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) {
134 static bool fib4_has_custom_rules(struct net *net)
136 return net->ipv4.fib_has_custom_rules;
138 #endif /* CONFIG_IP_MULTIPLE_TABLES */
140 static void fib_replace_table(struct net *net, struct fib_table *old,
141 struct fib_table *new)
143 #ifdef CONFIG_IP_MULTIPLE_TABLES
144 switch (new->tb_id) {
146 rcu_assign_pointer(net->ipv4.fib_main, new);
148 case RT_TABLE_DEFAULT:
149 rcu_assign_pointer(net->ipv4.fib_default, new);
156 /* replace the old table in the hlist */
157 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
160 int fib_unmerge(struct net *net)
162 struct fib_table *old, *new, *main_table;
164 /* attempt to fetch local table if it has been allocated */
165 old = fib_get_table(net, RT_TABLE_LOCAL);
169 new = fib_trie_unmerge(old);
173 /* table is already unmerged */
177 /* replace merged table with clean table */
178 fib_replace_table(net, old, new);
181 /* attempt to fetch main table if it has been allocated */
182 main_table = fib_get_table(net, RT_TABLE_MAIN);
186 /* flush local entries from main table */
187 fib_table_flush_external(main_table);
192 void fib_flush(struct net *net)
197 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
198 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
199 struct hlist_node *tmp;
200 struct fib_table *tb;
202 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
203 flushed += fib_table_flush(net, tb, false);
211 * Find address type as if only "dev" was present in the system. If
212 * on_dev is NULL then all interfaces are taken into consideration.
214 static inline unsigned int __inet_dev_addr_type(struct net *net,
215 const struct net_device *dev,
216 __be32 addr, u32 tb_id)
218 struct flowi4 fl4 = { .daddr = addr };
219 struct fib_result res;
220 unsigned int ret = RTN_BROADCAST;
221 struct fib_table *table;
223 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
224 return RTN_BROADCAST;
225 if (ipv4_is_multicast(addr))
226 return RTN_MULTICAST;
230 table = fib_get_table(net, tb_id);
233 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
234 struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
236 if (!dev || dev == nhc->nhc_dev)
245 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
247 return __inet_dev_addr_type(net, NULL, addr, tb_id);
249 EXPORT_SYMBOL(inet_addr_type_table);
251 unsigned int inet_addr_type(struct net *net, __be32 addr)
253 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
255 EXPORT_SYMBOL(inet_addr_type);
257 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
260 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
262 return __inet_dev_addr_type(net, dev, addr, rt_table);
264 EXPORT_SYMBOL(inet_dev_addr_type);
266 /* inet_addr_type with dev == NULL but using the table from a dev
267 * if one is associated
269 unsigned int inet_addr_type_dev_table(struct net *net,
270 const struct net_device *dev,
273 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
275 return __inet_dev_addr_type(net, NULL, addr, rt_table);
277 EXPORT_SYMBOL(inet_addr_type_dev_table);
279 __be32 fib_compute_spec_dst(struct sk_buff *skb)
281 struct net_device *dev = skb->dev;
282 struct in_device *in_dev;
283 struct fib_result res;
288 rt = skb_rtable(skb);
289 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
291 return ip_hdr(skb)->daddr;
293 in_dev = __in_dev_get_rcu(dev);
297 scope = RT_SCOPE_UNIVERSE;
298 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
299 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
300 struct flowi4 fl4 = {
301 .flowi4_iif = LOOPBACK_IFINDEX,
302 .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
303 .daddr = ip_hdr(skb)->saddr,
304 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
305 .flowi4_scope = scope,
306 .flowi4_mark = vmark ? skb->mark : 0,
308 if (!fib_lookup(net, &fl4, &res, 0))
309 return fib_result_prefsrc(net, &res);
311 scope = RT_SCOPE_LINK;
314 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
317 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
319 bool dev_match = false;
320 #ifdef CONFIG_IP_ROUTE_MULTIPATH
323 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
324 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
326 if (nhc->nhc_dev == dev) {
329 } else if (l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex) {
335 if (fib_info_nhc(fi, 0)->nhc_dev == dev)
341 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
343 /* Given (packet source, input interface) and optional (dst, oif, tos):
344 * - (main) check, that source is valid i.e. not broadcast or our local
346 * - figure out what "logical" interface this packet arrived
347 * and calculate "specific destination" address.
348 * - check, that packet arrived from expected physical interface.
349 * called with rcu_read_lock()
351 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
352 u8 tos, int oif, struct net_device *dev,
353 int rpf, struct in_device *idev, u32 *itag)
355 struct net *net = dev_net(dev);
356 struct flow_keys flkeys;
358 struct fib_result res;
363 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
365 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
368 fl4.flowi4_tos = tos;
369 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
370 fl4.flowi4_tun_key.tun_id = 0;
371 fl4.flowi4_flags = 0;
372 fl4.flowi4_uid = sock_net_uid(net, NULL);
374 no_addr = idev->ifa_list == NULL;
376 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
377 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
378 fl4.flowi4_proto = 0;
383 if (fib_lookup(net, &fl4, &res, 0))
385 if (res.type != RTN_UNICAST &&
386 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
388 fib_combine_itag(itag, &res);
390 dev_match = fib_info_nh_uses_dev(res.fi, dev);
391 /* This is not common, loopback packets retain skb_dst so normally they
392 * would not even hit this slow path.
394 dev_match = dev_match || (res.type == RTN_LOCAL &&
395 dev == net->loopback_dev);
397 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
404 fl4.flowi4_oif = dev->ifindex;
407 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
408 if (res.type == RTN_UNICAST)
409 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
425 /* Ignore rp_filter for packets protected by IPsec. */
426 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
427 u8 tos, int oif, struct net_device *dev,
428 struct in_device *idev, u32 *itag)
430 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
431 struct net *net = dev_net(dev);
433 if (!r && !fib_num_tclassid_users(net) &&
434 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
435 if (IN_DEV_ACCEPT_LOCAL(idev))
437 /* with custom local routes in place, checking local addresses
438 * only will be too optimistic, with custom rules, checking
439 * local addresses only can be too strict, e.g. due to vrf
441 if (net->ipv4.fib_has_custom_local_routes ||
442 fib4_has_custom_rules(net))
444 if (inet_lookup_ifaddr_rcu(net, src))
453 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
456 static inline __be32 sk_extract_addr(struct sockaddr *addr)
458 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
461 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
465 nla = (struct nlattr *) ((char *) mx + len);
466 nla->nla_type = type;
467 nla->nla_len = nla_attr_size(4);
468 *(u32 *) nla_data(nla) = value;
470 return len + nla_total_size(4);
473 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
474 struct fib_config *cfg)
479 memset(cfg, 0, sizeof(*cfg));
480 cfg->fc_nlinfo.nl_net = net;
482 if (rt->rt_dst.sa_family != AF_INET)
483 return -EAFNOSUPPORT;
486 * Check mask for validity:
487 * a) it must be contiguous.
488 * b) destination must have all host bits clear.
489 * c) if application forgot to set correct family (AF_INET),
490 * reject request unless it is absolutely clear i.e.
491 * both family and mask are zero.
494 addr = sk_extract_addr(&rt->rt_dst);
495 if (!(rt->rt_flags & RTF_HOST)) {
496 __be32 mask = sk_extract_addr(&rt->rt_genmask);
498 if (rt->rt_genmask.sa_family != AF_INET) {
499 if (mask || rt->rt_genmask.sa_family)
500 return -EAFNOSUPPORT;
503 if (bad_mask(mask, addr))
506 plen = inet_mask_len(mask);
509 cfg->fc_dst_len = plen;
512 if (cmd != SIOCDELRT) {
513 cfg->fc_nlflags = NLM_F_CREATE;
514 cfg->fc_protocol = RTPROT_BOOT;
518 cfg->fc_priority = rt->rt_metric - 1;
520 if (rt->rt_flags & RTF_REJECT) {
521 cfg->fc_scope = RT_SCOPE_HOST;
522 cfg->fc_type = RTN_UNREACHABLE;
526 cfg->fc_scope = RT_SCOPE_NOWHERE;
527 cfg->fc_type = RTN_UNICAST;
531 struct net_device *dev;
532 char devname[IFNAMSIZ];
534 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
537 devname[IFNAMSIZ-1] = 0;
538 colon = strchr(devname, ':');
541 dev = __dev_get_by_name(net, devname);
544 cfg->fc_oif = dev->ifindex;
545 cfg->fc_table = l3mdev_fib_table(dev);
547 const struct in_ifaddr *ifa;
548 struct in_device *in_dev;
550 in_dev = __in_dev_get_rtnl(dev);
557 in_dev_for_each_ifa_rcu(ifa, in_dev) {
558 if (strcmp(ifa->ifa_label, devname) == 0)
565 cfg->fc_prefsrc = ifa->ifa_local;
569 addr = sk_extract_addr(&rt->rt_gateway);
570 if (rt->rt_gateway.sa_family == AF_INET && addr) {
571 unsigned int addr_type;
574 cfg->fc_gw_family = AF_INET;
575 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
576 if (rt->rt_flags & RTF_GATEWAY &&
577 addr_type == RTN_UNICAST)
578 cfg->fc_scope = RT_SCOPE_UNIVERSE;
581 if (cmd == SIOCDELRT)
584 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
587 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
588 cfg->fc_scope = RT_SCOPE_LINK;
590 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
594 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
598 if (rt->rt_flags & RTF_MTU)
599 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
601 if (rt->rt_flags & RTF_WINDOW)
602 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
604 if (rt->rt_flags & RTF_IRTT)
605 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
608 cfg->fc_mx_len = len;
615 * Handle IP routing ioctl calls.
616 * These are used to manipulate the routing tables
618 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
620 struct fib_config cfg;
624 case SIOCADDRT: /* Add a route */
625 case SIOCDELRT: /* Delete a route */
626 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
630 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
632 struct fib_table *tb;
634 if (cmd == SIOCDELRT) {
635 tb = fib_get_table(net, cfg.fc_table);
637 err = fib_table_delete(net, tb, &cfg,
642 tb = fib_new_table(net, cfg.fc_table);
644 err = fib_table_insert(net, tb,
650 /* allocated by rtentry_to_fib_config() */
659 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
660 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 },
661 [RTA_DST] = { .type = NLA_U32 },
662 [RTA_SRC] = { .type = NLA_U32 },
663 [RTA_IIF] = { .type = NLA_U32 },
664 [RTA_OIF] = { .type = NLA_U32 },
665 [RTA_GATEWAY] = { .type = NLA_U32 },
666 [RTA_PRIORITY] = { .type = NLA_U32 },
667 [RTA_PREFSRC] = { .type = NLA_U32 },
668 [RTA_METRICS] = { .type = NLA_NESTED },
669 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
670 [RTA_FLOW] = { .type = NLA_U32 },
671 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
672 [RTA_ENCAP] = { .type = NLA_NESTED },
673 [RTA_UID] = { .type = NLA_U32 },
674 [RTA_MARK] = { .type = NLA_U32 },
675 [RTA_TABLE] = { .type = NLA_U32 },
676 [RTA_IP_PROTO] = { .type = NLA_U8 },
677 [RTA_SPORT] = { .type = NLA_U16 },
678 [RTA_DPORT] = { .type = NLA_U16 },
679 [RTA_NH_ID] = { .type = NLA_U32 },
682 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
683 struct netlink_ext_ack *extack)
688 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
689 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
694 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
696 switch (via->rtvia_family) {
698 if (alen != sizeof(__be32)) {
699 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
702 cfg->fc_gw_family = AF_INET;
703 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
707 if (alen != sizeof(struct in6_addr)) {
708 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
711 cfg->fc_gw_family = AF_INET6;
712 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
714 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
719 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
726 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
727 struct nlmsghdr *nlh, struct fib_config *cfg,
728 struct netlink_ext_ack *extack)
730 bool has_gw = false, has_via = false;
735 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
736 rtm_ipv4_policy, extack);
740 memset(cfg, 0, sizeof(*cfg));
742 rtm = nlmsg_data(nlh);
743 cfg->fc_dst_len = rtm->rtm_dst_len;
744 cfg->fc_tos = rtm->rtm_tos;
745 cfg->fc_table = rtm->rtm_table;
746 cfg->fc_protocol = rtm->rtm_protocol;
747 cfg->fc_scope = rtm->rtm_scope;
748 cfg->fc_type = rtm->rtm_type;
749 cfg->fc_flags = rtm->rtm_flags;
750 cfg->fc_nlflags = nlh->nlmsg_flags;
752 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
753 cfg->fc_nlinfo.nlh = nlh;
754 cfg->fc_nlinfo.nl_net = net;
756 if (cfg->fc_type > RTN_MAX) {
757 NL_SET_ERR_MSG(extack, "Invalid route type");
762 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
763 switch (nla_type(attr)) {
765 cfg->fc_dst = nla_get_be32(attr);
768 cfg->fc_oif = nla_get_u32(attr);
772 cfg->fc_gw4 = nla_get_be32(attr);
774 cfg->fc_gw_family = AF_INET;
778 err = fib_gw_from_via(cfg, attr, extack);
783 cfg->fc_priority = nla_get_u32(attr);
786 cfg->fc_prefsrc = nla_get_be32(attr);
789 cfg->fc_mx = nla_data(attr);
790 cfg->fc_mx_len = nla_len(attr);
793 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
798 cfg->fc_mp = nla_data(attr);
799 cfg->fc_mp_len = nla_len(attr);
802 cfg->fc_flow = nla_get_u32(attr);
805 cfg->fc_table = nla_get_u32(attr);
808 cfg->fc_encap = attr;
811 cfg->fc_encap_type = nla_get_u16(attr);
812 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
818 cfg->fc_nh_id = nla_get_u32(attr);
824 if (cfg->fc_oif || cfg->fc_gw_family ||
825 cfg->fc_encap || cfg->fc_mp) {
826 NL_SET_ERR_MSG(extack,
827 "Nexthop specification and nexthop id are mutually exclusive");
832 if (has_gw && has_via) {
833 NL_SET_ERR_MSG(extack,
834 "Nexthop configuration can not contain both GATEWAY and VIA");
843 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
844 struct netlink_ext_ack *extack)
846 struct net *net = sock_net(skb->sk);
847 struct fib_config cfg;
848 struct fib_table *tb;
851 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
855 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
856 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
861 tb = fib_get_table(net, cfg.fc_table);
863 NL_SET_ERR_MSG(extack, "FIB table does not exist");
868 err = fib_table_delete(net, tb, &cfg, extack);
873 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
874 struct netlink_ext_ack *extack)
876 struct net *net = sock_net(skb->sk);
877 struct fib_config cfg;
878 struct fib_table *tb;
881 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
885 tb = fib_new_table(net, cfg.fc_table);
891 err = fib_table_insert(net, tb, &cfg, extack);
892 if (!err && cfg.fc_type == RTN_LOCAL)
893 net->ipv4.fib_has_custom_local_routes = true;
898 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
899 struct fib_dump_filter *filter,
900 struct netlink_callback *cb)
902 struct netlink_ext_ack *extack = cb->extack;
903 struct nlattr *tb[RTA_MAX + 1];
909 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
910 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
914 rtm = nlmsg_data(nlh);
915 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
917 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
921 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
922 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
925 if (rtm->rtm_flags & RTM_F_CLONED)
926 filter->dump_routes = false;
928 filter->dump_exceptions = false;
930 filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
931 filter->flags = rtm->rtm_flags;
932 filter->protocol = rtm->rtm_protocol;
933 filter->rt_type = rtm->rtm_type;
934 filter->table_id = rtm->rtm_table;
936 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
937 rtm_ipv4_policy, extack);
941 for (i = 0; i <= RTA_MAX; ++i) {
949 filter->table_id = nla_get_u32(tb[i]);
952 ifindex = nla_get_u32(tb[i]);
953 filter->dev = __dev_get_by_index(net, ifindex);
958 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
963 if (filter->flags || filter->protocol || filter->rt_type ||
964 filter->table_id || filter->dev) {
965 filter->filter_set = 1;
966 cb->answer_flags = NLM_F_DUMP_FILTERED;
971 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
973 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
975 struct fib_dump_filter filter = { .dump_routes = true,
976 .dump_exceptions = true };
977 const struct nlmsghdr *nlh = cb->nlh;
978 struct net *net = sock_net(skb->sk);
980 unsigned int e = 0, s_e;
981 struct fib_table *tb;
982 struct hlist_head *head;
985 if (cb->strict_check) {
986 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
989 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
990 struct rtmsg *rtm = nlmsg_data(nlh);
992 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
995 /* ipv4 does not use prefix flag */
996 if (filter.flags & RTM_F_PREFIX)
999 if (filter.table_id) {
1000 tb = fib_get_table(net, filter.table_id);
1002 if (filter.dump_all_families)
1005 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1009 err = fib_table_dump(tb, skb, cb, &filter);
1010 return skb->len ? : err;
1018 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1020 head = &net->ipv4.fib_table_hash[h];
1021 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1025 memset(&cb->args[2], 0, sizeof(cb->args) -
1026 2 * sizeof(cb->args[0]));
1027 err = fib_table_dump(tb, skb, cb, &filter);
1029 if (likely(skb->len))
1050 /* Prepare and feed intra-kernel routing request.
1051 * Really, it should be netlink message, but :-( netlink
1052 * can be not configured, so that we feed it directly
1053 * to fib engine. It is legal, because all events occur
1054 * only when netlink is already locked.
1056 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1057 struct in_ifaddr *ifa, u32 rt_priority)
1059 struct net *net = dev_net(ifa->ifa_dev->dev);
1060 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1061 struct fib_table *tb;
1062 struct fib_config cfg = {
1063 .fc_protocol = RTPROT_KERNEL,
1066 .fc_dst_len = dst_len,
1067 .fc_priority = rt_priority,
1068 .fc_prefsrc = ifa->ifa_local,
1069 .fc_oif = ifa->ifa_dev->dev->ifindex,
1070 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1077 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1079 tb = fib_new_table(net, tb_id);
1083 cfg.fc_table = tb->tb_id;
1085 if (type != RTN_LOCAL)
1086 cfg.fc_scope = RT_SCOPE_LINK;
1088 cfg.fc_scope = RT_SCOPE_HOST;
1090 if (cmd == RTM_NEWROUTE)
1091 fib_table_insert(net, tb, &cfg, NULL);
1093 fib_table_delete(net, tb, &cfg, NULL);
1096 void fib_add_ifaddr(struct in_ifaddr *ifa)
1098 struct in_device *in_dev = ifa->ifa_dev;
1099 struct net_device *dev = in_dev->dev;
1100 struct in_ifaddr *prim = ifa;
1101 __be32 mask = ifa->ifa_mask;
1102 __be32 addr = ifa->ifa_local;
1103 __be32 prefix = ifa->ifa_address & mask;
1105 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1106 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1108 pr_warn("%s: bug: prim == NULL\n", __func__);
1113 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1115 if (!(dev->flags & IFF_UP))
1118 /* Add broadcast address, if it is explicitly assigned. */
1119 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1120 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1123 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1124 (prefix != addr || ifa->ifa_prefixlen < 32)) {
1125 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1126 fib_magic(RTM_NEWROUTE,
1127 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1128 prefix, ifa->ifa_prefixlen, prim,
1129 ifa->ifa_rt_priority);
1131 /* Add network specific broadcasts, when it takes a sense */
1132 if (ifa->ifa_prefixlen < 31) {
1133 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1135 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1141 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1143 __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1144 struct in_device *in_dev = ifa->ifa_dev;
1145 struct net_device *dev = in_dev->dev;
1147 if (!(dev->flags & IFF_UP) ||
1148 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1149 ipv4_is_zeronet(prefix) ||
1150 prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1154 fib_magic(RTM_NEWROUTE,
1155 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1156 prefix, ifa->ifa_prefixlen, ifa, new_metric);
1158 /* delete the old */
1159 fib_magic(RTM_DELROUTE,
1160 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1161 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1164 /* Delete primary or secondary address.
1165 * Optionally, on secondary address promotion consider the addresses
1166 * from subnet iprim as deleted, even if they are in device list.
1167 * In this case the secondary ifa can be in device list.
1169 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1171 struct in_device *in_dev = ifa->ifa_dev;
1172 struct net_device *dev = in_dev->dev;
1173 struct in_ifaddr *ifa1;
1174 struct in_ifaddr *prim = ifa, *prim1 = NULL;
1175 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1176 __be32 any = ifa->ifa_address & ifa->ifa_mask;
1181 unsigned int ok = 0;
1182 int subnet = 0; /* Primary network */
1183 int gone = 1; /* Address is missing */
1184 int same_prefsrc = 0; /* Another primary with same IP */
1186 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1187 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1189 /* if the device has been deleted, we don't perform
1193 pr_warn("%s: bug: prim == NULL\n", __func__);
1196 if (iprim && iprim != prim) {
1197 pr_warn("%s: bug: iprim != prim\n", __func__);
1200 } else if (!ipv4_is_zeronet(any) &&
1201 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1202 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1203 fib_magic(RTM_DELROUTE,
1204 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1205 any, ifa->ifa_prefixlen, prim, 0);
1212 /* Deletion is more complicated than add.
1213 * We should take care of not to delete too much :-)
1215 * Scan address list to be sure that addresses are really gone.
1218 in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1220 /* promotion, keep the IP */
1224 /* Ignore IFAs from our subnet */
1225 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1226 inet_ifa_match(ifa1->ifa_address, iprim))
1229 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1230 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1231 /* Another address from our subnet? */
1232 if (ifa1->ifa_mask == prim->ifa_mask &&
1233 inet_ifa_match(ifa1->ifa_address, prim))
1236 /* We reached the secondaries, so
1237 * same_prefsrc should be determined.
1241 /* Search new prim1 if ifa1 is not
1242 * using the current prim1
1245 ifa1->ifa_mask != prim1->ifa_mask ||
1246 !inet_ifa_match(ifa1->ifa_address, prim1))
1247 prim1 = inet_ifa_byprefix(in_dev,
1252 if (prim1->ifa_local != prim->ifa_local)
1256 if (prim->ifa_local != ifa1->ifa_local)
1262 if (ifa->ifa_local == ifa1->ifa_local)
1264 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1266 if (brd == ifa1->ifa_broadcast)
1268 if (any == ifa1->ifa_broadcast)
1270 /* primary has network specific broadcasts */
1271 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1272 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1273 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1275 if (!ipv4_is_zeronet(any1)) {
1276 if (ifa->ifa_broadcast == brd1 ||
1277 ifa->ifa_broadcast == any1)
1279 if (brd == brd1 || brd == any1)
1281 if (any == brd1 || any == any1)
1290 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1292 if (subnet && ifa->ifa_prefixlen < 31) {
1293 if (!(ok & BRD1_OK))
1294 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1296 if (!(ok & BRD0_OK))
1297 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1300 if (!(ok & LOCAL_OK)) {
1301 unsigned int addr_type;
1303 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1305 /* Check, that this local address finally disappeared. */
1306 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1308 if (gone && addr_type != RTN_LOCAL) {
1309 /* And the last, but not the least thing.
1310 * We must flush stray FIB entries.
1312 * First of all, we scan fib_info list searching
1313 * for stray nexthop entries, then ignite fib_flush.
1315 if (fib_sync_down_addr(dev, ifa->ifa_local))
1316 fib_flush(dev_net(dev));
1325 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1328 struct fib_result res;
1329 struct flowi4 fl4 = {
1330 .flowi4_mark = frn->fl_mark,
1331 .daddr = frn->fl_addr,
1332 .flowi4_tos = frn->fl_tos,
1333 .flowi4_scope = frn->fl_scope,
1335 struct fib_table *tb;
1339 tb = fib_get_table(net, frn->tb_id_in);
1345 frn->tb_id = tb->tb_id;
1346 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1349 frn->prefixlen = res.prefixlen;
1350 frn->nh_sel = res.nh_sel;
1351 frn->type = res.type;
1352 frn->scope = res.scope;
1360 static void nl_fib_input(struct sk_buff *skb)
1363 struct fib_result_nl *frn;
1364 struct nlmsghdr *nlh;
1367 net = sock_net(skb->sk);
1368 nlh = nlmsg_hdr(skb);
1369 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1370 skb->len < nlh->nlmsg_len ||
1371 nlmsg_len(nlh) < sizeof(*frn))
1374 skb = netlink_skb_clone(skb, GFP_KERNEL);
1377 nlh = nlmsg_hdr(skb);
1379 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1380 nl_fib_lookup(net, frn);
1382 portid = NETLINK_CB(skb).portid; /* netlink portid */
1383 NETLINK_CB(skb).portid = 0; /* from kernel */
1384 NETLINK_CB(skb).dst_group = 0; /* unicast */
1385 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1388 static int __net_init nl_fib_lookup_init(struct net *net)
1391 struct netlink_kernel_cfg cfg = {
1392 .input = nl_fib_input,
1395 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1397 return -EAFNOSUPPORT;
1398 net->ipv4.fibnl = sk;
1402 static void nl_fib_lookup_exit(struct net *net)
1404 netlink_kernel_release(net->ipv4.fibnl);
1405 net->ipv4.fibnl = NULL;
1408 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1411 if (fib_sync_down_dev(dev, event, force))
1412 fib_flush(dev_net(dev));
1414 rt_cache_flush(dev_net(dev));
1418 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1420 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1421 struct net_device *dev = ifa->ifa_dev->dev;
1422 struct net *net = dev_net(dev);
1426 fib_add_ifaddr(ifa);
1427 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1428 fib_sync_up(dev, RTNH_F_DEAD);
1430 atomic_inc(&net->ipv4.dev_addr_genid);
1431 rt_cache_flush(dev_net(dev));
1434 fib_del_ifaddr(ifa, NULL);
1435 atomic_inc(&net->ipv4.dev_addr_genid);
1436 if (!ifa->ifa_dev->ifa_list) {
1437 /* Last address was deleted from this interface.
1440 fib_disable_ip(dev, event, true);
1442 rt_cache_flush(dev_net(dev));
1449 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1451 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1452 struct netdev_notifier_changeupper_info *upper_info = ptr;
1453 struct netdev_notifier_info_ext *info_ext = ptr;
1454 struct in_device *in_dev;
1455 struct net *net = dev_net(dev);
1456 struct in_ifaddr *ifa;
1459 if (event == NETDEV_UNREGISTER) {
1460 fib_disable_ip(dev, event, true);
1465 in_dev = __in_dev_get_rtnl(dev);
1471 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1472 fib_add_ifaddr(ifa);
1474 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1475 fib_sync_up(dev, RTNH_F_DEAD);
1477 atomic_inc(&net->ipv4.dev_addr_genid);
1478 rt_cache_flush(net);
1481 fib_disable_ip(dev, event, false);
1484 flags = dev_get_flags(dev);
1485 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1486 fib_sync_up(dev, RTNH_F_LINKDOWN);
1488 fib_sync_down_dev(dev, event, false);
1489 rt_cache_flush(net);
1491 case NETDEV_CHANGEMTU:
1492 fib_sync_mtu(dev, info_ext->ext.mtu);
1493 rt_cache_flush(net);
1495 case NETDEV_CHANGEUPPER:
1497 /* flush all routes if dev is linked to or unlinked from
1498 * an L3 master device (e.g., VRF)
1500 if (upper_info->upper_dev &&
1501 netif_is_l3_master(upper_info->upper_dev))
1502 fib_disable_ip(dev, NETDEV_DOWN, true);
1508 static struct notifier_block fib_inetaddr_notifier = {
1509 .notifier_call = fib_inetaddr_event,
1512 static struct notifier_block fib_netdev_notifier = {
1513 .notifier_call = fib_netdev_event,
1516 static int __net_init ip_fib_net_init(struct net *net)
1519 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1521 err = fib4_notifier_init(net);
1525 /* Avoid false sharing : Use at least a full cache line */
1526 size = max_t(size_t, size, L1_CACHE_BYTES);
1528 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1529 if (!net->ipv4.fib_table_hash) {
1531 goto err_table_hash_alloc;
1534 err = fib4_rules_init(net);
1536 goto err_rules_init;
1540 kfree(net->ipv4.fib_table_hash);
1541 err_table_hash_alloc:
1542 fib4_notifier_exit(net);
1546 static void ip_fib_net_exit(struct net *net)
1551 #ifdef CONFIG_IP_MULTIPLE_TABLES
1552 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1553 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1555 /* Destroy the tables in reverse order to guarantee that the
1556 * local table, ID 255, is destroyed before the main table, ID
1557 * 254. This is necessary as the local table may contain
1558 * references to data contained in the main table.
1560 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1561 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1562 struct hlist_node *tmp;
1563 struct fib_table *tb;
1565 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1566 hlist_del(&tb->tb_hlist);
1567 fib_table_flush(net, tb, true);
1572 #ifdef CONFIG_IP_MULTIPLE_TABLES
1573 fib4_rules_exit(net);
1576 kfree(net->ipv4.fib_table_hash);
1577 fib4_notifier_exit(net);
1580 static int __net_init fib_net_init(struct net *net)
1584 #ifdef CONFIG_IP_ROUTE_CLASSID
1585 net->ipv4.fib_num_tclassid_users = 0;
1587 error = ip_fib_net_init(net);
1590 error = nl_fib_lookup_init(net);
1593 error = fib_proc_init(net);
1600 nl_fib_lookup_exit(net);
1602 ip_fib_net_exit(net);
1606 static void __net_exit fib_net_exit(struct net *net)
1609 nl_fib_lookup_exit(net);
1610 ip_fib_net_exit(net);
1613 static struct pernet_operations fib_net_ops = {
1614 .init = fib_net_init,
1615 .exit = fib_net_exit,
1618 void __init ip_fib_init(void)
1622 register_pernet_subsys(&fib_net_ops);
1624 register_netdevice_notifier(&fib_netdev_notifier);
1625 register_inetaddr_notifier(&fib_inetaddr_notifier);
1627 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1628 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1629 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);