chiark / gitweb /
test-example: Provide test which uses unshare(8)
[secnet.git] / site.c
diff --git a/site.c b/site.c
index c0f43879deea6a62775b8408ccf1bf489581cd9f..324a2061682e69188605c04650d6e012d4993624 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 +206,8 @@ 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_record_peer(struct site *st, transport_peers *peers,
                                  const struct comm_addr *addr, const char *m);
 
@@ -214,6 +217,12 @@ static void transport_xmit(struct site *st, transport_peers *peers,
  /***** END of transport peers declarations *****/
 
 
+struct data_key {
+    struct transform_inst_if *transform;
+    uint64_t key_timeout; /* End of life of current key */
+    uint32_t remote_session_id;
+};
+
 struct site {
     closure_t cl;
     struct site_if ops;
@@ -233,11 +242,13 @@ struct site {
     struct random_if *random;
     struct rsaprivkey_if *privkey;
     struct rsapubkey_if *pubkey;
-    struct transform_if *transform;
+    struct transform_if **transforms;
+    int ntransforms;
     struct dh_if *dh;
     struct hash_if *hash;
 
     uint32_t index; /* Index of this site */
+    uint32_t local_capabilities;
     int32_t setup_retries; /* How many times to send setup packets */
     int32_t setup_retry_interval; /* Initial timeout for setup packets */
     int32_t wait_timeout; /* How long to wait if setup unsuccessful */
@@ -247,9 +258,6 @@ struct site {
                                      after this time, initiate a new
                                      key exchange */
 
-    uint8_t *setupsig; /* Expected signature of incoming MSG1 packets */
-    int32_t setupsiglen; /* Allows us to discard packets quickly if
-                           they are not for us */
     bool_t setup_priority; /* Do we have precedence if both sites emit
                              message 1 simultaneously? */
     uint32_t log_events;
@@ -257,13 +265,14 @@ struct site {
 /* runtime information */
     uint32_t state;
     uint64_t now; /* Most recently seen time */
+    bool_t allow_send_prod;
 
     /* The currently established session */
-    uint32_t remote_session_id;
-    struct transform_inst_if *current_transform;
-    bool_t current_valid;
-    uint64_t current_key_timeout; /* End of life of current key */
+    struct data_key current;
+    struct data_key auxiliary_key;
+    bool_t auxiliary_is_new;
     uint64_t renegotiate_key_time; /* When we can negotiate a new key */
+    uint64_t auxiliary_renegotiate_key_time;
     transport_peers peers; /* Current address(es) of peer for data traffic */
 
     /* The current key setup protocol exchange.  We can only be
@@ -272,6 +281,8 @@ struct site {
        packet; we keep trying to continue the exchange, and have to
        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;
+    struct transform_if *chosen_transform;
     uint32_t setup_session_id;
     transport_peers setup_peers;
     uint8_t localN[NONCELEN]; /* Nonces for key exchange */
@@ -282,9 +293,12 @@ struct site {
     uint64_t timeout; /* Timeout for current state */
     uint8_t *dhsecret;
     uint8_t *sharedsecret;
+    uint32_t sharedsecretlen, sharedsecretallocd;
     struct transform_inst_if *new_transform; /* For key setup/verify */
 };
 
+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;
@@ -310,20 +324,61 @@ static void slog(struct site *st, uint32_t event, cstring_t msg, ...)
        }
 
        vsnprintf(buf,sizeof(buf),msg,ap);
-       st->log->log(st->log->st,class,"%s: %s",st->tunname,buf);
+       slilog(st->log,class,"%s: %s",st->tunname,buf);
     }
     va_end(ap);
 }
 
 static void set_link_quality(struct site *st);
-static void delete_key(struct site *st, cstring_t reason, uint32_t loglevel);
-static bool_t initiate_key_setup(struct site *st, cstring_t reason);
+static void delete_keys(struct site *st, cstring_t reason, uint32_t loglevel);
+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,
+                                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);
 static void enter_state_wait(struct site *st);
 static void activate_new_key(struct site *st);
 
+static bool_t is_transform_valid(struct transform_inst_if *transform)
+{
+    return transform && transform->valid(transform->st);
+}
+
+static bool_t current_valid(struct site *st)
+{
+    return is_transform_valid(st->current.transform);
+}
+
+#define DEFINE_CALL_TRANSFORM(fwdrev)                                  \
+static int call_transform_##fwdrev(struct site *st,                    \
+                                  struct transform_inst_if *transform, \
+                                  struct buffer_if *buf,               \
+                                  const char **errmsg)                 \
+{                                                                      \
+    if (!is_transform_valid(transform)) {                              \
+       *errmsg="transform not set up";                                 \
+       return 1;                                                       \
+    }                                                                  \
+    return transform->fwdrev(transform->st,buf,errmsg);                        \
+}
+
+DEFINE_CALL_TRANSFORM(forwards)
+DEFINE_CALL_TRANSFORM(reverse)
+
+static void dispose_transform(struct transform_inst_if **transform_var)
+{
+    struct transform_inst_if *transform=*transform_var;
+    if (transform) {
+       transform->delkey(transform->st);
+       transform->destroy(transform->st);
+    }
+    *transform_var = 0;
+}    
+
 #define CHECK_AVAIL(b,l) do { if ((b)->size<(l)) return False; } while(0)
 #define CHECK_EMPTY(b) do { if ((b)->size!=0) return False; } while(0)
 #define CHECK_TYPE(b,t) do { uint32_t type; \
@@ -331,14 +386,20 @@ static void activate_new_key(struct site *st);
     type=buf_unprepend_uint32((b)); \
     if (type!=(t)) return False; } while(0)
 
+struct parsedname {
+    int32_t len;
+    uint8_t *name;
+    struct buffer_if extrainfo;
+};
+
 struct msg {
     uint8_t *hashstart;
     uint32_t dest;
     uint32_t source;
-    int32_t remlen;
-    uint8_t *remote;
-    int32_t loclen;
-    uint8_t *local;
+    struct parsedname remote;
+    struct parsedname local;
+    uint32_t remote_capabilities;
+    int capab_transformnum;
     uint8_t *nR;
     uint8_t *nL;
     int32_t pklen;
@@ -348,6 +409,63 @@ struct msg {
     char *sig;
 };
 
+static void set_new_transform(struct site *st, char *pk)
+{
+    /* Make room for the shared key */
+    st->sharedsecretlen=st->chosen_transform->keylen?:st->dh->ceil_len;
+    assert(st->sharedsecretlen);
+    if (st->sharedsecretlen > st->sharedsecretallocd) {
+       st->sharedsecretallocd=st->sharedsecretlen;
+       st->sharedsecret=realloc(st->sharedsecret,st->sharedsecretallocd);
+    }
+    if (!st->sharedsecret) fatal_perror("site:sharedsecret");
+
+    /* Generate the shared key */
+    st->dh->makeshared(st->dh->st,st->dhsecret,st->dh->len,pk,
+                      st->sharedsecret,st->sharedsecretlen);
+
+    /* Set up the transform */
+    struct transform_if *generator=st->chosen_transform;
+    struct transform_inst_if *generated=generator->create(generator->st);
+    generated->setkey(generated->st,st->sharedsecret,
+                     st->sharedsecretlen,st->setup_priority);
+    dispose_transform(&st->new_transform);
+    st->new_transform=generated;
+
+    slog(st,LOG_SETUP_INIT,"key exchange negotiated transform"
+        " %d (capabilities ours=%#"PRIx32" theirs=%#"PRIx32")",
+        st->chosen_transform->capab_transformnum,
+        st->local_capabilities, st->remote_capabilities);
+}
+
+struct xinfoadd {
+    int32_t lenpos, afternul;
+};
+static void append_string_xinfo_start(struct buffer_if *buf,
+                                     struct xinfoadd *xia,
+                                     const char *str)
+    /* Helps construct one of the names with additional info as found
+     * in MSG1..4.  Call this function first, then append all the
+     * desired extra info (not including the nul byte) to the buffer,
+     * then call append_string_xinfo_done. */
+{
+    xia->lenpos = buf->size;
+    buf_append_string(buf,str);
+    buf_append_uint8(buf,0);
+    xia->afternul = buf->size;
+}
+static void append_string_xinfo_done(struct buffer_if *buf,
+                                    struct xinfoadd *xia)
+{
+    /* we just need to adjust the string length */
+    if (buf->size == xia->afternul) {
+       /* no extra info, strip the nul too */
+       buf_unappend_uint8(buf);
+    } else {
+       put_uint16(buf->start+xia->lenpos, buf->size-(xia->lenpos+2));
+    }
+}
+
 /* Build any of msg1 to msg4. msg5 and msg6 are built from the inside
    out using a transform of config data supplied by netlink */
 static bool_t generate_msg(struct site *st, uint32_t type, cstring_t what)
@@ -363,7 +481,14 @@ static bool_t generate_msg(struct site *st, uint32_t type, cstring_t what)
        (type==LABEL_MSG1?0:st->setup_session_id));
     buf_append_uint32(&st->buffer,st->index);
     buf_append_uint32(&st->buffer,type);
-    buf_append_string(&st->buffer,st->localname);
+
+    struct xinfoadd xia;
+    append_string_xinfo_start(&st->buffer,&xia,st->localname);
+    if ((st->local_capabilities & CAPAB_EARLY) || (type != LABEL_MSG1)) {
+       buf_append_uint32(&st->buffer,st->local_capabilities);
+    }
+    append_string_xinfo_done(&st->buffer,&xia);
+
     buf_append_string(&st->buffer,st->remotename);
     memcpy(buf_append(&st->buffer,NONCELEN),st->localN,NONCELEN);
     if (type==LABEL_MSG1) return True;
@@ -372,6 +497,9 @@ static bool_t generate_msg(struct site *st, uint32_t type, cstring_t what)
 
     if (hacky_par_mid_failnow()) return False;
 
+    if (type==LABEL_MSG3BIS)
+       buf_append_uint8(&st->buffer,st->chosen_transform->capab_transformnum);
+
     dhpub=st->dh->makepublic(st->dh->st,st->dhsecret,st->dh->len);
     buf_append_string(&st->buffer,dhpub);
     free(dhpub);
@@ -386,23 +514,43 @@ static bool_t generate_msg(struct site *st, uint32_t type, cstring_t what)
     return True;
 }
 
+static bool_t unpick_name(struct buffer_if *msg, struct parsedname *nm)
+{
+    CHECK_AVAIL(msg,2);
+    nm->len=buf_unprepend_uint16(msg);
+    CHECK_AVAIL(msg,nm->len);
+    nm->name=buf_unprepend(msg,nm->len);
+    uint8_t *nul=memchr(nm->name,0,nm->len);
+    if (!nul) {
+       buffer_readonly_view(&nm->extrainfo,0,0);
+    } else {
+       buffer_readonly_view(&nm->extrainfo, nul+1, msg->start-(nul+1));
+       nm->len=nul-nm->name;
+    }
+    return True;
+}
+
 static bool_t unpick_msg(struct site *st, uint32_t type,
                         struct buffer_if *msg, struct msg *m)
 {
+    m->capab_transformnum=-1;
     m->hashstart=msg->start;
     CHECK_AVAIL(msg,4);
     m->dest=buf_unprepend_uint32(msg);
     CHECK_AVAIL(msg,4);
     m->source=buf_unprepend_uint32(msg);
     CHECK_TYPE(msg,type);
-    CHECK_AVAIL(msg,2);
-    m->remlen=buf_unprepend_uint16(msg);
-    CHECK_AVAIL(msg,m->remlen);
-    m->remote=buf_unprepend(msg,m->remlen);
-    CHECK_AVAIL(msg,2);
-    m->loclen=buf_unprepend_uint16(msg);
-    CHECK_AVAIL(msg,m->loclen);
-    m->local=buf_unprepend(msg,m->loclen);
+    if (!unpick_name(msg,&m->remote)) return False;
+    m->remote_capabilities=0;
+    if (m->remote.extrainfo.size) {
+       CHECK_AVAIL(&m->remote.extrainfo,4);
+       m->remote_capabilities=buf_unprepend_uint32(&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) {
@@ -415,6 +563,12 @@ static bool_t unpick_msg(struct site *st, uint32_t type,
        CHECK_EMPTY(msg);
        return True;
     }
+    if (type==LABEL_MSG3BIS) {
+       CHECK_AVAIL(msg,1);
+       m->capab_transformnum = buf_unprepend_uint8(msg);
+    } else {
+       m->capab_transformnum = CAPAB_TRANSFORMNUM_ANCIENT;
+    }
     CHECK_AVAIL(msg,2);
     m->pklen=buf_unprepend_uint16(msg);
     CHECK_AVAIL(msg,m->pklen);
@@ -428,6 +582,14 @@ static bool_t unpick_msg(struct site *st, uint32_t type,
     return True;
 }
 
+static bool_t name_matches(const struct parsedname *nm, const char *expected)
+{
+    int expected_len=strlen(expected);
+    return
+       nm->len == expected_len &&
+       !memcmp(nm->name, expected, expected_len);
+}    
+
 static bool_t check_msg(struct site *st, uint32_t type, struct msg *m,
                        cstring_t *error)
 {
@@ -435,11 +597,11 @@ static bool_t check_msg(struct site *st, uint32_t type, struct msg *m,
 
     /* Check that the site names and our nonce have been sent
        back correctly, and then store our peer's nonce. */ 
-    if (memcmp(m->remote,st->remotename,strlen(st->remotename)!=0)) {
+    if (!name_matches(&m->remote,st->remotename)) {
        *error="wrong remote site name";
        return False;
     }
-    if (memcmp(m->local,st->localname,strlen(st->localname)!=0)) {
+    if (!name_matches(&m->local,st->localname)) {
        *error="wrong local site name";
        return False;
     }
@@ -448,11 +610,17 @@ static bool_t check_msg(struct site *st, uint32_t type, struct msg *m,
        return False;
     }
     if (type==LABEL_MSG2) return True;
-    if (memcmp(m->nR,st->remoteN,NONCELEN)!=0) {
+    if (!consttime_memeq(m->nR,st->remoteN,NONCELEN)!=0) {
        *error="wrong remotely-generated nonce";
        return False;
     }
-    if (type==LABEL_MSG3) return True;
+    /* MSG3 has complicated rules about capabilities, which are
+     * handled in process_msg3. */
+    if (type==LABEL_MSG3 || type==LABEL_MSG3BIS) return True;
+    if (m->remote_capabilities!=st->remote_capabilities) {
+       *error="remote capabilities changed";
+       return False;
+    }
     if (type==LABEL_MSG4) return True;
     *error="unknown message type";
     return False;
@@ -465,19 +633,16 @@ static bool_t generate_msg1(struct site *st)
 }
 
 static bool_t process_msg1(struct site *st, struct buffer_if *msg1,
-                          const struct comm_addr *src)
+                          const struct comm_addr *src, struct msg *m)
 {
-    struct msg m;
-
     /* We've already determined we're in an appropriate state to
        process an incoming MSG1, and that the MSG1 has correct values
        of A and B. */
 
-    if (!unpick_msg(st,LABEL_MSG1,msg1,&m)) return False;
-
     transport_record_peer(st,&st->setup_peers,src,"msg1");
-    st->setup_session_id=m.source;
-    memcpy(st->remoteN,m.nR,NONCELEN);
+    st->setup_session_id=m->source;
+    st->remote_capabilities=m->remote_capabilities;
+    memcpy(st->remoteN,m->nR,NONCELEN);
     return True;
 }
 
@@ -499,6 +664,30 @@ static bool_t process_msg2(struct site *st, struct buffer_if *msg2,
        return False;
     }
     st->setup_session_id=m.source;
+    st->remote_capabilities=m.remote_capabilities;
+
+    /* Select the transform to use */
+
+    uint32_t remote_transforms = st->remote_capabilities & CAPAB_TRANSFORM_MASK;
+    if (!remote_transforms)
+       /* old secnets only had this one transform */
+       remote_transforms = 1UL << CAPAB_TRANSFORMNUM_ANCIENT;
+
+    struct transform_if *ti;
+    int i;
+    for (i=0; i<st->ntransforms; i++) {
+       ti=st->transforms[i];
+       if ((1UL << ti->capab_transformnum) & remote_transforms)
+           goto transform_found;
+    }
+    slog(st,LOG_ERROR,"no transforms in common"
+        " (us %#"PRIx32"; them: %#"PRIx32")",
+        st->local_capabilities & CAPAB_TRANSFORM_MASK,
+        remote_transforms);
+    return False;
+ transform_found:
+    st->chosen_transform=ti;
+
     memcpy(st->remoteN,m.nR,NONCELEN);
     return True;
 }
@@ -508,22 +697,49 @@ static bool_t generate_msg3(struct site *st)
     /* Now we have our nonce and their nonce. Think of a secret key,
        and create message number 3. */
     st->random->generate(st->random->st,st->dh->len,st->dhsecret);
-    return generate_msg(st,LABEL_MSG3,"site:MSG3");
+    return generate_msg(st,
+                       (st->remote_capabilities & CAPAB_TRANSFORM_MASK
+                        ? LABEL_MSG3BIS : LABEL_MSG3),
+                       "site:MSG3");
 }
 
 static bool_t process_msg3(struct site *st, struct buffer_if *msg3,
-                          const struct comm_addr *src)
+                          const struct comm_addr *src, uint32_t msgtype)
 {
     struct msg m;
     uint8_t *hash;
     void *hst;
     cstring_t err;
 
-    if (!unpick_msg(st,LABEL_MSG3,msg3,&m)) return False;
-    if (!check_msg(st,LABEL_MSG3,&m,&err)) {
+    assert(msgtype==LABEL_MSG3 || msgtype==LABEL_MSG3BIS);
+
+    if (!unpick_msg(st,msgtype,msg3,&m)) return False;
+    if (!check_msg(st,msgtype,&m,&err)) {
        slog(st,LOG_SEC,"msg3: %s",err);
        return False;
     }
+    uint32_t capab_adv_late = m.remote_capabilities
+       & ~st->remote_capabilities & CAPAB_EARLY;
+    if (capab_adv_late) {
+       slog(st,LOG_SEC,"msg3 impermissibly adds early capability flag(s)"
+            " %#"PRIx32" (was %#"PRIx32", now %#"PRIx32")",
+            capab_adv_late, st->remote_capabilities, m.remote_capabilities);
+       return False;
+    }
+    st->remote_capabilities|=m.remote_capabilities;
+
+    struct transform_if *ti;
+    int i;
+    for (i=0; i<st->ntransforms; i++) {
+       ti=st->transforms[i];
+       if (ti->capab_transformnum == m.capab_transformnum)
+           goto transform_found;
+    }
+    slog(st,LOG_SEC,"peer chose unknown-to-us transform %d!",
+        m.capab_transformnum);
+    return False;
+ transform_found:
+    st->chosen_transform=ti;
 
     /* Check signature and store g^x mod m */
     hash=safe_malloc(st->hash->len, "process_msg3");
@@ -544,13 +760,8 @@ static bool_t process_msg3(struct site *st, struct buffer_if *msg3,
     /* Invent our DH secret key */
     st->random->generate(st->random->st,st->dh->len,st->dhsecret);
 
-    /* Generate the shared key */
-    st->dh->makeshared(st->dh->st,st->dhsecret,st->dh->len,m.pk,
-                      st->sharedsecret,st->transform->keylen);
-
-    /* Set up the transform */
-    st->new_transform->setkey(st->new_transform->st,st->sharedsecret,
-                             st->transform->keylen);
+    /* Generate the shared key and set up the transform */
+    set_new_transform(st,m.pk);
 
     return True;
 }
@@ -592,12 +803,9 @@ static bool_t process_msg4(struct site *st, struct buffer_if *msg4,
 
     /* Terminate their DH public key with a '0' */
     m.pk[m.pklen]=0;
-    /* Generate the shared key */
-    st->dh->makeshared(st->dh->st,st->dhsecret,st->dh->len,m.pk,
-                      st->sharedsecret,st->transform->keylen);
-    /* Set up the transform */
-    st->new_transform->setkey(st->new_transform->st,st->sharedsecret,
-                             st->transform->keylen);
+
+    /* Generate the shared key and set up the transform */
+    set_new_transform(st,m.pk);
 
     return True;
 }
@@ -627,12 +835,13 @@ 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->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);
-    st->new_transform->forwards(st->new_transform->st,&st->buffer,
-                               &transform_err);
+    if (call_transform_forwards(st,st->new_transform,
+                               &st->buffer,&transform_err))
+       return False;
     buf_prepend_uint32(&st->buffer,LABEL_MSG5);
     buf_prepend_uint32(&st->buffer,st->index);
     buf_prepend_uint32(&st->buffer,st->setup_session_id);
@@ -642,15 +851,15 @@ static bool_t generate_msg5(struct site *st)
 }
 
 static bool_t process_msg5(struct site *st, struct buffer_if *msg5,
-                          const struct comm_addr *src)
+                          const struct comm_addr *src,
+                          struct transform_inst_if *transform)
 {
     struct msg0 m;
     cstring_t transform_err;
 
     if (!unpick_msg0(st,msg5,&m)) return False;
 
-    if (st->new_transform->reverse(st->new_transform->st,
-                                  msg5,&transform_err)) {
+    if (call_transform_reverse(st,transform,msg5,&transform_err)) {
        /* There's a problem */
        slog(st,LOG_SEC,"process_msg5: transform: %s",transform_err);
        return False;
@@ -666,22 +875,30 @@ static bool_t process_msg5(struct site *st, struct buffer_if *msg5,
     return True;
 }
 
-static bool_t generate_msg6(struct site *st)
+static void create_msg6(struct site *st, struct transform_inst_if *transform,
+                       uint32_t session_id)
 {
     cstring_t transform_err;
 
     BUF_ALLOC(&st->buffer,"site:MSG6");
     /* We are going to add four words to the message */
-    buffer_init(&st->buffer,st->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);
-    st->new_transform->forwards(st->new_transform->st,&st->buffer,
-                               &transform_err);
+    int problem = call_transform_forwards(st,transform,
+                                         &st->buffer,&transform_err);
+    assert(!problem);
     buf_prepend_uint32(&st->buffer,LABEL_MSG6);
     buf_prepend_uint32(&st->buffer,st->index);
-    buf_prepend_uint32(&st->buffer,st->setup_session_id);
+    buf_prepend_uint32(&st->buffer,session_id);
+}
 
+static bool_t generate_msg6(struct site *st)
+{
+    if (!is_transform_valid(st->new_transform))
+       return False;
+    create_msg6(st,st->new_transform,st->setup_session_id);
     st->retries=1; /* Peer will retransmit MSG5 if this packet gets lost */
     return True;
 }
@@ -694,8 +911,7 @@ static bool_t process_msg6(struct site *st, struct buffer_if *msg6,
 
     if (!unpick_msg0(st,msg6,&m)) return False;
 
-    if (st->new_transform->reverse(st->new_transform->st,
-                                  msg6,&transform_err)) {
+    if (call_transform_reverse(st,st->new_transform,msg6,&transform_err)) {
        /* There's a problem */
        slog(st,LOG_SEC,"process_msg6: transform: %s",transform_err);
        return False;
@@ -711,9 +927,10 @@ static bool_t process_msg6(struct site *st, struct buffer_if *msg6,
     return True;
 }
 
-static bool_t decrypt_msg0(struct site *st, struct buffer_if *msg0)
+static bool_t decrypt_msg0(struct site *st, struct buffer_if *msg0,
+                          const struct comm_addr *src)
 {
-    cstring_t transform_err, newkey_err="n/a";
+    cstring_t transform_err, auxkey_err, newkey_err="n/a";
     struct msg0 m;
     uint32_t problem;
 
@@ -722,19 +939,47 @@ static bool_t decrypt_msg0(struct site *st, struct buffer_if *msg0)
     /* Keep a copy so we can try decrypting it with multiple keys */
     buffer_copy(&st->scratch, msg0);
 
-    problem = st->current_transform->reverse(st->current_transform->st,
-                                            msg0,&transform_err);
-    if (!problem) return True;
-
-    if (problem==2) {
-       slog(st,LOG_DROP,"transform: %s (merely skew)",transform_err);
-       return False;
+    problem = call_transform_reverse(st,st->current.transform,
+                                    msg0,&transform_err);
+    if (!problem) {
+       if (!st->auxiliary_is_new)
+           delete_one_key(st,&st->auxiliary_key,
+                          "peer has used new key","auxiliary key",LOG_SEC);
+       return True;
+    }
+    if (problem==2)
+       goto skew;
+
+    buffer_copy(msg0, &st->scratch);
+    problem = call_transform_reverse(st,st->auxiliary_key.transform,
+                                    msg0,&auxkey_err);
+    if (problem==0) {
+       slog(st,LOG_DROP,"processing packet which uses auxiliary key");
+       if (st->auxiliary_is_new) {
+           /* We previously timed out in state SENTMSG5 but it turns
+            * out that our peer did in fact get our MSG5 and is
+            * using the new key.  So we should switch to it too. */
+           /* This is a bit like activate_new_key. */
+           struct data_key t;
+           t=st->current;
+           st->current=st->auxiliary_key;
+           st->auxiliary_key=t;
+
+           delete_one_key(st,&st->auxiliary_key,"peer has used new key",
+                          "previous key",LOG_SEC);
+           st->auxiliary_is_new=0;
+           st->renegotiate_key_time=st->auxiliary_renegotiate_key_time;
+       }
+       return True;
     }
+    if (problem==2)
+       goto skew;
 
     if (st->state==SITE_SENTMSG5) {
        buffer_copy(msg0, &st->scratch);
-       if (!st->new_transform->reverse(st->new_transform->st,
-                                       msg0,&newkey_err)) {
+       problem = call_transform_reverse(st,st->new_transform,
+                                        msg0,&newkey_err);
+       if (!problem) {
            /* It looks like we didn't get the peer's MSG6 */
            /* This is like a cut-down enter_new_state(SITE_RUN) */
            slog(st,LOG_STATE,"will enter state RUN (MSG0 with new key)");
@@ -743,10 +988,18 @@ static bool_t decrypt_msg0(struct site *st, struct buffer_if *msg0)
            activate_new_key(st);
            return True; /* do process the data in this packet */
        }
+       if (problem==2)
+           goto skew;
     }
 
-    slog(st,LOG_SEC,"transform: %s (new: %s)",transform_err,newkey_err);
-    initiate_key_setup(st,"incoming message would not decrypt");
+    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",0);
+    send_nak(src,m.dest,m.source,m.type,msg0,"message would not decrypt");
+    return False;
+
+ skew:
+    slog(st,LOG_DROP,"transform: %s (merely skew)",transform_err);
     return False;
 }
 
@@ -755,7 +1008,7 @@ static bool_t process_msg0(struct site *st, struct buffer_if *msg0,
 {
     uint32_t type;
 
-    if (!decrypt_msg0(st,msg0))
+    if (!decrypt_msg0(st,msg0,src))
        return False;
 
     CHECK_AVAIL(msg0,4);
@@ -763,7 +1016,7 @@ static bool_t process_msg0(struct site *st, struct buffer_if *msg0,
     switch(type) {
     case LABEL_MSG7:
        /* We must forget about the current session. */
-       delete_key(st,"request from peer",LOG_SEC);
+       delete_keys(st,"request from peer",LOG_SEC);
        return True;
     case LABEL_MSG9:
        /* Deliver to netlink layer */
@@ -771,7 +1024,7 @@ 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 "
@@ -784,9 +1037,9 @@ static bool_t process_msg0(struct site *st, struct buffer_if *msg0,
 static void dump_packet(struct site *st, struct buffer_if *buf,
                        const struct comm_addr *addr, bool_t incoming)
 {
-    uint32_t dest=ntohl(*(uint32_t *)buf->start);
-    uint32_t source=ntohl(*(uint32_t *)(buf->start+4));
-    uint32_t msgtype=ntohl(*(uint32_t *)(buf->start+8));
+    uint32_t dest=get_uint32(buf->start);
+    uint32_t source=get_uint32(buf->start+4);
+    uint32_t msgtype=get_uint32(buf->start+8);
 
     if (st->log_events & LOG_DUMP)
        slilog(st->log,M_DEBUG,"%s: %s: %08x<-%08x: %08x:",
@@ -806,6 +1059,24 @@ static bool_t send_msg(struct site *st)
        st->timeout=st->now+st->setup_retry_interval;
        st->retries--;
        return True;
+    } else if (st->state==SITE_SENTMSG5) {
+       slog(st,LOG_SETUP_TIMEOUT,"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 */
+       struct transform_inst_if *t;
+       t=st->auxiliary_key.transform;
+       st->auxiliary_key.transform=st->new_transform;
+       st->new_transform=t;
+       dispose_transform(&st->new_transform);
+
+       st->auxiliary_is_new=1;
+       st->auxiliary_key.key_timeout=st->now+st->key_lifetime;
+       st->auxiliary_renegotiate_key_time=st->now+st->key_renegotiate_time;
+       st->auxiliary_key.remote_session_id=st->setup_session_id;
+
+       enter_state_wait(st);
+       return False;
     } else {
        slog(st,LOG_SETUP_TIMEOUT,"timed out sending key setup packet "
            "(in state %s)",state_name(st->state));
@@ -834,7 +1105,7 @@ static void site_resolve_callback(void *sst, struct in_addr *address)
        slog(st,LOG_ERROR,"resolution of %s failed",st->address);
        ca_use=0;
     }
-    if (transport_compute_setupinit_peers(st,ca_use)) {
+    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 */
@@ -843,14 +1114,15 @@ static void site_resolve_callback(void *sst, struct in_addr *address)
     }
 }
 
-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)
 {
     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");
@@ -861,34 +1133,44 @@ static void activate_new_key(struct site *st)
 {
     struct transform_inst_if *t;
 
-    /* We have two transform instances, which we swap between active
-       and setup */
-    t=st->current_transform;
-    st->current_transform=st->new_transform;
+    /* We have three transform instances, which we swap between old,
+       active and setup */
+    t=st->auxiliary_key.transform;
+    st->auxiliary_key.transform=st->current.transform;
+    st->current.transform=st->new_transform;
     st->new_transform=t;
+    dispose_transform(&st->new_transform);
 
-    t->delkey(t->st);
     st->timeout=0;
-    st->current_valid=True;
-    st->current_key_timeout=st->now+st->key_lifetime;
+    st->auxiliary_is_new=0;
+    st->auxiliary_key.key_timeout=st->current.key_timeout;
+    st->current.key_timeout=st->now+st->key_lifetime;
     st->renegotiate_key_time=st->now+st->key_renegotiate_time;
     transport_peers_copy(st,&st->peers,&st->setup_peers);
-    st->remote_session_id=st->setup_session_id;
+    st->current.remote_session_id=st->setup_session_id;
 
     slog(st,LOG_ACTIVATE_KEY,"new key activated");
     enter_state_run(st);
 }
 
-static void delete_key(struct site *st, cstring_t reason, uint32_t loglevel)
+static void delete_one_key(struct site *st, struct data_key *key,
+                          cstring_t reason, cstring_t which, uint32_t loglevel)
 {
-    if (st->current_valid) {
+    if (!is_transform_valid(key->transform)) return;
+    if (reason) slog(st,loglevel,"%s deleted (%s)",which,reason);
+    dispose_transform(&key->transform);
+    key->key_timeout=0;
+}
+
+static void delete_keys(struct site *st, cstring_t reason, uint32_t loglevel)
+{
+    if (current_valid(st)) {
        slog(st,loglevel,"session closed (%s)",reason);
 
-       st->current_valid=False;
-       st->current_transform->delkey(st->current_transform->st);
-       st->current_key_timeout=0;
+       delete_one_key(st,&st->current,0,0,0);
        set_link_quality(st);
     }
+    delete_one_key(st,&st->auxiliary_key,0,0,0);
 }
 
 static void state_assert(struct site *st, bool_t ok)
@@ -900,14 +1182,14 @@ static void enter_state_stop(struct site *st)
 {
     st->state=SITE_STOP;
     st->timeout=0;
-    delete_key(st,"entering state STOP",LOG_TIMEOUT_KEY);
-    st->new_transform->delkey(st->new_transform->st);
+    delete_keys(st,"entering state STOP",LOG_TIMEOUT_KEY);
+    dispose_transform(&st->new_transform);
 }
 
 static void set_link_quality(struct site *st)
 {
     uint32_t quality;
-    if (st->current_valid)
+    if (current_valid(st))
        quality=LINK_QUALITY_UP;
     else if (st->state==SITE_WAIT || st->state==SITE_STOP)
        quality=LINK_QUALITY_DOWN;
@@ -931,9 +1213,9 @@ static void enter_state_run(struct site *st)
     transport_peers_clear(st,&st->setup_peers);
     memset(st->localN,0,NONCELEN);
     memset(st->remoteN,0,NONCELEN);
-    st->new_transform->delkey(st->new_transform->st);
+    dispose_transform(&st->new_transform);
     memset(st->dhsecret,0,st->dh->len);
-    memset(st->sharedsecret,0,st->transform->keylen);
+    memset(st->sharedsecret,0,st->sharedsecretlen);
     set_link_quality(st);
 }
 
@@ -1019,19 +1301,21 @@ static bool_t send_msg7(struct site *st, cstring_t reason)
 {
     cstring_t transform_err;
 
-    if (st->current_valid && st->buffer.free
+    if (current_valid(st) && st->buffer.free
        && transport_peers_valid(&st->peers)) {
        BUF_ALLOC(&st->buffer,"site:MSG7");
-       buffer_init(&st->buffer,st->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);
-       st->current_transform->forwards(st->current_transform->st,
-                                       &st->buffer, &transform_err);
+       if (call_transform_forwards(st, st->current.transform,
+                                   &st->buffer, &transform_err))
+           goto free_out;
        buf_prepend_uint32(&st->buffer,LABEL_MSG0);
        buf_prepend_uint32(&st->buffer,st->index);
-       buf_prepend_uint32(&st->buffer,st->remote_session_id);
+       buf_prepend_uint32(&st->buffer,st->current.remote_session_id);
        transport_xmit(st,&st->peers,&st->buffer,True);
        BUF_FREE(&st->buffer);
+    free_out:
        return True;
     }
     return False;
@@ -1050,6 +1334,30 @@ 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) {
@@ -1070,14 +1378,23 @@ static int site_beforepoll(void *sst, struct pollfd *fds, int *nfds_io,
     st->now=*now;
 
     /* Work out when our next timeout is. The earlier of 'timeout' or
-       'current_key_timeout'. A stored value of '0' indicates no timeout
+       'current.key_timeout'. A stored value of '0' indicates no timeout
        active. */
     site_settimeout(st->timeout, timeout_io);
-    site_settimeout(st->current_key_timeout, timeout_io);
+    site_settimeout(st->current.key_timeout, timeout_io);
+    site_settimeout(st->auxiliary_key.key_timeout, timeout_io);
 
     return 0; /* success */
 }
 
+static void check_expiry(struct site *st, struct data_key *key,
+                        const char *which)
+{
+    if (key->key_timeout && *now>key->key_timeout) {
+       delete_one_key(st,key,"maximum life exceeded",which,LOG_TIMEOUT_KEY);
+    }
+}
+
 /* NB site_afterpoll will be called before site_beforepoll is ever called */
 static void site_afterpoll(void *sst, struct pollfd *fds, int nfds)
 {
@@ -1096,9 +1413,8 @@ static void site_afterpoll(void *sst, struct pollfd *fds, int nfds)
                 st->state);
        }
     }
-    if (st->current_key_timeout && *now>st->current_key_timeout) {
-       delete_key(st,"maximum key life exceeded",LOG_TIMEOUT_KEY);
-    }
+    check_expiry(st,&st->current,"current key");
+    check_expiry(st,&st->auxiliary_key,"auxiliary key");
 }
 
 /* This function is called by the netlink device to deliver packets
@@ -1114,30 +1430,49 @@ 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 (st->current_valid && transport_peers_valid(&st->peers)) {
+    if (current_valid(st) && transport_peers_valid(&st->peers)) {
        /* Transform it and send it */
        if (buf->size>0) {
            buf_prepend_uint32(buf,LABEL_MSG9);
-           st->current_transform->forwards(st->current_transform->st,
-                                           buf, &transform_err);
+           if (call_transform_forwards(st, st->current.transform,
+                                       buf, &transform_err))
+               goto free_out;
            buf_prepend_uint32(buf,LABEL_MSG0);
            buf_prepend_uint32(buf,st->index);
-           buf_prepend_uint32(buf,st->remote_session_id);
+           buf_prepend_uint32(buf,st->current.remote_session_id);
            transport_xmit(st,&st->peers,buf,False);
        }
+    free_out:
        BUF_FREE(buf);
        return;
     }
 
     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,
+                          uint32_t type, struct msg *m)
+    /* For packets which are identified by the local and remote names.
+     * If it has our name and our peer's name in it it's for us. */
+{
+    struct buffer_if buf[1];
+    buffer_readonly_clone(buf,buf_in);
+    return unpick_msg(st,type,buf,m)
+       && name_matches(&m->remote,st->remotename)
+       && name_matches(&m->local,st->localname);
 }
 
 /* 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)
 {
@@ -1145,67 +1480,80 @@ static bool_t site_incoming(void *sst, struct buffer_if *buf,
 
     if (buf->size < 12) return False;
 
-    uint32_t dest=ntohl(*(uint32_t *)buf->start);
-
-    if (dest==0) {
-       /* It could be for any site - it should have LABEL_MSG1 and
-          might have our name and our peer's name in it */
-       if (buf->size<(st->setupsiglen+8+NONCELEN)) return False;
-       if (memcmp(buf->start+8,st->setupsig,st->setupsiglen)==0) {
-           /* It's addressed to us. Decide what to do about it. */
-           dump_packet(st,buf,source,True);
-           if (st->state==SITE_RUN || st->state==SITE_RESOLVE ||
-               st->state==SITE_WAIT) {
-               /* We should definitely process it */
-               if (process_msg1(st,buf,source)) {
-                   slog(st,LOG_SETUP_INIT,"key setup initiated by peer");
+    uint32_t dest=get_uint32(buf->start);
+    uint32_t msgtype=get_uint32(buf->start+8);
+    struct msg named_msg;
+
+    if (msgtype==LABEL_MSG1) {
+       if (!named_for_us(st,buf,msgtype,&named_msg))
+           return False;
+       /* It's a MSG1 addressed to us. Decide what to do about it. */
+       dump_packet(st,buf,source,True);
+       if (st->state==SITE_RUN || st->state==SITE_RESOLVE ||
+           st->state==SITE_WAIT) {
+           /* We should definitely process it */
+           if (process_msg1(st,buf,source,&named_msg)) {
+               slog(st,LOG_SETUP_INIT,"key setup initiated by peer");
+               enter_new_state(st,SITE_SENTMSG2);
+           } else {
+               slog(st,LOG_ERROR,"failed to process incoming msg1");
+           }
+           BUF_FREE(buf);
+           return True;
+       } else if (st->state==SITE_SENTMSG1) {
+           /* We've just sent a message 1! They may have crossed on
+              the wire. If we have priority then we ignore the
+              incoming one, otherwise we process it as usual. */
+           if (st->setup_priority) {
+               BUF_FREE(buf);
+               slog(st,LOG_DUMP,"crossed msg1s; we are higher "
+                    "priority => ignore incoming msg1");
+               return True;
+           } else {
+               slog(st,LOG_DUMP,"crossed msg1s; we are lower "
+                    "priority => use incoming msg1");
+               if (process_msg1(st,buf,source,&named_msg)) {
+                   BUF_FREE(&st->buffer); /* Free our old message 1 */
                    enter_new_state(st,SITE_SENTMSG2);
                } else {
-                   slog(st,LOG_ERROR,"failed to process incoming msg1");
+                   slog(st,LOG_ERROR,"failed to process an incoming "
+                        "crossed msg1 (we have low priority)");
                }
                BUF_FREE(buf);
                return True;
-           } else if (st->state==SITE_SENTMSG1) {
-               /* We've just sent a message 1! They may have crossed on
-                  the wire. If we have priority then we ignore the
-                  incoming one, otherwise we process it as usual. */
-               if (st->setup_priority) {
-                   BUF_FREE(buf);
-                   slog(st,LOG_DUMP,"crossed msg1s; we are higher "
-                        "priority => ignore incoming msg1");
-                   return True;
-               } else {
-                   slog(st,LOG_DUMP,"crossed msg1s; we are lower "
-                        "priority => use incoming msg1");
-                   if (process_msg1(st,buf,source)) {
-                       BUF_FREE(&st->buffer); /* Free our old message 1 */
-                       enter_new_state(st,SITE_SENTMSG2);
-                   } else {
-                       slog(st,LOG_ERROR,"failed to process an incoming "
-                            "crossed msg1 (we have low priority)");
-                   }
-                   BUF_FREE(buf);
-                   return True;
-               }
            }
-           /* The message 1 was received at an unexpected stage of the
-              key setup. XXX POLICY - what do we do? */
-           slog(st,LOG_UNEXPECTED,"unexpected incoming message 1");
-           BUF_FREE(buf);
-           return True;
        }
-       return False; /* Not for us. */
+       /* The message 1 was received at an unexpected stage of the
+          key setup. XXX POLICY - what do we do? */
+       slog(st,LOG_UNEXPECTED,"unexpected incoming message 1");
+       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 */
-       uint32_t msgtype=ntohl(get_uint32(buf->start+8));
        if (msgtype!=LABEL_MSG0) dump_packet(st,buf,source,True);
        switch (msgtype) {
-       case 0: /* NAK */
+       case LABEL_NAK:
            /* 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->remote_session_id) {
-               initiate_key_setup(st,"received a NAK");
+           if (get_uint32(buf->start+4)==st->current.remote_session_id) {
+               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");
            }
@@ -1230,10 +1578,11 @@ static bool_t site_incoming(void *sst, struct buffer_if *buf,
            }
            break;
        case LABEL_MSG3:
+       case LABEL_MSG3BIS:
            /* Setup packet: expected only in state SENTMSG2 */
            if (st->state!=SITE_SENTMSG2) {
                slog(st,LOG_UNEXPECTED,"unexpected MSG3");
-           } else if (process_msg3(st,buf,source)) {
+           } else if (process_msg3(st,buf,source,msgtype)) {
                transport_setup_msgok(st,source);
                enter_new_state(st,SITE_SENTMSG4);
            } else {
@@ -1258,13 +1607,26 @@ static bool_t site_incoming(void *sst, struct buffer_if *buf,
               case we discard it. The peer will realise that we
               are using the new key when they see our data packets.
               Until then the peer's data packets to us get discarded. */
-           if (st->state!=SITE_SENTMSG4) {
-               slog(st,LOG_UNEXPECTED,"unexpected MSG5");
-           } else if (process_msg5(st,buf,source)) {
-               transport_setup_msgok(st,source);
-               enter_new_state(st,SITE_RUN);
+           if (st->state==SITE_SENTMSG4) {
+               if (process_msg5(st,buf,source,st->new_transform)) {
+                   transport_setup_msgok(st,source);
+                   enter_new_state(st,SITE_RUN);
+               } else {
+                   slog(st,LOG_SEC,"invalid MSG5");
+               }
+           } else if (st->state==SITE_RUN) {
+               if (process_msg5(st,buf,source,st->current.transform)) {
+                   slog(st,LOG_DROP,"got MSG5, retransmitting MSG6");
+                   transport_setup_msgok(st,source);
+                   create_msg6(st,st->current.transform,
+                               st->current.remote_session_id);
+                   transport_xmit(st,&st->peers,&st->buffer,True);
+                   BUF_FREE(&st->buffer);
+               } else {
+                   slog(st,LOG_SEC,"invalid MSG5 (in state RUN)");
+               }
            } else {
-               slog(st,LOG_SEC,"invalid MSG5");
+               slog(st,LOG_UNEXPECTED,"unexpected MSG5");
            }
            break;
        case LABEL_MSG6:
@@ -1360,21 +1722,28 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
 
     assert(index_sequence < 0xffffffffUL);
     st->index = ++index_sequence;
+    st->local_capabilities = 0;
     st->netlink=find_cl_if(dict,"link",CL_NETLINK,True,"site",loc);
 
-    list_t *comms_cfg=dict_lookup(dict,"comm");
-    if (!comms_cfg) cfgfatal(loc,"site","closure list \"comm\" not found\n");
-    st->ncomms=list_length(comms_cfg);
-    st->comms=safe_malloc_ary(sizeof(*st->comms),st->ncomms,"comms");
-    assert(st->ncomms);
-    for (i=0; i<st->ncomms; i++) {
-       item_t *item=list_elem(comms_cfg,i);
-       if (item->type!=t_closure)
-           cfgfatal(loc,"site","comm is not a closure\n");
-       closure_t *cl=item->data.closure;
-       if (cl->type!=CL_COMM) cfgfatal(loc,"site","comm closure wrong type\n");
-       st->comms[i]=cl->interface;
-    }
+#define GET_CLOSURE_LIST(dictkey,things,nthings,CL_TYPE) do{           \
+    list_t *things##_cfg=dict_lookup(dict,dictkey);                    \
+    if (!things##_cfg)                                                 \
+       cfgfatal(loc,"site","closure list \"%s\" not found\n",dictkey); \
+    st->nthings=list_length(things##_cfg);                             \
+    st->things=safe_malloc_ary(sizeof(*st->things),st->nthings,dictkey "s"); \
+    assert(st->nthings);                                               \
+    for (i=0; i<st->nthings; i++) {                                    \
+       item_t *item=list_elem(things##_cfg,i);                         \
+       if (item->type!=t_closure)                                      \
+           cfgfatal(loc,"site","%s is not a closure\n",dictkey);       \
+       closure_t *cl=item->data.closure;                               \
+       if (cl->type!=CL_TYPE)                                          \
+           cfgfatal(loc,"site","%s closure wrong type\n",dictkey);     \
+       st->things[i]=cl->interface;                                    \
+    }                                                                  \
+}while(0)
+
+    GET_CLOSURE_LIST("comm",comms,ncomms,CL_COMM);
 
     st->resolver=find_cl_if(dict,"resolver",CL_RESOLVER,True,"site",loc);
     st->log=find_cl_if(dict,"log",CL_LOG,True,"site",loc);
@@ -1387,8 +1756,7 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     else st->remoteport=0;
     st->pubkey=find_cl_if(dict,"key",CL_RSAPUBKEY,True,"site",loc);
 
-    st->transform=
-       find_cl_if(dict,"transform",CL_TRANSFORM,True,"site",loc);
+    GET_CLOSURE_LIST("transform",transforms,ntransforms,CL_TRANSFORM);
 
     st->dh=find_cl_if(dict,"dh",CL_DH,True,"site",loc);
     st->hash=find_cl_if(dict,"hash",CL_HASH,True,"site",loc);
@@ -1427,6 +1795,8 @@ 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->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);
@@ -1434,19 +1804,11 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     /* The information we expect to see in incoming messages of type 1 */
     /* fixme: lots of unchecked overflows here, but the results are only
        corrupted packets rather than undefined behaviour */
-    st->setupsiglen=strlen(st->remotename)+strlen(st->localname)+8;
-    st->setupsig=safe_malloc(st->setupsiglen,"site_apply");
-    put_uint32(st->setupsig+0,LABEL_MSG1);
-    put_uint16(st->setupsig+4,strlen(st->remotename));
-    memcpy(&st->setupsig[6],st->remotename,strlen(st->remotename));
-    put_uint16(st->setupsig+(6+strlen(st->remotename)),strlen(st->localname));
-    memcpy(&st->setupsig[8+strlen(st->remotename)],st->localname,
-          strlen(st->localname));
     st->setup_priority=(strcmp(st->localname,st->remotename)>0);
 
     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
@@ -1454,36 +1816,36 @@ static list_t *site_apply(closure_t *self, struct cloc loc, dict_t *context,
     register_for_poll(st, site_beforepoll, site_afterpoll, 0, "site");
     st->timeout=0;
 
-    st->current_valid=False;
-    st->current_key_timeout=0;
+    st->remote_capabilities=0;
+    st->chosen_transform=0;
+    st->current.key_timeout=0;
+    st->auxiliary_key.key_timeout=0;
     transport_peers_clear(st,&st->peers);
     transport_peers_clear(st,&st->setup_peers);
     /* XXX mlock these */
     st->dhsecret=safe_malloc(st->dh->len,"site:dhsecret");
-    st->sharedsecret=safe_malloc(st->transform->keylen,"site:sharedsecret");
-
-    /* We need to compute some properties of our comms */
-#define COMPUTE_WORST(pad)                     \
-    int worst_##pad=0;                         \
-    for (i=0; i<st->ncomms; i++) {             \
-       int thispad=st->comms[i]->pad;          \
-       if (thispad > worst_##pad)              \
-           worst_##pad=thispad;                \
+    st->sharedsecretlen=st->sharedsecretallocd=0;
+    st->sharedsecret=0;
+
+    for (i=0; i<st->ntransforms; i++) {
+       struct transform_if *ti=st->transforms[i];
+       uint32_t capbit = 1UL << ti->capab_transformnum;
+       if (st->local_capabilities & capbit)
+           slog(st,LOG_ERROR,"transformnum capability bit"
+                " %d (%#"PRIx32") reused", ti->capab_transformnum, capbit);
+       st->local_capabilities |= capbit;
     }
-    COMPUTE_WORST(min_start_pad)
-    COMPUTE_WORST(min_end_pad)
 
     /* We need to register the remote networks with the netlink device */
-    st->netlink->reg(st->netlink->st, site_outgoing, st,
-                    st->transform->max_start_pad+(4*4)+
-                    worst_min_start_pad,
-                    st->transform->max_end_pad+worst_min_end_pad);
+    st->netlink->reg(st->netlink->st, site_outgoing, st);
     
     for (i=0; i<st->ncomms; i++)
        st->comms[i]->request_notify(st->comms[i]->st, st, site_incoming);
 
-    st->current_transform=st->transform->create(st->transform->st);
-    st->new_transform=st->transform->create(st->transform->st);
+    st->current.transform=0;
+    st->auxiliary_key.transform=0;
+    st->new_transform=0;
+    st->auxiliary_is_new=0;
 
     enter_state_stop(st);
 
@@ -1582,23 +1944,30 @@ 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
      * 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");