[pve-devel] r6388 - in pve-cluster/trunk/data: . PVE

[svn-commits at proxmox.com]

Author: dietmar
Date: 2011-08-01 13:33:10 +0200 (Mon, 01 Aug 2011)
New Revision: 6388

Modified:
   pve-cluster/trunk/data/ChangeLog
   pve-cluster/trunk/data/PVE/Cluster.pm
Log:
cfs_lock improvements 


Modified: pve-cluster/trunk/data/ChangeLog
===================================================================
--- pve-cluster/trunk/data/ChangeLog	2011-08-01 09:29:24 UTC (rev 6387)
+++ pve-cluster/trunk/data/ChangeLog	2011-08-01 11:33:10 UTC (rev 6388)
@@ -1,3 +1,9 @@
+2011-08-01  Proxmox Support Team  <support at proxmox.com>
+
+	* PVE/Cluster.pm (check_cfs_quorum): helper to check quorum
+	(cfs_lock_file): remove $text parameter
+	(cfs_lock_storage): remove $text parameter
+
 2011-06-29  Proxmox Support Team  <support at proxmox.com>
 
 	* PVE/Cluster.pm (run_silent_cmd): supress output when running

Modified: pve-cluster/trunk/data/PVE/Cluster.pm
===================================================================
--- pve-cluster/trunk/data/PVE/Cluster.pm	2011-08-01 09:29:24 UTC (rev 6387)
+++ pve-cluster/trunk/data/PVE/Cluster.pm	2011-08-01 11:33:10 UTC (rev 6388)
@@ -2,6 +2,7 @@
 
 use strict;
 use POSIX;
+use File::stat qw();
 use Socket;
 use Storable qw(dclone);
 use IO::File;
@@ -73,6 +74,13 @@
     }
 }
 
+sub check_cfs_quorum {
+    # note: -w $lockdir always return 1 for root, so wee need
+    # to use File::stat here
+    my $st = File::stat::stat($lockdir);
+    return (($st->mode & 0200) != 0);
+}
+
 sub check_cfs_is_mounted {
     my ($noerr) = @_;
 
@@ -727,21 +735,25 @@
 }
 
 my $cfs_lock = sub {
-    my ($lockid, $timeout, $text, $code, @param) = @_;
+    my ($lockid, $timeout, $code, @param) = @_;
 
     my $res;
 
-    # fixed timeout: cfs locks have a timeout of 120
-    # using 60 gives us another 60 seconds to abort the task
-
+    # this timeout is for aquire the lock
     $timeout = 10 if !$timeout;
 
     my $filename = "$lockdir/$lockid";
 
+    my $msg = "can't aquire cfs lock '$lockid'";
+
     eval {
 
-        local $SIG{ALRM} = sub { die "can't aquire lock for $text '$lockid': got timeout\n"; };
+	if (! -d $lockdir) {
+	    die "$msg: pve cluster filesystem not online.\n";
+	}
 
+        local $SIG{ALRM} = sub { die "got lock request timeout\n"; };
+
         alarm ($timeout);
 
 	if (!(mkdir $filename)) {
@@ -751,13 +763,13 @@
 		    (utime 0, 0, $filename); # cfs unlock request
 		} else {
 		    print STDERR " OK\n";
- 		    last;
+		    last;
 		}
 		sleep(1);
 	    }
 	}
 
-	# fixed timeout: cfs locks have a timeout of 120
+	# fixed command timeout: cfs locks have a timeout of 120
 	# using 60 gives us another 60 seconds to abort the task
 	alarm(60);
 	local $SIG{ALRM} = sub { die "got lock timeout - aborting command\n"; };
@@ -771,6 +783,11 @@
 
     alarm(0);
 
+    if ($err && ($err eq "got lock request timeout\n") &&
+	!check_cfs_quorum()){
+	$err = "$msg: no quorum!\n";
+    }	
+
     if (!$err || $err !~ /^got lock timeout -/) {
 	rmdir $filename; # cfs unlock
     }
@@ -786,22 +803,22 @@
 };
 
 sub cfs_lock_file {
-    my ($filename, $timeout, $text, $code, @param) = @_;
+    my ($filename, $timeout, $code, @param) = @_;
 
     my $info = $observed->{$filename} || die "unknown file '$filename'";
 
     my $lockid = "file-$filename";
     $lockid =~ s/[.\/]/_/g;
 
-    &$cfs_lock($lockid, $timeout, $text, $code, @param);
+    &$cfs_lock($lockid, $timeout, $code, @param);
 }
 
 sub cfs_lock_storage {
-    my ($storeid, $timeout, $text, $code, @param) = @_;
+    my ($storeid, $timeout, $code, @param) = @_;
 
     my $lockid = "storage-$storeid";
 
-    &$cfs_lock($lockid, $timeout, $text, $code, @param);
+    &$cfs_lock($lockid, $timeout, $code, @param);
 }
 
 my $log_levels = {