306 lines
11 KiB
Bash
306 lines
11 KiB
Bash
#!/bin/bash
|
|
#
|
|
# Copyright (c) 2020-2023 Wind River Systems, Inc.
|
|
#
|
|
# SPDX-License-Identifier: Apache-2.0
|
|
#
|
|
|
|
CRASHDUMPMGR_TAG=${CRASHDUMPMGR_TAG:-"crashDumpMgr"}
|
|
|
|
RETVAL=0
|
|
|
|
# Default values and constants
|
|
DEFAULT_MAX_SIZE=5368709120 # "5GiB"
|
|
DEFAULT_MAX_FILES=4
|
|
UNLIMITED="unlimited"
|
|
DEFAULT_MAX_USED="${UNLIMITED}" # Assign UNLIMITED to DEFAULT_MAX_USED
|
|
DEFAULT_MIN_REMAINDER_PERCENT=10
|
|
MIN_REMAINDER_MINIMUM=1073741824 # 1GiB in bytes
|
|
|
|
# number format to/from human readable commands.
|
|
NUMFMT_TO_HR="/usr/bin/numfmt --to=iec"
|
|
NUMFMT_FROM_HR="/usr/bin/numfmt --from=auto"
|
|
|
|
#############################################################################
|
|
# 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 manages the crash dump bundles, creating tar archives for
|
|
# storage, handling maximum file count and storage limits, and preserving
|
|
# summaries of the crash dumps.
|
|
#
|
|
# The first bundle is tar'ed as vmcore_first.tar and preserved.
|
|
# Subsequent crash bundles are nicely tar'ed as vmcore_<date>.tar.
|
|
#
|
|
# Save the crash dump vmcore summary for all crash dumps.
|
|
#
|
|
# Assumptions: log rotation is used to compress these bundles in the background
|
|
#
|
|
# Parameters:
|
|
# $1 = max_size ; maximum vmcore size to keep (human-readable size or "unlimited")
|
|
# $2 = max_files ; maximum number of crash dump files to keep
|
|
# $3 = max_used ; maximum used storage size (human-readable size or "unlimited")
|
|
# $4 = min_available ; minimum available storage size (human-readable size)
|
|
#
|
|
############################################################################
|
|
|
|
|
|
function manage_crash_dumps()
|
|
{
|
|
CRASH_DIR="/var/crash"
|
|
CRASH_BUNDLE_DIR="/var/log/crash"
|
|
OTHER_BUNDLE="${CRASH_BUNDLE_DIR}/vmcore"
|
|
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"
|
|
CRASH_BUNDLE_SUMMARY_DEB="dmesg."
|
|
|
|
if [ "${4}" = "" ] ; then
|
|
# Get the size of the filesystem assigned to /var/log/crash in bytes
|
|
fs_size=$(df -B1 ${CRASH_BUNDLE_DIR} | awk 'NR==2 {print $2}')
|
|
# Calculate min_available as the percentage of the filesystem size
|
|
min_available=$((${fs_size} * ${DEFAULT_MIN_REMAINDER_PERCENT} / 100))
|
|
else
|
|
min_available=${4}
|
|
fi
|
|
|
|
# Set a minimum value for min_available
|
|
if [ ${min_available} -lt ${MIN_REMAINDER_MINIMUM} ]; then
|
|
min_available=${MIN_REMAINDER_MINIMUM}
|
|
fi
|
|
|
|
max_size_hr=""
|
|
max_used_hr=""
|
|
if [ "${max_size}" != "${UNLIMITED}" ]; then
|
|
max_size_hr="($(${NUMFMT_TO_HR} ${max_size}))"
|
|
fi
|
|
if [ "${max_used}" != "${UNLIMITED}" ]; then
|
|
max_used_hr="($(${NUMFMT_TO_HR} ${max_used}))"
|
|
fi
|
|
min_available_hr=$(${NUMFMT_TO_HR} ${min_available})
|
|
|
|
log "max crash dump files set to ${max_files}"
|
|
log "max crash dump vmcore size is ${max_size} ${max_size_hr}"
|
|
log "max used storage size is ${max_used} ${max_used_hr}"
|
|
log "minimum available storage size is ${min_available} (${min_available_hr})"
|
|
|
|
# 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}"
|
|
|
|
# create dir if it does not exist
|
|
if [ ! -d ${CRASH_BUNDLE_DIR} ] ; then
|
|
mkdir ${CRASH_BUNDLE_DIR}
|
|
fi
|
|
for entry in $(ls -rt ${CRASH_DIR}/);
|
|
do
|
|
entry=${CRASH_DIR}/${entry}
|
|
remove_entry=false
|
|
max_files_saved=false
|
|
if [ -d "${entry}" ] ; then
|
|
time=${entry##*/}
|
|
if [ -e "${entry}/${CRASH_BUNDLE_SUMMARY_DEB}${time}" ] ; then
|
|
log "saving summary: ${CRASH_DIR}/$(basename ${time})_${CRASH_BUNDLE_SUMMARY_DEB}${time}"
|
|
|
|
# save the crash dump dmesg.<date> for debian for all crash dumps
|
|
cp -a ${entry}/${CRASH_BUNDLE_SUMMARY_DEB}${time} ${CRASH_DIR}/$(basename ${time})_${CRASH_BUNDLE_SUMMARY_DEB}${time}
|
|
fi
|
|
if [ -e "${entry}/${CRASH_BUNDLE_SUMMARY}" ] ; then
|
|
log "saving summary: ${CRASH_DIR}/$(basename ${entry})_${CRASH_BUNDLE_SUMMARY}"
|
|
|
|
# save the crash dump vmcore summary for all crash dumps
|
|
cp -a ${entry}/${CRASH_BUNDLE_SUMMARY} ${CRASH_DIR}/$(basename ${entry})_${CRASH_BUNDLE_SUMMARY}
|
|
fi
|
|
if [ -e "${entry}/dump.${time}" ] || [ -e "${entry}/vmcore" ] ; then
|
|
|
|
# get the size of this vmcore file ; raw and human readable
|
|
if [ -e "${entry}/dump.${time}" ] ; then
|
|
vmcore_size=$(stat --format='%s' ${entry}/dump.${time})
|
|
else
|
|
vmcore_size=$(stat --format='%s' ${entry}/vmcore)
|
|
fi
|
|
vmcore_size_hr=$(${NUMFMT_TO_HR} ${vmcore_size})
|
|
|
|
# Manage max number of files
|
|
if [ "$(ls -A ${CRASH_BUNDLE_DIR} | wc -l)" -ge "${max_files}" ]; then
|
|
oldest_vmcore=$(ls -t ${CRASH_BUNDLE_DIR} | tail -2 | head -1)
|
|
oldest_vmcore_size=$(stat --format='%s' ${oldest_vmcore})
|
|
max_files_saved=true
|
|
fi
|
|
|
|
# get available ${CRASH_BUNDLE_DIR} fs space in bytes
|
|
available=$(df -B1 ${CRASH_BUNDLE_DIR} | grep -v Available | awk '{ print $4 }')
|
|
|
|
# get the current used space in the ${CRASH_BUNDLE_DIR} fs
|
|
used_space=$(du -sb ${CRASH_BUNDLE_DIR} | awk '{print $1}')
|
|
|
|
# if the ${CRASH_BUNDLE_DIR} contains the maximum number of files, the available and used_space
|
|
# need to be updated to the value after deleting the oldest crash dump file.
|
|
if [ "${max_files_saved}" = true ] ; then
|
|
available=$((available+oldest_vmcore_size))
|
|
used_space=$((used_space-oldest_vmcore_size))
|
|
fi
|
|
available_hr=$(${NUMFMT_TO_HR} ${available})
|
|
log "new vmcore detected (size:${vmcore_size}:${vmcore_size_hr}) ;" \
|
|
"${CRASH_BUNDLE_DIR} avail:${available}:${available_hr}"
|
|
|
|
# Don't save this crash dump if it would leave the
|
|
# ${CRASH_BUNDLE_DIR} filesystem with less than 1GiB.
|
|
if [ ${available} -gt ${vmcore_size} ]; then
|
|
remaining=$((available-vmcore_size))
|
|
else
|
|
remaining=0
|
|
fi
|
|
|
|
# Check if adding the file would exceed the maximum used space limit
|
|
total_used_space=$((used_space + file_size))
|
|
if [ "${max_used}" != "$UNLIMITED" ] && [ ${total_used_space} -gt ${max_used} ]; then
|
|
|
|
log "The last crash dump is not saved because it would exceed the maximum" \
|
|
"used space limit specified in the max_used parameter (${max_used} bytes)."
|
|
remove_entry=true
|
|
|
|
# check for min required 'remaining' ${CRASH_BUNDLE_DIR} filesystem space
|
|
elif [ ${remaining} -lt ${min_available} ] ; then
|
|
log "insufficient space in ${CRASH_BUNDLE_DIR} for ${vmcore_size_hr} ${entry};" \
|
|
"would leave only ${remaining} bytes"
|
|
remove_entry=true
|
|
|
|
# create a new crash bundle if the vmcore file isn't oversized
|
|
elif [ ${vmcore_size} -lt ${max_size} ] || [ "${max_size}" = "${UNLIMITED}" ] ; then
|
|
|
|
if [ ! -e "${FIRST_BUNDLE}" ] && [ ! -e "${FIRST_BUNDLE_ROTATED}" ]; then
|
|
log "creating first bundle from ${entry}"
|
|
${IONICE_CMD} ${NICE_CMD} ${TAR_CMD} ${FIRST_BUNDLE} -C ${CRASH_DIR} $(basename ${entry})
|
|
|
|
else
|
|
if [ "${max_files_saved}" = true ] ; then
|
|
# delete oldest vmcore file
|
|
log "removing oldest vmcore file: ${oldest_vmcore}"
|
|
rm -rf "${CRASH_BUNDLE_DIR}/${oldest_vmcore}"
|
|
fi
|
|
log "creating bundle from ${entry}"
|
|
${IONICE_CMD} ${NICE_CMD} ${TAR_CMD} ${OTHER_BUNDLE}_${time}.tar -C ${CRASH_DIR} $(basename ${entry})
|
|
fi
|
|
remove_entry=true
|
|
else
|
|
log "deleting oversize (${vmcore_size_hr}) vmcore file $(basename ${entry})"
|
|
remove_entry=true
|
|
fi
|
|
elif [[ "$entry" == *"_dmesg."* ]] || [[ "$entry" == *"_vmcore-dmesg.txt"* ]] ; then
|
|
log "saved old $entry summary"
|
|
elif [[ "$entry" != "$CRASH_DIR/*" ]] ; then
|
|
# removes vmcore files not named properly
|
|
# i.e vmcore.incomplete
|
|
remove_entry=true
|
|
fi
|
|
elif [[ "$entry" != *"_dmesg."* ]] && [[ "$entry" != *"_vmcore-dmesg.txt"* ]] ; then
|
|
# removes files in /var/crash that are not crash dumps related
|
|
remove_entry=true
|
|
fi
|
|
|
|
if [ "${remove_entry}" = true ] ; then
|
|
log "removing ${entry}"
|
|
rm -rf "${entry}"
|
|
fi
|
|
done
|
|
}
|
|
|
|
function print_help() {
|
|
echo "Usage: $(basename "$0") [OPTIONS]"
|
|
echo "Options:"
|
|
echo " --max-size <size> Set maximum vmcore size (human-readable size or \"$UNLIMITED\")"
|
|
echo " --max-files <number> Set maximum number of crash dump files"
|
|
echo " --max-used <size> Set maximum used storage size (human-readable size or \"$UNLIMITED\")"
|
|
echo " --min-available <size> Set minimum available storage size (human-readable size)"
|
|
}
|
|
|
|
normalize_size_format() {
|
|
echo "$1" | tr ',' '.'
|
|
}
|
|
|
|
# Initialize default values
|
|
max_size="${DEFAULT_MAX_SIZE}"
|
|
max_files="${DEFAULT_MAX_FILES}"
|
|
max_used="${DEFAULT_MAX_USED}"
|
|
min_available=""
|
|
|
|
# Initialize a flag to indicate if help has been shown
|
|
help_shown=false
|
|
|
|
# Parse the command line
|
|
while [[ $# -gt 0 ]]; do
|
|
case "${1}" in
|
|
-h|--help)
|
|
print_help
|
|
help_shown=true
|
|
shift
|
|
;;
|
|
|
|
--max-size)
|
|
shift
|
|
if [ "${1}" = "${UNLIMITED}" ]; then
|
|
max_size=${UNLIMITED}
|
|
else
|
|
max_size=$(${NUMFMT_FROM_HR} "$(normalize_size_format "${1}")")
|
|
fi
|
|
shift
|
|
;;
|
|
|
|
--max-files)
|
|
shift
|
|
max_files="${1}"
|
|
shift
|
|
;;
|
|
|
|
--max-used)
|
|
shift
|
|
if [ "${1}" = "${UNLIMITED}" ]; then
|
|
max_used=${UNLIMITED}
|
|
else
|
|
max_used=$(${NUMFMT_FROM_HR} "$(normalize_size_format "${1}")")
|
|
fi
|
|
shift
|
|
;;
|
|
|
|
--min-available)
|
|
shift
|
|
min_available=$(${NUMFMT_FROM_HR} "$(normalize_size_format "${1}")")
|
|
shift
|
|
;;
|
|
|
|
*)
|
|
if [ "${help_shown}" = false ]; then
|
|
print_help
|
|
help_shown=true
|
|
else
|
|
# Handle unknown options or arguments
|
|
shift
|
|
fi
|
|
;;
|
|
esac
|
|
done
|
|
|
|
manage_crash_dumps ${max_size} ${max_files} ${max_used} ${min_available}
|
|
|
|
exit $RETVAL
|