| 1 | #!/bin/bash
|
|---|
| 2 | # $Id: testbox-maintenance.sh 64523 2016-11-02 20:07:37Z vboxsync $
|
|---|
| 3 | ## @file
|
|---|
| 4 | # VirtualBox Validation Kit - testbox mainenance service
|
|---|
| 5 | #
|
|---|
| 6 |
|
|---|
| 7 | #
|
|---|
| 8 | # Copyright (C) 2006-2016 Oracle Corporation
|
|---|
| 9 | #
|
|---|
| 10 | # This file is part of VirtualBox Open Source Edition (OSE), as
|
|---|
| 11 | # available from http://www.virtualbox.org. This file is free software;
|
|---|
| 12 | # you can redistribute it and/or modify it under the terms of the GNU
|
|---|
| 13 | # General Public License (GPL) as published by the Free Software
|
|---|
| 14 | # Foundation, in version 2 as it comes in the "COPYING" file of the
|
|---|
| 15 | # VirtualBox OSE distribution. VirtualBox OSE is distributed in the
|
|---|
| 16 | # hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
|
|---|
| 17 | #
|
|---|
| 18 | # The contents of this file may alternatively be used under the terms
|
|---|
| 19 | # of the Common Development and Distribution License Version 1.0
|
|---|
| 20 | # (CDDL) only, as it comes in the "COPYING.CDDL" file of the
|
|---|
| 21 | # VirtualBox OSE distribution, in which case the provisions of the
|
|---|
| 22 | # CDDL are applicable instead of those of the GPL.
|
|---|
| 23 | #
|
|---|
| 24 | # You may elect to license modified versions of this file under the
|
|---|
| 25 | # terms and conditions of either the GPL or the CDDL or both.
|
|---|
| 26 | #
|
|---|
| 27 |
|
|---|
| 28 |
|
|---|
| 29 | #
|
|---|
| 30 | # Global Variables (config first).
|
|---|
| 31 | #
|
|---|
| 32 | MY_REBOOT_WHEN_DONE="yes"
|
|---|
| 33 | #MY_REBOOT_WHEN_DONE=""
|
|---|
| 34 |
|
|---|
| 35 | MY_TFTP_ROOT="/mnt/testbox-tftp"
|
|---|
| 36 | MY_BACKUP_ROOT="/mnt/testbox-backup"
|
|---|
| 37 | MY_BACKUP_MNT_TEST_FILE="/mnt/testbox-backup/testbox-backup"
|
|---|
| 38 | MY_GLOBAL_LOG_FILE="${MY_BACKUP_ROOT}/maintenance.log"
|
|---|
| 39 |
|
|---|
| 40 | MY_IP=""
|
|---|
| 41 | MY_BACKUP_DIR=""
|
|---|
| 42 | MY_LOG_FILE=""
|
|---|
| 43 | MY_PXELINUX_CFG_FILE=""
|
|---|
| 44 |
|
|---|
| 45 |
|
|---|
| 46 | ##
|
|---|
| 47 | # Info message.
|
|---|
| 48 | #
|
|---|
| 49 | InfoMsg()
|
|---|
| 50 | {
|
|---|
| 51 | echo "testbox-maintenance.sh: debug:" $*;
|
|---|
| 52 | if test -n "${MY_LOG_FILE}"; then
|
|---|
| 53 | echo "`date -uIsec`: ${MY_IP}: info:" $* >> ${MY_LOG_FILE};
|
|---|
| 54 | fi
|
|---|
| 55 | }
|
|---|
| 56 |
|
|---|
| 57 |
|
|---|
| 58 | ##
|
|---|
| 59 | # Error message and reboot+exit. First argument is exit code.
|
|---|
| 60 | #
|
|---|
| 61 | ErrorMsgExit()
|
|---|
| 62 | {
|
|---|
| 63 | MY_RET=$1
|
|---|
| 64 | shift
|
|---|
| 65 | echo "testbox-maintenance.sh: error:" $* >&2;
|
|---|
| 66 | # Append to the testbox log.
|
|---|
| 67 | if test -n "${MY_LOG_FILE}"; then
|
|---|
| 68 | echo "`date -uIsec`: ${MY_IP}: error:" $* >> "${MY_LOG_FILE}";
|
|---|
| 69 | fi
|
|---|
| 70 | # Append to the global log.
|
|---|
| 71 | if test -f "${MY_BACKUP_MNT_TEST_FILE}"; then
|
|---|
| 72 | echo "`date -uIsec`: ${MY_IP}: error:" $* >> "${MY_GLOBAL_LOG_FILE}";
|
|---|
| 73 | fi
|
|---|
| 74 |
|
|---|
| 75 | #
|
|---|
| 76 | # On error we normally wait 5min before rebooting to avoid repeating the
|
|---|
| 77 | # same error too many time before the admin finds out. We choose NOT to
|
|---|
| 78 | # remove the PXE config file here because (a) the admin might otherwise
|
|---|
| 79 | # not notice something went wrong, (b) the system could easily be in a
|
|---|
| 80 | # weird unbootable state, (c) the problem might be temporary.
|
|---|
| 81 | #
|
|---|
| 82 | # While debugging, we just exit here.
|
|---|
| 83 | #
|
|---|
| 84 | if test -n "${MY_REBOOT_WHEN_DONE}"; then
|
|---|
| 85 | sleep 5m
|
|---|
| 86 | echo "testbox-maintenance.sh: rebooting (after error)" >&2;
|
|---|
| 87 | reboot
|
|---|
| 88 | fi
|
|---|
| 89 | exit ${MY_RET}
|
|---|
| 90 | }
|
|---|
| 91 |
|
|---|
| 92 | #
|
|---|
| 93 | # Try figure out the IP address of the box and the hostname from it again.
|
|---|
| 94 | #
|
|---|
| 95 | MY_IP=` hostname -I | cut -f1 -d' ' | head -1 `
|
|---|
| 96 | if test -z "${MY_IP}" -o `echo "${MY_IP}" | wc -w` -ne "1" -o "${MY_IP}" = "127.0.0.1"; then
|
|---|
| 97 | ErrorMsgExit 10 "Failed to get a good IP! (MY_IP=${MY_IP})"
|
|---|
| 98 | fi
|
|---|
| 99 | MY_HOSTNAME=`getent hosts "${MY_IP}" | sed -s 's/[[:space:]][[:space:]]*/ /g' | cut -d' ' -f2 `
|
|---|
| 100 | if test -z "${MY_HOSTNAME}"; then
|
|---|
| 101 | MY_HOSTNAME="unknown";
|
|---|
| 102 | fi
|
|---|
| 103 |
|
|---|
| 104 | # Derive the backup dir and log file name from it.
|
|---|
| 105 | if test ! -f "${MY_BACKUP_MNT_TEST_FILE}"; then
|
|---|
| 106 | ErrorMsgExit 11 "Backup directory is not mounted."
|
|---|
| 107 | fi
|
|---|
| 108 | MY_BACKUP_DIR="${MY_BACKUP_ROOT}/${MY_IP}"
|
|---|
| 109 | MY_LOG_FILE="${MY_BACKUP_DIR}/maintenance.log"
|
|---|
| 110 | mkdir -p "${MY_BACKUP_DIR}"
|
|---|
| 111 | InfoMsg "MY_IP=${MY_IP}<eol>"
|
|---|
| 112 | echo "`date -uIsec`: ${MY_IP}: ${MY_HOSTNAME} says hi." >> "${MY_GLOBAL_LOG_FILE}"
|
|---|
| 113 |
|
|---|
| 114 | #
|
|---|
| 115 | # Convert the IP address to PXELINUX hex format, then check that we've got
|
|---|
| 116 | # a config file on the TFTP share that we later can remove. We consider it a
|
|---|
| 117 | # fatal failure if we don't because we've probably got the wrong IP and we'll
|
|---|
| 118 | # be stuck doing the same stuff over and over again.
|
|---|
| 119 | #
|
|---|
| 120 | MY_TMP=`echo "${MY_IP}" | sed -e 's/\./ /g' `
|
|---|
| 121 | MY_IP_HEX=`printf "%02X%02X%02X%02X" ${MY_TMP}`
|
|---|
| 122 | InfoMsg "MY_IP_HEX=${MY_IP_HEX}<eol>"
|
|---|
| 123 |
|
|---|
| 124 | if test ! -f "${MY_TFTP_ROOT}/pxelinux.0"; then
|
|---|
| 125 | ErrorMsgExit 12 "TFTP share mounted or mixxing pxelinux.0 in the root."
|
|---|
| 126 | fi
|
|---|
| 127 |
|
|---|
| 128 | MY_PXELINUX_CFG_FILE="${MY_TFTP_ROOT}/pxelinux.cfg/${MY_IP_HEX}"
|
|---|
| 129 | if test ! -f "${MY_PXELINUX_CFG_FILE}"; then
|
|---|
| 130 | ErrorMsgExit 13 "No pxelinux.cfg file found (${MY_PXELINUX_CFG_FILE}) - wrong IP?"
|
|---|
| 131 | fi
|
|---|
| 132 |
|
|---|
| 133 | #
|
|---|
| 134 | # Dig the action out of from the kernel command line.
|
|---|
| 135 | #
|
|---|
| 136 | InfoMsg "/proc/cmdline: `cat /proc/cmdline`"
|
|---|
| 137 | set `cat /proc/cmdline`
|
|---|
| 138 | MY_ACTION=not-found
|
|---|
| 139 | while test $# -ge 1; do
|
|---|
| 140 | case "$1" in
|
|---|
| 141 | testbox-action-*)
|
|---|
| 142 | MY_ACTION="$1"
|
|---|
| 143 | ;;
|
|---|
| 144 | esac
|
|---|
| 145 | shift
|
|---|
| 146 | done
|
|---|
| 147 | if test "${MY_ACTION}" = "not-found"; then
|
|---|
| 148 | ErrorMsgExit 14 "No action given. Expected testbox-action-backup, testbox-action-backup-again, testbox-action-restore," \
|
|---|
| 149 | "testbox-action-refresh-info, or testbox-action-rescue on the kernel command line.";
|
|---|
| 150 | fi
|
|---|
| 151 |
|
|---|
| 152 | # Validate and shorten the action.
|
|---|
| 153 | case "${MY_ACTION}" in
|
|---|
| 154 | testbox-action-backup)
|
|---|
| 155 | MY_ACTION="backup";
|
|---|
| 156 | ;;
|
|---|
| 157 | testbox-action-backup-again)
|
|---|
| 158 | MY_ACTION="backup-again";
|
|---|
| 159 | ;;
|
|---|
| 160 | testbox-action-restore)
|
|---|
| 161 | MY_ACTION="restore";
|
|---|
| 162 | ;;
|
|---|
| 163 | testbox-action-refresh-info)
|
|---|
| 164 | MY_ACTION="refresh-info";
|
|---|
| 165 | ;;
|
|---|
| 166 | testbox-action-rescue)
|
|---|
| 167 | MY_ACTION="rescue";
|
|---|
| 168 | ;;
|
|---|
| 169 | *) ErrorMsgExit 15 "Invalid action '${MY_ACTION}'";
|
|---|
| 170 | ;;
|
|---|
| 171 | esac
|
|---|
| 172 |
|
|---|
| 173 | # Log the action in both logs.
|
|---|
| 174 | echo "`date -uIsec`: ${MY_IP}: info: Executing '${MY_ACTION}'." >> "${MY_GLOBAL_LOG_FILE}";
|
|---|
| 175 |
|
|---|
| 176 | #
|
|---|
| 177 | # Generate missing info for this testbox if backing up.
|
|---|
| 178 | #
|
|---|
| 179 | MY_INFO_FILE="${MY_BACKUP_DIR}/testbox-info.txt"
|
|---|
| 180 | if test '!' -f "${MY_INFO_FILE}" \
|
|---|
| 181 | -o "${MY_ACTION}" = "backup" \
|
|---|
| 182 | -o "${MY_ACTION}" = "backup-again" \
|
|---|
| 183 | -o "${MY_ACTION}" = "refresh-info" ;
|
|---|
| 184 | then
|
|---|
| 185 | echo "IP: ${MY_IP}" > ${MY_INFO_FILE};
|
|---|
| 186 | echo "HEX-IP: ${MY_IP_HEX}" >> ${MY_INFO_FILE};
|
|---|
| 187 | echo "Hostname: ${MY_HOSTNAME}" >> ${MY_INFO_FILE};
|
|---|
| 188 | echo "" >> ${MY_INFO_FILE};
|
|---|
| 189 | echo "**** cat /proc/cpuinfo ****" >> ${MY_INFO_FILE};
|
|---|
| 190 | echo "**** cat /proc/cpuinfo ****" >> ${MY_INFO_FILE};
|
|---|
| 191 | echo "**** cat /proc/cpuinfo ****" >> ${MY_INFO_FILE};
|
|---|
| 192 | cat /proc/cpuinfo >> ${MY_INFO_FILE};
|
|---|
| 193 | echo "" >> ${MY_INFO_FILE};
|
|---|
| 194 | echo "**** lspci -vvv ****" >> ${MY_INFO_FILE};
|
|---|
| 195 | echo "**** lspci -vvv ****" >> ${MY_INFO_FILE};
|
|---|
| 196 | echo "**** lspci -vvv ****" >> ${MY_INFO_FILE};
|
|---|
| 197 | lspci -vvv >> ${MY_INFO_FILE} 2>&1;
|
|---|
| 198 | echo "" >> ${MY_INFO_FILE};
|
|---|
| 199 | echo "**** biosdecode ****" >> ${MY_INFO_FILE};
|
|---|
| 200 | echo "**** biosdecode ****" >> ${MY_INFO_FILE};
|
|---|
| 201 | echo "**** biosdecode ****" >> ${MY_INFO_FILE};
|
|---|
| 202 | biosdecode >> ${MY_INFO_FILE} 2>&1;
|
|---|
| 203 | echo "" >> ${MY_INFO_FILE};
|
|---|
| 204 | echo "**** dmidecode ****" >> ${MY_INFO_FILE};
|
|---|
| 205 | echo "**** dmidecode ****" >> ${MY_INFO_FILE};
|
|---|
| 206 | echo "**** dmidecode ****" >> ${MY_INFO_FILE};
|
|---|
| 207 | dmidecode >> ${MY_INFO_FILE} 2>&1;
|
|---|
| 208 | echo "" >> ${MY_INFO_FILE};
|
|---|
| 209 | echo "**** fdisk -l ****" >> ${MY_INFO_FILE};
|
|---|
| 210 | echo "**** fdisk -l ****" >> ${MY_INFO_FILE};
|
|---|
| 211 | echo "**** fdisk -l ****" >> ${MY_INFO_FILE};
|
|---|
| 212 | fdisk -l >> ${MY_INFO_FILE} 2>&1;
|
|---|
| 213 |
|
|---|
| 214 | # Get the raw ACPI tables and whatnot since we can. Use zip as tar will
|
|---|
| 215 | # zero pad virtual files due to wrong misleading size returned by stat (4K).
|
|---|
| 216 | zip -r9 "${MY_BACKUP_DIR}/testbox-info.zip" \
|
|---|
| 217 | /sys/firmware/ \
|
|---|
| 218 | /proc/cpuinfo
|
|---|
| 219 | fi
|
|---|
| 220 |
|
|---|
| 221 | if test '!' -f "${MY_BACKUP_DIR}/${MY_HOSTNAME}" -a "${MY_HOSTNAME}" != "unknown"; then
|
|---|
| 222 | echo "${MY_HOSTNAME}" > "${MY_BACKUP_DIR}/${MY_HOSTNAME}"
|
|---|
| 223 | fi
|
|---|
| 224 |
|
|---|
| 225 | if test '!' -f "${MY_BACKUP_DIR}/${MY_IP_HEX}"; then
|
|---|
| 226 | echo "${MY_IP}" > "${MY_BACKUP_DIR}/${MY_IP_HEX}"
|
|---|
| 227 | fi
|
|---|
| 228 |
|
|---|
| 229 | #
|
|---|
| 230 | # Assemble a list of block devices using /sys/block/* and some filtering.
|
|---|
| 231 | #
|
|---|
| 232 | if test -f "${MY_BACKUP_DIR}/disk-devices.lst"; then
|
|---|
| 233 | MY_BLOCK_DEVS=`cat ${MY_BACKUP_DIR}/disk-devices.lst \
|
|---|
| 234 | | sed -e 's/[[:space:]][::space::]]*/ /g' -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//' `;
|
|---|
| 235 | if test -z "${MY_BLOCK_DEVS}"; then
|
|---|
| 236 | ErrorMsgExit 17 "No block devices found via sys/block."
|
|---|
| 237 | fi
|
|---|
| 238 | InfoMsg "disk-device.lst: MY_BLOCK_DEVS=${MY_BLOCK_DEVS}";
|
|---|
| 239 | else
|
|---|
| 240 | MY_BLOCK_DEVS="";
|
|---|
| 241 | for MY_DEV in `ls /sys/block`; do
|
|---|
| 242 | case "${MY_DEV}" in
|
|---|
| 243 | [sh]d*)
|
|---|
| 244 | MY_BLOCK_DEVS="${MY_BLOCK_DEVS} ${MY_DEV}"
|
|---|
| 245 | ;;
|
|---|
| 246 | *) InfoMsg "Ignoring /sys/block/${MY_DEV}";
|
|---|
| 247 | ;;
|
|---|
| 248 | esac
|
|---|
| 249 | done
|
|---|
| 250 | if test -z "${MY_BLOCK_DEVS}"; then
|
|---|
| 251 | ErrorMsgExit 17 "No block devices found via /sys/block."
|
|---|
| 252 | fi
|
|---|
| 253 | InfoMsg "/sys/block: MY_BLOCK_DEVS=${MY_BLOCK_DEVS}";
|
|---|
| 254 | fi
|
|---|
| 255 |
|
|---|
| 256 | #
|
|---|
| 257 | # Take action
|
|---|
| 258 | #
|
|---|
| 259 | case "${MY_ACTION}" in
|
|---|
| 260 | #
|
|---|
| 261 | # Create a backup. The 'backup' action refuses to overwrite an
|
|---|
| 262 | # existing backup, but is otherwise identical to 'backup-again'.
|
|---|
| 263 | #
|
|---|
| 264 | backup|backup-again)
|
|---|
| 265 | for MY_DEV in ${MY_BLOCK_DEVS}; do
|
|---|
| 266 | MY_DST="${MY_BACKUP_DIR}/${MY_DEV}.gz"
|
|---|
| 267 | if test -f "${MY_DST}"; then
|
|---|
| 268 | if test "${MY_ACTION}" != 'backup-again'; then
|
|---|
| 269 | ErrorMsgExit 18 "${MY_DST} already exists"
|
|---|
| 270 | fi
|
|---|
| 271 | InfoMsg "${MY_DST} already exists"
|
|---|
| 272 | fi
|
|---|
| 273 | done
|
|---|
| 274 |
|
|---|
| 275 | # Do the backing up.
|
|---|
| 276 | for MY_DEV in ${MY_BLOCK_DEVS}; do
|
|---|
| 277 | MY_SRC="/dev/${MY_DEV}"
|
|---|
| 278 | MY_DST="${MY_BACKUP_DIR}/${MY_DEV}.gz"
|
|---|
| 279 | if test -b "${MY_SRC}"; then
|
|---|
| 280 | InfoMsg "Backing up ${MY_SRC} to ${MY_DST}...";
|
|---|
| 281 | dd if="${MY_SRC}" bs=2M count=1024 | gzip -c > "${MY_DST}";
|
|---|
| 282 | MY_RCS=("${PIPESTATUS[@]}");
|
|---|
| 283 | if test "${MY_RCS[0]}" -eq 0 -a "${MY_RCS[1]}" -eq 0; then
|
|---|
| 284 | InfoMsg "Successfully backed up ${MY_SRC} to ${MY_DST}";
|
|---|
| 285 | else
|
|---|
| 286 | rm -f "${MY_DST}";
|
|---|
| 287 | ErrorMsgExit 19 "There was a problem backing up ${MY_SRC} to ${MY_DST}: dd => ${MY_RCS[0]}; gzip => ${MY_RCS[1]}";
|
|---|
| 288 | fi
|
|---|
| 289 | else
|
|---|
| 290 | InfoMsg "Skipping ${MY_SRC} as it either doesn't exist or isn't a block device";
|
|---|
| 291 | fi
|
|---|
| 292 | done
|
|---|
| 293 | ;;
|
|---|
| 294 |
|
|---|
| 295 | #
|
|---|
| 296 | # Restore existing.
|
|---|
| 297 | #
|
|---|
| 298 | restore)
|
|---|
| 299 | for MY_DEV in ${MY_BLOCK_DEVS}; do
|
|---|
| 300 | MY_SRC="${MY_BACKUP_DIR}/${MY_DEV}.gz"
|
|---|
| 301 | MY_DST="/dev/${MY_DEV}"
|
|---|
| 302 | if test -b "${MY_DST}"; then
|
|---|
| 303 | if test -f "${MY_SRC}"; then
|
|---|
| 304 | InfoMsg "Restoring ${MY_SRC} onto ${MY_DST}...";
|
|---|
| 305 | gunzip -c "${MY_SRC}" | dd of="${MY_DST}" bs=64K;
|
|---|
| 306 | MY_RCS=("${PIPESTATUS[@]}");
|
|---|
| 307 | if test ${MY_RCS[0]} -eq 0 -a ${MY_RCS[1]} -eq 0; then
|
|---|
| 308 | InfoMsg "Successfully restored ${MY_SRC} onto ${MY_DST}";
|
|---|
| 309 | else
|
|---|
| 310 | ErrorMsgExit 20 "There was a problem restoring ${MY_SRC} onto ${MY_DST}: dd => ${MY_RCS[1]}; gunzip => ${MY_RCS[0]}";
|
|---|
| 311 | fi
|
|---|
| 312 | else
|
|---|
| 313 | InfoMsg "Skipping ${MY_DST} because ${MY_SRC} does not exist.";
|
|---|
| 314 | fi
|
|---|
| 315 | else
|
|---|
| 316 | InfoMsg "Skipping ${MY_DST} as it either doesn't exist or isn't a block device.";
|
|---|
| 317 | fi
|
|---|
| 318 | done
|
|---|
| 319 | ;;
|
|---|
| 320 |
|
|---|
| 321 | #
|
|---|
| 322 | # Nothing else to do for refresh-info.
|
|---|
| 323 | #
|
|---|
| 324 | refresh-info)
|
|---|
| 325 | ;;
|
|---|
| 326 |
|
|---|
| 327 | #
|
|---|
| 328 | # For the rescue action, we just quit without removing the PXE config or
|
|---|
| 329 | # rebooting the box. The admin will do that once the system has been rescued.
|
|---|
| 330 | #
|
|---|
| 331 | rescue)
|
|---|
| 332 | InfoMsg "rescue: exiting. Admin must remove PXE config and reboot manually when done."
|
|---|
| 333 | exit 0;
|
|---|
| 334 | ;;
|
|---|
| 335 |
|
|---|
| 336 | *) ErrorMsgExit 98 "Huh? MY_ACTION='${MY_ACTION}'"
|
|---|
| 337 | ;;
|
|---|
| 338 | esac
|
|---|
| 339 |
|
|---|
| 340 | #
|
|---|
| 341 | # If we get here, remove the PXE config and reboot immediately.
|
|---|
| 342 | #
|
|---|
| 343 | InfoMsg "'${MY_ACTION}' - done";
|
|---|
| 344 | if test -n "${MY_REBOOT_WHEN_DONE}"; then
|
|---|
| 345 | sync
|
|---|
| 346 | if rm -f "${MY_PXELINUX_CFG_FILE}"; then
|
|---|
| 347 | InfoMsg "removed ${MY_PXELINUX_CFG_FILE}";
|
|---|
| 348 | else
|
|---|
| 349 | ErrorMsgExit 99 "failed to remove ${MY_PXELINUX_CFG_FILE}";
|
|---|
| 350 | fi
|
|---|
| 351 | sync
|
|---|
| 352 | InfoMsg "rebooting";
|
|---|
| 353 | reboot
|
|---|
| 354 | fi
|
|---|
| 355 | exit 0
|
|---|
| 356 |
|
|---|
