| 1 | #!/bin/ksh |
|---|
| 2 | # IBM_PROLOG_BEGIN_TAG |
|---|
| 3 | # This is an automatically generated prolog. |
|---|
| 4 | # |
|---|
| 5 | # |
|---|
| 6 | # |
|---|
| 7 | # Licensed Materials - Property of IBM |
|---|
| 8 | # |
|---|
| 9 | # (C) COPYRIGHT International Business Machines Corp. 2000,2006 |
|---|
| 10 | # All Rights Reserved |
|---|
| 11 | # |
|---|
| 12 | # US Government Users Restricted Rights - Use, duplication or |
|---|
| 13 | # disclosure restricted by GSA ADP Schedule Contract with IBM Corp. |
|---|
| 14 | # |
|---|
| 15 | # IBM_PROLOG_END_TAG |
|---|
| 16 | # @(#)19 1.32 src/avs/fs/mmfs/ts/admin/runmmfs.sh, mmfs, avs_rgpfs24, rgpfs240610b 10/26/05 19:20:55 |
|---|
| 17 | ####################################################################### |
|---|
| 18 | # |
|---|
| 19 | # runmmfs is used to start up the GPFS daemon. |
|---|
| 20 | # |
|---|
| 21 | # Usage: |
|---|
| 22 | # runmmfs [options] |
|---|
| 23 | # |
|---|
| 24 | # where |
|---|
| 25 | # options if specified, passed through to the mmfsenv script. |
|---|
| 26 | # |
|---|
| 27 | ####################################################################### |
|---|
| 28 | |
|---|
| 29 | |
|---|
| 30 | #------------------------------------------------------------------------ |
|---|
| 31 | # Collect options, if any, to be passed on to mmfsenv. |
|---|
| 32 | #------------------------------------------------------------------------ |
|---|
| 33 | mmfsenvOpt=$@ |
|---|
| 34 | |
|---|
| 35 | #------------------------------------------------------------------------ |
|---|
| 36 | # Decide where to look for the commands. The normal install directory |
|---|
| 37 | # is /usr/lpp/mmfs/bin. This can be changed by setting the MMFSDIR |
|---|
| 38 | # environment variable to something other than /usr/lpp/mmfs. |
|---|
| 39 | # If MMFSDIR is not used, before assuming the default install directory, |
|---|
| 40 | # we will check if the mmfs binaries are in the same place where this |
|---|
| 41 | # script resides. This is a development aid for running out of a build tree. |
|---|
| 42 | #------------------------------------------------------------------------ |
|---|
| 43 | if [[ -n $MMFSDIR ]] |
|---|
| 44 | then |
|---|
| 45 | mmcmdDir="${MMFSDIR}/bin" |
|---|
| 46 | else |
|---|
| 47 | if [[ -z ${0%%/*} ]] |
|---|
| 48 | then |
|---|
| 49 | fullname="$0" |
|---|
| 50 | else |
|---|
| 51 | fullname="${PWD%/}/$0" |
|---|
| 52 | fi |
|---|
| 53 | progDir=${fullname%/*} |
|---|
| 54 | |
|---|
| 55 | if [[ -f ${progDir}/mmfs ]] |
|---|
| 56 | then |
|---|
| 57 | mmcmdDir="$progDir" |
|---|
| 58 | else |
|---|
| 59 | mmcmdDir="/usr/lpp/mmfs/bin" |
|---|
| 60 | fi |
|---|
| 61 | fi |
|---|
| 62 | |
|---|
| 63 | # Determine the values of mmcmdSubdir and mmcmdSuffix. |
|---|
| 64 | set -f ; set -- $(/bin/uname -a) ; set +f |
|---|
| 65 | osName=$1 |
|---|
| 66 | osVersion=$4 |
|---|
| 67 | set -- |
|---|
| 68 | if [[ -f ${mmcmdDir}/mmfs || $osName != AIX || $osVersion < 5 ]] |
|---|
| 69 | then |
|---|
| 70 | mmcmdSubdir="" |
|---|
| 71 | mmcmdSuffix="" |
|---|
| 72 | |
|---|
| 73 | else |
|---|
| 74 | # Must be running on top of AIX 5. |
|---|
| 75 | # Determine whether the current kernel is 64-bit. |
|---|
| 76 | kernelMode=$(${mmcmdDir}/mmkerninfo) |
|---|
| 77 | if [[ $kernelMode = 64 ]] |
|---|
| 78 | then |
|---|
| 79 | mmcmdSubdir="aix64" |
|---|
| 80 | mmcmdSuffix="64" |
|---|
| 81 | else |
|---|
| 82 | mmcmdSubdir="aix32" |
|---|
| 83 | mmcmdSuffix="" |
|---|
| 84 | fi |
|---|
| 85 | fi |
|---|
| 86 | |
|---|
| 87 | # Include global declarations and service routines. |
|---|
| 88 | . ${mmcmdDir}/mmglobfuncs |
|---|
| 89 | . ${mmcmdDir}/mmsdrfsdef |
|---|
| 90 | . ${mmcmdDir}/mmfsfuncs |
|---|
| 91 | |
|---|
| 92 | commandPath=set |
|---|
| 93 | |
|---|
| 94 | sourceFile="runmmfs.sh" |
|---|
| 95 | [[ -n $DEBUG || -n $DEBUGrunmmfs ]] && set -x |
|---|
| 96 | $mmTRACE_ENTER "$*" |
|---|
| 97 | |
|---|
| 98 | |
|---|
| 99 | #------------------------------------------------------------------------ |
|---|
| 100 | # Function to catch stop signals issued while we are in this script. |
|---|
| 101 | # The daemon, once started, is responsible for signal handling. |
|---|
| 102 | #------------------------------------------------------------------------ |
|---|
| 103 | function sigcatch |
|---|
| 104 | { |
|---|
| 105 | print "$(date) ${mmcmd}: Received a stop request; exiting" |
|---|
| 106 | $mmTRACE_EXIT "rc=1 stop request" |
|---|
| 107 | exit 1 |
|---|
| 108 | } |
|---|
| 109 | |
|---|
| 110 | #------------------------------------------------------------------------ |
|---|
| 111 | # Set the default sleep backoff to 7 seconds. This ensures that the |
|---|
| 112 | # first three attempts to start will not occur within 20 seconds. |
|---|
| 113 | #------------------------------------------------------------------------ |
|---|
| 114 | (( backoff = 7 )) |
|---|
| 115 | |
|---|
| 116 | #------------------------------------------------------------------------ |
|---|
| 117 | # If running under the control of the startup daemon (daemonize command), |
|---|
| 118 | # set up the console output (standout and standerr) files. |
|---|
| 119 | #------------------------------------------------------------------------ |
|---|
| 120 | foreground=true |
|---|
| 121 | waittime=20 |
|---|
| 122 | |
|---|
| 123 | # DAEMONIZE is set by daemonize command when it execs us |
|---|
| 124 | if [[ -n $DAEMONIZE ]]; then |
|---|
| 125 | |
|---|
| 126 | #-------------------------------------------------------------- |
|---|
| 127 | # If we were started by daemonize then we are responsible for |
|---|
| 128 | # restarting mmfsd. The rerunmmfs file enables that function |
|---|
| 129 | #-------------------------------------------------------------- |
|---|
| 130 | $touch $rerunmmfsFile |
|---|
| 131 | |
|---|
| 132 | foreground=false |
|---|
| 133 | |
|---|
| 134 | #------------------------------------------------------ |
|---|
| 135 | # Find the setting of the logDir mmfs.cfg parameter. |
|---|
| 136 | # It can be used to override the default log location. |
|---|
| 137 | #------------------------------------------------------ |
|---|
| 138 | if [[ -f $mmfscfg ]] |
|---|
| 139 | then |
|---|
| 140 | logDir=$($awk '$1 == "logDir" {value = $2} END {print value}' $mmfscfg) |
|---|
| 141 | [[ -n $logDir ]] && rasDir="${logDir}/" |
|---|
| 142 | fi |
|---|
| 143 | |
|---|
| 144 | standin="/dev/console" |
|---|
| 145 | standout="${rasDir}mmfs.log" |
|---|
| 146 | standerr="${rasDir}mmfs.log" |
|---|
| 147 | signorm=31 |
|---|
| 148 | sigforce=15 |
|---|
| 149 | fi |
|---|
| 150 | |
|---|
| 151 | #--------------------------------------------------------------------- |
|---|
| 152 | # Set the stop request signal handlers. |
|---|
| 153 | # Use the "waittime" value to set the respawn backoff. |
|---|
| 154 | #--------------------------------------------------------------------- |
|---|
| 155 | trap sigcatch $signorm $sigforce |
|---|
| 156 | (( backoff = (waittime / 3) + 1 )) |
|---|
| 157 | |
|---|
| 158 | #------------------------------------------------------------------ |
|---|
| 159 | # If we were started by daemonize, then this script may be responsible |
|---|
| 160 | # for respawning mmfsd after a death ($rerunmmfsFile is present). |
|---|
| 161 | #------------------------------------------------------------------ |
|---|
| 162 | while true |
|---|
| 163 | do |
|---|
| 164 | # Move the respawnlog file, if any. |
|---|
| 165 | $mv "$respawnlog" "$respawnlog2" 2>/dev/null |
|---|
| 166 | |
|---|
| 167 | if [[ -n $DAEMONIZE ]]; then |
|---|
| 168 | exec 0<$standin |
|---|
| 169 | exec 1>$standout |
|---|
| 170 | exec 2>$standerr |
|---|
| 171 | |
|---|
| 172 | cd ${mmcmdDir} |
|---|
| 173 | fi |
|---|
| 174 | |
|---|
| 175 | print -- "$(date) $mmcmd starting" |
|---|
| 176 | |
|---|
| 177 | if [[ $foreground = false ]] |
|---|
| 178 | then |
|---|
| 179 | #--------------------------------------------------------------------- |
|---|
| 180 | # Clean out console log files not accessed in the last 11 days. |
|---|
| 181 | #--------------------------------------------------------------------- |
|---|
| 182 | savepwd="$PWD" |
|---|
| 183 | if [[ -f "$standout" ]]; then |
|---|
| 184 | dir="${standout%/*}" |
|---|
| 185 | base="${standout##*/}" |
|---|
| 186 | print "Removing old $standout.* files:" |
|---|
| 187 | if cd "$dir"; then |
|---|
| 188 | $find . -type f -name "${base}"'.*' -atime +10 -mtime +10 -exec $rm {} \; -print |
|---|
| 189 | fi |
|---|
| 190 | fi |
|---|
| 191 | |
|---|
| 192 | if [[ "$standout" != "$standerr" && -f "$standerr" ]]; then |
|---|
| 193 | dir="${standerr%/*}" |
|---|
| 194 | base="${standerr##*/}" |
|---|
| 195 | print "Removing old $standerr.* files:" |
|---|
| 196 | if cd "$dir"; then |
|---|
| 197 | $find . -type f -name "${base}"'.*' -atime +10 -mtime +10 -exec $rm {} \; -print |
|---|
| 198 | fi |
|---|
| 199 | fi |
|---|
| 200 | cd "$savepwd" |
|---|
| 201 | |
|---|
| 202 | #--------------------------------------------------------------------- |
|---|
| 203 | # Standard out and standard error are opened in non-append mode. If |
|---|
| 204 | # standard out and standard error are directed at files, rename the |
|---|
| 205 | # files, so that on the next restart these files are not overwritten. |
|---|
| 206 | # Also, re-open the files in append mode. |
|---|
| 207 | #--------------------------------------------------------------------- |
|---|
| 208 | suffix="$($date +%Y.%m.%d.%H.%M.%S).$($hostname -s)" |
|---|
| 209 | changeprevious=1 |
|---|
| 210 | if [[ -f "$respawnlog2" ]]; then |
|---|
| 211 | # When respawning, use previous log file to prevent the directory |
|---|
| 212 | # from filling up with lots of nearly identical files. |
|---|
| 213 | if [[ -L "${standout}.latest" ]]; then |
|---|
| 214 | latest=$($ls -l "${standout}.latest") |
|---|
| 215 | latest=${latest##*-\> } |
|---|
| 216 | prefix=$(basename ${standout}) |
|---|
| 217 | latest=${latest#${prefix}.} |
|---|
| 218 | [[ -n "$latest" ]] && suffix="$latest" |
|---|
| 219 | changeprevious=0 |
|---|
| 220 | fi |
|---|
| 221 | fi |
|---|
| 222 | |
|---|
| 223 | if [[ -f "$standout" ]]; then |
|---|
| 224 | $cat "$standout" >> "${standout}.${suffix}" |
|---|
| 225 | $chmod 644 "${standout}.${suffix}" |
|---|
| 226 | exec 1>>"${standout}.${suffix}" |
|---|
| 227 | [[ "$standout" = "$standerr" ]] && exec 2>&1 |
|---|
| 228 | $rm "$standout" |
|---|
| 229 | |
|---|
| 230 | # Create a symbolic link with suffix "latest" to the current log |
|---|
| 231 | # and suffix "previous" to the previous log. |
|---|
| 232 | if [[ $changeprevious = 1 ]]; then |
|---|
| 233 | if [[ -L "${standout}.previous" ]]; then |
|---|
| 234 | $rm -f "${standout}.previous" |
|---|
| 235 | else |
|---|
| 236 | $mv -f "${standout}.previous" "${standout}.previous.save" |
|---|
| 237 | fi |
|---|
| 238 | [[ ! -r "${standout}.previous" ]] && [[ -L "${standout}.latest" ]] && \ |
|---|
| 239 | $mv -f "${standout}.latest" "${standout}.previous" |
|---|
| 240 | fi |
|---|
| 241 | |
|---|
| 242 | [[ ! -r "${standout}.latest" ]] && \ |
|---|
| 243 | $ln -sf $(basename "${standout}.${suffix}") "${standout}.latest" |
|---|
| 244 | [[ ! -f "${mmsdrfsDir}mmfslog" ]] && \ |
|---|
| 245 | $ln -sf "${standout}.latest" "${mmsdrfsDir}mmfslog" |
|---|
| 246 | fi |
|---|
| 247 | |
|---|
| 248 | if [[ "$standout" != "$standerr" && -f "$standerr" ]]; then |
|---|
| 249 | $cat "$standerr" >> "${standerr}.${suffix}" |
|---|
| 250 | $chmod 644 "${standerr}.${suffix}" |
|---|
| 251 | exec 2>>"${standerr}.${suffix}" |
|---|
| 252 | $rm "$standerr" |
|---|
| 253 | fi |
|---|
| 254 | fi # end of if [[ $foreground = false ]] |
|---|
| 255 | |
|---|
| 256 | #------------------------------------------------------------------------- |
|---|
| 257 | # To prevent start/crash loops, this script creates/modifies a file |
|---|
| 258 | # that keeps a record of how often the MMFS daemon has been respawned |
|---|
| 259 | # and the daemon has not come up completely due to external influences. |
|---|
| 260 | # This file is removed by the daemon. |
|---|
| 261 | # |
|---|
| 262 | # If the file does not exist, create it with values "0 ${backoff}"; where |
|---|
| 263 | # the backoff is the initial sleep (default 7 seconds), designed so that |
|---|
| 264 | # the first three respawns will not occur within 20 seconds. |
|---|
| 265 | # |
|---|
| 266 | # If the file already exists, and the sleep we're about to take is less |
|---|
| 267 | # than 56 seconds, double the time for the next sleep. If the upcoming |
|---|
| 268 | # sleep is greater than or equal to 56 seconds but less than 336 seconds, |
|---|
| 269 | # add another 56 seconds to the next sleep. If the upcoming sleep is |
|---|
| 270 | # 336 seconds or more, the next sleep will be the same. The count of the |
|---|
| 271 | # number of times we've slept waiting to respawn is always incremented |
|---|
| 272 | # by one. |
|---|
| 273 | # |
|---|
| 274 | # Note: file respawnlog isn't restored until just before an "exit" or |
|---|
| 275 | # "exec"; until then file respawnlog2 is used. This helps to prevent |
|---|
| 276 | # the persistence of the respawn log between stop and start commands. |
|---|
| 277 | #------------------------------------------------------------------------- |
|---|
| 278 | if [[ -s "$respawnlog2" ]]; then |
|---|
| 279 | read count waitPeriod < "$respawnlog2" |
|---|
| 280 | if [[ $count != +([0-9]) || $waitPeriod != +([0-9]) ]]; then |
|---|
| 281 | (( count = 1 )) |
|---|
| 282 | (( waitPeriod = backoff )) |
|---|
| 283 | else |
|---|
| 284 | (( count += 1 )) |
|---|
| 285 | fi |
|---|
| 286 | if (( waitPeriod < 56 )); then |
|---|
| 287 | (( newWaitPeriod = waitPeriod * 2 )) |
|---|
| 288 | elif (( waitPeriod < 336 )); then |
|---|
| 289 | (( newWaitPeriod = waitPeriod + 56 )) |
|---|
| 290 | else |
|---|
| 291 | (( newWaitPeriod = waitPeriod )) |
|---|
| 292 | fi |
|---|
| 293 | print ${count} ${newWaitPeriod} > "$respawnlog2" |
|---|
| 294 | print "$(date) ${mmcmd}: respawn ${count} waiting ${waitPeriod} seconds before restarting mmfsd" |
|---|
| 295 | |
|---|
| 296 | #--------------------------------------------------------------------- |
|---|
| 297 | # Run the sleep in the background and then wait for it. This lets |
|---|
| 298 | # the parent script get signals exactly when they're delivered. |
|---|
| 299 | # A side effect is that the sleep itself is protected from the signals, |
|---|
| 300 | # and will continue to run, but a "sleep" process consumes almost no |
|---|
| 301 | # resources, and will eventually wakeup and exit. |
|---|
| 302 | #--------------------------------------------------------------------- |
|---|
| 303 | $sleep ${waitPeriod} & |
|---|
| 304 | wait |
|---|
| 305 | |
|---|
| 306 | else |
|---|
| 307 | # file does not exist yet, create it. |
|---|
| 308 | (( newWaitPeriod = backoff )) |
|---|
| 309 | print 0 ${newWaitPeriod} > "$respawnlog2" |
|---|
| 310 | |
|---|
| 311 | fi # end of if [[ -s "$respawnlog2" ]] |
|---|
| 312 | |
|---|
| 313 | #------------------------------------------------------ |
|---|
| 314 | # Perform standard initialization. |
|---|
| 315 | # If something goes wrong, do not bring the daemon up. |
|---|
| 316 | #------------------------------------------------------ |
|---|
| 317 | $mmfsenv $mmfsenvOpt |
|---|
| 318 | mmfsenvrc=$? |
|---|
| 319 | sourceFile="runmmfs.sh" # restore value after in-line call to mmfsenv |
|---|
| 320 | |
|---|
| 321 | #--------------------------------------------------------------------------- |
|---|
| 322 | # Now check the return status from "mmfsenv"; if it's bad, restore the |
|---|
| 323 | # respawnlog file just before the exit and delete any old respawnlog2 files. |
|---|
| 324 | # If the error is in setting up the kernel extension, give up all together. |
|---|
| 325 | #--------------------------------------------------------------------------- |
|---|
| 326 | if [[ $mmfsenvrc -eq $MM_KExtFailure ]]; then |
|---|
| 327 | print "$(date) ${mmcmd}: error in loading or unloading the mmfs kernel extension" |
|---|
| 328 | print "$(date) ${mmcmd}: stopping GPFS" |
|---|
| 329 | |
|---|
| 330 | $rm -f "${respawnlog}"* $rerunmmfsFile |
|---|
| 331 | $mmTRACE_EXIT "rc=$mmfsenvrc mmfsenv error" |
|---|
| 332 | exit $mmfsenvrc |
|---|
| 333 | |
|---|
| 334 | elif [[ $mmfsenvrc != 0 ]]; then |
|---|
| 335 | print "$(date) ${mmcmd}: error preparing environment for mmfsd" |
|---|
| 336 | print "$(date) ${mmcmd}: will try again in ${newWaitPeriod} seconds" |
|---|
| 337 | $mv -f "$respawnlog2" "$respawnlog" |
|---|
| 338 | $rm -f "${respawnlog}".* |
|---|
| 339 | $mmTRACE_EXIT "rc=$mmfsenvrc mmfsenv error" |
|---|
| 340 | |
|---|
| 341 | # If the rerunmmfsFile is present, go back to the top |
|---|
| 342 | # of the loop for another iteration. Otherwise we exit. |
|---|
| 343 | [[ -n $DAEMONIZE && -f "$rerunmmfsFile" ]] && continue |
|---|
| 344 | |
|---|
| 345 | $rm -f $rerunmmfsFile |
|---|
| 346 | exit $mmfsenvrc |
|---|
| 347 | fi |
|---|
| 348 | |
|---|
| 349 | #----------------------------------------------------- |
|---|
| 350 | # Start the MMFS daemon. |
|---|
| 351 | # Restore the respawnlog file just before the "exec". |
|---|
| 352 | #----------------------------------------------------- |
|---|
| 353 | ulimit -d unlimited |
|---|
| 354 | ulimit -m unlimited |
|---|
| 355 | ulimit -f unlimited |
|---|
| 356 | ulimit -t unlimited |
|---|
| 357 | ulimit -n 1000000 # virtually unlimited while still an acceptable value |
|---|
| 358 | ulimit -c 1024 # keep core files small |
|---|
| 359 | |
|---|
| 360 | # Start tracing if the environment variable is set. |
|---|
| 361 | [[ -n "$MMTRACE" ]] && $mmtrace |
|---|
| 362 | |
|---|
| 363 | if [[ -n "$MMFSCONFIG" ]]; then |
|---|
| 364 | MMARGS="-d $MMFSCONFIG" |
|---|
| 365 | else |
|---|
| 366 | MMARGS="" |
|---|
| 367 | fi |
|---|
| 368 | $mmTRACE "---> Starting GPFS daemon from runmmfs." |
|---|
| 369 | $mv -f "$respawnlog2" "$respawnlog" |
|---|
| 370 | $rm -f "${respawnlog}".* |
|---|
| 371 | |
|---|
| 372 | # If we are spawned by daemonize and $rerunmmfsFile exists, |
|---|
| 373 | # then we'll be responsible for restarting GPFS. |
|---|
| 374 | if [[ -n $DAEMONIZE ]]; then |
|---|
| 375 | $mmfsd $MMARGS |
|---|
| 376 | rc=$? |
|---|
| 377 | |
|---|
| 378 | # If rc is 101 (i.e., this daemon discovered a previously-running |
|---|
| 379 | # incarnation of mmfsd), just get out and do not erase $rerunmmfsFile, |
|---|
| 380 | # since it belongs to the other instance. |
|---|
| 381 | [[ $rc = 101 ]] && exit $rc |
|---|
| 382 | |
|---|
| 383 | # Continue to the beginning of the loop for restart of mmfsd. |
|---|
| 384 | [[ -f "$rerunmmfsFile" ]] && continue |
|---|
| 385 | exit |
|---|
| 386 | fi |
|---|
| 387 | |
|---|
| 388 | exec $mmfsd $MMARGS |
|---|
| 389 | |
|---|
| 390 | done |
|---|
| 391 | |
|---|
