2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * IPv4 Forwarding Information Base: FIB frontend.
8 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
16 #include <linux/module.h>
17 #include <linux/uaccess.h>
18 #include <linux/bitops.h>
19 #include <linux/capability.h>
20 #include <linux/types.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/errno.h>
28 #include <linux/inet.h>
29 #include <linux/inetdevice.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_addr.h>
32 #include <linux/if_arp.h>
33 #include <linux/skbuff.h>
34 #include <linux/cache.h>
35 #include <linux/init.h>
36 #include <linux/list.h>
37 #include <linux/slab.h>
40 #include <net/protocol.h>
41 #include <net/route.h>
45 #include <net/ip_fib.h>
46 #include <net/rtnetlink.h>
48 #include <net/l3mdev.h>
49 #include <net/lwtunnel.h>
50 #include <trace/events/fib.h>
52 #ifndef CONFIG_IP_MULTIPLE_TABLES
54 static int __net_init fib4_rules_init(struct net *net)
56 struct fib_table *local_table, *main_table;
58 main_table = fib_trie_table(RT_TABLE_MAIN, NULL);
62 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
66 hlist_add_head_rcu(&local_table->tb_hlist,
67 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
68 hlist_add_head_rcu(&main_table->tb_hlist,
69 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
73 fib_free_table(main_table);
78 struct fib_table *fib_new_table(struct net *net, u32 id)
80 struct fib_table *tb, *alias = NULL;
85 tb = fib_get_table(net, id);
89 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
90 alias = fib_new_table(net, RT_TABLE_MAIN);
92 tb = fib_trie_table(id, alias);
98 rcu_assign_pointer(net->ipv4.fib_main, tb);
100 case RT_TABLE_DEFAULT:
101 rcu_assign_pointer(net->ipv4.fib_default, tb);
107 h = id & (FIB_TABLE_HASHSZ - 1);
108 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
111 EXPORT_SYMBOL_GPL(fib_new_table);
113 /* caller must hold either rtnl or rcu read lock */
114 struct fib_table *fib_get_table(struct net *net, u32 id)
116 struct fib_table *tb;
117 struct hlist_head *head;
122 h = id & (FIB_TABLE_HASHSZ - 1);
124 head = &net->ipv4.fib_table_hash[h];
125 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
131 #endif /* CONFIG_IP_MULTIPLE_TABLES */
133 static void fib_replace_table(struct net *net, struct fib_table *old,
134 struct fib_table *new)
136 #ifdef CONFIG_IP_MULTIPLE_TABLES
137 switch (new->tb_id) {
139 rcu_assign_pointer(net->ipv4.fib_main, new);
141 case RT_TABLE_DEFAULT:
142 rcu_assign_pointer(net->ipv4.fib_default, new);
149 /* replace the old table in the hlist */
150 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
153 int fib_unmerge(struct net *net)
155 struct fib_table *old, *new, *main_table;
157 /* attempt to fetch local table if it has been allocated */
158 old = fib_get_table(net, RT_TABLE_LOCAL);
162 new = fib_trie_unmerge(old);
166 /* table is already unmerged */
170 /* replace merged table with clean table */
171 fib_replace_table(net, old, new);
174 /* attempt to fetch main table if it has been allocated */
175 main_table = fib_get_table(net, RT_TABLE_MAIN);
179 /* flush local entries from main table */
180 fib_table_flush_external(main_table);
185 static void fib_flush(struct net *net)
190 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
191 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
192 struct hlist_node *tmp;
193 struct fib_table *tb;
195 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
196 flushed += fib_table_flush(net, tb, false);
204 * Find address type as if only "dev" was present in the system. If
205 * on_dev is NULL then all interfaces are taken into consideration.
207 static inline unsigned int __inet_dev_addr_type(struct net *net,
208 const struct net_device *dev,
209 __be32 addr, u32 tb_id)
211 struct flowi4 fl4 = { .daddr = addr };
212 struct fib_result res;
213 unsigned int ret = RTN_BROADCAST;
214 struct fib_table *table;
216 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
217 return RTN_BROADCAST;
218 if (ipv4_is_multicast(addr))
219 return RTN_MULTICAST;
223 table = fib_get_table(net, tb_id);
226 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
227 if (!dev || dev == res.fi->fib_dev)
236 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
238 return __inet_dev_addr_type(net, NULL, addr, tb_id);
240 EXPORT_SYMBOL(inet_addr_type_table);
242 unsigned int inet_addr_type(struct net *net, __be32 addr)
244 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
246 EXPORT_SYMBOL(inet_addr_type);
248 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
251 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
253 return __inet_dev_addr_type(net, dev, addr, rt_table);
255 EXPORT_SYMBOL(inet_dev_addr_type);
257 /* inet_addr_type with dev == NULL but using the table from a dev
258 * if one is associated
260 unsigned int inet_addr_type_dev_table(struct net *net,
261 const struct net_device *dev,
264 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
266 return __inet_dev_addr_type(net, NULL, addr, rt_table);
268 EXPORT_SYMBOL(inet_addr_type_dev_table);
270 __be32 fib_compute_spec_dst(struct sk_buff *skb)
272 struct net_device *dev = skb->dev;
273 struct in_device *in_dev;
274 struct fib_result res;
279 rt = skb_rtable(skb);
280 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
282 return ip_hdr(skb)->daddr;
284 in_dev = __in_dev_get_rcu(dev);
288 scope = RT_SCOPE_UNIVERSE;
289 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
290 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
291 struct flowi4 fl4 = {
292 .flowi4_iif = LOOPBACK_IFINDEX,
293 .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
294 .daddr = ip_hdr(skb)->saddr,
295 .flowi4_tos = ip_hdr(skb)->tos & IPTOS_RT_MASK,
296 .flowi4_scope = scope,
297 .flowi4_mark = vmark ? skb->mark : 0,
299 if (!fib_lookup(net, &fl4, &res, 0))
300 return FIB_RES_PREFSRC(net, res);
302 scope = RT_SCOPE_LINK;
305 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
308 /* Given (packet source, input interface) and optional (dst, oif, tos):
309 * - (main) check, that source is valid i.e. not broadcast or our local
311 * - figure out what "logical" interface this packet arrived
312 * and calculate "specific destination" address.
313 * - check, that packet arrived from expected physical interface.
314 * called with rcu_read_lock()
316 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
317 u8 tos, int oif, struct net_device *dev,
318 int rpf, struct in_device *idev, u32 *itag)
321 struct fib_result res;
323 struct net *net = dev_net(dev);
327 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
329 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
332 fl4.flowi4_tos = tos;
333 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
334 fl4.flowi4_tun_key.tun_id = 0;
335 fl4.flowi4_flags = 0;
336 fl4.flowi4_uid = sock_net_uid(net, NULL);
338 no_addr = idev->ifa_list == NULL;
340 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
342 trace_fib_validate_source(dev, &fl4);
344 if (fib_lookup(net, &fl4, &res, 0))
346 if (res.type != RTN_UNICAST &&
347 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
349 if (!rpf && !fib_num_tclassid_users(net) &&
350 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
352 fib_combine_itag(itag, &res);
355 #ifdef CONFIG_IP_ROUTE_MULTIPATH
356 for (ret = 0; ret < res.fi->fib_nhs; ret++) {
357 struct fib_nh *nh = &res.fi->fib_nh[ret];
359 if (nh->nh_dev == dev) {
362 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
368 if (FIB_RES_DEV(res) == dev)
372 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
379 fl4.flowi4_oif = dev->ifindex;
382 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
383 if (res.type == RTN_UNICAST)
384 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
400 /* Ignore rp_filter for packets protected by IPsec. */
401 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
402 u8 tos, int oif, struct net_device *dev,
403 struct in_device *idev, u32 *itag)
405 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
407 if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
408 IN_DEV_ACCEPT_LOCAL(idev) &&
409 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
413 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
416 static inline __be32 sk_extract_addr(struct sockaddr *addr)
418 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
421 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
425 nla = (struct nlattr *) ((char *) mx + len);
426 nla->nla_type = type;
427 nla->nla_len = nla_attr_size(4);
428 *(u32 *) nla_data(nla) = value;
430 return len + nla_total_size(4);
433 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
434 struct fib_config *cfg)
439 memset(cfg, 0, sizeof(*cfg));
440 cfg->fc_nlinfo.nl_net = net;
442 if (rt->rt_dst.sa_family != AF_INET)
443 return -EAFNOSUPPORT;
446 * Check mask for validity:
447 * a) it must be contiguous.
448 * b) destination must have all host bits clear.
449 * c) if application forgot to set correct family (AF_INET),
450 * reject request unless it is absolutely clear i.e.
451 * both family and mask are zero.
454 addr = sk_extract_addr(&rt->rt_dst);
455 if (!(rt->rt_flags & RTF_HOST)) {
456 __be32 mask = sk_extract_addr(&rt->rt_genmask);
458 if (rt->rt_genmask.sa_family != AF_INET) {
459 if (mask || rt->rt_genmask.sa_family)
460 return -EAFNOSUPPORT;
463 if (bad_mask(mask, addr))
466 plen = inet_mask_len(mask);
469 cfg->fc_dst_len = plen;
472 if (cmd != SIOCDELRT) {
473 cfg->fc_nlflags = NLM_F_CREATE;
474 cfg->fc_protocol = RTPROT_BOOT;
478 cfg->fc_priority = rt->rt_metric - 1;
480 if (rt->rt_flags & RTF_REJECT) {
481 cfg->fc_scope = RT_SCOPE_HOST;
482 cfg->fc_type = RTN_UNREACHABLE;
486 cfg->fc_scope = RT_SCOPE_NOWHERE;
487 cfg->fc_type = RTN_UNICAST;
491 struct net_device *dev;
492 char devname[IFNAMSIZ];
494 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
497 devname[IFNAMSIZ-1] = 0;
498 colon = strchr(devname, ':');
501 dev = __dev_get_by_name(net, devname);
504 cfg->fc_oif = dev->ifindex;
505 cfg->fc_table = l3mdev_fib_table(dev);
507 struct in_ifaddr *ifa;
508 struct in_device *in_dev = __in_dev_get_rtnl(dev);
512 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
513 if (strcmp(ifa->ifa_label, devname) == 0)
517 cfg->fc_prefsrc = ifa->ifa_local;
521 addr = sk_extract_addr(&rt->rt_gateway);
522 if (rt->rt_gateway.sa_family == AF_INET && addr) {
523 unsigned int addr_type;
526 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
527 if (rt->rt_flags & RTF_GATEWAY &&
528 addr_type == RTN_UNICAST)
529 cfg->fc_scope = RT_SCOPE_UNIVERSE;
533 cfg->fc_table = RT_TABLE_MAIN;
535 if (cmd == SIOCDELRT)
538 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
541 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
542 cfg->fc_scope = RT_SCOPE_LINK;
544 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
548 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
552 if (rt->rt_flags & RTF_MTU)
553 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
555 if (rt->rt_flags & RTF_WINDOW)
556 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
558 if (rt->rt_flags & RTF_IRTT)
559 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
562 cfg->fc_mx_len = len;
569 * Handle IP routing ioctl calls.
570 * These are used to manipulate the routing tables
572 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
574 struct fib_config cfg;
579 case SIOCADDRT: /* Add a route */
580 case SIOCDELRT: /* Delete a route */
581 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
584 if (copy_from_user(&rt, arg, sizeof(rt)))
588 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
590 struct fib_table *tb;
592 if (cmd == SIOCDELRT) {
593 tb = fib_get_table(net, cfg.fc_table);
595 err = fib_table_delete(net, tb, &cfg,
600 tb = fib_new_table(net, cfg.fc_table);
602 err = fib_table_insert(net, tb,
608 /* allocated by rtentry_to_fib_config() */
617 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
618 [RTA_DST] = { .type = NLA_U32 },
619 [RTA_SRC] = { .type = NLA_U32 },
620 [RTA_IIF] = { .type = NLA_U32 },
621 [RTA_OIF] = { .type = NLA_U32 },
622 [RTA_GATEWAY] = { .type = NLA_U32 },
623 [RTA_PRIORITY] = { .type = NLA_U32 },
624 [RTA_PREFSRC] = { .type = NLA_U32 },
625 [RTA_METRICS] = { .type = NLA_NESTED },
626 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
627 [RTA_FLOW] = { .type = NLA_U32 },
628 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
629 [RTA_ENCAP] = { .type = NLA_NESTED },
630 [RTA_UID] = { .type = NLA_U32 },
631 [RTA_MARK] = { .type = NLA_U32 },
632 [RTA_TABLE] = { .type = NLA_U32 },
635 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
636 struct nlmsghdr *nlh, struct fib_config *cfg,
637 struct netlink_ext_ack *extack)
643 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy,
648 memset(cfg, 0, sizeof(*cfg));
650 rtm = nlmsg_data(nlh);
651 cfg->fc_dst_len = rtm->rtm_dst_len;
652 cfg->fc_tos = rtm->rtm_tos;
653 cfg->fc_table = rtm->rtm_table;
654 cfg->fc_protocol = rtm->rtm_protocol;
655 cfg->fc_scope = rtm->rtm_scope;
656 cfg->fc_type = rtm->rtm_type;
657 cfg->fc_flags = rtm->rtm_flags;
658 cfg->fc_nlflags = nlh->nlmsg_flags;
660 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
661 cfg->fc_nlinfo.nlh = nlh;
662 cfg->fc_nlinfo.nl_net = net;
664 if (cfg->fc_type > RTN_MAX) {
665 NL_SET_ERR_MSG(extack, "Invalid route type");
670 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
671 switch (nla_type(attr)) {
673 cfg->fc_dst = nla_get_be32(attr);
676 cfg->fc_oif = nla_get_u32(attr);
679 cfg->fc_gw = nla_get_be32(attr);
682 NL_SET_ERR_MSG(extack, "IPv4 does not support RTA_VIA attribute");
686 cfg->fc_priority = nla_get_u32(attr);
689 cfg->fc_prefsrc = nla_get_be32(attr);
692 cfg->fc_mx = nla_data(attr);
693 cfg->fc_mx_len = nla_len(attr);
696 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
701 cfg->fc_mp = nla_data(attr);
702 cfg->fc_mp_len = nla_len(attr);
705 cfg->fc_flow = nla_get_u32(attr);
708 cfg->fc_table = nla_get_u32(attr);
711 cfg->fc_encap = attr;
714 cfg->fc_encap_type = nla_get_u16(attr);
715 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
728 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
729 struct netlink_ext_ack *extack)
731 struct net *net = sock_net(skb->sk);
732 struct fib_config cfg;
733 struct fib_table *tb;
736 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
740 tb = fib_get_table(net, cfg.fc_table);
742 NL_SET_ERR_MSG(extack, "FIB table does not exist");
747 err = fib_table_delete(net, tb, &cfg, extack);
752 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
753 struct netlink_ext_ack *extack)
755 struct net *net = sock_net(skb->sk);
756 struct fib_config cfg;
757 struct fib_table *tb;
760 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
764 tb = fib_new_table(net, cfg.fc_table);
770 err = fib_table_insert(net, tb, &cfg, extack);
775 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
777 struct net *net = sock_net(skb->sk);
779 unsigned int e = 0, s_e;
780 struct fib_table *tb;
781 struct hlist_head *head;
784 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
785 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
793 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
795 head = &net->ipv4.fib_table_hash[h];
796 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
800 memset(&cb->args[2], 0, sizeof(cb->args) -
801 2 * sizeof(cb->args[0]));
802 err = fib_table_dump(tb, skb, cb);
804 if (likely(skb->len))
825 /* Prepare and feed intra-kernel routing request.
826 * Really, it should be netlink message, but :-( netlink
827 * can be not configured, so that we feed it directly
828 * to fib engine. It is legal, because all events occur
829 * only when netlink is already locked.
831 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
833 struct net *net = dev_net(ifa->ifa_dev->dev);
834 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
835 struct fib_table *tb;
836 struct fib_config cfg = {
837 .fc_protocol = RTPROT_KERNEL,
840 .fc_dst_len = dst_len,
841 .fc_prefsrc = ifa->ifa_local,
842 .fc_oif = ifa->ifa_dev->dev->ifindex,
843 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
850 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
852 tb = fib_new_table(net, tb_id);
856 cfg.fc_table = tb->tb_id;
858 if (type != RTN_LOCAL)
859 cfg.fc_scope = RT_SCOPE_LINK;
861 cfg.fc_scope = RT_SCOPE_HOST;
863 if (cmd == RTM_NEWROUTE)
864 fib_table_insert(net, tb, &cfg, NULL);
866 fib_table_delete(net, tb, &cfg, NULL);
869 void fib_add_ifaddr(struct in_ifaddr *ifa)
871 struct in_device *in_dev = ifa->ifa_dev;
872 struct net_device *dev = in_dev->dev;
873 struct in_ifaddr *prim = ifa;
874 __be32 mask = ifa->ifa_mask;
875 __be32 addr = ifa->ifa_local;
876 __be32 prefix = ifa->ifa_address & mask;
878 if (ifa->ifa_flags & IFA_F_SECONDARY) {
879 prim = inet_ifa_byprefix(in_dev, prefix, mask);
881 pr_warn("%s: bug: prim == NULL\n", __func__);
886 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
888 if (!(dev->flags & IFF_UP))
891 /* Add broadcast address, if it is explicitly assigned. */
892 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
893 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
895 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
896 (prefix != addr || ifa->ifa_prefixlen < 32)) {
897 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
898 fib_magic(RTM_NEWROUTE,
899 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
900 prefix, ifa->ifa_prefixlen, prim);
902 /* Add network specific broadcasts, when it takes a sense */
903 if (ifa->ifa_prefixlen < 31) {
904 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
905 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
911 /* Delete primary or secondary address.
912 * Optionally, on secondary address promotion consider the addresses
913 * from subnet iprim as deleted, even if they are in device list.
914 * In this case the secondary ifa can be in device list.
916 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
918 struct in_device *in_dev = ifa->ifa_dev;
919 struct net_device *dev = in_dev->dev;
920 struct in_ifaddr *ifa1;
921 struct in_ifaddr *prim = ifa, *prim1 = NULL;
922 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
923 __be32 any = ifa->ifa_address & ifa->ifa_mask;
929 int subnet = 0; /* Primary network */
930 int gone = 1; /* Address is missing */
931 int same_prefsrc = 0; /* Another primary with same IP */
933 if (ifa->ifa_flags & IFA_F_SECONDARY) {
934 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
936 /* if the device has been deleted, we don't perform
940 pr_warn("%s: bug: prim == NULL\n", __func__);
943 if (iprim && iprim != prim) {
944 pr_warn("%s: bug: iprim != prim\n", __func__);
947 } else if (!ipv4_is_zeronet(any) &&
948 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
949 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
950 fib_magic(RTM_DELROUTE,
951 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
952 any, ifa->ifa_prefixlen, prim);
959 /* Deletion is more complicated than add.
960 * We should take care of not to delete too much :-)
962 * Scan address list to be sure that addresses are really gone.
965 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
967 /* promotion, keep the IP */
971 /* Ignore IFAs from our subnet */
972 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
973 inet_ifa_match(ifa1->ifa_address, iprim))
976 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
977 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
978 /* Another address from our subnet? */
979 if (ifa1->ifa_mask == prim->ifa_mask &&
980 inet_ifa_match(ifa1->ifa_address, prim))
983 /* We reached the secondaries, so
984 * same_prefsrc should be determined.
988 /* Search new prim1 if ifa1 is not
989 * using the current prim1
992 ifa1->ifa_mask != prim1->ifa_mask ||
993 !inet_ifa_match(ifa1->ifa_address, prim1))
994 prim1 = inet_ifa_byprefix(in_dev,
999 if (prim1->ifa_local != prim->ifa_local)
1003 if (prim->ifa_local != ifa1->ifa_local)
1009 if (ifa->ifa_local == ifa1->ifa_local)
1011 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1013 if (brd == ifa1->ifa_broadcast)
1015 if (any == ifa1->ifa_broadcast)
1017 /* primary has network specific broadcasts */
1018 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1019 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1020 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1022 if (!ipv4_is_zeronet(any1)) {
1023 if (ifa->ifa_broadcast == brd1 ||
1024 ifa->ifa_broadcast == any1)
1026 if (brd == brd1 || brd == any1)
1028 if (any == brd1 || any == any1)
1036 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1037 if (subnet && ifa->ifa_prefixlen < 31) {
1038 if (!(ok & BRD1_OK))
1039 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1040 if (!(ok & BRD0_OK))
1041 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1043 if (!(ok & LOCAL_OK)) {
1044 unsigned int addr_type;
1046 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1048 /* Check, that this local address finally disappeared. */
1049 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1051 if (gone && addr_type != RTN_LOCAL) {
1052 /* And the last, but not the least thing.
1053 * We must flush stray FIB entries.
1055 * First of all, we scan fib_info list searching
1056 * for stray nexthop entries, then ignite fib_flush.
1058 if (fib_sync_down_addr(dev, ifa->ifa_local))
1059 fib_flush(dev_net(dev));
1068 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1071 struct fib_result res;
1072 struct flowi4 fl4 = {
1073 .flowi4_mark = frn->fl_mark,
1074 .daddr = frn->fl_addr,
1075 .flowi4_tos = frn->fl_tos,
1076 .flowi4_scope = frn->fl_scope,
1078 struct fib_table *tb;
1082 tb = fib_get_table(net, frn->tb_id_in);
1088 frn->tb_id = tb->tb_id;
1089 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1092 frn->prefixlen = res.prefixlen;
1093 frn->nh_sel = res.nh_sel;
1094 frn->type = res.type;
1095 frn->scope = res.scope;
1103 static void nl_fib_input(struct sk_buff *skb)
1106 struct fib_result_nl *frn;
1107 struct nlmsghdr *nlh;
1110 net = sock_net(skb->sk);
1111 nlh = nlmsg_hdr(skb);
1112 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1113 skb->len < nlh->nlmsg_len ||
1114 nlmsg_len(nlh) < sizeof(*frn))
1117 skb = netlink_skb_clone(skb, GFP_KERNEL);
1120 nlh = nlmsg_hdr(skb);
1122 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1123 nl_fib_lookup(net, frn);
1125 portid = NETLINK_CB(skb).portid; /* netlink portid */
1126 NETLINK_CB(skb).portid = 0; /* from kernel */
1127 NETLINK_CB(skb).dst_group = 0; /* unicast */
1128 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1131 static int __net_init nl_fib_lookup_init(struct net *net)
1134 struct netlink_kernel_cfg cfg = {
1135 .input = nl_fib_input,
1138 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1140 return -EAFNOSUPPORT;
1141 net->ipv4.fibnl = sk;
1145 static void nl_fib_lookup_exit(struct net *net)
1147 netlink_kernel_release(net->ipv4.fibnl);
1148 net->ipv4.fibnl = NULL;
1151 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1154 if (fib_sync_down_dev(dev, event, force))
1155 fib_flush(dev_net(dev));
1157 rt_cache_flush(dev_net(dev));
1161 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1163 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1164 struct net_device *dev = ifa->ifa_dev->dev;
1165 struct net *net = dev_net(dev);
1169 fib_add_ifaddr(ifa);
1170 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1171 fib_sync_up(dev, RTNH_F_DEAD);
1173 atomic_inc(&net->ipv4.dev_addr_genid);
1174 rt_cache_flush(dev_net(dev));
1177 fib_del_ifaddr(ifa, NULL);
1178 atomic_inc(&net->ipv4.dev_addr_genid);
1179 if (!ifa->ifa_dev->ifa_list) {
1180 /* Last address was deleted from this interface.
1183 fib_disable_ip(dev, event, true);
1185 rt_cache_flush(dev_net(dev));
1192 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1194 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1195 struct netdev_notifier_changeupper_info *upper_info = ptr;
1196 struct netdev_notifier_info_ext *info_ext = ptr;
1197 struct in_device *in_dev;
1198 struct net *net = dev_net(dev);
1201 if (event == NETDEV_UNREGISTER) {
1202 fib_disable_ip(dev, event, true);
1207 in_dev = __in_dev_get_rtnl(dev);
1214 fib_add_ifaddr(ifa);
1215 } endfor_ifa(in_dev);
1216 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1217 fib_sync_up(dev, RTNH_F_DEAD);
1219 atomic_inc(&net->ipv4.dev_addr_genid);
1220 rt_cache_flush(net);
1223 fib_disable_ip(dev, event, false);
1226 flags = dev_get_flags(dev);
1227 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1228 fib_sync_up(dev, RTNH_F_LINKDOWN);
1230 fib_sync_down_dev(dev, event, false);
1231 rt_cache_flush(net);
1233 case NETDEV_CHANGEMTU:
1234 fib_sync_mtu(dev, info_ext->ext.mtu);
1235 rt_cache_flush(net);
1237 case NETDEV_CHANGEUPPER:
1239 /* flush all routes if dev is linked to or unlinked from
1240 * an L3 master device (e.g., VRF)
1242 if (upper_info->upper_dev &&
1243 netif_is_l3_master(upper_info->upper_dev))
1244 fib_disable_ip(dev, NETDEV_DOWN, true);
1250 static struct notifier_block fib_inetaddr_notifier = {
1251 .notifier_call = fib_inetaddr_event,
1254 static struct notifier_block fib_netdev_notifier = {
1255 .notifier_call = fib_netdev_event,
1258 static int __net_init ip_fib_net_init(struct net *net)
1261 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1263 err = fib4_notifier_init(net);
1267 /* Avoid false sharing : Use at least a full cache line */
1268 size = max_t(size_t, size, L1_CACHE_BYTES);
1270 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1271 if (!net->ipv4.fib_table_hash) {
1273 goto err_table_hash_alloc;
1276 err = fib4_rules_init(net);
1278 goto err_rules_init;
1282 kfree(net->ipv4.fib_table_hash);
1283 err_table_hash_alloc:
1284 fib4_notifier_exit(net);
1288 static void ip_fib_net_exit(struct net *net)
1293 #ifdef CONFIG_IP_MULTIPLE_TABLES
1294 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1295 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1297 /* Destroy the tables in reverse order to guarantee that the
1298 * local table, ID 255, is destroyed before the main table, ID
1299 * 254. This is necessary as the local table may contain
1300 * references to data contained in the main table.
1302 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1303 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1304 struct hlist_node *tmp;
1305 struct fib_table *tb;
1307 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1308 hlist_del(&tb->tb_hlist);
1309 fib_table_flush(net, tb, true);
1314 #ifdef CONFIG_IP_MULTIPLE_TABLES
1315 fib4_rules_exit(net);
1318 kfree(net->ipv4.fib_table_hash);
1319 fib4_notifier_exit(net);
1322 static int __net_init fib_net_init(struct net *net)
1326 #ifdef CONFIG_IP_ROUTE_CLASSID
1327 net->ipv4.fib_num_tclassid_users = 0;
1329 error = ip_fib_net_init(net);
1332 error = nl_fib_lookup_init(net);
1335 error = fib_proc_init(net);
1342 nl_fib_lookup_exit(net);
1344 ip_fib_net_exit(net);
1348 static void __net_exit fib_net_exit(struct net *net)
1351 nl_fib_lookup_exit(net);
1352 ip_fib_net_exit(net);
1355 static struct pernet_operations fib_net_ops = {
1356 .init = fib_net_init,
1357 .exit = fib_net_exit,
1360 void __init ip_fib_init(void)
1364 register_pernet_subsys(&fib_net_ops);
1366 register_netdevice_notifier(&fib_netdev_notifier);
1367 register_inetaddr_notifier(&fib_inetaddr_notifier);
1369 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1370 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1371 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);