OSDN Git Service

Merge 4.4.187 into android-4.4
[sagit-ice-cold/kernel_xiaomi_msm8998.git] / net / ipv4 / fib_frontend.c
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.
5  *
6  *              IPv4 Forwarding Information Base: FIB frontend.
7  *
8  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
9  *
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.
14  */
15
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>
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>
38
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 <trace/events/fib.h>
50
51 #ifndef CONFIG_IP_MULTIPLE_TABLES
52
53 static int __net_init fib4_rules_init(struct net *net)
54 {
55         struct fib_table *local_table, *main_table;
56
57         main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
58         if (!main_table)
59                 return -ENOMEM;
60
61         local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
62         if (!local_table)
63                 goto fail;
64
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]);
69         return 0;
70
71 fail:
72         fib_free_table(main_table);
73         return -ENOMEM;
74 }
75 #else
76
77 struct fib_table *fib_new_table(struct net *net, u32 id)
78 {
79         struct fib_table *tb, *alias = NULL;
80         unsigned int h;
81
82         if (id == 0)
83                 id = RT_TABLE_MAIN;
84         tb = fib_get_table(net, id);
85         if (tb)
86                 return tb;
87
88         if (id == RT_TABLE_LOCAL && !net->ipv4.fib_has_custom_rules)
89                 alias = fib_new_table(net, RT_TABLE_MAIN);
90
91         tb = fib_trie_table(id, alias);
92         if (!tb)
93                 return NULL;
94
95         switch (id) {
96         case RT_TABLE_LOCAL:
97                 rcu_assign_pointer(net->ipv4.fib_local, tb);
98                 break;
99         case RT_TABLE_MAIN:
100                 rcu_assign_pointer(net->ipv4.fib_main, tb);
101                 break;
102         case RT_TABLE_DEFAULT:
103                 rcu_assign_pointer(net->ipv4.fib_default, tb);
104                 break;
105         default:
106                 break;
107         }
108
109         h = id & (FIB_TABLE_HASHSZ - 1);
110         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
111         return tb;
112 }
113
114 /* caller must hold either rtnl or rcu read lock */
115 struct fib_table *fib_get_table(struct net *net, u32 id)
116 {
117         struct fib_table *tb;
118         struct hlist_head *head;
119         unsigned int h;
120
121         if (id == 0)
122                 id = RT_TABLE_MAIN;
123         h = id & (FIB_TABLE_HASHSZ - 1);
124
125         head = &net->ipv4.fib_table_hash[h];
126         hlist_for_each_entry_rcu(tb, head, tb_hlist) {
127                 if (tb->tb_id == id)
128                         return tb;
129         }
130         return NULL;
131 }
132 #endif /* CONFIG_IP_MULTIPLE_TABLES */
133
134 static void fib_replace_table(struct net *net, struct fib_table *old,
135                               struct fib_table *new)
136 {
137 #ifdef CONFIG_IP_MULTIPLE_TABLES
138         switch (new->tb_id) {
139         case RT_TABLE_LOCAL:
140                 rcu_assign_pointer(net->ipv4.fib_local, new);
141                 break;
142         case RT_TABLE_MAIN:
143                 rcu_assign_pointer(net->ipv4.fib_main, new);
144                 break;
145         case RT_TABLE_DEFAULT:
146                 rcu_assign_pointer(net->ipv4.fib_default, new);
147                 break;
148         default:
149                 break;
150         }
151
152 #endif
153         /* replace the old table in the hlist */
154         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
155 }
156
157 int fib_unmerge(struct net *net)
158 {
159         struct fib_table *old, *new;
160
161         /* attempt to fetch local table if it has been allocated */
162         old = fib_get_table(net, RT_TABLE_LOCAL);
163         if (!old)
164                 return 0;
165
166         new = fib_trie_unmerge(old);
167         if (!new)
168                 return -ENOMEM;
169
170         /* replace merged table with clean table */
171         if (new != old) {
172                 fib_replace_table(net, old, new);
173                 fib_free_table(old);
174         }
175
176         return 0;
177 }
178
179 static void fib_flush(struct net *net)
180 {
181         int flushed = 0;
182         unsigned int h;
183
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;
188
189                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
190                         flushed += fib_table_flush(tb, false);
191         }
192
193         if (flushed)
194                 rt_cache_flush(net);
195 }
196
197 void fib_flush_external(struct net *net)
198 {
199         struct fib_table *tb;
200         struct hlist_head *head;
201         unsigned int h;
202
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);
207         }
208 }
209
210 /*
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.
213  */
214 static inline unsigned int __inet_dev_addr_type(struct net *net,
215                                                 const struct net_device *dev,
216                                                 __be32 addr, u32 tb_id)
217 {
218         struct flowi4           fl4 = { .daddr = addr };
219         struct fib_result       res;
220         unsigned int ret = RTN_BROADCAST;
221         struct fib_table *table;
222
223         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
224                 return RTN_BROADCAST;
225         if (ipv4_is_multicast(addr))
226                 return RTN_MULTICAST;
227
228         rcu_read_lock();
229
230         table = fib_get_table(net, tb_id);
231         if (table) {
232                 ret = RTN_UNICAST;
233                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
234                         if (!dev || dev == res.fi->fib_dev)
235                                 ret = res.type;
236                 }
237         }
238
239         rcu_read_unlock();
240         return ret;
241 }
242
243 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
244 {
245         return __inet_dev_addr_type(net, NULL, addr, tb_id);
246 }
247 EXPORT_SYMBOL(inet_addr_type_table);
248
249 unsigned int inet_addr_type(struct net *net, __be32 addr)
250 {
251         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
252 }
253 EXPORT_SYMBOL(inet_addr_type);
254
255 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
256                                 __be32 addr)
257 {
258         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
259
260         return __inet_dev_addr_type(net, dev, addr, rt_table);
261 }
262 EXPORT_SYMBOL(inet_dev_addr_type);
263
264 /* inet_addr_type with dev == NULL but using the table from a dev
265  * if one is associated
266  */
267 unsigned int inet_addr_type_dev_table(struct net *net,
268                                       const struct net_device *dev,
269                                       __be32 addr)
270 {
271         u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
272
273         return __inet_dev_addr_type(net, NULL, addr, rt_table);
274 }
275 EXPORT_SYMBOL(inet_addr_type_dev_table);
276
277 __be32 fib_compute_spec_dst(struct sk_buff *skb)
278 {
279         struct net_device *dev = skb->dev;
280         struct in_device *in_dev;
281         struct fib_result res;
282         struct rtable *rt;
283         struct net *net;
284         int scope;
285
286         rt = skb_rtable(skb);
287         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
288             RTCF_LOCAL)
289                 return ip_hdr(skb)->daddr;
290
291         in_dev = __in_dev_get_rcu(dev);
292
293         net = dev_net(dev);
294
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,
305                 };
306                 if (!fib_lookup(net, &fl4, &res, 0))
307                         return FIB_RES_PREFSRC(net, res);
308         } else {
309                 scope = RT_SCOPE_LINK;
310         }
311
312         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
313 }
314
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
317  *   address.
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()
322  */
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)
326 {
327         int ret, no_addr;
328         struct fib_result res;
329         struct flowi4 fl4;
330         struct net *net;
331         bool dev_match;
332
333         fl4.flowi4_oif = 0;
334         fl4.flowi4_iif = l3mdev_master_ifindex_rcu(dev);
335         if (!fl4.flowi4_iif)
336                 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
337         fl4.daddr = src;
338         fl4.saddr = dst;
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;
343
344         no_addr = idev->ifa_list == NULL;
345
346         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
347
348         trace_fib_validate_source(dev, &fl4);
349
350         net = dev_net(dev);
351         if (fib_lookup(net, &fl4, &res, 0))
352                 goto last_resort;
353         if (res.type != RTN_UNICAST &&
354             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
355                 goto e_inval;
356         if (!rpf && !fib_num_tclassid_users(dev_net(dev)) &&
357             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
358                 goto last_resort;
359         fib_combine_itag(itag, &res);
360         dev_match = false;
361
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];
365
366                 if (nh->nh_dev == dev) {
367                         dev_match = true;
368                         break;
369                 } else if (l3mdev_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
370                         dev_match = true;
371                         break;
372                 }
373         }
374 #else
375         if (FIB_RES_DEV(res) == dev)
376                 dev_match = true;
377 #endif
378         if (dev_match) {
379                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
380                 return ret;
381         }
382         if (no_addr)
383                 goto last_resort;
384         if (rpf == 1)
385                 goto e_rpf;
386         fl4.flowi4_oif = dev->ifindex;
387
388         ret = 0;
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;
392         }
393         return ret;
394
395 last_resort:
396         if (rpf)
397                 goto e_rpf;
398         *itag = 0;
399         return 0;
400
401 e_inval:
402         return -EINVAL;
403 e_rpf:
404         return -EXDEV;
405 }
406
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)
411 {
412         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
413
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))) {
417                 *itag = 0;
418                 return 0;
419         }
420         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
421 }
422
423 static inline __be32 sk_extract_addr(struct sockaddr *addr)
424 {
425         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
426 }
427
428 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
429 {
430         struct nlattr *nla;
431
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;
436
437         return len + nla_total_size(4);
438 }
439
440 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
441                                  struct fib_config *cfg)
442 {
443         __be32 addr;
444         int plen;
445
446         memset(cfg, 0, sizeof(*cfg));
447         cfg->fc_nlinfo.nl_net = net;
448
449         if (rt->rt_dst.sa_family != AF_INET)
450                 return -EAFNOSUPPORT;
451
452         /*
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.
459          */
460         plen = 32;
461         addr = sk_extract_addr(&rt->rt_dst);
462         if (!(rt->rt_flags & RTF_HOST)) {
463                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
464
465                 if (rt->rt_genmask.sa_family != AF_INET) {
466                         if (mask || rt->rt_genmask.sa_family)
467                                 return -EAFNOSUPPORT;
468                 }
469
470                 if (bad_mask(mask, addr))
471                         return -EINVAL;
472
473                 plen = inet_mask_len(mask);
474         }
475
476         cfg->fc_dst_len = plen;
477         cfg->fc_dst = addr;
478
479         if (cmd != SIOCDELRT) {
480                 cfg->fc_nlflags = NLM_F_CREATE;
481                 cfg->fc_protocol = RTPROT_BOOT;
482         }
483
484         if (rt->rt_metric)
485                 cfg->fc_priority = rt->rt_metric - 1;
486
487         if (rt->rt_flags & RTF_REJECT) {
488                 cfg->fc_scope = RT_SCOPE_HOST;
489                 cfg->fc_type = RTN_UNREACHABLE;
490                 return 0;
491         }
492
493         cfg->fc_scope = RT_SCOPE_NOWHERE;
494         cfg->fc_type = RTN_UNICAST;
495
496         if (rt->rt_dev) {
497                 char *colon;
498                 struct net_device *dev;
499                 char devname[IFNAMSIZ];
500
501                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
502                         return -EFAULT;
503
504                 devname[IFNAMSIZ-1] = 0;
505                 colon = strchr(devname, ':');
506                 if (colon)
507                         *colon = 0;
508                 dev = __dev_get_by_name(net, devname);
509                 if (!dev)
510                         return -ENODEV;
511                 cfg->fc_oif = dev->ifindex;
512                 if (colon) {
513                         struct in_ifaddr *ifa;
514                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
515                         if (!in_dev)
516                                 return -ENODEV;
517                         *colon = ':';
518                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
519                                 if (strcmp(ifa->ifa_label, devname) == 0)
520                                         break;
521                         if (!ifa)
522                                 return -ENODEV;
523                         cfg->fc_prefsrc = ifa->ifa_local;
524                 }
525         }
526
527         addr = sk_extract_addr(&rt->rt_gateway);
528         if (rt->rt_gateway.sa_family == AF_INET && addr) {
529                 unsigned int addr_type;
530
531                 cfg->fc_gw = addr;
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;
536         }
537
538         if (cmd == SIOCDELRT)
539                 return 0;
540
541         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
542                 return -EINVAL;
543
544         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
545                 cfg->fc_scope = RT_SCOPE_LINK;
546
547         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
548                 struct nlattr *mx;
549                 int len = 0;
550
551                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
552                 if (!mx)
553                         return -ENOMEM;
554
555                 if (rt->rt_flags & RTF_MTU)
556                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
557
558                 if (rt->rt_flags & RTF_WINDOW)
559                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
560
561                 if (rt->rt_flags & RTF_IRTT)
562                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
563
564                 cfg->fc_mx = mx;
565                 cfg->fc_mx_len = len;
566         }
567
568         return 0;
569 }
570
571 /*
572  * Handle IP routing ioctl calls.
573  * These are used to manipulate the routing tables
574  */
575 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
576 {
577         struct fib_config cfg;
578         struct rtentry rt;
579         int err;
580
581         switch (cmd) {
582         case SIOCADDRT:         /* Add a route */
583         case SIOCDELRT:         /* Delete a route */
584                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
585                         return -EPERM;
586
587                 if (copy_from_user(&rt, arg, sizeof(rt)))
588                         return -EFAULT;
589
590                 rtnl_lock();
591                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
592                 if (err == 0) {
593                         struct fib_table *tb;
594
595                         if (cmd == SIOCDELRT) {
596                                 tb = fib_get_table(net, cfg.fc_table);
597                                 if (tb)
598                                         err = fib_table_delete(tb, &cfg);
599                                 else
600                                         err = -ESRCH;
601                         } else {
602                                 tb = fib_new_table(net, cfg.fc_table);
603                                 if (tb)
604                                         err = fib_table_insert(tb, &cfg);
605                                 else
606                                         err = -ENOBUFS;
607                         }
608
609                         /* allocated by rtentry_to_fib_config() */
610                         kfree(cfg.fc_mx);
611                 }
612                 rtnl_unlock();
613                 return err;
614         }
615         return -EINVAL;
616 }
617
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 },
631         [RTA_UID]               = { .type = NLA_U32 },
632 };
633
634 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
635                              struct nlmsghdr *nlh, struct fib_config *cfg)
636 {
637         struct nlattr *attr;
638         int err, remaining;
639         struct rtmsg *rtm;
640
641         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
642         if (err < 0)
643                 goto errout;
644
645         memset(cfg, 0, sizeof(*cfg));
646
647         rtm = nlmsg_data(nlh);
648         cfg->fc_dst_len = rtm->rtm_dst_len;
649         cfg->fc_tos = rtm->rtm_tos;
650         cfg->fc_table = rtm->rtm_table;
651         cfg->fc_protocol = rtm->rtm_protocol;
652         cfg->fc_scope = rtm->rtm_scope;
653         cfg->fc_type = rtm->rtm_type;
654         cfg->fc_flags = rtm->rtm_flags;
655         cfg->fc_nlflags = nlh->nlmsg_flags;
656
657         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
658         cfg->fc_nlinfo.nlh = nlh;
659         cfg->fc_nlinfo.nl_net = net;
660
661         if (cfg->fc_type > RTN_MAX) {
662                 err = -EINVAL;
663                 goto errout;
664         }
665
666         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
667                 switch (nla_type(attr)) {
668                 case RTA_DST:
669                         cfg->fc_dst = nla_get_be32(attr);
670                         break;
671                 case RTA_OIF:
672                         cfg->fc_oif = nla_get_u32(attr);
673                         break;
674                 case RTA_GATEWAY:
675                         cfg->fc_gw = nla_get_be32(attr);
676                         break;
677                 case RTA_PRIORITY:
678                         cfg->fc_priority = nla_get_u32(attr);
679                         break;
680                 case RTA_PREFSRC:
681                         cfg->fc_prefsrc = nla_get_be32(attr);
682                         break;
683                 case RTA_METRICS:
684                         cfg->fc_mx = nla_data(attr);
685                         cfg->fc_mx_len = nla_len(attr);
686                         break;
687                 case RTA_MULTIPATH:
688                         cfg->fc_mp = nla_data(attr);
689                         cfg->fc_mp_len = nla_len(attr);
690                         break;
691                 case RTA_FLOW:
692                         cfg->fc_flow = nla_get_u32(attr);
693                         break;
694                 case RTA_TABLE:
695                         cfg->fc_table = nla_get_u32(attr);
696                         break;
697                 case RTA_ENCAP:
698                         cfg->fc_encap = attr;
699                         break;
700                 case RTA_ENCAP_TYPE:
701                         cfg->fc_encap_type = nla_get_u16(attr);
702                         break;
703                 }
704         }
705
706         return 0;
707 errout:
708         return err;
709 }
710
711 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
712 {
713         struct net *net = sock_net(skb->sk);
714         struct fib_config cfg;
715         struct fib_table *tb;
716         int err;
717
718         err = rtm_to_fib_config(net, skb, nlh, &cfg);
719         if (err < 0)
720                 goto errout;
721
722         tb = fib_get_table(net, cfg.fc_table);
723         if (!tb) {
724                 err = -ESRCH;
725                 goto errout;
726         }
727
728         err = fib_table_delete(tb, &cfg);
729 errout:
730         return err;
731 }
732
733 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
734 {
735         struct net *net = sock_net(skb->sk);
736         struct fib_config cfg;
737         struct fib_table *tb;
738         int err;
739
740         err = rtm_to_fib_config(net, skb, nlh, &cfg);
741         if (err < 0)
742                 goto errout;
743
744         tb = fib_new_table(net, cfg.fc_table);
745         if (!tb) {
746                 err = -ENOBUFS;
747                 goto errout;
748         }
749
750         err = fib_table_insert(tb, &cfg);
751 errout:
752         return err;
753 }
754
755 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
756 {
757         struct net *net = sock_net(skb->sk);
758         unsigned int h, s_h;
759         unsigned int e = 0, s_e;
760         struct fib_table *tb;
761         struct hlist_head *head;
762         int dumped = 0, err;
763
764         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
765             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
766                 return skb->len;
767
768         s_h = cb->args[0];
769         s_e = cb->args[1];
770
771         rcu_read_lock();
772
773         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
774                 e = 0;
775                 head = &net->ipv4.fib_table_hash[h];
776                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
777                         if (e < s_e)
778                                 goto next;
779                         if (dumped)
780                                 memset(&cb->args[2], 0, sizeof(cb->args) -
781                                                  2 * sizeof(cb->args[0]));
782                         err = fib_table_dump(tb, skb, cb);
783                         if (err < 0) {
784                                 if (likely(skb->len))
785                                         goto out;
786
787                                 goto out_err;
788                         }
789                         dumped = 1;
790 next:
791                         e++;
792                 }
793         }
794 out:
795         err = skb->len;
796 out_err:
797         rcu_read_unlock();
798
799         cb->args[1] = e;
800         cb->args[0] = h;
801
802         return err;
803 }
804
805 /* Prepare and feed intra-kernel routing request.
806  * Really, it should be netlink message, but :-( netlink
807  * can be not configured, so that we feed it directly
808  * to fib engine. It is legal, because all events occur
809  * only when netlink is already locked.
810  */
811 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
812 {
813         struct net *net = dev_net(ifa->ifa_dev->dev);
814         u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
815         struct fib_table *tb;
816         struct fib_config cfg = {
817                 .fc_protocol = RTPROT_KERNEL,
818                 .fc_type = type,
819                 .fc_dst = dst,
820                 .fc_dst_len = dst_len,
821                 .fc_prefsrc = ifa->ifa_local,
822                 .fc_oif = ifa->ifa_dev->dev->ifindex,
823                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
824                 .fc_nlinfo = {
825                         .nl_net = net,
826                 },
827         };
828
829         if (!tb_id)
830                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
831
832         tb = fib_new_table(net, tb_id);
833         if (!tb)
834                 return;
835
836         cfg.fc_table = tb->tb_id;
837
838         if (type != RTN_LOCAL)
839                 cfg.fc_scope = RT_SCOPE_LINK;
840         else
841                 cfg.fc_scope = RT_SCOPE_HOST;
842
843         if (cmd == RTM_NEWROUTE)
844                 fib_table_insert(tb, &cfg);
845         else
846                 fib_table_delete(tb, &cfg);
847 }
848
849 void fib_add_ifaddr(struct in_ifaddr *ifa)
850 {
851         struct in_device *in_dev = ifa->ifa_dev;
852         struct net_device *dev = in_dev->dev;
853         struct in_ifaddr *prim = ifa;
854         __be32 mask = ifa->ifa_mask;
855         __be32 addr = ifa->ifa_local;
856         __be32 prefix = ifa->ifa_address & mask;
857
858         if (ifa->ifa_flags & IFA_F_SECONDARY) {
859                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
860                 if (!prim) {
861                         pr_warn("%s: bug: prim == NULL\n", __func__);
862                         return;
863                 }
864         }
865
866         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
867
868         if (!(dev->flags & IFF_UP))
869                 return;
870
871         /* Add broadcast address, if it is explicitly assigned. */
872         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
873                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
874
875         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
876             (prefix != addr || ifa->ifa_prefixlen < 32)) {
877                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
878                         fib_magic(RTM_NEWROUTE,
879                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
880                                   prefix, ifa->ifa_prefixlen, prim);
881
882                 /* Add network specific broadcasts, when it takes a sense */
883                 if (ifa->ifa_prefixlen < 31) {
884                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
885                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
886                                   32, prim);
887                 }
888         }
889 }
890
891 /* Delete primary or secondary address.
892  * Optionally, on secondary address promotion consider the addresses
893  * from subnet iprim as deleted, even if they are in device list.
894  * In this case the secondary ifa can be in device list.
895  */
896 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
897 {
898         struct in_device *in_dev = ifa->ifa_dev;
899         struct net_device *dev = in_dev->dev;
900         struct in_ifaddr *ifa1;
901         struct in_ifaddr *prim = ifa, *prim1 = NULL;
902         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
903         __be32 any = ifa->ifa_address & ifa->ifa_mask;
904 #define LOCAL_OK        1
905 #define BRD_OK          2
906 #define BRD0_OK         4
907 #define BRD1_OK         8
908         unsigned int ok = 0;
909         int subnet = 0;         /* Primary network */
910         int gone = 1;           /* Address is missing */
911         int same_prefsrc = 0;   /* Another primary with same IP */
912
913         if (ifa->ifa_flags & IFA_F_SECONDARY) {
914                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
915                 if (!prim) {
916                         /* if the device has been deleted, we don't perform
917                          * address promotion
918                          */
919                         if (!in_dev->dead)
920                                 pr_warn("%s: bug: prim == NULL\n", __func__);
921                         return;
922                 }
923                 if (iprim && iprim != prim) {
924                         pr_warn("%s: bug: iprim != prim\n", __func__);
925                         return;
926                 }
927         } else if (!ipv4_is_zeronet(any) &&
928                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
929                 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
930                         fib_magic(RTM_DELROUTE,
931                                   dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
932                                   any, ifa->ifa_prefixlen, prim);
933                 subnet = 1;
934         }
935
936         if (in_dev->dead)
937                 goto no_promotions;
938
939         /* Deletion is more complicated than add.
940          * We should take care of not to delete too much :-)
941          *
942          * Scan address list to be sure that addresses are really gone.
943          */
944
945         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
946                 if (ifa1 == ifa) {
947                         /* promotion, keep the IP */
948                         gone = 0;
949                         continue;
950                 }
951                 /* Ignore IFAs from our subnet */
952                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
953                     inet_ifa_match(ifa1->ifa_address, iprim))
954                         continue;
955
956                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
957                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
958                         /* Another address from our subnet? */
959                         if (ifa1->ifa_mask == prim->ifa_mask &&
960                             inet_ifa_match(ifa1->ifa_address, prim))
961                                 prim1 = prim;
962                         else {
963                                 /* We reached the secondaries, so
964                                  * same_prefsrc should be determined.
965                                  */
966                                 if (!same_prefsrc)
967                                         continue;
968                                 /* Search new prim1 if ifa1 is not
969                                  * using the current prim1
970                                  */
971                                 if (!prim1 ||
972                                     ifa1->ifa_mask != prim1->ifa_mask ||
973                                     !inet_ifa_match(ifa1->ifa_address, prim1))
974                                         prim1 = inet_ifa_byprefix(in_dev,
975                                                         ifa1->ifa_address,
976                                                         ifa1->ifa_mask);
977                                 if (!prim1)
978                                         continue;
979                                 if (prim1->ifa_local != prim->ifa_local)
980                                         continue;
981                         }
982                 } else {
983                         if (prim->ifa_local != ifa1->ifa_local)
984                                 continue;
985                         prim1 = ifa1;
986                         if (prim != prim1)
987                                 same_prefsrc = 1;
988                 }
989                 if (ifa->ifa_local == ifa1->ifa_local)
990                         ok |= LOCAL_OK;
991                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
992                         ok |= BRD_OK;
993                 if (brd == ifa1->ifa_broadcast)
994                         ok |= BRD1_OK;
995                 if (any == ifa1->ifa_broadcast)
996                         ok |= BRD0_OK;
997                 /* primary has network specific broadcasts */
998                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
999                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1000                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1001
1002                         if (!ipv4_is_zeronet(any1)) {
1003                                 if (ifa->ifa_broadcast == brd1 ||
1004                                     ifa->ifa_broadcast == any1)
1005                                         ok |= BRD_OK;
1006                                 if (brd == brd1 || brd == any1)
1007                                         ok |= BRD1_OK;
1008                                 if (any == brd1 || any == any1)
1009                                         ok |= BRD0_OK;
1010                         }
1011                 }
1012         }
1013
1014 no_promotions:
1015         if (!(ok & BRD_OK))
1016                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
1017         if (subnet && ifa->ifa_prefixlen < 31) {
1018                 if (!(ok & BRD1_OK))
1019                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
1020                 if (!(ok & BRD0_OK))
1021                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1022         }
1023         if (!(ok & LOCAL_OK)) {
1024                 unsigned int addr_type;
1025
1026                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1027
1028                 /* Check, that this local address finally disappeared. */
1029                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1030                                                      ifa->ifa_local);
1031                 if (gone && addr_type != RTN_LOCAL) {
1032                         /* And the last, but not the least thing.
1033                          * We must flush stray FIB entries.
1034                          *
1035                          * First of all, we scan fib_info list searching
1036                          * for stray nexthop entries, then ignite fib_flush.
1037                          */
1038                         if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
1039                                 fib_flush(dev_net(dev));
1040                 }
1041         }
1042 #undef LOCAL_OK
1043 #undef BRD_OK
1044 #undef BRD0_OK
1045 #undef BRD1_OK
1046 }
1047
1048 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1049 {
1050
1051         struct fib_result       res;
1052         struct flowi4           fl4 = {
1053                 .flowi4_mark = frn->fl_mark,
1054                 .daddr = frn->fl_addr,
1055                 .flowi4_tos = frn->fl_tos,
1056                 .flowi4_scope = frn->fl_scope,
1057         };
1058         struct fib_table *tb;
1059
1060         rcu_read_lock();
1061
1062         tb = fib_get_table(net, frn->tb_id_in);
1063
1064         frn->err = -ENOENT;
1065         if (tb) {
1066                 local_bh_disable();
1067
1068                 frn->tb_id = tb->tb_id;
1069                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1070
1071                 if (!frn->err) {
1072                         frn->prefixlen = res.prefixlen;
1073                         frn->nh_sel = res.nh_sel;
1074                         frn->type = res.type;
1075                         frn->scope = res.scope;
1076                 }
1077                 local_bh_enable();
1078         }
1079
1080         rcu_read_unlock();
1081 }
1082
1083 static void nl_fib_input(struct sk_buff *skb)
1084 {
1085         struct net *net;
1086         struct fib_result_nl *frn;
1087         struct nlmsghdr *nlh;
1088         u32 portid;
1089
1090         net = sock_net(skb->sk);
1091         nlh = nlmsg_hdr(skb);
1092         if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1093             skb->len < nlh->nlmsg_len ||
1094             nlmsg_len(nlh) < sizeof(*frn))
1095                 return;
1096
1097         skb = netlink_skb_clone(skb, GFP_KERNEL);
1098         if (!skb)
1099                 return;
1100         nlh = nlmsg_hdr(skb);
1101
1102         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1103         nl_fib_lookup(net, frn);
1104
1105         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1106         NETLINK_CB(skb).portid = 0;        /* from kernel */
1107         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1108         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1109 }
1110
1111 static int __net_init nl_fib_lookup_init(struct net *net)
1112 {
1113         struct sock *sk;
1114         struct netlink_kernel_cfg cfg = {
1115                 .input  = nl_fib_input,
1116         };
1117
1118         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1119         if (!sk)
1120                 return -EAFNOSUPPORT;
1121         net->ipv4.fibnl = sk;
1122         return 0;
1123 }
1124
1125 static void nl_fib_lookup_exit(struct net *net)
1126 {
1127         netlink_kernel_release(net->ipv4.fibnl);
1128         net->ipv4.fibnl = NULL;
1129 }
1130
1131 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1132                            bool force)
1133 {
1134         if (fib_sync_down_dev(dev, event, force))
1135                 fib_flush(dev_net(dev));
1136         rt_cache_flush(dev_net(dev));
1137         arp_ifdown(dev);
1138 }
1139
1140 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1141 {
1142         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1143         struct net_device *dev = ifa->ifa_dev->dev;
1144         struct net *net = dev_net(dev);
1145
1146         switch (event) {
1147         case NETDEV_UP:
1148                 fib_add_ifaddr(ifa);
1149 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1150                 fib_sync_up(dev, RTNH_F_DEAD);
1151 #endif
1152                 atomic_inc(&net->ipv4.dev_addr_genid);
1153                 rt_cache_flush(dev_net(dev));
1154                 break;
1155         case NETDEV_DOWN:
1156                 fib_del_ifaddr(ifa, NULL);
1157                 atomic_inc(&net->ipv4.dev_addr_genid);
1158                 if (!ifa->ifa_dev->ifa_list) {
1159                         /* Last address was deleted from this interface.
1160                          * Disable IP.
1161                          */
1162                         fib_disable_ip(dev, event, true);
1163                 } else {
1164                         rt_cache_flush(dev_net(dev));
1165                 }
1166                 break;
1167         }
1168         return NOTIFY_DONE;
1169 }
1170
1171 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1172 {
1173         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1174         struct netdev_notifier_changeupper_info *upper_info = ptr;
1175         struct netdev_notifier_info_ext *info_ext = ptr;
1176         struct in_device *in_dev;
1177         struct net *net = dev_net(dev);
1178         unsigned int flags;
1179
1180         if (event == NETDEV_UNREGISTER) {
1181                 fib_disable_ip(dev, event, true);
1182                 rt_flush_dev(dev);
1183                 return NOTIFY_DONE;
1184         }
1185
1186         in_dev = __in_dev_get_rtnl(dev);
1187         if (!in_dev)
1188                 return NOTIFY_DONE;
1189
1190         switch (event) {
1191         case NETDEV_UP:
1192                 for_ifa(in_dev) {
1193                         fib_add_ifaddr(ifa);
1194                 } endfor_ifa(in_dev);
1195 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1196                 fib_sync_up(dev, RTNH_F_DEAD);
1197 #endif
1198                 atomic_inc(&net->ipv4.dev_addr_genid);
1199                 rt_cache_flush(net);
1200                 break;
1201         case NETDEV_DOWN:
1202                 fib_disable_ip(dev, event, false);
1203                 break;
1204         case NETDEV_CHANGE:
1205                 flags = dev_get_flags(dev);
1206                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1207                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1208                 else
1209                         fib_sync_down_dev(dev, event, false);
1210                 rt_cache_flush(net);
1211                 break;
1212         case NETDEV_CHANGEMTU:
1213                 fib_sync_mtu(dev, info_ext->ext.mtu);
1214                 rt_cache_flush(net);
1215                 break;
1216         case NETDEV_CHANGEUPPER:
1217                 upper_info = ptr;
1218                 /* flush all routes if dev is linked to or unlinked from
1219                  * an L3 master device (e.g., VRF)
1220                  */
1221                 if (upper_info->upper_dev &&
1222                     netif_is_l3_master(upper_info->upper_dev))
1223                         fib_disable_ip(dev, NETDEV_DOWN, true);
1224                 break;
1225         }
1226         return NOTIFY_DONE;
1227 }
1228
1229 static struct notifier_block fib_inetaddr_notifier = {
1230         .notifier_call = fib_inetaddr_event,
1231 };
1232
1233 static struct notifier_block fib_netdev_notifier = {
1234         .notifier_call = fib_netdev_event,
1235 };
1236
1237 static int __net_init ip_fib_net_init(struct net *net)
1238 {
1239         int err;
1240         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1241
1242         /* Avoid false sharing : Use at least a full cache line */
1243         size = max_t(size_t, size, L1_CACHE_BYTES);
1244
1245         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1246         if (!net->ipv4.fib_table_hash)
1247                 return -ENOMEM;
1248
1249         err = fib4_rules_init(net);
1250         if (err < 0)
1251                 goto fail;
1252         return 0;
1253
1254 fail:
1255         kfree(net->ipv4.fib_table_hash);
1256         return err;
1257 }
1258
1259 static void ip_fib_net_exit(struct net *net)
1260 {
1261         int i;
1262
1263         rtnl_lock();
1264 #ifdef CONFIG_IP_MULTIPLE_TABLES
1265         RCU_INIT_POINTER(net->ipv4.fib_local, NULL);
1266         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1267         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1268 #endif
1269         /* Destroy the tables in reverse order to guarantee that the
1270          * local table, ID 255, is destroyed before the main table, ID
1271          * 254. This is necessary as the local table may contain
1272          * references to data contained in the main table.
1273          */
1274         for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1275                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1276                 struct hlist_node *tmp;
1277                 struct fib_table *tb;
1278
1279                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1280                         hlist_del(&tb->tb_hlist);
1281                         fib_table_flush(tb, true);
1282                         fib_free_table(tb);
1283                 }
1284         }
1285
1286 #ifdef CONFIG_IP_MULTIPLE_TABLES
1287         fib4_rules_exit(net);
1288 #endif
1289         rtnl_unlock();
1290         kfree(net->ipv4.fib_table_hash);
1291 }
1292
1293 static int __net_init fib_net_init(struct net *net)
1294 {
1295         int error;
1296
1297 #ifdef CONFIG_IP_ROUTE_CLASSID
1298         net->ipv4.fib_num_tclassid_users = 0;
1299 #endif
1300         error = ip_fib_net_init(net);
1301         if (error < 0)
1302                 goto out;
1303         error = nl_fib_lookup_init(net);
1304         if (error < 0)
1305                 goto out_nlfl;
1306         error = fib_proc_init(net);
1307         if (error < 0)
1308                 goto out_proc;
1309 out:
1310         return error;
1311
1312 out_proc:
1313         nl_fib_lookup_exit(net);
1314 out_nlfl:
1315         ip_fib_net_exit(net);
1316         goto out;
1317 }
1318
1319 static void __net_exit fib_net_exit(struct net *net)
1320 {
1321         fib_proc_exit(net);
1322         nl_fib_lookup_exit(net);
1323         ip_fib_net_exit(net);
1324 }
1325
1326 static struct pernet_operations fib_net_ops = {
1327         .init = fib_net_init,
1328         .exit = fib_net_exit,
1329 };
1330
1331 void __init ip_fib_init(void)
1332 {
1333         fib_trie_init();
1334
1335         register_pernet_subsys(&fib_net_ops);
1336
1337         register_netdevice_notifier(&fib_netdev_notifier);
1338         register_inetaddr_notifier(&fib_inetaddr_notifier);
1339
1340         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1341         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1342         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1343 }