granicus.if.org Git - zfs/blob - cmd/zpool/zpool_vdev.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2013 by Delphix. All rights reserved.
  25  */
  26
  27 /*
  28  * Functions to convert between a list of vdevs and an nvlist representing the
  29  * configuration.  Each entry in the list can be one of:
  30  *
  31  *      Device vdevs
  32  *              disk=(path=..., devid=...)
  33  *              file=(path=...)
  34  *
  35  *      Group vdevs
  36  *              raidz[1|2]=(...)
  37  *              mirror=(...)
  38  *
  39  *      Hot spares
  40  *
  41  * While the underlying implementation supports it, group vdevs cannot contain
  42  * other group vdevs.  All userland verification of devices is contained within
  43  * this file.  If successful, the nvlist returned can be passed directly to the
  44  * kernel; we've done as much verification as possible in userland.
  45  *
  46  * Hot spares are a special case, and passed down as an array of disk vdevs, at
  47  * the same level as the root of the vdev tree.
  48  *
  49  * The only function exported by this file is 'make_root_vdev'.  The
  50  * function performs several passes:
  51  *
  52  *      1. Construct the vdev specification.  Performs syntax validation and
  53  *         makes sure each device is valid.
  54  *      2. Check for devices in use.  Using libblkid to make sure that no
  55  *         devices are also in use.  Some can be overridden using the 'force'
  56  *         flag, others cannot.
  57  *      3. Check for replication errors if the 'force' flag is not specified.
  58  *         validates that the replication level is consistent across the
  59  *         entire pool.
  60  *      4. Call libzfs to label any whole disks with an EFI label.
  61  */
  62
  63 #include <assert.h>
  64 #include <ctype.h>
  65 #include <devid.h>
  66 #include <errno.h>
  67 #include <fcntl.h>
  68 #include <libintl.h>
  69 #include <libnvpair.h>
  70 #include <limits.h>
  71 #include <scsi/scsi.h>
  72 #include <scsi/sg.h>
  73 #include <stdio.h>
  74 #include <string.h>
  75 #include <unistd.h>
  76 #include <sys/efi_partition.h>
  77 #include <sys/stat.h>
  78 #include <sys/vtoc.h>
  79 #include <sys/mntent.h>
  80 #include <uuid/uuid.h>
  81 #include <blkid/blkid.h>
  82 #include "zpool_util.h"
  83 #include <sys/zfs_context.h>
  84
  85 /*
  86  * For any given vdev specification, we can have multiple errors.  The
  87  * vdev_error() function keeps track of whether we have seen an error yet, and
  88  * prints out a header if its the first error we've seen.
  89  */
  90 boolean_t error_seen;
  91 boolean_t is_force;
  92
  93 typedef struct vdev_disk_db_entry
  94 {
  95         char id[24];
  96         int sector_size;
  97 } vdev_disk_db_entry_t;
  98
  99 /*
 100  * Database of block devices that lie about physical sector sizes.  The
 101  * identification string must be precisely 24 characters to avoid false
 102  * negatives
 103  */
 104 static vdev_disk_db_entry_t vdev_disk_database[] = {
 105         {"ATA     ADATA SSD S396 3", 8192},
 106         {"ATA     APPLE SSD SM128E", 8192},
 107         {"ATA     APPLE SSD SM256E", 8192},
 108         {"ATA     APPLE SSD SM512E", 8192},
 109         {"ATA     APPLE SSD SM768E", 8192},
 110         {"ATA     C400-MTFDDAC064M", 8192},
 111         {"ATA     C400-MTFDDAC128M", 8192},
 112         {"ATA     C400-MTFDDAC256M", 8192},
 113         {"ATA     C400-MTFDDAC512M", 8192},
 114         {"ATA     Corsair Force 3 ", 8192},
 115         {"ATA     Corsair Force GS", 8192},
 116         {"ATA     INTEL SSDSA2CT04", 8192},
 117         {"ATA     INTEL SSDSA2BZ10", 8192},
 118         {"ATA     INTEL SSDSA2BZ20", 8192},
 119         {"ATA     INTEL SSDSA2BZ30", 8192},
 120         {"ATA     INTEL SSDSA2CW04", 8192},
 121         {"ATA     INTEL SSDSA2CW08", 8192},
 122         {"ATA     INTEL SSDSA2CW12", 8192},
 123         {"ATA     INTEL SSDSA2CW16", 8192},
 124         {"ATA     INTEL SSDSA2CW30", 8192},
 125         {"ATA     INTEL SSDSA2CW60", 8192},
 126         {"ATA     INTEL SSDSC2CT06", 8192},
 127         {"ATA     INTEL SSDSC2CT12", 8192},
 128         {"ATA     INTEL SSDSC2CT18", 8192},
 129         {"ATA     INTEL SSDSC2CT24", 8192},
 130         {"ATA     INTEL SSDSC2CW06", 8192},
 131         {"ATA     INTEL SSDSC2CW12", 8192},
 132         {"ATA     INTEL SSDSC2CW18", 8192},
 133         {"ATA     INTEL SSDSC2CW24", 8192},
 134         {"ATA     INTEL SSDSC2CW48", 8192},
 135         {"ATA     KINGSTON SH100S3", 8192},
 136         {"ATA     KINGSTON SH103S3", 8192},
 137         {"ATA     M4-CT064M4SSD2  ", 8192},
 138         {"ATA     M4-CT128M4SSD2  ", 8192},
 139         {"ATA     M4-CT256M4SSD2  ", 8192},
 140         {"ATA     M4-CT512M4SSD2  ", 8192},
 141         {"ATA     OCZ-AGILITY2    ", 8192},
 142         {"ATA     OCZ-AGILITY3    ", 8192},
 143         {"ATA     OCZ-VERTEX2 3.5 ", 8192},
 144         {"ATA     OCZ-VERTEX3     ", 8192},
 145         {"ATA     OCZ-VERTEX3 LT  ", 8192},
 146         {"ATA     OCZ-VERTEX3 MI  ", 8192},
 147         {"ATA     OCZ-VERTEX4     ", 8192},
 148         {"ATA     SAMSUNG MZ7WD120", 8192},
 149         {"ATA     SAMSUNG MZ7WD240", 8192},
 150         {"ATA     SAMSUNG MZ7WD480", 8192},
 151         {"ATA     SAMSUNG MZ7WD960", 8192},
 152         {"ATA     SAMSUNG SSD 830 ", 8192},
 153         {"ATA     Samsung SSD 840 ", 8192},
 154         {"ATA     SanDisk SSD U100", 8192},
 155         {"ATA     TOSHIBA THNSNH06", 8192},
 156         {"ATA     TOSHIBA THNSNH12", 8192},
 157         {"ATA     TOSHIBA THNSNH25", 8192},
 158         {"ATA     TOSHIBA THNSNH51", 8192},
 159         {"ATA     APPLE SSD TS064C", 4096},
 160         {"ATA     APPLE SSD TS128C", 4096},
 161         {"ATA     APPLE SSD TS256C", 4096},
 162         {"ATA     APPLE SSD TS512C", 4096},
 163         {"ATA     INTEL SSDSA2M040", 4096},
 164         {"ATA     INTEL SSDSA2M080", 4096},
 165         {"ATA     INTEL SSDSA2M160", 4096},
 166         {"ATA     INTEL SSDSC2MH12", 4096},
 167         {"ATA     INTEL SSDSC2MH25", 4096},
 168         {"ATA     OCZ CORE_SSD    ", 4096},
 169         {"ATA     OCZ-VERTEX      ", 4096},
 170         {"ATA     SAMSUNG MCCOE32G", 4096},
 171         {"ATA     SAMSUNG MCCOE64G", 4096},
 172         {"ATA     SAMSUNG SSD PM80", 4096},
 173         /* Flash drives optimized for 4KB IOs on larger pages */
 174         {"ATA     INTEL SSDSC2BA10", 4096},
 175         {"ATA     INTEL SSDSC2BA20", 4096},
 176         {"ATA     INTEL SSDSC2BA40", 4096},
 177         {"ATA     INTEL SSDSC2BA80", 4096},
 178         {"ATA     INTEL SSDSC2BB08", 4096},
 179         {"ATA     INTEL SSDSC2BB12", 4096},
 180         {"ATA     INTEL SSDSC2BB16", 4096},
 181         {"ATA     INTEL SSDSC2BB24", 4096},
 182         {"ATA     INTEL SSDSC2BB30", 4096},
 183         {"ATA     INTEL SSDSC2BB40", 4096},
 184         {"ATA     INTEL SSDSC2BB48", 4096},
 185         {"ATA     INTEL SSDSC2BB60", 4096},
 186         {"ATA     INTEL SSDSC2BB80", 4096},
 187         {"ATA     INTEL SSDSC2BW24", 4096},
 188         {"ATA     INTEL SSDSC2BP24", 4096},
 189         {"ATA     INTEL SSDSC2BP48", 4096},
 190         {"NA      SmrtStorSDLKAE9W", 4096},
 191         /* Imported from Open Solaris */
 192         {"ATA     MARVELL SD88SA02", 4096},
 193         /* Advanced format Hard drives */
 194         {"ATA     Hitachi HDS5C303", 4096},
 195         {"ATA     SAMSUNG HD204UI ", 4096},
 196         {"ATA     ST2000DL004 HD20", 4096},
 197         {"ATA     WDC WD10EARS-00M", 4096},
 198         {"ATA     WDC WD10EARS-00S", 4096},
 199         {"ATA     WDC WD10EARS-00Z", 4096},
 200         {"ATA     WDC WD15EARS-00M", 4096},
 201         {"ATA     WDC WD15EARS-00S", 4096},
 202         {"ATA     WDC WD15EARS-00Z", 4096},
 203         {"ATA     WDC WD20EARS-00M", 4096},
 204         {"ATA     WDC WD20EARS-00S", 4096},
 205         {"ATA     WDC WD20EARS-00Z", 4096},
 206         {"ATA     WDC WD1600BEVT-0", 4096},
 207         {"ATA     WDC WD2500BEVT-0", 4096},
 208         {"ATA     WDC WD3200BEVT-0", 4096},
 209         {"ATA     WDC WD5000BEVT-0", 4096},
 210         /* Virtual disks: Assume zvols with default volblocksize */
 211 #if 0
 212         {"ATA     QEMU HARDDISK   ", 8192},
 213         {"IET     VIRTUAL-DISK    ", 8192},
 214         {"OI      COMSTAR         ", 8192},
 215         {"SUN     COMSTAR         ", 8192},
 216         {"NETAPP  LUN             ", 8192},
 217 #endif
 218 };
 219
 220 static const int vdev_disk_database_size =
 221         sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
 222
 223 #define INQ_REPLY_LEN   96
 224 #define INQ_CMD_LEN     6
 225
 226 static boolean_t
 227 check_sector_size_database(char *path, int *sector_size)
 228 {
 229         unsigned char inq_buff[INQ_REPLY_LEN];
 230         unsigned char sense_buffer[32];
 231         unsigned char inq_cmd_blk[INQ_CMD_LEN] =
 232             {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
 233         sg_io_hdr_t io_hdr;
 234         int error;
 235         int fd;
 236         int i;
 237
 238         /* Prepare INQUIRY command */
 239         memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
 240         io_hdr.interface_id = 'S';
 241         io_hdr.cmd_len = sizeof (inq_cmd_blk);
 242         io_hdr.mx_sb_len = sizeof (sense_buffer);
 243         io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 244         io_hdr.dxfer_len = INQ_REPLY_LEN;
 245         io_hdr.dxferp = inq_buff;
 246         io_hdr.cmdp = inq_cmd_blk;
 247         io_hdr.sbp = sense_buffer;
 248         io_hdr.timeout = 10;            /* 10 milliseconds is ample time */
 249
 250         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 251                 return (B_FALSE);
 252
 253         error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
 254
 255         (void) close(fd);
 256
 257         if (error < 0)
 258                 return (B_FALSE);
 259
 260         if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
 261                 return (B_FALSE);
 262
 263         for (i = 0; i < vdev_disk_database_size; i++) {
 264                 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
 265                         continue;
 266
 267                 *sector_size = vdev_disk_database[i].sector_size;
 268                 return (B_TRUE);
 269         }
 270
 271         return (B_FALSE);
 272 }
 273
 274 /*PRINTFLIKE1*/
 275 static void
 276 vdev_error(const char *fmt, ...)
 277 {
 278         va_list ap;
 279
 280         if (!error_seen) {
 281                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
 282                 if (!is_force)
 283                         (void) fprintf(stderr, gettext("use '-f' to override "
 284                             "the following errors:\n"));
 285                 else
 286                         (void) fprintf(stderr, gettext("the following errors "
 287                             "must be manually repaired:\n"));
 288                 error_seen = B_TRUE;
 289         }
 290
 291         va_start(ap, fmt);
 292         (void) vfprintf(stderr, fmt, ap);
 293         va_end(ap);
 294 }
 295
 296 /*
 297  * Check that a file is valid.  All we can do in this case is check that it's
 298  * not in use by another pool, and not in use by swap.
 299  */
 300 static int
 301 check_file(const char *file, boolean_t force, boolean_t isspare)
 302 {
 303         char  *name;
 304         int fd;
 305         int ret = 0;
 306         pool_state_t state;
 307         boolean_t inuse;
 308
 309         if ((fd = open(file, O_RDONLY)) < 0)
 310                 return (0);
 311
 312         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
 313                 const char *desc;
 314
 315                 switch (state) {
 316                 case POOL_STATE_ACTIVE:
 317                         desc = gettext("active");
 318                         break;
 319
 320                 case POOL_STATE_EXPORTED:
 321                         desc = gettext("exported");
 322                         break;
 323
 324                 case POOL_STATE_POTENTIALLY_ACTIVE:
 325                         desc = gettext("potentially active");
 326                         break;
 327
 328                 default:
 329                         desc = gettext("unknown");
 330                         break;
 331                 }
 332
 333                 /*
 334                  * Allow hot spares to be shared between pools.
 335                  */
 336                 if (state == POOL_STATE_SPARE && isspare)
 337                         return (0);
 338
 339                 if (state == POOL_STATE_ACTIVE ||
 340                     state == POOL_STATE_SPARE || !force) {
 341                         switch (state) {
 342                         case POOL_STATE_SPARE:
 343                                 vdev_error(gettext("%s is reserved as a hot "
 344                                     "spare for pool %s\n"), file, name);
 345                                 break;
 346                         default:
 347                                 vdev_error(gettext("%s is part of %s pool "
 348                                     "'%s'\n"), file, desc, name);
 349                                 break;
 350                         }
 351                         ret = -1;
 352                 }
 353
 354                 free(name);
 355         }
 356
 357         (void) close(fd);
 358         return (ret);
 359 }
 360
 361 static void
 362 check_error(int err)
 363 {
 364         (void) fprintf(stderr, gettext("warning: device in use checking "
 365             "failed: %s\n"), strerror(err));
 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 a whole disk.  Iterate over all slices on the disk and make sure
 403  * that none is in use by calling check_slice().
 404  */
 405 static int
 406 check_disk(const char *path, blkid_cache cache, int force,
 407     boolean_t isspare, boolean_t iswholedisk)
 408 {
 409         struct dk_gpt *vtoc;
 410         char slice_path[MAXPATHLEN];
 411         int err = 0;
 412         int fd, i;
 413
 414         /* This is not a wholedisk we only check the given partition */
 415         if (!iswholedisk)
 416                 return (check_slice(path, cache, force, isspare));
 417
 418         /*
 419          * When the device is a whole disk try to read the efi partition
 420          * label.  If this is successful we safely check the all of the
 421          * partitions.  However, when it fails it may simply be because
 422          * the disk is partitioned via the MBR.  Since we currently can
 423          * not easily decode the MBR return a failure and prompt to the
 424          * user to use force option since we cannot check the partitions.
 425          */
 426         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
 427                 check_error(errno);
 428                 return (-1);
 429         }
 430
 431         if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
 432                 (void) close(fd);
 433
 434                 if (force) {
 435                         return (0);
 436                 } else {
 437                         vdev_error(gettext("%s does not contain an EFI "
 438                             "label but it may contain partition\n"
 439                             "information in the MBR.\n"), path);
 440                         return (-1);
 441                 }
 442         }
 443
 444         /*
 445          * The primary efi partition label is damaged however the secondary
 446          * label at the end of the device is intact.  Rather than use this
 447          * label we should play it safe and treat this as a non efi device.
 448          */
 449         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
 450                 efi_free(vtoc);
 451                 (void) close(fd);
 452
 453                 if (force) {
 454                         /* Partitions will no be created using the backup */
 455                         return (0);
 456                 } else {
 457                         vdev_error(gettext("%s contains a corrupt primary "
 458                             "EFI label.\n"), path);
 459                         return (-1);
 460                 }
 461         }
 462
 463         for (i = 0; i < vtoc->efi_nparts; i++) {
 464
 465                 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
 466                     uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
 467                         continue;
 468
 469                 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
 470                         (void) snprintf(slice_path, sizeof (slice_path),
 471                             "%s%s%d", path, "-part", i+1);
 472                 else
 473                         (void) snprintf(slice_path, sizeof (slice_path),
 474                             "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
 475                             "p" : "", i+1);
 476
 477                 err = check_slice(slice_path, cache, force, isspare);
 478                 if (err)
 479                         break;
 480         }
 481
 482         efi_free(vtoc);
 483         (void) close(fd);
 484
 485         return (err);
 486 }
 487
 488 static int
 489 check_device(const char *path, boolean_t force,
 490     boolean_t isspare, boolean_t iswholedisk)
 491 {
 492         static blkid_cache cache = NULL;
 493
 494         /*
 495          * There is no easy way to add a correct blkid_put_cache() call,
 496          * memory will be reclaimed when the command exits.
 497          */
 498         if (cache == NULL) {
 499                 int err;
 500
 501                 if ((err = blkid_get_cache(&cache, NULL)) != 0) {
 502                         check_error(err);
 503                         return (-1);
 504                 }
 505
 506                 if ((err = blkid_probe_all(cache)) != 0) {
 507                         blkid_put_cache(cache);
 508                         check_error(err);
 509                         return (-1);
 510                 }
 511         }
 512
 513         return (check_disk(path, cache, force, isspare, iswholedisk));
 514 }
 515
 516 /*
 517  * By "whole disk" we mean an entire physical disk (something we can
 518  * label, toggle the write cache on, etc.) as opposed to the full
 519  * capacity of a pseudo-device such as lofi or did.  We act as if we
 520  * are labeling the disk, which should be a pretty good test of whether
 521  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 522  * it isn't.
 523  */
 524 static boolean_t
 525 is_whole_disk(const char *path)
 526 {
 527         struct dk_gpt *label;
 528         int fd;
 529
 530         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 531                 return (B_FALSE);
 532         if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
 533                 (void) close(fd);
 534                 return (B_FALSE);
 535         }
 536         efi_free(label);
 537         (void) close(fd);
 538         return (B_TRUE);
 539 }
 540
 541 /*
 542  * This may be a shorthand device path or it could be total gibberish.
 543  * Check to see if it is a known device available in zfs_vdev_paths.
 544  * As part of this check, see if we've been given an entire disk
 545  * (minus the slice number).
 546  */
 547 static int
 548 is_shorthand_path(const char *arg, char *path,
 549     struct stat64 *statbuf, boolean_t *wholedisk)
 550 {
 551         int error;
 552
 553         error = zfs_resolve_shortname(arg, path, MAXPATHLEN);
 554         if (error == 0) {
 555                 *wholedisk = is_whole_disk(path);
 556                 if (*wholedisk || (stat64(path, statbuf) == 0))
 557                         return (0);
 558         }
 559
 560         strlcpy(path, arg, sizeof (path));
 561         memset(statbuf, 0, sizeof (*statbuf));
 562         *wholedisk = B_FALSE;
 563
 564         return (error);
 565 }
 566
 567 /*
 568  * Determine if the given path is a hot spare within the given configuration.
 569  * If no configuration is given we rely solely on the label.
 570  */
 571 static boolean_t
 572 is_spare(nvlist_t *config, const char *path)
 573 {
 574         int fd;
 575         pool_state_t state;
 576         char *name = NULL;
 577         nvlist_t *label;
 578         uint64_t guid, spareguid;
 579         nvlist_t *nvroot;
 580         nvlist_t **spares;
 581         uint_t i, nspares;
 582         boolean_t inuse;
 583
 584         if ((fd = open(path, O_RDONLY)) < 0)
 585                 return (B_FALSE);
 586
 587         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
 588             !inuse ||
 589             state != POOL_STATE_SPARE ||
 590             zpool_read_label(fd, &label, NULL) != 0) {
 591                 free(name);
 592                 (void) close(fd);
 593                 return (B_FALSE);
 594         }
 595         free(name);
 596         (void) close(fd);
 597
 598         if (config == NULL)
 599                 return (B_TRUE);
 600
 601         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
 602         nvlist_free(label);
 603
 604         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 605             &nvroot) == 0);
 606         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 607             &spares, &nspares) == 0) {
 608                 for (i = 0; i < nspares; i++) {
 609                         verify(nvlist_lookup_uint64(spares[i],
 610                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
 611                         if (spareguid == guid)
 612                                 return (B_TRUE);
 613                 }
 614         }
 615
 616         return (B_FALSE);
 617 }
 618
 619 /*
 620  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 621  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 622  * leaf vdev are:
 623  *
 624  *      /dev/xxx        Complete disk path
 625  *      /xxx            Full path to file
 626  *      xxx             Shorthand for <zfs_vdev_paths>/xxx
 627  */
 628 static nvlist_t *
 629 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
 630 {
 631         char path[MAXPATHLEN];
 632         struct stat64 statbuf;
 633         nvlist_t *vdev = NULL;
 634         char *type = NULL;
 635         boolean_t wholedisk = B_FALSE;
 636         uint64_t ashift = 0;
 637         int err;
 638
 639         /*
 640          * Determine what type of vdev this is, and put the full path into
 641          * 'path'.  We detect whether this is a device of file afterwards by
 642          * checking the st_mode of the file.
 643          */
 644         if (arg[0] == '/') {
 645                 /*
 646                  * Complete device or file path.  Exact type is determined by
 647                  * examining the file descriptor afterwards.  Symbolic links
 648                  * are resolved to their real paths for the is_whole_disk()
 649                  * and S_ISBLK/S_ISREG type checks.  However, we are careful
 650                  * to store the given path as ZPOOL_CONFIG_PATH to ensure we
 651                  * can leverage udev's persistent device labels.
 652                  */
 653                 if (realpath(arg, path) == NULL) {
 654                         (void) fprintf(stderr,
 655                             gettext("cannot resolve path '%s'\n"), arg);
 656                         return (NULL);
 657                 }
 658
 659                 wholedisk = is_whole_disk(path);
 660                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 661                         (void) fprintf(stderr,
 662                             gettext("cannot open '%s': %s\n"),
 663                             path, strerror(errno));
 664                         return (NULL);
 665                 }
 666
 667                 /* After is_whole_disk() check restore original passed path */
 668                 strlcpy(path, arg, MAXPATHLEN);
 669         } else {
 670                 err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
 671                 if (err != 0) {
 672                         /*
 673                          * If we got ENOENT, then the user gave us
 674                          * gibberish, so try to direct them with a
 675                          * reasonable error message.  Otherwise,
 676                          * regurgitate strerror() since it's the best we
 677                          * can do.
 678                          */
 679                         if (err == ENOENT) {
 680                                 (void) fprintf(stderr,
 681                                     gettext("cannot open '%s': no such "
 682                                     "device in %s\n"), arg, DISK_ROOT);
 683                                 (void) fprintf(stderr,
 684                                     gettext("must be a full path or "
 685                                     "shorthand device name\n"));
 686                                 return (NULL);
 687                         } else {
 688                                 (void) fprintf(stderr,
 689                                     gettext("cannot open '%s': %s\n"),
 690                                     path, strerror(errno));
 691                                 return (NULL);
 692                         }
 693                 }
 694         }
 695
 696         /*
 697          * Determine whether this is a device or a file.
 698          */
 699         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 700                 type = VDEV_TYPE_DISK;
 701         } else if (S_ISREG(statbuf.st_mode)) {
 702                 type = VDEV_TYPE_FILE;
 703         } else {
 704                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 705                     "block device or regular file\n"), path);
 706                 return (NULL);
 707         }
 708
 709         /*
 710          * Finally, we have the complete device or file, and we know that it is
 711          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 712          * vdevs have a 'path' element, and devices also have a 'devid' element.
 713          */
 714         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 715         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 716         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 717         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 718         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 719                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 720                     (uint64_t)wholedisk) == 0);
 721
 722         /*
 723          * Override defaults if custom properties are provided.
 724          */
 725         if (props != NULL) {
 726                 char *value = NULL;
 727
 728                 if (nvlist_lookup_string(props,
 729                     zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
 730                         zfs_nicestrtonum(NULL, value, &ashift);
 731         }
 732
 733         /*
 734          * If the device is known to incorrectly report its physical sector
 735          * size explicitly provide the known correct value.
 736          */
 737         if (ashift == 0) {
 738                 int sector_size;
 739
 740                 if (check_sector_size_database(path, &sector_size) == B_TRUE)
 741                         ashift = highbit64(sector_size) - 1;
 742         }
 743
 744         if (ashift > 0)
 745                 nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
 746
 747         return (vdev);
 748 }
 749
 750 /*
 751  * Go through and verify the replication level of the pool is consistent.
 752  * Performs the following checks:
 753  *
 754  *      For the new spec, verifies that devices in mirrors and raidz are the
 755  *      same size.
 756  *
 757  *      If the current configuration already has inconsistent replication
 758  *      levels, ignore any other potential problems in the new spec.
 759  *
 760  *      Otherwise, make sure that the current spec (if there is one) and the new
 761  *      spec have consistent replication levels.
 762  */
 763 typedef struct replication_level {
 764         char *zprl_type;
 765         uint64_t zprl_children;
 766         uint64_t zprl_parity;
 767 } replication_level_t;
 768
 769 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 770
 771 /*
 772  * Given a list of toplevel vdevs, return the current replication level.  If
 773  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 774  * an error message will be displayed for each self-inconsistent vdev.
 775  */
 776 static replication_level_t *
 777 get_replication(nvlist_t *nvroot, boolean_t fatal)
 778 {
 779         nvlist_t **top;
 780         uint_t t, toplevels;
 781         nvlist_t **child;
 782         uint_t c, children;
 783         nvlist_t *nv;
 784         char *type;
 785         replication_level_t lastrep = { 0 }, rep, *ret;
 786         boolean_t dontreport;
 787
 788         ret = safe_malloc(sizeof (replication_level_t));
 789
 790         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 791             &top, &toplevels) == 0);
 792
 793         lastrep.zprl_type = NULL;
 794         for (t = 0; t < toplevels; t++) {
 795                 uint64_t is_log = B_FALSE;
 796
 797                 nv = top[t];
 798
 799                 /*
 800                  * For separate logs we ignore the top level vdev replication
 801                  * constraints.
 802                  */
 803                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 804                 if (is_log)
 805                         continue;
 806
 807                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 808                     &type) == 0);
 809                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 810                     &child, &children) != 0) {
 811                         /*
 812                          * This is a 'file' or 'disk' vdev.
 813                          */
 814                         rep.zprl_type = type;
 815                         rep.zprl_children = 1;
 816                         rep.zprl_parity = 0;
 817                 } else {
 818                         uint64_t vdev_size;
 819
 820                         /*
 821                          * This is a mirror or RAID-Z vdev.  Go through and make
 822                          * sure the contents are all the same (files vs. disks),
 823                          * keeping track of the number of elements in the
 824                          * process.
 825                          *
 826                          * We also check that the size of each vdev (if it can
 827                          * be determined) is the same.
 828                          */
 829                         rep.zprl_type = type;
 830                         rep.zprl_children = 0;
 831
 832                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 833                                 verify(nvlist_lookup_uint64(nv,
 834                                     ZPOOL_CONFIG_NPARITY,
 835                                     &rep.zprl_parity) == 0);
 836                                 assert(rep.zprl_parity != 0);
 837                         } else {
 838                                 rep.zprl_parity = 0;
 839                         }
 840
 841                         /*
 842                          * The 'dontreport' variable indicates that we've
 843                          * already reported an error for this spec, so don't
 844                          * bother doing it again.
 845                          */
 846                         type = NULL;
 847                         dontreport = 0;
 848                         vdev_size = -1ULL;
 849                         for (c = 0; c < children; c++) {
 850                                 nvlist_t *cnv = child[c];
 851                                 char *path;
 852                                 struct stat64 statbuf;
 853                                 uint64_t size = -1ULL;
 854                                 char *childtype;
 855                                 int fd, err;
 856
 857                                 rep.zprl_children++;
 858
 859                                 verify(nvlist_lookup_string(cnv,
 860                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 861
 862                                 /*
 863                                  * If this is a replacing or spare vdev, then
 864                                  * get the real first child of the vdev.
 865                                  */
 866                                 if (strcmp(childtype,
 867                                     VDEV_TYPE_REPLACING) == 0 ||
 868                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 869                                         nvlist_t **rchild;
 870                                         uint_t rchildren;
 871
 872                                         verify(nvlist_lookup_nvlist_array(cnv,
 873                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 874                                             &rchildren) == 0);
 875                                         assert(rchildren == 2);
 876                                         cnv = rchild[0];
 877
 878                                         verify(nvlist_lookup_string(cnv,
 879                                             ZPOOL_CONFIG_TYPE,
 880                                             &childtype) == 0);
 881                                 }
 882
 883                                 verify(nvlist_lookup_string(cnv,
 884                                     ZPOOL_CONFIG_PATH, &path) == 0);
 885
 886                                 /*
 887                                  * If we have a raidz/mirror that combines disks
 888                                  * with files, report it as an error.
 889                                  */
 890                                 if (!dontreport && type != NULL &&
 891                                     strcmp(type, childtype) != 0) {
 892                                         if (ret != NULL)
 893                                                 free(ret);
 894                                         ret = NULL;
 895                                         if (fatal)
 896                                                 vdev_error(gettext(
 897                                                     "mismatched replication "
 898                                                     "level: %s contains both "
 899                                                     "files and devices\n"),
 900                                                     rep.zprl_type);
 901                                         else
 902                                                 return (NULL);
 903                                         dontreport = B_TRUE;
 904                                 }
 905
 906                                 /*
 907                                  * According to stat(2), the value of 'st_size'
 908                                  * is undefined for block devices and character
 909                                  * devices.  But there is no effective way to
 910                                  * determine the real size in userland.
 911                                  *
 912                                  * Instead, we'll take advantage of an
 913                                  * implementation detail of spec_size().  If the
 914                                  * device is currently open, then we (should)
 915                                  * return a valid size.
 916                                  *
 917                                  * If we still don't get a valid size (indicated
 918                                  * by a size of 0 or MAXOFFSET_T), then ignore
 919                                  * this device altogether.
 920                                  */
 921                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 922                                         err = fstat64(fd, &statbuf);
 923                                         (void) close(fd);
 924                                 } else {
 925                                         err = stat64(path, &statbuf);
 926                                 }
 927
 928                                 if (err != 0 ||
 929                                     statbuf.st_size == 0 ||
 930                                     statbuf.st_size == MAXOFFSET_T)
 931                                         continue;
 932
 933                                 size = statbuf.st_size;
 934
 935                                 /*
 936                                  * Also make sure that devices and
 937                                  * slices have a consistent size.  If
 938                                  * they differ by a significant amount
 939                                  * (~16MB) then report an error.
 940                                  */
 941                                 if (!dontreport &&
 942                                     (vdev_size != -1ULL &&
 943                                     (labs(size - vdev_size) >
 944                                     ZPOOL_FUZZ))) {
 945                                         if (ret != NULL)
 946                                                 free(ret);
 947                                         ret = NULL;
 948                                         if (fatal)
 949                                                 vdev_error(gettext(
 950                                                     "%s contains devices of "
 951                                                     "different sizes\n"),
 952                                                     rep.zprl_type);
 953                                         else
 954                                                 return (NULL);
 955                                         dontreport = B_TRUE;
 956                                 }
 957
 958                                 type = childtype;
 959                                 vdev_size = size;
 960                         }
 961                 }
 962
 963                 /*
 964                  * At this point, we have the replication of the last toplevel
 965                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 966                  * different.
 967                  */
 968                 if (lastrep.zprl_type != NULL) {
 969                         if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
 970                                 if (ret != NULL)
 971                                         free(ret);
 972                                 ret = NULL;
 973                                 if (fatal)
 974                                         vdev_error(gettext(
 975                                             "mismatched replication level: "
 976                                             "both %s and %s vdevs are "
 977                                             "present\n"),
 978                                             lastrep.zprl_type, rep.zprl_type);
 979                                 else
 980                                         return (NULL);
 981                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 982                                 if (ret)
 983                                         free(ret);
 984                                 ret = NULL;
 985                                 if (fatal)
 986                                         vdev_error(gettext(
 987                                             "mismatched replication level: "
 988                                             "both %llu and %llu device parity "
 989                                             "%s vdevs are present\n"),
 990                                             lastrep.zprl_parity,
 991                                             rep.zprl_parity,
 992                                             rep.zprl_type);
 993                                 else
 994                                         return (NULL);
 995                         } else if (lastrep.zprl_children != rep.zprl_children) {
 996                                 if (ret)
 997                                         free(ret);
 998                                 ret = NULL;
 999                                 if (fatal)
1000                                         vdev_error(gettext(
1001                                             "mismatched replication level: "
1002                                             "both %llu-way and %llu-way %s "
1003                                             "vdevs are present\n"),
1004                                             lastrep.zprl_children,
1005                                             rep.zprl_children,
1006                                             rep.zprl_type);
1007                                 else
1008                                         return (NULL);
1009                         }
1010                 }
1011                 lastrep = rep;
1012         }
1013
1014         if (ret != NULL)
1015                 *ret = rep;
1016
1017         return (ret);
1018 }
1019
1020 /*
1021  * Check the replication level of the vdev spec against the current pool.  Calls
1022  * get_replication() to make sure the new spec is self-consistent.  If the pool
1023  * has a consistent replication level, then we ignore any errors.  Otherwise,
1024  * report any difference between the two.
1025  */
1026 static int
1027 check_replication(nvlist_t *config, nvlist_t *newroot)
1028 {
1029         nvlist_t **child;
1030         uint_t  children;
1031         replication_level_t *current = NULL, *new;
1032         int ret;
1033
1034         /*
1035          * If we have a current pool configuration, check to see if it's
1036          * self-consistent.  If not, simply return success.
1037          */
1038         if (config != NULL) {
1039                 nvlist_t *nvroot;
1040
1041                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1042                     &nvroot) == 0);
1043                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1044                         return (0);
1045         }
1046         /*
1047          * for spares there may be no children, and therefore no
1048          * replication level to check
1049          */
1050         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1051             &child, &children) != 0) || (children == 0)) {
1052                 free(current);
1053                 return (0);
1054         }
1055
1056         /*
1057          * If all we have is logs then there's no replication level to check.
1058          */
1059         if (num_logs(newroot) == children) {
1060                 free(current);
1061                 return (0);
1062         }
1063
1064         /*
1065          * Get the replication level of the new vdev spec, reporting any
1066          * inconsistencies found.
1067          */
1068         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1069                 free(current);
1070                 return (-1);
1071         }
1072
1073         /*
1074          * Check to see if the new vdev spec matches the replication level of
1075          * the current pool.
1076          */
1077         ret = 0;
1078         if (current != NULL) {
1079                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1080                         vdev_error(gettext(
1081                             "mismatched replication level: pool uses %s "
1082                             "and new vdev is %s\n"),
1083                             current->zprl_type, new->zprl_type);
1084                         ret = -1;
1085                 } else if (current->zprl_parity != new->zprl_parity) {
1086                         vdev_error(gettext(
1087                             "mismatched replication level: pool uses %llu "
1088                             "device parity and new vdev uses %llu\n"),
1089                             current->zprl_parity, new->zprl_parity);
1090                         ret = -1;
1091                 } else if (current->zprl_children != new->zprl_children) {
1092                         vdev_error(gettext(
1093                             "mismatched replication level: pool uses %llu-way "
1094                             "%s and new vdev uses %llu-way %s\n"),
1095                             current->zprl_children, current->zprl_type,
1096                             new->zprl_children, new->zprl_type);
1097                         ret = -1;
1098                 }
1099         }
1100
1101         free(new);
1102         if (current != NULL)
1103                 free(current);
1104
1105         return (ret);
1106 }
1107
1108 static int
1109 zero_label(char *path)
1110 {
1111         const int size = 4096;
1112         char buf[size];
1113         int err, fd;
1114
1115         if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1116                 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1117                     path, strerror(errno));
1118                 return (-1);
1119         }
1120
1121         memset(buf, 0, size);
1122         err = write(fd, buf, size);
1123         (void) fdatasync(fd);
1124         (void) close(fd);
1125
1126         if (err == -1) {
1127                 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1128                     "of '%s': %s\n"), size, path, strerror(errno));
1129                 return (-1);
1130         }
1131
1132         if (err != size) {
1133                 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1134                     "of '%s'\n"), err, size, path);
1135                 return (-1);
1136         }
1137
1138         return (0);
1139 }
1140
1141 /*
1142  * Go through and find any whole disks in the vdev specification, labelling them
1143  * as appropriate.  When constructing the vdev spec, we were unable to open this
1144  * device in order to provide a devid.  Now that we have labelled the disk and
1145  * know that slice 0 is valid, we can construct the devid now.
1146  *
1147  * If the disk was already labeled with an EFI label, we will have gotten the
1148  * devid already (because we were able to open the whole disk).  Otherwise, we
1149  * need to get the devid after we label the disk.
1150  */
1151 static int
1152 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1153 {
1154         nvlist_t **child;
1155         uint_t c, children;
1156         char *type, *path;
1157         char devpath[MAXPATHLEN];
1158         char udevpath[MAXPATHLEN];
1159         uint64_t wholedisk;
1160         struct stat64 statbuf;
1161         int is_exclusive = 0;
1162         int fd;
1163         int ret;
1164
1165         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1166
1167         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1168             &child, &children) != 0) {
1169
1170                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1171                         return (0);
1172
1173                 /*
1174                  * We have a disk device.  If this is a whole disk write
1175                  * out the efi partition table, otherwise write zero's to
1176                  * the first 4k of the partition.  This is to ensure that
1177                  * libblkid will not misidentify the partition due to a
1178                  * magic value left by the previous filesystem.
1179                  */
1180                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1181                 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1182                     &wholedisk));
1183
1184                 if (!wholedisk) {
1185                         (void) zero_label(path);
1186                         return (0);
1187                 }
1188
1189                 if (realpath(path, devpath) == NULL) {
1190                         ret = errno;
1191                         (void) fprintf(stderr,
1192                             gettext("cannot resolve path '%s'\n"), path);
1193                         return (ret);
1194                 }
1195
1196                 /*
1197                  * Remove any previously existing symlink from a udev path to
1198                  * the device before labeling the disk.  This makes
1199                  * zpool_label_disk_wait() truly wait for the new link to show
1200                  * up instead of returning if it finds an old link still in
1201                  * place.  Otherwise there is a window between when udev
1202                  * deletes and recreates the link during which access attempts
1203                  * will fail with ENOENT.
1204                  */
1205                 strncpy(udevpath, path, MAXPATHLEN);
1206                 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1207
1208                 fd = open(devpath, O_RDWR|O_EXCL);
1209                 if (fd == -1) {
1210                         if (errno == EBUSY)
1211                                 is_exclusive = 1;
1212                 } else {
1213                         (void) close(fd);
1214                 }
1215
1216                 /*
1217                  * If the partition exists, contains a valid spare label,
1218                  * and is opened exclusively there is no need to partition
1219                  * it.  Hot spares have already been partitioned and are
1220                  * held open exclusively by the kernel as a safety measure.
1221                  *
1222                  * If the provided path is for a /dev/disk/ device its
1223                  * symbolic link will be removed, partition table created,
1224                  * and then block until udev creates the new link.
1225                  */
1226                 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1227                         ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1228                         if (ret == 0) {
1229                                 ret = lstat64(udevpath, &statbuf);
1230                                 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1231                                         (void) unlink(udevpath);
1232                         }
1233
1234                         if (zpool_label_disk(g_zfs, zhp,
1235                             strrchr(devpath, '/') + 1) == -1)
1236                                 return (-1);
1237
1238                         ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1239                         if (ret) {
1240                                 (void) fprintf(stderr, gettext("cannot "
1241                                     "resolve path '%s': %d\n"), udevpath, ret);
1242                                 return (-1);
1243                         }
1244
1245                         (void) zero_label(udevpath);
1246                 }
1247
1248                 /*
1249                  * Update the path to refer to the partition.  The presence of
1250                  * the 'whole_disk' field indicates to the CLI that we should
1251                  * chop off the partition number when displaying the device in
1252                  * future output.
1253                  */
1254                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1255
1256                 return (0);
1257         }
1258
1259         for (c = 0; c < children; c++)
1260                 if ((ret = make_disks(zhp, child[c])) != 0)
1261                         return (ret);
1262
1263         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1264             &child, &children) == 0)
1265                 for (c = 0; c < children; c++)
1266                         if ((ret = make_disks(zhp, child[c])) != 0)
1267                                 return (ret);
1268
1269         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1270             &child, &children) == 0)
1271                 for (c = 0; c < children; c++)
1272                         if ((ret = make_disks(zhp, child[c])) != 0)
1273                                 return (ret);
1274
1275         return (0);
1276 }
1277
1278 /*
1279  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1280  * the majority of this task.
1281  */
1282 static boolean_t
1283 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1284     boolean_t replacing, boolean_t isspare)
1285 {
1286         nvlist_t **child;
1287         uint_t c, children;
1288         char *type, *path;
1289         int ret = 0;
1290         char buf[MAXPATHLEN];
1291         uint64_t wholedisk = B_FALSE;
1292         boolean_t anyinuse = B_FALSE;
1293
1294         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1295
1296         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1297             &child, &children) != 0) {
1298
1299                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1300                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1301                         verify(!nvlist_lookup_uint64(nv,
1302                             ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1303
1304                 /*
1305                  * As a generic check, we look to see if this is a replace of a
1306                  * hot spare within the same pool.  If so, we allow it
1307                  * regardless of what libblkid or zpool_in_use() says.
1308                  */
1309                 if (replacing) {
1310                         (void) strlcpy(buf, path, sizeof (buf));
1311                         if (wholedisk) {
1312                                 ret = zfs_append_partition(buf,  sizeof (buf));
1313                                 if (ret == -1)
1314                                         return (-1);
1315                         }
1316
1317                         if (is_spare(config, buf))
1318                                 return (B_FALSE);
1319                 }
1320
1321                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1322                         ret = check_device(path, force, isspare, wholedisk);
1323
1324                 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1325                         ret = check_file(path, force, isspare);
1326
1327                 return (ret != 0);
1328         }
1329
1330         for (c = 0; c < children; c++)
1331                 if (is_device_in_use(config, child[c], force, replacing,
1332                     B_FALSE))
1333                         anyinuse = B_TRUE;
1334
1335         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1336             &child, &children) == 0)
1337                 for (c = 0; c < children; c++)
1338                         if (is_device_in_use(config, child[c], force, replacing,
1339                             B_TRUE))
1340                                 anyinuse = B_TRUE;
1341
1342         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1343             &child, &children) == 0)
1344                 for (c = 0; c < children; c++)
1345                         if (is_device_in_use(config, child[c], force, replacing,
1346                             B_FALSE))
1347                                 anyinuse = B_TRUE;
1348
1349         return (anyinuse);
1350 }
1351
1352 static const char *
1353 is_grouping(const char *type, int *mindev, int *maxdev)
1354 {
1355         if (strncmp(type, "raidz", 5) == 0) {
1356                 const char *p = type + 5;
1357                 char *end;
1358                 long nparity;
1359
1360                 if (*p == '\0') {
1361                         nparity = 1;
1362                 } else if (*p == '0') {
1363                         return (NULL); /* no zero prefixes allowed */
1364                 } else {
1365                         errno = 0;
1366                         nparity = strtol(p, &end, 10);
1367                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1368                             *end != '\0')
1369                                 return (NULL);
1370                 }
1371
1372                 if (mindev != NULL)
1373                         *mindev = nparity + 1;
1374                 if (maxdev != NULL)
1375                         *maxdev = 255;
1376                 return (VDEV_TYPE_RAIDZ);
1377         }
1378
1379         if (maxdev != NULL)
1380                 *maxdev = INT_MAX;
1381
1382         if (strcmp(type, "mirror") == 0) {
1383                 if (mindev != NULL)
1384                         *mindev = 2;
1385                 return (VDEV_TYPE_MIRROR);
1386         }
1387
1388         if (strcmp(type, "spare") == 0) {
1389                 if (mindev != NULL)
1390                         *mindev = 1;
1391                 return (VDEV_TYPE_SPARE);
1392         }
1393
1394         if (strcmp(type, "log") == 0) {
1395                 if (mindev != NULL)
1396                         *mindev = 1;
1397                 return (VDEV_TYPE_LOG);
1398         }
1399
1400         if (strcmp(type, "cache") == 0) {
1401                 if (mindev != NULL)
1402                         *mindev = 1;
1403                 return (VDEV_TYPE_L2CACHE);
1404         }
1405
1406         return (NULL);
1407 }
1408
1409 /*
1410  * Construct a syntactically valid vdev specification,
1411  * and ensure that all devices and files exist and can be opened.
1412  * Note: we don't bother freeing anything in the error paths
1413  * because the program is just going to exit anyway.
1414  */
1415 nvlist_t *
1416 construct_spec(nvlist_t *props, int argc, char **argv)
1417 {
1418         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1419         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1420         const char *type;
1421         uint64_t is_log;
1422         boolean_t seen_logs;
1423
1424         top = NULL;
1425         toplevels = 0;
1426         spares = NULL;
1427         l2cache = NULL;
1428         nspares = 0;
1429         nlogs = 0;
1430         nl2cache = 0;
1431         is_log = B_FALSE;
1432         seen_logs = B_FALSE;
1433
1434         while (argc > 0) {
1435                 nv = NULL;
1436
1437                 /*
1438                  * If it's a mirror or raidz, the subsequent arguments are
1439                  * its leaves -- until we encounter the next mirror or raidz.
1440                  */
1441                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1442                         nvlist_t **child = NULL;
1443                         int c, children = 0;
1444
1445                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1446                                 if (spares != NULL) {
1447                                         (void) fprintf(stderr,
1448                                             gettext("invalid vdev "
1449                                             "specification: 'spare' can be "
1450                                             "specified only once\n"));
1451                                         return (NULL);
1452                                 }
1453                                 is_log = B_FALSE;
1454                         }
1455
1456                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1457                                 if (seen_logs) {
1458                                         (void) fprintf(stderr,
1459                                             gettext("invalid vdev "
1460                                             "specification: 'log' can be "
1461                                             "specified only once\n"));
1462                                         return (NULL);
1463                                 }
1464                                 seen_logs = B_TRUE;
1465                                 is_log = B_TRUE;
1466                                 argc--;
1467                                 argv++;
1468                                 /*
1469                                  * A log is not a real grouping device.
1470                                  * We just set is_log and continue.
1471                                  */
1472                                 continue;
1473                         }
1474
1475                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1476                                 if (l2cache != NULL) {
1477                                         (void) fprintf(stderr,
1478                                             gettext("invalid vdev "
1479                                             "specification: 'cache' can be "
1480                                             "specified only once\n"));
1481                                         return (NULL);
1482                                 }
1483                                 is_log = B_FALSE;
1484                         }
1485
1486                         if (is_log) {
1487                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1488                                         (void) fprintf(stderr,
1489                                             gettext("invalid vdev "
1490                                             "specification: unsupported 'log' "
1491                                             "device: %s\n"), type);
1492                                         return (NULL);
1493                                 }
1494                                 nlogs++;
1495                         }
1496
1497                         for (c = 1; c < argc; c++) {
1498                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1499                                         break;
1500                                 children++;
1501                                 child = realloc(child,
1502                                     children * sizeof (nvlist_t *));
1503                                 if (child == NULL)
1504                                         zpool_no_memory();
1505                                 if ((nv = make_leaf_vdev(props, argv[c],
1506                                     B_FALSE)) == NULL)
1507                                         return (NULL);
1508                                 child[children - 1] = nv;
1509                         }
1510
1511                         if (children < mindev) {
1512                                 (void) fprintf(stderr, gettext("invalid vdev "
1513                                     "specification: %s requires at least %d "
1514                                     "devices\n"), argv[0], mindev);
1515                                 return (NULL);
1516                         }
1517
1518                         if (children > maxdev) {
1519                                 (void) fprintf(stderr, gettext("invalid vdev "
1520                                     "specification: %s supports no more than "
1521                                     "%d devices\n"), argv[0], maxdev);
1522                                 return (NULL);
1523                         }
1524
1525                         argc -= c;
1526                         argv += c;
1527
1528                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1529                                 spares = child;
1530                                 nspares = children;
1531                                 continue;
1532                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1533                                 l2cache = child;
1534                                 nl2cache = children;
1535                                 continue;
1536                         } else {
1537                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1538                                     0) == 0);
1539                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1540                                     type) == 0);
1541                                 verify(nvlist_add_uint64(nv,
1542                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1543                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1544                                         verify(nvlist_add_uint64(nv,
1545                                             ZPOOL_CONFIG_NPARITY,
1546                                             mindev - 1) == 0);
1547                                 }
1548                                 verify(nvlist_add_nvlist_array(nv,
1549                                     ZPOOL_CONFIG_CHILDREN, child,
1550                                     children) == 0);
1551
1552                                 for (c = 0; c < children; c++)
1553                                         nvlist_free(child[c]);
1554                                 free(child);
1555                         }
1556                 } else {
1557                         /*
1558                          * We have a device.  Pass off to make_leaf_vdev() to
1559                          * construct the appropriate nvlist describing the vdev.
1560                          */
1561                         if ((nv = make_leaf_vdev(props, argv[0],
1562                             is_log)) == NULL)
1563                                 return (NULL);
1564                         if (is_log)
1565                                 nlogs++;
1566                         argc--;
1567                         argv++;
1568                 }
1569
1570                 toplevels++;
1571                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1572                 if (top == NULL)
1573                         zpool_no_memory();
1574                 top[toplevels - 1] = nv;
1575         }
1576
1577         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1578                 (void) fprintf(stderr, gettext("invalid vdev "
1579                     "specification: at least one toplevel vdev must be "
1580                     "specified\n"));
1581                 return (NULL);
1582         }
1583
1584         if (seen_logs && nlogs == 0) {
1585                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1586                     "log requires at least 1 device\n"));
1587                 return (NULL);
1588         }
1589
1590         /*
1591          * Finally, create nvroot and add all top-level vdevs to it.
1592          */
1593         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1594         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1595             VDEV_TYPE_ROOT) == 0);
1596         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1597             top, toplevels) == 0);
1598         if (nspares != 0)
1599                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1600                     spares, nspares) == 0);
1601         if (nl2cache != 0)
1602                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1603                     l2cache, nl2cache) == 0);
1604
1605         for (t = 0; t < toplevels; t++)
1606                 nvlist_free(top[t]);
1607         for (t = 0; t < nspares; t++)
1608                 nvlist_free(spares[t]);
1609         for (t = 0; t < nl2cache; t++)
1610                 nvlist_free(l2cache[t]);
1611         if (spares)
1612                 free(spares);
1613         if (l2cache)
1614                 free(l2cache);
1615         free(top);
1616
1617         return (nvroot);
1618 }
1619
1620 nvlist_t *
1621 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1622     splitflags_t flags, int argc, char **argv)
1623 {
1624         nvlist_t *newroot = NULL, **child;
1625         uint_t c, children;
1626
1627         if (argc > 0) {
1628                 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1629                         (void) fprintf(stderr, gettext("Unable to build a "
1630                             "pool from the specified devices\n"));
1631                         return (NULL);
1632                 }
1633
1634                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1635                         nvlist_free(newroot);
1636                         return (NULL);
1637                 }
1638
1639                 /* avoid any tricks in the spec */
1640                 verify(nvlist_lookup_nvlist_array(newroot,
1641                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1642                 for (c = 0; c < children; c++) {
1643                         char *path;
1644                         const char *type;
1645                         int min, max;
1646
1647                         verify(nvlist_lookup_string(child[c],
1648                             ZPOOL_CONFIG_PATH, &path) == 0);
1649                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1650                                 (void) fprintf(stderr, gettext("Cannot use "
1651                                     "'%s' as a device for splitting\n"), type);
1652                                 nvlist_free(newroot);
1653                                 return (NULL);
1654                         }
1655                 }
1656         }
1657
1658         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1659                 if (newroot != NULL)
1660                         nvlist_free(newroot);
1661                 return (NULL);
1662         }
1663
1664         return (newroot);
1665 }
1666
1667 /*
1668  * Get and validate the contents of the given vdev specification.  This ensures
1669  * that the nvlist returned is well-formed, that all the devices exist, and that
1670  * they are not currently in use by any other known consumer.  The 'poolconfig'
1671  * parameter is the current configuration of the pool when adding devices
1672  * existing pool, and is used to perform additional checks, such as changing the
1673  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1674  * new pool.  The 'force' flag controls whether devices should be forcefully
1675  * added, even if they appear in use.
1676  */
1677 nvlist_t *
1678 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1679     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1680 {
1681         nvlist_t *newroot;
1682         nvlist_t *poolconfig = NULL;
1683         is_force = force;
1684
1685         /*
1686          * Construct the vdev specification.  If this is successful, we know
1687          * that we have a valid specification, and that all devices can be
1688          * opened.
1689          */
1690         if ((newroot = construct_spec(props, argc, argv)) == NULL)
1691                 return (NULL);
1692
1693         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1694                 nvlist_free(newroot);
1695                 return (NULL);
1696         }
1697
1698         /*
1699          * Validate each device to make sure that its not shared with another
1700          * subsystem.  We do this even if 'force' is set, because there are some
1701          * uses (such as a dedicated dump device) that even '-f' cannot
1702          * override.
1703          */
1704         if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1705                 nvlist_free(newroot);
1706                 return (NULL);
1707         }
1708
1709         /*
1710          * Check the replication level of the given vdevs and report any errors
1711          * found.  We include the existing pool spec, if any, as we need to
1712          * catch changes against the existing replication level.
1713          */
1714         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1715                 nvlist_free(newroot);
1716                 return (NULL);
1717         }
1718
1719         /*
1720          * Run through the vdev specification and label any whole disks found.
1721          */
1722         if (!dryrun && make_disks(zhp, newroot) != 0) {
1723                 nvlist_free(newroot);
1724                 return (NULL);
1725         }
1726
1727         return (newroot);
1728 }