| 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. 2005,2006 |
|---|
| 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 | # @(#)93 1.1.1.3 src/avs/fs/mmfs/ts/admin/mmsdrrestore.sh, mmfs, avs_rgpfs24, rgpfs240610b 2/17/06 16:51:07 |
|---|
| 17 | ############################################################################## |
|---|
| 18 | # |
|---|
| 19 | # Ensure that the GPFS system files on the specified nodes are at the |
|---|
| 20 | # most recent level. If the local mmsdrfs file is missing, use the file |
|---|
| 21 | # specified with the -F option from the node specified with the -p option. |
|---|
| 22 | # |
|---|
| 23 | # Usage: |
|---|
| 24 | # mmsdrrestore [-p NodeName] [-F mmsdrfsFile] [-R remoteFileCopyCommand] |
|---|
| 25 | # [-a | -N {Node[,Node...] | NodeFile | NodeClass}] |
|---|
| 26 | # |
|---|
| 27 | # where: |
|---|
| 28 | # -p NodeName specifies the node from which to obtain a valid |
|---|
| 29 | # mmsdrfs file. This should be either the primary |
|---|
| 30 | # configuration server or a node that contains a |
|---|
| 31 | # valid backup copy of the mmsdrfs file. |
|---|
| 32 | # If not specified, the local node is assumed. |
|---|
| 33 | # |
|---|
| 34 | # -F mmsdrfsFile specifies the pathname of the mmsdrfs file to use. |
|---|
| 35 | # If not specified, /var/mmfs/gen/mmsdrfs is assumed. |
|---|
| 36 | # |
|---|
| 37 | # -R remoteFileCopyCommand specifies the fully-qualified pathname for |
|---|
| 38 | # the remote file copy program to be used for staging |
|---|
| 39 | # the mmsdrfs file. The default is /usr/bin/rcp. |
|---|
| 40 | # |
|---|
| 41 | # Future options: |
|---|
| 42 | # -a Restore all nodes in the GPFS cluster. |
|---|
| 43 | # |
|---|
| 44 | # -N Node,Node,... specifies the affected nodes in a list or in a file. |
|---|
| 45 | # -N NodeFile NodeClass may be one of several possible node classes |
|---|
| 46 | # -N NodeClass (e.g., quorumnodes, managernodes, nsdnodes, etc.) |
|---|
| 47 | # |
|---|
| 48 | # Notes: |
|---|
| 49 | # If not explicitly specified otherwise, only the local node is affected. |
|---|
| 50 | # |
|---|
| 51 | # If the -p and/or -F parameters are specified, i.e., if the local mmsdrfs |
|---|
| 52 | # file will be replaced, the command must be first run on the primary and |
|---|
| 53 | # backup cluster configuration servers. |
|---|
| 54 | # |
|---|
| 55 | # If the local mmsdrfs file is going to be replaced, the command saves |
|---|
| 56 | # the original file as /var/mmfs/gen/mmsdrfs.mmsdrrestore.backup |
|---|
| 57 | # |
|---|
| 58 | # The -R option has to be specified only if the mmsdrfs file needs to be |
|---|
| 59 | # copied from a different node and the cluster was set to use ssh/scp. |
|---|
| 60 | # Once the correct mmsdrfs file is staged, the rest of the code will use |
|---|
| 61 | # the remote commands that have been defined for the cluster. In other |
|---|
| 62 | # words, you can not change the remote shell and file copy programs with |
|---|
| 63 | # the mmsdrrestore command; use mmchcluster for this purpose once the |
|---|
| 64 | # cluster has been restored. |
|---|
| 65 | # |
|---|
| 66 | ############################################################################## |
|---|
| 67 | |
|---|
| 68 | # Include global declarations and service routines. |
|---|
| 69 | . /usr/lpp/mmfs/bin/mmglobfuncs |
|---|
| 70 | . /usr/lpp/mmfs/bin/mmsdrfsdef |
|---|
| 71 | . /usr/lpp/mmfs/bin/mmfsfuncs |
|---|
| 72 | |
|---|
| 73 | sourceFile="mmsdrrestore.sh" |
|---|
| 74 | [[ -n $DEBUG || -n $DEBUGmmsdrrestore ]] && set -x |
|---|
| 75 | $mmTRACE_ENTER "$*" |
|---|
| 76 | |
|---|
| 77 | |
|---|
| 78 | # Local variables |
|---|
| 79 | |
|---|
| 80 | usageMsg=284 |
|---|
| 81 | rc=0 |
|---|
| 82 | |
|---|
| 83 | |
|---|
| 84 | # Local routines |
|---|
| 85 | |
|---|
| 86 | |
|---|
| 87 | ##################################################################### |
|---|
| 88 | # |
|---|
| 89 | # Function: Verifies that a command executed successfully. If the |
|---|
| 90 | # return code from the command is not zero, the function |
|---|
| 91 | # issues a message, restores the original mmsdrfs file, |
|---|
| 92 | # and exits. |
|---|
| 93 | # |
|---|
| 94 | # Input: $1 - text to be printed in the message |
|---|
| 95 | # $2 - return code from the execution of the command |
|---|
| 96 | # |
|---|
| 97 | ##################################################################### |
|---|
| 98 | function checkForErrorsAndRestoreOldFile |
|---|
| 99 | { |
|---|
| 100 | sourceFile="mmsdrrestore.sh" |
|---|
| 101 | [[ -n $DEBUG || -n $DEBUGcheckForErrorsAndRestoreOldFile ]] && set -x |
|---|
| 102 | $mmTRACE_ENTER "$*" |
|---|
| 103 | typeset msgSub=$1 |
|---|
| 104 | typeset rc=$2 |
|---|
| 105 | |
|---|
| 106 | if [ $rc != "0" ] |
|---|
| 107 | then |
|---|
| 108 | # If asked, issue an "unexpected error" message. |
|---|
| 109 | [[ $msgSub != NOMSG ]] && \ |
|---|
| 110 | printErrorMsg 171 "$mmcmd" "$msgSub" $rc |
|---|
| 111 | |
|---|
| 112 | # Remove the mmsdrfs file if it was copied from another place. |
|---|
| 113 | [[ -n $userProvidedFile ]] && \ |
|---|
| 114 | $rm -f $mmsdrfsFile |
|---|
| 115 | |
|---|
| 116 | # Restore the original mmsdrfs file. |
|---|
| 117 | [[ -f ${mmsdrfsFile}.${mmcmd}.backup ]] && \ |
|---|
| 118 | $mv -f ${mmsdrfsFile}.${mmcmd}.backup $mmsdrfsFile |
|---|
| 119 | |
|---|
| 120 | cleanupAndExit $2 |
|---|
| 121 | fi |
|---|
| 122 | |
|---|
| 123 | } #------ end of function checkForErrorsAndRestoreOldFile ------- |
|---|
| 124 | |
|---|
| 125 | |
|---|
| 126 | |
|---|
| 127 | ####################### |
|---|
| 128 | # Mainline processing |
|---|
| 129 | ####################### |
|---|
| 130 | |
|---|
| 131 | |
|---|
| 132 | ################################## |
|---|
| 133 | # Process each of the arguments. |
|---|
| 134 | ################################## |
|---|
| 135 | [[ $arg1 = '-?' || $arg1 = '-h' || $arg1 = '--help' || $arg1 = '--' ]] && \ |
|---|
| 136 | syntaxError "help" $usageMsg |
|---|
| 137 | |
|---|
| 138 | while getopts :aF:N:p:R: OPT |
|---|
| 139 | do |
|---|
| 140 | case $OPT in |
|---|
| 141 | a) [[ -n $aflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 142 | aflag="-$OPT" |
|---|
| 143 | [[ -n $Nflag ]] && \ |
|---|
| 144 | syntaxError "invalidCombination" $usageMsg $aflag $Nflag |
|---|
| 145 | syntaxError "invalidOption" $usageMsg $OPT #esjx option not allowed yet |
|---|
| 146 | ;; |
|---|
| 147 | |
|---|
| 148 | F) # source mmsdrfs file |
|---|
| 149 | [[ -n $Farg ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 150 | Farg=$OPTARG |
|---|
| 151 | ;; |
|---|
| 152 | |
|---|
| 153 | N) [[ -n $Nflag ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 154 | Nflag="-$OPT" |
|---|
| 155 | nodenames="$OPTARG" |
|---|
| 156 | [[ -n $aflag ]] && \ |
|---|
| 157 | syntaxError "invalidCombination" $usageMsg $Nflag $aflag |
|---|
| 158 | syntaxError "invalidOption" $usageMsg $OPT #esjx option not allowed yet |
|---|
| 159 | ;; |
|---|
| 160 | |
|---|
| 161 | p) # primary server |
|---|
| 162 | [[ -n $parg ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 163 | parg=$OPTARG |
|---|
| 164 | ;; |
|---|
| 165 | |
|---|
| 166 | R) # remote file copy command |
|---|
| 167 | [[ -n $Rarg ]] && syntaxError "multiple" $noUsageMsg "-$OPT" |
|---|
| 168 | Rarg=$OPTARG |
|---|
| 169 | [[ $Rarg != /* ]] && \ |
|---|
| 170 | syntaxError "absolutePath_2" $noUsageMsg "-$OPT" "$OPTARG" |
|---|
| 171 | ;; |
|---|
| 172 | |
|---|
| 173 | :) syntaxError "missingValue" $usageMsg $OPTARG |
|---|
| 174 | ;; |
|---|
| 175 | |
|---|
| 176 | +[aFNpR]) |
|---|
| 177 | syntaxError "invalidOption" $usageMsg "$OPT" |
|---|
| 178 | ;; |
|---|
| 179 | |
|---|
| 180 | *) syntaxError "invalidOption" $usageMsg $OPTARG |
|---|
| 181 | ;; |
|---|
| 182 | |
|---|
| 183 | esac |
|---|
| 184 | done # end of while getopts :afF:N:p:r:R: OPT do |
|---|
| 185 | |
|---|
| 186 | shift OPTIND-1 |
|---|
| 187 | [[ $# != 0 ]] && syntaxError "extraArg" $usageMsg $1 |
|---|
| 188 | |
|---|
| 189 | |
|---|
| 190 | ########################################################### |
|---|
| 191 | # Process the optional remote file copy command parameter. |
|---|
| 192 | ########################################################### |
|---|
| 193 | if [[ -n $Rarg ]] |
|---|
| 194 | then |
|---|
| 195 | if [[ ! -x $Rarg ]] |
|---|
| 196 | then |
|---|
| 197 | # The specified remote file copy command is incorrect |
|---|
| 198 | # (it is not an executable). |
|---|
| 199 | printErrorMsg 423 $mmcmd ${Rarg%% *} |
|---|
| 200 | cleanupAndExit |
|---|
| 201 | fi |
|---|
| 202 | rcp="$Rarg" |
|---|
| 203 | export GPFS_rcpPath="$rcp" |
|---|
| 204 | fi |
|---|
| 205 | |
|---|
| 206 | |
|---|
| 207 | ################################################################### |
|---|
| 208 | # If an mmsdrfs file exists and we are told not to use it, create |
|---|
| 209 | # a backup of the local mmsdrfs file before proceding any further. |
|---|
| 210 | ################################################################### |
|---|
| 211 | if [[ ( -n $parg || -n $Farg && $Farg != $mmsdrfsFile ) && -f $mmsdrfsFile ]] |
|---|
| 212 | then |
|---|
| 213 | $cp $mmsdrfsFile ${mmsdrfsFile}.${mmcmd}.backup |
|---|
| 214 | checkForErrors "cp $mmsdrfsFile ${mmsdrfsFile}.${mmcmd}.backup" $? |
|---|
| 215 | fi |
|---|
| 216 | |
|---|
| 217 | |
|---|
| 218 | ############################################################## |
|---|
| 219 | # If we are told where to find the good mmsdrfs file, get it. |
|---|
| 220 | ############################################################## |
|---|
| 221 | sourceNode=$parg |
|---|
| 222 | sourceFile=$Farg |
|---|
| 223 | [[ -z $sourceFile ]] && sourceFile=$mmsdrfsFile |
|---|
| 224 | |
|---|
| 225 | if [[ -n $parg ]] |
|---|
| 226 | then |
|---|
| 227 | # The file must be copied from a remote node. |
|---|
| 228 | isNodeReachable $sourceNode |
|---|
| 229 | if [[ $? -ne 0 ]] |
|---|
| 230 | then |
|---|
| 231 | # The node is not reachable. |
|---|
| 232 | printErrorMsg 340 $mmcmd $sourceNode |
|---|
| 233 | checkForErrorsAndRestoreOldFile NOMSG 1 |
|---|
| 234 | fi |
|---|
| 235 | |
|---|
| 236 | rcpResult=$(LC_ALL=C $rcp ${sourceNode}:${sourceFile} $mmsdrfsFile 2>&1) |
|---|
| 237 | rc=$? |
|---|
| 238 | if [[ $rc -ne 0 ]] |
|---|
| 239 | then |
|---|
| 240 | # If there was no need to copy the file, that's fine. |
|---|
| 241 | print -- "$rcpResult" | $grep -e "refer to the same file" >/dev/null |
|---|
| 242 | if [[ $? -ne 0 ]] |
|---|
| 243 | then |
|---|
| 244 | # Show the error from the rcp command and give up. |
|---|
| 245 | print -u2 "$rcpResult" |
|---|
| 246 | checkForErrorsAndRestoreOldFile \ |
|---|
| 247 | "$rcp ${sourceNode}:${sourceFile} $mmsdrfsFile" $rc |
|---|
| 248 | fi |
|---|
| 249 | fi # end of if [[ $rc -ne 0 ]] |
|---|
| 250 | userProvidedFile=yes |
|---|
| 251 | |
|---|
| 252 | elif [[ $sourceFile != $mmsdrfsFile ]] |
|---|
| 253 | then |
|---|
| 254 | # The file must be copied from some other place within the local node. |
|---|
| 255 | $cp $sourceFile $mmsdrfsFile |
|---|
| 256 | checkForErrorsAndRestoreOldFile "cp $sourceFile $mmsdrfsFile" $? |
|---|
| 257 | userProvidedFile=yes |
|---|
| 258 | |
|---|
| 259 | else |
|---|
| 260 | : # The local mmsdrfs file is supposed to be good. |
|---|
| 261 | fi # end of if [[ -n $parg ]] |
|---|
| 262 | |
|---|
| 263 | # At this point we must have an mmsdrfs file to work with. |
|---|
| 264 | if [[ ! -f $mmsdrfsFile ]] |
|---|
| 265 | then |
|---|
| 266 | # Either the node does not belong to a cluster |
|---|
| 267 | # or the config information is lost/corrupted. |
|---|
| 268 | printErrorMsg 282 $mmcmd |
|---|
| 269 | checkForErrors "$mmcmd: Missing /var/mmfs/gen/mmsdrfs" 1 |
|---|
| 270 | fi |
|---|
| 271 | |
|---|
| 272 | # If this is a user-specified mmsdrfs file, perform a quick sanity check. |
|---|
| 273 | if [[ -n $userProvidedFile ]] |
|---|
| 274 | then |
|---|
| 275 | checkSdrfsFile $mmsdrfsFile |
|---|
| 276 | checkForErrorsAndRestoreOldFile "$mmcmd: Invalid input file" $? |
|---|
| 277 | fi |
|---|
| 278 | |
|---|
| 279 | |
|---|
| 280 | ######################################################################## |
|---|
| 281 | # See if this node belongs to the cluster defined by this mmsdrfs file. |
|---|
| 282 | ######################################################################## |
|---|
| 283 | # Remove the local mmfsNodeData file. We will derive the information |
|---|
| 284 | # based on the currently active IP addresses of the node. |
|---|
| 285 | $rm -f $mmfsNodeData |
|---|
| 286 | |
|---|
| 287 | # Determine the node's GPFS node number by checking all local addresses. |
|---|
| 288 | nodeData=$(guessLocalNodeNumber) |
|---|
| 289 | rc=$? |
|---|
| 290 | set -f ; set -- $nodeData ; set +f |
|---|
| 291 | ourNodeNumber=$1 |
|---|
| 292 | ourNodeName=$3 |
|---|
| 293 | |
|---|
| 294 | if [[ -z $ourNodeNumber ]] |
|---|
| 295 | then |
|---|
| 296 | # The node does not appear to belong to the cluster rpresented |
|---|
| 297 | # by this mmsdrfs file. Either the file is from a different |
|---|
| 298 | # cluster, or not all adapter interfaces have been activated. |
|---|
| 299 | # Or the node indeed is not part of the cluster. |
|---|
| 300 | |
|---|
| 301 | # Determine the name of the source file for the message. |
|---|
| 302 | sdrfs="$sourceFile" |
|---|
| 303 | [[ -n $sourceNode ]] && sdrfs="${sourceNode}:${sourceFile}" |
|---|
| 304 | |
|---|
| 305 | # There is no record for this node in the file. |
|---|
| 306 | printErrorMsg 283 $mmcmd $sdrfs |
|---|
| 307 | checkForErrorsAndRestoreOldFile NOMSG 1 |
|---|
| 308 | fi |
|---|
| 309 | |
|---|
| 310 | |
|---|
| 311 | ############################################################################# |
|---|
| 312 | # At this point, the current mmsdrfs file is assumed to be the correct one. |
|---|
| 313 | # The code continues with the standard gpfsInit processing. This means |
|---|
| 314 | # that the primary or backup servers, as per the current mmsdrfs file, |
|---|
| 315 | # will be consulted to ensure this is indeed the latest version of the file. |
|---|
| 316 | ############################################################################# |
|---|
| 317 | # Remove the current versions of the system files to force the environment |
|---|
| 318 | # to be rebuilt based on the information in the new mmsdrfs file. |
|---|
| 319 | $rm -f $mmfscfgFile $mmfsNodeData ${mmfsEnvLevel}+([0-9]) \ |
|---|
| 320 | ${mmfsNewKeyLevel}+([0-9]) ${mmfsCommittedKeyLevel}+([0-9]) \ |
|---|
| 321 | $nsdpvol |
|---|
| 322 | |
|---|
| 323 | |
|---|
| 324 | ######################################################################## |
|---|
| 325 | # Set up trap exception handling and call the gpfsInit function. |
|---|
| 326 | # It will ensure that the local copy of the mmsdrfs and the rest of the |
|---|
| 327 | # GPFS system files are up-to-date. There is no need to lock the sdr. |
|---|
| 328 | ######################################################################## |
|---|
| 329 | trap pretrap2 HUP INT QUIT KILL |
|---|
| 330 | gpfsInitOutput=$(gpfsInit nolock) |
|---|
| 331 | setGlobalVar $? $gpfsInitOutput |
|---|
| 332 | |
|---|
| 333 | |
|---|
| 334 | ########################################################### |
|---|
| 335 | # Ensure the sdrserv daemon is running on this node only |
|---|
| 336 | # if the node is a primary or backup configuration server. |
|---|
| 337 | ########################################################### |
|---|
| 338 | killSdrServ >/dev/null 2>&1 |
|---|
| 339 | [[ $ourNodeName = $primaryServer || $ourNodeName = $backupServer ]] && \ |
|---|
| 340 | startSdrServ CURRENT |
|---|
| 341 | |
|---|
| 342 | # The node was successfully restored. |
|---|
| 343 | printErrorMsg 285 $mmcmd $ourNodeName |
|---|
| 344 | |
|---|
| 345 | # Done for now. The -a and -N options are not implemented yet. |
|---|
| 346 | cleanupAndExit 0 |
|---|
| 347 | |
|---|
| 348 | |
|---|
| 349 | ####################################################### |
|---|
| 350 | # Create a file containing all of the affected nodes. |
|---|
| 351 | ####################################################### |
|---|
| 352 | if [[ -n $aflag ]] |
|---|
| 353 | then |
|---|
| 354 | # Get a list of all nodes in the cluster. |
|---|
| 355 | getNodeList $REL_HOSTNAME_Field $GLOBAL_ID $mmsdrfsFile > $nodefile |
|---|
| 356 | |
|---|
| 357 | elif [[ -n $Nflag ]] |
|---|
| 358 | then |
|---|
| 359 | # Convert the passed data into a file containing admin node names. |
|---|
| 360 | createVerifiedNodefile $nodenames $REL_HOSTNAME_Field no $nodefile |
|---|
| 361 | [[ $? -ne 0 ]] && cleanupAndExit |
|---|
| 362 | |
|---|
| 363 | else |
|---|
| 364 | : # Only the local node is affected. |
|---|
| 365 | fi # end of if [[ -n $aflag ]] |
|---|
| 366 | |
|---|
| 367 | |
|---|
| 368 | # Ensure we have the proper credentials. |
|---|
| 369 | [[ $getCredCalled = no ]] && getCred |
|---|
| 370 | |
|---|
| 371 | |
|---|
| 372 | #esjxx - If more than one node is involved AND a new mmsdrfs file was copied, |
|---|
| 373 | #esjxx push the file to the rest of the affected nodes. This will be done |
|---|
| 374 | #esjxx with a new mmremote call that will take care of mmsdrserv as well. |
|---|
| 375 | #esjxx We will need to pass the names of the primary and backup servers |
|---|
| 376 | #esjxx and clusterID to ensure that the individual nodes will go to the |
|---|
| 377 | #esjxx right place for the correct version of the mmsdrfs file. |
|---|
| 378 | #esjxx If the individual nodes have different information, they should go |
|---|
| 379 | #esjxx to the correct primary only and specify current gen number 0 or 1. |
|---|
| 380 | #esjxx This should apply to the backup too ??? |
|---|
| 381 | |
|---|
| 382 | |
|---|
| 383 | ########################## |
|---|
| 384 | # Restore the GPFS nodes. |
|---|
| 385 | ########################## |
|---|
| 386 | if [[ ! -s $nodefile ]] |
|---|
| 387 | then |
|---|
| 388 | : # This is a request to restore only the local node. We are done. |
|---|
| 389 | else |
|---|
| 390 | # This is a request to restore other nodes as well. |
|---|
| 391 | $mmcommon onall $nodefile $unreachedNodes cfg -f |
|---|
| 392 | rc=$? |
|---|
| 393 | fi # end of if [[ ! -s $nodefile ]] |
|---|
| 394 | |
|---|
| 395 | # If any nodes could not be reached, tell the user which ones. |
|---|
| 396 | if [[ -s $unreachedNodes ]] |
|---|
| 397 | then |
|---|
| 398 | # The following nodes could not be reached: . . . |
|---|
| 399 | printErrorMsg 270 $mmcmd |
|---|
| 400 | $cat $unreachedNodes 1>&2 |
|---|
| 401 | fi |
|---|
| 402 | |
|---|
| 403 | cleanupAndExit $rc |
|---|
| 404 | |
|---|
