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)
89 alias = fib_new_table(net, RT_TABLE_MAIN);
91 tb = fib_trie_table(id, alias);
97 rcu_assign_pointer(net->ipv4.fib_main, tb);
99 case RT_TABLE_DEFAULT:
100 rcu_assign_pointer(net->ipv4.fib_default, tb);
106 h = id & (FIB_TABLE_HASHSZ - 1);
107 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
110 EXPORT_SYMBOL_GPL(fib_new_table);
112 /* caller must hold either rtnl or rcu read lock */
113 struct fib_table *fib_get_table(struct net *net, u32 id)
115 struct fib_table *tb;
116 struct hlist_head *head;
121 h = id & (FIB_TABLE_HASHSZ - 1);
123 head = &net->ipv4.fib_table_hash[h];
124 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
130 #endif /* CONFIG_IP_MULTIPLE_TABLES */
132 static void fib_replace_table(struct net *net, struct fib_table *old,
133 struct fib_table *new)
135 #ifdef CONFIG_IP_MULTIPLE_TABLES
136 switch (new->tb_id) {
138 rcu_assign_pointer(net->ipv4.fib_main, new);
140 case RT_TABLE_DEFAULT:
141 rcu_assign_pointer(net->ipv4.fib_default, new);
148 /* replace the old table in the hlist */
149 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
152 int fib_unmerge(struct net *net)
154 struct fib_table *old, *new;
156 /* attempt to fetch local table if it has been allocated */
157 old = fib_get_table(net, RT_TABLE_LOCAL);
161 new = fib_trie_unmerge(old);
165 /* replace merged table with clean table */
167 fib_replace_table(net, old, new);
174 static void fib_flush(struct net *net)
179 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
180 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
181 struct hlist_node *tmp;
182 struct fib_table *tb;
184 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
185 flushed += fib_table_flush(net, tb);
192 void fib_flush_external(struct net *net)
194 struct fib_table *tb;
195 struct hlist_head *head;
198 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
199 head = &net->ipv4.fib_table_hash[h];
200 hlist_for_each_entry(tb, head, tb_hlist)
201 fib_table_flush_external(tb);
206 * Find address type as if only "dev" was present in the system. If
207 * on_dev is NULL then all interfaces are taken into consideration.
209 static inline unsigned int __inet_dev_addr_type(struct net *net,
210 const struct net_device *dev,
211 __be32 addr, u32 tb_id)
213 struct flowi4 fl4 = { .daddr = addr };
214 struct fib_result res;
215 unsigned int ret = RTN_BROADCAST;
216 struct fib_table *table;
218 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
219 return RTN_BROADCAST;
220 if (ipv4_is_multicast(addr))
221 return RTN_MULTICAST;
225 table = fib_get_table(net, tb_id);
228 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
229 if (!dev || dev == res.fi->fib_dev)
238 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
240 return __inet_dev_addr_type(net, NULL, addr, tb_id);
242 EXPORT_SYMBOL(inet_addr_type_table);
244 unsigned int inet_addr_type(struct net *net, __be32 addr)
246 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
248 EXPORT_SYMBOL(inet_addr_type);
250 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
253 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
255 return __inet_dev_addr_type(net, dev, addr, rt_table);
257 EXPORT_SYMBOL(inet_dev_addr_type);
259 /* inet_addr_type with dev == NULL but using the table from a dev
260 * if one is associated
262 unsigned int inet_addr_type_dev_table(struct net *net,
263 const struct net_device *dev,
266 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
268 return __inet_dev_addr_type(net, NULL, addr, rt_table);
270 EXPORT_SYMBOL(inet_addr_type_dev_table);
272 __be32 fib_compute_spec_dst(struct sk_buff *skb)
274 struct net_device *dev = skb->dev;
275 struct in_device *in_dev;
276 struct fib_result res;
281 rt = skb_rtable(skb);
282 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
284 return ip_hdr(skb)->daddr;
286 in_dev = __in_dev_get_rcu(dev);
291 scope = RT_SCOPE_UNIVERSE;
292 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
293 struct flowi4 fl4 = {
294 .flowi4_iif = LOOPBACK_IFINDEX,
295 .daddr = ip_hdr(skb)->saddr,
296 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
297 .flowi4_scope = scope,
298 .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0,
300 if (!fib_lookup(net, &fl4, &res, 0))
301 return FIB_RES_PREFSRC(net, res);
303 scope = RT_SCOPE_LINK;
306 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
309 /* Given (packet source, input interface) and optional (dst, oif, tos):
310 * - (main) check, that source is valid i.e. not broadcast or our local
312 * - figure out what "logical" interface this packet arrived
313 * and calculate "specific destination" address.
314 * - check, that packet arrived from expected physical interface.
315 * called with rcu_read_lock()
317 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
318 u8 tos, int oif, struct net_device *dev,
319 int rpf, struct in_device *idev, u32 *itag)
322 struct fib_result res;
328 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
330 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
333 fl4.flowi4_tos = tos;
334 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
335 fl4.flowi4_tun_key.tun_id = 0;
336 fl4.flowi4_flags = 0;
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);
345 if (fib_lookup(net, &fl4, &res, 0))
347 if (res.type != RTN_UNICAST &&
348 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
350 if (!rpf && !fib_num_tclassid_users(dev_net(dev)) &&
351 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
353 fib_combine_itag(itag, &res);
356 #ifdef CONFIG_IP_ROUTE_MULTIPATH
357 for (ret = 0; ret < res.fi->fib_nhs; ret++) {
358 struct fib_nh *nh = &res.fi->fib_nh[ret];
360 if (nh->nh_dev == dev) {
363 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
369 if (FIB_RES_DEV(res) == dev)
373 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
380 fl4.flowi4_oif = dev->ifindex;
383 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
384 if (res.type == RTN_UNICAST)
385 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
401 /* Ignore rp_filter for packets protected by IPsec. */
402 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
403 u8 tos, int oif, struct net_device *dev,
404 struct in_device *idev, u32 *itag)
406 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
408 if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
409 IN_DEV_ACCEPT_LOCAL(idev) &&
410 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
414 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
417 static inline __be32 sk_extract_addr(struct sockaddr *addr)
419 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
422 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
426 nla = (struct nlattr *) ((char *) mx + len);
427 nla->nla_type = type;
428 nla->nla_len = nla_attr_size(4);
429 *(u32 *) nla_data(nla) = value;
431 return len + nla_total_size(4);
434 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
435 struct fib_config *cfg)
440 memset(cfg, 0, sizeof(*cfg));
441 cfg->fc_nlinfo.nl_net = net;
443 if (rt->rt_dst.sa_family != AF_INET)
444 return -EAFNOSUPPORT;
447 * Check mask for validity:
448 * a) it must be contiguous.
449 * b) destination must have all host bits clear.
450 * c) if application forgot to set correct family (AF_INET),
451 * reject request unless it is absolutely clear i.e.
452 * both family and mask are zero.
455 addr = sk_extract_addr(&rt->rt_dst);
456 if (!(rt->rt_flags & RTF_HOST)) {
457 __be32 mask = sk_extract_addr(&rt->rt_genmask);
459 if (rt->rt_genmask.sa_family != AF_INET) {
460 if (mask || rt->rt_genmask.sa_family)
461 return -EAFNOSUPPORT;
464 if (bad_mask(mask, addr))
467 plen = inet_mask_len(mask);
470 cfg->fc_dst_len = plen;
473 if (cmd != SIOCDELRT) {
474 cfg->fc_nlflags = NLM_F_CREATE;
475 cfg->fc_protocol = RTPROT_BOOT;
479 cfg->fc_priority = rt->rt_metric - 1;
481 if (rt->rt_flags & RTF_REJECT) {
482 cfg->fc_scope = RT_SCOPE_HOST;
483 cfg->fc_type = RTN_UNREACHABLE;
487 cfg->fc_scope = RT_SCOPE_NOWHERE;
488 cfg->fc_type = RTN_UNICAST;
492 struct net_device *dev;
493 char devname[IFNAMSIZ];
495 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
498 devname[IFNAMSIZ-1] = 0;
499 colon = strchr(devname, ':');
502 dev = __dev_get_by_name(net, devname);
505 cfg->fc_oif = dev->ifindex;
506 cfg->fc_table = l3mdev_fib_table(dev);
508 struct in_ifaddr *ifa;
509 struct in_device *in_dev = __in_dev_get_rtnl(dev);
513 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
514 if (strcmp(ifa->ifa_label, devname) == 0)
518 cfg->fc_prefsrc = ifa->ifa_local;
522 addr = sk_extract_addr(&rt->rt_gateway);
523 if (rt->rt_gateway.sa_family == AF_INET && addr) {
524 unsigned int addr_type;
527 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
528 if (rt->rt_flags & RTF_GATEWAY &&
529 addr_type == RTN_UNICAST)
530 cfg->fc_scope = RT_SCOPE_UNIVERSE;
533 if (cmd == SIOCDELRT)
536 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
539 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
540 cfg->fc_scope = RT_SCOPE_LINK;
542 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
546 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
550 if (rt->rt_flags & RTF_MTU)
551 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
553 if (rt->rt_flags & RTF_WINDOW)
554 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
556 if (rt->rt_flags & RTF_IRTT)
557 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
560 cfg->fc_mx_len = len;
567 * Handle IP routing ioctl calls.
568 * These are used to manipulate the routing tables
570 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
572 struct fib_config cfg;
577 case SIOCADDRT: /* Add a route */
578 case SIOCDELRT: /* Delete a route */
579 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
582 if (copy_from_user(&rt, arg, sizeof(rt)))
586 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
588 struct fib_table *tb;
590 if (cmd == SIOCDELRT) {
591 tb = fib_get_table(net, cfg.fc_table);
593 err = fib_table_delete(net, tb, &cfg);
597 tb = fib_new_table(net, cfg.fc_table);
599 err = fib_table_insert(net, tb, &cfg);
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 },
628 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
629 struct nlmsghdr *nlh, struct fib_config *cfg)
635 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
639 memset(cfg, 0, sizeof(*cfg));
641 rtm = nlmsg_data(nlh);
642 cfg->fc_dst_len = rtm->rtm_dst_len;
643 cfg->fc_tos = rtm->rtm_tos;
644 cfg->fc_table = rtm->rtm_table;
645 cfg->fc_protocol = rtm->rtm_protocol;
646 cfg->fc_scope = rtm->rtm_scope;
647 cfg->fc_type = rtm->rtm_type;
648 cfg->fc_flags = rtm->rtm_flags;
649 cfg->fc_nlflags = nlh->nlmsg_flags;
651 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
652 cfg->fc_nlinfo.nlh = nlh;
653 cfg->fc_nlinfo.nl_net = net;
655 if (cfg->fc_type > RTN_MAX) {
660 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
661 switch (nla_type(attr)) {
663 cfg->fc_dst = nla_get_be32(attr);
666 cfg->fc_oif = nla_get_u32(attr);
669 cfg->fc_gw = nla_get_be32(attr);
672 cfg->fc_priority = nla_get_u32(attr);
675 cfg->fc_prefsrc = nla_get_be32(attr);
678 cfg->fc_mx = nla_data(attr);
679 cfg->fc_mx_len = nla_len(attr);
682 cfg->fc_mp = nla_data(attr);
683 cfg->fc_mp_len = nla_len(attr);
686 cfg->fc_flow = nla_get_u32(attr);
689 cfg->fc_table = nla_get_u32(attr);
692 cfg->fc_encap = attr;
695 cfg->fc_encap_type = nla_get_u16(attr);
705 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
707 struct net *net = sock_net(skb->sk);
708 struct fib_config cfg;
709 struct fib_table *tb;
712 err = rtm_to_fib_config(net, skb, nlh, &cfg);
716 tb = fib_get_table(net, cfg.fc_table);
722 err = fib_table_delete(net, tb, &cfg);
727 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
729 struct net *net = sock_net(skb->sk);
730 struct fib_config cfg;
731 struct fib_table *tb;
734 err = rtm_to_fib_config(net, skb, nlh, &cfg);
738 tb = fib_new_table(net, cfg.fc_table);
744 err = fib_table_insert(net, tb, &cfg);
749 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
751 struct net *net = sock_net(skb->sk);
753 unsigned int e = 0, s_e;
754 struct fib_table *tb;
755 struct hlist_head *head;
758 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
759 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
767 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
769 head = &net->ipv4.fib_table_hash[h];
770 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
774 memset(&cb->args[2], 0, sizeof(cb->args) -
775 2 * sizeof(cb->args[0]));
776 if (fib_table_dump(tb, skb, cb) < 0)
792 /* Prepare and feed intra-kernel routing request.
793 * Really, it should be netlink message, but :-( netlink
794 * can be not configured, so that we feed it directly
795 * to fib engine. It is legal, because all events occur
796 * only when netlink is already locked.
798 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
800 struct net *net = dev_net(ifa->ifa_dev->dev);
801 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
802 struct fib_table *tb;
803 struct fib_config cfg = {
804 .fc_protocol = RTPROT_KERNEL,
807 .fc_dst_len = dst_len,
808 .fc_prefsrc = ifa->ifa_local,
809 .fc_oif = ifa->ifa_dev->dev->ifindex,
810 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
817 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
819 tb = fib_new_table(net, tb_id);
823 cfg.fc_table = tb->tb_id;
825 if (type != RTN_LOCAL)
826 cfg.fc_scope = RT_SCOPE_LINK;
828 cfg.fc_scope = RT_SCOPE_HOST;
830 if (cmd == RTM_NEWROUTE)
831 fib_table_insert(net, tb, &cfg);
833 fib_table_delete(net, tb, &cfg);
836 void fib_add_ifaddr(struct in_ifaddr *ifa)
838 struct in_device *in_dev = ifa->ifa_dev;
839 struct net_device *dev = in_dev->dev;
840 struct in_ifaddr *prim = ifa;
841 __be32 mask = ifa->ifa_mask;
842 __be32 addr = ifa->ifa_local;
843 __be32 prefix = ifa->ifa_address & mask;
845 if (ifa->ifa_flags & IFA_F_SECONDARY) {
846 prim = inet_ifa_byprefix(in_dev, prefix, mask);
848 pr_warn("%s: bug: prim == NULL\n", __func__);
853 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
855 if (!(dev->flags & IFF_UP))
858 /* Add broadcast address, if it is explicitly assigned. */
859 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
860 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
862 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
863 (prefix != addr || ifa->ifa_prefixlen < 32)) {
864 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
865 fib_magic(RTM_NEWROUTE,
866 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
867 prefix, ifa->ifa_prefixlen, prim);
869 /* Add network specific broadcasts, when it takes a sense */
870 if (ifa->ifa_prefixlen < 31) {
871 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
872 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
878 /* Delete primary or secondary address.
879 * Optionally, on secondary address promotion consider the addresses
880 * from subnet iprim as deleted, even if they are in device list.
881 * In this case the secondary ifa can be in device list.
883 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
885 struct in_device *in_dev = ifa->ifa_dev;
886 struct net_device *dev = in_dev->dev;
887 struct in_ifaddr *ifa1;
888 struct in_ifaddr *prim = ifa, *prim1 = NULL;
889 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
890 __be32 any = ifa->ifa_address & ifa->ifa_mask;
896 int subnet = 0; /* Primary network */
897 int gone = 1; /* Address is missing */
898 int same_prefsrc = 0; /* Another primary with same IP */
900 if (ifa->ifa_flags & IFA_F_SECONDARY) {
901 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
903 /* if the device has been deleted, we don't perform
907 pr_warn("%s: bug: prim == NULL\n", __func__);
910 if (iprim && iprim != prim) {
911 pr_warn("%s: bug: iprim != prim\n", __func__);
914 } else if (!ipv4_is_zeronet(any) &&
915 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
916 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
917 fib_magic(RTM_DELROUTE,
918 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
919 any, ifa->ifa_prefixlen, prim);
926 /* Deletion is more complicated than add.
927 * We should take care of not to delete too much :-)
929 * Scan address list to be sure that addresses are really gone.
932 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
934 /* promotion, keep the IP */
938 /* Ignore IFAs from our subnet */
939 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
940 inet_ifa_match(ifa1->ifa_address, iprim))
943 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
944 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
945 /* Another address from our subnet? */
946 if (ifa1->ifa_mask == prim->ifa_mask &&
947 inet_ifa_match(ifa1->ifa_address, prim))
950 /* We reached the secondaries, so
951 * same_prefsrc should be determined.
955 /* Search new prim1 if ifa1 is not
956 * using the current prim1
959 ifa1->ifa_mask != prim1->ifa_mask ||
960 !inet_ifa_match(ifa1->ifa_address, prim1))
961 prim1 = inet_ifa_byprefix(in_dev,
966 if (prim1->ifa_local != prim->ifa_local)
970 if (prim->ifa_local != ifa1->ifa_local)
976 if (ifa->ifa_local == ifa1->ifa_local)
978 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
980 if (brd == ifa1->ifa_broadcast)
982 if (any == ifa1->ifa_broadcast)
984 /* primary has network specific broadcasts */
985 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
986 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
987 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
989 if (!ipv4_is_zeronet(any1)) {
990 if (ifa->ifa_broadcast == brd1 ||
991 ifa->ifa_broadcast == any1)
993 if (brd == brd1 || brd == any1)
995 if (any == brd1 || any == any1)
1003 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1004 if (subnet && ifa->ifa_prefixlen < 31) {
1005 if (!(ok & BRD1_OK))
1006 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1007 if (!(ok & BRD0_OK))
1008 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1010 if (!(ok & LOCAL_OK)) {
1011 unsigned int addr_type;
1013 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1015 /* Check, that this local address finally disappeared. */
1016 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1018 if (gone && addr_type != RTN_LOCAL) {
1019 /* And the last, but not the least thing.
1020 * We must flush stray FIB entries.
1022 * First of all, we scan fib_info list searching
1023 * for stray nexthop entries, then ignite fib_flush.
1025 if (fib_sync_down_addr(dev, ifa->ifa_local))
1026 fib_flush(dev_net(dev));
1035 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1038 struct fib_result res;
1039 struct flowi4 fl4 = {
1040 .flowi4_mark = frn->fl_mark,
1041 .daddr = frn->fl_addr,
1042 .flowi4_tos = frn->fl_tos,
1043 .flowi4_scope = frn->fl_scope,
1045 struct fib_table *tb;
1049 tb = fib_get_table(net, frn->tb_id_in);
1055 frn->tb_id = tb->tb_id;
1056 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1059 frn->prefixlen = res.prefixlen;
1060 frn->nh_sel = res.nh_sel;
1061 frn->type = res.type;
1062 frn->scope = res.scope;
1070 static void nl_fib_input(struct sk_buff *skb)
1073 struct fib_result_nl *frn;
1074 struct nlmsghdr *nlh;
1077 net = sock_net(skb->sk);
1078 nlh = nlmsg_hdr(skb);
1079 if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len ||
1080 nlmsg_len(nlh) < sizeof(*frn))
1083 skb = netlink_skb_clone(skb, GFP_KERNEL);
1086 nlh = nlmsg_hdr(skb);
1088 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1089 nl_fib_lookup(net, frn);
1091 portid = NETLINK_CB(skb).portid; /* netlink portid */
1092 NETLINK_CB(skb).portid = 0; /* from kernel */
1093 NETLINK_CB(skb).dst_group = 0; /* unicast */
1094 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1097 static int __net_init nl_fib_lookup_init(struct net *net)
1100 struct netlink_kernel_cfg cfg = {
1101 .input = nl_fib_input,
1104 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1106 return -EAFNOSUPPORT;
1107 net->ipv4.fibnl = sk;
1111 static void nl_fib_lookup_exit(struct net *net)
1113 netlink_kernel_release(net->ipv4.fibnl);
1114 net->ipv4.fibnl = NULL;
1117 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1120 if (fib_sync_down_dev(dev, event, force))
1121 fib_flush(dev_net(dev));
1122 rt_cache_flush(dev_net(dev));
1126 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1128 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1129 struct net_device *dev = ifa->ifa_dev->dev;
1130 struct net *net = dev_net(dev);
1134 fib_add_ifaddr(ifa);
1135 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1136 fib_sync_up(dev, RTNH_F_DEAD);
1138 atomic_inc(&net->ipv4.dev_addr_genid);
1139 rt_cache_flush(dev_net(dev));
1142 fib_del_ifaddr(ifa, NULL);
1143 atomic_inc(&net->ipv4.dev_addr_genid);
1144 if (!ifa->ifa_dev->ifa_list) {
1145 /* Last address was deleted from this interface.
1148 fib_disable_ip(dev, event, true);
1150 rt_cache_flush(dev_net(dev));
1157 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1159 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1160 struct netdev_notifier_changeupper_info *info;
1161 struct in_device *in_dev;
1162 struct net *net = dev_net(dev);
1165 if (event == NETDEV_UNREGISTER) {
1166 fib_disable_ip(dev, event, true);
1171 in_dev = __in_dev_get_rtnl(dev);
1178 fib_add_ifaddr(ifa);
1179 } endfor_ifa(in_dev);
1180 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1181 fib_sync_up(dev, RTNH_F_DEAD);
1183 atomic_inc(&net->ipv4.dev_addr_genid);
1184 rt_cache_flush(net);
1187 fib_disable_ip(dev, event, false);
1190 flags = dev_get_flags(dev);
1191 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1192 fib_sync_up(dev, RTNH_F_LINKDOWN);
1194 fib_sync_down_dev(dev, event, false);
1196 case NETDEV_CHANGEMTU:
1197 rt_cache_flush(net);
1199 case NETDEV_CHANGEUPPER:
1201 /* flush all routes if dev is linked to or unlinked from
1202 * an L3 master device (e.g., VRF)
1204 if (info->upper_dev && netif_is_l3_master(info->upper_dev))
1205 fib_disable_ip(dev, NETDEV_DOWN, true);
1211 static struct notifier_block fib_inetaddr_notifier = {
1212 .notifier_call = fib_inetaddr_event,
1215 static struct notifier_block fib_netdev_notifier = {
1216 .notifier_call = fib_netdev_event,
1219 static int __net_init ip_fib_net_init(struct net *net)
1222 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1224 /* Avoid false sharing : Use at least a full cache line */
1225 size = max_t(size_t, size, L1_CACHE_BYTES);
1227 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1228 if (!net->ipv4.fib_table_hash)
1231 err = fib4_rules_init(net);
1237 kfree(net->ipv4.fib_table_hash);
1241 static void ip_fib_net_exit(struct net *net)
1246 #ifdef CONFIG_IP_MULTIPLE_TABLES
1247 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1248 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1250 for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1251 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1252 struct hlist_node *tmp;
1253 struct fib_table *tb;
1255 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1256 hlist_del(&tb->tb_hlist);
1257 fib_table_flush(net, tb);
1262 #ifdef CONFIG_IP_MULTIPLE_TABLES
1263 fib4_rules_exit(net);
1266 kfree(net->ipv4.fib_table_hash);
1269 static int __net_init fib_net_init(struct net *net)
1273 #ifdef CONFIG_IP_ROUTE_CLASSID
1274 net->ipv4.fib_num_tclassid_users = 0;
1276 error = ip_fib_net_init(net);
1279 error = nl_fib_lookup_init(net);
1282 error = fib_proc_init(net);
1289 nl_fib_lookup_exit(net);
1291 ip_fib_net_exit(net);
1295 static void __net_exit fib_net_exit(struct net *net)
1298 nl_fib_lookup_exit(net);
1299 ip_fib_net_exit(net);
1302 static struct pernet_operations fib_net_ops = {
1303 .init = fib_net_init,
1304 .exit = fib_net_exit,
1307 void __init ip_fib_init(void)
1309 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1310 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1311 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1313 register_pernet_subsys(&fib_net_ops);
1314 register_netdevice_notifier(&fib_netdev_notifier);
1315 register_inetaddr_notifier(&fib_inetaddr_notifier);