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,st->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 iph=(struct iphdr *)buf->start;
297 icmph=(struct icmphdr *)buf->start;
298 if (iph->protocol==1) {
299 switch(icmph->type) {
300 case 3: /* Destination unreachable */
301 case 11: /* Time Exceeded */
302 case 12: /* Parameter Problem */
306 /* How do we spot broadcast destination addresses? */
307 if (ntohs(iph->frag_off)&0x1fff) return False; /* Non-initial fragment */
308 source=ntohl(iph->saddr);
309 if (source==0) return False;
310 if ((source&0xff000000)==0x7f000000) return False;
311 /* How do we spot broadcast source addresses? */
312 if ((source&0xf0000000)==0xe0000000) return False; /* Multicast */
313 if ((source&0xf0000000)==0xf0000000) return False; /* Class E */
317 /* How much of the original IP packet do we include in its ICMP
318 response? The header plus up to 64 bits. */
321 4.3.2.3 Original Message Header
323 Historically, every ICMP error message has included the Internet
324 header and at least the first 8 data bytes of the datagram that
325 triggered the error. This is no longer adequate, due to the use of
326 IP-in-IP tunneling and other technologies. Therefore, the ICMP
327 datagram SHOULD contain as much of the original datagram as possible
328 without the length of the ICMP datagram exceeding 576 bytes. The
329 returned IP header (and user data) MUST be identical to that which
330 was received, except that the router is not required to undo any
331 modifications to the IP header that are normally performed in
332 forwarding that were performed before the error was detected (e.g.,
333 decrementing the TTL, or updating options). Note that the
334 requirements of Section [4.3.3.5] supersede this requirement in some
335 cases (i.e., for a Parameter Problem message, if the problem is in a
336 modified field, the router must undo the modification). See Section
339 static uint16_t netlink_icmp_reply_len(struct buffer_if *buf)
341 struct iphdr *iph=(struct iphdr *)buf->start;
345 /* We include the first 8 bytes of the packet data, provided they exist */
347 plen=ntohs(iph->tot_len);
348 return (hlen>plen?plen:hlen);
351 /* client indicates where the packet we're constructing a response to
352 comes from. NULL indicates the host. */
353 static void netlink_icmp_simple(struct netlink *st, struct buffer_if *buf,
354 struct netlink_client *client,
355 uint8_t type, uint8_t code)
357 struct iphdr *iph=(struct iphdr *)buf->start;
361 if (netlink_icmp_may_reply(buf)) {
362 len=netlink_icmp_reply_len(buf);
363 h=netlink_icmp_tmpl(st,ntohl(iph->saddr),len);
364 h->type=type; h->code=code;
365 memcpy(buf_append(&st->icmp,len),buf->start,len);
366 netlink_icmp_csum(h);
367 netlink_packet_deliver(st,NULL,&st->icmp);
368 BUF_ASSERT_FREE(&st->icmp);
373 * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the
375 * RFC1812: 4.2.2.5 MUST discard messages containing invalid checksums.
377 * Is the datagram acceptable?
379 * 1. Length at least the size of an ip header
381 * 3. Checksums correctly.
382 * 4. Doesn't have a bogus length
384 static bool_t netlink_check(struct netlink *st, struct buffer_if *buf,
385 char *errmsgbuf, int errmsgbuflen)
387 #define BAD(...) do{ \
388 snprintf(errmsgbuf,errmsgbuflen,__VA_ARGS__); \
392 struct iphdr *iph=(struct iphdr *)buf->start;
395 if (iph->ihl < 5) BAD("ihl %u",iph->ihl);
396 if (iph->version != 4) BAD("version %u",iph->version);
397 if (buf->size < iph->ihl*4) BAD("size %"PRId32"<%u*4",buf->size,iph->ihl);
398 if (ip_fast_csum((uint8_t *)iph, iph->ihl)!=0) BAD("csum");
399 len=ntohs(iph->tot_len);
400 /* There should be no padding */
401 if (buf->size!=len) BAD("len %"PRId32"!=%"PRId32,buf->size,len);
402 if (len<(iph->ihl<<2)) BAD("len %"PRId32"<(%u<<2)",len,iph->ihl);
403 /* XXX check that there's no source route specified */
409 /* Deliver a packet. "client" is the _origin_ of the packet, not its
410 destination, and is NULL for packets from the host and packets
411 generated internally in secnet. */
412 static void netlink_packet_deliver(struct netlink *st,
413 struct netlink_client *client,
414 struct buffer_if *buf)
416 struct iphdr *iph=(struct iphdr *)buf->start;
417 uint32_t dest=ntohl(iph->daddr);
418 uint32_t source=ntohl(iph->saddr);
419 uint32_t best_quality;
420 bool_t allow_route=False;
421 bool_t found_allowed=False;
425 BUF_ASSERT_USED(buf);
427 if (dest==st->secnet_address) {
428 Message(M_ERR,"%s: trying to deliver a packet to myself!\n",st->name);
433 /* Packets from the host (client==NULL) may always be routed. Packets
434 from clients with the allow_route option will also be routed. */
435 if (!client || (client && (client->options & OPT_ALLOWROUTE)))
438 /* If !allow_route, we check the routing table anyway, and if
439 there's a suitable route with OPT_ALLOWROUTE set we use it. If
440 there's a suitable route, but none with OPT_ALLOWROUTE set then
441 we generate ICMP 'communication with destination network
442 administratively prohibited'. */
446 for (i=0; i<st->n_clients; i++) {
447 if (st->routes[i]->up &&
448 ipset_contains_addr(st->routes[i]->networks,dest)) {
449 /* It's an available route to the correct destination. But is
450 it better than the one we already have? */
452 /* If we have already found an allowed route then we don't
453 bother looking at routes we're not allowed to use. If
454 we don't yet have an allowed route we'll consider any. */
455 if (!allow_route && found_allowed) {
456 if (!(st->routes[i]->options&OPT_ALLOWROUTE)) continue;
459 if (st->routes[i]->link_quality>best_quality
460 || best_quality==0) {
461 best_quality=st->routes[i]->link_quality;
463 if (st->routes[i]->options&OPT_ALLOWROUTE)
465 /* If quality isn't perfect we may wish to
466 consider kicking the tunnel with a 0-length
467 packet to prompt it to perform a key setup.
468 Then it'll eventually decide it's up or
470 /* If quality is perfect and we're allowed to use the
471 route we don't need to search any more. */
472 if (best_quality>=MAXIMUM_LINK_QUALITY &&
473 (allow_route || found_allowed)) break;
477 if (best_match==-1) {
478 /* The packet's not going down a tunnel. It might (ought to)
480 if (ipset_contains_addr(st->networks,dest)) {
481 st->deliver_to_host(st->dst,buf);
483 BUF_ASSERT_FREE(buf);
486 s=ipaddr_to_string(source);
487 d=ipaddr_to_string(dest);
488 Message(M_DEBUG,"%s: don't know where to deliver packet "
489 "(s=%s, d=%s)\n", st->name, s, d);
491 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
492 ICMP_CODE_NET_UNREACHABLE);
497 !(st->routes[best_match]->options&OPT_ALLOWROUTE)) {
499 s=ipaddr_to_string(source);
500 d=ipaddr_to_string(dest);
501 /* We have a usable route but aren't allowed to use it.
502 Generate ICMP destination unreachable: communication
503 with destination network administratively prohibited */
504 Message(M_NOTICE,"%s: denied forwarding for packet (s=%s, d=%s)\n",
508 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
509 ICMP_CODE_NET_PROHIBITED);
512 if (best_quality>0) {
513 /* XXX Fragment if required */
514 st->routes[best_match]->deliver(
515 st->routes[best_match]->dst, buf);
516 st->routes[best_match]->outcount++;
517 BUF_ASSERT_FREE(buf);
519 /* Generate ICMP destination unreachable */
520 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
521 ICMP_CODE_NET_UNREACHABLE); /* client==NULL */
526 BUF_ASSERT_FREE(buf);
529 static void netlink_packet_forward(struct netlink *st,
530 struct netlink_client *client,
531 struct buffer_if *buf)
533 struct iphdr *iph=(struct iphdr *)buf->start;
535 BUF_ASSERT_USED(buf);
537 /* Packet has already been checked */
539 /* Generate ICMP time exceeded */
540 netlink_icmp_simple(st,buf,client,ICMP_TYPE_TIME_EXCEEDED,
541 ICMP_CODE_TTL_EXCEEDED);
547 iph->check=ip_fast_csum((uint8_t *)iph,iph->ihl);
549 netlink_packet_deliver(st,client,buf);
550 BUF_ASSERT_FREE(buf);
553 /* Deal with packets addressed explicitly to us */
554 static void netlink_packet_local(struct netlink *st,
555 struct netlink_client *client,
556 struct buffer_if *buf)
562 h=(struct icmphdr *)buf->start;
564 if ((ntohs(h->iph.frag_off)&0xbfff)!=0) {
565 Message(M_WARNING,"%s: fragmented packet addressed to secnet; "
566 "ignoring it\n",st->name);
571 if (h->iph.protocol==1) {
573 if (h->type==ICMP_TYPE_ECHO_REQUEST && h->code==0) {
574 /* ICMP echo-request. Special case: we re-use the buffer
575 to construct the reply. */
576 h->type=ICMP_TYPE_ECHO_REPLY;
577 h->iph.daddr=h->iph.saddr;
578 h->iph.saddr=htonl(st->secnet_address);
581 h->iph.check=ip_fast_csum((uint8_t *)h,h->iph.ihl);
582 netlink_icmp_csum(h);
583 netlink_packet_deliver(st,NULL,buf);
586 Message(M_WARNING,"%s: unknown incoming ICMP\n",st->name);
588 /* Send ICMP protocol unreachable */
589 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
590 ICMP_CODE_PROTOCOL_UNREACHABLE);
598 /* If cid==NULL packet is from host, otherwise cid specifies which tunnel
600 static void netlink_incoming(struct netlink *st, struct netlink_client *client,
601 struct buffer_if *buf)
603 uint32_t source,dest;
607 BUF_ASSERT_USED(buf);
608 if (!netlink_check(st,buf,errmsgbuf,sizeof(errmsgbuf))) {
609 Message(M_WARNING,"%s: bad IP packet from %s: %s\n",
610 st->name,client?client->name:"host",
615 iph=(struct iphdr *)buf->start;
617 source=ntohl(iph->saddr);
618 dest=ntohl(iph->daddr);
620 /* Check source. If we don't like the source, there's no point
621 generating ICMP because we won't know how to get it to the
622 source of the packet. */
624 /* Check that the packet source is appropriate for the tunnel
626 if (!ipset_contains_addr(client->networks,source)) {
628 s=ipaddr_to_string(source);
629 d=ipaddr_to_string(dest);
630 Message(M_WARNING,"%s: packet from tunnel %s with bad "
631 "source address (s=%s,d=%s)\n",st->name,client->name,s,d);
637 /* Check that the packet originates in our configured local
638 network, and hasn't been forwarded from elsewhere or
639 generated with the wrong source address */
640 if (!ipset_contains_addr(st->networks,source)) {
642 s=ipaddr_to_string(source);
643 d=ipaddr_to_string(dest);
644 Message(M_WARNING,"%s: outgoing packet with bad source address "
645 "(s=%s,d=%s)\n",st->name,s,d);
652 /* If this is a point-to-point device we don't examine the
653 destination address at all; we blindly send it down our
654 one-and-only registered tunnel, or to the host, depending on
655 where it came from. It's up to external software to check
656 address validity and generate ICMP, etc. */
659 st->deliver_to_host(st->dst,buf);
661 st->clients->deliver(st->clients->dst,buf);
663 BUF_ASSERT_FREE(buf);
667 /* st->secnet_address needs checking before matching destination
669 if (dest==st->secnet_address) {
670 netlink_packet_local(st,client,buf);
671 BUF_ASSERT_FREE(buf);
674 netlink_packet_forward(st,client,buf);
675 BUF_ASSERT_FREE(buf);
678 static void netlink_inst_incoming(void *sst, struct buffer_if *buf)
680 struct netlink_client *c=sst;
681 struct netlink *st=c->nst;
683 netlink_incoming(st,c,buf);
686 static void netlink_dev_incoming(void *sst, struct buffer_if *buf)
688 struct netlink *st=sst;
690 netlink_incoming(st,NULL,buf);
693 static void netlink_set_quality(void *sst, uint32_t quality)
695 struct netlink_client *c=sst;
696 struct netlink *st=c->nst;
698 c->link_quality=quality;
699 c->up=(c->link_quality==LINK_QUALITY_DOWN)?False:True;
700 if (c->options&OPT_SOFTROUTE) {
701 st->set_routes(st->dst,c);
705 static void netlink_output_subnets(struct netlink *st, uint32_t loglevel,
706 struct subnet_list *snets)
711 for (i=0; i<snets->entries; i++) {
712 net=subnet_to_string(snets->list[i]);
713 Message(loglevel,"%s ",net);
718 static void netlink_dump_routes(struct netlink *st, bool_t requested)
724 if (requested) c=M_WARNING;
726 net=ipaddr_to_string(st->secnet_address);
727 Message(c,"%s: point-to-point (remote end is %s); routes:\n",
730 netlink_output_subnets(st,c,st->clients->subnets);
733 Message(c,"%s: routing table:\n",st->name);
734 for (i=0; i<st->n_clients; i++) {
735 netlink_output_subnets(st,c,st->routes[i]->subnets);
736 Message(c,"-> tunnel %s (%s,mtu %d,%s routes,%s,"
737 "quality %d,use %d,pri %lu)\n",
739 st->routes[i]->up?"up":"down",
741 st->routes[i]->options&OPT_SOFTROUTE?"soft":"hard",
742 st->routes[i]->options&OPT_ALLOWROUTE?"free":"restricted",
743 st->routes[i]->link_quality,
744 st->routes[i]->outcount,
745 (unsigned long)st->routes[i]->priority);
747 net=ipaddr_to_string(st->secnet_address);
748 Message(c,"%s/32 -> netlink \"%s\" (use %d)\n",
749 net,st->name,st->localcount);
751 for (i=0; i<st->subnets->entries; i++) {
752 net=subnet_to_string(st->subnets->list[i]);
753 Message(c,"%s ",net);
757 Message(c,"-> host (use %d)\n",st->outcount);
761 /* ap is a pointer to a member of the routes array */
762 static int netlink_compare_client_priority(const void *ap, const void *bp)
764 const struct netlink_client *const*a=ap;
765 const struct netlink_client *const*b=bp;
767 if ((*a)->priority==(*b)->priority) return 0;
768 if ((*a)->priority<(*b)->priority) return 1;
772 static void netlink_phase_hook(void *sst, uint32_t new_phase)
774 struct netlink *st=sst;
775 struct netlink_client *c;
778 /* All the networks serviced by the various tunnels should now
779 * have been registered. We build a routing table by sorting the
780 * clients by priority. */
781 st->routes=safe_malloc_ary(sizeof(*st->routes),st->n_clients,
782 "netlink_phase_hook");
785 for (c=st->clients; c; c=c->next) {
789 /* Sort the table in descending order of priority */
790 qsort(st->routes,st->n_clients,sizeof(*st->routes),
791 netlink_compare_client_priority);
793 netlink_dump_routes(st,False);
796 static void netlink_signal_handler(void *sst, int signum)
798 struct netlink *st=sst;
799 Message(M_INFO,"%s: route dump requested by SIGUSR1\n",st->name);
800 netlink_dump_routes(st,True);
803 static void netlink_inst_set_mtu(void *sst, int32_t new_mtu)
805 struct netlink_client *c=sst;
810 static void netlink_inst_reg(void *sst, netlink_deliver_fn *deliver,
811 void *dst, int32_t max_start_pad,
814 struct netlink_client *c=sst;
815 struct netlink *st=c->nst;
817 if (max_start_pad > st->max_start_pad) st->max_start_pad=max_start_pad;
818 if (max_end_pad > st->max_end_pad) st->max_end_pad=max_end_pad;
823 static struct flagstr netlink_option_table[]={
824 { "soft", OPT_SOFTROUTE },
825 { "allow-route", OPT_ALLOWROUTE },
828 /* This is the routine that gets called when the closure that's
829 returned by an invocation of a netlink device closure (eg. tun,
830 userv-ipif) is invoked. It's used to create routes and pass in
831 information about them; the closure it returns is used by site
833 static closure_t *netlink_inst_create(struct netlink *st,
834 struct cloc loc, dict_t *dict)
836 struct netlink_client *c;
838 struct ipset *networks;
839 uint32_t options,priority;
843 name=dict_read_string(dict, "name", True, st->name, loc);
845 l=dict_lookup(dict,"routes");
847 cfgfatal(loc,st->name,"required parameter \"routes\" not found\n");
848 networks=string_list_to_ipset(l,loc,st->name,"routes");
849 options=string_list_to_word(dict_lookup(dict,"options"),
850 netlink_option_table,st->name);
852 priority=dict_read_number(dict,"priority",False,st->name,loc,0);
853 mtu=dict_read_number(dict,"mtu",False,st->name,loc,0);
855 if ((options&OPT_SOFTROUTE) && !st->set_routes) {
856 cfgfatal(loc,st->name,"this netlink device does not support "
861 if (options&OPT_SOFTROUTE) {
862 /* XXX for now we assume that soft routes require root privilege;
863 this may not always be true. The device driver can tell us. */
864 require_root_privileges=True;
865 require_root_privileges_explanation="netlink: soft routes";
867 cfgfatal(loc,st->name,"point-to-point netlinks do not support "
873 /* Check that nets are a subset of st->remote_networks;
874 refuse to register if they are not. */
875 if (!ipset_is_subset(st->remote_networks,networks)) {
876 cfgfatal(loc,st->name,"routes are not allowed\n");
880 c=safe_malloc(sizeof(*c),"netlink_inst_create");
881 c->cl.description=name;
882 c->cl.type=CL_NETLINK;
884 c->cl.interface=&c->ops;
886 c->ops.reg=netlink_inst_reg;
887 c->ops.deliver=netlink_inst_incoming;
888 c->ops.set_quality=netlink_set_quality;
889 c->ops.set_mtu=netlink_inst_set_mtu;
892 c->networks=networks;
893 c->subnets=ipset_to_subnet_list(networks);
894 c->priority=priority;
898 c->link_quality=LINK_QUALITY_UNUSED;
899 c->mtu=mtu?mtu:st->mtu;
906 assert(st->n_clients < INT_MAX);
912 static list_t *netlink_inst_apply(closure_t *self, struct cloc loc,
913 dict_t *context, list_t *args)
915 struct netlink *st=self->interface;
921 item=list_elem(args,0);
922 if (!item || item->type!=t_dict) {
923 cfgfatal(loc,st->name,"must have a dictionary argument\n");
925 dict=item->data.dict;
927 cl=netlink_inst_create(st,loc,dict);
929 return new_closure(cl);
932 netlink_deliver_fn *netlink_init(struct netlink *st,
933 void *dst, struct cloc loc,
934 dict_t *dict, cstring_t description,
935 netlink_route_fn *set_routes,
936 netlink_deliver_fn *to_host)
942 st->cl.description=description;
944 st->cl.apply=netlink_inst_apply;
951 st->set_routes=set_routes;
952 st->deliver_to_host=to_host;
954 st->name=dict_read_string(dict,"name",False,description,loc);
955 if (!st->name) st->name=description;
956 l=dict_lookup(dict,"networks");
958 st->networks=string_list_to_ipset(l,loc,st->name,"networks");
962 st->networks=ipset_complement(empty);
965 l=dict_lookup(dict,"remote-networks");
967 st->remote_networks=string_list_to_ipset(l,loc,st->name,
972 st->remote_networks=ipset_complement(empty);
976 sa=dict_find_item(dict,"secnet-address",False,"netlink",loc);
977 ptpa=dict_find_item(dict,"ptp-address",False,"netlink",loc);
979 cfgfatal(loc,st->name,"you may not specify secnet-address and "
980 "ptp-address in the same netlink device\n");
983 cfgfatal(loc,st->name,"you must specify secnet-address or "
984 "ptp-address for this netlink device\n");
987 st->secnet_address=string_item_to_ipaddr(sa,"netlink");
990 st->secnet_address=string_item_to_ipaddr(ptpa,"netlink");
993 /* To be strictly correct we could subtract secnet_address from
994 networks here. It shouldn't make any practical difference,
995 though, and will make the route dump look complicated... */
996 st->subnets=ipset_to_subnet_list(st->networks);
997 st->mtu=dict_read_number(dict, "mtu", False, "netlink", loc, DEFAULT_MTU);
998 buffer_new(&st->icmp,ICMP_BUFSIZE);
1002 add_hook(PHASE_SETUP,netlink_phase_hook,st);
1003 request_signal_notification(SIGUSR1, netlink_signal_handler, st);
1005 /* If we're point-to-point then we return a CL_NETLINK directly,
1006 rather than a CL_NETLINK_OLD or pure closure (depending on
1007 compatibility). This CL_NETLINK is for our one and only
1008 client. Our cl.apply function is NULL. */
1011 cl=netlink_inst_create(st,loc,dict);
1014 return netlink_dev_incoming;
1017 /* No connection to the kernel at all... */
1023 static bool_t null_set_route(void *sst, struct netlink_client *routes)
1025 struct null *st=sst;
1027 if (routes->up!=routes->kup) {
1028 Message(M_INFO,"%s: setting routes for tunnel %s to state %s\n",
1029 st->nl.name,routes->name,
1030 routes->up?"up":"down");
1031 routes->kup=routes->up;
1037 static void null_deliver(void *sst, struct buffer_if *buf)
1042 static list_t *null_apply(closure_t *self, struct cloc loc, dict_t *context,
1049 st=safe_malloc(sizeof(*st),"null_apply");
1051 item=list_elem(args,0);
1052 if (!item || item->type!=t_dict)
1053 cfgfatal(loc,"null-netlink","parameter must be a dictionary\n");
1055 dict=item->data.dict;
1057 netlink_init(&st->nl,st,loc,dict,"null-netlink",null_set_route,
1060 return new_closure(&st->nl.cl);
1063 void netlink_module(dict_t *dict)
1065 add_closure(dict,"null-netlink",null_apply);