o Build dmeventd against multilog

to retrieve which devices got registered with the daemon;
  needs locking changes as well

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 @@
-2.00.16-cvs (2004-05-24)
+2.01.10-cvs (2005-04-04)

WHATS_NEW

@@ -1,5 +1,259 @@
-Version 2.00.17 -
+Version 2.01.10 - 
+================================
+  Tidy lv_segment interface.
+  Initial pv_segment support.
+  vgchange --physicalextentsize
+  Internal snapshot restructuring.
+  Remove unused internal non-persistent snapshot option.
+  Allow offline extension of snapshot volumes.
+  Move from 2-step to 3-step on-disk metadata commit.
+  Scan ramdisks too and allow non-O_DIRECT fallback.
+  Annotate, tidy and extend list.h.
+  Alignment tidying.
+  Make clvmd work around some "bugs" in gulm's node state notifications.
+  Tidy clvmd's SIGHUP handler
+
+Version 2.01.09 - 4th April 2005
+================================
+  Add --ignorelockingfailure to vgmknodes.
+  clvmd: Don't allow user operations to start until the lvm thread is fully up.
+  clvmd-gulm: set KEEPALIVE on sockets.
+
+Version 2.01.08 - 22nd March 2005
+=================================
+  Add clustered attribute so vgchange can identify clustered VGs w/o locking.
+  Improve detection of external changes affecting internal cache.
+  Add 'already in device cache' debug message.
+  Add -a to pvdisplay -C.
+  Avoid rmdir opendir error messsages when dir was already removed.
+  Tighten signal handlers.
+  Avoid some compiler warnings.
+  Additional rename failure error message.
+  read/write may be macros.
+  clvmd: don't take out lvm thread lock at startup, it only protects jobs list.
+
+Version 2.01.07 - 8th March 2005
+================================
+  Cope with new devices appearing by rescanning /dev if a uuid can't be found.
+  Remove DESTDIR from LVM_SHARED_PATH.
+  clvmd fixes: make FDs close-on-exec
+               gulm unlocks VG & orphan locks at startup in case they are stale
+               gulm now unlocks VG & orphan locks if client dies.
+
+Version 2.01.06 - 1st March 2005
+================================
+  Suppress 'open failed' error messages during scanning.
+  Option to suppress warnings of file descriptors left open.
+  Fix default value of metadatacopies in documentation (2->1).
+  Fix clvmd-gulm locking.
+  ./configure --enable-debug now enables debugging code in clvmd.
+  Fix clvmd-gulm node up/down code so it actually works.
+  clvmd-gulm now releases locks when shut down.
+
+Version 2.01.05 - 18th February 2005
+====================================
+  Static binary invokes dynamic binary if appropriate.
+  Make clvmd config check a little more tolerant.
+  gulm clvmd can now cope with >1 message arriving in a TCP message.
+
+Version 2.01.04 - 9th February 2005
+===================================
+  Add fixed offset to imported pool minor numbers.
+  Update binary pathnames in clvmd_init_rhel4.
+  lvm2cmd.so should skip the check for open fds.
+  Remove unused -f from pvmove.
+  Gulm clvmd doesn't report "connection refused" errors.
+  clvmd does a basic config file sanity check at startup.
+  Fix potential thread shutdown race in clvmd.
+
+Version 2.01.03 - 1st February 2005
+===================================
+  More 64-bit display/report fixes.
+  More informative startup mesg if can't create /etc/lvm.
+  Fix snapshot device size bug (since 2.01.01).
+  clvmd announces startup and cluster connection in syslog.
+  Gulm clvmd doesn't hang trying to talk to a rebooted node.
+  Gulm clvmd doesn't print cman error on startup.
+
+Version 2.01.02 - 21st January 2005
+===================================
+  Update clvmd_init_rhel4: use lvm.static and don't load dlm.
+  Fix some size_t printing.
+  Fix 64 bit xlate consts.
+  Split out pool sptype_names to avoid unused const.
+  Always fail if random id generation fails.
+  Recognise gnbd devices.
+  Fix clvmd startup bug introduced in cman/gulm amalgamation.
+  Improve reporting of node-specific locking errors.
+
+Version 2.01.01 - 19th January 2005
+===================================
+  Fix clvmd lv_info_by_lvid open_count.
+  Store snapshot and origin sizes separately.
+  Update vgcreate man page.
+
+Version 2.01.00 - 17th January 2005
+===================================
+  Fix vgscan metadata auto-correction.
+  Only ask libdevmapper for open_count when we need it.
+  Adjust RHEL4 clvmd init script priority.
+  Enable building of CMAN & GULM versions of clvmd into a single binary
+
+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.
+  Display all filtered devices, not just PVs, with pvs -a.
+  Fix sync_dir() when no / in filename
+  vgcfgbackup -f accepts template with %s for VG name.
+  Extend hash functions to handle non-null-terminated data.
+  Add local activation support.
+  Tidy relative paths in makefile includes.
+  fsadm support for fsck and resizing - needs testing.
   Add read-only GFS pool support.
   Add lvm2create_initrd script from http://poochiereds.net/svn/lvm2/
   Fix rounding of large diplayed sizes.

WHATS_NEW_DM

@@ -1,5 +1,42 @@
-Version 1.00.18
+Version 1.01.02 - 
 =============================
+
+Version 1.01.01 - 29 Mar 2005
+=============================
+  Update dmsetup man page.
+  Drop-in devmap_name replacement.
+  Add option to compile without ioctl for testing.
+  Fix DM_LIB_VERSION sed.
+
+Version 1.01.00 - 17 Jan 2005
+=============================
+  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.
+  Fix a targets string size calc in driver.
+  Add -c to dmsetup for column-based output.
   Add target message-passing ioctl.
 
 Version 1.00.17 - 17 Apr 2004

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
@@ -31,213 +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 ;;
-	darwin*)
-		CFLAGS="-no-cpp-precomp -fno-common"
-		CLDFLAGS=
-		CLDWHOLEARCHIVE="-all_load"
-		CLDNOWHOLEARCHIVE=
-		LD_DEPS=
-		LD_FLAGS=
-		SOFLAG="-dynamiclib"
-		DEVMAPPER=no
-		ODIRECT=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""$DEVMAPPER" 1>&6
+################################################################################
+dnl -- Build cluster LVM daemon
+AC_MSG_CHECKING(whether to build cluster LVM daemon)
+AC_ARG_WITH(clvmd,
+  [  --with-clvmd=TYPE       Build cluster LVM Daemon: cman/gulm/none/all
+                          [TYPE=none] ],
+  [ CLVMD="$withval" ],
+  [ CLVMD="none" ])
+if test x$CLVMD = xyes; then
+	CLVMD=all
+fi
+AC_MSG_RESULT($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
 
-dnl Mess with default exec_prefix
+################################################################################
+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
 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(
@@ -250,42 +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$READLINE = xyes; then
 	AC_CHECK_LIB(readline, readline, ,
 		AC_MSG_ERROR(
@@ -296,17 +437,17 @@ support with --disable-readline or download and install readline from:
 Note: if you are using precompiled packages you will also need the development
 package as well (which may be called readline-devel or something similar).
 )
-		exit
 	)
 	AC_CHECK_FUNC(rl_completion_matches, CFLAGS="$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$INTL = xyes; then
 	INTL_PACKAGE="lvm2"
@@ -315,7 +456,6 @@ if test x$INTL = xyes; then
 		then  AC_MSG_ERROR(
 		msgfmt not found in path $PATH
 		)
-		exit  
 	fi;
 
 	AC_ARG_WITH(localedir,
@@ -324,6 +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="$withval" ],
@@ -334,13 +475,54 @@ AC_ARG_WITH(staticdir,
   	    [ STATICDIR="$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$CLVMD != xnone; then
+	AC_CHECK_HEADERS(mntent.h netdb.h netinet/in.h pthread.h search.h sys/mount.h sys/socket.h sys/uio.h sys/un.h utmpx.h,,AC_MSG_ERROR(bailing out))
+	AC_CHECK_FUNCS(dup2 getmntent memmove select socket,,AC_MSG_ERROR(bailing out))
+	AC_FUNC_GETMNTENT
+#	AC_FUNC_REALLOC
+	AC_FUNC_SELECT_ARGTYPES
+fi
+
+if test x$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)
@@ -350,11 +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)
@@ -370,23 +553,31 @@ 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							 	\
 po/Makefile								\
 tools/Makefile							 	\
 tools/version.h								\
+tools/fsadm/Makefile							\
 test/mm/Makefile							\
 test/device/Makefile							\
 test/format1/Makefile							\
@@ -395,8 +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
+  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 +1,76 @@
+#
+# 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          \
+	lvm-functions.c  \
+	system-lv.c
+
+ifeq ("@CLVMD@", "gulm")
+	GULM = yes
+endif
+
+ifeq ("@CLVMD@", "cman")
+	CMAN = yes
+endif
+
+ifeq ("@CLVMD@", "all")
+	GULM = yes
+	CMAN = yes
+endif
+
+ifeq ("@DEBUG@", "yes")
+	CFLAGS += -DDEBUG
+endif
+
+ifeq ("$(GULM)", "yes")
+	SOURCES += clvmd-gulm.c tcp-comms.c
+	LMLIBS += -lccs -lgulm
+	CFLAGS += -DUSE_GULM
+endif
+
+ifeq ("$(CMAN)", "yes")
+	SOURCES += clvmd-cman.c
+	LMLIBS += -ldlm
+	CFLAGS += -DUSE_CMAN
+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 +1,66 @@
+/*
+ * 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 */
+#define CLVMD_FLAG_NODEERRS     4       /* Reply has errors in node-specific portion */
+
+/* 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 +1,540 @@
+/*
+ * 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;
+};
+
+static int _init_cluster(void)
+{
+	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) {
+		/* Don't print an error here because we could be just probing for CMAN */
+		return -1;
+	}
+	/* Set Close-on-exec */
+	fcntl(cluster_sock, F_SETFD, 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;
+}
+
+static void _cluster_init_completed(void)
+{
+	clvmd_cluster_init_completed();
+}
+
+static int _get_main_cluster_fd()
+{
+	return cluster_sock;
+}
+
+static int _get_num_nodes()
+{
+	return num_nodes;
+}
+
+/* send_message with the fd check removed */
+static 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, CMAN_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;
+}
+
+static void _get_our_csid(char *csid)
+{
+	int i;
+	memset(csid, 0, CMAN_MAX_CSID_LEN);
+
+	for (i = 0; i < num_nodes; i++) {
+		if (nodes[i].us)
+			memcpy(csid, &nodes[i].node_id, CMAN_MAX_CSID_LEN);
+	}
+}
+
+/* Call a callback routine for each node that known (down mean not running a clvmd) */
+static 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]);
+	}
+}
+
+static 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;
+}
+
+static 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);
+}
+
+static 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 */
+static 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 */
+static 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, CMAN_MAX_CSID_LEN);
+			return 0;
+		}
+	}
+	return -1;
+}
+
+/* Convert a CSID to a node name */
+static int _name_from_csid(char *csid, char *name)
+{
+	int i;
+
+	for (i = 0; i < num_nodes; i++) {
+		if (memcmp(csid, &nodes[i].node_id, CMAN_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 */
+static 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, CMAN_MAX_CSID_LEN);
+
+	return nodeid;
+}
+
+static 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);
+}
+
+static 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;
+}
+
+static 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;
+
+}
+
+static struct cluster_ops _cluster_cman_ops = {
+	.cluster_init_completed   = _cluster_init_completed,
+	.cluster_send_message     = _cluster_send_message,
+	.name_from_csid           = _name_from_csid,
+	.csid_from_name           = _csid_from_name,
+	.get_num_nodes            = _get_num_nodes,
+	.cluster_fd_callback      = _cluster_fd_callback,
+	.get_main_cluster_fd      = _get_main_cluster_fd,
+	.cluster_do_node_callback = _cluster_do_node_callback,
+	.is_quorate               = _is_quorate,
+	.get_our_csid             = _get_our_csid,
+	.add_up_node              = _add_up_node,
+	.cluster_closedown        = _cluster_closedown,
+	.sync_lock                = _sync_lock,
+	.sync_unlock              = _sync_unlock,
+};
+
+struct cluster_ops *init_cman_cluster(void)
+{
+	if (!_init_cluster())
+		return &_cluster_cman_ops;
+	else
+		return NULL;
+}

daemons/clvmd/clvmd-command.c

@@ -0,0 +1,286 @@
+/*
+ * 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);
+		char *lockname = hash_get_key(lock_hash, v);
+
+		DEBUGLOG("cleanup: Unlocking lock %s %x\n", lockname, lkid);
+		sync_unlock(lockname, lkid);
+	}
+
+	hash_destroy(lock_hash);
+	client->bits.localsock.private = 0;
+    }
+}

daemons/clvmd/clvmd-comms.h

@@ -0,0 +1,70 @@
+/*
+ * 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;
+
+struct cluster_ops {
+	void (*cluster_init_completed) (void);
+
+	int (*cluster_send_message) (void *buf, int msglen, char *csid,
+				const char *errtext);
+	int (*name_from_csid) (char *csid, char *name);
+	int (*csid_from_name) (char *csid, char *name);
+	int (*get_num_nodes) (void);
+	int (*cluster_fd_callback) (struct local_client *fd, char *buf, int len,
+			       char *csid, struct local_client **new_client);
+	int (*get_main_cluster_fd) (void);	/* gets accept FD or cman cluster socket */
+	int (*cluster_do_node_callback) (struct local_client *client,
+				    void (*callback) (struct local_client *,
+						      char *csid, int node_up));
+	int (*is_quorate) (void);
+
+	void (*get_our_csid) (char *csid);
+	void (*add_up_node) (char *csid);
+	void (*reread_config) (void);
+	void (*cluster_closedown) (void);
+
+	int (*sync_lock) (const char *resource, int mode, int flags, int *lockid);
+	int (*sync_unlock) (const char *resource, int lockid);
+
+};
+
+#ifdef USE_GULM
+#  include "tcp-comms.h"
+struct cluster_ops *init_gulm_cluster(void);
+#define MAX_CSID_LEN 			GULM_MAX_CSID_LEN
+#define MAX_CLUSTER_MEMBER_NAME_LEN	GULM_MAX_CLUSTER_MEMBER_NAME_LEN
+#endif
+
+#ifdef USE_CMAN
+#  include "cnxman-socket.h"
+#  define CMAN_MAX_CSID_LEN 4
+#  ifndef MAX_CSID_LEN
+#    define MAX_CSID_LEN CMAN_MAX_CSID_LEN
+#  endif
+#  undef MAX_CLUSTER_MEMBER_NAME_LEN
+#  define MAX_CLUSTER_MEMBER_NAME_LEN	CMAN_MAX_CLUSTER_MEMBER_NAME_LEN
+struct cluster_ops *init_cman_cluster(void);
+#endif
+
+
+#endif

daemons/clvmd/clvmd-gulm.c

@@ -0,0 +1,997 @@
+/******************************************************************************
+*******************************************************************************
+**
+**  Copyright (C) Sistina Software, Inc.  2002-2003  All rights reserved.
+**  Copyright (C) 2004-2005 Red Hat, Inc. 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 "lvm-functions.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 quorate state */
+static uint8_t gulm_quorate = 0;
+static enum {INIT_NOTDONE, INIT_DONE, INIT_WAITQUORATE} init_state = INIT_NOTDONE;
+
+/* Number of active nodes */
+static int num_nodes;
+
+static char *cluster_name;
+static int in_shutdown = 0;
+
+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[GULM_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);
+static int _csid_from_name(char *csid, char *name);
+static void _cluster_closedown(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);
+
+    /* Set Close-on-exec */
+    fcntl(fd, F_SETFD, 1);
+
+    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 _reread_config(void)
+{
+        /* Re-read CCS node list */
+	DEBUGLOG("Re-reading CCS config\n");
+	get_all_cluster_nodes();
+}
+
+static int _init_cluster(void)
+{
+    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);
+
+    return 0;
+}
+
+static void _cluster_closedown(void)
+{
+    DEBUGLOG("cluster_closedown\n");
+    in_shutdown = 1;
+    unlock_all();
+    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];
+
+    DEBUGLOG("Dropping expired locks for %s\n", nodename?nodename:"(null)");
+    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);
+
+   /* Get the current core state (for quorum) */
+   lg_core_corestate(gulm_if);
+
+   return 0;
+}
+
+static void set_node_state(struct node_info *ninfo, char *csid, uint8_t nodestate)
+{
+    int oldstate = ninfo->state;
+
+    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;
+	}
+    }
+    /* Gulm doesn't always send node DOWN events, so even if this a a node UP we must
+     * assume (ahem) that it prevously went down at some time. So we close
+     * the sockets here to make sure that we don't have any dead connections
+     * to that node.
+     */
+    tcp_remove_client(csid);
+
+    DEBUGLOG("set_node_state, '%s' state = %d (oldstate=%d), num_nodes=%d\n",
+	     ninfo->name, ninfo->state, oldstate, 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, GULM_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, GULM_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 void _get_our_csid(char *csid)
+{
+	get_our_gulm_csid(csid);
+}
+
+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[GULM_MAX_CSID_LEN];
+
+		DEBUGLOG("Got Nodelist, stop\n");
+		if (gulm_quorate)
+		{
+			clvmd_cluster_init_completed();
+			init_state = INIT_DONE;
+		}
+		else
+		{
+			if (init_state == INIT_NOTDONE)
+				init_state = INIT_WAITQUORATE;
+		}
+
+		/* Mark ourself as up */
+		_get_our_csid(ourcsid);
+		gulm_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. quorate:%d, corestate:%x mastername:%s\n",
+	     quorate, corestate, mastername);
+
+    gulm_quorate = quorate;
+    if (quorate && init_state == INIT_WAITQUORATE)
+    {
+	    clvmd_cluster_init_completed();
+	    init_state = INIT_DONE;
+    }
+    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);
+
+    /* No waiting process to wake up when we are shutting down */
+    if (in_shutdown)
+	    return 0;
+
+    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), GULM_MAX_CSID_LEN);
+    return 1;
+}
+
+int gulm_name_from_csid(char *csid, char *name)
+{
+    struct node_info *ninfo;
+
+    ninfo = hash_lookup_binary(node_hash, csid, GULM_MAX_CSID_LEN);
+    if (!ninfo)
+    {
+        sprintf(name, "UNKNOWN %s", print_csid(csid));
+	return -1;
+    }
+
+    strcpy(name, ninfo->name);
+    return 0;
+}
+
+
+static 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), GULM_MAX_CSID_LEN);
+	    return 0;
+	}
+    }
+    return -1;
+}
+
+static int _get_num_nodes()
+{
+    DEBUGLOG("num_nodes = %d\n", num_nodes);
+    return num_nodes;
+}
+
+/* Node is now known to be running a clvmd */
+void gulm_add_up_node(char *csid)
+{
+    struct node_info *ninfo;
+
+    ninfo = hash_lookup_binary(node_hash, csid, GULM_MAX_CSID_LEN);
+    if (!ninfo) {
+	    DEBUGLOG("gulm_add_up_node no node_hash entry for csid %s\n", print_csid(csid));
+	return;
+    }
+
+    DEBUGLOG("gulm_add_up_node %s\n", ninfo->name);
+
+    if (ninfo->state == NODE_DOWN)
+	    num_nodes++;
+    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, GULM_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;
+    drop_expired_locks(ninfo->name);
+    return;
+
+}
+
+/* Call a callback for each node, so the caller knows whether it's up or down */
+static 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[GULM_MAX_CSID_LEN];
+	struct local_client *client;
+
+	ninfo = hash_get_data(node_hash, hn);
+	memcpy(csid, hash_get_key(node_hash, hn), GULM_MAX_CSID_LEN);
+
+	DEBUGLOG("down_callback. node %s, state = %d\n", ninfo->name, ninfo->state);
+
+	client = hash_lookup_binary(sock_hash, csid, GULM_MAX_CSID_LEN);
+	if (!client)
+	{
+	    /* If it's up but not connected, try to make contact */
+	    if (ninfo->state == NODE_UP)
+		    gulm_connect_csid(csid, &client);
+
+	    client = hash_lookup_binary(sock_hash, csid, GULM_MAX_CSID_LEN);
+
+	}
+	if (ninfo->state != NODE_DOWN)
+		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:
+    case lg_err_AlreadyPend:
+	    errno = EAGAIN;
+	    break;
+
+	/* 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;
+    }
+
+    /* When we are shutting down, don't wait for unlocks
+       to be acknowledged, just do it. */
+    if (in_shutdown)
+	    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).
+*/
+static 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 too 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);
+	if (status)
+		goto out;
+	status = _unlock_resource(lock2, *lockid);
+	break;
+
+    case LCK_WRITE:
+	status = _lock_resource(lock2, lg_lock_state_Exclusive, flags, lockid);
+	if (status)
+		goto out;
+	status = _unlock_resource(lock1, *lockid);
+	break;
+
+    default:
+	status = -1;
+	errno = EINVAL;
+	break;
+    }
+ out:
+    *lockid = mode;
+    return status;
+}
+
+static 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);
+
+    status = _unlock_resource(lock1, lockid);
+    if (!status)
+	    status = _unlock_resource(lock2, lockid);
+
+    return status;
+}
+
+static int _is_quorate()
+{
+	return gulm_quorate;
+}
+
+/* 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[GULM_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, GULM_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, GULM_MAX_CSID_LEN, ninfo);
+	    }
+	}
+	else
+	{
+	    DEBUGLOG("node %s has clvm disabled\n", nodename);
+	}
+	free(nodename);
+    }
+
+    /* Finished with config file */
+    ccs_disconnect(ctree);
+
+    return 0;
+}
+
+static int _get_main_cluster_fd(void)
+{
+	return get_main_gulm_cluster_fd();
+}
+
+static int _cluster_fd_callback(struct local_client *fd, char *buf, int len, char *csid, struct local_client **new_client)
+{
+	return cluster_fd_gulm_callback(fd, buf, len, csid, new_client);
+}
+
+static int _cluster_send_message(void *buf, int msglen, char *csid, const char *errtext)
+{
+	return gulm_cluster_send_message(buf, msglen, csid, errtext);
+}
+
+static struct cluster_ops _cluster_gulm_ops = {
+	.cluster_init_completed   = NULL,
+	.cluster_send_message     = _cluster_send_message,
+	.name_from_csid           = gulm_name_from_csid,
+	.csid_from_name           = _csid_from_name,
+	.get_num_nodes            = _get_num_nodes,
+	.cluster_fd_callback      = _cluster_fd_callback,
+	.get_main_cluster_fd      = _get_main_cluster_fd,
+	.cluster_do_node_callback = _cluster_do_node_callback,
+	.is_quorate               = _is_quorate,
+	.get_our_csid             = _get_our_csid,
+	.add_up_node              = gulm_add_up_node,
+	.reread_config            = _reread_config,
+	.cluster_closedown        = _cluster_closedown,
+	.sync_lock                = _sync_lock,
+	.sync_unlock              = _sync_unlock,
+};
+
+struct cluster_ops *init_gulm_cluster(void)
+{
+	if (!_init_cluster())
+		return &_cluster_gulm_ops;
+	else
+		return NULL;
+}

daemons/clvmd/clvmd-gulm.h

@@ -0,0 +1,12 @@
+
+
+
+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);
+
+void gulm_add_up_node(char *csid);
+int gulm_name_from_csid(char *csid, char *name);

daemons/clvmd/clvmd.h

@@ -0,0 +1,124 @@
+/*
+ * 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;
+	uint8_t removeme;
+
+	union {
+		struct localsock_bits localsock;
+		struct pipe_bits pipe;
+		struct netsock_bits net;
+	} bits;
+};
+
+#ifdef DEBUG
+#define DEBUGLOG(fmt, args...) {time_t P; time(&P); fprintf(stderr, "CLVMD[%x]: %.15s ", (int)pthread_self(), ctime(&P)+4 ); 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);
+
+int sync_lock(const char *resource, int mode, int flags, int *lockid);
+int sync_unlock(const char *resource, int lockid);
+
+#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 CMAN_MAX_CLUSTER_MESSAGE          1500
+#define CMAN_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[CMAN_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/lvm-functions.c

@@ -0,0 +1,532 @@
+/*
+ * 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, 0))
+		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, 0))
+		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, 0))
+				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;
+}
+
+
+/* Only called at gulm startup. Drop any leftover VG or P_orphan locks
+   that might be hanging around if we died for any reason
+*/
+static void drop_vg_locks()
+{
+	char vg[128];
+	char line[255];
+	FILE *vgs =
+	    popen
+	    ("lvm pvs --nolocking --noheadings -o vg_name", "r");
+
+	sync_unlock("P_orphans", LCK_EXCL);
+
+	if (!vgs)
+		return;
+
+	while (fgets(line, sizeof(line), vgs)) {
+		char *vgend;
+		char *vgstart;
+
+		if (line[strlen(line)-1] == '\n')
+			line[strlen(line)-1] = '\0';
+
+		vgstart = line + strspn(line, " ");
+		vgend = vgstart + strcspn(vgstart, " ");
+		*vgend = '\0';
+
+		if (strncmp(vgstart, "WARNING:", 8) == 0)
+			continue;
+
+		sprintf(vg, "V_%s", vgstart);
+		sync_unlock(vg, LCK_EXCL);
+
+	}
+	fclose(vgs);
+}
+
+/*
+ * 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
+	    ("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;
+}
+
+/* This checks some basic cluster-LVM configuration stuff */
+static void check_config()
+{
+	int locking_type;
+
+	locking_type = find_config_int(cmd->cft->root, "global/locking_type", 1);
+
+	if (locking_type == 3) /* compiled-in cluster support */
+		return;
+
+	if (locking_type == 2) { /* External library, check name */
+		const char *libname;
+
+		libname = find_config_str(cmd->cft->root, "global/locking_library",
+					  "");
+		if (strstr(libname, "liblvm2clusterlock.so"))
+			return;
+
+		log_error("Incorrect LVM locking library specified in lvm.conf, cluster operations may not work.");
+		return;
+	}
+	log_error("locking_type not set correctly in lvm.conf, cluster operations will not work.");
+}
+
+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(int using_gulm)
+{
+	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);
+
+	/* Check lvm.conf is setup for cluster-LVM */
+	check_config();
+
+	/* Remove any non-LV locks that may have been left around */
+	if (using_gulm)
+		drop_vg_locks();
+
+	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(int using_gulm);
+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 +1,371 @@
+/*
+ * 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 "clvm.h"
+#include "clvmd-comms.h"
+#include "clvmd.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 +1,484 @@
+/******************************************************************************
+*******************************************************************************
+**
+**  Copyright (C) Sistina Software, Inc.  2002-2003  All rights reserved.
+**  Copyright (C) 2004-2005 Red Hat, Inc. 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));
+	setsockopt(listen_fd, SOL_SOCKET, SO_KEEPALIVE, &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);
+
+    /* Set Close-on-exec */
+    fcntl(listen_fd, F_SETFD, 1);
+
+    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, GULM_MAX_CSID_LEN);
+    if (client)
+    {
+	hash_remove_binary(sock_hash, csid, GULM_MAX_CSID_LEN);
+	client->removeme = 1;
+    }
+
+    /* Look for a mangled one too */
+    csid[0] ^= 0x80;
+
+    client = hash_lookup_binary(sock_hash, csid, GULM_MAX_CSID_LEN);
+    if (client)
+    {
+	hash_remove_binary(sock_hash, csid, GULM_MAX_CSID_LEN);
+	client->removeme = 1;
+    }
+
+    /* 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, GULM_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, GULM_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, GULM_MAX_CSID_LEN, client);
+
+    return 0;
+}
+
+int get_main_gulm_cluster_fd()
+{
+    return listen_fd;
+}
+
+
+/* Read on main comms (listen) socket, accept it */
+int cluster_fd_gulm_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[GULM_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 (gulm_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);
+    struct clvm_header *header = (struct clvm_header *)buf;
+    int status;
+    uint32_t arglen;
+
+    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, GULM_MAX_CSID_LEN);
+
+    /* Read just the header first, then get the rest if there is any.
+     * Stream sockets, sigh.
+     */
+    status = read(client->fd, buf, sizeof(struct clvm_header));
+    if (status > 0)
+    {
+	    int status2;
+
+	    arglen = ntohl(header->arglen);
+
+	    /* Get the rest */
+	    if (arglen && arglen < GULM_MAX_CLUSTER_MESSAGE)
+	    {
+		    status2 = read(client->fd, buf+status, arglen);
+		    if (status2 > 0)
+			    status += status2;
+		    else
+			    status = status2;
+	    }
+    }
+
+    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[GULM_MAX_CSID_LEN];
+
+	memcpy(remcsid, csid, GULM_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, GULM_MAX_CSID_LEN);
+
+	/* Tell cluster manager layer */
+	add_down_node(remcsid);
+    }
+    else {
+	    gulm_add_up_node(csid);
+	    /* Send it back to clvmd */
+	    process_message(client, buf, status, csid);
+    }
+    return status;
+}
+
+int gulm_connect_csid(char *csid, struct local_client **newclient)
+{
+    int fd;
+    struct sockaddr_in6 addr;
+    int status;
+    int one = 1;
+
+    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, GULM_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)
+    {
+	/* "Connection refused" is "normal" because clvmd may not yet be running
+	 * on that node.
+	 */
+	if (errno != ECONNREFUSED)
+	{
+	    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;
+    }
+
+    /* Set Close-on-exec */
+    fcntl(fd, F_SETFD, 1);
+    setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(int));
+
+    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 */
+    gulm_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[GULM_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_gulm_csid(ourcsid);
+    if (memcmp(csid, ourcsid, GULM_MAX_CSID_LEN) == 0)
+	return msglen;
+
+    client = hash_lookup_binary(sock_hash, csid, GULM_MAX_CSID_LEN);
+    if (!client)
+    {
+	status = gulm_connect_csid(csid, &client);
+	if (status)
+	    return -1;
+    }
+    DEBUGLOG("tcp_send_message, fd = %d\n", client->fd);
+
+    return write(client->fd, buf, msglen);
+}
+
+
+int gulm_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[GULM_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_gulm_csid(char *csid)
+{
+    static char our_csid[GULM_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, GULM_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, GULM_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 +1,13 @@
+#include <netinet/in.h>
+
+#define GULM_MAX_CLUSTER_MESSAGE 1600
+#define GULM_MAX_CSID_LEN sizeof(struct in6_addr)
+#define GULM_MAX_CLUSTER_MEMBER_NAME_LEN 128
+
+extern int init_comms(unsigned short);
+extern char *print_csid(char *);
+int get_main_gulm_cluster_fd(void);
+int cluster_fd_gulm_callback(struct local_client *fd, char *buf, int len, char *csid, struct local_client **new_client);
+int gulm_cluster_send_message(void *buf, int msglen, char *csid, const char *errtext);
+void get_our_gulm_csid(char *csid);
+int gulm_connect_csid(char *csid, struct local_client **newclient);

daemons/dmeventd/.exported_symbols

@@ -0,0 +1,3 @@
+process_event
+register_device
+unregister_device

daemons/dmeventd/Makefile.in

@@ -0,0 +1,51 @@
+#
+# Copyright (C) 2001-2004 Sistina Software, Inc. All rights reserved.
+# Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+#
+# This file is part of the device-mapper userspace tools.
+#
+# 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@
+
+TARGETS = dmevent dmeventd
+INSTALL_TYPE = install_dynamic
+
+SOURCES = noop.c
+CLEAN_TARGETS = dmevent
+
+ifeq ("@LIB_SUFFIX@","dylib")
+  LIB_SHARED = libdmeventnoop.dylib
+else
+  LIB_SHARED = libdmeventnoop.so
+endif
+ 
+LDFLAGS += -ldl -ldevmapper -lpthread
+
+include ../make.tmpl
+
+libdmeventnoop.so: noop.o
+
+dmevent: dmevent.o $(interfacedir)/libdevmapper.$(LIB_SUFFIX) $(top_srcdir)/lib/event/libdmevent.$(LIB_SUFFIX)
+	$(CC) -o $@ dmevent.o $(LDFLAGS) \
+	      -L$(interfacedir) -L$(DESTDIR)/lib -L$(top_srcdir)/lib/event -L$(top_srcdir)/multilog -lmultilog -ldmevent $(LIBS)
+
+dmeventd: dmeventd.o $(interfacedir)/libdevmapper.$(LIB_SUFFIX) $(top_srcdir)/lib/event/libdmevent.$(LIB_SUFFIX)
+	$(CC) -o $@ dmeventd.o $(LDFLAGS) \
+	      -L$(interfacedir) -L$(DESTDIR)/lib -L$(top_srcdir)/lib/event -L$(top_srcdir)/multilog -ldmevent -lmultilog $(LIBS)
+
+install: $(INSTALL_TYPE)
+
+.PHONY: install_dynamic
+
+install_dynamic: dmeventd
+	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) dmeventd $(sbindir)/dmeventd
+

daemons/dmeventd/dmevent.c

@@ -0,0 +1,137 @@
+/*
+ * Copyright (C) 2005 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of the device-mapper userspace tools.
+ *
+ * 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 "libdevmapper.h"
+#include "libdm-event.h"
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/file.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+static enum event_type events = ALL_ERRORS; /* All until we can distinguish. */
+static char *default_dso_name = "noop";  /* default DSO is noop */
+static int default_reg = 1;		 /* default action is register */
+
+/* Display help. */
+static void print_usage(char *name)
+{
+	char *cmd = strrchr(name, '/');
+
+	cmd = cmd ? cmd + 1 : name;
+	printf("Usage::\n"
+	       "%s [options] <device>\n"
+	       "\n"
+	       "Options:\n"
+	       "  -d <dso>           Specify the DSO to use.\n"
+	       "  -h                 Print this usage.\n"
+	       "  -l                 List registered devices.\n"
+	       "  -r                 Register for event (default).\n"
+	       "  -u                 Unregister for event.\n"
+	       "\n", cmd);
+}
+
+/* Parse command line arguments. */
+static int parse_argv(int argc, char **argv,
+		      char **dso_name, char **device, int *reg, int *list)
+{
+	int c;
+	const char *options = "d:hlru";
+
+	while ((c = getopt(argc, argv, options)) != -1) {
+		switch (c) {
+		case 'd':
+			*dso_name = optarg;
+			break;
+		case 'h':
+			print_usage(argv[0]);
+			exit(EXIT_SUCCESS);
+		case 'l':
+			*list = 1;
+			break;
+		case 'r':
+			*reg = 1;
+			break;
+		case 'u':
+			*reg = 0;
+			break;
+		default:
+			fprintf(stderr, "Unknown option '%c'.\n"
+				"Try '-h' for help.\n", c);
+
+			return 0;
+		}
+	}
+
+	if (!*list) {
+		if (optind >= argc) {
+			fprintf(stderr, "You need to specify a device.\n");
+			return 0;
+		}
+
+		*device = argv[optind];
+	}
+
+	return 1;
+}
+
+int main(int argc, char **argv)
+{
+	int list = 0, ret, reg = default_reg;
+	char *device, *dso_name = default_dso_name;
+
+	if (!parse_argv(argc, argv, &dso_name, &device, &reg, &list))
+		exit(EXIT_FAILURE);
+	
+	if (list) {
+		do {
+			ret = dm_get_next_registered_device(&dso_name, &device, &events);
+			if (!ret) {
+				printf("%s %s 0x%x\n", dso_name, device, events);
+				free(dso_name);
+				free(device);
+			}
+		} while (ret);
+
+		exit(EXIT_SUCCESS);
+	}
+	if ((ret = reg ? dm_register_for_event(dso_name, device, events) :
+			 dm_unregister_for_event(dso_name, device, events))) {
+		fprintf(stderr, "Failed to %sregister %s: %s\n",
+			reg ? "": "un", device, strerror(-ret));
+
+		exit(EXIT_FAILURE);
+	}
+
+	printf("%s %sregistered successfully.\n", device, reg ? "" : "un");
+
+	exit(EXIT_SUCCESS);
+}
+
+/*
+ * Overrides for Emacs so that we follow Linus's tabbing style.
+ * Emacs will notice this stuff at the end of the file and automatically
+ * adjust the settings for this buffer only.  This must remain at the end
+ * of the file.
+ * ---------------------------------------------------------------------------
+ * Local variables:
+ * c-file-style: "linux"
+ * End:
+ */

daemons/dmeventd/mktestdevices

@@ -0,0 +1,12 @@
+#!/bin/sh
+#
+# Create test devices for dmeventd
+#
+
+trap "rm -f /tmp/tmp.$$" 0 1 2 3 15
+
+echo "0 1024 zero" > /tmp/tmp.$$
+dmsetup create test /tmp/tmp.$$
+dmsetup create test1 /tmp/tmp.$$
+
+kill -15 $$

daemons/dmeventd/noop.c

@@ -0,0 +1,39 @@
+/*
+ * Copyright (C) 2005 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of the device-mapper userspace tools.
+ *
+ * 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 "libdm-event.h"
+
+#include <stdio.h>
+
+void process_event(char *device, enum event_type event)
+{
+	fprintf(stderr, "[%s] %s(%d) - Device: %s, Event %d\n",
+		__FILE__, __func__, __LINE__, device, event);
+}
+
+int register_device(char *device)
+{
+	fprintf(stderr, "[%s] %s(%d) - Device: %s\n",
+		__FILE__, __func__, __LINE__, device);
+
+	return 1;
+}
+
+int unregister_device(char *device)
+{
+	fprintf(stderr, "[%s] %s(%d) - Device: %s\n",
+		__FILE__, __func__, __LINE__, device);
+
+	return 1;
+}

doc/Makefile.in

@@ -18,7 +18,7 @@ VPATH = @srcdir@
 CONFSRC=example.conf
 CONFDEST=lvm.conf
 
-include ../make.tmpl
+include $(top_srcdir)/make.tmpl
 
 install:
 	@if [ ! -e $(confdir)/$(CONFDEST) ]; then \

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).
 
@@ -252,11 +258,10 @@ activation {
 #
 # metadata {
     # Default number of copies of metadata to hold on each PV.  0, 1 or 2.
-    # It's best to leave this at 2.
-    # You might want to override it from the command line with 0 or 1 
+    # You might want to override it from the command line with 0 
     # when running pvcreate on new PVs which are to be added to large VGs.
 
-    # pvmetadatacopies = 2
+    # pvmetadatacopies = 1
 
     # Approximate default size of on-disk metadata areas in sectors.
     # You should increase this if you have large volume groups or

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 +31,8 @@
 ../lib/log/log.h
 ../lib/metadata/lv_alloc.h
 ../lib/metadata/metadata.h
-../lib/metadata/segtypes.h
+../lib/metadata/pv_alloc.h
+../lib/metadata/segtype.h
 ../lib/mm/dbg_malloc.h
 ../lib/mm/memlock.h
 ../lib/mm/pool.h
@@ -46,3 +48,4 @@
 ../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 \
@@ -69,12 +70,14 @@ SOURCES =\
 	metadata/merge.c \
 	metadata/metadata.c \
 	metadata/mirror.c \
+	metadata/pv_manip.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 +107,14 @@ 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 +123,6 @@ 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 \
@@ -136,5 +143,5 @@ LIB_STATIC = liblvm.a
 
 $(SUBDIRS): $(LIB_STATIC)
 
-include ../make.tmpl
+include $(top_srcdir)/make.tmpl
 

lib/activate/activate.c

@@ -78,12 +78,13 @@ int target_present(const char *target_name)
 {
 	return 0;
 }
-int lv_info(const struct logical_volume *lv, struct lvinfo *info)
+int lv_info(const struct logical_volume *lv, struct lvinfo *info,
+	    int with_open_count)
 {
 	return 0;
 }
 int lv_info_by_lvid(struct cmd_context *cmd, const char *lvid_s,
-		    struct lvinfo *info)
+		    struct lvinfo *info, int with_open_count)
 {
 	return 0;
 }
@@ -333,7 +334,7 @@ int target_present(const char *target_name)
  * Returns 1 if info structure populated, else 0 on failure.
  */
 static int _lv_info(const struct logical_volume *lv, int mknodes,
-		    struct lvinfo *info)
+		    struct lvinfo *info, int with_open_count)
 {
 	int r;
 	struct dev_manager *dm;
@@ -347,7 +348,7 @@ static int _lv_info(const struct logical_volume *lv, int mknodes,
 		return 0;
 	}
 
-	if (!(r = dev_manager_info(dm, lv, mknodes, &dminfo)))
+	if (!(r = dev_manager_info(dm, lv, mknodes, with_open_count, &dminfo)))
 		stack;
 
 	info->exists = dminfo.exists;
@@ -361,20 +362,21 @@ static int _lv_info(const struct logical_volume *lv, int mknodes,
 	return r;
 }
 
-int lv_info(const struct logical_volume *lv, struct lvinfo *info)
+int lv_info(const struct logical_volume *lv, struct lvinfo *info,
+	    int with_open_count)
 {
-	return _lv_info(lv, 0, info);
+	return _lv_info(lv, 0, info, with_open_count);
 }
 
 int lv_info_by_lvid(struct cmd_context *cmd, const char *lvid_s,
-		    struct lvinfo *info)
+		    struct lvinfo *info, int with_open_count)
 {
 	struct logical_volume *lv;
 
 	if (!(lv = lv_from_lvid(cmd, lvid_s)))
 		return 0;
 
-	return _lv_info(lv, 0, info);
+	return _lv_info(lv, 0, info, with_open_count);
 }
 
 /*
@@ -412,7 +414,7 @@ int lv_mirror_percent(struct logical_volume *lv, int wait, float *percent,
 	if (!activation())
 		return 0;
 
-	if (!lv_info(lv, &info)) {
+	if (!lv_info(lv, &info, 0)) {
 		stack;
 		return 0;
 	}
@@ -437,7 +439,7 @@ static int _lv_active(struct logical_volume *lv)
 {
 	struct lvinfo info;
 
-	if (!lv_info(lv, &info)) {
+	if (!lv_info(lv, &info, 0)) {
 		stack;
 		return -1;
 	}
@@ -449,7 +451,7 @@ static int _lv_open_count(struct logical_volume *lv)
 {
 	struct lvinfo info;
 
-	if (!lv_info(lv, &info)) {
+	if (!lv_info(lv, &info, 1)) {
 		stack;
 		return -1;
 	}
@@ -566,7 +568,7 @@ static int _lv_suspend(struct cmd_context *cmd, const char *lvid_s,
 		return 1;
 	}
 
-	if (!lv_info(lv, &info)) {
+	if (!lv_info(lv, &info, 0)) {
 		stack;
 		return 0;
 	}
@@ -612,7 +614,7 @@ static int _lv_resume(struct cmd_context *cmd, const char *lvid_s,
 		return 1;
 	}
 
-	if (!lv_info(lv, &info)) {
+	if (!lv_info(lv, &info, 0)) {
 		stack;
 		return 0;
 	}
@@ -657,7 +659,7 @@ int lv_deactivate(struct cmd_context *cmd, const char *lvid_s)
 		return 1;
 	}
 
-	if (!lv_info(lv, &info)) {
+	if (!lv_info(lv, &info, 1)) {
 		stack;
 		return 0;
 	}
@@ -726,7 +728,7 @@ static int _lv_activate(struct cmd_context *cmd, const char *lvid_s, int filter)
 		return 1;
 	}
 
-	if (!lv_info(lv, &info)) {
+	if (!lv_info(lv, &info, 0)) {
 		stack;
 		return 0;
 	}
@@ -765,7 +767,7 @@ int lv_mknodes(struct cmd_context *cmd, const struct logical_volume *lv)
 		return r;
 	}
 
-	if (!_lv_info(lv, 1, &info)) {
+	if (!_lv_info(lv, 1, &info, 0)) {
 		stack;
 		return 0;
 	}

lib/activate/activate.h

@@ -55,9 +55,10 @@ int lv_mknodes(struct cmd_context *cmd, const struct logical_volume *lv);
 /*
  * Returns 1 if info structure has been populated, else 0.
  */
-int lv_info(const struct logical_volume *lv, struct lvinfo *info);
+int lv_info(const struct logical_volume *lv, struct lvinfo *info,
+	    int with_open_count);
 int lv_info_by_lvid(struct cmd_context *cmd, const char *lvid_s,
-		    struct lvinfo *info);
+		    struct lvinfo *info, int with_open_count);
 
 /*
  * Returns 1 if activate_lv has been set: 1 = activate; 0 = don't.

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;
@@ -294,7 +211,8 @@ static struct dm_task *_setup_task(const char *name, const char *uuid,
 }
 
 static int _info_run(const char *name, const char *uuid, struct dm_info *info,
-		     int mknodes, struct pool *mem, char **uuid_out)
+		     int mknodes, int with_open_count, struct pool *mem,
+		     char **uuid_out)
 {
 	int r = 0;
 	struct dm_task *dmt;
@@ -308,6 +226,10 @@ static int _info_run(const char *name, const char *uuid, struct dm_info *info,
 		return 0;
 	}
 
+	if (!with_open_count)
+		if (!dm_task_no_open_count(dmt))
+			log_error("Failed to disable open_count");
+
 	if (!dm_task_run(dmt)) {
 		stack;
 		goto out;
@@ -333,14 +255,17 @@ static int _info_run(const char *name, const char *uuid, struct dm_info *info,
 }
 
 static int _info(const char *name, const char *uuid, int mknodes,
-		 struct dm_info *info, struct pool *mem, char **uuid_out)
+		 int with_open_count, struct dm_info *info,
+		 struct pool *mem, char **uuid_out)
 {
 	if (!mknodes && uuid && *uuid &&
-	    _info_run(NULL, uuid, info, 0, mem, uuid_out) && info->exists)
+	    _info_run(NULL, uuid, info, 0, with_open_count, mem, uuid_out) &&
+	    	      info->exists)
 		return 1;
 
 	if (name)
-		return _info_run(name, NULL, info, mknodes, mem, uuid_out);
+		return _info_run(name, NULL, info, mknodes, with_open_count,
+				 mem, uuid_out);
 
 	return 0;
 }
@@ -362,6 +287,9 @@ static int _status_run(const char *name, const char *uuid,
 		return 0;
 	}
 
+	if (!dm_task_no_open_count(dmt))
+		log_error("Failed to disable open_count");
+
 	if (!dm_task_run(dmt)) {
 		stack;
 		goto out;
@@ -440,6 +368,9 @@ static int _percent_run(struct dev_manager *dm, const char *name,
 		return 0;
 	}
 
+	if (!dm_task_no_open_count(dmt))
+		log_error("Failed to disable open_count");
+
 	if (!dm_task_run(dmt)) {
 		stack;
 		goto out;
@@ -519,10 +450,16 @@ 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 +474,16 @@ static int _rename(struct dev_layer *dl, char *newname)
 		goto out;
 	}
 
-	if (!(r = dm_task_run(dmt)))
+	if (!dm_task_no_open_count(dmt))
+		log_error("Failed to disable open_count");
+
+	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;
 
@@ -563,6 +505,9 @@ static int _suspend_or_resume(const char *name, action_t suspend)
 		return 0;
 	}
 
+	if (!dm_task_no_open_count(dmt))
+		log_error("Failed to disable open_count");
+
 	if (!(r = dm_task_run(dmt)))
 		log_error("Couldn't %s device '%s'", sus ? "suspend" : "resume",
 			  name);
@@ -654,6 +599,9 @@ static int _load(struct dev_manager *dm, struct dev_layer *dl, int task)
 			log_very_verbose("Activating %s read-only", dl->name);
 	}
 
+	if (!dm_task_no_open_count(dmt))
+		log_error("Failed to disable open_count");
+
 	if (!(r = dm_task_run(dmt))) {
 		log_error("Couldn't load device '%s'.", dl->name);
 		if ((dl->lv->minor >= 0 || dl->lv->major >= 0) &&
@@ -710,6 +658,9 @@ static int _remove(struct dev_layer *dl)
 		return 0;
 	}
 
+	if (!dm_task_no_open_count(dmt))
+		log_error("Failed to disable open_count");
+
 	/* Suppress error message if it's still in use - we'll log it later */
 	log_suppress(1);
 
@@ -776,6 +727,25 @@ static int _emit_target_line(struct dev_manager *dm, struct dm_task *dmt,
 	return 1;
 }
 
+int compose_log_line(struct dev_manager *dm, struct lv_segment *seg,
+		     char *params, size_t paramsize, int *pos, int areas,
+		     uint32_t region_size)
+{
+	int tw;
+
+	tw = lvm_snprintf(params, paramsize, "core 1 %u %u ",
+			  region_size, areas);
+
+	if (tw < 0) {
+		stack;
+		return -1;
+	}
+
+	*pos += tw;
+
+	return 1;
+}
+
 int compose_areas_line(struct dev_manager *dm, struct lv_segment *seg,
 		       char *params, size_t paramsize, int *pos, int start_area,
 		       int areas)
@@ -924,21 +894,23 @@ static int _populate_snapshot(struct dev_manager *dm,
 {
 	char *origin, *cow;
 	char params[PATH_MAX * 2 + 32];
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct dev_layer *dlo, *dlc;
 	char devbufo[10], devbufc[10];
+	uint64_t size;
 
-	if (!(s = find_cow(dl->lv))) {
+	if (!(snap_seg = find_cow(dl->lv))) {
 		log_error("Couldn't find snapshot for '%s'.", dl->lv->name);
 		return 0;
 	}
 
-	if (!(origin = _build_dlid(dm->mem, s->origin->lvid.s, "real"))) {
+	if (!(origin = _build_dlid(dm->mem, snap_seg->origin->lvid.s,
+				   "real"))) {
 		stack;
 		return 0;
 	}
 
-	if (!(cow = _build_dlid(dm->mem, s->cow->lvid.s, "cow"))) {
+	if (!(cow = _build_dlid(dm->mem, snap_seg->cow->lvid.s, "cow"))) {
 		stack;
 		return 0;
 	}
@@ -957,27 +929,27 @@ static int _populate_snapshot(struct dev_manager *dm,
 	if (!dm_format_dev(devbufo, sizeof(devbufo), dlo->info.major,
 			   dlo->info.minor)) {
 		log_error("Couldn't create origin device parameters for '%s'.",
-			  s->origin->name);
+			  snap_seg->origin->name);
 		return 0;
 	}
 
 	if (!dm_format_dev(devbufc, sizeof(devbufc), dlc->info.major,
 			   dlc->info.minor)) {
 		log_error("Couldn't create cow device parameters for '%s'.",
-			  s->cow->name);
+			  snap_seg->cow->name);
 		return 0;
 	}
 
 	if (lvm_snprintf(params, sizeof(params), "%s %s P %d",
-			 devbufo, devbufc, s->chunk_size) == -1) {
+			 devbufo, devbufc, snap_seg->chunk_size) == -1) {
 		stack;
 		return 0;
 	}
 
-	log_debug("Adding target: 0 %" PRIu64 " snapshot %s",
-		  s->origin->size, params);
-	if (!dm_task_add_target
-	    (dmt, UINT64_C(0), s->origin->size, "snapshot", params)) {
+	size = (uint64_t) snap_seg->len * snap_seg->origin->vg->extent_size;
+
+	log_debug("Adding target: 0 %" PRIu64 " snapshot %s", size, params);
+	if (!dm_task_add_target(dmt, UINT64_C(0), size, "snapshot", params)) {
 		stack;
 		return 0;
 	}
@@ -994,7 +966,7 @@ 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;
 	}
@@ -1045,14 +1017,14 @@ void dev_manager_destroy(struct dev_manager *dm)
 }
 
 int dev_manager_info(struct dev_manager *dm, const struct logical_volume *lv,
-		     int mknodes, struct dm_info *info)
+		     int mknodes, int with_open_count, struct dm_info *info)
 {
 	char *name;
 
 	/*
 	 * 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;
 	}
@@ -1061,7 +1033,8 @@ int dev_manager_info(struct dev_manager *dm, const struct logical_volume *lv,
 	 * Try and get some info on this device.
 	 */
 	log_debug("Getting device info for %s", name);
-	if (!_info(name, lv->lvid.s, mknodes, info, NULL, NULL)) {
+	if (!_info(name, lv->lvid.s, mknodes, with_open_count, info, NULL,
+		   NULL)) {
 		stack;
 		return 0;
 	}
@@ -1077,7 +1050,7 @@ 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 +1082,7 @@ 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;
 	}
@@ -1140,7 +1113,7 @@ static struct dev_layer *_create_dev(struct dev_manager *dm, char *name,
 	dl->name = name;
 
 	log_debug("Getting device info for %s", dl->name);
-	if (!_info(dl->name, dlid, 0, &dl->info, dm->mem, &uuid)) {
+	if (!_info(dl->name, dlid, 0, 0, &dl->info, dm->mem, &uuid)) {
 		stack;
 		return NULL;
 	}
@@ -1173,7 +1146,7 @@ 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;
 	}
@@ -1321,7 +1294,7 @@ static int _expand_origin_real(struct dev_manager *dm,
 static int _expand_origin(struct dev_manager *dm, struct logical_volume *lv)
 {
 	struct logical_volume *active;
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct list *sh;
 
 	/*
@@ -1330,7 +1303,7 @@ static int _expand_origin(struct dev_manager *dm, struct logical_volume *lv)
 	 */
 	list_iterate(sh, &dm->active_list) {
 		active = list_item(sh, struct lv_list)->lv;
-		if ((s = find_cow(active)) && (s->origin == lv))
+		if ((snap_seg = find_cow(active)) && (snap_seg->origin == lv))
 			return _expand_origin_real(dm, lv);
 	}
 
@@ -1341,7 +1314,7 @@ static int _expand_origin(struct dev_manager *dm, struct logical_volume *lv)
 }
 
 static int _expand_snapshot(struct dev_manager *dm, struct logical_volume *lv,
-			    struct snapshot *s)
+			    struct lv_segment *snap_seg)
 {
 	/*
 	 * snapshot(org, cow)
@@ -1378,13 +1351,15 @@ static int _expand_snapshot(struct dev_manager *dm, struct logical_volume *lv,
 
 	/* add the dependency on the real origin device */
 	if (!str_list_add(dm->mem, &dl->pre_create,
-			  _build_dlid(dm->mem, s->origin->lvid.s, "real"))) {
+			  _build_dlid(dm->mem, snap_seg->origin->lvid.s,
+				      "real"))) {
 		stack;
 		return 0;
 	}
 
 	/* add the dependency on the visible origin device */
-	if (!str_list_add(dm->mem, &dl->pre_suspend, s->origin->lvid.s)) {
+	if (!str_list_add(dm->mem, &dl->pre_suspend,
+			  snap_seg->origin->lvid.s)) {
 		stack;
 		return 0;
 	}
@@ -1397,13 +1372,13 @@ static int _expand_snapshot(struct dev_manager *dm, struct logical_volume *lv,
  */
 static int _expand_lv(struct dev_manager *dm, struct logical_volume *lv)
 {
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 
 	/*
 	 * FIXME: this doesn't cope with recursive snapshots yet.
 	 */
-	if ((s = find_cow(lv)))
-		return _expand_snapshot(dm, lv, s);
+	if ((snap_seg = find_cow(lv)))
+		return _expand_snapshot(dm, lv, snap_seg);
 
 	else if (lv_is_origin(lv))
 		return _expand_origin(dm, lv);
@@ -1492,6 +1467,8 @@ static int _mark_lvs(struct dev_manager *dm, struct list *lvs, int flag)
 
 	list_iterate(lvh, lvs) {
 		lv = list_item(lvh, struct lv_list)->lv;
+		if (lv->status & SNAPSHOT)
+			continue;
 
 		if (!(dl = _lookup(dm, lv->lvid.s, NULL))) {
 			stack;
@@ -1618,11 +1595,11 @@ 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;
 			}
@@ -1653,14 +1630,16 @@ static int _create_rec(struct dev_manager *dm, struct dev_layer *dl)
 static int _build_all_layers(struct dev_manager *dm, struct volume_group *vg)
 {
 	struct list *lvh;
-	struct logical_volume *lvt;
+	struct logical_volume *lv;
 
 	/*
 	 * Build layers for complete vg.
 	 */
 	list_iterate(lvh, &vg->lvs) {
-		lvt = list_item(lvh, struct lv_list)->lv;
-		if (!_expand_lv(dm, lvt)) {
+		lv = list_item(lvh, struct lv_list)->lv;
+		if (lv->status & SNAPSHOT)
+			continue;
+		if (!_expand_lv(dm, lv)) {
 			stack;
 			return 0;
 		}
@@ -1962,12 +1941,14 @@ static int _add_lvs(struct pool *mem,
 		    struct list *head, struct logical_volume *origin)
 {
 	struct logical_volume *lv;
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct list *lvh;
 
 	list_iterate(lvh, &origin->vg->lvs) {
 		lv = list_item(lvh, struct lv_list)->lv;
-		if ((s = find_cow(lv)) && s->origin == origin)
+		if (lv->status & SNAPSHOT)
+			continue;
+		if ((snap_seg = find_cow(lv)) && snap_seg->origin == origin)
 			if (!_add_lv(mem, head, lv))
 				return 0;
 	}
@@ -1992,7 +1973,7 @@ static void _remove_lv(struct list *head, struct logical_volume *lv)
 static int _remove_lvs(struct dev_manager *dm, struct logical_volume *lv)
 {
 	struct logical_volume *active, *old_origin;
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct list *sh, *active_head;
 
 	active_head = &dm->active_list;
@@ -2000,22 +1981,23 @@ static int _remove_lvs(struct dev_manager *dm, struct logical_volume *lv)
 	/* Remove any snapshots with given origin */
 	list_iterate(sh, active_head) {
 		active = list_item(sh, struct lv_list)->lv;
-		if ((s = find_cow(active)) && s->origin == lv) {
+		if ((snap_seg = find_cow(active)) && snap_seg->origin == lv) {
 			_remove_lv(active_head, active);
 		}
 	}
 
 	_remove_lv(active_head, lv);
 
-	if (!(s = find_cow(lv)))
+	if (!(snap_seg = find_cow(lv)))
 		return 1;
 
-	old_origin = s->origin;
+	old_origin = snap_seg->origin;
 
 	/* Was this the last active snapshot with this origin? */
 	list_iterate(sh, active_head) {
 		active = list_item(sh, struct lv_list)->lv;
-		if ((s = find_cow(active)) && s->origin == old_origin) {
+		if ((snap_seg = find_cow(active)) &&
+		    snap_seg->origin == old_origin) {
 			return 1;
 		}
 	}
@@ -2027,7 +2009,7 @@ static int _remove_suspended_lvs(struct dev_manager *dm,
 				 struct logical_volume *lv)
 {
 	struct logical_volume *suspended;
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct list *sh, *suspend_head;
 
 	suspend_head = &dm->suspend_list;
@@ -2035,7 +2017,8 @@ static int _remove_suspended_lvs(struct dev_manager *dm,
 	/* Remove from list any snapshots with given origin */
 	list_iterate(sh, suspend_head) {
 		suspended = list_item(sh, struct lv_list)->lv;
-		if ((s = find_cow(suspended)) && s->origin == lv) {
+		if ((snap_seg = find_cow(suspended)) &&
+		    snap_seg->origin == lv) {
 			_remove_lv(suspend_head, suspended);
 		}
 	}
@@ -2121,6 +2104,8 @@ static int _fill_in_active_list(struct dev_manager *dm, struct volume_group *vg)
 
 	list_iterate(lvh, &vg->lvs) {
 		lv = list_item(lvh, struct lv_list)->lv;
+		if (lv->status & SNAPSHOT)
+			continue;
 
 		if (!(dlid = _build_dlid(dm->mem, lv->lvid.s, NULL))) {
 			stack;
@@ -2227,8 +2212,8 @@ 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/dev_manager.h

@@ -36,7 +36,7 @@ void dev_manager_exit(void);
  * unsuspended until the snapshot is also created.)
  */
 int dev_manager_info(struct dev_manager *dm, const struct logical_volume *lv,
-		     int mknodes, struct dm_info *info);
+		     int mknodes, int with_open_count, struct dm_info *info);
 int dev_manager_snapshot_percent(struct dev_manager *dm,
 				 struct logical_volume *lv, float *percent);
 int dev_manager_mirror_percent(struct dev_manager *dm,

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 (dir_exists(vg_path) && 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/activate/targets.h

@@ -19,7 +19,12 @@
 struct dev_manager;
 struct lv_segment;
 
-int compose_areas_line(struct dev_manager *dm, struct lv_segment *seg,                                 char *params, size_t paramsize, int *pos,
-		                        int start_area, int areas);
+int compose_areas_line(struct dev_manager *dm, struct lv_segment *seg,
+		       char *params, size_t paramsize, int *pos,
+		       int start_area, int areas);
+
+int compose_log_line(struct dev_manager *dm, struct lv_segment *seg,
+		     char *params, size_t paramsize, int *pos, int areas,
+		     uint32_t region_size);
 
 #endif

lib/cache/lvmcache.c

@@ -104,10 +104,30 @@ struct lvmcache_vginfo *vginfo_from_vgname(const char *vgname)
 const struct format_type *fmt_from_vgname(const char *vgname)
 {
 	struct lvmcache_vginfo *vginfo;
+	struct label *label;
+	struct list *ih, *devh, *tmp;
+	struct list devs;
+	struct device_list *devl;
 
 	if (!(vginfo = vginfo_from_vgname(vgname)))
 		return NULL;
 
+	/* This function is normally called before reading metadata so
+ 	 * we check cached labels here. Unfortunately vginfo is volatile. */
+	list_init(&devs);
+	list_iterate(ih, &vginfo->infos) {
+		devl = malloc(sizeof(*devl));
+		devl->dev = list_item(ih, struct lvmcache_info)->dev;
+		list_add(&devs, &devl->list);
+	}
+
+	list_iterate_safe(devh, tmp, &devs) {
+		devl = list_item(devh, struct device_list);
+		label_read(devl->dev, &label);
+		list_del(&devl->list);
+		free(devl);
+	}
+
 	return vginfo->fmt;
 }
 
@@ -188,7 +208,7 @@ int lvmcache_label_scan(struct cmd_context *cmd, int full_scan)
 		goto out;
 	}
 
-	if (!(iter = dev_iter_create(cmd->filter))) {
+	if (!(iter = dev_iter_create(cmd->filter, (full_scan == 2) ? 1: 0))) {
 		log_error("dev_iter creation failed");
 		goto out;
 	}
@@ -266,7 +286,7 @@ struct device *device_from_pvid(struct cmd_context *cmd, struct id *pvid)
 	if (memlock())
 		return NULL;
 
-	lvmcache_label_scan(cmd, 1);
+	lvmcache_label_scan(cmd, 2);
 
 	/* Try again */
 	if ((info = info_from_pvid((char *) pvid))) {
@@ -403,6 +423,9 @@ int lvmcache_update_vgname(struct lvmcache_info *info, const char *vgname)
 	/* FIXME Check consistency of list! */
 	vginfo->fmt = info->fmt;
 
+	log_debug("lvmcache: %s now %s%s", dev_name(info->dev),
+		  *vgname ? "in VG " : "orphaned", vgname);
+
 	return 1;
 }
 
@@ -551,6 +574,8 @@ static void _lvmcache_destroy_lockname(int present)
 
 void lvmcache_destroy(void)
 {
+	log_verbose("Wiping internal VG cache");
+
 	_has_scanned = 0;
 
 	if (_vgid_hash) {

lib/cache/lvmcache.h

@@ -36,9 +36,10 @@ struct volume_group;
 struct lvmcache_vginfo {
 	struct list list;	/* Join these vginfos together */
 	struct list infos;	/* List head for lvmcache_infos */
+	const struct format_type *fmt;
 	char *vgname;		/* "" == orphan */
 	char vgid[ID_LEN + 1];
-	const struct format_type *fmt;
+	char _padding[7];
 };
 
 struct lvmcache_info {
@@ -56,7 +57,8 @@ struct lvmcache_info {
 int lvmcache_init(void);
 void lvmcache_destroy(void);
 
-/* Set full_scan to 1 to reread every filtered device label */
+/* Set full_scan to 1 to reread every filtered device label or
+ * 2 to rescan /dev for new devices */
 int lvmcache_label_scan(struct cmd_context *cmd, int full_scan);
 
 /* Add/delete a device */

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
@@ -181,6 +181,12 @@ static int _process_config(struct cmd_context *cmd)
 		return 0;
 	}
 
+	if (*cmd->proc_dir && !dir_exists(cmd->proc_dir)) {
+		log_error("Warning: proc dir %s not found - some checks will be bypassed",
+			  cmd->proc_dir);
+		cmd->proc_dir[0] = '\0';
+	}
+
 	/* activation? */
 	cmd->default_settings.activation = find_config_int(cmd->cft->root,
 							   "global/activation",
@@ -519,8 +525,8 @@ 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 +543,7 @@ 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 +805,7 @@ 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);
@@ -824,10 +831,15 @@ struct cmd_context *create_toolcontext(struct arg *the_args)
 		goto error;
 
 	/* Create system directory if it doesn't already exist */
-	if (*cmd->sys_dir && !create_dir(cmd->sys_dir))
+	if (*cmd->sys_dir && !create_dir(cmd->sys_dir)) {
+		log_error("Failed to create LVM2 system dir for metadata backups, config "
+			  "files and internal cache.");
+		log_error("Set environment variable LVM_SYSTEM_DIR to alternative location "
+			  "or empty string.");
 		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 +870,7 @@ 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/commands/toolcontext.h

@@ -32,17 +32,15 @@ struct config_info {
 	int syslog;
 	int activation;
 	int suffix;
-	uint64_t unit_factor;
-	char unit_type;
-	const char *msg_prefix;
-	int cmd_name;		/* Show command name? */
-
 	int archive;		/* should we archive ? */
 	int backup;		/* should we backup ? */
-
+	const char *msg_prefix;
 	struct format_type *fmt;
-
+	uint64_t unit_factor;
+	int cmd_name;		/* Show command name? */
 	mode_t umask;
+	char unit_type;
+	char _padding[1];
 };
 
 struct config_tree;

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

@@ -95,15 +95,18 @@
 #define DEFAULT_VGS_COLS "vg_name,pv_count,lv_count,snap_count,vg_attr,vg_size,vg_free"
 #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_PVSEGS_COLS "pv_name,vg_name,pv_fmt,pv_attr,pv_size,pv_free,pvseg_start,pvseg_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_PVSEGS_COLS_VERB "pv_name,vg_name,pv_fmt,pv_attr,pv_size,pv_free,pvseg_start,pvseg_size"
 
 #define DEFAULT_LVS_SORT "vg_name,lv_name"
 #define DEFAULT_VGS_SORT "vg_name"
 #define DEFAULT_PVS_SORT "pv_name"
 #define DEFAULT_SEGS_SORT "vg_name,lv_name,seg_start"
+#define DEFAULT_PVSEGS_SORT "pv_name,pvseg_start"
 
 #endif				/* _LVM_DEFAULTS_H */

lib/datastruct/hash.c

@@ -19,7 +19,8 @@
 struct hash_node {
 	struct hash_node *next;
 	void *data;
-	char key[1];
+	int keylen;
+	char key[0];
 };
 
 struct hash_table {
@@ -56,22 +57,23 @@ static unsigned char _nums[] = {
 	209
 };
 
-static struct hash_node *_create_node(const char *str)
+static struct hash_node *_create_node(const char *str, int len)
 {
-	/* remember sizeof(n) includes an extra char from key[1],
-	   so not adding 1 to the strlen as you would expect */
-	struct hash_node *n = dbg_malloc(sizeof(*n) + strlen(str));
+	struct hash_node *n = dbg_malloc(sizeof(*n) + len);
 
-	if (n)
-		strcpy(n->key, str);
+	if (n) {
+		memcpy(n->key, str, len);
+		n->keylen = len;
+	}
 
 	return n;
 }
 
-static unsigned _hash(const char *str)
+static unsigned _hash(const char *str, uint32_t len)
 {
-	unsigned long h = 0, g;
-	while (*str) {
+	unsigned long h = 0, g, i;
+
+	for (i = 0; i < len; i++) {
 		h <<= 4;
 		h += _nums[(int) *str++];
 		g = h & ((unsigned long) 0xf << 16u);
@@ -80,6 +82,7 @@ static unsigned _hash(const char *str)
 			h ^= g >> 5u;
 		}
 	}
+
 	return h;
 }
 
@@ -134,32 +137,35 @@ void hash_destroy(struct hash_table *t)
 	dbg_free(t);
 }
 
-static struct hash_node **_find(struct hash_table *t, const char *key)
+static inline struct hash_node **_find(struct hash_table *t, const char *key,
+				       uint32_t len)
 {
-	unsigned h = _hash(key) & (t->num_slots - 1);
+	unsigned h = _hash(key, len) & (t->num_slots - 1);
 	struct hash_node **c;
 
 	for (c = &t->slots[h]; *c; c = &((*c)->next))
-		if (!strcmp(key, (*c)->key))
+		if (!memcmp(key, (*c)->key, len))
 			break;
 
 	return c;
 }
 
-void *hash_lookup(struct hash_table *t, const char *key)
+void *hash_lookup_binary(struct hash_table *t, const char *key,
+			 uint32_t len)
 {
-	struct hash_node **c = _find(t, key);
+	struct hash_node **c = _find(t, key, len);
 	return *c ? (*c)->data : 0;
 }
 
-int hash_insert(struct hash_table *t, const char *key, void *data)
+int hash_insert_binary(struct hash_table *t, const char *key,
+		       uint32_t len, void *data)
 {
-	struct hash_node **c = _find(t, key);
+	struct hash_node **c = _find(t, key, len);
 
 	if (*c)
 		(*c)->data = data;
 	else {
-		struct hash_node *n = _create_node(key);
+		struct hash_node *n = _create_node(key, len);
 
 		if (!n)
 			return 0;
@@ -173,9 +179,10 @@ int hash_insert(struct hash_table *t, const char *key, void *data)
 	return 1;
 }
 
-void hash_remove(struct hash_table *t, const char *key)
+void hash_remove_binary(struct hash_table *t, const char *key,
+			uint32_t len)
 {
-	struct hash_node **c = _find(t, key);
+	struct hash_node **c = _find(t, key, len);
 
 	if (*c) {
 		struct hash_node *old = *c;
@@ -185,6 +192,21 @@ void hash_remove(struct hash_table *t, const char *key)
 	}
 }
 
+void *hash_lookup(struct hash_table *t, const char *key)
+{
+	return hash_lookup_binary(t, key, strlen(key) + 1);
+}
+
+int hash_insert(struct hash_table *t, const char *key, void *data)
+{
+	return hash_insert_binary(t, key, strlen(key) + 1, data);
+}
+
+void hash_remove(struct hash_table *t, const char *key)
+{
+	hash_remove_binary(t, key, strlen(key) + 1);
+}
+
 unsigned hash_get_num_entries(struct hash_table *t)
 {
 	return t->num_nodes;
@@ -235,6 +257,6 @@ struct hash_node *hash_get_first(struct hash_table *t)
 
 struct hash_node *hash_get_next(struct hash_table *t, struct hash_node *n)
 {
-	unsigned h = _hash(n->key) & (t->num_slots - 1);
+	unsigned h = _hash(n->key, n->keylen) & (t->num_slots - 1);
 	return n->next ? n->next : _next_slot(t, h + 1);
 }

lib/datastruct/hash.h

@@ -26,10 +26,14 @@ void hash_destroy(struct hash_table *t);
 void hash_wipe(struct hash_table *t);
 
 void *hash_lookup(struct hash_table *t, const char *key);
-void *hash_lookup_fixed(struct hash_table *t, const char *key, uint32_t len);
 int hash_insert(struct hash_table *t, const char *key, void *data);
 void hash_remove(struct hash_table *t, const char *key);
 
+void *hash_lookup_binary(struct hash_table *t, const char *key, uint32_t len);
+int hash_insert_binary(struct hash_table *t, const char *key, uint32_t len,
+		       void *data);
+void hash_remove_binary(struct hash_table *t, const char *key, uint32_t len);
+
 unsigned hash_get_num_entries(struct hash_table *t);
 void hash_iter(struct hash_table *t, iterate_fn f);
 

lib/datastruct/list.h

@@ -18,17 +18,30 @@
 
 #include <assert.h>
 
+/*
+ * A list consists of a list head plus elements.
+ * Each element has 'next' and 'previous' pointers.
+ * The list head's pointers point to the first and the last element.
+ */
+
 struct list {
 	struct list *n, *p;
 };
 
+/*
+ * Initialise a list before use.
+ * The list head's next and previous pointers point back to itself.
+ */
 #define LIST_INIT(name)	struct list name = { &(name), &(name) }
-
 static inline void list_init(struct list *head)
 {
 	head->n = head->p = head;
 }
 
+/*
+ * Insert an element before 'head'.
+ * If 'head' is the list head, this adds an element to the end of the list.
+ */
 static inline void list_add(struct list *head, struct list *elem)
 {
 	assert(head->n);
@@ -40,6 +53,10 @@ static inline void list_add(struct list *head, struct list *elem)
 	head->p = elem;
 }
 
+/*
+ * Insert an element after 'head'.
+ * If 'head' is the list head, this adds an element to the front of the list.
+ */
 static inline void list_add_h(struct list *head, struct list *elem)
 {
 	assert(head->n);
@@ -51,53 +68,127 @@ static inline void list_add_h(struct list *head, struct list *elem)
 	head->n = elem;
 }
 
+/*
+ * Delete an element from its list.
+ * Note that this doesn't change the element itself - it may still be safe
+ * to follow its pointers.
+ */
 static inline void list_del(struct list *elem)
 {
 	elem->n->p = elem->p;
 	elem->p->n = elem->n;
 }
 
+/*
+ * Is the list empty?
+ */
 static inline int list_empty(struct list *head)
 {
 	return head->n == head;
 }
 
+/*
+ * Is this the first element of the list?
+ */
+static inline int list_start(struct list *head, struct list *elem)
+{
+	return elem->p == head;
+}
+
+/*
+ * Is this the last element of the list?
+ */
 static inline int list_end(struct list *head, struct list *elem)
 {
 	return elem->n == head;
 }
 
+/*
+ * Return the previous element of the list, or NULL if we've reached the start.
+ */
+static inline struct list *list_prev(struct list *head, struct list *elem)
+{
+	return (list_start(head, elem) ? NULL : elem->p);
+}
+
+/*
+ * Return the next element of the list, or NULL if we've reached the end.
+ */
 static inline struct list *list_next(struct list *head, struct list *elem)
 {
 	return (list_end(head, elem) ? NULL : elem->n);
 }
 
-#define list_item(v, t) \
-    ((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->list))
+/*
+ * Given the address v of an instance of 'struct list' called 'head' 
+ * contained in a structure of type t, return the containing structure.
+ */
+#define list_struct_base(v, t, head) \
+    ((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->head))
 
-#define list_struct_base(v, t, h) \
-    ((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->h))
+/*
+ * Given the address v of an instance of 'struct list list' contained in
+ * a structure of type t, return the containing structure.
+ */
+#define list_item(v, t) list_struct_base((v), t, list)
 
-/* Given a known element in a known structure, locate another */
+/*
+ * Given the address v of one known element e in a known structure of type t,
+ * return another element f.
+ */
 #define struct_field(v, t, e, f) \
     (((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->e))->f)
 
-/* Given a known element in a known structure, locate the list head */
+/*
+ * Given the address v of a known element e in a known structure of type t,
+ * return the list head 'list'
+ */
 #define list_head(v, t, e) struct_field(v, t, e, list)
 
+/*
+ * Set v to each element of a list in turn.
+ */
 #define list_iterate(v, head) \
 	for (v = (head)->n; v != head; v = v->n)
 
+/*
+ * Set v to each element in a list in turn, starting from the element 
+ * in front of 'start'.
+ * You can use this to 'unwind' a list_iterate and back out actions on
+ * already-processed elements.
+ * If 'start' is 'head' it walks the list backwards.
+ */
 #define list_uniterate(v, head, start) \
 	for (v = (start)->p; v != head; v = v->p)
 
+/*
+ * A safe way to walk a list and delete and free some elements along
+ * the way.
+ * t must be defined as a temporary variable of the same type as v.
+ */
 #define list_iterate_safe(v, t, head) \
 	for (v = (head)->n, t = v->n; v != head; v = t, t = v->n)
 
-#define list_iterate_items(v, head) \
-	for (v = list_item((head)->n, typeof(*v)); &v->list != (head); \
-	     v = list_item(v->list.n, typeof(*v)))
+/*
+ * Walk a list, setting 'v' in turn to the containing structure of each item.
+ * The containing structure should be the same type as 'v'.
+ * The 'struct list' variable within the containing structure is 'field'.
+ */
+#define list_iterate_items_gen(v, head, field) \
+	for (v = list_struct_base((head)->n, typeof(*v), field); \
+	     &v->field != (head); \
+	     v = list_struct_base(v->field.n, typeof(*v), field))
 
+/*
+ * Walk a list, setting 'v' in turn to the containing structure of each item.
+ * The containing structure should be the same type as 'v'.
+ * The list should be 'struct list list' within the containing structure.
+ */
+#define list_iterate_items(v, head) list_iterate_items_gen(v, (head), list)
+
+/*
+ * Return the number of elements in a list by walking it.
+ */
 static inline unsigned int list_size(const struct list *head)
 {
 	unsigned int s = 0;

lib/device/dev-cache.c

@@ -21,6 +21,7 @@
 #include "lvm-types.h"
 #include "btree.h"
 #include "filter.h"
+#include "filter-persistent.h"
 
 #include <unistd.h>
 #include <sys/param.h>
@@ -82,6 +83,7 @@ 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 +103,7 @@ 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);
@@ -184,7 +187,7 @@ static int _add_alias(struct device *dev, const char *path)
 	/* Is name already there? */
 	list_iterate(ah, &dev->aliases) {
 		if (!strcmp(list_item(ah, struct str_list)->str, path)) {
-			stack;
+			log_debug("%s: Already in device cache", path);
 			return 1;
 		}
 	}
@@ -353,11 +356,11 @@ static int _insert(const char *path, int rec)
 	return r;
 }
 
-static void _full_scan(void)
+static void _full_scan(int dev_scan)
 {
 	struct list *dh;
 
-	if (_cache.has_scanned)
+	if (_cache.has_scanned && !dev_scan)
 		return;
 
 	list_iterate(dh, &_cache.dirs) {
@@ -366,6 +369,7 @@ static void _full_scan(void)
 	};
 
 	_cache.has_scanned = 1;
+	init_full_scan_done(1);
 }
 
 int dev_cache_has_scanned(void)
@@ -377,17 +381,16 @@ void dev_cache_scan(int do_scan)
 {
 	if (!do_scan)
 		_cache.has_scanned = 1;
-	else {
-		_cache.has_scanned = 0;
-		_full_scan();
-	}
+	else
+		_full_scan(1);
 }
 
 int dev_cache_init(void)
 {
 	_cache.names = NULL;
+	_cache.has_scanned = 0;
 
-	if (!(_cache.mem = pool_create(10 * 1024))) {
+	if (!(_cache.mem = pool_create("dev_cache", 10 * 1024))) {
 		stack;
 		return 0;
 	}
@@ -538,7 +541,7 @@ struct device *dev_cache_get(const char *name, struct dev_filter *f)
 	return (d && (!f || f->passes_filter(f, d))) ? d : NULL;
 }
 
-struct dev_iter *dev_iter_create(struct dev_filter *f)
+struct dev_iter *dev_iter_create(struct dev_filter *f, int dev_scan)
 {
 	struct dev_iter *di = dbg_malloc(sizeof(*di));
 
@@ -547,7 +550,14 @@ struct dev_iter *dev_iter_create(struct dev_filter *f)
 		return NULL;
 	}
 
-	_full_scan();
+
+	if (dev_scan) {
+		/* Flag gets reset between each command */
+		if (!full_scan_done())
+			persistent_filter_wipe(f); /* Calls _full_scan(1) */
+	} else
+		_full_scan(0);
+
 	di->current = btree_first(_cache.devices);
 	di->filter = f;
 

lib/device/dev-cache.h

@@ -45,7 +45,7 @@ struct device *dev_cache_get(const char *name, struct dev_filter *f);
  * Object for iterating through the cache.
  */
 struct dev_iter;
-struct dev_iter *dev_iter_create(struct dev_filter *f);
+struct dev_iter *dev_iter_create(struct dev_filter *f, int dev_scan);
 void dev_iter_destroy(struct dev_iter *iter);
 struct device *dev_iter_get(struct dev_iter *iter);
 

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())
@@ -296,23 +330,57 @@ int dev_open_flags(struct device *dev, int flags, int direct, int quiet)
 	}
 
 #ifdef O_DIRECT_SUPPORT
-	if (direct)
-		flags |= O_DIRECT;
+	if (direct) {
+		if (!(dev->flags & DEV_O_DIRECT_TESTED))
+			dev->flags |= DEV_O_DIRECT;
+
+		if ((dev->flags & DEV_O_DIRECT))
+			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);
+#ifdef O_DIRECT_SUPPORT
+		if (direct && !(dev->flags & DEV_O_DIRECT_TESTED)) {
+			flags &= ~O_DIRECT;
+			if ((dev->fd = open(name, flags, 0777)) >= 0) {
+				dev->flags &= ~DEV_O_DIRECT;
+				log_debug("%s: Not using O_DIRECT", name);
+				goto opened;
+			}
+		}
+#endif
+		if (quiet)
+			log_sys_debug("open", name);
+		else
+			log_sys_error("open", name);
 		return 0;
 	}
 
-	dev->open_count = 1;
+#ifdef O_DIRECT_SUPPORT
+      opened:
+	if (direct)
+		dev->flags |= DEV_O_DIRECT_TESTED;
+#endif
+	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 +389,14 @@ 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%s", dev_name(dev),
+		  dev->flags & DEV_OPENED_RW ? "RW" : "RO",
+		  dev->flags & DEV_O_DIRECT ? " O_DIRECT" : "");
 
 	return 1;
 }
@@ -354,6 +424,7 @@ 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 +450,11 @@ 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 +486,10 @@ 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 +507,10 @@ 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 +525,10 @@ 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 +575,5 @@ 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 +22,9 @@
 #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 */
+#define DEV_O_DIRECT		0x00000010	/* Use O_DIRECT */
+#define DEV_O_DIRECT_TESTED	0x00000020	/* DEV_O_DIRECT is reliable */
 
 /*
  * All devices in LVM will be represented by one of these.
@@ -34,11 +37,13 @@ struct device {
 	/* private */
 	int fd;
 	int open_count;
+	int block_size;
 	uint32_t flags;
 	uint64_t end;
 	struct list open_list;
 
 	char pvid[ID_LEN + 1];
+	char _padding[7];
 };
 
 struct device_list {
@@ -55,13 +60,13 @@ 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 */
 int dev_open(struct device *dev);
 int dev_open_quiet(struct device *dev);
-int dev_open_flags(struct device *dev, int flags, int append, int quiet);
+int dev_open_flags(struct device *dev, int flags, int direct, int quiet);
 int dev_close(struct device *dev);
 int dev_close_immediate(struct device *dev);
 void dev_close_all(void);
@@ -89,15 +94,14 @@ 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
 
@@ -258,7 +258,7 @@ void pvdisplay_full(struct cmd_context *cmd, struct physical_volume *pv,
 		log_print("PV Size               %s" " / not usable %s",	/*  [LVM: %s]", */
 			  size,
 			  display_size(cmd, (pv->size -
-					     pv->pe_count * pv->pe_size),
+				       (uint64_t) pv->pe_count * pv->pe_size),
 				       SIZE_SHORT));
 
 	} else
@@ -317,7 +317,7 @@ void lvdisplay_colons(struct logical_volume *lv)
 {
 	int inkernel;
 	struct lvinfo info;
-	inkernel = lv_info(lv, &info) && info.exists;
+	inkernel = lv_info(lv, &info, 1) && info.exists;
 
 	log_print("%s%s/%s:%s:%d:%d:-1:%d:%" PRIu64 ":%d:-1:%d:%d:%d:%d",
 		  lv->vg->cmd->dev_dir,
@@ -337,10 +337,9 @@ int lvdisplay_full(struct cmd_context *cmd, struct logical_volume *lv,
 		   void *handle)
 {
 	struct lvinfo info;
-	int inkernel, snap_active;
+	int inkernel, snap_active = 0;
 	char uuid[64];
-	struct snapshot *snap = NULL;
-	struct list *slh, *snaplist;
+	struct lv_segment *snap_seg = NULL;
 	float snap_percent;	/* fused, fsize; */
 
 	if (!id_write_format(&lv->lvid.id[1], uuid, sizeof(uuid))) {
@@ -348,7 +347,7 @@ int lvdisplay_full(struct cmd_context *cmd, struct logical_volume *lv,
 		return 0;
 	}
 
-	inkernel = lv_info(lv, &info) && info.exists;
+	inkernel = lv_info(lv, &info, 1) && info.exists;
 
 	log_print("--- Logical volume ---");
 
@@ -364,27 +363,30 @@ int lvdisplay_full(struct cmd_context *cmd, struct logical_volume *lv,
 	if (lv_is_origin(lv)) {
 		log_print("LV snapshot status     source of");
 
-		snaplist = find_snapshots(lv);
-		list_iterate(slh, snaplist) {
-			snap = list_item(slh, struct snapshot_list)->snapshot;
-			snap_active = lv_snapshot_percent(snap->cow,
-							  &snap_percent);
-			if (!snap_active || snap_percent < 0 ||
-			    snap_percent >= 100) snap_active = 0;
+		list_iterate_items_gen(snap_seg, &lv->snapshot_segs,
+				       origin_list) {
+			if (inkernel &&
+			    (snap_active = lv_snapshot_percent(snap_seg->cow,
+							       &snap_percent)))
+				if (snap_percent < 0 || snap_percent >= 100)
+					snap_active = 0;
 			log_print("                       %s%s/%s [%s]",
 				  lv->vg->cmd->dev_dir, lv->vg->name,
-				  snap->cow->name,
+				  snap_seg->cow->name,
 				  (snap_active > 0) ? "active" : "INACTIVE");
 		}
-		snap = NULL;
-	} else if ((snap = find_cow(lv))) {
-		snap_active = lv_snapshot_percent(lv, &snap_percent);
-		if (!snap_active || snap_percent < 0 || snap_percent >= 100)
-			snap_active = 0;
+		snap_seg = NULL;
+	} else if ((snap_seg = find_cow(lv))) {
+		if (inkernel &&
+		    (snap_active = lv_snapshot_percent(snap_seg->cow,
+						       &snap_percent)))
+			if (snap_percent < 0 || snap_percent >= 100)
+				snap_active = 0;
+
 		log_print("LV snapshot status     %s destination for %s%s/%s",
 			  (snap_active > 0) ? "active" : "INACTIVE",
 			  lv->vg->cmd->dev_dir, lv->vg->name,
-			  snap->origin->name);
+			  snap_seg->origin->name);
 	}
 
 	if (inkernel && info.suspended)
@@ -402,15 +404,24 @@ int lvdisplay_full(struct cmd_context *cmd, struct logical_volume *lv,
 
 	log_print("LV Size                %s",
 		  display_size(cmd,
-			       snap ? snap->origin->size : lv->size,
+			       snap_seg ? snap_seg->origin->size : lv->size,
 			       SIZE_SHORT));
 
 	log_print("Current LE             %u",
-		  snap ? snap->origin->le_count : lv->le_count);
+		  snap_seg ? snap_seg->origin->le_count : lv->le_count);
 
-/********** FIXME allocation
-    log_print("Allocated LE           %u", lv->allocated_le);
-**********/
+	if (snap_seg) {
+		log_print("COW-table size         %s",
+			  display_size(cmd, (uint64_t) lv->size, SIZE_SHORT));
+		log_print("COW-table LE           %u", lv->le_count);
+
+		if (snap_active)
+			log_print("Allocated to snapshot  %.2f%% ", snap_percent);	
+
+		log_print("Snapshot chunk size    %s",
+			  display_size(cmd, (uint64_t) snap_seg->chunk_size,
+				       SIZE_SHORT));
+	}
 
 	log_print("Segments               %u", list_size(&lv->segments));
 
@@ -418,31 +429,6 @@ int lvdisplay_full(struct cmd_context *cmd, struct logical_volume *lv,
 	log_print("Stripe size (KByte)    %u", lv->stripesize / 2);
 ***********/
 
-	if (snap) {
-		if (snap_percent == -1)
-			snap_percent = 100;
-
-		log_print("Snapshot chunk size    %s",
-			  display_size(cmd, (uint64_t) snap->chunk_size,
-				       SIZE_SHORT));
-
-/*
-	size = display_size(lv->size, SIZE_SHORT);
-	sscanf(size, "%f", &fsize);
-	fused = fsize * snap_percent / 100;
-*/
-		log_print("Allocated to snapshot  %.2f%% ",	/* [%.2f/%s]", */
-			  snap_percent);	/*, fused, size); */
-		/* dbg_free(size); */
-	}
-
-/********** FIXME Snapshot
-	size = ???
-	log_print("Allocated to COW-table %s", size);
-	dbg_free(size);
-    }
-******************/
-
 	log_print("Allocation             %s", get_alloc_string(lv->alloc));
 	log_print("Read ahead sectors     %u", lv->read_ahead);
 
@@ -550,7 +536,7 @@ void vgdisplay_full(struct volume_group *vg)
 			  vg->status & SHARED ? "yes" : "no");
 	}
 	log_print("MAX LV                %u", vg->max_lv);
-	log_print("Cur LV                %u", vg->lv_count);
+	log_print("Cur LV                %u", vg->lv_count + vg->snapshot_count);
 	log_print("Open LV               %u", lvs_in_vg_opened(vg));
 /****** FIXME Max LV Size
       log_print ( "MAX LV Size           %s",
@@ -662,3 +648,22 @@ 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

@@ -53,7 +53,9 @@ int persistent_filter_wipe(struct dev_filter *f)
 {
 	struct pfilter *pf = (struct pfilter *) f->private;
 
+	log_verbose("Wiping cache of LVM-capable devices");
 	hash_wipe(pf->devices);
+
 	/* Trigger complete device scan */
 	dev_cache_scan(1);
 
@@ -212,13 +214,10 @@ 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

@@ -169,8 +169,10 @@ 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;
+	dev_t dev = { 0 };
 	int r = 1;
 
         if (!(dr = opendir(dir))) {
@@ -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 +65,61 @@ 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 */
+	{"gnbd", 1},		/* Network block device */
+	{"ramdisk", 1},		/* RAM disk */
 	{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, 1)) {
+		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 +129,31 @@ 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 +188,13 @@ 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 +202,7 @@ 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 +211,29 @@ 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 +248,9 @@ 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 +260,6 @@ 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

@@ -27,9 +27,7 @@ SOURCES =\
 
 LIB_SHARED = liblvm2format1.so
 
-include ../../make.tmpl
-
-.PHONY: install
+include $(top_srcdir)/make.tmpl
 
 install: liblvm2format1.so
 	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) $< \

lib/format1/disk-rep.c

@@ -482,7 +482,7 @@ int read_pvs_in_vg(const struct format_type *fmt, const char *vg_name,
 		/* vgcache_del(vg_name); */
 	}
 
-	if (!(iter = dev_iter_create(filter))) {
+	if (!(iter = dev_iter_create(filter, 1))) {
 		log_error("read_pvs_in_vg: dev_iter_create failed");
 		return 0;
 	}

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"
 
@@ -147,7 +148,6 @@ static struct volume_group *_build_vg(struct format_instance *fid,
 	vg->seqno = 0;
 	list_init(&vg->pvs);
 	list_init(&vg->lvs);
-	list_init(&vg->snapshots);
 	list_init(&vg->tags);
 
 	if (!_check_vgs(pvs, &partial))
@@ -182,7 +182,7 @@ 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 +275,7 @@ 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 +298,7 @@ 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 +420,7 @@ 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 +492,17 @@ 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 +553,7 @@ 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 +590,7 @@ 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 +25,8 @@
 #include "lvm-string.h"
 #include "filter.h"
 #include "toolcontext.h"
-#include "segtypes.h"
+#include "segtype.h"
+#include "pv_alloc.h"
 
 #include <time.h>
 
@@ -89,6 +90,13 @@ int import_pv(struct pool *mem, struct device *dev,
 	pv->pe_alloc_count = pvd->pe_allocated;
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
+
+	if (!alloc_pv_segment_whole_pv(mem, pv)) {
+		stack;
+		return 0;
+	}
 
 	return 1;
 }
@@ -323,6 +331,8 @@ int import_lv(struct pool *mem, struct logical_volume *lv, struct lv_disk *lvd)
 	lv->size = lvd->lv_size;
 	lv->le_count = lvd->lv_allocated_le;
 
+	lv->snapshot = NULL;
+	list_init(&lv->snapshot_segs);
 	list_init(&lv->segments);
 	list_init(&lv->tags);
 
@@ -495,7 +505,7 @@ int export_lvs(struct disk_list *dl, struct volume_group *vg,
 	       struct physical_volume *pv, const char *dev_dir)
 {
 	int r = 0;
-	struct list *lvh, *sh;
+	struct list *lvh;
 	struct lv_list *ll;
 	struct lvd_list *lvdl;
 	size_t len;
@@ -524,6 +534,9 @@ int export_lvs(struct disk_list *dl, struct volume_group *vg,
 
 	list_iterate(lvh, &vg->lvs) {
 		ll = list_item(lvh, struct lv_list);
+		if (ll->lv->status & SNAPSHOT)
+			continue;
+
 		if (!(lvdl = pool_alloc(dl->mem, sizeof(*lvdl)))) {
 			stack;
 			goto out;
@@ -532,7 +545,6 @@ int export_lvs(struct disk_list *dl, struct volume_group *vg,
 		_export_lv(&lvdl->lvd, vg, ll->lv, dev_dir);
 
 		lv_num = lvnum_from_lvid(&ll->lv->lvid);
-
 		lvdl->lvd.lv_number = lv_num;
 
 		if (!hash_insert(lvd_hash, ll->lv->name, &lvdl->lvd)) {
@@ -545,37 +557,20 @@ int export_lvs(struct disk_list *dl, struct volume_group *vg,
 			goto out;
 		}
 
+		if (lv_is_origin(ll->lv))
+			lvdl->lvd.lv_access |= LV_SNAPSHOT_ORG;
+
+		if (lv_is_cow(ll->lv)) {
+			lvdl->lvd.lv_access |= LV_SNAPSHOT;
+			lvdl->lvd.lv_chunk_size = ll->lv->snapshot->chunk_size;
+			lvdl->lvd.lv_snapshot_minor =
+			    lvnum_from_lvid(&ll->lv->snapshot->origin->lvid);
+		}
+
 		list_add(&dl->lvds, &lvdl->list);
 		dl->pvd.lv_cur++;
 	}
 
-	/*
-	 * Now we need to run through the snapshots, exporting
-	 * the SNAPSHOT_ORG flags etc.
-	 */
-	list_iterate(sh, &vg->snapshots) {
-		struct lv_disk *org, *cow;
-		struct snapshot *s = list_item(sh,
-					       struct snapshot_list)->snapshot;
-
-		if (!(org = hash_lookup(lvd_hash, s->origin->name))) {
-			log_err("Couldn't find snapshot origin '%s'.",
-				s->origin->name);
-			goto out;
-		}
-
-		if (!(cow = hash_lookup(lvd_hash, s->cow->name))) {
-			log_err("Couldn't find snapshot cow store '%s'.",
-				s->cow->name);
-			goto out;
-		}
-
-		org->lv_access |= LV_SNAPSHOT_ORG;
-		cow->lv_access |= LV_SNAPSHOT;
-		cow->lv_snapshot_minor = org->lv_number;
-		cow->lv_chunk_size = s->chunk_size;
-	}
-
 	r = 1;
 
       out:
@@ -646,7 +641,8 @@ int import_snapshots(struct pool *mem, struct volume_group *vg,
 				continue;
 
 			/* insert the snapshot */
-			if (!vg_add_snapshot(org, cow, 1, NULL,
+			if (!vg_add_snapshot(vg->fid, NULL, org, cow, NULL,
+					     org->le_count, 
 					     lvd->lv_chunk_size)) {
 				log_err("Couldn't add snapshot.");
 				return 0;

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,
@@ -60,6 +60,8 @@ static struct hash_table *_create_lv_maps(struct pool *mem,
 
 	list_iterate(llh, &vg->lvs) {
 		ll = list_item(llh, struct lv_list);
+		if (ll->lv->status & SNAPSHOT)
+			continue;
 
 		if (!(lvm = pool_alloc(mem, sizeof(*lvm)))) {
 			stack;
@@ -205,58 +207,55 @@ static int _check_maps_are_complete(struct hash_table *maps)
 
 static int _read_linear(struct cmd_context *cmd, struct lv_map *lvm)
 {
-	uint32_t le = 0;
+	uint32_t le = 0, len;
 	struct lv_segment *seg;
+	struct segment_type *segtype;
+
+	if (!(segtype = get_segtype_from_string(cmd, "striped"))) {
+		stack;
+		return 0;
+	}
 
 	while (le < lvm->lv->le_count) {
-		seg = alloc_lv_segment(cmd->mem, 1);
+		len = 0;
 
-		seg->lv = lvm->lv;
-		if (!(seg->segtype = get_segtype_from_string(cmd, "striped"))) {
-			stack;
+		do
+			len++;
+		while ((lvm->map[le + len].pv == lvm->map[le].pv) &&
+			 (lvm->map[le].pv &&
+			  lvm->map[le + len].pe == lvm->map[le].pe + len));
+
+		if (!(seg = alloc_lv_segment(cmd->mem, segtype, lvm->lv, le,
+					     len, 0, 0, 1, len, 0, 0))) {
+			log_error("Failed to allocate linear segment.");
 			return 0;
 		}
 
-		seg->le = le;
-		seg->len = 0;
-		seg->area_len = 0;
-		seg->stripe_size = 0;
-
-		seg->area[0].type = AREA_PV;
-		seg->area[0].u.pv.pv = lvm->map[le].pv;
-		seg->area[0].u.pv.pe = lvm->map[le].pe;
-
-		do {
-			seg->len++;
-			seg->area_len++;
-		} while ((lvm->map[le + seg->len].pv == seg->area[0].u.pv.pv) &&
-			 (seg->area[0].u.pv.pv &&
-			  lvm->map[le + seg->len].pe == seg->area[0].u.pv.pe +
-			  seg->len));
-
-		le += seg->len;
+		set_lv_segment_area_pv(seg, 0, lvm->map[le].pv, lvm->map[le].pe);
 
 		list_add(&lvm->lv->segments, &seg->list);
+
+		le += seg->len;
 	}
 
 	return 1;
 }
 
-static int _check_stripe(struct lv_map *lvm, struct lv_segment *seg,
-			 uint32_t base_le, uint32_t len)
+static int _check_stripe(struct lv_map *lvm, uint32_t area_count,
+			 uint32_t seg_len, uint32_t base_le, uint32_t len)
 {
-	uint32_t le, st;
-
-	le = base_le + seg->len;
+	uint32_t st;
 
 	/*
 	 * Is the next physical extent in every stripe adjacent to the last?
 	 */
-	for (st = 0; st < seg->area_count; st++)
-		if ((lvm->map[le + st * len].pv != seg->area[st].u.pv.pv) ||
-		    (seg->area[st].u.pv.pv &&
-		     lvm->map[le + st * len].pe !=
-		     seg->area[st].u.pv.pe + seg->len)) return 0;
+	for (st = 0; st < area_count; st++)
+		if ((lvm->map[base_le + st * len + seg_len].pv !=
+		     lvm->map[base_le + st * len].pv) ||
+		    (lvm->map[base_le + st * len].pv &&
+		     lvm->map[base_le + st * len + seg_len].pe !=
+		     lvm->map[base_le + st * len].pe + seg_len))
+			return 0;
 
 	return 1;
 }
@@ -264,7 +263,9 @@ static int _check_stripe(struct lv_map *lvm, struct lv_segment *seg,
 static int _read_stripes(struct cmd_context *cmd, struct lv_map *lvm)
 {
 	uint32_t st, le = 0, len;
+	uint32_t area_len;
 	struct lv_segment *seg;
+	struct segment_type *segtype;
 
 	/*
 	 * Work out overall striped length
@@ -276,43 +277,42 @@ static int _read_stripes(struct cmd_context *cmd, struct lv_map *lvm)
 	}
 	len = lvm->lv->le_count / lvm->stripes;
 
+	if (!(segtype = get_segtype_from_string(cmd, "striped"))) {
+		stack;
+		return 0;
+	}
+
 	while (le < len) {
-		if (!(seg = alloc_lv_segment(cmd->mem, lvm->stripes))) {
-			stack;
-			return 0;
-		}
-
-		seg->lv = lvm->lv;
-		if (!(seg->segtype = get_segtype_from_string(cmd, "striped"))) {
-			stack;
-			return 0;
-		}
-		seg->stripe_size = lvm->stripe_size;
-		seg->le = seg->area_count * le;
-		seg->len = 1;
-		seg->area_len = 1;
-
-		/*
-		 * Set up start positions of each stripe in this segment
-		 */
-		for (st = 0; st < seg->area_count; st++) {
-			seg->area[st].u.pv.pv = lvm->map[le + st * len].pv;
-			seg->area[st].u.pv.pe = lvm->map[le + st * len].pe;
-		}
+		area_len = 1;
 
 		/* 
 		 * Find how many blocks are contiguous in all stripes
 		 * and so can form part of this segment
 		 */
-		while (_check_stripe(lvm, seg, le, len)) {
-			seg->len++;
-			seg->area_len++;
+		while (_check_stripe(lvm, lvm->stripes,
+				     area_len * lvm->stripes, le, len))
+			area_len++;
+
+		if (!(seg = alloc_lv_segment(cmd->mem, segtype, lvm->lv,
+					     lvm->stripes * le,
+					     lvm->stripes * area_len,
+					     0, lvm->stripe_size, lvm->stripes,
+					     area_len, 0, 0))) {
+			log_error("Failed to allocate striped segment.");
+			return 0;
 		}
 
-		le += seg->len;
-		seg->len *= seg->area_count;
+		/*
+		 * Set up start positions of each stripe in this segment
+		 */
+		for (st = 0; st < seg->area_count; st++)
+			set_lv_segment_area_pv(seg, st,
+					       lvm->map[le + st * len].pv,
+					       lvm->map[le + st * len].pe);
 
 		list_add(&lvm->lv->segments, &seg->list);
+
+		le += seg->len;
 	}
 
 	return 1;
@@ -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

@@ -24,9 +24,7 @@ SOURCES =\
 
 LIB_SHARED = liblvm2formatpool.so
 
-include ../../make.tmpl
-
-.PHONY: install
+include $(top_srcdir)/make.tmpl
 
 install: liblvm2formatpool.so
 	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) $< \

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/disk_rep.h

@@ -20,6 +20,8 @@
 #include "metadata.h"
 #include "pool.h"
 
+#define MINOR_OFFSET 65536
+
 /* From NSP.cf */
 #define NSPMajorVersion	4
 #define NSPMinorVersion	1
@@ -66,29 +68,6 @@ struct pool_list;
 struct user_subpool;
 struct user_device;
 
-/* This must be kept up to date with sistina/pool/module/pool_sptypes.h */
-
-/*  Generic Labels  */
-#define SPTYPE_DATA                (0x00000000)
-
-/*  GFS specific labels  */
-#define SPTYPE_GFS_DATA            (0x68011670)
-#define SPTYPE_GFS_JOURNAL         (0x69011670)
-
-struct sptype_name {
-	const char *name;
-	uint32_t label;
-};
-
-static const struct sptype_name sptype_names[] = {
-	{"data",	SPTYPE_DATA},
-
-	{"gfs_data",	SPTYPE_GFS_DATA},
-	{"gfs_journal",	SPTYPE_GFS_JOURNAL},
-
-	{"", 0x0}		/*  This must be the last flag.  */
-};
-
 struct pool_disk {
 	uint64_t pl_magic;	/* Pool magic number */
 	uint64_t pl_pool_id;	/* Unique pool identifier */

lib/format_pool/format_pool.c

@@ -132,7 +132,6 @@ static struct volume_group *_build_vg_from_pds(struct format_instance
 	vg->system_id = NULL;
 	list_init(&vg->pvs);
 	list_init(&vg->lvs);
-	list_init(&vg->snapshots);
 	list_init(&vg->tags);
 
 	if (!import_pool_vg(vg, smem, pds)) {
@@ -181,7 +180,7 @@ 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 +226,7 @@ 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 +356,7 @@ 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

@@ -19,10 +19,12 @@
 #include "metadata.h"
 #include "lvmcache.h"
 #include "disk_rep.h"
+#include "sptype_names.h"
 #include "lv_alloc.h"
+#include "pv_alloc.h"
 #include "str_list.h"
 #include "display.h"
-#include "segtypes.h"
+#include "segtype.h"
 
 /* This file contains only imports at the moment... */
 
@@ -81,6 +83,8 @@ int import_pool_lvs(struct volume_group *vg, struct pool *mem, struct list *pls)
 	lv->name = NULL;
 	lv->le_count = 0;
 	lv->read_ahead = 0;
+	lv->snapshot = NULL;
+	list_init(&lv->snapshot_segs);
 	list_init(&lv->segments);
 	list_init(&lv->tags);
 
@@ -107,10 +111,12 @@ int import_pool_lvs(struct volume_group *vg, struct pool *mem, struct list *pls)
 		/* for pool a minor of 0 is dynamic */
 		if (pl->pd.pl_minor) {
 			lv->status |= FIXED_MINOR;
-			lv->minor = pl->pd.pl_minor;
+			lv->minor = pl->pd.pl_minor + MINOR_OFFSET;
 		} else {
 			lv->minor = -1;
 		}
+		lv->snapshot = NULL;
+		list_init(&lv->snapshot_segs);
 		list_init(&lv->segments);
 		list_init(&lv->tags);
 	}
@@ -146,7 +152,7 @@ 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);
 	}
 
@@ -177,6 +183,13 @@ int import_pool_pv(const struct format_type *fmt, struct pool *mem,
 	pv->pe_alloc_count = pv->pe_count;
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
+
+	if (!alloc_pv_segment_whole_pv(mem, pv)) {
+		stack;
+		return 0;
+	}
 
 	return 1;
 }
@@ -199,40 +212,41 @@ static int _add_stripe_seg(struct pool *mem,
 			   uint32_t *le_cur)
 {
 	struct lv_segment *seg;
+	struct segment_type *segtype;
 	int j;
+	uint32_t area_len;
 
-	if (!(seg = alloc_lv_segment(mem, usp->num_devs))) {
-		log_error("Unable to allocate striped lv_segment structure");
-		return 0;
-	}
-	if(usp->striping & (usp->striping - 1)) {
+	if (usp->striping & (usp->striping - 1)) {
 		log_error("Stripe size must be a power of 2");
 		return 0;
 	}
-	seg->stripe_size = usp->striping;
-	seg->status |= 0;
-	seg->le += *le_cur;
+
+	area_len = (usp->devs[0].blocks) / POOL_PE_SIZE;
+
+	if (!(segtype = get_segtype_from_string(lv->vg->cmd,
+						     "striped"))) {
+		stack;
+		return 0;
+	}
+
+	if (!(seg = alloc_lv_segment(mem, segtype, lv, *le_cur, 
+				     area_len * usp->num_devs, 0,
+				     usp->striping, usp->num_devs, area_len,
+				     0, 0))) {
+		log_error("Unable to allocate striped lv_segment structure");
+		return 0;
+	}
+
+	for (j = 0; j < usp->num_devs; j++)
+		set_lv_segment_area_pv(seg, j, usp->devs[j].pv, 0);
 
 	/* add the subpool type to the segment tag list */
 	str_list_add(mem, &seg->tags, _cvt_sptype(usp->type));
 
-	for (j = 0; j < usp->num_devs; j++) {
-		if (!(seg->segtype = get_segtype_from_string(lv->vg->cmd,
-							     "striped"))) {
-			stack;
-			return 0;
-		}
-
-		seg->area_len = (usp->devs[j].blocks) / POOL_PE_SIZE;
-		seg->len += seg->area_len;
-		*le_cur += seg->area_len;
-		seg->lv = lv;
-
-		seg->area[j].type = AREA_PV;
-		seg->area[j].u.pv.pv = usp->devs[j].pv;
-		seg->area[j].u.pv.pe = 0;
-	}
 	list_add(&lv->segments, &seg->list);
+
+	*le_cur += seg->len;
+
 	return 1;
 }
 
@@ -241,37 +255,35 @@ static int _add_linear_seg(struct pool *mem,
 			   uint32_t *le_cur)
 {
 	struct lv_segment *seg;
+	struct segment_type *segtype;
 	int j;
+	uint32_t area_len;
+
+	if (!(segtype = get_segtype_from_string(lv->vg->cmd, "striped"))) {
+		stack;
+		return 0;
+	}
 
 	for (j = 0; j < usp->num_devs; j++) {
-		/* linear segments only have 1 data area */
-		if (!(seg = alloc_lv_segment(mem, 1))) {
+		area_len = (usp->devs[j].blocks) / POOL_PE_SIZE;
+
+		if (!(seg = alloc_lv_segment(mem, segtype, lv, *le_cur,
+					     area_len, 0, usp->striping,
+					     1, area_len, POOL_PE_SIZE, 0))) {
 			log_error("Unable to allocate linear lv_segment "
 				  "structure");
 			return 0;
 		}
-		seg->stripe_size = usp->striping;
-		seg->le += *le_cur;
-		seg->chunk_size = POOL_PE_SIZE;
-		seg->status |= 0;
-		if (!(seg->segtype = get_segtype_from_string(lv->vg->cmd,
-							     "striped"))) {
-			stack;
-			return 0;
-		}
+
 		/* add the subpool type to the segment tag list */
 		str_list_add(mem, &seg->tags, _cvt_sptype(usp->type));
 
-		seg->lv = lv;
-
-		seg->area_len = (usp->devs[j].blocks) / POOL_PE_SIZE;
-		seg->len = seg->area_len;
-		*le_cur += seg->len;
-		seg->area[0].type = AREA_PV;
-		seg->area[0].u.pv.pv = usp->devs[j].pv;
-		seg->area[0].u.pv.pe = 0;
+		set_lv_segment_area_pv(seg, 0, usp->devs[j].pv, 0);
 		list_add(&lv->segments, &seg->list);
+
+		*le_cur += seg->len;
 	}
+
 	return 1;
 }
 
@@ -287,8 +299,11 @@ int import_pool_segments(struct list *lvs, struct pool *mem,
 
 	list_iterate(lvhs, lvs) {
 		lvl = list_item(lvhs, struct lv_list);
-
 		lv = lvl->lv;
+
+		if (lv->status & SNAPSHOT)
+			continue;
+
 		for (i = 0; i < subpools; i++) {
 			if (usp[i].striping) {
 				if (!_add_stripe_seg(mem, &usp[i], lv, &le_cur)) {

lib/format_pool/sptype_names.h

@@ -0,0 +1,42 @@
+/*
+ * Copyright (C) 1997-2004 Sistina Software, Inc. All rights reserved.  
+ * Copyright (C) 2004-2005 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 SPTYPE_NAMES_H
+#define SPTYPE_NAMES_H
+
+/* This must be kept up to date with sistina/pool/module/pool_sptypes.h */
+
+/*  Generic Labels  */
+#define SPTYPE_DATA                (0x00000000)
+
+/*  GFS specific labels  */
+#define SPTYPE_GFS_DATA            (0x68011670)
+#define SPTYPE_GFS_JOURNAL         (0x69011670)
+
+struct sptype_name {
+	const char *name;
+	uint32_t label;
+};
+
+static const struct sptype_name sptype_names[] = {
+	{"data",	SPTYPE_DATA},
+
+	{"gfs_data",	SPTYPE_GFS_DATA},
+	{"gfs_journal",	SPTYPE_GFS_JOURNAL},
+
+	{"", 0x0}		/*  This must be the last flag.  */
+};
+
+#endif

lib/format_text/archive.c

@@ -53,7 +53,7 @@ struct archive_file {
 	struct list list;
 
 	char *path;
-	int index;
+	uint32_t index;
 };
 
 /*
@@ -132,7 +132,8 @@ static char *_join(struct pool *mem, const char *dir, const char *name)
 static struct list *_scan_archive(struct pool *mem,
 				  const char *vgname, const char *dir)
 {
-	int i, count, ix;
+	int i, count;
+	uint32_t ix;
 	char vgname_found[64], *path;
 	struct dirent **dirent;
 	struct archive_file *af;
@@ -240,7 +241,7 @@ int archive_vg(struct volume_group *vg,
 	       uint32_t retain_days, uint32_t min_archive)
 {
 	int i, fd, renamed = 0;
-	unsigned int ix = 0;
+	uint32_t ix = 0;
 	struct archive_file *last;
 	FILE *fp = NULL;
 	char temp_file[PATH_MAX], archive_name[PATH_MAX];
@@ -285,7 +286,7 @@ int archive_vg(struct volume_group *vg,
 
 	for (i = 0; i < 10; i++) {
 		if (lvm_snprintf(archive_name, sizeof(archive_name),
-				 "%s/%s_%05d.vg", dir, vg->name, ix) < 0) {
+				 "%s/%s_%05u.vg", dir, vg->name, ix) < 0) {
 			log_error("Archive file name too long.");
 			return 0;
 		}

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>
@@ -477,87 +477,6 @@ static int _count_segments(struct logical_volume *lv)
 	return r;
 }
 
-static int _print_snapshot(struct formatter *f, struct snapshot *snap,
-			   unsigned int count)
-{
-	char buffer[256];
-	struct lv_segment seg;
-
-	f->nl(f);
-
-	outf(f, "snapshot%u {", count);
-	_inc_indent(f);
-
-	if (!id_write_format(&snap->id, buffer, sizeof(buffer))) {
-		stack;
-		return 0;
-	}
-
-	outf(f, "id = \"%s\"", buffer);
-
-	seg.status = LVM_READ | LVM_WRITE | VISIBLE_LV;
-	if (!print_flags(seg.status, LV_FLAGS, buffer, sizeof(buffer))) {
-		stack;
-		return 0;
-	}
-
-	outf(f, "status = %s", buffer);
-	outf(f, "segment_count = 1");
-
-	f->nl(f);
-
-	if (!(seg.segtype = get_segtype_from_string(snap->origin->vg->cmd,
-						    "snapshot"))) {
-		stack;
-		return 0;
-	}
-
-	seg.le = 0;
-	seg.len = snap->origin->le_count;
-	seg.origin = snap->origin;
-	seg.cow = snap->cow;
-	seg.chunk_size = snap->chunk_size;
-
-	/* FIXME Dummy values */
-	list_init(&seg.list);
-	seg.lv = snap->cow;
-	seg.stripe_size = 0;
-	seg.area_count = 0;
-	seg.area_len = 0;
-	seg.extents_copied = 0;
-
-	/* Can't tag a snapshot independently of its origin */
-	list_init(&seg.tags);
-
-	if (!_print_segment(f, snap->origin->vg, 1, &seg)) {
-		stack;
-		return 0;
-	}
-
-	_dec_indent(f);
-	outf(f, "}");
-
-	return 1;
-}
-
-static int _print_snapshots(struct formatter *f, struct volume_group *vg)
-{
-	struct list *sh;
-	struct snapshot *s;
-	unsigned int count = 0;
-
-	list_iterate(sh, &vg->snapshots) {
-		s = list_item(sh, struct snapshot_list)->snapshot;
-
-		if (!_print_snapshot(f, s, count++)) {
-			stack;
-			return 0;
-		}
-	}
-
-	return 1;
-}
-
 static int _print_lvs(struct formatter *f, struct volume_group *vg)
 {
 	struct list *lvh;
@@ -629,11 +548,6 @@ static int _print_lvs(struct formatter *f, struct volume_group *vg)
 		outf(f, "}");
 	}
 
-	if (!_print_snapshots(f, vg)) {
-		stack;
-		return 0;
-	}
-
 	_dec_indent(f);
 	outf(f, "}");
 
@@ -652,7 +566,7 @@ 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/flags.c

@@ -54,6 +54,7 @@ static struct flag _lv_flags[] = {
 	{LOCKED, "LOCKED"},
 	{MIRRORED, NULL},
 	{VIRTUAL, NULL},
+	{SNAPSHOT, NULL},
 	{0, NULL}
 };
 

lib/format_text/format-text.c

@@ -88,8 +88,11 @@ static int _lv_setup(struct format_instance *fid, struct logical_volume *lv)
 	}
 */
 
-	if (!*lv->lvid.s)
-		lvid_create(&lv->lvid, &lv->vg->id);
+	if (!*lv->lvid.s && !lvid_create(&lv->lvid, &lv->vg->id)) {
+		log_error("Random lvid creation failed for %s/%s.",
+			  lv->vg->name, lv->name);
+		return 0;
+	}
 
 	return 1;
 }
@@ -180,57 +183,74 @@ static int _raw_write_mda_header(const struct format_type *fmt,
 
 static struct raw_locn *_find_vg_rlocn(struct device_area *dev_area,
 				       struct mda_header *mdah,
-				       const char *vgname)
+				       const char *vgname,
+				       int precommit)
 {
 	size_t len;
 	char vgnamebuf[NAME_LEN + 2];
 	struct raw_locn *rlocn;
+	struct lvmcache_info *info;
 
-	rlocn = mdah->raw_locns;
+	rlocn = mdah->raw_locns;	/* Slot 0 */
 
+	if (precommit)
+		rlocn++;		/* Slot 1 */
+
+	/* FIXME Loop through rlocns two-at-a-time.  List null-terminated. */
 	/* FIXME Ignore if checksum incorrect!!! */
-	while (rlocn->offset) {
-		if (!dev_read(dev_area->dev, dev_area->start + rlocn->offset,
-			      sizeof(vgnamebuf), vgnamebuf)) {
-			stack;
-			return NULL;
-		}
-		if (!strncmp(vgnamebuf, vgname, len = strlen(vgname)) &&
-		    (isspace(vgnamebuf[len]) || vgnamebuf[len] == '{')) {
-			return rlocn;
-		}
-		rlocn++;
+	if (!dev_read(dev_area->dev, dev_area->start + rlocn->offset,
+		      sizeof(vgnamebuf), vgnamebuf)) {
+		stack;
+		goto error;
 	}
 
+	if (!strncmp(vgnamebuf, vgname, len = strlen(vgname)) &&
+	    (isspace(vgnamebuf[len]) || vgnamebuf[len] == '{')) {
+		return rlocn;
+	}
+
+      error:
+	if ((info = info_from_pvid(dev_area->dev->pvid)))
+		lvmcache_update_vgname(info, ORPHAN);
+
 	return NULL;
 }
 
-static struct raw_locn *_vg_posn(struct format_instance *fid,
-				 struct device_area *dev_area,
-				 const char *vgname)
+/*
+ * Determine offset for uncommitted metadata
+ */
+static uint64_t _next_rlocn_offset(struct raw_locn *rlocn,
+				   struct mda_header *mdah)
 {
+	if (!rlocn)
+		/* Find an empty slot */
+		/* FIXME Assume only one VG per mdah for now */
+		return MDA_HEADER_SIZE;
 
-	struct mda_header *mdah;
-
-	if (!(mdah = _raw_read_mda_header(fid->fmt, dev_area))) {
-		stack;
-		return NULL;
-	}
-
-	return _find_vg_rlocn(dev_area, mdah, vgname);
+	/* Start of free space - round up to next sector; circular */
+	return ((rlocn->offset + rlocn->size +
+		(SECTOR_SIZE - rlocn->size % SECTOR_SIZE) -
+		MDA_HEADER_SIZE) % (mdah->size - MDA_HEADER_SIZE))
+	       + MDA_HEADER_SIZE;
 }
 
 static int _raw_holds_vgname(struct format_instance *fid,
 			     struct device_area *dev_area, const char *vgname)
 {
 	int r = 0;
+	struct mda_header *mdah;
 
 	if (!dev_open(dev_area->dev)) {
 		stack;
 		return 0;
 	}
 
-	if (_vg_posn(fid, dev_area, vgname))
+	if (!(mdah = _raw_read_mda_header(fid->fmt, dev_area))) {
+		stack;
+		return 0;
+	}
+
+	if (_find_vg_rlocn(dev_area, mdah, vgname, 0))
 		r = 1;
 
 	if (!dev_close(dev_area->dev))
@@ -241,7 +261,8 @@ static int _raw_holds_vgname(struct format_instance *fid,
 
 static struct volume_group *_vg_read_raw_area(struct format_instance *fid,
 					      const char *vgname,
-					      struct device_area *area)
+					      struct device_area *area,
+					      int precommit)
 {
 	struct volume_group *vg = NULL;
 	struct raw_locn *rlocn;
@@ -260,8 +281,8 @@ static struct volume_group *_vg_read_raw_area(struct format_instance *fid,
 		goto out;
 	}
 
-	if (!(rlocn = _vg_posn(fid, area, vgname))) {
-		stack;
+	if (!(rlocn = _find_vg_rlocn(area, mdah, vgname, precommit))) {
+		log_debug("VG %s not found on %s", vgname, dev_name(area->dev));
 		goto out;
 	}
 
@@ -284,8 +305,9 @@ static struct volume_group *_vg_read_raw_area(struct format_instance *fid,
 		stack;
 		goto out;
 	}
-	log_debug("Read %s metadata (%u) from %s at %" PRIu64 " size %" PRIu64,
-		  vg->name, vg->seqno, dev_name(area->dev),
+	log_debug("Read %s %smetadata (%u) from %s at %" PRIu64 " size %"
+		  PRIu64, vg->name, precommit ? "pre-commit " : "",
+		  vg->seqno, dev_name(area->dev),
 		  area->start + rlocn->offset, rlocn->size);
 
       out:
@@ -301,7 +323,16 @@ static struct volume_group *_vg_read_raw(struct format_instance *fid,
 {
 	struct mda_context *mdac = (struct mda_context *) mda->metadata_locn;
 
-	return _vg_read_raw_area(fid, vgname, &mdac->area);
+	return _vg_read_raw_area(fid, vgname, &mdac->area, 0);
+}
+
+static struct volume_group *_vg_read_precommit_raw(struct format_instance *fid,
+						   const char *vgname,
+						   struct metadata_area *mda)
+{
+	struct mda_context *mdac = (struct mda_context *) mda->metadata_locn;
+
+	return _vg_read_raw_area(fid, vgname, &mdac->area, 1);
 }
 
 static int _vg_write_raw(struct format_instance *fid, struct volume_group *vg,
@@ -341,18 +372,8 @@ static int _vg_write_raw(struct format_instance *fid, struct volume_group *vg,
 		goto out;
 	}
 
-	if ((rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name))) {
-		/* Start of free space - round up to next sector; circular */
-		mdac->rlocn.offset =
-		    ((rlocn->offset + rlocn->size +
-		      (SECTOR_SIZE - rlocn->size % SECTOR_SIZE) -
-		      MDA_HEADER_SIZE) % (mdah->size - MDA_HEADER_SIZE))
-		    + MDA_HEADER_SIZE;
-	} else {
-		/* Find an empty slot */
-		/* FIXME Assume only one VG per mdah for now */
-		mdac->rlocn.offset = MDA_HEADER_SIZE;
-	}
+	rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name, 0);
+	mdac->rlocn.offset = _next_rlocn_offset(rlocn, mdah);
 
 	if (!(mdac->rlocn.size = text_vg_export_raw(vg, "", buf, sizeof(buf)))) {
 		log_error("VG %s metadata writing failed", vg->name);
@@ -417,8 +438,10 @@ static int _vg_write_raw(struct format_instance *fid, struct volume_group *vg,
 	return r;
 }
 
-static int _vg_commit_raw(struct format_instance *fid, struct volume_group *vg,
-			  struct metadata_area *mda)
+static int _vg_commit_raw_rlocn(struct format_instance *fid,
+				struct volume_group *vg,
+				struct metadata_area *mda,
+				int precommit)
 {
 	struct mda_context *mdac = (struct mda_context *) mda->metadata_locn;
 	struct mda_header *mdah;
@@ -445,18 +468,23 @@ static int _vg_commit_raw(struct format_instance *fid, struct volume_group *vg,
 		goto out;
 	}
 
-	if (!(rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name))) {
-		rlocn = &mdah->raw_locns[0];
+	if (!(rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name, 0))) {
+		mdah->raw_locns[0].offset = 0;
 		mdah->raw_locns[1].offset = 0;
+		mdah->raw_locns[2].offset = 0;
+		rlocn = &mdah->raw_locns[0];
 	}
 
+	if (precommit)
+		rlocn++;
+
 	rlocn->offset = mdac->rlocn.offset;
 	rlocn->size = mdac->rlocn.size;
 	rlocn->checksum = mdac->rlocn.checksum;
 
-	log_debug("Committing %s metadata (%u) to %s header at %" PRIu64,
-		  vg->name, vg->seqno, dev_name(mdac->area.dev),
-		  mdac->area.start);
+	log_debug("%sCommitting %s metadata (%u) to %s header at %" PRIu64,
+		  precommit ? "Pre-" : "", vg->name, vg->seqno,
+		  dev_name(mdac->area.dev), mdac->area.start);
 	if (!_raw_write_mda_header(fid->fmt, mdac->area.dev, mdac->area.start,
 				   mdah)) {
 		log_error("Failed to write metadata area header");
@@ -466,12 +494,25 @@ static int _vg_commit_raw(struct format_instance *fid, struct volume_group *vg,
 	r = 1;
 
       out:
-	if (!dev_close(mdac->area.dev))
+	if (!precommit && !dev_close(mdac->area.dev))
 		stack;
 
 	return r;
 }
 
+static int _vg_commit_raw(struct format_instance *fid, struct volume_group *vg,
+			  struct metadata_area *mda)
+{
+	return _vg_commit_raw_rlocn(fid, vg, mda, 0);
+}
+
+static int _vg_precommit_raw(struct format_instance *fid,
+			     struct volume_group *vg,
+			     struct metadata_area *mda)
+{
+	return _vg_commit_raw_rlocn(fid, vg, mda, 1);
+}
+
 /* Close metadata area devices */
 static int _vg_revert_raw(struct format_instance *fid, struct volume_group *vg,
 			  struct metadata_area *mda)
@@ -517,7 +558,7 @@ static int _vg_remove_raw(struct format_instance *fid, struct volume_group *vg,
 		goto out;
 	}
 
-	if (!(rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name))) {
+	if (!(rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name, 0))) {
 		rlocn = &mdah->raw_locns[0];
 		mdah->raw_locns[1].offset = 0;
 	}
@@ -543,13 +584,13 @@ static int _vg_remove_raw(struct format_instance *fid, struct volume_group *vg,
 
 static struct volume_group *_vg_read_file_name(struct format_instance *fid,
 					       const char *vgname,
-					       const char *path_live)
+					       const char *read_path)
 {
 	struct volume_group *vg;
 	time_t when;
 	char *desc;
 
-	if (!(vg = text_vg_import_file(fid, path_live, &when, &desc))) {
+	if (!(vg = text_vg_import_file(fid, read_path, &when, &desc))) {
 		stack;
 		return NULL;
 	}
@@ -562,10 +603,10 @@ static struct volume_group *_vg_read_file_name(struct format_instance *fid,
 	if (vgname && strcmp(vgname, vg->name)) {
 		pool_free(fid->fmt->cmd->mem, vg);
 		log_err("'%s' does not contain volume group '%s'.",
-			path_live, vgname);
+			read_path, vgname);
 		return NULL;
 	} else
-		log_debug("Read volume group %s from %s", vg->name, path_live);
+		log_debug("Read volume group %s from %s", vg->name, read_path);
 
 	return vg;
 }
@@ -579,6 +620,15 @@ static struct volume_group *_vg_read_file(struct format_instance *fid,
 	return _vg_read_file_name(fid, vgname, tc->path_live);
 }
 
+static struct volume_group *_vg_read_precommit_file(struct format_instance *fid,
+						    const char *vgname,
+						    struct metadata_area *mda)
+{
+	struct text_context *tc = (struct text_context *) mda->metadata_locn;
+
+	return _vg_read_file_name(fid, vgname, tc->path_edit);
+}
+
 static int _vg_write_file(struct format_instance *fid, struct volume_group *vg,
 			  struct metadata_area *mda)
 {
@@ -850,7 +900,7 @@ static int _scan_raw(const struct format_type *fmt)
 		if (vgname_from_mda(fmt, &rl->dev_area, vgnamebuf,
 				    sizeof(vgnamebuf))) {
 			if ((vg = _vg_read_raw_area(&fid, vgnamebuf,
-						    &rl->dev_area)))
+						    &rl->dev_area, 0)))
 				lvmcache_update_vg(vg);
 		}
 	}
@@ -1099,34 +1149,6 @@ static int _pv_write(const struct format_type *fmt, struct physical_volume *pv,
 	return 1;
 }
 
-static int _get_pv_from_vg(const struct format_type *fmt, const char *vg_name,
-			   const char *id, struct physical_volume *pv)
-{
-	struct volume_group *vg;
-	struct list *pvh;
-	struct pv_list *pvl;
-	int consistent = 0;
-
-	if (!(vg = vg_read(fmt->cmd, vg_name, &consistent))) {
-		log_error("format_text: _vg_read failed to read VG %s",
-			  vg_name);
-		return 0;
-	}
-
-	if (!consistent)
-		log_error("Warning: Volume group %s is not consistent",
-			  vg_name);
-
-	list_iterate(pvh, &vg->pvs) {
-		pvl = list_item(pvh, struct pv_list);
-		if (id_equal(&pvl->pv->id, (const struct id *) id)) {
-			memcpy(pv, pvl->pv, sizeof(*pv));
-			return 1;
-		}
-	}
-	return 0;
-}
-
 static int _add_raw(struct list *raw_list, struct device_area *dev_area)
 {
 	struct raw_list *rl;
@@ -1175,19 +1197,19 @@ static int _pv_read(const struct format_type *fmt, const char *pv_name,
 
 	/* Have we already cached vgname? */
 	if (info->vginfo && info->vginfo->vgname && *info->vginfo->vgname &&
-	    _get_pv_from_vg(info->fmt, info->vginfo->vgname, info->dev->pvid,
-			    pv)) {
+	    get_pv_from_vg_by_id(info->fmt, info->vginfo->vgname,
+				 info->dev->pvid, pv)) {
 		return 1;
 	}
 
-	/* Perform full scan and try again */
-	if (!memlock()) {
-		lvmcache_label_scan(fmt->cmd, 1);
+	/* Perform full scan (just the first time) and try again */
+	if (!memlock() && !full_scan_done()) {
+		lvmcache_label_scan(fmt->cmd, 2);
 
 		if (info->vginfo && info->vginfo->vgname &&
 		    *info->vginfo->vgname &&
-		    _get_pv_from_vg(info->fmt, info->vginfo->vgname,
-				    info->dev->pvid, pv)) {
+		    get_pv_from_vg_by_id(info->fmt, info->vginfo->vgname,
+					 info->dev->pvid, pv)) {
 			return 1;
 		}
 	}
@@ -1272,6 +1294,7 @@ static void _destroy(const struct format_type *fmt)
 
 static struct metadata_area_ops _metadata_text_file_ops = {
 	vg_read:_vg_read_file,
+	vg_read_precommit:_vg_read_precommit_file,
 	vg_write:_vg_write_file,
 	vg_remove:_vg_remove_file,
 	vg_commit:_vg_commit_file
@@ -1286,8 +1309,10 @@ static struct metadata_area_ops _metadata_text_file_backup_ops = {
 
 static struct metadata_area_ops _metadata_text_raw_ops = {
 	vg_read:_vg_read_raw,
+	vg_read_precommit:_vg_read_precommit_raw,
 	vg_write:_vg_write_raw,
 	vg_remove:_vg_remove_raw,
+	vg_precommit:_vg_precommit_raw,
 	vg_commit:_vg_commit_raw,
 	vg_revert:_vg_revert_raw
 };
@@ -1643,7 +1668,8 @@ struct format_type *create_text_format(struct cmd_context *cmd)
 	fmt->ops = &_text_handler;
 	fmt->name = FMT_TEXT_NAME;
 	fmt->alias = FMT_TEXT_ALIAS;
-	fmt->features = FMT_SEGMENTS | FMT_MDAS | FMT_TAGS | FMT_UNLIMITED_VOLS;
+	fmt->features = FMT_SEGMENTS | FMT_MDAS | FMT_TAGS | FMT_PRECOMMIT |
+			FMT_UNLIMITED_VOLS;
 
 	if (!(mda_lists = dbg_malloc(sizeof(struct mda_lists)))) {
 		log_error("Failed to allocate dir_list");
@@ -1682,14 +1708,15 @@ 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 +22,8 @@
 #include "toolcontext.h"
 #include "lvmcache.h"
 #include "lv_alloc.h"
-#include "segtypes.h"
+#include "pv_alloc.h"
+#include "segtype.h"
 #include "text_import.h"
 
 typedef int (*section_fn) (struct format_instance * fid, struct pool * mem,
@@ -31,7 +32,7 @@ typedef int (*section_fn) (struct format_instance * fid, struct pool * mem,
 			   struct hash_table * pv_hash);
 
 #define _read_int32(root, path, result) \
-	get_config_uint32(root, path, result)
+	get_config_uint32(root, path, (uint32_t *) result)
 
 #define _read_uint32(root, path, result) \
 	get_config_uint32(root, path, result)
@@ -190,6 +191,8 @@ static int _read_pv(struct format_instance *fid, struct pool *mem,
 	}
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
 
 	/* Optional tags */
 	if ((cn = find_config_node(pvn, "tags")) &&
@@ -208,6 +211,11 @@ static int _read_pv(struct format_instance *fid, struct pool *mem,
 	pv->pe_alloc_count = 0;
 	pv->fmt = fid->fmt;
 
+	if (!alloc_pv_segment_whole_pv(mem, pv)) {
+		stack;
+		return 0;
+	}
+
 	vg->pv_count++;
 	list_add(&vg->pvs, &pvl->list);
 
@@ -283,19 +291,13 @@ static int _read_segment(struct pool *mem, struct volume_group *vg,
 		return 0;
 	}
 
-	if (!(seg = alloc_lv_segment(mem, area_count))) {
+	if (!(seg = alloc_lv_segment(mem, segtype, lv, start_extent,
+				     extent_count, 0, 0, area_count,
+				     extent_count, 0, 0))) {
 		log_error("Segment allocation failed");
 		return 0;
 	}
 
-	seg->lv = lv;
-	seg->le = start_extent;
-	seg->len = extent_count;
-	seg->area_len = extent_count;
-	seg->status = 0u;
-	seg->segtype = segtype;
-	seg->extents_copied = 0u;
-
 	if (seg->segtype->ops->text_import &&
 	    !seg->segtype->ops->text_import(seg, sn, pv_hash)) {
 		stack;
@@ -361,19 +363,14 @@ int text_import_areas(struct lv_segment *seg, const struct config_node *sn,
 
 		/* FIXME Cope if LV not yet read in */
 		if ((pv = hash_lookup(pv_hash, cv->v.str))) {
-			seg->area[s].type = AREA_PV;
-			seg->area[s].u.pv.pv = pv;
-			seg->area[s].u.pv.pe = cv->next->v.i;
+			set_lv_segment_area_pv(seg, s, pv, cv->next->v.i);
 			/*
 			 * Adjust extent counts in the pv and vg.
 			 */
 			pv->pe_alloc_count += seg->area_len;
 			seg->lv->vg->free_count -= seg->area_len;
-
 		} else if ((lv1 = find_lv(seg->lv->vg, cv->v.str))) {
-			seg->area[s].type = AREA_LV;
-			seg->area[s].u.lv.lv = lv1;
-			seg->area[s].u.lv.le = cv->next->v.i;
+			set_lv_segment_area_lv(seg, s, lv1, cv->next->v.i);
 		} else {
 			log_error("Couldn't find volume '%s' "
 				  "for segment '%s'.",
@@ -508,6 +505,8 @@ static int _read_lvnames(struct format_instance *fid, struct pool *mem,
 	if (!_read_int32(lvn, "read_ahead", &lv->read_ahead))
 		lv->read_ahead = 0;
 
+	lv->snapshot = NULL;
+	list_init(&lv->snapshot_segs);
 	list_init(&lv->segments);
 	list_init(&lv->tags);
 
@@ -561,24 +560,29 @@ static int _read_lvsegs(struct format_instance *fid, struct pool *mem,
 
 	lv->size = (uint64_t) lv->le_count * (uint64_t) vg->extent_size;
 
-	/* Skip this for now for snapshots */
-	if (!(lv->status & SNAPSHOT)) {
-		lv->minor = -1;
-		if ((lv->status & FIXED_MINOR) &&
-		    !_read_int32(lvn, "minor", &lv->minor)) {
-			log_error("Couldn't read minor number for logical "
-				  "volume %s.", lv->name);
-			return 0;
-		}
-		lv->major = -1;
-		if ((lv->status & FIXED_MINOR) &&
-		    !_read_int32(lvn, "major", &lv->major)) {
-			log_error("Couldn't read major number for logical "
-				  "volume %s.", lv->name);
-		}
-	} else {
+	/*
+	 * FIXME We now have 2 LVs for each snapshot. The real one was
+	 * created by vg_add_snapshot from the segment text_import.
+	 */
+	if (lv->status & SNAPSHOT) {
 		vg->lv_count--;
 		list_del(&lvl->list);
+		return 1;
+	}
+
+	lv->minor = -1;
+	if ((lv->status & FIXED_MINOR) &&
+	    !_read_int32(lvn, "minor", &lv->minor)) {
+		log_error("Couldn't read minor number for logical "
+			  "volume %s.", lv->name);
+		return 0;
+	}
+
+	lv->major = -1;
+	if ((lv->status & FIXED_MINOR) &&
+	    !_read_int32(lvn, "major", &lv->major)) {
+		log_error("Couldn't read major number for logical "
+			  "volume %s.", lv->name);
 	}
 
 	return 1;
@@ -736,7 +740,6 @@ static struct volume_group *_read_vg(struct format_instance *fid,
 	}
 
 	list_init(&vg->lvs);
-	list_init(&vg->snapshots);
 	list_init(&vg->tags);
 
 	/* Optional tags */

lib/label/label.c

@@ -117,12 +117,17 @@ static struct labeller *_find_labeller(struct device *dev, char *buf,
 	struct labeller_i *li;
 	struct labeller *r = NULL;
 	struct label_header *lh;
+	struct lvmcache_info *info;
 	uint64_t sector;
 	int found = 0;
 	char readbuf[LABEL_SCAN_SIZE];
 
 	if (!dev_open(dev)) {
 		stack;
+
+		if ((info = info_from_pvid(dev->pvid)))
+			lvmcache_update_vgname(info, ORPHAN);
+
 		return NULL;
 	}
 
@@ -182,10 +187,13 @@ static struct labeller *_find_labeller(struct device *dev, char *buf,
 		}
 	}
 
-	if (!found)
-		log_very_verbose("%s: No label detected", dev_name(dev));
-
       out:
+	if (!found) {
+		if ((info = info_from_pvid(dev->pvid)))
+			lvmcache_update_vgname(info, ORPHAN);
+		log_very_verbose("%s: No label detected", dev_name(dev));
+	}
+
 	if (!dev_close(dev))
 		stack;
 
@@ -279,7 +287,7 @@ int label_read(struct device *dev, struct label **result)
 		return 0;
 	}
 
-	if ((r = l->ops->read(l, dev, buf, result)) && result && *result)
+	if ((r = (l->ops->read)(l, dev, buf, result)) && result && *result)
 		(*result)->sector = sector;
 
 	return r;
@@ -309,7 +317,7 @@ int label_write(struct device *dev, struct label *label)
 	lh->sector_xl = xlate64(label->sector);
 	lh->offset_xl = xlate32(sizeof(*lh));
 
-	if (!label->labeller->ops->write(label, buf)) {
+	if (!(label->labeller->ops->write)(label, buf)) {
 		stack;
 		return 0;
 	}

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 +1,470 @@
+/*
+ * 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;
+
+		/* Only return an error here if there are no node-specific
+		   errors present in the message that might have more detail */
+		if (!(outheader->flags & CLVMD_FLAG_NODEERRS)) {
+			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;
+			errno = response[i].status;
+		} 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;
+			errno = response[i].status;
+		}
+	}
+
+	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 | LCK_CLUSTERED;
+
+	_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;
+	*flags |= LCK_CLUSTERED;
+
+	return 1;
+}
+#endif

lib/locking/file_locking.c

@@ -41,7 +41,7 @@ static char _lock_dir[NAME_LEN];
 
 static sig_t _oldhandler;
 static sigset_t _fullsigset, _intsigset;
-static int _handler_installed;
+static volatile sig_atomic_t _handler_installed;
 
 static int _release_lock(const char *file, int unlock)
 {
@@ -95,38 +95,40 @@ static void _reset_file_locking(void)
 static void _remove_ctrl_c_handler()
 {
 	siginterrupt(SIGINT, 0);
-	if (!_handler_installed || _oldhandler == SIG_ERR)
+	if (!_handler_installed)
 		return;
 
+	_handler_installed = 0;
+
 	sigprocmask(SIG_SETMASK, &_fullsigset, NULL);
 	if (signal(SIGINT, _oldhandler) == SIG_ERR)
 		log_sys_error("signal", "_remove_ctrl_c_handler");
-
-	_handler_installed = 0;
 }
 
 static void _trap_ctrl_c(int sig)
 {
 	_remove_ctrl_c_handler();
 	log_error("CTRL-c detected: giving up waiting for lock");
-	return;
 }
 
 static void _install_ctrl_c_handler()
 {
-	if ((_oldhandler = signal(SIGINT, _trap_ctrl_c)) == SIG_ERR)
+	_handler_installed = 1;
+
+	if ((_oldhandler = signal(SIGINT, _trap_ctrl_c)) == SIG_ERR) {
+		_handler_installed = 0;
 		return;
+	}
 
 	sigprocmask(SIG_SETMASK, &_intsigset, NULL);
 	siginterrupt(SIGINT, 1);
-
-	_handler_installed = 1;
 }
 
 static int _lock_file(const char *file, int flags)
 {
 	int operation;
 	int r = 1;
+	int old_errno;
 
 	struct lock_list *ll;
 	struct stat buf1, buf2;
@@ -176,10 +178,12 @@ static int _lock_file(const char *file, int flags)
 			_install_ctrl_c_handler();
 
 		r = flock(ll->lf, operation);
+		old_errno = errno;
 		if (!(flags & LCK_NONBLOCK))
 			_remove_ctrl_c_handler();
 
 		if (r) {
+			errno = old_errno;
 			log_sys_error("flock", ll->res);
 			goto err;
 		}

lib/locking/locking.c

@@ -124,6 +124,8 @@ static inline void _update_vg_lock_count(int flags)
  */
 int init_locking(int type, struct config_tree *cft)
 {
+	init_lockingfailed(0);
+
 	switch (type) {
 	case 0:
 		init_no_locking(&_locking, cft);
@@ -145,6 +147,14 @@ 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;
@@ -157,6 +167,7 @@ int init_locking(int type, struct config_tree *cft)
 	log_verbose("Locking disabled - only read operations permitted.");
 
 	init_no_locking(&_locking, cft);
+	init_lockingfailed(1);
 
 	return 1;
 }
@@ -313,3 +324,9 @@ int vg_write_lock_held(void)
 {
 	return _vg_write_lock_held;
 }
+
+int locking_is_clustered(void)
+{
+	return (_locking.flags & LCK_CLUSTERED) ? 1 : 0;
+}
+

lib/locking/locking.h

@@ -20,6 +20,7 @@ int init_locking(int type, struct config_tree *cf);
 void fin_locking(void);
 void reset_locking(void);
 int vg_write_lock_held(void);
+int locking_is_clustered(void);
 
 /*
  * LCK_VG:
@@ -63,6 +64,7 @@ int check_lvm1_vg_inactive(struct cmd_context *cmd, const char *vgname);
  */
 #define LCK_NONBLOCK	0x00000010	/* Don't block waiting for lock? */
 #define LCK_HOLD	0x00000020	/* Hold lock when lock_vol returns? */
+#define LCK_LOCAL	0x00000040	/* Don't propagate to other nodes */
 
 /*
  * Common combinations
@@ -85,9 +87,14 @@ int check_lvm1_vg_inactive(struct cmd_context *cmd, const char *vgname);
 #define activate_lv(cmd, vol)	lock_vol(cmd, vol, LCK_LV_ACTIVATE | LCK_HOLD)
 #define activate_lv_excl(cmd, vol)	\
 				lock_vol(cmd, vol, LCK_LV_EXCLUSIVE | LCK_HOLD)
+#define activate_lv_local(cmd, vol)	\
+	lock_vol(cmd, vol, LCK_LV_ACTIVATE | LCK_HOLD | LCK_LOCAL)
+#define deactivate_lv_local(cmd, vol)	\
+	lock_vol(cmd, vol, LCK_LV_DEACTIVATE | LCK_LOCAL)
 
 /* Process list of LVs */
 int suspend_lvs(struct cmd_context *cmd, struct list *lvs);
 int resume_lvs(struct cmd_context *cmd, struct list *lvs);
 int activate_lvs_excl(struct cmd_context *cmd, struct list *lvs);
 
+

lib/locking/locking_types.h

@@ -23,6 +23,7 @@ typedef void (*fin_lock_fn) (void);
 typedef void (*reset_lock_fn) (void);
 
 #define LCK_PRE_MEMLOCK	0x00000001	/* Is memlock() needed before calls? */
+#define LCK_CLUSTERED	0x00000002
 
 struct locking_type {
 	uint32_t flags;
@@ -40,3 +41,4 @@ 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

@@ -28,7 +28,9 @@ 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 _full_scan_done = 0;	/* Restrict to one full scan during each cmd */
 static int _debug_level = 0;
 static int _syslog = 0;
 static int _log_to_file = 0;
@@ -38,6 +40,8 @@ static int _indent = 1;
 static int _log_cmd_name = 0;
 static int _log_suppress = 0;
 static int _ignorelockingfailure = 0;
+static int _lockingfailed = 0;
+static int _security_level = SECURITY_LEVEL;
 static char _cmd_name[30] = "";
 static char _msg_prefix[30] = "  ";
 static int _already_logging = 0;
@@ -137,16 +141,36 @@ void init_partial(int level)
 	_partial = level;
 }
 
+void init_md_filtering(int level)
+{
+	_md_filtering = level;
+}
+
 void init_pvmove(int level)
 {
 	_pvmove = level;
 }
 
+void init_full_scan_done(int level)
+{
+	_full_scan_done = level;
+}
+
 void init_ignorelockingfailure(int level)
 {
 	_ignorelockingfailure = level;
 }
 
+void init_lockingfailed(int level)
+{
+	_lockingfailed = level;
+}
+
+void init_security_level(int level)
+{
+	_security_level = level;
+}
+
 void init_cmd_name(int status)
 {
 	_log_cmd_name = status;
@@ -181,16 +205,36 @@ int partial_mode()
 	return _partial;
 }
 
+int md_filtering()
+{
+	return _md_filtering;
+}
+
 int pvmove_mode()
 {
 	return _pvmove;
 }
 
+int full_scan_done()
+{
+	return _full_scan_done;
+}
+
+int lockingfailed()
+{
+	return _lockingfailed;
+}
+
 int ignorelockingfailure()
 {
 	return _ignorelockingfailure;
 }
 
+int security_level()
+{
+	return _security_level;
+}
+
 void init_debug(int level)
 {
 	_debug_level = level;
@@ -204,7 +248,7 @@ 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 */
@@ -232,36 +276,46 @@ 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:
@@ -273,7 +327,8 @@ 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);
 			}
@@ -281,7 +336,8 @@ 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

@@ -49,6 +49,7 @@
 #define _LOG_FATAL 2
 
 #define VERBOSE_BASE_LEVEL _LOG_WARN
+#define SECURITY_LEVEL 0
 
 void init_log_file(const char *log_file, int append);
 void init_log_direct(const char *log_file, int append);
@@ -62,20 +63,28 @@ 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_full_scan_done(int level);
 void init_debug(int level);
 void init_cmd_name(int status);
 void init_msg_prefix(const char *prefix);
 void init_indent(int indent);
 void init_ignorelockingfailure(int level);
+void init_lockingfailed(int level);
+void init_security_level(int level);
 
 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 full_scan_done(void);
 int debug_level(void);
 int ignorelockingfailure(void);
+int lockingfailed(void);
+int security_level(void);
 
 /* Suppress messages to stdout/stderr */
 void log_suppress(int suppress);

lib/metadata/lv_alloc.h

@@ -16,5 +16,23 @@
 #ifndef _LVM_LV_ALLOC_H
 #include "pool.h"
 
-struct lv_segment *alloc_lv_segment(struct pool *mem, uint32_t num_areas);
+struct lv_segment *alloc_lv_segment(struct pool *mem,
+				    struct segment_type *segtype,
+				    struct logical_volume *lv,
+				    uint32_t le, uint32_t len,
+				    uint32_t status,
+				    uint32_t stripe_size,
+				    uint32_t area_count,
+				    uint32_t area_len,
+				    uint32_t chunk_size,
+				    uint32_t extents_copied);
+
+struct lv_segment *alloc_snapshot_seg(struct logical_volume *lv,
+				      uint32_t allocated);
+
+void set_lv_segment_area_pv(struct lv_segment *seg, uint32_t area_num,
+			    struct physical_volume *pv, uint32_t pe);
+void set_lv_segment_area_lv(struct lv_segment *seg, uint32_t area_num,
+			    struct logical_volume *lv, uint32_t le);
+
 #endif

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
@@ -61,22 +61,57 @@ static void _put_extents(struct lv_segment *seg)
 	}
 }
 
-struct lv_segment *alloc_lv_segment(struct pool *mem, uint32_t num_areas)
+struct lv_segment *alloc_lv_segment(struct pool *mem,
+				    struct segment_type *segtype,
+				    struct logical_volume *lv,
+				    uint32_t le, uint32_t len,
+				    uint32_t status,
+				    uint32_t stripe_size,
+				    uint32_t area_count,
+				    uint32_t area_len,
+				    uint32_t chunk_size,
+				    uint32_t extents_copied)
 {
 	struct lv_segment *seg;
-	uint32_t len = sizeof(*seg) + (num_areas * sizeof(seg->area[0]));
+	uint32_t sz = sizeof(*seg) + (area_count * sizeof(seg->area[0]));
 
-	if (!(seg = pool_zalloc(mem, len))) {
+	if (!(seg = pool_zalloc(mem, sz))) {
 		stack;
 		return NULL;
 	}
 
-	seg->area_count = num_areas;
+	seg->segtype = segtype;
+	seg->lv = lv;
+	seg->le = le;
+	seg->len = len;
+	seg->status = status;
+	seg->stripe_size = stripe_size;
+	seg->area_count = area_count;
+	seg->area_len = area_len;
+	seg->chunk_size = chunk_size;
+	seg->extents_copied = extents_copied;
 	list_init(&seg->tags);
 
 	return seg;
 }
 
+void set_lv_segment_area_pv(struct lv_segment *seg, uint32_t area_num,
+			    struct physical_volume *pv, uint32_t pe)
+{
+	seg->area[area_num].type = AREA_PV;
+	seg->area[area_num].u.pv.pv = pv;
+	seg->area[area_num].u.pv.pe = pe;
+	seg->area[area_num].u.pv.pvseg = NULL;
+}
+
+void set_lv_segment_area_lv(struct lv_segment *seg, uint32_t area_num,
+			    struct logical_volume *lv, uint32_t le)
+{
+	seg->area[area_num].type = AREA_LV;
+	seg->area[area_num].u.lv.lv = lv;
+	seg->area[area_num].u.lv.le = le;
+}
+
 static int _alloc_parallel_area(struct logical_volume *lv, uint32_t area_count,
 				uint32_t stripe_size,
 				struct segment_type *segtype,
@@ -98,24 +133,17 @@ static int _alloc_parallel_area(struct logical_volume *lv, uint32_t area_count,
 	if (smallest < area_len)
 		area_len = smallest;
 
-	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, area_count))) {
-		log_err("Couldn't allocate new parallel segment.");
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, *ix,
+				     area_len * (striped ? area_count : 1),
+				     0u, stripe_size, area_count, area_len,
+				     0u, 0u))) {
+		log_error("Couldn't allocate new parallel segment.");
 		return 0;
 	}
 
-	seg->lv = lv;
-	seg->segtype = segtype;
-	seg->le = *ix;
-	seg->len = area_len * (striped ? area_count : 1);
-	seg->area_len = area_len;
-	seg->stripe_size = stripe_size;
-	seg->extents_copied = 0u;
-
 	for (s = 0; s < area_count; s++) {
 		struct pv_area *pva = areas[s];
-		seg->area[s].type = AREA_PV;
-		seg->area[s].u.pv.pv = pva->map->pvl->pv;
-		seg->area[s].u.pv.pe = pva->start;
+		set_lv_segment_area_pv(seg, s, pva->map->pvl->pv, pva->start);
 		consume_pv_area(pva, area_len);
 	}
 
@@ -221,34 +249,31 @@ static int _alloc_linear_area(struct logical_volume *lv, uint32_t *ix,
 {
 	uint32_t count, remaining;
 	struct lv_segment *seg;
+	struct segment_type *segtype;
 
 	count = pva->count;
 	remaining = lv->le_count - *ix;
 	if (count > remaining)
 		count = remaining;
 
-	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, 1))) {
-		log_err("Couldn't allocate new stripe segment.");
-		return 0;
-	}
-
-	seg->lv = lv;
-	if (!(seg->segtype = get_segtype_from_string(lv->vg->cmd, "striped"))) {
+	if (!(segtype = get_segtype_from_string(lv->vg->cmd, "striped"))) {
 		stack;
 		return 0;
 	}
-	seg->le = *ix;
-	seg->len = count;
-	seg->area_len = count;
-	seg->stripe_size = 0;
-	seg->area[0].type = AREA_PV;
-	seg->area[0].u.pv.pv = map->pvl->pv;
-	seg->area[0].u.pv.pe = pva->start;
+
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, *ix,
+				     count, 0, 0, 1, count, 0, 0))) {
+		log_error("Couldn't allocate new stripe segment.");
+		return 0;
+	}
+
+	set_lv_segment_area_pv(seg, 0, map->pvl->pv, pva->start);
 
 	list_add(&lv->segments, &seg->list);
 
-	consume_pv_area(pva, count);
-	*ix += count;
+	consume_pv_area(pva, seg->len);
+	*ix += seg->len;
+
 	return 1;
 }
 
@@ -266,30 +291,21 @@ static int _alloc_mirrored_area(struct logical_volume *lv, uint32_t *ix,
 	if (count > remaining)
 		count = remaining;
 
-	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, 2))) {
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, *ix,
+				     count, 0, 0, 2, count, 0, 0))) {
 		log_err("Couldn't allocate new mirrored segment.");
 		return 0;
 	}
 
-	seg->lv = lv;
-	seg->segtype = segtype;
-	seg->le = *ix;
-	seg->status = 0u;
-	seg->len = count;
-	seg->area_len = count;
-	seg->stripe_size = 0;
-	seg->extents_copied = 0u;
 	/* FIXME Remove AREA_PV restriction here? */
-	seg->area[0].type = AREA_PV;
-	seg->area[0].u.pv.pv = mirrored_pv;
-	seg->area[0].u.pv.pe = mirrored_pe;
-	seg->area[1].type = AREA_PV;
-	seg->area[1].u.pv.pv = map->pvl->pv;
-	seg->area[1].u.pv.pe = pva->start;
+	set_lv_segment_area_pv(seg, 0, mirrored_pv, mirrored_pe);
+	set_lv_segment_area_pv(seg, 1, map->pvl->pv, pva->start);
+
 	list_add(&lv->segments, &seg->list);
 
-	consume_pv_area(pva, count);
-	*ix += count;
+	consume_pv_area(pva, seg->len);
+	*ix += seg->len;
+
 	return 1;
 }
 
@@ -421,25 +437,44 @@ static int _alloc_virtual(struct logical_volume *lv,
 {
 	struct lv_segment *seg;
 
-	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, 0))) {
-		log_err("Couldn't allocate new zero segment.");
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, allocated,
+				     lv->le_count - allocated, 0, 0, 0,
+				     lv->le_count - allocated, 0, 0))) {
+		log_error("Couldn't allocate new zero segment.");
 		return 0;
 	}
 
-	seg->lv = lv;
-	seg->segtype = segtype;
-	seg->status = 0u;
-	seg->le = allocated;
-	seg->len = lv->le_count - allocated;
-	seg->area_len = seg->len;
-	seg->stripe_size = 0;
-	seg->extents_copied = 0u;
 	list_add(&lv->segments, &seg->list);
 	lv->status |= VIRTUAL;
 
 	return 1;
 }
 
+struct lv_segment *alloc_snapshot_seg(struct logical_volume *lv,
+				      uint32_t allocated)
+{
+	struct lv_segment *seg;
+	struct segment_type *segtype;
+
+	segtype = get_segtype_from_string(lv->vg->cmd, "snapshot");
+	if (!segtype) {
+		log_error("Failed to find snapshot segtype");
+		return NULL;
+	}
+
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, allocated,
+				     lv->le_count - allocated, 0, 0, 0,
+				     lv->le_count - allocated, 0, 0))) {
+		log_error("Couldn't allocate new snapshot segment.");
+		return NULL;
+	}
+
+	list_add(&lv->segments, &seg->list);
+	lv->status |= VIRTUAL;
+
+	return seg;
+}
+
 /*
  * Chooses a correct allocation policy.
  */
@@ -458,7 +493,7 @@ 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;
 	}
@@ -543,8 +578,10 @@ static char *_generate_lv_name(struct volume_group *vg, const char *format,
 struct logical_volume *lv_create_empty(struct format_instance *fi,
 				       const char *name,
 				       const char *name_format,
+				       union lvid *lvid,
 				       uint32_t status,
 				       alloc_policy_t alloc,
+				       int import,
 				       struct volume_group *vg)
 {
 	struct cmd_context *cmd = vg->cmd;
@@ -565,7 +602,8 @@ struct logical_volume *lv_create_empty(struct format_instance *fi,
 		return NULL;
 	}
 
-	log_verbose("Creating logical volume %s", name);
+	if (!import)
+		log_verbose("Creating logical volume %s", name);
 
 	if (!(ll = pool_zalloc(cmd->mem, sizeof(*ll))) ||
 	    !(ll->lv = pool_zalloc(cmd->mem, sizeof(*ll->lv)))) {
@@ -592,9 +630,14 @@ struct logical_volume *lv_create_empty(struct format_instance *fi,
 	lv->minor = -1;
 	lv->size = UINT64_C(0);
 	lv->le_count = 0;
+	lv->snapshot = NULL;
+	list_init(&lv->snapshot_segs);
 	list_init(&lv->segments);
 	list_init(&lv->tags);
 
+	if (lvid)
+		lv->lvid = *lvid;
+
 	if (fi->fmt->ops->lv_setup && !fi->fmt->ops->lv_setup(fi, lv)) {
 		stack;
 		if (ll)
@@ -602,7 +645,9 @@ struct logical_volume *lv_create_empty(struct format_instance *fi,
 		return NULL;
 	}
 
-	vg->lv_count++;
+	if (!import)
+		vg->lv_count++;
+
 	list_add(&vg->lvs, &ll->list);
 
 	return lv;
@@ -623,6 +668,16 @@ 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
@@ -93,11 +98,17 @@ static int _lv_split_segment(struct logical_volume *lv, struct lv_segment *seg,
 	}
 
 	/* Clone the existing segment */
-	if (!(split_seg = alloc_lv_segment(lv->vg->cmd->mem, seg->area_count))) {
-		log_error("Couldn't allocate new LV segment.");
+	if (!(split_seg = alloc_lv_segment(lv->vg->cmd->mem, seg->segtype,
+					   seg->lv, seg->le, seg->len,
+					   seg->status, seg->stripe_size,
+					   seg->area_count, seg->area_len,
+					   seg->chunk_size,
+					   seg->extents_copied))) {
+		log_error("Couldn't allocate cloned LV segment.");
 		return 0;
 	}
 
+	/* FIXME Avoid the memcpy */
 	len = sizeof(*seg) + (seg->area_count * sizeof(seg->area[0]));
 	memcpy(split_seg, seg, len);
 
@@ -107,8 +118,9 @@ 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 +128,19 @@ 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 +150,13 @@ 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.c

@@ -21,6 +21,8 @@
 #include "lvm-string.h"
 #include "lvmcache.h"
 #include "memlock.h"
+#include "str_list.h"
+#include "pv_alloc.h"
 
 static int _add_pv_to_vg(struct format_instance *fid, struct volume_group *vg,
 			 const char *pv_name)
@@ -39,7 +41,7 @@ static int _add_pv_to_vg(struct format_instance *fid, struct volume_group *vg,
 	}
 
 	list_init(&mdas);
-	if (!(pv = pv_read(fid->fmt->cmd, pv_name, &mdas, NULL))) {
+	if (!(pv = pv_read(fid->fmt->cmd, pv_name, &mdas, NULL, 1))) {
 		log_error("%s not identified as an existing physical volume",
 			  pv_name);
 		return 0;
@@ -109,6 +111,57 @@ static int _add_pv_to_vg(struct format_instance *fid, struct volume_group *vg,
 	return 1;
 }
 
+static int _copy_pv(struct physical_volume *pv_to,
+		    struct physical_volume *pv_from)
+{
+	memcpy(pv_to, pv_from, sizeof(*pv_to));
+
+	if (!str_list_dup(pv_to->fmt->cmd->mem, &pv_to->tags, &pv_from->tags)) {
+		log_error("PV tags duplication failed");
+		return 0;
+	}
+
+	if (!peg_dup(pv_to->fmt->cmd->mem, &pv_to->segments,
+		     &pv_to->free_segments, &pv_from->segments)) {
+		stack;
+		return 0;
+	}
+
+	return 1;
+}
+
+int get_pv_from_vg_by_id(const struct format_type *fmt, const char *vg_name,
+			 const char *id, struct physical_volume *pv)
+{
+	struct volume_group *vg;
+	struct list *pvh;
+	struct pv_list *pvl;
+	int consistent = 0;
+
+	if (!(vg = vg_read(fmt->cmd, vg_name, &consistent))) {
+		log_error("get_pv_from_vg_by_id: vg_read failed to read VG %s",
+			  vg_name);
+		return 0;
+	}
+
+	if (!consistent)
+		log_error("Warning: Volume group %s is not consistent",
+			  vg_name);
+
+	list_iterate(pvh, &vg->pvs) {
+		pvl = list_item(pvh, struct pv_list);
+		if (id_equal(&pvl->pv->id, (const struct id *) id)) {
+			if (!_copy_pv(pv, pvl->pv)) {
+				stack;
+				return 0;
+			}
+			return 1;
+		}
+	}
+
+	return 0;
+}
+
 int vg_rename(struct cmd_context *cmd, struct volume_group *vg,
 	      const char *new_name)
 {
@@ -221,7 +274,6 @@ struct volume_group *vg_create(struct cmd_context *cmd, const char *vg_name,
 	list_init(&vg->lvs);
 
 	vg->snapshot_count = 0;
-	list_init(&vg->snapshots);
 
 	list_init(&vg->tags);
 
@@ -249,6 +301,153 @@ struct volume_group *vg_create(struct cmd_context *cmd, const char *vg_name,
 	return NULL;
 }
 
+static int _recalc_extents(uint32_t *extents, const char *desc1,
+			   const char *desc2, uint32_t old_size,
+			   uint32_t new_size)
+{
+	uint64_t size = (uint64_t) old_size * (*extents);
+
+	if (size % new_size) {
+		log_error("New size %" PRIu64 " for %s%s not an exact number "
+			  "of new extents.", size, desc1, desc2);
+		return 0;
+	}
+
+	size /= new_size;
+
+	if (size > UINT32_MAX) {
+		log_error("New extent count %" PRIu64 " for %s%s exceeds "
+			  "32 bits.", size, desc1, desc2);
+		return 0;
+	}
+
+	*extents = (uint32_t) size;
+
+	return 1;
+}
+
+int vg_change_pesize(struct cmd_context *cmd, struct volume_group *vg,
+		     uint32_t new_size)
+{
+	uint32_t old_size = vg->extent_size;
+	struct pv_list *pvl;
+	struct lv_list *lvl;
+	struct physical_volume *pv;
+	struct logical_volume *lv;
+	struct lv_segment *seg;
+	uint32_t s;
+
+	vg->extent_size = new_size;
+
+	if (vg->fid->fmt->ops->vg_setup &&
+	    !vg->fid->fmt->ops->vg_setup(vg->fid, vg)) {
+		stack;
+		return 0;
+	}
+
+	if (!_recalc_extents(&vg->extent_count, vg->name, "", old_size,
+			     new_size)) {
+		stack;
+		return 0;
+	}
+
+	if (!_recalc_extents(&vg->free_count, vg->name, " free space",
+			     old_size, new_size)) {
+		stack;
+		return 0;
+	}
+
+	/* foreach PV */
+	list_iterate_items(pvl, &vg->pvs) {
+		pv = pvl->pv;
+
+		pv->pe_size = new_size;
+		if (!_recalc_extents(&pv->pe_count, dev_name(pv->dev), "",
+				     old_size, new_size)) {
+			stack;
+			return 0;
+		}
+
+		if (!_recalc_extents(&pv->pe_alloc_count, dev_name(pv->dev),
+				     " allocated space", old_size, new_size)) {
+			stack;
+			return 0;
+		}
+	}
+
+	/* foreach LV */
+	list_iterate_items(lvl, &vg->lvs) {
+		lv = lvl->lv;
+
+		if (!_recalc_extents(&lv->le_count, lv->name, "", old_size,
+				     new_size)) {
+			stack;
+			return 0;
+		}
+
+		list_iterate_items(seg, &lv->segments) {
+			if (!_recalc_extents(&seg->le, lv->name,
+					     " segment start", old_size,
+					     new_size)) {
+				stack;
+				return 0;
+			}
+
+			if (!_recalc_extents(&seg->len, lv->name,
+					     " segment length", old_size,
+					     new_size)) {
+				stack;
+				return 0;
+			}
+
+			if (!_recalc_extents(&seg->area_len, lv->name,
+					     " area length", old_size,
+					     new_size)) {
+				stack;
+				return 0;
+			}
+
+			if (!_recalc_extents(&seg->extents_copied, lv->name,
+					     " extents moved", old_size,
+					     new_size)) {
+				stack;
+				return 0;
+			}
+
+			/* foreach area */
+			for (s = 0; s < seg->area_count; s++) {
+				switch (seg->area[s].type) {
+				case AREA_PV:
+					if (!_recalc_extents
+					    (&seg->area[s].u.pv.pe, lv->name,
+					     " area start", old_size,
+					     new_size)) {
+						stack;
+						return 0;
+					}
+					break;
+				case AREA_LV:
+					if (!_recalc_extents
+					    (&seg->area[s].u.lv.le, lv->name,
+					     " area start", old_size,
+					     new_size)) {
+						stack;
+						return 0;
+					}
+					break;
+				default:
+					log_error("Unrecognised segment type "
+						  "%u", seg->area[s].type);
+					return 0;
+				}
+			}
+		}
+
+	}
+
+	return 1;
+}
+
 /* Sizes in sectors */
 struct physical_volume *pv_create(const struct format_type *fmt,
 				  struct device *dev,
@@ -267,10 +466,13 @@ struct physical_volume *pv_create(const struct format_type *fmt,
 		return NULL;
 	}
 
-	if (!id)
-		id_create(&pv->id);
-	else
+	if (id)
 		memcpy(&pv->id, id, sizeof(*id));
+	else if (!id_create(&pv->id)) {
+		log_error("Failed to create random uuid for %s.",
+			  dev_name(dev));
+		return NULL;
+	}
 
 	pv->dev = dev;
 
@@ -308,6 +510,8 @@ struct physical_volume *pv_create(const struct format_type *fmt,
 	pv->fmt = fmt;
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
 
 	if (!fmt->ops->pv_setup(fmt, pe_start, existing_extent_count,
 				existing_extent_size,
@@ -426,7 +630,7 @@ struct physical_volume *find_pv_by_name(struct cmd_context *cmd,
 {
 	struct physical_volume *pv;
 
-	if (!(pv = pv_read(cmd, pv_name, NULL, NULL))) {
+	if (!(pv = pv_read(cmd, pv_name, NULL, NULL, 1))) {
 		log_error("Physical volume %s not found", pv_name);
 		return NULL;
 	}
@@ -504,6 +708,7 @@ int vg_write(struct volume_group *vg)
 			/* Revert */
 			list_uniterate(mdah2, &vg->fid->metadata_areas, mdah) {
 				mda = list_item(mdah2, struct metadata_area);
+
 				if (mda->ops->vg_revert &&
 				    !mda->ops->vg_revert(vg->fid, vg, mda)) {
 					stack;
@@ -525,6 +730,24 @@ int vg_write(struct volume_group *vg)
 		}
 	}
 
+	/* Now pre-commit each copy of the new metadata */
+	list_iterate(mdah, &vg->fid->metadata_areas) {
+		mda = list_item(mdah, struct metadata_area);
+		if (mda->ops->vg_precommit &&
+		    !mda->ops->vg_precommit(vg->fid, vg, mda)) {
+			stack;
+			/* Revert */
+			list_iterate(mdah2, &vg->fid->metadata_areas) {
+				mda = list_item(mdah2, struct metadata_area);
+				if (mda->ops->vg_revert &&
+				    !mda->ops->vg_revert(vg->fid, vg, mda)) {
+					stack;
+				}
+			}
+			return 0;
+		}
+	}
+
 	return 1;
 }
 
@@ -595,7 +818,6 @@ static struct volume_group *_vg_read_orphans(struct cmd_context *cmd)
 	}
 	list_init(&vg->pvs);
 	list_init(&vg->lvs);
-	list_init(&vg->snapshots);
 	list_init(&vg->tags);
 	vg->cmd = cmd;
 	if (!(vg->name = pool_strdup(cmd->mem, ORPHAN))) {
@@ -605,7 +827,7 @@ static struct volume_group *_vg_read_orphans(struct cmd_context *cmd)
 
 	list_iterate(ih, &vginfo->infos) {
 		dev = list_item(ih, struct lvmcache_info)->dev;
-		if (!(pv = pv_read(cmd, dev_name(dev), NULL, NULL))) {
+		if (!(pv = pv_read(cmd, dev_name(dev), NULL, NULL, 1))) {
 			continue;
 		}
 		if (!(pvl = pool_zalloc(cmd->mem, sizeof(*pvl)))) {
@@ -627,8 +849,9 @@ static struct volume_group *_vg_read_orphans(struct cmd_context *cmd)
  * and take appropriate action if it isn't (e.g. abort; get write lock 
  * and call vg_read again).
  */
-struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
-			     int *consistent)
+static struct volume_group *_vg_read(struct cmd_context *cmd,
+				     const char *vgname,
+				     int *consistent, int precommitted)
 {
 	struct format_instance *fid;
 	const struct format_type *fmt;
@@ -638,6 +861,11 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 	int inconsistent = 0;
 
 	if (!*vgname) {
+		if (precommitted) {
+			log_error("Internal error: vg_read requires vgname "
+				  "with pre-commit.");
+			return NULL;
+		}
 		*consistent = 1;
 		return _vg_read_orphans(cmd);
 	}
@@ -651,7 +879,7 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 				stack;
 				return NULL;
 			}
-			lvmcache_label_scan(cmd, 1);
+			lvmcache_label_scan(cmd, 2);
 			if (!(fmt = fmt_from_vgname(vgname))) {
 				stack;
 				return NULL;
@@ -659,6 +887,12 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 		}
 	}
 
+	if (precommitted && !(fmt->features & FMT_PRECOMMIT)) {
+		log_error("Internal error: %s doesn't support "
+			  "pre-commit", fmt->name);
+		return NULL;
+	}
+
 	/* create format instance with appropriate metadata area */
 	if (!(fid = fmt->ops->create_instance(fmt, vgname, NULL))) {
 		log_error("Failed to create format instance");
@@ -668,7 +902,10 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 	/* Ensure contents of all metadata areas match - else do recovery */
 	list_iterate(mdah, &fid->metadata_areas) {
 		mda = list_item(mdah, struct metadata_area);
-		if (!(vg = mda->ops->vg_read(fid, vgname, mda))) {
+		if ((precommitted &&
+		     !(vg = mda->ops->vg_read_precommit(fid, vgname, mda))) ||
+		    (!precommitted &&
+		     !(vg = mda->ops->vg_read(fid, vgname, mda)))) {
 			inconsistent = 1;
 			continue;
 		}
@@ -684,15 +921,67 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 		}
 	}
 
-	/* Failed to find VG */
+	/* Failed to find VG where we expected it - full scan and retry */
 	if (!correct_vg) {
-		stack;
-		return NULL;
+		inconsistent = 0;
+
+		lvmcache_label_scan(cmd, 2);
+		if (!(fmt = fmt_from_vgname(vgname))) {
+			stack;
+			return NULL;
+		}
+
+		if (precommitted && !(fmt->features & FMT_PRECOMMIT)) {
+			log_error("Internal error: %s doesn't support "
+				  "pre-commit", fmt->name);
+			return NULL;
+		}
+
+		/* create format instance with appropriate metadata area */
+		if (!(fid = fmt->ops->create_instance(fmt, vgname, NULL))) {
+			log_error("Failed to create format instance");
+			return NULL;
+		}
+
+		/* Ensure contents of all metadata areas match - else recover */
+		list_iterate(mdah, &fid->metadata_areas) {
+			mda = list_item(mdah, struct metadata_area);
+			if ((precommitted &&
+			     !(vg = mda->ops->vg_read_precommit(fid, vgname,
+								mda))) ||
+			    (!precommitted &&
+			     !(vg = mda->ops->vg_read(fid, vgname, mda)))) {
+				inconsistent = 1;
+				continue;
+			}
+			if (!correct_vg) {
+				correct_vg = vg;
+				continue;
+			}
+			/* FIXME Also ensure contents same - checksums same? */
+			if (correct_vg->seqno != vg->seqno) {
+				inconsistent = 1;
+				if (vg->seqno > correct_vg->seqno)
+					correct_vg = vg;
+			}
+		}
+
+		/* Give up looking */
+		if (!correct_vg) {
+			stack;
+			return NULL;
+		}
 	}
 
 	lvmcache_update_vg(correct_vg);
 
 	if (inconsistent) {
+		if (precommitted) {
+			log_error("Inconsistent pre-commit metadata copies "
+				  "for volume group %s", vgname);
+			return NULL;
+		}
+
 		if (!*consistent)
 			return correct_vg;
 
@@ -711,6 +1000,11 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 			log_error("Automatic metadata correction failed");
 			return NULL;
 		}
+		if (!vg_commit(correct_vg)) {
+			log_error("Automatic metadata correction commit "
+				  "failed");
+			return NULL;
+		}
 	}
 
 	if ((correct_vg->status & PVMOVE) && !pvmove_mode()) {
@@ -725,6 +1019,19 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 	return correct_vg;
 }
 
+struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
+			     int *consistent)
+{
+	return _vg_read(cmd, vgname, consistent, 0);
+}
+
+struct volume_group *vg_read_precommitted(struct cmd_context *cmd,
+					  const char *vgname,
+					  int *consistent)
+{
+	return _vg_read(cmd, vgname, consistent, 1);
+}
+
 /* This is only called by lv_from_lvid, which is only called from 
  * activate.c so we know the appropriate VG lock is already held and 
  * the vg_read is therefore safe.
@@ -760,7 +1067,7 @@ struct volume_group *vg_read_by_vgid(struct cmd_context *cmd, const char *vgid)
 	 *       allowed to do a full scan here any more. */
 
 	// The slow way - full scan required to cope with vgrename 
-	if (!(vgnames = get_vgs(cmd, 1))) {
+	if (!(vgnames = get_vgs(cmd, 2))) {
 		log_error("vg_read_by_vgid: get_vgs failed");
 		return NULL;
 	}
@@ -814,7 +1121,8 @@ struct logical_volume *lv_from_lvid(struct cmd_context *cmd, const char *lvid_s)
 
 /* FIXME Use label functions instead of PV functions */
 struct physical_volume *pv_read(struct cmd_context *cmd, const char *pv_name,
-				struct list *mdas, uint64_t *label_sector)
+				struct list *mdas, uint64_t *label_sector,
+				int warnings)
 {
 	struct physical_volume *pv;
 	struct label *label;
@@ -823,12 +1131,14 @@ struct physical_volume *pv_read(struct cmd_context *cmd, const char *pv_name,
 
 	if (!(dev = dev_cache_get(pv_name, cmd->filter))) {
 		stack;
-		return 0;
+		return NULL;
 	}
 
 	if (!(label_read(dev, &label))) {
-		log_error("No physical volume label read from %s", pv_name);
-		return 0;
+		if (warnings)
+			log_error("No physical volume label read from %s",
+				  pv_name);
+		return NULL;
 	}
 
 	info = (struct lvmcache_info *) label->info;
@@ -837,22 +1147,29 @@ struct physical_volume *pv_read(struct cmd_context *cmd, const char *pv_name,
 
 	if (!(pv = pool_zalloc(cmd->mem, sizeof(*pv)))) {
 		log_error("pv allocation for '%s' failed", pv_name);
-		return 0;
+		return NULL;
 	}
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
 
 	/* FIXME Move more common code up here */
 	if (!(info->fmt->ops->pv_read(info->fmt, pv_name, pv, mdas))) {
 		log_error("Failed to read existing physical volume '%s'",
 			  pv_name);
-		return 0;
+		return NULL;
 	}
 
 	if (!pv->size)
 		return NULL;
-	else
-		return pv;
+	
+	if (!alloc_pv_segment_whole_pv(cmd->mem, pv)) {
+		stack;
+		return NULL;
+	}
+
+	return pv;
 }
 
 /* May return empty list */

lib/metadata/metadata.h

@@ -50,7 +50,7 @@
 #define VISIBLE_LV		0x00000040	/* LV */
 #define FIXED_MINOR		0x00000080	/* LV */
 /* FIXME Remove when metadata restructuring is completed */
-#define SNAPSHOT		0x00001000	/* LV - tmp internal use only */
+#define SNAPSHOT		0x00001000	/* LV - internal use only */
 #define PVMOVE			0x00002000	/* VG LV SEG */
 #define LOCKED			0x00004000	/* LV */
 #define MIRRORED		0x00008000	/* LV - internal use only */
@@ -68,6 +68,7 @@
 #define FMT_UNLIMITED_VOLS	0x00000008	/* Unlimited PVs/LVs? */
 #define FMT_RESTRICTED_LVIDS	0x00000010	/* LVID <= 255 */
 #define FMT_ORPHAN_ALLOCATABLE	0x00000020	/* Orphan PV allocatable? */
+#define FMT_PRECOMMIT		0x00000040	/* Supports pre-commit? */
 
 typedef enum {
 	ALLOC_INVALID,
@@ -98,6 +99,20 @@ struct format_type {
 	void *private;
 };
 
+struct pv_segment {
+	struct list list;	/* Member of pv->segments: ordered list
+				 * covering entire data area on this PV */
+
+	struct physical_volume *pv;
+	uint32_t pe;
+	uint32_t len;
+
+	struct lv_segment *lvseg;	/* NULL if free space */
+	uint32_t lv_area;	/* Index to area in LV segment */
+
+	struct list freelist;	/* Member of pv->free_segments */
+};
+
 struct physical_volume {
 	struct id id;
 	struct device *dev;
@@ -113,6 +128,8 @@ struct physical_volume {
 	uint32_t pe_count;
 	uint32_t pe_alloc_count;
 
+	struct list segments;	/* Ordered pv_segments covering complete PV */
+	struct list free_segments;	/* Free pv_segments for this PV */
 	struct list tags;
 };
 
@@ -124,6 +141,9 @@ struct metadata_area_ops {
 	struct volume_group *(*vg_read) (struct format_instance * fi,
 					 const char *vg_name,
 					 struct metadata_area * mda);
+	struct volume_group *(*vg_read_precommit) (struct format_instance * fi,
+					 const char *vg_name,
+					 struct metadata_area * mda);
 	/*
 	 * Write out complete VG metadata.  You must ensure internal
 	 * consistency before calling. eg. PEs can't refer to PVs not
@@ -140,6 +160,9 @@ struct metadata_area_ops {
 	 */
 	int (*vg_write) (struct format_instance * fid, struct volume_group * vg,
 			 struct metadata_area * mda);
+	int (*vg_precommit) (struct format_instance * fid,
+			     struct volume_group * vg,
+			     struct metadata_area * mda);
 	int (*vg_commit) (struct format_instance * fid,
 			  struct volume_group * vg, struct metadata_area * mda);
 	int (*vg_revert) (struct format_instance * fid,
@@ -184,11 +207,8 @@ struct volume_group {
 
 	/* logical volumes */
 	uint32_t lv_count;
-	struct list lvs;
-
-	/* snapshots */
 	uint32_t snapshot_count;
-	struct list snapshots;
+	struct list lvs;
 
 	struct list tags;
 };
@@ -210,7 +230,8 @@ struct lv_segment {
 	uint32_t area_len;
 	struct logical_volume *origin;
 	struct logical_volume *cow;
-	uint32_t chunk_size;
+	struct list origin_list;
+	uint32_t chunk_size;	/* In sectors */
 	uint32_t extents_copied;
 
 	struct list tags;
@@ -222,6 +243,7 @@ struct lv_segment {
 			struct {
 				struct physical_volume *pv;
 				uint32_t pe;
+				struct pv_segment *pvseg;
 			} pv;
 			struct {
 				struct logical_volume *lv;
@@ -246,26 +268,20 @@ struct logical_volume {
 	uint64_t size;
 	uint32_t le_count;
 
+	uint32_t origin_count;
+	struct list snapshot_segs;
+	struct lv_segment *snapshot;
+
 	struct list segments;
 	struct list tags;
 };
 
-struct snapshot {
-	struct id id;
-
-	int persistent;		/* boolean */
-	uint32_t chunk_size;	/* in 512 byte sectors */
-
-	struct logical_volume *origin;
-	struct logical_volume *cow;
-};
-
 struct name_list {
 	struct list list;
 	char *name;
 };
 
-struct alloc_area {
+struct pe_range {
 	struct list list;
 	uint32_t start;		/* PEs */
 	uint32_t count;		/* PEs */
@@ -275,7 +291,7 @@ 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 {
@@ -283,17 +299,16 @@ struct lv_list {
 	struct logical_volume *lv;
 };
 
-struct snapshot_list {
-	struct list list;
-
-	struct snapshot *snapshot;
-};
-
 struct mda_list {
 	struct list list;
 	struct device_area mda;
 };
 
+struct peg_list {
+	struct list list;
+	struct pv_segment *peg;
+};
+
 /*
  * Ownership of objects passes to caller.
  */
@@ -341,6 +356,12 @@ 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
 	 */
@@ -367,9 +388,13 @@ int vg_commit(struct volume_group *vg);
 int vg_revert(struct volume_group *vg);
 struct volume_group *vg_read(struct cmd_context *cmd, const char *vg_name,
 			     int *consistent);
+struct volume_group *vg_read_precommitted(struct cmd_context *cmd,
+					  const char *vg_name,
+					  int *consistent);
 struct volume_group *vg_read_by_vgid(struct cmd_context *cmd, const char *vgid);
 struct physical_volume *pv_read(struct cmd_context *cmd, const char *pv_name,
-				struct list *mdas, uint64_t *label_sector);
+				struct list *mdas, uint64_t *label_sector,
+				int warnings);
 struct list *get_pvs(struct cmd_context *cmd);
 
 /* Set full_scan to 1 to re-read every (filtered) device label */
@@ -399,13 +424,17 @@ int vg_rename(struct cmd_context *cmd, struct volume_group *vg,
 	      const char *new_name);
 int vg_extend(struct format_instance *fi, struct volume_group *vg,
 	      int pv_count, char **pv_names);
+int vg_change_pesize(struct cmd_context *cmd, struct volume_group *vg,
+		     uint32_t new_extent_size);
 
 /* Manipulate LVs */
 struct logical_volume *lv_create_empty(struct format_instance *fi,
 				       const char *name,
 				       const char *name_format,
+				       union lvid *lvid,
 				       uint32_t status,
 				       alloc_policy_t alloc,
+				       int import,
 				       struct volume_group *vg);
 
 int lv_reduce(struct format_instance *fi,
@@ -432,6 +461,8 @@ struct physical_volume *pv_find(struct volume_group *vg, const char *pv_name);
 struct pv_list *find_pv_in_vg(struct volume_group *vg, const char *pv_name);
 struct physical_volume *find_pv_in_vg_by_uuid(struct volume_group *vg,
 					      struct id *id);
+int get_pv_from_vg_by_id(const struct format_type *fmt, const char *vg_name,
+			 const char *id, struct physical_volume *pv);
 
 /* Find an LV within a given VG */
 struct lv_list *find_lv_in_vg(struct volume_group *vg, const char *lv_name);
@@ -485,13 +516,12 @@ int lv_is_cow(const struct logical_volume *lv);
 
 int pv_is_in_vg(struct volume_group *vg, struct physical_volume *pv);
 
-struct snapshot *find_cow(const struct logical_volume *lv);
-struct snapshot *find_origin(const struct logical_volume *lv);
-struct list *find_snapshots(const struct logical_volume *lv);
+struct lv_segment *find_cow(const struct logical_volume *lv);
 
-int vg_add_snapshot(struct logical_volume *origin,
-		    struct logical_volume *cow,
-		    int persistent, struct id *id, uint32_t chunk_size);
+int vg_add_snapshot(struct format_instance *fid, const char *name,
+		    struct logical_volume *origin, struct logical_volume *cow,
+		    union lvid *lvid, uint32_t extent_count,
+		    uint32_t chunk_size);
 
 int vg_remove_snapshot(struct volume_group *vg, struct logical_volume *cow);
 
@@ -500,7 +530,7 @@ 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

@@ -16,9 +16,10 @@
 #include "lib.h"
 #include "metadata.h"
 #include "toolcontext.h"
-#include "segtypes.h"
+#include "segtype.h"
 #include "display.h"
 #include "activate.h"
+#include "lv_alloc.h"
 
 /* 
  * Replace any LV segments on given PV with temporary mirror.
@@ -26,7 +27,7 @@
  */
 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 +35,15 @@ 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 +57,111 @@ 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;
+				set_lv_segment_area_lv(seg, s, lv_mirr, 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;
 		}
 	}
 
@@ -149,9 +224,9 @@ int remove_pvmove_mirrors(struct volume_group *vg,
 				else
 					c = 0;
 
-				seg->area[s].type = AREA_PV;
-				seg->area[s].u.pv.pv = mir_seg->area[c].u.pv.pv;
-				seg->area[s].u.pv.pe = mir_seg->area[c].u.pv.pe;
+				set_lv_segment_area_pv(seg, s,
+						       mir_seg->area[c].u.pv.pv,
+						       mir_seg->area[c].u.pv.pe);
 
 				/* Replace mirror with old area */
 				if (!
@@ -167,8 +242,12 @@ int remove_pvmove_mirrors(struct volume_group *vg,
 				lv1->status &= ~LOCKED;
 			}
 		}
+		if (!lv_merge_segments(lv1))
+			stack;
+
 	}
 
+
 	return 1;
 }