Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / 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
49 #ifndef CONFIG_IP_MULTIPLE_TABLES
50
51 static int __net_init fib4_rules_init(struct net *net)
52 {
53         struct fib_table *local_table, *main_table;
54
55         main_table  = fib_trie_table(RT_TABLE_MAIN, NULL);
56         if (!main_table)
57                 return -ENOMEM;
58
59         local_table = fib_trie_table(RT_TABLE_LOCAL, main_table);
60         if (!local_table)
61                 goto fail;
62
63         hlist_add_head_rcu(&local_table->tb_hlist,
64                                 &net->ipv4.fib_table_hash[TABLE_LOCAL_INDEX]);
65         hlist_add_head_rcu(&main_table->tb_hlist,
66                                 &net->ipv4.fib_table_hash[TABLE_MAIN_INDEX]);
67         return 0;
68
69 fail:
70         fib_free_table(main_table);
71         return -ENOMEM;
72 }
73 #else
74
75 struct fib_table *fib_new_table(struct net *net, u32 id)
76 {
77         struct fib_table *tb, *alias = NULL;
78         unsigned int h;
79
80         if (id == 0)
81                 id = RT_TABLE_MAIN;
82         tb = fib_get_table(net, id);
83         if (tb)
84                 return tb;
85
86         if (id == RT_TABLE_LOCAL)
87                 alias = fib_new_table(net, RT_TABLE_MAIN);
88
89         tb = fib_trie_table(id, alias);
90         if (!tb)
91                 return NULL;
92
93         switch (id) {
94         case RT_TABLE_LOCAL:
95                 rcu_assign_pointer(net->ipv4.fib_local, tb);
96                 break;
97         case RT_TABLE_MAIN:
98                 rcu_assign_pointer(net->ipv4.fib_main, tb);
99                 break;
100         case RT_TABLE_DEFAULT:
101                 rcu_assign_pointer(net->ipv4.fib_default, tb);
102                 break;
103         default:
104                 break;
105         }
106
107         h = id & (FIB_TABLE_HASHSZ - 1);
108         hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
109         return tb;
110 }
111
112 /* caller must hold either rtnl or rcu read lock */
113 struct fib_table *fib_get_table(struct net *net, u32 id)
114 {
115         struct fib_table *tb;
116         struct hlist_head *head;
117         unsigned int h;
118
119         if (id == 0)
120                 id = RT_TABLE_MAIN;
121         h = id & (FIB_TABLE_HASHSZ - 1);
122
123         head = &net->ipv4.fib_table_hash[h];
124         hlist_for_each_entry_rcu(tb, head, tb_hlist) {
125                 if (tb->tb_id == id)
126                         return tb;
127         }
128         return NULL;
129 }
130 #endif /* CONFIG_IP_MULTIPLE_TABLES */
131
132 static void fib_replace_table(struct net *net, struct fib_table *old,
133                               struct fib_table *new)
134 {
135 #ifdef CONFIG_IP_MULTIPLE_TABLES
136         switch (new->tb_id) {
137         case RT_TABLE_LOCAL:
138                 rcu_assign_pointer(net->ipv4.fib_local, new);
139                 break;
140         case RT_TABLE_MAIN:
141                 rcu_assign_pointer(net->ipv4.fib_main, new);
142                 break;
143         case RT_TABLE_DEFAULT:
144                 rcu_assign_pointer(net->ipv4.fib_default, new);
145                 break;
146         default:
147                 break;
148         }
149
150 #endif
151         /* replace the old table in the hlist */
152         hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
153 }
154
155 int fib_unmerge(struct net *net)
156 {
157         struct fib_table *old, *new;
158
159         /* attempt to fetch local table if it has been allocated */
160         old = fib_get_table(net, RT_TABLE_LOCAL);
161         if (!old)
162                 return 0;
163
164         new = fib_trie_unmerge(old);
165         if (!new)
166                 return -ENOMEM;
167
168         /* replace merged table with clean table */
169         if (new != old) {
170                 fib_replace_table(net, old, new);
171                 fib_free_table(old);
172         }
173
174         return 0;
175 }
176
177 static void fib_flush(struct net *net)
178 {
179         int flushed = 0;
180         unsigned int h;
181
182         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
183                 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
184                 struct hlist_node *tmp;
185                 struct fib_table *tb;
186
187                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
188                         flushed += fib_table_flush(tb);
189         }
190
191         if (flushed)
192                 rt_cache_flush(net);
193 }
194
195 void fib_flush_external(struct net *net)
196 {
197         struct fib_table *tb;
198         struct hlist_head *head;
199         unsigned int h;
200
201         for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
202                 head = &net->ipv4.fib_table_hash[h];
203                 hlist_for_each_entry(tb, head, tb_hlist)
204                         fib_table_flush_external(tb);
205         }
206 }
207
208 /*
209  * Find address type as if only "dev" was present in the system. If
210  * on_dev is NULL then all interfaces are taken into consideration.
211  */
212 static inline unsigned int __inet_dev_addr_type(struct net *net,
213                                                 const struct net_device *dev,
214                                                 __be32 addr)
215 {
216         struct flowi4           fl4 = { .daddr = addr };
217         struct fib_result       res;
218         unsigned int ret = RTN_BROADCAST;
219         struct fib_table *local_table;
220
221         if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
222                 return RTN_BROADCAST;
223         if (ipv4_is_multicast(addr))
224                 return RTN_MULTICAST;
225
226         rcu_read_lock();
227
228         local_table = fib_get_table(net, RT_TABLE_LOCAL);
229         if (local_table) {
230                 ret = RTN_UNICAST;
231                 if (!fib_table_lookup(local_table, &fl4, &res, FIB_LOOKUP_NOREF)) {
232                         if (!dev || dev == res.fi->fib_dev)
233                                 ret = res.type;
234                 }
235         }
236
237         rcu_read_unlock();
238         return ret;
239 }
240
241 unsigned int inet_addr_type(struct net *net, __be32 addr)
242 {
243         return __inet_dev_addr_type(net, NULL, addr);
244 }
245 EXPORT_SYMBOL(inet_addr_type);
246
247 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
248                                 __be32 addr)
249 {
250         return __inet_dev_addr_type(net, dev, addr);
251 }
252 EXPORT_SYMBOL(inet_dev_addr_type);
253
254 __be32 fib_compute_spec_dst(struct sk_buff *skb)
255 {
256         struct net_device *dev = skb->dev;
257         struct in_device *in_dev;
258         struct fib_result res;
259         struct rtable *rt;
260         struct flowi4 fl4;
261         struct net *net;
262         int scope;
263
264         rt = skb_rtable(skb);
265         if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
266             RTCF_LOCAL)
267                 return ip_hdr(skb)->daddr;
268
269         in_dev = __in_dev_get_rcu(dev);
270         BUG_ON(!in_dev);
271
272         net = dev_net(dev);
273
274         scope = RT_SCOPE_UNIVERSE;
275         if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
276                 fl4.flowi4_oif = 0;
277                 fl4.flowi4_iif = LOOPBACK_IFINDEX;
278                 fl4.daddr = ip_hdr(skb)->saddr;
279                 fl4.saddr = 0;
280                 fl4.flowi4_tos = RT_TOS(ip_hdr(skb)->tos);
281                 fl4.flowi4_scope = scope;
282                 fl4.flowi4_mark = IN_DEV_SRC_VMARK(in_dev) ? skb->mark : 0;
283                 if (!fib_lookup(net, &fl4, &res))
284                         return FIB_RES_PREFSRC(net, res);
285         } else {
286                 scope = RT_SCOPE_LINK;
287         }
288
289         return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
290 }
291
292 /* Given (packet source, input interface) and optional (dst, oif, tos):
293  * - (main) check, that source is valid i.e. not broadcast or our local
294  *   address.
295  * - figure out what "logical" interface this packet arrived
296  *   and calculate "specific destination" address.
297  * - check, that packet arrived from expected physical interface.
298  * called with rcu_read_lock()
299  */
300 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
301                                  u8 tos, int oif, struct net_device *dev,
302                                  int rpf, struct in_device *idev, u32 *itag)
303 {
304         int ret, no_addr;
305         struct fib_result res;
306         struct flowi4 fl4;
307         struct net *net;
308         bool dev_match;
309
310         fl4.flowi4_oif = 0;
311         fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
312         fl4.daddr = src;
313         fl4.saddr = dst;
314         fl4.flowi4_tos = tos;
315         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
316
317         no_addr = idev->ifa_list == NULL;
318
319         fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
320
321         net = dev_net(dev);
322         if (fib_lookup(net, &fl4, &res))
323                 goto last_resort;
324         if (res.type != RTN_UNICAST &&
325             (res.type != RTN_LOCAL || !IN_DEV_ACCEPT_LOCAL(idev)))
326                 goto e_inval;
327         if (!rpf && !fib_num_tclassid_users(dev_net(dev)) &&
328             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev)))
329                 goto last_resort;
330         fib_combine_itag(itag, &res);
331         dev_match = false;
332
333 #ifdef CONFIG_IP_ROUTE_MULTIPATH
334         for (ret = 0; ret < res.fi->fib_nhs; ret++) {
335                 struct fib_nh *nh = &res.fi->fib_nh[ret];
336
337                 if (nh->nh_dev == dev) {
338                         dev_match = true;
339                         break;
340                 }
341         }
342 #else
343         if (FIB_RES_DEV(res) == dev)
344                 dev_match = true;
345 #endif
346         if (dev_match) {
347                 ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
348                 return ret;
349         }
350         if (no_addr)
351                 goto last_resort;
352         if (rpf == 1)
353                 goto e_rpf;
354         fl4.flowi4_oif = dev->ifindex;
355
356         ret = 0;
357         if (fib_lookup(net, &fl4, &res) == 0) {
358                 if (res.type == RTN_UNICAST)
359                         ret = FIB_RES_NH(res).nh_scope >= RT_SCOPE_HOST;
360         }
361         return ret;
362
363 last_resort:
364         if (rpf)
365                 goto e_rpf;
366         *itag = 0;
367         return 0;
368
369 e_inval:
370         return -EINVAL;
371 e_rpf:
372         return -EXDEV;
373 }
374
375 /* Ignore rp_filter for packets protected by IPsec. */
376 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
377                         u8 tos, int oif, struct net_device *dev,
378                         struct in_device *idev, u32 *itag)
379 {
380         int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
381
382         if (!r && !fib_num_tclassid_users(dev_net(dev)) &&
383             IN_DEV_ACCEPT_LOCAL(idev) &&
384             (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
385                 *itag = 0;
386                 return 0;
387         }
388         return __fib_validate_source(skb, src, dst, tos, oif, dev, r, idev, itag);
389 }
390
391 static inline __be32 sk_extract_addr(struct sockaddr *addr)
392 {
393         return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
394 }
395
396 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
397 {
398         struct nlattr *nla;
399
400         nla = (struct nlattr *) ((char *) mx + len);
401         nla->nla_type = type;
402         nla->nla_len = nla_attr_size(4);
403         *(u32 *) nla_data(nla) = value;
404
405         return len + nla_total_size(4);
406 }
407
408 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
409                                  struct fib_config *cfg)
410 {
411         __be32 addr;
412         int plen;
413
414         memset(cfg, 0, sizeof(*cfg));
415         cfg->fc_nlinfo.nl_net = net;
416
417         if (rt->rt_dst.sa_family != AF_INET)
418                 return -EAFNOSUPPORT;
419
420         /*
421          * Check mask for validity:
422          * a) it must be contiguous.
423          * b) destination must have all host bits clear.
424          * c) if application forgot to set correct family (AF_INET),
425          *    reject request unless it is absolutely clear i.e.
426          *    both family and mask are zero.
427          */
428         plen = 32;
429         addr = sk_extract_addr(&rt->rt_dst);
430         if (!(rt->rt_flags & RTF_HOST)) {
431                 __be32 mask = sk_extract_addr(&rt->rt_genmask);
432
433                 if (rt->rt_genmask.sa_family != AF_INET) {
434                         if (mask || rt->rt_genmask.sa_family)
435                                 return -EAFNOSUPPORT;
436                 }
437
438                 if (bad_mask(mask, addr))
439                         return -EINVAL;
440
441                 plen = inet_mask_len(mask);
442         }
443
444         cfg->fc_dst_len = plen;
445         cfg->fc_dst = addr;
446
447         if (cmd != SIOCDELRT) {
448                 cfg->fc_nlflags = NLM_F_CREATE;
449                 cfg->fc_protocol = RTPROT_BOOT;
450         }
451
452         if (rt->rt_metric)
453                 cfg->fc_priority = rt->rt_metric - 1;
454
455         if (rt->rt_flags & RTF_REJECT) {
456                 cfg->fc_scope = RT_SCOPE_HOST;
457                 cfg->fc_type = RTN_UNREACHABLE;
458                 return 0;
459         }
460
461         cfg->fc_scope = RT_SCOPE_NOWHERE;
462         cfg->fc_type = RTN_UNICAST;
463
464         if (rt->rt_dev) {
465                 char *colon;
466                 struct net_device *dev;
467                 char devname[IFNAMSIZ];
468
469                 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
470                         return -EFAULT;
471
472                 devname[IFNAMSIZ-1] = 0;
473                 colon = strchr(devname, ':');
474                 if (colon)
475                         *colon = 0;
476                 dev = __dev_get_by_name(net, devname);
477                 if (!dev)
478                         return -ENODEV;
479                 cfg->fc_oif = dev->ifindex;
480                 if (colon) {
481                         struct in_ifaddr *ifa;
482                         struct in_device *in_dev = __in_dev_get_rtnl(dev);
483                         if (!in_dev)
484                                 return -ENODEV;
485                         *colon = ':';
486                         for (ifa = in_dev->ifa_list; ifa; ifa = ifa->ifa_next)
487                                 if (strcmp(ifa->ifa_label, devname) == 0)
488                                         break;
489                         if (!ifa)
490                                 return -ENODEV;
491                         cfg->fc_prefsrc = ifa->ifa_local;
492                 }
493         }
494
495         addr = sk_extract_addr(&rt->rt_gateway);
496         if (rt->rt_gateway.sa_family == AF_INET && addr) {
497                 cfg->fc_gw = addr;
498                 if (rt->rt_flags & RTF_GATEWAY &&
499                     inet_addr_type(net, addr) == RTN_UNICAST)
500                         cfg->fc_scope = RT_SCOPE_UNIVERSE;
501         }
502
503         if (cmd == SIOCDELRT)
504                 return 0;
505
506         if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw)
507                 return -EINVAL;
508
509         if (cfg->fc_scope == RT_SCOPE_NOWHERE)
510                 cfg->fc_scope = RT_SCOPE_LINK;
511
512         if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
513                 struct nlattr *mx;
514                 int len = 0;
515
516                 mx = kzalloc(3 * nla_total_size(4), GFP_KERNEL);
517                 if (!mx)
518                         return -ENOMEM;
519
520                 if (rt->rt_flags & RTF_MTU)
521                         len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
522
523                 if (rt->rt_flags & RTF_WINDOW)
524                         len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
525
526                 if (rt->rt_flags & RTF_IRTT)
527                         len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
528
529                 cfg->fc_mx = mx;
530                 cfg->fc_mx_len = len;
531         }
532
533         return 0;
534 }
535
536 /*
537  * Handle IP routing ioctl calls.
538  * These are used to manipulate the routing tables
539  */
540 int ip_rt_ioctl(struct net *net, unsigned int cmd, void __user *arg)
541 {
542         struct fib_config cfg;
543         struct rtentry rt;
544         int err;
545
546         switch (cmd) {
547         case SIOCADDRT:         /* Add a route */
548         case SIOCDELRT:         /* Delete a route */
549                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
550                         return -EPERM;
551
552                 if (copy_from_user(&rt, arg, sizeof(rt)))
553                         return -EFAULT;
554
555                 rtnl_lock();
556                 err = rtentry_to_fib_config(net, cmd, &rt, &cfg);
557                 if (err == 0) {
558                         struct fib_table *tb;
559
560                         if (cmd == SIOCDELRT) {
561                                 tb = fib_get_table(net, cfg.fc_table);
562                                 if (tb)
563                                         err = fib_table_delete(tb, &cfg);
564                                 else
565                                         err = -ESRCH;
566                         } else {
567                                 tb = fib_new_table(net, cfg.fc_table);
568                                 if (tb)
569                                         err = fib_table_insert(tb, &cfg);
570                                 else
571                                         err = -ENOBUFS;
572                         }
573
574                         /* allocated by rtentry_to_fib_config() */
575                         kfree(cfg.fc_mx);
576                 }
577                 rtnl_unlock();
578                 return err;
579         }
580         return -EINVAL;
581 }
582
583 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
584         [RTA_DST]               = { .type = NLA_U32 },
585         [RTA_SRC]               = { .type = NLA_U32 },
586         [RTA_IIF]               = { .type = NLA_U32 },
587         [RTA_OIF]               = { .type = NLA_U32 },
588         [RTA_GATEWAY]           = { .type = NLA_U32 },
589         [RTA_PRIORITY]          = { .type = NLA_U32 },
590         [RTA_PREFSRC]           = { .type = NLA_U32 },
591         [RTA_METRICS]           = { .type = NLA_NESTED },
592         [RTA_MULTIPATH]         = { .len = sizeof(struct rtnexthop) },
593         [RTA_FLOW]              = { .type = NLA_U32 },
594 };
595
596 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
597                              struct nlmsghdr *nlh, struct fib_config *cfg)
598 {
599         struct nlattr *attr;
600         int err, remaining;
601         struct rtmsg *rtm;
602
603         err = nlmsg_validate(nlh, sizeof(*rtm), RTA_MAX, rtm_ipv4_policy);
604         if (err < 0)
605                 goto errout;
606
607         memset(cfg, 0, sizeof(*cfg));
608
609         rtm = nlmsg_data(nlh);
610         cfg->fc_dst_len = rtm->rtm_dst_len;
611         cfg->fc_tos = rtm->rtm_tos;
612         cfg->fc_table = rtm->rtm_table;
613         cfg->fc_protocol = rtm->rtm_protocol;
614         cfg->fc_scope = rtm->rtm_scope;
615         cfg->fc_type = rtm->rtm_type;
616         cfg->fc_flags = rtm->rtm_flags;
617         cfg->fc_nlflags = nlh->nlmsg_flags;
618
619         cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
620         cfg->fc_nlinfo.nlh = nlh;
621         cfg->fc_nlinfo.nl_net = net;
622
623         if (cfg->fc_type > RTN_MAX) {
624                 err = -EINVAL;
625                 goto errout;
626         }
627
628         nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
629                 switch (nla_type(attr)) {
630                 case RTA_DST:
631                         cfg->fc_dst = nla_get_be32(attr);
632                         break;
633                 case RTA_OIF:
634                         cfg->fc_oif = nla_get_u32(attr);
635                         break;
636                 case RTA_GATEWAY:
637                         cfg->fc_gw = nla_get_be32(attr);
638                         break;
639                 case RTA_PRIORITY:
640                         cfg->fc_priority = nla_get_u32(attr);
641                         break;
642                 case RTA_PREFSRC:
643                         cfg->fc_prefsrc = nla_get_be32(attr);
644                         break;
645                 case RTA_METRICS:
646                         cfg->fc_mx = nla_data(attr);
647                         cfg->fc_mx_len = nla_len(attr);
648                         break;
649                 case RTA_MULTIPATH:
650                         cfg->fc_mp = nla_data(attr);
651                         cfg->fc_mp_len = nla_len(attr);
652                         break;
653                 case RTA_FLOW:
654                         cfg->fc_flow = nla_get_u32(attr);
655                         break;
656                 case RTA_TABLE:
657                         cfg->fc_table = nla_get_u32(attr);
658                         break;
659                 }
660         }
661
662         return 0;
663 errout:
664         return err;
665 }
666
667 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
668 {
669         struct net *net = sock_net(skb->sk);
670         struct fib_config cfg;
671         struct fib_table *tb;
672         int err;
673
674         err = rtm_to_fib_config(net, skb, nlh, &cfg);
675         if (err < 0)
676                 goto errout;
677
678         tb = fib_get_table(net, cfg.fc_table);
679         if (!tb) {
680                 err = -ESRCH;
681                 goto errout;
682         }
683
684         err = fib_table_delete(tb, &cfg);
685 errout:
686         return err;
687 }
688
689 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
690 {
691         struct net *net = sock_net(skb->sk);
692         struct fib_config cfg;
693         struct fib_table *tb;
694         int err;
695
696         err = rtm_to_fib_config(net, skb, nlh, &cfg);
697         if (err < 0)
698                 goto errout;
699
700         tb = fib_new_table(net, cfg.fc_table);
701         if (!tb) {
702                 err = -ENOBUFS;
703                 goto errout;
704         }
705
706         err = fib_table_insert(tb, &cfg);
707 errout:
708         return err;
709 }
710
711 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
712 {
713         struct net *net = sock_net(skb->sk);
714         unsigned int h, s_h;
715         unsigned int e = 0, s_e;
716         struct fib_table *tb;
717         struct hlist_head *head;
718         int dumped = 0;
719
720         if (nlmsg_len(cb->nlh) >= sizeof(struct rtmsg) &&
721             ((struct rtmsg *) nlmsg_data(cb->nlh))->rtm_flags & RTM_F_CLONED)
722                 return skb->len;
723
724         s_h = cb->args[0];
725         s_e = cb->args[1];
726
727         rcu_read_lock();
728
729         for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
730                 e = 0;
731                 head = &net->ipv4.fib_table_hash[h];
732                 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
733                         if (e < s_e)
734                                 goto next;
735                         if (dumped)
736                                 memset(&cb->args[2], 0, sizeof(cb->args) -
737                                                  2 * sizeof(cb->args[0]));
738                         if (fib_table_dump(tb, skb, cb) < 0)
739                                 goto out;
740                         dumped = 1;
741 next:
742                         e++;
743                 }
744         }
745 out:
746         rcu_read_unlock();
747
748         cb->args[1] = e;
749         cb->args[0] = h;
750
751         return skb->len;
752 }
753
754 /* Prepare and feed intra-kernel routing request.
755  * Really, it should be netlink message, but :-( netlink
756  * can be not configured, so that we feed it directly
757  * to fib engine. It is legal, because all events occur
758  * only when netlink is already locked.
759  */
760 static void fib_magic(int cmd, int type, __be32 dst, int dst_len, struct in_ifaddr *ifa)
761 {
762         struct net *net = dev_net(ifa->ifa_dev->dev);
763         struct fib_table *tb;
764         struct fib_config cfg = {
765                 .fc_protocol = RTPROT_KERNEL,
766                 .fc_type = type,
767                 .fc_dst = dst,
768                 .fc_dst_len = dst_len,
769                 .fc_prefsrc = ifa->ifa_local,
770                 .fc_oif = ifa->ifa_dev->dev->ifindex,
771                 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
772                 .fc_nlinfo = {
773                         .nl_net = net,
774                 },
775         };
776
777         if (type == RTN_UNICAST)
778                 tb = fib_new_table(net, RT_TABLE_MAIN);
779         else
780                 tb = fib_new_table(net, RT_TABLE_LOCAL);
781
782         if (!tb)
783                 return;
784
785         cfg.fc_table = tb->tb_id;
786
787         if (type != RTN_LOCAL)
788                 cfg.fc_scope = RT_SCOPE_LINK;
789         else
790                 cfg.fc_scope = RT_SCOPE_HOST;
791
792         if (cmd == RTM_NEWROUTE)
793                 fib_table_insert(tb, &cfg);
794         else
795                 fib_table_delete(tb, &cfg);
796 }
797
798 void fib_add_ifaddr(struct in_ifaddr *ifa)
799 {
800         struct in_device *in_dev = ifa->ifa_dev;
801         struct net_device *dev = in_dev->dev;
802         struct in_ifaddr *prim = ifa;
803         __be32 mask = ifa->ifa_mask;
804         __be32 addr = ifa->ifa_local;
805         __be32 prefix = ifa->ifa_address & mask;
806
807         if (ifa->ifa_flags & IFA_F_SECONDARY) {
808                 prim = inet_ifa_byprefix(in_dev, prefix, mask);
809                 if (!prim) {
810                         pr_warn("%s: bug: prim == NULL\n", __func__);
811                         return;
812                 }
813         }
814
815         fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim);
816
817         if (!(dev->flags & IFF_UP))
818                 return;
819
820         /* Add broadcast address, if it is explicitly assigned. */
821         if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF))
822                 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
823
824         if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
825             (prefix != addr || ifa->ifa_prefixlen < 32)) {
826                 fib_magic(RTM_NEWROUTE,
827                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
828                           prefix, ifa->ifa_prefixlen, prim);
829
830                 /* Add network specific broadcasts, when it takes a sense */
831                 if (ifa->ifa_prefixlen < 31) {
832                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix, 32, prim);
833                         fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
834                                   32, prim);
835                 }
836         }
837 }
838
839 /* Delete primary or secondary address.
840  * Optionally, on secondary address promotion consider the addresses
841  * from subnet iprim as deleted, even if they are in device list.
842  * In this case the secondary ifa can be in device list.
843  */
844 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
845 {
846         struct in_device *in_dev = ifa->ifa_dev;
847         struct net_device *dev = in_dev->dev;
848         struct in_ifaddr *ifa1;
849         struct in_ifaddr *prim = ifa, *prim1 = NULL;
850         __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
851         __be32 any = ifa->ifa_address & ifa->ifa_mask;
852 #define LOCAL_OK        1
853 #define BRD_OK          2
854 #define BRD0_OK         4
855 #define BRD1_OK         8
856         unsigned int ok = 0;
857         int subnet = 0;         /* Primary network */
858         int gone = 1;           /* Address is missing */
859         int same_prefsrc = 0;   /* Another primary with same IP */
860
861         if (ifa->ifa_flags & IFA_F_SECONDARY) {
862                 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
863                 if (!prim) {
864                         pr_warn("%s: bug: prim == NULL\n", __func__);
865                         return;
866                 }
867                 if (iprim && iprim != prim) {
868                         pr_warn("%s: bug: iprim != prim\n", __func__);
869                         return;
870                 }
871         } else if (!ipv4_is_zeronet(any) &&
872                    (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
873                 fib_magic(RTM_DELROUTE,
874                           dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
875                           any, ifa->ifa_prefixlen, prim);
876                 subnet = 1;
877         }
878
879         /* Deletion is more complicated than add.
880          * We should take care of not to delete too much :-)
881          *
882          * Scan address list to be sure that addresses are really gone.
883          */
884
885         for (ifa1 = in_dev->ifa_list; ifa1; ifa1 = ifa1->ifa_next) {
886                 if (ifa1 == ifa) {
887                         /* promotion, keep the IP */
888                         gone = 0;
889                         continue;
890                 }
891                 /* Ignore IFAs from our subnet */
892                 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
893                     inet_ifa_match(ifa1->ifa_address, iprim))
894                         continue;
895
896                 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
897                 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
898                         /* Another address from our subnet? */
899                         if (ifa1->ifa_mask == prim->ifa_mask &&
900                             inet_ifa_match(ifa1->ifa_address, prim))
901                                 prim1 = prim;
902                         else {
903                                 /* We reached the secondaries, so
904                                  * same_prefsrc should be determined.
905                                  */
906                                 if (!same_prefsrc)
907                                         continue;
908                                 /* Search new prim1 if ifa1 is not
909                                  * using the current prim1
910                                  */
911                                 if (!prim1 ||
912                                     ifa1->ifa_mask != prim1->ifa_mask ||
913                                     !inet_ifa_match(ifa1->ifa_address, prim1))
914                                         prim1 = inet_ifa_byprefix(in_dev,
915                                                         ifa1->ifa_address,
916                                                         ifa1->ifa_mask);
917                                 if (!prim1)
918                                         continue;
919                                 if (prim1->ifa_local != prim->ifa_local)
920                                         continue;
921                         }
922                 } else {
923                         if (prim->ifa_local != ifa1->ifa_local)
924                                 continue;
925                         prim1 = ifa1;
926                         if (prim != prim1)
927                                 same_prefsrc = 1;
928                 }
929                 if (ifa->ifa_local == ifa1->ifa_local)
930                         ok |= LOCAL_OK;
931                 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
932                         ok |= BRD_OK;
933                 if (brd == ifa1->ifa_broadcast)
934                         ok |= BRD1_OK;
935                 if (any == ifa1->ifa_broadcast)
936                         ok |= BRD0_OK;
937                 /* primary has network specific broadcasts */
938                 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
939                         __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
940                         __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
941
942                         if (!ipv4_is_zeronet(any1)) {
943                                 if (ifa->ifa_broadcast == brd1 ||
944                                     ifa->ifa_broadcast == any1)
945                                         ok |= BRD_OK;
946                                 if (brd == brd1 || brd == any1)
947                                         ok |= BRD1_OK;
948                                 if (any == brd1 || any == any1)
949                                         ok |= BRD0_OK;
950                         }
951                 }
952         }
953
954         if (!(ok & BRD_OK))
955                 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32, prim);
956         if (subnet && ifa->ifa_prefixlen < 31) {
957                 if (!(ok & BRD1_OK))
958                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32, prim);
959                 if (!(ok & BRD0_OK))
960                         fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32, prim);
961         }
962         if (!(ok & LOCAL_OK)) {
963                 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim);
964
965                 /* Check, that this local address finally disappeared. */
966                 if (gone &&
967                     inet_addr_type(dev_net(dev), ifa->ifa_local) != RTN_LOCAL) {
968                         /* And the last, but not the least thing.
969                          * We must flush stray FIB entries.
970                          *
971                          * First of all, we scan fib_info list searching
972                          * for stray nexthop entries, then ignite fib_flush.
973                          */
974                         if (fib_sync_down_addr(dev_net(dev), ifa->ifa_local))
975                                 fib_flush(dev_net(dev));
976                 }
977         }
978 #undef LOCAL_OK
979 #undef BRD_OK
980 #undef BRD0_OK
981 #undef BRD1_OK
982 }
983
984 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
985 {
986
987         struct fib_result       res;
988         struct flowi4           fl4 = {
989                 .flowi4_mark = frn->fl_mark,
990                 .daddr = frn->fl_addr,
991                 .flowi4_tos = frn->fl_tos,
992                 .flowi4_scope = frn->fl_scope,
993         };
994         struct fib_table *tb;
995
996         rcu_read_lock();
997
998         tb = fib_get_table(net, frn->tb_id_in);
999
1000         frn->err = -ENOENT;
1001         if (tb) {
1002                 local_bh_disable();
1003
1004                 frn->tb_id = tb->tb_id;
1005                 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1006
1007                 if (!frn->err) {
1008                         frn->prefixlen = res.prefixlen;
1009                         frn->nh_sel = res.nh_sel;
1010                         frn->type = res.type;
1011                         frn->scope = res.scope;
1012                 }
1013                 local_bh_enable();
1014         }
1015
1016         rcu_read_unlock();
1017 }
1018
1019 static void nl_fib_input(struct sk_buff *skb)
1020 {
1021         struct net *net;
1022         struct fib_result_nl *frn;
1023         struct nlmsghdr *nlh;
1024         u32 portid;
1025
1026         net = sock_net(skb->sk);
1027         nlh = nlmsg_hdr(skb);
1028         if (skb->len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len ||
1029             nlmsg_len(nlh) < sizeof(*frn))
1030                 return;
1031
1032         skb = netlink_skb_clone(skb, GFP_KERNEL);
1033         if (!skb)
1034                 return;
1035         nlh = nlmsg_hdr(skb);
1036
1037         frn = (struct fib_result_nl *) nlmsg_data(nlh);
1038         nl_fib_lookup(net, frn);
1039
1040         portid = NETLINK_CB(skb).portid;      /* netlink portid */
1041         NETLINK_CB(skb).portid = 0;        /* from kernel */
1042         NETLINK_CB(skb).dst_group = 0;  /* unicast */
1043         netlink_unicast(net->ipv4.fibnl, skb, portid, MSG_DONTWAIT);
1044 }
1045
1046 static int __net_init nl_fib_lookup_init(struct net *net)
1047 {
1048         struct sock *sk;
1049         struct netlink_kernel_cfg cfg = {
1050                 .input  = nl_fib_input,
1051         };
1052
1053         sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1054         if (!sk)
1055                 return -EAFNOSUPPORT;
1056         net->ipv4.fibnl = sk;
1057         return 0;
1058 }
1059
1060 static void nl_fib_lookup_exit(struct net *net)
1061 {
1062         netlink_kernel_release(net->ipv4.fibnl);
1063         net->ipv4.fibnl = NULL;
1064 }
1065
1066 static void fib_disable_ip(struct net_device *dev, int force)
1067 {
1068         if (fib_sync_down_dev(dev, force))
1069                 fib_flush(dev_net(dev));
1070         rt_cache_flush(dev_net(dev));
1071         arp_ifdown(dev);
1072 }
1073
1074 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1075 {
1076         struct in_ifaddr *ifa = (struct in_ifaddr *)ptr;
1077         struct net_device *dev = ifa->ifa_dev->dev;
1078         struct net *net = dev_net(dev);
1079
1080         switch (event) {
1081         case NETDEV_UP:
1082                 fib_add_ifaddr(ifa);
1083 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1084                 fib_sync_up(dev);
1085 #endif
1086                 atomic_inc(&net->ipv4.dev_addr_genid);
1087                 rt_cache_flush(dev_net(dev));
1088                 break;
1089         case NETDEV_DOWN:
1090                 fib_del_ifaddr(ifa, NULL);
1091                 atomic_inc(&net->ipv4.dev_addr_genid);
1092                 if (!ifa->ifa_dev->ifa_list) {
1093                         /* Last address was deleted from this interface.
1094                          * Disable IP.
1095                          */
1096                         fib_disable_ip(dev, 1);
1097                 } else {
1098                         rt_cache_flush(dev_net(dev));
1099                 }
1100                 break;
1101         }
1102         return NOTIFY_DONE;
1103 }
1104
1105 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1106 {
1107         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1108         struct in_device *in_dev;
1109         struct net *net = dev_net(dev);
1110
1111         if (event == NETDEV_UNREGISTER) {
1112                 fib_disable_ip(dev, 2);
1113                 rt_flush_dev(dev);
1114                 return NOTIFY_DONE;
1115         }
1116
1117         in_dev = __in_dev_get_rtnl(dev);
1118         if (!in_dev)
1119                 return NOTIFY_DONE;
1120
1121         switch (event) {
1122         case NETDEV_UP:
1123                 for_ifa(in_dev) {
1124                         fib_add_ifaddr(ifa);
1125                 } endfor_ifa(in_dev);
1126 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1127                 fib_sync_up(dev);
1128 #endif
1129                 atomic_inc(&net->ipv4.dev_addr_genid);
1130                 rt_cache_flush(net);
1131                 break;
1132         case NETDEV_DOWN:
1133                 fib_disable_ip(dev, 0);
1134                 break;
1135         case NETDEV_CHANGEMTU:
1136         case NETDEV_CHANGE:
1137                 rt_cache_flush(net);
1138                 break;
1139         }
1140         return NOTIFY_DONE;
1141 }
1142
1143 static struct notifier_block fib_inetaddr_notifier = {
1144         .notifier_call = fib_inetaddr_event,
1145 };
1146
1147 static struct notifier_block fib_netdev_notifier = {
1148         .notifier_call = fib_netdev_event,
1149 };
1150
1151 static int __net_init ip_fib_net_init(struct net *net)
1152 {
1153         int err;
1154         size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1155
1156         /* Avoid false sharing : Use at least a full cache line */
1157         size = max_t(size_t, size, L1_CACHE_BYTES);
1158
1159         net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1160         if (!net->ipv4.fib_table_hash)
1161                 return -ENOMEM;
1162
1163         err = fib4_rules_init(net);
1164         if (err < 0)
1165                 goto fail;
1166         return 0;
1167
1168 fail:
1169         kfree(net->ipv4.fib_table_hash);
1170         return err;
1171 }
1172
1173 static void ip_fib_net_exit(struct net *net)
1174 {
1175         unsigned int i;
1176
1177         rtnl_lock();
1178 #ifdef CONFIG_IP_MULTIPLE_TABLES
1179         RCU_INIT_POINTER(net->ipv4.fib_local, NULL);
1180         RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1181         RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1182 #endif
1183         for (i = 0; i < FIB_TABLE_HASHSZ; i++) {
1184                 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1185                 struct hlist_node *tmp;
1186                 struct fib_table *tb;
1187
1188                 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1189                         hlist_del(&tb->tb_hlist);
1190                         fib_table_flush(tb);
1191                         fib_free_table(tb);
1192                 }
1193         }
1194
1195 #ifdef CONFIG_IP_MULTIPLE_TABLES
1196         fib4_rules_exit(net);
1197 #endif
1198         rtnl_unlock();
1199         kfree(net->ipv4.fib_table_hash);
1200 }
1201
1202 static int __net_init fib_net_init(struct net *net)
1203 {
1204         int error;
1205
1206 #ifdef CONFIG_IP_ROUTE_CLASSID
1207         net->ipv4.fib_num_tclassid_users = 0;
1208 #endif
1209         error = ip_fib_net_init(net);
1210         if (error < 0)
1211                 goto out;
1212         error = nl_fib_lookup_init(net);
1213         if (error < 0)
1214                 goto out_nlfl;
1215         error = fib_proc_init(net);
1216         if (error < 0)
1217                 goto out_proc;
1218 out:
1219         return error;
1220
1221 out_proc:
1222         nl_fib_lookup_exit(net);
1223 out_nlfl:
1224         ip_fib_net_exit(net);
1225         goto out;
1226 }
1227
1228 static void __net_exit fib_net_exit(struct net *net)
1229 {
1230         fib_proc_exit(net);
1231         nl_fib_lookup_exit(net);
1232         ip_fib_net_exit(net);
1233 }
1234
1235 static struct pernet_operations fib_net_ops = {
1236         .init = fib_net_init,
1237         .exit = fib_net_exit,
1238 };
1239
1240 void __init ip_fib_init(void)
1241 {
1242         rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, NULL);
1243         rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, NULL);
1244         rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib, NULL);
1245
1246         register_pernet_subsys(&fib_net_ops);
1247         register_netdevice_notifier(&fib_netdev_notifier);
1248         register_inetaddr_notifier(&fib_inetaddr_notifier);
1249
1250         fib_trie_init();
1251 }