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 | |
469fd1d9 SE |
17 | #define OPT_SOFTROUTE 1 |
18 | #define OPT_ALLOWROUTE 2 | |
19 | ||
4efd681a SE |
20 | /* Generic IP checksum routine */ |
21 | static inline uint16_t ip_csum(uint8_t *iph,uint32_t count) | |
2fe58dfd | 22 | { |
4efd681a SE |
23 | register uint32_t sum=0; |
24 | ||
25 | while (count>1) { | |
26 | sum+=ntohs(*(uint16_t *)iph); | |
27 | iph+=2; | |
28 | count-=2; | |
29 | } | |
30 | if(count>0) | |
31 | sum+=*(uint8_t *)iph; | |
32 | while (sum>>16) | |
33 | sum=(sum&0xffff)+(sum>>16); | |
34 | return htons(~sum); | |
2fe58dfd SE |
35 | } |
36 | ||
4efd681a SE |
37 | #ifdef i386 |
38 | /* | |
39 | * This is a version of ip_compute_csum() optimized for IP headers, | |
40 | * which always checksum on 4 octet boundaries. | |
41 | * | |
42 | * By Jorge Cwik <jorge@laser.satlink.net>, adapted for linux by | |
43 | * Arnt Gulbrandsen. | |
44 | */ | |
45 | static inline uint16_t ip_fast_csum(uint8_t *iph, uint32_t ihl) { | |
46 | uint32_t sum; | |
47 | ||
48 | __asm__ __volatile__(" | |
49 | movl (%1), %0 | |
50 | subl $4, %2 | |
51 | jbe 2f | |
52 | addl 4(%1), %0 | |
53 | adcl 8(%1), %0 | |
54 | adcl 12(%1), %0 | |
55 | 1: adcl 16(%1), %0 | |
56 | lea 4(%1), %1 | |
57 | decl %2 | |
58 | jne 1b | |
59 | adcl $0, %0 | |
60 | movl %0, %2 | |
61 | shrl $16, %0 | |
62 | addw %w2, %w0 | |
63 | adcl $0, %0 | |
64 | notl %0 | |
65 | 2: | |
66 | " | |
67 | /* Since the input registers which are loaded with iph and ipl | |
68 | are modified, we must also specify them as outputs, or gcc | |
69 | will assume they contain their original values. */ | |
70 | : "=r" (sum), "=r" (iph), "=r" (ihl) | |
71 | : "1" (iph), "2" (ihl)); | |
72 | return sum; | |
73 | } | |
74 | #else | |
75 | static inline uint16_t ip_fast_csum(uint8_t *iph, uint32_t ihl) | |
2fe58dfd | 76 | { |
4efd681a SE |
77 | return ip_csum(iph,ihl*4); |
78 | } | |
79 | #endif | |
80 | ||
81 | struct iphdr { | |
82 | #if defined (WORDS_BIGENDIAN) | |
83 | uint8_t version:4, | |
84 | ihl:4; | |
85 | #else | |
86 | uint8_t ihl:4, | |
87 | version:4; | |
88 | #endif | |
89 | uint8_t tos; | |
90 | uint16_t tot_len; | |
91 | uint16_t id; | |
92 | uint16_t frag_off; | |
93 | uint8_t ttl; | |
94 | uint8_t protocol; | |
95 | uint16_t check; | |
96 | uint32_t saddr; | |
97 | uint32_t daddr; | |
98 | /* The options start here. */ | |
99 | }; | |
100 | ||
101 | struct icmphdr { | |
102 | struct iphdr iph; | |
103 | uint8_t type; | |
104 | uint8_t code; | |
105 | uint16_t check; | |
106 | union { | |
107 | uint32_t unused; | |
108 | struct { | |
109 | uint8_t pointer; | |
110 | uint8_t unused1; | |
111 | uint16_t unused2; | |
112 | } pprob; | |
113 | uint32_t gwaddr; | |
114 | struct { | |
115 | uint16_t id; | |
116 | uint16_t seq; | |
117 | } echo; | |
118 | } d; | |
119 | }; | |
120 | ||
70dc107b SE |
121 | static void netlink_packet_deliver(struct netlink *st, |
122 | struct netlink_client *client, | |
123 | struct buffer_if *buf); | |
4efd681a SE |
124 | |
125 | static struct icmphdr *netlink_icmp_tmpl(struct netlink *st, | |
126 | uint32_t dest,uint16_t len) | |
127 | { | |
128 | struct icmphdr *h; | |
129 | ||
130 | BUF_ALLOC(&st->icmp,"netlink_icmp_tmpl"); | |
131 | buffer_init(&st->icmp,st->max_start_pad); | |
132 | h=buf_append(&st->icmp,sizeof(*h)); | |
133 | ||
134 | h->iph.version=4; | |
135 | h->iph.ihl=5; | |
136 | h->iph.tos=0; | |
137 | h->iph.tot_len=htons(len+(h->iph.ihl*4)+8); | |
138 | h->iph.id=0; | |
139 | h->iph.frag_off=0; | |
140 | h->iph.ttl=255; | |
141 | h->iph.protocol=1; | |
142 | h->iph.saddr=htonl(st->secnet_address); | |
143 | h->iph.daddr=htonl(dest); | |
144 | h->iph.check=0; | |
145 | h->iph.check=ip_fast_csum((uint8_t *)&h->iph,h->iph.ihl); | |
146 | h->check=0; | |
147 | h->d.unused=0; | |
148 | ||
149 | return h; | |
150 | } | |
151 | ||
152 | /* Fill in the ICMP checksum field correctly */ | |
153 | static void netlink_icmp_csum(struct icmphdr *h) | |
154 | { | |
155 | uint32_t len; | |
156 | ||
157 | len=ntohs(h->iph.tot_len)-(4*h->iph.ihl); | |
158 | h->check=0; | |
159 | h->check=ip_csum(&h->type,len); | |
160 | } | |
161 | ||
162 | /* RFC1122: | |
163 | * An ICMP error message MUST NOT be sent as the result of | |
164 | * receiving: | |
165 | * | |
166 | * * an ICMP error message, or | |
167 | * | |
168 | * * a datagram destined to an IP broadcast or IP multicast | |
169 | * address, or | |
170 | * | |
171 | * * a datagram sent as a link-layer broadcast, or | |
172 | * | |
173 | * * a non-initial fragment, or | |
174 | * | |
175 | * * a datagram whose source address does not define a single | |
176 | * host -- e.g., a zero address, a loopback address, a | |
177 | * broadcast address, a multicast address, or a Class E | |
178 | * address. | |
179 | */ | |
180 | static bool_t netlink_icmp_may_reply(struct buffer_if *buf) | |
181 | { | |
182 | struct iphdr *iph; | |
8dea8d37 | 183 | struct icmphdr *icmph; |
4efd681a SE |
184 | uint32_t source; |
185 | ||
186 | iph=(struct iphdr *)buf->start; | |
8dea8d37 SE |
187 | icmph=(struct icmphdr *)buf->start; |
188 | if (iph->protocol==1) { | |
189 | switch(icmph->type) { | |
190 | case 3: /* Destination unreachable */ | |
191 | case 11: /* Time Exceeded */ | |
192 | case 12: /* Parameter Problem */ | |
193 | return False; | |
194 | } | |
195 | } | |
4efd681a SE |
196 | /* How do we spot broadcast destination addresses? */ |
197 | if (ntohs(iph->frag_off)&0x1fff) return False; /* Non-initial fragment */ | |
198 | source=ntohl(iph->saddr); | |
199 | if (source==0) return False; | |
200 | if ((source&0xff000000)==0x7f000000) return False; | |
201 | /* How do we spot broadcast source addresses? */ | |
202 | if ((source&0xf0000000)==0xe0000000) return False; /* Multicast */ | |
203 | if ((source&0xf0000000)==0xf0000000) return False; /* Class E */ | |
204 | return True; | |
205 | } | |
206 | ||
207 | /* How much of the original IP packet do we include in its ICMP | |
208 | response? The header plus up to 64 bits. */ | |
209 | static uint16_t netlink_icmp_reply_len(struct buffer_if *buf) | |
210 | { | |
211 | struct iphdr *iph=(struct iphdr *)buf->start; | |
212 | uint16_t hlen,plen; | |
213 | ||
214 | hlen=iph->ihl*4; | |
215 | /* We include the first 8 bytes of the packet data, provided they exist */ | |
216 | hlen+=8; | |
217 | plen=ntohs(iph->tot_len); | |
218 | return (hlen>plen?plen:hlen); | |
219 | } | |
220 | ||
70dc107b SE |
221 | /* client indicates where the packet we're constructing a response to |
222 | comes from. NULL indicates the host. */ | |
4efd681a | 223 | static void netlink_icmp_simple(struct netlink *st, struct buffer_if *buf, |
70dc107b | 224 | struct netlink_client *client, |
4efd681a SE |
225 | uint8_t type, uint8_t code) |
226 | { | |
227 | struct iphdr *iph=(struct iphdr *)buf->start; | |
228 | struct icmphdr *h; | |
229 | uint16_t len; | |
230 | ||
231 | if (netlink_icmp_may_reply(buf)) { | |
232 | len=netlink_icmp_reply_len(buf); | |
233 | h=netlink_icmp_tmpl(st,ntohl(iph->saddr),len); | |
234 | h->type=type; h->code=code; | |
235 | memcpy(buf_append(&st->icmp,len),buf->start,len); | |
236 | netlink_icmp_csum(h); | |
70dc107b | 237 | netlink_packet_deliver(st,NULL,&st->icmp); |
4efd681a SE |
238 | BUF_ASSERT_FREE(&st->icmp); |
239 | } | |
240 | } | |
241 | ||
242 | /* | |
243 | * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the | |
244 | * checksum. | |
245 | * | |
246 | * Is the datagram acceptable? | |
247 | * | |
248 | * 1. Length at least the size of an ip header | |
249 | * 2. Version of 4 | |
250 | * 3. Checksums correctly. | |
251 | * 4. Doesn't have a bogus length | |
252 | */ | |
253 | static bool_t netlink_check(struct netlink *st, struct buffer_if *buf) | |
254 | { | |
255 | struct iphdr *iph=(struct iphdr *)buf->start; | |
256 | uint32_t len; | |
257 | ||
9d3a4132 SE |
258 | if (iph->ihl < 5 || iph->version != 4) return False; |
259 | if (buf->size < iph->ihl*4) return False; | |
260 | if (ip_fast_csum((uint8_t *)iph, iph->ihl)!=0) return False; | |
4efd681a SE |
261 | len=ntohs(iph->tot_len); |
262 | /* There should be no padding */ | |
9d3a4132 | 263 | if (buf->size!=len || len<(iph->ihl<<2)) return False; |
4efd681a SE |
264 | /* XXX check that there's no source route specified */ |
265 | return True; | |
266 | } | |
267 | ||
469fd1d9 SE |
268 | /* Deliver a packet. "client" is the _origin_ of the packet, not its |
269 | destination. */ | |
70dc107b SE |
270 | static void netlink_packet_deliver(struct netlink *st, |
271 | struct netlink_client *client, | |
272 | struct buffer_if *buf) | |
4efd681a SE |
273 | { |
274 | struct iphdr *iph=(struct iphdr *)buf->start; | |
275 | uint32_t dest=ntohl(iph->daddr); | |
70dc107b SE |
276 | uint32_t source=ntohl(iph->saddr); |
277 | uint32_t best_quality; | |
469fd1d9 SE |
278 | bool_t allow_route=False; |
279 | bool_t found_allowed=False; | |
70dc107b SE |
280 | int best_match; |
281 | int i; | |
2fe58dfd | 282 | |
4efd681a | 283 | BUF_ASSERT_USED(buf); |
2fe58dfd | 284 | |
4efd681a | 285 | if (dest==st->secnet_address) { |
469fd1d9 | 286 | Message(M_ERR,"%s: trying to deliver a packet to myself!\n"); |
4efd681a | 287 | BUF_FREE(buf); |
2fe58dfd SE |
288 | return; |
289 | } | |
4efd681a | 290 | |
469fd1d9 SE |
291 | /* Packets from the host (client==NULL) will always be routed. Packets |
292 | from clients with the allow_route option will also be routed. */ | |
293 | if (!client || (client && (client->options & OPT_ALLOWROUTE))) | |
294 | allow_route=True; | |
295 | ||
296 | /* If !allow_route, we check the routing table anyway, and if | |
297 | there's a suitable route with OPT_ALLOWROUTE set we use it. If | |
298 | there's a suitable route, but none with OPT_ALLOWROUTE set then | |
299 | we generate ICMP 'communication with destination network | |
300 | administratively prohibited'. */ | |
301 | ||
302 | best_quality=0; | |
303 | best_match=-1; | |
304 | for (i=0; i<st->n_routes; i++) { | |
305 | if (st->routes[i].up && subnet_match(&st->routes[i].net,dest)) { | |
306 | /* It's an available route to the correct destination. But is | |
307 | it better than the one we already have? */ | |
308 | ||
309 | /* If we have already found an allowed route then we don't | |
310 | bother looking at routes we're not allowed to use. If | |
311 | we don't yet have an allowed route we'll consider any. */ | |
312 | if (!allow_route && found_allowed) { | |
313 | if (!(st->routes[i].c->options&OPT_ALLOWROUTE)) continue; | |
70dc107b | 314 | } |
469fd1d9 SE |
315 | |
316 | if (st->routes[i].c->link_quality>best_quality | |
317 | || best_quality==0) { | |
318 | best_quality=st->routes[i].c->link_quality; | |
319 | best_match=i; | |
320 | if (st->routes[i].c->options&OPT_ALLOWROUTE) | |
321 | found_allowed=True; | |
322 | /* If quality isn't perfect we may wish to | |
323 | consider kicking the tunnel with a 0-length | |
324 | packet to prompt it to perform a key setup. | |
325 | Then it'll eventually decide it's up or | |
326 | down. */ | |
327 | /* If quality is perfect and we're allowed to use the | |
328 | route we don't need to search any more. */ | |
329 | if (best_quality>=MAXIMUM_LINK_QUALITY && | |
330 | (allow_route || found_allowed)) break; | |
4efd681a | 331 | } |
70dc107b | 332 | } |
469fd1d9 SE |
333 | } |
334 | if (best_match==-1) { | |
335 | /* The packet's not going down a tunnel. It might (ought to) | |
336 | be for the host. */ | |
70dc107b | 337 | if (subnet_matches_list(&st->networks,dest)) { |
469fd1d9 SE |
338 | st->deliver_to_host(st->dst,buf); |
339 | st->outcount++; | |
70dc107b SE |
340 | BUF_ASSERT_FREE(buf); |
341 | } else { | |
469fd1d9 SE |
342 | string_t s,d; |
343 | s=ipaddr_to_string(source); | |
344 | d=ipaddr_to_string(dest); | |
345 | Message(M_ERR,"%s: don't know where to deliver packet " | |
346 | "(s=%s, d=%s)\n", st->name, s, d); | |
347 | free(s); free(d); | |
70dc107b SE |
348 | netlink_icmp_simple(st,buf,client,3,0); |
349 | BUF_FREE(buf); | |
2fe58dfd | 350 | } |
469fd1d9 SE |
351 | } else { |
352 | if (!allow_route && | |
353 | !(st->routes[best_match].c->options&OPT_ALLOWROUTE)) { | |
354 | string_t s,d; | |
355 | s=ipaddr_to_string(source); | |
356 | d=ipaddr_to_string(dest); | |
357 | /* We have a usable route but aren't allowed to use it. | |
358 | Generate ICMP destination unreachable: communication | |
359 | with destination network administratively prohibited */ | |
360 | Message(M_NOTICE,"%s: denied forwarding for packet (s=%s, d=%s)\n", | |
361 | st->name,s,d); | |
362 | free(s); free(d); | |
363 | ||
364 | netlink_icmp_simple(st,buf,client,3,9); | |
365 | BUF_FREE(buf); | |
366 | } | |
367 | if (best_quality>0) { | |
368 | st->routes[best_match].c->deliver( | |
369 | st->routes[best_match].c->dst, buf); | |
370 | st->routes[best_match].outcount++; | |
371 | BUF_ASSERT_FREE(buf); | |
372 | } else { | |
373 | /* Generate ICMP destination unreachable */ | |
374 | netlink_icmp_simple(st,buf,client,3,0); /* client==NULL */ | |
375 | BUF_FREE(buf); | |
376 | } | |
2fe58dfd | 377 | } |
70dc107b | 378 | BUF_ASSERT_FREE(buf); |
4efd681a SE |
379 | } |
380 | ||
70dc107b SE |
381 | static void netlink_packet_forward(struct netlink *st, |
382 | struct netlink_client *client, | |
383 | struct buffer_if *buf) | |
4efd681a SE |
384 | { |
385 | struct iphdr *iph=(struct iphdr *)buf->start; | |
386 | ||
387 | BUF_ASSERT_USED(buf); | |
388 | ||
389 | /* Packet has already been checked */ | |
390 | if (iph->ttl<=1) { | |
391 | /* Generate ICMP time exceeded */ | |
70dc107b | 392 | netlink_icmp_simple(st,buf,client,11,0); |
4efd681a SE |
393 | BUF_FREE(buf); |
394 | return; | |
395 | } | |
396 | iph->ttl--; | |
397 | iph->check=0; | |
398 | iph->check=ip_fast_csum((uint8_t *)iph,iph->ihl); | |
399 | ||
70dc107b | 400 | netlink_packet_deliver(st,client,buf); |
4efd681a SE |
401 | BUF_ASSERT_FREE(buf); |
402 | } | |
403 | ||
9d3a4132 | 404 | /* Deal with packets addressed explicitly to us */ |
70dc107b SE |
405 | static void netlink_packet_local(struct netlink *st, |
406 | struct netlink_client *client, | |
407 | struct buffer_if *buf) | |
4efd681a SE |
408 | { |
409 | struct icmphdr *h; | |
410 | ||
469fd1d9 SE |
411 | st->localcount++; |
412 | ||
4efd681a SE |
413 | h=(struct icmphdr *)buf->start; |
414 | ||
415 | if ((ntohs(h->iph.frag_off)&0xbfff)!=0) { | |
9d3a4132 SE |
416 | Message(M_WARNING,"%s: fragmented packet addressed to secnet; " |
417 | "ignoring it\n",st->name); | |
4efd681a SE |
418 | BUF_FREE(buf); |
419 | return; | |
420 | } | |
421 | ||
422 | if (h->iph.protocol==1) { | |
423 | /* It's ICMP */ | |
424 | if (h->type==8 && h->code==0) { | |
425 | /* ICMP echo-request. Special case: we re-use the buffer | |
426 | to construct the reply. */ | |
427 | h->type=0; | |
428 | h->iph.daddr=h->iph.saddr; | |
429 | h->iph.saddr=htonl(st->secnet_address); | |
430 | h->iph.ttl=255; /* Be nice and bump it up again... */ | |
431 | h->iph.check=0; | |
432 | h->iph.check=ip_fast_csum((uint8_t *)h,h->iph.ihl); | |
433 | netlink_icmp_csum(h); | |
70dc107b | 434 | netlink_packet_deliver(st,NULL,buf); |
4efd681a SE |
435 | return; |
436 | } | |
437 | Message(M_WARNING,"%s: unknown incoming ICMP\n",st->name); | |
438 | } else { | |
439 | /* Send ICMP protocol unreachable */ | |
70dc107b | 440 | netlink_icmp_simple(st,buf,client,3,2); |
4efd681a SE |
441 | BUF_FREE(buf); |
442 | return; | |
443 | } | |
444 | ||
445 | BUF_FREE(buf); | |
446 | } | |
447 | ||
9d3a4132 SE |
448 | /* If cid==NULL packet is from host, otherwise cid specifies which tunnel |
449 | it came from. */ | |
469fd1d9 SE |
450 | static void netlink_incoming(struct netlink *st, struct netlink_client *client, |
451 | struct buffer_if *buf) | |
4efd681a | 452 | { |
4efd681a SE |
453 | uint32_t source,dest; |
454 | struct iphdr *iph; | |
455 | ||
456 | BUF_ASSERT_USED(buf); | |
457 | if (!netlink_check(st,buf)) { | |
9d3a4132 SE |
458 | Message(M_WARNING,"%s: bad IP packet from %s\n", |
459 | st->name,client?client->name:"host"); | |
4efd681a SE |
460 | BUF_FREE(buf); |
461 | return; | |
462 | } | |
463 | iph=(struct iphdr *)buf->start; | |
464 | ||
465 | source=ntohl(iph->saddr); | |
466 | dest=ntohl(iph->daddr); | |
467 | ||
9d3a4132 SE |
468 | /* Check source */ |
469 | if (client) { | |
c6f79b17 SE |
470 | /* Check that the packet source is appropriate for the tunnel |
471 | it came down */ | |
469fd1d9 | 472 | if (!subnet_matches_list(&client->networks,source)) { |
9d3a4132 SE |
473 | string_t s,d; |
474 | s=ipaddr_to_string(source); | |
475 | d=ipaddr_to_string(dest); | |
476 | Message(M_WARNING,"%s: packet from tunnel %s with bad " | |
477 | "source address (s=%s,d=%s)\n",st->name,client->name,s,d); | |
478 | free(s); free(d); | |
479 | BUF_FREE(buf); | |
480 | return; | |
481 | } | |
482 | } else { | |
c6f79b17 SE |
483 | /* Check that the packet originates in our configured local |
484 | network, and hasn't been forwarded from elsewhere or | |
485 | generated with the wrong source address */ | |
9d3a4132 SE |
486 | if (!subnet_matches_list(&st->networks,source)) { |
487 | string_t s,d; | |
488 | s=ipaddr_to_string(source); | |
489 | d=ipaddr_to_string(dest); | |
490 | Message(M_WARNING,"%s: outgoing packet with bad source address " | |
491 | "(s=%s,d=%s)\n",st->name,s,d); | |
492 | free(s); free(d); | |
493 | BUF_FREE(buf); | |
494 | return; | |
495 | } | |
4efd681a | 496 | } |
c6f79b17 SE |
497 | |
498 | /* If this is a point-to-point device we don't examine the packet at | |
499 | all; we blindly send it down our one-and-only registered tunnel, | |
500 | or to the host, depending on where it came from. */ | |
501 | if (st->ptp) { | |
502 | if (client) { | |
469fd1d9 | 503 | st->deliver_to_host(st->dst,buf); |
c6f79b17 | 504 | } else { |
469fd1d9 | 505 | st->clients->deliver(st->clients->dst,buf); |
c6f79b17 SE |
506 | } |
507 | BUF_ASSERT_FREE(buf); | |
508 | return; | |
509 | } | |
510 | ||
9d3a4132 SE |
511 | /* (st->secnet_address needs checking before matching destination |
512 | addresses) */ | |
2fe58dfd | 513 | if (dest==st->secnet_address) { |
9d3a4132 | 514 | netlink_packet_local(st,client,buf); |
4efd681a | 515 | BUF_ASSERT_FREE(buf); |
2fe58dfd SE |
516 | return; |
517 | } | |
9d3a4132 | 518 | if (client) { |
3454dce4 | 519 | /* Check for free routing */ |
9d3a4132 SE |
520 | if (!subnet_matches_list(&st->networks,dest)) { |
521 | string_t s,d; | |
522 | s=ipaddr_to_string(source); | |
523 | d=ipaddr_to_string(dest); | |
524 | Message(M_WARNING,"%s: incoming packet from tunnel %s " | |
525 | "with bad destination address " | |
526 | "(s=%s,d=%s)\n",st->name,client->name,s,d); | |
527 | free(s); free(d); | |
528 | BUF_FREE(buf); | |
529 | return; | |
530 | } | |
2fe58dfd | 531 | } |
70dc107b | 532 | netlink_packet_forward(st,client,buf); |
4efd681a SE |
533 | BUF_ASSERT_FREE(buf); |
534 | } | |
535 | ||
469fd1d9 SE |
536 | static void netlink_inst_incoming(void *sst, struct buffer_if *buf) |
537 | { | |
538 | struct netlink_client *c=sst; | |
539 | struct netlink *st=c->nst; | |
540 | ||
541 | netlink_incoming(st,c,buf); | |
542 | } | |
543 | ||
544 | static void netlink_dev_incoming(void *sst, struct buffer_if *buf) | |
545 | { | |
546 | struct netlink *st=sst; | |
547 | ||
548 | netlink_incoming(st,NULL,buf); | |
549 | } | |
550 | ||
9d3a4132 | 551 | static void netlink_set_softlinks(struct netlink *st, struct netlink_client *c, |
042a8da9 | 552 | bool_t up, uint32_t quality) |
4efd681a | 553 | { |
9d3a4132 | 554 | uint32_t i; |
4efd681a | 555 | |
9d3a4132 SE |
556 | if (!st->routes) return; /* Table has not yet been created */ |
557 | for (i=0; i<st->n_routes; i++) { | |
042a8da9 SE |
558 | if (st->routes[i].c==c) { |
559 | st->routes[i].quality=quality; | |
560 | if (!st->routes[i].hard) { | |
561 | st->routes[i].up=up; | |
562 | st->set_route(st->dst,&st->routes[i]); | |
563 | } | |
9d3a4132 | 564 | } |
4efd681a | 565 | } |
4efd681a SE |
566 | } |
567 | ||
469fd1d9 | 568 | static void netlink_set_quality(void *sst, uint32_t quality) |
4efd681a | 569 | { |
469fd1d9 SE |
570 | struct netlink_client *c=sst; |
571 | struct netlink *st=c->nst; | |
4efd681a | 572 | |
70dc107b | 573 | c->link_quality=quality; |
9d3a4132 | 574 | if (c->link_quality==LINK_QUALITY_DOWN) { |
042a8da9 | 575 | netlink_set_softlinks(st,c,False,c->link_quality); |
9d3a4132 | 576 | } else { |
042a8da9 | 577 | netlink_set_softlinks(st,c,True,c->link_quality); |
9d3a4132 | 578 | } |
4efd681a SE |
579 | } |
580 | ||
042a8da9 | 581 | static void netlink_dump_routes(struct netlink *st, bool_t requested) |
9d3a4132 SE |
582 | { |
583 | int i; | |
584 | string_t net; | |
042a8da9 | 585 | uint32_t c=M_INFO; |
9d3a4132 | 586 | |
042a8da9 | 587 | if (requested) c=M_WARNING; |
469fd1d9 SE |
588 | if (st->ptp) { |
589 | net=ipaddr_to_string(st->secnet_address); | |
590 | Message(c,"%s: point-to-point (remote end is %s); routes:\n", | |
591 | st->name, net); | |
9d3a4132 | 592 | free(net); |
469fd1d9 SE |
593 | for (i=0; i<st->n_routes; i++) { |
594 | net=subnet_to_string(&st->routes[i].net); | |
595 | Message(c,"%s ",net); | |
596 | free(net); | |
597 | } | |
598 | Message(c,"\n"); | |
599 | } else { | |
600 | Message(c,"%s: routing table:\n",st->name); | |
601 | for (i=0; i<st->n_routes; i++) { | |
602 | net=subnet_to_string(&st->routes[i].net); | |
603 | Message(c,"%s -> tunnel %s (%s,%s route,%s,quality %d,use %d)\n",net, | |
604 | st->routes[i].c->name, | |
605 | st->routes[i].hard?"hard":"soft", | |
606 | st->routes[i].allow_route?"free":"restricted", | |
607 | st->routes[i].up?"up":"down", | |
608 | st->routes[i].quality, | |
609 | st->routes[i].outcount); | |
610 | free(net); | |
611 | } | |
612 | net=ipaddr_to_string(st->secnet_address); | |
613 | Message(c,"%s/32 -> netlink \"%s\" (use %d)\n", | |
614 | net,st->name,st->localcount); | |
9d3a4132 | 615 | free(net); |
469fd1d9 SE |
616 | for (i=0; i<st->networks.entries; i++) { |
617 | net=subnet_to_string(&st->networks.list[i]); | |
618 | Message(c,"%s -> host (use %d)\n",net,st->outcount); | |
619 | free(net); | |
620 | } | |
9d3a4132 SE |
621 | } |
622 | } | |
623 | ||
70dc107b SE |
624 | static int netlink_compare_route_specificity(const void *ap, const void *bp) |
625 | { | |
626 | const struct netlink_route *a=ap; | |
627 | const struct netlink_route *b=bp; | |
628 | ||
629 | if (a->net.len==b->net.len) return 0; | |
630 | if (a->net.len<b->net.len) return 1; | |
631 | return -1; | |
632 | } | |
633 | ||
634 | static void netlink_phase_hook(void *sst, uint32_t new_phase) | |
635 | { | |
636 | struct netlink *st=sst; | |
637 | struct netlink_client *c; | |
638 | uint32_t i,j; | |
639 | ||
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) { | |
469fd1d9 SE |
648 | for (j=0; j<c->networks.entries; j++) { |
649 | st->routes[i].net=c->networks.list[j]; | |
70dc107b | 650 | st->routes[i].c=c; |
3454dce4 | 651 | /* Hard routes are always up; |
469fd1d9 SE |
652 | soft routes default to down; routes with no 'deliver' function |
653 | default to down */ | |
654 | st->routes[i].up=c->deliver? | |
655 | (c->options&OPT_SOFTROUTE?False:True): | |
656 | False; | |
9d3a4132 | 657 | st->routes[i].kup=False; |
469fd1d9 SE |
658 | st->routes[i].hard=c->options&OPT_SOFTROUTE?False:True; |
659 | st->routes[i].allow_route=c->options&OPT_ALLOWROUTE? | |
3454dce4 | 660 | True:False; |
042a8da9 | 661 | st->routes[i].quality=c->link_quality; |
469fd1d9 | 662 | st->routes[i].outcount=0; |
70dc107b SE |
663 | i++; |
664 | } | |
665 | } | |
666 | /* ASSERT i==st->n_routes */ | |
667 | if (i!=st->n_routes) { | |
668 | fatal("netlink: route count error: expected %d got %d\n", | |
669 | st->n_routes,i); | |
670 | } | |
671 | /* Sort the table in descending order of specificity */ | |
672 | qsort(st->routes,st->n_routes,sizeof(*st->routes), | |
673 | netlink_compare_route_specificity); | |
9d3a4132 | 674 | |
042a8da9 SE |
675 | netlink_dump_routes(st,False); |
676 | } | |
677 | ||
678 | static void netlink_signal_handler(void *sst, int signum) | |
679 | { | |
680 | struct netlink *st=sst; | |
681 | Message(M_INFO,"%s: route dump requested by SIGUSR1\n",st->name); | |
682 | netlink_dump_routes(st,True); | |
70dc107b SE |
683 | } |
684 | ||
469fd1d9 SE |
685 | static void netlink_inst_reg(void *sst, netlink_deliver_fn *deliver, |
686 | void *dst, uint32_t max_start_pad, | |
687 | uint32_t max_end_pad) | |
688 | { | |
689 | struct netlink_client *c=sst; | |
690 | struct netlink *st=c->nst; | |
691 | ||
692 | if (max_start_pad > st->max_start_pad) st->max_start_pad=max_start_pad; | |
693 | if (max_end_pad > st->max_end_pad) st->max_end_pad=max_end_pad; | |
694 | c->deliver=deliver; | |
695 | c->dst=dst; | |
696 | } | |
697 | ||
698 | static struct flagstr netlink_option_table[]={ | |
699 | { "soft", OPT_SOFTROUTE }, | |
700 | { "allow-route", OPT_ALLOWROUTE }, | |
701 | { NULL, 0} | |
702 | }; | |
703 | /* This is the routine that gets called when the closure that's | |
704 | returned by an invocation of a netlink device closure (eg. tun, | |
705 | userv-ipif) is invoked. It's used to create routes and pass in | |
706 | information about them; the closure it returns is used by site | |
707 | code. */ | |
708 | static closure_t *netlink_inst_create(struct netlink *st, | |
709 | struct cloc loc, dict_t *dict) | |
710 | { | |
711 | struct netlink_client *c; | |
712 | string_t name; | |
713 | struct subnet_list networks; | |
714 | uint32_t options; | |
715 | ||
716 | name=dict_read_string(dict, "name", True, st->name, loc); | |
717 | ||
718 | dict_read_subnet_list(dict, "routes", True, st->name, loc, | |
719 | &networks); | |
720 | options=string_list_to_word(dict_lookup(dict,"options"), | |
721 | netlink_option_table,st->name); | |
722 | ||
723 | if ((options&OPT_SOFTROUTE) && !st->set_route) { | |
724 | cfgfatal(loc,st->name,"this netlink device does not support " | |
725 | "soft routes.\n"); | |
726 | return NULL; | |
727 | } | |
728 | ||
729 | if (options&OPT_SOFTROUTE) { | |
730 | /* XXX for now we assume that soft routes require root privilege; | |
731 | this may not always be true. The device driver can tell us. */ | |
732 | require_root_privileges=True; | |
733 | require_root_privileges_explanation="netlink: soft routes"; | |
734 | if (st->ptp) { | |
735 | cfgfatal(loc,st->name,"point-to-point netlinks do not support " | |
736 | "soft routes.\n"); | |
737 | return NULL; | |
738 | } | |
739 | } | |
740 | ||
741 | /* Check that nets do not intersect st->exclude_remote_networks; | |
742 | refuse to register if they do. */ | |
743 | if (subnet_lists_intersect(&st->exclude_remote_networks,&networks)) { | |
744 | cfgfatal(loc,st->name,"networks intersect with the explicitly " | |
745 | "excluded remote networks\n"); | |
746 | return NULL; | |
747 | } | |
748 | ||
749 | c=safe_malloc(sizeof(*c),"netlink_inst_create"); | |
750 | c->cl.description=name; | |
751 | c->cl.type=CL_NETLINK; | |
752 | c->cl.apply=NULL; | |
753 | c->cl.interface=&c->ops; | |
754 | c->ops.st=c; | |
755 | c->ops.reg=netlink_inst_reg; | |
756 | c->ops.deliver=netlink_inst_incoming; | |
757 | c->ops.set_quality=netlink_set_quality; | |
758 | c->nst=st; | |
759 | ||
760 | c->networks=networks; | |
761 | c->deliver=NULL; | |
762 | c->dst=NULL; | |
763 | c->name=name; | |
764 | c->options=options; | |
765 | c->link_quality=LINK_QUALITY_DOWN; | |
766 | c->next=st->clients; | |
767 | st->clients=c; | |
768 | st->n_routes+=networks.entries; | |
769 | ||
770 | return &c->cl; | |
771 | } | |
772 | ||
773 | static list_t *netlink_inst_apply(closure_t *self, struct cloc loc, | |
774 | dict_t *context, list_t *args) | |
775 | { | |
776 | struct netlink *st=self->interface; | |
777 | ||
778 | dict_t *dict; | |
779 | item_t *item; | |
780 | closure_t *cl; | |
781 | ||
782 | Message(M_DEBUG_CONFIG,"netlink_inst_apply\n"); | |
783 | ||
784 | item=list_elem(args,0); | |
785 | if (!item || item->type!=t_dict) { | |
786 | cfgfatal(loc,st->name,"must have a dictionary argument\n"); | |
787 | } | |
788 | dict=item->data.dict; | |
789 | ||
790 | cl=netlink_inst_create(st,loc,dict); | |
791 | ||
792 | return new_closure(cl); | |
793 | } | |
794 | ||
9d3a4132 SE |
795 | netlink_deliver_fn *netlink_init(struct netlink *st, |
796 | void *dst, struct cloc loc, | |
797 | dict_t *dict, string_t description, | |
798 | netlink_route_fn *set_route, | |
799 | netlink_deliver_fn *to_host) | |
4efd681a | 800 | { |
c6f79b17 SE |
801 | item_t *sa, *ptpa; |
802 | ||
4efd681a SE |
803 | st->dst=dst; |
804 | st->cl.description=description; | |
469fd1d9 SE |
805 | st->cl.type=CL_PURE; |
806 | st->cl.apply=netlink_inst_apply; | |
807 | st->cl.interface=st; | |
4efd681a SE |
808 | st->max_start_pad=0; |
809 | st->max_end_pad=0; | |
810 | st->clients=NULL; | |
9d3a4132 | 811 | st->set_route=set_route; |
4efd681a SE |
812 | st->deliver_to_host=to_host; |
813 | ||
814 | st->name=dict_read_string(dict,"name",False,"netlink",loc); | |
815 | if (!st->name) st->name=description; | |
816 | dict_read_subnet_list(dict, "networks", True, "netlink", loc, | |
817 | &st->networks); | |
baa06aeb SE |
818 | dict_read_subnet_list(dict, "exclude-remote-networks", False, "netlink", |
819 | loc, &st->exclude_remote_networks); | |
c6f79b17 | 820 | sa=dict_find_item(dict,"secnet-address",False,"netlink",loc); |
469fd1d9 | 821 | ptpa=dict_find_item(dict,"ptp-address",False,"netlink",loc); |
c6f79b17 SE |
822 | if (sa && ptpa) { |
823 | cfgfatal(loc,st->name,"you may not specify secnet-address and " | |
824 | "ptp-address in the same netlink device\n"); | |
825 | } | |
826 | if (!(sa || ptpa)) { | |
827 | cfgfatal(loc,st->name,"you must specify secnet-address or " | |
828 | "ptp-address for this netlink device\n"); | |
829 | } | |
830 | if (sa) { | |
831 | st->secnet_address=string_to_ipaddr(sa,"netlink"); | |
832 | st->ptp=False; | |
833 | } else { | |
834 | st->secnet_address=string_to_ipaddr(ptpa,"netlink"); | |
835 | st->ptp=True; | |
836 | } | |
4efd681a SE |
837 | st->mtu=dict_read_number(dict, "mtu", False, "netlink", loc, DEFAULT_MTU); |
838 | buffer_new(&st->icmp,ICMP_BUFSIZE); | |
70dc107b SE |
839 | st->n_routes=0; |
840 | st->routes=NULL; | |
469fd1d9 SE |
841 | st->outcount=0; |
842 | st->localcount=0; | |
70dc107b SE |
843 | |
844 | add_hook(PHASE_SETUP,netlink_phase_hook,st); | |
042a8da9 | 845 | request_signal_notification(SIGUSR1, netlink_signal_handler, st); |
4efd681a | 846 | |
469fd1d9 SE |
847 | /* If we're point-to-point then we return a CL_NETLINK directly, |
848 | rather than a CL_NETLINK_OLD or pure closure (depending on | |
849 | compatibility). This CL_NETLINK is for our one and only | |
850 | client. Our cl.apply function is NULL. */ | |
851 | if (st->ptp) { | |
852 | closure_t *cl; | |
853 | cl=netlink_inst_create(st,loc,dict); | |
854 | st->cl=*cl; | |
855 | } | |
856 | return netlink_dev_incoming; | |
2fe58dfd SE |
857 | } |
858 | ||
9d3a4132 | 859 | /* No connection to the kernel at all... */ |
2fe58dfd | 860 | |
9d3a4132 | 861 | struct null { |
4efd681a | 862 | struct netlink nl; |
4efd681a | 863 | }; |
2fe58dfd | 864 | |
9d3a4132 | 865 | static bool_t null_set_route(void *sst, struct netlink_route *route) |
4efd681a | 866 | { |
9d3a4132 SE |
867 | struct null *st=sst; |
868 | string_t t; | |
4efd681a | 869 | |
9d3a4132 SE |
870 | if (route->up!=route->kup) { |
871 | t=subnet_to_string(&route->net); | |
872 | Message(M_INFO,"%s: setting route %s to state %s\n",st->nl.name, | |
873 | t, route->up?"up":"down"); | |
874 | free(t); | |
875 | route->kup=route->up; | |
876 | return True; | |
2fe58dfd | 877 | } |
9d3a4132 | 878 | return False; |
2fe58dfd | 879 | } |
9d3a4132 | 880 | |
469fd1d9 | 881 | static void null_deliver(void *sst, struct buffer_if *buf) |
2fe58dfd SE |
882 | { |
883 | return; | |
884 | } | |
885 | ||
886 | static list_t *null_apply(closure_t *self, struct cloc loc, dict_t *context, | |
887 | list_t *args) | |
888 | { | |
889 | struct null *st; | |
4efd681a SE |
890 | item_t *item; |
891 | dict_t *dict; | |
2fe58dfd | 892 | |
4efd681a | 893 | st=safe_malloc(sizeof(*st),"null_apply"); |
2fe58dfd | 894 | |
4efd681a SE |
895 | item=list_elem(args,0); |
896 | if (!item || item->type!=t_dict) | |
897 | cfgfatal(loc,"null-netlink","parameter must be a dictionary\n"); | |
898 | ||
899 | dict=item->data.dict; | |
900 | ||
9d3a4132 SE |
901 | netlink_init(&st->nl,st,loc,dict,"null-netlink",null_set_route, |
902 | null_deliver); | |
4efd681a SE |
903 | |
904 | return new_closure(&st->nl.cl); | |
2fe58dfd SE |
905 | } |
906 | ||
907 | init_module netlink_module; | |
908 | void netlink_module(dict_t *dict) | |
909 | { | |
4efd681a | 910 | add_closure(dict,"null-netlink",null_apply); |
2fe58dfd | 911 | } |