X-Git-Url: http://www.project-moonshot.org/gitweb/?p=trust_router.git;a=blobdiff_plain;f=tr%2Ftr_tid.c;h=9d32edcaf307da52af469779587db69b5d0d4f42;hp=3e1577e5fd2db7b3350083dc8ef293b69fc2bc9c;hb=HEAD;hpb=0705cc5174a47715ade9e2b91b051d6c067f0bd5 diff --git a/tr/tr_tid.c b/tr/tr_tid.c index 3e1577e..9d32edc 100644 --- a/tr/tr_tid.c +++ b/tr/tr_tid.c @@ -1,21 +1,61 @@ +/* + * Copyright (c) 2016, JANET(UK) + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * 3. Neither the name of JANET(UK) nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED + * OF THE POSSIBILITY OF SUCH DAMAGE. + * + */ + #include +#include #include #include #include #include #include +#include #include #include #include +#include #include #include +#include +#include #include +#include -/* Structure to hold TR instance and original request in one cookie */ +/* Structure to hold data for the tid response callback */ typedef struct tr_resp_cookie { - TIDS_INSTANCE *tids; - TID_REQ *orig_req; + int thread_id; + TID_RESP *resp; } TR_RESP_COOKIE; /* hold a tids instance and a config manager */ @@ -25,43 +65,292 @@ struct tr_tids_event_cookie { TRPS_INSTANCE *trps; }; +static void tr_tidc_resp_handler(TIDC_INSTANCE *tidc, + TID_REQ *req, + TID_RESP *resp, + void *resp_cookie) +{ + TR_RESP_COOKIE *cookie=talloc_get_type_abort(resp_cookie, TR_RESP_COOKIE); + + tr_debug("tr_tidc_resp_handler: Response received! Realm = %s, Community = %s, result = %s.", + resp->realm->buf, + resp->comm->buf, + (TID_SUCCESS==resp->result)?"success":"error"); + + if (resp->error_path!=NULL) + tr_debug("tr_tids_resp_handler: error_path is set."); + cookie->resp=tid_resp_dup(cookie, resp); +} + +/* data for AAA req forwarding threads */ +struct tr_tids_fwd_cookie { + int thread_id; + pthread_mutex_t mutex; /* lock on the mq (separate from the locking within the mq, see below) */ + TR_MQ *mq; /* messages from thread to main process; set to NULL to disable response */ + TR_AAA_SERVER *aaa; /* AAA server to contact */ + DH *dh_params; + TID_REQ *fwd_req; /* the req to duplicate */ +}; + +static int tr_tids_fwd_cookie_destructor(void *obj) +{ + struct tr_tids_fwd_cookie *c=talloc_get_type_abort(obj, struct tr_tids_fwd_cookie); + if (c->dh_params!=NULL) + tr_destroy_dh_params(c->dh_params); + return 0; +} + +/* Block until we get the lock, returns 0 on success. + * The mutex is used to protect changes to the mq pointer in + * a thread's cookie. The master thread sets this to null to indicate + * that it has abandoned the thread and the message queue is no longer + * valid. This is unrelated to the locking in the message queue + * implementation itself. */ +static int tr_tids_fwd_get_mutex(struct tr_tids_fwd_cookie *cookie) +{ + if (cookie==NULL) + return -1; + + return (pthread_mutex_lock(&(cookie->mutex))); +} + +static int tr_tids_fwd_release_mutex(struct tr_tids_fwd_cookie *cookie) +{ + if (cookie==NULL) + return -1; + + return (pthread_mutex_unlock(&(cookie->mutex))); +} + +/* values for messages */ +#define TR_TID_MQMSG_SUCCESS "tid success" +#define TR_TID_MQMSG_FAILURE "tid failure" + +/* Thread main for sending and receiving a request to a single AAA server */ +static void *tr_tids_req_fwd_thread(void *arg) +{ + TALLOC_CTX *tmp_ctx=talloc_new(NULL); + struct tr_tids_fwd_cookie *args=talloc_get_type_abort(arg, struct tr_tids_fwd_cookie); + TIDC_INSTANCE *tidc=tidc_create(); + TR_MQ_MSG *msg=NULL; + TR_RESP_COOKIE *cookie=NULL; + char *aaa_hostname = NULL; + int aaa_port; + int rc=0; + int success=0; + + talloc_steal(tmp_ctx, args); /* take responsibility for the cookie */ + + if (tidc!=NULL) + talloc_steal(tmp_ctx, tidc); + + /* create the cookie we will use for our response */ + cookie=talloc(tmp_ctx, TR_RESP_COOKIE); + if (cookie==NULL) { + tr_crit("tr_tids_req_fwd_thread: unable to allocate response cookie."); + success=0; + goto cleanup; + } + cookie->thread_id=args->thread_id; + tr_debug("tr_tids_req_fwd_thread: thread %d started.", cookie->thread_id); + + /* Create a TID client instance */ + if (tidc==NULL) { + tr_crit("tr_tids_req_fwd_thread: Unable to allocate TIDC instance."); + /*tids_send_err_response(tids, orig_req, "Memory allocation failure");*/ + /* TODO: encode reason for failure */ + success=0; + goto cleanup; + } + + /* Set-up TID connection */ + aaa_hostname = tr_name_strdup(tr_aaa_server_get_hostname(args->aaa)); + if (aaa_hostname == NULL) { + tr_crit("tr_tids_req_fwd_thread: unable to allocate AAA hostname string"); + success=0; + goto cleanup; + } + aaa_port = tr_aaa_server_get_port(args->aaa); + if ((aaa_port <= 0) || (aaa_port > 65535)) { + tr_notice("tr_tids_req_fwd_thread: invalid port (%d) for %s", aaa_port, aaa_hostname); + success=0; + goto cleanup; + } + + if (-1==(args->fwd_req->conn = tidc_open_connection(tidc, + aaa_hostname, + aaa_port, + &(args->fwd_req->gssctx)))) { + tr_notice("tr_tids_req_fwd_thread: Error in tidc_open_connection."); + /* tids_send_err_response(tids, orig_req, "Can't open connection to next hop TIDS"); */ + /* TODO: encode reason for failure */ + success=0; + goto cleanup; + }; + tr_debug("tr_tids_req_fwd_thread: thread %d opened TID connection to %s:%d.", + cookie->thread_id, aaa_hostname, aaa_port); + + /* Send a TID request. */ + if (0 > (rc = tidc_fwd_request(tidc, args->fwd_req, tr_tidc_resp_handler, (void *)cookie))) { + tr_notice("Error from tidc_fwd_request, rc = %d.", rc); + success=0; + goto cleanup; + } + /* cookie->resp should now contain our copy of the response */ + success=1; + tr_debug("tr_tids_req_fwd_thread: thread %d received response.", cookie->thread_id); -static void tr_tidc_resp_handler (TIDC_INSTANCE *tidc, - TID_REQ *req, - TID_RESP *resp, - void *resp_cookie) +cleanup: + /* Notify parent thread of the response, if it's still listening. */ + if (0!=tr_tids_fwd_get_mutex(args)) { + tr_notice("tr_tids_req_fwd_thread: thread %d unable to acquire mutex.", cookie->thread_id); + } else if (NULL!=args->mq) { + /* mq is still valid, so we can queue our response */ + tr_debug("tr_tids_req_fwd_thread: thread %d using valid msg queue.", cookie->thread_id); + if (success) + msg= tr_mq_msg_new(tmp_ctx, TR_TID_MQMSG_SUCCESS); + else + msg= tr_mq_msg_new(tmp_ctx, TR_TID_MQMSG_FAILURE); + + if (msg==NULL) + tr_notice("tr_tids_req_fwd_thread: thread %d unable to allocate response msg.", cookie->thread_id); + + tr_mq_msg_set_payload(msg, (void *)cookie, NULL); + if (NULL!=cookie) + talloc_steal(msg, cookie); /* attach this to the msg so we can forget about it */ + tr_mq_add(args->mq, msg); + talloc_steal(NULL, args); /* take out of our tmp_ctx; master thread now responsible for freeing */ + tr_debug("tr_tids_req_fwd_thread: thread %d queued response message.", cookie->thread_id); + if (0!=tr_tids_fwd_release_mutex(args)) + tr_notice("tr_tids_req_fwd_thread: Error releasing mutex."); + } + + if (aaa_hostname != NULL) + free(aaa_hostname); + + talloc_free(tmp_ctx); + return NULL; +} + +/* Merges r2 into r1 if they are compatible. */ +static TID_RC tr_tids_merge_resps(TID_RESP *r1, TID_RESP *r2) { - tr_debug("tr_tidc_resp_handler: Response received (conn = %d)! Realm = %s, Community = %s.", ((TR_RESP_COOKIE *)resp_cookie)->orig_req->conn, resp->realm->buf, resp->comm->buf); - req->resp_rcvd = 1; - - /* TBD -- handle concatentation of multiple responses to single req */ - tids_send_response(((TR_RESP_COOKIE *)resp_cookie)->tids, - ((TR_RESP_COOKIE *)resp_cookie)->orig_req, - resp); - - return; + /* ensure these are compatible replies */ + if ((r1->result!=TID_SUCCESS) || (r2->result!=TID_SUCCESS)) + return TID_ERROR; + + if ((0 == tr_name_cmp(r1->rp_realm, r2->rp_realm)) + && (0 == tr_name_cmp(r1->realm, r2->realm)) + && ( (0 == tr_name_cmp(r1->comm, r2->comm)) + || (0 == tr_name_cmp(r1->comm, r2->orig_coi)) + || (0 == tr_name_cmp(r1->orig_coi, r2->comm)))) { + + tid_srvr_blk_add(r1->servers, tid_srvr_blk_dup(r1, r2->servers)); + return TID_SUCCESS; + } + + return TID_ERROR; } -static int tr_tids_req_handler (TIDS_INSTANCE *tids, - TID_REQ *orig_req, - TID_RESP *resp, - void *cookie_in) +enum map_coi_result { + MAP_COI_SUCCESS = 0, + MAP_COI_MAP_NOT_REQUIRED, + MAP_COI_ALREADY_MAPPED, + MAP_COI_NO_APC, + MAP_COI_INVALID_APC, + MAP_COI_UNKNOWN_COMM, + MAP_COI_ERROR +}; + +static enum map_coi_result map_coi(TR_COMM_TABLE *ctable, TID_REQ *req) +{ + TR_COMM *orig_comm; + TR_NAME *apc_name; + TR_COMM *apc; + TR_APC *apcs; + + if (tid_req_get_orig_coi(req) != NULL) + return MAP_COI_ALREADY_MAPPED; + + /* look up the community */ + orig_comm = tr_comm_table_find_comm(ctable, tid_req_get_comm(req)); + if (orig_comm == NULL) + return MAP_COI_UNKNOWN_COMM; + + if (tr_comm_get_type(orig_comm) == TR_COMM_APC) + return MAP_COI_MAP_NOT_REQUIRED; /* it was already an APC, no mapping to do */ + + /* use first (only) APC. These are just APC names */ + apcs = tr_comm_get_apcs(orig_comm); + if ((!apcs) || (!tr_apc_get_id(apcs))) + return MAP_COI_NO_APC; + + /* get our own copy of the APC name */ + apc_name = tr_dup_name(tr_apc_get_id(apcs)); + if (apc_name == NULL) { + tr_err("map_coi: Error allocating apc_name"); + return MAP_COI_ERROR; + } + + /* Check that the APC is configured */ + apc = tr_comm_table_find_comm(ctable, apc_name); + if (apc == NULL) { + tr_free_name(apc_name); + return MAP_COI_INVALID_APC; + } + + tid_req_set_orig_coi(req, tid_req_get_comm(req)); /* was null, so no need to free anything */ + tid_req_set_comm(req, apc_name); /* original contents will be freed via orig_coi */ + + return MAP_COI_SUCCESS; /* successfully mapped */ +} + +/** + * Process a TID request + * + * Return value of -1 means to send a TID_ERROR response. Fill in resp->err_msg or it will + * be returned as a generic error. + * + * @param tids + * @param orig_req + * @param resp + * @param cookie_in + * @return + */ +static int tr_tids_req_handler(TIDS_INSTANCE *tids, + TID_REQ *orig_req, + TID_RESP *resp, + void *cookie_in) { - TALLOC_CTX *tmp_ctx=talloc_new(NULL);; - TIDC_INSTANCE *tidc = NULL; - TR_RESP_COOKIE resp_cookie; - TR_AAA_SERVER *aaa_servers = NULL; - TR_NAME *apc = NULL; + TALLOC_CTX *tmp_ctx=talloc_new(NULL); + TR_AAA_SERVER *aaa_servers=NULL, *this_aaa=NULL; + int n_aaa=0; + int idp_shared=0; + TR_AAA_SERVER_ITER *aaa_iter=NULL; + pthread_t aaa_thread[TR_TID_MAX_AAA_SERVERS]; + struct tr_tids_fwd_cookie *aaa_cookie[TR_TID_MAX_AAA_SERVERS]={NULL}; + TID_RESP *aaa_resp[TR_TID_MAX_AAA_SERVERS]={NULL}; + TR_RP_CLIENT *rp_client=NULL; + TR_RP_CLIENT_ITER *rpc_iter=NULL; TID_REQ *fwd_req = NULL; TR_COMM *cfg_comm = NULL; TR_COMM *cfg_apc = NULL; - int oaction = TR_FILTER_ACTION_REJECT; - int rc = 0; + TR_FILTER_ACTION oaction = TR_FILTER_ACTION_REJECT; time_t expiration_interval=0; struct tr_tids_event_cookie *cookie=talloc_get_type_abort(cookie_in, struct tr_tids_event_cookie); TR_CFG_MGR *cfg_mgr=cookie->cfg_mgr; TRPS_INSTANCE *trps=cookie->trps; TRP_ROUTE *route=NULL; + TR_MQ *mq=NULL; + TR_MQ_MSG *msg=NULL; + unsigned int n_responses=0; + unsigned int n_failed=0; + struct timespec ts_abort={0}; + unsigned int resp_frac_numer=cfg_mgr->active->internal->tid_resp_numer; + unsigned int resp_frac_denom=cfg_mgr->active->internal->tid_resp_denom; + TR_RESP_COOKIE *payload=NULL; + TR_FILTER_TARGET *target=NULL; + int ii=0; int retval=-1; if ((!tids) || (!orig_req) || (!resp)) { @@ -72,174 +361,387 @@ static int tr_tids_req_handler (TIDS_INSTANCE *tids, tr_debug("tr_tids_req_handler: Request received (conn = %d)! Realm = %s, Comm = %s", orig_req->conn, orig_req->realm->buf, orig_req->comm->buf); + if (orig_req->request_id) + tr_debug("tr_tids_req_handler: TID request ID: %.*s", orig_req->request_id->len, orig_req->request_id->buf); + else + tr_debug("tr_tids_req_handler: TID request ID: none"); + tids->req_count++; /* Duplicate the request, so we can modify and forward it */ - if (NULL == (fwd_req = tid_dup_req(orig_req))) { + if (NULL == (fwd_req=tid_dup_req(orig_req))) { tr_debug("tr_tids_req_handler: Unable to duplicate request."); - retval=-1; + retval=-1; /* response will be a generic internal error */ goto cleanup; } + talloc_steal(tmp_ctx, fwd_req); - if (NULL == (cfg_comm = tr_comm_lookup(cfg_mgr->active->comms, orig_req->comm))) { + /* cfg_comm is now the community (APC or CoI) of the incoming request */ + if (NULL == (cfg_comm=tr_comm_table_find_comm(cfg_mgr->active->ctable, orig_req->comm))) { tr_notice("tr_tids_req_hander: Request for unknown comm: %s.", orig_req->comm->buf); - tids_send_err_response(tids, orig_req, "Unknown community"); + tid_resp_set_err_msg(resp, tr_new_name("Unknown community")); retval=-1; goto cleanup; } - /* Check that the rp_realm matches the filter for the GSS name that - * was received. */ + /* We now need to apply the filters associated with the RP client handing us the request. + * It is possible (or even likely) that more than one client is associated with the GSS + * name we got from the authentication. We will apply all of them in an arbitrary order. + * For this to result in well-defined behavior, either only accept or only reject filter + * lines should be used, or a unique GSS name must be given for each RP realm. */ - if ((!tids->rp_gss) || - (!tids->rp_gss->filter)) { + if (!tids->gss_name) { tr_notice("tr_tids_req_handler: No GSS name for incoming request."); - tids_send_err_response(tids, orig_req, "No GSS name for request"); + tid_resp_set_err_msg(resp, tr_new_name("No GSS name for request")); retval=-1; goto cleanup; } - if ((TR_FILTER_NO_MATCH == tr_filter_process_rp_permitted(orig_req->rp_realm, - tids->rp_gss->filter, - orig_req->cons, - &fwd_req->cons, - &oaction)) || - (TR_FILTER_ACTION_REJECT == oaction)) { - tr_notice("tr_tids_req_handler: RP realm (%s) does not match RP Realm filter for GSS name", orig_req->rp_realm->buf); - tids_send_err_response(tids, orig_req, "RP Realm filter error"); + /* Keep original constraints, may add more from the filter. These will be added to orig_req as + * well. Need to verify that this is acceptable behavior, but it's what we've always done. */ + fwd_req->cons=orig_req->cons; + + target=tr_filter_target_tid_req(tmp_ctx, orig_req); + if (target==NULL) { + tr_crit("tid_req_handler: Unable to allocate filter target, cannot apply filter!"); + tid_resp_set_err_msg(resp, tr_new_name("Incoming TID request filter error")); retval=-1; goto cleanup; } + + rpc_iter=tr_rp_client_iter_new(tmp_ctx); + if (rpc_iter==NULL) { + tr_err("tid_req_handler: Unable to allocate RP client iterator."); + retval=-1; + goto cleanup; + } + for (rp_client=tr_rp_client_iter_first(rpc_iter, cfg_mgr->active->rp_clients); + rp_client != NULL; + rp_client=tr_rp_client_iter_next(rpc_iter)) { + + if (!tr_gss_names_matches(rp_client->gss_names, tids->gss_name)) + continue; /* skip any that don't match the GSS name */ + + if (TR_FILTER_MATCH == tr_filter_apply(target, + tr_filter_set_get(rp_client->filters, + TR_FILTER_TYPE_TID_INBOUND), + &(fwd_req->cons), + &oaction)) + break; /* Stop looking, oaction is set */ + } + + /* We get here whether or not a filter matched. If tr_filter_apply() doesn't match, it returns + * a default action of reject, so we don't have to check why we exited the loop. */ + if (oaction != TR_FILTER_ACTION_ACCEPT) { + tr_notice("tr_tids_req_handler: Incoming TID request rejected by RP client filter for GSS name %.*s", + tids->gss_name->len, tids->gss_name->buf); + tid_resp_set_err_msg(resp, tr_new_name("Incoming TID request filter error")); + retval = -1; + goto cleanup; + } + /* Check that the rp_realm is a member of the community in the request */ - if (NULL == (tr_find_comm_rp(cfg_comm, orig_req->rp_realm))) { - tr_notice("tr_tids_req_handler: RP Realm (%s) not member of community (%s).", orig_req->rp_realm->buf, orig_req->comm->buf); - tids_send_err_response(tids, orig_req, "RP COI membership error"); + if (NULL == tr_comm_find_rp(cfg_mgr->active->ctable, cfg_comm, orig_req->rp_realm)) { + tr_notice("tr_tids_req_handler: RP Realm (%s) not member of community (%s).", + orig_req->rp_realm->buf, orig_req->comm->buf); + tid_resp_set_err_msg(resp, tr_new_name("RP community membership error")); retval=-1; goto cleanup; } - /* Map the comm in the request from a COI to an APC, if needed */ - if (TR_COMM_COI == cfg_comm->type) { - tr_debug("tr_tids_req_handler: Community was a COI, switching."); - /* TBD -- In theory there can be more than one? How would that work? */ - if ((!cfg_comm->apcs) || (!cfg_comm->apcs->id)) { - tr_notice("No valid APC for COI %s.", orig_req->comm->buf); - tids_send_err_response(tids, orig_req, "No valid APC for community"); - retval=-1; + switch(map_coi(cfg_mgr->active->ctable, fwd_req)) { + case MAP_COI_MAP_NOT_REQUIRED: + cfg_apc = cfg_comm; + break; + + case MAP_COI_SUCCESS: + cfg_apc = tr_comm_table_find_comm(cfg_mgr->active->ctable, tid_req_get_comm(fwd_req)); + tr_debug("tr_tids_req_handler: Community %.*s is a COI, mapping to APC %.*s.", + tid_req_get_orig_coi(fwd_req)->len, tid_req_get_orig_coi(fwd_req)->buf, + tr_comm_get_id(cfg_apc)->len, tr_comm_get_id(cfg_apc)->buf); + break; + + case MAP_COI_ALREADY_MAPPED: + tr_notice("tr_tids_req_handler: community %.*s is COI but COI to APC mapping already occurred. Dropping request.", + tid_req_get_comm(orig_req)->len, tid_req_get_comm(orig_req)->buf); + tid_resp_set_err_msg(resp, tr_new_name("Second COI to APC mapping would result, permitted only once.")); + retval = -1; goto cleanup; - } - apc = tr_dup_name(cfg_comm->apcs->id); - /* Check that the APC is configured */ - if (NULL == (cfg_apc = tr_comm_lookup(cfg_mgr->active->comms, apc))) { - tr_notice("tr_tids_req_hander: Request for unknown comm: %s.", apc->buf); - tids_send_err_response(tids, orig_req, "Unknown APC"); - retval=-1; + case MAP_COI_NO_APC: + tr_notice("No valid APC for COI %.*s.", + tid_req_get_comm(orig_req)->len, tid_req_get_comm(orig_req)->buf); + tid_resp_set_err_msg(resp, tr_new_name("No valid APC for community")); + retval = -1; goto cleanup; - } - fwd_req->comm = apc; - fwd_req->orig_coi = orig_req->comm; + case MAP_COI_INVALID_APC: + tr_notice("tr_tids_req_hander: Request for unknown APC."); + tid_resp_set_err_msg(resp, tr_new_name("Unknown APC")); + retval = -1; + goto cleanup; - /* Check that rp_realm is a member of this APC */ - if (NULL == (tr_find_comm_rp(cfg_apc, orig_req->rp_realm))) { - tr_notice("tr_tids_req_hander: RP Realm (%s) not member of community (%s).", orig_req->rp_realm->buf, orig_req->comm->buf); - tids_send_err_response(tids, orig_req, "RP APC membership error"); - retval=-1; + default: + tr_notice("tr_tids_req_hander: Unexpected error mapping COI to APC."); + retval = -1; goto cleanup; - } } - /* Look up the route for this community/realm. */ - tr_debug("tr_tids_req_handler: looking up route."); - route=trps_get_selected_route(trps, orig_req->comm, orig_req->realm); - if (route==NULL) { - tr_notice("tr_tids_req_handler: no route table entry found for realm (%s) in community (%s).", - orig_req->realm->buf, orig_req->comm->buf); - tids_send_err_response(tids, orig_req, "Missing trust route error"); + /* cfg_comm is now the original community, and cfg_apc is the APC it belongs to. These + * may both be the same. If not, check that rp_realm is a member of the mapped APC */ + if ((cfg_apc != cfg_comm) + && (NULL == tr_comm_find_rp(cfg_mgr->active->ctable, + cfg_apc, + tid_req_get_rp_realm(fwd_req)))) { + tr_notice("tr_tids_req_hander: RP Realm (%.*s) not member of mapped APC (%.*s).", + tid_req_get_rp_realm(fwd_req)->len, tid_req_get_rp_realm(fwd_req)->buf, + tr_comm_get_id(cfg_apc)->len, tr_comm_get_id(cfg_apc)->buf); + tid_resp_set_err_msg(resp, tr_new_name("RP community membership error")); retval=-1; goto cleanup; } - tr_debug("tr_tids_req_handler: found route."); - if (trp_route_is_local(route)) { - tr_debug("tr_tids_req_handler: route is local."); - aaa_servers = tr_idp_aaa_server_lookup(cfg_mgr->active->idp_realms, - orig_req->realm, - orig_req->comm); - } else { - tr_debug("tr_tids_req_handler: route not local."); - aaa_servers = tr_aaa_server_new(tmp_ctx, trp_route_get_next_hop(route)); - } - /* Find the AAA server(s) for this request */ - if (NULL == aaa_servers) { - tr_debug("tr_tids_req_handler: No AAA Servers for realm %s, defaulting.", orig_req->realm->buf); - if (NULL == (aaa_servers = tr_default_server_lookup (cfg_mgr->active->default_servers, - orig_req->comm))) { + /* Look up the route for forwarding request's community/realm. */ + tr_debug("tr_tids_req_handler: looking up route."); + route=trps_get_selected_route(trps, fwd_req->comm, fwd_req->realm); + if (route==NULL) { + /* No route. Use default AAA servers if we have them. */ + tr_debug("tr_tids_req_handler: No route for realm %s, defaulting.", fwd_req->realm->buf); + if (NULL == (aaa_servers = tr_default_server_lookup(cfg_mgr->active->default_servers, + fwd_req->comm))) { tr_notice("tr_tids_req_handler: No default AAA servers, discarded."); - tids_send_err_response(tids, orig_req, "No path to AAA Server(s) for realm"); - retval=-1; + tid_resp_set_err_msg(resp, tr_new_name("No path to AAA Server(s) for realm")); + retval = -1; goto cleanup; } + idp_shared = 0; } else { - /* if we aren't defaulting, check idp coi and apc membership */ - if (NULL == (tr_find_comm_idp(cfg_comm, fwd_req->realm))) { - tr_notice("tr_tids_req_handler: IDP Realm (%s) not member of community (%s).", orig_req->realm->buf, orig_req->comm->buf); - tids_send_err_response(tids, orig_req, "IDP community membership error"); + /* Found a route. Determine the AAA servers or next hop address for the request we are forwarding. */ + tr_debug("tr_tids_req_handler: found route."); + if (trp_route_is_local(route)) { + tr_debug("tr_tids_req_handler: route is local."); + aaa_servers = tr_idp_aaa_server_lookup(cfg_mgr->active->ctable->idp_realms, + fwd_req->realm, + fwd_req->comm, + &idp_shared); + } else { + tr_debug("tr_tids_req_handler: route not local."); + aaa_servers = tr_aaa_server_new(tmp_ctx); /* cleaned up via talloc */ + if (aaa_servers == NULL) { + tr_err("tr_tids_req_handler: error allocating next hop"); + retval=-1; + goto cleanup; + } + tr_aaa_server_set_hostname(aaa_servers, trp_route_dup_next_hop(route)); + if (tr_aaa_server_get_hostname(aaa_servers) == NULL) { + tr_err("tr_tids_req_handler: error allocating next hop"); + retval=-1; + goto cleanup; + } + tr_aaa_server_set_port(aaa_servers, trp_route_get_next_hop_port(route)); + idp_shared = 0; + } + + /* Since we aren't defaulting, check idp coi and apc membership of the original request */ + if (NULL == (tr_comm_find_idp(cfg_mgr->active->ctable, cfg_comm, orig_req->realm))) { + tr_notice("tr_tids_req_handler: IDP Realm (%s) not member of community (%s).", orig_req->realm->buf, cfg_comm->id->buf); + tid_resp_set_err_msg(resp, tr_new_name("IDP community membership error")); retval=-1; goto cleanup; } - if ( cfg_apc && (NULL == (tr_find_comm_idp(cfg_apc, fwd_req->realm)))) { - tr_notice("tr_tids_req_handler: IDP Realm (%s) not member of APC (%s).", orig_req->realm->buf, orig_req->comm->buf); - tids_send_err_response(tids, orig_req, "IDP APC membership error"); + if ( cfg_apc && (NULL == (tr_comm_find_idp(cfg_mgr->active->ctable, cfg_apc, orig_req->realm)))) { + tr_notice("tr_tids_req_handler: IDP Realm (%s) not member of APC (%s).", orig_req->realm->buf, cfg_apc->id->buf); + tid_resp_set_err_msg(resp, tr_new_name("IDP APC membership error")); retval=-1; goto cleanup; } } - /* send a TID request to the AAA server(s), and get the answer(s) */ - /* TBD -- Handle multiple servers */ + /* Make sure we came through with a AAA server. If not, we can't handle the request. + * Report using the original request, not translated values. */ + if (NULL == aaa_servers) { + tr_notice("tr_tids_req_handler: no route or AAA server for realm (%s) in community (%s).", + orig_req->realm->buf, orig_req->comm->buf); + tid_resp_set_err_msg(resp, tr_new_name("Missing trust route error")); + retval = -1; + goto cleanup; + } - if (cfg_apc) - expiration_interval = cfg_apc->expiration_interval; - else expiration_interval = cfg_comm->expiration_interval; + /* send a TID request to the AAA server(s), and get the answer(s) */ + tr_debug("tr_tids_req_handler: sending TID request(s)."); + /* Use the smaller of the APC's expiration interval and the expiration interval of the incoming request */ + expiration_interval = cfg_apc->expiration_interval; if (fwd_req->expiration_interval) fwd_req->expiration_interval = (expiration_interval < fwd_req->expiration_interval) ? expiration_interval : fwd_req->expiration_interval; - else fwd_req->expiration_interval = expiration_interval; - /* Create a TID client instance */ - if (NULL == (tidc = tidc_create())) { - tr_crit("tr_tids_req_hander: Unable to allocate TIDC instance."); - tids_send_err_response(tids, orig_req, "Memory allocation failure"); + else + fwd_req->expiration_interval = expiration_interval; + + /* Set up message queue for replies from req forwarding threads */ + mq=tr_mq_new(tmp_ctx); + if (mq==NULL) { + tr_notice("tr_tids_req_handler: unable to allocate message queue."); retval=-1; goto cleanup; } - /* Use the DH parameters from the original request */ - /* TBD -- this needs to be fixed when we handle more than one req per conn */ - tidc->client_dh = orig_req->tidc_dh; - - /* Save information about this request for the response */ - resp_cookie.tids = tids; - resp_cookie.orig_req = orig_req; + tr_debug("tr_tids_req_handler: message queue allocated."); - /* Set-up TID connection */ - if (-1 == (fwd_req->conn = tidc_open_connection(tidc, - aaa_servers->hostname->buf, - TID_PORT, - &(fwd_req->gssctx)))) { - tr_notice("tr_tids_req_handler: Error in tidc_open_connection."); - tids_send_err_response(tids, orig_req, "Can't open connection to next hop TIDS"); + /* start threads */ + aaa_iter=tr_aaa_server_iter_new(tmp_ctx); + if (aaa_iter==NULL) { + tr_notice("tr_tids_req_handler: unable to allocate AAA server iterator."); retval=-1; goto cleanup; - }; + } + for (n_aaa=0, this_aaa=tr_aaa_server_iter_first(aaa_iter, aaa_servers); + this_aaa!=NULL; + n_aaa++, this_aaa=tr_aaa_server_iter_next(aaa_iter)) { + tr_debug("tr_tids_req_handler: Preparing to start thread %d.", n_aaa); + + aaa_cookie[n_aaa]=talloc(tmp_ctx, struct tr_tids_fwd_cookie); + if (aaa_cookie[n_aaa]==NULL) { + tr_notice("tr_tids_req_handler: unable to allocate cookie for AAA thread %d.", n_aaa); + retval=-1; + goto cleanup; + } + talloc_set_destructor((void *)(aaa_cookie[n_aaa]), tr_tids_fwd_cookie_destructor); + /* fill in the cookie. To ensure the thread has valid data even if we exit first and + * abandon it, duplicate anything pointed to (except the mq). */ + aaa_cookie[n_aaa]->thread_id=n_aaa; + if (0!=pthread_mutex_init(&(aaa_cookie[n_aaa]->mutex), NULL)) { + tr_notice("tr_tids_req_handler: unable to init mutex for AAA thread %d.", n_aaa); + retval=-1; + goto cleanup; + } + aaa_cookie[n_aaa]->mq=mq; + aaa_cookie[n_aaa]->aaa=this_aaa; + aaa_cookie[n_aaa]->dh_params=tr_dh_dup(orig_req->tidc_dh); + aaa_cookie[n_aaa]->fwd_req=tid_dup_req(fwd_req); + talloc_steal(aaa_cookie[n_aaa], aaa_cookie[n_aaa]->fwd_req); + tr_debug("tr_tids_req_handler: cookie %d initialized.", n_aaa); + + /* Take the cookie out of tmp_ctx before starting thread. If thread starts, it becomes + * responsible for freeing it until it queues a response. If we did not do this, the possibility + * exists that this function exits, freeing the cookie, before the thread takes the cookie + * out of our tmp_ctx. This would cause a segfault or talloc error in the thread. */ + talloc_steal(NULL, aaa_cookie[n_aaa]); + if (0!=pthread_create(&(aaa_thread[n_aaa]), NULL, tr_tids_req_fwd_thread, aaa_cookie[n_aaa])) { + talloc_steal(tmp_ctx, aaa_cookie[n_aaa]); /* thread start failed; steal this back */ + tr_notice("tr_tids_req_handler: unable to start AAA thread %d.", n_aaa); + retval=-1; + goto cleanup; + } + tr_debug("tr_tids_req_handler: thread %d started.", n_aaa); + } - /* Send a TID request */ - if (0 > (rc = tidc_fwd_request(tidc, fwd_req, &tr_tidc_resp_handler, (void *)&resp_cookie))) { - tr_notice("Error from tidc_fwd_request, rc = %d.", rc); - tids_send_err_response(tids, orig_req, "Can't forward request to next hop TIDS"); + /* determine expiration time */ + if (0!=tr_mq_pop_timeout(cfg_mgr->active->internal->tid_req_timeout, &ts_abort)) { + tr_notice("tr_tids_req_handler: unable to read clock for timeout."); retval=-1; goto cleanup; } + /* wait for responses */ + tr_debug("tr_tids_req_handler: waiting for response(s)."); + n_responses=0; + n_failed=0; + while (((n_responses+n_failed)thread_id]=payload->resp; /* save pointers to these */ + + if (payload->resp->result==TID_SUCCESS) { + tr_tids_merge_resps(resp, payload->resp); + n_responses++; + } else { + n_failed++; + tr_notice("tr_tids_req_handler: TID error received from AAA server %d: %.*s", + payload->thread_id, + payload->resp->err_msg->len, + payload->resp->err_msg->buf); + } + } else if (0==strcmp(tr_mq_msg_get_message(msg), TR_TID_MQMSG_FAILURE)) { + /* failure */ + n_failed++; + payload=talloc_get_type(tr_mq_msg_get_payload(msg), TR_RESP_COOKIE); + if (payload!=NULL) + talloc_steal(tmp_ctx, payload); /* put this back in our context */ + else { + /* this means the thread was unable to allocate a response cookie, and we thus cannot determine which thread it was. This is bad and should never happen in a working system.. Give up. */ + tr_notice("tr_tids_req_handler: TID request thread sent invalid reply. Aborting!"); + retval=-1; + goto cleanup; + } + tr_notice("tr_tids_req_handler: TID request for AAA server %d failed.", + payload->thread_id); + } else { + /* unexpected message */ + tr_err("tr_tids_req_handler: Unexpected message received. Aborting!"); + retval=-1; + goto cleanup; + } + + /* Set the cookie pointer to NULL so we know we've dealt with this one. The + * cookie itself is in our tmp_ctx, which we'll free before exiting. Let it hang + * around in case we are still using pointers to elements of the cookie. */ + aaa_cookie[payload->thread_id]=NULL; + + tr_mq_msg_free(msg); + + /* check whether we've received enough responses to exit */ + if ((idp_shared && (n_responses>0)) || + (resp_frac_denom*n_responses>=resp_frac_numer*n_aaa)) + break; + } + + tr_debug("tr_tids_req_handler: done waiting for responses. %d responses, %d failures.", + n_responses, n_failed); + /* Inform any remaining threads that we will no longer handle their responses. */ + for (ii=0; iimq=NULL; /* threads will not try to respond through a null mq */ + + if (0!=tr_tids_fwd_release_mutex(aaa_cookie[ii])) + tr_notice("tr_tids_req_handler: unable to release mutex for AAA thread %d.", ii); + } + } + + /* Now all threads have either replied (and aaa_cookie[ii] is null) or have been told not to + * reply (by setting their mq pointer to null). However, some may have responded by placing + * a message on the mq after we last checked but before we set their mq pointer to null. These + * will not know that we gave up on them, so we must free their cookies for them. We can just + * go through any remaining messages on the mq to identify these threads. By putting them in + * our context instead of freeing them directly, we ensure we don't accidentally invalidate + * any of our own pointers into the structure before this function exits. */ + while (NULL!=(msg=tr_mq_pop(mq, NULL))) { + payload=(TR_RESP_COOKIE *)tr_mq_msg_get_payload(msg); + if (aaa_cookie[payload->thread_id]!=NULL) + talloc_steal(tmp_ctx, aaa_cookie[payload->thread_id]); + + tr_mq_msg_free(msg); + } + + if (n_responses==0) { + /* No requests succeeded, so this will be an error */ + retval = -1; + + /* If we got any error responses, send an arbitrarily chosen one. */ + for (ii=0; iitids; TR_CFG_MGR *cfg_mgr = cookie->cfg_mgr; @@ -261,15 +762,15 @@ static int tr_tids_gss_handler(gss_name_t client_name, TR_NAME *gss_name, return -1; } - /* look up the RP client matching the GSS name */ - if ((NULL == (rp = tr_rp_client_lookup(cfg_mgr->active->rp_clients, gss_name)))) { - tr_debug("tr_tids_gss_handler: Unknown GSS name %s", gss_name->buf); + /* Ensure at least one client exists using this GSS name */ + if (NULL == tr_rp_client_lookup(cfg_mgr->active->rp_clients, gss_name)) { + tr_debug("tr_tids_gss_handler: Unknown GSS name %.*s", gss_name->len, gss_name->buf); return -1; } - /* Store the rp client */ - tids->rp_gss = rp; - tr_debug("Client's GSS Name: %s", gss_name->buf); + /* Store the GSS name */ + tids->gss_name = tr_dup_name(gss_name); + tr_debug("Client's GSS Name: %.*s", gss_name->len, gss_name->buf); return 0; } @@ -280,7 +781,7 @@ static int tr_tids_gss_handler(gss_name_t client_name, TR_NAME *gss_name, /* called when a connection to the TIDS port is received */ static void tr_tids_event_cb(int listener, short event, void *arg) { - TIDS_INSTANCE *tids = (TIDS_INSTANCE *)arg; + TIDS_INSTANCE *tids = talloc_get_type_abort(arg, TIDS_INSTANCE); if (0==(event & EV_READ)) tr_debug("tr_tids_event_cb: unexpected event on TIDS socket (event=0x%X)", event); @@ -288,18 +789,28 @@ static void tr_tids_event_cb(int listener, short event, void *arg) tids_accept(tids, listener); } -/* Configure the tids instance and set up its event handler. +/* called when it's time to sweep for completed TID child processes */ +static void tr_tids_sweep_cb(int listener, short event, void *arg) +{ + TIDS_INSTANCE *tids = talloc_get_type_abort(arg, TIDS_INSTANCE); + + if (0==(event & EV_TIMEOUT)) + tr_debug("tr_tids_event_cb: unexpected event on TID process sweep timer (event=0x%X)", event); + else + tids_sweep_procs(tids); +} + +/* Configure the tids instance and set up its event handlers. * Returns 0 on success, nonzero on failure. Fills in * *tids_event (which should be allocated by caller). */ -int tr_tids_event_init(struct event_base *base, - TIDS_INSTANCE *tids, - TR_CFG_MGR *cfg_mgr, - TRPS_INSTANCE *trps, - struct tr_socket_event *tids_ev) +int tr_tids_event_init(struct event_base *base, TIDS_INSTANCE *tids, TR_CFG_MGR *cfg_mgr, TRPS_INSTANCE *trps, + struct tr_socket_event *tids_ev, struct event **sweep_ev) { TALLOC_CTX *tmp_ctx=talloc_new(NULL); struct tr_tids_event_cookie *cookie=NULL; + struct timeval sweep_interval; int retval=0; + int ii=0; if (tids_ev == NULL) { tr_debug("tr_tids_event_init: Null tids_ev."); @@ -307,6 +818,12 @@ int tr_tids_event_init(struct event_base *base, goto cleanup; } + if (sweep_ev == NULL) { + tr_debug("tr_tids_event_init: Null sweep_ev."); + retval = 1; + goto cleanup; + } + /* Create the cookie for callbacks. We'll put it in the tids context, so it will * be cleaned up when tids is freed by talloc_free. */ cookie=talloc(tmp_ctx, struct tr_tids_event_cookie); @@ -321,25 +838,35 @@ int tr_tids_event_init(struct event_base *base, talloc_steal(tids, cookie); /* get a tids listener */ - tids_ev->sock_fd=tids_get_listener(tids, - tr_tids_req_handler, - tr_tids_gss_handler, - cfg_mgr->active->internal->hostname, - cfg_mgr->active->internal->tids_port, - (void *)cookie); - if (tids_ev->sock_fd < 0) { + tids_ev->n_sock_fd = (int)tids_get_listener(tids, + tr_tids_req_handler, + tr_tids_gss_handler, + cfg_mgr->active->internal->hostname, + cfg_mgr->active->internal->tids_port, + (void *)cookie, + tids_ev->sock_fd, + TR_MAX_SOCKETS); + if (tids_ev->n_sock_fd==0) { tr_crit("Error opening TID server socket."); retval=1; goto cleanup; } - /* and its event */ - tids_ev->ev=event_new(base, - tids_ev->sock_fd, - EV_READ|EV_PERSIST, - tr_tids_event_cb, - (void *)tids); - event_add(tids_ev->ev, NULL); + /* Set up listener events */ + for (ii=0; iin_sock_fd; ii++) { + tids_ev->ev[ii]=event_new(base, + tids_ev->sock_fd[ii], + EV_READ|EV_PERSIST, + tr_tids_event_cb, + (void *)tids); + event_add(tids_ev->ev[ii], NULL); + } + + /* Set up a periodic check for completed TID handler processes */ + *sweep_ev = event_new(base, -1, EV_TIMEOUT|EV_PERSIST, tr_tids_sweep_cb, tids); + sweep_interval.tv_sec = 10; + sweep_interval.tv_usec = 0; + event_add(*sweep_ev, &sweep_interval); cleanup: talloc_free(tmp_ctx);