Add dm_task_no_open_count() to skip getting open_count.

Allow user to kill clvmd during initialisation.

Makefile.in

@@ -22,11 +22,13 @@ ifeq ("@INTL@", "yes")
   SUBDIRS += po
 endif
 
-SUBDIRS += lib tools
+SUBDIRS += lib tools daemons
 
 ifeq ($(MAKECMDGOALS),distclean)
-  SUBDIRS += lib/format1 \
+  SUBDIRS += daemons/clvmd \
+	     lib/format1 \
 	     lib/format_pool \
+	     lib/locking \
 	     lib/mirror \
 	     lib/snapshot \
 	     po \
@@ -35,14 +37,16 @@ endif
 
 include make.tmpl
 
+daemons: lib
 lib: include
 tools: lib
-po: lib tools
+po: tools daemons
 
 ifeq ("@INTL@", "yes")
 lib.pofile: include.pofile
 tools.pofile: lib.pofile
-po.pofile: lib.pofile tools.pofile
+daemons.pofile: lib.pofile
+po.pofile: tools.pofile daemons.pofile
 pofile: po.pofile
 endif
 

README

@@ -9,8 +9,8 @@ Installation instructions are in INSTALL.
 There is no warranty - see COPYING and COPYING.LIB.
 
 Tarballs are available from:
-  ftp://ftp.sistina.com/pub/LVM2/tools/
-  ftp://ftp.sistina.com/pub/LVM2/device-mapper/
+  ftp://sources.redhat.com/pub/lvm2/
+  ftp://sources.redhat.com/pub/dm/
 
 To access the CVS tree use:
   cvs -d :pserver:cvs@sources.redhat.com:/cvs/lvm2 login

VERSION

@@ -1 +1 @@
-2.00.17-cvs (2004-06-20)
+2.00.34-cvs (2005-01-07)

WHATS_NEW

@@ -1,3 +1,152 @@
+Version 2.00.34 - 
+==================================
+  Adjust RHEL4 clvmd init script priority.
+
+Version 2.00.33 - 7th January 2005
+==================================
+  pvcreate wipes first 4 sectors unless given --zero n.
+  gulm clvmd now uses new ccsd key names.
+  gulm clvmd now doesn't ignore the first node in cluster.conf
+  Improve clvmd failure message if it's already running.
+  Allow user to kill clvmd during initialisation.
+  Fix off-by-one error in cluster_locking that could cause read hangs.
+
+Version 2.00.32 - 22nd December 2004
+====================================
+  Drop static/dl restriction for now.
+  Fix an error fprintf.
+  Fix vgdisplay -s. Breaks (undocumented) lvs/pvs/vgs -s instead for now.
+  Fix device reference counting on re-opens.
+  Ignore sysfs symlinks when DT_UNKNOWN.
+  Add clvmd init script for RHEL4.
+  Skip devices that are too small to be PVs.
+  Fix pvchange -x segfault with lvm2-format orphan.
+  Cope with empty msdos partition tables.
+  Add CONTRIBUTORS file.
+
+Version 2.00.31 - 12th December 2004
+====================================
+  Reopen RO file descriptors RW if necessary.
+
+Version 2.00.30 - 10th December 2004
+====================================
+  Additional device-handling debug messages.
+  Additional verbosity level -vvvv includes line numbers and backtraces.
+  Verbose messages now go to stderr not stdout.
+  Close any stray file descriptors before starting.
+  Refine partitionable checks for certain device types.
+  Allow devices/types to override built-ins.
+  Fix lvreduce man page .i->.I
+  Fix vgsplit man page title.
+  Fix clvmd man makefile.
+  Extend dev_open logging.
+  Make clvmd_fix_conf.sh UNDOable.
+
+Version 2.00.29 - 27th November 2004
+====================================
+  xlate compilation fix.
+
+Version 2.00.28 - 27th November 2004
+====================================
+  Fix partition table & md signature detection.
+  Minor configure/makefile tidy.
+  Export version.h from tools for clvmd.
+
+Version 2.00.27 - 24th November 2004
+====================================
+  Trap large memory allocation requests.
+  Fix to partition table detection code.
+  Improve filter debug mesgs.
+  Make clvmd_fix_conf.sh UNDOable
+
+Version 2.00.26 - 23rd November 2004
+====================================
+  Improve pool debugging stats.
+  Detect partition table signature.
+  pvcreate wipes md superblocks. (With --uuid or --restorefile it prompts.)
+  Separate out md superblock detection code.
+  Prevent snapshot origin resizing.
+  Improve a vgremove error message.
+  Update some man pages.
+  Allow y/n with -ae args (exclusive activation).
+  Fixes to lvcreate vgname parsing.
+  Fix dm_name string size calculation.
+  Improve clvmd error reporting during startup.
+  Make clvmd cope with large gaps in node numbers IDs.
+  Make clvmd initialisation cope better with debugging output.
+  Tidy clvmd socket callbacks so all work happens outside main loop.
+  clvmd -V now displays lvm version too.
+  Add optional gulm build for clvmd
+
+Version 2.00.25 - 29th September 2004
+=====================================
+  Fix return code from rm_link for vgmknodes.
+  Make clvmd LV hash table thread-safe.
+  Fix clvmd locking so it will lock out multiple users on the same node.
+  Fix clvmd VG locking to it can cope with multiple VG locks.
+  Remove spurious trailing dot in lvreduce man page.
+  Fix vgremove locking.
+
+Version 2.00.24 - 16th September 2004
+=====================================
+  Fix pool_empty so it really does empty the memory pool.
+  Rename old segtypes files to segtype.
+  Some fixes to memory debugging code.
+  Exclude internal commands formats & segtypes from install.
+
+Version 2.00.23 - 15th September 2004
+=====================================
+  Export dm name build & split functions.
+  Use O_NOATIME on devices if available.
+  Write log message when each segtype/format gets initialised.
+  New commands 'segtypes' and 'formats'.
+  Suppress pvmove abort message in test mode.
+  Improve pvcreate/remove device not found error message.
+  Allow pvmove to move data within the same PV.
+  Describe how pvmove works on man page.
+  Test for incompatible format/segtype combinations in lv_extend.
+  Fix lvchange example on man page.
+
+Version 2.00.22 - 3rd September 2004
+====================================
+  Fix /dev/vgname perms.
+  Restructure xlate.h.
+  Add clvmd man page.
+
+Version 2.00.21 - 19th August 2004
+==================================
+  Update cnxman-socket.h from cman.
+  Recognise iseries/vd devices.
+  Use 'make install_cluster' to install cluster extensions only.
+  Cope with DT_UNKNOWN in sysfs.
+  Fix extents_moved metadata size comment.
+  Remove duplicate line in pvremove help text.
+  Support variable mirror region size.
+  Support PE ranges in pvmove source PV.
+  Fixes to as-yet-unused LV segment splitting code.
+  Change alloc_areas to pe_ranges and allow suppression of availability checks.
+  Add dev_size column to pvs.
+  Add report columns for in-kernel device number.
+
+Version 2.00.20 - 3 July 2004
+=============================
+  More autoconf fixes.
+  Fix device number handling for 2.6 kernels.
+
+Version 2.00.19 - 29 June 2004
+==============================
+  Reduce severity of setlocale failure message.
+  Recognise argv[0] "initrd-lvm" (pld-linux).
+  Make -O2 configurable.
+  Added --disable-selinux to configure script.
+  LD_FLAGS->LDFLAGS & LD_DEPS->LDDEPS in configure script.
+  Add init_debug to clvmd.
+
+Version 2.00.18 - 24 June 2004
+==============================
+  Fix vgchange activation.
+  Add cluster support.
+
 Version 2.00.17 - 20 June 2004
 ==============================
   configure --enable-fsadm to try out fsadm.  fsadm is not tested yet.

WHATS_NEW_DM

@@ -1,3 +1,27 @@
+Version 1.01.00 -
+============================
+  Add dm_task_no_open_count() to skip getting open_count.
+
+Version 1.00.21 - 7 Jan 2005
+============================
+  Fix /proc/devices parsing.
+
+Version 1.00.20 - 6 Jan 2005
+============================
+  Attempt to fix /dev/mapper/control transparently if it's wrong.
+  Configuration-time option for setting uid/gid/mode for /dev/mapper nodes.
+  Update kernel patches for 2.4.27/2.4.28-pre-4 (includes minor fixes).
+  Add --noheadings columns option for colon-separated dmsetup output.
+  Support device referencing by uuid or major/minor.
+  Warn if kernel data didn't fit in buffer.
+  Fix a printf.
+
+Version 1.00.19 - 3 July 2004
+=============================
+  More autoconf fixes.
+  Fix a dmsetup newline.
+  Fix device number handling for 2.6 kernels.
+
 Version 1.00.18 - 20 Jun 2004
 =============================
   Fix a uuid free in libdm-iface.

configure.in

@@ -1,29 +1,64 @@
 ##
-##    Copyright 1999-2000 Sistina Software, Inc.
+## Copyright (C) 2000-2004 Sistina Software, Inc. All rights reserved.
+## Copyright (C) 2004 Red Hat, Inc. All rights reserved.
 ##
-##    This is free software released under the GNU General Public License.
-##    There is no warranty for this software.  See the file COPYING for
-##    details.
+## This file is part of the LVM2.
 ##
-##    See the file CONTRIBUTORS for a list of contributors.
+## This copyrighted material is made available to anyone wishing to use,
+## modify, copy, or redistribute it subject to the terms and conditions
+## of the GNU General Public License v.2.
 ##
-##    This file is maintained by:
-##      AJ Lewis <lewis@sistina.com>
-## 
-##    File name: configure.in
-##
-##    Description: Input file for autoconf.  Generates the configure script 
-##                 that tries to keep everything nice and portable.  It also
-##                 simplifies distribution package building considerably.
+## You should have received a copy of the GNU General Public License
+## along with this program; if not, write to the Free Software Foundation,
+## Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 ################################################################################
 
-dnl Process this file with autoconf to produce a configure script.
+AC_PREREQ(2.53)
+################################################################################
+dnl -- Process this file with autoconf to produce a configure script.
 AC_INIT(lib/device/dev-cache.h)
 
-dnl setup the directory where autoconf has auxilary files
+################################################################################
+dnl -- Setup the directory where autoconf has auxilary files
 AC_CONFIG_AUX_DIR(autoconf) 
 
-dnl Checks for programs.
+################################################################################
+dnl -- Get system type
+AC_CANONICAL_SYSTEM
+
+case "$host_os" in
+	linux*)
+		CFLAGS="$CFLAGS"
+		COPTIMISE_FLAG="-O2"
+		CLDFLAGS="$CLDFLAGS -Wl,--version-script,.export.sym"
+		CLDWHOLEARCHIVE="-Wl,-whole-archive"
+		CLDNOWHOLEARCHIVE="-Wl,-no-whole-archive"
+		LDDEPS="$LDDEPS .export.sym"
+		LDFLAGS="$LDFLAGS -Wl,--export-dynamic"
+		SOFLAG="-shared"
+		DEVMAPPER=yes
+		ODIRECT=yes
+		SELINUX=yes
+		CLUSTER=internal
+		FSADM=no ;;
+	darwin*)
+		CFLAGS="$CFLAGS -no-cpp-precomp -fno-common"
+		COPTIMISE_FLAG="-O2"
+		CLDFLAGS="$CLDFLAGS"
+		CLDWHOLEARCHIVE="-all_load"
+		CLDNOWHOLEARCHIVE=
+		LDDEPS="$LDDEPS"
+		LDFLAGS="$LDFLAGS"
+		SOFLAG="-dynamiclib"
+		DEVMAPPER=no
+		ODIRECT=no
+		SELINUX=no
+		CLUSTER=none
+		FSADM=no ;;
+esac
+
+################################################################################
+dnl -- Checks for programs.
 AC_PROG_AWK
 AC_PROG_CC
 AC_PROG_INSTALL
@@ -31,220 +66,310 @@ AC_PROG_LN_S
 AC_PROG_MAKE_SET
 AC_PROG_RANLIB
 
-dnl Checks for header files.
+################################################################################
+dnl -- Checks for header files.
 AC_HEADER_DIRENT
 AC_HEADER_STDC
-AC_CHECK_HEADERS(fcntl.h malloc.h sys/ioctl.h unistd.h)
+AC_HEADER_SYS_WAIT
+AC_HEADER_TIME
 
-dnl Checks for typedefs, structures, and compiler characteristics.
+AC_CHECK_HEADERS(fcntl.h limits.h locale.h stddef.h syslog.h sys/file.h sys/ioctl.h sys/param.h sys/time.h,,AC_MSG_ERROR(bailing out))
+AC_CHECK_HEADERS(assert.h ctype.h libgen.h signal.h stdio.h sys/mman.h sys/resource.h sys/stat.h sys/types.h sys/utsname.h sys/wait.h time.h,,AC_MSG_ERROR(bailing out))
+
+case "$host_os" in
+	linux*)
+		AC_CHECK_HEADERS(asm/byteorder.h linux/fs.h malloc.h,,AC_MSG_ERROR(bailing out)) ;;
+	darwin*)
+		AC_CHECK_HEADERS(machine/endian.h sys/disk.h,,AC_MSG_ERROR(bailing out)) ;;
+esac
+
+################################################################################
+dnl -- Checks for typedefs, structures, and compiler characteristics.
 AC_C_CONST
 AC_C_INLINE
 AC_TYPE_OFF_T
 AC_TYPE_PID_T
 AC_TYPE_SIZE_T
+AC_TYPE_MODE_T
 AC_STRUCT_ST_RDEV
-AC_HEADER_TIME
+AC_STRUCT_TM
 
-dnl Get system type
-AC_CANONICAL_SYSTEM
+################################################################################
+dnl -- Check for functions
+AC_CHECK_FUNCS(gethostname getpagesize memset munmap setlocale strcasecmp strchr strdup strncasecmp strerror strrchr strstr strtol strtoul,,AC_MSG_ERROR(bailing out))
+AC_FUNC_ALLOCA
+AC_FUNC_CLOSEDIR_VOID
+AC_FUNC_FORK
+AC_FUNC_LSTAT
+AC_FUNC_MALLOC
+AC_FUNC_MEMCMP
+AC_FUNC_MMAP
+AC_FUNC_STAT
+AC_FUNC_STRTOD
 
-case "$host_os" in
-	linux*)
-		CFLAGS= 
-		CLDFLAGS="-Wl,--version-script,.export.sym"
-		CLDWHOLEARCHIVE="-Wl,-whole-archive"
-		CLDNOWHOLEARCHIVE="-Wl,-no-whole-archive"
-		LD_DEPS=".export.sym"
-		LD_FLAGS="-Wl,--export-dynamic"
-		SOFLAG="-shared"
-		DEVMAPPER=yes
-		ODIRECT=yes
-		FSADM=no ;;
-	darwin*)
-		CFLAGS="-no-cpp-precomp -fno-common"
-		CLDFLAGS=
-		CLDWHOLEARCHIVE="-all_load"
-		CLDNOWHOLEARCHIVE=
-		LD_DEPS=
-		LD_FLAGS=
-		SOFLAG="-dynamiclib"
-		DEVMAPPER=no
-		ODIRECT=no
-		FSADM=no ;;
-esac
-
-dnl -- prefix is /usr by default, the exec_prefix default is setup later
+################################################################################
+dnl -- Prefix is /usr by default, the exec_prefix default is setup later
 AC_PREFIX_DEFAULT(/usr)
 
-OWNER="root"
-GROUP="root"
+################################################################################
+dnl -- Parallel make jobs?
+AC_ARG_ENABLE(jobs, [  --enable-jobs=NUM       Number of jobs to run simultaneously], JOBS=-j$enableval, JOBS=-j2)
+
+################################################################################
+dnl -- Setup the ownership of the files
+AC_MSG_CHECKING(file owner)
+OWNER="root"
 
-dnl -- setup the ownership of the files
 AC_ARG_WITH(user,
   [  --with-user=USER        Set the owner of installed files ],
   [ OWNER="$withval" ])
+AC_MSG_RESULT($OWNER)
 
 if test x$OWNER != x; then
 	OWNER="-o $OWNER"
 fi
 
-dnl -- setup the group ownership of the files
+################################################################################
+dnl -- Setup the group ownership of the files
+AC_MSG_CHECKING(group owner)
+GROUP="root"
 AC_ARG_WITH(group,
   [  --with-group=GROUP      Set the group owner of installed files ],
   [ GROUP="$withval" ])
+AC_MSG_RESULT($GROUP)
 
 if test x$GROUP != x; then
 	GROUP="-g $GROUP"
 fi
 
+################################################################################
 dnl -- LVM1 tool fallback option
+AC_MSG_CHECKING(whether to enable lvm1 fallback)
 AC_ARG_ENABLE(lvm1_fallback, [  --enable-lvm1_fallback  Use this to fall back and use LVM1 binaries if
                           device-mapper is missing from the kernel],  LVM1_FALLBACK=$enableval, LVM1_FALLBACK=no)
+AC_MSG_RESULT($LVM1_FALLBACK)
 
 if test x$LVM1_FALLBACK = xyes; then
 	CFLAGS="$CFLAGS -DLVM1_FALLBACK"
 fi
 
+################################################################################
 dnl -- format1 inclusion type
+AC_MSG_CHECKING(whether to include support for lvm1 metadata)
 AC_ARG_WITH(lvm1,
   [  --with-lvm1=TYPE        LVM1 metadata support: internal/shared/none
                           [TYPE=internal] ],
   [ LVM1="$withval" ],
   [ LVM1="internal" ])
+AC_MSG_RESULT($LVM1)
 
 if [[ "x$LVM1" != xnone -a "x$LVM1" != xinternal -a "x$LVM1" != xshared ]];
  then  AC_MSG_ERROR(
 --with-lvm1 parameter invalid
 )
- exit
 fi;
 
 if test x$LVM1 = xinternal; then
 	CFLAGS="$CFLAGS -DLVM1_INTERNAL"
 fi
 
+################################################################################
 dnl -- format_pool inclusion type
+AC_MSG_CHECKING(whether to include support for GFS pool metadata)
 AC_ARG_WITH(pool,
   [  --with-pool=TYPE        GFS pool read-only support: internal/shared/none
                           [TYPE=internal] ],
   [ POOL="$withval" ],
   [ POOL="internal" ])
+AC_MSG_RESULT($POOL)
 
 if [[ "x$POOL" != xnone -a "x$POOL" != xinternal -a "x$POOL" != xshared ]];
  then  AC_MSG_ERROR(
 --with-pool parameter invalid
 )
- exit
 fi;
 
 if test x$POOL = xinternal; then
 	CFLAGS="$CFLAGS -DPOOL_INTERNAL"
 fi
 
+################################################################################
+dnl -- cluster_locking inclusion type
+AC_MSG_CHECKING(whether to include support for cluster locking)
+AC_ARG_WITH(cluster,
+  [  --with-cluster=TYPE     Cluster LVM locking support: internal/shared/none
+                          [TYPE=internal] ],
+  [ CLUSTER="$withval" ])
+AC_MSG_RESULT($CLUSTER)
 
-AC_ARG_ENABLE(jobs, [  --enable-jobs=NUM       Number of jobs to run simultaneously], JOBS=-j$enableval, JOBS=-j2)
+if [[ "x$CLUSTER" != xnone -a "x$CLUSTER" != xinternal -a "x$CLUSTER" != xshared ]];
+ then  AC_MSG_ERROR(
+--with-cluster parameter invalid
+)
+fi;
 
+if test x$CLUSTER = xinternal; then
+	CFLAGS="$CFLAGS -DCLUSTER_LOCKING_INTERNAL"
+fi
+
+################################################################################
 dnl -- snapshots inclusion type
+AC_MSG_CHECKING(whether to include snapshots)
 AC_ARG_WITH(snapshots,
   [  --with-snapshots=TYPE   Snapshot support: internal/shared/none
                           [TYPE=internal] ],
   [ SNAPSHOTS="$withval" ],
   [ SNAPSHOTS="internal" ])
+AC_MSG_RESULT($SNAPSHOTS)
 
 if [[ "x$SNAPSHOTS" != xnone -a "x$SNAPSHOTS" != xinternal -a "x$SNAPSHOTS" != xshared ]];
  then  AC_MSG_ERROR(
 --with-snapshots parameter invalid
 )
- exit
 fi;
 
 if test x$SNAPSHOTS = xinternal; then
 	CFLAGS="$CFLAGS -DSNAPSHOT_INTERNAL"
 fi
 
+################################################################################
 dnl -- mirrors inclusion type
+AC_MSG_CHECKING(whether to include mirrors)
 AC_ARG_WITH(mirrors,
   [  --with-mirrors=TYPE     Mirror support: internal/shared/none
                           [TYPE=internal] ],
   [ MIRRORS="$withval" ],
   [ MIRRORS="internal" ])
+AC_MSG_RESULT($MIRRORS)
 
 if [[ "x$MIRRORS" != xnone -a "x$MIRRORS" != xinternal -a "x$MIRRORS" != xshared ]];
  then  AC_MSG_ERROR(
 --with-mirrors parameter invalid
 )
- exit
 fi;
 
 if test x$MIRRORS = xinternal; then
 	CFLAGS="$CFLAGS -DMIRRORED_INTERNAL"
 fi
 
-dnl Enables staticly-linked tools
+################################################################################
+dnl -- Enables staticly-linked tools
+AC_MSG_CHECKING(whether to use static linking)
 AC_ARG_ENABLE(static_link, [  --enable-static_link    Use this to link the tools to their libraries
                           statically.  Default is dynamic linking],  STATIC_LINK=$enableval, STATIC_LINK=no)
+AC_MSG_RESULT($STATIC_LINK)
 
-dnl Enable readline
-AC_ARG_ENABLE(readline, [  --enable-readline       Enable readline support],  \
+################################################################################
+dnl -- Enable readline
+AC_MSG_CHECKING(whether to enable readline)
+AC_ARG_ENABLE(readline, [  --enable-readline       Enable readline support],
 READLINE=$enableval, READLINE=no)
+AC_MSG_RESULT($READLINE)
 
 if test x$READLINE = xyes; then
 	CFLAGS="$CFLAGS -DREADLINE_SUPPORT"
 fi
 
-echo $ac_n "checking whether to enable debugging""... $ac_c" 1>&6
-dnl Enable Debugging
-AC_ARG_ENABLE(debug,    [  --enable-debug          Enable debugging],  \
-DEBUG=yes, DEBUG=no)
-echo "$ac_t""$DEBUG" 1>&6
+################################################################################
+dnl -- Disable selinux
+AC_MSG_CHECKING(whether to enable selinux support)
+AC_ARG_ENABLE(selinux, [  --disable-selinux       Disable selinux support],
+SELINUX=$enableval)
+AC_MSG_RESULT($SELINUX)
 
-echo $ac_n "checking whether to enable device-mapper""... $ac_c" 1>&6
-dnl Disable devmapper
-AC_ARG_ENABLE(devmapper, [  --disable-devmapper     Disable device-mapper interaction],  \
-DEVMAPPER=no)
-echo "$ac_t""$DEVMAPPER" 1>&6
+################################################################################
+dnl -- Build cluster LVM daemon
+AC_MSG_CHECKING(whether to build cluster LVM daemon)
+AC_ARG_WITH(clvmd,
+  [  --with-clvmd=TYPE       Build cluster LVM Daemon: cman/gulm/none
+                          [TYPE=none] ],
+  [ CLVMD="$withval" ],
+  [ CLVMD="none" ])
+if test x$CLVMD = xyes; then
+	CLVMD=cman
+fi
+AC_MSG_RESULT($CLVMD)
+
+dnl -- If clvmd enabled without cluster locking, automagically include it
+if  test x$CLVMD != xnone && test x$CLUSTER = xnone; then
+	CLUSTER=internal
+fi
+
+################################################################################
+dnl -- Enable debugging
+AC_MSG_CHECKING(whether to enable debugging)
+AC_ARG_ENABLE(debug,    [  --enable-debug          Enable debugging],
+DEBUG=$enableval, DEBUG=no)
+AC_MSG_RESULT($DEBUG)
+
+dnl -- Normally turn off optimisation for debug builds
+if test x$DEBUG = xyes; then
+	COPTIMISE_FLAG=
+fi
+
+################################################################################
+dnl -- Override optimisation
+AC_MSG_CHECKING(for C optimisation flag)
+AC_ARG_WITH(optimisation,
+  [  --with-optimisation=OPT C optimisation flag [OPT=-O2] ],
+  [ COPTIMISE_FLAG="$withval" ])
+AC_MSG_RESULT($COPTIMISE_FLAG)
+
+################################################################################
+dnl -- Disable devmapper
+AC_MSG_CHECKING(whether to use device-mapper)
+AC_ARG_ENABLE(devmapper, [  --disable-devmapper     Disable device-mapper interaction],
+DEVMAPPER=$enableval)
+AC_MSG_RESULT($DEVMAPPER)
 
 if test x$DEVMAPPER = xyes; then
 	CFLAGS="$CFLAGS -DDEVMAPPER_SUPPORT"
 fi
 
-echo $ac_n "checking whether to enable O_DIRECT""... $ac_c" 1>&6
-dnl Disable O_DIRECT
-AC_ARG_ENABLE(o_direct, [  --disable-o_direct      Disable O_DIRECT],  \
-ODIRECT=no)
-echo "$ac_t""$ODIRECT" 1>&6
+################################################################################
+dnl -- Disable O_DIRECT
+AC_MSG_CHECKING(whether to enable O_DIRECT)
+AC_ARG_ENABLE(o_direct, [  --disable-o_direct      Disable O_DIRECT],
+ODIRECT=$enableval)
+AC_MSG_RESULT($ODIRECT)
 
 if test x$ODIRECT = xyes; then
 	CFLAGS="$CFLAGS -DO_DIRECT_SUPPORT"
 fi
 
-echo $ac_n "checking whether to compile liblvm2cmd.so""... $ac_c" 1>&6
-dnl Enable cmdlib
-AC_ARG_ENABLE(cmdlib, [  --enable-cmdlib         Build shared command library],  \
-CMDLIB=yes, CMDLIB=no)
-echo "$ac_t""$CMDLIB" 1>&6
+################################################################################
+dnl -- Enable cmdlib
+AC_MSG_CHECKING(whether to compile liblvm2cmd.so)
+AC_ARG_ENABLE(cmdlib, [  --enable-cmdlib         Build shared command library],
+CMDLIB=$enableval, CMDLIB=no)
+AC_MSG_RESULT($CMDLIB)
 
 if test x$CMDLIB = xyes; then
 	CFLAGS="$CFLAGS -DCMDLIB"
 fi
 
-echo $ac_n "checking whether to enable fsadm""... $ac_c" 1>&6
-dnl Enable fsadm
-AC_ARG_ENABLE(fsadm, [  --enable-fsadm          Enable fsadm], FSADM=yes)
-echo "$ac_t""$FSADM" 1>&6
+################################################################################
+dnl -- Enable fsadm
+AC_MSG_CHECKING(whether to build fsadm)
+AC_ARG_ENABLE(fsadm, [  --enable-fsadm          Enable fsadm],
+FSADM=$enableval)
+AC_MSG_RESULT($FSADM)
 
-dnl Mess with default exec_prefix
+################################################################################
+dnl -- Mess with default exec_prefix
 if [[ "x$exec_prefix" = xNONE -a "x$prefix" = xNONE ]];
  then  exec_prefix="";
 fi;
 
-dnl Checks for library functions.
+################################################################################
+dnl -- Checks for library functions.
 AC_PROG_GCC_TRADITIONAL
 AC_TYPE_SIGNAL
 AC_FUNC_VPRINTF
-AC_CHECK_FUNCS(mkdir rmdir uname)
+AC_CHECK_FUNCS(mkdir rmdir uname,,AC_MSG_ERROR(bailing out))
 
-dnl check for termcap (Shamelessly copied from parted 1.4.17)
+################################################################################
+dnl -- Check for termcap (Shamelessly copied from parted 1.4.17)
 if test x$READLINE = xyes; then
 	AC_SEARCH_LIBS(tgetent, ncurses curses termcap termlib, ,
 		AC_MSG_ERROR(
@@ -257,42 +382,51 @@ Note: if you are using precompiled packages you will also need the development
 Note: (n)curses also seems to work as a substitute for termcap.  This was
   not found either - but you could try installing that as well.
 )
-	exit
 	)
 fi
 
-dnl Check for dlopen
+################################################################################
+dnl -- Check for dlopen
 AC_CHECK_LIB(dl, dlopen, HAVE_LIBDL=yes, HAVE_LIBDL=no)
 
-if [[ "x$HAVE_LIBDL" = xyes -a "x$STATIC_LINK" = xno ]]; then
+if [[ "x$HAVE_LIBDL" = xyes ]]; then
 	CFLAGS="$CFLAGS -DHAVE_LIBDL"
 	LIBS="-ldl $LIBS"
 else
 	HAVE_LIBDL=no
 fi
 
-dnl Check for shared/static conflicts
-if [[ \( "x$LVM1" = xshared -o "x$POOL" = xshared -o \
-      "x$SNAPSHOTS" = xshared -o "x$MIRRORS" = xshared \
+################################################################################
+dnl -- Check for shared/static conflicts
+if [[ \( "x$LVM1" = xshared -o "x$POOL" = xshared -o "x$CLUSTER" = xshared \
+      -o "x$SNAPSHOTS" = xshared -o "x$MIRRORS" = xshared \
       \) -a "x$STATIC_LINK" = xyes ]];
  then  AC_MSG_ERROR(
 Features cannot be 'shared' when building statically
 )
- exit
 fi
 
-dnl Check for is_selinux_enabled
-AC_CHECK_LIB(selinux, is_selinux_enabled, HAVE_SELINUX=yes, HAVE_SELINUX=no)
+################################################################################
+dnl -- Check for is_selinux_enabled
+if test x$SELINUX = xyes; then
+	AC_MSG_CHECKING(for is_selinux_enabled function)
+	AC_CHECK_LIB(selinux, is_selinux_enabled, HAVE_SELINUX=yes, HAVE_SELINUX=no)
+	AC_MSG_RESULT($HAVE_SELINUX)
 
-if test x$HAVE_SELINUX = xyes; then
-	CFLAGS="$CFLAGS -DHAVE_SELINUX"
-	LIBS="-lselinux $LIBS"
+	if test x$HAVE_SELINUX = xyes; then
+		CFLAGS="$CFLAGS -DHAVE_SELINUX"
+		LIBS="-lselinux $LIBS"
+	else
+		AC_MSG_WARN(Disabling selinux)
+	fi
 fi
 
-dnl Check for getopt
+################################################################################
+dnl -- Check for getopt
 AC_CHECK_HEADERS(getopt.h, CFLAGS="$CFLAGS -DHAVE_GETOPTLONG")
 
-dnl Check for readline (Shamelessly copied from parted 1.4.17)
+################################################################################
+dnl -- Check for readline (Shamelessly copied from parted 1.4.17)
 if test x$READLINE = xyes; then
 	AC_CHECK_LIB(readline, readline, ,
 		AC_MSG_ERROR(
@@ -303,17 +437,17 @@ support with --disable-readline or download and install readline from:
 Note: if you are using precompiled packages you will also need the development
 package as well (which may be called readline-devel or something similar).
 )
-		exit
 	)
 	AC_CHECK_FUNC(rl_completion_matches, CFLAGS="$CFLAGS -DHAVE_RL_COMPLETION_MATCHES")
 		
 fi
 
-echo $ac_n "checking whether to enable internationalisation""... $ac_c" 1>&6
-dnl Internationalisation stuff
-AC_ARG_ENABLE(nls, [  --enable-nls            Enable Native Language Support],\
-		INTL=yes, INTL=no)
-echo "$ac_t""$INTL" 1>&6
+################################################################################
+dnl -- Internationalisation stuff
+AC_MSG_CHECKING(whether to enable internationalisation)
+AC_ARG_ENABLE(nls, [  --enable-nls            Enable Native Language Support],
+		INTL=$enableval, INTL=no)
+AC_MSG_RESULT($INTL)
 
 if test x$INTL = xyes; then
 	INTL_PACKAGE="lvm2"
@@ -322,7 +456,6 @@ if test x$INTL = xyes; then
 		then  AC_MSG_ERROR(
 		msgfmt not found in path $PATH
 		)
-		exit  
 	fi;
 
 	AC_ARG_WITH(localedir,
@@ -331,6 +464,7 @@ if test x$INTL = xyes; then
   		    [ LOCALEDIR='${prefix}/share/locale' ])
 fi
 
+################################################################################
 AC_ARG_WITH(confdir,
 	    [  --with-confdir=DIR      Configuration files in DIR [/etc]],
   	    [ CONFDIR="$withval" ],
@@ -341,13 +475,54 @@ AC_ARG_WITH(staticdir,
   	    [ STATICDIR="$withval" ],
  	    [ STATICDIR='${exec_prefix}/sbin' ])
 
+################################################################################
+dnl -- Ensure additional headers required
+if test x$READLINE = xyes; then
+	AC_CHECK_HEADERS(readline/readline.h readline/history.h,,AC_MSG_ERROR(bailing out))
+fi
 
+if test x$CLVMD = xyes; then
+	AC_CHECK_HEADERS(mntent.h netdb.h netinet/in.h pthread.h search.h sys/mount.h sys/socket.h sys/uio.h sys/un.h utmpx.h,,AC_MSG_ERROR(bailing out))
+	AC_CHECK_FUNCS(dup2 getmntent memmove select socket,,AC_MSG_ERROR(bailing out))
+	AC_FUNC_GETMNTENT
+#	AC_FUNC_REALLOC
+	AC_FUNC_SELECT_ARGTYPES
+fi
+
+if test x$FSADM = xyes; then
+	AC_CHECK_HEADERS(fstab.h sys/mount.h sys/vfs.h,,AC_MSG_ERROR(bailing out))
+	AC_CHECK_FUNCS(memmove,,AC_MSG_ERROR(bailing out))
+fi
+
+if test x$CLUSTER != xnone; then
+	AC_CHECK_HEADERS(sys/socket.h sys/un.h,,AC_MSG_ERROR(bailing out))
+	AC_CHECK_FUNCS(socket,,AC_MSG_ERROR(bailing out))
+fi
+
+if test x$HAVE_LIBDL = xyes; then
+	AC_CHECK_HEADERS(dlfcn.h,,AC_MSG_ERROR(bailing out))
+fi
+
+if test x$INTL = xyes; then
+	AC_CHECK_HEADERS(libintl.h,,AC_MSG_ERROR(bailing out))
+fi
+
+if test x$DEVMAPPER = xyes; then
+	AC_CHECK_HEADERS(libdevmapper.h,,AC_MSG_ERROR(bailing out))
+fi
+
+if test x$HAVE_SELINUX = xyes; then
+	AC_CHECK_HEADERS(selinux/selinux.h,,AC_MSG_ERROR(bailing out))
+fi
+
+################################################################################
 if test "-f VERSION"; then
   LVM_VERSION="\"`cat VERSION`\""
 else
   LVM_VERSION="Unknown"
 fi
 
+################################################################################
 AC_SUBST(JOBS)
 AC_SUBST(STATIC_LINK)
 AC_SUBST(LVM1)
@@ -357,11 +532,12 @@ AC_SUBST(MIRRORS)
 AC_SUBST(OWNER)
 AC_SUBST(GROUP)
 AC_SUBST(CFLAGS)
+AC_SUBST(COPTIMISE_FLAG)
 AC_SUBST(CLDFLAGS)
 AC_SUBST(CLDWHOLEARCHIVE)
 AC_SUBST(CLDNOWHOLEARCHIVE)
-AC_SUBST(LD_DEPS)
-AC_SUBST(LD_FLAGS)
+AC_SUBST(LDDEPS)
+AC_SUBST(LDFLAGS)
 AC_SUBST(SOFLAG)
 AC_SUBST(LIBS)
 AC_SUBST(LVM_VERSION)
@@ -377,18 +553,24 @@ AC_SUBST(CONFDIR)
 AC_SUBST(STATICDIR)
 AC_SUBST(INTL_PACKAGE)
 AC_SUBST(INTL)
+AC_SUBST(CLVMD)
+AC_SUBST(CLUSTER)
 AC_SUBST(FSADM)
 
-dnl First and last lines should not contain files to generate in order to 
-dnl keep utility scripts running properly
+################################################################################
+dnl -- First and last lines should not contain files to generate in order to 
+dnl -- keep utility scripts running properly
 AC_OUTPUT( 								\
 Makefile								\
-make.tmpl                                                               \
+make.tmpl								\
+daemons/Makefile							\
+daemons/clvmd/Makefile							\
 doc/Makefile								\
 include/Makefile						 	\
 lib/Makefile							 	\
 lib/format1/Makefile						 	\
 lib/format_pool/Makefile						\
+lib/locking/Makefile							\
 lib/mirror/Makefile							\
 lib/snapshot/Makefile							\
 man/Makefile							 	\
@@ -404,14 +586,9 @@ test/filters/Makefile                                                   \
 )
 
 if test x$ODIRECT != xyes; then
-  echo
-  echo Warning: O_DIRECT disabled.
-  echo Use of pvmove may cause machine to lock up under low memory conditions.
-  echo
+  AC_MSG_WARN(Warning: O_DIRECT disabled: low-memory pvmove may lock up)
 fi
 
 if test x$FSADM == xyes; then
-  echo
-  echo Warning: fsadm support is untested.
-  echo
+  AC_MSG_WARN(fsadm support is untested)
 fi

daemons/Makefile.in

@@ -0,0 +1,23 @@
+#
+# Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+#
+# This file is part of the LVM2.
+#
+# This copyrighted material is made available to anyone wishing to use,
+# modify, copy, or redistribute it subject to the terms and conditions
+# of the GNU General Public License v.2.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software Foundation,
+# Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+srcdir = @srcdir@
+top_srcdir = @top_srcdir@
+VPATH = @srcdir@
+
+ifneq ("@CLVMD@", "none")
+  SUBDIRS = clvmd
+endif
+
+include $(top_srcdir)/make.tmpl
+

daemons/clvmd/Makefile.in

@@ -0,0 +1,59 @@
+#
+# Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+#
+# This file is part of the LVM2.
+#
+# This copyrighted material is made available to anyone wishing to use,
+# modify, copy, or redistribute it subject to the terms and conditions
+# of the GNU General Public License v.2.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software Foundation,
+# Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+srcdir = @srcdir@
+top_srcdir = @top_srcdir@
+VPATH = @srcdir@
+
+SOURCES = \
+	clvmd-command.c  \
+	clvmd.c          \
+	libclvm.c        \
+	lvm-functions.c  \
+	system-lv.c
+
+ifeq ("@CLVMD@", "gulm")
+	SOURCES += clvmd-gulm.c tcp-comms.c
+	LMLIBS += -lccs -lgulm
+	CFLAGS += -DUSE_GULM
+endif
+
+ifeq ("@CLVMD@", "cman")
+	SOURCES += clvmd-cman.c
+	LMLIBS += -ldlm
+endif
+
+TARGETS = \
+	clvmd
+
+include $(top_srcdir)/make.tmpl
+
+CFLAGS += -D_REENTRANT -fno-strict-aliasing
+LIBS += -ldevmapper -llvm -lpthread
+
+INSTALL_TARGETS = \
+	install_clvmd
+
+clvmd: $(OBJECTS) $(top_srcdir)/lib/liblvm.a
+	$(CC) -o clvmd $(OBJECTS) $(LDFLAGS) $(LVMLIBS) $(LMLIBS) $(LIBS)
+
+.PHONY: install_clvmd
+
+install_clvmd: $(TARGETS)
+	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) clvmd \
+		$(sbindir)/clvmd
+
+install: $(INSTALL_TARGETS)
+
+install_cluster: $(INSTALL_TARGETS)
+

daemons/clvmd/clvm.h

@@ -0,0 +1,65 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/* Definitions for CLVMD server and clients */
+
+/*
+ * The protocol spoken over the cluster and across the local socket.
+ */
+
+#ifndef _CLVM_H
+#define _CLVM_H
+
+struct clvm_header {
+	uint8_t  cmd;	        /* See below */
+	uint8_t  flags;	        /* See below */
+	uint16_t xid;	        /* Transaction ID */
+	uint32_t clientid;	/* Only used in Daemon->Daemon comms */
+	int32_t  status;	/* For replies, whether request succeeded */
+	uint32_t arglen;	/* Length of argument below. 
+				   If >1500 then it will be passed 
+				   around the cluster in the system LV */
+	char node[1];		/* Actually a NUL-terminated string, node name.
+				   If this is empty then the command is 
+				   forwarded to all cluster nodes unless 
+				   FLAG_LOCAL is also set. */
+	char args[1];		/* Arguments for the command follow the 
+				   node name, This member is only
+				   valid if the node name is empty */
+} __attribute__ ((packed));
+
+/* Flags */
+#define CLVMD_FLAG_LOCAL        1	/* Only do this on the local node */
+#define CLVMD_FLAG_SYSTEMLV     2	/* Data in system LV under my node name */
+
+/* Name of the local socket to communicate between libclvm and clvmd */
+//static const char CLVMD_SOCKNAME[]="/var/run/clvmd";
+static const char CLVMD_SOCKNAME[] = "\0clvmd";
+
+/* Internal commands & replies */
+#define CLVMD_CMD_REPLY    1
+#define CLVMD_CMD_VERSION  2	/* Send version around cluster when we start */
+#define CLVMD_CMD_GOAWAY   3	/* Die if received this - we are running 
+				   an incompatible version */
+#define CLVMD_CMD_TEST     4	/* Just for mucking about */
+
+#define CLVMD_CMD_LOCK              30
+#define CLVMD_CMD_UNLOCK            31
+
+/* Lock/Unlock commands */
+#define CLVMD_CMD_LOCK_LV           50
+#define CLVMD_CMD_LOCK_VG           51
+
+#endif

daemons/clvmd/clvmd-cman.c

@@ -0,0 +1,507 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/*
+ * CMAN communication layer for clvmd.
+ */
+
+#include <pthread.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/socket.h>
+#include <sys/uio.h>
+#include <sys/un.h>
+#include <sys/time.h>
+#include <sys/ioctl.h>
+#include <sys/utsname.h>
+#include <syslog.h>
+#include <netinet/in.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <signal.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <getopt.h>
+#include <errno.h>
+
+#include "clvmd-comms.h"
+#include "clvm.h"
+#include "libdlm.h"
+#include "log.h"
+#include "clvmd.h"
+#include "lvm-functions.h"
+
+#define LOCKSPACE_NAME "clvmd"
+
+static int cluster_sock;
+static int num_nodes;
+static struct cl_cluster_node *nodes = NULL;
+static int count_nodes; /* size of allocated nodes array */
+static int max_updown_nodes = 50;	/* Current size of the allocated array */
+/* Node up/down status, indexed by nodeid */
+static int *node_updown = NULL;
+static dlm_lshandle_t *lockspace;
+
+static void count_clvmds_running(void);
+static void get_members(void);
+static int nodeid_from_csid(char *csid);
+static int name_from_nodeid(int nodeid, char *name);
+
+struct lock_wait {
+	pthread_cond_t cond;
+	pthread_mutex_t mutex;
+	struct dlm_lksb lksb;
+};
+
+int init_cluster()
+{
+	struct sockaddr_cl saddr;
+	int port = CLUSTER_PORT_CLVMD;
+
+	/* Open the cluster communication socket */
+	cluster_sock = socket(AF_CLUSTER, SOCK_DGRAM, CLPROTO_CLIENT);
+	if (cluster_sock == -1) {
+		syslog(LOG_ERR, "Can't open cluster manager socket: %m");
+		return -1;
+	}
+
+	/* Bind to our port number on the cluster.
+	   Writes to this will block if the cluster loses quorum */
+	saddr.scl_family = AF_CLUSTER;
+	saddr.scl_port = port;
+
+	if (bind
+	    (cluster_sock, (struct sockaddr *) &saddr,
+	     sizeof(struct sockaddr_cl))) {
+		syslog(LOG_ERR, "Can't bind cluster socket: %m");
+		return -1;
+	}
+
+	/* Get the cluster members list */
+	get_members();
+	count_clvmds_running();
+
+	/* Create a lockspace for LV & VG locks to live in */
+	lockspace = dlm_create_lockspace(LOCKSPACE_NAME, 0600);
+	if (!lockspace) {
+		syslog(LOG_ERR, "Unable to create lockspace for CLVM: %m");
+		return -1;
+	}
+	dlm_ls_pthread_init(lockspace);
+	return 0;
+}
+
+int get_main_cluster_fd()
+{
+	return cluster_sock;
+}
+
+int get_num_nodes()
+{
+	return num_nodes;
+}
+
+/* send_message with the fd check removed */
+int cluster_send_message(void *buf, int msglen, char *csid, const char *errtext)
+{
+	struct iovec iov[2];
+	struct msghdr msg;
+	struct sockaddr_cl saddr;
+	int len = 0;
+
+	msg.msg_control = NULL;
+	msg.msg_controllen = 0;
+	msg.msg_iovlen = 1;
+	msg.msg_iov = iov;
+	msg.msg_flags = 0;
+	iov[0].iov_len = msglen;
+	iov[0].iov_base = buf;
+
+	saddr.scl_family = AF_CLUSTER;
+	saddr.scl_port = CLUSTER_PORT_CLVMD;
+	if (csid) {
+		msg.msg_name = &saddr;
+		msg.msg_namelen = sizeof(saddr);
+		memcpy(&saddr.scl_nodeid, csid, MAX_CSID_LEN);
+	} else {		/* Cluster broadcast */
+
+		msg.msg_name = NULL;
+		msg.msg_namelen = 0;
+	}
+
+	do {
+		len = sendmsg(cluster_sock, &msg, 0);
+		if (len < 0 && errno != EAGAIN)
+			log_error(errtext);
+
+	} while (len == -1 && errno == EAGAIN);
+	return len;
+}
+
+void get_our_csid(char *csid)
+{
+	int i;
+	memset(csid, 0, MAX_CSID_LEN);
+
+	for (i = 0; i < num_nodes; i++) {
+		if (nodes[i].us)
+			memcpy(csid, &nodes[i].node_id, MAX_CSID_LEN);
+	}
+}
+
+/* Call a callback routine for each node that known (down mean not running a clvmd) */
+int cluster_do_node_callback(struct local_client *client,
+			     void (*callback) (struct local_client *, char *,
+					       int))
+{
+	int i;
+	int somedown = 0;
+
+	for (i = 0; i < get_num_nodes(); i++) {
+		callback(client, (char *)&nodes[i].node_id, node_updown[nodes[i].node_id]);
+		if (!node_updown[nodes[i].node_id])
+			somedown = -1;
+	}
+	return somedown;
+}
+
+/* Process OOB message from the cluster socket,
+   this currently just means that a node has stopped listening on our port */
+static void process_oob_msg(char *buf, int len, int nodeid)
+{
+	char namebuf[256];
+	switch (buf[0]) {
+        case CLUSTER_OOB_MSG_PORTCLOSED:
+		name_from_nodeid(nodeid, namebuf);
+		log_notice("clvmd on node %s has died\n", namebuf);
+		DEBUGLOG("Got OOB message, removing node %s\n", namebuf);
+
+		node_updown[nodeid] = 0;
+		break;
+
+	case CLUSTER_OOB_MSG_STATECHANGE:
+		DEBUGLOG("Got OOB message, Cluster state change\n");
+		get_members();
+		break;
+	default:
+		/* ERROR */
+		DEBUGLOG("Got unknown OOB message: %d\n", buf[0]);
+	}
+}
+
+int cluster_fd_callback(struct local_client *client, char *buf, int len, char *csid,
+			struct local_client **new_client)
+{
+	struct iovec iov[2];
+	struct msghdr msg;
+	struct sockaddr_cl saddr;
+
+	/* We never return a new client */
+	*new_client = NULL;
+
+	msg.msg_control = NULL;
+	msg.msg_controllen = 0;
+	msg.msg_iovlen = 1;
+	msg.msg_iov = iov;
+	msg.msg_name = &saddr;
+	msg.msg_flags = 0;
+	msg.msg_namelen = sizeof(saddr);
+	iov[0].iov_len = len;
+	iov[0].iov_base = buf;
+
+	len = recvmsg(cluster_sock, &msg, MSG_OOB | O_NONBLOCK);
+	if (len < 0 && errno == EAGAIN)
+		return len;
+
+	DEBUGLOG("Read on cluster socket, len = %d\n", len);
+
+	/* A real error */
+	if (len < 0) {
+		log_error("read error on cluster socket: %m");
+		return 0;
+	}
+
+	/* EOF - we have left the cluster */
+	if (len == 0)
+		return 0;
+
+	/* Is it OOB? probably a node gone down */
+	if (msg.msg_flags & MSG_OOB) {
+		process_oob_msg(iov[0].iov_base, len, saddr.scl_nodeid);
+
+		/* Tell the upper layer to ignore this message */
+		len = -1;
+		errno = EAGAIN;
+	}
+	else {
+		memcpy(csid, &saddr.scl_nodeid, sizeof(saddr.scl_nodeid));
+		/* Send it back to clvmd */
+		process_message(client, buf, len, csid);
+	}
+	return len;
+}
+
+void add_up_node(char *csid)
+{
+	/* It's up ! */
+	int nodeid = nodeid_from_csid(csid);
+
+	if (nodeid >= max_updown_nodes) {
+	        int new_size = nodeid + 10;
+		int *new_updown = realloc(node_updown, new_size);
+
+		if (new_updown) {
+			node_updown = new_updown;
+			max_updown_nodes = new_size;
+			DEBUGLOG("realloced more space for nodes. now %d\n",
+				 max_updown_nodes);
+		} else {
+			log_error
+			    ("Realloc failed. Node status for clvmd will be wrong. quitting\n");
+			exit(999);
+		}
+	}
+	node_updown[nodeid] = 1;
+	DEBUGLOG("Added new node %d to updown list\n", nodeid);
+}
+
+void cluster_closedown()
+{
+	unlock_all();
+	dlm_release_lockspace(LOCKSPACE_NAME, lockspace, 1);
+	close(cluster_sock);
+}
+
+static int is_listening(int nodeid)
+{
+	struct cl_listen_request rq;
+	int status;
+
+	rq.port = CLUSTER_PORT_CLVMD;
+	rq.nodeid = nodeid;
+
+	do {
+		status = ioctl(cluster_sock, SIOCCLUSTER_ISLISTENING, &rq);
+		if (status < 0 && errno == EBUSY) {	/* Don't busywait */
+			sleep(1);
+			errno = EBUSY;	/* In case sleep trashes it */
+		}
+	}
+	while (status < 0 && errno == EBUSY);
+
+	return status;
+}
+
+/* Populate the list of CLVMDs running.
+   called only at startup time */
+void count_clvmds_running(void)
+{
+	int i;
+
+	for (i = 0; i < num_nodes; i++) {
+		node_updown[nodes[i].node_id] = is_listening(nodes[i].node_id);
+	}
+}
+
+/* Get a list of active cluster members */
+static void get_members()
+{
+	struct cl_cluster_nodelist nodelist;
+
+	num_nodes = ioctl(cluster_sock, SIOCCLUSTER_GETMEMBERS, 0);
+	if (num_nodes == -1) {
+		log_error("Unable to get node count");
+	} else {
+	        /* Not enough room for new nodes list ? */
+	        if (num_nodes > count_nodes && nodes) {
+			free(nodes);
+			nodes = NULL;
+		}
+
+		if (nodes == NULL) {
+		        count_nodes = num_nodes + 10; /* Overallocate a little */
+		        nodes = malloc(count_nodes * sizeof(struct cl_cluster_node));
+			if (!nodes) {
+			        log_error("Unable to allocate nodes array\n");
+				exit(5);
+			}
+		}
+		nodelist.max_members = count_nodes;
+		nodelist.nodes = nodes;
+
+		num_nodes = ioctl(cluster_sock, SIOCCLUSTER_GETMEMBERS, &nodelist);
+		if (num_nodes <= 0) {
+		        log_error("Unable to get node details");
+			exit(6);
+		}
+
+		/* Sanity check struct */
+		if (nodes[0].size != sizeof(struct cl_cluster_node)) {
+			log_error
+			    ("sizeof(cl_cluster_node) does not match size returned from the kernel: aborting\n");
+			exit(10);
+		}
+
+		if (node_updown == NULL) {
+			node_updown =
+			    (int *) malloc(sizeof(int) *
+					   max(num_nodes, max_updown_nodes));
+			memset(node_updown, 0,
+			       sizeof(int) * max(num_nodes, max_updown_nodes));
+		}
+	}
+}
+
+/* Convert a node name to a CSID */
+int csid_from_name(char *csid, char *name)
+{
+	int i;
+
+	for (i = 0; i < num_nodes; i++) {
+		if (strcmp(name, nodes[i].name) == 0) {
+			memcpy(csid, &nodes[i].node_id, MAX_CSID_LEN);
+			return 0;
+		}
+	}
+	return -1;
+}
+
+/* Convert a CSID to a node name */
+int name_from_csid(char *csid, char *name)
+{
+	int i;
+
+	for (i = 0; i < num_nodes; i++) {
+		if (memcmp(csid, &nodes[i].node_id, MAX_CSID_LEN) == 0) {
+			strcpy(name, nodes[i].name);
+			return 0;
+		}
+	}
+	/* Who?? */
+	strcpy(name, "Unknown");
+	return -1;
+}
+
+/* Convert a node ID to a node name */
+int name_from_nodeid(int nodeid, char *name)
+{
+	int i;
+
+	for (i = 0; i < num_nodes; i++) {
+		if (nodeid == nodes[i].node_id) {
+			strcpy(name, nodes[i].name);
+			return 0;
+		}
+	}
+	/* Who?? */
+	strcpy(name, "Unknown");
+	return -1;
+}
+
+/* Convert a CSID to a node ID */
+static int nodeid_from_csid(char *csid)
+{
+        int nodeid;
+
+	memcpy(&nodeid, csid, MAX_CSID_LEN);
+
+	return nodeid;
+}
+
+int is_quorate()
+{
+	return ioctl(cluster_sock, SIOCCLUSTER_ISQUORATE, 0);
+}
+
+static void sync_ast_routine(void *arg)
+{
+	struct lock_wait *lwait = arg;
+
+	pthread_mutex_lock(&lwait->mutex);
+	pthread_cond_signal(&lwait->cond);
+	pthread_mutex_unlock(&lwait->mutex);
+}
+
+int sync_lock(const char *resource, int mode, int flags, int *lockid)
+{
+	int status;
+	struct lock_wait lwait;
+
+	if (!lockid) {
+		errno = EINVAL;
+		return -1;
+	}
+
+	DEBUGLOG("sync_lock: '%s' mode:%d flags=%d\n", resource,mode,flags);
+	/* Conversions need the lockid in the LKSB */
+	if (flags & LKF_CONVERT)
+		lwait.lksb.sb_lkid = *lockid;
+
+	pthread_cond_init(&lwait.cond, NULL);
+	pthread_mutex_init(&lwait.mutex, NULL);
+	pthread_mutex_lock(&lwait.mutex);
+
+	status = dlm_ls_lock(lockspace,
+			     mode,
+			     &lwait.lksb,
+			     flags,
+			     resource,
+			     strlen(resource),
+			     0, sync_ast_routine, &lwait, NULL, NULL);
+	if (status)
+		return status;
+
+	/* Wait for it to complete */
+	pthread_cond_wait(&lwait.cond, &lwait.mutex);
+	pthread_mutex_unlock(&lwait.mutex);
+
+	*lockid = lwait.lksb.sb_lkid;
+
+	errno = lwait.lksb.sb_status;
+	DEBUGLOG("sync_lock: returning lkid %x\n", *lockid);
+	if (lwait.lksb.sb_status)
+		return -1;
+	else
+		return 0;
+}
+
+int sync_unlock(const char *resource /* UNUSED */, int lockid)
+{
+	int status;
+	struct lock_wait lwait;
+
+	DEBUGLOG("sync_unlock: '%s' lkid:%x\n", resource, lockid);
+
+	pthread_cond_init(&lwait.cond, NULL);
+	pthread_mutex_init(&lwait.mutex, NULL);
+	pthread_mutex_lock(&lwait.mutex);
+
+	status = dlm_ls_unlock(lockspace, lockid, 0, &lwait.lksb, &lwait);
+
+	if (status)
+		return status;
+
+	/* Wait for it to complete */
+	pthread_cond_wait(&lwait.cond, &lwait.mutex);
+	pthread_mutex_unlock(&lwait.mutex);
+
+	errno = lwait.lksb.sb_status;
+	if (lwait.lksb.sb_status != EUNLOCK)
+		return -1;
+	else
+		return 0;
+
+}

daemons/clvmd/clvmd-command.c

@@ -0,0 +1,285 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/*
+
+  CLVMD Cluster LVM daemon command processor.
+
+  To add commands to the daemon simply add a processor in do_command and return
+  and messages back in buf and the length in *retlen. The initial value of
+  buflen is the maximum size of the buffer. if buf is not large enough then it
+  may be reallocated by the functions in here to a suitable size bearing in
+  mind that anything larger than the passed-in size will have to be returned
+  using the system LV and so performance will suffer.
+
+  The status return will be negated and passed back to the originating node.
+
+  pre- and post- command routines are called only on the local node. The
+  purpose is primarily to get and release locks, though the pre- routine should
+  also do any other local setups required by the command (if any) and can
+  return a failure code that prevents the command from being distributed around
+  the cluster
+
+  The pre- and post- routines are run in their own thread so can block as long
+  they like, do_command is run in the main clvmd thread so should not block for
+  too long. If the pre-command returns an error code (!=0) then the command
+  will not be propogated around the cluster but the post-command WILL be called
+
+  Also note that the pre and post routine are *always* called on the local
+  node, even if the command to be executed was only requested to run on a
+  remote node. It may peek inside the client structure to check the status of
+  the command.
+
+  The clients of the daemon must, naturally, understand the return messages and
+  codes.
+
+  Routines in here may only READ the values in the client structure passed in
+  apart from client->private which they are free to do what they like with.
+
+*/
+
+#include <pthread.h>
+#include <sys/types.h>
+#include <sys/utsname.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <fcntl.h>
+#include <string.h>
+#include <stddef.h>
+#include <unistd.h>
+#include <errno.h>
+
+#include "list.h"
+#include "hash.h"
+#include "locking.h"
+#include "log.h"
+#include "lvm-functions.h"
+#include "clvmd-comms.h"
+#include "clvm.h"
+#include "clvmd.h"
+#include "libdlm.h"
+
+/* This is where all the real work happens:
+   NOTE: client will be NULL when this is executed on a remote node */
+int do_command(struct local_client *client, struct clvm_header *msg, int msglen,
+	       char **buf, int buflen, int *retlen)
+{
+	char *args = msg->node + strlen(msg->node) + 1;
+	int arglen = msglen - sizeof(struct clvm_header) - strlen(msg->node);
+	int status = 0;
+	char *lockname;
+	struct utsname nodeinfo;
+	unsigned char lock_cmd;
+	unsigned char lock_flags;
+
+	/* Do the command */
+	switch (msg->cmd) {
+		/* Just a test message */
+	case CLVMD_CMD_TEST:
+		if (arglen > buflen) {
+			buflen = arglen + 200;
+			*buf = realloc(*buf, buflen);
+		}
+		uname(&nodeinfo);
+		*retlen = 1 + snprintf(*buf, buflen, "TEST from %s: %s v%s",
+				       nodeinfo.nodename, args,
+				       nodeinfo.release);
+		break;
+
+	case CLVMD_CMD_LOCK_VG:
+		/* Check to see if the VG is in use by LVM1 */
+		status = do_check_lvm1(&args[2]);
+		break;
+
+	case CLVMD_CMD_LOCK_LV:
+		/* This is the biggie */
+		lock_cmd = args[0];
+		lock_flags = args[1];
+		lockname = &args[2];
+		status = do_lock_lv(lock_cmd, lock_flags, lockname);
+		/* Replace EIO with something less scary */
+		if (status == EIO) {
+			*retlen =
+			    1 + snprintf(*buf, buflen,
+					 "Internal lvm error, check syslog");
+			return EIO;
+		}
+		break;
+
+	default:
+		/* Won't get here because command is validated in pre_command */
+		break;
+	}
+
+	/* Check the status of the command and return the error text */
+	if (status) {
+		*retlen = 1 + snprintf(*buf, buflen, strerror(status));
+	}
+
+	return status;
+
+}
+
+static int lock_vg(struct local_client *client)
+{
+    struct hash_table *lock_hash;
+    struct clvm_header *header =
+	(struct clvm_header *) client->bits.localsock.cmd;
+    unsigned char lock_cmd;
+    unsigned char lock_flags;
+    char *args = header->node + strlen(header->node) + 1;
+    int lkid;
+    int status = 0;
+    char *lockname;
+
+    /* Keep a track of VG locks in our own hash table. In current
+       practice there should only ever be more than two VGs locked
+       if a user tries to merge lots of them at once */
+    if (client->bits.localsock.private) {
+	lock_hash = (struct hash_table *)client->bits.localsock.private;
+    }
+    else {
+	lock_hash = hash_create(3);
+	if (!lock_hash)
+	    return ENOMEM;
+	client->bits.localsock.private = (void *)lock_hash;
+    }
+
+    lock_cmd = args[0];
+    lock_flags = args[1];
+    lockname = &args[2];
+    DEBUGLOG("doing PRE command LOCK_VG '%s' at %x (client=%p)\n", lockname, lock_cmd, client);
+
+    if (lock_cmd == LCK_UNLOCK) {
+
+	lkid = (int)(long)hash_lookup(lock_hash, lockname);
+	if (lkid == 0)
+	    return EINVAL;
+
+	status = sync_unlock(lockname, lkid);
+	if (status)
+	    status = errno;
+	else
+	    hash_remove(lock_hash, lockname);
+    }
+    else {
+
+	status = sync_lock(lockname, (int)lock_cmd, (int)lock_flags, &lkid);
+	if (status)
+	    status = errno;
+	else
+	    hash_insert(lock_hash, lockname, (void *)lkid);
+    }
+
+    return status;
+}
+
+
+/* Pre-command is a good place to get locks that are needed only for the duration
+   of the commands around the cluster (don't forget to free them in post-command),
+   and to sanity check the command arguments */
+int do_pre_command(struct local_client *client)
+{
+	struct clvm_header *header =
+	    (struct clvm_header *) client->bits.localsock.cmd;
+	unsigned char lock_cmd;
+	unsigned char lock_flags;
+	char *args = header->node + strlen(header->node) + 1;
+	int lockid;
+	int status = 0;
+	char *lockname;
+
+	switch (header->cmd) {
+	case CLVMD_CMD_TEST:
+		status = sync_lock("CLVMD_TEST", LKM_EXMODE, 0, &lockid);
+		client->bits.localsock.private = (void *) lockid;
+		break;
+
+	case CLVMD_CMD_LOCK_VG:
+       	        status = lock_vg(client);
+		break;
+
+	case CLVMD_CMD_LOCK_LV:
+		lock_cmd = args[0];
+		lock_flags = args[1];
+		lockname = &args[2];
+		status = pre_lock_lv(lock_cmd, lock_flags, lockname);
+		break;
+
+	default:
+		log_error("Unknown command %d received\n", header->cmd);
+		status = EINVAL;
+	}
+	return status;
+}
+
+/* Note that the post-command routine is called even if the pre-command or the real command
+   failed */
+int do_post_command(struct local_client *client)
+{
+	struct clvm_header *header =
+	    (struct clvm_header *) client->bits.localsock.cmd;
+	int status = 0;
+	unsigned char lock_cmd;
+	unsigned char lock_flags;
+	char *args = header->node + strlen(header->node) + 1;
+	char *lockname;
+
+	switch (header->cmd) {
+	case CLVMD_CMD_TEST:
+		status =
+		    sync_unlock("CLVMD_TEST", (int) (long) client->bits.localsock.private);
+		client->bits.localsock.private = 0;
+		break;
+
+	case CLVMD_CMD_LOCK_VG:
+		/* Nothing to do here */
+		break;
+
+	case CLVMD_CMD_LOCK_LV:
+		lock_cmd = args[0];
+		lock_flags = args[1];
+		lockname = &args[2];
+		status = post_lock_lv(lock_cmd, lock_flags, lockname);
+		break;
+	}
+	return status;
+}
+
+
+/* Called when the client is about to be deleted */
+void cmd_client_cleanup(struct local_client *client)
+{
+    if (client->bits.localsock.private) {
+
+	struct hash_node *v;
+	struct hash_table *lock_hash =
+	    (struct hash_table *)client->bits.localsock.private;
+
+	hash_iterate(v, lock_hash) {
+		int lkid = (int)(long)hash_get_data(lock_hash, v);
+
+		DEBUGLOG("cleanup: Unlocking lkid %x\n", lkid);
+		sync_unlock("DUMMY", lkid);
+	}
+
+	hash_destroy(lock_hash);
+	client->bits.localsock.private = 0;
+    }
+}

daemons/clvmd/clvmd-comms.h

@@ -0,0 +1,55 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/*
+ * Abstraction layer for clvmd cluster communications
+ */
+
+#ifndef _CLVMD_COMMS_H
+#define _CLVMD_COMMS_H
+
+struct local_client;
+
+extern int cluster_send_message(void *buf, int msglen, char *csid,
+				const char *errtext);
+extern int name_from_csid(char *csid, char *name);
+extern int csid_from_name(char *csid, char *name);
+extern int get_num_nodes(void);
+extern int cluster_fd_callback(struct local_client *fd, char *buf, int len,
+			       char *csid, struct local_client **new_client);
+extern int init_cluster(void);
+extern int get_main_cluster_fd(void);	/* gets accept FD or cman cluster socket */
+extern int cluster_do_node_callback(struct local_client *client,
+				    void (*callback) (struct local_client *,
+						      char *csid, int node_up));
+extern int is_quorate(void);
+
+extern void get_our_csid(char *csid);
+extern void add_up_node(char *csid);
+extern void cluster_closedown(void);
+
+extern int sync_lock(const char *resource, int mode, int flags, int *lockid);
+extern int sync_unlock(const char *resource, int lockid);
+
+#ifdef USE_GULM
+#include "tcp-comms.h"
+#else
+/* cman */
+#include "cnxman-socket.h"
+#define MAX_CSID_LEN 4
+#endif
+
+
+#endif

daemons/clvmd/clvmd-gulm.c

@@ -0,0 +1,908 @@
+/******************************************************************************
+*******************************************************************************
+**
+**  Copyright (C) Sistina Software, Inc.  2002-2003  All rights reserved.
+**
+*******************************************************************************
+******************************************************************************/
+
+/* This provides the interface between clvmd and gulm as the cluster
+ * and lock manager.
+ *
+ * It also provides the "liblm" functions too as it's hard (and pointless)
+ * to seperate them out when using gulm.
+ *
+ * What it does /not/ provide is the communications between clvmd daemons
+ * on the cluster nodes. That is done in tcp-comms.c
+ */
+
+
+#include <pthread.h>
+#include <sys/types.h>
+#include <sys/utsname.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/file.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <signal.h>
+#include <fcntl.h>
+#include <string.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <errno.h>
+#include <utmpx.h>
+#include <syslog.h>
+#include <assert.h>
+
+#include "ccs.h"
+#include "list.h"
+#include "locking.h"
+#include "log.h"
+#include "clvm.h"
+#include "clvmd-comms.h"
+#include "clvmd.h"
+#include "hash.h"
+#include "clvmd-gulm.h"
+#include "libgulm.h"
+#include "hash.h"
+
+/* Hash list of nodes in the cluster */
+static struct hash_table *node_hash;
+
+/* hash list of outstanding lock requests */
+static struct hash_table *lock_hash;
+
+/* Copy of the current core state */
+static uint8_t current_corestate;
+
+/* Number of active nodes */
+static int num_nodes;
+
+static char *cluster_name;
+
+static pthread_mutex_t lock_start_mutex;
+static volatile int lock_start_flag;
+
+struct node_info
+{
+    enum {NODE_UNKNOWN, NODE_DOWN, NODE_UP, NODE_CLVMD} state;
+    char name[MAX_CLUSTER_MEMBER_NAME_LEN];
+};
+
+struct lock_wait
+{
+    pthread_cond_t cond;
+    pthread_mutex_t mutex;
+    int status;
+};
+
+/* Forward */
+static int read_from_core_sock(struct local_client *client, char *buf, int len, char *csid,
+			       struct local_client **new_client);
+static int read_from_lock_sock(struct local_client *client, char *buf, int len, char *csid,
+			       struct local_client **new_client);
+static int get_all_cluster_nodes(void);
+
+/* In tcp-comms.c */
+extern struct hash_table *sock_hash;
+
+static int add_internal_client(int fd, fd_callback_t callback)
+{
+    struct local_client *client;
+
+    DEBUGLOG("Add_internal_client, fd = %d\n", fd);
+
+    /* Add a GULM file descriptor it to the main loop */
+    client = malloc(sizeof(struct local_client));
+    if (!client)
+    {
+	DEBUGLOG("malloc failed\n");
+	return -1;
+    }
+
+    memset(client, 0, sizeof(struct local_client));
+    client->fd = fd;
+    client->type = CLUSTER_INTERNAL;
+    client->callback = callback;
+    add_client(client);
+
+    return 0;
+}
+
+/* Gulm library handle */
+static gulm_interface_p gulm_if;
+static lg_core_callbacks_t core_callbacks;
+static lg_lockspace_callbacks_t lock_callbacks;
+
+static void badsig_handler(int sig)
+{
+    DEBUGLOG("got sig %d\n", sig);
+    cluster_closedown();
+    exit(0);
+}
+
+static void sighup_handler(int sig)
+{
+    DEBUGLOG("got SIGHUP\n");
+
+    /* Re-read CCS node list */
+    get_all_cluster_nodes();
+}
+
+int init_cluster()
+{
+    int status;
+    int ccs_h;
+    int port = 0;
+    char *portstr;
+
+    /* Get cluster name from CCS */
+    ccs_h = ccs_force_connect(NULL, 0);
+    if (ccs_h < 0)
+    {
+	syslog(LOG_ERR, "Cannot login in to CCSD server\n");
+	return -1;
+    }
+
+    ccs_get(ccs_h, "//cluster/@name", &cluster_name);
+    DEBUGLOG("got cluster name %s\n", cluster_name);
+
+    if (!ccs_get(ccs_h, "//cluster/clvm/@port", &portstr))
+    {
+	port = atoi(portstr);
+	free(portstr);
+	DEBUGLOG("got port number %d\n", port);
+
+	if (port <= 0 && port >= 65536)
+	    port = 0;
+    }
+
+    ccs_disconnect(ccs_h);
+
+    /* Block locking until we are logged in */
+    pthread_mutex_init(&lock_start_mutex, NULL);
+    pthread_mutex_lock(&lock_start_mutex);
+    lock_start_flag = 1;
+
+    node_hash = hash_create(100);
+    lock_hash = hash_create(10);
+
+    /* Get all nodes from CCS */
+    if (get_all_cluster_nodes())
+	return -1;
+
+    /* Initialise GULM library */
+    status = lg_initialize(&gulm_if, cluster_name, "clvmd");
+    if (status)
+    {
+	DEBUGLOG("lg_initialize failed: %d\n", status);
+	return status;
+    }
+
+    /* Connect to core - we are not "important" :-) */
+    status = lg_core_login(gulm_if, 0);
+    if (status)
+    {
+	DEBUGLOG("lg_core_login failed: %d\n", status);
+	return status;
+    }
+
+    /* Initialise the inter-node comms */
+    status = init_comms(port);
+    if (status)
+	return status;
+
+    /* Add core FD to the list */
+    status = add_internal_client(lg_core_selector(gulm_if), read_from_core_sock);
+    if (status)
+    {
+	DEBUGLOG("can't allocate client space\n");
+	return status;
+    }
+
+    /* Connect to the lock server */
+    if (lg_lock_login(gulm_if, "CLVM"))
+    {
+	syslog(LOG_ERR, "Cannot login in to LOCK server\n");
+	DEBUGLOG("Cannot login in to LOCK server\n");
+	exit(88);
+    }
+
+    /* Add lockspace FD to the list */
+    status = add_internal_client(lg_lock_selector(gulm_if), read_from_lock_sock);
+    if (status)
+    {
+	DEBUGLOG("can't allocate client space\n");
+	exit(status);
+    }
+
+    /* Request a list of nodes, we can;t really do anything until
+       this comes back */
+    status = lg_core_nodelist(gulm_if);
+    if (status)
+    {
+	DEBUGLOG("lg_core_nodelist failed: %d\n", status);
+	return status;
+    }
+
+    /* So I can kill it without taking GULM down too */
+    signal(SIGINT, badsig_handler);
+    signal(SIGTERM, badsig_handler);
+
+    /* Re-read the node list on SIGHUP */
+    signal(SIGHUP, sighup_handler);
+
+    return 0;
+}
+
+void cluster_closedown()
+{
+    DEBUGLOG("cluster_closedown\n");
+    lg_lock_logout(gulm_if);
+    lg_core_logout(gulm_if);
+    lg_release(gulm_if);
+}
+
+/* Expire locks for a named node, or us */
+#define GIO_KEY_SIZE 46
+static void drop_expired_locks(char *nodename)
+{
+    struct utsname nodeinfo;
+    uint8_t mask[GIO_KEY_SIZE];
+
+    memset(mask, 0xff, GIO_KEY_SIZE);
+
+    if (!nodename)
+    {
+	uname(&nodeinfo);
+	nodename = nodeinfo.nodename;
+    }
+
+    if (lg_lock_drop_exp(gulm_if, nodename, mask, GIO_KEY_SIZE))
+    {
+	DEBUGLOG("Error calling lg_lock_drop_exp()\n");
+    }
+}
+
+
+static int read_from_core_sock(struct local_client *client, char *buf, int len, char *csid,
+			       struct local_client **new_client)
+{
+    int status;
+
+    *new_client = NULL;
+    status = lg_core_handle_messages(gulm_if, &core_callbacks, NULL);
+    return status<0 ? status : 1;
+}
+
+static int read_from_lock_sock(struct local_client *client, char *buf, int len, char *csid,
+			       struct local_client **new_client)
+{
+    int status;
+
+    *new_client = NULL;
+    status = lg_lock_handle_messages(gulm_if, &lock_callbacks, NULL);
+    return status<0 ? status : 1;
+}
+
+
+/* CORE callback routines */
+static int core_login_reply(void *misc, uint64_t gen, uint32_t error, uint32_t rank, uint8_t corestate)
+{
+   DEBUGLOG("CORE Got a Login reply.  gen:%lld err:%d rank:%d corestate:%d\n",
+         gen, error, rank, corestate);
+
+   if (error)
+       exit(error);
+
+   current_corestate = corestate;
+   return 0;
+}
+
+static void set_node_state(struct node_info *ninfo, char *csid, uint8_t nodestate)
+{
+    if (nodestate == lg_core_Logged_in)
+    {
+	/* Don't clobber NODE_CLVMD state */
+	if (ninfo->state != NODE_CLVMD)
+	{
+	    if (ninfo->state == NODE_UNKNOWN ||
+		ninfo->state == NODE_DOWN)
+		num_nodes++;
+
+	    ninfo->state = NODE_UP;
+	}
+    }
+    else
+    {
+	if (nodestate == lg_core_Expired ||
+	    nodestate == lg_core_Fenced ||
+	    nodestate == lg_core_Logged_out)
+	{
+	    if (ninfo->state != NODE_DOWN)
+		num_nodes--;
+	    ninfo->state = NODE_DOWN;
+	    tcp_remove_client(csid);
+	}
+    }
+    DEBUGLOG("set_node_state, '%s' state = %d, num_nodes=%d\n",
+	     ninfo->name, ninfo->state, num_nodes);
+}
+
+static struct node_info *add_or_set_node(char *name, struct in6_addr *ip, uint8_t state)
+{
+    struct node_info *ninfo;
+
+    ninfo = hash_lookup_binary(node_hash, (char *)ip, MAX_CSID_LEN);
+    if (!ninfo)
+    {
+	/* If we can't find that node then re-read the config file in case it
+	   was added after we were started */
+	DEBUGLOG("Node %s not found, re-reading config file\n", name);
+	get_all_cluster_nodes();
+
+	/* Now try again */
+	ninfo = hash_lookup_binary(node_hash, (char *)ip, MAX_CSID_LEN);
+	if (!ninfo)
+	{
+	    DEBUGLOG("Ignoring node %s, not part of the SAN cluster\n", name);
+	    return NULL;
+	}
+    }
+
+    set_node_state(ninfo, (char *)&ip, state);
+
+    return ninfo;
+}
+
+static int core_nodelist(void *misc, lglcb_t type, char *name, struct in6_addr *ip, uint8_t state)
+{
+    DEBUGLOG("CORE nodelist\n");
+
+    if (type == lglcb_start)
+    {
+	DEBUGLOG("Got Nodelist, start\n");
+    }
+    else
+    {
+	if (type == lglcb_item)
+	{
+	    DEBUGLOG("Got nodelist, item: %s, %#x\n", name, state);
+
+	    add_or_set_node(name, ip, state);
+	}
+	else
+	{
+	    if (type == lglcb_stop)
+	    {
+		char ourcsid[MAX_CSID_LEN];
+
+		DEBUGLOG("Got Nodelist, stop\n");
+		clvmd_cluster_init_completed();
+
+		/* Mark ourself as up */
+		get_our_csid(ourcsid);
+		add_up_node(ourcsid);
+	    }
+	    else
+	    {
+		DEBUGLOG("Unknown lglcb_t %#x\n", type);
+	    }
+	}
+    }
+
+    return 0;
+}
+
+static int core_statechange(void *misc, uint8_t corestate, uint8_t quorate, struct in6_addr *masterip, char *mastername)
+{
+    DEBUGLOG("CORE Got statechange  corestate:%#x mastername:%s\n",
+	     corestate, mastername);
+
+    current_corestate = corestate;
+    return 0;
+}
+
+static int core_nodechange(void *misc, char *nodename, struct in6_addr *nodeip, uint8_t nodestate)
+{
+    struct node_info *ninfo;
+
+    DEBUGLOG("CORE node change, name=%s, state = %d\n", nodename, nodestate);
+
+    /* If we don't get nodeip here, try a lookup by name */
+    if (!nodeip)
+	csid_from_name((char *)nodeip, nodename);
+    if (!nodeip)
+	return 0;
+
+    ninfo = add_or_set_node(nodename, nodeip, nodestate);
+    if (!ninfo)
+	return 0;
+
+    /* Check if we need to drop any expired locks */
+    if (ninfo->state == NODE_DOWN)
+    {
+	drop_expired_locks(nodename);
+    }
+
+    return 0;
+}
+static int core_error(void *misc, uint32_t err)
+{
+    DEBUGLOG("CORE error: %d\n", err);
+    // Not sure what happens here
+    return 0;
+}
+
+/* LOCK callback routines */
+static int lock_login_reply(void *misc, uint32_t error, uint8_t which)
+{
+    DEBUGLOG("LOCK Got a Login reply.  err:%d which:%d\n",
+	     error, which);
+
+    if (error)
+	exit(error);
+
+    /* Drop any expired locks for us that might be hanging around */
+    drop_expired_locks(NULL);
+
+    /* Enable locking operations in other threads */
+    if (lock_start_flag)
+    {
+	lock_start_flag = 0;
+	pthread_mutex_unlock(&lock_start_mutex);
+    }
+
+    return 0;
+}
+
+static int lock_lock_state(void *misc, uint8_t *key, uint16_t keylen,
+			   uint64_t subid, uint64_t start, uint64_t stop,
+			   uint8_t state, uint32_t flags, uint32_t error,
+			   uint8_t *LVB, uint16_t LVBlen)
+{
+    struct lock_wait *lwait;
+
+    DEBUGLOG("LOCK lock state: %s, error = %d\n", key, error);
+
+    lwait = hash_lookup(lock_hash, key);
+    if (!lwait)
+    {
+	DEBUGLOG("Can't find hash entry for resource %s\n", key);
+	return 0;
+    }
+    lwait->status = error;
+    pthread_mutex_lock(&lwait->mutex);
+    pthread_cond_signal(&lwait->cond);
+    pthread_mutex_unlock(&lwait->mutex);
+
+    return 0;
+}
+static int lock_error(void *misc, uint32_t err)
+{
+    DEBUGLOG("LOCK error: %d\n", err);
+    // Not sure what happens here
+    return 0;
+}
+
+
+/* CORE callbacks */
+static lg_core_callbacks_t core_callbacks = {
+    .login_reply  = core_login_reply,
+    .nodelist     = core_nodelist,
+    .statechange  = core_statechange,
+    .nodechange   = core_nodechange,
+    .error        = core_error,
+};
+
+/* LOCK callbacks */
+static lg_lockspace_callbacks_t lock_callbacks = {
+    .login_reply   = lock_login_reply,
+    .lock_state    = lock_lock_state,
+    .error         = lock_error,
+};
+
+/* Allow tcp-comms to loop round the list of active nodes */
+int get_next_node_csid(void **context, char *csid)
+{
+    struct node_info *ninfo = NULL;
+
+    /* First node */
+    if (!*context)
+    {
+	*context = hash_get_first(node_hash);
+    }
+    else
+    {
+	*context = hash_get_next(node_hash, *context);
+    }
+    if (*context)
+	ninfo = hash_get_data(node_hash, *context);
+
+    /* Find a node that is UP */
+    while (*context && ninfo->state == NODE_DOWN)
+    {
+	*context = hash_get_next(node_hash, *context);
+	if (*context)
+	{
+	    ninfo = hash_get_data(node_hash, *context);
+	}
+    }
+
+    if (!*context || ninfo->state == NODE_DOWN)
+    {
+	return 0;
+    }
+
+    memcpy(csid, hash_get_key(node_hash, *context), MAX_CSID_LEN);
+    return 1;
+}
+
+int name_from_csid(char *csid, char *name)
+{
+    struct node_info *ninfo;
+
+    ninfo = hash_lookup_binary(node_hash, csid, MAX_CSID_LEN);
+    if (!ninfo)
+    {
+        sprintf(name, "UNKNOWN %s", print_csid(csid));
+	return -1;
+    }
+
+    strcpy(name, ninfo->name);
+    return 0;
+}
+
+
+int csid_from_name(char *csid, char *name)
+{
+    struct hash_node *hn;
+    struct node_info *ninfo;
+
+    hash_iterate(hn, node_hash)
+    {
+	ninfo = hash_get_data(node_hash, hn);
+	if (strcmp(ninfo->name, name) == 0)
+	{
+	    memcpy(csid, hash_get_key(node_hash, hn), MAX_CSID_LEN);
+	    return 0;
+	}
+    }
+    return -1;
+}
+
+int get_num_nodes()
+{
+    DEBUGLOG("num_nodes = %d\n", num_nodes);
+    return num_nodes;
+}
+
+/* Node is now known to be running a clvmd */
+void add_up_node(char *csid)
+{
+    struct node_info *ninfo;
+
+    ninfo = hash_lookup_binary(node_hash, csid, MAX_CSID_LEN);
+    if (!ninfo)
+	return;
+
+    ninfo->state = NODE_CLVMD;
+    return;
+
+}
+/* Node is now known to be NOT running a clvmd */
+void add_down_node(char *csid)
+{
+    struct node_info *ninfo;
+
+    ninfo = hash_lookup_binary(node_hash, csid, MAX_CSID_LEN);
+    if (!ninfo)
+	return;
+
+    /* Only set it to UP if it was previously known to be
+       running clvmd - gulm may set it DOWN quite soon */
+    if (ninfo->state == NODE_CLVMD)
+	ninfo->state = NODE_UP;
+    return;
+
+}
+
+/* Call a callback for each node, so the caller knows whether it's up or down */
+int cluster_do_node_callback(struct local_client *master_client,
+			     void (*callback)(struct local_client *, char *csid, int node_up))
+{
+    struct hash_node *hn;
+    struct node_info *ninfo;
+
+    hash_iterate(hn, node_hash)
+    {
+	char csid[MAX_CSID_LEN];
+	struct local_client *client;
+
+	ninfo = hash_get_data(node_hash, hn);
+	memcpy(csid, hash_get_key(node_hash, hn), MAX_CSID_LEN);
+
+	DEBUGLOG("down_callback. node %s, state = %d\n", ninfo->name, ninfo->state);
+
+	client = hash_lookup_binary(sock_hash, csid, MAX_CSID_LEN);
+	if (client)
+	    callback(master_client, csid, ninfo->state == NODE_CLVMD);
+    }
+    return 0;
+}
+
+/* Convert gulm error codes to unix errno numbers */
+static int gulm_to_errno(int gulm_ret)
+{
+    switch (gulm_ret)
+    {
+    case lg_err_TryFailed:
+	errno = EAGAIN;
+	break;
+
+    case lg_err_AlreadyPend:
+	errno = EBUSY;
+
+	/* More?? */
+    default:
+	errno = EINVAL;
+    }
+
+    return gulm_ret ? -1 : 0;
+}
+
+/* Real locking */
+static int _lock_resource(char *resource, int mode, int flags, int *lockid)
+{
+    int status;
+    struct lock_wait lwait;
+
+    /* Wait until the lock module is ready */
+    if (lock_start_flag)
+    {
+	pthread_mutex_lock(&lock_start_mutex);
+	pthread_mutex_unlock(&lock_start_mutex);
+    }
+
+    pthread_cond_init(&lwait.cond, NULL);
+    pthread_mutex_init(&lwait.mutex, NULL);
+    pthread_mutex_lock(&lwait.mutex);
+
+    /* This needs to be converted from DLM/LVM2 value for GULM */
+    if (flags == LCK_NONBLOCK) flags = lg_lock_flag_Try;
+
+    hash_insert(lock_hash, resource, &lwait);
+    DEBUGLOG("lock_resource '%s', flags=%d, mode=%d\n", resource, flags, mode);
+
+    status = lg_lock_state_req(gulm_if, resource, strlen(resource)+1,
+			       0, 0, 0,
+			       mode, flags, NULL, 0);
+    if (status)
+    {
+	DEBUGLOG("lg_lock_state returned %d\n", status);
+	return status;
+    }
+
+    /* Wait for it to complete */
+    pthread_cond_wait(&lwait.cond, &lwait.mutex);
+    pthread_mutex_unlock(&lwait.mutex);
+
+    hash_remove(lock_hash, resource);
+    DEBUGLOG("lock-resource returning %d\n", lwait.status);
+
+    return gulm_to_errno(lwait.status);
+}
+
+
+static int _unlock_resource(char *resource, int lockid)
+{
+    int status;
+    struct lock_wait lwait;
+
+    pthread_cond_init(&lwait.cond, NULL);
+    pthread_mutex_init(&lwait.mutex, NULL);
+    pthread_mutex_lock(&lwait.mutex);
+
+    hash_insert(lock_hash, resource, &lwait);
+
+    DEBUGLOG("unlock_resource %s\n", resource);
+    status = lg_lock_state_req(gulm_if, resource, strlen(resource)+1,
+			       0, 0, 0,
+			       lg_lock_state_Unlock, 0, NULL, 0);
+
+    if (status)
+    {
+	DEBUGLOG("lg_lock_state(unlock) returned %d\n", status);
+	return status;
+    }
+
+    /* Wait for it to complete */
+
+    pthread_cond_wait(&lwait.cond, &lwait.mutex);
+    pthread_mutex_unlock(&lwait.mutex);
+
+    hash_remove(lock_hash, resource);
+
+    return gulm_to_errno(lwait.status);
+}
+
+
+/* These two locking functions MUST be called in a seperate thread from
+   the clvmd main loop because they expect to be woken up by it.
+
+   These are abstractions around the real locking functions (above)
+   as we need to emulate the DLM's EX/PW/CW interaction with GULM using
+   two locks.
+   To aid unlocking, we store the lock mode in the lockid (as GULM
+   doesn't use this).
+*/
+int sync_lock(const char *resource, int mode, int flags, int *lockid)
+{
+    int status;
+    char lock1[strlen(resource)+3];
+    char lock2[strlen(resource)+3];
+
+    snprintf(lock1, sizeof(lock1), "%s-1", resource);
+    snprintf(lock2, sizeof(lock2), "%s-2", resource);
+
+    switch (mode)
+    {
+    case LCK_EXCL:
+	status = _lock_resource(lock1, lg_lock_state_Exclusive, flags, lockid);
+	if (status)
+	    goto out;
+
+	/* If we can't get this lock then bail out */
+	status = _lock_resource(lock2, lg_lock_state_Exclusive, LCK_NONBLOCK, lockid);
+        if (status == lg_err_TryFailed)
+        {
+           _unlock_resource(lock1, *lockid);
+           status = -1;
+           errno = EAGAIN;
+        }
+	break;
+
+    case LCK_READ:
+	status = _lock_resource(lock1, lg_lock_state_Shared, flags, lockid);
+	break;
+
+    case LCK_WRITE:
+	status = _lock_resource(lock2, lg_lock_state_Exclusive, flags, lockid);
+	break;
+
+    default:
+	status = -1;
+	errno = EINVAL;
+	break;
+    }
+ out:
+    *lockid = mode;
+    return status;
+}
+
+int sync_unlock(const char *resource, int lockid)
+{
+    int status = 0;
+    char lock1[strlen(resource)+3];
+    char lock2[strlen(resource)+3];
+
+    snprintf(lock1, sizeof(lock1), "%s-1", resource);
+    snprintf(lock2, sizeof(lock2), "%s-2", resource);
+
+    /* The held lock mode is in the lock id */
+    assert(lockid == LCK_EXCL ||
+	   lockid == LCK_READ ||
+	   lockid == LCK_WRITE);
+
+    switch (lockid)
+    {
+    case LCK_EXCL:
+	status = _unlock_resource(lock1, lockid);
+	if (status)
+	    goto out;
+	status = _unlock_resource(lock2, lockid);
+	break;
+
+    case LCK_READ:
+	status = _unlock_resource(lock1, lockid);
+	break;
+
+    case LCK_WRITE:
+	status = _unlock_resource(lock2, lockid);
+	break;
+    }
+
+ out:
+    return status;
+}
+
+int is_quorate()
+{
+    if (current_corestate == lg_core_Slave ||
+	current_corestate == lg_core_Master ||
+	current_corestate == lg_core_Client)
+	return 1;
+    else
+	return 0;
+}
+
+/* Get all the cluster node names & IPs from CCS and
+   add them to our node list so we know who to talk to.
+   Called when we start up and if we get sent SIGHUP.
+*/
+static int get_all_cluster_nodes()
+{
+    int ctree;
+    char *nodename;
+    int error;
+    int i;
+
+    /* Open the config file */
+    ctree = ccs_force_connect(NULL, 1);
+    if (ctree < 0)
+    {
+	log_error("Error connecting to CCS");
+	return -1;
+    }
+
+    for (i=1;;i++)
+    {
+	char nodekey[256];
+	char nodeip[MAX_CSID_LEN];
+	int  clvmflag = 1;
+	char *clvmflagstr;
+	char key[256];
+
+	sprintf(nodekey, "//cluster/clusternodes/clusternode[%d]/@name", i);
+	error = ccs_get(ctree, nodekey, &nodename);
+	if (error)
+	    break;
+
+	sprintf(key, "//cluster/clusternodes/clusternode[@name=\"%s\"]/clvm", nodename);
+	if (!ccs_get(ctree, key, &clvmflagstr))
+	{
+	    clvmflag = atoi(clvmflagstr);
+	    free(clvmflagstr);
+	}
+
+	DEBUGLOG("Got node %s from ccs(clvmflag = %d)\n", nodename, clvmflag);
+	if ((get_ip_address(nodename, nodeip) == 0) && clvmflag)
+	{
+	    struct node_info *ninfo;
+
+	    /* If it's not in the list, then add it */
+	    ninfo = hash_lookup_binary(node_hash, nodeip, MAX_CSID_LEN);
+	    if (!ninfo)
+	    {
+		ninfo = malloc(sizeof(struct node_info));
+		if (!ninfo)
+		{
+		    syslog(LOG_ERR, "Cannot alloc memory for node info\n");
+		    ccs_disconnect(ctree);
+		    return -1;
+		}
+		strcpy(ninfo->name, nodename);
+
+		ninfo->state = NODE_DOWN;
+		hash_insert_binary(node_hash, nodeip, MAX_CSID_LEN, ninfo);
+	    }
+	}
+	else
+	{
+	    DEBUGLOG("node %s has clvm disabled\n", nodename);
+	}
+	free(nodename);
+    }
+
+    /* Finished with config file */
+    ccs_disconnect(ctree);
+
+    return 0;
+}
+

daemons/clvmd/clvmd-gulm.h

@@ -0,0 +1,9 @@
+
+
+
+extern int get_next_node_csid(void **context, char *csid);
+extern void add_down_node(char *csid);
+extern int gulm_fd(void);
+extern int get_ip_address(char *node, char *addr);
+extern void tcp_remove_client(char *csid);
+extern int alloc_client(int fd, char *csid, struct local_client **new_client);

daemons/clvmd/clvmd.h

@@ -0,0 +1,119 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifndef _CLVMD_H
+#define _CLVMD_H
+
+#define CLVMD_MAJOR_VERSION 0
+#define CLVMD_MINOR_VERSION 2
+#define CLVMD_PATCH_VERSION 1
+
+/* Name of the cluster LVM admin lock */
+#define ADMIN_LOCK_NAME "CLVMD_ADMIN"
+
+/* Default time (in seconds) we will wait for all remote commands to execute
+   before declaring them dead */
+#define DEFAULT_CMD_TIMEOUT 60
+
+/* One of these for each reply we get from command execution on a node */
+struct node_reply {
+	char node[MAX_CLUSTER_MEMBER_NAME_LEN];
+	char *replymsg;
+	int status;
+	struct node_reply *next;
+};
+
+/*
+ * These exist for the use of local sockets only when we are
+ * collecting responses from all cluster nodes
+ */
+struct localsock_bits {
+	struct node_reply *replies;
+	int num_replies;
+	int expected_replies;
+	time_t sent_time;	/* So we can check for timeouts */
+	int in_progress;	/* Only execute one cmd at a time per client */
+	int sent_out;		/* Flag to indicate that a command was sent
+				   to remote nodes */
+	void *private;		/* Private area for command processor use */
+	void *cmd;		/* Whole command as passed down local socket */
+	int cmd_len;		/* Length of above */
+	int pipe;		/* Pipe to send PRE completion status down */
+	int finished;		/* Flag to tell subthread to exit */
+	int all_success;	/* Set to 0 if any node (or the pre_command)
+				   failed */
+	struct local_client *pipe_client;
+	pthread_t threadid;
+	enum { PRE_COMMAND, POST_COMMAND, QUIT } state;
+	pthread_mutex_t mutex;	/* Main thread and worker synchronisation */
+	pthread_cond_t cond;
+
+	pthread_mutex_t reply_mutex;	/* Protect reply structure */
+};
+
+/* Entries for PIPE clients */
+struct pipe_bits {
+	struct local_client *client;	/* Actual (localsock) client */
+	pthread_t threadid;		/* Our own copy of the thread id */
+};
+
+/* Entries for Network socket clients */
+struct netsock_bits {
+	void *private;
+	int flags;
+};
+
+typedef int (*fd_callback_t) (struct local_client * fd, char *buf, int len,
+			      char *csid, struct local_client ** new_client);
+
+/* One of these for each fd we are listening on */
+struct local_client {
+	int fd;
+	enum { CLUSTER_MAIN_SOCK, CLUSTER_DATA_SOCK, LOCAL_RENDEZVOUS,
+		    LOCAL_SOCK, THREAD_PIPE, CLUSTER_INTERNAL } type;
+	struct local_client *next;
+	unsigned short xid;
+	fd_callback_t callback;
+
+	union {
+		struct localsock_bits localsock;
+		struct pipe_bits pipe;
+		struct netsock_bits net;
+	} bits;
+};
+
+#ifdef DEBUG
+#define DEBUGLOG(fmt, args...) fprintf(stderr, "CLVMD[%d]: %ld ", getpid(), time(NULL) ); fprintf(stderr, fmt, ## args)
+#else
+#define DEBUGLOG(fmt, args...)
+#endif
+
+#ifndef max
+#define max(a,b) ((a)>(b)?(a):(b))
+#endif
+
+/* The real command processor is in clvmd-command.c */
+extern int do_command(struct local_client *client, struct clvm_header *msg,
+		      int msglen, char **buf, int buflen, int *retlen);
+
+/* Pre and post command routines are called only on the local node */
+extern int do_pre_command(struct local_client *client);
+extern int do_post_command(struct local_client *client);
+extern void cmd_client_cleanup(struct local_client *client);
+extern int add_client(struct local_client *new_client);
+
+extern void clvmd_cluster_init_completed(void);
+extern void process_message(struct local_client *client, char *buf, int len, char *csid);
+#endif

daemons/clvmd/cnxman-socket.h

@@ -0,0 +1,226 @@
+/******************************************************************************
+*******************************************************************************
+**
+**  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
+**  Copyright (C) 2004 Red Hat, Inc.  All rights reserved.
+**
+**  This copyrighted material is made available to anyone wishing to use,
+**  modify, copy, or redistribute it subject to the terms and conditions
+**  of the GNU General Public License v.2.
+**
+*******************************************************************************
+******************************************************************************/
+
+/* CMAN socket interface header,
+   may be include by user or kernel code */
+
+#ifndef __CNXMAN_SOCKET_H
+#define __CNXMAN_SOCKET_H
+
+/* A currently unused number. TIPC also uses this number and you're unlikely
+   to be using both.
+ */
+#define AF_CLUSTER 30
+#define PF_CLUSTER AF_CLUSTER
+
+/* Protocol(socket) types */
+#define CLPROTO_MASTER 2
+#define CLPROTO_CLIENT 3
+
+/* ioctls -- should register these properly */
+#define SIOCCLUSTER_NOTIFY            _IOW('x', 0x01, int)
+#define SIOCCLUSTER_REMOVENOTIFY      _IO( 'x', 0x02)
+#define SIOCCLUSTER_GETMEMBERS        _IOR('x', 0x03, struct cl_cluster_nodelist)
+#define SIOCCLUSTER_SETEXPECTED_VOTES _IOW('x', 0x04, int)
+#define SIOCCLUSTER_ISQUORATE         _IO( 'x', 0x05)
+#define SIOCCLUSTER_ISLISTENING       _IOW('x', 0x06, struct cl_listen_request)
+#define SIOCCLUSTER_GETALLMEMBERS     _IOR('x', 0x07, struct cl_cluster_nodelist)
+#define SIOCCLUSTER_SET_VOTES         _IOW('x', 0x08, int)
+#define SIOCCLUSTER_GET_VERSION       _IOR('x', 0x09, struct cl_version)
+#define SIOCCLUSTER_SET_VERSION       _IOW('x', 0x0a, struct cl_version)
+#define SIOCCLUSTER_ISACTIVE          _IO( 'x', 0x0b)
+#define SIOCCLUSTER_KILLNODE          _IOW('x', 0x0c, int)
+#define SIOCCLUSTER_GET_JOINCOUNT     _IO( 'x', 0x0d)
+#define SIOCCLUSTER_SERVICE_REGISTER  _IOW('x', 0x0e, char)
+#define SIOCCLUSTER_SERVICE_UNREGISTER _IO('x', 0x0f)
+#define SIOCCLUSTER_SERVICE_JOIN      _IO( 'x', 0x10)
+#define SIOCCLUSTER_SERVICE_LEAVE     _IO( 'x', 0x20)
+#define SIOCCLUSTER_SERVICE_SETSIGNAL _IOW('x', 0x30, int)
+#define SIOCCLUSTER_SERVICE_STARTDONE _IOW('x', 0x40, unsigned int)
+#define SIOCCLUSTER_SERVICE_GETEVENT  _IOR('x', 0x50, struct cl_service_event)
+#define SIOCCLUSTER_SERVICE_GETMEMBERS _IOR('x', 0x60, struct cl_cluster_nodelist)
+#define SIOCCLUSTER_SERVICE_GLOBALID  _IOR('x', 0x70, uint32_t)
+#define SIOCCLUSTER_SERVICE_SETLEVEL  _IOR('x', 0x80, int)
+#define SIOCCLUSTER_GETNODE	      _IOWR('x', 0x90, struct cl_cluster_node)
+#define SIOCCLUSTER_BARRIER           _IOW('x', 0x0a0, struct cl_barrier_info)
+
+/* These were setsockopts */
+#define SIOCCLUSTER_PASS_SOCKET       _IOW('x', 0x0b0, struct cl_passed_sock)
+#define SIOCCLUSTER_SET_NODENAME      _IOW('x', 0x0b1, char *)
+#define SIOCCLUSTER_SET_NODEID        _IOW('x', 0x0b2, int)
+#define SIOCCLUSTER_JOIN_CLUSTER      _IOW('x', 0x0b3, struct cl_join_cluster_info)
+#define SIOCCLUSTER_LEAVE_CLUSTER     _IOW('x', 0x0b4, int)
+
+
+/* Maximum size of a cluster message */
+#define MAX_CLUSTER_MESSAGE          1500
+#define MAX_CLUSTER_MEMBER_NAME_LEN   255
+#define MAX_BARRIER_NAME_LEN           33
+#define MAX_SA_ADDR_LEN                12
+#define MAX_CLUSTER_NAME_LEN           16
+
+/* Well-known cluster port numbers */
+#define CLUSTER_PORT_MEMBERSHIP  1	/* Mustn't block during cluster
+					 * transitions! */
+#define CLUSTER_PORT_SERVICES    2
+#define CLUSTER_PORT_SYSMAN      10	/* Remote execution daemon */
+#define CLUSTER_PORT_CLVMD       11	/* Cluster LVM daemon */
+#define CLUSTER_PORT_SLM         12	/* LVM SLM (simple lock manager) */
+
+/* Port numbers above this will be blocked when the cluster is inquorate or in
+ * transition */
+#define HIGH_PROTECTED_PORT      9
+
+/* Reasons for leaving the cluster */
+#define CLUSTER_LEAVEFLAG_DOWN     0	/* Normal shutdown */
+#define CLUSTER_LEAVEFLAG_KILLED   1
+#define CLUSTER_LEAVEFLAG_PANIC    2
+#define CLUSTER_LEAVEFLAG_REMOVED  3	/* This one can reduce quorum */
+#define CLUSTER_LEAVEFLAG_REJECTED 4	/* Not allowed into the cluster in the
+					 * first place */
+#define CLUSTER_LEAVEFLAG_INCONSISTENT 5	/* Our view of the cluster is
+						 * in a minority */
+#define CLUSTER_LEAVEFLAG_DEAD         6	/* Discovered to be dead */
+#define CLUSTER_LEAVEFLAG_FORCE     0x10	/* Forced by command-line */
+
+/* OOB messages sent to a local socket */
+#define CLUSTER_OOB_MSG_PORTCLOSED  1
+#define CLUSTER_OOB_MSG_STATECHANGE 2
+#define CLUSTER_OOB_MSG_SERVICEEVENT 3
+
+/* Sendmsg flags, these are above the normal sendmsg flags so they don't
+ * interfere */
+#define MSG_NOACK     0x010000	/* Don't need an ACK for this message */
+#define MSG_QUEUE     0x020000	/* Queue the message for sending later */
+#define MSG_MULTICAST 0x080000	/* Message was sent to all nodes in the cluster
+				 */
+#define MSG_ALLINT    0x100000	/* Send out of all interfaces */
+#define MSG_REPLYEXP  0x200000	/* Reply is expected */
+
+typedef enum { NODESTATE_REMOTEMEMBER, NODESTATE_JOINING, NODESTATE_MEMBER,
+	       NODESTATE_DEAD } nodestate_t;
+
+
+struct sockaddr_cl {
+	unsigned short scl_family;
+	unsigned char scl_flags;
+	unsigned char scl_port;
+	int           scl_nodeid;
+};
+
+/*
+ * This is how we pass the multicast & receive sockets into kernel space.
+ */
+struct cl_passed_sock {
+	int fd;			/* FD of master socket to do multicast on */
+	int number;		/* Socket number, to match up recvonly & bcast
+				 * sockets */
+        int multicast;          /* Is it multicast or receive ? */
+};
+
+/* Cluster configuration info passed when we join the cluster */
+struct cl_join_cluster_info {
+	unsigned char votes;
+	unsigned int expected_votes;
+	unsigned int two_node;
+	unsigned int config_version;
+
+        char cluster_name[17];
+};
+
+
+/* This is the structure, per node, returned from the membership ioctl */
+struct cl_cluster_node {
+	unsigned int size;
+	unsigned int node_id;
+	unsigned int us;
+	unsigned int leave_reason;
+	unsigned int incarnation;
+	nodestate_t state;
+	char name[MAX_CLUSTER_MEMBER_NAME_LEN];
+	unsigned char votes;
+};
+
+/* The struct passed to the membership ioctls */
+struct cl_cluster_nodelist {
+        uint32_t max_members;
+        struct cl_cluster_node *nodes;
+};
+
+/* Structure passed to SIOCCLUSTER_ISLISTENING */
+struct cl_listen_request {
+	unsigned char port;
+        int           nodeid;
+};
+
+/* A Cluster PORTCLOSED message - received by a local user as an OOB message */
+struct cl_portclosed_oob {
+	unsigned char cmd;	/* CLUSTER_OOB_MSG_PORTCLOSED */
+	unsigned char port;
+};
+
+/* Get all version numbers or set the config version */
+struct cl_version {
+	unsigned int major;
+	unsigned int minor;
+	unsigned int patch;
+	unsigned int config;
+};
+
+/* structure passed to barrier ioctls */
+struct cl_barrier_info {
+	char cmd;
+	char name[MAX_BARRIER_NAME_LEN];
+	unsigned int flags;
+	unsigned long arg;
+};
+
+typedef enum { SERVICE_EVENT_STOP, SERVICE_EVENT_START, SERVICE_EVENT_FINISH,
+		SERVICE_EVENT_LEAVEDONE } service_event_t;
+
+typedef enum { SERVICE_START_FAILED, SERVICE_START_JOIN, SERVICE_START_LEAVE }
+		service_start_t;
+
+struct cl_service_event {
+	service_event_t type;
+	service_start_t start_type;
+	unsigned int event_id;
+	unsigned int last_stop;
+	unsigned int last_start;
+	unsigned int last_finish;
+	unsigned int node_count;
+};
+
+
+/* Commands to the barrier ioctl */
+#define BARRIER_IOCTL_REGISTER 1
+#define BARRIER_IOCTL_CHANGE   2
+#define BARRIER_IOCTL_DELETE   3
+#define BARRIER_IOCTL_WAIT     4
+
+/* Attributes of a barrier - bitmask */
+#define BARRIER_ATTR_AUTODELETE 1
+#define BARRIER_ATTR_MULTISTEP  2
+#define BARRIER_ATTR_MANUAL     4
+#define BARRIER_ATTR_ENABLED    8
+#define BARRIER_ATTR_CALLBACK  16
+
+/* Attribute setting commands */
+#define BARRIER_SETATTR_AUTODELETE 1
+#define BARRIER_SETATTR_MULTISTEP  2
+#define BARRIER_SETATTR_ENABLED    3
+#define BARRIER_SETATTR_NODES      4
+#define BARRIER_SETATTR_CALLBACK   5
+#define BARRIER_SETATTR_TIMEOUT    6
+
+#endif

daemons/clvmd/libclvm.c

@@ -0,0 +1,446 @@
+/*
+ * Copyright (C) 1997-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/* library functions for Cluster LVM Daemon */
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/socket.h>
+#include <sys/uio.h>
+#include <sys/un.h>
+#include <sys/time.h>
+#include <sys/ioctl.h>
+#include <sys/utsname.h>
+#include <syslog.h>
+#include <netinet/in.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <signal.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <search.h>
+#include <errno.h>
+
+#include "clvm.h"
+#include "libclvm.h"
+
+/* CLVM in hex! */
+#define LVM_SIGNATURE 0x434C564D
+
+#define MAX_CLUSTER_MEMBER_NAME_LEN 255
+
+/* NOTE: the LVMD uses the socket FD as the client ID, this means
+   that any client that calls fork() will inherit the context of
+   it's parent. */
+static int clvmd_sock = -1;
+
+static int open_local_sock(void)
+{
+	int local_socket;
+	struct sockaddr_un sockaddr;
+
+	/* Open local socket */
+	local_socket = socket(PF_UNIX, SOCK_STREAM, 0);
+	if (local_socket < 0) {
+		perror("Can't create local socket");
+		return -1;
+	}
+
+	fcntl(local_socket, F_SETFD, !FD_CLOEXEC);
+
+	strcpy(sockaddr.sun_path, CLVMD_SOCKNAME);
+	sockaddr.sun_family = AF_UNIX;
+	if (connect
+	    (local_socket, (struct sockaddr *) &sockaddr, sizeof(sockaddr))) {
+		int saved_errno = errno;
+
+		close(local_socket);
+
+		errno = saved_errno;
+		return -1;
+	}
+	return local_socket;
+}
+
+/* Send a request and return the status */
+static int send_request(char *inbuf, int inlen, char **retbuf)
+{
+	char outbuf[PIPE_BUF];
+	struct clvm_header *outheader = (struct clvm_header *) outbuf;
+	int len;
+	int off;
+	fd_set fds;
+
+	FD_ZERO(&fds);
+	FD_SET(clvmd_sock, &fds);
+
+	/* Send it to CLVMD */
+	if (write(clvmd_sock, inbuf, inlen) != inlen) {
+		perror("Error writing to CLVMD");
+		return -1;
+	}
+
+	/* Get the response */
+	if ((len = read(clvmd_sock, outbuf, sizeof(struct clvm_header))) < 0) {
+		perror("Error reading CLVMD");
+		return -1;
+	}
+	if (len == 0) {
+		fprintf(stderr, "EOF reading CLVMD");
+		errno = ENOTCONN;
+		return -1;
+	}
+
+	/* Allocate buffer */
+	*retbuf = malloc(len + outheader->arglen);
+	if (!*retbuf) {
+		errno = ENOMEM;
+		return -1;
+	}
+
+	/* Copy the header */
+	memcpy(*retbuf, outbuf, len);
+	outheader = (struct clvm_header *) *retbuf;
+
+	/* Read the returned values */
+	off = 1;		/* we've already read the first byte */
+
+	while (off < outheader->arglen && len > 0) {
+		len = read(clvmd_sock, outheader->args + off, PIPE_BUF);
+		if (len > 0)
+			off += len;
+	}
+
+	/* Was it an error ? */
+	if (outheader->status < 0) {
+		errno = -outheader->status;
+		return -2;
+	}
+	return 0;
+}
+
+/* Build the structure header and parse-out wildcard node names */
+static void build_header(struct clvm_header *head, int cmd, const char *node,
+			 void *data, int len)
+{
+	head->cmd = cmd;
+	head->status = 0;
+	head->flags = 0;
+	head->clientid = 0;
+	head->arglen = len;
+	if (node) {
+		/* Allow a couple of special node names:
+		   "*" for all nodes,
+		   "." for the local node only
+		 */
+		if (strcmp(node, "*") == 0) {
+			head->node[0] = '\0';
+		} else if (strcmp(node, ".") == 0) {
+			head->node[0] = '\0';
+			head->flags = CLVMD_FLAG_LOCAL;
+		} else {
+			strcpy(head->node, node);
+		}
+	} else {
+		head->node[0] = '\0';
+	}
+}
+
+/* Send a message to a(or all) node(s) in the cluster */
+int lvm_cluster_write(char cmd, char *node, void *data, int len)
+{
+	char outbuf[sizeof(struct clvm_header) + len + strlen(node) + 1];
+	char *retbuf = NULL;
+	int status;
+	struct clvm_header *head = (struct clvm_header *) outbuf;
+
+	if (clvmd_sock == -1)
+		clvmd_sock = open_local_sock();
+	if (clvmd_sock == -1)
+		return -1;
+
+	build_header(head, cmd, node, data, len);
+	memcpy(head->node + strlen(head->node) + 1, data, len);
+
+	status =
+	    send_request(outbuf,
+			 sizeof(struct clvm_header) + strlen(head->node) + len,
+			 &retbuf);
+	if (retbuf)
+		free(retbuf);
+
+	return status;
+}
+
+/* API: Send a message to a(or all) node(s) in the cluster
+   and wait for replies */
+int lvm_cluster_request(char cmd, const char *node, void *data, int len,
+			lvm_response_t ** response, int *num)
+{
+	char outbuf[sizeof(struct clvm_header) + len + strlen(node) + 1];
+	int *outptr;
+	char *inptr;
+	char *retbuf = NULL;
+	int status;
+	int i;
+	int num_responses = 0;
+	struct clvm_header *head = (struct clvm_header *) outbuf;
+	lvm_response_t *rarray;
+
+	*num = 0;
+
+	if (clvmd_sock == -1)
+		clvmd_sock = open_local_sock();
+	if (clvmd_sock == -1)
+		return -1;
+
+	build_header(head, cmd, node, data, len);
+	memcpy(head->node + strlen(head->node) + 1, data, len);
+
+	status =
+	    send_request(outbuf,
+			 sizeof(struct clvm_header) + strlen(head->node) + len,
+			 &retbuf);
+	if (status == 0 || status == -2) {
+		/* Count the number of responses we got */
+		head = (struct clvm_header *) retbuf;
+		inptr = head->args;
+		while (inptr[0]) {
+			num_responses++;
+			inptr += strlen(inptr) + 1;
+			inptr += sizeof(int);
+			inptr += strlen(inptr) + 1;
+		}
+
+		/* Allocate response array. With an extra pair of INTs on the front to sanity
+		   check the pointer when we are given it back to free */
+		outptr =
+		    malloc(sizeof(lvm_response_t) * num_responses +
+			   sizeof(int) * 2);
+		if (!outptr) {
+			if (retbuf)
+				free(retbuf);
+			errno = ENOMEM;
+			return -1;
+		}
+
+		*response = (lvm_response_t *) (outptr + 2);
+		outptr[0] = LVM_SIGNATURE;
+		outptr[1] = num_responses;
+		rarray = *response;
+
+		/* Unpack the response into an lvm_response_t array */
+		inptr = head->args;
+		i = 0;
+		while (inptr[0]) {
+			strcpy(rarray[i].node, inptr);
+			inptr += strlen(inptr) + 1;
+
+			rarray[i].status = *(int *) inptr;
+			inptr += sizeof(int);
+
+			rarray[i].response = malloc(strlen(inptr) + 1);
+			if (rarray[i].response == NULL) {
+				/* Free up everything else and return error */
+				int j;
+				for (j = 0; j < i; j++)
+					free(rarray[i].response);
+				free(outptr);
+				errno = ENOMEM;
+				return -1;
+			}
+
+			strcpy(rarray[i].response, inptr);
+			rarray[i].len = strlen(inptr);
+			inptr += strlen(inptr) + 1;
+			i++;
+		}
+		*num = num_responses;
+		*response = rarray;
+	}
+
+	if (retbuf)
+		free(retbuf);
+	return status;
+}
+
+/* API: Free reply array */
+int lvm_cluster_free_request(lvm_response_t * response)
+{
+	int *ptr = (int *) response - 2;
+	int i;
+	int num;
+
+	/* Check it's ours to free */
+	if (response == NULL || *ptr != LVM_SIGNATURE) {
+		errno = EINVAL;
+		return -1;
+	}
+
+	num = ptr[1];
+	for (i = 0; i < num; i++) {
+		free(response[i].response);
+	}
+	free(ptr);
+
+	return 0;
+}
+
+/* These are a "higher-level" API providing black-box lock/unlock
+   functions for cluster LVM...maybe */
+
+/* Set by lock(), used by unlock() */
+static int num_responses;
+static lvm_response_t *response;
+
+int lvm_lock_for_cluster(char scope, char *name, int verbosity)
+{
+	int status;
+	int i;
+	char *args;
+	int len;
+
+	if (name) {
+		len = strlen(name) + 2;
+		args = alloca(len);
+		strcpy(args + 1, name);
+	} else {
+		len = 2;
+		args = alloca(len);
+		args[1] = '\0';
+	}
+	args[0] = scope;
+
+	status = lvm_cluster_request(CLVMD_CMD_LOCK,
+				     "", args, len, &response, &num_responses);
+
+	/* If any nodes were down then display them and return an error */
+	for (i = 0; i < num_responses; i++) {
+		if (response[i].status == -EHOSTDOWN) {
+			if (verbosity)
+				fprintf(stderr,
+					"clvmd not running on node %s\n",
+					response[i].node);
+			status = -1;
+		}
+	}
+
+	/* If there was an error then free the memory now as the caller won't
+	   want to do the unlock */
+	if (status) {
+		int saved_errno = errno;
+		lvm_cluster_free_request(response);
+		num_responses = 0;
+		errno = saved_errno;
+	}
+	return status;
+}
+
+int lvm_unlock_for_cluster(char scope, char *name, int verbosity)
+{
+	int status;
+	int i;
+	int len;
+	int failed;
+	int num_unlock_responses;
+	char *args;
+	lvm_response_t *unlock_response;
+
+	/* We failed - this should not have been called */
+	if (num_responses == 0)
+		return 0;
+
+	if (name) {
+		len = strlen(name) + 2;
+		args = alloca(len);
+		strcpy(args + 1, name);
+	} else {
+		len = 2;
+		args = alloca(len);
+		args[1] = '\0';
+	}
+	args[0] = scope;
+
+	/* See if it failed anywhere */
+	failed = 0;
+	for (i = 0; i < num_responses; i++) {
+		if (response[i].status != 0)
+			failed++;
+	}
+
+	/* If it failed on any nodes then we only unlock on
+	   the nodes that succeeded */
+	if (failed) {
+		for (i = 0; i < num_responses; i++) {
+			/* Unlock the ones that succeeded */
+			if (response[i].status == 0) {
+				status = lvm_cluster_request(CLVMD_CMD_UNLOCK,
+							     response[i].node,
+							     args, len,
+							     &unlock_response,
+							     &num_unlock_responses);
+				if (status) {
+					if (verbosity)
+						fprintf(stderr,
+							"cluster command to node %s failed: %s\n",
+							response[i].node,
+							strerror(errno));
+				} else if (unlock_response[0].status != 0) {
+					if (verbosity > 1)
+						fprintf(stderr,
+							"unlock on node %s failed: %s\n",
+							response[i].node,
+							strerror(unlock_response
+								 [0].status));
+				}
+				lvm_cluster_free_request(unlock_response);
+			} else {
+				if (verbosity)
+					fprintf(stderr,
+						"command on node %s failed: '%s' - will be left locked\n",
+						response[i].node,
+						strerror(response[i].status));
+			}
+		}
+	} else {
+		/* All OK, we can do a full cluster unlock */
+		status = lvm_cluster_request(CLVMD_CMD_UNLOCK,
+					     "",
+					     args, len,
+					     &unlock_response,
+					     &num_unlock_responses);
+		if (status) {
+			if (verbosity > 1)
+				fprintf(stderr, "cluster command failed: %s\n",
+					strerror(errno));
+		} else {
+			for (i = 0; i < num_unlock_responses; i++) {
+				if (unlock_response[i].status != 0) {
+					if (verbosity > 1)
+						fprintf(stderr,
+							"unlock on node %s failed: %s\n",
+							response[i].node,
+							strerror(unlock_response
+								 [0].status));
+				}
+			}
+		}
+		lvm_cluster_free_request(unlock_response);
+	}
+	lvm_cluster_free_request(response);
+
+	return 0;
+}

daemons/clvmd/libclvm.h

@@ -0,0 +1,36 @@
+/*
+ * Copyright (C) 1997-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifndef _LIBCLVM_H
+#define _LIBCLVM_H
+
+typedef struct lvm_response {
+	char node[255];
+	char *response;
+	int status;
+	int len;
+
+} lvm_response_t;
+
+extern int lvm_cluster_request(char cmd, const char *node, void *data, int len,
+			       lvm_response_t ** response, int *num);
+extern int lvm_cluster_write(char cmd, char *node, void *data, int len);
+extern int lvm_cluster_free_request(lvm_response_t * response);
+
+/* The "high-level" API */
+extern int lvm_lock_for_cluster(char scope, char *name, int verbosity);
+extern int lvm_unlock_for_cluster(char scope, char *name, int verbosity);
+
+#endif

daemons/clvmd/lvm-functions.c

@@ -0,0 +1,463 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include <pthread.h>
+#include <sys/types.h>
+#include <sys/utsname.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <fcntl.h>
+#include <string.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <errno.h>
+#include <syslog.h>
+#include <assert.h>
+
+#include "libdlm.h"
+#include "clvm.h"
+#include "clvmd-comms.h"
+#include "clvmd.h"
+#include "lvm-functions.h"
+
+/* LVM2 headers */
+#include "toolcontext.h"
+#include "log.h"
+#include "activate.h"
+#include "hash.h"
+#include "locking.h"
+
+static struct cmd_context *cmd = NULL;
+static struct hash_table *lv_hash = NULL;
+static pthread_mutex_t lv_hash_lock;
+
+struct lv_info {
+	int lock_id;
+	int lock_mode;
+};
+
+/* Return the mode a lock is currently held at (or -1 if not held) */
+static int get_current_lock(char *resource)
+{
+	struct lv_info *lvi;
+
+	pthread_mutex_lock(&lv_hash_lock);
+	lvi = hash_lookup(lv_hash, resource);
+	pthread_mutex_unlock(&lv_hash_lock);
+	if (lvi) {
+		return lvi->lock_mode;
+	} else {
+		return -1;
+	}
+}
+
+/* Called at shutdown to tidy the lockspace */
+void unlock_all()
+{
+	struct hash_node *v;
+
+	pthread_mutex_lock(&lv_hash_lock);
+	hash_iterate(v, lv_hash) {
+		struct lv_info *lvi = hash_get_data(lv_hash, v);
+
+		sync_unlock(hash_get_key(lv_hash, v), lvi->lock_id);
+	}
+	pthread_mutex_unlock(&lv_hash_lock);
+}
+
+/* Gets a real lock and keeps the info in the hash table */
+int hold_lock(char *resource, int mode, int flags)
+{
+	int status;
+	int saved_errno;
+	struct lv_info *lvi;
+
+	flags &= LKF_NOQUEUE;	/* Only LKF_NOQUEUE is valid here */
+
+	pthread_mutex_lock(&lv_hash_lock);
+	lvi = hash_lookup(lv_hash, resource);
+	pthread_mutex_unlock(&lv_hash_lock);
+	if (lvi) {
+		/* Already exists - convert it */
+		status =
+		    sync_lock(resource, mode, LKF_CONVERT | flags,
+			      &lvi->lock_id);
+		saved_errno = errno;
+		if (!status)
+			lvi->lock_mode = mode;
+
+		if (status) {
+			DEBUGLOG("hold_lock. convert to %d failed: %s\n", mode,
+				 strerror(errno));
+		}
+		errno = saved_errno;
+	} else {
+		lvi = malloc(sizeof(struct lv_info));
+		if (!lvi)
+			return -1;
+
+		lvi->lock_mode = mode;
+		status = sync_lock(resource, mode, flags, &lvi->lock_id);
+		saved_errno = errno;
+		if (status) {
+			free(lvi);
+			DEBUGLOG("hold_lock. lock at %d failed: %s\n", mode,
+				 strerror(errno));
+		} else {
+		        pthread_mutex_lock(&lv_hash_lock);
+			hash_insert(lv_hash, resource, lvi);
+			pthread_mutex_unlock(&lv_hash_lock);
+		}
+		errno = saved_errno;
+	}
+	return status;
+}
+
+/* Unlock and remove it from the hash table */
+int hold_unlock(char *resource)
+{
+	struct lv_info *lvi;
+	int status;
+	int saved_errno;
+
+	pthread_mutex_lock(&lv_hash_lock);
+	lvi = hash_lookup(lv_hash, resource);
+	pthread_mutex_unlock(&lv_hash_lock);
+	if (!lvi) {
+		DEBUGLOG("hold_unlock, lock not already held\n");
+		return 0;
+	}
+
+	status = sync_unlock(resource, lvi->lock_id);
+	saved_errno = errno;
+	if (!status) {
+	    	pthread_mutex_lock(&lv_hash_lock);
+		hash_remove(lv_hash, resource);
+		pthread_mutex_unlock(&lv_hash_lock);
+		free(lvi);
+	} else {
+		DEBUGLOG("hold_unlock. unlock failed(%d): %s\n", status,
+			 strerror(errno));
+	}
+
+	errno = saved_errno;
+	return status;
+}
+
+/* Watch the return codes here.
+   liblvm API functions return 1(true) for success, 0(false) for failure and don't set errno.
+   libdlm API functions return 0 for success, -1 for failure and do set errno.
+   These functions here return 0 for success or >0 for failure (where the retcode is errno)
+*/
+
+/* Activate LV exclusive or non-exclusive */
+static int do_activate_lv(char *resource, int mode)
+{
+	int oldmode;
+	int status;
+	int activate_lv;
+	struct lvinfo lvi;
+
+	/* Is it already open ? */
+	oldmode = get_current_lock(resource);
+	if (oldmode == mode) {
+		return 0;	/* Nothing to do */
+	}
+
+	/* Does the config file want us to activate this LV ? */
+	if (!lv_activation_filter(cmd, resource, &activate_lv))
+		return EIO;
+
+	if (!activate_lv)
+		return 0;	/* Success, we did nothing! */
+
+	/* Do we need to activate exclusively? */
+	if (activate_lv == 2)
+		mode = LKM_EXMODE;
+
+	/* OK, try to get the lock */
+	status = hold_lock(resource, mode, LKF_NOQUEUE);
+	if (status)
+		return errno;
+
+	/* If it's suspended then resume it */
+	if (!lv_info_by_lvid(cmd, resource, &lvi))
+		return EIO;
+
+	if (lvi.suspended)
+		if (!lv_resume(cmd, resource))
+			return EIO;
+
+	/* Now activate it */
+	if (!lv_activate(cmd, resource))
+		return EIO;
+
+	return 0;
+}
+
+/* Resume the LV if it was active */
+static int do_resume_lv(char *resource)
+{
+	int oldmode;
+
+	/* Is it open ? */
+	oldmode = get_current_lock(resource);
+	if (oldmode == -1) {
+		DEBUGLOG("do_deactivate_lock, lock not already held\n");
+		return 0;	/* We don't need to do anything */
+	}
+
+	if (!lv_resume_if_active(cmd, resource))
+		return EIO;
+
+	return 0;
+}
+
+/* Suspend the device if active */
+static int do_suspend_lv(char *resource)
+{
+	int oldmode;
+	struct lvinfo lvi;
+
+	/* Is it open ? */
+	oldmode = get_current_lock(resource);
+	if (oldmode == -1) {
+		DEBUGLOG("do_suspend_lv, lock held at %d\n", oldmode);
+		return 0; /* Not active, so it's OK */
+	}
+
+	/* Only suspend it if it exists */
+	if (!lv_info_by_lvid(cmd, resource, &lvi))
+		return EIO;
+
+	if (lvi.exists) {
+		if (!lv_suspend_if_active(cmd, resource)) {
+			return EIO;
+		}
+	}
+	return 0;
+}
+
+static int do_deactivate_lv(char *resource)
+{
+	int oldmode;
+	int status;
+
+	/* Is it open ? */
+	oldmode = get_current_lock(resource);
+	if (oldmode == -1) {
+		DEBUGLOG("do_deactivate_lock, lock not already held\n");
+		return 0;	/* We don't need to do anything */
+	}
+
+	if (!lv_deactivate(cmd, resource))
+		return EIO;
+
+	status = hold_unlock(resource);
+	if (status)
+		return errno;
+
+	return 0;
+}
+
+/* This is the LOCK_LV part that happens on all nodes in the cluster -
+   it is responsible for the interaction with device-mapper and LVM */
+int do_lock_lv(unsigned char command, unsigned char lock_flags, char *resource)
+{
+	int status = 0;
+
+	DEBUGLOG("do_lock_lv: resource '%s', cmd = 0x%x, flags = %d\n",
+		 resource, command, lock_flags);
+
+	if (!cmd->config_valid || config_files_changed(cmd)) {
+		/* Reinitialise various settings inc. logging, filters */
+		if (!refresh_toolcontext(cmd)) {
+			log_error("Updated config file invalid. Aborting.");
+			return EINVAL;
+		}
+	}
+
+	switch (command) {
+	case LCK_LV_EXCLUSIVE:
+		status = do_activate_lv(resource, LKM_EXMODE);
+		break;
+
+	case LCK_LV_SUSPEND:
+		status = do_suspend_lv(resource);
+		break;
+
+	case LCK_UNLOCK:
+	case LCK_LV_RESUME:	/* if active */
+		status = do_resume_lv(resource);
+		break;
+
+	case LCK_LV_ACTIVATE:
+		status = do_activate_lv(resource, LKM_CRMODE);
+		break;
+
+	case LCK_LV_DEACTIVATE:
+		status = do_deactivate_lv(resource);
+		break;
+
+	default:
+		DEBUGLOG("Invalid LV command 0x%x\n", command);
+		status = EINVAL;
+		break;
+	}
+
+	/* clean the pool for another command */
+	pool_empty(cmd->mem);
+
+	DEBUGLOG("Command return is %d\n", status);
+	return status;
+}
+
+/* Functions to do on the local node only BEFORE the cluster-wide stuff above happens */
+int pre_lock_lv(unsigned char command, unsigned char lock_flags, char *resource)
+{
+	/* Nearly all the stuff happens cluster-wide. Apart from SUSPEND. Here we get the
+	   lock out on this node (because we are the node modifying the metadata)
+	   before suspending cluster-wide.
+	 */
+	if (command == LCK_LV_SUSPEND) {
+		DEBUGLOG("pre_lock_lv: resource '%s', cmd = 0x%x, flags = %d\n",
+			 resource, command, lock_flags);
+
+		if (hold_lock(resource, LKM_PWMODE, LKF_NOQUEUE))
+			return errno;
+	}
+	return 0;
+}
+
+/* Functions to do on the local node only AFTER the cluster-wide stuff above happens */
+int post_lock_lv(unsigned char command, unsigned char lock_flags,
+		 char *resource)
+{
+	/* Opposite of above, done on resume after a metadata update */
+	if (command == LCK_LV_RESUME) {
+		int oldmode;
+
+		DEBUGLOG
+		    ("post_lock_lv: resource '%s', cmd = 0x%x, flags = %d\n",
+		     resource, command, lock_flags);
+
+		/* If the lock state is PW then restore it to what it was */
+		oldmode = get_current_lock(resource);
+		if (oldmode == LKM_PWMODE) {
+			struct lvinfo lvi;
+
+			if (!lv_info_by_lvid(cmd, resource, &lvi))
+				return EIO;
+
+			if (lvi.exists) {
+				if (hold_lock(resource, LKM_CRMODE, 0))
+					return errno;
+			} else {
+				if (hold_unlock(resource))
+					return errno;
+			}
+		}
+	}
+	return 0;
+}
+
+/* Check if a VG is un use by LVM1 so we don't stomp on it */
+int do_check_lvm1(char *vgname)
+{
+	int status;
+
+	status = check_lvm1_vg_inactive(cmd, vgname);
+
+	return status == 1 ? 0 : EBUSY;
+}
+
+/*
+ * Ideally, clvmd should be started before any LVs are active
+ * but this may not be the case...
+ * I suppose this also comes in handy if clvmd crashes, not that it would!
+ */
+static void *get_initial_state()
+{
+	char lv[64], vg[64], flags[25];
+	char uuid[65];
+	char line[255];
+	FILE *lvs =
+	    popen
+	    ("/sbin/lvm lvs --nolocking --noheadings -o vg_uuid,lv_uuid,lv_attr",
+	     "r");
+
+	if (!lvs)
+		return NULL;
+
+	while (fgets(line, sizeof(line), lvs)) {
+	        if (sscanf(line, "%s %s %s\n", vg, lv, flags) == 3) {
+
+			/* States: s:suspended a:active S:dropped snapshot I:invalid snapshot */
+		        if (strlen(vg) == 38 &&                         /* is is a valid UUID ? */
+			    (flags[4] == 'a' || flags[4] == 's')) {	/* is it active or suspended? */
+				/* Convert hyphen-separated UUIDs into one */
+				memcpy(&uuid[0], &vg[0], 6);
+				memcpy(&uuid[6], &vg[7], 4);
+				memcpy(&uuid[10], &vg[12], 4);
+				memcpy(&uuid[14], &vg[17], 4);
+				memcpy(&uuid[18], &vg[22], 4);
+				memcpy(&uuid[22], &vg[27], 4);
+				memcpy(&uuid[26], &vg[32], 6);
+				memcpy(&uuid[32], &lv[0], 6);
+				memcpy(&uuid[38], &lv[7], 4);
+				memcpy(&uuid[42], &lv[12], 4);
+				memcpy(&uuid[46], &lv[17], 4);
+				memcpy(&uuid[50], &lv[22], 4);
+				memcpy(&uuid[54], &lv[27], 4);
+				memcpy(&uuid[58], &lv[32], 6);
+				uuid[64] = '\0';
+
+				DEBUGLOG("getting initial lock for %s\n", uuid);
+				hold_lock(uuid, LKM_CRMODE, LKF_NOQUEUE);
+			}
+		}
+	}
+	fclose(lvs);
+	return NULL;
+}
+
+void init_lvhash()
+{
+	/* Create hash table for keeping LV locks & status */
+	lv_hash = hash_create(100);
+	pthread_mutex_init(&lv_hash_lock, NULL);
+}
+
+/* Called to initialise the LVM context of the daemon */
+int init_lvm(void)
+{
+	if (!(cmd = create_toolcontext(NULL))) {
+		log_error("Failed to allocate command context");
+		return 0;
+	}
+
+	/* Use LOG_DAEMON for syslog messages instead of LOG_USER */
+	init_syslog(LOG_DAEMON);
+	init_debug(_LOG_ERR);
+
+	get_initial_state();
+
+	return 1;
+}

daemons/clvmd/lvm-functions.h

@@ -0,0 +1,35 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/* Functions in lvm-functions.c */
+
+#ifndef _LVM_FUNCTIONS_H
+#define _LVM_FUNCTIONS_H
+
+extern int pre_lock_lv(unsigned char lock_cmd, unsigned char lock_flags,
+		       char *resource);
+extern int do_lock_lv(unsigned char lock_cmd, unsigned char lock_flags,
+		      char *resource);
+extern int post_lock_lv(unsigned char lock_cmd, unsigned char lock_flags,
+			char *resource);
+extern int do_check_lvm1(char *vgname);
+extern int init_lvm(void);
+extern void init_lvhash(void);
+
+extern int hold_unlock(char *resource);
+extern int hold_lock(char *resource, int mode, int flags);
+extern void unlock_all(void);
+
+#endif

daemons/clvmd/system-lv.c

@@ -0,0 +1,369 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/* Routines dealing with the System LV */
+
+#include <pthread.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/socket.h>
+#include <sys/uio.h>
+#include <sys/un.h>
+#include <sys/time.h>
+#include <sys/ioctl.h>
+#include <sys/mount.h>
+#include <sys/utsname.h>
+#include <syslog.h>
+#include <netinet/in.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <signal.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <dirent.h>
+#include <errno.h>
+#include <mntent.h>
+
+#include "libdlm.h"
+#include "log.h"
+#include "list.h"
+#include "locking.h"
+#include "system-lv.h"
+#include "clvmd-comms.h"
+#ifdef HAVE_CCS
+#include "ccs.h"
+#endif
+
+#define SYSTEM_LV_FILESYSTEM "ext2"
+#define SYSTEM_LV_MOUNTPOINT "/tmp/.clvmd-XXXXXX"
+
+extern char *config_filename(void);
+
+static char system_lv_name[PATH_MAX] = { '\0' };
+static char mount_point[PATH_MAX] = { '\0' };
+static int mounted = 0;
+static int mounted_rw = 0;
+static int lockid;
+static const char *lock_name = "CLVM_SYSTEM_LV";
+
+/* Look in /proc/mounts or (as a last resort) /etc/mtab to
+   see if the system-lv is mounted. If it is mounted and we
+   think it's not then abort because we don't have the right
+   lock status and we don't know what other processes are doing with it.
+
+   Returns 1 for mounted, 0 for not mounted so it matches the condition
+   of the "mounted" static variable above.
+*/
+static int is_really_mounted(void)
+{
+	FILE *mountfile;
+	struct mntent *ment;
+
+	mountfile = setmntent("/proc/mounts", "r");
+	if (!mountfile) {
+		mountfile = setmntent("/etc/mtab", "r");
+		if (!mountfile) {
+			log_error("Unable to open /proc/mounts or /etc/mtab");
+			return -1;
+		}
+	}
+
+	/* Look for system LV name in the file */
+	do {
+		ment = getmntent(mountfile);
+		if (ment) {
+			if (strcmp(ment->mnt_fsname, system_lv_name) == 0) {
+				endmntent(mountfile);
+				return 1;
+			}
+		}
+	}
+	while (ment);
+
+	endmntent(mountfile);
+	return 0;
+}
+
+/* Get the system LV name from the config file */
+static int find_system_lv(void)
+{
+	if (system_lv_name[0] == '\0') {
+#ifdef HAVE_CCS
+		int error;
+		ccs_node_t *ctree;
+
+		/* Read the cluster config file */
+		/* Open the config file */
+		error = open_ccs_file(&ctree, "clvm.ccs");
+		if (error) {
+			perror("reading config file");
+			return -1;
+		}
+
+		strcpy(system_lv_name, find_ccs_str(ctree,
+						    "cluster/systemlv", '/',
+						    "/dev/vg/system_lv"));
+
+		/* Finished with config file */
+		close_ccs_file(ctree);
+#else
+		if (getenv("CLVMD_SYSTEM_LV"))
+			strcpy(system_lv_name, getenv("CLVMD_SYSTEM_LV"));
+		else
+			return -1;
+#endif
+	}
+
+	/* See if it has been mounted outside our control */
+	if (is_really_mounted() != mounted) {
+		log_error
+		    ("The system LV state has been mounted/umounted outside the control of clvmd\n"
+		     "it cannot not be used for cluster communications until this is fixed.\n");
+		return -1;
+	}
+	return 0;
+}
+
+/* No prizes */
+int system_lv_umount(void)
+{
+	if (!mounted)
+		return 0;
+
+	if (umount(mount_point) < 0) {
+		log_error("umount of system LV (%s) failed: %m\n",
+			  system_lv_name);
+		return -1;
+	}
+
+	sync_unlock(lock_name, lockid);
+	mounted = 0;
+
+	/* Remove the mount point */
+	rmdir(mount_point);
+
+	return 0;
+}
+
+int system_lv_mount(int readwrite)
+{
+	int status;
+	int saved_errno;
+	int fd;
+
+	if (find_system_lv()) {
+		errno = EBUSY;
+		return -1;
+	}
+
+	/* Is it already mounted suitably? */
+	if (mounted) {
+		if (!readwrite || (readwrite && mounted_rw)) {
+			return 0;
+		} else {
+			/* Mounted RO and we need RW */
+			if (system_lv_umount() < 0)
+				return -1;
+		}
+	}
+
+	/* Randomize the mount point */
+	strcpy(mount_point, SYSTEM_LV_MOUNTPOINT);
+	fd = mkstemp(mount_point);
+	if (fd < 0) {
+		log_error("mkstemp for system LV mount point failed: %m\n");
+		return -1;
+	}
+
+	/* Race condition here but there's no mkstemp for directories */
+	close(fd);
+	unlink(mount_point);
+	mkdir(mount_point, 0600);
+
+	/* Make sure we have a system-lv lock */
+	status =
+	    sync_lock(lock_name, (readwrite) ? LKM_EXMODE : LKM_CRMODE, 0,
+		      &lockid);
+	if (status < 0)
+		return -1;
+
+	/* Mount it */
+	if (mount(system_lv_name, mount_point, SYSTEM_LV_FILESYSTEM,
+		  MS_MGC_VAL | MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_SYNCHRONOUS
+		  | (readwrite ? 0 : MS_RDONLY), NULL) < 0) {
+		/* mount(2) returns EINVAL if the volume has no FS on it. So, if we want to
+		   write to it we try to make a filesystem in it and retry the mount */
+		if (errno == EINVAL && readwrite) {
+			char cmd[256];
+
+			log_error("Attempting mkfs on system LV device %s\n",
+				  system_lv_name);
+			snprintf(cmd, sizeof(cmd), "/sbin/mkfs -t %s %s",
+				 SYSTEM_LV_FILESYSTEM, system_lv_name);
+			system(cmd);
+
+			if (mount
+			    (system_lv_name, mount_point, SYSTEM_LV_FILESYSTEM,
+			     MS_MGC_VAL | MS_NOSUID | MS_NODEV | MS_NOEXEC |
+			     MS_SYNCHRONOUS | (readwrite ? 0 : MS_RDONLY),
+			     NULL) == 0)
+				goto mounted;
+		}
+
+		saved_errno = errno;
+		log_error("mount of system LV (%s, %s, %s) failed: %m\n",
+			  system_lv_name, mount_point, SYSTEM_LV_FILESYSTEM);
+		sync_unlock(lock_name, lockid);
+		errno = saved_errno;
+		return -1;
+	}
+
+      mounted:
+/* Set the internal flags */
+	mounted = 1;
+	mounted_rw = readwrite;
+
+	return 0;
+}
+
+/* Erase *all* files in the root directory of the system LV.
+   This *MUST* be called with an appropriate lock held!
+   The LV is left mounted RW because it is assumed that the
+   caller wants to write something here after clearing some space */
+int system_lv_eraseall(void)
+{
+	DIR *dir;
+	struct dirent *ent;
+	char fname[PATH_MAX];
+
+	/* Must be mounted R/W */
+	system_lv_mount(1);
+
+	dir = opendir(mount_point);
+	if (!dir)
+		return -1;
+
+	while ((ent = readdir(dir))) {
+		struct stat st;
+		snprintf(fname, sizeof(fname), "%s/%s", mount_point,
+			 ent->d_name);
+
+		if (stat(fname, &st)) {
+			if (S_ISREG(st.st_mode))
+				unlink(fname);
+		}
+	}
+	closedir(dir);
+	return 0;
+}
+
+/* This is a "high-level" routine - it mounts the system LV, writes
+   the data into a file named after this node and then umounts the LV
+   again */
+int system_lv_write_data(char *data, ssize_t len)
+{
+	struct utsname nodeinfo;
+	char fname[PATH_MAX];
+	int outfile;
+	ssize_t thiswrite;
+	ssize_t written;
+
+	if (system_lv_mount(1))
+		return -1;
+
+	/* Build the file name we are goingto use. */
+	uname(&nodeinfo);
+	snprintf(fname, sizeof(fname), "%s/%s", mount_point, nodeinfo.nodename);
+
+	/* Open the file for output */
+	outfile = open(fname, O_RDWR | O_CREAT | O_TRUNC, 0600);
+	if (outfile < 0) {
+		int saved_errno = errno;
+		system_lv_umount();
+		errno = saved_errno;
+		return -1;
+	}
+
+	written = 0;
+	do {
+		thiswrite = write(outfile, data + written, len - written);
+		if (thiswrite > 0)
+			written += thiswrite;
+
+	} while (written < len && thiswrite > 0);
+
+	close(outfile);
+
+	system_lv_umount();
+	return (thiswrite < 0) ? -1 : 0;
+}
+
+/* This is a "high-level" routine - it mounts the system LV, reads
+   the data from a named file and then umounts the LV
+   again */
+int system_lv_read_data(char *fname_base, char *data, ssize_t *len)
+{
+	char fname[PATH_MAX];
+	int outfile;
+	struct stat st;
+	ssize_t filesize;
+	ssize_t thisread;
+	ssize_t readbytes;
+
+	if (system_lv_mount(0))
+		return -1;
+
+	/* Build the file name we are going to use. */
+	snprintf(fname, sizeof(fname), "%s/%s", mount_point, fname_base);
+
+	/* Get the file size and stuff. Actually we only need the file size but
+	   this will also check that the file exists */
+	if (stat(fname, &st) < 0) {
+		int saved_errno = errno;
+
+		log_error("stat of file %s on system LV failed: %m\n", fname);
+		system_lv_umount();
+		errno = saved_errno;
+		return -1;
+	}
+	filesize = st.st_size;
+
+	outfile = open(fname, O_RDONLY);
+	if (outfile < 0) {
+		int saved_errno = errno;
+
+		log_error("open of file %s on system LV failed: %m\n", fname);
+		system_lv_umount();
+		errno = saved_errno;
+		return -1;
+	}
+
+	readbytes = 0;
+	do {
+		thisread =
+		    read(outfile, data + readbytes, filesize - readbytes);
+		if (thisread > 0)
+			readbytes += thisread;
+
+	} while (readbytes < filesize && thisread > 0);
+
+	close(outfile);
+
+	system_lv_umount();
+
+	*len = readbytes;
+	return (thisread < 0) ? -1 : 0;
+}

daemons/clvmd/system-lv.h

@@ -0,0 +1,30 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifndef _CLVM_SYSTEM_LV_H
+#define _CLVM_SYSTEM_LV_H
+
+/* Prototypes for System-LV functions */
+
+/* "low-level" functions */
+extern int system_lv_umount(void);
+extern int system_lv_mount(int readwrite);
+extern int system_lv_eraseall(void);
+
+/* "high-level" functions */
+extern int system_lv_write_data(char *data, ssize_t len);
+extern int system_lv_read_data(char *fname_base, char *data, ssize_t *len);
+
+#endif

daemons/clvmd/tcp-comms.c

@@ -0,0 +1,444 @@
+/******************************************************************************
+*******************************************************************************
+**
+**  Copyright (C) Sistina Software, Inc.  2002-2003  All rights reserved.
+**
+*******************************************************************************
+******************************************************************************/
+
+/* This provides the inter-clvmd communications for a system without CMAN.
+   There is a listening TCP socket which accepts new connections in the
+   normal way.
+   It can also make outgoing connnections to the other clvmd nodes.
+*/
+
+
+#include <pthread.h>
+#include <sys/types.h>
+#include <sys/utsname.h>
+#include <sys/ioctl.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <fcntl.h>
+#include <string.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <errno.h>
+#include <syslog.h>
+#include <netdb.h>
+#include <assert.h>
+
+#include "clvm.h"
+#include "clvmd-comms.h"
+#include "clvmd.h"
+#include "clvmd-gulm.h"
+#include "hash.h"
+
+#define DEFAULT_TCP_PORT 21064
+
+static int listen_fd = -1;
+static int tcp_port;
+struct hash_table *sock_hash;
+
+static int get_our_ip_address(char *addr, int *family);
+static int read_from_tcpsock(struct local_client *fd, char *buf, int len, char *csid,
+			     struct local_client **new_client);
+
+/* Called by init_cluster() to open up the listening socket */
+int init_comms(unsigned short port)
+{
+    struct sockaddr_in6 addr;
+
+    sock_hash = hash_create(100);
+    tcp_port = port ? port : DEFAULT_TCP_PORT;
+
+    listen_fd = socket(AF_INET6, SOCK_STREAM, 0);
+
+    if (listen_fd < 0)
+    {
+	return -1;
+    }
+    else
+    {
+	int one = 1;
+	setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(int));
+    }
+
+    memset(&addr, 0, sizeof(addr)); // Bind to INADDR_ANY
+    addr.sin6_family = AF_INET6;
+    addr.sin6_port = htons(tcp_port);
+
+    if (bind(listen_fd, (struct sockaddr *)&addr, sizeof(addr)) < 0)
+    {
+	DEBUGLOG("Can't bind to port: %s\n", strerror(errno));
+	syslog(LOG_ERR, "Can't bind to port %d, is clvmd already running ?", tcp_port);
+	close(listen_fd);
+	return -1;
+    }
+
+    listen(listen_fd, 5);
+
+    return 0;
+}
+
+void tcp_remove_client(char *csid)
+ {
+    struct local_client *client;
+    DEBUGLOG("tcp_remove_client\n");
+
+    /* Don't actually close the socket here - that's the
+       job of clvmd.c whch will do the job when it notices the
+       other end has gone. We just need to remove the client(s) from
+       the hash table so we don't try to use it for sending any more */
+    client = hash_lookup_binary(sock_hash, csid, MAX_CSID_LEN);
+    if (client)
+    {
+	hash_remove_binary(sock_hash, csid, MAX_CSID_LEN);
+    }
+
+    /* Look for a mangled one too */
+    csid[0] ^= 0x80;
+
+    client = hash_lookup_binary(sock_hash, csid, MAX_CSID_LEN);
+    if (client)
+    {
+	hash_remove_binary(sock_hash, csid, MAX_CSID_LEN);
+    }
+
+    /* Put it back as we found it */
+    csid[0] ^= 0x80;
+}
+
+int alloc_client(int fd, char *csid, struct local_client **new_client)
+{
+    struct local_client *client;
+
+    DEBUGLOG("alloc_client %d csid = %s\n", fd, print_csid(csid));
+
+    /* Create a local_client and return it */
+    client = malloc(sizeof(struct local_client));
+    if (!client)
+    {
+	DEBUGLOG("malloc failed\n");
+	return -1;
+    }
+
+    memset(client, 0, sizeof(struct local_client));
+    client->fd = fd;
+    client->type = CLUSTER_DATA_SOCK;
+    client->callback = read_from_tcpsock;
+    if (new_client)
+	*new_client = client;
+
+    /* Add to our list of node sockets */
+    if (hash_lookup_binary(sock_hash, csid, MAX_CSID_LEN))
+    {
+	DEBUGLOG("alloc_client mangling CSID for second connection\n");
+	/* This is a duplicate connection but we can't close it because
+	   the other end may already have started sending.
+	   So, we mangle the IP address and keep it, all sending will
+	   go out of the main FD
+	*/
+	csid[0] ^= 0x80;
+	client->bits.net.flags = 1; /* indicate mangled CSID */
+
+        /* If it still exists then kill the connection as we should only
+           ever have one incoming connection from each node */
+        if (hash_lookup_binary(sock_hash, csid, MAX_CSID_LEN))
+        {
+	    DEBUGLOG("Multiple incoming connections from node\n");
+            syslog(LOG_ERR, " Bogus incoming connection from %d.%d.%d.%d\n", csid[0],csid[1],csid[2],csid[3]);
+
+	    free(client);
+            errno = ECONNREFUSED;
+            return -1;
+        }
+    }
+    hash_insert_binary(sock_hash, csid, MAX_CSID_LEN, client);
+
+    return 0;
+}
+
+int get_main_cluster_fd()
+{
+    return listen_fd;
+}
+
+
+/* Read on main comms (listen) socket, accept it */
+int cluster_fd_callback(struct local_client *fd, char *buf, int len, char *csid,
+			struct local_client **new_client)
+{
+    int newfd;
+    struct sockaddr_in6 addr;
+    socklen_t addrlen = sizeof(addr);
+    int status;
+    char name[MAX_CLUSTER_MEMBER_NAME_LEN];
+
+    DEBUGLOG("cluster_fd_callback\n");
+    *new_client = NULL;
+    newfd = accept(listen_fd, (struct sockaddr *)&addr, &addrlen);
+
+    DEBUGLOG("cluster_fd_callback, newfd=%d (errno=%d)\n", newfd, errno);
+    if (!newfd)
+    {
+	syslog(LOG_ERR, "error in accept: %m");
+	errno = EAGAIN;
+	return -1; /* Don't return an error or clvmd will close the listening FD */
+    }
+
+    /* Check that the client is a member of the cluster
+       and reject if not.
+    */
+    if (name_from_csid((char *)&addr.sin6_addr, name) < 0)
+    {
+	syslog(LOG_ERR, "Got connect from non-cluster node %s\n",
+	       print_csid((char *)&addr.sin6_addr));
+	DEBUGLOG("Got connect from non-cluster node %s\n",
+		 print_csid((char *)&addr.sin6_addr));
+	close(newfd);
+
+	errno = EAGAIN;
+	return -1;
+    }
+
+    status = alloc_client(newfd, (char *)&addr.sin6_addr, new_client);
+    if (status)
+    {
+	DEBUGLOG("cluster_fd_callback, alloc_client failed, status = %d\n", status);
+	close(newfd);
+	/* See above... */
+	errno = EAGAIN;
+	return -1;
+    }
+    DEBUGLOG("cluster_fd_callback, returning %d, %p\n", newfd, *new_client);
+    return newfd;
+}
+
+
+static int read_from_tcpsock(struct local_client *client, char *buf, int len, char *csid,
+			     struct local_client **new_client)
+{
+    struct sockaddr_in6 addr;
+    socklen_t slen = sizeof(addr);
+    int status;
+
+    DEBUGLOG("read_from_tcpsock fd %d\n", client->fd);
+    *new_client = NULL;
+
+    /* Get "csid" */
+    getpeername(client->fd, (struct sockaddr *)&addr, &slen);
+    memcpy(csid, &addr.sin6_addr, MAX_CSID_LEN);
+
+    status = read(client->fd, buf, len);
+
+    DEBUGLOG("read_from_tcpsock, status = %d(errno = %d)\n", status, errno);
+
+    /* Remove it from the hash table if there's an error, clvmd will
+       remove the socket from its lists and free the client struct */
+    if (status == 0 ||
+	(status < 0 && errno != EAGAIN && errno != EINTR))
+    {
+	char remcsid[MAX_CSID_LEN];
+
+	memcpy(remcsid, csid, MAX_CSID_LEN);
+	close(client->fd);
+
+	/* If the csid was mangled, then make sure we remove the right entry */
+	if (client->bits.net.flags)
+	    remcsid[0] ^= 0x80;
+	hash_remove_binary(sock_hash, remcsid, MAX_CSID_LEN);
+
+	/* Tell cluster manager layer */
+	add_down_node(remcsid);
+    }
+    else {
+	    /* Send it back to clvmd */
+	    process_message(client, buf, len, csid);
+    }
+    return status;
+}
+
+static int connect_csid(char *csid, struct local_client **newclient)
+{
+    int fd;
+    struct sockaddr_in6 addr;
+    int status;
+
+    DEBUGLOG("Connecting socket\n");
+    fd = socket(PF_INET6, SOCK_STREAM, 0);
+
+    if (fd < 0)
+    {
+	syslog(LOG_ERR, "Unable to create new socket: %m");
+	return -1;
+    }
+
+    addr.sin6_family = AF_INET6;
+    memcpy(&addr.sin6_addr, csid, MAX_CSID_LEN);
+    addr.sin6_port = htons(tcp_port);
+
+    DEBUGLOG("Connecting socket %d\n", fd);
+    if (connect(fd, (struct sockaddr *)&addr, sizeof(struct sockaddr_in6)) < 0)
+    {
+	syslog(LOG_ERR, "Unable to connect to remote node: %m");
+	DEBUGLOG("Unable to connect to remote node: %s\n", strerror(errno));
+	close(fd);
+	return -1;
+    }
+
+    status = alloc_client(fd, csid, newclient);
+    if (status)
+	close(fd);
+    else
+	add_client(*newclient);
+
+    /* If we can connect to it, it must be running a clvmd */
+    add_up_node(csid);
+    return status;
+}
+
+/* Send a message to a known CSID */
+static int tcp_send_message(void *buf, int msglen, unsigned char *csid, const char *errtext)
+{
+    int status;
+    struct local_client *client;
+    char ourcsid[MAX_CSID_LEN];
+
+    assert(csid);
+
+    DEBUGLOG("tcp_send_message, csid = %s, msglen = %d\n", print_csid(csid), msglen);
+
+    /* Don't connect to ourself */
+    get_our_csid(ourcsid);
+    if (memcmp(csid, ourcsid, MAX_CSID_LEN) == 0)
+	return msglen;
+
+    client = hash_lookup_binary(sock_hash, csid, MAX_CSID_LEN);
+    if (!client)
+    {
+	status = connect_csid(csid, &client);
+	if (status)
+	    return -1;
+    }
+    DEBUGLOG("tcp_send_message, fd = %d\n", client->fd);
+
+    return write(client->fd, buf, msglen);
+}
+
+
+int cluster_send_message(void *buf, int msglen, char *csid, const char *errtext)
+{
+    int status=0;
+
+    DEBUGLOG("cluster send message, csid = %p, msglen = %d\n", csid, msglen);
+
+    /* If csid is NULL then send to all known (not just connected) nodes */
+    if (!csid)
+    {
+	void *context = NULL;
+	char loop_csid[MAX_CSID_LEN];
+
+	/* Loop round all gulm-known nodes */
+	while (get_next_node_csid(&context, loop_csid))
+	{
+	    status = tcp_send_message(buf, msglen, loop_csid, errtext);
+	    if (status == 0 ||
+		(status < 0 && (errno == EAGAIN || errno == EINTR)))
+		break;
+	}
+    }
+    else
+    {
+
+	status = tcp_send_message(buf, msglen, csid, errtext);
+    }
+    return status;
+}
+
+/* To get our own IP address we get the locally bound address of the
+   socket that's talking to GULM in the assumption(eek) that it will
+   be on the "right" network in a multi-homed system */
+static int get_our_ip_address(char *addr, int *family)
+{
+	struct utsname info;
+
+	uname(&info);
+	get_ip_address(info.nodename, addr);
+
+	return 0;
+}
+
+/* Public version of above for those that don't care what protocol
+   we're using */
+void get_our_csid(char *csid)
+{
+    static char our_csid[MAX_CSID_LEN];
+    static int got_csid = 0;
+
+    if (!got_csid)
+    {
+	int family;
+
+	memset(our_csid, 0, sizeof(our_csid));
+	if (get_our_ip_address(our_csid, &family))
+	{
+	    got_csid = 1;
+	}
+    }
+    memcpy(csid, our_csid, MAX_CSID_LEN);
+}
+
+static void map_v4_to_v6(struct in_addr *ip4, struct in6_addr *ip6)
+{
+   ip6->s6_addr32[0] = 0;
+   ip6->s6_addr32[1] = 0;
+   ip6->s6_addr32[2] = htonl(0xffff);
+   ip6->s6_addr32[3] = ip4->s_addr;
+}
+
+/* Get someone else's IP address from DNS */
+int get_ip_address(char *node, char *addr)
+{
+    struct hostent *he;
+
+    memset(addr, 0, MAX_CSID_LEN);
+
+    // TODO: what do we do about multi-homed hosts ???
+    // CCSs ip_interfaces solved this but some bugger removed it.
+
+    /* Try IPv6 first. The man page for gethostbyname implies that
+       it will lookup ip6 & ip4 names, but it seems not to */
+    he = gethostbyname2(node, AF_INET6);
+    if (he)
+    {
+	memcpy(addr, he->h_addr_list[0],
+	       he->h_length);
+    }
+    else
+    {
+	he = gethostbyname2(node, AF_INET);
+	if (!he)
+	    return -1;
+	map_v4_to_v6((struct in_addr *)he->h_addr_list[0], (struct in6_addr *)addr);
+    }
+
+    return 0;
+}
+
+char *print_csid(char *csid)
+{
+    static char buf[128];
+    int *icsid = (int *)csid;
+
+    sprintf(buf, "[%x.%x.%x.%x]",
+	    icsid[0],icsid[1],icsid[2],icsid[3]);
+
+    return buf;
+}

daemons/clvmd/tcp-comms.h

@@ -0,0 +1,8 @@
+#include <netinet/in.h>
+
+#define MAX_CLUSTER_MESSAGE 1600
+#define MAX_CSID_LEN sizeof(struct in6_addr)
+#define MAX_CLUSTER_MEMBER_NAME_LEN 128
+
+extern int init_comms(unsigned short);
+extern char *print_csid(char *);

doc/example.conf

@@ -27,6 +27,12 @@ devices {
     # the device will be accepted or rejected (ignored).  Devices that
     # don't match any patterns are accepted.
 
+    # Be careful if there there are symbolic links or multiple filesystem 
+    # entries for the same device as each name is checked separately against
+    # the list of patterns.  The effect is that if any name matches any 'a'
+    # pattern, the device is accepted; otherwise if any name matches any 'r'
+    # pattern it is rejected; otherwise it is accepted.
+
     # Remember to run vgscan after you change this parameter to ensure 
     # that the cache file gets regenerated (see below).
 

doc/tagging.txt

@@ -0,0 +1,165 @@
+Tagging aims
+============
+  1) Ability to attach an unordered list of tags to LVM metadata objects.
+  2) Ability to add or remove tags easily.
+  3) Ability to select LVM objects for processing according to presence/absence
+     of specific tags.
+  4) Ability to control through the config file which VGs/LVs are activated 
+     on different machines using names or tags.
+  5) Ability to overlay settings from different config files e.g. override
+     some settings in a global config file locally.
+
+Clarifications
+==============
+  1) Tag character set: A-Za-z0-9_+.- 
+     Can't start with hyphen & max length is 128 (NAME_LEN).
+  2) LVM object types that can be tagged:
+       VG, LV, LV segment
+       PV - tags are stored in VG metadata so disappear when PV becomes orphaned
+     Snapshots can't be tagged, but their origin may be.
+  3) A tag can be used in place of any command line LVM object reference that
+     accepts (a) a list of objects; or (b) a single object as long as the
+     tag expands to a single object.  This is not supported everywhere yet.
+     Duplicate arguments in a list after argument expansion may get removed 
+     retaining the first copy of each argument.
+  4) Wherever there may be ambiguity of argument type, a tag must be prefixed 
+     by '@'; elsewhere an '@' prefix is optional.
+  5) LVM1 objects cannot be tagged, as the disk format doesn't support it.
+  6) Tags can be added or removed with --addtag or --deltag.
+
+Config file Extensions
+======================
+  To define host tags in config file:
+
+  tags {
+  	# Set a tag with the hostname
+	hosttags = 1
+
+	tag1 { }
+
+  	tag2 {
+		# If no exact match, tag is not set.
+		host_list = [ "hostname", "dbase" ]
+	}
+  }
+
+Activation config file example
+==============================
+  activation {
+      volume_list = [ "vg1/lvol0", "@database" ]
+  }
+
+  Matches against vgname, vgname/lvname or @tag set in *metadata*.
+  @* matches exactly against *any* tag set on the host.
+  The VG or LV only gets activated if a metadata tag matches.
+  The default if there is no match is not to activate.
+  If volume_list is not present and any tags are defined on the host 
+  then it only activates if a host tag matches a metadata tag.
+  If volume_list is not present and no tags are defined on the host 
+  then it does activate.
+
+Multiple config files
+=====================
+  (a) lvm.conf
+  (b) lvm_<host_tag>.conf
+
+  At startup, load lvm.conf.
+  Process tag settings.
+  If any host tags were defined, load lvm_tag.conf for each tag, if present.
+
+  When searching for a specific config file entry, search order is (b)
+  then (a), stopping at the first match.  
+  Within (b) use reverse order tags got set, so file for last tag set is
+  searched first.
+  New tags set in (b) *do* trigger additional config file loads. 
+
+Usage Examples
+==============
+  1) Simple activation control via metadata with static config files
+
+  lvm.conf:  (Identical on every machine - global settings)
+    tags {
+      hostname_tags = 1
+    }
+
+  From any machine in the cluster, add db1 to the list of machines that
+  activate vg1/lvol2:
+
+  lvchange --tag @db1 vg1/lvol2
+  (followed by lvchange -ay to actually activate it)
+
+
+  2) Multiple hosts.  
+
+    Activate vg1 only on the database hosts, db1 and db2.
+    Activate vg2 only on the fileserver host fs1.
+    Activate nothing initially on the fileserver backup host fsb1, but be
+    prepared for it to take over from fs1.
+
+  Option (i) - centralised admin, static configuration replicated between hosts  
+    # Add @database tag to vg1's metadata
+    vgchange --tag @database vg1
+
+    # Add @fileserver tag to vg2's metadata
+    vgchange --tag @fileserver vg2
+
+    lvm.conf:  (Identical on every machine)
+      tags {
+        database {
+          host_list = [ "db1", "db2" ]
+        }
+        fileserver {
+	  host_list = [ "fs1" ]
+        }
+        fileserverbackup {
+          host_list = [ "fsb1" ]
+        }
+      }
+
+      activation {
+        # Only activate if host has a tag that matches a metadata tag
+        volume_list = [ "@*" ]
+      }
+  
+  In the event of the fileserver host going down, vg2 can be brought up
+  on fsb1 by running *on any node* 'vgchange --tag @fileserverbackup vg2'
+  followed by 'vgchange -ay vg2'
+  
+  
+  Option (ii) - localised admin & configuation
+  (i.e. each host holds *locally* which classes of volumes to activate)
+    # Add @database tag to vg1's metadata
+    vgchange --tag @database vg1
+  
+    # Add @fileserver tag to vg2's metadata
+    vgchange --tag @fileserver vg2
+  
+    lvm.conf:  (Identical on every machine - global settings)
+      tags {
+        hosttags = 1
+      }
+  
+    lvm_db1.conf: (only needs to be on db1 - could be symlink to lvm_db.conf)
+      activation {
+        volume_list = [ "@database" ]
+      }
+  
+    lvm_db2.conf: (only needs to be on db2 - could be symlink to lvm_db.conf)
+      activation {
+        volume_list = [ "@database" ]
+      }
+  
+    lvm_fs1.conf: (only needs to be on fs1 - could be symlink to lvm_fs.conf)
+      activation {
+        volume_list = [ "@fileserver" ]
+      }
+  
+    If fileserver goes down, to bring a spare machine fsb1 in as fileserver,
+    create lvm_fsb1.conf on fsb1 (or symlink to lvm_fs.conf):
+
+      activation {
+        volume_list = [ "@fileserver" ]
+      }
+
+    and run 'vgchange -ay vg2' or 'vgchange -ay @fileserver'
+

include/.symlinks

@@ -1,3 +1,4 @@
+../daemons/clvmd/clvm.h
 ../lib/activate/activate.h
 ../lib/activate/targets.h
 ../lib/cache/lvmcache.h
@@ -30,7 +31,7 @@
 ../lib/log/log.h
 ../lib/metadata/lv_alloc.h
 ../lib/metadata/metadata.h
-../lib/metadata/segtypes.h
+../lib/metadata/segtype.h
 ../lib/mm/dbg_malloc.h
 ../lib/mm/memlock.h
 ../lib/mm/pool.h
@@ -46,3 +47,4 @@
 ../lib/report/report.h
 ../lib/uuid/uuid.h
 ../po/pogen.h
+../tools/version.h

include/Makefile.in

@@ -20,6 +20,8 @@ VPATH = @srcdir@
 
 LN_S = @LN_S@
 
+.PHONY: clean distclean all install pofile install_cluster
+
 all: .symlinks_created
 
 .symlinks_created: .symlinks
@@ -37,5 +39,5 @@ clean:
 
 install:
 
-.PHONY: clean distclean all install pofile
+install_cluster:
 

lib/Makefile.in

@@ -43,6 +43,7 @@ SOURCES =\
 	datastruct/str_list.c \
 	device/dev-cache.c \
 	device/dev-io.c \
+	device/dev-md.c \
 	device/device.c \
 	display/display.c \
 	error/errseg.c \
@@ -70,11 +71,12 @@ SOURCES =\
 	metadata/metadata.c \
 	metadata/mirror.c \
 	metadata/pv_map.c \
-	metadata/segtypes.c \
+	metadata/segtype.c \
 	metadata/snapshot_manip.c \
 	misc/crc.c \
 	misc/lvm-file.c \
 	misc/lvm-string.c \
+	mm/dbg_malloc.c \
 	mm/memlock.c \
 	mm/pool.c \
 	regex/matcher.c \
@@ -104,6 +106,14 @@ ifeq ("@POOL@", "internal")
 	format_pool/pool_label.c
 endif
 
+ifeq ("@CLUSTER@", "internal")
+  SOURCES += locking/cluster_locking.c
+endif
+
+ifeq ("@CLUSTER@", "shared")
+  SUBDIRS += locking
+endif
+
 ifeq ("@SNAPSHOTS@", "internal")
   SOURCES += snapshot/snapshot.c
 endif
@@ -112,10 +122,6 @@ ifeq ("@MIRRORS@", "internal")
   SOURCES += mirror/mirrored.c
 endif
 
-ifeq ("@DEBUG@", "yes")
-  SOURCES += mm/dbg_malloc.c
-endif
-
 ifeq ("@DEVMAPPER@", "yes")
   SOURCES +=\
 	activate/dev_manager.c \

lib/activate/dev_manager.c

@@ -21,7 +21,7 @@
 #include "lvm-string.h"
 #include "fs.h"
 #include "defaults.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "display.h"
 #include "toolcontext.h"
 #include "targets.h"
@@ -165,89 +165,6 @@ static inline void _clear_flag(struct dev_layer *dl, int bit)
 	dl->flags &= ~(1 << bit);
 }
 
-/*
- * Device layer names are all of the form <vg>-<lv>-<layer>, any
- * other hyphens that appear in these names are quoted with yet
- * another hyphen.  The top layer of any device has no layer
- * name.  eg, vg0-lvol0.
- */
-static void _count_hyphens(const char *str, size_t *len, int *hyphens)
-{
-	const char *ptr;
-
-	for (ptr = str; *ptr; ptr++, (*len)++)
-		if (*ptr == '-')
-			(*hyphens)++;
-}
-
-/*
- * Copies a string, quoting hyphens with hyphens.
- */
-static void _quote_hyphens(char **out, const char *src)
-{
-	while (*src) {
-		if (*src == '-')
-			*(*out)++ = '-';
-
-		*(*out)++ = *src++;
-	}
-}
-
-/*
- * <vg>-<lv>-<layer> or if !layer just <vg>-<lv>.
- */
-static char *_build_name(struct pool *mem, const char *vg,
-			 const char *lv, const char *layer)
-{
-	size_t len = 0;
-	int hyphens = 0;
-	char *r, *out;
-
-	_count_hyphens(vg, &len, &hyphens);
-	_count_hyphens(lv, &len, &hyphens);
-
-	if (layer && *layer)
-		_count_hyphens(layer, &len, &hyphens);
-
-	len += hyphens + 2;
-
-	if (!(r = pool_alloc(mem, len))) {
-		stack;
-		return NULL;
-	}
-
-	out = r;
-	_quote_hyphens(&out, vg);
-	*out++ = '-';
-	_quote_hyphens(&out, lv);
-
-	if (layer && *layer) {
-		*out++ = '-';
-		_quote_hyphens(&out, layer);
-	}
-	*out = '\0';
-
-	return r;
-}
-
-/* Find start of LV component in hyphenated name */
-static char *_find_lv_name(char *vg)
-{
-	char *c = vg;
-
-	while (*c && *(c + 1)) {
-		if (*c == '-') {
-			if (*(c + 1) == '-')
-				c++;
-			else
-				return (c + 1);
-		}
-		c++;
-	}
-
-	return NULL;
-}
-
 static char *_build_dlid(struct pool *mem, const char *lvid, const char *layer)
 {
 	char *dlid;
@@ -519,10 +436,16 @@ static int _percent(struct dev_manager *dm, const char *name, const char *uuid,
 	return 0;
 }
 
-static int _rename(struct dev_layer *dl, char *newname)
+static int _rename(struct dev_manager *dm, struct dev_layer *dl, char *newname)
 {
 	int r = 1;
 	struct dm_task *dmt;
+	char *vgname, *lvname, *layer;
+
+	if (!split_dm_name(dm->mem, dl->name, &vgname, &lvname, &layer)) {
+		log_error("Couldn't split up dm layer name %s", dl->name);
+		return 0;
+	}
 
 	log_verbose("Renaming %s to %s", dl->name, newname);
 
@@ -537,11 +460,13 @@ static int _rename(struct dev_layer *dl, char *newname)
 		goto out;
 	}
 
-	if (!(r = dm_task_run(dmt)))
+	if (!(r = dm_task_run(dmt))) {
 		log_error("Couldn't rename device '%s'.", dl->name);
+		goto out;
+	}
 
 	if (r && _get_flag(dl, VISIBLE))
-		fs_rename_lv(dl->lv, newname, _find_lv_name(dl->name));
+		fs_rename_lv(dl->lv, newname, lvname);
 
 	dl->name = newname;
 
@@ -994,7 +919,7 @@ struct dev_manager *dev_manager_create(struct cmd_context *cmd,
 	struct pool *mem;
 	struct dev_manager *dm;
 
-	if (!(mem = pool_create(16 * 1024))) {
+	if (!(mem = pool_create("dev_manager", 16 * 1024))) {
 		stack;
 		return NULL;
 	}
@@ -1052,7 +977,7 @@ int dev_manager_info(struct dev_manager *dm, const struct logical_volume *lv,
 	/*
 	 * Build a name for the top layer.
 	 */
-	if (!(name = _build_name(dm->mem, lv->vg->name, lv->name, NULL))) {
+	if (!(name = build_dm_name(dm->mem, lv->vg->name, lv->name, NULL))) {
 		stack;
 		return 0;
 	}
@@ -1077,7 +1002,7 @@ int dev_manager_snapshot_percent(struct dev_manager *dm,
 	/*
 	 * Build a name for the top layer.
 	 */
-	if (!(name = _build_name(dm->mem, lv->vg->name, lv->name, NULL))) {
+	if (!(name = build_dm_name(dm->mem, lv->vg->name, lv->name, NULL))) {
 		stack;
 		return 0;
 	}
@@ -1109,7 +1034,7 @@ int dev_manager_mirror_percent(struct dev_manager *dm,
 	/*
 	 * Build a name for the top layer.
 	 */
-	if (!(name = _build_name(dm->mem, lv->vg->name, lv->name, NULL))) {
+	if (!(name = build_dm_name(dm->mem, lv->vg->name, lv->name, NULL))) {
 		stack;
 		return 0;
 	}
@@ -1173,7 +1098,7 @@ static struct dev_layer *_create_layer(struct dev_manager *dm,
 	char *name, *dlid;
 	struct dev_layer *dl;
 
-	if (!(name = _build_name(dm->mem, lv->vg->name, lv->name, layer))) {
+	if (!(name = build_dm_name(dm->mem, lv->vg->name, lv->name, layer))) {
 		stack;
 		return NULL;
 	}
@@ -1618,11 +1543,11 @@ static int _create_rec(struct dev_manager *dm, struct dev_layer *dl)
 	if (dl->info.exists) {
 		if ((suffix = rindex(dl->dlid, '-')))
 			suffix++;
-		newname = _build_name(dm->mem, dm->vg_name, dl->lv->name,
-				      suffix);
+		newname = build_dm_name(dm->mem, dm->vg_name, dl->lv->name,
+					suffix);
 		if (strcmp(newname, dl->name)) {
 			if (!_suspend_parents(dm, dl) ||
-			    !_suspend(dl) || !_rename(dl, newname)) {
+			    !_suspend(dl) || !_rename(dm, dl, newname)) {
 				stack;
 				return 0;
 			}
@@ -2227,8 +2152,8 @@ int dev_manager_lv_mknodes(const struct logical_volume *lv)
 {
 	char *name;
 
-	if (!(name = _build_name(lv->vg->cmd->mem, lv->vg->name,
-				 lv->name, NULL))) {
+	if (!(name = build_dm_name(lv->vg->cmd->mem, lv->vg->name,
+				   lv->name, NULL))) {
 		stack;
 		return 0;
 	}

lib/activate/fs.c

@@ -46,7 +46,7 @@ static int _mk_dir(const char *dev_dir, const char *vg_name)
 		return 1;
 
 	log_very_verbose("Creating directory %s", vg_path);
-	if (mkdir(vg_path, 0555)) {
+	if (mkdir(vg_path, 0777)) {
 		log_sys_error("mkdir", vg_path);
 		return 0;
 	}
@@ -65,10 +65,10 @@ static int _rm_dir(const char *dev_dir, const char *vg_name)
 		return 0;
 	}
 
-	log_very_verbose("Removing directory %s", vg_path);
-
-	if (is_empty_dir(vg_path))
+	if (is_empty_dir(vg_path)) {
+		log_very_verbose("Removing directory %s", vg_path);
 		rmdir(vg_path);
+	}
 
 	return 1;
 }
@@ -202,9 +202,9 @@ static int _rm_link(const char *dev_dir, const char *vg_name,
 	}
 
 	if (lstat(lv_path, &buf) || !S_ISLNK(buf.st_mode)) {
-		if (errno != ENOENT)
-			log_error("%s not symbolic link - not removing",
-				  lv_path);
+		if (errno == ENOENT)
+			return 1;
+		log_error("%s not symbolic link - not removing", lv_path);
 		return 0;
 	}
 

lib/commands/toolcontext.c

@@ -33,7 +33,7 @@
 #include "display.h"
 #include "memlock.h"
 #include "str_list.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "lvmcache.h"
 
 #ifdef HAVE_LIBDL
@@ -519,8 +519,8 @@ static struct dev_filter *_init_filter_components(struct cmd_context *cmd)
 
 	/* regex filter. Optional. */
 	if (!(cn = find_config_node(cmd->cft->root, "devices/filter")))
-		log_debug("devices/filter not found in config file: no regex "
-			  "filter installed");
+		log_very_verbose("devices/filter not found in config file: "
+				 "no regex filter installed");
 
 	else if (!(filters[nr_filt++] = regex_filter_create(cn->v))) {
 		log_error("Failed to create regex device filter");
@@ -537,6 +537,7 @@ static struct dev_filter *_init_filter_components(struct cmd_context *cmd)
 	/* md component filter. Optional, non-critical. */
 	if (find_config_bool(cmd->cft->root, "devices/md_component_detection",
 			     DEFAULT_MD_COMPONENT_DETECTION)) {
+		init_md_filtering(1);
 		if ((filters[nr_filt] = md_filter_create()))
 			nr_filt++;
 	}
@@ -798,7 +799,7 @@ struct cmd_context *create_toolcontext(struct arg *the_args)
 #endif
 
 	if (!setlocale(LC_ALL, ""))
-		log_error("setlocale failed");
+		log_very_verbose("setlocale failed");
 
 #ifdef INTL_PACKAGE
 	bindtextdomain(INTL_PACKAGE, LOCALEDIR);
@@ -827,7 +828,7 @@ struct cmd_context *create_toolcontext(struct arg *the_args)
 	if (*cmd->sys_dir && !create_dir(cmd->sys_dir))
 		goto error;
 
-	if (!(cmd->libmem = pool_create(4 * 1024))) {
+	if (!(cmd->libmem = pool_create("library", 4 * 1024))) {
 		log_error("Library memory pool creation failed");
 		return 0;
 	}
@@ -858,7 +859,7 @@ struct cmd_context *create_toolcontext(struct arg *the_args)
 	if (!_init_filters(cmd))
 		goto error;
 
-	if (!(cmd->mem = pool_create(4 * 1024))) {
+	if (!(cmd->mem = pool_create("command", 4 * 1024))) {
 		log_error("Command memory pool creation failed");
 		return 0;
 	}

lib/config/config.c

@@ -99,7 +99,7 @@ static int _tok_match(const char *str, const char *b, const char *e)
 struct config_tree *create_config_tree(const char *filename)
 {
 	struct cs *c;
-	struct pool *mem = pool_create(10 * 1024);
+	struct pool *mem = pool_create("config", 10 * 1024);
 
 	if (!mem) {
 		stack;

lib/config/defaults.h

@@ -96,9 +96,9 @@
 #define DEFAULT_PVS_COLS "pv_name,vg_name,pv_fmt,pv_attr,pv_size,pv_free"
 #define DEFAULT_SEGS_COLS "lv_name,vg_name,lv_attr,stripes,segtype,seg_size"
 
-#define DEFAULT_LVS_COLS_VERB "lv_name,vg_name,seg_count,lv_attr,lv_size,lv_major,lv_minor,origin,snap_percent,move_pv,copy_percent,lv_uuid"
+#define DEFAULT_LVS_COLS_VERB "lv_name,vg_name,seg_count,lv_attr,lv_size,lv_major,lv_minor,lv_kernel_major,lv_kernel_minor,origin,snap_percent,move_pv,copy_percent,lv_uuid"
 #define DEFAULT_VGS_COLS_VERB "vg_name,vg_attr,vg_extent_size,pv_count,lv_count,snap_count,vg_size,vg_free,vg_uuid"
-#define DEFAULT_PVS_COLS_VERB "pv_name,vg_name,pv_fmt,pv_attr,pv_size,pv_free,pv_uuid"
+#define DEFAULT_PVS_COLS_VERB "pv_name,vg_name,pv_fmt,pv_attr,pv_size,pv_free,dev_size,pv_uuid"
 #define DEFAULT_SEGS_COLS_VERB "lv_name,vg_name,lv_attr,seg_start,seg_size,stripes,segtype,stripesize,chunksize"
 
 #define DEFAULT_LVS_SORT "vg_name,lv_name"

lib/device/dev-cache.c

@@ -82,6 +82,7 @@ struct device *dev_create_file(const char *filename, struct device *dev,
 	dev->dev = 0;
 	dev->fd = -1;
 	dev->open_count = 0;
+	dev->block_size = -1;
 	memset(dev->pvid, 0, sizeof(dev->pvid));
 	list_init(&dev->open_list);
 
@@ -101,6 +102,7 @@ static struct device *_dev_create(dev_t d)
 	dev->dev = d;
 	dev->fd = -1;
 	dev->open_count = 0;
+	dev->block_size = -1;
 	dev->end = UINT64_C(0);
 	memset(dev->pvid, 0, sizeof(dev->pvid));
 	list_init(&dev->open_list);
@@ -387,7 +389,7 @@ int dev_cache_init(void)
 {
 	_cache.names = NULL;
 
-	if (!(_cache.mem = pool_create(10 * 1024))) {
+	if (!(_cache.mem = pool_create("dev_cache", 10 * 1024))) {
 		stack;
 		return 0;
 	}

lib/device/dev-io.c

@@ -81,7 +81,9 @@ static int _io(struct device_area *where, void *buffer, int should_write)
 	}
 
 	if (lseek(fd, (off_t) where->start, SEEK_SET) < 0) {
-		log_sys_error("lseek", dev_name(where->dev));
+		log_error("%s: lseek %" PRIu64 " failed: %s",
+			  dev_name(where->dev), (uint64_t) where->start,
+			  strerror(errno));
 		return 0;
 	}
 
@@ -92,6 +94,14 @@ static int _io(struct device_area *where, void *buffer, int should_write)
 			    read(fd, buffer, (size_t) where->size - total);
 		while ((n < 0) && ((errno == EINTR) || (errno == EAGAIN)));
 
+		if (n < 0)
+			log_error("%s: %s failed after %" PRIu64 " of %" PRIu64
+				  " at %" PRIu64 ": %s", dev_name(where->dev),
+				  should_write ? "write" : "read",
+				  (uint64_t) total,
+				  (uint64_t) where->size,
+				  (uint64_t) where->start, strerror(errno));
+
 		if (n <= 0)
 			break;
 
@@ -114,14 +124,18 @@ static int _io(struct device_area *where, void *buffer, int should_write)
  */
 static int _get_block_size(struct device *dev, unsigned int *size)
 {
-	int s;
+	const char *name = dev_name(dev);
 
-	if (ioctl(dev_fd(dev), BLKBSZGET, &s) < 0) {
-		log_sys_error("ioctl BLKBSZGET", dev_name(dev));
-		return 0;
+	if ((dev->block_size == -1)) {
+		if (ioctl(dev_fd(dev), BLKBSZGET, &dev->block_size) < 0) {
+			log_sys_error("ioctl BLKBSZGET", name);
+			return 0;
+		}
+		log_debug("%s: block size is %u bytes", name, dev->block_size);
 	}
 
-	*size = (unsigned int) s;
+	*size = (unsigned int) dev->block_size;
+
 	return 1;
 }
 
@@ -212,12 +226,11 @@ static int _aligned_io(struct device_area *where, void *buffer,
  * Public functions
  *---------------------------------------------------------------*/
 
-int dev_get_size(struct device *dev, uint64_t *size)
+int dev_get_size(const struct device *dev, uint64_t *size)
 {
 	int fd;
 	const char *name = dev_name(dev);
 
-	log_very_verbose("Getting size of %s", name);
 	if ((fd = open(name, O_RDONLY)) < 0) {
 		log_sys_error("open", name);
 		return 0;
@@ -225,22 +238,27 @@ int dev_get_size(struct device *dev, uint64_t *size)
 
 	if (ioctl(fd, BLKGETSIZE64, size) < 0) {
 		log_sys_error("ioctl BLKGETSIZE64", name);
-		close(fd);
+		if (close(fd))
+			log_sys_error("close", name);
 		return 0;
 	}
 
 	*size >>= BLKSIZE_SHIFT;	/* Convert to sectors */
-	close(fd);
+	if (close(fd))
+		log_sys_error("close", name);
+
+	log_very_verbose("%s: size is %" PRIu64 " sectors", name, *size);
+
 	return 1;
 }
 
+/* FIXME Unused
 int dev_get_sectsize(struct device *dev, uint32_t *size)
 {
 	int fd;
 	int s;
 	const char *name = dev_name(dev);
 
-	log_very_verbose("Getting size of %s", name);
 	if ((fd = open(name, O_RDONLY)) < 0) {
 		log_sys_error("open", name);
 		return 0;
@@ -254,8 +272,12 @@ int dev_get_sectsize(struct device *dev, uint32_t *size)
 
 	close(fd);
 	*size = (uint32_t) s;
+
+	log_very_verbose("%s: sector size is %" PRIu32 " bytes", name, *size);
+
 	return 1;
 }
+*/
 
 void dev_flush(struct device *dev)
 {
@@ -274,8 +296,20 @@ int dev_open_flags(struct device *dev, int flags, int direct, int quiet)
 	const char *name;
 
 	if (dev->fd >= 0) {
-		dev->open_count++;
-		return 1;
+		if ((dev->flags & DEV_OPENED_RW) ||
+		    ((flags & O_ACCMODE) != O_RDWR)) {
+			dev->open_count++;
+			return 1;
+		}
+
+		if (dev->open_count) {
+			/* FIXME Ensure we never get here */
+			log_debug("WARNING: %s already opened read-only", 
+				  dev_name(dev));
+			dev->open_count++;
+		}
+
+		dev_close_immediate(dev);
 	}
 
 	if (memlock())
@@ -300,19 +334,29 @@ int dev_open_flags(struct device *dev, int flags, int direct, int quiet)
 		flags |= O_DIRECT;
 #endif
 
+#ifdef O_NOATIME
+	/* Don't update atime on device inodes */
+	if (!(dev->flags & DEV_REGULAR))
+		flags |= O_NOATIME;
+#endif
+
 	if ((dev->fd = open(name, flags, 0777)) < 0) {
 		log_sys_error("open", name);
 		return 0;
 	}
 
-	dev->open_count = 1;
+	dev->open_count++;
 	dev->flags &= ~DEV_ACCESSED_W;
+	if ((flags & O_ACCMODE) == O_RDWR)
+		dev->flags |= DEV_OPENED_RW;
+	else
+		dev->flags &= ~DEV_OPENED_RW;
 
 	if (!(dev->flags & DEV_REGULAR) &&
 	    ((fstat(dev->fd, &buf) < 0) || (buf.st_rdev != dev->dev))) {
 		log_error("%s: fstat failed: Has device name changed?", name);
-		dev_close(dev);
-		dev->fd = -1;
+		dev_close_immediate(dev);
+		dev->open_count = 0;
 		return 0;
 	}
 
@@ -321,12 +365,13 @@ int dev_open_flags(struct device *dev, int flags, int direct, int quiet)
 		dev_flush(dev);
 #endif
 
-	if ((flags & O_CREAT) && !(flags & O_TRUNC)) {
+	if ((flags & O_CREAT) && !(flags & O_TRUNC))
 		dev->end = lseek(dev->fd, (off_t) 0, SEEK_END);
-	}
 
 	list_add(&_open_devices, &dev->open_list);
-	log_debug("Opened %s", dev_name(dev));
+
+	log_debug("Opened %s %s", dev_name(dev),
+		  dev->flags & DEV_OPENED_RW ? "RW" : "RO");
 
 	return 1;
 }
@@ -354,6 +399,7 @@ static void _close(struct device *dev)
 	if (close(dev->fd))
 		log_sys_error("close", dev_name(dev));
 	dev->fd = -1;
+	dev->block_size = -1;
 	list_del(&dev->open_list);
 
 	log_debug("Closed %s", dev_name(dev));
@@ -379,8 +425,11 @@ static int _dev_close(struct device *dev, int immediate)
 		dev_flush(dev);
 #endif
 
+	if (dev->open_count > 0)
+		dev->open_count--;
+
 	/* FIXME lookup device in cache to get vgname and see if it's locked? */
-	if (--dev->open_count < 1 && (immediate || !vgs_locked()))
+	if (immediate || (dev->open_count < 1 && !vgs_locked()))
 		_close(dev);
 
 	return 1;
@@ -412,8 +461,10 @@ int dev_read(struct device *dev, uint64_t offset, size_t len, void *buffer)
 {
 	struct device_area where;
 
-	if (!dev->open_count)
+	if (!dev->open_count) {
+		stack;
 		return 0;
+	}
 
 	where.dev = dev;
 	where.start = offset;
@@ -431,8 +482,10 @@ int dev_append(struct device *dev, size_t len, void *buffer)
 {
 	int r;
 
-	if (!dev->open_count)
+	if (!dev->open_count) {
+		stack;
 		return 0;
+	}
 
 	r = dev_write(dev, dev->end, len, buffer);
 	dev->end += (uint64_t) len;
@@ -447,8 +500,10 @@ int dev_write(struct device *dev, uint64_t offset, size_t len, void *buffer)
 {
 	struct device_area where;
 
-	if (!dev->open_count)
+	if (!dev->open_count) {
+		stack;
 		return 0;
+	}
 
 	where.dev = dev;
 	where.start = offset;
@@ -495,6 +550,5 @@ int dev_zero(struct device *dev, uint64_t offset, size_t len)
 	if (!dev_close(dev))
 		stack;
 
-	/* FIXME: Always display error */
 	return (len == 0);
 }

lib/device/dev-md.c

@@ -0,0 +1,69 @@
+/*
+ * Copyright (C) 2004 Luca Berra
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU Lesser General Public License v.2.1.
+ *
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "lib.h"
+#include "metadata.h"
+#include "xlate.h"
+
+/* Lifted from <linux/raid/md_p.h> because of difficulty including it */
+
+#define MD_SB_MAGIC 0xa92b4efc
+#define MD_RESERVED_BYTES (64 * 1024)
+#define MD_RESERVED_SECTORS (MD_RESERVED_BYTES / 512)
+#define MD_NEW_SIZE_SECTORS(x) ((x & ~(MD_RESERVED_SECTORS - 1)) \
+				- MD_RESERVED_SECTORS)
+
+/*
+ * Returns -1 on error
+ */
+int dev_is_md(struct device *dev, uint64_t *sb)
+{
+	int ret = 0;
+
+#ifdef linux
+
+	uint64_t size, sb_offset;
+	uint32_t md_magic;
+
+	if (!dev_get_size(dev, &size)) {
+		stack;
+		return -1;
+	}
+
+	if (size < MD_RESERVED_SECTORS * 2)
+		return 0;
+
+	if (!dev_open(dev)) {
+		stack;
+		return -1;
+	}
+
+	sb_offset = MD_NEW_SIZE_SECTORS(size) << SECTOR_SHIFT;
+
+	/* Check if it is an md component device. */
+	if (dev_read(dev, sb_offset, sizeof(uint32_t), &md_magic) &&
+	    (md_magic == xlate32(MD_SB_MAGIC))) {
+		if (sb)
+			*sb = sb_offset;
+		ret = 1;
+	}
+
+	if (!dev_close(dev))
+		stack;
+
+#endif
+	return ret;
+}
+

lib/device/device.c

@@ -13,6 +13,93 @@
  * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  */
 
+#include "lib.h"
+#include "lvm-types.h"
+#include "device.h"
+#include "metadata.h"
+#include "filter.h"
+#include "xlate.h"
+
+/* See linux/genhd.h and fs/partitions/msdos */
+
+#define PART_MAGIC 0xAA55
+#define PART_MAGIC_OFFSET UINT64_C(0x1FE)
+#define PART_OFFSET UINT64_C(0x1BE)
+
+struct partition {
+        uint8_t boot_ind;
+        uint8_t head;
+        uint8_t sector;
+        uint8_t cyl;
+        uint8_t sys_ind;	/* partition type */
+        uint8_t end_head;
+        uint8_t end_sector;
+        uint8_t end_cyl;
+        uint32_t start_sect;
+        uint32_t nr_sects;
+} __attribute__((packed));
+
+static int _is_partitionable(struct device *dev)
+{
+	int parts = max_partitions(MAJOR(dev->dev));
+
+	if ((parts <= 1) || (MINOR(dev->dev) % parts))
+		return 0;
+
+	return 1;
+}
+
+static int _has_partition_table(struct device *dev)
+{
+	int ret = 0;
+	unsigned p;
+	uint8_t buf[SECTOR_SIZE];
+	uint16_t *part_magic;
+	struct partition *part;
+
+	if (!dev_open(dev)) {
+		stack;
+		return -1;
+	}
+
+	if (!dev_read(dev, 0, sizeof(buf), &buf)) {
+		stack;
+		goto out;
+	}
+
+	/* FIXME Check for other types of partition table too */
+
+	/* Check for msdos partition table */
+	part_magic = (uint16_t *)(buf + PART_MAGIC_OFFSET);
+	if ((*part_magic == xlate16(PART_MAGIC))) {
+		part = (struct partition *) (buf + PART_OFFSET);
+		for (p = 0; p < 4; p++, part++) {
+			/* Table is invalid if boot indicator not 0 or 0x80 */
+			if ((part->boot_ind & 0x7f)) {
+				ret = 0;
+				break;
+			}
+			/* Must have at least one non-empty partition */
+			if (part->nr_sects)
+				ret = 1;
+		}
+	}
+
+      out:
+	if (!dev_close(dev))
+		stack;
+
+	return ret;
+}
+
+int is_partitioned_dev(struct device *dev)
+{
+	if (!_is_partitionable(dev))
+		return 0;
+
+	return _has_partition_table(dev);
+}
+
 #if 0
 #include <sys/stat.h>
 #include <sys/mman.h>
@@ -27,24 +114,13 @@
 #include <linux/major.h>
 #include <linux/genhd.h>
 
-#include "dbg_malloc.h"
-#include "log.h"
-#include "dev-cache.h"
-#include "metadata.h"
-#include "device.h"
-
 int _get_partition_type(struct dev_filter *filter, struct device *d);
 
-#define MINOR_PART(dm, d) (MINOR((d)->dev) % dev_max_partitions(dm, (d)->dev))
+#define MINOR_PART(dev) (MINOR((dev)->dev) % max_partitions(MINOR((dev)->dev)))
 
-int is_whole_disk(struct dev_filter *filter, struct device *d)
+int is_extended_partition(struct device *d)
 {
-	return (MINOR_PART(dm, d)) ? 0 : 1;
-}
-
-int is_extended_partition(struct dev_mgr *dm, struct device *d)
-{
-	return (MINOR_PART(dm, d) > 4) ? 1 : 0;
+	return (MINOR_PART(d) > 4) ? 1 : 0;
 }
 
 struct device *dev_primary(struct dev_mgr *dm, struct device *d)

lib/device/device.h

@@ -22,6 +22,7 @@
 #define DEV_ACCESSED_W		0x00000001	/* Device written to? */
 #define DEV_REGULAR		0x00000002	/* Regular file? */
 #define DEV_ALLOCED		0x00000004	/* dbg_malloc used */
+#define DEV_OPENED_RW		0x00000008	/* Opened RW */
 
 /*
  * All devices in LVM will be represented by one of these.
@@ -34,6 +35,7 @@ struct device {
 	/* private */
 	int fd;
 	int open_count;
+	int block_size;
 	uint32_t flags;
 	uint64_t end;
 	struct list open_list;
@@ -55,7 +57,7 @@ struct device_area {
 /*
  * All io should use these routines.
  */
-int dev_get_size(struct device *dev, uint64_t *size);
+int dev_get_size(const struct device *dev, uint64_t *size);
 int dev_get_sectsize(struct device *dev, uint32_t *size);
 
 /* Use quiet version if device number could change e.g. when opening LV */
@@ -89,15 +91,14 @@ static inline const char *dev_name(const struct device *dev)
 /* Return a valid device name from the alias list; NULL otherwise */
 const char *dev_name_confirmed(struct device *dev, int quiet);
 
+/* Does device contain md superblock?  If so, where? */
+int dev_is_md(struct device *dev, uint64_t *sb);
+
 /* FIXME Check partition type if appropriate */
 
 #define is_lvm_partition(a) 1
+/* int is_lvm_partition(const char *name); */
 
-/*
-static inline int is_lvm_partition(const char *name)
-{
-	return 1;
-}
-*/
+int is_partitioned_dev(struct device *dev);
 
 #endif

lib/display/display.c

@@ -18,7 +18,7 @@
 #include "display.h"
 #include "activate.h"
 #include "toolcontext.h"
-#include "segtypes.h"
+#include "segtype.h"
 
 #define SIZE_BUF 128
 
@@ -662,3 +662,22 @@ void vgdisplay_short(struct volume_group *vg)
 			       SIZE_SHORT));
 	return;
 }
+
+void display_formats(struct cmd_context *cmd)
+{
+	struct format_type *fmt;
+
+	list_iterate_items(fmt, &cmd->formats) {
+		log_print("%s", fmt->name);
+	}
+}
+
+void display_segtypes(struct cmd_context *cmd)
+{
+	struct segment_type *segtype;
+
+	list_iterate_items(segtype, &cmd->segtypes) {
+		log_print("%s", segtype->name);
+	}
+}
+

lib/display/display.h

@@ -45,6 +45,9 @@ void vgdisplay_full(struct volume_group *vg);
 void vgdisplay_colons(struct volume_group *vg);
 void vgdisplay_short(struct volume_group *vg);
 
+void display_formats(struct cmd_context *cmd);
+void display_segtypes(struct cmd_context *cmd);
+
 /*
  * Allocation policy display conversion routines.
  */

lib/error/errseg.c

@@ -16,7 +16,7 @@
 #include "pool.h"
 #include "list.h"
 #include "toolcontext.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "display.h"
 #include "text_export.h"
 #include "text_import.h"
@@ -97,5 +97,7 @@ struct segment_type *init_error_segtype(struct cmd_context *cmd)
 	segtype->private = NULL;
 	segtype->flags = SEG_CAN_SPLIT | SEG_VIRTUAL;
 
+	log_very_verbose("Initialised segtype: %s", segtype->name);
+
 	return segtype;
 }

lib/filters/filter-md.c

@@ -28,42 +28,24 @@
 
 static int _ignore_md(struct dev_filter *f, struct device *dev)
 {
-	uint64_t size, sector;
-	uint32_t md_magic;
-
-	if (!dev_get_size(dev, &size)) {
-		stack;
-		return 0;
-	}
-
-	if (size < MD_RESERVED_SECTORS * 2)
-		/*
-		 * We could ignore it since it is obviously too
-		 * small, but that's not our job.
-		 */
+	int ret;
+	
+	if (!md_filtering())
 		return 1;
+	
+	ret = dev_is_md(dev, NULL);
 
-	if (!dev_open(dev)) {
-		stack;
+	if (ret == 1) {
+		log_debug("%s: Skipping md component device", dev_name(dev));
 		return 0;
 	}
 
-	sector = MD_NEW_SIZE_SECTORS(size);
-
-	/* Check if it is an md component device. */
-	if (dev_read(dev, sector << SECTOR_SHIFT, sizeof(uint32_t), &md_magic)) {
-		if (md_magic == MD_SB_MAGIC) {
-			log_debug("%s: Skipping md component device",
-				  dev_name(dev));
-			if (!dev_close(dev))
-				stack;
-			return 0;
-		}
+	if (ret < 0) {
+		log_debug("%s: Skipping: error in md component detection",
+			  dev_name(dev));
+		return 0;
 	}
 
-	if (!dev_close(dev))
-		stack;
-
 	return 1;
 }
 

lib/filters/filter-persistent.c

@@ -212,13 +212,10 @@ static int _lookup_p(struct dev_filter *f, struct device *dev)
 			sl = list_item(ah, struct str_list);
 			hash_insert(pf->devices, sl->str, l);
 		}
-	}
+	} else if (l == PF_BAD_DEVICE)
+			log_debug("%s: Skipping (cached)", dev_name(dev));
 
-	if (l == PF_BAD_DEVICE) {
-		log_debug("%s: Skipping (cached)", dev_name(dev));
-		return 0;
-	} else
-		return 1;
+	return (l == PF_BAD_DEVICE) ? 0 : 1;
 }
 
 static void _destroy(struct dev_filter *f)

lib/filters/filter-regex.c

@@ -101,7 +101,7 @@ static int _build_matcher(struct rfilter *rf, struct config_value *val)
 	unsigned count = 0;
 	int i, r = 0;
 
-	if (!(scratch = pool_create(1024))) {
+	if (!(scratch = pool_create("filter matcher", 1024))) {
 		stack;
 		return 0;
 	}
@@ -186,6 +186,9 @@ static int _accept_p(struct dev_filter *f, struct device *dev)
 		first = 0;
 	}
 
+	if (rejected)
+		log_debug("%s: Skipping (regex)", dev_name(dev));
+
 	/*
 	 * pass everything that doesn't match
 	 * anything.
@@ -201,7 +204,7 @@ static void _destroy(struct dev_filter *f)
 
 struct dev_filter *regex_filter_create(struct config_value *patterns)
 {
-	struct pool *mem = pool_create(10 * 1024);
+	struct pool *mem = pool_create("filter regex", 10 * 1024);
 	struct rfilter *rf;
 	struct dev_filter *f;
 

lib/filters/filter-sysfs.c

@@ -169,6 +169,8 @@ static int _read_devs(struct dev_set *ds, const char *dir)
 {
         struct dirent *d;
         DIR *dr;
+	unsigned char dtype;
+	struct stat info;
 	char path[PATH_MAX];
 	dev_t dev;
 	int r = 1;
@@ -189,19 +191,31 @@ static int _read_devs(struct dev_set *ds, const char *dir)
 			continue;
 		}
 
-		if (d->d_type == DT_DIR) {
+		dtype = d->d_type;
+
+		if (dtype == DT_UNKNOWN) {
+			if (lstat(path, &info) >= 0) {
+				if (S_ISLNK(info.st_mode))
+					dtype = DT_LNK;
+				else if (S_ISDIR(info.st_mode))
+					dtype = DT_DIR;
+				else if (S_ISREG(info.st_mode))
+					dtype = DT_REG;
+			}
+		}
+
+		if (dtype == DT_DIR) {
 			if (!_read_devs(ds, path)) {
 				r = 0;
 				break;
 			}
 		}
 
-		if ((d->d_type == DT_REG && !strcmp(d->d_name, "dev")))
+		if ((dtype == DT_REG && !strcmp(d->d_name, "dev")))
 			if (!_read_dev(path, &dev) || !_set_insert(ds, dev)) {
 				r = 0;
 				break;
 			}
-
 	}
 
         if (closedir(dr))
@@ -234,7 +248,11 @@ static int _accept_p(struct dev_filter *f, struct device *dev)
 	if (ds->initialised != 1)
 		return 1;
 
-	return _set_lookup(ds, dev->dev);
+	if (!_set_lookup(ds, dev->dev)) {
+		log_debug("%s: Skipping (sysfs)", dev_name(dev));
+		return 0;
+	} else
+		return 1;
 }
 
 static void _destroy(struct dev_filter *f)
@@ -253,7 +271,7 @@ struct dev_filter *sysfs_filter_create(const char *proc)
 	if (!_locate_sysfs_blocks(proc, sys_block, sizeof(sys_block)))
 		return NULL;
 
-	if (!(mem = pool_create(256))) {
+	if (!(mem = pool_create("sysfs", 256))) {
 		log_error("sysfs pool creation failed");
 		return NULL;
 	}

lib/filters/filter.c

@@ -18,6 +18,7 @@
 #include "filter.h"
 #include "lvm-string.h"
 #include "config.h"
+#include "metadata.h"
 
 #include <dirent.h>
 #include <unistd.h>
@@ -25,7 +26,11 @@
 #include <fcntl.h>
 #include <limits.h>
 
-#define NUMBER_OF_MAJORS 256
+#define NUMBER_OF_MAJORS 4096
+
+/* FIXME Make this sparse */
+/* 0 means LVM won't use this major number. */
+static int _max_partitions_by_major[NUMBER_OF_MAJORS];
 
 typedef struct {
 	const char *name;
@@ -39,12 +44,19 @@ int md_major(void)
 	return _md_major;
 }
 
-/* This list can be supplemented with devices/types in the config file */
+/*
+ * Devices are only checked for partition tables if their minor number
+ * is a multiple of the number corresponding to their type below
+ * i.e. this gives the granularity of whole-device minor numbers.
+ * Use 1 if the device is not partitionable.
+ *
+ * The list can be supplemented with devices/types in the config file.
+ */
 static const device_info_t device_info[] = {
-	{"ide", 16},		/* IDE disk */
+	{"ide", 64},		/* IDE disk */
 	{"sd", 16},		/* SCSI disk */
-	{"md", 16},		/* Multiple Disk driver (SoftRAID) */
-	{"loop", 16},		/* Loop device */
+	{"md", 1},		/* Multiple Disk driver (SoftRAID) */
+	{"loop", 1},		/* Loop device */
 	{"dasd", 4},		/* DASD disk (IBM S/390, zSeries) */
 	{"dac960", 8},		/* DAC960 */
 	{"nbd", 16},		/* Network Block Device */
@@ -53,37 +65,59 @@ static const device_info_t device_info[] = {
 	{"ubd", 16},		/* User-mode virtual block device */
 	{"ataraid", 16},	/* ATA Raid */
 	{"drbd", 16},		/* Distributed Replicated Block Device */
+	{"emcpower", 16},	/* EMC Powerpath */
 	{"power2", 16},		/* EMC Powerpath */
 	{"i2o_block", 16},	/* i2o Block Disk */
+	{"iseries/vd", 8},	/* iSeries disks */
 	{NULL, 0}
 };
 
 static int _passes_lvm_type_device_filter(struct dev_filter *f,
 					  struct device *dev)
 {
-	int fd;
 	const char *name = dev_name(dev);
+	int ret = 0;
+	uint64_t size;
 
 	/* Is this a recognised device type? */
-	if (!(((int *) f->private)[MAJOR(dev->dev)])) {
+	if (!_max_partitions_by_major[MAJOR(dev->dev)]) {
 		log_debug("%s: Skipping: Unrecognised LVM device type %"
 			  PRIu64, name, (uint64_t) MAJOR(dev->dev));
 		return 0;
 	}
 
 	/* Check it's accessible */
-	if ((fd = open(name, O_RDONLY)) < 0) {
-		log_debug("%s: Skipping: open failed: %s", name,
-			  strerror(errno));
+	if (!dev_open_flags(dev, O_RDONLY, 0, 0)) {
+		log_debug("%s: Skipping: open failed", name);
 		return 0;
 	}
+	
+	/* Check it's not too small */
+	if (!dev_get_size(dev, &size)) {
+		log_debug("%s: Skipping: dev_get_size failed", name);
+		goto out;
+	}
 
-	close(fd);
+	if (size < PV_MIN_SIZE) {
+		log_debug("%s: Skipping: Too small to hold a PV", name);
+		goto out;
+	}
 
-	return 1;
+	if (is_partitioned_dev(dev)) {
+		log_debug("%s: Skipping: Partition table signature found",
+			  name);
+		goto out;
+	}
+
+	ret = 1;
+
+      out:
+	dev_close(dev);
+
+	return ret;
 }
 
-static int *_scan_proc_dev(const char *proc, const struct config_node *cn)
+static int _scan_proc_dev(const char *proc, const struct config_node *cn)
 {
 	char line[80];
 	char proc_devices[PATH_MAX];
@@ -93,35 +127,31 @@ static int *_scan_proc_dev(const char *proc, const struct config_node *cn)
 	int blocksection = 0;
 	size_t dev_len = 0;
 	struct config_value *cv;
-	int *max_partitions_by_major;
 	char *name;
 
-	if (!(max_partitions_by_major =
-	      dbg_malloc(sizeof(int) * NUMBER_OF_MAJORS))) {
-		log_error("Filter failed to allocate max_partitions_by_major");
-		return NULL;
-	}
 
 	if (!*proc) {
 		log_verbose("No proc filesystem found: using all block device "
 			    "types");
 		for (i = 0; i < NUMBER_OF_MAJORS; i++)
-			max_partitions_by_major[i] = 1;
-		return max_partitions_by_major;
+			_max_partitions_by_major[i] = 1;
+		return 1;
 	}
 
+	/* All types unrecognised initially */
+	memset(_max_partitions_by_major, 0, sizeof(int) * NUMBER_OF_MAJORS);
+
 	if (lvm_snprintf(proc_devices, sizeof(proc_devices),
 			 "%s/devices", proc) < 0) {
 		log_error("Failed to create /proc/devices string");
-		return NULL;
+		return 0;
 	}
 
 	if (!(pd = fopen(proc_devices, "r"))) {
 		log_sys_error("fopen", proc_devices);
-		return NULL;
+		return 0;
 	}
 
-	memset(max_partitions_by_major, 0, sizeof(int) * NUMBER_OF_MAJORS);
 	while (fgets(line, 80, pd) != NULL) {
 		i = 0;
 		while (line[i] == ' ' && line[i] != '\0')
@@ -156,13 +186,13 @@ static int *_scan_proc_dev(const char *proc, const struct config_node *cn)
 			if (dev_len <= strlen(line + i) &&
 			    !strncmp(device_info[j].name, line + i, dev_len) &&
 			    (line_maj < NUMBER_OF_MAJORS)) {
-				max_partitions_by_major[line_maj] =
+				_max_partitions_by_major[line_maj] =
 				    device_info[j].max_partitions;
 				break;
 			}
 		}
 
-		if (max_partitions_by_major[line_maj] || !cn)
+		if (!cn)
 			continue;
 
 		/* Check devices/types for local variations */
@@ -170,7 +200,7 @@ static int *_scan_proc_dev(const char *proc, const struct config_node *cn)
 			if (cv->type != CFG_STRING) {
 				log_error("Expecting string in devices/types "
 					  "in config file");
-				return NULL;
+				return 0;
 			}
 			dev_len = strlen(cv->v.str);
 			name = cv->v.str;
@@ -179,24 +209,29 @@ static int *_scan_proc_dev(const char *proc, const struct config_node *cn)
 				log_error("Max partition count missing for %s "
 					  "in devices/types in config file",
 					  name);
-				return NULL;
+				return 0;
 			}
 			if (!cv->v.i) {
 				log_error("Zero partition count invalid for "
 					  "%s in devices/types in config file",
 					  name);
-				return NULL;
+				return 0;
 			}
 			if (dev_len <= strlen(line + i) &&
 			    !strncmp(name, line + i, dev_len) &&
 			    (line_maj < NUMBER_OF_MAJORS)) {
-				max_partitions_by_major[line_maj] = cv->v.i;
+				_max_partitions_by_major[line_maj] = cv->v.i;
 				break;
 			}
 		}
 	}
 	fclose(pd);
-	return max_partitions_by_major;
+	return 1;
+}
+
+int max_partitions(int major)
+{
+	return _max_partitions_by_major[major];
 }
 
 struct dev_filter *lvm_type_filter_create(const char *proc,
@@ -211,8 +246,9 @@ struct dev_filter *lvm_type_filter_create(const char *proc,
 
 	f->passes_filter = _passes_lvm_type_device_filter;
 	f->destroy = lvm_type_filter_destroy;
+	f->private = NULL;
 
-	if (!(f->private = _scan_proc_dev(proc, cn))) {
+	if (!_scan_proc_dev(proc, cn)) {
 		stack;
 		return NULL;
 	}
@@ -222,7 +258,6 @@ struct dev_filter *lvm_type_filter_create(const char *proc,
 
 void lvm_type_filter_destroy(struct dev_filter *f)
 {
-	dbg_free(f->private);
 	dbg_free(f);
 	return;
 }

lib/filters/filter.h

@@ -21,7 +21,9 @@
 #include <sys/stat.h>
 
 #ifdef linux
-#  include <linux/kdev_t.h>
+#  define MAJOR(dev)	((dev & 0xfff00) >> 8)
+#  define MINOR(dev)	((dev & 0xff) | ((dev >> 12) & 0xfff00))
+#  define MKDEV(ma,mi)	((mi & 0xff) | (ma << 8) | ((mi & ~0xff) << 12))
 #else
 #  define MAJOR(x) major((x))
 #  define MINOR(x) minor((x))
@@ -35,4 +37,6 @@ void lvm_type_filter_destroy(struct dev_filter *f);
 
 int md_major(void);
 
+int max_partitions(int major);
+
 #endif

lib/format1/Makefile.in

@@ -29,8 +29,6 @@ LIB_SHARED = liblvm2format1.so
 
 include $(top_srcdir)/make.tmpl
 
-.PHONY: install
-
 install: liblvm2format1.so
 	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) $< \
 		$(libdir)/liblvm2format1.so.$(LIB_VERSION)

lib/format1/format1.c

@@ -24,6 +24,7 @@
 #include "lvmcache.h"
 #include "lvm1-label.h"
 #include "format1.h"
+#include "segtype.h"
 
 #define FMT_LVM1_NAME "lvm1"
 
@@ -182,7 +183,7 @@ static struct volume_group *_vg_read(struct format_instance *fid,
 				     const char *vg_name,
 				     struct metadata_area *mda)
 {
-	struct pool *mem = pool_create(1024 * 10);
+	struct pool *mem = pool_create("lvm1 vg_read", 1024 * 10);
 	struct list pvs;
 	struct volume_group *vg = NULL;
 	list_init(&pvs);
@@ -275,7 +276,7 @@ static int _flatten_vg(struct format_instance *fid, struct pool *mem,
 static int _vg_write(struct format_instance *fid, struct volume_group *vg,
 		     struct metadata_area *mda)
 {
-	struct pool *mem = pool_create(1024 * 10);
+	struct pool *mem = pool_create("lvm1 vg_write", 1024 * 10);
 	struct list pvds;
 	int r = 0;
 
@@ -298,7 +299,7 @@ static int _vg_write(struct format_instance *fid, struct volume_group *vg,
 static int _pv_read(const struct format_type *fmt, const char *pv_name,
 		    struct physical_volume *pv, struct list *mdas)
 {
-	struct pool *mem = pool_create(1024);
+	struct pool *mem = pool_create("lvm1 pv_read", 1024);
 	struct disk_list *dl;
 	struct device *dev;
 	int r = 0;
@@ -420,7 +421,7 @@ static int _pv_write(const struct format_type *fmt, struct physical_volume *pv,
 	pv->pe_size = pv->pe_count = 0;
 	pv->pe_start = PE_ALIGN;
 
-	if (!(mem = pool_create(1024))) {
+	if (!(mem = pool_create("lvm1 pv_write", 1024))) {
 		stack;
 		return 0;
 	}
@@ -492,6 +493,17 @@ static int _vg_setup(struct format_instance *fid, struct volume_group *vg)
 	return 1;
 }
 
+static int _segtype_supported (struct format_instance *fid, 
+			       struct segment_type *segtype)
+{
+	if (!(segtype->flags & SEG_FORMAT1_SUPPORT)) {
+		stack;
+		return 0;
+	}
+
+	return 1;
+}
+
 static struct metadata_area_ops _metadata_format1_ops = {
 	vg_read:_vg_read,
 	vg_write:_vg_write,
@@ -542,6 +554,7 @@ static struct format_handler _format1_ops = {
 	pv_write:_pv_write,
 	lv_setup:_lv_setup,
 	vg_setup:_vg_setup,
+	segtype_supported:_segtype_supported,
 	create_instance:_create_instance,
 	destroy_instance:_destroy_instance,
 	destroy:_destroy,
@@ -578,5 +591,7 @@ struct format_type *init_format(struct cmd_context *cmd)
 		return NULL;
 	}
 
+	log_very_verbose("Initialised format: %s", fmt->name);
+
 	return fmt;
 }

lib/format1/import-export.c

@@ -25,7 +25,7 @@
 #include "lvm-string.h"
 #include "filter.h"
 #include "toolcontext.h"
-#include "segtypes.h"
+#include "segtype.h"
 
 #include <time.h>
 

lib/format1/import-extents.c

@@ -20,7 +20,7 @@
 #include "disk-rep.h"
 #include "lv_alloc.h"
 #include "display.h"
-#include "segtypes.h"
+#include "segtype.h"
 
 /*
  * After much thought I have decided it is easier,
@@ -344,7 +344,7 @@ int import_extents(struct cmd_context *cmd, struct volume_group *vg,
 		   struct list *pvds)
 {
 	int r = 0;
-	struct pool *scratch = pool_create(10 * 1024);
+	struct pool *scratch = pool_create("lvm1 import_extents", 10 * 1024);
 	struct hash_table *maps;
 
 	if (!scratch) {

lib/format1/vg_number.c

@@ -30,7 +30,7 @@ int get_free_vg_number(struct format_instance *fid, struct dev_filter *filter,
 	struct list *pvh;
 	struct list all_pvs;
 	struct disk_list *dl;
-	struct pool *mem = pool_create(10 * 1024);
+	struct pool *mem = pool_create("lvm1 vg_number", 10 * 1024);
 	int numbers[MAX_VG], i, r = 0;
 
 	list_init(&all_pvs);

lib/format_pool/Makefile.in

@@ -26,8 +26,6 @@ LIB_SHARED = liblvm2formatpool.so
 
 include $(top_srcdir)/make.tmpl
 
-.PHONY: install
-
 install: liblvm2formatpool.so
 	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) $< \
 		$(libdir)/liblvm2formatpool.so.$(LIB_VERSION)

lib/format_pool/disk_rep.c

@@ -259,7 +259,7 @@ static int _read_vg_pds(const struct format_type *fmt, struct pool *mem,
 
 	/* FIXME: maybe should return a different error in memory
 	 * allocation failure */
-	if (!(tmpmem = pool_create(512))) {
+	if (!(tmpmem = pool_create("pool read_vg", 512))) {
 		stack;
 		return 0;
 	}

lib/format_pool/format_pool.c

@@ -181,7 +181,7 @@ static struct volume_group *_vg_read(struct format_instance *fid,
 				     const char *vg_name,
 				     struct metadata_area *mda)
 {
-	struct pool *mem = pool_create(1024);
+	struct pool *mem = pool_create("pool vg_read", 1024);
 	struct list pds;
 	struct volume_group *vg = NULL;
 
@@ -227,7 +227,7 @@ static int _pv_setup(const struct format_type *fmt,
 static int _pv_read(const struct format_type *fmt, const char *pv_name,
 		    struct physical_volume *pv, struct list *mdas)
 {
-	struct pool *mem = pool_create(1024);
+	struct pool *mem = pool_create("pool pv_read", 1024);
 	struct pool_list *pl;
 	struct device *dev;
 	int r = 0;
@@ -357,5 +357,7 @@ struct format_type *init_format(struct cmd_context *cmd)
 		return NULL;
 	}
 
+	log_very_verbose("Initialised format: %s", fmt->name);
+
 	return fmt;
 }

lib/format_pool/import_export.c

@@ -22,7 +22,7 @@
 #include "lv_alloc.h"
 #include "str_list.h"
 #include "display.h"
-#include "segtypes.h"
+#include "segtype.h"
 
 /* This file contains only imports at the moment... */
 
@@ -146,7 +146,7 @@ int import_pool_pvs(const struct format_type *fmt, struct volume_group *vg,
 		}
 		pl->pv = pvl->pv;
 		pvl->mdas = NULL;
-		pvl->alloc_areas = NULL;
+		pvl->pe_ranges = NULL;
 		list_add(pvs, &pvl->list);
 	}
 

lib/format_text/export.c

@@ -20,7 +20,7 @@
 #include "pool.h"
 #include "display.h"
 #include "lvm-string.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "text_export.h"
 
 #include <stdarg.h>
@@ -652,7 +652,7 @@ static int _build_pv_names(struct formatter *f, struct volume_group *vg)
 	struct physical_volume *pv;
 	char buffer[32], *name;
 
-	if (!(f->mem = pool_create(512))) {
+	if (!(f->mem = pool_create("text pv_names", 512))) {
 		stack;
 		goto bad;
 	}

lib/format_text/format-text.c

@@ -1682,14 +1682,15 @@ struct format_type *create_text_format(struct cmd_context *cmd)
 		}
 	}
 
-	if (!(cn = find_config_node(cmd->cft->root, "metadata/disk_areas")))
-		return fmt;
-
-	for (cn = cn->child; cn; cn = cn->sib) {
-		if (!_get_config_disk_area(cmd, cn, &mda_lists->raws))
-			goto err;
+	if ((cn = find_config_node(cmd->cft->root, "metadata/disk_areas"))) {
+		for (cn = cn->child; cn; cn = cn->sib) {
+			if (!_get_config_disk_area(cmd, cn, &mda_lists->raws))
+				goto err;
+		}
 	}
 
+	log_very_verbose("Initialised format: %s", fmt->name);
+
 	return fmt;
 
       err:

lib/format_text/import_vsn1.c

@@ -22,7 +22,7 @@
 #include "toolcontext.h"
 #include "lvmcache.h"
 #include "lv_alloc.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "text_import.h"
 
 typedef int (*section_fn) (struct format_instance * fid, struct pool * mem,

lib/locking/.exported_symbols

@@ -0,0 +1,4 @@
+locking_init
+locking_end
+lock_resource
+reset_locking

lib/locking/Makefile.in

@@ -0,0 +1,30 @@
+#
+# Copyright (C) 2003-2004 Sistina Software, Inc. All rights reserved.
+# Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+#
+# This file is part of the LVM2.
+#
+# This copyrighted material is made available to anyone wishing to use,
+# modify, copy, or redistribute it subject to the terms and conditions
+# of the GNU General Public License v.2.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software Foundation,
+# Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+srcdir = @srcdir@
+top_srcdir = @top_srcdir@
+VPATH = @srcdir@
+
+SOURCES = cluster_locking.c
+
+LIB_SHARED = liblvm2clusterlock.so
+
+include $(top_srcdir)/make.tmpl
+
+install install_cluster: liblvm2clusterlock.so
+	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) $< \
+		$(libdir)/liblvm2clusterlock.so.$(LIB_VERSION)
+	$(LN_S) -f liblvm2clusterlock.so.$(LIB_VERSION) \
+		$(libdir)/liblvm2clusterlock.so
+

lib/locking/cluster_locking.c

@@ -0,0 +1,461 @@
+/*
+ * Copyright (C) 2002-2004 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/*
+ * Locking functions for LVM.
+ * The main purpose of this part of the library is to serialise LVM
+ * management operations across a cluster.
+ */
+
+#include "lib.h"
+#include "clvm.h"
+#include "lvm-string.h"
+#include "locking.h"
+#include "locking_types.h"
+
+#include <stddef.h>
+#include <sys/socket.h>
+#include <sys/un.h>
+#include <unistd.h>
+
+#ifndef CLUSTER_LOCKING_INTERNAL
+int lock_resource(struct cmd_context *cmd, const char *resource, int flags);
+void locking_end(void);
+int locking_init(int type, struct config_tree *cf, uint32_t *flags);
+#endif
+
+typedef struct lvm_response {
+	char node[255];
+	char *response;
+	int status;
+	int len;
+} lvm_response_t;
+
+/*
+ * This gets stuck at the start of memory we allocate so we
+ * can sanity-check it at deallocation time
+ */
+#define LVM_SIGNATURE 0x434C564D
+
+/*
+ * NOTE: the LVMD uses the socket FD as the client ID, this means
+ * that any client that calls fork() will inherit the context of
+ * it's parent.
+ */
+static int _clvmd_sock = -1;
+
+/* FIXME Install SIGPIPE handler? */
+
+/* Open connection to the Cluster Manager daemon */
+static int _open_local_sock(void)
+{
+	int local_socket;
+	struct sockaddr_un sockaddr;
+
+	/* Open local socket */
+	if ((local_socket = socket(PF_UNIX, SOCK_STREAM, 0)) < 0) {
+		log_error("Local socket creation failed: %s", strerror(errno));
+		return -1;
+	}
+
+	memset(&sockaddr, 0, sizeof(sockaddr));
+	memcpy(sockaddr.sun_path, CLVMD_SOCKNAME, sizeof(CLVMD_SOCKNAME));
+
+	sockaddr.sun_family = AF_UNIX;
+
+	if (connect(local_socket,(struct sockaddr *) &sockaddr,
+		    sizeof(sockaddr))) {
+		int saved_errno = errno;
+
+		log_error("connect() failed on local socket: %s",
+			  strerror(errno));
+		if (close(local_socket))
+			stack;
+
+		errno = saved_errno;
+		return -1;
+	}
+
+	return local_socket;
+}
+
+/* Send a request and return the status */
+static int _send_request(char *inbuf, int inlen, char **retbuf)
+{
+	char outbuf[PIPE_BUF];
+	struct clvm_header *outheader = (struct clvm_header *) outbuf;
+	int len;
+	int off;
+	int buflen;
+	int err;
+
+	/* Send it to CLVMD */
+ rewrite:
+	if ( (err = write(_clvmd_sock, inbuf, inlen)) != inlen) {
+	        if (err == -1 && errno == EINTR)
+		        goto rewrite;
+		log_error("Error writing data to clvmd: %s", strerror(errno));
+		return 0;
+	}
+
+	/* Get the response */
+ reread:
+	if ((len = read(_clvmd_sock, outbuf, sizeof(struct clvm_header))) < 0) {
+	        if (errno == EINTR)
+		        goto reread;
+		log_error("Error reading data from clvmd: %s", strerror(errno));
+		return 0;
+	}
+
+	if (len == 0) {
+		log_error("EOF reading CLVMD");
+		errno = ENOTCONN;
+		return 0;
+	}
+
+	/* Allocate buffer */
+	buflen = len + outheader->arglen;
+	*retbuf = dbg_malloc(buflen);
+	if (!*retbuf) {
+		errno = ENOMEM;
+		return 0;
+	}
+
+	/* Copy the header */
+	memcpy(*retbuf, outbuf, len);
+	outheader = (struct clvm_header *) *retbuf;
+
+	/* Read the returned values */
+	off = 1;		/* we've already read the first byte */
+	while (off <= outheader->arglen && len > 0) {
+		len = read(_clvmd_sock, outheader->args + off,
+			   buflen - off - offsetof(struct clvm_header, args));
+		if (len > 0)
+			off += len;
+	}
+
+	/* Was it an error ? */
+	if (outheader->status < 0) {
+		errno = -outheader->status;
+		log_error("cluster request failed: %s", strerror(errno));
+		return 0;
+	}
+
+	return 1;
+}
+
+/* Build the structure header and parse-out wildcard node names */
+static void _build_header(struct clvm_header *head, int cmd, const char *node,
+			  int len)
+{
+	head->cmd = cmd;
+	head->status = 0;
+	head->flags = 0;
+	head->clientid = 0;
+	head->arglen = len;
+
+	if (node) {
+		/*
+		 * Allow a couple of special node names:
+		 * "*" for all nodes,
+		 * "." for the local node only
+		 */
+		if (strcmp(node, "*") == 0) {
+			head->node[0] = '\0';
+		} else if (strcmp(node, ".") == 0) {
+			head->node[0] = '\0';
+			head->flags = CLVMD_FLAG_LOCAL;
+		} else
+			strcpy(head->node, node);
+	} else
+		head->node[0] = '\0';
+}
+
+/*
+ * Send a message to a(or all) node(s) in the cluster and wait for replies
+ */
+static int _cluster_request(char cmd, const char *node, void *data, int len,
+			   lvm_response_t ** response, int *num)
+{
+	char outbuf[sizeof(struct clvm_header) + len + strlen(node) + 1];
+	int *outptr;
+	char *inptr;
+	char *retbuf = NULL;
+	int status;
+	int i;
+	int num_responses = 0;
+	struct clvm_header *head = (struct clvm_header *) outbuf;
+	lvm_response_t *rarray;
+
+	*num = 0;
+
+	if (_clvmd_sock == -1)
+		_clvmd_sock = _open_local_sock();
+
+	if (_clvmd_sock == -1)
+		return 0;
+
+	_build_header(head, cmd, node, len);
+	memcpy(head->node + strlen(head->node) + 1, data, len);
+
+	status = _send_request(outbuf, sizeof(struct clvm_header) +
+			      strlen(head->node) + len, &retbuf);
+	if (!status)
+		goto out;
+
+	/* Count the number of responses we got */
+	head = (struct clvm_header *) retbuf;
+	inptr = head->args;
+	while (inptr[0]) {
+		num_responses++;
+		inptr += strlen(inptr) + 1;
+		inptr += sizeof(int);
+		inptr += strlen(inptr) + 1;
+	}
+
+	/*
+	 * Allocate response array.
+	 * With an extra pair of INTs on the front to sanity
+	 * check the pointer when we are given it back to free
+	 */
+	outptr = dbg_malloc(sizeof(lvm_response_t) * num_responses +
+			    sizeof(int) * 2);
+	if (!outptr) {
+		errno = ENOMEM;
+		status = 0;
+		goto out;
+	}
+
+	*response = (lvm_response_t *) (outptr + 2);
+	outptr[0] = LVM_SIGNATURE;
+	outptr[1] = num_responses;
+	rarray = *response;
+
+	/* Unpack the response into an lvm_response_t array */
+	inptr = head->args;
+	i = 0;
+	while (inptr[0]) {
+		strcpy(rarray[i].node, inptr);
+		inptr += strlen(inptr) + 1;
+
+		rarray[i].status = *(int *) inptr;
+		inptr += sizeof(int);
+
+		rarray[i].response = dbg_malloc(strlen(inptr) + 1);
+		if (rarray[i].response == NULL) {
+			/* Free up everything else and return error */
+			int j;
+			for (j = 0; j < i; j++)
+				dbg_free(rarray[i].response);
+			free(outptr);
+			errno = ENOMEM;
+			status = -1;
+			goto out;
+		}
+
+		strcpy(rarray[i].response, inptr);
+		rarray[i].len = strlen(inptr);
+		inptr += strlen(inptr) + 1;
+		i++;
+	}
+	*num = num_responses;
+	*response = rarray;
+
+      out:
+	if (retbuf)
+		dbg_free(retbuf);
+
+	return status;
+}
+
+/* Free reply array */
+static int _cluster_free_request(lvm_response_t * response)
+{
+	int *ptr = (int *) response - 2;
+	int i;
+	int num;
+
+	/* Check it's ours to free */
+	if (response == NULL || *ptr != LVM_SIGNATURE) {
+		errno = EINVAL;
+		return 0;
+	}
+
+	num = ptr[1];
+
+	for (i = 0; i < num; i++) {
+		dbg_free(response[i].response);
+	}
+
+	dbg_free(ptr);
+
+	return 1;
+}
+
+static int _lock_for_cluster(unsigned char cmd, unsigned int flags, char *name)
+{
+	int status;
+	int i;
+	char *args;
+	const char *node = "";
+	int len;
+	int saved_errno = errno;
+	lvm_response_t *response = NULL;
+	int num_responses;
+
+	assert(name);
+
+	len = strlen(name) + 3;
+	args = alloca(len);
+	strcpy(args + 2, name);
+
+	args[0] = flags & 0xBF; /* Maskoff LOCAL flag */
+	args[1] = 0;		/* Not used now */
+
+	/*
+	 * VG locks are just that: locks, and have no side effects
+	 * so we only need to do them on the local node because all
+	 * locks are cluster-wide.
+	 * Also, if the lock is exclusive it makes no sense to try to 
+	 * acquire it on all nodes, so just do that on the local node too.
+	 */
+	if (cmd == CLVMD_CMD_LOCK_VG ||
+	    (flags & LCK_TYPE_MASK) == LCK_EXCL ||
+	    (flags & LCK_LOCAL))
+		node = ".";
+
+	status = _cluster_request(cmd, node, args, len,
+				  &response, &num_responses);
+
+	/* If any nodes were down then display them and return an error */
+	for (i = 0; i < num_responses; i++) {
+		if (response[i].status == -EHOSTDOWN) {
+			log_error("clvmd not running on node %s",
+				  response[i].node);
+			status = 0;
+		} else if (response[i].status) {
+			log_error("Error locking on node %s: %s",
+				  response[i].node,
+				  response[i].response[0] ?
+				  	response[i].response :
+				  	strerror(response[i].status));
+			status = 0;
+		}
+	}
+
+	saved_errno = errno;
+	_cluster_free_request(response);
+	errno = saved_errno;
+
+	return status;
+}
+
+/* API entry point for LVM */
+#ifdef CLUSTER_LOCKING_INTERNAL
+static int _lock_resource(struct cmd_context *cmd, const char *resource,
+			  int flags)
+#else
+int lock_resource(struct cmd_context *cmd, const char *resource, int flags)
+#endif
+{
+	char lockname[PATH_MAX];
+	int cluster_cmd = 0;
+
+	assert(strlen(resource) < sizeof(lockname));
+
+	switch (flags & LCK_SCOPE_MASK) {
+	case LCK_VG:
+		/* If the VG name is empty then lock the unused PVs */
+		if (!resource || !*resource)
+			lvm_snprintf(lockname, sizeof(lockname), "P_orphans");
+		else
+			lvm_snprintf(lockname, sizeof(lockname), "V_%s",
+				     resource);
+
+		cluster_cmd = CLVMD_CMD_LOCK_VG;
+		flags &= LCK_TYPE_MASK;
+		break;
+
+	case LCK_LV:
+		cluster_cmd = CLVMD_CMD_LOCK_LV;
+		strcpy(lockname, resource);
+		flags &= 0xffdf;	/* Mask off HOLD flag */
+		break;
+
+	default:
+		log_error("Unrecognised lock scope: %d",
+			  flags & LCK_SCOPE_MASK);
+		return 0;
+	}
+
+	/* Send a message to the cluster manager */
+	log_very_verbose("Locking %s at 0x%x", lockname, flags);
+
+	return _lock_for_cluster(cluster_cmd, flags, lockname);
+}
+
+#ifdef CLUSTER_LOCKING_INTERNAL
+static void _locking_end(void)
+#else
+void locking_end(void)
+#endif
+{
+	if (_clvmd_sock != -1 && close(_clvmd_sock))
+		stack;
+
+	_clvmd_sock = -1;
+}
+
+#ifdef CLUSTER_LOCKING_INTERNAL
+static void _reset_locking(void)
+#else
+void reset_locking(void)
+#endif
+{
+	if (close(_clvmd_sock))
+		stack;
+
+	_clvmd_sock = _open_local_sock();
+	if (_clvmd_sock == -1)
+	        stack;
+}
+
+#ifdef CLUSTER_LOCKING_INTERNAL
+int init_cluster_locking(struct locking_type *locking, struct config_tree *cft)
+{
+	locking->lock_resource = _lock_resource;
+	locking->fin_locking = _locking_end;
+	locking->reset_locking = _reset_locking;
+	locking->flags = LCK_PRE_MEMLOCK;
+
+	_clvmd_sock = _open_local_sock();
+	if (_clvmd_sock == -1)
+		return 0;
+
+	return 1;
+}
+#else
+int locking_init(int type, struct config_tree *cf, uint32_t *flags)
+{
+	_clvmd_sock = _open_local_sock();
+	if (_clvmd_sock == -1)
+		return 0;
+
+	/* Ask LVM to lock memory before calling us */
+	*flags |= LCK_PRE_MEMLOCK;
+
+	return 1;
+}
+#endif

lib/locking/locking.c

@@ -145,6 +145,14 @@ int init_locking(int type, struct config_tree *cft)
 		return 1;
 #endif
 
+#ifdef CLUSTER_LOCKING_INTERNAL
+	case 3:
+		if (!init_cluster_locking(&_locking, cft))
+			break;
+		log_very_verbose("Cluster locking enabled.");
+		return 1;
+#endif
+
 	default:
 		log_error("Unknown locking type requested.");
 		return 0;

lib/locking/locking_types.h

@@ -40,3 +40,4 @@ int init_no_locking(struct locking_type *locking, struct config_tree *cf);
 int init_file_locking(struct locking_type *locking, struct config_tree *cf);
 
 int init_external_locking(struct locking_type *locking, struct config_tree *cf);
+int init_cluster_locking(struct locking_type *locking, struct config_tree *cf);

lib/log/log.c

@@ -28,6 +28,7 @@ static struct str_list _log_dev_alias;
 static int _verbose_level = VERBOSE_BASE_LEVEL;
 static int _test = 0;
 static int _partial = 0;
+static int _md_filtering = 0;
 static int _pvmove = 0;
 static int _debug_level = 0;
 static int _syslog = 0;
@@ -138,6 +139,11 @@ void init_partial(int level)
 	_partial = level;
 }
 
+void init_md_filtering(int level)
+{
+	_md_filtering = level;
+}
+
 void init_pvmove(int level)
 {
 	_pvmove = level;
@@ -187,6 +193,11 @@ int partial_mode()
 	return _partial;
 }
 
+int md_filtering()
+{
+	return _md_filtering;
+}
+
 int pvmove_mode()
 {
 	return _pvmove;
@@ -215,7 +226,7 @@ int debug_level()
 void print_log(int level, const char *file, int line, const char *format, ...)
 {
 	va_list ap;
-	char buf[1024], buf2[4096];
+	char buf[1024], buf2[4096], locn[4096];
 	int bufused, n;
 	const char *message;
 	const char *trformat;		/* Translated format string */
@@ -243,36 +254,46 @@ void print_log(int level, const char *file, int line, const char *format, ...)
 
       log_it:
 	if (!_log_suppress) {
+		if (_verbose_level > _LOG_DEBUG)
+			lvm_snprintf(locn, sizeof(locn), "#%s:%d ",
+				     file, line);
+		else
+			locn[0] = '\0';
+
 		va_start(ap, format);
 		switch (level) {
 		case _LOG_DEBUG:
-			if (!strcmp("<backtrace>", format))
+			if (!strcmp("<backtrace>", format) &&
+			    _verbose_level <= _LOG_DEBUG)
 				break;
 			if (_verbose_level >= _LOG_DEBUG) {
-				printf("%s%s", _cmd_name, _msg_prefix);
+				fprintf(stderr, "%s%s%s", locn, _cmd_name,
+					_msg_prefix);
 				if (_indent)
-					printf("      ");
-				vprintf(trformat, ap);
-				putchar('\n');
+					fprintf(stderr, "      ");
+				vfprintf(stderr, trformat, ap);
+				fputc('\n', stderr);
 			}
 			break;
 
 		case _LOG_INFO:
 			if (_verbose_level >= _LOG_INFO) {
-				printf("%s%s", _cmd_name, _msg_prefix);
+				fprintf(stderr, "%s%s%s", locn, _cmd_name,
+					_msg_prefix);
 				if (_indent)
-					printf("    ");
-				vprintf(trformat, ap);
-				putchar('\n');
+					fprintf(stderr, "    ");
+				vfprintf(stderr, trformat, ap);
+				fputc('\n', stderr);
 			}
 			break;
 		case _LOG_NOTICE:
 			if (_verbose_level >= _LOG_NOTICE) {
-				printf("%s%s", _cmd_name, _msg_prefix);
+				fprintf(stderr, "%s%s%s", locn, _cmd_name,
+					_msg_prefix);
 				if (_indent)
-					printf("  ");
-				vprintf(trformat, ap);
-				putchar('\n');
+					fprintf(stderr, "  ");
+				vfprintf(stderr, trformat, ap);
+				fputc('\n', stderr);
 			}
 			break;
 		case _LOG_WARN:
@@ -284,7 +305,8 @@ void print_log(int level, const char *file, int line, const char *format, ...)
 			break;
 		case _LOG_ERR:
 			if (_verbose_level >= _LOG_ERR) {
-				fprintf(stderr, "%s%s", _cmd_name, _msg_prefix);
+				fprintf(stderr, "%s%s%s", locn, _cmd_name,
+					_msg_prefix);
 				vfprintf(stderr, trformat, ap);
 				fputc('\n', stderr);
 			}
@@ -292,7 +314,8 @@ void print_log(int level, const char *file, int line, const char *format, ...)
 		case _LOG_FATAL:
 		default:
 			if (_verbose_level >= _LOG_FATAL) {
-				fprintf(stderr, "%s%s", _cmd_name, _msg_prefix);
+				fprintf(stderr, "%s%s%s", locn, _cmd_name,
+					_msg_prefix);
 				vfprintf(stderr, trformat, ap);
 				fputc('\n', stderr);
 			}

lib/log/log.h

@@ -63,6 +63,7 @@ void fin_syslog(void);
 void init_verbose(int level);
 void init_test(int level);
 void init_partial(int level);
+void init_md_filtering(int level);
 void init_pvmove(int level);
 void init_debug(int level);
 void init_cmd_name(int status);
@@ -75,6 +76,7 @@ void set_cmd_name(const char *cmd_name);
 
 int test_mode(void);
 int partial_mode(void);
+int md_filtering(void);
 int pvmove_mode(void);
 int debug_level(void);
 int ignorelockingfailure(void);

lib/metadata/lv_manip.c

@@ -21,7 +21,7 @@
 #include "toolcontext.h"
 #include "lv_alloc.h"
 #include "display.h"
-#include "segtypes.h"
+#include "segtype.h"
 
 /*
  * These functions adjust the pe counts in pv's
@@ -458,7 +458,7 @@ static int _allocate(struct volume_group *vg, struct logical_volume *lv,
 	if (segtype->flags & SEG_VIRTUAL)
 		return _alloc_virtual(lv, allocated, segtype);
 
-	if (!(scratch = pool_create(1024))) {
+	if (!(scratch = pool_create("allocation", 1024))) {
 		stack;
 		return 0;
 	}
@@ -623,6 +623,16 @@ int lv_extend(struct format_instance *fid,
 	lv->le_count += extents;
 	lv->size += (uint64_t) extents *lv->vg->extent_size;
 
+	if (fid->fmt->ops->segtype_supported &&
+	    !fid->fmt->ops->segtype_supported(fid, segtype)) {
+		log_error("Metadata format (%s) does not support required "
+			  "LV segment type (%s).", fid->fmt->name,
+			  segtype->name);
+		log_error("Consider changing the metadata format by running "
+			  "vgconvert.");
+		return 0;
+	}
+
 	if (!_allocate(lv->vg, lv, allocatable_pvs, old_le_count, alloc,
 		       segtype, stripes, stripe_size, mirrors, mirrored_pv,
 		       mirrored_pe, status)) {

lib/metadata/merge.c

@@ -18,7 +18,7 @@
 #include "toolcontext.h"
 #include "lv_alloc.h"
 #include "str_list.h"
-#include "segtypes.h"
+#include "segtype.h"
 
 /*
  * Attempt to merge two adjacent segments.
@@ -39,6 +39,9 @@ int lv_merge_segments(struct logical_volume *lv)
 	struct list *segh, *t;
 	struct lv_segment *current, *prev = NULL;
 
+	if (lv->status & LOCKED || lv->status & PVMOVE)
+		return 1;
+
 	list_iterate_safe(segh, t, &lv->segments) {
 		current = list_item(segh, struct lv_segment);
 
@@ -77,6 +80,7 @@ int lv_check_segments(struct logical_volume *lv)
 
 /*
  * Split the supplied segment at the supplied logical extent
+ * NB Use LE numbering that works across stripes PV1: 0,2,4 PV2: 1,3,5 etc.
  */
 static int _lv_split_segment(struct logical_volume *lv, struct lv_segment *seg,
 			     uint32_t le)
@@ -85,6 +89,7 @@ static int _lv_split_segment(struct logical_volume *lv, struct lv_segment *seg,
 	struct lv_segment *split_seg;
 	uint32_t s;
 	uint32_t offset = le - seg->le;
+	uint32_t area_offset;
 
 	if (!(seg->segtype->flags & SEG_CAN_SPLIT)) {
 		log_error("Unable to split the %s segment at LE %" PRIu32
@@ -107,8 +112,9 @@ static int _lv_split_segment(struct logical_volume *lv, struct lv_segment *seg,
 	}
 
 	/* In case of a striped segment, the offset has to be / stripes */
+	area_offset = offset;
 	if (seg->segtype->flags & SEG_AREAS_STRIPED)
-		offset /= seg->area_count;
+		area_offset /= seg->area_count;
 
 	/* Adjust the PV mapping */
 	for (s = 0; s < seg->area_count; s++) {
@@ -116,12 +122,19 @@ static int _lv_split_segment(struct logical_volume *lv, struct lv_segment *seg,
 		switch (seg->area[s].type) {
 		case AREA_LV:
 			split_seg->area[s].u.lv.le =
-			    seg->area[s].u.lv.le + offset;
+			    seg->area[s].u.lv.le + area_offset;
+			log_debug("Split %s:%u[%u] at %u: %s LE %u", lv->name,
+				  seg->le, s, le, seg->area[s].u.lv.lv->name,
+				  split_seg->area[s].u.lv.le);
 			break;
 
 		case AREA_PV:
 			split_seg->area[s].u.pv.pe =
-			    seg->area[s].u.pv.pe + offset;
+			    seg->area[s].u.pv.pe + area_offset;
+			log_debug("Split %s:%u[%u] at %u: %s PE %u", lv->name,
+				  seg->le, s, le,
+				  dev_name(seg->area[s].u.pv.pv->dev),
+				  split_seg->area[s].u.pv.pe);
 			break;
 
 		default:
@@ -131,8 +144,13 @@ static int _lv_split_segment(struct logical_volume *lv, struct lv_segment *seg,
 		}
 	}
 
-	split_seg->area_len = seg->area_len - offset;
-	seg->area_len = offset;
+	split_seg->area_len -= area_offset;
+	seg->area_len = area_offset;
+
+	split_seg->len -= offset;
+	seg->len = offset;
+
+	split_seg->le = seg->le + seg->len;
 
 	/* Add split off segment to the list _after_ the original one */
 	list_add_h(&seg->list, &split_seg->list);

lib/metadata/metadata.h

@@ -265,7 +265,7 @@ struct name_list {
 	char *name;
 };
 
-struct alloc_area {
+struct pe_range {
 	struct list list;
 	uint32_t start;		/* PEs */
 	uint32_t count;		/* PEs */
@@ -275,7 +275,7 @@ struct pv_list {
 	struct list list;
 	struct physical_volume *pv;
 	struct list *mdas;	/* Metadata areas */
-	struct list *alloc_areas;	/* Areas we may allocate from */
+	struct list *pe_ranges;	/* Ranges of PEs e.g. for allocation */
 };
 
 struct lv_list {
@@ -341,6 +341,12 @@ struct format_handler {
 	 */
 	int (*vg_setup) (struct format_instance * fi, struct volume_group * vg);
 
+	/*
+	 * Check whether particular segment type is supported.
+	 */
+	int (*segtype_supported) (struct format_instance *fid,
+				  struct segment_type *segtype);
+
 	/*
 	 * Create format instance with a particular metadata area
 	 */
@@ -501,7 +507,7 @@ int vg_remove_snapshot(struct volume_group *vg, struct logical_volume *cow);
  */
 int insert_pvmove_mirrors(struct cmd_context *cmd,
 			  struct logical_volume *lv_mirr,
-			  struct physical_volume *pv,
+			  struct list *source_pvl,
 			  struct logical_volume *lv,
 			  struct list *allocatable_pvs,
 			  struct list *lvs_changed);

lib/metadata/mirror.c

@@ -16,7 +16,7 @@
 #include "lib.h"
 #include "metadata.h"
 #include "toolcontext.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "display.h"
 #include "activate.h"
 
@@ -26,7 +26,7 @@
  */
 int insert_pvmove_mirrors(struct cmd_context *cmd,
 			  struct logical_volume *lv_mirr,
-			  struct physical_volume *pv,
+			  struct list *source_pvl,
 			  struct logical_volume *lv,
 			  struct list *allocatable_pvs,
 			  struct list *lvs_changed)
@@ -34,9 +34,15 @@ int insert_pvmove_mirrors(struct cmd_context *cmd,
 	struct list *segh;
 	struct lv_segment *seg;
 	struct lv_list *lvl;
+	struct pv_list *pvl;
 	int lv_used = 0;
 	uint32_t s, start_le, extent_count = 0u;
 	struct segment_type *segtype;
+	struct pe_range *per;
+	uint32_t pe_start, pe_end, per_end, stripe_multiplier;
+
+	/* Only 1 PV may feature in source_pvl */
+	pvl = list_item(source_pvl->n, struct pv_list);
 
 	if (!(segtype = get_segtype_from_string(lv->vg->cmd, "mirror"))) {
 		stack;
@@ -50,43 +56,113 @@ int insert_pvmove_mirrors(struct cmd_context *cmd,
                 return 0;
         }
 
+	/* Split LV segments to match PE ranges */
+	list_iterate(segh, &lv->segments) {
+		seg = list_item(segh, struct lv_segment);
+		for (s = 0; s < seg->area_count; s++) {
+			if (seg->area[s].type != AREA_PV ||
+			    seg->area[s].u.pv.pv->dev != pvl->pv->dev)
+				continue;
+
+			/* Do these PEs need moving? */
+			list_iterate_items(per, pvl->pe_ranges) {
+				pe_start = seg->area[s].u.pv.pe;
+				pe_end = pe_start + seg->area_len - 1;
+				per_end = per->start + per->count - 1;
+
+				/* No overlap? */
+				if ((pe_end < per->start) ||
+				    (pe_start > per_end))
+					continue;
+
+				if (seg->segtype->flags & SEG_AREAS_STRIPED)
+					stripe_multiplier = seg->area_count;
+				else
+					stripe_multiplier = 1;
+
+				if ((per->start != pe_start &&
+				     per->start > pe_start) &&
+				    !lv_split_segment(lv, seg->le +
+						      (per->start - pe_start) *
+						      stripe_multiplier)) {
+					stack;
+					return 0;
+				}
+
+				if ((per_end != pe_end &&
+				     per_end < pe_end) &&
+				    !lv_split_segment(lv, seg->le +
+						      (per_end - pe_start + 1) *
+						      stripe_multiplier)) {
+					stack;
+					return 0;
+				}
+			}
+		}
+	}
+
 	/* Work through all segments on the supplied PV */
 	list_iterate(segh, &lv->segments) {
 		seg = list_item(segh, struct lv_segment);
 		for (s = 0; s < seg->area_count; s++) {
 			if (seg->area[s].type != AREA_PV ||
-			    seg->area[s].u.pv.pv->dev != pv->dev)
+			    seg->area[s].u.pv.pv->dev != pvl->pv->dev)
 				continue;
 
-			/* First time, add LV to list of LVs affected */
-			if (!lv_used) {
-				if (!(lvl = pool_alloc(cmd->mem, sizeof(*lvl)))) {
-					log_error("lv_list alloc failed");
+			pe_start = seg->area[s].u.pv.pe;
+
+			/* Do these PEs need moving? */
+			list_iterate_items(per, pvl->pe_ranges) {
+				per_end = per->start + per->count - 1;
+
+				if ((pe_start < per->start) ||
+				    (pe_start > per_end))
+					continue;
+
+				log_debug("Matched PE range %u-%u against "
+					  "%s %u len %u", per->start, per_end,
+					  dev_name(seg->area[s].u.pv.pv->dev),
+					  seg->area[s].u.pv.pe, seg->area_len);
+
+				/* First time, add LV to list of LVs affected */
+				if (!lv_used) {
+					if (!(lvl = pool_alloc(cmd->mem, sizeof(*lvl)))) {
+						log_error("lv_list alloc failed");
+						return 0;
+					}
+					lvl->lv = lv;
+					list_add(lvs_changed, &lvl->list);
+					lv_used = 1;
+				}
+	
+				log_very_verbose("Moving %s:%u-%u of %s/%s",
+						 dev_name(pvl->pv->dev),
+						 seg->area[s].u.pv.pe,
+						 seg->area[s].u.pv.pe +
+						     seg->area_len - 1,
+						 lv->vg->name, lv->name);
+
+				start_le = lv_mirr->le_count;
+				if (!lv_extend(lv->vg->fid, lv_mirr, segtype, 1,
+				       	seg->area_len, 0u, seg->area_len,
+				       	seg->area[s].u.pv.pv,
+				       	seg->area[s].u.pv.pe,
+				       	PVMOVE, allocatable_pvs,
+				       	lv->alloc)) {
+					log_error("Unable to allocate "
+						  "temporary LV for pvmove.");
 					return 0;
 				}
-				lvl->lv = lv;
-				list_add(lvs_changed, &lvl->list);
-				lv_used = 1;
+				seg->area[s].type = AREA_LV;
+				seg->area[s].u.lv.lv = lv_mirr;
+				seg->area[s].u.lv.le = start_le;
+	
+				extent_count += seg->area_len;
+	
+				lv->status |= LOCKED;
+
+				break;
 			}
-
-			start_le = lv_mirr->le_count;
-			if (!lv_extend(lv->vg->fid, lv_mirr, segtype, 1,
-				       seg->area_len, 0u, seg->area_len,
-				       seg->area[s].u.pv.pv,
-				       seg->area[s].u.pv.pe,
-				       PVMOVE, allocatable_pvs,
-				       lv->alloc)) {
-				log_error("Allocation for temporary "
-					  "pvmove LV failed");
-				return 0;
-			}
-			seg->area[s].type = AREA_LV;
-			seg->area[s].u.lv.lv = lv_mirr;
-			seg->area[s].u.lv.le = start_le;
-
-			extent_count += seg->area_len;
-
-			lv->status |= LOCKED;
 		}
 	}
 
@@ -167,8 +243,12 @@ int remove_pvmove_mirrors(struct volume_group *vg,
 				lv1->status &= ~LOCKED;
 			}
 		}
+		if (!lv_merge_segments(lv1))
+			stack;
+
 	}
 
+
 	return 1;
 }
 

lib/metadata/pv_map.c

@@ -206,13 +206,13 @@ static int _create_areas(struct pool *mem, struct pv_map *pvm, uint32_t start,
 static int _create_allocatable_areas(struct pool *mem, struct pv_map *pvm)
 {
 	struct list *alloc_areas, *aah;
-	struct alloc_area *aa;
+	struct pe_range *aa;
 
-	alloc_areas = pvm->pvl->alloc_areas;
+	alloc_areas = pvm->pvl->pe_ranges;
 
 	if (alloc_areas) {
 		list_iterate(aah, alloc_areas) {
-			aa = list_item(aah, struct alloc_area);
+			aa = list_item(aah, struct pe_range);
 			if (!_create_areas(mem, pvm, aa->start, aa->count)) {
 				stack;
 				return 0;

lib/metadata/segtype.c

@@ -15,7 +15,7 @@
 
 #include "lib.h"
 #include "toolcontext.h"
-#include "segtypes.h"
+#include "segtype.h"
 
 struct segment_type *get_segtype_from_string(struct cmd_context *cmd,
 					     const char *str)

lib/metadata/segtype.h

@@ -85,4 +85,8 @@ struct segment_type *init_snapshot_segtype(struct cmd_context *cmd);
 struct segment_type *init_mirrored_segtype(struct cmd_context *cmd);
 #endif
 
+#ifdef CRYPT_INTERNAL
+struct segment_type *init_crypt_segtype(struct cmd_context *cmd);
+#endif
+
 #endif

lib/mirror/Makefile.in

@@ -22,8 +22,6 @@ LIB_SHARED = liblvm2mirror.so
 
 include $(top_srcdir)/make.tmpl
 
-.PHONY: install
-
 install: liblvm2mirror.so
 	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) $< \
 		$(libdir)/liblvm2mirror.so.$(LIB_VERSION)

lib/mirror/mirrored.c

@@ -18,7 +18,7 @@
 #include "list.h"
 #include "toolcontext.h"
 #include "metadata.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "display.h"
 #include "text_export.h"
 #include "text_import.h"
@@ -95,7 +95,7 @@ static int _text_export(const struct lv_segment *seg, struct formatter *f)
 {
 	outf(f, "mirror_count = %u", seg->area_count);
 	if (seg->status & PVMOVE)
-		out_size(f, (uint64_t) seg->extents_copied,
+		out_size(f, (uint64_t) seg->extents_copied * seg->lv->vg->extent_size,
 			 "extents_moved = %u", seg->extents_copied);
 
 	return out_areas(f, seg, "mirror");
@@ -130,6 +130,7 @@ static int _compose_target_line(struct dev_manager *dm, struct pool *mem,
 	int mirror_status = MIRR_RUNNING;
 	int areas = seg->area_count;
 	int start_area = 0u;
+	uint32_t region_size, region_max;
 
 	if (!*target_state)
 		*target_state = _init_target(mem, cft);
@@ -143,7 +144,7 @@ static int _compose_target_line(struct dev_manager *dm, struct pool *mem,
 		if (seg->extents_copied == seg->area_len) {
 			mirror_status = MIRR_COMPLETED;
 			start_area = 1;
-		} else if (*pvmove_mirror_count++) {
+		} else if ((*pvmove_mirror_count)++) {
 			mirror_status = MIRR_DISABLED;
 			areas = 1;
 		}
@@ -153,8 +154,20 @@ static int _compose_target_line(struct dev_manager *dm, struct pool *mem,
 		*target = "linear";
 	} else {
 		*target = "mirror";
+
+		/* Find largest power of 2 region size unit we can use */
+		region_max = (1 << (ffs(seg->area_len) - 1)) *
+			      seg->lv->vg->extent_size;
+
+		region_size = mirr_state->region_size;
+		if (region_max < region_size) {
+			region_size = region_max;
+			log_verbose("Using reduced mirror region size of %u sectors",
+				    region_size);
+		}
+
 		if ((*pos = lvm_snprintf(params, paramsize, "core 1 %u %u ",
-					 mirr_state->region_size, areas)) < 0) {
+					 region_size, areas)) < 0) {
 			stack;
 			return -1;
 		}
@@ -162,7 +175,6 @@ static int _compose_target_line(struct dev_manager *dm, struct pool *mem,
 
 	return compose_areas_line(dm, seg, params, paramsize, pos, start_area,
 				  areas);
-
 }
 
 static int _target_percent(void **target_state, struct pool *mem,
@@ -188,8 +200,7 @@ static int _target_percent(void **target_state, struct pool *mem,
 	*total_denominator += denominator;
 
 	if (seg)
-		seg->extents_copied = mirr_state->region_size *
-		    numerator / seg->lv->vg->extent_size;
+		seg->extents_copied = seg->area_len * numerator / denominator;
 
 	return 1;
 }
@@ -247,5 +258,7 @@ struct segment_type *init_segtype(struct cmd_context *cmd)
 	segtype->private = NULL;
 	segtype->flags = SEG_AREAS_MIRRORED;
 
+	log_very_verbose("Initialised segtype: %s", segtype->name);
+
 	return segtype;
 }

lib/misc/lvm-string.c

@@ -16,6 +16,7 @@
 #include "lib.h"
 #include "lvm-types.h"
 #include "lvm-string.h"
+#include "pool.h"
 
 #include <ctype.h>
 
@@ -98,3 +99,109 @@ int split_words(char *buffer, unsigned max, char **argv)
 
 	return arg;
 }
+
+/*
+ * Device layer names are all of the form <vg>-<lv>-<layer>, any
+ * other hyphens that appear in these names are quoted with yet
+ * another hyphen.  The top layer of any device has no layer
+ * name.  eg, vg0-lvol0.
+ */
+static void _count_hyphens(const char *str, size_t *len, int *hyphens)
+{
+	const char *ptr;
+
+	for (ptr = str; *ptr; ptr++, (*len)++)
+		if (*ptr == '-')
+			(*hyphens)++;
+}
+
+/*
+ * Copies a string, quoting hyphens with hyphens.
+ */
+static void _quote_hyphens(char **out, const char *src)
+{
+	while (*src) {
+		if (*src == '-')
+			*(*out)++ = '-';
+
+		*(*out)++ = *src++;
+	}
+}
+
+/*
+ * <vg>-<lv>-<layer> or if !layer just <vg>-<lv>.
+ */
+char *build_dm_name(struct pool *mem, const char *vg,
+		    const char *lv, const char *layer)
+{
+	size_t len = 1;
+	int hyphens = 1;
+	char *r, *out;
+
+	_count_hyphens(vg, &len, &hyphens);
+	_count_hyphens(lv, &len, &hyphens);
+
+	if (layer && *layer) {
+		_count_hyphens(layer, &len, &hyphens);
+		hyphens++;
+	}
+
+	len += hyphens;
+
+	if (!(r = pool_alloc(mem, len))) {
+		stack;
+		return NULL;
+	}
+
+	out = r;
+	_quote_hyphens(&out, vg);
+	*out++ = '-';
+	_quote_hyphens(&out, lv);
+
+	if (layer && *layer) {
+		*out++ = '-';
+		_quote_hyphens(&out, layer);
+	}
+	*out = '\0';
+
+	return r;
+}
+
+/*
+ * Remove hyphen quoting from a component of a name.
+ * NULL-terminates the component and returns start of next component.
+ */
+static char *_unquote(char *component)
+{
+	char *c = component;
+	char *o = c;
+
+	while (*c) {
+		if (*(c + 1)) {
+			if (*c == '-') {
+				if (*(c + 1) == '-')
+					c++;
+				else
+					break;
+			}
+		}
+		*o = *c;
+		o++;
+		c++;
+	}
+
+	*o = '\0';
+	return (c + 1);
+}
+
+int split_dm_name(struct pool *mem, const char *dmname,
+		  char **vgname, char **lvname, char **layer)
+{
+	if (!(*vgname = pool_strdup(mem, dmname)))
+		return 0;
+
+	_unquote(*layer = _unquote(*lvname = _unquote(*vgname)));
+
+	return 1;
+}
+

lib/misc/lvm-string.h

@@ -19,6 +19,8 @@
 #include <stdio.h>
 #include <stdarg.h>
 
+struct pool;
+
 /*
  * On error, up to glibc 2.0.6, snprintf returned -1 if buffer was too small;
  * From glibc 2.1 it returns number of chars (excl. trailing null) that would 
@@ -32,4 +34,10 @@ int emit_to_buffer(char **buffer, size_t *size, const char *fmt, ...);
 
 int split_words(char *buffer, unsigned max, char **argv);
 
+char *build_dm_name(struct pool *mem, const char *vg,
+                    const char *lv, const char *layer);
+
+int split_dm_name(struct pool *mem, const char *dmname,
+                  char **vgname, char **lvname, char **layer);
+
 #endif

lib/mm/dbg_malloc.c

@@ -19,6 +19,8 @@
 
 #include <stdarg.h>
 
+#ifdef DEBUG_MEM
+
 struct memblock {
 	struct memblock *prev, *next;	/* All allocated blocks are linked */
 	size_t length;		/* Size of the requested block */
@@ -29,11 +31,13 @@ struct memblock {
 };
 
 static struct {
-	unsigned int blocks, mblocks;
+	unsigned block_serialno;/* Non-decreasing serialno of block */
+	unsigned blocks_allocated; /* Current number of blocks allocated */
+	unsigned blocks_max;	/* Max no of concurrently-allocated blocks */
 	unsigned int bytes, mbytes;
 
 } _mem_stats = {
-0, 0, 0, 0};
+0, 0, 0, 0, 0};
 
 static struct memblock *_head = 0;
 static struct memblock *_tail = 0;
@@ -45,7 +49,7 @@ void *malloc_aux(size_t s, const char *file, int line)
 
 	if (s > 50000000) {
 		log_error("Huge memory allocation (size %" PRIuPTR
-			  ") rejected - bug?", s);
+			  ") rejected - metadata corruption?", s);
 		return 0;
 	}
 
@@ -65,7 +69,7 @@ void *malloc_aux(size_t s, const char *file, int line)
 	/* setup fields */
 	nb->magic = nb + 1;
 	nb->length = s;
-	nb->id = ++_mem_stats.blocks;
+	nb->id = ++_mem_stats.block_serialno;
 	nb->next = 0;
 	nb->prev = _tail;
 
@@ -89,8 +93,9 @@ void *malloc_aux(size_t s, const char *file, int line)
 			*ptr++ = (char) nb->id;
 	}
 
-	if (_mem_stats.blocks > _mem_stats.mblocks)
-		_mem_stats.mblocks = _mem_stats.blocks;
+	_mem_stats.blocks_allocated++;
+	if (_mem_stats.blocks_allocated > _mem_stats.blocks_max)
+		_mem_stats.blocks_max = _mem_stats.blocks_allocated;
 
 	_mem_stats.bytes += s;
 	if (_mem_stats.bytes > _mem_stats.mbytes)
@@ -140,8 +145,8 @@ void free_aux(void *p)
 	else
 		_tail = mb->prev;
 
-	assert(_mem_stats.blocks);
-	_mem_stats.blocks--;
+	assert(_mem_stats.blocks_allocated);
+	_mem_stats.blocks_allocated--;
 	_mem_stats.bytes -= mb->length;
 
 	/* free the memory */
@@ -163,7 +168,6 @@ void *realloc_aux(void *p, unsigned int s, const char *file, int line)
 	return r;
 }
 
-#ifdef DEBUG_MEM
 int dump_memory(void)
 {
 	unsigned long tot = 0;
@@ -210,10 +214,18 @@ void bounds_check(void)
 		mb = mb->next;
 	}
 }
-#endif
 
-/*
- * Local variables:
- * c-file-style: "linux"
- * End:
- */
+#else
+
+void *malloc_aux(size_t s, const char *file, int line)
+{
+	if (s > 50000000) {
+		log_error("Huge memory allocation (size %" PRIuPTR
+			  ") rejected - metadata corruption?", s);
+		return 0;
+	}
+
+	return malloc(s);
+}
+
+#endif

lib/mm/dbg_malloc.h

@@ -20,22 +20,26 @@
 #include <stdlib.h>
 #include <string.h>
 
-#ifdef DEBUG_MEM
 void *malloc_aux(size_t s, const char *file, int line);
+#  define dbg_malloc(s) malloc_aux((s), __FILE__, __LINE__)
+
+#ifdef DEBUG_MEM
+
 void free_aux(void *p);
 void *realloc_aux(void *p, unsigned int s, const char *file, int line);
 int dump_memory(void);
 void bounds_check(void);
 
-#  define dbg_malloc(s) malloc_aux((s), __FILE__, __LINE__)
 #  define dbg_free(p) free_aux(p)
 #  define dbg_realloc(p, s) realloc_aux(p, s, __FILE__, __LINE__)
+
 #else
-#  define dbg_malloc(s) malloc(s)
+
 #  define dbg_free(p) free(p)
 #  define dbg_realloc(p, s) realloc(p, s)
 #  define dump_memory()
 #  define bounds_check()
+
 #endif
 
 static inline char *dbg_strdup(const char *str)

lib/mm/pool-debug.c

@@ -22,38 +22,64 @@ struct block {
 	void *data;
 };
 
+typedef struct {
+	unsigned block_serialno;	/* Non-decreasing serialno of block */
+	unsigned blocks_allocated;	/* Current number of blocks allocated */
+	unsigned blocks_max;	/* Max no of concurrently-allocated blocks */
+	unsigned int bytes, maxbytes;
+} pool_stats;
+
 struct pool {
+	const char *name;
+
 	int begun;
 	struct block *object;
 
 	struct block *blocks;
 	struct block *tail;
+
+	pool_stats stats;
 };
 
 /* by default things come out aligned for doubles */
 #define DEFAULT_ALIGNMENT __alignof__ (double)
 
-struct pool *pool_create(size_t chunk_hint)
+struct pool *pool_create(const char *name, size_t chunk_hint)
 {
 	struct pool *mem = dbg_malloc(sizeof(*mem));
 
 	if (!mem) {
-		log_error("Couldn't create memory pool (size %" PRIsize_t ")",
-			  sizeof(*mem));
+		log_error("Couldn't create memory pool %s (size %"
+			  PRIsize_t ")", name, sizeof(*mem));
 		return NULL;
 	}
 
+	mem->name = name;
 	mem->begun = 0;
 	mem->object = 0;
 	mem->blocks = mem->tail = NULL;
+
+	mem->stats.block_serialno = 0;
+	mem->stats.blocks_allocated = 0;
+	mem->stats.blocks_max = 0;
+	mem->stats.bytes = 0;
+	mem->stats.maxbytes = 0;
+
+#ifdef DEBUG_POOL
+	log_debug("Created mempool %s", name);
+#endif
+
 	return mem;
 }
 
-static void _free_blocks(struct block *b)
+static void _free_blocks(struct pool *p, struct block *b)
 {
 	struct block *n;
 
 	while (b) {
+		p->stats.bytes -= b->size;
+		p->stats.blocks_allocated--;
+
 		n = b->next;
 		dbg_free(b->data);
 		dbg_free(b);
@@ -61,9 +87,22 @@ static void _free_blocks(struct block *b)
 	}
 }
 
+static void _pool_stats(struct pool *p, const char *action)
+{
+#ifdef DEBUG_POOL
+	log_debug("%s mempool %s: %u/%u bytes, %u/%u blocks, "
+		  "%u allocations)", action, p->name, p->stats.bytes,
+		  p->stats.maxbytes, p->stats.blocks_allocated,
+		  p->stats.blocks_max, p->stats.block_serialno);
+#else
+	;
+#endif
+}
+
 void pool_destroy(struct pool *p)
 {
-	_free_blocks(p->blocks);
+	_pool_stats(p, "Destroying");
+	_free_blocks(p, p->blocks);
 	dbg_free(p);
 }
 
@@ -79,11 +118,20 @@ static void _append_block(struct pool *p, struct block *b)
 		p->tail = b;
 	} else
 		p->blocks = p->tail = b;
+
+	p->stats.block_serialno++;
+	p->stats.blocks_allocated++;
+	if (p->stats.blocks_allocated > p->stats.blocks_max)
+		p->stats.blocks_max = p->stats.blocks_allocated;
+
+	p->stats.bytes += b->size;
+	if (p->stats.bytes > p->stats.maxbytes)
+		p->stats.maxbytes = p->stats.bytes;
 }
 
 static struct block *_new_block(size_t s, unsigned alignment)
 {
-	static char *_oom = "Out of memory";
+	static const char *_oom = "Out of memory";
 
 	/* FIXME: I'm currently ignoring the alignment arg. */
 	size_t len = sizeof(struct block) + s;
@@ -121,12 +169,14 @@ void *pool_alloc_aligned(struct pool *p, size_t s, unsigned alignment)
 		return NULL;
 
 	_append_block(p, b);
+
 	return b->data;
 }
 
 void pool_empty(struct pool *p)
 {
-	_free_blocks(p->blocks);
+	_pool_stats(p, "Emptying");
+	_free_blocks(p, p->blocks);
 	p->blocks = p->tail = NULL;
 }
 
@@ -134,6 +184,8 @@ void pool_free(struct pool *p, void *ptr)
 {
 	struct block *b, *prev = NULL;
 
+	_pool_stats(p, "Freeing (before)");
+
 	for (b = p->blocks; b; b = b->next) {
 		if (b->data == ptr)
 			break;
@@ -147,13 +199,15 @@ void pool_free(struct pool *p, void *ptr)
 	 */
 	assert(b);
 
-	_free_blocks(b);
+	_free_blocks(p, b);
 
 	if (prev) {
 		p->tail = prev;
 		prev->next = NULL;
 	} else
 		p->blocks = p->tail = NULL;
+
+	_pool_stats(p, "Freeing (after)");
 }
 
 int pool_begin_object(struct pool *p, size_t init_size)
@@ -180,10 +234,13 @@ int pool_grow_object(struct pool *p, const void *buffer, size_t delta)
 
 	if (p->object) {
 		memcpy(new->data, p->object->data, p->object->size);
+		dbg_free(p->object->data);
 		dbg_free(p->object);
 	}
 	p->object = new;
 
+	memcpy(new->data + size - delta, buffer, delta);
+
 	return 1;
 }
 

lib/mm/pool-fast.c

@@ -36,14 +36,14 @@ struct chunk *_new_chunk(struct pool *p, size_t s);
 /* by default things come out aligned for doubles */
 #define DEFAULT_ALIGNMENT __alignof__ (double)
 
-struct pool *pool_create(size_t chunk_hint)
+struct pool *pool_create(const char *name, size_t chunk_hint)
 {
 	size_t new_size = 1024;
 	struct pool *p = dbg_malloc(sizeof(*p));
 
 	if (!p) {
-		log_error("Couldn't create memory pool (size %" PRIsize_t ")",
-			  sizeof(*p));
+		log_error("Couldn't create memory pool %s (size %"
+			  PRIsize_t ")", name, sizeof(*p));
 		return 0;
 	}
 	memset(p, 0, sizeof(*p));
@@ -109,8 +109,8 @@ void pool_empty(struct pool *p)
 	for (c = p->chunk; c && c->prev; c = c->prev)
 		;
 
-	if (p->chunk)
-		pool_free(p, (char *) (p->chunk + 1));
+	if (c)
+		pool_free(p, (char *) (c + 1));
 }
 
 void pool_free(struct pool *p, void *ptr)

lib/mm/pool.h

@@ -55,7 +55,7 @@
 struct pool;
 
 /* constructor and destructor */
-struct pool *pool_create(size_t chunk_hint);
+struct pool *pool_create(const char *name, size_t chunk_hint);
 void pool_destroy(struct pool *p);
 
 /* simple allocation/free routines */

lib/mm/xlate.h

@@ -17,6 +17,41 @@
 #define _LVM_XLATE_H
 
 #ifdef linux
+#  include <endian.h>
+#  include <byteswap.h>
+#else
+#  include <machine/endian.h>
+#  define bswap_16(x) (((x) & 0x00ffU) << 8 | \
+		       ((x) & 0xff00U) >> 8)
+#  define bswap_32(x) (((x) & 0x000000ffU) << 24 | \
+		       ((x) & 0xff000000U) >> 24 | \
+		       ((x) & 0x0000ff00U) << 8  | \
+		       ((x) & 0x00ff0000U) >> 8)
+#  define bswap_64(x) (((x) & 0x00000000000000ffU) << 56 | \
+		       ((x) & 0xff00000000000000U) >> 56 | \
+		       ((x) & 0x000000000000ff00U) << 40 | \
+		       ((x) & 0x00ff000000000000U) >> 40 | \
+		       ((x) & 0x0000000000ff0000U) << 24 | \
+		       ((x) & 0x0000ff0000000000U) >> 24 | \
+		       ((x) & 0x00000000ff000000U) << 8 | \
+		       ((x) & 0x000000ff00000000U) >> 8)
+#endif
+
+#if BYTE_ORDER == LITTLE_ENDIAN
+#  define xlate16(x) (x)
+#  define xlate32(x) (x)
+#  define xlate64(x) (x)
+#  define xlate16_be(x) bswap_16(x)
+#  define xlate32_be(x) bswap_32(x)
+#  define xlate64_be(x) bswap_64(x)
+#elif BYTE_ORDER == BIG_ENDIAN
+#  define xlate16(x) bswap_16(x)
+#  define xlate32(x) bswap_32(x)
+#  define xlate64(x) bswap_64(x)
+#  define xlate16_be(x) (x)
+#  define xlate32_be(x) (x)
+#  define xlate64_be(x) (x)
+#else
 #  include <asm/byteorder.h>
 #  define xlate16(x) __cpu_to_le16((x))
 #  define xlate32(x) __cpu_to_le32((x))
@@ -24,41 +59,6 @@
 #  define xlate16_be(x) __cpu_to_be16((x))
 #  define xlate32_be(x) __cpu_to_be32((x))
 #  define xlate64_be(x) __cpu_to_be64((x))
-#else
-#  include <machine/endian.h>
-#  if !defined(BYTE_ORDER) || \
-      (BYTE_ORDER != BIG_ENDIAN && BYTE_ORDER != LITTLE_ENDIAN)
-#    error "Undefined or unrecognised BYTE_ORDER";
-#  endif
-#  define __xlate16(x) (((x) & 0x00ffU) << 8 | \
-			((x) & 0xff00U) >> 8)
-#  define __xlate32(x) (((x) & 0x000000ffU) << 24 | \
-			((x) & 0xff000000U) >> 24 | \
-			((x) & 0x0000ff00U) << 8  | \
-			((x) & 0x00ff0000U) >> 8)
-#  define __xlate64(x) (((x) & 0x00000000000000ffU) << 56 | \
-			((x) & 0xff00000000000000U) >> 56 | \
-			((x) & 0x000000000000ff00U) << 40 | \
-			((x) & 0x00ff000000000000U) >> 40 | \
-			((x) & 0x0000000000ff0000U) << 24 | \
-			((x) & 0x0000ff0000000000U) >> 24 | \
-			((x) & 0x00000000ff000000U) << 8 | \
-			((x) & 0x000000ff00000000U) >> 8)
-#  if BYTE_ORDER == LITTLE_ENDIAN
-#    define xlate16(x) (x)
-#    define xlate32(x) (x)
-#    define xlate64(x) (x)
-#    define xlate16_be(x) __xlate16(x)
-#    define xlate32_be(x) __xlate32(x)
-#    define xlate64_be(x) __xlate64(x)
-#  else
-#    define xlate16(x) __xlate16(x)
-#    define xlate32(x) __xlate32(x)
-#    define xlate64(x) __xlate64(x)
-#    define xlate16_be(x) (x)
-#    define xlate32_be(x) (x)
-#    define xlate64_be(x) (x)
-#  endif
 #endif
 
 #endif

lib/regex/matcher.c

@@ -275,7 +275,7 @@ struct matcher *matcher_create(struct pool *mem, const char **patterns,
 	int i;
 	size_t len = 0;
 	struct rx_node *rx;
-	struct pool *scratch = pool_create(10 * 1024);
+	struct pool *scratch = pool_create("regex matcher", 10 * 1024);
 	struct matcher *m;
 
 	if (!scratch) {

lib/report/columns.h

@@ -23,6 +23,8 @@ FIELD(LVS, lv, STR, "LV", name, 4, string, "lv_name")
 FIELD(LVS, lv, STR, "Attr", lvid, 4, lvstatus, "lv_attr")
 FIELD(LVS, lv, NUM, "Maj", major, 3, int32, "lv_major")
 FIELD(LVS, lv, NUM, "Min", minor, 3, int32, "lv_minor")
+FIELD(LVS, lv, STR, "KMaj", lvid, 4, lvkmaj, "lv_kernel_major")
+FIELD(LVS, lv, STR, "KMin", lvid, 4, lvkmin, "lv_kernel_minor")
 FIELD(LVS, lv, NUM, "LSize", size, 5, size64, "lv_size")
 FIELD(LVS, lv, NUM, "#Seg", lvid, 4, lvsegcount, "seg_count")
 FIELD(LVS, lv, STR, "Origin", lvid, 6, origin, "origin")
@@ -34,6 +36,7 @@ FIELD(LVS, lv, STR, "LV Tags", tags, 7, tags, "lv_tags")
 FIELD(PVS, pv, STR, "Fmt", id, 3, pvfmt, "pv_fmt")
 FIELD(PVS, pv, STR, "PV UUID", id, 38, uuid, "pv_uuid")
 FIELD(PVS, pv, NUM, "PSize", id, 5, pvsize, "pv_size")
+FIELD(PVS, pv, NUM, "DevSize", dev, 7, devsize, "dev_size")
 FIELD(PVS, pv, NUM, "PFree", id, 5, pvfree, "pv_free")
 FIELD(PVS, pv, NUM, "Used", id, 4, pvused, "pv_used")
 FIELD(PVS, pv, STR, "PV", dev, 10, dev_name, "pv_name")

lib/report/report.c

@@ -21,7 +21,7 @@
 #include "lvm-string.h"
 #include "display.h"
 #include "activate.h"
-#include "segtypes.h"
+#include "segtype.h"
 
 /* 
  * For macro use
@@ -259,6 +259,65 @@ static int _pvfmt_disp(struct report_handle *rh, struct field *field,
 	return _string_disp(rh, field, &pv->fmt->name);
 }
 
+static int _int_disp(struct report_handle *rh, struct field *field,
+		     const void *data)
+{
+	const int value = *(const int *) data;
+	uint64_t *sortval;
+	char *repstr;
+
+	if (!(repstr = pool_zalloc(rh->mem, 13))) {
+		log_error("pool_alloc failed");
+		return 0;
+	}
+
+	if (!(sortval = pool_alloc(rh->mem, sizeof(int64_t)))) {
+		log_error("pool_alloc failed");
+		return 0;
+	}
+
+	if (lvm_snprintf(repstr, 12, "%d", value) < 0) {
+		log_error("int too big: %d", value);
+		return 0;
+	}
+
+	*sortval = (const uint64_t) value;
+	field->sort_value = sortval;
+	field->report_string = repstr;
+
+	return 1;
+}
+
+static int _lvkmaj_disp(struct report_handle *rh, struct field *field,
+			const void *data)
+{
+	const struct logical_volume *lv = (const struct logical_volume *) data;
+	struct lvinfo info;
+	uint64_t minusone = UINT64_C(-1);
+
+	if (lv_info(lv, &info) && info.exists)
+		return _int_disp(rh, field, &info.major);
+	else
+		return _int_disp(rh, field, &minusone);
+
+	return 1;
+}
+
+static int _lvkmin_disp(struct report_handle *rh, struct field *field,
+			const void *data)
+{
+	const struct logical_volume *lv = (const struct logical_volume *) data;
+	struct lvinfo info;
+	uint64_t minusone = UINT64_C(-1);
+
+	if (lv_info(lv, &info) && info.exists)
+		return _int_disp(rh, field, &info.minor);
+	else
+		return _int_disp(rh, field, &minusone);
+
+	return 1;
+}
+
 static int _lvstatus_disp(struct report_handle *rh, struct field *field,
 			  const void *data)
 {
@@ -586,6 +645,18 @@ static int _pvsize_disp(struct report_handle *rh, struct field *field,
 	return _size64_disp(rh, field, &size);
 }
 
+static int _devsize_disp(struct report_handle *rh, struct field *field,
+			const void *data)
+{
+	const struct device *dev = *(const struct device **) data;
+	uint64_t size;
+
+	if (!dev_get_size(dev, &size))
+		size = 0;
+
+	return _size64_disp(rh, field, &size);
+}
+
 static int _vgfree_disp(struct report_handle *rh, struct field *field,
 			const void *data)
 {
@@ -1010,7 +1081,7 @@ void *report_init(struct cmd_context *cmd, const char *format, const char *keys,
 		rh->field_prefix = "";
 	}
 
-	if (!(rh->mem = pool_create(10 * 1024))) {
+	if (!(rh->mem = pool_create("report", 10 * 1024))) {
 		log_error("Allocation of memory pool for report failed");
 		return NULL;
 	}

lib/snapshot/Makefile.in

@@ -22,8 +22,6 @@ LIB_SHARED = liblvm2snapshot.so
 
 include $(top_srcdir)/make.tmpl
 
-.PHONY: install
-
 install: liblvm2snapshot.so
 	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) $< \
 		$(libdir)/liblvm2snapshot.so.$(LIB_VERSION)

lib/snapshot/snapshot.c

@@ -18,7 +18,7 @@
 #include "list.h"
 #include "toolcontext.h"
 #include "metadata.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "text_export.h"
 #include "config.h"
 #include "activate.h"
@@ -161,5 +161,7 @@ struct segment_type *init_segtype(struct cmd_context *cmd)
 	segtype->private = NULL;
 	segtype->flags = SEG_SNAPSHOT;
 
+	log_very_verbose("Initialised segtype: %s", segtype->name);
+
 	return segtype;
 }

lib/striped/striped.c

@@ -17,7 +17,7 @@
 #include "pool.h"
 #include "list.h"
 #include "toolcontext.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "display.h"
 #include "text_export.h"
 #include "text_import.h"
@@ -217,5 +217,7 @@ struct segment_type *init_striped_segtype(struct cmd_context *cmd)
 	segtype->flags =
 	    SEG_CAN_SPLIT | SEG_AREAS_STRIPED | SEG_FORMAT1_SUPPORT;
 
+	log_very_verbose("Initialised segtype: %s", segtype->name);
+
 	return segtype;
 }

lib/zero/zero.c

@@ -16,7 +16,7 @@
 #include "pool.h"
 #include "list.h"
 #include "toolcontext.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "display.h"
 #include "text_export.h"
 #include "text_import.h"
@@ -97,5 +97,7 @@ struct segment_type *init_zero_segtype(struct cmd_context *cmd)
 	segtype->private = NULL;
 	segtype->flags = SEG_CAN_SPLIT | SEG_VIRTUAL;
 
+	log_very_verbose("Initialised segtype: %s", segtype->name);
+
 	return segtype;
 }

libdm/.exported_symbols

@@ -20,6 +20,7 @@ dm_task_set_minor
 dm_task_set_sector
 dm_task_set_message
 dm_task_add_target
+dm_task_no_open_count
 dm_get_next_target
 dm_task_run
 dm_set_dev_dir

libdm/Makefile.in

@@ -1,6 +1,6 @@
 #
 # Copyright (C) 2001-2004 Sistina Software, Inc. All rights reserved.
-# Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+# Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
 #
 # This file is part of the device-mapper userspace tools.
 #
@@ -17,13 +17,16 @@ top_srcdir = @top_srcdir@
 VPATH = @srcdir@
 interface = @interface@
 
-SOURCES = libdm-common.c $(interface)/libdm-iface.c
+SOURCES = libdm-common.c libdm-file.c $(interface)/libdm-iface.c
 
 INCLUDES = -I$(interface)
 
 LIB_STATIC = $(interface)/libdevmapper.a
 LIB_SHARED = $(interface)/libdevmapper.so
 
+CFLAGS += -DDEVICE_UID=@DEVICE_UID@ -DDEVICE_GID=@DEVICE_GID@ \
+	  -DDEVICE_MODE=@DEVICE_MODE@
+
 include ../make.tmpl
 
 .PHONY: install_dynamic install_static \

libdm/ioctl/libdm-compat.h

@@ -16,9 +16,9 @@
 #ifndef _LINUX_LIBDM_COMPAT_H
 #define _LINUX_LIBDM_COMPAT_H
 
+#include "kdev_t.h"
 #include <inttypes.h>
 #include <linux/dm-ioctl.h>
-#include <linux/kdev_t.h>
 #include <sys/ioctl.h>
 
 struct dm_task;

libdm/ioctl/libdm-iface.c

@@ -1,6 +1,6 @@
 /*
  * Copyright (C) 2001-2004 Sistina Software, Inc. All rights reserved.
- * Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+ * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
  *
  * This file is part of the device-mapper userspace tools.
  *
@@ -13,28 +13,23 @@
  * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  */
 
+#include "lib.h"
 #include "libdm-targets.h"
 #include "libdm-common.h"
-#include "log.h"
+#include "libdm-file.h"
 
 #ifdef DM_COMPAT
 #  include "libdm-compat.h"
 #endif
 
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <unistd.h>
 #include <fcntl.h>
 #include <dirent.h>
-#include <errno.h>
-#include <sys/stat.h>
 #include <sys/ioctl.h>
 #include <limits.h>
 
 #ifdef linux
+#  include "kdev_t.h"
 #  include <linux/limits.h>
-#  include <linux/kdev_t.h>
 #  include <linux/dm-ioctl.h>
 #else
 #  define MAJOR(x) major((x))
@@ -53,6 +48,13 @@
 #error The version of dm-ioctl.h included is incompatible.
 #endif
 
+/* FIXME This should be exported in device-mapper.h */
+#define DM_NAME "device-mapper"
+
+#define PROC_MISC "/proc/misc"
+#define PROC_DEVICES "/proc/devices"
+#define MISC_NAME "misc"
+
 /* dm major version no for running kernel */
 static int _dm_version = DM_VERSION_MAJOR;
 static int _log_suppress = 0;
@@ -114,22 +116,143 @@ static void *_align(void *ptr, unsigned int a)
 	return (void *) (((unsigned long) ptr + agn) & ~agn);
 }
 
+static int _get_proc_number(const char *file, const char *name,
+			    uint32_t *number)
+{
+	FILE *fl;
+	char nm[256];
+	int c;
+
+	if (!(fl = fopen(file, "r"))) {
+		log_error("%s: fopen failed: %s", file, strerror(errno));
+		return 0;
+	}
+
+	while (!feof(fl)) {
+		if (fscanf(fl, "%d %255s\n", number, &nm[0]) == 2) {
+			if (!strcmp(name, nm)) {
+				fclose(fl);
+				return 1;
+			}
+		} else do {
+			c = fgetc(fl);
+		} while (c != EOF && c != '\n');
+	}
+	fclose(fl);
+
+	log_error("%s: No entry for %s found", file, name);
+	return 0;
+}
+
+static int _control_device_number(uint32_t *major, uint32_t *minor)
+{
+	if (!_get_proc_number(PROC_DEVICES, MISC_NAME, major) ||
+	    !_get_proc_number(PROC_MISC, DM_NAME, minor)) {
+		*major = 0;
+		return 0;
+	}
+
+	return 1;
+}
+
+/*
+ * Returns 1 if exists; 0 if it doesn't; -1 if it's wrong
+ */
+static int _control_exists(const char *control, uint32_t major, uint32_t minor)
+{
+	struct stat buf;
+
+	if (stat(control, &buf) < 0) {
+		if (errno != ENOENT)
+			log_error("%s: stat failed: %s", control,
+				  strerror(errno));
+		return 0;
+	}
+
+	if (!S_ISCHR(buf.st_mode)) {
+		log_verbose("%s: Wrong inode type", control);
+		if (!unlink(control))
+			return 0;
+		log_error("%s: unlink failed: %s", control,
+			  strerror(errno));
+		return -1;
+	}
+
+	if (major && buf.st_rdev != MKDEV(major, minor)) {
+		log_verbose("%s: Wrong device number: (%u, %u) instead of "
+			    "(%u, %u)", control,
+			    MAJOR(buf.st_mode), MINOR(buf.st_mode),
+			    major, minor);
+		if (!unlink(control))
+			return 0;
+		log_error("%s: unlink failed: %s", control,
+			  strerror(errno));
+		return -1;
+	}
+
+	return 1;
+}
+
+static int _create_control(const char *control, uint32_t major, uint32_t minor)
+{
+	int ret;
+	mode_t old_umask;
+
+	if (!major)
+		return 0;
+
+	old_umask = umask(0022);
+	ret = create_dir(dm_dir());
+	umask(old_umask);
+
+	if (!ret)
+		return 0;
+
+	log_verbose("Creating device %s (%u, %u)", control, major, minor);
+
+	if (mknod(control, S_IFCHR | S_IRUSR | S_IWUSR,
+		  MKDEV(major, minor)) < 0)  {
+		log_error("%s: mknod failed: %s", control, strerror(errno));
+		return 0;
+	}
+
+#ifdef HAVE_SELINUX
+        if (!set_selinux_context(control)) {
+                stack;
+                return 0;
+        }
+#endif
+
+	return 1;
+}
+
 static int _open_control(void)
 {
 	char control[PATH_MAX];
+	uint32_t major = 0, minor;
 
 	if (_control_fd != -1)
 		return 1;
 
 	snprintf(control, sizeof(control), "%s/control", dm_dir());
 
+	if (!_control_device_number(&major, &minor))
+		log_error("Is device-mapper driver missing from kernel?");
+
+	if (!_control_exists(control, major, minor) &&
+	    !_create_control(control, major, minor))
+		goto error;
+
 	if ((_control_fd = open(control, O_RDWR)) < 0) {
 		log_error("%s: open failed: %s", control, strerror(errno));
-		log_error("Is device-mapper driver missing from kernel?");
-		return 0;
+		goto error;
 	}
 
 	return 1;
+
+error:
+	log_error("Failure to communicate with kernel device-mapper driver.");
+	return 0;
 }
 
 void dm_task_destroy(struct dm_task *dmt)
@@ -495,9 +618,14 @@ static int _dm_task_run_v1(struct dm_task *dmt)
 		goto bad;
 	}
 
+	if (dmi->flags & DM_BUFFER_FULL_FLAG)
+		/* FIXME Increase buffer size and retry operation (if query) */
+		log_error("Warning: libdevmapper buffer too small for data");
+
 	switch (dmt->type) {
 	case DM_DEVICE_CREATE:
-		add_dev_node(dmt->dev_name, MAJOR(dmi->dev), MINOR(dmi->dev));
+		add_dev_node(dmt->dev_name, MAJOR(dmi->dev), MINOR(dmi->dev),
+			     dmt->uid, dmt->gid, dmt->mode);
 		break;
 
 	case DM_DEVICE_REMOVE:
@@ -511,7 +639,8 @@ static int _dm_task_run_v1(struct dm_task *dmt)
 	case DM_DEVICE_MKNODES:
 		if (dmi->flags & DM_EXISTS_FLAG)
 			add_dev_node(dmt->dev_name, MAJOR(dmi->dev),
-				     MINOR(dmi->dev));
+				     MINOR(dmi->dev),
+				     dmt->uid, dmt->gid, dmt->mode);
 		else
 			rm_dev_node(dmt->dev_name);
 		break;
@@ -800,6 +929,13 @@ int dm_task_set_sector(struct dm_task *dmt, uint64_t sector)
 	return 1;
 }
 
+int dm_task_no_open_count(struct dm_task *dmt)
+{
+	dmt->no_open_count = 1;
+
+	return 1;
+}
+
 int dm_task_set_event_nr(struct dm_task *dmt, uint32_t event_nr)
 {
 	dmt->event_nr = event_nr;
@@ -1152,7 +1288,8 @@ int dm_task_run(struct dm_task *dmt)
 	if (dmt->type == DM_DEVICE_CREATE && dmt->head)
 		return _create_and_load_v4(dmt);
 
-	if (dmt->type == DM_DEVICE_MKNODES && !dmt->dev_name)
+	if (dmt->type == DM_DEVICE_MKNODES && !dmt->dev_name &&
+	    !dmt->uuid && dmt->major <= 0)
 		return _mknodes_v4(dmt);
 
 	if (!_open_control())
@@ -1168,8 +1305,13 @@ int dm_task_run(struct dm_task *dmt)
 		dmi->flags |= DM_STATUS_TABLE_FLAG;
 
 	dmi->flags |= DM_EXISTS_FLAG;	/* FIXME */
-	log_debug("dm %s %s %s %s%.0llu %s", _cmd_data_v4[dmt->type].name,
+
+	if (dmt->no_open_count)
+		dmi->flags |= DM_SKIP_BDGET_FLAG;
+
+	log_debug("dm %s %s %s %s%c %.0llu %s", _cmd_data_v4[dmt->type].name,
 		  dmi->name, dmi->uuid, dmt->newname ? dmt->newname : "",
+		  dmt->no_open_count ? 'N' : 'O',
 		  dmt->sector, dmt->message ? dmt->message : "");
 	if (ioctl(_control_fd, command, dmi) < 0) {
 		if (errno == ENXIO && ((dmt->type == DM_DEVICE_INFO) ||
@@ -1189,22 +1331,28 @@ int dm_task_run(struct dm_task *dmt)
       ignore_error:
 	switch (dmt->type) {
 	case DM_DEVICE_CREATE:
-		add_dev_node(dmt->dev_name, MAJOR(dmi->dev), MINOR(dmi->dev));
+		add_dev_node(dmt->dev_name, MAJOR(dmi->dev), MINOR(dmi->dev),
+			     dmt->uid, dmt->gid, dmt->mode);
 		break;
 
 	case DM_DEVICE_REMOVE:
-		rm_dev_node(dmt->dev_name);
+		/* FIXME Kernel needs to fill in dmi->name */
+		if (dmt->dev_name)
+			rm_dev_node(dmt->dev_name);
 		break;
 
 	case DM_DEVICE_RENAME:
-		rename_dev_node(dmt->dev_name, dmt->newname);
+		/* FIXME Kernel needs to fill in dmi->name */
+		if (dmt->dev_name)
+			rename_dev_node(dmt->dev_name, dmt->newname);
 		break;
 
 	case DM_DEVICE_MKNODES:
 		if (dmi->flags & DM_EXISTS_FLAG)
-			add_dev_node(dmt->dev_name, MAJOR(dmi->dev),
-				     MINOR(dmi->dev));
-		else
+			add_dev_node(dmi->name, MAJOR(dmi->dev),
+				     MINOR(dmi->dev),
+				     dmt->uid, dmt->gid, dmt->mode);
+		else if (dmt->dev_name)
 			rm_dev_node(dmt->dev_name);
 		break;
 

libdm/ioctl/libdm-targets.h

@@ -17,6 +17,7 @@
 #define LIB_DMTARGETS_H
 
 #include <inttypes.h>
+#include <sys/types.h>
 
 struct dm_ioctl;
 struct dm_ioctl_v1;
@@ -40,6 +41,9 @@ struct dm_task {
 	uint32_t event_nr;
 	int major;
 	int minor;
+	uid_t uid;
+	gid_t gid;
+	mode_t mode;
 	union {
 		struct dm_ioctl *v4;
 		struct dm_ioctl_v1 *v1;
@@ -47,6 +51,7 @@ struct dm_task {
 	char *newname;
 	char *message;
 	uint64_t sector;
+	int no_open_count;
 
 	char *uuid;
 };

libdm/libdevmapper.h

@@ -134,6 +134,7 @@ int dm_task_set_major(struct dm_task *dmt, int major);
 int dm_task_set_event_nr(struct dm_task *dmt, uint32_t event_nr);
 int dm_task_set_message(struct dm_task *dmt, const char *message);
 int dm_task_set_sector(struct dm_task *dmt, uint64_t sector);
+int dm_task_no_open_count(struct dm_task *dmt);
 
 /*
  * Use these to prepare for a create or reload.