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 <asm/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 <trace/events/fib.h>
51 #ifndef CONFIG_IP_MULTIPLE_TABLES
53 static int __net_init fib4_rules_init(struct net *net)
55 struct fib_table *local_table, *main_table;
57 main_table = fib_trie_table(RT_TABLE_MAIN, NULL);
61 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
65 hlist_add_head_rcu(&local_table->tb_hlist,
66 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
67 hlist_add_head_rcu(&main_table->tb_hlist,
68 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
72 fib_free_table(main_table);
77 struct fib_table *fib_new_table(struct net *net, u32 id)
79 struct fib_table *tb, *alias = NULL;
84 tb = fib_get_table(net, id);
88 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
89 alias = fib_new_table(net, RT_TABLE_MAIN);
91 tb = fib_trie_table(id, alias);
97 rcu_assign_pointer(net->ipv4.fib_local, tb);
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]);
114 /* caller must hold either rtnl or rcu read lock */
115 struct fib_table *fib_get_table(struct net *net, u32 id)
117 struct fib_table *tb;
118 struct hlist_head *head;
123 h = id & (FIB_TABLE_HASHSZ - 1);
125 head = &net->ipv4.fib_table_hash[h];
126 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
132 #endif /* CONFIG_IP_MULTIPLE_TABLES */
134 static void fib_replace_table(struct net *net, struct fib_table *old,
135 struct fib_table *new)
137 #ifdef CONFIG_IP_MULTIPLE_TABLES
138 switch (new->tb_id) {
140 rcu_assign_pointer(net->ipv4.fib_local, new);
143 rcu_assign_pointer(net->ipv4.fib_main, new);
145 case RT_TABLE_DEFAULT:
146 rcu_assign_pointer(net->ipv4.fib_default, new);
153 /* replace the old table in the hlist */
154 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
157 int fib_unmerge(struct net *net)
159 struct fib_table *old, *new;
161 /* attempt to fetch local table if it has been allocated */
162 old = fib_get_table(net, RT_TABLE_LOCAL);
166 new = fib_trie_unmerge(old);
170 /* replace merged table with clean table */
172 fib_replace_table(net, old, new);
179 static void fib_flush(struct net *net)
184 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
185 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
186 struct hlist_node *tmp;
187 struct fib_table *tb;
189 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
190 flushed += fib_table_flush(tb);
197 void fib_flush_external(struct net *net)
199 struct fib_table *tb;
200 struct hlist_head *head;
203 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
204 head = &net->ipv4.fib_table_hash[h];
205 hlist_for_each_entry(tb, head, tb_hlist)
206 fib_table_flush_external(tb);
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 if (!dev || dev == res.fi->fib_dev)
243 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
245 return __inet_dev_addr_type(net, NULL, addr, tb_id);
247 EXPORT_SYMBOL(inet_addr_type_table);
249 unsigned int inet_addr_type(struct net *net, __be32 addr)
251 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
253 EXPORT_SYMBOL(inet_addr_type);
255 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
258 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
260 return __inet_dev_addr_type(net, dev, addr, rt_table);
262 EXPORT_SYMBOL(inet_dev_addr_type);
264 /* inet_addr_type with dev == NULL but using the table from a dev
265 * if one is associated
267 unsigned int inet_addr_type_dev_table(struct net *net,
268 const struct net_device *dev,
271 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
273 return __inet_dev_addr_type(net, NULL, addr, rt_table);
275 EXPORT_SYMBOL(inet_addr_type_dev_table);
277 __be32 fib_compute_spec_dst(struct sk_buff *skb)
279 struct net_device *dev = skb->dev;
280 struct in_device *in_dev;
281 struct fib_result res;
286 rt = skb_rtable(skb);
287 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
289 return ip_hdr(skb)->daddr;
291 in_dev = __in_dev_get_rcu(dev);
295 scope = RT_SCOPE_UNIVERSE;
296 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
297 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
298 struct flowi4 fl4 = {
299 .flowi4_iif = LOOPBACK_IFINDEX,
300 .flowi4_oif = l3mdev_master_ifindex_rcu(dev),
301 .daddr = ip_hdr(skb)->saddr,
302 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
303 .flowi4_scope = scope,
304 .flowi4_mark = vmark ? skb->mark : 0,
306 if (!fib_lookup(net, &fl4, &res, 0))
307 return FIB_RES_PREFSRC(net, res);
309 scope = RT_SCOPE_LINK;
312 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
315 /* Given (packet source, input interface) and optional (dst, oif, tos):
316 * - (main) check, that source is valid i.e. not broadcast or our local
318 * - figure out what "logical" interface this packet arrived
319 * and calculate "specific destination" address.
320 * - check, that packet arrived from expected physical interface.
321 * called with rcu_read_lock()
323 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
324 u8 tos, int oif, struct net_device *dev,
325 int rpf, struct in_device *idev, u32 *itag)
328 struct fib_result res;
334 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
336 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
339 fl4.flowi4_tos = tos;
340 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
341 fl4.flowi4_tun_key.tun_id = 0;
342 fl4.flowi4_flags = 0;
344 no_addr = idev->ifa_list == NULL;
346 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
348 trace_fib_validate_source(dev, &fl4);
351 if (fib_lookup(net, &fl4, &res, 0))
353 if (res.type != RTN_UNICAST &&
354 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
356 if (!rpf && !fib_num_tclassid_users(dev_net(dev)) &&
357 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
359 fib_combine_itag(itag, &res);
362 #ifdef CONFIG_IP_ROUTE_MULTIPATH
363 for (ret = 0; ret < res.fi->fib_nhs; ret++) {
364 struct fib_nh *nh = &res.fi->fib_nh[ret];
366 if (nh->nh_dev == dev) {
369 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
375 if (FIB_RES_DEV(res) == dev)
379 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
386 fl4.flowi4_oif = dev->ifindex;
389 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
390 if (res.type == RTN_UNICAST)
391 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
407 /* Ignore rp_filter for packets protected by IPsec. */
408 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
409 u8 tos, int oif, struct net_device *dev,
410 struct in_device *idev, u32 *itag)
412 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
414 if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
415 IN_DEV_ACCEPT_LOCAL(idev) &&
416 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
420 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
423 static inline __be32 sk_extract_addr(struct sockaddr *addr)
425 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
428 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
432 nla = (struct nlattr *) ((char *) mx + len);
433 nla->nla_type = type;
434 nla->nla_len = nla_attr_size(4);
435 *(u32 *) nla_data(nla) = value;
437 return len + nla_total_size(4);
440 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
441 struct fib_config *cfg)
446 memset(cfg, 0, sizeof(*cfg));
447 cfg->fc_nlinfo.nl_net = net;
449 if (rt->rt_dst.sa_family != AF_INET)
450 return -EAFNOSUPPORT;
453 * Check mask for validity:
454 * a) it must be contiguous.
455 * b) destination must have all host bits clear.
456 * c) if application forgot to set correct family (AF_INET),
457 * reject request unless it is absolutely clear i.e.
458 * both family and mask are zero.
461 addr = sk_extract_addr(&rt->rt_dst);
462 if (!(rt->rt_flags & RTF_HOST)) {
463 __be32 mask = sk_extract_addr(&rt->rt_genmask);
465 if (rt->rt_genmask.sa_family != AF_INET) {
466 if (mask || rt->rt_genmask.sa_family)
467 return -EAFNOSUPPORT;
470 if (bad_mask(mask, addr))
473 plen = inet_mask_len(mask);
476 cfg->fc_dst_len = plen;
479 if (cmd != SIOCDELRT) {
480 cfg->fc_nlflags = NLM_F_CREATE;
481 cfg->fc_protocol = RTPROT_BOOT;
485 cfg->fc_priority = rt->rt_metric - 1;
487 if (rt->rt_flags & RTF_REJECT) {
488 cfg->fc_scope = RT_SCOPE_HOST;
489 cfg->fc_type = RTN_UNREACHABLE;
493 cfg->fc_scope = RT_SCOPE_NOWHERE;
494 cfg->fc_type = RTN_UNICAST;
498 struct net_device *dev;
499 char devname[IFNAMSIZ];
501 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
504 devname[IFNAMSIZ-1] = 0;
505 colon = strchr(devname, ':');
508 dev = __dev_get_by_name(net, devname);
511 cfg->fc_oif = dev->ifindex;
513 struct in_ifaddr *ifa;
514 struct in_device *in_dev = __in_dev_get_rtnl(dev);
518 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
519 if (strcmp(ifa->ifa_label, devname) == 0)
523 cfg->fc_prefsrc = ifa->ifa_local;
527 addr = sk_extract_addr(&rt->rt_gateway);
528 if (rt->rt_gateway.sa_family == AF_INET && addr) {
529 unsigned int addr_type;
532 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
533 if (rt->rt_flags & RTF_GATEWAY &&
534 addr_type == RTN_UNICAST)
535 cfg->fc_scope = RT_SCOPE_UNIVERSE;
538 if (cmd == SIOCDELRT)
541 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
544 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
545 cfg->fc_scope = RT_SCOPE_LINK;
547 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
551 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
555 if (rt->rt_flags & RTF_MTU)
556 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
558 if (rt->rt_flags & RTF_WINDOW)
559 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
561 if (rt->rt_flags & RTF_IRTT)
562 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
565 cfg->fc_mx_len = len;
572 * Handle IP routing ioctl calls.
573 * These are used to manipulate the routing tables
575 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
577 struct fib_config cfg;
582 case SIOCADDRT: /* Add a route */
583 case SIOCDELRT: /* Delete a route */
584 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
587 if (copy_from_user(&rt, arg, sizeof(rt)))
591 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
593 struct fib_table *tb;
595 if (cmd == SIOCDELRT) {
596 tb = fib_get_table(net, cfg.fc_table);
598 err = fib_table_delete(tb, &cfg);
602 tb = fib_new_table(net, cfg.fc_table);
604 err = fib_table_insert(tb, &cfg);
609 /* allocated by rtentry_to_fib_config() */
618 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
619 [RTA_DST] = { .type = NLA_U32 },
620 [RTA_SRC] = { .type = NLA_U32 },
621 [RTA_IIF] = { .type = NLA_U32 },
622 [RTA_OIF] = { .type = NLA_U32 },
623 [RTA_GATEWAY] = { .type = NLA_U32 },
624 [RTA_PRIORITY] = { .type = NLA_U32 },
625 [RTA_PREFSRC] = { .type = NLA_U32 },
626 [RTA_METRICS] = { .type = NLA_NESTED },
627 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
628 [RTA_FLOW] = { .type = NLA_U32 },
629 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
630 [RTA_ENCAP] = { .type = NLA_NESTED },
633 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
634 struct nlmsghdr *nlh, struct fib_config *cfg)
640 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
644 memset(cfg, 0, sizeof(*cfg));
646 rtm = nlmsg_data(nlh);
647 cfg->fc_dst_len = rtm->rtm_dst_len;
648 cfg->fc_tos = rtm->rtm_tos;
649 cfg->fc_table = rtm->rtm_table;
650 cfg->fc_protocol = rtm->rtm_protocol;
651 cfg->fc_scope = rtm->rtm_scope;
652 cfg->fc_type = rtm->rtm_type;
653 cfg->fc_flags = rtm->rtm_flags;
654 cfg->fc_nlflags = nlh->nlmsg_flags;
656 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
657 cfg->fc_nlinfo.nlh = nlh;
658 cfg->fc_nlinfo.nl_net = net;
660 if (cfg->fc_type > RTN_MAX) {
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 cfg->fc_mp = nla_data(attr);
688 cfg->fc_mp_len = nla_len(attr);
691 cfg->fc_flow = nla_get_u32(attr);
694 cfg->fc_table = nla_get_u32(attr);
697 cfg->fc_encap = attr;
700 cfg->fc_encap_type = nla_get_u16(attr);
710 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
712 struct net *net = sock_net(skb->sk);
713 struct fib_config cfg;
714 struct fib_table *tb;
717 err = rtm_to_fib_config(net, skb, nlh, &cfg);
721 tb = fib_get_table(net, cfg.fc_table);
727 err = fib_table_delete(tb, &cfg);
732 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
734 struct net *net = sock_net(skb->sk);
735 struct fib_config cfg;
736 struct fib_table *tb;
739 err = rtm_to_fib_config(net, skb, nlh, &cfg);
743 tb = fib_new_table(net, cfg.fc_table);
749 err = fib_table_insert(tb, &cfg);
754 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
756 struct net *net = sock_net(skb->sk);
758 unsigned int e = 0, s_e;
759 struct fib_table *tb;
760 struct hlist_head *head;
763 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
764 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
772 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
774 head = &net->ipv4.fib_table_hash[h];
775 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
779 memset(&cb->args[2], 0, sizeof(cb->args) -
780 2 * sizeof(cb->args[0]));
781 err = fib_table_dump(tb, skb, cb);
783 if (likely(skb->len))
804 /* Prepare and feed intra-kernel routing request.
805 * Really, it should be netlink message, but :-( netlink
806 * can be not configured, so that we feed it directly
807 * to fib engine. It is legal, because all events occur
808 * only when netlink is already locked.
810 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
812 struct net *net = dev_net(ifa->ifa_dev->dev);
813 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
814 struct fib_table *tb;
815 struct fib_config cfg = {
816 .fc_protocol = RTPROT_KERNEL,
819 .fc_dst_len = dst_len,
820 .fc_prefsrc = ifa->ifa_local,
821 .fc_oif = ifa->ifa_dev->dev->ifindex,
822 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
829 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
831 tb = fib_new_table(net, tb_id);
835 cfg.fc_table = tb->tb_id;
837 if (type != RTN_LOCAL)
838 cfg.fc_scope = RT_SCOPE_LINK;
840 cfg.fc_scope = RT_SCOPE_HOST;
842 if (cmd == RTM_NEWROUTE)
843 fib_table_insert(tb, &cfg);
845 fib_table_delete(tb, &cfg);
848 void fib_add_ifaddr(struct in_ifaddr *ifa)
850 struct in_device *in_dev = ifa->ifa_dev;
851 struct net_device *dev = in_dev->dev;
852 struct in_ifaddr *prim = ifa;
853 __be32 mask = ifa->ifa_mask;
854 __be32 addr = ifa->ifa_local;
855 __be32 prefix = ifa->ifa_address & mask;
857 if (ifa->ifa_flags & IFA_F_SECONDARY) {
858 prim = inet_ifa_byprefix(in_dev, prefix, mask);
860 pr_warn("%s: bug: prim == NULL\n", __func__);
865 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
867 if (!(dev->flags & IFF_UP))
870 /* Add broadcast address, if it is explicitly assigned. */
871 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
872 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
874 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
875 (prefix != addr || ifa->ifa_prefixlen < 32)) {
876 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
877 fib_magic(RTM_NEWROUTE,
878 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
879 prefix, ifa->ifa_prefixlen, prim);
881 /* Add network specific broadcasts, when it takes a sense */
882 if (ifa->ifa_prefixlen < 31) {
883 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
884 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
890 /* Delete primary or secondary address.
891 * Optionally, on secondary address promotion consider the addresses
892 * from subnet iprim as deleted, even if they are in device list.
893 * In this case the secondary ifa can be in device list.
895 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
897 struct in_device *in_dev = ifa->ifa_dev;
898 struct net_device *dev = in_dev->dev;
899 struct in_ifaddr *ifa1;
900 struct in_ifaddr *prim = ifa, *prim1 = NULL;
901 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
902 __be32 any = ifa->ifa_address & ifa->ifa_mask;
908 int subnet = 0; /* Primary network */
909 int gone = 1; /* Address is missing */
910 int same_prefsrc = 0; /* Another primary with same IP */
912 if (ifa->ifa_flags & IFA_F_SECONDARY) {
913 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
915 /* if the device has been deleted, we don't perform
919 pr_warn("%s: bug: prim == NULL\n", __func__);
922 if (iprim && iprim != prim) {
923 pr_warn("%s: bug: iprim != prim\n", __func__);
926 } else if (!ipv4_is_zeronet(any) &&
927 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
928 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
929 fib_magic(RTM_DELROUTE,
930 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
931 any, ifa->ifa_prefixlen, prim);
938 /* Deletion is more complicated than add.
939 * We should take care of not to delete too much :-)
941 * Scan address list to be sure that addresses are really gone.
944 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
946 /* promotion, keep the IP */
950 /* Ignore IFAs from our subnet */
951 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
952 inet_ifa_match(ifa1->ifa_address, iprim))
955 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
956 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
957 /* Another address from our subnet? */
958 if (ifa1->ifa_mask == prim->ifa_mask &&
959 inet_ifa_match(ifa1->ifa_address, prim))
962 /* We reached the secondaries, so
963 * same_prefsrc should be determined.
967 /* Search new prim1 if ifa1 is not
968 * using the current prim1
971 ifa1->ifa_mask != prim1->ifa_mask ||
972 !inet_ifa_match(ifa1->ifa_address, prim1))
973 prim1 = inet_ifa_byprefix(in_dev,
978 if (prim1->ifa_local != prim->ifa_local)
982 if (prim->ifa_local != ifa1->ifa_local)
988 if (ifa->ifa_local == ifa1->ifa_local)
990 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
992 if (brd == ifa1->ifa_broadcast)
994 if (any == ifa1->ifa_broadcast)
996 /* primary has network specific broadcasts */
997 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
998 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
999 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1001 if (!ipv4_is_zeronet(any1)) {
1002 if (ifa->ifa_broadcast == brd1 ||
1003 ifa->ifa_broadcast == any1)
1005 if (brd == brd1 || brd == any1)
1007 if (any == brd1 || any == any1)
1015 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1016 if (subnet && ifa->ifa_prefixlen < 31) {
1017 if (!(ok & BRD1_OK))
1018 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1019 if (!(ok & BRD0_OK))
1020 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1022 if (!(ok & LOCAL_OK)) {
1023 unsigned int addr_type;
1025 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1027 /* Check, that this local address finally disappeared. */
1028 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1030 if (gone && addr_type != RTN_LOCAL) {
1031 /* And the last, but not the least thing.
1032 * We must flush stray FIB entries.
1034 * First of all, we scan fib_info list searching
1035 * for stray nexthop entries, then ignite fib_flush.
1037 if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
1038 fib_flush(dev_net(dev));
1047 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1050 struct fib_result res;
1051 struct flowi4 fl4 = {
1052 .flowi4_mark = frn->fl_mark,
1053 .daddr = frn->fl_addr,
1054 .flowi4_tos = frn->fl_tos,
1055 .flowi4_scope = frn->fl_scope,
1057 struct fib_table *tb;
1061 tb = fib_get_table(net, frn->tb_id_in);
1067 frn->tb_id = tb->tb_id;
1068 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1071 frn->prefixlen = res.prefixlen;
1072 frn->nh_sel = res.nh_sel;
1073 frn->type = res.type;
1074 frn->scope = res.scope;
1082 static void nl_fib_input(struct sk_buff *skb)
1085 struct fib_result_nl *frn;
1086 struct nlmsghdr *nlh;
1089 net = sock_net(skb->sk);
1090 nlh = nlmsg_hdr(skb);
1091 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1092 skb->len < nlh->nlmsg_len ||
1093 nlmsg_len(nlh) < sizeof(*frn))
1096 skb = netlink_skb_clone(skb, GFP_KERNEL);
1099 nlh = nlmsg_hdr(skb);
1101 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1102 nl_fib_lookup(net, frn);
1104 portid = NETLINK_CB(skb).portid; /* netlink portid */
1105 NETLINK_CB(skb).portid = 0; /* from kernel */
1106 NETLINK_CB(skb).dst_group = 0; /* unicast */
1107 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1110 static int __net_init nl_fib_lookup_init(struct net *net)
1113 struct netlink_kernel_cfg cfg = {
1114 .input = nl_fib_input,
1117 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1119 return -EAFNOSUPPORT;
1120 net->ipv4.fibnl = sk;
1124 static void nl_fib_lookup_exit(struct net *net)
1126 netlink_kernel_release(net->ipv4.fibnl);
1127 net->ipv4.fibnl = NULL;
1130 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1133 if (fib_sync_down_dev(dev, event, force))
1134 fib_flush(dev_net(dev));
1135 rt_cache_flush(dev_net(dev));
1139 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1141 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1142 struct net_device *dev = ifa->ifa_dev->dev;
1143 struct net *net = dev_net(dev);
1147 fib_add_ifaddr(ifa);
1148 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1149 fib_sync_up(dev, RTNH_F_DEAD);
1151 atomic_inc(&net->ipv4.dev_addr_genid);
1152 rt_cache_flush(dev_net(dev));
1155 fib_del_ifaddr(ifa, NULL);
1156 atomic_inc(&net->ipv4.dev_addr_genid);
1157 if (!ifa->ifa_dev->ifa_list) {
1158 /* Last address was deleted from this interface.
1161 fib_disable_ip(dev, event, true);
1163 rt_cache_flush(dev_net(dev));
1170 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1172 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1173 struct netdev_notifier_changeupper_info *info;
1174 struct in_device *in_dev;
1175 struct net *net = dev_net(dev);
1178 if (event == NETDEV_UNREGISTER) {
1179 fib_disable_ip(dev, event, true);
1184 in_dev = __in_dev_get_rtnl(dev);
1191 fib_add_ifaddr(ifa);
1192 } endfor_ifa(in_dev);
1193 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1194 fib_sync_up(dev, RTNH_F_DEAD);
1196 atomic_inc(&net->ipv4.dev_addr_genid);
1197 rt_cache_flush(net);
1200 fib_disable_ip(dev, event, false);
1203 flags = dev_get_flags(dev);
1204 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1205 fib_sync_up(dev, RTNH_F_LINKDOWN);
1207 fib_sync_down_dev(dev, event, false);
1209 case NETDEV_CHANGEMTU:
1210 rt_cache_flush(net);
1212 case NETDEV_CHANGEUPPER:
1214 /* flush all routes if dev is linked to or unlinked from
1215 * an L3 master device (e.g., VRF)
1217 if (info->upper_dev && netif_is_l3_master(info->upper_dev))
1218 fib_disable_ip(dev, NETDEV_DOWN, true);
1224 static struct notifier_block fib_inetaddr_notifier = {
1225 .notifier_call = fib_inetaddr_event,
1228 static struct notifier_block fib_netdev_notifier = {
1229 .notifier_call = fib_netdev_event,
1232 static int __net_init ip_fib_net_init(struct net *net)
1235 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1237 /* Avoid false sharing : Use at least a full cache line */
1238 size = max_t(size_t, size, L1_CACHE_BYTES);
1240 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1241 if (!net->ipv4.fib_table_hash)
1244 err = fib4_rules_init(net);
1250 kfree(net->ipv4.fib_table_hash);
1254 static void ip_fib_net_exit(struct net *net)
1259 #ifdef CONFIG_IP_MULTIPLE_TABLES
1260 RCU_INIT_POINTER(net->ipv4.fib_local, NULL);
1261 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1262 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1264 /* Destroy the tables in reverse order to guarantee that the
1265 * local table, ID 255, is destroyed before the main table, ID
1266 * 254. This is necessary as the local table may contain
1267 * references to data contained in the main table.
1269 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1270 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1271 struct hlist_node *tmp;
1272 struct fib_table *tb;
1274 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1275 hlist_del(&tb->tb_hlist);
1276 fib_table_flush(tb);
1281 #ifdef CONFIG_IP_MULTIPLE_TABLES
1282 fib4_rules_exit(net);
1285 kfree(net->ipv4.fib_table_hash);
1288 static int __net_init fib_net_init(struct net *net)
1292 #ifdef CONFIG_IP_ROUTE_CLASSID
1293 net->ipv4.fib_num_tclassid_users = 0;
1295 error = ip_fib_net_init(net);
1298 error = nl_fib_lookup_init(net);
1301 error = fib_proc_init(net);
1308 nl_fib_lookup_exit(net);
1310 ip_fib_net_exit(net);
1314 static void __net_exit fib_net_exit(struct net *net)
1317 nl_fib_lookup_exit(net);
1318 ip_fib_net_exit(net);
1321 static struct pernet_operations fib_net_ops = {
1322 .init = fib_net_init,
1323 .exit = fib_net_exit,
1326 void __init ip_fib_init(void)
1330 register_pernet_subsys(&fib_net_ops);
1332 register_netdevice_notifier(&fib_netdev_notifier);
1333 register_inetaddr_notifier(&fib_inetaddr_notifier);
1335 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1336 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1337 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);