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  * Copyright (c) 2013 by Delphix. All rights reserved.
  25  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>.
  26  */
  27
  28 /*
  29  * Functions to convert between a list of vdevs and an nvlist representing the
  30  * configuration.  Each entry in the list can be one of:
  31  *
  32  *      Device vdevs
  33  *              disk=(path=..., devid=...)
  34  *              file=(path=...)
  35  *
  36  *      Group vdevs
  37  *              raidz[1|2]=(...)
  38  *              mirror=(...)
  39  *
  40  *      Hot spares
  41  *
  42  * While the underlying implementation supports it, group vdevs cannot contain
  43  * other group vdevs.  All userland verification of devices is contained within
  44  * this file.  If successful, the nvlist returned can be passed directly to the
  45  * kernel; we've done as much verification as possible in userland.
  46  *
  47  * Hot spares are a special case, and passed down as an array of disk vdevs, at
  48  * the same level as the root of the vdev tree.
  49  *
  50  * The only function exported by this file is 'make_root_vdev'.  The
  51  * function performs several passes:
  52  *
  53  *      1. Construct the vdev specification.  Performs syntax validation and
  54  *         makes sure each device is valid.
  55  *      2. Check for devices in use.  Using libdiskmgt, makes sure that no
  56  *         devices are also in use.  Some can be overridden using the 'force'
  57  *         flag, others cannot.
  58  *      3. Check for replication errors if the 'force' flag is not specified.
  59  *         validates that the replication level is consistent across the
  60  *         entire pool.
  61  *      4. Call libzfs to label any whole disks with an EFI label.
  62  */
  63
  64 #include <assert.h>
  65 #include <devid.h>
  66 #include <errno.h>
  67 #include <fcntl.h>
  68 #include <libintl.h>
  69 #include <libnvpair.h>
  70 #include <limits.h>
  71 #include <stdio.h>
  72 #include <string.h>
  73 #include <unistd.h>
  74 #include <paths.h>
  75 #include <sys/stat.h>
  76 #include <sys/disk.h>
  77 #include <sys/mntent.h>
  78 #include <libgeom.h>
  79
  80 #include "zpool_util.h"
  81
  82 #define DISK_ROOT       "/dev/dsk"
  83 #define RDISK_ROOT      "/dev/rdsk"
  84 #define BACKUP_SLICE    "s2"
  85
  86 /*
  87  * For any given vdev specification, we can have multiple errors.  The
  88  * vdev_error() function keeps track of whether we have seen an error yet, and
  89  * prints out a header if its the first error we've seen.
  90  */
  91 boolean_t error_seen;
  92 boolean_t is_force;
  93
  94 /*PRINTFLIKE1*/
  95 static void
  96 vdev_error(const char *fmt, ...)
  97 {
  98         va_list ap;
  99
 100         if (!error_seen) {
 101                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
 102                 if (!is_force)
 103                         (void) fprintf(stderr, gettext("use '-f' to override "
 104                             "the following errors:\n"));
 105                 else
 106                         (void) fprintf(stderr, gettext("the following errors "
 107                             "must be manually repaired:\n"));
 108                 error_seen = B_TRUE;
 109         }
 110
 111         va_start(ap, fmt);
 112         (void) vfprintf(stderr, fmt, ap);
 113         va_end(ap);
 114 }
 115
 116 #ifdef illumos
 117 static void
 118 libdiskmgt_error(int error)
 119 {
 120         /*
 121          * ENXIO/ENODEV is a valid error message if the device doesn't live in
 122          * /dev/dsk.  Don't bother printing an error message in this case.
 123          */
 124         if (error == ENXIO || error == ENODEV)
 125                 return;
 126
 127         (void) fprintf(stderr, gettext("warning: device in use checking "
 128             "failed: %s\n"), strerror(error));
 129 }
 130
 131 /*
 132  * Validate a device, passing the bulk of the work off to libdiskmgt.
 133  */
 134 static int
 135 check_slice(const char *path, int force, boolean_t wholedisk, boolean_t isspare)
 136 {
 137         char *msg;
 138         int error = 0;
 139         dm_who_type_t who;
 140
 141         if (force)
 142                 who = DM_WHO_ZPOOL_FORCE;
 143         else if (isspare)
 144                 who = DM_WHO_ZPOOL_SPARE;
 145         else
 146                 who = DM_WHO_ZPOOL;
 147
 148         if (dm_inuse((char *)path, &msg, who, &error) || error) {
 149                 if (error != 0) {
 150                         libdiskmgt_error(error);
 151                         return (0);
 152                 } else {
 153                         vdev_error("%s", msg);
 154                         free(msg);
 155                         return (-1);
 156                 }
 157         }
 158
 159         /*
 160          * If we're given a whole disk, ignore overlapping slices since we're
 161          * about to label it anyway.
 162          */
 163         error = 0;
 164         if (!wholedisk && !force &&
 165             (dm_isoverlapping((char *)path, &msg, &error) || error)) {
 166                 if (error == 0) {
 167                         /* dm_isoverlapping returned -1 */
 168                         vdev_error(gettext("%s overlaps with %s\n"), path, msg);
 169                         free(msg);
 170                         return (-1);
 171                 } else if (error != ENODEV) {
 172                         /* libdiskmgt's devcache only handles physical drives */
 173                         libdiskmgt_error(error);
 174                         return (0);
 175                 }
 176         }
 177
 178         return (0);
 179 }
 180
 181
 182 /*
 183  * Validate a whole disk.  Iterate over all slices on the disk and make sure
 184  * that none is in use by calling check_slice().
 185  */
 186 static int
 187 check_disk(const char *name, dm_descriptor_t disk, int force, int isspare)
 188 {
 189         dm_descriptor_t *drive, *media, *slice;
 190         int err = 0;
 191         int i;
 192         int ret;
 193
 194         /*
 195          * Get the drive associated with this disk.  This should never fail,
 196          * because we already have an alias handle open for the device.
 197          */
 198         if ((drive = dm_get_associated_descriptors(disk, DM_DRIVE,
 199             &err)) == NULL || *drive == NULL) {
 200                 if (err)
 201                         libdiskmgt_error(err);
 202                 return (0);
 203         }
 204
 205         if ((media = dm_get_associated_descriptors(*drive, DM_MEDIA,
 206             &err)) == NULL) {
 207                 dm_free_descriptors(drive);
 208                 if (err)
 209                         libdiskmgt_error(err);
 210                 return (0);
 211         }
 212
 213         dm_free_descriptors(drive);
 214
 215         /*
 216          * It is possible that the user has specified a removable media drive,
 217          * and the media is not present.
 218          */
 219         if (*media == NULL) {
 220                 dm_free_descriptors(media);
 221                 vdev_error(gettext("'%s' has no media in drive\n"), name);
 222                 return (-1);
 223         }
 224
 225         if ((slice = dm_get_associated_descriptors(*media, DM_SLICE,
 226             &err)) == NULL) {
 227                 dm_free_descriptors(media);
 228                 if (err)
 229                         libdiskmgt_error(err);
 230                 return (0);
 231         }
 232
 233         dm_free_descriptors(media);
 234
 235         ret = 0;
 236
 237         /*
 238          * Iterate over all slices and report any errors.  We don't care about
 239          * overlapping slices because we are using the whole disk.
 240          */
 241         for (i = 0; slice[i] != NULL; i++) {
 242                 char *name = dm_get_name(slice[i], &err);
 243
 244                 if (check_slice(name, force, B_TRUE, isspare) != 0)
 245                         ret = -1;
 246
 247                 dm_free_name(name);
 248         }
 249
 250         dm_free_descriptors(slice);
 251         return (ret);
 252 }
 253
 254 /*
 255  * Validate a device.
 256  */
 257 static int
 258 check_device(const char *path, boolean_t force, boolean_t isspare)
 259 {
 260         dm_descriptor_t desc;
 261         int err;
 262         char *dev;
 263
 264         /*
 265          * For whole disks, libdiskmgt does not include the leading dev path.
 266          */
 267         dev = strrchr(path, '/');
 268         assert(dev != NULL);
 269         dev++;
 270         if ((desc = dm_get_descriptor_by_name(DM_ALIAS, dev, &err)) != NULL) {
 271                 err = check_disk(path, desc, force, isspare);
 272                 dm_free_descriptor(desc);
 273                 return (err);
 274         }
 275
 276         return (check_slice(path, force, B_FALSE, isspare));
 277 }
 278 #endif  /* illumos */
 279
 280 /*
 281  * Check that a file is valid.  All we can do in this case is check that it's
 282  * not in use by another pool, and not in use by swap.
 283  */
 284 static int
 285 check_file(const char *file, boolean_t force, boolean_t isspare)
 286 {
 287         char  *name;
 288         int fd;
 289         int ret = 0;
 290         int err;
 291         pool_state_t state;
 292         boolean_t inuse;
 293
 294 #ifdef illumos
 295         if (dm_inuse_swap(file, &err)) {
 296                 if (err)
 297                         libdiskmgt_error(err);
 298                 else
 299                         vdev_error(gettext("%s is currently used by swap. "
 300                             "Please see swap(1M).\n"), file);
 301                 return (-1);
 302         }
 303 #endif
 304
 305         if ((fd = open(file, O_RDONLY)) < 0)
 306                 return (0);
 307
 308         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
 309                 const char *desc;
 310
 311                 switch (state) {
 312                 case POOL_STATE_ACTIVE:
 313                         desc = gettext("active");
 314                         break;
 315
 316                 case POOL_STATE_EXPORTED:
 317                         desc = gettext("exported");
 318                         break;
 319
 320                 case POOL_STATE_POTENTIALLY_ACTIVE:
 321                         desc = gettext("potentially active");
 322                         break;
 323
 324                 default:
 325                         desc = gettext("unknown");
 326                         break;
 327                 }
 328
 329                 /*
 330                  * Allow hot spares to be shared between pools.
 331                  */
 332                 if (state == POOL_STATE_SPARE && isspare)
 333                         return (0);
 334
 335                 if (state == POOL_STATE_ACTIVE ||
 336                     state == POOL_STATE_SPARE || !force) {
 337                         switch (state) {
 338                         case POOL_STATE_SPARE:
 339                                 vdev_error(gettext("%s is reserved as a hot "
 340                                     "spare for pool %s\n"), file, name);
 341                                 break;
 342                         default:
 343                                 vdev_error(gettext("%s is part of %s pool "
 344                                     "'%s'\n"), file, desc, name);
 345                                 break;
 346                         }
 347                         ret = -1;
 348                 }
 349
 350                 free(name);
 351         }
 352
 353         (void) close(fd);
 354         return (ret);
 355 }
 356
 357 static int
 358 check_device(const char *name, boolean_t force, boolean_t isspare)
 359 {
 360         char path[MAXPATHLEN];
 361
 362         if (strncmp(name, _PATH_DEV, sizeof(_PATH_DEV) - 1) != 0)
 363                 snprintf(path, sizeof(path), "%s%s", _PATH_DEV, name);
 364         else
 365                 strlcpy(path, name, sizeof(path));
 366
 367         return (check_file(path, force, isspare));
 368 }
 369
 370 /*
 371  * By "whole disk" we mean an entire physical disk (something we can
 372  * label, toggle the write cache on, etc.) as opposed to the full
 373  * capacity of a pseudo-device such as lofi or did.  We act as if we
 374  * are labeling the disk, which should be a pretty good test of whether
 375  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 376  * it isn't.
 377  */
 378 static boolean_t
 379 is_whole_disk(const char *arg)
 380 {
 381 #ifdef illumos
 382         struct dk_gpt *label;
 383         int     fd;
 384         char    path[MAXPATHLEN];
 385
 386         (void) snprintf(path, sizeof (path), "%s%s%s",
 387             RDISK_ROOT, strrchr(arg, '/'), BACKUP_SLICE);
 388         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0)
 389                 return (B_FALSE);
 390         if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
 391                 (void) close(fd);
 392                 return (B_FALSE);
 393         }
 394         efi_free(label);
 395         (void) close(fd);
 396         return (B_TRUE);
 397 #else
 398         int fd;
 399
 400         fd = g_open(arg, 0);
 401         if (fd >= 0) {
 402                 g_close(fd);
 403                 return (B_TRUE);
 404         }
 405         return (B_FALSE);
 406 #endif
 407 }
 408
 409 /*
 410  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 411  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 412  * leaf vdev are:
 413  *
 414  *      /dev/dsk/xxx    Complete disk path
 415  *      /xxx            Full path to file
 416  *      xxx             Shorthand for /dev/dsk/xxx
 417  */
 418 static nvlist_t *
 419 make_leaf_vdev(const char *arg, uint64_t is_log)
 420 {
 421         char path[MAXPATHLEN];
 422         struct stat64 statbuf;
 423         nvlist_t *vdev = NULL;
 424         char *type = NULL;
 425         boolean_t wholedisk = B_FALSE;
 426
 427         /*
 428          * Determine what type of vdev this is, and put the full path into
 429          * 'path'.  We detect whether this is a device of file afterwards by
 430          * checking the st_mode of the file.
 431          */
 432         if (arg[0] == '/') {
 433                 /*
 434                  * Complete device or file path.  Exact type is determined by
 435                  * examining the file descriptor afterwards.
 436                  */
 437                 wholedisk = is_whole_disk(arg);
 438                 if (!wholedisk && (stat64(arg, &statbuf) != 0)) {
 439                         (void) fprintf(stderr,
 440                             gettext("cannot open '%s': %s\n"),
 441                             arg, strerror(errno));
 442                         return (NULL);
 443                 }
 444
 445                 (void) strlcpy(path, arg, sizeof (path));
 446         } else {
 447                 /*
 448                  * This may be a short path for a device, or it could be total
 449                  * gibberish.  Check to see if it's a known device in
 450                  * /dev/dsk/.  As part of this check, see if we've been given a
 451                  * an entire disk (minus the slice number).
 452                  */
 453                 if (strncmp(arg, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
 454                         strlcpy(path, arg, sizeof (path));
 455                 else
 456                         snprintf(path, sizeof (path), "%s%s", _PATH_DEV, arg);
 457                 wholedisk = is_whole_disk(path);
 458                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 459                         /*
 460                          * If we got ENOENT, then the user gave us
 461                          * gibberish, so try to direct them with a
 462                          * reasonable error message.  Otherwise,
 463                          * regurgitate strerror() since it's the best we
 464                          * can do.
 465                          */
 466                         if (errno == ENOENT) {
 467                                 (void) fprintf(stderr,
 468                                     gettext("cannot open '%s': no such "
 469                                     "GEOM provider\n"), arg);
 470                                 (void) fprintf(stderr,
 471                                     gettext("must be a full path or "
 472                                     "shorthand device name\n"));
 473                                 return (NULL);
 474                         } else {
 475                                 (void) fprintf(stderr,
 476                                     gettext("cannot open '%s': %s\n"),
 477                                     path, strerror(errno));
 478                                 return (NULL);
 479                         }
 480                 }
 481         }
 482
 483 #ifdef __FreeBSD__
 484         if (S_ISCHR(statbuf.st_mode)) {
 485                 statbuf.st_mode &= ~S_IFCHR;
 486                 statbuf.st_mode |= S_IFBLK;
 487                 wholedisk = B_FALSE;
 488         }
 489 #endif
 490
 491         /*
 492          * Determine whether this is a device or a file.
 493          */
 494         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 495                 type = VDEV_TYPE_DISK;
 496         } else if (S_ISREG(statbuf.st_mode)) {
 497                 type = VDEV_TYPE_FILE;
 498         } else {
 499                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 500                     "GEOM provider or regular file\n"), path);
 501                 return (NULL);
 502         }
 503
 504         /*
 505          * Finally, we have the complete device or file, and we know that it is
 506          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 507          * vdevs have a 'path' element, and devices also have a 'devid' element.
 508          */
 509         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 510         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 511         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 512         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 513         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 514                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 515                     (uint64_t)wholedisk) == 0);
 516
 517 #ifdef have_devid
 518         /*
 519          * For a whole disk, defer getting its devid until after labeling it.
 520          */
 521         if (S_ISBLK(statbuf.st_mode) && !wholedisk) {
 522                 /*
 523                  * Get the devid for the device.
 524                  */
 525                 int fd;
 526                 ddi_devid_t devid;
 527                 char *minor = NULL, *devid_str = NULL;
 528
 529                 if ((fd = open(path, O_RDONLY)) < 0) {
 530                         (void) fprintf(stderr, gettext("cannot open '%s': "
 531                             "%s\n"), path, strerror(errno));
 532                         nvlist_free(vdev);
 533                         return (NULL);
 534                 }
 535
 536                 if (devid_get(fd, &devid) == 0) {
 537                         if (devid_get_minor_name(fd, &minor) == 0 &&
 538                             (devid_str = devid_str_encode(devid, minor)) !=
 539                             NULL) {
 540                                 verify(nvlist_add_string(vdev,
 541                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 542                         }
 543                         if (devid_str != NULL)
 544                                 devid_str_free(devid_str);
 545                         if (minor != NULL)
 546                                 devid_str_free(minor);
 547                         devid_free(devid);
 548                 }
 549
 550                 (void) close(fd);
 551         }
 552 #endif
 553
 554         return (vdev);
 555 }
 556
 557 /*
 558  * Go through and verify the replication level of the pool is consistent.
 559  * Performs the following checks:
 560  *
 561  *      For the new spec, verifies that devices in mirrors and raidz are the
 562  *      same size.
 563  *
 564  *      If the current configuration already has inconsistent replication
 565  *      levels, ignore any other potential problems in the new spec.
 566  *
 567  *      Otherwise, make sure that the current spec (if there is one) and the new
 568  *      spec have consistent replication levels.
 569  */
 570 typedef struct replication_level {
 571         char *zprl_type;
 572         uint64_t zprl_children;
 573         uint64_t zprl_parity;
 574 } replication_level_t;
 575
 576 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 577
 578 /*
 579  * Given a list of toplevel vdevs, return the current replication level.  If
 580  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 581  * an error message will be displayed for each self-inconsistent vdev.
 582  */
 583 static replication_level_t *
 584 get_replication(nvlist_t *nvroot, boolean_t fatal)
 585 {
 586         nvlist_t **top;
 587         uint_t t, toplevels;
 588         nvlist_t **child;
 589         uint_t c, children;
 590         nvlist_t *nv;
 591         char *type;
 592         replication_level_t lastrep = {0};
 593         replication_level_t rep;
 594         replication_level_t *ret;
 595         boolean_t dontreport;
 596
 597         ret = safe_malloc(sizeof (replication_level_t));
 598
 599         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 600             &top, &toplevels) == 0);
 601
 602         lastrep.zprl_type = NULL;
 603         for (t = 0; t < toplevels; t++) {
 604                 uint64_t is_log = B_FALSE;
 605
 606                 nv = top[t];
 607
 608                 /*
 609                  * For separate logs we ignore the top level vdev replication
 610                  * constraints.
 611                  */
 612                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 613                 if (is_log)
 614                         continue;
 615
 616                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 617                     &type) == 0);
 618                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 619                     &child, &children) != 0) {
 620                         /*
 621                          * This is a 'file' or 'disk' vdev.
 622                          */
 623                         rep.zprl_type = type;
 624                         rep.zprl_children = 1;
 625                         rep.zprl_parity = 0;
 626                 } else {
 627                         uint64_t vdev_size;
 628
 629                         /*
 630                          * This is a mirror or RAID-Z vdev.  Go through and make
 631                          * sure the contents are all the same (files vs. disks),
 632                          * keeping track of the number of elements in the
 633                          * process.
 634                          *
 635                          * We also check that the size of each vdev (if it can
 636                          * be determined) is the same.
 637                          */
 638                         rep.zprl_type = type;
 639                         rep.zprl_children = 0;
 640
 641                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 642                                 verify(nvlist_lookup_uint64(nv,
 643                                     ZPOOL_CONFIG_NPARITY,
 644                                     &rep.zprl_parity) == 0);
 645                                 assert(rep.zprl_parity != 0);
 646                         } else {
 647                                 rep.zprl_parity = 0;
 648                         }
 649
 650                         /*
 651                          * The 'dontreport' variable indicates that we've
 652                          * already reported an error for this spec, so don't
 653                          * bother doing it again.
 654                          */
 655                         type = NULL;
 656                         dontreport = 0;
 657                         vdev_size = -1ULL;
 658                         for (c = 0; c < children; c++) {
 659                                 nvlist_t *cnv = child[c];
 660                                 char *path;
 661                                 struct stat64 statbuf;
 662                                 uint64_t size = -1ULL;
 663                                 char *childtype;
 664                                 int fd, err;
 665
 666                                 rep.zprl_children++;
 667
 668                                 verify(nvlist_lookup_string(cnv,
 669                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 670
 671                                 /*
 672                                  * If this is a replacing or spare vdev, then
 673                                  * get the real first child of the vdev.
 674                                  */
 675                                 if (strcmp(childtype,
 676                                     VDEV_TYPE_REPLACING) == 0 ||
 677                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 678                                         nvlist_t **rchild;
 679                                         uint_t rchildren;
 680
 681                                         verify(nvlist_lookup_nvlist_array(cnv,
 682                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 683                                             &rchildren) == 0);
 684                                         assert(rchildren == 2);
 685                                         cnv = rchild[0];
 686
 687                                         verify(nvlist_lookup_string(cnv,
 688                                             ZPOOL_CONFIG_TYPE,
 689                                             &childtype) == 0);
 690                                 }
 691
 692                                 verify(nvlist_lookup_string(cnv,
 693                                     ZPOOL_CONFIG_PATH, &path) == 0);
 694
 695                                 /*
 696                                  * If we have a raidz/mirror that combines disks
 697                                  * with files, report it as an error.
 698                                  */
 699                                 if (!dontreport && type != NULL &&
 700                                     strcmp(type, childtype) != 0) {
 701                                         if (ret != NULL)
 702                                                 free(ret);
 703                                         ret = NULL;
 704                                         if (fatal)
 705                                                 vdev_error(gettext(
 706                                                     "mismatched replication "
 707                                                     "level: %s contains both "
 708                                                     "files and devices\n"),
 709                                                     rep.zprl_type);
 710                                         else
 711                                                 return (NULL);
 712                                         dontreport = B_TRUE;
 713                                 }
 714
 715                                 /*
 716                                  * According to stat(2), the value of 'st_size'
 717                                  * is undefined for block devices and character
 718                                  * devices.  But there is no effective way to
 719                                  * determine the real size in userland.
 720                                  *
 721                                  * Instead, we'll take advantage of an
 722                                  * implementation detail of spec_size().  If the
 723                                  * device is currently open, then we (should)
 724                                  * return a valid size.
 725                                  *
 726                                  * If we still don't get a valid size (indicated
 727                                  * by a size of 0 or MAXOFFSET_T), then ignore
 728                                  * this device altogether.
 729                                  */
 730                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 731                                         err = fstat64(fd, &statbuf);
 732                                         (void) close(fd);
 733                                 } else {
 734                                         err = stat64(path, &statbuf);
 735                                 }
 736
 737                                 if (err != 0 ||
 738                                     statbuf.st_size == 0 ||
 739                                     statbuf.st_size == MAXOFFSET_T)
 740                                         continue;
 741
 742                                 size = statbuf.st_size;
 743
 744                                 /*
 745                                  * Also make sure that devices and
 746                                  * slices have a consistent size.  If
 747                                  * they differ by a significant amount
 748                                  * (~16MB) then report an error.
 749                                  */
 750                                 if (!dontreport &&
 751                                     (vdev_size != -1ULL &&
 752                                     (labs(size - vdev_size) >
 753                                     ZPOOL_FUZZ))) {
 754                                         if (ret != NULL)
 755                                                 free(ret);
 756                                         ret = NULL;
 757                                         if (fatal)
 758                                                 vdev_error(gettext(
 759                                                     "%s contains devices of "
 760                                                     "different sizes\n"),
 761                                                     rep.zprl_type);
 762                                         else
 763                                                 return (NULL);
 764                                         dontreport = B_TRUE;
 765                                 }
 766
 767                                 type = childtype;
 768                                 vdev_size = size;
 769                         }
 770                 }
 771
 772                 /*
 773                  * At this point, we have the replication of the last toplevel
 774                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 775                  * different.
 776                  */
 777                 if (lastrep.zprl_type != NULL) {
 778                         if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
 779                                 if (ret != NULL)
 780                                         free(ret);
 781                                 ret = NULL;
 782                                 if (fatal)
 783                                         vdev_error(gettext(
 784                                             "mismatched replication level: "
 785                                             "both %s and %s vdevs are "
 786                                             "present\n"),
 787                                             lastrep.zprl_type, rep.zprl_type);
 788                                 else
 789                                         return (NULL);
 790                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 791                                 if (ret)
 792                                         free(ret);
 793                                 ret = NULL;
 794                                 if (fatal)
 795                                         vdev_error(gettext(
 796                                             "mismatched replication level: "
 797                                             "both %llu and %llu device parity "
 798                                             "%s vdevs are present\n"),
 799                                             lastrep.zprl_parity,
 800                                             rep.zprl_parity,
 801                                             rep.zprl_type);
 802                                 else
 803                                         return (NULL);
 804                         } else if (lastrep.zprl_children != rep.zprl_children) {
 805                                 if (ret)
 806                                         free(ret);
 807                                 ret = NULL;
 808                                 if (fatal)
 809                                         vdev_error(gettext(
 810                                             "mismatched replication level: "
 811                                             "both %llu-way and %llu-way %s "
 812                                             "vdevs are present\n"),
 813                                             lastrep.zprl_children,
 814                                             rep.zprl_children,
 815                                             rep.zprl_type);
 816                                 else
 817                                         return (NULL);
 818                         }
 819                 }
 820                 lastrep = rep;
 821         }
 822
 823         if (ret != NULL)
 824                 *ret = rep;
 825
 826         return (ret);
 827 }
 828
 829 /*
 830  * Check the replication level of the vdev spec against the current pool.  Calls
 831  * get_replication() to make sure the new spec is self-consistent.  If the pool
 832  * has a consistent replication level, then we ignore any errors.  Otherwise,
 833  * report any difference between the two.
 834  */
 835 static int
 836 check_replication(nvlist_t *config, nvlist_t *newroot)
 837 {
 838         nvlist_t **child;
 839         uint_t  children;
 840         replication_level_t *current = NULL, *new;
 841         int ret;
 842
 843         /*
 844          * If we have a current pool configuration, check to see if it's
 845          * self-consistent.  If not, simply return success.
 846          */
 847         if (config != NULL) {
 848                 nvlist_t *nvroot;
 849
 850                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 851                     &nvroot) == 0);
 852                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
 853                         return (0);
 854         }
 855         /*
 856          * for spares there may be no children, and therefore no
 857          * replication level to check
 858          */
 859         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
 860             &child, &children) != 0) || (children == 0)) {
 861                 free(current);
 862                 return (0);
 863         }
 864
 865         /*
 866          * If all we have is logs then there's no replication level to check.
 867          */
 868         if (num_logs(newroot) == children) {
 869                 free(current);
 870                 return (0);
 871         }
 872
 873         /*
 874          * Get the replication level of the new vdev spec, reporting any
 875          * inconsistencies found.
 876          */
 877         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
 878                 free(current);
 879                 return (-1);
 880         }
 881
 882         /*
 883          * Check to see if the new vdev spec matches the replication level of
 884          * the current pool.
 885          */
 886         ret = 0;
 887         if (current != NULL) {
 888                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
 889                         vdev_error(gettext(
 890                             "mismatched replication level: pool uses %s "
 891                             "and new vdev is %s\n"),
 892                             current->zprl_type, new->zprl_type);
 893                         ret = -1;
 894                 } else if (current->zprl_parity != new->zprl_parity) {
 895                         vdev_error(gettext(
 896                             "mismatched replication level: pool uses %llu "
 897                             "device parity and new vdev uses %llu\n"),
 898                             current->zprl_parity, new->zprl_parity);
 899                         ret = -1;
 900                 } else if (current->zprl_children != new->zprl_children) {
 901                         vdev_error(gettext(
 902                             "mismatched replication level: pool uses %llu-way "
 903                             "%s and new vdev uses %llu-way %s\n"),
 904                             current->zprl_children, current->zprl_type,
 905                             new->zprl_children, new->zprl_type);
 906                         ret = -1;
 907                 }
 908         }
 909
 910         free(new);
 911         if (current != NULL)
 912                 free(current);
 913
 914         return (ret);
 915 }
 916
 917 #ifdef illumos
 918 /*
 919  * Go through and find any whole disks in the vdev specification, labelling them
 920  * as appropriate.  When constructing the vdev spec, we were unable to open this
 921  * device in order to provide a devid.  Now that we have labelled the disk and
 922  * know that slice 0 is valid, we can construct the devid now.
 923  *
 924  * If the disk was already labeled with an EFI label, we will have gotten the
 925  * devid already (because we were able to open the whole disk).  Otherwise, we
 926  * need to get the devid after we label the disk.
 927  */
 928 static int
 929 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 930 {
 931         nvlist_t **child;
 932         uint_t c, children;
 933         char *type, *path, *diskname;
 934         char buf[MAXPATHLEN];
 935         uint64_t wholedisk;
 936         int fd;
 937         int ret;
 938         ddi_devid_t devid;
 939         char *minor = NULL, *devid_str = NULL;
 940
 941         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
 942
 943         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 944             &child, &children) != 0) {
 945
 946                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
 947                         return (0);
 948
 949                 /*
 950                  * We have a disk device.  Get the path to the device
 951                  * and see if it's a whole disk by appending the backup
 952                  * slice and stat()ing the device.
 953                  */
 954                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
 955                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
 956                     &wholedisk) != 0 || !wholedisk)
 957                         return (0);
 958
 959                 diskname = strrchr(path, '/');
 960                 assert(diskname != NULL);
 961                 diskname++;
 962                 if (zpool_label_disk(g_zfs, zhp, diskname) == -1)
 963                         return (-1);
 964
 965                 /*
 966                  * Fill in the devid, now that we've labeled the disk.
 967                  */
 968                 (void) snprintf(buf, sizeof (buf), "%ss0", path);
 969                 if ((fd = open(buf, O_RDONLY)) < 0) {
 970                         (void) fprintf(stderr,
 971                             gettext("cannot open '%s': %s\n"),
 972                             buf, strerror(errno));
 973                         return (-1);
 974                 }
 975
 976                 if (devid_get(fd, &devid) == 0) {
 977                         if (devid_get_minor_name(fd, &minor) == 0 &&
 978                             (devid_str = devid_str_encode(devid, minor)) !=
 979                             NULL) {
 980                                 verify(nvlist_add_string(nv,
 981                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 982                         }
 983                         if (devid_str != NULL)
 984                                 devid_str_free(devid_str);
 985                         if (minor != NULL)
 986                                 devid_str_free(minor);
 987                         devid_free(devid);
 988                 }
 989
 990                 /*
 991                  * Update the path to refer to the 's0' slice.  The presence of
 992                  * the 'whole_disk' field indicates to the CLI that we should
 993                  * chop off the slice number when displaying the device in
 994                  * future output.
 995                  */
 996                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0);
 997
 998                 (void) close(fd);
 999
1000                 return (0);
1001         }
1002
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_SPARES,
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         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1014             &child, &children) == 0)
1015                 for (c = 0; c < children; c++)
1016                         if ((ret = make_disks(zhp, child[c])) != 0)
1017                                 return (ret);
1018
1019         return (0);
1020 }
1021 #endif  /* illumos */
1022
1023 /*
1024  * Determine if the given path is a hot spare within the given configuration.
1025  */
1026 static boolean_t
1027 is_spare(nvlist_t *config, const char *path)
1028 {
1029         int fd;
1030         pool_state_t state;
1031         char *name = NULL;
1032         nvlist_t *label;
1033         uint64_t guid, spareguid;
1034         nvlist_t *nvroot;
1035         nvlist_t **spares;
1036         uint_t i, nspares;
1037         boolean_t inuse;
1038
1039         if ((fd = open(path, O_RDONLY)) < 0)
1040                 return (B_FALSE);
1041
1042         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
1043             !inuse ||
1044             state != POOL_STATE_SPARE ||
1045             zpool_read_label(fd, &label) != 0) {
1046                 free(name);
1047                 (void) close(fd);
1048                 return (B_FALSE);
1049         }
1050         free(name);
1051         (void) close(fd);
1052
1053         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
1054         nvlist_free(label);
1055
1056         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1057             &nvroot) == 0);
1058         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1059             &spares, &nspares) == 0) {
1060                 for (i = 0; i < nspares; i++) {
1061                         verify(nvlist_lookup_uint64(spares[i],
1062                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
1063                         if (spareguid == guid)
1064                                 return (B_TRUE);
1065                 }
1066         }
1067
1068         return (B_FALSE);
1069 }
1070
1071 /*
1072  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1073  * the majority of this task.
1074  */
1075 static boolean_t
1076 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1077     boolean_t replacing, boolean_t isspare)
1078 {
1079         nvlist_t **child;
1080         uint_t c, children;
1081         char *type, *path;
1082         int ret = 0;
1083         char buf[MAXPATHLEN];
1084         uint64_t wholedisk;
1085         boolean_t anyinuse = B_FALSE;
1086
1087         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1088
1089         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1090             &child, &children) != 0) {
1091
1092                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
1093
1094                 /*
1095                  * As a generic check, we look to see if this is a replace of a
1096                  * hot spare within the same pool.  If so, we allow it
1097                  * regardless of what libdiskmgt or zpool_in_use() says.
1098                  */
1099                 if (replacing) {
1100 #ifdef illumos
1101                         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1102                             &wholedisk) == 0 && wholedisk)
1103                                 (void) snprintf(buf, sizeof (buf), "%ss0",
1104                                     path);
1105                         else
1106 #endif
1107                                 (void) strlcpy(buf, path, sizeof (buf));
1108
1109                         if (is_spare(config, buf))
1110                                 return (B_FALSE);
1111                 }
1112
1113                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1114                         ret = check_device(path, force, isspare);
1115                 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1116                         ret = check_file(path, force, isspare);
1117
1118                 return (ret != 0);
1119         }
1120
1121         for (c = 0; c < children; c++)
1122                 if (is_device_in_use(config, child[c], force, replacing,
1123                     B_FALSE))
1124                         anyinuse = B_TRUE;
1125
1126         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1127             &child, &children) == 0)
1128                 for (c = 0; c < children; c++)
1129                         if (is_device_in_use(config, child[c], force, replacing,
1130                             B_TRUE))
1131                                 anyinuse = B_TRUE;
1132
1133         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1134             &child, &children) == 0)
1135                 for (c = 0; c < children; c++)
1136                         if (is_device_in_use(config, child[c], force, replacing,
1137                             B_FALSE))
1138                                 anyinuse = B_TRUE;
1139
1140         return (anyinuse);
1141 }
1142
1143 static const char *
1144 is_grouping(const char *type, int *mindev, int *maxdev)
1145 {
1146         if (strncmp(type, "raidz", 5) == 0) {
1147                 const char *p = type + 5;
1148                 char *end;
1149                 long nparity;
1150
1151                 if (*p == '\0') {
1152                         nparity = 1;
1153                 } else if (*p == '0') {
1154                         return (NULL); /* no zero prefixes allowed */
1155                 } else {
1156                         errno = 0;
1157                         nparity = strtol(p, &end, 10);
1158                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1159                             *end != '\0')
1160                                 return (NULL);
1161                 }
1162
1163                 if (mindev != NULL)
1164                         *mindev = nparity + 1;
1165                 if (maxdev != NULL)
1166                         *maxdev = 255;
1167                 return (VDEV_TYPE_RAIDZ);
1168         }
1169
1170         if (maxdev != NULL)
1171                 *maxdev = INT_MAX;
1172
1173         if (strcmp(type, "mirror") == 0) {
1174                 if (mindev != NULL)
1175                         *mindev = 2;
1176                 return (VDEV_TYPE_MIRROR);
1177         }
1178
1179         if (strcmp(type, "spare") == 0) {
1180                 if (mindev != NULL)
1181                         *mindev = 1;
1182                 return (VDEV_TYPE_SPARE);
1183         }
1184
1185         if (strcmp(type, "log") == 0) {
1186                 if (mindev != NULL)
1187                         *mindev = 1;
1188                 return (VDEV_TYPE_LOG);
1189         }
1190
1191         if (strcmp(type, "cache") == 0) {
1192                 if (mindev != NULL)
1193                         *mindev = 1;
1194                 return (VDEV_TYPE_L2CACHE);
1195         }
1196
1197         return (NULL);
1198 }
1199
1200 /*
1201  * Construct a syntactically valid vdev specification,
1202  * and ensure that all devices and files exist and can be opened.
1203  * Note: we don't bother freeing anything in the error paths
1204  * because the program is just going to exit anyway.
1205  */
1206 nvlist_t *
1207 construct_spec(int argc, char **argv)
1208 {
1209         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1210         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1211         const char *type;
1212         uint64_t is_log;
1213         boolean_t seen_logs;
1214
1215         top = NULL;
1216         toplevels = 0;
1217         spares = NULL;
1218         l2cache = NULL;
1219         nspares = 0;
1220         nlogs = 0;
1221         nl2cache = 0;
1222         is_log = B_FALSE;
1223         seen_logs = B_FALSE;
1224
1225         while (argc > 0) {
1226                 nv = NULL;
1227
1228                 /*
1229                  * If it's a mirror or raidz, the subsequent arguments are
1230                  * its leaves -- until we encounter the next mirror or raidz.
1231                  */
1232                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1233                         nvlist_t **child = NULL;
1234                         int c, children = 0;
1235
1236                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1237                                 if (spares != NULL) {
1238                                         (void) fprintf(stderr,
1239                                             gettext("invalid vdev "
1240                                             "specification: 'spare' can be "
1241                                             "specified only once\n"));
1242                                         return (NULL);
1243                                 }
1244                                 is_log = B_FALSE;
1245                         }
1246
1247                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1248                                 if (seen_logs) {
1249                                         (void) fprintf(stderr,
1250                                             gettext("invalid vdev "
1251                                             "specification: 'log' can be "
1252                                             "specified only once\n"));
1253                                         return (NULL);
1254                                 }
1255                                 seen_logs = B_TRUE;
1256                                 is_log = B_TRUE;
1257                                 argc--;
1258                                 argv++;
1259                                 /*
1260                                  * A log is not a real grouping device.
1261                                  * We just set is_log and continue.
1262                                  */
1263                                 continue;
1264                         }
1265
1266                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1267                                 if (l2cache != NULL) {
1268                                         (void) fprintf(stderr,
1269                                             gettext("invalid vdev "
1270                                             "specification: 'cache' can be "
1271                                             "specified only once\n"));
1272                                         return (NULL);
1273                                 }
1274                                 is_log = B_FALSE;
1275                         }
1276
1277                         if (is_log) {
1278                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1279                                         (void) fprintf(stderr,
1280                                             gettext("invalid vdev "
1281                                             "specification: unsupported 'log' "
1282                                             "device: %s\n"), type);
1283                                         return (NULL);
1284                                 }
1285                                 nlogs++;
1286                         }
1287
1288                         for (c = 1; c < argc; c++) {
1289                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1290                                         break;
1291                                 children++;
1292                                 child = realloc(child,
1293                                     children * sizeof (nvlist_t *));
1294                                 if (child == NULL)
1295                                         zpool_no_memory();
1296                                 if ((nv = make_leaf_vdev(argv[c], B_FALSE))
1297                                     == NULL)
1298                                         return (NULL);
1299                                 child[children - 1] = nv;
1300                         }
1301
1302                         if (children < mindev) {
1303                                 (void) fprintf(stderr, gettext("invalid vdev "
1304                                     "specification: %s requires at least %d "
1305                                     "devices\n"), argv[0], mindev);
1306                                 return (NULL);
1307                         }
1308
1309                         if (children > maxdev) {
1310                                 (void) fprintf(stderr, gettext("invalid vdev "
1311                                     "specification: %s supports no more than "
1312                                     "%d devices\n"), argv[0], maxdev);
1313                                 return (NULL);
1314                         }
1315
1316                         argc -= c;
1317                         argv += c;
1318
1319                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1320                                 spares = child;
1321                                 nspares = children;
1322                                 continue;
1323                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1324                                 l2cache = child;
1325                                 nl2cache = children;
1326                                 continue;
1327                         } else {
1328                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1329                                     0) == 0);
1330                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1331                                     type) == 0);
1332                                 verify(nvlist_add_uint64(nv,
1333                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1334                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1335                                         verify(nvlist_add_uint64(nv,
1336                                             ZPOOL_CONFIG_NPARITY,
1337                                             mindev - 1) == 0);
1338                                 }
1339                                 verify(nvlist_add_nvlist_array(nv,
1340                                     ZPOOL_CONFIG_CHILDREN, child,
1341                                     children) == 0);
1342
1343                                 for (c = 0; c < children; c++)
1344                                         nvlist_free(child[c]);
1345                                 free(child);
1346                         }
1347                 } else {
1348                         /*
1349                          * We have a device.  Pass off to make_leaf_vdev() to
1350                          * construct the appropriate nvlist describing the vdev.
1351                          */
1352                         if ((nv = make_leaf_vdev(argv[0], is_log)) == NULL)
1353                                 return (NULL);
1354                         if (is_log)
1355                                 nlogs++;
1356                         argc--;
1357                         argv++;
1358                 }
1359
1360                 toplevels++;
1361                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1362                 if (top == NULL)
1363                         zpool_no_memory();
1364                 top[toplevels - 1] = nv;
1365         }
1366
1367         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1368                 (void) fprintf(stderr, gettext("invalid vdev "
1369                     "specification: at least one toplevel vdev must be "
1370                     "specified\n"));
1371                 return (NULL);
1372         }
1373
1374         if (seen_logs && nlogs == 0) {
1375                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1376                     "log requires at least 1 device\n"));
1377                 return (NULL);
1378         }
1379
1380         /*
1381          * Finally, create nvroot and add all top-level vdevs to it.
1382          */
1383         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1384         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1385             VDEV_TYPE_ROOT) == 0);
1386         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1387             top, toplevels) == 0);
1388         if (nspares != 0)
1389                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1390                     spares, nspares) == 0);
1391         if (nl2cache != 0)
1392                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1393                     l2cache, nl2cache) == 0);
1394
1395         for (t = 0; t < toplevels; t++)
1396                 nvlist_free(top[t]);
1397         for (t = 0; t < nspares; t++)
1398                 nvlist_free(spares[t]);
1399         for (t = 0; t < nl2cache; t++)
1400                 nvlist_free(l2cache[t]);
1401         if (spares)
1402                 free(spares);
1403         if (l2cache)
1404                 free(l2cache);
1405         free(top);
1406
1407         return (nvroot);
1408 }
1409
1410 nvlist_t *
1411 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1412     splitflags_t flags, int argc, char **argv)
1413 {
1414         nvlist_t *newroot = NULL, **child;
1415         uint_t c, children;
1416
1417         if (argc > 0) {
1418                 if ((newroot = construct_spec(argc, argv)) == NULL) {
1419                         (void) fprintf(stderr, gettext("Unable to build a "
1420                             "pool from the specified devices\n"));
1421                         return (NULL);
1422                 }
1423
1424 #ifdef illumos
1425                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1426                         nvlist_free(newroot);
1427                         return (NULL);
1428                 }
1429 #endif
1430
1431                 /* avoid any tricks in the spec */
1432                 verify(nvlist_lookup_nvlist_array(newroot,
1433                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1434                 for (c = 0; c < children; c++) {
1435                         char *path;
1436                         const char *type;
1437                         int min, max;
1438
1439                         verify(nvlist_lookup_string(child[c],
1440                             ZPOOL_CONFIG_PATH, &path) == 0);
1441                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1442                                 (void) fprintf(stderr, gettext("Cannot use "
1443                                     "'%s' as a device for splitting\n"), type);
1444                                 nvlist_free(newroot);
1445                                 return (NULL);
1446                         }
1447                 }
1448         }
1449
1450         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1451                 nvlist_free(newroot);
1452                 return (NULL);
1453         }
1454
1455         return (newroot);
1456 }
1457
1458 /*
1459  * Get and validate the contents of the given vdev specification.  This ensures
1460  * that the nvlist returned is well-formed, that all the devices exist, and that
1461  * they are not currently in use by any other known consumer.  The 'poolconfig'
1462  * parameter is the current configuration of the pool when adding devices
1463  * existing pool, and is used to perform additional checks, such as changing the
1464  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1465  * new pool.  The 'force' flag controls whether devices should be forcefully
1466  * added, even if they appear in use.
1467  */
1468 nvlist_t *
1469 make_root_vdev(zpool_handle_t *zhp, int force, int check_rep,
1470     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1471 {
1472         nvlist_t *newroot;
1473         nvlist_t *poolconfig = NULL;
1474         is_force = force;
1475
1476         /*
1477          * Construct the vdev specification.  If this is successful, we know
1478          * that we have a valid specification, and that all devices can be
1479          * opened.
1480          */
1481         if ((newroot = construct_spec(argc, argv)) == NULL)
1482                 return (NULL);
1483
1484         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
1485                 return (NULL);
1486
1487         /*
1488          * Validate each device to make sure that its not shared with another
1489          * subsystem.  We do this even if 'force' is set, because there are some
1490          * uses (such as a dedicated dump device) that even '-f' cannot
1491          * override.
1492          */
1493         if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1494                 nvlist_free(newroot);
1495                 return (NULL);
1496         }
1497
1498         /*
1499          * Check the replication level of the given vdevs and report any errors
1500          * found.  We include the existing pool spec, if any, as we need to
1501          * catch changes against the existing replication level.
1502          */
1503         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1504                 nvlist_free(newroot);
1505                 return (NULL);
1506         }
1507
1508 #ifdef illumos
1509         /*
1510          * Run through the vdev specification and label any whole disks found.
1511          */
1512         if (!dryrun && make_disks(zhp, newroot) != 0) {
1513                 nvlist_free(newroot);
1514                 return (NULL);
1515         }
1516 #endif
1517
1518         return (newroot);
1519 }