]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Linux INET6 implementation | |
3 | * FIB front-end. | |
4 | * | |
5 | * Authors: | |
1ab1457c | 6 | * Pedro Roque <roque@di.fc.ul.pt> |
1da177e4 | 7 | * |
1da177e4 LT |
8 | * This program is free software; you can redistribute it and/or |
9 | * modify it under the terms of the GNU General Public License | |
10 | * as published by the Free Software Foundation; either version | |
11 | * 2 of the License, or (at your option) any later version. | |
12 | */ | |
13 | ||
14 | /* Changes: | |
15 | * | |
16 | * YOSHIFUJI Hideaki @USAGI | |
17 | * reworked default router selection. | |
18 | * - respect outgoing interface | |
19 | * - select from (probably) reachable routers (i.e. | |
20 | * routers in REACHABLE, STALE, DELAY or PROBE states). | |
21 | * - always select the same router if it is (probably) | |
22 | * reachable. otherwise, round-robin the list. | |
c0bece9f YH |
23 | * Ville Nuorvala |
24 | * Fixed routing subtrees. | |
1da177e4 LT |
25 | */ |
26 | ||
4fc268d2 | 27 | #include <linux/capability.h> |
1da177e4 LT |
28 | #include <linux/errno.h> |
29 | #include <linux/types.h> | |
30 | #include <linux/times.h> | |
31 | #include <linux/socket.h> | |
32 | #include <linux/sockios.h> | |
33 | #include <linux/net.h> | |
34 | #include <linux/route.h> | |
35 | #include <linux/netdevice.h> | |
36 | #include <linux/in6.h> | |
7bc570c8 | 37 | #include <linux/mroute6.h> |
1da177e4 | 38 | #include <linux/init.h> |
1da177e4 | 39 | #include <linux/if_arp.h> |
1da177e4 LT |
40 | #include <linux/proc_fs.h> |
41 | #include <linux/seq_file.h> | |
5b7c931d | 42 | #include <linux/nsproxy.h> |
5a0e3ad6 | 43 | #include <linux/slab.h> |
457c4cbc | 44 | #include <net/net_namespace.h> |
1da177e4 LT |
45 | #include <net/snmp.h> |
46 | #include <net/ipv6.h> | |
47 | #include <net/ip6_fib.h> | |
48 | #include <net/ip6_route.h> | |
49 | #include <net/ndisc.h> | |
50 | #include <net/addrconf.h> | |
51 | #include <net/tcp.h> | |
52 | #include <linux/rtnetlink.h> | |
53 | #include <net/dst.h> | |
54 | #include <net/xfrm.h> | |
8d71740c | 55 | #include <net/netevent.h> |
21713ebc | 56 | #include <net/netlink.h> |
1da177e4 LT |
57 | |
58 | #include <asm/uaccess.h> | |
59 | ||
60 | #ifdef CONFIG_SYSCTL | |
61 | #include <linux/sysctl.h> | |
62 | #endif | |
63 | ||
64 | /* Set to 3 to get tracing. */ | |
65 | #define RT6_DEBUG 2 | |
66 | ||
67 | #if RT6_DEBUG >= 3 | |
68 | #define RDBG(x) printk x | |
69 | #define RT6_TRACE(x...) printk(KERN_DEBUG x) | |
70 | #else | |
71 | #define RDBG(x) | |
72 | #define RT6_TRACE(x...) do { ; } while (0) | |
73 | #endif | |
74 | ||
21efcfa0 ED |
75 | static struct rt6_info *ip6_rt_copy(const struct rt6_info *ort, |
76 | const struct in6_addr *dest); | |
1da177e4 | 77 | static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie); |
0dbaee3b | 78 | static unsigned int ip6_default_advmss(const struct dst_entry *dst); |
d33e4553 | 79 | static unsigned int ip6_default_mtu(const struct dst_entry *dst); |
1da177e4 LT |
80 | static struct dst_entry *ip6_negative_advice(struct dst_entry *); |
81 | static void ip6_dst_destroy(struct dst_entry *); | |
82 | static void ip6_dst_ifdown(struct dst_entry *, | |
83 | struct net_device *dev, int how); | |
569d3645 | 84 | static int ip6_dst_gc(struct dst_ops *ops); |
1da177e4 LT |
85 | |
86 | static int ip6_pkt_discard(struct sk_buff *skb); | |
87 | static int ip6_pkt_discard_out(struct sk_buff *skb); | |
88 | static void ip6_link_failure(struct sk_buff *skb); | |
89 | static void ip6_rt_update_pmtu(struct dst_entry *dst, u32 mtu); | |
90 | ||
70ceb4f5 | 91 | #ifdef CONFIG_IPV6_ROUTE_INFO |
efa2cea0 | 92 | static struct rt6_info *rt6_add_route_info(struct net *net, |
b71d1d42 ED |
93 | const struct in6_addr *prefix, int prefixlen, |
94 | const struct in6_addr *gwaddr, int ifindex, | |
70ceb4f5 | 95 | unsigned pref); |
efa2cea0 | 96 | static struct rt6_info *rt6_get_route_info(struct net *net, |
b71d1d42 ED |
97 | const struct in6_addr *prefix, int prefixlen, |
98 | const struct in6_addr *gwaddr, int ifindex); | |
70ceb4f5 YH |
99 | #endif |
100 | ||
06582540 DM |
101 | static u32 *ipv6_cow_metrics(struct dst_entry *dst, unsigned long old) |
102 | { | |
103 | struct rt6_info *rt = (struct rt6_info *) dst; | |
104 | struct inet_peer *peer; | |
105 | u32 *p = NULL; | |
106 | ||
8e2ec639 YZ |
107 | if (!(rt->dst.flags & DST_HOST)) |
108 | return NULL; | |
109 | ||
06582540 DM |
110 | if (!rt->rt6i_peer) |
111 | rt6_bind_peer(rt, 1); | |
112 | ||
113 | peer = rt->rt6i_peer; | |
114 | if (peer) { | |
115 | u32 *old_p = __DST_METRICS_PTR(old); | |
116 | unsigned long prev, new; | |
117 | ||
118 | p = peer->metrics; | |
119 | if (inet_metrics_new(peer)) | |
120 | memcpy(p, old_p, sizeof(u32) * RTAX_MAX); | |
121 | ||
122 | new = (unsigned long) p; | |
123 | prev = cmpxchg(&dst->_metrics, old, new); | |
124 | ||
125 | if (prev != old) { | |
126 | p = __DST_METRICS_PTR(prev); | |
127 | if (prev & DST_METRICS_READ_ONLY) | |
128 | p = NULL; | |
129 | } | |
130 | } | |
131 | return p; | |
132 | } | |
133 | ||
d3aaeb38 DM |
134 | static struct neighbour *ip6_neigh_lookup(const struct dst_entry *dst, const void *daddr) |
135 | { | |
136 | return __neigh_lookup_errno(&nd_tbl, daddr, dst->dev); | |
137 | } | |
138 | ||
9a7ec3a9 | 139 | static struct dst_ops ip6_dst_ops_template = { |
1da177e4 | 140 | .family = AF_INET6, |
09640e63 | 141 | .protocol = cpu_to_be16(ETH_P_IPV6), |
1da177e4 LT |
142 | .gc = ip6_dst_gc, |
143 | .gc_thresh = 1024, | |
144 | .check = ip6_dst_check, | |
0dbaee3b | 145 | .default_advmss = ip6_default_advmss, |
d33e4553 | 146 | .default_mtu = ip6_default_mtu, |
06582540 | 147 | .cow_metrics = ipv6_cow_metrics, |
1da177e4 LT |
148 | .destroy = ip6_dst_destroy, |
149 | .ifdown = ip6_dst_ifdown, | |
150 | .negative_advice = ip6_negative_advice, | |
151 | .link_failure = ip6_link_failure, | |
152 | .update_pmtu = ip6_rt_update_pmtu, | |
1ac06e03 | 153 | .local_out = __ip6_local_out, |
d3aaeb38 | 154 | .neigh_lookup = ip6_neigh_lookup, |
1da177e4 LT |
155 | }; |
156 | ||
ec831ea7 RD |
157 | static unsigned int ip6_blackhole_default_mtu(const struct dst_entry *dst) |
158 | { | |
159 | return 0; | |
160 | } | |
161 | ||
14e50e57 DM |
162 | static void ip6_rt_blackhole_update_pmtu(struct dst_entry *dst, u32 mtu) |
163 | { | |
164 | } | |
165 | ||
0972ddb2 HB |
166 | static u32 *ip6_rt_blackhole_cow_metrics(struct dst_entry *dst, |
167 | unsigned long old) | |
168 | { | |
169 | return NULL; | |
170 | } | |
171 | ||
14e50e57 DM |
172 | static struct dst_ops ip6_dst_blackhole_ops = { |
173 | .family = AF_INET6, | |
09640e63 | 174 | .protocol = cpu_to_be16(ETH_P_IPV6), |
14e50e57 DM |
175 | .destroy = ip6_dst_destroy, |
176 | .check = ip6_dst_check, | |
ec831ea7 | 177 | .default_mtu = ip6_blackhole_default_mtu, |
214f45c9 | 178 | .default_advmss = ip6_default_advmss, |
14e50e57 | 179 | .update_pmtu = ip6_rt_blackhole_update_pmtu, |
0972ddb2 | 180 | .cow_metrics = ip6_rt_blackhole_cow_metrics, |
d3aaeb38 | 181 | .neigh_lookup = ip6_neigh_lookup, |
14e50e57 DM |
182 | }; |
183 | ||
62fa8a84 DM |
184 | static const u32 ip6_template_metrics[RTAX_MAX] = { |
185 | [RTAX_HOPLIMIT - 1] = 255, | |
186 | }; | |
187 | ||
bdb3289f | 188 | static struct rt6_info ip6_null_entry_template = { |
d8d1f30b CG |
189 | .dst = { |
190 | .__refcnt = ATOMIC_INIT(1), | |
191 | .__use = 1, | |
192 | .obsolete = -1, | |
193 | .error = -ENETUNREACH, | |
d8d1f30b CG |
194 | .input = ip6_pkt_discard, |
195 | .output = ip6_pkt_discard_out, | |
1da177e4 LT |
196 | }, |
197 | .rt6i_flags = (RTF_REJECT | RTF_NONEXTHOP), | |
4f724279 | 198 | .rt6i_protocol = RTPROT_KERNEL, |
1da177e4 LT |
199 | .rt6i_metric = ~(u32) 0, |
200 | .rt6i_ref = ATOMIC_INIT(1), | |
201 | }; | |
202 | ||
101367c2 TG |
203 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES |
204 | ||
6723ab54 DM |
205 | static int ip6_pkt_prohibit(struct sk_buff *skb); |
206 | static int ip6_pkt_prohibit_out(struct sk_buff *skb); | |
6723ab54 | 207 | |
280a34c8 | 208 | static struct rt6_info ip6_prohibit_entry_template = { |
d8d1f30b CG |
209 | .dst = { |
210 | .__refcnt = ATOMIC_INIT(1), | |
211 | .__use = 1, | |
212 | .obsolete = -1, | |
213 | .error = -EACCES, | |
d8d1f30b CG |
214 | .input = ip6_pkt_prohibit, |
215 | .output = ip6_pkt_prohibit_out, | |
101367c2 TG |
216 | }, |
217 | .rt6i_flags = (RTF_REJECT | RTF_NONEXTHOP), | |
4f724279 | 218 | .rt6i_protocol = RTPROT_KERNEL, |
101367c2 TG |
219 | .rt6i_metric = ~(u32) 0, |
220 | .rt6i_ref = ATOMIC_INIT(1), | |
221 | }; | |
222 | ||
bdb3289f | 223 | static struct rt6_info ip6_blk_hole_entry_template = { |
d8d1f30b CG |
224 | .dst = { |
225 | .__refcnt = ATOMIC_INIT(1), | |
226 | .__use = 1, | |
227 | .obsolete = -1, | |
228 | .error = -EINVAL, | |
d8d1f30b CG |
229 | .input = dst_discard, |
230 | .output = dst_discard, | |
101367c2 TG |
231 | }, |
232 | .rt6i_flags = (RTF_REJECT | RTF_NONEXTHOP), | |
4f724279 | 233 | .rt6i_protocol = RTPROT_KERNEL, |
101367c2 TG |
234 | .rt6i_metric = ~(u32) 0, |
235 | .rt6i_ref = ATOMIC_INIT(1), | |
236 | }; | |
237 | ||
238 | #endif | |
239 | ||
1da177e4 | 240 | /* allocate dst with ip6_dst_ops */ |
5c1e6aa3 | 241 | static inline struct rt6_info *ip6_dst_alloc(struct dst_ops *ops, |
957c665f DM |
242 | struct net_device *dev, |
243 | int flags) | |
1da177e4 | 244 | { |
957c665f | 245 | struct rt6_info *rt = dst_alloc(ops, dev, 0, 0, flags); |
cf911662 | 246 | |
fbe58186 MB |
247 | if (rt != NULL) |
248 | memset(&rt->rt6i_table, 0, | |
249 | sizeof(*rt) - sizeof(struct dst_entry)); | |
cf911662 DM |
250 | |
251 | return rt; | |
1da177e4 LT |
252 | } |
253 | ||
254 | static void ip6_dst_destroy(struct dst_entry *dst) | |
255 | { | |
256 | struct rt6_info *rt = (struct rt6_info *)dst; | |
257 | struct inet6_dev *idev = rt->rt6i_idev; | |
b3419363 | 258 | struct inet_peer *peer = rt->rt6i_peer; |
1da177e4 | 259 | |
8e2ec639 YZ |
260 | if (!(rt->dst.flags & DST_HOST)) |
261 | dst_destroy_metrics_generic(dst); | |
262 | ||
1da177e4 LT |
263 | if (idev != NULL) { |
264 | rt->rt6i_idev = NULL; | |
265 | in6_dev_put(idev); | |
1ab1457c | 266 | } |
b3419363 | 267 | if (peer) { |
b3419363 DM |
268 | rt->rt6i_peer = NULL; |
269 | inet_putpeer(peer); | |
270 | } | |
271 | } | |
272 | ||
6431cbc2 DM |
273 | static atomic_t __rt6_peer_genid = ATOMIC_INIT(0); |
274 | ||
275 | static u32 rt6_peer_genid(void) | |
276 | { | |
277 | return atomic_read(&__rt6_peer_genid); | |
278 | } | |
279 | ||
b3419363 DM |
280 | void rt6_bind_peer(struct rt6_info *rt, int create) |
281 | { | |
282 | struct inet_peer *peer; | |
283 | ||
b3419363 DM |
284 | peer = inet_getpeer_v6(&rt->rt6i_dst.addr, create); |
285 | if (peer && cmpxchg(&rt->rt6i_peer, NULL, peer) != NULL) | |
286 | inet_putpeer(peer); | |
6431cbc2 DM |
287 | else |
288 | rt->rt6i_peer_genid = rt6_peer_genid(); | |
1da177e4 LT |
289 | } |
290 | ||
291 | static void ip6_dst_ifdown(struct dst_entry *dst, struct net_device *dev, | |
292 | int how) | |
293 | { | |
294 | struct rt6_info *rt = (struct rt6_info *)dst; | |
295 | struct inet6_dev *idev = rt->rt6i_idev; | |
5a3e55d6 | 296 | struct net_device *loopback_dev = |
c346dca1 | 297 | dev_net(dev)->loopback_dev; |
1da177e4 | 298 | |
5a3e55d6 DL |
299 | if (dev != loopback_dev && idev != NULL && idev->dev == dev) { |
300 | struct inet6_dev *loopback_idev = | |
301 | in6_dev_get(loopback_dev); | |
1da177e4 LT |
302 | if (loopback_idev != NULL) { |
303 | rt->rt6i_idev = loopback_idev; | |
304 | in6_dev_put(idev); | |
305 | } | |
306 | } | |
307 | } | |
308 | ||
309 | static __inline__ int rt6_check_expired(const struct rt6_info *rt) | |
310 | { | |
a02cec21 ED |
311 | return (rt->rt6i_flags & RTF_EXPIRES) && |
312 | time_after(jiffies, rt->rt6i_expires); | |
1da177e4 LT |
313 | } |
314 | ||
b71d1d42 | 315 | static inline int rt6_need_strict(const struct in6_addr *daddr) |
c71099ac | 316 | { |
a02cec21 ED |
317 | return ipv6_addr_type(daddr) & |
318 | (IPV6_ADDR_MULTICAST | IPV6_ADDR_LINKLOCAL | IPV6_ADDR_LOOPBACK); | |
c71099ac TG |
319 | } |
320 | ||
1da177e4 | 321 | /* |
c71099ac | 322 | * Route lookup. Any table->tb6_lock is implied. |
1da177e4 LT |
323 | */ |
324 | ||
8ed67789 DL |
325 | static inline struct rt6_info *rt6_device_match(struct net *net, |
326 | struct rt6_info *rt, | |
b71d1d42 | 327 | const struct in6_addr *saddr, |
1da177e4 | 328 | int oif, |
d420895e | 329 | int flags) |
1da177e4 LT |
330 | { |
331 | struct rt6_info *local = NULL; | |
332 | struct rt6_info *sprt; | |
333 | ||
dd3abc4e YH |
334 | if (!oif && ipv6_addr_any(saddr)) |
335 | goto out; | |
336 | ||
d8d1f30b | 337 | for (sprt = rt; sprt; sprt = sprt->dst.rt6_next) { |
dd3abc4e YH |
338 | struct net_device *dev = sprt->rt6i_dev; |
339 | ||
340 | if (oif) { | |
1da177e4 LT |
341 | if (dev->ifindex == oif) |
342 | return sprt; | |
343 | if (dev->flags & IFF_LOOPBACK) { | |
344 | if (sprt->rt6i_idev == NULL || | |
345 | sprt->rt6i_idev->dev->ifindex != oif) { | |
d420895e | 346 | if (flags & RT6_LOOKUP_F_IFACE && oif) |
1da177e4 | 347 | continue; |
1ab1457c | 348 | if (local && (!oif || |
1da177e4 LT |
349 | local->rt6i_idev->dev->ifindex == oif)) |
350 | continue; | |
351 | } | |
352 | local = sprt; | |
353 | } | |
dd3abc4e YH |
354 | } else { |
355 | if (ipv6_chk_addr(net, saddr, dev, | |
356 | flags & RT6_LOOKUP_F_IFACE)) | |
357 | return sprt; | |
1da177e4 | 358 | } |
dd3abc4e | 359 | } |
1da177e4 | 360 | |
dd3abc4e | 361 | if (oif) { |
1da177e4 LT |
362 | if (local) |
363 | return local; | |
364 | ||
d420895e | 365 | if (flags & RT6_LOOKUP_F_IFACE) |
8ed67789 | 366 | return net->ipv6.ip6_null_entry; |
1da177e4 | 367 | } |
dd3abc4e | 368 | out: |
1da177e4 LT |
369 | return rt; |
370 | } | |
371 | ||
27097255 YH |
372 | #ifdef CONFIG_IPV6_ROUTER_PREF |
373 | static void rt6_probe(struct rt6_info *rt) | |
374 | { | |
f2c31e32 | 375 | struct neighbour *neigh; |
27097255 YH |
376 | /* |
377 | * Okay, this does not seem to be appropriate | |
378 | * for now, however, we need to check if it | |
379 | * is really so; aka Router Reachability Probing. | |
380 | * | |
381 | * Router Reachability Probe MUST be rate-limited | |
382 | * to no more than one per minute. | |
383 | */ | |
f2c31e32 ED |
384 | rcu_read_lock(); |
385 | neigh = rt ? dst_get_neighbour(&rt->dst) : NULL; | |
27097255 | 386 | if (!neigh || (neigh->nud_state & NUD_VALID)) |
f2c31e32 | 387 | goto out; |
27097255 YH |
388 | read_lock_bh(&neigh->lock); |
389 | if (!(neigh->nud_state & NUD_VALID) && | |
52e16356 | 390 | time_after(jiffies, neigh->updated + rt->rt6i_idev->cnf.rtr_probe_interval)) { |
27097255 YH |
391 | struct in6_addr mcaddr; |
392 | struct in6_addr *target; | |
393 | ||
394 | neigh->updated = jiffies; | |
395 | read_unlock_bh(&neigh->lock); | |
396 | ||
397 | target = (struct in6_addr *)&neigh->primary_key; | |
398 | addrconf_addr_solict_mult(target, &mcaddr); | |
399 | ndisc_send_ns(rt->rt6i_dev, NULL, target, &mcaddr, NULL); | |
f2c31e32 | 400 | } else { |
27097255 | 401 | read_unlock_bh(&neigh->lock); |
f2c31e32 ED |
402 | } |
403 | out: | |
404 | rcu_read_unlock(); | |
27097255 YH |
405 | } |
406 | #else | |
407 | static inline void rt6_probe(struct rt6_info *rt) | |
408 | { | |
27097255 YH |
409 | } |
410 | #endif | |
411 | ||
1da177e4 | 412 | /* |
554cfb7e | 413 | * Default Router Selection (RFC 2461 6.3.6) |
1da177e4 | 414 | */ |
b6f99a21 | 415 | static inline int rt6_check_dev(struct rt6_info *rt, int oif) |
554cfb7e YH |
416 | { |
417 | struct net_device *dev = rt->rt6i_dev; | |
161980f4 | 418 | if (!oif || dev->ifindex == oif) |
554cfb7e | 419 | return 2; |
161980f4 DM |
420 | if ((dev->flags & IFF_LOOPBACK) && |
421 | rt->rt6i_idev && rt->rt6i_idev->dev->ifindex == oif) | |
422 | return 1; | |
423 | return 0; | |
554cfb7e | 424 | } |
1da177e4 | 425 | |
b6f99a21 | 426 | static inline int rt6_check_neigh(struct rt6_info *rt) |
1da177e4 | 427 | { |
f2c31e32 | 428 | struct neighbour *neigh; |
398bcbeb | 429 | int m; |
f2c31e32 ED |
430 | |
431 | rcu_read_lock(); | |
432 | neigh = dst_get_neighbour(&rt->dst); | |
4d0c5911 YH |
433 | if (rt->rt6i_flags & RTF_NONEXTHOP || |
434 | !(rt->rt6i_flags & RTF_GATEWAY)) | |
435 | m = 1; | |
436 | else if (neigh) { | |
554cfb7e YH |
437 | read_lock_bh(&neigh->lock); |
438 | if (neigh->nud_state & NUD_VALID) | |
4d0c5911 | 439 | m = 2; |
398bcbeb YH |
440 | #ifdef CONFIG_IPV6_ROUTER_PREF |
441 | else if (neigh->nud_state & NUD_FAILED) | |
442 | m = 0; | |
443 | #endif | |
444 | else | |
ea73ee23 | 445 | m = 1; |
554cfb7e | 446 | read_unlock_bh(&neigh->lock); |
398bcbeb YH |
447 | } else |
448 | m = 0; | |
f2c31e32 | 449 | rcu_read_unlock(); |
554cfb7e | 450 | return m; |
1da177e4 LT |
451 | } |
452 | ||
554cfb7e YH |
453 | static int rt6_score_route(struct rt6_info *rt, int oif, |
454 | int strict) | |
1da177e4 | 455 | { |
4d0c5911 | 456 | int m, n; |
1ab1457c | 457 | |
4d0c5911 | 458 | m = rt6_check_dev(rt, oif); |
77d16f45 | 459 | if (!m && (strict & RT6_LOOKUP_F_IFACE)) |
554cfb7e | 460 | return -1; |
ebacaaa0 YH |
461 | #ifdef CONFIG_IPV6_ROUTER_PREF |
462 | m |= IPV6_DECODE_PREF(IPV6_EXTRACT_PREF(rt->rt6i_flags)) << 2; | |
463 | #endif | |
4d0c5911 | 464 | n = rt6_check_neigh(rt); |
557e92ef | 465 | if (!n && (strict & RT6_LOOKUP_F_REACHABLE)) |
554cfb7e YH |
466 | return -1; |
467 | return m; | |
468 | } | |
469 | ||
f11e6659 DM |
470 | static struct rt6_info *find_match(struct rt6_info *rt, int oif, int strict, |
471 | int *mpri, struct rt6_info *match) | |
554cfb7e | 472 | { |
f11e6659 DM |
473 | int m; |
474 | ||
475 | if (rt6_check_expired(rt)) | |
476 | goto out; | |
477 | ||
478 | m = rt6_score_route(rt, oif, strict); | |
479 | if (m < 0) | |
480 | goto out; | |
481 | ||
482 | if (m > *mpri) { | |
483 | if (strict & RT6_LOOKUP_F_REACHABLE) | |
484 | rt6_probe(match); | |
485 | *mpri = m; | |
486 | match = rt; | |
487 | } else if (strict & RT6_LOOKUP_F_REACHABLE) { | |
488 | rt6_probe(rt); | |
489 | } | |
490 | ||
491 | out: | |
492 | return match; | |
493 | } | |
494 | ||
495 | static struct rt6_info *find_rr_leaf(struct fib6_node *fn, | |
496 | struct rt6_info *rr_head, | |
497 | u32 metric, int oif, int strict) | |
498 | { | |
499 | struct rt6_info *rt, *match; | |
554cfb7e | 500 | int mpri = -1; |
1da177e4 | 501 | |
f11e6659 DM |
502 | match = NULL; |
503 | for (rt = rr_head; rt && rt->rt6i_metric == metric; | |
d8d1f30b | 504 | rt = rt->dst.rt6_next) |
f11e6659 DM |
505 | match = find_match(rt, oif, strict, &mpri, match); |
506 | for (rt = fn->leaf; rt && rt != rr_head && rt->rt6i_metric == metric; | |
d8d1f30b | 507 | rt = rt->dst.rt6_next) |
f11e6659 | 508 | match = find_match(rt, oif, strict, &mpri, match); |
1da177e4 | 509 | |
f11e6659 DM |
510 | return match; |
511 | } | |
1da177e4 | 512 | |
f11e6659 DM |
513 | static struct rt6_info *rt6_select(struct fib6_node *fn, int oif, int strict) |
514 | { | |
515 | struct rt6_info *match, *rt0; | |
8ed67789 | 516 | struct net *net; |
1da177e4 | 517 | |
f11e6659 | 518 | RT6_TRACE("%s(fn->leaf=%p, oif=%d)\n", |
0dc47877 | 519 | __func__, fn->leaf, oif); |
554cfb7e | 520 | |
f11e6659 DM |
521 | rt0 = fn->rr_ptr; |
522 | if (!rt0) | |
523 | fn->rr_ptr = rt0 = fn->leaf; | |
1da177e4 | 524 | |
f11e6659 | 525 | match = find_rr_leaf(fn, rt0, rt0->rt6i_metric, oif, strict); |
1da177e4 | 526 | |
554cfb7e | 527 | if (!match && |
f11e6659 | 528 | (strict & RT6_LOOKUP_F_REACHABLE)) { |
d8d1f30b | 529 | struct rt6_info *next = rt0->dst.rt6_next; |
f11e6659 | 530 | |
554cfb7e | 531 | /* no entries matched; do round-robin */ |
f11e6659 DM |
532 | if (!next || next->rt6i_metric != rt0->rt6i_metric) |
533 | next = fn->leaf; | |
534 | ||
535 | if (next != rt0) | |
536 | fn->rr_ptr = next; | |
1da177e4 | 537 | } |
1da177e4 | 538 | |
f11e6659 | 539 | RT6_TRACE("%s() => %p\n", |
0dc47877 | 540 | __func__, match); |
1da177e4 | 541 | |
c346dca1 | 542 | net = dev_net(rt0->rt6i_dev); |
a02cec21 | 543 | return match ? match : net->ipv6.ip6_null_entry; |
1da177e4 LT |
544 | } |
545 | ||
70ceb4f5 YH |
546 | #ifdef CONFIG_IPV6_ROUTE_INFO |
547 | int rt6_route_rcv(struct net_device *dev, u8 *opt, int len, | |
b71d1d42 | 548 | const struct in6_addr *gwaddr) |
70ceb4f5 | 549 | { |
c346dca1 | 550 | struct net *net = dev_net(dev); |
70ceb4f5 YH |
551 | struct route_info *rinfo = (struct route_info *) opt; |
552 | struct in6_addr prefix_buf, *prefix; | |
553 | unsigned int pref; | |
4bed72e4 | 554 | unsigned long lifetime; |
70ceb4f5 YH |
555 | struct rt6_info *rt; |
556 | ||
557 | if (len < sizeof(struct route_info)) { | |
558 | return -EINVAL; | |
559 | } | |
560 | ||
561 | /* Sanity check for prefix_len and length */ | |
562 | if (rinfo->length > 3) { | |
563 | return -EINVAL; | |
564 | } else if (rinfo->prefix_len > 128) { | |
565 | return -EINVAL; | |
566 | } else if (rinfo->prefix_len > 64) { | |
567 | if (rinfo->length < 2) { | |
568 | return -EINVAL; | |
569 | } | |
570 | } else if (rinfo->prefix_len > 0) { | |
571 | if (rinfo->length < 1) { | |
572 | return -EINVAL; | |
573 | } | |
574 | } | |
575 | ||
576 | pref = rinfo->route_pref; | |
577 | if (pref == ICMPV6_ROUTER_PREF_INVALID) | |
3933fc95 | 578 | return -EINVAL; |
70ceb4f5 | 579 | |
4bed72e4 | 580 | lifetime = addrconf_timeout_fixup(ntohl(rinfo->lifetime), HZ); |
70ceb4f5 YH |
581 | |
582 | if (rinfo->length == 3) | |
583 | prefix = (struct in6_addr *)rinfo->prefix; | |
584 | else { | |
585 | /* this function is safe */ | |
586 | ipv6_addr_prefix(&prefix_buf, | |
587 | (struct in6_addr *)rinfo->prefix, | |
588 | rinfo->prefix_len); | |
589 | prefix = &prefix_buf; | |
590 | } | |
591 | ||
efa2cea0 DL |
592 | rt = rt6_get_route_info(net, prefix, rinfo->prefix_len, gwaddr, |
593 | dev->ifindex); | |
70ceb4f5 YH |
594 | |
595 | if (rt && !lifetime) { | |
e0a1ad73 | 596 | ip6_del_rt(rt); |
70ceb4f5 YH |
597 | rt = NULL; |
598 | } | |
599 | ||
600 | if (!rt && lifetime) | |
efa2cea0 | 601 | rt = rt6_add_route_info(net, prefix, rinfo->prefix_len, gwaddr, dev->ifindex, |
70ceb4f5 YH |
602 | pref); |
603 | else if (rt) | |
604 | rt->rt6i_flags = RTF_ROUTEINFO | | |
605 | (rt->rt6i_flags & ~RTF_PREF_MASK) | RTF_PREF(pref); | |
606 | ||
607 | if (rt) { | |
4bed72e4 | 608 | if (!addrconf_finite_timeout(lifetime)) { |
70ceb4f5 YH |
609 | rt->rt6i_flags &= ~RTF_EXPIRES; |
610 | } else { | |
611 | rt->rt6i_expires = jiffies + HZ * lifetime; | |
612 | rt->rt6i_flags |= RTF_EXPIRES; | |
613 | } | |
d8d1f30b | 614 | dst_release(&rt->dst); |
70ceb4f5 YH |
615 | } |
616 | return 0; | |
617 | } | |
618 | #endif | |
619 | ||
8ed67789 | 620 | #define BACKTRACK(__net, saddr) \ |
982f56f3 | 621 | do { \ |
8ed67789 | 622 | if (rt == __net->ipv6.ip6_null_entry) { \ |
982f56f3 | 623 | struct fib6_node *pn; \ |
e0eda7bb | 624 | while (1) { \ |
982f56f3 YH |
625 | if (fn->fn_flags & RTN_TL_ROOT) \ |
626 | goto out; \ | |
627 | pn = fn->parent; \ | |
628 | if (FIB6_SUBTREE(pn) && FIB6_SUBTREE(pn) != fn) \ | |
8bce65b9 | 629 | fn = fib6_lookup(FIB6_SUBTREE(pn), NULL, saddr); \ |
982f56f3 YH |
630 | else \ |
631 | fn = pn; \ | |
632 | if (fn->fn_flags & RTN_RTINFO) \ | |
633 | goto restart; \ | |
c71099ac | 634 | } \ |
c71099ac | 635 | } \ |
982f56f3 | 636 | } while(0) |
c71099ac | 637 | |
8ed67789 DL |
638 | static struct rt6_info *ip6_pol_route_lookup(struct net *net, |
639 | struct fib6_table *table, | |
4c9483b2 | 640 | struct flowi6 *fl6, int flags) |
1da177e4 LT |
641 | { |
642 | struct fib6_node *fn; | |
643 | struct rt6_info *rt; | |
644 | ||
c71099ac | 645 | read_lock_bh(&table->tb6_lock); |
4c9483b2 | 646 | fn = fib6_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr); |
c71099ac TG |
647 | restart: |
648 | rt = fn->leaf; | |
4c9483b2 DM |
649 | rt = rt6_device_match(net, rt, &fl6->saddr, fl6->flowi6_oif, flags); |
650 | BACKTRACK(net, &fl6->saddr); | |
c71099ac | 651 | out: |
d8d1f30b | 652 | dst_use(&rt->dst, jiffies); |
c71099ac | 653 | read_unlock_bh(&table->tb6_lock); |
c71099ac TG |
654 | return rt; |
655 | ||
656 | } | |
657 | ||
9acd9f3a YH |
658 | struct rt6_info *rt6_lookup(struct net *net, const struct in6_addr *daddr, |
659 | const struct in6_addr *saddr, int oif, int strict) | |
c71099ac | 660 | { |
4c9483b2 DM |
661 | struct flowi6 fl6 = { |
662 | .flowi6_oif = oif, | |
663 | .daddr = *daddr, | |
c71099ac TG |
664 | }; |
665 | struct dst_entry *dst; | |
77d16f45 | 666 | int flags = strict ? RT6_LOOKUP_F_IFACE : 0; |
c71099ac | 667 | |
adaa70bb | 668 | if (saddr) { |
4c9483b2 | 669 | memcpy(&fl6.saddr, saddr, sizeof(*saddr)); |
adaa70bb TG |
670 | flags |= RT6_LOOKUP_F_HAS_SADDR; |
671 | } | |
672 | ||
4c9483b2 | 673 | dst = fib6_rule_lookup(net, &fl6, flags, ip6_pol_route_lookup); |
c71099ac TG |
674 | if (dst->error == 0) |
675 | return (struct rt6_info *) dst; | |
676 | ||
677 | dst_release(dst); | |
678 | ||
1da177e4 LT |
679 | return NULL; |
680 | } | |
681 | ||
7159039a YH |
682 | EXPORT_SYMBOL(rt6_lookup); |
683 | ||
c71099ac | 684 | /* ip6_ins_rt is called with FREE table->tb6_lock. |
1da177e4 LT |
685 | It takes new route entry, the addition fails by any reason the |
686 | route is freed. In any case, if caller does not hold it, it may | |
687 | be destroyed. | |
688 | */ | |
689 | ||
86872cb5 | 690 | static int __ip6_ins_rt(struct rt6_info *rt, struct nl_info *info) |
1da177e4 LT |
691 | { |
692 | int err; | |
c71099ac | 693 | struct fib6_table *table; |
1da177e4 | 694 | |
c71099ac TG |
695 | table = rt->rt6i_table; |
696 | write_lock_bh(&table->tb6_lock); | |
86872cb5 | 697 | err = fib6_add(&table->tb6_root, rt, info); |
c71099ac | 698 | write_unlock_bh(&table->tb6_lock); |
1da177e4 LT |
699 | |
700 | return err; | |
701 | } | |
702 | ||
40e22e8f TG |
703 | int ip6_ins_rt(struct rt6_info *rt) |
704 | { | |
4d1169c1 | 705 | struct nl_info info = { |
c346dca1 | 706 | .nl_net = dev_net(rt->rt6i_dev), |
4d1169c1 | 707 | }; |
528c4ceb | 708 | return __ip6_ins_rt(rt, &info); |
40e22e8f TG |
709 | } |
710 | ||
21efcfa0 ED |
711 | static struct rt6_info *rt6_alloc_cow(const struct rt6_info *ort, |
712 | const struct in6_addr *daddr, | |
b71d1d42 | 713 | const struct in6_addr *saddr) |
1da177e4 | 714 | { |
1da177e4 LT |
715 | struct rt6_info *rt; |
716 | ||
717 | /* | |
718 | * Clone the route. | |
719 | */ | |
720 | ||
21efcfa0 | 721 | rt = ip6_rt_copy(ort, daddr); |
1da177e4 LT |
722 | |
723 | if (rt) { | |
14deae41 DM |
724 | struct neighbour *neigh; |
725 | int attempts = !in_softirq(); | |
726 | ||
58c4fb86 YH |
727 | if (!(rt->rt6i_flags&RTF_GATEWAY)) { |
728 | if (rt->rt6i_dst.plen != 128 && | |
21efcfa0 | 729 | ipv6_addr_equal(&ort->rt6i_dst.addr, daddr)) |
58c4fb86 | 730 | rt->rt6i_flags |= RTF_ANYCAST; |
1da177e4 | 731 | ipv6_addr_copy(&rt->rt6i_gateway, daddr); |
58c4fb86 | 732 | } |
1da177e4 | 733 | |
1da177e4 | 734 | rt->rt6i_flags |= RTF_CACHE; |
1da177e4 LT |
735 | |
736 | #ifdef CONFIG_IPV6_SUBTREES | |
737 | if (rt->rt6i_src.plen && saddr) { | |
738 | ipv6_addr_copy(&rt->rt6i_src.addr, saddr); | |
739 | rt->rt6i_src.plen = 128; | |
740 | } | |
741 | #endif | |
742 | ||
14deae41 DM |
743 | retry: |
744 | neigh = ndisc_get_neigh(rt->rt6i_dev, &rt->rt6i_gateway); | |
745 | if (IS_ERR(neigh)) { | |
746 | struct net *net = dev_net(rt->rt6i_dev); | |
747 | int saved_rt_min_interval = | |
748 | net->ipv6.sysctl.ip6_rt_gc_min_interval; | |
749 | int saved_rt_elasticity = | |
750 | net->ipv6.sysctl.ip6_rt_gc_elasticity; | |
751 | ||
752 | if (attempts-- > 0) { | |
753 | net->ipv6.sysctl.ip6_rt_gc_elasticity = 1; | |
754 | net->ipv6.sysctl.ip6_rt_gc_min_interval = 0; | |
755 | ||
86393e52 | 756 | ip6_dst_gc(&net->ipv6.ip6_dst_ops); |
14deae41 DM |
757 | |
758 | net->ipv6.sysctl.ip6_rt_gc_elasticity = | |
759 | saved_rt_elasticity; | |
760 | net->ipv6.sysctl.ip6_rt_gc_min_interval = | |
761 | saved_rt_min_interval; | |
762 | goto retry; | |
763 | } | |
764 | ||
765 | if (net_ratelimit()) | |
766 | printk(KERN_WARNING | |
7e1b33e5 | 767 | "ipv6: Neighbour table overflow.\n"); |
d8d1f30b | 768 | dst_free(&rt->dst); |
14deae41 DM |
769 | return NULL; |
770 | } | |
69cce1d1 | 771 | dst_set_neighbour(&rt->dst, neigh); |
1da177e4 | 772 | |
95a9a5ba | 773 | } |
1da177e4 | 774 | |
95a9a5ba YH |
775 | return rt; |
776 | } | |
1da177e4 | 777 | |
21efcfa0 ED |
778 | static struct rt6_info *rt6_alloc_clone(struct rt6_info *ort, |
779 | const struct in6_addr *daddr) | |
299d9939 | 780 | { |
21efcfa0 ED |
781 | struct rt6_info *rt = ip6_rt_copy(ort, daddr); |
782 | ||
299d9939 | 783 | if (rt) { |
299d9939 | 784 | rt->rt6i_flags |= RTF_CACHE; |
f2c31e32 | 785 | dst_set_neighbour(&rt->dst, neigh_clone(dst_get_neighbour_raw(&ort->dst))); |
299d9939 YH |
786 | } |
787 | return rt; | |
788 | } | |
789 | ||
8ed67789 | 790 | static struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table, int oif, |
4c9483b2 | 791 | struct flowi6 *fl6, int flags) |
1da177e4 LT |
792 | { |
793 | struct fib6_node *fn; | |
519fbd87 | 794 | struct rt6_info *rt, *nrt; |
c71099ac | 795 | int strict = 0; |
1da177e4 | 796 | int attempts = 3; |
519fbd87 | 797 | int err; |
53b7997f | 798 | int reachable = net->ipv6.devconf_all->forwarding ? 0 : RT6_LOOKUP_F_REACHABLE; |
1da177e4 | 799 | |
77d16f45 | 800 | strict |= flags & RT6_LOOKUP_F_IFACE; |
1da177e4 LT |
801 | |
802 | relookup: | |
c71099ac | 803 | read_lock_bh(&table->tb6_lock); |
1da177e4 | 804 | |
8238dd06 | 805 | restart_2: |
4c9483b2 | 806 | fn = fib6_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr); |
1da177e4 LT |
807 | |
808 | restart: | |
4acad72d | 809 | rt = rt6_select(fn, oif, strict | reachable); |
8ed67789 | 810 | |
4c9483b2 | 811 | BACKTRACK(net, &fl6->saddr); |
8ed67789 | 812 | if (rt == net->ipv6.ip6_null_entry || |
8238dd06 | 813 | rt->rt6i_flags & RTF_CACHE) |
1ddef044 | 814 | goto out; |
1da177e4 | 815 | |
d8d1f30b | 816 | dst_hold(&rt->dst); |
c71099ac | 817 | read_unlock_bh(&table->tb6_lock); |
fb9de91e | 818 | |
f2c31e32 | 819 | if (!dst_get_neighbour_raw(&rt->dst) && !(rt->rt6i_flags & RTF_NONEXTHOP)) |
4c9483b2 | 820 | nrt = rt6_alloc_cow(rt, &fl6->daddr, &fl6->saddr); |
7343ff31 | 821 | else if (!(rt->dst.flags & DST_HOST)) |
4c9483b2 | 822 | nrt = rt6_alloc_clone(rt, &fl6->daddr); |
7343ff31 DM |
823 | else |
824 | goto out2; | |
e40cf353 | 825 | |
d8d1f30b | 826 | dst_release(&rt->dst); |
8ed67789 | 827 | rt = nrt ? : net->ipv6.ip6_null_entry; |
1da177e4 | 828 | |
d8d1f30b | 829 | dst_hold(&rt->dst); |
519fbd87 | 830 | if (nrt) { |
40e22e8f | 831 | err = ip6_ins_rt(nrt); |
519fbd87 | 832 | if (!err) |
1da177e4 | 833 | goto out2; |
1da177e4 | 834 | } |
1da177e4 | 835 | |
519fbd87 YH |
836 | if (--attempts <= 0) |
837 | goto out2; | |
838 | ||
839 | /* | |
c71099ac | 840 | * Race condition! In the gap, when table->tb6_lock was |
519fbd87 YH |
841 | * released someone could insert this route. Relookup. |
842 | */ | |
d8d1f30b | 843 | dst_release(&rt->dst); |
519fbd87 YH |
844 | goto relookup; |
845 | ||
846 | out: | |
8238dd06 YH |
847 | if (reachable) { |
848 | reachable = 0; | |
849 | goto restart_2; | |
850 | } | |
d8d1f30b | 851 | dst_hold(&rt->dst); |
c71099ac | 852 | read_unlock_bh(&table->tb6_lock); |
1da177e4 | 853 | out2: |
d8d1f30b CG |
854 | rt->dst.lastuse = jiffies; |
855 | rt->dst.__use++; | |
c71099ac TG |
856 | |
857 | return rt; | |
1da177e4 LT |
858 | } |
859 | ||
8ed67789 | 860 | static struct rt6_info *ip6_pol_route_input(struct net *net, struct fib6_table *table, |
4c9483b2 | 861 | struct flowi6 *fl6, int flags) |
4acad72d | 862 | { |
4c9483b2 | 863 | return ip6_pol_route(net, table, fl6->flowi6_iif, fl6, flags); |
4acad72d PE |
864 | } |
865 | ||
c71099ac TG |
866 | void ip6_route_input(struct sk_buff *skb) |
867 | { | |
b71d1d42 | 868 | const struct ipv6hdr *iph = ipv6_hdr(skb); |
c346dca1 | 869 | struct net *net = dev_net(skb->dev); |
adaa70bb | 870 | int flags = RT6_LOOKUP_F_HAS_SADDR; |
4c9483b2 DM |
871 | struct flowi6 fl6 = { |
872 | .flowi6_iif = skb->dev->ifindex, | |
873 | .daddr = iph->daddr, | |
874 | .saddr = iph->saddr, | |
875 | .flowlabel = (* (__be32 *) iph)&IPV6_FLOWINFO_MASK, | |
876 | .flowi6_mark = skb->mark, | |
877 | .flowi6_proto = iph->nexthdr, | |
c71099ac | 878 | }; |
adaa70bb | 879 | |
1d6e55f1 | 880 | if (rt6_need_strict(&iph->daddr) && skb->dev->type != ARPHRD_PIMREG) |
adaa70bb | 881 | flags |= RT6_LOOKUP_F_IFACE; |
c71099ac | 882 | |
4c9483b2 | 883 | skb_dst_set(skb, fib6_rule_lookup(net, &fl6, flags, ip6_pol_route_input)); |
c71099ac TG |
884 | } |
885 | ||
8ed67789 | 886 | static struct rt6_info *ip6_pol_route_output(struct net *net, struct fib6_table *table, |
4c9483b2 | 887 | struct flowi6 *fl6, int flags) |
1da177e4 | 888 | { |
4c9483b2 | 889 | return ip6_pol_route(net, table, fl6->flowi6_oif, fl6, flags); |
c71099ac TG |
890 | } |
891 | ||
9c7a4f9c | 892 | struct dst_entry * ip6_route_output(struct net *net, const struct sock *sk, |
4c9483b2 | 893 | struct flowi6 *fl6) |
c71099ac TG |
894 | { |
895 | int flags = 0; | |
896 | ||
4c9483b2 | 897 | if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl6->daddr)) |
77d16f45 | 898 | flags |= RT6_LOOKUP_F_IFACE; |
c71099ac | 899 | |
4c9483b2 | 900 | if (!ipv6_addr_any(&fl6->saddr)) |
adaa70bb | 901 | flags |= RT6_LOOKUP_F_HAS_SADDR; |
0c9a2ac1 YH |
902 | else if (sk) |
903 | flags |= rt6_srcprefs2flags(inet6_sk(sk)->srcprefs); | |
adaa70bb | 904 | |
4c9483b2 | 905 | return fib6_rule_lookup(net, fl6, flags, ip6_pol_route_output); |
1da177e4 LT |
906 | } |
907 | ||
7159039a | 908 | EXPORT_SYMBOL(ip6_route_output); |
1da177e4 | 909 | |
2774c131 | 910 | struct dst_entry *ip6_blackhole_route(struct net *net, struct dst_entry *dst_orig) |
14e50e57 | 911 | { |
5c1e6aa3 | 912 | struct rt6_info *rt, *ort = (struct rt6_info *) dst_orig; |
14e50e57 DM |
913 | struct dst_entry *new = NULL; |
914 | ||
5c1e6aa3 | 915 | rt = dst_alloc(&ip6_dst_blackhole_ops, ort->dst.dev, 1, 0, 0); |
14e50e57 | 916 | if (rt) { |
cf911662 DM |
917 | memset(&rt->rt6i_table, 0, sizeof(*rt) - sizeof(struct dst_entry)); |
918 | ||
d8d1f30b | 919 | new = &rt->dst; |
14e50e57 | 920 | |
14e50e57 | 921 | new->__use = 1; |
352e512c HX |
922 | new->input = dst_discard; |
923 | new->output = dst_discard; | |
14e50e57 | 924 | |
21efcfa0 ED |
925 | if (dst_metrics_read_only(&ort->dst)) |
926 | new->_metrics = ort->dst._metrics; | |
927 | else | |
928 | dst_copy_metrics(new, &ort->dst); | |
14e50e57 DM |
929 | rt->rt6i_idev = ort->rt6i_idev; |
930 | if (rt->rt6i_idev) | |
931 | in6_dev_hold(rt->rt6i_idev); | |
932 | rt->rt6i_expires = 0; | |
933 | ||
934 | ipv6_addr_copy(&rt->rt6i_gateway, &ort->rt6i_gateway); | |
935 | rt->rt6i_flags = ort->rt6i_flags & ~RTF_EXPIRES; | |
936 | rt->rt6i_metric = 0; | |
937 | ||
938 | memcpy(&rt->rt6i_dst, &ort->rt6i_dst, sizeof(struct rt6key)); | |
939 | #ifdef CONFIG_IPV6_SUBTREES | |
940 | memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key)); | |
941 | #endif | |
942 | ||
943 | dst_free(new); | |
944 | } | |
945 | ||
69ead7af DM |
946 | dst_release(dst_orig); |
947 | return new ? new : ERR_PTR(-ENOMEM); | |
14e50e57 | 948 | } |
14e50e57 | 949 | |
1da177e4 LT |
950 | /* |
951 | * Destination cache support functions | |
952 | */ | |
953 | ||
954 | static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie) | |
955 | { | |
956 | struct rt6_info *rt; | |
957 | ||
958 | rt = (struct rt6_info *) dst; | |
959 | ||
6431cbc2 DM |
960 | if (rt->rt6i_node && (rt->rt6i_node->fn_sernum == cookie)) { |
961 | if (rt->rt6i_peer_genid != rt6_peer_genid()) { | |
962 | if (!rt->rt6i_peer) | |
963 | rt6_bind_peer(rt, 0); | |
964 | rt->rt6i_peer_genid = rt6_peer_genid(); | |
965 | } | |
1da177e4 | 966 | return dst; |
6431cbc2 | 967 | } |
1da177e4 LT |
968 | return NULL; |
969 | } | |
970 | ||
971 | static struct dst_entry *ip6_negative_advice(struct dst_entry *dst) | |
972 | { | |
973 | struct rt6_info *rt = (struct rt6_info *) dst; | |
974 | ||
975 | if (rt) { | |
54c1a859 YH |
976 | if (rt->rt6i_flags & RTF_CACHE) { |
977 | if (rt6_check_expired(rt)) { | |
978 | ip6_del_rt(rt); | |
979 | dst = NULL; | |
980 | } | |
981 | } else { | |
1da177e4 | 982 | dst_release(dst); |
54c1a859 YH |
983 | dst = NULL; |
984 | } | |
1da177e4 | 985 | } |
54c1a859 | 986 | return dst; |
1da177e4 LT |
987 | } |
988 | ||
989 | static void ip6_link_failure(struct sk_buff *skb) | |
990 | { | |
991 | struct rt6_info *rt; | |
992 | ||
3ffe533c | 993 | icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_ADDR_UNREACH, 0); |
1da177e4 | 994 | |
adf30907 | 995 | rt = (struct rt6_info *) skb_dst(skb); |
1da177e4 LT |
996 | if (rt) { |
997 | if (rt->rt6i_flags&RTF_CACHE) { | |
d8d1f30b | 998 | dst_set_expires(&rt->dst, 0); |
1da177e4 LT |
999 | rt->rt6i_flags |= RTF_EXPIRES; |
1000 | } else if (rt->rt6i_node && (rt->rt6i_flags & RTF_DEFAULT)) | |
1001 | rt->rt6i_node->fn_sernum = -1; | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | static void ip6_rt_update_pmtu(struct dst_entry *dst, u32 mtu) | |
1006 | { | |
1007 | struct rt6_info *rt6 = (struct rt6_info*)dst; | |
1008 | ||
1009 | if (mtu < dst_mtu(dst) && rt6->rt6i_dst.plen == 128) { | |
1010 | rt6->rt6i_flags |= RTF_MODIFIED; | |
1011 | if (mtu < IPV6_MIN_MTU) { | |
defb3519 | 1012 | u32 features = dst_metric(dst, RTAX_FEATURES); |
1da177e4 | 1013 | mtu = IPV6_MIN_MTU; |
defb3519 DM |
1014 | features |= RTAX_FEATURE_ALLFRAG; |
1015 | dst_metric_set(dst, RTAX_FEATURES, features); | |
1da177e4 | 1016 | } |
defb3519 | 1017 | dst_metric_set(dst, RTAX_MTU, mtu); |
1da177e4 LT |
1018 | } |
1019 | } | |
1020 | ||
0dbaee3b | 1021 | static unsigned int ip6_default_advmss(const struct dst_entry *dst) |
1da177e4 | 1022 | { |
0dbaee3b DM |
1023 | struct net_device *dev = dst->dev; |
1024 | unsigned int mtu = dst_mtu(dst); | |
1025 | struct net *net = dev_net(dev); | |
1026 | ||
1da177e4 LT |
1027 | mtu -= sizeof(struct ipv6hdr) + sizeof(struct tcphdr); |
1028 | ||
5578689a DL |
1029 | if (mtu < net->ipv6.sysctl.ip6_rt_min_advmss) |
1030 | mtu = net->ipv6.sysctl.ip6_rt_min_advmss; | |
1da177e4 LT |
1031 | |
1032 | /* | |
1ab1457c YH |
1033 | * Maximal non-jumbo IPv6 payload is IPV6_MAXPLEN and |
1034 | * corresponding MSS is IPV6_MAXPLEN - tcp_header_size. | |
1035 | * IPV6_MAXPLEN is also valid and means: "any MSS, | |
1da177e4 LT |
1036 | * rely only on pmtu discovery" |
1037 | */ | |
1038 | if (mtu > IPV6_MAXPLEN - sizeof(struct tcphdr)) | |
1039 | mtu = IPV6_MAXPLEN; | |
1040 | return mtu; | |
1041 | } | |
1042 | ||
d33e4553 DM |
1043 | static unsigned int ip6_default_mtu(const struct dst_entry *dst) |
1044 | { | |
1045 | unsigned int mtu = IPV6_MIN_MTU; | |
1046 | struct inet6_dev *idev; | |
1047 | ||
1048 | rcu_read_lock(); | |
1049 | idev = __in6_dev_get(dst->dev); | |
1050 | if (idev) | |
1051 | mtu = idev->cnf.mtu6; | |
1052 | rcu_read_unlock(); | |
1053 | ||
1054 | return mtu; | |
1055 | } | |
1056 | ||
3b00944c YH |
1057 | static struct dst_entry *icmp6_dst_gc_list; |
1058 | static DEFINE_SPINLOCK(icmp6_dst_lock); | |
5d0bbeeb | 1059 | |
3b00944c | 1060 | struct dst_entry *icmp6_dst_alloc(struct net_device *dev, |
1da177e4 | 1061 | struct neighbour *neigh, |
9acd9f3a | 1062 | const struct in6_addr *addr) |
1da177e4 LT |
1063 | { |
1064 | struct rt6_info *rt; | |
1065 | struct inet6_dev *idev = in6_dev_get(dev); | |
c346dca1 | 1066 | struct net *net = dev_net(dev); |
1da177e4 LT |
1067 | |
1068 | if (unlikely(idev == NULL)) | |
1069 | return NULL; | |
1070 | ||
957c665f | 1071 | rt = ip6_dst_alloc(&net->ipv6.ip6_dst_ops, dev, 0); |
1da177e4 LT |
1072 | if (unlikely(rt == NULL)) { |
1073 | in6_dev_put(idev); | |
1074 | goto out; | |
1075 | } | |
1076 | ||
1da177e4 LT |
1077 | if (neigh) |
1078 | neigh_hold(neigh); | |
14deae41 | 1079 | else { |
1da177e4 | 1080 | neigh = ndisc_get_neigh(dev, addr); |
14deae41 DM |
1081 | if (IS_ERR(neigh)) |
1082 | neigh = NULL; | |
1083 | } | |
1da177e4 | 1084 | |
8e2ec639 YZ |
1085 | rt->dst.flags |= DST_HOST; |
1086 | rt->dst.output = ip6_output; | |
69cce1d1 | 1087 | dst_set_neighbour(&rt->dst, neigh); |
d8d1f30b | 1088 | atomic_set(&rt->dst.__refcnt, 1); |
defb3519 | 1089 | dst_metric_set(&rt->dst, RTAX_HOPLIMIT, 255); |
8e2ec639 YZ |
1090 | |
1091 | ipv6_addr_copy(&rt->rt6i_dst.addr, addr); | |
1092 | rt->rt6i_dst.plen = 128; | |
1093 | rt->rt6i_idev = idev; | |
1da177e4 | 1094 | |
3b00944c | 1095 | spin_lock_bh(&icmp6_dst_lock); |
d8d1f30b CG |
1096 | rt->dst.next = icmp6_dst_gc_list; |
1097 | icmp6_dst_gc_list = &rt->dst; | |
3b00944c | 1098 | spin_unlock_bh(&icmp6_dst_lock); |
1da177e4 | 1099 | |
5578689a | 1100 | fib6_force_start_gc(net); |
1da177e4 LT |
1101 | |
1102 | out: | |
d8d1f30b | 1103 | return &rt->dst; |
1da177e4 LT |
1104 | } |
1105 | ||
3d0f24a7 | 1106 | int icmp6_dst_gc(void) |
1da177e4 | 1107 | { |
e9476e95 | 1108 | struct dst_entry *dst, **pprev; |
3d0f24a7 | 1109 | int more = 0; |
1da177e4 | 1110 | |
3b00944c YH |
1111 | spin_lock_bh(&icmp6_dst_lock); |
1112 | pprev = &icmp6_dst_gc_list; | |
5d0bbeeb | 1113 | |
1da177e4 LT |
1114 | while ((dst = *pprev) != NULL) { |
1115 | if (!atomic_read(&dst->__refcnt)) { | |
1116 | *pprev = dst->next; | |
1117 | dst_free(dst); | |
1da177e4 LT |
1118 | } else { |
1119 | pprev = &dst->next; | |
3d0f24a7 | 1120 | ++more; |
1da177e4 LT |
1121 | } |
1122 | } | |
1123 | ||
3b00944c | 1124 | spin_unlock_bh(&icmp6_dst_lock); |
5d0bbeeb | 1125 | |
3d0f24a7 | 1126 | return more; |
1da177e4 LT |
1127 | } |
1128 | ||
1e493d19 DM |
1129 | static void icmp6_clean_all(int (*func)(struct rt6_info *rt, void *arg), |
1130 | void *arg) | |
1131 | { | |
1132 | struct dst_entry *dst, **pprev; | |
1133 | ||
1134 | spin_lock_bh(&icmp6_dst_lock); | |
1135 | pprev = &icmp6_dst_gc_list; | |
1136 | while ((dst = *pprev) != NULL) { | |
1137 | struct rt6_info *rt = (struct rt6_info *) dst; | |
1138 | if (func(rt, arg)) { | |
1139 | *pprev = dst->next; | |
1140 | dst_free(dst); | |
1141 | } else { | |
1142 | pprev = &dst->next; | |
1143 | } | |
1144 | } | |
1145 | spin_unlock_bh(&icmp6_dst_lock); | |
1146 | } | |
1147 | ||
569d3645 | 1148 | static int ip6_dst_gc(struct dst_ops *ops) |
1da177e4 | 1149 | { |
1da177e4 | 1150 | unsigned long now = jiffies; |
86393e52 | 1151 | struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops); |
7019b78e DL |
1152 | int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval; |
1153 | int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size; | |
1154 | int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity; | |
1155 | int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout; | |
1156 | unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc; | |
fc66f95c | 1157 | int entries; |
7019b78e | 1158 | |
fc66f95c | 1159 | entries = dst_entries_get_fast(ops); |
7019b78e | 1160 | if (time_after(rt_last_gc + rt_min_interval, now) && |
fc66f95c | 1161 | entries <= rt_max_size) |
1da177e4 LT |
1162 | goto out; |
1163 | ||
6891a346 BT |
1164 | net->ipv6.ip6_rt_gc_expire++; |
1165 | fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net); | |
1166 | net->ipv6.ip6_rt_last_gc = now; | |
fc66f95c ED |
1167 | entries = dst_entries_get_slow(ops); |
1168 | if (entries < ops->gc_thresh) | |
7019b78e | 1169 | net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1; |
1da177e4 | 1170 | out: |
7019b78e | 1171 | net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity; |
fc66f95c | 1172 | return entries > rt_max_size; |
1da177e4 LT |
1173 | } |
1174 | ||
1175 | /* Clean host part of a prefix. Not necessary in radix tree, | |
1176 | but results in cleaner routing tables. | |
1177 | ||
1178 | Remove it only when all the things will work! | |
1179 | */ | |
1180 | ||
6b75d090 | 1181 | int ip6_dst_hoplimit(struct dst_entry *dst) |
1da177e4 | 1182 | { |
5170ae82 | 1183 | int hoplimit = dst_metric_raw(dst, RTAX_HOPLIMIT); |
a02e4b7d | 1184 | if (hoplimit == 0) { |
6b75d090 | 1185 | struct net_device *dev = dst->dev; |
c68f24cc ED |
1186 | struct inet6_dev *idev; |
1187 | ||
1188 | rcu_read_lock(); | |
1189 | idev = __in6_dev_get(dev); | |
1190 | if (idev) | |
6b75d090 | 1191 | hoplimit = idev->cnf.hop_limit; |
c68f24cc | 1192 | else |
53b7997f | 1193 | hoplimit = dev_net(dev)->ipv6.devconf_all->hop_limit; |
c68f24cc | 1194 | rcu_read_unlock(); |
1da177e4 LT |
1195 | } |
1196 | return hoplimit; | |
1197 | } | |
abbf46ae | 1198 | EXPORT_SYMBOL(ip6_dst_hoplimit); |
1da177e4 LT |
1199 | |
1200 | /* | |
1201 | * | |
1202 | */ | |
1203 | ||
86872cb5 | 1204 | int ip6_route_add(struct fib6_config *cfg) |
1da177e4 LT |
1205 | { |
1206 | int err; | |
5578689a | 1207 | struct net *net = cfg->fc_nlinfo.nl_net; |
1da177e4 LT |
1208 | struct rt6_info *rt = NULL; |
1209 | struct net_device *dev = NULL; | |
1210 | struct inet6_dev *idev = NULL; | |
c71099ac | 1211 | struct fib6_table *table; |
1da177e4 LT |
1212 | int addr_type; |
1213 | ||
86872cb5 | 1214 | if (cfg->fc_dst_len > 128 || cfg->fc_src_len > 128) |
1da177e4 LT |
1215 | return -EINVAL; |
1216 | #ifndef CONFIG_IPV6_SUBTREES | |
86872cb5 | 1217 | if (cfg->fc_src_len) |
1da177e4 LT |
1218 | return -EINVAL; |
1219 | #endif | |
86872cb5 | 1220 | if (cfg->fc_ifindex) { |
1da177e4 | 1221 | err = -ENODEV; |
5578689a | 1222 | dev = dev_get_by_index(net, cfg->fc_ifindex); |
1da177e4 LT |
1223 | if (!dev) |
1224 | goto out; | |
1225 | idev = in6_dev_get(dev); | |
1226 | if (!idev) | |
1227 | goto out; | |
1228 | } | |
1229 | ||
86872cb5 TG |
1230 | if (cfg->fc_metric == 0) |
1231 | cfg->fc_metric = IP6_RT_PRIO_USER; | |
1da177e4 | 1232 | |
5578689a | 1233 | table = fib6_new_table(net, cfg->fc_table); |
c71099ac TG |
1234 | if (table == NULL) { |
1235 | err = -ENOBUFS; | |
1236 | goto out; | |
1237 | } | |
1238 | ||
957c665f | 1239 | rt = ip6_dst_alloc(&net->ipv6.ip6_dst_ops, NULL, DST_NOCOUNT); |
1da177e4 LT |
1240 | |
1241 | if (rt == NULL) { | |
1242 | err = -ENOMEM; | |
1243 | goto out; | |
1244 | } | |
1245 | ||
d8d1f30b | 1246 | rt->dst.obsolete = -1; |
6f704992 YH |
1247 | rt->rt6i_expires = (cfg->fc_flags & RTF_EXPIRES) ? |
1248 | jiffies + clock_t_to_jiffies(cfg->fc_expires) : | |
1249 | 0; | |
1da177e4 | 1250 | |
86872cb5 TG |
1251 | if (cfg->fc_protocol == RTPROT_UNSPEC) |
1252 | cfg->fc_protocol = RTPROT_BOOT; | |
1253 | rt->rt6i_protocol = cfg->fc_protocol; | |
1254 | ||
1255 | addr_type = ipv6_addr_type(&cfg->fc_dst); | |
1da177e4 LT |
1256 | |
1257 | if (addr_type & IPV6_ADDR_MULTICAST) | |
d8d1f30b | 1258 | rt->dst.input = ip6_mc_input; |
ab79ad14 MŻ |
1259 | else if (cfg->fc_flags & RTF_LOCAL) |
1260 | rt->dst.input = ip6_input; | |
1da177e4 | 1261 | else |
d8d1f30b | 1262 | rt->dst.input = ip6_forward; |
1da177e4 | 1263 | |
d8d1f30b | 1264 | rt->dst.output = ip6_output; |
1da177e4 | 1265 | |
86872cb5 TG |
1266 | ipv6_addr_prefix(&rt->rt6i_dst.addr, &cfg->fc_dst, cfg->fc_dst_len); |
1267 | rt->rt6i_dst.plen = cfg->fc_dst_len; | |
1da177e4 | 1268 | if (rt->rt6i_dst.plen == 128) |
11d53b49 | 1269 | rt->dst.flags |= DST_HOST; |
1da177e4 | 1270 | |
8e2ec639 YZ |
1271 | if (!(rt->dst.flags & DST_HOST) && cfg->fc_mx) { |
1272 | u32 *metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL); | |
1273 | if (!metrics) { | |
1274 | err = -ENOMEM; | |
1275 | goto out; | |
1276 | } | |
1277 | dst_init_metrics(&rt->dst, metrics, 0); | |
1278 | } | |
1da177e4 | 1279 | #ifdef CONFIG_IPV6_SUBTREES |
86872cb5 TG |
1280 | ipv6_addr_prefix(&rt->rt6i_src.addr, &cfg->fc_src, cfg->fc_src_len); |
1281 | rt->rt6i_src.plen = cfg->fc_src_len; | |
1da177e4 LT |
1282 | #endif |
1283 | ||
86872cb5 | 1284 | rt->rt6i_metric = cfg->fc_metric; |
1da177e4 LT |
1285 | |
1286 | /* We cannot add true routes via loopback here, | |
1287 | they would result in kernel looping; promote them to reject routes | |
1288 | */ | |
86872cb5 | 1289 | if ((cfg->fc_flags & RTF_REJECT) || |
ab79ad14 MŻ |
1290 | (dev && (dev->flags&IFF_LOOPBACK) && !(addr_type&IPV6_ADDR_LOOPBACK) |
1291 | && !(cfg->fc_flags&RTF_LOCAL))) { | |
1da177e4 | 1292 | /* hold loopback dev/idev if we haven't done so. */ |
5578689a | 1293 | if (dev != net->loopback_dev) { |
1da177e4 LT |
1294 | if (dev) { |
1295 | dev_put(dev); | |
1296 | in6_dev_put(idev); | |
1297 | } | |
5578689a | 1298 | dev = net->loopback_dev; |
1da177e4 LT |
1299 | dev_hold(dev); |
1300 | idev = in6_dev_get(dev); | |
1301 | if (!idev) { | |
1302 | err = -ENODEV; | |
1303 | goto out; | |
1304 | } | |
1305 | } | |
d8d1f30b CG |
1306 | rt->dst.output = ip6_pkt_discard_out; |
1307 | rt->dst.input = ip6_pkt_discard; | |
1308 | rt->dst.error = -ENETUNREACH; | |
1da177e4 LT |
1309 | rt->rt6i_flags = RTF_REJECT|RTF_NONEXTHOP; |
1310 | goto install_route; | |
1311 | } | |
1312 | ||
86872cb5 | 1313 | if (cfg->fc_flags & RTF_GATEWAY) { |
b71d1d42 | 1314 | const struct in6_addr *gw_addr; |
1da177e4 LT |
1315 | int gwa_type; |
1316 | ||
86872cb5 TG |
1317 | gw_addr = &cfg->fc_gateway; |
1318 | ipv6_addr_copy(&rt->rt6i_gateway, gw_addr); | |
1da177e4 LT |
1319 | gwa_type = ipv6_addr_type(gw_addr); |
1320 | ||
1321 | if (gwa_type != (IPV6_ADDR_LINKLOCAL|IPV6_ADDR_UNICAST)) { | |
1322 | struct rt6_info *grt; | |
1323 | ||
1324 | /* IPv6 strictly inhibits using not link-local | |
1325 | addresses as nexthop address. | |
1326 | Otherwise, router will not able to send redirects. | |
1327 | It is very good, but in some (rare!) circumstances | |
1328 | (SIT, PtP, NBMA NOARP links) it is handy to allow | |
1329 | some exceptions. --ANK | |
1330 | */ | |
1331 | err = -EINVAL; | |
1332 | if (!(gwa_type&IPV6_ADDR_UNICAST)) | |
1333 | goto out; | |
1334 | ||
5578689a | 1335 | grt = rt6_lookup(net, gw_addr, NULL, cfg->fc_ifindex, 1); |
1da177e4 LT |
1336 | |
1337 | err = -EHOSTUNREACH; | |
1338 | if (grt == NULL) | |
1339 | goto out; | |
1340 | if (dev) { | |
1341 | if (dev != grt->rt6i_dev) { | |
d8d1f30b | 1342 | dst_release(&grt->dst); |
1da177e4 LT |
1343 | goto out; |
1344 | } | |
1345 | } else { | |
1346 | dev = grt->rt6i_dev; | |
1347 | idev = grt->rt6i_idev; | |
1348 | dev_hold(dev); | |
1349 | in6_dev_hold(grt->rt6i_idev); | |
1350 | } | |
1351 | if (!(grt->rt6i_flags&RTF_GATEWAY)) | |
1352 | err = 0; | |
d8d1f30b | 1353 | dst_release(&grt->dst); |
1da177e4 LT |
1354 | |
1355 | if (err) | |
1356 | goto out; | |
1357 | } | |
1358 | err = -EINVAL; | |
1359 | if (dev == NULL || (dev->flags&IFF_LOOPBACK)) | |
1360 | goto out; | |
1361 | } | |
1362 | ||
1363 | err = -ENODEV; | |
1364 | if (dev == NULL) | |
1365 | goto out; | |
1366 | ||
c3968a85 DW |
1367 | if (!ipv6_addr_any(&cfg->fc_prefsrc)) { |
1368 | if (!ipv6_chk_addr(net, &cfg->fc_prefsrc, dev, 0)) { | |
1369 | err = -EINVAL; | |
1370 | goto out; | |
1371 | } | |
1372 | ipv6_addr_copy(&rt->rt6i_prefsrc.addr, &cfg->fc_prefsrc); | |
1373 | rt->rt6i_prefsrc.plen = 128; | |
1374 | } else | |
1375 | rt->rt6i_prefsrc.plen = 0; | |
1376 | ||
86872cb5 | 1377 | if (cfg->fc_flags & (RTF_GATEWAY | RTF_NONEXTHOP)) { |
69cce1d1 DM |
1378 | struct neighbour *n = __neigh_lookup_errno(&nd_tbl, &rt->rt6i_gateway, dev); |
1379 | if (IS_ERR(n)) { | |
1380 | err = PTR_ERR(n); | |
1da177e4 LT |
1381 | goto out; |
1382 | } | |
69cce1d1 | 1383 | dst_set_neighbour(&rt->dst, n); |
1da177e4 LT |
1384 | } |
1385 | ||
86872cb5 | 1386 | rt->rt6i_flags = cfg->fc_flags; |
1da177e4 LT |
1387 | |
1388 | install_route: | |
86872cb5 TG |
1389 | if (cfg->fc_mx) { |
1390 | struct nlattr *nla; | |
1391 | int remaining; | |
1392 | ||
1393 | nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) { | |
8f4c1f9b | 1394 | int type = nla_type(nla); |
86872cb5 TG |
1395 | |
1396 | if (type) { | |
1397 | if (type > RTAX_MAX) { | |
1da177e4 LT |
1398 | err = -EINVAL; |
1399 | goto out; | |
1400 | } | |
86872cb5 | 1401 | |
defb3519 | 1402 | dst_metric_set(&rt->dst, type, nla_get_u32(nla)); |
1da177e4 | 1403 | } |
1da177e4 LT |
1404 | } |
1405 | } | |
1406 | ||
d8d1f30b | 1407 | rt->dst.dev = dev; |
1da177e4 | 1408 | rt->rt6i_idev = idev; |
c71099ac | 1409 | rt->rt6i_table = table; |
63152fc0 | 1410 | |
c346dca1 | 1411 | cfg->fc_nlinfo.nl_net = dev_net(dev); |
63152fc0 | 1412 | |
86872cb5 | 1413 | return __ip6_ins_rt(rt, &cfg->fc_nlinfo); |
1da177e4 LT |
1414 | |
1415 | out: | |
1416 | if (dev) | |
1417 | dev_put(dev); | |
1418 | if (idev) | |
1419 | in6_dev_put(idev); | |
1420 | if (rt) | |
d8d1f30b | 1421 | dst_free(&rt->dst); |
1da177e4 LT |
1422 | return err; |
1423 | } | |
1424 | ||
86872cb5 | 1425 | static int __ip6_del_rt(struct rt6_info *rt, struct nl_info *info) |
1da177e4 LT |
1426 | { |
1427 | int err; | |
c71099ac | 1428 | struct fib6_table *table; |
c346dca1 | 1429 | struct net *net = dev_net(rt->rt6i_dev); |
1da177e4 | 1430 | |
8ed67789 | 1431 | if (rt == net->ipv6.ip6_null_entry) |
6c813a72 PM |
1432 | return -ENOENT; |
1433 | ||
c71099ac TG |
1434 | table = rt->rt6i_table; |
1435 | write_lock_bh(&table->tb6_lock); | |
1da177e4 | 1436 | |
86872cb5 | 1437 | err = fib6_del(rt, info); |
d8d1f30b | 1438 | dst_release(&rt->dst); |
1da177e4 | 1439 | |
c71099ac | 1440 | write_unlock_bh(&table->tb6_lock); |
1da177e4 LT |
1441 | |
1442 | return err; | |
1443 | } | |
1444 | ||
e0a1ad73 TG |
1445 | int ip6_del_rt(struct rt6_info *rt) |
1446 | { | |
4d1169c1 | 1447 | struct nl_info info = { |
c346dca1 | 1448 | .nl_net = dev_net(rt->rt6i_dev), |
4d1169c1 | 1449 | }; |
528c4ceb | 1450 | return __ip6_del_rt(rt, &info); |
e0a1ad73 TG |
1451 | } |
1452 | ||
86872cb5 | 1453 | static int ip6_route_del(struct fib6_config *cfg) |
1da177e4 | 1454 | { |
c71099ac | 1455 | struct fib6_table *table; |
1da177e4 LT |
1456 | struct fib6_node *fn; |
1457 | struct rt6_info *rt; | |
1458 | int err = -ESRCH; | |
1459 | ||
5578689a | 1460 | table = fib6_get_table(cfg->fc_nlinfo.nl_net, cfg->fc_table); |
c71099ac TG |
1461 | if (table == NULL) |
1462 | return err; | |
1463 | ||
1464 | read_lock_bh(&table->tb6_lock); | |
1da177e4 | 1465 | |
c71099ac | 1466 | fn = fib6_locate(&table->tb6_root, |
86872cb5 TG |
1467 | &cfg->fc_dst, cfg->fc_dst_len, |
1468 | &cfg->fc_src, cfg->fc_src_len); | |
1ab1457c | 1469 | |
1da177e4 | 1470 | if (fn) { |
d8d1f30b | 1471 | for (rt = fn->leaf; rt; rt = rt->dst.rt6_next) { |
86872cb5 | 1472 | if (cfg->fc_ifindex && |
1da177e4 | 1473 | (rt->rt6i_dev == NULL || |
86872cb5 | 1474 | rt->rt6i_dev->ifindex != cfg->fc_ifindex)) |
1da177e4 | 1475 | continue; |
86872cb5 TG |
1476 | if (cfg->fc_flags & RTF_GATEWAY && |
1477 | !ipv6_addr_equal(&cfg->fc_gateway, &rt->rt6i_gateway)) | |
1da177e4 | 1478 | continue; |
86872cb5 | 1479 | if (cfg->fc_metric && cfg->fc_metric != rt->rt6i_metric) |
1da177e4 | 1480 | continue; |
d8d1f30b | 1481 | dst_hold(&rt->dst); |
c71099ac | 1482 | read_unlock_bh(&table->tb6_lock); |
1da177e4 | 1483 | |
86872cb5 | 1484 | return __ip6_del_rt(rt, &cfg->fc_nlinfo); |
1da177e4 LT |
1485 | } |
1486 | } | |
c71099ac | 1487 | read_unlock_bh(&table->tb6_lock); |
1da177e4 LT |
1488 | |
1489 | return err; | |
1490 | } | |
1491 | ||
1492 | /* | |
1493 | * Handle redirects | |
1494 | */ | |
a6279458 | 1495 | struct ip6rd_flowi { |
4c9483b2 | 1496 | struct flowi6 fl6; |
a6279458 YH |
1497 | struct in6_addr gateway; |
1498 | }; | |
1499 | ||
8ed67789 DL |
1500 | static struct rt6_info *__ip6_route_redirect(struct net *net, |
1501 | struct fib6_table *table, | |
4c9483b2 | 1502 | struct flowi6 *fl6, |
a6279458 | 1503 | int flags) |
1da177e4 | 1504 | { |
4c9483b2 | 1505 | struct ip6rd_flowi *rdfl = (struct ip6rd_flowi *)fl6; |
a6279458 | 1506 | struct rt6_info *rt; |
e843b9e1 | 1507 | struct fib6_node *fn; |
c71099ac | 1508 | |
1da177e4 | 1509 | /* |
e843b9e1 YH |
1510 | * Get the "current" route for this destination and |
1511 | * check if the redirect has come from approriate router. | |
1512 | * | |
1513 | * RFC 2461 specifies that redirects should only be | |
1514 | * accepted if they come from the nexthop to the target. | |
1515 | * Due to the way the routes are chosen, this notion | |
1516 | * is a bit fuzzy and one might need to check all possible | |
1517 | * routes. | |
1da177e4 | 1518 | */ |
1da177e4 | 1519 | |
c71099ac | 1520 | read_lock_bh(&table->tb6_lock); |
4c9483b2 | 1521 | fn = fib6_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr); |
e843b9e1 | 1522 | restart: |
d8d1f30b | 1523 | for (rt = fn->leaf; rt; rt = rt->dst.rt6_next) { |
e843b9e1 YH |
1524 | /* |
1525 | * Current route is on-link; redirect is always invalid. | |
1526 | * | |
1527 | * Seems, previous statement is not true. It could | |
1528 | * be node, which looks for us as on-link (f.e. proxy ndisc) | |
1529 | * But then router serving it might decide, that we should | |
1530 | * know truth 8)8) --ANK (980726). | |
1531 | */ | |
1532 | if (rt6_check_expired(rt)) | |
1533 | continue; | |
1534 | if (!(rt->rt6i_flags & RTF_GATEWAY)) | |
1535 | continue; | |
4c9483b2 | 1536 | if (fl6->flowi6_oif != rt->rt6i_dev->ifindex) |
e843b9e1 | 1537 | continue; |
a6279458 | 1538 | if (!ipv6_addr_equal(&rdfl->gateway, &rt->rt6i_gateway)) |
e843b9e1 YH |
1539 | continue; |
1540 | break; | |
1541 | } | |
a6279458 | 1542 | |
cb15d9c2 | 1543 | if (!rt) |
8ed67789 | 1544 | rt = net->ipv6.ip6_null_entry; |
4c9483b2 | 1545 | BACKTRACK(net, &fl6->saddr); |
cb15d9c2 | 1546 | out: |
d8d1f30b | 1547 | dst_hold(&rt->dst); |
a6279458 | 1548 | |
c71099ac | 1549 | read_unlock_bh(&table->tb6_lock); |
e843b9e1 | 1550 | |
a6279458 YH |
1551 | return rt; |
1552 | }; | |
1553 | ||
b71d1d42 ED |
1554 | static struct rt6_info *ip6_route_redirect(const struct in6_addr *dest, |
1555 | const struct in6_addr *src, | |
1556 | const struct in6_addr *gateway, | |
a6279458 YH |
1557 | struct net_device *dev) |
1558 | { | |
adaa70bb | 1559 | int flags = RT6_LOOKUP_F_HAS_SADDR; |
c346dca1 | 1560 | struct net *net = dev_net(dev); |
a6279458 | 1561 | struct ip6rd_flowi rdfl = { |
4c9483b2 DM |
1562 | .fl6 = { |
1563 | .flowi6_oif = dev->ifindex, | |
1564 | .daddr = *dest, | |
1565 | .saddr = *src, | |
a6279458 | 1566 | }, |
a6279458 | 1567 | }; |
adaa70bb | 1568 | |
86c36ce4 BH |
1569 | ipv6_addr_copy(&rdfl.gateway, gateway); |
1570 | ||
adaa70bb TG |
1571 | if (rt6_need_strict(dest)) |
1572 | flags |= RT6_LOOKUP_F_IFACE; | |
a6279458 | 1573 | |
4c9483b2 | 1574 | return (struct rt6_info *)fib6_rule_lookup(net, &rdfl.fl6, |
58f09b78 | 1575 | flags, __ip6_route_redirect); |
a6279458 YH |
1576 | } |
1577 | ||
b71d1d42 ED |
1578 | void rt6_redirect(const struct in6_addr *dest, const struct in6_addr *src, |
1579 | const struct in6_addr *saddr, | |
a6279458 YH |
1580 | struct neighbour *neigh, u8 *lladdr, int on_link) |
1581 | { | |
1582 | struct rt6_info *rt, *nrt = NULL; | |
1583 | struct netevent_redirect netevent; | |
c346dca1 | 1584 | struct net *net = dev_net(neigh->dev); |
a6279458 YH |
1585 | |
1586 | rt = ip6_route_redirect(dest, src, saddr, neigh->dev); | |
1587 | ||
8ed67789 | 1588 | if (rt == net->ipv6.ip6_null_entry) { |
1da177e4 LT |
1589 | if (net_ratelimit()) |
1590 | printk(KERN_DEBUG "rt6_redirect: source isn't a valid nexthop " | |
1591 | "for redirect target\n"); | |
a6279458 | 1592 | goto out; |
1da177e4 LT |
1593 | } |
1594 | ||
1da177e4 LT |
1595 | /* |
1596 | * We have finally decided to accept it. | |
1597 | */ | |
1598 | ||
1ab1457c | 1599 | neigh_update(neigh, lladdr, NUD_STALE, |
1da177e4 LT |
1600 | NEIGH_UPDATE_F_WEAK_OVERRIDE| |
1601 | NEIGH_UPDATE_F_OVERRIDE| | |
1602 | (on_link ? 0 : (NEIGH_UPDATE_F_OVERRIDE_ISROUTER| | |
1603 | NEIGH_UPDATE_F_ISROUTER)) | |
1604 | ); | |
1605 | ||
1606 | /* | |
1607 | * Redirect received -> path was valid. | |
1608 | * Look, redirects are sent only in response to data packets, | |
1609 | * so that this nexthop apparently is reachable. --ANK | |
1610 | */ | |
d8d1f30b | 1611 | dst_confirm(&rt->dst); |
1da177e4 LT |
1612 | |
1613 | /* Duplicate redirect: silently ignore. */ | |
f2c31e32 | 1614 | if (neigh == dst_get_neighbour_raw(&rt->dst)) |
1da177e4 LT |
1615 | goto out; |
1616 | ||
21efcfa0 | 1617 | nrt = ip6_rt_copy(rt, dest); |
1da177e4 LT |
1618 | if (nrt == NULL) |
1619 | goto out; | |
1620 | ||
1621 | nrt->rt6i_flags = RTF_GATEWAY|RTF_UP|RTF_DYNAMIC|RTF_CACHE; | |
1622 | if (on_link) | |
1623 | nrt->rt6i_flags &= ~RTF_GATEWAY; | |
1624 | ||
1da177e4 | 1625 | ipv6_addr_copy(&nrt->rt6i_gateway, (struct in6_addr*)neigh->primary_key); |
69cce1d1 | 1626 | dst_set_neighbour(&nrt->dst, neigh_clone(neigh)); |
1da177e4 | 1627 | |
40e22e8f | 1628 | if (ip6_ins_rt(nrt)) |
1da177e4 LT |
1629 | goto out; |
1630 | ||
d8d1f30b CG |
1631 | netevent.old = &rt->dst; |
1632 | netevent.new = &nrt->dst; | |
8d71740c TT |
1633 | call_netevent_notifiers(NETEVENT_REDIRECT, &netevent); |
1634 | ||
1da177e4 | 1635 | if (rt->rt6i_flags&RTF_CACHE) { |
e0a1ad73 | 1636 | ip6_del_rt(rt); |
1da177e4 LT |
1637 | return; |
1638 | } | |
1639 | ||
1640 | out: | |
d8d1f30b | 1641 | dst_release(&rt->dst); |
1da177e4 LT |
1642 | } |
1643 | ||
1644 | /* | |
1645 | * Handle ICMP "packet too big" messages | |
1646 | * i.e. Path MTU discovery | |
1647 | */ | |
1648 | ||
b71d1d42 | 1649 | static void rt6_do_pmtu_disc(const struct in6_addr *daddr, const struct in6_addr *saddr, |
ae878ae2 | 1650 | struct net *net, u32 pmtu, int ifindex) |
1da177e4 LT |
1651 | { |
1652 | struct rt6_info *rt, *nrt; | |
1653 | int allfrag = 0; | |
d3052b55 | 1654 | again: |
ae878ae2 | 1655 | rt = rt6_lookup(net, daddr, saddr, ifindex, 0); |
1da177e4 LT |
1656 | if (rt == NULL) |
1657 | return; | |
1658 | ||
d3052b55 AV |
1659 | if (rt6_check_expired(rt)) { |
1660 | ip6_del_rt(rt); | |
1661 | goto again; | |
1662 | } | |
1663 | ||
d8d1f30b | 1664 | if (pmtu >= dst_mtu(&rt->dst)) |
1da177e4 LT |
1665 | goto out; |
1666 | ||
1667 | if (pmtu < IPV6_MIN_MTU) { | |
1668 | /* | |
1ab1457c | 1669 | * According to RFC2460, PMTU is set to the IPv6 Minimum Link |
1da177e4 LT |
1670 | * MTU (1280) and a fragment header should always be included |
1671 | * after a node receiving Too Big message reporting PMTU is | |
1672 | * less than the IPv6 Minimum Link MTU. | |
1673 | */ | |
1674 | pmtu = IPV6_MIN_MTU; | |
1675 | allfrag = 1; | |
1676 | } | |
1677 | ||
1678 | /* New mtu received -> path was valid. | |
1679 | They are sent only in response to data packets, | |
1680 | so that this nexthop apparently is reachable. --ANK | |
1681 | */ | |
d8d1f30b | 1682 | dst_confirm(&rt->dst); |
1da177e4 LT |
1683 | |
1684 | /* Host route. If it is static, it would be better | |
1685 | not to override it, but add new one, so that | |
1686 | when cache entry will expire old pmtu | |
1687 | would return automatically. | |
1688 | */ | |
1689 | if (rt->rt6i_flags & RTF_CACHE) { | |
defb3519 DM |
1690 | dst_metric_set(&rt->dst, RTAX_MTU, pmtu); |
1691 | if (allfrag) { | |
1692 | u32 features = dst_metric(&rt->dst, RTAX_FEATURES); | |
1693 | features |= RTAX_FEATURE_ALLFRAG; | |
1694 | dst_metric_set(&rt->dst, RTAX_FEATURES, features); | |
1695 | } | |
d8d1f30b | 1696 | dst_set_expires(&rt->dst, net->ipv6.sysctl.ip6_rt_mtu_expires); |
1da177e4 LT |
1697 | rt->rt6i_flags |= RTF_MODIFIED|RTF_EXPIRES; |
1698 | goto out; | |
1699 | } | |
1700 | ||
1701 | /* Network route. | |
1702 | Two cases are possible: | |
1703 | 1. It is connected route. Action: COW | |
1704 | 2. It is gatewayed route or NONEXTHOP route. Action: clone it. | |
1705 | */ | |
f2c31e32 | 1706 | if (!dst_get_neighbour_raw(&rt->dst) && !(rt->rt6i_flags & RTF_NONEXTHOP)) |
a1e78363 | 1707 | nrt = rt6_alloc_cow(rt, daddr, saddr); |
d5315b50 YH |
1708 | else |
1709 | nrt = rt6_alloc_clone(rt, daddr); | |
a1e78363 | 1710 | |
d5315b50 | 1711 | if (nrt) { |
defb3519 DM |
1712 | dst_metric_set(&nrt->dst, RTAX_MTU, pmtu); |
1713 | if (allfrag) { | |
1714 | u32 features = dst_metric(&nrt->dst, RTAX_FEATURES); | |
1715 | features |= RTAX_FEATURE_ALLFRAG; | |
1716 | dst_metric_set(&nrt->dst, RTAX_FEATURES, features); | |
1717 | } | |
a1e78363 YH |
1718 | |
1719 | /* According to RFC 1981, detecting PMTU increase shouldn't be | |
1720 | * happened within 5 mins, the recommended timer is 10 mins. | |
1721 | * Here this route expiration time is set to ip6_rt_mtu_expires | |
1722 | * which is 10 mins. After 10 mins the decreased pmtu is expired | |
1723 | * and detecting PMTU increase will be automatically happened. | |
1724 | */ | |
d8d1f30b | 1725 | dst_set_expires(&nrt->dst, net->ipv6.sysctl.ip6_rt_mtu_expires); |
a1e78363 YH |
1726 | nrt->rt6i_flags |= RTF_DYNAMIC|RTF_EXPIRES; |
1727 | ||
40e22e8f | 1728 | ip6_ins_rt(nrt); |
1da177e4 | 1729 | } |
1da177e4 | 1730 | out: |
d8d1f30b | 1731 | dst_release(&rt->dst); |
1da177e4 LT |
1732 | } |
1733 | ||
b71d1d42 | 1734 | void rt6_pmtu_discovery(const struct in6_addr *daddr, const struct in6_addr *saddr, |
ae878ae2 MŻ |
1735 | struct net_device *dev, u32 pmtu) |
1736 | { | |
1737 | struct net *net = dev_net(dev); | |
1738 | ||
1739 | /* | |
1740 | * RFC 1981 states that a node "MUST reduce the size of the packets it | |
1741 | * is sending along the path" that caused the Packet Too Big message. | |
1742 | * Since it's not possible in the general case to determine which | |
1743 | * interface was used to send the original packet, we update the MTU | |
1744 | * on the interface that will be used to send future packets. We also | |
1745 | * update the MTU on the interface that received the Packet Too Big in | |
1746 | * case the original packet was forced out that interface with | |
1747 | * SO_BINDTODEVICE or similar. This is the next best thing to the | |
1748 | * correct behaviour, which would be to update the MTU on all | |
1749 | * interfaces. | |
1750 | */ | |
1751 | rt6_do_pmtu_disc(daddr, saddr, net, pmtu, 0); | |
1752 | rt6_do_pmtu_disc(daddr, saddr, net, pmtu, dev->ifindex); | |
1753 | } | |
1754 | ||
1da177e4 LT |
1755 | /* |
1756 | * Misc support functions | |
1757 | */ | |
1758 | ||
21efcfa0 ED |
1759 | static struct rt6_info *ip6_rt_copy(const struct rt6_info *ort, |
1760 | const struct in6_addr *dest) | |
1da177e4 | 1761 | { |
c346dca1 | 1762 | struct net *net = dev_net(ort->rt6i_dev); |
5c1e6aa3 | 1763 | struct rt6_info *rt = ip6_dst_alloc(&net->ipv6.ip6_dst_ops, |
957c665f | 1764 | ort->dst.dev, 0); |
1da177e4 LT |
1765 | |
1766 | if (rt) { | |
d8d1f30b CG |
1767 | rt->dst.input = ort->dst.input; |
1768 | rt->dst.output = ort->dst.output; | |
8e2ec639 | 1769 | rt->dst.flags |= DST_HOST; |
d8d1f30b | 1770 | |
21efcfa0 | 1771 | ipv6_addr_copy(&rt->rt6i_dst.addr, dest); |
8e2ec639 | 1772 | rt->rt6i_dst.plen = 128; |
defb3519 | 1773 | dst_copy_metrics(&rt->dst, &ort->dst); |
d8d1f30b | 1774 | rt->dst.error = ort->dst.error; |
1da177e4 LT |
1775 | rt->rt6i_idev = ort->rt6i_idev; |
1776 | if (rt->rt6i_idev) | |
1777 | in6_dev_hold(rt->rt6i_idev); | |
d8d1f30b | 1778 | rt->dst.lastuse = jiffies; |
1da177e4 LT |
1779 | rt->rt6i_expires = 0; |
1780 | ||
1781 | ipv6_addr_copy(&rt->rt6i_gateway, &ort->rt6i_gateway); | |
1782 | rt->rt6i_flags = ort->rt6i_flags & ~RTF_EXPIRES; | |
1783 | rt->rt6i_metric = 0; | |
1784 | ||
1da177e4 LT |
1785 | #ifdef CONFIG_IPV6_SUBTREES |
1786 | memcpy(&rt->rt6i_src, &ort->rt6i_src, sizeof(struct rt6key)); | |
1787 | #endif | |
0f6c6392 | 1788 | memcpy(&rt->rt6i_prefsrc, &ort->rt6i_prefsrc, sizeof(struct rt6key)); |
c71099ac | 1789 | rt->rt6i_table = ort->rt6i_table; |
1da177e4 LT |
1790 | } |
1791 | return rt; | |
1792 | } | |
1793 | ||
70ceb4f5 | 1794 | #ifdef CONFIG_IPV6_ROUTE_INFO |
efa2cea0 | 1795 | static struct rt6_info *rt6_get_route_info(struct net *net, |
b71d1d42 ED |
1796 | const struct in6_addr *prefix, int prefixlen, |
1797 | const struct in6_addr *gwaddr, int ifindex) | |
70ceb4f5 YH |
1798 | { |
1799 | struct fib6_node *fn; | |
1800 | struct rt6_info *rt = NULL; | |
c71099ac TG |
1801 | struct fib6_table *table; |
1802 | ||
efa2cea0 | 1803 | table = fib6_get_table(net, RT6_TABLE_INFO); |
c71099ac TG |
1804 | if (table == NULL) |
1805 | return NULL; | |
70ceb4f5 | 1806 | |
c71099ac TG |
1807 | write_lock_bh(&table->tb6_lock); |
1808 | fn = fib6_locate(&table->tb6_root, prefix ,prefixlen, NULL, 0); | |
70ceb4f5 YH |
1809 | if (!fn) |
1810 | goto out; | |
1811 | ||
d8d1f30b | 1812 | for (rt = fn->leaf; rt; rt = rt->dst.rt6_next) { |
70ceb4f5 YH |
1813 | if (rt->rt6i_dev->ifindex != ifindex) |
1814 | continue; | |
1815 | if ((rt->rt6i_flags & (RTF_ROUTEINFO|RTF_GATEWAY)) != (RTF_ROUTEINFO|RTF_GATEWAY)) | |
1816 | continue; | |
1817 | if (!ipv6_addr_equal(&rt->rt6i_gateway, gwaddr)) | |
1818 | continue; | |
d8d1f30b | 1819 | dst_hold(&rt->dst); |
70ceb4f5 YH |
1820 | break; |
1821 | } | |
1822 | out: | |
c71099ac | 1823 | write_unlock_bh(&table->tb6_lock); |
70ceb4f5 YH |
1824 | return rt; |
1825 | } | |
1826 | ||
efa2cea0 | 1827 | static struct rt6_info *rt6_add_route_info(struct net *net, |
b71d1d42 ED |
1828 | const struct in6_addr *prefix, int prefixlen, |
1829 | const struct in6_addr *gwaddr, int ifindex, | |
70ceb4f5 YH |
1830 | unsigned pref) |
1831 | { | |
86872cb5 TG |
1832 | struct fib6_config cfg = { |
1833 | .fc_table = RT6_TABLE_INFO, | |
238fc7ea | 1834 | .fc_metric = IP6_RT_PRIO_USER, |
86872cb5 TG |
1835 | .fc_ifindex = ifindex, |
1836 | .fc_dst_len = prefixlen, | |
1837 | .fc_flags = RTF_GATEWAY | RTF_ADDRCONF | RTF_ROUTEINFO | | |
1838 | RTF_UP | RTF_PREF(pref), | |
efa2cea0 DL |
1839 | .fc_nlinfo.pid = 0, |
1840 | .fc_nlinfo.nlh = NULL, | |
1841 | .fc_nlinfo.nl_net = net, | |
86872cb5 TG |
1842 | }; |
1843 | ||
1844 | ipv6_addr_copy(&cfg.fc_dst, prefix); | |
1845 | ipv6_addr_copy(&cfg.fc_gateway, gwaddr); | |
70ceb4f5 | 1846 | |
e317da96 YH |
1847 | /* We should treat it as a default route if prefix length is 0. */ |
1848 | if (!prefixlen) | |
86872cb5 | 1849 | cfg.fc_flags |= RTF_DEFAULT; |
70ceb4f5 | 1850 | |
86872cb5 | 1851 | ip6_route_add(&cfg); |
70ceb4f5 | 1852 | |
efa2cea0 | 1853 | return rt6_get_route_info(net, prefix, prefixlen, gwaddr, ifindex); |
70ceb4f5 YH |
1854 | } |
1855 | #endif | |
1856 | ||
b71d1d42 | 1857 | struct rt6_info *rt6_get_dflt_router(const struct in6_addr *addr, struct net_device *dev) |
1ab1457c | 1858 | { |
1da177e4 | 1859 | struct rt6_info *rt; |
c71099ac | 1860 | struct fib6_table *table; |
1da177e4 | 1861 | |
c346dca1 | 1862 | table = fib6_get_table(dev_net(dev), RT6_TABLE_DFLT); |
c71099ac TG |
1863 | if (table == NULL) |
1864 | return NULL; | |
1da177e4 | 1865 | |
c71099ac | 1866 | write_lock_bh(&table->tb6_lock); |
d8d1f30b | 1867 | for (rt = table->tb6_root.leaf; rt; rt=rt->dst.rt6_next) { |
1da177e4 | 1868 | if (dev == rt->rt6i_dev && |
045927ff | 1869 | ((rt->rt6i_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) && |
1da177e4 LT |
1870 | ipv6_addr_equal(&rt->rt6i_gateway, addr)) |
1871 | break; | |
1872 | } | |
1873 | if (rt) | |
d8d1f30b | 1874 | dst_hold(&rt->dst); |
c71099ac | 1875 | write_unlock_bh(&table->tb6_lock); |
1da177e4 LT |
1876 | return rt; |
1877 | } | |
1878 | ||
b71d1d42 | 1879 | struct rt6_info *rt6_add_dflt_router(const struct in6_addr *gwaddr, |
ebacaaa0 YH |
1880 | struct net_device *dev, |
1881 | unsigned int pref) | |
1da177e4 | 1882 | { |
86872cb5 TG |
1883 | struct fib6_config cfg = { |
1884 | .fc_table = RT6_TABLE_DFLT, | |
238fc7ea | 1885 | .fc_metric = IP6_RT_PRIO_USER, |
86872cb5 TG |
1886 | .fc_ifindex = dev->ifindex, |
1887 | .fc_flags = RTF_GATEWAY | RTF_ADDRCONF | RTF_DEFAULT | | |
1888 | RTF_UP | RTF_EXPIRES | RTF_PREF(pref), | |
5578689a DL |
1889 | .fc_nlinfo.pid = 0, |
1890 | .fc_nlinfo.nlh = NULL, | |
c346dca1 | 1891 | .fc_nlinfo.nl_net = dev_net(dev), |
86872cb5 | 1892 | }; |
1da177e4 | 1893 | |
86872cb5 | 1894 | ipv6_addr_copy(&cfg.fc_gateway, gwaddr); |
1da177e4 | 1895 | |
86872cb5 | 1896 | ip6_route_add(&cfg); |
1da177e4 | 1897 | |
1da177e4 LT |
1898 | return rt6_get_dflt_router(gwaddr, dev); |
1899 | } | |
1900 | ||
7b4da532 | 1901 | void rt6_purge_dflt_routers(struct net *net) |
1da177e4 LT |
1902 | { |
1903 | struct rt6_info *rt; | |
c71099ac TG |
1904 | struct fib6_table *table; |
1905 | ||
1906 | /* NOTE: Keep consistent with rt6_get_dflt_router */ | |
7b4da532 | 1907 | table = fib6_get_table(net, RT6_TABLE_DFLT); |
c71099ac TG |
1908 | if (table == NULL) |
1909 | return; | |
1da177e4 LT |
1910 | |
1911 | restart: | |
c71099ac | 1912 | read_lock_bh(&table->tb6_lock); |
d8d1f30b | 1913 | for (rt = table->tb6_root.leaf; rt; rt = rt->dst.rt6_next) { |
1da177e4 | 1914 | if (rt->rt6i_flags & (RTF_DEFAULT | RTF_ADDRCONF)) { |
d8d1f30b | 1915 | dst_hold(&rt->dst); |
c71099ac | 1916 | read_unlock_bh(&table->tb6_lock); |
e0a1ad73 | 1917 | ip6_del_rt(rt); |
1da177e4 LT |
1918 | goto restart; |
1919 | } | |
1920 | } | |
c71099ac | 1921 | read_unlock_bh(&table->tb6_lock); |
1da177e4 LT |
1922 | } |
1923 | ||
5578689a DL |
1924 | static void rtmsg_to_fib6_config(struct net *net, |
1925 | struct in6_rtmsg *rtmsg, | |
86872cb5 TG |
1926 | struct fib6_config *cfg) |
1927 | { | |
1928 | memset(cfg, 0, sizeof(*cfg)); | |
1929 | ||
1930 | cfg->fc_table = RT6_TABLE_MAIN; | |
1931 | cfg->fc_ifindex = rtmsg->rtmsg_ifindex; | |
1932 | cfg->fc_metric = rtmsg->rtmsg_metric; | |
1933 | cfg->fc_expires = rtmsg->rtmsg_info; | |
1934 | cfg->fc_dst_len = rtmsg->rtmsg_dst_len; | |
1935 | cfg->fc_src_len = rtmsg->rtmsg_src_len; | |
1936 | cfg->fc_flags = rtmsg->rtmsg_flags; | |
1937 | ||
5578689a | 1938 | cfg->fc_nlinfo.nl_net = net; |
f1243c2d | 1939 | |
86872cb5 TG |
1940 | ipv6_addr_copy(&cfg->fc_dst, &rtmsg->rtmsg_dst); |
1941 | ipv6_addr_copy(&cfg->fc_src, &rtmsg->rtmsg_src); | |
1942 | ipv6_addr_copy(&cfg->fc_gateway, &rtmsg->rtmsg_gateway); | |
1943 | } | |
1944 | ||
5578689a | 1945 | int ipv6_route_ioctl(struct net *net, unsigned int cmd, void __user *arg) |
1da177e4 | 1946 | { |
86872cb5 | 1947 | struct fib6_config cfg; |
1da177e4 LT |
1948 | struct in6_rtmsg rtmsg; |
1949 | int err; | |
1950 | ||
1951 | switch(cmd) { | |
1952 | case SIOCADDRT: /* Add a route */ | |
1953 | case SIOCDELRT: /* Delete a route */ | |
1954 | if (!capable(CAP_NET_ADMIN)) | |
1955 | return -EPERM; | |
1956 | err = copy_from_user(&rtmsg, arg, | |
1957 | sizeof(struct in6_rtmsg)); | |
1958 | if (err) | |
1959 | return -EFAULT; | |
86872cb5 | 1960 | |
5578689a | 1961 | rtmsg_to_fib6_config(net, &rtmsg, &cfg); |
86872cb5 | 1962 | |
1da177e4 LT |
1963 | rtnl_lock(); |
1964 | switch (cmd) { | |
1965 | case SIOCADDRT: | |
86872cb5 | 1966 | err = ip6_route_add(&cfg); |
1da177e4 LT |
1967 | break; |
1968 | case SIOCDELRT: | |
86872cb5 | 1969 | err = ip6_route_del(&cfg); |
1da177e4 LT |
1970 | break; |
1971 | default: | |
1972 | err = -EINVAL; | |
1973 | } | |
1974 | rtnl_unlock(); | |
1975 | ||
1976 | return err; | |
3ff50b79 | 1977 | } |
1da177e4 LT |
1978 | |
1979 | return -EINVAL; | |
1980 | } | |
1981 | ||
1982 | /* | |
1983 | * Drop the packet on the floor | |
1984 | */ | |
1985 | ||
d5fdd6ba | 1986 | static int ip6_pkt_drop(struct sk_buff *skb, u8 code, int ipstats_mib_noroutes) |
1da177e4 | 1987 | { |
612f09e8 | 1988 | int type; |
adf30907 | 1989 | struct dst_entry *dst = skb_dst(skb); |
612f09e8 YH |
1990 | switch (ipstats_mib_noroutes) { |
1991 | case IPSTATS_MIB_INNOROUTES: | |
0660e03f | 1992 | type = ipv6_addr_type(&ipv6_hdr(skb)->daddr); |
45bb0060 | 1993 | if (type == IPV6_ADDR_ANY) { |
3bd653c8 DL |
1994 | IP6_INC_STATS(dev_net(dst->dev), ip6_dst_idev(dst), |
1995 | IPSTATS_MIB_INADDRERRORS); | |
612f09e8 YH |
1996 | break; |
1997 | } | |
1998 | /* FALLTHROUGH */ | |
1999 | case IPSTATS_MIB_OUTNOROUTES: | |
3bd653c8 DL |
2000 | IP6_INC_STATS(dev_net(dst->dev), ip6_dst_idev(dst), |
2001 | ipstats_mib_noroutes); | |
612f09e8 YH |
2002 | break; |
2003 | } | |
3ffe533c | 2004 | icmpv6_send(skb, ICMPV6_DEST_UNREACH, code, 0); |
1da177e4 LT |
2005 | kfree_skb(skb); |
2006 | return 0; | |
2007 | } | |
2008 | ||
9ce8ade0 TG |
2009 | static int ip6_pkt_discard(struct sk_buff *skb) |
2010 | { | |
612f09e8 | 2011 | return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_INNOROUTES); |
9ce8ade0 TG |
2012 | } |
2013 | ||
20380731 | 2014 | static int ip6_pkt_discard_out(struct sk_buff *skb) |
1da177e4 | 2015 | { |
adf30907 | 2016 | skb->dev = skb_dst(skb)->dev; |
612f09e8 | 2017 | return ip6_pkt_drop(skb, ICMPV6_NOROUTE, IPSTATS_MIB_OUTNOROUTES); |
1da177e4 LT |
2018 | } |
2019 | ||
6723ab54 DM |
2020 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES |
2021 | ||
9ce8ade0 TG |
2022 | static int ip6_pkt_prohibit(struct sk_buff *skb) |
2023 | { | |
612f09e8 | 2024 | return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_INNOROUTES); |
9ce8ade0 TG |
2025 | } |
2026 | ||
2027 | static int ip6_pkt_prohibit_out(struct sk_buff *skb) | |
2028 | { | |
adf30907 | 2029 | skb->dev = skb_dst(skb)->dev; |
612f09e8 | 2030 | return ip6_pkt_drop(skb, ICMPV6_ADM_PROHIBITED, IPSTATS_MIB_OUTNOROUTES); |
9ce8ade0 TG |
2031 | } |
2032 | ||
6723ab54 DM |
2033 | #endif |
2034 | ||
1da177e4 LT |
2035 | /* |
2036 | * Allocate a dst for local (unicast / anycast) address. | |
2037 | */ | |
2038 | ||
2039 | struct rt6_info *addrconf_dst_alloc(struct inet6_dev *idev, | |
2040 | const struct in6_addr *addr, | |
2041 | int anycast) | |
2042 | { | |
c346dca1 | 2043 | struct net *net = dev_net(idev->dev); |
5c1e6aa3 | 2044 | struct rt6_info *rt = ip6_dst_alloc(&net->ipv6.ip6_dst_ops, |
957c665f | 2045 | net->loopback_dev, 0); |
14deae41 | 2046 | struct neighbour *neigh; |
1da177e4 | 2047 | |
40385653 BG |
2048 | if (rt == NULL) { |
2049 | if (net_ratelimit()) | |
2050 | pr_warning("IPv6: Maximum number of routes reached," | |
2051 | " consider increasing route/max_size.\n"); | |
1da177e4 | 2052 | return ERR_PTR(-ENOMEM); |
40385653 | 2053 | } |
1da177e4 | 2054 | |
1da177e4 LT |
2055 | in6_dev_hold(idev); |
2056 | ||
11d53b49 | 2057 | rt->dst.flags |= DST_HOST; |
d8d1f30b CG |
2058 | rt->dst.input = ip6_input; |
2059 | rt->dst.output = ip6_output; | |
1da177e4 | 2060 | rt->rt6i_idev = idev; |
d8d1f30b | 2061 | rt->dst.obsolete = -1; |
1da177e4 LT |
2062 | |
2063 | rt->rt6i_flags = RTF_UP | RTF_NONEXTHOP; | |
58c4fb86 YH |
2064 | if (anycast) |
2065 | rt->rt6i_flags |= RTF_ANYCAST; | |
2066 | else | |
1da177e4 | 2067 | rt->rt6i_flags |= RTF_LOCAL; |
14deae41 DM |
2068 | neigh = ndisc_get_neigh(rt->rt6i_dev, &rt->rt6i_gateway); |
2069 | if (IS_ERR(neigh)) { | |
d8d1f30b | 2070 | dst_free(&rt->dst); |
14deae41 | 2071 | |
29546a64 | 2072 | return ERR_CAST(neigh); |
1da177e4 | 2073 | } |
69cce1d1 | 2074 | dst_set_neighbour(&rt->dst, neigh); |
1da177e4 LT |
2075 | |
2076 | ipv6_addr_copy(&rt->rt6i_dst.addr, addr); | |
2077 | rt->rt6i_dst.plen = 128; | |
5578689a | 2078 | rt->rt6i_table = fib6_get_table(net, RT6_TABLE_LOCAL); |
1da177e4 | 2079 | |
d8d1f30b | 2080 | atomic_set(&rt->dst.__refcnt, 1); |
1da177e4 LT |
2081 | |
2082 | return rt; | |
2083 | } | |
2084 | ||
c3968a85 DW |
2085 | int ip6_route_get_saddr(struct net *net, |
2086 | struct rt6_info *rt, | |
b71d1d42 | 2087 | const struct in6_addr *daddr, |
c3968a85 DW |
2088 | unsigned int prefs, |
2089 | struct in6_addr *saddr) | |
2090 | { | |
2091 | struct inet6_dev *idev = ip6_dst_idev((struct dst_entry*)rt); | |
2092 | int err = 0; | |
2093 | if (rt->rt6i_prefsrc.plen) | |
2094 | ipv6_addr_copy(saddr, &rt->rt6i_prefsrc.addr); | |
2095 | else | |
2096 | err = ipv6_dev_get_saddr(net, idev ? idev->dev : NULL, | |
2097 | daddr, prefs, saddr); | |
2098 | return err; | |
2099 | } | |
2100 | ||
2101 | /* remove deleted ip from prefsrc entries */ | |
2102 | struct arg_dev_net_ip { | |
2103 | struct net_device *dev; | |
2104 | struct net *net; | |
2105 | struct in6_addr *addr; | |
2106 | }; | |
2107 | ||
2108 | static int fib6_remove_prefsrc(struct rt6_info *rt, void *arg) | |
2109 | { | |
2110 | struct net_device *dev = ((struct arg_dev_net_ip *)arg)->dev; | |
2111 | struct net *net = ((struct arg_dev_net_ip *)arg)->net; | |
2112 | struct in6_addr *addr = ((struct arg_dev_net_ip *)arg)->addr; | |
2113 | ||
2114 | if (((void *)rt->rt6i_dev == dev || dev == NULL) && | |
2115 | rt != net->ipv6.ip6_null_entry && | |
2116 | ipv6_addr_equal(addr, &rt->rt6i_prefsrc.addr)) { | |
2117 | /* remove prefsrc entry */ | |
2118 | rt->rt6i_prefsrc.plen = 0; | |
2119 | } | |
2120 | return 0; | |
2121 | } | |
2122 | ||
2123 | void rt6_remove_prefsrc(struct inet6_ifaddr *ifp) | |
2124 | { | |
2125 | struct net *net = dev_net(ifp->idev->dev); | |
2126 | struct arg_dev_net_ip adni = { | |
2127 | .dev = ifp->idev->dev, | |
2128 | .net = net, | |
2129 | .addr = &ifp->addr, | |
2130 | }; | |
2131 | fib6_clean_all(net, fib6_remove_prefsrc, 0, &adni); | |
2132 | } | |
2133 | ||
8ed67789 DL |
2134 | struct arg_dev_net { |
2135 | struct net_device *dev; | |
2136 | struct net *net; | |
2137 | }; | |
2138 | ||
1da177e4 LT |
2139 | static int fib6_ifdown(struct rt6_info *rt, void *arg) |
2140 | { | |
bc3ef660 | 2141 | const struct arg_dev_net *adn = arg; |
2142 | const struct net_device *dev = adn->dev; | |
8ed67789 | 2143 | |
bc3ef660 | 2144 | if ((rt->rt6i_dev == dev || dev == NULL) && |
2145 | rt != adn->net->ipv6.ip6_null_entry) { | |
1da177e4 LT |
2146 | RT6_TRACE("deleted by ifdown %p\n", rt); |
2147 | return -1; | |
2148 | } | |
2149 | return 0; | |
2150 | } | |
2151 | ||
f3db4851 | 2152 | void rt6_ifdown(struct net *net, struct net_device *dev) |
1da177e4 | 2153 | { |
8ed67789 DL |
2154 | struct arg_dev_net adn = { |
2155 | .dev = dev, | |
2156 | .net = net, | |
2157 | }; | |
2158 | ||
2159 | fib6_clean_all(net, fib6_ifdown, 0, &adn); | |
1e493d19 | 2160 | icmp6_clean_all(fib6_ifdown, &adn); |
1da177e4 LT |
2161 | } |
2162 | ||
2163 | struct rt6_mtu_change_arg | |
2164 | { | |
2165 | struct net_device *dev; | |
2166 | unsigned mtu; | |
2167 | }; | |
2168 | ||
2169 | static int rt6_mtu_change_route(struct rt6_info *rt, void *p_arg) | |
2170 | { | |
2171 | struct rt6_mtu_change_arg *arg = (struct rt6_mtu_change_arg *) p_arg; | |
2172 | struct inet6_dev *idev; | |
2173 | ||
2174 | /* In IPv6 pmtu discovery is not optional, | |
2175 | so that RTAX_MTU lock cannot disable it. | |
2176 | We still use this lock to block changes | |
2177 | caused by addrconf/ndisc. | |
2178 | */ | |
2179 | ||
2180 | idev = __in6_dev_get(arg->dev); | |
2181 | if (idev == NULL) | |
2182 | return 0; | |
2183 | ||
2184 | /* For administrative MTU increase, there is no way to discover | |
2185 | IPv6 PMTU increase, so PMTU increase should be updated here. | |
2186 | Since RFC 1981 doesn't include administrative MTU increase | |
2187 | update PMTU increase is a MUST. (i.e. jumbo frame) | |
2188 | */ | |
2189 | /* | |
2190 | If new MTU is less than route PMTU, this new MTU will be the | |
2191 | lowest MTU in the path, update the route PMTU to reflect PMTU | |
2192 | decreases; if new MTU is greater than route PMTU, and the | |
2193 | old MTU is the lowest MTU in the path, update the route PMTU | |
2194 | to reflect the increase. In this case if the other nodes' MTU | |
2195 | also have the lowest MTU, TOO BIG MESSAGE will be lead to | |
2196 | PMTU discouvery. | |
2197 | */ | |
2198 | if (rt->rt6i_dev == arg->dev && | |
d8d1f30b CG |
2199 | !dst_metric_locked(&rt->dst, RTAX_MTU) && |
2200 | (dst_mtu(&rt->dst) >= arg->mtu || | |
2201 | (dst_mtu(&rt->dst) < arg->mtu && | |
2202 | dst_mtu(&rt->dst) == idev->cnf.mtu6))) { | |
defb3519 | 2203 | dst_metric_set(&rt->dst, RTAX_MTU, arg->mtu); |
566cfd8f | 2204 | } |
1da177e4 LT |
2205 | return 0; |
2206 | } | |
2207 | ||
2208 | void rt6_mtu_change(struct net_device *dev, unsigned mtu) | |
2209 | { | |
c71099ac TG |
2210 | struct rt6_mtu_change_arg arg = { |
2211 | .dev = dev, | |
2212 | .mtu = mtu, | |
2213 | }; | |
1da177e4 | 2214 | |
c346dca1 | 2215 | fib6_clean_all(dev_net(dev), rt6_mtu_change_route, 0, &arg); |
1da177e4 LT |
2216 | } |
2217 | ||
ef7c79ed | 2218 | static const struct nla_policy rtm_ipv6_policy[RTA_MAX+1] = { |
5176f91e | 2219 | [RTA_GATEWAY] = { .len = sizeof(struct in6_addr) }, |
86872cb5 | 2220 | [RTA_OIF] = { .type = NLA_U32 }, |
ab364a6f | 2221 | [RTA_IIF] = { .type = NLA_U32 }, |
86872cb5 TG |
2222 | [RTA_PRIORITY] = { .type = NLA_U32 }, |
2223 | [RTA_METRICS] = { .type = NLA_NESTED }, | |
2224 | }; | |
2225 | ||
2226 | static int rtm_to_fib6_config(struct sk_buff *skb, struct nlmsghdr *nlh, | |
2227 | struct fib6_config *cfg) | |
1da177e4 | 2228 | { |
86872cb5 TG |
2229 | struct rtmsg *rtm; |
2230 | struct nlattr *tb[RTA_MAX+1]; | |
2231 | int err; | |
1da177e4 | 2232 | |
86872cb5 TG |
2233 | err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv6_policy); |
2234 | if (err < 0) | |
2235 | goto errout; | |
1da177e4 | 2236 | |
86872cb5 TG |
2237 | err = -EINVAL; |
2238 | rtm = nlmsg_data(nlh); | |
2239 | memset(cfg, 0, sizeof(*cfg)); | |
2240 | ||
2241 | cfg->fc_table = rtm->rtm_table; | |
2242 | cfg->fc_dst_len = rtm->rtm_dst_len; | |
2243 | cfg->fc_src_len = rtm->rtm_src_len; | |
2244 | cfg->fc_flags = RTF_UP; | |
2245 | cfg->fc_protocol = rtm->rtm_protocol; | |
2246 | ||
2247 | if (rtm->rtm_type == RTN_UNREACHABLE) | |
2248 | cfg->fc_flags |= RTF_REJECT; | |
2249 | ||
ab79ad14 MŻ |
2250 | if (rtm->rtm_type == RTN_LOCAL) |
2251 | cfg->fc_flags |= RTF_LOCAL; | |
2252 | ||
86872cb5 TG |
2253 | cfg->fc_nlinfo.pid = NETLINK_CB(skb).pid; |
2254 | cfg->fc_nlinfo.nlh = nlh; | |
3b1e0a65 | 2255 | cfg->fc_nlinfo.nl_net = sock_net(skb->sk); |
86872cb5 TG |
2256 | |
2257 | if (tb[RTA_GATEWAY]) { | |
2258 | nla_memcpy(&cfg->fc_gateway, tb[RTA_GATEWAY], 16); | |
2259 | cfg->fc_flags |= RTF_GATEWAY; | |
1da177e4 | 2260 | } |
86872cb5 TG |
2261 | |
2262 | if (tb[RTA_DST]) { | |
2263 | int plen = (rtm->rtm_dst_len + 7) >> 3; | |
2264 | ||
2265 | if (nla_len(tb[RTA_DST]) < plen) | |
2266 | goto errout; | |
2267 | ||
2268 | nla_memcpy(&cfg->fc_dst, tb[RTA_DST], plen); | |
1da177e4 | 2269 | } |
86872cb5 TG |
2270 | |
2271 | if (tb[RTA_SRC]) { | |
2272 | int plen = (rtm->rtm_src_len + 7) >> 3; | |
2273 | ||
2274 | if (nla_len(tb[RTA_SRC]) < plen) | |
2275 | goto errout; | |
2276 | ||
2277 | nla_memcpy(&cfg->fc_src, tb[RTA_SRC], plen); | |
1da177e4 | 2278 | } |
86872cb5 | 2279 | |
c3968a85 DW |
2280 | if (tb[RTA_PREFSRC]) |
2281 | nla_memcpy(&cfg->fc_prefsrc, tb[RTA_PREFSRC], 16); | |
2282 | ||
86872cb5 TG |
2283 | if (tb[RTA_OIF]) |
2284 | cfg->fc_ifindex = nla_get_u32(tb[RTA_OIF]); | |
2285 | ||
2286 | if (tb[RTA_PRIORITY]) | |
2287 | cfg->fc_metric = nla_get_u32(tb[RTA_PRIORITY]); | |
2288 | ||
2289 | if (tb[RTA_METRICS]) { | |
2290 | cfg->fc_mx = nla_data(tb[RTA_METRICS]); | |
2291 | cfg->fc_mx_len = nla_len(tb[RTA_METRICS]); | |
1da177e4 | 2292 | } |
86872cb5 TG |
2293 | |
2294 | if (tb[RTA_TABLE]) | |
2295 | cfg->fc_table = nla_get_u32(tb[RTA_TABLE]); | |
2296 | ||
2297 | err = 0; | |
2298 | errout: | |
2299 | return err; | |
1da177e4 LT |
2300 | } |
2301 | ||
c127ea2c | 2302 | static int inet6_rtm_delroute(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg) |
1da177e4 | 2303 | { |
86872cb5 TG |
2304 | struct fib6_config cfg; |
2305 | int err; | |
1da177e4 | 2306 | |
86872cb5 TG |
2307 | err = rtm_to_fib6_config(skb, nlh, &cfg); |
2308 | if (err < 0) | |
2309 | return err; | |
2310 | ||
2311 | return ip6_route_del(&cfg); | |
1da177e4 LT |
2312 | } |
2313 | ||
c127ea2c | 2314 | static int inet6_rtm_newroute(struct sk_buff *skb, struct nlmsghdr* nlh, void *arg) |
1da177e4 | 2315 | { |
86872cb5 TG |
2316 | struct fib6_config cfg; |
2317 | int err; | |
1da177e4 | 2318 | |
86872cb5 TG |
2319 | err = rtm_to_fib6_config(skb, nlh, &cfg); |
2320 | if (err < 0) | |
2321 | return err; | |
2322 | ||
2323 | return ip6_route_add(&cfg); | |
1da177e4 LT |
2324 | } |
2325 | ||
339bf98f TG |
2326 | static inline size_t rt6_nlmsg_size(void) |
2327 | { | |
2328 | return NLMSG_ALIGN(sizeof(struct rtmsg)) | |
2329 | + nla_total_size(16) /* RTA_SRC */ | |
2330 | + nla_total_size(16) /* RTA_DST */ | |
2331 | + nla_total_size(16) /* RTA_GATEWAY */ | |
2332 | + nla_total_size(16) /* RTA_PREFSRC */ | |
2333 | + nla_total_size(4) /* RTA_TABLE */ | |
2334 | + nla_total_size(4) /* RTA_IIF */ | |
2335 | + nla_total_size(4) /* RTA_OIF */ | |
2336 | + nla_total_size(4) /* RTA_PRIORITY */ | |
6a2b9ce0 | 2337 | + RTAX_MAX * nla_total_size(4) /* RTA_METRICS */ |
339bf98f TG |
2338 | + nla_total_size(sizeof(struct rta_cacheinfo)); |
2339 | } | |
2340 | ||
191cd582 BH |
2341 | static int rt6_fill_node(struct net *net, |
2342 | struct sk_buff *skb, struct rt6_info *rt, | |
0d51aa80 JHS |
2343 | struct in6_addr *dst, struct in6_addr *src, |
2344 | int iif, int type, u32 pid, u32 seq, | |
7bc570c8 | 2345 | int prefix, int nowait, unsigned int flags) |
1da177e4 LT |
2346 | { |
2347 | struct rtmsg *rtm; | |
2d7202bf | 2348 | struct nlmsghdr *nlh; |
e3703b3d | 2349 | long expires; |
9e762a4a | 2350 | u32 table; |
f2c31e32 | 2351 | struct neighbour *n; |
1da177e4 LT |
2352 | |
2353 | if (prefix) { /* user wants prefix routes only */ | |
2354 | if (!(rt->rt6i_flags & RTF_PREFIX_RT)) { | |
2355 | /* success since this is not a prefix route */ | |
2356 | return 1; | |
2357 | } | |
2358 | } | |
2359 | ||
2d7202bf TG |
2360 | nlh = nlmsg_put(skb, pid, seq, type, sizeof(*rtm), flags); |
2361 | if (nlh == NULL) | |
26932566 | 2362 | return -EMSGSIZE; |
2d7202bf TG |
2363 | |
2364 | rtm = nlmsg_data(nlh); | |
1da177e4 LT |
2365 | rtm->rtm_family = AF_INET6; |
2366 | rtm->rtm_dst_len = rt->rt6i_dst.plen; | |
2367 | rtm->rtm_src_len = rt->rt6i_src.plen; | |
2368 | rtm->rtm_tos = 0; | |
c71099ac | 2369 | if (rt->rt6i_table) |
9e762a4a | 2370 | table = rt->rt6i_table->tb6_id; |
c71099ac | 2371 | else |
9e762a4a PM |
2372 | table = RT6_TABLE_UNSPEC; |
2373 | rtm->rtm_table = table; | |
2d7202bf | 2374 | NLA_PUT_U32(skb, RTA_TABLE, table); |
1da177e4 LT |
2375 | if (rt->rt6i_flags&RTF_REJECT) |
2376 | rtm->rtm_type = RTN_UNREACHABLE; | |
ab79ad14 MŻ |
2377 | else if (rt->rt6i_flags&RTF_LOCAL) |
2378 | rtm->rtm_type = RTN_LOCAL; | |
1da177e4 LT |
2379 | else if (rt->rt6i_dev && (rt->rt6i_dev->flags&IFF_LOOPBACK)) |
2380 | rtm->rtm_type = RTN_LOCAL; | |
2381 | else | |
2382 | rtm->rtm_type = RTN_UNICAST; | |
2383 | rtm->rtm_flags = 0; | |
2384 | rtm->rtm_scope = RT_SCOPE_UNIVERSE; | |
2385 | rtm->rtm_protocol = rt->rt6i_protocol; | |
2386 | if (rt->rt6i_flags&RTF_DYNAMIC) | |
2387 | rtm->rtm_protocol = RTPROT_REDIRECT; | |
2388 | else if (rt->rt6i_flags & RTF_ADDRCONF) | |
2389 | rtm->rtm_protocol = RTPROT_KERNEL; | |
2390 | else if (rt->rt6i_flags&RTF_DEFAULT) | |
2391 | rtm->rtm_protocol = RTPROT_RA; | |
2392 | ||
2393 | if (rt->rt6i_flags&RTF_CACHE) | |
2394 | rtm->rtm_flags |= RTM_F_CLONED; | |
2395 | ||
2396 | if (dst) { | |
2d7202bf | 2397 | NLA_PUT(skb, RTA_DST, 16, dst); |
1ab1457c | 2398 | rtm->rtm_dst_len = 128; |
1da177e4 | 2399 | } else if (rtm->rtm_dst_len) |
2d7202bf | 2400 | NLA_PUT(skb, RTA_DST, 16, &rt->rt6i_dst.addr); |
1da177e4 LT |
2401 | #ifdef CONFIG_IPV6_SUBTREES |
2402 | if (src) { | |
2d7202bf | 2403 | NLA_PUT(skb, RTA_SRC, 16, src); |
1ab1457c | 2404 | rtm->rtm_src_len = 128; |
1da177e4 | 2405 | } else if (rtm->rtm_src_len) |
2d7202bf | 2406 | NLA_PUT(skb, RTA_SRC, 16, &rt->rt6i_src.addr); |
1da177e4 | 2407 | #endif |
7bc570c8 YH |
2408 | if (iif) { |
2409 | #ifdef CONFIG_IPV6_MROUTE | |
2410 | if (ipv6_addr_is_multicast(&rt->rt6i_dst.addr)) { | |
8229efda | 2411 | int err = ip6mr_get_route(net, skb, rtm, nowait); |
7bc570c8 YH |
2412 | if (err <= 0) { |
2413 | if (!nowait) { | |
2414 | if (err == 0) | |
2415 | return 0; | |
2416 | goto nla_put_failure; | |
2417 | } else { | |
2418 | if (err == -EMSGSIZE) | |
2419 | goto nla_put_failure; | |
2420 | } | |
2421 | } | |
2422 | } else | |
2423 | #endif | |
2424 | NLA_PUT_U32(skb, RTA_IIF, iif); | |
2425 | } else if (dst) { | |
1da177e4 | 2426 | struct in6_addr saddr_buf; |
c3968a85 | 2427 | if (ip6_route_get_saddr(net, rt, dst, 0, &saddr_buf) == 0) |
2d7202bf | 2428 | NLA_PUT(skb, RTA_PREFSRC, 16, &saddr_buf); |
1da177e4 | 2429 | } |
2d7202bf | 2430 | |
c3968a85 DW |
2431 | if (rt->rt6i_prefsrc.plen) { |
2432 | struct in6_addr saddr_buf; | |
2433 | ipv6_addr_copy(&saddr_buf, &rt->rt6i_prefsrc.addr); | |
2434 | NLA_PUT(skb, RTA_PREFSRC, 16, &saddr_buf); | |
2435 | } | |
2436 | ||
defb3519 | 2437 | if (rtnetlink_put_metrics(skb, dst_metrics_ptr(&rt->dst)) < 0) |
2d7202bf TG |
2438 | goto nla_put_failure; |
2439 | ||
f2c31e32 ED |
2440 | rcu_read_lock(); |
2441 | n = dst_get_neighbour(&rt->dst); | |
2442 | if (n) | |
2443 | NLA_PUT(skb, RTA_GATEWAY, 16, &n->primary_key); | |
2444 | rcu_read_unlock(); | |
2d7202bf | 2445 | |
d8d1f30b | 2446 | if (rt->dst.dev) |
2d7202bf TG |
2447 | NLA_PUT_U32(skb, RTA_OIF, rt->rt6i_dev->ifindex); |
2448 | ||
2449 | NLA_PUT_U32(skb, RTA_PRIORITY, rt->rt6i_metric); | |
e3703b3d | 2450 | |
36e3deae YH |
2451 | if (!(rt->rt6i_flags & RTF_EXPIRES)) |
2452 | expires = 0; | |
2453 | else if (rt->rt6i_expires - jiffies < INT_MAX) | |
2454 | expires = rt->rt6i_expires - jiffies; | |
2455 | else | |
2456 | expires = INT_MAX; | |
69cdf8f9 | 2457 | |
d8d1f30b CG |
2458 | if (rtnl_put_cacheinfo(skb, &rt->dst, 0, 0, 0, |
2459 | expires, rt->dst.error) < 0) | |
e3703b3d | 2460 | goto nla_put_failure; |
2d7202bf TG |
2461 | |
2462 | return nlmsg_end(skb, nlh); | |
2463 | ||
2464 | nla_put_failure: | |
26932566 PM |
2465 | nlmsg_cancel(skb, nlh); |
2466 | return -EMSGSIZE; | |
1da177e4 LT |
2467 | } |
2468 | ||
1b43af54 | 2469 | int rt6_dump_route(struct rt6_info *rt, void *p_arg) |
1da177e4 LT |
2470 | { |
2471 | struct rt6_rtnl_dump_arg *arg = (struct rt6_rtnl_dump_arg *) p_arg; | |
2472 | int prefix; | |
2473 | ||
2d7202bf TG |
2474 | if (nlmsg_len(arg->cb->nlh) >= sizeof(struct rtmsg)) { |
2475 | struct rtmsg *rtm = nlmsg_data(arg->cb->nlh); | |
1da177e4 LT |
2476 | prefix = (rtm->rtm_flags & RTM_F_PREFIX) != 0; |
2477 | } else | |
2478 | prefix = 0; | |
2479 | ||
191cd582 BH |
2480 | return rt6_fill_node(arg->net, |
2481 | arg->skb, rt, NULL, NULL, 0, RTM_NEWROUTE, | |
1da177e4 | 2482 | NETLINK_CB(arg->cb->skb).pid, arg->cb->nlh->nlmsg_seq, |
7bc570c8 | 2483 | prefix, 0, NLM_F_MULTI); |
1da177e4 LT |
2484 | } |
2485 | ||
c127ea2c | 2486 | static int inet6_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void *arg) |
1da177e4 | 2487 | { |
3b1e0a65 | 2488 | struct net *net = sock_net(in_skb->sk); |
ab364a6f TG |
2489 | struct nlattr *tb[RTA_MAX+1]; |
2490 | struct rt6_info *rt; | |
1da177e4 | 2491 | struct sk_buff *skb; |
ab364a6f | 2492 | struct rtmsg *rtm; |
4c9483b2 | 2493 | struct flowi6 fl6; |
ab364a6f | 2494 | int err, iif = 0; |
1da177e4 | 2495 | |
ab364a6f TG |
2496 | err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_ipv6_policy); |
2497 | if (err < 0) | |
2498 | goto errout; | |
1da177e4 | 2499 | |
ab364a6f | 2500 | err = -EINVAL; |
4c9483b2 | 2501 | memset(&fl6, 0, sizeof(fl6)); |
1da177e4 | 2502 | |
ab364a6f TG |
2503 | if (tb[RTA_SRC]) { |
2504 | if (nla_len(tb[RTA_SRC]) < sizeof(struct in6_addr)) | |
2505 | goto errout; | |
2506 | ||
4c9483b2 | 2507 | ipv6_addr_copy(&fl6.saddr, nla_data(tb[RTA_SRC])); |
ab364a6f TG |
2508 | } |
2509 | ||
2510 | if (tb[RTA_DST]) { | |
2511 | if (nla_len(tb[RTA_DST]) < sizeof(struct in6_addr)) | |
2512 | goto errout; | |
2513 | ||
4c9483b2 | 2514 | ipv6_addr_copy(&fl6.daddr, nla_data(tb[RTA_DST])); |
ab364a6f TG |
2515 | } |
2516 | ||
2517 | if (tb[RTA_IIF]) | |
2518 | iif = nla_get_u32(tb[RTA_IIF]); | |
2519 | ||
2520 | if (tb[RTA_OIF]) | |
4c9483b2 | 2521 | fl6.flowi6_oif = nla_get_u32(tb[RTA_OIF]); |
1da177e4 LT |
2522 | |
2523 | if (iif) { | |
2524 | struct net_device *dev; | |
5578689a | 2525 | dev = __dev_get_by_index(net, iif); |
1da177e4 LT |
2526 | if (!dev) { |
2527 | err = -ENODEV; | |
ab364a6f | 2528 | goto errout; |
1da177e4 LT |
2529 | } |
2530 | } | |
2531 | ||
ab364a6f TG |
2532 | skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); |
2533 | if (skb == NULL) { | |
2534 | err = -ENOBUFS; | |
2535 | goto errout; | |
2536 | } | |
1da177e4 | 2537 | |
ab364a6f TG |
2538 | /* Reserve room for dummy headers, this skb can pass |
2539 | through good chunk of routing engine. | |
2540 | */ | |
459a98ed | 2541 | skb_reset_mac_header(skb); |
ab364a6f | 2542 | skb_reserve(skb, MAX_HEADER + sizeof(struct ipv6hdr)); |
1da177e4 | 2543 | |
4c9483b2 | 2544 | rt = (struct rt6_info*) ip6_route_output(net, NULL, &fl6); |
d8d1f30b | 2545 | skb_dst_set(skb, &rt->dst); |
1da177e4 | 2546 | |
4c9483b2 | 2547 | err = rt6_fill_node(net, skb, rt, &fl6.daddr, &fl6.saddr, iif, |
1da177e4 | 2548 | RTM_NEWROUTE, NETLINK_CB(in_skb).pid, |
7bc570c8 | 2549 | nlh->nlmsg_seq, 0, 0, 0); |
1da177e4 | 2550 | if (err < 0) { |
ab364a6f TG |
2551 | kfree_skb(skb); |
2552 | goto errout; | |
1da177e4 LT |
2553 | } |
2554 | ||
5578689a | 2555 | err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).pid); |
ab364a6f | 2556 | errout: |
1da177e4 | 2557 | return err; |
1da177e4 LT |
2558 | } |
2559 | ||
86872cb5 | 2560 | void inet6_rt_notify(int event, struct rt6_info *rt, struct nl_info *info) |
1da177e4 LT |
2561 | { |
2562 | struct sk_buff *skb; | |
5578689a | 2563 | struct net *net = info->nl_net; |
528c4ceb DL |
2564 | u32 seq; |
2565 | int err; | |
2566 | ||
2567 | err = -ENOBUFS; | |
2568 | seq = info->nlh != NULL ? info->nlh->nlmsg_seq : 0; | |
86872cb5 | 2569 | |
339bf98f | 2570 | skb = nlmsg_new(rt6_nlmsg_size(), gfp_any()); |
21713ebc TG |
2571 | if (skb == NULL) |
2572 | goto errout; | |
2573 | ||
191cd582 | 2574 | err = rt6_fill_node(net, skb, rt, NULL, NULL, 0, |
7bc570c8 | 2575 | event, info->pid, seq, 0, 0, 0); |
26932566 PM |
2576 | if (err < 0) { |
2577 | /* -EMSGSIZE implies BUG in rt6_nlmsg_size() */ | |
2578 | WARN_ON(err == -EMSGSIZE); | |
2579 | kfree_skb(skb); | |
2580 | goto errout; | |
2581 | } | |
1ce85fe4 PNA |
2582 | rtnl_notify(skb, net, info->pid, RTNLGRP_IPV6_ROUTE, |
2583 | info->nlh, gfp_any()); | |
2584 | return; | |
21713ebc TG |
2585 | errout: |
2586 | if (err < 0) | |
5578689a | 2587 | rtnl_set_sk_err(net, RTNLGRP_IPV6_ROUTE, err); |
1da177e4 LT |
2588 | } |
2589 | ||
8ed67789 DL |
2590 | static int ip6_route_dev_notify(struct notifier_block *this, |
2591 | unsigned long event, void *data) | |
2592 | { | |
2593 | struct net_device *dev = (struct net_device *)data; | |
c346dca1 | 2594 | struct net *net = dev_net(dev); |
8ed67789 DL |
2595 | |
2596 | if (event == NETDEV_REGISTER && (dev->flags & IFF_LOOPBACK)) { | |
d8d1f30b | 2597 | net->ipv6.ip6_null_entry->dst.dev = dev; |
8ed67789 DL |
2598 | net->ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(dev); |
2599 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES | |
d8d1f30b | 2600 | net->ipv6.ip6_prohibit_entry->dst.dev = dev; |
8ed67789 | 2601 | net->ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(dev); |
d8d1f30b | 2602 | net->ipv6.ip6_blk_hole_entry->dst.dev = dev; |
8ed67789 DL |
2603 | net->ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(dev); |
2604 | #endif | |
2605 | } | |
2606 | ||
2607 | return NOTIFY_OK; | |
2608 | } | |
2609 | ||
1da177e4 LT |
2610 | /* |
2611 | * /proc | |
2612 | */ | |
2613 | ||
2614 | #ifdef CONFIG_PROC_FS | |
2615 | ||
1da177e4 LT |
2616 | struct rt6_proc_arg |
2617 | { | |
2618 | char *buffer; | |
2619 | int offset; | |
2620 | int length; | |
2621 | int skip; | |
2622 | int len; | |
2623 | }; | |
2624 | ||
2625 | static int rt6_info_route(struct rt6_info *rt, void *p_arg) | |
2626 | { | |
33120b30 | 2627 | struct seq_file *m = p_arg; |
69cce1d1 | 2628 | struct neighbour *n; |
1da177e4 | 2629 | |
4b7a4274 | 2630 | seq_printf(m, "%pi6 %02x ", &rt->rt6i_dst.addr, rt->rt6i_dst.plen); |
1da177e4 LT |
2631 | |
2632 | #ifdef CONFIG_IPV6_SUBTREES | |
4b7a4274 | 2633 | seq_printf(m, "%pi6 %02x ", &rt->rt6i_src.addr, rt->rt6i_src.plen); |
1da177e4 | 2634 | #else |
33120b30 | 2635 | seq_puts(m, "00000000000000000000000000000000 00 "); |
1da177e4 | 2636 | #endif |
f2c31e32 | 2637 | rcu_read_lock(); |
69cce1d1 DM |
2638 | n = dst_get_neighbour(&rt->dst); |
2639 | if (n) { | |
2640 | seq_printf(m, "%pi6", n->primary_key); | |
1da177e4 | 2641 | } else { |
33120b30 | 2642 | seq_puts(m, "00000000000000000000000000000000"); |
1da177e4 | 2643 | } |
f2c31e32 | 2644 | rcu_read_unlock(); |
33120b30 | 2645 | seq_printf(m, " %08x %08x %08x %08x %8s\n", |
d8d1f30b CG |
2646 | rt->rt6i_metric, atomic_read(&rt->dst.__refcnt), |
2647 | rt->dst.__use, rt->rt6i_flags, | |
33120b30 | 2648 | rt->rt6i_dev ? rt->rt6i_dev->name : ""); |
1da177e4 LT |
2649 | return 0; |
2650 | } | |
2651 | ||
33120b30 | 2652 | static int ipv6_route_show(struct seq_file *m, void *v) |
1da177e4 | 2653 | { |
f3db4851 DL |
2654 | struct net *net = (struct net *)m->private; |
2655 | fib6_clean_all(net, rt6_info_route, 0, m); | |
33120b30 AD |
2656 | return 0; |
2657 | } | |
1da177e4 | 2658 | |
33120b30 AD |
2659 | static int ipv6_route_open(struct inode *inode, struct file *file) |
2660 | { | |
de05c557 | 2661 | return single_open_net(inode, file, ipv6_route_show); |
f3db4851 DL |
2662 | } |
2663 | ||
33120b30 AD |
2664 | static const struct file_operations ipv6_route_proc_fops = { |
2665 | .owner = THIS_MODULE, | |
2666 | .open = ipv6_route_open, | |
2667 | .read = seq_read, | |
2668 | .llseek = seq_lseek, | |
b6fcbdb4 | 2669 | .release = single_release_net, |
33120b30 AD |
2670 | }; |
2671 | ||
1da177e4 LT |
2672 | static int rt6_stats_seq_show(struct seq_file *seq, void *v) |
2673 | { | |
69ddb805 | 2674 | struct net *net = (struct net *)seq->private; |
1da177e4 | 2675 | seq_printf(seq, "%04x %04x %04x %04x %04x %04x %04x\n", |
69ddb805 DL |
2676 | net->ipv6.rt6_stats->fib_nodes, |
2677 | net->ipv6.rt6_stats->fib_route_nodes, | |
2678 | net->ipv6.rt6_stats->fib_rt_alloc, | |
2679 | net->ipv6.rt6_stats->fib_rt_entries, | |
2680 | net->ipv6.rt6_stats->fib_rt_cache, | |
fc66f95c | 2681 | dst_entries_get_slow(&net->ipv6.ip6_dst_ops), |
69ddb805 | 2682 | net->ipv6.rt6_stats->fib_discarded_routes); |
1da177e4 LT |
2683 | |
2684 | return 0; | |
2685 | } | |
2686 | ||
2687 | static int rt6_stats_seq_open(struct inode *inode, struct file *file) | |
2688 | { | |
de05c557 | 2689 | return single_open_net(inode, file, rt6_stats_seq_show); |
69ddb805 DL |
2690 | } |
2691 | ||
9a32144e | 2692 | static const struct file_operations rt6_stats_seq_fops = { |
1da177e4 LT |
2693 | .owner = THIS_MODULE, |
2694 | .open = rt6_stats_seq_open, | |
2695 | .read = seq_read, | |
2696 | .llseek = seq_lseek, | |
b6fcbdb4 | 2697 | .release = single_release_net, |
1da177e4 LT |
2698 | }; |
2699 | #endif /* CONFIG_PROC_FS */ | |
2700 | ||
2701 | #ifdef CONFIG_SYSCTL | |
2702 | ||
1da177e4 | 2703 | static |
8d65af78 | 2704 | int ipv6_sysctl_rtcache_flush(ctl_table *ctl, int write, |
1da177e4 LT |
2705 | void __user *buffer, size_t *lenp, loff_t *ppos) |
2706 | { | |
c486da34 LAG |
2707 | struct net *net; |
2708 | int delay; | |
2709 | if (!write) | |
1da177e4 | 2710 | return -EINVAL; |
c486da34 LAG |
2711 | |
2712 | net = (struct net *)ctl->extra1; | |
2713 | delay = net->ipv6.sysctl.flush_delay; | |
2714 | proc_dointvec(ctl, write, buffer, lenp, ppos); | |
2715 | fib6_run_gc(delay <= 0 ? ~0UL : (unsigned long)delay, net); | |
2716 | return 0; | |
1da177e4 LT |
2717 | } |
2718 | ||
760f2d01 | 2719 | ctl_table ipv6_route_table_template[] = { |
1ab1457c | 2720 | { |
1da177e4 | 2721 | .procname = "flush", |
4990509f | 2722 | .data = &init_net.ipv6.sysctl.flush_delay, |
1da177e4 | 2723 | .maxlen = sizeof(int), |
89c8b3a1 | 2724 | .mode = 0200, |
6d9f239a | 2725 | .proc_handler = ipv6_sysctl_rtcache_flush |
1da177e4 LT |
2726 | }, |
2727 | { | |
1da177e4 | 2728 | .procname = "gc_thresh", |
9a7ec3a9 | 2729 | .data = &ip6_dst_ops_template.gc_thresh, |
1da177e4 LT |
2730 | .maxlen = sizeof(int), |
2731 | .mode = 0644, | |
6d9f239a | 2732 | .proc_handler = proc_dointvec, |
1da177e4 LT |
2733 | }, |
2734 | { | |
1da177e4 | 2735 | .procname = "max_size", |
4990509f | 2736 | .data = &init_net.ipv6.sysctl.ip6_rt_max_size, |
1da177e4 LT |
2737 | .maxlen = sizeof(int), |
2738 | .mode = 0644, | |
6d9f239a | 2739 | .proc_handler = proc_dointvec, |
1da177e4 LT |
2740 | }, |
2741 | { | |
1da177e4 | 2742 | .procname = "gc_min_interval", |
4990509f | 2743 | .data = &init_net.ipv6.sysctl.ip6_rt_gc_min_interval, |
1da177e4 LT |
2744 | .maxlen = sizeof(int), |
2745 | .mode = 0644, | |
6d9f239a | 2746 | .proc_handler = proc_dointvec_jiffies, |
1da177e4 LT |
2747 | }, |
2748 | { | |
1da177e4 | 2749 | .procname = "gc_timeout", |
4990509f | 2750 | .data = &init_net.ipv6.sysctl.ip6_rt_gc_timeout, |
1da177e4 LT |
2751 | .maxlen = sizeof(int), |
2752 | .mode = 0644, | |
6d9f239a | 2753 | .proc_handler = proc_dointvec_jiffies, |
1da177e4 LT |
2754 | }, |
2755 | { | |
1da177e4 | 2756 | .procname = "gc_interval", |
4990509f | 2757 | .data = &init_net.ipv6.sysctl.ip6_rt_gc_interval, |
1da177e4 LT |
2758 | .maxlen = sizeof(int), |
2759 | .mode = 0644, | |
6d9f239a | 2760 | .proc_handler = proc_dointvec_jiffies, |
1da177e4 LT |
2761 | }, |
2762 | { | |
1da177e4 | 2763 | .procname = "gc_elasticity", |
4990509f | 2764 | .data = &init_net.ipv6.sysctl.ip6_rt_gc_elasticity, |
1da177e4 LT |
2765 | .maxlen = sizeof(int), |
2766 | .mode = 0644, | |
f3d3f616 | 2767 | .proc_handler = proc_dointvec, |
1da177e4 LT |
2768 | }, |
2769 | { | |
1da177e4 | 2770 | .procname = "mtu_expires", |
4990509f | 2771 | .data = &init_net.ipv6.sysctl.ip6_rt_mtu_expires, |
1da177e4 LT |
2772 | .maxlen = sizeof(int), |
2773 | .mode = 0644, | |
6d9f239a | 2774 | .proc_handler = proc_dointvec_jiffies, |
1da177e4 LT |
2775 | }, |
2776 | { | |
1da177e4 | 2777 | .procname = "min_adv_mss", |
4990509f | 2778 | .data = &init_net.ipv6.sysctl.ip6_rt_min_advmss, |
1da177e4 LT |
2779 | .maxlen = sizeof(int), |
2780 | .mode = 0644, | |
f3d3f616 | 2781 | .proc_handler = proc_dointvec, |
1da177e4 LT |
2782 | }, |
2783 | { | |
1da177e4 | 2784 | .procname = "gc_min_interval_ms", |
4990509f | 2785 | .data = &init_net.ipv6.sysctl.ip6_rt_gc_min_interval, |
1da177e4 LT |
2786 | .maxlen = sizeof(int), |
2787 | .mode = 0644, | |
6d9f239a | 2788 | .proc_handler = proc_dointvec_ms_jiffies, |
1da177e4 | 2789 | }, |
f8572d8f | 2790 | { } |
1da177e4 LT |
2791 | }; |
2792 | ||
2c8c1e72 | 2793 | struct ctl_table * __net_init ipv6_route_sysctl_init(struct net *net) |
760f2d01 DL |
2794 | { |
2795 | struct ctl_table *table; | |
2796 | ||
2797 | table = kmemdup(ipv6_route_table_template, | |
2798 | sizeof(ipv6_route_table_template), | |
2799 | GFP_KERNEL); | |
5ee09105 YH |
2800 | |
2801 | if (table) { | |
2802 | table[0].data = &net->ipv6.sysctl.flush_delay; | |
c486da34 | 2803 | table[0].extra1 = net; |
86393e52 | 2804 | table[1].data = &net->ipv6.ip6_dst_ops.gc_thresh; |
5ee09105 YH |
2805 | table[2].data = &net->ipv6.sysctl.ip6_rt_max_size; |
2806 | table[3].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval; | |
2807 | table[4].data = &net->ipv6.sysctl.ip6_rt_gc_timeout; | |
2808 | table[5].data = &net->ipv6.sysctl.ip6_rt_gc_interval; | |
2809 | table[6].data = &net->ipv6.sysctl.ip6_rt_gc_elasticity; | |
2810 | table[7].data = &net->ipv6.sysctl.ip6_rt_mtu_expires; | |
2811 | table[8].data = &net->ipv6.sysctl.ip6_rt_min_advmss; | |
9c69fabe | 2812 | table[9].data = &net->ipv6.sysctl.ip6_rt_gc_min_interval; |
5ee09105 YH |
2813 | } |
2814 | ||
760f2d01 DL |
2815 | return table; |
2816 | } | |
1da177e4 LT |
2817 | #endif |
2818 | ||
2c8c1e72 | 2819 | static int __net_init ip6_route_net_init(struct net *net) |
cdb18761 | 2820 | { |
633d424b | 2821 | int ret = -ENOMEM; |
8ed67789 | 2822 | |
86393e52 AD |
2823 | memcpy(&net->ipv6.ip6_dst_ops, &ip6_dst_ops_template, |
2824 | sizeof(net->ipv6.ip6_dst_ops)); | |
f2fc6a54 | 2825 | |
fc66f95c ED |
2826 | if (dst_entries_init(&net->ipv6.ip6_dst_ops) < 0) |
2827 | goto out_ip6_dst_ops; | |
2828 | ||
8ed67789 DL |
2829 | net->ipv6.ip6_null_entry = kmemdup(&ip6_null_entry_template, |
2830 | sizeof(*net->ipv6.ip6_null_entry), | |
2831 | GFP_KERNEL); | |
2832 | if (!net->ipv6.ip6_null_entry) | |
fc66f95c | 2833 | goto out_ip6_dst_entries; |
d8d1f30b | 2834 | net->ipv6.ip6_null_entry->dst.path = |
8ed67789 | 2835 | (struct dst_entry *)net->ipv6.ip6_null_entry; |
d8d1f30b | 2836 | net->ipv6.ip6_null_entry->dst.ops = &net->ipv6.ip6_dst_ops; |
62fa8a84 DM |
2837 | dst_init_metrics(&net->ipv6.ip6_null_entry->dst, |
2838 | ip6_template_metrics, true); | |
8ed67789 DL |
2839 | |
2840 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES | |
2841 | net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template, | |
2842 | sizeof(*net->ipv6.ip6_prohibit_entry), | |
2843 | GFP_KERNEL); | |
68fffc67 PZ |
2844 | if (!net->ipv6.ip6_prohibit_entry) |
2845 | goto out_ip6_null_entry; | |
d8d1f30b | 2846 | net->ipv6.ip6_prohibit_entry->dst.path = |
8ed67789 | 2847 | (struct dst_entry *)net->ipv6.ip6_prohibit_entry; |
d8d1f30b | 2848 | net->ipv6.ip6_prohibit_entry->dst.ops = &net->ipv6.ip6_dst_ops; |
62fa8a84 DM |
2849 | dst_init_metrics(&net->ipv6.ip6_prohibit_entry->dst, |
2850 | ip6_template_metrics, true); | |
8ed67789 DL |
2851 | |
2852 | net->ipv6.ip6_blk_hole_entry = kmemdup(&ip6_blk_hole_entry_template, | |
2853 | sizeof(*net->ipv6.ip6_blk_hole_entry), | |
2854 | GFP_KERNEL); | |
68fffc67 PZ |
2855 | if (!net->ipv6.ip6_blk_hole_entry) |
2856 | goto out_ip6_prohibit_entry; | |
d8d1f30b | 2857 | net->ipv6.ip6_blk_hole_entry->dst.path = |
8ed67789 | 2858 | (struct dst_entry *)net->ipv6.ip6_blk_hole_entry; |
d8d1f30b | 2859 | net->ipv6.ip6_blk_hole_entry->dst.ops = &net->ipv6.ip6_dst_ops; |
62fa8a84 DM |
2860 | dst_init_metrics(&net->ipv6.ip6_blk_hole_entry->dst, |
2861 | ip6_template_metrics, true); | |
8ed67789 DL |
2862 | #endif |
2863 | ||
b339a47c PZ |
2864 | net->ipv6.sysctl.flush_delay = 0; |
2865 | net->ipv6.sysctl.ip6_rt_max_size = 4096; | |
2866 | net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2; | |
2867 | net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ; | |
2868 | net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ; | |
2869 | net->ipv6.sysctl.ip6_rt_gc_elasticity = 9; | |
2870 | net->ipv6.sysctl.ip6_rt_mtu_expires = 10*60*HZ; | |
2871 | net->ipv6.sysctl.ip6_rt_min_advmss = IPV6_MIN_MTU - 20 - 40; | |
2872 | ||
cdb18761 DL |
2873 | #ifdef CONFIG_PROC_FS |
2874 | proc_net_fops_create(net, "ipv6_route", 0, &ipv6_route_proc_fops); | |
2875 | proc_net_fops_create(net, "rt6_stats", S_IRUGO, &rt6_stats_seq_fops); | |
2876 | #endif | |
6891a346 BT |
2877 | net->ipv6.ip6_rt_gc_expire = 30*HZ; |
2878 | ||
8ed67789 DL |
2879 | ret = 0; |
2880 | out: | |
2881 | return ret; | |
f2fc6a54 | 2882 | |
68fffc67 PZ |
2883 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES |
2884 | out_ip6_prohibit_entry: | |
2885 | kfree(net->ipv6.ip6_prohibit_entry); | |
2886 | out_ip6_null_entry: | |
2887 | kfree(net->ipv6.ip6_null_entry); | |
2888 | #endif | |
fc66f95c ED |
2889 | out_ip6_dst_entries: |
2890 | dst_entries_destroy(&net->ipv6.ip6_dst_ops); | |
f2fc6a54 | 2891 | out_ip6_dst_ops: |
f2fc6a54 | 2892 | goto out; |
cdb18761 DL |
2893 | } |
2894 | ||
2c8c1e72 | 2895 | static void __net_exit ip6_route_net_exit(struct net *net) |
cdb18761 DL |
2896 | { |
2897 | #ifdef CONFIG_PROC_FS | |
2898 | proc_net_remove(net, "ipv6_route"); | |
2899 | proc_net_remove(net, "rt6_stats"); | |
2900 | #endif | |
8ed67789 DL |
2901 | kfree(net->ipv6.ip6_null_entry); |
2902 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES | |
2903 | kfree(net->ipv6.ip6_prohibit_entry); | |
2904 | kfree(net->ipv6.ip6_blk_hole_entry); | |
2905 | #endif | |
41bb78b4 | 2906 | dst_entries_destroy(&net->ipv6.ip6_dst_ops); |
cdb18761 DL |
2907 | } |
2908 | ||
2909 | static struct pernet_operations ip6_route_net_ops = { | |
2910 | .init = ip6_route_net_init, | |
2911 | .exit = ip6_route_net_exit, | |
2912 | }; | |
2913 | ||
8ed67789 DL |
2914 | static struct notifier_block ip6_route_dev_notifier = { |
2915 | .notifier_call = ip6_route_dev_notify, | |
2916 | .priority = 0, | |
2917 | }; | |
2918 | ||
433d49c3 | 2919 | int __init ip6_route_init(void) |
1da177e4 | 2920 | { |
433d49c3 DL |
2921 | int ret; |
2922 | ||
9a7ec3a9 DL |
2923 | ret = -ENOMEM; |
2924 | ip6_dst_ops_template.kmem_cachep = | |
e5d679f3 | 2925 | kmem_cache_create("ip6_dst_cache", sizeof(struct rt6_info), 0, |
f845ab6b | 2926 | SLAB_HWCACHE_ALIGN, NULL); |
9a7ec3a9 | 2927 | if (!ip6_dst_ops_template.kmem_cachep) |
c19a28e1 | 2928 | goto out; |
14e50e57 | 2929 | |
fc66f95c | 2930 | ret = dst_entries_init(&ip6_dst_blackhole_ops); |
8ed67789 | 2931 | if (ret) |
bdb3289f | 2932 | goto out_kmem_cache; |
bdb3289f | 2933 | |
fc66f95c ED |
2934 | ret = register_pernet_subsys(&ip6_route_net_ops); |
2935 | if (ret) | |
2936 | goto out_dst_entries; | |
2937 | ||
5dc121e9 AE |
2938 | ip6_dst_blackhole_ops.kmem_cachep = ip6_dst_ops_template.kmem_cachep; |
2939 | ||
8ed67789 DL |
2940 | /* Registering of the loopback is done before this portion of code, |
2941 | * the loopback reference in rt6_info will not be taken, do it | |
2942 | * manually for init_net */ | |
d8d1f30b | 2943 | init_net.ipv6.ip6_null_entry->dst.dev = init_net.loopback_dev; |
8ed67789 DL |
2944 | init_net.ipv6.ip6_null_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev); |
2945 | #ifdef CONFIG_IPV6_MULTIPLE_TABLES | |
d8d1f30b | 2946 | init_net.ipv6.ip6_prohibit_entry->dst.dev = init_net.loopback_dev; |
8ed67789 | 2947 | init_net.ipv6.ip6_prohibit_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev); |
d8d1f30b | 2948 | init_net.ipv6.ip6_blk_hole_entry->dst.dev = init_net.loopback_dev; |
8ed67789 DL |
2949 | init_net.ipv6.ip6_blk_hole_entry->rt6i_idev = in6_dev_get(init_net.loopback_dev); |
2950 | #endif | |
433d49c3 DL |
2951 | ret = fib6_init(); |
2952 | if (ret) | |
8ed67789 | 2953 | goto out_register_subsys; |
433d49c3 | 2954 | |
433d49c3 DL |
2955 | ret = xfrm6_init(); |
2956 | if (ret) | |
cdb18761 | 2957 | goto out_fib6_init; |
c35b7e72 | 2958 | |
433d49c3 DL |
2959 | ret = fib6_rules_init(); |
2960 | if (ret) | |
2961 | goto xfrm6_init; | |
7e5449c2 | 2962 | |
433d49c3 | 2963 | ret = -ENOBUFS; |
c7ac8679 GR |
2964 | if (__rtnl_register(PF_INET6, RTM_NEWROUTE, inet6_rtm_newroute, NULL, NULL) || |
2965 | __rtnl_register(PF_INET6, RTM_DELROUTE, inet6_rtm_delroute, NULL, NULL) || | |
2966 | __rtnl_register(PF_INET6, RTM_GETROUTE, inet6_rtm_getroute, NULL, NULL)) | |
433d49c3 | 2967 | goto fib6_rules_init; |
c127ea2c | 2968 | |
8ed67789 | 2969 | ret = register_netdevice_notifier(&ip6_route_dev_notifier); |
cdb18761 DL |
2970 | if (ret) |
2971 | goto fib6_rules_init; | |
8ed67789 | 2972 | |
433d49c3 DL |
2973 | out: |
2974 | return ret; | |
2975 | ||
2976 | fib6_rules_init: | |
433d49c3 DL |
2977 | fib6_rules_cleanup(); |
2978 | xfrm6_init: | |
433d49c3 | 2979 | xfrm6_fini(); |
433d49c3 | 2980 | out_fib6_init: |
433d49c3 | 2981 | fib6_gc_cleanup(); |
8ed67789 DL |
2982 | out_register_subsys: |
2983 | unregister_pernet_subsys(&ip6_route_net_ops); | |
fc66f95c ED |
2984 | out_dst_entries: |
2985 | dst_entries_destroy(&ip6_dst_blackhole_ops); | |
433d49c3 | 2986 | out_kmem_cache: |
f2fc6a54 | 2987 | kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep); |
433d49c3 | 2988 | goto out; |
1da177e4 LT |
2989 | } |
2990 | ||
2991 | void ip6_route_cleanup(void) | |
2992 | { | |
8ed67789 | 2993 | unregister_netdevice_notifier(&ip6_route_dev_notifier); |
101367c2 | 2994 | fib6_rules_cleanup(); |
1da177e4 | 2995 | xfrm6_fini(); |
1da177e4 | 2996 | fib6_gc_cleanup(); |
8ed67789 | 2997 | unregister_pernet_subsys(&ip6_route_net_ops); |
41bb78b4 | 2998 | dst_entries_destroy(&ip6_dst_blackhole_ops); |
f2fc6a54 | 2999 | kmem_cache_destroy(ip6_dst_ops_template.kmem_cachep); |
1da177e4 | 3000 | } |