FIX: prevent so-sensor-clean runaway loop and concurrent instances

The cleanup loop had no progress check or pass limit, so once /nsm was over
threshold with nothing left to reclaim it spun at full speed, writing 1.1 GB /
12.5M lines to sensor_clean.log in five hours. Stop when a pass removes
nothing, when a pass frees no space, or at MAX_PASSES, and drop the per-pass
"no old files" logging in favor of one actionable line.

Replace the pgrep guard with flock -n. A find|while read subshell inherits the
parent's argv, so pgrep -cf counted one instance as hundreds; it was also
check-then-act, which let cron stack up overlapping runs.

Paths now derive from SENSOR_DIR with env-overridable LOG/LOCK so the
over-threshold path can be tested against a scratch filesystem.

salt/common/tools/sbin/so-sensor-clean

@@ -8,21 +8,37 @@
 # Elastic License 2.0.
 
 
-SENSOR_DIR='/nsm'
+SENSOR_DIR="${SENSOR_DIR:-/nsm}"
 CRIT_DISK_USAGE=90
-CUR_USAGE=$(df -P $SENSOR_DIR | tail -1 | awk '{print $5}' | tr -d %)
-LOG="/opt/so/log/sensor_clean.log"
-TODAY=$(date -u "+%Y-%m-%d")
+LOG="${LOG:-/opt/so/log/sensor_clean.log}"
+LOCK="${LOCK:-/var/tmp/so-sensor-clean.lock}"
+MAX_PASSES=100
 
+ZEEK_LOGS="$SENSOR_DIR/zeek/logs"
+STRELKA_FILES="$SENSOR_DIR/strelka/processed"
+SURICATA_LOGS="$SENSOR_DIR/suricata"
+PCAPS="$SENSOR_DIR/pcapout"
+
+log() {
+	echo "$(date) - $*" >>"$LOG"
+}
+
+disk_usage() {
+	df -P "$SENSOR_DIR" | tail -1 | awk '{print $5}' | tr -d %
+}
+
+disk_avail() {
+	df -P "$SENSOR_DIR" | tail -1 | awk '{print $4}'
+}
+
+# sets REMOVED=1 if anything was actually deleted
 clean() {
 	## find the oldest Zeek logs directory
-	OLDEST_DIR=$(ls /nsm/zeek/logs/ | grep -v "current" | grep -v "stats" | grep -v "packetloss" | grep -v "zeek_clean" | sort | head -n 1)
-	if [ -z "$OLDEST_DIR" -o "$OLDEST_DIR" == ".." -o "$OLDEST_DIR" == "." ]; then
-		echo "$(date) - No old Zeek logs available to clean up in /nsm/zeek/logs/" >>$LOG
-		#exit 0
-	else
-		echo "$(date) - Removing directory: /nsm/zeek/logs/$OLDEST_DIR" >>$LOG
-		rm -rf /nsm/zeek/logs/"$OLDEST_DIR"
+	OLDEST_DIR=$(ls "$ZEEK_LOGS" 2>/dev/null | grep -v "current" | grep -v "stats" | grep -v "packetloss" | grep -v "zeek_clean" | sort | head -n 1)
+	if [ -n "$OLDEST_DIR" ]; then
+		log "Removing directory: $ZEEK_LOGS/$OLDEST_DIR"
+		rm -rf "$ZEEK_LOGS/$OLDEST_DIR"
+		REMOVED=1
 	fi
 
 	## Remarking for now, as we are moving extracted files to /nsm/strelka/processed
@@ -43,58 +59,73 @@ clean() {
 	#fi
 
 	## Clean up Zeek extracted files processed by Strelka
-	STRELKA_FILES='/nsm/strelka/processed'
-	OLDEST_STRELKA=$(find $STRELKA_FILES -type f -printf '%T+ %p\n' | sort -n | head -n 1)
-	if [ -z "$OLDEST_STRELKA" -o "$OLDEST_STRELKA" == ".." -o "$OLDEST_STRELKA" == "." ]; then
-		echo "$(date) - No old files available to clean up in $STRELKA_FILES" >>$LOG
-	else
+	OLDEST_STRELKA=$(find "$STRELKA_FILES" -type f -printf '%T+ %p\n' 2>/dev/null | sort -n | head -n 1)
+	if [ -n "$OLDEST_STRELKA" ]; then
 		OLDEST_STRELKA_DATE=$(echo $OLDEST_STRELKA | awk '{print $1}' | cut -d+ -f1)
-		OLDEST_STRELKA_FILE=$(echo $OLDEST_STRELKA | awk '{print $2}')
-		echo "$(date) - Removing extracted files for $OLDEST_STRELKA_DATE" >>$LOG
-		find $STRELKA_FILES -type f -printf '%T+ %p\n' | grep $OLDEST_STRELKA_DATE | awk '{print $2}' | while read FILE; do
-			echo "$(date) - Removing file: $FILE" >>$LOG
+		log "Removing extracted files for $OLDEST_STRELKA_DATE"
+		REMOVED=1
+		find "$STRELKA_FILES" -type f -printf '%T+ %p\n' 2>/dev/null | grep $OLDEST_STRELKA_DATE | awk '{print $2}' | while read FILE; do
+			log "Removing file: $FILE"
 			rm -f "$FILE"
 		done
 	fi
 
 	## Clean up Suricata log files
-	SURICATA_LOGS='/nsm/suricata'
-	OLDEST_SURICATA=$(find $SURICATA_LOGS -type f -printf '%T+ %p\n' | sort -n | head -n 1)
-	if [[ -z "$OLDEST_SURICATA" ]] || [[ "$OLDEST_SURICATA" == ".." ]] || [[ "$OLDEST_SURICATA" == "." ]]; then
-		echo "$(date) - No old files available to clean up in $SURICATA_LOGS" >>$LOG
-	else
+	OLDEST_SURICATA=$(find "$SURICATA_LOGS" -type f -printf '%T+ %p\n' 2>/dev/null | sort -n | head -n 1)
+	if [ -n "$OLDEST_SURICATA" ]; then
 		OLDEST_SURICATA_DATE=$(echo $OLDEST_SURICATA | awk '{print $1}' | cut -d+ -f1)
-		OLDEST_SURICATA_FILE=$(echo $OLDEST_SURICATA | awk '{print $2}')
-		echo "$(date) - Removing logs for $OLDEST_SURICATA_DATE" >>$LOG
-		find $SURICATA_LOGS -type f -printf '%T+ %p\n' | grep $OLDEST_SURICATA_DATE | awk '{print $2}' | while read FILE; do
-			echo "$(date) - Removing file: $FILE" >>$LOG
+		log "Removing logs for $OLDEST_SURICATA_DATE"
+		REMOVED=1
+		find "$SURICATA_LOGS" -type f -printf '%T+ %p\n' 2>/dev/null | grep $OLDEST_SURICATA_DATE | awk '{print $2}' | while read FILE; do
+			log "Removing file: $FILE"
 			rm -f "$FILE"
 		done
 	fi
 
 	## Clean up extracted pcaps
-	PCAPS='/nsm/pcapout'
-	OLDEST_PCAP=$(find $PCAPS -type f -printf '%T+ %p\n' | sort -n | head -n 1)
-	if [ -z "$OLDEST_PCAP" -o "$OLDEST_PCAP" == ".." -o "$OLDEST_PCAP" == "." ]; then
-		echo "$(date) - No old files available to clean up in $PCAPS" >>$LOG
-	else
+	OLDEST_PCAP=$(find "$PCAPS" -type f -printf '%T+ %p\n' 2>/dev/null | sort -n | head -n 1)
+	if [ -n "$OLDEST_PCAP" ]; then
 		OLDEST_PCAP_DATE=$(echo $OLDEST_PCAP | awk '{print $1}' | cut -d+ -f1)
-		OLDEST_PCAP_FILE=$(echo $OLDEST_PCAP | awk '{print $2}')
-		echo "$(date) - Removing extracted files for $OLDEST_PCAP_DATE" >>$LOG
-		find $PCAPS -type f -printf '%T+ %p\n' | grep $OLDEST_PCAP_DATE | awk '{print $2}' | while read FILE; do
-			echo "$(date) - Removing file: $FILE" >>$LOG
+		log "Removing extracted files for $OLDEST_PCAP_DATE"
+		REMOVED=1
+		find "$PCAPS" -type f -printf '%T+ %p\n' 2>/dev/null | grep $OLDEST_PCAP_DATE | awk '{print $2}' | while read FILE; do
+			log "Removing file: $FILE"
 			rm -f "$FILE"
 		done
 	fi
 }
 
-# Check to see if we are already running
-NUM_RUNNING=$(pgrep -cf "/bin/bash /usr/sbin/so-sensor-clean")
-[ "$NUM_RUNNING" -gt 1 ] && echo "$(date) - $NUM_RUNNING sensor clean script processes running...exiting." >>$LOG && exit 0
-
-if [ "$CUR_USAGE" -gt "$CRIT_DISK_USAGE" ]; then
-	while [ "$CUR_USAGE" -gt "$CRIT_DISK_USAGE" ]; do
-		clean
-		CUR_USAGE=$(df -P $SENSOR_DIR | tail -1 | awk '{print $5}' | tr -d %)
-	done
+# Only one instance at a time; the lock is the fd, so it releases on any exit
+exec 9>"$LOCK" || exit 1
+if ! flock -n 9; then
+	log "another so-sensor-clean is already running (lock $LOCK held); exiting"
+	exit 0
 fi
+
+CUR_USAGE=$(disk_usage)
+[ "$CUR_USAGE" -gt "$CRIT_DISK_USAGE" ] || exit 0
+
+log "$SENSOR_DIR at ${CUR_USAGE}% (threshold ${CRIT_DISK_USAGE}%); starting cleanup"
+
+PASS=0
+while [ "$CUR_USAGE" -gt "$CRIT_DISK_USAGE" ]; do
+	PASS=$((PASS + 1))
+	if [ "$PASS" -gt "$MAX_PASSES" ]; then
+		log "stopping after $MAX_PASSES passes; $SENSOR_DIR still at ${CUR_USAGE}%"
+		break
+	fi
+
+	REMOVED=0
+	BEFORE=$(disk_avail)
+	clean
+	CUR_USAGE=$(disk_usage)
+
+	if [ "$REMOVED" -eq 0 ]; then
+		log "nothing left to remove in $ZEEK_LOGS, $STRELKA_FILES, $SURICATA_LOGS, $PCAPS; $SENSOR_DIR still at ${CUR_USAGE}% - space is consumed outside of NSM cleanup scope"
+		break
+	fi
+	if [ "$(disk_avail)" -le "$BEFORE" ]; then
+		log "pass $PASS freed no space; $SENSOR_DIR still at ${CUR_USAGE}% - stopping until next run"
+		break
+	fi
+done

salt/zeek/defaults.yaml

@@ -18,7 +18,6 @@ zeek:
       StatsLogEnable: 0
       StatsLogExpireInterval: 0
       StatusCmdShowAll: 0
-      StopWait: 1
       CrashExpireInterval: 0
       SitePolicyScripts: local.zeek
       LogDir: /nsm/zeek/logs

salt/zeek/disabled.sls

@@ -9,19 +9,9 @@
 include:
   - zeek.sostatus
   
-# Stop first so the entrypoint's SIGTERM trap can archive the final logs; docker_container.absent
-# with force is a 'docker rm -f', which never delivers SIGTERM. force stays so the state still
-# converges if the stop overruns.
-so-zeek_stopped:
-  docker_container.stopped:
-    - name: so-zeek
-    - error_on_absent: False
-
 so-zeek:
   docker_container.absent:
     - force: True
-    - require:
-      - docker_container: so-zeek_stopped
 
 so-zeek_so-status.disabled:
   file.comment:

salt/zeek/enabled.sls

@@ -19,10 +19,6 @@ so-zeek:
     - restart_policy: unless-stopped
     - start: True
     - privileged: True
-    # Docker's default 10s grace is not enough for the entrypoint's SIGTERM trap to run
-    # 'zeekctl stop' and let StopWait archive the final logs. Overrunning it means SIGKILL,
-    # which strands those logs in spool/tmp and marks every node crashed on the next start.
-    - stop_timeout: 180
     {% if DOCKERMERGED.containers['so-zeek'].ulimits %}
     - ulimits:
     {%   for ULIMIT in DOCKERMERGED.containers['so-zeek'].ulimits %}

salt/zeek/soc_zeek.yaml

@@ -99,18 +99,6 @@ zeek:
         regexFailureMessage: You must enter a whole number of days, or 0 to keep crash directories forever.
         helpLink: zeek
         advanced: True
-      StopWait:
-        description: >-
-          Set to 1 to make "zeekctl stop" wait for the final logs to be archived instead of
-          letting that finish in the background. Security Onion stops Zeek by stopping its
-          container, so anything still running in the background is killed when the container
-          exits - without this, the last logs of each run are stranded unarchived in
-          /nsm/zeek/spool/tmp and never reach Elasticsearch. It is read only for that reason.
-        regex: ^[01]$
-        regexFailureMessage: You must enter 0 or 1.
-        helpLink: zeek
-        advanced: True
-        readonly: True
       MinDiskSpace:
         description: >-
           Percentage of free disk space below which ZeekControl reports a warning, or 0 to disable the check