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 = RT_TOS(ip_hdr(skb)->tos),
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
324 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
325 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
327 if (nhc->nhc_dev == dev) {
330 } else if (l3mdev_master_ifindex_rcu(nhc->nhc_dev) == dev->ifindex) {
336 if (fib_info_nhc(fi, 0)->nhc_dev == dev)
342 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
344 /* Given (packet source, input interface) and optional (dst, oif, tos):
345 * - (main) check, that source is valid i.e. not broadcast or our local
347 * - figure out what "logical" interface this packet arrived
348 * and calculate "specific destination" address.
349 * - check, that packet arrived from expected physical interface.
350 * called with rcu_read_lock()
352 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
353 u8 tos, int oif, struct net_device *dev,
354 int rpf, struct in_device *idev, u32 *itag)
356 struct net *net = dev_net(dev);
357 struct flow_keys flkeys;
359 struct fib_result res;
364 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
366 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
369 fl4.flowi4_tos = tos;
370 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
371 fl4.flowi4_tun_key.tun_id = 0;
372 fl4.flowi4_flags = 0;
373 fl4.flowi4_uid = sock_net_uid(net, NULL);
375 no_addr = idev->ifa_list == NULL;
377 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
378 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
379 fl4.flowi4_proto = 0;
384 if (fib_lookup(net, &fl4, &res, 0))
386 if (res.type != RTN_UNICAST &&
387 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
389 fib_combine_itag(itag, &res);
391 dev_match = fib_info_nh_uses_dev(res.fi, dev);
392 /* This is not common, loopback packets retain skb_dst so normally they
393 * would not even hit this slow path.
395 dev_match = dev_match || (res.type == RTN_LOCAL &&
396 dev == net->loopback_dev);
398 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
405 fl4.flowi4_oif = dev->ifindex;
408 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
409 if (res.type == RTN_UNICAST)
410 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
426 /* Ignore rp_filter for packets protected by IPsec. */
427 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
428 u8 tos, int oif, struct net_device *dev,
429 struct in_device *idev, u32 *itag)
431 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
432 struct net *net = dev_net(dev);
434 if (!r && !fib_num_tclassid_users(net) &&
435 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
436 if (IN_DEV_ACCEPT_LOCAL(idev))
438 /* with custom local routes in place, checking local addresses
439 * only will be too optimistic, with custom rules, checking
440 * local addresses only can be too strict, e.g. due to vrf
442 if (net->ipv4.fib_has_custom_local_routes ||
443 fib4_has_custom_rules(net))
445 if (inet_lookup_ifaddr_rcu(net, src))
454 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
457 static inline __be32 sk_extract_addr(struct sockaddr *addr)
459 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
462 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
466 nla = (struct nlattr *) ((char *) mx + len);
467 nla->nla_type = type;
468 nla->nla_len = nla_attr_size(4);
469 *(u32 *) nla_data(nla) = value;
471 return len + nla_total_size(4);
474 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
475 struct fib_config *cfg)
480 memset(cfg, 0, sizeof(*cfg));
481 cfg->fc_nlinfo.nl_net = net;
483 if (rt->rt_dst.sa_family != AF_INET)
484 return -EAFNOSUPPORT;
487 * Check mask for validity:
488 * a) it must be contiguous.
489 * b) destination must have all host bits clear.
490 * c) if application forgot to set correct family (AF_INET),
491 * reject request unless it is absolutely clear i.e.
492 * both family and mask are zero.
495 addr = sk_extract_addr(&rt->rt_dst);
496 if (!(rt->rt_flags & RTF_HOST)) {
497 __be32 mask = sk_extract_addr(&rt->rt_genmask);
499 if (rt->rt_genmask.sa_family != AF_INET) {
500 if (mask || rt->rt_genmask.sa_family)
501 return -EAFNOSUPPORT;
504 if (bad_mask(mask, addr))
507 plen = inet_mask_len(mask);
510 cfg->fc_dst_len = plen;
513 if (cmd != SIOCDELRT) {
514 cfg->fc_nlflags = NLM_F_CREATE;
515 cfg->fc_protocol = RTPROT_BOOT;
519 cfg->fc_priority = rt->rt_metric - 1;
521 if (rt->rt_flags & RTF_REJECT) {
522 cfg->fc_scope = RT_SCOPE_HOST;
523 cfg->fc_type = RTN_UNREACHABLE;
527 cfg->fc_scope = RT_SCOPE_NOWHERE;
528 cfg->fc_type = RTN_UNICAST;
532 struct net_device *dev;
533 char devname[IFNAMSIZ];
535 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
538 devname[IFNAMSIZ-1] = 0;
539 colon = strchr(devname, ':');
542 dev = __dev_get_by_name(net, devname);
545 cfg->fc_oif = dev->ifindex;
546 cfg->fc_table = l3mdev_fib_table(dev);
548 const struct in_ifaddr *ifa;
549 struct in_device *in_dev;
551 in_dev = __in_dev_get_rtnl(dev);
558 in_dev_for_each_ifa_rcu(ifa, in_dev) {
559 if (strcmp(ifa->ifa_label, devname) == 0)
566 cfg->fc_prefsrc = ifa->ifa_local;
570 addr = sk_extract_addr(&rt->rt_gateway);
571 if (rt->rt_gateway.sa_family == AF_INET && addr) {
572 unsigned int addr_type;
575 cfg->fc_gw_family = AF_INET;
576 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
577 if (rt->rt_flags & RTF_GATEWAY &&
578 addr_type == RTN_UNICAST)
579 cfg->fc_scope = RT_SCOPE_UNIVERSE;
582 if (cmd == SIOCDELRT)
585 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
588 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
589 cfg->fc_scope = RT_SCOPE_LINK;
591 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
595 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
599 if (rt->rt_flags & RTF_MTU)
600 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
602 if (rt->rt_flags & RTF_WINDOW)
603 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
605 if (rt->rt_flags & RTF_IRTT)
606 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
609 cfg->fc_mx_len = len;
616 * Handle IP routing ioctl calls.
617 * These are used to manipulate the routing tables
619 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
621 struct fib_config cfg;
625 case SIOCADDRT: /* Add a route */
626 case SIOCDELRT: /* Delete a route */
627 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
631 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
633 struct fib_table *tb;
635 if (cmd == SIOCDELRT) {
636 tb = fib_get_table(net, cfg.fc_table);
638 err = fib_table_delete(net, tb, &cfg,
643 tb = fib_new_table(net, cfg.fc_table);
645 err = fib_table_insert(net, tb,
651 /* allocated by rtentry_to_fib_config() */
660 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
661 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 },
662 [RTA_DST] = { .type = NLA_U32 },
663 [RTA_SRC] = { .type = NLA_U32 },
664 [RTA_IIF] = { .type = NLA_U32 },
665 [RTA_OIF] = { .type = NLA_U32 },
666 [RTA_GATEWAY] = { .type = NLA_U32 },
667 [RTA_PRIORITY] = { .type = NLA_U32 },
668 [RTA_PREFSRC] = { .type = NLA_U32 },
669 [RTA_METRICS] = { .type = NLA_NESTED },
670 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
671 [RTA_FLOW] = { .type = NLA_U32 },
672 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
673 [RTA_ENCAP] = { .type = NLA_NESTED },
674 [RTA_UID] = { .type = NLA_U32 },
675 [RTA_MARK] = { .type = NLA_U32 },
676 [RTA_TABLE] = { .type = NLA_U32 },
677 [RTA_IP_PROTO] = { .type = NLA_U8 },
678 [RTA_SPORT] = { .type = NLA_U16 },
679 [RTA_DPORT] = { .type = NLA_U16 },
680 [RTA_NH_ID] = { .type = NLA_U32 },
683 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
684 struct netlink_ext_ack *extack)
689 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
690 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
695 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
697 switch (via->rtvia_family) {
699 if (alen != sizeof(__be32)) {
700 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
703 cfg->fc_gw_family = AF_INET;
704 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
708 if (alen != sizeof(struct in6_addr)) {
709 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
712 cfg->fc_gw_family = AF_INET6;
713 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
715 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
720 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
727 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
728 struct nlmsghdr *nlh, struct fib_config *cfg,
729 struct netlink_ext_ack *extack)
731 bool has_gw = false, has_via = false;
736 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
737 rtm_ipv4_policy, extack);
741 memset(cfg, 0, sizeof(*cfg));
743 rtm = nlmsg_data(nlh);
744 cfg->fc_dst_len = rtm->rtm_dst_len;
745 cfg->fc_tos = rtm->rtm_tos;
746 cfg->fc_table = rtm->rtm_table;
747 cfg->fc_protocol = rtm->rtm_protocol;
748 cfg->fc_scope = rtm->rtm_scope;
749 cfg->fc_type = rtm->rtm_type;
750 cfg->fc_flags = rtm->rtm_flags;
751 cfg->fc_nlflags = nlh->nlmsg_flags;
753 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
754 cfg->fc_nlinfo.nlh = nlh;
755 cfg->fc_nlinfo.nl_net = net;
757 if (cfg->fc_type > RTN_MAX) {
758 NL_SET_ERR_MSG(extack, "Invalid route type");
763 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
764 switch (nla_type(attr)) {
766 cfg->fc_dst = nla_get_be32(attr);
769 cfg->fc_oif = nla_get_u32(attr);
773 cfg->fc_gw4 = nla_get_be32(attr);
775 cfg->fc_gw_family = AF_INET;
779 err = fib_gw_from_via(cfg, attr, extack);
784 cfg->fc_priority = nla_get_u32(attr);
787 cfg->fc_prefsrc = nla_get_be32(attr);
790 cfg->fc_mx = nla_data(attr);
791 cfg->fc_mx_len = nla_len(attr);
794 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
799 cfg->fc_mp = nla_data(attr);
800 cfg->fc_mp_len = nla_len(attr);
803 cfg->fc_flow = nla_get_u32(attr);
806 cfg->fc_table = nla_get_u32(attr);
809 cfg->fc_encap = attr;
812 cfg->fc_encap_type = nla_get_u16(attr);
813 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
819 cfg->fc_nh_id = nla_get_u32(attr);
825 if (cfg->fc_oif || cfg->fc_gw_family ||
826 cfg->fc_encap || cfg->fc_mp) {
827 NL_SET_ERR_MSG(extack,
828 "Nexthop specification and nexthop id are mutually exclusive");
833 if (has_gw && has_via) {
834 NL_SET_ERR_MSG(extack,
835 "Nexthop configuration can not contain both GATEWAY and VIA");
844 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
845 struct netlink_ext_ack *extack)
847 struct net *net = sock_net(skb->sk);
848 struct fib_config cfg;
849 struct fib_table *tb;
852 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
856 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
857 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
862 tb = fib_get_table(net, cfg.fc_table);
864 NL_SET_ERR_MSG(extack, "FIB table does not exist");
869 err = fib_table_delete(net, tb, &cfg, extack);
874 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
875 struct netlink_ext_ack *extack)
877 struct net *net = sock_net(skb->sk);
878 struct fib_config cfg;
879 struct fib_table *tb;
882 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
886 tb = fib_new_table(net, cfg.fc_table);
892 err = fib_table_insert(net, tb, &cfg, extack);
893 if (!err && cfg.fc_type == RTN_LOCAL)
894 net->ipv4.fib_has_custom_local_routes = true;
899 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
900 struct fib_dump_filter *filter,
901 struct netlink_callback *cb)
903 struct netlink_ext_ack *extack = cb->extack;
904 struct nlattr *tb[RTA_MAX + 1];
910 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
911 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
915 rtm = nlmsg_data(nlh);
916 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
918 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
922 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
923 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
926 if (rtm->rtm_flags & RTM_F_CLONED)
927 filter->dump_routes = false;
929 filter->dump_exceptions = false;
931 filter->dump_all_families = (rtm->rtm_family == AF_UNSPEC);
932 filter->flags = rtm->rtm_flags;
933 filter->protocol = rtm->rtm_protocol;
934 filter->rt_type = rtm->rtm_type;
935 filter->table_id = rtm->rtm_table;
937 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
938 rtm_ipv4_policy, extack);
942 for (i = 0; i <= RTA_MAX; ++i) {
950 filter->table_id = nla_get_u32(tb[i]);
953 ifindex = nla_get_u32(tb[i]);
954 filter->dev = __dev_get_by_index(net, ifindex);
959 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
964 if (filter->flags || filter->protocol || filter->rt_type ||
965 filter->table_id || filter->dev) {
966 filter->filter_set = 1;
967 cb->answer_flags = NLM_F_DUMP_FILTERED;
972 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
974 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
976 struct fib_dump_filter filter = { .dump_routes = true,
977 .dump_exceptions = true };
978 const struct nlmsghdr *nlh = cb->nlh;
979 struct net *net = sock_net(skb->sk);
981 unsigned int e = 0, s_e;
982 struct fib_table *tb;
983 struct hlist_head *head;
986 if (cb->strict_check) {
987 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
990 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
991 struct rtmsg *rtm = nlmsg_data(nlh);
993 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
996 /* ipv4 does not use prefix flag */
997 if (filter.flags & RTM_F_PREFIX)
1000 if (filter.table_id) {
1001 tb = fib_get_table(net, filter.table_id);
1003 if (filter.dump_all_families)
1006 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1010 err = fib_table_dump(tb, skb, cb, &filter);
1011 return skb->len ? : err;
1019 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1021 head = &net->ipv4.fib_table_hash[h];
1022 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1026 memset(&cb->args[2], 0, sizeof(cb->args) -
1027 2 * sizeof(cb->args[0]));
1028 err = fib_table_dump(tb, skb, cb, &filter);
1030 if (likely(skb->len))
1051 /* Prepare and feed intra-kernel routing request.
1052 * Really, it should be netlink message, but :-( netlink
1053 * can be not configured, so that we feed it directly
1054 * to fib engine. It is legal, because all events occur
1055 * only when netlink is already locked.
1057 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1058 struct in_ifaddr *ifa, u32 rt_priority)
1060 struct net *net = dev_net(ifa->ifa_dev->dev);
1061 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1062 struct fib_table *tb;
1063 struct fib_config cfg = {
1064 .fc_protocol = RTPROT_KERNEL,
1067 .fc_dst_len = dst_len,
1068 .fc_priority = rt_priority,
1069 .fc_prefsrc = ifa->ifa_local,
1070 .fc_oif = ifa->ifa_dev->dev->ifindex,
1071 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1078 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1080 tb = fib_new_table(net, tb_id);
1084 cfg.fc_table = tb->tb_id;
1086 if (type != RTN_LOCAL)
1087 cfg.fc_scope = RT_SCOPE_LINK;
1089 cfg.fc_scope = RT_SCOPE_HOST;
1091 if (cmd == RTM_NEWROUTE)
1092 fib_table_insert(net, tb, &cfg, NULL);
1094 fib_table_delete(net, tb, &cfg, NULL);
1097 void fib_add_ifaddr(struct in_ifaddr *ifa)
1099 struct in_device *in_dev = ifa->ifa_dev;
1100 struct net_device *dev = in_dev->dev;
1101 struct in_ifaddr *prim = ifa;
1102 __be32 mask = ifa->ifa_mask;
1103 __be32 addr = ifa->ifa_local;
1104 __be32 prefix = ifa->ifa_address & mask;
1106 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1107 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1109 pr_warn("%s: bug: prim == NULL\n", __func__);
1114 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1116 if (!(dev->flags & IFF_UP))
1119 /* Add broadcast address, if it is explicitly assigned. */
1120 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
1121 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1124 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1125 (prefix != addr || ifa->ifa_prefixlen < 32)) {
1126 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1127 fib_magic(RTM_NEWROUTE,
1128 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1129 prefix, ifa->ifa_prefixlen, prim,
1130 ifa->ifa_rt_priority);
1132 /* Add network specific broadcasts, when it takes a sense */
1133 if (ifa->ifa_prefixlen < 31) {
1134 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32,
1136 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1142 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1144 __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1145 struct in_device *in_dev = ifa->ifa_dev;
1146 struct net_device *dev = in_dev->dev;
1148 if (!(dev->flags & IFF_UP) ||
1149 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1150 ipv4_is_zeronet(prefix) ||
1151 prefix == ifa->ifa_local || ifa->ifa_prefixlen == 32)
1155 fib_magic(RTM_NEWROUTE,
1156 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1157 prefix, ifa->ifa_prefixlen, ifa, new_metric);
1159 /* delete the old */
1160 fib_magic(RTM_DELROUTE,
1161 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1162 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1165 /* Delete primary or secondary address.
1166 * Optionally, on secondary address promotion consider the addresses
1167 * from subnet iprim as deleted, even if they are in device list.
1168 * In this case the secondary ifa can be in device list.
1170 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1172 struct in_device *in_dev = ifa->ifa_dev;
1173 struct net_device *dev = in_dev->dev;
1174 struct in_ifaddr *ifa1;
1175 struct in_ifaddr *prim = ifa, *prim1 = NULL;
1176 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1177 __be32 any = ifa->ifa_address & ifa->ifa_mask;
1182 unsigned int ok = 0;
1183 int subnet = 0; /* Primary network */
1184 int gone = 1; /* Address is missing */
1185 int same_prefsrc = 0; /* Another primary with same IP */
1187 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1188 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1190 /* if the device has been deleted, we don't perform
1194 pr_warn("%s: bug: prim == NULL\n", __func__);
1197 if (iprim && iprim != prim) {
1198 pr_warn("%s: bug: iprim != prim\n", __func__);
1201 } else if (!ipv4_is_zeronet(any) &&
1202 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1203 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1204 fib_magic(RTM_DELROUTE,
1205 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1206 any, ifa->ifa_prefixlen, prim, 0);
1213 /* Deletion is more complicated than add.
1214 * We should take care of not to delete too much :-)
1216 * Scan address list to be sure that addresses are really gone.
1219 in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1221 /* promotion, keep the IP */
1225 /* Ignore IFAs from our subnet */
1226 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1227 inet_ifa_match(ifa1->ifa_address, iprim))
1230 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1231 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1232 /* Another address from our subnet? */
1233 if (ifa1->ifa_mask == prim->ifa_mask &&
1234 inet_ifa_match(ifa1->ifa_address, prim))
1237 /* We reached the secondaries, so
1238 * same_prefsrc should be determined.
1242 /* Search new prim1 if ifa1 is not
1243 * using the current prim1
1246 ifa1->ifa_mask != prim1->ifa_mask ||
1247 !inet_ifa_match(ifa1->ifa_address, prim1))
1248 prim1 = inet_ifa_byprefix(in_dev,
1253 if (prim1->ifa_local != prim->ifa_local)
1257 if (prim->ifa_local != ifa1->ifa_local)
1263 if (ifa->ifa_local == ifa1->ifa_local)
1265 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1267 if (brd == ifa1->ifa_broadcast)
1269 if (any == ifa1->ifa_broadcast)
1271 /* primary has network specific broadcasts */
1272 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1273 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1274 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1276 if (!ipv4_is_zeronet(any1)) {
1277 if (ifa->ifa_broadcast == brd1 ||
1278 ifa->ifa_broadcast == any1)
1280 if (brd == brd1 || brd == any1)
1282 if (any == brd1 || any == any1)
1291 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1293 if (subnet && ifa->ifa_prefixlen < 31) {
1294 if (!(ok & BRD1_OK))
1295 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1297 if (!(ok & BRD0_OK))
1298 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1301 if (!(ok & LOCAL_OK)) {
1302 unsigned int addr_type;
1304 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1306 /* Check, that this local address finally disappeared. */
1307 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1309 if (gone && addr_type != RTN_LOCAL) {
1310 /* And the last, but not the least thing.
1311 * We must flush stray FIB entries.
1313 * First of all, we scan fib_info list searching
1314 * for stray nexthop entries, then ignite fib_flush.
1316 if (fib_sync_down_addr(dev, ifa->ifa_local))
1317 fib_flush(dev_net(dev));
1326 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1329 struct fib_result res;
1330 struct flowi4 fl4 = {
1331 .flowi4_mark = frn->fl_mark,
1332 .daddr = frn->fl_addr,
1333 .flowi4_tos = frn->fl_tos,
1334 .flowi4_scope = frn->fl_scope,
1336 struct fib_table *tb;
1340 tb = fib_get_table(net, frn->tb_id_in);
1346 frn->tb_id = tb->tb_id;
1347 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1350 frn->prefixlen = res.prefixlen;
1351 frn->nh_sel = res.nh_sel;
1352 frn->type = res.type;
1353 frn->scope = res.scope;
1361 static void nl_fib_input(struct sk_buff *skb)
1364 struct fib_result_nl *frn;
1365 struct nlmsghdr *nlh;
1368 net = sock_net(skb->sk);
1369 nlh = nlmsg_hdr(skb);
1370 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1371 skb->len < nlh->nlmsg_len ||
1372 nlmsg_len(nlh) < sizeof(*frn))
1375 skb = netlink_skb_clone(skb, GFP_KERNEL);
1378 nlh = nlmsg_hdr(skb);
1380 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1381 nl_fib_lookup(net, frn);
1383 portid = NETLINK_CB(skb).portid; /* netlink portid */
1384 NETLINK_CB(skb).portid = 0; /* from kernel */
1385 NETLINK_CB(skb).dst_group = 0; /* unicast */
1386 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1389 static int __net_init nl_fib_lookup_init(struct net *net)
1392 struct netlink_kernel_cfg cfg = {
1393 .input = nl_fib_input,
1396 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1398 return -EAFNOSUPPORT;
1399 net->ipv4.fibnl = sk;
1403 static void nl_fib_lookup_exit(struct net *net)
1405 netlink_kernel_release(net->ipv4.fibnl);
1406 net->ipv4.fibnl = NULL;
1409 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1412 if (fib_sync_down_dev(dev, event, force))
1413 fib_flush(dev_net(dev));
1415 rt_cache_flush(dev_net(dev));
1419 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1421 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1422 struct net_device *dev = ifa->ifa_dev->dev;
1423 struct net *net = dev_net(dev);
1427 fib_add_ifaddr(ifa);
1428 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1429 fib_sync_up(dev, RTNH_F_DEAD);
1431 atomic_inc(&net->ipv4.dev_addr_genid);
1432 rt_cache_flush(dev_net(dev));
1435 fib_del_ifaddr(ifa, NULL);
1436 atomic_inc(&net->ipv4.dev_addr_genid);
1437 if (!ifa->ifa_dev->ifa_list) {
1438 /* Last address was deleted from this interface.
1441 fib_disable_ip(dev, event, true);
1443 rt_cache_flush(dev_net(dev));
1450 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1452 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1453 struct netdev_notifier_changeupper_info *upper_info = ptr;
1454 struct netdev_notifier_info_ext *info_ext = ptr;
1455 struct in_device *in_dev;
1456 struct net *net = dev_net(dev);
1457 struct in_ifaddr *ifa;
1460 if (event == NETDEV_UNREGISTER) {
1461 fib_disable_ip(dev, event, true);
1466 in_dev = __in_dev_get_rtnl(dev);
1472 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1473 fib_add_ifaddr(ifa);
1475 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1476 fib_sync_up(dev, RTNH_F_DEAD);
1478 atomic_inc(&net->ipv4.dev_addr_genid);
1479 rt_cache_flush(net);
1482 fib_disable_ip(dev, event, false);
1485 flags = dev_get_flags(dev);
1486 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1487 fib_sync_up(dev, RTNH_F_LINKDOWN);
1489 fib_sync_down_dev(dev, event, false);
1490 rt_cache_flush(net);
1492 case NETDEV_CHANGEMTU:
1493 fib_sync_mtu(dev, info_ext->ext.mtu);
1494 rt_cache_flush(net);
1496 case NETDEV_CHANGEUPPER:
1498 /* flush all routes if dev is linked to or unlinked from
1499 * an L3 master device (e.g., VRF)
1501 if (upper_info->upper_dev &&
1502 netif_is_l3_master(upper_info->upper_dev))
1503 fib_disable_ip(dev, NETDEV_DOWN, true);
1509 static struct notifier_block fib_inetaddr_notifier = {
1510 .notifier_call = fib_inetaddr_event,
1513 static struct notifier_block fib_netdev_notifier = {
1514 .notifier_call = fib_netdev_event,
1517 static int __net_init ip_fib_net_init(struct net *net)
1520 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1522 err = fib4_notifier_init(net);
1526 /* Avoid false sharing : Use at least a full cache line */
1527 size = max_t(size_t, size, L1_CACHE_BYTES);
1529 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1530 if (!net->ipv4.fib_table_hash) {
1532 goto err_table_hash_alloc;
1535 err = fib4_rules_init(net);
1537 goto err_rules_init;
1541 kfree(net->ipv4.fib_table_hash);
1542 err_table_hash_alloc:
1543 fib4_notifier_exit(net);
1547 static void ip_fib_net_exit(struct net *net)
1552 #ifdef CONFIG_IP_MULTIPLE_TABLES
1553 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1554 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1556 /* Destroy the tables in reverse order to guarantee that the
1557 * local table, ID 255, is destroyed before the main table, ID
1558 * 254. This is necessary as the local table may contain
1559 * references to data contained in the main table.
1561 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1562 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1563 struct hlist_node *tmp;
1564 struct fib_table *tb;
1566 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1567 hlist_del(&tb->tb_hlist);
1568 fib_table_flush(net, tb, true);
1573 #ifdef CONFIG_IP_MULTIPLE_TABLES
1574 fib4_rules_exit(net);
1577 kfree(net->ipv4.fib_table_hash);
1578 fib4_notifier_exit(net);
1581 static int __net_init fib_net_init(struct net *net)
1585 #ifdef CONFIG_IP_ROUTE_CLASSID
1586 net->ipv4.fib_num_tclassid_users = 0;
1588 error = ip_fib_net_init(net);
1591 error = nl_fib_lookup_init(net);
1594 error = fib_proc_init(net);
1601 nl_fib_lookup_exit(net);
1603 ip_fib_net_exit(net);
1607 static void __net_exit fib_net_exit(struct net *net)
1610 nl_fib_lookup_exit(net);
1611 ip_fib_net_exit(net);
1614 static struct pernet_operations fib_net_ops = {
1615 .init = fib_net_init,
1616 .exit = fib_net_exit,
1619 void __init ip_fib_init(void)
1623 register_pernet_subsys(&fib_net_ops);
1625 register_netdevice_notifier(&fib_netdev_notifier);
1626 register_inetaddr_notifier(&fib_inetaddr_notifier);
1628 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1629 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1630 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);