * using the current prim1
1253 ifa1->ifa_mask != prim1->ifa_mask ||
1254 !inet_ifa_match(ifa1->ifa_address, prim1))
1255 prim1 = inet_ifa_byprefix(in_dev,
1260 if (prim1->ifa_local != prim->ifa_local)
1264 if (prim->ifa_local != ifa1->ifa_local)
1270 if (ifa->ifa_local == ifa1->ifa_local)
1272 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1274 if (brd == ifa1->ifa_broadcast)
1276 if (any == ifa1->ifa_broadcast)
1278 /* primary has network specific broadcasts */
1279 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1280 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1281 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1283 if (!ipv4_is_zeronet(any1)) {
1284 if (ifa->ifa_broadcast == brd1 ||
1285 ifa->ifa_broadcast == any1)
1287 if (brd == brd1 || brd == any1)
1289 if (any == brd1 || any == any1)
1298 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1300 if (subnet && ifa->ifa_prefixlen < 31) {
1301 if (!(ok & BRD1_OK))
1302 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1304 if (!(ok & BRD0_OK))
1305 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1308 if (!(ok & LOCAL_OK)) {
1309 unsigned int addr_type;
1311 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1313 /* Check, that this local address finally disappeared. */
1314 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1316 if (gone && addr_type != RTN_LOCAL) {
1317 /* And the last, but not the least thing.
1318 * We must flush stray FIB entries.
1320 * First of all, we scan fib_info list searching
1321 * for stray nexthop entries, then ignite fib_flush.
1323 if (fib_sync_down_addr(dev, ifa->ifa_local))
1324 fib_flush(dev_net(dev));
1333 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1336 struct fib_result res;
1337 struct flowi4 fl4 = {
1338 .flowi4_mark = frn->fl_mark,
1339 .daddr = frn->fl_addr,
1340 .flowi4_tos = frn->fl_tos,
1341 .flowi4_scope = frn->fl_scope,
1343 struct fib_table *tb;
1347 tb = fib_get_table(net, frn->tb_id_in);
1353 frn->tb_id = tb->tb_id;
1354 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1357 frn->prefixlen = res.prefixlen;
1358 frn->nh_sel = res.nh_sel;
1359 frn->type = res.type;
1360 frn->scope = res.scope;
1368 static void nl_fib_input(struct sk_buff *skb)
1371 struct fib_result_nl *frn;
1372 struct nlmsghdr *nlh;
1375 net = sock_net(skb->sk);
1376 nlh = nlmsg_hdr(skb);
1377 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1378 skb->len < nlh->nlmsg_len ||
1379 nlmsg_len(nlh) < sizeof(*frn))
1382 skb = netlink_skb_clone(skb, GFP_KERNEL);
1385 nlh = nlmsg_hdr(skb);
1387 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1388 nl_fib_lookup(net, frn);
1390 portid = NETLINK_CB(skb).portid; /* netlink portid */
1391 NETLINK_CB(skb).portid = 0; /* from kernel */
1392 NETLINK_CB(skb).dst_group = 0; /* unicast */
1393 nlmsg_unicast(net->ipv4.fibnl, skb, portid);
1396 static int __net_init nl_fib_lookup_init(struct net *net)
1399 struct netlink_kernel_cfg cfg = {
1400 .input = nl_fib_input,
1403 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1405 return -EAFNOSUPPORT;
1406 net->ipv4.fibnl = sk;
1410 static void nl_fib_lookup_exit(struct net *net)
1412 netlink_kernel_release(net->ipv4.fibnl);
1413 net->ipv4.fibnl = NULL;
1416 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1419 if (fib_sync_down_dev(dev, event, force))
1420 fib_flush(dev_net(dev));
1422 rt_cache_flush(dev_net(dev));
1426 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1428 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1429 struct net_device *dev = ifa->ifa_dev->dev;
1430 struct net *net = dev_net(dev);
1434 fib_add_ifaddr(ifa);
1435 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1436 fib_sync_up(dev, RTNH_F_DEAD);
1438 atomic_inc(&net->ipv4.dev_addr_genid);
1439 rt_cache_flush(dev_net(dev));
1442 fib_del_ifaddr(ifa, NULL);
1443 atomic_inc(&net->ipv4.dev_addr_genid);
1444 if (!ifa->ifa_dev->ifa_list) {
1445 /* Last address was deleted from this interface.
1448 fib_disable_ip(dev, event, true);
1450 rt_cache_flush(dev_net(dev));
1457 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1459 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1460 struct netdev_notifier_changeupper_info *upper_info = ptr;
1461 struct netdev_notifier_info_ext *info_ext = ptr;
1462 struct in_device *in_dev;
1463 struct net *net = dev_net(dev);
1464 struct in_ifaddr *ifa;
1467 if (event == NETDEV_UNREGISTER) {
1468 fib_disable_ip(dev, event, true);
1473 in_dev = __in_dev_get_rtnl(dev);
1479 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1480 fib_add_ifaddr(ifa);
1482 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1483 fib_sync_up(dev, RTNH_F_DEAD);
1485 atomic_inc(&net->ipv4.dev_addr_genid);
1486 rt_cache_flush(net);
1489 fib_disable_ip(dev, event, false);
1492 flags = dev_get_flags(dev);
1493 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1494 fib_sync_up(dev, RTNH_F_LINKDOWN);
1496 fib_sync_down_dev(dev, event, false);
1497 rt_cache_flush(net);
1499 case NETDEV_CHANGEMTU:
1500 fib_sync_mtu(dev, info_ext->ext.mtu);
1501 rt_cache_flush(net);
1503 case NETDEV_CHANGEUPPER:
1505 /* flush all routes if dev is linked to or unlinked from
1506 * an L3 master device (e.g., VRF)
1508 if (upper_info->upper_dev &&
1509 netif_is_l3_master(upper_info->upper_dev))
1510 fib_disable_ip(dev, NETDEV_DOWN, true);
1516 static struct notifier_block fib_inetaddr_notifier = {
1517 .notifier_call = fib_inetaddr_event,
1520 static struct notifier_block fib_netdev_notifier = {
1521 .notifier_call = fib_netdev_event,
1524 static int __net_init ip_fib_net_init(struct net *net)
1527 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1529 err = fib4_notifier_init(net);
1533 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1534 /* Default to 3-tuple */
1535 net->ipv4.sysctl_fib_multipath_hash_fields =
1536 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
1539 /* Avoid false sharing : Use at least a full cache line */
1540 size = max_t(size_t, size, L1_CACHE_BYTES);
1542 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1543 if (!net->ipv4.fib_table_hash) {
1545 goto err_table_hash_alloc;
1548 err = fib4_rules_init(net);
1550 goto err_rules_init;
1554 kfree(net->ipv4.fib_table_hash);
1555 err_table_hash_alloc:
1556 fib4_notifier_exit(net);
1560 static void ip_fib_net_exit(struct net *net)
1565 #ifdef CONFIG_IP_MULTIPLE_TABLES
1566 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1567 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1569 /* Destroy the tables in reverse order to guarantee that the
1570 * local table, ID 255, is destroyed before the main table, ID
1571 * 254. This is necessary as the local table may contain
1572 * references to data contained in the main table.
1574 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1575 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1576 struct hlist_node *tmp;
1577 struct fib_table *tb;
1579 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1580 hlist_del(&tb->tb_hlist);
1581 fib_table_flush(net, tb, true);
1586 #ifdef CONFIG_IP_MULTIPLE_TABLES
1587 fib4_rules_exit(net);
1590 kfree(net->ipv4.fib_table_hash);
1591 fib4_notifier_exit(net);
1594 static int __net_init fib_net_init(struct net *net)
1598 #ifdef CONFIG_IP_ROUTE_CLASSID
1599 atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1601 error = ip_fib_net_init(net);
1604 error = nl_fib_lookup_init(net);
1607 error = fib_proc_init(net);
1614 nl_fib_lookup_exit(net);
1617 ip_fib_net_exit(net);
1622 static void __net_exit fib_net_exit(struct net *net)
1625 nl_fib_lookup_exit(net);
1628 static void __net_exit fib_net_exit_batch(struct list_head *net_list)
1633 list_for_each_entry(net, net_list, exit_list)
1634 ip_fib_net_exit(net);
1639 static struct pernet_operations fib_net_ops = {
1640 .init = fib_net_init,
1641 .exit = fib_net_exit,
1642 .exit_batch = fib_net_exit_batch,
1645 void __init ip_fib_init(void)
1649 register_pernet_subsys(&fib_net_ops);
1651 register_netdevice_notifier(&fib_netdev_notifier);
1652 register_inetaddr_notifier(&fib_inetaddr_notifier);
1654 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1655 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1656 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);