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 that a disk including all partitions are safe to use.
 403  *
 404  * For EFI labeled disks this can done relatively easily with the libefi
 405  * library.  The partition numbers are extracted from the label and used
 406  * to generate the expected /dev/ paths.  Each partition can then be
 407  * checked for conflicts.
 408  *
 409  * For non-EFI labeled disks (MBR/EBR/etc) the same process is possible
 410  * but due to the lack of a readily available libraries this scanning is
 411  * not implemented.  Instead only the device path as given is checked.
 412  */
 413 static int
 414 check_disk(const char *path, blkid_cache cache, int force,
 415     boolean_t isspare, boolean_t iswholedisk)
 416 {
 417         struct dk_gpt *vtoc;
 418         char slice_path[MAXPATHLEN];
 419         int err = 0;
 420         int fd, i;
 421
 422         if (!iswholedisk)
 423                 return (check_slice(path, cache, force, isspare));
 424
 425         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0) {
 426                 check_error(errno);
 427                 return (-1);
 428         }
 429
 430         /*
 431          * Expected to fail for non-EFI labled disks.  Just check the device
 432          * as given and do not attempt to detect and scan partitions.
 433          */
 434         err = efi_alloc_and_read(fd, &vtoc);
 435         if (err) {
 436                 (void) close(fd);
 437                 return (check_slice(path, cache, force, isspare));
 438         }
 439
 440         /*
 441          * The primary efi partition label is damaged however the secondary
 442          * label at the end of the device is intact.  Rather than use this
 443          * label we should play it safe and treat this as a non efi device.
 444          */
 445         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
 446                 efi_free(vtoc);
 447                 (void) close(fd);
 448
 449                 if (force) {
 450                         /* Partitions will no be created using the backup */
 451                         return (0);
 452                 } else {
 453                         vdev_error(gettext("%s contains a corrupt primary "
 454                             "EFI label.\n"), path);
 455                         return (-1);
 456                 }
 457         }
 458
 459         for (i = 0; i < vtoc->efi_nparts; i++) {
 460
 461                 if (vtoc->efi_parts[i].p_tag == V_UNASSIGNED ||
 462                     uuid_is_null((uchar_t *)&vtoc->efi_parts[i].p_guid))
 463                         continue;
 464
 465                 if (strncmp(path, UDISK_ROOT, strlen(UDISK_ROOT)) == 0)
 466                         (void) snprintf(slice_path, sizeof (slice_path),
 467                             "%s%s%d", path, "-part", i+1);
 468                 else
 469                         (void) snprintf(slice_path, sizeof (slice_path),
 470                             "%s%s%d", path, isdigit(path[strlen(path)-1]) ?
 471                             "p" : "", i+1);
 472
 473                 err = check_slice(slice_path, cache, force, isspare);
 474                 if (err)
 475                         break;
 476         }
 477
 478         efi_free(vtoc);
 479         (void) close(fd);
 480
 481         return (err);
 482 }
 483
 484 static int
 485 check_device(const char *path, boolean_t force,
 486     boolean_t isspare, boolean_t iswholedisk)
 487 {
 488         static blkid_cache cache = NULL;
 489
 490         /*
 491          * There is no easy way to add a correct blkid_put_cache() call,
 492          * memory will be reclaimed when the command exits.
 493          */
 494         if (cache == NULL) {
 495                 int err;
 496
 497                 if ((err = blkid_get_cache(&cache, NULL)) != 0) {
 498                         check_error(err);
 499                         return (-1);
 500                 }
 501
 502                 if ((err = blkid_probe_all(cache)) != 0) {
 503                         blkid_put_cache(cache);
 504                         check_error(err);
 505                         return (-1);
 506                 }
 507         }
 508
 509         return (check_disk(path, cache, force, isspare, iswholedisk));
 510 }
 511
 512 /*
 513  * By "whole disk" we mean an entire physical disk (something we can
 514  * label, toggle the write cache on, etc.) as opposed to the full
 515  * capacity of a pseudo-device such as lofi or did.  We act as if we
 516  * are labeling the disk, which should be a pretty good test of whether
 517  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 518  * it isn't.
 519  */
 520 static boolean_t
 521 is_whole_disk(const char *path)
 522 {
 523         struct dk_gpt *label;
 524         int fd;
 525
 526         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
 527                 return (B_FALSE);
 528         if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
 529                 (void) close(fd);
 530                 return (B_FALSE);
 531         }
 532         efi_free(label);
 533         (void) close(fd);
 534         return (B_TRUE);
 535 }
 536
 537 /*
 538  * This may be a shorthand device path or it could be total gibberish.
 539  * Check to see if it is a known device available in zfs_vdev_paths.
 540  * As part of this check, see if we've been given an entire disk
 541  * (minus the slice number).
 542  */
 543 static int
 544 is_shorthand_path(const char *arg, char *path,
 545     struct stat64 *statbuf, boolean_t *wholedisk)
 546 {
 547         int error;
 548
 549         error = zfs_resolve_shortname(arg, path, MAXPATHLEN);
 550         if (error == 0) {
 551                 *wholedisk = is_whole_disk(path);
 552                 if (*wholedisk || (stat64(path, statbuf) == 0))
 553                         return (0);
 554         }
 555
 556         strlcpy(path, arg, sizeof (path));
 557         memset(statbuf, 0, sizeof (*statbuf));
 558         *wholedisk = B_FALSE;
 559
 560         return (error);
 561 }
 562
 563 /*
 564  * Determine if the given path is a hot spare within the given configuration.
 565  * If no configuration is given we rely solely on the label.
 566  */
 567 static boolean_t
 568 is_spare(nvlist_t *config, const char *path)
 569 {
 570         int fd;
 571         pool_state_t state;
 572         char *name = NULL;
 573         nvlist_t *label;
 574         uint64_t guid, spareguid;
 575         nvlist_t *nvroot;
 576         nvlist_t **spares;
 577         uint_t i, nspares;
 578         boolean_t inuse;
 579
 580         if ((fd = open(path, O_RDONLY)) < 0)
 581                 return (B_FALSE);
 582
 583         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
 584             !inuse ||
 585             state != POOL_STATE_SPARE ||
 586             zpool_read_label(fd, &label, NULL) != 0) {
 587                 free(name);
 588                 (void) close(fd);
 589                 return (B_FALSE);
 590         }
 591         free(name);
 592         (void) close(fd);
 593
 594         if (config == NULL)
 595                 return (B_TRUE);
 596
 597         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
 598         nvlist_free(label);
 599
 600         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 601             &nvroot) == 0);
 602         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 603             &spares, &nspares) == 0) {
 604                 for (i = 0; i < nspares; i++) {
 605                         verify(nvlist_lookup_uint64(spares[i],
 606                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
 607                         if (spareguid == guid)
 608                                 return (B_TRUE);
 609                 }
 610         }
 611
 612         return (B_FALSE);
 613 }
 614
 615 /*
 616  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 617  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 618  * leaf vdev are:
 619  *
 620  *      /dev/xxx        Complete disk path
 621  *      /xxx            Full path to file
 622  *      xxx             Shorthand for <zfs_vdev_paths>/xxx
 623  */
 624 static nvlist_t *
 625 make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
 626 {
 627         char path[MAXPATHLEN];
 628         struct stat64 statbuf;
 629         nvlist_t *vdev = NULL;
 630         char *type = NULL;
 631         boolean_t wholedisk = B_FALSE;
 632         uint64_t ashift = 0;
 633         int err;
 634
 635         /*
 636          * Determine what type of vdev this is, and put the full path into
 637          * 'path'.  We detect whether this is a device of file afterwards by
 638          * checking the st_mode of the file.
 639          */
 640         if (arg[0] == '/') {
 641                 /*
 642                  * Complete device or file path.  Exact type is determined by
 643                  * examining the file descriptor afterwards.  Symbolic links
 644                  * are resolved to their real paths for the is_whole_disk()
 645                  * and S_ISBLK/S_ISREG type checks.  However, we are careful
 646                  * to store the given path as ZPOOL_CONFIG_PATH to ensure we
 647                  * can leverage udev's persistent device labels.
 648                  */
 649                 if (realpath(arg, path) == NULL) {
 650                         (void) fprintf(stderr,
 651                             gettext("cannot resolve path '%s'\n"), arg);
 652                         return (NULL);
 653                 }
 654
 655                 wholedisk = is_whole_disk(path);
 656                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 657                         (void) fprintf(stderr,
 658                             gettext("cannot open '%s': %s\n"),
 659                             path, strerror(errno));
 660                         return (NULL);
 661                 }
 662
 663                 /* After is_whole_disk() check restore original passed path */
 664                 strlcpy(path, arg, MAXPATHLEN);
 665         } else {
 666                 err = is_shorthand_path(arg, path, &statbuf, &wholedisk);
 667                 if (err != 0) {
 668                         /*
 669                          * If we got ENOENT, then the user gave us
 670                          * gibberish, so try to direct them with a
 671                          * reasonable error message.  Otherwise,
 672                          * regurgitate strerror() since it's the best we
 673                          * can do.
 674                          */
 675                         if (err == ENOENT) {
 676                                 (void) fprintf(stderr,
 677                                     gettext("cannot open '%s': no such "
 678                                     "device in %s\n"), arg, DISK_ROOT);
 679                                 (void) fprintf(stderr,
 680                                     gettext("must be a full path or "
 681                                     "shorthand device name\n"));
 682                                 return (NULL);
 683                         } else {
 684                                 (void) fprintf(stderr,
 685                                     gettext("cannot open '%s': %s\n"),
 686                                     path, strerror(errno));
 687                                 return (NULL);
 688                         }
 689                 }
 690         }
 691
 692         /*
 693          * Determine whether this is a device or a file.
 694          */
 695         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 696                 type = VDEV_TYPE_DISK;
 697         } else if (S_ISREG(statbuf.st_mode)) {
 698                 type = VDEV_TYPE_FILE;
 699         } else {
 700                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 701                     "block device or regular file\n"), path);
 702                 return (NULL);
 703         }
 704
 705         /*
 706          * Finally, we have the complete device or file, and we know that it is
 707          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 708          * vdevs have a 'path' element, and devices also have a 'devid' element.
 709          */
 710         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 711         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 712         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 713         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 714         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 715                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 716                     (uint64_t)wholedisk) == 0);
 717
 718         /*
 719          * Override defaults if custom properties are provided.
 720          */
 721         if (props != NULL) {
 722                 char *value = NULL;
 723
 724                 if (nvlist_lookup_string(props,
 725                     zpool_prop_to_name(ZPOOL_PROP_ASHIFT), &value) == 0)
 726                         zfs_nicestrtonum(NULL, value, &ashift);
 727         }
 728
 729         /*
 730          * If the device is known to incorrectly report its physical sector
 731          * size explicitly provide the known correct value.
 732          */
 733         if (ashift == 0) {
 734                 int sector_size;
 735
 736                 if (check_sector_size_database(path, &sector_size) == B_TRUE)
 737                         ashift = highbit64(sector_size) - 1;
 738         }
 739
 740         if (ashift > 0)
 741                 nvlist_add_uint64(vdev, ZPOOL_CONFIG_ASHIFT, ashift);
 742
 743         return (vdev);
 744 }
 745
 746 /*
 747  * Go through and verify the replication level of the pool is consistent.
 748  * Performs the following checks:
 749  *
 750  *      For the new spec, verifies that devices in mirrors and raidz are the
 751  *      same size.
 752  *
 753  *      If the current configuration already has inconsistent replication
 754  *      levels, ignore any other potential problems in the new spec.
 755  *
 756  *      Otherwise, make sure that the current spec (if there is one) and the new
 757  *      spec have consistent replication levels.
 758  */
 759 typedef struct replication_level {
 760         char *zprl_type;
 761         uint64_t zprl_children;
 762         uint64_t zprl_parity;
 763 } replication_level_t;
 764
 765 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 766
 767 /*
 768  * Given a list of toplevel vdevs, return the current replication level.  If
 769  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 770  * an error message will be displayed for each self-inconsistent vdev.
 771  */
 772 static replication_level_t *
 773 get_replication(nvlist_t *nvroot, boolean_t fatal)
 774 {
 775         nvlist_t **top;
 776         uint_t t, toplevels;
 777         nvlist_t **child;
 778         uint_t c, children;
 779         nvlist_t *nv;
 780         char *type;
 781         replication_level_t lastrep = { 0 }, rep, *ret;
 782         boolean_t dontreport;
 783
 784         ret = safe_malloc(sizeof (replication_level_t));
 785
 786         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 787             &top, &toplevels) == 0);
 788
 789         lastrep.zprl_type = NULL;
 790         for (t = 0; t < toplevels; t++) {
 791                 uint64_t is_log = B_FALSE;
 792
 793                 nv = top[t];
 794
 795                 /*
 796                  * For separate logs we ignore the top level vdev replication
 797                  * constraints.
 798                  */
 799                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 800                 if (is_log)
 801                         continue;
 802
 803                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 804                     &type) == 0);
 805                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 806                     &child, &children) != 0) {
 807                         /*
 808                          * This is a 'file' or 'disk' vdev.
 809                          */
 810                         rep.zprl_type = type;
 811                         rep.zprl_children = 1;
 812                         rep.zprl_parity = 0;
 813                 } else {
 814                         uint64_t vdev_size;
 815
 816                         /*
 817                          * This is a mirror or RAID-Z vdev.  Go through and make
 818                          * sure the contents are all the same (files vs. disks),
 819                          * keeping track of the number of elements in the
 820                          * process.
 821                          *
 822                          * We also check that the size of each vdev (if it can
 823                          * be determined) is the same.
 824                          */
 825                         rep.zprl_type = type;
 826                         rep.zprl_children = 0;
 827
 828                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 829                                 verify(nvlist_lookup_uint64(nv,
 830                                     ZPOOL_CONFIG_NPARITY,
 831                                     &rep.zprl_parity) == 0);
 832                                 assert(rep.zprl_parity != 0);
 833                         } else {
 834                                 rep.zprl_parity = 0;
 835                         }
 836
 837                         /*
 838                          * The 'dontreport' variable indicates that we've
 839                          * already reported an error for this spec, so don't
 840                          * bother doing it again.
 841                          */
 842                         type = NULL;
 843                         dontreport = 0;
 844                         vdev_size = -1ULL;
 845                         for (c = 0; c < children; c++) {
 846                                 nvlist_t *cnv = child[c];
 847                                 char *path;
 848                                 struct stat64 statbuf;
 849                                 uint64_t size = -1ULL;
 850                                 char *childtype;
 851                                 int fd, err;
 852
 853                                 rep.zprl_children++;
 854
 855                                 verify(nvlist_lookup_string(cnv,
 856                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 857
 858                                 /*
 859                                  * If this is a replacing or spare vdev, then
 860                                  * get the real first child of the vdev.
 861                                  */
 862                                 if (strcmp(childtype,
 863                                     VDEV_TYPE_REPLACING) == 0 ||
 864                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 865                                         nvlist_t **rchild;
 866                                         uint_t rchildren;
 867
 868                                         verify(nvlist_lookup_nvlist_array(cnv,
 869                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 870                                             &rchildren) == 0);
 871                                         assert(rchildren == 2);
 872                                         cnv = rchild[0];
 873
 874                                         verify(nvlist_lookup_string(cnv,
 875                                             ZPOOL_CONFIG_TYPE,
 876                                             &childtype) == 0);
 877                                 }
 878
 879                                 verify(nvlist_lookup_string(cnv,
 880                                     ZPOOL_CONFIG_PATH, &path) == 0);
 881
 882                                 /*
 883                                  * If we have a raidz/mirror that combines disks
 884                                  * with files, report it as an error.
 885                                  */
 886                                 if (!dontreport && type != NULL &&
 887                                     strcmp(type, childtype) != 0) {
 888                                         if (ret != NULL)
 889                                                 free(ret);
 890                                         ret = NULL;
 891                                         if (fatal)
 892                                                 vdev_error(gettext(
 893                                                     "mismatched replication "
 894                                                     "level: %s contains both "
 895                                                     "files and devices\n"),
 896                                                     rep.zprl_type);
 897                                         else
 898                                                 return (NULL);
 899                                         dontreport = B_TRUE;
 900                                 }
 901
 902                                 /*
 903                                  * According to stat(2), the value of 'st_size'
 904                                  * is undefined for block devices and character
 905                                  * devices.  But there is no effective way to
 906                                  * determine the real size in userland.
 907                                  *
 908                                  * Instead, we'll take advantage of an
 909                                  * implementation detail of spec_size().  If the
 910                                  * device is currently open, then we (should)
 911                                  * return a valid size.
 912                                  *
 913                                  * If we still don't get a valid size (indicated
 914                                  * by a size of 0 or MAXOFFSET_T), then ignore
 915                                  * this device altogether.
 916                                  */
 917                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 918                                         err = fstat64(fd, &statbuf);
 919                                         (void) close(fd);
 920                                 } else {
 921                                         err = stat64(path, &statbuf);
 922                                 }
 923
 924                                 if (err != 0 ||
 925                                     statbuf.st_size == 0 ||
 926                                     statbuf.st_size == MAXOFFSET_T)
 927                                         continue;
 928
 929                                 size = statbuf.st_size;
 930
 931                                 /*
 932                                  * Also make sure that devices and
 933                                  * slices have a consistent size.  If
 934                                  * they differ by a significant amount
 935                                  * (~16MB) then report an error.
 936                                  */
 937                                 if (!dontreport &&
 938                                     (vdev_size != -1ULL &&
 939                                     (labs(size - vdev_size) >
 940                                     ZPOOL_FUZZ))) {
 941                                         if (ret != NULL)
 942                                                 free(ret);
 943                                         ret = NULL;
 944                                         if (fatal)
 945                                                 vdev_error(gettext(
 946                                                     "%s contains devices of "
 947                                                     "different sizes\n"),
 948                                                     rep.zprl_type);
 949                                         else
 950                                                 return (NULL);
 951                                         dontreport = B_TRUE;
 952                                 }
 953
 954                                 type = childtype;
 955                                 vdev_size = size;
 956                         }
 957                 }
 958
 959                 /*
 960                  * At this point, we have the replication of the last toplevel
 961                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 962                  * different.
 963                  */
 964                 if (lastrep.zprl_type != NULL) {
 965                         if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
 966                                 if (ret != NULL)
 967                                         free(ret);
 968                                 ret = NULL;
 969                                 if (fatal)
 970                                         vdev_error(gettext(
 971                                             "mismatched replication level: "
 972                                             "both %s and %s vdevs are "
 973                                             "present\n"),
 974                                             lastrep.zprl_type, rep.zprl_type);
 975                                 else
 976                                         return (NULL);
 977                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 978                                 if (ret)
 979                                         free(ret);
 980                                 ret = NULL;
 981                                 if (fatal)
 982                                         vdev_error(gettext(
 983                                             "mismatched replication level: "
 984                                             "both %llu and %llu device parity "
 985                                             "%s vdevs are present\n"),
 986                                             lastrep.zprl_parity,
 987                                             rep.zprl_parity,
 988                                             rep.zprl_type);
 989                                 else
 990                                         return (NULL);
 991                         } else if (lastrep.zprl_children != rep.zprl_children) {
 992                                 if (ret)
 993                                         free(ret);
 994                                 ret = NULL;
 995                                 if (fatal)
 996                                         vdev_error(gettext(
 997                                             "mismatched replication level: "
 998                                             "both %llu-way and %llu-way %s "
 999                                             "vdevs are present\n"),
1000                                             lastrep.zprl_children,
1001                                             rep.zprl_children,
1002                                             rep.zprl_type);
1003                                 else
1004                                         return (NULL);
1005                         }
1006                 }
1007                 lastrep = rep;
1008         }
1009
1010         if (ret != NULL)
1011                 *ret = rep;
1012
1013         return (ret);
1014 }
1015
1016 /*
1017  * Check the replication level of the vdev spec against the current pool.  Calls
1018  * get_replication() to make sure the new spec is self-consistent.  If the pool
1019  * has a consistent replication level, then we ignore any errors.  Otherwise,
1020  * report any difference between the two.
1021  */
1022 static int
1023 check_replication(nvlist_t *config, nvlist_t *newroot)
1024 {
1025         nvlist_t **child;
1026         uint_t  children;
1027         replication_level_t *current = NULL, *new;
1028         int ret;
1029
1030         /*
1031          * If we have a current pool configuration, check to see if it's
1032          * self-consistent.  If not, simply return success.
1033          */
1034         if (config != NULL) {
1035                 nvlist_t *nvroot;
1036
1037                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1038                     &nvroot) == 0);
1039                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
1040                         return (0);
1041         }
1042         /*
1043          * for spares there may be no children, and therefore no
1044          * replication level to check
1045          */
1046         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
1047             &child, &children) != 0) || (children == 0)) {
1048                 free(current);
1049                 return (0);
1050         }
1051
1052         /*
1053          * If all we have is logs then there's no replication level to check.
1054          */
1055         if (num_logs(newroot) == children) {
1056                 free(current);
1057                 return (0);
1058         }
1059
1060         /*
1061          * Get the replication level of the new vdev spec, reporting any
1062          * inconsistencies found.
1063          */
1064         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
1065                 free(current);
1066                 return (-1);
1067         }
1068
1069         /*
1070          * Check to see if the new vdev spec matches the replication level of
1071          * the current pool.
1072          */
1073         ret = 0;
1074         if (current != NULL) {
1075                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
1076                         vdev_error(gettext(
1077                             "mismatched replication level: pool uses %s "
1078                             "and new vdev is %s\n"),
1079                             current->zprl_type, new->zprl_type);
1080                         ret = -1;
1081                 } else if (current->zprl_parity != new->zprl_parity) {
1082                         vdev_error(gettext(
1083                             "mismatched replication level: pool uses %llu "
1084                             "device parity and new vdev uses %llu\n"),
1085                             current->zprl_parity, new->zprl_parity);
1086                         ret = -1;
1087                 } else if (current->zprl_children != new->zprl_children) {
1088                         vdev_error(gettext(
1089                             "mismatched replication level: pool uses %llu-way "
1090                             "%s and new vdev uses %llu-way %s\n"),
1091                             current->zprl_children, current->zprl_type,
1092                             new->zprl_children, new->zprl_type);
1093                         ret = -1;
1094                 }
1095         }
1096
1097         free(new);
1098         if (current != NULL)
1099                 free(current);
1100
1101         return (ret);
1102 }
1103
1104 static int
1105 zero_label(char *path)
1106 {
1107         const int size = 4096;
1108         char buf[size];
1109         int err, fd;
1110
1111         if ((fd = open(path, O_WRONLY|O_EXCL)) < 0) {
1112                 (void) fprintf(stderr, gettext("cannot open '%s': %s\n"),
1113                     path, strerror(errno));
1114                 return (-1);
1115         }
1116
1117         memset(buf, 0, size);
1118         err = write(fd, buf, size);
1119         (void) fdatasync(fd);
1120         (void) close(fd);
1121
1122         if (err == -1) {
1123                 (void) fprintf(stderr, gettext("cannot zero first %d bytes "
1124                     "of '%s': %s\n"), size, path, strerror(errno));
1125                 return (-1);
1126         }
1127
1128         if (err != size) {
1129                 (void) fprintf(stderr, gettext("could only zero %d/%d bytes "
1130                     "of '%s'\n"), err, size, path);
1131                 return (-1);
1132         }
1133
1134         return (0);
1135 }
1136
1137 /*
1138  * Go through and find any whole disks in the vdev specification, labelling them
1139  * as appropriate.  When constructing the vdev spec, we were unable to open this
1140  * device in order to provide a devid.  Now that we have labelled the disk and
1141  * know that slice 0 is valid, we can construct the devid now.
1142  *
1143  * If the disk was already labeled with an EFI label, we will have gotten the
1144  * devid already (because we were able to open the whole disk).  Otherwise, we
1145  * need to get the devid after we label the disk.
1146  */
1147 static int
1148 make_disks(zpool_handle_t *zhp, nvlist_t *nv)
1149 {
1150         nvlist_t **child;
1151         uint_t c, children;
1152         char *type, *path;
1153         char devpath[MAXPATHLEN];
1154         char udevpath[MAXPATHLEN];
1155         uint64_t wholedisk;
1156         struct stat64 statbuf;
1157         int is_exclusive = 0;
1158         int fd;
1159         int ret;
1160
1161         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1162
1163         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1164             &child, &children) != 0) {
1165
1166                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1167                         return (0);
1168
1169                 /*
1170                  * We have a disk device.  If this is a whole disk write
1171                  * out the efi partition table, otherwise write zero's to
1172                  * the first 4k of the partition.  This is to ensure that
1173                  * libblkid will not misidentify the partition due to a
1174                  * magic value left by the previous filesystem.
1175                  */
1176                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1177                 verify(!nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1178                     &wholedisk));
1179
1180                 if (!wholedisk) {
1181                         (void) zero_label(path);
1182                         return (0);
1183                 }
1184
1185                 if (realpath(path, devpath) == NULL) {
1186                         ret = errno;
1187                         (void) fprintf(stderr,
1188                             gettext("cannot resolve path '%s'\n"), path);
1189                         return (ret);
1190                 }
1191
1192                 /*
1193                  * Remove any previously existing symlink from a udev path to
1194                  * the device before labeling the disk.  This makes
1195                  * zpool_label_disk_wait() truly wait for the new link to show
1196                  * up instead of returning if it finds an old link still in
1197                  * place.  Otherwise there is a window between when udev
1198                  * deletes and recreates the link during which access attempts
1199                  * will fail with ENOENT.
1200                  */
1201                 strncpy(udevpath, path, MAXPATHLEN);
1202                 (void) zfs_append_partition(udevpath, MAXPATHLEN);
1203
1204                 fd = open(devpath, O_RDWR|O_EXCL);
1205                 if (fd == -1) {
1206                         if (errno == EBUSY)
1207                                 is_exclusive = 1;
1208                 } else {
1209                         (void) close(fd);
1210                 }
1211
1212                 /*
1213                  * If the partition exists, contains a valid spare label,
1214                  * and is opened exclusively there is no need to partition
1215                  * it.  Hot spares have already been partitioned and are
1216                  * held open exclusively by the kernel as a safety measure.
1217                  *
1218                  * If the provided path is for a /dev/disk/ device its
1219                  * symbolic link will be removed, partition table created,
1220                  * and then block until udev creates the new link.
1221                  */
1222                 if (!is_exclusive || !is_spare(NULL, udevpath)) {
1223                         ret = strncmp(udevpath, UDISK_ROOT, strlen(UDISK_ROOT));
1224                         if (ret == 0) {
1225                                 ret = lstat64(udevpath, &statbuf);
1226                                 if (ret == 0 && S_ISLNK(statbuf.st_mode))
1227                                         (void) unlink(udevpath);
1228                         }
1229
1230                         if (zpool_label_disk(g_zfs, zhp,
1231                             strrchr(devpath, '/') + 1) == -1)
1232                                 return (-1);
1233
1234                         ret = zpool_label_disk_wait(udevpath, DISK_LABEL_WAIT);
1235                         if (ret) {
1236                                 (void) fprintf(stderr, gettext("cannot "
1237                                     "resolve path '%s': %d\n"), udevpath, ret);
1238                                 return (-1);
1239                         }
1240
1241                         (void) zero_label(udevpath);
1242                 }
1243
1244                 /*
1245                  * Update the path to refer to the partition.  The presence of
1246                  * the 'whole_disk' field indicates to the CLI that we should
1247                  * chop off the partition number when displaying the device in
1248                  * future output.
1249                  */
1250                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
1251
1252                 return (0);
1253         }
1254
1255         for (c = 0; c < children; c++)
1256                 if ((ret = make_disks(zhp, child[c])) != 0)
1257                         return (ret);
1258
1259         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1260             &child, &children) == 0)
1261                 for (c = 0; c < children; c++)
1262                         if ((ret = make_disks(zhp, child[c])) != 0)
1263                                 return (ret);
1264
1265         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1266             &child, &children) == 0)
1267                 for (c = 0; c < children; c++)
1268                         if ((ret = make_disks(zhp, child[c])) != 0)
1269                                 return (ret);
1270
1271         return (0);
1272 }
1273
1274 /*
1275  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1276  * the majority of this task.
1277  */
1278 static boolean_t
1279 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1280     boolean_t replacing, boolean_t isspare)
1281 {
1282         nvlist_t **child;
1283         uint_t c, children;
1284         char *type, *path;
1285         int ret = 0;
1286         char buf[MAXPATHLEN];
1287         uint64_t wholedisk = B_FALSE;
1288         boolean_t anyinuse = B_FALSE;
1289
1290         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1291
1292         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1293             &child, &children) != 0) {
1294
1295                 verify(!nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path));
1296                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1297                         verify(!nvlist_lookup_uint64(nv,
1298                             ZPOOL_CONFIG_WHOLE_DISK, &wholedisk));
1299
1300                 /*
1301                  * As a generic check, we look to see if this is a replace of a
1302                  * hot spare within the same pool.  If so, we allow it
1303                  * regardless of what libblkid or zpool_in_use() says.
1304                  */
1305                 if (replacing) {
1306                         (void) strlcpy(buf, path, sizeof (buf));
1307                         if (wholedisk) {
1308                                 ret = zfs_append_partition(buf,  sizeof (buf));
1309                                 if (ret == -1)
1310                                         return (-1);
1311                         }
1312
1313                         if (is_spare(config, buf))
1314                                 return (B_FALSE);
1315                 }
1316
1317                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1318                         ret = check_device(path, force, isspare, wholedisk);
1319
1320                 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1321                         ret = check_file(path, force, isspare);
1322
1323                 return (ret != 0);
1324         }
1325
1326         for (c = 0; c < children; c++)
1327                 if (is_device_in_use(config, child[c], force, replacing,
1328                     B_FALSE))
1329                         anyinuse = B_TRUE;
1330
1331         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1332             &child, &children) == 0)
1333                 for (c = 0; c < children; c++)
1334                         if (is_device_in_use(config, child[c], force, replacing,
1335                             B_TRUE))
1336                                 anyinuse = B_TRUE;
1337
1338         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1339             &child, &children) == 0)
1340                 for (c = 0; c < children; c++)
1341                         if (is_device_in_use(config, child[c], force, replacing,
1342                             B_FALSE))
1343                                 anyinuse = B_TRUE;
1344
1345         return (anyinuse);
1346 }
1347
1348 static const char *
1349 is_grouping(const char *type, int *mindev, int *maxdev)
1350 {
1351         if (strncmp(type, "raidz", 5) == 0) {
1352                 const char *p = type + 5;
1353                 char *end;
1354                 long nparity;
1355
1356                 if (*p == '\0') {
1357                         nparity = 1;
1358                 } else if (*p == '0') {
1359                         return (NULL); /* no zero prefixes allowed */
1360                 } else {
1361                         errno = 0;
1362                         nparity = strtol(p, &end, 10);
1363                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1364                             *end != '\0')
1365                                 return (NULL);
1366                 }
1367
1368                 if (mindev != NULL)
1369                         *mindev = nparity + 1;
1370                 if (maxdev != NULL)
1371                         *maxdev = 255;
1372                 return (VDEV_TYPE_RAIDZ);
1373         }
1374
1375         if (maxdev != NULL)
1376                 *maxdev = INT_MAX;
1377
1378         if (strcmp(type, "mirror") == 0) {
1379                 if (mindev != NULL)
1380                         *mindev = 2;
1381                 return (VDEV_TYPE_MIRROR);
1382         }
1383
1384         if (strcmp(type, "spare") == 0) {
1385                 if (mindev != NULL)
1386                         *mindev = 1;
1387                 return (VDEV_TYPE_SPARE);
1388         }
1389
1390         if (strcmp(type, "log") == 0) {
1391                 if (mindev != NULL)
1392                         *mindev = 1;
1393                 return (VDEV_TYPE_LOG);
1394         }
1395
1396         if (strcmp(type, "cache") == 0) {
1397                 if (mindev != NULL)
1398                         *mindev = 1;
1399                 return (VDEV_TYPE_L2CACHE);
1400         }
1401
1402         return (NULL);
1403 }
1404
1405 /*
1406  * Construct a syntactically valid vdev specification,
1407  * and ensure that all devices and files exist and can be opened.
1408  * Note: we don't bother freeing anything in the error paths
1409  * because the program is just going to exit anyway.
1410  */
1411 nvlist_t *
1412 construct_spec(nvlist_t *props, int argc, char **argv)
1413 {
1414         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1415         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1416         const char *type;
1417         uint64_t is_log;
1418         boolean_t seen_logs;
1419
1420         top = NULL;
1421         toplevels = 0;
1422         spares = NULL;
1423         l2cache = NULL;
1424         nspares = 0;
1425         nlogs = 0;
1426         nl2cache = 0;
1427         is_log = B_FALSE;
1428         seen_logs = B_FALSE;
1429
1430         while (argc > 0) {
1431                 nv = NULL;
1432
1433                 /*
1434                  * If it's a mirror or raidz, the subsequent arguments are
1435                  * its leaves -- until we encounter the next mirror or raidz.
1436                  */
1437                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1438                         nvlist_t **child = NULL;
1439                         int c, children = 0;
1440
1441                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1442                                 if (spares != NULL) {
1443                                         (void) fprintf(stderr,
1444                                             gettext("invalid vdev "
1445                                             "specification: 'spare' can be "
1446                                             "specified only once\n"));
1447                                         return (NULL);
1448                                 }
1449                                 is_log = B_FALSE;
1450                         }
1451
1452                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1453                                 if (seen_logs) {
1454                                         (void) fprintf(stderr,
1455                                             gettext("invalid vdev "
1456                                             "specification: 'log' can be "
1457                                             "specified only once\n"));
1458                                         return (NULL);
1459                                 }
1460                                 seen_logs = B_TRUE;
1461                                 is_log = B_TRUE;
1462                                 argc--;
1463                                 argv++;
1464                                 /*
1465                                  * A log is not a real grouping device.
1466                                  * We just set is_log and continue.
1467                                  */
1468                                 continue;
1469                         }
1470
1471                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1472                                 if (l2cache != NULL) {
1473                                         (void) fprintf(stderr,
1474                                             gettext("invalid vdev "
1475                                             "specification: 'cache' can be "
1476                                             "specified only once\n"));
1477                                         return (NULL);
1478                                 }
1479                                 is_log = B_FALSE;
1480                         }
1481
1482                         if (is_log) {
1483                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1484                                         (void) fprintf(stderr,
1485                                             gettext("invalid vdev "
1486                                             "specification: unsupported 'log' "
1487                                             "device: %s\n"), type);
1488                                         return (NULL);
1489                                 }
1490                                 nlogs++;
1491                         }
1492
1493                         for (c = 1; c < argc; c++) {
1494                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1495                                         break;
1496                                 children++;
1497                                 child = realloc(child,
1498                                     children * sizeof (nvlist_t *));
1499                                 if (child == NULL)
1500                                         zpool_no_memory();
1501                                 if ((nv = make_leaf_vdev(props, argv[c],
1502                                     B_FALSE)) == NULL)
1503                                         return (NULL);
1504                                 child[children - 1] = nv;
1505                         }
1506
1507                         if (children < mindev) {
1508                                 (void) fprintf(stderr, gettext("invalid vdev "
1509                                     "specification: %s requires at least %d "
1510                                     "devices\n"), argv[0], mindev);
1511                                 return (NULL);
1512                         }
1513
1514                         if (children > maxdev) {
1515                                 (void) fprintf(stderr, gettext("invalid vdev "
1516                                     "specification: %s supports no more than "
1517                                     "%d devices\n"), argv[0], maxdev);
1518                                 return (NULL);
1519                         }
1520
1521                         argc -= c;
1522                         argv += c;
1523
1524                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1525                                 spares = child;
1526                                 nspares = children;
1527                                 continue;
1528                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1529                                 l2cache = child;
1530                                 nl2cache = children;
1531                                 continue;
1532                         } else {
1533                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1534                                     0) == 0);
1535                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1536                                     type) == 0);
1537                                 verify(nvlist_add_uint64(nv,
1538                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1539                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1540                                         verify(nvlist_add_uint64(nv,
1541                                             ZPOOL_CONFIG_NPARITY,
1542                                             mindev - 1) == 0);
1543                                 }
1544                                 verify(nvlist_add_nvlist_array(nv,
1545                                     ZPOOL_CONFIG_CHILDREN, child,
1546                                     children) == 0);
1547
1548                                 for (c = 0; c < children; c++)
1549                                         nvlist_free(child[c]);
1550                                 free(child);
1551                         }
1552                 } else {
1553                         /*
1554                          * We have a device.  Pass off to make_leaf_vdev() to
1555                          * construct the appropriate nvlist describing the vdev.
1556                          */
1557                         if ((nv = make_leaf_vdev(props, argv[0],
1558                             is_log)) == NULL)
1559                                 return (NULL);
1560                         if (is_log)
1561                                 nlogs++;
1562                         argc--;
1563                         argv++;
1564                 }
1565
1566                 toplevels++;
1567                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1568                 if (top == NULL)
1569                         zpool_no_memory();
1570                 top[toplevels - 1] = nv;
1571         }
1572
1573         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1574                 (void) fprintf(stderr, gettext("invalid vdev "
1575                     "specification: at least one toplevel vdev must be "
1576                     "specified\n"));
1577                 return (NULL);
1578         }
1579
1580         if (seen_logs && nlogs == 0) {
1581                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1582                     "log requires at least 1 device\n"));
1583                 return (NULL);
1584         }
1585
1586         /*
1587          * Finally, create nvroot and add all top-level vdevs to it.
1588          */
1589         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1590         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1591             VDEV_TYPE_ROOT) == 0);
1592         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1593             top, toplevels) == 0);
1594         if (nspares != 0)
1595                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1596                     spares, nspares) == 0);
1597         if (nl2cache != 0)
1598                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1599                     l2cache, nl2cache) == 0);
1600
1601         for (t = 0; t < toplevels; t++)
1602                 nvlist_free(top[t]);
1603         for (t = 0; t < nspares; t++)
1604                 nvlist_free(spares[t]);
1605         for (t = 0; t < nl2cache; t++)
1606                 nvlist_free(l2cache[t]);
1607         if (spares)
1608                 free(spares);
1609         if (l2cache)
1610                 free(l2cache);
1611         free(top);
1612
1613         return (nvroot);
1614 }
1615
1616 nvlist_t *
1617 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1618     splitflags_t flags, int argc, char **argv)
1619 {
1620         nvlist_t *newroot = NULL, **child;
1621         uint_t c, children;
1622
1623         if (argc > 0) {
1624                 if ((newroot = construct_spec(props, argc, argv)) == NULL) {
1625                         (void) fprintf(stderr, gettext("Unable to build a "
1626                             "pool from the specified devices\n"));
1627                         return (NULL);
1628                 }
1629
1630                 if (!flags.dryrun && make_disks(zhp, newroot) != 0) {
1631                         nvlist_free(newroot);
1632                         return (NULL);
1633                 }
1634
1635                 /* avoid any tricks in the spec */
1636                 verify(nvlist_lookup_nvlist_array(newroot,
1637                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1638                 for (c = 0; c < children; c++) {
1639                         char *path;
1640                         const char *type;
1641                         int min, max;
1642
1643                         verify(nvlist_lookup_string(child[c],
1644                             ZPOOL_CONFIG_PATH, &path) == 0);
1645                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1646                                 (void) fprintf(stderr, gettext("Cannot use "
1647                                     "'%s' as a device for splitting\n"), type);
1648                                 nvlist_free(newroot);
1649                                 return (NULL);
1650                         }
1651                 }
1652         }
1653
1654         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1655                 if (newroot != NULL)
1656                         nvlist_free(newroot);
1657                 return (NULL);
1658         }
1659
1660         return (newroot);
1661 }
1662
1663 /*
1664  * Get and validate the contents of the given vdev specification.  This ensures
1665  * that the nvlist returned is well-formed, that all the devices exist, and that
1666  * they are not currently in use by any other known consumer.  The 'poolconfig'
1667  * parameter is the current configuration of the pool when adding devices
1668  * existing pool, and is used to perform additional checks, such as changing the
1669  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1670  * new pool.  The 'force' flag controls whether devices should be forcefully
1671  * added, even if they appear in use.
1672  */
1673 nvlist_t *
1674 make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
1675     boolean_t replacing, boolean_t dryrun, int argc, char **argv)
1676 {
1677         nvlist_t *newroot;
1678         nvlist_t *poolconfig = NULL;
1679         is_force = force;
1680
1681         /*
1682          * Construct the vdev specification.  If this is successful, we know
1683          * that we have a valid specification, and that all devices can be
1684          * opened.
1685          */
1686         if ((newroot = construct_spec(props, argc, argv)) == NULL)
1687                 return (NULL);
1688
1689         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL)) {
1690                 nvlist_free(newroot);
1691                 return (NULL);
1692         }
1693
1694         /*
1695          * Validate each device to make sure that its not shared with another
1696          * subsystem.  We do this even if 'force' is set, because there are some
1697          * uses (such as a dedicated dump device) that even '-f' cannot
1698          * override.
1699          */
1700         if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1701                 nvlist_free(newroot);
1702                 return (NULL);
1703         }
1704
1705         /*
1706          * Check the replication level of the given vdevs and report any errors
1707          * found.  We include the existing pool spec, if any, as we need to
1708          * catch changes against the existing replication level.
1709          */
1710         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1711                 nvlist_free(newroot);
1712                 return (NULL);
1713         }
1714
1715         /*
1716          * Run through the vdev specification and label any whole disks found.
1717          */
1718         if (!dryrun && make_disks(zhp, newroot) != 0) {
1719                 nvlist_free(newroot);
1720                 return (NULL);
1721         }
1722
1723         return (newroot);
1724 }