Add some code that prevents node from going into hwdown when the image

being loaded is a user image.

db/libdb.pm.in

@@ -28,7 +28,7 @@ use vars qw(@ISA @EXPORT);
 	 PROJMEMBERTRUST_ROOT PROJMEMBERTRUST_GROUPROOT
 	 PROJMEMBERTRUST_PROJROOT
 
-	 PROJROOT GROUPROOT USERROOT TBOPSPID
+	 PROJROOT GROUPROOT USERROOT TBOPSPID 
 	 PLABMOND_PID PLABMOND_EID PLABHOLDING_PID PLABHOLDING_EID
 
 	 TBTrustConvert TBMinTrust TBGrpTrust TBProjTrust
@@ -152,7 +152,7 @@ use vars qw(@ISA @EXPORT);
 	 TBValidNodeLogType TBValidNodeName TBSetNodeLogEntry
 	 TBSetSchedReload MapNodeOSID TBLockExp TBUnLockExp TBSetExpSwapTime
 	 TBUnixGroupList TBOSID TBOSMaxConcurrent TBOSCountInstances
-	 TBResolveNextOSID
+	 TBResolveNextOSID TBOsidToPid
 	 TBOSLoadMaxOkay TBImageLoadMaxOkay TBImageID ExpSwapper
 	 TBdbfork VnameToNodeid TBExpLocked
 	 TBIsNodeRemote TBExptSetLogFile TBExptClearLogFile TBExptGetLogFile
@@ -1922,6 +1922,27 @@ sub TBOSID ($$) {
     return $row[0];
 }
 
+#
+# Return pid of an osid (internal name).
+#
+# usage: TBOsidToPid(char *osid, \$pid)
+#        returns 1 if osid is valid; store pid into return arg.
+#        returns 0 if osid is not valid.
+#
+sub TBOsidToPid ($$) {
+    my($osid, $ppid) = @_;
+
+    my $query_result =
+	DBQueryFatal("select pid from os_info where osid='$osid'");
+
+    if (! $query_result->num_rows) {
+	return 0;
+    }
+    my ($pid) = $query_result->fetchrow_array();
+    $$ppid = $pid;
+    return 1;
+}
+
 #
 # Returns the maximum number of concurrent instantiations of an image.
 #

tbsetup/os_setup.in

@@ -692,8 +692,22 @@ while ( @nodelist ) {
 	next;
     }
 
-    # Reserve it to down experiment.
-    MarkNodeDown($node);
+    #
+    # If the user has picked a standard image and it fails to boot,
+    # something is wrong, so reserve it to hwdwon experiment. If the
+    # image belongs to the user, then we assume its the image at fault,
+    # and allow it to be returned to the pool (caller, tbswap will end
+    # doing the nfree on nodes with a DOWN allocstate).
+    #
+    my $pidofosid;
+    if (! TBOsidToPid($osids{$node}, \$pidofosid) ||
+	$pidofosid eq TBOPSPID()) {
+	MarkNodeDown($node);
+	TBSetNodeLogEntry($node, $dbuid, TB_DEFAULT_NODELOGTYPE(),
+			  "'Moved to hwdown by os_setup; ".
+			  "failed to boot image for osid " . $osids{$node} .
+			  " in $pid/$eid'");
+    }
     TBSetNodeAllocState( $node, TBDB_ALLOCSTATE_DOWN() );
     $nodeAllocStates{$node} = TBDB_ALLOCSTATE_DOWN();