X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=secnet.git;a=blobdiff_plain;f=netlink.c;h=601686425124cb7d7bb46cb03500f9df75e7c909;hp=0cbb27d10a09b025e7341cd2c65709b0b1477321;hb=46d06c3918b0080af7e87db986612e23bde4a341;hpb=baa06aeb963965b4b6a8a8051ec15b72372080dd

diff --git a/netlink.c b/netlink.c
index 0cbb27d..6016864 100644
--- a/netlink.c
+++ b/netlink.c
@@ -1,80 +1,132 @@
 /* User-kernel network link */
 
-/* We support a variety of methods for extracting packets from the
-   kernel: userv-ipif, ipif on its own (when we run as root), the
-   kernel TUN driver.  Possible future methods: SLIP to a pty, an
-   external netlink daemon.  There is a performance/security
-   tradeoff. */
+/* See RFCs 791, 792, 1123 and 1812 */
 
-/* When dealing with SLIP (to a pty, or ipif) we have separate rx, tx
-   and client buffers.  When receiving we may read() any amount, not
-   just whole packets.  When transmitting we need to bytestuff anyway,
-   and may be part-way through receiving. */
+/* The netlink device is actually a router.  Tunnels are unnumbered
+   point-to-point lines (RFC1812 section 2.2.7); the router has a
+   single address (the 'router-id'). */
 
-/* Each netlink device is actually a router, with its own IP address.
-   We do things like decreasing the TTL and recalculating the header
-   checksum, generating ICMP, responding to pings, etc. */
+/* This is where we currently have the anti-spoofing paranoia - before
+   sending a packet to the kernel we check that the tunnel it came
+   over could reasonably have produced it. */
 
-/* This is where we have the anti-spoofing paranoia - before sending a
-   packet to the kernel we check that the tunnel it came over could
-   reasonably have produced it. */
 
-#include "secnet.h"
-#include <stdio.h>
-#include <string.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <sys/ioctl.h>
-#include "util.h"
+/* Points to note from RFC1812 (which may require changes in this
+   file):
 
-#ifdef HAVE_LINUX_IF_H
-#include <linux/if.h>
-#include <linux/if_tun.h>
-#endif
+3.3.4 Maximum Transmission Unit - MTU
 
-/* XXX where do we find if_tun on other platforms? */
+   The MTU of each logical interface MUST be configurable within the
+   range of legal MTUs for the interface.
 
-#define DEFAULT_BUFSIZE 2048
-#define DEFAULT_MTU 1000
-#define ICMP_BUFSIZE 1024
+   Many Link Layer protocols define a maximum frame size that may be
+   sent.  In such cases, a router MUST NOT allow an MTU to be set which
+   would allow sending of frames larger than those allowed by the Link
+   Layer protocol.  However, a router SHOULD be willing to receive a
+   packet as large as the maximum frame size even if that is larger than
+   the MTU.
 
-#define SLIP_END    192
-#define SLIP_ESC    219
-#define SLIP_ESCEND 220
-#define SLIP_ESCESC 221
+4.2.1  A router SHOULD count datagrams discarded.
 
-struct netlink_client {
-    struct subnet_list *networks;
-    netlink_deliver_fn *deliver;
-    void *dst;
-    string_t name;
-    bool_t can_deliver;
-    struct netlink_client *next;
-};
+4.2.2.1 Source route options - we probably should implement processing
+of source routes, even though mostly the security policy will prevent
+their use.
 
-/* Netlink provides one function to the device driver, to call to deliver
-   a packet from the device. The device driver provides one function to
-   netlink, for it to call to deliver a packet to the device. */
+5.3.13.4 Source Route Options
 
-struct netlink {
-    closure_t cl;
-    struct netlink_if ops;
-    void *dst; /* Pointer to host interface state */
-    string_t name;
-    uint32_t max_start_pad;
-    uint32_t max_end_pad;
-    struct subnet_list networks;
-    struct subnet_list exclude_remote_networks;
-    uint32_t local_address; /* host interface address */
-    uint32_t secnet_address; /* our own address */
-    uint32_t mtu;
-    struct netlink_client *clients;
-    netlink_deliver_fn *deliver_to_host; /* Provided by driver */
-    struct buffer_if icmp; /* Buffer for assembly of outgoing ICMP */
-};
+   A router MUST implement support for source route options in forwarded
+   packets.  A router MAY implement a configuration option that, when
+   enabled, causes all source-routed packets to be discarded.  However,
+   such an option MUST NOT be enabled by default.
+
+5.3.13.5 Record Route Option
+
+   Routers MUST support the Record Route option in forwarded packets.
+
+   A router MAY provide a configuration option that, if enabled, will
+   cause the router to ignore (i.e., pass through unchanged) Record
+   Route options in forwarded packets.  If provided, such an option MUST
+   default to enabling the record-route.  This option should not affect
+   the processing of Record Route options in datagrams received by the
+   router itself (in particular, Record Route options in ICMP echo
+   requests will still be processed according to Section [4.3.3.6]).
+
+5.3.13.6 Timestamp Option
+
+   Routers MUST support the timestamp option in forwarded packets.  A
+   timestamp value MUST follow the rules given [INTRO:2].
+
+   If the flags field = 3 (timestamp and prespecified address), the
+   router MUST add its timestamp if the next prespecified address
+   matches any of the router's IP addresses.  It is not necessary that
+   the prespecified address be either the address of the interface on
+   which the packet arrived or the address of the interface over which
+   it will be sent.
+
+
+4.2.2.7 Fragmentation: RFC 791 Section 3.2
+
+   Fragmentation, as described in [INTERNET:1], MUST be supported by a
+   router.
+
+4.2.2.8 Reassembly: RFC 791 Section 3.2
+
+   As specified in the corresponding section of [INTRO:2], a router MUST
+   support reassembly of datagrams that it delivers to itself.
+
+4.2.2.9 Time to Live: RFC 791 Section 3.2
+
+   Note in particular that a router MUST NOT check the TTL of a packet
+   except when forwarding it.
+
+   A router MUST NOT discard a datagram just because it was received
+   with TTL equal to zero or one; if it is to the router and otherwise
+   valid, the router MUST attempt to receive it.
+
+   On messages the router originates, the IP layer MUST provide a means
+   for the transport layer to set the TTL field of every datagram that
+   is sent.  When a fixed TTL value is used, it MUST be configurable.
+
+
+8.1 The Simple Network Management Protocol - SNMP
+8.1.1 SNMP Protocol Elements
+
+   Routers MUST be manageable by SNMP [MGT:3].  The SNMP MUST operate
+   using UDP/IP as its transport and network protocols.
+
+
+*/
+
+#include <string.h>
+#include <assert.h>
+#include <limits.h>
+#include "secnet.h"
+#include "util.h"
+#include "ipaddr.h"
+#include "netlink.h"
+#include "process.h"
+
+#ifdef NETLINK_DEBUG
+#define MDEBUG(...) Message(M_DEBUG, __VA_ARGS__)
+#else /* !NETLINK_DEBUG */
+#define MDEBUG(...) ((void)0)
+#endif /* !NETLINK_DEBUG */
+
+#define ICMP_TYPE_ECHO_REPLY             0
+
+#define ICMP_TYPE_UNREACHABLE            3
+#define ICMP_CODE_NET_UNREACHABLE        0
+#define ICMP_CODE_PROTOCOL_UNREACHABLE   2
+#define ICMP_CODE_FRAGMENTATION_REQUIRED 4
+#define ICMP_CODE_NET_PROHIBITED        13
+
+#define ICMP_TYPE_ECHO_REQUEST           8
+
+#define ICMP_TYPE_TIME_EXCEEDED         11
+#define ICMP_CODE_TTL_EXCEEDED           0
 
 /* Generic IP checksum routine */
-static inline uint16_t ip_csum(uint8_t *iph,uint32_t count)
+static inline uint16_t ip_csum(const uint8_t *iph,int32_t count)
 {
     register uint32_t sum=0;
 
@@ -98,38 +150,39 @@ static inline uint16_t ip_csum(uint8_t *iph,uint32_t count)
  *      By Jorge Cwik <jorge@laser.satlink.net>, adapted for linux by
  *      Arnt Gulbrandsen.
  */
-static inline uint16_t ip_fast_csum(uint8_t *iph, uint32_t ihl) {
+static inline uint16_t ip_fast_csum(const uint8_t *iph, int32_t ihl) {
     uint32_t sum;
 
-    __asm__ __volatile__("
-            movl (%1), %0
-            subl $4, %2
-            jbe 2f
-            addl 4(%1), %0
-            adcl 8(%1), %0
-            adcl 12(%1), %0
-1:          adcl 16(%1), %0
-            lea 4(%1), %1
-            decl %2
-            jne 1b
-            adcl $0, %0
-            movl %0, %2
-            shrl $16, %0
-            addw %w2, %w0
-            adcl $0, %0
-            notl %0
-2:
-            "
+    __asm__ __volatile__(
+            "movl (%1), %0      ;\n"
+            "subl $4, %2        ;\n"
+            "jbe 2f             ;\n"
+            "addl 4(%1), %0     ;\n"
+            "adcl 8(%1), %0     ;\n"
+            "adcl 12(%1), %0    ;\n"
+"1:         adcl 16(%1), %0     ;\n"
+            "lea 4(%1), %1      ;\n"
+            "decl %2            ;\n"
+            "jne 1b             ;\n"
+            "adcl $0, %0        ;\n"
+            "movl %0, %2        ;\n"
+            "shrl $16, %0       ;\n"
+            "addw %w2, %w0      ;\n"
+            "adcl $0, %0        ;\n"
+            "notl %0            ;\n"
+"2:                             ;\n"
         /* Since the input registers which are loaded with iph and ipl
            are modified, we must also specify them as outputs, or gcc
            will assume they contain their original values. */
         : "=r" (sum), "=r" (iph), "=r" (ihl)
-        : "1" (iph), "2" (ihl));
+        : "1" (iph), "2" (ihl)
+	: "memory");
     return sum;
 }
 #else
-static inline uint16_t ip_fast_csum(uint8_t *iph, uint32_t ihl)
+static inline uint16_t ip_fast_csum(const uint8_t *iph, int32_t ihl)
 {
+    assert(ihl < INT_MAX/4);
     return ip_csum(iph,ihl*4);
 }
 #endif
@@ -145,7 +198,11 @@ struct iphdr {
     uint8_t    tos;
     uint16_t   tot_len;
     uint16_t   id;
-    uint16_t   frag_off;
+    uint16_t   frag;
+#define IPHDR_FRAG_OFF  ((uint16_t)0x1fff)
+#define IPHDR_FRAG_MORE ((uint16_t)0x2000)
+#define IPHDR_FRAG_DONT ((uint16_t)0x4000)
+/*                 reserved        0x8000 */
     uint8_t    ttl;
     uint8_t    protocol;
     uint16_t   check;
@@ -159,7 +216,7 @@ struct icmphdr {
     uint8_t type;
     uint8_t code;
     uint16_t check;
-    union {
+    union icmpinfofield {
 	uint32_t unused;
 	struct {
 	    uint8_t pointer;
@@ -171,18 +228,48 @@ struct icmphdr {
 	    uint16_t id;
 	    uint16_t seq;
 	} echo;
+	struct {
+	    uint16_t unused;
+	    uint16_t mtu;
+	} fragneeded;
     } d;
 };
+
+static const union icmpinfofield icmp_noinfo;
     
-static void netlink_packet_deliver(struct netlink *st, struct buffer_if *buf);
+static void netlink_client_deliver(struct netlink *st,
+				   struct netlink_client *client,
+				   uint32_t source, uint32_t dest,
+				   struct buffer_if *buf);
+static void netlink_host_deliver(struct netlink *st,
+				 struct netlink_client *sender,
+				 uint32_t source, uint32_t dest,
+				 struct buffer_if *buf);
+
+static const char *sender_name(struct netlink_client *sender /* or NULL */)
+{
+    return sender?sender->name:"(local)";
+}
 
+static void netlink_packet_deliver(struct netlink *st,
+				   struct netlink_client *client,
+				   struct buffer_if *buf);
+
+/* XXX RFC1812 4.3.2.5:
+   All other ICMP error messages (Destination Unreachable,
+   Redirect, Time Exceeded, and Parameter Problem) SHOULD have their
+   precedence value set to 6 (INTERNETWORK CONTROL) or 7 (NETWORK
+   CONTROL).  The IP Precedence value for these error messages MAY be
+   settable.
+   */
 static struct icmphdr *netlink_icmp_tmpl(struct netlink *st,
-					 uint32_t dest,uint16_t len)
+					 uint32_t source, uint32_t dest,
+					 uint16_t len)
 {
     struct icmphdr *h;
 
     BUF_ALLOC(&st->icmp,"netlink_icmp_tmpl");
-    buffer_init(&st->icmp,st->max_start_pad);
+    buffer_init(&st->icmp,calculate_max_start_pad());
     h=buf_append(&st->icmp,sizeof(*h));
 
     h->iph.version=4;
@@ -190,10 +277,10 @@ static struct icmphdr *netlink_icmp_tmpl(struct netlink *st,
     h->iph.tos=0;
     h->iph.tot_len=htons(len+(h->iph.ihl*4)+8);
     h->iph.id=0;
-    h->iph.frag_off=0;
-    h->iph.ttl=255;
+    h->iph.frag=0;
+    h->iph.ttl=255; /* XXX should be configurable */
     h->iph.protocol=1;
-    h->iph.saddr=htonl(st->secnet_address);
+    h->iph.saddr=htonl(source);
     h->iph.daddr=htonl(dest);
     h->iph.check=0;
     h->iph.check=ip_fast_csum((uint8_t *)&h->iph,h->iph.ihl);
@@ -206,7 +293,7 @@ static struct icmphdr *netlink_icmp_tmpl(struct netlink *st,
 /* Fill in the ICMP checksum field correctly */
 static void netlink_icmp_csum(struct icmphdr *h)
 {
-    uint32_t len;
+    int32_t len;
 
     len=ntohs(h->iph.tot_len)-(4*h->iph.ihl);
     h->check=0;
@@ -234,13 +321,30 @@ static void netlink_icmp_csum(struct icmphdr *h)
 static bool_t netlink_icmp_may_reply(struct buffer_if *buf)
 {
     struct iphdr *iph;
+    struct icmphdr *icmph;
     uint32_t source;
 
+    if (buf->size < (int)sizeof(struct icmphdr)) return False;
     iph=(struct iphdr *)buf->start;
-    if (iph->protocol==1) return False; /* Overly-broad; we may reply to
-					   eg. icmp echo-request */
+    icmph=(struct icmphdr *)buf->start;
+    if (iph->protocol==1) {
+	switch(icmph->type) {
+	    /* Based on http://www.iana.org/assignments/icmp-parameters/icmp-parameters.xhtml#icmp-parameters-types
+	     * as retrieved Thu, 20 Mar 2014 00:16:44 +0000.
+	     * Deprecated, reserved, unassigned and experimental
+	     * options are treated as not safe to reply to.
+	     */
+	case 0: /* Echo Reply */
+	case 8: /* Echo */
+	case 13: /* Timestamp */
+	case 14: /* Timestamp Reply */
+	    return True;
+	default:
+	    return False;
+	}
+    }
     /* How do we spot broadcast destination addresses? */
-    if (ntohs(iph->frag_off)&0x1fff) return False; /* Non-initial fragment */
+    if (ntohs(iph->frag)&IPHDR_FRAG_OFF) return False;
     source=ntohl(iph->saddr);
     if (source==0) return False;
     if ((source&0xff000000)==0x7f000000) return False;
@@ -252,8 +356,29 @@ static bool_t netlink_icmp_may_reply(struct buffer_if *buf)
 
 /* How much of the original IP packet do we include in its ICMP
    response? The header plus up to 64 bits. */
+
+/* XXX TODO RFC1812:
+4.3.2.3 Original Message Header
+
+   Historically, every ICMP error message has included the Internet
+   header and at least the first 8 data bytes of the datagram that
+   triggered the error.  This is no longer adequate, due to the use of
+   IP-in-IP tunneling and other technologies.  Therefore, the ICMP
+   datagram SHOULD contain as much of the original datagram as possible
+   without the length of the ICMP datagram exceeding 576 bytes.  The
+   returned IP header (and user data) MUST be identical to that which
+   was received, except that the router is not required to undo any
+   modifications to the IP header that are normally performed in
+   forwarding that were performed before the error was detected (e.g.,
+   decrementing the TTL, or updating options).  Note that the
+   requirements of Section [4.3.3.5] supersede this requirement in some
+   cases (i.e., for a Parameter Problem message, if the problem is in a
+   modified field, the router must undo the modification).  See Section
+   [4.3.3.5]).
+   */
 static uint16_t netlink_icmp_reply_len(struct buffer_if *buf)
 {
+    if (buf->size < (int)sizeof(struct iphdr)) return 0;
     struct iphdr *iph=(struct iphdr *)buf->start;
     uint16_t hlen,plen;
 
@@ -261,23 +386,60 @@ static uint16_t netlink_icmp_reply_len(struct buffer_if *buf)
     /* We include the first 8 bytes of the packet data, provided they exist */
     hlen+=8;
     plen=ntohs(iph->tot_len);
-    return (hlen>plen?plen:hlen);
+    return MIN(hlen,plen);
 }
 
-static void netlink_icmp_simple(struct netlink *st, struct buffer_if *buf,
-				uint8_t type, uint8_t code)
+/* client indicates where the packet we're constructing a response to
+   comes from. NULL indicates the host. */
+static void netlink_icmp_simple(struct netlink *st,
+				struct netlink_client *origsender,
+				struct buffer_if *buf,
+				uint8_t type, uint8_t code,
+				union icmpinfofield info)
 {
-    struct iphdr *iph=(struct iphdr *)buf->start;
     struct icmphdr *h;
     uint16_t len;
 
     if (netlink_icmp_may_reply(buf)) {
+	struct iphdr *iph=(struct iphdr *)buf->start;
+
+	uint32_t icmpdest = ntohl(iph->saddr);
+	uint32_t icmpsource;
+	const char *icmpsourcedebugprefix;
+	if (!st->ptp) {
+	    icmpsource=st->secnet_address;
+	    icmpsourcedebugprefix="";
+	} else if (origsender) {
+	    /* was from peer, send reply as if from host */
+	    icmpsource=st->local_address;
+	    icmpsourcedebugprefix="L!";
+	} else {
+	    /* was from host, send reply as if from peer */
+	    icmpsource=st->secnet_address; /* actually, peer address */
+	    icmpsourcedebugprefix="P!";
+	}
+	MDEBUG("%s: generating ICMP re %s[%s]->[%s]:"
+	       " from %s%s type=%u code=%u\n",
+	       st->name, sender_name(origsender),
+	       ipaddr_to_string(ntohl(iph->saddr)),
+	       ipaddr_to_string(ntohl(iph->daddr)),
+	       icmpsourcedebugprefix,
+	       ipaddr_to_string(icmpsource),
+	       type, code);
+
 	len=netlink_icmp_reply_len(buf);
-	h=netlink_icmp_tmpl(st,ntohl(iph->saddr),len);
-	h->type=type; h->code=code;
-	memcpy(buf_append(&st->icmp,len),buf->start,len);
+	h=netlink_icmp_tmpl(st,icmpsource,icmpdest,len);
+	h->type=type; h->code=code; h->d=info;
+	BUF_ADD_BYTES(append,&st->icmp,buf->start,len);
 	netlink_icmp_csum(h);
-	netlink_packet_deliver(st,&st->icmp);
+
+	if (!st->ptp) {
+	    netlink_packet_deliver(st,NULL,&st->icmp);
+	} else if (origsender) {
+	    netlink_client_deliver(st,origsender,icmpsource,icmpdest,&st->icmp);
+	} else {
+	    netlink_host_deliver(st,NULL,icmpsource,icmpdest,&st->icmp);
+	}
 	BUF_ASSERT_FREE(&st->icmp);
     }
 }
@@ -285,6 +447,7 @@ static void netlink_icmp_simple(struct netlink *st, struct buffer_if *buf,
 /*
  * RFC1122: 3.1.2.2 MUST silently discard any IP frame that fails the
  * checksum.
+ * RFC1812: 4.2.2.5 MUST discard messages containing invalid checksums.
  *
  * Is the datagram acceptable?
  *
@@ -293,73 +456,335 @@ static void netlink_icmp_simple(struct netlink *st, struct buffer_if *buf,
  * 3. Checksums correctly.
  * 4. Doesn't have a bogus length
  */
-static bool_t netlink_check(struct netlink *st, struct buffer_if *buf)
+static bool_t netlink_check(struct netlink *st, struct buffer_if *buf,
+			    char *errmsgbuf, int errmsgbuflen)
 {
+#define BAD(...) do{					\
+	snprintf(errmsgbuf,errmsgbuflen,__VA_ARGS__);	\
+	return False;					\
+    }while(0)
+
+    if (buf->size < (int)sizeof(struct iphdr)) BAD("len %"PRIu32"",buf->size);
     struct iphdr *iph=(struct iphdr *)buf->start;
-    uint32_t len;
+    int32_t len;
+
+    if (iph->ihl < 5) BAD("ihl %u",iph->ihl);
+    if (iph->version != 4) BAD("version %u",iph->version);
+    if (buf->size < iph->ihl*4) BAD("size %"PRId32"<%u*4",buf->size,iph->ihl);
+    if (ip_fast_csum((uint8_t *)iph, iph->ihl)!=0) BAD("csum");
+    len=ntohs(iph->tot_len);
+    /* There should be no padding */
+    if (buf->size!=len) BAD("len %"PRId32"!=%"PRId32,buf->size,len);
+    if (len<(iph->ihl<<2)) BAD("len %"PRId32"<(%u<<2)",len,iph->ihl);
+    /* XXX check that there's no source route specified */
+    return True;
+
+#undef BAD
+}
 
-    if (iph->ihl < 5 || iph->version != 4) {
-	printf("ihl/version check failed\n");
-	return False;
+static const char *fragment_filter_header(uint8_t *base, long *hlp)
+{
+    const int fixedhl = sizeof(struct iphdr);
+    long hl = *hlp;
+    const uint8_t *ipend = base + hl;
+    uint8_t *op = base + fixedhl;
+    const uint8_t *ip = op;
+
+    while (ip < ipend) {
+	uint8_t opt = ip[0];
+	int remain = ipend - ip;
+	if (opt == 0x00) /* End of Options List */ break;
+	if (opt == 0x01) /* No Operation */ continue;
+	if (remain < 2) return "IPv4 options truncated at length";
+	int optlen = ip[1];
+	if (remain < optlen) return "IPv4 options truncated in option";
+	if (opt & 0x80) /* copy */ {
+	    memmove(op, ip, optlen);
+	    op += optlen;
+	}
+	ip += optlen;
     }
-    if (buf->size < iph->ihl*4) {
-	printf("buffer size check failed\n");
-	return False;
+    while ((hl = (op - base)) & 0x3)
+	*op++ = 0x00 /* End of Option List */;
+    ((struct iphdr*)base)->ihl = hl >> 2;
+    *hlp = hl;
+
+    return 0;
+}
+
+/* Fragment or send ICMP Fragmentation Needed */
+static void netlink_maybe_fragment(struct netlink *st,
+				   struct netlink_client *sender,
+				   netlink_deliver_fn *deliver,
+				   void *deliver_dst,
+				   const char *delivery_name,
+				   int32_t mtu,
+				   uint32_t source, uint32_t dest,
+				   struct buffer_if *buf)
+{
+    struct iphdr *iph=(struct iphdr*)buf->start;
+    long hl = iph->ihl*4;
+    const char *ssource = ipaddr_to_string(source);
+
+    if (buf->size <= mtu) {
+	deliver(deliver_dst, buf);
+	return;
     }
-    if (ip_fast_csum((uint8_t *)iph, iph->ihl)!=0) {
-	printf("checksum failed\n");
-	return False;
+
+    MDEBUG("%s: fragmenting %s->%s org.size=%"PRId32"\n",
+	   st->name, ssource, delivery_name, buf->size);
+
+#define BADFRAG(m, ...)					\
+	Message(M_WARNING,				\
+	        "%s: fragmenting packet from source %s"	\
+		" for transmission via %s: " m "\n",	\
+		st->name, ssource, delivery_name,	\
+		## __VA_ARGS__);
+
+    unsigned orig_frag = ntohs(iph->frag);
+
+    if (orig_frag&IPHDR_FRAG_DONT) {
+	union icmpinfofield info =
+	    { .fragneeded = { .unused = 0, .mtu = htons(mtu) } };
+	netlink_icmp_simple(st,sender,buf,
+			    ICMP_TYPE_UNREACHABLE,
+			    ICMP_CODE_FRAGMENTATION_REQUIRED,
+			    info);
+	BUF_FREE(buf);
+	return;
     }
-    len=ntohs(iph->tot_len);
-    /* There should be no padding */
-    if (buf->size!=len || len<(iph->ihl<<2)) {
-	printf("length check failed buf->size=%d len=%d\n",buf->size,len);
-	return False;
+    if (mtu < hl + 8) {
+	BADFRAG("mtu %"PRId32" too small", mtu);
+	BUF_FREE(buf);
+	return;
     }
 
-    /* XXX check that there's no source route specified */
-    return True;
+    /* we (ab)use the icmp buffer to stash the original packet */
+    struct buffer_if *orig = &st->icmp;
+    BUF_ALLOC(orig,"netlink_client_deliver fragment orig");
+    buffer_copy(orig,buf);
+    BUF_FREE(buf);
+
+    const uint8_t *startindata = orig->start + hl;
+    const uint8_t *indata =      startindata;
+    const uint8_t *endindata =   orig->start + orig->size;
+    _Bool filtered = 0;
+
+    for (;;) {
+	/* compute our fragment offset */
+	long dataoffset = indata - startindata
+	    + (orig_frag & IPHDR_FRAG_OFF)*8;
+	assert(!(dataoffset & 7));
+	if (dataoffset > IPHDR_FRAG_OFF*8) {
+	    BADFRAG("ultimate fragment offset out of range");
+	    break;
+	}
+
+	BUF_ALLOC(buf,"netlink_client_deliver fragment frag");
+	buffer_init(buf,calculate_max_start_pad());
+
+	/* copy header (possibly filtered); will adjust in a bit */
+	struct iphdr *fragh = buf_append(buf, hl);
+	memcpy(fragh, orig->start, hl);
+
+	/* decide how much payload to copy and copy it */
+	long avail = mtu - hl;
+	long remain = endindata - indata;
+	long use = avail < remain ? (avail & ~(long)7) : remain;
+	BUF_ADD_BYTES(append, buf, indata, use);
+	indata += use;
+
+	_Bool last_frag = indata >= endindata;
+
+	/* adjust the header */
+	fragh->tot_len = htons(buf->size);
+	fragh->frag =
+	    htons((orig_frag & ~IPHDR_FRAG_OFF) |
+		  (last_frag ? 0 : IPHDR_FRAG_MORE) |
+		  (dataoffset >> 3));
+	fragh->check = 0;
+	fragh->check = ip_fast_csum((const void*)fragh, fragh->ihl);
+
+	/* actually send it */
+	deliver(deliver_dst, buf);
+	if (last_frag)
+	    break;
+
+	/* after copying the header for the first frag,
+	 * we filter the header for the remaining frags */
+	if (!filtered++) {
+	    const char *bad = fragment_filter_header(orig->start, &hl);
+	    if (bad) { BADFRAG("%s", bad); break; }
+	}
+    }
+
+    BUF_FREE(orig);
+
+#undef BADFRAG
 }
 
-static void netlink_packet_deliver(struct netlink *st, struct buffer_if *buf)
+/* Deliver a packet _to_ client; used after we have decided
+ * what to do with it (and just to check that the client has
+ * actually registered a delivery function with us). */
+static void netlink_client_deliver(struct netlink *st,
+				   struct netlink_client *client,
+				   uint32_t source, uint32_t dest,
+				   struct buffer_if *buf)
 {
+    if (!client->deliver) {
+	string_t s,d;
+	s=ipaddr_to_string(source);
+	d=ipaddr_to_string(dest);
+	Message(M_ERR,"%s: dropping %s->%s, client not registered\n",
+		st->name,s,d);
+	BUF_FREE(buf);
+	return;
+    }
+    netlink_maybe_fragment(st,NULL, client->deliver,client->dst,client->name,
+			   client->mtu, source,dest,buf);
+    client->outcount++;
+}
+
+/* Deliver a packet to the host; used after we have decided that that
+ * is what to do with it. */
+static void netlink_host_deliver(struct netlink *st,
+				 struct netlink_client *sender,
+				 uint32_t source, uint32_t dest,
+				 struct buffer_if *buf)
+{
+    netlink_maybe_fragment(st,sender, st->deliver_to_host,st->dst,"(host)",
+			   st->mtu, source,dest,buf);
+    st->outcount++;
+}
+
+/* Deliver a packet. "sender"==NULL for packets from the host and packets
+   generated internally in secnet.  */
+static void netlink_packet_deliver(struct netlink *st,
+				   struct netlink_client *sender,
+				   struct buffer_if *buf)
+{
+    if (buf->size < (int)sizeof(struct iphdr)) {
+	Message(M_ERR,"%s: trying to deliver a too-short packet"
+		" from %s!\n",st->name, sender_name(sender));
+	BUF_FREE(buf);
+	return;
+    }
+
     struct iphdr *iph=(struct iphdr *)buf->start;
     uint32_t dest=ntohl(iph->daddr);
-    struct netlink_client *c;
+    uint32_t source=ntohl(iph->saddr);
+    uint32_t best_quality;
+    bool_t allow_route=False;
+    bool_t found_allowed=False;
+    int best_match;
+    int i;
 
     BUF_ASSERT_USED(buf);
 
     if (dest==st->secnet_address) {
-	Message(M_ERROR,"%s: trying to deliver a packet to myself!\n");
+	Message(M_ERR,"%s: trying to deliver a packet to myself!\n",st->name);
 	BUF_FREE(buf);
 	return;
     }
     
-    for (c=st->clients; c; c=c->next) {
-	if (subnet_match(c->networks,dest)) {
-	    if (c->can_deliver) {
-		c->deliver(c->dst,c,buf);
+    /* Packets from the host (sender==NULL) may always be routed.  Packets
+       from clients with the allow_route option will also be routed. */
+    if (!sender || (sender && (sender->options & OPT_ALLOWROUTE)))
+	allow_route=True;
+
+    /* If !allow_route, we check the routing table anyway, and if
+       there's a suitable route with OPT_ALLOWROUTE set we use it.  If
+       there's a suitable route, but none with OPT_ALLOWROUTE set then
+       we generate ICMP 'communication with destination network
+       administratively prohibited'. */
+
+    best_quality=0;
+    best_match=-1;
+    for (i=0; i<st->n_clients; i++) {
+	if (st->routes[i]->up &&
+	    ipset_contains_addr(st->routes[i]->networks,dest)) {
+	    /* It's an available route to the correct destination. But is
+	       it better than the one we already have? */
+
+	    /* If we have already found an allowed route then we don't
+	       bother looking at routes we're not allowed to use.  If
+	       we don't yet have an allowed route we'll consider any.  */
+	    if (!allow_route && found_allowed) {
+		if (!(st->routes[i]->options&OPT_ALLOWROUTE)) continue;
+	    }
+	    
+	    if (st->routes[i]->link_quality>best_quality
+		|| best_quality==0) {
+		best_quality=st->routes[i]->link_quality;
+		best_match=i;
+		if (st->routes[i]->options&OPT_ALLOWROUTE)
+		    found_allowed=True;
+		/* If quality isn't perfect we may wish to
+		   consider kicking the tunnel with a 0-length
+		   packet to prompt it to perform a key setup.
+		   Then it'll eventually decide it's up or
+		   down. */
+		/* If quality is perfect and we're allowed to use the
+		   route we don't need to search any more. */
+		if (best_quality>=MAXIMUM_LINK_QUALITY && 
+		    (allow_route || found_allowed)) break;
+	    }
+	}
+    }
+    if (best_match==-1) {
+	/* The packet's not going down a tunnel.  It might (ought to)
+	   be for the host.   */
+	if (ipset_contains_addr(st->networks,dest)) {
+	    netlink_host_deliver(st,sender,source,dest,buf);
+	    BUF_ASSERT_FREE(buf);
+	} else {
+	    string_t s,d;
+	    s=ipaddr_to_string(source);
+	    d=ipaddr_to_string(dest);
+	    Message(M_DEBUG,"%s: don't know where to deliver packet "
+		    "(s=%s, d=%s)\n", st->name, s, d);
+	    netlink_icmp_simple(st,sender,buf,ICMP_TYPE_UNREACHABLE,
+				ICMP_CODE_NET_UNREACHABLE, icmp_noinfo);
+	    BUF_FREE(buf);
+	}
+    } else {
+	if (!allow_route &&
+	    !(st->routes[best_match]->options&OPT_ALLOWROUTE)) {
+	    string_t s,d;
+	    s=ipaddr_to_string(source);
+	    d=ipaddr_to_string(dest);
+	    /* We have a usable route but aren't allowed to use it.
+	       Generate ICMP destination unreachable: communication
+	       with destination network administratively prohibited */
+	    Message(M_NOTICE,"%s: denied forwarding for packet (s=%s, d=%s)\n",
+		    st->name,s,d);
+		    
+	    netlink_icmp_simple(st,sender,buf,ICMP_TYPE_UNREACHABLE,
+				ICMP_CODE_NET_PROHIBITED, icmp_noinfo);
+	    BUF_FREE(buf);
+	} else {
+	    if (best_quality>0) {
+		netlink_client_deliver(st,st->routes[best_match],
+				       source,dest,buf);
 		BUF_ASSERT_FREE(buf);
 	    } else {
 		/* Generate ICMP destination unreachable */
-		netlink_icmp_simple(st,buf,3,0);
+		netlink_icmp_simple(st,sender,buf,
+				    ICMP_TYPE_UNREACHABLE,
+				    ICMP_CODE_NET_UNREACHABLE,
+				    icmp_noinfo);
 		BUF_FREE(buf);
 	    }
-	    return;
 	}
     }
-    if (subnet_match(&st->networks,dest)) {
-	st->deliver_to_host(st->dst,NULL,buf);
-	BUF_ASSERT_FREE(buf);
-	return;
-    }
-    Message(M_ERROR,"%s: failed to deliver a packet (bad destination address)"
-	    "\nXXX make this message clearer\n");
-    BUF_FREE(buf);
+    BUF_ASSERT_FREE(buf);
 }
 
-static void netlink_packet_forward(struct netlink *st, struct buffer_if *buf)
+static void netlink_packet_forward(struct netlink *st, 
+				   struct netlink_client *sender,
+				   struct buffer_if *buf)
 {
+    if (buf->size < (int)sizeof(struct iphdr)) return;
     struct iphdr *iph=(struct iphdr *)buf->start;
     
     BUF_ASSERT_USED(buf);
@@ -367,7 +792,8 @@ static void netlink_packet_forward(struct netlink *st, struct buffer_if *buf)
     /* Packet has already been checked */
     if (iph->ttl<=1) {
 	/* Generate ICMP time exceeded */
-	netlink_icmp_simple(st,buf,11,0);
+	netlink_icmp_simple(st,sender,buf,ICMP_TYPE_TIME_EXCEEDED,
+			    ICMP_CODE_TTL_EXCEEDED,icmp_noinfo);
 	BUF_FREE(buf);
 	return;
     }
@@ -375,43 +801,57 @@ static void netlink_packet_forward(struct netlink *st, struct buffer_if *buf)
     iph->check=0;
     iph->check=ip_fast_csum((uint8_t *)iph,iph->ihl);
 
-    netlink_packet_deliver(st,buf);
+    netlink_packet_deliver(st,sender,buf);
     BUF_ASSERT_FREE(buf);
 }
 
-/* Someone has been foolish enough to address a packet to us. I
-   suppose we should reply to it, just to be polite. */
-static void netlink_packet_local(struct netlink *st, struct buffer_if *buf)
+/* Deal with packets addressed explicitly to us */
+static void netlink_packet_local(struct netlink *st,
+				 struct netlink_client *sender,
+				 struct buffer_if *buf)
 {
     struct icmphdr *h;
 
+    st->localcount++;
+
+    if (buf->size < (int)sizeof(struct icmphdr)) {
+	Message(M_WARNING,"%s: short packet addressed to secnet; "
+		"ignoring it\n",st->name);
+	BUF_FREE(buf);
+	return;
+    }
     h=(struct icmphdr *)buf->start;
 
-    if ((ntohs(h->iph.frag_off)&0xbfff)!=0) {
-	Message(M_WARNING,"%s: fragmented packet addressed to us\n",st->name);
+    unsigned fraginfo = ntohs(h->iph.frag);
+    if ((fraginfo&(IPHDR_FRAG_OFF|IPHDR_FRAG_MORE))!=0) {
+	if (!(fraginfo & IPHDR_FRAG_OFF))
+	    /* report only for first fragment */
+	    Message(M_WARNING,"%s: fragmented packet addressed to secnet; "
+		    "ignoring it\n",st->name);
 	BUF_FREE(buf);
 	return;
     }
 
     if (h->iph.protocol==1) {
 	/* It's ICMP */
-	if (h->type==8 && h->code==0) {
+	if (h->type==ICMP_TYPE_ECHO_REQUEST && h->code==0) {
 	    /* ICMP echo-request. Special case: we re-use the buffer
 	       to construct the reply. */
-	    h->type=0;
+	    h->type=ICMP_TYPE_ECHO_REPLY;
 	    h->iph.daddr=h->iph.saddr;
 	    h->iph.saddr=htonl(st->secnet_address);
-	    h->iph.ttl=255; /* Be nice and bump it up again... */
+	    h->iph.ttl=255;
 	    h->iph.check=0;
 	    h->iph.check=ip_fast_csum((uint8_t *)h,h->iph.ihl);
 	    netlink_icmp_csum(h);
-	    netlink_packet_deliver(st,buf);
+	    netlink_packet_deliver(st,NULL,buf);
 	    return;
 	}
 	Message(M_WARNING,"%s: unknown incoming ICMP\n",st->name);
     } else {
 	/* Send ICMP protocol unreachable */
-	netlink_icmp_simple(st,buf,3,2);
+	netlink_icmp_simple(st,sender,buf,ICMP_TYPE_UNREACHABLE,
+			    ICMP_CODE_PROTOCOL_UNREACHABLE,icmp_noinfo);
 	BUF_FREE(buf);
 	return;
     }
@@ -419,695 +859,420 @@ static void netlink_packet_local(struct netlink *st, struct buffer_if *buf)
     BUF_FREE(buf);
 }
 
-/* Called by site code when remote packet is available */
-/* buf is allocated on entry and free on return */
-static void netlink_from_tunnel(void *sst, void *cst, struct buffer_if *buf)
+/* If cid==NULL packet is from host, otherwise cid specifies which tunnel 
+   it came from. */
+static void netlink_incoming(struct netlink *st, struct netlink_client *sender,
+			     struct buffer_if *buf)
 {
-    struct netlink *st=sst;
-    struct netlink_client *client=cst;
     uint32_t source,dest;
     struct iphdr *iph;
+    char errmsgbuf[50];
+    const char *sourcedesc=sender?sender->name:"host";
 
     BUF_ASSERT_USED(buf);
-    if (!netlink_check(st,buf)) {
-	Message(M_WARNING,"%s: bad IP packet from tunnel %s\n",
-		st->name,client->name);
+
+    if (!netlink_check(st,buf,errmsgbuf,sizeof(errmsgbuf))) {
+	Message(M_WARNING,"%s: bad IP packet from %s: %s\n",
+		st->name,sourcedesc,
+		errmsgbuf);
 	BUF_FREE(buf);
 	return;
     }
+    assert(buf->size >= (int)sizeof(struct iphdr));
     iph=(struct iphdr *)buf->start;
 
     source=ntohl(iph->saddr);
     dest=ntohl(iph->daddr);
 
-    /* Check that the packet source is in 'nets' and its destination is
-       in client->networks */
-    if (!subnet_match(client->networks,source)) {
-	string_t s,d;
-	s=ipaddr_to_string(source);
-	d=ipaddr_to_string(dest);
-	Message(M_WARNING,"%s: packet from tunnel %s with bad source address "
-		"(s=%s,d=%s)\n",st->name,client->name,s,d);
-	free(s); free(d);
-	BUF_FREE(buf);
-	return;
-    }
-    /* (st->secnet_address needs checking before matching against
-       st->networks because secnet's IP address may not be in the
-       range the host is willing to deal with) */
-    if (dest==st->secnet_address) {
-        netlink_packet_local(st,buf);
-	BUF_ASSERT_FREE(buf);
-	return;
-    }
-    if (!subnet_match(&st->networks,dest)) {
-	string_t s,d;
-	s=ipaddr_to_string(source);
-	d=ipaddr_to_string(dest);
-	Message(M_WARNING,"%s: incoming packet from tunnel %s "
-		"with bad destination address "
-		"(s=%s,d=%s)\n",st->name,client->name,s,d);
-	free(s); free(d);
-	BUF_FREE(buf);
-	return;
+    /* Check source. If we don't like the source, there's no point
+       generating ICMP because we won't know how to get it to the
+       source of the packet. */
+    if (sender) {
+	/* Check that the packet source is appropriate for the tunnel
+	   it came down */
+	if (!ipset_contains_addr(sender->networks,source)) {
+	    string_t s,d;
+	    s=ipaddr_to_string(source);
+	    d=ipaddr_to_string(dest);
+	    Message(M_WARNING,"%s: packet from tunnel %s with bad "
+		    "source address (s=%s,d=%s)\n",st->name,sender->name,s,d);
+	    BUF_FREE(buf);
+	    return;
+	}
+    } else {
+	/* Check that the packet originates in our configured local
+	   network, and hasn't been forwarded from elsewhere or
+	   generated with the wrong source address */
+	if (!ipset_contains_addr(st->networks,source)) {
+	    string_t s,d;
+	    s=ipaddr_to_string(source);
+	    d=ipaddr_to_string(dest);
+	    Message(M_WARNING,"%s: outgoing packet with bad source address "
+		    "(s=%s,d=%s)\n",st->name,s,d);
+	    BUF_FREE(buf);
+	    return;
+	}
     }
 
-    netlink_packet_forward(st,buf);
-
-    BUF_ASSERT_FREE(buf);
-}
-
-/* Called by driver code when packet is received from kernel */
-/* cid should be NULL */
-/* buf should be allocated on entry, and is free on return */
-static void netlink_from_host(void *sst, void *cid, struct buffer_if *buf)
-{
-    struct netlink *st=sst;
-    uint32_t source,dest;
-    struct iphdr *iph;
-
-    BUF_ASSERT_USED(buf);
-    if (!netlink_check(st,buf)) {
-	Message(M_WARNING,"%s: bad IP packet from host\n",
-		st->name);
-	BUF_FREE(buf);
+    /* If this is a point-to-point device we don't examine the
+       destination address at all; we blindly send it down our
+       one-and-only registered tunnel, or to the host, depending on
+       where it came from.  It's up to external software to check
+       address validity and generate ICMP, etc. */
+    if (st->ptp) {
+	if (sender) {
+	    netlink_host_deliver(st,sender,source,dest,buf);
+	} else {
+	    netlink_client_deliver(st,st->clients,source,dest,buf);
+	}
+	BUF_ASSERT_FREE(buf);
 	return;
     }
-    iph=(struct iphdr *)buf->start;
 
-    source=ntohl(iph->saddr);
-    dest=ntohl(iph->daddr);
-
-    if (!subnet_match(&st->networks,source)) {
-	string_t s,d;
-	s=ipaddr_to_string(source);
-	d=ipaddr_to_string(dest);
-	Message(M_WARNING,"%s: outgoing packet with bad source address "
-		"(s=%s,d=%s)\n",st->name,s,d);
-	free(s); free(d);
-	BUF_FREE(buf);
-	return;
-    }
+    /* st->secnet_address needs checking before matching destination
+       addresses */
     if (dest==st->secnet_address) {
-	netlink_packet_local(st,buf);
+	netlink_packet_local(st,sender,buf);
 	BUF_ASSERT_FREE(buf);
 	return;
     }
-    netlink_packet_forward(st,buf);
+    netlink_packet_forward(st,sender,buf);
     BUF_ASSERT_FREE(buf);
 }
 
-static void netlink_set_delivery(void *sst, void *cid, bool_t can_deliver)
+static void netlink_inst_incoming(void *sst, struct buffer_if *buf)
 {
-    struct netlink_client *c=cid;
+    struct netlink_client *c=sst;
+    struct netlink *st=c->nst;
 
-    c->can_deliver=can_deliver;
+    netlink_incoming(st,c,buf);
 }
 
-static void *netlink_regnets(void *sst, struct subnet_list *nets,
-			     netlink_deliver_fn *deliver, void *dst,
-			     uint32_t max_start_pad, uint32_t max_end_pad,
-			     string_t client_name)
+static void netlink_dev_incoming(void *sst, struct buffer_if *buf)
 {
     struct netlink *st=sst;
-    struct netlink_client *c;
-
-    Message(M_DEBUG_CONFIG,"netlink_regnets: request for %d networks, "
-	    "max_start_pad=%d, max_end_pad=%d\n",
-	    nets->entries,max_start_pad,max_end_pad);
-
-    /* Check that nets does not intersect with st->networks or
-       st->exclude_remote_networks; refuse to register if it does. */
-    if (subnet_lists_intersect(&st->networks,nets)) {
-	Message(M_ERROR,"%s: site %s specifies networks that "
-		"intersect with our local networks\n",st->name,client_name);
-	return False;
-    }
-    if (subnet_lists_intersect(&st->exclude_remote_networks,nets)) {
-	Message(M_ERROR,"%s: site %s specifies networks that "
-		"intersect with the explicitly excluded remote networks\n",
-		st->name,client_name);
-	return False;
-    }
 
-    c=safe_malloc(sizeof(*c),"netlink_regnets");
-    c->networks=nets;
-    c->deliver=deliver;
-    c->dst=dst;
-    c->name=client_name; /* XXX copy it? */
-    c->can_deliver=False;
-    c->next=st->clients;
-    st->clients=c;
-    if (max_start_pad > st->max_start_pad) st->max_start_pad=max_start_pad;
-    if (max_end_pad > st->max_end_pad) st->max_end_pad=max_end_pad;
-
-    return c;
+    netlink_incoming(st,NULL,buf);
 }
 
-static netlink_deliver_fn *netlink_init(struct netlink *st,
-					void *dst, struct cloc loc,
-					dict_t *dict, string_t description,
-					netlink_deliver_fn *to_host)
+static void netlink_set_quality(void *sst, uint32_t quality)
 {
-    st->dst=dst;
-    st->cl.description=description;
-    st->cl.type=CL_NETLINK;
-    st->cl.apply=NULL;
-    st->cl.interface=&st->ops;
-    st->ops.st=st;
-    st->ops.regnets=netlink_regnets;
-    st->ops.deliver=netlink_from_tunnel;
-    st->ops.set_delivery=netlink_set_delivery;
-    st->max_start_pad=0;
-    st->max_end_pad=0;
-    st->clients=NULL;
-    st->deliver_to_host=to_host;
-
-    st->name=dict_read_string(dict,"name",False,"netlink",loc);
-    if (!st->name) st->name=description;
-    dict_read_subnet_list(dict, "networks", True, "netlink", loc,
-			  &st->networks);
-    dict_read_subnet_list(dict, "exclude-remote-networks", False, "netlink",
-			  loc, &st->exclude_remote_networks);
-    /* local-address and secnet-address do not have to be in local-networks;
-       however, they should be advertised in the 'sites' file for the
-       local site. */
-    st->local_address=string_to_ipaddr(
-	dict_find_item(dict,"local-address", True, "netlink", loc),"netlink");
-    st->secnet_address=string_to_ipaddr(
-	dict_find_item(dict,"secnet-address", True, "netlink", loc),"netlink");
-    st->mtu=dict_read_number(dict, "mtu", False, "netlink", loc, DEFAULT_MTU);
-    buffer_new(&st->icmp,ICMP_BUFSIZE);
+    struct netlink_client *c=sst;
+    struct netlink *st=c->nst;
 
-    return netlink_from_host;
+    c->link_quality=quality;
+    c->up=(c->link_quality==LINK_QUALITY_DOWN)?False:True;
+    if (c->options&OPT_SOFTROUTE) {
+	st->set_routes(st->dst,c);
+    }
 }
 
-/* Connection to the kernel through userv-ipif */
-
-struct userv {
-    struct netlink nl;
-    int txfd; /* We transmit to userv */
-    int rxfd; /* We receive from userv */
-    string_t userv_path;
-    string_t service_user;
-    string_t service_name;
-    uint32_t txbuflen;
-    struct buffer_if *buff; /* We unstuff received packets into here
-			       and send them to the site code. */
-    bool_t pending_esc;
-    netlink_deliver_fn *netlink_to_tunnel;
-};
-
-static int userv_beforepoll(void *sst, struct pollfd *fds, int *nfds_io,
-			    int *timeout_io, const struct timeval *tv_now,
-			    uint64_t *now)
+static void netlink_output_subnets(struct netlink *st, uint32_t loglevel,
+				   struct subnet_list *snets)
 {
-    struct userv *st=sst;
-    *nfds_io=2;
-    fds[0].fd=st->txfd;
-    fds[0].events=POLLERR; /* Might want to pick up POLLOUT sometime */
-    fds[1].fd=st->rxfd;
-    fds[1].events=POLLIN|POLLERR|POLLHUP;
-    return 0;
+    int32_t i;
+    string_t net;
+
+    for (i=0; i<snets->entries; i++) {
+	net=subnet_to_string(snets->list[i]);
+	Message(loglevel,"%s ",net);
+    }
 }
 
-static void userv_afterpoll(void *sst, struct pollfd *fds, int nfds,
-			    const struct timeval *tv_now, uint64_t *now)
+static void netlink_dump_routes(struct netlink *st, bool_t requested)
 {
-    struct userv *st=sst;
-    uint8_t rxbuf[DEFAULT_BUFSIZE];
-    int l,i;
-
-    if (fds[1].revents&POLLERR) {
-	printf("userv_afterpoll: hup!\n");
-    }
-    if (fds[1].revents&POLLIN) {
-	l=read(st->rxfd,rxbuf,DEFAULT_BUFSIZE);
-	if (l<0) {
-	    fatal_perror("userv_afterpoll: read(rxfd)");
-	}
-	if (l==0) {
-	    fatal("userv_afterpoll: read(rxfd)=0; userv gone away?\n");
+    int i;
+    string_t net;
+    uint32_t c=M_INFO;
+
+    if (requested) c=M_WARNING;
+    if (st->ptp) {
+	net=ipaddr_to_string(st->secnet_address);
+	Message(c,"%s: point-to-point (remote end is %s); routes: ",
+		st->name, net);
+	netlink_output_subnets(st,c,st->clients->subnets);
+	Message(c,"\n");
+    } else {
+	Message(c,"%s: routing table:\n",st->name);
+	for (i=0; i<st->n_clients; i++) {
+	    netlink_output_subnets(st,c,st->routes[i]->subnets);
+	    Message(c,"-> tunnel %s (%s,mtu %d,%s routes,%s,"
+		    "quality %d,use %d,pri %lu)\n",
+		    st->routes[i]->name,
+		    st->routes[i]->up?"up":"down",
+		    st->routes[i]->mtu,
+		    st->routes[i]->options&OPT_SOFTROUTE?"soft":"hard",
+		    st->routes[i]->options&OPT_ALLOWROUTE?"free":"restricted",
+		    st->routes[i]->link_quality,
+		    st->routes[i]->outcount,
+		    (unsigned long)st->routes[i]->priority);
 	}
-	/* XXX really crude unstuff code */
-	/* XXX check for buffer overflow */
-	BUF_ASSERT_USED(st->buff);
-	for (i=0; i<l; i++) {
-	    if (st->pending_esc) {
-		st->pending_esc=False;
-		switch(rxbuf[i]) {
-		case SLIP_ESCEND:
-		    *(uint8_t *)buf_append(st->buff,1)=SLIP_END;
-		    break;
-		case SLIP_ESCESC:
-		    *(uint8_t *)buf_append(st->buff,1)=SLIP_ESC;
-		    break;
-		default:
-		    fatal("userv_afterpoll: bad SLIP escape character\n");
-		}
-	    } else {
-		switch (rxbuf[i]) {
-		case SLIP_END:
-		    if (st->buff->size>0) {
-			st->netlink_to_tunnel(&st->nl,NULL,
-					      st->buff);
-			BUF_ALLOC(st->buff,"userv_afterpoll");
-		    }
-		    buffer_init(st->buff,st->nl.max_start_pad);
-		    break;
-		case SLIP_ESC:
-		    st->pending_esc=True;
-		    break;
-		default:
-		    *(uint8_t *)buf_append(st->buff,1)=rxbuf[i];
-		    break;
-		}
-	    }
+	net=ipaddr_to_string(st->secnet_address);
+	Message(c,"%s/32 -> netlink \"%s\" (use %d)\n",
+		net,st->name,st->localcount);
+	for (i=0; i<st->subnets->entries; i++) {
+	    net=subnet_to_string(st->subnets->list[i]);
+	    Message(c,"%s ",net);
 	}
+	if (i>0)
+	    Message(c,"-> host (use %d)\n",st->outcount);
     }
 }
 
-/* Send buf to the kernel. Free buf before returning. */
-static void userv_deliver_to_kernel(void *sst, void *cid,
-				    struct buffer_if *buf)
+/* ap is a pointer to a member of the routes array */
+static int netlink_compare_client_priority(const void *ap, const void *bp)
 {
-    struct userv *st=sst;
-    uint8_t txbuf[DEFAULT_BUFSIZE];
-    uint8_t *i;
-    uint32_t j;
-
-    BUF_ASSERT_USED(buf);
+    const struct netlink_client *const*a=ap;
+    const struct netlink_client *const*b=bp;
 
-    /* Spit the packet at userv-ipif: SLIP start marker, then
-       bytestuff the packet, then SLIP end marker */
-    /* XXX crunchy bytestuff code */
-    j=0;
-    txbuf[j++]=SLIP_END;
-    for (i=buf->start; i<(buf->start+buf->size); i++) {
-	switch (*i) {
-	case SLIP_END:
-	    txbuf[j++]=SLIP_ESC;
-	    txbuf[j++]=SLIP_ESCEND;
-	    break;
-	case SLIP_ESC:
-	    txbuf[j++]=SLIP_ESC;
-	    txbuf[j++]=SLIP_ESCESC;
-	    break;
-	default:
-	    txbuf[j++]=*i;
-	    break;
-	}
-    }
-    txbuf[j++]=SLIP_END;
-    if (write(st->txfd,txbuf,j)<0) {
-	fatal_perror("userv_deliver_to_kernel: write()");
-    }
-    BUF_FREE(buf);
+    if ((*a)->priority==(*b)->priority) return 0;
+    if ((*a)->priority<(*b)->priority) return 1;
+    return -1;
 }
 
-static void userv_phase_hook(void *sst, uint32_t newphase)
+static void netlink_phase_hook(void *sst, uint32_t new_phase)
 {
-    struct userv *st=sst;
-    pid_t child;
-    int c_stdin[2];
-    int c_stdout[2];
-    string_t addrs;
-    string_t nets;
-    string_t s;
+    struct netlink *st=sst;
     struct netlink_client *c;
-    int i;
-
-    /* This is where we actually invoke userv - all the networks we'll
-       be using should already have been registered. */
-
-    addrs=safe_malloc(512,"userv_phase_hook:addrs");
-    snprintf(addrs,512,"%s,%s,%d,slip",ipaddr_to_string(st->nl.local_address),
-	     ipaddr_to_string(st->nl.secnet_address),st->nl.mtu);
-
-    nets=safe_malloc(1024,"userv_phase_hook:nets");
-    *nets=0;
-    for (c=st->nl.clients; c; c=c->next) {
-	for (i=0; i<c->networks->entries; i++) {
-	    s=subnet_to_string(&c->networks->list[i]);
-	    strcat(nets,s);
-	    strcat(nets,",");
-	    free(s);
-	}
-    }
-    nets[strlen(nets)-1]=0;
-
-    Message(M_INFO,"\nuserv_phase_hook: %s %s %s %s %s\n",st->userv_path,
-	   st->service_user,st->service_name,addrs,nets);
-
-    /* Allocate buffer, plus space for padding. Make sure we end up
-       with the start of the packet well-aligned. */
-    /* ALIGN(st->max_start_pad,16); */
-    /* ALIGN(st->max_end_pad,16); */
-
-    st->pending_esc=False;
-
-    /* Invoke userv */
-    if (pipe(c_stdin)!=0) {
-	fatal_perror("userv_phase_hook: pipe(c_stdin)");
-    }
-    if (pipe(c_stdout)!=0) {
-	fatal_perror("userv_phase_hook: pipe(c_stdout)");
+    int32_t i;
+
+    /* All the networks serviced by the various tunnels should now
+     * have been registered.  We build a routing table by sorting the
+     * clients by priority.  */
+    st->routes=safe_malloc_ary(sizeof(*st->routes),st->n_clients,
+			       "netlink_phase_hook");
+    /* Fill the table */
+    i=0;
+    for (c=st->clients; c; c=c->next) {
+	assert(i<INT_MAX);
+	st->routes[i++]=c;
     }
-    st->txfd=c_stdin[1];
-    st->rxfd=c_stdout[0];
+    /* Sort the table in descending order of priority */
+    qsort(st->routes,st->n_clients,sizeof(*st->routes),
+	  netlink_compare_client_priority);
 
-    child=fork();
-    if (child==-1) {
-	fatal_perror("userv_phase_hook: fork()");
-    }
-    if (child==0) {
-	char **argv;
-
-	/* We are the child. Modify our stdin and stdout, then exec userv */
-	dup2(c_stdin[0],0);
-	dup2(c_stdout[1],1);
-	close(c_stdin[1]);
-	close(c_stdout[0]);
-
-	/* The arguments are:
-	   userv
-	   service-user
-	   service-name
-	   local-addr,secnet-addr,mtu,protocol
-	   route1,route2,... */
-	argv=malloc(sizeof(*argv)*6);
-	argv[0]=st->userv_path;
-	argv[1]=st->service_user;
-	argv[2]=st->service_name;
-	argv[3]=addrs;
-	argv[4]=nets;
-	argv[5]=NULL;
-	execvp(st->userv_path,argv);
-	perror("netlink-userv-ipif: execvp");
-
-	exit(1);
-    }
-    /* We are the parent... */
-	   
-    /* Register for poll() */
-    register_for_poll(st, userv_beforepoll, userv_afterpoll, 2, st->nl.name);
+    netlink_dump_routes(st,False);
 }
 
-static list_t *userv_apply(closure_t *self, struct cloc loc, dict_t *context,
-			   list_t *args)
+static void netlink_signal_handler(void *sst, int signum)
 {
-    struct userv *st;
-    item_t *item;
-    dict_t *dict;
-
-    st=safe_malloc(sizeof(*st),"userv_apply");
-
-    /* First parameter must be a dict */
-    item=list_elem(args,0);
-    if (!item || item->type!=t_dict)
-	cfgfatal(loc,"userv-ipif","parameter must be a dictionary\n");
-    
-    dict=item->data.dict;
-
-    st->netlink_to_tunnel=
-	netlink_init(&st->nl,st,loc,dict,
-		     "netlink-userv-ipif",userv_deliver_to_kernel);
-
-    st->userv_path=dict_read_string(dict,"userv-path",False,"userv-netlink",
-				    loc);
-    st->service_user=dict_read_string(dict,"service-user",False,
-				      "userv-netlink",loc);
-    st->service_name=dict_read_string(dict,"service-name",False,
-				      "userv-netlink",loc);
-    if (!st->userv_path) st->userv_path="userv";
-    if (!st->service_user) st->service_user="root";
-    if (!st->service_name) st->service_name="ipif";
-    st->buff=find_cl_if(dict,"buffer",CL_BUFFER,True,"userv-netlink",loc);
-    BUF_ALLOC(st->buff,"netlink:userv_apply");
-
-    st->rxfd=-1; st->txfd=-1;
-    add_hook(PHASE_DROPPRIV,userv_phase_hook,st);
-
-    return new_closure(&st->nl.cl);
-}
-
-/* Connection to the kernel through the universal TUN/TAP driver */
-
-struct tun {
-    struct netlink nl;
-    int fd;
-    string_t device_path;
-    string_t interface_name;
-    string_t ifconfig_path;
-    string_t route_path;
-    bool_t tun_old;
-    bool_t search_for_if; /* Applies to tun-old only */
-    struct buffer_if *buff; /* We receive packets into here
-			       and send them to the netlink code. */
-    netlink_deliver_fn *netlink_to_tunnel;
-};
-
-static int tun_beforepoll(void *sst, struct pollfd *fds, int *nfds_io,
-			  int *timeout_io, const struct timeval *tv_now,
-			  uint64_t *now)
-{
-    struct tun *st=sst;
-    *nfds_io=1;
-    fds[0].fd=st->fd;
-    fds[0].events=POLLIN|POLLERR|POLLHUP;
-    return 0;
+    struct netlink *st=sst;
+    Message(M_INFO,"%s: route dump requested by SIGUSR1\n",st->name);
+    netlink_dump_routes(st,True);
 }
 
-static void tun_afterpoll(void *sst, struct pollfd *fds, int nfds,
-			    const struct timeval *tv_now, uint64_t *now)
+static void netlink_inst_set_mtu(void *sst, int32_t new_mtu)
 {
-    struct tun *st=sst;
-    int l;
+    struct netlink_client *c=sst;
 
-    if (fds[0].revents&POLLERR) {
-	printf("tun_afterpoll: hup!\n");
-    }
-    if (fds[0].revents&POLLIN) {
-	BUF_ALLOC(st->buff,"tun_afterpoll");
-	buffer_init(st->buff,st->nl.max_start_pad);
-	l=read(st->fd,st->buff->start,st->buff->len-st->nl.max_start_pad);
-	if (l<0) {
-	    fatal_perror("tun_afterpoll: read()");
-	}
-	if (l==0) {
-	    fatal("tun_afterpoll: read()=0; device gone away?\n");
-	}
-	if (l>0) {
-	    st->buff->size=l;
-	    st->netlink_to_tunnel(&st->nl,NULL,st->buff);
-	    BUF_ASSERT_FREE(st->buff);
-	}
-    }
+    c->mtu=new_mtu;
 }
 
-static void tun_deliver_to_kernel(void *sst, void *cid,
-				  struct buffer_if *buf)
+static void netlink_inst_reg(void *sst, netlink_deliver_fn *deliver, 
+			     void *dst, uint32_t *localmtu_r)
 {
-    struct tun *st=sst;
+    struct netlink_client *c=sst;
+    struct netlink *st=c->nst;
 
-    BUF_ASSERT_USED(buf);
+    c->deliver=deliver;
+    c->dst=dst;
 
-    /* No error checking, because we'd just throw the packet away anyway */
-    write(st->fd,buf->start,buf->size);
-    BUF_FREE(buf);
+    if (localmtu_r)
+	*localmtu_r=st->mtu;
 }
 
-static void tun_phase_hook(void *sst, uint32_t newphase)
+static struct flagstr netlink_option_table[]={
+    { "soft", OPT_SOFTROUTE },
+    { "allow-route", OPT_ALLOWROUTE },
+    { NULL, 0}
+};
+/* This is the routine that gets called when the closure that's
+   returned by an invocation of a netlink device closure (eg. tun,
+   userv-ipif) is invoked.  It's used to create routes and pass in
+   information about them; the closure it returns is used by site
+   code.  */
+static closure_t *netlink_inst_create(struct netlink *st,
+				      struct cloc loc, dict_t *dict)
 {
-    struct tun *st=sst;
-    string_t hostaddr,secnetaddr;
-    uint8_t mtu[6];
-    string_t network,mask;
     struct netlink_client *c;
-    int i;
+    string_t name;
+    struct ipset *networks;
+    uint32_t options,priority;
+    int32_t mtu;
+    list_t *l;
+
+    name=dict_read_string(dict, "name", True, st->name, loc);
+
+    l=dict_lookup(dict,"routes");
+    if (!l)
+	cfgfatal(loc,st->name,"required parameter \"routes\" not found\n");
+    networks=string_list_to_ipset(l,loc,st->name,"routes");
+    options=string_list_to_word(dict_lookup(dict,"options"),
+				netlink_option_table,st->name);
+
+    priority=dict_read_number(dict,"priority",False,st->name,loc,0);
+    mtu=dict_read_number(dict,"mtu",False,st->name,loc,0);
+
+    if ((options&OPT_SOFTROUTE) && !st->set_routes) {
+	cfgfatal(loc,st->name,"this netlink device does not support "
+		 "soft routes.\n");
+	return NULL;
+    }
 
-    if (st->tun_old) {
-	if (st->search_for_if) {
-	    string_t dname;
-	    int i;
-
-	    /* ASSERT st->interface_name */
-	    dname=safe_malloc(strlen(st->device_path)+4,"tun_old_apply");
-	    st->interface_name=safe_malloc(8,"tun_phase_hook");
-	
-	    for (i=0; i<255; i++) {
-		sprintf(dname,"%s%d",st->device_path,i);
-		if ((st->fd=open(dname,O_RDWR))>0) {
-		    sprintf(st->interface_name,"tun%d",i);
-		    Message(M_INFO,"%s: allocated network interface %s "
-			    "through %s\n",st->nl.name,st->interface_name,
-			    dname);
-		    break;
-		}
-	    }
-	    if (st->fd==-1) {
-		fatal("%s: unable to open any TUN device (%s...)\n",
-		      st->nl.name,st->device_path);
-	    }
-	} else {
-	    st->fd=open(st->device_path,O_RDWR);
-	    if (st->fd==-1) {
-		fatal_perror("%s: unable to open TUN device file %s",
-			     st->nl.name,st->device_path);
-	    }
+    if (options&OPT_SOFTROUTE) {
+	/* XXX for now we assume that soft routes require root privilege;
+	   this may not always be true. The device driver can tell us. */
+	require_root_privileges=True;
+	require_root_privileges_explanation="netlink: soft routes";
+	if (st->ptp) {
+	    cfgfatal(loc,st->name,"point-to-point netlinks do not support "
+		     "soft routes.\n");
+	    return NULL;
 	}
-    } else {
-#ifdef HAVE_LINUX_IF_H
-	struct ifreq ifr;
-
-	/* New TUN interface: open the device, then do ioctl TUNSETIFF
-	   to set or find out the network interface name. */
-	st->fd=open(st->device_path,O_RDWR);
-	if (st->fd==-1) {
-	    fatal_perror("%s: can't open device file %s",st->nl.name,
-			 st->device_path);
-	}
-	memset(&ifr,0,sizeof(ifr));
-	ifr.ifr_flags = IFF_TUN | IFF_NO_PI; /* Just send/receive IP packets,
-						no extra headers */
-	if (st->interface_name)
-	    strncpy(ifr.ifr_name,st->interface_name,IFNAMSIZ);
-	if (ioctl(st->fd,TUNSETIFF,&ifr)<0) {
-	    fatal_perror("%s: ioctl(TUNSETIFF)",st->nl.name);
-	}
-	if (!st->interface_name) {
-	    st->interface_name=safe_malloc(strlen(ifr.ifr_name)+1,"tun_apply");
-	    strcpy(st->interface_name,ifr.ifr_name);
-	    Message(M_INFO,"%s: allocated network interface %s\n",st->nl.name,
-		    st->interface_name);
-	}
-#else
-	fatal("netlink.c:tun_phase_hook:!tun_old unexpected\n");
-#endif /* HAVE_LINUX_IF_H */
     }
-    /* All the networks we'll be using have been registered. Invoke ifconfig
-       to set the TUN device's address, and route to add routes to all
-       our networks. */
-
-    hostaddr=ipaddr_to_string(st->nl.local_address);
-    secnetaddr=ipaddr_to_string(st->nl.secnet_address);
-    snprintf(mtu,6,"%d",st->nl.mtu);
-    mtu[5]=0;
-
-    sys_cmd(st->ifconfig_path,"ifconfig",st->interface_name,
-	    hostaddr,"netmask","255.255.255.255","-broadcast",
-	    "pointopoint",secnetaddr,"mtu",mtu,"up",(char *)0);
-
-    for (c=st->nl.clients; c; c=c->next) {
-	for (i=0; i<c->networks->entries; i++) {
-	    network=ipaddr_to_string(c->networks->list[i].prefix);
-	    mask=ipaddr_to_string(c->networks->list[i].mask);
-	    sys_cmd(st->route_path,"route","add","-net",network,
-		    "netmask",mask,"gw",secnetaddr,(char *)0);
-	}
+
+    /* Check that nets are a subset of st->remote_networks;
+       refuse to register if they are not. */
+    if (!ipset_is_subset(st->remote_networks,networks)) {
+	cfgfatal(loc,st->name,"routes are not allowed\n");
+	return NULL;
     }
 
-    /* Register for poll() */
-    register_for_poll(st, tun_beforepoll, tun_afterpoll, 1, st->nl.name);
+    c=safe_malloc(sizeof(*c),"netlink_inst_create");
+    c->cl.description=name;
+    c->cl.type=CL_NETLINK;
+    c->cl.apply=NULL;
+    c->cl.interface=&c->ops;
+    c->ops.st=c;
+    c->ops.reg=netlink_inst_reg;
+    c->ops.deliver=netlink_inst_incoming;
+    c->ops.set_quality=netlink_set_quality;
+    c->ops.set_mtu=netlink_inst_set_mtu;
+    c->nst=st;
+
+    c->networks=networks;
+    c->subnets=ipset_to_subnet_list(networks);
+    c->priority=priority;
+    c->deliver=NULL;
+    c->dst=NULL;
+    c->name=name;
+    c->link_quality=LINK_QUALITY_UNUSED;
+    c->mtu=mtu?mtu:st->mtu;
+    c->options=options;
+    c->outcount=0;
+    c->up=False;
+    c->kup=False;
+    c->next=st->clients;
+    st->clients=c;
+    assert(st->n_clients < INT_MAX);
+    st->n_clients++;
+
+    return &c->cl;
 }
 
-#ifdef HAVE_LINUX_IF_H
-static list_t *tun_apply(closure_t *self, struct cloc loc, dict_t *context,
-			 list_t *args)
+static list_t *netlink_inst_apply(closure_t *self, struct cloc loc,
+				  dict_t *context, list_t *args)
 {
-    struct tun *st;
-    item_t *item;
-    dict_t *dict;
+    struct netlink *st=self->interface;
 
-    st=safe_malloc(sizeof(*st),"tun_apply");
+    dict_t *dict;
+    item_t *item;
+    closure_t *cl;
 
-    /* First parameter must be a dict */
     item=list_elem(args,0);
-    if (!item || item->type!=t_dict)
-	cfgfatal(loc,"tun","parameter must be a dictionary\n");
-    
+    if (!item || item->type!=t_dict) {
+	cfgfatal(loc,st->name,"must have a dictionary argument\n");
+    }
     dict=item->data.dict;
 
-    st->netlink_to_tunnel=
-	netlink_init(&st->nl,st,loc,dict,
-		     "netlink-tun",tun_deliver_to_kernel);
-
-    st->tun_old=False;
-    st->device_path=dict_read_string(dict,"device",False,"tun-netlink",loc);
-    st->interface_name=dict_read_string(dict,"interface",False,
-					"tun-netlink",loc);
-    st->ifconfig_path=dict_read_string(dict,"ifconfig-path",
-				       False,"tun-netlink",loc);
-    st->route_path=dict_read_string(dict,"route-path",
-				    False,"tun-netlink",loc);
-
-    if (!st->device_path) st->device_path="/dev/net/tun";
-    if (!st->ifconfig_path) st->ifconfig_path="ifconfig";
-    if (!st->route_path) st->route_path="route";
-    st->buff=find_cl_if(dict,"buffer",CL_BUFFER,True,"tun-netlink",loc);
-
-    add_hook(PHASE_GETRESOURCES,tun_phase_hook,st);
+    cl=netlink_inst_create(st,loc,dict);
 
-    return new_closure(&st->nl.cl);
+    return new_closure(cl);
 }
-#endif /* HAVE_LINUX_IF_H */
 
-static list_t *tun_old_apply(closure_t *self, struct cloc loc, dict_t *context,
-			     list_t *args)
+netlink_deliver_fn *netlink_init(struct netlink *st,
+				 void *dst, struct cloc loc,
+				 dict_t *dict, cstring_t description,
+				 netlink_route_fn *set_routes,
+				 netlink_deliver_fn *to_host)
 {
-    struct tun *st;
-    item_t *item;
-    dict_t *dict;
-
-    st=safe_malloc(sizeof(*st),"tun_old_apply");
-
-    Message(M_WARNING,"the tun-old code has never been tested. Please report "
-	    "success or failure to steve@greenend.org.uk\n");
+    item_t *sa, *ptpa;
+    list_t *l;
 
-    /* First parameter must be a dict */
-    item=list_elem(args,0);
-    if (!item || item->type!=t_dict)
-	cfgfatal(loc,"tun","parameter must be a dictionary\n");
-    
-    dict=item->data.dict;
+    st->dst=dst;
+    st->cl.description=description;
+    st->cl.type=CL_PURE;
+    st->cl.apply=netlink_inst_apply;
+    st->cl.interface=st;
+    st->clients=NULL;
+    st->routes=NULL;
+    st->n_clients=0;
+    st->set_routes=set_routes;
+    st->deliver_to_host=to_host;
 
-    st->netlink_to_tunnel=
-	netlink_init(&st->nl,st,loc,dict,
-		     "netlink-tun",tun_deliver_to_kernel);
-
-    st->tun_old=True;
-    st->device_path=dict_read_string(dict,"device",False,"tun-netlink",loc);
-    st->interface_name=dict_read_string(dict,"interface",False,
-					"tun-netlink",loc);
-    st->search_for_if=dict_read_bool(dict,"interface-search",False,
-				     "tun-netlink",loc,st->device_path==NULL);
-    st->ifconfig_path=dict_read_string(dict,"ifconfig-path",False,
-				       "tun-netlink",loc);
-    st->route_path=dict_read_string(dict,"route-path",False,"tun-netlink",loc);
-
-    if (!st->device_path) st->device_path="/dev/tun";
-    if (!st->ifconfig_path) st->ifconfig_path="ifconfig";
-    if (!st->route_path) st->route_path="route";
-    st->buff=find_cl_if(dict,"buffer",CL_BUFFER,True,"tun-netlink",loc);
-
-    /* Old TUN interface: the network interface name depends on which
-       /dev/tunX file we open. If 'interface-search' is set to true, treat
-       'device' as the prefix and try numbers from 0--255. If it's set
-       to false, treat 'device' as the whole name, and require than an
-       appropriate interface name be specified. */
-    if (st->search_for_if && st->interface_name) {
-	cfgfatal(loc,"tun-old","you may not specify an interface name "
-		 "in interface-search mode\n");
+    st->name=dict_read_string(dict,"name",False,description,loc);
+    if (!st->name) st->name=description;
+    l=dict_lookup(dict,"networks");
+    if (l) 
+	st->networks=string_list_to_ipset(l,loc,st->name,"networks");
+    else {
+	struct ipset *empty;
+	empty=ipset_new();
+	st->networks=ipset_complement(empty);
+	ipset_free(empty);
     }
-    if (!st->search_for_if && !st->interface_name) {
-	cfgfatal(loc,"tun-old","you must specify an interface name "
-		 "when you explicitly specify a TUN device file\n");
+    l=dict_lookup(dict,"remote-networks");
+    if (l) {
+	st->remote_networks=string_list_to_ipset(l,loc,st->name,
+						 "remote-networks");
+    } else {
+	struct ipset *empty;
+	empty=ipset_new();
+	st->remote_networks=ipset_complement(empty);
+	ipset_free(empty);
     }
+    st->local_address=string_item_to_ipaddr(
+	dict_find_item(dict,"local-address", True, "netlink", loc),"netlink");
 
-
-    add_hook(PHASE_GETRESOURCES,tun_phase_hook,st);
-
-    return new_closure(&st->nl.cl);
+    sa=dict_find_item(dict,"secnet-address",False,"netlink",loc);
+    ptpa=dict_find_item(dict,"ptp-address",False,"netlink",loc);
+    if (sa && ptpa) {
+	cfgfatal(loc,st->name,"you may not specify secnet-address and "
+		 "ptp-address in the same netlink device\n");
+    }
+    if (!(sa || ptpa)) {
+	cfgfatal(loc,st->name,"you must specify secnet-address or "
+		 "ptp-address for this netlink device\n");
+    }
+    if (sa) {
+	st->secnet_address=string_item_to_ipaddr(sa,"netlink");
+	st->ptp=False;
+    } else {
+	st->secnet_address=string_item_to_ipaddr(ptpa,"netlink");
+	st->ptp=True;
+    }
+    /* To be strictly correct we could subtract secnet_address from
+       networks here.  It shouldn't make any practical difference,
+       though, and will make the route dump look complicated... */
+    st->subnets=ipset_to_subnet_list(st->networks);
+    st->mtu=dict_read_number(dict, "mtu", False, "netlink", loc, DEFAULT_MTU);
+    buffer_new(&st->icmp,MAX(ICMP_BUFSIZE,st->mtu));
+    st->outcount=0;
+    st->localcount=0;
+
+    add_hook(PHASE_SETUP,netlink_phase_hook,st);
+    request_signal_notification(SIGUSR1, netlink_signal_handler, st);
+
+    /* If we're point-to-point then we return a CL_NETLINK directly,
+       rather than a CL_NETLINK_OLD or pure closure (depending on
+       compatibility).  This CL_NETLINK is for our one and only
+       client.  Our cl.apply function is NULL. */
+    if (st->ptp) {
+	closure_t *cl;
+	cl=netlink_inst_create(st,loc,dict);
+	st->cl=*cl;
+    }
+    return netlink_dev_incoming;
 }
 
 /* No connection to the kernel at all... */
@@ -1116,7 +1281,21 @@ struct null {
     struct netlink nl;
 };
 
-static void null_deliver(void *sst, void *cid, struct buffer_if *buf)
+static bool_t null_set_route(void *sst, struct netlink_client *routes)
+{
+    struct null *st=sst;
+
+    if (routes->up!=routes->kup) {
+	Message(M_INFO,"%s: setting routes for tunnel %s to state %s\n",
+		st->nl.name,routes->name,
+		routes->up?"up":"down");
+	routes->kup=routes->up;
+	return True;
+    }
+    return False;
+}
+	    
+static void null_deliver(void *sst, struct buffer_if *buf)
 {
     return;
 }
@@ -1136,23 +1315,13 @@ static list_t *null_apply(closure_t *self, struct cloc loc, dict_t *context,
     
     dict=item->data.dict;
 
-    netlink_init(&st->nl,st,loc,dict,"null-netlink",null_deliver);
+    netlink_init(&st->nl,st,loc,dict,"null-netlink",null_set_route,
+		 null_deliver);
 
     return new_closure(&st->nl.cl);
 }
 
-init_module netlink_module;
 void netlink_module(dict_t *dict)
 {
-    add_closure(dict,"userv-ipif",userv_apply);
-#ifdef HAVE_LINUX_IF_H
-    add_closure(dict,"tun",tun_apply);
-#endif
-    add_closure(dict,"tun-old",tun_old_apply);
     add_closure(dict,"null-netlink",null_apply);
-#if 0
-    /* TODO */
-    add_closure(dict,"pty-slip",ptyslip_apply);
-    add_closure(dict,"slipd",slipd_apply);
-#endif /* 0 */
 }