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  */
  25
  26 /*
  27  * Functions to convert between a list of vdevs and an nvlist representing the
  28  * configuration.  Each entry in the list can be one of:
  29  *
  30  *      Device vdevs
  31  *              disk=(path=..., devid=...)
  32  *              file=(path=...)
  33  *
  34  *      Group vdevs
  35  *              raidz[1|2]=(...)
  36  *              mirror=(...)
  37  *
  38  *      Hot spares
  39  *
  40  * While the underlying implementation supports it, group vdevs cannot contain
  41  * other group vdevs.  All userland verification of devices is contained within
  42  * this file.  If successful, the nvlist returned can be passed directly to the
  43  * kernel; we've done as much verification as possible in userland.
  44  *
  45  * Hot spares are a special case, and passed down as an array of disk vdevs, at
  46  * the same level as the root of the vdev tree.
  47  *
  48  * The only function exported by this file is 'make_root_vdev'.  The
  49  * function performs several passes:
  50  *
  51  *      1. Construct the vdev specification.  Performs syntax validation and
  52  *         makes sure each device is valid.
  53  *      2. Check for devices in use.  Using libblkid to make sure that no
  54  *         devices are also in use.  Some can be overridden using the 'force'
  55  *         flag, others cannot.
  56  *      3. Check for replication errors if the 'force' flag is not specified.
  57  *         validates that the replication level is consistent across the
  58  *         entire pool.
  59  *      4. Call libzfs to label any whole disks with an EFI label.
  60  */
  61
  62 #include <assert.h>
  63 #include <ctype.h>
  64 #include <devid.h>
  65 #include <errno.h>
  66 #include <fcntl.h>
  67 #include <libintl.h>
  68 #include <libnvpair.h>
  69 #include <limits.h>
  70 #include <scsi/scsi.h>
  71 #include <scsi/sg.h>
  72 #include <stdio.h>
  73 #include <string.h>
  74 #include <unistd.h>
  75 #include <sys/efi_partition.h>
  76 #include <sys/stat.h>
  77 #include <sys/vtoc.h>
  78 #include <sys/mntent.h>
  79 #include <uuid/uuid.h>
  80 #ifdef HAVE_LIBBLKID
  81 #include <blkid/blkid.h>
  82 #else
  83 #define blkid_cache void *
  84 #endif /* HAVE_LIBBLKID */
  85
  86 #include "zpool_util.h"
  87 #include <sys/zfs_context.h>
  88
  89 /*
  90  * For any given vdev specification, we can have multiple errors.  The
  91  * vdev_error() function keeps track of whether we have seen an error yet, and
  92  * prints out a header if its the first error we've seen.
  93  */
  94 boolean_t error_seen;
  95 boolean_t is_force;
  96
  97 typedef struct vdev_disk_db_entry
  98 {
  99         char id[24];
 100         int sector_size;
 101 } vdev_disk_db_entry_t;
 102
 103 /*
 104  * Database of block devices that lie about physical sector sizes.  The
 105  * identification string must be precisely 24 characters to avoid false
 106  * negatives
 107  */
 108 static vdev_disk_db_entry_t vdev_disk_database[] = {
 109         {"ATA     ADATA SSD S396 3", 8192},
 110         {"ATA     APPLE SSD SM128E", 8192},
 111         {"ATA     APPLE SSD SM256E", 8192},
 112         {"ATA     APPLE SSD SM512E", 8192},
 113         {"ATA     APPLE SSD SM768E", 8192},
 114         {"ATA     C400-MTFDDAC064M", 8192},
 115         {"ATA     C400-MTFDDAC128M", 8192},
 116         {"ATA     C400-MTFDDAC256M", 8192},
 117         {"ATA     C400-MTFDDAC512M", 8192},
 118         {"ATA     Corsair Force 3 ", 8192},
 119         {"ATA     Corsair Force GS", 8192},
 120         {"ATA     INTEL SSDSA2CT04", 8192},
 121         {"ATA     INTEL SSDSA2BZ10", 8192},
 122         {"ATA     INTEL SSDSA2BZ20", 8192},
 123         {"ATA     INTEL SSDSA2BZ30", 8192},
 124         {"ATA     INTEL SSDSA2CW04", 8192},
 125         {"ATA     INTEL SSDSA2CW08", 8192},
 126         {"ATA     INTEL SSDSA2CW12", 8192},
 127         {"ATA     INTEL SSDSA2CW16", 8192},
 128         {"ATA     INTEL SSDSA2CW30", 8192},
 129         {"ATA     INTEL SSDSA2CW60", 8192},
 130         {"ATA     INTEL SSDSC2BA10", 8192},
 131         {"ATA     INTEL SSDSC2BA20", 8192},
 132         {"ATA     INTEL SSDSC2BA40", 8192},
 133         {"ATA     INTEL SSDSC2BA80", 8192},
 134         {"ATA     INTEL SSDSC2BB08", 8192},
 135         {"ATA     INTEL SSDSC2BB12", 8192},
 136         {"ATA     INTEL SSDSC2BB16", 8192},
 137         {"ATA     INTEL SSDSC2BB24", 8192},
 138         {"ATA     INTEL SSDSC2BB30", 8192},
 139         {"ATA     INTEL SSDSC2BB40", 8192},
 140         {"ATA     INTEL SSDSC2BB48", 8192},
 141         {"ATA     INTEL SSDSC2BB60", 8192},
 142         {"ATA     INTEL SSDSC2BB80", 8192},
 143         {"ATA     INTEL SSDSC2CT06", 8192},
 144         {"ATA     INTEL SSDSC2CT12", 8192},
 145         {"ATA     INTEL SSDSC2CT18", 8192},
 146         {"ATA     INTEL SSDSC2CT24", 8192},
 147         {"ATA     INTEL SSDSC2CW06", 8192},
 148         {"ATA     INTEL SSDSC2CW12", 8192},
 149         {"ATA     INTEL SSDSC2CW18", 8192},
 150         {"ATA     INTEL SSDSC2CW24", 8192},
 151         {"ATA     INTEL SSDSC2CW48", 8192},
 152         {"ATA     KINGSTON SH100S3", 8192},
 153         {"ATA     KINGSTON SH103S3", 8192},
 154         {"ATA     M4-CT064M4SSD2  ", 8192},
 155         {"ATA     M4-CT128M4SSD2  ", 8192},
 156         {"ATA     M4-CT256M4SSD2  ", 8192},
 157         {"ATA     M4-CT512M4SSD2  ", 8192},
 158         {"ATA     OCZ-AGILITY2    ", 8192},
 159         {"ATA     OCZ-AGILITY3    ", 8192},
 160         {"ATA     OCZ-VERTEX2 3.5 ", 8192},
 161         {"ATA     OCZ-VERTEX3     ", 8192},
 162         {"ATA     OCZ-VERTEX3 LT  ", 8192},
 163         {"ATA     OCZ-VERTEX3 MI  ", 8192},
 164         {"ATA     OCZ-VERTEX4     ", 8192},
 165         {"ATA     SAMSUNG MZ7WD120", 8192},
 166         {"ATA     SAMSUNG MZ7WD240", 8192},
 167         {"ATA     SAMSUNG MZ7WD480", 8192},
 168         {"ATA     SAMSUNG MZ7WD960", 8192},
 169         {"ATA     SAMSUNG SSD 830 ", 8192},
 170         {"ATA     Samsung SSD 840 ", 8192},
 171         {"ATA     SanDisk SSD U100", 8192},
 172         {"ATA     TOSHIBA THNSNH06", 8192},
 173         {"ATA     TOSHIBA THNSNH12", 8192},
 174         {"ATA     TOSHIBA THNSNH25", 8192},
 175         {"ATA     TOSHIBA THNSNH51", 8192},
 176         {"ATA     APPLE SSD TS064C", 4096},
 177         {"ATA     APPLE SSD TS128C", 4096},
 178         {"ATA     APPLE SSD TS256C", 4096},
 179         {"ATA     APPLE SSD TS512C", 4096},
 180         {"ATA     INTEL SSDSA2M040", 4096},
 181         {"ATA     INTEL SSDSA2M080", 4096},
 182         {"ATA     INTEL SSDSA2M160", 4096},
 183         {"ATA     INTEL SSDSC2MH12", 4096},
 184         {"ATA     INTEL SSDSC2MH25", 4096},
 185         {"ATA     OCZ CORE_SSD    ", 4096},
 186         {"ATA     OCZ-VERTEX      ", 4096},
 187         {"ATA     SAMSUNG MCCOE32G", 4096},
 188         {"ATA     SAMSUNG MCCOE64G", 4096},
 189         {"ATA     SAMSUNG SSD PM80", 4096},
 190         /* Imported from Open Solaris */
 191         {"ATA     MARVELL SD88SA02", 4096},
 192         /* Advanced format Hard drives */
 193         {"ATA     Hitachi HDS5C303", 4096},
 194         {"ATA     SAMSUNG HD204UI ", 4096},
 195         {"ATA     ST2000DL004 HD20", 4096},
 196         {"ATA     WDC WD10EARS-00M", 4096},
 197         {"ATA     WDC WD10EARS-00S", 4096},
 198         {"ATA     WDC WD10EARS-00Z", 4096},
 199         {"ATA     WDC WD15EARS-00M", 4096},
 200         {"ATA     WDC WD15EARS-00S", 4096},
 201         {"ATA     WDC WD15EARS-00Z", 4096},
 202         {"ATA     WDC WD20EARS-00M", 4096},
 203         {"ATA     WDC WD20EARS-00S", 4096},
 204         {"ATA     WDC WD20EARS-00Z", 4096},
 205         /* Virtual disks: Assume zvols with default volblocksize */
 206 #if 0
 207         {"ATA     QEMU HARDDISK   ", 8192},
 208         {"IET     VIRTUAL-DISK    ", 8192},
 209         {"OI      COMSTAR         ", 8192},
 210         {"SUN     COMSTAR         ", 8192},
 211         {"NETAPP  LUN             ", 8192},
 212 #endif
 213 };
 214
 215 static const int vdev_disk_database_size =
 216         sizeof (vdev_disk_database) / sizeof (vdev_disk_database[0]);
 217
 218 #define INQ_REPLY_LEN   96
 219 #define INQ_CMD_LEN     6
 220
 221 static boolean_t
 222 check_sector_size_database(char *path, int *sector_size)
 223 {
 224         unsigned char inq_buff[INQ_REPLY_LEN];
 225         unsigned char sense_buffer[32];
 226         unsigned char inq_cmd_blk[INQ_CMD_LEN] =
 227             {INQUIRY, 0, 0, 0, INQ_REPLY_LEN, 0};
 228         sg_io_hdr_t io_hdr;
 229         int error;
 230         int fd;
 231         int i;
 232
 233         /* Prepare INQUIRY command */
 234         memset(&io_hdr, 0, sizeof (sg_io_hdr_t));
 235         io_hdr.interface_id = 'S';
 236         io_hdr.cmd_len = sizeof (inq_cmd_blk);
 237         io_hdr.mx_sb_len = sizeof (sense_buffer);
 238         io_hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 239         io_hdr.dxfer_len = INQ_REPLY_LEN;
 240         io_hdr.dxferp = inq_buff;
 241         io_hdr.cmdp = inq_cmd_blk;
 242         io_hdr.sbp = sense_buffer;
 243         io_hdr.timeout = 10;            /* 10 milliseconds is ample time */
 244
 245         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 246                 return (B_FALSE);
 247
 248         error = ioctl(fd, SG_IO, (unsigned long) &io_hdr);
 249
 250         (void) close(fd);
 251
 252         if (error < 0)
 253                 return (B_FALSE);
 254
 255         if ((io_hdr.info & SG_INFO_OK_MASK) != SG_INFO_OK)
 256                 return (B_FALSE);
 257
 258         for (i = 0; i < vdev_disk_database_size; i++) {
 259                 if (memcmp(inq_buff + 8, vdev_disk_database[i].id, 24))
 260                         continue;
 261
 262                 *sector_size = vdev_disk_database[i].sector_size;
 263                 return (B_TRUE);
 264         }
 265
 266         return (B_FALSE);
 267 }
 268
 269 /*PRINTFLIKE1*/
 270 static void
 271 vdev_error(const char *fmt, ...)
 272 {
 273         va_list ap;
 274
 275         if (!error_seen) {
 276                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
 277                 if (!is_force)
 278                         (void) fprintf(stderr, gettext("use '-f' to override "
 279                             "the following errors:\n"));
 280                 else
 281                         (void) fprintf(stderr, gettext("the following errors "
 282                             "must be manually repaired:\n"));
 283                 error_seen = B_TRUE;
 284         }
 285
 286         va_start(ap, fmt);
 287         (void) vfprintf(stderr, fmt, ap);
 288         va_end(ap);
 289 }
 290
 291 /*
 292  * Check that a file is valid.  All we can do in this case is check that it's
 293  * not in use by another pool, and not in use by swap.
 294  */
 295 static int
 296 check_file(const char *file, boolean_t force, boolean_t isspare)
 297 {
 298         char  *name;
 299         int fd;
 300         int ret = 0;
 301         pool_state_t state;
 302         boolean_t inuse;
 303
 304         if ((fd = open(file, O_RDONLY)) < 0)
 305                 return (0);
 306
 307         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
 308                 const char *desc;
 309
 310                 switch (state) {
 311                 case POOL_STATE_ACTIVE:
 312                         desc = gettext("active");
 313                         break;
 314
 315                 case POOL_STATE_EXPORTED:
 316                         desc = gettext("exported");
 317                         break;
 318
 319                 case POOL_STATE_POTENTIALLY_ACTIVE:
 320                         desc = gettext("potentially active");
 321                         break;
 322
 323                 default:
 324                         desc = gettext("unknown");
 325                         break;
 326                 }
 327
 328                 /*
 329                  * Allow hot spares to be shared between pools.
 330                  */
 331                 if (state == POOL_STATE_SPARE && isspare)
 332                         return (0);
 333
 334                 if (state == POOL_STATE_ACTIVE ||
 335                     state == POOL_STATE_SPARE || !force) {
 336                         switch (state) {
 337                         case POOL_STATE_SPARE:
 338                                 vdev_error(gettext("%s is reserved as a hot "
 339                                     "spare for pool %s\n"), file, name);
 340                                 break;
 341                         default:
 342                                 vdev_error(gettext("%s is part of %s pool "
 343                                     "'%s'\n"), file, desc, name);
 344                                 break;
 345                         }
 346                         ret = -1;
 347                 }
 348
 349                 free(name);
 350         }
 351
 352         (void) close(fd);
 353         return (ret);
 354 }
 355
 356 static void
 357 check_error(int err)
 358 {
 359         (void) fprintf(stderr, gettext("warning: device in use checking "
 360             "failed: %s\n"), strerror(err));
 361 }
 362
 363 static int
 364 check_slice(const char *path, blkid_cache cache, int force, boolean_t isspare)
 365 {
 366         int err;
 367 #ifdef HAVE_LIBBLKID
 368         char *value;
 369
 370         /* No valid type detected device is safe to use */
 371         value = blkid_get_tag_value(cache, "TYPE", path);
 372         if (value == NULL)
 373                 return (0);
 374
 375         /*
 376          * If libblkid detects a ZFS device, we check the device
 377          * using check_file() to see if it's safe.  The one safe
 378          * case is a spare device shared between multiple pools.
 379          */
 380         if (strcmp(value, "zfs") == 0) {
 381                 err = check_file(path, force, isspare);
 382         } else {
 383                 if (force) {
 384                         err = 0;
 385                 } else {
 386                         err = -1;
 387                         vdev_error(gettext("%s contains a filesystem of "
 388                             "type '%s'\n"), path, value);
 389                 }
 390         }
 391
 392         free(value);
 393 #else
 394         err = check_file(path, force, isspare);
 395 #endif /* HAVE_LIBBLKID */
 396
 397         return (err);
 398 }
 399
 400 /*
 401  * Validate a whole disk.  Iterate over all slices on the disk and make sure
 402  * that none is in use by calling check_slice().
 403  */
 404 static int
 405 check_disk(const char *path, blkid_cache cache, int force,
 406     boolean_t isspare, boolean_t iswholedisk)
 407 {
 408         struct dk_gpt *vtoc;
 409         char slice_path[MAXPATHLEN];
 410         int err = 0;
 411         int fd, i;
 412
 413         /* This is not a wholedisk we only check the given partition */
 414         if (!iswholedisk)
 415                 return (check_slice(path, cache, force, isspare));
 416
 417         /*
 418          * When the device is a whole disk try to read the efi partition
 419          * label.  If this is successful we safely check the all of the
 420          * partitions.  However, when it fails it may simply be because
 421          * the disk is partitioned via the MBR.  Since we currently can
 422          * not easily decode the MBR return a failure and prompt to the
 423          * user to use force option since we cannot check the partitions.
 424          */
 425         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
 426                 check_error(errno);
 427                 return (-1);
 428         }
 429
 430         if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
 431                 (void) close(fd);
 432
 433                 if (force) {
 434                         return (0);
 435                 } else {
 436                         vdev_error(gettext("%s does not contain an EFI "
 437                             "label but it may contain partition\n"
 438                             "information in the MBR.\n"), path);
 439                         return (-1);
 440                 }
 441         }
 442
 443         /*
 444          * The primary efi partition label is damaged however the secondary
 445          * label at the end of the device is intact.  Rather than use this
 446          * label we should play it safe and treat this as a non efi device.
 447          */
 448         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
 449                 efi_free(vtoc);
 450                 (void) close(fd);
 451
 452                 if (force) {
 453                         /* Partitions will no be created using the backup */
 454                         return (0);
 455                 } else {
 456                         vdev_error(gettext("%s contains a corrupt primary "
 457                             "EFI label.\n"), path);
 458                         return (-1);
 459                 }
 460         }
 461
 462         for (i = 0; i < vtoc->efi_nparts; i++) {
 463
 464                 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
 465                     uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
 466                         continue;
 467
 468                 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
 469                         (void) snprintf(slice_path, sizeof (slice_path),
 470                             "%s%s%d", path, "-part", i+1);
 471                 else
 472                         (void) snprintf(slice_path, sizeof (slice_path),
 473                             "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
 474                             "p" : "", i+1);
 475
 476                 err = check_slice(slice_path, cache, force, isspare);
 477                 if (err)
 478                         break;
 479         }
 480
 481         efi_free(vtoc);
 482         (void) close(fd);
 483
 484         return (err);
 485 }
 486
 487 static int
 488 check_device(const char *path, boolean_t force,
 489     boolean_t isspare, boolean_t iswholedisk)
 490 {
 491         static blkid_cache cache = NULL;
 492
 493 #ifdef HAVE_LIBBLKID
 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 #endif /* HAVE_LIBBLKID */
 513
 514         return (check_disk(path, cache, force, isspare, iswholedisk));
 515 }
 516
 517 /*
 518  * By "whole disk" we mean an entire physical disk (something we can
 519  * label, toggle the write cache on, etc.) as opposed to the full
 520  * capacity of a pseudo-device such as lofi or did.  We act as if we
 521  * are labeling the disk, which should be a pretty good test of whether
 522  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 523  * it isn't.
 524  */
 525 static boolean_t
 526 is_whole_disk(const char *path)
 527 {
 528         struct dk_gpt *label;
 529         int fd;
 530
 531         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 532                 return (B_FALSE);
 533         if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
 534                 (void) close(fd);
 535                 return (B_FALSE);
 536         }
 537         efi_free(label);
 538         (void) close(fd);
 539         return (B_TRUE);
 540 }
 541
 542 /*
 543  * This may be a shorthand device path or it could be total gibberish.
 544  * Check to see if it is a known device available in zfs_vdev_paths.
 545  * As part of this check, see if we've been given an entire disk
 546  * (minus the slice number).
 547  */
 548 static int
 549 is_shorthand_path(const char *arg, char *path,
 550     struct stat64 *statbuf, boolean_t *wholedisk)
 551 {
 552         int error;
 553
 554         error = zfs_resolve_shortname(arg, path, MAXPATHLEN);
 555         if (error == 0) {
 556                 *wholedisk = is_whole_disk(path);
 557                 if (*wholedisk || (stat64(path, statbuf) == 0))
 558                         return (0);
 559         }
 560
 561         strlcpy(path, arg, sizeof (path));
 562         memset(statbuf, 0, sizeof (*statbuf));
 563         *wholedisk = B_FALSE;
 564
 565         return (error);
 566 }
 567
 568 /*
 569  * Determine if the given path is a hot spare within the given configuration.
 570  * If no configuration is given we rely solely on the label.
 571  */
 572 static boolean_t
 573 is_spare(nvlist_t *config, const char *path)
 574 {
 575         int fd;
 576         pool_state_t state;
 577         char *name = NULL;
 578         nvlist_t *label;
 579         uint64_t guid, spareguid;
 580         nvlist_t *nvroot;
 581         nvlist_t **spares;
 582         uint_t i, nspares;
 583         boolean_t inuse;
 584
 585         if ((fd = open(path, O_RDONLY)) < 0)
 586                 return (B_FALSE);
 587
 588         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
 589             !inuse ||
 590             state != POOL_STATE_SPARE ||
 591             zpool_read_label(fd, &label) != 0) {
 592                 free(name);
 593                 (void) close(fd);
 594                 return (B_FALSE);
 595         }
 596         free(name);
 597         (void) close(fd);
 598
 599         if (config == NULL)
 600                 return (B_TRUE);
 601
 602         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
 603         nvlist_free(label);
 604
 605         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 606             &nvroot) == 0);
 607         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 608             &spares, &nspares) == 0) {
 609                 for (i = 0; i < nspares; i++) {
 610                         verify(nvlist_lookup_uint64(spares[i],
 611                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
 612                         if (spareguid == guid)
 613                                 return (B_TRUE);
 614                 }
 615         }
 616
 617         return (B_FALSE);
 618 }
 619
 620 /*
 621  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 622  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 623  * leaf vdev are:
 624  *
 625  *      /dev/xxx        Complete disk path
 626  *      /xxx            Full path to file
 627  *      xxx             Shorthand for <zfs_vdev_paths>/xxx
 628  */
 629 static nvlist_t *
 630 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
 631 {
 632         char path[MAXPATHLEN];
 633         struct stat64 statbuf;
 634         nvlist_t *vdev = NULL;
 635         char *type = NULL;
 636         boolean_t wholedisk = B_FALSE;
 637         uint64_t ashift = 0;
 638         int err;
 639
 640         /*
 641          * Determine what type of vdev this is, and put the full path into
 642          * 'path'.  We detect whether this is a device of file afterwards by
 643          * checking the st_mode of the file.
 644          */
 645         if (arg[0] == '/') {
 646                 /*
 647                  * Complete device or file path.  Exact type is determined by
 648                  * examining the file descriptor afterwards.  Symbolic links
 649                  * are resolved to their real paths for the is_whole_disk()
 650                  * and S_ISBLK/S_ISREG type checks.  However, we are careful
 651                  * to store the given path as ZPOOL_CONFIG_PATH to ensure we
 652                  * can leverage udev's persistent device labels.
 653                  */
 654                 if (realpath(arg, path) == NULL) {
 655                         (void) fprintf(stderr,
 656                             gettext("cannot resolve path '%s'\n"), arg);
 657                         return (NULL);
 658                 }
 659
 660                 wholedisk = is_whole_disk(path);
 661                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 662                         (void) fprintf(stderr,
 663                             gettext("cannot open '%s': %s\n"),
 664                             path, strerror(errno));
 665                         return (NULL);
 666                 }
 667
 668                 /* After is_whole_disk() check restore original passed path */
 669                 strlcpy(path, arg, MAXPATHLEN);
 670         } else {
 671                 err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
 672                 if (err != 0) {
 673                         /*
 674                          * If we got ENOENT, then the user gave us
 675                          * gibberish, so try to direct them with a
 676                          * reasonable error message.  Otherwise,
 677                          * regurgitate strerror() since it's the best we
 678                          * can do.
 679                          */
 680                         if (err == ENOENT) {
 681                                 (void) fprintf(stderr,
 682                                     gettext("cannot open '%s': no such "
 683                                     "device in %s\n"), arg, DISK_ROOT);
 684                                 (void) fprintf(stderr,
 685                                     gettext("must be a full path or "
 686                                     "shorthand device name\n"));
 687                                 return (NULL);
 688                         } else {
 689                                 (void) fprintf(stderr,
 690                                     gettext("cannot open '%s': %s\n"),
 691                                     path, strerror(errno));
 692                                 return (NULL);
 693                         }
 694                 }
 695         }
 696
 697         /*
 698          * Determine whether this is a device or a file.
 699          */
 700         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 701                 type = VDEV_TYPE_DISK;
 702         } else if (S_ISREG(statbuf.st_mode)) {
 703                 type = VDEV_TYPE_FILE;
 704         } else {
 705                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 706                     "block device or regular file\n"), path);
 707                 return (NULL);
 708         }
 709
 710         /*
 711          * Finally, we have the complete device or file, and we know that it is
 712          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 713          * vdevs have a 'path' element, and devices also have a 'devid' element.
 714          */
 715         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 716         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 717         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 718         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 719         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 720                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 721                     (uint64_t)wholedisk) == 0);
 722
 723         /*
 724          * Override defaults if custom properties are provided.
 725          */
 726         if (props != NULL) {
 727                 char *value = NULL;
 728
 729                 if (nvlist_lookup_string(props,
 730                     zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
 731                         zfs_nicestrtonum(NULL, value, &ashift);
 732         }
 733
 734         /*
 735          * If the device is known to incorrectly report its physical sector
 736          * size explicitly provide the known correct value.
 737          */
 738         if (ashift == 0) {
 739                 int sector_size;
 740
 741                 if (check_sector_size_database(path, &sector_size) == B_TRUE)
 742                         ashift = highbit(sector_size) - 1;
 743         }
 744
 745         if (ashift > 0)
 746                 nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
 747
 748         return (vdev);
 749 }
 750
 751 /*
 752  * Go through and verify the replication level of the pool is consistent.
 753  * Performs the following checks:
 754  *
 755  *      For the new spec, verifies that devices in mirrors and raidz are the
 756  *      same size.
 757  *
 758  *      If the current configuration already has inconsistent replication
 759  *      levels, ignore any other potential problems in the new spec.
 760  *
 761  *      Otherwise, make sure that the current spec (if there is one) and the new
 762  *      spec have consistent replication levels.
 763  */
 764 typedef struct replication_level {
 765         char *zprl_type;
 766         uint64_t zprl_children;
 767         uint64_t zprl_parity;
 768 } replication_level_t;
 769
 770 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 771
 772 /*
 773  * Given a list of toplevel vdevs, return the current replication level.  If
 774  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 775  * an error message will be displayed for each self-inconsistent vdev.
 776  */
 777 static replication_level_t *
 778 get_replication(nvlist_t *nvroot, boolean_t fatal)
 779 {
 780         nvlist_t **top;
 781         uint_t t, toplevels;
 782         nvlist_t **child;
 783         uint_t c, children;
 784         nvlist_t *nv;
 785         char *type;
 786         replication_level_t lastrep = { 0 }, rep, *ret;
 787         boolean_t dontreport;
 788
 789         ret = safe_malloc(sizeof (replication_level_t));
 790
 791         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 792             &top, &toplevels) == 0);
 793
 794         lastrep.zprl_type = NULL;
 795         for (t = 0; t < toplevels; t++) {
 796                 uint64_t is_log = B_FALSE;
 797
 798                 nv = top[t];
 799
 800                 /*
 801                  * For separate logs we ignore the top level vdev replication
 802                  * constraints.
 803                  */
 804                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 805                 if (is_log)
 806                         continue;
 807
 808                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 809                     &type) == 0);
 810                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 811                     &child, &children) != 0) {
 812                         /*
 813                          * This is a 'file' or 'disk' vdev.
 814                          */
 815                         rep.zprl_type = type;
 816                         rep.zprl_children = 1;
 817                         rep.zprl_parity = 0;
 818                 } else {
 819                         uint64_t vdev_size;
 820
 821                         /*
 822                          * This is a mirror or RAID-Z vdev.  Go through and make
 823                          * sure the contents are all the same (files vs. disks),
 824                          * keeping track of the number of elements in the
 825                          * process.
 826                          *
 827                          * We also check that the size of each vdev (if it can
 828                          * be determined) is the same.
 829                          */
 830                         rep.zprl_type = type;
 831                         rep.zprl_children = 0;
 832
 833                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 834                                 verify(nvlist_lookup_uint64(nv,
 835                                     ZPOOL_CONFIG_NPARITY,
 836                                     &rep.zprl_parity) == 0);
 837                                 assert(rep.zprl_parity != 0);
 838                         } else {
 839                                 rep.zprl_parity = 0;
 840                         }
 841
 842                         /*
 843                          * The 'dontreport' variable indicates that we've
 844                          * already reported an error for this spec, so don't
 845                          * bother doing it again.
 846                          */
 847                         type = NULL;
 848                         dontreport = 0;
 849                         vdev_size = -1ULL;
 850                         for (c = 0; c < children; c++) {
 851                                 nvlist_t *cnv = child[c];
 852                                 char *path;
 853                                 struct stat64 statbuf;
 854                                 uint64_t size = -1ULL;
 855                                 char *childtype;
 856                                 int fd, err;
 857
 858                                 rep.zprl_children++;
 859
 860                                 verify(nvlist_lookup_string(cnv,
 861                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 862
 863                                 /*
 864                                  * If this is a replacing or spare vdev, then
 865                                  * get the real first child of the vdev.
 866                                  */
 867                                 if (strcmp(childtype,
 868                                     VDEV_TYPE_REPLACING) == 0 ||
 869                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 870                                         nvlist_t **rchild;
 871                                         uint_t rchildren;
 872
 873                                         verify(nvlist_lookup_nvlist_array(cnv,
 874                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 875                                             &rchildren) == 0);
 876                                         assert(rchildren == 2);
 877                                         cnv = rchild[0];
 878
 879                                         verify(nvlist_lookup_string(cnv,
 880                                             ZPOOL_CONFIG_TYPE,
 881                                             &childtype) == 0);
 882                                 }
 883
 884                                 verify(nvlist_lookup_string(cnv,
 885                                     ZPOOL_CONFIG_PATH, &path) == 0);
 886
 887                                 /*
 888                                  * If we have a raidz/mirror that combines disks
 889                                  * with files, report it as an error.
 890                                  */
 891                                 if (!dontreport && type != NULL &&
 892                                     strcmp(type, childtype) != 0) {
 893                                         if (ret != NULL)
 894                                                 free(ret);
 895                                         ret = NULL;
 896                                         if (fatal)
 897                                                 vdev_error(gettext(
 898                                                     "mismatched replication "
 899                                                     "level: %s contains both "
 900                                                     "files and devices\n"),
 901                                                     rep.zprl_type);
 902                                         else
 903                                                 return (NULL);
 904                                         dontreport = B_TRUE;
 905                                 }
 906
 907                                 /*
 908                                  * According to stat(2), the value of 'st_size'
 909                                  * is undefined for block devices and character
 910                                  * devices.  But there is no effective way to
 911                                  * determine the real size in userland.
 912                                  *
 913                                  * Instead, we'll take advantage of an
 914                                  * implementation detail of spec_size().  If the
 915                                  * device is currently open, then we (should)
 916                                  * return a valid size.
 917                                  *
 918                                  * If we still don't get a valid size (indicated
 919                                  * by a size of 0 or MAXOFFSET_T), then ignore
 920                                  * this device altogether.
 921                                  */
 922                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 923                                         err = fstat64(fd, &statbuf);
 924                                         (void) close(fd);
 925                                 } else {
 926                                         err = stat64(path, &statbuf);
 927                                 }
 928
 929                                 if (err != 0 ||
 930                                     statbuf.st_size == 0 ||
 931                                     statbuf.st_size == MAXOFFSET_T)
 932                                         continue;
 933
 934                                 size = statbuf.st_size;
 935
 936                                 /*
 937                                  * Also make sure that devices and
 938                                  * slices have a consistent size.  If
 939                                  * they differ by a significant amount
 940                                  * (~16MB) then report an error.
 941                                  */
 942                                 if (!dontreport &&
 943                                     (vdev_size != -1ULL &&
 944                                     (labs(size - vdev_size) >
 945                                     ZPOOL_FUZZ))) {
 946                                         if (ret != NULL)
 947                                                 free(ret);
 948                                         ret = NULL;
 949                                         if (fatal)
 950                                                 vdev_error(gettext(
 951                                                     "%s contains devices of "
 952                                                     "different sizes\n"),
 953                                                     rep.zprl_type);
 954                                         else
 955                                                 return (NULL);
 956                                         dontreport = B_TRUE;
 957                                 }
 958
 959                                 type = childtype;
 960                                 vdev_size = size;
 961                         }
 962                 }
 963
 964                 /*
 965                  * At this point, we have the replication of the last toplevel
 966                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 967                  * different.
 968                  */
 969                 if (lastrep.zprl_type != NULL) {
 970                         if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
 971                                 if (ret != NULL)
 972                                         free(ret);
 973                                 ret = NULL;
 974                                 if (fatal)
 975                                         vdev_error(gettext(
 976                                             "mismatched replication level: "
 977                                             "both %s and %s vdevs are "
 978                                             "present\n"),
 979                                             lastrep.zprl_type, rep.zprl_type);
 980                                 else
 981                                         return (NULL);
 982                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 983                                 if (ret)
 984                                         free(ret);
 985                                 ret = NULL;
 986                                 if (fatal)
 987                                         vdev_error(gettext(
 988                                             "mismatched replication level: "
 989                                             "both %llu and %llu device parity "
 990                                             "%s vdevs are present\n"),
 991                                             lastrep.zprl_parity,
 992                                             rep.zprl_parity,
 993                                             rep.zprl_type);
 994                                 else
 995                                         return (NULL);
 996                         } else if (lastrep.zprl_children != rep.zprl_children) {
 997                                 if (ret)
 998                                         free(ret);
 999                                 ret = NULL;
1000                                 if (fatal)
1001                                         vdev_error(gettext(
1002                                             "mismatched replication level: "
1003                                             "both %llu-way and %llu-way %s "
1004                                             "vdevs are present\n"),
1005                                             lastrep.zprl_children,
1006                                             rep.zprl_children,
1007                                             rep.zprl_type);
1008                                 else
1009                                         return (NULL);
1010                         }
1011                 }
1012                 lastrep = rep;
1013         }
1014
1015         if (ret != NULL)
1016                 *ret = rep;
1017
1018         return (ret);
1019 }
1020
1021 /*
1022  * Check the replication level of the vdev spec against the current pool.  Calls
1023  * get_replication() to make sure the new spec is self-consistent.  If the pool
1024  * has a consistent replication level, then we ignore any errors.  Otherwise,
1025  * report any difference between the two.
1026  */
1027 static int
1028 check_replication(nvlist_t *config, nvlist_t *newroot)
1029 {
1030         nvlist_t **child;
1031         uint_t  children;
1032         replication_level_t *current = NULL, *new;
1033         int ret;
1034
1035         /*
1036          * If we have a current pool configuration, check to see if it's
1037          * self-consistent.  If not, simply return success.
1038          */
1039         if (config != NULL) {
1040                 nvlist_t *nvroot;
1041
1042                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1043                     &nvroot) == 0);
1044                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1045                         return (0);
1046         }
1047         /*
1048          * for spares there may be no children, and therefore no
1049          * replication level to check
1050          */
1051         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1052             &child, &children) != 0) || (children == 0)) {
1053                 free(current);
1054                 return (0);
1055         }
1056
1057         /*
1058          * If all we have is logs then there's no replication level to check.
1059          */
1060         if (num_logs(newroot) == children) {
1061                 free(current);
1062                 return (0);
1063         }
1064
1065         /*
1066          * Get the replication level of the new vdev spec, reporting any
1067          * inconsistencies found.
1068          */
1069         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1070                 free(current);
1071                 return (-1);
1072         }
1073
1074         /*
1075          * Check to see if the new vdev spec matches the replication level of
1076          * the current pool.
1077          */
1078         ret = 0;
1079         if (current != NULL) {
1080                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1081                         vdev_error(gettext(
1082                             "mismatched replication level: pool uses %s "
1083                             "and new vdev is %s\n"),
1084                             current->zprl_type, new->zprl_type);
1085                         ret = -1;
1086                 } else if (current->zprl_parity != new->zprl_parity) {
1087                         vdev_error(gettext(
1088                             "mismatched replication level: pool uses %llu "
1089                             "device parity and new vdev uses %llu\n"),
1090                             current->zprl_parity, new->zprl_parity);
1091                         ret = -1;
1092                 } else if (current->zprl_children != new->zprl_children) {
1093                         vdev_error(gettext(
1094                             "mismatched replication level: pool uses %llu-way "
1095                             "%s and new vdev uses %llu-way %s\n"),
1096                             current->zprl_children, current->zprl_type,
1097                             new->zprl_children, new->zprl_type);
1098                         ret = -1;
1099                 }
1100         }
1101
1102         free(new);
1103         if (current != NULL)
1104                 free(current);
1105
1106         return (ret);
1107 }
1108
1109 static int
1110 zero_label(char *path)
1111 {
1112         const int size = 4096;
1113         char buf[size];
1114         int err, fd;
1115
1116         if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1117                 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1118                     path, strerror(errno));
1119                 return (-1);
1120         }
1121
1122         memset(buf, 0, size);
1123         err = write(fd, buf, size);
1124         (void) fdatasync(fd);
1125         (void) close(fd);
1126
1127         if (err == -1) {
1128                 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1129                     "of '%s': %s\n"), size, path, strerror(errno));
1130                 return (-1);
1131         }
1132
1133         if (err != size) {
1134                 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1135                     "of '%s'\n"), err, size, path);
1136                 return (-1);
1137         }
1138
1139         return (0);
1140 }
1141
1142 /*
1143  * Go through and find any whole disks in the vdev specification, labelling them
1144  * as appropriate.  When constructing the vdev spec, we were unable to open this
1145  * device in order to provide a devid.  Now that we have labelled the disk and
1146  * know that slice 0 is valid, we can construct the devid now.
1147  *
1148  * If the disk was already labeled with an EFI label, we will have gotten the
1149  * devid already (because we were able to open the whole disk).  Otherwise, we
1150  * need to get the devid after we label the disk.
1151  */
1152 static int
1153 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1154 {
1155         nvlist_t **child;
1156         uint_t c, children;
1157         char *type, *path;
1158         char devpath[MAXPATHLEN];
1159         char udevpath[MAXPATHLEN];
1160         uint64_t wholedisk;
1161         struct stat64 statbuf;
1162         int is_exclusive = 0;
1163         int fd;
1164         int ret;
1165
1166         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1167
1168         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1169             &child, &children) != 0) {
1170
1171                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1172                         return (0);
1173
1174                 /*
1175                  * We have a disk device.  If this is a whole disk write
1176                  * out the efi partition table, otherwise write zero's to
1177                  * the first 4k of the partition.  This is to ensure that
1178                  * libblkid will not misidentify the partition due to a
1179                  * magic value left by the previous filesystem.
1180                  */
1181                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1182                 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1183                     &wholedisk));
1184
1185                 if (!wholedisk) {
1186                         (void) zero_label(path);
1187                         return (0);
1188                 }
1189
1190                 if (realpath(path, devpath) == NULL) {
1191                         ret = errno;
1192                         (void) fprintf(stderr,
1193                             gettext("cannot resolve path '%s'\n"), path);
1194                         return (ret);
1195                 }
1196
1197                 /*
1198                  * Remove any previously existing symlink from a udev path to
1199                  * the device before labeling the disk.  This makes
1200                  * zpool_label_disk_wait() truly wait for the new link to show
1201                  * up instead of returning if it finds an old link still in
1202                  * place.  Otherwise there is a window between when udev
1203                  * deletes and recreates the link during which access attempts
1204                  * will fail with ENOENT.
1205                  */
1206                 strncpy(udevpath, path, MAXPATHLEN);
1207                 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1208
1209                 fd = open(devpath, O_RDWR|O_EXCL);
1210                 if (fd == -1) {
1211                         if (errno == EBUSY)
1212                                 is_exclusive = 1;
1213                 } else {
1214                         (void) close(fd);
1215                 }
1216
1217                 /*
1218                  * If the partition exists, contains a valid spare label,
1219                  * and is opened exclusively there is no need to partition
1220                  * it.  Hot spares have already been partitioned and are
1221                  * held open exclusively by the kernel as a safety measure.
1222                  *
1223                  * If the provided path is for a /dev/disk/ device its
1224                  * symbolic link will be removed, partition table created,
1225                  * and then block until udev creates the new link.
1226                  */
1227                 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1228                         ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1229                         if (ret == 0) {
1230                                 ret = lstat64(udevpath, &statbuf);
1231                                 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1232                                         (void) unlink(udevpath);
1233                         }
1234
1235                         if (zpool_label_disk(g_zfs, zhp,
1236                             strrchr(devpath, '/') + 1) == -1)
1237                                 return (-1);
1238
1239                         ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1240                         if (ret) {
1241                                 (void) fprintf(stderr, gettext("cannot "
1242                                     "resolve path '%s': %d\n"), udevpath, ret);
1243                                 return (-1);
1244                         }
1245
1246                         (void) zero_label(udevpath);
1247                 }
1248
1249                 /*
1250                  * Update the path to refer to the partition.  The presence of
1251                  * the 'whole_disk' field indicates to the CLI that we should
1252                  * chop off the partition number when displaying the device in
1253                  * future output.
1254                  */
1255                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1256
1257                 return (0);
1258         }
1259
1260         for (c = 0; c < children; c++)
1261                 if ((ret = make_disks(zhp, child[c])) != 0)
1262                         return (ret);
1263
1264         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1265             &child, &children) == 0)
1266                 for (c = 0; c < children; c++)
1267                         if ((ret = make_disks(zhp, child[c])) != 0)
1268                                 return (ret);
1269
1270         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1271             &child, &children) == 0)
1272                 for (c = 0; c < children; c++)
1273                         if ((ret = make_disks(zhp, child[c])) != 0)
1274                                 return (ret);
1275
1276         return (0);
1277 }
1278
1279 /*
1280  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1281  * the majority of this task.
1282  */
1283 static int
1284 check_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1285     boolean_t replacing, boolean_t isspare)
1286 {
1287         nvlist_t **child;
1288         uint_t c, children;
1289         char *type, *path;
1290         int ret = 0;
1291         char buf[MAXPATHLEN];
1292         uint64_t wholedisk = 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 (0);
1319                 }
1320
1321                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1322                         ret = check_device(path, force, isspare, wholedisk);
1323
1324                 if (strcmp(type, VDEV_TYPE_FILE) == 0)
1325                         ret = check_file(path, force, isspare);
1326
1327                 return (ret);
1328         }
1329
1330         for (c = 0; c < children; c++)
1331                 if ((ret = check_in_use(config, child[c], force,
1332                     replacing, B_FALSE)) != 0)
1333                         return (ret);
1334
1335         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1336             &child, &children) == 0)
1337                 for (c = 0; c < children; c++)
1338                         if ((ret = check_in_use(config, child[c], force,
1339                             replacing, B_TRUE)) != 0)
1340                                 return (ret);
1341
1342         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1343             &child, &children) == 0)
1344                 for (c = 0; c < children; c++)
1345                         if ((ret = check_in_use(config, child[c], force,
1346                             replacing, B_FALSE)) != 0)
1347                                 return (ret);
1348
1349         return (0);
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                 return (NULL);
1695
1696         /*
1697          * Validate each device to make sure that its not shared with another
1698          * subsystem.  We do this even if 'force' is set, because there are some
1699          * uses (such as a dedicated dump device) that even '-f' cannot
1700          * override.
1701          */
1702         if (check_in_use(poolconfig, newroot, force, replacing, B_FALSE) != 0) {
1703                 nvlist_free(newroot);
1704                 return (NULL);
1705         }
1706
1707         /*
1708          * Check the replication level of the given vdevs and report any errors
1709          * found.  We include the existing pool spec, if any, as we need to
1710          * catch changes against the existing replication level.
1711          */
1712         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1713                 nvlist_free(newroot);
1714                 return (NULL);
1715         }
1716
1717         /*
1718          * Run through the vdev specification and label any whole disks found.
1719          */
1720         if (!dryrun && make_disks(zhp, newroot) != 0) {
1721                 nvlist_free(newroot);
1722                 return (NULL);
1723         }
1724
1725         return (newroot);
1726 }