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 SSDSC2BW48", 4096},
 192         {"ATA     INTEL SSDSC2BP24", 4096},
 193         {"ATA     INTEL SSDSC2BP48", 4096},
 194         {"NA      SmrtStorSDLKAE9W", 4096},
 195         /* Imported from Open Solaris */
 196         {"ATA     MARVELL SD88SA02", 4096},
 197         /* Advanced format Hard drives */
 198         {"ATA     Hitachi HDS5C303", 4096},
 199         {"ATA     SAMSUNG HD204UI ", 4096},
 200         {"ATA     ST2000DL004 HD20", 4096},
 201         {"ATA     WDC WD10EARS-00M", 4096},
 202         {"ATA     WDC WD10EARS-00S", 4096},
 203         {"ATA     WDC WD10EARS-00Z", 4096},
 204         {"ATA     WDC WD15EARS-00M", 4096},
 205         {"ATA     WDC WD15EARS-00S", 4096},
 206         {"ATA     WDC WD15EARS-00Z", 4096},
 207         {"ATA     WDC WD20EARS-00M", 4096},
 208         {"ATA     WDC WD20EARS-00S", 4096},
 209         {"ATA     WDC WD20EARS-00Z", 4096},
 210         {"ATA     WDC WD1600BEVT-0", 4096},
 211         {"ATA     WDC WD2500BEVT-0", 4096},
 212         {"ATA     WDC WD3200BEVT-0", 4096},
 213         {"ATA     WDC WD5000BEVT-0", 4096},
 214         /* Virtual disks: Assume zvols with default volblocksize */
 215 #if 0
 216         {"ATA     QEMU HARDDISK   ", 8192},
 217         {"IET     VIRTUAL-DISK    ", 8192},
 218         {"OI      COMSTAR         ", 8192},
 219         {"SUN     COMSTAR         ", 8192},
 220         {"NETAPP  LUN             ", 8192},
 221 #endif
 222 };
 223
 224 static const int vdev_disk_database_size =
 225         sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
 226
 227 #define INQ_REPLY_LEN   96
 228 #define INQ_CMD_LEN     6
 229
 230 static boolean_t
 231 check_sector_size_database(char *path, int *sector_size)
 232 {
 233         unsigned char inq_buff[INQ_REPLY_LEN];
 234         unsigned char sense_buffer[32];
 235         unsigned char inq_cmd_blk[INQ_CMD_LEN] =
 236             {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
 237         sg_io_hdr_t io_hdr;
 238         int error;
 239         int fd;
 240         int i;
 241
 242         /* Prepare INQUIRY command */
 243         memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
 244         io_hdr.interface_id = 'S';
 245         io_hdr.cmd_len = sizeof (inq_cmd_blk);
 246         io_hdr.mx_sb_len = sizeof (sense_buffer);
 247         io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 248         io_hdr.dxfer_len = INQ_REPLY_LEN;
 249         io_hdr.dxferp = inq_buff;
 250         io_hdr.cmdp = inq_cmd_blk;
 251         io_hdr.sbp = sense_buffer;
 252         io_hdr.timeout = 10;            /* 10 milliseconds is ample time */
 253
 254         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 255                 return (B_FALSE);
 256
 257         error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
 258
 259         (void) close(fd);
 260
 261         if (error < 0)
 262                 return (B_FALSE);
 263
 264         if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
 265                 return (B_FALSE);
 266
 267         for (i = 0; i < vdev_disk_database_size; i++) {
 268                 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
 269                         continue;
 270
 271                 *sector_size = vdev_disk_database[i].sector_size;
 272                 return (B_TRUE);
 273         }
 274
 275         return (B_FALSE);
 276 }
 277
 278 /*PRINTFLIKE1*/
 279 static void
 280 vdev_error(const char *fmt, ...)
 281 {
 282         va_list ap;
 283
 284         if (!error_seen) {
 285                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
 286                 if (!is_force)
 287                         (void) fprintf(stderr, gettext("use '-f' to override "
 288                             "the following errors:\n"));
 289                 else
 290                         (void) fprintf(stderr, gettext("the following errors "
 291                             "must be manually repaired:\n"));
 292                 error_seen = B_TRUE;
 293         }
 294
 295         va_start(ap, fmt);
 296         (void) vfprintf(stderr, fmt, ap);
 297         va_end(ap);
 298 }
 299
 300 /*
 301  * Check that a file is valid.  All we can do in this case is check that it's
 302  * not in use by another pool, and not in use by swap.
 303  */
 304 static int
 305 check_file(const char *file, boolean_t force, boolean_t isspare)
 306 {
 307         char  *name;
 308         int fd;
 309         int ret = 0;
 310         pool_state_t state;
 311         boolean_t inuse;
 312
 313         if ((fd = open(file, O_RDONLY)) < 0)
 314                 return (0);
 315
 316         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
 317                 const char *desc;
 318
 319                 switch (state) {
 320                 case POOL_STATE_ACTIVE:
 321                         desc = gettext("active");
 322                         break;
 323
 324                 case POOL_STATE_EXPORTED:
 325                         desc = gettext("exported");
 326                         break;
 327
 328                 case POOL_STATE_POTENTIALLY_ACTIVE:
 329                         desc = gettext("potentially active");
 330                         break;
 331
 332                 default:
 333                         desc = gettext("unknown");
 334                         break;
 335                 }
 336
 337                 /*
 338                  * Allow hot spares to be shared between pools.
 339                  */
 340                 if (state == POOL_STATE_SPARE && isspare) {
 341                         free(name);
 342                         (void) close(fd);
 343                         return (0);
 344                 }
 345
 346                 if (state == POOL_STATE_ACTIVE ||
 347                     state == POOL_STATE_SPARE || !force) {
 348                         switch (state) {
 349                         case POOL_STATE_SPARE:
 350                                 vdev_error(gettext("%s is reserved as a hot "
 351                                     "spare for pool %s\n"), file, name);
 352                                 break;
 353                         default:
 354                                 vdev_error(gettext("%s is part of %s pool "
 355                                     "'%s'\n"), file, desc, name);
 356                                 break;
 357                         }
 358                         ret = -1;
 359                 }
 360
 361                 free(name);
 362         }
 363
 364         (void) close(fd);
 365         return (ret);
 366 }
 367
 368 static int
 369 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
 370 {
 371         int err;
 372         char *value;
 373
 374         /* No valid type detected device is safe to use */
 375         value = blkid_get_tag_value(cache, "TYPE", path);
 376         if (value == NULL)
 377                 return (0);
 378
 379         /*
 380          * If libblkid detects a ZFS device, we check the device
 381          * using check_file() to see if it's safe.  The one safe
 382          * case is a spare device shared between multiple pools.
 383          */
 384         if (strcmp(value, "zfs_member") == 0) {
 385                 err = check_file(path, force, isspare);
 386         } else {
 387                 if (force) {
 388                         err = 0;
 389                 } else {
 390                         err = -1;
 391                         vdev_error(gettext("%s contains a filesystem of "
 392                             "type '%s'\n"), path, value);
 393                 }
 394         }
 395
 396         free(value);
 397
 398         return (err);
 399 }
 400
 401 /*
 402  * Validate that a disk including all partitions are safe to use.
 403  *
 404  * For EFI labeled disks this can done relatively easily with the libefi
 405  * library.  The partition numbers are extracted from the label and used
 406  * to generate the expected /dev/ paths.  Each partition can then be
 407  * checked for conflicts.
 408  *
 409  * For non-EFI labeled disks (MBR/EBR/etc) the same process is possible
 410  * but due to the lack of a readily available libraries this scanning is
 411  * not implemented.  Instead only the device path as given is checked.
 412  */
 413 static int
 414 check_disk(const char *path, blkid_cache cache, int force,
 415     boolean_t isspare, boolean_t iswholedisk)
 416 {
 417         struct dk_gpt *vtoc;
 418         char slice_path[MAXPATHLEN];
 419         int err = 0;
 420         int fd, i;
 421
 422         if (!iswholedisk)
 423                 return (check_slice(path, cache, force, isspare));
 424
 425         if ((fd = open(path, O_RDONLY|O_DIRECT|O_EXCL)) < 0) {
 426                 char *value = blkid_get_tag_value(cache, "TYPE", path);
 427                 (void) fprintf(stderr, gettext("%s is in use and contains "
 428                     "a %s filesystem.\n"), path, value ? value : "unknown");
 429                 return (-1);
 430         }
 431
 432         /*
 433          * Expected to fail for non-EFI labled disks.  Just check the device
 434          * as given and do not attempt to detect and scan partitions.
 435          */
 436         err = efi_alloc_and_read(fd, &vtoc);
 437         if (err) {
 438                 (void) close(fd);
 439                 return (check_slice(path, cache, force, isspare));
 440         }
 441
 442         /*
 443          * The primary efi partition label is damaged however the secondary
 444          * label at the end of the device is intact.  Rather than use this
 445          * label we should play it safe and treat this as a non efi device.
 446          */
 447         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
 448                 efi_free(vtoc);
 449                 (void) close(fd);
 450
 451                 if (force) {
 452                         /* Partitions will now be created using the backup */
 453                         return (0);
 454                 } else {
 455                         vdev_error(gettext("%s contains a corrupt primary "
 456                             "EFI label.\n"), path);
 457                         return (-1);
 458                 }
 459         }
 460
 461         for (i = 0; i < vtoc->efi_nparts; i++) {
 462
 463                 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
 464                     uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
 465                         continue;
 466
 467                 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
 468                         (void) snprintf(slice_path, sizeof (slice_path),
 469                             "%s%s%d", path, "-part", i+1);
 470                 else
 471                         (void) snprintf(slice_path, sizeof (slice_path),
 472                             "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
 473                             "p" : "", i+1);
 474
 475                 err = check_slice(slice_path, cache, force, isspare);
 476                 if (err)
 477                         break;
 478         }
 479
 480         efi_free(vtoc);
 481         (void) close(fd);
 482
 483         return (err);
 484 }
 485
 486 static int
 487 check_device(const char *path, boolean_t force,
 488     boolean_t isspare, boolean_t iswholedisk)
 489 {
 490         blkid_cache cache;
 491         int error;
 492
 493         error = blkid_get_cache(&cache, NULL);
 494         if (error != 0) {
 495                 (void) fprintf(stderr, gettext("unable to access the blkid "
 496                     "cache.\n"));
 497                 return (-1);
 498         }
 499
 500         error = check_disk(path, cache, force, isspare, iswholedisk);
 501         blkid_put_cache(cache);
 502
 503         return (error);
 504 }
 505
 506 /*
 507  * This may be a shorthand device path or it could be total gibberish.
 508  * Check to see if it is a known device available in zfs_vdev_paths.
 509  * As part of this check, see if we've been given an entire disk
 510  * (minus the slice number).
 511  */
 512 static int
 513 is_shorthand_path(const char *arg, char *path, size_t path_size,
 514     struct stat64 *statbuf, boolean_t *wholedisk)
 515 {
 516         int error;
 517
 518         error = zfs_resolve_shortname(arg, path, path_size);
 519         if (error == 0) {
 520                 *wholedisk = zfs_dev_is_whole_disk(path);
 521                 if (*wholedisk || (stat64(path, statbuf) == 0))
 522                         return (0);
 523         }
 524
 525         strlcpy(path, arg, path_size);
 526         memset(statbuf, 0, sizeof (*statbuf));
 527         *wholedisk = B_FALSE;
 528
 529         return (error);
 530 }
 531
 532 /*
 533  * Determine if the given path is a hot spare within the given configuration.
 534  * If no configuration is given we rely solely on the label.
 535  */
 536 static boolean_t
 537 is_spare(nvlist_t *config, const char *path)
 538 {
 539         int fd;
 540         pool_state_t state;
 541         char *name = NULL;
 542         nvlist_t *label;
 543         uint64_t guid, spareguid;
 544         nvlist_t *nvroot;
 545         nvlist_t **spares;
 546         uint_t i, nspares;
 547         boolean_t inuse;
 548
 549         if ((fd = open(path, O_RDONLY)) < 0)
 550                 return (B_FALSE);
 551
 552         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
 553             !inuse ||
 554             state != POOL_STATE_SPARE ||
 555             zpool_read_label(fd, &label, NULL) != 0) {
 556                 free(name);
 557                 (void) close(fd);
 558                 return (B_FALSE);
 559         }
 560         free(name);
 561         (void) close(fd);
 562
 563         if (config == NULL) {
 564                 nvlist_free(label);
 565                 return (B_TRUE);
 566         }
 567
 568         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
 569         nvlist_free(label);
 570
 571         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 572             &nvroot) == 0);
 573         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 574             &spares, &nspares) == 0) {
 575                 for (i = 0; i < nspares; i++) {
 576                         verify(nvlist_lookup_uint64(spares[i],
 577                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
 578                         if (spareguid == guid)
 579                                 return (B_TRUE);
 580                 }
 581         }
 582
 583         return (B_FALSE);
 584 }
 585
 586 /*
 587  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 588  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 589  * leaf vdev are:
 590  *
 591  *      /dev/xxx        Complete disk path
 592  *      /xxx            Full path to file
 593  *      xxx             Shorthand for <zfs_vdev_paths>/xxx
 594  */
 595 static nvlist_t *
 596 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
 597 {
 598         char path[MAXPATHLEN];
 599         struct stat64 statbuf;
 600         nvlist_t *vdev = NULL;
 601         char *type = NULL;
 602         boolean_t wholedisk = B_FALSE;
 603         uint64_t ashift = 0;
 604         int err;
 605
 606         /*
 607          * Determine what type of vdev this is, and put the full path into
 608          * 'path'.  We detect whether this is a device of file afterwards by
 609          * checking the st_mode of the file.
 610          */
 611         if (arg[0] == '/') {
 612                 /*
 613                  * Complete device or file path.  Exact type is determined by
 614                  * examining the file descriptor afterwards.  Symbolic links
 615                  * are resolved to their real paths to determine whole disk
 616                  * and S_ISBLK/S_ISREG type checks.  However, we are careful
 617                  * to store the given path as ZPOOL_CONFIG_PATH to ensure we
 618                  * can leverage udev's persistent device labels.
 619                  */
 620                 if (realpath(arg, path) == NULL) {
 621                         (void) fprintf(stderr,
 622                             gettext("cannot resolve path '%s'\n"), arg);
 623                         return (NULL);
 624                 }
 625
 626                 wholedisk = zfs_dev_is_whole_disk(path);
 627                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 628                         (void) fprintf(stderr,
 629                             gettext("cannot open '%s': %s\n"),
 630                             path, strerror(errno));
 631                         return (NULL);
 632                 }
 633
 634                 /* After whole disk check restore original passed path */
 635                 strlcpy(path, arg, sizeof (path));
 636         } else {
 637                 err = is_shorthand_path(arg, path, sizeof (path),
 638                     &statbuf, &wholedisk);
 639                 if (err != 0) {
 640                         /*
 641                          * If we got ENOENT, then the user gave us
 642                          * gibberish, so try to direct them with a
 643                          * reasonable error message.  Otherwise,
 644                          * regurgitate strerror() since it's the best we
 645                          * can do.
 646                          */
 647                         if (err == ENOENT) {
 648                                 (void) fprintf(stderr,
 649                                     gettext("cannot open '%s': no such "
 650                                     "device in %s\n"), arg, DISK_ROOT);
 651                                 (void) fprintf(stderr,
 652                                     gettext("must be a full path or "
 653                                     "shorthand device name\n"));
 654                                 return (NULL);
 655                         } else {
 656                                 (void) fprintf(stderr,
 657                                     gettext("cannot open '%s': %s\n"),
 658                                     path, strerror(errno));
 659                                 return (NULL);
 660                         }
 661                 }
 662         }
 663
 664         /*
 665          * Determine whether this is a device or a file.
 666          */
 667         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 668                 type = VDEV_TYPE_DISK;
 669         } else if (S_ISREG(statbuf.st_mode)) {
 670                 type = VDEV_TYPE_FILE;
 671         } else {
 672                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 673                     "block device or regular file\n"), path);
 674                 return (NULL);
 675         }
 676
 677         /*
 678          * Finally, we have the complete device or file, and we know that it is
 679          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 680          * vdevs have a 'path' element, and devices also have a 'devid' element.
 681          */
 682         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 683         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 684         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 685         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 686         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 687                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 688                     (uint64_t)wholedisk) == 0);
 689
 690         /*
 691          * Override defaults if custom properties are provided.
 692          */
 693         if (props != NULL) {
 694                 char *value = NULL;
 695
 696                 if (nvlist_lookup_string(props,
 697                     zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0) {
 698                         if (zfs_nicestrtonum(NULL, value, &ashift) != 0) {
 699                                 (void) fprintf(stderr,
 700                                     gettext("ashift must be a number.\n"));
 701                                 return (NULL);
 702                         }
 703                         if (ashift != 0 &&
 704                             (ashift < ASHIFT_MIN || ashift > ASHIFT_MAX)) {
 705                                 (void) fprintf(stderr,
 706                                     gettext("invalid 'ashift=%" PRIu64 "' "
 707                                     "property: only values between %" PRId32 " "
 708                                     "and %" PRId32 " are allowed.\n"),
 709                                     ashift, ASHIFT_MIN, ASHIFT_MAX);
 710                                 return (NULL);
 711                         }
 712                 }
 713         }
 714
 715         /*
 716          * If the device is known to incorrectly report its physical sector
 717          * size explicitly provide the known correct value.
 718          */
 719         if (ashift == 0) {
 720                 int sector_size;
 721
 722                 if (check_sector_size_database(path, &sector_size) == B_TRUE)
 723                         ashift = highbit64(sector_size) - 1;
 724         }
 725
 726         if (ashift > 0)
 727                 (void) nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
 728
 729         return (vdev);
 730 }
 731
 732 /*
 733  * Go through and verify the replication level of the pool is consistent.
 734  * Performs the following checks:
 735  *
 736  *      For the new spec, verifies that devices in mirrors and raidz are the
 737  *      same size.
 738  *
 739  *      If the current configuration already has inconsistent replication
 740  *      levels, ignore any other potential problems in the new spec.
 741  *
 742  *      Otherwise, make sure that the current spec (if there is one) and the new
 743  *      spec have consistent replication levels.
 744  */
 745 typedef struct replication_level {
 746         char *zprl_type;
 747         uint64_t zprl_children;
 748         uint64_t zprl_parity;
 749 } replication_level_t;
 750
 751 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 752
 753 static boolean_t
 754 is_raidz_mirror(replication_level_t *a, replication_level_t *b,
 755     replication_level_t **raidz, replication_level_t **mirror)
 756 {
 757         if (strcmp(a->zprl_type, "raidz") == 0 &&
 758             strcmp(b->zprl_type, "mirror") == 0) {
 759                 *raidz = a;
 760                 *mirror = b;
 761                 return (B_TRUE);
 762         }
 763         return (B_FALSE);
 764 }
 765
 766 /*
 767  * Given a list of toplevel vdevs, return the current replication level.  If
 768  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 769  * an error message will be displayed for each self-inconsistent vdev.
 770  */
 771 static replication_level_t *
 772 get_replication(nvlist_t *nvroot, boolean_t fatal)
 773 {
 774         nvlist_t **top;
 775         uint_t t, toplevels;
 776         nvlist_t **child;
 777         uint_t c, children;
 778         nvlist_t *nv;
 779         char *type;
 780         replication_level_t lastrep = {0};
 781         replication_level_t rep;
 782         replication_level_t *ret;
 783         replication_level_t *raidz, *mirror;
 784         boolean_t dontreport;
 785
 786         ret = safe_malloc(sizeof (replication_level_t));
 787
 788         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 789             &top, &toplevels) == 0);
 790
 791         for (t = 0; t < toplevels; t++) {
 792                 uint64_t is_log = B_FALSE;
 793
 794                 nv = top[t];
 795
 796                 /*
 797                  * For separate logs we ignore the top level vdev replication
 798                  * constraints.
 799                  */
 800                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 801                 if (is_log)
 802                         continue;
 803
 804                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 805                     &type) == 0);
 806                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 807                     &child, &children) != 0) {
 808                         /*
 809                          * This is a 'file' or 'disk' vdev.
 810                          */
 811                         rep.zprl_type = type;
 812                         rep.zprl_children = 1;
 813                         rep.zprl_parity = 0;
 814                 } else {
 815                         uint64_t vdev_size;
 816
 817                         /*
 818                          * This is a mirror or RAID-Z vdev.  Go through and make
 819                          * sure the contents are all the same (files vs. disks),
 820                          * keeping track of the number of elements in the
 821                          * process.
 822                          *
 823                          * We also check that the size of each vdev (if it can
 824                          * be determined) is the same.
 825                          */
 826                         rep.zprl_type = type;
 827                         rep.zprl_children = 0;
 828
 829                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 830                                 verify(nvlist_lookup_uint64(nv,
 831                                     ZPOOL_CONFIG_NPARITY,
 832                                     &rep.zprl_parity) == 0);
 833                                 assert(rep.zprl_parity != 0);
 834                         } else {
 835                                 rep.zprl_parity = 0;
 836                         }
 837
 838                         /*
 839                          * The 'dontreport' variable indicates that we've
 840                          * already reported an error for this spec, so don't
 841                          * bother doing it again.
 842                          */
 843                         type = NULL;
 844                         dontreport = 0;
 845                         vdev_size = -1ULL;
 846                         for (c = 0; c < children; c++) {
 847                                 nvlist_t *cnv = child[c];
 848                                 char *path;
 849                                 struct stat64 statbuf;
 850                                 uint64_t size = -1ULL;
 851                                 char *childtype;
 852                                 int fd, err;
 853
 854                                 rep.zprl_children++;
 855
 856                                 verify(nvlist_lookup_string(cnv,
 857                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 858
 859                                 /*
 860                                  * If this is a replacing or spare vdev, then
 861                                  * get the real first child of the vdev.
 862                                  */
 863                                 if (strcmp(childtype,
 864                                     VDEV_TYPE_REPLACING) == 0 ||
 865                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 866                                         nvlist_t **rchild;
 867                                         uint_t rchildren;
 868
 869                                         verify(nvlist_lookup_nvlist_array(cnv,
 870                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 871                                             &rchildren) == 0);
 872                                         assert(rchildren == 2);
 873                                         cnv = rchild[0];
 874
 875                                         verify(nvlist_lookup_string(cnv,
 876                                             ZPOOL_CONFIG_TYPE,
 877                                             &childtype) == 0);
 878                                 }
 879
 880                                 verify(nvlist_lookup_string(cnv,
 881                                     ZPOOL_CONFIG_PATH, &path) == 0);
 882
 883                                 /*
 884                                  * If we have a raidz/mirror that combines disks
 885                                  * with files, report it as an error.
 886                                  */
 887                                 if (!dontreport && type != NULL &&
 888                                     strcmp(type, childtype) != 0) {
 889                                         if (ret != NULL)
 890                                                 free(ret);
 891                                         ret = NULL;
 892                                         if (fatal)
 893                                                 vdev_error(gettext(
 894                                                     "mismatched replication "
 895                                                     "level: %s contains both "
 896                                                     "files and devices\n"),
 897                                                     rep.zprl_type);
 898                                         else
 899                                                 return (NULL);
 900                                         dontreport = B_TRUE;
 901                                 }
 902
 903                                 /*
 904                                  * According to stat(2), the value of 'st_size'
 905                                  * is undefined for block devices and character
 906                                  * devices.  But there is no effective way to
 907                                  * determine the real size in userland.
 908                                  *
 909                                  * Instead, we'll take advantage of an
 910                                  * implementation detail of spec_size().  If the
 911                                  * device is currently open, then we (should)
 912                                  * return a valid size.
 913                                  *
 914                                  * If we still don't get a valid size (indicated
 915                                  * by a size of 0 or MAXOFFSET_T), then ignore
 916                                  * this device altogether.
 917                                  */
 918                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 919                                         err = fstat64_blk(fd, &statbuf);
 920                                         (void) close(fd);
 921                                 } else {
 922                                         err = stat64(path, &statbuf);
 923                                 }
 924
 925                                 if (err != 0 ||
 926                                     statbuf.st_size == 0 ||
 927                                     statbuf.st_size == MAXOFFSET_T)
 928                                         continue;
 929
 930                                 size = statbuf.st_size;
 931
 932                                 /*
 933                                  * Also make sure that devices and
 934                                  * slices have a consistent size.  If
 935                                  * they differ by a significant amount
 936                                  * (~16MB) then report an error.
 937                                  */
 938                                 if (!dontreport &&
 939                                     (vdev_size != -1ULL &&
 940                                     (labs(size - vdev_size) >
 941                                     ZPOOL_FUZZ))) {
 942                                         if (ret != NULL)
 943                                                 free(ret);
 944                                         ret = NULL;
 945                                         if (fatal)
 946                                                 vdev_error(gettext(
 947                                                     "%s contains devices of "
 948                                                     "different sizes\n"),
 949                                                     rep.zprl_type);
 950                                         else
 951                                                 return (NULL);
 952                                         dontreport = B_TRUE;
 953                                 }
 954
 955                                 type = childtype;
 956                                 vdev_size = size;
 957                         }
 958                 }
 959
 960                 /*
 961                  * At this point, we have the replication of the last toplevel
 962                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 963                  * different.
 964                  */
 965                 if (lastrep.zprl_type != NULL) {
 966                         if (is_raidz_mirror(&lastrep, &rep, &raidz, &mirror) ||
 967                             is_raidz_mirror(&rep, &lastrep, &raidz, &mirror)) {
 968                                 /*
 969                                  * Accepted raidz and mirror when they can
 970                                  * handle the same number of disk failures.
 971                                  */
 972                                 if (raidz->zprl_parity !=
 973                                     mirror->zprl_children - 1) {
 974                                         if (ret != NULL)
 975                                                 free(ret);
 976                                         ret = NULL;
 977                                         if (fatal)
 978                                                 vdev_error(gettext(
 979                                                     "mismatched replication "
 980                                                     "level: "
 981                                                     "%s and %s vdevs with "
 982                                                     "different redundancy, "
 983                                                     "%llu vs. %llu (%llu-way) "
 984                                                     "are present\n"),
 985                                                     raidz->zprl_type,
 986                                                     mirror->zprl_type,
 987                                                     raidz->zprl_parity,
 988                                                     mirror->zprl_children - 1,
 989                                                     mirror->zprl_children);
 990                                         else
 991                                                 return (NULL);
 992                                 }
 993                         } else if (strcmp(lastrep.zprl_type, rep.zprl_type) !=
 994                             0) {
 995                                 if (ret != NULL)
 996                                         free(ret);
 997                                 ret = NULL;
 998                                 if (fatal)
 999                                         vdev_error(gettext(
1000                                             "mismatched replication level: "
1001                                             "both %s and %s vdevs are "
1002                                             "present\n"),
1003                                             lastrep.zprl_type, rep.zprl_type);
1004                                 else
1005                                         return (NULL);
1006                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
1007                                 if (ret)
1008                                         free(ret);
1009                                 ret = NULL;
1010                                 if (fatal)
1011                                         vdev_error(gettext(
1012                                             "mismatched replication level: "
1013                                             "both %llu and %llu device parity "
1014                                             "%s vdevs are present\n"),
1015                                             lastrep.zprl_parity,
1016                                             rep.zprl_parity,
1017                                             rep.zprl_type);
1018                                 else
1019                                         return (NULL);
1020                         } else if (lastrep.zprl_children != rep.zprl_children) {
1021                                 if (ret)
1022                                         free(ret);
1023                                 ret = NULL;
1024                                 if (fatal)
1025                                         vdev_error(gettext(
1026                                             "mismatched replication level: "
1027                                             "both %llu-way and %llu-way %s "
1028                                             "vdevs are present\n"),
1029                                             lastrep.zprl_children,
1030                                             rep.zprl_children,
1031                                             rep.zprl_type);
1032                                 else
1033                                         return (NULL);
1034                         }
1035                 }
1036                 lastrep = rep;
1037         }
1038
1039         if (ret != NULL)
1040                 *ret = rep;
1041
1042         return (ret);
1043 }
1044
1045 /*
1046  * Check the replication level of the vdev spec against the current pool.  Calls
1047  * get_replication() to make sure the new spec is self-consistent.  If the pool
1048  * has a consistent replication level, then we ignore any errors.  Otherwise,
1049  * report any difference between the two.
1050  */
1051 static int
1052 check_replication(nvlist_t *config, nvlist_t *newroot)
1053 {
1054         nvlist_t **child;
1055         uint_t  children;
1056         replication_level_t *current = NULL, *new;
1057         replication_level_t *raidz, *mirror;
1058         int ret;
1059
1060         /*
1061          * If we have a current pool configuration, check to see if it's
1062          * self-consistent.  If not, simply return success.
1063          */
1064         if (config != NULL) {
1065                 nvlist_t *nvroot;
1066
1067                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1068                     &nvroot) == 0);
1069                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1070                         return (0);
1071         }
1072         /*
1073          * for spares there may be no children, and therefore no
1074          * replication level to check
1075          */
1076         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1077             &child, &children) != 0) || (children == 0)) {
1078                 free(current);
1079                 return (0);
1080         }
1081
1082         /*
1083          * If all we have is logs then there's no replication level to check.
1084          */
1085         if (num_logs(newroot) == children) {
1086                 free(current);
1087                 return (0);
1088         }
1089
1090         /*
1091          * Get the replication level of the new vdev spec, reporting any
1092          * inconsistencies found.
1093          */
1094         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1095                 free(current);
1096                 return (-1);
1097         }
1098
1099         /*
1100          * Check to see if the new vdev spec matches the replication level of
1101          * the current pool.
1102          */
1103         ret = 0;
1104         if (current != NULL) {
1105                 if (is_raidz_mirror(current, new, &raidz, &mirror) ||
1106                     is_raidz_mirror(new, current, &raidz, &mirror)) {
1107                         if (raidz->zprl_parity != mirror->zprl_children - 1) {
1108                                 vdev_error(gettext(
1109                                     "mismatched replication level: pool and "
1110                                     "new vdev with different redundancy, %s "
1111                                     "and %s vdevs, %llu vs. %llu (%llu-way)\n"),
1112                                     raidz->zprl_type,
1113                                     mirror->zprl_type,
1114                                     raidz->zprl_parity,
1115                                     mirror->zprl_children - 1,
1116                                     mirror->zprl_children);
1117                                 ret = -1;
1118                         }
1119                 } else if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1120                         vdev_error(gettext(
1121                             "mismatched replication level: pool uses %s "
1122                             "and new vdev is %s\n"),
1123                             current->zprl_type, new->zprl_type);
1124                         ret = -1;
1125                 } else if (current->zprl_parity != new->zprl_parity) {
1126                         vdev_error(gettext(
1127                             "mismatched replication level: pool uses %llu "
1128                             "device parity and new vdev uses %llu\n"),
1129                             current->zprl_parity, new->zprl_parity);
1130                         ret = -1;
1131                 } else if (current->zprl_children != new->zprl_children) {
1132                         vdev_error(gettext(
1133                             "mismatched replication level: pool uses %llu-way "
1134                             "%s and new vdev uses %llu-way %s\n"),
1135                             current->zprl_children, current->zprl_type,
1136                             new->zprl_children, new->zprl_type);
1137                         ret = -1;
1138                 }
1139         }
1140
1141         free(new);
1142         if (current != NULL)
1143                 free(current);
1144
1145         return (ret);
1146 }
1147
1148 static int
1149 zero_label(char *path)
1150 {
1151         const int size = 4096;
1152         char buf[size];
1153         int err, fd;
1154
1155         if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1156                 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1157                     path, strerror(errno));
1158                 return (-1);
1159         }
1160
1161         memset(buf, 0, size);
1162         err = write(fd, buf, size);
1163         (void) fdatasync(fd);
1164         (void) close(fd);
1165
1166         if (err == -1) {
1167                 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1168                     "of '%s': %s\n"), size, path, strerror(errno));
1169                 return (-1);
1170         }
1171
1172         if (err != size) {
1173                 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1174                     "of '%s'\n"), err, size, path);
1175                 return (-1);
1176         }
1177
1178         return (0);
1179 }
1180
1181 /*
1182  * Go through and find any whole disks in the vdev specification, labelling them
1183  * as appropriate.  When constructing the vdev spec, we were unable to open this
1184  * device in order to provide a devid.  Now that we have labelled the disk and
1185  * know that slice 0 is valid, we can construct the devid now.
1186  *
1187  * If the disk was already labeled with an EFI label, we will have gotten the
1188  * devid already (because we were able to open the whole disk).  Otherwise, we
1189  * need to get the devid after we label the disk.
1190  */
1191 static int
1192 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1193 {
1194         nvlist_t **child;
1195         uint_t c, children;
1196         char *type, *path;
1197         char devpath[MAXPATHLEN];
1198         char udevpath[MAXPATHLEN];
1199         uint64_t wholedisk;
1200         struct stat64 statbuf;
1201         int is_exclusive = 0;
1202         int fd;
1203         int ret;
1204
1205         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1206
1207         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1208             &child, &children) != 0) {
1209
1210                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1211                         return (0);
1212
1213                 /*
1214                  * We have a disk device.  If this is a whole disk write
1215                  * out the efi partition table, otherwise write zero's to
1216                  * the first 4k of the partition.  This is to ensure that
1217                  * libblkid will not misidentify the partition due to a
1218                  * magic value left by the previous filesystem.
1219                  */
1220                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1221                 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1222                     &wholedisk));
1223
1224                 if (!wholedisk) {
1225                         /*
1226                          * Update device id string for mpath nodes (Linux only)
1227                          */
1228                         if (is_mpath_whole_disk(path))
1229                                 update_vdev_config_dev_strs(nv);
1230
1231                         if (!is_spare(NULL, path))
1232                                 (void) zero_label(path);
1233                         return (0);
1234                 }
1235
1236                 if (realpath(path, devpath) == NULL) {
1237                         ret = errno;
1238                         (void) fprintf(stderr,
1239                             gettext("cannot resolve path '%s'\n"), path);
1240                         return (ret);
1241                 }
1242
1243                 /*
1244                  * Remove any previously existing symlink from a udev path to
1245                  * the device before labeling the disk.  This ensures that
1246                  * only newly created links are used.  Otherwise there is a
1247                  * window between when udev deletes and recreates the link
1248                  * during which access attempts will fail with ENOENT.
1249                  */
1250                 strlcpy(udevpath, path, MAXPATHLEN);
1251                 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1252
1253                 fd = open(devpath, O_RDWR|O_EXCL);
1254                 if (fd == -1) {
1255                         if (errno == EBUSY)
1256                                 is_exclusive = 1;
1257                 } else {
1258                         (void) close(fd);
1259                 }
1260
1261                 /*
1262                  * If the partition exists, contains a valid spare label,
1263                  * and is opened exclusively there is no need to partition
1264                  * it.  Hot spares have already been partitioned and are
1265                  * held open exclusively by the kernel as a safety measure.
1266                  *
1267                  * If the provided path is for a /dev/disk/ device its
1268                  * symbolic link will be removed, partition table created,
1269                  * and then block until udev creates the new link.
1270                  */
1271                 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1272                         char *devnode = strrchr(devpath, '/') + 1;
1273
1274                         ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1275                         if (ret == 0) {
1276                                 ret = lstat64(udevpath, &statbuf);
1277                                 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1278                                         (void) unlink(udevpath);
1279                         }
1280
1281                         /*
1282                          * When labeling a pool the raw device node name
1283                          * is provided as it appears under /dev/.
1284                          */
1285                         if (zpool_label_disk(g_zfs, zhp, devnode) == -1)
1286                                 return (-1);
1287
1288                         /*
1289                          * Wait for udev to signal the device is available
1290                          * by the provided path.
1291                          */
1292                         ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1293                         if (ret) {
1294                                 (void) fprintf(stderr,
1295                                     gettext("missing link: %s was "
1296                                     "partitioned but %s is missing\n"),
1297                                     devnode, udevpath);
1298                                 return (ret);
1299                         }
1300
1301                         ret = zero_label(udevpath);
1302                         if (ret)
1303                                 return (ret);
1304                 }
1305
1306                 /*
1307                  * Update the path to refer to the partition.  The presence of
1308                  * the 'whole_disk' field indicates to the CLI that we should
1309                  * chop off the partition number when displaying the device in
1310                  * future output.
1311                  */
1312                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1313
1314                 /*
1315                  * Update device id strings for whole disks (Linux only)
1316                  */
1317                 update_vdev_config_dev_strs(nv);
1318
1319                 return (0);
1320         }
1321
1322         for (c = 0; c < children; c++)
1323                 if ((ret = make_disks(zhp, child[c])) != 0)
1324                         return (ret);
1325
1326         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1327             &child, &children) == 0)
1328                 for (c = 0; c < children; c++)
1329                         if ((ret = make_disks(zhp, child[c])) != 0)
1330                                 return (ret);
1331
1332         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1333             &child, &children) == 0)
1334                 for (c = 0; c < children; c++)
1335                         if ((ret = make_disks(zhp, child[c])) != 0)
1336                                 return (ret);
1337
1338         return (0);
1339 }
1340
1341 /*
1342  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1343  * the majority of this task.
1344  */
1345 static boolean_t
1346 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1347     boolean_t replacing, boolean_t isspare)
1348 {
1349         nvlist_t **child;
1350         uint_t c, children;
1351         char *type, *path;
1352         int ret = 0;
1353         char buf[MAXPATHLEN];
1354         uint64_t wholedisk = B_FALSE;
1355         boolean_t anyinuse = B_FALSE;
1356
1357         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1358
1359         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1360             &child, &children) != 0) {
1361
1362                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1363                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1364                         verify(!nvlist_lookup_uint64(nv,
1365                             ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1366
1367                 /*
1368                  * As a generic check, we look to see if this is a replace of a
1369                  * hot spare within the same pool.  If so, we allow it
1370                  * regardless of what libblkid or zpool_in_use() says.
1371                  */
1372                 if (replacing) {
1373                         (void) strlcpy(buf, path, sizeof (buf));
1374                         if (wholedisk) {
1375                                 ret = zfs_append_partition(buf,  sizeof (buf));
1376                                 if (ret == -1)
1377                                         return (-1);
1378                         }
1379
1380                         if (is_spare(config, buf))
1381                                 return (B_FALSE);
1382                 }
1383
1384                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1385                         ret = check_device(path, force, isspare, wholedisk);
1386
1387                 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1388                         ret = check_file(path, force, isspare);
1389
1390                 return (ret != 0);
1391         }
1392
1393         for (c = 0; c < children; c++)
1394                 if (is_device_in_use(config, child[c], force, replacing,
1395                     B_FALSE))
1396                         anyinuse = B_TRUE;
1397
1398         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1399             &child, &children) == 0)
1400                 for (c = 0; c < children; c++)
1401                         if (is_device_in_use(config, child[c], force, replacing,
1402                             B_TRUE))
1403                                 anyinuse = B_TRUE;
1404
1405         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1406             &child, &children) == 0)
1407                 for (c = 0; c < children; c++)
1408                         if (is_device_in_use(config, child[c], force, replacing,
1409                             B_FALSE))
1410                                 anyinuse = B_TRUE;
1411
1412         return (anyinuse);
1413 }
1414
1415 static const char *
1416 is_grouping(const char *type, int *mindev, int *maxdev)
1417 {
1418         if (strncmp(type, "raidz", 5) == 0) {
1419                 const char *p = type + 5;
1420                 char *end;
1421                 long nparity;
1422
1423                 if (*p == '\0') {
1424                         nparity = 1;
1425                 } else if (*p == '0') {
1426                         return (NULL); /* no zero prefixes allowed */
1427                 } else {
1428                         errno = 0;
1429                         nparity = strtol(p, &end, 10);
1430                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1431                             *end != '\0')
1432                                 return (NULL);
1433                 }
1434
1435                 if (mindev != NULL)
1436                         *mindev = nparity + 1;
1437                 if (maxdev != NULL)
1438                         *maxdev = 255;
1439                 return (VDEV_TYPE_RAIDZ);
1440         }
1441
1442         if (maxdev != NULL)
1443                 *maxdev = INT_MAX;
1444
1445         if (strcmp(type, "mirror") == 0) {
1446                 if (mindev != NULL)
1447                         *mindev = 2;
1448                 return (VDEV_TYPE_MIRROR);
1449         }
1450
1451         if (strcmp(type, "spare") == 0) {
1452                 if (mindev != NULL)
1453                         *mindev = 1;
1454                 return (VDEV_TYPE_SPARE);
1455         }
1456
1457         if (strcmp(type, "log") == 0) {
1458                 if (mindev != NULL)
1459                         *mindev = 1;
1460                 return (VDEV_TYPE_LOG);
1461         }
1462
1463         if (strcmp(type, "cache") == 0) {
1464                 if (mindev != NULL)
1465                         *mindev = 1;
1466                 return (VDEV_TYPE_L2CACHE);
1467         }
1468
1469         return (NULL);
1470 }
1471
1472 /*
1473  * Construct a syntactically valid vdev specification,
1474  * and ensure that all devices and files exist and can be opened.
1475  * Note: we don't bother freeing anything in the error paths
1476  * because the program is just going to exit anyway.
1477  */
1478 nvlist_t *
1479 construct_spec(nvlist_t *props, int argc, char **argv)
1480 {
1481         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1482         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1483         const char *type;
1484         uint64_t is_log;
1485         boolean_t seen_logs;
1486
1487         top = NULL;
1488         toplevels = 0;
1489         spares = NULL;
1490         l2cache = NULL;
1491         nspares = 0;
1492         nlogs = 0;
1493         nl2cache = 0;
1494         is_log = B_FALSE;
1495         seen_logs = B_FALSE;
1496         nvroot = NULL;
1497
1498         while (argc > 0) {
1499                 nv = NULL;
1500
1501                 /*
1502                  * If it's a mirror or raidz, the subsequent arguments are
1503                  * its leaves -- until we encounter the next mirror or raidz.
1504                  */
1505                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1506                         nvlist_t **child = NULL;
1507                         int c, children = 0;
1508
1509                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1510                                 if (spares != NULL) {
1511                                         (void) fprintf(stderr,
1512                                             gettext("invalid vdev "
1513                                             "specification: 'spare' can be "
1514                                             "specified only once\n"));
1515                                         goto spec_out;
1516                                 }
1517                                 is_log = B_FALSE;
1518                         }
1519
1520                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1521                                 if (seen_logs) {
1522                                         (void) fprintf(stderr,
1523                                             gettext("invalid vdev "
1524                                             "specification: 'log' can be "
1525                                             "specified only once\n"));
1526                                         goto spec_out;
1527                                 }
1528                                 seen_logs = B_TRUE;
1529                                 is_log = B_TRUE;
1530                                 argc--;
1531                                 argv++;
1532                                 /*
1533                                  * A log is not a real grouping device.
1534                                  * We just set is_log and continue.
1535                                  */
1536                                 continue;
1537                         }
1538
1539                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1540                                 if (l2cache != NULL) {
1541                                         (void) fprintf(stderr,
1542                                             gettext("invalid vdev "
1543                                             "specification: 'cache' can be "
1544                                             "specified only once\n"));
1545                                         goto spec_out;
1546                                 }
1547                                 is_log = B_FALSE;
1548                         }
1549
1550                         if (is_log) {
1551                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1552                                         (void) fprintf(stderr,
1553                                             gettext("invalid vdev "
1554                                             "specification: unsupported 'log' "
1555                                             "device: %s\n"), type);
1556                                         goto spec_out;
1557                                 }
1558                                 nlogs++;
1559                         }
1560
1561                         for (c = 1; c < argc; c++) {
1562                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1563                                         break;
1564                                 children++;
1565                                 child = realloc(child,
1566                                     children * sizeof (nvlist_t *));
1567                                 if (child == NULL)
1568                                         zpool_no_memory();
1569                                 if ((nv = make_leaf_vdev(props, argv[c],
1570                                     B_FALSE)) == NULL) {
1571                                         for (c = 0; c < children - 1; c++)
1572                                                 nvlist_free(child[c]);
1573                                         free(child);
1574                                         goto spec_out;
1575                                 }
1576
1577                                 child[children - 1] = nv;
1578                         }
1579
1580                         if (children < mindev) {
1581                                 (void) fprintf(stderr, gettext("invalid vdev "
1582                                     "specification: %s requires at least %d "
1583                                     "devices\n"), argv[0], mindev);
1584                                 for (c = 0; c < children; c++)
1585                                         nvlist_free(child[c]);
1586                                 free(child);
1587                                 goto spec_out;
1588                         }
1589
1590                         if (children > maxdev) {
1591                                 (void) fprintf(stderr, gettext("invalid vdev "
1592                                     "specification: %s supports no more than "
1593                                     "%d devices\n"), argv[0], maxdev);
1594                                 for (c = 0; c < children; c++)
1595                                         nvlist_free(child[c]);
1596                                 free(child);
1597                                 goto spec_out;
1598                         }
1599
1600                         argc -= c;
1601                         argv += c;
1602
1603                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1604                                 spares = child;
1605                                 nspares = children;
1606                                 continue;
1607                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1608                                 l2cache = child;
1609                                 nl2cache = children;
1610                                 continue;
1611                         } else {
1612                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1613                                     0) == 0);
1614                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1615                                     type) == 0);
1616                                 verify(nvlist_add_uint64(nv,
1617                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1618                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1619                                         verify(nvlist_add_uint64(nv,
1620                                             ZPOOL_CONFIG_NPARITY,
1621                                             mindev - 1) == 0);
1622                                 }
1623                                 verify(nvlist_add_nvlist_array(nv,
1624                                     ZPOOL_CONFIG_CHILDREN, child,
1625                                     children) == 0);
1626
1627                                 for (c = 0; c < children; c++)
1628                                         nvlist_free(child[c]);
1629                                 free(child);
1630                         }
1631                 } else {
1632                         /*
1633                          * We have a device.  Pass off to make_leaf_vdev() to
1634                          * construct the appropriate nvlist describing the vdev.
1635                          */
1636                         if ((nv = make_leaf_vdev(props, argv[0],
1637                             is_log)) == NULL)
1638                                 goto spec_out;
1639
1640                         if (is_log)
1641                                 nlogs++;
1642                         argc--;
1643                         argv++;
1644                 }
1645
1646                 toplevels++;
1647                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1648                 if (top == NULL)
1649                         zpool_no_memory();
1650                 top[toplevels - 1] = nv;
1651         }
1652
1653         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1654                 (void) fprintf(stderr, gettext("invalid vdev "
1655                     "specification: at least one toplevel vdev must be "
1656                     "specified\n"));
1657                 goto spec_out;
1658         }
1659
1660         if (seen_logs && nlogs == 0) {
1661                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1662                     "log requires at least 1 device\n"));
1663                 goto spec_out;
1664         }
1665
1666         /*
1667          * Finally, create nvroot and add all top-level vdevs to it.
1668          */
1669         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1670         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1671             VDEV_TYPE_ROOT) == 0);
1672         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1673             top, toplevels) == 0);
1674         if (nspares != 0)
1675                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1676                     spares, nspares) == 0);
1677         if (nl2cache != 0)
1678                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1679                     l2cache, nl2cache) == 0);
1680
1681 spec_out:
1682         for (t = 0; t < toplevels; t++)
1683                 nvlist_free(top[t]);
1684         for (t = 0; t < nspares; t++)
1685                 nvlist_free(spares[t]);
1686         for (t = 0; t < nl2cache; t++)
1687                 nvlist_free(l2cache[t]);
1688
1689         free(spares);
1690         free(l2cache);
1691         free(top);
1692
1693         return (nvroot);
1694 }
1695
1696 nvlist_t *
1697 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1698     splitflags_t flags, int argc, char **argv)
1699 {
1700         nvlist_t *newroot = NULL, **child;
1701         uint_t c, children;
1702
1703         if (argc > 0) {
1704                 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1705                         (void) fprintf(stderr, gettext("Unable to build a "
1706                             "pool from the specified devices\n"));
1707                         return (NULL);
1708                 }
1709
1710                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1711                         nvlist_free(newroot);
1712                         return (NULL);
1713                 }
1714
1715                 /* avoid any tricks in the spec */
1716                 verify(nvlist_lookup_nvlist_array(newroot,
1717                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1718                 for (c = 0; c < children; c++) {
1719                         char *path;
1720                         const char *type;
1721                         int min, max;
1722
1723                         verify(nvlist_lookup_string(child[c],
1724                             ZPOOL_CONFIG_PATH, &path) == 0);
1725                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1726                                 (void) fprintf(stderr, gettext("Cannot use "
1727                                     "'%s' as a device for splitting\n"), type);
1728                                 nvlist_free(newroot);
1729                                 return (NULL);
1730                         }
1731                 }
1732         }
1733
1734         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1735                 nvlist_free(newroot);
1736                 return (NULL);
1737         }
1738
1739         return (newroot);
1740 }
1741
1742 /*
1743  * Get and validate the contents of the given vdev specification.  This ensures
1744  * that the nvlist returned is well-formed, that all the devices exist, and that
1745  * they are not currently in use by any other known consumer.  The 'poolconfig'
1746  * parameter is the current configuration of the pool when adding devices
1747  * existing pool, and is used to perform additional checks, such as changing the
1748  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1749  * new pool.  The 'force' flag controls whether devices should be forcefully
1750  * added, even if they appear in use.
1751  */
1752 nvlist_t *
1753 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1754     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1755 {
1756         nvlist_t *newroot;
1757         nvlist_t *poolconfig = NULL;
1758         is_force = force;
1759
1760         /*
1761          * Construct the vdev specification.  If this is successful, we know
1762          * that we have a valid specification, and that all devices can be
1763          * opened.
1764          */
1765         if ((newroot = construct_spec(props, argc, argv)) == NULL)
1766                 return (NULL);
1767
1768         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1769                 nvlist_free(newroot);
1770                 return (NULL);
1771         }
1772
1773         /*
1774          * Validate each device to make sure that its not shared with another
1775          * subsystem.  We do this even if 'force' is set, because there are some
1776          * uses (such as a dedicated dump device) that even '-f' cannot
1777          * override.
1778          */
1779         if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1780                 nvlist_free(newroot);
1781                 return (NULL);
1782         }
1783
1784         /*
1785          * Check the replication level of the given vdevs and report any errors
1786          * found.  We include the existing pool spec, if any, as we need to
1787          * catch changes against the existing replication level.
1788          */
1789         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1790                 nvlist_free(newroot);
1791                 return (NULL);
1792         }
1793
1794         /*
1795          * Run through the vdev specification and label any whole disks found.
1796          */
1797         if (!dryrun && make_disks(zhp, newroot) != 0) {
1798                 nvlist_free(newroot);
1799                 return (NULL);
1800         }
1801
1802         return (newroot);
1803 }