tr_debug("tr_trps_gss_handler()");
- if ((!client_name) || (!gss_name) || (!trps) || (!cfg_mgr)) {
+ if ((!client_name) || (!trps) || (!cfg_mgr)) {
tr_debug("tr_trps_gss_handler: Bad parameters.");
return -1;
}
static void tr_trps_cleanup_trpc(TRPS_INSTANCE *trps, TRPC_INSTANCE *trpc)
{
- TR_MQ_MSG *msg;
-
- /* tell the trpc thread to exit */
- msg = tr_mq_msg_new(NULL, TR_MQMSG_TRPC_EXIT_OK, TR_MQ_PRIO_NORMAL);
- if (msg) {
- tr_debug("tr_trps_cleanup_trpc: Notifying thread that it may now exit");
- trpc_mq_add(trpc, msg);
- } else {
- tr_crit("tr_trps_cleanup_trpc: Unable to acknowledge disconnection, thread will probably never terminate");
- }
-
pthread_join(*trp_connection_get_thread(trpc_get_conn(trpc)), NULL);
trps_remove_trpc(trps, trpc);
trpc_free(trpc);
tr_err("tr_trps_process_mq: incoming connection from unknown peer (%s) reported.", tmp);
else {
trp_peer_set_incoming_status(peer, PEER_CONNECTED);
- tr_err("tr_trps_process_mq: incoming connection from %s established.", tmp);
+ tr_notice("tr_trps_process_mq: incoming connection from %s established.", tmp);
}
free(tmp);
}
} else {
trp_peer_set_incoming_status(peer, PEER_DISCONNECTED);
tr_trps_cleanup_conn(trps, conn);
- tr_err("tr_trps_process_mq: incoming connection from %s lost.", tmp);
+ tr_notice("tr_trps_process_mq: incoming connection from %s lost.", tmp);
}
free(tmp);
}
tr_err("tr_trps_process_mq: outgoing connection to unknown peer (%s) reported.", tmp);
else {
trp_peer_set_outgoing_status(peer, PEER_CONNECTED);
- tr_err("tr_trps_process_mq: outgoing connection to %s established.", tmp);
+ tr_notice("tr_trps_process_mq: outgoing connection to %s established.", tmp);
}
free(tmp);
}
else if (0==strcmp(s, TR_MQMSG_TRPC_DISCONNECTED)) {
- /* The trpc connection died - see note above tr_trpc_thread() regarding the shutdown protocol.
- * The EXIT_OK message is sent in tr_trps_cleanup_trpc(). */
TRPC_INSTANCE *trpc=talloc_get_type_abort(tr_mq_msg_get_payload(msg), TRPC_INSTANCE);
TR_NAME *svcname=trpc_get_gssname(trpc);
peer=trps_get_peer_by_servicename(trps, svcname);
tr_err("tr_trps_process_mq: outgoing connection to unknown peer (%s) lost.", tmp);
else {
trp_peer_set_outgoing_status(peer, PEER_DISCONNECTED);
- tr_err("tr_trps_process_mq: outgoing connection to %s lost.", tmp);
+ tr_notice("tr_trps_process_mq: outgoing connection to %s lost.", tmp);
tr_trps_cleanup_trpc(trps, trpc);
}
free(tmp);
return retval;
}
-
- struct trpc_notify_cb_data {
- int msg_ready;
- pthread_cond_t cond;
- pthread_mutex_t mutex;
- };
-
- static void tr_trpc_mq_cb(TR_MQ *mq, void *arg)
- {
- struct trpc_notify_cb_data *cb_data=(struct trpc_notify_cb_data *) arg;
- pthread_mutex_lock(&(cb_data->mutex));
- if (!cb_data->msg_ready) {
- cb_data->msg_ready=1;
- pthread_cond_signal(&(cb_data->cond));
- }
- pthread_mutex_unlock(&(cb_data->mutex));
- }
-
/* data passed to thread */
struct trpc_thread_data {
TRPC_INSTANCE *trpc;
};
/**
- * Returns when a message is ready
- */
- static void trpc_thread_wait_for_message(struct trpc_notify_cb_data *cb)
- {
- cb->msg_ready = 0;
- do
- pthread_cond_wait(&(cb->cond), &(cb->mutex));
- while (!cb->msg_ready);
- }
-
- /**
* Thread for handling TRPC (outgoing) connections
*
- * Locking protocol:
- *
- * This thread uses a mutex and condition variable to wait for incoming messages
- * on its queue. The main thread interacts with these through the tr_trpc_mq_cb()
- * callback. This callback is called by the TR_MQ whenever the queue goes from empty
- * to non-empty. The callback locks the thread's mutex, then sets a msg_ready flag
- * signals the condition variable, then unlocks the mutex.
- *
- * This thread waits for the condition variable to be signaled, then checks that the
- * msg_ready flag has actually been set (pthreads does not guarantee against false
- * wakes). If so, it holds the mutex lock while reading any messages in its queue and
- * sending responses over the GSS connection. If it receives an ABORT message from
- * the main thread (not currently used), or if sending the GSS message fails (which
- * usually indicates that the connection has been lost), the thread begins the shutdown
- * process. It still holds the mutex lock at the start of this process.
+ * Opens a connection to a peer. If successful, notifies the trps thread by
+ * posting a TR_MQMSG_TRPC_CONNECTED message to the trps message queue.
+ * It then waits for messages on trpc->mq. Normally these will be TR_MQMSG_TRPC_SEND
+ * messages, which this thread forwards to the peer. If its connection is lost or
+ * a TR_MQMSG_ABORT message is received on trpc->mq, the thread sends a
+ * TR_MQMSG_TRPC_DISCONNECTED message to the trps thread, then cleans up and
+ * terminates.
*
- * This begins by setting trpc->shutting_down = 1, then sending a message to the
- * TRPS (main) thread indicating that it has DISCONNECTED. It then releases the mutex
- * and waits for more messages, again using the mutex and the condition variable. It
- * ignores any except for an EXIT_OK message. When that is received, it exits, terminating
- * the thread.
- *
- * When the main thread receives the DISCONNECTED message, it must stop sending messages
- * to this thread's message queue. When it is certain that it will not queue any more messages,
- * it sends the EXIT_OK message as its last interaction with this thread. Once that message
- * has been queued, queueing additional messages may result in deadlock or segfaults.
- * (It is ok if there are other messages in the queue, but it must not queue more.)
+ * The trps may continue queueing messages for this client even when the
+ * connection is down. To prevent the queue from growing endlessly, this thread
+ * should clear its queue after failed connection attempts.
*/
static void *tr_trpc_thread(void *arg)
{
const char *msg_type=NULL;
char *encoded_msg=NULL;
TR_NAME *peer_gssname=NULL;
- int n_sent = 0;
- int n_popped = 0;
+ struct timespec wait_until = {0};
int exit_loop=0;
- struct trpc_notify_cb_data cb_data={0,
- PTHREAD_COND_INITIALIZER,
- PTHREAD_MUTEX_INITIALIZER};
-
tr_debug("tr_trpc_thread: started");
- /* set up the mq for receiving */
- pthread_mutex_lock(&(cb_data.mutex)); /* hold this lock until we enter the main loop */
-
- tr_mq_lock(trpc->mq);
- tr_mq_set_notify_cb(trpc->mq, tr_trpc_mq_cb, (void *) &cb_data);
- tr_mq_unlock(trpc->mq);
-
+ /* Try to make the outgoing connection */
rc=trpc_connect(trpc);
if (rc!=TRP_SUCCESS) {
tr_notice("tr_trpc_thread: failed to initiate connection to %s:%d.",
trpc_get_server(trpc),
trpc_get_port(trpc));
+ trpc_mq_clear(trpc); /* clear the queue even though we did not connect */
} else {
+ /* Retrieve the GSS name used by the peer for authentication */
peer_gssname=trp_connection_get_peer(trpc_get_conn(trpc));
if (peer_gssname==NULL) {
tr_err("tr_trpc_thread: could not duplicate peer_gssname.");
talloc_free(tmp_ctx);
return NULL;
}
- tr_debug("tr_trpc_thread: connected to peer %s", peer_gssname->buf);
+ tr_debug("tr_trpc_thread: connected to peer %.*s",
+ peer_gssname->len, peer_gssname->buf);
msg=tr_mq_msg_new(tmp_ctx, TR_MQMSG_TRPC_CONNECTED, TR_MQ_PRIO_HIGH);
tr_mq_msg_set_payload(msg, (void *)tr_dup_name(peer_gssname), tr_free_name_helper);
trps_mq_add(trps, msg); /* steals msg context */
msg=NULL;
+ /* Loop until we get an abort message or until the connection is lost. */
while(!exit_loop) {
- trpc_thread_wait_for_message(&cb_data); /* handles locking */
-
- n_popped = 0; /* have not popped any messages from the queue */
- n_sent = 0; /* have not sent any messages yet */
- for (msg = trpc_mq_pop(trpc);
- msg != NULL;
- msg = trpc_mq_pop(trpc)) {
- n_popped++;
+ /* Wait up to 10 minutes for a message to be queued to send to the peer.
+ * Log a warning if we go longer than that, but don't give up. */
+ if (tr_mq_pop_timeout(10 * 60, &wait_until) != 0) {
+ tr_err("tr_trpc_thread: unable to set abort timeout");
+ break; /* immediately exit the loop, don't go through cleanup */
+ }
+
+ /* Pop a message from the queue. */
+ msg = trpc_mq_pop(trpc, &wait_until);
+ if (msg) {
msg_type = tr_mq_msg_get_message(msg);
if (0 == strcmp(msg_type, TR_MQMSG_ABORT)) {
+ tr_debug("tr_trpc_thread: received abort message from main thread.");
exit_loop = 1;
- break;
} else if (0 == strcmp(msg_type, TR_MQMSG_TRPC_SEND)) {
encoded_msg = tr_mq_msg_get_payload(msg);
if (encoded_msg == NULL)
else {
rc = trpc_send_msg(trpc, encoded_msg);
if (rc == TRP_SUCCESS) {
- n_sent++;
+ tr_debug("tr_trpc_thread: sent message.");
} else {
tr_notice("tr_trpc_thread: trpc_send_msg failed.");
/* Assume this means we lost the connection. */
exit_loop = 1;
- break;
}
}
} else
tr_notice("tr_trpc_thread: unknown message '%s' received.", msg_type);
tr_mq_msg_free(msg);
+ } else {
+ tr_warning("tr_trpc_thread: no outgoing messages to %.*s for 10 minutes",
+ peer_gssname->len, peer_gssname->buf);
}
-
- /* if n_popped == 0, then n_sent must be zero (it's only set after at
- * least one msg is popped) */
- if (n_popped==0)
- tr_err("tr_trpc_thread: notified of message, but queue empty");
- else
- tr_debug("tr_trpc_thread: sent %d messages.", n_sent);
}
}
- tr_debug("tr_trpc_thread: Disconnected. Waiting to terminate thread.");
- trpc->shutting_down = 1;
-
/* Send a DISCONNECTED message to the main thread */
+ tr_debug("tr_trpc_thread: notifying main thread of disconnection.");
msg=tr_mq_msg_new(tmp_ctx, TR_MQMSG_TRPC_DISCONNECTED, TR_MQ_PRIO_NORMAL);
tr_mq_msg_set_payload(msg, (void *)trpc, NULL); /* do not pass a free routine */
if (msg==NULL) {
- /* can't notify main thread of exit - just do it and hope for the best */
+ /* can't notify main thread */
tr_err("tr_trpc_thread: error allocating TR_MQ_MSG");
} else {
trps_mq_add(trps, msg);
-
- /* DISCONNECTED sent, now wait for an acknowledgement before exiting */
- exit_loop = 0;
- while (!exit_loop) {
- trpc_thread_wait_for_message(&cb_data);
-
- while (NULL != (msg = trpc_mq_pop(trpc))) {
- msg_type = tr_mq_msg_get_message(msg);
-
- /* ignore anything except an EXIT_OK */
- if (0 == strcmp(msg_type, TR_MQMSG_TRPC_EXIT_OK)) {
- exit_loop = 1;
- break; /* skip any further messages */
- }
- }
- }
}
- tr_debug("tr_trpc_thread: thread terminating.");
talloc_free(tmp_ctx);
+ tr_debug("tr_trpc_thread: thread terminating.");
return NULL;
}
tr_debug("tr_config_changed: freeing tr->mons->authorized_gss_names");
tr_gss_names_free(tr->mons->authorized_gss_names);
}
- tr->mons->authorized_gss_names = tr_gss_names_dup(tr->mons, new_cfg->internal->monitoring_credentials);
+ if (new_cfg->internal->monitoring_credentials != NULL) {
+ tr->mons->authorized_gss_names = tr_gss_names_dup(tr->mons, new_cfg->internal->monitoring_credentials);
+ } else {
+ tr->mons->authorized_gss_names = tr_gss_names_new(tr->mons);
+ }
if (tr->mons->authorized_gss_names == NULL) {
tr_err("tr_config_changed: Error configuring monitoring credentials");
}
/* get the connection for this peer */
trpc=trps_find_trpc(trps, peer);
- /* instead, let's let that happen and then clear the queue when an attempt to
- * connect fails */
+ /* The peer connection (trpc) usually exists even if the connection is down.
+ * We will queue messages even if the connection is down. To prevent this from
+ * endlessly increasing the size of the queue, the trpc handler needs to clear
+ * its queue periodically, even if it is unable to send the messages
+ */
if (trpc==NULL) {
tr_warning("trps_send_msg: skipping message queued for missing TRP client entry.");
- } else if (trpc->shutting_down) {
- tr_debug("trps_send_msg: skipping message because TRP client is shutting down.");
- rc = TRP_SUCCESS; /* it's ok that this didn't get sent, the connection will be gone in a moment */
} else {
mq_msg=tr_mq_msg_new(tmp_ctx, TR_MQMSG_TRPC_SEND, TR_MQ_PRIO_NORMAL);
msg_dup=talloc_strdup(mq_msg, msg); /* get local copy in mq_msg context */
static TRP_RC trps_handle_inforec_route(TRPS_INSTANCE *trps, TRP_UPD *upd, TRP_INFOREC *rec)
{
TRP_ROUTE *route=NULL;
+ TR_COMM *comm = NULL;
unsigned int feas=0;
/* determine feasibility */
feas=trps_check_feasibility(trps, trp_upd_get_realm(upd), trp_upd_get_comm(upd), rec);
tr_debug("trps_handle_update: record feasibility=%d", feas);
- /* do we have an existing route? */
- route=trps_get_route(trps,
- trp_upd_get_comm(upd),
- trp_upd_get_realm(upd),
- trp_upd_get_peer(upd));
- if (route!=NULL) {
- /* there was a route table entry already */
- tr_debug("trps_handle_updates: route entry already exists.");
- if (feas) {
- /* Update is feasible. Accept it. */
- trps_accept_update(trps, upd, rec);
- } else {
- /* Update is infeasible. Ignore it unless the trust router has changed. */
- if (0!=tr_name_cmp(trp_route_get_trust_router(route),
- trp_inforec_get_trust_router(rec))) {
- /* the trust router associated with the route has changed, treat update as a retraction */
- trps_retract_route(trps, route);
+ /* verify that the community is an APC */
+ comm = tr_comm_table_find_comm(trps->ctable, trp_upd_get_comm(upd));
+ if (comm == NULL) {
+ /* We don't know this community. Reject the route. */
+ tr_debug("trps_handle_updates: community %.*s unknown, ignoring route for %.*s",
+ trp_upd_get_comm(upd)->len, trp_upd_get_comm(upd)->buf,
+ trp_upd_get_realm(upd)->len, trp_upd_get_realm(upd)->buf);
+ } else if (tr_comm_get_type(comm) != TR_COMM_APC) {
+ /* The community in a route request *must* be an APC. This was not - ignore it. */
+ tr_debug("trps_handle_updates: community %.*s is not an APC, ignoring route for %.*s",
+ trp_upd_get_comm(upd)->len, trp_upd_get_comm(upd)->buf,
+ trp_upd_get_realm(upd)->len, trp_upd_get_realm(upd)->buf);
+ } else {
+ /* do we have an existing route? */
+ route=trps_get_route(trps,
+ trp_upd_get_comm(upd),
+ trp_upd_get_realm(upd),
+ trp_upd_get_peer(upd));
+ if (route!=NULL) {
+ /* there was a route table entry already */
+ tr_debug("trps_handle_updates: route entry already exists.");
+ if (feas) {
+ /* Update is feasible. Accept it. */
+ trps_accept_update(trps, upd, rec);
+ } else {
+ /* Update is infeasible. Ignore it unless the trust router has changed. */
+ if (0!=tr_name_cmp(trp_route_get_trust_router(route),
+ trp_inforec_get_trust_router(rec))) {
+ /* the trust router associated with the route has changed, treat update as a retraction */
+ trps_retract_route(trps, route);
+ }
}
+ } else {
+ /* No existing route table entry. Ignore it unless it is feasible and not a retraction. */
+ tr_debug("trps_handle_update: no route entry exists yet.");
+ if (feas && trp_metric_is_finite(trp_inforec_get_metric(rec)))
+ trps_accept_update(trps, upd, rec);
}
- } else {
- /* No existing route table entry. Ignore it unless it is feasible and not a retraction. */
- tr_debug("trps_handle_update: no route entry exists yet.");
- if (feas && trp_metric_is_finite(trp_inforec_get_metric(rec)))
- trps_accept_update(trps, upd, rec);
}
return TRP_SUCCESS;
return comm;
}
-static TR_RP_REALM *trps_create_new_rp_realm(TALLOC_CTX *mem_ctx, TR_NAME *realm_id, TRP_INFOREC *rec)
+static TR_RP_REALM *trps_create_new_rp_realm(TALLOC_CTX *mem_ctx, TR_NAME *comm, TR_NAME *realm_id, TRP_INFOREC *rec)
{
TALLOC_CTX *tmp_ctx=talloc_new(NULL);
TR_RP_REALM *rp=tr_rp_realm_new(tmp_ctx);
return rp;
}
-static TR_IDP_REALM *trps_create_new_idp_realm(TALLOC_CTX *mem_ctx, TR_NAME *realm_id, TRP_INFOREC *rec)
+static TR_IDP_REALM *trps_create_new_idp_realm(TALLOC_CTX *mem_ctx,
+ TR_NAME *comm_id,
+ TR_NAME *realm_id,
+ TRP_INFOREC *rec)
{
TALLOC_CTX *tmp_ctx=talloc_new(NULL);
TR_IDP_REALM *idp=tr_idp_realm_new(tmp_ctx);
-
+ TR_APC *realm_apcs = NULL;
+
if (idp==NULL) {
tr_debug("trps_create_new_idp_realm: unable to allocate new realm.");
goto cleanup;
idp=NULL;
goto cleanup;
}
- if (trp_inforec_get_apcs(rec)!=NULL) {
- tr_idp_realm_set_apcs(idp, tr_apc_dup(tmp_ctx, trp_inforec_get_apcs(rec)));
- if (tr_idp_realm_get_apcs(idp)==NULL) {
- tr_debug("trps_create_new_idp_realm: unable to allocate APC list.");
- idp=NULL;
+
+ /* Set the APCs. If the community is a CoI, copy its APCs. If it is an APC, then
+ * that community itself is the APC for the realm. */
+ if (trp_inforec_get_comm_type(rec) == TR_COMM_APC) {
+ /* the community is an APC for this realm */
+ realm_apcs = tr_apc_new(tmp_ctx);
+ if (realm_apcs == NULL) {
+ tr_debug("trps_create_new_idp_realm: unable to allocate new APC list.");
+ idp = NULL;
+ goto cleanup;
+ }
+
+ tr_apc_set_id(realm_apcs, tr_dup_name(comm_id));
+ if (tr_apc_get_id(realm_apcs) == NULL) {
+ tr_debug("trps_create_new_idp_realm: unable to allocate new APC name.");
+ idp = NULL;
+ goto cleanup;
+ }
+ } else {
+ /* the community is not an APC for this realm */
+ realm_apcs = trp_inforec_get_apcs(rec);
+ if (realm_apcs == NULL) {
+ tr_debug("trps_create_new_idp_realm: no APCs for realm %.*s/%.*s, cannot add.",
+ realm_id->len, realm_id->buf,
+ comm_id->len, comm_id->buf);
+ idp = NULL;
+ goto cleanup;
+ }
+
+ /* we have APCs, make our own copy */
+ realm_apcs = tr_apc_dup(tmp_ctx, realm_apcs);
+ if (realm_apcs == NULL) {
+ tr_debug("trps_create_new_idp_realm: unable to duplicate APC list.");
+ idp = NULL;
goto cleanup;
}
}
+
+ /* Whether the community is an APC or CoI, the APCs for the realm are in realm_apcs */
+ tr_idp_realm_set_apcs(idp, realm_apcs); /* takes realm_apcs out of tmp_ctx on success */
+ if (tr_idp_realm_get_apcs(idp) == NULL) {
+ tr_debug("trps_create_new_idp_realm: unable to set APC list for new realm.");
+ idp=NULL;
+ goto cleanup;
+ }
+
idp->origin=TR_REALM_DISCOVERED;
talloc_steal(mem_ctx, idp);
tr_debug("trps_handle_inforec_comm: unable to create new community.");
goto cleanup;
}
- tr_comm_table_add_comm(trps->ctable, comm);
+ if (tr_comm_table_add_comm(trps->ctable, comm) != 0)
+ {
+ tr_debug("trps_handle_inforec_comm: unable to add community to community table.");
+ goto cleanup;
+ }
}
/* TODO: see if other comm data match the new inforec and update or complain */
if (rp_realm==NULL) {
tr_debug("trps_handle_inforec_comm: unknown RP realm %.*s in inforec, creating it.",
realm_id->len, realm_id->buf);
- rp_realm=trps_create_new_rp_realm(tmp_ctx, realm_id, rec);
+ rp_realm= trps_create_new_rp_realm(tmp_ctx, tr_comm_get_id(comm), realm_id, rec);
if (rp_realm==NULL) {
tr_debug("trps_handle_inforec_comm: unable to create new RP realm.");
/* we may leave an unused community in the table, but it will only last until
if (idp_realm==NULL) {
tr_debug("trps_handle_inforec_comm: unknown IDP realm %.*s in inforec, creating it.",
realm_id->len, realm_id->buf);
- idp_realm=trps_create_new_idp_realm(tmp_ctx, realm_id, rec);
+ idp_realm= trps_create_new_idp_realm(tmp_ctx, tr_comm_get_id(comm), realm_id, rec);
if (idp_realm==NULL) {
tr_debug("trps_handle_inforec_comm: unable to create new IDP realm.");
/* we may leave an unused community in the table, but it will only last until
/* choose the best route to comm/realm, optionally excluding routes to a particular peer */
static TRP_ROUTE *trps_find_best_route(TRPS_INSTANCE *trps,
- TR_NAME *comm,
- TR_NAME *realm,
- TR_NAME *exclude_peer)
+ TR_NAME *comm,
+ TR_NAME *realm,
+ TR_NAME *exclude_peer_label)
{
TRP_ROUTE **entry=NULL;
TRP_ROUTE *best=NULL;
+ TRP_PEER *route_peer = NULL;
size_t n_entry=0;
unsigned int kk=0;
unsigned int kk_min=0;
entry=trp_rtable_get_realm_entries(trps->rtable, comm, realm, &n_entry);
for (kk=0; kk<n_entry; kk++) {
if (trp_route_get_metric(entry[kk]) < min_metric) {
- if ((exclude_peer==NULL) || (0!=tr_name_cmp(trp_route_get_peer(entry[kk]),
- exclude_peer))) {
- kk_min=kk;
- min_metric=trp_route_get_metric(entry[kk]);
- }
+ if (exclude_peer_label != NULL) {
+ if (!trp_route_is_local(entry[kk])) {
+ /* route is not local, check the peer label */
+ route_peer = trp_ptable_find_gss_name(trps->ptable,
+ trp_route_get_peer(entry[kk]));
+ if (route_peer == NULL) {
+ tr_err("trps_find_best_route: unknown peer GSS name (%.*s) for route %d to %.*s/%.*s",
+ trp_route_get_peer(entry[kk])->len, trp_route_get_peer(entry[kk])->buf,
+ kk,
+ realm->len, realm->buf,
+ comm->len, comm->buf);
+ continue; /* unknown peer, skip the route */
+ }
+ if (0 == tr_name_cmp(exclude_peer_label, trp_peer_get_label(route_peer))) {
+ /* we're excluding this peer - skip the route */
+ continue;
+ }
+ }
+ }
+ /* if we get here, we're not excluding the route */
+ kk_min = kk;
+ min_metric = trp_route_get_metric(entry[kk]);
}
}
+
if (trp_metric_is_finite(min_metric))
best=entry[kk_min];
{
TALLOC_CTX *tmp_ctx=talloc_new(NULL);
struct timespec sweep_time={0,0};
+ struct timespec tmp = {0};
TR_COMM_MEMB *memb=NULL;
TR_COMM_ITER *iter=NULL;
TRP_RC rc=TRP_ERROR;
tr_comm_memb_get_realm_id(memb)->len, tr_comm_memb_get_realm_id(memb)->buf,
tr_comm_get_id(tr_comm_memb_get_comm(memb))->len, tr_comm_get_id(tr_comm_memb_get_comm(memb))->buf,
tr_comm_memb_get_origin(memb)->len, tr_comm_memb_get_origin(memb)->buf,
- timespec_to_str(tr_comm_memb_get_expiry(memb)));
+ timespec_to_str(tr_comm_memb_get_expiry_realtime(memb, &tmp)));
tr_comm_table_remove_memb(trps->ctable, memb);
tr_comm_memb_free(memb);
} else {
tr_comm_memb_expire(memb);
trps_compute_expiry(trps, tr_comm_memb_get_interval(memb), tr_comm_memb_get_expiry(memb));
tr_debug("trps_sweep_ctable: community membership expired at %s, resetting expiry to %s (%.*s in %.*s, origin %.*s).",
- timespec_to_str(&sweep_time),
- timespec_to_str(tr_comm_memb_get_expiry(memb)),
+ timespec_to_str(tr_clock_convert(TRP_CLOCK, &sweep_time, CLOCK_REALTIME, &tmp)),
+ timespec_to_str(tr_comm_memb_get_expiry_realtime(memb, &tmp)),
tr_comm_memb_get_realm_id(memb)->len, tr_comm_memb_get_realm_id(memb)->buf,
tr_comm_get_id(tr_comm_memb_get_comm(memb))->len, tr_comm_get_id(tr_comm_memb_get_comm(memb))->buf,
tr_comm_memb_get_origin(memb)->len, tr_comm_memb_get_origin(memb)->buf);
}
/* select the correct route to comm/realm to be announced to peer */
-static TRP_ROUTE *trps_select_realm_update(TRPS_INSTANCE *trps, TR_NAME *comm, TR_NAME *realm, TR_NAME *peer_gssname)
+static TRP_ROUTE *trps_select_realm_update(TRPS_INSTANCE *trps, TR_NAME *comm, TR_NAME *realm, TR_NAME *peer_label)
{
- TRP_ROUTE *route;
+ TRP_ROUTE *route = NULL;
+ TRP_PEER *route_peer = NULL;
+ TR_NAME *route_peer_label = NULL;
/* Take the currently selected route unless it is through the peer we're sending the update to.
- * I.e., enforce the split horizon rule. */
+ * I.e., enforce the split horizon rule. Start by looking up the currently selected route. */
route=trp_rtable_get_selected_entry(trps->rtable, comm, realm);
if (route==NULL) {
/* No selected route, this should only happen if the only route has been retracted,
* in which case we do not want to advertise it. */
return NULL;
}
- tr_debug("trps_select_realm_update: %s vs %s", peer_gssname->buf,
- trp_route_get_peer(route)->buf);
- if (0==tr_name_cmp(peer_gssname, trp_route_get_peer(route))) {
- tr_debug("trps_select_realm_update: matched, finding alternate route");
- /* the selected entry goes through the peer we're reporting to, choose an alternate */
- route=trps_find_best_route(trps, comm, realm, peer_gssname);
- if ((route==NULL) || (!trp_metric_is_finite(trp_route_get_metric(route))))
- return NULL; /* don't advertise a nonexistent or retracted route */
+
+ /* Check whether it's local. */
+ if (trp_route_is_local(route)) {
+ /* It is always ok to announce a local route */
+ tr_debug("trps_select_realm_update: selected route for %.*s/%.*s is local",
+ realm->len, realm->buf,
+ comm->len, comm->buf);
+ } else {
+ /* It's not local. Get the route's peer and check whether it's the same place we
+ * got the selected route from. Peer should always correspond to an entry in our
+ * peer table. */
+ tr_debug("trps_select_realm_update: selected route for %.*s/%.*s is not local",
+ realm->len, realm->buf,
+ comm->len, comm->buf);
+ route_peer = trp_ptable_find_gss_name(trps->ptable, trp_route_get_peer(route));
+ if (route_peer == NULL) {
+ tr_err("trps_select_realm_update: unknown peer GSS name (%.*s) for selected route for %.*s/%.*s",
+ trp_route_get_peer(route)->len, trp_route_get_peer(route)->buf,
+ realm->len, realm->buf,
+ comm->len, comm->buf);
+ return NULL;
+ }
+ route_peer_label = trp_peer_get_label(route_peer);
+ if (route_peer_label == NULL) {
+ tr_err("trps_select_realm_update: error retrieving peer label for selected route for %.*s/%.*s",
+ realm->len, realm->buf,
+ comm->len, comm->buf);
+ return NULL;
+ }
+
+ /* see if these match */
+ tr_debug("trps_select_realm_update: %.*s vs %.*s",
+ peer_label->len, peer_label->buf,
+ route_peer_label->len, route_peer_label->buf);
+
+ if (0==tr_name_cmp(peer_label, route_peer_label)) {
+ /* the selected entry goes through the peer we're reporting to, choose an alternate */
+ tr_debug("trps_select_realm_update: matched, finding alternate route");
+ route=trps_find_best_route(trps, comm, realm, peer_label);
+ if ((route==NULL) || (!trp_metric_is_finite(trp_route_get_metric(route)))) {
+ tr_debug("trps_select_realm_update: no route to %.*s/%.*s suitable to advertise to %.*s",
+ realm->len, realm->buf,
+ comm->len, comm->buf,
+ peer_label->len, peer_label->buf);
+ return NULL; /* don't advertise a nonexistent or retracted route */
+ }
+ }
}
return route;
}
static TRP_RC trps_select_route_updates_for_peer(TALLOC_CTX *mem_ctx,
GPtrArray *updates,
TRPS_INSTANCE *trps,
- TR_NAME *peer_gssname,
+ TR_NAME *peer_label,
int triggered)
{
size_t n_comm=0;
for (ii=0; ii<n_comm; ii++) {
realm=trp_rtable_get_comm_realms(trps->rtable, comm[ii], &n_realm);
for (jj=0; jj<n_realm; jj++) {
- best=trps_select_realm_update(trps, comm[ii], realm[jj], peer_gssname);
+ best=trps_select_realm_update(trps, comm[ii], realm[jj], peer_label);
/* If we found a route, add it to the list. If triggered!=0, then only
* add triggered routes. */
if ((best!=NULL) && ((!triggered) || trp_route_is_triggered(best))) {
}
/* construct an update with all the inforecs for comm/realm/role to be sent to peer */
-static TRP_UPD *trps_comm_update(TALLOC_CTX *mem_ctx, TRPS_INSTANCE *trps, TR_NAME *peer_gssname, TR_COMM *comm, TR_REALM *realm)
+static TRP_UPD *trps_comm_update(TALLOC_CTX *mem_ctx,
+ TRPS_INSTANCE *trps,
+ TR_NAME *peer_label,
+ TR_COMM *comm,
+ TR_REALM *realm)
{
TALLOC_CTX *tmp_ctx=talloc_new(NULL);
TRP_UPD *upd=trp_upd_new(tmp_ctx);
static TRP_RC trps_select_comm_updates_for_peer(TALLOC_CTX *mem_ctx,
GPtrArray *updates,
TRPS_INSTANCE *trps,
- TR_NAME *peer_gssname,
+ TR_NAME *peer_label,
int triggered)
{
TALLOC_CTX *tmp_ctx=talloc_new(NULL);
tr_debug("trps_select_comm_updates_for_peer: adding realm %.*s",
tr_realm_get_id(realm)->len,
tr_realm_get_id(realm)->buf);
- upd=trps_comm_update(mem_ctx, trps, peer_gssname, comm, realm);
+ upd=trps_comm_update(mem_ctx, trps, peer_label, comm, realm);
if (upd!=NULL)
g_ptr_array_add(updates, upd);
}