1 /* User-kernel network link */
3 /* See RFCs 791, 792, 1123 and 1812 */
5 /* The netlink device is actually a router. Tunnels are unnumbered
6 point-to-point lines (RFC1812 section 2.2.7); the router has a
7 single address (the 'router-id'). */
9 /* This is where we currently have the anti-spoofing paranoia - before
10 sending a packet to the kernel we check that the tunnel it came
11 over could reasonably have produced it. */
14 /* Points to note from RFC1812 (which may require changes in this
17 3.3.4 Maximum Transmission Unit - MTU
19 The MTU of each logical interface MUST be configurable within the
20 range of legal MTUs for the interface.
22 Many Link Layer protocols define a maximum frame size that may be
23 sent. In such cases, a router MUST NOT allow an MTU to be set which
24 would allow sending of frames larger than those allowed by the Link
25 Layer protocol. However, a router SHOULD be willing to receive a
26 packet as large as the maximum frame size even if that is larger than
29 4.2.1 A router SHOULD count datagrams discarded.
31 4.2.2.1 Source route options - we probably should implement processing
32 of source routes, even though mostly the security policy will prevent
35 5.3.13.4 Source Route Options
37 A router MUST implement support for source route options in forwarded
38 packets. A router MAY implement a configuration option that, when
39 enabled, causes all source-routed packets to be discarded. However,
40 such an option MUST NOT be enabled by default.
42 5.3.13.5 Record Route Option
44 Routers MUST support the Record Route option in forwarded packets.
46 A router MAY provide a configuration option that, if enabled, will
47 cause the router to ignore (i.e., pass through unchanged) Record
48 Route options in forwarded packets. If provided, such an option MUST
49 default to enabling the record-route. This option should not affect
50 the processing of Record Route options in datagrams received by the
51 router itself (in particular, Record Route options in ICMP echo
52 requests will still be processed according to Section [4.3.3.6]).
54 5.3.13.6 Timestamp Option
56 Routers MUST support the timestamp option in forwarded packets. A
57 timestamp value MUST follow the rules given [INTRO:2].
59 If the flags field = 3 (timestamp and prespecified address), the
60 router MUST add its timestamp if the next prespecified address
61 matches any of the router's IP addresses. It is not necessary that
62 the prespecified address be either the address of the interface on
63 which the packet arrived or the address of the interface over which
67 4.2.2.7 Fragmentation: RFC 791 Section 3.2
69 Fragmentation, as described in [INTERNET:1], MUST be supported by a
72 4.2.2.8 Reassembly: RFC 791 Section 3.2
74 As specified in the corresponding section of [INTRO:2], a router MUST
75 support reassembly of datagrams that it delivers to itself.
77 4.2.2.9 Time to Live: RFC 791 Section 3.2
79 Note in particular that a router MUST NOT check the TTL of a packet
80 except when forwarding it.
82 A router MUST NOT discard a datagram just because it was received
83 with TTL equal to zero or one; if it is to the router and otherwise
84 valid, the router MUST attempt to receive it.
86 On messages the router originates, the IP layer MUST provide a means
87 for the transport layer to set the TTL field of every datagram that
88 is sent. When a fixed TTL value is used, it MUST be configurable.
91 8.1 The Simple Network Management Protocol - SNMP
92 8.1.1 SNMP Protocol Elements
94 Routers MUST be manageable by SNMP [MGT:3]. The SNMP MUST operate
95 using UDP/IP as its transport and network protocols.
109 #define ICMP_TYPE_ECHO_REPLY 0
111 #define ICMP_TYPE_UNREACHABLE 3
112 #define ICMP_CODE_NET_UNREACHABLE 0
113 #define ICMP_CODE_PROTOCOL_UNREACHABLE 2
114 #define ICMP_CODE_FRAGMENTATION_REQUIRED 4
115 #define ICMP_CODE_NET_PROHIBITED 13
117 #define ICMP_TYPE_ECHO_REQUEST 8
119 #define ICMP_TYPE_TIME_EXCEEDED 11
120 #define ICMP_CODE_TTL_EXCEEDED 0
122 /* Generic IP checksum routine */
123 static inline uint16_t ip_csum(uint8_t *iph,int32_t count)
125 register uint32_t sum=0;
128 sum+=ntohs(*(uint16_t *)iph);
133 sum+=*(uint8_t *)iph;
135 sum=(sum&0xffff)+(sum>>16);
141 * This is a version of ip_compute_csum() optimized for IP headers,
142 * which always checksum on 4 octet boundaries.
144 * By Jorge Cwik <jorge@laser.satlink.net>, adapted for linux by
147 static inline uint16_t ip_fast_csum(uint8_t *iph, int32_t ihl) {
150 __asm__ __volatile__(
156 "adcl 12(%1), %0 ;\n"
157 "1: adcl 16(%1), %0 ;\n"
168 /* Since the input registers which are loaded with iph and ipl
169 are modified, we must also specify them as outputs, or gcc
170 will assume they contain their original values. */
171 : "=r" (sum), "=r" (iph), "=r" (ihl)
172 : "1" (iph), "2" (ihl)
177 static inline uint16_t ip_fast_csum(uint8_t *iph, int32_t ihl)
179 assert(ihl < INT_MAX/4);
180 return ip_csum(iph,ihl*4);
185 #if defined (WORDS_BIGENDIAN)
201 /* The options start here. */
224 static void netlink_packet_deliver(struct netlink *st,
225 struct netlink_client *client,
226 struct buffer_if *buf);
228 /* XXX RFC1812 4.3.2.5:
229 All other ICMP error messages (Destination Unreachable,
230 Redirect, Time Exceeded, and Parameter Problem) SHOULD have their
231 precedence value set to 6 (INTERNETWORK CONTROL) or 7 (NETWORK
232 CONTROL). The IP Precedence value for these error messages MAY be
235 static struct icmphdr *netlink_icmp_tmpl(struct netlink *st,
236 uint32_t dest,uint16_t len)
240 BUF_ALLOC(&st->icmp,"netlink_icmp_tmpl");
241 buffer_init(&st->icmp,calculate_max_start_pad());
242 h=buf_append(&st->icmp,sizeof(*h));
247 h->iph.tot_len=htons(len+(h->iph.ihl*4)+8);
250 h->iph.ttl=255; /* XXX should be configurable */
252 h->iph.saddr=htonl(st->secnet_address);
253 h->iph.daddr=htonl(dest);
255 h->iph.check=ip_fast_csum((uint8_t *)&h->iph,h->iph.ihl);
262 /* Fill in the ICMP checksum field correctly */
263 static void netlink_icmp_csum(struct icmphdr *h)
267 len=ntohs(h->iph.tot_len)-(4*h->iph.ihl);
269 h->check=ip_csum(&h->type,len);
273 * An ICMP error message MUST NOT be sent as the result of
276 * * an ICMP error message, or
278 * * a datagram destined to an IP broadcast or IP multicast
281 * * a datagram sent as a link-layer broadcast, or
283 * * a non-initial fragment, or
285 * * a datagram whose source address does not define a single
286 * host -- e.g., a zero address, a loopback address, a
287 * broadcast address, a multicast address, or a Class E
290 static bool_t netlink_icmp_may_reply(struct buffer_if *buf)
293 struct icmphdr *icmph;
296 if (buf->size < (int)sizeof(struct icmphdr)) return False;
297 iph=(struct iphdr *)buf->start;
298 icmph=(struct icmphdr *)buf->start;
299 if (iph->protocol==1) {
300 switch(icmph->type) {
301 case 3: /* Destination unreachable */
302 case 11: /* Time Exceeded */
303 case 12: /* Parameter Problem */
307 /* How do we spot broadcast destination addresses? */
308 if (ntohs(iph->frag_off)&0x1fff) return False; /* Non-initial fragment */
309 source=ntohl(iph->saddr);
310 if (source==0) return False;
311 if ((source&0xff000000)==0x7f000000) return False;
312 /* How do we spot broadcast source addresses? */
313 if ((source&0xf0000000)==0xe0000000) return False; /* Multicast */
314 if ((source&0xf0000000)==0xf0000000) return False; /* Class E */
318 /* How much of the original IP packet do we include in its ICMP
319 response? The header plus up to 64 bits. */
322 4.3.2.3 Original Message Header
324 Historically, every ICMP error message has included the Internet
325 header and at least the first 8 data bytes of the datagram that
326 triggered the error. This is no longer adequate, due to the use of
327 IP-in-IP tunneling and other technologies. Therefore, the ICMP
328 datagram SHOULD contain as much of the original datagram as possible
329 without the length of the ICMP datagram exceeding 576 bytes. The
330 returned IP header (and user data) MUST be identical to that which
331 was received, except that the router is not required to undo any
332 modifications to the IP header that are normally performed in
333 forwarding that were performed before the error was detected (e.g.,
334 decrementing the TTL, or updating options). Note that the
335 requirements of Section [4.3.3.5] supersede this requirement in some
336 cases (i.e., for a Parameter Problem message, if the problem is in a
337 modified field, the router must undo the modification). See Section
340 static uint16_t netlink_icmp_reply_len(struct buffer_if *buf)
342 if (buf->size < (int)sizeof(struct iphdr)) return 0;
343 struct iphdr *iph=(struct iphdr *)buf->start;
347 /* We include the first 8 bytes of the packet data, provided they exist */
349 plen=ntohs(iph->tot_len);
350 return (hlen>plen?plen:hlen);
353 /* client indicates where the packet we're constructing a response to
354 comes from. NULL indicates the host. */
355 static void netlink_icmp_simple(struct netlink *st, struct buffer_if *buf,
356 struct netlink_client *client,
357 uint8_t type, uint8_t code)
362 if (netlink_icmp_may_reply(buf)) {
363 struct iphdr *iph=(struct iphdr *)buf->start;
364 len=netlink_icmp_reply_len(buf);
365 h=netlink_icmp_tmpl(st,ntohl(iph->saddr),len);
366 h->type=type; h->code=code;
367 memcpy(buf_append(&st->icmp,len),buf->start,len);
368 netlink_icmp_csum(h);
369 netlink_packet_deliver(st,NULL,&st->icmp);
370 BUF_ASSERT_FREE(&st->icmp);
375 * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the
377 * RFC1812: 4.2.2.5 MUST discard messages containing invalid checksums.
379 * Is the datagram acceptable?
381 * 1. Length at least the size of an ip header
383 * 3. Checksums correctly.
384 * 4. Doesn't have a bogus length
386 static bool_t netlink_check(struct netlink *st, struct buffer_if *buf,
387 char *errmsgbuf, int errmsgbuflen)
389 #define BAD(...) do{ \
390 snprintf(errmsgbuf,errmsgbuflen,__VA_ARGS__); \
394 if (buf->size < (int)sizeof(struct iphdr)) BAD("len %"PRIu32"",buf->size);
395 struct iphdr *iph=(struct iphdr *)buf->start;
398 if (iph->ihl < 5) BAD("ihl %u",iph->ihl);
399 if (iph->version != 4) BAD("version %u",iph->version);
400 if (buf->size < iph->ihl*4) BAD("size %"PRId32"<%u*4",buf->size,iph->ihl);
401 if (ip_fast_csum((uint8_t *)iph, iph->ihl)!=0) BAD("csum");
402 len=ntohs(iph->tot_len);
403 /* There should be no padding */
404 if (buf->size!=len) BAD("len %"PRId32"!=%"PRId32,buf->size,len);
405 if (len<(iph->ihl<<2)) BAD("len %"PRId32"<(%u<<2)",len,iph->ihl);
406 /* XXX check that there's no source route specified */
412 /* Deliver a packet _to_ client; used after we have decided
413 * what to do with it. */
414 static void netlink_client_deliver(struct netlink *st,
415 struct netlink_client *client,
416 uint32_t source, uint32_t dest,
417 struct buffer_if *buf)
419 client->deliver(client->dst, buf);
423 /* Deliver a packet. "client" is the _origin_ of the packet, not its
424 destination, and is NULL for packets from the host and packets
425 generated internally in secnet. */
426 static void netlink_packet_deliver(struct netlink *st,
427 struct netlink_client *client,
428 struct buffer_if *buf)
430 if (buf->size < (int)sizeof(struct iphdr)) {
431 Message(M_ERR,"%s: trying to deliver a too-short packet"
432 " from %s!\n",st->name, client?client->name:"(local)");
437 struct iphdr *iph=(struct iphdr *)buf->start;
438 uint32_t dest=ntohl(iph->daddr);
439 uint32_t source=ntohl(iph->saddr);
440 uint32_t best_quality;
441 bool_t allow_route=False;
442 bool_t found_allowed=False;
446 BUF_ASSERT_USED(buf);
448 if (dest==st->secnet_address) {
449 Message(M_ERR,"%s: trying to deliver a packet to myself!\n",st->name);
454 /* Packets from the host (client==NULL) may always be routed. Packets
455 from clients with the allow_route option will also be routed. */
456 if (!client || (client && (client->options & OPT_ALLOWROUTE)))
459 /* If !allow_route, we check the routing table anyway, and if
460 there's a suitable route with OPT_ALLOWROUTE set we use it. If
461 there's a suitable route, but none with OPT_ALLOWROUTE set then
462 we generate ICMP 'communication with destination network
463 administratively prohibited'. */
467 for (i=0; i<st->n_clients; i++) {
468 if (st->routes[i]->up &&
469 ipset_contains_addr(st->routes[i]->networks,dest)) {
470 /* It's an available route to the correct destination. But is
471 it better than the one we already have? */
473 /* If we have already found an allowed route then we don't
474 bother looking at routes we're not allowed to use. If
475 we don't yet have an allowed route we'll consider any. */
476 if (!allow_route && found_allowed) {
477 if (!(st->routes[i]->options&OPT_ALLOWROUTE)) continue;
480 if (st->routes[i]->link_quality>best_quality
481 || best_quality==0) {
482 best_quality=st->routes[i]->link_quality;
484 if (st->routes[i]->options&OPT_ALLOWROUTE)
486 /* If quality isn't perfect we may wish to
487 consider kicking the tunnel with a 0-length
488 packet to prompt it to perform a key setup.
489 Then it'll eventually decide it's up or
491 /* If quality is perfect and we're allowed to use the
492 route we don't need to search any more. */
493 if (best_quality>=MAXIMUM_LINK_QUALITY &&
494 (allow_route || found_allowed)) break;
498 if (best_match==-1) {
499 /* The packet's not going down a tunnel. It might (ought to)
501 if (ipset_contains_addr(st->networks,dest)) {
502 st->deliver_to_host(st->dst,buf);
504 BUF_ASSERT_FREE(buf);
507 s=ipaddr_to_string(source);
508 d=ipaddr_to_string(dest);
509 Message(M_DEBUG,"%s: don't know where to deliver packet "
510 "(s=%s, d=%s)\n", st->name, s, d);
512 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
513 ICMP_CODE_NET_UNREACHABLE);
518 !(st->routes[best_match]->options&OPT_ALLOWROUTE)) {
520 s=ipaddr_to_string(source);
521 d=ipaddr_to_string(dest);
522 /* We have a usable route but aren't allowed to use it.
523 Generate ICMP destination unreachable: communication
524 with destination network administratively prohibited */
525 Message(M_NOTICE,"%s: denied forwarding for packet (s=%s, d=%s)\n",
529 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
530 ICMP_CODE_NET_PROHIBITED);
533 if (best_quality>0) {
534 /* XXX Fragment if required */
535 netlink_client_deliver(st,st->routes[best_match],
537 BUF_ASSERT_FREE(buf);
539 /* Generate ICMP destination unreachable */
540 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
541 ICMP_CODE_NET_UNREACHABLE); /* client==NULL */
546 BUF_ASSERT_FREE(buf);
549 static void netlink_packet_forward(struct netlink *st,
550 struct netlink_client *client,
551 struct buffer_if *buf)
553 if (buf->size < (int)sizeof(struct iphdr)) return;
554 struct iphdr *iph=(struct iphdr *)buf->start;
556 BUF_ASSERT_USED(buf);
558 /* Packet has already been checked */
560 /* Generate ICMP time exceeded */
561 netlink_icmp_simple(st,buf,client,ICMP_TYPE_TIME_EXCEEDED,
562 ICMP_CODE_TTL_EXCEEDED);
568 iph->check=ip_fast_csum((uint8_t *)iph,iph->ihl);
570 netlink_packet_deliver(st,client,buf);
571 BUF_ASSERT_FREE(buf);
574 /* Deal with packets addressed explicitly to us */
575 static void netlink_packet_local(struct netlink *st,
576 struct netlink_client *client,
577 struct buffer_if *buf)
583 if (buf->size < (int)sizeof(struct icmphdr)) {
584 Message(M_WARNING,"%s: short packet addressed to secnet; "
585 "ignoring it\n",st->name);
589 h=(struct icmphdr *)buf->start;
591 if ((ntohs(h->iph.frag_off)&0xbfff)!=0) {
592 Message(M_WARNING,"%s: fragmented packet addressed to secnet; "
593 "ignoring it\n",st->name);
598 if (h->iph.protocol==1) {
600 if (h->type==ICMP_TYPE_ECHO_REQUEST && h->code==0) {
601 /* ICMP echo-request. Special case: we re-use the buffer
602 to construct the reply. */
603 h->type=ICMP_TYPE_ECHO_REPLY;
604 h->iph.daddr=h->iph.saddr;
605 h->iph.saddr=htonl(st->secnet_address);
608 h->iph.check=ip_fast_csum((uint8_t *)h,h->iph.ihl);
609 netlink_icmp_csum(h);
610 netlink_packet_deliver(st,NULL,buf);
613 Message(M_WARNING,"%s: unknown incoming ICMP\n",st->name);
615 /* Send ICMP protocol unreachable */
616 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
617 ICMP_CODE_PROTOCOL_UNREACHABLE);
625 /* If cid==NULL packet is from host, otherwise cid specifies which tunnel
627 static void netlink_incoming(struct netlink *st, struct netlink_client *client,
628 struct buffer_if *buf)
630 uint32_t source,dest;
633 const char *sourcedesc=client?client->name:"host";
635 BUF_ASSERT_USED(buf);
637 if (!netlink_check(st,buf,errmsgbuf,sizeof(errmsgbuf))) {
638 Message(M_WARNING,"%s: bad IP packet from %s: %s\n",
644 assert(buf->size >= (int)sizeof(struct icmphdr));
645 iph=(struct iphdr *)buf->start;
647 source=ntohl(iph->saddr);
648 dest=ntohl(iph->daddr);
650 /* Check source. If we don't like the source, there's no point
651 generating ICMP because we won't know how to get it to the
652 source of the packet. */
654 /* Check that the packet source is appropriate for the tunnel
656 if (!ipset_contains_addr(client->networks,source)) {
658 s=ipaddr_to_string(source);
659 d=ipaddr_to_string(dest);
660 Message(M_WARNING,"%s: packet from tunnel %s with bad "
661 "source address (s=%s,d=%s)\n",st->name,client->name,s,d);
667 /* Check that the packet originates in our configured local
668 network, and hasn't been forwarded from elsewhere or
669 generated with the wrong source address */
670 if (!ipset_contains_addr(st->networks,source)) {
672 s=ipaddr_to_string(source);
673 d=ipaddr_to_string(dest);
674 Message(M_WARNING,"%s: outgoing packet with bad source address "
675 "(s=%s,d=%s)\n",st->name,s,d);
682 /* If this is a point-to-point device we don't examine the
683 destination address at all; we blindly send it down our
684 one-and-only registered tunnel, or to the host, depending on
685 where it came from. It's up to external software to check
686 address validity and generate ICMP, etc. */
689 st->deliver_to_host(st->dst,buf);
691 netlink_client_deliver(st,st->clients,source,dest,buf);
693 BUF_ASSERT_FREE(buf);
697 /* st->secnet_address needs checking before matching destination
699 if (dest==st->secnet_address) {
700 netlink_packet_local(st,client,buf);
701 BUF_ASSERT_FREE(buf);
704 netlink_packet_forward(st,client,buf);
705 BUF_ASSERT_FREE(buf);
708 static void netlink_inst_incoming(void *sst, struct buffer_if *buf)
710 struct netlink_client *c=sst;
711 struct netlink *st=c->nst;
713 netlink_incoming(st,c,buf);
716 static void netlink_dev_incoming(void *sst, struct buffer_if *buf)
718 struct netlink *st=sst;
720 netlink_incoming(st,NULL,buf);
723 static void netlink_set_quality(void *sst, uint32_t quality)
725 struct netlink_client *c=sst;
726 struct netlink *st=c->nst;
728 c->link_quality=quality;
729 c->up=(c->link_quality==LINK_QUALITY_DOWN)?False:True;
730 if (c->options&OPT_SOFTROUTE) {
731 st->set_routes(st->dst,c);
735 static void netlink_output_subnets(struct netlink *st, uint32_t loglevel,
736 struct subnet_list *snets)
741 for (i=0; i<snets->entries; i++) {
742 net=subnet_to_string(snets->list[i]);
743 Message(loglevel,"%s ",net);
748 static void netlink_dump_routes(struct netlink *st, bool_t requested)
754 if (requested) c=M_WARNING;
756 net=ipaddr_to_string(st->secnet_address);
757 Message(c,"%s: point-to-point (remote end is %s); routes:\n",
760 netlink_output_subnets(st,c,st->clients->subnets);
763 Message(c,"%s: routing table:\n",st->name);
764 for (i=0; i<st->n_clients; i++) {
765 netlink_output_subnets(st,c,st->routes[i]->subnets);
766 Message(c,"-> tunnel %s (%s,mtu %d,%s routes,%s,"
767 "quality %d,use %d,pri %lu)\n",
769 st->routes[i]->up?"up":"down",
771 st->routes[i]->options&OPT_SOFTROUTE?"soft":"hard",
772 st->routes[i]->options&OPT_ALLOWROUTE?"free":"restricted",
773 st->routes[i]->link_quality,
774 st->routes[i]->outcount,
775 (unsigned long)st->routes[i]->priority);
777 net=ipaddr_to_string(st->secnet_address);
778 Message(c,"%s/32 -> netlink \"%s\" (use %d)\n",
779 net,st->name,st->localcount);
781 for (i=0; i<st->subnets->entries; i++) {
782 net=subnet_to_string(st->subnets->list[i]);
783 Message(c,"%s ",net);
787 Message(c,"-> host (use %d)\n",st->outcount);
791 /* ap is a pointer to a member of the routes array */
792 static int netlink_compare_client_priority(const void *ap, const void *bp)
794 const struct netlink_client *const*a=ap;
795 const struct netlink_client *const*b=bp;
797 if ((*a)->priority==(*b)->priority) return 0;
798 if ((*a)->priority<(*b)->priority) return 1;
802 static void netlink_phase_hook(void *sst, uint32_t new_phase)
804 struct netlink *st=sst;
805 struct netlink_client *c;
808 /* All the networks serviced by the various tunnels should now
809 * have been registered. We build a routing table by sorting the
810 * clients by priority. */
811 st->routes=safe_malloc_ary(sizeof(*st->routes),st->n_clients,
812 "netlink_phase_hook");
815 for (c=st->clients; c; c=c->next) {
819 /* Sort the table in descending order of priority */
820 qsort(st->routes,st->n_clients,sizeof(*st->routes),
821 netlink_compare_client_priority);
823 netlink_dump_routes(st,False);
826 static void netlink_signal_handler(void *sst, int signum)
828 struct netlink *st=sst;
829 Message(M_INFO,"%s: route dump requested by SIGUSR1\n",st->name);
830 netlink_dump_routes(st,True);
833 static void netlink_inst_set_mtu(void *sst, int32_t new_mtu)
835 struct netlink_client *c=sst;
840 static void netlink_inst_reg(void *sst, netlink_deliver_fn *deliver,
843 struct netlink_client *c=sst;
849 static struct flagstr netlink_option_table[]={
850 { "soft", OPT_SOFTROUTE },
851 { "allow-route", OPT_ALLOWROUTE },
854 /* This is the routine that gets called when the closure that's
855 returned by an invocation of a netlink device closure (eg. tun,
856 userv-ipif) is invoked. It's used to create routes and pass in
857 information about them; the closure it returns is used by site
859 static closure_t *netlink_inst_create(struct netlink *st,
860 struct cloc loc, dict_t *dict)
862 struct netlink_client *c;
864 struct ipset *networks;
865 uint32_t options,priority;
869 name=dict_read_string(dict, "name", True, st->name, loc);
871 l=dict_lookup(dict,"routes");
873 cfgfatal(loc,st->name,"required parameter \"routes\" not found\n");
874 networks=string_list_to_ipset(l,loc,st->name,"routes");
875 options=string_list_to_word(dict_lookup(dict,"options"),
876 netlink_option_table,st->name);
878 priority=dict_read_number(dict,"priority",False,st->name,loc,0);
879 mtu=dict_read_number(dict,"mtu",False,st->name,loc,0);
881 if ((options&OPT_SOFTROUTE) && !st->set_routes) {
882 cfgfatal(loc,st->name,"this netlink device does not support "
887 if (options&OPT_SOFTROUTE) {
888 /* XXX for now we assume that soft routes require root privilege;
889 this may not always be true. The device driver can tell us. */
890 require_root_privileges=True;
891 require_root_privileges_explanation="netlink: soft routes";
893 cfgfatal(loc,st->name,"point-to-point netlinks do not support "
899 /* Check that nets are a subset of st->remote_networks;
900 refuse to register if they are not. */
901 if (!ipset_is_subset(st->remote_networks,networks)) {
902 cfgfatal(loc,st->name,"routes are not allowed\n");
906 c=safe_malloc(sizeof(*c),"netlink_inst_create");
907 c->cl.description=name;
908 c->cl.type=CL_NETLINK;
910 c->cl.interface=&c->ops;
912 c->ops.reg=netlink_inst_reg;
913 c->ops.deliver=netlink_inst_incoming;
914 c->ops.set_quality=netlink_set_quality;
915 c->ops.set_mtu=netlink_inst_set_mtu;
918 c->networks=networks;
919 c->subnets=ipset_to_subnet_list(networks);
920 c->priority=priority;
924 c->link_quality=LINK_QUALITY_UNUSED;
925 c->mtu=mtu?mtu:st->mtu;
932 assert(st->n_clients < INT_MAX);
938 static list_t *netlink_inst_apply(closure_t *self, struct cloc loc,
939 dict_t *context, list_t *args)
941 struct netlink *st=self->interface;
947 item=list_elem(args,0);
948 if (!item || item->type!=t_dict) {
949 cfgfatal(loc,st->name,"must have a dictionary argument\n");
951 dict=item->data.dict;
953 cl=netlink_inst_create(st,loc,dict);
955 return new_closure(cl);
958 netlink_deliver_fn *netlink_init(struct netlink *st,
959 void *dst, struct cloc loc,
960 dict_t *dict, cstring_t description,
961 netlink_route_fn *set_routes,
962 netlink_deliver_fn *to_host)
968 st->cl.description=description;
970 st->cl.apply=netlink_inst_apply;
975 st->set_routes=set_routes;
976 st->deliver_to_host=to_host;
978 st->name=dict_read_string(dict,"name",False,description,loc);
979 if (!st->name) st->name=description;
980 l=dict_lookup(dict,"networks");
982 st->networks=string_list_to_ipset(l,loc,st->name,"networks");
986 st->networks=ipset_complement(empty);
989 l=dict_lookup(dict,"remote-networks");
991 st->remote_networks=string_list_to_ipset(l,loc,st->name,
996 st->remote_networks=ipset_complement(empty);
1000 sa=dict_find_item(dict,"secnet-address",False,"netlink",loc);
1001 ptpa=dict_find_item(dict,"ptp-address",False,"netlink",loc);
1003 cfgfatal(loc,st->name,"you may not specify secnet-address and "
1004 "ptp-address in the same netlink device\n");
1006 if (!(sa || ptpa)) {
1007 cfgfatal(loc,st->name,"you must specify secnet-address or "
1008 "ptp-address for this netlink device\n");
1011 st->secnet_address=string_item_to_ipaddr(sa,"netlink");
1014 st->secnet_address=string_item_to_ipaddr(ptpa,"netlink");
1017 /* To be strictly correct we could subtract secnet_address from
1018 networks here. It shouldn't make any practical difference,
1019 though, and will make the route dump look complicated... */
1020 st->subnets=ipset_to_subnet_list(st->networks);
1021 st->mtu=dict_read_number(dict, "mtu", False, "netlink", loc, DEFAULT_MTU);
1022 buffer_new(&st->icmp,ICMP_BUFSIZE);
1026 add_hook(PHASE_SETUP,netlink_phase_hook,st);
1027 request_signal_notification(SIGUSR1, netlink_signal_handler, st);
1029 /* If we're point-to-point then we return a CL_NETLINK directly,
1030 rather than a CL_NETLINK_OLD or pure closure (depending on
1031 compatibility). This CL_NETLINK is for our one and only
1032 client. Our cl.apply function is NULL. */
1035 cl=netlink_inst_create(st,loc,dict);
1038 return netlink_dev_incoming;
1041 /* No connection to the kernel at all... */
1047 static bool_t null_set_route(void *sst, struct netlink_client *routes)
1049 struct null *st=sst;
1051 if (routes->up!=routes->kup) {
1052 Message(M_INFO,"%s: setting routes for tunnel %s to state %s\n",
1053 st->nl.name,routes->name,
1054 routes->up?"up":"down");
1055 routes->kup=routes->up;
1061 static void null_deliver(void *sst, struct buffer_if *buf)
1066 static list_t *null_apply(closure_t *self, struct cloc loc, dict_t *context,
1073 st=safe_malloc(sizeof(*st),"null_apply");
1075 item=list_elem(args,0);
1076 if (!item || item->type!=t_dict)
1077 cfgfatal(loc,"null-netlink","parameter must be a dictionary\n");
1079 dict=item->data.dict;
1081 netlink_init(&st->nl,st,loc,dict,"null-netlink",null_set_route,
1084 return new_closure(&st->nl.cl);
1087 void netlink_module(dict_t *dict)
1089 add_closure(dict,"null-netlink",null_apply);