Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * IPv4 Forwarding Information Base: semantics. | |
7 | * | |
1da177e4 LT |
8 | * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public License | |
12 | * as published by the Free Software Foundation; either version | |
13 | * 2 of the License, or (at your option) any later version. | |
14 | */ | |
15 | ||
1da177e4 | 16 | #include <asm/uaccess.h> |
1da177e4 LT |
17 | #include <linux/bitops.h> |
18 | #include <linux/types.h> | |
19 | #include <linux/kernel.h> | |
20 | #include <linux/jiffies.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/string.h> | |
23 | #include <linux/socket.h> | |
24 | #include <linux/sockios.h> | |
25 | #include <linux/errno.h> | |
26 | #include <linux/in.h> | |
27 | #include <linux/inet.h> | |
14c85021 | 28 | #include <linux/inetdevice.h> |
1da177e4 LT |
29 | #include <linux/netdevice.h> |
30 | #include <linux/if_arp.h> | |
31 | #include <linux/proc_fs.h> | |
32 | #include <linux/skbuff.h> | |
1da177e4 | 33 | #include <linux/init.h> |
5a0e3ad6 | 34 | #include <linux/slab.h> |
1da177e4 | 35 | |
14c85021 | 36 | #include <net/arp.h> |
1da177e4 LT |
37 | #include <net/ip.h> |
38 | #include <net/protocol.h> | |
39 | #include <net/route.h> | |
40 | #include <net/tcp.h> | |
41 | #include <net/sock.h> | |
42 | #include <net/ip_fib.h> | |
f21c7bc5 | 43 | #include <net/netlink.h> |
4e902c57 | 44 | #include <net/nexthop.h> |
571e7226 | 45 | #include <net/lwtunnel.h> |
1da177e4 LT |
46 | |
47 | #include "fib_lookup.h" | |
48 | ||
832b4c5e | 49 | static DEFINE_SPINLOCK(fib_info_lock); |
1da177e4 LT |
50 | static struct hlist_head *fib_info_hash; |
51 | static struct hlist_head *fib_info_laddrhash; | |
123b9731 | 52 | static unsigned int fib_info_hash_size; |
1da177e4 LT |
53 | static unsigned int fib_info_cnt; |
54 | ||
55 | #define DEVINDEX_HASHBITS 8 | |
56 | #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS) | |
57 | static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE]; | |
58 | ||
59 | #ifdef CONFIG_IP_ROUTE_MULTIPATH | |
60 | ||
61 | static DEFINE_SPINLOCK(fib_multipath_lock); | |
62 | ||
6a31d2a9 ED |
63 | #define for_nexthops(fi) { \ |
64 | int nhsel; const struct fib_nh *nh; \ | |
65 | for (nhsel = 0, nh = (fi)->fib_nh; \ | |
66 | nhsel < (fi)->fib_nhs; \ | |
67 | nh++, nhsel++) | |
68 | ||
69 | #define change_nexthops(fi) { \ | |
70 | int nhsel; struct fib_nh *nexthop_nh; \ | |
71 | for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ | |
72 | nhsel < (fi)->fib_nhs; \ | |
73 | nexthop_nh++, nhsel++) | |
1da177e4 LT |
74 | |
75 | #else /* CONFIG_IP_ROUTE_MULTIPATH */ | |
76 | ||
77 | /* Hope, that gcc will optimize it to get rid of dummy loop */ | |
78 | ||
6a31d2a9 ED |
79 | #define for_nexthops(fi) { \ |
80 | int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \ | |
81 | for (nhsel = 0; nhsel < 1; nhsel++) | |
1da177e4 | 82 | |
6a31d2a9 ED |
83 | #define change_nexthops(fi) { \ |
84 | int nhsel; \ | |
85 | struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \ | |
86 | for (nhsel = 0; nhsel < 1; nhsel++) | |
1da177e4 LT |
87 | |
88 | #endif /* CONFIG_IP_ROUTE_MULTIPATH */ | |
89 | ||
90 | #define endfor_nexthops(fi) } | |
91 | ||
92 | ||
3be0686b | 93 | const struct fib_prop fib_props[RTN_MAX + 1] = { |
6a31d2a9 | 94 | [RTN_UNSPEC] = { |
1da177e4 LT |
95 | .error = 0, |
96 | .scope = RT_SCOPE_NOWHERE, | |
6a31d2a9 ED |
97 | }, |
98 | [RTN_UNICAST] = { | |
1da177e4 LT |
99 | .error = 0, |
100 | .scope = RT_SCOPE_UNIVERSE, | |
6a31d2a9 ED |
101 | }, |
102 | [RTN_LOCAL] = { | |
1da177e4 LT |
103 | .error = 0, |
104 | .scope = RT_SCOPE_HOST, | |
6a31d2a9 ED |
105 | }, |
106 | [RTN_BROADCAST] = { | |
1da177e4 LT |
107 | .error = 0, |
108 | .scope = RT_SCOPE_LINK, | |
6a31d2a9 ED |
109 | }, |
110 | [RTN_ANYCAST] = { | |
1da177e4 LT |
111 | .error = 0, |
112 | .scope = RT_SCOPE_LINK, | |
6a31d2a9 ED |
113 | }, |
114 | [RTN_MULTICAST] = { | |
1da177e4 LT |
115 | .error = 0, |
116 | .scope = RT_SCOPE_UNIVERSE, | |
6a31d2a9 ED |
117 | }, |
118 | [RTN_BLACKHOLE] = { | |
1da177e4 LT |
119 | .error = -EINVAL, |
120 | .scope = RT_SCOPE_UNIVERSE, | |
6a31d2a9 ED |
121 | }, |
122 | [RTN_UNREACHABLE] = { | |
1da177e4 LT |
123 | .error = -EHOSTUNREACH, |
124 | .scope = RT_SCOPE_UNIVERSE, | |
6a31d2a9 ED |
125 | }, |
126 | [RTN_PROHIBIT] = { | |
1da177e4 LT |
127 | .error = -EACCES, |
128 | .scope = RT_SCOPE_UNIVERSE, | |
6a31d2a9 ED |
129 | }, |
130 | [RTN_THROW] = { | |
1da177e4 LT |
131 | .error = -EAGAIN, |
132 | .scope = RT_SCOPE_UNIVERSE, | |
6a31d2a9 ED |
133 | }, |
134 | [RTN_NAT] = { | |
1da177e4 LT |
135 | .error = -EINVAL, |
136 | .scope = RT_SCOPE_NOWHERE, | |
6a31d2a9 ED |
137 | }, |
138 | [RTN_XRESOLVE] = { | |
1da177e4 LT |
139 | .error = -EINVAL, |
140 | .scope = RT_SCOPE_NOWHERE, | |
6a31d2a9 | 141 | }, |
1da177e4 LT |
142 | }; |
143 | ||
c5038a83 DM |
144 | static void rt_fibinfo_free(struct rtable __rcu **rtp) |
145 | { | |
146 | struct rtable *rt = rcu_dereference_protected(*rtp, 1); | |
147 | ||
148 | if (!rt) | |
149 | return; | |
150 | ||
151 | /* Not even needed : RCU_INIT_POINTER(*rtp, NULL); | |
152 | * because we waited an RCU grace period before calling | |
153 | * free_fib_info_rcu() | |
154 | */ | |
155 | ||
156 | dst_free(&rt->dst); | |
157 | } | |
158 | ||
4895c771 DM |
159 | static void free_nh_exceptions(struct fib_nh *nh) |
160 | { | |
caa41527 | 161 | struct fnhe_hash_bucket *hash; |
4895c771 DM |
162 | int i; |
163 | ||
caa41527 ED |
164 | hash = rcu_dereference_protected(nh->nh_exceptions, 1); |
165 | if (!hash) | |
166 | return; | |
4895c771 DM |
167 | for (i = 0; i < FNHE_HASH_SIZE; i++) { |
168 | struct fib_nh_exception *fnhe; | |
169 | ||
5abf7f7e | 170 | fnhe = rcu_dereference_protected(hash[i].chain, 1); |
4895c771 DM |
171 | while (fnhe) { |
172 | struct fib_nh_exception *next; | |
173 | ||
5abf7f7e | 174 | next = rcu_dereference_protected(fnhe->fnhe_next, 1); |
c5038a83 | 175 | |
2ffae99d TT |
176 | rt_fibinfo_free(&fnhe->fnhe_rth_input); |
177 | rt_fibinfo_free(&fnhe->fnhe_rth_output); | |
c5038a83 | 178 | |
4895c771 DM |
179 | kfree(fnhe); |
180 | ||
181 | fnhe = next; | |
182 | } | |
183 | } | |
184 | kfree(hash); | |
185 | } | |
186 | ||
c5038a83 | 187 | static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp) |
d26b3a7c ED |
188 | { |
189 | int cpu; | |
190 | ||
191 | if (!rtp) | |
192 | return; | |
193 | ||
194 | for_each_possible_cpu(cpu) { | |
195 | struct rtable *rt; | |
196 | ||
197 | rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1); | |
198 | if (rt) | |
199 | dst_free(&rt->dst); | |
200 | } | |
201 | free_percpu(rtp); | |
202 | } | |
203 | ||
1da177e4 | 204 | /* Release a nexthop info record */ |
19c1ea14 YZ |
205 | static void free_fib_info_rcu(struct rcu_head *head) |
206 | { | |
207 | struct fib_info *fi = container_of(head, struct fib_info, rcu); | |
208 | ||
e49cc0da YZ |
209 | change_nexthops(fi) { |
210 | if (nexthop_nh->nh_dev) | |
211 | dev_put(nexthop_nh->nh_dev); | |
5a6228a0 | 212 | lwtstate_put(nexthop_nh->nh_lwtstate); |
caa41527 | 213 | free_nh_exceptions(nexthop_nh); |
c5038a83 DM |
214 | rt_fibinfo_free_cpus(nexthop_nh->nh_pcpu_rth_output); |
215 | rt_fibinfo_free(&nexthop_nh->nh_rth_input); | |
e49cc0da YZ |
216 | } endfor_nexthops(fi); |
217 | ||
19c1ea14 YZ |
218 | if (fi->fib_metrics != (u32 *) dst_default_metrics) |
219 | kfree(fi->fib_metrics); | |
220 | kfree(fi); | |
221 | } | |
1da177e4 LT |
222 | |
223 | void free_fib_info(struct fib_info *fi) | |
224 | { | |
225 | if (fi->fib_dead == 0) { | |
058bd4d2 | 226 | pr_warn("Freeing alive fib_info %p\n", fi); |
1da177e4 LT |
227 | return; |
228 | } | |
1da177e4 | 229 | fib_info_cnt--; |
7a9bc9b8 DM |
230 | #ifdef CONFIG_IP_ROUTE_CLASSID |
231 | change_nexthops(fi) { | |
232 | if (nexthop_nh->nh_tclassid) | |
f4530fa5 | 233 | fi->fib_net->ipv4.fib_num_tclassid_users--; |
7a9bc9b8 DM |
234 | } endfor_nexthops(fi); |
235 | #endif | |
19c1ea14 | 236 | call_rcu(&fi->rcu, free_fib_info_rcu); |
1da177e4 LT |
237 | } |
238 | ||
239 | void fib_release_info(struct fib_info *fi) | |
240 | { | |
832b4c5e | 241 | spin_lock_bh(&fib_info_lock); |
1da177e4 LT |
242 | if (fi && --fi->fib_treeref == 0) { |
243 | hlist_del(&fi->fib_hash); | |
244 | if (fi->fib_prefsrc) | |
245 | hlist_del(&fi->fib_lhash); | |
246 | change_nexthops(fi) { | |
71fceff0 | 247 | if (!nexthop_nh->nh_dev) |
1da177e4 | 248 | continue; |
71fceff0 | 249 | hlist_del(&nexthop_nh->nh_hash); |
1da177e4 LT |
250 | } endfor_nexthops(fi) |
251 | fi->fib_dead = 1; | |
252 | fib_info_put(fi); | |
253 | } | |
832b4c5e | 254 | spin_unlock_bh(&fib_info_lock); |
1da177e4 LT |
255 | } |
256 | ||
6a31d2a9 | 257 | static inline int nh_comp(const struct fib_info *fi, const struct fib_info *ofi) |
1da177e4 LT |
258 | { |
259 | const struct fib_nh *onh = ofi->fib_nh; | |
260 | ||
261 | for_nexthops(fi) { | |
262 | if (nh->nh_oif != onh->nh_oif || | |
263 | nh->nh_gw != onh->nh_gw || | |
264 | nh->nh_scope != onh->nh_scope || | |
265 | #ifdef CONFIG_IP_ROUTE_MULTIPATH | |
266 | nh->nh_weight != onh->nh_weight || | |
267 | #endif | |
c7066f70 | 268 | #ifdef CONFIG_IP_ROUTE_CLASSID |
1da177e4 LT |
269 | nh->nh_tclassid != onh->nh_tclassid || |
270 | #endif | |
571e7226 | 271 | lwtunnel_cmp_encap(nh->nh_lwtstate, onh->nh_lwtstate) || |
8a3d0316 | 272 | ((nh->nh_flags ^ onh->nh_flags) & ~RTNH_COMPARE_MASK)) |
1da177e4 LT |
273 | return -1; |
274 | onh++; | |
275 | } endfor_nexthops(fi); | |
276 | return 0; | |
277 | } | |
278 | ||
88ebc72f DM |
279 | static inline unsigned int fib_devindex_hashfn(unsigned int val) |
280 | { | |
281 | unsigned int mask = DEVINDEX_HASHSIZE - 1; | |
282 | ||
283 | return (val ^ | |
284 | (val >> DEVINDEX_HASHBITS) ^ | |
285 | (val >> (DEVINDEX_HASHBITS * 2))) & mask; | |
286 | } | |
287 | ||
1da177e4 LT |
288 | static inline unsigned int fib_info_hashfn(const struct fib_info *fi) |
289 | { | |
123b9731 | 290 | unsigned int mask = (fib_info_hash_size - 1); |
1da177e4 LT |
291 | unsigned int val = fi->fib_nhs; |
292 | ||
37e826c5 | 293 | val ^= (fi->fib_protocol << 8) | fi->fib_scope; |
81f7bf6c | 294 | val ^= (__force u32)fi->fib_prefsrc; |
1da177e4 | 295 | val ^= fi->fib_priority; |
88ebc72f DM |
296 | for_nexthops(fi) { |
297 | val ^= fib_devindex_hashfn(nh->nh_oif); | |
298 | } endfor_nexthops(fi) | |
1da177e4 LT |
299 | |
300 | return (val ^ (val >> 7) ^ (val >> 12)) & mask; | |
301 | } | |
302 | ||
303 | static struct fib_info *fib_find_info(const struct fib_info *nfi) | |
304 | { | |
305 | struct hlist_head *head; | |
1da177e4 LT |
306 | struct fib_info *fi; |
307 | unsigned int hash; | |
308 | ||
309 | hash = fib_info_hashfn(nfi); | |
310 | head = &fib_info_hash[hash]; | |
311 | ||
b67bfe0d | 312 | hlist_for_each_entry(fi, head, fib_hash) { |
09ad9bc7 | 313 | if (!net_eq(fi->fib_net, nfi->fib_net)) |
4814bdbd | 314 | continue; |
1da177e4 LT |
315 | if (fi->fib_nhs != nfi->fib_nhs) |
316 | continue; | |
317 | if (nfi->fib_protocol == fi->fib_protocol && | |
37e826c5 | 318 | nfi->fib_scope == fi->fib_scope && |
1da177e4 LT |
319 | nfi->fib_prefsrc == fi->fib_prefsrc && |
320 | nfi->fib_priority == fi->fib_priority && | |
f4ef85bb | 321 | nfi->fib_type == fi->fib_type && |
1da177e4 | 322 | memcmp(nfi->fib_metrics, fi->fib_metrics, |
fcd13f42 | 323 | sizeof(u32) * RTAX_MAX) == 0 && |
8a3d0316 | 324 | !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) && |
1da177e4 LT |
325 | (nfi->fib_nhs == 0 || nh_comp(fi, nfi) == 0)) |
326 | return fi; | |
327 | } | |
328 | ||
329 | return NULL; | |
330 | } | |
331 | ||
1da177e4 | 332 | /* Check, that the gateway is already configured. |
6a31d2a9 | 333 | * Used only by redirect accept routine. |
1da177e4 | 334 | */ |
d878e72e | 335 | int ip_fib_check_default(__be32 gw, struct net_device *dev) |
1da177e4 LT |
336 | { |
337 | struct hlist_head *head; | |
1da177e4 LT |
338 | struct fib_nh *nh; |
339 | unsigned int hash; | |
340 | ||
832b4c5e | 341 | spin_lock(&fib_info_lock); |
1da177e4 LT |
342 | |
343 | hash = fib_devindex_hashfn(dev->ifindex); | |
344 | head = &fib_info_devhash[hash]; | |
b67bfe0d | 345 | hlist_for_each_entry(nh, head, nh_hash) { |
1da177e4 LT |
346 | if (nh->nh_dev == dev && |
347 | nh->nh_gw == gw && | |
6a31d2a9 | 348 | !(nh->nh_flags & RTNH_F_DEAD)) { |
832b4c5e | 349 | spin_unlock(&fib_info_lock); |
1da177e4 LT |
350 | return 0; |
351 | } | |
352 | } | |
353 | ||
832b4c5e | 354 | spin_unlock(&fib_info_lock); |
1da177e4 LT |
355 | |
356 | return -1; | |
357 | } | |
358 | ||
339bf98f TG |
359 | static inline size_t fib_nlmsg_size(struct fib_info *fi) |
360 | { | |
361 | size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg)) | |
362 | + nla_total_size(4) /* RTA_TABLE */ | |
363 | + nla_total_size(4) /* RTA_DST */ | |
364 | + nla_total_size(4) /* RTA_PRIORITY */ | |
ea697639 DB |
365 | + nla_total_size(4) /* RTA_PREFSRC */ |
366 | + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */ | |
339bf98f TG |
367 | |
368 | /* space for nested metrics */ | |
369 | payload += nla_total_size((RTAX_MAX * nla_total_size(4))); | |
370 | ||
371 | if (fi->fib_nhs) { | |
571e7226 | 372 | size_t nh_encapsize = 0; |
339bf98f TG |
373 | /* Also handles the special case fib_nhs == 1 */ |
374 | ||
375 | /* each nexthop is packed in an attribute */ | |
376 | size_t nhsize = nla_total_size(sizeof(struct rtnexthop)); | |
377 | ||
378 | /* may contain flow and gateway attribute */ | |
379 | nhsize += 2 * nla_total_size(4); | |
380 | ||
571e7226 RP |
381 | /* grab encap info */ |
382 | for_nexthops(fi) { | |
383 | if (nh->nh_lwtstate) { | |
384 | /* RTA_ENCAP_TYPE */ | |
385 | nh_encapsize += lwtunnel_get_encap_size( | |
386 | nh->nh_lwtstate); | |
387 | /* RTA_ENCAP */ | |
388 | nh_encapsize += nla_total_size(2); | |
389 | } | |
390 | } endfor_nexthops(fi); | |
391 | ||
339bf98f | 392 | /* all nexthops are packed in a nested attribute */ |
571e7226 RP |
393 | payload += nla_total_size((fi->fib_nhs * nhsize) + |
394 | nh_encapsize); | |
395 | ||
339bf98f TG |
396 | } |
397 | ||
398 | return payload; | |
399 | } | |
400 | ||
81f7bf6c | 401 | void rtmsg_fib(int event, __be32 key, struct fib_alias *fa, |
9877b253 | 402 | int dst_len, u32 tb_id, const struct nl_info *info, |
b8f55831 | 403 | unsigned int nlm_flags) |
1da177e4 LT |
404 | { |
405 | struct sk_buff *skb; | |
4e902c57 | 406 | u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; |
f21c7bc5 | 407 | int err = -ENOBUFS; |
1da177e4 | 408 | |
339bf98f | 409 | skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL); |
51456b29 | 410 | if (!skb) |
f21c7bc5 | 411 | goto errout; |
1da177e4 | 412 | |
15e47304 | 413 | err = fib_dump_info(skb, info->portid, seq, event, tb_id, |
37e826c5 | 414 | fa->fa_type, key, dst_len, |
b8f55831 | 415 | fa->fa_tos, fa->fa_info, nlm_flags); |
26932566 PM |
416 | if (err < 0) { |
417 | /* -EMSGSIZE implies BUG in fib_nlmsg_size() */ | |
418 | WARN_ON(err == -EMSGSIZE); | |
419 | kfree_skb(skb); | |
420 | goto errout; | |
421 | } | |
15e47304 | 422 | rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE, |
1ce85fe4 PNA |
423 | info->nlh, GFP_KERNEL); |
424 | return; | |
f21c7bc5 TG |
425 | errout: |
426 | if (err < 0) | |
4d1169c1 | 427 | rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err); |
1da177e4 LT |
428 | } |
429 | ||
c9cb6b6e SH |
430 | static int fib_detect_death(struct fib_info *fi, int order, |
431 | struct fib_info **last_resort, int *last_idx, | |
432 | int dflt) | |
1da177e4 LT |
433 | { |
434 | struct neighbour *n; | |
435 | int state = NUD_NONE; | |
436 | ||
437 | n = neigh_lookup(&arp_tbl, &fi->fib_nh[0].nh_gw, fi->fib_dev); | |
438 | if (n) { | |
439 | state = n->nud_state; | |
440 | neigh_release(n); | |
88f64320 JA |
441 | } else { |
442 | return 0; | |
1da177e4 | 443 | } |
d9319100 | 444 | if (state == NUD_REACHABLE) |
1da177e4 | 445 | return 0; |
6a31d2a9 | 446 | if ((state & NUD_VALID) && order != dflt) |
1da177e4 | 447 | return 0; |
6a31d2a9 | 448 | if ((state & NUD_VALID) || |
88f64320 | 449 | (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) { |
1da177e4 LT |
450 | *last_resort = fi; |
451 | *last_idx = order; | |
452 | } | |
453 | return 1; | |
454 | } | |
455 | ||
456 | #ifdef CONFIG_IP_ROUTE_MULTIPATH | |
457 | ||
4e902c57 | 458 | static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining) |
1da177e4 LT |
459 | { |
460 | int nhs = 0; | |
1da177e4 | 461 | |
4e902c57 | 462 | while (rtnh_ok(rtnh, remaining)) { |
1da177e4 | 463 | nhs++; |
4e902c57 TG |
464 | rtnh = rtnh_next(rtnh, &remaining); |
465 | } | |
466 | ||
467 | /* leftover implies invalid nexthop configuration, discard it */ | |
468 | return remaining > 0 ? 0 : nhs; | |
1da177e4 LT |
469 | } |
470 | ||
4e902c57 TG |
471 | static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh, |
472 | int remaining, struct fib_config *cfg) | |
1da177e4 | 473 | { |
571e7226 RP |
474 | struct net *net = cfg->fc_nlinfo.nl_net; |
475 | int ret; | |
476 | ||
1da177e4 | 477 | change_nexthops(fi) { |
4e902c57 TG |
478 | int attrlen; |
479 | ||
480 | if (!rtnh_ok(rtnh, remaining)) | |
1da177e4 | 481 | return -EINVAL; |
4e902c57 | 482 | |
71fceff0 DM |
483 | nexthop_nh->nh_flags = |
484 | (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags; | |
485 | nexthop_nh->nh_oif = rtnh->rtnh_ifindex; | |
486 | nexthop_nh->nh_weight = rtnh->rtnh_hops + 1; | |
4e902c57 TG |
487 | |
488 | attrlen = rtnh_attrlen(rtnh); | |
489 | if (attrlen > 0) { | |
490 | struct nlattr *nla, *attrs = rtnh_attrs(rtnh); | |
491 | ||
492 | nla = nla_find(attrs, attrlen, RTA_GATEWAY); | |
67b61f6c | 493 | nexthop_nh->nh_gw = nla ? nla_get_in_addr(nla) : 0; |
c7066f70 | 494 | #ifdef CONFIG_IP_ROUTE_CLASSID |
4e902c57 | 495 | nla = nla_find(attrs, attrlen, RTA_FLOW); |
71fceff0 | 496 | nexthop_nh->nh_tclassid = nla ? nla_get_u32(nla) : 0; |
7a9bc9b8 | 497 | if (nexthop_nh->nh_tclassid) |
f4530fa5 | 498 | fi->fib_net->ipv4.fib_num_tclassid_users++; |
1da177e4 | 499 | #endif |
571e7226 RP |
500 | nla = nla_find(attrs, attrlen, RTA_ENCAP); |
501 | if (nla) { | |
502 | struct lwtunnel_state *lwtstate; | |
503 | struct net_device *dev = NULL; | |
504 | struct nlattr *nla_entype; | |
505 | ||
506 | nla_entype = nla_find(attrs, attrlen, | |
507 | RTA_ENCAP_TYPE); | |
508 | if (!nla_entype) | |
509 | goto err_inval; | |
510 | if (cfg->fc_oif) | |
511 | dev = __dev_get_by_index(net, cfg->fc_oif); | |
512 | ret = lwtunnel_build_state(dev, nla_get_u16( | |
513 | nla_entype), | |
514 | nla, &lwtstate); | |
515 | if (ret) | |
516 | goto errout; | |
5a6228a0 ND |
517 | nexthop_nh->nh_lwtstate = |
518 | lwtstate_get(lwtstate); | |
571e7226 | 519 | } |
1da177e4 | 520 | } |
4e902c57 TG |
521 | |
522 | rtnh = rtnh_next(rtnh, &remaining); | |
1da177e4 | 523 | } endfor_nexthops(fi); |
4e902c57 | 524 | |
1da177e4 | 525 | return 0; |
571e7226 RP |
526 | |
527 | err_inval: | |
528 | ret = -EINVAL; | |
529 | ||
530 | errout: | |
531 | return ret; | |
1da177e4 LT |
532 | } |
533 | ||
534 | #endif | |
535 | ||
571e7226 RP |
536 | int fib_encap_match(struct net *net, u16 encap_type, |
537 | struct nlattr *encap, | |
538 | int oif, const struct fib_nh *nh) | |
539 | { | |
540 | struct lwtunnel_state *lwtstate; | |
541 | struct net_device *dev = NULL; | |
542 | int ret; | |
543 | ||
544 | if (encap_type == LWTUNNEL_ENCAP_NONE) | |
545 | return 0; | |
546 | ||
547 | if (oif) | |
548 | dev = __dev_get_by_index(net, oif); | |
549 | ret = lwtunnel_build_state(dev, encap_type, | |
550 | encap, &lwtstate); | |
551 | if (!ret) | |
552 | return lwtunnel_cmp_encap(lwtstate, nh->nh_lwtstate); | |
553 | ||
554 | return 0; | |
555 | } | |
556 | ||
4e902c57 | 557 | int fib_nh_match(struct fib_config *cfg, struct fib_info *fi) |
1da177e4 | 558 | { |
571e7226 | 559 | struct net *net = cfg->fc_nlinfo.nl_net; |
1da177e4 | 560 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
4e902c57 TG |
561 | struct rtnexthop *rtnh; |
562 | int remaining; | |
1da177e4 LT |
563 | #endif |
564 | ||
4e902c57 | 565 | if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority) |
1da177e4 LT |
566 | return 1; |
567 | ||
4e902c57 | 568 | if (cfg->fc_oif || cfg->fc_gw) { |
571e7226 RP |
569 | if (cfg->fc_encap) { |
570 | if (fib_encap_match(net, cfg->fc_encap_type, | |
571 | cfg->fc_encap, cfg->fc_oif, | |
572 | fi->fib_nh)) | |
573 | return 1; | |
574 | } | |
4e902c57 TG |
575 | if ((!cfg->fc_oif || cfg->fc_oif == fi->fib_nh->nh_oif) && |
576 | (!cfg->fc_gw || cfg->fc_gw == fi->fib_nh->nh_gw)) | |
1da177e4 LT |
577 | return 0; |
578 | return 1; | |
579 | } | |
580 | ||
581 | #ifdef CONFIG_IP_ROUTE_MULTIPATH | |
51456b29 | 582 | if (!cfg->fc_mp) |
1da177e4 | 583 | return 0; |
4e902c57 TG |
584 | |
585 | rtnh = cfg->fc_mp; | |
586 | remaining = cfg->fc_mp_len; | |
e905a9ed | 587 | |
1da177e4 | 588 | for_nexthops(fi) { |
4e902c57 | 589 | int attrlen; |
1da177e4 | 590 | |
4e902c57 | 591 | if (!rtnh_ok(rtnh, remaining)) |
1da177e4 | 592 | return -EINVAL; |
4e902c57 TG |
593 | |
594 | if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->nh_oif) | |
1da177e4 | 595 | return 1; |
4e902c57 TG |
596 | |
597 | attrlen = rtnh_attrlen(rtnh); | |
f76936d0 | 598 | if (attrlen > 0) { |
4e902c57 TG |
599 | struct nlattr *nla, *attrs = rtnh_attrs(rtnh); |
600 | ||
601 | nla = nla_find(attrs, attrlen, RTA_GATEWAY); | |
67b61f6c | 602 | if (nla && nla_get_in_addr(nla) != nh->nh_gw) |
1da177e4 | 603 | return 1; |
c7066f70 | 604 | #ifdef CONFIG_IP_ROUTE_CLASSID |
4e902c57 TG |
605 | nla = nla_find(attrs, attrlen, RTA_FLOW); |
606 | if (nla && nla_get_u32(nla) != nh->nh_tclassid) | |
1da177e4 LT |
607 | return 1; |
608 | #endif | |
609 | } | |
4e902c57 TG |
610 | |
611 | rtnh = rtnh_next(rtnh, &remaining); | |
1da177e4 LT |
612 | } endfor_nexthops(fi); |
613 | #endif | |
614 | return 0; | |
615 | } | |
616 | ||
617 | ||
618 | /* | |
6a31d2a9 ED |
619 | * Picture |
620 | * ------- | |
621 | * | |
622 | * Semantics of nexthop is very messy by historical reasons. | |
623 | * We have to take into account, that: | |
624 | * a) gateway can be actually local interface address, | |
625 | * so that gatewayed route is direct. | |
626 | * b) gateway must be on-link address, possibly | |
627 | * described not by an ifaddr, but also by a direct route. | |
628 | * c) If both gateway and interface are specified, they should not | |
629 | * contradict. | |
630 | * d) If we use tunnel routes, gateway could be not on-link. | |
631 | * | |
632 | * Attempt to reconcile all of these (alas, self-contradictory) conditions | |
633 | * results in pretty ugly and hairy code with obscure logic. | |
634 | * | |
635 | * I chose to generalized it instead, so that the size | |
636 | * of code does not increase practically, but it becomes | |
637 | * much more general. | |
638 | * Every prefix is assigned a "scope" value: "host" is local address, | |
639 | * "link" is direct route, | |
640 | * [ ... "site" ... "interior" ... ] | |
641 | * and "universe" is true gateway route with global meaning. | |
642 | * | |
643 | * Every prefix refers to a set of "nexthop"s (gw, oif), | |
644 | * where gw must have narrower scope. This recursion stops | |
645 | * when gw has LOCAL scope or if "nexthop" is declared ONLINK, | |
646 | * which means that gw is forced to be on link. | |
647 | * | |
648 | * Code is still hairy, but now it is apparently logically | |
649 | * consistent and very flexible. F.e. as by-product it allows | |
650 | * to co-exists in peace independent exterior and interior | |
651 | * routing processes. | |
652 | * | |
653 | * Normally it looks as following. | |
654 | * | |
655 | * {universe prefix} -> (gw, oif) [scope link] | |
656 | * | | |
657 | * |-> {link prefix} -> (gw, oif) [scope local] | |
658 | * | | |
659 | * |-> {local prefix} (terminal node) | |
1da177e4 | 660 | */ |
4e902c57 TG |
661 | static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi, |
662 | struct fib_nh *nh) | |
1da177e4 LT |
663 | { |
664 | int err; | |
86167a37 | 665 | struct net *net; |
6a31d2a9 | 666 | struct net_device *dev; |
1da177e4 | 667 | |
86167a37 | 668 | net = cfg->fc_nlinfo.nl_net; |
1da177e4 LT |
669 | if (nh->nh_gw) { |
670 | struct fib_result res; | |
671 | ||
6a31d2a9 | 672 | if (nh->nh_flags & RTNH_F_ONLINK) { |
30bbaa19 | 673 | unsigned int addr_type; |
1da177e4 | 674 | |
4e902c57 | 675 | if (cfg->fc_scope >= RT_SCOPE_LINK) |
1da177e4 | 676 | return -EINVAL; |
6a31d2a9 ED |
677 | dev = __dev_get_by_index(net, nh->nh_oif); |
678 | if (!dev) | |
1da177e4 | 679 | return -ENODEV; |
6a31d2a9 | 680 | if (!(dev->flags & IFF_UP)) |
1da177e4 | 681 | return -ENETDOWN; |
30bbaa19 DA |
682 | addr_type = inet_addr_type_dev_table(net, dev, nh->nh_gw); |
683 | if (addr_type != RTN_UNICAST) | |
684 | return -EINVAL; | |
8a3d0316 AG |
685 | if (!netif_carrier_ok(dev)) |
686 | nh->nh_flags |= RTNH_F_LINKDOWN; | |
1da177e4 LT |
687 | nh->nh_dev = dev; |
688 | dev_hold(dev); | |
689 | nh->nh_scope = RT_SCOPE_LINK; | |
690 | return 0; | |
691 | } | |
ebc0ffae | 692 | rcu_read_lock(); |
1da177e4 | 693 | { |
3bfd8472 | 694 | struct fib_table *tbl = NULL; |
9ade2286 DM |
695 | struct flowi4 fl4 = { |
696 | .daddr = nh->nh_gw, | |
697 | .flowi4_scope = cfg->fc_scope + 1, | |
698 | .flowi4_oif = nh->nh_oif, | |
6a662719 | 699 | .flowi4_iif = LOOPBACK_IFINDEX, |
4e902c57 | 700 | }; |
1da177e4 LT |
701 | |
702 | /* It is not necessary, but requires a bit of thinking */ | |
9ade2286 DM |
703 | if (fl4.flowi4_scope < RT_SCOPE_LINK) |
704 | fl4.flowi4_scope = RT_SCOPE_LINK; | |
3bfd8472 DA |
705 | |
706 | if (cfg->fc_table) | |
707 | tbl = fib_get_table(net, cfg->fc_table); | |
708 | ||
709 | if (tbl) | |
710 | err = fib_table_lookup(tbl, &fl4, &res, | |
711 | FIB_LOOKUP_IGNORE_LINKSTATE); | |
712 | else | |
713 | err = fib_lookup(net, &fl4, &res, | |
714 | FIB_LOOKUP_IGNORE_LINKSTATE); | |
ebc0ffae ED |
715 | if (err) { |
716 | rcu_read_unlock(); | |
1da177e4 | 717 | return err; |
ebc0ffae | 718 | } |
1da177e4 LT |
719 | } |
720 | err = -EINVAL; | |
721 | if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) | |
722 | goto out; | |
723 | nh->nh_scope = res.scope; | |
724 | nh->nh_oif = FIB_RES_OIF(res); | |
6a31d2a9 ED |
725 | nh->nh_dev = dev = FIB_RES_DEV(res); |
726 | if (!dev) | |
1da177e4 | 727 | goto out; |
6a31d2a9 | 728 | dev_hold(dev); |
8a3d0316 AG |
729 | if (!netif_carrier_ok(dev)) |
730 | nh->nh_flags |= RTNH_F_LINKDOWN; | |
8723e1b4 | 731 | err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN; |
1da177e4 LT |
732 | } else { |
733 | struct in_device *in_dev; | |
734 | ||
6a31d2a9 | 735 | if (nh->nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) |
1da177e4 LT |
736 | return -EINVAL; |
737 | ||
8723e1b4 ED |
738 | rcu_read_lock(); |
739 | err = -ENODEV; | |
86167a37 | 740 | in_dev = inetdev_by_index(net, nh->nh_oif); |
51456b29 | 741 | if (!in_dev) |
8723e1b4 ED |
742 | goto out; |
743 | err = -ENETDOWN; | |
744 | if (!(in_dev->dev->flags & IFF_UP)) | |
745 | goto out; | |
1da177e4 LT |
746 | nh->nh_dev = in_dev->dev; |
747 | dev_hold(nh->nh_dev); | |
748 | nh->nh_scope = RT_SCOPE_HOST; | |
8a3d0316 AG |
749 | if (!netif_carrier_ok(nh->nh_dev)) |
750 | nh->nh_flags |= RTNH_F_LINKDOWN; | |
8723e1b4 | 751 | err = 0; |
1da177e4 | 752 | } |
8723e1b4 ED |
753 | out: |
754 | rcu_read_unlock(); | |
755 | return err; | |
1da177e4 LT |
756 | } |
757 | ||
81f7bf6c | 758 | static inline unsigned int fib_laddr_hashfn(__be32 val) |
1da177e4 | 759 | { |
123b9731 | 760 | unsigned int mask = (fib_info_hash_size - 1); |
1da177e4 | 761 | |
6a31d2a9 ED |
762 | return ((__force u32)val ^ |
763 | ((__force u32)val >> 7) ^ | |
764 | ((__force u32)val >> 14)) & mask; | |
1da177e4 LT |
765 | } |
766 | ||
123b9731 | 767 | static struct hlist_head *fib_info_hash_alloc(int bytes) |
1da177e4 LT |
768 | { |
769 | if (bytes <= PAGE_SIZE) | |
88f83491 | 770 | return kzalloc(bytes, GFP_KERNEL); |
1da177e4 LT |
771 | else |
772 | return (struct hlist_head *) | |
6a31d2a9 ED |
773 | __get_free_pages(GFP_KERNEL | __GFP_ZERO, |
774 | get_order(bytes)); | |
1da177e4 LT |
775 | } |
776 | ||
123b9731 | 777 | static void fib_info_hash_free(struct hlist_head *hash, int bytes) |
1da177e4 LT |
778 | { |
779 | if (!hash) | |
780 | return; | |
781 | ||
782 | if (bytes <= PAGE_SIZE) | |
783 | kfree(hash); | |
784 | else | |
785 | free_pages((unsigned long) hash, get_order(bytes)); | |
786 | } | |
787 | ||
123b9731 DM |
788 | static void fib_info_hash_move(struct hlist_head *new_info_hash, |
789 | struct hlist_head *new_laddrhash, | |
790 | unsigned int new_size) | |
1da177e4 | 791 | { |
b7656e7f | 792 | struct hlist_head *old_info_hash, *old_laddrhash; |
123b9731 | 793 | unsigned int old_size = fib_info_hash_size; |
b7656e7f | 794 | unsigned int i, bytes; |
1da177e4 | 795 | |
832b4c5e | 796 | spin_lock_bh(&fib_info_lock); |
b7656e7f DM |
797 | old_info_hash = fib_info_hash; |
798 | old_laddrhash = fib_info_laddrhash; | |
123b9731 | 799 | fib_info_hash_size = new_size; |
1da177e4 LT |
800 | |
801 | for (i = 0; i < old_size; i++) { | |
802 | struct hlist_head *head = &fib_info_hash[i]; | |
b67bfe0d | 803 | struct hlist_node *n; |
1da177e4 LT |
804 | struct fib_info *fi; |
805 | ||
b67bfe0d | 806 | hlist_for_each_entry_safe(fi, n, head, fib_hash) { |
1da177e4 LT |
807 | struct hlist_head *dest; |
808 | unsigned int new_hash; | |
809 | ||
1da177e4 LT |
810 | new_hash = fib_info_hashfn(fi); |
811 | dest = &new_info_hash[new_hash]; | |
812 | hlist_add_head(&fi->fib_hash, dest); | |
813 | } | |
814 | } | |
815 | fib_info_hash = new_info_hash; | |
816 | ||
817 | for (i = 0; i < old_size; i++) { | |
818 | struct hlist_head *lhead = &fib_info_laddrhash[i]; | |
b67bfe0d | 819 | struct hlist_node *n; |
1da177e4 LT |
820 | struct fib_info *fi; |
821 | ||
b67bfe0d | 822 | hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) { |
1da177e4 LT |
823 | struct hlist_head *ldest; |
824 | unsigned int new_hash; | |
825 | ||
1da177e4 LT |
826 | new_hash = fib_laddr_hashfn(fi->fib_prefsrc); |
827 | ldest = &new_laddrhash[new_hash]; | |
828 | hlist_add_head(&fi->fib_lhash, ldest); | |
829 | } | |
830 | } | |
831 | fib_info_laddrhash = new_laddrhash; | |
832 | ||
832b4c5e | 833 | spin_unlock_bh(&fib_info_lock); |
b7656e7f DM |
834 | |
835 | bytes = old_size * sizeof(struct hlist_head *); | |
123b9731 DM |
836 | fib_info_hash_free(old_info_hash, bytes); |
837 | fib_info_hash_free(old_laddrhash, bytes); | |
1da177e4 LT |
838 | } |
839 | ||
436c3b66 DM |
840 | __be32 fib_info_update_nh_saddr(struct net *net, struct fib_nh *nh) |
841 | { | |
842 | nh->nh_saddr = inet_select_addr(nh->nh_dev, | |
843 | nh->nh_gw, | |
37e826c5 | 844 | nh->nh_parent->fib_scope); |
436c3b66 DM |
845 | nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid); |
846 | ||
847 | return nh->nh_saddr; | |
848 | } | |
849 | ||
021dd3b8 DA |
850 | static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc) |
851 | { | |
852 | if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst || | |
853 | fib_prefsrc != cfg->fc_dst) { | |
854 | int tb_id = cfg->fc_table; | |
855 | ||
856 | if (tb_id == RT_TABLE_MAIN) | |
857 | tb_id = RT_TABLE_LOCAL; | |
858 | ||
859 | if (inet_addr_type_table(cfg->fc_nlinfo.nl_net, | |
860 | fib_prefsrc, tb_id) != RTN_LOCAL) { | |
861 | return false; | |
862 | } | |
863 | } | |
864 | return true; | |
865 | } | |
866 | ||
4e902c57 | 867 | struct fib_info *fib_create_info(struct fib_config *cfg) |
1da177e4 LT |
868 | { |
869 | int err; | |
870 | struct fib_info *fi = NULL; | |
871 | struct fib_info *ofi; | |
1da177e4 | 872 | int nhs = 1; |
7462bd74 | 873 | struct net *net = cfg->fc_nlinfo.nl_net; |
1da177e4 | 874 | |
4c8237cd DM |
875 | if (cfg->fc_type > RTN_MAX) |
876 | goto err_inval; | |
877 | ||
1da177e4 | 878 | /* Fast check to catch the most weird cases */ |
4e902c57 | 879 | if (fib_props[cfg->fc_type].scope > cfg->fc_scope) |
1da177e4 LT |
880 | goto err_inval; |
881 | ||
882 | #ifdef CONFIG_IP_ROUTE_MULTIPATH | |
4e902c57 TG |
883 | if (cfg->fc_mp) { |
884 | nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len); | |
1da177e4 LT |
885 | if (nhs == 0) |
886 | goto err_inval; | |
887 | } | |
888 | #endif | |
1da177e4 LT |
889 | |
890 | err = -ENOBUFS; | |
123b9731 DM |
891 | if (fib_info_cnt >= fib_info_hash_size) { |
892 | unsigned int new_size = fib_info_hash_size << 1; | |
1da177e4 LT |
893 | struct hlist_head *new_info_hash; |
894 | struct hlist_head *new_laddrhash; | |
895 | unsigned int bytes; | |
896 | ||
897 | if (!new_size) | |
d94ce9b2 | 898 | new_size = 16; |
1da177e4 | 899 | bytes = new_size * sizeof(struct hlist_head *); |
123b9731 DM |
900 | new_info_hash = fib_info_hash_alloc(bytes); |
901 | new_laddrhash = fib_info_hash_alloc(bytes); | |
1da177e4 | 902 | if (!new_info_hash || !new_laddrhash) { |
123b9731 DM |
903 | fib_info_hash_free(new_info_hash, bytes); |
904 | fib_info_hash_free(new_laddrhash, bytes); | |
88f83491 | 905 | } else |
123b9731 | 906 | fib_info_hash_move(new_info_hash, new_laddrhash, new_size); |
1da177e4 | 907 | |
123b9731 | 908 | if (!fib_info_hash_size) |
1da177e4 LT |
909 | goto failure; |
910 | } | |
911 | ||
0da974f4 | 912 | fi = kzalloc(sizeof(*fi)+nhs*sizeof(struct fib_nh), GFP_KERNEL); |
51456b29 | 913 | if (!fi) |
1da177e4 | 914 | goto failure; |
aeefa1ec | 915 | fib_info_cnt++; |
725d1e1b DM |
916 | if (cfg->fc_mx) { |
917 | fi->fib_metrics = kzalloc(sizeof(u32) * RTAX_MAX, GFP_KERNEL); | |
918 | if (!fi->fib_metrics) | |
919 | goto failure; | |
920 | } else | |
921 | fi->fib_metrics = (u32 *) dst_default_metrics; | |
1da177e4 | 922 | |
efd7ef1c | 923 | fi->fib_net = net; |
4e902c57 | 924 | fi->fib_protocol = cfg->fc_protocol; |
37e826c5 | 925 | fi->fib_scope = cfg->fc_scope; |
4e902c57 TG |
926 | fi->fib_flags = cfg->fc_flags; |
927 | fi->fib_priority = cfg->fc_priority; | |
928 | fi->fib_prefsrc = cfg->fc_prefsrc; | |
f4ef85bb | 929 | fi->fib_type = cfg->fc_type; |
1da177e4 LT |
930 | |
931 | fi->fib_nhs = nhs; | |
932 | change_nexthops(fi) { | |
71fceff0 | 933 | nexthop_nh->nh_parent = fi; |
d26b3a7c | 934 | nexthop_nh->nh_pcpu_rth_output = alloc_percpu(struct rtable __rcu *); |
f8a17175 JA |
935 | if (!nexthop_nh->nh_pcpu_rth_output) |
936 | goto failure; | |
1da177e4 LT |
937 | } endfor_nexthops(fi) |
938 | ||
4e902c57 TG |
939 | if (cfg->fc_mx) { |
940 | struct nlattr *nla; | |
941 | int remaining; | |
942 | ||
943 | nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) { | |
8f4c1f9b | 944 | int type = nla_type(nla); |
4e902c57 TG |
945 | |
946 | if (type) { | |
6fac2625 DM |
947 | u32 val; |
948 | ||
4e902c57 | 949 | if (type > RTAX_MAX) |
1da177e4 | 950 | goto err_inval; |
ea697639 DB |
951 | if (type == RTAX_CC_ALGO) { |
952 | char tmp[TCP_CA_NAME_MAX]; | |
953 | ||
954 | nla_strlcpy(tmp, nla, sizeof(tmp)); | |
955 | val = tcp_ca_get_key_by_name(tmp); | |
956 | if (val == TCP_CA_UNSPEC) | |
957 | goto err_inval; | |
958 | } else { | |
959 | val = nla_get_u32(nla); | |
960 | } | |
6fac2625 DM |
961 | if (type == RTAX_ADVMSS && val > 65535 - 40) |
962 | val = 65535 - 40; | |
710ab6c0 DM |
963 | if (type == RTAX_MTU && val > 65535 - 15) |
964 | val = 65535 - 15; | |
6fac2625 | 965 | fi->fib_metrics[type - 1] = val; |
1da177e4 | 966 | } |
1da177e4 LT |
967 | } |
968 | } | |
1da177e4 | 969 | |
4e902c57 | 970 | if (cfg->fc_mp) { |
1da177e4 | 971 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
4e902c57 TG |
972 | err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg); |
973 | if (err != 0) | |
1da177e4 | 974 | goto failure; |
4e902c57 | 975 | if (cfg->fc_oif && fi->fib_nh->nh_oif != cfg->fc_oif) |
1da177e4 | 976 | goto err_inval; |
4e902c57 | 977 | if (cfg->fc_gw && fi->fib_nh->nh_gw != cfg->fc_gw) |
1da177e4 | 978 | goto err_inval; |
c7066f70 | 979 | #ifdef CONFIG_IP_ROUTE_CLASSID |
4e902c57 | 980 | if (cfg->fc_flow && fi->fib_nh->nh_tclassid != cfg->fc_flow) |
1da177e4 LT |
981 | goto err_inval; |
982 | #endif | |
983 | #else | |
984 | goto err_inval; | |
985 | #endif | |
986 | } else { | |
987 | struct fib_nh *nh = fi->fib_nh; | |
4e902c57 | 988 | |
571e7226 RP |
989 | if (cfg->fc_encap) { |
990 | struct lwtunnel_state *lwtstate; | |
991 | struct net_device *dev = NULL; | |
992 | ||
993 | if (cfg->fc_encap_type == LWTUNNEL_ENCAP_NONE) | |
994 | goto err_inval; | |
995 | if (cfg->fc_oif) | |
996 | dev = __dev_get_by_index(net, cfg->fc_oif); | |
997 | err = lwtunnel_build_state(dev, cfg->fc_encap_type, | |
998 | cfg->fc_encap, &lwtstate); | |
999 | if (err) | |
1000 | goto failure; | |
1001 | ||
5a6228a0 | 1002 | nh->nh_lwtstate = lwtstate_get(lwtstate); |
571e7226 | 1003 | } |
4e902c57 TG |
1004 | nh->nh_oif = cfg->fc_oif; |
1005 | nh->nh_gw = cfg->fc_gw; | |
1006 | nh->nh_flags = cfg->fc_flags; | |
c7066f70 | 1007 | #ifdef CONFIG_IP_ROUTE_CLASSID |
4e902c57 | 1008 | nh->nh_tclassid = cfg->fc_flow; |
7a9bc9b8 | 1009 | if (nh->nh_tclassid) |
f4530fa5 | 1010 | fi->fib_net->ipv4.fib_num_tclassid_users++; |
1da177e4 | 1011 | #endif |
1da177e4 LT |
1012 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
1013 | nh->nh_weight = 1; | |
1014 | #endif | |
1015 | } | |
1016 | ||
4e902c57 TG |
1017 | if (fib_props[cfg->fc_type].error) { |
1018 | if (cfg->fc_gw || cfg->fc_oif || cfg->fc_mp) | |
1da177e4 LT |
1019 | goto err_inval; |
1020 | goto link_it; | |
4c8237cd DM |
1021 | } else { |
1022 | switch (cfg->fc_type) { | |
1023 | case RTN_UNICAST: | |
1024 | case RTN_LOCAL: | |
1025 | case RTN_BROADCAST: | |
1026 | case RTN_ANYCAST: | |
1027 | case RTN_MULTICAST: | |
1028 | break; | |
1029 | default: | |
1030 | goto err_inval; | |
1031 | } | |
1da177e4 LT |
1032 | } |
1033 | ||
4e902c57 | 1034 | if (cfg->fc_scope > RT_SCOPE_HOST) |
1da177e4 LT |
1035 | goto err_inval; |
1036 | ||
4e902c57 | 1037 | if (cfg->fc_scope == RT_SCOPE_HOST) { |
1da177e4 LT |
1038 | struct fib_nh *nh = fi->fib_nh; |
1039 | ||
1040 | /* Local address is added. */ | |
1041 | if (nhs != 1 || nh->nh_gw) | |
1042 | goto err_inval; | |
1043 | nh->nh_scope = RT_SCOPE_NOWHERE; | |
7462bd74 | 1044 | nh->nh_dev = dev_get_by_index(net, fi->fib_nh->nh_oif); |
1da177e4 | 1045 | err = -ENODEV; |
51456b29 | 1046 | if (!nh->nh_dev) |
1da177e4 LT |
1047 | goto failure; |
1048 | } else { | |
8a3d0316 AG |
1049 | int linkdown = 0; |
1050 | ||
1da177e4 | 1051 | change_nexthops(fi) { |
6a31d2a9 ED |
1052 | err = fib_check_nh(cfg, fi, nexthop_nh); |
1053 | if (err != 0) | |
1da177e4 | 1054 | goto failure; |
8a3d0316 AG |
1055 | if (nexthop_nh->nh_flags & RTNH_F_LINKDOWN) |
1056 | linkdown++; | |
1da177e4 | 1057 | } endfor_nexthops(fi) |
8a3d0316 AG |
1058 | if (linkdown == fi->fib_nhs) |
1059 | fi->fib_flags |= RTNH_F_LINKDOWN; | |
1da177e4 LT |
1060 | } |
1061 | ||
021dd3b8 DA |
1062 | if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) |
1063 | goto err_inval; | |
1da177e4 | 1064 | |
1fc050a1 | 1065 | change_nexthops(fi) { |
436c3b66 | 1066 | fib_info_update_nh_saddr(net, nexthop_nh); |
1fc050a1 DM |
1067 | } endfor_nexthops(fi) |
1068 | ||
1da177e4 | 1069 | link_it: |
6a31d2a9 ED |
1070 | ofi = fib_find_info(fi); |
1071 | if (ofi) { | |
1da177e4 LT |
1072 | fi->fib_dead = 1; |
1073 | free_fib_info(fi); | |
1074 | ofi->fib_treeref++; | |
1075 | return ofi; | |
1076 | } | |
1077 | ||
1078 | fi->fib_treeref++; | |
1079 | atomic_inc(&fi->fib_clntref); | |
832b4c5e | 1080 | spin_lock_bh(&fib_info_lock); |
1da177e4 LT |
1081 | hlist_add_head(&fi->fib_hash, |
1082 | &fib_info_hash[fib_info_hashfn(fi)]); | |
1083 | if (fi->fib_prefsrc) { | |
1084 | struct hlist_head *head; | |
1085 | ||
1086 | head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)]; | |
1087 | hlist_add_head(&fi->fib_lhash, head); | |
1088 | } | |
1089 | change_nexthops(fi) { | |
1090 | struct hlist_head *head; | |
1091 | unsigned int hash; | |
1092 | ||
71fceff0 | 1093 | if (!nexthop_nh->nh_dev) |
1da177e4 | 1094 | continue; |
71fceff0 | 1095 | hash = fib_devindex_hashfn(nexthop_nh->nh_dev->ifindex); |
1da177e4 | 1096 | head = &fib_info_devhash[hash]; |
71fceff0 | 1097 | hlist_add_head(&nexthop_nh->nh_hash, head); |
1da177e4 | 1098 | } endfor_nexthops(fi) |
832b4c5e | 1099 | spin_unlock_bh(&fib_info_lock); |
1da177e4 LT |
1100 | return fi; |
1101 | ||
1102 | err_inval: | |
1103 | err = -EINVAL; | |
1104 | ||
1105 | failure: | |
e905a9ed | 1106 | if (fi) { |
1da177e4 LT |
1107 | fi->fib_dead = 1; |
1108 | free_fib_info(fi); | |
1109 | } | |
4e902c57 TG |
1110 | |
1111 | return ERR_PTR(err); | |
1da177e4 LT |
1112 | } |
1113 | ||
15e47304 | 1114 | int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event, |
37e826c5 | 1115 | u32 tb_id, u8 type, __be32 dst, int dst_len, u8 tos, |
be403ea1 | 1116 | struct fib_info *fi, unsigned int flags) |
1da177e4 | 1117 | { |
be403ea1 | 1118 | struct nlmsghdr *nlh; |
1da177e4 | 1119 | struct rtmsg *rtm; |
1da177e4 | 1120 | |
15e47304 | 1121 | nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags); |
51456b29 | 1122 | if (!nlh) |
26932566 | 1123 | return -EMSGSIZE; |
be403ea1 TG |
1124 | |
1125 | rtm = nlmsg_data(nlh); | |
1da177e4 LT |
1126 | rtm->rtm_family = AF_INET; |
1127 | rtm->rtm_dst_len = dst_len; | |
1128 | rtm->rtm_src_len = 0; | |
1129 | rtm->rtm_tos = tos; | |
709772e6 KPO |
1130 | if (tb_id < 256) |
1131 | rtm->rtm_table = tb_id; | |
1132 | else | |
1133 | rtm->rtm_table = RT_TABLE_COMPAT; | |
f3756b79 DM |
1134 | if (nla_put_u32(skb, RTA_TABLE, tb_id)) |
1135 | goto nla_put_failure; | |
1da177e4 LT |
1136 | rtm->rtm_type = type; |
1137 | rtm->rtm_flags = fi->fib_flags; | |
37e826c5 | 1138 | rtm->rtm_scope = fi->fib_scope; |
1da177e4 | 1139 | rtm->rtm_protocol = fi->fib_protocol; |
be403ea1 | 1140 | |
f3756b79 | 1141 | if (rtm->rtm_dst_len && |
930345ea | 1142 | nla_put_in_addr(skb, RTA_DST, dst)) |
f3756b79 DM |
1143 | goto nla_put_failure; |
1144 | if (fi->fib_priority && | |
1145 | nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority)) | |
1146 | goto nla_put_failure; | |
1da177e4 | 1147 | if (rtnetlink_put_metrics(skb, fi->fib_metrics) < 0) |
be403ea1 TG |
1148 | goto nla_put_failure; |
1149 | ||
f3756b79 | 1150 | if (fi->fib_prefsrc && |
930345ea | 1151 | nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc)) |
f3756b79 | 1152 | goto nla_put_failure; |
1da177e4 | 1153 | if (fi->fib_nhs == 1) { |
0eeb075f AG |
1154 | struct in_device *in_dev; |
1155 | ||
f3756b79 | 1156 | if (fi->fib_nh->nh_gw && |
930345ea | 1157 | nla_put_in_addr(skb, RTA_GATEWAY, fi->fib_nh->nh_gw)) |
f3756b79 DM |
1158 | goto nla_put_failure; |
1159 | if (fi->fib_nh->nh_oif && | |
1160 | nla_put_u32(skb, RTA_OIF, fi->fib_nh->nh_oif)) | |
1161 | goto nla_put_failure; | |
0eeb075f | 1162 | if (fi->fib_nh->nh_flags & RTNH_F_LINKDOWN) { |
96ac5cc9 | 1163 | in_dev = __in_dev_get_rtnl(fi->fib_nh->nh_dev); |
0eeb075f AG |
1164 | if (in_dev && |
1165 | IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev)) | |
1166 | rtm->rtm_flags |= RTNH_F_DEAD; | |
1167 | } | |
c7066f70 | 1168 | #ifdef CONFIG_IP_ROUTE_CLASSID |
f3756b79 DM |
1169 | if (fi->fib_nh[0].nh_tclassid && |
1170 | nla_put_u32(skb, RTA_FLOW, fi->fib_nh[0].nh_tclassid)) | |
1171 | goto nla_put_failure; | |
8265abc0 | 1172 | #endif |
571e7226 RP |
1173 | if (fi->fib_nh->nh_lwtstate) |
1174 | lwtunnel_fill_encap(skb, fi->fib_nh->nh_lwtstate); | |
1da177e4 LT |
1175 | } |
1176 | #ifdef CONFIG_IP_ROUTE_MULTIPATH | |
1177 | if (fi->fib_nhs > 1) { | |
be403ea1 TG |
1178 | struct rtnexthop *rtnh; |
1179 | struct nlattr *mp; | |
1180 | ||
1181 | mp = nla_nest_start(skb, RTA_MULTIPATH); | |
51456b29 | 1182 | if (!mp) |
be403ea1 | 1183 | goto nla_put_failure; |
1da177e4 LT |
1184 | |
1185 | for_nexthops(fi) { | |
0eeb075f AG |
1186 | struct in_device *in_dev; |
1187 | ||
be403ea1 | 1188 | rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh)); |
51456b29 | 1189 | if (!rtnh) |
be403ea1 TG |
1190 | goto nla_put_failure; |
1191 | ||
1192 | rtnh->rtnh_flags = nh->nh_flags & 0xFF; | |
0eeb075f | 1193 | if (nh->nh_flags & RTNH_F_LINKDOWN) { |
96ac5cc9 | 1194 | in_dev = __in_dev_get_rtnl(nh->nh_dev); |
0eeb075f AG |
1195 | if (in_dev && |
1196 | IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev)) | |
1197 | rtnh->rtnh_flags |= RTNH_F_DEAD; | |
1198 | } | |
be403ea1 TG |
1199 | rtnh->rtnh_hops = nh->nh_weight - 1; |
1200 | rtnh->rtnh_ifindex = nh->nh_oif; | |
1201 | ||
f3756b79 | 1202 | if (nh->nh_gw && |
930345ea | 1203 | nla_put_in_addr(skb, RTA_GATEWAY, nh->nh_gw)) |
f3756b79 | 1204 | goto nla_put_failure; |
c7066f70 | 1205 | #ifdef CONFIG_IP_ROUTE_CLASSID |
f3756b79 DM |
1206 | if (nh->nh_tclassid && |
1207 | nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid)) | |
1208 | goto nla_put_failure; | |
8265abc0 | 1209 | #endif |
571e7226 RP |
1210 | if (nh->nh_lwtstate) |
1211 | lwtunnel_fill_encap(skb, nh->nh_lwtstate); | |
be403ea1 TG |
1212 | /* length of rtnetlink header + attributes */ |
1213 | rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *) rtnh; | |
1da177e4 | 1214 | } endfor_nexthops(fi); |
be403ea1 TG |
1215 | |
1216 | nla_nest_end(skb, mp); | |
1da177e4 LT |
1217 | } |
1218 | #endif | |
053c095a JB |
1219 | nlmsg_end(skb, nlh); |
1220 | return 0; | |
1da177e4 | 1221 | |
be403ea1 | 1222 | nla_put_failure: |
26932566 PM |
1223 | nlmsg_cancel(skb, nlh); |
1224 | return -EMSGSIZE; | |
1da177e4 LT |
1225 | } |
1226 | ||
1da177e4 | 1227 | /* |
6a31d2a9 ED |
1228 | * Update FIB if: |
1229 | * - local address disappeared -> we must delete all the entries | |
1230 | * referring to it. | |
1231 | * - device went down -> we must shutdown all nexthops going via it. | |
1da177e4 | 1232 | */ |
4814bdbd | 1233 | int fib_sync_down_addr(struct net *net, __be32 local) |
1da177e4 LT |
1234 | { |
1235 | int ret = 0; | |
85326fa5 DL |
1236 | unsigned int hash = fib_laddr_hashfn(local); |
1237 | struct hlist_head *head = &fib_info_laddrhash[hash]; | |
85326fa5 | 1238 | struct fib_info *fi; |
1da177e4 | 1239 | |
51456b29 | 1240 | if (!fib_info_laddrhash || local == 0) |
85326fa5 | 1241 | return 0; |
1da177e4 | 1242 | |
b67bfe0d | 1243 | hlist_for_each_entry(fi, head, fib_lhash) { |
09ad9bc7 | 1244 | if (!net_eq(fi->fib_net, net)) |
4814bdbd | 1245 | continue; |
85326fa5 DL |
1246 | if (fi->fib_prefsrc == local) { |
1247 | fi->fib_flags |= RTNH_F_DEAD; | |
1248 | ret++; | |
1da177e4 LT |
1249 | } |
1250 | } | |
85326fa5 DL |
1251 | return ret; |
1252 | } | |
1253 | ||
8a3d0316 | 1254 | int fib_sync_down_dev(struct net_device *dev, unsigned long event) |
85326fa5 DL |
1255 | { |
1256 | int ret = 0; | |
1257 | int scope = RT_SCOPE_NOWHERE; | |
1258 | struct fib_info *prev_fi = NULL; | |
1259 | unsigned int hash = fib_devindex_hashfn(dev->ifindex); | |
1260 | struct hlist_head *head = &fib_info_devhash[hash]; | |
85326fa5 | 1261 | struct fib_nh *nh; |
1da177e4 | 1262 | |
8a3d0316 AG |
1263 | if (event == NETDEV_UNREGISTER || |
1264 | event == NETDEV_DOWN) | |
85326fa5 | 1265 | scope = -1; |
1da177e4 | 1266 | |
b67bfe0d | 1267 | hlist_for_each_entry(nh, head, nh_hash) { |
85326fa5 DL |
1268 | struct fib_info *fi = nh->nh_parent; |
1269 | int dead; | |
1da177e4 | 1270 | |
85326fa5 DL |
1271 | BUG_ON(!fi->fib_nhs); |
1272 | if (nh->nh_dev != dev || fi == prev_fi) | |
1273 | continue; | |
1274 | prev_fi = fi; | |
1275 | dead = 0; | |
1276 | change_nexthops(fi) { | |
6a31d2a9 | 1277 | if (nexthop_nh->nh_flags & RTNH_F_DEAD) |
85326fa5 | 1278 | dead++; |
71fceff0 DM |
1279 | else if (nexthop_nh->nh_dev == dev && |
1280 | nexthop_nh->nh_scope != scope) { | |
8a3d0316 AG |
1281 | switch (event) { |
1282 | case NETDEV_DOWN: | |
1283 | case NETDEV_UNREGISTER: | |
1284 | nexthop_nh->nh_flags |= RTNH_F_DEAD; | |
1285 | /* fall through */ | |
1286 | case NETDEV_CHANGE: | |
1287 | nexthop_nh->nh_flags |= RTNH_F_LINKDOWN; | |
1288 | break; | |
1289 | } | |
1da177e4 | 1290 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
85326fa5 | 1291 | spin_lock_bh(&fib_multipath_lock); |
71fceff0 DM |
1292 | fi->fib_power -= nexthop_nh->nh_power; |
1293 | nexthop_nh->nh_power = 0; | |
85326fa5 | 1294 | spin_unlock_bh(&fib_multipath_lock); |
1da177e4 | 1295 | #endif |
85326fa5 DL |
1296 | dead++; |
1297 | } | |
1da177e4 | 1298 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
8a3d0316 AG |
1299 | if (event == NETDEV_UNREGISTER && |
1300 | nexthop_nh->nh_dev == dev) { | |
85326fa5 DL |
1301 | dead = fi->fib_nhs; |
1302 | break; | |
1da177e4 | 1303 | } |
85326fa5 DL |
1304 | #endif |
1305 | } endfor_nexthops(fi) | |
1306 | if (dead == fi->fib_nhs) { | |
8a3d0316 AG |
1307 | switch (event) { |
1308 | case NETDEV_DOWN: | |
1309 | case NETDEV_UNREGISTER: | |
1310 | fi->fib_flags |= RTNH_F_DEAD; | |
1311 | /* fall through */ | |
1312 | case NETDEV_CHANGE: | |
1313 | fi->fib_flags |= RTNH_F_LINKDOWN; | |
1314 | break; | |
1315 | } | |
85326fa5 | 1316 | ret++; |
1da177e4 LT |
1317 | } |
1318 | } | |
1319 | ||
1320 | return ret; | |
1321 | } | |
1322 | ||
0c838ff1 | 1323 | /* Must be invoked inside of an RCU protected region. */ |
2392debc | 1324 | void fib_select_default(const struct flowi4 *flp, struct fib_result *res) |
0c838ff1 DM |
1325 | { |
1326 | struct fib_info *fi = NULL, *last_resort = NULL; | |
56315f9e | 1327 | struct hlist_head *fa_head = res->fa_head; |
0c838ff1 | 1328 | struct fib_table *tb = res->table; |
18a912e9 | 1329 | u8 slen = 32 - res->prefixlen; |
0c838ff1 | 1330 | int order = -1, last_idx = -1; |
2392debc JA |
1331 | struct fib_alias *fa, *fa1 = NULL; |
1332 | u32 last_prio = res->fi->fib_priority; | |
1333 | u8 last_tos = 0; | |
0c838ff1 | 1334 | |
56315f9e | 1335 | hlist_for_each_entry_rcu(fa, fa_head, fa_list) { |
0c838ff1 DM |
1336 | struct fib_info *next_fi = fa->fa_info; |
1337 | ||
18a912e9 JA |
1338 | if (fa->fa_slen != slen) |
1339 | continue; | |
2392debc JA |
1340 | if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos) |
1341 | continue; | |
18a912e9 JA |
1342 | if (fa->tb_id != tb->tb_id) |
1343 | continue; | |
2392debc JA |
1344 | if (next_fi->fib_priority > last_prio && |
1345 | fa->fa_tos == last_tos) { | |
1346 | if (last_tos) | |
1347 | continue; | |
1348 | break; | |
1349 | } | |
1350 | if (next_fi->fib_flags & RTNH_F_DEAD) | |
1351 | continue; | |
1352 | last_tos = fa->fa_tos; | |
1353 | last_prio = next_fi->fib_priority; | |
1354 | ||
37e826c5 | 1355 | if (next_fi->fib_scope != res->scope || |
0c838ff1 DM |
1356 | fa->fa_type != RTN_UNICAST) |
1357 | continue; | |
0c838ff1 DM |
1358 | if (!next_fi->fib_nh[0].nh_gw || |
1359 | next_fi->fib_nh[0].nh_scope != RT_SCOPE_LINK) | |
1360 | continue; | |
1361 | ||
1362 | fib_alias_accessed(fa); | |
1363 | ||
51456b29 | 1364 | if (!fi) { |
0c838ff1 DM |
1365 | if (next_fi != res->fi) |
1366 | break; | |
2392debc | 1367 | fa1 = fa; |
0c838ff1 | 1368 | } else if (!fib_detect_death(fi, order, &last_resort, |
2392debc | 1369 | &last_idx, fa1->fa_default)) { |
0c838ff1 | 1370 | fib_result_assign(res, fi); |
2392debc | 1371 | fa1->fa_default = order; |
0c838ff1 DM |
1372 | goto out; |
1373 | } | |
1374 | fi = next_fi; | |
1375 | order++; | |
1376 | } | |
1377 | ||
51456b29 | 1378 | if (order <= 0 || !fi) { |
2392debc JA |
1379 | if (fa1) |
1380 | fa1->fa_default = -1; | |
0c838ff1 DM |
1381 | goto out; |
1382 | } | |
1383 | ||
1384 | if (!fib_detect_death(fi, order, &last_resort, &last_idx, | |
2392debc | 1385 | fa1->fa_default)) { |
0c838ff1 | 1386 | fib_result_assign(res, fi); |
2392debc | 1387 | fa1->fa_default = order; |
0c838ff1 DM |
1388 | goto out; |
1389 | } | |
1390 | ||
1391 | if (last_idx >= 0) | |
1392 | fib_result_assign(res, last_resort); | |
2392debc | 1393 | fa1->fa_default = last_idx; |
0c838ff1 | 1394 | out: |
31d40937 | 1395 | return; |
0c838ff1 DM |
1396 | } |
1397 | ||
1da177e4 | 1398 | /* |
6a31d2a9 ED |
1399 | * Dead device goes up. We wake up dead nexthops. |
1400 | * It takes sense only on multipath routes. | |
1da177e4 | 1401 | */ |
8a3d0316 | 1402 | int fib_sync_up(struct net_device *dev, unsigned int nh_flags) |
1da177e4 LT |
1403 | { |
1404 | struct fib_info *prev_fi; | |
1405 | unsigned int hash; | |
1406 | struct hlist_head *head; | |
1da177e4 LT |
1407 | struct fib_nh *nh; |
1408 | int ret; | |
1409 | ||
6a31d2a9 | 1410 | if (!(dev->flags & IFF_UP)) |
1da177e4 LT |
1411 | return 0; |
1412 | ||
1413 | prev_fi = NULL; | |
1414 | hash = fib_devindex_hashfn(dev->ifindex); | |
1415 | head = &fib_info_devhash[hash]; | |
1416 | ret = 0; | |
1417 | ||
b67bfe0d | 1418 | hlist_for_each_entry(nh, head, nh_hash) { |
1da177e4 LT |
1419 | struct fib_info *fi = nh->nh_parent; |
1420 | int alive; | |
1421 | ||
1422 | BUG_ON(!fi->fib_nhs); | |
1423 | if (nh->nh_dev != dev || fi == prev_fi) | |
1424 | continue; | |
1425 | ||
1426 | prev_fi = fi; | |
1427 | alive = 0; | |
1428 | change_nexthops(fi) { | |
8a3d0316 | 1429 | if (!(nexthop_nh->nh_flags & nh_flags)) { |
1da177e4 LT |
1430 | alive++; |
1431 | continue; | |
1432 | } | |
51456b29 | 1433 | if (!nexthop_nh->nh_dev || |
6a31d2a9 | 1434 | !(nexthop_nh->nh_dev->flags & IFF_UP)) |
1da177e4 | 1435 | continue; |
71fceff0 DM |
1436 | if (nexthop_nh->nh_dev != dev || |
1437 | !__in_dev_get_rtnl(dev)) | |
1da177e4 LT |
1438 | continue; |
1439 | alive++; | |
8a3d0316 | 1440 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
1da177e4 | 1441 | spin_lock_bh(&fib_multipath_lock); |
71fceff0 | 1442 | nexthop_nh->nh_power = 0; |
8a3d0316 | 1443 | nexthop_nh->nh_flags &= ~nh_flags; |
1da177e4 | 1444 | spin_unlock_bh(&fib_multipath_lock); |
8a3d0316 AG |
1445 | #else |
1446 | nexthop_nh->nh_flags &= ~nh_flags; | |
1447 | #endif | |
1da177e4 LT |
1448 | } endfor_nexthops(fi) |
1449 | ||
1450 | if (alive > 0) { | |
8a3d0316 | 1451 | fi->fib_flags &= ~nh_flags; |
1da177e4 LT |
1452 | ret++; |
1453 | } | |
1454 | } | |
1455 | ||
1456 | return ret; | |
1457 | } | |
1458 | ||
8a3d0316 AG |
1459 | #ifdef CONFIG_IP_ROUTE_MULTIPATH |
1460 | ||
1da177e4 | 1461 | /* |
6a31d2a9 ED |
1462 | * The algorithm is suboptimal, but it provides really |
1463 | * fair weighted route distribution. | |
1da177e4 | 1464 | */ |
1b7fe593 | 1465 | void fib_select_multipath(struct fib_result *res) |
1da177e4 LT |
1466 | { |
1467 | struct fib_info *fi = res->fi; | |
0eeb075f | 1468 | struct in_device *in_dev; |
1da177e4 LT |
1469 | int w; |
1470 | ||
1471 | spin_lock_bh(&fib_multipath_lock); | |
1472 | if (fi->fib_power <= 0) { | |
1473 | int power = 0; | |
1474 | change_nexthops(fi) { | |
0eeb075f AG |
1475 | in_dev = __in_dev_get_rcu(nexthop_nh->nh_dev); |
1476 | if (nexthop_nh->nh_flags & RTNH_F_DEAD) | |
1477 | continue; | |
1478 | if (in_dev && | |
1479 | IN_DEV_IGNORE_ROUTES_WITH_LINKDOWN(in_dev) && | |
1480 | nexthop_nh->nh_flags & RTNH_F_LINKDOWN) | |
1481 | continue; | |
1482 | power += nexthop_nh->nh_weight; | |
1483 | nexthop_nh->nh_power = nexthop_nh->nh_weight; | |
1da177e4 LT |
1484 | } endfor_nexthops(fi); |
1485 | fi->fib_power = power; | |
1486 | if (power <= 0) { | |
1487 | spin_unlock_bh(&fib_multipath_lock); | |
1488 | /* Race condition: route has just become dead. */ | |
1489 | res->nh_sel = 0; | |
1490 | return; | |
1491 | } | |
1492 | } | |
1493 | ||
1494 | ||
1495 | /* w should be random number [0..fi->fib_power-1], | |
6a31d2a9 | 1496 | * it is pretty bad approximation. |
1da177e4 LT |
1497 | */ |
1498 | ||
1499 | w = jiffies % fi->fib_power; | |
1500 | ||
1501 | change_nexthops(fi) { | |
6a31d2a9 | 1502 | if (!(nexthop_nh->nh_flags & RTNH_F_DEAD) && |
71fceff0 | 1503 | nexthop_nh->nh_power) { |
6a31d2a9 ED |
1504 | w -= nexthop_nh->nh_power; |
1505 | if (w <= 0) { | |
71fceff0 | 1506 | nexthop_nh->nh_power--; |
1da177e4 LT |
1507 | fi->fib_power--; |
1508 | res->nh_sel = nhsel; | |
1509 | spin_unlock_bh(&fib_multipath_lock); | |
1510 | return; | |
1511 | } | |
1512 | } | |
1513 | } endfor_nexthops(fi); | |
1514 | ||
1515 | /* Race condition: route has just become dead. */ | |
1516 | res->nh_sel = 0; | |
1517 | spin_unlock_bh(&fib_multipath_lock); | |
1518 | } | |
1519 | #endif |