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  */
  26
  27 /*
  28  * Functions to convert between a list of vdevs and an nvlist representing the
  29  * configuration.  Each entry in the list can be one of:
  30  *
  31  *      Device vdevs
  32  *              disk=(path=..., devid=...)
  33  *              file=(path=...)
  34  *
  35  *      Group vdevs
  36  *              raidz[1|2]=(...)
  37  *              mirror=(...)
  38  *
  39  *      Hot spares
  40  *
  41  * While the underlying implementation supports it, group vdevs cannot contain
  42  * other group vdevs.  All userland verification of devices is contained within
  43  * this file.  If successful, the nvlist returned can be passed directly to the
  44  * kernel; we've done as much verification as possible in userland.
  45  *
  46  * Hot spares are a special case, and passed down as an array of disk vdevs, at
  47  * the same level as the root of the vdev tree.
  48  *
  49  * The only function exported by this file is 'make_root_vdev'.  The
  50  * function performs several passes:
  51  *
  52  *      1. Construct the vdev specification.  Performs syntax validation and
  53  *         makes sure each device is valid.
  54  *      2. Check for devices in use.  Using libdiskmgt, makes sure that no
  55  *         devices are also in use.  Some can be overridden using the 'force'
  56  *         flag, others cannot.
  57  *      3. Check for replication errors if the 'force' flag is not specified.
  58  *         validates that the replication level is consistent across the
  59  *         entire pool.
  60  *      4. Call libzfs to label any whole disks with an EFI label.
  61  */
  62
  63 #include <assert.h>
  64 #include <devid.h>
  65 #include <errno.h>
  66 #include <fcntl.h>
  67 #include <libintl.h>
  68 #include <libnvpair.h>
  69 #include <limits.h>
  70 #include <stdio.h>
  71 #include <string.h>
  72 #include <unistd.h>
  73 #include <paths.h>
  74 #include <sys/stat.h>
  75 #include <sys/disk.h>
  76 #include <sys/mntent.h>
  77 #include <libgeom.h>
  78
  79 #include "zpool_util.h"
  80
  81 #define DISK_ROOT       "/dev/dsk"
  82 #define RDISK_ROOT      "/dev/rdsk"
  83 #define BACKUP_SLICE    "s2"
  84
  85 /*
  86  * For any given vdev specification, we can have multiple errors.  The
  87  * vdev_error() function keeps track of whether we have seen an error yet, and
  88  * prints out a header if its the first error we've seen.
  89  */
  90 boolean_t error_seen;
  91 boolean_t is_force;
  92
  93 /*PRINTFLIKE1*/
  94 static void
  95 vdev_error(const char *fmt, ...)
  96 {
  97         va_list ap;
  98
  99         if (!error_seen) {
 100                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
 101                 if (!is_force)
 102                         (void) fprintf(stderr, gettext("use '-f' to override "
 103                             "the following errors:\n"));
 104                 else
 105                         (void) fprintf(stderr, gettext("the following errors "
 106                             "must be manually repaired:\n"));
 107                 error_seen = B_TRUE;
 108         }
 109
 110         va_start(ap, fmt);
 111         (void) vfprintf(stderr, fmt, ap);
 112         va_end(ap);
 113 }
 114
 115 #ifdef sun
 116 static void
 117 libdiskmgt_error(int error)
 118 {
 119         /*
 120          * ENXIO/ENODEV is a valid error message if the device doesn't live in
 121          * /dev/dsk.  Don't bother printing an error message in this case.
 122          */
 123         if (error == ENXIO || error == ENODEV)
 124                 return;
 125
 126         (void) fprintf(stderr, gettext("warning: device in use checking "
 127             "failed: %s\n"), strerror(error));
 128 }
 129
 130 /*
 131  * Validate a device, passing the bulk of the work off to libdiskmgt.
 132  */
 133 static int
 134 check_slice(const char *path, int force, boolean_t wholedisk, boolean_t isspare)
 135 {
 136         char *msg;
 137         int error = 0;
 138         dm_who_type_t who;
 139
 140         if (force)
 141                 who = DM_WHO_ZPOOL_FORCE;
 142         else if (isspare)
 143                 who = DM_WHO_ZPOOL_SPARE;
 144         else
 145                 who = DM_WHO_ZPOOL;
 146
 147         if (dm_inuse((char *)path, &msg, who, &error) || error) {
 148                 if (error != 0) {
 149                         libdiskmgt_error(error);
 150                         return (0);
 151                 } else {
 152                         vdev_error("%s", msg);
 153                         free(msg);
 154                         return (-1);
 155                 }
 156         }
 157
 158         /*
 159          * If we're given a whole disk, ignore overlapping slices since we're
 160          * about to label it anyway.
 161          */
 162         error = 0;
 163         if (!wholedisk && !force &&
 164             (dm_isoverlapping((char *)path, &msg, &error) || error)) {
 165                 if (error == 0) {
 166                         /* dm_isoverlapping returned -1 */
 167                         vdev_error(gettext("%s overlaps with %s\n"), path, msg);
 168                         free(msg);
 169                         return (-1);
 170                 } else if (error != ENODEV) {
 171                         /* libdiskmgt's devcache only handles physical drives */
 172                         libdiskmgt_error(error);
 173                         return (0);
 174                 }
 175         }
 176
 177         return (0);
 178 }
 179
 180
 181 /*
 182  * Validate a whole disk.  Iterate over all slices on the disk and make sure
 183  * that none is in use by calling check_slice().
 184  */
 185 static int
 186 check_disk(const char *name, dm_descriptor_t disk, int force, int isspare)
 187 {
 188         dm_descriptor_t *drive, *media, *slice;
 189         int err = 0;
 190         int i;
 191         int ret;
 192
 193         /*
 194          * Get the drive associated with this disk.  This should never fail,
 195          * because we already have an alias handle open for the device.
 196          */
 197         if ((drive = dm_get_associated_descriptors(disk, DM_DRIVE,
 198             &err)) == NULL || *drive == NULL) {
 199                 if (err)
 200                         libdiskmgt_error(err);
 201                 return (0);
 202         }
 203
 204         if ((media = dm_get_associated_descriptors(*drive, DM_MEDIA,
 205             &err)) == NULL) {
 206                 dm_free_descriptors(drive);
 207                 if (err)
 208                         libdiskmgt_error(err);
 209                 return (0);
 210         }
 211
 212         dm_free_descriptors(drive);
 213
 214         /*
 215          * It is possible that the user has specified a removable media drive,
 216          * and the media is not present.
 217          */
 218         if (*media == NULL) {
 219                 dm_free_descriptors(media);
 220                 vdev_error(gettext("'%s' has no media in drive\n"), name);
 221                 return (-1);
 222         }
 223
 224         if ((slice = dm_get_associated_descriptors(*media, DM_SLICE,
 225             &err)) == NULL) {
 226                 dm_free_descriptors(media);
 227                 if (err)
 228                         libdiskmgt_error(err);
 229                 return (0);
 230         }
 231
 232         dm_free_descriptors(media);
 233
 234         ret = 0;
 235
 236         /*
 237          * Iterate over all slices and report any errors.  We don't care about
 238          * overlapping slices because we are using the whole disk.
 239          */
 240         for (i = 0; slice[i] != NULL; i++) {
 241                 char *name = dm_get_name(slice[i], &err);
 242
 243                 if (check_slice(name, force, B_TRUE, isspare) != 0)
 244                         ret = -1;
 245
 246                 dm_free_name(name);
 247         }
 248
 249         dm_free_descriptors(slice);
 250         return (ret);
 251 }
 252
 253 /*
 254  * Validate a device.
 255  */
 256 static int
 257 check_device(const char *path, boolean_t force, boolean_t isspare)
 258 {
 259         dm_descriptor_t desc;
 260         int err;
 261         char *dev;
 262
 263         /*
 264          * For whole disks, libdiskmgt does not include the leading dev path.
 265          */
 266         dev = strrchr(path, '/');
 267         assert(dev != NULL);
 268         dev++;
 269         if ((desc = dm_get_descriptor_by_name(DM_ALIAS, dev, &err)) != NULL) {
 270                 err = check_disk(path, desc, force, isspare);
 271                 dm_free_descriptor(desc);
 272                 return (err);
 273         }
 274
 275         return (check_slice(path, force, B_FALSE, isspare));
 276 }
 277 #endif  /* sun */
 278
 279 /*
 280  * Check that a file is valid.  All we can do in this case is check that it's
 281  * not in use by another pool, and not in use by swap.
 282  */
 283 static int
 284 check_file(const char *file, boolean_t force, boolean_t isspare)
 285 {
 286         char  *name;
 287         int fd;
 288         int ret = 0;
 289         int err;
 290         pool_state_t state;
 291         boolean_t inuse;
 292
 293 #ifdef sun
 294         if (dm_inuse_swap(file, &err)) {
 295                 if (err)
 296                         libdiskmgt_error(err);
 297                 else
 298                         vdev_error(gettext("%s is currently used by swap. "
 299                             "Please see swap(1M).\n"), file);
 300                 return (-1);
 301         }
 302 #endif
 303
 304         if ((fd = open(file, O_RDONLY)) < 0)
 305                 return (0);
 306
 307         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
 308                 const char *desc;
 309
 310                 switch (state) {
 311                 case POOL_STATE_ACTIVE:
 312                         desc = gettext("active");
 313                         break;
 314
 315                 case POOL_STATE_EXPORTED:
 316                         desc = gettext("exported");
 317                         break;
 318
 319                 case POOL_STATE_POTENTIALLY_ACTIVE:
 320                         desc = gettext("potentially active");
 321                         break;
 322
 323                 default:
 324                         desc = gettext("unknown");
 325                         break;
 326                 }
 327
 328                 /*
 329                  * Allow hot spares to be shared between pools.
 330                  */
 331                 if (state == POOL_STATE_SPARE && isspare)
 332                         return (0);
 333
 334                 if (state == POOL_STATE_ACTIVE ||
 335                     state == POOL_STATE_SPARE || !force) {
 336                         switch (state) {
 337                         case POOL_STATE_SPARE:
 338                                 vdev_error(gettext("%s is reserved as a hot "
 339                                     "spare for pool %s\n"), file, name);
 340                                 break;
 341                         default:
 342                                 vdev_error(gettext("%s is part of %s pool "
 343                                     "'%s'\n"), file, desc, name);
 344                                 break;
 345                         }
 346                         ret = -1;
 347                 }
 348
 349                 free(name);
 350         }
 351
 352         (void) close(fd);
 353         return (ret);
 354 }
 355
 356 static int
 357 check_device(const char *name, boolean_t force, boolean_t isspare)
 358 {
 359         char path[MAXPATHLEN];
 360
 361         if (strncmp(name, _PATH_DEV, sizeof(_PATH_DEV) - 1) != 0)
 362                 snprintf(path, sizeof(path), "%s%s", _PATH_DEV, name);
 363         else
 364                 strlcpy(path, name, sizeof(path));
 365
 366         return (check_file(path, force, isspare));
 367 }
 368
 369 /*
 370  * By "whole disk" we mean an entire physical disk (something we can
 371  * label, toggle the write cache on, etc.) as opposed to the full
 372  * capacity of a pseudo-device such as lofi or did.  We act as if we
 373  * are labeling the disk, which should be a pretty good test of whether
 374  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 375  * it isn't.
 376  */
 377 static boolean_t
 378 is_whole_disk(const char *arg)
 379 {
 380 #ifdef sun
 381         struct dk_gpt *label;
 382         int     fd;
 383         char    path[MAXPATHLEN];
 384
 385         (void) snprintf(path, sizeof (path), "%s%s%s",
 386             RDISK_ROOT, strrchr(arg, '/'), BACKUP_SLICE);
 387         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0)
 388                 return (B_FALSE);
 389         if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
 390                 (void) close(fd);
 391                 return (B_FALSE);
 392         }
 393         efi_free(label);
 394         (void) close(fd);
 395         return (B_TRUE);
 396 #else
 397         int fd;
 398
 399         fd = g_open(arg, 0);
 400         if (fd >= 0) {
 401                 g_close(fd);
 402                 return (B_TRUE);
 403         }
 404         return (B_FALSE);
 405 #endif
 406 }
 407
 408 /*
 409  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 410  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 411  * leaf vdev are:
 412  *
 413  *      /dev/dsk/xxx    Complete disk path
 414  *      /xxx            Full path to file
 415  *      xxx             Shorthand for /dev/dsk/xxx
 416  */
 417 static nvlist_t *
 418 make_leaf_vdev(const char *arg, uint64_t is_log)
 419 {
 420         char path[MAXPATHLEN];
 421         struct stat64 statbuf;
 422         nvlist_t *vdev = NULL;
 423         char *type = NULL;
 424         boolean_t wholedisk = B_FALSE;
 425
 426         /*
 427          * Determine what type of vdev this is, and put the full path into
 428          * 'path'.  We detect whether this is a device of file afterwards by
 429          * checking the st_mode of the file.
 430          */
 431         if (arg[0] == '/') {
 432                 /*
 433                  * Complete device or file path.  Exact type is determined by
 434                  * examining the file descriptor afterwards.
 435                  */
 436                 wholedisk = is_whole_disk(arg);
 437                 if (!wholedisk && (stat64(arg, &statbuf) != 0)) {
 438                         (void) fprintf(stderr,
 439                             gettext("cannot open '%s': %s\n"),
 440                             arg, strerror(errno));
 441                         return (NULL);
 442                 }
 443
 444                 (void) strlcpy(path, arg, sizeof (path));
 445         } else {
 446                 /*
 447                  * This may be a short path for a device, or it could be total
 448                  * gibberish.  Check to see if it's a known device in
 449                  * /dev/dsk/.  As part of this check, see if we've been given a
 450                  * an entire disk (minus the slice number).
 451                  */
 452                 if (strncmp(arg, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
 453                         strlcpy(path, arg, sizeof (path));
 454                 else
 455                         snprintf(path, sizeof (path), "%s%s", _PATH_DEV, arg);
 456                 wholedisk = is_whole_disk(path);
 457                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 458                         /*
 459                          * If we got ENOENT, then the user gave us
 460                          * gibberish, so try to direct them with a
 461                          * reasonable error message.  Otherwise,
 462                          * regurgitate strerror() since it's the best we
 463                          * can do.
 464                          */
 465                         if (errno == ENOENT) {
 466                                 (void) fprintf(stderr,
 467                                     gettext("cannot open '%s': no such "
 468                                     "GEOM provider\n"), arg);
 469                                 (void) fprintf(stderr,
 470                                     gettext("must be a full path or "
 471                                     "shorthand device name\n"));
 472                                 return (NULL);
 473                         } else {
 474                                 (void) fprintf(stderr,
 475                                     gettext("cannot open '%s': %s\n"),
 476                                     path, strerror(errno));
 477                                 return (NULL);
 478                         }
 479                 }
 480         }
 481
 482 #ifdef __FreeBSD__
 483         if (S_ISCHR(statbuf.st_mode)) {
 484                 statbuf.st_mode &= ~S_IFCHR;
 485                 statbuf.st_mode |= S_IFBLK;
 486                 wholedisk = B_FALSE;
 487         }
 488 #endif
 489
 490         /*
 491          * Determine whether this is a device or a file.
 492          */
 493         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 494                 type = VDEV_TYPE_DISK;
 495         } else if (S_ISREG(statbuf.st_mode)) {
 496                 type = VDEV_TYPE_FILE;
 497         } else {
 498                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 499                     "GEOM provider or regular file\n"), path);
 500                 return (NULL);
 501         }
 502
 503         /*
 504          * Finally, we have the complete device or file, and we know that it is
 505          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 506          * vdevs have a 'path' element, and devices also have a 'devid' element.
 507          */
 508         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 509         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 510         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 511         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 512         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 513                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 514                     (uint64_t)wholedisk) == 0);
 515
 516 #ifdef have_devid
 517         /*
 518          * For a whole disk, defer getting its devid until after labeling it.
 519          */
 520         if (S_ISBLK(statbuf.st_mode) && !wholedisk) {
 521                 /*
 522                  * Get the devid for the device.
 523                  */
 524                 int fd;
 525                 ddi_devid_t devid;
 526                 char *minor = NULL, *devid_str = NULL;
 527
 528                 if ((fd = open(path, O_RDONLY)) < 0) {
 529                         (void) fprintf(stderr, gettext("cannot open '%s': "
 530                             "%s\n"), path, strerror(errno));
 531                         nvlist_free(vdev);
 532                         return (NULL);
 533                 }
 534
 535                 if (devid_get(fd, &devid) == 0) {
 536                         if (devid_get_minor_name(fd, &minor) == 0 &&
 537                             (devid_str = devid_str_encode(devid, minor)) !=
 538                             NULL) {
 539                                 verify(nvlist_add_string(vdev,
 540                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 541                         }
 542                         if (devid_str != NULL)
 543                                 devid_str_free(devid_str);
 544                         if (minor != NULL)
 545                                 devid_str_free(minor);
 546                         devid_free(devid);
 547                 }
 548
 549                 (void) close(fd);
 550         }
 551 #endif
 552
 553         return (vdev);
 554 }
 555
 556 /*
 557  * Go through and verify the replication level of the pool is consistent.
 558  * Performs the following checks:
 559  *
 560  *      For the new spec, verifies that devices in mirrors and raidz are the
 561  *      same size.
 562  *
 563  *      If the current configuration already has inconsistent replication
 564  *      levels, ignore any other potential problems in the new spec.
 565  *
 566  *      Otherwise, make sure that the current spec (if there is one) and the new
 567  *      spec have consistent replication levels.
 568  */
 569 typedef struct replication_level {
 570         char *zprl_type;
 571         uint64_t zprl_children;
 572         uint64_t zprl_parity;
 573 } replication_level_t;
 574
 575 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 576
 577 /*
 578  * Given a list of toplevel vdevs, return the current replication level.  If
 579  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 580  * an error message will be displayed for each self-inconsistent vdev.
 581  */
 582 static replication_level_t *
 583 get_replication(nvlist_t *nvroot, boolean_t fatal)
 584 {
 585         nvlist_t **top;
 586         uint_t t, toplevels;
 587         nvlist_t **child;
 588         uint_t c, children;
 589         nvlist_t *nv;
 590         char *type;
 591         replication_level_t lastrep, rep, *ret;
 592         boolean_t dontreport;
 593
 594         ret = safe_malloc(sizeof (replication_level_t));
 595
 596         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 597             &top, &toplevels) == 0);
 598
 599         lastrep.zprl_type = NULL;
 600         for (t = 0; t < toplevels; t++) {
 601                 uint64_t is_log = B_FALSE;
 602
 603                 nv = top[t];
 604
 605                 /*
 606                  * For separate logs we ignore the top level vdev replication
 607                  * constraints.
 608                  */
 609                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 610                 if (is_log)
 611                         continue;
 612
 613                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 614                     &type) == 0);
 615                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 616                     &child, &children) != 0) {
 617                         /*
 618                          * This is a 'file' or 'disk' vdev.
 619                          */
 620                         rep.zprl_type = type;
 621                         rep.zprl_children = 1;
 622                         rep.zprl_parity = 0;
 623                 } else {
 624                         uint64_t vdev_size;
 625
 626                         /*
 627                          * This is a mirror or RAID-Z vdev.  Go through and make
 628                          * sure the contents are all the same (files vs. disks),
 629                          * keeping track of the number of elements in the
 630                          * process.
 631                          *
 632                          * We also check that the size of each vdev (if it can
 633                          * be determined) is the same.
 634                          */
 635                         rep.zprl_type = type;
 636                         rep.zprl_children = 0;
 637
 638                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 639                                 verify(nvlist_lookup_uint64(nv,
 640                                     ZPOOL_CONFIG_NPARITY,
 641                                     &rep.zprl_parity) == 0);
 642                                 assert(rep.zprl_parity != 0);
 643                         } else {
 644                                 rep.zprl_parity = 0;
 645                         }
 646
 647                         /*
 648                          * The 'dontreport' variable indicates that we've
 649                          * already reported an error for this spec, so don't
 650                          * bother doing it again.
 651                          */
 652                         type = NULL;
 653                         dontreport = 0;
 654                         vdev_size = -1ULL;
 655                         for (c = 0; c < children; c++) {
 656                                 nvlist_t *cnv = child[c];
 657                                 char *path;
 658                                 struct stat64 statbuf;
 659                                 uint64_t size = -1ULL;
 660                                 char *childtype;
 661                                 int fd, err;
 662
 663                                 rep.zprl_children++;
 664
 665                                 verify(nvlist_lookup_string(cnv,
 666                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 667
 668                                 /*
 669                                  * If this is a replacing or spare vdev, then
 670                                  * get the real first child of the vdev.
 671                                  */
 672                                 if (strcmp(childtype,
 673                                     VDEV_TYPE_REPLACING) == 0 ||
 674                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 675                                         nvlist_t **rchild;
 676                                         uint_t rchildren;
 677
 678                                         verify(nvlist_lookup_nvlist_array(cnv,
 679                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 680                                             &rchildren) == 0);
 681                                         assert(rchildren == 2);
 682                                         cnv = rchild[0];
 683
 684                                         verify(nvlist_lookup_string(cnv,
 685                                             ZPOOL_CONFIG_TYPE,
 686                                             &childtype) == 0);
 687                                 }
 688
 689                                 verify(nvlist_lookup_string(cnv,
 690                                     ZPOOL_CONFIG_PATH, &path) == 0);
 691
 692                                 /*
 693                                  * If we have a raidz/mirror that combines disks
 694                                  * with files, report it as an error.
 695                                  */
 696                                 if (!dontreport && type != NULL &&
 697                                     strcmp(type, childtype) != 0) {
 698                                         if (ret != NULL)
 699                                                 free(ret);
 700                                         ret = NULL;
 701                                         if (fatal)
 702                                                 vdev_error(gettext(
 703                                                     "mismatched replication "
 704                                                     "level: %s contains both "
 705                                                     "files and devices\n"),
 706                                                     rep.zprl_type);
 707                                         else
 708                                                 return (NULL);
 709                                         dontreport = B_TRUE;
 710                                 }
 711
 712                                 /*
 713                                  * According to stat(2), the value of 'st_size'
 714                                  * is undefined for block devices and character
 715                                  * devices.  But there is no effective way to
 716                                  * determine the real size in userland.
 717                                  *
 718                                  * Instead, we'll take advantage of an
 719                                  * implementation detail of spec_size().  If the
 720                                  * device is currently open, then we (should)
 721                                  * return a valid size.
 722                                  *
 723                                  * If we still don't get a valid size (indicated
 724                                  * by a size of 0 or MAXOFFSET_T), then ignore
 725                                  * this device altogether.
 726                                  */
 727                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 728                                         err = fstat64(fd, &statbuf);
 729                                         (void) close(fd);
 730                                 } else {
 731                                         err = stat64(path, &statbuf);
 732                                 }
 733
 734                                 if (err != 0 ||
 735                                     statbuf.st_size == 0 ||
 736                                     statbuf.st_size == MAXOFFSET_T)
 737                                         continue;
 738
 739                                 size = statbuf.st_size;
 740
 741                                 /*
 742                                  * Also make sure that devices and
 743                                  * slices have a consistent size.  If
 744                                  * they differ by a significant amount
 745                                  * (~16MB) then report an error.
 746                                  */
 747                                 if (!dontreport &&
 748                                     (vdev_size != -1ULL &&
 749                                     (labs(size - vdev_size) >
 750                                     ZPOOL_FUZZ))) {
 751                                         if (ret != NULL)
 752                                                 free(ret);
 753                                         ret = NULL;
 754                                         if (fatal)
 755                                                 vdev_error(gettext(
 756                                                     "%s contains devices of "
 757                                                     "different sizes\n"),
 758                                                     rep.zprl_type);
 759                                         else
 760                                                 return (NULL);
 761                                         dontreport = B_TRUE;
 762                                 }
 763
 764                                 type = childtype;
 765                                 vdev_size = size;
 766                         }
 767                 }
 768
 769                 /*
 770                  * At this point, we have the replication of the last toplevel
 771                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 772                  * different.
 773                  */
 774                 if (lastrep.zprl_type != NULL) {
 775                         if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
 776                                 if (ret != NULL)
 777                                         free(ret);
 778                                 ret = NULL;
 779                                 if (fatal)
 780                                         vdev_error(gettext(
 781                                             "mismatched replication level: "
 782                                             "both %s and %s vdevs are "
 783                                             "present\n"),
 784                                             lastrep.zprl_type, rep.zprl_type);
 785                                 else
 786                                         return (NULL);
 787                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 788                                 if (ret)
 789                                         free(ret);
 790                                 ret = NULL;
 791                                 if (fatal)
 792                                         vdev_error(gettext(
 793                                             "mismatched replication level: "
 794                                             "both %llu and %llu device parity "
 795                                             "%s vdevs are present\n"),
 796                                             lastrep.zprl_parity,
 797                                             rep.zprl_parity,
 798                                             rep.zprl_type);
 799                                 else
 800                                         return (NULL);
 801                         } else if (lastrep.zprl_children != rep.zprl_children) {
 802                                 if (ret)
 803                                         free(ret);
 804                                 ret = NULL;
 805                                 if (fatal)
 806                                         vdev_error(gettext(
 807                                             "mismatched replication level: "
 808                                             "both %llu-way and %llu-way %s "
 809                                             "vdevs are present\n"),
 810                                             lastrep.zprl_children,
 811                                             rep.zprl_children,
 812                                             rep.zprl_type);
 813                                 else
 814                                         return (NULL);
 815                         }
 816                 }
 817                 lastrep = rep;
 818         }
 819
 820         if (ret != NULL)
 821                 *ret = rep;
 822
 823         return (ret);
 824 }
 825
 826 /*
 827  * Check the replication level of the vdev spec against the current pool.  Calls
 828  * get_replication() to make sure the new spec is self-consistent.  If the pool
 829  * has a consistent replication level, then we ignore any errors.  Otherwise,
 830  * report any difference between the two.
 831  */
 832 static int
 833 check_replication(nvlist_t *config, nvlist_t *newroot)
 834 {
 835         nvlist_t **child;
 836         uint_t  children;
 837         replication_level_t *current = NULL, *new;
 838         int ret;
 839
 840         /*
 841          * If we have a current pool configuration, check to see if it's
 842          * self-consistent.  If not, simply return success.
 843          */
 844         if (config != NULL) {
 845                 nvlist_t *nvroot;
 846
 847                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 848                     &nvroot) == 0);
 849                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
 850                         return (0);
 851         }
 852         /*
 853          * for spares there may be no children, and therefore no
 854          * replication level to check
 855          */
 856         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
 857             &child, &children) != 0) || (children == 0)) {
 858                 free(current);
 859                 return (0);
 860         }
 861
 862         /*
 863          * If all we have is logs then there's no replication level to check.
 864          */
 865         if (num_logs(newroot) == children) {
 866                 free(current);
 867                 return (0);
 868         }
 869
 870         /*
 871          * Get the replication level of the new vdev spec, reporting any
 872          * inconsistencies found.
 873          */
 874         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
 875                 free(current);
 876                 return (-1);
 877         }
 878
 879         /*
 880          * Check to see if the new vdev spec matches the replication level of
 881          * the current pool.
 882          */
 883         ret = 0;
 884         if (current != NULL) {
 885                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
 886                         vdev_error(gettext(
 887                             "mismatched replication level: pool uses %s "
 888                             "and new vdev is %s\n"),
 889                             current->zprl_type, new->zprl_type);
 890                         ret = -1;
 891                 } else if (current->zprl_parity != new->zprl_parity) {
 892                         vdev_error(gettext(
 893                             "mismatched replication level: pool uses %llu "
 894                             "device parity and new vdev uses %llu\n"),
 895                             current->zprl_parity, new->zprl_parity);
 896                         ret = -1;
 897                 } else if (current->zprl_children != new->zprl_children) {
 898                         vdev_error(gettext(
 899                             "mismatched replication level: pool uses %llu-way "
 900                             "%s and new vdev uses %llu-way %s\n"),
 901                             current->zprl_children, current->zprl_type,
 902                             new->zprl_children, new->zprl_type);
 903                         ret = -1;
 904                 }
 905         }
 906
 907         free(new);
 908         if (current != NULL)
 909                 free(current);
 910
 911         return (ret);
 912 }
 913
 914 #ifdef sun
 915 /*
 916  * Go through and find any whole disks in the vdev specification, labelling them
 917  * as appropriate.  When constructing the vdev spec, we were unable to open this
 918  * device in order to provide a devid.  Now that we have labelled the disk and
 919  * know that slice 0 is valid, we can construct the devid now.
 920  *
 921  * If the disk was already labeled with an EFI label, we will have gotten the
 922  * devid already (because we were able to open the whole disk).  Otherwise, we
 923  * need to get the devid after we label the disk.
 924  */
 925 static int
 926 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 927 {
 928         nvlist_t **child;
 929         uint_t c, children;
 930         char *type, *path, *diskname;
 931         char buf[MAXPATHLEN];
 932         uint64_t wholedisk;
 933         int fd;
 934         int ret;
 935         ddi_devid_t devid;
 936         char *minor = NULL, *devid_str = NULL;
 937
 938         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
 939
 940         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 941             &child, &children) != 0) {
 942
 943                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
 944                         return (0);
 945
 946                 /*
 947                  * We have a disk device.  Get the path to the device
 948                  * and see if it's a whole disk by appending the backup
 949                  * slice and stat()ing the device.
 950                  */
 951                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
 952                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
 953                     &wholedisk) != 0 || !wholedisk)
 954                         return (0);
 955
 956                 diskname = strrchr(path, '/');
 957                 assert(diskname != NULL);
 958                 diskname++;
 959                 if (zpool_label_disk(g_zfs, zhp, diskname) == -1)
 960                         return (-1);
 961
 962                 /*
 963                  * Fill in the devid, now that we've labeled the disk.
 964                  */
 965                 (void) snprintf(buf, sizeof (buf), "%ss0", path);
 966                 if ((fd = open(buf, O_RDONLY)) < 0) {
 967                         (void) fprintf(stderr,
 968                             gettext("cannot open '%s': %s\n"),
 969                             buf, strerror(errno));
 970                         return (-1);
 971                 }
 972
 973                 if (devid_get(fd, &devid) == 0) {
 974                         if (devid_get_minor_name(fd, &minor) == 0 &&
 975                             (devid_str = devid_str_encode(devid, minor)) !=
 976                             NULL) {
 977                                 verify(nvlist_add_string(nv,
 978                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 979                         }
 980                         if (devid_str != NULL)
 981                                 devid_str_free(devid_str);
 982                         if (minor != NULL)
 983                                 devid_str_free(minor);
 984                         devid_free(devid);
 985                 }
 986
 987                 /*
 988                  * Update the path to refer to the 's0' slice.  The presence of
 989                  * the 'whole_disk' field indicates to the CLI that we should
 990                  * chop off the slice number when displaying the device in
 991                  * future output.
 992                  */
 993                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0);
 994
 995                 (void) close(fd);
 996
 997                 return (0);
 998         }
 999
1000         for (c = 0; c < children; c++)
1001                 if ((ret = make_disks(zhp, child[c])) != 0)
1002                         return (ret);
1003
1004         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1005             &child, &children) == 0)
1006                 for (c = 0; c < children; c++)
1007                         if ((ret = make_disks(zhp, child[c])) != 0)
1008                                 return (ret);
1009
1010         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1011             &child, &children) == 0)
1012                 for (c = 0; c < children; c++)
1013                         if ((ret = make_disks(zhp, child[c])) != 0)
1014                                 return (ret);
1015
1016         return (0);
1017 }
1018 #endif  /* sun */
1019
1020 /*
1021  * Determine if the given path is a hot spare within the given configuration.
1022  */
1023 static boolean_t
1024 is_spare(nvlist_t *config, const char *path)
1025 {
1026         int fd;
1027         pool_state_t state;
1028         char *name = NULL;
1029         nvlist_t *label;
1030         uint64_t guid, spareguid;
1031         nvlist_t *nvroot;
1032         nvlist_t **spares;
1033         uint_t i, nspares;
1034         boolean_t inuse;
1035
1036         if ((fd = open(path, O_RDONLY)) < 0)
1037                 return (B_FALSE);
1038
1039         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
1040             !inuse ||
1041             state != POOL_STATE_SPARE ||
1042             zpool_read_label(fd, &label) != 0) {
1043                 free(name);
1044                 (void) close(fd);
1045                 return (B_FALSE);
1046         }
1047         free(name);
1048         (void) close(fd);
1049
1050         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
1051         nvlist_free(label);
1052
1053         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1054             &nvroot) == 0);
1055         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1056             &spares, &nspares) == 0) {
1057                 for (i = 0; i < nspares; i++) {
1058                         verify(nvlist_lookup_uint64(spares[i],
1059                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
1060                         if (spareguid == guid)
1061                                 return (B_TRUE);
1062                 }
1063         }
1064
1065         return (B_FALSE);
1066 }
1067
1068 /*
1069  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1070  * the majority of this task.
1071  */
1072 static boolean_t
1073 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1074     boolean_t replacing, boolean_t isspare)
1075 {
1076         nvlist_t **child;
1077         uint_t c, children;
1078         char *type, *path;
1079         int ret;
1080         char buf[MAXPATHLEN];
1081         uint64_t wholedisk;
1082         boolean_t anyinuse = B_FALSE;
1083
1084         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1085
1086         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1087             &child, &children) != 0) {
1088
1089                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
1090
1091                 /*
1092                  * As a generic check, we look to see if this is a replace of a
1093                  * hot spare within the same pool.  If so, we allow it
1094                  * regardless of what libdiskmgt or zpool_in_use() says.
1095                  */
1096                 if (replacing) {
1097 #ifdef sun
1098                         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1099                             &wholedisk) == 0 && wholedisk)
1100                                 (void) snprintf(buf, sizeof (buf), "%ss0",
1101                                     path);
1102                         else
1103 #endif
1104                                 (void) strlcpy(buf, path, sizeof (buf));
1105
1106                         if (is_spare(config, buf))
1107                                 return (B_FALSE);
1108                 }
1109
1110                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1111                         ret = check_device(path, force, isspare);
1112                 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1113                         ret = check_file(path, force, isspare);
1114
1115                 return (ret != 0);
1116         }
1117
1118         for (c = 0; c < children; c++)
1119                 if (is_device_in_use(config, child[c], force, replacing,
1120                     B_FALSE))
1121                         anyinuse = B_TRUE;
1122
1123         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1124             &child, &children) == 0)
1125                 for (c = 0; c < children; c++)
1126                         if (is_device_in_use(config, child[c], force, replacing,
1127                             B_TRUE))
1128                                 anyinuse = B_TRUE;
1129
1130         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1131             &child, &children) == 0)
1132                 for (c = 0; c < children; c++)
1133                         if (is_device_in_use(config, child[c], force, replacing,
1134                             B_FALSE))
1135                                 anyinuse = B_TRUE;
1136
1137         return (anyinuse);
1138 }
1139
1140 static const char *
1141 is_grouping(const char *type, int *mindev, int *maxdev)
1142 {
1143         if (strncmp(type, "raidz", 5) == 0) {
1144                 const char *p = type + 5;
1145                 char *end;
1146                 long nparity;
1147
1148                 if (*p == '\0') {
1149                         nparity = 1;
1150                 } else if (*p == '0') {
1151                         return (NULL); /* no zero prefixes allowed */
1152                 } else {
1153                         errno = 0;
1154                         nparity = strtol(p, &end, 10);
1155                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1156                             *end != '\0')
1157                                 return (NULL);
1158                 }
1159
1160                 if (mindev != NULL)
1161                         *mindev = nparity + 1;
1162                 if (maxdev != NULL)
1163                         *maxdev = 255;
1164                 return (VDEV_TYPE_RAIDZ);
1165         }
1166
1167         if (maxdev != NULL)
1168                 *maxdev = INT_MAX;
1169
1170         if (strcmp(type, "mirror") == 0) {
1171                 if (mindev != NULL)
1172                         *mindev = 2;
1173                 return (VDEV_TYPE_MIRROR);
1174         }
1175
1176         if (strcmp(type, "spare") == 0) {
1177                 if (mindev != NULL)
1178                         *mindev = 1;
1179                 return (VDEV_TYPE_SPARE);
1180         }
1181
1182         if (strcmp(type, "log") == 0) {
1183                 if (mindev != NULL)
1184                         *mindev = 1;
1185                 return (VDEV_TYPE_LOG);
1186         }
1187
1188         if (strcmp(type, "cache") == 0) {
1189                 if (mindev != NULL)
1190                         *mindev = 1;
1191                 return (VDEV_TYPE_L2CACHE);
1192         }
1193
1194         return (NULL);
1195 }
1196
1197 /*
1198  * Construct a syntactically valid vdev specification,
1199  * and ensure that all devices and files exist and can be opened.
1200  * Note: we don't bother freeing anything in the error paths
1201  * because the program is just going to exit anyway.
1202  */
1203 nvlist_t *
1204 construct_spec(int argc, char **argv)
1205 {
1206         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1207         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1208         const char *type;
1209         uint64_t is_log;
1210         boolean_t seen_logs;
1211
1212         top = NULL;
1213         toplevels = 0;
1214         spares = NULL;
1215         l2cache = NULL;
1216         nspares = 0;
1217         nlogs = 0;
1218         nl2cache = 0;
1219         is_log = B_FALSE;
1220         seen_logs = B_FALSE;
1221
1222         while (argc > 0) {
1223                 nv = NULL;
1224
1225                 /*
1226                  * If it's a mirror or raidz, the subsequent arguments are
1227                  * its leaves -- until we encounter the next mirror or raidz.
1228                  */
1229                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1230                         nvlist_t **child = NULL;
1231                         int c, children = 0;
1232
1233                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1234                                 if (spares != NULL) {
1235                                         (void) fprintf(stderr,
1236                                             gettext("invalid vdev "
1237                                             "specification: 'spare' can be "
1238                                             "specified only once\n"));
1239                                         return (NULL);
1240                                 }
1241                                 is_log = B_FALSE;
1242                         }
1243
1244                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1245                                 if (seen_logs) {
1246                                         (void) fprintf(stderr,
1247                                             gettext("invalid vdev "
1248                                             "specification: 'log' can be "
1249                                             "specified only once\n"));
1250                                         return (NULL);
1251                                 }
1252                                 seen_logs = B_TRUE;
1253                                 is_log = B_TRUE;
1254                                 argc--;
1255                                 argv++;
1256                                 /*
1257                                  * A log is not a real grouping device.
1258                                  * We just set is_log and continue.
1259                                  */
1260                                 continue;
1261                         }
1262
1263                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1264                                 if (l2cache != NULL) {
1265                                         (void) fprintf(stderr,
1266                                             gettext("invalid vdev "
1267                                             "specification: 'cache' can be "
1268                                             "specified only once\n"));
1269                                         return (NULL);
1270                                 }
1271                                 is_log = B_FALSE;
1272                         }
1273
1274                         if (is_log) {
1275                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1276                                         (void) fprintf(stderr,
1277                                             gettext("invalid vdev "
1278                                             "specification: unsupported 'log' "
1279                                             "device: %s\n"), type);
1280                                         return (NULL);
1281                                 }
1282                                 nlogs++;
1283                         }
1284
1285                         for (c = 1; c < argc; c++) {
1286                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1287                                         break;
1288                                 children++;
1289                                 child = realloc(child,
1290                                     children * sizeof (nvlist_t *));
1291                                 if (child == NULL)
1292                                         zpool_no_memory();
1293                                 if ((nv = make_leaf_vdev(argv[c], B_FALSE))
1294                                     == NULL)
1295                                         return (NULL);
1296                                 child[children - 1] = nv;
1297                         }
1298
1299                         if (children < mindev) {
1300                                 (void) fprintf(stderr, gettext("invalid vdev "
1301                                     "specification: %s requires at least %d "
1302                                     "devices\n"), argv[0], mindev);
1303                                 return (NULL);
1304                         }
1305
1306                         if (children > maxdev) {
1307                                 (void) fprintf(stderr, gettext("invalid vdev "
1308                                     "specification: %s supports no more than "
1309                                     "%d devices\n"), argv[0], maxdev);
1310                                 return (NULL);
1311                         }
1312
1313                         argc -= c;
1314                         argv += c;
1315
1316                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1317                                 spares = child;
1318                                 nspares = children;
1319                                 continue;
1320                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1321                                 l2cache = child;
1322                                 nl2cache = children;
1323                                 continue;
1324                         } else {
1325                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1326                                     0) == 0);
1327                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1328                                     type) == 0);
1329                                 verify(nvlist_add_uint64(nv,
1330                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1331                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1332                                         verify(nvlist_add_uint64(nv,
1333                                             ZPOOL_CONFIG_NPARITY,
1334                                             mindev - 1) == 0);
1335                                 }
1336                                 verify(nvlist_add_nvlist_array(nv,
1337                                     ZPOOL_CONFIG_CHILDREN, child,
1338                                     children) == 0);
1339
1340                                 for (c = 0; c < children; c++)
1341                                         nvlist_free(child[c]);
1342                                 free(child);
1343                         }
1344                 } else {
1345                         /*
1346                          * We have a device.  Pass off to make_leaf_vdev() to
1347                          * construct the appropriate nvlist describing the vdev.
1348                          */
1349                         if ((nv = make_leaf_vdev(argv[0], is_log)) == NULL)
1350                                 return (NULL);
1351                         if (is_log)
1352                                 nlogs++;
1353                         argc--;
1354                         argv++;
1355                 }
1356
1357                 toplevels++;
1358                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1359                 if (top == NULL)
1360                         zpool_no_memory();
1361                 top[toplevels - 1] = nv;
1362         }
1363
1364         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1365                 (void) fprintf(stderr, gettext("invalid vdev "
1366                     "specification: at least one toplevel vdev must be "
1367                     "specified\n"));
1368                 return (NULL);
1369         }
1370
1371         if (seen_logs && nlogs == 0) {
1372                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1373                     "log requires at least 1 device\n"));
1374                 return (NULL);
1375         }
1376
1377         /*
1378          * Finally, create nvroot and add all top-level vdevs to it.
1379          */
1380         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1381         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1382             VDEV_TYPE_ROOT) == 0);
1383         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1384             top, toplevels) == 0);
1385         if (nspares != 0)
1386                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1387                     spares, nspares) == 0);
1388         if (nl2cache != 0)
1389                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1390                     l2cache, nl2cache) == 0);
1391
1392         for (t = 0; t < toplevels; t++)
1393                 nvlist_free(top[t]);
1394         for (t = 0; t < nspares; t++)
1395                 nvlist_free(spares[t]);
1396         for (t = 0; t < nl2cache; t++)
1397                 nvlist_free(l2cache[t]);
1398         if (spares)
1399                 free(spares);
1400         if (l2cache)
1401                 free(l2cache);
1402         free(top);
1403
1404         return (nvroot);
1405 }
1406
1407 nvlist_t *
1408 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1409     splitflags_t flags, int argc, char **argv)
1410 {
1411         nvlist_t *newroot = NULL, **child;
1412         uint_t c, children;
1413
1414         if (argc > 0) {
1415                 if ((newroot = construct_spec(argc, argv)) == NULL) {
1416                         (void) fprintf(stderr, gettext("Unable to build a "
1417                             "pool from the specified devices\n"));
1418                         return (NULL);
1419                 }
1420
1421 #ifdef sun
1422                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1423                         nvlist_free(newroot);
1424                         return (NULL);
1425                 }
1426 #endif
1427
1428                 /* avoid any tricks in the spec */
1429                 verify(nvlist_lookup_nvlist_array(newroot,
1430                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1431                 for (c = 0; c < children; c++) {
1432                         char *path;
1433                         const char *type;
1434                         int min, max;
1435
1436                         verify(nvlist_lookup_string(child[c],
1437                             ZPOOL_CONFIG_PATH, &path) == 0);
1438                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1439                                 (void) fprintf(stderr, gettext("Cannot use "
1440                                     "'%s' as a device for splitting\n"), type);
1441                                 nvlist_free(newroot);
1442                                 return (NULL);
1443                         }
1444                 }
1445         }
1446
1447         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1448                 if (newroot != NULL)
1449                         nvlist_free(newroot);
1450                 return (NULL);
1451         }
1452
1453         return (newroot);
1454 }
1455
1456 /*
1457  * Get and validate the contents of the given vdev specification.  This ensures
1458  * that the nvlist returned is well-formed, that all the devices exist, and that
1459  * they are not currently in use by any other known consumer.  The 'poolconfig'
1460  * parameter is the current configuration of the pool when adding devices
1461  * existing pool, and is used to perform additional checks, such as changing the
1462  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1463  * new pool.  The 'force' flag controls whether devices should be forcefully
1464  * added, even if they appear in use.
1465  */
1466 nvlist_t *
1467 make_root_vdev(zpool_handle_t *zhp, int force, int check_rep,
1468     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1469 {
1470         nvlist_t *newroot;
1471         nvlist_t *poolconfig = NULL;
1472         is_force = force;
1473
1474         /*
1475          * Construct the vdev specification.  If this is successful, we know
1476          * that we have a valid specification, and that all devices can be
1477          * opened.
1478          */
1479         if ((newroot = construct_spec(argc, argv)) == NULL)
1480                 return (NULL);
1481
1482         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
1483                 return (NULL);
1484
1485         /*
1486          * Validate each device to make sure that its not shared with another
1487          * subsystem.  We do this even if 'force' is set, because there are some
1488          * uses (such as a dedicated dump device) that even '-f' cannot
1489          * override.
1490          */
1491         if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1492                 nvlist_free(newroot);
1493                 return (NULL);
1494         }
1495
1496         /*
1497          * Check the replication level of the given vdevs and report any errors
1498          * found.  We include the existing pool spec, if any, as we need to
1499          * catch changes against the existing replication level.
1500          */
1501         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1502                 nvlist_free(newroot);
1503                 return (NULL);
1504         }
1505
1506 #ifdef sun
1507         /*
1508          * Run through the vdev specification and label any whole disks found.
1509          */
1510         if (!dryrun && make_disks(zhp, newroot) != 0) {
1511                 nvlist_free(newroot);
1512                 return (NULL);
1513         }
1514 #endif
1515
1516         return (newroot);
1517 }