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         /*
 516          * For a whole disk, defer getting its devid until after labeling it.
 517          */
 518         if (S_ISBLK(statbuf.st_mode) && !wholedisk) {
 519                 /*
 520                  * Get the devid for the device.
 521                  */
 522                 int fd;
 523                 ddi_devid_t devid;
 524                 char *minor = NULL, *devid_str = NULL;
 525
 526                 if ((fd = open(path, O_RDONLY)) < 0) {
 527                         (void) fprintf(stderr, gettext("cannot open '%s': "
 528                             "%s\n"), path, strerror(errno));
 529                         nvlist_free(vdev);
 530                         return (NULL);
 531                 }
 532
 533                 if (devid_get(fd, &devid) == 0) {
 534                         if (devid_get_minor_name(fd, &minor) == 0 &&
 535                             (devid_str = devid_str_encode(devid, minor)) !=
 536                             NULL) {
 537                                 verify(nvlist_add_string(vdev,
 538                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 539                         }
 540                         if (devid_str != NULL)
 541                                 devid_str_free(devid_str);
 542                         if (minor != NULL)
 543                                 devid_str_free(minor);
 544                         devid_free(devid);
 545                 }
 546
 547                 (void) close(fd);
 548         }
 549
 550         return (vdev);
 551 }
 552
 553 /*
 554  * Go through and verify the replication level of the pool is consistent.
 555  * Performs the following checks:
 556  *
 557  *      For the new spec, verifies that devices in mirrors and raidz are the
 558  *      same size.
 559  *
 560  *      If the current configuration already has inconsistent replication
 561  *      levels, ignore any other potential problems in the new spec.
 562  *
 563  *      Otherwise, make sure that the current spec (if there is one) and the new
 564  *      spec have consistent replication levels.
 565  */
 566 typedef struct replication_level {
 567         char *zprl_type;
 568         uint64_t zprl_children;
 569         uint64_t zprl_parity;
 570 } replication_level_t;
 571
 572 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 573
 574 /*
 575  * Given a list of toplevel vdevs, return the current replication level.  If
 576  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 577  * an error message will be displayed for each self-inconsistent vdev.
 578  */
 579 static replication_level_t *
 580 get_replication(nvlist_t *nvroot, boolean_t fatal)
 581 {
 582         nvlist_t **top;
 583         uint_t t, toplevels;
 584         nvlist_t **child;
 585         uint_t c, children;
 586         nvlist_t *nv;
 587         char *type;
 588         replication_level_t lastrep, rep, *ret;
 589         boolean_t dontreport;
 590
 591         ret = safe_malloc(sizeof (replication_level_t));
 592
 593         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 594             &top, &toplevels) == 0);
 595
 596         lastrep.zprl_type = NULL;
 597         for (t = 0; t < toplevels; t++) {
 598                 uint64_t is_log = B_FALSE;
 599
 600                 nv = top[t];
 601
 602                 /*
 603                  * For separate logs we ignore the top level vdev replication
 604                  * constraints.
 605                  */
 606                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 607                 if (is_log)
 608                         continue;
 609
 610                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 611                     &type) == 0);
 612                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 613                     &child, &children) != 0) {
 614                         /*
 615                          * This is a 'file' or 'disk' vdev.
 616                          */
 617                         rep.zprl_type = type;
 618                         rep.zprl_children = 1;
 619                         rep.zprl_parity = 0;
 620                 } else {
 621                         uint64_t vdev_size;
 622
 623                         /*
 624                          * This is a mirror or RAID-Z vdev.  Go through and make
 625                          * sure the contents are all the same (files vs. disks),
 626                          * keeping track of the number of elements in the
 627                          * process.
 628                          *
 629                          * We also check that the size of each vdev (if it can
 630                          * be determined) is the same.
 631                          */
 632                         rep.zprl_type = type;
 633                         rep.zprl_children = 0;
 634
 635                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 636                                 verify(nvlist_lookup_uint64(nv,
 637                                     ZPOOL_CONFIG_NPARITY,
 638                                     &rep.zprl_parity) == 0);
 639                                 assert(rep.zprl_parity != 0);
 640                         } else {
 641                                 rep.zprl_parity = 0;
 642                         }
 643
 644                         /*
 645                          * The 'dontreport' variable indicates that we've
 646                          * already reported an error for this spec, so don't
 647                          * bother doing it again.
 648                          */
 649                         type = NULL;
 650                         dontreport = 0;
 651                         vdev_size = -1ULL;
 652                         for (c = 0; c < children; c++) {
 653                                 nvlist_t *cnv = child[c];
 654                                 char *path;
 655                                 struct stat64 statbuf;
 656                                 uint64_t size = -1ULL;
 657                                 char *childtype;
 658                                 int fd, err;
 659
 660                                 rep.zprl_children++;
 661
 662                                 verify(nvlist_lookup_string(cnv,
 663                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 664
 665                                 /*
 666                                  * If this is a replacing or spare vdev, then
 667                                  * get the real first child of the vdev.
 668                                  */
 669                                 if (strcmp(childtype,
 670                                     VDEV_TYPE_REPLACING) == 0 ||
 671                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 672                                         nvlist_t **rchild;
 673                                         uint_t rchildren;
 674
 675                                         verify(nvlist_lookup_nvlist_array(cnv,
 676                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 677                                             &rchildren) == 0);
 678                                         assert(rchildren == 2);
 679                                         cnv = rchild[0];
 680
 681                                         verify(nvlist_lookup_string(cnv,
 682                                             ZPOOL_CONFIG_TYPE,
 683                                             &childtype) == 0);
 684                                 }
 685
 686                                 verify(nvlist_lookup_string(cnv,
 687                                     ZPOOL_CONFIG_PATH, &path) == 0);
 688
 689                                 /*
 690                                  * If we have a raidz/mirror that combines disks
 691                                  * with files, report it as an error.
 692                                  */
 693                                 if (!dontreport && type != NULL &&
 694                                     strcmp(type, childtype) != 0) {
 695                                         if (ret != NULL)
 696                                                 free(ret);
 697                                         ret = NULL;
 698                                         if (fatal)
 699                                                 vdev_error(gettext(
 700                                                     "mismatched replication "
 701                                                     "level: %s contains both "
 702                                                     "files and devices\n"),
 703                                                     rep.zprl_type);
 704                                         else
 705                                                 return (NULL);
 706                                         dontreport = B_TRUE;
 707                                 }
 708
 709                                 /*
 710                                  * According to stat(2), the value of 'st_size'
 711                                  * is undefined for block devices and character
 712                                  * devices.  But there is no effective way to
 713                                  * determine the real size in userland.
 714                                  *
 715                                  * Instead, we'll take advantage of an
 716                                  * implementation detail of spec_size().  If the
 717                                  * device is currently open, then we (should)
 718                                  * return a valid size.
 719                                  *
 720                                  * If we still don't get a valid size (indicated
 721                                  * by a size of 0 or MAXOFFSET_T), then ignore
 722                                  * this device altogether.
 723                                  */
 724                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 725                                         err = fstat64(fd, &statbuf);
 726                                         (void) close(fd);
 727                                 } else {
 728                                         err = stat64(path, &statbuf);
 729                                 }
 730
 731                                 if (err != 0 ||
 732                                     statbuf.st_size == 0 ||
 733                                     statbuf.st_size == MAXOFFSET_T)
 734                                         continue;
 735
 736                                 size = statbuf.st_size;
 737
 738                                 /*
 739                                  * Also make sure that devices and
 740                                  * slices have a consistent size.  If
 741                                  * they differ by a significant amount
 742                                  * (~16MB) then report an error.
 743                                  */
 744                                 if (!dontreport &&
 745                                     (vdev_size != -1ULL &&
 746                                     (labs(size - vdev_size) >
 747                                     ZPOOL_FUZZ))) {
 748                                         if (ret != NULL)
 749                                                 free(ret);
 750                                         ret = NULL;
 751                                         if (fatal)
 752                                                 vdev_error(gettext(
 753                                                     "%s contains devices of "
 754                                                     "different sizes\n"),
 755                                                     rep.zprl_type);
 756                                         else
 757                                                 return (NULL);
 758                                         dontreport = B_TRUE;
 759                                 }
 760
 761                                 type = childtype;
 762                                 vdev_size = size;
 763                         }
 764                 }
 765
 766                 /*
 767                  * At this point, we have the replication of the last toplevel
 768                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 769                  * different.
 770                  */
 771                 if (lastrep.zprl_type != NULL) {
 772                         if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
 773                                 if (ret != NULL)
 774                                         free(ret);
 775                                 ret = NULL;
 776                                 if (fatal)
 777                                         vdev_error(gettext(
 778                                             "mismatched replication level: "
 779                                             "both %s and %s vdevs are "
 780                                             "present\n"),
 781                                             lastrep.zprl_type, rep.zprl_type);
 782                                 else
 783                                         return (NULL);
 784                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 785                                 if (ret)
 786                                         free(ret);
 787                                 ret = NULL;
 788                                 if (fatal)
 789                                         vdev_error(gettext(
 790                                             "mismatched replication level: "
 791                                             "both %llu and %llu device parity "
 792                                             "%s vdevs are present\n"),
 793                                             lastrep.zprl_parity,
 794                                             rep.zprl_parity,
 795                                             rep.zprl_type);
 796                                 else
 797                                         return (NULL);
 798                         } else if (lastrep.zprl_children != rep.zprl_children) {
 799                                 if (ret)
 800                                         free(ret);
 801                                 ret = NULL;
 802                                 if (fatal)
 803                                         vdev_error(gettext(
 804                                             "mismatched replication level: "
 805                                             "both %llu-way and %llu-way %s "
 806                                             "vdevs are present\n"),
 807                                             lastrep.zprl_children,
 808                                             rep.zprl_children,
 809                                             rep.zprl_type);
 810                                 else
 811                                         return (NULL);
 812                         }
 813                 }
 814                 lastrep = rep;
 815         }
 816
 817         if (ret != NULL)
 818                 *ret = rep;
 819
 820         return (ret);
 821 }
 822
 823 /*
 824  * Check the replication level of the vdev spec against the current pool.  Calls
 825  * get_replication() to make sure the new spec is self-consistent.  If the pool
 826  * has a consistent replication level, then we ignore any errors.  Otherwise,
 827  * report any difference between the two.
 828  */
 829 static int
 830 check_replication(nvlist_t *config, nvlist_t *newroot)
 831 {
 832         nvlist_t **child;
 833         uint_t  children;
 834         replication_level_t *current = NULL, *new;
 835         int ret;
 836
 837         /*
 838          * If we have a current pool configuration, check to see if it's
 839          * self-consistent.  If not, simply return success.
 840          */
 841         if (config != NULL) {
 842                 nvlist_t *nvroot;
 843
 844                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 845                     &nvroot) == 0);
 846                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
 847                         return (0);
 848         }
 849         /*
 850          * for spares there may be no children, and therefore no
 851          * replication level to check
 852          */
 853         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
 854             &child, &children) != 0) || (children == 0)) {
 855                 free(current);
 856                 return (0);
 857         }
 858
 859         /*
 860          * If all we have is logs then there's no replication level to check.
 861          */
 862         if (num_logs(newroot) == children) {
 863                 free(current);
 864                 return (0);
 865         }
 866
 867         /*
 868          * Get the replication level of the new vdev spec, reporting any
 869          * inconsistencies found.
 870          */
 871         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
 872                 free(current);
 873                 return (-1);
 874         }
 875
 876         /*
 877          * Check to see if the new vdev spec matches the replication level of
 878          * the current pool.
 879          */
 880         ret = 0;
 881         if (current != NULL) {
 882                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
 883                         vdev_error(gettext(
 884                             "mismatched replication level: pool uses %s "
 885                             "and new vdev is %s\n"),
 886                             current->zprl_type, new->zprl_type);
 887                         ret = -1;
 888                 } else if (current->zprl_parity != new->zprl_parity) {
 889                         vdev_error(gettext(
 890                             "mismatched replication level: pool uses %llu "
 891                             "device parity and new vdev uses %llu\n"),
 892                             current->zprl_parity, new->zprl_parity);
 893                         ret = -1;
 894                 } else if (current->zprl_children != new->zprl_children) {
 895                         vdev_error(gettext(
 896                             "mismatched replication level: pool uses %llu-way "
 897                             "%s and new vdev uses %llu-way %s\n"),
 898                             current->zprl_children, current->zprl_type,
 899                             new->zprl_children, new->zprl_type);
 900                         ret = -1;
 901                 }
 902         }
 903
 904         free(new);
 905         if (current != NULL)
 906                 free(current);
 907
 908         return (ret);
 909 }
 910
 911 #ifdef sun
 912 /*
 913  * Go through and find any whole disks in the vdev specification, labelling them
 914  * as appropriate.  When constructing the vdev spec, we were unable to open this
 915  * device in order to provide a devid.  Now that we have labelled the disk and
 916  * know that slice 0 is valid, we can construct the devid now.
 917  *
 918  * If the disk was already labeled with an EFI label, we will have gotten the
 919  * devid already (because we were able to open the whole disk).  Otherwise, we
 920  * need to get the devid after we label the disk.
 921  */
 922 static int
 923 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 924 {
 925         nvlist_t **child;
 926         uint_t c, children;
 927         char *type, *path, *diskname;
 928         char buf[MAXPATHLEN];
 929         uint64_t wholedisk;
 930         int fd;
 931         int ret;
 932         ddi_devid_t devid;
 933         char *minor = NULL, *devid_str = NULL;
 934
 935         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
 936
 937         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 938             &child, &children) != 0) {
 939
 940                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
 941                         return (0);
 942
 943                 /*
 944                  * We have a disk device.  Get the path to the device
 945                  * and see if it's a whole disk by appending the backup
 946                  * slice and stat()ing the device.
 947                  */
 948                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
 949                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
 950                     &wholedisk) != 0 || !wholedisk)
 951                         return (0);
 952
 953                 diskname = strrchr(path, '/');
 954                 assert(diskname != NULL);
 955                 diskname++;
 956                 if (zpool_label_disk(g_zfs, zhp, diskname) == -1)
 957                         return (-1);
 958
 959                 /*
 960                  * Fill in the devid, now that we've labeled the disk.
 961                  */
 962                 (void) snprintf(buf, sizeof (buf), "%ss0", path);
 963                 if ((fd = open(buf, O_RDONLY)) < 0) {
 964                         (void) fprintf(stderr,
 965                             gettext("cannot open '%s': %s\n"),
 966                             buf, strerror(errno));
 967                         return (-1);
 968                 }
 969
 970                 if (devid_get(fd, &devid) == 0) {
 971                         if (devid_get_minor_name(fd, &minor) == 0 &&
 972                             (devid_str = devid_str_encode(devid, minor)) !=
 973                             NULL) {
 974                                 verify(nvlist_add_string(nv,
 975                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 976                         }
 977                         if (devid_str != NULL)
 978                                 devid_str_free(devid_str);
 979                         if (minor != NULL)
 980                                 devid_str_free(minor);
 981                         devid_free(devid);
 982                 }
 983
 984                 /*
 985                  * Update the path to refer to the 's0' slice.  The presence of
 986                  * the 'whole_disk' field indicates to the CLI that we should
 987                  * chop off the slice number when displaying the device in
 988                  * future output.
 989                  */
 990                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0);
 991
 992                 (void) close(fd);
 993
 994                 return (0);
 995         }
 996
 997         for (c = 0; c < children; c++)
 998                 if ((ret = make_disks(zhp, child[c])) != 0)
 999                         return (ret);
1000
1001         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1002             &child, &children) == 0)
1003                 for (c = 0; c < children; c++)
1004                         if ((ret = make_disks(zhp, child[c])) != 0)
1005                                 return (ret);
1006
1007         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1008             &child, &children) == 0)
1009                 for (c = 0; c < children; c++)
1010                         if ((ret = make_disks(zhp, child[c])) != 0)
1011                                 return (ret);
1012
1013         return (0);
1014 }
1015 #endif  /* sun */
1016
1017 /*
1018  * Determine if the given path is a hot spare within the given configuration.
1019  */
1020 static boolean_t
1021 is_spare(nvlist_t *config, const char *path)
1022 {
1023         int fd;
1024         pool_state_t state;
1025         char *name = NULL;
1026         nvlist_t *label;
1027         uint64_t guid, spareguid;
1028         nvlist_t *nvroot;
1029         nvlist_t **spares;
1030         uint_t i, nspares;
1031         boolean_t inuse;
1032
1033         if ((fd = open(path, O_RDONLY)) < 0)
1034                 return (B_FALSE);
1035
1036         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
1037             !inuse ||
1038             state != POOL_STATE_SPARE ||
1039             zpool_read_label(fd, &label) != 0) {
1040                 free(name);
1041                 (void) close(fd);
1042                 return (B_FALSE);
1043         }
1044         free(name);
1045         (void) close(fd);
1046
1047         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
1048         nvlist_free(label);
1049
1050         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1051             &nvroot) == 0);
1052         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1053             &spares, &nspares) == 0) {
1054                 for (i = 0; i < nspares; i++) {
1055                         verify(nvlist_lookup_uint64(spares[i],
1056                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
1057                         if (spareguid == guid)
1058                                 return (B_TRUE);
1059                 }
1060         }
1061
1062         return (B_FALSE);
1063 }
1064
1065 /*
1066  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1067  * the majority of this task.
1068  */
1069 static int
1070 check_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1071     boolean_t replacing, boolean_t isspare)
1072 {
1073         nvlist_t **child;
1074         uint_t c, children;
1075         char *type, *path;
1076         int ret;
1077         char buf[MAXPATHLEN];
1078         uint64_t wholedisk;
1079
1080         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1081
1082         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1083             &child, &children) != 0) {
1084
1085                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
1086
1087                 /*
1088                  * As a generic check, we look to see if this is a replace of a
1089                  * hot spare within the same pool.  If so, we allow it
1090                  * regardless of what libdiskmgt or zpool_in_use() says.
1091                  */
1092                 if (replacing) {
1093 #ifdef sun
1094                         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1095                             &wholedisk) == 0 && wholedisk)
1096                                 (void) snprintf(buf, sizeof (buf), "%ss0",
1097                                     path);
1098                         else
1099 #endif
1100                                 (void) strlcpy(buf, path, sizeof (buf));
1101
1102                         if (is_spare(config, buf))
1103                                 return (0);
1104                 }
1105
1106                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1107                         ret = check_device(path, force, isspare);
1108
1109                 if (strcmp(type, VDEV_TYPE_FILE) == 0)
1110                         ret = check_file(path, force, isspare);
1111
1112                 return (ret);
1113         }
1114
1115         for (c = 0; c < children; c++)
1116                 if ((ret = check_in_use(config, child[c], force,
1117                     replacing, B_FALSE)) != 0)
1118                         return (ret);
1119
1120         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1121             &child, &children) == 0)
1122                 for (c = 0; c < children; c++)
1123                         if ((ret = check_in_use(config, child[c], force,
1124                             replacing, B_TRUE)) != 0)
1125                                 return (ret);
1126
1127         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1128             &child, &children) == 0)
1129                 for (c = 0; c < children; c++)
1130                         if ((ret = check_in_use(config, child[c], force,
1131                             replacing, B_FALSE)) != 0)
1132                                 return (ret);
1133
1134         return (0);
1135 }
1136
1137 static const char *
1138 is_grouping(const char *type, int *mindev, int *maxdev)
1139 {
1140         if (strncmp(type, "raidz", 5) == 0) {
1141                 const char *p = type + 5;
1142                 char *end;
1143                 long nparity;
1144
1145                 if (*p == '\0') {
1146                         nparity = 1;
1147                 } else if (*p == '0') {
1148                         return (NULL); /* no zero prefixes allowed */
1149                 } else {
1150                         errno = 0;
1151                         nparity = strtol(p, &end, 10);
1152                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1153                             *end != '\0')
1154                                 return (NULL);
1155                 }
1156
1157                 if (mindev != NULL)
1158                         *mindev = nparity + 1;
1159                 if (maxdev != NULL)
1160                         *maxdev = 255;
1161                 return (VDEV_TYPE_RAIDZ);
1162         }
1163
1164         if (maxdev != NULL)
1165                 *maxdev = INT_MAX;
1166
1167         if (strcmp(type, "mirror") == 0) {
1168                 if (mindev != NULL)
1169                         *mindev = 2;
1170                 return (VDEV_TYPE_MIRROR);
1171         }
1172
1173         if (strcmp(type, "spare") == 0) {
1174                 if (mindev != NULL)
1175                         *mindev = 1;
1176                 return (VDEV_TYPE_SPARE);
1177         }
1178
1179         if (strcmp(type, "log") == 0) {
1180                 if (mindev != NULL)
1181                         *mindev = 1;
1182                 return (VDEV_TYPE_LOG);
1183         }
1184
1185         if (strcmp(type, "cache") == 0) {
1186                 if (mindev != NULL)
1187                         *mindev = 1;
1188                 return (VDEV_TYPE_L2CACHE);
1189         }
1190
1191         return (NULL);
1192 }
1193
1194 /*
1195  * Construct a syntactically valid vdev specification,
1196  * and ensure that all devices and files exist and can be opened.
1197  * Note: we don't bother freeing anything in the error paths
1198  * because the program is just going to exit anyway.
1199  */
1200 nvlist_t *
1201 construct_spec(int argc, char **argv)
1202 {
1203         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1204         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1205         const char *type;
1206         uint64_t is_log;
1207         boolean_t seen_logs;
1208
1209         top = NULL;
1210         toplevels = 0;
1211         spares = NULL;
1212         l2cache = NULL;
1213         nspares = 0;
1214         nlogs = 0;
1215         nl2cache = 0;
1216         is_log = B_FALSE;
1217         seen_logs = B_FALSE;
1218
1219         while (argc > 0) {
1220                 nv = NULL;
1221
1222                 /*
1223                  * If it's a mirror or raidz, the subsequent arguments are
1224                  * its leaves -- until we encounter the next mirror or raidz.
1225                  */
1226                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1227                         nvlist_t **child = NULL;
1228                         int c, children = 0;
1229
1230                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1231                                 if (spares != NULL) {
1232                                         (void) fprintf(stderr,
1233                                             gettext("invalid vdev "
1234                                             "specification: 'spare' can be "
1235                                             "specified only once\n"));
1236                                         return (NULL);
1237                                 }
1238                                 is_log = B_FALSE;
1239                         }
1240
1241                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1242                                 if (seen_logs) {
1243                                         (void) fprintf(stderr,
1244                                             gettext("invalid vdev "
1245                                             "specification: 'log' can be "
1246                                             "specified only once\n"));
1247                                         return (NULL);
1248                                 }
1249                                 seen_logs = B_TRUE;
1250                                 is_log = B_TRUE;
1251                                 argc--;
1252                                 argv++;
1253                                 /*
1254                                  * A log is not a real grouping device.
1255                                  * We just set is_log and continue.
1256                                  */
1257                                 continue;
1258                         }
1259
1260                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1261                                 if (l2cache != NULL) {
1262                                         (void) fprintf(stderr,
1263                                             gettext("invalid vdev "
1264                                             "specification: 'cache' can be "
1265                                             "specified only once\n"));
1266                                         return (NULL);
1267                                 }
1268                                 is_log = B_FALSE;
1269                         }
1270
1271                         if (is_log) {
1272                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1273                                         (void) fprintf(stderr,
1274                                             gettext("invalid vdev "
1275                                             "specification: unsupported 'log' "
1276                                             "device: %s\n"), type);
1277                                         return (NULL);
1278                                 }
1279                                 nlogs++;
1280                         }
1281
1282                         for (c = 1; c < argc; c++) {
1283                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1284                                         break;
1285                                 children++;
1286                                 child = realloc(child,
1287                                     children * sizeof (nvlist_t *));
1288                                 if (child == NULL)
1289                                         zpool_no_memory();
1290                                 if ((nv = make_leaf_vdev(argv[c], B_FALSE))
1291                                     == NULL)
1292                                         return (NULL);
1293                                 child[children - 1] = nv;
1294                         }
1295
1296                         if (children < mindev) {
1297                                 (void) fprintf(stderr, gettext("invalid vdev "
1298                                     "specification: %s requires at least %d "
1299                                     "devices\n"), argv[0], mindev);
1300                                 return (NULL);
1301                         }
1302
1303                         if (children > maxdev) {
1304                                 (void) fprintf(stderr, gettext("invalid vdev "
1305                                     "specification: %s supports no more than "
1306                                     "%d devices\n"), argv[0], maxdev);
1307                                 return (NULL);
1308                         }
1309
1310                         argc -= c;
1311                         argv += c;
1312
1313                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1314                                 spares = child;
1315                                 nspares = children;
1316                                 continue;
1317                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1318                                 l2cache = child;
1319                                 nl2cache = children;
1320                                 continue;
1321                         } else {
1322                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1323                                     0) == 0);
1324                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1325                                     type) == 0);
1326                                 verify(nvlist_add_uint64(nv,
1327                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1328                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1329                                         verify(nvlist_add_uint64(nv,
1330                                             ZPOOL_CONFIG_NPARITY,
1331                                             mindev - 1) == 0);
1332                                 }
1333                                 verify(nvlist_add_nvlist_array(nv,
1334                                     ZPOOL_CONFIG_CHILDREN, child,
1335                                     children) == 0);
1336
1337                                 for (c = 0; c < children; c++)
1338                                         nvlist_free(child[c]);
1339                                 free(child);
1340                         }
1341                 } else {
1342                         /*
1343                          * We have a device.  Pass off to make_leaf_vdev() to
1344                          * construct the appropriate nvlist describing the vdev.
1345                          */
1346                         if ((nv = make_leaf_vdev(argv[0], is_log)) == NULL)
1347                                 return (NULL);
1348                         if (is_log)
1349                                 nlogs++;
1350                         argc--;
1351                         argv++;
1352                 }
1353
1354                 toplevels++;
1355                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1356                 if (top == NULL)
1357                         zpool_no_memory();
1358                 top[toplevels - 1] = nv;
1359         }
1360
1361         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1362                 (void) fprintf(stderr, gettext("invalid vdev "
1363                     "specification: at least one toplevel vdev must be "
1364                     "specified\n"));
1365                 return (NULL);
1366         }
1367
1368         if (seen_logs && nlogs == 0) {
1369                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1370                     "log requires at least 1 device\n"));
1371                 return (NULL);
1372         }
1373
1374         /*
1375          * Finally, create nvroot and add all top-level vdevs to it.
1376          */
1377         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1378         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1379             VDEV_TYPE_ROOT) == 0);
1380         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1381             top, toplevels) == 0);
1382         if (nspares != 0)
1383                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1384                     spares, nspares) == 0);
1385         if (nl2cache != 0)
1386                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1387                     l2cache, nl2cache) == 0);
1388
1389         for (t = 0; t < toplevels; t++)
1390                 nvlist_free(top[t]);
1391         for (t = 0; t < nspares; t++)
1392                 nvlist_free(spares[t]);
1393         for (t = 0; t < nl2cache; t++)
1394                 nvlist_free(l2cache[t]);
1395         if (spares)
1396                 free(spares);
1397         if (l2cache)
1398                 free(l2cache);
1399         free(top);
1400
1401         return (nvroot);
1402 }
1403
1404 nvlist_t *
1405 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1406     splitflags_t flags, int argc, char **argv)
1407 {
1408         nvlist_t *newroot = NULL, **child;
1409         uint_t c, children;
1410
1411         if (argc > 0) {
1412                 if ((newroot = construct_spec(argc, argv)) == NULL) {
1413                         (void) fprintf(stderr, gettext("Unable to build a "
1414                             "pool from the specified devices\n"));
1415                         return (NULL);
1416                 }
1417
1418 #ifdef sun
1419                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1420                         nvlist_free(newroot);
1421                         return (NULL);
1422                 }
1423 #endif
1424
1425                 /* avoid any tricks in the spec */
1426                 verify(nvlist_lookup_nvlist_array(newroot,
1427                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1428                 for (c = 0; c < children; c++) {
1429                         char *path;
1430                         const char *type;
1431                         int min, max;
1432
1433                         verify(nvlist_lookup_string(child[c],
1434                             ZPOOL_CONFIG_PATH, &path) == 0);
1435                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1436                                 (void) fprintf(stderr, gettext("Cannot use "
1437                                     "'%s' as a device for splitting\n"), type);
1438                                 nvlist_free(newroot);
1439                                 return (NULL);
1440                         }
1441                 }
1442         }
1443
1444         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1445                 if (newroot != NULL)
1446                         nvlist_free(newroot);
1447                 return (NULL);
1448         }
1449
1450         return (newroot);
1451 }
1452
1453 /*
1454  * Get and validate the contents of the given vdev specification.  This ensures
1455  * that the nvlist returned is well-formed, that all the devices exist, and that
1456  * they are not currently in use by any other known consumer.  The 'poolconfig'
1457  * parameter is the current configuration of the pool when adding devices
1458  * existing pool, and is used to perform additional checks, such as changing the
1459  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1460  * new pool.  The 'force' flag controls whether devices should be forcefully
1461  * added, even if they appear in use.
1462  */
1463 nvlist_t *
1464 make_root_vdev(zpool_handle_t *zhp, int force, int check_rep,
1465     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1466 {
1467         nvlist_t *newroot;
1468         nvlist_t *poolconfig = NULL;
1469         is_force = force;
1470
1471         /*
1472          * Construct the vdev specification.  If this is successful, we know
1473          * that we have a valid specification, and that all devices can be
1474          * opened.
1475          */
1476         if ((newroot = construct_spec(argc, argv)) == NULL)
1477                 return (NULL);
1478
1479         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
1480                 return (NULL);
1481
1482         /*
1483          * Validate each device to make sure that its not shared with another
1484          * subsystem.  We do this even if 'force' is set, because there are some
1485          * uses (such as a dedicated dump device) that even '-f' cannot
1486          * override.
1487          */
1488         if (check_in_use(poolconfig, newroot, force, replacing, B_FALSE) != 0) {
1489                 nvlist_free(newroot);
1490                 return (NULL);
1491         }
1492
1493         /*
1494          * Check the replication level of the given vdevs and report any errors
1495          * found.  We include the existing pool spec, if any, as we need to
1496          * catch changes against the existing replication level.
1497          */
1498         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1499                 nvlist_free(newroot);
1500                 return (NULL);
1501         }
1502
1503 #ifdef sun
1504         /*
1505          * Run through the vdev specification and label any whole disks found.
1506          */
1507         if (!dryrun && make_disks(zhp, newroot) != 0) {
1508                 nvlist_free(newroot);
1509                 return (NULL);
1510         }
1511 #endif
1512
1513         return (newroot);
1514 }