X-Git-Url: https://code.citadel.org/?a=blobdiff_plain;f=citadel%2Fhousekeeping.c;h=0ca383202c2bb1f27275e2d92cedb5f7d5a860f5;hb=fa2dd842abb5feedea3e2253255722fcaecc3c6f;hp=b4d76e970cd7e3cac98ad37566e9e8aa4a27b19b;hpb=ae822088bca5cce3d6c3ef7f6d312a6d5de7a77c;p=citadel.git

diff --git a/citadel/housekeeping.c b/citadel/housekeeping.c
index b4d76e970..0ca383202 100644
--- a/citadel/housekeeping.c
+++ b/citadel/housekeeping.c
@@ -1,94 +1,217 @@
 /*
- * This file contains housekeeping tasks which periodically
- * need to be executed.
+ * This file contains miscellaneous housekeeping tasks.
  *
+ * Copyright (c) 1987-2021 by the citadel.org team
+ *
+ * This program is open source software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 3.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
  */
 
-#include <stdlib.h>
-#include <unistd.h>
 #include <stdio.h>
-#include <time.h>
-#include <ctype.h>
-#include <string.h>
-#include <errno.h>
-#include <pthread.h>
-#include "citadel.h"
-#include "server.h"
-#include "proto.h"
-#include "citserver.h"
+#include <libcitadel.h>
+
+#include "ctdl_module.h"
+#include "serv_extensions.h"
+#include "room_ops.h"
+#include "internet_addressing.h"
+#include "config.h"
+#include "journaling.h"
+#include "citadel_ldap.h"
+
+void check_sched_shutdown(void) {
+	if ((ScheduledShutdown == 1) && (ContextList == NULL)) {
+		syslog(LOG_NOTICE, "housekeeping: scheduled shutdown initiating");
+		server_shutting_down = 1;
+	}
+}
+
 
 /*
- * Terminate idle sessions.  This function pounds through the session table
- * comparing the current time to each session's time-of-last-command.  If an
- * idle session is found it is terminated, then the search restarts at the
- * beginning because the pointer to our place in the list becomes invalid.
+ * Check (and fix) floor reference counts.  This doesn't need to be done
+ * very often, since the counts should remain correct during normal operation.
  */
-void terminate_idle_sessions(void) {
-	struct CitContext *ccptr;
-	time_t now;
-	
-	time(&now);
-	for (ccptr = ContextList; ccptr != NULL; ccptr = ccptr->next) {
-		if (  (ccptr!=CC)
-		   && (config.c_sleeping > 0)
-		   && (now - (ccptr->lastcmd) > config.c_sleeping) ) {
-			lprintf(3, "Session %d timed out\n", ccptr->cs_pid);
-			kill_session(ccptr->cs_pid);
-			ccptr = ContextList;
-			}
+void check_ref_counts_backend(struct ctdlroom *qrbuf, void *data) {
+
+	int *new_refcounts;
+
+	new_refcounts = (int *) data;
+
+	++new_refcounts[(int)qrbuf->QRfloor];
+}
+
+
+void check_ref_counts(void) {
+	struct floor flbuf;
+	int a;
+
+	int new_refcounts[MAXFLOORS];
+
+	syslog(LOG_DEBUG, "housekeeping: checking floor reference counts");
+	for (a=0; a<MAXFLOORS; ++a) {
+		new_refcounts[a] = 0;
+	}
+
+	cdb_begin_transaction();
+	CtdlForEachRoom(check_ref_counts_backend, (void *)new_refcounts );
+	cdb_end_transaction();
+
+	for (a=0; a<MAXFLOORS; ++a) {
+		lgetfloor(&flbuf, a);
+		flbuf.f_ref_count = new_refcounts[a];
+		if (new_refcounts[a] > 0) {
+			flbuf.f_flags = flbuf.f_flags | QR_INUSE;
+		}
+		else {
+			flbuf.f_flags = flbuf.f_flags & ~QR_INUSE;
 		}
+		lputfloor(&flbuf, a);
+		syslog(LOG_DEBUG, "housekeeping: floor %d has %d rooms", a, new_refcounts[a]);
 	}
+}
 
 
 /*
- * Main housekeeping function.  This gets run whenever a session terminates.
+ * Provide hints as to whether we have any memory leaks
  */
+void keep_an_eye_on_memory_usage(void) {
+	static void *original_brk = NULL;
+	if (!original_brk) original_brk = sbrk(0);	// Remember the original program break so we can test for leaks
+	syslog(LOG_DEBUG, "original_brk=%lx, current_brk=%lx, addl=%ld", (long)original_brk, (long)sbrk(0), (long)(sbrk(0)-original_brk));	// FIXME not so noisy please
+}
+
+
+/*
+ * This is the housekeeping loop.  Worker threads come through here after
+ * processing client requests but before jumping back into the pool.  We
+ * only allow housekeeping to execute once per minute, and we only allow one
+ * instance to run at a time.
+ */
+static int housekeeping_in_progress = 0;
+static time_t last_timer = 0L;
 void do_housekeeping(void) {
+	int do_housekeeping_now = 0;
+	int do_perminute_housekeeping_now = 0;
+	time_t now;
 
-	begin_critical_section(S_HOUSEKEEPING);
 	/*
-	 * Terminate idle sessions.
+	 * We do it this way instead of wrapping the whole loop in an
+	 * S_HOUSEKEEPING critical section because it eliminates the need to
+	 * potentially have multiple concurrent mutexes in progress.
 	 */
-	lprintf(7, "Calling terminate_idle_sessions()\n");
-	terminate_idle_sessions();
+	begin_critical_section(S_HOUSEKEEPING);
+	if (housekeeping_in_progress == 0) {
+		do_housekeeping_now = 1;
+		housekeeping_in_progress = 1;
+	}
+	end_critical_section(S_HOUSEKEEPING);
+
+	now = time(NULL);
+	if ( (do_housekeeping_now == 0) && (!CtdlIsSingleUser()) ) {
+		if ( (now - last_timer) > (time_t)300 ) {
+			syslog(LOG_WARNING,
+				"housekeeping: WARNING: housekeeping loop has not run for %ld minutes.  Is something stuck?",
+				((now - last_timer) / 60)
+			);
+		}
+		return;
+	}
 
 	/*
-	 * If the server is scheduled to shut down the next time all
-	 * users are logged out, now's the time to do it.
+	 * Ok, at this point we've made the decision to run the housekeeping
+	 * loop.  Everything below this point is real work.
 	 */
-	if ((ScheduledShutdown == 1) && (ContextList == NULL)) {
-		lprintf(3, "Scheduled shutdown initiating.\n");
-		master_cleanup();
+
+	if ( (now - last_timer) > (time_t)60 ) {
+		do_perminute_housekeeping_now = 1;
+		last_timer = time(NULL);
+	}
+
+	/* First, do the "as often as needed" stuff... */
+	JournalRunQueue();
+	PerformSessionHooks(EVT_HOUSE);
+
+	/* Then, do the "once per minute" stuff... */
+	if (do_perminute_housekeeping_now) {
+		cdb_check_handles();
+		PerformSessionHooks(EVT_TIMER);		// Run all registered TIMER hooks
+
+#ifdef HAVE_LDAP					// LDAP sync isn't in a module so we can put it here
+		static time_t last_ldap_sync = 0L;
+		if ( (now - last_ldap_sync) > (time_t)CtdlGetConfigLong("c_ldap_sync_freq") ) {
+			CtdlSynchronizeUsersFromLDAP();
+			last_ldap_sync = time(NULL);
 		}
-	end_critical_section(S_HOUSEKEEPING);
+#endif
+
+	keep_an_eye_on_memory_usage();
 	}
 
+	/*
+	 * All done.
+	 */
+	begin_critical_section(S_HOUSEKEEPING);
+	housekeeping_in_progress = 0;
+	end_critical_section(S_HOUSEKEEPING);
+}
 
 
-/*
- * Check (and fix) floor reference counts.  This doesn't need to be done
- * very often, since the counts should remain correct during normal operation.
- */
-void check_ref_counts(void) {
-	int ref[MAXFLOORS];
-	struct quickroom qrbuf;
-	struct floor flbuf;
-	int a;
+void CtdlDisableHouseKeeping(void) {
+	int ActiveBackgroundJobs;
+	int do_housekeeping_now = 0;
+	struct CitContext *nptr;
+	int nContexts, i;
+
+retry_block_housekeeping:
+	syslog(LOG_INFO, "housekeeping: trying to disable services");
+	begin_critical_section(S_HOUSEKEEPING);
+	if (housekeeping_in_progress == 0) {
+		do_housekeeping_now = 1;
+		housekeeping_in_progress = 1;
+	}
+	end_critical_section(S_HOUSEKEEPING);
+	if (do_housekeeping_now == 0) {
+		usleep(1000000);
+		goto retry_block_housekeeping;
+	}
+	
+	syslog(LOG_INFO, "housekeeping: checking for running server jobs");
 
-	for (a=0; a<MAXFLOORS; ++a) ref[a] = 0;
+retry_wait_for_contexts:
+	/* So that we don't keep the context list locked for a long time
+	 * we create a copy of it first
+	 */
+	ActiveBackgroundJobs = 0;
+	nptr = CtdlGetContextArray(&nContexts) ;
+	if (nptr)
+	{
+		for (i=0; i<nContexts; i++) 
+		{
+			if ((nptr[i].state != CON_SYS) || (nptr[i].lastcmd == 0))
+				continue;
+			ActiveBackgroundJobs ++;
+			syslog(LOG_INFO, "jousekeeping: job CC[%d] active; use TERM if you don't want to wait for it", nptr[i].cs_pid);
 		
-	for (a=0; a<MAXROOMS; ++a) {
-		getroom(&qrbuf, a);
-		if (qrbuf.QRflags & QR_INUSE) {
-			++ref[(int)qrbuf.QRfloor];
-			}
 		}
+	
+		free(nptr);
+
+	}
+	if (ActiveBackgroundJobs != 0) {
+		syslog(LOG_INFO, "housekeeping: found %d running jobs, need to wait", ActiveBackgroundJobs);
+		usleep(5000000);
+		goto retry_wait_for_contexts;
+	}
+	syslog(LOG_INFO, "housekeeping: disabled now.");
+}
 
-	for (a=0; a<MAXFLOORS; ++a) {
-		lgetfloor(&flbuf, a);
-		flbuf.f_ref_count = ref[a];
-		if (ref[a] > 0) flbuf.f_flags = flbuf.f_flags | QR_INUSE ;
-		lputfloor(&flbuf, a);
-		}
-	}	
 
+void CtdlEnableHouseKeeping(void) {
+	begin_critical_section(S_HOUSEKEEPING);
+	housekeeping_in_progress = 0;
+	end_critical_section(S_HOUSEKEEPING);
+}