Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[cris-mirror.git] / net / ipv4 / fib_frontend.c
blobf05afaf3235c0500a9087eae6365b7001aa64663
1 /*
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>
22 #include <linux/mm.h>
23 #include <linux/string.h>
24 #include <linux/socket.h>
25 #include <linux/sockios.h>
26 #include <linux/errno.h>
27 #include <linux/in.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>
39 #include <net/ip.h>
40 #include <net/protocol.h>
41 #include <net/route.h>
42 #include <net/tcp.h>
43 #include <net/sock.h>
44 #include <net/arp.h>
45 #include <net/ip_fib.h>
46 #include <net/rtnetlink.h>
47 #include <net/xfrm.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);
59 if (!main_table)
60 return -ENOMEM;
62 local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
63 if (!local_table)
64 goto fail;
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]);
70 return 0;
72 fail:
73 fib_free_table(main_table);
74 return -ENOMEM;
77 static bool fib4_has_custom_rules(struct net *net)
79 return false;
81 #else
83 struct fib_table *fib_new_table(struct net *net, u32 id)
85 struct fib_table *tb, *alias = NULL;
86 unsigned int h;
88 if (id == 0)
89 id = RT_TABLE_MAIN;
90 tb = fib_get_table(net, id);
91 if (tb)
92 return tb;
94 if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
95 alias = fib_new_table(net, RT_TABLE_MAIN);
97 tb = fib_trie_table(id, alias);
98 if (!tb)
99 return NULL;
101 switch (id) {
102 case RT_TABLE_MAIN:
103 rcu_assign_pointer(net->ipv4.fib_main, tb);
104 break;
105 case RT_TABLE_DEFAULT:
106 rcu_assign_pointer(net->ipv4.fib_default, tb);
107 break;
108 default:
109 break;
112 h = id & (FIB_TABLE_HASHSZ - 1);
113 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
114 return tb;
116 EXPORT_SYMBOL_GPL(fib_new_table);
118 /* caller must hold either rtnl or rcu read lock */
119 struct fib_table *fib_get_table(struct net *net, u32 id)
121 struct fib_table *tb;
122 struct hlist_head *head;
123 unsigned int h;
125 if (id == 0)
126 id = RT_TABLE_MAIN;
127 h = id & (FIB_TABLE_HASHSZ - 1);
129 head = &net->ipv4.fib_table_hash[h];
130 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
131 if (tb->tb_id == id)
132 return tb;
134 return NULL;
137 static bool fib4_has_custom_rules(struct net *net)
139 return net->ipv4.fib_has_custom_rules;
141 #endif /* CONFIG_IP_MULTIPLE_TABLES */
143 static void fib_replace_table(struct net *net, struct fib_table *old,
144 struct fib_table *new)
146 #ifdef CONFIG_IP_MULTIPLE_TABLES
147 switch (new->tb_id) {
148 case RT_TABLE_MAIN:
149 rcu_assign_pointer(net->ipv4.fib_main, new);
150 break;
151 case RT_TABLE_DEFAULT:
152 rcu_assign_pointer(net->ipv4.fib_default, new);
153 break;
154 default:
155 break;
158 #endif
159 /* replace the old table in the hlist */
160 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
163 int fib_unmerge(struct net *net)
165 struct fib_table *old, *new, *main_table;
167 /* attempt to fetch local table if it has been allocated */
168 old = fib_get_table(net, RT_TABLE_LOCAL);
169 if (!old)
170 return 0;
172 new = fib_trie_unmerge(old);
173 if (!new)
174 return -ENOMEM;
176 /* table is already unmerged */
177 if (new == old)
178 return 0;
180 /* replace merged table with clean table */
181 fib_replace_table(net, old, new);
182 fib_free_table(old);
184 /* attempt to fetch main table if it has been allocated */
185 main_table = fib_get_table(net, RT_TABLE_MAIN);
186 if (!main_table)
187 return 0;
189 /* flush local entries from main table */
190 fib_table_flush_external(main_table);
192 return 0;
195 static void fib_flush(struct net *net)
197 int flushed = 0;
198 unsigned int h;
200 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
201 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
202 struct hlist_node *tmp;
203 struct fib_table *tb;
205 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
206 flushed += fib_table_flush(net, tb);
209 if (flushed)
210 rt_cache_flush(net);
214 * Find address type as if only "dev" was present in the system. If
215 * on_dev is NULL then all interfaces are taken into consideration.
217 static inline unsigned int __inet_dev_addr_type(struct net *net,
218 const struct net_device *dev,
219 __be32 addr, u32 tb_id)
221 struct flowi4 fl4 = { .daddr = addr };
222 struct fib_result res;
223 unsigned int ret = RTN_BROADCAST;
224 struct fib_table *table;
226 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
227 return RTN_BROADCAST;
228 if (ipv4_is_multicast(addr))
229 return RTN_MULTICAST;
231 rcu_read_lock();
233 table = fib_get_table(net, tb_id);
234 if (table) {
235 ret = RTN_UNICAST;
236 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
237 if (!dev || dev == res.fi->fib_dev)
238 ret = res.type;
242 rcu_read_unlock();
243 return ret;
246 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
248 return __inet_dev_addr_type(net, NULL, addr, tb_id);
250 EXPORT_SYMBOL(inet_addr_type_table);
252 unsigned int inet_addr_type(struct net *net, __be32 addr)
254 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
256 EXPORT_SYMBOL(inet_addr_type);
258 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
259 __be32 addr)
261 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
263 return __inet_dev_addr_type(net, dev, addr, rt_table);
265 EXPORT_SYMBOL(inet_dev_addr_type);
267 /* inet_addr_type with dev == NULL but using the table from a dev
268 * if one is associated
270 unsigned int inet_addr_type_dev_table(struct net *net,
271 const struct net_device *dev,
272 __be32 addr)
274 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
276 return __inet_dev_addr_type(net, NULL, addr, rt_table);
278 EXPORT_SYMBOL(inet_addr_type_dev_table);
280 __be32 fib_compute_spec_dst(struct sk_buff *skb)
282 struct net_device *dev = skb->dev;
283 struct in_device *in_dev;
284 struct fib_result res;
285 struct rtable *rt;
286 struct net *net;
287 int scope;
289 rt = skb_rtable(skb);
290 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
291 RTCF_LOCAL)
292 return ip_hdr(skb)->daddr;
294 in_dev = __in_dev_get_rcu(dev);
295 BUG_ON(!in_dev);
297 net = dev_net(dev);
299 scope = RT_SCOPE_UNIVERSE;
300 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
301 struct flowi4 fl4 = {
302 .flowi4_iif = LOOPBACK_IFINDEX,
303 .daddr = ip_hdr(skb)->saddr,
304 .flowi4_tos = RT_TOS(ip_hdr(skb)->tos),
305 .flowi4_scope = scope,
306 .flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0,
308 if (!fib_lookup(net, &fl4, &res, 0))
309 return FIB_RES_PREFSRC(net, res);
310 } else {
311 scope = RT_SCOPE_LINK;
314 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
317 /* Given (packet source, input interface) and optional (dst, oif, tos):
318 * - (main) check, that source is valid i.e. not broadcast or our local
319 * address.
320 * - figure out what "logical" interface this packet arrived
321 * and calculate "specific destination" address.
322 * - check, that packet arrived from expected physical interface.
323 * called with rcu_read_lock()
325 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
326 u8 tos, int oif, struct net_device *dev,
327 int rpf, struct in_device *idev, u32 *itag)
329 int ret, no_addr;
330 struct fib_result res;
331 struct flowi4 fl4;
332 struct net *net = dev_net(dev);
333 bool dev_match;
335 fl4.flowi4_oif = 0;
336 fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
337 if (!fl4.flowi4_iif)
338 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
339 fl4.daddr = src;
340 fl4.saddr = dst;
341 fl4.flowi4_tos = tos;
342 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
343 fl4.flowi4_tun_key.tun_id = 0;
344 fl4.flowi4_flags = 0;
345 fl4.flowi4_uid = sock_net_uid(net, NULL);
347 no_addr = idev->ifa_list == NULL;
349 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
351 trace_fib_validate_source(dev, &fl4);
353 if (fib_lookup(net, &fl4, &res, 0))
354 goto last_resort;
355 if (res.type != RTN_UNICAST &&
356 (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
357 goto e_inval;
358 fib_combine_itag(itag, &res);
359 dev_match = false;
361 #ifdef CONFIG_IP_ROUTE_MULTIPATH
362 for (ret = 0; ret < res.fi->fib_nhs; ret++) {
363 struct fib_nh *nh = &res.fi->fib_nh[ret];
365 if (nh->nh_dev == dev) {
366 dev_match = true;
367 break;
368 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
369 dev_match = true;
370 break;
373 #else
374 if (FIB_RES_DEV(res) == dev)
375 dev_match = true;
376 #endif
377 if (dev_match) {
378 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
379 return ret;
381 if (no_addr)
382 goto last_resort;
383 if (rpf == 1)
384 goto e_rpf;
385 fl4.flowi4_oif = dev->ifindex;
387 ret = 0;
388 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
389 if (res.type == RTN_UNICAST)
390 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
392 return ret;
394 last_resort:
395 if (rpf)
396 goto e_rpf;
397 *itag = 0;
398 return 0;
400 e_inval:
401 return -EINVAL;
402 e_rpf:
403 return -EXDEV;
406 /* Ignore rp_filter for packets protected by IPsec. */
407 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
408 u8 tos, int oif, struct net_device *dev,
409 struct in_device *idev, u32 *itag)
411 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
412 struct net *net = dev_net(dev);
414 if (!r && !fib_num_tclassid_users(net) &&
415 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
416 if (IN_DEV_ACCEPT_LOCAL(idev))
417 goto ok;
418 /* with custom local routes in place, checking local addresses
419 * only will be too optimistic, with custom rules, checking
420 * local addresses only can be too strict, e.g. due to vrf
422 if (net->ipv4.fib_has_custom_local_routes ||
423 fib4_has_custom_rules(net))
424 goto full_check;
425 if (inet_lookup_ifaddr_rcu(net, src))
426 return -EINVAL;
429 *itag = 0;
430 return 0;
433 full_check:
434 return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
437 static inline __be32 sk_extract_addr(struct sockaddr *addr)
439 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
442 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
444 struct nlattr *nla;
446 nla = (struct nlattr *) ((char *) mx + len);
447 nla->nla_type = type;
448 nla->nla_len = nla_attr_size(4);
449 *(u32 *) nla_data(nla) = value;
451 return len + nla_total_size(4);
454 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
455 struct fib_config *cfg)
457 __be32 addr;
458 int plen;
460 memset(cfg, 0, sizeof(*cfg));
461 cfg->fc_nlinfo.nl_net = net;
463 if (rt->rt_dst.sa_family != AF_INET)
464 return -EAFNOSUPPORT;
467 * Check mask for validity:
468 * a) it must be contiguous.
469 * b) destination must have all host bits clear.
470 * c) if application forgot to set correct family (AF_INET),
471 * reject request unless it is absolutely clear i.e.
472 * both family and mask are zero.
474 plen = 32;
475 addr = sk_extract_addr(&rt->rt_dst);
476 if (!(rt->rt_flags & RTF_HOST)) {
477 __be32 mask = sk_extract_addr(&rt->rt_genmask);
479 if (rt->rt_genmask.sa_family != AF_INET) {
480 if (mask || rt->rt_genmask.sa_family)
481 return -EAFNOSUPPORT;
484 if (bad_mask(mask, addr))
485 return -EINVAL;
487 plen = inet_mask_len(mask);
490 cfg->fc_dst_len = plen;
491 cfg->fc_dst = addr;
493 if (cmd != SIOCDELRT) {
494 cfg->fc_nlflags = NLM_F_CREATE;
495 cfg->fc_protocol = RTPROT_BOOT;
498 if (rt->rt_metric)
499 cfg->fc_priority = rt->rt_metric - 1;
501 if (rt->rt_flags & RTF_REJECT) {
502 cfg->fc_scope = RT_SCOPE_HOST;
503 cfg->fc_type = RTN_UNREACHABLE;
504 return 0;
507 cfg->fc_scope = RT_SCOPE_NOWHERE;
508 cfg->fc_type = RTN_UNICAST;
510 if (rt->rt_dev) {
511 char *colon;
512 struct net_device *dev;
513 char devname[IFNAMSIZ];
515 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
516 return -EFAULT;
518 devname[IFNAMSIZ-1] = 0;
519 colon = strchr(devname, ':');
520 if (colon)
521 *colon = 0;
522 dev = __dev_get_by_name(net, devname);
523 if (!dev)
524 return -ENODEV;
525 cfg->fc_oif = dev->ifindex;
526 cfg->fc_table = l3mdev_fib_table(dev);
527 if (colon) {
528 struct in_ifaddr *ifa;
529 struct in_device *in_dev = __in_dev_get_rtnl(dev);
530 if (!in_dev)
531 return -ENODEV;
532 *colon = ':';
533 for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
534 if (strcmp(ifa->ifa_label, devname) == 0)
535 break;
536 if (!ifa)
537 return -ENODEV;
538 cfg->fc_prefsrc = ifa->ifa_local;
542 addr = sk_extract_addr(&rt->rt_gateway);
543 if (rt->rt_gateway.sa_family == AF_INET && addr) {
544 unsigned int addr_type;
546 cfg->fc_gw = addr;
547 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
548 if (rt->rt_flags & RTF_GATEWAY &&
549 addr_type == RTN_UNICAST)
550 cfg->fc_scope = RT_SCOPE_UNIVERSE;
553 if (cmd == SIOCDELRT)
554 return 0;
556 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
557 return -EINVAL;
559 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
560 cfg->fc_scope = RT_SCOPE_LINK;
562 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
563 struct nlattr *mx;
564 int len = 0;
566 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
567 if (!mx)
568 return -ENOMEM;
570 if (rt->rt_flags & RTF_MTU)
571 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
573 if (rt->rt_flags & RTF_WINDOW)
574 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
576 if (rt->rt_flags & RTF_IRTT)
577 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
579 cfg->fc_mx = mx;
580 cfg->fc_mx_len = len;
583 return 0;
587 * Handle IP routing ioctl calls.
588 * These are used to manipulate the routing tables
590 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
592 struct fib_config cfg;
593 int err;
595 switch (cmd) {
596 case SIOCADDRT: /* Add a route */
597 case SIOCDELRT: /* Delete a route */
598 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
599 return -EPERM;
601 rtnl_lock();
602 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
603 if (err == 0) {
604 struct fib_table *tb;
606 if (cmd == SIOCDELRT) {
607 tb = fib_get_table(net, cfg.fc_table);
608 if (tb)
609 err = fib_table_delete(net, tb, &cfg,
610 NULL);
611 else
612 err = -ESRCH;
613 } else {
614 tb = fib_new_table(net, cfg.fc_table);
615 if (tb)
616 err = fib_table_insert(net, tb,
617 &cfg, NULL);
618 else
619 err = -ENOBUFS;
622 /* allocated by rtentry_to_fib_config() */
623 kfree(cfg.fc_mx);
625 rtnl_unlock();
626 return err;
628 return -EINVAL;
631 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
632 [RTA_DST] = { .type = NLA_U32 },
633 [RTA_SRC] = { .type = NLA_U32 },
634 [RTA_IIF] = { .type = NLA_U32 },
635 [RTA_OIF] = { .type = NLA_U32 },
636 [RTA_GATEWAY] = { .type = NLA_U32 },
637 [RTA_PRIORITY] = { .type = NLA_U32 },
638 [RTA_PREFSRC] = { .type = NLA_U32 },
639 [RTA_METRICS] = { .type = NLA_NESTED },
640 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
641 [RTA_FLOW] = { .type = NLA_U32 },
642 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
643 [RTA_ENCAP] = { .type = NLA_NESTED },
644 [RTA_UID] = { .type = NLA_U32 },
645 [RTA_MARK] = { .type = NLA_U32 },
648 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
649 struct nlmsghdr *nlh, struct fib_config *cfg,
650 struct netlink_ext_ack *extack)
652 struct nlattr *attr;
653 int err, remaining;
654 struct rtmsg *rtm;
656 err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy,
657 extack);
658 if (err < 0)
659 goto errout;
661 memset(cfg, 0, sizeof(*cfg));
663 rtm = nlmsg_data(nlh);
664 cfg->fc_dst_len = rtm->rtm_dst_len;
665 cfg->fc_tos = rtm->rtm_tos;
666 cfg->fc_table = rtm->rtm_table;
667 cfg->fc_protocol = rtm->rtm_protocol;
668 cfg->fc_scope = rtm->rtm_scope;
669 cfg->fc_type = rtm->rtm_type;
670 cfg->fc_flags = rtm->rtm_flags;
671 cfg->fc_nlflags = nlh->nlmsg_flags;
673 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
674 cfg->fc_nlinfo.nlh = nlh;
675 cfg->fc_nlinfo.nl_net = net;
677 if (cfg->fc_type > RTN_MAX) {
678 NL_SET_ERR_MSG(extack, "Invalid route type");
679 err = -EINVAL;
680 goto errout;
683 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
684 switch (nla_type(attr)) {
685 case RTA_DST:
686 cfg->fc_dst = nla_get_be32(attr);
687 break;
688 case RTA_OIF:
689 cfg->fc_oif = nla_get_u32(attr);
690 break;
691 case RTA_GATEWAY:
692 cfg->fc_gw = nla_get_be32(attr);
693 break;
694 case RTA_PRIORITY:
695 cfg->fc_priority = nla_get_u32(attr);
696 break;
697 case RTA_PREFSRC:
698 cfg->fc_prefsrc = nla_get_be32(attr);
699 break;
700 case RTA_METRICS:
701 cfg->fc_mx = nla_data(attr);
702 cfg->fc_mx_len = nla_len(attr);
703 break;
704 case RTA_MULTIPATH:
705 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
706 nla_len(attr),
707 extack);
708 if (err < 0)
709 goto errout;
710 cfg->fc_mp = nla_data(attr);
711 cfg->fc_mp_len = nla_len(attr);
712 break;
713 case RTA_FLOW:
714 cfg->fc_flow = nla_get_u32(attr);
715 break;
716 case RTA_TABLE:
717 cfg->fc_table = nla_get_u32(attr);
718 break;
719 case RTA_ENCAP:
720 cfg->fc_encap = attr;
721 break;
722 case RTA_ENCAP_TYPE:
723 cfg->fc_encap_type = nla_get_u16(attr);
724 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
725 extack);
726 if (err < 0)
727 goto errout;
728 break;
732 return 0;
733 errout:
734 return err;
737 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
738 struct netlink_ext_ack *extack)
740 struct net *net = sock_net(skb->sk);
741 struct fib_config cfg;
742 struct fib_table *tb;
743 int err;
745 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
746 if (err < 0)
747 goto errout;
749 tb = fib_get_table(net, cfg.fc_table);
750 if (!tb) {
751 NL_SET_ERR_MSG(extack, "FIB table does not exist");
752 err = -ESRCH;
753 goto errout;
756 err = fib_table_delete(net, tb, &cfg, extack);
757 errout:
758 return err;
761 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
762 struct netlink_ext_ack *extack)
764 struct net *net = sock_net(skb->sk);
765 struct fib_config cfg;
766 struct fib_table *tb;
767 int err;
769 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
770 if (err < 0)
771 goto errout;
773 tb = fib_new_table(net, cfg.fc_table);
774 if (!tb) {
775 err = -ENOBUFS;
776 goto errout;
779 err = fib_table_insert(net, tb, &cfg, extack);
780 if (!err && cfg.fc_type == RTN_LOCAL)
781 net->ipv4.fib_has_custom_local_routes = true;
782 errout:
783 return err;
786 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
788 struct net *net = sock_net(skb->sk);
789 unsigned int h, s_h;
790 unsigned int e = 0, s_e;
791 struct fib_table *tb;
792 struct hlist_head *head;
793 int dumped = 0, err;
795 if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
796 ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
797 return skb->len;
799 s_h = cb->args[0];
800 s_e = cb->args[1];
802 rcu_read_lock();
804 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
805 e = 0;
806 head = &net->ipv4.fib_table_hash[h];
807 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
808 if (e < s_e)
809 goto next;
810 if (dumped)
811 memset(&cb->args[2], 0, sizeof(cb->args) -
812 2 * sizeof(cb->args[0]));
813 err = fib_table_dump(tb, skb, cb);
814 if (err < 0) {
815 if (likely(skb->len))
816 goto out;
818 goto out_err;
820 dumped = 1;
821 next:
822 e++;
825 out:
826 err = skb->len;
827 out_err:
828 rcu_read_unlock();
830 cb->args[1] = e;
831 cb->args[0] = h;
833 return err;
836 /* Prepare and feed intra-kernel routing request.
837 * Really, it should be netlink message, but :-( netlink
838 * can be not configured, so that we feed it directly
839 * to fib engine. It is legal, because all events occur
840 * only when netlink is already locked.
842 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
844 struct net *net = dev_net(ifa->ifa_dev->dev);
845 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
846 struct fib_table *tb;
847 struct fib_config cfg = {
848 .fc_protocol = RTPROT_KERNEL,
849 .fc_type = type,
850 .fc_dst = dst,
851 .fc_dst_len = dst_len,
852 .fc_prefsrc = ifa->ifa_local,
853 .fc_oif = ifa->ifa_dev->dev->ifindex,
854 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
855 .fc_nlinfo = {
856 .nl_net = net,
860 if (!tb_id)
861 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
863 tb = fib_new_table(net, tb_id);
864 if (!tb)
865 return;
867 cfg.fc_table = tb->tb_id;
869 if (type != RTN_LOCAL)
870 cfg.fc_scope = RT_SCOPE_LINK;
871 else
872 cfg.fc_scope = RT_SCOPE_HOST;
874 if (cmd == RTM_NEWROUTE)
875 fib_table_insert(net, tb, &cfg, NULL);
876 else
877 fib_table_delete(net, tb, &cfg, NULL);
880 void fib_add_ifaddr(struct in_ifaddr *ifa)
882 struct in_device *in_dev = ifa->ifa_dev;
883 struct net_device *dev = in_dev->dev;
884 struct in_ifaddr *prim = ifa;
885 __be32 mask = ifa->ifa_mask;
886 __be32 addr = ifa->ifa_local;
887 __be32 prefix = ifa->ifa_address & mask;
889 if (ifa->ifa_flags & IFA_F_SECONDARY) {
890 prim = inet_ifa_byprefix(in_dev, prefix, mask);
891 if (!prim) {
892 pr_warn("%s: bug: prim == NULL\n", __func__);
893 return;
897 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
899 if (!(dev->flags & IFF_UP))
900 return;
902 /* Add broadcast address, if it is explicitly assigned. */
903 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
904 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
906 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
907 (prefix != addr || ifa->ifa_prefixlen < 32)) {
908 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
909 fib_magic(RTM_NEWROUTE,
910 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
911 prefix, ifa->ifa_prefixlen, prim);
913 /* Add network specific broadcasts, when it takes a sense */
914 if (ifa->ifa_prefixlen < 31) {
915 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
916 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
917 32, prim);
922 /* Delete primary or secondary address.
923 * Optionally, on secondary address promotion consider the addresses
924 * from subnet iprim as deleted, even if they are in device list.
925 * In this case the secondary ifa can be in device list.
927 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
929 struct in_device *in_dev = ifa->ifa_dev;
930 struct net_device *dev = in_dev->dev;
931 struct in_ifaddr *ifa1;
932 struct in_ifaddr *prim = ifa, *prim1 = NULL;
933 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
934 __be32 any = ifa->ifa_address & ifa->ifa_mask;
935 #define LOCAL_OK 1
936 #define BRD_OK 2
937 #define BRD0_OK 4
938 #define BRD1_OK 8
939 unsigned int ok = 0;
940 int subnet = 0; /* Primary network */
941 int gone = 1; /* Address is missing */
942 int same_prefsrc = 0; /* Another primary with same IP */
944 if (ifa->ifa_flags & IFA_F_SECONDARY) {
945 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
946 if (!prim) {
947 /* if the device has been deleted, we don't perform
948 * address promotion
950 if (!in_dev->dead)
951 pr_warn("%s: bug: prim == NULL\n", __func__);
952 return;
954 if (iprim && iprim != prim) {
955 pr_warn("%s: bug: iprim != prim\n", __func__);
956 return;
958 } else if (!ipv4_is_zeronet(any) &&
959 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
960 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
961 fib_magic(RTM_DELROUTE,
962 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
963 any, ifa->ifa_prefixlen, prim);
964 subnet = 1;
967 if (in_dev->dead)
968 goto no_promotions;
970 /* Deletion is more complicated than add.
971 * We should take care of not to delete too much :-)
973 * Scan address list to be sure that addresses are really gone.
976 for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
977 if (ifa1 == ifa) {
978 /* promotion, keep the IP */
979 gone = 0;
980 continue;
982 /* Ignore IFAs from our subnet */
983 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
984 inet_ifa_match(ifa1->ifa_address, iprim))
985 continue;
987 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
988 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
989 /* Another address from our subnet? */
990 if (ifa1->ifa_mask == prim->ifa_mask &&
991 inet_ifa_match(ifa1->ifa_address, prim))
992 prim1 = prim;
993 else {
994 /* We reached the secondaries, so
995 * same_prefsrc should be determined.
997 if (!same_prefsrc)
998 continue;
999 /* Search new prim1 if ifa1 is not
1000 * using the current prim1
1002 if (!prim1 ||
1003 ifa1->ifa_mask != prim1->ifa_mask ||
1004 !inet_ifa_match(ifa1->ifa_address, prim1))
1005 prim1 = inet_ifa_byprefix(in_dev,
1006 ifa1->ifa_address,
1007 ifa1->ifa_mask);
1008 if (!prim1)
1009 continue;
1010 if (prim1->ifa_local != prim->ifa_local)
1011 continue;
1013 } else {
1014 if (prim->ifa_local != ifa1->ifa_local)
1015 continue;
1016 prim1 = ifa1;
1017 if (prim != prim1)
1018 same_prefsrc = 1;
1020 if (ifa->ifa_local == ifa1->ifa_local)
1021 ok |= LOCAL_OK;
1022 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1023 ok |= BRD_OK;
1024 if (brd == ifa1->ifa_broadcast)
1025 ok |= BRD1_OK;
1026 if (any == ifa1->ifa_broadcast)
1027 ok |= BRD0_OK;
1028 /* primary has network specific broadcasts */
1029 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1030 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1031 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1033 if (!ipv4_is_zeronet(any1)) {
1034 if (ifa->ifa_broadcast == brd1 ||
1035 ifa->ifa_broadcast == any1)
1036 ok |= BRD_OK;
1037 if (brd == brd1 || brd == any1)
1038 ok |= BRD1_OK;
1039 if (any == brd1 || any == any1)
1040 ok |= BRD0_OK;
1045 no_promotions:
1046 if (!(ok & BRD_OK))
1047 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1048 if (subnet && ifa->ifa_prefixlen < 31) {
1049 if (!(ok & BRD1_OK))
1050 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1051 if (!(ok & BRD0_OK))
1052 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1054 if (!(ok & LOCAL_OK)) {
1055 unsigned int addr_type;
1057 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1059 /* Check, that this local address finally disappeared. */
1060 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1061 ifa->ifa_local);
1062 if (gone && addr_type != RTN_LOCAL) {
1063 /* And the last, but not the least thing.
1064 * We must flush stray FIB entries.
1066 * First of all, we scan fib_info list searching
1067 * for stray nexthop entries, then ignite fib_flush.
1069 if (fib_sync_down_addr(dev, ifa->ifa_local))
1070 fib_flush(dev_net(dev));
1073 #undef LOCAL_OK
1074 #undef BRD_OK
1075 #undef BRD0_OK
1076 #undef BRD1_OK
1079 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1082 struct fib_result res;
1083 struct flowi4 fl4 = {
1084 .flowi4_mark = frn->fl_mark,
1085 .daddr = frn->fl_addr,
1086 .flowi4_tos = frn->fl_tos,
1087 .flowi4_scope = frn->fl_scope,
1089 struct fib_table *tb;
1091 rcu_read_lock();
1093 tb = fib_get_table(net, frn->tb_id_in);
1095 frn->err = -ENOENT;
1096 if (tb) {
1097 local_bh_disable();
1099 frn->tb_id = tb->tb_id;
1100 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1102 if (!frn->err) {
1103 frn->prefixlen = res.prefixlen;
1104 frn->nh_sel = res.nh_sel;
1105 frn->type = res.type;
1106 frn->scope = res.scope;
1108 local_bh_enable();
1111 rcu_read_unlock();
1114 static void nl_fib_input(struct sk_buff *skb)
1116 struct net *net;
1117 struct fib_result_nl *frn;
1118 struct nlmsghdr *nlh;
1119 u32 portid;
1121 net = sock_net(skb->sk);
1122 nlh = nlmsg_hdr(skb);
1123 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1124 skb->len < nlh->nlmsg_len ||
1125 nlmsg_len(nlh) < sizeof(*frn))
1126 return;
1128 skb = netlink_skb_clone(skb, GFP_KERNEL);
1129 if (!skb)
1130 return;
1131 nlh = nlmsg_hdr(skb);
1133 frn = (struct fib_result_nl *) nlmsg_data(nlh);
1134 nl_fib_lookup(net, frn);
1136 portid = NETLINK_CB(skb).portid; /* netlink portid */
1137 NETLINK_CB(skb).portid = 0; /* from kernel */
1138 NETLINK_CB(skb).dst_group = 0; /* unicast */
1139 netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1142 static int __net_init nl_fib_lookup_init(struct net *net)
1144 struct sock *sk;
1145 struct netlink_kernel_cfg cfg = {
1146 .input = nl_fib_input,
1149 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1150 if (!sk)
1151 return -EAFNOSUPPORT;
1152 net->ipv4.fibnl = sk;
1153 return 0;
1156 static void nl_fib_lookup_exit(struct net *net)
1158 netlink_kernel_release(net->ipv4.fibnl);
1159 net->ipv4.fibnl = NULL;
1162 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1163 bool force)
1165 if (fib_sync_down_dev(dev, event, force))
1166 fib_flush(dev_net(dev));
1167 else
1168 rt_cache_flush(dev_net(dev));
1169 arp_ifdown(dev);
1172 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1174 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1175 struct net_device *dev = ifa->ifa_dev->dev;
1176 struct net *net = dev_net(dev);
1178 switch (event) {
1179 case NETDEV_UP:
1180 fib_add_ifaddr(ifa);
1181 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1182 fib_sync_up(dev, RTNH_F_DEAD);
1183 #endif
1184 atomic_inc(&net->ipv4.dev_addr_genid);
1185 rt_cache_flush(dev_net(dev));
1186 break;
1187 case NETDEV_DOWN:
1188 fib_del_ifaddr(ifa, NULL);
1189 atomic_inc(&net->ipv4.dev_addr_genid);
1190 if (!ifa->ifa_dev->ifa_list) {
1191 /* Last address was deleted from this interface.
1192 * Disable IP.
1194 fib_disable_ip(dev, event, true);
1195 } else {
1196 rt_cache_flush(dev_net(dev));
1198 break;
1200 return NOTIFY_DONE;
1203 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1205 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1206 struct netdev_notifier_changeupper_info *info;
1207 struct in_device *in_dev;
1208 struct net *net = dev_net(dev);
1209 unsigned int flags;
1211 if (event == NETDEV_UNREGISTER) {
1212 fib_disable_ip(dev, event, true);
1213 rt_flush_dev(dev);
1214 return NOTIFY_DONE;
1217 in_dev = __in_dev_get_rtnl(dev);
1218 if (!in_dev)
1219 return NOTIFY_DONE;
1221 switch (event) {
1222 case NETDEV_UP:
1223 for_ifa(in_dev) {
1224 fib_add_ifaddr(ifa);
1225 } endfor_ifa(in_dev);
1226 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1227 fib_sync_up(dev, RTNH_F_DEAD);
1228 #endif
1229 atomic_inc(&net->ipv4.dev_addr_genid);
1230 rt_cache_flush(net);
1231 break;
1232 case NETDEV_DOWN:
1233 fib_disable_ip(dev, event, false);
1234 break;
1235 case NETDEV_CHANGE:
1236 flags = dev_get_flags(dev);
1237 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1238 fib_sync_up(dev, RTNH_F_LINKDOWN);
1239 else
1240 fib_sync_down_dev(dev, event, false);
1241 /* fall through */
1242 case NETDEV_CHANGEMTU:
1243 rt_cache_flush(net);
1244 break;
1245 case NETDEV_CHANGEUPPER:
1246 info = ptr;
1247 /* flush all routes if dev is linked to or unlinked from
1248 * an L3 master device (e.g., VRF)
1250 if (info->upper_dev && netif_is_l3_master(info->upper_dev))
1251 fib_disable_ip(dev, NETDEV_DOWN, true);
1252 break;
1254 return NOTIFY_DONE;
1257 static struct notifier_block fib_inetaddr_notifier = {
1258 .notifier_call = fib_inetaddr_event,
1261 static struct notifier_block fib_netdev_notifier = {
1262 .notifier_call = fib_netdev_event,
1265 static int __net_init ip_fib_net_init(struct net *net)
1267 int err;
1268 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1270 err = fib4_notifier_init(net);
1271 if (err)
1272 return err;
1274 /* Avoid false sharing : Use at least a full cache line */
1275 size = max_t(size_t, size, L1_CACHE_BYTES);
1277 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1278 if (!net->ipv4.fib_table_hash) {
1279 err = -ENOMEM;
1280 goto err_table_hash_alloc;
1283 err = fib4_rules_init(net);
1284 if (err < 0)
1285 goto err_rules_init;
1286 return 0;
1288 err_rules_init:
1289 kfree(net->ipv4.fib_table_hash);
1290 err_table_hash_alloc:
1291 fib4_notifier_exit(net);
1292 return err;
1295 static void ip_fib_net_exit(struct net *net)
1297 int i;
1299 rtnl_lock();
1300 #ifdef CONFIG_IP_MULTIPLE_TABLES
1301 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1302 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1303 #endif
1304 /* Destroy the tables in reverse order to guarantee that the
1305 * local table, ID 255, is destroyed before the main table, ID
1306 * 254. This is necessary as the local table may contain
1307 * references to data contained in the main table.
1309 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1310 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1311 struct hlist_node *tmp;
1312 struct fib_table *tb;
1314 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1315 hlist_del(&tb->tb_hlist);
1316 fib_table_flush(net, tb);
1317 fib_free_table(tb);
1321 #ifdef CONFIG_IP_MULTIPLE_TABLES
1322 fib4_rules_exit(net);
1323 #endif
1324 rtnl_unlock();
1325 kfree(net->ipv4.fib_table_hash);
1326 fib4_notifier_exit(net);
1329 static int __net_init fib_net_init(struct net *net)
1331 int error;
1333 #ifdef CONFIG_IP_ROUTE_CLASSID
1334 net->ipv4.fib_num_tclassid_users = 0;
1335 #endif
1336 error = ip_fib_net_init(net);
1337 if (error < 0)
1338 goto out;
1339 error = nl_fib_lookup_init(net);
1340 if (error < 0)
1341 goto out_nlfl;
1342 error = fib_proc_init(net);
1343 if (error < 0)
1344 goto out_proc;
1345 out:
1346 return error;
1348 out_proc:
1349 nl_fib_lookup_exit(net);
1350 out_nlfl:
1351 ip_fib_net_exit(net);
1352 goto out;
1355 static void __net_exit fib_net_exit(struct net *net)
1357 fib_proc_exit(net);
1358 nl_fib_lookup_exit(net);
1359 ip_fib_net_exit(net);
1362 static struct pernet_operations fib_net_ops = {
1363 .init = fib_net_init,
1364 .exit = fib_net_exit,
1367 void __init ip_fib_init(void)
1369 fib_trie_init();
1371 register_pernet_subsys(&fib_net_ops);
1373 register_netdevice_notifier(&fib_netdev_notifier);
1374 register_inetaddr_notifier(&fib_inetaddr_notifier);
1376 rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
1377 rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
1378 rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, 0);