move assertion to top of function. Found by PVS-Studio
[freeradius.git] / src / main / process.c
index 0841ad8..7b4196c 100644 (file)
@@ -70,11 +70,16 @@ static char const *action_codes[] = {
 };
 
 #ifdef DEBUG_STATE_MACHINE
-#define TRACE_STATE_MACHINE if (debug_flag) do { struct timeval debug_tv; \
-                                                gettimeofday(&debug_tv, NULL);\
-                                                debug_tv.tv_sec -= fr_start_time;\
-                                                printf("(%u) %d.%06d ********\tSTATE %s action %s live M-%s C-%s\t********\n",\
-                                                       request->number, (int) debug_tv.tv_sec, (int) debug_tv.tv_usec,  __FUNCTION__, action_codes[action], master_state_names[request->master_state], child_state_names[request->child_state]); } while (0)
+#  define TRACE_STATE_MACHINE \
+if (rad_debug_lvl) do { \
+       struct timeval debug_tv; \
+       gettimeofday(&debug_tv, NULL); \
+       debug_tv.tv_sec -= fr_start_time; \
+       printf("(%u) %d.%06d ********\tSTATE %s action %s live M-%s C-%s\t********\n",\
+              request->number, (int) debug_tv.tv_sec, (int) debug_tv.tv_usec, \
+              __FUNCTION__, action_codes[action], master_state_names[request->master_state], \
+              child_state_names[request->child_state]); \
+} while (0)
 
 static char const *master_state_names[REQUEST_MASTER_NUM_STATES] = {
        "?",
@@ -94,18 +99,49 @@ static char const *child_state_names[REQUEST_CHILD_NUM_STATES] = {
 };
 
 #else
-#define TRACE_STATE_MACHINE {}
+#  define TRACE_STATE_MACHINE {}
 #endif
 
-/*
- *     Declare a state in the state machine.
+static NEVER_RETURNS void _rad_panic(char const *file, unsigned int line, char const *msg)
+{
+       ERROR("%s[%u]: %s", file, line, msg);
+       fr_exit_now(1);
+}
+
+#define rad_panic(x) _rad_panic(__FILE__, __LINE__, x)
+
+/** Declare a state in the state machine
+ *
+ * Expands to the start of a function definition for a given state.
  *
+ * @param _x the name of the state.
  */
-#define STATE_MACHINE_DECL(_x) static void CC_HINT(nonnull) _x(REQUEST *request, int action)
+#define STATE_MACHINE_DECL(_x) static void _x(REQUEST *request, int action)
 
-#define STATE_MACHINE_TIMER(_x) request->timer_action = _x; \
-               fr_event_insert(el, request_timer, request, \
-                               &when, &request->ev);
+static void request_timer(void *ctx);
+
+/** Insert #REQUEST back into the event heap, to continue executing at a future time
+ *
+ * @param file the state machine timer call occurred in.
+ * @param line the state machine timer call occurred on.
+ * @param request to set add the timer event for.
+ * @param when the event should fine.
+ * @param action to perform when we resume processing the request.
+ */
+static inline void state_machine_timer(char const *file, int line, REQUEST *request,
+                                      struct timeval *when, fr_state_action_t action)
+{
+       request->timer_action = action;
+       if (!fr_event_insert(el, request_timer, request, when, &request->ev)) {
+               _rad_panic(file, line, "Failed to insert event");
+       }
+}
+
+/** @copybrief state_machine_timer
+ *
+ * @param _x the action to perform when we resume processing the request.
+ */
+#define STATE_MACHINE_TIMER(_x) state_machine_timer(__FILE__, __LINE__, request, &when, _x)
 
 /*
  *     We need a different VERIFY_REQUEST macro in process.c
@@ -194,32 +230,30 @@ static char const *child_state_names[REQUEST_CHILD_NUM_STATES] = {
  *     being called from a child thread or the master, and then do
  *     different things based on that.
  */
-
-
 #ifdef WITH_PROXY
 static fr_packet_list_t *proxy_list = NULL;
 static TALLOC_CTX *proxy_ctx = NULL;
 #endif
 
 #ifdef HAVE_PTHREAD_H
-#ifdef WITH_PROXY
+#  ifdef WITH_PROXY
 static pthread_mutex_t proxy_mutex;
 static bool proxy_no_new_sockets = false;
-#endif
+#  endif
 
-#define PTHREAD_MUTEX_LOCK if (spawn_flag) pthread_mutex_lock
-#define PTHREAD_MUTEX_UNLOCK if (spawn_flag) pthread_mutex_unlock
+#  define PTHREAD_MUTEX_LOCK if (spawn_flag) pthread_mutex_lock
+#  define PTHREAD_MUTEX_UNLOCK if (spawn_flag) pthread_mutex_unlock
 
 static pthread_t NO_SUCH_CHILD_PID;
-#define NO_CHILD_THREAD request->child_pid = NO_SUCH_CHILD_PID
+#  define NO_CHILD_THREAD request->child_pid = NO_SUCH_CHILD_PID
 
 #else
 /*
  *     This is easier than ifdef's throughout the code.
  */
-#define PTHREAD_MUTEX_LOCK(_x)
-#define PTHREAD_MUTEX_UNLOCK(_x)
-#define NO_CHILD_THREAD
+#  define PTHREAD_MUTEX_LOCK(_x)
+#  define PTHREAD_MUTEX_UNLOCK(_x)
+#  define NO_CHILD_THREAD
 #endif
 
 #ifdef HAVE_PTHREAD_H
@@ -236,21 +270,27 @@ static bool we_are_master(void)
 /*
  *     Assertions are debug checks.
  */
-#ifndef NDEBUG
-#define ASSERT_MASTER  if (!we_are_master()) rad_panic("We are not master")
-#endif
+#  ifndef NDEBUG
+#    define ASSERT_MASTER      if (!we_are_master()) rad_panic("We are not master")
+#    endif
 #else
 
 /*
  *     No threads: we're always master.
  */
-#define we_are_master(_x) (1)
+#  define we_are_master(_x) (1)
 #endif /* HAVE_PTHREAD_H */
 
 #ifndef ASSERT_MASTER
-#define ASSERT_MASTER
+#  define ASSERT_MASTER
 #endif
 
+/*
+ *     Make state transitions simpler.
+ */
+#define FINAL_STATE(_x) NO_CHILD_THREAD; request->component = "<" #_x ">"; request->module = ""; request->child_state = _x
+
+
 static int event_new_fd(rad_listen_t *this);
 
 /*
@@ -261,8 +301,8 @@ static int event_new_fd(rad_listen_t *this);
 static rad_listen_t *new_listeners = NULL;
 
 static pthread_mutex_t fd_mutex;
-#define FD_MUTEX_LOCK if (spawn_flag) pthread_mutex_lock
-#define FD_MUTEX_UNLOCK if (spawn_flag) pthread_mutex_unlock
+#  define FD_MUTEX_LOCK if (spawn_flag) pthread_mutex_lock
+#  define FD_MUTEX_UNLOCK if (spawn_flag) pthread_mutex_unlock
 
 void radius_update_listener(rad_listen_t *this)
 {
@@ -303,38 +343,44 @@ void radius_update_listener(rad_listen_t *this)
 /*
  *     This is easier than ifdef's throughout the code.
  */
-#define FD_MUTEX_LOCK(_x)
-#define FD_MUTEX_UNLOCK(_x)
+#  define FD_MUTEX_LOCK(_x)
+#  define FD_MUTEX_UNLOCK(_x)
 #endif
 
 static int request_num_counter = 1;
 #ifdef WITH_PROXY
-static int request_will_proxy(REQUEST *request);
-static int request_proxy(REQUEST *request, int retransmit);
-STATE_MACHINE_DECL(proxy_wait_for_reply);
-STATE_MACHINE_DECL(proxy_no_reply);
-STATE_MACHINE_DECL(proxy_running);
-static int process_proxy_reply(REQUEST *request, RADIUS_PACKET *reply);
-static void remove_from_proxy_hash(REQUEST *request);
-static void remove_from_proxy_hash_nl(REQUEST *request, bool yank);
-static int insert_into_proxy_hash(REQUEST *request);
+static int request_will_proxy(REQUEST *request) CC_HINT(nonnull);
+static int request_proxy(REQUEST *request) CC_HINT(nonnull);
+STATE_MACHINE_DECL(request_ping) CC_HINT(nonnull);
+
+STATE_MACHINE_DECL(request_response_delay) CC_HINT(nonnull);
+STATE_MACHINE_DECL(request_cleanup_delay) CC_HINT(nonnull);
+STATE_MACHINE_DECL(request_running) CC_HINT(nonnull);
+STATE_MACHINE_DECL(request_done) CC_HINT(nonnull);
+
+STATE_MACHINE_DECL(proxy_no_reply) CC_HINT(nonnull);
+STATE_MACHINE_DECL(proxy_running) CC_HINT(nonnull);
+STATE_MACHINE_DECL(proxy_wait_for_reply) CC_HINT(nonnull);
+
+static int process_proxy_reply(REQUEST *request, RADIUS_PACKET *reply) CC_HINT(nonnull (1));
+static void remove_from_proxy_hash(REQUEST *request) CC_HINT(nonnull);
+static void remove_from_proxy_hash_nl(REQUEST *request, bool yank) CC_HINT(nonnull);
+static int insert_into_proxy_hash(REQUEST *request) CC_HINT(nonnull);
 #endif
 
 static REQUEST *request_setup(TALLOC_CTX *ctx, rad_listen_t *listener, RADIUS_PACKET *packet,
                              RADCLIENT *client, RAD_REQUEST_FUNP fun);
+static int request_pre_handler(REQUEST *request, UNUSED int action) CC_HINT(nonnull);
 
-STATE_MACHINE_DECL(request_response_delay);
-STATE_MACHINE_DECL(request_cleanup_delay);
-STATE_MACHINE_DECL(request_running);
 #ifdef WITH_COA
-static void request_coa_originate(REQUEST *request);
-STATE_MACHINE_DECL(coa_running);
-STATE_MACHINE_DECL(coa_wait_for_reply);
-STATE_MACHINE_DECL(coa_no_reply);
-static void coa_separate(REQUEST *request);
-#define COA_SEPARATE if (request->coa) coa_separate(request->coa);
+static void request_coa_originate(REQUEST *request) CC_HINT(nonnull);
+STATE_MACHINE_DECL(coa_wait_for_reply) CC_HINT(nonnull);
+STATE_MACHINE_DECL(coa_no_reply) CC_HINT(nonnull);
+STATE_MACHINE_DECL(coa_running) CC_HINT(nonnull);
+static void coa_separate(REQUEST *request) CC_HINT(nonnull);
+#  define COA_SEPARATE if (request->coa) coa_separate(request->coa);
 #else
-#define COA_SEPARATE
+#  define COA_SEPARATE
 #endif
 
 #define CHECK_FOR_STOP do { if (request->master_state == REQUEST_STOP_PROCESSING) {request_done(request, FR_ACTION_DONE);return;}} while (0)
@@ -344,17 +390,6 @@ static void coa_separate(REQUEST *request);
 
 #define INSERT_EVENT(_function, _ctx) if (!fr_event_insert(el, _function, _ctx, &((_ctx)->when), &((_ctx)->ev))) { _rad_panic(__FILE__, __LINE__, "Failed to insert event"); }
 
-static NEVER_RETURNS void _rad_panic(char const *file, unsigned int line, char const *msg)
-{
-       ERROR("[%s:%d] %s", file, line, msg);
-#ifndef NDEBUG
-       rad_assert(0 == 1);
-#endif
-       fr_exit(1);
-}
-
-#define rad_panic(x) _rad_panic(__FILE__, __LINE__, x)
-
 static void tv_add(struct timeval *tv, int usec_delay)
 {
        if (usec_delay >= USEC) {
@@ -380,6 +415,14 @@ static void debug_packet(REQUEST *request, RADIUS_PACKET *packet, bool received)
        if (!packet) return;
        if (!RDEBUG_ENABLED) return;
 
+#ifdef WITH_DETAIL
+       /*
+        *      Don't print IP addresses for detail files.
+        */
+       if (request->listener &&
+           (request->listener->type == RAD_LISTEN_DETAIL)) return;
+
+#endif
        /*
         *      Client-specific debugging re-prints the input
         *      packet into the client log.
@@ -387,31 +430,39 @@ static void debug_packet(REQUEST *request, RADIUS_PACKET *packet, bool received)
         *      This really belongs in a utility library
         */
        if (is_radius_code(packet->code)) {
-               RDEBUG("%s %s Id %i from %s:%i to %s:%i length %zu",
+               RDEBUG("%s %s Id %i from %s%s%s:%i to %s%s%s:%i length %zu",
                       received ? "Received" : "Sent",
                       fr_packet_codes[packet->code],
                       packet->id,
+                      packet->src_ipaddr.af == AF_INET6 ? "[" : "",
                       inet_ntop(packet->src_ipaddr.af,
                                 &packet->src_ipaddr.ipaddr,
                                 src_ipaddr, sizeof(src_ipaddr)),
+                      packet->src_ipaddr.af == AF_INET6 ? "]" : "",
                       packet->src_port,
+                      packet->dst_ipaddr.af == AF_INET6 ? "[" : "",
                       inet_ntop(packet->dst_ipaddr.af,
                                 &packet->dst_ipaddr.ipaddr,
                                 dst_ipaddr, sizeof(dst_ipaddr)),
+                      packet->dst_ipaddr.af == AF_INET6 ? "]" : "",
                       packet->dst_port,
                       packet->data_len);
        } else {
-               RDEBUG("%s code %i Id %i from %s:%i to %s:%i length %zu",
+               RDEBUG("%s code %u Id %i from %s%s%s:%i to %s%s%s:%i length %zu\n",
                       received ? "Received" : "Sent",
                       packet->code,
                       packet->id,
+                      packet->src_ipaddr.af == AF_INET6 ? "[" : "",
                       inet_ntop(packet->src_ipaddr.af,
                                 &packet->src_ipaddr.ipaddr,
                                 src_ipaddr, sizeof(src_ipaddr)),
+                      packet->src_ipaddr.af == AF_INET6 ? "]" : "",
                       packet->src_port,
+                      packet->dst_ipaddr.af == AF_INET6 ? "[" : "",
                       inet_ntop(packet->dst_ipaddr.af,
                                 &packet->dst_ipaddr.ipaddr,
                                 dst_ipaddr, sizeof(dst_ipaddr)),
+                      packet->dst_ipaddr.af == AF_INET6 ? "]" : "",
                       packet->dst_port,
                       packet->data_len);
        }
@@ -434,6 +485,8 @@ static struct timeval *request_response_window(REQUEST *request)
 {
        VERIFY_REQUEST(request);
 
+       rad_assert(request->home_server != NULL);
+
        if (request->client) {
                /*
                 *      The client hasn't set the response window.  Return
@@ -449,7 +502,6 @@ static struct timeval *request_response_window(REQUEST *request)
                }
        }
 
-       rad_assert(request->home_server != NULL);
        return &request->home_server->response_window;
 }
 
@@ -530,10 +582,21 @@ static void proxy_reply_too_late(REQUEST *request)
 #endif
 
 
-/*
- *     Only ever called from the master thread.
+/** Mark a request DONE and clean it up.
+ *
+ *  When a request is DONE, it can have ties to a number of other
+ *  portions of the server.  The request hash, proxy hash, events,
+ *  child threads, etc.  This function takes care of either cleaning
+ *  up the request, or managing the timers to wait for the ties to be
+ *  removed.
+ *
+ *  \dot
+ *     digraph done {
+ *             done -> done [ label = "still running" ];
+ *     }
+ *  \enddot
  */
-STATE_MACHINE_DECL(request_done)
+static void request_done(REQUEST *request, int action)
 {
        struct timeval now, when;
 
@@ -565,8 +628,7 @@ STATE_MACHINE_DECL(request_done)
         *      and wait for the master thread timer to clean us up.
         */
        if (!we_are_master()) {
-               NO_CHILD_THREAD;
-               request->child_state = REQUEST_DONE;
+               FINAL_STATE(REQUEST_DONE);
                return;
        }
 #endif
@@ -619,7 +681,7 @@ STATE_MACHINE_DECL(request_done)
 #endif
 
 #ifdef DEBUG_STATE_MACHINE
-               if (debug_flag) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n",
+               if (rad_debug_lvl) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n",
                                       request->number, __FUNCTION__,
                                       child_state_names[request->child_state],
                                       child_state_names[REQUEST_DONE]);
@@ -736,8 +798,7 @@ STATE_MACHINE_DECL(request_done)
                 *      If we're the last one, remove the listener now.
                 */
                if ((request->listener->count == 0) &&
-                   (request->listener->status == RAD_LISTEN_STATUS_EOL)) {
-                       request->listener->status = RAD_LISTEN_STATUS_REMOVE_NOW;
+                   (request->listener->status >= RAD_LISTEN_STATUS_FROZEN)) {
                        event_new_fd(request->listener);
                }
        }
@@ -761,13 +822,27 @@ static void request_cleanup_delay_init(REQUEST *request)
 
        VERIFY_REQUEST(request);
 
-       if (request->packet->code == PW_CODE_ACCOUNTING_REQUEST) goto done;
+       /*
+        *      Do cleanup delay ONLY for RADIUS packets from a real
+        *      client.  Everything else just gets cleaned up
+        *      immediately.
+        */
+       if (request->packet->dst_port == 0) goto done;
 
-#ifdef WITH_DETAIL
        /*
-        *      If the packets are from the detail file, we can clean them up now.
+        *      Accounting packets shouldn't be retransmitted.  They
+        *      should always be updated with Acct-Delay-Time.
         */
-       if (request->listener->type == RAD_LISTEN_DETAIL) goto done;
+#ifdef WITH_ACCOUNTING
+       if (request->packet->code == PW_CODE_ACCOUNTING_REQUEST) goto done;
+#endif
+
+#ifdef WITH_DHCP
+       if (request->listener->type == RAD_LISTEN_DHCP) goto done;
+#endif
+
+#ifdef WITH_VMPS
+       if (request->listener->type == RAD_LISTEN_VQP) goto done;
 #endif
 
        if (!request->root->cleanup_delay) goto done;
@@ -785,19 +860,23 @@ static void request_cleanup_delay_init(REQUEST *request)
         */
        if (timercmp(&when, &now, >)) {
 #ifdef DEBUG_STATE_MACHINE
-               if (debug_flag) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_cleanup_delay");
+               if (rad_debug_lvl) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_cleanup_delay");
 #endif
                request->process = request_cleanup_delay;
-               request->child_state = REQUEST_CLEANUP_DELAY;
+
+               if (!we_are_master()) {
+                       FINAL_STATE(REQUEST_CLEANUP_DELAY);
+                       return;
+               }
 
                /*
                 *      Update this if we can, otherwise let the timers pick it up.
                 */
-               if (we_are_master()) {
-                       STATE_MACHINE_TIMER(FR_ACTION_TIMER);
-               } else {
-                       NO_CHILD_THREAD;
-               }
+               request->child_state = REQUEST_CLEANUP_DELAY;
+#ifdef HAVE_PTHREAD_H
+               rad_assert(request->child_pid == NO_SUCH_CHILD_PID);
+#endif
+               STATE_MACHINE_TIMER(FR_ACTION_TIMER);
                return;
        }
 
@@ -812,7 +891,7 @@ done:
 /*
  *     Enforce max_request_time.
  */
-static void request_max_time(REQUEST *request)
+static bool request_max_time(REQUEST *request)
 {
        struct timeval now, when;
        rad_assert(request->magic == REQUEST_MAGIC);
@@ -835,7 +914,7 @@ static void request_max_time(REQUEST *request)
        if (request->child_state == REQUEST_DONE) {
        done:
                request_done(request, FR_ACTION_DONE);
-               return;
+               return true;
        }
 
        /*
@@ -878,6 +957,7 @@ static void request_max_time(REQUEST *request)
        tv_add(&when, request->delay);
        request->delay += request->delay >> 1;
        STATE_MACHINE_TIMER(FR_ACTION_TIMER);
+       return false;
 }
 
 static void request_queue_or_run(REQUEST *request,
@@ -897,7 +977,7 @@ static void request_queue_or_run(REQUEST *request,
         */
        if (request->master_state == REQUEST_STOP_PROCESSING) {
 #ifdef DEBUG_STATE_MACHINE
-               if (debug_flag) printf("(%u) ********\tSTATE %s M-%s causes C-%s-> C-%s\t********\n",
+               if (rad_debug_lvl) printf("(%u) ********\tSTATE %s M-%s causes C-%s-> C-%s\t********\n",
                                       request->number, __FUNCTION__,
                                       master_state_names[request->master_state],
                                       child_state_names[request->child_state],
@@ -964,7 +1044,33 @@ static void request_dup(REQUEST *request)
              request->component, request->module);
 }
 
-STATE_MACHINE_DECL(request_cleanup_delay)
+
+/** Sit on a request until it's time to clean it up.
+ *
+ *  A NAS may not see a response from the server.  When the NAS
+ *  retransmits, we want to be able to send a cached reply back.  The
+ *  alternative is to re-process the packet, which does bad things for
+ *  EAP, among others.
+ *
+ *  IF we do see a NAS retransmit, we extend the cleanup delay,
+ *  because the NAS might miss our cached reply.
+ *
+ *  Otherwise, once we reach cleanup_delay, we transition to DONE.
+ *
+ *  \dot
+ *     digraph cleanup_delay {
+ *             cleanup_delay;
+ *             send_reply [ label = "send_reply\nincrease cleanup delay" ];
+ *
+ *             cleanup_delay -> send_reply [ label = "DUP" ];
+ *             send_reply -> cleanup_delay;
+ *             cleanup_delay -> proxy_reply_too_late [ label = "PROXY_REPLY", arrowhead = "none" ];
+ *             cleanup_delay -> cleanup_delay [ label = "TIMER < timeout" ];
+ *             cleanup_delay -> done [ label = "TIMER >= timeout" ];
+ *     }
+ *  \enddot
+ */
+static void request_cleanup_delay(REQUEST *request, int action)
 {
        struct timeval when, now;
 
@@ -987,7 +1093,7 @@ STATE_MACHINE_DECL(request_cleanup_delay)
                 *      Double the cleanup_delay to catch retransmits.
                 */
                when = request->reply->timestamp;
-               request->delay += request->delay ;
+               request->delay += request->delay;
                when.tv_sec += request->delay;
 
                STATE_MACHINE_TIMER(FR_ACTION_TIMER);
@@ -1009,13 +1115,13 @@ STATE_MACHINE_DECL(request_cleanup_delay)
 
                if (timercmp(&when, &now, >)) {
 #ifdef DEBUG_STATE_MACHINE
-                       if (debug_flag) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_cleanup_delay");
+                       if (rad_debug_lvl) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_cleanup_delay");
 #endif
                        STATE_MACHINE_TIMER(FR_ACTION_TIMER);
                        return;
                } /* else it's time to clean up */
 
-               request_done(request, REQUEST_DONE);
+               request_done(request, FR_ACTION_DONE);
                break;
 
        default:
@@ -1024,7 +1130,27 @@ STATE_MACHINE_DECL(request_cleanup_delay)
        }
 }
 
-STATE_MACHINE_DECL(request_response_delay)
+
+/** Sit on a request until it's time to respond to it.
+ *
+ *  For security reasons, rejects (and maybe some other) packets are
+ *  delayed for a while before we respond.  This delay means that
+ *  badly behaved NASes don't hammer the server with authentication
+ *  attempts.
+ *
+ *  Otherwise, once we reach response_delay, we send the reply, and
+ *  transition to cleanup_delay.
+ *
+ *  \dot
+ *     digraph response_delay {
+ *             response_delay -> proxy_reply_too_late [ label = "PROXY_REPLY", arrowhead = "none" ];
+ *             response_delay -> response_delay [ label = "DUP, TIMER < timeout" ];
+ *             response_delay -> send_reply [ label = "TIMER >= timeout" ];
+ *             send_reply -> cleanup_delay;
+ *     }
+ *  \enddot
+ */
+static void request_response_delay(REQUEST *request, int action)
 {
        struct timeval when, now;
 
@@ -1037,7 +1163,7 @@ STATE_MACHINE_DECL(request_response_delay)
 
        switch (action) {
        case FR_ACTION_DUP:
-               ERROR("(%u) Discarding duplicate request from "
+               RDEBUG("(%u) Discarding duplicate request from "
                      "client %s port %d - ID: %u due to delayed response",
                      request->number, request->client->shortname,
                      request->packet->src_port,request->packet->id);
@@ -1052,8 +1178,6 @@ STATE_MACHINE_DECL(request_response_delay)
        case FR_ACTION_TIMER:
                fr_event_now(el, &now);
 
-               rad_assert(request->response_delay.tv_sec > 0);
-
                /*
                 *      See if it's time to send the reply.  If not,
                 *      we wait some more.
@@ -1065,7 +1189,7 @@ STATE_MACHINE_DECL(request_response_delay)
 
                if (timercmp(&when, &now, >)) {
 #ifdef DEBUG_STATE_MACHINE
-                       if (debug_flag) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_response_delay");
+                       if (rad_debug_lvl) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_response_delay");
 #endif
                        STATE_MACHINE_TIMER(FR_ACTION_TIMER);
                        return;
@@ -1088,7 +1212,7 @@ STATE_MACHINE_DECL(request_response_delay)
 }
 
 
-static int CC_HINT(nonnull) request_pre_handler(REQUEST *request, UNUSED int action)
+static int request_pre_handler(REQUEST *request, UNUSED int action)
 {
        int rcode;
 
@@ -1104,8 +1228,8 @@ static int CC_HINT(nonnull) request_pre_handler(REQUEST *request, UNUSED int act
         *      process it.
         */
        if (request->packet->dst_port == 0) {
-               request->username = pairfind(request->packet->vps, PW_USER_NAME, 0, TAG_ANY);
-               request->password = pairfind(request->packet->vps, PW_USER_PASSWORD, 0, TAG_ANY);
+               request->username = fr_pair_find_by_num(request->packet->vps, PW_USER_NAME, 0, TAG_ANY);
+               request->password = fr_pair_find_by_num(request->packet->vps, PW_USER_PASSWORD, 0, TAG_ANY);
                return 1;
        }
 
@@ -1136,13 +1260,19 @@ static int CC_HINT(nonnull) request_pre_handler(REQUEST *request, UNUSED int act
        }
 
        if (!request->username) {
-               request->username = pairfind(request->packet->vps, PW_USER_NAME, 0, TAG_ANY);
+               request->username = fr_pair_find_by_num(request->packet->vps, PW_USER_NAME, 0, TAG_ANY);
        }
 
        return 1;
 }
 
-STATE_MACHINE_DECL(request_finish)
+
+/**  Do the final processing of a request before we reply to the NAS.
+ *
+ *  Various cleanups, suppress responses, copy Proxy-State, and set
+ *  response_delay or cleanup_delay;
+ */
+static void request_finish(REQUEST *request, int action)
 {
        VALUE_PAIR *vp;
 
@@ -1165,7 +1295,7 @@ STATE_MACHINE_DECL(request_finish)
        /*
         *      Override the response code if a control:Response-Packet-Type attribute is present.
         */
-       vp = pairfind(request->config_items, PW_RESPONSE_PACKET_TYPE, 0, TAG_ANY);
+       vp = fr_pair_find_by_num(request->config, PW_RESPONSE_PACKET_TYPE, 0, TAG_ANY);
        if (vp) {
                if (vp->vp_integer == 256) {
                        RDEBUG2("Not responding to request");
@@ -1179,8 +1309,7 @@ STATE_MACHINE_DECL(request_finish)
         */
        else if (request->packet->code == PW_CODE_ACCESS_REQUEST) {
                if (request->reply->code == 0) {
-                       vp = pairfind(request->config_items, PW_AUTH_TYPE, 0, TAG_ANY);
-
+                       vp = fr_pair_find_by_num(request->config, PW_AUTH_TYPE, 0, TAG_ANY);
                        if (!vp || (vp->vp_integer != 5)) {
                                RDEBUG2("There was no response configured: "
                                        "rejecting request");
@@ -1193,9 +1322,9 @@ STATE_MACHINE_DECL(request_finish)
        /*
         *      Copy Proxy-State from the request to the reply.
         */
-       vp = paircopy_by_num(request->reply, request->packet->vps,
+       vp = fr_pair_list_copy_by_num(request->reply, request->packet->vps,
                       PW_PROXY_STATE, 0, TAG_ANY);
-       if (vp) pairadd(&request->reply->vps, vp);
+       if (vp) fr_pair_add(&request->reply->vps, vp);
 
        /*
         *      Call Post-Auth for Access-Request packets.
@@ -1204,6 +1333,14 @@ STATE_MACHINE_DECL(request_finish)
                rad_postauth(request);
        }
 
+#ifdef WITH_COA
+       /*
+        *      Maybe originate a CoA request.
+        */
+       if ((action == FR_ACTION_RUN) && !request->proxy && request->coa) {
+               request_coa_originate(request);
+       }
+#endif
 
        /*
         *      Clean up.  These are no longer needed.
@@ -1217,8 +1354,7 @@ STATE_MACHINE_DECL(request_finish)
         */
        if (request->packet->dst_port == 0) {
                RDEBUG("Finished internally proxied request.");
-               NO_CHILD_THREAD;
-               request->child_state = REQUEST_DONE;
+               FINAL_STATE(REQUEST_DONE);
                return;
        }
 
@@ -1228,13 +1364,15 @@ STATE_MACHINE_DECL(request_finish)
         */
        if (request->listener->type == RAD_LISTEN_DETAIL) {
                request->simul_max = 1;
-               request->listener->send(request->listener, request);
+
                /*
                 *      But only print the reply if there is one.
                 */
                if (request->reply->code != 0) {
                        debug_packet(request, request->reply, false);
                }
+
+               request->listener->send(request->listener, request);
                goto done;
        }
 #endif
@@ -1279,6 +1417,27 @@ STATE_MACHINE_DECL(request_finish)
            (request->root->reject_delay.tv_sec > 0)) {
                request->response_delay = request->root->reject_delay;
 
+               vp = fr_pair_find_by_num(request->reply->vps, PW_FREERADIUS_RESPONSE_DELAY, 0, TAG_ANY);
+               if (vp) {
+                       if (vp->vp_integer <= 10) {
+                               request->response_delay.tv_sec = vp->vp_integer;
+                       } else {
+                               request->response_delay.tv_sec = 10;
+                       }
+                       request->response_delay.tv_usec = 0;
+               } else {
+                       vp = fr_pair_find_by_num(request->reply->vps, PW_FREERADIUS_RESPONSE_DELAY_USEC, 0, TAG_ANY);
+                       if (vp) {
+                               if (vp->vp_integer <= 10 * USEC) {
+                                       request->response_delay.tv_sec = vp->vp_integer / USEC;
+                                       request->response_delay.tv_usec = vp->vp_integer % USEC;
+                               } else {
+                                       request->response_delay.tv_sec = 10;
+                                       request->response_delay.tv_usec = 0;
+                               }
+                       }
+               }
+
 #ifdef WITH_PROXY
                /*
                 *      If we timed out a proxy packet, don't delay
@@ -1294,62 +1453,26 @@ STATE_MACHINE_DECL(request_finish)
        /*
         *      Send the reply.
         */
-       if (request->response_delay.tv_sec == 0) {
-               rad_assert(request->response_delay.tv_usec == 0);
+       if ((request->response_delay.tv_sec == 0) &&
+           (request->response_delay.tv_usec == 0)) {
 
                /*
                 *      Don't print a reply if there's none to send.
                 */
                if (request->reply->code != 0) {
-                       request->listener->send(request->listener, request);
+                       if (rad_debug_lvl && request->state &&
+                           (request->reply->code == PW_CODE_ACCESS_ACCEPT)) {
+                               if (!fr_pair_find_by_num(request->packet->vps, PW_STATE, 0, TAG_ANY)) {
+                                       RWDEBUG2("Unused attributes found in &session-state:");
+                               }
+                       }
+
                        debug_packet(request, request->reply, false);
+                       request->listener->send(request->listener, request);
                }
 
        done:
                RDEBUG2("Finished request");
-               request->component = "<core>";
-               request->module = "<done>";
-
-#ifdef WITH_ACCOUNTING
-               /*
-                *      Accounting packets can be cleaned up now.
-                */
-               if (request->packet->code == PW_CODE_ACCOUNTING_REQUEST) {
-                       NO_CHILD_THREAD;
-                       request->child_state = REQUEST_DONE;
-                       return;
-               }
-#endif
-
-#ifdef WITH_DETAIL
-               /*
-                *      If the packets are from the detail file, we can clean them up now.
-                */
-               if (request->listener->type == RAD_LISTEN_DETAIL) {
-                       NO_CHILD_THREAD;
-                       request->child_state = REQUEST_DONE;
-                       return;
-               }
-#endif
-
-#ifdef WITH_COA
-               /*
-                *      If we've originated this CoA request, it gets
-                *      cleaned up now.
-                */
-               if (request->proxy &&
-                   ((request->proxy->code == PW_CODE_COA_REQUEST) ||
-                    (request->proxy->code == PW_CODE_DISCONNECT_REQUEST)) &&
-                   (request->packet->code != request->proxy->code)) {
-                       NO_CHILD_THREAD;
-                       request->child_state = REQUEST_DONE;
-                       return;
-               }
-#endif
-
-               /*
-                *      Clean up the request.
-                */
                request_cleanup_delay_init(request);
 
        } else {
@@ -1361,15 +1484,26 @@ STATE_MACHINE_DECL(request_finish)
                RDEBUG2("Delaying response for %d.%06d seconds",
                        (int) request->response_delay.tv_sec, (int) request->response_delay.tv_usec);
                request->listener->encode(request->listener, request);
-               request->component = "<core>";
-               request->module = "<delay>";
                request->process = request_response_delay;
-               NO_CHILD_THREAD;
-               request->child_state = REQUEST_RESPONSE_DELAY;
+
+               FINAL_STATE(REQUEST_RESPONSE_DELAY);
        }
 }
 
-STATE_MACHINE_DECL(request_running)
+/** Process a request from a client.
+ *
+ *  The outcome might be that the request is proxied.
+ *
+ *  \dot
+ *     digraph running {
+ *             running -> running [ label = "TIMER < max_request_time" ];
+ *             running -> done [ label = "TIMER >= max_request_time" ];
+ *             running -> proxy [ label = "proxied" ];
+ *             running -> dup [ label = "DUP", arrowhead = "none" ];
+ *     }
+ *  \enddot
+ */
+static void request_running(REQUEST *request, int action)
 {
        VERIFY_REQUEST(request);
 
@@ -1379,7 +1513,7 @@ STATE_MACHINE_DECL(request_running)
        switch (action) {
        case FR_ACTION_TIMER:
                COA_SEPARATE;
-               request_max_time(request);
+               (void) request_max_time(request);
                break;
 
        case FR_ACTION_DUP:
@@ -1389,14 +1523,12 @@ STATE_MACHINE_DECL(request_running)
        case FR_ACTION_RUN:
                if (!request_pre_handler(request, action)) {
 #ifdef DEBUG_STATE_MACHINE
-                       if (debug_flag) printf("(%u) ********\tSTATE %s failed in pre-handler C-%s -> C-%s\t********\n",
+                       if (rad_debug_lvl) printf("(%u) ********\tSTATE %s failed in pre-handler C-%s -> C-%s\t********\n",
                                               request->number, __FUNCTION__,
                                               child_state_names[request->child_state],
                                               child_state_names[REQUEST_DONE]);
 #endif
-
-                       NO_CHILD_THREAD;
-                       request->child_state = REQUEST_DONE;
+                       FINAL_STATE(REQUEST_DONE);
                        break;
                }
 
@@ -1410,7 +1542,7 @@ STATE_MACHINE_DECL(request_running)
                if ((action == FR_ACTION_RUN) &&
                    request_will_proxy(request)) {
 #ifdef DEBUG_STATE_MACHINE
-                       if (debug_flag) printf("(%u) ********\tWill Proxy\t********\n", request->number);
+                       if (rad_debug_lvl) printf("(%u) ********\tWill Proxy\t********\n", request->number);
 #endif
                        /*
                         *      If this fails, it
@@ -1418,21 +1550,12 @@ STATE_MACHINE_DECL(request_running)
                         *      up the post proxy fail
                         *      handler.
                         */
-                       if (request_proxy(request, 0) < 0) goto req_finished;
+                       if (request_proxy(request) < 0) goto req_finished;
                } else
 #endif
                {
 #ifdef DEBUG_STATE_MACHINE
-                       if (debug_flag) printf("(%u) ********\tFinished\t********\n", request->number);
-#endif
-
-#ifdef WITH_COA
-                       /*
-                        *      Maybe originate a CoA request.
-                        */
-                       if ((action == FR_ACTION_RUN) && request->coa) {
-                               request_coa_originate(request);
-                       }
+                       if (rad_debug_lvl) printf("(%u) ********\tFinished\t********\n", request->number);
 #endif
 
 #ifdef WITH_PROXY
@@ -1469,10 +1592,15 @@ int request_receive(TALLOC_CTX *ctx, rad_listen_t *listener, RADIUS_PACKET *pack
 #ifdef WITH_ACCOUNTING
        if (listener->type != RAD_LISTEN_DETAIL)
 #endif
+
+#ifdef WITH_TCP
        {
                sock = listener->data;
                sock->last_packet = now.tv_sec;
+
+               packet->proto = sock->proto;
        }
+#endif
 
        /*
         *      Skip everything if required.
@@ -1597,6 +1725,7 @@ skip_dup:
        if (!ctx) {
                ctx = talloc_pool(NULL, main_config.talloc_pool_size);
                if (!ctx) return 0;
+               talloc_set_name_const(ctx, "request_receive_pool");
 
                /*
                 *      The packet is still allocated from a different
@@ -1638,8 +1767,8 @@ skip_dup:
 #endif
 
                request->listener->decode(request->listener, request);
-               request->username = pairfind(request->packet->vps, PW_USER_NAME, 0, TAG_ANY);
-               request->password = pairfind(request->packet->vps, PW_USER_PASSWORD, 0, TAG_ANY);
+               request->username = fr_pair_find_by_num(request->packet->vps, PW_USER_NAME, 0, TAG_ANY);
+               request->password = fr_pair_find_by_num(request->packet->vps, PW_USER_PASSWORD, 0, TAG_ANY);
 
                fun(request);
 
@@ -1679,7 +1808,7 @@ static REQUEST *request_setup(TALLOC_CTX *ctx, rad_listen_t *listener, RADIUS_PA
                ERROR("No memory");
                return NULL;
        }
-       request->reply = rad_alloc(request, false);
+       request->reply = rad_alloc_reply(request, packet);
        if (!request->reply) {
                ERROR("No memory");
                talloc_free(request);
@@ -1694,14 +1823,11 @@ static REQUEST *request_setup(TALLOC_CTX *ctx, rad_listen_t *listener, RADIUS_PA
        request->master_state = REQUEST_ACTIVE;
        request->child_state = REQUEST_RUNNING;
 #ifdef DEBUG_STATE_MACHINE
-       if (debug_flag) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n",
+       if (rad_debug_lvl) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n",
                               request->number, __FUNCTION__,
                               child_state_names[request->child_state],
                               child_state_names[REQUEST_RUNNING]);
 #endif
-#ifdef HAVE_PTHREAD_H
-       request->child_pid = NO_SUCH_CHILD_PID;
-#endif
        request->handle = fun;
        NO_CHILD_THREAD;
 
@@ -1786,10 +1912,10 @@ static void tcp_socket_timer(void *ctx)
 
        ASSERT_MASTER;
 
-       fr_event_now(el, &now);
-
        if (listener->status != RAD_LISTEN_STATUS_KNOWN) return;
 
+       fr_event_now(el, &now);
+
        switch (listener->type) {
 #ifdef WITH_PROXY
        case RAD_LISTEN_PROXY:
@@ -1821,7 +1947,28 @@ static void tcp_socket_timer(void *ctx)
 
                do_close:
 
-                       listener->status = RAD_LISTEN_STATUS_EOL;
+#ifdef WITH_PROXY
+                       /*
+                        *      Proxy sockets get frozen, so that we don't use
+                        *      them for new requests.  But we do keep them
+                        *      open to listen for replies to requests we had
+                        *      previously sent.
+                        */
+                       if (listener->type == RAD_LISTEN_PROXY) {
+                               PTHREAD_MUTEX_LOCK(&proxy_mutex);
+                               if (!fr_packet_list_socket_freeze(proxy_list,
+                                                                 listener->fd)) {
+                                       ERROR("Fatal error freezing socket: %s", fr_strerror());
+                                       fr_exit(1);
+                               }
+                               PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
+                       }
+#endif
+
+                       /*
+                        *      Mark the socket as "don't use if at all possible".
+                        */
+                       listener->status = RAD_LISTEN_STATUS_FROZEN;
                        event_new_fd(listener);
                        return;
                }
@@ -1869,26 +2016,6 @@ static void tcp_socket_timer(void *ctx)
 
 #ifdef WITH_PROXY
 /*
- *     Add +/- 2s of jitter, as suggested in RFC 3539
- *     and in RFC 5080.
- */
-static void add_jitter(struct timeval *when)
-{
-       uint32_t jitter;
-
-       when->tv_sec -= 2;
-
-       jitter = fr_rand();
-       jitter ^= (jitter >> 10);
-       jitter &= ((1 << 22) - 1); /* 22 bits of 1 */
-
-       /*
-        *      Add in ~ (4 * USEC) of jitter.
-        */
-       tv_add(when, jitter);
-}
-
-/*
  *     Called by socket_del to remove requests with this socket
  */
 static int eol_proxy_listener(void *ctx, void *data)
@@ -1996,8 +2123,9 @@ static void remove_from_proxy_hash_nl(REQUEST *request, bool yank)
        }
 
 #ifdef WITH_TCP
-       rad_assert(request->proxy_listener != NULL);
-       request->proxy_listener->count--;
+       if (request->proxy_listener) {
+               request->proxy_listener->count--;
+       }
 #endif
        request->proxy_listener = NULL;
 
@@ -2006,7 +2134,6 @@ static void remove_from_proxy_hash_nl(REQUEST *request, bool yank)
         *      the mutex.  This guarantees that when another thread
         *      grabs the mutex, the "not in hash" flag is correct.
         */
-       RDEBUG3("proxy: request is no longer in proxy hash");
 }
 
 static void remove_from_proxy_hash(REQUEST *request)
@@ -2166,13 +2293,13 @@ static int process_proxy_reply(REQUEST *request, RADIUS_PACKET *reply)
         *      Delete any reply we had accumulated until now.
         */
        RDEBUG2("Clearing existing &reply: attributes");
-       pairfree(&request->reply->vps);
+       fr_pair_list_free(&request->reply->vps);
 
        /*
         *      Run the packet through the post-proxy stage,
         *      BEFORE playing games with the attributes.
         */
-       vp = pairfind(request->config_items, PW_POST_PROXY_TYPE, 0, TAG_ANY);
+       vp = fr_pair_find_by_num(request->config, PW_POST_PROXY_TYPE, 0, TAG_ANY);
        if (vp) {
                post_proxy_type = vp->vp_integer;
        /*
@@ -2209,7 +2336,7 @@ static int process_proxy_reply(REQUEST *request, RADIUS_PACKET *reply)
                 *      Create config:Post-Proxy-Type
                 */
                if (dval) {
-                       vp = radius_paircreate(request, &request->config_items, PW_POST_PROXY_TYPE, 0);
+                       vp = radius_pair_create(request, &request->config, PW_POST_PROXY_TYPE, 0);
                        vp->vp_integer = dval->value;
                }
        }
@@ -2259,7 +2386,7 @@ static int process_proxy_reply(REQUEST *request, RADIUS_PACKET *reply)
        }
 
 #ifdef WITH_COA
-       if (request->packet->code == request->proxy->code)
+       if (request->packet->code == request->proxy->code) {
          /*
           *    Don't run the next bit if we originated a CoA
           *    packet, after receiving an Access-Request or
@@ -2267,21 +2394,30 @@ static int process_proxy_reply(REQUEST *request, RADIUS_PACKET *reply)
           */
 #endif
 
-       /*
-        *      There may NOT be a proxy reply, as we may be
-        *      running Post-Proxy-Type = Fail.
-        */
-       if (reply) {
-               pairadd(&request->reply->vps, paircopy(request->reply, reply->vps));
-
                /*
-                *      Delete the Proxy-State Attributes from
-                *      the reply.  These include Proxy-State
-                *      attributes from us and remote server.
+                *      There may NOT be a proxy reply, as we may be
+                *      running Post-Proxy-Type = Fail.
                 */
-               pairdelete(&request->reply->vps, PW_PROXY_STATE, 0, TAG_ANY);
-       }
+               if (reply) {
+                       fr_pair_add(&request->reply->vps, fr_pair_list_copy(request->reply, reply->vps));
+
+                       /*
+                        *      Delete the Proxy-State Attributes from
+                        *      the reply.  These include Proxy-State
+                        *      attributes from us and remote server.
+                        */
+                       fr_pair_delete_by_num(&request->reply->vps, PW_PROXY_STATE, 0, TAG_ANY);
 
+               } else {
+                       vp = fr_pair_find_by_num(request->config, PW_RESPONSE_PACKET_TYPE, 0, TAG_ANY);
+                       if (vp && (vp->vp_integer != 256)) {
+                               request->proxy_reply = rad_alloc_reply(request, request->proxy);
+                               request->proxy_reply->code = vp->vp_integer;
+                       }
+               }
+#ifdef WITH_COA
+       }
+#endif
        switch (rcode) {
        default:  /* Don't do anything */
                break;
@@ -2295,6 +2431,28 @@ static int process_proxy_reply(REQUEST *request, RADIUS_PACKET *reply)
        return 1;
 }
 
+static void mark_home_server_alive(REQUEST *request, home_server_t *home)
+{
+       char buffer[128];
+
+       home->state = HOME_STATE_ALIVE;
+       home->response_timeouts = 0;
+       exec_trigger(request, home->cs, "home_server.alive", false);
+       home->currently_outstanding = 0;
+       home->num_sent_pings = 0;
+       home->num_received_pings = 0;
+       gettimeofday(&home->revive_time, NULL);
+
+       fr_event_delete(el, &home->ev);
+
+       RPROXY("Marking home server %s port %d alive",
+              inet_ntop(request->proxy->dst_ipaddr.af,
+                        &request->proxy->dst_ipaddr.ipaddr,
+                        buffer, sizeof(buffer)),
+              request->proxy->dst_port);
+}
+
+
 int request_proxy_reply(RADIUS_PACKET *packet)
 {
        RADIUS_PACKET **proxy_p;
@@ -2309,7 +2467,8 @@ int request_proxy_reply(RADIUS_PACKET *packet)
 
        if (!proxy_p) {
                PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
-               PROXY("No outstanding request was found for reply from host %s port %d - ID %u",
+               PROXY("No outstanding request was found for %s packet from host %s port %d - ID %u",
+                      fr_packet_codes[packet->code],
                       inet_ntop(packet->src_ipaddr.af,
                                 &packet->src_ipaddr.ipaddr,
                                 buffer, sizeof(buffer)),
@@ -2318,7 +2477,6 @@ int request_proxy_reply(RADIUS_PACKET *packet)
        }
 
        request = fr_packet2myptr(REQUEST, proxy, proxy_p);
-       request->num_proxied_responses++; /* needs to be protected by lock */
 
        PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
 
@@ -2353,12 +2511,16 @@ int request_proxy_reply(RADIUS_PACKET *packet)
         *      Status-Server packets don't count as real packets.
         */
        if (request->proxy->code != PW_CODE_STATUS_SERVER) {
+#ifdef WITH_TCP
                listen_socket_t *sock = request->proxy_listener->data;
 
-               request->home_server->last_packet_recv = now.tv_sec;
                sock->last_packet = now.tv_sec;
+#endif
+               request->home_server->last_packet_recv = now.tv_sec;
        }
 
+       request->num_proxied_responses++;
+
        /*
         *      If we have previously seen a reply, ignore the
         *      duplicate.
@@ -2409,6 +2571,8 @@ int request_proxy_reply(RADIUS_PACKET *packet)
 
 #ifdef WITH_ACCOUNTING
        case PW_CODE_ACCOUNTING_REQUEST:
+               proxy_acct_stats.last_packet = packet->timestamp.tv_sec;
+
                request->proxy_listener->stats.total_responses++;
                proxy_acct_stats.last_packet = packet->timestamp.tv_sec;
                break;
@@ -2433,12 +2597,13 @@ int request_proxy_reply(RADIUS_PACKET *packet)
 #endif
 
        /*
-        *      We've received a reply.  If we hadn't been sending it
-        *      packets for a while, just mark it alive.
+        *      If we hadn't been sending the home server packets for
+        *      a while, just mark it alive.  Or, if it was zombie,
+        *      it's now responded, and is therefore alive.
         */
-       if (request->home_server->state == HOME_STATE_UNKNOWN) {
-               request->home_server->state = HOME_STATE_ALIVE;
-               request->home_server->response_timeouts = 0;
+       if ((request->home_server->state == HOME_STATE_UNKNOWN) ||
+           (request->home_server->state == HOME_STATE_ZOMBIE)) {
+               mark_home_server_alive(request, request->home_server);
        }
 
        /*
@@ -2462,10 +2627,12 @@ static int setup_post_proxy_fail(REQUEST *request)
        if (request->proxy->code == PW_CODE_ACCESS_REQUEST) {
                dval = dict_valbyname(PW_POST_PROXY_TYPE, 0,
                                      "Fail-Authentication");
-
+#ifdef WITH_ACCOUNTING
        } else if (request->proxy->code == PW_CODE_ACCOUNTING_REQUEST) {
                dval = dict_valbyname(PW_POST_PROXY_TYPE, 0,
                                      "Fail-Accounting");
+#endif
+
 #ifdef WITH_COA
        } else if (request->proxy->code == PW_CODE_COA_REQUEST) {
                dval = dict_valbyname(PW_POST_PROXY_TYPE, 0, "Fail-CoA");
@@ -2481,19 +2648,36 @@ static int setup_post_proxy_fail(REQUEST *request)
        if (!dval) dval = dict_valbyname(PW_POST_PROXY_TYPE, 0, "Fail");
 
        if (!dval) {
-               pairdelete(&request->config_items, PW_POST_PROXY_TYPE, 0, TAG_ANY);
+               fr_pair_delete_by_num(&request->config, PW_POST_PROXY_TYPE, 0, TAG_ANY);
                return 0;
        }
 
-       vp = pairfind(request->config_items, PW_POST_PROXY_TYPE, 0, TAG_ANY);
-       if (!vp) vp = radius_paircreate(request, &request->config_items,
+       vp = fr_pair_find_by_num(request->config, PW_POST_PROXY_TYPE, 0, TAG_ANY);
+       if (!vp) vp = radius_pair_create(request, &request->config,
                                        PW_POST_PROXY_TYPE, 0);
        vp->vp_integer = dval->value;
 
        return 1;
 }
 
-STATE_MACHINE_DECL(proxy_no_reply)
+
+/** Process a request after the proxy has timed out.
+ *
+ *  Run the packet through Post-Proxy-Type Fail
+ *
+ *  \dot
+ *     digraph proxy_no_reply {
+ *             proxy_no_reply;
+ *
+ *             proxy_no_reply -> dup [ label = "DUP", arrowhead = "none" ];
+ *             proxy_no_reply -> timer [ label = "TIMER < max_request_time" ];
+ *             proxy_no_reply -> proxy_reply_too_late [ label = "PROXY_REPLY" arrowhead = "none"];
+ *             proxy_no_reply -> process_proxy_reply [ label = "RUN" ];
+ *             proxy_no_reply -> done [ label = "TIMER >= timeout" ];
+ *     }
+ *  \enddot
+ */
+static void proxy_no_reply(REQUEST *request, int action)
 {
        VERIFY_REQUEST(request);
 
@@ -2506,7 +2690,7 @@ STATE_MACHINE_DECL(proxy_no_reply)
                break;
 
        case FR_ACTION_TIMER:
-               request_max_time(request);
+               (void) request_max_time(request);
                break;
 
        case FR_ACTION_PROXY_REPLY:
@@ -2526,7 +2710,23 @@ STATE_MACHINE_DECL(proxy_no_reply)
        }
 }
 
-STATE_MACHINE_DECL(proxy_running)
+/** Process the request after receiving a proxy reply.
+ *
+ *  Throught the post-proxy section, and the through the handler
+ *  function.
+ *
+ *  \dot
+ *     digraph proxy_running {
+ *             proxy_running;
+ *
+ *             proxy_running -> dup [ label = "DUP", arrowhead = "none" ];
+ *             proxy_running -> timer [ label = "TIMER < max_request_time" ];
+ *             proxy_running -> process_proxy_reply [ label = "RUN" ];
+ *             proxy_running -> done [ label = "TIMER >= timeout" ];
+ *     }
+ *  \enddot
+ */
+static void proxy_running(REQUEST *request, int action)
 {
        VERIFY_REQUEST(request);
 
@@ -2539,7 +2739,7 @@ STATE_MACHINE_DECL(proxy_running)
                break;
 
        case FR_ACTION_TIMER:
-               request_max_time(request);
+               (void) request_max_time(request);
                break;
 
        case FR_ACTION_RUN:
@@ -2555,6 +2755,34 @@ STATE_MACHINE_DECL(proxy_running)
        }
 }
 
+/** Determine if a #REQUEST needs to be proxied, and perform pre-proxy operations
+ *
+ * Whether a request will be proxied is determined by the attributes present
+ * in request->config. If any of the following attributes are found, the
+ * request may be proxied.
+ *
+ * The key attributes are:
+ *   - PW_PROXY_TO_REALM          - Specifies a realm the request should be proxied to.
+ *   - PW_HOME_SERVER_POOL        - Specifies a specific home server pool to proxy to.
+ *   - PW_PACKET_DST_IP_ADDRESS   - Specifies a specific IPv4 home server to proxy to.
+ *   - PW_PACKET_DST_IPV6_ADDRESS - Specifies a specific IPv6 home server to proxy to.
+ *
+ * Certain packet types such as #PW_CODE_STATUS_SERVER will never be proxied.
+ *
+ * If request should be proxied, will:
+ *   - Add request:Proxy-State
+ *   - Strip the current username value of its realm (depending on config)
+ *   - Create a CHAP-Challenge from the original request vector, if one doesn't already
+ *     exist.
+ *   - Call the pre-process section in the current server, or in the virtual server
+ *     associated with the home server pool we're proxying to.
+ *
+ * @todo A lot of this logic is RADIUS specific, and should be moved out into a protocol
+ *     specific function.
+ *
+ * @param request The #REQUEST to evaluate for proxying.
+ * @return 0 if not proxying, 1 if request should be proxied, -1 on error.
+ */
 static int request_will_proxy(REQUEST *request)
 {
        int rcode, pre_proxy_type = 0;
@@ -2576,7 +2804,7 @@ static int request_will_proxy(REQUEST *request)
         */
        if (request->reply->code != 0) return 0;
 
-       vp = pairfind(request->config_items, PW_PROXY_TO_REALM, 0, TAG_ANY);
+       vp = fr_pair_find_by_num(request->config, PW_PROXY_TO_REALM, 0, TAG_ANY);
        if (vp) {
                realm = realm_find2(vp->vp_strvalue);
                if (!realm) {
@@ -2608,7 +2836,7 @@ static int request_will_proxy(REQUEST *request)
                        return 0;
                }
 
-       } else if ((vp = pairfind(request->config_items, PW_HOME_SERVER_POOL, 0, TAG_ANY)) != NULL) {
+       } else if ((vp = fr_pair_find_by_num(request->config, PW_HOME_SERVER_POOL, 0, TAG_ANY)) != NULL) {
                int pool_type;
 
                switch (request->packet->code) {
@@ -2638,9 +2866,8 @@ static int request_will_proxy(REQUEST *request)
                /*
                 *      Send it directly to a home server (i.e. NAS)
                 */
-       } else if (((vp = pairfind(request->config_items, PW_PACKET_DST_IP_ADDRESS, 0, TAG_ANY)) != NULL) ||
-                  ((vp = pairfind(request->config_items, PW_PACKET_DST_IPV6_ADDRESS, 0, TAG_ANY)) != NULL)) {
-               VALUE_PAIR *port;
+       } else if (((vp = fr_pair_find_by_num(request->config, PW_PACKET_DST_IP_ADDRESS, 0, TAG_ANY)) != NULL) ||
+                  ((vp = fr_pair_find_by_num(request->config, PW_PACKET_DST_IPV6_ADDRESS, 0, TAG_ANY)) != NULL)) {
                uint16_t dst_port;
                fr_ipaddr_t dst_ipaddr;
 
@@ -2656,27 +2883,55 @@ static int request_will_proxy(REQUEST *request)
                        dst_ipaddr.prefix = 128;
                }
 
-               port = pairfind(request->config_items, PW_PACKET_DST_PORT, 0, TAG_ANY);
-               if (!port) {
-               dst_port = PW_COA_UDP_PORT;
-               } else {
-                       dst_port = vp->vp_integer;
-               }
+               vp = fr_pair_find_by_num(request->config, PW_PACKET_DST_PORT, 0, TAG_ANY);
+               if (!vp) {
+                       if (request->packet->code == PW_CODE_ACCESS_REQUEST) {
+                               dst_port = PW_AUTH_UDP_PORT;
 
-               /*
-                *      Nothing does CoA over TCP.
-                */
+#ifdef WITH_ACCOUNTING
+                       } else if (request->packet->code == PW_CODE_ACCOUNTING_REQUEST) {
+                               dst_port = PW_ACCT_UDP_PORT;
+#endif
+
+#ifdef WITH_COA
+                       } else if ((request->packet->code == PW_CODE_COA_REQUEST) ||
+                                  (request->packet->code == PW_CODE_DISCONNECT_REQUEST)) {
+                               dst_port = PW_COA_UDP_PORT;
+#endif
+                       } else { /* shouldn't happen for RADIUS... */
+                               return 0;
+                       }
+
+               } else {
+                       dst_port = vp->vp_integer;
+               }
+
+               /*
+                *      Nothing does CoA over TCP.
+                */
                home = home_server_find(&dst_ipaddr, dst_port, IPPROTO_UDP);
                if (!home) {
                        char buffer[256];
 
-                       WARN("No such CoA home server %s port %u",
+                       WARN("No such home server %s port %u",
                             inet_ntop(dst_ipaddr.af, &dst_ipaddr.ipaddr, buffer, sizeof(buffer)),
                             (unsigned int) dst_port);
                        return 0;
                }
 
-               goto do_home;
+               /*
+                *      The home server is alive (or may be alive).
+                *      Send the packet to the IP.
+                */
+               if (home->state != HOME_STATE_IS_DEAD) goto do_home;
+
+               /*
+                *      The home server is dead.  If you wanted
+                *      fail-over, you should have proxied to a pool.
+                *      Sucks to be you.
+                */
+
+               return 0;
 
        } else {
                return 0;
@@ -2715,7 +2970,7 @@ do_home:
        /*
         *      Remember that we sent the request to a Realm.
         */
-       if (realmname) pairmake_packet("Realm", realmname, T_OP_EQ);
+       if (realmname) pair_make_request("Realm", realmname, T_OP_EQ);
 
        /*
         *      Strip the name, if told to.
@@ -2724,7 +2979,7 @@ do_home:
         *      requests.
         */
        if (realm && (realm->strip_realm == true) &&
-          (strippedname = pairfind(request->proxy->vps, PW_STRIPPED_USER_NAME, 0, TAG_ANY)) != NULL) {
+          (strippedname = fr_pair_find_by_num(request->proxy->vps, PW_STRIPPED_USER_NAME, 0, TAG_ANY)) != NULL) {
                /*
                 *      If there's a Stripped-User-Name attribute in
                 *      the request, then use THAT as the User-Name
@@ -2738,10 +2993,10 @@ do_home:
                 *      from the vps list, and making the new
                 *      User-Name the head of the vps list.
                 */
-               vp = pairfind(request->proxy->vps, PW_USER_NAME, 0, TAG_ANY);
+               vp = fr_pair_find_by_num(request->proxy->vps, PW_USER_NAME, 0, TAG_ANY);
                if (!vp) {
                        vp_cursor_t cursor;
-                       vp = radius_paircreate(NULL, NULL,
+                       vp = radius_pair_create(NULL, NULL,
                                               PW_USER_NAME, 0);
                        rad_assert(vp != NULL); /* handled by above function */
                        /* Insert at the START of the list */
@@ -2750,7 +3005,7 @@ do_home:
                        fr_cursor_merge(&cursor, request->proxy->vps);
                        request->proxy->vps = vp;
                }
-               pairstrcpy(vp, strippedname->vp_strvalue);
+               fr_pair_value_strcpy(vp, strippedname->vp_strvalue);
 
                /*
                 *      Do NOT delete Stripped-User-Name.
@@ -2764,18 +3019,18 @@ do_home:
         *      anymore - we changed it.
         */
        if ((request->packet->code == PW_CODE_ACCESS_REQUEST) &&
-           pairfind(request->proxy->vps, PW_CHAP_PASSWORD, 0, TAG_ANY) &&
-           pairfind(request->proxy->vps, PW_CHAP_CHALLENGE, 0, TAG_ANY) == NULL) {
-               vp = radius_paircreate(request->proxy, &request->proxy->vps, PW_CHAP_CHALLENGE, 0);
-               pairmemcpy(vp, request->packet->vector, sizeof(request->packet->vector));
+           fr_pair_find_by_num(request->proxy->vps, PW_CHAP_PASSWORD, 0, TAG_ANY) &&
+           fr_pair_find_by_num(request->proxy->vps, PW_CHAP_CHALLENGE, 0, TAG_ANY) == NULL) {
+               vp = radius_pair_create(request->proxy, &request->proxy->vps, PW_CHAP_CHALLENGE, 0);
+               fr_pair_value_memcpy(vp, request->packet->vector, sizeof(request->packet->vector));
        }
 
        /*
         *      The RFC's say we have to do this, but FreeRADIUS
         *      doesn't need it.
         */
-       vp = radius_paircreate(request->proxy, &request->proxy->vps, PW_PROXY_STATE, 0);
-       pairsprintf(vp, "%u", request->packet->id);
+       vp = radius_pair_create(request->proxy, &request->proxy->vps, PW_PROXY_STATE, 0);
+       fr_pair_value_sprintf(vp, "%u", request->packet->id);
 
        /*
         *      Should be done BEFORE inserting into proxy hash, as
@@ -2786,7 +3041,7 @@ do_home:
        /*
         *      Call the pre-proxy routines.
         */
-       vp = pairfind(request->config_items, PW_PRE_PROXY_TYPE, 0, TAG_ANY);
+       vp = fr_pair_find_by_num(request->config, PW_PRE_PROXY_TYPE, 0, TAG_ANY);
        if (vp) {
                DICT_VALUE const *dval = dict_valbyattr(vp->da->attr, vp->da->vendor, vp->vp_integer);
                /* Must be a validation issue */
@@ -2812,8 +3067,17 @@ do_home:
 
                request->server = old_server;
        } else {
+               char buffer[128];
+
+               RDEBUG2("Starting proxy to home server %s port %d",
+                       inet_ntop(request->proxy->dst_ipaddr.af,
+                                 &request->proxy->dst_ipaddr.ipaddr,
+                                 buffer, sizeof(buffer)),
+                       request->proxy->dst_port);
+
                rcode = process_pre_proxy(pre_proxy_type, request);
        }
+
        switch (rcode) {
        case RLM_MODULE_FAIL:
        case RLM_MODULE_INVALID:
@@ -2833,13 +3097,72 @@ do_home:
        case RLM_MODULE_NOOP:
        case RLM_MODULE_OK:
        case RLM_MODULE_UPDATED:
-               break;
+               return 1;
        }
+}
 
-       return 1;
+static int proxy_to_virtual_server(REQUEST *request)
+{
+       REQUEST *fake;
+
+       if (request->packet->dst_port == 0) {
+               WARN("Cannot proxy an internal request");
+               return 0;
+       }
+
+       DEBUG("Proxying to virtual server %s",
+             request->home_server->server);
+
+       /*
+        *      Packets to virtual servers don't get
+        *      retransmissions sent to them.  And the virtual
+        *      server is run ONLY if we have no child
+        *      threads, or we're running in a child thread.
+        */
+       rad_assert(!spawn_flag || !we_are_master());
+
+       fake = request_alloc_fake(request);
+
+       fake->packet->vps = fr_pair_list_copy(fake->packet, request->packet->vps);
+       talloc_free(request->proxy);
+
+       fake->server = request->home_server->server;
+       fake->handle = request->handle;
+       fake->process = NULL; /* should never be run for anything */
+
+       /*
+        *      Run the virtual server.
+        */
+       request_running(fake, FR_ACTION_RUN);
+
+       request->proxy = talloc_steal(request, fake->packet);
+       fake->packet = NULL;
+       request->proxy_reply = talloc_steal(request, fake->reply);
+       fake->reply = NULL;
+
+       talloc_free(fake);
+
+       /*
+        *      No reply code, toss the reply we have,
+        *      and do post-proxy-type Fail.
+        */
+       if (!request->proxy_reply->code) {
+               TALLOC_FREE(request->proxy_reply);
+               setup_post_proxy_fail(request);
+       }
+
+       /*
+        *      Do the proxy reply (if any)
+        */
+       if (process_proxy_reply(request, request->proxy_reply)) {
+               request->handle(request);
+       }
+
+       return -1;      /* so we call request_finish */
 }
 
-static int request_proxy(REQUEST *request, int retransmit)
+
+static int request_proxy(REQUEST *request)
 {
        char buffer[128];
 
@@ -2868,83 +3191,19 @@ static int request_proxy(REQUEST *request, int retransmit)
         *
         *      So, we have some horrible hacks to get around that.
         */
-       if (request->home_server->server) {
-               REQUEST *fake;
-
-               if (request->packet->dst_port == 0) {
-                       WARN("Cannot proxy an internal request");
-                       return 0;
-               }
-
-               DEBUG("Proxying to virtual server %s",
-                     request->home_server->server);
-
-               /*
-                *      Packets to virtual serrers don't get
-                *      retransmissions sent to them.  And the virtual
-                *      server is run ONLY if we have no child
-                *      threads, or we're running in a child thread.
-                */
-               rad_assert(retransmit == 0);
-               rad_assert(!spawn_flag || !we_are_master());
-
-               fake = request_alloc_fake(request);
-
-               fake->packet->vps = paircopy(fake->packet, request->packet->vps);
-               talloc_free(request->proxy);
-
-               fake->server = request->home_server->server;
-               fake->handle = request->handle;
-               fake->process = NULL; /* should never be run for anything */
-
-               /*
-                *      Run the virtual server.
-                */
-               request_running(fake, FR_ACTION_RUN);
-
-               request->proxy = talloc_steal(request, fake->packet);
-               fake->packet = NULL;
-               request->proxy_reply = talloc_steal(request, fake->reply);
-               fake->reply = NULL;
-
-               talloc_free(fake);
-
-               /*
-                *      No reply code, toss the reply we have,
-                *      and do post-proxy-type Fail.
-                */
-               if (!request->proxy_reply->code) {
-                       TALLOC_FREE(request->proxy_reply);
-                       setup_post_proxy_fail(request);
-               }
-
-               /*
-                *      Just do the work here, rather than trying to
-                *      run the "decode proxy reply" stuff...
-                */
-               process_proxy_reply(request, request->proxy_reply);
-
-               /*
-                *      If we have a reply, run it through the handler.
-                */
-               if (request->proxy_reply) {
-                       request->handle(request); /* to do more post-proxy stuff */
-               }
-
-               return -1;      /* so we call request_finish */
-       }
+       if (request->home_server->server) return proxy_to_virtual_server(request);
 
        /*
         *      We're actually sending a proxied packet.  Do that now.
         */
        if (!request->in_proxy_hash && !insert_into_proxy_hash(request)) {
-               ERROR("Failed to insert request into the proxy list");
+               RPROXY("Failed to insert request into the proxy list");
                return -1;
        }
 
        rad_assert(request->proxy->id >= 0);
 
-       if (debug_flag) {
+       if (rad_debug_lvl) {
                struct timeval *response_window;
 
                response_window = request_response_window(request);
@@ -2969,11 +3228,8 @@ static int request_proxy(REQUEST *request, int retransmit)
 
        }
 
-       gettimeofday(&request->proxy_retransmit, NULL);
-       if (!retransmit) {
-               request->proxy->timestamp = request->proxy_retransmit;
-       }
-       request->home_server->last_packet_sent = request->proxy_retransmit.tv_sec;
+       gettimeofday(&request->proxy->timestamp, NULL);
+       request->home_server->last_packet_sent = request->proxy->timestamp.tv_sec;
 
        /*
         *      Encode the packet before we do anything else.
@@ -2984,9 +3240,15 @@ static int request_proxy(REQUEST *request, int retransmit)
        /*
         *      Set the state function, then the state, no child, and
         *      send the packet.
+        *
+        *      The order here is different from other state changes
+        *      due to race conditions with replies from the home
+        *      server.
         */
        request->process = proxy_wait_for_reply;
        request->child_state = REQUEST_PROXIED;
+       request->component = "<REQUEST_PROXIED>";
+       request->module = "";
        NO_CHILD_THREAD;
 
        /*
@@ -3031,46 +3293,65 @@ static int request_proxy_anew(REQUEST *request)
                }
                return 0;
        }
-       home_server_update_request(home, request);
-
-       if (!insert_into_proxy_hash(request)) {
-               RPROXY("Failed to insert retransmission into the proxy list");
-               goto post_proxy_fail;
-       }
-
-       /*
-        *      Free the old packet, to force re-encoding
-        */
-       talloc_free(request->proxy->data);
-       request->proxy->data = NULL;
-       request->proxy->data_len = 0;
 
 #ifdef WITH_ACCOUNTING
        /*
-        *      Update the Acct-Delay-Time attribute.
+        *      Update the Acct-Delay-Time attribute, since the LAST
+        *      time we tried to retransmit this packet.
         */
        if (request->packet->code == PW_CODE_ACCOUNTING_REQUEST) {
                VALUE_PAIR *vp;
 
-               vp = pairfind(request->proxy->vps, PW_ACCT_DELAY_TIME, 0, TAG_ANY);
-               if (!vp) vp = radius_paircreate(request->proxy,
+               vp = fr_pair_find_by_num(request->proxy->vps, PW_ACCT_DELAY_TIME, 0, TAG_ANY);
+               if (!vp) vp = radius_pair_create(request->proxy,
                                                &request->proxy->vps,
                                                PW_ACCT_DELAY_TIME, 0);
                if (vp) {
                        struct timeval now;
 
                        gettimeofday(&now, NULL);
-                       vp->vp_integer += now.tv_sec - request->proxy_retransmit.tv_sec;
+                       vp->vp_integer += now.tv_sec - request->proxy->timestamp.tv_sec;
                }
        }
 #endif
 
-       if (request_proxy(request, 1) != 1) goto post_proxy_fail;
+       /*
+        *      May have failed over to a "fallback" virtual server.
+        *      If so, run that instead of doing proxying to a real
+        *      server.
+        */
+       if (home->server) {
+               request->home_server = home;
+               TALLOC_FREE(request->proxy);
+
+               (void) proxy_to_virtual_server(request);
+               return 0;
+       }
+
+       home_server_update_request(home, request);
+
+       if (!insert_into_proxy_hash(request)) {
+               RPROXY("Failed to insert retransmission into the proxy list");
+               goto post_proxy_fail;
+       }
+
+       /*
+        *      Free the old packet, to force re-encoding
+        */
+       talloc_free(request->proxy->data);
+       request->proxy->data = NULL;
+       request->proxy->data_len = 0;
+
+       if (request_proxy(request) != 1) goto post_proxy_fail;
 
        return 1;
 }
 
-STATE_MACHINE_DECL(request_ping)
+
+/** Ping a home server.
+ *
+ */
+static void request_ping(REQUEST *request, int action)
 {
        home_server_t *home = request->home_server;
        char buffer[128];
@@ -3082,20 +3363,22 @@ STATE_MACHINE_DECL(request_ping)
 
        switch (action) {
        case FR_ACTION_TIMER:
-               ERROR("No response to status check %d for home server %s port %d",
+               ERROR("No response to status check %d ID %u for home server %s port %d",
                       request->number,
+                      request->proxy->id,
                       inet_ntop(request->proxy->dst_ipaddr.af,
                                 &request->proxy->dst_ipaddr.ipaddr,
                                 buffer, sizeof(buffer)),
                       request->proxy->dst_port);
+               remove_from_proxy_hash(request);
                break;
 
        case FR_ACTION_PROXY_REPLY:
                rad_assert(request->in_proxy_hash);
 
                request->home_server->num_received_pings++;
-               RPROXY("Received response to status check %d (%d in current sequence)",
-                      request->number, home->num_received_pings);
+               RPROXY("Received response to status check %d ID %u (%d in current sequence)",
+                      request->number, request->proxy->id, home->num_received_pings);
 
                /*
                 *      Remove the request from any hashes
@@ -3126,21 +3409,7 @@ STATE_MACHINE_DECL(request_ping)
                 *      Mark it alive and delete any outstanding
                 *      pings.
                 */
-               home->state = HOME_STATE_ALIVE;
-               home->response_timeouts = 0;
-               exec_trigger(request, home->cs, "home_server.alive", false);
-               home->currently_outstanding = 0;
-               home->num_sent_pings = 0;
-               home->num_received_pings = 0;
-               gettimeofday(&home->revive_time, NULL);
-
-               fr_event_delete(el, &home->ev);
-
-               RPROXY("Marking home server %s port %d alive",
-                      inet_ntop(request->proxy->dst_ipaddr.af,
-                                &request->proxy->dst_ipaddr.ipaddr,
-                                buffer, sizeof(buffer)),
-                      request->proxy->dst_port);
+               mark_home_server_alive(request, home);
                break;
 
        default:
@@ -3149,12 +3418,33 @@ STATE_MACHINE_DECL(request_ping)
        }
 
        rad_assert(!request->in_request_hash);
+       rad_assert(!request->in_proxy_hash);
        rad_assert(request->ev == NULL);
        NO_CHILD_THREAD;
        request_done(request, FR_ACTION_DONE);
 }
 
 /*
+ *     Add +/- 2s of jitter, as suggested in RFC 3539
+ *     and in RFC 5080.
+ */
+static void add_jitter(struct timeval *when)
+{
+       uint32_t jitter;
+
+       when->tv_sec -= 2;
+
+       jitter = fr_rand();
+       jitter ^= (jitter >> 10);
+       jitter &= ((1 << 22) - 1); /* 22 bits of 1 */
+
+       /*
+        *      Add in ~ (4 * USEC) of jitter.
+        */
+       tv_add(when, jitter);
+}
+
+/*
  *     Called from start of zombie period, OR after control socket
  *     marks the home server dead.
  */
@@ -3166,9 +3456,6 @@ static void ping_home_server(void *ctx)
        struct timeval when, now;
 
        if ((home->state == HOME_STATE_ALIVE) ||
-#ifdef WITH_TCP
-           (home->proto == IPPROTO_TCP) ||
-#endif
            (home->ev != NULL)) {
                return;
        }
@@ -3195,8 +3482,8 @@ static void ping_home_server(void *ctx)
         */
        if (home->ping_check == HOME_PING_CHECK_NONE) {
                if (home->state == HOME_STATE_ZOMBIE) {
-                       when = home->zombie_period_start;
-                       when.tv_sec += home->zombie_period;
+                       home->when = home->zombie_period_start;
+                       home->when.tv_sec += home->zombie_period;
                        INSERT_EVENT(ping_home_server, home);
                }
 
@@ -3219,55 +3506,64 @@ static void ping_home_server(void *ctx)
        if (home->ping_check == HOME_PING_CHECK_STATUS_SERVER) {
                request->proxy->code = PW_CODE_STATUS_SERVER;
 
-               pairmake(request->proxy, &request->proxy->vps,
+               fr_pair_make(request->proxy, &request->proxy->vps,
                         "Message-Authenticator", "0x00", T_OP_SET);
 
-       } else if (home->type == HOME_TYPE_AUTH) {
+       } else if ((home->type == HOME_TYPE_AUTH) ||
+                  (home->type == HOME_TYPE_AUTH_ACCT)) {
                request->proxy->code = PW_CODE_ACCESS_REQUEST;
 
-               pairmake(request->proxy, &request->proxy->vps,
+               fr_pair_make(request->proxy, &request->proxy->vps,
                         "User-Name", home->ping_user_name, T_OP_SET);
-               pairmake(request->proxy, &request->proxy->vps,
+               fr_pair_make(request->proxy, &request->proxy->vps,
                         "User-Password", home->ping_user_password, T_OP_SET);
-               pairmake(request->proxy, &request->proxy->vps,
+               fr_pair_make(request->proxy, &request->proxy->vps,
                         "Service-Type", "Authenticate-Only", T_OP_SET);
-               pairmake(request->proxy, &request->proxy->vps,
+               fr_pair_make(request->proxy, &request->proxy->vps,
                         "Message-Authenticator", "0x00", T_OP_SET);
 
-       } else {
 #ifdef WITH_ACCOUNTING
+       } else if (home->type == HOME_TYPE_ACCT) {
                request->proxy->code = PW_CODE_ACCOUNTING_REQUEST;
 
-               pairmake(request->proxy, &request->proxy->vps,
+               fr_pair_make(request->proxy, &request->proxy->vps,
                         "User-Name", home->ping_user_name, T_OP_SET);
-               pairmake(request->proxy, &request->proxy->vps,
+               fr_pair_make(request->proxy, &request->proxy->vps,
                         "Acct-Status-Type", "Stop", T_OP_SET);
-               pairmake(request->proxy, &request->proxy->vps,
+               fr_pair_make(request->proxy, &request->proxy->vps,
                         "Acct-Session-Id", "00000000", T_OP_SET);
-               vp = pairmake(request->proxy, &request->proxy->vps,
+               vp = fr_pair_make(request->proxy, &request->proxy->vps,
                              "Event-Timestamp", "0", T_OP_SET);
                vp->vp_date = now.tv_sec;
-#else
-               rad_assert("Internal sanity check failed");
 #endif
+
+       } else {
+               /*
+                *      Unkown home server type.
+                */
+               talloc_free(request);
+               return;
        }
 
-       vp = pairmake(request->proxy, &request->proxy->vps,
+       vp = fr_pair_make(request->proxy, &request->proxy->vps,
                      "NAS-Identifier", "", T_OP_SET);
        if (vp) {
-               pairsprintf(vp, "Status Check %u. Are you alive?",
+               fr_pair_value_sprintf(vp, "Status Check %u. Are you alive?",
                            home->num_sent_pings);
        }
 
+#ifdef WITH_TCP
+       request->proxy->proto = home->proto;
+#endif
        request->proxy->src_ipaddr = home->src_ipaddr;
        request->proxy->dst_ipaddr = home->ipaddr;
        request->proxy->dst_port = home->port;
        request->home_server = home;
 #ifdef DEBUG_STATE_MACHINE
-       if (debug_flag) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n", request->number, __FUNCTION__,
+       if (rad_debug_lvl) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n", request->number, __FUNCTION__,
                               child_state_names[request->child_state],
                               child_state_names[REQUEST_DONE]);
-       if (debug_flag) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_ping");
+       if (rad_debug_lvl) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_ping");
 #endif
 #ifdef HAVE_PTHREAD_H
        rad_assert(request->child_pid == NO_SUCH_CHILD_PID);
@@ -3301,6 +3597,7 @@ static void ping_home_server(void *ctx)
        home->num_sent_pings++;
 
        rad_assert(request->proxy_listener != NULL);
+       debug_packet(request, request->proxy, false);
        request->proxy_listener->send(request->proxy_listener,
                                      request);
 
@@ -3342,13 +3639,6 @@ static void mark_home_server_zombie(home_server_t *home, struct timeval *now, st
        rad_assert((home->state == HOME_STATE_ALIVE) ||
                   (home->state == HOME_STATE_UNKNOWN));
 
-#ifdef WITH_TCP
-       if (home->proto == IPPROTO_TCP) {
-               WARN("Not marking TCP server %s zombie", home->log_name);
-               return;
-       }
-#endif
-
        /*
         *      We've received a real packet recently.  Don't mark the
         *      server as zombie until we've received NO packets for a
@@ -3360,7 +3650,7 @@ static void mark_home_server_zombie(home_server_t *home, struct timeval *now, st
         */
        start = now->tv_sec - ((home->zombie_period + 3) / 4);
        if (home->last_packet_recv >= start) {
-               DEBUG("Recieved reply from home server %d seconds ago.  Might not be zombie.",
+               DEBUG("Received reply from home server %d seconds ago.  Might not be zombie.",
                      (int) (now->tv_sec - home->last_packet_recv));
                return;
        }
@@ -3394,10 +3684,6 @@ void revive_home_server(void *ctx)
        home_server_t *home = talloc_get_type_abort(ctx, home_server_t);
        char buffer[128];
 
-#ifdef WITH_TCP
-       rad_assert(home->proto != IPPROTO_TCP);
-#endif
-
        home->state = HOME_STATE_ALIVE;
        home->response_timeouts = 0;
        home_trigger(home, "home_server.alive");
@@ -3421,13 +3707,6 @@ void mark_home_server_dead(home_server_t *home, struct timeval *when)
        int previous_state = home->state;
        char buffer[128];
 
-#ifdef WITH_TCP
-       if (home->proto == IPPROTO_TCP) {
-               WARN("Not marking TCP server dead");
-               return;
-       }
-#endif
-
        PROXY( "Marking home server %s port %d as dead.",
               inet_ntop(home->ipaddr.af, &home->ipaddr.ipaddr,
                         buffer, sizeof(buffer)),
@@ -3462,7 +3741,26 @@ void mark_home_server_dead(home_server_t *home, struct timeval *when)
        }
 }
 
-STATE_MACHINE_DECL(proxy_wait_for_reply)
+/** Wait for a reply after proxying a request.
+ *
+ *  Retransmit the proxied packet, or time out and go to
+ *  proxy_no_reply.  Mark the home server unresponsive, etc.
+ *
+ *  If we do receive a reply, we transition to proxy_running.
+ *
+ *  \dot
+ *     digraph proxy_wait_for_reply {
+ *             proxy_wait_for_reply;
+ *
+ *             proxy_wait_for_reply -> retransmit_proxied_request [ label = "DUP", arrowhead = "none" ];
+ *             proxy_wait_for_reply -> proxy_no_reply [ label = "TIMER >= response_window" ];
+ *             proxy_wait_for_reply -> timer [ label = "TIMER < max_request_time" ];
+ *             proxy_wait_for_reply -> proxy_running [ label = "PROXY_REPLY" arrowhead = "none"];
+ *             proxy_wait_for_reply -> done [ label = "TIMER >= max_request_time" ];
+ *     }
+ *  \enddot
+ */
+static void proxy_wait_for_reply(REQUEST *request, int action)
 {
        struct timeval now, when;
        struct timeval *response_window = NULL;
@@ -3492,14 +3790,27 @@ STATE_MACHINE_DECL(proxy_wait_for_reply)
                 */
                if (request->home_server->server) return;
 
+               /*
+                *      Use a new connection when the home server is
+                *      dead, or when there's no proxy listener, or
+                *      when the listener is failed or dead.
+                *
+                *      If the listener is known or frozen, use it for
+                *      retransmits.
+                */
                if ((home->state == HOME_STATE_IS_DEAD) ||
                    !request->proxy_listener ||
-                   (request->proxy_listener->status != RAD_LISTEN_STATUS_KNOWN)) {
+                   (request->proxy_listener->status >= RAD_LISTEN_STATUS_EOL)) {
                        request_proxy_anew(request);
                        return;
                }
 
 #ifdef WITH_TCP
+               /*
+                *      The home server is still alive, but TCP.  We
+                *      rely on TCP to get the request and reply back.
+                *      So there's no need to retransmit.
+                */
                if (home->proto == IPPROTO_TCP) {
                        DEBUG2("Suppressing duplicate proxied request (tcp) to home server %s port %d proto TCP - ID: %d",
                               inet_ntop(request->proxy->dst_ipaddr.af,
@@ -3515,7 +3826,7 @@ STATE_MACHINE_DECL(proxy_wait_for_reply)
                 *      More than one retransmit a second is stupid,
                 *      and should be suppressed by the proxy.
                 */
-               when = request->proxy_retransmit;
+               when = request->proxy->timestamp;
                when.tv_sec++;
 
                if (timercmp(&now, &when, <)) {
@@ -3534,7 +3845,7 @@ STATE_MACHINE_DECL(proxy_wait_for_reply)
                 *      get a new ID.
                 */
                if ((request->packet->code == PW_CODE_ACCOUNTING_REQUEST) &&
-                   pairfind(request->proxy->vps, PW_ACCT_DELAY_TIME, 0, TAG_ANY)) {
+                   fr_pair_find_by_num(request->proxy->vps, PW_ACCT_DELAY_TIME, 0, TAG_ANY)) {
                        request_proxy_anew(request);
                        return;
                }
@@ -3551,9 +3862,9 @@ STATE_MACHINE_DECL(proxy_wait_for_reply)
                rad_assert(request->proxy_listener != NULL);
                FR_STATS_TYPE_INC(home->stats.total_requests);
                home->last_packet_sent = now.tv_sec;
-               request->proxy_retransmit = now;
-               request->proxy_listener->send(request->proxy_listener, request);
+               request->proxy->timestamp = now;
                debug_packet(request, request->proxy, false);
+               request->proxy_listener->send(request->proxy_listener, request);
                break;
 
        case FR_ACTION_TIMER:
@@ -3561,7 +3872,7 @@ STATE_MACHINE_DECL(proxy_wait_for_reply)
 
 #ifdef WITH_TCP
                if (!request->proxy_listener ||
-                   (request->proxy_listener->status != RAD_LISTEN_STATUS_KNOWN)) {
+                   (request->proxy_listener->status >= RAD_LISTEN_STATUS_EOL)) {
                        remove_from_proxy_hash(request);
 
                        when = request->packet->timestamp;
@@ -3608,18 +3919,11 @@ STATE_MACHINE_DECL(proxy_wait_for_reply)
                 *      "response_window", then mark the home server
                 *      as zombie.
                 *
-                *      If the connection is TCP, then another
-                *      "watchdog timer" function takes care of pings,
-                *      etc.  So we don't need to do it here.
-                *
                 *      This check should really be part of a home
                 *      server state machine.
                 */
                if (((home->state == HOME_STATE_ALIVE) ||
                     (home->state == HOME_STATE_UNKNOWN))
-#ifdef WITH_TCP
-                   && (home->proto != IPPROTO_TCP)
-#endif
                        ) {
                        home->response_timeouts++;
                        if (home->response_timeouts >= home->max_response_timeouts)
@@ -3637,6 +3941,17 @@ STATE_MACHINE_DECL(proxy_wait_for_reply)
                        FR_STATS_TYPE_INC(proxy_acct_stats.total_timeouts);
                }
 #endif
+#ifdef WITH_COA
+               else if (home->type == HOME_TYPE_COA) {
+                       if (request->proxy_listener) FR_STATS_TYPE_INC(request->proxy_listener->stats.total_timeouts);
+
+                       if (request->packet->code == PW_CODE_COA_REQUEST) {
+                               FR_STATS_TYPE_INC(proxy_coa_stats.total_timeouts);
+                       } else {
+                               FR_STATS_TYPE_INC(proxy_dsc_stats.total_timeouts);
+                       }
+               }
+#endif
 
                /*
                 *      There was no response within the window.  Stop
@@ -3687,6 +4002,7 @@ STATE_MACHINE_DECL(proxy_wait_for_reply)
 }
 #endif /* WITH_PROXY */
 
+
 /***********************************************************************
  *
  *  CoA code
@@ -3719,9 +4035,9 @@ static void request_coa_originate(REQUEST *request)
        /*
         *      Check whether we want to originate one, or cancel one.
         */
-       vp = pairfind(request->config_items, PW_SEND_COA_REQUEST, 0, TAG_ANY);
+       vp = fr_pair_find_by_num(request->config, PW_SEND_COA_REQUEST, 0, TAG_ANY);
        if (!vp) {
-               vp = pairfind(request->coa->proxy->vps, PW_SEND_COA_REQUEST, 0, TAG_ANY);
+               vp = fr_pair_find_by_num(request->coa->proxy->vps, PW_SEND_COA_REQUEST, 0, TAG_ANY);
        }
 
        if (vp) {
@@ -3732,22 +4048,28 @@ static void request_coa_originate(REQUEST *request)
                }
        }
 
+       if (!main_config.proxy_requests) {
+               RWDEBUG("Cannot originate CoA packets unless 'proxy_requests = yes'");
+                       TALLOC_FREE(request->coa);
+               return;
+       }
+
        coa = request->coa;
 
        /*
         *      src_ipaddr will be set up in proxy_encode.
         */
        memset(&ipaddr, 0, sizeof(ipaddr));
-       vp = pairfind(coa->proxy->vps, PW_PACKET_DST_IP_ADDRESS, 0, TAG_ANY);
+       vp = fr_pair_find_by_num(coa->proxy->vps, PW_PACKET_DST_IP_ADDRESS, 0, TAG_ANY);
        if (vp) {
                ipaddr.af = AF_INET;
                ipaddr.ipaddr.ip4addr.s_addr = vp->vp_ipaddr;
                ipaddr.prefix = 32;
-       } else if ((vp = pairfind(coa->proxy->vps, PW_PACKET_DST_IPV6_ADDRESS, 0, TAG_ANY)) != NULL) {
+       } else if ((vp = fr_pair_find_by_num(coa->proxy->vps, PW_PACKET_DST_IPV6_ADDRESS, 0, TAG_ANY)) != NULL) {
                ipaddr.af = AF_INET6;
                ipaddr.ipaddr.ip6addr = vp->vp_ipv6addr;
                ipaddr.prefix = 128;
-       } else if ((vp = pairfind(coa->proxy->vps, PW_HOME_SERVER_POOL, 0, TAG_ANY)) != NULL) {
+       } else if ((vp = fr_pair_find_by_num(coa->proxy->vps, PW_HOME_SERVER_POOL, 0, TAG_ANY)) != NULL) {
                coa->home_pool = home_pool_byname(vp->vp_strvalue,
                                                  HOME_TYPE_COA);
                if (!coa->home_pool) {
@@ -3787,7 +4109,7 @@ static void request_coa_originate(REQUEST *request)
        } else if (!coa->home_server) {
                uint16_t port = PW_COA_UDP_PORT;
 
-               vp = pairfind(coa->proxy->vps, PW_PACKET_DST_PORT, 0, TAG_ANY);
+               vp = fr_pair_find_by_num(coa->proxy->vps, PW_PACKET_DST_PORT, 0, TAG_ANY);
                if (vp) port = vp->vp_integer;
 
                coa->home_server = home_server_find(&ipaddr, port, IPPROTO_UDP);
@@ -3799,7 +4121,7 @@ static void request_coa_originate(REQUEST *request)
                }
        }
 
-       vp = pairfind(coa->proxy->vps, PW_PACKET_TYPE, 0, TAG_ANY);
+       vp = fr_pair_find_by_num(coa->proxy->vps, PW_PACKET_TYPE, 0, TAG_ANY);
        if (vp) {
                switch (vp->vp_integer) {
                case PW_CODE_COA_REQUEST:
@@ -3827,7 +4149,7 @@ static void request_coa_originate(REQUEST *request)
        coa->packet = rad_copy_packet(coa, request->packet);
        coa->reply = rad_copy_packet(coa, request->reply);
 
-       coa->config_items = paircopy(coa, request->config_items);
+       coa->config = fr_pair_list_copy(coa, request->config);
        coa->num_coa_requests = 0;
        coa->handle = null_handler;
        coa->number = request->number; /* it's associated with the same request */
@@ -3835,7 +4157,7 @@ static void request_coa_originate(REQUEST *request)
        /*
         *      Call the pre-proxy routines.
         */
-       vp = pairfind(request->config_items, PW_PRE_PROXY_TYPE, 0, TAG_ANY);
+       vp = fr_pair_find_by_num(request->config, PW_PRE_PROXY_TYPE, 0, TAG_ANY);
        if (vp) {
                DICT_VALUE const *dval = dict_valbyattr(vp->da->attr, vp->da->vendor, vp->vp_integer);
                /* Must be a validation issue */
@@ -3909,7 +4231,7 @@ static void request_coa_originate(REQUEST *request)
        debug_packet(coa, coa->proxy, false);
 
 #ifdef DEBUG_STATE_MACHINE
-       if (debug_flag) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n", request->number, __FUNCTION__,
+       if (rad_debug_lvl) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n", request->number, __FUNCTION__,
                               child_state_names[request->child_state],
                               child_state_names[REQUEST_PROXIED]);
 #endif
@@ -3925,6 +4247,8 @@ static void request_coa_originate(REQUEST *request)
        coa->child_pid = NO_SUCH_CHILD_PID;
 #endif
 
+       if (we_are_master()) coa_separate(request->coa);
+
        /*
         *      And send the packet.
         */
@@ -3936,15 +4260,12 @@ static void coa_retransmit(REQUEST *request)
 {
        uint32_t delay, frac;
        struct timeval now, when, mrd;
+       char buffer[128];
 
        VERIFY_REQUEST(request);
 
        fr_event_now(el, &now);
 
-       /*
-        *      FIXME: Enforce max_request_time
-        */
-
        if (request->delay == 0) {
                /*
                 *      Implement re-transmit algorithm as per RFC 5080
@@ -3981,8 +4302,6 @@ static void coa_retransmit(REQUEST *request)
         */
        if (request->home_server->coa_mrc &&
            (request->num_coa_requests >= request->home_server->coa_mrc)) {
-               char buffer[128];
-
                RERROR("Failing request - originate-coa ID %u, due to lack of any response from coa server %s port %d",
                       request->proxy->id,
                               inet_ntop(request->proxy->dst_ipaddr.af,
@@ -4049,11 +4368,37 @@ static void coa_retransmit(REQUEST *request)
 
        FR_STATS_TYPE_INC(request->home_server->stats.total_requests);
 
+       RDEBUG2("Sending duplicate CoA request to home server %s port %d - ID: %d",
+               inet_ntop(request->proxy->dst_ipaddr.af,
+                         &request->proxy->dst_ipaddr.ipaddr,
+                         buffer, sizeof(buffer)),
+               request->proxy->dst_port,
+               request->proxy->id);
+
        request->proxy_listener->send(request->proxy_listener,
                                      request);
 }
 
-STATE_MACHINE_DECL(coa_wait_for_reply)
+
+/** Wait for a reply after originating a CoA a request.
+ *
+ *  Retransmit the proxied packet, or time out and go to
+ *  coa_no_reply.  Mark the home server unresponsive, etc.
+ *
+ *  If we do receive a reply, we transition to coa_running.
+ *
+ *  \dot
+ *     digraph coa_wait_for_reply {
+ *             coa_wait_for_reply;
+ *
+ *             coa_wait_for_reply -> coa_no_reply [ label = "TIMER >= response_window" ];
+ *             coa_wait_for_reply -> timer [ label = "TIMER < max_request_time" ];
+ *             coa_wait_for_reply -> coa_running [ label = "PROXY_REPLY" arrowhead = "none"];
+ *             coa_wait_for_reply -> done [ label = "TIMER >= max_request_time" ];
+ *     }
+ *  \enddot
+ */
+static void coa_wait_for_reply(REQUEST *request, int action)
 {
        VERIFY_REQUEST(request);
 
@@ -4061,16 +4406,27 @@ STATE_MACHINE_DECL(coa_wait_for_reply)
        ASSERT_MASTER;
        CHECK_FOR_STOP;
 
+       if (request->parent) coa_separate(request);
+
        switch (action) {
        case FR_ACTION_TIMER:
-               if (request->parent) coa_separate(request);
+               if (request_max_time(request)) break;
+
+               /*
+                *      Don't do fail-over.  This is a 3.1 feature.
+                */
+               if (!request->home_server ||
+                   (request->home_server->state == HOME_STATE_IS_DEAD) ||
+                   !request->proxy_listener ||
+                   (request->proxy_listener->status >= RAD_LISTEN_STATUS_EOL)) {
+                       request_done(request, FR_ACTION_DONE);
+                       break;
+               }
 
                coa_retransmit(request);
                break;
 
        case FR_ACTION_PROXY_REPLY:
-               if (request->parent) coa_separate(request);
-
                request_queue_or_run(request, coa_running);
                break;
 
@@ -4101,9 +4457,31 @@ static void coa_separate(REQUEST *request)
        (void) talloc_steal(NULL, request);
        request->parent->coa = NULL;
        request->parent = NULL;
+
+       if (we_are_master()) {
+               request->delay = 0;
+               coa_retransmit(request);
+       }
 }
 
-STATE_MACHINE_DECL(coa_no_reply)
+
+/** Process a request after the CoA has timed out.
+ *
+ *  Run the packet through Post-Proxy-Type Fail
+ *
+ *  \dot
+ *     digraph coa_no_reply {
+ *             coa_no_reply;
+ *
+ *             coa_no_reply -> dup [ label = "DUP", arrowhead = "none" ];
+ *             coa_no_reply -> timer [ label = "TIMER < max_request_time" ];
+ *             coa_no_reply -> coa_reply_too_late [ label = "PROXY_REPLY" arrowhead = "none"];
+ *             coa_no_reply -> process_proxy_reply [ label = "RUN" ];
+ *             coa_no_reply -> done [ label = "TIMER >= timeout" ];
+ *     }
+ *  \enddot
+ */
+static void coa_no_reply(REQUEST *request, int action)
 {
        char buffer[128];
 
@@ -4114,7 +4492,7 @@ STATE_MACHINE_DECL(coa_no_reply)
 
        switch (action) {
        case FR_ACTION_TIMER:
-               request_max_time(request);
+               (void) request_max_time(request);
                break;
 
        case FR_ACTION_PROXY_REPLY: /* too late! */
@@ -4138,7 +4516,23 @@ STATE_MACHINE_DECL(coa_no_reply)
        }
 }
 
-STATE_MACHINE_DECL(coa_running)
+
+/** Process the request after receiving a coa reply.
+ *
+ *  Throught the post-proxy section, and the through the handler
+ *  function.
+ *
+ *  \dot
+ *     digraph coa_running {
+ *             coa_running;
+ *
+ *             coa_running -> timer [ label = "TIMER < max_request_time" ];
+ *             coa_running -> process_proxy_reply [ label = "RUN" ];
+ *             coa_running -> done [ label = "TIMER >= timeout" ];
+ *     }
+ *  \enddot
+ */
+static void coa_running(REQUEST *request, int action)
 {
        VERIFY_REQUEST(request);
 
@@ -4147,7 +4541,7 @@ STATE_MACHINE_DECL(coa_running)
 
        switch (action) {
        case FR_ACTION_TIMER:
-               request_max_time(request);
+               (void) request_max_time(request);
                break;
 
        case FR_ACTION_RUN:
@@ -4248,7 +4642,7 @@ static void event_status(struct timeval *wake)
        int argval;
 #endif
 
-       if (debug_flag == 0) {
+       if (rad_debug_lvl == 0) {
                if (just_started) {
                        INFO("Ready to process requests");
                        just_started = false;
@@ -4316,7 +4710,6 @@ static void listener_free_cb(void *ctx)
        rad_assert(this->next == NULL);
        talloc_free(this);
 }
-#endif
 
 #ifdef WITH_PROXY
 static int proxy_eol_cb(void *ctx, void *data)
@@ -4356,7 +4749,8 @@ static int proxy_eol_cb(void *ctx, void *data)
         */
        return 0;
 }
-#endif
+#endif /* WITH_PROXY */
+#endif /* WITH_TCP */
 
 static int event_new_fd(rad_listen_t *this)
 {
@@ -4374,11 +4768,14 @@ static int event_new_fd(rad_listen_t *this)
                rad_assert(sock != NULL);
                if (just_started) {
                        DEBUG("Listening on %s", buffer);
+               } else {
+                       INFO(" ... adding new socket %s", buffer);
+               }
 
 #ifdef WITH_PROXY
-               } else if (this->type == RAD_LISTEN_PROXY) {
+               if (!just_started && (this->type == RAD_LISTEN_PROXY)) {
                        home_server_t *home;
-
+                       
                        home = sock->home;
                        if (!home || !home->limit.max_connections) {
                                INFO(" ... adding new socket %s", buffer);
@@ -4388,8 +4785,6 @@ static int event_new_fd(rad_listen_t *this)
                        }
 
 #endif
-               } else {
-                       INFO(" ... adding new socket %s", buffer);
                }
 
                switch (this->type) {
@@ -4438,7 +4833,7 @@ static int event_new_fd(rad_listen_t *this)
                                        rad_panic("Failed to insert event");
                                }
                        }
-#endif
+#endif /* WITH_TCP */
                        break;
 #endif /* WITH_PROXY */
 
@@ -4465,7 +4860,7 @@ static int event_new_fd(rad_listen_t *this)
                                        fr_exit(1);
                                }
                        }
-#endif
+#endif /* WITH_TCP */
                        break;
                } /* switch over listener types */
 
@@ -4484,7 +4879,40 @@ static int event_new_fd(rad_listen_t *this)
 
 #ifdef WITH_TCP
        /*
-        *      Stop using this socket, if at all possible.
+        *      The socket has reached a timeout.  Try to close it.
+        */
+       if (this->status == RAD_LISTEN_STATUS_FROZEN) {
+               /*
+                *      Requests are still using the socket.  Wait for
+                *      them to finish.
+                */
+               if (this->count > 0) {
+                       struct timeval when;
+                       listen_socket_t *sock = this->data;
+
+                       /*
+                        *      Try again to clean up the socket in 30
+                        *      seconds.
+                        */
+                       gettimeofday(&when, NULL);
+                       when.tv_sec += 30;
+
+                       ASSERT_MASTER;
+                       if (!fr_event_insert(el,
+                                            (fr_event_callback_t) event_new_fd,
+                                            this, &when, &sock->ev)) {
+                               rad_panic("Failed to insert event");
+                       }
+
+                       return 1;
+               }
+
+               fr_event_fd_delete(el, 0, this->fd);
+               this->status = RAD_LISTEN_STATUS_REMOVE_NOW;
+       }
+
+       /*
+        *      The socket has had a catastrophic error.  Close it.
         */
        if (this->status == RAD_LISTEN_STATUS_EOL) {
                /*
@@ -4494,10 +4922,7 @@ static int event_new_fd(rad_listen_t *this)
 
 #ifdef WITH_PROXY
                /*
-                *      Proxy sockets get frozen, so that we don't use
-                *      them for new requests.  But we do keep them
-                *      open to listen for replies to requests we had
-                *      previously sent.
+                *      Tell all requests using this socket that the socket is dead.
                 */
                if (this->type == RAD_LISTEN_PROXY) {
                        PTHREAD_MUTEX_LOCK(&proxy_mutex);
@@ -4507,10 +4932,12 @@ static int event_new_fd(rad_listen_t *this)
                                fr_exit(1);
                        }
 
-                       fr_packet_list_walk(proxy_list, this, proxy_eol_cb);
+                       if (this->count > 0) {
+                               fr_packet_list_walk(proxy_list, this, proxy_eol_cb);
+                       }
                        PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
                }
-#endif
+#endif /* WITH_PROXY */
 
                /*
                 *      Requests are still using the socket.  Wait for
@@ -4542,7 +4969,7 @@ static int event_new_fd(rad_listen_t *this)
                 */
                this->status = RAD_LISTEN_STATUS_REMOVE_NOW;
        } /* socket is at EOL */
-#endif
+#endif   /* WITH_TCP */
 
        /*
         *      Nuke the socket.
@@ -4551,8 +4978,8 @@ static int event_new_fd(rad_listen_t *this)
                int devnull;
 #ifdef WITH_TCP
                listen_socket_t *sock = this->data;
-#endif
                struct timeval when;
+#endif
 
                /*
                 *      Re-open the socket, pointing it to /dev/null.
@@ -4613,7 +5040,7 @@ static int event_new_fd(rad_listen_t *this)
                        }
                        PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
                } else
-#endif
+#endif /* WITH_PROXY */
                {
                        INFO(" ... shutting down socket %s", buffer);
 
@@ -4644,8 +5071,8 @@ static int event_new_fd(rad_listen_t *this)
                                     &(sock->ev))) {
                        rad_panic("Failed to insert event");
                }
-       }
 #endif /* WITH_TCP */
+       }
 
        return 1;
 }
@@ -4693,8 +5120,7 @@ static void handle_signal_self(int flag)
                fr_event_loop_exit(el, 0x80);
        }
 
-#ifdef WITH_DETAIL
-#ifndef WITH_DETAIL_THREAD
+#if defined(WITH_DETAIL) && !defined(WITH_DETAIL_THREAD)
        if ((flag & RADIUS_SIGNAL_SELF_DETAIL) != 0) {
                rad_listen_t *this;
 
@@ -4719,11 +5145,8 @@ static void handle_signal_self(int flag)
                }
        }
 #endif
-#endif
 
-#ifdef WITH_TCP
-#ifdef WITH_PROXY
-#ifdef HAVE_PTHREAD_H
+#if defined(WITH_TCP) && defined(WITH_PROXY) && defined(HAVE_PTHREAD_H)
        /*
         *      There are new listeners in the list.  Run
         *      event_new_fd() on them.
@@ -4747,14 +5170,16 @@ static void handle_signal_self(int flag)
                new_listeners = NULL;
                FD_MUTEX_UNLOCK(&fd_mutex);
        }
-#endif /* HAVE_PTHREAD_H */
-#endif /* WITH_PROXY */
-#endif /* WITH_TCP */
+#endif
 }
 
 #ifndef HAVE_PTHREAD_H
 void radius_signal_self(int flag)
 {
+       if (flag == RADIUS_SIGNAL_SELF_TERM) {
+               main_config.exiting = true;
+       }
+
        return handle_signal_self(flag);
 }
 
@@ -4769,6 +5194,10 @@ void radius_signal_self(int flag)
        ssize_t rcode;
        uint8_t buffer[16];
 
+       if (flag == RADIUS_SIGNAL_SELF_TERM) {
+               main_config.exiting = true;
+       }
+
        /*
         *      The read MUST be non-blocking for this to work.
         */
@@ -4833,6 +5262,119 @@ static int packet_entry_cmp(void const *one, void const *two)
        return fr_packet_cmp(*a, *b);
 }
 
+#ifdef WITH_PROXY
+/*
+ *     They haven't defined a proxy listener.  Automatically
+ *     add one for them, with the correct address family.
+ */
+static void create_default_proxy_listener(int af)
+{
+       uint16_t        port = 0;
+       home_server_t   home;
+       listen_socket_t *sock;
+       rad_listen_t    *this;
+
+       memset(&home, 0, sizeof(home));
+
+       /*
+        *      Open a default UDP port
+        */
+       home.proto = IPPROTO_UDP;
+       port = 0;
+
+       /*
+        *      Set the address family.
+        */
+       home.src_ipaddr.af = af;
+       home.ipaddr.af = af;
+
+       /*
+        *      Get the correct listener.
+        */
+       this = proxy_new_listener(proxy_ctx, &home, port);
+       if (!this) {
+               fr_exit_now(1);
+       }
+
+       sock = this->data;
+       if (!fr_packet_list_socket_add(proxy_list, this->fd,
+                                      sock->proto,
+                                      &sock->other_ipaddr, sock->other_port,
+                                      this)) {
+               ERROR("Failed adding proxy socket");
+               fr_exit_now(1);
+       }
+
+       /*
+        *      Insert the FD into list of FDs to listen on.
+        */
+       radius_update_listener(this);
+}
+
+/*
+ *     See if we automatically need to open a proxy socket.
+ */
+static void check_proxy(rad_listen_t *head)
+{
+       bool            defined_proxy;
+       bool            has_v4, has_v6;
+       rad_listen_t    *this;
+
+       if (check_config) return;
+       if (!main_config.proxy_requests) return;
+       if (!head) return;
+       if (!home_servers_udp) return;
+
+       /*
+        *      We passed "-i" on the command line.  Use that address
+        *      family for the proxy socket.
+        */
+       if (main_config.myip.af != AF_UNSPEC) {
+               create_default_proxy_listener(main_config.myip.af);
+               return;
+       }
+
+       defined_proxy = has_v4 = has_v6 = false;
+
+       /*
+        *      Figure out if we need to open a proxy socket, and if
+        *      so, which one.
+        */
+       for (this = head; this != NULL; this = this->next) {
+               listen_socket_t *sock;
+
+               switch (this->type) {
+               case RAD_LISTEN_PROXY:
+                       defined_proxy = true;
+                       break;
+
+               case RAD_LISTEN_AUTH:
+#ifdef WITH_ACCT
+               case RAD_LISTEN_ACCT:
+#endif
+#ifdef WITH_COA
+               case RAD_LISTEN_COA:
+#endif
+                       sock = this->data;
+                       if (sock->my_ipaddr.af == AF_INET) has_v4 = true;
+                       if (sock->my_ipaddr.af == AF_INET6) has_v6 = true;
+                       break;
+                       
+               default:
+                       break;
+               }
+       }
+
+       /*
+        *      Assume they know what they're doing.
+        */
+       if (defined_proxy) return;
+
+       if (has_v4) create_default_proxy_listener(AF_INET);
+
+       if (has_v6) create_default_proxy_listener(AF_INET6);
+}
+#endif
 
 int radius_event_start(CONF_SECTION *cs, bool have_children)
 {
@@ -4855,7 +5397,7 @@ int radius_event_start(CONF_SECTION *cs, bool have_children)
        request_num_counter = 0;
 
 #ifdef WITH_PROXY
-       if (main_config.proxy_requests) {
+       if (main_config.proxy_requests && !check_config) {
                /*
                 *      Create the tree for managing proxied requests and
                 *      responses.
@@ -4942,23 +5484,27 @@ int radius_event_start(CONF_SECTION *cs, bool have_children)
        }
 #endif
 
-       DEBUG("%s: #### Opening IP addresses and Ports ####", main_config.name);
-
-       /*
-             The server temporarily switches to an unprivileged
-             user very early in the bootstrapping process.
-             However, some sockets MAY require privileged access
-             (bind to device, or to port < 1024, or to raw
-             sockets).  Those sockets need to call suid up/down
-             themselves around the functions that need a privileged
-             uid.
-       */
-       if (listen_init(cs, &head, spawn_flag) < 0) {
+       DEBUG("%s: #### Opening IP addresses and Ports ####", main_config.name);
+
+       /*
+        *      The server temporarily switches to an unprivileged
+        *      user very early in the bootstrapping process.
+        *      However, some sockets MAY require privileged access
+        *      (bind to device, or to port < 1024, or to raw
+        *      sockets).  Those sockets need to call suid up/down
+        *      themselves around the functions that need a privileged
+        *      uid.
+        */
+       if (listen_init(cs, &head, spawn_flag) < 0) {
                fr_exit_now(1);
        }
 
        main_config.listen = head;
 
+#ifdef WITH_PROXY
+       check_proxy(head);
+#endif
+
        /*
         *      At this point, no one has any business *ever* going
         *      back to root uid.