1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * IPv4 Forwarding Information Base: FIB frontend.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 */
11
12 #include <linux/module.h>
13 #include <linux/uaccess.h>
14 #include <linux/bitops.h>
15 #include <linux/capability.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/string.h>
20 #include <linux/socket.h>
21 #include <linux/sockios.h>
22 #include <linux/errno.h>
23 #include <linux/in.h>
24 #include <linux/inet.h>
25 #include <linux/inetdevice.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_addr.h>
28 #include <linux/if_arp.h>
29 #include <linux/skbuff.h>
30 #include <linux/cache.h>
31 #include <linux/init.h>
32 #include <linux/list.h>
33 #include <linux/slab.h>
34
35 #include <net/inet_dscp.h>
36 #include <net/ip.h>
37 #include <net/protocol.h>
38 #include <net/route.h>
39 #include <net/tcp.h>
40 #include <net/sock.h>
41 #include <net/arp.h>
42 #include <net/ip_fib.h>
43 #include <net/nexthop.h>
44 #include <net/rtnetlink.h>
45 #include <net/xfrm.h>
46 #include <net/l3mdev.h>
47 #include <net/lwtunnel.h>
48 #include <trace/events/fib.h>
49
50 #ifndef CONFIG_IP_MULTIPLE_TABLES
51
fib4_rules_init(struct net * net)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
fib_new_table(struct net * net,u32 id)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 && !net->ipv4.fib_has_custom_rules)
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_MAIN:
96 rcu_assign_pointer(net->ipv4.fib_main, tb);
97 break;
98 case RT_TABLE_DEFAULT:
99 rcu_assign_pointer(net->ipv4.fib_default, tb);
100 break;
101 default:
102 break;
103 }
104
105 h = id & (FIB_TABLE_HASHSZ - 1);
106 hlist_add_head_rcu(&tb->tb_hlist, &net->ipv4.fib_table_hash[h]);
107 return tb;
108 }
109 EXPORT_SYMBOL_GPL(fib_new_table);
110
111 /* caller must hold either rtnl or rcu read lock */
fib_get_table(struct net * net,u32 id)112 struct fib_table *fib_get_table(struct net *net, u32 id)
113 {
114 struct fib_table *tb;
115 struct hlist_head *head;
116 unsigned int h;
117
118 if (id == 0)
119 id = RT_TABLE_MAIN;
120 h = id & (FIB_TABLE_HASHSZ - 1);
121
122 head = &net->ipv4.fib_table_hash[h];
123 hlist_for_each_entry_rcu(tb, head, tb_hlist,
124 lockdep_rtnl_is_held()) {
125 if (tb->tb_id == id)
126 return tb;
127 }
128 return NULL;
129 }
130 #endif /* CONFIG_IP_MULTIPLE_TABLES */
131
fib_replace_table(struct net * net,struct fib_table * old,struct fib_table * new)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_MAIN:
138 rcu_assign_pointer(net->ipv4.fib_main, new);
139 break;
140 case RT_TABLE_DEFAULT:
141 rcu_assign_pointer(net->ipv4.fib_default, new);
142 break;
143 default:
144 break;
145 }
146
147 #endif
148 /* replace the old table in the hlist */
149 hlist_replace_rcu(&old->tb_hlist, &new->tb_hlist);
150 }
151
fib_unmerge(struct net * net)152 int fib_unmerge(struct net *net)
153 {
154 struct fib_table *old, *new, *main_table;
155
156 /* attempt to fetch local table if it has been allocated */
157 old = fib_get_table(net, RT_TABLE_LOCAL);
158 if (!old)
159 return 0;
160
161 new = fib_trie_unmerge(old);
162 if (!new)
163 return -ENOMEM;
164
165 /* table is already unmerged */
166 if (new == old)
167 return 0;
168
169 /* replace merged table with clean table */
170 fib_replace_table(net, old, new);
171 fib_free_table(old);
172
173 /* attempt to fetch main table if it has been allocated */
174 main_table = fib_get_table(net, RT_TABLE_MAIN);
175 if (!main_table)
176 return 0;
177
178 /* flush local entries from main table */
179 fib_table_flush_external(main_table);
180
181 return 0;
182 }
183
fib_flush(struct net * net)184 void fib_flush(struct net *net)
185 {
186 int flushed = 0;
187 unsigned int h;
188
189 for (h = 0; h < FIB_TABLE_HASHSZ; h++) {
190 struct hlist_head *head = &net->ipv4.fib_table_hash[h];
191 struct hlist_node *tmp;
192 struct fib_table *tb;
193
194 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist)
195 flushed += fib_table_flush(net, tb, false);
196 }
197
198 if (flushed)
199 rt_cache_flush(net);
200 }
201
202 /*
203 * Find address type as if only "dev" was present in the system. If
204 * on_dev is NULL then all interfaces are taken into consideration.
205 */
__inet_dev_addr_type(struct net * net,const struct net_device * dev,__be32 addr,u32 tb_id)206 static inline unsigned int __inet_dev_addr_type(struct net *net,
207 const struct net_device *dev,
208 __be32 addr, u32 tb_id)
209 {
210 struct flowi4 fl4 = { .daddr = addr };
211 struct fib_result res;
212 unsigned int ret = RTN_BROADCAST;
213 struct fib_table *table;
214
215 if (ipv4_is_zeronet(addr) || ipv4_is_lbcast(addr))
216 return RTN_BROADCAST;
217 if (ipv4_is_multicast(addr))
218 return RTN_MULTICAST;
219
220 rcu_read_lock();
221
222 table = fib_get_table(net, tb_id);
223 if (table) {
224 ret = RTN_UNICAST;
225 if (!fib_table_lookup(table, &fl4, &res, FIB_LOOKUP_NOREF)) {
226 struct fib_nh_common *nhc = fib_info_nhc(res.fi, 0);
227
228 if (!dev || dev == nhc->nhc_dev)
229 ret = res.type;
230 }
231 }
232
233 rcu_read_unlock();
234 return ret;
235 }
236
inet_addr_type_table(struct net * net,__be32 addr,u32 tb_id)237 unsigned int inet_addr_type_table(struct net *net, __be32 addr, u32 tb_id)
238 {
239 return __inet_dev_addr_type(net, NULL, addr, tb_id);
240 }
241 EXPORT_SYMBOL(inet_addr_type_table);
242
inet_addr_type(struct net * net,__be32 addr)243 unsigned int inet_addr_type(struct net *net, __be32 addr)
244 {
245 return __inet_dev_addr_type(net, NULL, addr, RT_TABLE_LOCAL);
246 }
247 EXPORT_SYMBOL(inet_addr_type);
248
inet_dev_addr_type(struct net * net,const struct net_device * dev,__be32 addr)249 unsigned int inet_dev_addr_type(struct net *net, const struct net_device *dev,
250 __be32 addr)
251 {
252 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
253
254 return __inet_dev_addr_type(net, dev, addr, rt_table);
255 }
256 EXPORT_SYMBOL(inet_dev_addr_type);
257
258 /* inet_addr_type with dev == NULL but using the table from a dev
259 * if one is associated
260 */
inet_addr_type_dev_table(struct net * net,const struct net_device * dev,__be32 addr)261 unsigned int inet_addr_type_dev_table(struct net *net,
262 const struct net_device *dev,
263 __be32 addr)
264 {
265 u32 rt_table = l3mdev_fib_table(dev) ? : RT_TABLE_LOCAL;
266
267 return __inet_dev_addr_type(net, NULL, addr, rt_table);
268 }
269 EXPORT_SYMBOL(inet_addr_type_dev_table);
270
fib_compute_spec_dst(struct sk_buff * skb)271 __be32 fib_compute_spec_dst(struct sk_buff *skb)
272 {
273 struct net_device *dev = skb->dev;
274 struct in_device *in_dev;
275 struct fib_result res;
276 struct rtable *rt;
277 struct net *net;
278 int scope;
279
280 rt = skb_rtable(skb);
281 if ((rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST | RTCF_LOCAL)) ==
282 RTCF_LOCAL)
283 return ip_hdr(skb)->daddr;
284
285 in_dev = __in_dev_get_rcu(dev);
286
287 net = dev_net(dev);
288
289 scope = RT_SCOPE_UNIVERSE;
290 if (!ipv4_is_zeronet(ip_hdr(skb)->saddr)) {
291 bool vmark = in_dev && IN_DEV_SRC_VMARK(in_dev);
292 struct flowi4 fl4 = {
293 .flowi4_iif = LOOPBACK_IFINDEX,
294 .flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev),
295 .daddr = ip_hdr(skb)->saddr,
296 .flowi4_tos = inet_dscp_to_dsfield(ip4h_dscp(ip_hdr(skb))),
297 .flowi4_scope = scope,
298 .flowi4_mark = vmark ? skb->mark : 0,
299 };
300 if (!fib_lookup(net, &fl4, &res, 0))
301 return fib_result_prefsrc(net, &res);
302 } else {
303 scope = RT_SCOPE_LINK;
304 }
305
306 return inet_select_addr(dev, ip_hdr(skb)->saddr, scope);
307 }
308
fib_info_nh_uses_dev(struct fib_info * fi,const struct net_device * dev)309 bool fib_info_nh_uses_dev(struct fib_info *fi, const struct net_device *dev)
310 {
311 bool dev_match = false;
312 #ifdef CONFIG_IP_ROUTE_MULTIPATH
313 if (unlikely(fi->nh)) {
314 dev_match = nexthop_uses_dev(fi->nh, dev);
315 } else {
316 int ret;
317
318 for (ret = 0; ret < fib_info_num_path(fi); ret++) {
319 const struct fib_nh_common *nhc = fib_info_nhc(fi, ret);
320
321 if (nhc_l3mdev_matches_dev(nhc, dev)) {
322 dev_match = true;
323 break;
324 }
325 }
326 }
327 #else
328 if (fib_info_nhc(fi, 0)->nhc_dev == dev)
329 dev_match = true;
330 #endif
331
332 return dev_match;
333 }
334 EXPORT_SYMBOL_GPL(fib_info_nh_uses_dev);
335
336 /* Given (packet source, input interface) and optional (dst, oif, tos):
337 * - (main) check, that source is valid i.e. not broadcast or our local
338 * address.
339 * - figure out what "logical" interface this packet arrived
340 * and calculate "specific destination" address.
341 * - check, that packet arrived from expected physical interface.
342 * called with rcu_read_lock()
343 */
__fib_validate_source(struct sk_buff * skb,__be32 src,__be32 dst,dscp_t dscp,int oif,struct net_device * dev,int rpf,struct in_device * idev,u32 * itag)344 static int __fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
345 dscp_t dscp, int oif, struct net_device *dev,
346 int rpf, struct in_device *idev, u32 *itag)
347 {
348 struct net *net = dev_net(dev);
349 enum skb_drop_reason reason;
350 struct flow_keys flkeys;
351 int ret, no_addr;
352 struct fib_result res;
353 struct flowi4 fl4;
354 bool dev_match;
355
356 fl4.flowi4_oif = 0;
357 fl4.flowi4_l3mdev = l3mdev_master_ifindex_rcu(dev);
358 fl4.flowi4_iif = oif ? : LOOPBACK_IFINDEX;
359 fl4.daddr = src;
360 fl4.saddr = dst;
361 fl4.flowi4_tos = inet_dscp_to_dsfield(dscp);
362 fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
363 fl4.flowi4_tun_key.tun_id = 0;
364 fl4.flowi4_flags = 0;
365 fl4.flowi4_uid = sock_net_uid(net, NULL);
366 fl4.flowi4_multipath_hash = 0;
367
368 no_addr = idev->ifa_list == NULL;
369
370 fl4.flowi4_mark = IN_DEV_SRC_VMARK(idev) ? skb->mark : 0;
371 if (!fib4_rules_early_flow_dissect(net, skb, &fl4, &flkeys)) {
372 fl4.flowi4_proto = 0;
373 fl4.fl4_sport = 0;
374 fl4.fl4_dport = 0;
375 } else {
376 swap(fl4.fl4_sport, fl4.fl4_dport);
377 }
378
379 if (fib_lookup(net, &fl4, &res, 0))
380 goto last_resort;
381 if (res.type != RTN_UNICAST) {
382 if (res.type != RTN_LOCAL) {
383 reason = SKB_DROP_REASON_IP_INVALID_SOURCE;
384 goto e_inval;
385 } else if (!IN_DEV_ACCEPT_LOCAL(idev)) {
386 reason = SKB_DROP_REASON_IP_LOCAL_SOURCE;
387 goto e_inval;
388 }
389 }
390 fib_combine_itag(itag, &res);
391
392 dev_match = fib_info_nh_uses_dev(res.fi, dev);
393 /* This is not common, loopback packets retain skb_dst so normally they
394 * would not even hit this slow path.
395 */
396 dev_match = dev_match || (res.type == RTN_LOCAL &&
397 dev == net->loopback_dev);
398 if (dev_match) {
399 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
400 return ret;
401 }
402 if (no_addr)
403 goto last_resort;
404 if (rpf == 1)
405 goto e_rpf;
406 fl4.flowi4_oif = dev->ifindex;
407
408 ret = 0;
409 if (fib_lookup(net, &fl4, &res, FIB_LOOKUP_IGNORE_LINKSTATE) == 0) {
410 if (res.type == RTN_UNICAST)
411 ret = FIB_RES_NHC(res)->nhc_scope >= RT_SCOPE_HOST;
412 }
413 return ret;
414
415 last_resort:
416 if (rpf)
417 goto e_rpf;
418 *itag = 0;
419 return 0;
420
421 e_inval:
422 return -reason;
423 e_rpf:
424 return -SKB_DROP_REASON_IP_RPFILTER;
425 }
426
427 /* Ignore rp_filter for packets protected by IPsec. */
fib_validate_source(struct sk_buff * skb,__be32 src,__be32 dst,dscp_t dscp,int oif,struct net_device * dev,struct in_device * idev,u32 * itag)428 int fib_validate_source(struct sk_buff *skb, __be32 src, __be32 dst,
429 dscp_t dscp, int oif, struct net_device *dev,
430 struct in_device *idev, u32 *itag)
431 {
432 int r = secpath_exists(skb) ? 0 : IN_DEV_RPFILTER(idev);
433 struct net *net = dev_net(dev);
434
435 if (!r && !fib_num_tclassid_users(net) &&
436 (dev->ifindex != oif || !IN_DEV_TX_REDIRECTS(idev))) {
437 if (IN_DEV_ACCEPT_LOCAL(idev))
438 goto ok;
439 /* with custom local routes in place, checking local addresses
440 * only will be too optimistic, with custom rules, checking
441 * local addresses only can be too strict, e.g. due to vrf
442 */
443 if (net->ipv4.fib_has_custom_local_routes ||
444 fib4_has_custom_rules(net))
445 goto full_check;
446 /* Within the same container, it is regarded as a martian source,
447 * and the same host but different containers are not.
448 */
449 if (inet_lookup_ifaddr_rcu(net, src))
450 return -SKB_DROP_REASON_IP_LOCAL_SOURCE;
451
452 ok:
453 *itag = 0;
454 return 0;
455 }
456
457 full_check:
458 return __fib_validate_source(skb, src, dst, dscp, oif, dev, r, idev,
459 itag);
460 }
461
sk_extract_addr(struct sockaddr * addr)462 static inline __be32 sk_extract_addr(struct sockaddr *addr)
463 {
464 return ((struct sockaddr_in *) addr)->sin_addr.s_addr;
465 }
466
put_rtax(struct nlattr * mx,int len,int type,u32 value)467 static int put_rtax(struct nlattr *mx, int len, int type, u32 value)
468 {
469 struct nlattr *nla;
470
471 nla = (struct nlattr *) ((char *) mx + len);
472 nla->nla_type = type;
473 nla->nla_len = nla_attr_size(4);
474 *(u32 *) nla_data(nla) = value;
475
476 return len + nla_total_size(4);
477 }
478
rtentry_to_fib_config(struct net * net,int cmd,struct rtentry * rt,struct fib_config * cfg)479 static int rtentry_to_fib_config(struct net *net, int cmd, struct rtentry *rt,
480 struct fib_config *cfg)
481 {
482 __be32 addr;
483 int plen;
484
485 memset(cfg, 0, sizeof(*cfg));
486 cfg->fc_nlinfo.nl_net = net;
487
488 if (rt->rt_dst.sa_family != AF_INET)
489 return -EAFNOSUPPORT;
490
491 /*
492 * Check mask for validity:
493 * a) it must be contiguous.
494 * b) destination must have all host bits clear.
495 * c) if application forgot to set correct family (AF_INET),
496 * reject request unless it is absolutely clear i.e.
497 * both family and mask are zero.
498 */
499 plen = 32;
500 addr = sk_extract_addr(&rt->rt_dst);
501 if (!(rt->rt_flags & RTF_HOST)) {
502 __be32 mask = sk_extract_addr(&rt->rt_genmask);
503
504 if (rt->rt_genmask.sa_family != AF_INET) {
505 if (mask || rt->rt_genmask.sa_family)
506 return -EAFNOSUPPORT;
507 }
508
509 if (bad_mask(mask, addr))
510 return -EINVAL;
511
512 plen = inet_mask_len(mask);
513 }
514
515 cfg->fc_dst_len = plen;
516 cfg->fc_dst = addr;
517
518 if (cmd != SIOCDELRT) {
519 cfg->fc_nlflags = NLM_F_CREATE;
520 cfg->fc_protocol = RTPROT_BOOT;
521 }
522
523 if (rt->rt_metric)
524 cfg->fc_priority = rt->rt_metric - 1;
525
526 if (rt->rt_flags & RTF_REJECT) {
527 cfg->fc_scope = RT_SCOPE_HOST;
528 cfg->fc_type = RTN_UNREACHABLE;
529 return 0;
530 }
531
532 cfg->fc_scope = RT_SCOPE_NOWHERE;
533 cfg->fc_type = RTN_UNICAST;
534
535 if (rt->rt_dev) {
536 char *colon;
537 struct net_device *dev;
538 char devname[IFNAMSIZ];
539
540 if (copy_from_user(devname, rt->rt_dev, IFNAMSIZ-1))
541 return -EFAULT;
542
543 devname[IFNAMSIZ-1] = 0;
544 colon = strchr(devname, ':');
545 if (colon)
546 *colon = 0;
547 dev = __dev_get_by_name(net, devname);
548 if (!dev)
549 return -ENODEV;
550 cfg->fc_oif = dev->ifindex;
551 cfg->fc_table = l3mdev_fib_table(dev);
552 if (colon) {
553 const struct in_ifaddr *ifa;
554 struct in_device *in_dev;
555
556 in_dev = __in_dev_get_rtnl_net(dev);
557 if (!in_dev)
558 return -ENODEV;
559
560 *colon = ':';
561
562 in_dev_for_each_ifa_rtnl_net(net, ifa, in_dev) {
563 if (strcmp(ifa->ifa_label, devname) == 0)
564 break;
565 }
566
567 if (!ifa)
568 return -ENODEV;
569 cfg->fc_prefsrc = ifa->ifa_local;
570 }
571 }
572
573 addr = sk_extract_addr(&rt->rt_gateway);
574 if (rt->rt_gateway.sa_family == AF_INET && addr) {
575 unsigned int addr_type;
576
577 cfg->fc_gw4 = addr;
578 cfg->fc_gw_family = AF_INET;
579 addr_type = inet_addr_type_table(net, addr, cfg->fc_table);
580 if (rt->rt_flags & RTF_GATEWAY &&
581 addr_type == RTN_UNICAST)
582 cfg->fc_scope = RT_SCOPE_UNIVERSE;
583 }
584
585 if (!cfg->fc_table)
586 cfg->fc_table = RT_TABLE_MAIN;
587
588 if (cmd == SIOCDELRT)
589 return 0;
590
591 if (rt->rt_flags & RTF_GATEWAY && !cfg->fc_gw_family)
592 return -EINVAL;
593
594 if (cfg->fc_scope == RT_SCOPE_NOWHERE)
595 cfg->fc_scope = RT_SCOPE_LINK;
596
597 if (rt->rt_flags & (RTF_MTU | RTF_WINDOW | RTF_IRTT)) {
598 struct nlattr *mx;
599 int len = 0;
600
601 mx = kcalloc(3, nla_total_size(4), GFP_KERNEL);
602 if (!mx)
603 return -ENOMEM;
604
605 if (rt->rt_flags & RTF_MTU)
606 len = put_rtax(mx, len, RTAX_ADVMSS, rt->rt_mtu - 40);
607
608 if (rt->rt_flags & RTF_WINDOW)
609 len = put_rtax(mx, len, RTAX_WINDOW, rt->rt_window);
610
611 if (rt->rt_flags & RTF_IRTT)
612 len = put_rtax(mx, len, RTAX_RTT, rt->rt_irtt << 3);
613
614 cfg->fc_mx = mx;
615 cfg->fc_mx_len = len;
616 }
617
618 return 0;
619 }
620
621 /*
622 * Handle IP routing ioctl calls.
623 * These are used to manipulate the routing tables
624 */
ip_rt_ioctl(struct net * net,unsigned int cmd,struct rtentry * rt)625 int ip_rt_ioctl(struct net *net, unsigned int cmd, struct rtentry *rt)
626 {
627 struct fib_config cfg;
628 int err;
629
630 switch (cmd) {
631 case SIOCADDRT: /* Add a route */
632 case SIOCDELRT: /* Delete a route */
633 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
634 return -EPERM;
635
636 rtnl_net_lock(net);
637 err = rtentry_to_fib_config(net, cmd, rt, &cfg);
638 if (err == 0) {
639 struct fib_table *tb;
640
641 if (cmd == SIOCDELRT) {
642 tb = fib_get_table(net, cfg.fc_table);
643 if (tb)
644 err = fib_table_delete(net, tb, &cfg,
645 NULL);
646 else
647 err = -ESRCH;
648 } else {
649 tb = fib_new_table(net, cfg.fc_table);
650 if (tb)
651 err = fib_table_insert(net, tb,
652 &cfg, NULL);
653 else
654 err = -ENOBUFS;
655 }
656
657 /* allocated by rtentry_to_fib_config() */
658 kfree(cfg.fc_mx);
659 }
660 rtnl_net_unlock(net);
661 return err;
662 }
663 return -EINVAL;
664 }
665
666 const struct nla_policy rtm_ipv4_policy[RTA_MAX + 1] = {
667 [RTA_UNSPEC] = { .strict_start_type = RTA_DPORT + 1 },
668 [RTA_DST] = { .type = NLA_U32 },
669 [RTA_SRC] = { .type = NLA_U32 },
670 [RTA_IIF] = { .type = NLA_U32 },
671 [RTA_OIF] = { .type = NLA_U32 },
672 [RTA_GATEWAY] = { .type = NLA_U32 },
673 [RTA_PRIORITY] = { .type = NLA_U32 },
674 [RTA_PREFSRC] = { .type = NLA_U32 },
675 [RTA_METRICS] = { .type = NLA_NESTED },
676 [RTA_MULTIPATH] = { .len = sizeof(struct rtnexthop) },
677 [RTA_FLOW] = { .type = NLA_U32 },
678 [RTA_ENCAP_TYPE] = { .type = NLA_U16 },
679 [RTA_ENCAP] = { .type = NLA_NESTED },
680 [RTA_UID] = { .type = NLA_U32 },
681 [RTA_MARK] = { .type = NLA_U32 },
682 [RTA_TABLE] = { .type = NLA_U32 },
683 [RTA_IP_PROTO] = { .type = NLA_U8 },
684 [RTA_SPORT] = { .type = NLA_U16 },
685 [RTA_DPORT] = { .type = NLA_U16 },
686 [RTA_NH_ID] = { .type = NLA_U32 },
687 };
688
fib_gw_from_via(struct fib_config * cfg,struct nlattr * nla,struct netlink_ext_ack * extack)689 int fib_gw_from_via(struct fib_config *cfg, struct nlattr *nla,
690 struct netlink_ext_ack *extack)
691 {
692 struct rtvia *via;
693 int alen;
694
695 if (nla_len(nla) < offsetof(struct rtvia, rtvia_addr)) {
696 NL_SET_ERR_MSG(extack, "Invalid attribute length for RTA_VIA");
697 return -EINVAL;
698 }
699
700 via = nla_data(nla);
701 alen = nla_len(nla) - offsetof(struct rtvia, rtvia_addr);
702
703 switch (via->rtvia_family) {
704 case AF_INET:
705 if (alen != sizeof(__be32)) {
706 NL_SET_ERR_MSG(extack, "Invalid IPv4 address in RTA_VIA");
707 return -EINVAL;
708 }
709 cfg->fc_gw_family = AF_INET;
710 cfg->fc_gw4 = *((__be32 *)via->rtvia_addr);
711 break;
712 case AF_INET6:
713 #if IS_ENABLED(CONFIG_IPV6)
714 if (alen != sizeof(struct in6_addr)) {
715 NL_SET_ERR_MSG(extack, "Invalid IPv6 address in RTA_VIA");
716 return -EINVAL;
717 }
718 cfg->fc_gw_family = AF_INET6;
719 cfg->fc_gw6 = *((struct in6_addr *)via->rtvia_addr);
720 #else
721 NL_SET_ERR_MSG(extack, "IPv6 support not enabled in kernel");
722 return -EINVAL;
723 #endif
724 break;
725 default:
726 NL_SET_ERR_MSG(extack, "Unsupported address family in RTA_VIA");
727 return -EINVAL;
728 }
729
730 return 0;
731 }
732
rtm_to_fib_config(struct net * net,struct sk_buff * skb,struct nlmsghdr * nlh,struct fib_config * cfg,struct netlink_ext_ack * extack)733 static int rtm_to_fib_config(struct net *net, struct sk_buff *skb,
734 struct nlmsghdr *nlh, struct fib_config *cfg,
735 struct netlink_ext_ack *extack)
736 {
737 bool has_gw = false, has_via = false;
738 struct nlattr *attr;
739 int err, remaining;
740 struct rtmsg *rtm;
741
742 err = nlmsg_validate_deprecated(nlh, sizeof(*rtm), RTA_MAX,
743 rtm_ipv4_policy, extack);
744 if (err < 0)
745 goto errout;
746
747 memset(cfg, 0, sizeof(*cfg));
748
749 rtm = nlmsg_data(nlh);
750
751 if (!inet_validate_dscp(rtm->rtm_tos)) {
752 NL_SET_ERR_MSG(extack,
753 "Invalid dsfield (tos): ECN bits must be 0");
754 err = -EINVAL;
755 goto errout;
756 }
757 cfg->fc_dscp = inet_dsfield_to_dscp(rtm->rtm_tos);
758
759 cfg->fc_dst_len = rtm->rtm_dst_len;
760 cfg->fc_table = rtm->rtm_table;
761 cfg->fc_protocol = rtm->rtm_protocol;
762 cfg->fc_scope = rtm->rtm_scope;
763 cfg->fc_type = rtm->rtm_type;
764 cfg->fc_flags = rtm->rtm_flags;
765 cfg->fc_nlflags = nlh->nlmsg_flags;
766
767 cfg->fc_nlinfo.portid = NETLINK_CB(skb).portid;
768 cfg->fc_nlinfo.nlh = nlh;
769 cfg->fc_nlinfo.nl_net = net;
770
771 if (cfg->fc_type > RTN_MAX) {
772 NL_SET_ERR_MSG(extack, "Invalid route type");
773 err = -EINVAL;
774 goto errout;
775 }
776
777 nlmsg_for_each_attr(attr, nlh, sizeof(struct rtmsg), remaining) {
778 switch (nla_type(attr)) {
779 case RTA_DST:
780 cfg->fc_dst = nla_get_be32(attr);
781 break;
782 case RTA_OIF:
783 cfg->fc_oif = nla_get_u32(attr);
784 break;
785 case RTA_GATEWAY:
786 has_gw = true;
787 cfg->fc_gw4 = nla_get_be32(attr);
788 if (cfg->fc_gw4)
789 cfg->fc_gw_family = AF_INET;
790 break;
791 case RTA_VIA:
792 has_via = true;
793 err = fib_gw_from_via(cfg, attr, extack);
794 if (err)
795 goto errout;
796 break;
797 case RTA_PRIORITY:
798 cfg->fc_priority = nla_get_u32(attr);
799 break;
800 case RTA_PREFSRC:
801 cfg->fc_prefsrc = nla_get_be32(attr);
802 break;
803 case RTA_METRICS:
804 cfg->fc_mx = nla_data(attr);
805 cfg->fc_mx_len = nla_len(attr);
806 break;
807 case RTA_MULTIPATH:
808 err = lwtunnel_valid_encap_type_attr(nla_data(attr),
809 nla_len(attr),
810 extack, false);
811 if (err < 0)
812 goto errout;
813 cfg->fc_mp = nla_data(attr);
814 cfg->fc_mp_len = nla_len(attr);
815 break;
816 case RTA_FLOW:
817 cfg->fc_flow = nla_get_u32(attr);
818 break;
819 case RTA_TABLE:
820 cfg->fc_table = nla_get_u32(attr);
821 break;
822 case RTA_ENCAP:
823 cfg->fc_encap = attr;
824 break;
825 case RTA_ENCAP_TYPE:
826 cfg->fc_encap_type = nla_get_u16(attr);
827 err = lwtunnel_valid_encap_type(cfg->fc_encap_type,
828 extack, false);
829 if (err < 0)
830 goto errout;
831 break;
832 case RTA_NH_ID:
833 cfg->fc_nh_id = nla_get_u32(attr);
834 break;
835 }
836 }
837
838 if (cfg->fc_dst_len > 32) {
839 NL_SET_ERR_MSG(extack, "Invalid prefix length");
840 err = -EINVAL;
841 goto errout;
842 }
843
844 if (cfg->fc_dst_len < 32 && (ntohl(cfg->fc_dst) << cfg->fc_dst_len)) {
845 NL_SET_ERR_MSG(extack, "Invalid prefix for given prefix length");
846 err = -EINVAL;
847 goto errout;
848 }
849
850 if (cfg->fc_nh_id) {
851 if (cfg->fc_oif || cfg->fc_gw_family ||
852 cfg->fc_encap || cfg->fc_mp) {
853 NL_SET_ERR_MSG(extack,
854 "Nexthop specification and nexthop id are mutually exclusive");
855 err = -EINVAL;
856 goto errout;
857 }
858 }
859
860 if (has_gw && has_via) {
861 NL_SET_ERR_MSG(extack,
862 "Nexthop configuration can not contain both GATEWAY and VIA");
863 err = -EINVAL;
864 goto errout;
865 }
866
867 if (!cfg->fc_table)
868 cfg->fc_table = RT_TABLE_MAIN;
869
870 return 0;
871 errout:
872 return err;
873 }
874
inet_rtm_delroute(struct sk_buff * skb,struct nlmsghdr * nlh,struct netlink_ext_ack * extack)875 static int inet_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
876 struct netlink_ext_ack *extack)
877 {
878 struct net *net = sock_net(skb->sk);
879 struct fib_config cfg;
880 struct fib_table *tb;
881 int err;
882
883 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
884 if (err < 0)
885 goto errout;
886
887 rtnl_net_lock(net);
888
889 if (cfg.fc_nh_id && !nexthop_find_by_id(net, cfg.fc_nh_id)) {
890 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
891 err = -EINVAL;
892 goto unlock;
893 }
894
895 tb = fib_get_table(net, cfg.fc_table);
896 if (!tb) {
897 NL_SET_ERR_MSG(extack, "FIB table does not exist");
898 err = -ESRCH;
899 goto unlock;
900 }
901
902 err = fib_table_delete(net, tb, &cfg, extack);
903 unlock:
904 rtnl_net_unlock(net);
905 errout:
906 return err;
907 }
908
inet_rtm_newroute(struct sk_buff * skb,struct nlmsghdr * nlh,struct netlink_ext_ack * extack)909 static int inet_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
910 struct netlink_ext_ack *extack)
911 {
912 struct net *net = sock_net(skb->sk);
913 struct fib_config cfg;
914 struct fib_table *tb;
915 int err;
916
917 err = rtm_to_fib_config(net, skb, nlh, &cfg, extack);
918 if (err < 0)
919 goto errout;
920
921 rtnl_net_lock(net);
922
923 tb = fib_new_table(net, cfg.fc_table);
924 if (!tb) {
925 err = -ENOBUFS;
926 goto unlock;
927 }
928
929 err = fib_table_insert(net, tb, &cfg, extack);
930 if (!err && cfg.fc_type == RTN_LOCAL)
931 net->ipv4.fib_has_custom_local_routes = true;
932
933 unlock:
934 rtnl_net_unlock(net);
935 errout:
936 return err;
937 }
938
ip_valid_fib_dump_req(struct net * net,const struct nlmsghdr * nlh,struct fib_dump_filter * filter,struct netlink_callback * cb)939 int ip_valid_fib_dump_req(struct net *net, const struct nlmsghdr *nlh,
940 struct fib_dump_filter *filter,
941 struct netlink_callback *cb)
942 {
943 struct netlink_ext_ack *extack = cb->extack;
944 struct nlattr *tb[RTA_MAX + 1];
945 struct rtmsg *rtm;
946 int err, i;
947
948 if (filter->rtnl_held)
949 ASSERT_RTNL();
950
951 if (nlh->nlmsg_len < nlmsg_msg_size(sizeof(*rtm))) {
952 NL_SET_ERR_MSG(extack, "Invalid header for FIB dump request");
953 return -EINVAL;
954 }
955
956 rtm = nlmsg_data(nlh);
957 if (rtm->rtm_dst_len || rtm->rtm_src_len || rtm->rtm_tos ||
958 rtm->rtm_scope) {
959 NL_SET_ERR_MSG(extack, "Invalid values in header for FIB dump request");
960 return -EINVAL;
961 }
962
963 if (rtm->rtm_flags & ~(RTM_F_CLONED | RTM_F_PREFIX)) {
964 NL_SET_ERR_MSG(extack, "Invalid flags for FIB dump request");
965 return -EINVAL;
966 }
967 if (rtm->rtm_flags & RTM_F_CLONED)
968 filter->dump_routes = false;
969 else
970 filter->dump_exceptions = false;
971
972 filter->flags = rtm->rtm_flags;
973 filter->protocol = rtm->rtm_protocol;
974 filter->rt_type = rtm->rtm_type;
975 filter->table_id = rtm->rtm_table;
976
977 err = nlmsg_parse_deprecated_strict(nlh, sizeof(*rtm), tb, RTA_MAX,
978 rtm_ipv4_policy, extack);
979 if (err < 0)
980 return err;
981
982 for (i = 0; i <= RTA_MAX; ++i) {
983 int ifindex;
984
985 if (!tb[i])
986 continue;
987
988 switch (i) {
989 case RTA_TABLE:
990 filter->table_id = nla_get_u32(tb[i]);
991 break;
992 case RTA_OIF:
993 ifindex = nla_get_u32(tb[i]);
994 if (filter->rtnl_held)
995 filter->dev = __dev_get_by_index(net, ifindex);
996 else
997 filter->dev = dev_get_by_index_rcu(net, ifindex);
998 if (!filter->dev)
999 return -ENODEV;
1000 break;
1001 default:
1002 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
1003 return -EINVAL;
1004 }
1005 }
1006
1007 if (filter->flags || filter->protocol || filter->rt_type ||
1008 filter->table_id || filter->dev) {
1009 filter->filter_set = 1;
1010 cb->answer_flags = NLM_F_DUMP_FILTERED;
1011 }
1012
1013 return 0;
1014 }
1015 EXPORT_SYMBOL_GPL(ip_valid_fib_dump_req);
1016
inet_dump_fib(struct sk_buff * skb,struct netlink_callback * cb)1017 static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
1018 {
1019 struct fib_dump_filter filter = {
1020 .dump_routes = true,
1021 .dump_exceptions = true,
1022 .rtnl_held = false,
1023 };
1024 const struct nlmsghdr *nlh = cb->nlh;
1025 struct net *net = sock_net(skb->sk);
1026 unsigned int h, s_h;
1027 unsigned int e = 0, s_e;
1028 struct fib_table *tb;
1029 struct hlist_head *head;
1030 int dumped = 0, err = 0;
1031
1032 rcu_read_lock();
1033 if (cb->strict_check) {
1034 err = ip_valid_fib_dump_req(net, nlh, &filter, cb);
1035 if (err < 0)
1036 goto unlock;
1037 } else if (nlmsg_len(nlh) >= sizeof(struct rtmsg)) {
1038 struct rtmsg *rtm = nlmsg_data(nlh);
1039
1040 filter.flags = rtm->rtm_flags & (RTM_F_PREFIX | RTM_F_CLONED);
1041 }
1042
1043 /* ipv4 does not use prefix flag */
1044 if (filter.flags & RTM_F_PREFIX)
1045 goto unlock;
1046
1047 if (filter.table_id) {
1048 tb = fib_get_table(net, filter.table_id);
1049 if (!tb) {
1050 if (rtnl_msg_family(cb->nlh) != PF_INET)
1051 goto unlock;
1052
1053 NL_SET_ERR_MSG(cb->extack, "ipv4: FIB table does not exist");
1054 err = -ENOENT;
1055 goto unlock;
1056 }
1057 err = fib_table_dump(tb, skb, cb, &filter);
1058 goto unlock;
1059 }
1060
1061 s_h = cb->args[0];
1062 s_e = cb->args[1];
1063
1064 err = 0;
1065 for (h = s_h; h < FIB_TABLE_HASHSZ; h++, s_e = 0) {
1066 e = 0;
1067 head = &net->ipv4.fib_table_hash[h];
1068 hlist_for_each_entry_rcu(tb, head, tb_hlist) {
1069 if (e < s_e)
1070 goto next;
1071 if (dumped)
1072 memset(&cb->args[2], 0, sizeof(cb->args) -
1073 2 * sizeof(cb->args[0]));
1074 err = fib_table_dump(tb, skb, cb, &filter);
1075 if (err < 0)
1076 goto out;
1077 dumped = 1;
1078 next:
1079 e++;
1080 }
1081 }
1082 out:
1083
1084 cb->args[1] = e;
1085 cb->args[0] = h;
1086
1087 unlock:
1088 rcu_read_unlock();
1089 return err;
1090 }
1091
1092 /* Prepare and feed intra-kernel routing request.
1093 * Really, it should be netlink message, but :-( netlink
1094 * can be not configured, so that we feed it directly
1095 * to fib engine. It is legal, because all events occur
1096 * only when netlink is already locked.
1097 */
fib_magic(int cmd,int type,__be32 dst,int dst_len,struct in_ifaddr * ifa,u32 rt_priority)1098 static void fib_magic(int cmd, int type, __be32 dst, int dst_len,
1099 struct in_ifaddr *ifa, u32 rt_priority)
1100 {
1101 struct net *net = dev_net(ifa->ifa_dev->dev);
1102 u32 tb_id = l3mdev_fib_table(ifa->ifa_dev->dev);
1103 struct fib_table *tb;
1104 struct fib_config cfg = {
1105 .fc_protocol = RTPROT_KERNEL,
1106 .fc_type = type,
1107 .fc_dst = dst,
1108 .fc_dst_len = dst_len,
1109 .fc_priority = rt_priority,
1110 .fc_prefsrc = ifa->ifa_local,
1111 .fc_oif = ifa->ifa_dev->dev->ifindex,
1112 .fc_nlflags = NLM_F_CREATE | NLM_F_APPEND,
1113 .fc_nlinfo = {
1114 .nl_net = net,
1115 },
1116 };
1117
1118 if (!tb_id)
1119 tb_id = (type == RTN_UNICAST) ? RT_TABLE_MAIN : RT_TABLE_LOCAL;
1120
1121 tb = fib_new_table(net, tb_id);
1122 if (!tb)
1123 return;
1124
1125 cfg.fc_table = tb->tb_id;
1126
1127 if (type != RTN_LOCAL)
1128 cfg.fc_scope = RT_SCOPE_LINK;
1129 else
1130 cfg.fc_scope = RT_SCOPE_HOST;
1131
1132 if (cmd == RTM_NEWROUTE)
1133 fib_table_insert(net, tb, &cfg, NULL);
1134 else
1135 fib_table_delete(net, tb, &cfg, NULL);
1136 }
1137
fib_add_ifaddr(struct in_ifaddr * ifa)1138 void fib_add_ifaddr(struct in_ifaddr *ifa)
1139 {
1140 struct in_device *in_dev = ifa->ifa_dev;
1141 struct net_device *dev = in_dev->dev;
1142 struct in_ifaddr *prim = ifa;
1143 __be32 mask = ifa->ifa_mask;
1144 __be32 addr = ifa->ifa_local;
1145 __be32 prefix = ifa->ifa_address & mask;
1146
1147 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1148 prim = inet_ifa_byprefix(in_dev, prefix, mask);
1149 if (!prim) {
1150 pr_warn("%s: bug: prim == NULL\n", __func__);
1151 return;
1152 }
1153 }
1154
1155 fib_magic(RTM_NEWROUTE, RTN_LOCAL, addr, 32, prim, 0);
1156
1157 if (!(dev->flags & IFF_UP))
1158 return;
1159
1160 /* Add broadcast address, if it is explicitly assigned. */
1161 if (ifa->ifa_broadcast && ifa->ifa_broadcast != htonl(0xFFFFFFFF)) {
1162 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1163 prim, 0);
1164 arp_invalidate(dev, ifa->ifa_broadcast, false);
1165 }
1166
1167 if (!ipv4_is_zeronet(prefix) && !(ifa->ifa_flags & IFA_F_SECONDARY) &&
1168 (prefix != addr || ifa->ifa_prefixlen < 32)) {
1169 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1170 fib_magic(RTM_NEWROUTE,
1171 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1172 prefix, ifa->ifa_prefixlen, prim,
1173 ifa->ifa_rt_priority);
1174
1175 /* Add the network broadcast address, when it makes sense */
1176 if (ifa->ifa_prefixlen < 31) {
1177 fib_magic(RTM_NEWROUTE, RTN_BROADCAST, prefix | ~mask,
1178 32, prim, 0);
1179 arp_invalidate(dev, prefix | ~mask, false);
1180 }
1181 }
1182 }
1183
fib_modify_prefix_metric(struct in_ifaddr * ifa,u32 new_metric)1184 void fib_modify_prefix_metric(struct in_ifaddr *ifa, u32 new_metric)
1185 {
1186 __be32 prefix = ifa->ifa_address & ifa->ifa_mask;
1187 struct in_device *in_dev = ifa->ifa_dev;
1188 struct net_device *dev = in_dev->dev;
1189
1190 if (!(dev->flags & IFF_UP) ||
1191 ifa->ifa_flags & (IFA_F_SECONDARY | IFA_F_NOPREFIXROUTE) ||
1192 ipv4_is_zeronet(prefix) ||
1193 (prefix == ifa->ifa_local && ifa->ifa_prefixlen == 32))
1194 return;
1195
1196 /* add the new */
1197 fib_magic(RTM_NEWROUTE,
1198 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1199 prefix, ifa->ifa_prefixlen, ifa, new_metric);
1200
1201 /* delete the old */
1202 fib_magic(RTM_DELROUTE,
1203 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1204 prefix, ifa->ifa_prefixlen, ifa, ifa->ifa_rt_priority);
1205 }
1206
1207 /* Delete primary or secondary address.
1208 * Optionally, on secondary address promotion consider the addresses
1209 * from subnet iprim as deleted, even if they are in device list.
1210 * In this case the secondary ifa can be in device list.
1211 */
fib_del_ifaddr(struct in_ifaddr * ifa,struct in_ifaddr * iprim)1212 void fib_del_ifaddr(struct in_ifaddr *ifa, struct in_ifaddr *iprim)
1213 {
1214 struct in_device *in_dev = ifa->ifa_dev;
1215 struct net_device *dev = in_dev->dev;
1216 struct in_ifaddr *ifa1;
1217 struct in_ifaddr *prim = ifa, *prim1 = NULL;
1218 __be32 brd = ifa->ifa_address | ~ifa->ifa_mask;
1219 __be32 any = ifa->ifa_address & ifa->ifa_mask;
1220 #define LOCAL_OK 1
1221 #define BRD_OK 2
1222 #define BRD0_OK 4
1223 #define BRD1_OK 8
1224 unsigned int ok = 0;
1225 int subnet = 0; /* Primary network */
1226 int gone = 1; /* Address is missing */
1227 int same_prefsrc = 0; /* Another primary with same IP */
1228
1229 if (ifa->ifa_flags & IFA_F_SECONDARY) {
1230 prim = inet_ifa_byprefix(in_dev, any, ifa->ifa_mask);
1231 if (!prim) {
1232 /* if the device has been deleted, we don't perform
1233 * address promotion
1234 */
1235 if (!in_dev->dead)
1236 pr_warn("%s: bug: prim == NULL\n", __func__);
1237 return;
1238 }
1239 if (iprim && iprim != prim) {
1240 pr_warn("%s: bug: iprim != prim\n", __func__);
1241 return;
1242 }
1243 } else if (!ipv4_is_zeronet(any) &&
1244 (any != ifa->ifa_local || ifa->ifa_prefixlen < 32)) {
1245 if (!(ifa->ifa_flags & IFA_F_NOPREFIXROUTE))
1246 fib_magic(RTM_DELROUTE,
1247 dev->flags & IFF_LOOPBACK ? RTN_LOCAL : RTN_UNICAST,
1248 any, ifa->ifa_prefixlen, prim, 0);
1249 subnet = 1;
1250 }
1251
1252 if (in_dev->dead)
1253 goto no_promotions;
1254
1255 /* Deletion is more complicated than add.
1256 * We should take care of not to delete too much :-)
1257 *
1258 * Scan address list to be sure that addresses are really gone.
1259 */
1260 rcu_read_lock();
1261 in_dev_for_each_ifa_rcu(ifa1, in_dev) {
1262 if (ifa1 == ifa) {
1263 /* promotion, keep the IP */
1264 gone = 0;
1265 continue;
1266 }
1267 /* Ignore IFAs from our subnet */
1268 if (iprim && ifa1->ifa_mask == iprim->ifa_mask &&
1269 inet_ifa_match(ifa1->ifa_address, iprim))
1270 continue;
1271
1272 /* Ignore ifa1 if it uses different primary IP (prefsrc) */
1273 if (ifa1->ifa_flags & IFA_F_SECONDARY) {
1274 /* Another address from our subnet? */
1275 if (ifa1->ifa_mask == prim->ifa_mask &&
1276 inet_ifa_match(ifa1->ifa_address, prim))
1277 prim1 = prim;
1278 else {
1279 /* We reached the secondaries, so
1280 * same_prefsrc should be determined.
1281 */
1282 if (!same_prefsrc)
1283 continue;
1284 /* Search new prim1 if ifa1 is not
1285 * using the current prim1
1286 */
1287 if (!prim1 ||
1288 ifa1->ifa_mask != prim1->ifa_mask ||
1289 !inet_ifa_match(ifa1->ifa_address, prim1))
1290 prim1 = inet_ifa_byprefix(in_dev,
1291 ifa1->ifa_address,
1292 ifa1->ifa_mask);
1293 if (!prim1)
1294 continue;
1295 if (prim1->ifa_local != prim->ifa_local)
1296 continue;
1297 }
1298 } else {
1299 if (prim->ifa_local != ifa1->ifa_local)
1300 continue;
1301 prim1 = ifa1;
1302 if (prim != prim1)
1303 same_prefsrc = 1;
1304 }
1305 if (ifa->ifa_local == ifa1->ifa_local)
1306 ok |= LOCAL_OK;
1307 if (ifa->ifa_broadcast == ifa1->ifa_broadcast)
1308 ok |= BRD_OK;
1309 if (brd == ifa1->ifa_broadcast)
1310 ok |= BRD1_OK;
1311 if (any == ifa1->ifa_broadcast)
1312 ok |= BRD0_OK;
1313 /* primary has network specific broadcasts */
1314 if (prim1 == ifa1 && ifa1->ifa_prefixlen < 31) {
1315 __be32 brd1 = ifa1->ifa_address | ~ifa1->ifa_mask;
1316 __be32 any1 = ifa1->ifa_address & ifa1->ifa_mask;
1317
1318 if (!ipv4_is_zeronet(any1)) {
1319 if (ifa->ifa_broadcast == brd1 ||
1320 ifa->ifa_broadcast == any1)
1321 ok |= BRD_OK;
1322 if (brd == brd1 || brd == any1)
1323 ok |= BRD1_OK;
1324 if (any == brd1 || any == any1)
1325 ok |= BRD0_OK;
1326 }
1327 }
1328 }
1329 rcu_read_unlock();
1330
1331 no_promotions:
1332 if (!(ok & BRD_OK))
1333 fib_magic(RTM_DELROUTE, RTN_BROADCAST, ifa->ifa_broadcast, 32,
1334 prim, 0);
1335 if (subnet && ifa->ifa_prefixlen < 31) {
1336 if (!(ok & BRD1_OK))
1337 fib_magic(RTM_DELROUTE, RTN_BROADCAST, brd, 32,
1338 prim, 0);
1339 if (!(ok & BRD0_OK))
1340 fib_magic(RTM_DELROUTE, RTN_BROADCAST, any, 32,
1341 prim, 0);
1342 }
1343 if (!(ok & LOCAL_OK)) {
1344 unsigned int addr_type;
1345
1346 fib_magic(RTM_DELROUTE, RTN_LOCAL, ifa->ifa_local, 32, prim, 0);
1347
1348 /* Check, that this local address finally disappeared. */
1349 addr_type = inet_addr_type_dev_table(dev_net(dev), dev,
1350 ifa->ifa_local);
1351 if (gone && addr_type != RTN_LOCAL) {
1352 /* And the last, but not the least thing.
1353 * We must flush stray FIB entries.
1354 *
1355 * First of all, we scan fib_info list searching
1356 * for stray nexthop entries, then ignite fib_flush.
1357 */
1358 if (fib_sync_down_addr(dev, ifa->ifa_local))
1359 fib_flush(dev_net(dev));
1360 }
1361 }
1362 #undef LOCAL_OK
1363 #undef BRD_OK
1364 #undef BRD0_OK
1365 #undef BRD1_OK
1366 }
1367
nl_fib_lookup(struct net * net,struct fib_result_nl * frn)1368 static void nl_fib_lookup(struct net *net, struct fib_result_nl *frn)
1369 {
1370
1371 struct fib_result res;
1372 struct flowi4 fl4 = {
1373 .flowi4_mark = frn->fl_mark,
1374 .daddr = frn->fl_addr,
1375 .flowi4_tos = frn->fl_tos & INET_DSCP_MASK,
1376 .flowi4_scope = frn->fl_scope,
1377 };
1378 struct fib_table *tb;
1379
1380 rcu_read_lock();
1381
1382 tb = fib_get_table(net, frn->tb_id_in);
1383
1384 frn->err = -ENOENT;
1385 if (tb) {
1386 local_bh_disable();
1387
1388 frn->tb_id = tb->tb_id;
1389 frn->err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
1390
1391 if (!frn->err) {
1392 frn->prefixlen = res.prefixlen;
1393 frn->nh_sel = res.nh_sel;
1394 frn->type = res.type;
1395 frn->scope = res.scope;
1396 }
1397 local_bh_enable();
1398 }
1399
1400 rcu_read_unlock();
1401 }
1402
nl_fib_input(struct sk_buff * skb)1403 static void nl_fib_input(struct sk_buff *skb)
1404 {
1405 struct net *net;
1406 struct fib_result_nl *frn;
1407 struct nlmsghdr *nlh;
1408 u32 portid;
1409
1410 net = sock_net(skb->sk);
1411 nlh = nlmsg_hdr(skb);
1412 if (skb->len < nlmsg_total_size(sizeof(*frn)) ||
1413 skb->len < nlh->nlmsg_len ||
1414 nlmsg_len(nlh) < sizeof(*frn))
1415 return;
1416
1417 skb = netlink_skb_clone(skb, GFP_KERNEL);
1418 if (!skb)
1419 return;
1420 nlh = nlmsg_hdr(skb);
1421
1422 frn = nlmsg_data(nlh);
1423 nl_fib_lookup(net, frn);
1424
1425 portid = NETLINK_CB(skb).portid; /* netlink portid */
1426 NETLINK_CB(skb).portid = 0; /* from kernel */
1427 NETLINK_CB(skb).dst_group = 0; /* unicast */
1428 nlmsg_unicast(net->ipv4.fibnl, skb, portid);
1429 }
1430
nl_fib_lookup_init(struct net * net)1431 static int __net_init nl_fib_lookup_init(struct net *net)
1432 {
1433 struct sock *sk;
1434 struct netlink_kernel_cfg cfg = {
1435 .input = nl_fib_input,
1436 };
1437
1438 sk = netlink_kernel_create(net, NETLINK_FIB_LOOKUP, &cfg);
1439 if (!sk)
1440 return -EAFNOSUPPORT;
1441 net->ipv4.fibnl = sk;
1442 return 0;
1443 }
1444
nl_fib_lookup_exit(struct net * net)1445 static void nl_fib_lookup_exit(struct net *net)
1446 {
1447 netlink_kernel_release(net->ipv4.fibnl);
1448 net->ipv4.fibnl = NULL;
1449 }
1450
fib_disable_ip(struct net_device * dev,unsigned long event,bool force)1451 static void fib_disable_ip(struct net_device *dev, unsigned long event,
1452 bool force)
1453 {
1454 if (fib_sync_down_dev(dev, event, force))
1455 fib_flush(dev_net(dev));
1456 else
1457 rt_cache_flush(dev_net(dev));
1458 arp_ifdown(dev);
1459 }
1460
fib_inetaddr_event(struct notifier_block * this,unsigned long event,void * ptr)1461 static int fib_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
1462 {
1463 struct in_ifaddr *ifa = ptr;
1464 struct net_device *dev = ifa->ifa_dev->dev;
1465 struct net *net = dev_net(dev);
1466
1467 switch (event) {
1468 case NETDEV_UP:
1469 fib_add_ifaddr(ifa);
1470 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1471 fib_sync_up(dev, RTNH_F_DEAD);
1472 #endif
1473 atomic_inc(&net->ipv4.dev_addr_genid);
1474 rt_cache_flush(net);
1475 break;
1476 case NETDEV_DOWN:
1477 fib_del_ifaddr(ifa, NULL);
1478 atomic_inc(&net->ipv4.dev_addr_genid);
1479 if (!ifa->ifa_dev->ifa_list) {
1480 /* Last address was deleted from this interface.
1481 * Disable IP.
1482 */
1483 fib_disable_ip(dev, event, true);
1484 } else {
1485 rt_cache_flush(net);
1486 }
1487 break;
1488 }
1489 return NOTIFY_DONE;
1490 }
1491
fib_netdev_event(struct notifier_block * this,unsigned long event,void * ptr)1492 static int fib_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
1493 {
1494 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1495 struct netdev_notifier_changeupper_info *upper_info = ptr;
1496 struct netdev_notifier_info_ext *info_ext = ptr;
1497 struct in_device *in_dev;
1498 struct net *net = dev_net(dev);
1499 struct in_ifaddr *ifa;
1500 unsigned int flags;
1501
1502 if (event == NETDEV_UNREGISTER) {
1503 fib_disable_ip(dev, event, true);
1504 rt_flush_dev(dev);
1505 return NOTIFY_DONE;
1506 }
1507
1508 in_dev = __in_dev_get_rtnl(dev);
1509 if (!in_dev)
1510 return NOTIFY_DONE;
1511
1512 switch (event) {
1513 case NETDEV_UP:
1514 in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1515 fib_add_ifaddr(ifa);
1516 }
1517 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1518 fib_sync_up(dev, RTNH_F_DEAD);
1519 #endif
1520 atomic_inc(&net->ipv4.dev_addr_genid);
1521 rt_cache_flush(net);
1522 break;
1523 case NETDEV_DOWN:
1524 fib_disable_ip(dev, event, false);
1525 break;
1526 case NETDEV_CHANGE:
1527 flags = dev_get_flags(dev);
1528 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
1529 fib_sync_up(dev, RTNH_F_LINKDOWN);
1530 else
1531 fib_sync_down_dev(dev, event, false);
1532 rt_cache_flush(net);
1533 break;
1534 case NETDEV_CHANGEMTU:
1535 fib_sync_mtu(dev, info_ext->ext.mtu);
1536 rt_cache_flush(net);
1537 break;
1538 case NETDEV_CHANGEUPPER:
1539 upper_info = ptr;
1540 /* flush all routes if dev is linked to or unlinked from
1541 * an L3 master device (e.g., VRF)
1542 */
1543 if (upper_info->upper_dev &&
1544 netif_is_l3_master(upper_info->upper_dev))
1545 fib_disable_ip(dev, NETDEV_DOWN, true);
1546 break;
1547 }
1548 return NOTIFY_DONE;
1549 }
1550
1551 static struct notifier_block fib_inetaddr_notifier = {
1552 .notifier_call = fib_inetaddr_event,
1553 };
1554
1555 static struct notifier_block fib_netdev_notifier = {
1556 .notifier_call = fib_netdev_event,
1557 };
1558
ip_fib_net_init(struct net * net)1559 static int __net_init ip_fib_net_init(struct net *net)
1560 {
1561 int err;
1562 size_t size = sizeof(struct hlist_head) * FIB_TABLE_HASHSZ;
1563
1564 err = fib4_notifier_init(net);
1565 if (err)
1566 return err;
1567
1568 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1569 /* Default to 3-tuple */
1570 net->ipv4.sysctl_fib_multipath_hash_fields =
1571 FIB_MULTIPATH_HASH_FIELD_DEFAULT_MASK;
1572 #endif
1573
1574 /* Avoid false sharing : Use at least a full cache line */
1575 size = max_t(size_t, size, L1_CACHE_BYTES);
1576
1577 net->ipv4.fib_table_hash = kzalloc(size, GFP_KERNEL);
1578 if (!net->ipv4.fib_table_hash) {
1579 err = -ENOMEM;
1580 goto err_table_hash_alloc;
1581 }
1582
1583 err = fib4_rules_init(net);
1584 if (err < 0)
1585 goto err_rules_init;
1586 return 0;
1587
1588 err_rules_init:
1589 kfree(net->ipv4.fib_table_hash);
1590 err_table_hash_alloc:
1591 fib4_notifier_exit(net);
1592 return err;
1593 }
1594
ip_fib_net_exit(struct net * net)1595 static void ip_fib_net_exit(struct net *net)
1596 {
1597 int i;
1598
1599 ASSERT_RTNL_NET(net);
1600 #ifdef CONFIG_IP_MULTIPLE_TABLES
1601 RCU_INIT_POINTER(net->ipv4.fib_main, NULL);
1602 RCU_INIT_POINTER(net->ipv4.fib_default, NULL);
1603 #endif
1604 /* Destroy the tables in reverse order to guarantee that the
1605 * local table, ID 255, is destroyed before the main table, ID
1606 * 254. This is necessary as the local table may contain
1607 * references to data contained in the main table.
1608 */
1609 for (i = FIB_TABLE_HASHSZ - 1; i >= 0; i--) {
1610 struct hlist_head *head = &net->ipv4.fib_table_hash[i];
1611 struct hlist_node *tmp;
1612 struct fib_table *tb;
1613
1614 hlist_for_each_entry_safe(tb, tmp, head, tb_hlist) {
1615 hlist_del(&tb->tb_hlist);
1616 fib_table_flush(net, tb, true);
1617 fib_free_table(tb);
1618 }
1619 }
1620
1621 #ifdef CONFIG_IP_MULTIPLE_TABLES
1622 fib4_rules_exit(net);
1623 #endif
1624
1625 kfree(net->ipv4.fib_table_hash);
1626 fib4_notifier_exit(net);
1627 }
1628
fib_net_init(struct net * net)1629 static int __net_init fib_net_init(struct net *net)
1630 {
1631 int error;
1632
1633 #ifdef CONFIG_IP_ROUTE_CLASSID
1634 atomic_set(&net->ipv4.fib_num_tclassid_users, 0);
1635 #endif
1636 error = ip_fib_net_init(net);
1637 if (error < 0)
1638 goto out;
1639
1640 error = fib4_semantics_init(net);
1641 if (error)
1642 goto out_semantics;
1643
1644 error = nl_fib_lookup_init(net);
1645 if (error < 0)
1646 goto out_nlfl;
1647
1648 error = fib_proc_init(net);
1649 if (error < 0)
1650 goto out_proc;
1651 out:
1652 return error;
1653
1654 out_proc:
1655 nl_fib_lookup_exit(net);
1656 out_nlfl:
1657 fib4_semantics_exit(net);
1658 out_semantics:
1659 rtnl_net_lock(net);
1660 ip_fib_net_exit(net);
1661 rtnl_net_unlock(net);
1662 goto out;
1663 }
1664
fib_net_exit(struct net * net)1665 static void __net_exit fib_net_exit(struct net *net)
1666 {
1667 fib_proc_exit(net);
1668 nl_fib_lookup_exit(net);
1669 }
1670
fib_net_exit_batch(struct list_head * net_list)1671 static void __net_exit fib_net_exit_batch(struct list_head *net_list)
1672 {
1673 struct net *net;
1674
1675 rtnl_lock();
1676 list_for_each_entry(net, net_list, exit_list) {
1677 __rtnl_net_lock(net);
1678 ip_fib_net_exit(net);
1679 __rtnl_net_unlock(net);
1680 }
1681 rtnl_unlock();
1682
1683 list_for_each_entry(net, net_list, exit_list)
1684 fib4_semantics_exit(net);
1685 }
1686
1687 static struct pernet_operations fib_net_ops = {
1688 .init = fib_net_init,
1689 .exit = fib_net_exit,
1690 .exit_batch = fib_net_exit_batch,
1691 };
1692
1693 static const struct rtnl_msg_handler fib_rtnl_msg_handlers[] __initconst = {
1694 {.protocol = PF_INET, .msgtype = RTM_NEWROUTE,
1695 .doit = inet_rtm_newroute, .flags = RTNL_FLAG_DOIT_PERNET},
1696 {.protocol = PF_INET, .msgtype = RTM_DELROUTE,
1697 .doit = inet_rtm_delroute, .flags = RTNL_FLAG_DOIT_PERNET},
1698 {.protocol = PF_INET, .msgtype = RTM_GETROUTE, .dumpit = inet_dump_fib,
1699 .flags = RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE},
1700 };
1701
ip_fib_init(void)1702 void __init ip_fib_init(void)
1703 {
1704 fib_trie_init();
1705
1706 register_pernet_subsys(&fib_net_ops);
1707
1708 register_netdevice_notifier(&fib_netdev_notifier);
1709 register_inetaddr_notifier(&fib_inetaddr_notifier);
1710
1711 rtnl_register_many(fib_rtnl_msg_handlers);
1712 }
1713