xcat-core/xCAT-server/lib/xcat/plugins/esx.pm
2013-10-02 13:46:15 -04:00

5100 lines
200 KiB
Perl

package xCAT_plugin::esx;
use strict;
use warnings;
use xCAT::Table;
use xCAT::Utils;
use xCAT::TableUtils;
use xCAT::ServiceNodeUtils;
use xCAT::TZUtils;
use Time::HiRes qw (sleep);
use xCAT::Template;
use xCAT::MsgUtils;
use xCAT::SvrUtils;
use xCAT::NodeRange;
use xCAT::Common;
use xCAT::VMCommon;
use POSIX "WNOHANG";
use Getopt::Long;
use Thread qw(yield);
use POSIX qw(WNOHANG nice);
use File::Path qw/mkpath rmtree/;
use File::Temp qw/tempdir/;
use File::Copy;
use Fcntl qw/:flock/;
use IO::Socket; #Need name resolution
use Scalar::Util qw/looks_like_number/;
#use Data::Dumper;
Getopt::Long::Configure("bundling");
Getopt::Long::Configure("pass_through");
my @cpiopid;
our @ISA = 'xCAT::Common';
#in xCAT, the lifetime of a process ends on every request
#therefore, the lifetime of assignments to these glabals as architected
#is to be cleared on every request
#my %esx_comm_pids;
my %limbonodes; #nodes in limbo during a forced migration due to missing parent
my %hyphash; #A data structure to hold hypervisor-wide variables (i.e. the current resource pool, virtual machine folder, connection object
my %vcenterhash; #A data structure to reflect the state of vcenter connectivity to hypervisors
my %vmhash; #store per vm info of interest
my %clusterhash;
my %hypready; #A structure for hypervisor readiness to be tracked before proceeding to normal operations
my %running_tasks; #A struct to track this processes
my $output_handler; #Pointer to the function to drive results to client
my $executerequest;
my $usehostnamesforvcenter;
my %tablecfg; #to hold the tables
my %hostrefbynode;
my $currkey;
my $requester;
my $viavcenter;
my $viavcenterbyhyp;
my $vcenterautojoin=1;
my $datastoreautomount=1;
my $vcenterforceremove=0; #used in rmhypervisor
my $reconfigreset=1;
my $vmwaresdkdetect = eval {
require VMware::VIRuntime;
VMware::VIRuntime->import();
1;
};
my %lockhandles;
sub recursion_copy {
my $source = shift;
my $destination = shift;
my $dirhandle;
opendir($dirhandle,$source);
my $entry;
foreach $entry (readdir($dirhandle)) {
if ($entry eq '.' or $entry eq '..') { next; }
my $tempsource = "$source/$entry";
my $tempdestination = "$destination/$entry";
if ( -d $tempsource ) {
unless (-d $tempdestination) { mkdir $tempdestination or die "failure creating directory $tempdestination, $!"; }
recursion_copy($tempsource,$tempdestination);
} else {
copy($tempsource,$tempdestination) or die "failed copy from $tempsource to $tempdestination, $!";
}
}
}
sub lockbyname {
my $name = shift;
my $lckh;
mkpath("/tmp/xcat/locks/");
while (-e "/tmp/xcat/locks/$name") { sleep 1; }
open($lockhandles{$name},">>","/tmp/xcat/locks/$name");
flock($lockhandles{$name},LOCK_EX);
}
sub unlockbyname {
my $name = shift;
unlink("/tmp/xcat/locks/$name");
close($lockhandles{$name});
}
my %guestidmap = (
"rhel.6.*" => "rhel6_",
"rhel.5.*" => "rhel5_",
"rhel4.*" => "rhel4_",
"centos6.*" => "rhel6_",
"centos5.*" => "rhel5_",
"centos4.*" => "rhel4_",
"sles11.*" => "sles11_",
"sles10.*" => "sles10_",
"win2k8" => "winLonghorn",
"win2k8r2" => "windows7Server",
"win2012" => "windows8Server",
"hyperv2012" => "windows8Server",
"esix5.*" => "vmkernel5",
"esix4.*" => "vmkernel",
"win8" => "windows8_",
"win7" => "windows7_",
"win2k3" => "winNetStandard",
"imagex" => "winNetStandard",
"boottarget" => "otherLinux"
#otherGuest, otherGuest64, otherLinuxGuest, otherLinux64Guest
);
sub handled_commands{
return {
copycd => 'esx',
mknetboot => "nodetype:os=(esxi.*)",
mkinstall => "nodetype:os=(esxi5.*)",
rpower => 'nodehm:power,mgt',
rsetboot => 'nodehm:power,mgt',
rmigrate => 'nodehm:power,mgt',
mkvm => 'nodehm:mgt',
rmvm => 'nodehm:mgt',
clonevm => 'nodehm:mgt',
createvcluster => 'esx',
lsvcluster => 'esx',
rinv => 'nodehm:mgt',
chvm => 'nodehm:mgt',
rshutdown => "nodetype:os=(esxi.*)",
lsvm => ['hypervisor:type','nodetype:os=(esx.*)'],
rmhypervisor => ['hypervisor:type','nodetype:os=(esx.*)'],
chhypervisor => ['hypervisor:type','nodetype:os=(esx.*)'],
#lsvm => 'nodehm:mgt', not really supported yet
};
}
sub preprocess_request {
my $request = shift;
my $callback = shift;
if ($request->{command}->[0] eq 'createvcluster' or $request->{command}->[0] eq 'lsvcluster') {
return [$request];
}
#if already preprocessed, go straight to request
if ( (defined($request->{_xcatpreprocessed}))
&& ($request->{_xcatpreprocessed}->[0] == 1))
{
return [$request];
}
my $username = 'root';
my $password = '';
my $vusername = "Administrator";
my $vpassword = "";
unless ($request and $request->{command} and $request->{command}->[0]) { return; }
if ($request->{command}->[0] eq 'copycd')
{ #don't farm out copycd
return [$request];
}elsif($request->{command}->[0] eq 'mknetboot'
or $request->{command}->[0] eq 'mkinstall'){
return [$request];
}
xCAT::Common::usage_noderange($request,$callback);
if ($request->{_xcatpreprocessed} and $request->{_xcatpreprocessed}->[0] == 1) { return [$request]; }
# exit if preprocesses
my @requests;
my $noderange = $request->{node}; # array ref
my $command = $request->{command}->[0];
my $extraargs = $request->{arg};
my @exargs=($request->{arg});
my %hyp_hash = ();
my %cluster_hash=();
# Get nodes from mp table and assign nodes to mp hash.
my $passtab = xCAT::Table->new('passwd');
my $tmp;
if ($passtab) {
($tmp) = $passtab->getAttribs({'key'=>'vmware'},'username','password');
if (defined($tmp)) {
$username = $tmp->{username};
$password = $tmp->{password};
}
($tmp) = $passtab->getAttribs({'key'=>'vcenter'},'username','password');
if (defined($tmp)) {
$vusername = $tmp->{username};
$vpassword = $tmp->{password};
}
}
my $vmtab = xCAT::Table->new("vm");
unless($vmtab){
$callback->({data=>["Cannot open vm table"]});
$request = {};
return;
}
my $vmtabhash = $vmtab->getNodesAttribs($noderange,['host','migrationdest']);
foreach my $node (@$noderange){
if ($command eq "rmhypervisor" or $command eq 'lsvm' or $command eq 'rshutdown' or $command eq "chhypervisor") {
$hyp_hash{$node}{nodes} = [$node];
} else {
my $ent = $vmtabhash->{$node}->[0];
if(defined($ent->{host})) {
push @{$hyp_hash{$ent->{host}}{nodes}}, $node;
} elsif (defined($ent->{migrationdest})) {
$cluster_hash{$ent->{migrationdest}}->{nodes}->{$node}=1;
} else {
xCAT::SvrUtils::sendmsg([1,": no host or cluster defined for guest"], $callback,$node);
}
}
}
# find service nodes for the MMs
# build an individual request for each service node
my $service = "xcat";
my @hyps=keys(%hyp_hash);
my %targethyps;
if ($command eq 'rmigrate' and (scalar @{$extraargs} >= 1)) {
@ARGV=@{$extraargs};
my $offline;
my $junk;
GetOptions(
"f" => \$offline,
"s=s" => \$junk #wo don't care about it, but suck up nfs:// targets so they don't get added
);
my $dsthyp = $ARGV[0];
if ($dsthyp) {
push @hyps,$dsthyp;
$targethyps{$dsthyp}=1;
}
}
#TODO: per hypervisor table password lookup
my @allnodes;
push @allnodes,@hyps;
push @allnodes,@$noderange;
my $sn = xCAT::ServiceNodeUtils->get_ServiceNode(\@allnodes, $service, "MN");
#vmtabhash was from when we had vm.host do double duty for hypervisor data
#$vmtabhash = $vmtab->getNodesAttribs(\@hyps,['host']);
#We now use hypervisor fields to be unambiguous
my $hyptab = xCAT::Table->new('hypervisor');
my $hyptabhash={};
if ($hyptab) {
$hyptabhash = $hyptab->getNodesAttribs(\@hyps,['mgr']);
}
# build each request for each service node
foreach my $snkey (keys %$sn){
my $reqcopy = {%$request};
$reqcopy->{'_xcatdest'} = $snkey;
$reqcopy->{_xcatpreprocessed}->[0] = 1;
my $hyps1=$sn->{$snkey};
my @moreinfo=();
my @nodes=();
foreach (@$hyps1) { #This preserves the constructed data to avoid redundant table lookup
my $cfgdata;
if (not $targethyps{$_} and not $hyp_hash{$_}) { #a vm, skip it
next;
} elsif ($hyp_hash{$_}{nodes}) {
push @nodes, @{$hyp_hash{$_}{nodes}};
$cfgdata = "[$_][".join(',',@{$hyp_hash{$_}{nodes}})."][$username][$password][$vusername][$vpassword]"; #TODO: not use vm.host?
} else {
$cfgdata = "[$_][][$username][$password][$vusername][$vpassword]"; #TODO: not use vm.host?
}
if (defined $hyptabhash->{$_}->[0]->{mgr}) {
$cfgdata .= "[". $hyptabhash->{$_}->[0]->{mgr}."]";
} else {
$cfgdata .= "[]";
}
push @moreinfo, $cfgdata; #"[$_][".join(',',@{$hyp_hash{$_}{nodes}})."][$username][$password]";
}
foreach (keys %cluster_hash) {
my $cluster;
my $vcenter;
if (/@/) {
($cluster,$vcenter) = split /@/,$_,2;
} else {
die "TODO: implement default vcenter (for now, user, do vm.migratiodest=cluster".'@'."vcentername)";
}
push @moreinfo,"[CLUSTER:$cluster][".join(',',keys %{$cluster_hash{$_}->{nodes}})."][$username][$password][$vusername][$vpassword][$vcenter]";
}
if (scalar @nodes) {
$reqcopy->{node} = \@nodes;
}
#print "nodes=@nodes\n";
$reqcopy->{moreinfo}=\@moreinfo;
push @requests, $reqcopy;
}
return \@requests;
}
sub process_request {
#$SIG{INT} = $SIG{TERM} = sub{
# foreach (keys %esx_comm_pids){
# kill 2,$_;
# }
# exit 0;
#};
my $request = shift;
$output_handler = shift;
$executerequest = shift;
if ($request->{_xcat_authname}->[0]) {
$requester=$request->{_xcat_authname}->[0];
}
my $level = shift;
my $distname = undef;
my $arch = undef;
my $path = undef;
my $command = $request->{command}->[0];
#The first segment is fulfilling the role of this plugin as
#a hypervisor provisioning plugin (akin to anaconda, windows, sles plugins)
if($command eq 'copycd'){
return copycd($request,$executerequest);
}elsif($command eq 'mkinstall'){
return mkinstall($request,$executerequest);
}elsif($command eq 'mknetboot'){
return mknetboot($request,$executerequest);
}
#From here on out, code for managing guests under VMware
#Detect whether or not the VMware SDK is available on this specific system
unless ($vmwaresdkdetect) {
$vmwaresdkdetect = eval {
require VMware::VIRuntime;
VMware::VIRuntime->import();
1;
};
}
unless ($vmwaresdkdetect) {
xCAT::SvrUtils::sendmsg([1,"VMWare SDK required for operation, but not installed"], $output_handler);
return;
}
if ($command eq 'createvcluster') {
create_new_cluster($request);
return;
}
if ($command eq 'lsvcluster') {
list_clusters($request);
return;
}
my $moreinfo;
my $noderange = $request->{node};
xCAT::VMCommon::grab_table_data($noderange,\%tablecfg,$output_handler);
my @exargs;
unless($command){
return; # Empty request
}
if (ref($request->{arg})) {
@exargs = @{$request->{arg}};
} else {
@exargs = ($request->{arg});
}
#my $sitetab = xCAT::Table->new('site');
#if($sitetab){
#(my $ref) = $sitetab->getAttribs({key => 'usehostnamesforvcenter'}, 'value');
my @entries = xCAT::TableUtils->get_site_attribute("usehostnamesforvcenter");
my $t_entry = $entries[0];
if ( defined($t_entry) ) {
$usehostnamesforvcenter = $t_entry;
}
#($ref) = $sitetab->getAttribs({key => 'vcenterautojoin'}, 'value');
@entries = xCAT::TableUtils->get_site_attribute("vcenterautojoin");
$t_entry = $entries[0];
if ( defined($t_entry) ) {
$vcenterautojoin = $t_entry;
if ($vcenterautojoin =~ /^n/ or $vcenterautojoin =~ /^dis/) {
$vcenterautojoin=0;
}
}
#($ref) = $sitetab->getAttribs({key => 'vmwaredatastoreautomount'}, 'value');
@entries = xCAT::TableUtils->get_site_attribute("vmwaredatastoreautomount");
$t_entry = $entries[0];
if ( defined($t_entry) ) {
$datastoreautomount = $t_entry;
if ($datastoreautomount =~ /^n/ or $datastoreautomount =~ /^dis/) {
$datastoreautomount=0;
}
}
#($ref) = $sitetab->getAttribs({key => 'vmwarereconfigonpower'},'value');
@entries = xCAT::TableUtils->get_site_attribute("vmwarereconfigonpower");
$t_entry = $entries[0];
if ( defined($t_entry) ) {
$reconfigreset=$t_entry;
if ($reconfigreset =~ /^(n|d)/i) { #if no or disable, skip it
$reconfigreset=0;
}
}
# }
if ($request->{moreinfo}) { $moreinfo=$request->{moreinfo}; }
else { $moreinfo=build_more_info($noderange,$output_handler);}
foreach my $info (@$moreinfo) {
$info=~/^\[(.*?)\]\[(.*?)\]\[(.*?)\]\[(.*?)\]\[(.*?)\]\[(.*?)\]\[(.*?)\]/;
my $hyp=$1;
my @nodes=split(',', $2);
my $username = $3;
my $password = $4;
my $tmpvcname=$7;
my $tmpvcuname=$5;
my $tmpvcpass=$6;
if ($hyp =~ /^CLUSTER:/) { #a cluster, not a host.
$hyp =~ s/^CLUSTER://;
$clusterhash{$hyp}->{vcenter}->{name} = $tmpvcname;
$clusterhash{$hyp}->{vcenter}->{username} = $tmpvcuname;
$clusterhash{$hyp}->{vcenter}->{password} = $tmpvcpass;
foreach (@nodes) {
$clusterhash{$hyp}->{nodes}->{$_}=1;
}
next;
}
$hyphash{$hyp}->{vcenter}->{name} = $tmpvcname;
$hyphash{$hyp}->{vcenter}->{username} = $tmpvcuname;
$hyphash{$hyp}->{vcenter}->{password} = $tmpvcpass;
$hyphash{$hyp}->{username}=$username;# $nodeid;
$hyphash{$hyp}->{password}=$password;# $nodeid;
unless ($hyphash{$hyp}->{vcenter}->{password}) {
$hyphash{$hyp}->{vcenter}->{password} = "";
}
my $ent;
for (my $i=0; $i<@nodes; $i++){
if ($command eq 'rmigrate' and grep /-f/, @exargs) { #offline migration,
$hyphash{$hyp}->{offline} = 1; #if it is migrate and it has nodes, it is a source hypervisor apt to be offline
#this will hint to relevant code to operate under the assumption of a
#downed hypervisor source
#note this will make dangerous assumptions, it will make a very minimal attempt
#to operate normally, but really should only be called if the source is down and
#fenced (i.e. storage, network, or turned off and stateless
}
my $node = $nodes[$i];
#my $nodeid = $ids[$i];
$hyphash{$hyp}->{nodes}->{$node}=1;# $nodeid;
}
}
my $hyptab = xCAT::Table->new('hypervisor',create=>0);
if ($hyptab) {
my @hyps = keys %hyphash;
$tablecfg{hypervisor} = $hyptab->getNodesAttribs(\@hyps,['mgr','netmap','defaultnet','cluster','preferdirect','datacenter']);
}
my $hoststab = xCAT::Table->new('hosts',create=>0);
if ($hoststab) {
my @hyps = keys %hyphash;
$tablecfg{hosts} = $hoststab->getNodesAttribs(\@hyps,['hostnames']);
}
#my $children = 0;
#my $vmmaxp = 84;
#$SIG{CHLD} = sub { my $cpid; while ($cpid = waitpid(-1, WNOHANG) > 0) { delete $esx_comm_pids{$cpid}; $children--; } };
$viavcenter = 0;
if ($command eq 'rmigrate' or $command eq 'rmhypervisor') { #Only use vcenter when required, fewer prereqs
$viavcenter = 1;
}
if ($command eq 'rmhypervisor' and grep /-f/, @exargs) { #force remove of hypervisor
$vcenterforceremove=1;
}
my $keytab = xCAT::Table->new('prodkey');
if ($keytab) {
my @hypes = keys %hyphash;
$tablecfg{prodkey} = $keytab->getNodesAttribs(\@hypes,[qw/product key/]);
}
my $hyp;
my %needvcentervalidation;
my $cluster;
foreach $cluster (keys %clusterhash) {
my $vcenter = $clusterhash{$cluster}->{vcenter}->{name};
unless ($vcenterhash{$vcenter}->{conn}) {
eval {
$vcenterhash{$vcenter}->{conn} = Vim->new(service_url=>"https://$vcenter/sdk");
$vcenterhash{$vcenter}->{conn}->login(user_name => $clusterhash{$cluster}->{vcenter}->{username},
password => $clusterhash{$cluster}->{vcenter}->{password});
};
if ($@) {
$vcenterhash{$vcenter}->{conn} = undef;
xCAT::SvrUtils::sendmsg([1,"Unable to reach $vcenter vCenter server to manage cluster $cluster: $@"], $output_handler);
next;
}
my $clusternode;
}
$clusterhash{$cluster}->{conn}=$vcenterhash{$vcenter}->{conn};
foreach my $clusternode (keys %{$clusterhash{$cluster}->{nodes}}) {
$vmhash{$clusternode}->{conn}=$vcenterhash{$vcenter}->{conn};
}
}
foreach $hyp (sort(keys %hyphash)){
#if($pid == 0){
if ($viavcenter or (defined $tablecfg{hypervisor}->{$hyp}->[0]->{mgr} and not $tablecfg{hypervisor}->{$hyp}->[0]->{preferdirect})) {
$viavcenterbyhyp->{$hyp}=1;
$hypready{$hyp} = 0; #This hypervisor requires a flag be set to signify vCenter sanenes before proceeding
my $vcenter = $hyphash{$hyp}->{vcenter}->{name};
unless ($vcenterhash{$vcenter}->{conn}) {
eval {
$vcenterhash{$vcenter}->{conn} =
Vim->new(service_url=>"https://$vcenter/sdk");
$vcenterhash{$vcenter}->{conn}->login(
user_name => $hyphash{$hyp}->{vcenter}->{username},
password => $hyphash{$hyp}->{vcenter}->{password}
);
};
if ($@) {
$vcenterhash{$vcenter}->{conn} = undef;
xCAT::SvrUtils::sendmsg([1,"Unable to reach $vcenter vCenter server to manage $hyp: $@"], $output_handler);
next;
}
}
my $hypnode;
foreach $hypnode (keys %{$hyphash{$hyp}->{nodes}}) {
$vmhash{$hypnode}->{conn}=$vcenterhash{$hyphash{$hyp}->{vcenter}->{name}}->{conn};
}
$hyphash{$hyp}->{conn} = $vcenterhash{$hyphash{$hyp}->{vcenter}->{name}}->{conn};
$hyphash{$hyp}->{vcenter}->{conn} = $vcenterhash{$hyphash{$hyp}->{vcenter}->{name}}->{conn};
$needvcentervalidation{$hyp}=$vcenter;
$vcenterhash{$vcenter}->{allhyps}->{$hyp}=1;
} else {
eval {
$hyphash{$hyp}->{conn} = Vim->new(service_url=>"https://$hyp/sdk");
$hyphash{$hyp}->{conn}->login(user_name=>$hyphash{$hyp}->{username},password=>$hyphash{$hyp}->{password});
};
if ($@) {
$hyphash{$hyp}->{conn} = undef;
xCAT::SvrUtils::sendmsg([1,"Unable to reach $hyp to perform operation due to $@"], $output_handler);
$hypready{$hyp} = -1;
next;
}
my $localnode;
foreach $localnode (keys %{$hyphash{$hyp}->{nodes}}) {
$vmhash{$localnode}->{conn}=$hyphash{$hyp}->{conn};
}
validate_licenses($hyp);
}
#}else{
# $esx_comm_pids{$pid} = 1;
#}
}
foreach $hyp (keys %needvcentervalidation) {
my $vcenter = $needvcentervalidation{$hyp};
if (not defined $vcenterhash{$vcenter}->{hostviews}) {
populate_vcenter_hostviews($vcenter);
}
if (validate_vcenter_prereqs($hyp, \&declare_ready, {
hyp=>$hyp,
vcenter=>$vcenter
}) eq "failed") {
$hypready{$hyp} = -1;
}
}
while (grep { $_ == 0 } values %hypready) {
wait_for_tasks();
sleep (1); #We'll check back in every second. Unfortunately, we have to poll since we are in web service land
}
my @badhypes;
if (grep { $_ == -1 } values %hypready) {
foreach (keys %hypready) {
if ($hypready{$_} == -1) {
unless ($hyphash{$_}->{offline}) {
push @badhypes,$_;
}
my @relevant_nodes = sort (keys %{$hyphash{$_}->{nodes}});
my $sadhypervisor=$_;
foreach (@relevant_nodes) {
if ($command eq "rmigrate" and grep /-f/,@exargs) { $limbonodes{$_}=$needvcentervalidation{$sadhypervisor}; } else {
xCAT::SvrUtils::sendmsg([1,": hypervisor unreachable"], $output_handler,$_);
}
if ($command eq "rpower" and grep /stat/,@exargs) { $limbonodes{$_}=$needvcentervalidation{$sadhypervisor}; } #try to stat power anyway through vcenter of interest...
}
delete $hyphash{$_};
}
}
if (@badhypes) {
xCAT::SvrUtils::sendmsg([1,": The following hypervisors failed to become ready for the operation: ".join(',',@badhypes)], $output_handler);
}
}
do_cmd($command,@exargs);
foreach (@badhypes) { delete $hyphash{$_}; }
foreach my $vm (sort(keys %vmhash)){
$vmhash{$vm}->{conn}->logout();
}
}
sub validate_licenses {
my $hyp = shift;
my $conn = $hyphash{$hyp}->{conn}; #This can't possibly be called via a cluster stack, so hyphash is appropriate here
unless ($tablecfg{prodkey}->{$hyp}) { #if no license specified, no-op
return;
}
my $hv = get_hostview(hypname=>$hyp,conn=>$conn,properties=>['configManager','name']);
my $lm = $conn->get_view(mo_ref=>$hv->configManager->licenseManager);
my @licenses;
foreach (@{$lm->licenses}) {
push @licenses,uc($_->licenseKey);
}
my @newlicenses;
foreach (@{$tablecfg{prodkey}->{$hyp}}) {
if (defined($_->{product}) and $_->{product} eq 'esx') {
my $key = uc($_->{key});
unless (grep /$key/,@licenses) {
push @newlicenses,$key;
}
}
}
foreach (@newlicenses) {
$lm->UpdateLicense(licenseKey=>$_);
}
}
sub do_cmd {
my $command = shift;
my @exargs = @_;
if ($command eq 'rpower') {
generic_vm_operation(['config.name','config.guestId','config.hardware.memoryMB','config.hardware.numCPU','runtime.powerState','runtime.host'],\&power,@exargs);
} elsif ($command eq 'rmvm') {
generic_vm_operation(['config.name','runtime.powerState','runtime.host'],\&rmvm,@exargs);
} elsif ($command eq 'rsetboot') {
generic_vm_operation(['config.name','runtime.host'],\&setboot,@exargs);
} elsif ($command eq 'rinv') {
generic_vm_operation(['config.name','config','runtime.host','layoutEx'],\&inv,@exargs);
} elsif ($command eq 'rmhypervisor') {
generic_hyp_operation(\&rmhypervisor,@exargs);
} elsif ($command eq 'rshutdown') {
generic_hyp_operation(\&rshutdown,@exargs);
} elsif ($command eq 'chhypervisor') {
generic_hyp_operation(\&chhypervisor,@exargs);
} elsif ($command eq 'lsvm') {
generic_hyp_operation(\&lsvm,@exargs);
} elsif ($command eq 'clonevm') {
generic_hyp_operation(\&clonevms,@exargs);
} elsif ($command eq 'mkvm') {
generic_hyp_operation(\&mkvms,@exargs);
} elsif ($command eq 'chvm') {
generic_vm_operation(['config.name','config','runtime.host'],\&chvm,@exargs);
#generic_hyp_operation(\&chvm,@exargs);
} elsif ($command eq 'rmigrate') { #Technically, on a host view, but vcenter path is 'weirder'
generic_hyp_operation(\&migrate,@exargs);
}
wait_for_tasks();
if ($command eq 'clonevm') { #TODO: unconditional, remove mkvms hosted copy
my @dhcpnodes;
foreach (keys %{$tablecfg{dhcpneeded}}) {
push @dhcpnodes,$_;
delete $tablecfg{dhcpneeded}->{$_};
}
unless ($::XCATSITEVALS{'dhcpsetup'} and ($::XCATSITEVALS{'dhcpsetup'} =~ /^n/i or $::XCATSITEVALS{'dhcpsetup'} =~ /^d/i or $::XCATSITEVALS{'dhcpsetup'} eq '0')) {
$executerequest->({command=>['makedhcp'],node=>\@dhcpnodes});
}
}
}
#inventory request for esx
sub inv {
my %args = @_;
my $node = $args{node};
my $hyp = $args{hyp};
if (not defined $args{vmview}) { #attempt one refresh
$args{vmview} = $vmhash{$node}->{conn}->find_entity_view(view_type => 'VirtualMachine',properties=>['config.name','runtime.powerState'],filter=>{name=>$node});
if (not defined $args{vmview}) {
xCAT::SvrUtils::sendmsg([1,"VM does not appear to exist"], $output_handler,$node);
return;
}
}
if (not $args{vmview}->{config}) {
xCAT::SvrUtils::sendmsg([1,"VM is in an invalid state"], $output_handler,$node);
return;
}
@ARGV= @{$args{exargs}};
require Getopt::Long;
my $tableUpdate;
my $rc = GetOptions(
't' => \$tableUpdate,
);
$SIG{__WARN__} = 'DEFAULT';
if(@ARGV > 1) {
xCAT::SvrUtils::sendmsg("Invalid arguments: @ARGV", $output_handler);
return;
}
if(!$rc) {
return;
}
my $vmview = $args{vmview};
my $moref = $vmview->{mo_ref}->value;
xCAT::SvrUtils::sendmsg("Managed Object Reference: $moref", $output_handler,$node);
my $uuid = $vmview->config->uuid;
$uuid =~ s/(..)(..)(..)(..)-(..)(..)-(..)(..)/$4$3$2$1-$6$5-$8$7/;
xCAT::SvrUtils::sendmsg("UUID/GUID: $uuid", $output_handler,$node);
my $cpuCount = $vmview->config->hardware->numCPU;
xCAT::SvrUtils::sendmsg("CPUs: $cpuCount", $output_handler,$node);
my $memory = $vmview->config->hardware->memoryMB;
xCAT::SvrUtils::sendmsg("Memory: $memory MB", $output_handler,$node);
my %updatehash = ( cpus => $cpuCount, memory=>$memory);
my $devices = $vmview->config->hardware->device;
my $label;
my $size;
my $fileName;
my $device;
if ($tableUpdate and $hyp) {
validate_datastore_prereqs([$node],$hyp); #need datastoremaps to verify names...
}
my %vmstorageurls;
foreach $device (@$devices) {
$label = $device->deviceInfo->label;
if($label =~ /^Hard disk/) {
$label .= " (d".$device->controllerKey.":".$device->unitNumber.")";
$size = $device->capacityInKB / 1024;
$fileName = $device->backing->fileName;
$output_handler->({
node=>{
name=>$node,
data=>{
desc=>$label,
contents=>"$size MB @ $fileName"
}
}
});
#if ($tableUpdate) {
# $fileName =~ /\[([^\]]+)\]/;
# $vmstorageurls{$hyphash{$hyp}->{datastoreurlmap}->{$1}}=1;
#}
} elsif ($label =~ /Network/) {
xCAT::SvrUtils::sendmsg("$label: ".$device->macAddress, $output_handler,$node);
}
}
if ($tableUpdate) {
my $cfgdatastore;
foreach (@{$vmview->layoutEx->file}) {
#TODO, track ALL layoutEx->file....
if ($_->type eq 'config') {
$_->name =~ /\[([^\]]+)\]/;
$cfgdatastore = $hyphash{$hyp}->{datastoreurlmap}->{$1};
last;
}
}
my $cfgkey;
if ($tablecfg{vm}->{$node}->[0]->{cfgstore}) { #check the config file explicitly, ignore the rest
$cfgkey='cfgstore';
} elsif ($tablecfg{vm}->{$node}->[0]->{storage}) { #check the config file explicitly, ignore the rest
$cfgkey='storage';
}
my $configuration = $tablecfg{vm}->{$node}->[0]->{$cfgkey}; #TODO: prune urls that map to no layoutEx->file entries anymore
my $configappend = $configuration;
$configappend =~ s/^[^,=]*//;
$tablecfg{vm}->{$node}->[0]->{$cfgkey} =~ m!nfs://([^/]+)/!;
my $tablecfgserver =$1;
my $cfgserver = inet_aton($tablecfgserver);
if ($cfgserver) {
$cfgserver = inet_ntoa($cfgserver); #get the IP address (TODO: really need to wrap getaddrinfo this handily...
my $cfgurl = $tablecfg{vm}->{$node}->[0]->{$cfgkey};
$cfgurl =~ s/$tablecfgserver/$cfgserver/;
if ($cfgurl ne $cfgdatastore) {
$updatehash{$cfgkey} = $cfgdatastore.$configappend;
}
}
}
if($tableUpdate){
my $vm=xCAT::Table->new('vm',-create=>1);
$vm->setNodeAttribs($node,\%updatehash);
}
}
#changes the memory, number of cpus and device size
#can also add,resize and remove disks
sub chvm {
my %args = @_;
my $node = $args{node};
my $hyp = $args{hyp};
if (not defined $args{vmview}) { #attempt one refresh
$args{vmview} = $vmhash{$node}->{conn}->find_entity_view(view_type => 'VirtualMachine',
properties=>['config.name','runtime.powerState'],
filter=>{name=>$node});
if (not defined $args{vmview}) {
xCAT::SvrUtils::sendmsg([1,"VM does not appear to exist"], $output_handler,$node);
return;
}
}
@ARGV= @{$args{exargs}};
my @deregister;
my @purge;
my @add;
my %resize;
my $cpuCount;
my $memory;
my $vmview = $args{vmview};
require Getopt::Long;
$SIG{__WARN__} = sub {
xCAT::SvrUtils::sendmsg([1,"Could not parse options, ".shift()], $output_handler);
};
my @otherparams;
my $cdrom;
my $eject;
my $rc = GetOptions(
"d=s" => \@deregister,
"p=s" => \@purge,
"a=s" => \@add,
"o=s" => \@otherparams,
"resize=s%" => \%resize,
"optical|cdrom|c=s" => \$cdrom,
"eject" => \$eject,
"cpus=s" => \$cpuCount,
"mem=s" => \$memory
);
$SIG{__WARN__} = 'DEFAULT';
if(@ARGV) {
xCAT::SvrUtils::sendmsg("Invalid arguments: @ARGV", $output_handler);
return;
}
if(!$rc) {
return;
}
#use Data::Dumper;
#xCAT::SvrUtils::sendmsg("dereg = ".Dumper(\@deregister));
#xCAT::SvrUtils::sendmsg("purge = ".Dumper(\@purge));
#xCAT::SvrUtils::sendmsg("add = ".Dumper(\@add));
#xCAT::SvrUtils::sendmsg("resize = ".Dumper(\%resize));
#xCAT::SvrUtils::sendmsg("cpus = $cpuCount");
#xCAT::SvrUtils::sendmsg("mem = ".getUnits($memory,"K",1024));
my %conargs;
if($cpuCount) {
if ($cpuCount =~ /^\+(\d+)/) {
$cpuCount = $vmview->config->hardware->numCPU+$1;
} elsif ($cpuCount =~ /^-(\d+)/) {
$cpuCount = $vmview->config->hardware->numCPU-$1;
}
$conargs{numCPUs} = $cpuCount;
}
if($memory) {
if ($memory =~ /^\+(.+)/) {
$conargs{memoryMB} = $vmview->config->hardware->memoryMB + getUnits($1,"M",1048576);
} elsif ($memory =~ /^-(\d+)/) {
$conargs{memoryMB} = $vmview->config->hardware->memoryMB - getUnits($1,"M",1048576);
} else {
$conargs{memoryMB} = getUnits($memory, "M", 1048576);
}
}
my $disk;
my $devices = $vmview->config->hardware->device;
my $label;
my $device;
my $cmdLabel;
my $newSize;
my @devChanges;
if(@deregister) {
for $disk (@deregister) {
$device = getDiskByLabel($disk, $devices);
unless($device) {
xCAT::SvrUtils::sendmsg([1,"Disk: $disk does not exist"], $output_handler,$node);
return;
}
#xCAT::SvrUtils::sendmsg(Dumper($device));
push @devChanges, VirtualDeviceConfigSpec->new(
device => $device,
operation => VirtualDeviceConfigSpecOperation->new('remove'));
}
}
if(@purge) {
for $disk (@purge) {
$device = getDiskByLabel($disk, $devices);
unless($device) {
xCAT::SvrUtils::sendmsg([1,"Disk: $disk does not exist"], $output_handler,$node);
return;
}
#xCAT::SvrUtils::sendmsg(Dumper($device));
push @devChanges, VirtualDeviceConfigSpec->new(
device => $device,
operation => VirtualDeviceConfigSpecOperation->new('remove'),
fileOperation => VirtualDeviceConfigSpecFileOperation->new('destroy'));
}
}
if(@add) {
my $addSizes = join(',',@add);
my $scsiCont;
my $scsiUnit;
my $ideCont;
my $ideUnit;
my $label;
my $idefull=0;
my $scsifull=0;
foreach $device (@$devices) {
$label = $device->deviceInfo->label;
if($label =~ /^SCSI controller/) {
my $tmpu=getAvailUnit($device->{key},$devices,maxnum=>15);
if ($tmpu > 0) {
$scsiCont = $device;
$scsiUnit=$tmpu;
} else {
$scsifull=1;
}
#ignore scsiControllers that are full, problem still remains if trying to add across two controllers in one go
}
if($label =~ /^IDE/ and not $ideCont) {
my $tmpu=getAvailUnit($device->{key},$devices,maxnum=>1);
if ($tmpu >= 0) {
$ideCont = $device;
$ideUnit = $tmpu;
} elsif ($device->{key} == 201) {
$idefull=1;
}
}
}
unless ($hyphash{$hyp}->{datastoremap}) { validate_datastore_prereqs([],$hyp); }
push @devChanges, create_storage_devs($node,$hyphash{$hyp}->{datastoremap},$addSizes,$scsiCont,$scsiUnit,$ideCont,$ideUnit,$devices,idefull=>$idefull,scsifull=>$scsifull);
}
if ($cdrom or $eject) {
my $opticalbackingif;
my $opticalconnectable;
if ($cdrom) {
my $storageurl;
if ($cdrom =~ m!://!) {
$storageurl=$cdrom;
$storageurl =~ s!/[^/]*\z!!;
unless (validate_datastore_prereqs([],$hyp,{$storageurl=>[$node]})) {
xCAT::SvrUtils::sendmsg([1,"Unable to find/mount datastore holding $cdrom"], $output_handler,$node);
return;
}
$cdrom =~ s!.*/!!;
} else {
$storageurl = $tablecfg{vm}->{$node}->[0]->{storage};
$storageurl =~ s/=.*//;
$storageurl =~ s/.*,//;
$storageurl =~ s/\/\z//;
}
$opticalbackingif = VirtualCdromIsoBackingInfo->new( fileName => "[".$hyphash{$hyp}->{datastoremap}->{$storageurl}."] $cdrom");
$opticalconnectable = VirtualDeviceConnectInfo->new(startConnected=>1,allowGuestControl=>1,connected=>1);
} elsif ($eject) {
$opticalbackingif=VirtualCdromRemoteAtapiBackingInfo->new(deviceName=>"");
$opticalconnectable=VirtualDeviceConnectInfo->new(startConnected=>0,allowGuestControl=>1,connected=>0);
}
my $oldcd;
foreach my $dev (@$devices) {
if ($dev->deviceInfo->label eq "CD/DVD drive 1") {
$oldcd=$dev;
last;
}
}
unless ($oldcd) {
if ($cdrom) {
xCAT::SvrUtils::sendmsg([1,"Unable to find Optical drive in VM to insert ISO image"], $output_handler,$node);
} else {
xCAT::SvrUtils::sendmsg([1,"Unable to find Optical drive in VM to perform eject"], $output_handler,$node);
}
return;
}
my $newDevice = VirtualCdrom->new(backing => $opticalbackingif,
key=>$oldcd->key,
controllerKey=>201,
unitNumber=>0,
connectable=>$opticalconnectable,
);
push @devChanges, VirtualDeviceConfigSpec->new(
device => $newDevice,
operation => VirtualDeviceConfigSpecOperation->new('edit'));
}
if(%resize) {
while( my ($key, $value) = each(%resize) ) {
my @drives = split(/,/, $key);
for my $device ( @drives ) {
my $disk = $device;
$device = getDiskByLabel($disk, $devices);
unless($device) {
xCAT::SvrUtils::sendmsg([1,"Disk: $disk does not exist"], $output_handler,$node);
return;
}
if ($value =~ /^\+(.+)/) {
$value = $device->capacityInKB + getUnits($1,"G",1024);
} else {
$value = getUnits($value, "G", 1024);
}
my $newDevice = VirtualDisk->new(deviceInfo => $device->deviceInfo,
key => $device->key,
controllerKey => $device->controllerKey,
unitNumber => $device->unitNumber,
deviceInfo => $device->deviceInfo,
backing => $device->backing,
capacityInKB => $value);
push @devChanges, VirtualDeviceConfigSpec->new(
device => $newDevice,
operation => VirtualDeviceConfigSpecOperation->new('edit'));
}
}
}
if(@devChanges) {
$conargs{deviceChange} = \@devChanges;
}
if (@otherparams) {
my $key;
my $value;
my @optionvals;
foreach (@otherparams) {
($key,$value) = split /=/;
unless ($key) {
xCAT::SvrUtils::sendmsg([1,"Invalid format for other parameter specification"], $output_handler,$node);
return;
}
if ($value) {
push @optionvals,OptionValue->new(key=>$key,value=>$value);
} else {
push @optionvals,OptionValue->new(key=>$key); #the api doc says this is *supposed* to delete a key, don't think it works though, e.g. http://communities.vmware.com/message/1602644
}
}
$conargs{extraConfig} = \@optionvals;
}
my $reconfigspec = VirtualMachineConfigSpec->new(%conargs);
#xCAT::SvrUtils::sendmsg("reconfigspec = ".Dumper($reconfigspec));
my $task = $vmview->ReconfigVM_Task(spec=>$reconfigspec);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&chvm_task_callback;
$running_tasks{$task}->{hyp} = $hyp;
$running_tasks{$task}->{data} = { node => $node, successtext => "node successfully changed",cpus=>$cpuCount,mem=>$memory };
}
sub getUsedUnits {
my $contKey = shift;
my $devices = shift;
my %usedids;
$usedids{7}=1;
$usedids{'7'}=1; #TODO: figure out which of these is redundant, the string or the number variant
for my $device (@$devices) {
if($device->{controllerKey} eq $contKey) {
$usedids{$device->{unitNumber}}=1;
}
}
return \%usedids;
}
sub getAvailUnit {
my $contKey = shift;
my $devices = shift;
my %args = @_;
my $maxunit=-1;
if (defined $args{maxnum}) {
$maxunit=$args{maxnum};
}
my %usedids;
$usedids{7}=1;
$usedids{'7'}=1; #TODO: figure out which of these is redundant, the string or the number variant
for my $device (@$devices) {
if($device->{controllerKey} eq $contKey) {
$usedids{$device->{unitNumber}}=1;
}
}
my $highestUnit=0;
while ($usedids{$highestUnit}) {
if ($highestUnit == $maxunit) {
return -1;
}
$highestUnit++;
}
return $highestUnit;
}
#given a device list from a vm and a label for a hard disk, returns the device object
sub getDiskByLabel {
my $cmdLabel = shift;
my $devices = shift;
my $device;
my $label;
$cmdLabel = commandLabel($cmdLabel);
foreach $device (@$devices) {
$label = $device->deviceInfo->label;
if($cmdLabel eq $label) {
return $device;
} elsif (($label =~ /^Hard disk/) and ($cmdLabel =~ /^d(.*)/)) {
my $desc = $1;
if ($desc =~ /(.*):(.*)/) {#specific
my $controller=$1;
my $unit=$2;
if ($device->unitNumber == $unit and $device->controllerKey == $controller) {
return $device;
}
} elsif ($desc =~ /\d+/ and $device->unitNumber == $desc) { #not specific
return $device;
}
}
}
return undef;
}
#takes a label for a hard disk and prepends "Hard disk " if it's not there already
sub commandLabel {
my $label = shift;
if(($label =~ /^Hard disk/) or ($label =~ /^d\d+/)) {
return $label;
}
return "Hard disk ".$label;
}
#this function will check pending task status
sub process_tasks {
foreach (keys %running_tasks) {
my $curcon;
if (defined $running_tasks{$_}->{conn}) {
$curcon = $running_tasks{$_}->{conn};
} elsif ($running_tasks{$_}->{hyp}) {
$curcon = $hyphash{$running_tasks{$_}->{hyp}}->{conn};
} elsif ($running_tasks{$_}->{vm}) {
$curcon = $vmhash{$running_tasks{$_}->{vm}}->{conn};
} elsif ($running_tasks{$_}->{cluster}) {
$curcon = $clusterhash{$running_tasks{$_}->{cluster}}->{conn};
} else {
use Carp qw/confess/;
confess "This stack trace indicates a cluster unfriendly path";
}
my $curt = $curcon->get_view(mo_ref=>$running_tasks{$_}->{task});
my $state = $curt->info->state->val;
unless ($state eq 'running' or $state eq 'queued') {
$running_tasks{$_}->{callback}->($curt,$running_tasks{$_}->{data});
delete $running_tasks{$_};
}
if ($state eq 'running' and not $running_tasks{$_}->{questionasked}) { # and $curt->info->progress == 95) { #This is unfortunate, there should be a 'state' to indicate a question is blocking
#however there isn't, so if we see something running at 95%, we just manually see if a question blocked the rest
my $vm;
$@="";
eval {
$vm = $curcon->get_view(mo_ref=>$curt->info->entity);
};
if ($@) { $vm = 0; }
if ($vm and $vm->{summary} and $vm->summary->{runtime} and $vm->summary->runtime->{question} and $vm->summary->runtime->question) {
$running_tasks{$_}->{questionasked}=1;
$running_tasks{$_}->{callback}->($curt,$running_tasks{$_}->{data},$vm->summary->runtime->question,$vm);
}
}
}
}
#this function is a barrier to ensure prerequisites are met
sub wait_for_tasks {
while (scalar keys %running_tasks) {
process_tasks;
sleep (1); #We'll check back in every second. Unfortunately, we have to poll since we are in web service land
}
}
sub connecthost_callback {
my $task = shift;
my $args = shift;
my $hv = $args->{hostview};
my $state = $task->info->state->val;
if ($state eq "success") {
$hypready{$args->{hypname}}=1; #declare readiness
enable_vmotion(hypname=>$args->{hypname},hostview=>$args->{hostview},conn=>$args->{conn});
$vcenterhash{$args->{vcenter}}->{goodhyps}->{$args->{hypname}} = 1;
if (defined $args->{depfun}) { #If a function is waiting for the host connect to go valid, call it
$args->{depfun}->($args->{depargs});
}
return;
}
my $thumbprint;
eval {
$thumbprint = $task->{info}->error->fault->thumbprint;
};
if ($thumbprint) { #was an unknown certificate error, retry and accept the unknown certificate
$args->{connspec}->{sslThumbprint}=$task->info->error->fault->thumbprint;
my $task;
if (defined $args->{hostview}) {#It was a reconnect request
$task = $hv->ReconnectHost_Task(cnxSpec=>$args->{connspec});
} elsif (defined $args->{foldview}) {#was an add host request
$task = $args->{foldview}->AddStandaloneHost_Task(spec=>$args->{connspec},addConnected=>1);
} elsif (defined $args->{cluster}) {#was an add host to cluster request
$task = $args->{cluster}->AddHost_Task(spec=>$args->{connspec},asConnected=>1);
}
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&connecthost_callback;
$running_tasks{$task}->{conn} = $args->{conn};
$running_tasks{$task}->{data} = $args; #{ conn_spec=>$connspec,hostview=>$hv,hypname=>$args->{hypname},vcenter=>$args->{vcenter} };
} elsif ($state eq 'error') {
my $error = $task->info->error->localizedMessage;
if (defined ($task->info->error->fault->faultMessage)) { #Only in 4.0, support of 3.5 must be careful?
foreach(@{$task->info->error->fault->faultMessage}) {
$error.=$_->message;
}
}
xCAT::SvrUtils::sendmsg([1,$error], $output_handler); #,$node);
$hypready{$args->{hypname}} = -1; #Impossible for this hypervisor to ever be ready
$vcenterhash{$args->{vcenter}}->{badhyps}->{$args->{hypname}} = 1;
}
}
sub delhost_callback { #only called in rmhypervisor -f case during validate vcenter phase
my $task = shift;
my $args = shift;
my $hv = $args->{hostview};
my $state = $task->info->state->val;
if ($state eq "success") {
xCAT::SvrUtils::sendmsg("removed", $output_handler,$args->{hypname});
$hypready{$args->{hypname}} = -1; #Impossible for this hypervisor to ever be ready
$vcenterhash{$args->{vcenter}}->{badhyps}->{$args->{hypname}} = 1;
} elsif ($state eq 'error') {
my $error = $task->info->error->localizedMessage;
if (defined ($task->info->error->fault->faultMessage)) { #Only in 4.0, support of 3.5 must be careful?
foreach(@{$task->info->error->fault->faultMessage}) {
$error.=$_->message;
}
}
xCAT::SvrUtils::sendmsg([1,$error], $output_handler); #,$node);
$hypready{$args->{hypname}} = -1; #Impossible for this hypervisor to ever be ready
$vcenterhash{$args->{vcenter}}->{badhyps}->{$args->{hypname}} = 1;
}
}
sub get_clusterview {
my %args = @_;
my $clustname = $args{clustname};
my %subargs = (
view_type=>'ClusterComputeResource',
);
if ($args{properties}) {
$subargs{properties}=$args{properties};
}
$subargs{filter}={name=>$clustname};
my $view = $args{conn}->find_entity_view(%subargs);
return $view;
#foreach (@{$args{conn}->find_entity_views(%subargs)}) {
# if ($_->name eq "$clustname") {
# return $_;
# last;
# }
#}
}
sub get_hostview {
my %args = @_;
my $host = $args{hypname};
my %subargs = (
view_type=>'HostSystem',
);
if ($args{properties}) {
$subargs{properties}=$args{properties};
}
my @addrs = gethostbyname($host);
my $ip;
my $name;
my $aliases;
if ($addrs[4]) {
$ip=inet_ntoa($addrs[4]);
($name, $aliases) = gethostbyaddr($addrs[4],AF_INET); #TODO: IPv6
} else {
($ip,$name,$aliases) = ($host,$host,"");
}
my @matchvalues = ($host,$ip,$name);
foreach (split /\s+/,$aliases) {
push @matchvalues,$_;
}
my $view;
$subargs{filter}={'name' => qr/$host(?:\.|\z)/};
$view = $args{conn}->find_entity_view(%subargs);
if ($view) { return $view; }
foreach (@matchvalues) {
$subargs{filter}={'name' => qr/$_(?:\.|\z)/};
$view = $args{conn}->find_entity_view(%subargs);
if ($view) { return $view; }
}
$subargs{filter}={'name' => qr/localhost(?:\.|\z)/};
$view = $args{conn}->find_entity_view(%subargs);
if ($view) { return $view; }
return undef; #rest of function should be obsoleted, going to run with that assumption for 2.5 at least
# $subargs{filter}={'name' =~ qr/.*/};
# foreach (@{$args{conn}->find_entity_views(%subargs)}) {
# my $view = $_;
# if ($_->name =~ /$host(?:\.|\z)/ or $_->name =~ /localhost(?:\.|\z)/ or grep { $view->name =~ /$_(?:\.|\z)/ } @matchvalues) {
# return $view;
# last;
# }
# }
}
sub enable_vmotion {
#TODO: vmware 3.x semantics too? this is 4.0...
my %args = @_;
unless ($args{hostview}) {
$args{hostview} = get_hostview(conn=>$args{conn},hypname=>$args{hypname},properties=>['configManager','name']);
}
my $nicmgr=$args{conn}->get_view(mo_ref=>$args{hostview}->configManager->virtualNicManager);
my $qnc = $nicmgr->QueryNetConfig(nicType=>"vmotion");
if ($qnc->{selectedVnic}) {
return 1;
} else {
my $vniccount=scalar @{$qnc->candidateVnic};
if ($vniccount==1 or ($vniccount==2 and $qnc->candidateVnic->[1]->spec->ip->ipAddress =~ /^169.254/)) { #There is only one possible path, use it
$nicmgr->SelectVnicForNicType(nicType=>"vmotion",device=>$qnc->candidateVnic->[0]->device);
return 1;
} else {
xCAT::SvrUtils::sendmsg([1,"TODO: use configuration to pick the nic ".$args{hypname}], $output_handler);
}
return 0;
}
}
sub mkvm_callback {
my $task = shift;
my $args = shift;
my $node = $args->{node};
my $hyp = $args->{hyp};
if ($task->info->state->val eq 'error') {
my $error = $task->info->error->localizedMessage;
xCAT::SvrUtils::sendmsg([1,$error], $output_handler,$node);
}
}
sub relay_vmware_err {
my $task = shift;
my $extratext = shift;
my @nodes = @_;
my $error = $task->info->error->localizedMessage;
if (defined ($task->info->error->fault->faultMessage)) { #Only in 4.0, support of 3.5 must be careful?
foreach(@{$task->info->error->fault->faultMessage}) {
$error.=$_->message;
}
}
if (@nodes) {
foreach (@nodes) {
xCAT::SvrUtils::sendmsg([1,$extratext.$error], $output_handler,$_);
}
}else {
xCAT::SvrUtils::sendmsg([1,$extratext.$error], $output_handler);
}
}
sub relocate_callback {
my $task = shift;
my $parms = shift;
my $state = $task->info->state->val;
if ($state eq 'success') {
my $vmtab = xCAT::Table->new('vm'); #TODO: update vm.storage?
my $prevloc = $tablecfg{vm}->{$parms->{node}}->[0]->{storage};
my $model;
($prevloc,$model) = split /=/,$prevloc;
my $target = $parms->{target};
if ($model) {
$target.="=$model";
}
$vmtab->setNodeAttribs($parms->{node},{storage=>$target});
xCAT::SvrUtils::sendmsg(":relocated to to ".$parms->{target}, $output_handler,$parms->{node});
} else {
relay_vmware_err($task,"Relocating to ".$parms->{target}." ",$parms->{node});
}
}
sub migrate_ok { #look like a successful migrate, callback for registering a vm
my %args = @_;
my $vmtab = xCAT::Table->new('vm');
$vmtab->setNodeAttribs($args{nodes}->[0],{host=>$args{target}});
xCAT::SvrUtils::sendmsg("migrated to ".$args{target}, $output_handler,$args{nodes}->[0]);
}
sub migrate_callback {
my $task = shift;
my $parms = shift;
my $state = $task->info->state->val;
if (not $parms->{skiptodeadsource} and $state eq 'success') {
my $vmtab = xCAT::Table->new('vm');
$vmtab->setNodeAttribs($parms->{node},{host=>$parms->{target}});
xCAT::SvrUtils::sendmsg("migrated to ".$parms->{target}, $output_handler,$parms->{node});
} elsif($parms->{offline}) { #try a forceful RegisterVM instead
my $target = $parms->{target};
my $hostview = $hyphash{$target}->{conn}->find_entity_view(view_type=>'VirtualMachine',properties=>['config.name'],filter=>{name=>$parms->{node}});
if ($hostview) { #this means vcenter still has it in inventory, but on a dead node...
#unfortunately, vcenter won't give up the old one until we zap the dead hypervisor
#also unfortunately, it doesn't make it easy to find said hypervisor..
$hostview = $hyphash{$parms->{src}}->{conn}->get_view(mo_ref=>$hyphash{$parms->{src}}->{deletionref});
$task = $hostview->Destroy_Task();
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&migrate_callback;
$running_tasks{$task}->{conn} = $hyphash{$target}->{vcenter}->{conn};
$running_tasks{$task}->{data} = { offline=>1, src=>$parms->{src}, node=>$parms->{node}, target=>$target, skiptodeadsource=>1 };
} else { #it is completely gone, attempt a register_vm strategy
register_vm($target,$parms->{node},undef,\&migrate_ok,{ nodes => [$parms->{node}], target=>$target, },"failonerror");
}
} else {
relay_vmware_err($task,"Migrating to ".$parms->{target}." ",$parms->{node});
}
}
sub poweron_task_callback {
my $task = shift;
my $parms = shift;
my $q = shift; #question if blocked
my $vm = shift; #path to answer questions if asked
my $state = $task->info->state->val;
my $node = $parms->{node};
my $intent = $parms->{successtext};
if ($state eq 'success') {
xCAT::SvrUtils::sendmsg($intent, $output_handler,$node);
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
} elsif ($q and $q->text =~ /^msg.uuid.altered:/ and ($q->choice->choiceInfo->[0]->summary eq 'Cancel' and ($q->choice->choiceInfo->[0]->key eq '0'))) { #make sure it is what is what we have seen it to be
if ($parms->{forceon} and $q->choice->choiceInfo->[1]->summary =~ /I (_)?moved it/ and $q->choice->choiceInfo->[1]->key eq '1') { #answer the question as 'moved'
$vm->AnswerVM(questionId=>$q->id,answerChoice=>'1');
} else {
$vm->AnswerVM(questionId=>$q->id,answerChoice=>'0');
xCAT::SvrUtils::sendmsg([1,"Failure powering on VM, it mismatched against the hypervisor. If positive VM is not running on another hypervisor, use -f to force VM on"], $output_handler,$node);
}
} elsif ($q) {
if ($q->choice->choiceInfo->[0]->summary eq 'Cancel') {
xCAT::SvrUtils::sendmsg([1,":Cancelling due to unexpected question executing task: ".$q->text], $output_handler,$node);
} else {
xCAT::SvrUtils::sendmsg([1,":Task hang due to unexpected question executing task, need to use VMware tools to clean up the mess for now: ".$q->text], $output_handler,$node);
}
}
}
sub chvm_task_callback {
my $task = shift;
my $parms = shift;
my $state = $task->info->state->val;
my $node = $parms->{node};
my $intent = $parms->{successtext};
if ($state eq 'success') {
my $updatehash;
if ($parms->{cpus} and $tablecfg{vm}->{$node}->[0]->{cpus}) { #need to update
$updatehash->{cpus}=$parms->{cpus};
}
if ($parms->{mem} and $tablecfg{vm}->{$node}->[0]->{memory}) { #need to update
$updatehash->{memory}=$parms->{mem};
}
if ($updatehash) {
my $vmtab = xCAT::Table->new('vm',-create=>1);
$vmtab->setNodeAttribs($node,$updatehash);
}
xCAT::SvrUtils::sendmsg($intent, $output_handler,$node);
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
}
}
sub generic_task_callback {
my $task = shift;
my $parms = shift;
my $state = $task->info->state->val;
my $node = $parms->{node};
my $intent = $parms->{successtext};
if ($state eq 'success') {
xCAT::SvrUtils::sendmsg($intent, $output_handler,$node);
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
}
}
sub migrate {
my %args = @_;
my @nodes = @{$args{nodes}};
my $hyp = $args{hyp};
my $vcenter = $hyphash{$hyp}->{vcenter}->{name};
my $datastoredest;
my $offline;
@ARGV=@{$args{exargs}};
unless (GetOptions(
's=s' => \$datastoredest,
'f' => \$offline,
)) {
xCAT::SvrUtils::sendmsg([1,"Error parsing arguments"], $output_handler);
return;
}
my $target=$hyp; #case for storage migration
if ($datastoredest) { $datastoredest =~ s/=.*//; }#remove =scsi and similar if specified
if ($datastoredest and scalar @ARGV) {
xCAT::SvrUtils::sendmsg([1,"Unable to mix storage migration and processing of arguments ".join(' ',@ARGV)], $output_handler);
return;
} elsif (@ARGV) {
$target=shift @ARGV;
if (@ARGV) {
xCAT::SvrUtils::sendmsg([1,"Unrecognized arguments ".join(' ',@ARGV)], $output_handler);
return;
}
} elsif ($datastoredest) { #storage migration only
unless (validate_datastore_prereqs([],$hyp,{$datastoredest=>\@nodes})) {
xCAT::SvrUtils::sendmsg([1,"Unable to find/mount target datastore $datastoredest"], $output_handler);
return;
}
foreach (@nodes) {
my $hostview = $hyphash{$hyp}->{conn}->find_entity_view(view_type=>'VirtualMachine',properties=>['config.name'],filter=>{name=>$_});
my $relocatspec = VirtualMachineRelocateSpec->new(
datastore=>$hyphash{$hyp}->{datastorerefmap}->{$datastoredest},
);
my $task = $hostview->RelocateVM_Task(spec=>$relocatspec);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&relocate_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{data} = { node => $_, target=>$datastoredest };
process_tasks; #check for tasks needing followup actions before the task is forgotten (VMWare's memory is fairly short at times
}
return;
}
if ((not $offline and $vcenterhash{$vcenter}->{badhyps}->{$hyp}) or $vcenterhash{$vcenter}->{badhyps}->{$target}) {
xCAT::SvrUtils::sendmsg([1,"Unable to migrate ".join(',',@nodes)." to $target due to inability to validate vCenter connectivity"], $output_handler);
return;
}
if (($offline or $vcenterhash{$vcenter}->{goodhyps}->{$hyp}) and $vcenterhash{$vcenter}->{goodhyps}->{$target}) {
unless (validate_datastore_prereqs(\@nodes,$target)) {
xCAT::SvrUtils::sendmsg([1,"Unable to verify storage state on target system"], $output_handler);
return;
}
unless (validate_network_prereqs(\@nodes,$target)) {
xCAT::SvrUtils::sendmsg([1,"Unable to verify target network state"], $output_handler);
return;
}
my $dstview = get_hostview(conn=>$hyphash{$target}->{conn},hypname=>$target,properties=>['name','parent']);
unless ($hyphash{$target}->{pool}) {
$hyphash{$target}->{pool} = $hyphash{$target}->{conn}->get_view(mo_ref=>$dstview->parent,properties=>['resourcePool'])->resourcePool;
}
foreach (@nodes) {
process_tasks; #check for tasks needing followup actions before the task is forgotten (VMWare's memory is fairly short at times
my $srcview = $hyphash{$target}->{conn}->find_entity_view(view_type=>'VirtualMachine',properties=>['config.name'],filter=>{name=>$_});
if ($offline and not $srcview) { #we have a request to resurrect the dead..
register_vm($target,$_,undef,\&migrate_ok,{ nodes => [$_], exargs => $args{exargs}, target=>$target, hyp => $args{hyp}, offline => $offline },"failonerror");
return;
} elsif (not $srcview) {
$srcview = $hyphash{$hyp}->{conn}->find_entity_view(view_type=>'VirtualMachine',properties=>['config.name'],filter=>{name=>$_});
}
unless ($srcview) {
xCAT::SvrUtils::sendmsg([1,"Unable to locate node in vCenter"], $output_handler,$_);
next;
}
my $task = $srcview->MigrateVM_Task(
host=>$dstview,
pool=>$hyphash{$target}->{pool},
priority=>VirtualMachineMovePriority->new('highPriority'));
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&migrate_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{data} = { node => $_, src=>$hyp, target=>$target, offline => $offline };
}
} else {
#xCAT::SvrUtils::sendmsg("Waiting for BOTH to be 'good'");
return; #One of them is still 'pending'
}
}
sub reconfig_callback {
my $task = shift;
my $args = shift;
#$args->{reconfig_args}->{vmview}->update_view_data();
delete $args->{reconfig_args}->{vmview}; #Force a reload of the view, update_view_data seems to not work as advertised..
$args->{reconfig_fun}->(%{$args->{reconfig_args}});
}
sub repower { #Called to try power again after power down for reconfig
my $task = shift;
my $args = shift;
my $powargs=$args->{power_args};
$powargs->{pretendop}=1;
#$powargs->{vmview}->update_view_data();
delete $powargs->{vmview}; #Force a reload of the view, update_view_data seems to not work as advertised..
power(%$powargs);
}
sub retry_rmvm {
my $task = shift;
my $args = shift;
my $node = $args->{node};
my $state = $task->info->state->val;
if ($state eq "success") {
#$Data::Dumper::Maxdepth=2;
delete $args->{args}->{vmview};
rmvm(%{$args->{args}});
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
}
}
sub rmvm {
my %args = @_;
my $node = $args{node};
my $hyp = $args{hyp};
if (not defined $args{vmview}) { #attempt one refresh
$args{vmview} = $vmhash{$node}->{conn}->find_entity_view(view_type => 'VirtualMachine',properties=>['config.name','runtime.powerState'],filter=>{name=>$node});
if (not defined $args{vmview}) {
xCAT::SvrUtils::sendmsg([1,"VM does not appear to exist"], $output_handler,$node);
return;
}
}
@ARGV= @{$args{exargs}};
require Getopt::Long;
my $forceremove;
my $purge;
GetOptions(
'f' => \$forceremove,
'p' => \$purge,
);
my $task;
unless ($args{vmview}->{'runtime.powerState'}->val eq 'poweredOff') {
if ($forceremove) {
$task = $args{vmview}->PowerOffVM_Task();
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&retry_rmvm,
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{vm} = $node;
$running_tasks{$task}->{data} = { node => $node, args=>\%args };
return;
} else {
xCAT::SvrUtils::sendmsg([1,"Cannot rmvm active guest (use -f argument to force)"], $output_handler,$node);
return;
}
}
if ($purge) {
$task = $args{vmview}->Destroy_Task();
$running_tasks{$task}->{data} = { node => $node, successtext => 'purged' };
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp}; #$hyp_conns->{$hyp};
$running_tasks{$task}->{vm} = $node;
} else {
$task = $args{vmview}->UnregisterVM();
}
}
sub getreconfigspec {
my %args = @_;
my $node = $args{node};
my $vmview = $args{view};
my $currid=$args{view}->{'config.guestId'};
my $rightid=getguestid($node);
my %conargs;
my $reconfigneeded=0;
if ($currid ne $rightid) {
$reconfigneeded=1;
$conargs{guestId}=$rightid;
}
my $newmem;
if ($tablecfg{vm}->{$node}->[0]->{memory} and $newmem = getUnits($tablecfg{vm}->{$node}->[0]->{memory},"M",1048576)) {
my $currmem = $vmview->{'config.hardware.memoryMB'};
if ($newmem ne $currmem) {
$conargs{memoryMB} = $newmem;
$reconfigneeded=1;
}
}
my $newcpus;
if ($tablecfg{vm}->{$node}->[0]->{cpus} and $newcpus = $tablecfg{vm}->{$node}->[0]->{cpus}) {
my $currncpu = $vmview->{'config.hardware.numCPU'};
if ($newcpus ne $currncpu) {
$conargs{numCPUs} = $newcpus;
$reconfigneeded=1;
}
}
if ($reconfigneeded) {
return VirtualMachineConfigSpec->new(%conargs);
} else {
return 0;
}
}
#This routine takes a single node, managing vmv instance, and task tracking hash to submit a power on request
sub power {
my %args = @_;
my $node = $args{node};
my $hyp = $args{hyp};
my $pretendop = $args{pretendop}; #to pretend a system was on for reset or boot when we have to turn it off internally for reconfig
if (not defined $args{vmview}) { #attempt one refresh
$args{vmview} = $vmhash{$node}->{conn}->find_entity_view(view_type => 'VirtualMachine',properties=>['config.name','config.guestId','config.hardware.memoryMB','config.hardware.numCPU','runtime.powerState'],filter=>{name=>$node});
#vmview not existing now is not an issue, this function
#is designed to handle that and correct if reasonably possible
#comes into play particularly in a stateless context
}
@ARGV = @{$args{exargs}}; #for getoptions;
my $forceon;
require Getopt::Long;
GetOptions(
'force|f' => \$forceon,
);
my $subcmd = $ARGV[0];
my $intent="";
my $task;
my $currstat;
if ($args{vmview}) {
$currstat = $args{vmview}->{'runtime.powerState'}->val;
if (grep /$subcmd/,qw/on reset boot/) {
my $reconfigspec;
if ($reconfigreset and ($reconfigspec = getreconfigspec(node=>$node,view=>$args{vmview}))) {
if ($currstat eq 'poweredOff') {
#xCAT::SvrUtils::sendmsg("Correcting guestId because $currid and $rightid are not the same...");#DEBUG
my $task = $args{vmview}->ReconfigVM_Task(spec=>$reconfigspec);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&reconfig_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{vm} = $node;
$running_tasks{$task}->{data} = { node => $node, reconfig_fun=>\&power, reconfig_args=>\%args };
return;
} elsif (grep /$subcmd/,qw/reset boot/) { #going to have to do a 'cycle' and present it up normally..
#xCAT::SvrUtils::sendmsg("DEBUG: forcing a cycle");
$task = $args{vmview}->PowerOffVM_Task();
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&repower;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{vm} = $node;
$running_tasks{$task}->{data} = { node => $node, power_args=>\%args};
return; #we have to wait
}
#TODO: fixit
#xCAT::SvrUtils::sendmsg("I see vm has $currid and I want it to be $rightid");
}
}
} else {
$currstat = 'off';
}
if ($currstat eq 'poweredOff') {
$currstat = 'off';
} elsif ($currstat eq 'poweredOn') {
$currstat = 'on';
} elsif ($currstat eq 'suspended') {
$currstat = 'suspend';
}
if ($subcmd =~ /^stat/) {
xCAT::SvrUtils::sendmsg($currstat, $output_handler,$node);
return;
}
if ($subcmd =~ /boot/) {
$intent = "$currstat ";
if ($currstat eq 'on' or $args{pretendop}) {
$intent = "on ";
$subcmd = 'reset';
} else {
$subcmd = 'on';
}
}
if ($subcmd =~ /on/) {
if ($currstat eq 'off' or $currstat eq 'suspend') {
if (not $args{vmview}) { #We are asking to turn on a system the hypervisor
#doesn't know, attempt to register it first
register_vm($hyp,$node,undef,\&power,\%args);
return; #We'll pick it up on the retry if it gets registered
}
eval {
if ($hyp) {
$task = $args{vmview}->PowerOnVM_Task(host=>$hyphash{$hyp}->{hostview});
} else {
$task = $args{vmview}->PowerOnVM_Task(); #DRS may have it's way with me
}
};
if ($@) {
xCAT::SvrUtils::sendmsg([1,":".$@], $output_handler,$node);
return;
}
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&poweron_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp}; #$hyp_conns->{$hyp};
$running_tasks{$task}->{vm} = $node;
$running_tasks{$task}->{data} = { node => $node, successtext => $intent.'on', forceon=>$forceon };
} else {
xCAT::SvrUtils::sendmsg($currstat, $output_handler,$node);
}
} elsif ($subcmd =~ /softoff/) {
if ($currstat eq 'on') {
$args{vmview}->ShutdownGuest();
xCAT::SvrUtils::sendmsg("softoff", $output_handler,$node);
} else {
xCAT::SvrUtils::sendmsg($currstat, $output_handler,$node);
}
} elsif ($subcmd =~ /off/) {
if ($currstat eq 'on') {
$task = $args{vmview}->PowerOffVM_Task();
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{vm} = $node;
$running_tasks{$task}->{data} = { node => $node, successtext => 'off' };
} else {
xCAT::SvrUtils::sendmsg($currstat, $output_handler,$node);
}
} elsif ($subcmd =~ /suspend/) {
if ($currstat eq 'on') {
$task = $args{vmview}->SuspendVM_Task();
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{vm} = $node;
$running_tasks{$task}->{data} = { node => $node, successtext => 'suspend' };
} else {
xCAT::SvrUtils::sendmsg("off", $output_handler,$node);
}
} elsif ($subcmd =~ /reset/) {
if ($currstat eq 'on') {
$task = $args{vmview}->ResetVM_Task();
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{vm} = $node;
$running_tasks{$task}->{data} = { node => $node, successtext => $intent.'reset' };
} elsif ($args{pretendop}) { #It is off, but pretend it was on
eval {
if ($hyp) {
$task = $args{vmview}->PowerOnVM_Task(host=>$hyphash{$hyp}->{hostview});
} else {
$task = $args{vmview}->PowerOnVM_Task(); #allow DRS
}
};
if ($@) {
xCAT::SvrUtils::sendmsg([1,":".$@], $output_handler,$node);
return;
}
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{vm} = $node;
$running_tasks{$task}->{data} = { node => $node, successtext => $intent.'reset' };
} else {
xCAT::SvrUtils::sendmsg($currstat, $output_handler,$node);
}
}
}
sub generic_vm_operation { #The general form of firing per-vm requests to ESX hypervisor
my $properties = shift; #The relevant properties to the general task, MUST INCLUDE config.name
my $function = shift; #The function to actually run against the right VM view
my @exargs = @_; #Store the rest to pass on
my $hyp;
my $vmviews;
my %vcviews; #views populated once per vcenter server for improved performance
my $node;
foreach $hyp (keys %hyphash) {
if ($viavcenterbyhyp->{$hyp}) {
foreach $node (keys %{$hyphash{$hyp}->{nodes}}){
$vcenterhash{$hyphash{$hyp}->{vcenter}->{name}}->{vms}->{$node}=1;
}
}
}
foreach (keys %limbonodes) {
$vcenterhash{$limbonodes{$_}}->{vms}->{$_}=1;
}
my $cluster;
foreach $cluster (keys %clusterhash) {
foreach $node (keys %{$clusterhash{$cluster}->{nodes}}) {
$vcenterhash{$clusterhash{$cluster}->{vcenter}->{name}}->{vms}->{$node}=1;
}
}
my $currentvcenter;
my %foundlimbo;
foreach $currentvcenter (keys %vcenterhash) {
#retrieve all vm views in one gulp
my $vmsearchstring = join(")|(",keys %{$vcenterhash{$currentvcenter}->{vms}});
$vmsearchstring = '^(('.$vmsearchstring.'))(\z|\.)';
my $regex = qr/$vmsearchstring/;
$vcviews{$currentvcenter} = $vcenterhash{$currentvcenter}->{conn}->find_entity_views(view_type => 'VirtualMachine',properties=>$properties,filter=>{'config.name'=>$regex});
foreach (@{$vcviews{$currentvcenter}}) {
my $node = $_->{'config.name'};
unless (defined $tablecfg{vm}->{$node}) {
$node =~ s/\..*//; #try the short name;
}
if (defined $tablecfg{vm}->{$node}) { #see if the host pointer requires a refresh
my $hostref = $hostrefbynode{$node};
if ($hostref and $hostref eq $_->{'runtime.host'}->value) { next; } #the actual host reference matches the one that we got when populating hostviews based on what the table had to say #TODO: does this mean it is buggy if we want to mkvm/rmigrate/etc if the current vm.host is wrong and the noderange doesn't have something on the right hostview making us not get it in the
#mass request? Or is it just slower because it hand gets host views?
my $host = $vcenterhash{$currentvcenter}->{conn}->get_view(mo_ref=>$_->{'runtime.host'},properties=>['summary.config.name']);
$host = $host->{'summary.config.name'};
my $shost = $host;
$shost =~ s/\..*//;
#time to figure out which of these is a node
my @nodes = noderange("$host,$shost");
my $vmtab = xCAT::Table->new("vm",-create=>1);
unless($vmtab){
die "Error opening vm table";
}
if ($nodes[0]) {
if ($limbonodes{$node}) { $foundlimbo{$node}=$currentvcenter; }
$vmtab->setNodeAttribs($node,{host=>$nodes[0]});
} #else {
# $vmtab->setNodeAttribs($node,{host=>$host});
#}
}
}
}
foreach my $lnode (keys %foundlimbo) {
$vmviews= $vcviews{$foundlimbo{$lnode}};
my %mgdvms; #sort into a hash for convenience
foreach (@$vmviews) {
$mgdvms{$_->{'config.name'}} = $_;
}
$function->(
node=>$lnode,
vm=>$lnode,
vmview=>$mgdvms{$node},
exargs=>\@exargs
);
}
my @entitylist;
push @entitylist,keys %hyphash;
push @entitylist,keys %clusterhash;
foreach my $entity (@entitylist) {
if ($hyphash{$entity}) {
$hyp=$entity; #save some retyping...
if ($viavcenterbyhyp->{$hyp}) {
$vmviews= $vcviews{$hyphash{$hyp}->{vcenter}->{name}}
} else {
$vmviews = [];
my $node;
foreach $node (sort (keys %{$hyphash{$hyp}->{nodes}})){
push @{$vmviews},$hyphash{$hyp}->{conn}->find_entity_view(view_type => 'VirtualMachine',properties=>$properties,filter=>{'config.name'=>qr/^$node/});
}
#$vmviews = $hyphash{$hyp}->{conn}->find_entity_views(view_type => 'VirtualMachine',properties=>$properties);
}
my %mgdvms; #sort into a hash for convenience
foreach (@$vmviews) {
$mgdvms{$_->{'config.name'}} = $_;
}
my $node;
foreach $node (sort (keys %{$hyphash{$hyp}->{nodes}})){
$function->(
node=>$node,
hyp=>$hyp,
vmview=>$mgdvms{$node},
exargs=>\@exargs
);
process_tasks; #check for tasks needing followup actions before the task is forgotten (VMWare's memory is fairly short at times
}
} else { #a cluster.
$vmviews= $vcviews{$clusterhash{$entity}->{vcenter}->{name}};
my %mgdvms; #sort into a hash for convenience
foreach (@$vmviews) {
$mgdvms{$_->{'config.name'}} = $_;
}
my $node;
foreach $node (sort (keys %{$clusterhash{$entity}->{nodes}})){
$function->(
node=>$node,
cluster=>$entity,
vm=>$node,
vmview=>$mgdvms{$node},
exargs=>\@exargs
);
}
}
}
}
sub generic_hyp_operation { #The general form of firing per-hypervisor requests to ESX hypervisor
my $function = shift; #The function to actually run against the right VM view
my @exargs = @_; #Store the rest to pass on
my $hyp;
if (scalar keys %limbonodes) { #we are in forced migration with dead sources, try to register them
@ARGV=@exargs;
my $datastoredest;
my $offline;
unless (GetOptions(
's=s' => \$datastoredest,
'f' => \$offline,
)) {
xCAT::SvrUtils::sendmsg([1,"Error parsing arguments"], $output_handler);
return;
}
if ($datastoredest) {
xCAT::SvrUtils::sendmsg([1,"Storage migration impossible with dead hypervisor, must be migrated to live hypervisor first"], $output_handler);
return;
} elsif (@ARGV) {
my $target=shift @ARGV;
if (@ARGV) {
xCAT::SvrUtils::sendmsg([1,"Unrecognized arguments ".join(' ',@ARGV)], $output_handler);
return;
}
foreach (keys %limbonodes) {
register_vm($target,$_,undef,\&migrate_ok,{ nodes => [$_], target=>$target, },"failonerror");
}
} else { #storage migration only
xCAT::SvrUtils::sendmsg([1,"No target hypervisor specified"], $output_handler);
}
}
foreach $hyp (keys %hyphash) {
process_tasks; #check for tasks needing followup actions before the task is forgotten (VMWare's memory is fairly short at times
my @relevant_nodes = sort (keys %{$hyphash{$hyp}->{nodes}});
unless (scalar @relevant_nodes) {
next;
}
$function->(
nodes => \@relevant_nodes,
hyp => $hyp,
exargs => \@exargs
);
#my $vmviews = $hyp_conns->{$hyp}->find_entity_views(view_type => 'VirtualMachine',properties=>['runtime.powerState','config.name']);
#my %mgdvms; #sort into a hash for convenience
#foreach (@$vmviews) {
# $mgdvms{$_->{'config.name'}} = $_;
#}
#my $node;
#foreach $node (sort (keys %{$hyp_hash->{$hyp}->{nodes}})){
# $function->($node,$mgdvms{$node},$taskstotrack,$callback,@exargs);
#REMINDER FOR RINV TO COME
# foreach (@nothing) { #@{$mgdvms{$node}->config->hardware->device}) {
# if (defined $_->{macAddress}) {
# print "\nFound a mac: ".$_->macAddress."\n";
# }
# }
# }
}
foreach $hyp (keys %clusterhash) { #clonevm, mkvm, rmigrate could land here in clustered mode with DRS/HA
process_tasks;
my @relevant_nodes = sort (keys %{$clusterhash{$hyp}->{nodes}});
unless (scalar @relevant_nodes) {
next;
}
$function->(nodes => \@relevant_nodes,cluster=>$hyp,exargs => \@exargs,conn=>$clusterhash{$hyp}->{conn});
}
}
sub rmhypervisor_disconnected {
my $task = shift;
my $parms = shift;
my $node = $parms->{node};
my $hyp = $node;
my $state = $task->info->state->val;
if ($state eq 'success') {
my $task = $hyphash{$hyp}->{hostview}->Destroy_Task();
$running_tasks{$task}->{data} = { node => $node, successtext => 'removed' };
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} =$hyp;
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
}
}
sub rmhypervisor_inmaintenance {
my $task = shift;
my $parms = shift;
my $state = $task->info->state->val;
my $node = $parms->{node};
my $intent = $parms->{successtext};
if ($state eq 'success') {
my $hyp = $parms->{node};
my $task = $hyphash{$hyp}->{hostview}->DisconnectHost_Task();
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&rmhypervisor_disconnected;
$running_tasks{$task}->{hyp} = $hyp;
$running_tasks{$task}->{data} = { node => $hyp };
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
}
}
sub lsvm {
my %args = @_;
my $hyp = $args{hyp};
$hyphash{$hyp}->{hostview} = get_hostview(hypname=>$hyp,conn=>$hyphash{$hyp}->{conn}); #,properties=>['config','configManager']);
use Data::Dumper;
my $vms = $hyphash{$hyp}->{hostview}->vm;
unless ($vms) {
return;
}
foreach (@$vms) {
my $vmv = $hyphash{$hyp}->{conn}->get_view(mo_ref=>$_);
xCAT::SvrUtils::sendmsg($vmv->name, $output_handler,$hyp);
}
return;
}
sub chhypervisor {
my %args = @_;
@ARGV = @{$args{exargs}}; #for getoptions;
my $maintenance;
my $online;
my $stat;
my $vlanaddspec;
my $vlanremspec;
require Getopt::Long;
GetOptions(
'maintenance|m' => \$maintenance,
'online|o' => \$online,
'show|s' => \$stat,
'show|s' => \$stat,
'addvlan=s' => \$vlanaddspec,
'removevlan=s' => \$vlanremspec,
);
my $hyp = $args{hyp};
$hyphash{$hyp}->{hostview} = get_hostview(hypname=>$hyp,conn=>$hyphash{$hyp}->{conn}); #,properties=>['config','configManager']);
if ($maintenance) {
if (defined $hyphash{$hyp}->{hostview}) {
my $task = $hyphash{$hyp}->{hostview}->EnterMaintenanceMode_Task(timeout=>0);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{data} = { node => $hyp , successtext => "hypervisor in maintenance mode"};
}
} elsif ($online) {
if (defined $hyphash{$hyp}->{hostview}) {
my $task = $hyphash{$hyp}->{hostview}->ExitMaintenanceMode_Task(timeout=>0);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{data} = { node => $hyp , successtext => "hypervisor online"};
}
} elsif ($stat) {
if (defined $hyphash{$hyp}->{hostview}) {
if ($hyphash{$hyp}->{hostview}->runtime->inMaintenanceMode) {
xCAT::SvrUtils::sendmsg("hypervisor in maintenance mode", $output_handler,$hyp);
} else {
xCAT::SvrUtils::sendmsg("hypervisor online", $output_handler,$hyp);
}
}
} elsif ($vlanaddspec) {
fixup_hostportgroup($vlanaddspec, $hyp);
} elsif ($vlanremspec) {
fixup_hostportgroup($vlanremspec, $hyp, action=>'remove');
}
return;
}
sub rshutdown { #TODO: refactor with next function too
my %args = @_;
my $hyp = $args{hyp};
$hyphash{$hyp}->{hostview} = get_hostview(hypname=>$hyp,conn=>$hyphash{$hyp}->{conn}); #,properties=>['config','configManager']);
if (defined $hyphash{$hyp}->{hostview}) {
my $task = $hyphash{$hyp}->{hostview}->EnterMaintenanceMode_Task(timeout=>0);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&rshutdown_inmaintenance;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{data} = { node => $hyp };
}
return;
}
sub rshutdown_inmaintenance {
my $task = shift;
my $parms = shift;
my $state = $task->info->state->val;
my $node = $parms->{node};
my $intent = $parms->{successtext};
if ($state eq 'success') {
my $hyp = $parms->{node};
if (defined $hyphash{$hyp}->{hostview}) {
my $task = $hyphash{$hyp}->{hostview}->ShutdownHost_Task(force=>0);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $hyp;
$running_tasks{$task}->{data} = { node => $hyp, successtext => "shutdown initiated" };
}
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
}
return;
}
sub rmhypervisor {
my %args = @_;
my $hyp = $args{hyp};
$hyphash{$hyp}->{hostview} = get_hostview(hypname=>$hyp,conn=>$hyphash{$hyp}->{conn}); #,properties=>['config','configManager']);
if (defined $hyphash{$hyp}->{hostview}) {
my $task = $hyphash{$hyp}->{hostview}->EnterMaintenanceMode_Task(timeout=>0);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&rmhypervisor_inmaintenance;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{data} = { node => $hyp };
}
return;
}
sub clonevms {
my %args=@_;
my $nodes = $args{nodes};
my $hyp = $args{hyp};
my $cluster = $args{cluster};
@ARGV = @{$args{exargs}}; #for getoptions;
my $base;
my $force;
my $detach;
my $specialize;
my $target;
require Getopt::Long;
GetOptions(
'b=s' => \$base,
'f' => \$force,
'd' => \$detach,
'specialize' => \$specialize,
't=s' => \$target,
);
if ($base and $target) {
foreach my $node (@$nodes) {
xCAT::SvrUtils::sendmsg([1,"Cannot specify both base (-b) and target (-t)"], $output_handler,$node);
}
return;
}
unless ($base or $target) {
foreach my $node (@$nodes) {
xCAT::SvrUtils::sendmsg([1,"Must specify one of base (-b) or target (-t)"], $output_handler,$node);
}
return;
}
if ($target and (scalar @{$nodes} != 1)) {
foreach my $node (@$nodes) {
xCAT::SvrUtils::sendmsg([1,"Cannot specify mulitple nodes to create a master from"], $output_handler,$node);
}
return;
}
if ($hyp) {
$hyphash{$hyp}->{hostview} = get_hostview(hypname=>$hyp,conn=>$hyphash{$hyp}->{conn});
}
my $newdatastores;
my $mastername;
my $url;
my $masterref;
if ($base) { #if base, we need to pull in the target datastores
my $mastertab=xCAT::Table->new('vmmaster');
$masterref=$mastertab->getNodeAttribs( $base,[qw/storage os arch profile storagemodel nics specializeparameters/]);
unless ($masterref) {
foreach my $node (@$nodes) {
xCAT::SvrUtils::sendmsg([1,"Cannot find master $base in vmmaster table"], $output_handler,$node);
}
return;
}
$newdatastores->{$masterref->{storage}}=[]; #make sure that the master datastore is mounted...
foreach (@$nodes) {
my $url;
if ($tablecfg{vm}->{$_}->[0]->{storage}) {
$url=$tablecfg{vm}->{$_}->[0]->{storage};
$url =~ s/=.*//;
} else {
$url=$masterref->{storage};
}
unless ($url) { die "Shouldn't be possible"; }
if (ref $newdatastores->{$_}) {
push @{$newdatastores->{$url}},$_;
} else {
$newdatastores->{$url}=[$_];
}
}
} elsif ($target) {
if ($url =~ m!/!) {
$url=$target;
$url =~ s!/([^/]*)\z!!;
$mastername=$1;
} else {
$url = $tablecfg{vm}->{$nodes->[0]}->[0]->{storage};
$url =~ s/.*\|//;
$url =~ s/=(.*)//;
$url =~ s/,.*//;
$mastername=$target
}
$newdatastores->{$url}=[$nodes->[0]];
}
if ($hyp) {
unless (validate_datastore_prereqs($nodes,$hyp,$newdatastores)) {
return;
}
} else { #need to build datastore map for cluster
refreshclusterdatastoremap($cluster);
}
sortoutdatacenters(nodes=>$nodes,hyp=>$hyp,cluster=>$cluster);
if ($target) {
return promote_vm_to_master(node=>$nodes->[0],target=>$target,force=>$force,detach=>$detach,cluster=>$cluster,hyp=>$hyp,url=>$url,mastername=>$mastername);
} elsif ($base) {
return clone_vms_from_master(nodes=>$nodes,base=>$base,detach=>$detach,cluster=>$cluster,hyp=>$hyp,mastername=>$base,masterent=>$masterref,specialize=>$specialize);
}
}
sub sortoutdatacenters { #figure out all the vmfolders for all the nodes passed in
my %args=@_;
my $nodes=$args{nodes};
my $hyp=$args{hyp};
my $cluster=$args{cluster};
my %nondefaultdcs;
my $deffolder;
my $conn;
if ($hyp) {
unless (defined $hyphash{$hyp}->{vmfolder}) {
$hyphash{$hyp}->{vmfolder} = $hyphash{$hyp}->{conn}->get_view(mo_ref=>$hyphash{$hyp}->{conn}->find_entity_view(view_type=>'Datacenter',properties=>['vmFolder'])->vmFolder);
}
$conn= $hyphash{$hyp}->{conn};
$deffolder=$hyphash{$hyp}->{vmfolder};
} else { #clustered
unless (defined $clusterhash{$cluster}->{vmfolder}) {
$clusterhash{$cluster}->{vmfolder} = $clusterhash{$cluster}->{conn}->get_view(mo_ref=>$clusterhash{$cluster}->{conn}->find_entity_view(view_type=>'Datacenter',properties=>['vmFolder'])->vmFolder);
}
$deffolder=$clusterhash{$cluster}->{vmfolder};
$conn= $clusterhash{$cluster}->{conn};
}
foreach (@$nodes) {
if ($tablecfg{vm}->{$_}->[0]->{datacenter}) {
$nondefaultdcs{$tablecfg{vm}->{$_}->[0]->{datacenter}}->{$_}=1;
} else {
$vmhash{$_}->{vmfolder}=$deffolder;
}
}
my $datacenter;
foreach $datacenter (keys %nondefaultdcs) {
my $vmfolder= $conn->get_view(mo_ref=>$conn->find_entity_view(view_type=>'Datacenter',properties=>['vmFolder'],filter=>{name=>$datacenter})->vmFolder,filter=>{name=>$datacenter});
foreach (keys %{$nondefaultdcs{$datacenter}}) {
$vmhash{$_}->{vmfolder}=$vmfolder;
}
}
}
sub clone_vms_from_master {
my %args = @_;
my $mastername=$args{mastername};
my $specialize=$args{specialize};
my $hyp = $args{hyp};
my $cluster=$args{cluster};
my $regex=qr/^$mastername\z/;
my @nodes=@{$args{nodes}};
my $node;
my $conn;
if ($hyp) {
$conn=$hyphash{$hyp}->{conn};
} else {
$conn=$clusterhash{$cluster}->{conn};
}
my $masterviews = $conn->find_entity_views(view_type => 'VirtualMachine',filter=>{'config.name'=>$regex});
if (scalar(@$masterviews) != 1) {
foreach $node (@nodes) {
xCAT::SvrUtils::sendmsg([1,"Unable to find master $mastername in VMWare infrastructure"], $output_handler,$node);
}
return;
}
my $masterview=$masterviews->[0];
my $masterent=$args{masterent};
my $ostype;
foreach $node (@nodes) {
my $destination=$tablecfg{vm}->{$node}->[0]->{storage};
my $nodetypeent;
my $vment;
$ostype=$masterent->{'os'};
foreach (qw/os arch profile/) {
$nodetypeent->{$_}=$masterent->{$_};
}
foreach (qw/storagemodel nics/) {
$vment->{$_}=$masterent->{$_};
}
$vment->{master}=$args{mastername};
unless ($destination) {
$destination=$masterent->{storage};
$vment->{storage}=$destination;
}
$destination =~ s/=.*//;
my $placement_resources=get_placement_resources(hyp=>$hyp,cluster=>$cluster,destination=>$destination);
my $pool=$placement_resources->{pool};
my $dstore=$placement_resources->{datastore};
my %relocatespecargs = (
datastore=>$dstore, #$hyphash{$hyp}->{datastorerefmap}->{$destination},
pool=>$pool,
#diskMoveType=>"createNewChildDiskBacking", #fyi, requires a snapshot, which isn't compatible with templates, moveChildMostDiskBacking would potentially be fine, but either way is ha incopmatible and limited to 8, arbitrary limitations hard to work around...
);
unless ($args{detach}) {
$relocatespecargs{diskMoveType}="createNewChildDiskBacking";
}
if ($hyp) { $relocatespecargs{host}=$hyphash{$hyp}->{hostview} }
my $relocatespec = VirtualMachineRelocateSpec->new(%relocatespecargs);
my %clonespecargs = (
location=>$relocatespec,
template=>0,
powerOn=>0
);
unless ($args{detach}) {
$clonespecargs{snapshot}=$masterview->snapshot->currentSnapshot;
}
if ($specialize) {
my %custargs;
if ($masterent->{specializeparameters}) { %custargs = ( parameters=>$masterent->{specializeparameters} ); }
$clonespecargs{customization} = make_customization_spec($node,ostype=>$ostype,%custargs);
}
my $clonespec = VirtualMachineCloneSpec->new(%clonespecargs);
my $vmfolder = $vmhash{$node}->{vmfolder};
my $task = $masterview->CloneVM_Task(folder=>$vmfolder,name=>$node,spec=>$clonespec);
$running_tasks{$task}->{data} = { node => $node, conn=>$conn, successtext => 'Successfully cloned from '.$args{mastername},
mastername=>$args{mastername}, nodetypeent=>$nodetypeent,vment=>$vment,
hyp=>$args{hyp},
};
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&clone_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp}; #$hyp_conns->{$hyp};
$running_tasks{$task}->{vm} = $node; #$hyp_conns->{$hyp};
}
}
sub make_customization_spec {
my $node = shift;
my %args = @_;
my $password="Passw0rd";
my $wintimezone;
#map of number to strings can be found at
#http://osman-shener-en.blogspot.com/2008/02/unattendedtxt-time-zone-index.html
my $fullname="Unspecified User";
my $orgName="Unspecified Organization";
if ($::XCATSITEVALS{winfullname}) { $fullname = $::XCATSITEVALS{winfullname}; }
if ($::XCATSITEVALS{winorgname}) { $orgName = $::XCATSITEVALS{winorgname}; }
my @runonce=(); #to be read in from postscripts table
$wintimezone=xCAT::TZUtils::get_wintimezonenum();
my $ptab=xCAT::Table->new('postscripts',-create=>0);
if ($ptab) {
my $psent = $ptab->getNodeAttribs($node,[qw/postscripts postbootscripts/]);
if ($psent and $psent->{postscripts}) {
push @runonce,split /,/,$psent->{postscripts};
}
if ($psent and $psent->{postbootscripts}) {
push @runonce,split /,/,$psent->{postbootscripts};
}
}
$ptab = xCAT::Table->new('passwd',-create=>0);
unless ($ptab) {
die "passwd table needed";
}
my ($passent) = $ptab->getAttribs({"key"=>"system",username=>"Administrator"},'password');
unless ($passent) {
die "need passwd table entry for system account Administrator";
}
$password=$passent->{password};
my %lfpd;
if ($args{ostype} and $args{ostype} =~ /win2k3/) {
%lfpd = (
licenseFilePrintData=>CustomizationLicenseFilePrintData->new(
autoMode=>CustomizationLicenseDataMode->new(
'perSeat'
)
)
);
}
my %runonce;
if (scalar @runonce) { #skip section if no postscripts or postbootscripts
%runonce=(
guiRunOnce=>CustomizationGuiRunOnce->new(
commandList=>\@runonce,
)
);
}
my %autologonargs = ( autoLogon=>0, autoLogonCount=>1, );
if ($args{parameters} and $args{parameters} =~ /autoLogonCount=([^,]*)/i) {
my $count = $1;
if ($count) {
$autologonargs{autoLogon}=1;
$autologonargs{autoLogonCount}=$count;
}
}
my $identity = CustomizationSysprep->new(
%runonce,
%lfpd,
guiUnattended => CustomizationGuiUnattended->new(
%autologonargs,
password=>CustomizationPassword->new(
plainText=>1,
value=>$password,
),
timeZone=>$wintimezone,
),
identification=>get_customizedidentification(),
userData=>CustomizationUserData->new(
computerName=>CustomizationFixedName->new(name=>$node),
fullName=>$fullname,
orgName=>$orgName,
productId=>"",
),
);
my $options = CustomizationWinOptions->new(changeSID=>1,deleteAccounts=>0);
my $customizationspec = CustomizationSpec->new(
globalIPSettings=>CustomizationGlobalIPSettings->new(),
identity=>$identity,
nicSettingMap=>[
CustomizationAdapterMapping->new(adapter=>CustomizationIPSettings->new(ip=>CustomizationDhcpIpGenerator->new()))
],
options=>$options,
);
return $customizationspec;
}
sub get_customizedidentification {
#for now, just do a 'TBD' workgroup. VMWare not supporting joining without domain admin password is rather unfortunate
return CustomizationIdentification->new(
joinWorkgroup=>"TBD",
);
}
sub get_placement_resources {
my %args = @_;
my $pool;
my $dstore;
my $hyp = $args{hyp};
my $cluster = $args{cluster};
my $destination=$args{destination};
if ($hyp) {
unless (defined $hyphash{$hyp}->{pool}) {
$hyphash{$hyp}->{pool} = $hyphash{$hyp}->{conn}->get_view(mo_ref=>$hyphash{$hyp}->{hostview}->parent,properties=>['resourcePool'])->resourcePool;
}
$pool=$hyphash{$hyp}->{pool};
if ($destination) { $dstore=$hyphash{$hyp}->{datastorerefmap}->{$destination} };
} else {#clustered...
unless (defined $clusterhash{$cluster}->{pool}) {
my $cview = get_clusterview(clustname=>$cluster,conn=>$clusterhash{$cluster}->{conn});
$clusterhash{$cluster}->{pool}=$cview->resourcePool;
}
$pool=$clusterhash{$cluster}->{pool};
if ($destination) { $dstore=$clusterhash{$cluster}->{datastorerefmap}->{$destination} };
}
return {
pool=>$pool,
datastore=>$dstore,
}
}
sub clone_task_callback {
my $task = shift;
my $parms = shift;
my $state = $task->info->state->val;
my $node = $parms->{node};
my $conn = $parms->{conn};
my $intent = $parms->{successtext};
if ($state eq 'success') {
#xCAT::SvrUtils::sendmsg($intent, $output_handler,$node);
my $nodetype=xCAT::Table->new('nodetype',-create=>1);
my $vm=xCAT::Table->new('vm',-create=>1);
$vm->setAttribs({node=>$node},$parms->{vment});
$nodetype->setAttribs({node=>$node},$parms->{nodetypeent});
foreach (keys %{$parms->{vment}}) {
$tablecfg{vm}->{$node}->[0]->{$_}=$parms->{vment}->{$_};
}
my @networks = split /,/,$tablecfg{vm}->{$node}->[0]->{nics};
my @macs = xCAT::VMCommon::getMacAddresses(\%tablecfg,$node,scalar @networks);
#now with macs, change all macs in the vm to match our generated macs
my $regex = qr/^$node(\z|\.)/;
#have to do an expensive pull of the vm view, since it is brand new
my $nodeviews = $conn->find_entity_views(view_type => 'VirtualMachine',filter=>{'config.name'=>$regex});
unless (scalar @$nodeviews == 1) { die "this should be impossible"; }
my $vpdtab=xCAT::Table->new('vpd',-create=>1);
$vpdtab->setAttribs({node=>$node},{uuid=>$nodeviews->[0]->config->uuid});
my $ndev;
my @devstochange;
foreach $ndev (@{$nodeviews->[0]->config->hardware->device}) {
unless ($ndev->{macAddress}) { next; } #not an ndev
$ndev->{macAddress}=shift @macs;
$ndev->{addressType}="manual";
push @devstochange, VirtualDeviceConfigSpec->new(
device => $ndev,
operation => VirtualDeviceConfigSpecOperation->new('edit'));
}
if (@devstochange) {
my $reconfigspec = VirtualMachineConfigSpec->new(deviceChange=>\@devstochange);
my $task = $nodeviews->[0]->ReconfigVM_Task(spec=>$reconfigspec);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $parms->{hyp};
$running_tasks{$task}->{conn} = $parms->{conn};
$running_tasks{$task}->{data} = { node => $node, successtext => $intent};
}
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
}
}
sub promote_vm_to_master {
my %args = @_;
my $node=$args{node};
my $hyp=$args{hyp};
my $cluster=$args{cluster};
my $regex=qr/^$node(\z|\.)/;
my $conn;
if ($hyp) {
$conn=$hyphash{$hyp}->{conn};
} else {
$conn=$clusterhash{$cluster}->{conn};
}
my $nodeviews = $conn->find_entity_views(view_type => 'VirtualMachine',filter=>{'config.name'=>$regex});
if (scalar(@$nodeviews) != 1) {
xCAT::SvrUtils::sendmsg([1,"Cannot find $node in VMWare infrastructure"], $output_handler,$node);
return;
}
my $nodeview = shift @$nodeviews;
my $dstore;
if ($hyp) {
$dstore=$hyphash{$hyp}->{datastorerefmap}->{$args{url}},
} else {
$dstore=$clusterhash{$cluster}->{datastorerefmap}->{$args{url}},
}
my $relocatespec = VirtualMachineRelocateSpec->new(
datastore=>$dstore,
);
my $clonespec = VirtualMachineCloneSpec->new(
location=>$relocatespec,
template=>0, #can't go straight to template, need to clone, then snap, then templatify
powerOn=>0
);
my $vmfolder=$vmhash{$node}->{vmfolder};
my $task = $nodeview->CloneVM_Task(folder=>$vmfolder,name=>$args{mastername},spec=>$clonespec);
$running_tasks{$task}->{data} = { node => $node, hyp => $args{hyp}, conn => $conn, successtext => 'Successfully copied to '.$args{mastername}, mastername=>$args{mastername}, url=>$args{url} };
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&promote_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp}; #$hyp_conns->{$hyp};
$running_tasks{$task}->{vm}=$node;
}
sub promote_task_callback {
my $task = shift;
my $parms = shift;
my $state = $task->info->state->val;
my $node = $parms->{node};
my $intent = $parms->{successtext};
if ($state eq 'success') { #now, we have to make one snapshot for linked clones
my $mastername=$parms->{mastername};
my $regex=qr/^$mastername\z/;
my $masterviews = $parms->{conn}->find_entity_views(view_type => 'VirtualMachine',filter=>{'config.name'=>$regex});
unless (scalar @$masterviews == 1) {
die "Impossible";
}
my $masterview = $masterviews->[0];
my $task = $masterview->CreateSnapshot_Task(name=>"xcatsnap",memory=>"false",quiesce=>"false");
$parms->{masterview}=$masterview;
$running_tasks{$task}->{data} = $parms;
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&promotesnap_task_callback;
$running_tasks{$task}->{hyp} = $parms->{hyp}; #$hyp_conns->{$hyp};
$running_tasks{$task}->{vm}=$parms->{node};
#xCAT::SvrUtils::sendmsg($intent, $output_handler,$node);
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
}
}
sub promotesnap_task_callback {
my $task = shift;
my $parms = shift;
my $state = $task->info->state->val;
my $node = $parms->{node};
my $intent = $parms->{successtext};
if ($state eq 'success') {
$parms->{masterview}->MarkAsTemplate; #time to be a template
xCAT::SvrUtils::sendmsg($intent, $output_handler,$node);
my $mastertabentry = {
originator=>$requester,
vintage=>scalar(localtime),
storage=>$parms->{url},
};
foreach (qw/os arch profile/) {
if (defined ($tablecfg{nodetype}->{$node}->[0]->{$_})) {
$mastertabentry->{$_}=$tablecfg{nodetype}->{$node}->[0]->{$_};
}
}
foreach (qw/storagemodel nics/) {
if (defined ($tablecfg{vm}->{$node}->[0]->{$_})) {
$mastertabentry->{$_}=$tablecfg{vm}->{$node}->[0]->{$_};
}
}
my $vmmastertab=xCAT::Table->new('vmmaster',-create=>1);
my $date=scalar(localtime);
$vmmastertab->setAttribs({name=>$parms->{mastername}},$mastertabentry);
} elsif ($state eq 'error') {
relay_vmware_err($task,"",$node);
}
}
sub mkvms {
my %args = @_;
my $nodes = $args{nodes};
my $hyp = $args{hyp};
my $cluster = $args{cluster};
@ARGV = @{$args{exargs}}; #for getoptions;
my $disksize;
require Getopt::Long;
my $cpuCount;
my $memory;
GetOptions(
'size|s=s' => \$disksize,
"cpus=s" => \$cpuCount,
"mem=s" => \$memory
);
my $node;
my $conn;
if ($hyp) {
$hyphash{$hyp}->{hostview} = get_hostview(hypname=>$hyp,conn=>$hyphash{$hyp}->{conn}); #,properties=>['config','configManager']);
unless (validate_datastore_prereqs($nodes,$hyp)) {
return;
}
$conn=$hyphash{$hyp}->{conn};
} else {
refreshclusterdatastoremap($cluster);
$conn=$clusterhash{$cluster}->{conn};
}
sortoutdatacenters(nodes=>$nodes,hyp=>$hyp,cluster=>$cluster);
my $placement_resources=get_placement_resources(hyp=>$hyp,cluster=>$cluster);
#$hyphash{$hyp}->{pool} = $hyphash{$hyp}->{conn}->get_view(mo_ref=>$hyphash{$hyp}->{hostview}->parent,properties=>['resourcePool'])->resourcePool;
my $cfg;
foreach $node (@$nodes) {
process_tasks; #check for tasks needing followup actions before the task is forgotten (VMWare's memory is fairly short at times
if ($conn->find_entity_view(view_type=>"VirtualMachine",filter=>{name=>$node})) {
xCAT::SvrUtils::sendmsg([1,"Virtual Machine already exists"], $output_handler,$node);
next;
} else {
register_vm($hyp,$node,$disksize,undef,undef,undef,cpus=>$cpuCount,memory=>$memory,cluster=>$cluster);
}
}
my @dhcpnodes;
foreach (keys %{$tablecfg{dhcpneeded}}) {
push @dhcpnodes,$_;
delete $tablecfg{dhcpneeded}->{$_};
}
unless ($::XCATSITEVALS{'dhcpsetup'} and ($::XCATSITEVALS{'dhcpsetup'} =~ /^n/i or $::XCATSITEVALS{'dhcpsetup'} =~ /^d/i or $::XCATSITEVALS{'dhcpsetup'} eq '0')) {
$executerequest->({command=>['makedhcp'],node=>\@dhcpnodes});
}
}
sub setboot {
my %args = @_;
my $node = $args{node};
my $hyp = $args{hyp};
if (not defined $args{vmview}) { #attempt one refresh
$args{vmview} = $vmhash{$node}->{conn}->find_entity_view(view_type => 'VirtualMachine',properties=>['config.name'],filter=>{name=>$node});
if (not defined $args{vmview}) {
xCAT::SvrUtils::sendmsg([1,"VM does not appear to exist"], $output_handler,$node);
return;
}
}
my $bootorder = ${$args{exargs}}[0];
#NOTE: VMware simply does not currently seem to allow programatically changing the boot
#order like other virtualization solutions supported by xCAT.
#This doesn't behave quite like any existing mechanism:
#vm.bootorder was meant to take the place of system nvram, vmware imitates that unfortunate aspect of bare metal too well..
#rsetboot was created to describe the ipmi scenario of a transient boot device, this is persistant *except* for setup, which is not
#rbootseq was meant to be entirely persistant and ordered.
#rsetboot is picked, the usage scenario matches about as good as I could think of
my $reconfigspec;
if ($bootorder =~ /setup/) {
unless ($bootorder eq 'setup') {
xCAT::SvrUtils::sendmsg([1,"rsetboot parameter may not contain 'setup' with other items, assuming vm.bootorder is just 'setup'"], $output_handler,$node);
}
$reconfigspec = VirtualMachineConfigSpec->new(
bootOptions=>VirtualMachineBootOptions->new(enterBIOSSetup=>1),
);
} else {
$bootorder = "allow:".$bootorder;
$reconfigspec = VirtualMachineConfigSpec->new(
bootOptions=>VirtualMachineBootOptions->new(enterBIOSSetup=>0),
extraConfig => [OptionValue->new(key => 'bios.bootDeviceClasses',value=>$bootorder)]
);
}
my $task = $args{vmview}->ReconfigVM_Task(spec=>$reconfigspec);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&generic_task_callback;
$running_tasks{$task}->{hyp} = $args{hyp};
$running_tasks{$task}->{data} = { node => $node, successtext => ${$args{exargs}}[0] };
}
sub register_vm {#Attempt to register existing instance of a VM
my $hyp = shift;
my $node = shift;
my $disksize = shift;
my $blockedfun = shift; #a pointer to a blocked function to call on success
my $blockedargs = shift; #hash reference to call blocked function with
my $failonerr = shift;
my %args=@_; #ok, went overboard with positional arguments, from now on, named arguments
my $task;
if ($hyp) {
validate_network_prereqs([keys %{$hyphash{$hyp}->{nodes}}],$hyp);
unless (defined $hyphash{$hyp}->{datastoremap} or validate_datastore_prereqs([keys %{$hyphash{$hyp}->{nodes}}],$hyp)) {
die "unexpected condition";
}
} else {
scan_cluster_networks($args{cluster});
}
sortoutdatacenters(nodes=>[$node],hyp=>$hyp,cluster=>$args{cluster});
my $placement_resources=get_placement_resources(hyp=>$hyp,cluster=>$args{cluster});
# Try to add an existing VM to the machine folder
my $success = eval {
if ($hyp) {
$task = $vmhash{$node}->{vmfolder}->RegisterVM_Task(path=>getcfgdatastore($node,$hyphash{$hyp}->{datastoremap})." /$node/$node.vmx",name=>$node,pool=>$hyphash{$hyp}->{pool},host=>$hyphash{$hyp}->{hostview},asTemplate=>0);
} else {
$task = $vmhash{$node}->{vmfolder}->RegisterVM_Task(path=>getcfgdatastore($node,$clusterhash{$args{cluster}}->{datastoremap})." /$node/$node.vmx",name=>$node,pool=>$placement_resources->{pool},asTemplate=>0);
}
};
# if we couldn't add it then it means it wasn't created yet. So we create it.
my $cluster=$args{cluster};
if ($@ or not $success) {
#if (ref($@) eq 'SoapFault') {
# if (ref($@->detail) eq 'NotFound') {
register_vm_callback(undef, {
node => $node,
disksize => $disksize,
blockedfun => $blockedfun,
blockedargs => $blockedargs,
errregister=>$failonerr,
cpus=>$args{cpus},
memory=>$args{memory},
hyp => $hyp,
cluster=>$cluster,
});
}
if ($task) {
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&register_vm_callback;
$running_tasks{$task}->{hyp} = $hyp;
$running_tasks{$task}->{cluster} = $cluster;
$running_tasks{$task}->{data} = {
node => $node,
disksize => $disksize,
blockedfun => $blockedfun,
blockedargs => $blockedargs,
errregister=>$failonerr,
cpus=>$args{cpus},
memory=>$args{memory},
hyp => $hyp,
cluster=>$cluster,
};
}
}
sub register_vm_callback {
my $task = shift;
my $args = shift;
if (not $task or $task->info->state->val eq 'error') { #TODO: fail for 'rpower' flow, mkvm is too invasive in VMWare to be induced by 'rpower on'
if (not defined $args->{blockedfun}) {
mknewvm($args->{node},$args->{disksize},$args->{hyp},$args);
} elsif ($args->{errregister}) {
relay_vmware_err($task,"",$args->{node});
} else {
xCAT::SvrUtils::sendmsg([1,"mkvm must be called before use of this function"], $output_handler,$args->{node});
}
} elsif (defined $args->{blockedfun}) { #If there is a blocked function, call it here)
$args->{blockedfun}->(%{$args->{blockedargs}});
}
}
sub getURI {
my $method = shift;
my $location = shift;
my $uri = '';
if($method =~ /nfs/){
(my $server,my $path) = split/\//,$location,2;
$server =~ s/:$//; #tolerate habitual colons
my $servern = inet_aton($server);
unless ($servern) {
xCAT::SvrUtils::sendmsg([1,"could not resolve '$server' to an address from vm.storage/vm.cfgstore"], $output_handler);
}
$server = inet_ntoa($servern);
$uri = "nfs://$server/$path";
}elsif($method =~ /vmfs/){
(my $name, undef) = split /\//,$location,2;
$name =~ s/:$//; #remove a : if someone put it in for some reason.
$uri = "vmfs://$name";
}else{
xCAT::SvrUtils::sendmsg([1,"Unsupported VMware Storage Method: $method. Please use 'vmfs or nfs'"], $output_handler);
}
return $uri;
}
sub getcfgdatastore {
my $node = shift;
my $dses = shift;
my $cfgdatastore = $tablecfg{vm}->{$node}->[0]->{cfgstore};
unless ($cfgdatastore) {
$cfgdatastore = $tablecfg{vm}->{$node}->[0]->{storage};
#TODO: if multiple drives are specified, make sure to split this out
#DONE: I believe the regex after this conditional takes care of that case already..
}
$cfgdatastore =~ s/=.*//;
(my $method,my $location) = split /:\/\//,$cfgdatastore,2;
my $uri = $cfgdatastore;
unless ($dses->{$uri}) { #don't call getURI if map works out fine already
$uri = getURI($method,$location);
}
$cfgdatastore = "[".$dses->{$uri}."]";
#$cfgdatastore =~ s/,.*$//; #these two lines of code were kinda pointless
#$cfgdatastore =~ s/\/$//;
return $cfgdatastore;
}
sub mknewvm {
my $node=shift;
my $disksize=shift;
my $hyp=shift;
my $otherargs=shift;
my $cluster=$otherargs->{cluster};
my $placement_resources=get_placement_resources(hyp=>$hyp,cluster=>$cluster);
my $pool=$placement_resources->{pool};
my $cfg;
if ($hyp) {
$cfg = build_cfgspec($node,$hyphash{$hyp}->{datastoremap},$hyphash{$hyp}->{nets},$disksize,$hyp,$otherargs);
} else { #cluster based..
$cfg = build_cfgspec($node,$clusterhash{$cluster}->{datastoremap},$clusterhash{$cluster}->{nets},$disksize,$hyp,$otherargs);
}
my $task;
if ($hyp) {
$task = $vmhash{$node}->{vmfolder}->CreateVM_Task(config=>$cfg,pool=>$hyphash{$hyp}->{pool},host=>$hyphash{$hyp}->{hostview});
} else {
$task = $vmhash{$node}->{vmfolder}->CreateVM_Task(config=>$cfg,pool=>$pool); #drs away
}
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&mkvm_callback;
$running_tasks{$task}->{hyp} = $hyp;
$running_tasks{$task}->{cluster} = $cluster;
$running_tasks{$task}->{data} = { hyp=>$hyp, cluster=>$cluster, node => $node };
}
sub getUnits {
my $amount = shift;
my $defunit = shift;
my $divisor=shift;
unless ($amount) { return; }
unless ($divisor) {
$divisor = 1;
}
if ($amount =~ /(\D)$/) { #If unitless, add unit
$defunit=$1;
chop $amount;
}
if ($defunit =~ /k/i) {
return $amount*1024/$divisor;
} elsif ($defunit =~ /m/i) {
return $amount*1048576/$divisor;
} elsif ($defunit =~ /g/i) {
return $amount*1073741824/$divisor;
}
}
sub getguestid {
my $osfound=0;
my $node = shift;
if ($tablecfg{vm}->{$node}->[0]->{guestostype}) { #if admin wants to skip derivation from nodetype.os value, let em
return $tablecfg{vm}->{$node}->[0]->{guestostype};
}
my $nodeos = $tablecfg{nodetype}->{$node}->[0]->{os};
my $nodearch = $tablecfg{nodetype}->{$node}->[0]->{arch};
foreach (keys %guestidmap) {
if (defined($nodeos) and $nodeos =~ /$_/) {
if ($nodearch eq 'x86_64' and $_ !~ /vmkernel/) {
$nodeos=$guestidmap{$_}."64Guest";
} else {
$nodeos=$guestidmap{$_};
$nodeos =~ s/_$//;
$nodeos .= "Guest";
}
$osfound=1;
last;
}
}
unless ($osfound) {
if (defined($nodearch) and $nodearch eq 'x86_64') {
$nodeos="otherGuest64";
} else {
$nodeos="otherGuest";
}
}
return $nodeos;
}
sub build_cfgspec {
my $node = shift;
my $dses = shift; #map to match vm table to datastore names
my $netmap = shift;
my $disksize = shift;
my $hyp = shift;
my $otherargs=shift;
my $memory;
my $ncpus;
my $updatehash;
if ($otherargs->{memory}) {
$memory=getUnits($otherargs->{memory},"M",1048576);
if ($tablecfg{vm}->{$node}->[0]->{memory}) {
$updatehash->{memory}=$memory;
}
} elsif ($tablecfg{vm}->{$node}->[0]->{memory}) {
$memory = getUnits($tablecfg{vm}->{$node}->[0]->{memory},"M",1048576);
} else {
$memory = 512;
}
if ($otherargs->{cpus}) {
$ncpus=$otherargs->{cpus};
if ($tablecfg{vm}->{$node}->[0]->{cpus}) {
$updatehash->{cpus}=$ncpus;
}
} elsif ($tablecfg{vm}->{$node}->[0]->{cpus}) {
$ncpus = $tablecfg{vm}->{$node}->[0]->{cpus};
} else {
$ncpus = 1;
}
if ($updatehash) {
my $vmtab = xCAT::Table->new('vm',-create=>1);
$vmtab->setNodeAttribs($node,$updatehash);
}
my @devices;
$currkey=0;
my $opticalbacking = VirtualCdromRemoteAtapiBackingInfo->new(deviceName=>"");
my $opticalconnectable = VirtualDeviceConnectInfo->new(startConnected=>0,allowGuestControl=>1,connected=>0);
my $optical =VirtualCdrom->new( controllerKey => 201,
connectable=>$opticalconnectable,
backing=>$opticalbacking,
key => $currkey++,
unitNumber => 0, );
push @devices,VirtualDeviceConfigSpec->new(device => $optical, operation => VirtualDeviceConfigSpecOperation->new('add'));
push @devices,create_storage_devs($node,$dses,$disksize);
push @devices,create_nic_devs($node,$netmap,$hyp);
#my $cfgdatastore = $tablecfg{vm}->{$node}->[0]->{storage}; #TODO: need a new cfglocation field in case of stateless guest?
#$cfgdatastore =~ s/,.*$//;
#$cfgdatastore =~ s/\/$//;
#$cfgdatastore = "[".$dses->{$cfgdatastore}."]";
my $cfgdatastore = getcfgdatastore($node,$dses);
my $vfiles = VirtualMachineFileInfo->new(vmPathName=>$cfgdatastore);
#my $nodeos = $tablecfg{nodetype}->{$node}->[0]->{os};
#my $nodearch = $tablecfg{nodetype}->{$node}->[0]->{arch};
my $nodeos = getguestid($node); #nodeos=>$nodeos,nodearch=>$nodearch);
my $uuid;
if ($tablecfg{vpd}->{$node}->[0]->{uuid}) {
$uuid = $tablecfg{vpd}->{$node}->[0]->{uuid};
} else {
if ($tablecfg{mac}->{$node}->[0]->{mac}) { #a uuidv1 is possible, generate that for absolute uniqueness guarantee
my $mac = $tablecfg{mac}->{$node}->[0]->{mac};
$mac =~ s/\|.*//;
$mac =~ s/!.*//;
$uuid=xCAT::Utils::genUUID(mac=>$mac);
} else {
$uuid=xCAT::Utils::genUUID();
}
my $vpdtab = xCAT::Table->new('vpd');
$vpdtab->setNodeAttribs($node,{uuid=>$uuid});
}
$uuid =~ s/^(..)(..)(..)(..)-(..)(..)-(..)(..)/$4$3$2$1-$6$5-$8$7/;
my @optionvals;
if ($tablecfg{vm}->{$node}->[0]->{othersettings}) {
my $key;
my $value;
foreach (split /;/,$tablecfg{vm}->{$node}->[0]->{othersettings}) {
($key,$value)=split /=/;
if ($value) {
push @optionvals,OptionValue->new(key=>$key,value=>$value);
} else {
push @optionvals,OptionValue->new(key=>$key);
}
}
}
my %specargs = (
name => $node,
files => $vfiles,
guestId=>$nodeos,
memoryMB => $memory,
numCPUs => $ncpus,
deviceChange => \@devices,
uuid=>$uuid,
);
if (@optionvals) {
$specargs{extraConfig}=\@optionvals;
}
return VirtualMachineConfigSpec->new(%specargs);
}
sub create_nic_devs {
my $node = shift;
my $netmap = shift;
my $hyp = shift;
my @networks = split /,/,$tablecfg{vm}->{$node}->[0]->{nics};
my @devs;
my $idx = 0;
my @macs = xCAT::VMCommon::getMacAddresses(\%tablecfg,$node,scalar @networks);
my $connprefs=VirtualDeviceConnectInfo->new(
allowGuestControl=>1,
connected=>0,
startConnected => 1
);
my $model=$tablecfg{vm}->{$node}->[0]->{nicmodel};
unless ($model) {
$model='e1000';
}
foreach (@networks) {
my $pgname=$_;
if ($hyp) {
$pgname = $hyphash{$hyp}->{pgnames}->{$_};
}
s/.*://;
my $hadspecmodel=0;
if (m/=/) {
$hadspecmodel=1;
s/=(.*)$//;
}
my $tmpmodel=$model;
if ($hadspecmodel) { $tmpmodel=$1; }
my $netname = $_;
my $backing = VirtualEthernetCardNetworkBackingInfo->new(
network => $netmap->{$pgname},
deviceName=>$pgname,
);
my %newcardargs=(
key=>0,#3, #$currkey++,
backing=>$backing,
addressType=>"manual",
macAddress=>shift @macs,
connectable=>$connprefs,
wakeOnLanEnabled=>1, #TODO: configurable in tables?
);
my $newcard;
if ($tmpmodel eq 'e1000') {
$newcard=VirtualE1000->new(%newcardargs);
} elsif ($tmpmodel eq 'vmxnet3') {
$newcard=VirtualVmxnet3->new(%newcardargs);
} elsif ($tmpmodel eq 'pcnet32') {
$newcard=VirtualPCNet32->new(%newcardargs);
} elsif ($tmpmodel eq 'vmxnet2') {
$newcard=VirtualVmxnet2->new(%newcardargs);
} elsif ($tmpmodel eq 'vmxnet') {
$newcard=VirtualVmxnet->new(%newcardargs);
} else {
xCAT::SvrUtils::sendmsg([1,"$tmpmodel not a recognized nic type, falling back to e1000 (vmxnet3, e1000, pcnet32, vmxnet2, vmxnet are recognized"], $output_handler,$node);
$newcard=VirtualE1000->new(%newcardargs);
}
push @devs,VirtualDeviceConfigSpec->new(device => $newcard,
operation => VirtualDeviceConfigSpecOperation->new('add'));
$idx++;
}
return @devs;
die "Stop running for test";
}
sub create_storage_devs {
my $node = shift;
my $sdmap = shift;
my $sizes = shift;
my @sizes = split /[,:]/, $sizes;
my $existingScsiCont = shift;
my $scsiUnit = shift;
my $existingIdeCont = shift;
my $ideUnit = shift;
my $devices = shift;
my %args=@_;
my $scsicontrollerkey=0;
my $idecontrollerkey=200; #IDE 'controllers' exist at 200 and 201 invariably, with no flexibility?
#Cannot find documentation that declares this absolute, but attempts to do otherwise
#lead in failure, also of note, these are single-channel controllers, so two devs per controller
my $backingif;
my @devs;
my $havescsidevs =0;
my $disktype = 'ide';
my $ideunitnum=0;
my $scsiunitnum=0;
my $havescsicontroller=0;
my %usedideunits;
my %usedscsiunits=(7=>1,'7'=>1);
if (defined $existingScsiCont) {
$havescsicontroller=1;
$scsicontrollerkey = $existingScsiCont->{key};
$scsiunitnum = $scsiUnit;
%usedscsiunits = %{getUsedUnits($scsicontrollerkey,$devices)};
}
if (defined $existingIdeCont) {
$idecontrollerkey = $existingIdeCont->{key};
$ideunitnum = $ideUnit;
%usedideunits = %{getUsedUnits($idecontrollerkey,$devices)};
}
my $unitnum;
my %disktocont;
my $dev;
my @storelocs = split /,/,$tablecfg{vm}->{$node}->[0]->{storage};
my $globaldisktype = $tablecfg{vm}->{$node}->[0]->{storagemodel};
unless ($globaldisktype) { $globaldisktype='ide'; }
#number of devices is the larger of the specified sizes (TODO: masters) or storage pools to span
my $numdevs = (scalar @storelocs > scalar @sizes ? scalar @storelocs : scalar @sizes);
my $controllertype='scsi';
while ($numdevs-- > 0) {
my $storeloc = shift @storelocs;
unless (scalar @storelocs) { @storelocs = ($storeloc); } #allow reuse of one cfg specified pool for multiple devs
my $disksize = shift @sizes;
unless (scalar @sizes) { @sizes = ($disksize); } #if we emptied the array, stick the last entry back on to allow it to specify all remaining disks
$disksize = getUnits($disksize,'G',1024);
$disktype = $globaldisktype;
if ($storeloc =~ /=/) {
($storeloc,$disktype) = split /=/,$storeloc;
}
if ($disktype eq 'ide' and $args{idefull}) {
xCAT::SvrUtils::sendmsg([1,"VM is at capacity for IDE devices, a drive was not added"], $output_handler,$node);
return;
} elsif (($disktype eq 'scsi' or $disktype eq 'sas' or $disktype eq 'pvscsi') and $args{scsifull}) {
xCAT::SvrUtils::sendmsg([1,"SCSI Controller at capacity, a drive was not added"], $output_handler,$node);
return;
}
$storeloc =~ s/\/$//;
(my $method,my $location) = split /:\/\//,$storeloc,2;
my $uri = $storeloc;
unless ($sdmap->{$uri}) { #don't call getURI if map works out fine already
$uri = getURI($method,$location);
}
#(my $server,my $path) = split/\//,$location,2;
#$server =~ s/:$//; #tolerate habitual colons
#my $servern = inet_aton($server);
#unless ($servern) {
# xCAT::SvrUtils::sendmsg([1,"could not resolve '$server' to an address from vm.storage"]);
# return;
#}
#$server = inet_ntoa($servern);
#my $uri = "nfs://$server/$path";
$backingif = VirtualDiskFlatVer2BackingInfo->new(diskMode => 'persistent',
thinProvisioned => 1,
fileName => "[".$sdmap->{$uri}."]");
if ($disktype eq 'ide' and $idecontrollerkey == 1 and $ideunitnum == 0) { #reserve a spot for CD
$ideunitnum = 1;
} elsif ($disktype eq 'ide' and $ideunitnum == 2) { #go from current to next ide 'controller'
$idecontrollerkey++;
$ideunitnum=0;
}
unless ($disktype eq 'ide') {
push @{$disktocont{$scsicontrollerkey}},$currkey;
}
my $controllerkey;
if ($disktype eq 'ide') {
$controllerkey = $idecontrollerkey;
$unitnum = 0;
while ($usedideunits{$unitnum}) {
$unitnum++;
}
if ($unitnum == 2) {
$idecontrollerkey++;
$ideunitnum=1;
$unitnum=1;
$controllerkey=$idecontrollerkey;
}
$usedideunits{$unitnum}=1;
} else {
$controllertype=$disktype;
$controllerkey = $scsicontrollerkey;
$unitnum = 0;
while ($usedscsiunits{$unitnum}) {
$unitnum++;
}
$usedscsiunits{$unitnum}=1;
$havescsidevs=1;
}
$dev =VirtualDisk->new(backing=>$backingif,
controllerKey => $controllerkey,
key => $currkey++,
unitNumber => $unitnum,
capacityInKB => $disksize);
push @devs,VirtualDeviceConfigSpec->new(device => $dev,
fileOperation => VirtualDeviceConfigSpecFileOperation->new('create'),
operation => VirtualDeviceConfigSpecOperation->new('add'));
}
#It *seems* that IDE controllers are not subject to require creation, so we skip it
if ($havescsidevs and not $havescsicontroller) { #need controllers to attach the disks to
foreach(0..$scsicontrollerkey) {
if ($controllertype eq 'scsi') {
$dev=VirtualLsiLogicController->new(key => $_,
device => \@{$disktocont{$_}},
sharedBus => VirtualSCSISharing->new('noSharing'),
busNumber => $_);
} elsif ($controllertype eq 'sas') {
$dev=VirtualLsiLogicSASController->new(key => $_,
device => \@{$disktocont{$_}},
sharedBus => VirtualSCSISharing->new('noSharing'),
busNumber => $_);
} elsif ($controllertype eq 'pvscsi') {
$dev=ParaVirtualSCSIController->new(key => $_,
device => \@{$disktocont{$_}},
sharedBus => VirtualSCSISharing->new('noSharing'),
busNumber => $_);
}
push @devs,VirtualDeviceConfigSpec->new(device => $dev,
operation => VirtualDeviceConfigSpecOperation->new('add'));
}
}
return @devs;
# my $ctlr = VirtualIDEController->new(
}
sub declare_ready {
my %args = %{shift()};
$hypready{$args{hyp}}=1;
}
sub populate_vcenter_hostviews {
my $vcenter = shift;
my @hypervisors;
my %nametohypmap;
my $iterations=1;
if ($usehostnamesforvcenter and $usehostnamesforvcenter !~ /no/i) {
$iterations=2; #two passes possible
my $hyp;
foreach $hyp (keys %{$vcenterhash{$vcenter}->{allhyps}}) {
if ($tablecfg{hosts}->{$hyp}->[0]->{hostnames}) {
$nametohypmap{$tablecfg{hosts}->{$hyp}->[0]->{hostnames}}=$hyp;
}
}
@hypervisors = keys %nametohypmap;
} else {
@hypervisors = keys %{$vcenterhash{$vcenter}->{allhyps}};
}
while ($iterations and scalar(@hypervisors)) {
my $hosts = join(")|(",@hypervisors);
$hosts = '^(('.$hosts.'))(\z|\.)';
my $search = qr/$hosts/;
my @hypviews = @{$vcenterhash{$vcenter}->{conn}->find_entity_views(view_type=>'HostSystem',properties=>['summary.config.name','summary.runtime.connectionState','runtime.inMaintenanceMode','parent','configManager','summary.host'],filter=>{'summary.config.name'=>$search})};
foreach (@hypviews) {
my $hypname = $_->{'summary.config.name'};
my $hypv=$_;
my $hyp;
if ($vcenterhash{$vcenter}->{allhyps}->{$hypname}) { #simplest case, config.name is exactly the same as node name
$vcenterhash{$vcenter}->{hostviews}->{$hypname} = $_;
$hyp=$hypname;
} elsif ($nametohypmap{$hypname}) { #second case, there is a name mapping this to a real name
$vcenterhash{$vcenter}->{hostviews}->{$nametohypmap{$hypname}} = $_;
$hyp=$nametohypmap{$hypname};
} else { #name as-is doesn't work, start stripping domain and hope for the best
$hypname =~ s/\..*//;
if ($vcenterhash{$vcenter}->{allhyps}->{$hypname}) { #shortname is a node
$vcenterhash{$vcenter}->{hostviews}->{$hypname} = $_;
$hyp=$hypname;
} elsif ($nametohypmap{$hypname}) { #alias for node
$vcenterhash{$vcenter}->{hostviews}->{$nametohypmap{$hypname}} = $_;
$hyp=$nametohypmap{$hypname};
}
}
foreach my $nodename (keys %{$hyphash{$hyp}->{nodes}}) {
$hostrefbynode{$nodename}=$hypv->{'summary.host'}->value;
}
}
$iterations--;
@hypervisors=();
if ($usehostnamesforvcenter and $usehostnamesforvcenter !~ /no/i) { #check for hypervisors by native node name if missed above
foreach my $hyp (keys %{$vcenterhash{$vcenter}->{allhyps}}) {
unless ($vcenterhash{$vcenter}->{hostviews}->{$hyp}) {
push @hypervisors,$hyp;
}
}
}
}
}
sub create_new_cluster {
my $req = shift;
@ARGV = @{$req->{arg}};
my $vcenter;
my $password;
my $user;
my $datacenter;
GetOptions(
'vcenter=s' => \$vcenter,
'password=s' => \$password,
'datacenter=s' => \$datacenter,
'username=s' => \$user,
);
my $clustername = shift @ARGV;
my $conn = Vim->new(service_url=>"https://$vcenter/sdk");
$conn->login(user_name=>$user, password=>$password);
if ($datacenter) {
$datacenter = $conn->find_entity_view(view_type => 'Datacenter', properties=>['hostFolder'],filter=>{name=>$datacenter});
unless ($datacenter) {
xCAT::SvrUtils::sendmsg([1,": Unable to find requested datacenter"], $output_handler);
return;
}
} else {
$datacenter = $conn->find_entity_view(view_type => 'Datacenter', properties=>['hostFolder']);
}
my $hfolder = $conn->get_view(mo_ref=>$datacenter->hostFolder);
my $cfgspec = ClusterConfigSpecEx->new();
$hfolder->CreateClusterEx(name=>$clustername, spec=>$cfgspec);
}
sub list_clusters {
my $req = shift;
@ARGV = @{$req->{arg}};
my $vcenter;
my $password;
my $user;
my $datacenter;
GetOptions(
'vcenter=s' => \$vcenter,
'password=s' => \$password,
'datacenter=s' => \$datacenter,
'username=s' => \$user,
);
my $clustername = shift @ARGV;
my $conn = Vim->new(service_url=>"https://$vcenter/sdk");
$conn->login(user_name=>$user, password=>$password);
use Data::Dumper;
my $clustviews = $conn->find_entity_views(view_type=> 'ClusterComputeResource');
foreach (@$clustviews) {
xCAT::SvrUtils::sendmsg($_->{name}, $output_handler);
}
return;
}
sub validate_vcenter_prereqs { #Communicate with vCenter and ensure this host is added correctly to a vCenter instance when an operation requires it
my $hyp = shift;
my $depfun = shift;
my $depargs = shift;
my $vcenter = $hyphash{$hyp}->{vcenter}->{name};
unless ($hyphash{$hyp}->{vcenter}->{conn}) {
eval {
$hyphash{$hyp}->{vcenter}->{conn} = Vim->new(service_url=>"https://$vcenter/sdk");
$hyphash{$hyp}->{vcenter}->{conn}->login(user_name=>$hyphash{$hyp}->{vcenter}->{username},password=>$hyphash{$hyp}->{vcenter}->{password});
};
if ($@) {
$hyphash{$hyp}->{vcenter}->{conn} = undef;
}
}
unless ($hyphash{$hyp}->{vcenter}->{conn}) {
xCAT::SvrUtils::sendmsg([1,": Unable to reach vCenter server managing $hyp"], $output_handler);
return undef;
}
my $foundhyp;
my $name=$hyp;
if ($usehostnamesforvcenter and $usehostnamesforvcenter !~ /no/i) {
if ($tablecfg{hosts}->{$hyp}->[0]->{hostnames}) {
$name = $tablecfg{hosts}->{$hyp}->[0]->{hostnames};
}
}
my $connspec = HostConnectSpec->new(
hostName=>$name,
password=>$hyphash{$hyp}->{password},
userName=>$hyphash{$hyp}->{username},
force=>1,
);
my $hview;
$hview = $vcenterhash{$vcenter}->{hostviews}->{$hyp};
if ($hview) {
if ($hview->{'summary.config.name'} =~ /^$hyp(?:\.|\z)/ or $hview->{'summary.config.name'} =~ /^$name(?:\.|\z)/) { #Looks good, call the dependent function after declaring the state of vcenter to hypervisor as good
if ($hview->{'summary.runtime.connectionState'}->val eq 'connected') {
if ($vcenterautojoin) { #admin has requested manual vcenter management, don't mess with vmotion settings
enable_vmotion(hypname=>$hyp,hostview=>$hview,conn=>$hyphash{$hyp}->{vcenter}->{conn});
}
$vcenterhash{$vcenter}->{goodhyps}->{$hyp} = 1;
$depfun->($depargs);
if ($hview->parent->type eq 'ClusterComputeResource') { #if it is in a cluster, we can directly remove it
$hyphash{$hyp}->{deletionref} = $hview->{mo_ref};
} elsif ($hview->parent->type eq 'ComputeResource') { #For some reason, we must delete the container instead
$hyphash{$hyp}->{deletionref} = $hview->{parent}; #save off a reference to delete hostview off just in case
}
return 1;
} elsif ($vcenterautojoin or $vcenterforceremove) { #if allowed autojoin and the current view seems corrupt, throw it away and rejoin
my $ref_to_delete;
if ($hview->parent->type eq 'ClusterComputeResource') { #We are allowed to specifically kill a host in a cluster
$ref_to_delete = $hview->{mo_ref};
} elsif ($hview->parent->type eq 'ComputeResource') { #For some reason, we must delete the container instead
$ref_to_delete = $hview->{parent};
}
my $task = $hyphash{$hyp}->{vcenter}->{conn}->get_view(mo_ref=>$ref_to_delete)->Destroy_Task();
$running_tasks{$task}->{task} = $task;
if ($vcenterautojoin) {
$running_tasks{$task}->{callback} = \&addhosttovcenter;
} elsif ($vcenterforceremove) {
$running_tasks{$task}->{callback} = \&delhost_callback;
}
$running_tasks{$task}->{conn} = $hyphash{$hyp}->{vcenter}->{conn};
$running_tasks{$task}->{data} = { depfun => $depfun, depargs => $depargs, conn=> $hyphash{$hyp}->{vcenter}->{conn}, connspec=>$connspec,hostview=>$hview,hypname=>$hyp,vcenter=>$vcenter };
return undef;
#The rest would be shorter/ideal, but seems to be confused a lot by stateless
#Maybe in a future VMWare technology level the following would work better
#than it does today
# my $task = $hview_->ReconnectHost_Task(cnxSpec=>$connspec);
# my $task = $hview->DisconnectHost_Task();
# $running_tasks{$task}->{task} = $task;
# $running_tasks{$task}->{callback} = \&disconnecthost_callback;
# $running_tasks{$task}->{conn} = $hyphash{$hyp}->{vcenter}->{conn};
# $running_tasks{$task}->{data} = { depfun => $depfun, depargs => $depargs, conn=> $hyphash{$hyp}->{vcenter}->{conn}, connspec=>$connspec,hostview=>$hview,hypname=>$hyp,vcenter=>$vcenter };
#ADDHOST
} else {
if ($hyphash{$hyp}->{offline}) {
xCAT::SvrUtils::sendmsg(": Failed to communicate with $hyp, vCenter reports it as in inventory but not connected and xCAT is set to not autojoin", $output_handler);
} else {
xCAT::SvrUtils::sendmsg([1,": Failed to communicate with $hyp, vCenter reports it as in inventory but not connected and xCAT is set to not autojoin"], $output_handler);
}
$hyphash{$hyp}->{conn} = undef;
return "failed";
}
}
}
unless ($vcenterautojoin) {
if ($hyphash{$hyp}->{offline}) {
xCAT::SvrUtils::sendmsg(": Failed to communicate with $hyp, vCenter does not have it in inventory and xCAT is set to not autojoin", $output_handler);
} else {
xCAT::SvrUtils::sendmsg([1,": Failed to communicate with $hyp, vCenter does not have it in inventory and xCAT is set to not autojoin"], $output_handler);
}
$hyphash{$hyp}->{conn} = undef;
return "failed";
}
#If still in function, haven't found any likely host entries, make a new one
unless ($hyphash{$hyp}->{offline}) {
eval {
$hyphash{$hyp}->{conn} = Vim->new(service_url=>"https://$hyp/sdk"); #Direct connect to install/check licenses
$hyphash{$hyp}->{conn}->login(user_name=>$hyphash{$hyp}->{username},password=>$hyphash{$hyp}->{password});
};
if ($@) {
xCAT::SvrUtils::sendmsg([1,": Failed to communicate with $hyp due to $@"], $output_handler);
$hyphash{$hyp}->{conn} = undef;
return "failed";
}
validate_licenses($hyp);
}
addhosttovcenter(undef,{
depfun => $depfun,
depargs => $depargs,
conn=>$hyphash{$hyp}->{vcenter}->{conn},
connspec=>$connspec,
hypname=>$hyp,
vcenter=>$vcenter,
});
}
sub addhosttovcenter {
my $task = shift;
my $args = shift;
my $hyp = $args->{hypname};
my $depfun = $args->{depfun};
my $depargs = $args->{depargs};
my $connspec = $args->{connspec};
my $vcenter = $args->{vcenter};
if ($task) {
my $state = $task->info->state->val;
if ($state eq 'error') {
die;
}
}
if ($hyphash{$args->{hypname}}->{offline}) { #let it stay offline
$hypready{$args->{hypname}}=1; #declare readiness
#enable_vmotion(hypname=>$args->{hypname},hostview=>$args->{hostview},conn=>$args->{conn});
$vcenterhash{$args->{vcenter}}->{goodhyps}->{$args->{hypname}} = 1;
if (defined $args->{depfun}) { #If a function is waiting for the host connect to go valid, call it
$args->{depfun}->($args->{depargs});
}
return;
}
if ($tablecfg{hypervisor}->{$hyp}->[0]->{cluster}) {
my $cluster = get_clusterview(clustname=>$tablecfg{hypervisor}->{$hyp}->[0]->{cluster},conn=>$hyphash{$hyp}->{vcenter}->{conn});
unless ($cluster) {
xCAT::SvrUtils::sendmsg([1,$tablecfg{hypervisor}->{$hyp}->[0]->{cluster}. " is not a known cluster to the vCenter server."], $output_handler);
$hypready{$hyp}=-1; #Declare impossiblility to be ready
return;
}
$task = $cluster->AddHost_Task(spec=>$connspec,asConnected=>1);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&connecthost_callback;
$running_tasks{$task}->{conn} = $hyphash{$hyp}->{vcenter}->{conn};
$running_tasks{$task}->{data} = { depfun => $depfun, depargs=> $depargs, conn=> $hyphash{$hyp}->{vcenter}->{conn}, connspec=>$connspec, cluster=>$cluster, hypname=>$hyp, vcenter=>$vcenter };
} else {
my $datacenter = validate_datacenter_prereqs($hyp);
unless ($datacenter) { return; }
my $hfolder = $datacenter->hostFolder; #$hyphash{$hyp}->{vcenter}->{conn}->find_entity_view(view_type=>'Datacenter',properties=>['hostFolder'])->hostFolder;
$hfolder = $hyphash{$hyp}->{vcenter}->{conn}->get_view(mo_ref=>$hfolder);
$task = $hfolder->AddStandaloneHost_Task(spec=>$connspec,addConnected=>1);
$running_tasks{$task}->{task} = $task;
$running_tasks{$task}->{callback} = \&connecthost_callback;
$running_tasks{$task}->{conn} = $hyphash{$hyp}->{vcenter}->{conn};
$running_tasks{$task}->{data} = { depfun => $depfun, depargs=> $depargs, conn=> $hyphash{$hyp}->{vcenter}->{conn}, connspec=>$connspec, foldview=>$hfolder, hypname=>$hyp, vcenter=>$vcenter };
}
#print Dumper @{$hyphash{$hyp}->{vcenter}->{conn}->find_entity_views(view_type=>'HostSystem',properties=>['runtime.connectionState'])};
}
sub validate_datacenter_prereqs {
my ($hyp) = @_;
my $datacenter;
if ($tablecfg{hypervisor}->{$hyp}->[0]->{datacenter}) {
$datacenter = $hyphash{$hyp}->{vcenter}->{conn}->find_entity_view(view_type => 'Datacenter', properties=>['hostFolder'],filter=>{name=>$tablecfg{hypervisor}->{$hyp}->[0]->{datacenter}});
unless ($datacenter) {
xCAT::SvrUtils::sendmsg([1,": Unable to find requested datacenter (hypervisor.datacenter for $hyp is ".$tablecfg{hypervisor}->{$hyp}->[0]->{datacenter}.")"], $output_handler);
return;
}
} else {
$datacenter = $hyphash{$hyp}->{vcenter}->{conn}->find_entity_view(view_type => 'Datacenter', properties=>['hostFolder']);
}
if (!defined $datacenter) {
my $vconn = $hyphash{$hyp}->{vcenter}->{conn};
my $root_folder = $vconn->get_view(mo_ref=>$vconn->get_service_content()->rootFolder);
$root_folder->CreateDatacenter(name=>'xcat-datacenter');
$datacenter = $hyphash{$hyp}->{vcenter}->{conn}->find_entity_view(view_type => 'Datacenter', properties=>['hostFolder']);
}
return $datacenter;
}
sub get_default_switch_for_hypervisor {
#This will make sure the default, implicit switch is in order in accordance
#with the configuration. If nothing specified, it just spits out vSwitch0
#if something specified, make sure it exists
#if it doesn't exist, and the syntax explains how to build it, build it
#return undef if something is specified, doesn't exist, and lacks instruction
my $hyp = shift;
my $defswitch = 'vSwitch0';
my $switchmembers;
if ($tablecfg{hypervisor}->{$hyp}->[0]->{defaultnet}) {
$defswitch = $tablecfg{hypervisor}->{$hyp}->[0]->{defaultnet};
($defswitch,$switchmembers) = split /=/,$defswitch,2;
my $vswitch;
my $hostview = $hyphash{$hyp}->{hostview};
foreach $vswitch (@{$hostview->config->network->vswitch}) {
if ($vswitch->name eq $defswitch) {
return $defswitch;
}
}
#If still here, means we need to build the switch
unless ($switchmembers) { return undef; } #No hope, no idea how to make it
return create_vswitch($hyp,$defswitch,split(/&/,$switchmembers));
} else {
return 'vSwitch0';
}
}
sub get_switchname_for_portdesc {
#Thisk function will examine all current switches to find or create a switch to match the described requirement
my $hyp = shift;
my $portdesc = shift;
my $description; #actual name to use for the virtual switch
if ($tablecfg{hypervisor}->{$hyp}->[0]->{netmap}) {
foreach (split /,/,$tablecfg{hypervisor}->{$hyp}->[0]->{netmap}) {
if (/^$portdesc=/) {
($description,$portdesc) = split /=/,$_,2;
last;
}
}
} else {
$description = 'vsw'.$portdesc;
}
unless ($description) {
xCAT::SvrUtils::sendmsg([1,": Invalid format for hypervisor.netmap detected for $hyp"], $output_handler);
return undef;
}
my %requiredports;
my %portkeys;
foreach (split /&/,$portdesc) {
$requiredports{$_}=1;
}
my $hostview = $hyphash{$hyp}->{hostview};
unless ($hostview) {
$hyphash{$hyp}->{hostview} = get_hostview(hypname=>$hyp,conn=>$hyphash{$hyp}->{conn}); #,properties=>['config','configManager']); #clustered can't run here, hyphash conn reference good
$hostview = $hyphash{$hyp}->{hostview};
}
foreach (@{$hostview->config->network->pnic}) {
if ($requiredports{$_->device}) { #We establish lookups both ways
$portkeys{$_->key}=$_->device;
delete $requiredports{$_->device};
}
}
if (keys %requiredports) {
xCAT::SvrUtils::sendmsg([1,":Unable to locate the following nics on $hyp: ".join(',',keys %requiredports)], $output_handler);
return undef;
}
my $foundmatchswitch;
my $cfgmismatch=0;
my $vswitch;
foreach $vswitch (@{$hostview->config->network->vswitch}) {
$cfgmismatch=0; #new switch, no sign of mismatch
foreach (@{$vswitch->pnic}) {
if ($portkeys{$_}) {
$foundmatchswitch=$vswitch->name;
delete $requiredports{$portkeys{$_}};
delete $portkeys{$_};
} else {
$cfgmismatch=1; #If this turns out to have anything, it is bad
}
}
if ($foundmatchswitch) { last; }
}
if ($foundmatchswitch) {
if ($cfgmismatch) {
xCAT::SvrUtils::sendmsg([1,": Aggregation mismatch detected, request nic is aggregated with a nic not requested"], $output_handler);
return undef;
}
unless (keys %portkeys) {
return $foundmatchswitch;
}
die "TODO: add physical nics to aggregation if requested";
} else {
return create_vswitch($hyp,$description,values %portkeys);
}
die "impossible occurance";
return undef;
}
sub create_vswitch {
my $hyp = shift;
my $description = shift;
my @ports = @_;
my $vswitch = HostVirtualSwitchBondBridge->new(
nicDevice=>\@ports
);
my $vswspec = HostVirtualSwitchSpec->new(
bridge=>$vswitch,
mtu=>1500,
numPorts=>64
);
my $hostview = $hyphash{$hyp}->{hostview};
my $netman=$hyphash{$hyp}->{conn}->get_view(mo_ref=>$hostview->configManager->networkSystem); #can't run in clustered mode, fine path..
$netman->AddVirtualSwitch(
vswitchName=>$description,
spec=>$vswspec
);
return $description;
}
sub scan_cluster_networks {
my $cluster = shift;
use Data::Dumper;
my $conn = $clusterhash{$cluster}->{conn};
my $cview = get_clusterview(clustname=>$cluster,conn=>$conn);
if (defined $cview->{network}) {
foreach (@{$cview->network}) {
my $nvw = $conn->get_view(mo_ref=>$_);
if (defined $nvw->name) {
$clusterhash{$cluster}->{nets}->{$nvw->name}=$_;
}
}
}
}
sub fixup_hostportgroup {
my $vlanspec = shift;
my $hyp = shift;
my %args = @_;
my $action = 'add';
if ($args{action}) { $action = $args{action} }
my $hostview = $hyphash{$hyp}->{hostview};
my $switchsupport = 0;
eval {
require xCAT::SwitchHandler;
$switchsupport = 1;
};
my $hypconn = $hyphash{$hyp}->{conn}; #this function can't work in clustered mode anyway, so this is appropriote.
my $vldata = $vlanspec;
my $switchname = get_default_switch_for_hypervisor($hyp);
my $pgname;
$vldata =~ s/=.*//; #TODO specify nic model with <blah>=model
if ($vldata =~ /:/) { #The config specifies a particular path in some way
$vldata =~ s/(.*)://;
$switchname = get_switchname_for_portdesc($hyp,$1);
$pgname=$switchname."-".$vldata;
} else { #Use the default vswitch per table config to connect this through, use the same name we did before to maintain compatibility
$pgname=$vldata;
}
my $netsys;
$hyphash{$hyp}->{pgnames}->{$vlanspec}=$pgname;
my $policy = HostNetworkPolicy->new();
unless ($hyphash{$hyp}->{nets}->{$pgname}) {
my $vlanid;
if (looks_like_number($vldata)) {
$vlanid = $vldata;
} elsif ($vldata =~ /trunk/) {
$vlanid=4095;
} elsif ($vldata =~ /vl(an)?(\d+)$/) {
$vlanid=$2;
} else {
$vlanid = 0;
}
if ($vlanid > 0 and $vlanid < 4095 and $switchsupport) {
my $switchtab = xCAT::Table->new("switch", -create=>0);
if ($switchtab) {
my $swent = $switchtab->getNodeAttribs($hyp, [qw/switch port/]);
if ($swent and $swent->{'switch'}) {
my $swh = new xCAT::SwitchHandler->new($swent->{'switch'});
my @vlids = $swh->get_vlan_ids();
if ($action eq 'add') {
unless (grep {$_ eq $vlanid} @vlids) {
$swh->create_vlan($vlanid);
}
$swh->add_ports_to_vlan($vlanid, $swent->{'port'});
} elsif ($action eq 'remove') {
$swh->remove_ports_from_vlan($vlanid, $swent->{'port'});
}
}
}
}
my $hostgroupdef = HostPortGroupSpec->new(
name =>$pgname,
vlanId=>$vlanid,
policy=>$policy,
vswitchName=>$switchname
);
unless ($netsys) {
$netsys = $hyphash{$hyp}->{conn}->get_view(mo_ref=>$hostview->configManager->networkSystem);
}
if ($action eq 'remove') {
$netsys->RemovePortGroup(pgName=>$pgname);
return;
} elsif ($action eq 'add') {
$netsys->AddPortGroup(portgrp=>$hostgroupdef);
}
#$hyphash{$hyp}->{nets}->{$netname}=1;
while ((not defined $hyphash{$hyp}->{nets}->{$pgname}) and sleep 1) { #we will only sleep if we know something will be waiting for
$hostview->update_view_data(); #pull in changes induced by previous activity
if (defined $hostview->{network}) { #We load the new object references
foreach (@{$hostview->network}) {
my $nvw = $hypconn->get_view(mo_ref=>$_);
if (defined $nvw->name) {
$hyphash{$hyp}->{nets}->{$nvw->name}=$_;
}
}
}
}
}
}
sub validate_network_prereqs {
my $nodes = shift;
my $hyp = shift;
my $hypconn = $hyphash{$hyp}->{conn}; #this function can't work in clustered mode anyway, so this is appropriote.
my $hostview = $hyphash{$hyp}->{hostview};
if ($hostview) {
$hostview->update_view_data(); #pull in changes induced by previous activity
} else {
$hyphash{$hyp}->{hostview} = get_hostview(hypname=>$hyp,conn=>$hyphash{$hyp}->{conn}); #,properties=>['config','configManager','network']);
$hostview = $hyphash{$hyp}->{hostview};
}
my $node;
my $method;
my $location;
if (defined $hostview->{network}) {
foreach (@{$hostview->network}) {
my $nvw = $hypconn->get_view(mo_ref=>$_);
if (defined $nvw->name) {
$hyphash{$hyp}->{nets}->{$nvw->name}=$_;
}
}
}
foreach $node (@$nodes) {
my @networks = split /,/,$tablecfg{vm}->{$node}->[0]->{nics};
foreach (@networks) {
fixup_hostportgroup($_, $hyp);
}
}
return 1;
}
sub refreshclusterdatastoremap {
my $cluster = shift;
my $conn=$clusterhash{$cluster}->{conn};
my $cview = get_clusterview(clustname=>$cluster,conn=>$conn);
if (defined $cview->{datastore}) {
foreach (@{$cview->datastore}) {
my $dsv = $conn->get_view(mo_ref=>$_);
if (defined $dsv->info->{nas}) {
if ($dsv->info->nas->type eq 'NFS') {
my $mnthost = $dsv->info->nas->remoteHost;
# my $mnthost = inet_aton($dsv->info->nas->remoteHost);
# if ($mnthost) {
# $mnthost = inet_ntoa($mnthost);
# } else {
# $mnthost = $dsv->info->nas->remoteHost;
# xCAT::SvrUtils::sendmsg([1,"Unable to resolve VMware specified host '".$dsv->info->nas->remoteHost."' to an address, problems may occur"], $output_handler);
# }
$clusterhash{$cluster}->{datastoremap}->{"nfs://".$mnthost.$dsv->info->nas->remotePath}=$dsv->info->name;
$clusterhash{$cluster}->{datastoreurlmap}->{$dsv->info->name} = "nfs://".$mnthost.$dsv->info->nas->remotePath; #save off a suitable URL if needed
$clusterhash{$cluster}->{datastorerefmap}->{"nfs://".$mnthost.$dsv->info->nas->remotePath}=$_;
} #TODO: care about SMB
}elsif(defined $dsv->info->{vmfs}){
my $name = $dsv->info->vmfs->name;
$clusterhash{$cluster}->{datastoremap}->{"vmfs://".$name} = $dsv->info->name;
$clusterhash{$cluster}->{datastoreurlmap}->{$dsv->info->name} = "vmfs://".$name;
$clusterhash{$cluster}->{datastorerefmap}->{"vmfs://".$name} = $_;
}
}
}
#that's... about it... not doing any of the fancy mounting and stuff, if you do it cluster style, you are on your own. It's simply too terrifying to try to fixup
#a whole cluster instead of chasing one host, a whole lot slower. One would hope vmware would've done this, but they don't
}
sub validate_datastore_prereqs {
my $hyp = $_[1];
lockbyname($hyp.".datastores");
$@="";
my $rc;
eval { $rc=validate_datastore_prereqs_inlock(@_); };
unlockbyname($hyp.".datastores");
if ($@) { die $@; }
return $rc;
}
sub validate_datastore_prereqs_inlock {
my $nodes = shift;
my $hyp = shift;
my $newdatastores = shift; # a hash reference of URLs to afflicted nodes outside of table space
my $hypconn = $hyphash{$hyp}->{conn};
my $hostview = $hyphash{$hyp}->{hostview};
unless ($hostview) {
$hyphash{$hyp}->{hostview} = get_hostview(hypname=>$hyp,conn=>$hypconn); #,properties=>['config','configManager']);
$hostview = $hyphash{$hyp}->{hostview};
}
my $node;
my $method;
my $location;
# get all of the datastores that are currently available on this node.
# and put them into a hash
if (defined $hostview->{datastore}) { # only iterate if it exists
foreach (@{$hostview->datastore}) {
my $dsv = $hypconn->get_view(mo_ref=>$_);
if (defined $dsv->info->{nas}) {
if ($dsv->info->nas->type eq 'NFS') {
my $mnthost = inet_aton($dsv->info->nas->remoteHost);
if ($mnthost) {
$mnthost = inet_ntoa($mnthost);
} else {
$mnthost = $dsv->info->nas->remoteHost;
xCAT::SvrUtils::sendmsg([1,"Unable to resolve VMware specified host '".$dsv->info->nas->remoteHost."' to an address, problems may occur"], $output_handler);
}
$hyphash{$hyp}->{datastoremap}->{"nfs://".$mnthost.$dsv->info->nas->remotePath}=$dsv->info->name;
$hyphash{$hyp}->{datastoreurlmap}->{$dsv->info->name} = "nfs://".$mnthost.$dsv->info->nas->remotePath;
$hyphash{$hyp}->{datastorerefmap}->{"nfs://".$mnthost.$dsv->info->nas->remotePath}=$_;
} #TODO: care about SMB
}elsif(defined $dsv->info->{vmfs}){
my $name = $dsv->info->vmfs->name;
$hyphash{$hyp}->{datastoremap}->{"vmfs://".$name} = $dsv->info->name;
$hyphash{$hyp}->{datastoreurlmap}->{$dsv->info->name} = "vmfs://".$name;
$hyphash{$hyp}->{datastorerefmap}->{"vmfs://".$name} = $_;
}
}
}
my $refresh_names=0;
# now go through the nodes and make sure that we have matching datastores.
# E.g.: if its NFS, then mount it (if not mounted)
# E.g.: if its VMFS, then create it if not created already. Note: VMFS will persist on
# machine reboots, unless its destroyed by being overwritten.
foreach $node (@$nodes) {
my @storage = split /,/,$tablecfg{vm}->{$node}->[0]->{storage};
if ($tablecfg{vm}->{$node}->[0]->{cfgstore}) {
push @storage,$tablecfg{vm}->{$node}->[0]->{cfgstore};
}
foreach (@storage) { #TODO: merge this with foreach loop below. Here we could build onto $newdatastores instead, for faster operation at scale
s/=.*//; #remove device type information from configuration
s/\/$//; #Strip trailing slash if specified, to align to VMware semantics
if (/:\/\//) {
($method,$location) = split /:\/\//,$_,2;
if($method =~ /nfs/){
# go through and see if NFS is mounted, if not, then mount it.
(my $server, my $path) = split /\//,$location,2;
$server =~ s/:$//; #remove a : if someone put it in out of nfs mount habit
my $servern = inet_aton($server);
unless ($servern) {
xCAT::SvrUtils::sendmsg([1,": Unable to resolve '$server' to an address, check vm.cfgstore/vm.storage"], $output_handler);
return 0;
}
$server = inet_ntoa($servern);
my $uri = "nfs://$server/$path";
unless ($hyphash{$hyp}->{datastoremap}->{$uri}) { #If not already there, must mount it
unless ($datastoreautomount) {
xCAT::SvrUtils::sendmsg([1,": $uri is not currently accessible at the given location and automount is disabled in site table"], $output_handler,$node);
return 0;
}
$refresh_names=1;
($hyphash{$hyp}->{datastoremap}->{$uri},$hyphash{$hyp}->{datastorerefmap}->{$uri})=mount_nfs_datastore($hostview,$location);
$hyphash{$hyp}->{datastoreurlmap}->{$hyphash{$hyp}->{datastoremap}->{$uri}} = $uri;
}
}elsif($method =~ /vmfs/){
(my $name, undef) = split /\//,$location,2;
$name =~ s/:$//; #remove a : if someone put it in for some reason.
my $uri = "vmfs://$name";
# check and see if this vmfs is on the node.
unless ($hyphash{$hyp}->{datastoremap}->{$uri}) { #If not already there, try creating it.
unless ($datastoreautomount) {
xCAT::SvrUtils::sendmsg([1,": $uri is not currently accessible at the given location and automount is disabled in site table"], $output_handler,$node);
return 0;
}
$refresh_names=1;
($hyphash{$hyp}->{datastoremap}->{$uri},$hyphash{$hyp}->{datastorerefmap}->{$uri})=create_vmfs_datastore($hostview,$name,$hyp);
unless($hyphash{hyp}->{datastoremap}->{$uri}){ return 0; }
$hyphash{$hyp}->{datastoreurlmap}->{$hyphash{$hyp}->{datastoremap}->{$uri}} = $uri;
}
}else{
xCAT::SvrUtils::sendmsg([1,": $method is unsupported at this time (nfs would be)"], $output_handler,$node);
return 0;
}
} else {
xCAT::SvrUtils::sendmsg([1,": $_ not supported storage specification for ESX plugin,\n\t'nfs://<server>/<path>'\n\t\tor\n\t'vmfs://<vmfs>'\n only currently supported vm.storage supported for ESX at the moment"], $output_handler,$node);
return 0;
} #TODO: raw device mapping, VMFS via iSCSI, VMFS via FC?
}
}
# newdatastores are for migrations or changing vms.
# TODO: make this work for VMFS. Right now only NFS.
if (ref $newdatastores) {
foreach (keys %$newdatastores) {
my $origurl=$_;
s/\/$//; #Strip trailing slash if specified, to align to VMware semantics
if (/:\/\//) {
($method,$location) = split /:\/\//,$_,2;
if($method =~ /nfs/){
(my $server, my $path) = split /\//,$location,2;
$server =~ s/:$//; #remove a : if someone put it in out of nfs mount habit
my $servern = inet_aton($server);
unless ($servern) {
xCAT::SvrUtils::sendmsg([1,": Unable to resolve '$server' to an address, check vm.cfgstore/vm.storage"], $output_handler);
return 0;
}
$server = inet_ntoa($servern);
my $uri = "nfs://$server/$path";
unless ($method =~ /nfs/) {
foreach (@{$newdatastores->{$_}}) {
xCAT::SvrUtils::sendmsg([1,": $method is unsupported at this time (nfs would be)"], $output_handler,$_);
}
return 0;
}
unless ($hyphash{$hyp}->{datastoremap}->{$uri}) { #If not already there, must mount it
unless ($datastoreautomount) {
xCAT::SvrUtils::sendmsg([1,":) $uri is not currently accessible at the given location and automount is disabled in site table"], $output_handler,$node);
return 0;
}
$refresh_names=1;
($hyphash{$hyp}->{datastoremap}->{$uri},$hyphash{$hyp}->{datastorerefmap}->{$uri})=mount_nfs_datastore($hostview,$location);
}
$hyphash{$hyp}->{datastoreurlmap}->{$hyphash{$hyp}->{datastoremap}->{$uri}} = $uri;
$hyphash{$hyp}->{datastoremap}->{$origurl}=$hyphash{$hyp}->{datastoremap}->{$uri}; #we track both the uri xCAT expected and the one vCenter actually ended up with
$hyphash{$hyp}->{datastorerefmap}->{$origurl}=$hyphash{$hyp}->{datastorerefmap}->{$uri};
}elsif($method =~ /vmfs/){
(my $name, undef) = split /\//,$location,2;
$name =~ s/:$//; #remove a : if someone put it in for some reason.
my $uri = "vmfs://$name";
unless ($hyphash{$hyp}->{datastoremap}->{$uri}) { #If not already there, it should be!
unless ($datastoreautomount) {
xCAT::SvrUtils::sendmsg([1,": $uri is not currently accessible at the given location and automount is disabled in site table"], $output_handler,$node);
return 0;
}
$refresh_names=1;
($hyphash{$hyp}->{datastoremap}->{$uri},$hyphash{$hyp}->{datastorerefmap}->{$uri})=create_vmfs_datastore($hostview,$name,$hyp);
unless($hyphash{hyp}->{datastoremap}->{$uri}){ return 0; }
}
$hyphash{$hyp}->{datastoreurlmap}->{$hyphash{$hyp}->{datastoremap}->{$uri}} = $uri;
$hyphash{$hyp}->{datastoremap}->{$origurl}=$hyphash{$hyp}->{datastoremap}->{$uri};
$hyphash{$hyp}->{datastorerefmap}->{$origurl}=$hyphash{$hyp}->{datastorerefmap}->{$uri};
}else{
print "$method: not NFS and not VMFS here!\n";
}
} else {
my $datastore=$_;
foreach my $ds (@{$newdatastores->{$_}}) {
xCAT::SvrUtils::sendmsg([1,": $datastore not supported storage specification for ESX plugin, 'nfs://<server>/<path>' only currently supported vm.storage supported for ESX at the moment"], $output_handler,$ds);
}
return 0;
} #TODO: raw device mapping, VMFS via iSCSI, VMFS via FC, VMFS on same local drive?
}
}
if ($refresh_names) { #if we are in a vcenter context, vmware can rename a datastore behind our backs immediately after adding
$hostview->update_view_data();
if (defined $hostview->{datastore}) { # only iterate if it exists
foreach (@{$hostview->datastore}) {
my $dsv = $hypconn->get_view(mo_ref=>$_);
if (defined $dsv->info->{nas}) {
if ($dsv->info->nas->type eq 'NFS') {
my $mnthost = inet_aton($dsv->info->nas->remoteHost);
if ($mnthost) {
$mnthost = inet_ntoa($mnthost);
} else {
$mnthost = $dsv->info->nas->remoteHost;
xCAT::SvrUtils::sendmsg([1,"Unable to resolve VMware specified host '".$dsv->info->nas->remoteHost."' to an address, problems may occur"], $output_handler);
}
$hyphash{$hyp}->{datastoremap}->{"nfs://".$mnthost.$dsv->info->nas->remotePath}=$dsv->info->name;
$hyphash{$hyp}->{datastoreurlmap}->{$dsv->info->name} = "nfs://".$mnthost.$dsv->info->nas->remotePath;
$hyphash{$hyp}->{datastorerefmap}->{"nfs://".$mnthost.$dsv->info->nas->remotePath}=$_;
} #TODO: care about SMB
} #TODO: care about VMFS
}
}
}
return 1;
}
sub getlabel_for_datastore {
my $method = shift;
my $location = shift;
$location =~ s/\//_/g;
$location= $method.'_'.$location;
#VMware has a 42 character limit, we will start mangling to get under 42.
#Will try to preserve as much informative detail as possible, hence several conditionals instead of taking the easy way out
if (length($location) > 42) {
$location =~ s/nfs_//; #Ditch unique names for different protocols to the same path, seems unbelievably unlikely
}
if (length($location) > 42) {
$location =~ s/\.//g; #Next, ditch host delimiter, it is unlikely that hosts will have unique names if their dots are removed
}
if (length($location) > 42) {
$location =~ s/_//g; #Next, ditch path delimiter, it is unlikely that two paths will happen to look the same without delimiters
}
if (length($location) > 42) { #finally, replace the middle with ellipsis
substr($location,20,-20,'..');
}
return $location;
}
sub mount_nfs_datastore {
my $hostview = shift;
my $location = shift;
my $server;
my $path;
unless ($datastoreautomount) {
die "automount of VMware datastores is disabled in site configuration, not continuing";
}
($server,$path) = split /\//,$location,2;
$location = getlabel_for_datastore('nfs',$location);
my $nds = HostNasVolumeSpec->new(accessMode=>'readWrite',
remoteHost=>$server,
localPath=>$location,
remotePath=>"/".$path);
my $dsmv = $hostview->{vim}->get_view(mo_ref=>$hostview->configManager->datastoreSystem);
my $dsref;
eval {
$dsref=$dsmv->CreateNasDatastore(spec=>$nds);
};
if ($@) {
die "$@" unless $@ =~ m/Fault detail: DuplicateNameFault/;
die "esx plugin: a datastore was discovered with the same name referring to a different nominatum- cannot continue\n$@"
unless &match_nfs_datastore($server,"/$path",$hostview->{vim});
}
return ($location,$dsref);
}
# create a VMFS data store on a node so that VMs can live locally instead of NFS
sub create_vmfs_datastore {
my $hostview = shift; # VM object
my $name = shift; # name of storage we wish to create.
my $hyp = shift;
unless ($datastoreautomount) {
die "automount of VMware datastores is disabled in site configuration, not continuing";
}
# call some VMware API here to create
my $hdss = $hostview->{vim}->get_view(mo_ref=>$hostview->configManager->datastoreSystem);
my $diskList = $hdss->QueryAvailableDisksForVmfs();
my $count = scalar(@$diskList); # get the number of disks available for formatting.
unless($count >0){
#die "No disks are available to create VMFS volume for $name";
$output_handler->({error=>["No disks are available on $hyp to create VMFS volume for $name"],errorcode=>1});
return 0;
}
foreach my $disk(@$diskList){
my $options = $hdss->QueryVmfsDatastoreCreateOptions(devicePath => $disk->devicePath);
@$options[0]->spec->vmfs->volumeName($name);
my $newDatastore = $hdss->CreateVmfsDatastore(spec => @$options[0]->spec );
#return $newDatastore;
# create it on the first disk we see.
return ($name, $newDatastore);
}
return 0;
}
sub build_more_info{
die("TODO: fix this function if called");
print "Does this acually get called????**********************************\n";
my $noderange=shift;
my $callback=shift;
my $vmtab = xCAT::Table->new("vm");
my @moreinfo=();
unless ($vmtab) {
$callback->({data=>["Cannot open mp table"]});
return @moreinfo;
}
my %mpa_hash=();
foreach my $node (@$noderange) {
my $ent=$vmtab->getNodeAttribs($node,['mpa', 'id']);
if (defined($ent->{mpa})) { push @{$mpa_hash{$ent->{mpa}}{nodes}}, $node;}
else {
$callback->({data=>["no mpa defined for node $node"]});
return @moreinfo;
}
if (defined($ent->{id})) { push @{$mpa_hash{$ent->{mpa}}{ids}}, $ent->{id};}
else { push @{$mpa_hash{$ent->{mpa}}{ids}}, "";}
}
foreach (keys %mpa_hash) {
push @moreinfo, "\[$_\]\[" . join(',',@{$mpa_hash{$_}{nodes}}) ."\]\[" . join(',',@{$mpa_hash{$_}{ids}}) . "\]";
}
return \@moreinfo;
}
sub copycd {
my $request = shift;
my $doreq = shift;
my $distname = "";
my $path;
my $arch;
my $darch;
my $installroot;
$installroot = "/install";
#my $sitetab = xCAT::Table->new('site');
#if($sitetab){
#(my $ref) = $sitetab->getAttribs({key => 'installdir'}, 'value');
my @entries = xCAT::TableUtils->get_site_attribute("installdir");
my $t_entry = $entries[0];
if ( defined($t_entry) ) {
$installroot = $t_entry;
}
#}
@ARGV = @{$request->{arg}};
GetOptions(
'n=s' => \$distname,
'a=s' => \$arch,
'm=s' => \$path
);
# run a few tests to see if the copycds should use this plugin
unless ($path){
# can't use us cause we need a path and you didn't provide one!
return;
}
if( $distname and $distname !~ /^esx/ ){
# we're for esx, so if you didn't specify that its not us!
return;
}
my $found = 0;
if (-r $path . "/README" and -r $path . "/build_number" and -d $path . "/VMware" and -r $path . "/packages.xml") { #We have a probable new style ESX media
open(LINE,$path."/packages.xml");
my $product;
my $version;
while (<LINE>) {
if (/roductLineId>([^<]*)<\/Prod/) {
$product = $1;
}
if (/ersion>([^<]*)<\/version/) {
$version = $1;
$version =~ s/\.0$//;
}
if (/arch>([^>]*)<\/arch/) {
unless ($darch and $darch =~ /x86_64/) { #prefer to be characterized as x86_64
$darch = $1;
$arch = $1;
}
}
}
close(LINE);
if ($product and $version) {
unless ($distname) { $distname = $product.$version; }
$found = 1;
}
} elsif (-r $path . "/README" and -r $path . "/open_source_licenses.txt" and -d $path . "/VMware") { #Candidate to be ESX 3.5
open(LINE,$path."/README");
while(<LINE>) {
if (/VMware ESX Server 3.5\s*$/) {
$darch ='x86';
$arch = 'x86';
unless ($distname) { $distname = 'esx3.5'; }
$found = 1;
last;
}
}
close(LINE);
} elsif (-r $path . "/README.txt" and -r $path . "/vmkernel.gz"){
# its an esxi dvd!
# if we got here its probably ESX they want to copy
my $line;
my $darch;
open(LINE, $path . "/README.txt") or die "couldn't open!";
while($line = <LINE>){
chomp($line);
if($line =~ /VMware ESXi(?: version)? 4\.(\d+)/){
$darch = "x86_64";
unless ($distname) {
$distname = "esxi4";
if ($1) {
$distname .= '.'.$1;
}
}
$found = 1;
if( $arch and $arch ne $darch){
xCAT::SvrUtils::sendmsg([1, "Requested distribution architecture $arch, but media is $darch"], $output_handler);
return;
}
$arch = $darch;
last; # we found our distro! end this loop madness.
}
}
close(LINE);
unless($found){
xCAT::SvrUtils::sendmsg([1,"I don't recognize this VMware ESX DVD"], $output_handler);
return; # doesn't seem to be a valid DVD or CD
}
} elsif (-r $path . "/vmkernel.gz" and -r $path . "/isolinux.cfg"){
open(LINE,$path . "/isolinux.cfg");
while (<LINE>) {
if (/ThinESX Installer/) {
$darch = 'x86';
$arch='x86';
unless ($distname) { $distname='esxi3.5'; }
$found=1;
last;
}
}
close(LINE);
} elsif (-r $path . "/vmware-esx-base-readme") {
open(LINE,$path."/vmware-esx-base-readme");
while (<LINE>) {
if (/VMware ESXi 5\.0/) {
$darch="x86_64";
$arch="x86_64";
unless ($distname) { $distname='esxi5'; }
$found=1;
last;
}
if (/VMware ESXi 5\.1/) {
$darch="x86_64";
$arch="x86_64";
unless ($distname) { $distname='esxi5.1'; }
$found=1;
last;
}
if (/VMware ESXi 5\.5/) {
$darch="x86_64";
$arch="x86_64";
unless ($distname) { $distname='esxi5.5'; }
$found=1;
last;
}
}
}
unless ($found) { return; } #not our media
if ($::XCATSITEVALS{osimagerequired}){
my ($nohaveimages,$errstr)=xCAT::SvrUtils->update_tables_with_templates($distname, $arch,"","",checkonly=>1);
if ($nohaveimages) {
$output_handler->({error => "No Templates found to support $distname($arch)",errorcode=>2});
}
}
xCAT::SvrUtils::sendmsg("Copying media to $installroot/$distname/$arch/", $output_handler);
my $omask = umask 0022;
mkpath("$installroot/$distname/$arch");
umask $omask;
my $rc;
my $reaped = 0;
$SIG{INT} = $SIG{TERM} = sub {
foreach(@cpiopid){
kill 2, $_;
}
if ($path) {
chdir("/");
system("umount $path");
}
};
my $KID;
chdir $path;
my $numFiles = `find . -print | wc -l`;
my $child = open($KID, "|-");
unless (defined $child)
{
xCAT::SvrUtils::sendmsg([1,"Media copy operation fork failure"], $output_handler);
return;
}
if ($child)
{
push @cpiopid, $child;
my @finddata = `find .`;
for (@finddata)
{
print $KID $_;
}
close($KID);
$rc = $?;
}
else
{
nice 10;
my $c = "nice -n 20 cpio -vdump $installroot/$distname/$arch";
my $k2 = open(PIPE, "$c 2>&1 |") ||
xCAT::SvrUtils::sendmsg([1,"Media copy operation fork failure"], $output_handler);
push @cpiopid, $k2;
my $copied = 0;
my ($percent, $fout);
while(<PIPE>){
next if /^cpio:/;
$percent = $copied / $numFiles;
$fout = sprintf "%0.2f%%", $percent * 100;
$output_handler->({sinfo => "$fout"});
++$copied;
}
exit;
}
# let everyone read it
#chdir "/tmp";
chmod 0755, "$installroot/$distname/$arch";
if ($distname =~ /esxi5/) { #going to tweak boot.cfg for install and default stateless case
if (! -r "$installroot/$distname/$arch/boot.cfg.stateless") {
copy("$installroot/$distname/$arch/boot.cfg","$installroot/$distname/$arch/boot.cfg.stateless");
my $bootcfg;
open($bootcfg,"<","$installroot/$distname/$arch/boot.cfg");
my @bootcfg = <$bootcfg>;
close($bootcfg);
foreach (@bootcfg) { #no point in optimizing trivial, infrequent code, readable this way
s!kernel=/!kernel=!; # remove leading /
s!modules=/!modules=!; #remove leading /
s!--- /!--- !g; #remove all the 'absolute' slashes
}
open($bootcfg,">","$installroot/$distname/$arch/boot.cfg.install");
foreach (@bootcfg) {
if (/^modules=/ and $_ !~ /xcatmod.tgz/ and not $::XCATSITEVALS{xcatesximoddisable}) {
chomp();
s! *\z! --- xcatmod.tgz\n!;
}
print $bootcfg $_;
}
close($bootcfg);
foreach (@bootcfg) { #no point in optimizing trivial, infrequent code, readable this way
s/runweasel//; #don't run the installer in stateless mode
s!--- imgdb.tgz!!; #don't need the imgdb for stateless
s!--- imgpayld.tgz!!; #don't need the boot payload since we aren't installing
s!--- tools.t00!!; #tools could be useful, but for now skip the memory requirement
s!--- weaselin.i00!!; #and also don't need the weasel install images if... not installing
if (/^modules=/ and $_ !~ /xcatmod.tgz/ and not $::XCATSITEVALS{xcatesximoddisable}) {
chomp();
s! *\z! --- xcatmod.tgz\n!;
}
s!Loading ESXi installer!xCAT is loading ESXi stateless!;
}
open($bootcfg,">","$installroot/$distname/$arch/boot.cfg.stateless");
foreach (@bootcfg) {
print $bootcfg $_;
}
close($bootcfg);
if (grep /LSIProvi.v00/,@bootcfg and ! -r "$installroot/$distname/$arch/LSIProvi.v00" and -r "$installroot/$distname/$arch/lsiprovi.v00") { #there is media with LSIProv.v00 expected, but the install media was mal-constructed, fix it
move("$installroot/$distname/$arch/lsiprovi.v00","$installroot/$distname/$arch/LSIProvi.v00");
}
}
}
if ($rc != 0){
xCAT::SvrUtils::sendmsg([1,"Media copy operation failed, status $rc"], $output_handler);
}else{
xCAT::SvrUtils::sendmsg("Media copy operation successful", $output_handler);
my @ret=xCAT::SvrUtils->update_tables_with_templates($distname, $arch);
if ($ret[0] != 0) {
xCAT::SvrUtils::sendmsg("Error when updating the osimage tables: " . $ret[1], $output_handler);
}
}
}
sub makecustomizedmod {
my $osver = shift;
my $dest = shift;
if ($::XCATSITEVALS{xcatesximoddisable}) { return 1; }
my $modname;
if ($osver =~ /esxi4/) { #want more descriptive name,but don't break esxi4 setups.
$modname="mod.tgz";
} else {
$modname="xcatmod.tgz";
}
# if it already exists, do not overwrite it because it may be someone
# else's custom image
if(-f "$dest/$modname"){ return 1; }
my $passtab = xCAT::Table->new('passwd');
my $tmp;
my $password;
if ($passtab) {
($tmp) = $passtab->getAttribs({'key'=>'vmware'},'username','password');
if (defined($tmp)) {
$password = $tmp->{password};
}
}
unless ($password) {
xCAT::SvrUtils::sendmsg([1,": Unable to find a password entry for esxi in passwd table"], $output_handler);
return 0;
}
mkpath("/tmp/xcat");
my $tempdir = tempdir("/tmp/xcat/esxmodcustXXXXXXXX");
my $shadow;
mkpath($tempdir."/etc/");
my $oldmask=umask(0077);
open($shadow,">",$tempdir."/etc/shadow");
$password = crypt($password,'$1$'.xCAT::Utils::genpassword(8));
my $dayssince1970 = int(time()/86400); #Be truthful about /etc/shadow
my @otherusers = qw/nobody nfsnobody dcui daemon/;
if ($osver =~ /esxi4/) {
push @otherusers,"vimuser";
} elsif ($osver =~ /esxi5/) {
push @otherusers,"vpxuser";
}
print $shadow "root:$password:$dayssince1970:0:99999:7:::\n";
foreach (@otherusers) {
print $shadow "$_:*:$dayssince1970:0:99999:7:::\n";
}
close($shadow);
umask($oldmask);
if ($osver =~ /esxi4/ and -e "$::XCATROOT/share/xcat/netboot/esxi/38.xcat-enableipv6") {
mkpath($tempdir."/etc/vmware/init/init.d");
copy( "$::XCATROOT/share/xcat/netboot/esxi/38.xcat-enableipv6",$tempdir."/etc/vmware/init/init.d/38.xcat-enableipv6");
} elsif ($osver =~ /esxi5/ and -e "$::XCATROOT/share/xcat/netboot/esxi/xcat-ipv6.json") {
mkpath($tempdir."/usr/libexec/jumpstart/plugins/");
copy( "$::XCATROOT/share/xcat/netboot/esxi/xcat-ipv6.json",$tempdir."/usr/libexec/jumpstart/plugins/xcat-ipv6.json");
}
if ($osver =~ /esxi4/ and -e "$::XCATROOT/share/xcat/netboot/esxi/47.xcat-networking") {
copy( "$::XCATROOT/share/xcat/netboot/esxi/47.xcat-networking",$tempdir."/etc/vmware/init/init.d/47.xcat-networking");
} elsif ($osver =~ /esxi5/ and -e "$::XCATROOT/share/xcat/netboot/esxi/39.ipv6fixup") {
mkpath($tempdir."/etc/init.d");
copy( "$::XCATROOT/share/xcat/netboot/esxi/39.ipv6fixup",$tempdir."/etc/init.d/39.ipv6fixup");
chmod(0755,"$tempdir/etc/init.d/39.ipv6fixup");
}
if ($osver =~ /esxi5/ and -e "$::XCATROOT/share/xcat/netboot/esxi/48.esxifixup") {
mkpath($tempdir."/etc/init.d");
copy( "$::XCATROOT/share/xcat/netboot/esxi/48.esxifixup",$tempdir."/etc/init.d/48.esxifixup");
chmod(0755,"$tempdir/etc/init.d/48.esxifixup");
}
if (-e "$::XCATROOT/share/xcat/netboot/esxi/xcatsplash") {
mkpath($tempdir."/etc/vmware/");
copy( "$::XCATROOT/share/xcat/netboot/esxi/xcatsplash",$tempdir."/etc/vmware/welcome");
}
my $dossh=0;
if (-r "/root/.ssh/id_rsa.pub") {
$dossh=1;
my $umask = umask(0077);#don't remember if dropbear is picky, but just in case
if ($osver =~ /esxi4/) { #esxi4 used more typical path
mkpath($tempdir."/.ssh");
copy("/root/.ssh/id_rsa.pub",$tempdir."/.ssh/authorized_keys");
} elsif ($osver =~ /esxi5/) { #weird path to keys
mkpath($tempdir."/etc/ssh/keys-root");
copy("/root/.ssh/id_rsa.pub",$tempdir."/etc/ssh/keys-root/authorized_keys");
}
umask($umask);
}
my $tfile;
mkpath($tempdir."/var/run/vmware");
open $tfile,">",$tempdir."/var/run/vmware/show-tech-support-login";
close($tfile);
#TODO: auto-enable ssh and request boot-time customization rather than on-demand?
require Cwd;
my $dir=Cwd::cwd();
chdir($tempdir);
if (-e "$dest/$modname") {
unlink("$dest/$modname");
}
if ($dossh and $osver =~ /esxi4/) {
system("tar czf $dest/$modname * .ssh");
} else {
system("tar czf $dest/$modname *");
}
chdir($dir);
rmtree($tempdir);
return 1;
}
sub getplatform {
my $os = shift;
if ($os =~ /esxi/) {
return "esxi";
}
return $os;
}
sub esxi_kickstart_from_template {
my %args=@_;
my $installdir = "/install";
if ($::XCATSITEVALS{installdir}) { $installdir = $::XCATSITEVALS{installdir}; }
my $plat = getplatform($args{os});
my $template = xCAT::SvrUtils::get_tmpl_file_name("$installdir/custom/install/$plat",$args{profile},$args{os},$args{arch},$args{os});
unless ($template) {
$template = xCAT::SvrUtils::get_tmpl_file_name("$::XCATROOT/share/xcat/install/$plat",$args{profile},$args{os},$args{arch},$args{os});
}
my $tmperr;
if (-r "$template") {
$tmperr=xCAT::Template->subvars($template,"$installdir/autoinst/".$args{node},$args{node},undef);
} else {
$tmperr="Unable to find template in /install/custom/install/$plat or $::XCATROOT/share/xcat/install/$plat (for $args{profile}/$args{os}/$args{arch} combination)";
}
if ($tmperr) {
xCAT::SvrUtils::sendmsg([1,$tmperr], $output_handler,$args{node});
}
}
sub mkinstall {
return mkcommonboot("install",@_);
}
sub mknetboot {
return mkcommonboot("stateless",@_);
}
sub merge_esxi5_append {
my $tmpl = shift;
my $append = shift;
my $outfile = shift;
my $in;
my $out;
open($in,"<",$tmpl);
open($out,">",$outfile);
my $line;
while ($line = <$in>) {
if ($line =~ /kernelopt=/) {
chomp($line);
$line .= $append."\n";
#if ($line =~ /modules=b.b00/) {
# $line =~ s/modules=b.b00/modules=b.b00 $append/;
}
unless ($line =~ /^prefix=/) {
print $out $line;
}
}
}
sub mkcommonboot {
my $bootmode = shift;
my $req = shift;
my $doreq = shift;
my $globaltftpdir = "/tftpboot";
my @nodes = @{$req->{node}};
my $ostab = xCAT::Table->new('nodetype');
#my $sitetab = xCAT::Table->new('site');
my $bptab = xCAT::Table->new('bootparams',-create=>1);
my $installroot = "/install";
#if ($sitetab){
#(my $ref) = $sitetab->getAttribs({key => 'installdir'}, 'value');
my @entries = xCAT::TableUtils->get_site_attribute("installdir");
my $t_entry = $entries[0];
if ( defined($t_entry) ) {
$installroot = $t_entry;
}
#($ref) = $sitetab->getAttribs({key => 'tftpdir'}, 'value');
@entries = xCAT::TableUtils->get_site_attribute("tftpdir");
$t_entry = $entries[0];
if ( defined($t_entry) ) {
$globaltftpdir = $t_entry;
}
#}
my %donetftp=();
my $bpadds = $bptab->getNodesAttribs(\@nodes,['addkcmdline']);
my $nodehmtab = xCAT::Table->new('nodehm',-create=>0);
my $serialconfig;
if ($nodehmtab) {
$serialconfig = $nodehmtab->getNodesAttribs(\@nodes,['serialport','serialspeed']);
}
my $restab = xCAT::Table->new('noderes',-create=>0);
my $resents;
if ($restab) {
$resents = $restab->getNodesAttribs(\@nodes,['tftpdir','nfsserver']);
}
my %tablecolumnsneededforaddkcmdline;
my %nodesubdata;
foreach my $key (keys %$bpadds){ #First, we identify all needed table.columns needed to aggregate database call
my $add = $bpadds->{$key}->[0]->{addkcmdline};
next if ! defined $add;
while ($add =~ /#NODEATTRIB:([^:#]+):([^:#]+)#/) {
push @{$tablecolumnsneededforaddkcmdline{$1}},$2;
$add =~ s/#NODEATTRIB:([^:#]+):([^:#]+)#//;
}
}
foreach my $table (keys %tablecolumnsneededforaddkcmdline) {
my $tab = xCAT::Table->new($table,-create=>0);
if ($tab) {
$nodesubdata{$table}=$tab->getNodesAttribs(\@nodes,$tablecolumnsneededforaddkcmdline{$table});
}
}
my $osents = $ostab->getNodesAttribs(\@nodes, ['os', 'arch', 'profile']);
foreach my $node (@nodes){
my $ent = $osents->{$node}->[0];
my $arch = $ent->{'arch'};
my $profile = $ent->{'profile'};
my $osver = $ent->{'os'};
my $tftpdir;
my $ksserver;
if ($resents and $resents->{$node}->[0]->{nfsserver}) {
$ksserver=$resents->{$node}->[0]->{nfsserver};
} else {
$ksserver='!myipfn!';
}
if ($resents and $resents->{$node}->[0]->{tftpdir}) {
$tftpdir = $resents->{$node}->[0]->{tftpdir};
} else {
$tftpdir = $globaltftpdir;
}
#if($arch ne 'x86'){
# xCAT::SvrUtils::sendmsg([1,"VMware ESX hypervisors are x86, please change the nodetype.arch value to x86 instead of $arch for $node before proceeding:
#e.g: nodech $node nodetype.arch=x86\n"]);
# return;
#}
# first make sure copycds was done:
my $custprofpath = $profile;
unless ($custprofpath =~ /^\//) {#If profile begins with a /, assume it already is a path
$custprofpath = $installroot."/custom/install/$osver/$arch/$profile";
unless(-d $custprofpath) {
$custprofpath = $installroot."/custom/install/esxi/$arch/$profile";
}
}
unless(
-r "$custprofpath/vmkboot.gz"
or -r "$custprofpath/b.z"
or -r "$custprofpath/mboot.c32"
or -r "$custprofpath/install.tgz"
or -r "$installroot/$osver/$arch/mboot.c32"
or -r "$installroot/$osver/$arch/install.tgz" ){
xCAT::SvrUtils::sendmsg([1,"Please run copycds first for $osver or create custom image in $custprofpath/"], $output_handler);
}
my @reqmods = qw/vmkboot.gz vmk.gz sys.vgz cim.vgz/; #Required modules for an image to be considered complete
if ( -r "$custprofpath/b.z" ) { #if someone hand extracts from imagedd, a different name scheme is used
@reqmods = qw/b.z k.z s.z c.z/;
}
my %mods;
foreach (@reqmods) {
$mods{$_} = 1;
}
my $shortprofname = $profile;
$shortprofname =~ s/\/\z//;
$shortprofname =~ s/.*\///;
mkpath("$tftpdir/xcat/netboot/$osver/$arch/$shortprofname/");
my $havemod=0;
unless($donetftp{$osver,$arch,$profile,$tftpdir}) {
my $srcdir = "$installroot/$osver/$arch";
my $dest = "$tftpdir/xcat/netboot/$osver/$arch/$shortprofname";
cpNetbootImages($osver,$srcdir,$dest,$custprofpath,\%mods,bootmode=>$bootmode);
if ($havemod = makecustomizedmod($osver,$dest)) {
push @reqmods,"mod.tgz";
$mods{"mod.tgz"}=1;
}
if ($osver =~ /esxi4/ and -r "$::XCATROOT/share/xcat/netboot/syslinux/mboot.c32") { #prefer xCAT patched mboot.c32 with BOOTIF for mboot
copy("$::XCATROOT/share/xcat/netboot/syslinux/mboot.c32", $dest);
} elsif (-r "$custprofpath/mboot.c32") {
copy("$custprofpath/mboot.c32", $dest);
} elsif (-r "$srcdir/mboot.c32") {
copy("$srcdir/mboot.c32", $dest);
}
if (-f "$srcdir/efiboot.img") {
copy("$srcdir/efiboot.img",$dest);
print("$srcdir/efi");
mkpath("$dest/efi");
recursion_copy("$srcdir/efi","$dest/efi");
}
$donetftp{$osver,$arch,$profile,$tftpdir} = 1;
}
my $tp = "xcat/netboot/$osver/$arch/$shortprofname";
my $kernel;
my $kcmdline;
my $append;
my $shortappend;
if ($osver =~ /esxi4/) {
my $bail=0;
foreach (@reqmods) {
unless (-r "$tftpdir/$tp/$_") {
xCAT::SvrUtils::sendmsg([1,"$_ is missing from the target destination, ensure that either copycds has been run or that $custprofpath contains this file"], $output_handler);
$bail=1; #only flag to bail, present as many messages as possible to user
}
}
if ($bail) { #if the above loop detected one or more failures, bail out
return;
}
# now make <HEX> file entry stuff
$kernel = "$tp/mboot.c32";
my $prepend;
if ($reqmods[0] eq "vmkboot.gz") {
$prepend = "$tp/vmkboot.gz";
delete $mods{"vmkboot.gz"};
$append = " --- $tp/vmk.gz";
delete $mods{"vmk.gz"};
$append .= " --- $tp/sys.vgz";
delete $mods{"sys.vgz"};
$append .= " --- $tp/cim.vgz";
delete $mods{"cim.vgz"};
} else { #the single letter style
$prepend = "$tp/b.z";
delete $mods{"b.z"};
$append = " --- $tp/k.z";
delete $mods{"k.z"};
$append .= " --- $tp/s.z";
delete $mods{"s.z"};
$append .= " --- $tp/c.z";
delete $mods{"c.z"};
}
if ($mods{"mod.tgz"}) {
$append .= " --- $tp/mod.tgz";
delete $mods{"mod.tgz"};
}
foreach (keys %mods) {
$append .= " --- $tp/$_";
}
if (defined $bpadds->{$node}->[0]->{addkcmdline}) {
my $modules;
($kcmdline,$modules) = split /---/,$bpadds->{$node}->[0]->{addkcmdline},2;
$kcmdline =~ s/#NODEATTRIB:([^:#]+):([^:#]+)#/$nodesubdata{$1}->{$node}->[0]->{$2}/eg;
if ($modules) {
$append .= " --- ".$modules;
}
$prepend .= " ".$kcmdline;
}
$append = $prepend.$append;
}
elsif ($osver =~ /esxi5/) { #do a more straightforward thing..
$kernel = "$tp/mboot.c32";
if (-r "$tftpdir/$tp/boot.cfg.$bootmode.tmpl") { #so much for straightforward..
$shortappend = "-c $tp/boot.cfg.$bootmode.$node";
} else {
$append = "-c $tp/boot.cfg.$bootmode";
}
if ($bootmode eq "install") {
$append .= " ks=http://$ksserver/install/autoinst/$node";
esxi_kickstart_from_template(node=>$node,os=>$osver,arch=>$arch,profile=>$profile);
}
if ($bootmode ne "install" and $serialconfig->{$node}) { #don't do it for install, installer croaks currently
my $comport = 1;
if (defined $serialconfig->{$node}->[0]->{serialport}) {
$comport = $serialconfig->{$node}->[0]->{serialport}+1;
$append .= " -S $comport tty2port=com$comport";
}
if (defined $serialconfig->{$node}->[0]->{serialspeed}) {
$append .= " -s ".$serialconfig->{$node}->[0]->{serialspeed}." com".$comport."_baud=".$serialconfig->{$node}->[0]->{serialspeed};
}
}
if (defined $bpadds->{$node}->[0]->{addkcmdline}) {
$append .= " ".$bpadds->{$node}->[0]->{addkcmdline};
$append =~ s/#NODEATTRIB:([^:#]+):([^:#]+)#/$nodesubdata{$1}->{$node}->[0]->{$2}/eg;
}
}
if ($shortappend) { #esxi5 user desiring to put everything in one boot config file. . .
merge_esxi5_append("$tftpdir/$tp/boot.cfg.$bootmode.tmpl",$append,"$tftpdir/$tp/boot.cfg.$bootmode.$node");
$append=$shortappend;
}
$output_handler->({node=>[{name=>[$node],'_addkcmdlinehandled'=>[1]}]});
$bptab->setNodeAttribs(
$node,
{
kernel => $kernel,
initrd => "",
kcmdline => $append
}
);
} # end of node loop
}
# this is where we extract the netboot images out of the copied ISO image
sub cpNetbootImages {
my $osver = shift;
my $srcDir = shift;
my $destDir = shift;
my $overridedir = shift;
my $modulestoadd = shift;
my %parmargs = @_;
my $bootmode="stateless";
if ($parmargs{bootmode}) { $bootmode = $parmargs{bootmode} }
my $tmpDir = "/tmp/xcat.$$";
if($osver =~ /esxi4/){
# we don't want to go through this all the time, so if its already
# there we're not going to extract:
unless( -r "$destDir/vmk.gz"
and -r "$destDir/vmkboot.gz"
and -r "$destDir/sys.vgz"
and -r "$destDir/cim.vgz"
and -r "$destDir/cimstg.tgz"
){
if (-r "$srcDir/image.tgz") { #it still may work without image.tgz if profile customization has everything replaced
mkdir($tmpDir);
chdir($tmpDir);
xCAT::SvrUtils::sendmsg("extracting netboot files from OS image. This may take about a minute or two...hopefully you have ~1GB free in your /tmp dir\n", $output_handler);
my $cmd = "tar zxf $srcDir/image.tgz";
print "\n$cmd\n";
if(system($cmd)){
xCAT::SvrUtils::sendmsg([1,"Unable to extract $srcDir/image.tgz\n"], $output_handler);
}
# this has the big image and may take a while.
# this should now create:
# /tmp/xcat.1234/usr/lib/vmware/installer/VMware-VMvisor-big-164009-x86_64.dd.bz2 or some other version. We need to extract partition 5 from it.
system("bunzip2 $tmpDir/usr/lib/vmware/installer/*bz2");
xCAT::SvrUtils::sendmsg("finished extracting, now copying files...\n", $output_handler);
# now we need to get partition 5 which has the installation goods in it.
my $scmd = "fdisk -lu $tmpDir/usr/lib/vmware/installer/*dd 2>&1 | grep dd5 | awk '{print \$2}'";
print "running: $scmd\n";
my $sector = `$scmd`;
chomp($sector);
my $offset = $sector * 512;
mkdir "/mnt/xcat";
my $mntcmd = "mount $tmpDir/usr/lib/vmware/installer/*dd /mnt/xcat -o loop,offset=$offset";
print "$mntcmd\n";
if(system($mntcmd)){
xCAT::SvrUtils::sendmsg([1,"unable to mount partition 5 of the ESX netboot image to /mnt/xcat"], $output_handler);
return;
}
if (! -d $destDir) {
if ( -e $destDir ) {
xCAT::SvrUtils::sendmsg([1,"Could not copy netboot contents to $destDir, it exists but is not currently a directory"], $output_handler);
return;
}
mkpath($destDir);
}
if(system("cp /mnt/xcat/* $destDir/")){
xCAT::SvrUtils::sendmsg([1,"Could not copy netboot contents to $destDir"], $output_handler);
chdir("/");
system("umount /mnt/xcat");
return;
}
chdir("/tmp");
system("umount /mnt/xcat");
print "tempDir: $tmpDir\n";
system("rm -rf $tmpDir");
} elsif (-r "$srcDir/cim.vgz" and -r "$srcDir/vmkernel.gz" and -r "$srcDir/vmkboot.gz" and -r "$srcDir/sys.vgz") {
use File::Basename;
if (! -d $destDir) {
mkpath($destDir);
}
#In ESXI 4.1, the above breaks, this seems to work, much simpler too
foreach ("$srcDir/cim.vgz","$srcDir/vmkernel.gz","$srcDir/vmkboot.gz","$srcDir/sys.vgz","$srcDir/sys.vgz") {
my $mod = scalar fileparse($_);
if ($mod =~ /vmkernel.gz/) {
copy($_,"$destDir/vmk.gz") or xCAT::SvrUtils::sendmsg([1,"Could not copy netboot contents from $_ to $destDir/$mod"], $output_handler);
} else {
copy($_,"$destDir/$mod") or xCAT::SvrUtils::sendmsg([1,"Could not copy netboot contents from $_ to $destDir/$mod"], $output_handler);
}
}
}
}
#this is the override directory if there is one, otherwise it's actually the default dir
if (-d $overridedir) {
mkdir($overridedir);
}
#Copy over all modules
use File::Basename;
foreach (glob "$overridedir/*") {
my $mod = scalar fileparse($_);
if ($mod =~ /gz\z/ and $mod !~ /pkgdb.tgz/ and $mod !~ /vmkernel.gz/) {
$modulestoadd->{$mod}=1;
copy($_,"$destDir/$mod") or xCAT::SvrUtils::sendmsg([1,"Could not copy netboot contents from $overridedir to $destDir"], $output_handler);
} elsif ($mod =~ /vmkernel.gz/) {
$modulestoadd->{"vmk.gz"}=1;
copy($_,"$destDir/vmk.gz") or xCAT::SvrUtils::sendmsg([1,"Could not copy netboot contents from $overridedir to $destDir"], $output_handler);
}
}
}elsif ($osver =~ /esxi5/) { #we need boot.cfg.stateles
my @filestocopy = ("boot.cfg.$bootmode");
if (-r "$srcDir/boot.cfg.$bootmode" or -r "$overridedir/boot.cfg.$bootmode") {
@filestocopy = ("boot.cfg.$bootmode");
} elsif (-r "$srcDir/boot.cfg.$bootmode.tmpl" or -r "$overridedir/boot.cfg.$bootmode.tmpl") {
@filestocopy = ("boot.cfg.$bootmode.tmpl");
} else {
xCAT::SvrUtils::sendmsg([1,"$srcDir is missing boot.cfg.$bootmode file required for $bootmode boot"], $output_handler);
return;
}
my $statelesscfg;
if (-r "$overridedir/boot.cfg.$bootmode.tmpl") {
open ($statelesscfg,"<","$overridedir/boot.cfg.$bootmode.tmpl");
@filestocopy = ("boot.cfg.$bootmode.tmpl");
} elsif (-r "$overridedir/boot.cfg.$bootmode") {
open ($statelesscfg,"<","$overridedir/boot.cfg.$bootmode");
} elsif (-r "$srcDir/boot.cfg.$bootmode.tmpl") {
@filestocopy = ("boot.cfg.$bootmode.tmpl");
open ($statelesscfg,"<","$srcDir/boot.cfg.$bootmode.tmpl");
} elsif (-r "$srcDir/boot.cfg.$bootmode") {
open ($statelesscfg,"<","$srcDir/boot.cfg.$bootmode");
} else {
die "boot.cfg.$bootmode was missing from $srcDir???";
}
my @statelesscfg=<$statelesscfg>;
foreach (@statelesscfg) { #search for files specified by the boot cfg and pull them in
if (/^kernel=(.*)/) {
push @filestocopy,$1;
} elsif (/^modules=(.*)/) {
foreach (split / --- /,$1) {
s/^\s*//;
s/\s.*//;
push @filestocopy,$_;
}
}
}
#now that we have a list, do the copy (mostly redundant, but PXE needs them tftp accessible)
foreach (@filestocopy) {
chomp;
s/ *\z//;
my $mod = scalar fileparse($_);
if (-r "$overridedir/$mod") {
copyIfNewer("$overridedir/$mod","$destDir/$mod") or xCAT::SvrUtils::sendmsg([1,"Could not copy netboot contents from $overridedir/$mod to $destDir/$mod, $!"], $output_handler);
} elsif (-r "$srcDir/$mod") {
copyIfNewer($srcDir."/".$mod,"$destDir/$mod") or xCAT::SvrUtils::sendmsg([1,"Could not copy netboot contents from $srcDir/$mod to $destDir/$mod, $!"], $output_handler);
} elsif ($mod ne "xcatmod.tgz") {
xCAT::SvrUtils::sendmsg([1,"Could not copy netboot contents from $srcDir/$mod to $destDir/$mod, $srcDir/$mod not found"], $output_handler);
}
}
} else {
xCAT::SvrUtils::sendmsg([1,"VMware $osver is not supported for netboot"], $output_handler);
}
}
sub copyIfNewer {
my $source = shift;
my $dest = shift;
if (! -e $dest or -C $source < -C $dest) {
return copy($source,$dest);
}
return 1;
}
# compares nfs target described by parameters to every share mounted by target hypervisor
# returns 1 if matching datastore is present and 0 otherwise
sub match_nfs_datastore {
my ($host, $path, $hypconn) = @_;
die "esx plugin bug: no host provided for match_datastore" unless defined $host;
die "esx plugin bug: no path provided for match_datastore" unless defined $path;
my @ip;
eval {
if ($host =~ m/\b(?:(?:25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)\.){3}(?:25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)\//) {
use Socket;
@ip = ( $host );
$host = gethostbyaddr(inet_aton($host, AF_INET), AF_INET);
} else {
use Socket;
(undef, undef, undef, undef, @ip) = gethostbyname($host);
my @ip_ntoa = ();
foreach (@ip) {
push (@ip_ntoa, inet_ntoa($_));
}
@ip = @ip_ntoa;
}
};
if ($@) {
die "error while resolving datastore host: $@\n";
}
my %viewcrit = (
view_type => 'HostSystem',
properties => [ 'config.fileSystemVolume' ],
);
my $dsviews = $hypconn->find_entity_views(%viewcrit);
foreach (@$dsviews) {
foreach my $mount (@{$_->get_property('config.fileSystemVolume.mountInfo')}) {
next unless $mount->{'volume'}{'type'} eq 'NFS';
my $hostMatch = 0;
HOSTMATCH: foreach (@ip, $host) {
next HOSTMATCH unless $mount->{'volume'}{'remoteHost'} eq $_;
$hostMatch = 1;
last HOSTMATCH;
}
next unless $hostMatch;
next unless $mount->{'volume'}{'remotePath'} eq $path;
return 1;
}
}
return 0;
}
1;
# vi: set ts=4 sw=4 filetype=perl: