Commit | Line | Data |
---|---|---|
2fe58dfd SE |
1 | /* User-kernel network link */ |
2 | ||
baa06aeb SE |
3 | /* Each netlink device is actually a router, with its own IP address. |
4 | We do things like decreasing the TTL and recalculating the header | |
5 | checksum, generating ICMP, responding to pings, etc. */ | |
2fe58dfd SE |
6 | |
7 | /* This is where we have the anti-spoofing paranoia - before sending a | |
8 | packet to the kernel we check that the tunnel it came over could | |
9 | reasonably have produced it. */ | |
10 | ||
8689b3a9 | 11 | #include "secnet.h" |
2fe58dfd | 12 | #include "util.h" |
7138d0c5 | 13 | #include "ipaddr.h" |
9d3a4132 | 14 | #include "netlink.h" |
2fe58dfd | 15 | |
4efd681a SE |
16 | /* Generic IP checksum routine */ |
17 | static inline uint16_t ip_csum(uint8_t *iph,uint32_t count) | |
2fe58dfd | 18 | { |
4efd681a SE |
19 | register uint32_t sum=0; |
20 | ||
21 | while (count>1) { | |
22 | sum+=ntohs(*(uint16_t *)iph); | |
23 | iph+=2; | |
24 | count-=2; | |
25 | } | |
26 | if(count>0) | |
27 | sum+=*(uint8_t *)iph; | |
28 | while (sum>>16) | |
29 | sum=(sum&0xffff)+(sum>>16); | |
30 | return htons(~sum); | |
2fe58dfd SE |
31 | } |
32 | ||
4efd681a SE |
33 | #ifdef i386 |
34 | /* | |
35 | * This is a version of ip_compute_csum() optimized for IP headers, | |
36 | * which always checksum on 4 octet boundaries. | |
37 | * | |
38 | * By Jorge Cwik <jorge@laser.satlink.net>, adapted for linux by | |
39 | * Arnt Gulbrandsen. | |
40 | */ | |
41 | static inline uint16_t ip_fast_csum(uint8_t *iph, uint32_t ihl) { | |
42 | uint32_t sum; | |
43 | ||
44 | __asm__ __volatile__(" | |
45 | movl (%1), %0 | |
46 | subl $4, %2 | |
47 | jbe 2f | |
48 | addl 4(%1), %0 | |
49 | adcl 8(%1), %0 | |
50 | adcl 12(%1), %0 | |
51 | 1: adcl 16(%1), %0 | |
52 | lea 4(%1), %1 | |
53 | decl %2 | |
54 | jne 1b | |
55 | adcl $0, %0 | |
56 | movl %0, %2 | |
57 | shrl $16, %0 | |
58 | addw %w2, %w0 | |
59 | adcl $0, %0 | |
60 | notl %0 | |
61 | 2: | |
62 | " | |
63 | /* Since the input registers which are loaded with iph and ipl | |
64 | are modified, we must also specify them as outputs, or gcc | |
65 | will assume they contain their original values. */ | |
66 | : "=r" (sum), "=r" (iph), "=r" (ihl) | |
67 | : "1" (iph), "2" (ihl)); | |
68 | return sum; | |
69 | } | |
70 | #else | |
71 | static inline uint16_t ip_fast_csum(uint8_t *iph, uint32_t ihl) | |
2fe58dfd | 72 | { |
4efd681a SE |
73 | return ip_csum(iph,ihl*4); |
74 | } | |
75 | #endif | |
76 | ||
77 | struct iphdr { | |
78 | #if defined (WORDS_BIGENDIAN) | |
79 | uint8_t version:4, | |
80 | ihl:4; | |
81 | #else | |
82 | uint8_t ihl:4, | |
83 | version:4; | |
84 | #endif | |
85 | uint8_t tos; | |
86 | uint16_t tot_len; | |
87 | uint16_t id; | |
88 | uint16_t frag_off; | |
89 | uint8_t ttl; | |
90 | uint8_t protocol; | |
91 | uint16_t check; | |
92 | uint32_t saddr; | |
93 | uint32_t daddr; | |
94 | /* The options start here. */ | |
95 | }; | |
96 | ||
97 | struct icmphdr { | |
98 | struct iphdr iph; | |
99 | uint8_t type; | |
100 | uint8_t code; | |
101 | uint16_t check; | |
102 | union { | |
103 | uint32_t unused; | |
104 | struct { | |
105 | uint8_t pointer; | |
106 | uint8_t unused1; | |
107 | uint16_t unused2; | |
108 | } pprob; | |
109 | uint32_t gwaddr; | |
110 | struct { | |
111 | uint16_t id; | |
112 | uint16_t seq; | |
113 | } echo; | |
114 | } d; | |
115 | }; | |
116 | ||
70dc107b SE |
117 | static void netlink_packet_deliver(struct netlink *st, |
118 | struct netlink_client *client, | |
119 | struct buffer_if *buf); | |
4efd681a SE |
120 | |
121 | static struct icmphdr *netlink_icmp_tmpl(struct netlink *st, | |
122 | uint32_t dest,uint16_t len) | |
123 | { | |
124 | struct icmphdr *h; | |
125 | ||
126 | BUF_ALLOC(&st->icmp,"netlink_icmp_tmpl"); | |
127 | buffer_init(&st->icmp,st->max_start_pad); | |
128 | h=buf_append(&st->icmp,sizeof(*h)); | |
129 | ||
130 | h->iph.version=4; | |
131 | h->iph.ihl=5; | |
132 | h->iph.tos=0; | |
133 | h->iph.tot_len=htons(len+(h->iph.ihl*4)+8); | |
134 | h->iph.id=0; | |
135 | h->iph.frag_off=0; | |
136 | h->iph.ttl=255; | |
137 | h->iph.protocol=1; | |
138 | h->iph.saddr=htonl(st->secnet_address); | |
139 | h->iph.daddr=htonl(dest); | |
140 | h->iph.check=0; | |
141 | h->iph.check=ip_fast_csum((uint8_t *)&h->iph,h->iph.ihl); | |
142 | h->check=0; | |
143 | h->d.unused=0; | |
144 | ||
145 | return h; | |
146 | } | |
147 | ||
148 | /* Fill in the ICMP checksum field correctly */ | |
149 | static void netlink_icmp_csum(struct icmphdr *h) | |
150 | { | |
151 | uint32_t len; | |
152 | ||
153 | len=ntohs(h->iph.tot_len)-(4*h->iph.ihl); | |
154 | h->check=0; | |
155 | h->check=ip_csum(&h->type,len); | |
156 | } | |
157 | ||
158 | /* RFC1122: | |
159 | * An ICMP error message MUST NOT be sent as the result of | |
160 | * receiving: | |
161 | * | |
162 | * * an ICMP error message, or | |
163 | * | |
164 | * * a datagram destined to an IP broadcast or IP multicast | |
165 | * address, or | |
166 | * | |
167 | * * a datagram sent as a link-layer broadcast, or | |
168 | * | |
169 | * * a non-initial fragment, or | |
170 | * | |
171 | * * a datagram whose source address does not define a single | |
172 | * host -- e.g., a zero address, a loopback address, a | |
173 | * broadcast address, a multicast address, or a Class E | |
174 | * address. | |
175 | */ | |
176 | static bool_t netlink_icmp_may_reply(struct buffer_if *buf) | |
177 | { | |
178 | struct iphdr *iph; | |
179 | uint32_t source; | |
180 | ||
181 | iph=(struct iphdr *)buf->start; | |
182 | if (iph->protocol==1) return False; /* Overly-broad; we may reply to | |
183 | eg. icmp echo-request */ | |
184 | /* How do we spot broadcast destination addresses? */ | |
185 | if (ntohs(iph->frag_off)&0x1fff) return False; /* Non-initial fragment */ | |
186 | source=ntohl(iph->saddr); | |
187 | if (source==0) return False; | |
188 | if ((source&0xff000000)==0x7f000000) return False; | |
189 | /* How do we spot broadcast source addresses? */ | |
190 | if ((source&0xf0000000)==0xe0000000) return False; /* Multicast */ | |
191 | if ((source&0xf0000000)==0xf0000000) return False; /* Class E */ | |
192 | return True; | |
193 | } | |
194 | ||
195 | /* How much of the original IP packet do we include in its ICMP | |
196 | response? The header plus up to 64 bits. */ | |
197 | static uint16_t netlink_icmp_reply_len(struct buffer_if *buf) | |
198 | { | |
199 | struct iphdr *iph=(struct iphdr *)buf->start; | |
200 | uint16_t hlen,plen; | |
201 | ||
202 | hlen=iph->ihl*4; | |
203 | /* We include the first 8 bytes of the packet data, provided they exist */ | |
204 | hlen+=8; | |
205 | plen=ntohs(iph->tot_len); | |
206 | return (hlen>plen?plen:hlen); | |
207 | } | |
208 | ||
70dc107b SE |
209 | /* client indicates where the packet we're constructing a response to |
210 | comes from. NULL indicates the host. */ | |
4efd681a | 211 | static void netlink_icmp_simple(struct netlink *st, struct buffer_if *buf, |
70dc107b | 212 | struct netlink_client *client, |
4efd681a SE |
213 | uint8_t type, uint8_t code) |
214 | { | |
215 | struct iphdr *iph=(struct iphdr *)buf->start; | |
216 | struct icmphdr *h; | |
217 | uint16_t len; | |
218 | ||
219 | if (netlink_icmp_may_reply(buf)) { | |
220 | len=netlink_icmp_reply_len(buf); | |
221 | h=netlink_icmp_tmpl(st,ntohl(iph->saddr),len); | |
222 | h->type=type; h->code=code; | |
223 | memcpy(buf_append(&st->icmp,len),buf->start,len); | |
224 | netlink_icmp_csum(h); | |
70dc107b | 225 | netlink_packet_deliver(st,NULL,&st->icmp); |
4efd681a SE |
226 | BUF_ASSERT_FREE(&st->icmp); |
227 | } | |
228 | } | |
229 | ||
230 | /* | |
231 | * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the | |
232 | * checksum. | |
233 | * | |
234 | * Is the datagram acceptable? | |
235 | * | |
236 | * 1. Length at least the size of an ip header | |
237 | * 2. Version of 4 | |
238 | * 3. Checksums correctly. | |
239 | * 4. Doesn't have a bogus length | |
240 | */ | |
241 | static bool_t netlink_check(struct netlink *st, struct buffer_if *buf) | |
242 | { | |
243 | struct iphdr *iph=(struct iphdr *)buf->start; | |
244 | uint32_t len; | |
245 | ||
9d3a4132 SE |
246 | if (iph->ihl < 5 || iph->version != 4) return False; |
247 | if (buf->size < iph->ihl*4) return False; | |
248 | if (ip_fast_csum((uint8_t *)iph, iph->ihl)!=0) return False; | |
4efd681a SE |
249 | len=ntohs(iph->tot_len); |
250 | /* There should be no padding */ | |
9d3a4132 | 251 | if (buf->size!=len || len<(iph->ihl<<2)) return False; |
4efd681a SE |
252 | /* XXX check that there's no source route specified */ |
253 | return True; | |
254 | } | |
255 | ||
70dc107b SE |
256 | /* Deliver a packet. "client" points to the _origin_ of the packet, not |
257 | its destination. (May be used when sending ICMP response - avoid | |
258 | asymmetric routing.) */ | |
259 | static void netlink_packet_deliver(struct netlink *st, | |
260 | struct netlink_client *client, | |
261 | struct buffer_if *buf) | |
4efd681a SE |
262 | { |
263 | struct iphdr *iph=(struct iphdr *)buf->start; | |
264 | uint32_t dest=ntohl(iph->daddr); | |
70dc107b SE |
265 | uint32_t source=ntohl(iph->saddr); |
266 | uint32_t best_quality; | |
267 | int best_match; | |
268 | int i; | |
2fe58dfd | 269 | |
4efd681a | 270 | BUF_ASSERT_USED(buf); |
2fe58dfd | 271 | |
4efd681a SE |
272 | if (dest==st->secnet_address) { |
273 | Message(M_ERROR,"%s: trying to deliver a packet to myself!\n"); | |
274 | BUF_FREE(buf); | |
2fe58dfd SE |
275 | return; |
276 | } | |
4efd681a | 277 | |
3454dce4 SE |
278 | /* XXX we're going to need an extra value 'allow_route' for the |
279 | source of the packet. It's always True for packets from the | |
280 | host. For packets from tunnels, we consult the client | |
281 | options. If !allow_route and the destination is a tunnel that | |
282 | also doesn't allow routing, we must reject the packet with an | |
283 | 'administratively prohibited' or something similar ICMP. */ | |
70dc107b SE |
284 | if (!client) { |
285 | /* Origin of packet is host or secnet. Might be for a tunnel. */ | |
286 | best_quality=0; | |
287 | best_match=-1; | |
288 | for (i=0; i<st->n_routes; i++) { | |
9d3a4132 | 289 | if (st->routes[i].up && subnet_match(&st->routes[i].net,dest)) { |
70dc107b SE |
290 | if (st->routes[i].c->link_quality>best_quality |
291 | || best_quality==0) { | |
292 | best_quality=st->routes[i].c->link_quality; | |
293 | best_match=i; | |
294 | /* If quality isn't perfect we may wish to | |
295 | consider kicking the tunnel with a 0-length | |
296 | packet to prompt it to perform a key setup. | |
297 | Then it'll eventually decide it's up or | |
298 | down. */ | |
299 | /* If quality is perfect we don't need to search | |
300 | any more. */ | |
301 | if (best_quality>=MAXIMUM_LINK_QUALITY) break; | |
302 | } | |
303 | } | |
304 | } | |
305 | if (best_match==-1) { | |
306 | /* Not going down a tunnel. Might be for the host. | |
307 | XXX think about this - only situation should be if we're | |
308 | sending ICMP. */ | |
309 | if (source!=st->secnet_address) { | |
310 | Message(M_ERROR,"netlink_packet_deliver: outgoing packet " | |
311 | "from host that won't fit down any of our tunnels!\n"); | |
9d3a4132 SE |
312 | /* XXX I think this could also occur if a soft tunnel just |
313 | went down, but still had packets queued in the kernel. */ | |
70dc107b SE |
314 | BUF_FREE(buf); |
315 | } else { | |
316 | st->deliver_to_host(st->dst,NULL,buf); | |
317 | BUF_ASSERT_FREE(buf); | |
318 | } | |
319 | } else { | |
320 | if (best_quality>0) { | |
321 | st->routes[best_match].c->deliver( | |
322 | st->routes[best_match].c->dst, | |
323 | st->routes[best_match].c, buf); | |
4efd681a SE |
324 | BUF_ASSERT_FREE(buf); |
325 | } else { | |
326 | /* Generate ICMP destination unreachable */ | |
70dc107b | 327 | netlink_icmp_simple(st,buf,client,3,0); /* client==NULL */ |
4efd681a SE |
328 | BUF_FREE(buf); |
329 | } | |
70dc107b SE |
330 | } |
331 | } else { /* client is set */ | |
332 | /* We know the origin is a tunnel - packet must be for the host */ | |
3454dce4 SE |
333 | /* XXX THIS IS NOT NECESSARILY TRUE, AND NEEDS FIXING */ |
334 | /* THIS FUNCTION MUST JUST DELIVER THE PACKET: IT MUST ASSUME | |
335 | THE PACKET HAS ALREADY BEEN CHECKED */ | |
70dc107b SE |
336 | if (subnet_matches_list(&st->networks,dest)) { |
337 | st->deliver_to_host(st->dst,NULL,buf); | |
338 | BUF_ASSERT_FREE(buf); | |
339 | } else { | |
340 | Message(M_ERROR,"%s: packet from tunnel %s can't be delivered " | |
341 | "to the host\n",st->name,client->name); | |
342 | netlink_icmp_simple(st,buf,client,3,0); | |
343 | BUF_FREE(buf); | |
2fe58dfd SE |
344 | } |
345 | } | |
70dc107b | 346 | BUF_ASSERT_FREE(buf); |
4efd681a SE |
347 | } |
348 | ||
70dc107b SE |
349 | static void netlink_packet_forward(struct netlink *st, |
350 | struct netlink_client *client, | |
351 | struct buffer_if *buf) | |
4efd681a SE |
352 | { |
353 | struct iphdr *iph=(struct iphdr *)buf->start; | |
354 | ||
355 | BUF_ASSERT_USED(buf); | |
356 | ||
357 | /* Packet has already been checked */ | |
358 | if (iph->ttl<=1) { | |
359 | /* Generate ICMP time exceeded */ | |
70dc107b | 360 | netlink_icmp_simple(st,buf,client,11,0); |
4efd681a SE |
361 | BUF_FREE(buf); |
362 | return; | |
363 | } | |
364 | iph->ttl--; | |
365 | iph->check=0; | |
366 | iph->check=ip_fast_csum((uint8_t *)iph,iph->ihl); | |
367 | ||
70dc107b | 368 | netlink_packet_deliver(st,client,buf); |
4efd681a SE |
369 | BUF_ASSERT_FREE(buf); |
370 | } | |
371 | ||
9d3a4132 | 372 | /* Deal with packets addressed explicitly to us */ |
70dc107b SE |
373 | static void netlink_packet_local(struct netlink *st, |
374 | struct netlink_client *client, | |
375 | struct buffer_if *buf) | |
4efd681a SE |
376 | { |
377 | struct icmphdr *h; | |
378 | ||
379 | h=(struct icmphdr *)buf->start; | |
380 | ||
381 | if ((ntohs(h->iph.frag_off)&0xbfff)!=0) { | |
9d3a4132 SE |
382 | Message(M_WARNING,"%s: fragmented packet addressed to secnet; " |
383 | "ignoring it\n",st->name); | |
4efd681a SE |
384 | BUF_FREE(buf); |
385 | return; | |
386 | } | |
387 | ||
388 | if (h->iph.protocol==1) { | |
389 | /* It's ICMP */ | |
390 | if (h->type==8 && h->code==0) { | |
391 | /* ICMP echo-request. Special case: we re-use the buffer | |
392 | to construct the reply. */ | |
393 | h->type=0; | |
394 | h->iph.daddr=h->iph.saddr; | |
395 | h->iph.saddr=htonl(st->secnet_address); | |
396 | h->iph.ttl=255; /* Be nice and bump it up again... */ | |
397 | h->iph.check=0; | |
398 | h->iph.check=ip_fast_csum((uint8_t *)h,h->iph.ihl); | |
399 | netlink_icmp_csum(h); | |
70dc107b | 400 | netlink_packet_deliver(st,NULL,buf); |
4efd681a SE |
401 | return; |
402 | } | |
403 | Message(M_WARNING,"%s: unknown incoming ICMP\n",st->name); | |
404 | } else { | |
405 | /* Send ICMP protocol unreachable */ | |
70dc107b | 406 | netlink_icmp_simple(st,buf,client,3,2); |
4efd681a SE |
407 | BUF_FREE(buf); |
408 | return; | |
409 | } | |
410 | ||
411 | BUF_FREE(buf); | |
412 | } | |
413 | ||
9d3a4132 SE |
414 | /* If cid==NULL packet is from host, otherwise cid specifies which tunnel |
415 | it came from. */ | |
416 | static void netlink_incoming(void *sst, void *cid, struct buffer_if *buf) | |
4efd681a SE |
417 | { |
418 | struct netlink *st=sst; | |
9d3a4132 | 419 | struct netlink_client *client=cid; |
4efd681a SE |
420 | uint32_t source,dest; |
421 | struct iphdr *iph; | |
422 | ||
423 | BUF_ASSERT_USED(buf); | |
424 | if (!netlink_check(st,buf)) { | |
9d3a4132 SE |
425 | Message(M_WARNING,"%s: bad IP packet from %s\n", |
426 | st->name,client?client->name:"host"); | |
4efd681a SE |
427 | BUF_FREE(buf); |
428 | return; | |
429 | } | |
430 | iph=(struct iphdr *)buf->start; | |
431 | ||
432 | source=ntohl(iph->saddr); | |
433 | dest=ntohl(iph->daddr); | |
434 | ||
9d3a4132 SE |
435 | /* Check source */ |
436 | if (client) { | |
437 | /* Check that the packet source is in 'nets' and its destination is | |
438 | in st->networks */ | |
439 | if (!subnet_matches_list(client->networks,source)) { | |
440 | string_t s,d; | |
441 | s=ipaddr_to_string(source); | |
442 | d=ipaddr_to_string(dest); | |
443 | Message(M_WARNING,"%s: packet from tunnel %s with bad " | |
444 | "source address (s=%s,d=%s)\n",st->name,client->name,s,d); | |
445 | free(s); free(d); | |
446 | BUF_FREE(buf); | |
447 | return; | |
448 | } | |
449 | } else { | |
450 | if (!subnet_matches_list(&st->networks,source)) { | |
451 | string_t s,d; | |
452 | s=ipaddr_to_string(source); | |
453 | d=ipaddr_to_string(dest); | |
454 | Message(M_WARNING,"%s: outgoing packet with bad source address " | |
455 | "(s=%s,d=%s)\n",st->name,s,d); | |
456 | free(s); free(d); | |
457 | BUF_FREE(buf); | |
458 | return; | |
459 | } | |
4efd681a | 460 | } |
9d3a4132 SE |
461 | /* (st->secnet_address needs checking before matching destination |
462 | addresses) */ | |
2fe58dfd | 463 | if (dest==st->secnet_address) { |
9d3a4132 | 464 | netlink_packet_local(st,client,buf); |
4efd681a | 465 | BUF_ASSERT_FREE(buf); |
2fe58dfd SE |
466 | return; |
467 | } | |
9d3a4132 | 468 | if (client) { |
3454dce4 | 469 | /* Check for free routing */ |
9d3a4132 SE |
470 | if (!subnet_matches_list(&st->networks,dest)) { |
471 | string_t s,d; | |
472 | s=ipaddr_to_string(source); | |
473 | d=ipaddr_to_string(dest); | |
474 | Message(M_WARNING,"%s: incoming packet from tunnel %s " | |
475 | "with bad destination address " | |
476 | "(s=%s,d=%s)\n",st->name,client->name,s,d); | |
477 | free(s); free(d); | |
478 | BUF_FREE(buf); | |
479 | return; | |
480 | } | |
2fe58dfd | 481 | } |
70dc107b | 482 | netlink_packet_forward(st,client,buf); |
4efd681a SE |
483 | BUF_ASSERT_FREE(buf); |
484 | } | |
485 | ||
9d3a4132 SE |
486 | static void netlink_set_softlinks(struct netlink *st, struct netlink_client *c, |
487 | bool_t up) | |
4efd681a | 488 | { |
9d3a4132 | 489 | uint32_t i; |
4efd681a | 490 | |
9d3a4132 SE |
491 | if (!st->routes) return; /* Table has not yet been created */ |
492 | for (i=0; i<st->n_routes; i++) { | |
493 | if (!st->routes[i].hard && st->routes[i].c==c) { | |
494 | st->routes[i].up=up; | |
495 | st->set_route(st->dst,&st->routes[i]); | |
496 | } | |
4efd681a | 497 | } |
4efd681a SE |
498 | } |
499 | ||
70dc107b | 500 | static void netlink_set_quality(void *sst, void *cid, uint32_t quality) |
4efd681a | 501 | { |
9d3a4132 | 502 | struct netlink *st=sst; |
4efd681a SE |
503 | struct netlink_client *c=cid; |
504 | ||
70dc107b | 505 | c->link_quality=quality; |
9d3a4132 SE |
506 | if (c->link_quality==LINK_QUALITY_DOWN) { |
507 | netlink_set_softlinks(st,c,False); | |
508 | } else { | |
509 | netlink_set_softlinks(st,c,True); | |
510 | } | |
4efd681a SE |
511 | } |
512 | ||
513 | static void *netlink_regnets(void *sst, struct subnet_list *nets, | |
514 | netlink_deliver_fn *deliver, void *dst, | |
515 | uint32_t max_start_pad, uint32_t max_end_pad, | |
3454dce4 | 516 | uint32_t options, string_t client_name) |
4efd681a SE |
517 | { |
518 | struct netlink *st=sst; | |
519 | struct netlink_client *c; | |
520 | ||
521 | Message(M_DEBUG_CONFIG,"netlink_regnets: request for %d networks, " | |
522 | "max_start_pad=%d, max_end_pad=%d\n", | |
523 | nets->entries,max_start_pad,max_end_pad); | |
524 | ||
3454dce4 | 525 | if ((options&NETLINK_OPTION_SOFTROUTE) && !st->set_route) { |
9d3a4132 SE |
526 | Message(M_ERROR,"%s: this netlink device does not support " |
527 | "soft routes.\n"); | |
528 | return NULL; | |
529 | } | |
530 | ||
3454dce4 | 531 | if (options&NETLINK_OPTION_SOFTROUTE) { |
9d3a4132 | 532 | /* XXX for now we assume that soft routes require root privilege; |
3454dce4 | 533 | this may not always be true. The device driver can tell us. */ |
9d3a4132 SE |
534 | require_root_privileges=True; |
535 | require_root_privileges_explanation="netlink: soft routes"; | |
536 | } | |
537 | ||
9d3a4132 SE |
538 | /* Check that nets do not intersect st->exclude_remote_networks; |
539 | refuse to register if they do. */ | |
baa06aeb SE |
540 | if (subnet_lists_intersect(&st->exclude_remote_networks,nets)) { |
541 | Message(M_ERROR,"%s: site %s specifies networks that " | |
542 | "intersect with the explicitly excluded remote networks\n", | |
543 | st->name,client_name); | |
544 | return False; | |
545 | } | |
546 | ||
4efd681a SE |
547 | c=safe_malloc(sizeof(*c),"netlink_regnets"); |
548 | c->networks=nets; | |
549 | c->deliver=deliver; | |
550 | c->dst=dst; | |
551 | c->name=client_name; /* XXX copy it? */ | |
3454dce4 | 552 | c->options=options; |
70dc107b | 553 | c->link_quality=LINK_QUALITY_DOWN; |
4efd681a SE |
554 | c->next=st->clients; |
555 | st->clients=c; | |
556 | if (max_start_pad > st->max_start_pad) st->max_start_pad=max_start_pad; | |
557 | if (max_end_pad > st->max_end_pad) st->max_end_pad=max_end_pad; | |
70dc107b | 558 | st->n_routes+=nets->entries; |
4efd681a SE |
559 | |
560 | return c; | |
561 | } | |
562 | ||
9d3a4132 SE |
563 | static void netlink_dump_routes(struct netlink *st) |
564 | { | |
565 | int i; | |
566 | string_t net; | |
567 | ||
568 | Message(M_INFO,"%s: routing table:\n",st->name); | |
569 | for (i=0; i<st->n_routes; i++) { | |
570 | net=subnet_to_string(&st->routes[i].net); | |
3454dce4 SE |
571 | Message(M_INFO,"%s -> tunnel %s (%s,%s route,%s)\n",net, |
572 | st->routes[i].c->name, | |
9d3a4132 | 573 | st->routes[i].hard?"hard":"soft", |
3454dce4 | 574 | st->routes[i].allow_route?"free":"restricted", |
9d3a4132 SE |
575 | st->routes[i].up?"up":"down"); |
576 | free(net); | |
577 | } | |
578 | Message(M_INFO,"%s/32 -> netlink \"%s\"\n", | |
579 | ipaddr_to_string(st->secnet_address),st->name); | |
580 | for (i=0; i<st->networks.entries; i++) { | |
581 | net=subnet_to_string(&st->networks.list[i]); | |
582 | Message(M_INFO,"%s -> host\n",net); | |
583 | free(net); | |
584 | } | |
585 | } | |
586 | ||
70dc107b SE |
587 | static int netlink_compare_route_specificity(const void *ap, const void *bp) |
588 | { | |
589 | const struct netlink_route *a=ap; | |
590 | const struct netlink_route *b=bp; | |
591 | ||
592 | if (a->net.len==b->net.len) return 0; | |
593 | if (a->net.len<b->net.len) return 1; | |
594 | return -1; | |
595 | } | |
596 | ||
597 | static void netlink_phase_hook(void *sst, uint32_t new_phase) | |
598 | { | |
599 | struct netlink *st=sst; | |
600 | struct netlink_client *c; | |
601 | uint32_t i,j; | |
602 | ||
603 | /* All the networks serviced by the various tunnels should now | |
604 | * have been registered. We build a routing table by sorting the | |
605 | * routes into most-specific-first order. */ | |
606 | st->routes=safe_malloc(st->n_routes*sizeof(*st->routes), | |
607 | "netlink_phase_hook"); | |
608 | /* Fill the table */ | |
609 | i=0; | |
610 | for (c=st->clients; c; c=c->next) { | |
611 | for (j=0; j<c->networks->entries; j++) { | |
612 | st->routes[i].net=c->networks->list[j]; | |
613 | st->routes[i].c=c; | |
3454dce4 SE |
614 | /* Hard routes are always up; |
615 | soft routes default to down */ | |
616 | st->routes[i].up=c->options&NETLINK_OPTION_SOFTROUTE?False:True; | |
9d3a4132 | 617 | st->routes[i].kup=False; |
3454dce4 SE |
618 | st->routes[i].hard=c->options&NETLINK_OPTION_SOFTROUTE?False:True; |
619 | st->routes[i].allow_route=c->options&NETLINK_OPTION_ALLOW_ROUTE? | |
620 | True:False; | |
70dc107b SE |
621 | i++; |
622 | } | |
623 | } | |
624 | /* ASSERT i==st->n_routes */ | |
625 | if (i!=st->n_routes) { | |
626 | fatal("netlink: route count error: expected %d got %d\n", | |
627 | st->n_routes,i); | |
628 | } | |
629 | /* Sort the table in descending order of specificity */ | |
630 | qsort(st->routes,st->n_routes,sizeof(*st->routes), | |
631 | netlink_compare_route_specificity); | |
9d3a4132 SE |
632 | |
633 | netlink_dump_routes(st); | |
70dc107b SE |
634 | } |
635 | ||
9d3a4132 SE |
636 | netlink_deliver_fn *netlink_init(struct netlink *st, |
637 | void *dst, struct cloc loc, | |
638 | dict_t *dict, string_t description, | |
639 | netlink_route_fn *set_route, | |
640 | netlink_deliver_fn *to_host) | |
4efd681a SE |
641 | { |
642 | st->dst=dst; | |
643 | st->cl.description=description; | |
644 | st->cl.type=CL_NETLINK; | |
645 | st->cl.apply=NULL; | |
646 | st->cl.interface=&st->ops; | |
647 | st->ops.st=st; | |
648 | st->ops.regnets=netlink_regnets; | |
9d3a4132 | 649 | st->ops.deliver=netlink_incoming; |
70dc107b | 650 | st->ops.set_quality=netlink_set_quality; |
4efd681a SE |
651 | st->max_start_pad=0; |
652 | st->max_end_pad=0; | |
653 | st->clients=NULL; | |
9d3a4132 | 654 | st->set_route=set_route; |
4efd681a SE |
655 | st->deliver_to_host=to_host; |
656 | ||
657 | st->name=dict_read_string(dict,"name",False,"netlink",loc); | |
658 | if (!st->name) st->name=description; | |
659 | dict_read_subnet_list(dict, "networks", True, "netlink", loc, | |
660 | &st->networks); | |
baa06aeb SE |
661 | dict_read_subnet_list(dict, "exclude-remote-networks", False, "netlink", |
662 | loc, &st->exclude_remote_networks); | |
b2a56f7c SE |
663 | /* secnet-address does not have to be in local-networks; |
664 | however, it should be advertised in the 'sites' file for the | |
baa06aeb | 665 | local site. */ |
4efd681a SE |
666 | st->secnet_address=string_to_ipaddr( |
667 | dict_find_item(dict,"secnet-address", True, "netlink", loc),"netlink"); | |
4efd681a SE |
668 | st->mtu=dict_read_number(dict, "mtu", False, "netlink", loc, DEFAULT_MTU); |
669 | buffer_new(&st->icmp,ICMP_BUFSIZE); | |
70dc107b SE |
670 | st->n_routes=0; |
671 | st->routes=NULL; | |
672 | ||
673 | add_hook(PHASE_SETUP,netlink_phase_hook,st); | |
4efd681a | 674 | |
9d3a4132 | 675 | return netlink_incoming; |
2fe58dfd SE |
676 | } |
677 | ||
9d3a4132 | 678 | /* No connection to the kernel at all... */ |
2fe58dfd | 679 | |
9d3a4132 | 680 | struct null { |
4efd681a | 681 | struct netlink nl; |
4efd681a | 682 | }; |
2fe58dfd | 683 | |
9d3a4132 | 684 | static bool_t null_set_route(void *sst, struct netlink_route *route) |
4efd681a | 685 | { |
9d3a4132 SE |
686 | struct null *st=sst; |
687 | string_t t; | |
4efd681a | 688 | |
9d3a4132 SE |
689 | if (route->up!=route->kup) { |
690 | t=subnet_to_string(&route->net); | |
691 | Message(M_INFO,"%s: setting route %s to state %s\n",st->nl.name, | |
692 | t, route->up?"up":"down"); | |
693 | free(t); | |
694 | route->kup=route->up; | |
695 | return True; | |
2fe58dfd | 696 | } |
9d3a4132 | 697 | return False; |
2fe58dfd | 698 | } |
9d3a4132 | 699 | |
2fe58dfd SE |
700 | static void null_deliver(void *sst, void *cid, struct buffer_if *buf) |
701 | { | |
702 | return; | |
703 | } | |
704 | ||
705 | static list_t *null_apply(closure_t *self, struct cloc loc, dict_t *context, | |
706 | list_t *args) | |
707 | { | |
708 | struct null *st; | |
4efd681a SE |
709 | item_t *item; |
710 | dict_t *dict; | |
2fe58dfd | 711 | |
4efd681a | 712 | st=safe_malloc(sizeof(*st),"null_apply"); |
2fe58dfd | 713 | |
4efd681a SE |
714 | item=list_elem(args,0); |
715 | if (!item || item->type!=t_dict) | |
716 | cfgfatal(loc,"null-netlink","parameter must be a dictionary\n"); | |
717 | ||
718 | dict=item->data.dict; | |
719 | ||
9d3a4132 SE |
720 | netlink_init(&st->nl,st,loc,dict,"null-netlink",null_set_route, |
721 | null_deliver); | |
4efd681a SE |
722 | |
723 | return new_closure(&st->nl.cl); | |
2fe58dfd SE |
724 | } |
725 | ||
726 | init_module netlink_module; | |
727 | void netlink_module(dict_t *dict) | |
728 | { | |
4efd681a | 729 | add_closure(dict,"null-netlink",null_apply); |
2fe58dfd | 730 | } |