libzfs/common/libzfs_vdev.c

/*
 * 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 (c) 2005, 2012, Oracle and/or its affiliates. All rights reserved.
 */

/*
 * Functions to convert between a list of vdevs and an nvlist representing the
 * configuration.  Each entry in the list can be one of:
 *
 *  Device vdevs
 *      disk=(path=..., devid=...)
 *      file=(path=...)
 *
 *  Group vdevs
 *      raidz[1|2]=(...)
 *      mirror=(...)
 *
 *  Hot spares
 *
 * While the underlying implementation supports it, group vdevs cannot contain
 * other group vdevs.  All userland verification of devices is contained within
 * this file.  If successful, the nvlist returned can be passed directly to the
 * kernel; we've done as much verification as possible in userland.
 *
 * Hot spares are a special case, and passed down as an array of disk vdevs, at
 * the same level as the root of the vdev tree.
 *
 * Functions exported by this file:
 * 'zpool_make_root_vdev' - the function performs several passes:
 *
 *  1. Construct the vdev specification.  Performs syntax validation and
 *         makes sure each device is valid.
 *  2. Check for devices in use.  Using libdiskmgt, makes sure that no
 *         devices are also in use.  Some can be overridden using the 'force'
 *         flag, others cannot.
 *  3. Check for replication errors if the 'force' flag is not specified.
 *         validates that the replication level is consistent across the
 *         entire pool.
 *  4. Label any whole disks with an EFI label.
 *
 * 'zpool_split_root_vdev' - the function performs several passes:
 *
 *  1. Construct the vdev specification.  Performs syntax validation and
 *         makes sure each device is valid.
 *  2. Check target device if its eligable for splitting.
 *  3. Initiate the actual split.
 */

#include <assert.h>
#include <ctype.h>
#include <devid.h>
#include <errno.h>
#include <fcntl.h>
#include <libdiskmgt.h>
#include <libdevinfo.h>
#include <libintl.h>
#include <libnvpair.h>
#include <limits.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <sys/efi_partition.h>
#include <sys/stat.h>
#include <sys/vtoc.h>
#include <sys/mntent.h>
#include <sys/paths.h>

#include <libzfs_impl.h>

#define DISK_ROOT   _PATH_DEV "dsk"
#define RDISK_ROOT  _PATH_DEV "rdsk"
#define BACKUP_SLICE    "s2"

/*
 * For any given vdev specification, we can have multiple errors.  The
 * vdev_error() function keeps track of whether we have seen an error yet, and
 * prints out a header if its the first error we've seen.
 */
/*PRINTFLIKE4*/
static void
vdev_error(libzfs_handle_t *hdl, boolean_t force, int error,
    const char *fmt, ...)
{
    static boolean_t error_seen = B_FALSE;
    char msg[1024];
    va_list ap;

    if (!error_seen) {
        error_seen = B_TRUE;
        if (!force) {
            (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
                "use -f to override the following errors:"));
        } else {
            (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
                "the following errors must be manually repaired:"));
        }
        zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s"), msg);
        (void) zfs_error(hdl, error, dgettext(TEXT_DOMAIN,
            "vdev verification failed"));
    }
    va_start(ap, fmt);
    (void) vfprintf(stderr, fmt, ap);
    va_end(ap);
}

static void
internal_error(libzfs_handle_t *hdl, int error)
{
    /*
     * ENXIO/ENODEV is a valid error message if the device doesn't live in
     * /dev/dsk.  Don't bother printing an error message in this case.
     */
    if (error == ENXIO || error == ENODEV)
        return;
    (void) zfs_error_fmt(hdl, EZFS_INVALCONFIG, dgettext(TEXT_DOMAIN,
        "device in use checking failed: %s\n"), strerror(error));
}

/*
 * Validate a device, passing the bulk of the work off to libdiskmgt.
 */
static int
check_slice(libzfs_handle_t *hdl, const char *path, boolean_t force,
    boolean_t wholedisk, boolean_t isspare)
{
    char *msg;
    int error = 0;
    dm_who_type_t who;

    if (force)
        who = isspare ? DM_WHO_ZPOOL_SPARE_FORCE : DM_WHO_ZPOOL_FORCE;
    else
        who = isspare ? DM_WHO_ZPOOL_SPARE : DM_WHO_ZPOOL;

    if (dm_inuse((char *)path, &msg, who, &error) || error) {
        if (error != 0) {
            internal_error(hdl, error);
            return (0);
        }
        vdev_error(hdl, force, EZFS_DEV_INUSE, "%s", msg);
        free(msg);
        return (-1);
    }

    /*
     * If we're given a whole disk, ignore overlapping slices since we're
     * about to label it anyway.
     */
    error = 0;
    if (!wholedisk && !force &&
        (dm_isoverlapping((char *)path, &msg, &error) || error)) {
        if (error == 0) {
            /* dm_isoverlapping returned -1 */
            vdev_error(hdl, force, EZFS_DEV_INUSE,
                gettext("%s overlaps with %s\n"), path, msg);
            free(msg);
            return (-1);
        } else if (error != ENODEV) {
            /* libdiskmgt's devcache only handles physical drives */
            internal_error(hdl, error);
            return (0);
        }
    }

    return (0);
}


/*
 * Validate a whole disk.  Iterate over all slices on the disk and make sure
 * that none is in use by calling check_slice().
 */
static int
check_disk(libzfs_handle_t *hdl, const char *name, dm_descriptor_t disk,
    boolean_t force, int isspare)
{
    dm_descriptor_t *drive, *media, *slice;
    int err = 0;
    int i, ret;

    /*
     * Get the drive associated with this disk.  This should never fail,
     * because we already have an alias handle open for the device.
     */
    if ((drive = dm_get_associated_descriptors(disk, DM_DRIVE,
        &err)) == NULL || *drive == NULL) {
        if (err)
            internal_error(hdl, err);
        return (0);
    }

    if ((media = dm_get_associated_descriptors(*drive, DM_MEDIA,
        &err)) == NULL) {
        dm_free_descriptors(drive);
        if (err)
            internal_error(hdl, err);
        return (0);
    }

    dm_free_descriptors(drive);

    /*
     * It is possible that the user has specified a removable media drive,
     * and the media is not present.
     */
    if (*media == NULL) {
        dm_free_descriptors(media);
        vdev_error(hdl, force, EZFS_BADDEV,
            gettext("'%s' has no media in drive\n"), name);
        return (-1);
    }

    if ((slice = dm_get_associated_descriptors(*media, DM_SLICE,
        &err)) == NULL) {
        dm_free_descriptors(media);
        if (err)
            internal_error(hdl, err);
        return (0);
    }

    dm_free_descriptors(media);

    ret = 0;

    /*
     * Iterate over all slices and report any errors.  We don't care about
     * overlapping slices because we are using the whole disk.
     */
    for (i = 0; slice[i] != NULL; i++) {
        char *name = dm_get_name(slice[i], &err);

        if (check_slice(hdl, name, force, B_TRUE, isspare) != 0)
            ret = -1;

        dm_free_name(name);
    }

    dm_free_descriptors(slice);
    return (ret);
}

/*
 * Validate a device.
 */
static int
check_device(libzfs_handle_t *hdl, const char *path, boolean_t force,
    boolean_t isspare)
{
    dm_descriptor_t desc;
    int err;
    char *dev;

    /*
     * For whole disks, libdiskmgt does not include the leading dev path.
     */
    dev = strrchr(path, '/');
    assert(dev != NULL);
    dev++;
    if ((desc = dm_get_descriptor_by_name(DM_ALIAS, dev, &err)) != NULL) {
        err = check_disk(hdl, path, desc, force, isspare);
        dm_free_descriptor(desc);
        return (err);
    }

    return (check_slice(hdl, path, force, B_FALSE, isspare));
}

/*
 * Check that a file is valid.  All we can do in this case is check that it's
 * not in use by another pool, and not in use by swap.
 */
static int
check_file(libzfs_handle_t *hdl, const char *file, boolean_t force,
    boolean_t isspare)
{
    char  *name;
    int fd;
    int ret = 0;
    int err;
    pool_state_t state;
    boolean_t inuse;

    if (dm_inuse_swap(file, &err)) {
        if (err) {
            internal_error(hdl, err);
        } else {
            vdev_error(hdl, force, EZFS_DEV_INUSE,
                gettext("%s is currently used "
                "by swap. Please see swap(1M).\n"), file);
        }
        return (-1);
    }

    if ((fd = open64(file, O_RDONLY)) < 0) {
        (void) zfs_error_fmt(hdl, EZFS_OPENFAILED,
            dgettext(TEXT_DOMAIN, "cannot open '%s': %s\n"),
            file, strerror(errno));
        return (-1);
    }

    if (zpool_in_use(hdl, fd, &state, &name, &inuse) == 0 && inuse) {
        const char *desc;

        switch (state) {
        case POOL_STATE_ACTIVE:
            desc = gettext("active");
            break;

        case POOL_STATE_EXPORTED:
            desc = gettext("exported");
            break;

        case POOL_STATE_POTENTIALLY_ACTIVE:
            desc = gettext("potentially active");
            break;

        default:
            desc = gettext("unknown");
            break;
        }

        /*
         * Allow hot spares to be shared between pools.
         */
        if (state == POOL_STATE_SPARE && isspare)
            return (0);

        if (state == POOL_STATE_ACTIVE ||
            state == POOL_STATE_SPARE || !force) {
            switch (state) {
            case POOL_STATE_SPARE:
                vdev_error(hdl, force, EZFS_DEV_INUSE,
                    gettext("%s is reserved as a hot spare "
                    "for ZFS pool '%s'. Please see "
                    "zpool(1M)\n"), file, name);
                break;
            default:
                vdev_error(hdl, force, EZFS_DEV_INUSE,
                    gettext("%s is part of %s ZFS pool "
                    "'%s'. Please see zpool(1M)\n"),
                    file, desc, name);
                break;
            }
            ret = -1;
        }

        free(name);
    }

    (void) close(fd);
    return (ret);
}


/*
 * By "whole disk" we mean an entire physical disk (something we can
 * label, toggle the write cache on, etc.) as opposed to the full
 * capacity of a pseudo-device such as lofi or did.  We act as if we
 * are labeling the disk, which should be a pretty good test of whether
 * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 * it isn't.
 */
static boolean_t
is_whole_disk(const char *arg)
{
    struct dk_gpt *label;
    int fd;
    char    path[MAXPATHLEN];

    (void) snprintf(path, sizeof (path), "%s%s%s",
        RDISK_ROOT, strrchr(arg, '/'), BACKUP_SLICE);
    if ((fd = open(path, O_RDWR | O_NDELAY)) < 0)
        return (B_FALSE);
    if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
        (void) close(fd);
        return (B_FALSE);
    }
    efi_free(label);
    (void) close(fd);
    return (B_TRUE);
}

/*
 * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 * device, fill in the device id to make a complete nvlist.  Valid forms for a
 * leaf vdev are:
 *
 *  /dev/dsk/xxx        Complete disk path
 *  /xxx            Full path to file
 *  xxx         Shorthand for /dev/dsk/xxx
 *  /dev/chassis/.../disk   Symlink to whole-disk (xxx shorthand)
 *
 * It is sufficient to only record the failure cause as we fail immediately,
 * and the caller will issue the entire error message with zfs_error().
 */
static nvlist_t *
make_leaf_vdev(libzfs_handle_t *hdl, const char *arg, uint64_t is_log)
{
    char *lvdev = NULL;
    struct stat64 statbuf;
    char *rp;
    char path[MAXPATHLEN];
    nvlist_t *vdev = NULL;
    char *type = NULL;
    boolean_t wholedisk = B_FALSE;
    char *tmp;

    /*
     * If 'arg' is a symlink, read the link and try to convert it into
     * a whole-disk ctd name.  This is done to support references to
     * /dev/chassis/.../disk, where the link value resolves to the
     * /devices path. If the di_cro interfaces can convert the link value
     * into a disk ctd name, then we proceed with that ctd name.
     *
     * NOTE: To insulate us from new cro fields, we use the string-based
     * query interface.
     */
    if (libzfs_get_cro_hdl(hdl) && (lstat64(arg, &statbuf) == 0) &&
        S_ISLNK(statbuf.st_mode) &&
        (rp = realpath(arg, NULL))) {
        di_cro_reca_t ra;
        di_cro_rec_t r;
        int query_len;
        char *query;
        static const char *query_fmt =
            DI_CRO_Q_OCCUPANT_TYPE  DI_CRO_QREFMT   ","
            DI_CRO_Q_OCCUPANT_DEVICES   DI_CRO_QREFMT;

        query_len = strlen(query_fmt) +
            strlen("disk") + strlen(rp) + 1;
        if ((query = zfs_alloc(hdl, query_len)) == NULL)
            goto noloc;
        (void) snprintf(query, query_len, query_fmt, "disk", rp);
        ra = di_cro_reca_create_query((di_cro_hdl_t)
            libzfs_get_cro_hdl(hdl), 0, query);
        r = di_cro_reca_next(ra, NULL);
        lvdev = di_cro_rec_fgeti_occupant_compdev(r, 0, NULL, NULL);
        di_cro_reca_destroy(ra);
        free(query);
noloc:      free(rp);
    }

    if (lvdev == NULL)
        lvdev = (char *)arg;

    /*
     * Determine what type of vdev this is, and put the full path into
     * 'path'.  We detect whether this is a device of file afterwards by
     * checking the st_mode of the file.
     */
    if (lvdev[0] == '/') {
        /*
         * Complete device or file path.  Exact type is determined by
         * examining the file descriptor afterwards.
         */
        wholedisk = is_whole_disk(lvdev);
        if (!wholedisk && (stat64(lvdev, &statbuf) != 0)) {
            zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                "cannot open '%s': %s"), arg, strerror(errno));
            return (NULL);
        }

        (void) strlcpy(path, lvdev, sizeof (path));
    } else {
        /*
         * This may be a short path for a device, or it could be total
         * gibberish.  Check to see if it's a known device in
         * /dev/dsk/.  As part of this check, see if we've been given a
         * an entire disk (minus the slice number).
         */
        (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT,
            lvdev);
        wholedisk = is_whole_disk(path);
        if (!wholedisk && (stat64(path, &statbuf) != 0)) {
            /*
             * If we got ENOENT, then the user gave us
             * gibberish, so try to direct them with a
             * reasonable error message.  Otherwise,
             * regurgitate strerror() since it's the best we
             * can do.
             */
            if (errno == ENOENT) {
                zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                    "cannot open '%s': no such device in %s\n"
                    "must be a full path or shorthand device "
                    "name"), arg, DISK_ROOT);
            } else {
                zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                    "cannot open '%s': %s"),
                    path, strerror(errno));
            }
            return (NULL);
        }
    }

    /*
     * Determine whether this is a device or a file.
     */
    if (wholedisk || S_ISBLK(statbuf.st_mode)) {
        type = VDEV_TYPE_DISK;
    } else if (S_ISREG(statbuf.st_mode)) {
        type = VDEV_TYPE_FILE;
    } else {
        zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot use '%s': "
            "must be a block device or regular file"), path);
        return (NULL);
    }

    /*
     * Finally, we have the complete device or file, and we know that it is
     * acceptable to use.  Construct the nvlist to describe this vdev.  All
     * vdevs have a 'path' element, and devices also have a 'devid' element.
     */
    verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
    verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
    verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
    verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
    if (strcmp(type, VDEV_TYPE_DISK) == 0)
        verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
            (uint64_t)wholedisk) == 0);

    /*
     * Use di_cro interfaces to get /dev/chassis path.
     */
    if ((tmp = zfs_alloc(hdl, strlen(path) + 1)) != NULL) {
        if (strncmp(path, "/dev/dsk/", 9) == 0)
            (void) strcpy(tmp, path + 9);
        else
            (void) strcpy(tmp, path);
        if (tmp[0] == 'c' && isdigit(tmp[1])) {
            di_cro_reca_t ra;
            di_cro_rec_t r;
            int query_len;
            char *query, *slice, *dp = NULL;
            static const char *query_fmt =
                DI_CRO_Q_OCCUPANT_TYPE  DI_CRO_QREFMT   ","
                DI_CRO_Q_OCCUPANT_COMPDEV   DI_CRO_QREFMT;

            if ((slice = strchr(tmp, 's')) != NULL)
                *slice = '\0';
            query_len = strlen(query_fmt) + strlen("disk") +
                strlen(tmp) + 1;
            if ((query = zfs_alloc(hdl, query_len)) != NULL) {
                (void) snprintf(query, query_len, query_fmt,
                    "disk", tmp);
                ra = di_cro_reca_create_query((di_cro_hdl_t)
                    libzfs_get_cro_hdl(hdl), 0, query);
                r = di_cro_reca_next(ra, NULL);
                if ((dp = di_cro_rec_fgeti_devchassis_path(r, 0,
                    NULL, NULL)) != NULL)
                    verify(nvlist_add_string(vdev,
                        ZPOOL_CONFIG_DEVCHASSIS, dp) == 0);
                if ((dp = di_cro_rec_fgeti_chassis_id(r, 0,
                    NULL, NULL)) != NULL)
                    verify(nvlist_add_string(vdev,
                        ZPOOL_CONFIG_CHASSISSN, dp) == 0);
                if ((dp = di_cro_rec_fgeti_receptacle_name(r, 0,
                    NULL, NULL)) != NULL)
                    verify(nvlist_add_string(vdev,
                        ZPOOL_CONFIG_LOCATION, dp) == 0);
                di_cro_reca_destroy(ra);
                free(query);
            }
        }
        free(tmp);
    }

    /*
     * For a whole disk, defer getting its devid until after labeling it.
     */
    if (S_ISBLK(statbuf.st_mode) && !wholedisk) {
        /*
         * Get the devid for the device.
         */
        int fd;
        ddi_devid_t devid;
        char *minor = NULL, *devid_str = NULL;

        if ((fd = open(path, O_RDONLY)) < 0) {
            zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                "cannot open '%s': %s"), path, strerror(errno));
            nvlist_free(vdev);
            return (NULL);
        }

        if (devid_get(fd, &devid) == 0) {
            if (devid_get_minor_name(fd, &minor) == 0 &&
                (devid_str = devid_str_encode(devid, minor)) !=
                NULL) {
                verify(nvlist_add_string(vdev,
                    ZPOOL_CONFIG_DEVID, devid_str) == 0);
            }
            if (devid_str != NULL)
                devid_str_free(devid_str);
            if (minor != NULL)
                devid_str_free(minor);
            devid_free(devid);
        }

        (void) close(fd);
    }

    return (vdev);
}

/*
 * Go through and verify the replication level of the pool is consistent.
 * Performs the following checks:
 *
 *  For the new spec, verifies that devices in mirrors and raidz are the
 *  same size.
 *
 *  If the current configuration already has inconsistent replication
 *  levels, ignore any other potential problems in the new spec.
 *
 *  Otherwise, make sure that the current spec (if there is one) and the new
 *  spec have consistent replication levels.
 */
typedef struct replication_level {
    const char *zprl_type;
    uint64_t zprl_children;
    uint64_t zprl_parity;
} replication_level_t;

#define ZPOOL_FUZZ  (16 * 1024 * 1024)

/*
 * Check one top-level vdev and its children to ensure we don't mix devices
 * and files, and that the leaf vdevs are of approximately the same size.
 */
static boolean_t
check_tvdevs(libzfs_handle_t *hdl, nvlist_t **child, uint_t children,
    replication_level_t *rep, boolean_t fatal, boolean_t force)
{
    uint64_t vdev_size;
    boolean_t reported;
    const char *type;
    int c;
    /*
     * This is a mirror or RAID-Z vdev.  Go through and make
     * sure the contents are all the same (files vs. disks),
     * keeping track of the number of elements in the process.
     *
     * We also check that the size of each vdev (if it can
     * be determined) is the same.
     *
     * The 'reported' variable indicates that we've
     * already reported an error for this spec, so don't
     * bother doing it again.
     */
    type = NULL;
    reported = B_FALSE;
    vdev_size = -1ULL;
    rep->zprl_children = 0;

    for (c = 0; c < children; c++) {
        nvlist_t *cnv = child[c];
        char *path, *childtype;
        struct stat64 statbuf;
        uint64_t size = -1ULL;
        int fd, err;

        rep->zprl_children++;

        verify(nvlist_lookup_string(cnv,
            ZPOOL_CONFIG_TYPE, &childtype) == 0);

        /*
         * If this is a replacing or spare vdev, then
         * get the real first child of the vdev.
         */
        while (strcmp(childtype,
            VDEV_TYPE_REPLACING) == 0 ||
            strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
            nvlist_t **rchild;
            uint_t rchildren;

            verify(nvlist_lookup_nvlist_array(cnv,
                ZPOOL_CONFIG_CHILDREN, &rchild,
                &rchildren) == 0);
            assert(rchildren > 1);
            cnv = rchild[0];

            verify(nvlist_lookup_string(cnv,
                ZPOOL_CONFIG_TYPE,
                &childtype) == 0);
        }

        verify(nvlist_lookup_string(cnv,
            ZPOOL_CONFIG_PATH, &path) == 0);

        /*
         * If we have a raidz/mirror that combines disks
         * with files, report it as an error.
         */
        if (!reported && type != NULL && strcmp(type, childtype) != 0) {
            if (!fatal)
                return (B_FALSE);
            vdev_error(hdl, force, EZFS_INVALCONFIG, gettext(
                "mismatched replication level: %s contains both "
                "files and devices\n"), rep->zprl_type);
            reported = B_TRUE;
        }
        type = childtype;

        /*
         * According to stat(2), the value of 'st_size' is undefined
         * for block devices and character devices. But there is no
         * effective way to determine the real size in userland.
         *
         * Instead, we'll take advantage of an implementation detail of
         * spec_size(). If the device is currently open, then we
         * (should) return a valid size.
         *
         * If we still don't get a valid size (indicated by a size of 0
         * or MAXOFFSET_T), then ignore this device altogether.
         */
        if ((fd = open(path, O_RDONLY)) >= 0) {
            err = fstat64(fd, &statbuf);
            (void) close(fd);
        } else {
            err = stat64(path, &statbuf);
        }

        if (err != 0 ||
            statbuf.st_size == 0 ||
            statbuf.st_size == MAXOFFSET_T)
            continue;

        size = statbuf.st_size;

        /*
         * Also make sure that devices and slices have a consistent
         * size.  If they differ by a significant amount then report
         * an error.
         */
        if (!reported &&
            (vdev_size != -1ULL &&
            (labs(size - vdev_size) > ZPOOL_FUZZ))) {
            if (!fatal)
                return (B_FALSE);
            vdev_error(hdl, force, EZFS_INVALCONFIG, gettext(
                "%s contains devices of different sizes\n"),
                rep->zprl_type);
            reported = B_TRUE;
        }

        vdev_size = size;
    }

    /* success if we haven't reported an error */
    return (!reported);
}

/*
 * Check the replication levels in the existing pool's config and report
 * an error if there is a mismatch.
 */
static boolean_t
compare_rep_levels(libzfs_handle_t *hdl, replication_level_t *lastrep,
    replication_level_t *rep, boolean_t fatal, boolean_t force)
{
    if (lastrep->zprl_type == NULL)
        return (B_TRUE);
    if (strcmp(lastrep->zprl_type, rep->zprl_type) != 0) {
        if (fatal) {
            vdev_error(hdl, force, EZFS_INVALCONFIG, gettext(
                "mismatched replication level: both %s "
                "and %s vdevs are present\n"),
                lastrep->zprl_type, rep->zprl_type);
        }
    } else if (lastrep->zprl_parity != rep->zprl_parity) {
        if (fatal) {
            vdev_error(hdl, force, EZFS_INVALCONFIG, gettext(
                "mismatched replication level: both %llu "
                "and %llu and %llu device parity %s "
                "vdevs are present\n"),
                lastrep->zprl_parity,
                rep->zprl_parity, rep->zprl_type);
        }
    } else if (lastrep->zprl_children != rep->zprl_children) {
        if (fatal) {
            vdev_error(hdl, force, EZFS_INVALCONFIG, gettext(
                "mismatched replication level: both "
                "%llu-way and %llu-way %s vdevs are "
                "present\n"),
                lastrep->zprl_children,
                rep->zprl_children, rep->zprl_type);
        }
    } else {
        return (B_TRUE);
    }

    return (B_FALSE);
}

/*
 * Check the replication levels between the existing config and the
 * config that the user wants to have.  This differs from compare_rep_levels()
 * only in the strings we output.
 */
static boolean_t
check_new_levels(libzfs_handle_t *hdl, replication_level_t *current,
    replication_level_t *new, boolean_t force)
{
    if (current->zprl_type == NULL || new->zprl_type == NULL)
        return (B_TRUE);
    if (strcmp(current->zprl_type, new->zprl_type) != 0) {
        vdev_error(hdl, force, EZFS_INVALCONFIG, gettext(
            "mismatched replication level: pool uses %s and new "
            "vdev is %s\n"),
            current->zprl_type, new->zprl_type);
    } else if (current->zprl_parity != new->zprl_parity) {
        vdev_error(hdl, force, EZFS_INVALCONFIG, gettext(
            "mismatched replication level: pool uses %llu device "
            "parity and new vdev uses %llu\n"),
            current->zprl_parity, new->zprl_parity);
    } else if (current->zprl_children != new->zprl_children) {
        vdev_error(hdl, force, EZFS_INVALCONFIG, gettext(
            "mismatched replication level: pool uses %llu-way %s "
            "and new vdev uses %llu-way %s\n"),
            current->zprl_children, current->zprl_type,
            new->zprl_children, new->zprl_type);
    } else {
        return (B_TRUE);
    }

    return (B_FALSE);
}

/*
 * Given a list of toplevel vdevs, determine if the configuration is
 * self-consistent. If the config is inconsistent, return false. If 'fatal' is
 * set, then an error message will be displayed for each self-inconsistent
 * vdev.
 */
static boolean_t
valid_replication(libzfs_handle_t *hdl, nvlist_t *nvroot,
    replication_level_t *pool, replication_level_t *slog, boolean_t fatal,
    boolean_t force)
{
    nvlist_t **top;
    uint_t t, toplevels;
    nvlist_t **child;
    uint_t children;
    nvlist_t *nv;
    char *type;
    replication_level_t *prev, rep;
    boolean_t ret = B_TRUE;

    pool->zprl_type = NULL;
    slog->zprl_type = NULL;

    verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
        &top, &toplevels) == 0);

    for (t = 0; t < toplevels; t++) {
        uint64_t is_log = B_FALSE;

        nv = top[t];

        /*
         * For separate logs we ignore the top level vdev replication
         * constraints.
         */
        (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);

        verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
            &type) == 0);
        rep.zprl_type = type;
        rep.zprl_parity = 0;
        if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
            &child, &children) != 0) {
            /*
             * This is a 'file' or 'disk' vdev.
             */
            rep.zprl_children = 1;
        } else {
            /*
             * If this is a raidz device, update the parity.
             */
            if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
                verify(nvlist_lookup_uint64(nv,
                    ZPOOL_CONFIG_NPARITY,
                    &rep.zprl_parity) == 0);
                assert(rep.zprl_parity != 0);
            }

            if (!check_tvdevs(hdl, child, children, &rep, fatal,
                force)) {
                if (!fatal)
                    return (B_FALSE);
                ret = B_FALSE;
            }
        }

        /*
         * At this point, we have the replication of the last toplevel
         * vdev in 'rep'.  Compare it to the pool or the slog.
         */
        prev = is_log ? slog : pool;
        if (!compare_rep_levels(hdl, prev, &rep, fatal, force)) {
            if (!fatal)
                return (B_FALSE);
            ret = B_FALSE;
        }
        *prev = rep;
    }

    return (ret);
}

/*
 * Check the replication level of the vdev spec against the current pool. Calls
 * valid_replication() to make sure the new spec is self-consistent.  If the
 * pool has a consistent replication level, then we ignore any errors.
 * Otherwise, report any difference between the two.
 */
static int
check_replication(libzfs_handle_t *hdl, nvlist_t *config, nvlist_t *newroot,
    boolean_t force)
{
    nvlist_t **child;
    uint_t children;
    replication_level_t p_curr = {0}, p_new;    /* pool info */
    replication_level_t s_curr = {0}, s_new;    /* slog info */

    /*
     * If we have a current pool configuration, check to see if it's
     * self-consistent.  If not, simply return success.
     */
    if (config != NULL) {
        nvlist_t *nvroot;

        verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
            &nvroot) == 0);
        if (!valid_replication(hdl, nvroot, &p_curr, &s_curr, B_FALSE,
            force))
            return (0);
    }

    /*
     * for spares there may be no children, and therefore no
     * replication level to check
     */
    if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
        &child, &children) != 0) || (children == 0))
        return (0);

    /*
     * Get the replication level of the new vdev spec, reporting any
     * inconsistencies found.
     */
    if (!valid_replication(hdl, newroot, &p_new, &s_new, B_TRUE, force))
        return (-1);

    /*
     * Check to see if the new vdev spec matches the replication level of
     * the current pool.
     */
    if (!check_new_levels(hdl, &p_curr, &p_new, force) ||
        !check_new_levels(hdl, &s_curr, &s_new, force))
        return (-1);

    return (0);
}

/*
 * Go through and find any whole disks in the vdev specification, labelling them
 * as appropriate.  When constructing the vdev spec, we were unable to open this
 * device in order to provide a devid.  Now that we have labelled the disk and
 * know that slice 0 is valid, we can construct the devid now.
 *
 * If the disk was already labeled with an EFI label, we will have gotten the
 * devid already (because we were able to open the whole disk).  Otherwise, we
 * need to get the devid after we label the disk.
 */
static int
make_disks(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
    int create_req_part)
{
    nvlist_t **child;
    uint_t c, children;
    char *type, *path, *diskname;
    char buf[MAXPATHLEN];
    uint64_t wholedisk;
    int fd;
    int ret;
    ddi_devid_t devid;
    char *minor = NULL, *devid_str = NULL;
    int slicenum;

    verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);

    if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
        &child, &children) != 0) {

        if (strcmp(type, VDEV_TYPE_DISK) != 0)
            return (0);

        /*
         * We have a disk device.  Get the path to the device
         * and see if it's a whole disk by appending the backup
         * slice and stat()ing the device.
         */
        verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
        if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
            &wholedisk) != 0 || !wholedisk)
            return (0);

        diskname = strrchr(path, '/');
        assert(diskname != NULL);
        diskname++;

        if (zpool_label_disk(hdl, zhp, diskname, create_req_part,
            &slicenum) == -1)
            return (-1);

        /*
         * Fill in the devid, now that we've labeled the disk.
         */
        (void) snprintf(buf, sizeof (buf), "%ss%d", path, slicenum);
        if ((fd = open(buf, O_RDONLY)) < 0) {
            (void) zfs_error_fmt(hdl, EZFS_OPENFAILED,
                dgettext(TEXT_DOMAIN, "cannot open '%s': %s\n"),
                buf, strerror(errno));
            return (-1);
        }

        if (devid_get(fd, &devid) == 0) {
            if (devid_get_minor_name(fd, &minor) == 0 &&
                (devid_str = devid_str_encode(devid, minor)) !=
                NULL) {
                verify(nvlist_add_string(nv,
                    ZPOOL_CONFIG_DEVID, devid_str) == 0);
            }
            if (devid_str != NULL)
                devid_str_free(devid_str);
            if (minor != NULL)
                devid_str_free(minor);
            devid_free(devid);
        }

        /*
         * Update the path to refer to the slice.  The presence of
         * the 'whole_disk' field indicates to the CLI that we should
         * chop off the slice number when displaying the device in
         * future output.
         */
        verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0);
        (void) close(fd);

        return (0);
    }

    for (c = 0; c < children; c++) {
        if ((ret = make_disks(hdl, zhp, child[c],
            create_req_part)) != 0)
            return (ret);
    }
    if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
        &child, &children) == 0) {
        for (c = 0; c < children; c++) {
            if ((ret = make_disks(hdl, zhp, child[c],
                create_req_part)) != 0)
                return (ret);
        }
    }
    if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
        &child, &children) == 0) {
        for (c = 0; c < children; c++) {
            if ((ret = make_disks(hdl, zhp, child[c],
                create_req_part)) != 0)
                return (ret);
        }
    }
    return (0);
}

/*
 * Determine if the given path is a hot spare within the given configuration.
 */
static boolean_t
is_spare(libzfs_handle_t *hdl, nvlist_t *config, const char *path)
{
    int fd;
    pool_state_t state;
    char *name = NULL;
    nvlist_t *label;
    uint64_t guid, spareguid;
    nvlist_t *nvroot;
    nvlist_t **spares;
    uint_t i, nspares;
    boolean_t inuse;

    if ((fd = open(path, O_RDONLY)) < 0)
        return (B_FALSE);

    if (zpool_in_use(hdl, fd, &state, &name, &inuse) != 0 ||
        !inuse ||
        state != POOL_STATE_SPARE ||
        zpool_read_label(fd, &label) != 0) {
        free(name);
        (void) close(fd);
        return (B_FALSE);
    }
    free(name);
    (void) close(fd);

    verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
    nvlist_free(label);

    verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
        &nvroot) == 0);
    if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
        &spares, &nspares) == 0) {
        for (i = 0; i < nspares; i++) {
            verify(nvlist_lookup_uint64(spares[i],
                ZPOOL_CONFIG_GUID, &spareguid) == 0);
            if (spareguid == guid)
                return (B_TRUE);
        }
    }

    return (B_FALSE);
}

/*
 * Go through and find any devices that are in use.  We rely on libdiskmgt for
 * the majority of this task.
 */
static int
check_in_use(libzfs_handle_t *hdl, nvlist_t *config, nvlist_t *nv,
    boolean_t force, boolean_t replacing, boolean_t isspare)
{
    nvlist_t **child;
    uint_t c, children;
    int ret;

    if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
        &child, &children) != 0) {
        char buf[MAXPATHLEN];
        char *type, *path;
        uint64_t wholedisk;

        verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
        verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
        /*
         * As a generic check, we look to see if this is a replace of a
         * hot spare within the same pool.  If so, we allow it
         * regardless of what libdiskmgt or zpool_in_use() says.
         */
        if (replacing) {
            if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
                &wholedisk) == 0 && wholedisk) {

                    (void) snprintf(buf, sizeof (buf),
                        "%ss%d", path, 0);

                    if (is_spare(hdl, config, buf)) {
                        return (0);
                    } else {
                        (void) snprintf(buf,
                            sizeof (buf),
                            "%ss%d", path, 1);
                        if (is_spare(hdl, config, buf))
                            return (0);
                    }

            } else {
                (void) strlcpy(buf, path, sizeof (buf));

                if (is_spare(hdl, config, buf))
                    return (0);
            }
        }

        if (strcmp(type, VDEV_TYPE_DISK) == 0)
            ret = check_device(hdl, path, force, isspare);

        if (strcmp(type, VDEV_TYPE_FILE) == 0)
            ret = check_file(hdl, path, force, isspare);

        return (ret);
    }

    for (c = 0; c < children; c++) {
        if ((ret = check_in_use(hdl, config, child[c], force,
            replacing, B_FALSE)) != 0)
            return (ret);
    }
    if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
        &child, &children) == 0) {
        for (c = 0; c < children; c++) {
            if ((ret = check_in_use(hdl, config, child[c], force,
                replacing, B_TRUE)) != 0)
                return (ret);
        }
    }
    if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
        &child, &children) == 0) {
        for (c = 0; c < children; c++) {
            if ((ret = check_in_use(hdl, config, child[c], force,
                replacing, B_FALSE)) != 0)
                return (ret);
        }
    }
    return (0);
}

static const char *
is_grouping(const char *type, int *mindev, int *maxdev)
{
    if (strncmp(type, "raidz", 5) == 0) {
        const char *p = type + 5;
        char *end;
        long nparity;

        if (*p == '\0') {
            nparity = 1;
        } else if (*p == '0') {
            return (NULL); /* no zero prefixes allowed */
        } else {
            errno = 0;
            nparity = strtol(p, &end, 10);
            if (errno != 0 || nparity < 1 || nparity >= 255 ||
                *end != '\0')
                return (NULL);
        }

        if (mindev != NULL)
            *mindev = nparity + 1;
        if (maxdev != NULL)
            *maxdev = 255;
        return (VDEV_TYPE_RAIDZ);
    }

    if (maxdev != NULL)
        *maxdev = INT_MAX;

    if (strcmp(type, "mirror") == 0) {
        if (mindev != NULL)
            *mindev = 2;
        return (VDEV_TYPE_MIRROR);
    }

    if (strcmp(type, "spare") == 0) {
        if (mindev != NULL)
            *mindev = 1;
        return (VDEV_TYPE_SPARE);
    }

    if (strcmp(type, "log") == 0) {
        if (mindev != NULL)
            *mindev = 1;
        return (VDEV_TYPE_LOG);
    }

    if (strcmp(type, "cache") == 0) {
        if (mindev != NULL)
            *mindev = 1;
        return (VDEV_TYPE_L2CACHE);
    }

    return (NULL);
}

/*
 * Construct a syntactically valid vdev specification,
 * and ensure that all devices and files exist and can be opened.
 * It is sufficient to only record the failure cause as we fail immediately,
 * and the caller will issue the entire error message with zfs_error().
 */
static nvlist_t *
construct_spec(libzfs_handle_t *hdl, int vdevcount, char **vdevlist)
{
    nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
    int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
    const char *type;
    uint64_t is_log;
    boolean_t seen_logs;
    char msg[1024];

    nvroot = NULL;
    top = NULL;
    toplevels = 0;
    spares = NULL;
    l2cache = NULL;
    nspares = 0;
    nlogs = 0;
    nl2cache = 0;
    is_log = B_FALSE;
    seen_logs = B_FALSE;
    (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
        "invalid vdev specification: "));

    while (vdevcount > 0) {
        nv = NULL;
        /*
         * If it's a mirror or raidz, the subsequent arguments are
         * its leaves -- until we encounter the next mirror or raidz.
         */
        if ((type = is_grouping(vdevlist[0], &mindev, &maxdev))
            != NULL) {
            nvlist_t **child = NULL;
            int c, children = 0;

            if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
                if (spares != NULL) {
                    zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                        "%s'%s' can be specified only "
                        "once"), msg, VDEV_TYPE_SPARE);
                    goto errout;
                }
                is_log = B_FALSE;
            }
            if (strcmp(type, VDEV_TYPE_LOG) == 0) {
                if (seen_logs) {
                    zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                        "%s'%s' can be specified only "
                        "once"), msg, VDEV_TYPE_LOG);
                    goto errout;
                }
                seen_logs = B_TRUE;
                is_log = B_TRUE;
                vdevcount--;
                vdevlist++;
                /*
                 * A log is not a real grouping device.
                 * We just set is_log and continue.
                 */
                continue;
            }
            if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
                if (l2cache != NULL) {
                    zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                        "%s'%s' can be specified only "
                        "once"), msg, VDEV_TYPE_L2CACHE);
                    goto errout;
                }
                is_log = B_FALSE;
            }
            if (is_log) {
                if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
                    zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                        "%sunsupported '%s' device: %s"),
                        msg, VDEV_TYPE_LOG, type);
                    goto errout;
                }
                nlogs++;
            }
            for (c = 1; c < vdevcount; c++) {
                if (is_grouping(vdevlist[c], NULL, NULL)
                    != NULL)
                    break;
                child = zfs_realloc(hdl, child, children *
                    sizeof (nvlist_t *), (children + 1) *
                    sizeof (nvlist_t *));
                if ((nv = make_leaf_vdev(hdl, vdevlist[c],
                    B_FALSE)) == NULL) {
                    for (int i = 0; i < children; i++)
                        nvlist_free(child[i]);
                    free(child);
                    goto errout;
                }
                child[children] = nv;
                children++;
            }
            if (children < mindev) {
                zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                    "%s%s requires at least %d devices"),
                    msg, vdevlist[0], mindev);
                if (child) {
                    for (int i = 0; i < children; i++)
                        nvlist_free(child[i]);
                    free(child);
                }
                goto errout;
            }
            if (children > maxdev) {
                zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                    "%s%s supports no more then %d devices"),
                    msg, vdevlist[0], maxdev);
                if (child) {
                    for (int i = 0; i < children; i++)
                        nvlist_free(child[i]);
                    free(child);
                }
                goto errout;
            }

            vdevcount -= c;
            vdevlist += c;

            if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
                spares = child;
                nspares = children;
                continue;
            } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
                l2cache = child;
                nl2cache = children;
                continue;
            } else {
                verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
                    0) == 0);
                verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
                    type) == 0);
                verify(nvlist_add_uint64(nv,
                    ZPOOL_CONFIG_IS_LOG, is_log) == 0);
                if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
                    verify(nvlist_add_uint64(nv,
                        ZPOOL_CONFIG_NPARITY,
                        mindev - 1) == 0);
                }
                verify(nvlist_add_nvlist_array(nv,
                    ZPOOL_CONFIG_CHILDREN, child,
                    children) == 0);

                for (c = 0; c < children; c++)
                    nvlist_free(child[c]);
                free(child);
            }
        } else {
            /*
             * We have a device.  Pass off to make_leaf_vdev() to
             * construct the appropriate nvlist describing the vdev.
             */
            if ((nv = make_leaf_vdev(hdl, vdevlist[0], is_log))
                == NULL)
                goto errout;
            if (is_log)
                nlogs++;
            vdevcount--;
            vdevlist++;
        }

        top = zfs_realloc(hdl, top, toplevels * sizeof (nvlist_t *),
            (toplevels + 1) * sizeof (nvlist_t *));
        top[toplevels] = nv;
        toplevels++;
    }

    if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
        zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
            "%sat least one toplevel vdev must be specified"), msg);
        goto errout;
    }

    if (seen_logs && nlogs == 0) {
        zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s'%s' "
            "requires at least 1 device"), msg, VDEV_TYPE_LOG);
        goto errout;
    }

    /*
     * Finally, create nvroot and add all top-level vdevs to it.
     */
    verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
    verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
        VDEV_TYPE_ROOT) == 0);
    verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
        top, toplevels) == 0);
    if (nspares != 0)
        verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
            spares, nspares) == 0);
    if (nl2cache != 0)
        verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
            l2cache, nl2cache) == 0);
errout:
    for (t = 0; t < toplevels; t++)
        nvlist_free(top[t]);
    for (t = 0; t < nspares; t++)
        nvlist_free(spares[t]);
    for (t = 0; t < nl2cache; t++)
        nvlist_free(l2cache[t]);
    free(spares);
    free(l2cache);
    free(top);

    return (nvroot);
}

nvlist_t *
zpool_split_root_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
    splitflags_t flags, int vdevcount, char **vdevlist)
{
    nvlist_t *newroot = NULL;

    assert(zhp != NULL);

    if (vdevcount > 0) {
        char msg[1024];
        nvlist_t **child;
        uint_t c, children;
        libzfs_handle_t *hdl = zhp->zpool_hdl;

        (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
            "Unable to split '%s'"), zhp->zpool_name);

        if ((newroot = construct_spec(hdl, vdevcount, vdevlist))
            == NULL) {
            (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
            return (NULL);
        }
        if (!flags.dryrun && make_disks(hdl, zhp, newroot,
            ZPOOL_LABEL_MATCH_REQ_PART) != 0) {
            (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
            nvlist_free(newroot);
            return (NULL);
        }
        /* avoid any tricks in the spec */
        verify(nvlist_lookup_nvlist_array(newroot,
            ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
        for (c = 0; c < children; c++) {
            char *path;
            const char *type;
            int min, max;

            verify(nvlist_lookup_string(child[c],
                ZPOOL_CONFIG_PATH, &path) == 0);
            if ((type = is_grouping(path, &min, &max)) != NULL) {
                zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
                    "invalid vdev specification: cannot use "
                    "'%s' as a device for splitting"),
                    type);
                (void) zfs_error(hdl, EZFS_BADDEV, msg);
                nvlist_free(newroot);
                return (NULL);
            }
        }
    }

    if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
        /* zpool_vdev_split() does the error messaging here */
        if (newroot != NULL)
            nvlist_free(newroot);
        return (NULL);
    }

    return (newroot);
}

/*
 * Get and validate the contents of the given vdev specification.  This ensures
 * that the nvlist returned is well-formed, that all the devices exist, and that
 * they are not currently in use by any other known consumer.  The 'poolconfig'
 * parameter is the current configuration of the pool when adding devices
 * existing pool, and is used to perform additional checks, such as changing the
 * replication level of the pool.  It can be 'NULL' to indicate that this is a
 * new pool.  The 'force' flag controls whether devices should be forcefully
 * added, even if they appear in use.
 */
nvlist_t *
zpool_make_root_vdev(libzfs_handle_t *hdl, zpool_handle_t *zhp, boolean_t force,
    int check_rep, boolean_t replacing, boolean_t dryrun, int create_req_part,
    int vdevcount, char **vdevlist)
{
    nvlist_t *newroot;
    nvlist_t *poolconfig = NULL;
    char msg[1024];

    (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
        "Unable to build pool from specified devices"));
    /*
     * Construct the vdev specification.  If this is successful, we know
     * that we have a valid specification, and that all devices can be
     * opened.
     */
    if ((newroot = construct_spec(hdl, vdevcount, vdevlist)) == NULL) {
        (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
        return (NULL);
    }
    if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
        (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
        return (NULL);
    }
    /*
     * Validate each device to make sure that its not shared with another
     * subsystem.  We do this even if 'force' is set, because there are some
     * uses (such as a dedicated dump device) that even '-f' cannot
     * override.
     */
    if (check_in_use(hdl, poolconfig, newroot, force, replacing, B_FALSE)
        != 0) {
        nvlist_free(newroot);
        (void) zfs_error(hdl, EZFS_DEV_INUSE, msg);
        return (NULL);
    }
    /*
     * Check the replication level of the given vdevs and report any errors
     * found.  We include the existing pool spec, if any, as we need to
     * catch changes against the existing replication level.
     */
    if (check_rep && check_replication(hdl, poolconfig, newroot, force)
        != 0) {
        nvlist_free(newroot);
        (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
        return (NULL);
    }
    /*
     * Run through the vdev specification and label any whole disks found.
     */
    if (!dryrun && make_disks(hdl, zhp, newroot, create_req_part) != 0) {
        nvlist_free(newroot);
        (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
        return (NULL);
    }

    return (newroot);
}