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, 2015 by Delphix. All rights reserved.
  25  * Copyright (c) 2016 Intel Corporation.
  26  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>.
  27  */
  28
  29 /*
  30  * Functions to convert between a list of vdevs and an nvlist representing the
  31  * configuration.  Each entry in the list can be one of:
  32  *
  33  *      Device vdevs
  34  *              disk=(path=..., devid=...)
  35  *              file=(path=...)
  36  *
  37  *      Group vdevs
  38  *              raidz[1|2]=(...)
  39  *              mirror=(...)
  40  *
  41  *      Hot spares
  42  *
  43  * While the underlying implementation supports it, group vdevs cannot contain
  44  * other group vdevs.  All userland verification of devices is contained within
  45  * this file.  If successful, the nvlist returned can be passed directly to the
  46  * kernel; we've done as much verification as possible in userland.
  47  *
  48  * Hot spares are a special case, and passed down as an array of disk vdevs, at
  49  * the same level as the root of the vdev tree.
  50  *
  51  * The only function exported by this file is 'make_root_vdev'.  The
  52  * function performs several passes:
  53  *
  54  *      1. Construct the vdev specification.  Performs syntax validation and
  55  *         makes sure each device is valid.
  56  *      2. Check for devices in use.  Using libblkid to make sure that no
  57  *         devices are also in use.  Some can be overridden using the 'force'
  58  *         flag, others cannot.
  59  *      3. Check for replication errors if the 'force' flag is not specified.
  60  *         validates that the replication level is consistent across the
  61  *         entire pool.
  62  *      4. Call libzfs to label any whole disks with an EFI label.
  63  */
  64
  65 #include <assert.h>
  66 #include <ctype.h>
  67 #include <devid.h>
  68 #include <errno.h>
  69 #include <fcntl.h>
  70 #include <libintl.h>
  71 #include <libnvpair.h>
  72 #include <limits.h>
  73 #include <sys/spa.h>
  74 #include <scsi/scsi.h>
  75 #include <scsi/sg.h>
  76 #include <stdio.h>
  77 #include <string.h>
  78 #include <unistd.h>
  79 #include <sys/efi_partition.h>
  80 #include <sys/stat.h>
  81 #include <sys/vtoc.h>
  82 #include <sys/mntent.h>
  83 #include <uuid/uuid.h>
  84 #include <blkid/blkid.h>
  85 #include "zpool_util.h"
  86 #include <sys/zfs_context.h>
  87
  88 /*
  89  * For any given vdev specification, we can have multiple errors.  The
  90  * vdev_error() function keeps track of whether we have seen an error yet, and
  91  * prints out a header if its the first error we've seen.
  92  */
  93 boolean_t error_seen;
  94 boolean_t is_force;
  95
  96 typedef struct vdev_disk_db_entry
  97 {
  98         char id[24];
  99         int sector_size;
 100 } vdev_disk_db_entry_t;
 101
 102 /*
 103  * Database of block devices that lie about physical sector sizes.  The
 104  * identification string must be precisely 24 characters to avoid false
 105  * negatives
 106  */
 107 static vdev_disk_db_entry_t vdev_disk_database[] = {
 108         {"ATA     ADATA SSD S396 3", 8192},
 109         {"ATA     APPLE SSD SM128E", 8192},
 110         {"ATA     APPLE SSD SM256E", 8192},
 111         {"ATA     APPLE SSD SM512E", 8192},
 112         {"ATA     APPLE SSD SM768E", 8192},
 113         {"ATA     C400-MTFDDAC064M", 8192},
 114         {"ATA     C400-MTFDDAC128M", 8192},
 115         {"ATA     C400-MTFDDAC256M", 8192},
 116         {"ATA     C400-MTFDDAC512M", 8192},
 117         {"ATA     Corsair Force 3 ", 8192},
 118         {"ATA     Corsair Force GS", 8192},
 119         {"ATA     INTEL SSDSA2CT04", 8192},
 120         {"ATA     INTEL SSDSA2BZ10", 8192},
 121         {"ATA     INTEL SSDSA2BZ20", 8192},
 122         {"ATA     INTEL SSDSA2BZ30", 8192},
 123         {"ATA     INTEL SSDSA2CW04", 8192},
 124         {"ATA     INTEL SSDSA2CW08", 8192},
 125         {"ATA     INTEL SSDSA2CW12", 8192},
 126         {"ATA     INTEL SSDSA2CW16", 8192},
 127         {"ATA     INTEL SSDSA2CW30", 8192},
 128         {"ATA     INTEL SSDSA2CW60", 8192},
 129         {"ATA     INTEL SSDSC2CT06", 8192},
 130         {"ATA     INTEL SSDSC2CT12", 8192},
 131         {"ATA     INTEL SSDSC2CT18", 8192},
 132         {"ATA     INTEL SSDSC2CT24", 8192},
 133         {"ATA     INTEL SSDSC2CW06", 8192},
 134         {"ATA     INTEL SSDSC2CW12", 8192},
 135         {"ATA     INTEL SSDSC2CW18", 8192},
 136         {"ATA     INTEL SSDSC2CW24", 8192},
 137         {"ATA     INTEL SSDSC2CW48", 8192},
 138         {"ATA     KINGSTON SH100S3", 8192},
 139         {"ATA     KINGSTON SH103S3", 8192},
 140         {"ATA     M4-CT064M4SSD2  ", 8192},
 141         {"ATA     M4-CT128M4SSD2  ", 8192},
 142         {"ATA     M4-CT256M4SSD2  ", 8192},
 143         {"ATA     M4-CT512M4SSD2  ", 8192},
 144         {"ATA     OCZ-AGILITY2    ", 8192},
 145         {"ATA     OCZ-AGILITY3    ", 8192},
 146         {"ATA     OCZ-VERTEX2 3.5 ", 8192},
 147         {"ATA     OCZ-VERTEX3     ", 8192},
 148         {"ATA     OCZ-VERTEX3 LT  ", 8192},
 149         {"ATA     OCZ-VERTEX3 MI  ", 8192},
 150         {"ATA     OCZ-VERTEX4     ", 8192},
 151         {"ATA     SAMSUNG MZ7WD120", 8192},
 152         {"ATA     SAMSUNG MZ7WD240", 8192},
 153         {"ATA     SAMSUNG MZ7WD480", 8192},
 154         {"ATA     SAMSUNG MZ7WD960", 8192},
 155         {"ATA     SAMSUNG SSD 830 ", 8192},
 156         {"ATA     Samsung SSD 840 ", 8192},
 157         {"ATA     SanDisk SSD U100", 8192},
 158         {"ATA     TOSHIBA THNSNH06", 8192},
 159         {"ATA     TOSHIBA THNSNH12", 8192},
 160         {"ATA     TOSHIBA THNSNH25", 8192},
 161         {"ATA     TOSHIBA THNSNH51", 8192},
 162         {"ATA     APPLE SSD TS064C", 4096},
 163         {"ATA     APPLE SSD TS128C", 4096},
 164         {"ATA     APPLE SSD TS256C", 4096},
 165         {"ATA     APPLE SSD TS512C", 4096},
 166         {"ATA     INTEL SSDSA2M040", 4096},
 167         {"ATA     INTEL SSDSA2M080", 4096},
 168         {"ATA     INTEL SSDSA2M160", 4096},
 169         {"ATA     INTEL SSDSC2MH12", 4096},
 170         {"ATA     INTEL SSDSC2MH25", 4096},
 171         {"ATA     OCZ CORE_SSD    ", 4096},
 172         {"ATA     OCZ-VERTEX      ", 4096},
 173         {"ATA     SAMSUNG MCCOE32G", 4096},
 174         {"ATA     SAMSUNG MCCOE64G", 4096},
 175         {"ATA     SAMSUNG SSD PM80", 4096},
 176         /* Flash drives optimized for 4KB IOs on larger pages */
 177         {"ATA     INTEL SSDSC2BA10", 4096},
 178         {"ATA     INTEL SSDSC2BA20", 4096},
 179         {"ATA     INTEL SSDSC2BA40", 4096},
 180         {"ATA     INTEL SSDSC2BA80", 4096},
 181         {"ATA     INTEL SSDSC2BB08", 4096},
 182         {"ATA     INTEL SSDSC2BB12", 4096},
 183         {"ATA     INTEL SSDSC2BB16", 4096},
 184         {"ATA     INTEL SSDSC2BB24", 4096},
 185         {"ATA     INTEL SSDSC2BB30", 4096},
 186         {"ATA     INTEL SSDSC2BB40", 4096},
 187         {"ATA     INTEL SSDSC2BB48", 4096},
 188         {"ATA     INTEL SSDSC2BB60", 4096},
 189         {"ATA     INTEL SSDSC2BB80", 4096},
 190         {"ATA     INTEL SSDSC2BW24", 4096},
 191         {"ATA     INTEL SSDSC2BP24", 4096},
 192         {"ATA     INTEL SSDSC2BP48", 4096},
 193         {"NA      SmrtStorSDLKAE9W", 4096},
 194         /* Imported from Open Solaris */
 195         {"ATA     MARVELL SD88SA02", 4096},
 196         /* Advanced format Hard drives */
 197         {"ATA     Hitachi HDS5C303", 4096},
 198         {"ATA     SAMSUNG HD204UI ", 4096},
 199         {"ATA     ST2000DL004 HD20", 4096},
 200         {"ATA     WDC WD10EARS-00M", 4096},
 201         {"ATA     WDC WD10EARS-00S", 4096},
 202         {"ATA     WDC WD10EARS-00Z", 4096},
 203         {"ATA     WDC WD15EARS-00M", 4096},
 204         {"ATA     WDC WD15EARS-00S", 4096},
 205         {"ATA     WDC WD15EARS-00Z", 4096},
 206         {"ATA     WDC WD20EARS-00M", 4096},
 207         {"ATA     WDC WD20EARS-00S", 4096},
 208         {"ATA     WDC WD20EARS-00Z", 4096},
 209         {"ATA     WDC WD1600BEVT-0", 4096},
 210         {"ATA     WDC WD2500BEVT-0", 4096},
 211         {"ATA     WDC WD3200BEVT-0", 4096},
 212         {"ATA     WDC WD5000BEVT-0", 4096},
 213         /* Virtual disks: Assume zvols with default volblocksize */
 214 #if 0
 215         {"ATA     QEMU HARDDISK   ", 8192},
 216         {"IET     VIRTUAL-DISK    ", 8192},
 217         {"OI      COMSTAR         ", 8192},
 218         {"SUN     COMSTAR         ", 8192},
 219         {"NETAPP  LUN             ", 8192},
 220 #endif
 221 };
 222
 223 static const int vdev_disk_database_size =
 224         sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
 225
 226 #define INQ_REPLY_LEN   96
 227 #define INQ_CMD_LEN     6
 228
 229 static boolean_t
 230 check_sector_size_database(char *path, int *sector_size)
 231 {
 232         unsigned char inq_buff[INQ_REPLY_LEN];
 233         unsigned char sense_buffer[32];
 234         unsigned char inq_cmd_blk[INQ_CMD_LEN] =
 235             {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
 236         sg_io_hdr_t io_hdr;
 237         int error;
 238         int fd;
 239         int i;
 240
 241         /* Prepare INQUIRY command */
 242         memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
 243         io_hdr.interface_id = 'S';
 244         io_hdr.cmd_len = sizeof (inq_cmd_blk);
 245         io_hdr.mx_sb_len = sizeof (sense_buffer);
 246         io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 247         io_hdr.dxfer_len = INQ_REPLY_LEN;
 248         io_hdr.dxferp = inq_buff;
 249         io_hdr.cmdp = inq_cmd_blk;
 250         io_hdr.sbp = sense_buffer;
 251         io_hdr.timeout = 10;            /* 10 milliseconds is ample time */
 252
 253         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 254                 return (B_FALSE);
 255
 256         error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
 257
 258         (void) close(fd);
 259
 260         if (error < 0)
 261                 return (B_FALSE);
 262
 263         if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
 264                 return (B_FALSE);
 265
 266         for (i = 0; i < vdev_disk_database_size; i++) {
 267                 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
 268                         continue;
 269
 270                 *sector_size = vdev_disk_database[i].sector_size;
 271                 return (B_TRUE);
 272         }
 273
 274         return (B_FALSE);
 275 }
 276
 277 /*PRINTFLIKE1*/
 278 static void
 279 vdev_error(const char *fmt, ...)
 280 {
 281         va_list ap;
 282
 283         if (!error_seen) {
 284                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
 285                 if (!is_force)
 286                         (void) fprintf(stderr, gettext("use '-f' to override "
 287                             "the following errors:\n"));
 288                 else
 289                         (void) fprintf(stderr, gettext("the following errors "
 290                             "must be manually repaired:\n"));
 291                 error_seen = B_TRUE;
 292         }
 293
 294         va_start(ap, fmt);
 295         (void) vfprintf(stderr, fmt, ap);
 296         va_end(ap);
 297 }
 298
 299 /*
 300  * Check that a file is valid.  All we can do in this case is check that it's
 301  * not in use by another pool, and not in use by swap.
 302  */
 303 static int
 304 check_file(const char *file, boolean_t force, boolean_t isspare)
 305 {
 306         char  *name;
 307         int fd;
 308         int ret = 0;
 309         pool_state_t state;
 310         boolean_t inuse;
 311
 312         if ((fd = open(file, O_RDONLY)) < 0)
 313                 return (0);
 314
 315         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
 316                 const char *desc;
 317
 318                 switch (state) {
 319                 case POOL_STATE_ACTIVE:
 320                         desc = gettext("active");
 321                         break;
 322
 323                 case POOL_STATE_EXPORTED:
 324                         desc = gettext("exported");
 325                         break;
 326
 327                 case POOL_STATE_POTENTIALLY_ACTIVE:
 328                         desc = gettext("potentially active");
 329                         break;
 330
 331                 default:
 332                         desc = gettext("unknown");
 333                         break;
 334                 }
 335
 336                 /*
 337                  * Allow hot spares to be shared between pools.
 338                  */
 339                 if (state == POOL_STATE_SPARE && isspare) {
 340                         free(name);
 341                         (void) close(fd);
 342                         return (0);
 343                 }
 344
 345                 if (state == POOL_STATE_ACTIVE ||
 346                     state == POOL_STATE_SPARE || !force) {
 347                         switch (state) {
 348                         case POOL_STATE_SPARE:
 349                                 vdev_error(gettext("%s is reserved as a hot "
 350                                     "spare for pool %s\n"), file, name);
 351                                 break;
 352                         default:
 353                                 vdev_error(gettext("%s is part of %s pool "
 354                                     "'%s'\n"), file, desc, name);
 355                                 break;
 356                         }
 357                         ret = -1;
 358                 }
 359
 360                 free(name);
 361         }
 362
 363         (void) close(fd);
 364         return (ret);
 365 }
 366
 367 static void
 368 check_error(int err)
 369 {
 370         (void) fprintf(stderr, gettext("warning: device in use checking "
 371             "failed: %s\n"), strerror(err));
 372 }
 373
 374 static int
 375 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
 376 {
 377         int err;
 378         char *value;
 379
 380         /* No valid type detected device is safe to use */
 381         value = blkid_get_tag_value(cache, "TYPE", path);
 382         if (value == NULL)
 383                 return (0);
 384
 385         /*
 386          * If libblkid detects a ZFS device, we check the device
 387          * using check_file() to see if it's safe.  The one safe
 388          * case is a spare device shared between multiple pools.
 389          */
 390         if (strcmp(value, "zfs_member") == 0) {
 391                 err = check_file(path, force, isspare);
 392         } else {
 393                 if (force) {
 394                         err = 0;
 395                 } else {
 396                         err = -1;
 397                         vdev_error(gettext("%s contains a filesystem of "
 398                             "type '%s'\n"), path, value);
 399                 }
 400         }
 401
 402         free(value);
 403
 404         return (err);
 405 }
 406
 407 /*
 408  * Validate that a disk including all partitions are safe to use.
 409  *
 410  * For EFI labeled disks this can done relatively easily with the libefi
 411  * library.  The partition numbers are extracted from the label and used
 412  * to generate the expected /dev/ paths.  Each partition can then be
 413  * checked for conflicts.
 414  *
 415  * For non-EFI labeled disks (MBR/EBR/etc) the same process is possible
 416  * but due to the lack of a readily available libraries this scanning is
 417  * not implemented.  Instead only the device path as given is checked.
 418  */
 419 static int
 420 check_disk(const char *path, blkid_cache cache, int force,
 421     boolean_t isspare, boolean_t iswholedisk)
 422 {
 423         struct dk_gpt *vtoc;
 424         char slice_path[MAXPATHLEN];
 425         int err = 0;
 426         int fd, i;
 427
 428         if (!iswholedisk)
 429                 return (check_slice(path, cache, force, isspare));
 430
 431         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
 432                 check_error(errno);
 433                 return (-1);
 434         }
 435
 436         /*
 437          * Expected to fail for non-EFI labled disks.  Just check the device
 438          * as given and do not attempt to detect and scan partitions.
 439          */
 440         err = efi_alloc_and_read(fd, &vtoc);
 441         if (err) {
 442                 (void) close(fd);
 443                 return (check_slice(path, cache, force, isspare));
 444         }
 445
 446         /*
 447          * The primary efi partition label is damaged however the secondary
 448          * label at the end of the device is intact.  Rather than use this
 449          * label we should play it safe and treat this as a non efi device.
 450          */
 451         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
 452                 efi_free(vtoc);
 453                 (void) close(fd);
 454
 455                 if (force) {
 456                         /* Partitions will now be created using the backup */
 457                         return (0);
 458                 } else {
 459                         vdev_error(gettext("%s contains a corrupt primary "
 460                             "EFI label.\n"), path);
 461                         return (-1);
 462                 }
 463         }
 464
 465         for (i = 0; i < vtoc->efi_nparts; i++) {
 466
 467                 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
 468                     uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
 469                         continue;
 470
 471                 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
 472                         (void) snprintf(slice_path, sizeof (slice_path),
 473                             "%s%s%d", path, "-part", i+1);
 474                 else
 475                         (void) snprintf(slice_path, sizeof (slice_path),
 476                             "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
 477                             "p" : "", i+1);
 478
 479                 err = check_slice(slice_path, cache, force, isspare);
 480                 if (err)
 481                         break;
 482         }
 483
 484         efi_free(vtoc);
 485         (void) close(fd);
 486
 487         return (err);
 488 }
 489
 490 static int
 491 check_device(const char *path, boolean_t force,
 492     boolean_t isspare, boolean_t iswholedisk)
 493 {
 494         blkid_cache cache;
 495         int error;
 496
 497         error = blkid_get_cache(&cache, NULL);
 498         if (error != 0) {
 499                 check_error(error);
 500                 return (-1);
 501         }
 502
 503         error = check_disk(path, cache, force, isspare, iswholedisk);
 504         blkid_put_cache(cache);
 505
 506         return (error);
 507 }
 508
 509 /*
 510  * This may be a shorthand device path or it could be total gibberish.
 511  * Check to see if it is a known device available in zfs_vdev_paths.
 512  * As part of this check, see if we've been given an entire disk
 513  * (minus the slice number).
 514  */
 515 static int
 516 is_shorthand_path(const char *arg, char *path, size_t path_size,
 517     struct stat64 *statbuf, boolean_t *wholedisk)
 518 {
 519         int error;
 520
 521         error = zfs_resolve_shortname(arg, path, path_size);
 522         if (error == 0) {
 523                 *wholedisk = zfs_dev_is_whole_disk(path);
 524                 if (*wholedisk || (stat64(path, statbuf) == 0))
 525                         return (0);
 526         }
 527
 528         strlcpy(path, arg, path_size);
 529         memset(statbuf, 0, sizeof (*statbuf));
 530         *wholedisk = B_FALSE;
 531
 532         return (error);
 533 }
 534
 535 /*
 536  * Determine if the given path is a hot spare within the given configuration.
 537  * If no configuration is given we rely solely on the label.
 538  */
 539 static boolean_t
 540 is_spare(nvlist_t *config, const char *path)
 541 {
 542         int fd;
 543         pool_state_t state;
 544         char *name = NULL;
 545         nvlist_t *label;
 546         uint64_t guid, spareguid;
 547         nvlist_t *nvroot;
 548         nvlist_t **spares;
 549         uint_t i, nspares;
 550         boolean_t inuse;
 551
 552         if ((fd = open(path, O_RDONLY)) < 0)
 553                 return (B_FALSE);
 554
 555         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
 556             !inuse ||
 557             state != POOL_STATE_SPARE ||
 558             zpool_read_label(fd, &label, NULL) != 0) {
 559                 free(name);
 560                 (void) close(fd);
 561                 return (B_FALSE);
 562         }
 563         free(name);
 564         (void) close(fd);
 565
 566         if (config == NULL) {
 567                 nvlist_free(label);
 568                 return (B_TRUE);
 569         }
 570
 571         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
 572         nvlist_free(label);
 573
 574         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 575             &nvroot) == 0);
 576         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 577             &spares, &nspares) == 0) {
 578                 for (i = 0; i < nspares; i++) {
 579                         verify(nvlist_lookup_uint64(spares[i],
 580                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
 581                         if (spareguid == guid)
 582                                 return (B_TRUE);
 583                 }
 584         }
 585
 586         return (B_FALSE);
 587 }
 588
 589 /*
 590  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 591  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 592  * leaf vdev are:
 593  *
 594  *      /dev/xxx        Complete disk path
 595  *      /xxx            Full path to file
 596  *      xxx             Shorthand for <zfs_vdev_paths>/xxx
 597  */
 598 static nvlist_t *
 599 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
 600 {
 601         char path[MAXPATHLEN];
 602         struct stat64 statbuf;
 603         nvlist_t *vdev = NULL;
 604         char *type = NULL;
 605         boolean_t wholedisk = B_FALSE;
 606         uint64_t ashift = 0;
 607         int err;
 608
 609         /*
 610          * Determine what type of vdev this is, and put the full path into
 611          * 'path'.  We detect whether this is a device of file afterwards by
 612          * checking the st_mode of the file.
 613          */
 614         if (arg[0] == '/') {
 615                 /*
 616                  * Complete device or file path.  Exact type is determined by
 617                  * examining the file descriptor afterwards.  Symbolic links
 618                  * are resolved to their real paths to determine whole disk
 619                  * and S_ISBLK/S_ISREG type checks.  However, we are careful
 620                  * to store the given path as ZPOOL_CONFIG_PATH to ensure we
 621                  * can leverage udev's persistent device labels.
 622                  */
 623                 if (realpath(arg, path) == NULL) {
 624                         (void) fprintf(stderr,
 625                             gettext("cannot resolve path '%s'\n"), arg);
 626                         return (NULL);
 627                 }
 628
 629                 wholedisk = zfs_dev_is_whole_disk(path);
 630                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 631                         (void) fprintf(stderr,
 632                             gettext("cannot open '%s': %s\n"),
 633                             path, strerror(errno));
 634                         return (NULL);
 635                 }
 636
 637                 /* After whole disk check restore original passed path */
 638                 strlcpy(path, arg, sizeof (path));
 639         } else {
 640                 err = is_shorthand_path(arg, path, sizeof (path),
 641                     &statbuf, &wholedisk);
 642                 if (err != 0) {
 643                         /*
 644                          * If we got ENOENT, then the user gave us
 645                          * gibberish, so try to direct them with a
 646                          * reasonable error message.  Otherwise,
 647                          * regurgitate strerror() since it's the best we
 648                          * can do.
 649                          */
 650                         if (err == ENOENT) {
 651                                 (void) fprintf(stderr,
 652                                     gettext("cannot open '%s': no such "
 653                                     "device in %s\n"), arg, DISK_ROOT);
 654                                 (void) fprintf(stderr,
 655                                     gettext("must be a full path or "
 656                                     "shorthand device name\n"));
 657                                 return (NULL);
 658                         } else {
 659                                 (void) fprintf(stderr,
 660                                     gettext("cannot open '%s': %s\n"),
 661                                     path, strerror(errno));
 662                                 return (NULL);
 663                         }
 664                 }
 665         }
 666
 667         /*
 668          * Determine whether this is a device or a file.
 669          */
 670         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 671                 type = VDEV_TYPE_DISK;
 672         } else if (S_ISREG(statbuf.st_mode)) {
 673                 type = VDEV_TYPE_FILE;
 674         } else {
 675                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 676                     "block device or regular file\n"), path);
 677                 return (NULL);
 678         }
 679
 680         /*
 681          * Finally, we have the complete device or file, and we know that it is
 682          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 683          * vdevs have a 'path' element, and devices also have a 'devid' element.
 684          */
 685         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 686         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 687         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 688         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 689         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 690                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 691                     (uint64_t)wholedisk) == 0);
 692
 693         /*
 694          * Override defaults if custom properties are provided.
 695          */
 696         if (props != NULL) {
 697                 char *value = NULL;
 698
 699                 if (nvlist_lookup_string(props,
 700                     zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0) {
 701                         if (zfs_nicestrtonum(NULL, value, &ashift) != 0) {
 702                                 (void) fprintf(stderr,
 703                                     gettext("ashift must be a number.\n"));
 704                                 return (NULL);
 705                         }
 706                         if (ashift != 0 &&
 707                             (ashift < ASHIFT_MIN || ashift > ASHIFT_MAX)) {
 708                                 (void) fprintf(stderr,
 709                                     gettext("invalid 'ashift=%" PRIu64 "' "
 710                                     "property: only values between %" PRId32 " "
 711                                     "and %" PRId32 " are allowed.\n"),
 712                                     ashift, ASHIFT_MIN, ASHIFT_MAX);
 713                                 return (NULL);
 714                         }
 715                 }
 716         }
 717
 718         /*
 719          * If the device is known to incorrectly report its physical sector
 720          * size explicitly provide the known correct value.
 721          */
 722         if (ashift == 0) {
 723                 int sector_size;
 724
 725                 if (check_sector_size_database(path, &sector_size) == B_TRUE)
 726                         ashift = highbit64(sector_size) - 1;
 727         }
 728
 729         if (ashift > 0)
 730                 (void) nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
 731
 732         return (vdev);
 733 }
 734
 735 /*
 736  * Go through and verify the replication level of the pool is consistent.
 737  * Performs the following checks:
 738  *
 739  *      For the new spec, verifies that devices in mirrors and raidz are the
 740  *      same size.
 741  *
 742  *      If the current configuration already has inconsistent replication
 743  *      levels, ignore any other potential problems in the new spec.
 744  *
 745  *      Otherwise, make sure that the current spec (if there is one) and the new
 746  *      spec have consistent replication levels.
 747  */
 748 typedef struct replication_level {
 749         char *zprl_type;
 750         uint64_t zprl_children;
 751         uint64_t zprl_parity;
 752 } replication_level_t;
 753
 754 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 755
 756 static boolean_t
 757 is_raidz_mirror(replication_level_t *a, replication_level_t *b,
 758     replication_level_t **raidz, replication_level_t **mirror)
 759 {
 760         if (strcmp(a->zprl_type, "raidz") == 0 &&
 761             strcmp(b->zprl_type, "mirror") == 0) {
 762                 *raidz = a;
 763                 *mirror = b;
 764                 return (B_TRUE);
 765         }
 766         return (B_FALSE);
 767 }
 768
 769 /*
 770  * Given a list of toplevel vdevs, return the current replication level.  If
 771  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 772  * an error message will be displayed for each self-inconsistent vdev.
 773  */
 774 static replication_level_t *
 775 get_replication(nvlist_t *nvroot, boolean_t fatal)
 776 {
 777         nvlist_t **top;
 778         uint_t t, toplevels;
 779         nvlist_t **child;
 780         uint_t c, children;
 781         nvlist_t *nv;
 782         char *type;
 783         replication_level_t lastrep = {0};
 784         replication_level_t rep;
 785         replication_level_t *ret;
 786         replication_level_t *raidz, *mirror;
 787         boolean_t dontreport;
 788
 789         ret = safe_malloc(sizeof (replication_level_t));
 790
 791         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 792             &top, &toplevels) == 0);
 793
 794         for (t = 0; t < toplevels; t++) {
 795                 uint64_t is_log = B_FALSE;
 796
 797                 nv = top[t];
 798
 799                 /*
 800                  * For separate logs we ignore the top level vdev replication
 801                  * constraints.
 802                  */
 803                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 804                 if (is_log)
 805                         continue;
 806
 807                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 808                     &type) == 0);
 809                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 810                     &child, &children) != 0) {
 811                         /*
 812                          * This is a 'file' or 'disk' vdev.
 813                          */
 814                         rep.zprl_type = type;
 815                         rep.zprl_children = 1;
 816                         rep.zprl_parity = 0;
 817                 } else {
 818                         uint64_t vdev_size;
 819
 820                         /*
 821                          * This is a mirror or RAID-Z vdev.  Go through and make
 822                          * sure the contents are all the same (files vs. disks),
 823                          * keeping track of the number of elements in the
 824                          * process.
 825                          *
 826                          * We also check that the size of each vdev (if it can
 827                          * be determined) is the same.
 828                          */
 829                         rep.zprl_type = type;
 830                         rep.zprl_children = 0;
 831
 832                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 833                                 verify(nvlist_lookup_uint64(nv,
 834                                     ZPOOL_CONFIG_NPARITY,
 835                                     &rep.zprl_parity) == 0);
 836                                 assert(rep.zprl_parity != 0);
 837                         } else {
 838                                 rep.zprl_parity = 0;
 839                         }
 840
 841                         /*
 842                          * The 'dontreport' variable indicates that we've
 843                          * already reported an error for this spec, so don't
 844                          * bother doing it again.
 845                          */
 846                         type = NULL;
 847                         dontreport = 0;
 848                         vdev_size = -1ULL;
 849                         for (c = 0; c < children; c++) {
 850                                 nvlist_t *cnv = child[c];
 851                                 char *path;
 852                                 struct stat64 statbuf;
 853                                 uint64_t size = -1ULL;
 854                                 char *childtype;
 855                                 int fd, err;
 856
 857                                 rep.zprl_children++;
 858
 859                                 verify(nvlist_lookup_string(cnv,
 860                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 861
 862                                 /*
 863                                  * If this is a replacing or spare vdev, then
 864                                  * get the real first child of the vdev.
 865                                  */
 866                                 if (strcmp(childtype,
 867                                     VDEV_TYPE_REPLACING) == 0 ||
 868                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 869                                         nvlist_t **rchild;
 870                                         uint_t rchildren;
 871
 872                                         verify(nvlist_lookup_nvlist_array(cnv,
 873                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 874                                             &rchildren) == 0);
 875                                         assert(rchildren == 2);
 876                                         cnv = rchild[0];
 877
 878                                         verify(nvlist_lookup_string(cnv,
 879                                             ZPOOL_CONFIG_TYPE,
 880                                             &childtype) == 0);
 881                                 }
 882
 883                                 verify(nvlist_lookup_string(cnv,
 884                                     ZPOOL_CONFIG_PATH, &path) == 0);
 885
 886                                 /*
 887                                  * If we have a raidz/mirror that combines disks
 888                                  * with files, report it as an error.
 889                                  */
 890                                 if (!dontreport && type != NULL &&
 891                                     strcmp(type, childtype) != 0) {
 892                                         if (ret != NULL)
 893                                                 free(ret);
 894                                         ret = NULL;
 895                                         if (fatal)
 896                                                 vdev_error(gettext(
 897                                                     "mismatched replication "
 898                                                     "level: %s contains both "
 899                                                     "files and devices\n"),
 900                                                     rep.zprl_type);
 901                                         else
 902                                                 return (NULL);
 903                                         dontreport = B_TRUE;
 904                                 }
 905
 906                                 /*
 907                                  * According to stat(2), the value of 'st_size'
 908                                  * is undefined for block devices and character
 909                                  * devices.  But there is no effective way to
 910                                  * determine the real size in userland.
 911                                  *
 912                                  * Instead, we'll take advantage of an
 913                                  * implementation detail of spec_size().  If the
 914                                  * device is currently open, then we (should)
 915                                  * return a valid size.
 916                                  *
 917                                  * If we still don't get a valid size (indicated
 918                                  * by a size of 0 or MAXOFFSET_T), then ignore
 919                                  * this device altogether.
 920                                  */
 921                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 922                                         err = fstat64(fd, &statbuf);
 923                                         (void) close(fd);
 924                                 } else {
 925                                         err = stat64(path, &statbuf);
 926                                 }
 927
 928                                 if (err != 0 ||
 929                                     statbuf.st_size == 0 ||
 930                                     statbuf.st_size == MAXOFFSET_T)
 931                                         continue;
 932
 933                                 size = statbuf.st_size;
 934
 935                                 /*
 936                                  * Also make sure that devices and
 937                                  * slices have a consistent size.  If
 938                                  * they differ by a significant amount
 939                                  * (~16MB) then report an error.
 940                                  */
 941                                 if (!dontreport &&
 942                                     (vdev_size != -1ULL &&
 943                                     (labs(size - vdev_size) >
 944                                     ZPOOL_FUZZ))) {
 945                                         if (ret != NULL)
 946                                                 free(ret);
 947                                         ret = NULL;
 948                                         if (fatal)
 949                                                 vdev_error(gettext(
 950                                                     "%s contains devices of "
 951                                                     "different sizes\n"),
 952                                                     rep.zprl_type);
 953                                         else
 954                                                 return (NULL);
 955                                         dontreport = B_TRUE;
 956                                 }
 957
 958                                 type = childtype;
 959                                 vdev_size = size;
 960                         }
 961                 }
 962
 963                 /*
 964                  * At this point, we have the replication of the last toplevel
 965                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 966                  * different.
 967                  */
 968                 if (lastrep.zprl_type != NULL) {
 969                         if (is_raidz_mirror(&lastrep, &rep, &raidz, &mirror) ||
 970                             is_raidz_mirror(&rep, &lastrep, &raidz, &mirror)) {
 971                                 /*
 972                                  * Accepted raidz and mirror when they can
 973                                  * handle the same number of disk failures.
 974                                  */
 975                                 if (raidz->zprl_parity !=
 976                                     mirror->zprl_children - 1) {
 977                                         if (ret != NULL)
 978                                                 free(ret);
 979                                         ret = NULL;
 980                                         if (fatal)
 981                                                 vdev_error(gettext(
 982                                                     "mismatched replication "
 983                                                     "level: "
 984                                                     "%s and %s vdevs with "
 985                                                     "different redundancy, "
 986                                                     "%llu vs. %llu (%llu-way) "
 987                                                     "are present\n"),
 988                                                     raidz->zprl_type,
 989                                                     mirror->zprl_type,
 990                                                     raidz->zprl_parity,
 991                                                     mirror->zprl_children - 1,
 992                                                     mirror->zprl_children);
 993                                         else
 994                                                 return (NULL);
 995                                 }
 996                         } else if (strcmp(lastrep.zprl_type, rep.zprl_type) !=
 997                             0) {
 998                                 if (ret != NULL)
 999                                         free(ret);
1000                                 ret = NULL;
1001                                 if (fatal)
1002                                         vdev_error(gettext(
1003                                             "mismatched replication level: "
1004                                             "both %s and %s vdevs are "
1005                                             "present\n"),
1006                                             lastrep.zprl_type, rep.zprl_type);
1007                                 else
1008                                         return (NULL);
1009                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
1010                                 if (ret)
1011                                         free(ret);
1012                                 ret = NULL;
1013                                 if (fatal)
1014                                         vdev_error(gettext(
1015                                             "mismatched replication level: "
1016                                             "both %llu and %llu device parity "
1017                                             "%s vdevs are present\n"),
1018                                             lastrep.zprl_parity,
1019                                             rep.zprl_parity,
1020                                             rep.zprl_type);
1021                                 else
1022                                         return (NULL);
1023                         } else if (lastrep.zprl_children != rep.zprl_children) {
1024                                 if (ret)
1025                                         free(ret);
1026                                 ret = NULL;
1027                                 if (fatal)
1028                                         vdev_error(gettext(
1029                                             "mismatched replication level: "
1030                                             "both %llu-way and %llu-way %s "
1031                                             "vdevs are present\n"),
1032                                             lastrep.zprl_children,
1033                                             rep.zprl_children,
1034                                             rep.zprl_type);
1035                                 else
1036                                         return (NULL);
1037                         }
1038                 }
1039                 lastrep = rep;
1040         }
1041
1042         if (ret != NULL)
1043                 *ret = rep;
1044
1045         return (ret);
1046 }
1047
1048 /*
1049  * Check the replication level of the vdev spec against the current pool.  Calls
1050  * get_replication() to make sure the new spec is self-consistent.  If the pool
1051  * has a consistent replication level, then we ignore any errors.  Otherwise,
1052  * report any difference between the two.
1053  */
1054 static int
1055 check_replication(nvlist_t *config, nvlist_t *newroot)
1056 {
1057         nvlist_t **child;
1058         uint_t  children;
1059         replication_level_t *current = NULL, *new;
1060         int ret;
1061
1062         /*
1063          * If we have a current pool configuration, check to see if it's
1064          * self-consistent.  If not, simply return success.
1065          */
1066         if (config != NULL) {
1067                 nvlist_t *nvroot;
1068
1069                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1070                     &nvroot) == 0);
1071                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1072                         return (0);
1073         }
1074         /*
1075          * for spares there may be no children, and therefore no
1076          * replication level to check
1077          */
1078         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1079             &child, &children) != 0) || (children == 0)) {
1080                 free(current);
1081                 return (0);
1082         }
1083
1084         /*
1085          * If all we have is logs then there's no replication level to check.
1086          */
1087         if (num_logs(newroot) == children) {
1088                 free(current);
1089                 return (0);
1090         }
1091
1092         /*
1093          * Get the replication level of the new vdev spec, reporting any
1094          * inconsistencies found.
1095          */
1096         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1097                 free(current);
1098                 return (-1);
1099         }
1100
1101         /*
1102          * Check to see if the new vdev spec matches the replication level of
1103          * the current pool.
1104          */
1105         ret = 0;
1106         if (current != NULL) {
1107                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1108                         vdev_error(gettext(
1109                             "mismatched replication level: pool uses %s "
1110                             "and new vdev is %s\n"),
1111                             current->zprl_type, new->zprl_type);
1112                         ret = -1;
1113                 } else if (current->zprl_parity != new->zprl_parity) {
1114                         vdev_error(gettext(
1115                             "mismatched replication level: pool uses %llu "
1116                             "device parity and new vdev uses %llu\n"),
1117                             current->zprl_parity, new->zprl_parity);
1118                         ret = -1;
1119                 } else if (current->zprl_children != new->zprl_children) {
1120                         vdev_error(gettext(
1121                             "mismatched replication level: pool uses %llu-way "
1122                             "%s and new vdev uses %llu-way %s\n"),
1123                             current->zprl_children, current->zprl_type,
1124                             new->zprl_children, new->zprl_type);
1125                         ret = -1;
1126                 }
1127         }
1128
1129         free(new);
1130         if (current != NULL)
1131                 free(current);
1132
1133         return (ret);
1134 }
1135
1136 static int
1137 zero_label(char *path)
1138 {
1139         const int size = 4096;
1140         char buf[size];
1141         int err, fd;
1142
1143         if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1144                 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1145                     path, strerror(errno));
1146                 return (-1);
1147         }
1148
1149         memset(buf, 0, size);
1150         err = write(fd, buf, size);
1151         (void) fdatasync(fd);
1152         (void) close(fd);
1153
1154         if (err == -1) {
1155                 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1156                     "of '%s': %s\n"), size, path, strerror(errno));
1157                 return (-1);
1158         }
1159
1160         if (err != size) {
1161                 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1162                     "of '%s'\n"), err, size, path);
1163                 return (-1);
1164         }
1165
1166         return (0);
1167 }
1168
1169 /*
1170  * Go through and find any whole disks in the vdev specification, labelling them
1171  * as appropriate.  When constructing the vdev spec, we were unable to open this
1172  * device in order to provide a devid.  Now that we have labelled the disk and
1173  * know that slice 0 is valid, we can construct the devid now.
1174  *
1175  * If the disk was already labeled with an EFI label, we will have gotten the
1176  * devid already (because we were able to open the whole disk).  Otherwise, we
1177  * need to get the devid after we label the disk.
1178  */
1179 static int
1180 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1181 {
1182         nvlist_t **child;
1183         uint_t c, children;
1184         char *type, *path;
1185         char devpath[MAXPATHLEN];
1186         char udevpath[MAXPATHLEN];
1187         uint64_t wholedisk;
1188         struct stat64 statbuf;
1189         int is_exclusive = 0;
1190         int fd;
1191         int ret;
1192
1193         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1194
1195         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1196             &child, &children) != 0) {
1197
1198                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1199                         return (0);
1200
1201                 /*
1202                  * We have a disk device.  If this is a whole disk write
1203                  * out the efi partition table, otherwise write zero's to
1204                  * the first 4k of the partition.  This is to ensure that
1205                  * libblkid will not misidentify the partition due to a
1206                  * magic value left by the previous filesystem.
1207                  */
1208                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1209                 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1210                     &wholedisk));
1211
1212                 if (!wholedisk) {
1213                         /*
1214                          * Update device id string for mpath nodes (Linux only)
1215                          */
1216                         if (is_mpath_whole_disk(path))
1217                                 update_vdev_config_dev_strs(nv);
1218
1219                         (void) zero_label(path);
1220                         return (0);
1221                 }
1222
1223                 if (realpath(path, devpath) == NULL) {
1224                         ret = errno;
1225                         (void) fprintf(stderr,
1226                             gettext("cannot resolve path '%s'\n"), path);
1227                         return (ret);
1228                 }
1229
1230                 /*
1231                  * Remove any previously existing symlink from a udev path to
1232                  * the device before labeling the disk.  This ensures that
1233                  * only newly created links are used.  Otherwise there is a
1234                  * window between when udev deletes and recreates the link
1235                  * during which access attempts will fail with ENOENT.
1236                  */
1237                 strlcpy(udevpath, path, MAXPATHLEN);
1238                 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1239
1240                 fd = open(devpath, O_RDWR|O_EXCL);
1241                 if (fd == -1) {
1242                         if (errno == EBUSY)
1243                                 is_exclusive = 1;
1244                 } else {
1245                         (void) close(fd);
1246                 }
1247
1248                 /*
1249                  * If the partition exists, contains a valid spare label,
1250                  * and is opened exclusively there is no need to partition
1251                  * it.  Hot spares have already been partitioned and are
1252                  * held open exclusively by the kernel as a safety measure.
1253                  *
1254                  * If the provided path is for a /dev/disk/ device its
1255                  * symbolic link will be removed, partition table created,
1256                  * and then block until udev creates the new link.
1257                  */
1258                 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1259                         char *devnode = strrchr(devpath, '/') + 1;
1260
1261                         ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1262                         if (ret == 0) {
1263                                 ret = lstat64(udevpath, &statbuf);
1264                                 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1265                                         (void) unlink(udevpath);
1266                         }
1267
1268                         /*
1269                          * When labeling a pool the raw device node name
1270                          * is provided as it appears under /dev/.
1271                          */
1272                         if (zpool_label_disk(g_zfs, zhp, devnode) == -1)
1273                                 return (-1);
1274
1275                         /*
1276                          * Wait for udev to signal the device is available
1277                          * by the provided path.
1278                          */
1279                         ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1280                         if (ret) {
1281                                 (void) fprintf(stderr,
1282                                     gettext("missing link: %s was "
1283                                     "partitioned but %s is missing\n"),
1284                                     devnode, udevpath);
1285                                 return (ret);
1286                         }
1287
1288                         ret = zero_label(udevpath);
1289                         if (ret)
1290                                 return (ret);
1291                 }
1292
1293                 /*
1294                  * Update the path to refer to the partition.  The presence of
1295                  * the 'whole_disk' field indicates to the CLI that we should
1296                  * chop off the partition number when displaying the device in
1297                  * future output.
1298                  */
1299                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1300
1301                 /*
1302                  * Update device id strings for whole disks (Linux only)
1303                  */
1304                 update_vdev_config_dev_strs(nv);
1305
1306                 return (0);
1307         }
1308
1309         for (c = 0; c < children; c++)
1310                 if ((ret = make_disks(zhp, child[c])) != 0)
1311                         return (ret);
1312
1313         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1314             &child, &children) == 0)
1315                 for (c = 0; c < children; c++)
1316                         if ((ret = make_disks(zhp, child[c])) != 0)
1317                                 return (ret);
1318
1319         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1320             &child, &children) == 0)
1321                 for (c = 0; c < children; c++)
1322                         if ((ret = make_disks(zhp, child[c])) != 0)
1323                                 return (ret);
1324
1325         return (0);
1326 }
1327
1328 /*
1329  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1330  * the majority of this task.
1331  */
1332 static boolean_t
1333 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1334     boolean_t replacing, boolean_t isspare)
1335 {
1336         nvlist_t **child;
1337         uint_t c, children;
1338         char *type, *path;
1339         int ret = 0;
1340         char buf[MAXPATHLEN];
1341         uint64_t wholedisk = B_FALSE;
1342         boolean_t anyinuse = B_FALSE;
1343
1344         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1345
1346         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1347             &child, &children) != 0) {
1348
1349                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1350                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1351                         verify(!nvlist_lookup_uint64(nv,
1352                             ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1353
1354                 /*
1355                  * As a generic check, we look to see if this is a replace of a
1356                  * hot spare within the same pool.  If so, we allow it
1357                  * regardless of what libblkid or zpool_in_use() says.
1358                  */
1359                 if (replacing) {
1360                         (void) strlcpy(buf, path, sizeof (buf));
1361                         if (wholedisk) {
1362                                 ret = zfs_append_partition(buf,  sizeof (buf));
1363                                 if (ret == -1)
1364                                         return (-1);
1365                         }
1366
1367                         if (is_spare(config, buf))
1368                                 return (B_FALSE);
1369                 }
1370
1371                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1372                         ret = check_device(path, force, isspare, wholedisk);
1373
1374                 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1375                         ret = check_file(path, force, isspare);
1376
1377                 return (ret != 0);
1378         }
1379
1380         for (c = 0; c < children; c++)
1381                 if (is_device_in_use(config, child[c], force, replacing,
1382                     B_FALSE))
1383                         anyinuse = B_TRUE;
1384
1385         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1386             &child, &children) == 0)
1387                 for (c = 0; c < children; c++)
1388                         if (is_device_in_use(config, child[c], force, replacing,
1389                             B_TRUE))
1390                                 anyinuse = B_TRUE;
1391
1392         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1393             &child, &children) == 0)
1394                 for (c = 0; c < children; c++)
1395                         if (is_device_in_use(config, child[c], force, replacing,
1396                             B_FALSE))
1397                                 anyinuse = B_TRUE;
1398
1399         return (anyinuse);
1400 }
1401
1402 static const char *
1403 is_grouping(const char *type, int *mindev, int *maxdev)
1404 {
1405         if (strncmp(type, "raidz", 5) == 0) {
1406                 const char *p = type + 5;
1407                 char *end;
1408                 long nparity;
1409
1410                 if (*p == '\0') {
1411                         nparity = 1;
1412                 } else if (*p == '0') {
1413                         return (NULL); /* no zero prefixes allowed */
1414                 } else {
1415                         errno = 0;
1416                         nparity = strtol(p, &end, 10);
1417                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1418                             *end != '\0')
1419                                 return (NULL);
1420                 }
1421
1422                 if (mindev != NULL)
1423                         *mindev = nparity + 1;
1424                 if (maxdev != NULL)
1425                         *maxdev = 255;
1426                 return (VDEV_TYPE_RAIDZ);
1427         }
1428
1429         if (maxdev != NULL)
1430                 *maxdev = INT_MAX;
1431
1432         if (strcmp(type, "mirror") == 0) {
1433                 if (mindev != NULL)
1434                         *mindev = 2;
1435                 return (VDEV_TYPE_MIRROR);
1436         }
1437
1438         if (strcmp(type, "spare") == 0) {
1439                 if (mindev != NULL)
1440                         *mindev = 1;
1441                 return (VDEV_TYPE_SPARE);
1442         }
1443
1444         if (strcmp(type, "log") == 0) {
1445                 if (mindev != NULL)
1446                         *mindev = 1;
1447                 return (VDEV_TYPE_LOG);
1448         }
1449
1450         if (strcmp(type, "cache") == 0) {
1451                 if (mindev != NULL)
1452                         *mindev = 1;
1453                 return (VDEV_TYPE_L2CACHE);
1454         }
1455
1456         return (NULL);
1457 }
1458
1459 /*
1460  * Construct a syntactically valid vdev specification,
1461  * and ensure that all devices and files exist and can be opened.
1462  * Note: we don't bother freeing anything in the error paths
1463  * because the program is just going to exit anyway.
1464  */
1465 nvlist_t *
1466 construct_spec(nvlist_t *props, int argc, char **argv)
1467 {
1468         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1469         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1470         const char *type;
1471         uint64_t is_log;
1472         boolean_t seen_logs;
1473
1474         top = NULL;
1475         toplevels = 0;
1476         spares = NULL;
1477         l2cache = NULL;
1478         nspares = 0;
1479         nlogs = 0;
1480         nl2cache = 0;
1481         is_log = B_FALSE;
1482         seen_logs = B_FALSE;
1483         nvroot = NULL;
1484
1485         while (argc > 0) {
1486                 nv = NULL;
1487
1488                 /*
1489                  * If it's a mirror or raidz, the subsequent arguments are
1490                  * its leaves -- until we encounter the next mirror or raidz.
1491                  */
1492                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1493                         nvlist_t **child = NULL;
1494                         int c, children = 0;
1495
1496                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1497                                 if (spares != NULL) {
1498                                         (void) fprintf(stderr,
1499                                             gettext("invalid vdev "
1500                                             "specification: 'spare' can be "
1501                                             "specified only once\n"));
1502                                         goto spec_out;
1503                                 }
1504                                 is_log = B_FALSE;
1505                         }
1506
1507                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1508                                 if (seen_logs) {
1509                                         (void) fprintf(stderr,
1510                                             gettext("invalid vdev "
1511                                             "specification: 'log' can be "
1512                                             "specified only once\n"));
1513                                         goto spec_out;
1514                                 }
1515                                 seen_logs = B_TRUE;
1516                                 is_log = B_TRUE;
1517                                 argc--;
1518                                 argv++;
1519                                 /*
1520                                  * A log is not a real grouping device.
1521                                  * We just set is_log and continue.
1522                                  */
1523                                 continue;
1524                         }
1525
1526                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1527                                 if (l2cache != NULL) {
1528                                         (void) fprintf(stderr,
1529                                             gettext("invalid vdev "
1530                                             "specification: 'cache' can be "
1531                                             "specified only once\n"));
1532                                         goto spec_out;
1533                                 }
1534                                 is_log = B_FALSE;
1535                         }
1536
1537                         if (is_log) {
1538                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1539                                         (void) fprintf(stderr,
1540                                             gettext("invalid vdev "
1541                                             "specification: unsupported 'log' "
1542                                             "device: %s\n"), type);
1543                                         goto spec_out;
1544                                 }
1545                                 nlogs++;
1546                         }
1547
1548                         for (c = 1; c < argc; c++) {
1549                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1550                                         break;
1551                                 children++;
1552                                 child = realloc(child,
1553                                     children * sizeof (nvlist_t *));
1554                                 if (child == NULL)
1555                                         zpool_no_memory();
1556                                 if ((nv = make_leaf_vdev(props, argv[c],
1557                                     B_FALSE)) == NULL) {
1558                                         for (c = 0; c < children - 1; c++)
1559                                                 nvlist_free(child[c]);
1560                                         free(child);
1561                                         goto spec_out;
1562                                 }
1563
1564                                 child[children - 1] = nv;
1565                         }
1566
1567                         if (children < mindev) {
1568                                 (void) fprintf(stderr, gettext("invalid vdev "
1569                                     "specification: %s requires at least %d "
1570                                     "devices\n"), argv[0], mindev);
1571                                 for (c = 0; c < children; c++)
1572                                         nvlist_free(child[c]);
1573                                 free(child);
1574                                 goto spec_out;
1575                         }
1576
1577                         if (children > maxdev) {
1578                                 (void) fprintf(stderr, gettext("invalid vdev "
1579                                     "specification: %s supports no more than "
1580                                     "%d devices\n"), argv[0], maxdev);
1581                                 for (c = 0; c < children; c++)
1582                                         nvlist_free(child[c]);
1583                                 free(child);
1584                                 goto spec_out;
1585                         }
1586
1587                         argc -= c;
1588                         argv += c;
1589
1590                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1591                                 spares = child;
1592                                 nspares = children;
1593                                 continue;
1594                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1595                                 l2cache = child;
1596                                 nl2cache = children;
1597                                 continue;
1598                         } else {
1599                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1600                                     0) == 0);
1601                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1602                                     type) == 0);
1603                                 verify(nvlist_add_uint64(nv,
1604                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1605                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1606                                         verify(nvlist_add_uint64(nv,
1607                                             ZPOOL_CONFIG_NPARITY,
1608                                             mindev - 1) == 0);
1609                                 }
1610                                 verify(nvlist_add_nvlist_array(nv,
1611                                     ZPOOL_CONFIG_CHILDREN, child,
1612                                     children) == 0);
1613
1614                                 for (c = 0; c < children; c++)
1615                                         nvlist_free(child[c]);
1616                                 free(child);
1617                         }
1618                 } else {
1619                         /*
1620                          * We have a device.  Pass off to make_leaf_vdev() to
1621                          * construct the appropriate nvlist describing the vdev.
1622                          */
1623                         if ((nv = make_leaf_vdev(props, argv[0],
1624                             is_log)) == NULL)
1625                                 goto spec_out;
1626
1627                         if (is_log)
1628                                 nlogs++;
1629                         argc--;
1630                         argv++;
1631                 }
1632
1633                 toplevels++;
1634                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1635                 if (top == NULL)
1636                         zpool_no_memory();
1637                 top[toplevels - 1] = nv;
1638         }
1639
1640         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1641                 (void) fprintf(stderr, gettext("invalid vdev "
1642                     "specification: at least one toplevel vdev must be "
1643                     "specified\n"));
1644                 goto spec_out;
1645         }
1646
1647         if (seen_logs && nlogs == 0) {
1648                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1649                     "log requires at least 1 device\n"));
1650                 goto spec_out;
1651         }
1652
1653         /*
1654          * Finally, create nvroot and add all top-level vdevs to it.
1655          */
1656         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1657         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1658             VDEV_TYPE_ROOT) == 0);
1659         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1660             top, toplevels) == 0);
1661         if (nspares != 0)
1662                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1663                     spares, nspares) == 0);
1664         if (nl2cache != 0)
1665                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1666                     l2cache, nl2cache) == 0);
1667
1668 spec_out:
1669         for (t = 0; t < toplevels; t++)
1670                 nvlist_free(top[t]);
1671         for (t = 0; t < nspares; t++)
1672                 nvlist_free(spares[t]);
1673         for (t = 0; t < nl2cache; t++)
1674                 nvlist_free(l2cache[t]);
1675
1676         free(spares);
1677         free(l2cache);
1678         free(top);
1679
1680         return (nvroot);
1681 }
1682
1683 nvlist_t *
1684 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1685     splitflags_t flags, int argc, char **argv)
1686 {
1687         nvlist_t *newroot = NULL, **child;
1688         uint_t c, children;
1689
1690         if (argc > 0) {
1691                 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1692                         (void) fprintf(stderr, gettext("Unable to build a "
1693                             "pool from the specified devices\n"));
1694                         return (NULL);
1695                 }
1696
1697                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1698                         nvlist_free(newroot);
1699                         return (NULL);
1700                 }
1701
1702                 /* avoid any tricks in the spec */
1703                 verify(nvlist_lookup_nvlist_array(newroot,
1704                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1705                 for (c = 0; c < children; c++) {
1706                         char *path;
1707                         const char *type;
1708                         int min, max;
1709
1710                         verify(nvlist_lookup_string(child[c],
1711                             ZPOOL_CONFIG_PATH, &path) == 0);
1712                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1713                                 (void) fprintf(stderr, gettext("Cannot use "
1714                                     "'%s' as a device for splitting\n"), type);
1715                                 nvlist_free(newroot);
1716                                 return (NULL);
1717                         }
1718                 }
1719         }
1720
1721         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1722                 nvlist_free(newroot);
1723                 return (NULL);
1724         }
1725
1726         return (newroot);
1727 }
1728
1729 /*
1730  * Get and validate the contents of the given vdev specification.  This ensures
1731  * that the nvlist returned is well-formed, that all the devices exist, and that
1732  * they are not currently in use by any other known consumer.  The 'poolconfig'
1733  * parameter is the current configuration of the pool when adding devices
1734  * existing pool, and is used to perform additional checks, such as changing the
1735  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1736  * new pool.  The 'force' flag controls whether devices should be forcefully
1737  * added, even if they appear in use.
1738  */
1739 nvlist_t *
1740 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1741     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1742 {
1743         nvlist_t *newroot;
1744         nvlist_t *poolconfig = NULL;
1745         is_force = force;
1746
1747         /*
1748          * Construct the vdev specification.  If this is successful, we know
1749          * that we have a valid specification, and that all devices can be
1750          * opened.
1751          */
1752         if ((newroot = construct_spec(props, argc, argv)) == NULL)
1753                 return (NULL);
1754
1755         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1756                 nvlist_free(newroot);
1757                 return (NULL);
1758         }
1759
1760         /*
1761          * Validate each device to make sure that its not shared with another
1762          * subsystem.  We do this even if 'force' is set, because there are some
1763          * uses (such as a dedicated dump device) that even '-f' cannot
1764          * override.
1765          */
1766         if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1767                 nvlist_free(newroot);
1768                 return (NULL);
1769         }
1770
1771         /*
1772          * Check the replication level of the given vdevs and report any errors
1773          * found.  We include the existing pool spec, if any, as we need to
1774          * catch changes against the existing replication level.
1775          */
1776         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1777                 nvlist_free(newroot);
1778                 return (NULL);
1779         }
1780
1781         /*
1782          * Run through the vdev specification and label any whole disks found.
1783          */
1784         if (!dryrun && make_disks(zhp, newroot) != 0) {
1785                 nvlist_free(newroot);
1786                 return (NULL);
1787         }
1788
1789         return (newroot);
1790 }