cddl/contrib/opensolaris/cmd/zpool/zpool_vdev.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  */
  25
  26 /*
  27  * Functions to convert between a list of vdevs and an nvlist representing the
  28  * configuration.  Each entry in the list can be one of:
  29  *
  30  *      Device vdevs
  31  *              disk=(path=..., devid=...)
  32  *              file=(path=...)
  33  *
  34  *      Group vdevs
  35  *              raidz[1|2]=(...)
  36  *              mirror=(...)
  37  *
  38  *      Hot spares
  39  *
  40  * While the underlying implementation supports it, group vdevs cannot contain
  41  * other group vdevs.  All userland verification of devices is contained within
  42  * this file.  If successful, the nvlist returned can be passed directly to the
  43  * kernel; we've done as much verification as possible in userland.
  44  *
  45  * Hot spares are a special case, and passed down as an array of disk vdevs, at
  46  * the same level as the root of the vdev tree.
  47  *
  48  * The only function exported by this file is 'make_root_vdev'.  The
  49  * function performs several passes:
  50  *
  51  *      1. Construct the vdev specification.  Performs syntax validation and
  52  *         makes sure each device is valid.
  53  *      2. Check for devices in use.  Using libdiskmgt, makes sure that no
  54  *         devices are also in use.  Some can be overridden using the 'force'
  55  *         flag, others cannot.
  56  *      3. Check for replication errors if the 'force' flag is not specified.
  57  *         validates that the replication level is consistent across the
  58  *         entire pool.
  59  *      4. Call libzfs to label any whole disks with an EFI label.
  60  */
  61
  62 #include <assert.h>
  63 #include <devid.h>
  64 #include <errno.h>
  65 #include <fcntl.h>
  66 #include <libintl.h>
  67 #include <libnvpair.h>
  68 #include <limits.h>
  69 #include <stdio.h>
  70 #include <string.h>
  71 #include <unistd.h>
  72 #include <paths.h>
  73 #include <sys/stat.h>
  74 #include <sys/disk.h>
  75 #include <sys/mntent.h>
  76 #include <libgeom.h>
  77
  78 #include "zpool_util.h"
  79
  80 #define DISK_ROOT       "/dev/dsk"
  81 #define RDISK_ROOT      "/dev/rdsk"
  82 #define BACKUP_SLICE    "s2"
  83
  84 /*
  85  * For any given vdev specification, we can have multiple errors.  The
  86  * vdev_error() function keeps track of whether we have seen an error yet, and
  87  * prints out a header if its the first error we've seen.
  88  */
  89 boolean_t error_seen;
  90 boolean_t is_force;
  91
  92 /*PRINTFLIKE1*/
  93 static void
  94 vdev_error(const char *fmt, ...)
  95 {
  96         va_list ap;
  97
  98         if (!error_seen) {
  99                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
 100                 if (!is_force)
 101                         (void) fprintf(stderr, gettext("use '-f' to override "
 102                             "the following errors:\n"));
 103                 else
 104                         (void) fprintf(stderr, gettext("the following errors "
 105                             "must be manually repaired:\n"));
 106                 error_seen = B_TRUE;
 107         }
 108
 109         va_start(ap, fmt);
 110         (void) vfprintf(stderr, fmt, ap);
 111         va_end(ap);
 112 }
 113
 114 #ifdef sun
 115 static void
 116 libdiskmgt_error(int error)
 117 {
 118         /*
 119          * ENXIO/ENODEV is a valid error message if the device doesn't live in
 120          * /dev/dsk.  Don't bother printing an error message in this case.
 121          */
 122         if (error == ENXIO || error == ENODEV)
 123                 return;
 124
 125         (void) fprintf(stderr, gettext("warning: device in use checking "
 126             "failed: %s\n"), strerror(error));
 127 }
 128
 129 /*
 130  * Validate a device, passing the bulk of the work off to libdiskmgt.
 131  */
 132 static int
 133 check_slice(const char *path, int force, boolean_t wholedisk, boolean_t isspare)
 134 {
 135         char *msg;
 136         int error = 0;
 137         dm_who_type_t who;
 138
 139         if (force)
 140                 who = DM_WHO_ZPOOL_FORCE;
 141         else if (isspare)
 142                 who = DM_WHO_ZPOOL_SPARE;
 143         else
 144                 who = DM_WHO_ZPOOL;
 145
 146         if (dm_inuse((char *)path, &msg, who, &error) || error) {
 147                 if (error != 0) {
 148                         libdiskmgt_error(error);
 149                         return (0);
 150                 } else {
 151                         vdev_error("%s", msg);
 152                         free(msg);
 153                         return (-1);
 154                 }
 155         }
 156
 157         /*
 158          * If we're given a whole disk, ignore overlapping slices since we're
 159          * about to label it anyway.
 160          */
 161         error = 0;
 162         if (!wholedisk && !force &&
 163             (dm_isoverlapping((char *)path, &msg, &error) || error)) {
 164                 if (error == 0) {
 165                         /* dm_isoverlapping returned -1 */
 166                         vdev_error(gettext("%s overlaps with %s\n"), path, msg);
 167                         free(msg);
 168                         return (-1);
 169                 } else if (error != ENODEV) {
 170                         /* libdiskmgt's devcache only handles physical drives */
 171                         libdiskmgt_error(error);
 172                         return (0);
 173                 }
 174         }
 175
 176         return (0);
 177 }
 178
 179
 180 /*
 181  * Validate a whole disk.  Iterate over all slices on the disk and make sure
 182  * that none is in use by calling check_slice().
 183  */
 184 static int
 185 check_disk(const char *name, dm_descriptor_t disk, int force, int isspare)
 186 {
 187         dm_descriptor_t *drive, *media, *slice;
 188         int err = 0;
 189         int i;
 190         int ret;
 191
 192         /*
 193          * Get the drive associated with this disk.  This should never fail,
 194          * because we already have an alias handle open for the device.
 195          */
 196         if ((drive = dm_get_associated_descriptors(disk, DM_DRIVE,
 197             &err)) == NULL || *drive == NULL) {
 198                 if (err)
 199                         libdiskmgt_error(err);
 200                 return (0);
 201         }
 202
 203         if ((media = dm_get_associated_descriptors(*drive, DM_MEDIA,
 204             &err)) == NULL) {
 205                 dm_free_descriptors(drive);
 206                 if (err)
 207                         libdiskmgt_error(err);
 208                 return (0);
 209         }
 210
 211         dm_free_descriptors(drive);
 212
 213         /*
 214          * It is possible that the user has specified a removable media drive,
 215          * and the media is not present.
 216          */
 217         if (*media == NULL) {
 218                 dm_free_descriptors(media);
 219                 vdev_error(gettext("'%s' has no media in drive\n"), name);
 220                 return (-1);
 221         }
 222
 223         if ((slice = dm_get_associated_descriptors(*media, DM_SLICE,
 224             &err)) == NULL) {
 225                 dm_free_descriptors(media);
 226                 if (err)
 227                         libdiskmgt_error(err);
 228                 return (0);
 229         }
 230
 231         dm_free_descriptors(media);
 232
 233         ret = 0;
 234
 235         /*
 236          * Iterate over all slices and report any errors.  We don't care about
 237          * overlapping slices because we are using the whole disk.
 238          */
 239         for (i = 0; slice[i] != NULL; i++) {
 240                 char *name = dm_get_name(slice[i], &err);
 241
 242                 if (check_slice(name, force, B_TRUE, isspare) != 0)
 243                         ret = -1;
 244
 245                 dm_free_name(name);
 246         }
 247
 248         dm_free_descriptors(slice);
 249         return (ret);
 250 }
 251
 252 /*
 253  * Validate a device.
 254  */
 255 static int
 256 check_device(const char *path, boolean_t force, boolean_t isspare)
 257 {
 258         dm_descriptor_t desc;
 259         int err;
 260         char *dev;
 261
 262         /*
 263          * For whole disks, libdiskmgt does not include the leading dev path.
 264          */
 265         dev = strrchr(path, '/');
 266         assert(dev != NULL);
 267         dev++;
 268         if ((desc = dm_get_descriptor_by_name(DM_ALIAS, dev, &err)) != NULL) {
 269                 err = check_disk(path, desc, force, isspare);
 270                 dm_free_descriptor(desc);
 271                 return (err);
 272         }
 273
 274         return (check_slice(path, force, B_FALSE, isspare));
 275 }
 276 #endif  /* sun */
 277
 278 /*
 279  * Check that a file is valid.  All we can do in this case is check that it's
 280  * not in use by another pool, and not in use by swap.
 281  */
 282 static int
 283 check_file(const char *file, boolean_t force, boolean_t isspare)
 284 {
 285         char  *name;
 286         int fd;
 287         int ret = 0;
 288         int err;
 289         pool_state_t state;
 290         boolean_t inuse;
 291
 292 #ifdef sun
 293         if (dm_inuse_swap(file, &err)) {
 294                 if (err)
 295                         libdiskmgt_error(err);
 296                 else
 297                         vdev_error(gettext("%s is currently used by swap. "
 298                             "Please see swap(1M).\n"), file);
 299                 return (-1);
 300         }
 301 #endif
 302
 303         if ((fd = open(file, O_RDONLY)) < 0)
 304                 return (0);
 305
 306         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
 307                 const char *desc;
 308
 309                 switch (state) {
 310                 case POOL_STATE_ACTIVE:
 311                         desc = gettext("active");
 312                         break;
 313
 314                 case POOL_STATE_EXPORTED:
 315                         desc = gettext("exported");
 316                         break;
 317
 318                 case POOL_STATE_POTENTIALLY_ACTIVE:
 319                         desc = gettext("potentially active");
 320                         break;
 321
 322                 default:
 323                         desc = gettext("unknown");
 324                         break;
 325                 }
 326
 327                 /*
 328                  * Allow hot spares to be shared between pools.
 329                  */
 330                 if (state == POOL_STATE_SPARE && isspare)
 331                         return (0);
 332
 333                 if (state == POOL_STATE_ACTIVE ||
 334                     state == POOL_STATE_SPARE || !force) {
 335                         switch (state) {
 336                         case POOL_STATE_SPARE:
 337                                 vdev_error(gettext("%s is reserved as a hot "
 338                                     "spare for pool %s\n"), file, name);
 339                                 break;
 340                         default:
 341                                 vdev_error(gettext("%s is part of %s pool "
 342                                     "'%s'\n"), file, desc, name);
 343                                 break;
 344                         }
 345                         ret = -1;
 346                 }
 347
 348                 free(name);
 349         }
 350
 351         (void) close(fd);
 352         return (ret);
 353 }
 354
 355 static int
 356 check_device(const char *name, boolean_t force, boolean_t isspare)
 357 {
 358         char path[MAXPATHLEN];
 359
 360         if (strncmp(name, _PATH_DEV, sizeof(_PATH_DEV) - 1) != 0)
 361                 snprintf(path, sizeof(path), "%s%s", _PATH_DEV, name);
 362         else
 363                 strlcpy(path, name, sizeof(path));
 364
 365         return (check_file(path, force, isspare));
 366 }
 367
 368 /*
 369  * By "whole disk" we mean an entire physical disk (something we can
 370  * label, toggle the write cache on, etc.) as opposed to the full
 371  * capacity of a pseudo-device such as lofi or did.  We act as if we
 372  * are labeling the disk, which should be a pretty good test of whether
 373  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 374  * it isn't.
 375  */
 376 static boolean_t
 377 is_whole_disk(const char *arg)
 378 {
 379 #ifdef sun
 380         struct dk_gpt *label;
 381         int     fd;
 382         char    path[MAXPATHLEN];
 383
 384         (void) snprintf(path, sizeof (path), "%s%s%s",
 385             RDISK_ROOT, strrchr(arg, '/'), BACKUP_SLICE);
 386         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0)
 387                 return (B_FALSE);
 388         if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
 389                 (void) close(fd);
 390                 return (B_FALSE);
 391         }
 392         efi_free(label);
 393         (void) close(fd);
 394         return (B_TRUE);
 395 #else
 396         int fd;
 397
 398         fd = g_open(arg, 0);
 399         if (fd >= 0) {
 400                 g_close(fd);
 401                 return (B_TRUE);
 402         }
 403         return (B_FALSE);
 404 #endif
 405 }
 406
 407 /*
 408  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 409  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 410  * leaf vdev are:
 411  *
 412  *      /dev/dsk/xxx    Complete disk path
 413  *      /xxx            Full path to file
 414  *      xxx             Shorthand for /dev/dsk/xxx
 415  */
 416 static nvlist_t *
 417 make_leaf_vdev(const char *arg, uint64_t is_log)
 418 {
 419         char path[MAXPATHLEN];
 420         struct stat64 statbuf;
 421         nvlist_t *vdev = NULL;
 422         char *type = NULL;
 423         boolean_t wholedisk = B_FALSE;
 424
 425         /*
 426          * Determine what type of vdev this is, and put the full path into
 427          * 'path'.  We detect whether this is a device of file afterwards by
 428          * checking the st_mode of the file.
 429          */
 430         if (arg[0] == '/') {
 431                 /*
 432                  * Complete device or file path.  Exact type is determined by
 433                  * examining the file descriptor afterwards.
 434                  */
 435                 wholedisk = is_whole_disk(arg);
 436                 if (!wholedisk && (stat64(arg, &statbuf) != 0)) {
 437                         (void) fprintf(stderr,
 438                             gettext("cannot open '%s': %s\n"),
 439                             arg, strerror(errno));
 440                         return (NULL);
 441                 }
 442
 443                 (void) strlcpy(path, arg, sizeof (path));
 444         } else {
 445                 /*
 446                  * This may be a short path for a device, or it could be total
 447                  * gibberish.  Check to see if it's a known device in
 448                  * /dev/dsk/.  As part of this check, see if we've been given a
 449                  * an entire disk (minus the slice number).
 450                  */
 451                 if (strncmp(arg, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
 452                         strlcpy(path, arg, sizeof (path));
 453                 else
 454                         snprintf(path, sizeof (path), "%s%s", _PATH_DEV, arg);
 455                 wholedisk = is_whole_disk(path);
 456                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 457                         /*
 458                          * If we got ENOENT, then the user gave us
 459                          * gibberish, so try to direct them with a
 460                          * reasonable error message.  Otherwise,
 461                          * regurgitate strerror() since it's the best we
 462                          * can do.
 463                          */
 464                         if (errno == ENOENT) {
 465                                 (void) fprintf(stderr,
 466                                     gettext("cannot open '%s': no such "
 467                                     "GEOM provider\n"), arg);
 468                                 (void) fprintf(stderr,
 469                                     gettext("must be a full path or "
 470                                     "shorthand device name\n"));
 471                                 return (NULL);
 472                         } else {
 473                                 (void) fprintf(stderr,
 474                                     gettext("cannot open '%s': %s\n"),
 475                                     path, strerror(errno));
 476                                 return (NULL);
 477                         }
 478                 }
 479         }
 480
 481 #ifdef __FreeBSD__
 482         if (S_ISCHR(statbuf.st_mode)) {
 483                 statbuf.st_mode &= ~S_IFCHR;
 484                 statbuf.st_mode |= S_IFBLK;
 485                 wholedisk = B_FALSE;
 486         }
 487 #endif
 488
 489         /*
 490          * Determine whether this is a device or a file.
 491          */
 492         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 493                 type = VDEV_TYPE_DISK;
 494         } else if (S_ISREG(statbuf.st_mode)) {
 495                 type = VDEV_TYPE_FILE;
 496         } else {
 497                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 498                     "GEOM provider or regular file\n"), path);
 499                 return (NULL);
 500         }
 501
 502         /*
 503          * Finally, we have the complete device or file, and we know that it is
 504          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 505          * vdevs have a 'path' element, and devices also have a 'devid' element.
 506          */
 507         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 508         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 509         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 510         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 511         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 512                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 513                     (uint64_t)wholedisk) == 0);
 514
 515 #ifdef have_devid
 516         /*
 517          * For a whole disk, defer getting its devid until after labeling it.
 518          */
 519         if (S_ISBLK(statbuf.st_mode) && !wholedisk) {
 520                 /*
 521                  * Get the devid for the device.
 522                  */
 523                 int fd;
 524                 ddi_devid_t devid;
 525                 char *minor = NULL, *devid_str = NULL;
 526
 527                 if ((fd = open(path, O_RDONLY)) < 0) {
 528                         (void) fprintf(stderr, gettext("cannot open '%s': "
 529                             "%s\n"), path, strerror(errno));
 530                         nvlist_free(vdev);
 531                         return (NULL);
 532                 }
 533
 534                 if (devid_get(fd, &devid) == 0) {
 535                         if (devid_get_minor_name(fd, &minor) == 0 &&
 536                             (devid_str = devid_str_encode(devid, minor)) !=
 537                             NULL) {
 538                                 verify(nvlist_add_string(vdev,
 539                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 540                         }
 541                         if (devid_str != NULL)
 542                                 devid_str_free(devid_str);
 543                         if (minor != NULL)
 544                                 devid_str_free(minor);
 545                         devid_free(devid);
 546                 }
 547
 548                 (void) close(fd);
 549         }
 550 #endif
 551
 552         return (vdev);
 553 }
 554
 555 /*
 556  * Go through and verify the replication level of the pool is consistent.
 557  * Performs the following checks:
 558  *
 559  *      For the new spec, verifies that devices in mirrors and raidz are the
 560  *      same size.
 561  *
 562  *      If the current configuration already has inconsistent replication
 563  *      levels, ignore any other potential problems in the new spec.
 564  *
 565  *      Otherwise, make sure that the current spec (if there is one) and the new
 566  *      spec have consistent replication levels.
 567  */
 568 typedef struct replication_level {
 569         char *zprl_type;
 570         uint64_t zprl_children;
 571         uint64_t zprl_parity;
 572 } replication_level_t;
 573
 574 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 575
 576 /*
 577  * Given a list of toplevel vdevs, return the current replication level.  If
 578  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 579  * an error message will be displayed for each self-inconsistent vdev.
 580  */
 581 static replication_level_t *
 582 get_replication(nvlist_t *nvroot, boolean_t fatal)
 583 {
 584         nvlist_t **top;
 585         uint_t t, toplevels;
 586         nvlist_t **child;
 587         uint_t c, children;
 588         nvlist_t *nv;
 589         char *type;
 590         replication_level_t lastrep, rep, *ret;
 591         boolean_t dontreport;
 592
 593         ret = safe_malloc(sizeof (replication_level_t));
 594
 595         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 596             &top, &toplevels) == 0);
 597
 598         lastrep.zprl_type = NULL;
 599         for (t = 0; t < toplevels; t++) {
 600                 uint64_t is_log = B_FALSE;
 601
 602                 nv = top[t];
 603
 604                 /*
 605                  * For separate logs we ignore the top level vdev replication
 606                  * constraints.
 607                  */
 608                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 609                 if (is_log)
 610                         continue;
 611
 612                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 613                     &type) == 0);
 614                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 615                     &child, &children) != 0) {
 616                         /*
 617                          * This is a 'file' or 'disk' vdev.
 618                          */
 619                         rep.zprl_type = type;
 620                         rep.zprl_children = 1;
 621                         rep.zprl_parity = 0;
 622                 } else {
 623                         uint64_t vdev_size;
 624
 625                         /*
 626                          * This is a mirror or RAID-Z vdev.  Go through and make
 627                          * sure the contents are all the same (files vs. disks),
 628                          * keeping track of the number of elements in the
 629                          * process.
 630                          *
 631                          * We also check that the size of each vdev (if it can
 632                          * be determined) is the same.
 633                          */
 634                         rep.zprl_type = type;
 635                         rep.zprl_children = 0;
 636
 637                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 638                                 verify(nvlist_lookup_uint64(nv,
 639                                     ZPOOL_CONFIG_NPARITY,
 640                                     &rep.zprl_parity) == 0);
 641                                 assert(rep.zprl_parity != 0);
 642                         } else {
 643                                 rep.zprl_parity = 0;
 644                         }
 645
 646                         /*
 647                          * The 'dontreport' variable indicates that we've
 648                          * already reported an error for this spec, so don't
 649                          * bother doing it again.
 650                          */
 651                         type = NULL;
 652                         dontreport = 0;
 653                         vdev_size = -1ULL;
 654                         for (c = 0; c < children; c++) {
 655                                 nvlist_t *cnv = child[c];
 656                                 char *path;
 657                                 struct stat64 statbuf;
 658                                 uint64_t size = -1ULL;
 659                                 char *childtype;
 660                                 int fd, err;
 661
 662                                 rep.zprl_children++;
 663
 664                                 verify(nvlist_lookup_string(cnv,
 665                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 666
 667                                 /*
 668                                  * If this is a replacing or spare vdev, then
 669                                  * get the real first child of the vdev.
 670                                  */
 671                                 if (strcmp(childtype,
 672                                     VDEV_TYPE_REPLACING) == 0 ||
 673                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 674                                         nvlist_t **rchild;
 675                                         uint_t rchildren;
 676
 677                                         verify(nvlist_lookup_nvlist_array(cnv,
 678                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 679                                             &rchildren) == 0);
 680                                         assert(rchildren == 2);
 681                                         cnv = rchild[0];
 682
 683                                         verify(nvlist_lookup_string(cnv,
 684                                             ZPOOL_CONFIG_TYPE,
 685                                             &childtype) == 0);
 686                                 }
 687
 688                                 verify(nvlist_lookup_string(cnv,
 689                                     ZPOOL_CONFIG_PATH, &path) == 0);
 690
 691                                 /*
 692                                  * If we have a raidz/mirror that combines disks
 693                                  * with files, report it as an error.
 694                                  */
 695                                 if (!dontreport && type != NULL &&
 696                                     strcmp(type, childtype) != 0) {
 697                                         if (ret != NULL)
 698                                                 free(ret);
 699                                         ret = NULL;
 700                                         if (fatal)
 701                                                 vdev_error(gettext(
 702                                                     "mismatched replication "
 703                                                     "level: %s contains both "
 704                                                     "files and devices\n"),
 705                                                     rep.zprl_type);
 706                                         else
 707                                                 return (NULL);
 708                                         dontreport = B_TRUE;
 709                                 }
 710
 711                                 /*
 712                                  * According to stat(2), the value of 'st_size'
 713                                  * is undefined for block devices and character
 714                                  * devices.  But there is no effective way to
 715                                  * determine the real size in userland.
 716                                  *
 717                                  * Instead, we'll take advantage of an
 718                                  * implementation detail of spec_size().  If the
 719                                  * device is currently open, then we (should)
 720                                  * return a valid size.
 721                                  *
 722                                  * If we still don't get a valid size (indicated
 723                                  * by a size of 0 or MAXOFFSET_T), then ignore
 724                                  * this device altogether.
 725                                  */
 726                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 727                                         err = fstat64(fd, &statbuf);
 728                                         (void) close(fd);
 729                                 } else {
 730                                         err = stat64(path, &statbuf);
 731                                 }
 732
 733                                 if (err != 0 ||
 734                                     statbuf.st_size == 0 ||
 735                                     statbuf.st_size == MAXOFFSET_T)
 736                                         continue;
 737
 738                                 size = statbuf.st_size;
 739
 740                                 /*
 741                                  * Also make sure that devices and
 742                                  * slices have a consistent size.  If
 743                                  * they differ by a significant amount
 744                                  * (~16MB) then report an error.
 745                                  */
 746                                 if (!dontreport &&
 747                                     (vdev_size != -1ULL &&
 748                                     (labs(size - vdev_size) >
 749                                     ZPOOL_FUZZ))) {
 750                                         if (ret != NULL)
 751                                                 free(ret);
 752                                         ret = NULL;
 753                                         if (fatal)
 754                                                 vdev_error(gettext(
 755                                                     "%s contains devices of "
 756                                                     "different sizes\n"),
 757                                                     rep.zprl_type);
 758                                         else
 759                                                 return (NULL);
 760                                         dontreport = B_TRUE;
 761                                 }
 762
 763                                 type = childtype;
 764                                 vdev_size = size;
 765                         }
 766                 }
 767
 768                 /*
 769                  * At this point, we have the replication of the last toplevel
 770                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 771                  * different.
 772                  */
 773                 if (lastrep.zprl_type != NULL) {
 774                         if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
 775                                 if (ret != NULL)
 776                                         free(ret);
 777                                 ret = NULL;
 778                                 if (fatal)
 779                                         vdev_error(gettext(
 780                                             "mismatched replication level: "
 781                                             "both %s and %s vdevs are "
 782                                             "present\n"),
 783                                             lastrep.zprl_type, rep.zprl_type);
 784                                 else
 785                                         return (NULL);
 786                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 787                                 if (ret)
 788                                         free(ret);
 789                                 ret = NULL;
 790                                 if (fatal)
 791                                         vdev_error(gettext(
 792                                             "mismatched replication level: "
 793                                             "both %llu and %llu device parity "
 794                                             "%s vdevs are present\n"),
 795                                             lastrep.zprl_parity,
 796                                             rep.zprl_parity,
 797                                             rep.zprl_type);
 798                                 else
 799                                         return (NULL);
 800                         } else if (lastrep.zprl_children != rep.zprl_children) {
 801                                 if (ret)
 802                                         free(ret);
 803                                 ret = NULL;
 804                                 if (fatal)
 805                                         vdev_error(gettext(
 806                                             "mismatched replication level: "
 807                                             "both %llu-way and %llu-way %s "
 808                                             "vdevs are present\n"),
 809                                             lastrep.zprl_children,
 810                                             rep.zprl_children,
 811                                             rep.zprl_type);
 812                                 else
 813                                         return (NULL);
 814                         }
 815                 }
 816                 lastrep = rep;
 817         }
 818
 819         if (ret != NULL)
 820                 *ret = rep;
 821
 822         return (ret);
 823 }
 824
 825 /*
 826  * Check the replication level of the vdev spec against the current pool.  Calls
 827  * get_replication() to make sure the new spec is self-consistent.  If the pool
 828  * has a consistent replication level, then we ignore any errors.  Otherwise,
 829  * report any difference between the two.
 830  */
 831 static int
 832 check_replication(nvlist_t *config, nvlist_t *newroot)
 833 {
 834         nvlist_t **child;
 835         uint_t  children;
 836         replication_level_t *current = NULL, *new;
 837         int ret;
 838
 839         /*
 840          * If we have a current pool configuration, check to see if it's
 841          * self-consistent.  If not, simply return success.
 842          */
 843         if (config != NULL) {
 844                 nvlist_t *nvroot;
 845
 846                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 847                     &nvroot) == 0);
 848                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
 849                         return (0);
 850         }
 851         /*
 852          * for spares there may be no children, and therefore no
 853          * replication level to check
 854          */
 855         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
 856             &child, &children) != 0) || (children == 0)) {
 857                 free(current);
 858                 return (0);
 859         }
 860
 861         /*
 862          * If all we have is logs then there's no replication level to check.
 863          */
 864         if (num_logs(newroot) == children) {
 865                 free(current);
 866                 return (0);
 867         }
 868
 869         /*
 870          * Get the replication level of the new vdev spec, reporting any
 871          * inconsistencies found.
 872          */
 873         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
 874                 free(current);
 875                 return (-1);
 876         }
 877
 878         /*
 879          * Check to see if the new vdev spec matches the replication level of
 880          * the current pool.
 881          */
 882         ret = 0;
 883         if (current != NULL) {
 884                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
 885                         vdev_error(gettext(
 886                             "mismatched replication level: pool uses %s "
 887                             "and new vdev is %s\n"),
 888                             current->zprl_type, new->zprl_type);
 889                         ret = -1;
 890                 } else if (current->zprl_parity != new->zprl_parity) {
 891                         vdev_error(gettext(
 892                             "mismatched replication level: pool uses %llu "
 893                             "device parity and new vdev uses %llu\n"),
 894                             current->zprl_parity, new->zprl_parity);
 895                         ret = -1;
 896                 } else if (current->zprl_children != new->zprl_children) {
 897                         vdev_error(gettext(
 898                             "mismatched replication level: pool uses %llu-way "
 899                             "%s and new vdev uses %llu-way %s\n"),
 900                             current->zprl_children, current->zprl_type,
 901                             new->zprl_children, new->zprl_type);
 902                         ret = -1;
 903                 }
 904         }
 905
 906         free(new);
 907         if (current != NULL)
 908                 free(current);
 909
 910         return (ret);
 911 }
 912
 913 #ifdef sun
 914 /*
 915  * Go through and find any whole disks in the vdev specification, labelling them
 916  * as appropriate.  When constructing the vdev spec, we were unable to open this
 917  * device in order to provide a devid.  Now that we have labelled the disk and
 918  * know that slice 0 is valid, we can construct the devid now.
 919  *
 920  * If the disk was already labeled with an EFI label, we will have gotten the
 921  * devid already (because we were able to open the whole disk).  Otherwise, we
 922  * need to get the devid after we label the disk.
 923  */
 924 static int
 925 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 926 {
 927         nvlist_t **child;
 928         uint_t c, children;
 929         char *type, *path, *diskname;
 930         char buf[MAXPATHLEN];
 931         uint64_t wholedisk;
 932         int fd;
 933         int ret;
 934         ddi_devid_t devid;
 935         char *minor = NULL, *devid_str = NULL;
 936
 937         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
 938
 939         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 940             &child, &children) != 0) {
 941
 942                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
 943                         return (0);
 944
 945                 /*
 946                  * We have a disk device.  Get the path to the device
 947                  * and see if it's a whole disk by appending the backup
 948                  * slice and stat()ing the device.
 949                  */
 950                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
 951                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
 952                     &wholedisk) != 0 || !wholedisk)
 953                         return (0);
 954
 955                 diskname = strrchr(path, '/');
 956                 assert(diskname != NULL);
 957                 diskname++;
 958                 if (zpool_label_disk(g_zfs, zhp, diskname) == -1)
 959                         return (-1);
 960
 961                 /*
 962                  * Fill in the devid, now that we've labeled the disk.
 963                  */
 964                 (void) snprintf(buf, sizeof (buf), "%ss0", path);
 965                 if ((fd = open(buf, O_RDONLY)) < 0) {
 966                         (void) fprintf(stderr,
 967                             gettext("cannot open '%s': %s\n"),
 968                             buf, strerror(errno));
 969                         return (-1);
 970                 }
 971
 972                 if (devid_get(fd, &devid) == 0) {
 973                         if (devid_get_minor_name(fd, &minor) == 0 &&
 974                             (devid_str = devid_str_encode(devid, minor)) !=
 975                             NULL) {
 976                                 verify(nvlist_add_string(nv,
 977                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 978                         }
 979                         if (devid_str != NULL)
 980                                 devid_str_free(devid_str);
 981                         if (minor != NULL)
 982                                 devid_str_free(minor);
 983                         devid_free(devid);
 984                 }
 985
 986                 /*
 987                  * Update the path to refer to the 's0' slice.  The presence of
 988                  * the 'whole_disk' field indicates to the CLI that we should
 989                  * chop off the slice number when displaying the device in
 990                  * future output.
 991                  */
 992                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0);
 993
 994                 (void) close(fd);
 995
 996                 return (0);
 997         }
 998
 999         for (c = 0; c < children; c++)
1000                 if ((ret = make_disks(zhp, child[c])) != 0)
1001                         return (ret);
1002
1003         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1004             &child, &children) == 0)
1005                 for (c = 0; c < children; c++)
1006                         if ((ret = make_disks(zhp, child[c])) != 0)
1007                                 return (ret);
1008
1009         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1010             &child, &children) == 0)
1011                 for (c = 0; c < children; c++)
1012                         if ((ret = make_disks(zhp, child[c])) != 0)
1013                                 return (ret);
1014
1015         return (0);
1016 }
1017 #endif  /* sun */
1018
1019 /*
1020  * Determine if the given path is a hot spare within the given configuration.
1021  */
1022 static boolean_t
1023 is_spare(nvlist_t *config, const char *path)
1024 {
1025         int fd;
1026         pool_state_t state;
1027         char *name = NULL;
1028         nvlist_t *label;
1029         uint64_t guid, spareguid;
1030         nvlist_t *nvroot;
1031         nvlist_t **spares;
1032         uint_t i, nspares;
1033         boolean_t inuse;
1034
1035         if ((fd = open(path, O_RDONLY)) < 0)
1036                 return (B_FALSE);
1037
1038         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
1039             !inuse ||
1040             state != POOL_STATE_SPARE ||
1041             zpool_read_label(fd, &label) != 0) {
1042                 free(name);
1043                 (void) close(fd);
1044                 return (B_FALSE);
1045         }
1046         free(name);
1047         (void) close(fd);
1048
1049         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
1050         nvlist_free(label);
1051
1052         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1053             &nvroot) == 0);
1054         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1055             &spares, &nspares) == 0) {
1056                 for (i = 0; i < nspares; i++) {
1057                         verify(nvlist_lookup_uint64(spares[i],
1058                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
1059                         if (spareguid == guid)
1060                                 return (B_TRUE);
1061                 }
1062         }
1063
1064         return (B_FALSE);
1065 }
1066
1067 /*
1068  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1069  * the majority of this task.
1070  */
1071 static int
1072 check_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1073     boolean_t replacing, boolean_t isspare)
1074 {
1075         nvlist_t **child;
1076         uint_t c, children;
1077         char *type, *path;
1078         int ret;
1079         char buf[MAXPATHLEN];
1080         uint64_t wholedisk;
1081
1082         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1083
1084         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1085             &child, &children) != 0) {
1086
1087                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
1088
1089                 /*
1090                  * As a generic check, we look to see if this is a replace of a
1091                  * hot spare within the same pool.  If so, we allow it
1092                  * regardless of what libdiskmgt or zpool_in_use() says.
1093                  */
1094                 if (replacing) {
1095 #ifdef sun
1096                         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1097                             &wholedisk) == 0 && wholedisk)
1098                                 (void) snprintf(buf, sizeof (buf), "%ss0",
1099                                     path);
1100                         else
1101 #endif
1102                                 (void) strlcpy(buf, path, sizeof (buf));
1103
1104                         if (is_spare(config, buf))
1105                                 return (0);
1106                 }
1107
1108                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1109                         ret = check_device(path, force, isspare);
1110
1111                 if (strcmp(type, VDEV_TYPE_FILE) == 0)
1112                         ret = check_file(path, force, isspare);
1113
1114                 return (ret);
1115         }
1116
1117         for (c = 0; c < children; c++)
1118                 if ((ret = check_in_use(config, child[c], force,
1119                     replacing, B_FALSE)) != 0)
1120                         return (ret);
1121
1122         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1123             &child, &children) == 0)
1124                 for (c = 0; c < children; c++)
1125                         if ((ret = check_in_use(config, child[c], force,
1126                             replacing, B_TRUE)) != 0)
1127                                 return (ret);
1128
1129         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1130             &child, &children) == 0)
1131                 for (c = 0; c < children; c++)
1132                         if ((ret = check_in_use(config, child[c], force,
1133                             replacing, B_FALSE)) != 0)
1134                                 return (ret);
1135
1136         return (0);
1137 }
1138
1139 static const char *
1140 is_grouping(const char *type, int *mindev, int *maxdev)
1141 {
1142         if (strncmp(type, "raidz", 5) == 0) {
1143                 const char *p = type + 5;
1144                 char *end;
1145                 long nparity;
1146
1147                 if (*p == '\0') {
1148                         nparity = 1;
1149                 } else if (*p == '0') {
1150                         return (NULL); /* no zero prefixes allowed */
1151                 } else {
1152                         errno = 0;
1153                         nparity = strtol(p, &end, 10);
1154                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1155                             *end != '\0')
1156                                 return (NULL);
1157                 }
1158
1159                 if (mindev != NULL)
1160                         *mindev = nparity + 1;
1161                 if (maxdev != NULL)
1162                         *maxdev = 255;
1163                 return (VDEV_TYPE_RAIDZ);
1164         }
1165
1166         if (maxdev != NULL)
1167                 *maxdev = INT_MAX;
1168
1169         if (strcmp(type, "mirror") == 0) {
1170                 if (mindev != NULL)
1171                         *mindev = 2;
1172                 return (VDEV_TYPE_MIRROR);
1173         }
1174
1175         if (strcmp(type, "spare") == 0) {
1176                 if (mindev != NULL)
1177                         *mindev = 1;
1178                 return (VDEV_TYPE_SPARE);
1179         }
1180
1181         if (strcmp(type, "log") == 0) {
1182                 if (mindev != NULL)
1183                         *mindev = 1;
1184                 return (VDEV_TYPE_LOG);
1185         }
1186
1187         if (strcmp(type, "cache") == 0) {
1188                 if (mindev != NULL)
1189                         *mindev = 1;
1190                 return (VDEV_TYPE_L2CACHE);
1191         }
1192
1193         return (NULL);
1194 }
1195
1196 /*
1197  * Construct a syntactically valid vdev specification,
1198  * and ensure that all devices and files exist and can be opened.
1199  * Note: we don't bother freeing anything in the error paths
1200  * because the program is just going to exit anyway.
1201  */
1202 nvlist_t *
1203 construct_spec(int argc, char **argv)
1204 {
1205         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1206         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1207         const char *type;
1208         uint64_t is_log;
1209         boolean_t seen_logs;
1210
1211         top = NULL;
1212         toplevels = 0;
1213         spares = NULL;
1214         l2cache = NULL;
1215         nspares = 0;
1216         nlogs = 0;
1217         nl2cache = 0;
1218         is_log = B_FALSE;
1219         seen_logs = B_FALSE;
1220
1221         while (argc > 0) {
1222                 nv = NULL;
1223
1224                 /*
1225                  * If it's a mirror or raidz, the subsequent arguments are
1226                  * its leaves -- until we encounter the next mirror or raidz.
1227                  */
1228                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1229                         nvlist_t **child = NULL;
1230                         int c, children = 0;
1231
1232                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1233                                 if (spares != NULL) {
1234                                         (void) fprintf(stderr,
1235                                             gettext("invalid vdev "
1236                                             "specification: 'spare' can be "
1237                                             "specified only once\n"));
1238                                         return (NULL);
1239                                 }
1240                                 is_log = B_FALSE;
1241                         }
1242
1243                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1244                                 if (seen_logs) {
1245                                         (void) fprintf(stderr,
1246                                             gettext("invalid vdev "
1247                                             "specification: 'log' can be "
1248                                             "specified only once\n"));
1249                                         return (NULL);
1250                                 }
1251                                 seen_logs = B_TRUE;
1252                                 is_log = B_TRUE;
1253                                 argc--;
1254                                 argv++;
1255                                 /*
1256                                  * A log is not a real grouping device.
1257                                  * We just set is_log and continue.
1258                                  */
1259                                 continue;
1260                         }
1261
1262                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1263                                 if (l2cache != NULL) {
1264                                         (void) fprintf(stderr,
1265                                             gettext("invalid vdev "
1266                                             "specification: 'cache' can be "
1267                                             "specified only once\n"));
1268                                         return (NULL);
1269                                 }
1270                                 is_log = B_FALSE;
1271                         }
1272
1273                         if (is_log) {
1274                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1275                                         (void) fprintf(stderr,
1276                                             gettext("invalid vdev "
1277                                             "specification: unsupported 'log' "
1278                                             "device: %s\n"), type);
1279                                         return (NULL);
1280                                 }
1281                                 nlogs++;
1282                         }
1283
1284                         for (c = 1; c < argc; c++) {
1285                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1286                                         break;
1287                                 children++;
1288                                 child = realloc(child,
1289                                     children * sizeof (nvlist_t *));
1290                                 if (child == NULL)
1291                                         zpool_no_memory();
1292                                 if ((nv = make_leaf_vdev(argv[c], B_FALSE))
1293                                     == NULL)
1294                                         return (NULL);
1295                                 child[children - 1] = nv;
1296                         }
1297
1298                         if (children < mindev) {
1299                                 (void) fprintf(stderr, gettext("invalid vdev "
1300                                     "specification: %s requires at least %d "
1301                                     "devices\n"), argv[0], mindev);
1302                                 return (NULL);
1303                         }
1304
1305                         if (children > maxdev) {
1306                                 (void) fprintf(stderr, gettext("invalid vdev "
1307                                     "specification: %s supports no more than "
1308                                     "%d devices\n"), argv[0], maxdev);
1309                                 return (NULL);
1310                         }
1311
1312                         argc -= c;
1313                         argv += c;
1314
1315                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1316                                 spares = child;
1317                                 nspares = children;
1318                                 continue;
1319                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1320                                 l2cache = child;
1321                                 nl2cache = children;
1322                                 continue;
1323                         } else {
1324                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1325                                     0) == 0);
1326                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1327                                     type) == 0);
1328                                 verify(nvlist_add_uint64(nv,
1329                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1330                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1331                                         verify(nvlist_add_uint64(nv,
1332                                             ZPOOL_CONFIG_NPARITY,
1333                                             mindev - 1) == 0);
1334                                 }
1335                                 verify(nvlist_add_nvlist_array(nv,
1336                                     ZPOOL_CONFIG_CHILDREN, child,
1337                                     children) == 0);
1338
1339                                 for (c = 0; c < children; c++)
1340                                         nvlist_free(child[c]);
1341                                 free(child);
1342                         }
1343                 } else {
1344                         /*
1345                          * We have a device.  Pass off to make_leaf_vdev() to
1346                          * construct the appropriate nvlist describing the vdev.
1347                          */
1348                         if ((nv = make_leaf_vdev(argv[0], is_log)) == NULL)
1349                                 return (NULL);
1350                         if (is_log)
1351                                 nlogs++;
1352                         argc--;
1353                         argv++;
1354                 }
1355
1356                 toplevels++;
1357                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1358                 if (top == NULL)
1359                         zpool_no_memory();
1360                 top[toplevels - 1] = nv;
1361         }
1362
1363         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1364                 (void) fprintf(stderr, gettext("invalid vdev "
1365                     "specification: at least one toplevel vdev must be "
1366                     "specified\n"));
1367                 return (NULL);
1368         }
1369
1370         if (seen_logs && nlogs == 0) {
1371                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1372                     "log requires at least 1 device\n"));
1373                 return (NULL);
1374         }
1375
1376         /*
1377          * Finally, create nvroot and add all top-level vdevs to it.
1378          */
1379         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1380         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1381             VDEV_TYPE_ROOT) == 0);
1382         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1383             top, toplevels) == 0);
1384         if (nspares != 0)
1385                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1386                     spares, nspares) == 0);
1387         if (nl2cache != 0)
1388                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1389                     l2cache, nl2cache) == 0);
1390
1391         for (t = 0; t < toplevels; t++)
1392                 nvlist_free(top[t]);
1393         for (t = 0; t < nspares; t++)
1394                 nvlist_free(spares[t]);
1395         for (t = 0; t < nl2cache; t++)
1396                 nvlist_free(l2cache[t]);
1397         if (spares)
1398                 free(spares);
1399         if (l2cache)
1400                 free(l2cache);
1401         free(top);
1402
1403         return (nvroot);
1404 }
1405
1406 nvlist_t *
1407 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1408     splitflags_t flags, int argc, char **argv)
1409 {
1410         nvlist_t *newroot = NULL, **child;
1411         uint_t c, children;
1412
1413         if (argc > 0) {
1414                 if ((newroot = construct_spec(argc, argv)) == NULL) {
1415                         (void) fprintf(stderr, gettext("Unable to build a "
1416                             "pool from the specified devices\n"));
1417                         return (NULL);
1418                 }
1419
1420 #ifdef sun
1421                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1422                         nvlist_free(newroot);
1423                         return (NULL);
1424                 }
1425 #endif
1426
1427                 /* avoid any tricks in the spec */
1428                 verify(nvlist_lookup_nvlist_array(newroot,
1429                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1430                 for (c = 0; c < children; c++) {
1431                         char *path;
1432                         const char *type;
1433                         int min, max;
1434
1435                         verify(nvlist_lookup_string(child[c],
1436                             ZPOOL_CONFIG_PATH, &path) == 0);
1437                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1438                                 (void) fprintf(stderr, gettext("Cannot use "
1439                                     "'%s' as a device for splitting\n"), type);
1440                                 nvlist_free(newroot);
1441                                 return (NULL);
1442                         }
1443                 }
1444         }
1445
1446         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1447                 if (newroot != NULL)
1448                         nvlist_free(newroot);
1449                 return (NULL);
1450         }
1451
1452         return (newroot);
1453 }
1454
1455 /*
1456  * Get and validate the contents of the given vdev specification.  This ensures
1457  * that the nvlist returned is well-formed, that all the devices exist, and that
1458  * they are not currently in use by any other known consumer.  The 'poolconfig'
1459  * parameter is the current configuration of the pool when adding devices
1460  * existing pool, and is used to perform additional checks, such as changing the
1461  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1462  * new pool.  The 'force' flag controls whether devices should be forcefully
1463  * added, even if they appear in use.
1464  */
1465 nvlist_t *
1466 make_root_vdev(zpool_handle_t *zhp, int force, int check_rep,
1467     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1468 {
1469         nvlist_t *newroot;
1470         nvlist_t *poolconfig = NULL;
1471         is_force = force;
1472
1473         /*
1474          * Construct the vdev specification.  If this is successful, we know
1475          * that we have a valid specification, and that all devices can be
1476          * opened.
1477          */
1478         if ((newroot = construct_spec(argc, argv)) == NULL)
1479                 return (NULL);
1480
1481         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
1482                 return (NULL);
1483
1484         /*
1485          * Validate each device to make sure that its not shared with another
1486          * subsystem.  We do this even if 'force' is set, because there are some
1487          * uses (such as a dedicated dump device) that even '-f' cannot
1488          * override.
1489          */
1490         if (check_in_use(poolconfig, newroot, force, replacing, B_FALSE) != 0) {
1491                 nvlist_free(newroot);
1492                 return (NULL);
1493         }
1494
1495         /*
1496          * Check the replication level of the given vdevs and report any errors
1497          * found.  We include the existing pool spec, if any, as we need to
1498          * catch changes against the existing replication level.
1499          */
1500         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1501                 nvlist_free(newroot);
1502                 return (NULL);
1503         }
1504
1505 #ifdef sun
1506         /*
1507          * Run through the vdev specification and label any whole disks found.
1508          */
1509         if (!dryrun && make_disks(zhp, newroot) != 0) {
1510                 nvlist_free(newroot);
1511                 return (NULL);
1512         }
1513 #endif
1514
1515         return (newroot);
1516 }