chiark / gitweb /
site: transport peers: Break out transport_resolve_complete,_tardy
[secnet.git] / site.c
diff --git a/site.c b/site.c
index 6c92193..afa85e8 100644 (file)
--- a/site.c
+++ b/site.c
@@ -89,6 +89,8 @@
 #define SITE_SENTMSG5 7
 #define SITE_WAIT     8
 
+int32_t site_max_start_pad = 4*4;
+
 static cstring_t state_name(uint32_t state)
 {
     switch (state) {
@@ -204,7 +206,12 @@ static void transport_peers_copy(struct site *st, transport_peers *dst,
 static void transport_setup_msgok(struct site *st, const struct comm_addr *a);
 static void transport_data_msgok(struct site *st, const struct comm_addr *a);
 static bool_t transport_compute_setupinit_peers(struct site *st,
-        const struct comm_addr *configured_addr /* 0 if none or not found */);
+        const struct comm_addr *configured_addr /* 0 if none or not found */,
+        const struct comm_addr *prod_hint_addr /* 0 if none */);
+static void transport_resolve_complete(struct site *st,
+                                      const struct comm_addr *a);
+static void transport_resolve_complete_tardy(struct site *st,
+                                            const struct comm_addr *ca_use);
 static void transport_record_peer(struct site *st, transport_peers *peers,
                                  const struct comm_addr *addr, const char *m);
 
@@ -226,11 +233,12 @@ struct site {
 /* configuration information */
     string_t localname;
     string_t remotename;
-    bool_t peer_mobile; /* Mobile client support */
+    bool_t local_mobile, peer_mobile; /* Mobile client support */
     int32_t transport_peers_max;
     string_t tunname; /* localname<->remotename by default, used in logs */
     string_t address; /* DNS name for bootstrapping, optional */
     int remoteport; /* Port for bootstrapping, optional */
+    uint32_t mtu_target;
     struct netlink_if *netlink;
     struct comm_if **comms;
     int ncomms;
@@ -262,6 +270,8 @@ struct site {
 /* runtime information */
     uint32_t state;
     uint64_t now; /* Most recently seen time */
+    bool_t allow_send_prod;
+    bool_t resolving;
 
     /* The currently established session */
     struct data_key current;
@@ -278,6 +288,7 @@ struct site {
        timeout before we can listen for another setup packet); perhaps
        we should keep a list of 'bad' sources for setup packets. */
     uint32_t remote_capabilities;
+    uint16_t remote_adv_mtu;
     struct transform_if *chosen_transform;
     uint32_t setup_session_id;
     transport_peers setup_peers;
@@ -293,33 +304,75 @@ struct site {
     struct transform_inst_if *new_transform; /* For key setup/verify */
 };
 
-static void slog(struct site *st, uint32_t event, cstring_t msg, ...)
+static uint32_t event_log_priority(struct site *st, uint32_t event)
+{
+    if (!(event&st->log_events))
+       return 0;
+    switch(event) {
+    case LOG_UNEXPECTED:    return M_INFO;
+    case LOG_SETUP_INIT:    return M_INFO;
+    case LOG_SETUP_TIMEOUT: return M_NOTICE;
+    case LOG_ACTIVATE_KEY:  return M_INFO;
+    case LOG_TIMEOUT_KEY:   return M_INFO;
+    case LOG_SEC:           return M_SECURITY;
+    case LOG_STATE:         return M_DEBUG;
+    case LOG_DROP:          return M_DEBUG;
+    case LOG_DUMP:          return M_DEBUG;
+    case LOG_ERROR:         return M_ERR;
+    case LOG_PEER_ADDRS:    return M_DEBUG;
+    default:                return M_ERR;
+    }
+}
+
+static void vslog(struct site *st, uint32_t event, cstring_t msg, va_list ap)
+FORMAT(printf,3,0);
+static void vslog(struct site *st, uint32_t event, cstring_t msg, va_list ap)
 {
-    va_list ap;
-    char buf[240];
     uint32_t class;
 
+    class=event_log_priority(st, event);
+    if (class) {
+       slilog_part(st->log,class,"%s: ",st->tunname);
+       vslilog_part(st->log,class,msg,ap);
+       slilog_part(st->log,class,"\n");
+    }
+}
+
+static void slog(struct site *st, uint32_t event, cstring_t msg, ...)
+FORMAT(printf,3,4);
+static void slog(struct site *st, uint32_t event, cstring_t msg, ...)
+{
+    va_list ap;
     va_start(ap,msg);
+    vslog(st,event,msg,ap);
+    va_end(ap);
+}
 
-    if (event&st->log_events) {
-       switch(event) {
-       case LOG_UNEXPECTED: class=M_INFO; break;
-       case LOG_SETUP_INIT: class=M_INFO; break;
-       case LOG_SETUP_TIMEOUT: class=M_NOTICE; break;
-       case LOG_ACTIVATE_KEY: class=M_INFO; break;
-       case LOG_TIMEOUT_KEY: class=M_INFO; break;
-       case LOG_SEC: class=M_SECURITY; break;
-       case LOG_STATE: class=M_DEBUG; break;
-       case LOG_DROP: class=M_DEBUG; break;
-       case LOG_DUMP: class=M_DEBUG; break;
-       case LOG_ERROR: class=M_ERR; break;
-       case LOG_PEER_ADDRS: class=M_DEBUG; break;
-       default: class=M_ERR; break;
-       }
+static void logtimeout(struct site *st, const char *fmt, ...)
+FORMAT(printf,2,3);
+static void logtimeout(struct site *st, const char *fmt, ...)
+{
+    uint32_t class=event_log_priority(st,LOG_SETUP_TIMEOUT);
+    if (!class)
+       return;
 
-       vsnprintf(buf,sizeof(buf),msg,ap);
-       st->log->log(st->log->st,class,"%s: %s",st->tunname,buf);
+    va_list ap;
+    va_start(ap,fmt);
+
+    slilog_part(st->log,class,"%s: ",st->tunname);
+    vslilog_part(st->log,class,fmt,ap);
+
+    const char *delim;
+    int i;
+    for (i=0, delim=" (tried ";
+        i<st->setup_peers.npeers;
+        i++, delim=", ") {
+       transport_peer *peer=&st->setup_peers.peers[i];
+       const char *s=comm_addr_to_string(&peer->addr);
+       slilog_part(st->log,class,"%s%s",delim,s);
     }
+
+    slilog_part(st->log,class,")\n");
     va_end(ap);
 }
 
@@ -329,7 +382,8 @@ static void delete_one_key(struct site *st, struct data_key *key,
                           const char *reason /* may be 0 meaning don't log*/,
                           const char *which /* ignored if !reasonn */,
                           uint32_t loglevel /* ignored if !reasonn */);
-static bool_t initiate_key_setup(struct site *st, cstring_t reason);
+static bool_t initiate_key_setup(struct site *st, cstring_t reason,
+                                const struct comm_addr *prod_hint);
 static void enter_state_run(struct site *st);
 static bool_t enter_state_resolve(struct site *st);
 static bool_t enter_new_state(struct site *st,uint32_t next);
@@ -379,6 +433,14 @@ static void dispose_transform(struct transform_inst_if **transform_var)
     type=buf_unprepend_uint32((b)); \
     if (type!=(t)) return False; } while(0)
 
+static _Bool type_is_msg34(uint32_t type)
+{
+    return
+       type == LABEL_MSG3 ||
+       type == LABEL_MSG3BIS ||
+       type == LABEL_MSG4;
+}
+
 struct parsedname {
     int32_t len;
     uint8_t *name;
@@ -392,6 +454,7 @@ struct msg {
     struct parsedname remote;
     struct parsedname local;
     uint32_t remote_capabilities;
+    uint16_t remote_mtu;
     int capab_transformnum;
     uint8_t *nR;
     uint8_t *nL;
@@ -480,6 +543,9 @@ static bool_t generate_msg(struct site *st, uint32_t type, cstring_t what)
     if ((st->local_capabilities & CAPAB_EARLY) || (type != LABEL_MSG1)) {
        buf_append_uint32(&st->buffer,st->local_capabilities);
     }
+    if (type_is_msg34(type)) {
+       buf_append_uint16(&st->buffer,st->mtu_target);
+    }
     append_string_xinfo_done(&st->buffer,&xia);
 
     buf_append_string(&st->buffer,st->remotename);
@@ -535,11 +601,20 @@ static bool_t unpick_msg(struct site *st, uint32_t type,
     CHECK_TYPE(msg,type);
     if (!unpick_name(msg,&m->remote)) return False;
     m->remote_capabilities=0;
+    m->remote_mtu=0;
     if (m->remote.extrainfo.size) {
        CHECK_AVAIL(&m->remote.extrainfo,4);
        m->remote_capabilities=buf_unprepend_uint32(&m->remote.extrainfo);
     }
+    if (type_is_msg34(type) && m->remote.extrainfo.size) {
+       CHECK_AVAIL(&m->remote.extrainfo,2);
+       m->remote_mtu=buf_unprepend_uint16(&m->remote.extrainfo);
+    }
     if (!unpick_name(msg,&m->local)) return False;
+    if (type==LABEL_PROD) {
+       CHECK_EMPTY(msg);
+       return True;
+    }
     CHECK_AVAIL(msg,NONCELEN);
     m->nR=buf_unprepend(msg,NONCELEN);
     if (type==LABEL_MSG1) {
@@ -628,7 +703,6 @@ static bool_t process_msg1(struct site *st, struct buffer_if *msg1,
        process an incoming MSG1, and that the MSG1 has correct values
        of A and B. */
 
-    transport_record_peer(st,&st->setup_peers,src,"msg1");
     st->setup_session_id=m->source;
     st->remote_capabilities=m->remote_capabilities;
     memcpy(st->remoteN,m->nR,NONCELEN);
@@ -692,12 +766,34 @@ static bool_t generate_msg3(struct site *st)
                        "site:MSG3");
 }
 
+static bool_t process_msg3_msg4(struct site *st, struct msg *m)
+{
+    uint8_t *hash;
+    void *hst;
+
+    /* Check signature and store g^x mod m */
+    hash=safe_malloc(st->hash->len, "process_msg3_msg4");
+    hst=st->hash->init();
+    st->hash->update(hst,m->hashstart,m->hashlen);
+    st->hash->final(hst,hash);
+    /* Terminate signature with a '0' - cheating, but should be ok */
+    m->sig[m->siglen]=0;
+    if (!st->pubkey->check(st->pubkey->st,hash,st->hash->len,m->sig)) {
+       slog(st,LOG_SEC,"msg3/msg4 signature failed check!");
+       free(hash);
+       return False;
+    }
+    free(hash);
+
+    st->remote_adv_mtu=m->remote_mtu;
+
+    return True;
+}
+
 static bool_t process_msg3(struct site *st, struct buffer_if *msg3,
                           const struct comm_addr *src, uint32_t msgtype)
 {
     struct msg m;
-    uint8_t *hash;
-    void *hst;
     cstring_t err;
 
     assert(msgtype==LABEL_MSG3 || msgtype==LABEL_MSG3BIS);
@@ -730,19 +826,8 @@ static bool_t process_msg3(struct site *st, struct buffer_if *msg3,
  transform_found:
     st->chosen_transform=ti;
 
-    /* Check signature and store g^x mod m */
-    hash=safe_malloc(st->hash->len, "process_msg3");
-    hst=st->hash->init();
-    st->hash->update(hst,m.hashstart,m.hashlen);
-    st->hash->final(hst,hash);
-    /* Terminate signature with a '0' - cheating, but should be ok */
-    m.sig[m.siglen]=0;
-    if (!st->pubkey->check(st->pubkey->st,hash,st->hash->len,m.sig)) {
-       slog(st,LOG_SEC,"msg3 signature failed check!");
-       free(hash);
+    if (!process_msg3_msg4(st,&m))
        return False;
-    }
-    free(hash);
 
     /* Terminate their DH public key with a '0' */
     m.pk[m.pklen]=0;
@@ -766,8 +851,6 @@ static bool_t process_msg4(struct site *st, struct buffer_if *msg4,
                           const struct comm_addr *src)
 {
     struct msg m;
-    uint8_t *hash;
-    void *hst;
     cstring_t err;
 
     if (!unpick_msg(st,LABEL_MSG4,msg4,&m)) return False;
@@ -776,19 +859,8 @@ static bool_t process_msg4(struct site *st, struct buffer_if *msg4,
        return False;
     }
     
-    /* Check signature and store g^x mod m */
-    hash=safe_malloc(st->hash->len, "process_msg4");
-    hst=st->hash->init();
-    st->hash->update(hst,m.hashstart,m.hashlen);
-    st->hash->final(hst,hash);
-    /* Terminate signature with a '0' - cheating, but should be ok */
-    m.sig[m.siglen]=0;
-    if (!st->pubkey->check(st->pubkey->st,hash,st->hash->len,m.sig)) {
-       slog(st,LOG_SEC,"msg4 signature failed check!");
-       free(hash);
+    if (!process_msg3_msg4(st,&m))
        return False;
-    }
-    free(hash);
 
     /* Terminate their DH public key with a '0' */
     m.pk[m.pklen]=0;
@@ -824,7 +896,7 @@ static bool_t generate_msg5(struct site *st)
 
     BUF_ALLOC(&st->buffer,"site:MSG5");
     /* We are going to add four words to the message */
-    buffer_init(&st->buffer,st->new_transform->max_start_pad+(4*4));
+    buffer_init(&st->buffer,calculate_max_start_pad());
     /* Give the netlink code an opportunity to put its own stuff in the
        message (configuration information, etc.) */
     buf_prepend_uint32(&st->buffer,LABEL_MSG5);
@@ -871,7 +943,7 @@ static void create_msg6(struct site *st, struct transform_inst_if *transform,
 
     BUF_ALLOC(&st->buffer,"site:MSG6");
     /* We are going to add four words to the message */
-    buffer_init(&st->buffer,transform->max_start_pad+(4*4));
+    buffer_init(&st->buffer,calculate_max_start_pad());
     /* Give the netlink code an opportunity to put its own stuff in the
        message (configuration information, etc.) */
     buf_prepend_uint32(&st->buffer,LABEL_MSG6);
@@ -983,7 +1055,7 @@ static bool_t decrypt_msg0(struct site *st, struct buffer_if *msg0,
 
     slog(st,LOG_SEC,"transform: %s (aux: %s, new: %s)",
         transform_err,auxkey_err,newkey_err);
-    initiate_key_setup(st,"incoming message would not decrypt");
+    initiate_key_setup(st,"incoming message would not decrypt",0);
     send_nak(src,m.dest,m.source,m.type,msg0,"message would not decrypt");
     return False;
 
@@ -1013,7 +1085,7 @@ static bool_t process_msg0(struct site *st, struct buffer_if *msg0,
        transport_data_msgok(st,src);
        /* See whether we should start negotiating a new key */
        if (st->now > st->renegotiate_key_time)
-           initiate_key_setup(st,"incoming packet in renegotiation window");
+           initiate_key_setup(st,"incoming packet in renegotiation window",0);
        return True;
     default:
        slog(st,LOG_SEC,"incoming encrypted message of type %08x "
@@ -1049,7 +1121,7 @@ static bool_t send_msg(struct site *st)
        st->retries--;
        return True;
     } else if (st->state==SITE_SENTMSG5) {
-       slog(st,LOG_SETUP_TIMEOUT,"timed out sending MSG5, stashing new key");
+       logtimeout(st,"timed out sending MSG5, stashing new key");
        /* We stash the key we have produced, in case it turns out that
         * our peer did see our MSG5 after all and starts using it. */
        /* This is a bit like some of activate_new_key */
@@ -1067,7 +1139,7 @@ static bool_t send_msg(struct site *st)
        enter_state_wait(st);
        return False;
     } else {
-       slog(st,LOG_SETUP_TIMEOUT,"timed out sending key setup packet "
+       logtimeout(st,"timed out sending key setup packet "
            "(in state %s)",state_name(st->state));
        enter_state_wait(st);
        return False;
@@ -1079,10 +1151,8 @@ static void site_resolve_callback(void *sst, struct in_addr *address)
     struct site *st=sst;
     struct comm_addr ca_buf, *ca_use;
 
-    if (st->state!=SITE_RESOLVE) {
-       slog(st,LOG_UNEXPECTED,"site_resolve_callback called unexpectedly");
-       return;
-    }
+    st->resolving=False;
+
     if (address) {
        FILLZERO(ca_buf);
        ca_buf.comm=st->comms[0];
@@ -1090,27 +1160,78 @@ static void site_resolve_callback(void *sst, struct in_addr *address)
        ca_buf.sin.sin_port=htons(st->remoteport);
        ca_buf.sin.sin_addr=*address;
        ca_use=&ca_buf;
+       slog(st,LOG_STATE,"resolution of %s completed: %s",
+            st->address, comm_addr_to_string(ca_use));;
     } else {
        slog(st,LOG_ERROR,"resolution of %s failed",st->address);
        ca_use=0;
     }
-    if (transport_compute_setupinit_peers(st,ca_use)) {
-       enter_new_state(st,SITE_SENTMSG1);
-    } else {
-       /* Can't figure out who to try to to talk to */
-       slog(st,LOG_SETUP_INIT,"key exchange failed: cannot find peer address");
-       enter_state_run(st);
+
+    switch (st->state) {
+    case SITE_RESOLVE:
+        if (transport_compute_setupinit_peers(st,ca_use,0)) {
+           enter_new_state(st,SITE_SENTMSG1);
+       } else {
+           /* Can't figure out who to try to to talk to */
+           slog(st,LOG_SETUP_INIT,
+                "key exchange failed: cannot find peer address");
+           enter_state_run(st);
+       }
+       break;
+    case SITE_SENTMSG1: case SITE_SENTMSG2:
+    case SITE_SENTMSG3: case SITE_SENTMSG4:
+    case SITE_SENTMSG5:
+       if (ca_use) {
+           /* We start using the address immediately for data too.
+            * It's best to store it in st->peers now because we might
+            * go via SENTMSG5, WAIT, and a MSG0, straight into using
+            * the new key (without updating the data peer addrs). */
+           transport_resolve_complete(st,ca_use);
+       } else if (st->local_mobile) {
+           /* We can't let this rest because we may have a peer
+            * address which will break in the future. */
+           slog(st,LOG_SETUP_INIT,"resolution of %s failed: "
+                "abandoning key exchange",st->address);
+           enter_state_wait(st);
+       } else {
+           slog(st,LOG_SETUP_INIT,"resolution of %s failed: "
+                " continuing to use source address of peer's packets"
+                " for key exchange and ultimately data",
+                st->address);
+       }
+       break;
+    case SITE_RUN:
+       if (ca_use) {
+           slog(st,LOG_SETUP_INIT,"resolution of %s completed tardily,"
+                " updating peer address(es)",st->address);
+           transport_resolve_complete_tardy(st,ca_use);
+       } else if (st->local_mobile) {
+           /* Not very good.  We should queue (another) renegotiation
+            * so that we can update the peer address. */
+           st->key_renegotiate_time=st->now+st->wait_timeout;
+       } else {
+           slog(st,LOG_SETUP_INIT,"resolution of %s failed: "
+                " continuing to use source address of peer's packets",
+                st->address);
+       }
+       break;
+    case SITE_WAIT:
+    case SITE_STOP:
+       /* oh well */
+       break;
     }
 }
 
-static bool_t initiate_key_setup(struct site *st, cstring_t reason)
+static bool_t initiate_key_setup(struct site *st, cstring_t reason,
+                                const struct comm_addr *prod_hint)
 {
+    /* Reentrancy hazard: can call enter_new_state/enter_state_* */
     if (st->state!=SITE_RUN) return False;
     slog(st,LOG_SETUP_INIT,"initiating key exchange (%s)",reason);
     if (st->address) {
        slog(st,LOG_SETUP_INIT,"resolving peer address");
        return enter_state_resolve(st);
-    } else if (transport_compute_setupinit_peers(st,0)) {
+    } else if (transport_compute_setupinit_peers(st,0,prod_hint)) {
        return enter_new_state(st,SITE_SENTMSG1);
     }
     slog(st,LOG_SETUP_INIT,"key exchange failed: no address for peer");
@@ -1137,7 +1258,15 @@ static void activate_new_key(struct site *st)
     transport_peers_copy(st,&st->peers,&st->setup_peers);
     st->current.remote_session_id=st->setup_session_id;
 
-    slog(st,LOG_ACTIVATE_KEY,"new key activated");
+    /* Compute the inter-site MTU.  This is min( our_mtu, their_mtu ).
+     * But their mtu be unspecified, in which case we just use ours. */
+    uint32_t intersite_mtu=
+       MIN(st->mtu_target, st->remote_adv_mtu ?: ~(uint32_t)0);
+    st->netlink->set_mtu(st->netlink->st,intersite_mtu);
+
+    slog(st,LOG_ACTIVATE_KEY,"new key activated"
+        " (mtu ours=%"PRId32" theirs=%"PRId32" intersite=%"PRId32")",
+        st->mtu_target, st->remote_adv_mtu, intersite_mtu);
     enter_state_run(st);
 }
 
@@ -1207,14 +1336,35 @@ static void enter_state_run(struct site *st)
     set_link_quality(st);
 }
 
+static bool_t ensure_resolving(struct site *st)
+{
+    /* Reentrancy hazard: may call site_resolve_callback and hence
+     * enter_new_state, enter_state_* and generate_msg*. */
+    if (st->resolving)
+        return True;
+
+    assert(st->address);
+
+    /* resolver->request might reentrantly call site_resolve_callback
+     * which will clear st->resolving, so we need to set it beforehand
+     * rather than afterwards; also, it might return False, in which
+     * case we have to clear ->resolving again. */
+    st->resolving=True;
+    bool_t ok = st->resolver->request(st->resolver->st,st->address,
+                                     site_resolve_callback,st);
+    if (!ok)
+       st->resolving=False;
+
+    return ok;
+}
+
 static bool_t enter_state_resolve(struct site *st)
 {
+    /* Reentrancy hazard!  See ensure_resolving. */
     state_assert(st,st->state==SITE_RUN);
     slog(st,LOG_STATE,"entering state RESOLVE");
     st->state=SITE_RESOLVE;
-    st->resolver->request(st->resolver->st,st->address,
-                         site_resolve_callback,st);
-    return True;
+    return ensure_resolving(st);
 }
 
 static bool_t enter_new_state(struct site *st, uint32_t next)
@@ -1292,7 +1442,7 @@ static bool_t send_msg7(struct site *st, cstring_t reason)
     if (current_valid(st) && st->buffer.free
        && transport_peers_valid(&st->peers)) {
        BUF_ALLOC(&st->buffer,"site:MSG7");
-       buffer_init(&st->buffer,st->current.transform->max_start_pad+(4*3));
+       buffer_init(&st->buffer,calculate_max_start_pad());
        buf_append_uint32(&st->buffer,LABEL_MSG7);
        buf_append_string(&st->buffer,reason);
        if (call_transform_forwards(st, st->current.transform,
@@ -1322,6 +1472,30 @@ static void enter_state_wait(struct site *st)
     /* XXX Erase keys etc. */
 }
 
+static void generate_prod(struct site *st, struct buffer_if *buf)
+{
+    buffer_init(buf,0);
+    buf_append_uint32(buf,0);
+    buf_append_uint32(buf,0);
+    buf_append_uint32(buf,LABEL_PROD);
+    buf_append_string(buf,st->localname);
+    buf_append_string(buf,st->remotename);
+}
+
+static void generate_send_prod(struct site *st,
+                              const struct comm_addr *source)
+{
+    if (!st->allow_send_prod) return; /* too soon */
+    if (!(st->state==SITE_RUN || st->state==SITE_RESOLVE ||
+         st->state==SITE_WAIT)) return; /* we'd ignore peer's MSG1 */
+
+    slog(st,LOG_SETUP_INIT,"prodding peer for key exchange");
+    st->allow_send_prod=0;
+    generate_prod(st,&st->scratch);
+    dump_packet(st,&st->scratch,source,False);
+    source->comm->sendmsg(source->comm->st, &st->scratch, source);
+}
+
 static inline void site_settimeout(uint64_t timeout, int *timeout_io)
 {
     if (timeout) {
@@ -1394,6 +1568,8 @@ static void site_outgoing(void *sst, struct buffer_if *buf)
        return;
     }
 
+    st->allow_send_prod=1;
+
     /* In all other states we consider delivering the packet if we have
        a valid key and a valid address to send it to. */
     if (current_valid(st) && transport_peers_valid(&st->peers)) {
@@ -1415,7 +1591,7 @@ static void site_outgoing(void *sst, struct buffer_if *buf)
 
     slog(st,LOG_DROP,"discarding outgoing packet of size %d",buf->size);
     BUF_FREE(buf);
-    initiate_key_setup(st,"outgoing packet");
+    initiate_key_setup(st,"outgoing packet",0);
 }
 
 static bool_t named_for_us(struct site *st, const struct buffer_if *buf_in,
@@ -1431,7 +1607,10 @@ static bool_t named_for_us(struct site *st, const struct buffer_if *buf_in,
 }
 
 /* This function is called by the communication device to deliver
-   packets from our peers. */
+   packets from our peers.
+   It should return True if the packet is recognised as being for
+   this current site instance (and should therefore not be processed
+   by other sites), even if the packet was otherwise ignored. */
 static bool_t site_incoming(void *sst, struct buffer_if *buf,
                            const struct comm_addr *source)
 {
@@ -1451,9 +1630,14 @@ static bool_t site_incoming(void *sst, struct buffer_if *buf,
        if (st->state==SITE_RUN || st->state==SITE_RESOLVE ||
            st->state==SITE_WAIT) {
            /* We should definitely process it */
+           transport_record_peer(st,&st->setup_peers,source,"msg1");
            if (process_msg1(st,buf,source,&named_msg)) {
                slog(st,LOG_SETUP_INIT,"key setup initiated by peer");
-               enter_new_state(st,SITE_SENTMSG2);
+               bool_t entered=enter_new_state(st,SITE_SENTMSG2);
+               if (entered && st->address && st->local_mobile)
+                   /* We must do this as the very last thing, because
+                      the resolver callback might reenter us. */
+                   ensure_resolving(st);
            } else {
                slog(st,LOG_ERROR,"failed to process incoming msg1");
            }
@@ -1473,6 +1657,7 @@ static bool_t site_incoming(void *sst, struct buffer_if *buf,
                     "priority => use incoming msg1");
                if (process_msg1(st,buf,source,&named_msg)) {
                    BUF_FREE(&st->buffer); /* Free our old message 1 */
+                   transport_setup_msgok(st,source);
                    enter_new_state(st,SITE_SENTMSG2);
                } else {
                    slog(st,LOG_ERROR,"failed to process an incoming "
@@ -1488,6 +1673,20 @@ static bool_t site_incoming(void *sst, struct buffer_if *buf,
        BUF_FREE(buf);
        return True;
     }
+    if (msgtype==LABEL_PROD) {
+       if (!named_for_us(st,buf,msgtype,&named_msg))
+           return False;
+       dump_packet(st,buf,source,True);
+       if (st->state!=SITE_RUN) {
+           slog(st,LOG_DROP,"ignoring PROD when not in state RUN");
+       } else if (current_valid(st)) {
+           slog(st,LOG_DROP,"ignoring PROD when we think we have a key");
+       } else {
+           initiate_key_setup(st,"peer sent PROD packet",source);
+       }
+       BUF_FREE(buf);
+       return True;
+    }
     if (dest==st->index) {
        /* Explicitly addressed to us */
        if (msgtype!=LABEL_MSG0) dump_packet(st,buf,source,True);
@@ -1496,7 +1695,9 @@ static bool_t site_incoming(void *sst, struct buffer_if *buf,
            /* If the source is our current peer then initiate a key setup,
               because our peer's forgotten the key */
            if (get_uint32(buf->start+4)==st->current.remote_session_id) {
-               initiate_key_setup(st,"received a NAK");
+               bool_t initiated;
+               initiated = initiate_key_setup(st,"received a NAK",0);
+               if (!initiated) generate_send_prod(st,source);
            } else {
                slog(st,LOG_SEC,"bad incoming NAK");
            }
@@ -1640,7 +1841,7 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     st->remotename=dict_read_string(dict, "name", True, "site", loc);
 
     st->peer_mobile=dict_read_bool(dict,"mobile",False,"site",loc,False);
-    bool_t local_mobile=
+    st->local_mobile=
        dict_read_bool(dict,"local-mobile",False,"site",loc,False);
 
     /* Sanity check (which also allows the 'sites' file to include
@@ -1649,14 +1850,14 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     if (strcmp(st->localname,st->remotename)==0) {
        Message(M_DEBUG,"site %s: local-name==name -> ignoring this site\n",
                st->localname);
-       if (st->peer_mobile != local_mobile)
+       if (st->peer_mobile != st->local_mobile)
            cfgfatal(loc,"site","site %s's peer-mobile=%d"
                    " but our local-mobile=%d\n",
-                   st->localname, st->peer_mobile, local_mobile);
+                   st->localname, st->peer_mobile, st->local_mobile);
        free(st);
        return NULL;
     }
-    if (st->peer_mobile && local_mobile) {
+    if (st->peer_mobile && st->local_mobile) {
        Message(M_WARNING,"site %s: site is mobile but so are we"
                " -> ignoring this site\n", st->remotename);
        free(st);
@@ -1704,7 +1905,7 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     st->dh=find_cl_if(dict,"dh",CL_DH,True,"site",loc);
     st->hash=find_cl_if(dict,"hash",CL_HASH,True,"site",loc);
 
-#define DEFAULT(D) (st->peer_mobile || local_mobile    \
+#define DEFAULT(D) (st->peer_mobile || st->local_mobile        \
                     ? DEFAULT_MOBILE_##D : DEFAULT_##D)
 #define CFG_NUMBER(k,D) dict_read_number(dict,(k),False,"site",loc,DEFAULT(D));
 
@@ -1712,6 +1913,7 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     st->setup_retries=        CFG_NUMBER("setup-retries", SETUP_RETRIES);
     st->setup_retry_interval= CFG_NUMBER("setup-timeout", SETUP_RETRY_INTERVAL);
     st->wait_timeout=         CFG_NUMBER("wait-time",     WAIT_TIME);
+    st->mtu_target= dict_read_number(dict,"mtu-target",False,"site",loc,0);
 
     st->mobile_peer_expiry= dict_read_number(
        dict,"mobile-peer-expiry",False,"site",loc,DEFAULT_MOBILE_PEER_EXPIRY);
@@ -1738,6 +1940,9 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     st->log_events=string_list_to_word(dict_lookup(dict,"log-events"),
                                       log_event_table,"site");
 
+    st->resolving=False;
+    st->allow_send_prod=0;
+
     st->tunname=safe_malloc(strlen(st->localname)+strlen(st->remotename)+5,
                            "site_apply");
     sprintf(st->tunname,"%s<->%s",st->localname,st->remotename);
@@ -1749,7 +1954,7 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
 
     buffer_new(&st->buffer,SETUP_BUFFER_LEN);
 
-    buffer_new(&st->scratch,0);
+    buffer_new(&st->scratch,SETUP_BUFFER_LEN);
     BUF_ALLOC(&st->scratch,"site:scratch");
 
     /* We are interested in poll(), but only for timeouts. We don't have
@@ -1768,17 +1973,6 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     st->sharedsecretlen=st->sharedsecretallocd=0;
     st->sharedsecret=0;
 
-    /* We need to compute some properties of our comms and transports */
-#define COMPUTE_WORST(things,pad)              \
-    int things##_worst_##pad=0;                        \
-    for (i=0; i<st->n##things; i++) {          \
-       int thispad=st->things[i]->pad;         \
-       if (thispad > things##_worst_##pad)     \
-           things##_worst_##pad=thispad;       \
-    }
-    COMPUTE_WORST(comms,min_start_pad)
-    COMPUTE_WORST(transforms,max_start_pad)
-
     for (i=0; i<st->ntransforms; i++) {
        struct transform_if *ti=st->transforms[i];
        uint32_t capbit = 1UL << ti->capab_transformnum;
@@ -1789,9 +1983,10 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     }
 
     /* We need to register the remote networks with the netlink device */
-    st->netlink->reg(st->netlink->st, site_outgoing, st,
-                    transforms_worst_max_start_pad+(4*4)+
-                    comms_worst_min_start_pad);
+    uint32_t netlink_mtu; /* local virtual interface mtu */
+    st->netlink->reg(st->netlink->st, site_outgoing, st, &netlink_mtu);
+    if (!st->mtu_target)
+       st->mtu_target=netlink_mtu;
     
     for (i=0; i<st->ncomms; i++)
        st->comms[i]->request_notify(st->comms[i]->st, st, site_incoming);
@@ -1836,14 +2031,14 @@ static void transport_peers_debug(struct site *st, transport_peers *dst,
         i++, (argp+=stride?stride:sizeof(*args))) {
        const struct comm_addr *ca=(void*)argp;
        slog(st, LOG_PEER_ADDRS, " args: addrs[%d]=%s",
-            i, ca->comm->addr_to_string(ca->comm->st,ca));
+            i, comm_addr_to_string(ca));
     }
     for (i=0; i<dst->npeers; i++) {
        struct timeval diff;
        timersub(tv_now,&dst->peers[i].last,&diff);
        const struct comm_addr *ca=&dst->peers[i].addr;
        slog(st, LOG_PEER_ADDRS, " peers: addrs[%d]=%s T-%ld.%06ld",
-            i, ca->comm->addr_to_string(ca->comm->st,ca),
+            i, comm_addr_to_string(ca),
             (unsigned long)diff.tv_sec, (unsigned long)diff.tv_usec);
     }
 }
@@ -1880,7 +2075,7 @@ static void transport_record_peer(struct site *st, transport_peers *peers,
 
     changed=1;
     if (peers->npeers==st->transport_peers_max)
-       slot=st->transport_peers_max;
+       slot=st->transport_peers_max-1;
     else
        slot=peers->npeers++;
 
@@ -1898,23 +2093,30 @@ static void transport_record_peer(struct site *st, transport_peers *peers,
 }
 
 static bool_t transport_compute_setupinit_peers(struct site *st,
-        const struct comm_addr *configured_addr /* 0 if none or not found */) {
+        const struct comm_addr *configured_addr /* 0 if none or not found */,
+        const struct comm_addr *prod_hint_addr /* 0 if none */) {
 
-    if (!configured_addr && !transport_peers_valid(&st->peers))
+    if (!configured_addr && !prod_hint_addr &&
+       !transport_peers_valid(&st->peers))
        return False;
 
     slog(st,LOG_SETUP_INIT,
-        (!configured_addr ? "using only %d old peer address(es)"
-         : "using configured address, and/or perhaps %d old peer address(es)"),
-        st->peers);
+        "using:%s%s %d old peer address(es)",
+        configured_addr ? " configured address;" : "",
+        prod_hint_addr ? " PROD hint address;" : "",
+        st->peers.npeers);
 
-    /* Non-mobile peers havve st->peers.npeers==0 or ==1, since they
+    /* Non-mobile peers have st->peers.npeers==0 or ==1, since they
      * have transport_peers_max==1.  The effect is that this code
      * always uses the configured address if supplied, or otherwise
-     * the existing data peer if one exists; this is as desired. */
+     * the address of the incoming PROD, or the existing data peer if
+     * one exists; this is as desired. */
 
     transport_peers_copy(st,&st->setup_peers,&st->peers);
 
+    if (prod_hint_addr)
+       transport_record_peer(st,&st->setup_peers,prod_hint_addr,"prod");
+
     if (configured_addr)
        transport_record_peer(st,&st->setup_peers,configured_addr,"setupinit");
 
@@ -1946,6 +2148,17 @@ static void transport_peers_copy(struct site *st, transport_peers *dst,
                          src->npeers, &src->peers->addr, sizeof(*src->peers));
 }
 
+static void transport_resolve_complete(struct site *st,
+                                      const struct comm_addr *ca_use) {
+    transport_record_peer(st,&st->peers,ca_use,"resolved data");
+    transport_record_peer(st,&st->setup_peers,ca_use,"resolved setup");
+}
+
+static void transport_resolve_complete_tardy(struct site *st,
+                                            const struct comm_addr *ca_use) {
+    transport_record_peer(st,&st->peers,ca_use,"resolved tardily");
+}
+
 void transport_xmit(struct site *st, transport_peers *peers,
                    struct buffer_if *buf, bool_t candebug) {
     int slot;