2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License as published by
4 * the Free Software Foundation; either version 2 of the License, or
5 * (at your option) any later version.
7 * This program is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10 * GNU General Public License for more details.
12 * You should have received a copy of the GNU General Public License
13 * along with this program; if not, write to the Free Software
14 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
21 * @brief Defines the state machines that control how requests are processed.
23 * @copyright 2012 The FreeRADIUS server project
24 * @copyright 2012 Alan DeKok <aland@deployingradius.com>
29 #include <freeradius-devel/radiusd.h>
30 #include <freeradius-devel/process.h>
31 #include <freeradius-devel/modules.h>
32 #include <freeradius-devel/state.h>
34 #include <freeradius-devel/rad_assert.h>
37 #include <freeradius-devel/detail.h>
43 #ifdef HAVE_SYS_WAIT_H
44 # include <sys/wait.h>
47 extern pid_t radius_pid;
48 extern fr_cond_t *debug_condition;
50 static bool spawn_flag = false;
51 static bool just_started = true;
52 time_t fr_start_time = (time_t)-1;
53 static rbtree_t *pl = NULL;
54 static fr_event_list_t *el = NULL;
56 fr_event_list_t *radius_event_list_corral(UNUSED event_corral_t hint) {
57 /* Currently we do not run a second event loop for modules. */
61 static char const *action_codes[] = {
72 #ifdef DEBUG_STATE_MACHINE
73 # define TRACE_STATE_MACHINE \
74 if (rad_debug_lvl) do { \
75 struct timeval debug_tv; \
76 gettimeofday(&debug_tv, NULL); \
77 debug_tv.tv_sec -= fr_start_time; \
78 printf("(%u) %d.%06d ********\tSTATE %s action %s live M-%s C-%s\t********\n",\
79 request->number, (int) debug_tv.tv_sec, (int) debug_tv.tv_usec, \
80 __FUNCTION__, action_codes[action], master_state_names[request->master_state], \
81 child_state_names[request->child_state]); \
84 static char const *master_state_names[REQUEST_MASTER_NUM_STATES] = {
91 static char const *child_state_names[REQUEST_CHILD_NUM_STATES] = {
102 # define TRACE_STATE_MACHINE {}
105 static NEVER_RETURNS void _rad_panic(char const *file, unsigned int line, char const *msg)
107 ERROR("%s[%u]: %s", file, line, msg);
111 #define rad_panic(x) _rad_panic(__FILE__, __LINE__, x)
113 /** Declare a state in the state machine
115 * Expands to the start of a function definition for a given state.
117 * @param _x the name of the state.
119 #define STATE_MACHINE_DECL(_x) static void _x(REQUEST *request, int action)
121 static void request_timer(void *ctx);
123 /** Insert #REQUEST back into the event heap, to continue executing at a future time
125 * @param file the state machine timer call occurred in.
126 * @param line the state machine timer call occurred on.
127 * @param request to set add the timer event for.
128 * @param when the event should fine.
129 * @param action to perform when we resume processing the request.
131 static inline void state_machine_timer(char const *file, int line, REQUEST *request,
132 struct timeval *when, fr_state_action_t action)
134 request->timer_action = action;
135 if (!fr_event_insert(el, request_timer, request, when, &request->ev)) {
136 _rad_panic(file, line, "Failed to insert event");
140 /** @copybrief state_machine_timer
142 * @param _x the action to perform when we resume processing the request.
144 #define STATE_MACHINE_TIMER(_x) state_machine_timer(__FILE__, __LINE__, request, &when, _x)
147 * We need a different VERIFY_REQUEST macro in process.c
148 * To avoid the race conditions with the master thread
149 * checking the REQUEST whilst it's being worked on by
152 #if defined(WITH_VERIFY_PTR) && defined(HAVE_PTHREAD_H)
153 # undef VERIFY_REQUEST
154 # define VERIFY_REQUEST(_x) if (pthread_equal(pthread_self(), _x->child_pid) != 0) verify_request(__FILE__, __LINE__, _x)
158 * @section request_timeline
160 * Time sequence of a request
163 * RQ-----------------P=============================Y-J-C
164 * ::::::::::::::::::::::::::::::::::::::::::::::::::::::::M
167 * - R: received. Duplicate detection is done, and request is
170 * - Q: Request is placed onto a queue for child threads to pick up.
171 * If there are no child threads, the request goes immediately
174 * - P: Processing the request through the modules.
176 * - Y: Reply is ready. Rejects MAY be delayed here. All other
177 * replies are sent immediately.
179 * - J: Reject is sent "response_delay" after the reply is ready.
181 * - C: For Access-Requests, After "cleanup_delay", the request is
182 * deleted. Accounting-Request packets go directly from Y to C.
184 * - M: Max request time. If the request hits this timer, it is
187 * Other considerations include duplicate and conflicting
188 * packets. When a dupicate packet is received, it is ignored
189 * until we've reached Y, as no response is ready. If the reply
190 * is a reject, duplicates are ignored until J, when we're ready
191 * to send the reply. In between the reply being sent (Y or J),
192 * and C, the server responds to duplicates by sending the cached
195 * Conflicting packets are sent in 2 situations.
197 * The first is in between R and Y. In that case, we consider
198 * it as a hint that we're taking too long, and the NAS has given
199 * up on the request. We then behave just as if the M timer was
200 * reached, and we discard the current request. This allows us
201 * to process the new one.
203 * The second case is when we're at Y, but we haven't yet
204 * finished processing the request. This is a race condition in
205 * the threading code (avoiding locks is faster). It means that
206 * a thread has actually encoded and sent the reply, and that the
207 * NAS has responded with a new packet. The server can then
208 * safely mark the current request as "OK to delete", and behaves
209 * just as if the M timer was reached. This usually happens only
210 * in high-load situations.
212 * Duplicate packets are sent when the NAS thinks we're taking
213 * too long, and wants a reply. From R-Y, duplicates are
214 * ignored. From Y-J (for Access-Rejects), duplicates are also
215 * ignored. From Y-C, duplicates get a duplicate reply. *And*,
216 * they cause the "cleanup_delay" time to be extended. This
217 * extension means that we're more likely to send a duplicate
218 * reply (if we have one), or to suppress processing the packet
219 * twice if we didn't reply to it.
221 * All functions in this file should be thread-safe, and should
222 * assume thet the REQUEST structure is being accessed
223 * simultaneously by the main thread, and by the child worker
224 * threads. This means that timers, etc. cannot be updated in
227 * Instead, the master thread periodically calls request->process
228 * with action TIMER. It's up to the individual functions to
229 * determine how to handle that. They need to check if they're
230 * being called from a child thread or the master, and then do
231 * different things based on that.
234 static fr_packet_list_t *proxy_list = NULL;
235 static TALLOC_CTX *proxy_ctx = NULL;
238 #ifdef HAVE_PTHREAD_H
240 static pthread_mutex_t proxy_mutex;
241 static bool proxy_no_new_sockets = false;
244 # define PTHREAD_MUTEX_LOCK if (spawn_flag) pthread_mutex_lock
245 # define PTHREAD_MUTEX_UNLOCK if (spawn_flag) pthread_mutex_unlock
247 static pthread_t NO_SUCH_CHILD_PID;
248 # define NO_CHILD_THREAD request->child_pid = NO_SUCH_CHILD_PID
252 * This is easier than ifdef's throughout the code.
254 # define PTHREAD_MUTEX_LOCK(_x)
255 # define PTHREAD_MUTEX_UNLOCK(_x)
256 # define NO_CHILD_THREAD
259 #ifdef HAVE_PTHREAD_H
260 static bool we_are_master(void)
263 (pthread_equal(pthread_self(), NO_SUCH_CHILD_PID) == 0)) {
271 * Assertions are debug checks.
274 # define ASSERT_MASTER if (!we_are_master()) rad_panic("We are not master")
279 * No threads: we're always master.
281 # define we_are_master(_x) (1)
282 #endif /* HAVE_PTHREAD_H */
284 #ifndef ASSERT_MASTER
285 # define ASSERT_MASTER
289 * Make state transitions simpler.
291 #define FINAL_STATE(_x) NO_CHILD_THREAD; request->component = "<" #_x ">"; request->module = ""; request->child_state = _x
294 static int event_new_fd(rad_listen_t *this);
297 * We need mutexes around the event FD list *only* in certain
300 #if defined (HAVE_PTHREAD_H) && (defined(WITH_PROXY) || defined(WITH_TCP))
301 static rad_listen_t *new_listeners = NULL;
303 static pthread_mutex_t fd_mutex;
304 # define FD_MUTEX_LOCK if (spawn_flag) pthread_mutex_lock
305 # define FD_MUTEX_UNLOCK if (spawn_flag) pthread_mutex_unlock
307 void radius_update_listener(rad_listen_t *this)
310 * Just do it ourselves.
312 if (we_are_master()) {
317 FD_MUTEX_LOCK(&fd_mutex);
320 * If it's already in the list, don't add it again.
323 FD_MUTEX_UNLOCK(&fd_mutex);
328 * Otherwise, add it to the list
330 this->next = new_listeners;
331 new_listeners = this;
332 FD_MUTEX_UNLOCK(&fd_mutex);
333 radius_signal_self(RADIUS_SIGNAL_SELF_NEW_FD);
336 void radius_update_listener(rad_listen_t *this)
339 * No threads. Just insert it.
344 * This is easier than ifdef's throughout the code.
346 # define FD_MUTEX_LOCK(_x)
347 # define FD_MUTEX_UNLOCK(_x)
350 static int request_num_counter = 1;
352 static int request_will_proxy(REQUEST *request) CC_HINT(nonnull);
353 static int request_proxy(REQUEST *request, int retransmit) CC_HINT(nonnull);
354 STATE_MACHINE_DECL(request_ping) CC_HINT(nonnull);
356 STATE_MACHINE_DECL(request_response_delay) CC_HINT(nonnull);
357 STATE_MACHINE_DECL(request_cleanup_delay) CC_HINT(nonnull);
358 STATE_MACHINE_DECL(request_running) CC_HINT(nonnull);
359 STATE_MACHINE_DECL(request_done) CC_HINT(nonnull);
361 STATE_MACHINE_DECL(proxy_no_reply) CC_HINT(nonnull);
362 STATE_MACHINE_DECL(proxy_running) CC_HINT(nonnull);
363 STATE_MACHINE_DECL(proxy_wait_for_reply) CC_HINT(nonnull);
365 static int process_proxy_reply(REQUEST *request, RADIUS_PACKET *reply) CC_HINT(nonnull (1));
366 static void remove_from_proxy_hash(REQUEST *request) CC_HINT(nonnull);
367 static void remove_from_proxy_hash_nl(REQUEST *request, bool yank) CC_HINT(nonnull);
368 static int insert_into_proxy_hash(REQUEST *request) CC_HINT(nonnull);
371 static REQUEST *request_setup(TALLOC_CTX *ctx, rad_listen_t *listener, RADIUS_PACKET *packet,
372 RADCLIENT *client, RAD_REQUEST_FUNP fun);
373 static int request_pre_handler(REQUEST *request, UNUSED int action) CC_HINT(nonnull);
376 static void request_coa_originate(REQUEST *request) CC_HINT(nonnull);
377 STATE_MACHINE_DECL(coa_wait_for_reply) CC_HINT(nonnull);
378 STATE_MACHINE_DECL(coa_no_reply) CC_HINT(nonnull);
379 STATE_MACHINE_DECL(coa_running) CC_HINT(nonnull);
380 static void coa_separate(REQUEST *request) CC_HINT(nonnull);
381 # define COA_SEPARATE if (request->coa) coa_separate(request->coa);
383 # define COA_SEPARATE
386 #define CHECK_FOR_STOP do { if (request->master_state == REQUEST_STOP_PROCESSING) {request_done(request, FR_ACTION_DONE);return;}} while (0)
389 #define USEC (1000000)
391 #define INSERT_EVENT(_function, _ctx) if (!fr_event_insert(el, _function, _ctx, &((_ctx)->when), &((_ctx)->ev))) { _rad_panic(__FILE__, __LINE__, "Failed to insert event"); }
393 static void tv_add(struct timeval *tv, int usec_delay)
395 if (usec_delay >= USEC) {
396 tv->tv_sec += usec_delay / USEC;
399 tv->tv_usec += usec_delay;
401 if (tv->tv_usec >= USEC) {
402 tv->tv_sec += tv->tv_usec / USEC;
408 * Debug the packet if requested.
410 static void debug_packet(REQUEST *request, RADIUS_PACKET *packet, bool received)
412 char src_ipaddr[128];
413 char dst_ipaddr[128];
416 if (!RDEBUG_ENABLED) return;
419 * Client-specific debugging re-prints the input
420 * packet into the client log.
422 * This really belongs in a utility library
424 if (is_radius_code(packet->code)) {
425 RDEBUG("%s %s Id %i from %s%s%s:%i to %s%s%s:%i length %zu",
426 received ? "Received" : "Sent",
427 fr_packet_codes[packet->code],
429 packet->src_ipaddr.af == AF_INET6 ? "[" : "",
430 inet_ntop(packet->src_ipaddr.af,
431 &packet->src_ipaddr.ipaddr,
432 src_ipaddr, sizeof(src_ipaddr)),
433 packet->src_ipaddr.af == AF_INET6 ? "]" : "",
435 packet->dst_ipaddr.af == AF_INET6 ? "[" : "",
436 inet_ntop(packet->dst_ipaddr.af,
437 &packet->dst_ipaddr.ipaddr,
438 dst_ipaddr, sizeof(dst_ipaddr)),
439 packet->dst_ipaddr.af == AF_INET6 ? "]" : "",
443 RDEBUG("%s code %u Id %i from %s%s%s:%i to %s%s%s:%i length %zu\n",
444 received ? "Received" : "Sent",
447 packet->src_ipaddr.af == AF_INET6 ? "[" : "",
448 inet_ntop(packet->src_ipaddr.af,
449 &packet->src_ipaddr.ipaddr,
450 src_ipaddr, sizeof(src_ipaddr)),
451 packet->src_ipaddr.af == AF_INET6 ? "]" : "",
453 packet->dst_ipaddr.af == AF_INET6 ? "[" : "",
454 inet_ntop(packet->dst_ipaddr.af,
455 &packet->dst_ipaddr.ipaddr,
456 dst_ipaddr, sizeof(dst_ipaddr)),
457 packet->dst_ipaddr.af == AF_INET6 ? "]" : "",
463 rdebug_pair_list(L_DBG_LVL_1, request, packet->vps, NULL);
465 rdebug_proto_pair_list(L_DBG_LVL_1, request, packet->vps);
470 /***********************************************************************
472 * Start of RADIUS server state machine.
474 ***********************************************************************/
476 static struct timeval *request_response_window(REQUEST *request)
478 VERIFY_REQUEST(request);
480 if (request->client) {
482 * The client hasn't set the response window. Return
483 * either the home server one, if set, or the global one.
485 if (!timerisset(&request->client->response_window)) {
486 return &request->home_server->response_window;
489 if (timercmp(&request->client->response_window,
490 &request->home_server->response_window, <)) {
491 return &request->client->response_window;
495 rad_assert(request->home_server != NULL);
496 return &request->home_server->response_window;
500 * Determine initial request processing delay.
502 static int request_init_delay(REQUEST *request)
504 struct timeval half_response_window;
506 VERIFY_REQUEST(request);
508 /* Allow client response window to lower initial delay */
509 if (timerisset(&request->client->response_window)) {
510 half_response_window.tv_sec = request->client->response_window.tv_sec >> 1;
511 half_response_window.tv_usec =
512 ((request->client->response_window.tv_sec & 1) * USEC +
513 request->client->response_window.tv_usec) >> 1;
514 if (timercmp(&half_response_window, &request->root->init_delay, <))
515 return (int)half_response_window.tv_sec * USEC +
516 (int)half_response_window.tv_usec;
519 return (int)request->root->init_delay.tv_sec * USEC +
520 (int)request->root->init_delay.tv_usec;
524 * Callback for ALL timer events related to the request.
526 static void request_timer(void *ctx)
528 REQUEST *request = talloc_get_type_abort(ctx, REQUEST);
531 action = request->timer_action;
535 request->process(request, action);
539 * Wrapper for talloc pools. If there's no parent, just free the
540 * request. If there is a parent, free the parent INSTEAD of the
543 static void request_free(REQUEST *request)
547 rad_assert(request->ev == NULL);
548 rad_assert(!request->in_request_hash);
549 rad_assert(!request->in_proxy_hash);
551 if ((request->options & RAD_REQUEST_OPTION_CTX) == 0) {
552 talloc_free(request);
556 ptr = talloc_parent(request);
557 rad_assert(ptr != NULL);
563 static void proxy_reply_too_late(REQUEST *request)
567 RDEBUG2("Reply from home server %s port %d - ID: %d arrived too late. Try increasing 'retry_delay' or 'max_request_time'",
568 inet_ntop(request->proxy->dst_ipaddr.af,
569 &request->proxy->dst_ipaddr.ipaddr,
570 buffer, sizeof(buffer)),
571 request->proxy->dst_port, request->proxy->id);
576 /** Mark a request DONE and clean it up.
578 * When a request is DONE, it can have ties to a number of other
579 * portions of the server. The request hash, proxy hash, events,
580 * child threads, etc. This function takes care of either cleaning
581 * up the request, or managing the timers to wait for the ties to be
586 * done -> done [ label = "still running" ];
590 static void request_done(REQUEST *request, int action)
592 struct timeval now, when;
594 VERIFY_REQUEST(request);
599 * Force this no matter what.
601 request->process = request_done;
605 * Tell the detail listener that we're done.
607 if (request->listener &&
608 (request->listener->type == RAD_LISTEN_DETAIL) &&
609 (request->simul_max != 1)) {
610 request->simul_max = 1;
611 request->listener->send(request->listener,
616 #ifdef HAVE_PTHREAD_H
618 * If called from a child thread, mark ourselves as done,
619 * and wait for the master thread timer to clean us up.
621 if (!we_are_master()) {
622 FINAL_STATE(REQUEST_DONE);
628 * Mark the request as STOP.
630 request->master_state = REQUEST_STOP_PROCESSING;
634 * Move the CoA request to its own handler.
637 coa_separate(request->coa);
638 } else if (request->parent && (request->parent->coa == request)) {
639 coa_separate(request);
644 * It doesn't hurt to send duplicate replies. All other
645 * signals are ignored, as the request will be cleaned up
651 rad_assert(request->listener != NULL);
653 if (request->reply->code != 0) {
654 request->listener->send(request->listener, request);
657 RDEBUG("No reply. Ignoring retransmit");
662 * Mark the request as done.
665 #ifdef HAVE_PTHREAD_H
667 * If the child is still running, leave it alone.
669 if (spawn_flag && (request->child_state <= REQUEST_RUNNING)) {
674 #ifdef DEBUG_STATE_MACHINE
675 if (rad_debug_lvl) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n",
676 request->number, __FUNCTION__,
677 child_state_names[request->child_state],
678 child_state_names[REQUEST_DONE]);
680 request->child_state = REQUEST_DONE;
684 * Called when the child is taking too long to
685 * finish. We've already marked it "please
686 * stop", so we don't complain any more.
688 case FR_ACTION_TIMER:
692 case FR_ACTION_PROXY_REPLY:
693 proxy_reply_too_late(request);
702 * Remove it from the request hash.
704 if (request->in_request_hash) {
705 if (!rbtree_deletebydata(pl, &request->packet)) {
708 request->in_request_hash = false;
713 * Wait for the proxy ID to expire. This allows us to
714 * avoid re-use of proxy IDs for a while.
716 if (request->in_proxy_hash) {
717 rad_assert(request->proxy != NULL);
719 fr_event_now(el, &now);
720 when = request->proxy->timestamp;
723 if (((request->proxy->code == PW_CODE_COA_REQUEST) ||
724 (request->proxy->code == PW_CODE_DISCONNECT_REQUEST)) &&
725 (request->packet->code != request->proxy->code)) {
726 when.tv_sec += request->home_server->coa_mrd;
729 timeradd(&when, request_response_window(request), &when);
732 * We haven't received all responses, AND there's still
733 * time to wait. Do so.
735 if ((request->num_proxied_requests > request->num_proxied_responses) &&
737 (request->home_server->proto != IPPROTO_TCP) &&
739 timercmp(&now, &when, <)) {
740 RDEBUG("Waiting for more responses from the home server");
747 remove_from_proxy_hash(request);
751 #ifdef HAVE_PTHREAD_H
753 * If there's no children, we can mark the request as done.
755 if (!spawn_flag) request->child_state = REQUEST_DONE;
759 * If the child is still running, wait for it to be finished.
761 if (request->child_state <= REQUEST_RUNNING) {
762 gettimeofday(&now, NULL);
767 if (request->delay < (USEC / 3)) request->delay = USEC / 3;
768 tv_add(&when, request->delay);
769 request->delay += request->delay >> 1;
770 if (request->delay > (10 * USEC)) request->delay = 10 * USEC;
772 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
776 #ifdef HAVE_PTHREAD_H
777 rad_assert(request->child_pid == NO_SUCH_CHILD_PID);
781 * @todo: do final states for TCP sockets, too?
783 request_stats_final(request);
785 if (request->listener) {
786 request->listener->count--;
789 * If we're the last one, remove the listener now.
791 if ((request->listener->count == 0) &&
792 (request->listener->status >= RAD_LISTEN_STATUS_FROZEN)) {
793 event_new_fd(request->listener);
798 if (request->packet) {
799 RDEBUG2("Cleaning up request packet ID %u with timestamp +%d",
801 (unsigned int) (request->timestamp - fr_start_time));
802 } /* else don't print anything */
805 fr_event_delete(el, &request->ev);
806 request_free(request);
810 static void request_cleanup_delay_init(REQUEST *request)
812 struct timeval now, when;
814 VERIFY_REQUEST(request);
817 * Do cleanup delay ONLY for RADIUS packets from a real
818 * client. Everything else just gets cleaned up
821 if (request->packet->dst_port == 0) goto done;
824 * Accounting packets shouldn't be retransmitted. They
825 * should always be updated with Acct-Delay-Time.
827 #ifdef WITH_ACCOUNTING
828 if (request->packet->code == PW_CODE_ACCOUNTING_REQUEST) goto done;
832 if (request->listener->type == RAD_LISTEN_DHCP) goto done;
836 if (request->listener->type == RAD_LISTEN_VQP) goto done;
839 if (!request->root->cleanup_delay) goto done;
841 gettimeofday(&now, NULL);
843 rad_assert(request->reply->timestamp.tv_sec != 0);
844 when = request->reply->timestamp;
846 request->delay = request->root->cleanup_delay;
847 when.tv_sec += request->delay;
850 * Set timer for when we need to clean it up.
852 if (timercmp(&when, &now, >)) {
853 #ifdef DEBUG_STATE_MACHINE
854 if (rad_debug_lvl) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_cleanup_delay");
856 request->process = request_cleanup_delay;
858 if (!we_are_master()) {
859 FINAL_STATE(REQUEST_CLEANUP_DELAY);
864 * Update this if we can, otherwise let the timers pick it up.
866 request->child_state = REQUEST_CLEANUP_DELAY;
867 #ifdef HAVE_PTHREAD_H
868 rad_assert(request->child_pid == NO_SUCH_CHILD_PID);
870 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
875 * Otherwise just clean it up.
878 request_done(request, FR_ACTION_DONE);
883 * Enforce max_request_time.
885 static bool request_max_time(REQUEST *request)
887 struct timeval now, when;
888 rad_assert(request->magic == REQUEST_MAGIC);
889 #ifdef DEBUG_STATE_MACHINE
890 int action = FR_ACTION_TIMER;
893 VERIFY_REQUEST(request);
899 * The child thread has acknowledged it's done.
900 * Transition to the DONE state.
902 * If the request was marked STOP, then the "check for
903 * stop" macro already took care of it.
905 if (request->child_state == REQUEST_DONE) {
907 request_done(request, FR_ACTION_DONE);
912 * The request is still running. Enforce max_request_time.
914 fr_event_now(el, &now);
915 when = request->packet->timestamp;
916 when.tv_sec += request->root->max_request_time;
919 * Taking too long: tell it to die.
921 if (timercmp(&now, &when, >=)) {
922 #ifdef HAVE_PTHREAD_H
924 * If there's a child thread processing it,
928 (pthread_equal(request->child_pid, NO_SUCH_CHILD_PID) == 0)) {
929 ERROR("Unresponsive child for request %u, in component %s module %s",
931 request->component ? request->component : "<core>",
932 request->module ? request->module : "<core>");
933 exec_trigger(request, NULL, "server.thread.unresponsive", true);
937 * Tell the request that it's done.
943 * Sleep for some more. We HOPE that the child will
944 * become responsive at some point in the future. We do
945 * this by adding 50% to the current timer.
948 tv_add(&when, request->delay);
949 request->delay += request->delay >> 1;
950 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
954 static void request_queue_or_run(REQUEST *request,
955 fr_request_process_t process)
957 #ifdef DEBUG_STATE_MACHINE
958 int action = FR_ACTION_TIMER;
961 VERIFY_REQUEST(request);
966 * Do this here so that fewer other functions need to do
969 if (request->master_state == REQUEST_STOP_PROCESSING) {
970 #ifdef DEBUG_STATE_MACHINE
971 if (rad_debug_lvl) printf("(%u) ********\tSTATE %s M-%s causes C-%s-> C-%s\t********\n",
972 request->number, __FUNCTION__,
973 master_state_names[request->master_state],
974 child_state_names[request->child_state],
975 child_state_names[REQUEST_DONE]);
977 request_done(request, FR_ACTION_DONE);
981 request->process = process;
983 if (we_are_master()) {
987 * (re) set the initial delay.
989 request->delay = request_init_delay(request);
990 if (request->delay > USEC) request->delay = USEC;
991 gettimeofday(&when, NULL);
992 tv_add(&when, request->delay);
993 request->delay += request->delay >> 1;
995 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
997 #ifdef HAVE_PTHREAD_H
1000 * A child thread will eventually pick it up.
1002 if (request_enqueue(request)) return;
1005 * Otherwise we're not going to do anything with
1008 request_done(request, FR_ACTION_DONE);
1014 request->child_state = REQUEST_RUNNING;
1015 request->process(request, FR_ACTION_RUN);
1019 * Requests that care about child process exit
1020 * codes have already either called
1021 * rad_waitpid(), or they've given up.
1023 while (waitpid(-1, NULL, WNOHANG) > 0);
1028 static void request_dup(REQUEST *request)
1030 ERROR("(%u) Ignoring duplicate packet from "
1031 "client %s port %d - ID: %u due to unfinished request "
1032 "in component %s module %s",
1033 request->number, request->client->shortname,
1034 request->packet->src_port,request->packet->id,
1035 request->component, request->module);
1039 /** Sit on a request until it's time to clean it up.
1041 * A NAS may not see a response from the server. When the NAS
1042 * retransmits, we want to be able to send a cached reply back. The
1043 * alternative is to re-process the packet, which does bad things for
1044 * EAP, among others.
1046 * IF we do see a NAS retransmit, we extend the cleanup delay,
1047 * because the NAS might miss our cached reply.
1049 * Otherwise, once we reach cleanup_delay, we transition to DONE.
1052 * digraph cleanup_delay {
1054 * send_reply [ label = "send_reply\nincrease cleanup delay" ];
1056 * cleanup_delay -> send_reply [ label = "DUP" ];
1057 * send_reply -> cleanup_delay;
1058 * cleanup_delay -> proxy_reply_too_late [ label = "PROXY_REPLY", arrowhead = "none" ];
1059 * cleanup_delay -> cleanup_delay [ label = "TIMER < timeout" ];
1060 * cleanup_delay -> done [ label = "TIMER >= timeout" ];
1064 static void request_cleanup_delay(REQUEST *request, int action)
1066 struct timeval when, now;
1068 VERIFY_REQUEST(request);
1070 TRACE_STATE_MACHINE;
1077 if (request->reply->code != 0) {
1078 request->listener->send(request->listener, request);
1080 RDEBUG("No reply. Ignoring retransmit");
1084 * Double the cleanup_delay to catch retransmits.
1086 when = request->reply->timestamp;
1087 request->delay += request->delay;
1088 when.tv_sec += request->delay;
1090 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
1094 case FR_ACTION_PROXY_REPLY:
1095 proxy_reply_too_late(request);
1099 case FR_ACTION_TIMER:
1100 fr_event_now(el, &now);
1102 rad_assert(request->root->cleanup_delay > 0);
1104 when = request->reply->timestamp;
1105 when.tv_sec += request->root->cleanup_delay;
1107 if (timercmp(&when, &now, >)) {
1108 #ifdef DEBUG_STATE_MACHINE
1109 if (rad_debug_lvl) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_cleanup_delay");
1111 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
1113 } /* else it's time to clean up */
1115 request_done(request, REQUEST_DONE);
1119 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
1125 /** Sit on a request until it's time to respond to it.
1127 * For security reasons, rejects (and maybe some other) packets are
1128 * delayed for a while before we respond. This delay means that
1129 * badly behaved NASes don't hammer the server with authentication
1132 * Otherwise, once we reach response_delay, we send the reply, and
1133 * transition to cleanup_delay.
1136 * digraph response_delay {
1137 * response_delay -> proxy_reply_too_late [ label = "PROXY_REPLY", arrowhead = "none" ];
1138 * response_delay -> response_delay [ label = "DUP, TIMER < timeout" ];
1139 * response_delay -> send_reply [ label = "TIMER >= timeout" ];
1140 * send_reply -> cleanup_delay;
1144 static void request_response_delay(REQUEST *request, int action)
1146 struct timeval when, now;
1148 VERIFY_REQUEST(request);
1150 TRACE_STATE_MACHINE;
1157 ERROR("(%u) Discarding duplicate request from "
1158 "client %s port %d - ID: %u due to delayed response",
1159 request->number, request->client->shortname,
1160 request->packet->src_port,request->packet->id);
1164 case FR_ACTION_PROXY_REPLY:
1165 proxy_reply_too_late(request);
1169 case FR_ACTION_TIMER:
1170 fr_event_now(el, &now);
1173 * See if it's time to send the reply. If not,
1174 * we wait some more.
1176 when = request->reply->timestamp;
1178 tv_add(&when, request->response_delay.tv_sec * USEC);
1179 tv_add(&when, request->response_delay.tv_usec);
1181 if (timercmp(&when, &now, >)) {
1182 #ifdef DEBUG_STATE_MACHINE
1183 if (rad_debug_lvl) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_response_delay");
1185 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
1187 } /* else it's time to send the reject */
1189 RDEBUG2("Sending delayed response");
1190 debug_packet(request, request->reply, false);
1191 request->listener->send(request->listener, request);
1194 * Clean up the request.
1196 request_cleanup_delay_init(request);
1200 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
1206 static int request_pre_handler(REQUEST *request, UNUSED int action)
1210 VERIFY_REQUEST(request);
1212 TRACE_STATE_MACHINE;
1214 if (request->master_state == REQUEST_STOP_PROCESSING) return 0;
1217 * Don't decode the packet if it's an internal "fake"
1218 * request. Instead, just return so that the caller can
1221 if (request->packet->dst_port == 0) {
1222 request->username = fr_pair_find_by_num(request->packet->vps, PW_USER_NAME, 0, TAG_ANY);
1223 request->password = fr_pair_find_by_num(request->packet->vps, PW_USER_PASSWORD, 0, TAG_ANY);
1227 if (!request->packet->vps) { /* FIXME: check for correct state */
1228 rcode = request->listener->decode(request->listener, request);
1231 if (debug_condition) {
1233 * Ignore parse errors.
1235 if (radius_evaluate_cond(request, RLM_MODULE_OK, 0, debug_condition)) {
1236 request->log.lvl = L_DBG_LVL_2;
1237 request->log.func = vradlog_request;
1242 debug_packet(request, request->packet, true);
1248 RATE_LIMIT(INFO("Dropping packet without response because of error: %s", fr_strerror()));
1249 request->reply->offset = -2; /* bad authenticator */
1253 if (!request->username) {
1254 request->username = fr_pair_find_by_num(request->packet->vps, PW_USER_NAME, 0, TAG_ANY);
1261 /** Do the final processing of a request before we reply to the NAS.
1263 * Various cleanups, suppress responses, copy Proxy-State, and set
1264 * response_delay or cleanup_delay;
1266 static void request_finish(REQUEST *request, int action)
1270 VERIFY_REQUEST(request);
1272 TRACE_STATE_MACHINE;
1275 (void) action; /* -Wunused */
1279 * Don't do post-auth if we're a CoA request originated
1280 * from an Access-Request. See request_alloc_coa() for
1283 if ((request->options & RAD_REQUEST_OPTION_COA) != 0) goto done;
1287 * Override the response code if a control:Response-Packet-Type attribute is present.
1289 vp = fr_pair_find_by_num(request->config, PW_RESPONSE_PACKET_TYPE, 0, TAG_ANY);
1291 if (vp->vp_integer == 256) {
1292 RDEBUG2("Not responding to request");
1293 request->reply->code = 0;
1295 request->reply->code = vp->vp_integer;
1299 * Catch Auth-Type := Reject BEFORE proxying the packet.
1301 else if (request->packet->code == PW_CODE_ACCESS_REQUEST) {
1302 if (request->reply->code == 0) {
1303 vp = fr_pair_find_by_num(request->config, PW_AUTH_TYPE, 0, TAG_ANY);
1304 if (!vp || (vp->vp_integer != 5)) {
1305 RDEBUG2("There was no response configured: "
1306 "rejecting request");
1309 request->reply->code = PW_CODE_ACCESS_REJECT;
1314 * Copy Proxy-State from the request to the reply.
1316 vp = fr_pair_list_copy_by_num(request->reply, request->packet->vps,
1317 PW_PROXY_STATE, 0, TAG_ANY);
1318 if (vp) fr_pair_add(&request->reply->vps, vp);
1321 * Call Post-Auth for Access-Request packets.
1323 if (request->packet->code == PW_CODE_ACCESS_REQUEST) {
1324 rad_postauth(request);
1329 * Maybe originate a CoA request.
1331 if ((action == FR_ACTION_RUN) && !request->proxy && request->coa) {
1332 request_coa_originate(request);
1337 * Clean up. These are no longer needed.
1339 gettimeofday(&request->reply->timestamp, NULL);
1342 * Fake packets get marked as "done", and have the
1343 * proxy-reply section deal with the reply attributes.
1344 * We therefore don't free the reply attributes.
1346 if (request->packet->dst_port == 0) {
1347 RDEBUG("Finished internally proxied request.");
1348 FINAL_STATE(REQUEST_DONE);
1354 * Always send the reply to the detail listener.
1356 if (request->listener->type == RAD_LISTEN_DETAIL) {
1357 request->simul_max = 1;
1360 * But only print the reply if there is one.
1362 if (request->reply->code != 0) {
1363 debug_packet(request, request->reply, false);
1366 request->listener->send(request->listener, request);
1372 * Ignore all "do not respond" packets.
1373 * Except for the detail ones, which need to ping
1374 * the detail file reader so that it will retransmit.
1376 if (!request->reply->code) {
1377 RDEBUG("Not sending reply to client.");
1382 * If it's not in the request hash, we MIGHT not want to
1385 * If duplicate packets are allowed, then then only
1386 * reason to NOT be in the request hash is because we
1387 * don't want to send a reply.
1389 * FIXME: this is crap. The rest of the state handling
1390 * should use a different field so that we don't have two
1393 * Otherwise duplicates are forbidden, and the request is
1394 * SUPPOSED to avoid the request hash.
1396 * In that case, we need to send a reply.
1398 if (!request->in_request_hash &&
1399 !request->listener->nodup) {
1400 RDEBUG("Suppressing reply to client.");
1405 * See if we need to delay an Access-Reject packet.
1407 if ((request->reply->code == PW_CODE_ACCESS_REJECT) &&
1408 (request->root->reject_delay.tv_sec > 0)) {
1409 request->response_delay = request->root->reject_delay;
1411 vp = fr_pair_find_by_num(request->reply->vps, PW_FREERADIUS_RESPONSE_DELAY, 0, TAG_ANY);
1413 if (vp->vp_integer <= 10) {
1414 request->response_delay.tv_sec = vp->vp_integer;
1416 request->response_delay.tv_sec = 10;
1418 request->response_delay.tv_usec = 0;
1420 vp = fr_pair_find_by_num(request->reply->vps, PW_FREERADIUS_RESPONSE_DELAY_USEC, 0, TAG_ANY);
1422 if (vp->vp_integer <= 10 * USEC) {
1423 request->response_delay.tv_sec = vp->vp_integer / USEC;
1424 request->response_delay.tv_usec = vp->vp_integer % USEC;
1426 request->response_delay.tv_sec = 10;
1427 request->response_delay.tv_usec = 0;
1434 * If we timed out a proxy packet, don't delay
1435 * the reject any more.
1437 if (request->proxy && !request->proxy_reply) {
1438 request->response_delay.tv_sec = 0;
1439 request->response_delay.tv_usec = 0;
1447 if ((request->response_delay.tv_sec == 0) &&
1448 (request->response_delay.tv_usec == 0)) {
1451 * Don't print a reply if there's none to send.
1453 if (request->reply->code != 0) {
1454 if (rad_debug_lvl && request->state &&
1455 (request->reply->code == PW_CODE_ACCESS_ACCEPT)) {
1456 if (!fr_pair_find_by_num(request->packet->vps, PW_STATE, 0, TAG_ANY)) {
1457 RWDEBUG2("Unused attributes found in &session-state:");
1461 debug_packet(request, request->reply, false);
1462 request->listener->send(request->listener, request);
1466 RDEBUG2("Finished request");
1467 request_cleanup_delay_init(request);
1471 * Encode and sign it here, so that the master
1472 * thread can just send the encoded data, which
1473 * means it does less work.
1475 RDEBUG2("Delaying response for %d.%06d seconds",
1476 (int) request->response_delay.tv_sec, (int) request->response_delay.tv_usec);
1477 request->listener->encode(request->listener, request);
1478 request->process = request_response_delay;
1480 FINAL_STATE(REQUEST_RESPONSE_DELAY);
1484 /** Process a request from a client.
1486 * The outcome might be that the request is proxied.
1490 * running -> running [ label = "TIMER < max_request_time" ];
1491 * running -> done [ label = "TIMER >= max_request_time" ];
1492 * running -> proxy [ label = "proxied" ];
1493 * running -> dup [ label = "DUP", arrowhead = "none" ];
1497 static void request_running(REQUEST *request, int action)
1499 VERIFY_REQUEST(request);
1501 TRACE_STATE_MACHINE;
1505 case FR_ACTION_TIMER:
1507 (void) request_max_time(request);
1511 request_dup(request);
1515 if (!request_pre_handler(request, action)) {
1516 #ifdef DEBUG_STATE_MACHINE
1517 if (rad_debug_lvl) printf("(%u) ********\tSTATE %s failed in pre-handler C-%s -> C-%s\t********\n",
1518 request->number, __FUNCTION__,
1519 child_state_names[request->child_state],
1520 child_state_names[REQUEST_DONE]);
1522 FINAL_STATE(REQUEST_DONE);
1526 rad_assert(request->handle != NULL);
1527 request->handle(request);
1531 * We may need to send a proxied request.
1533 if ((action == FR_ACTION_RUN) &&
1534 request_will_proxy(request)) {
1535 #ifdef DEBUG_STATE_MACHINE
1536 if (rad_debug_lvl) printf("(%u) ********\tWill Proxy\t********\n", request->number);
1540 * takes care of setting
1541 * up the post proxy fail
1544 if (request_proxy(request, 0) < 0) goto req_finished;
1548 #ifdef DEBUG_STATE_MACHINE
1549 if (rad_debug_lvl) printf("(%u) ********\tFinished\t********\n", request->number);
1555 request_finish(request, action);
1560 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
1565 int request_receive(TALLOC_CTX *ctx, rad_listen_t *listener, RADIUS_PACKET *packet,
1566 RADCLIENT *client, RAD_REQUEST_FUNP fun)
1569 RADIUS_PACKET **packet_p;
1570 REQUEST *request = NULL;
1572 listen_socket_t *sock = NULL;
1574 VERIFY_PACKET(packet);
1577 * Set the last packet received.
1579 gettimeofday(&now, NULL);
1581 packet->timestamp = now;
1583 #ifdef WITH_ACCOUNTING
1584 if (listener->type != RAD_LISTEN_DETAIL)
1587 sock = listener->data;
1588 sock->last_packet = now.tv_sec;
1591 packet->proto = sock->proto;
1596 * Skip everything if required.
1598 if (listener->nodup) goto skip_dup;
1600 packet_p = rbtree_finddata(pl, &packet);
1602 rad_child_state_t child_state;
1604 request = fr_packet2myptr(REQUEST, packet, packet_p);
1605 rad_assert(request->in_request_hash);
1606 child_state = request->child_state;
1609 * Same src/dst ip/port, length, and
1610 * authentication vector: must be a duplicate.
1612 if ((request->packet->data_len == packet->data_len) &&
1613 (memcmp(request->packet->vector, packet->vector,
1614 sizeof(packet->vector)) == 0)) {
1617 switch (packet->code) {
1618 case PW_CODE_ACCESS_REQUEST:
1619 FR_STATS_INC(auth, total_dup_requests);
1622 #ifdef WITH_ACCOUNTING
1623 case PW_CODE_ACCOUNTING_REQUEST:
1624 FR_STATS_INC(acct, total_dup_requests);
1628 case PW_CODE_COA_REQUEST:
1629 FR_STATS_INC(coa, total_dup_requests);
1632 case PW_CODE_DISCONNECT_REQUEST:
1633 FR_STATS_INC(dsc, total_dup_requests);
1640 #endif /* WITH_STATS */
1643 * Tell the state machine that there's a
1644 * duplicate request.
1646 request->process(request, FR_ACTION_DUP);
1647 return 0; /* duplicate of live request */
1651 * Mark the request as done ASAP, and before we
1652 * log anything. The child may stop processing
1653 * the request just as we're logging the
1656 request_done(request, FR_ACTION_DONE);
1660 * It's a new request, not a duplicate. If the
1661 * old one is done, then we can clean it up.
1663 if (child_state <= REQUEST_RUNNING) {
1665 * The request is still QUEUED or RUNNING. That's a problem.
1667 ERROR("Received conflicting packet from "
1668 "client %s port %d - ID: %u due to "
1669 "unfinished request. Giving up on old request.",
1671 packet->src_port, packet->id);
1675 * Mark the old request as done. If there's no
1676 * child, the request will be cleaned up
1677 * immediately. If there is a child, we'll set a
1678 * timer to go clean up the request.
1680 } /* else the new packet is unique */
1683 * Quench maximum number of outstanding requests.
1685 if (main_config.max_requests &&
1686 ((count = rbtree_num_elements(pl)) > main_config.max_requests)) {
1687 RATE_LIMIT(ERROR("Dropping request (%d is too many): from client %s port %d - ID: %d", count,
1689 packet->src_port, packet->id);
1690 WARN("Please check the configuration file.\n"
1691 "\tThe value for 'max_requests' is probably set too low.\n"));
1693 exec_trigger(NULL, NULL, "server.max_requests", true);
1699 * Rate-limit the incoming packets
1701 if (sock && sock->max_rate) {
1704 pps = rad_pps(&sock->rate_pps_old, &sock->rate_pps_now, &sock->rate_time, &now);
1705 if (pps > sock->max_rate) {
1706 DEBUG("Dropping request due to rate limiting");
1709 sock->rate_pps_now++;
1713 * Allocate a pool for the request.
1716 ctx = talloc_pool(NULL, main_config.talloc_pool_size);
1718 talloc_set_name_const(ctx, "request_receive_pool");
1721 * The packet is still allocated from a different
1722 * context, but oh well.
1724 (void) talloc_steal(ctx, packet);
1727 request = request_setup(ctx, listener, packet, client, fun);
1734 * Mark it as a "real" request with a context.
1736 request->options |= RAD_REQUEST_OPTION_CTX;
1739 * Remember the request in the list.
1741 if (!listener->nodup) {
1742 if (!rbtree_insert(pl, &request->packet)) {
1743 RERROR("Failed to insert request in the list of live requests: discarding it");
1744 request_done(request, FR_ACTION_DONE);
1748 request->in_request_hash = true;
1752 * Process it. Send a response, and free it.
1754 if (listener->synchronous) {
1756 rad_assert(listener->type != RAD_LISTEN_DETAIL);
1759 request->listener->decode(request->listener, request);
1760 request->username = fr_pair_find_by_num(request->packet->vps, PW_USER_NAME, 0, TAG_ANY);
1761 request->password = fr_pair_find_by_num(request->packet->vps, PW_USER_PASSWORD, 0, TAG_ANY);
1765 if (request->reply->code != 0) {
1766 request->listener->send(request->listener, request);
1768 RDEBUG("Not sending reply");
1772 * Don't do delayed reject. Oh well.
1774 request_free(request);
1779 * Otherwise, insert it into the state machine.
1780 * The child threads will take care of processing it.
1782 request_queue_or_run(request, request_running);
1788 static REQUEST *request_setup(TALLOC_CTX *ctx, rad_listen_t *listener, RADIUS_PACKET *packet,
1789 RADCLIENT *client, RAD_REQUEST_FUNP fun)
1794 * Create and initialize the new request.
1796 request = request_alloc(ctx);
1801 request->reply = rad_alloc(request, false);
1802 if (!request->reply) {
1804 talloc_free(request);
1808 request->listener = listener;
1809 request->client = client;
1810 request->packet = talloc_steal(request, packet);
1811 request->number = request_num_counter++;
1812 request->priority = listener->type;
1813 request->master_state = REQUEST_ACTIVE;
1814 request->child_state = REQUEST_RUNNING;
1815 #ifdef DEBUG_STATE_MACHINE
1816 if (rad_debug_lvl) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n",
1817 request->number, __FUNCTION__,
1818 child_state_names[request->child_state],
1819 child_state_names[REQUEST_RUNNING]);
1821 request->handle = fun;
1825 request->listener->stats.last_packet = request->packet->timestamp.tv_sec;
1826 if (packet->code == PW_CODE_ACCESS_REQUEST) {
1827 request->client->auth.last_packet = request->packet->timestamp.tv_sec;
1828 radius_auth_stats.last_packet = request->packet->timestamp.tv_sec;
1829 #ifdef WITH_ACCOUNTING
1830 } else if (packet->code == PW_CODE_ACCOUNTING_REQUEST) {
1831 request->client->acct.last_packet = request->packet->timestamp.tv_sec;
1832 radius_acct_stats.last_packet = request->packet->timestamp.tv_sec;
1835 #endif /* WITH_STATS */
1838 * Status-Server packets go to the head of the queue.
1840 if (request->packet->code == PW_CODE_STATUS_SERVER) request->priority = 0;
1843 * Set virtual server identity
1845 if (client->server) {
1846 request->server = client->server;
1847 } else if (listener->server) {
1848 request->server = listener->server;
1850 request->server = NULL;
1853 request->root = &main_config;
1855 request->listener->count++;
1859 * The request passes many of our sanity checks.
1860 * From here on in, if anything goes wrong, we
1861 * send a reject message, instead of dropping the
1866 * Build the reply template from the request.
1869 request->reply->sockfd = request->packet->sockfd;
1870 request->reply->dst_ipaddr = request->packet->src_ipaddr;
1871 request->reply->src_ipaddr = request->packet->dst_ipaddr;
1872 request->reply->dst_port = request->packet->src_port;
1873 request->reply->src_port = request->packet->dst_port;
1874 request->reply->id = request->packet->id;
1875 request->reply->code = 0; /* UNKNOWN code */
1876 memcpy(request->reply->vector, request->packet->vector,
1877 sizeof(request->reply->vector));
1878 request->reply->vps = NULL;
1879 request->reply->data = NULL;
1880 request->reply->data_len = 0;
1886 /***********************************************************************
1890 ***********************************************************************/
1893 * Timer function for all TCP sockets.
1895 static void tcp_socket_timer(void *ctx)
1897 rad_listen_t *listener = talloc_get_type_abort(ctx, rad_listen_t);
1898 listen_socket_t *sock = listener->data;
1899 struct timeval end, now;
1901 fr_socket_limit_t *limit;
1905 if (listener->status != RAD_LISTEN_STATUS_KNOWN) return;
1907 fr_event_now(el, &now);
1909 switch (listener->type) {
1911 case RAD_LISTEN_PROXY:
1912 limit = &sock->home->limit;
1916 case RAD_LISTEN_AUTH:
1917 #ifdef WITH_ACCOUNTING
1918 case RAD_LISTEN_ACCT:
1920 limit = &sock->limit;
1928 * If we enforce a lifetime, do it now.
1930 if (limit->lifetime > 0) {
1931 end.tv_sec = sock->opened + limit->lifetime;
1934 if (timercmp(&end, &now, <=)) {
1935 listener->print(listener, buffer, sizeof(buffer));
1936 DEBUG("Reached maximum lifetime on socket %s", buffer);
1942 * Proxy sockets get frozen, so that we don't use
1943 * them for new requests. But we do keep them
1944 * open to listen for replies to requests we had
1947 if (listener->type == RAD_LISTEN_PROXY) {
1948 PTHREAD_MUTEX_LOCK(&proxy_mutex);
1949 if (!fr_packet_list_socket_freeze(proxy_list,
1951 ERROR("Fatal error freezing socket: %s", fr_strerror());
1954 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
1959 * Mark the socket as "don't use if at all possible".
1961 listener->status = RAD_LISTEN_STATUS_FROZEN;
1962 event_new_fd(listener);
1971 * Enforce an idle timeout.
1973 if (limit->idle_timeout > 0) {
1974 struct timeval idle;
1976 rad_assert(sock->last_packet != 0);
1977 idle.tv_sec = sock->last_packet + limit->idle_timeout;
1980 if (timercmp(&idle, &now, <=)) {
1981 listener->print(listener, buffer, sizeof(buffer));
1982 DEBUG("Reached idle timeout on socket %s", buffer);
1987 * Enforce the minimum of idle timeout or lifetime.
1989 if (timercmp(&idle, &end, <)) {
1995 * Wake up at t + 0.5s. The code above checks if the timers
1996 * are <= t. This addition gives us a bit of leeway.
1998 end.tv_usec = USEC / 2;
2001 if (!fr_event_insert(el, tcp_socket_timer, listener, &end, &sock->ev)) {
2002 rad_panic("Failed to insert event");
2009 * Add +/- 2s of jitter, as suggested in RFC 3539
2012 static void add_jitter(struct timeval *when)
2019 jitter ^= (jitter >> 10);
2020 jitter &= ((1 << 22) - 1); /* 22 bits of 1 */
2023 * Add in ~ (4 * USEC) of jitter.
2025 tv_add(when, jitter);
2029 * Called by socket_del to remove requests with this socket
2031 static int eol_proxy_listener(void *ctx, void *data)
2033 rad_listen_t *this = talloc_get_type_abort(ctx, rad_listen_t);
2034 RADIUS_PACKET **proxy_p = data;
2037 request = fr_packet2myptr(REQUEST, proxy, proxy_p);
2038 if (request->proxy_listener != this) return 0;
2041 * The normal "remove_from_proxy_hash" tries to grab the
2042 * proxy mutex. We already have it held, so grabbing it
2043 * again will cause a deadlock. Instead, call the "no
2044 * lock" version of the function.
2046 rad_assert(request->in_proxy_hash == true);
2047 remove_from_proxy_hash_nl(request, false);
2050 * Don't mark it as DONE. The client can retransmit, and
2051 * the packet SHOULD be re-proxied somewhere else.
2053 * Return "2" means that the rbtree code will remove it
2054 * from the tree, and we don't need to do it ourselves.
2058 #endif /* WITH_PROXY */
2060 static int eol_listener(void *ctx, void *data)
2062 rad_listen_t *this = talloc_get_type_abort(ctx, rad_listen_t);
2063 RADIUS_PACKET **packet_p = data;
2066 request = fr_packet2myptr(REQUEST, packet, packet_p);
2067 if (request->listener != this) return 0;
2069 request->master_state = REQUEST_STOP_PROCESSING;
2070 request->process = request_done;
2074 #endif /* WITH_TCP */
2077 /***********************************************************************
2079 * Proxy handlers for the state machine.
2081 ***********************************************************************/
2084 * Called with the proxy mutex held
2086 static void remove_from_proxy_hash_nl(REQUEST *request, bool yank)
2088 VERIFY_REQUEST(request);
2090 if (!request->in_proxy_hash) return;
2092 fr_packet_list_id_free(proxy_list, request->proxy, yank);
2093 request->in_proxy_hash = false;
2096 * On the FIRST reply, decrement the count of outstanding
2097 * requests. Note that this is NOT the count of sent
2098 * packets, but whether or not the home server has
2101 if (request->home_server &&
2102 request->home_server->currently_outstanding) {
2103 request->home_server->currently_outstanding--;
2106 * If we're NOT sending it packets, AND it's been
2107 * a while since we got a response, then we don't
2108 * know if it's alive or dead.
2110 if ((request->home_server->currently_outstanding == 0) &&
2111 (request->home_server->state == HOME_STATE_ALIVE)) {
2112 struct timeval when, now;
2114 when.tv_sec = request->home_server->last_packet_recv ;
2117 timeradd(&when, request_response_window(request), &when);
2118 gettimeofday(&now, NULL);
2121 * last_packet + response_window
2123 * We *administratively* mark the home
2124 * server as "unknown" state, because we
2125 * haven't seen a packet for a while.
2127 if (timercmp(&now, &when, >)) {
2128 request->home_server->state = HOME_STATE_UNKNOWN;
2129 request->home_server->last_packet_sent = 0;
2130 request->home_server->last_packet_recv = 0;
2136 rad_assert(request->proxy_listener != NULL);
2137 request->proxy_listener->count--;
2139 request->proxy_listener = NULL;
2142 * Got from YES in hash, to NO, not in hash while we hold
2143 * the mutex. This guarantees that when another thread
2144 * grabs the mutex, the "not in hash" flag is correct.
2148 static void remove_from_proxy_hash(REQUEST *request)
2150 VERIFY_REQUEST(request);
2153 * Check this without grabbing the mutex because it's a
2154 * lot faster that way.
2156 if (!request->in_proxy_hash) return;
2159 * The "not in hash" flag is definitive. However, if the
2160 * flag says that it IS in the hash, there might still be
2161 * a race condition where it isn't.
2163 PTHREAD_MUTEX_LOCK(&proxy_mutex);
2165 if (!request->in_proxy_hash) {
2166 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
2170 remove_from_proxy_hash_nl(request, true);
2172 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
2175 static int insert_into_proxy_hash(REQUEST *request)
2179 bool success = false;
2180 void *proxy_listener;
2182 VERIFY_REQUEST(request);
2184 rad_assert(request->proxy != NULL);
2185 rad_assert(request->home_server != NULL);
2186 rad_assert(proxy_list != NULL);
2189 PTHREAD_MUTEX_LOCK(&proxy_mutex);
2190 proxy_listener = NULL;
2191 request->num_proxied_requests = 1;
2192 request->num_proxied_responses = 0;
2194 for (tries = 0; tries < 2; tries++) {
2196 listen_socket_t *sock;
2198 RDEBUG3("proxy: Trying to allocate ID (%d/2)", tries);
2199 success = fr_packet_list_id_alloc(proxy_list,
2200 request->home_server->proto,
2201 &request->proxy, &proxy_listener);
2204 if (tries > 0) continue; /* try opening new socket only once */
2206 #ifdef HAVE_PTHREAD_H
2207 if (proxy_no_new_sockets) break;
2210 RDEBUG3("proxy: Trying to open a new listener to the home server");
2211 this = proxy_new_listener(proxy_ctx, request->home_server, 0);
2213 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
2217 request->proxy->src_port = 0; /* Use any new socket */
2218 proxy_listener = this;
2221 if (!fr_packet_list_socket_add(proxy_list, this->fd,
2223 &sock->other_ipaddr, sock->other_port,
2226 #ifdef HAVE_PTHREAD_H
2227 proxy_no_new_sockets = true;
2229 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
2232 * This is bad. However, the
2233 * packet list now supports 256
2234 * open sockets, which should
2235 * minimize this problem.
2237 ERROR("Failed adding proxy socket: %s",
2243 * Add it to the event loop. Ensure that we have
2244 * only one mutex locked at a time.
2246 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
2247 radius_update_listener(this);
2248 PTHREAD_MUTEX_LOCK(&proxy_mutex);
2251 if (!proxy_listener || !success) {
2252 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
2253 REDEBUG2("proxy: Failed allocating Id for proxied request");
2255 request->proxy_listener = NULL;
2256 request->in_proxy_hash = false;
2260 rad_assert(request->proxy->id >= 0);
2262 request->proxy_listener = proxy_listener;
2263 request->in_proxy_hash = true;
2264 RDEBUG3("proxy: request is now in proxy hash");
2267 * Keep track of maximum outstanding requests to a
2268 * particular home server. 'max_outstanding' is
2269 * enforced in home_server_ldb(), in realms.c.
2271 request->home_server->currently_outstanding++;
2274 request->proxy_listener->count++;
2277 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
2279 RDEBUG3("proxy: allocating destination %s port %d - Id %d",
2280 inet_ntop(request->proxy->dst_ipaddr.af,
2281 &request->proxy->dst_ipaddr.ipaddr, buf, sizeof(buf)),
2282 request->proxy->dst_port,
2283 request->proxy->id);
2288 static int process_proxy_reply(REQUEST *request, RADIUS_PACKET *reply)
2291 int post_proxy_type = 0;
2294 VERIFY_REQUEST(request);
2297 * There may be a proxy reply, but it may be too late.
2299 if (!request->home_server->server && !request->proxy_listener) return 0;
2302 * Delete any reply we had accumulated until now.
2304 RDEBUG2("Clearing existing &reply: attributes");
2305 fr_pair_list_free(&request->reply->vps);
2308 * Run the packet through the post-proxy stage,
2309 * BEFORE playing games with the attributes.
2311 vp = fr_pair_find_by_num(request->config, PW_POST_PROXY_TYPE, 0, TAG_ANY);
2313 post_proxy_type = vp->vp_integer;
2315 * If we have a proxy_reply, and it was a reject, or a NAK
2316 * setup Post-Proxy <type>.
2318 * If the <type> doesn't have a section, then the Post-Proxy
2319 * section is ignored.
2322 DICT_VALUE *dval = NULL;
2324 switch (reply->code) {
2325 case PW_CODE_ACCESS_REJECT:
2326 dval = dict_valbyname(PW_POST_PROXY_TYPE, 0, "Reject");
2327 if (dval) post_proxy_type = dval->value;
2330 case PW_CODE_DISCONNECT_NAK:
2331 dval = dict_valbyname(PW_POST_PROXY_TYPE, 0, fr_packet_codes[reply->code]);
2332 if (dval) post_proxy_type = dval->value;
2335 case PW_CODE_COA_NAK:
2336 dval = dict_valbyname(PW_POST_PROXY_TYPE, 0, fr_packet_codes[reply->code]);
2337 if (dval) post_proxy_type = dval->value;
2345 * Create config:Post-Proxy-Type
2348 vp = radius_pair_create(request, &request->config, PW_POST_PROXY_TYPE, 0);
2349 vp->vp_integer = dval->value;
2353 if (post_proxy_type > 0) RDEBUG2("Found Post-Proxy-Type %s",
2354 dict_valnamebyattr(PW_POST_PROXY_TYPE, 0, post_proxy_type));
2357 VERIFY_PACKET(reply);
2360 * Decode the packet if required.
2362 if (request->proxy_listener) {
2363 rcode = request->proxy_listener->decode(request->proxy_listener, request);
2364 debug_packet(request, reply, true);
2367 * Pro-actively remove it from the proxy hash.
2368 * This is later than in 2.1.x, but it means that
2369 * the replies are authenticated before being
2370 * removed from the hash.
2373 (request->num_proxied_requests <= request->num_proxied_responses)) {
2374 remove_from_proxy_hash(request);
2377 rad_assert(!request->in_proxy_hash);
2379 } else if (request->in_proxy_hash) {
2380 remove_from_proxy_hash(request);
2383 if (request->home_pool && request->home_pool->virtual_server) {
2384 char const *old_server = request->server;
2386 request->server = request->home_pool->virtual_server;
2387 RDEBUG2("server %s {", request->server);
2389 rcode = process_post_proxy(post_proxy_type, request);
2392 request->server = old_server;
2394 rcode = process_post_proxy(post_proxy_type, request);
2398 if (request->packet->code == request->proxy->code)
2400 * Don't run the next bit if we originated a CoA
2401 * packet, after receiving an Access-Request or
2402 * Accounting-Request.
2407 * There may NOT be a proxy reply, as we may be
2408 * running Post-Proxy-Type = Fail.
2411 fr_pair_add(&request->reply->vps, fr_pair_list_copy(request->reply, reply->vps));
2414 * Delete the Proxy-State Attributes from
2415 * the reply. These include Proxy-State
2416 * attributes from us and remote server.
2418 fr_pair_delete_by_num(&request->reply->vps, PW_PROXY_STATE, 0, TAG_ANY);
2422 default: /* Don't do anything */
2424 case RLM_MODULE_FAIL:
2427 case RLM_MODULE_HANDLED:
2434 static void mark_home_server_alive(REQUEST *request, home_server_t *home)
2438 home->state = HOME_STATE_ALIVE;
2439 home->response_timeouts = 0;
2440 exec_trigger(request, home->cs, "home_server.alive", false);
2441 home->currently_outstanding = 0;
2442 home->num_sent_pings = 0;
2443 home->num_received_pings = 0;
2444 gettimeofday(&home->revive_time, NULL);
2446 fr_event_delete(el, &home->ev);
2448 RPROXY("Marking home server %s port %d alive",
2449 inet_ntop(request->proxy->dst_ipaddr.af,
2450 &request->proxy->dst_ipaddr.ipaddr,
2451 buffer, sizeof(buffer)),
2452 request->proxy->dst_port);
2456 int request_proxy_reply(RADIUS_PACKET *packet)
2458 RADIUS_PACKET **proxy_p;
2463 VERIFY_PACKET(packet);
2465 PTHREAD_MUTEX_LOCK(&proxy_mutex);
2466 proxy_p = fr_packet_list_find_byreply(proxy_list, packet);
2469 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
2470 PROXY("No outstanding request was found for %s packet from host %s port %d - ID %u",
2471 fr_packet_codes[packet->code],
2472 inet_ntop(packet->src_ipaddr.af,
2473 &packet->src_ipaddr.ipaddr,
2474 buffer, sizeof(buffer)),
2475 packet->src_port, packet->id);
2479 request = fr_packet2myptr(REQUEST, proxy, proxy_p);
2480 request->num_proxied_responses++; /* needs to be protected by lock */
2482 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
2485 * No reply, BUT the current packet fails verification:
2486 * ignore it. This does the MD5 calculations in the
2487 * server core, but I guess we can fix that later.
2489 if (!request->proxy_reply &&
2490 (rad_verify(packet, request->proxy,
2491 request->home_server->secret) != 0)) {
2492 DEBUG("Ignoring spoofed proxy reply. Signature is invalid");
2497 * The home server sent us a packet which doesn't match
2498 * something we have: ignore it. This is done only to
2499 * catch the case of broken systems.
2501 if (request->proxy_reply &&
2502 (memcmp(request->proxy_reply->vector,
2504 sizeof(request->proxy_reply->vector)) != 0)) {
2505 RDEBUG2("Ignoring conflicting proxy reply");
2509 gettimeofday(&now, NULL);
2512 * Status-Server packets don't count as real packets.
2514 if (request->proxy->code != PW_CODE_STATUS_SERVER) {
2515 listen_socket_t *sock = request->proxy_listener->data;
2517 request->home_server->last_packet_recv = now.tv_sec;
2518 sock->last_packet = now.tv_sec;
2522 * If we have previously seen a reply, ignore the
2525 if (request->proxy_reply) {
2526 RDEBUG2("Discarding duplicate reply from host %s port %d - ID: %d",
2527 inet_ntop(packet->src_ipaddr.af,
2528 &packet->src_ipaddr.ipaddr,
2529 buffer, sizeof(buffer)),
2530 packet->src_port, packet->id);
2535 * Call the state machine to do something useful with the
2538 request->proxy_reply = talloc_steal(request, packet);
2539 packet->timestamp = now;
2540 request->priority = RAD_LISTEN_PROXY;
2544 * Update the proxy listener stats here, because only one
2545 * thread accesses that at a time. The home_server and
2546 * main proxy_*_stats structures are updated once the
2547 * request is cleaned up.
2549 request->proxy_listener->stats.total_responses++;
2551 request->home_server->stats.last_packet = packet->timestamp.tv_sec;
2552 request->proxy_listener->stats.last_packet = packet->timestamp.tv_sec;
2554 switch (request->proxy->code) {
2555 case PW_CODE_ACCESS_REQUEST:
2556 proxy_auth_stats.last_packet = packet->timestamp.tv_sec;
2558 if (request->proxy_reply->code == PW_CODE_ACCESS_ACCEPT) {
2559 request->proxy_listener->stats.total_access_accepts++;
2561 } else if (request->proxy_reply->code == PW_CODE_ACCESS_REJECT) {
2562 request->proxy_listener->stats.total_access_rejects++;
2564 } else if (request->proxy_reply->code == PW_CODE_ACCESS_CHALLENGE) {
2565 request->proxy_listener->stats.total_access_challenges++;
2569 #ifdef WITH_ACCOUNTING
2570 case PW_CODE_ACCOUNTING_REQUEST:
2571 proxy_acct_stats.last_packet = packet->timestamp.tv_sec;
2573 request->proxy_listener->stats.total_responses++;
2574 proxy_acct_stats.last_packet = packet->timestamp.tv_sec;
2580 case PW_CODE_COA_REQUEST:
2581 request->proxy_listener->stats.total_responses++;
2582 proxy_coa_stats.last_packet = packet->timestamp.tv_sec;
2585 case PW_CODE_DISCONNECT_REQUEST:
2586 request->proxy_listener->stats.total_responses++;
2587 proxy_dsc_stats.last_packet = packet->timestamp.tv_sec;
2597 * If we hadn't been sending the home server packets for
2598 * a while, just mark it alive. Or, if it was zombie,
2599 * it's now responded, and is therefore alive.
2601 if ((request->home_server->state == HOME_STATE_UNKNOWN) ||
2602 (request->home_server->state == HOME_STATE_ZOMBIE)) {
2603 mark_home_server_alive(request, request->home_server);
2607 * Tell the request state machine that we have a proxy
2608 * reply. Depending on the function, this should either
2609 * ignore it, or process it.
2611 request->process(request, FR_ACTION_PROXY_REPLY);
2617 static int setup_post_proxy_fail(REQUEST *request)
2619 DICT_VALUE const *dval = NULL;
2622 VERIFY_REQUEST(request);
2624 if (request->proxy->code == PW_CODE_ACCESS_REQUEST) {
2625 dval = dict_valbyname(PW_POST_PROXY_TYPE, 0,
2626 "Fail-Authentication");
2627 #ifdef WITH_ACCOUNTING
2628 } else if (request->proxy->code == PW_CODE_ACCOUNTING_REQUEST) {
2629 dval = dict_valbyname(PW_POST_PROXY_TYPE, 0,
2634 } else if (request->proxy->code == PW_CODE_COA_REQUEST) {
2635 dval = dict_valbyname(PW_POST_PROXY_TYPE, 0, "Fail-CoA");
2637 } else if (request->proxy->code == PW_CODE_DISCONNECT_REQUEST) {
2638 dval = dict_valbyname(PW_POST_PROXY_TYPE, 0, "Fail-Disconnect");
2641 WARN("Unknown packet type in Post-Proxy-Type Fail: ignoring");
2645 if (!dval) dval = dict_valbyname(PW_POST_PROXY_TYPE, 0, "Fail");
2648 fr_pair_delete_by_num(&request->config, PW_POST_PROXY_TYPE, 0, TAG_ANY);
2652 vp = fr_pair_find_by_num(request->config, PW_POST_PROXY_TYPE, 0, TAG_ANY);
2653 if (!vp) vp = radius_pair_create(request, &request->config,
2654 PW_POST_PROXY_TYPE, 0);
2655 vp->vp_integer = dval->value;
2661 /** Process a request after the proxy has timed out.
2663 * Run the packet through Post-Proxy-Type Fail
2666 * digraph proxy_no_reply {
2669 * proxy_no_reply -> dup [ label = "DUP", arrowhead = "none" ];
2670 * proxy_no_reply -> timer [ label = "TIMER < max_request_time" ];
2671 * proxy_no_reply -> proxy_reply_too_late [ label = "PROXY_REPLY" arrowhead = "none"];
2672 * proxy_no_reply -> process_proxy_reply [ label = "RUN" ];
2673 * proxy_no_reply -> done [ label = "TIMER >= timeout" ];
2677 static void proxy_no_reply(REQUEST *request, int action)
2679 VERIFY_REQUEST(request);
2681 TRACE_STATE_MACHINE;
2686 request_dup(request);
2689 case FR_ACTION_TIMER:
2690 (void) request_max_time(request);
2693 case FR_ACTION_PROXY_REPLY:
2694 proxy_reply_too_late(request);
2698 if (process_proxy_reply(request, NULL)) {
2699 request->handle(request);
2701 request_finish(request, action);
2705 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
2710 /** Process the request after receiving a proxy reply.
2712 * Throught the post-proxy section, and the through the handler
2716 * digraph proxy_running {
2719 * proxy_running -> dup [ label = "DUP", arrowhead = "none" ];
2720 * proxy_running -> timer [ label = "TIMER < max_request_time" ];
2721 * proxy_running -> process_proxy_reply [ label = "RUN" ];
2722 * proxy_running -> done [ label = "TIMER >= timeout" ];
2726 static void proxy_running(REQUEST *request, int action)
2728 VERIFY_REQUEST(request);
2730 TRACE_STATE_MACHINE;
2735 request_dup(request);
2738 case FR_ACTION_TIMER:
2739 (void) request_max_time(request);
2743 if (process_proxy_reply(request, request->proxy_reply)) {
2744 request->handle(request);
2746 request_finish(request, action);
2749 default: /* duplicate proxy replies are suppressed */
2750 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
2755 /** Determine if a #REQUEST needs to be proxied, and perform pre-proxy operations
2757 * Whether a request will be proxied is determined by the attributes present
2758 * in request->config. If any of the following attributes are found, the
2759 * request may be proxied.
2761 * The key attributes are:
2762 * - PW_PROXY_TO_REALM - Specifies a realm the request should be proxied to.
2763 * - PW_HOME_SERVER_POOL - Specifies a specific home server pool to proxy to.
2764 * - PW_PACKET_DST_IP_ADDRESS - Specifies a specific IPv4 home server to proxy to.
2765 * - PW_PACKET_DST_IPV6_ADDRESS - Specifies a specific IPv6 home server to proxy to.
2767 * Certain packet types such as #PW_CODE_STATUS_SERVER will never be proxied.
2769 * If request should be proxied, will:
2770 * - Add request:Proxy-State
2771 * - Strip the current username value of its realm (depending on config)
2772 * - Create a CHAP-Challenge from the original request vector, if one doesn't already
2774 * - Call the pre-process section in the current server, or in the virtual server
2775 * associated with the home server pool we're proxying to.
2777 * @todo A lot of this logic is RADIUS specific, and should be moved out into a protocol
2778 * specific function.
2780 * @param request The #REQUEST to evaluate for proxying.
2781 * @return 0 if not proxying, 1 if request should be proxied, -1 on error.
2783 static int request_will_proxy(REQUEST *request)
2785 int rcode, pre_proxy_type = 0;
2786 char const *realmname = NULL;
2787 VALUE_PAIR *vp, *strippedname;
2788 home_server_t *home;
2789 REALM *realm = NULL;
2790 home_pool_t *pool = NULL;
2792 VERIFY_REQUEST(request);
2794 if (!request->root->proxy_requests) return 0;
2795 if (request->packet->dst_port == 0) return 0;
2796 if (request->packet->code == PW_CODE_STATUS_SERVER) return 0;
2797 if (request->in_proxy_hash) return 0;
2800 * FIXME: for 3.0, allow this only for rejects?
2802 if (request->reply->code != 0) return 0;
2804 vp = fr_pair_find_by_num(request->config, PW_PROXY_TO_REALM, 0, TAG_ANY);
2806 realm = realm_find2(vp->vp_strvalue);
2808 REDEBUG2("Cannot proxy to unknown realm %s",
2813 realmname = vp->vp_strvalue;
2816 * Figure out which pool to use.
2818 if (request->packet->code == PW_CODE_ACCESS_REQUEST) {
2819 pool = realm->auth_pool;
2821 #ifdef WITH_ACCOUNTING
2822 } else if (request->packet->code == PW_CODE_ACCOUNTING_REQUEST) {
2823 pool = realm->acct_pool;
2827 } else if ((request->packet->code == PW_CODE_COA_REQUEST) ||
2828 (request->packet->code == PW_CODE_DISCONNECT_REQUEST)) {
2829 pool = realm->coa_pool;
2836 } else if ((vp = fr_pair_find_by_num(request->config, PW_HOME_SERVER_POOL, 0, TAG_ANY)) != NULL) {
2839 switch (request->packet->code) {
2840 case PW_CODE_ACCESS_REQUEST:
2841 pool_type = HOME_TYPE_AUTH;
2844 #ifdef WITH_ACCOUNTING
2845 case PW_CODE_ACCOUNTING_REQUEST:
2846 pool_type = HOME_TYPE_ACCT;
2851 case PW_CODE_COA_REQUEST:
2852 case PW_CODE_DISCONNECT_REQUEST:
2853 pool_type = HOME_TYPE_COA;
2861 pool = home_pool_byname(vp->vp_strvalue, pool_type);
2864 * Send it directly to a home server (i.e. NAS)
2866 } else if (((vp = fr_pair_find_by_num(request->config, PW_PACKET_DST_IP_ADDRESS, 0, TAG_ANY)) != NULL) ||
2867 ((vp = fr_pair_find_by_num(request->config, PW_PACKET_DST_IPV6_ADDRESS, 0, TAG_ANY)) != NULL)) {
2869 fr_ipaddr_t dst_ipaddr;
2871 memset(&dst_ipaddr, 0, sizeof(dst_ipaddr));
2873 if (vp->da->attr == PW_PACKET_DST_IP_ADDRESS) {
2874 dst_ipaddr.af = AF_INET;
2875 dst_ipaddr.ipaddr.ip4addr.s_addr = vp->vp_ipaddr;
2876 dst_ipaddr.prefix = 32;
2878 dst_ipaddr.af = AF_INET6;
2879 memcpy(&dst_ipaddr.ipaddr.ip6addr, &vp->vp_ipv6addr, sizeof(vp->vp_ipv6addr));
2880 dst_ipaddr.prefix = 128;
2883 vp = fr_pair_find_by_num(request->config, PW_PACKET_DST_PORT, 0, TAG_ANY);
2885 if (request->packet->code == PW_CODE_ACCESS_REQUEST) {
2886 dst_port = PW_AUTH_UDP_PORT;
2888 #ifdef WITH_ACCOUNTING
2889 } else if (request->packet->code == PW_CODE_ACCOUNTING_REQUEST) {
2890 dst_port = PW_ACCT_UDP_PORT;
2894 } else if ((request->packet->code == PW_CODE_COA_REQUEST) ||
2895 (request->packet->code == PW_CODE_DISCONNECT_REQUEST)) {
2896 dst_port = PW_COA_UDP_PORT;
2898 } else { /* shouldn't happen for RADIUS... */
2903 dst_port = vp->vp_integer;
2907 * Nothing does CoA over TCP.
2909 home = home_server_find(&dst_ipaddr, dst_port, IPPROTO_UDP);
2913 WARN("No such home server %s port %u",
2914 inet_ntop(dst_ipaddr.af, &dst_ipaddr.ipaddr, buffer, sizeof(buffer)),
2915 (unsigned int) dst_port);
2920 * The home server is alive (or may be alive).
2921 * Send the packet to the IP.
2923 if (home->state != HOME_STATE_IS_DEAD) goto do_home;
2926 * The home server is dead. If you wanted
2927 * fail-over, you should have proxied to a pool.
2938 RWDEBUG2("Cancelling proxy as no home pool exists");
2942 if (request->listener->synchronous) {
2943 WARN("Cannot proxy a request which is from a 'synchronous' socket");
2947 request->home_pool = pool;
2949 home = home_server_ldb(realmname, pool, request);
2952 REDEBUG2("Failed to find live home server: Cancelling proxy");
2957 home_server_update_request(home, request);
2961 * Once we've decided to proxy a request, we cannot send
2962 * a CoA packet. So we free up any CoA packet here.
2964 if (request->coa) request_done(request->coa, FR_ACTION_DONE);
2968 * Remember that we sent the request to a Realm.
2970 if (realmname) pair_make_request("Realm", realmname, T_OP_EQ);
2973 * Strip the name, if told to.
2975 * Doing it here catches the case of proxied tunneled
2978 if (realm && (realm->strip_realm == true) &&
2979 (strippedname = fr_pair_find_by_num(request->proxy->vps, PW_STRIPPED_USER_NAME, 0, TAG_ANY)) != NULL) {
2981 * If there's a Stripped-User-Name attribute in
2982 * the request, then use THAT as the User-Name
2983 * for the proxied request, instead of the
2986 * This is done by making a copy of the
2987 * Stripped-User-Name attribute, turning it into
2988 * a User-Name attribute, deleting the
2989 * Stripped-User-Name and User-Name attributes
2990 * from the vps list, and making the new
2991 * User-Name the head of the vps list.
2993 vp = fr_pair_find_by_num(request->proxy->vps, PW_USER_NAME, 0, TAG_ANY);
2996 vp = radius_pair_create(NULL, NULL,
2998 rad_assert(vp != NULL); /* handled by above function */
2999 /* Insert at the START of the list */
3000 /* FIXME: Can't make assumptions about ordering */
3001 fr_cursor_init(&cursor, &vp);
3002 fr_cursor_merge(&cursor, request->proxy->vps);
3003 request->proxy->vps = vp;
3005 fr_pair_value_strcpy(vp, strippedname->vp_strvalue);
3008 * Do NOT delete Stripped-User-Name.
3013 * If there is no PW_CHAP_CHALLENGE attribute but
3014 * there is a PW_CHAP_PASSWORD we need to add it
3015 * since we can't use the request authenticator
3016 * anymore - we changed it.
3018 if ((request->packet->code == PW_CODE_ACCESS_REQUEST) &&
3019 fr_pair_find_by_num(request->proxy->vps, PW_CHAP_PASSWORD, 0, TAG_ANY) &&
3020 fr_pair_find_by_num(request->proxy->vps, PW_CHAP_CHALLENGE, 0, TAG_ANY) == NULL) {
3021 vp = radius_pair_create(request->proxy, &request->proxy->vps, PW_CHAP_CHALLENGE, 0);
3022 fr_pair_value_memcpy(vp, request->packet->vector, sizeof(request->packet->vector));
3026 * The RFC's say we have to do this, but FreeRADIUS
3029 vp = radius_pair_create(request->proxy, &request->proxy->vps, PW_PROXY_STATE, 0);
3030 fr_pair_value_sprintf(vp, "%u", request->packet->id);
3033 * Should be done BEFORE inserting into proxy hash, as
3034 * pre-proxy may use this information, or change it.
3036 request->proxy->code = request->packet->code;
3039 * Call the pre-proxy routines.
3041 vp = fr_pair_find_by_num(request->config, PW_PRE_PROXY_TYPE, 0, TAG_ANY);
3043 DICT_VALUE const *dval = dict_valbyattr(vp->da->attr, vp->da->vendor, vp->vp_integer);
3044 /* Must be a validation issue */
3046 RDEBUG2("Found Pre-Proxy-Type %s", dval->name);
3047 pre_proxy_type = vp->vp_integer;
3051 * home_pool may be NULL when originating CoA packets,
3052 * because they go directly to an IP address.
3054 if (request->home_pool && request->home_pool->virtual_server) {
3055 char const *old_server = request->server;
3057 request->server = request->home_pool->virtual_server;
3059 RDEBUG2("server %s {", request->server);
3061 rcode = process_pre_proxy(pre_proxy_type, request);
3065 request->server = old_server;
3067 rcode = process_pre_proxy(pre_proxy_type, request);
3071 case RLM_MODULE_FAIL:
3072 case RLM_MODULE_INVALID:
3073 case RLM_MODULE_NOTFOUND:
3074 case RLM_MODULE_USERLOCK:
3076 /* FIXME: debug print failed stuff */
3079 case RLM_MODULE_REJECT:
3080 case RLM_MODULE_HANDLED:
3084 * Only proxy the packet if the pre-proxy code succeeded.
3086 case RLM_MODULE_NOOP:
3088 case RLM_MODULE_UPDATED:
3093 static int proxy_to_virtual_server(REQUEST *request)
3097 if (request->packet->dst_port == 0) {
3098 WARN("Cannot proxy an internal request");
3102 DEBUG("Proxying to virtual server %s",
3103 request->home_server->server);
3106 * Packets to virtual servers don't get
3107 * retransmissions sent to them. And the virtual
3108 * server is run ONLY if we have no child
3109 * threads, or we're running in a child thread.
3111 rad_assert(!spawn_flag || !we_are_master());
3113 fake = request_alloc_fake(request);
3115 fake->packet->vps = fr_pair_list_copy(fake->packet, request->packet->vps);
3116 talloc_free(request->proxy);
3118 fake->server = request->home_server->server;
3119 fake->handle = request->handle;
3120 fake->process = NULL; /* should never be run for anything */
3123 * Run the virtual server.
3125 request_running(fake, FR_ACTION_RUN);
3127 request->proxy = talloc_steal(request, fake->packet);
3128 fake->packet = NULL;
3129 request->proxy_reply = talloc_steal(request, fake->reply);
3135 * No reply code, toss the reply we have,
3136 * and do post-proxy-type Fail.
3138 if (!request->proxy_reply->code) {
3139 TALLOC_FREE(request->proxy_reply);
3140 setup_post_proxy_fail(request);
3144 * Do the proxy reply (if any)
3146 if (process_proxy_reply(request, request->proxy_reply)) {
3147 request->handle(request);
3150 return -1; /* so we call request_finish */
3154 static int request_proxy(REQUEST *request, int retransmit)
3158 VERIFY_REQUEST(request);
3160 rad_assert(request->parent == NULL);
3161 rad_assert(request->home_server != NULL);
3163 if (request->master_state == REQUEST_STOP_PROCESSING) return 0;
3167 RWDEBUG("Cannot proxy and originate CoA packets at the same time. Cancelling CoA request");
3168 request_done(request->coa, FR_ACTION_DONE);
3173 * The request may need sending to a virtual server.
3174 * This code is more than a little screwed up. The rest
3175 * of the state machine doesn't handle parent / child
3176 * relationships well. i.e. if the child request takes
3177 * too long, the core will mark the *parent* as "stop
3178 * processing". And the child will continue without
3179 * knowing anything...
3181 * So, we have some horrible hacks to get around that.
3183 if (request->home_server->server) return proxy_to_virtual_server(request);
3186 * We're actually sending a proxied packet. Do that now.
3188 if (!request->in_proxy_hash && !insert_into_proxy_hash(request)) {
3189 RPROXY("Failed to insert request into the proxy list");
3193 rad_assert(request->proxy->id >= 0);
3195 if (rad_debug_lvl) {
3196 struct timeval *response_window;
3198 response_window = request_response_window(request);
3201 if (request->home_server->tls) {
3202 RDEBUG2("Proxying request to home server %s port %d (TLS) timeout %d.%06d",
3203 inet_ntop(request->proxy->dst_ipaddr.af,
3204 &request->proxy->dst_ipaddr.ipaddr,
3205 buffer, sizeof(buffer)),
3206 request->proxy->dst_port,
3207 (int) response_window->tv_sec, (int) response_window->tv_usec);
3210 RDEBUG2("Proxying request to home server %s port %d timeout %d.%06d",
3211 inet_ntop(request->proxy->dst_ipaddr.af,
3212 &request->proxy->dst_ipaddr.ipaddr,
3213 buffer, sizeof(buffer)),
3214 request->proxy->dst_port,
3215 (int) response_window->tv_sec, (int) response_window->tv_usec);
3220 gettimeofday(&request->proxy_retransmit, NULL);
3222 request->proxy->timestamp = request->proxy_retransmit;
3224 request->home_server->last_packet_sent = request->proxy_retransmit.tv_sec;
3227 * Encode the packet before we do anything else.
3229 request->proxy_listener->encode(request->proxy_listener, request);
3230 debug_packet(request, request->proxy, false);
3233 * Set the state function, then the state, no child, and
3236 * The order here is different from other state changes
3237 * due to race conditions with replies from the home
3240 request->process = proxy_wait_for_reply;
3241 request->child_state = REQUEST_PROXIED;
3242 request->component = "<REQUEST_PROXIED>";
3243 request->module = "";
3247 * And send the packet.
3249 request->proxy_listener->send(request->proxy_listener, request);
3254 * Proxy the packet as if it was new.
3256 static int request_proxy_anew(REQUEST *request)
3258 home_server_t *home;
3260 VERIFY_REQUEST(request);
3263 * Delete the request from the proxy list.
3265 * The packet list code takes care of ensuring that IDs
3266 * aren't reused until all 256 IDs have been used. So
3267 * there's a 1/256 chance of re-using the same ID when
3268 * we're sending to the same home server. Which is
3271 remove_from_proxy_hash(request);
3274 * Find a live home server for the request.
3276 home = home_server_ldb(NULL, request->home_pool, request);
3278 REDEBUG2("Failed to find live home server for request");
3280 if (setup_post_proxy_fail(request)) {
3281 request_queue_or_run(request, proxy_running);
3283 gettimeofday(&request->reply->timestamp, NULL);
3284 request_cleanup_delay_init(request);
3289 #ifdef WITH_ACCOUNTING
3291 * Update the Acct-Delay-Time attribute.
3293 if (request->packet->code == PW_CODE_ACCOUNTING_REQUEST) {
3296 vp = fr_pair_find_by_num(request->proxy->vps, PW_ACCT_DELAY_TIME, 0, TAG_ANY);
3297 if (!vp) vp = radius_pair_create(request->proxy,
3298 &request->proxy->vps,
3299 PW_ACCT_DELAY_TIME, 0);
3303 gettimeofday(&now, NULL);
3304 vp->vp_integer += now.tv_sec - request->proxy_retransmit.tv_sec;
3310 * May have failed over to a "fallback" virtual server.
3311 * If so, run that instead of doing proxying to a real
3315 request->home_server = home;
3316 TALLOC_FREE(request->proxy);
3318 (void) proxy_to_virtual_server(request);
3322 home_server_update_request(home, request);
3324 if (!insert_into_proxy_hash(request)) {
3325 RPROXY("Failed to insert retransmission into the proxy list");
3326 goto post_proxy_fail;
3330 * Free the old packet, to force re-encoding
3332 talloc_free(request->proxy->data);
3333 request->proxy->data = NULL;
3334 request->proxy->data_len = 0;
3336 if (request_proxy(request, 1) != 1) goto post_proxy_fail;
3342 /** Ping a home server.
3345 static void request_ping(REQUEST *request, int action)
3347 home_server_t *home = request->home_server;
3350 VERIFY_REQUEST(request);
3352 TRACE_STATE_MACHINE;
3356 case FR_ACTION_TIMER:
3357 ERROR("No response to status check %d ID %u for home server %s port %d",
3360 inet_ntop(request->proxy->dst_ipaddr.af,
3361 &request->proxy->dst_ipaddr.ipaddr,
3362 buffer, sizeof(buffer)),
3363 request->proxy->dst_port);
3366 case FR_ACTION_PROXY_REPLY:
3367 rad_assert(request->in_proxy_hash);
3369 request->home_server->num_received_pings++;
3370 RPROXY("Received response to status check %d ID %u (%d in current sequence)",
3371 request->number, request->proxy->id, home->num_received_pings);
3374 * Remove the request from any hashes
3376 fr_event_delete(el, &request->ev);
3377 remove_from_proxy_hash(request);
3380 * The control socket may have marked the home server as
3381 * alive. OR, it may have suddenly started responding to
3382 * requests again. If so, don't re-do the "make alive"
3385 if (home->state == HOME_STATE_ALIVE) break;
3388 * It's dead, and we haven't received enough ping
3389 * responses to mark it "alive". Wait a bit.
3391 * If it's zombie, we mark it alive immediately.
3393 if ((home->state == HOME_STATE_IS_DEAD) &&
3394 (home->num_received_pings < home->num_pings_to_alive)) {
3399 * Mark it alive and delete any outstanding
3402 mark_home_server_alive(request, home);
3406 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
3410 rad_assert(!request->in_request_hash);
3411 rad_assert(request->ev == NULL);
3413 request_done(request, FR_ACTION_DONE);
3417 * Called from start of zombie period, OR after control socket
3418 * marks the home server dead.
3420 static void ping_home_server(void *ctx)
3422 home_server_t *home = talloc_get_type_abort(ctx, home_server_t);
3425 struct timeval when, now;
3427 if ((home->state == HOME_STATE_ALIVE) ||
3428 (home->ev != NULL)) {
3432 gettimeofday(&now, NULL);
3436 * We've run out of zombie time. Mark it dead.
3438 if (home->state == HOME_STATE_ZOMBIE) {
3439 when = home->zombie_period_start;
3440 when.tv_sec += home->zombie_period;
3442 if (timercmp(&when, &now, <)) {
3443 DEBUG("PING: Zombie period is over for home server %s", home->log_name);
3444 mark_home_server_dead(home, &now);
3449 * We're not supposed to be pinging it. Just wake up
3450 * when we're supposed to mark it dead.
3452 if (home->ping_check == HOME_PING_CHECK_NONE) {
3453 if (home->state == HOME_STATE_ZOMBIE) {
3454 home->when = home->zombie_period_start;
3455 home->when.tv_sec += home->zombie_period;
3456 INSERT_EVENT(ping_home_server, home);
3460 * Else mark_home_server_dead will set a timer
3461 * for revive_interval.
3467 request = request_alloc(NULL);
3468 if (!request) return;
3469 request->number = request_num_counter++;
3472 request->proxy = rad_alloc(request, true);
3473 rad_assert(request->proxy != NULL);
3475 if (home->ping_check == HOME_PING_CHECK_STATUS_SERVER) {
3476 request->proxy->code = PW_CODE_STATUS_SERVER;
3478 fr_pair_make(request->proxy, &request->proxy->vps,
3479 "Message-Authenticator", "0x00", T_OP_SET);
3481 } else if ((home->type == HOME_TYPE_AUTH) ||
3482 (home->type == HOME_TYPE_AUTH_ACCT)) {
3483 request->proxy->code = PW_CODE_ACCESS_REQUEST;
3485 fr_pair_make(request->proxy, &request->proxy->vps,
3486 "User-Name", home->ping_user_name, T_OP_SET);
3487 fr_pair_make(request->proxy, &request->proxy->vps,
3488 "User-Password", home->ping_user_password, T_OP_SET);
3489 fr_pair_make(request->proxy, &request->proxy->vps,
3490 "Service-Type", "Authenticate-Only", T_OP_SET);
3491 fr_pair_make(request->proxy, &request->proxy->vps,
3492 "Message-Authenticator", "0x00", T_OP_SET);
3494 #ifdef WITH_ACCOUNTING
3495 } else if (home->type == HOME_TYPE_ACCT) {
3496 request->proxy->code = PW_CODE_ACCOUNTING_REQUEST;
3498 fr_pair_make(request->proxy, &request->proxy->vps,
3499 "User-Name", home->ping_user_name, T_OP_SET);
3500 fr_pair_make(request->proxy, &request->proxy->vps,
3501 "Acct-Status-Type", "Stop", T_OP_SET);
3502 fr_pair_make(request->proxy, &request->proxy->vps,
3503 "Acct-Session-Id", "00000000", T_OP_SET);
3504 vp = fr_pair_make(request->proxy, &request->proxy->vps,
3505 "Event-Timestamp", "0", T_OP_SET);
3506 vp->vp_date = now.tv_sec;
3511 * Unkown home server type.
3513 talloc_free(request);
3517 vp = fr_pair_make(request->proxy, &request->proxy->vps,
3518 "NAS-Identifier", "", T_OP_SET);
3520 fr_pair_value_sprintf(vp, "Status Check %u. Are you alive?",
3521 home->num_sent_pings);
3525 request->proxy->proto = home->proto;
3527 request->proxy->src_ipaddr = home->src_ipaddr;
3528 request->proxy->dst_ipaddr = home->ipaddr;
3529 request->proxy->dst_port = home->port;
3530 request->home_server = home;
3531 #ifdef DEBUG_STATE_MACHINE
3532 if (rad_debug_lvl) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n", request->number, __FUNCTION__,
3533 child_state_names[request->child_state],
3534 child_state_names[REQUEST_DONE]);
3535 if (rad_debug_lvl) printf("(%u) ********\tNEXT-STATE %s -> %s\n", request->number, __FUNCTION__, "request_ping");
3537 #ifdef HAVE_PTHREAD_H
3538 rad_assert(request->child_pid == NO_SUCH_CHILD_PID);
3540 request->child_state = REQUEST_PROXIED;
3541 request->process = request_ping;
3543 rad_assert(request->proxy_listener == NULL);
3545 if (!insert_into_proxy_hash(request)) {
3546 RPROXY("Failed to insert status check %d into proxy list. Discarding it.",
3549 rad_assert(!request->in_request_hash);
3550 rad_assert(!request->in_proxy_hash);
3551 rad_assert(request->ev == NULL);
3552 talloc_free(request);
3557 * Set up the timer callback.
3560 when.tv_sec += home->ping_timeout;
3562 DEBUG("PING: Waiting %u seconds for response to ping",
3563 home->ping_timeout);
3565 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
3566 home->num_sent_pings++;
3568 rad_assert(request->proxy_listener != NULL);
3569 debug_packet(request, request->proxy, false);
3570 request->proxy_listener->send(request->proxy_listener,
3574 * Add +/- 2s of jitter, as suggested in RFC 3539
3575 * and in the Issues and Fixes draft.
3578 home->when.tv_sec += home->ping_interval;
3580 add_jitter(&home->when);
3582 DEBUG("PING: Next status packet in %u seconds", home->ping_interval);
3583 INSERT_EVENT(ping_home_server, home);
3586 static void home_trigger(home_server_t *home, char const *trigger)
3588 REQUEST *my_request;
3589 RADIUS_PACKET *my_packet;
3591 my_request = talloc_zero(NULL, REQUEST);
3592 my_packet = talloc_zero(my_request, RADIUS_PACKET);
3593 my_request->proxy = my_packet;
3594 my_packet->dst_ipaddr = home->ipaddr;
3595 my_packet->src_ipaddr = home->src_ipaddr;
3597 exec_trigger(my_request, home->cs, trigger, false);
3598 talloc_free(my_request);
3601 static void mark_home_server_zombie(home_server_t *home, struct timeval *now, struct timeval *response_window)
3608 rad_assert((home->state == HOME_STATE_ALIVE) ||
3609 (home->state == HOME_STATE_UNKNOWN));
3612 * We've received a real packet recently. Don't mark the
3613 * server as zombie until we've received NO packets for a
3614 * while. The "1/4" of zombie period was chosen rather
3615 * arbitrarily. It's a balance between too short, which
3616 * gives quick fail-over and fail-back, or too long,
3617 * where the proxy still sends packets to an unresponsive
3620 start = now->tv_sec - ((home->zombie_period + 3) / 4);
3621 if (home->last_packet_recv >= start) {
3622 DEBUG("Received reply from home server %d seconds ago. Might not be zombie.",
3623 (int) (now->tv_sec - home->last_packet_recv));
3627 home->state = HOME_STATE_ZOMBIE;
3628 home_trigger(home, "home_server.zombie");
3631 * Set the home server to "zombie", as of the time
3634 home->zombie_period_start.tv_sec = start;
3635 home->zombie_period_start.tv_usec = USEC / 2;
3637 fr_event_delete(el, &home->ev);
3639 home->num_sent_pings = 0;
3640 home->num_received_pings = 0;
3642 PROXY( "Marking home server %s port %d as zombie (it has not responded in %d.%06d seconds).",
3643 inet_ntop(home->ipaddr.af, &home->ipaddr.ipaddr,
3644 buffer, sizeof(buffer)),
3645 home->port, (int) response_window->tv_sec, (int) response_window->tv_usec);
3647 ping_home_server(home);
3651 void revive_home_server(void *ctx)
3653 home_server_t *home = talloc_get_type_abort(ctx, home_server_t);
3656 home->state = HOME_STATE_ALIVE;
3657 home->response_timeouts = 0;
3658 home_trigger(home, "home_server.alive");
3659 home->currently_outstanding = 0;
3660 gettimeofday(&home->revive_time, NULL);
3663 * Delete any outstanding events.
3666 if (home->ev) fr_event_delete(el, &home->ev);
3668 PROXY( "Marking home server %s port %d alive again... we have no idea if it really is alive or not.",
3669 inet_ntop(home->ipaddr.af, &home->ipaddr.ipaddr,
3670 buffer, sizeof(buffer)),
3674 void mark_home_server_dead(home_server_t *home, struct timeval *when)
3676 int previous_state = home->state;
3679 PROXY( "Marking home server %s port %d as dead.",
3680 inet_ntop(home->ipaddr.af, &home->ipaddr.ipaddr,
3681 buffer, sizeof(buffer)),
3684 home->state = HOME_STATE_IS_DEAD;
3685 home_trigger(home, "home_server.dead");
3687 if (home->ping_check != HOME_PING_CHECK_NONE) {
3689 * If the control socket marks us dead, start
3690 * pinging. Otherwise, we already started
3691 * pinging when it was marked "zombie".
3693 if (previous_state == HOME_STATE_ALIVE) {
3694 ping_home_server(home);
3696 DEBUG("PING: Already pinging home server %s", home->log_name);
3701 * Revive it after a fixed period of time. This
3702 * is very, very, bad.
3705 home->when.tv_sec += home->revive_interval;
3707 DEBUG("PING: Reviving home server %s in %u seconds", home->log_name, home->revive_interval);
3709 INSERT_EVENT(revive_home_server, home);
3713 /** Wait for a reply after proxying a request.
3715 * Retransmit the proxied packet, or time out and go to
3716 * proxy_no_reply. Mark the home server unresponsive, etc.
3718 * If we do receive a reply, we transition to proxy_running.
3721 * digraph proxy_wait_for_reply {
3722 * proxy_wait_for_reply;
3724 * proxy_wait_for_reply -> retransmit_proxied_request [ label = "DUP", arrowhead = "none" ];
3725 * proxy_wait_for_reply -> proxy_no_reply [ label = "TIMER >= response_window" ];
3726 * proxy_wait_for_reply -> timer [ label = "TIMER < max_request_time" ];
3727 * proxy_wait_for_reply -> proxy_running [ label = "PROXY_REPLY" arrowhead = "none"];
3728 * proxy_wait_for_reply -> done [ label = "TIMER >= max_request_time" ];
3732 static void proxy_wait_for_reply(REQUEST *request, int action)
3734 struct timeval now, when;
3735 struct timeval *response_window = NULL;
3736 home_server_t *home = request->home_server;
3739 VERIFY_REQUEST(request);
3741 TRACE_STATE_MACHINE;
3744 rad_assert(request->packet->code != PW_CODE_STATUS_SERVER);
3745 rad_assert(request->home_server != NULL);
3747 gettimeofday(&now, NULL);
3752 * We have a reply, ignore the retransmit.
3754 if (request->proxy_reply) return;
3757 * The request was proxied to a virtual server.
3758 * Ignore the retransmit.
3760 if (request->home_server->server) return;
3763 * Use a new connection when the home server is
3764 * dead, or when there's no proxy listener, or
3765 * when the listener is failed or dead.
3767 * If the listener is known or frozen, use it for
3770 if ((home->state == HOME_STATE_IS_DEAD) ||
3771 !request->proxy_listener ||
3772 (request->proxy_listener->status >= RAD_LISTEN_STATUS_EOL)) {
3773 request_proxy_anew(request);
3779 * The home server is still alive, but TCP. We
3780 * rely on TCP to get the request and reply back.
3781 * So there's no need to retransmit.
3783 if (home->proto == IPPROTO_TCP) {
3784 DEBUG2("Suppressing duplicate proxied request (tcp) to home server %s port %d proto TCP - ID: %d",
3785 inet_ntop(request->proxy->dst_ipaddr.af,
3786 &request->proxy->dst_ipaddr.ipaddr,
3787 buffer, sizeof(buffer)),
3788 request->proxy->dst_port,
3789 request->proxy->id);
3795 * More than one retransmit a second is stupid,
3796 * and should be suppressed by the proxy.
3798 when = request->proxy_retransmit;
3801 if (timercmp(&now, &when, <)) {
3802 DEBUG2("Suppressing duplicate proxied request (too fast) to home server %s port %d proto TCP - ID: %d",
3803 inet_ntop(request->proxy->dst_ipaddr.af,
3804 &request->proxy->dst_ipaddr.ipaddr,
3805 buffer, sizeof(buffer)),
3806 request->proxy->dst_port,
3807 request->proxy->id);
3811 #ifdef WITH_ACCOUNTING
3813 * If we update the Acct-Delay-Time, we need to
3816 if ((request->packet->code == PW_CODE_ACCOUNTING_REQUEST) &&
3817 fr_pair_find_by_num(request->proxy->vps, PW_ACCT_DELAY_TIME, 0, TAG_ANY)) {
3818 request_proxy_anew(request);
3823 RDEBUG2("Sending duplicate proxied request to home server %s port %d - ID: %d",
3824 inet_ntop(request->proxy->dst_ipaddr.af,
3825 &request->proxy->dst_ipaddr.ipaddr,
3826 buffer, sizeof(buffer)),
3827 request->proxy->dst_port,
3828 request->proxy->id);
3829 request->num_proxied_requests++;
3831 rad_assert(request->proxy_listener != NULL);
3832 FR_STATS_TYPE_INC(home->stats.total_requests);
3833 home->last_packet_sent = now.tv_sec;
3834 request->proxy_retransmit = now;
3835 debug_packet(request, request->proxy, false);
3836 request->proxy_listener->send(request->proxy_listener, request);
3839 case FR_ACTION_TIMER:
3840 response_window = request_response_window(request);
3843 if (!request->proxy_listener ||
3844 (request->proxy_listener->status >= RAD_LISTEN_STATUS_EOL)) {
3845 remove_from_proxy_hash(request);
3847 when = request->packet->timestamp;
3848 when.tv_sec += request->root->max_request_time;
3850 if (timercmp(&when, &now, >)) {
3851 RDEBUG("Waiting for client retransmission in order to do a proxy retransmit");
3852 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
3859 * Wake up "response_window" time in the future.
3860 * i.e. when MY packet hasn't received a response.
3862 * Note that we DO NOT mark the home server as
3863 * zombie if it doesn't respond to us. It may be
3864 * responding to other (better looking) packets.
3866 when = request->proxy->timestamp;
3867 timeradd(&when, response_window, &when);
3870 * Not at the response window. Set the timer for
3873 if (timercmp(&when, &now, >)) {
3874 struct timeval diff;
3875 timersub(&when, &now, &diff);
3877 RDEBUG("Expecting proxy response no later than %d.%06d seconds from now",
3878 (int) diff.tv_sec, (int) diff.tv_usec);
3879 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
3884 RDEBUG("No proxy response, giving up on request and marking it done");
3887 * If we haven't received any packets for
3888 * "response_window", then mark the home server
3891 * This check should really be part of a home
3892 * server state machine.
3894 if (((home->state == HOME_STATE_ALIVE) ||
3895 (home->state == HOME_STATE_UNKNOWN))
3897 home->response_timeouts++;
3898 if (home->response_timeouts >= home->max_response_timeouts)
3899 mark_home_server_zombie(home, &now, response_window);
3902 FR_STATS_TYPE_INC(home->stats.total_timeouts);
3903 if (home->type == HOME_TYPE_AUTH) {
3904 if (request->proxy_listener) FR_STATS_TYPE_INC(request->proxy_listener->stats.total_timeouts);
3905 FR_STATS_TYPE_INC(proxy_auth_stats.total_timeouts);
3908 else if (home->type == HOME_TYPE_ACCT) {
3909 if (request->proxy_listener) FR_STATS_TYPE_INC(request->proxy_listener->stats.total_timeouts);
3910 FR_STATS_TYPE_INC(proxy_acct_stats.total_timeouts);
3914 else if (home->type == HOME_TYPE_COA) {
3915 if (request->proxy_listener) FR_STATS_TYPE_INC(request->proxy_listener->stats.total_timeouts);
3917 if (request->packet->code == PW_CODE_COA_REQUEST) {
3918 FR_STATS_TYPE_INC(proxy_coa_stats.total_timeouts);
3920 FR_STATS_TYPE_INC(proxy_dsc_stats.total_timeouts);
3926 * There was no response within the window. Stop
3927 * the request. If the client retransmitted, it
3928 * may have failed over to another home server.
3929 * But that one may be dead, too.
3931 * The extra verbose message if we have a username,
3932 * is extremely useful if the proxy is part of a chain
3933 * and the final home server, is not the one we're
3936 if (request->username) {
3937 RERROR("Failing proxied request for user \"%s\", due to lack of any response from home "
3938 "server %s port %d",
3939 request->username->vp_strvalue,
3940 inet_ntop(request->proxy->dst_ipaddr.af,
3941 &request->proxy->dst_ipaddr.ipaddr,
3942 buffer, sizeof(buffer)),
3943 request->proxy->dst_port);
3945 RERROR("Failing proxied request, due to lack of any response from home server %s port %d",
3946 inet_ntop(request->proxy->dst_ipaddr.af,
3947 &request->proxy->dst_ipaddr.ipaddr,
3948 buffer, sizeof(buffer)),
3949 request->proxy->dst_port);
3952 if (setup_post_proxy_fail(request)) {
3953 request_queue_or_run(request, proxy_no_reply);
3955 gettimeofday(&request->reply->timestamp, NULL);
3956 request_cleanup_delay_init(request);
3961 * We received a new reply. Go process it.
3963 case FR_ACTION_PROXY_REPLY:
3964 request_queue_or_run(request, proxy_running);
3968 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
3972 #endif /* WITH_PROXY */
3975 /***********************************************************************
3979 ***********************************************************************/
3981 static int null_handler(UNUSED REQUEST *request)
3987 * See if we need to originate a CoA request.
3989 static void request_coa_originate(REQUEST *request)
3991 int rcode, pre_proxy_type = 0;
3997 VERIFY_REQUEST(request);
3999 rad_assert(request->coa != NULL);
4000 rad_assert(request->proxy == NULL);
4001 rad_assert(!request->in_proxy_hash);
4002 rad_assert(request->proxy_reply == NULL);
4005 * Check whether we want to originate one, or cancel one.
4007 vp = fr_pair_find_by_num(request->config, PW_SEND_COA_REQUEST, 0, TAG_ANY);
4009 vp = fr_pair_find_by_num(request->coa->proxy->vps, PW_SEND_COA_REQUEST, 0, TAG_ANY);
4013 if (vp->vp_integer == 0) {
4015 TALLOC_FREE(request->coa);
4023 * src_ipaddr will be set up in proxy_encode.
4025 memset(&ipaddr, 0, sizeof(ipaddr));
4026 vp = fr_pair_find_by_num(coa->proxy->vps, PW_PACKET_DST_IP_ADDRESS, 0, TAG_ANY);
4028 ipaddr.af = AF_INET;
4029 ipaddr.ipaddr.ip4addr.s_addr = vp->vp_ipaddr;
4031 } else if ((vp = fr_pair_find_by_num(coa->proxy->vps, PW_PACKET_DST_IPV6_ADDRESS, 0, TAG_ANY)) != NULL) {
4032 ipaddr.af = AF_INET6;
4033 ipaddr.ipaddr.ip6addr = vp->vp_ipv6addr;
4034 ipaddr.prefix = 128;
4035 } else if ((vp = fr_pair_find_by_num(coa->proxy->vps, PW_HOME_SERVER_POOL, 0, TAG_ANY)) != NULL) {
4036 coa->home_pool = home_pool_byname(vp->vp_strvalue,
4038 if (!coa->home_pool) {
4039 RWDEBUG2("No such home_server_pool %s",
4045 * Prefer the pool to one server
4047 } else if (request->client->coa_pool) {
4048 coa->home_pool = request->client->coa_pool;
4050 } else if (request->client->coa_server) {
4051 coa->home_server = request->client->coa_server;
4055 * If all else fails, send it to the client that
4056 * originated this request.
4058 memcpy(&ipaddr, &request->packet->src_ipaddr, sizeof(ipaddr));
4062 * Use the pool, if it exists.
4064 if (coa->home_pool) {
4065 coa->home_server = home_server_ldb(NULL, coa->home_pool, coa);
4066 if (!coa->home_server) {
4067 RWDEBUG("No live home server for home_server_pool %s", coa->home_pool->name);
4070 home_server_update_request(coa->home_server, coa);
4072 } else if (!coa->home_server) {
4073 uint16_t port = PW_COA_UDP_PORT;
4075 vp = fr_pair_find_by_num(coa->proxy->vps, PW_PACKET_DST_PORT, 0, TAG_ANY);
4076 if (vp) port = vp->vp_integer;
4078 coa->home_server = home_server_find(&ipaddr, port, IPPROTO_UDP);
4079 if (!coa->home_server) {
4080 RWDEBUG2("Unknown destination %s:%d for CoA request.",
4081 inet_ntop(ipaddr.af, &ipaddr.ipaddr,
4082 buffer, sizeof(buffer)), port);
4087 vp = fr_pair_find_by_num(coa->proxy->vps, PW_PACKET_TYPE, 0, TAG_ANY);
4089 switch (vp->vp_integer) {
4090 case PW_CODE_COA_REQUEST:
4091 case PW_CODE_DISCONNECT_REQUEST:
4092 coa->proxy->code = vp->vp_integer;
4096 DEBUG("Cannot set CoA Packet-Type to code %d",
4102 if (!coa->proxy->code) coa->proxy->code = PW_CODE_COA_REQUEST;
4105 * The rest of the server code assumes that
4106 * request->packet && request->reply exist. Copy them
4107 * from the original request.
4109 rad_assert(coa->packet != NULL);
4110 rad_assert(coa->packet->vps == NULL);
4112 coa->packet = rad_copy_packet(coa, request->packet);
4113 coa->reply = rad_copy_packet(coa, request->reply);
4115 coa->config = fr_pair_list_copy(coa, request->config);
4116 coa->num_coa_requests = 0;
4117 coa->handle = null_handler;
4118 coa->number = request->number; /* it's associated with the same request */
4121 * Call the pre-proxy routines.
4123 vp = fr_pair_find_by_num(request->config, PW_PRE_PROXY_TYPE, 0, TAG_ANY);
4125 DICT_VALUE const *dval = dict_valbyattr(vp->da->attr, vp->da->vendor, vp->vp_integer);
4126 /* Must be a validation issue */
4128 RDEBUG2("Found Pre-Proxy-Type %s", dval->name);
4129 pre_proxy_type = vp->vp_integer;
4132 if (coa->home_pool && coa->home_pool->virtual_server) {
4133 char const *old_server = coa->server;
4135 coa->server = coa->home_pool->virtual_server;
4136 RDEBUG2("server %s {", coa->server);
4138 rcode = process_pre_proxy(pre_proxy_type, coa);
4141 coa->server = old_server;
4143 rcode = process_pre_proxy(pre_proxy_type, coa);
4150 * Only send the CoA packet if the pre-proxy code succeeded.
4152 case RLM_MODULE_NOOP:
4154 case RLM_MODULE_UPDATED:
4159 * Source IP / port is set when the proxy socket
4162 coa->proxy->dst_ipaddr = coa->home_server->ipaddr;
4163 coa->proxy->dst_port = coa->home_server->port;
4165 if (!insert_into_proxy_hash(coa)) {
4166 radlog_request(L_PROXY, 0, coa, "Failed to insert CoA request into proxy list");
4171 * We CANNOT divorce the CoA request from the parent
4172 * request. This function is running in a child thread,
4173 * and we need access to the main event loop in order to
4174 * to add the timers for the CoA packet.
4176 * Instead, we wait for the timer on the parent request
4179 gettimeofday(&coa->proxy->timestamp, NULL);
4180 coa->packet->timestamp = coa->proxy->timestamp; /* for max_request_time */
4181 coa->home_server->last_packet_sent = coa->proxy->timestamp.tv_sec;
4182 coa->delay = 0; /* need to calculate a new delay */
4185 * If requested, put a State attribute into the packet,
4186 * and cache the VPS.
4188 fr_state_put_vps(coa, NULL, coa->packet);
4191 * Encode the packet before we do anything else.
4193 coa->proxy_listener->encode(coa->proxy_listener, coa);
4194 debug_packet(coa, coa->proxy, false);
4196 #ifdef DEBUG_STATE_MACHINE
4197 if (rad_debug_lvl) printf("(%u) ********\tSTATE %s C-%s -> C-%s\t********\n", request->number, __FUNCTION__,
4198 child_state_names[request->child_state],
4199 child_state_names[REQUEST_PROXIED]);
4203 * Set the state function, then the state, no child, and
4206 coa->process = coa_wait_for_reply;
4207 coa->child_state = REQUEST_PROXIED;
4209 #ifdef HAVE_PTHREAD_H
4210 coa->child_pid = NO_SUCH_CHILD_PID;
4213 if (we_are_master()) coa_separate(request->coa);
4216 * And send the packet.
4218 coa->proxy_listener->send(coa->proxy_listener, coa);
4222 static void coa_retransmit(REQUEST *request)
4224 uint32_t delay, frac;
4225 struct timeval now, when, mrd;
4228 VERIFY_REQUEST(request);
4230 fr_event_now(el, &now);
4232 if (request->delay == 0) {
4234 * Implement re-transmit algorithm as per RFC 5080
4237 * We want IRT + RAND*IRT
4238 * or 0.9 IRT + rand(0,.2) IRT
4240 * 2^20 ~ USEC, and we want 2.
4241 * rand(0,0.2) USEC ~ (rand(0,2^21) / 10)
4243 delay = (fr_rand() & ((1 << 22) - 1)) / 10;
4244 request->delay = delay * request->home_server->coa_irt;
4245 delay = request->home_server->coa_irt * USEC;
4246 delay -= delay / 10;
4247 delay += request->delay;
4248 request->delay = delay;
4250 when = request->proxy->timestamp;
4251 tv_add(&when, delay);
4253 if (timercmp(&when, &now, >)) {
4254 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
4260 * Retransmit CoA request.
4264 * Cap count at MRC, if it is non-zero.
4266 if (request->home_server->coa_mrc &&
4267 (request->num_coa_requests >= request->home_server->coa_mrc)) {
4268 RERROR("Failing request - originate-coa ID %u, due to lack of any response from coa server %s port %d",
4270 inet_ntop(request->proxy->dst_ipaddr.af,
4271 &request->proxy->dst_ipaddr.ipaddr,
4272 buffer, sizeof(buffer)),
4273 request->proxy->dst_port);
4275 if (setup_post_proxy_fail(request)) {
4276 request_queue_or_run(request, coa_no_reply);
4278 request_done(request, FR_ACTION_DONE);
4284 * RFC 5080 Section 2.2.1
4286 * RT = 2*RTprev + RAND*RTprev
4287 * = 1.9 * RTprev + rand(0,.2) * RTprev
4288 * = 1.9 * RTprev + rand(0,1) * (RTprev / 5)
4291 delay ^= (delay >> 16);
4293 frac = request->delay / 5;
4294 delay = ((frac >> 16) * delay) + (((frac & 0xffff) * delay) >> 16);
4296 delay += (2 * request->delay) - (request->delay / 10);
4299 * Cap delay at MRT, if MRT is non-zero.
4301 if (request->home_server->coa_mrt &&
4302 (delay > (request->home_server->coa_mrt * USEC))) {
4303 int mrt_usec = request->home_server->coa_mrt * USEC;
4306 * delay = MRT + RAND * MRT
4307 * = 0.9 MRT + rand(0,.2) * MRT
4310 delay ^= (delay >> 15);
4312 delay = ((mrt_usec >> 16) * delay) + (((mrt_usec & 0xffff) * delay) >> 16);
4313 delay += mrt_usec - (mrt_usec / 10);
4316 request->delay = delay;
4318 tv_add(&when, request->delay);
4319 mrd = request->proxy->timestamp;
4320 mrd.tv_sec += request->home_server->coa_mrd;
4323 * Cap duration at MRD.
4325 if (timercmp(&mrd, &when, <)) {
4328 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
4330 request->num_coa_requests++; /* is NOT reset by code 3 lines above! */
4332 FR_STATS_TYPE_INC(request->home_server->stats.total_requests);
4334 RDEBUG2("Sending duplicate CoA request to home server %s port %d - ID: %d",
4335 inet_ntop(request->proxy->dst_ipaddr.af,
4336 &request->proxy->dst_ipaddr.ipaddr,
4337 buffer, sizeof(buffer)),
4338 request->proxy->dst_port,
4339 request->proxy->id);
4341 request->proxy_listener->send(request->proxy_listener,
4346 /** Wait for a reply after originating a CoA a request.
4348 * Retransmit the proxied packet, or time out and go to
4349 * coa_no_reply. Mark the home server unresponsive, etc.
4351 * If we do receive a reply, we transition to coa_running.
4354 * digraph coa_wait_for_reply {
4355 * coa_wait_for_reply;
4357 * coa_wait_for_reply -> coa_no_reply [ label = "TIMER >= response_window" ];
4358 * coa_wait_for_reply -> timer [ label = "TIMER < max_request_time" ];
4359 * coa_wait_for_reply -> coa_running [ label = "PROXY_REPLY" arrowhead = "none"];
4360 * coa_wait_for_reply -> done [ label = "TIMER >= max_request_time" ];
4364 static void coa_wait_for_reply(REQUEST *request, int action)
4366 VERIFY_REQUEST(request);
4368 TRACE_STATE_MACHINE;
4372 if (request->parent) coa_separate(request);
4375 case FR_ACTION_TIMER:
4376 if (request_max_time(request)) break;
4378 coa_retransmit(request);
4381 case FR_ACTION_PROXY_REPLY:
4382 request_queue_or_run(request, coa_running);
4386 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
4391 static void coa_separate(REQUEST *request)
4393 VERIFY_REQUEST(request);
4394 #ifdef DEBUG_STATE_MACHINE
4395 int action = FR_ACTION_TIMER;
4398 TRACE_STATE_MACHINE;
4401 rad_assert(request->parent != NULL);
4402 rad_assert(request->parent->coa == request);
4403 rad_assert(request->ev == NULL);
4404 rad_assert(!request->in_request_hash);
4405 rad_assert(request->coa == NULL);
4407 rad_assert(request->proxy_reply || request->proxy_listener);
4409 (void) talloc_steal(NULL, request);
4410 request->parent->coa = NULL;
4411 request->parent = NULL;
4413 if (we_are_master()) {
4415 coa_retransmit(request);
4420 /** Process a request after the CoA has timed out.
4422 * Run the packet through Post-Proxy-Type Fail
4425 * digraph coa_no_reply {
4428 * coa_no_reply -> dup [ label = "DUP", arrowhead = "none" ];
4429 * coa_no_reply -> timer [ label = "TIMER < max_request_time" ];
4430 * coa_no_reply -> coa_reply_too_late [ label = "PROXY_REPLY" arrowhead = "none"];
4431 * coa_no_reply -> process_proxy_reply [ label = "RUN" ];
4432 * coa_no_reply -> done [ label = "TIMER >= timeout" ];
4436 static void coa_no_reply(REQUEST *request, int action)
4440 VERIFY_REQUEST(request);
4442 TRACE_STATE_MACHINE;
4446 case FR_ACTION_TIMER:
4447 (void) request_max_time(request);
4450 case FR_ACTION_PROXY_REPLY: /* too late! */
4451 RDEBUG2("Reply from CoA server %s port %d - ID: %d arrived too late.",
4452 inet_ntop(request->proxy->src_ipaddr.af,
4453 &request->proxy->src_ipaddr.ipaddr,
4454 buffer, sizeof(buffer)),
4455 request->proxy->dst_port, request->proxy->id);
4459 if (process_proxy_reply(request, NULL)) {
4460 request->handle(request);
4462 request_done(request, FR_ACTION_DONE);
4466 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
4472 /** Process the request after receiving a coa reply.
4474 * Throught the post-proxy section, and the through the handler
4478 * digraph coa_running {
4481 * coa_running -> timer [ label = "TIMER < max_request_time" ];
4482 * coa_running -> process_proxy_reply [ label = "RUN" ];
4483 * coa_running -> done [ label = "TIMER >= timeout" ];
4487 static void coa_running(REQUEST *request, int action)
4489 VERIFY_REQUEST(request);
4491 TRACE_STATE_MACHINE;
4495 case FR_ACTION_TIMER:
4496 (void) request_max_time(request);
4500 if (process_proxy_reply(request, request->proxy_reply)) {
4501 request->handle(request);
4503 request_done(request, FR_ACTION_DONE);
4507 RDEBUG3("%s: Ignoring action %s", __FUNCTION__, action_codes[action]);
4511 #endif /* WITH_COA */
4513 /***********************************************************************
4515 * End of the State machine. Start of additional helper code.
4517 ***********************************************************************/
4519 /***********************************************************************
4523 ***********************************************************************/
4524 static void event_socket_handler(fr_event_list_t *xel, UNUSED int fd, void *ctx)
4526 rad_listen_t *listener = talloc_get_type_abort(ctx, rad_listen_t);
4528 rad_assert(xel == el);
4530 if ((listener->fd < 0)
4532 #ifndef WITH_DETAIL_THREAD
4533 && (listener->type != RAD_LISTEN_DETAIL)
4539 listener->print(listener, buffer, sizeof(buffer));
4540 ERROR("FATAL: Asked to read from closed socket: %s",
4543 rad_panic("Socket was closed on us!");
4547 listener->recv(listener);
4551 #ifdef WITH_DETAIL_THREAD
4554 * This function is called periodically to see if this detail
4555 * file is available for reading.
4557 static void event_poll_detail(void *ctx)
4560 rad_listen_t *this = talloc_get_type_abort(ctx, rad_listen_t);
4561 struct timeval when, now;
4562 listen_detail_t *detail = this->data;
4564 rad_assert(this->type == RAD_LISTEN_DETAIL);
4567 event_socket_handler(el, this->fd, this);
4569 fr_event_now(el, &now);
4573 * Backdoor API to get the delay until the next poll
4576 delay = this->encode(this, NULL);
4577 if (delay == 0) goto redo;
4579 tv_add(&when, delay);
4582 if (!fr_event_insert(el, event_poll_detail, this,
4583 &when, &detail->ev)) {
4584 ERROR("Failed creating handler");
4588 #endif /* WITH_DETAIL_THREAD */
4589 #endif /* WITH_DETAIL */
4591 static void event_status(struct timeval *wake)
4593 #if !defined(HAVE_PTHREAD_H) && defined(WNOHANG)
4597 if (rad_debug_lvl == 0) {
4599 INFO("Ready to process requests");
4600 just_started = false;
4606 INFO("Ready to process requests");
4608 } else if ((wake->tv_sec != 0) ||
4609 (wake->tv_usec >= 100000)) {
4610 DEBUG("Waking up in %d.%01u seconds.",
4611 (int) wake->tv_sec, (unsigned int) wake->tv_usec / 100000);
4616 * FIXME: Put this somewhere else, where it isn't called
4617 * all of the time...
4620 #if !defined(HAVE_PTHREAD_H) && defined(WNOHANG)
4622 * If there are no child threads, then there may
4623 * be child processes. In that case, wait for
4624 * their exit status, and throw that exit status
4625 * away. This helps get rid of zxombie children.
4627 while (waitpid(-1, &argval, WNOHANG) > 0) {
4635 static void listener_free_cb(void *ctx)
4637 rad_listen_t *this = talloc_get_type_abort(ctx, rad_listen_t);
4640 if (this->count > 0) {
4641 struct timeval when;
4642 listen_socket_t *sock = this->data;
4644 fr_event_now(el, &when);
4648 if (!fr_event_insert(el, listener_free_cb, this, &when,
4650 rad_panic("Failed to insert event");
4657 * It's all free, close the socket.
4660 this->print(this, buffer, sizeof(buffer));
4661 DEBUG("... cleaning up socket %s", buffer);
4662 rad_assert(this->next == NULL);
4668 static int proxy_eol_cb(void *ctx, void *data)
4670 struct timeval when;
4671 REQUEST *request = fr_packet2myptr(REQUEST, proxy, data);
4673 if (request->proxy_listener != ctx) return 0;
4676 * We don't care if it's being processed in a child thread.
4679 #ifdef WITH_ACCOUNTING
4681 * Accounting packets should be deleted immediately.
4682 * They will never be retransmitted by the client.
4684 if (request->proxy->code == PW_CODE_ACCOUNTING_REQUEST) {
4685 RDEBUG("Stopping request due to failed connection to home server");
4686 request->master_state = REQUEST_STOP_PROCESSING;
4691 * Reset the timer to be now, so that the request is
4692 * quickly updated. But spread the requests randomly
4693 * over the next second, so that we don't overload the
4696 fr_event_now(el, &when);
4697 tv_add(&when, fr_rand() % USEC);
4698 STATE_MACHINE_TIMER(FR_ACTION_TIMER);
4701 * Don't delete it from the list.
4707 static int event_new_fd(rad_listen_t *this)
4713 if (this->status == RAD_LISTEN_STATUS_KNOWN) return 1;
4715 this->print(this, buffer, sizeof(buffer));
4717 if (this->status == RAD_LISTEN_STATUS_INIT) {
4718 listen_socket_t *sock = this->data;
4720 rad_assert(sock != NULL);
4722 DEBUG("Listening on %s", buffer);
4724 INFO(" ... adding new socket %s", buffer);
4728 if (!just_started && (this->type == RAD_LISTEN_PROXY)) {
4729 home_server_t *home;
4732 if (!home || !home->limit.max_connections) {
4733 INFO(" ... adding new socket %s", buffer);
4735 INFO(" ... adding new socket %s (%u of %u)", buffer,
4736 home->limit.num_connections, home->limit.max_connections);
4742 switch (this->type) {
4745 * Detail files are always known, and aren't
4746 * put into the socket event loop.
4748 case RAD_LISTEN_DETAIL:
4749 this->status = RAD_LISTEN_STATUS_KNOWN;
4751 #ifndef WITH_DETAIL_THREAD
4753 * Set up the first poll interval.
4755 event_poll_detail(this);
4758 break; /* add the FD to the list */
4760 #endif /* WITH_DETAIL */
4764 * Add it to the list of sockets we can use.
4765 * Server sockets (i.e. auth/acct) are never
4766 * added to the packet list.
4768 case RAD_LISTEN_PROXY:
4770 rad_assert((sock->proto == IPPROTO_UDP) || (sock->home != NULL));
4773 * Add timers to outgoing child sockets, if necessary.
4775 if (sock->proto == IPPROTO_TCP && sock->opened &&
4776 (sock->home->limit.lifetime || sock->home->limit.idle_timeout)) {
4777 struct timeval when;
4779 when.tv_sec = sock->opened + 1;
4783 if (!fr_event_insert(el, tcp_socket_timer, this, &when,
4785 rad_panic("Failed to insert event");
4790 #endif /* WITH_PROXY */
4793 * FIXME: put idle timers on command sockets.
4799 * Add timers to incoming child sockets, if necessary.
4801 if (sock->proto == IPPROTO_TCP && sock->opened &&
4802 (sock->limit.lifetime || sock->limit.idle_timeout)) {
4803 struct timeval when;
4805 when.tv_sec = sock->opened + 1;
4809 if (!fr_event_insert(el, tcp_socket_timer, this, &when,
4811 ERROR("Failed adding timer for socket: %s", fr_strerror());
4817 } /* switch over listener types */
4820 * All sockets: add the FD to the event handler.
4822 if (!fr_event_fd_insert(el, 0, this->fd,
4823 event_socket_handler, this)) {
4824 ERROR("Failed adding event handler for socket: %s", fr_strerror());
4828 this->status = RAD_LISTEN_STATUS_KNOWN;
4834 * The socket has reached a timeout. Try to close it.
4836 if (this->status == RAD_LISTEN_STATUS_FROZEN) {
4838 * Requests are still using the socket. Wait for
4841 if (this->count > 0) {
4842 struct timeval when;
4843 listen_socket_t *sock = this->data;
4846 * Try again to clean up the socket in 30
4849 gettimeofday(&when, NULL);
4853 if (!fr_event_insert(el,
4854 (fr_event_callback_t) event_new_fd,
4855 this, &when, &sock->ev)) {
4856 rad_panic("Failed to insert event");
4862 fr_event_fd_delete(el, 0, this->fd);
4863 this->status = RAD_LISTEN_STATUS_REMOVE_NOW;
4867 * The socket has had a catastrophic error. Close it.
4869 if (this->status == RAD_LISTEN_STATUS_EOL) {
4871 * Remove it from the list of live FD's.
4873 fr_event_fd_delete(el, 0, this->fd);
4877 * Tell all requests using this socket that the socket is dead.
4879 if (this->type == RAD_LISTEN_PROXY) {
4880 PTHREAD_MUTEX_LOCK(&proxy_mutex);
4881 if (!fr_packet_list_socket_freeze(proxy_list,
4883 ERROR("Fatal error freezing socket: %s", fr_strerror());
4887 if (this->count > 0) {
4888 fr_packet_list_walk(proxy_list, this, proxy_eol_cb);
4890 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
4895 * Requests are still using the socket. Wait for
4898 if (this->count > 0) {
4899 struct timeval when;
4900 listen_socket_t *sock = this->data;
4903 * Try again to clean up the socket in 30
4906 gettimeofday(&when, NULL);
4910 if (!fr_event_insert(el,
4911 (fr_event_callback_t) event_new_fd,
4912 this, &when, &sock->ev)) {
4913 rad_panic("Failed to insert event");
4920 * No one is using the socket. We can remove it now.
4922 this->status = RAD_LISTEN_STATUS_REMOVE_NOW;
4923 } /* socket is at EOL */
4929 if (this->status == RAD_LISTEN_STATUS_REMOVE_NOW) {
4932 listen_socket_t *sock = this->data;
4934 struct timeval when;
4937 * Re-open the socket, pointing it to /dev/null.
4938 * This means that all writes proceed without
4939 * blocking, and all reads return "no data".
4941 * This leaves the socket active, so any child
4942 * threads won't go insane. But it means that
4943 * they cannot send or receive any packets.
4945 * This is EXTRA work in the normal case, when
4946 * sockets are closed without error. But it lets
4947 * us have one simple processing method for all
4950 devnull = open("/dev/null", O_RDWR);
4952 ERROR("FATAL failure opening /dev/null: %s",
4953 fr_syserror(errno));
4956 if (dup2(devnull, this->fd) < 0) {
4957 ERROR("FATAL failure closing socket: %s",
4958 fr_syserror(errno));
4964 rad_assert(this->type != RAD_LISTEN_DETAIL);
4970 * The socket is dead. Force all proxied packets
4971 * to stop using it. And then remove it from the
4972 * list of outgoing sockets.
4974 if (this->type == RAD_LISTEN_PROXY) {
4975 home_server_t *home;
4978 if (!home || !home->limit.max_connections) {
4979 INFO(" ... shutting down socket %s", buffer);
4981 INFO(" ... shutting down socket %s (%u of %u)", buffer,
4982 home->limit.num_connections, home->limit.max_connections);
4985 PTHREAD_MUTEX_LOCK(&proxy_mutex);
4986 fr_packet_list_walk(proxy_list, this, eol_proxy_listener);
4988 if (!fr_packet_list_socket_del(proxy_list, this->fd)) {
4989 ERROR("Fatal error removing socket %s: %s",
4990 buffer, fr_strerror());
4993 PTHREAD_MUTEX_UNLOCK(&proxy_mutex);
4997 INFO(" ... shutting down socket %s", buffer);
5000 * EOL all requests using this socket.
5002 rbtree_walk(pl, RBTREE_DELETE_ORDER, eol_listener, this);
5006 * No child threads, clean it up now.
5010 if (sock->ev) fr_event_delete(el, &sock->ev);
5016 * Wait until all requests using this socket are done.
5018 gettimeofday(&when, NULL);
5022 if (!fr_event_insert(el, listener_free_cb, this, &when,
5024 rad_panic("Failed to insert event");
5027 #endif /* WITH_TCP */
5032 /***********************************************************************
5036 ***********************************************************************/
5038 static void handle_signal_self(int flag)
5042 if ((flag & (RADIUS_SIGNAL_SELF_EXIT | RADIUS_SIGNAL_SELF_TERM)) != 0) {
5043 if ((flag & RADIUS_SIGNAL_SELF_EXIT) != 0) {
5044 INFO("Signalled to exit");
5045 fr_event_loop_exit(el, 1);
5047 INFO("Signalled to terminate");
5048 fr_event_loop_exit(el, 2);
5052 } /* else exit/term flags weren't set */
5055 * Tell the even loop to stop processing.
5057 if ((flag & RADIUS_SIGNAL_SELF_HUP) != 0) {
5059 static time_t last_hup = 0;
5062 if ((int) (when - last_hup) < 5) {
5063 INFO("Ignoring HUP (less than 5s since last one)");
5067 INFO("Received HUP signal");
5071 exec_trigger(NULL, NULL, "server.signal.hup", true);
5072 fr_event_loop_exit(el, 0x80);
5075 #if defined(WITH_DETAIL) && !defined(WITH_DETAIL_THREAD)
5076 if ((flag & RADIUS_SIGNAL_SELF_DETAIL) != 0) {
5080 * FIXME: O(N) loops suck.
5082 for (this = main_config.listen;
5084 this = this->next) {
5085 if (this->type != RAD_LISTEN_DETAIL) continue;
5088 * This one didn't send the signal, skip
5091 if (!this->decode(this, NULL)) continue;
5094 * Go service the interrupt.
5096 event_poll_detail(this);
5101 #if defined(WITH_TCP) && defined(WITH_PROXY) && defined(HAVE_PTHREAD_H)
5103 * There are new listeners in the list. Run
5104 * event_new_fd() on them.
5106 if ((flag & RADIUS_SIGNAL_SELF_NEW_FD) != 0) {
5107 rad_listen_t *this, *next;
5109 FD_MUTEX_LOCK(&fd_mutex);
5112 * FIXME: unlock the mutex before calling
5115 for (this = new_listeners; this != NULL; this = next) {
5122 new_listeners = NULL;
5123 FD_MUTEX_UNLOCK(&fd_mutex);
5128 #ifndef HAVE_PTHREAD_H
5129 void radius_signal_self(int flag)
5131 return handle_signal_self(flag);
5135 static int self_pipe[2] = { -1, -1 };
5138 * Inform ourselves that we received a signal.
5140 void radius_signal_self(int flag)
5146 * The read MUST be non-blocking for this to work.
5148 rcode = read(self_pipe[0], buffer, sizeof(buffer));
5152 for (i = 0; i < rcode; i++) {
5153 buffer[0] |= buffer[i];
5161 if (write(self_pipe[1], buffer, 1) < 0) fr_exit(0);
5165 static void event_signal_handler(UNUSED fr_event_list_t *xel,
5166 UNUSED int fd, UNUSED void *ctx)
5171 rcode = read(self_pipe[0], buffer, sizeof(buffer));
5172 if (rcode <= 0) return;
5175 * Merge pending signals.
5177 for (i = 0; i < rcode; i++) {
5178 buffer[0] |= buffer[i];
5181 handle_signal_self(buffer[0]);
5183 #endif /* HAVE_PTHREAD_H */
5185 /***********************************************************************
5187 * Bootstrapping code.
5189 ***********************************************************************/
5192 * Externally-visibly functions.
5194 int radius_event_init(TALLOC_CTX *ctx) {
5195 el = fr_event_list_create(ctx, event_status);
5201 static int packet_entry_cmp(void const *one, void const *two)
5203 RADIUS_PACKET const * const *a = one;
5204 RADIUS_PACKET const * const *b = two;
5206 return fr_packet_cmp(*a, *b);
5211 * They haven't defined a proxy listener. Automatically
5212 * add one for them, with the correct address family.
5214 static void create_default_proxy_listener(int af)
5218 listen_socket_t *sock;
5221 memset(&home, 0, sizeof(home));
5224 * Open a default UDP port
5226 home.proto = IPPROTO_UDP;
5230 * Set the address family.
5232 home.src_ipaddr.af = af;
5233 home.ipaddr.af = af;
5236 * Get the correct listener.
5238 this = proxy_new_listener(proxy_ctx, &home, port);
5244 if (!fr_packet_list_socket_add(proxy_list, this->fd,
5246 &sock->other_ipaddr, sock->other_port,
5248 ERROR("Failed adding proxy socket");
5253 * Insert the FD into list of FDs to listen on.
5255 radius_update_listener(this);
5259 * See if we automatically need to open a proxy socket.
5261 static void check_proxy(rad_listen_t *head)
5264 bool has_v4, has_v6;
5267 if (check_config) return;
5268 if (!main_config.proxy_requests) return;
5270 if (!home_servers_udp) return;
5273 * We passed "-i" on the command line. Use that address
5274 * family for the proxy socket.
5276 if (main_config.myip.af != AF_UNSPEC) {
5277 create_default_proxy_listener(main_config.myip.af);
5281 defined_proxy = has_v4 = has_v6 = false;
5284 * Figure out if we need to open a proxy socket, and if
5287 for (this = head; this != NULL; this = this->next) {
5288 listen_socket_t *sock;
5290 switch (this->type) {
5291 case RAD_LISTEN_PROXY:
5292 defined_proxy = true;
5295 case RAD_LISTEN_AUTH:
5297 case RAD_LISTEN_ACCT:
5300 case RAD_LISTEN_COA:
5303 if (sock->my_ipaddr.af == AF_INET) has_v4 = true;
5304 if (sock->my_ipaddr.af == AF_INET6) has_v6 = true;
5313 * Assume they know what they're doing.
5315 if (defined_proxy) return;
5317 if (has_v4) create_default_proxy_listener(AF_INET);
5319 if (has_v6) create_default_proxy_listener(AF_INET6);
5323 int radius_event_start(CONF_SECTION *cs, bool have_children)
5325 rad_listen_t *head = NULL;
5327 if (fr_start_time != (time_t)-1) return 0;
5329 time(&fr_start_time);
5331 if (!check_config) {
5333 * radius_event_init() must be called first
5337 pl = rbtree_create(NULL, packet_entry_cmp, NULL, 0);
5338 if (!pl) return 0; /* leak el */
5341 request_num_counter = 0;
5344 if (main_config.proxy_requests && !check_config) {
5346 * Create the tree for managing proxied requests and
5349 proxy_list = fr_packet_list_create(1);
5350 if (!proxy_list) return 0;
5352 #ifdef HAVE_PTHREAD_H
5353 if (pthread_mutex_init(&proxy_mutex, NULL) != 0) {
5354 ERROR("FATAL: Failed to initialize proxy mutex: %s",
5355 fr_syserror(errno));
5361 * The "init_delay" is set to "response_window".
5362 * Reset it to half of "response_window" in order
5363 * to give the event loop enough time to service
5364 * the event before hitting "response_window".
5366 main_config.init_delay.tv_usec += (main_config.init_delay.tv_sec & 0x01) * USEC;
5367 main_config.init_delay.tv_usec >>= 1;
5368 main_config.init_delay.tv_sec >>= 1;
5370 proxy_ctx = talloc_init("proxy");
5375 * Move all of the thread calls to this file?
5377 * It may be best for the mutexes to be in this file...
5379 spawn_flag = have_children;
5381 #ifdef HAVE_PTHREAD_H
5382 NO_SUCH_CHILD_PID = pthread_self(); /* not a child thread */
5385 * Initialize the threads ONLY if we're spawning, AND
5386 * we're running normally.
5388 if (have_children && !check_config &&
5389 (thread_pool_init(cs, &spawn_flag) < 0)) {
5395 DEBUG("%s: #### Skipping IP addresses and Ports ####",
5397 if (listen_init(cs, &head, spawn_flag) < 0) {
5404 #ifdef HAVE_PTHREAD_H
5406 * Child threads need a pipe to signal us, as do the
5409 if (pipe(self_pipe) < 0) {
5410 ERROR("Error opening internal pipe: %s", fr_syserror(errno));
5413 if ((fcntl(self_pipe[0], F_SETFL, O_NONBLOCK) < 0) ||
5414 (fcntl(self_pipe[0], F_SETFD, FD_CLOEXEC) < 0)) {
5415 ERROR("Error setting internal flags: %s", fr_syserror(errno));
5418 if ((fcntl(self_pipe[1], F_SETFL, O_NONBLOCK) < 0) ||
5419 (fcntl(self_pipe[1], F_SETFD, FD_CLOEXEC) < 0)) {
5420 ERROR("Error setting internal flags: %s", fr_syserror(errno));
5423 DEBUG4("Created signal pipe. Read end FD %i, write end FD %i", self_pipe[0], self_pipe[1]);
5425 if (!fr_event_fd_insert(el, 0, self_pipe[0], event_signal_handler, el)) {
5426 ERROR("Failed creating signal pipe handler: %s", fr_strerror());
5431 DEBUG("%s: #### Opening IP addresses and Ports ####", main_config.name);
5434 * The server temporarily switches to an unprivileged
5435 * user very early in the bootstrapping process.
5436 * However, some sockets MAY require privileged access
5437 * (bind to device, or to port < 1024, or to raw
5438 * sockets). Those sockets need to call suid up/down
5439 * themselves around the functions that need a privileged
5442 if (listen_init(cs, &head, spawn_flag) < 0) {
5446 main_config.listen = head;
5453 * At this point, no one has any business *ever* going
5456 rad_suid_down_permanent();
5463 static int proxy_delete_cb(UNUSED void *ctx, void *data)
5465 REQUEST *request = fr_packet2myptr(REQUEST, proxy, data);
5467 VERIFY_REQUEST(request);
5469 request->master_state = REQUEST_STOP_PROCESSING;
5471 #ifdef HAVE_PTHREAD_H
5472 if (pthread_equal(request->child_pid, NO_SUCH_CHILD_PID) == 0) return 0;
5476 * If it's queued we can't delete it from the queue.
5478 * Otherwise, it's OK to delete it. Even RUNNING, because
5479 * that will get caught by the check above.
5481 if (request->child_state == REQUEST_QUEUED) return 0;
5483 request->in_proxy_hash = false;
5485 if (!request->in_request_hash) {
5486 request_done(request, FR_ACTION_DONE);
5490 * Delete it from the list.
5497 static int request_delete_cb(UNUSED void *ctx, void *data)
5499 REQUEST *request = fr_packet2myptr(REQUEST, packet, data);
5501 VERIFY_REQUEST(request);
5503 request->master_state = REQUEST_STOP_PROCESSING;
5506 * Not done, or the child thread is still processing it.
5508 if (request->child_state < REQUEST_RESPONSE_DELAY) return 0; /* continue */
5510 #ifdef HAVE_PTHREAD_H
5511 if (pthread_equal(request->child_pid, NO_SUCH_CHILD_PID) == 0) return 0;
5515 rad_assert(request->in_proxy_hash == false);
5518 request->in_request_hash = false;
5520 if (request->ev) fr_event_delete(el, &request->ev);
5522 if (main_config.memory_report) {
5523 RDEBUG2("Cleaning up request packet ID %u with timestamp +%d",
5524 request->packet->id,
5525 (unsigned int) (request->timestamp - fr_start_time));
5530 rad_assert(!request->coa->in_proxy_hash);
5534 request_free(request);
5537 * Delete it from the list, and continue;
5543 void radius_event_free(void)
5549 * There are requests in the proxy hash that aren't
5550 * referenced from anywhere else. Remove them first.
5553 fr_packet_list_walk(proxy_list, NULL, proxy_delete_cb);
5557 rbtree_walk(pl, RBTREE_DELETE_ORDER, request_delete_cb, NULL);
5561 * Now that all requests have been marked "please stop",
5562 * ensure that all of the threads have exited.
5564 #ifdef HAVE_PTHREAD_H
5569 * Walk the lists again, ensuring that all
5570 * requests are done.
5572 if (main_config.memory_report) {
5577 fr_packet_list_walk(proxy_list, NULL, proxy_delete_cb);
5578 num = fr_packet_list_num_elements(proxy_list);
5580 ERROR("Proxy list has %d requests still in it.", num);
5585 rbtree_walk(pl, RBTREE_DELETE_ORDER, request_delete_cb, NULL);
5586 num = rbtree_num_elements(pl);
5588 ERROR("Request list has %d requests still in it.", num);
5597 fr_packet_list_free(proxy_list);
5600 if (proxy_ctx) talloc_free(proxy_ctx);
5605 if (debug_condition) talloc_free(debug_condition);
5608 int radius_event_process(void)
5612 return fr_event_loop(el);