granicus.if.org Git - zfs/blob - 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 (c) 2016 Intel Corporation.
  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 libblkid to make 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 <ctype.h>
  66 #include <devid.h>
  67 #include <errno.h>
  68 #include <fcntl.h>
  69 #include <libintl.h>
  70 #include <libnvpair.h>
  71 #include <limits.h>
  72 #include <scsi/scsi.h>
  73 #include <scsi/sg.h>
  74 #include <stdio.h>
  75 #include <string.h>
  76 #include <unistd.h>
  77 #include <sys/efi_partition.h>
  78 #include <sys/stat.h>
  79 #include <sys/vtoc.h>
  80 #include <sys/mntent.h>
  81 #include <uuid/uuid.h>
  82 #include <blkid/blkid.h>
  83 #include "zpool_util.h"
  84 #include <sys/zfs_context.h>
  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 typedef struct vdev_disk_db_entry
  95 {
  96         char id[24];
  97         int sector_size;
  98 } vdev_disk_db_entry_t;
  99
 100 /*
 101  * Database of block devices that lie about physical sector sizes.  The
 102  * identification string must be precisely 24 characters to avoid false
 103  * negatives
 104  */
 105 static vdev_disk_db_entry_t vdev_disk_database[] = {
 106         {"ATA     ADATA SSD S396 3", 8192},
 107         {"ATA     APPLE SSD SM128E", 8192},
 108         {"ATA     APPLE SSD SM256E", 8192},
 109         {"ATA     APPLE SSD SM512E", 8192},
 110         {"ATA     APPLE SSD SM768E", 8192},
 111         {"ATA     C400-MTFDDAC064M", 8192},
 112         {"ATA     C400-MTFDDAC128M", 8192},
 113         {"ATA     C400-MTFDDAC256M", 8192},
 114         {"ATA     C400-MTFDDAC512M", 8192},
 115         {"ATA     Corsair Force 3 ", 8192},
 116         {"ATA     Corsair Force GS", 8192},
 117         {"ATA     INTEL SSDSA2CT04", 8192},
 118         {"ATA     INTEL SSDSA2BZ10", 8192},
 119         {"ATA     INTEL SSDSA2BZ20", 8192},
 120         {"ATA     INTEL SSDSA2BZ30", 8192},
 121         {"ATA     INTEL SSDSA2CW04", 8192},
 122         {"ATA     INTEL SSDSA2CW08", 8192},
 123         {"ATA     INTEL SSDSA2CW12", 8192},
 124         {"ATA     INTEL SSDSA2CW16", 8192},
 125         {"ATA     INTEL SSDSA2CW30", 8192},
 126         {"ATA     INTEL SSDSA2CW60", 8192},
 127         {"ATA     INTEL SSDSC2CT06", 8192},
 128         {"ATA     INTEL SSDSC2CT12", 8192},
 129         {"ATA     INTEL SSDSC2CT18", 8192},
 130         {"ATA     INTEL SSDSC2CT24", 8192},
 131         {"ATA     INTEL SSDSC2CW06", 8192},
 132         {"ATA     INTEL SSDSC2CW12", 8192},
 133         {"ATA     INTEL SSDSC2CW18", 8192},
 134         {"ATA     INTEL SSDSC2CW24", 8192},
 135         {"ATA     INTEL SSDSC2CW48", 8192},
 136         {"ATA     KINGSTON SH100S3", 8192},
 137         {"ATA     KINGSTON SH103S3", 8192},
 138         {"ATA     M4-CT064M4SSD2  ", 8192},
 139         {"ATA     M4-CT128M4SSD2  ", 8192},
 140         {"ATA     M4-CT256M4SSD2  ", 8192},
 141         {"ATA     M4-CT512M4SSD2  ", 8192},
 142         {"ATA     OCZ-AGILITY2    ", 8192},
 143         {"ATA     OCZ-AGILITY3    ", 8192},
 144         {"ATA     OCZ-VERTEX2 3.5 ", 8192},
 145         {"ATA     OCZ-VERTEX3     ", 8192},
 146         {"ATA     OCZ-VERTEX3 LT  ", 8192},
 147         {"ATA     OCZ-VERTEX3 MI  ", 8192},
 148         {"ATA     OCZ-VERTEX4     ", 8192},
 149         {"ATA     SAMSUNG MZ7WD120", 8192},
 150         {"ATA     SAMSUNG MZ7WD240", 8192},
 151         {"ATA     SAMSUNG MZ7WD480", 8192},
 152         {"ATA     SAMSUNG MZ7WD960", 8192},
 153         {"ATA     SAMSUNG SSD 830 ", 8192},
 154         {"ATA     Samsung SSD 840 ", 8192},
 155         {"ATA     SanDisk SSD U100", 8192},
 156         {"ATA     TOSHIBA THNSNH06", 8192},
 157         {"ATA     TOSHIBA THNSNH12", 8192},
 158         {"ATA     TOSHIBA THNSNH25", 8192},
 159         {"ATA     TOSHIBA THNSNH51", 8192},
 160         {"ATA     APPLE SSD TS064C", 4096},
 161         {"ATA     APPLE SSD TS128C", 4096},
 162         {"ATA     APPLE SSD TS256C", 4096},
 163         {"ATA     APPLE SSD TS512C", 4096},
 164         {"ATA     INTEL SSDSA2M040", 4096},
 165         {"ATA     INTEL SSDSA2M080", 4096},
 166         {"ATA     INTEL SSDSA2M160", 4096},
 167         {"ATA     INTEL SSDSC2MH12", 4096},
 168         {"ATA     INTEL SSDSC2MH25", 4096},
 169         {"ATA     OCZ CORE_SSD    ", 4096},
 170         {"ATA     OCZ-VERTEX      ", 4096},
 171         {"ATA     SAMSUNG MCCOE32G", 4096},
 172         {"ATA     SAMSUNG MCCOE64G", 4096},
 173         {"ATA     SAMSUNG SSD PM80", 4096},
 174         /* Flash drives optimized for 4KB IOs on larger pages */
 175         {"ATA     INTEL SSDSC2BA10", 4096},
 176         {"ATA     INTEL SSDSC2BA20", 4096},
 177         {"ATA     INTEL SSDSC2BA40", 4096},
 178         {"ATA     INTEL SSDSC2BA80", 4096},
 179         {"ATA     INTEL SSDSC2BB08", 4096},
 180         {"ATA     INTEL SSDSC2BB12", 4096},
 181         {"ATA     INTEL SSDSC2BB16", 4096},
 182         {"ATA     INTEL SSDSC2BB24", 4096},
 183         {"ATA     INTEL SSDSC2BB30", 4096},
 184         {"ATA     INTEL SSDSC2BB40", 4096},
 185         {"ATA     INTEL SSDSC2BB48", 4096},
 186         {"ATA     INTEL SSDSC2BB60", 4096},
 187         {"ATA     INTEL SSDSC2BB80", 4096},
 188         {"ATA     INTEL SSDSC2BW24", 4096},
 189         {"ATA     INTEL SSDSC2BP24", 4096},
 190         {"ATA     INTEL SSDSC2BP48", 4096},
 191         {"NA      SmrtStorSDLKAE9W", 4096},
 192         /* Imported from Open Solaris */
 193         {"ATA     MARVELL SD88SA02", 4096},
 194         /* Advanced format Hard drives */
 195         {"ATA     Hitachi HDS5C303", 4096},
 196         {"ATA     SAMSUNG HD204UI ", 4096},
 197         {"ATA     ST2000DL004 HD20", 4096},
 198         {"ATA     WDC WD10EARS-00M", 4096},
 199         {"ATA     WDC WD10EARS-00S", 4096},
 200         {"ATA     WDC WD10EARS-00Z", 4096},
 201         {"ATA     WDC WD15EARS-00M", 4096},
 202         {"ATA     WDC WD15EARS-00S", 4096},
 203         {"ATA     WDC WD15EARS-00Z", 4096},
 204         {"ATA     WDC WD20EARS-00M", 4096},
 205         {"ATA     WDC WD20EARS-00S", 4096},
 206         {"ATA     WDC WD20EARS-00Z", 4096},
 207         {"ATA     WDC WD1600BEVT-0", 4096},
 208         {"ATA     WDC WD2500BEVT-0", 4096},
 209         {"ATA     WDC WD3200BEVT-0", 4096},
 210         {"ATA     WDC WD5000BEVT-0", 4096},
 211         /* Virtual disks: Assume zvols with default volblocksize */
 212 #if 0
 213         {"ATA     QEMU HARDDISK   ", 8192},
 214         {"IET     VIRTUAL-DISK    ", 8192},
 215         {"OI      COMSTAR         ", 8192},
 216         {"SUN     COMSTAR         ", 8192},
 217         {"NETAPP  LUN             ", 8192},
 218 #endif
 219 };
 220
 221 static const int vdev_disk_database_size =
 222         sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
 223
 224 #define INQ_REPLY_LEN   96
 225 #define INQ_CMD_LEN     6
 226
 227 static boolean_t
 228 check_sector_size_database(char *path, int *sector_size)
 229 {
 230         unsigned char inq_buff[INQ_REPLY_LEN];
 231         unsigned char sense_buffer[32];
 232         unsigned char inq_cmd_blk[INQ_CMD_LEN] =
 233             {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
 234         sg_io_hdr_t io_hdr;
 235         int error;
 236         int fd;
 237         int i;
 238
 239         /* Prepare INQUIRY command */
 240         memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
 241         io_hdr.interface_id = 'S';
 242         io_hdr.cmd_len = sizeof (inq_cmd_blk);
 243         io_hdr.mx_sb_len = sizeof (sense_buffer);
 244         io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 245         io_hdr.dxfer_len = INQ_REPLY_LEN;
 246         io_hdr.dxferp = inq_buff;
 247         io_hdr.cmdp = inq_cmd_blk;
 248         io_hdr.sbp = sense_buffer;
 249         io_hdr.timeout = 10;            /* 10 milliseconds is ample time */
 250
 251         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 252                 return (B_FALSE);
 253
 254         error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
 255
 256         (void) close(fd);
 257
 258         if (error < 0)
 259                 return (B_FALSE);
 260
 261         if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
 262                 return (B_FALSE);
 263
 264         for (i = 0; i < vdev_disk_database_size; i++) {
 265                 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
 266                         continue;
 267
 268                 *sector_size = vdev_disk_database[i].sector_size;
 269                 return (B_TRUE);
 270         }
 271
 272         return (B_FALSE);
 273 }
 274
 275 /*PRINTFLIKE1*/
 276 static void
 277 vdev_error(const char *fmt, ...)
 278 {
 279         va_list ap;
 280
 281         if (!error_seen) {
 282                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
 283                 if (!is_force)
 284                         (void) fprintf(stderr, gettext("use '-f' to override "
 285                             "the following errors:\n"));
 286                 else
 287                         (void) fprintf(stderr, gettext("the following errors "
 288                             "must be manually repaired:\n"));
 289                 error_seen = B_TRUE;
 290         }
 291
 292         va_start(ap, fmt);
 293         (void) vfprintf(stderr, fmt, ap);
 294         va_end(ap);
 295 }
 296
 297 /*
 298  * Check that a file is valid.  All we can do in this case is check that it's
 299  * not in use by another pool, and not in use by swap.
 300  */
 301 static int
 302 check_file(const char *file, boolean_t force, boolean_t isspare)
 303 {
 304         char  *name;
 305         int fd;
 306         int ret = 0;
 307         pool_state_t state;
 308         boolean_t inuse;
 309
 310         if ((fd = open(file, O_RDONLY)) < 0)
 311                 return (0);
 312
 313         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
 314                 const char *desc;
 315
 316                 switch (state) {
 317                 case POOL_STATE_ACTIVE:
 318                         desc = gettext("active");
 319                         break;
 320
 321                 case POOL_STATE_EXPORTED:
 322                         desc = gettext("exported");
 323                         break;
 324
 325                 case POOL_STATE_POTENTIALLY_ACTIVE:
 326                         desc = gettext("potentially active");
 327                         break;
 328
 329                 default:
 330                         desc = gettext("unknown");
 331                         break;
 332                 }
 333
 334                 /*
 335                  * Allow hot spares to be shared between pools.
 336                  */
 337                 if (state == POOL_STATE_SPARE && isspare)
 338                         return (0);
 339
 340                 if (state == POOL_STATE_ACTIVE ||
 341                     state == POOL_STATE_SPARE || !force) {
 342                         switch (state) {
 343                         case POOL_STATE_SPARE:
 344                                 vdev_error(gettext("%s is reserved as a hot "
 345                                     "spare for pool %s\n"), file, name);
 346                                 break;
 347                         default:
 348                                 vdev_error(gettext("%s is part of %s pool "
 349                                     "'%s'\n"), file, desc, name);
 350                                 break;
 351                         }
 352                         ret = -1;
 353                 }
 354
 355                 free(name);
 356         }
 357
 358         (void) close(fd);
 359         return (ret);
 360 }
 361
 362 static void
 363 check_error(int err)
 364 {
 365         (void) fprintf(stderr, gettext("warning: device in use checking "
 366             "failed: %s\n"), strerror(err));
 367 }
 368
 369 static int
 370 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
 371 {
 372         int err;
 373         char *value;
 374
 375         /* No valid type detected device is safe to use */
 376         value = blkid_get_tag_value(cache, "TYPE", path);
 377         if (value == NULL)
 378                 return (0);
 379
 380         /*
 381          * If libblkid detects a ZFS device, we check the device
 382          * using check_file() to see if it's safe.  The one safe
 383          * case is a spare device shared between multiple pools.
 384          */
 385         if (strcmp(value, "zfs_member") == 0) {
 386                 err = check_file(path, force, isspare);
 387         } else {
 388                 if (force) {
 389                         err = 0;
 390                 } else {
 391                         err = -1;
 392                         vdev_error(gettext("%s contains a filesystem of "
 393                             "type '%s'\n"), path, value);
 394                 }
 395         }
 396
 397         free(value);
 398
 399         return (err);
 400 }
 401
 402 /*
 403  * Validate that a disk including all partitions are safe to use.
 404  *
 405  * For EFI labeled disks this can done relatively easily with the libefi
 406  * library.  The partition numbers are extracted from the label and used
 407  * to generate the expected /dev/ paths.  Each partition can then be
 408  * checked for conflicts.
 409  *
 410  * For non-EFI labeled disks (MBR/EBR/etc) the same process is possible
 411  * but due to the lack of a readily available libraries this scanning is
 412  * not implemented.  Instead only the device path as given is checked.
 413  */
 414 static int
 415 check_disk(const char *path, blkid_cache cache, int force,
 416     boolean_t isspare, boolean_t iswholedisk)
 417 {
 418         struct dk_gpt *vtoc;
 419         char slice_path[MAXPATHLEN];
 420         int err = 0;
 421         int fd, i;
 422
 423         if (!iswholedisk)
 424                 return (check_slice(path, cache, force, isspare));
 425
 426         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
 427                 check_error(errno);
 428                 return (-1);
 429         }
 430
 431         /*
 432          * Expected to fail for non-EFI labled disks.  Just check the device
 433          * as given and do not attempt to detect and scan partitions.
 434          */
 435         err = efi_alloc_and_read(fd, &vtoc);
 436         if (err) {
 437                 (void) close(fd);
 438                 return (check_slice(path, cache, force, isspare));
 439         }
 440
 441         /*
 442          * The primary efi partition label is damaged however the secondary
 443          * label at the end of the device is intact.  Rather than use this
 444          * label we should play it safe and treat this as a non efi device.
 445          */
 446         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
 447                 efi_free(vtoc);
 448                 (void) close(fd);
 449
 450                 if (force) {
 451                         /* Partitions will now be created using the backup */
 452                         return (0);
 453                 } else {
 454                         vdev_error(gettext("%s contains a corrupt primary "
 455                             "EFI label.\n"), path);
 456                         return (-1);
 457                 }
 458         }
 459
 460         for (i = 0; i < vtoc->efi_nparts; i++) {
 461
 462                 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
 463                     uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
 464                         continue;
 465
 466                 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
 467                         (void) snprintf(slice_path, sizeof (slice_path),
 468                             "%s%s%d", path, "-part", i+1);
 469                 else
 470                         (void) snprintf(slice_path, sizeof (slice_path),
 471                             "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
 472                             "p" : "", i+1);
 473
 474                 err = check_slice(slice_path, cache, force, isspare);
 475                 if (err)
 476                         break;
 477         }
 478
 479         efi_free(vtoc);
 480         (void) close(fd);
 481
 482         return (err);
 483 }
 484
 485 static int
 486 check_device(const char *path, boolean_t force,
 487     boolean_t isspare, boolean_t iswholedisk)
 488 {
 489         static blkid_cache cache = NULL;
 490
 491         /*
 492          * There is no easy way to add a correct blkid_put_cache() call,
 493          * memory will be reclaimed when the command exits.
 494          */
 495         if (cache == NULL) {
 496                 int err;
 497
 498                 if ((err = blkid_get_cache(&cache, NULL)) != 0) {
 499                         check_error(err);
 500                         return (-1);
 501                 }
 502
 503                 if ((err = blkid_probe_all(cache)) != 0) {
 504                         blkid_put_cache(cache);
 505                         check_error(err);
 506                         return (-1);
 507                 }
 508         }
 509
 510         return (check_disk(path, cache, force, isspare, iswholedisk));
 511 }
 512
 513 /*
 514  * By "whole disk" we mean an entire physical disk (something we can
 515  * label, toggle the write cache on, etc.) as opposed to the full
 516  * capacity of a pseudo-device such as lofi or did.  We act as if we
 517  * are labeling the disk, which should be a pretty good test of whether
 518  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 519  * it isn't.
 520  */
 521 static boolean_t
 522 is_whole_disk(const char *path)
 523 {
 524         struct dk_gpt *label;
 525         int fd;
 526
 527         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 528                 return (B_FALSE);
 529         if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
 530                 (void) close(fd);
 531                 return (B_FALSE);
 532         }
 533         efi_free(label);
 534         (void) close(fd);
 535         return (B_TRUE);
 536 }
 537
 538 /*
 539  * This may be a shorthand device path or it could be total gibberish.
 540  * Check to see if it is a known device available in zfs_vdev_paths.
 541  * As part of this check, see if we've been given an entire disk
 542  * (minus the slice number).
 543  */
 544 static int
 545 is_shorthand_path(const char *arg, char *path,
 546     struct stat64 *statbuf, boolean_t *wholedisk)
 547 {
 548         int error;
 549
 550         error = zfs_resolve_shortname(arg, path, MAXPATHLEN);
 551         if (error == 0) {
 552                 *wholedisk = is_whole_disk(path);
 553                 if (*wholedisk || (stat64(path, statbuf) == 0))
 554                         return (0);
 555         }
 556
 557         strlcpy(path, arg, sizeof (path));
 558         memset(statbuf, 0, sizeof (*statbuf));
 559         *wholedisk = B_FALSE;
 560
 561         return (error);
 562 }
 563
 564 /*
 565  * Determine if the given path is a hot spare within the given configuration.
 566  * If no configuration is given we rely solely on the label.
 567  */
 568 static boolean_t
 569 is_spare(nvlist_t *config, const char *path)
 570 {
 571         int fd;
 572         pool_state_t state;
 573         char *name = NULL;
 574         nvlist_t *label;
 575         uint64_t guid, spareguid;
 576         nvlist_t *nvroot;
 577         nvlist_t **spares;
 578         uint_t i, nspares;
 579         boolean_t inuse;
 580
 581         if ((fd = open(path, O_RDONLY)) < 0)
 582                 return (B_FALSE);
 583
 584         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
 585             !inuse ||
 586             state != POOL_STATE_SPARE ||
 587             zpool_read_label(fd, &label, NULL) != 0) {
 588                 free(name);
 589                 (void) close(fd);
 590                 return (B_FALSE);
 591         }
 592         free(name);
 593         (void) close(fd);
 594
 595         if (config == NULL)
 596                 return (B_TRUE);
 597
 598         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
 599         nvlist_free(label);
 600
 601         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 602             &nvroot) == 0);
 603         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 604             &spares, &nspares) == 0) {
 605                 for (i = 0; i < nspares; i++) {
 606                         verify(nvlist_lookup_uint64(spares[i],
 607                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
 608                         if (spareguid == guid)
 609                                 return (B_TRUE);
 610                 }
 611         }
 612
 613         return (B_FALSE);
 614 }
 615
 616 /*
 617  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 618  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 619  * leaf vdev are:
 620  *
 621  *      /dev/xxx        Complete disk path
 622  *      /xxx            Full path to file
 623  *      xxx             Shorthand for <zfs_vdev_paths>/xxx
 624  */
 625 static nvlist_t *
 626 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
 627 {
 628         char path[MAXPATHLEN];
 629         struct stat64 statbuf;
 630         nvlist_t *vdev = NULL;
 631         char *type = NULL;
 632         boolean_t wholedisk = B_FALSE;
 633         uint64_t ashift = 0;
 634         int err;
 635
 636         /*
 637          * Determine what type of vdev this is, and put the full path into
 638          * 'path'.  We detect whether this is a device of file afterwards by
 639          * checking the st_mode of the file.
 640          */
 641         if (arg[0] == '/') {
 642                 /*
 643                  * Complete device or file path.  Exact type is determined by
 644                  * examining the file descriptor afterwards.  Symbolic links
 645                  * are resolved to their real paths for the is_whole_disk()
 646                  * and S_ISBLK/S_ISREG type checks.  However, we are careful
 647                  * to store the given path as ZPOOL_CONFIG_PATH to ensure we
 648                  * can leverage udev's persistent device labels.
 649                  */
 650                 if (realpath(arg, path) == NULL) {
 651                         (void) fprintf(stderr,
 652                             gettext("cannot resolve path '%s'\n"), arg);
 653                         return (NULL);
 654                 }
 655
 656                 wholedisk = is_whole_disk(path);
 657                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 658                         (void) fprintf(stderr,
 659                             gettext("cannot open '%s': %s\n"),
 660                             path, strerror(errno));
 661                         return (NULL);
 662                 }
 663
 664                 /* After is_whole_disk() check restore original passed path */
 665                 strlcpy(path, arg, MAXPATHLEN);
 666         } else {
 667                 err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
 668                 if (err != 0) {
 669                         /*
 670                          * If we got ENOENT, then the user gave us
 671                          * gibberish, so try to direct them with a
 672                          * reasonable error message.  Otherwise,
 673                          * regurgitate strerror() since it's the best we
 674                          * can do.
 675                          */
 676                         if (err == ENOENT) {
 677                                 (void) fprintf(stderr,
 678                                     gettext("cannot open '%s': no such "
 679                                     "device in %s\n"), arg, DISK_ROOT);
 680                                 (void) fprintf(stderr,
 681                                     gettext("must be a full path or "
 682                                     "shorthand device name\n"));
 683                                 return (NULL);
 684                         } else {
 685                                 (void) fprintf(stderr,
 686                                     gettext("cannot open '%s': %s\n"),
 687                                     path, strerror(errno));
 688                                 return (NULL);
 689                         }
 690                 }
 691         }
 692
 693         /*
 694          * Determine whether this is a device or a file.
 695          */
 696         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 697                 type = VDEV_TYPE_DISK;
 698         } else if (S_ISREG(statbuf.st_mode)) {
 699                 type = VDEV_TYPE_FILE;
 700         } else {
 701                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 702                     "block device or regular file\n"), path);
 703                 return (NULL);
 704         }
 705
 706         /*
 707          * Finally, we have the complete device or file, and we know that it is
 708          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 709          * vdevs have a 'path' element, and devices also have a 'devid' element.
 710          */
 711         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 712         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 713         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 714         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 715         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 716                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 717                     (uint64_t)wholedisk) == 0);
 718
 719         /*
 720          * Override defaults if custom properties are provided.
 721          */
 722         if (props != NULL) {
 723                 char *value = NULL;
 724
 725                 if (nvlist_lookup_string(props,
 726                     zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
 727                         zfs_nicestrtonum(NULL, value, &ashift);
 728         }
 729
 730         /*
 731          * If the device is known to incorrectly report its physical sector
 732          * size explicitly provide the known correct value.
 733          */
 734         if (ashift == 0) {
 735                 int sector_size;
 736
 737                 if (check_sector_size_database(path, &sector_size) == B_TRUE)
 738                         ashift = highbit64(sector_size) - 1;
 739         }
 740
 741         if (ashift > 0)
 742                 nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
 743
 744         return (vdev);
 745 }
 746
 747 /*
 748  * Go through and verify the replication level of the pool is consistent.
 749  * Performs the following checks:
 750  *
 751  *      For the new spec, verifies that devices in mirrors and raidz are the
 752  *      same size.
 753  *
 754  *      If the current configuration already has inconsistent replication
 755  *      levels, ignore any other potential problems in the new spec.
 756  *
 757  *      Otherwise, make sure that the current spec (if there is one) and the new
 758  *      spec have consistent replication levels.
 759  */
 760 typedef struct replication_level {
 761         char *zprl_type;
 762         uint64_t zprl_children;
 763         uint64_t zprl_parity;
 764 } replication_level_t;
 765
 766 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 767
 768 /*
 769  * Given a list of toplevel vdevs, return the current replication level.  If
 770  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 771  * an error message will be displayed for each self-inconsistent vdev.
 772  */
 773 static replication_level_t *
 774 get_replication(nvlist_t *nvroot, boolean_t fatal)
 775 {
 776         nvlist_t **top;
 777         uint_t t, toplevels;
 778         nvlist_t **child;
 779         uint_t c, children;
 780         nvlist_t *nv;
 781         char *type;
 782         replication_level_t lastrep = { 0 }, rep, *ret;
 783         boolean_t dontreport;
 784
 785         ret = safe_malloc(sizeof (replication_level_t));
 786
 787         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 788             &top, &toplevels) == 0);
 789
 790         lastrep.zprl_type = NULL;
 791         for (t = 0; t < toplevels; t++) {
 792                 uint64_t is_log = B_FALSE;
 793
 794                 nv = top[t];
 795
 796                 /*
 797                  * For separate logs we ignore the top level vdev replication
 798                  * constraints.
 799                  */
 800                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 801                 if (is_log)
 802                         continue;
 803
 804                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 805                     &type) == 0);
 806                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 807                     &child, &children) != 0) {
 808                         /*
 809                          * This is a 'file' or 'disk' vdev.
 810                          */
 811                         rep.zprl_type = type;
 812                         rep.zprl_children = 1;
 813                         rep.zprl_parity = 0;
 814                 } else {
 815                         uint64_t vdev_size;
 816
 817                         /*
 818                          * This is a mirror or RAID-Z vdev.  Go through and make
 819                          * sure the contents are all the same (files vs. disks),
 820                          * keeping track of the number of elements in the
 821                          * process.
 822                          *
 823                          * We also check that the size of each vdev (if it can
 824                          * be determined) is the same.
 825                          */
 826                         rep.zprl_type = type;
 827                         rep.zprl_children = 0;
 828
 829                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 830                                 verify(nvlist_lookup_uint64(nv,
 831                                     ZPOOL_CONFIG_NPARITY,
 832                                     &rep.zprl_parity) == 0);
 833                                 assert(rep.zprl_parity != 0);
 834                         } else {
 835                                 rep.zprl_parity = 0;
 836                         }
 837
 838                         /*
 839                          * The 'dontreport' variable indicates that we've
 840                          * already reported an error for this spec, so don't
 841                          * bother doing it again.
 842                          */
 843                         type = NULL;
 844                         dontreport = 0;
 845                         vdev_size = -1ULL;
 846                         for (c = 0; c < children; c++) {
 847                                 nvlist_t *cnv = child[c];
 848                                 char *path;
 849                                 struct stat64 statbuf;
 850                                 uint64_t size = -1ULL;
 851                                 char *childtype;
 852                                 int fd, err;
 853
 854                                 rep.zprl_children++;
 855
 856                                 verify(nvlist_lookup_string(cnv,
 857                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 858
 859                                 /*
 860                                  * If this is a replacing or spare vdev, then
 861                                  * get the real first child of the vdev.
 862                                  */
 863                                 if (strcmp(childtype,
 864                                     VDEV_TYPE_REPLACING) == 0 ||
 865                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 866                                         nvlist_t **rchild;
 867                                         uint_t rchildren;
 868
 869                                         verify(nvlist_lookup_nvlist_array(cnv,
 870                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 871                                             &rchildren) == 0);
 872                                         assert(rchildren == 2);
 873                                         cnv = rchild[0];
 874
 875                                         verify(nvlist_lookup_string(cnv,
 876                                             ZPOOL_CONFIG_TYPE,
 877                                             &childtype) == 0);
 878                                 }
 879
 880                                 verify(nvlist_lookup_string(cnv,
 881                                     ZPOOL_CONFIG_PATH, &path) == 0);
 882
 883                                 /*
 884                                  * If we have a raidz/mirror that combines disks
 885                                  * with files, report it as an error.
 886                                  */
 887                                 if (!dontreport && type != NULL &&
 888                                     strcmp(type, childtype) != 0) {
 889                                         if (ret != NULL)
 890                                                 free(ret);
 891                                         ret = NULL;
 892                                         if (fatal)
 893                                                 vdev_error(gettext(
 894                                                     "mismatched replication "
 895                                                     "level: %s contains both "
 896                                                     "files and devices\n"),
 897                                                     rep.zprl_type);
 898                                         else
 899                                                 return (NULL);
 900                                         dontreport = B_TRUE;
 901                                 }
 902
 903                                 /*
 904                                  * According to stat(2), the value of 'st_size'
 905                                  * is undefined for block devices and character
 906                                  * devices.  But there is no effective way to
 907                                  * determine the real size in userland.
 908                                  *
 909                                  * Instead, we'll take advantage of an
 910                                  * implementation detail of spec_size().  If the
 911                                  * device is currently open, then we (should)
 912                                  * return a valid size.
 913                                  *
 914                                  * If we still don't get a valid size (indicated
 915                                  * by a size of 0 or MAXOFFSET_T), then ignore
 916                                  * this device altogether.
 917                                  */
 918                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 919                                         err = fstat64(fd, &statbuf);
 920                                         (void) close(fd);
 921                                 } else {
 922                                         err = stat64(path, &statbuf);
 923                                 }
 924
 925                                 if (err != 0 ||
 926                                     statbuf.st_size == 0 ||
 927                                     statbuf.st_size == MAXOFFSET_T)
 928                                         continue;
 929
 930                                 size = statbuf.st_size;
 931
 932                                 /*
 933                                  * Also make sure that devices and
 934                                  * slices have a consistent size.  If
 935                                  * they differ by a significant amount
 936                                  * (~16MB) then report an error.
 937                                  */
 938                                 if (!dontreport &&
 939                                     (vdev_size != -1ULL &&
 940                                     (labs(size - vdev_size) >
 941                                     ZPOOL_FUZZ))) {
 942                                         if (ret != NULL)
 943                                                 free(ret);
 944                                         ret = NULL;
 945                                         if (fatal)
 946                                                 vdev_error(gettext(
 947                                                     "%s contains devices of "
 948                                                     "different sizes\n"),
 949                                                     rep.zprl_type);
 950                                         else
 951                                                 return (NULL);
 952                                         dontreport = B_TRUE;
 953                                 }
 954
 955                                 type = childtype;
 956                                 vdev_size = size;
 957                         }
 958                 }
 959
 960                 /*
 961                  * At this point, we have the replication of the last toplevel
 962                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 963                  * different.
 964                  */
 965                 if (lastrep.zprl_type != NULL) {
 966                         if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
 967                                 if (ret != NULL)
 968                                         free(ret);
 969                                 ret = NULL;
 970                                 if (fatal)
 971                                         vdev_error(gettext(
 972                                             "mismatched replication level: "
 973                                             "both %s and %s vdevs are "
 974                                             "present\n"),
 975                                             lastrep.zprl_type, rep.zprl_type);
 976                                 else
 977                                         return (NULL);
 978                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 979                                 if (ret)
 980                                         free(ret);
 981                                 ret = NULL;
 982                                 if (fatal)
 983                                         vdev_error(gettext(
 984                                             "mismatched replication level: "
 985                                             "both %llu and %llu device parity "
 986                                             "%s vdevs are present\n"),
 987                                             lastrep.zprl_parity,
 988                                             rep.zprl_parity,
 989                                             rep.zprl_type);
 990                                 else
 991                                         return (NULL);
 992                         } else if (lastrep.zprl_children != rep.zprl_children) {
 993                                 if (ret)
 994                                         free(ret);
 995                                 ret = NULL;
 996                                 if (fatal)
 997                                         vdev_error(gettext(
 998                                             "mismatched replication level: "
 999                                             "both %llu-way and %llu-way %s "
1000                                             "vdevs are present\n"),
1001                                             lastrep.zprl_children,
1002                                             rep.zprl_children,
1003                                             rep.zprl_type);
1004                                 else
1005                                         return (NULL);
1006                         }
1007                 }
1008                 lastrep = rep;
1009         }
1010
1011         if (ret != NULL)
1012                 *ret = rep;
1013
1014         return (ret);
1015 }
1016
1017 /*
1018  * Check the replication level of the vdev spec against the current pool.  Calls
1019  * get_replication() to make sure the new spec is self-consistent.  If the pool
1020  * has a consistent replication level, then we ignore any errors.  Otherwise,
1021  * report any difference between the two.
1022  */
1023 static int
1024 check_replication(nvlist_t *config, nvlist_t *newroot)
1025 {
1026         nvlist_t **child;
1027         uint_t  children;
1028         replication_level_t *current = NULL, *new;
1029         int ret;
1030
1031         /*
1032          * If we have a current pool configuration, check to see if it's
1033          * self-consistent.  If not, simply return success.
1034          */
1035         if (config != NULL) {
1036                 nvlist_t *nvroot;
1037
1038                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1039                     &nvroot) == 0);
1040                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1041                         return (0);
1042         }
1043         /*
1044          * for spares there may be no children, and therefore no
1045          * replication level to check
1046          */
1047         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1048             &child, &children) != 0) || (children == 0)) {
1049                 free(current);
1050                 return (0);
1051         }
1052
1053         /*
1054          * If all we have is logs then there's no replication level to check.
1055          */
1056         if (num_logs(newroot) == children) {
1057                 free(current);
1058                 return (0);
1059         }
1060
1061         /*
1062          * Get the replication level of the new vdev spec, reporting any
1063          * inconsistencies found.
1064          */
1065         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1066                 free(current);
1067                 return (-1);
1068         }
1069
1070         /*
1071          * Check to see if the new vdev spec matches the replication level of
1072          * the current pool.
1073          */
1074         ret = 0;
1075         if (current != NULL) {
1076                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1077                         vdev_error(gettext(
1078                             "mismatched replication level: pool uses %s "
1079                             "and new vdev is %s\n"),
1080                             current->zprl_type, new->zprl_type);
1081                         ret = -1;
1082                 } else if (current->zprl_parity != new->zprl_parity) {
1083                         vdev_error(gettext(
1084                             "mismatched replication level: pool uses %llu "
1085                             "device parity and new vdev uses %llu\n"),
1086                             current->zprl_parity, new->zprl_parity);
1087                         ret = -1;
1088                 } else if (current->zprl_children != new->zprl_children) {
1089                         vdev_error(gettext(
1090                             "mismatched replication level: pool uses %llu-way "
1091                             "%s and new vdev uses %llu-way %s\n"),
1092                             current->zprl_children, current->zprl_type,
1093                             new->zprl_children, new->zprl_type);
1094                         ret = -1;
1095                 }
1096         }
1097
1098         free(new);
1099         if (current != NULL)
1100                 free(current);
1101
1102         return (ret);
1103 }
1104
1105 static int
1106 zero_label(char *path)
1107 {
1108         const int size = 4096;
1109         char buf[size];
1110         int err, fd;
1111
1112         if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1113                 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1114                     path, strerror(errno));
1115                 return (-1);
1116         }
1117
1118         memset(buf, 0, size);
1119         err = write(fd, buf, size);
1120         (void) fdatasync(fd);
1121         (void) close(fd);
1122
1123         if (err == -1) {
1124                 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1125                     "of '%s': %s\n"), size, path, strerror(errno));
1126                 return (-1);
1127         }
1128
1129         if (err != size) {
1130                 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1131                     "of '%s'\n"), err, size, path);
1132                 return (-1);
1133         }
1134
1135         return (0);
1136 }
1137
1138 /*
1139  * Go through and find any whole disks in the vdev specification, labelling them
1140  * as appropriate.  When constructing the vdev spec, we were unable to open this
1141  * device in order to provide a devid.  Now that we have labelled the disk and
1142  * know that slice 0 is valid, we can construct the devid now.
1143  *
1144  * If the disk was already labeled with an EFI label, we will have gotten the
1145  * devid already (because we were able to open the whole disk).  Otherwise, we
1146  * need to get the devid after we label the disk.
1147  */
1148 static int
1149 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1150 {
1151         nvlist_t **child;
1152         uint_t c, children;
1153         char *type, *path;
1154         char devpath[MAXPATHLEN];
1155         char udevpath[MAXPATHLEN];
1156         uint64_t wholedisk;
1157         struct stat64 statbuf;
1158         int is_exclusive = 0;
1159         int fd;
1160         int ret;
1161
1162         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1163
1164         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1165             &child, &children) != 0) {
1166
1167                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1168                         return (0);
1169
1170                 /*
1171                  * We have a disk device.  If this is a whole disk write
1172                  * out the efi partition table, otherwise write zero's to
1173                  * the first 4k of the partition.  This is to ensure that
1174                  * libblkid will not misidentify the partition due to a
1175                  * magic value left by the previous filesystem.
1176                  */
1177                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1178                 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1179                     &wholedisk));
1180
1181                 if (!wholedisk) {
1182                         /*
1183                          * Update device id string for mpath nodes (Linux only)
1184                          */
1185                         if (is_mpath_whole_disk(path))
1186                                 update_vdev_config_dev_strs(nv);
1187
1188                         (void) zero_label(path);
1189                         return (0);
1190                 }
1191
1192                 if (realpath(path, devpath) == NULL) {
1193                         ret = errno;
1194                         (void) fprintf(stderr,
1195                             gettext("cannot resolve path '%s'\n"), path);
1196                         return (ret);
1197                 }
1198
1199                 /*
1200                  * Remove any previously existing symlink from a udev path to
1201                  * the device before labeling the disk.  This ensures that
1202                  * only newly created links are used.  Otherwise there is a
1203                  * window between when udev deletes and recreates the link
1204                  * during which access attempts will fail with ENOENT.
1205                  */
1206                 strncpy(udevpath, path, MAXPATHLEN);
1207                 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1208
1209                 fd = open(devpath, O_RDWR|O_EXCL);
1210                 if (fd == -1) {
1211                         if (errno == EBUSY)
1212                                 is_exclusive = 1;
1213                 } else {
1214                         (void) close(fd);
1215                 }
1216
1217                 /*
1218                  * If the partition exists, contains a valid spare label,
1219                  * and is opened exclusively there is no need to partition
1220                  * it.  Hot spares have already been partitioned and are
1221                  * held open exclusively by the kernel as a safety measure.
1222                  *
1223                  * If the provided path is for a /dev/disk/ device its
1224                  * symbolic link will be removed, partition table created,
1225                  * and then block until udev creates the new link.
1226                  */
1227                 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1228                         char *devnode = strrchr(devpath, '/') + 1;
1229
1230                         ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1231                         if (ret == 0) {
1232                                 ret = lstat64(udevpath, &statbuf);
1233                                 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1234                                         (void) unlink(udevpath);
1235                         }
1236
1237                         /*
1238                          * When labeling a pool the raw device node name
1239                          * is provided as it appears under /dev/.
1240                          */
1241                         if (zpool_label_disk(g_zfs, zhp, devnode) == -1)
1242                                 return (-1);
1243
1244                         /*
1245                          * Wait for udev to signal the device is available
1246                          * by the provided path.
1247                          */
1248                         ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1249                         if (ret) {
1250                                 (void) fprintf(stderr,
1251                                     gettext("missing link: %s was "
1252                                     "partitioned but %s is missing\n"),
1253                                     devnode, udevpath);
1254                                 return (ret);
1255                         }
1256
1257                         ret = zero_label(udevpath);
1258                         if (ret)
1259                                 return (ret);
1260                 }
1261
1262                 /*
1263                  * Update the path to refer to the partition.  The presence of
1264                  * the 'whole_disk' field indicates to the CLI that we should
1265                  * chop off the partition number when displaying the device in
1266                  * future output.
1267                  */
1268                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1269
1270                 /*
1271                  * Update device id strings for whole disks (Linux only)
1272                  */
1273                 update_vdev_config_dev_strs(nv);
1274
1275                 return (0);
1276         }
1277
1278         for (c = 0; c < children; c++)
1279                 if ((ret = make_disks(zhp, child[c])) != 0)
1280                         return (ret);
1281
1282         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1283             &child, &children) == 0)
1284                 for (c = 0; c < children; c++)
1285                         if ((ret = make_disks(zhp, child[c])) != 0)
1286                                 return (ret);
1287
1288         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1289             &child, &children) == 0)
1290                 for (c = 0; c < children; c++)
1291                         if ((ret = make_disks(zhp, child[c])) != 0)
1292                                 return (ret);
1293
1294         return (0);
1295 }
1296
1297 /*
1298  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1299  * the majority of this task.
1300  */
1301 static boolean_t
1302 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1303     boolean_t replacing, boolean_t isspare)
1304 {
1305         nvlist_t **child;
1306         uint_t c, children;
1307         char *type, *path;
1308         int ret = 0;
1309         char buf[MAXPATHLEN];
1310         uint64_t wholedisk = B_FALSE;
1311         boolean_t anyinuse = B_FALSE;
1312
1313         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1314
1315         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1316             &child, &children) != 0) {
1317
1318                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1319                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1320                         verify(!nvlist_lookup_uint64(nv,
1321                             ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1322
1323                 /*
1324                  * As a generic check, we look to see if this is a replace of a
1325                  * hot spare within the same pool.  If so, we allow it
1326                  * regardless of what libblkid or zpool_in_use() says.
1327                  */
1328                 if (replacing) {
1329                         (void) strlcpy(buf, path, sizeof (buf));
1330                         if (wholedisk) {
1331                                 ret = zfs_append_partition(buf,  sizeof (buf));
1332                                 if (ret == -1)
1333                                         return (-1);
1334                         }
1335
1336                         if (is_spare(config, buf))
1337                                 return (B_FALSE);
1338                 }
1339
1340                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1341                         ret = check_device(path, force, isspare, wholedisk);
1342
1343                 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1344                         ret = check_file(path, force, isspare);
1345
1346                 return (ret != 0);
1347         }
1348
1349         for (c = 0; c < children; c++)
1350                 if (is_device_in_use(config, child[c], force, replacing,
1351                     B_FALSE))
1352                         anyinuse = B_TRUE;
1353
1354         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1355             &child, &children) == 0)
1356                 for (c = 0; c < children; c++)
1357                         if (is_device_in_use(config, child[c], force, replacing,
1358                             B_TRUE))
1359                                 anyinuse = B_TRUE;
1360
1361         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1362             &child, &children) == 0)
1363                 for (c = 0; c < children; c++)
1364                         if (is_device_in_use(config, child[c], force, replacing,
1365                             B_FALSE))
1366                                 anyinuse = B_TRUE;
1367
1368         return (anyinuse);
1369 }
1370
1371 static const char *
1372 is_grouping(const char *type, int *mindev, int *maxdev)
1373 {
1374         if (strncmp(type, "raidz", 5) == 0) {
1375                 const char *p = type + 5;
1376                 char *end;
1377                 long nparity;
1378
1379                 if (*p == '\0') {
1380                         nparity = 1;
1381                 } else if (*p == '0') {
1382                         return (NULL); /* no zero prefixes allowed */
1383                 } else {
1384                         errno = 0;
1385                         nparity = strtol(p, &end, 10);
1386                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1387                             *end != '\0')
1388                                 return (NULL);
1389                 }
1390
1391                 if (mindev != NULL)
1392                         *mindev = nparity + 1;
1393                 if (maxdev != NULL)
1394                         *maxdev = 255;
1395                 return (VDEV_TYPE_RAIDZ);
1396         }
1397
1398         if (maxdev != NULL)
1399                 *maxdev = INT_MAX;
1400
1401         if (strcmp(type, "mirror") == 0) {
1402                 if (mindev != NULL)
1403                         *mindev = 2;
1404                 return (VDEV_TYPE_MIRROR);
1405         }
1406
1407         if (strcmp(type, "spare") == 0) {
1408                 if (mindev != NULL)
1409                         *mindev = 1;
1410                 return (VDEV_TYPE_SPARE);
1411         }
1412
1413         if (strcmp(type, "log") == 0) {
1414                 if (mindev != NULL)
1415                         *mindev = 1;
1416                 return (VDEV_TYPE_LOG);
1417         }
1418
1419         if (strcmp(type, "cache") == 0) {
1420                 if (mindev != NULL)
1421                         *mindev = 1;
1422                 return (VDEV_TYPE_L2CACHE);
1423         }
1424
1425         return (NULL);
1426 }
1427
1428 /*
1429  * Construct a syntactically valid vdev specification,
1430  * and ensure that all devices and files exist and can be opened.
1431  * Note: we don't bother freeing anything in the error paths
1432  * because the program is just going to exit anyway.
1433  */
1434 nvlist_t *
1435 construct_spec(nvlist_t *props, int argc, char **argv)
1436 {
1437         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1438         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1439         const char *type;
1440         uint64_t is_log;
1441         boolean_t seen_logs;
1442
1443         top = NULL;
1444         toplevels = 0;
1445         spares = NULL;
1446         l2cache = NULL;
1447         nspares = 0;
1448         nlogs = 0;
1449         nl2cache = 0;
1450         is_log = B_FALSE;
1451         seen_logs = B_FALSE;
1452
1453         while (argc > 0) {
1454                 nv = NULL;
1455
1456                 /*
1457                  * If it's a mirror or raidz, the subsequent arguments are
1458                  * its leaves -- until we encounter the next mirror or raidz.
1459                  */
1460                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1461                         nvlist_t **child = NULL;
1462                         int c, children = 0;
1463
1464                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1465                                 if (spares != NULL) {
1466                                         (void) fprintf(stderr,
1467                                             gettext("invalid vdev "
1468                                             "specification: 'spare' can be "
1469                                             "specified only once\n"));
1470                                         return (NULL);
1471                                 }
1472                                 is_log = B_FALSE;
1473                         }
1474
1475                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1476                                 if (seen_logs) {
1477                                         (void) fprintf(stderr,
1478                                             gettext("invalid vdev "
1479                                             "specification: 'log' can be "
1480                                             "specified only once\n"));
1481                                         return (NULL);
1482                                 }
1483                                 seen_logs = B_TRUE;
1484                                 is_log = B_TRUE;
1485                                 argc--;
1486                                 argv++;
1487                                 /*
1488                                  * A log is not a real grouping device.
1489                                  * We just set is_log and continue.
1490                                  */
1491                                 continue;
1492                         }
1493
1494                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1495                                 if (l2cache != NULL) {
1496                                         (void) fprintf(stderr,
1497                                             gettext("invalid vdev "
1498                                             "specification: 'cache' can be "
1499                                             "specified only once\n"));
1500                                         return (NULL);
1501                                 }
1502                                 is_log = B_FALSE;
1503                         }
1504
1505                         if (is_log) {
1506                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1507                                         (void) fprintf(stderr,
1508                                             gettext("invalid vdev "
1509                                             "specification: unsupported 'log' "
1510                                             "device: %s\n"), type);
1511                                         return (NULL);
1512                                 }
1513                                 nlogs++;
1514                         }
1515
1516                         for (c = 1; c < argc; c++) {
1517                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1518                                         break;
1519                                 children++;
1520                                 child = realloc(child,
1521                                     children * sizeof (nvlist_t *));
1522                                 if (child == NULL)
1523                                         zpool_no_memory();
1524                                 if ((nv = make_leaf_vdev(props, argv[c],
1525                                     B_FALSE)) == NULL)
1526                                         return (NULL);
1527                                 child[children - 1] = nv;
1528                         }
1529
1530                         if (children < mindev) {
1531                                 (void) fprintf(stderr, gettext("invalid vdev "
1532                                     "specification: %s requires at least %d "
1533                                     "devices\n"), argv[0], mindev);
1534                                 return (NULL);
1535                         }
1536
1537                         if (children > maxdev) {
1538                                 (void) fprintf(stderr, gettext("invalid vdev "
1539                                     "specification: %s supports no more than "
1540                                     "%d devices\n"), argv[0], maxdev);
1541                                 return (NULL);
1542                         }
1543
1544                         argc -= c;
1545                         argv += c;
1546
1547                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1548                                 spares = child;
1549                                 nspares = children;
1550                                 continue;
1551                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1552                                 l2cache = child;
1553                                 nl2cache = children;
1554                                 continue;
1555                         } else {
1556                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1557                                     0) == 0);
1558                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1559                                     type) == 0);
1560                                 verify(nvlist_add_uint64(nv,
1561                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1562                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1563                                         verify(nvlist_add_uint64(nv,
1564                                             ZPOOL_CONFIG_NPARITY,
1565                                             mindev - 1) == 0);
1566                                 }
1567                                 verify(nvlist_add_nvlist_array(nv,
1568                                     ZPOOL_CONFIG_CHILDREN, child,
1569                                     children) == 0);
1570
1571                                 for (c = 0; c < children; c++)
1572                                         nvlist_free(child[c]);
1573                                 free(child);
1574                         }
1575                 } else {
1576                         /*
1577                          * We have a device.  Pass off to make_leaf_vdev() to
1578                          * construct the appropriate nvlist describing the vdev.
1579                          */
1580                         if ((nv = make_leaf_vdev(props, argv[0],
1581                             is_log)) == NULL)
1582                                 return (NULL);
1583                         if (is_log)
1584                                 nlogs++;
1585                         argc--;
1586                         argv++;
1587                 }
1588
1589                 toplevels++;
1590                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1591                 if (top == NULL)
1592                         zpool_no_memory();
1593                 top[toplevels - 1] = nv;
1594         }
1595
1596         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1597                 (void) fprintf(stderr, gettext("invalid vdev "
1598                     "specification: at least one toplevel vdev must be "
1599                     "specified\n"));
1600                 return (NULL);
1601         }
1602
1603         if (seen_logs && nlogs == 0) {
1604                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1605                     "log requires at least 1 device\n"));
1606                 return (NULL);
1607         }
1608
1609         /*
1610          * Finally, create nvroot and add all top-level vdevs to it.
1611          */
1612         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1613         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1614             VDEV_TYPE_ROOT) == 0);
1615         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1616             top, toplevels) == 0);
1617         if (nspares != 0)
1618                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1619                     spares, nspares) == 0);
1620         if (nl2cache != 0)
1621                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1622                     l2cache, nl2cache) == 0);
1623
1624         for (t = 0; t < toplevels; t++)
1625                 nvlist_free(top[t]);
1626         for (t = 0; t < nspares; t++)
1627                 nvlist_free(spares[t]);
1628         for (t = 0; t < nl2cache; t++)
1629                 nvlist_free(l2cache[t]);
1630         if (spares)
1631                 free(spares);
1632         if (l2cache)
1633                 free(l2cache);
1634         free(top);
1635
1636         return (nvroot);
1637 }
1638
1639 nvlist_t *
1640 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1641     splitflags_t flags, int argc, char **argv)
1642 {
1643         nvlist_t *newroot = NULL, **child;
1644         uint_t c, children;
1645
1646         if (argc > 0) {
1647                 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1648                         (void) fprintf(stderr, gettext("Unable to build a "
1649                             "pool from the specified devices\n"));
1650                         return (NULL);
1651                 }
1652
1653                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1654                         nvlist_free(newroot);
1655                         return (NULL);
1656                 }
1657
1658                 /* avoid any tricks in the spec */
1659                 verify(nvlist_lookup_nvlist_array(newroot,
1660                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1661                 for (c = 0; c < children; c++) {
1662                         char *path;
1663                         const char *type;
1664                         int min, max;
1665
1666                         verify(nvlist_lookup_string(child[c],
1667                             ZPOOL_CONFIG_PATH, &path) == 0);
1668                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1669                                 (void) fprintf(stderr, gettext("Cannot use "
1670                                     "'%s' as a device for splitting\n"), type);
1671                                 nvlist_free(newroot);
1672                                 return (NULL);
1673                         }
1674                 }
1675         }
1676
1677         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1678                 nvlist_free(newroot);
1679                 return (NULL);
1680         }
1681
1682         return (newroot);
1683 }
1684
1685 /*
1686  * Get and validate the contents of the given vdev specification.  This ensures
1687  * that the nvlist returned is well-formed, that all the devices exist, and that
1688  * they are not currently in use by any other known consumer.  The 'poolconfig'
1689  * parameter is the current configuration of the pool when adding devices
1690  * existing pool, and is used to perform additional checks, such as changing the
1691  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1692  * new pool.  The 'force' flag controls whether devices should be forcefully
1693  * added, even if they appear in use.
1694  */
1695 nvlist_t *
1696 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1697     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1698 {
1699         nvlist_t *newroot;
1700         nvlist_t *poolconfig = NULL;
1701         is_force = force;
1702
1703         /*
1704          * Construct the vdev specification.  If this is successful, we know
1705          * that we have a valid specification, and that all devices can be
1706          * opened.
1707          */
1708         if ((newroot = construct_spec(props, argc, argv)) == NULL)
1709                 return (NULL);
1710
1711         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1712                 nvlist_free(newroot);
1713                 return (NULL);
1714         }
1715
1716         /*
1717          * Validate each device to make sure that its not shared with another
1718          * subsystem.  We do this even if 'force' is set, because there are some
1719          * uses (such as a dedicated dump device) that even '-f' cannot
1720          * override.
1721          */
1722         if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1723                 nvlist_free(newroot);
1724                 return (NULL);
1725         }
1726
1727         /*
1728          * Check the replication level of the given vdevs and report any errors
1729          * found.  We include the existing pool spec, if any, as we need to
1730          * catch changes against the existing replication level.
1731          */
1732         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1733                 nvlist_free(newroot);
1734                 return (NULL);
1735         }
1736
1737         /*
1738          * Run through the vdev specification and label any whole disks found.
1739          */
1740         if (!dryrun && make_disks(zhp, newroot) != 0) {
1741                 nvlist_free(newroot);
1742                 return (NULL);
1743         }
1744
1745         return (newroot);
1746 }