Merge tag 'mac80211-next-for-davem-2015-08-14' of git://git.kernel.org/pub/scm/linux...
[firefly-linux-kernel-4.4.55.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/vrf.h>
49
50 #ifndef CONFIG_IP_MULTIPLE_TABLES
51
52 static int __net_init fib4_rules_init(struct net *net)
53 {
54         struct fib_table *local_table, *main_table;
55
56         main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
57         if (!main_table)
58                 return -ENOMEM;
59
60         local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
61         if (!local_table)
62                 goto fail;
63
64         hlist_add_head_rcu(&local_table->tb_hlist,
65                                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
66         hlist_add_head_rcu(&main_table->tb_hlist,
67                                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
68         return 0;
69
70 fail:
71         fib_free_table(main_table);
72         return -ENOMEM;
73 }
74 #else
75
76 struct fib_table *fib_new_table(struct net *net, u32 id)
77 {
78         struct fib_table *tb, *alias = NULL;
79         unsigned int h;
80
81         if (id == 0)
82                 id = RT_TABLE_MAIN;
83         tb = fib_get_table(net, id);
84         if (tb)
85                 return tb;
86
87         if (id == RT_TABLE_LOCAL)
88                 alias = fib_new_table(net, RT_TABLE_MAIN);
89
90         tb = fib_trie_table(id, alias);
91         if (!tb)
92                 return NULL;
93
94         switch (id) {
95         case RT_TABLE_LOCAL:
96                 rcu_assign_pointer(net->ipv4.fib_local, tb);
97                 break;
98         case RT_TABLE_MAIN:
99                 rcu_assign_pointer(net->ipv4.fib_main, tb);
100                 break;
101         case RT_TABLE_DEFAULT:
102                 rcu_assign_pointer(net->ipv4.fib_default, tb);
103                 break;
104         default:
105                 break;
106         }
107
108         h = id & (FIB_TABLE_HASHSZ - 1);
109         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
110         return tb;
111 }
112
113 /* caller must hold either rtnl or rcu read lock */
114 struct fib_table *fib_get_table(struct net *net, u32 id)
115 {
116         struct fib_table *tb;
117         struct hlist_head *head;
118         unsigned int h;
119
120         if (id == 0)
121                 id = RT_TABLE_MAIN;
122         h = id & (FIB_TABLE_HASHSZ - 1);
123
124         head = &net->ipv4.fib_table_hash[h];
125         hlist_for_each_entry_rcu(tb, head, tb_hlist) {
126                 if (tb->tb_id == id)
127                         return tb;
128         }
129         return NULL;
130 }
131 #endif /* CONFIG_IP_MULTIPLE_TABLES */
132
133 static void fib_replace_table(struct net *net, struct fib_table *old,
134                               struct fib_table *new)
135 {
136 #ifdef CONFIG_IP_MULTIPLE_TABLES
137         switch (new->tb_id) {
138         case RT_TABLE_LOCAL:
139                 rcu_assign_pointer(net->ipv4.fib_local, new);
140                 break;
141         case RT_TABLE_MAIN:
142                 rcu_assign_pointer(net->ipv4.fib_main, new);
143                 break;
144         case RT_TABLE_DEFAULT:
145                 rcu_assign_pointer(net->ipv4.fib_default, new);
146                 break;
147         default:
148                 break;
149         }
150
151 #endif
152         /* replace the old table in the hlist */
153         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
154 }
155
156 int fib_unmerge(struct net *net)
157 {
158         struct fib_table *old, *new;
159
160         /* attempt to fetch local table if it has been allocated */
161         old = fib_get_table(net, RT_TABLE_LOCAL);
162         if (!old)
163                 return 0;
164
165         new = fib_trie_unmerge(old);
166         if (!new)
167                 return -ENOMEM;
168
169         /* replace merged table with clean table */
170         if (new != old) {
171                 fib_replace_table(net, old, new);
172                 fib_free_table(old);
173         }
174
175         return 0;
176 }
177
178 static void fib_flush(struct net *net)
179 {
180         int flushed = 0;
181         unsigned int h;
182
183         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
184                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
185                 struct hlist_node *tmp;
186                 struct fib_table *tb;
187
188                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
189                         flushed += fib_table_flush(tb);
190         }
191
192         if (flushed)
193                 rt_cache_flush(net);
194 }
195
196 void fib_flush_external(struct net *net)
197 {
198         struct fib_table *tb;
199         struct hlist_head *head;
200         unsigned int h;
201
202         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
203                 head = &net->ipv4.fib_table_hash[h];
204                 hlist_for_each_entry(tb, head, tb_hlist)
205                         fib_table_flush_external(tb);
206         }
207 }
208
209 /*
210  * Find address type as if only "dev" was present in the system. If
211  * on_dev is NULL then all interfaces are taken into consideration.
212  */
213 static inline unsigned int __inet_dev_addr_type(struct net *net,
214                                                 const struct net_device *dev,
215                                                 __be32 addr, int tb_id)
216 {
217         struct flowi4           fl4 = { .daddr = addr };
218         struct fib_result       res;
219         unsigned int ret = RTN_BROADCAST;
220         struct fib_table *table;
221
222         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
223                 return RTN_BROADCAST;
224         if (ipv4_is_multicast(addr))
225                 return RTN_MULTICAST;
226
227         rcu_read_lock();
228
229         table = fib_get_table(net, tb_id);
230         if (table) {
231                 ret = RTN_UNICAST;
232                 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
233                         if (!dev || dev == res.fi->fib_dev)
234                                 ret = res.type;
235                 }
236         }
237
238         rcu_read_unlock();
239         return ret;
240 }
241
242 unsigned int inet_addr_type_table(struct net *net, __be32 addr, int tb_id)
243 {
244         return __inet_dev_addr_type(net, NULL, addr, tb_id);
245 }
246 EXPORT_SYMBOL(inet_addr_type_table);
247
248 unsigned int inet_addr_type(struct net *net, __be32 addr)
249 {
250         return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
251 }
252 EXPORT_SYMBOL(inet_addr_type);
253
254 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
255                                 __be32 addr)
256 {
257         int rt_table = vrf_dev_table(dev) ? : RT_TABLE_LOCAL;
258
259         return __inet_dev_addr_type(net, dev, addr, rt_table);
260 }
261 EXPORT_SYMBOL(inet_dev_addr_type);
262
263 /* inet_addr_type with dev == NULL but using the table from a dev
264  * if one is associated
265  */
266 unsigned int inet_addr_type_dev_table(struct net *net,
267                                       const struct net_device *dev,
268                                       __be32 addr)
269 {
270         int rt_table = vrf_dev_table(dev) ? : RT_TABLE_LOCAL;
271
272         return __inet_dev_addr_type(net, NULL, addr, rt_table);
273 }
274 EXPORT_SYMBOL(inet_addr_type_dev_table);
275
276 __be32 fib_compute_spec_dst(struct sk_buff *skb)
277 {
278         struct net_device *dev = skb->dev;
279         struct in_device *in_dev;
280         struct fib_result res;
281         struct rtable *rt;
282         struct flowi4 fl4;
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         BUG_ON(!in_dev);
293
294         net = dev_net(dev);
295
296         scope = RT_SCOPE_UNIVERSE;
297         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
298                 fl4.flowi4_oif = 0;
299                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
300                 fl4.daddr = ip_hdr(skb)->saddr;
301                 fl4.saddr = 0;
302                 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
303                 fl4.flowi4_scope = scope;
304                 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
305                 fl4.flowi4_tun_key.tun_id = 0;
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 = vrf_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
343         no_addr = idev->ifa_list == NULL;
344
345         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
346
347         net = dev_net(dev);
348         if (fib_lookup(net, &fl4, &res, 0))
349                 goto last_resort;
350         if (res.type != RTN_UNICAST &&
351             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
352                 goto e_inval;
353         if (!rpf && !fib_num_tclassid_users(dev_net(dev)) &&
354             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
355                 goto last_resort;
356         fib_combine_itag(itag, &res);
357         dev_match = false;
358
359 #ifdef CONFIG_IP_ROUTE_MULTIPATH
360         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
361                 struct fib_nh *nh = &res.fi->fib_nh[ret];
362
363                 if (nh->nh_dev == dev) {
364                         dev_match = true;
365                         break;
366                 } else if (vrf_master_ifindex_rcu(nh->nh_dev) == dev->ifindex) {
367                         dev_match = true;
368                         break;
369                 }
370         }
371 #else
372         if (FIB_RES_DEV(res) == dev)
373                 dev_match = true;
374 #endif
375         if (dev_match) {
376                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
377                 return ret;
378         }
379         if (no_addr)
380                 goto last_resort;
381         if (rpf == 1)
382                 goto e_rpf;
383         fl4.flowi4_oif = dev->ifindex;
384
385         ret = 0;
386         if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
387                 if (res.type == RTN_UNICAST)
388                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
389         }
390         return ret;
391
392 last_resort:
393         if (rpf)
394                 goto e_rpf;
395         *itag = 0;
396         return 0;
397
398 e_inval:
399         return -EINVAL;
400 e_rpf:
401         return -EXDEV;
402 }
403
404 /* Ignore rp_filter for packets protected by IPsec. */
405 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
406                         u8 tos, int oif, struct net_device *dev,
407                         struct in_device *idev, u32 *itag)
408 {
409         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
410
411         if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
412             IN_DEV_ACCEPT_LOCAL(idev) &&
413             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
414                 *itag = 0;
415                 return 0;
416         }
417         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
418 }
419
420 static inline __be32 sk_extract_addr(struct sockaddr *addr)
421 {
422         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
423 }
424
425 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
426 {
427         struct nlattr *nla;
428
429         nla = (struct nlattr *) ((char *) mx + len);
430         nla->nla_type = type;
431         nla->nla_len = nla_attr_size(4);
432         *(u32 *) nla_data(nla) = value;
433
434         return len + nla_total_size(4);
435 }
436
437 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
438                                  struct fib_config *cfg)
439 {
440         __be32 addr;
441         int plen;
442
443         memset(cfg, 0, sizeof(*cfg));
444         cfg->fc_nlinfo.nl_net = net;
445
446         if (rt->rt_dst.sa_family != AF_INET)
447                 return -EAFNOSUPPORT;
448
449         /*
450          * Check mask for validity:
451          * a) it must be contiguous.
452          * b) destination must have all host bits clear.
453          * c) if application forgot to set correct family (AF_INET),
454          *    reject request unless it is absolutely clear i.e.
455          *    both family and mask are zero.
456          */
457         plen = 32;
458         addr = sk_extract_addr(&rt->rt_dst);
459         if (!(rt->rt_flags & RTF_HOST)) {
460                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
461
462                 if (rt->rt_genmask.sa_family != AF_INET) {
463                         if (mask || rt->rt_genmask.sa_family)
464                                 return -EAFNOSUPPORT;
465                 }
466
467                 if (bad_mask(mask, addr))
468                         return -EINVAL;
469
470                 plen = inet_mask_len(mask);
471         }
472
473         cfg->fc_dst_len = plen;
474         cfg->fc_dst = addr;
475
476         if (cmd != SIOCDELRT) {
477                 cfg->fc_nlflags = NLM_F_CREATE;
478                 cfg->fc_protocol = RTPROT_BOOT;
479         }
480
481         if (rt->rt_metric)
482                 cfg->fc_priority = rt->rt_metric - 1;
483
484         if (rt->rt_flags & RTF_REJECT) {
485                 cfg->fc_scope = RT_SCOPE_HOST;
486                 cfg->fc_type = RTN_UNREACHABLE;
487                 return 0;
488         }
489
490         cfg->fc_scope = RT_SCOPE_NOWHERE;
491         cfg->fc_type = RTN_UNICAST;
492
493         if (rt->rt_dev) {
494                 char *colon;
495                 struct net_device *dev;
496                 char devname[IFNAMSIZ];
497
498                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
499                         return -EFAULT;
500
501                 devname[IFNAMSIZ-1] = 0;
502                 colon = strchr(devname, ':');
503                 if (colon)
504                         *colon = 0;
505                 dev = __dev_get_by_name(net, devname);
506                 if (!dev)
507                         return -ENODEV;
508                 cfg->fc_oif = dev->ifindex;
509                 if (colon) {
510                         struct in_ifaddr *ifa;
511                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
512                         if (!in_dev)
513                                 return -ENODEV;
514                         *colon = ':';
515                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
516                                 if (strcmp(ifa->ifa_label, devname) == 0)
517                                         break;
518                         if (!ifa)
519                                 return -ENODEV;
520                         cfg->fc_prefsrc = ifa->ifa_local;
521                 }
522         }
523
524         addr = sk_extract_addr(&rt->rt_gateway);
525         if (rt->rt_gateway.sa_family == AF_INET && addr) {
526                 unsigned int addr_type;
527
528                 cfg->fc_gw = addr;
529                 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
530                 if (rt->rt_flags & RTF_GATEWAY &&
531                     addr_type == RTN_UNICAST)
532                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
533         }
534
535         if (cmd == SIOCDELRT)
536                 return 0;
537
538         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
539                 return -EINVAL;
540
541         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
542                 cfg->fc_scope = RT_SCOPE_LINK;
543
544         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
545                 struct nlattr *mx;
546                 int len = 0;
547
548                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
549                 if (!mx)
550                         return -ENOMEM;
551
552                 if (rt->rt_flags & RTF_MTU)
553                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
554
555                 if (rt->rt_flags & RTF_WINDOW)
556                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
557
558                 if (rt->rt_flags & RTF_IRTT)
559                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
560
561                 cfg->fc_mx = mx;
562                 cfg->fc_mx_len = len;
563         }
564
565         return 0;
566 }
567
568 /*
569  * Handle IP routing ioctl calls.
570  * These are used to manipulate the routing tables
571  */
572 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
573 {
574         struct fib_config cfg;
575         struct rtentry rt;
576         int err;
577
578         switch (cmd) {
579         case SIOCADDRT:         /* Add a route */
580         case SIOCDELRT:         /* Delete a route */
581                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
582                         return -EPERM;
583
584                 if (copy_from_user(&rt, arg, sizeof(rt)))
585                         return -EFAULT;
586
587                 rtnl_lock();
588                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
589                 if (err == 0) {
590                         struct fib_table *tb;
591
592                         if (cmd == SIOCDELRT) {
593                                 tb = fib_get_table(net, cfg.fc_table);
594                                 if (tb)
595                                         err = fib_table_delete(tb, &cfg);
596                                 else
597                                         err = -ESRCH;
598                         } else {
599                                 tb = fib_new_table(net, cfg.fc_table);
600                                 if (tb)
601                                         err = fib_table_insert(tb, &cfg);
602                                 else
603                                         err = -ENOBUFS;
604                         }
605
606                         /* allocated by rtentry_to_fib_config() */
607                         kfree(cfg.fc_mx);
608                 }
609                 rtnl_unlock();
610                 return err;
611         }
612         return -EINVAL;
613 }
614
615 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
616         [RTA_DST]               = { .type = NLA_U32 },
617         [RTA_SRC]               = { .type = NLA_U32 },
618         [RTA_IIF]               = { .type = NLA_U32 },
619         [RTA_OIF]               = { .type = NLA_U32 },
620         [RTA_GATEWAY]           = { .type = NLA_U32 },
621         [RTA_PRIORITY]          = { .type = NLA_U32 },
622         [RTA_PREFSRC]           = { .type = NLA_U32 },
623         [RTA_METRICS]           = { .type = NLA_NESTED },
624         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
625         [RTA_FLOW]              = { .type = NLA_U32 },
626         [RTA_ENCAP_TYPE]        = { .type = NLA_U16 },
627         [RTA_ENCAP]             = { .type = NLA_NESTED },
628 };
629
630 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
631                              struct nlmsghdr *nlh, struct fib_config *cfg)
632 {
633         struct nlattr *attr;
634         int err, remaining;
635         struct rtmsg *rtm;
636
637         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
638         if (err < 0)
639                 goto errout;
640
641         memset(cfg, 0, sizeof(*cfg));
642
643         rtm = nlmsg_data(nlh);
644         cfg->fc_dst_len = rtm->rtm_dst_len;
645         cfg->fc_tos = rtm->rtm_tos;
646         cfg->fc_table = rtm->rtm_table;
647         cfg->fc_protocol = rtm->rtm_protocol;
648         cfg->fc_scope = rtm->rtm_scope;
649         cfg->fc_type = rtm->rtm_type;
650         cfg->fc_flags = rtm->rtm_flags;
651         cfg->fc_nlflags = nlh->nlmsg_flags;
652
653         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
654         cfg->fc_nlinfo.nlh = nlh;
655         cfg->fc_nlinfo.nl_net = net;
656
657         if (cfg->fc_type > RTN_MAX) {
658                 err = -EINVAL;
659                 goto errout;
660         }
661
662         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
663                 switch (nla_type(attr)) {
664                 case RTA_DST:
665                         cfg->fc_dst = nla_get_be32(attr);
666                         break;
667                 case RTA_OIF:
668                         cfg->fc_oif = nla_get_u32(attr);
669                         break;
670                 case RTA_GATEWAY:
671                         cfg->fc_gw = nla_get_be32(attr);
672                         break;
673                 case RTA_PRIORITY:
674                         cfg->fc_priority = nla_get_u32(attr);
675                         break;
676                 case RTA_PREFSRC:
677                         cfg->fc_prefsrc = nla_get_be32(attr);
678                         break;
679                 case RTA_METRICS:
680                         cfg->fc_mx = nla_data(attr);
681                         cfg->fc_mx_len = nla_len(attr);
682                         break;
683                 case RTA_MULTIPATH:
684                         cfg->fc_mp = nla_data(attr);
685                         cfg->fc_mp_len = nla_len(attr);
686                         break;
687                 case RTA_FLOW:
688                         cfg->fc_flow = nla_get_u32(attr);
689                         break;
690                 case RTA_TABLE:
691                         cfg->fc_table = nla_get_u32(attr);
692                         break;
693                 case RTA_ENCAP:
694                         cfg->fc_encap = attr;
695                         break;
696                 case RTA_ENCAP_TYPE:
697                         cfg->fc_encap_type = nla_get_u16(attr);
698                         break;
699                 }
700         }
701
702         return 0;
703 errout:
704         return err;
705 }
706
707 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
708 {
709         struct net *net = sock_net(skb->sk);
710         struct fib_config cfg;
711         struct fib_table *tb;
712         int err;
713
714         err = rtm_to_fib_config(net, skb, nlh, &cfg);
715         if (err < 0)
716                 goto errout;
717
718         tb = fib_get_table(net, cfg.fc_table);
719         if (!tb) {
720                 err = -ESRCH;
721                 goto errout;
722         }
723
724         err = fib_table_delete(tb, &cfg);
725 errout:
726         return err;
727 }
728
729 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
730 {
731         struct net *net = sock_net(skb->sk);
732         struct fib_config cfg;
733         struct fib_table *tb;
734         int err;
735
736         err = rtm_to_fib_config(net, skb, nlh, &cfg);
737         if (err < 0)
738                 goto errout;
739
740         tb = fib_new_table(net, cfg.fc_table);
741         if (!tb) {
742                 err = -ENOBUFS;
743                 goto errout;
744         }
745
746         err = fib_table_insert(tb, &cfg);
747 errout:
748         return err;
749 }
750
751 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
752 {
753         struct net *net = sock_net(skb->sk);
754         unsigned int h, s_h;
755         unsigned int e = 0, s_e;
756         struct fib_table *tb;
757         struct hlist_head *head;
758         int dumped = 0;
759
760         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
761             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
762                 return skb->len;
763
764         s_h = cb->args[0];
765         s_e = cb->args[1];
766
767         rcu_read_lock();
768
769         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
770                 e = 0;
771                 head = &net->ipv4.fib_table_hash[h];
772                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
773                         if (e < s_e)
774                                 goto next;
775                         if (dumped)
776                                 memset(&cb->args[2], 0, sizeof(cb->args) -
777                                                  2 * sizeof(cb->args[0]));
778                         if (fib_table_dump(tb, skb, cb) < 0)
779                                 goto out;
780                         dumped = 1;
781 next:
782                         e++;
783                 }
784         }
785 out:
786         rcu_read_unlock();
787
788         cb->args[1] = e;
789         cb->args[0] = h;
790
791         return skb->len;
792 }
793
794 /* Prepare and feed intra-kernel routing request.
795  * Really, it should be netlink message, but :-( netlink
796  * can be not configured, so that we feed it directly
797  * to fib engine. It is legal, because all events occur
798  * only when netlink is already locked.
799  */
800 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
801 {
802         struct net *net = dev_net(ifa->ifa_dev->dev);
803         int tb_id = vrf_dev_table_rtnl(ifa->ifa_dev->dev);
804         struct fib_table *tb;
805         struct fib_config cfg = {
806                 .fc_protocol = RTPROT_KERNEL,
807                 .fc_type = type,
808                 .fc_dst = dst,
809                 .fc_dst_len = dst_len,
810                 .fc_prefsrc = ifa->ifa_local,
811                 .fc_oif = ifa->ifa_dev->dev->ifindex,
812                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
813                 .fc_nlinfo = {
814                         .nl_net = net,
815                 },
816         };
817
818         if (!tb_id)
819                 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
820
821         tb = fib_new_table(net, tb_id);
822         if (!tb)
823                 return;
824
825         cfg.fc_table = tb->tb_id;
826
827         if (type != RTN_LOCAL)
828                 cfg.fc_scope = RT_SCOPE_LINK;
829         else
830                 cfg.fc_scope = RT_SCOPE_HOST;
831
832         if (cmd == RTM_NEWROUTE)
833                 fib_table_insert(tb, &cfg);
834         else
835                 fib_table_delete(tb, &cfg);
836 }
837
838 void fib_add_ifaddr(struct in_ifaddr *ifa)
839 {
840         struct in_device *in_dev = ifa->ifa_dev;
841         struct net_device *dev = in_dev->dev;
842         struct in_ifaddr *prim = ifa;
843         __be32 mask = ifa->ifa_mask;
844         __be32 addr = ifa->ifa_local;
845         __be32 prefix = ifa->ifa_address & mask;
846
847         if (ifa->ifa_flags & IFA_F_SECONDARY) {
848                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
849                 if (!prim) {
850                         pr_warn("%s: bug: prim == NULL\n", __func__);
851                         return;
852                 }
853         }
854
855         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
856
857         if (!(dev->flags & IFF_UP))
858                 return;
859
860         /* Add broadcast address, if it is explicitly assigned. */
861         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
862                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
863
864         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
865             (prefix != addr || ifa->ifa_prefixlen < 32)) {
866                 fib_magic(RTM_NEWROUTE,
867                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
868                           prefix, ifa->ifa_prefixlen, prim);
869
870                 /* Add network specific broadcasts, when it takes a sense */
871                 if (ifa->ifa_prefixlen < 31) {
872                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
873                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
874                                   32, prim);
875                 }
876         }
877 }
878
879 /* Delete primary or secondary address.
880  * Optionally, on secondary address promotion consider the addresses
881  * from subnet iprim as deleted, even if they are in device list.
882  * In this case the secondary ifa can be in device list.
883  */
884 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
885 {
886         struct in_device *in_dev = ifa->ifa_dev;
887         struct net_device *dev = in_dev->dev;
888         struct in_ifaddr *ifa1;
889         struct in_ifaddr *prim = ifa, *prim1 = NULL;
890         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
891         __be32 any = ifa->ifa_address & ifa->ifa_mask;
892 #define LOCAL_OK        1
893 #define BRD_OK          2
894 #define BRD0_OK         4
895 #define BRD1_OK         8
896         unsigned int ok = 0;
897         int subnet = 0;         /* Primary network */
898         int gone = 1;           /* Address is missing */
899         int same_prefsrc = 0;   /* Another primary with same IP */
900
901         if (ifa->ifa_flags & IFA_F_SECONDARY) {
902                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
903                 if (!prim) {
904                         pr_warn("%s: bug: prim == NULL\n", __func__);
905                         return;
906                 }
907                 if (iprim && iprim != prim) {
908                         pr_warn("%s: bug: iprim != prim\n", __func__);
909                         return;
910                 }
911         } else if (!ipv4_is_zeronet(any) &&
912                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
913                 fib_magic(RTM_DELROUTE,
914                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
915                           any, ifa->ifa_prefixlen, prim);
916                 subnet = 1;
917         }
918
919         /* Deletion is more complicated than add.
920          * We should take care of not to delete too much :-)
921          *
922          * Scan address list to be sure that addresses are really gone.
923          */
924
925         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
926                 if (ifa1 == ifa) {
927                         /* promotion, keep the IP */
928                         gone = 0;
929                         continue;
930                 }
931                 /* Ignore IFAs from our subnet */
932                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
933                     inet_ifa_match(ifa1->ifa_address, iprim))
934                         continue;
935
936                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
937                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
938                         /* Another address from our subnet? */
939                         if (ifa1->ifa_mask == prim->ifa_mask &&
940                             inet_ifa_match(ifa1->ifa_address, prim))
941                                 prim1 = prim;
942                         else {
943                                 /* We reached the secondaries, so
944                                  * same_prefsrc should be determined.
945                                  */
946                                 if (!same_prefsrc)
947                                         continue;
948                                 /* Search new prim1 if ifa1 is not
949                                  * using the current prim1
950                                  */
951                                 if (!prim1 ||
952                                     ifa1->ifa_mask != prim1->ifa_mask ||
953                                     !inet_ifa_match(ifa1->ifa_address, prim1))
954                                         prim1 = inet_ifa_byprefix(in_dev,
955                                                         ifa1->ifa_address,
956                                                         ifa1->ifa_mask);
957                                 if (!prim1)
958                                         continue;
959                                 if (prim1->ifa_local != prim->ifa_local)
960                                         continue;
961                         }
962                 } else {
963                         if (prim->ifa_local != ifa1->ifa_local)
964                                 continue;
965                         prim1 = ifa1;
966                         if (prim != prim1)
967                                 same_prefsrc = 1;
968                 }
969                 if (ifa->ifa_local == ifa1->ifa_local)
970                         ok |= LOCAL_OK;
971                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
972                         ok |= BRD_OK;
973                 if (brd == ifa1->ifa_broadcast)
974                         ok |= BRD1_OK;
975                 if (any == ifa1->ifa_broadcast)
976                         ok |= BRD0_OK;
977                 /* primary has network specific broadcasts */
978                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
979                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
980                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
981
982                         if (!ipv4_is_zeronet(any1)) {
983                                 if (ifa->ifa_broadcast == brd1 ||
984                                     ifa->ifa_broadcast == any1)
985                                         ok |= BRD_OK;
986                                 if (brd == brd1 || brd == any1)
987                                         ok |= BRD1_OK;
988                                 if (any == brd1 || any == any1)
989                                         ok |= BRD0_OK;
990                         }
991                 }
992         }
993
994         if (!(ok & BRD_OK))
995                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
996         if (subnet && ifa->ifa_prefixlen < 31) {
997                 if (!(ok & BRD1_OK))
998                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
999                 if (!(ok & BRD0_OK))
1000                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
1001         }
1002         if (!(ok & LOCAL_OK)) {
1003                 unsigned int addr_type;
1004
1005                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
1006
1007                 /* Check, that this local address finally disappeared. */
1008                 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1009                                                      ifa->ifa_local);
1010                 if (gone && addr_type != RTN_LOCAL) {
1011                         /* And the last, but not the least thing.
1012                          * We must flush stray FIB entries.
1013                          *
1014                          * First of all, we scan fib_info list searching
1015                          * for stray nexthop entries, then ignite fib_flush.
1016                          */
1017                         if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
1018                                 fib_flush(dev_net(dev));
1019                 }
1020         }
1021 #undef LOCAL_OK
1022 #undef BRD_OK
1023 #undef BRD0_OK
1024 #undef BRD1_OK
1025 }
1026
1027 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1028 {
1029
1030         struct fib_result       res;
1031         struct flowi4           fl4 = {
1032                 .flowi4_mark = frn->fl_mark,
1033                 .daddr = frn->fl_addr,
1034                 .flowi4_tos = frn->fl_tos,
1035                 .flowi4_scope = frn->fl_scope,
1036         };
1037         struct fib_table *tb;
1038
1039         rcu_read_lock();
1040
1041         tb = fib_get_table(net, frn->tb_id_in);
1042
1043         frn->err = -ENOENT;
1044         if (tb) {
1045                 local_bh_disable();
1046
1047                 frn->tb_id = tb->tb_id;
1048                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1049
1050                 if (!frn->err) {
1051                         frn->prefixlen = res.prefixlen;
1052                         frn->nh_sel = res.nh_sel;
1053                         frn->type = res.type;
1054                         frn->scope = res.scope;
1055                 }
1056                 local_bh_enable();
1057         }
1058
1059         rcu_read_unlock();
1060 }
1061
1062 static void nl_fib_input(struct sk_buff *skb)
1063 {
1064         struct net *net;
1065         struct fib_result_nl *frn;
1066         struct nlmsghdr *nlh;
1067         u32 portid;
1068
1069         net = sock_net(skb->sk);
1070         nlh = nlmsg_hdr(skb);
1071         if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len ||
1072             nlmsg_len(nlh) < sizeof(*frn))
1073                 return;
1074
1075         skb = netlink_skb_clone(skb, GFP_KERNEL);
1076         if (!skb)
1077                 return;
1078         nlh = nlmsg_hdr(skb);
1079
1080         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1081         nl_fib_lookup(net, frn);
1082
1083         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1084         NETLINK_CB(skb).portid = 0;        /* from kernel */
1085         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1086         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1087 }
1088
1089 static int __net_init nl_fib_lookup_init(struct net *net)
1090 {
1091         struct sock *sk;
1092         struct netlink_kernel_cfg cfg = {
1093                 .input  = nl_fib_input,
1094         };
1095
1096         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1097         if (!sk)
1098                 return -EAFNOSUPPORT;
1099         net->ipv4.fibnl = sk;
1100         return 0;
1101 }
1102
1103 static void nl_fib_lookup_exit(struct net *net)
1104 {
1105         netlink_kernel_release(net->ipv4.fibnl);
1106         net->ipv4.fibnl = NULL;
1107 }
1108
1109 static void fib_disable_ip(struct net_device *dev, unsigned long event)
1110 {
1111         if (fib_sync_down_dev(dev, event))
1112                 fib_flush(dev_net(dev));
1113         rt_cache_flush(dev_net(dev));
1114         arp_ifdown(dev);
1115 }
1116
1117 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1118 {
1119         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1120         struct net_device *dev = ifa->ifa_dev->dev;
1121         struct net *net = dev_net(dev);
1122
1123         switch (event) {
1124         case NETDEV_UP:
1125                 fib_add_ifaddr(ifa);
1126 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1127                 fib_sync_up(dev, RTNH_F_DEAD);
1128 #endif
1129                 atomic_inc(&net->ipv4.dev_addr_genid);
1130                 rt_cache_flush(dev_net(dev));
1131                 break;
1132         case NETDEV_DOWN:
1133                 fib_del_ifaddr(ifa, NULL);
1134                 atomic_inc(&net->ipv4.dev_addr_genid);
1135                 if (!ifa->ifa_dev->ifa_list) {
1136                         /* Last address was deleted from this interface.
1137                          * Disable IP.
1138                          */
1139                         fib_disable_ip(dev, event);
1140                 } else {
1141                         rt_cache_flush(dev_net(dev));
1142                 }
1143                 break;
1144         }
1145         return NOTIFY_DONE;
1146 }
1147
1148 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1149 {
1150         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1151         struct in_device *in_dev;
1152         struct net *net = dev_net(dev);
1153         unsigned int flags;
1154
1155         if (event == NETDEV_UNREGISTER) {
1156                 fib_disable_ip(dev, event);
1157                 rt_flush_dev(dev);
1158                 return NOTIFY_DONE;
1159         }
1160
1161         in_dev = __in_dev_get_rtnl(dev);
1162         if (!in_dev)
1163                 return NOTIFY_DONE;
1164
1165         switch (event) {
1166         case NETDEV_UP:
1167                 for_ifa(in_dev) {
1168                         fib_add_ifaddr(ifa);
1169                 } endfor_ifa(in_dev);
1170 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1171                 fib_sync_up(dev, RTNH_F_DEAD);
1172 #endif
1173                 atomic_inc(&net->ipv4.dev_addr_genid);
1174                 rt_cache_flush(net);
1175                 break;
1176         case NETDEV_DOWN:
1177                 fib_disable_ip(dev, event);
1178                 break;
1179         case NETDEV_CHANGE:
1180                 flags = dev_get_flags(dev);
1181                 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1182                         fib_sync_up(dev, RTNH_F_LINKDOWN);
1183                 else
1184                         fib_sync_down_dev(dev, event);
1185                 /* fall through */
1186         case NETDEV_CHANGEMTU:
1187                 rt_cache_flush(net);
1188                 break;
1189         }
1190         return NOTIFY_DONE;
1191 }
1192
1193 static struct notifier_block fib_inetaddr_notifier = {
1194         .notifier_call = fib_inetaddr_event,
1195 };
1196
1197 static struct notifier_block fib_netdev_notifier = {
1198         .notifier_call = fib_netdev_event,
1199 };
1200
1201 static int __net_init ip_fib_net_init(struct net *net)
1202 {
1203         int err;
1204         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1205
1206         /* Avoid false sharing : Use at least a full cache line */
1207         size = max_t(size_t, size, L1_CACHE_BYTES);
1208
1209         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1210         if (!net->ipv4.fib_table_hash)
1211                 return -ENOMEM;
1212
1213         err = fib4_rules_init(net);
1214         if (err < 0)
1215                 goto fail;
1216         return 0;
1217
1218 fail:
1219         kfree(net->ipv4.fib_table_hash);
1220         return err;
1221 }
1222
1223 static void ip_fib_net_exit(struct net *net)
1224 {
1225         unsigned int i;
1226
1227         rtnl_lock();
1228 #ifdef CONFIG_IP_MULTIPLE_TABLES
1229         RCU_INIT_POINTER(net->ipv4.fib_local, NULL);
1230         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1231         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1232 #endif
1233         for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1234                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1235                 struct hlist_node *tmp;
1236                 struct fib_table *tb;
1237
1238                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1239                         hlist_del(&tb->tb_hlist);
1240                         fib_table_flush(tb);
1241                         fib_free_table(tb);
1242                 }
1243         }
1244
1245 #ifdef CONFIG_IP_MULTIPLE_TABLES
1246         fib4_rules_exit(net);
1247 #endif
1248         rtnl_unlock();
1249         kfree(net->ipv4.fib_table_hash);
1250 }
1251
1252 static int __net_init fib_net_init(struct net *net)
1253 {
1254         int error;
1255
1256 #ifdef CONFIG_IP_ROUTE_CLASSID
1257         net->ipv4.fib_num_tclassid_users = 0;
1258 #endif
1259         error = ip_fib_net_init(net);
1260         if (error < 0)
1261                 goto out;
1262         error = nl_fib_lookup_init(net);
1263         if (error < 0)
1264                 goto out_nlfl;
1265         error = fib_proc_init(net);
1266         if (error < 0)
1267                 goto out_proc;
1268 out:
1269         return error;
1270
1271 out_proc:
1272         nl_fib_lookup_exit(net);
1273 out_nlfl:
1274         ip_fib_net_exit(net);
1275         goto out;
1276 }
1277
1278 static void __net_exit fib_net_exit(struct net *net)
1279 {
1280         fib_proc_exit(net);
1281         nl_fib_lookup_exit(net);
1282         ip_fib_net_exit(net);
1283 }
1284
1285 static struct pernet_operations fib_net_ops = {
1286         .init = fib_net_init,
1287         .exit = fib_net_exit,
1288 };
1289
1290 void __init ip_fib_init(void)
1291 {
1292         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1293         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1294         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1295
1296         register_pernet_subsys(&fib_net_ops);
1297         register_netdevice_notifier(&fib_netdev_notifier);
1298         register_inetaddr_notifier(&fib_inetaddr_notifier);
1299
1300         fib_trie_init();
1301 }