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, 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 int
 368 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
 369 {
 370         int err;
 371         char *value;
 372
 373         /* No valid type detected device is safe to use */
 374         value = blkid_get_tag_value(cache, "TYPE", path);
 375         if (value == NULL)
 376                 return (0);
 377
 378         /*
 379          * If libblkid detects a ZFS device, we check the device
 380          * using check_file() to see if it's safe.  The one safe
 381          * case is a spare device shared between multiple pools.
 382          */
 383         if (strcmp(value, "zfs_member") == 0) {
 384                 err = check_file(path, force, isspare);
 385         } else {
 386                 if (force) {
 387                         err = 0;
 388                 } else {
 389                         err = -1;
 390                         vdev_error(gettext("%s contains a filesystem of "
 391                             "type '%s'\n"), path, value);
 392                 }
 393         }
 394
 395         free(value);
 396
 397         return (err);
 398 }
 399
 400 /*
 401  * Validate that a disk including all partitions are safe to use.
 402  *
 403  * For EFI labeled disks this can done relatively easily with the libefi
 404  * library.  The partition numbers are extracted from the label and used
 405  * to generate the expected /dev/ paths.  Each partition can then be
 406  * checked for conflicts.
 407  *
 408  * For non-EFI labeled disks (MBR/EBR/etc) the same process is possible
 409  * but due to the lack of a readily available libraries this scanning is
 410  * not implemented.  Instead only the device path as given is checked.
 411  */
 412 static int
 413 check_disk(const char *path, blkid_cache cache, int force,
 414     boolean_t isspare, boolean_t iswholedisk)
 415 {
 416         struct dk_gpt *vtoc;
 417         char slice_path[MAXPATHLEN];
 418         int err = 0;
 419         int fd, i;
 420
 421         if (!iswholedisk)
 422                 return (check_slice(path, cache, force, isspare));
 423
 424         if ((fd = open(path, O_RDONLY|O_DIRECT|O_EXCL)) < 0) {
 425                 char *value = blkid_get_tag_value(cache, "TYPE", path);
 426                 (void) fprintf(stderr, gettext("%s is in use and contains "
 427                     "a %s filesystem.\n"), path, value ? value : "unknown");
 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         blkid_cache cache;
 490         int error;
 491
 492         error = blkid_get_cache(&cache, NULL);
 493         if (error != 0) {
 494                 (void) fprintf(stderr, gettext("unable to access the blkid "
 495                     "cache.\n"));
 496                 return (-1);
 497         }
 498
 499         error = check_disk(path, cache, force, isspare, iswholedisk);
 500         blkid_put_cache(cache);
 501
 502         return (error);
 503 }
 504
 505 /*
 506  * This may be a shorthand device path or it could be total gibberish.
 507  * Check to see if it is a known device available in zfs_vdev_paths.
 508  * As part of this check, see if we've been given an entire disk
 509  * (minus the slice number).
 510  */
 511 static int
 512 is_shorthand_path(const char *arg, char *path, size_t path_size,
 513     struct stat64 *statbuf, boolean_t *wholedisk)
 514 {
 515         int error;
 516
 517         error = zfs_resolve_shortname(arg, path, path_size);
 518         if (error == 0) {
 519                 *wholedisk = zfs_dev_is_whole_disk(path);
 520                 if (*wholedisk || (stat64(path, statbuf) == 0))
 521                         return (0);
 522         }
 523
 524         strlcpy(path, arg, path_size);
 525         memset(statbuf, 0, sizeof (*statbuf));
 526         *wholedisk = B_FALSE;
 527
 528         return (error);
 529 }
 530
 531 /*
 532  * Determine if the given path is a hot spare within the given configuration.
 533  * If no configuration is given we rely solely on the label.
 534  */
 535 static boolean_t
 536 is_spare(nvlist_t *config, const char *path)
 537 {
 538         int fd;
 539         pool_state_t state;
 540         char *name = NULL;
 541         nvlist_t *label;
 542         uint64_t guid, spareguid;
 543         nvlist_t *nvroot;
 544         nvlist_t **spares;
 545         uint_t i, nspares;
 546         boolean_t inuse;
 547
 548         if ((fd = open(path, O_RDONLY)) < 0)
 549                 return (B_FALSE);
 550
 551         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
 552             !inuse ||
 553             state != POOL_STATE_SPARE ||
 554             zpool_read_label(fd, &label, NULL) != 0) {
 555                 free(name);
 556                 (void) close(fd);
 557                 return (B_FALSE);
 558         }
 559         free(name);
 560         (void) close(fd);
 561
 562         if (config == NULL) {
 563                 nvlist_free(label);
 564                 return (B_TRUE);
 565         }
 566
 567         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
 568         nvlist_free(label);
 569
 570         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 571             &nvroot) == 0);
 572         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 573             &spares, &nspares) == 0) {
 574                 for (i = 0; i < nspares; i++) {
 575                         verify(nvlist_lookup_uint64(spares[i],
 576                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
 577                         if (spareguid == guid)
 578                                 return (B_TRUE);
 579                 }
 580         }
 581
 582         return (B_FALSE);
 583 }
 584
 585 /*
 586  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 587  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 588  * leaf vdev are:
 589  *
 590  *      /dev/xxx        Complete disk path
 591  *      /xxx            Full path to file
 592  *      xxx             Shorthand for <zfs_vdev_paths>/xxx
 593  */
 594 static nvlist_t *
 595 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
 596 {
 597         char path[MAXPATHLEN];
 598         struct stat64 statbuf;
 599         nvlist_t *vdev = NULL;
 600         char *type = NULL;
 601         boolean_t wholedisk = B_FALSE;
 602         uint64_t ashift = 0;
 603         int err;
 604
 605         /*
 606          * Determine what type of vdev this is, and put the full path into
 607          * 'path'.  We detect whether this is a device of file afterwards by
 608          * checking the st_mode of the file.
 609          */
 610         if (arg[0] == '/') {
 611                 /*
 612                  * Complete device or file path.  Exact type is determined by
 613                  * examining the file descriptor afterwards.  Symbolic links
 614                  * are resolved to their real paths to determine whole disk
 615                  * and S_ISBLK/S_ISREG type checks.  However, we are careful
 616                  * to store the given path as ZPOOL_CONFIG_PATH to ensure we
 617                  * can leverage udev's persistent device labels.
 618                  */
 619                 if (realpath(arg, path) == NULL) {
 620                         (void) fprintf(stderr,
 621                             gettext("cannot resolve path '%s'\n"), arg);
 622                         return (NULL);
 623                 }
 624
 625                 wholedisk = zfs_dev_is_whole_disk(path);
 626                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 627                         (void) fprintf(stderr,
 628                             gettext("cannot open '%s': %s\n"),
 629                             path, strerror(errno));
 630                         return (NULL);
 631                 }
 632
 633                 /* After whole disk check restore original passed path */
 634                 strlcpy(path, arg, sizeof (path));
 635         } else {
 636                 err = is_shorthand_path(arg, path, sizeof (path),
 637                     &statbuf, &wholedisk);
 638                 if (err != 0) {
 639                         /*
 640                          * If we got ENOENT, then the user gave us
 641                          * gibberish, so try to direct them with a
 642                          * reasonable error message.  Otherwise,
 643                          * regurgitate strerror() since it's the best we
 644                          * can do.
 645                          */
 646                         if (err == ENOENT) {
 647                                 (void) fprintf(stderr,
 648                                     gettext("cannot open '%s': no such "
 649                                     "device in %s\n"), arg, DISK_ROOT);
 650                                 (void) fprintf(stderr,
 651                                     gettext("must be a full path or "
 652                                     "shorthand device name\n"));
 653                                 return (NULL);
 654                         } else {
 655                                 (void) fprintf(stderr,
 656                                     gettext("cannot open '%s': %s\n"),
 657                                     path, strerror(errno));
 658                                 return (NULL);
 659                         }
 660                 }
 661         }
 662
 663         /*
 664          * Determine whether this is a device or a file.
 665          */
 666         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 667                 type = VDEV_TYPE_DISK;
 668         } else if (S_ISREG(statbuf.st_mode)) {
 669                 type = VDEV_TYPE_FILE;
 670         } else {
 671                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 672                     "block device or regular file\n"), path);
 673                 return (NULL);
 674         }
 675
 676         /*
 677          * Finally, we have the complete device or file, and we know that it is
 678          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 679          * vdevs have a 'path' element, and devices also have a 'devid' element.
 680          */
 681         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 682         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 683         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 684         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 685         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 686                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 687                     (uint64_t)wholedisk) == 0);
 688
 689         /*
 690          * Override defaults if custom properties are provided.
 691          */
 692         if (props != NULL) {
 693                 char *value = NULL;
 694
 695                 if (nvlist_lookup_string(props,
 696                     zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0) {
 697                         if (zfs_nicestrtonum(NULL, value, &ashift) != 0) {
 698                                 (void) fprintf(stderr,
 699                                     gettext("ashift must be a number.\n"));
 700                                 return (NULL);
 701                         }
 702                         if (ashift != 0 &&
 703                             (ashift < ASHIFT_MIN || ashift > ASHIFT_MAX)) {
 704                                 (void) fprintf(stderr,
 705                                     gettext("invalid 'ashift=%" PRIu64 "' "
 706                                     "property: only values between %" PRId32 " "
 707                                     "and %" PRId32 " are allowed.\n"),
 708                                     ashift, ASHIFT_MIN, ASHIFT_MAX);
 709                                 return (NULL);
 710                         }
 711                 }
 712         }
 713
 714         /*
 715          * If the device is known to incorrectly report its physical sector
 716          * size explicitly provide the known correct value.
 717          */
 718         if (ashift == 0) {
 719                 int sector_size;
 720
 721                 if (check_sector_size_database(path, &sector_size) == B_TRUE)
 722                         ashift = highbit64(sector_size) - 1;
 723         }
 724
 725         if (ashift > 0)
 726                 (void) nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
 727
 728         return (vdev);
 729 }
 730
 731 /*
 732  * Go through and verify the replication level of the pool is consistent.
 733  * Performs the following checks:
 734  *
 735  *      For the new spec, verifies that devices in mirrors and raidz are the
 736  *      same size.
 737  *
 738  *      If the current configuration already has inconsistent replication
 739  *      levels, ignore any other potential problems in the new spec.
 740  *
 741  *      Otherwise, make sure that the current spec (if there is one) and the new
 742  *      spec have consistent replication levels.
 743  */
 744 typedef struct replication_level {
 745         char *zprl_type;
 746         uint64_t zprl_children;
 747         uint64_t zprl_parity;
 748 } replication_level_t;
 749
 750 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 751
 752 static boolean_t
 753 is_raidz_mirror(replication_level_t *a, replication_level_t *b,
 754     replication_level_t **raidz, replication_level_t **mirror)
 755 {
 756         if (strcmp(a->zprl_type, "raidz") == 0 &&
 757             strcmp(b->zprl_type, "mirror") == 0) {
 758                 *raidz = a;
 759                 *mirror = b;
 760                 return (B_TRUE);
 761         }
 762         return (B_FALSE);
 763 }
 764
 765 /*
 766  * Given a list of toplevel vdevs, return the current replication level.  If
 767  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 768  * an error message will be displayed for each self-inconsistent vdev.
 769  */
 770 static replication_level_t *
 771 get_replication(nvlist_t *nvroot, boolean_t fatal)
 772 {
 773         nvlist_t **top;
 774         uint_t t, toplevels;
 775         nvlist_t **child;
 776         uint_t c, children;
 777         nvlist_t *nv;
 778         char *type;
 779         replication_level_t lastrep = {0};
 780         replication_level_t rep;
 781         replication_level_t *ret;
 782         replication_level_t *raidz, *mirror;
 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         for (t = 0; t < toplevels; t++) {
 791                 uint64_t is_log = B_FALSE;
 792
 793                 nv = top[t];
 794
 795                 /*
 796                  * For separate logs we ignore the top level vdev replication
 797                  * constraints.
 798                  */
 799                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 800                 if (is_log)
 801                         continue;
 802
 803                 /* Ignore holes introduced by removing aux devices */
 804                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
 805                 if (strcmp(type, VDEV_TYPE_HOLE) == 0)
 806                         continue;
 807
 808                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 809                     &child, &children) != 0) {
 810                         /*
 811                          * This is a 'file' or 'disk' vdev.
 812                          */
 813                         rep.zprl_type = type;
 814                         rep.zprl_children = 1;
 815                         rep.zprl_parity = 0;
 816                 } else {
 817                         uint64_t vdev_size;
 818
 819                         /*
 820                          * This is a mirror or RAID-Z vdev.  Go through and make
 821                          * sure the contents are all the same (files vs. disks),
 822                          * keeping track of the number of elements in the
 823                          * process.
 824                          *
 825                          * We also check that the size of each vdev (if it can
 826                          * be determined) is the same.
 827                          */
 828                         rep.zprl_type = type;
 829                         rep.zprl_children = 0;
 830
 831                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 832                                 verify(nvlist_lookup_uint64(nv,
 833                                     ZPOOL_CONFIG_NPARITY,
 834                                     &rep.zprl_parity) == 0);
 835                                 assert(rep.zprl_parity != 0);
 836                         } else {
 837                                 rep.zprl_parity = 0;
 838                         }
 839
 840                         /*
 841                          * The 'dontreport' variable indicates that we've
 842                          * already reported an error for this spec, so don't
 843                          * bother doing it again.
 844                          */
 845                         type = NULL;
 846                         dontreport = 0;
 847                         vdev_size = -1ULL;
 848                         for (c = 0; c < children; c++) {
 849                                 nvlist_t *cnv = child[c];
 850                                 char *path;
 851                                 struct stat64 statbuf;
 852                                 uint64_t size = -1ULL;
 853                                 char *childtype;
 854                                 int fd, err;
 855
 856                                 rep.zprl_children++;
 857
 858                                 verify(nvlist_lookup_string(cnv,
 859                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 860
 861                                 /*
 862                                  * If this is a replacing or spare vdev, then
 863                                  * get the real first child of the vdev: do this
 864                                  * in a loop because replacing and spare vdevs
 865                                  * can be nested.
 866                                  */
 867                                 while (strcmp(childtype,
 868                                     VDEV_TYPE_REPLACING) == 0 ||
 869                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 870                                         nvlist_t **rchild;
 871                                         uint_t rchildren;
 872
 873                                         verify(nvlist_lookup_nvlist_array(cnv,
 874                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 875                                             &rchildren) == 0);
 876                                         assert(rchildren == 2);
 877                                         cnv = rchild[0];
 878
 879                                         verify(nvlist_lookup_string(cnv,
 880                                             ZPOOL_CONFIG_TYPE,
 881                                             &childtype) == 0);
 882                                 }
 883
 884                                 verify(nvlist_lookup_string(cnv,
 885                                     ZPOOL_CONFIG_PATH, &path) == 0);
 886
 887                                 /*
 888                                  * If we have a raidz/mirror that combines disks
 889                                  * with files, report it as an error.
 890                                  */
 891                                 if (!dontreport && type != NULL &&
 892                                     strcmp(type, childtype) != 0) {
 893                                         if (ret != NULL)
 894                                                 free(ret);
 895                                         ret = NULL;
 896                                         if (fatal)
 897                                                 vdev_error(gettext(
 898                                                     "mismatched replication "
 899                                                     "level: %s contains both "
 900                                                     "files and devices\n"),
 901                                                     rep.zprl_type);
 902                                         else
 903                                                 return (NULL);
 904                                         dontreport = B_TRUE;
 905                                 }
 906
 907                                 /*
 908                                  * According to stat(2), the value of 'st_size'
 909                                  * is undefined for block devices and character
 910                                  * devices.  But there is no effective way to
 911                                  * determine the real size in userland.
 912                                  *
 913                                  * Instead, we'll take advantage of an
 914                                  * implementation detail of spec_size().  If the
 915                                  * device is currently open, then we (should)
 916                                  * return a valid size.
 917                                  *
 918                                  * If we still don't get a valid size (indicated
 919                                  * by a size of 0 or MAXOFFSET_T), then ignore
 920                                  * this device altogether.
 921                                  */
 922                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 923                                         err = fstat64_blk(fd, &statbuf);
 924                                         (void) close(fd);
 925                                 } else {
 926                                         err = stat64(path, &statbuf);
 927                                 }
 928
 929                                 if (err != 0 ||
 930                                     statbuf.st_size == 0 ||
 931                                     statbuf.st_size == MAXOFFSET_T)
 932                                         continue;
 933
 934                                 size = statbuf.st_size;
 935
 936                                 /*
 937                                  * Also make sure that devices and
 938                                  * slices have a consistent size.  If
 939                                  * they differ by a significant amount
 940                                  * (~16MB) then report an error.
 941                                  */
 942                                 if (!dontreport &&
 943                                     (vdev_size != -1ULL &&
 944                                     (labs(size - vdev_size) >
 945                                     ZPOOL_FUZZ))) {
 946                                         if (ret != NULL)
 947                                                 free(ret);
 948                                         ret = NULL;
 949                                         if (fatal)
 950                                                 vdev_error(gettext(
 951                                                     "%s contains devices of "
 952                                                     "different sizes\n"),
 953                                                     rep.zprl_type);
 954                                         else
 955                                                 return (NULL);
 956                                         dontreport = B_TRUE;
 957                                 }
 958
 959                                 type = childtype;
 960                                 vdev_size = size;
 961                         }
 962                 }
 963
 964                 /*
 965                  * At this point, we have the replication of the last toplevel
 966                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 967                  * different.
 968                  */
 969                 if (lastrep.zprl_type != NULL) {
 970                         if (is_raidz_mirror(&lastrep, &rep, &raidz, &mirror) ||
 971                             is_raidz_mirror(&rep, &lastrep, &raidz, &mirror)) {
 972                                 /*
 973                                  * Accepted raidz and mirror when they can
 974                                  * handle the same number of disk failures.
 975                                  */
 976                                 if (raidz->zprl_parity !=
 977                                     mirror->zprl_children - 1) {
 978                                         if (ret != NULL)
 979                                                 free(ret);
 980                                         ret = NULL;
 981                                         if (fatal)
 982                                                 vdev_error(gettext(
 983                                                     "mismatched replication "
 984                                                     "level: "
 985                                                     "%s and %s vdevs with "
 986                                                     "different redundancy, "
 987                                                     "%llu vs. %llu (%llu-way) "
 988                                                     "are present\n"),
 989                                                     raidz->zprl_type,
 990                                                     mirror->zprl_type,
 991                                                     raidz->zprl_parity,
 992                                                     mirror->zprl_children - 1,
 993                                                     mirror->zprl_children);
 994                                         else
 995                                                 return (NULL);
 996                                 }
 997                         } else if (strcmp(lastrep.zprl_type, rep.zprl_type) !=
 998                             0) {
 999                                 if (ret != NULL)
1000                                         free(ret);
1001                                 ret = NULL;
1002                                 if (fatal)
1003                                         vdev_error(gettext(
1004                                             "mismatched replication level: "
1005                                             "both %s and %s vdevs are "
1006                                             "present\n"),
1007                                             lastrep.zprl_type, rep.zprl_type);
1008                                 else
1009                                         return (NULL);
1010                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
1011                                 if (ret)
1012                                         free(ret);
1013                                 ret = NULL;
1014                                 if (fatal)
1015                                         vdev_error(gettext(
1016                                             "mismatched replication level: "
1017                                             "both %llu and %llu device parity "
1018                                             "%s vdevs are present\n"),
1019                                             lastrep.zprl_parity,
1020                                             rep.zprl_parity,
1021                                             rep.zprl_type);
1022                                 else
1023                                         return (NULL);
1024                         } else if (lastrep.zprl_children != rep.zprl_children) {
1025                                 if (ret)
1026                                         free(ret);
1027                                 ret = NULL;
1028                                 if (fatal)
1029                                         vdev_error(gettext(
1030                                             "mismatched replication level: "
1031                                             "both %llu-way and %llu-way %s "
1032                                             "vdevs are present\n"),
1033                                             lastrep.zprl_children,
1034                                             rep.zprl_children,
1035                                             rep.zprl_type);
1036                                 else
1037                                         return (NULL);
1038                         }
1039                 }
1040                 lastrep = rep;
1041         }
1042
1043         if (ret != NULL)
1044                 *ret = rep;
1045
1046         return (ret);
1047 }
1048
1049 /*
1050  * Check the replication level of the vdev spec against the current pool.  Calls
1051  * get_replication() to make sure the new spec is self-consistent.  If the pool
1052  * has a consistent replication level, then we ignore any errors.  Otherwise,
1053  * report any difference between the two.
1054  */
1055 static int
1056 check_replication(nvlist_t *config, nvlist_t *newroot)
1057 {
1058         nvlist_t **child;
1059         uint_t  children;
1060         replication_level_t *current = NULL, *new;
1061         replication_level_t *raidz, *mirror;
1062         int ret;
1063
1064         /*
1065          * If we have a current pool configuration, check to see if it's
1066          * self-consistent.  If not, simply return success.
1067          */
1068         if (config != NULL) {
1069                 nvlist_t *nvroot;
1070
1071                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1072                     &nvroot) == 0);
1073                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1074                         return (0);
1075         }
1076         /*
1077          * for spares there may be no children, and therefore no
1078          * replication level to check
1079          */
1080         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1081             &child, &children) != 0) || (children == 0)) {
1082                 free(current);
1083                 return (0);
1084         }
1085
1086         /*
1087          * If all we have is logs then there's no replication level to check.
1088          */
1089         if (num_logs(newroot) == children) {
1090                 free(current);
1091                 return (0);
1092         }
1093
1094         /*
1095          * Get the replication level of the new vdev spec, reporting any
1096          * inconsistencies found.
1097          */
1098         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1099                 free(current);
1100                 return (-1);
1101         }
1102
1103         /*
1104          * Check to see if the new vdev spec matches the replication level of
1105          * the current pool.
1106          */
1107         ret = 0;
1108         if (current != NULL) {
1109                 if (is_raidz_mirror(current, new, &raidz, &mirror) ||
1110                     is_raidz_mirror(new, current, &raidz, &mirror)) {
1111                         if (raidz->zprl_parity != mirror->zprl_children - 1) {
1112                                 vdev_error(gettext(
1113                                     "mismatched replication level: pool and "
1114                                     "new vdev with different redundancy, %s "
1115                                     "and %s vdevs, %llu vs. %llu (%llu-way)\n"),
1116                                     raidz->zprl_type,
1117                                     mirror->zprl_type,
1118                                     raidz->zprl_parity,
1119                                     mirror->zprl_children - 1,
1120                                     mirror->zprl_children);
1121                                 ret = -1;
1122                         }
1123                 } else if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1124                         vdev_error(gettext(
1125                             "mismatched replication level: pool uses %s "
1126                             "and new vdev is %s\n"),
1127                             current->zprl_type, new->zprl_type);
1128                         ret = -1;
1129                 } else if (current->zprl_parity != new->zprl_parity) {
1130                         vdev_error(gettext(
1131                             "mismatched replication level: pool uses %llu "
1132                             "device parity and new vdev uses %llu\n"),
1133                             current->zprl_parity, new->zprl_parity);
1134                         ret = -1;
1135                 } else if (current->zprl_children != new->zprl_children) {
1136                         vdev_error(gettext(
1137                             "mismatched replication level: pool uses %llu-way "
1138                             "%s and new vdev uses %llu-way %s\n"),
1139                             current->zprl_children, current->zprl_type,
1140                             new->zprl_children, new->zprl_type);
1141                         ret = -1;
1142                 }
1143         }
1144
1145         free(new);
1146         if (current != NULL)
1147                 free(current);
1148
1149         return (ret);
1150 }
1151
1152 static int
1153 zero_label(char *path)
1154 {
1155         const int size = 4096;
1156         char buf[size];
1157         int err, fd;
1158
1159         if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1160                 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1161                     path, strerror(errno));
1162                 return (-1);
1163         }
1164
1165         memset(buf, 0, size);
1166         err = write(fd, buf, size);
1167         (void) fdatasync(fd);
1168         (void) close(fd);
1169
1170         if (err == -1) {
1171                 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1172                     "of '%s': %s\n"), size, path, strerror(errno));
1173                 return (-1);
1174         }
1175
1176         if (err != size) {
1177                 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1178                     "of '%s'\n"), err, size, path);
1179                 return (-1);
1180         }
1181
1182         return (0);
1183 }
1184
1185 /*
1186  * Go through and find any whole disks in the vdev specification, labelling them
1187  * as appropriate.  When constructing the vdev spec, we were unable to open this
1188  * device in order to provide a devid.  Now that we have labelled the disk and
1189  * know that slice 0 is valid, we can construct the devid now.
1190  *
1191  * If the disk was already labeled with an EFI label, we will have gotten the
1192  * devid already (because we were able to open the whole disk).  Otherwise, we
1193  * need to get the devid after we label the disk.
1194  */
1195 static int
1196 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1197 {
1198         nvlist_t **child;
1199         uint_t c, children;
1200         char *type, *path;
1201         char devpath[MAXPATHLEN];
1202         char udevpath[MAXPATHLEN];
1203         uint64_t wholedisk;
1204         struct stat64 statbuf;
1205         int is_exclusive = 0;
1206         int fd;
1207         int ret;
1208
1209         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1210
1211         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1212             &child, &children) != 0) {
1213
1214                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1215                         return (0);
1216
1217                 /*
1218                  * We have a disk device.  If this is a whole disk write
1219                  * out the efi partition table, otherwise write zero's to
1220                  * the first 4k of the partition.  This is to ensure that
1221                  * libblkid will not misidentify the partition due to a
1222                  * magic value left by the previous filesystem.
1223                  */
1224                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1225                 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1226                     &wholedisk));
1227
1228                 if (!wholedisk) {
1229                         /*
1230                          * Update device id string for mpath nodes (Linux only)
1231                          */
1232                         if (is_mpath_whole_disk(path))
1233                                 update_vdev_config_dev_strs(nv);
1234
1235                         if (!is_spare(NULL, path))
1236                                 (void) zero_label(path);
1237                         return (0);
1238                 }
1239
1240                 if (realpath(path, devpath) == NULL) {
1241                         ret = errno;
1242                         (void) fprintf(stderr,
1243                             gettext("cannot resolve path '%s'\n"), path);
1244                         return (ret);
1245                 }
1246
1247                 /*
1248                  * Remove any previously existing symlink from a udev path to
1249                  * the device before labeling the disk.  This ensures that
1250                  * only newly created links are used.  Otherwise there is a
1251                  * window between when udev deletes and recreates the link
1252                  * during which access attempts will fail with ENOENT.
1253                  */
1254                 strlcpy(udevpath, path, MAXPATHLEN);
1255                 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1256
1257                 fd = open(devpath, O_RDWR|O_EXCL);
1258                 if (fd == -1) {
1259                         if (errno == EBUSY)
1260                                 is_exclusive = 1;
1261                 } else {
1262                         (void) close(fd);
1263                 }
1264
1265                 /*
1266                  * If the partition exists, contains a valid spare label,
1267                  * and is opened exclusively there is no need to partition
1268                  * it.  Hot spares have already been partitioned and are
1269                  * held open exclusively by the kernel as a safety measure.
1270                  *
1271                  * If the provided path is for a /dev/disk/ device its
1272                  * symbolic link will be removed, partition table created,
1273                  * and then block until udev creates the new link.
1274                  */
1275                 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1276                         char *devnode = strrchr(devpath, '/') + 1;
1277
1278                         ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1279                         if (ret == 0) {
1280                                 ret = lstat64(udevpath, &statbuf);
1281                                 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1282                                         (void) unlink(udevpath);
1283                         }
1284
1285                         /*
1286                          * When labeling a pool the raw device node name
1287                          * is provided as it appears under /dev/.
1288                          */
1289                         if (zpool_label_disk(g_zfs, zhp, devnode) == -1)
1290                                 return (-1);
1291
1292                         /*
1293                          * Wait for udev to signal the device is available
1294                          * by the provided path.
1295                          */
1296                         ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1297                         if (ret) {
1298                                 (void) fprintf(stderr,
1299                                     gettext("missing link: %s was "
1300                                     "partitioned but %s is missing\n"),
1301                                     devnode, udevpath);
1302                                 return (ret);
1303                         }
1304
1305                         ret = zero_label(udevpath);
1306                         if (ret)
1307                                 return (ret);
1308                 }
1309
1310                 /*
1311                  * Update the path to refer to the partition.  The presence of
1312                  * the 'whole_disk' field indicates to the CLI that we should
1313                  * chop off the partition number when displaying the device in
1314                  * future output.
1315                  */
1316                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1317
1318                 /*
1319                  * Update device id strings for whole disks (Linux only)
1320                  */
1321                 update_vdev_config_dev_strs(nv);
1322
1323                 return (0);
1324         }
1325
1326         for (c = 0; c < children; c++)
1327                 if ((ret = make_disks(zhp, child[c])) != 0)
1328                         return (ret);
1329
1330         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1331             &child, &children) == 0)
1332                 for (c = 0; c < children; c++)
1333                         if ((ret = make_disks(zhp, child[c])) != 0)
1334                                 return (ret);
1335
1336         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1337             &child, &children) == 0)
1338                 for (c = 0; c < children; c++)
1339                         if ((ret = make_disks(zhp, child[c])) != 0)
1340                                 return (ret);
1341
1342         return (0);
1343 }
1344
1345 /*
1346  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1347  * the majority of this task.
1348  */
1349 static boolean_t
1350 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1351     boolean_t replacing, boolean_t isspare)
1352 {
1353         nvlist_t **child;
1354         uint_t c, children;
1355         char *type, *path;
1356         int ret = 0;
1357         char buf[MAXPATHLEN];
1358         uint64_t wholedisk = B_FALSE;
1359         boolean_t anyinuse = B_FALSE;
1360
1361         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1362
1363         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1364             &child, &children) != 0) {
1365
1366                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1367                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1368                         verify(!nvlist_lookup_uint64(nv,
1369                             ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1370
1371                 /*
1372                  * As a generic check, we look to see if this is a replace of a
1373                  * hot spare within the same pool.  If so, we allow it
1374                  * regardless of what libblkid or zpool_in_use() says.
1375                  */
1376                 if (replacing) {
1377                         (void) strlcpy(buf, path, sizeof (buf));
1378                         if (wholedisk) {
1379                                 ret = zfs_append_partition(buf,  sizeof (buf));
1380                                 if (ret == -1)
1381                                         return (-1);
1382                         }
1383
1384                         if (is_spare(config, buf))
1385                                 return (B_FALSE);
1386                 }
1387
1388                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1389                         ret = check_device(path, force, isspare, wholedisk);
1390
1391                 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1392                         ret = check_file(path, force, isspare);
1393
1394                 return (ret != 0);
1395         }
1396
1397         for (c = 0; c < children; c++)
1398                 if (is_device_in_use(config, child[c], force, replacing,
1399                     B_FALSE))
1400                         anyinuse = B_TRUE;
1401
1402         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1403             &child, &children) == 0)
1404                 for (c = 0; c < children; c++)
1405                         if (is_device_in_use(config, child[c], force, replacing,
1406                             B_TRUE))
1407                                 anyinuse = B_TRUE;
1408
1409         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1410             &child, &children) == 0)
1411                 for (c = 0; c < children; c++)
1412                         if (is_device_in_use(config, child[c], force, replacing,
1413                             B_FALSE))
1414                                 anyinuse = B_TRUE;
1415
1416         return (anyinuse);
1417 }
1418
1419 static const char *
1420 is_grouping(const char *type, int *mindev, int *maxdev)
1421 {
1422         if (strncmp(type, "raidz", 5) == 0) {
1423                 const char *p = type + 5;
1424                 char *end;
1425                 long nparity;
1426
1427                 if (*p == '\0') {
1428                         nparity = 1;
1429                 } else if (*p == '0') {
1430                         return (NULL); /* no zero prefixes allowed */
1431                 } else {
1432                         errno = 0;
1433                         nparity = strtol(p, &end, 10);
1434                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1435                             *end != '\0')
1436                                 return (NULL);
1437                 }
1438
1439                 if (mindev != NULL)
1440                         *mindev = nparity + 1;
1441                 if (maxdev != NULL)
1442                         *maxdev = 255;
1443                 return (VDEV_TYPE_RAIDZ);
1444         }
1445
1446         if (maxdev != NULL)
1447                 *maxdev = INT_MAX;
1448
1449         if (strcmp(type, "mirror") == 0) {
1450                 if (mindev != NULL)
1451                         *mindev = 2;
1452                 return (VDEV_TYPE_MIRROR);
1453         }
1454
1455         if (strcmp(type, "spare") == 0) {
1456                 if (mindev != NULL)
1457                         *mindev = 1;
1458                 return (VDEV_TYPE_SPARE);
1459         }
1460
1461         if (strcmp(type, "log") == 0) {
1462                 if (mindev != NULL)
1463                         *mindev = 1;
1464                 return (VDEV_TYPE_LOG);
1465         }
1466
1467         if (strcmp(type, "cache") == 0) {
1468                 if (mindev != NULL)
1469                         *mindev = 1;
1470                 return (VDEV_TYPE_L2CACHE);
1471         }
1472
1473         return (NULL);
1474 }
1475
1476 /*
1477  * Construct a syntactically valid vdev specification,
1478  * and ensure that all devices and files exist and can be opened.
1479  * Note: we don't bother freeing anything in the error paths
1480  * because the program is just going to exit anyway.
1481  */
1482 nvlist_t *
1483 construct_spec(nvlist_t *props, int argc, char **argv)
1484 {
1485         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1486         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1487         const char *type;
1488         uint64_t is_log;
1489         boolean_t seen_logs;
1490
1491         top = NULL;
1492         toplevels = 0;
1493         spares = NULL;
1494         l2cache = NULL;
1495         nspares = 0;
1496         nlogs = 0;
1497         nl2cache = 0;
1498         is_log = B_FALSE;
1499         seen_logs = B_FALSE;
1500         nvroot = NULL;
1501
1502         while (argc > 0) {
1503                 nv = NULL;
1504
1505                 /*
1506                  * If it's a mirror or raidz, the subsequent arguments are
1507                  * its leaves -- until we encounter the next mirror or raidz.
1508                  */
1509                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1510                         nvlist_t **child = NULL;
1511                         int c, children = 0;
1512
1513                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1514                                 if (spares != NULL) {
1515                                         (void) fprintf(stderr,
1516                                             gettext("invalid vdev "
1517                                             "specification: 'spare' can be "
1518                                             "specified only once\n"));
1519                                         goto spec_out;
1520                                 }
1521                                 is_log = B_FALSE;
1522                         }
1523
1524                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1525                                 if (seen_logs) {
1526                                         (void) fprintf(stderr,
1527                                             gettext("invalid vdev "
1528                                             "specification: 'log' can be "
1529                                             "specified only once\n"));
1530                                         goto spec_out;
1531                                 }
1532                                 seen_logs = B_TRUE;
1533                                 is_log = B_TRUE;
1534                                 argc--;
1535                                 argv++;
1536                                 /*
1537                                  * A log is not a real grouping device.
1538                                  * We just set is_log and continue.
1539                                  */
1540                                 continue;
1541                         }
1542
1543                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1544                                 if (l2cache != NULL) {
1545                                         (void) fprintf(stderr,
1546                                             gettext("invalid vdev "
1547                                             "specification: 'cache' can be "
1548                                             "specified only once\n"));
1549                                         goto spec_out;
1550                                 }
1551                                 is_log = B_FALSE;
1552                         }
1553
1554                         if (is_log) {
1555                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1556                                         (void) fprintf(stderr,
1557                                             gettext("invalid vdev "
1558                                             "specification: unsupported 'log' "
1559                                             "device: %s\n"), type);
1560                                         goto spec_out;
1561                                 }
1562                                 nlogs++;
1563                         }
1564
1565                         for (c = 1; c < argc; c++) {
1566                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1567                                         break;
1568                                 children++;
1569                                 child = realloc(child,
1570                                     children * sizeof (nvlist_t *));
1571                                 if (child == NULL)
1572                                         zpool_no_memory();
1573                                 if ((nv = make_leaf_vdev(props, argv[c],
1574                                     B_FALSE)) == NULL) {
1575                                         for (c = 0; c < children - 1; c++)
1576                                                 nvlist_free(child[c]);
1577                                         free(child);
1578                                         goto spec_out;
1579                                 }
1580
1581                                 child[children - 1] = nv;
1582                         }
1583
1584                         if (children < mindev) {
1585                                 (void) fprintf(stderr, gettext("invalid vdev "
1586                                     "specification: %s requires at least %d "
1587                                     "devices\n"), argv[0], mindev);
1588                                 for (c = 0; c < children; c++)
1589                                         nvlist_free(child[c]);
1590                                 free(child);
1591                                 goto spec_out;
1592                         }
1593
1594                         if (children > maxdev) {
1595                                 (void) fprintf(stderr, gettext("invalid vdev "
1596                                     "specification: %s supports no more than "
1597                                     "%d devices\n"), argv[0], maxdev);
1598                                 for (c = 0; c < children; c++)
1599                                         nvlist_free(child[c]);
1600                                 free(child);
1601                                 goto spec_out;
1602                         }
1603
1604                         argc -= c;
1605                         argv += c;
1606
1607                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1608                                 spares = child;
1609                                 nspares = children;
1610                                 continue;
1611                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1612                                 l2cache = child;
1613                                 nl2cache = children;
1614                                 continue;
1615                         } else {
1616                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1617                                     0) == 0);
1618                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1619                                     type) == 0);
1620                                 verify(nvlist_add_uint64(nv,
1621                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1622                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1623                                         verify(nvlist_add_uint64(nv,
1624                                             ZPOOL_CONFIG_NPARITY,
1625                                             mindev - 1) == 0);
1626                                 }
1627                                 verify(nvlist_add_nvlist_array(nv,
1628                                     ZPOOL_CONFIG_CHILDREN, child,
1629                                     children) == 0);
1630
1631                                 for (c = 0; c < children; c++)
1632                                         nvlist_free(child[c]);
1633                                 free(child);
1634                         }
1635                 } else {
1636                         /*
1637                          * We have a device.  Pass off to make_leaf_vdev() to
1638                          * construct the appropriate nvlist describing the vdev.
1639                          */
1640                         if ((nv = make_leaf_vdev(props, argv[0],
1641                             is_log)) == NULL)
1642                                 goto spec_out;
1643
1644                         if (is_log)
1645                                 nlogs++;
1646                         argc--;
1647                         argv++;
1648                 }
1649
1650                 toplevels++;
1651                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1652                 if (top == NULL)
1653                         zpool_no_memory();
1654                 top[toplevels - 1] = nv;
1655         }
1656
1657         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1658                 (void) fprintf(stderr, gettext("invalid vdev "
1659                     "specification: at least one toplevel vdev must be "
1660                     "specified\n"));
1661                 goto spec_out;
1662         }
1663
1664         if (seen_logs && nlogs == 0) {
1665                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1666                     "log requires at least 1 device\n"));
1667                 goto spec_out;
1668         }
1669
1670         /*
1671          * Finally, create nvroot and add all top-level vdevs to it.
1672          */
1673         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1674         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1675             VDEV_TYPE_ROOT) == 0);
1676         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1677             top, toplevels) == 0);
1678         if (nspares != 0)
1679                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1680                     spares, nspares) == 0);
1681         if (nl2cache != 0)
1682                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1683                     l2cache, nl2cache) == 0);
1684
1685 spec_out:
1686         for (t = 0; t < toplevels; t++)
1687                 nvlist_free(top[t]);
1688         for (t = 0; t < nspares; t++)
1689                 nvlist_free(spares[t]);
1690         for (t = 0; t < nl2cache; t++)
1691                 nvlist_free(l2cache[t]);
1692
1693         free(spares);
1694         free(l2cache);
1695         free(top);
1696
1697         return (nvroot);
1698 }
1699
1700 nvlist_t *
1701 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1702     splitflags_t flags, int argc, char **argv)
1703 {
1704         nvlist_t *newroot = NULL, **child;
1705         uint_t c, children;
1706
1707         if (argc > 0) {
1708                 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1709                         (void) fprintf(stderr, gettext("Unable to build a "
1710                             "pool from the specified devices\n"));
1711                         return (NULL);
1712                 }
1713
1714                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1715                         nvlist_free(newroot);
1716                         return (NULL);
1717                 }
1718
1719                 /* avoid any tricks in the spec */
1720                 verify(nvlist_lookup_nvlist_array(newroot,
1721                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1722                 for (c = 0; c < children; c++) {
1723                         char *path;
1724                         const char *type;
1725                         int min, max;
1726
1727                         verify(nvlist_lookup_string(child[c],
1728                             ZPOOL_CONFIG_PATH, &path) == 0);
1729                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1730                                 (void) fprintf(stderr, gettext("Cannot use "
1731                                     "'%s' as a device for splitting\n"), type);
1732                                 nvlist_free(newroot);
1733                                 return (NULL);
1734                         }
1735                 }
1736         }
1737
1738         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1739                 nvlist_free(newroot);
1740                 return (NULL);
1741         }
1742
1743         return (newroot);
1744 }
1745
1746 /*
1747  * Get and validate the contents of the given vdev specification.  This ensures
1748  * that the nvlist returned is well-formed, that all the devices exist, and that
1749  * they are not currently in use by any other known consumer.  The 'poolconfig'
1750  * parameter is the current configuration of the pool when adding devices
1751  * existing pool, and is used to perform additional checks, such as changing the
1752  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1753  * new pool.  The 'force' flag controls whether devices should be forcefully
1754  * added, even if they appear in use.
1755  */
1756 nvlist_t *
1757 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1758     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1759 {
1760         nvlist_t *newroot;
1761         nvlist_t *poolconfig = NULL;
1762         is_force = force;
1763
1764         /*
1765          * Construct the vdev specification.  If this is successful, we know
1766          * that we have a valid specification, and that all devices can be
1767          * opened.
1768          */
1769         if ((newroot = construct_spec(props, argc, argv)) == NULL)
1770                 return (NULL);
1771
1772         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1773                 nvlist_free(newroot);
1774                 return (NULL);
1775         }
1776
1777         /*
1778          * Validate each device to make sure that its not shared with another
1779          * subsystem.  We do this even if 'force' is set, because there are some
1780          * uses (such as a dedicated dump device) that even '-f' cannot
1781          * override.
1782          */
1783         if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1784                 nvlist_free(newroot);
1785                 return (NULL);
1786         }
1787
1788         /*
1789          * Check the replication level of the given vdevs and report any errors
1790          * found.  We include the existing pool spec, if any, as we need to
1791          * catch changes against the existing replication level.
1792          */
1793         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1794                 nvlist_free(newroot);
1795                 return (NULL);
1796         }
1797
1798         /*
1799          * Run through the vdev specification and label any whole disks found.
1800          */
1801         if (!dryrun && make_disks(zhp, newroot) != 0) {
1802                 nvlist_free(newroot);
1803                 return (NULL);
1804         }
1805
1806         return (newroot);
1807 }