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" |
042a8da9 | 15 | #include "process.h" |
2fe58dfd | 16 | |
4efd681a SE |
17 | /* Generic IP checksum routine */ |
18 | static inline uint16_t ip_csum(uint8_t *iph,uint32_t count) | |
2fe58dfd | 19 | { |
4efd681a SE |
20 | register uint32_t sum=0; |
21 | ||
22 | while (count>1) { | |
23 | sum+=ntohs(*(uint16_t *)iph); | |
24 | iph+=2; | |
25 | count-=2; | |
26 | } | |
27 | if(count>0) | |
28 | sum+=*(uint8_t *)iph; | |
29 | while (sum>>16) | |
30 | sum=(sum&0xffff)+(sum>>16); | |
31 | return htons(~sum); | |
2fe58dfd SE |
32 | } |
33 | ||
4efd681a SE |
34 | #ifdef i386 |
35 | /* | |
36 | * This is a version of ip_compute_csum() optimized for IP headers, | |
37 | * which always checksum on 4 octet boundaries. | |
38 | * | |
39 | * By Jorge Cwik <jorge@laser.satlink.net>, adapted for linux by | |
40 | * Arnt Gulbrandsen. | |
41 | */ | |
42 | static inline uint16_t ip_fast_csum(uint8_t *iph, uint32_t ihl) { | |
43 | uint32_t sum; | |
44 | ||
45 | __asm__ __volatile__(" | |
46 | movl (%1), %0 | |
47 | subl $4, %2 | |
48 | jbe 2f | |
49 | addl 4(%1), %0 | |
50 | adcl 8(%1), %0 | |
51 | adcl 12(%1), %0 | |
52 | 1: adcl 16(%1), %0 | |
53 | lea 4(%1), %1 | |
54 | decl %2 | |
55 | jne 1b | |
56 | adcl $0, %0 | |
57 | movl %0, %2 | |
58 | shrl $16, %0 | |
59 | addw %w2, %w0 | |
60 | adcl $0, %0 | |
61 | notl %0 | |
62 | 2: | |
63 | " | |
64 | /* Since the input registers which are loaded with iph and ipl | |
65 | are modified, we must also specify them as outputs, or gcc | |
66 | will assume they contain their original values. */ | |
67 | : "=r" (sum), "=r" (iph), "=r" (ihl) | |
68 | : "1" (iph), "2" (ihl)); | |
69 | return sum; | |
70 | } | |
71 | #else | |
72 | static inline uint16_t ip_fast_csum(uint8_t *iph, uint32_t ihl) | |
2fe58dfd | 73 | { |
4efd681a SE |
74 | return ip_csum(iph,ihl*4); |
75 | } | |
76 | #endif | |
77 | ||
78 | struct iphdr { | |
79 | #if defined (WORDS_BIGENDIAN) | |
80 | uint8_t version:4, | |
81 | ihl:4; | |
82 | #else | |
83 | uint8_t ihl:4, | |
84 | version:4; | |
85 | #endif | |
86 | uint8_t tos; | |
87 | uint16_t tot_len; | |
88 | uint16_t id; | |
89 | uint16_t frag_off; | |
90 | uint8_t ttl; | |
91 | uint8_t protocol; | |
92 | uint16_t check; | |
93 | uint32_t saddr; | |
94 | uint32_t daddr; | |
95 | /* The options start here. */ | |
96 | }; | |
97 | ||
98 | struct icmphdr { | |
99 | struct iphdr iph; | |
100 | uint8_t type; | |
101 | uint8_t code; | |
102 | uint16_t check; | |
103 | union { | |
104 | uint32_t unused; | |
105 | struct { | |
106 | uint8_t pointer; | |
107 | uint8_t unused1; | |
108 | uint16_t unused2; | |
109 | } pprob; | |
110 | uint32_t gwaddr; | |
111 | struct { | |
112 | uint16_t id; | |
113 | uint16_t seq; | |
114 | } echo; | |
115 | } d; | |
116 | }; | |
117 | ||
70dc107b SE |
118 | static void netlink_packet_deliver(struct netlink *st, |
119 | struct netlink_client *client, | |
120 | struct buffer_if *buf); | |
4efd681a SE |
121 | |
122 | static struct icmphdr *netlink_icmp_tmpl(struct netlink *st, | |
123 | uint32_t dest,uint16_t len) | |
124 | { | |
125 | struct icmphdr *h; | |
126 | ||
127 | BUF_ALLOC(&st->icmp,"netlink_icmp_tmpl"); | |
128 | buffer_init(&st->icmp,st->max_start_pad); | |
129 | h=buf_append(&st->icmp,sizeof(*h)); | |
130 | ||
131 | h->iph.version=4; | |
132 | h->iph.ihl=5; | |
133 | h->iph.tos=0; | |
134 | h->iph.tot_len=htons(len+(h->iph.ihl*4)+8); | |
135 | h->iph.id=0; | |
136 | h->iph.frag_off=0; | |
137 | h->iph.ttl=255; | |
138 | h->iph.protocol=1; | |
139 | h->iph.saddr=htonl(st->secnet_address); | |
140 | h->iph.daddr=htonl(dest); | |
141 | h->iph.check=0; | |
142 | h->iph.check=ip_fast_csum((uint8_t *)&h->iph,h->iph.ihl); | |
143 | h->check=0; | |
144 | h->d.unused=0; | |
145 | ||
146 | return h; | |
147 | } | |
148 | ||
149 | /* Fill in the ICMP checksum field correctly */ | |
150 | static void netlink_icmp_csum(struct icmphdr *h) | |
151 | { | |
152 | uint32_t len; | |
153 | ||
154 | len=ntohs(h->iph.tot_len)-(4*h->iph.ihl); | |
155 | h->check=0; | |
156 | h->check=ip_csum(&h->type,len); | |
157 | } | |
158 | ||
159 | /* RFC1122: | |
160 | * An ICMP error message MUST NOT be sent as the result of | |
161 | * receiving: | |
162 | * | |
163 | * * an ICMP error message, or | |
164 | * | |
165 | * * a datagram destined to an IP broadcast or IP multicast | |
166 | * address, or | |
167 | * | |
168 | * * a datagram sent as a link-layer broadcast, or | |
169 | * | |
170 | * * a non-initial fragment, or | |
171 | * | |
172 | * * a datagram whose source address does not define a single | |
173 | * host -- e.g., a zero address, a loopback address, a | |
174 | * broadcast address, a multicast address, or a Class E | |
175 | * address. | |
176 | */ | |
177 | static bool_t netlink_icmp_may_reply(struct buffer_if *buf) | |
178 | { | |
179 | struct iphdr *iph; | |
180 | uint32_t source; | |
181 | ||
182 | iph=(struct iphdr *)buf->start; | |
183 | if (iph->protocol==1) return False; /* Overly-broad; we may reply to | |
184 | eg. icmp echo-request */ | |
185 | /* How do we spot broadcast destination addresses? */ | |
186 | if (ntohs(iph->frag_off)&0x1fff) return False; /* Non-initial fragment */ | |
187 | source=ntohl(iph->saddr); | |
188 | if (source==0) return False; | |
189 | if ((source&0xff000000)==0x7f000000) return False; | |
190 | /* How do we spot broadcast source addresses? */ | |
191 | if ((source&0xf0000000)==0xe0000000) return False; /* Multicast */ | |
192 | if ((source&0xf0000000)==0xf0000000) return False; /* Class E */ | |
193 | return True; | |
194 | } | |
195 | ||
196 | /* How much of the original IP packet do we include in its ICMP | |
197 | response? The header plus up to 64 bits. */ | |
198 | static uint16_t netlink_icmp_reply_len(struct buffer_if *buf) | |
199 | { | |
200 | struct iphdr *iph=(struct iphdr *)buf->start; | |
201 | uint16_t hlen,plen; | |
202 | ||
203 | hlen=iph->ihl*4; | |
204 | /* We include the first 8 bytes of the packet data, provided they exist */ | |
205 | hlen+=8; | |
206 | plen=ntohs(iph->tot_len); | |
207 | return (hlen>plen?plen:hlen); | |
208 | } | |
209 | ||
70dc107b SE |
210 | /* client indicates where the packet we're constructing a response to |
211 | comes from. NULL indicates the host. */ | |
4efd681a | 212 | static void netlink_icmp_simple(struct netlink *st, struct buffer_if *buf, |
70dc107b | 213 | struct netlink_client *client, |
4efd681a SE |
214 | uint8_t type, uint8_t code) |
215 | { | |
216 | struct iphdr *iph=(struct iphdr *)buf->start; | |
217 | struct icmphdr *h; | |
218 | uint16_t len; | |
219 | ||
220 | if (netlink_icmp_may_reply(buf)) { | |
221 | len=netlink_icmp_reply_len(buf); | |
222 | h=netlink_icmp_tmpl(st,ntohl(iph->saddr),len); | |
223 | h->type=type; h->code=code; | |
224 | memcpy(buf_append(&st->icmp,len),buf->start,len); | |
225 | netlink_icmp_csum(h); | |
70dc107b | 226 | netlink_packet_deliver(st,NULL,&st->icmp); |
4efd681a SE |
227 | BUF_ASSERT_FREE(&st->icmp); |
228 | } | |
229 | } | |
230 | ||
231 | /* | |
232 | * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the | |
233 | * checksum. | |
234 | * | |
235 | * Is the datagram acceptable? | |
236 | * | |
237 | * 1. Length at least the size of an ip header | |
238 | * 2. Version of 4 | |
239 | * 3. Checksums correctly. | |
240 | * 4. Doesn't have a bogus length | |
241 | */ | |
242 | static bool_t netlink_check(struct netlink *st, struct buffer_if *buf) | |
243 | { | |
244 | struct iphdr *iph=(struct iphdr *)buf->start; | |
245 | uint32_t len; | |
246 | ||
9d3a4132 SE |
247 | if (iph->ihl < 5 || iph->version != 4) return False; |
248 | if (buf->size < iph->ihl*4) return False; | |
249 | if (ip_fast_csum((uint8_t *)iph, iph->ihl)!=0) return False; | |
4efd681a SE |
250 | len=ntohs(iph->tot_len); |
251 | /* There should be no padding */ | |
9d3a4132 | 252 | if (buf->size!=len || len<(iph->ihl<<2)) return False; |
4efd681a SE |
253 | /* XXX check that there's no source route specified */ |
254 | return True; | |
255 | } | |
256 | ||
70dc107b SE |
257 | /* Deliver a packet. "client" points to the _origin_ of the packet, not |
258 | its destination. (May be used when sending ICMP response - avoid | |
259 | asymmetric routing.) */ | |
260 | static void netlink_packet_deliver(struct netlink *st, | |
261 | struct netlink_client *client, | |
262 | struct buffer_if *buf) | |
4efd681a SE |
263 | { |
264 | struct iphdr *iph=(struct iphdr *)buf->start; | |
265 | uint32_t dest=ntohl(iph->daddr); | |
70dc107b SE |
266 | uint32_t source=ntohl(iph->saddr); |
267 | uint32_t best_quality; | |
268 | int best_match; | |
269 | int i; | |
2fe58dfd | 270 | |
4efd681a | 271 | BUF_ASSERT_USED(buf); |
2fe58dfd | 272 | |
4efd681a SE |
273 | if (dest==st->secnet_address) { |
274 | Message(M_ERROR,"%s: trying to deliver a packet to myself!\n"); | |
275 | BUF_FREE(buf); | |
2fe58dfd SE |
276 | return; |
277 | } | |
4efd681a | 278 | |
3454dce4 SE |
279 | /* XXX we're going to need an extra value 'allow_route' for the |
280 | source of the packet. It's always True for packets from the | |
281 | host. For packets from tunnels, we consult the client | |
282 | options. If !allow_route and the destination is a tunnel that | |
283 | also doesn't allow routing, we must reject the packet with an | |
284 | 'administratively prohibited' or something similar ICMP. */ | |
70dc107b SE |
285 | if (!client) { |
286 | /* Origin of packet is host or secnet. Might be for a tunnel. */ | |
287 | best_quality=0; | |
288 | best_match=-1; | |
289 | for (i=0; i<st->n_routes; i++) { | |
9d3a4132 | 290 | if (st->routes[i].up && subnet_match(&st->routes[i].net,dest)) { |
70dc107b SE |
291 | if (st->routes[i].c->link_quality>best_quality |
292 | || best_quality==0) { | |
293 | best_quality=st->routes[i].c->link_quality; | |
294 | best_match=i; | |
295 | /* If quality isn't perfect we may wish to | |
296 | consider kicking the tunnel with a 0-length | |
297 | packet to prompt it to perform a key setup. | |
298 | Then it'll eventually decide it's up or | |
299 | down. */ | |
300 | /* If quality is perfect we don't need to search | |
301 | any more. */ | |
302 | if (best_quality>=MAXIMUM_LINK_QUALITY) break; | |
303 | } | |
304 | } | |
305 | } | |
306 | if (best_match==-1) { | |
307 | /* Not going down a tunnel. Might be for the host. | |
308 | XXX think about this - only situation should be if we're | |
309 | sending ICMP. */ | |
310 | if (source!=st->secnet_address) { | |
311 | Message(M_ERROR,"netlink_packet_deliver: outgoing packet " | |
312 | "from host that won't fit down any of our tunnels!\n"); | |
9d3a4132 SE |
313 | /* XXX I think this could also occur if a soft tunnel just |
314 | went down, but still had packets queued in the kernel. */ | |
70dc107b SE |
315 | BUF_FREE(buf); |
316 | } else { | |
317 | st->deliver_to_host(st->dst,NULL,buf); | |
318 | BUF_ASSERT_FREE(buf); | |
319 | } | |
320 | } else { | |
321 | if (best_quality>0) { | |
322 | st->routes[best_match].c->deliver( | |
323 | st->routes[best_match].c->dst, | |
324 | st->routes[best_match].c, buf); | |
4efd681a SE |
325 | BUF_ASSERT_FREE(buf); |
326 | } else { | |
327 | /* Generate ICMP destination unreachable */ | |
70dc107b | 328 | netlink_icmp_simple(st,buf,client,3,0); /* client==NULL */ |
4efd681a SE |
329 | BUF_FREE(buf); |
330 | } | |
70dc107b SE |
331 | } |
332 | } else { /* client is set */ | |
333 | /* We know the origin is a tunnel - packet must be for the host */ | |
3454dce4 SE |
334 | /* XXX THIS IS NOT NECESSARILY TRUE, AND NEEDS FIXING */ |
335 | /* THIS FUNCTION MUST JUST DELIVER THE PACKET: IT MUST ASSUME | |
336 | THE PACKET HAS ALREADY BEEN CHECKED */ | |
70dc107b SE |
337 | if (subnet_matches_list(&st->networks,dest)) { |
338 | st->deliver_to_host(st->dst,NULL,buf); | |
339 | BUF_ASSERT_FREE(buf); | |
340 | } else { | |
341 | Message(M_ERROR,"%s: packet from tunnel %s can't be delivered " | |
342 | "to the host\n",st->name,client->name); | |
343 | netlink_icmp_simple(st,buf,client,3,0); | |
344 | BUF_FREE(buf); | |
2fe58dfd SE |
345 | } |
346 | } | |
70dc107b | 347 | BUF_ASSERT_FREE(buf); |
4efd681a SE |
348 | } |
349 | ||
70dc107b SE |
350 | static void netlink_packet_forward(struct netlink *st, |
351 | struct netlink_client *client, | |
352 | struct buffer_if *buf) | |
4efd681a SE |
353 | { |
354 | struct iphdr *iph=(struct iphdr *)buf->start; | |
355 | ||
356 | BUF_ASSERT_USED(buf); | |
357 | ||
358 | /* Packet has already been checked */ | |
359 | if (iph->ttl<=1) { | |
360 | /* Generate ICMP time exceeded */ | |
70dc107b | 361 | netlink_icmp_simple(st,buf,client,11,0); |
4efd681a SE |
362 | BUF_FREE(buf); |
363 | return; | |
364 | } | |
365 | iph->ttl--; | |
366 | iph->check=0; | |
367 | iph->check=ip_fast_csum((uint8_t *)iph,iph->ihl); | |
368 | ||
70dc107b | 369 | netlink_packet_deliver(st,client,buf); |
4efd681a SE |
370 | BUF_ASSERT_FREE(buf); |
371 | } | |
372 | ||
9d3a4132 | 373 | /* Deal with packets addressed explicitly to us */ |
70dc107b SE |
374 | static void netlink_packet_local(struct netlink *st, |
375 | struct netlink_client *client, | |
376 | struct buffer_if *buf) | |
4efd681a SE |
377 | { |
378 | struct icmphdr *h; | |
379 | ||
380 | h=(struct icmphdr *)buf->start; | |
381 | ||
382 | if ((ntohs(h->iph.frag_off)&0xbfff)!=0) { | |
9d3a4132 SE |
383 | Message(M_WARNING,"%s: fragmented packet addressed to secnet; " |
384 | "ignoring it\n",st->name); | |
4efd681a SE |
385 | BUF_FREE(buf); |
386 | return; | |
387 | } | |
388 | ||
389 | if (h->iph.protocol==1) { | |
390 | /* It's ICMP */ | |
391 | if (h->type==8 && h->code==0) { | |
392 | /* ICMP echo-request. Special case: we re-use the buffer | |
393 | to construct the reply. */ | |
394 | h->type=0; | |
395 | h->iph.daddr=h->iph.saddr; | |
396 | h->iph.saddr=htonl(st->secnet_address); | |
397 | h->iph.ttl=255; /* Be nice and bump it up again... */ | |
398 | h->iph.check=0; | |
399 | h->iph.check=ip_fast_csum((uint8_t *)h,h->iph.ihl); | |
400 | netlink_icmp_csum(h); | |
70dc107b | 401 | netlink_packet_deliver(st,NULL,buf); |
4efd681a SE |
402 | return; |
403 | } | |
404 | Message(M_WARNING,"%s: unknown incoming ICMP\n",st->name); | |
405 | } else { | |
406 | /* Send ICMP protocol unreachable */ | |
70dc107b | 407 | netlink_icmp_simple(st,buf,client,3,2); |
4efd681a SE |
408 | BUF_FREE(buf); |
409 | return; | |
410 | } | |
411 | ||
412 | BUF_FREE(buf); | |
413 | } | |
414 | ||
9d3a4132 SE |
415 | /* If cid==NULL packet is from host, otherwise cid specifies which tunnel |
416 | it came from. */ | |
417 | static void netlink_incoming(void *sst, void *cid, struct buffer_if *buf) | |
4efd681a SE |
418 | { |
419 | struct netlink *st=sst; | |
9d3a4132 | 420 | struct netlink_client *client=cid; |
4efd681a SE |
421 | uint32_t source,dest; |
422 | struct iphdr *iph; | |
423 | ||
424 | BUF_ASSERT_USED(buf); | |
425 | if (!netlink_check(st,buf)) { | |
9d3a4132 SE |
426 | Message(M_WARNING,"%s: bad IP packet from %s\n", |
427 | st->name,client?client->name:"host"); | |
4efd681a SE |
428 | BUF_FREE(buf); |
429 | return; | |
430 | } | |
431 | iph=(struct iphdr *)buf->start; | |
432 | ||
433 | source=ntohl(iph->saddr); | |
434 | dest=ntohl(iph->daddr); | |
435 | ||
9d3a4132 SE |
436 | /* Check source */ |
437 | if (client) { | |
c6f79b17 SE |
438 | /* Check that the packet source is appropriate for the tunnel |
439 | it came down */ | |
9d3a4132 SE |
440 | if (!subnet_matches_list(client->networks,source)) { |
441 | string_t s,d; | |
442 | s=ipaddr_to_string(source); | |
443 | d=ipaddr_to_string(dest); | |
444 | Message(M_WARNING,"%s: packet from tunnel %s with bad " | |
445 | "source address (s=%s,d=%s)\n",st->name,client->name,s,d); | |
446 | free(s); free(d); | |
447 | BUF_FREE(buf); | |
448 | return; | |
449 | } | |
450 | } else { | |
c6f79b17 SE |
451 | /* Check that the packet originates in our configured local |
452 | network, and hasn't been forwarded from elsewhere or | |
453 | generated with the wrong source address */ | |
9d3a4132 SE |
454 | if (!subnet_matches_list(&st->networks,source)) { |
455 | string_t s,d; | |
456 | s=ipaddr_to_string(source); | |
457 | d=ipaddr_to_string(dest); | |
458 | Message(M_WARNING,"%s: outgoing packet with bad source address " | |
459 | "(s=%s,d=%s)\n",st->name,s,d); | |
460 | free(s); free(d); | |
461 | BUF_FREE(buf); | |
462 | return; | |
463 | } | |
4efd681a | 464 | } |
c6f79b17 SE |
465 | |
466 | /* If this is a point-to-point device we don't examine the packet at | |
467 | all; we blindly send it down our one-and-only registered tunnel, | |
468 | or to the host, depending on where it came from. */ | |
469 | if (st->ptp) { | |
470 | if (client) { | |
471 | st->deliver_to_host(st->dst,NULL,buf); | |
472 | } else { | |
473 | st->clients->deliver(st->clients->dst,NULL,buf); | |
474 | } | |
475 | BUF_ASSERT_FREE(buf); | |
476 | return; | |
477 | } | |
478 | ||
9d3a4132 SE |
479 | /* (st->secnet_address needs checking before matching destination |
480 | addresses) */ | |
2fe58dfd | 481 | if (dest==st->secnet_address) { |
9d3a4132 | 482 | netlink_packet_local(st,client,buf); |
4efd681a | 483 | BUF_ASSERT_FREE(buf); |
2fe58dfd SE |
484 | return; |
485 | } | |
9d3a4132 | 486 | if (client) { |
3454dce4 | 487 | /* Check for free routing */ |
9d3a4132 SE |
488 | if (!subnet_matches_list(&st->networks,dest)) { |
489 | string_t s,d; | |
490 | s=ipaddr_to_string(source); | |
491 | d=ipaddr_to_string(dest); | |
492 | Message(M_WARNING,"%s: incoming packet from tunnel %s " | |
493 | "with bad destination address " | |
494 | "(s=%s,d=%s)\n",st->name,client->name,s,d); | |
495 | free(s); free(d); | |
496 | BUF_FREE(buf); | |
497 | return; | |
498 | } | |
2fe58dfd | 499 | } |
70dc107b | 500 | netlink_packet_forward(st,client,buf); |
4efd681a SE |
501 | BUF_ASSERT_FREE(buf); |
502 | } | |
503 | ||
9d3a4132 | 504 | static void netlink_set_softlinks(struct netlink *st, struct netlink_client *c, |
042a8da9 | 505 | bool_t up, uint32_t quality) |
4efd681a | 506 | { |
9d3a4132 | 507 | uint32_t i; |
4efd681a | 508 | |
9d3a4132 SE |
509 | if (!st->routes) return; /* Table has not yet been created */ |
510 | for (i=0; i<st->n_routes; i++) { | |
042a8da9 SE |
511 | if (st->routes[i].c==c) { |
512 | st->routes[i].quality=quality; | |
513 | if (!st->routes[i].hard) { | |
514 | st->routes[i].up=up; | |
515 | st->set_route(st->dst,&st->routes[i]); | |
516 | } | |
9d3a4132 | 517 | } |
4efd681a | 518 | } |
4efd681a SE |
519 | } |
520 | ||
70dc107b | 521 | static void netlink_set_quality(void *sst, void *cid, uint32_t quality) |
4efd681a | 522 | { |
9d3a4132 | 523 | struct netlink *st=sst; |
4efd681a SE |
524 | struct netlink_client *c=cid; |
525 | ||
70dc107b | 526 | c->link_quality=quality; |
9d3a4132 | 527 | if (c->link_quality==LINK_QUALITY_DOWN) { |
042a8da9 | 528 | netlink_set_softlinks(st,c,False,c->link_quality); |
9d3a4132 | 529 | } else { |
042a8da9 | 530 | netlink_set_softlinks(st,c,True,c->link_quality); |
9d3a4132 | 531 | } |
4efd681a SE |
532 | } |
533 | ||
534 | static void *netlink_regnets(void *sst, struct subnet_list *nets, | |
535 | netlink_deliver_fn *deliver, void *dst, | |
536 | uint32_t max_start_pad, uint32_t max_end_pad, | |
3454dce4 | 537 | uint32_t options, string_t client_name) |
4efd681a SE |
538 | { |
539 | struct netlink *st=sst; | |
540 | struct netlink_client *c; | |
541 | ||
542 | Message(M_DEBUG_CONFIG,"netlink_regnets: request for %d networks, " | |
543 | "max_start_pad=%d, max_end_pad=%d\n", | |
544 | nets->entries,max_start_pad,max_end_pad); | |
545 | ||
3454dce4 | 546 | if ((options&NETLINK_OPTION_SOFTROUTE) && !st->set_route) { |
9d3a4132 SE |
547 | Message(M_ERROR,"%s: this netlink device does not support " |
548 | "soft routes.\n"); | |
549 | return NULL; | |
550 | } | |
551 | ||
3454dce4 | 552 | if (options&NETLINK_OPTION_SOFTROUTE) { |
9d3a4132 | 553 | /* XXX for now we assume that soft routes require root privilege; |
3454dce4 | 554 | this may not always be true. The device driver can tell us. */ |
9d3a4132 SE |
555 | require_root_privileges=True; |
556 | require_root_privileges_explanation="netlink: soft routes"; | |
557 | } | |
558 | ||
9d3a4132 SE |
559 | /* Check that nets do not intersect st->exclude_remote_networks; |
560 | refuse to register if they do. */ | |
baa06aeb SE |
561 | if (subnet_lists_intersect(&st->exclude_remote_networks,nets)) { |
562 | Message(M_ERROR,"%s: site %s specifies networks that " | |
563 | "intersect with the explicitly excluded remote networks\n", | |
564 | st->name,client_name); | |
c6f79b17 SE |
565 | return NULL; |
566 | } | |
567 | ||
568 | if (st->clients && st->ptp) { | |
569 | fatal("%s: only one site may use a point-to-point netlink device\n", | |
570 | st->name); | |
571 | return NULL; | |
baa06aeb SE |
572 | } |
573 | ||
4efd681a SE |
574 | c=safe_malloc(sizeof(*c),"netlink_regnets"); |
575 | c->networks=nets; | |
576 | c->deliver=deliver; | |
577 | c->dst=dst; | |
c6f79b17 | 578 | c->name=client_name; |
3454dce4 | 579 | c->options=options; |
70dc107b | 580 | c->link_quality=LINK_QUALITY_DOWN; |
4efd681a SE |
581 | c->next=st->clients; |
582 | st->clients=c; | |
583 | if (max_start_pad > st->max_start_pad) st->max_start_pad=max_start_pad; | |
584 | if (max_end_pad > st->max_end_pad) st->max_end_pad=max_end_pad; | |
70dc107b | 585 | st->n_routes+=nets->entries; |
4efd681a SE |
586 | |
587 | return c; | |
588 | } | |
589 | ||
042a8da9 | 590 | static void netlink_dump_routes(struct netlink *st, bool_t requested) |
9d3a4132 SE |
591 | { |
592 | int i; | |
593 | string_t net; | |
042a8da9 | 594 | uint32_t c=M_INFO; |
9d3a4132 | 595 | |
042a8da9 SE |
596 | if (requested) c=M_WARNING; |
597 | Message(c,"%s: routing table:\n",st->name); | |
9d3a4132 SE |
598 | for (i=0; i<st->n_routes; i++) { |
599 | net=subnet_to_string(&st->routes[i].net); | |
042a8da9 | 600 | Message(c,"%s -> tunnel %s (%s,%s route,%s,quality %d)\n",net, |
3454dce4 | 601 | st->routes[i].c->name, |
9d3a4132 | 602 | st->routes[i].hard?"hard":"soft", |
3454dce4 | 603 | st->routes[i].allow_route?"free":"restricted", |
042a8da9 SE |
604 | st->routes[i].up?"up":"down", |
605 | st->routes[i].quality); | |
9d3a4132 SE |
606 | free(net); |
607 | } | |
042a8da9 | 608 | Message(c,"%s/32 -> netlink \"%s\"\n", |
9d3a4132 SE |
609 | ipaddr_to_string(st->secnet_address),st->name); |
610 | for (i=0; i<st->networks.entries; i++) { | |
611 | net=subnet_to_string(&st->networks.list[i]); | |
042a8da9 | 612 | Message(c,"%s -> host\n",net); |
9d3a4132 SE |
613 | free(net); |
614 | } | |
615 | } | |
616 | ||
70dc107b SE |
617 | static int netlink_compare_route_specificity(const void *ap, const void *bp) |
618 | { | |
619 | const struct netlink_route *a=ap; | |
620 | const struct netlink_route *b=bp; | |
621 | ||
622 | if (a->net.len==b->net.len) return 0; | |
623 | if (a->net.len<b->net.len) return 1; | |
624 | return -1; | |
625 | } | |
626 | ||
627 | static void netlink_phase_hook(void *sst, uint32_t new_phase) | |
628 | { | |
629 | struct netlink *st=sst; | |
630 | struct netlink_client *c; | |
631 | uint32_t i,j; | |
632 | ||
c6f79b17 SE |
633 | if (!st->clients && st->ptp) { |
634 | /* Point-to-point netlink devices must have precisely one | |
635 | client. If none has registered by now, complain. */ | |
636 | fatal("%s: point-to-point netlink devices must have precisely " | |
637 | "one client. This one doesn't have any.\n",st->name); | |
638 | } | |
639 | ||
70dc107b SE |
640 | /* All the networks serviced by the various tunnels should now |
641 | * have been registered. We build a routing table by sorting the | |
642 | * routes into most-specific-first order. */ | |
643 | st->routes=safe_malloc(st->n_routes*sizeof(*st->routes), | |
644 | "netlink_phase_hook"); | |
645 | /* Fill the table */ | |
646 | i=0; | |
647 | for (c=st->clients; c; c=c->next) { | |
648 | for (j=0; j<c->networks->entries; j++) { | |
649 | st->routes[i].net=c->networks->list[j]; | |
650 | st->routes[i].c=c; | |
3454dce4 SE |
651 | /* Hard routes are always up; |
652 | soft routes default to down */ | |
653 | st->routes[i].up=c->options&NETLINK_OPTION_SOFTROUTE?False:True; | |
9d3a4132 | 654 | st->routes[i].kup=False; |
3454dce4 SE |
655 | st->routes[i].hard=c->options&NETLINK_OPTION_SOFTROUTE?False:True; |
656 | st->routes[i].allow_route=c->options&NETLINK_OPTION_ALLOW_ROUTE? | |
657 | True:False; | |
042a8da9 | 658 | st->routes[i].quality=c->link_quality; |
70dc107b SE |
659 | i++; |
660 | } | |
661 | } | |
662 | /* ASSERT i==st->n_routes */ | |
663 | if (i!=st->n_routes) { | |
664 | fatal("netlink: route count error: expected %d got %d\n", | |
665 | st->n_routes,i); | |
666 | } | |
667 | /* Sort the table in descending order of specificity */ | |
668 | qsort(st->routes,st->n_routes,sizeof(*st->routes), | |
669 | netlink_compare_route_specificity); | |
9d3a4132 | 670 | |
042a8da9 SE |
671 | netlink_dump_routes(st,False); |
672 | } | |
673 | ||
674 | static void netlink_signal_handler(void *sst, int signum) | |
675 | { | |
676 | struct netlink *st=sst; | |
677 | Message(M_INFO,"%s: route dump requested by SIGUSR1\n",st->name); | |
678 | netlink_dump_routes(st,True); | |
70dc107b SE |
679 | } |
680 | ||
9d3a4132 SE |
681 | netlink_deliver_fn *netlink_init(struct netlink *st, |
682 | void *dst, struct cloc loc, | |
683 | dict_t *dict, string_t description, | |
684 | netlink_route_fn *set_route, | |
685 | netlink_deliver_fn *to_host) | |
4efd681a | 686 | { |
c6f79b17 SE |
687 | item_t *sa, *ptpa; |
688 | ||
4efd681a SE |
689 | st->dst=dst; |
690 | st->cl.description=description; | |
691 | st->cl.type=CL_NETLINK; | |
692 | st->cl.apply=NULL; | |
693 | st->cl.interface=&st->ops; | |
694 | st->ops.st=st; | |
695 | st->ops.regnets=netlink_regnets; | |
9d3a4132 | 696 | st->ops.deliver=netlink_incoming; |
70dc107b | 697 | st->ops.set_quality=netlink_set_quality; |
4efd681a SE |
698 | st->max_start_pad=0; |
699 | st->max_end_pad=0; | |
700 | st->clients=NULL; | |
9d3a4132 | 701 | st->set_route=set_route; |
4efd681a SE |
702 | st->deliver_to_host=to_host; |
703 | ||
704 | st->name=dict_read_string(dict,"name",False,"netlink",loc); | |
705 | if (!st->name) st->name=description; | |
706 | dict_read_subnet_list(dict, "networks", True, "netlink", loc, | |
707 | &st->networks); | |
baa06aeb SE |
708 | dict_read_subnet_list(dict, "exclude-remote-networks", False, "netlink", |
709 | loc, &st->exclude_remote_networks); | |
b2a56f7c SE |
710 | /* secnet-address does not have to be in local-networks; |
711 | however, it should be advertised in the 'sites' file for the | |
baa06aeb | 712 | local site. */ |
c6f79b17 SE |
713 | sa=dict_find_item(dict,"secnet-address",False,"netlink",loc); |
714 | ptpa=dict_find_item(dict,"ptp-address", False, "netlink", loc); | |
715 | if (sa && ptpa) { | |
716 | cfgfatal(loc,st->name,"you may not specify secnet-address and " | |
717 | "ptp-address in the same netlink device\n"); | |
718 | } | |
719 | if (!(sa || ptpa)) { | |
720 | cfgfatal(loc,st->name,"you must specify secnet-address or " | |
721 | "ptp-address for this netlink device\n"); | |
722 | } | |
723 | if (sa) { | |
724 | st->secnet_address=string_to_ipaddr(sa,"netlink"); | |
725 | st->ptp=False; | |
726 | } else { | |
727 | st->secnet_address=string_to_ipaddr(ptpa,"netlink"); | |
728 | st->ptp=True; | |
729 | } | |
4efd681a SE |
730 | st->mtu=dict_read_number(dict, "mtu", False, "netlink", loc, DEFAULT_MTU); |
731 | buffer_new(&st->icmp,ICMP_BUFSIZE); | |
70dc107b SE |
732 | st->n_routes=0; |
733 | st->routes=NULL; | |
734 | ||
735 | add_hook(PHASE_SETUP,netlink_phase_hook,st); | |
042a8da9 | 736 | request_signal_notification(SIGUSR1, netlink_signal_handler, st); |
4efd681a | 737 | |
9d3a4132 | 738 | return netlink_incoming; |
2fe58dfd SE |
739 | } |
740 | ||
9d3a4132 | 741 | /* No connection to the kernel at all... */ |
2fe58dfd | 742 | |
9d3a4132 | 743 | struct null { |
4efd681a | 744 | struct netlink nl; |
4efd681a | 745 | }; |
2fe58dfd | 746 | |
9d3a4132 | 747 | static bool_t null_set_route(void *sst, struct netlink_route *route) |
4efd681a | 748 | { |
9d3a4132 SE |
749 | struct null *st=sst; |
750 | string_t t; | |
4efd681a | 751 | |
9d3a4132 SE |
752 | if (route->up!=route->kup) { |
753 | t=subnet_to_string(&route->net); | |
754 | Message(M_INFO,"%s: setting route %s to state %s\n",st->nl.name, | |
755 | t, route->up?"up":"down"); | |
756 | free(t); | |
757 | route->kup=route->up; | |
758 | return True; | |
2fe58dfd | 759 | } |
9d3a4132 | 760 | return False; |
2fe58dfd | 761 | } |
9d3a4132 | 762 | |
2fe58dfd SE |
763 | static void null_deliver(void *sst, void *cid, struct buffer_if *buf) |
764 | { | |
765 | return; | |
766 | } | |
767 | ||
768 | static list_t *null_apply(closure_t *self, struct cloc loc, dict_t *context, | |
769 | list_t *args) | |
770 | { | |
771 | struct null *st; | |
4efd681a SE |
772 | item_t *item; |
773 | dict_t *dict; | |
2fe58dfd | 774 | |
4efd681a | 775 | st=safe_malloc(sizeof(*st),"null_apply"); |
2fe58dfd | 776 | |
4efd681a SE |
777 | item=list_elem(args,0); |
778 | if (!item || item->type!=t_dict) | |
779 | cfgfatal(loc,"null-netlink","parameter must be a dictionary\n"); | |
780 | ||
781 | dict=item->data.dict; | |
782 | ||
9d3a4132 SE |
783 | netlink_init(&st->nl,st,loc,dict,"null-netlink",null_set_route, |
784 | null_deliver); | |
4efd681a SE |
785 | |
786 | return new_closure(&st->nl.cl); | |
2fe58dfd SE |
787 | } |
788 | ||
789 | init_module netlink_module; | |
790 | void netlink_module(dict_t *dict) | |
791 | { | |
4efd681a | 792 | add_closure(dict,"null-netlink",null_apply); |
2fe58dfd | 793 | } |