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. "client" is the _origin_ of the packet, not its
413 destination, and is NULL for packets from the host and packets
414 generated internally in secnet. */
415 static void netlink_packet_deliver(struct netlink *st,
416 struct netlink_client *client,
417 struct buffer_if *buf)
419 if (buf->size < (int)sizeof(struct iphdr)) {
420 Message(M_ERR,"%s: trying to deliver a too-short packet"
421 " from %s!\n",st->name, client?client->name:"(local)");
426 struct iphdr *iph=(struct iphdr *)buf->start;
427 uint32_t dest=ntohl(iph->daddr);
428 uint32_t source=ntohl(iph->saddr);
429 uint32_t best_quality;
430 bool_t allow_route=False;
431 bool_t found_allowed=False;
435 BUF_ASSERT_USED(buf);
437 if (dest==st->secnet_address) {
438 Message(M_ERR,"%s: trying to deliver a packet to myself!\n",st->name);
443 /* Packets from the host (client==NULL) may always be routed. Packets
444 from clients with the allow_route option will also be routed. */
445 if (!client || (client && (client->options & OPT_ALLOWROUTE)))
448 /* If !allow_route, we check the routing table anyway, and if
449 there's a suitable route with OPT_ALLOWROUTE set we use it. If
450 there's a suitable route, but none with OPT_ALLOWROUTE set then
451 we generate ICMP 'communication with destination network
452 administratively prohibited'. */
456 for (i=0; i<st->n_clients; i++) {
457 if (st->routes[i]->up &&
458 ipset_contains_addr(st->routes[i]->networks,dest)) {
459 /* It's an available route to the correct destination. But is
460 it better than the one we already have? */
462 /* If we have already found an allowed route then we don't
463 bother looking at routes we're not allowed to use. If
464 we don't yet have an allowed route we'll consider any. */
465 if (!allow_route && found_allowed) {
466 if (!(st->routes[i]->options&OPT_ALLOWROUTE)) continue;
469 if (st->routes[i]->link_quality>best_quality
470 || best_quality==0) {
471 best_quality=st->routes[i]->link_quality;
473 if (st->routes[i]->options&OPT_ALLOWROUTE)
475 /* If quality isn't perfect we may wish to
476 consider kicking the tunnel with a 0-length
477 packet to prompt it to perform a key setup.
478 Then it'll eventually decide it's up or
480 /* If quality is perfect and we're allowed to use the
481 route we don't need to search any more. */
482 if (best_quality>=MAXIMUM_LINK_QUALITY &&
483 (allow_route || found_allowed)) break;
487 if (best_match==-1) {
488 /* The packet's not going down a tunnel. It might (ought to)
490 if (ipset_contains_addr(st->networks,dest)) {
491 st->deliver_to_host(st->dst,buf);
493 BUF_ASSERT_FREE(buf);
496 s=ipaddr_to_string(source);
497 d=ipaddr_to_string(dest);
498 Message(M_DEBUG,"%s: don't know where to deliver packet "
499 "(s=%s, d=%s)\n", st->name, s, d);
501 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
502 ICMP_CODE_NET_UNREACHABLE);
507 !(st->routes[best_match]->options&OPT_ALLOWROUTE)) {
509 s=ipaddr_to_string(source);
510 d=ipaddr_to_string(dest);
511 /* We have a usable route but aren't allowed to use it.
512 Generate ICMP destination unreachable: communication
513 with destination network administratively prohibited */
514 Message(M_NOTICE,"%s: denied forwarding for packet (s=%s, d=%s)\n",
518 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
519 ICMP_CODE_NET_PROHIBITED);
522 if (best_quality>0) {
523 /* XXX Fragment if required */
524 st->routes[best_match]->deliver(
525 st->routes[best_match]->dst, buf);
526 st->routes[best_match]->outcount++;
527 BUF_ASSERT_FREE(buf);
529 /* Generate ICMP destination unreachable */
530 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
531 ICMP_CODE_NET_UNREACHABLE); /* client==NULL */
536 BUF_ASSERT_FREE(buf);
539 static void netlink_packet_forward(struct netlink *st,
540 struct netlink_client *client,
541 struct buffer_if *buf)
543 if (buf->size < (int)sizeof(struct iphdr)) return;
544 struct iphdr *iph=(struct iphdr *)buf->start;
546 BUF_ASSERT_USED(buf);
548 /* Packet has already been checked */
550 /* Generate ICMP time exceeded */
551 netlink_icmp_simple(st,buf,client,ICMP_TYPE_TIME_EXCEEDED,
552 ICMP_CODE_TTL_EXCEEDED);
558 iph->check=ip_fast_csum((uint8_t *)iph,iph->ihl);
560 netlink_packet_deliver(st,client,buf);
561 BUF_ASSERT_FREE(buf);
564 /* Deal with packets addressed explicitly to us */
565 static void netlink_packet_local(struct netlink *st,
566 struct netlink_client *client,
567 struct buffer_if *buf)
573 if (buf->size < (int)sizeof(struct icmphdr)) {
574 Message(M_WARNING,"%s: short packet addressed to secnet; "
575 "ignoring it\n",st->name);
579 h=(struct icmphdr *)buf->start;
581 if ((ntohs(h->iph.frag_off)&0xbfff)!=0) {
582 Message(M_WARNING,"%s: fragmented packet addressed to secnet; "
583 "ignoring it\n",st->name);
588 if (h->iph.protocol==1) {
590 if (h->type==ICMP_TYPE_ECHO_REQUEST && h->code==0) {
591 /* ICMP echo-request. Special case: we re-use the buffer
592 to construct the reply. */
593 h->type=ICMP_TYPE_ECHO_REPLY;
594 h->iph.daddr=h->iph.saddr;
595 h->iph.saddr=htonl(st->secnet_address);
598 h->iph.check=ip_fast_csum((uint8_t *)h,h->iph.ihl);
599 netlink_icmp_csum(h);
600 netlink_packet_deliver(st,NULL,buf);
603 Message(M_WARNING,"%s: unknown incoming ICMP\n",st->name);
605 /* Send ICMP protocol unreachable */
606 netlink_icmp_simple(st,buf,client,ICMP_TYPE_UNREACHABLE,
607 ICMP_CODE_PROTOCOL_UNREACHABLE);
615 /* If cid==NULL packet is from host, otherwise cid specifies which tunnel
617 static void netlink_incoming(struct netlink *st, struct netlink_client *client,
618 struct buffer_if *buf)
620 uint32_t source,dest;
623 const char *sourcedesc=client?client->name:"host";
625 BUF_ASSERT_USED(buf);
627 if (!netlink_check(st,buf,errmsgbuf,sizeof(errmsgbuf))) {
628 Message(M_WARNING,"%s: bad IP packet from %s: %s\n",
634 assert(buf->size >= (int)sizeof(struct icmphdr));
635 iph=(struct iphdr *)buf->start;
637 source=ntohl(iph->saddr);
638 dest=ntohl(iph->daddr);
640 /* Check source. If we don't like the source, there's no point
641 generating ICMP because we won't know how to get it to the
642 source of the packet. */
644 /* Check that the packet source is appropriate for the tunnel
646 if (!ipset_contains_addr(client->networks,source)) {
648 s=ipaddr_to_string(source);
649 d=ipaddr_to_string(dest);
650 Message(M_WARNING,"%s: packet from tunnel %s with bad "
651 "source address (s=%s,d=%s)\n",st->name,client->name,s,d);
657 /* Check that the packet originates in our configured local
658 network, and hasn't been forwarded from elsewhere or
659 generated with the wrong source address */
660 if (!ipset_contains_addr(st->networks,source)) {
662 s=ipaddr_to_string(source);
663 d=ipaddr_to_string(dest);
664 Message(M_WARNING,"%s: outgoing packet with bad source address "
665 "(s=%s,d=%s)\n",st->name,s,d);
672 /* If this is a point-to-point device we don't examine the
673 destination address at all; we blindly send it down our
674 one-and-only registered tunnel, or to the host, depending on
675 where it came from. It's up to external software to check
676 address validity and generate ICMP, etc. */
679 st->deliver_to_host(st->dst,buf);
681 st->clients->deliver(st->clients->dst,buf);
683 BUF_ASSERT_FREE(buf);
687 /* st->secnet_address needs checking before matching destination
689 if (dest==st->secnet_address) {
690 netlink_packet_local(st,client,buf);
691 BUF_ASSERT_FREE(buf);
694 netlink_packet_forward(st,client,buf);
695 BUF_ASSERT_FREE(buf);
698 static void netlink_inst_incoming(void *sst, struct buffer_if *buf)
700 struct netlink_client *c=sst;
701 struct netlink *st=c->nst;
703 netlink_incoming(st,c,buf);
706 static void netlink_dev_incoming(void *sst, struct buffer_if *buf)
708 struct netlink *st=sst;
710 netlink_incoming(st,NULL,buf);
713 static void netlink_set_quality(void *sst, uint32_t quality)
715 struct netlink_client *c=sst;
716 struct netlink *st=c->nst;
718 c->link_quality=quality;
719 c->up=(c->link_quality==LINK_QUALITY_DOWN)?False:True;
720 if (c->options&OPT_SOFTROUTE) {
721 st->set_routes(st->dst,c);
725 static void netlink_output_subnets(struct netlink *st, uint32_t loglevel,
726 struct subnet_list *snets)
731 for (i=0; i<snets->entries; i++) {
732 net=subnet_to_string(snets->list[i]);
733 Message(loglevel,"%s ",net);
738 static void netlink_dump_routes(struct netlink *st, bool_t requested)
744 if (requested) c=M_WARNING;
746 net=ipaddr_to_string(st->secnet_address);
747 Message(c,"%s: point-to-point (remote end is %s); routes:\n",
750 netlink_output_subnets(st,c,st->clients->subnets);
753 Message(c,"%s: routing table:\n",st->name);
754 for (i=0; i<st->n_clients; i++) {
755 netlink_output_subnets(st,c,st->routes[i]->subnets);
756 Message(c,"-> tunnel %s (%s,mtu %d,%s routes,%s,"
757 "quality %d,use %d,pri %lu)\n",
759 st->routes[i]->up?"up":"down",
761 st->routes[i]->options&OPT_SOFTROUTE?"soft":"hard",
762 st->routes[i]->options&OPT_ALLOWROUTE?"free":"restricted",
763 st->routes[i]->link_quality,
764 st->routes[i]->outcount,
765 (unsigned long)st->routes[i]->priority);
767 net=ipaddr_to_string(st->secnet_address);
768 Message(c,"%s/32 -> netlink \"%s\" (use %d)\n",
769 net,st->name,st->localcount);
771 for (i=0; i<st->subnets->entries; i++) {
772 net=subnet_to_string(st->subnets->list[i]);
773 Message(c,"%s ",net);
777 Message(c,"-> host (use %d)\n",st->outcount);
781 /* ap is a pointer to a member of the routes array */
782 static int netlink_compare_client_priority(const void *ap, const void *bp)
784 const struct netlink_client *const*a=ap;
785 const struct netlink_client *const*b=bp;
787 if ((*a)->priority==(*b)->priority) return 0;
788 if ((*a)->priority<(*b)->priority) return 1;
792 static void netlink_phase_hook(void *sst, uint32_t new_phase)
794 struct netlink *st=sst;
795 struct netlink_client *c;
798 /* All the networks serviced by the various tunnels should now
799 * have been registered. We build a routing table by sorting the
800 * clients by priority. */
801 st->routes=safe_malloc_ary(sizeof(*st->routes),st->n_clients,
802 "netlink_phase_hook");
805 for (c=st->clients; c; c=c->next) {
809 /* Sort the table in descending order of priority */
810 qsort(st->routes,st->n_clients,sizeof(*st->routes),
811 netlink_compare_client_priority);
813 netlink_dump_routes(st,False);
816 static void netlink_signal_handler(void *sst, int signum)
818 struct netlink *st=sst;
819 Message(M_INFO,"%s: route dump requested by SIGUSR1\n",st->name);
820 netlink_dump_routes(st,True);
823 static void netlink_inst_set_mtu(void *sst, int32_t new_mtu)
825 struct netlink_client *c=sst;
830 static void netlink_inst_reg(void *sst, netlink_deliver_fn *deliver,
833 struct netlink_client *c=sst;
839 static struct flagstr netlink_option_table[]={
840 { "soft", OPT_SOFTROUTE },
841 { "allow-route", OPT_ALLOWROUTE },
844 /* This is the routine that gets called when the closure that's
845 returned by an invocation of a netlink device closure (eg. tun,
846 userv-ipif) is invoked. It's used to create routes and pass in
847 information about them; the closure it returns is used by site
849 static closure_t *netlink_inst_create(struct netlink *st,
850 struct cloc loc, dict_t *dict)
852 struct netlink_client *c;
854 struct ipset *networks;
855 uint32_t options,priority;
859 name=dict_read_string(dict, "name", True, st->name, loc);
861 l=dict_lookup(dict,"routes");
863 cfgfatal(loc,st->name,"required parameter \"routes\" not found\n");
864 networks=string_list_to_ipset(l,loc,st->name,"routes");
865 options=string_list_to_word(dict_lookup(dict,"options"),
866 netlink_option_table,st->name);
868 priority=dict_read_number(dict,"priority",False,st->name,loc,0);
869 mtu=dict_read_number(dict,"mtu",False,st->name,loc,0);
871 if ((options&OPT_SOFTROUTE) && !st->set_routes) {
872 cfgfatal(loc,st->name,"this netlink device does not support "
877 if (options&OPT_SOFTROUTE) {
878 /* XXX for now we assume that soft routes require root privilege;
879 this may not always be true. The device driver can tell us. */
880 require_root_privileges=True;
881 require_root_privileges_explanation="netlink: soft routes";
883 cfgfatal(loc,st->name,"point-to-point netlinks do not support "
889 /* Check that nets are a subset of st->remote_networks;
890 refuse to register if they are not. */
891 if (!ipset_is_subset(st->remote_networks,networks)) {
892 cfgfatal(loc,st->name,"routes are not allowed\n");
896 c=safe_malloc(sizeof(*c),"netlink_inst_create");
897 c->cl.description=name;
898 c->cl.type=CL_NETLINK;
900 c->cl.interface=&c->ops;
902 c->ops.reg=netlink_inst_reg;
903 c->ops.deliver=netlink_inst_incoming;
904 c->ops.set_quality=netlink_set_quality;
905 c->ops.set_mtu=netlink_inst_set_mtu;
908 c->networks=networks;
909 c->subnets=ipset_to_subnet_list(networks);
910 c->priority=priority;
914 c->link_quality=LINK_QUALITY_UNUSED;
915 c->mtu=mtu?mtu:st->mtu;
922 assert(st->n_clients < INT_MAX);
928 static list_t *netlink_inst_apply(closure_t *self, struct cloc loc,
929 dict_t *context, list_t *args)
931 struct netlink *st=self->interface;
937 item=list_elem(args,0);
938 if (!item || item->type!=t_dict) {
939 cfgfatal(loc,st->name,"must have a dictionary argument\n");
941 dict=item->data.dict;
943 cl=netlink_inst_create(st,loc,dict);
945 return new_closure(cl);
948 netlink_deliver_fn *netlink_init(struct netlink *st,
949 void *dst, struct cloc loc,
950 dict_t *dict, cstring_t description,
951 netlink_route_fn *set_routes,
952 netlink_deliver_fn *to_host)
958 st->cl.description=description;
960 st->cl.apply=netlink_inst_apply;
965 st->set_routes=set_routes;
966 st->deliver_to_host=to_host;
968 st->name=dict_read_string(dict,"name",False,description,loc);
969 if (!st->name) st->name=description;
970 l=dict_lookup(dict,"networks");
972 st->networks=string_list_to_ipset(l,loc,st->name,"networks");
976 st->networks=ipset_complement(empty);
979 l=dict_lookup(dict,"remote-networks");
981 st->remote_networks=string_list_to_ipset(l,loc,st->name,
986 st->remote_networks=ipset_complement(empty);
990 sa=dict_find_item(dict,"secnet-address",False,"netlink",loc);
991 ptpa=dict_find_item(dict,"ptp-address",False,"netlink",loc);
993 cfgfatal(loc,st->name,"you may not specify secnet-address and "
994 "ptp-address in the same netlink device\n");
997 cfgfatal(loc,st->name,"you must specify secnet-address or "
998 "ptp-address for this netlink device\n");
1001 st->secnet_address=string_item_to_ipaddr(sa,"netlink");
1004 st->secnet_address=string_item_to_ipaddr(ptpa,"netlink");
1007 /* To be strictly correct we could subtract secnet_address from
1008 networks here. It shouldn't make any practical difference,
1009 though, and will make the route dump look complicated... */
1010 st->subnets=ipset_to_subnet_list(st->networks);
1011 st->mtu=dict_read_number(dict, "mtu", False, "netlink", loc, DEFAULT_MTU);
1012 buffer_new(&st->icmp,ICMP_BUFSIZE);
1016 add_hook(PHASE_SETUP,netlink_phase_hook,st);
1017 request_signal_notification(SIGUSR1, netlink_signal_handler, st);
1019 /* If we're point-to-point then we return a CL_NETLINK directly,
1020 rather than a CL_NETLINK_OLD or pure closure (depending on
1021 compatibility). This CL_NETLINK is for our one and only
1022 client. Our cl.apply function is NULL. */
1025 cl=netlink_inst_create(st,loc,dict);
1028 return netlink_dev_incoming;
1031 /* No connection to the kernel at all... */
1037 static bool_t null_set_route(void *sst, struct netlink_client *routes)
1039 struct null *st=sst;
1041 if (routes->up!=routes->kup) {
1042 Message(M_INFO,"%s: setting routes for tunnel %s to state %s\n",
1043 st->nl.name,routes->name,
1044 routes->up?"up":"down");
1045 routes->kup=routes->up;
1051 static void null_deliver(void *sst, struct buffer_if *buf)
1056 static list_t *null_apply(closure_t *self, struct cloc loc, dict_t *context,
1063 st=safe_malloc(sizeof(*st),"null_apply");
1065 item=list_elem(args,0);
1066 if (!item || item->type!=t_dict)
1067 cfgfatal(loc,"null-netlink","parameter must be a dictionary\n");
1069 dict=item->data.dict;
1071 netlink_init(&st->nl,st,loc,dict,"null-netlink",null_set_route,
1074 return new_closure(&st->nl.cl);
1077 void netlink_module(dict_t *dict)
1079 add_closure(dict,"null-netlink",null_apply);