Add support for autoexpand property

While the autoexpand property may seem like a small feature it
depends on a significant amount of system infrastructure.  Enough
of that infrastructure is now in place that with a few modifications
for Linux it can be supported.

Auto-expand works as follows; when a block device is modified
(re-sized, closed after being open r/w, etc) a change uevent is
generated for udev.  The ZED, which is monitoring udev events,
passes the change event along to zfs_deliver_dle() if the disk
or partition contains a zfs_member as identified by blkid.

From here the device is matched against all imported pool vdevs
using the vdev_guid which was read from the label by blkid.  If
a match is found the ZED reopens the pool vdev.  This re-opening
is important because it allows the vdev to be briefly closed so
the disk partition table can be re-read.  Otherwise, it wouldn't
be possible to report the maximum possible expansion size.

Finally, if the property autoexpand=on a vdev expansion will be
attempted.  After performing some sanity checks on the disk to
verify that it is safe to expand,  the primary partition (-part1)
will be expanded and the partition table updated.  The partition
is then re-opened (again) to detect the updated size which allows
the new capacity to be used.

In order to make all of the above possible the following changes
were required:

* Updated the zpool_expand_001_pos and zpool_expand_003_pos tests.
  These tests now create a pool which is layered on a loopback,
  scsi_debug, and file vdev.  This allows for testing of non-
  partitioned block device (loopback), a partition block device
  (scsi_debug), and a file which does not receive udev change
  events.  This provided for better test coverage, and by removing
  the layering on ZFS volumes there issues surrounding layering
  one pool on another are avoided.

* zpool_find_vdev_by_physpath() updated to accept a vdev guid.
  This allows for matching by guid rather than path which is a
  more reliable way for the ZED to reference a vdev.

* Fixed zfs_zevent_wait() signal handling which could result
  in the ZED spinning when a signal was not handled.

* Removed vdev_disk_rrpart() functionality which can be abandoned
  in favor of kernel provided blkdev_reread_part() function.

* Added a rwlock which is held as a writer while a disk is being
  reopened.  This is important to prevent errors from occurring
  for any configuration related IOs which bypass the SCL_ZIO lock.
  The zpool_reopen_007_pos.ksh test case was added to verify IO
  error are never observed when reopening.  This is not expected
  to impact IO performance.

Additional fixes which aren't critical but were discovered and
resolved in the course of developing this functionality.

* Added PHYS_PATH="/dev/zvol/dataset" to the vdev configuration for
  ZFS volumes.  This is as good as a unique physical path, while the
  volumes are not used in the test cases anymore for other reasons
  this improvement was included.

Reviewed by: Richard Elling <Richard.Elling@RichardElling.com>
Signed-off-by: Sara Hartse <sara.hartse@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #120
Closes #2437
Closes #5771
Closes #7366
Closes #7582
Closes #7629

cmd/zed/agents/zfs_mod.c

@@ -697,8 +697,8 @@ zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
 {
 	char *devname = data;
 	boolean_t avail_spare, l2cache;
-	vdev_state_t newstate;
 	nvlist_t *tgt;
+	int error;
 
 	zed_log_msg(LOG_INFO, "zfsdle_vdev_online: searching for '%s' in '%s'",
 	    devname, zpool_get_name(zhp));
@@ -706,42 +706,58 @@ zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
 	if ((tgt = zpool_find_vdev_by_physpath(zhp, devname,
 	    &avail_spare, &l2cache, NULL)) != NULL) {
 		char *path, fullpath[MAXPATHLEN];
-		uint64_t wholedisk = 0ULL;
+		uint64_t wholedisk;
 
-		verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
+		error = nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH, &path);
-		    &path) == 0);
+		if (error) {
-		verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
+			zpool_close(zhp);
-		    &wholedisk) == 0);
-
-		(void) strlcpy(fullpath, path, sizeof (fullpath));
-		if (wholedisk) {
-			char *spath = zfs_strip_partition(fullpath);
-			boolean_t scrub_restart = B_TRUE;
-
-			if (!spath) {
-				zed_log_msg(LOG_INFO, "%s: Can't alloc",
-				    __func__);
 			return (0);
 		}
 
-			(void) strlcpy(fullpath, spath, sizeof (fullpath));
+		error = nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
-			free(spath);
+		    &wholedisk);
+		if (error)
+			wholedisk = 0;
+
+		if (wholedisk) {
+			path = strrchr(path, '/');
+			if (path != NULL) {
+				path = zfs_strip_partition(path + 1);
+				if (path == NULL) {
+					zpool_close(zhp);
+					return (0);
+				}
+			} else {
+				zpool_close(zhp);
+				return (0);
+			}
+
+			(void) strlcpy(fullpath, path, sizeof (fullpath));
+			free(path);
 
 			/*
 			 * We need to reopen the pool associated with this
-			 * device so that the kernel can update the size
+			 * device so that the kernel can update the size of
-			 * of the expanded device.
+			 * the expanded device.  When expanding there is no
+			 * need to restart the scrub from the beginning.
 			 */
+			boolean_t scrub_restart = B_FALSE;
 			(void) zpool_reopen_one(zhp, &scrub_restart);
+		} else {
+			(void) strlcpy(fullpath, path, sizeof (fullpath));
 		}
 
 		if (zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
-			zed_log_msg(LOG_INFO, "zfsdle_vdev_online: setting "
+			vdev_state_t newstate;
-			    "device '%s' to ONLINE state in pool '%s'",
+
-			    fullpath, zpool_get_name(zhp));
+			if (zpool_get_state(zhp) != POOL_STATE_UNAVAIL) {
-			if (zpool_get_state(zhp) != POOL_STATE_UNAVAIL)
+				error = zpool_vdev_online(zhp, fullpath, 0,
-				(void) zpool_vdev_online(zhp, fullpath, 0,
 				    &newstate);
+				zed_log_msg(LOG_INFO, "zfsdle_vdev_online: "
+				    "setting device '%s' to ONLINE state "
+				    "in pool '%s': %d", fullpath,
+				    zpool_get_name(zhp), error);
+			}
 		}
 		zpool_close(zhp);
 		return (1);
@@ -751,23 +767,32 @@ zfsdle_vdev_online(zpool_handle_t *zhp, void *data)
 }
 
 /*
- * This function handles the ESC_DEV_DLE event.
+ * This function handles the ESC_DEV_DLE device change event.  Use the
+ * provided vdev guid when looking up a disk or partition, when the guid
+ * is not present assume the entire disk is owned by ZFS and append the
+ * expected -part1 partition information then lookup by physical path.
  */
 static int
 zfs_deliver_dle(nvlist_t *nvl)
 {
-	char *devname;
+	char *devname, name[MAXPATHLEN];
+	uint64_t guid;
 
-	if (nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devname) != 0) {
+	if (nvlist_lookup_uint64(nvl, ZFS_EV_VDEV_GUID, &guid) == 0) {
-		zed_log_msg(LOG_INFO, "zfs_deliver_dle: no physpath");
+		sprintf(name, "%llu", (u_longlong_t)guid);
-		return (-1);
+	} else if (nvlist_lookup_string(nvl, DEV_PHYS_PATH, &devname) == 0) {
+		strlcpy(name, devname, MAXPATHLEN);
+		zfs_append_partition(name, MAXPATHLEN);
+	} else {
+		zed_log_msg(LOG_INFO, "zfs_deliver_dle: no guid or physpath");
 	}
 
-	if (zpool_iter(g_zfshdl, zfsdle_vdev_online, devname) != 1) {
+	if (zpool_iter(g_zfshdl, zfsdle_vdev_online, name) != 1) {
 		zed_log_msg(LOG_INFO, "zfs_deliver_dle: device '%s' not "
-		    "found", devname);
+		    "found", name);
 		return (1);
 	}
+
 	return (0);
 }
 

config/kernel-blkdev-get.m4

@@ -1,19 +0,0 @@
-dnl #
-dnl # 2.6.37 API change
-dnl # Added 3rd argument for the active holder, previously this was
-dnl # hardcoded to NULL.
-dnl #
-AC_DEFUN([ZFS_AC_KERNEL_3ARG_BLKDEV_GET], [
-	AC_MSG_CHECKING([whether blkdev_get() wants 3 args])
-	ZFS_LINUX_TRY_COMPILE([
-		#include <linux/fs.h>
-	],[
-		struct block_device *bdev = NULL;
-		(void) blkdev_get(bdev, 0, NULL);
-	],[
-		AC_MSG_RESULT(yes)
-		AC_DEFINE(HAVE_3ARG_BLKDEV_GET, 1, [blkdev_get() wants 3 args])
-	],[
-		AC_MSG_RESULT(no)
-	])
-])

config/kernel-blkdev-reread-part.m4

@@ -0,0 +1,21 @@
+dnl #
+dnl # 4.1 API, exported blkdev_reread_part() symbol, backported to the
+dnl # 3.10.0 CentOS 7.x enterprise kernels.
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_BLKDEV_REREAD_PART], [
+	AC_MSG_CHECKING([whether blkdev_reread_part() is available])
+	ZFS_LINUX_TRY_COMPILE([
+		#include <linux/fs.h>
+	], [
+		struct block_device *bdev = NULL;
+		int error;
+
+		error = blkdev_reread_part(bdev);
+	], [
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_BLKDEV_REREAD_PART, 1,
+		    [blkdev_reread_part() is available])
+	], [
+		AC_MSG_RESULT(no)
+	])
+])

config/kernel-get-gendisk.m4

@@ -1,17 +0,0 @@
-dnl #
-dnl # 2.6.34 API change
-dnl # Verify the get_gendisk() symbol is available.
-dnl #
-AC_DEFUN([ZFS_AC_KERNEL_GET_GENDISK],
-	[AC_MSG_CHECKING([whether get_gendisk() is available])
-	ZFS_LINUX_TRY_COMPILE_SYMBOL([
-		#include <linux/genhd.h>
-	], [
-		get_gendisk(0, NULL);
-	], [get_gendisk], [block/genhd.c], [
-		AC_MSG_RESULT(yes)
-		AC_DEFINE(HAVE_GET_GENDISK, 1, [get_gendisk() is available])
-	], [
-		AC_MSG_RESULT(no)
-	])
-])

config/kernel.m4

@@ -44,8 +44,8 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [
 	ZFS_AC_KERNEL_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS
 	ZFS_AC_KERNEL_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID
 	ZFS_AC_KERNEL_TYPE_FMODE_T
-	ZFS_AC_KERNEL_3ARG_BLKDEV_GET
 	ZFS_AC_KERNEL_BLKDEV_GET_BY_PATH
+	ZFS_AC_KERNEL_BLKDEV_REREAD_PART
 	ZFS_AC_KERNEL_OPEN_BDEV_EXCLUSIVE
 	ZFS_AC_KERNEL_LOOKUP_BDEV
 	ZFS_AC_KERNEL_INVALIDATE_BDEV_ARGS
@@ -73,7 +73,6 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [
 	ZFS_AC_KERNEL_BLK_QUEUE_HAVE_BLK_PLUG
 	ZFS_AC_KERNEL_GET_DISK_AND_MODULE
 	ZFS_AC_KERNEL_GET_DISK_RO
-	ZFS_AC_KERNEL_GET_GENDISK
 	ZFS_AC_KERNEL_HAVE_BIO_SET_OP_ATTRS
 	ZFS_AC_KERNEL_GENERIC_READLINK_GLOBAL
 	ZFS_AC_KERNEL_DISCARD_GRANULARITY

include/linux/blkdev_compat.h

@@ -364,6 +364,20 @@ bio_set_bi_error(struct bio *bio, int error)
 #define	vdev_bdev_close(bdev, md)	close_bdev_excl(bdev)
 #endif /* HAVE_BLKDEV_GET_BY_PATH | HAVE_OPEN_BDEV_EXCLUSIVE */
 
+/*
+ * 4.1 - x.y.z API,
+ * 3.10.0 CentOS 7.x API,
+ *   blkdev_reread_part()
+ *
+ * For older kernels trigger a re-reading of the partition table by calling
+ * check_disk_change() which calls flush_disk() to invalidate the device.
+ */
+#ifdef HAVE_BLKDEV_REREAD_PART
+#define	vdev_bdev_reread_part(bdev)	blkdev_reread_part(bdev)
+#else
+#define	vdev_bdev_reread_part(bdev)	check_disk_change(bdev)
+#endif /* HAVE_BLKDEV_REREAD_PART */
+
 /*
  * 2.6.22 API change
  * The function invalidate_bdev() lost it's second argument because

include/sys/vdev_disk.h

@@ -47,6 +47,7 @@ typedef struct vdev_disk {
 	ddi_devid_t		vd_devid;
 	char			*vd_minor;
 	struct block_device	*vd_bdev;
+	krwlock_t		vd_lock;
 } vdev_disk_t;
 
 #endif /* _KERNEL */

lib/libzfs/libzfs_import.c

@@ -145,6 +145,21 @@ zfs_device_get_devid(struct udev_device *dev, char *bufptr, size_t buflen)
 			return (0);
 		}
 
+		/*
+		 * For volumes use the persistent /dev/zvol/dataset identifier
+		 */
+		entry = udev_device_get_devlinks_list_entry(dev);
+		while (entry != NULL) {
+			const char *name;
+
+			name = udev_list_entry_get_name(entry);
+			if (strncmp(name, ZVOL_ROOT, strlen(ZVOL_ROOT)) == 0) {
+				(void) strlcpy(bufptr, name, buflen);
+				return (0);
+			}
+			entry = udev_list_entry_get_next(entry);
+		}
+
 		/*
 		 * NVME 'by-id' symlinks are similar to bus case
 		 */
@@ -187,28 +202,59 @@ int
 zfs_device_get_physical(struct udev_device *dev, char *bufptr, size_t buflen)
 {
 	const char *physpath = NULL;
+	struct udev_list_entry *entry;
 
 	/*
-	 * Normal disks use ID_PATH for their physical path.  Device mapper
+	 * Normal disks use ID_PATH for their physical path.
-	 * devices are virtual and don't have a physical path.  For them we
-	 * use ID_VDEV instead, which is setup via the /etc/vdev_id.conf file.
-	 * ID_VDEV provides a persistent path to a virtual device.  If you
-	 * don't have vdev_id.conf setup, you cannot use multipath autoreplace.
 	 */
-	if (!((physpath = udev_device_get_property_value(dev, "ID_PATH")) &&
+	physpath = udev_device_get_property_value(dev, "ID_PATH");
-	    physpath[0])) {
+	if (physpath != NULL && strlen(physpath) > 0) {
-		if (!((physpath =
-		    udev_device_get_property_value(dev, "ID_VDEV")) &&
-		    physpath[0])) {
-			return (ENODATA);
-		}
-	}
-
 		(void) strlcpy(bufptr, physpath, buflen);
-
 		return (0);
 	}
 
+	/*
+	 * Device mapper devices are virtual and don't have a physical
+	 * path. For them we use ID_VDEV instead, which is setup via the
+	 * /etc/vdev_id.conf file.  ID_VDEV provides a persistent path
+	 * to a virtual device.  If you don't have vdev_id.conf setup,
+	 * you cannot use multipath autoreplace with device mapper.
+	 */
+	physpath = udev_device_get_property_value(dev, "ID_VDEV");
+	if (physpath != NULL && strlen(physpath) > 0) {
+		(void) strlcpy(bufptr, physpath, buflen);
+		return (0);
+	}
+
+	/*
+	 * For ZFS volumes use the persistent /dev/zvol/dataset identifier
+	 */
+	entry = udev_device_get_devlinks_list_entry(dev);
+	while (entry != NULL) {
+		physpath = udev_list_entry_get_name(entry);
+		if (strncmp(physpath, ZVOL_ROOT, strlen(ZVOL_ROOT)) == 0) {
+			(void) strlcpy(bufptr, physpath, buflen);
+			return (0);
+		}
+		entry = udev_list_entry_get_next(entry);
+	}
+
+	/*
+	 * For all other devices fallback to using the by-uuid name.
+	 */
+	entry = udev_device_get_devlinks_list_entry(dev);
+	while (entry != NULL) {
+		physpath = udev_list_entry_get_name(entry);
+		if (strncmp(physpath, "/dev/disk/by-uuid", 17) == 0) {
+			(void) strlcpy(bufptr, physpath, buflen);
+			return (0);
+		}
+		entry = udev_list_entry_get_next(entry);
+	}
+
+	return (ENODATA);
+}
+
 boolean_t
 udev_is_mpath(struct udev_device *dev)
 {

lib/libzfs/libzfs_pool.c

@@ -2283,17 +2283,25 @@ vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
 }
 
 /*
- * Given a physical path (minus the "/devices" prefix), find the
+ * Given a physical path or guid, find the associated vdev.
- * associated vdev.
  */
 nvlist_t *
 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
 {
 	nvlist_t *search, *nvroot, *ret;
+	uint64_t guid;
+	char *end;
 
 	verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
-	verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
+
+	guid = strtoull(ppath, &end, 0);
+	if (guid != 0 && *end == '\0') {
+		verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
+	} else {
+		verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH,
+		    ppath) == 0);
+	}
 
 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
 	    &nvroot) == 0);

module/zfs/fm.c

@@ -665,25 +665,37 @@ out:
 	return (error);
 }
 
+/*
+ * Wait in an interruptible state for any new events.
+ */
 int
 zfs_zevent_wait(zfs_zevent_t *ze)
 {
-	int error = 0;
+	int error = EAGAIN;
 
 	mutex_enter(&zevent_lock);
+	zevent_waiters++;
 
+	while (error == EAGAIN) {
 		if (zevent_flags & ZEVENT_SHUTDOWN) {
-		error = ESHUTDOWN;
+			error = SET_ERROR(ESHUTDOWN);
-		goto out;
+			break;
 		}
 
-	zevent_waiters++;
+		error = cv_timedwait_sig(&zevent_cv, &zevent_lock,
-	cv_wait_sig(&zevent_cv, &zevent_lock);
+		    ddi_get_lbolt() + MSEC_TO_TICK(10));
-	if (issig(JUSTLOOKING))
+		if (signal_pending(current)) {
-		error = EINTR;
+			error = SET_ERROR(EINTR);
+			break;
+		} else if (!list_is_empty(&zevent_list)) {
+			error = 0;
+			continue;
+		} else {
+			error = EAGAIN;
+		}
+	}
 
 	zevent_waiters--;
-out:
 	mutex_exit(&zevent_lock);
 
 	return (error);

module/zfs/vdev.c

@@ -3241,7 +3241,8 @@ vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
 	/* XXX - L2ARC 1.0 does not support expansion */
 	if (!vd->vdev_aux) {
 		for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
-			pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
+			pvd->vdev_expanding = !!((flags & ZFS_ONLINE_EXPAND) ||
+			    spa->spa_autoexpand);
 	}
 
 	vdev_reopen(tvd);

module/zfs/vdev_disk.c

@@ -85,50 +85,64 @@ vdev_bdev_mode(int smode)
 }
 #endif /* HAVE_OPEN_BDEV_EXCLUSIVE */
 
-/* The capacity (in bytes) of a bdev that is available to be used by a vdev */
+/*
+ * Returns the usable capacity (in bytes) for the partition or disk.
+ */
 static uint64_t
-bdev_capacity(struct block_device *bdev, boolean_t wholedisk)
+bdev_capacity(struct block_device *bdev)
 {
-	struct hd_struct *part = bdev->bd_part;
+	return (i_size_read(bdev->bd_inode));
-	uint64_t sectors = get_capacity(bdev->bd_disk);
-	/* If there are no paritions, return the entire device capacity */
-	if (part == NULL)
-		return (sectors << SECTOR_BITS);
-
-	/*
-	 * If there are partitions, decide if we are using a `wholedisk`
-	 * layout (composed of part1 and part9) or just a single partition.
-	 */
-	if (wholedisk) {
-		/* Verify the expected device layout */
-		ASSERT3P(bdev, !=, bdev->bd_contains);
-		/*
-		 * Sectors used by the EFI partition (part9) as well as
-		 * partion alignment.
-		 */
-		uint64_t used = EFI_MIN_RESV_SIZE + NEW_START_BLOCK +
-		    PARTITION_END_ALIGNMENT;
-
-		/* Space available to the vdev, i.e. the size of part1 */
-		if (sectors <= used)
-			return (0);
-		uint64_t available = sectors - used;
-		return (available << SECTOR_BITS);
-	} else {
-		/* The partition capacity referenced by the block device */
-		return (part->nr_sects << SECTOR_BITS);
 }
+
+/*
+ * Returns the maximum expansion capacity of the block device (in bytes).
+ *
+ * It is possible to expand a vdev when it has been created as a wholedisk
+ * and the containing block device has increased in capacity.  Or when the
+ * partition containing the pool has been manually increased in size.
+ *
+ * This function is only responsible for calculating the potential expansion
+ * size so it can be reported by 'zpool list'.  The efi_use_whole_disk() is
+ * responsible for verifying the expected partition layout in the wholedisk
+ * case, and updating the partition table if appropriate.  Once the partition
+ * size has been increased the additional capacity will be visible using
+ * bdev_capacity().
+ */
+static uint64_t
+bdev_max_capacity(struct block_device *bdev, uint64_t wholedisk)
+{
+	uint64_t psize;
+	int64_t available;
+
+	if (wholedisk && bdev->bd_part != NULL && bdev != bdev->bd_contains) {
+		/*
+		 * When reporting maximum expansion capacity for a wholedisk
+		 * deduct any capacity which is expected to be lost due to
+		 * alignment restrictions.  Over reporting this value isn't
+		 * harmful and would only result in slightly less capacity
+		 * than expected post expansion.
+		 */
+		available = i_size_read(bdev->bd_contains->bd_inode) -
+		    ((EFI_MIN_RESV_SIZE + NEW_START_BLOCK +
+		    PARTITION_END_ALIGNMENT) << SECTOR_BITS);
+		if (available > 0)
+			psize = available;
+		else
+			psize = bdev_capacity(bdev);
+	} else {
+		psize = bdev_capacity(bdev);
+	}
+
+	return (psize);
 }
 
 static void
 vdev_disk_error(zio_t *zio)
 {
-#ifdef ZFS_DEBUG
+	zfs_dbgmsg(KERN_WARNING "zio error=%d type=%d offset=%llu size=%llu "
-	printk(KERN_WARNING "ZFS: zio error=%d type=%d offset=%llu size=%llu "
 	    "flags=%x\n", zio->io_error, zio->io_type,
 	    (u_longlong_t)zio->io_offset, (u_longlong_t)zio->io_size,
 	    zio->io_flags);
-#endif
 }
 
 /*
@@ -200,109 +214,73 @@ vdev_elevator_switch(vdev_t *v, char *elevator)
 	}
 }
 
-/*
- * Expanding a whole disk vdev involves invoking BLKRRPART on the
- * whole disk device. This poses a problem, because BLKRRPART will
- * return EBUSY if one of the disk's partitions is open. That's why
- * we have to do it here, just before opening the data partition.
- * Unfortunately, BLKRRPART works by dropping all partitions and
- * recreating them, which means that for a short time window, all
- * /dev/sdxN device files disappear (until udev recreates them).
- * This means two things:
- *  - When we open the data partition just after a BLKRRPART, we
- *    can't do it using the normal device file path because of the
- *    obvious race condition with udev. Instead, we use reliable
- *    kernel APIs to get a handle to the new partition device from
- *    the whole disk device.
- *  - Because vdev_disk_open() initially needs to find the device
- *    using its path, multiple vdev_disk_open() invocations in
- *    short succession on the same disk with BLKRRPARTs in the
- *    middle have a high probability of failure (because of the
- *    race condition with udev). A typical situation where this
- *    might happen is when the zpool userspace tool does a
- *    TRYIMPORT immediately followed by an IMPORT. For this
- *    reason, we only invoke BLKRRPART in the module when strictly
- *    necessary (zpool online -e case), and rely on userspace to
- *    do it when possible.
- */
-static struct block_device *
-vdev_disk_rrpart(const char *path, int mode, vdev_disk_t *vd)
-{
-#if defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK)
-	struct block_device *bdev, *result = ERR_PTR(-ENXIO);
-	struct gendisk *disk;
-	int error, partno;
-
-	bdev = vdev_bdev_open(path, vdev_bdev_mode(mode), zfs_vdev_holder);
-	if (IS_ERR(bdev))
-		return (bdev);
-
-	disk = get_gendisk(bdev->bd_dev, &partno);
-	vdev_bdev_close(bdev, vdev_bdev_mode(mode));
-
-	if (disk) {
-		bdev = bdget(disk_devt(disk));
-		if (bdev) {
-			error = blkdev_get(bdev, vdev_bdev_mode(mode), vd);
-			if (error == 0)
-				error = ioctl_by_bdev(bdev, BLKRRPART, 0);
-			vdev_bdev_close(bdev, vdev_bdev_mode(mode));
-		}
-
-		bdev = bdget_disk(disk, partno);
-		if (bdev) {
-			error = blkdev_get(bdev,
-			    vdev_bdev_mode(mode) | FMODE_EXCL, vd);
-			if (error == 0)
-				result = bdev;
-		}
-		put_disk(disk);
-	}
-
-	return (result);
-#else
-	return (ERR_PTR(-EOPNOTSUPP));
-#endif /* defined(HAVE_3ARG_BLKDEV_GET) && defined(HAVE_GET_GENDISK) */
-}
-
 static int
 vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
     uint64_t *ashift)
 {
-	struct block_device *bdev = ERR_PTR(-ENXIO);
+	struct block_device *bdev;
+	fmode_t mode = vdev_bdev_mode(spa_mode(v->vdev_spa));
+	int count = 0, block_size;
+	int bdev_retry_count = 50;
 	vdev_disk_t *vd;
-	int count = 0, mode, block_size;
 
 	/* Must have a pathname and it must be absolute. */
 	if (v->vdev_path == NULL || v->vdev_path[0] != '/') {
 		v->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
-		vdev_dbgmsg(v, "vdev_disk_open: invalid "
+		vdev_dbgmsg(v, "invalid vdev_path");
-		    "vdev_path '%s'", v->vdev_path);
 		return (SET_ERROR(EINVAL));
 	}
 
 	/*
-	 * Reopen the device if it's not currently open. Otherwise,
+	 * Reopen the device if it is currently open.  When expanding a
-	 * just update the physical size of the device.
+	 * partition force re-scanning the partition table while closed
+	 * in order to get an accurate updated block device size.  Then
+	 * since udev may need to recreate the device links increase the
+	 * open retry count before reporting the device as unavailable.
 	 */
-	if (v->vdev_tsd != NULL) {
-		ASSERT(v->vdev_reopening);
 	vd = v->vdev_tsd;
-		goto skip_open;
+	if (vd) {
+		char disk_name[BDEVNAME_SIZE + 6] = "/dev/";
+		boolean_t reread_part = B_FALSE;
+
+		rw_enter(&vd->vd_lock, RW_WRITER);
+		bdev = vd->vd_bdev;
+		vd->vd_bdev = NULL;
+
+		if (bdev) {
+			if (v->vdev_expanding && bdev != bdev->bd_contains) {
+				bdevname(bdev->bd_contains, disk_name + 5);
+				reread_part = B_TRUE;
 			}
 
+			vdev_bdev_close(bdev, mode);
+		}
+
+		if (reread_part) {
+			bdev = vdev_bdev_open(disk_name, mode, zfs_vdev_holder);
+			if (!IS_ERR(bdev)) {
+				int error = vdev_bdev_reread_part(bdev);
+				vdev_bdev_close(bdev, mode);
+				if (error == 0)
+					bdev_retry_count = 100;
+			}
+		}
+	} else {
 		vd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
-	if (vd == NULL)
+
-		return (SET_ERROR(ENOMEM));
+		rw_init(&vd->vd_lock, NULL, RW_DEFAULT, NULL);
+		rw_enter(&vd->vd_lock, RW_WRITER);
+	}
 
 	/*
 	 * Devices are always opened by the path provided at configuration
 	 * time.  This means that if the provided path is a udev by-id path
-	 * then drives may be recabled without an issue.  If the provided
+	 * then drives may be re-cabled without an issue.  If the provided
 	 * path is a udev by-path path, then the physical location information
 	 * will be preserved.  This can be critical for more complicated
 	 * configurations where drives are located in specific physical
-	 * locations to maximize the systems tolerence to component failure.
+	 * locations to maximize the systems tolerance to component failure.
+	 *
 	 * Alternatively, you can provide your own udev rule to flexibly map
 	 * the drives as you see fit.  It is not advised that you use the
 	 * /dev/[hd]d devices which may be reordered due to probing order.
@@ -317,15 +295,11 @@ vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
 	 * and it is reasonable to sleep and retry before giving up.  In
 	 * practice delays have been observed to be on the order of 100ms.
 	 */
-	mode = spa_mode(v->vdev_spa);
+	bdev = ERR_PTR(-ENXIO);
-	if (v->vdev_wholedisk && v->vdev_expanding)
+	while (IS_ERR(bdev) && count < bdev_retry_count) {
-		bdev = vdev_disk_rrpart(v->vdev_path, mode, vd);
+		bdev = vdev_bdev_open(v->vdev_path, mode, zfs_vdev_holder);
-
-	while (IS_ERR(bdev) && count < 50) {
-		bdev = vdev_bdev_open(v->vdev_path,
-		    vdev_bdev_mode(mode), zfs_vdev_holder);
 		if (unlikely(PTR_ERR(bdev) == -ENOENT)) {
-			msleep(10);
+			schedule_timeout(MSEC_TO_TICK(10));
 			count++;
 		} else if (IS_ERR(bdev)) {
 			break;
@@ -333,16 +307,18 @@ vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *max_psize,
 	}
 
 	if (IS_ERR(bdev)) {
-		dprintf("failed open v->vdev_path=%s, error=%d count=%d\n",
+		int error = -PTR_ERR(bdev);
-		    v->vdev_path, -PTR_ERR(bdev), count);
+		vdev_dbgmsg(v, "open error=%d count=%d\n", error, count);
-		kmem_free(vd, sizeof (vdev_disk_t));
+		vd->vd_bdev = NULL;
-		return (SET_ERROR(-PTR_ERR(bdev)));
+		v->vdev_tsd = vd;
+		rw_exit(&vd->vd_lock);
+		return (SET_ERROR(error));
+	} else {
+		vd->vd_bdev = bdev;
+		v->vdev_tsd = vd;
+		rw_exit(&vd->vd_lock);
 	}
 
-	v->vdev_tsd = vd;
-	vd->vd_bdev = bdev;
-
-skip_open:
 	/*  Determine the physical block size */
 	block_size = vdev_bdev_block_size(vd->vd_bdev);
 
@@ -352,9 +328,11 @@ skip_open:
 	/* Inform the ZIO pipeline that we are non-rotational */
 	v->vdev_nonrot = blk_queue_nonrot(bdev_get_queue(vd->vd_bdev));
 
-	/* Physical volume size in bytes */
+	/* Physical volume size in bytes for the partition */
-	*psize = bdev_capacity(vd->vd_bdev, v->vdev_wholedisk);
+	*psize = bdev_capacity(vd->vd_bdev);
-	*max_psize = *psize;
+
+	/* Physical volume size in bytes including possible expansion space */
+	*max_psize = bdev_max_capacity(vd->vd_bdev, v->vdev_wholedisk);
 
 	/* Based on the minimum sector size set the block size */
 	*ashift = highbit64(MAX(block_size, SPA_MINBLOCKSIZE)) - 1;
@@ -373,10 +351,12 @@ vdev_disk_close(vdev_t *v)
 	if (v->vdev_reopening || vd == NULL)
 		return;
 
-	if (vd->vd_bdev != NULL)
+	if (vd->vd_bdev != NULL) {
 		vdev_bdev_close(vd->vd_bdev,
 		    vdev_bdev_mode(spa_mode(v->vdev_spa)));
+	}
 
+	rw_destroy(&vd->vd_lock);
 	kmem_free(vd, sizeof (vdev_disk_t));
 	v->vdev_tsd = NULL;
 }
@@ -562,9 +542,15 @@ __vdev_disk_physio(struct block_device *bdev, zio_t *zio,
 #if defined(HAVE_BLK_QUEUE_HAVE_BLK_PLUG)
 	struct blk_plug plug;
 #endif
-
+	/*
-	ASSERT(zio != NULL);
+	 * Accessing outside the block device is never allowed.
-	ASSERT3U(io_offset + io_size, <=, bdev->bd_inode->i_size);
+	 */
+	if (io_offset + io_size > bdev->bd_inode->i_size) {
+		vdev_dbgmsg(zio->io_vd,
+		    "Illegal access %llu size %llu, device size %llu",
+		    io_offset, io_size, i_size_read(bdev->bd_inode));
+		return (SET_ERROR(EIO));
+	}
 
 retry:
 	dr = vdev_disk_dio_alloc(bio_count);
@@ -705,10 +691,34 @@ vdev_disk_io_start(zio_t *zio)
 	vdev_disk_t *vd = v->vdev_tsd;
 	int rw, flags, error;
 
+	/*
+	 * If the vdev is closed, it's likely in the REMOVED or FAULTED state.
+	 * Nothing to be done here but return failure.
+	 */
+	if (vd == NULL) {
+		zio->io_error = ENXIO;
+		zio_interrupt(zio);
+		return;
+	}
+
+	rw_enter(&vd->vd_lock, RW_READER);
+
+	/*
+	 * If the vdev is closed, it's likely due to a failed reopen and is
+	 * in the UNAVAIL state.  Nothing to be done here but return failure.
+	 */
+	if (vd->vd_bdev == NULL) {
+		rw_exit(&vd->vd_lock);
+		zio->io_error = ENXIO;
+		zio_interrupt(zio);
+		return;
+	}
+
 	switch (zio->io_type) {
 	case ZIO_TYPE_IOCTL:
 
 		if (!vdev_readable(v)) {
+			rw_exit(&vd->vd_lock);
 			zio->io_error = SET_ERROR(ENXIO);
 			zio_interrupt(zio);
 			return;
@@ -726,8 +736,10 @@ vdev_disk_io_start(zio_t *zio)
 			}
 
 			error = vdev_disk_io_flush(vd->vd_bdev, zio);
-			if (error == 0)
+			if (error == 0) {
+				rw_exit(&vd->vd_lock);
 				return;
+			}
 
 			zio->io_error = error;
 
@@ -737,6 +749,7 @@ vdev_disk_io_start(zio_t *zio)
 			zio->io_error = SET_ERROR(ENOTSUP);
 		}
 
+		rw_exit(&vd->vd_lock);
 		zio_execute(zio);
 		return;
 	case ZIO_TYPE_WRITE:
@@ -762,6 +775,7 @@ vdev_disk_io_start(zio_t *zio)
 		break;
 
 	default:
+		rw_exit(&vd->vd_lock);
 		zio->io_error = SET_ERROR(ENOTSUP);
 		zio_interrupt(zio);
 		return;
@@ -770,6 +784,8 @@ vdev_disk_io_start(zio_t *zio)
 	zio->io_target_timestamp = zio_handle_io_delay(zio);
 	error = __vdev_disk_physio(vd->vd_bdev, zio,
 	    zio->io_size, zio->io_offset, rw, flags);
+	rw_exit(&vd->vd_lock);
+
 	if (error) {
 		zio->io_error = error;
 		zio_interrupt(zio);

tests/runfiles/linux.run

@@ -333,7 +333,7 @@ tags = ['functional', 'cli_root', 'zpool_events']
 
 [tests/functional/cli_root/zpool_expand]
 tests = ['zpool_expand_001_pos', 'zpool_expand_002_pos',
-    'zpool_expand_003_neg', 'zpool_expand_004_pos']
+    'zpool_expand_003_neg', 'zpool_expand_004_pos', 'zpool_expand_005_pos']
 tags = ['functional', 'cli_root', 'zpool_expand']
 
 [tests/functional/cli_root/zpool_export]
@@ -398,7 +398,7 @@ tags = ['functional', 'cli_root', 'zpool_remove']
 [tests/functional/cli_root/zpool_reopen]
 tests = ['zpool_reopen_001_pos', 'zpool_reopen_002_pos',
     'zpool_reopen_003_pos', 'zpool_reopen_004_pos', 'zpool_reopen_005_pos',
-    'zpool_reopen_006_neg']
+    'zpool_reopen_006_neg', 'zpool_reopen_007_pos']
 tags = ['functional', 'cli_root', 'zpool_reopen']
 
 [tests/functional/cli_root/zpool_replace]

tests/test-runner/bin/zts-report.py

@@ -81,6 +81,13 @@ python_deps_reason = 'Python modules missing: python-cffi'
 #
 tmpfile_reason = 'Kernel O_TMPFILE support required'
 
+#
+# Some tests may depend on udev change events being generated when block
+# devices change capacity.  This functionality wasn't available until the
+# 2.6.38 kernel.
+#
+udev_reason = 'Kernel block device udev change events required'
+
 #
 # Some tests require that the NFS client and server utilities be installed.
 #
@@ -159,8 +166,6 @@ known = {
     'cli_root/zfs_unshare/zfs_unshare_002_pos': ['SKIP', na_reason],
     'cli_root/zfs_unshare/zfs_unshare_006_pos': ['SKIP', na_reason],
     'cli_root/zpool_create/zpool_create_016_pos': ['SKIP', na_reason],
-    'cli_root/zpool_expand/zpool_expand_001_pos': ['SKIP', '5771'],
-    'cli_root/zpool_expand/zpool_expand_003_neg': ['SKIP', '5771'],
     'cli_user/misc/zfs_share_001_neg': ['SKIP', na_reason],
     'cli_user/misc/zfs_unshare_001_neg': ['SKIP', na_reason],
     'inuse/inuse_001_pos': ['SKIP', na_reason],
@@ -219,6 +224,7 @@ maybe = {
     'cli_root/zpool_create/setup': ['SKIP', disk_reason],
     'cli_root/zpool_create/zpool_create_008_pos': ['FAIL', known_reason],
     'cli_root/zpool_destroy/zpool_destroy_001_pos': ['SKIP', '6145'],
+    'cli_root/zpool_expand/setup': ['SKIP', udev_reason],
     'cli_root/zpool_export/setup': ['SKIP', disk_reason],
     'cli_root/zpool_import/setup': ['SKIP', disk_reason],
     'cli_root/zpool_import/import_rewind_device_replaced':

tests/zfs-tests/include/blkdev.shlib

@@ -312,6 +312,7 @@ function on_off_disk # disk state{online,offline} host
 					log_fail "Onlining $disk failed"
 				fi
 			elif is_real_device $disk; then
+				block_device_wait
 				typeset -i retries=0
 				while ! lsscsi | egrep -q $disk; do
 					if (( $retries > 2 )); then
@@ -410,9 +411,7 @@ function load_scsi_debug # dev_size_mb add_host num_tgts max_luns blksz
 #
 function unload_scsi_debug
 {
-	if lsmod | grep scsi_debug >/dev/null; then
+	log_must_retry "in use" 5 modprobe -r scsi_debug
-		log_must modprobe -r scsi_debug
-	fi
 }
 
 #

tests/zfs-tests/tests/functional/cli_root/zpool_expand/Makefile.am

@@ -5,7 +5,8 @@ dist_pkgdata_SCRIPTS = \
 	zpool_expand_001_pos.ksh \
 	zpool_expand_002_pos.ksh \
 	zpool_expand_003_neg.ksh \
-	zpool_expand_004_pos.ksh
+	zpool_expand_004_pos.ksh \
+	zpool_expand_005_pos.ksh
 
 dist_pkgdata_DATA = \
 	zpool_expand.cfg

tests/zfs-tests/tests/functional/cli_root/zpool_expand/setup.ksh

@@ -29,6 +29,15 @@
 
 verify_runnable "global"
 
+#
+# The pool expansion tests depend on udev change events being generated
+# when block devices change capacity.  Since this functionality wasn't
+# available until the 2.6.38 kernel skip this test group.
+#
+if [[ $(linux_version) -lt $(linux_version "2.6.38") ]]; then
+	log_unsupported "Requires block device udev change events"
+fi
+
 zed_setup
 zed_start
 

tests/zfs-tests/tests/functional/cli_root/zpool_expand/zpool_expand.cfg

@@ -29,7 +29,9 @@
 #
 
 
-export org_size=$MINVDEVSIZE
+export org_size=$((1024*1024*1024))
-export exp_size=$((2*$org_size))
+export exp_size=$((2*1024*1024*1024))
+export org_size_mb=$((org_size/(1024*1024)))
 
-export VFS=$TESTPOOL/$TESTFS
+export FILE_LO=$TEST_BASE_DIR/vdev_lo
+export FILE_RAW=$TEST_BASE_DIR/vdev_raw

tests/zfs-tests/tests/functional/cli_root/zpool_expand/zpool_expand_001_pos.ksh

@@ -27,6 +27,7 @@
 
 #
 # Copyright (c) 2012, 2016 by Delphix. All rights reserved.
+# Copyright (c) 2018 by Lawrence Livermore National Security, LLC.
 #
 
 . $STF_SUITE/include/libtest.shlib
@@ -35,68 +36,85 @@
 #
 # DESCRIPTION:
 # Once zpool set autoexpand=on poolname, zpool can autoexpand by
-# Dynamic LUN Expansion
+# Dynamic VDEV Expansion
 #
 #
 # STRATEGY:
-# 1) Create a pool
+# 1) Create three vdevs (loopback, scsi_debug, and file)
-# 2) Create volume on top of the pool
+# 2) Create pool by using the different devices and set autoexpand=on
-# 3) Create pool by using the zvols and set autoexpand=on
+# 3) Expand each device as appropriate
-# 4) Expand the vol size by 'zfs set volsize'
+# 4) Check that the pool size was expanded
-# 5) Check that the pool size was expanded
+#
+# NOTE: Three different device types are used in this test to verify
+# expansion of non-partitioned block devices (loopback), partitioned
+# block devices (scsi_debug), and non-disk file vdevs.  ZFS volumes
+# are not used in order to avoid a possible lock inversion when
+# layering pools on zvols.
 #
 
 verify_runnable "global"
 
-# See issue: https://github.com/zfsonlinux/zfs/issues/5771
-if is_linux; then
-	log_unsupported "Requires autoexpand property support"
-fi
-
 function cleanup
 {
-	if poolexists $TESTPOOL1; then
+	poolexists $TESTPOOL1 && destroy_pool $TESTPOOL1
-		log_must zpool destroy $TESTPOOL1
+
+	if losetup -a | grep -q $DEV1; then
+		losetup -d $DEV1
 	fi
 
-	for i in 1 2 3; do
+	rm -f $FILE_LO $FILE_RAW
-		if datasetexists $VFS/vol$i; then
+
-			log_must zfs destroy $VFS/vol$i
+	block_device_wait
-		fi
+	unload_scsi_debug
-	done
 }
 
 log_onexit cleanup
 
-log_assert "zpool can be autoexpanded after set autoexpand=on on LUN expansion"
+log_assert "zpool can be autoexpanded after set autoexpand=on on vdev expansion"
-
-for i in 1 2 3; do
-	log_must zfs create -V $org_size $VFS/vol$i
-done
-block_device_wait
 
 for type in " " mirror raidz raidz2; do
+	log_note "Setting up loopback, scsi_debug, and file vdevs"
+	log_must truncate -s $org_size $FILE_LO
+	DEV1=$(losetup -f)
+	log_must losetup $DEV1 $FILE_LO
 
-	log_must zpool create -o autoexpand=on $TESTPOOL1 $type \
+	load_scsi_debug $org_size_mb 1 1 1 '512b'
-	    ${ZVOL_DEVDIR}/$VFS/vol1  ${ZVOL_DEVDIR}/$VFS/vol2 \
+	block_device_wait
-	    ${ZVOL_DEVDIR}/$VFS/vol3
+	DEV2=$(get_debug_device)
+
+	log_must truncate -s $org_size $FILE_RAW
+	DEV3=$FILE_RAW
+
+	# The -f is required since we're mixing disk and file vdevs.
+	log_must zpool create -f -o autoexpand=on $TESTPOOL1 $type \
+	    $DEV1 $DEV2 $DEV3
 
 	typeset autoexp=$(get_pool_prop autoexpand $TESTPOOL1)
 	if [[ $autoexp != "on" ]]; then
-		log_fail "zpool $TESTPOOL1 autoexpand should on but is $autoexp"
+		log_fail "zpool $TESTPOOL1 autoexpand should be on but is " \
+		    "$autoexp"
 	fi
 
 	typeset prev_size=$(get_pool_prop size $TESTPOOL1)
 	typeset zfs_prev_size=$(zfs get -p avail $TESTPOOL1 | tail -1 | \
 	    awk '{print $3}')
 
-	for i in 1 2 3; do
+	# Expand each device as appropriate being careful to add an artificial
-		log_must zfs set volsize=$exp_size $VFS/vol$i
+	# delay to ensure we get a single history entry for each.  This makes
-	done
+	# is easier to verify each expansion for the striped pool case, since
+	# they will not be merged in to a single larger expansion.
+	log_note "Expanding loopback, scsi_debug, and file vdevs"
+	log_must truncate -s $exp_size $FILE_LO
+	log_must losetup -c $DEV1
+	sleep 3
 
-	sync
+	echo "2" > /sys/bus/pseudo/drivers/scsi_debug/virtual_gb
-	sleep 10
+	echo "1" > /sys/class/block/$DEV2/device/rescan
-	sync
+	block_device_wait
+	sleep 3
+
+	log_must truncate -s $exp_size $FILE_RAW
+	log_must zpool online -e $TESTPOOL1 $FILE_RAW
 
 	typeset expand_size=$(get_pool_prop size $TESTPOOL1)
 	typeset zfs_expand_size=$(zfs get -p avail $TESTPOOL1 | tail -1 | \
@@ -105,7 +123,7 @@ for type in " " mirror raidz raidz2; do
 	log_note "$TESTPOOL1 $type has previous size: $prev_size and " \
 	    "expanded size: $expand_size"
 	# compare available pool size from zfs
-	if [[ $zfs_expand_size > $zfs_prev_size ]]; then
+	if [[ $zfs_expand_size -gt $zfs_prev_size ]]; then
 		# check for zpool history for the pool size expansion
 		if [[ $type == " " ]]; then
 			typeset expansion_size=$(($exp_size-$org_size))
@@ -114,9 +132,9 @@ for type in " " mirror raidz raidz2; do
 			    grep "vdev online" | \
 			    grep "(+${expansion_size}" | wc -l)
 
-			if [[ $size_addition -ne $i ]]; then
+			if [[ $size_addition -ne 3 ]]; then
-				log_fail "pool $TESTPOOL1 is not autoexpand " \
+				log_fail "pool $TESTPOOL1 has not expanded, " \
-				    "after LUN expansion"
+				    "$size_addition/3 vdevs expanded"
 			fi
 		elif [[ $type == "mirror" ]]; then
 			typeset expansion_size=$(($exp_size-$org_size))
@@ -126,8 +144,7 @@ for type in " " mirror raidz raidz2; do
 			    grep "(+${expansion_size})" >/dev/null 2>&1
 
 			if [[ $? -ne 0 ]] ; then
-				log_fail "pool $TESTPOOL1 is not autoexpand " \
+				log_fail "pool $TESTPOOL1 has not expanded"
-				    "after LUN expansion"
 			fi
 		else
 			typeset expansion_size=$((3*($exp_size-$org_size)))
@@ -137,19 +154,16 @@ for type in " " mirror raidz raidz2; do
 			    grep "(+${expansion_size})" >/dev/null 2>&1
 
 			if [[ $? -ne 0 ]]; then
-				log_fail "pool $TESTPOOL is not autoexpand " \
+				log_fail "pool $TESTPOOL has not expanded"
-				    "after LUN expansion"
 			fi
 		fi
 	else
-		log_fail "pool $TESTPOOL1 is not autoexpanded after LUN " \
+		log_fail "pool $TESTPOOL1 is not autoexpanded after vdev " \
-		    "expansion"
+		    "expansion.  Previous size: $zfs_prev_size and expanded " \
+		    "size: $zfs_expand_size"
 	fi
 
-	log_must zpool destroy $TESTPOOL1
+	cleanup
-	for i in 1 2 3; do
-		log_must zfs set volsize=$org_size $VFS/vol$i
 done
 
-done
+log_pass "zpool can autoexpand if autoexpand=on after vdev expansion"
-log_pass "zpool can be autoexpanded after set autoexpand=on on LUN expansion"

tests/zfs-tests/tests/functional/cli_root/zpool_expand/zpool_expand_002_pos.ksh

@@ -36,7 +36,7 @@
 #
 # DESCRIPTION:
 # After zpool online -e poolname zvol vdevs, zpool can autoexpand by
-# Dynamic LUN Expansion
+# Dynamic VDEV Expansion
 #
 #
 # STRATEGY:
@@ -52,9 +52,7 @@ verify_runnable "global"
 
 function cleanup
 {
-        if poolexists $TESTPOOL1; then
+	poolexists $TESTPOOL1 && destroy_pool $TESTPOOL1
-                log_must zpool destroy $TESTPOOL1
-        fi
 
 	for i in 1 2 3; do
 		[ -e ${TEMPFILE}.$i ] && log_must rm ${TEMPFILE}.$i
@@ -63,7 +61,7 @@ function cleanup
 
 log_onexit cleanup
 
-log_assert "zpool can expand after zpool online -e zvol vdevs on LUN expansion"
+log_assert "zpool can expand after zpool online -e zvol vdevs on vdev expansion"
 
 for type in " " mirror raidz raidz2; do
 	# Initialize the file devices and the pool
@@ -77,7 +75,7 @@ for type in " " mirror raidz raidz2; do
 	typeset autoexp=$(get_pool_prop autoexpand $TESTPOOL1)
 
 	if [[ $autoexp != "off" ]]; then
-		log_fail "zpool $TESTPOOL1 autoexpand should off but is " \
+		log_fail "zpool $TESTPOOL1 autoexpand should be off but is " \
 		    "$autoexp"
 	fi
 	typeset prev_size=$(get_pool_prop size $TESTPOOL1)
@@ -109,15 +107,15 @@ for type in " " mirror raidz raidz2; do
 		    "expected $expected_zpool_expandsize"
 	fi
 
-	# Online the devices to add the new space to the pool
+	# Online the devices to add the new space to the pool.  Add an
+	# artificial delay between online commands order to prevent them
+	# from being merged in to a single history entry.  This makes
+	# is easier to verify each expansion for the striped pool case.
 	for i in 1 2 3; do
 		log_must zpool online -e $TESTPOOL1 ${TEMPFILE}.$i
+		sleep 3
 	done
 
-	sync
-	sleep 10
-	sync
-
 	typeset expand_size=$(get_pool_prop size $TESTPOOL1)
 	typeset zfs_expand_size=$(get_prop avail $TESTPOOL1)
 	log_note "$TESTPOOL1 $type has previous size: $prev_size and " \
@@ -134,8 +132,9 @@ for type in " " mirror raidz raidz2; do
 			    grep "(+${expansion_size}" | wc -l)
 
 			if [[ $size_addition -ne $i ]]; then
-				log_fail "pool $TESTPOOL1 did not expand " \
+				log_fail "pool $TESTPOOL1 has not expanded " \
-				    "after LUN expansion and zpool online -e"
+				    "after zpool online -e, " \
+				    "$size_addition/3 vdevs expanded"
 			fi
 		elif [[ $type == "mirror" ]]; then
 			typeset expansion_size=$(($exp_size-$org_size))
@@ -145,8 +144,8 @@ for type in " " mirror raidz raidz2; do
 			    grep "(+${expansion_size})" >/dev/null 2>&1
 
 			if [[ $? -ne 0 ]]; then
-				log_fail "pool $TESTPOOL1 did not expand " \
+				log_fail "pool $TESTPOOL1 has not expanded " \
-				    "after LUN expansion and zpool online -e"
+				    "after zpool online -e"
 			fi
 		else
 			typeset expansion_size=$((3*($exp_size-$org_size)))
@@ -156,14 +155,14 @@ for type in " " mirror raidz raidz2; do
 			    grep "(+${expansion_size})" >/dev/null 2>&1
 
 			if [[ $? -ne 0 ]] ; then
-				log_fail "pool $TESTPOOL1 did not expand " \
+				log_fail "pool $TESTPOOL1 has not expanded " \
-				    "after LUN expansion and zpool online -e"
+				    "after zpool online -e"
 			fi
 		fi
 	else
-		log_fail "pool $TESTPOOL1 did not expand after LUN expansion " \
+		log_fail "pool $TESTPOOL1 did not expand after vdev expansion " \
 		    "and zpool online -e"
 	fi
 	log_must zpool destroy $TESTPOOL1
 done
-log_pass "zpool can expand after zpool online -e zvol vdevs on LUN expansion"
+log_pass "zpool can expand after zpool online -e"

tests/zfs-tests/tests/functional/cli_root/zpool_expand/zpool_expand_003_neg.ksh

@@ -27,95 +27,112 @@
 
 #
 # Copyright (c) 2012, 2016 by Delphix. All rights reserved.
+# Copyright (c) 2018 by Lawrence Livermore National Security, LLC.
 #
 
+
 . $STF_SUITE/include/libtest.shlib
 . $STF_SUITE/tests/functional/cli_root/zpool_expand/zpool_expand.cfg
 
 #
 # Description:
 # Once set zpool autoexpand=off, zpool can *NOT* autoexpand by
-# Dynamic LUN Expansion
+# Dynamic VDEV Expansion
 #
 #
 # STRATEGY:
-# 1) Create a pool
+# 1) Create three vdevs (loopback, scsi_debug, and file)
-# 2) Create volumes on top of the pool
+# 2) Create pool by using the different devices and set autoexpand=off
-# 3) Create pool by using the zvols and set autoexpand=off
+# 3) Expand each device as appropriate
-# 4) Expand the vol size by zfs set volsize
+# 4) Check that the pool size is not expanded
-# 5) Check that the pool size is not changed
+#
+# NOTE: Three different device types are used in this test to verify
+# expansion of non-partitioned block devices (loopback), partitioned
+# block devices (scsi_debug), and non-disk file vdevs.  ZFS volumes
+# are not used in order to avoid a possible lock inversion when
+# layering pools on zvols.
 #
 
 verify_runnable "global"
 
-# See issue: https://github.com/zfsonlinux/zfs/issues/5771
-if is_linux; then
-	log_unsupported "Requires autoexpand property support"
-fi
-
 function cleanup
 {
-        if poolexists $TESTPOOL1; then
+	poolexists $TESTPOOL1 && destroy_pool $TESTPOOL1
-                log_must zpool destroy $TESTPOOL1
+
+	if losetup -a | grep -q $DEV1; then
+		losetup -d $DEV1
 	fi
 
-	for i in 1 2 3; do
+	rm -f $FILE_LO $FILE_RAW
-		if datasetexists $VFS/vol$i; then
+
-			log_must zfs destroy $VFS/vol$i
+	block_device_wait
-		fi
+	unload_scsi_debug
-	done
 }
 
 log_onexit cleanup
 
-log_assert "zpool can not expand if set autoexpand=off after LUN expansion"
+log_assert "zpool can not expand if set autoexpand=off after vdev expansion"
-
-for i  in 1 2 3; do
-	log_must zfs create -V $org_size $VFS/vol$i
-done
-block_device_wait
 
 for type in " " mirror raidz raidz2; do
-	log_must zpool create $TESTPOOL1 $type ${ZVOL_DEVDIR}/$VFS/vol1 \
+	log_note "Setting up loopback, scsi_debug, and file vdevs"
-	    ${ZVOL_DEVDIR}/$VFS/vol2 ${ZVOL_DEVDIR}/$VFS/vol3
+	log_must truncate -s $org_size $FILE_LO
+	DEV1=$(losetup -f)
+	log_must losetup $DEV1 $FILE_LO
+
+	load_scsi_debug $org_size_mb 1 1 1 '512b'
+	block_device_wait
+	DEV2=$(get_debug_device)
+
+	log_must truncate -s $org_size $FILE_RAW
+	DEV3=$FILE_RAW
+
+	# The -f is required since we're mixing disk and file vdevs.
+	log_must zpool create -f $TESTPOOL1 $type $DEV1 $DEV2 $DEV3
 
 	typeset autoexp=$(get_pool_prop autoexpand $TESTPOOL1)
 	if [[ $autoexp != "off" ]]; then
-		log_fail "zpool $TESTPOOL1 autoexpand should off but is " \
+		log_fail "zpool $TESTPOOL1 autoexpand should be off but is " \
 		    "$autoexp"
 	fi
 
 	typeset prev_size=$(get_pool_prop size $TESTPOOL1)
 
-	for i in 1 2 3; do
-		log_must zfs set volsize=$exp_size $VFS/vol$i
-	done
 
-	sync
+	# Expand each device as appropriate being careful to add an artificial
-	sleep 10
+	# delay to ensure we get a single history entry for each.  This makes
-	sync
+	# is easier to verify each expansion for the striped pool case, since
+	# they will not be merged in to a single larger expansion.
+	log_note "Expanding loopback, scsi_debug, and file vdevs"
+	log_must truncate -s $exp_size $FILE_LO
+	log_must losetup -c $DEV1
+	sleep 3
+
+	echo "2" > /sys/bus/pseudo/drivers/scsi_debug/virtual_gb
+	echo "1" > /sys/class/block/$DEV2/device/rescan
+	block_device_wait
+	sleep 3
+
+	log_must truncate -s $exp_size $FILE_RAW
+
+	# This is far longer than we should need to wait, but let's be sure.
+	sleep 5
 
 	# check for zpool history for the pool size expansion
 	zpool history -il $TESTPOOL1 | grep "pool '$TESTPOOL1' size:" | \
 	    grep "vdev online" >/dev/null 2>&1
 
 	if [[ $? -eq 0 ]]; then
-		log_fail "pool $TESTPOOL1 is not autoexpand after LUN " \
+		log_fail "pool $TESTPOOL1 is not autoexpand after vdev " \
 		    "expansion"
 	fi
 
 	typeset expand_size=$(get_pool_prop size $TESTPOOL1)
 
 	if [[ "$prev_size" != "$expand_size" ]]; then
-		log_fail "pool $TESTPOOL1 size changed after LUN expansion"
+		log_fail "pool $TESTPOOL1 size changed after vdev expansion"
 	fi
 
-	log_must zpool destroy $TESTPOOL1
+	cleanup
-
-	for i in 1 2 3; do
-		log_must zfs set volsize=$org_size $VFS/vol$i
 done
 
-done
+log_pass "zpool can not autoexpand if autoexpand=off after vdev expansion"
-
-log_pass "zpool can not expand if set autoexpand=off after LUN expansion"

tests/zfs-tests/tests/functional/cli_root/zpool_expand/zpool_expand_004_pos.ksh

@@ -50,9 +50,7 @@ verify_runnable "global"
 
 function cleanup
 {
-        if poolexists $TESTPOOL1; then
+	poolexists $TESTPOOL1 && destroy_pool $TESTPOOL1
-                log_must zpool destroy $TESTPOOL1
-        fi
 
 	for i in 1 2 3; do
 		[ -e ${TEMPFILE}.$i ] && log_must rm ${TEMPFILE}.$i

tests/zfs-tests/tests/functional/cli_root/zpool_expand/zpool_expand_005_pos.ksh

@@ -0,0 +1,99 @@
+#! /bin/ksh -p
+#
+# CDDL HEADER START
+#
+# The contents of this file are subject to the terms of the
+# Common Development and Distribution License (the "License").
+# You may not use this file except in compliance with the License.
+#
+# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+# or http://www.opensolaris.org/os/licensing.
+# See the License for the specific language governing permissions
+# and limitations under the License.
+#
+# When distributing Covered Code, include this CDDL HEADER in each
+# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+# If applicable, add the following below this CDDL HEADER, with the
+# fields enclosed by brackets "[]" replaced with your own identifying
+# information: Portions Copyright [yyyy] [name of copyright owner]
+#
+# CDDL HEADER END
+#
+
+#
+# Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+# Use is subject to license terms.
+#
+
+#
+# Copyright (c) 2012, 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/include/blkdev.shlib
+. $STF_SUITE/tests/functional/cli_root/zpool_expand/zpool_expand.cfg
+
+#
+# DESCRIPTION:
+#
+# STRATEGY:
+# 1) Create a scsi_debug device and a pool based on it
+# 2) Expand the device and rescan the scsi bus
+# 3) Reopen the pool and check that it detects new available space
+# 4) Online the device and check that the pool has been expanded
+#
+
+verify_runnable "global"
+
+function cleanup
+{
+	poolexists $TESTPOOL1 && destroy_pool $TESTPOOL1
+	unload_scsi_debug
+}
+
+log_onexit cleanup
+
+log_assert "zpool based on scsi device can be expanded with zpool online -e"
+
+# run scsi_debug to create a device
+MINVDEVSIZE_MB=$((MINVDEVSIZE / 1048576))
+load_scsi_debug $MINVDEVSIZE_MB 1 1 1 '512b'
+block_device_wait
+SDISK=$(get_debug_device)
+log_must zpool create $TESTPOOL1 $SDISK
+
+typeset autoexp=$(get_pool_prop autoexpand $TESTPOOL1)
+if [[ $autoexp != "off" ]]; then
+	log_fail "zpool $TESTPOOL1 autoexpand should be off but is $autoexp"
+fi
+
+typeset prev_size=$(get_pool_prop size $TESTPOOL1)
+log_note "original pool size: $prev_size"
+
+# resize the scsi_debug device
+echo "5" > /sys/bus/pseudo/drivers/scsi_debug/virtual_gb
+# rescan the device to detect the new size
+echo "1" > /sys/class/block/$SDISK/device/rescan
+block_device_wait
+
+# reopen the pool so ZFS can see the new space
+log_must zpool reopen $TESTPOOL1
+
+typeset expandsize=$(get_pool_prop expandsize $TESTPOOL1)
+log_note "pool expandsize: $expandsize"
+if [[ "$zpool_expandsize" = "-" ]]; then
+	log_fail "pool $TESTPOOL1 did not detect any " \
+	    "expandsize after reopen"
+fi
+
+# online the device so the zpool will use the new space
+log_must zpool online -e $TESTPOOL1 $SDISK
+
+typeset new_size=$(get_pool_prop size $TESTPOOL1)
+log_note "new pool size: $new_size"
+if [[ $new_size -le $prev_size ]]; then
+	log_fail "pool $TESTPOOL1 did not expand " \
+	    "after vdev expansion and zpool online -e"
+fi
+
+log_pass "zpool based on scsi_debug can be expanded with reopen and online -e"

tests/zfs-tests/tests/functional/cli_root/zpool_reopen/Makefile.am

@@ -7,7 +7,8 @@ dist_pkgdata_SCRIPTS = \
 	zpool_reopen_003_pos.ksh \
 	zpool_reopen_004_pos.ksh \
 	zpool_reopen_005_pos.ksh \
-	zpool_reopen_006_neg.ksh
+	zpool_reopen_006_neg.ksh \
+	zpool_reopen_007_pos.ksh
 
 dist_pkgdata_DATA = \
 	zpool_reopen.cfg \

tests/zfs-tests/tests/functional/cli_root/zpool_reopen/cleanup.ksh

@@ -25,7 +25,7 @@ cleanup_devices $DISKS
 # Unplug the disk and remove scsi_debug module
 if is_linux; then
 	for SDDEVICE in $(get_debug_device); do
-		unplug $SDDEVICE
+		remove_disk $SDDEVICE
 	done
 	unload_scsi_debug
 fi

tests/zfs-tests/tests/functional/cli_root/zpool_reopen/zpool_reopen_007_pos.ksh

@@ -0,0 +1,67 @@
+#!/bin/ksh -p
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2018 by Lawrence Livermore National Security, LLC.
+#
+
+. $STF_SUITE/tests/functional/cli_root/zpool_reopen/zpool_reopen.shlib
+
+#
+# DESCRIPTION:
+# Test zpool reopen while performing IO to the pool.
+# Verify that no IO errors of any kind of reported.
+#
+# STRATEGY:
+# 1. Create a non-redundant pool.
+# 2. Repeat:
+#   a. Write files to the pool.
+#   b. Execute 'zpool reopen'.
+# 3. Verify that no errors are reported by 'zpool status'.
+
+verify_runnable "global"
+
+function cleanup
+{
+	poolexists $TESTPOOL && destroy_pool $TESTPOOL
+}
+
+log_assert "Testing zpool reopen with concurrent user IO"
+log_onexit cleanup
+
+set_removed_disk
+scsi_host=$(get_scsi_host $REMOVED_DISK)
+
+# 1. Create a non-redundant pool.
+log_must zpool create $TESTPOOL $DISK1 $DISK2 $DISK3
+
+for i in $(seq 10); do
+	# 3a. Write files in the background to the pool.
+	mkfile 64m /$TESTPOOL/data.$i &
+
+	# 3b. Execute 'zpool reopen'.
+	log_must zpool reopen $TESTPOOL
+
+	for disk in $DISK1 $DISK2 $DISK3; do
+		zpool status -P -v $TESTPOOL | grep $disk | \
+		    read -r name state rd wr cksum
+		log_must [ $state = "ONLINE" ]
+	        log_must [ $rd -eq 0 ]
+	        log_must [ $wr -eq 0 ]
+	        log_must [ $cksum -eq 0 ]
+	done
+done
+
+wait
+
+log_pass "Zpool reopen with concurrent user IO successful"