From 85f605a7620392bfc8bca7ef8eff7864bddaffc3 Mon Sep 17 00:00:00 2001
From: Eric MacDonald <eric.macdonald@windriver.com>
Date: Thu, 15 Oct 2020 15:51:04 -0400
Subject: [PATCH] Add /var/crash dump management to maintenance.

The Linux kernel can be configured to perform a
crash dump and reboot in response to specific,
typically serious, events.

A crash dump event produces a crash dump report
bundle (directory) of files that represent the
state of the kernel at the time of the event ;
usefull for post-event root cause analysis.

The kernel directs new crash dump bundles to
/var/crash/<dated vmcore bundle>. Crash dump
bundles are quite large and, if too many occur,
can fill up its target filesystem.

This update adds crash dump bundle management
to the maintenance with a new crashDumpMgr
service script and installs a crash dump
logrotation configuration file to
compress/preserve the first crash bundle and
compress/rotate all subsequent bundles.

With repeated crash dumps and the help of
backgroud logrotation this update produces
the following compressed crash dump bundles

controller-1:~$ ls -lrth /var/log/crash
total 238M
-rw-r--r-- 1 root 77M <date> vmcore_first.tar.1.gz
-rw-r--r-- 1 root 75M <date> vmcore.tar.1.gz

Change-Id: I2741e610c6c417d7fc14dfada283a1edacd9327f
Partial-Fix: 1898602
Signed-off-by: Eric MacDonald <eric.macdonald@windriver.com>
---
 mtce/centos/mtce.spec                 |   7 ++
 mtce/src/scripts/crashDumpMgr         | 131 ++++++++++++++++++++++++++
 mtce/src/scripts/crashDumpMgr.service |  13 +++
 mtce/src/scripts/crashdump.logrotate  |  14 +++
 4 files changed, 165 insertions(+)
 create mode 100644 mtce/src/scripts/crashDumpMgr
 create mode 100644 mtce/src/scripts/crashDumpMgr.service
 create mode 100644 mtce/src/scripts/crashdump.logrotate

diff --git a/mtce/centos/mtce.spec b/mtce/centos/mtce.spec
index d7b0c683..96211df0 100644
--- a/mtce/centos/mtce.spec
+++ b/mtce/centos/mtce.spec
@@ -349,6 +349,8 @@ install -m 700 -p -D %{_buildsubdir}/alarm/scripts/mtcalarm.init %{buildroot}%{_
 # TODO: Init hack. Should move to proper module
 install -m 755 -p -D %{_buildsubdir}/scripts/hwclock.sh %{buildroot}%{_sysconfdir}/init.d/hwclock.sh
 install -m 644 -p -D %{_buildsubdir}/scripts/hwclock.service %{buildroot}%{_unitdir}/hwclock.service
+install -m 755 -p -D %{_buildsubdir}/scripts/crashDumpMgr %{buildroot}%{_sysconfdir}/init.d/crashDumpMgr
+install -m 644 -p -D %{_buildsubdir}/scripts/crashDumpMgr.service %{buildroot}%{_unitdir}/crashDumpMgr.service
 
 # systemd service files
 install -m 644 -p -D %{_buildsubdir}/fsmon/scripts/fsmon.service %{buildroot}%{_unitdir}/fsmon.service
@@ -395,6 +397,7 @@ install -m 644 -p -D %{_buildsubdir}/lmon/scripts/lmon.pmon.conf %{buildroot}%{l
 
 # log rotation
 install -m 755 -d %{buildroot}%{_sysconfdir}/logrotate.d
+install -m 644 -p -D %{_buildsubdir}/scripts/crashdump.logrotate %{buildroot}%{local_etc_logrotated}/crashdump.logrotate
 install -m 644 -p -D %{_buildsubdir}/scripts/mtce.logrotate %{buildroot}%{local_etc_logrotated}/mtce.logrotate
 install -m 644 -p -D %{_buildsubdir}/hostw/scripts/hostw.logrotate %{buildroot}%{local_etc_logrotated}/hostw.logrotate
 install -m 644 -p -D %{_buildsubdir}/pmon/scripts/pmon.logrotate %{buildroot}%{local_etc_logrotated}/pmon.logrotate
@@ -424,6 +427,7 @@ install -m 755 -d %{buildroot}/var/run
 /bin/systemctl enable rsyncd.service
 /bin/systemctl enable goenabled.service
 /bin/systemctl enable mtcalarm.service
+/bin/systemctl enable crashDumpMgr.service
 
 %post -n mtce-hostw
 /bin/systemctl enable hostw.service
@@ -470,6 +474,7 @@ install -m 755 -d %{buildroot}/var/run
 %{local_etc_logrotated}/fsmon.logrotate
 %{local_etc_logrotated}/mtce.logrotate
 %{local_etc_logrotated}/mtcalarm.logrotate
+%{local_etc_logrotated}/crashdump.logrotate
 
 # Maintenance start/stop services scripts
 %{local_etc_servicesd}/controller/mtcTest
@@ -490,6 +495,7 @@ install -m 755 -d %{buildroot}/var/run
 %{_sysconfdir}/init.d/mtcClient
 %{_sysconfdir}/init.d/mtcalarm
 %{_sysconfdir}/init.d/hwclock.sh
+%{_sysconfdir}/init.d/crashDumpMgr
 
 %{_unitdir}/runservices.service
 %{_unitdir}/goenabled.service
@@ -499,6 +505,7 @@ install -m 755 -d %{buildroot}/var/run
 %{_unitdir}/mtcClient.service
 %{_unitdir}/hbsClient.service
 %{_unitdir}/hwclock.service
+%{_unitdir}/crashDumpMgr.service
 
 # Binaries
 %{local_bindir}/mtcAgent
diff --git a/mtce/src/scripts/crashDumpMgr b/mtce/src/scripts/crashDumpMgr
new file mode 100644
index 00000000..d57be7ba
--- /dev/null
+++ b/mtce/src/scripts/crashDumpMgr
@@ -0,0 +1,131 @@
+#!/bin/bash
+#
+# Copyright (c) 2020 Wind River Systems, Inc.
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+
+# chkconfig: 2345 98 2
+#
+### BEGIN INIT INFO
+# Provides: crashDumpMgr
+# Required-Start: $null
+# Required-Stop: $null
+# Default-Start:     3 5
+# Default-Stop:      0 1 2 6
+# Short-Description: Maintenance 'Crash Dump' Manager script
+### END INIT INFO
+
+CRASHDUMPMGR_TAG=${CRASHDUMPMGR_TAG:-"crashDumpMgr"}
+
+RETVAL=0
+
+#############################################################################
+# Log message to syslog
+#############################################################################
+
+function log()
+{
+    logger -t ${CRASHDUMPMGR_TAG} $@
+}
+
+#############################################################################
+#
+# Name   : manage_crash_dumps
+#
+# Purpose: Prevent crash dumps from filling up the root fs
+#
+# The kernel directs new crash dump bundles to
+# /var/crash/<dated vmcore bundle>. Crash dump
+# bundles are quite large and, if too many occur,
+# can fill up its target filesystem.
+#
+# This function nicely tars a crash bundle found in /var/crash
+# to /var/log/crash.
+#
+# The first bundle is tar'ed as vmcore_first.tar and preserved.
+# Subsequent crash bundles are nicely tar'ed as vmcore.tar
+#
+# Save the crash dump vmcore summary for all crash dumps.
+#
+# Assumptions: logration is used to compress these bundles in the background
+#
+############################################################################
+
+function manage_crash_dumps()
+{
+    CRASH_DIR="/var/crash"
+    CRASH_BUNDLE_DIR="/var/log/crash"
+    OTHER_BUNDLE="${CRASH_BUNDLE_DIR}/vmcore.tar"
+    FIRST_BUNDLE="${CRASH_BUNDLE_DIR}/vmcore_first.tar"
+    FIRST_BUNDLE_ROTATED="${CRASH_BUNDLE_DIR}/vmcore_first.tar.1.gz"
+    CRASH_BUNDLE_SUMMARY="vmcore-dmesg.txt"
+
+    # tar command and nice levels
+    TAR_CMD="tar -cf"
+    NICE_CMD="/usr/bin/nice -n19"
+    IONICE_CMD="/usr/bin/ionice -c2 -n7"
+
+    log "managing ${CRASH_DIR}"
+    cleanup=false
+
+    # create dir if it does not exist
+    if [ ! -d ${CRASH_BUNDLE_DIR} ] ; then
+        mkdir ${CRASH_BUNDLE_DIR}
+    fi
+
+    for entry in ${CRASH_DIR}/*
+    do
+        if [ -d ${entry} ] ; then
+            if [ -e ${entry}/vmcore ] ; then
+
+                # save the crash dump vmcore summary for all crash dumps
+                cp -a ${entry}/${CRASH_BUNDLE_SUMMARY} ${CRASH_DIR}/$(basename ${entry})_${CRASH_BUNDLE_SUMMARY}
+
+                if [ "${cleanup}" != true ] ; then
+                    if [ -e ${FIRST_BUNDLE} -o -e ${FIRST_BUNDLE_ROTATED} ] ; then
+                        if [ ! -e ${OTHER_BUNDLE} ] ; then
+                            log "creating bundle from ${entry}"
+                            ${IONICE_CMD} ${NICE_CMD} ${TAR_CMD} ${OTHER_BUNDLE} -C ${CRASH_DIR} $(basename ${entry})
+                            cleanup=true
+                        fi
+                    else
+                        log "creating first bundle from ${entry}"
+                        ${IONICE_CMD} ${NICE_CMD} ${TAR_CMD} ${FIRST_BUNDLE} -C ${CRASH_DIR} $(basename ${entry})
+                        cleanup=true
+                    fi
+                fi
+                log "removing ${entry}"
+                rm -rf "${entry}"
+            fi
+        fi
+    done
+}
+
+# service case
+case "$1" in
+    start)
+        manage_crash_dumps
+        ;;
+
+    stop)
+        log "stop"
+        ;;
+
+    restart)
+        log "restart"
+        stop
+        start
+        ;;
+
+    status)
+        log "status"
+        ;;
+
+    *)
+        log "usage: $0 { start | stop | status | restart }"
+        RETVAL=1
+        ;;
+esac
+
+exit $RETVAL
diff --git a/mtce/src/scripts/crashDumpMgr.service b/mtce/src/scripts/crashDumpMgr.service
new file mode 100644
index 00000000..49f979e8
--- /dev/null
+++ b/mtce/src/scripts/crashDumpMgr.service
@@ -0,0 +1,13 @@
+[Unit]
+Description=Crash Dump Manager
+After=network.target
+Before=sshd.service
+
+[Service]
+Type=oneshot
+RemainAfterExit=no
+ExecStart=/etc/init.d/crashDumpMgr start
+ExecStop=/etc/init.d/crashDumpMgr stop
+
+[Install]
+WantedBy=multi-user.target
diff --git a/mtce/src/scripts/crashdump.logrotate b/mtce/src/scripts/crashdump.logrotate
new file mode 100644
index 00000000..a16bcb7c
--- /dev/null
+++ b/mtce/src/scripts/crashdump.logrotate
@@ -0,0 +1,14 @@
+/var/log/crash/vmcore.tar
+/var/log/crash/vmcore_first.tar
+{
+    nodateext
+    size 1K
+    start 1
+    rotate 1
+    missingok
+    notifempty
+    compress
+    postrotate
+      rm -f $1
+    endscript
+}