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);
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);
289 scope = RT_SCOPE_UNIVERSE;
290 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
291 struct flowi4 fl4 = {
292 .flowi4_iif = LOOPBACK_IFINDEX,
293 .daddr = ip_hdr(skb)->saddr,
294 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
295 .flowi4_scope = scope,
296 .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0,
298 if (!fib_lookup(net, &fl4, &res, 0))
299 return FIB_RES_PREFSRC(net, res);
301 scope = RT_SCOPE_LINK;
304 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
307 /* Given (packet source, input interface) and optional (dst, oif, tos):
308 * - (main) check, that source is valid i.e. not broadcast or our local
310 * - figure out what "logical" interface this packet arrived
311 * and calculate "specific destination" address.
312 * - check, that packet arrived from expected physical interface.
313 * called with rcu_read_lock()
315 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
316 u8 tos, int oif, struct net_device *dev,
317 int rpf, struct in_device *idev, u32 *itag)
320 struct fib_result res;
322 struct net *net = dev_net(dev);
326 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
328 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
331 fl4.flowi4_tos = tos;
332 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
333 fl4.flowi4_tun_key.tun_id = 0;
334 fl4.flowi4_flags = 0;
335 fl4.flowi4_uid = sock_net_uid(net, NULL);
337 no_addr = idev->ifa_list == NULL;
339 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
341 trace_fib_validate_source(dev, &fl4);
343 if (fib_lookup(net, &fl4, &res, 0))
345 if (res.type != RTN_UNICAST &&
346 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
348 if (!rpf && !fib_num_tclassid_users(net) &&
349 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
351 fib_combine_itag(itag, &res);
354 #ifdef CONFIG_IP_ROUTE_MULTIPATH
355 for (ret = 0; ret < res.fi->fib_nhs; ret++) {
356 struct fib_nh *nh = &res.fi->fib_nh[ret];
358 if (nh->nh_dev == dev) {
361 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
367 if (FIB_RES_DEV(res) == dev)
371 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
378 fl4.flowi4_oif = dev->ifindex;
381 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
382 if (res.type == RTN_UNICAST)
383 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
399 /* Ignore rp_filter for packets protected by IPsec. */
400 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
401 u8 tos, int oif, struct net_device *dev,
402 struct in_device *idev, u32 *itag)
404 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
406 if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
407 IN_DEV_ACCEPT_LOCAL(idev) &&
408 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
412 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
415 static inline __be32 sk_extract_addr(struct sockaddr *addr)
417 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
420 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
424 nla = (struct nlattr *) ((char *) mx + len);
425 nla->nla_type = type;
426 nla->nla_len = nla_attr_size(4);
427 *(u32 *) nla_data(nla) = value;
429 return len + nla_total_size(4);
432 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
433 struct fib_config *cfg)
438 memset(cfg, 0, sizeof(*cfg));
439 cfg->fc_nlinfo.nl_net = net;
441 if (rt->rt_dst.sa_family != AF_INET)
442 return -EAFNOSUPPORT;
445 * Check mask for validity:
446 * a) it must be contiguous.
447 * b) destination must have all host bits clear.
448 * c) if application forgot to set correct family (AF_INET),
449 * reject request unless it is absolutely clear i.e.
450 * both family and mask are zero.
453 addr = sk_extract_addr(&rt->rt_dst);
454 if (!(rt->rt_flags & RTF_HOST)) {
455 __be32 mask = sk_extract_addr(&rt->rt_genmask);
457 if (rt->rt_genmask.sa_family != AF_INET) {
458 if (mask || rt->rt_genmask.sa_family)
459 return -EAFNOSUPPORT;
462 if (bad_mask(mask, addr))
465 plen = inet_mask_len(mask);
468 cfg->fc_dst_len = plen;
471 if (cmd != SIOCDELRT) {
472 cfg->fc_nlflags = NLM_F_CREATE;
473 cfg->fc_protocol = RTPROT_BOOT;
477 cfg->fc_priority = rt->rt_metric - 1;
479 if (rt->rt_flags & RTF_REJECT) {
480 cfg->fc_scope = RT_SCOPE_HOST;
481 cfg->fc_type = RTN_UNREACHABLE;
485 cfg->fc_scope = RT_SCOPE_NOWHERE;
486 cfg->fc_type = RTN_UNICAST;
490 struct net_device *dev;
491 char devname[IFNAMSIZ];
493 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
496 devname[IFNAMSIZ-1] = 0;
497 colon = strchr(devname, ':');
500 dev = __dev_get_by_name(net, devname);
503 cfg->fc_oif = dev->ifindex;
504 cfg->fc_table = l3mdev_fib_table(dev);
506 struct in_ifaddr *ifa;
507 struct in_device *in_dev = __in_dev_get_rtnl(dev);
511 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
512 if (strcmp(ifa->ifa_label, devname) == 0)
516 cfg->fc_prefsrc = ifa->ifa_local;
520 addr = sk_extract_addr(&rt->rt_gateway);
521 if (rt->rt_gateway.sa_family == AF_INET && addr) {
522 unsigned int addr_type;
525 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
526 if (rt->rt_flags & RTF_GATEWAY &&
527 addr_type == RTN_UNICAST)
528 cfg->fc_scope = RT_SCOPE_UNIVERSE;
531 if (cmd == SIOCDELRT)
534 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
537 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
538 cfg->fc_scope = RT_SCOPE_LINK;
540 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
544 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
548 if (rt->rt_flags & RTF_MTU)
549 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
551 if (rt->rt_flags & RTF_WINDOW)
552 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
554 if (rt->rt_flags & RTF_IRTT)
555 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
558 cfg->fc_mx_len = len;
565 * Handle IP routing ioctl calls.
566 * These are used to manipulate the routing tables
568 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
570 struct fib_config cfg;
575 case SIOCADDRT: /* Add a route */
576 case SIOCDELRT: /* Delete a route */
577 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
580 if (copy_from_user(&rt, arg, sizeof(rt)))
584 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
586 struct fib_table *tb;
588 if (cmd == SIOCDELRT) {
589 tb = fib_get_table(net, cfg.fc_table);
591 err = fib_table_delete(net, tb, &cfg,
596 tb = fib_new_table(net, cfg.fc_table);
598 err = fib_table_insert(net, tb,
604 /* allocated by rtentry_to_fib_config() */
613 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
614 [RTA_DST] = { .type = NLA_U32 },
615 [RTA_SRC] = { .type = NLA_U32 },
616 [RTA_IIF] = { .type = NLA_U32 },
617 [RTA_OIF] = { .type = NLA_U32 },
618 [RTA_GATEWAY] = { .type = NLA_U32 },
619 [RTA_PRIORITY] = { .type = NLA_U32 },
620 [RTA_PREFSRC] = { .type = NLA_U32 },
621 [RTA_METRICS] = { .type = NLA_NESTED },
622 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
623 [RTA_FLOW] = { .type = NLA_U32 },
624 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
625 [RTA_ENCAP] = { .type = NLA_NESTED },
626 [RTA_UID] = { .type = NLA_U32 },
627 [RTA_MARK] = { .type = NLA_U32 },
630 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
631 struct nlmsghdr *nlh, struct fib_config *cfg,
632 struct netlink_ext_ack *extack)
638 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy,
643 memset(cfg, 0, sizeof(*cfg));
645 rtm = nlmsg_data(nlh);
646 cfg->fc_dst_len = rtm->rtm_dst_len;
647 cfg->fc_tos = rtm->rtm_tos;
648 cfg->fc_table = rtm->rtm_table;
649 cfg->fc_protocol = rtm->rtm_protocol;
650 cfg->fc_scope = rtm->rtm_scope;
651 cfg->fc_type = rtm->rtm_type;
652 cfg->fc_flags = rtm->rtm_flags;
653 cfg->fc_nlflags = nlh->nlmsg_flags;
655 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
656 cfg->fc_nlinfo.nlh = nlh;
657 cfg->fc_nlinfo.nl_net = net;
659 if (cfg->fc_type > RTN_MAX) {
660 NL_SET_ERR_MSG(extack, "Invalid route type");
665 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
666 switch (nla_type(attr)) {
668 cfg->fc_dst = nla_get_be32(attr);
671 cfg->fc_oif = nla_get_u32(attr);
674 cfg->fc_gw = nla_get_be32(attr);
677 cfg->fc_priority = nla_get_u32(attr);
680 cfg->fc_prefsrc = nla_get_be32(attr);
683 cfg->fc_mx = nla_data(attr);
684 cfg->fc_mx_len = nla_len(attr);
687 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
692 cfg->fc_mp = nla_data(attr);
693 cfg->fc_mp_len = nla_len(attr);
696 cfg->fc_flow = nla_get_u32(attr);
699 cfg->fc_table = nla_get_u32(attr);
702 cfg->fc_encap = attr;
705 cfg->fc_encap_type = nla_get_u16(attr);
706 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
719 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
720 struct netlink_ext_ack *extack)
722 struct net *net = sock_net(skb->sk);
723 struct fib_config cfg;
724 struct fib_table *tb;
727 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
731 tb = fib_get_table(net, cfg.fc_table);
733 NL_SET_ERR_MSG(extack, "FIB table does not exist");
738 err = fib_table_delete(net, tb, &cfg, extack);
743 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
744 struct netlink_ext_ack *extack)
746 struct net *net = sock_net(skb->sk);
747 struct fib_config cfg;
748 struct fib_table *tb;
751 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
755 tb = fib_new_table(net, cfg.fc_table);
761 err = fib_table_insert(net, tb, &cfg, extack);
766 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
768 struct net *net = sock_net(skb->sk);
770 unsigned int e = 0, s_e;
771 struct fib_table *tb;
772 struct hlist_head *head;
775 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
776 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
784 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
786 head = &net->ipv4.fib_table_hash[h];
787 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
791 memset(&cb->args[2], 0, sizeof(cb->args) -
792 2 * sizeof(cb->args[0]));
793 err = fib_table_dump(tb, skb, cb);
795 if (likely(skb->len))
816 /* Prepare and feed intra-kernel routing request.
817 * Really, it should be netlink message, but :-( netlink
818 * can be not configured, so that we feed it directly
819 * to fib engine. It is legal, because all events occur
820 * only when netlink is already locked.
822 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
824 struct net *net = dev_net(ifa->ifa_dev->dev);
825 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
826 struct fib_table *tb;
827 struct fib_config cfg = {
828 .fc_protocol = RTPROT_KERNEL,
831 .fc_dst_len = dst_len,
832 .fc_prefsrc = ifa->ifa_local,
833 .fc_oif = ifa->ifa_dev->dev->ifindex,
834 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
841 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
843 tb = fib_new_table(net, tb_id);
847 cfg.fc_table = tb->tb_id;
849 if (type != RTN_LOCAL)
850 cfg.fc_scope = RT_SCOPE_LINK;
852 cfg.fc_scope = RT_SCOPE_HOST;
854 if (cmd == RTM_NEWROUTE)
855 fib_table_insert(net, tb, &cfg, NULL);
857 fib_table_delete(net, tb, &cfg, NULL);
860 void fib_add_ifaddr(struct in_ifaddr *ifa)
862 struct in_device *in_dev = ifa->ifa_dev;
863 struct net_device *dev = in_dev->dev;
864 struct in_ifaddr *prim = ifa;
865 __be32 mask = ifa->ifa_mask;
866 __be32 addr = ifa->ifa_local;
867 __be32 prefix = ifa->ifa_address & mask;
869 if (ifa->ifa_flags & IFA_F_SECONDARY) {
870 prim = inet_ifa_byprefix(in_dev, prefix, mask);
872 pr_warn("%s: bug: prim == NULL\n", __func__);
877 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
879 if (!(dev->flags & IFF_UP))
882 /* Add broadcast address, if it is explicitly assigned. */
883 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
884 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
886 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
887 (prefix != addr || ifa->ifa_prefixlen < 32)) {
888 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
889 fib_magic(RTM_NEWROUTE,
890 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
891 prefix, ifa->ifa_prefixlen, prim);
893 /* Add network specific broadcasts, when it takes a sense */
894 if (ifa->ifa_prefixlen < 31) {
895 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
896 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
902 /* Delete primary or secondary address.
903 * Optionally, on secondary address promotion consider the addresses
904 * from subnet iprim as deleted, even if they are in device list.
905 * In this case the secondary ifa can be in device list.
907 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
909 struct in_device *in_dev = ifa->ifa_dev;
910 struct net_device *dev = in_dev->dev;
911 struct in_ifaddr *ifa1;
912 struct in_ifaddr *prim = ifa, *prim1 = NULL;
913 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
914 __be32 any = ifa->ifa_address & ifa->ifa_mask;
920 int subnet = 0; /* Primary network */
921 int gone = 1; /* Address is missing */
922 int same_prefsrc = 0; /* Another primary with same IP */
924 if (ifa->ifa_flags & IFA_F_SECONDARY) {
925 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
927 /* if the device has been deleted, we don't perform
931 pr_warn("%s: bug: prim == NULL\n", __func__);
934 if (iprim && iprim != prim) {
935 pr_warn("%s: bug: iprim != prim\n", __func__);
938 } else if (!ipv4_is_zeronet(any) &&
939 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
940 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
941 fib_magic(RTM_DELROUTE,
942 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
943 any, ifa->ifa_prefixlen, prim);
950 /* Deletion is more complicated than add.
951 * We should take care of not to delete too much :-)
953 * Scan address list to be sure that addresses are really gone.
956 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
958 /* promotion, keep the IP */
962 /* Ignore IFAs from our subnet */
963 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
964 inet_ifa_match(ifa1->ifa_address, iprim))
967 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
968 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
969 /* Another address from our subnet? */
970 if (ifa1->ifa_mask == prim->ifa_mask &&
971 inet_ifa_match(ifa1->ifa_address, prim))
974 /* We reached the secondaries, so
975 * same_prefsrc should be determined.
979 /* Search new prim1 if ifa1 is not
980 * using the current prim1
983 ifa1->ifa_mask != prim1->ifa_mask ||
984 !inet_ifa_match(ifa1->ifa_address, prim1))
985 prim1 = inet_ifa_byprefix(in_dev,
990 if (prim1->ifa_local != prim->ifa_local)
994 if (prim->ifa_local != ifa1->ifa_local)
1000 if (ifa->ifa_local == ifa1->ifa_local)
1002 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1004 if (brd == ifa1->ifa_broadcast)
1006 if (any == ifa1->ifa_broadcast)
1008 /* primary has network specific broadcasts */
1009 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1010 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1011 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1013 if (!ipv4_is_zeronet(any1)) {
1014 if (ifa->ifa_broadcast == brd1 ||
1015 ifa->ifa_broadcast == any1)
1017 if (brd == brd1 || brd == any1)
1019 if (any == brd1 || any == any1)
1027 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1028 if (subnet && ifa->ifa_prefixlen < 31) {
1029 if (!(ok & BRD1_OK))
1030 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1031 if (!(ok & BRD0_OK))
1032 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1034 if (!(ok & LOCAL_OK)) {
1035 unsigned int addr_type;
1037 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1039 /* Check, that this local address finally disappeared. */
1040 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1042 if (gone && addr_type != RTN_LOCAL) {
1043 /* And the last, but not the least thing.
1044 * We must flush stray FIB entries.
1046 * First of all, we scan fib_info list searching
1047 * for stray nexthop entries, then ignite fib_flush.
1049 if (fib_sync_down_addr(dev, ifa->ifa_local))
1050 fib_flush(dev_net(dev));
1059 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1062 struct fib_result res;
1063 struct flowi4 fl4 = {
1064 .flowi4_mark = frn->fl_mark,
1065 .daddr = frn->fl_addr,
1066 .flowi4_tos = frn->fl_tos,
1067 .flowi4_scope = frn->fl_scope,
1069 struct fib_table *tb;
1073 tb = fib_get_table(net, frn->tb_id_in);
1079 frn->tb_id = tb->tb_id;
1080 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1083 frn->prefixlen = res.prefixlen;
1084 frn->nh_sel = res.nh_sel;
1085 frn->type = res.type;
1086 frn->scope = res.scope;
1094 static void nl_fib_input(struct sk_buff *skb)
1097 struct fib_result_nl *frn;
1098 struct nlmsghdr *nlh;
1101 net = sock_net(skb->sk);
1102 nlh = nlmsg_hdr(skb);
1103 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1104 skb->len < nlh->nlmsg_len ||
1105 nlmsg_len(nlh) < sizeof(*frn))
1108 skb = netlink_skb_clone(skb, GFP_KERNEL);
1111 nlh = nlmsg_hdr(skb);
1113 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1114 nl_fib_lookup(net, frn);
1116 portid = NETLINK_CB(skb).portid; /* netlink portid */
1117 NETLINK_CB(skb).portid = 0; /* from kernel */
1118 NETLINK_CB(skb).dst_group = 0; /* unicast */
1119 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1122 static int __net_init nl_fib_lookup_init(struct net *net)
1125 struct netlink_kernel_cfg cfg = {
1126 .input = nl_fib_input,
1129 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1131 return -EAFNOSUPPORT;
1132 net->ipv4.fibnl = sk;
1136 static void nl_fib_lookup_exit(struct net *net)
1138 netlink_kernel_release(net->ipv4.fibnl);
1139 net->ipv4.fibnl = NULL;
1142 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1145 if (fib_sync_down_dev(dev, event, force))
1146 fib_flush(dev_net(dev));
1148 rt_cache_flush(dev_net(dev));
1152 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1154 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1155 struct net_device *dev = ifa->ifa_dev->dev;
1156 struct net *net = dev_net(dev);
1160 fib_add_ifaddr(ifa);
1161 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1162 fib_sync_up(dev, RTNH_F_DEAD);
1164 atomic_inc(&net->ipv4.dev_addr_genid);
1165 rt_cache_flush(dev_net(dev));
1168 fib_del_ifaddr(ifa, NULL);
1169 atomic_inc(&net->ipv4.dev_addr_genid);
1170 if (!ifa->ifa_dev->ifa_list) {
1171 /* Last address was deleted from this interface.
1174 fib_disable_ip(dev, event, true);
1176 rt_cache_flush(dev_net(dev));
1183 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1185 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1186 struct netdev_notifier_changeupper_info *info;
1187 struct in_device *in_dev;
1188 struct net *net = dev_net(dev);
1191 if (event == NETDEV_UNREGISTER) {
1192 fib_disable_ip(dev, event, true);
1197 in_dev = __in_dev_get_rtnl(dev);
1204 fib_add_ifaddr(ifa);
1205 } endfor_ifa(in_dev);
1206 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1207 fib_sync_up(dev, RTNH_F_DEAD);
1209 atomic_inc(&net->ipv4.dev_addr_genid);
1210 rt_cache_flush(net);
1213 fib_disable_ip(dev, event, false);
1216 flags = dev_get_flags(dev);
1217 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1218 fib_sync_up(dev, RTNH_F_LINKDOWN);
1220 fib_sync_down_dev(dev, event, false);
1222 case NETDEV_CHANGEMTU:
1223 rt_cache_flush(net);
1225 case NETDEV_CHANGEUPPER:
1227 /* flush all routes if dev is linked to or unlinked from
1228 * an L3 master device (e.g., VRF)
1230 if (info->upper_dev && netif_is_l3_master(info->upper_dev))
1231 fib_disable_ip(dev, NETDEV_DOWN, true);
1237 static struct notifier_block fib_inetaddr_notifier = {
1238 .notifier_call = fib_inetaddr_event,
1241 static struct notifier_block fib_netdev_notifier = {
1242 .notifier_call = fib_netdev_event,
1245 static int __net_init ip_fib_net_init(struct net *net)
1248 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1250 net->ipv4.fib_seq = 0;
1252 /* Avoid false sharing : Use at least a full cache line */
1253 size = max_t(size_t, size, L1_CACHE_BYTES);
1255 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1256 if (!net->ipv4.fib_table_hash)
1259 err = fib4_rules_init(net);
1265 kfree(net->ipv4.fib_table_hash);
1269 static void ip_fib_net_exit(struct net *net)
1274 #ifdef CONFIG_IP_MULTIPLE_TABLES
1275 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1276 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1278 for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1279 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1280 struct hlist_node *tmp;
1281 struct fib_table *tb;
1283 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1284 hlist_del(&tb->tb_hlist);
1285 fib_table_flush(net, tb);
1290 #ifdef CONFIG_IP_MULTIPLE_TABLES
1291 fib4_rules_exit(net);
1294 kfree(net->ipv4.fib_table_hash);
1297 static int __net_init fib_net_init(struct net *net)
1301 #ifdef CONFIG_IP_ROUTE_CLASSID
1302 net->ipv4.fib_num_tclassid_users = 0;
1304 error = ip_fib_net_init(net);
1307 error = nl_fib_lookup_init(net);
1310 error = fib_proc_init(net);
1317 nl_fib_lookup_exit(net);
1319 ip_fib_net_exit(net);
1323 static void __net_exit fib_net_exit(struct net *net)
1326 nl_fib_lookup_exit(net);
1327 ip_fib_net_exit(net);
1330 static struct pernet_operations fib_net_ops = {
1331 .init = fib_net_init,
1332 .exit = fib_net_exit,
1335 void __init ip_fib_init(void)
1337 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1338 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1339 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1341 register_pernet_subsys(&fib_net_ops);
1342 register_netdevice_notifier(&fib_netdev_notifier);
1343 register_inetaddr_notifier(&fib_inetaddr_notifier);