| 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. 1999,2006 |
|---|
| 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 | # @(#)78 1.40 src/avs/fs/mmfs/ts/admin/mmchdisk.sh, mmfs, avs_rgpfs24, rgpfs240610b 11/11/05 09:29:00 |
|---|
| 17 | ######################################################################### |
|---|
| 18 | # |
|---|
| 19 | # Usage: |
|---|
| 20 | # mmchdisk Device {resume | start} -a |
|---|
| 21 | # [-N {all | mount | Node[,Node...] | NodeFile | NodeClass}] |
|---|
| 22 | # or |
|---|
| 23 | # mmchdisk Device {suspend | resume | stop | start | change} |
|---|
| 24 | # -d "DiskDesc[;DiskDesc...]" |
|---|
| 25 | # [-N {all | mount | Node[,Node...] | NodeFile | NodeClass}] |
|---|
| 26 | # |
|---|
| 27 | ######################################################################### |
|---|
| 28 | |
|---|
| 29 | # Include global declarations and service routines. |
|---|
| 30 | . /usr/lpp/mmfs/bin/mmglobfuncs |
|---|
| 31 | . /usr/lpp/mmfs/bin/mmsdrfsdef |
|---|
| 32 | . /usr/lpp/mmfs/bin/mmfsfuncs |
|---|
| 33 | |
|---|
| 34 | sourceFile="mmchdisk.sh" |
|---|
| 35 | [[ -n $DEBUG || -n $DEBUGmmchdisk ]] && set -x |
|---|
| 36 | $mmTRACE_ENTER "$*" |
|---|
| 37 | |
|---|
| 38 | |
|---|
| 39 | # Local variables |
|---|
| 40 | |
|---|
| 41 | usageMsg=287 |
|---|
| 42 | integer rc=0 |
|---|
| 43 | integer nodeCount=0 |
|---|
| 44 | # argList="" # argList without quotes around disk list |
|---|
| 45 | # argListQ="" # argList with quotes around disk list |
|---|
| 46 | |
|---|
| 47 | |
|---|
| 48 | ##################################################################### |
|---|
| 49 | # Process the command arguments. |
|---|
| 50 | ##################################################################### |
|---|
| 51 | [[ $arg1 = '-?' || $arg1 = '-h' || $arg1 = '--help' || $arg1 = '--' ]] && \ |
|---|
| 52 | syntaxError "help" $usageMsg |
|---|
| 53 | |
|---|
| 54 | [[ $argc -lt 2 ]] && \ |
|---|
| 55 | syntaxError "missingArgs" $usageMsg |
|---|
| 56 | |
|---|
| 57 | device=$arg1 # file system name (always the first parameter) |
|---|
| 58 | action=$arg2 # action to be taken for the disks (always the 2nd parameter) |
|---|
| 59 | shift 2 # Drop the device name and disk state from the parameter list. |
|---|
| 60 | |
|---|
| 61 | # Parse the optional parameters. Special care must be taken when creating |
|---|
| 62 | # argList because the disk names in the -d option are separated by the ';' |
|---|
| 63 | # character. For invocations via mmcommon, the argument lists must be |
|---|
| 64 | # enclosed in quotes. For direct (local) invocations of tschdisk, |
|---|
| 65 | # the argument lists must not have surrounding quotes. Detailed checking |
|---|
| 66 | # of supported keywords is left to tschdisk. |
|---|
| 67 | while getopts :ad:N: OPT |
|---|
| 68 | do |
|---|
| 69 | case $OPT in |
|---|
| 70 | a) # Change the state of all disks in the file system. |
|---|
| 71 | [[ -n $aflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 72 | [[ -n $dflag ]] && \ |
|---|
| 73 | syntaxError "invalidCombination" $usageMsg "-a" "-d" |
|---|
| 74 | aflag=yes |
|---|
| 75 | argList="$action -$OPT" |
|---|
| 76 | argListQ="$action -$OPT" |
|---|
| 77 | ;; |
|---|
| 78 | |
|---|
| 79 | d) # Change the state of the specified disks. |
|---|
| 80 | [[ -n $dflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 81 | [[ -n $aflag ]] && \ |
|---|
| 82 | syntaxError "invalidCombination" $usageMsg "-a" "-d" |
|---|
| 83 | diskDescList=$OPTARG |
|---|
| 84 | dflag=yes |
|---|
| 85 | [[ "$diskDescList" = *+([,${BLANKchar}${TABchar}])* ]] && \ |
|---|
| 86 | syntaxError "badSeparator_notSemicolon" $noUsageMsg |
|---|
| 87 | argList="$action -$OPT $OPTARG " |
|---|
| 88 | argListQ="$action -$OPT \"$OPTARG\" " |
|---|
| 89 | ;; |
|---|
| 90 | |
|---|
| 91 | N) # Specify the nodes on which to restripe. |
|---|
| 92 | [[ -n $Nflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 93 | nodeList="$OPTARG" |
|---|
| 94 | Nflag="-N $OPTARG" |
|---|
| 95 | ;; |
|---|
| 96 | |
|---|
| 97 | :) # missing required value after an option |
|---|
| 98 | syntaxError "missingValue" $usageMsg $OPTARG |
|---|
| 99 | ;; |
|---|
| 100 | |
|---|
| 101 | +[adN]) # plus flags are not allowed |
|---|
| 102 | syntaxError "invalidOption" $usageMsg $OPT |
|---|
| 103 | ;; |
|---|
| 104 | |
|---|
| 105 | *) # invalid option specified |
|---|
| 106 | syntaxError "invalidOption" $usageMsg $OPTARG |
|---|
| 107 | ;; |
|---|
| 108 | esac |
|---|
| 109 | done |
|---|
| 110 | |
|---|
| 111 | shift OPTIND-1 |
|---|
| 112 | [[ $# != 0 ]] && syntaxError "extraArg" $usageMsg $1 |
|---|
| 113 | |
|---|
| 114 | |
|---|
| 115 | ######################################################################## |
|---|
| 116 | # Set up trap exception handling and call the gpfsInit function. |
|---|
| 117 | # It will ensure that the local copy of the mmsdrfs and the rest of the |
|---|
| 118 | # GPFS system files are up-to-date. We only lock the sdr if mmchdisk |
|---|
| 119 | # was invoked with the change action. |
|---|
| 120 | ######################################################################## |
|---|
| 121 | if [[ $action = change ]] |
|---|
| 122 | then |
|---|
| 123 | trap pretrap HUP INT QUIT KILL |
|---|
| 124 | gpfsInitOutput=$(gpfsInit $lockId) |
|---|
| 125 | else |
|---|
| 126 | trap pretrap2 HUP INT QUIT KILL |
|---|
| 127 | gpfsInitOutput=$(gpfsInit nolock) |
|---|
| 128 | fi |
|---|
| 129 | setGlobalVar $? $gpfsInitOutput |
|---|
| 130 | |
|---|
| 131 | # Determine the lookup order for resolving host names. |
|---|
| 132 | [[ $osName != AIX ]] && resolveOrder=$(setHostResolveOrder) |
|---|
| 133 | |
|---|
| 134 | |
|---|
| 135 | ########################################################### |
|---|
| 136 | # Make sure the specified file system exists and is local. |
|---|
| 137 | ########################################################### |
|---|
| 138 | findFSoutput=$(findFS "$device" $mmsdrfsFile) |
|---|
| 139 | [[ -z $findFSoutput ]] && cleanupAndExit |
|---|
| 140 | |
|---|
| 141 | # Parse the output from the findFS function. |
|---|
| 142 | set -f ; set -- $findFSoutput ; set +f |
|---|
| 143 | fqDeviceName=$1 |
|---|
| 144 | deviceName=$2 |
|---|
| 145 | fsHomeCluster=$3 |
|---|
| 146 | |
|---|
| 147 | # Exit with a message if the command was invoked for a remote file system. |
|---|
| 148 | if [[ $fsHomeCluster != $HOME_CLUSTER ]] |
|---|
| 149 | then |
|---|
| 150 | # Command is not allowed for remote file systems. |
|---|
| 151 | printErrorMsg 106 $mmcmd $device $fsHomeCluster |
|---|
| 152 | cleanupAndExit |
|---|
| 153 | fi |
|---|
| 154 | |
|---|
| 155 | |
|---|
| 156 | ################################################################### |
|---|
| 157 | # If a keyword or a list of nodes was specified via the -N option, |
|---|
| 158 | # verify the input and add it to the argList and argListQ strings. |
|---|
| 159 | ################################################################### |
|---|
| 160 | if [[ -n $Nflag ]] |
|---|
| 161 | then |
|---|
| 162 | if [[ $nodeList = all || $nodeList = mount ]] |
|---|
| 163 | then |
|---|
| 164 | # Add the keyword ("all" or "mount") to the argList and argListQ strings. |
|---|
| 165 | argList="$argList -N $nodeList " |
|---|
| 166 | argListQ="$argListQ -N \"$nodeList\" " |
|---|
| 167 | else |
|---|
| 168 | # Verify the node list and convert it to a file of daemon node names. |
|---|
| 169 | createVerifiedNodefile "$nodeList" $DAEMON_NODENAME_Field no $nodefile |
|---|
| 170 | [[ $? -ne 0 ]] && cleanupAndExit |
|---|
| 171 | |
|---|
| 172 | # Convert the output data from a file to a comma-separated list. |
|---|
| 173 | newNodeList=$(print -- $(cat $nodefile) | $sed 's/ /,/g') |
|---|
| 174 | |
|---|
| 175 | # Add the verified node list to the argList and argListQ strings. |
|---|
| 176 | argList="$argList -N $newNodeList " |
|---|
| 177 | argListQ="$argListQ -N \"$newNodeList\" " |
|---|
| 178 | fi # end of if [[ $nodeList = all || $nodeList = mount ]] |
|---|
| 179 | fi # end of if [[ -n $Nflag ]] |
|---|
| 180 | |
|---|
| 181 | |
|---|
| 182 | ####################################################################### |
|---|
| 183 | # If the cluster to which this node belongs is the same as the cluster |
|---|
| 184 | # in which the file system resides, invoke the command directly. |
|---|
| 185 | ####################################################################### |
|---|
| 186 | if [[ $nsId = $fsHomeCluster ]] |
|---|
| 187 | then |
|---|
| 188 | ${mmcmdDir}/${links}/mmchdisk $fqDeviceName $argList 2>$errMsg |
|---|
| 189 | rc=$(remapRC $?) |
|---|
| 190 | |
|---|
| 191 | # If the command completed successfully, or if there is an |
|---|
| 192 | # unacceptable error, display any error messages and get out. |
|---|
| 193 | # If the action was "change", update the mmsdrfs file before exiting. |
|---|
| 194 | if [[ $rc -ne $MM_DaemonDown && $rc -ne $MM_QuorumWait ]] |
|---|
| 195 | then |
|---|
| 196 | [[ -s $errMsg ]] && $cat $errMsg 1>&2 |
|---|
| 197 | if [[ $rc -eq $MM_ConnectionReset ]] |
|---|
| 198 | then |
|---|
| 199 | # An internode connection was reset. |
|---|
| 200 | printErrorMsg 257 $mmcmd |
|---|
| 201 | fi |
|---|
| 202 | |
|---|
| 203 | # Update the mmsdrfs file and unlock the sdr if the action was "change". |
|---|
| 204 | if [[ $action = change ]] |
|---|
| 205 | then |
|---|
| 206 | # Reconcile the sdrfs file with the GPFS daemon's view of the filesystem. |
|---|
| 207 | $cp $mmsdrfsFile $newsdrfs |
|---|
| 208 | reconcileSdrfsWithDaemon $deviceName $newsdrfs |
|---|
| 209 | rc2=$? |
|---|
| 210 | if [[ $rc2 -ne 0 ]] |
|---|
| 211 | then |
|---|
| 212 | # reconcileSdrfsWithDaemon failed. |
|---|
| 213 | printErrorMsg 171 $mmcmd reconcileSdrfsWithDaemon $rc2 |
|---|
| 214 | # Tell the user to run mmcommon recoverfs against the file system. |
|---|
| 215 | printErrorMsg 190 $mmcmd $deviceName $deviceName |
|---|
| 216 | cleanupAndExit $rc |
|---|
| 217 | fi |
|---|
| 218 | |
|---|
| 219 | # Get the generation number from the version line of the new sdrfs file. |
|---|
| 220 | versionLine=$($head -1 $newsdrfs) |
|---|
| 221 | IFS=':' |
|---|
| 222 | set -f ; set -- $versionLine ; set +f |
|---|
| 223 | newGenNumber=$6 |
|---|
| 224 | IFS="$IFS_sv" |
|---|
| 225 | |
|---|
| 226 | # Commit the reconciled version of the sdrfs file to the server |
|---|
| 227 | # so the admin scripts and the daemon are in sync. |
|---|
| 228 | trap "" HUP INT QUIT KILL # Disable interrupts until the commit is done. |
|---|
| 229 | gpfsObjectInfo=$(commitChanges $fsHomeCluster $nsId $gpfsObjectInfo \ |
|---|
| 230 | $newGenNumber $newsdrfs $primaryServer) |
|---|
| 231 | if [[ $? -ne 0 ]] |
|---|
| 232 | then |
|---|
| 233 | # We were unable to replace the file in the sdr. |
|---|
| 234 | printErrorMsg 381 $mmcmd |
|---|
| 235 | # Tell the user to run mmcommon recoverfs against the filesystem. |
|---|
| 236 | printErrorMsg 190 $mmcmd $deviceName $deviceName |
|---|
| 237 | fi |
|---|
| 238 | |
|---|
| 239 | # Unlock the sdr. |
|---|
| 240 | [[ $sdrLocked = yes ]] && \ |
|---|
| 241 | freeLockOnServer $primaryServer $ourNodeNumber >/dev/null |
|---|
| 242 | sdrLocked=no |
|---|
| 243 | trap localPosttrap HUP INT QUIT KILL |
|---|
| 244 | |
|---|
| 245 | # Propagate the new mmsdrfs file. This process is asynchronous. |
|---|
| 246 | propagateSdrfsFile async $nodefile $newsdrfs $newGenNumber |
|---|
| 247 | |
|---|
| 248 | fi # end of if [[ $action = change ]] |
|---|
| 249 | |
|---|
| 250 | cleanupAndExit $rc |
|---|
| 251 | |
|---|
| 252 | fi # end of if [[ $rc -ne $MM_DaemonDown && $rc -ne $MM_QuorumWait ]] |
|---|
| 253 | |
|---|
| 254 | fi # end of if [[ $nsId = $fsHomeCluster ]] |
|---|
| 255 | $rm -f $errMsg |
|---|
| 256 | |
|---|
| 257 | |
|---|
| 258 | ############################################################################ |
|---|
| 259 | # If the local daemon is not available, send the command to an active node. |
|---|
| 260 | ############################################################################ |
|---|
| 261 | |
|---|
| 262 | # Create a file with the reliable names of the nodes |
|---|
| 263 | # in the cluster to which the file system belongs. |
|---|
| 264 | nodeCount=$(getNodeFile $REL_HOSTNAME_Field $fsHomeCluster $mmsdrfsFile $nodefile) |
|---|
| 265 | if [[ $nodeCount -eq 0 ]] |
|---|
| 266 | then |
|---|
| 267 | # The cluster is empty; there is nobody to run the command. |
|---|
| 268 | printErrorMsg 171 $mmcmd "getNodeFile (nodeCount=0)" 1 |
|---|
| 269 | cleanupAndExit |
|---|
| 270 | fi |
|---|
| 271 | |
|---|
| 272 | # Try the nodes one by one until you find a node that can execute the command. |
|---|
| 273 | preferredNode=0 # We have no idea where to go first; let mmcommon decide. |
|---|
| 274 | $mmcommon linkCommand $preferredNode $nodefile mmchdisk $fqDeviceName "$argListQ" |
|---|
| 275 | rc=$? |
|---|
| 276 | |
|---|
| 277 | # Update the mmsdrfs file and unlock the sdr if the action was "change" |
|---|
| 278 | # and the cluster state allowed the mmchdisk to be attempted. |
|---|
| 279 | if [[ $action = change ]] |
|---|
| 280 | then |
|---|
| 281 | if [[ $rc -ne $MM_DaemonDown && $rc -ne $MM_QuorumWait ]] |
|---|
| 282 | then |
|---|
| 283 | # Reconcile the sdrfs file with the GPFS daemon's view of the filesystem. |
|---|
| 284 | $cp $mmsdrfsFile $newsdrfs |
|---|
| 285 | reconcileSdrfsWithDaemon $deviceName $newsdrfs |
|---|
| 286 | rc2=$? |
|---|
| 287 | if [[ $rc2 -ne 0 ]] |
|---|
| 288 | then |
|---|
| 289 | # reconcileSdrfsWithDaemon failed. |
|---|
| 290 | printErrorMsg 171 $mmcmd reconcileSdrfsWithDaemon $rc2 |
|---|
| 291 | # Tell the user to run mmcommon recoverfs against the file system. |
|---|
| 292 | printErrorMsg 190 $mmcmd $deviceName $deviceName |
|---|
| 293 | cleanupAndExit $rc |
|---|
| 294 | fi |
|---|
| 295 | |
|---|
| 296 | # Get the generation number from the version line of the new sdrfs file. |
|---|
| 297 | versionLine=$($head -1 $newsdrfs) |
|---|
| 298 | IFS=':' |
|---|
| 299 | set -f ; set -- $versionLine ; set +f |
|---|
| 300 | newGenNumber=$6 |
|---|
| 301 | IFS="$IFS_sv" |
|---|
| 302 | |
|---|
| 303 | # Commit the reconciled version of the sdrfs file to the server |
|---|
| 304 | # so the admin scripts and the daemon are in sync. |
|---|
| 305 | trap "" HUP INT QUIT KILL # Disable interrupts until the commit is done. |
|---|
| 306 | gpfsObjectInfo=$(commitChanges $fsHomeCluster $nsId $gpfsObjectInfo \ |
|---|
| 307 | $newGenNumber $newsdrfs $primaryServer) |
|---|
| 308 | if [[ $? -ne 0 ]] |
|---|
| 309 | then |
|---|
| 310 | # We were unable to replace the file in the sdr. |
|---|
| 311 | printErrorMsg 381 $mmcmd |
|---|
| 312 | # Tell the user to run mmcommon recoverfs against the filesystem. |
|---|
| 313 | printErrorMsg 190 $mmcmd $deviceName $deviceName |
|---|
| 314 | fi |
|---|
| 315 | |
|---|
| 316 | # Unlock the sdr. |
|---|
| 317 | [[ $sdrLocked = yes ]] && \ |
|---|
| 318 | freeLockOnServer $primaryServer $ourNodeNumber >/dev/null |
|---|
| 319 | sdrLocked=no |
|---|
| 320 | trap localPosttrap HUP INT QUIT KILL |
|---|
| 321 | |
|---|
| 322 | # Propagate the new mmsdrfs file. This process is asynchronous. |
|---|
| 323 | propagateSdrfsFile async $nodefile $newsdrfs $newGenNumber |
|---|
| 324 | |
|---|
| 325 | else |
|---|
| 326 | # The cluster state did not allow mmchdisk to run. Just unlock the sdr. |
|---|
| 327 | [[ $sdrLocked = yes ]] && \ |
|---|
| 328 | freeLockOnServer $primaryServer $ourNodeNumber >/dev/null |
|---|
| 329 | sdrLocked=no |
|---|
| 330 | trap localPosttrap HUP INT QUIT KILL |
|---|
| 331 | |
|---|
| 332 | fi # end of if [[ $rc -ne $MM_DaemonDown && $rc -ne $MM_QuorumWait ]] |
|---|
| 333 | |
|---|
| 334 | fi # end of if [[ $action = change ]] |
|---|
| 335 | |
|---|
| 336 | cleanupAndExit $rc |
|---|
| 337 | |
|---|
