Merge branch '2018-05-29-radix-tree-iterate' into 2018-05-23-radix-tree-remove

.gitignore

@@ -79,5 +79,55 @@ test/lib/vgrename
 test/lib/vgs
 test/lib/vgscan
 test/lib/vgsplit
-
+test/api/lvtest.t
+test/api/pe_start.t
+test/api/percent.t
+test/api/python_lvm_unit.py
+test/api/test
+test/api/thin_percent.t
+test/api/vglist.t
+test/api/vgtest.t
+test/lib/aux
+test/lib/check
+test/lib/clvmd
+test/lib/dm-version-expected
+test/lib/dmeventd
+test/lib/dmsetup
+test/lib/dmstats
+test/lib/fail
+test/lib/flavour-ndev-cluster
+test/lib/flavour-ndev-cluster-lvmpolld
+test/lib/flavour-ndev-lvmetad
+test/lib/flavour-ndev-lvmetad-lvmpolld
+test/lib/flavour-ndev-lvmpolld
+test/lib/flavour-ndev-vanilla
+test/lib/flavour-udev-cluster
+test/lib/flavour-udev-cluster-lvmpolld
+test/lib/flavour-udev-lvmetad
+test/lib/flavour-udev-lvmetad-lvmpolld
+test/lib/flavour-udev-lvmlockd-dlm
+test/lib/flavour-udev-lvmlockd-sanlock
+test/lib/flavour-udev-lvmlockd-test
+test/lib/flavour-udev-lvmpolld
+test/lib/flavour-udev-vanilla
+test/lib/fsadm
+test/lib/get
+test/lib/inittest
+test/lib/invalid
+test/lib/lvm
+test/lib/lvm-wrapper
+test/lib/lvmchange
+test/lib/lvmdbusd.profile
+test/lib/lvmetad
+test/lib/lvmpolld
+test/lib/not
+test/lib/paths
+test/lib/paths-common
+test/lib/runner
+test/lib/should
+test/lib/test
+test/lib/thin-performance.profile
+test/lib/utils
+test/lib/version-expected
+test/unit/dmraid_t.c
 test/unit/unit-test

VERSION

@@ -1 +1 @@
-2.02.178(2)-git (2017-12-18)
+2.02.178(2)-git (2018-05-24)

VERSION_DM

@@ -1 +1 @@
-1.02.147-git (2017-12-18)
+1.02.147-git (2018-05-24)

WHATS_NEW

@@ -1,16 +1,22 @@
 Version 2.02.178 -
-=====================================
+====================================
+  Use versionsort to fix archive file expiry beyond 100000 files.
+
+Version 2.02.178-rc1 - 24th May 2018
+====================================
+  Add libaio dependency for build.
   Remove lvm1 and pool format handling and add filter to ignore them.
   Move some filter checks to after disks are read.
   Rework disk scanning and when it is used.
   Add new io layer and shift code to using it.
-  lvconvert: don't return success on degraded -m raid1 conversion
+  Fix lvconvert's return code on degraded -m raid1 conversion.
   --enable-testing switch for ./configure has been removed.
   --with-snapshots switch for ./configure has been removed.
   --with-mirrors switch for ./configure has been removed.
   --with-raid switch for ./configure has been removed.
   --with-thin switch for ./configure has been removed.
   --with-cache switch for ./configure has been removed.
+  Include new unit-test framework and unit tests.
   Extend validation of region_size for mirror segment.
   Reload whole device stack when reinitilizing mirror log.
   Mirrors without monitoring are WARNING and not blocking on error.
@@ -34,8 +40,8 @@ Version 2.02.178 -
   Enhance mirror log initialization for old mirror target.
   Skip private crypto and stratis devices.
   Skip frozen raid devices from scanning.
-  Activate RAID SubLVs on read_only_volume_list readwrite
+  Activate RAID SubLVs on read_only_volume_list readwrite.
-  Offer convenience type raid5_n converting to raid10
+  Offer convenience type raid5_n converting to raid10.
   Automatically avoid reading invalid snapshots during device scan.
   Ensure COW device is writable even for read-only thick snapshots.
   Support activation of component LVs in read-only mode.
@@ -53,20 +59,13 @@ Version 2.02.178 -
   Improve validation of created strings in vgimportclone.
   Add missing initialisation of mem pool in systemd generator.
   Do not reopen output streams for multithreaded users of liblvm.
-  Use versionsort to fix archive file expiry beyond 100000 files.
-  Add devices/use_aio, aio_max, aio_memory to configure AIO limits.
-  Support asynchronous I/O when scanning devices.
-  Detect asynchronous I/O capability in configure or accept --disable-aio.
-  Add AIO_SUPPORTED_CODE_PATH to indicate whether AIO may be used.
   Configure ensures /usr/bin dir is checked for dmpd tools.
   Restore pvmove support for wide-clustered active volumes (2.02.177).
   Avoid non-exclusive activation of exclusive segment types.
   Fix trimming sibling PVs when doing a pvmove of raid subLVs.
   Preserve exclusive activation during thin snaphost merge.
-  Suppress some repeated reads of the same disk data at the device layer.
   Avoid exceeding array bounds in allocation tag processing.
-  Refactor metadata reading code to use callback functions.
+  Add --lockopt to common options and add option to skip selected locks.
-  Move memory allocation for the key dev_reads into the device layer.
 
 Version 2.02.177 - 18th December 2017
 =====================================

WHATS_NEW_DM

@@ -1,5 +1,8 @@
 Version 1.02.147 -
-=====================================
+====================================
+
+Version 1.02.147-rc1 - 24th May 2018
+====================================
   Reuse uname() result for mirror target.
   Recognize also mounted btrfs through dm_device_has_mounted_fs().
   Add missing log_error() into dm_stats_populate() returning 0.

base/data-struct/radix-tree.c

@@ -75,17 +75,21 @@ struct node256 {
 struct radix_tree {
 	unsigned nr_entries;
 	struct value root;
+	radix_value_dtr dtr;
+	void *dtr_context;
 };
 
 //----------------------------------------------------------------
 
-struct radix_tree *radix_tree_create(void)
+struct radix_tree *radix_tree_create(radix_value_dtr dtr, void *dtr_context)
 {
 	struct radix_tree *rt = malloc(sizeof(*rt));
 
 	if (rt) {
 		rt->nr_entries = 0;
 		rt->root.type = UNSET;
+		rt->dtr = dtr;
+		rt->dtr_context = dtr_context;
 	}
 
 	return rt;
@@ -154,9 +158,9 @@ static void _free_node(struct value v, radix_value_dtr dtr, void *context)
 	}
 }
 
-void radix_tree_destroy(struct radix_tree *rt, radix_value_dtr dtr, void *context)
+void radix_tree_destroy(struct radix_tree *rt)
 {
-	_free_node(rt->root, dtr, context);
+	_free_node(rt->root, rt->dtr, rt->dtr_context);
 	free(rt);
 }
 

base/data-struct/radix-tree.h

@@ -25,12 +25,11 @@ union radix_value {
 	uint64_t n;
 };
 
-struct radix_tree *radix_tree_create(void);
-
 typedef void (*radix_value_dtr)(void *context, union radix_value v);
 
-// dtr may be NULL
+// dtr will be called on any deleted entries.  dtr may be NULL.
-void radix_tree_destroy(struct radix_tree *rt, radix_value_dtr dtr, void *context);
+struct radix_tree *radix_tree_create(radix_value_dtr dtr, void *dtr_context);
+void radix_tree_destroy(struct radix_tree *rt);
 
 unsigned radix_tree_size(struct radix_tree *rt);
 bool radix_tree_insert(struct radix_tree *rt, uint8_t *kb, uint8_t *ke, union radix_value v);

doc/release-notes/2.02.178

@@ -0,0 +1,53 @@
+Version 2.02.178
+================
+
+There are going to be some large changes to the lvm2 codebase
+over the next year or so.  Starting with this release.  These
+changes should be internal rather than having a big effect on
+the command line.  Inevitably these changes will increase the
+chance of bugs, so please be on the alert.
+
+
+Remove support for obsolete metadata formats
+--------------------------------------------
+
+Support for the GFS pool format, and format used by the
+original 1990's version of LVM1 have been removed.
+
+Use asynchronous IO
+-------------------
+
+Almost all IO uses libaio now.
+
+Rewrite label scanning
+----------------------
+
+Dave Teigland has reworked the label scanning and metadata reading
+logic to minimise the amount of IOs issued.  Combined with the aio changes 
+this can greatly improve scanning speed for some systems.
+
+./configure options
+-------------------
+
+We're going to try and remove as many options from ./configure as we
+can.  Each option multiplies the number of possible configurations
+that we should test (this testing is currently not occurring).
+
+The first batch to be removed are:
+
+  --enable-testing
+  --with-snapshots
+  --with-mirrors
+  --with-raid
+  --with-thin
+  --with-cache
+
+Stable targets that are in the upstream kernel will just be supported.
+
+In future optional target flags will be given in two situations:
+
+1) The target is experimental, or not upstream at all (eg, vdo).
+2) The target is deprecated and support will be removed at some future date.
+
+This decision could well be contentious, so could distro maintainers feel
+free to comment.

lib/device/dev-cache.c

@@ -1080,6 +1080,8 @@ void dev_cache_scan(void)
 {
 	struct dir_list *dl;
 	
+	log_debug_devs("Creating list of system devices.");
+
 	_cache.has_scanned = 1;
 
 	_insert_dirs(&_cache.dirs);

lib/format_text/archive.c

@@ -135,8 +135,8 @@ static struct dm_list *_scan_archive(struct dm_pool *mem,
 
 	dm_list_init(results);
 
-	/* Sort fails beyond 5-digit indexes */
+	/* Use versionsort to handle numbers beyond 5 digits */
-	if ((count = scandir(dir, &dirent, NULL, alphasort)) < 0) {
+	if ((count = scandir(dir, &dirent, NULL, versionsort)) < 0) {
 		log_error("Couldn't scan the archive directory (%s).", dir);
 		return 0;
 	}

lib/label/label.c

@@ -427,7 +427,11 @@ static int _process_block(struct cmd_context *cmd, struct dev_filter *f,
 
 static int _scan_dev_open(struct device *dev)
 {
+	struct dm_list *name_list;
+	struct dm_str_list *name_sl;
 	const char *name;
+	struct stat sbuf;
+	int retried = 0;
 	int flags = 0;
 	int fd;
 
@@ -435,20 +439,30 @@ static int _scan_dev_open(struct device *dev)
 		return 0;
 
 	if (dev->flags & DEV_IN_BCACHE) {
-		log_error("scan_dev_open %s DEV_IN_BCACHE already set", dev_name(dev));
+		/* Shouldn't happen */
+		log_error("Device open %s has DEV_IN_BCACHE already set", dev_name(dev));
 		dev->flags &= ~DEV_IN_BCACHE;
 	}
 
 	if (dev->bcache_fd > 0) {
-		log_error("scan_dev_open %s already open with fd %d",
+		/* Shouldn't happen */
+		log_error("Device open %s already open with fd %d",
 			  dev_name(dev), dev->bcache_fd);
 		return 0;
 	}
 
-	if (!(name = dev_name_confirmed(dev, 1))) {
+	/*
-		log_error("scan_dev_open %s no name", dev_name(dev));
+	 * All the names for this device (major:minor) are kept on
+	 * dev->aliases, the first one is the primary/preferred name.
+	 */
+	if (!(name_list = dm_list_first(&dev->aliases))) {
+		/* Shouldn't happen */
+		log_error("Device open %s %d:%d has no path names.",
+			  dev_name(dev), (int)MAJOR(dev->dev), (int)MINOR(dev->dev));
 		return 0;
 	}
+	name_sl = dm_list_item(name_list, struct dm_str_list);
+	name = name_sl->str;
 
 	flags |= O_RDWR;
 	flags |= O_DIRECT;
@@ -457,6 +471,8 @@ static int _scan_dev_open(struct device *dev)
 	if (dev->flags & DEV_BCACHE_EXCL)
 		flags |= O_EXCL;
 
+retry_open:
+
 	fd = open(name, flags, 0777);
 
 	if (fd < 0) {
@@ -464,7 +480,48 @@ static int _scan_dev_open(struct device *dev)
 			log_error("Can't open %s exclusively.  Mounted filesystem?",
 				  dev_name(dev));
 		} else {
-			log_error("scan_dev_open %s failed errno %d", dev_name(dev), errno);
+			int major, minor;
+
+			/*
+			 * Shouldn't happen, if it does, print stat info to help figure
+			 * out what's wrong.
+			 */
+
+			major = (int)MAJOR(dev->dev);
+			minor = (int)MINOR(dev->dev);
+
+			log_error("Device open %s %d:%d failed errno %d", name, major, minor, errno);
+
+			if (stat(name, &sbuf)) {
+				log_debug_devs("Device open %s %d:%d stat failed errno %d",
+					       name, major, minor, errno);
+			} else if (sbuf.st_rdev != dev->dev) {
+				log_debug_devs("Device open %s %d:%d stat %d:%d does not match.",
+					       name, major, minor,
+					       (int)MAJOR(sbuf.st_rdev), (int)MINOR(sbuf.st_rdev));
+			}
+
+			/*
+			 * FIXME: do we want to try opening this device using
+			 * one of the other path aliases for the same
+			 * major:minor from dev->aliases?  We could iterate
+			 * through those aliases to try opening each of them to
+			 * find one that works.  What are the consequences of
+			 * using a different, non-preferred alias to a device?
+			 */
+
+			if (!retried) {
+				/*
+				 * FIXME: remove this, the theory for this retry is that
+				 * there may be a udev race that we can sometimes mask by
+				 * retrying.  This is here until we can figure out if it's
+				 * needed and if so fix the real problem.
+				 */
+				usleep(5000);
+				log_debug_devs("Device open %s retry", dev_name(dev));
+				retried = 1;
+				goto retry_open;
+			}
 		}
 		return 0;
 	}
@@ -509,9 +566,10 @@ static int _scan_list(struct cmd_context *cmd, struct dev_filter *f,
 {
 	struct dm_list wait_devs;
 	struct dm_list done_devs;
+	struct dm_list reopen_devs;
 	struct device_list *devl, *devl2;
 	struct block *bb;
-	int scan_open_errors = 0;
+	int retried_open = 0;
 	int scan_read_errors = 0;
 	int scan_process_errors = 0;
 	int scan_failed_count = 0;
@@ -524,6 +582,7 @@ static int _scan_list(struct cmd_context *cmd, struct dev_filter *f,
 
 	dm_list_init(&wait_devs);
 	dm_list_init(&done_devs);
+	dm_list_init(&reopen_devs);
 
 	log_debug_devs("Scanning %d devices for VG info", dm_list_size(devs));
 
@@ -547,9 +606,7 @@ static int _scan_list(struct cmd_context *cmd, struct dev_filter *f,
 			if (!_scan_dev_open(devl->dev)) {
 				log_debug_devs("Scan failed to open %s.", dev_name(devl->dev));
 				dm_list_del(&devl->list);
-				dm_list_add(&done_devs, &devl->list);
+				dm_list_add(&reopen_devs, &devl->list);
-				scan_open_errors++;
-				scan_failed_count++;
 				continue;
 			}
 		}
@@ -612,8 +669,44 @@ static int _scan_list(struct cmd_context *cmd, struct dev_filter *f,
 	if (!dm_list_empty(devs))
 		goto scan_more;
 
-	log_debug_devs("Scanned devices: open errors %d read errors %d process errors %d",
+	/*
-			scan_open_errors, scan_read_errors, scan_process_errors);
+	 * We're done scanning all the devs.  If we failed to open any of them
+	 * the first time through, refresh device paths and retry.  We failed
+	 * to open the devs on the reopen_devs list.
+	 *
+	 * FIXME: it's not clear if or why this helps.
+	 *
+	 * FIXME: should we delete the first path name from dev->aliases that
+	 * we failed to open the first time before retrying?  If that path
+	 * still exists on the system, dev_cache_scan should put it back, but
+	 * if it doesn't exist we don't want to try using it again.
+	 */
+	if (!dm_list_empty(&reopen_devs)) {
+		if (retried_open) {
+			/* Don't try again. */
+			scan_failed_count += dm_list_size(&reopen_devs);
+			dm_list_splice(&done_devs, &reopen_devs);
+			goto out;
+		}
+		retried_open = 1;
+
+		/*
+		 * This will search the system's /dev for new path names and
+		 * could help us reopen the device if it finds a new preferred
+		 * path name for this dev's major:minor.  It does that by
+		 * inserting a new preferred path name on dev->aliases.  open
+		 * uses the first name from that list.
+		 */
+		log_debug_devs("Scanning refreshing device paths.");
+		dev_cache_scan();
+
+		/* Put devs that failed to open back on the original list to retry. */
+		dm_list_splice(devs, &reopen_devs);
+		goto scan_more;
+	}
+out:
+	log_debug_devs("Scanned devices: read errors %d process errors %d failed %d",
+			scan_read_errors, scan_process_errors, scan_failed_count);
 
 	if (failed)
 		*failed = scan_failed_count;

man/dmeventd.8_main

@@ -123,7 +123,7 @@ Command is executed with environmental variable
 in this environment will not try to interact with dmeventd.
 To see the fullness of a thin pool command may check these
 two environmental variables
-\fBDMEVENTD_THIN_POOL_DATA\fP and \fBDMEVENTD_THIN_POOL_DATA\fP.
+\fBDMEVENTD_THIN_POOL_DATA\fP and \fBDMEVENTD_THIN_POOL_METADATA\fP.
 Command can also read status with tools like \fBlvs\fP(8).
 .
 .SH ENVIRONMENT VARIABLES
@@ -134,7 +134,7 @@ Variable is set by thin plugin and is available to executed program. Value prese
 actual usage of thin pool data volume. Variable is not set when error event
 is processed.
 .TP
-.B DMEVENTD_THIN_POOL_DATA
+.B DMEVENTD_THIN_POOL_METADATA
 Variable is set by thin plugin and is available to executed program. Value present
 actual usage of thin pool metadata volume. Variable is not set when error event
 is processed.

man/vgexport.8_pregen

@@ -8,11 +8,22 @@ vgexport - Unregister volume group(s) from the system
     [ \fIoption_args\fP ]
 .br
 .SH DESCRIPTION
-vgexport makes inactive VGs unknown to the system. In this state, all the
+vgexport changes a VG into the exported state, which ensures that the VG
-PVs in the VG can be moved to a different system, from which
+and its disks are not being used, and cannot be used until the VG is
-\fBvgimport\fP(8) can then be run.
+imported by \fBvgimport\fP(8).  Putting a VG into an unusable, offline
+state can be useful when doing things like moving a VG's disks to another
+system.  Exporting a VG provides some protection from its LVs being
+accidentally used, or being used by an automated system before it's ready.
 
-Most LVM tools ignore exported VGs.
+A VG cannot be exported until all of its LVs are inactive.
+
+LVM commands will ignore an exported VG or report an error if a command
+tries to use it.
+
+For an exported VG, the vgs command will display \"x\" in the third VG
+attribute, and the pvs command will display \"x\" in the second PV
+attribute.  Both vgs and pvs will display \"exported\" from the export
+report field.
 
 vgexport clears the VG system ID, and vgimport sets the VG system ID to
 match the host running vgimport (if the host has a system ID).

test/lib/aux.sh

@@ -755,11 +755,14 @@ prepare_md_dev() {
 	local coption="--chunk"
 	local maj
 	local mddev
+	local mddir="md/"
+	local mdname
+	local mddevdir
 	maj=$(mdadm --version 2>&1) || skip "mdadm tool is missing!"
 
 	cleanup_md_dev
 
-	rm -f debug.log strace.log MD_DEV MD_DEV_PV MD_DEVICES
+	rm -f debug.log strace.log
 
 	case "$level" in
 	"1")  coption="--bitmap-chunk" ;;
@@ -770,9 +773,11 @@ prepare_md_dev() {
 	# - newer mdadm _completely_ defers to udev to create the associated device node
 	maj=${maj##*- v}
 	maj=${maj%%.*}
-	[ "$maj" -ge 3 ] && \
+	[ "$maj" -ge 3 ] || mddir=""
-		mddev=/dev/md/md_lvm_test0 || \
+
-		mddev=/dev/md_lvm_test0
+	mdname="md_lvm_test0"
+	mddev="/dev/${mddir}$mdname"
+	mddevdir="$DM_DEV_DIR/$mddir"
 
 	mdadm --create --metadata=1.0 "$mddev" --auto=md --level "$level" $with_bitmap "$coption"="$rchunk" --raid-devices="$rdevs" "${@:4}" || {
 		# Some older 'mdadm' version managed to open and close devices internaly
@@ -791,10 +796,11 @@ prepare_md_dev() {
 
 	# LVM/DM will see this device
 	case "$DM_DEV_DIR" in
-	"/dev") readlink -f "$mddev" ;;
+	"/dev") readlink -f "$mddev" > MD_DEV_PV ;;
-	*)	cp -LR "$mddev" "$DM_DEV_DIR"
+	*)	mkdir -p "$mddevdir"
-		echo "$DM_DEV_DIR/md_lvm_test0" ;;
+		cp -LR "$mddev" "$mddevdir"
-	esac > MD_DEV_PV
+		echo "${mddevdir}${mdname}" > MD_DEV_PV ;;
+	esac
 	echo "$mddev" > MD_DEV
 	notify_lvmetad "$(< MD_DEV_PV)"
 	printf "%s\n" "${@:4}" > MD_DEVICES
@@ -809,12 +815,14 @@ cleanup_md_dev() {
 	local IFS=$IFS_NL
 	local dev
 	local mddev
+	local mddev_pv
 	mddev=$(< MD_DEV)
+	mddev_pv=$(< MD_DEV_PV)
 	udev_wait
 	mdadm --stop "$mddev" || true
-	test "$DM_DEV_DIR" != "/dev" && rm -f "$DM_DEV_DIR/$(basename "$mddev")"
+	notify_lvmetad "$mddev_pv"
-	notify_lvmetad "$(< MD_DEV_PV)"
 	udev_wait  # wait till events are process, not zeroing to early
+	test "$DM_DEV_DIR" != "/dev" && rm -rf "${mddev_pv%/*}"
 	for dev in $(< MD_DEVICES); do
 		mdadm --zero-superblock "$dev" || true
 		notify_lvmetad "$dev"
@@ -1167,7 +1175,7 @@ extend_filter() {
 }
 
 extend_filter_LVMTEST() {
-	extend_filter "a|$DM_DEV_DIR/$PREFIX|"
+	extend_filter "a|$DM_DEV_DIR/$PREFIX|" "$@"
 }
 
 hide_dev() {

test/shell/lvmetad-pvscan-md.sh

@@ -25,8 +25,7 @@ aux prepare_devs 2
 aux prepare_md_dev 0 64 2 "$dev1" "$dev2"
 
 aux lvmconf 'devices/md_component_detection = 1'
-aux extend_filter_LVMTEST
+aux extend_filter_LVMTEST "a|/dev/md|"
-aux extend_filter "a|/dev/md.*|"
 
 pvdev=$(< MD_DEV_PV)
 

test/shell/pvcreate-md-fake-hdr.sh

@@ -0,0 +1,94 @@
+#!/usr/bin/env bash
+
+# Copyright (C) 2018 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.
+#
+# 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+
+
+# TODO: once code get fixed, add matching 'check' calls
+SKIP_WITH_LVMLOCKD=1
+SKIP_WITH_LVMPOLLD=1
+
+. lib/inittest
+
+test -f /proc/mdstat && grep -q raid1 /proc/mdstat || \
+	modprobe raid1 || skip
+
+aux lvmconf 'devices/md_component_detection = 1'
+aux extend_filter_LVMTEST "a|/dev/md|"
+
+aux prepare_devs 4
+
+vgcreate $vg "$dev3" "$dev4"
+
+# create 2 disk MD raid1 array
+# by default using metadata format 1.0 with data at the end of device
+aux prepare_md_dev 1 64 2 "$dev1" "$dev2"
+
+mddev=$(< MD_DEV)
+pvdev=$(< MD_DEV_PV)
+sleep 3
+mdadm --stop "$mddev"
+
+# copy fake PV/VG header PV3 -> PV2 (which is however md raid1 leg)
+dd if="$dev3"  of="$dev2" bs=64k count=1 conv=fdatasync
+
+# remove VG on PV3 & PV4
+vgremove -f $vg
+
+sleep 3
+aux udev_wait
+# too bad  'dd' wakes up  md array reassembling
+should not mdadm --detail "$mddev"
+should not mdadm --stop "$mddev"
+sleep 3
+
+# print what  blkid thinks about each PV
+for i in "$dev1" "$dev2" "$dev3" "$dev4"
+do
+	blkid -c /dev/null -w /dev/null "$i" || echo "Unknown signature"
+done
+
+# expect open count for each PV to be 0
+dmsetup info -c
+
+pvs -vvvv  "$dev2" "$dev3" || true
+
+# still expect open count for each PV to be 0
+dmsetup info -c
+
+pvs -vvvv  "$dev3" "$dev2" || true
+
+# and again we expect open count for each PV to be 0
+dmsetup info -c
+dmsetup table
+
+# even after 3 second of possible hidden raid array assembling
+sleep 3
+dmsetup info -c
+
+# if for any reason array went up - stop it again
+if mdadm --detail "$mddev" ; then
+	mdadm --stop "$mddev"
+	aux udev_wait
+	should not mdadm --detail "$mddev"
+fi
+
+# now reassemble array from  PV1 & PV2
+mdadm --assemble --verbose "$mddev" "$dev1" "$dev2"
+aux udev_wait
+sleep 1
+
+# and let 'fake hdr' to be fixed from master/primary leg
+# (when mdadm supports repair)
+if mdadm --action=repair "$mddev" ; then
+	sleep 1
+	# should be showing correctly PV3 & PV4
+	pvs
+fi

test/shell/pvcreate-operation-md.sh

@@ -83,13 +83,19 @@ EOF
     if aux kernel_at_least 2 6 33 ; then
 	# in case the system is running without devtmpfs /dev
 	# wait here for created device node on tmpfs
-	test "$DM_DEV_DIR" != "/dev" && cp -LR "${mddev}p1" "$DM_DEV_DIR"
+	test "$DM_DEV_DIR" = "/dev" || cp -LR "${mddev}p1" "${pvdev%/*}"
 
 	pvcreate --metadatasize 128k "${pvdev}p1"
 
 	maj=$(($(stat -L --printf=0x%t "${mddev}p1")))
 	min=$(($(stat -L --printf=0x%T "${mddev}p1")))
 
+	ls /sys/dev/block/$maj:$min/
+	ls /sys/dev/block/$maj:$min/holders/
+	cat /sys/dev/block/$maj:$min/dev
+	cat /sys/dev/block/$maj:$min/stat
+	cat /sys/dev/block/$maj:$min/size
+
 	sysfs_alignment_offset="/sys/dev/block/$maj:$min/alignment_offset"
 	[ -f "$sysfs_alignment_offset" ] && \
 		alignment_offset=$(< "$sysfs_alignment_offset") || \
@@ -100,20 +106,25 @@ EOF
 	check pv_field "${pvdev}p1" pe_start $pv_align --units b --nosuffix
 
 	pvremove "${pvdev}p1"
-	test "$DM_DEV_DIR" != "/dev" && rm -f "$DM_DEV_DIR/${mddev}p1"
+	test "$DM_DEV_DIR" = "/dev" || rm -f "${pvdev}p1"
     fi
 fi
 
 # Test newer topology-aware alignment detection w/ --dataalignment override
 if aux kernel_at_least 2 6 33 ; then
     # make sure we're clean for another test
-    dd if=/dev/zero of="$mddev" bs=512 count=1
+    dd if=/dev/zero of="$mddev" bs=512 count=4 conv=fdatasync
+    partprobe -s "$mddev"
     aux prepare_md_dev 0 1024 2 "$dev1" "$dev2"
     pvdev=$(< MD_DEV_PV)
 
     # optimal_io_size=2097152, minimum_io_size=1048576
     pvcreate --metadatasize 128k \
 	--config 'devices { md_chunk_alignment=0 }' "$pvdev"
+
+    # to see the processing of scanning
+    pvs -vvvv
+
     check pv_field "$pvdev" pe_start "2.00m"
 
     # now verify pe_start alignment override using --dataalignment

test/unit/radix_tree_t.c

@@ -21,14 +21,14 @@
 
 static void *rt_init(void)
 {
-	struct radix_tree *rt = radix_tree_create();
+	struct radix_tree *rt = radix_tree_create(NULL, NULL);
 	T_ASSERT(rt);
 	return rt;
 }
 
 static void rt_exit(void *fixture)
 {
-	radix_tree_destroy(fixture, NULL, NULL);
+	radix_tree_destroy(fixture);
 }
 
 static void test_create_destroy(void *fixture)