cddl/contrib/opensolaris/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, 2018 by Delphix. All rights reserved.
  25  * Copyright (c) 2016, 2017 Intel Corporation.
  26  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>.
  27  */
  28
  29 /*
  30  * Functions to convert between a list of vdevs and an nvlist representing the
  31  * configuration.  Each entry in the list can be one of:
  32  *
  33  *      Device vdevs
  34  *              disk=(path=..., devid=...)
  35  *              file=(path=...)
  36  *
  37  *      Group vdevs
  38  *              raidz[1|2]=(...)
  39  *              mirror=(...)
  40  *
  41  *      Hot spares
  42  *
  43  * While the underlying implementation supports it, group vdevs cannot contain
  44  * other group vdevs.  All userland verification of devices is contained within
  45  * this file.  If successful, the nvlist returned can be passed directly to the
  46  * kernel; we've done as much verification as possible in userland.
  47  *
  48  * Hot spares are a special case, and passed down as an array of disk vdevs, at
  49  * the same level as the root of the vdev tree.
  50  *
  51  * The only function exported by this file is 'make_root_vdev'.  The
  52  * function performs several passes:
  53  *
  54  *      1. Construct the vdev specification.  Performs syntax validation and
  55  *         makes sure each device is valid.
  56  *      2. Check for devices in use.  Using libdiskmgt, makes sure that no
  57  *         devices are also in use.  Some can be overridden using the 'force'
  58  *         flag, others cannot.
  59  *      3. Check for replication errors if the 'force' flag is not specified.
  60  *         validates that the replication level is consistent across the
  61  *         entire pool.
  62  *      4. Call libzfs to label any whole disks with an EFI label.
  63  */
  64
  65 #include <assert.h>
  66 #include <devid.h>
  67 #include <errno.h>
  68 #include <fcntl.h>
  69 #include <libintl.h>
  70 #include <libnvpair.h>
  71 #include <limits.h>
  72 #include <stdio.h>
  73 #include <string.h>
  74 #include <unistd.h>
  75 #include <paths.h>
  76 #include <sys/stat.h>
  77 #include <sys/disk.h>
  78 #include <sys/mntent.h>
  79 #include <libgeom.h>
  80
  81 #include "zpool_util.h"
  82
  83 #define BACKUP_SLICE    "s2"
  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 /*PRINTFLIKE1*/
  94 static void
  95 vdev_error(const char *fmt, ...)
  96 {
  97         va_list ap;
  98
  99         if (!error_seen) {
 100                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
 101                 if (!is_force)
 102                         (void) fprintf(stderr, gettext("use '-f' to override "
 103                             "the following errors:\n"));
 104                 else
 105                         (void) fprintf(stderr, gettext("the following errors "
 106                             "must be manually repaired:\n"));
 107                 error_seen = B_TRUE;
 108         }
 109
 110         va_start(ap, fmt);
 111         (void) vfprintf(stderr, fmt, ap);
 112         va_end(ap);
 113 }
 114
 115 #ifdef illumos
 116 static void
 117 libdiskmgt_error(int error)
 118 {
 119         /*
 120          * ENXIO/ENODEV is a valid error message if the device doesn't live in
 121          * /dev/dsk.  Don't bother printing an error message in this case.
 122          */
 123         if (error == ENXIO || error == ENODEV)
 124                 return;
 125
 126         (void) fprintf(stderr, gettext("warning: device in use checking "
 127             "failed: %s\n"), strerror(error));
 128 }
 129
 130 /*
 131  * Validate a device, passing the bulk of the work off to libdiskmgt.
 132  */
 133 static int
 134 check_slice(const char *path, int force, boolean_t wholedisk, boolean_t isspare)
 135 {
 136         char *msg;
 137         int error = 0;
 138         dm_who_type_t who;
 139
 140         if (force)
 141                 who = DM_WHO_ZPOOL_FORCE;
 142         else if (isspare)
 143                 who = DM_WHO_ZPOOL_SPARE;
 144         else
 145                 who = DM_WHO_ZPOOL;
 146
 147         if (dm_inuse((char *)path, &msg, who, &error) || error) {
 148                 if (error != 0) {
 149                         libdiskmgt_error(error);
 150                         return (0);
 151                 } else {
 152                         vdev_error("%s", msg);
 153                         free(msg);
 154                         return (-1);
 155                 }
 156         }
 157
 158         /*
 159          * If we're given a whole disk, ignore overlapping slices since we're
 160          * about to label it anyway.
 161          */
 162         error = 0;
 163         if (!wholedisk && !force &&
 164             (dm_isoverlapping((char *)path, &msg, &error) || error)) {
 165                 if (error == 0) {
 166                         /* dm_isoverlapping returned -1 */
 167                         vdev_error(gettext("%s overlaps with %s\n"), path, msg);
 168                         free(msg);
 169                         return (-1);
 170                 } else if (error != ENODEV) {
 171                         /* libdiskmgt's devcache only handles physical drives */
 172                         libdiskmgt_error(error);
 173                         return (0);
 174                 }
 175         }
 176
 177         return (0);
 178 }
 179
 180
 181 /*
 182  * Validate a whole disk.  Iterate over all slices on the disk and make sure
 183  * that none is in use by calling check_slice().
 184  */
 185 static int
 186 check_disk(const char *name, dm_descriptor_t disk, int force, int isspare)
 187 {
 188         dm_descriptor_t *drive, *media, *slice;
 189         int err = 0;
 190         int i;
 191         int ret;
 192
 193         /*
 194          * Get the drive associated with this disk.  This should never fail,
 195          * because we already have an alias handle open for the device.
 196          */
 197         if ((drive = dm_get_associated_descriptors(disk, DM_DRIVE,
 198             &err)) == NULL || *drive == NULL) {
 199                 if (err)
 200                         libdiskmgt_error(err);
 201                 return (0);
 202         }
 203
 204         if ((media = dm_get_associated_descriptors(*drive, DM_MEDIA,
 205             &err)) == NULL) {
 206                 dm_free_descriptors(drive);
 207                 if (err)
 208                         libdiskmgt_error(err);
 209                 return (0);
 210         }
 211
 212         dm_free_descriptors(drive);
 213
 214         /*
 215          * It is possible that the user has specified a removable media drive,
 216          * and the media is not present.
 217          */
 218         if (*media == NULL) {
 219                 dm_free_descriptors(media);
 220                 vdev_error(gettext("'%s' has no media in drive\n"), name);
 221                 return (-1);
 222         }
 223
 224         if ((slice = dm_get_associated_descriptors(*media, DM_SLICE,
 225             &err)) == NULL) {
 226                 dm_free_descriptors(media);
 227                 if (err)
 228                         libdiskmgt_error(err);
 229                 return (0);
 230         }
 231
 232         dm_free_descriptors(media);
 233
 234         ret = 0;
 235
 236         /*
 237          * Iterate over all slices and report any errors.  We don't care about
 238          * overlapping slices because we are using the whole disk.
 239          */
 240         for (i = 0; slice[i] != NULL; i++) {
 241                 char *name = dm_get_name(slice[i], &err);
 242
 243                 if (check_slice(name, force, B_TRUE, isspare) != 0)
 244                         ret = -1;
 245
 246                 dm_free_name(name);
 247         }
 248
 249         dm_free_descriptors(slice);
 250         return (ret);
 251 }
 252
 253 /*
 254  * Validate a device.
 255  */
 256 static int
 257 check_device(const char *path, boolean_t force, boolean_t isspare)
 258 {
 259         dm_descriptor_t desc;
 260         int err;
 261         char *dev;
 262
 263         /*
 264          * For whole disks, libdiskmgt does not include the leading dev path.
 265          */
 266         dev = strrchr(path, '/');
 267         assert(dev != NULL);
 268         dev++;
 269         if ((desc = dm_get_descriptor_by_name(DM_ALIAS, dev, &err)) != NULL) {
 270                 err = check_disk(path, desc, force, isspare);
 271                 dm_free_descriptor(desc);
 272                 return (err);
 273         }
 274
 275         return (check_slice(path, force, B_FALSE, isspare));
 276 }
 277 #endif  /* illumos */
 278
 279 /*
 280  * Check that a file is valid.  All we can do in this case is check that it's
 281  * not in use by another pool, and not in use by swap.
 282  */
 283 static int
 284 check_file(const char *file, boolean_t force, boolean_t isspare)
 285 {
 286         char  *name;
 287         int fd;
 288         int ret = 0;
 289         int err;
 290         pool_state_t state;
 291         boolean_t inuse;
 292
 293 #ifdef illumos
 294         if (dm_inuse_swap(file, &err)) {
 295                 if (err)
 296                         libdiskmgt_error(err);
 297                 else
 298                         vdev_error(gettext("%s is currently used by swap. "
 299                             "Please see swap(1M).\n"), file);
 300                 return (-1);
 301         }
 302 #endif
 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 int
 357 check_device(const char *name, boolean_t force, boolean_t isspare)
 358 {
 359         char path[MAXPATHLEN];
 360
 361         if (strncmp(name, _PATH_DEV, sizeof(_PATH_DEV) - 1) != 0)
 362                 snprintf(path, sizeof(path), "%s%s", _PATH_DEV, name);
 363         else
 364                 strlcpy(path, name, sizeof(path));
 365
 366         return (check_file(path, force, isspare));
 367 }
 368
 369 /*
 370  * By "whole disk" we mean an entire physical disk (something we can
 371  * label, toggle the write cache on, etc.) as opposed to the full
 372  * capacity of a pseudo-device such as lofi or did.  We act as if we
 373  * are labeling the disk, which should be a pretty good test of whether
 374  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 375  * it isn't.
 376  */
 377 static boolean_t
 378 is_whole_disk(const char *arg)
 379 {
 380 #ifdef illumos
 381         struct dk_gpt *label;
 382         int     fd;
 383         char    path[MAXPATHLEN];
 384
 385         (void) snprintf(path, sizeof (path), "%s%s%s",
 386             ZFS_RDISK_ROOT, strrchr(arg, '/'), BACKUP_SLICE);
 387         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0)
 388                 return (B_FALSE);
 389         if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
 390                 (void) close(fd);
 391                 return (B_FALSE);
 392         }
 393         efi_free(label);
 394         (void) close(fd);
 395         return (B_TRUE);
 396 #else
 397         int fd;
 398
 399         fd = g_open(arg, 0);
 400         if (fd >= 0) {
 401                 g_close(fd);
 402                 return (B_TRUE);
 403         }
 404         return (B_FALSE);
 405 #endif
 406 }
 407
 408 /*
 409  * Create a leaf vdev.  Determine if this is a file or a device.  If it's a
 410  * device, fill in the device id to make a complete nvlist.  Valid forms for a
 411  * leaf vdev are:
 412  *
 413  *      /dev/dsk/xxx    Complete disk path
 414  *      /xxx            Full path to file
 415  *      xxx             Shorthand for /dev/dsk/xxx
 416  */
 417 static nvlist_t *
 418 make_leaf_vdev(const char *arg, uint64_t is_log)
 419 {
 420         char path[MAXPATHLEN];
 421         struct stat64 statbuf;
 422         nvlist_t *vdev = NULL;
 423         char *type = NULL;
 424         boolean_t wholedisk = B_FALSE;
 425
 426         /*
 427          * Determine what type of vdev this is, and put the full path into
 428          * 'path'.  We detect whether this is a device of file afterwards by
 429          * checking the st_mode of the file.
 430          */
 431         if (arg[0] == '/') {
 432                 /*
 433                  * Complete device or file path.  Exact type is determined by
 434                  * examining the file descriptor afterwards.
 435                  */
 436                 wholedisk = is_whole_disk(arg);
 437                 if (!wholedisk && (stat64(arg, &statbuf) != 0)) {
 438                         (void) fprintf(stderr,
 439                             gettext("cannot open '%s': %s\n"),
 440                             arg, strerror(errno));
 441                         return (NULL);
 442                 }
 443
 444                 (void) strlcpy(path, arg, sizeof (path));
 445         } else {
 446                 /*
 447                  * This may be a short path for a device, or it could be total
 448                  * gibberish.  Check to see if it's a known device in
 449                  * /dev/dsk/.  As part of this check, see if we've been given a
 450                  * an entire disk (minus the slice number).
 451                  */
 452                 if (strncmp(arg, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
 453                         strlcpy(path, arg, sizeof (path));
 454                 else
 455                         snprintf(path, sizeof (path), "%s%s", _PATH_DEV, arg);
 456                 wholedisk = is_whole_disk(path);
 457                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 458                         /*
 459                          * If we got ENOENT, then the user gave us
 460                          * gibberish, so try to direct them with a
 461                          * reasonable error message.  Otherwise,
 462                          * regurgitate strerror() since it's the best we
 463                          * can do.
 464                          */
 465                         if (errno == ENOENT) {
 466                                 (void) fprintf(stderr,
 467                                     gettext("cannot open '%s': no such "
 468                                     "GEOM provider\n"), arg);
 469                                 (void) fprintf(stderr,
 470                                     gettext("must be a full path or "
 471                                     "shorthand device name\n"));
 472                                 return (NULL);
 473                         } else {
 474                                 (void) fprintf(stderr,
 475                                     gettext("cannot open '%s': %s\n"),
 476                                     path, strerror(errno));
 477                                 return (NULL);
 478                         }
 479                 }
 480         }
 481
 482 #ifdef __FreeBSD__
 483         if (S_ISCHR(statbuf.st_mode)) {
 484                 statbuf.st_mode &= ~S_IFCHR;
 485                 statbuf.st_mode |= S_IFBLK;
 486                 wholedisk = B_FALSE;
 487         }
 488 #endif
 489
 490         /*
 491          * Determine whether this is a device or a file.
 492          */
 493         if (wholedisk || S_ISBLK(statbuf.st_mode)) {
 494                 type = VDEV_TYPE_DISK;
 495         } else if (S_ISREG(statbuf.st_mode)) {
 496                 type = VDEV_TYPE_FILE;
 497         } else {
 498                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 499                     "GEOM provider or regular file\n"), path);
 500                 return (NULL);
 501         }
 502
 503         /*
 504          * Finally, we have the complete device or file, and we know that it is
 505          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 506          * vdevs have a 'path' element, and devices also have a 'devid' element.
 507          */
 508         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 509         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 510         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 511         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 512         if (is_log)
 513                 verify(nvlist_add_string(vdev, ZPOOL_CONFIG_ALLOCATION_BIAS,
 514                     VDEV_ALLOC_BIAS_LOG) == 0);
 515         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 516                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 517                     (uint64_t)wholedisk) == 0);
 518
 519 #ifdef have_devid
 520         /*
 521          * For a whole disk, defer getting its devid until after labeling it.
 522          */
 523         if (S_ISBLK(statbuf.st_mode) && !wholedisk) {
 524                 /*
 525                  * Get the devid for the device.
 526                  */
 527                 int fd;
 528                 ddi_devid_t devid;
 529                 char *minor = NULL, *devid_str = NULL;
 530
 531                 if ((fd = open(path, O_RDONLY)) < 0) {
 532                         (void) fprintf(stderr, gettext("cannot open '%s': "
 533                             "%s\n"), path, strerror(errno));
 534                         nvlist_free(vdev);
 535                         return (NULL);
 536                 }
 537
 538                 if (devid_get(fd, &devid) == 0) {
 539                         if (devid_get_minor_name(fd, &minor) == 0 &&
 540                             (devid_str = devid_str_encode(devid, minor)) !=
 541                             NULL) {
 542                                 verify(nvlist_add_string(vdev,
 543                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 544                         }
 545                         if (devid_str != NULL)
 546                                 devid_str_free(devid_str);
 547                         if (minor != NULL)
 548                                 devid_str_free(minor);
 549                         devid_free(devid);
 550                 }
 551
 552                 (void) close(fd);
 553         }
 554 #endif
 555
 556         return (vdev);
 557 }
 558
 559 /*
 560  * Go through and verify the replication level of the pool is consistent.
 561  * Performs the following checks:
 562  *
 563  *      For the new spec, verifies that devices in mirrors and raidz are the
 564  *      same size.
 565  *
 566  *      If the current configuration already has inconsistent replication
 567  *      levels, ignore any other potential problems in the new spec.
 568  *
 569  *      Otherwise, make sure that the current spec (if there is one) and the new
 570  *      spec have consistent replication levels.
 571  *
 572  *      If there is no current spec (create), make sure new spec has at least
 573  *      one general purpose vdev.
 574  */
 575 typedef struct replication_level {
 576         char *zprl_type;
 577         uint64_t zprl_children;
 578         uint64_t zprl_parity;
 579 } replication_level_t;
 580
 581 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 582
 583 static boolean_t
 584 is_raidz_mirror(replication_level_t *a, replication_level_t *b,
 585     replication_level_t **raidz, replication_level_t **mirror)
 586 {
 587         if (strcmp(a->zprl_type, "raidz") == 0 &&
 588             strcmp(b->zprl_type, "mirror") == 0) {
 589                 *raidz = a;
 590                 *mirror = b;
 591                 return (B_TRUE);
 592         }
 593         return (B_FALSE);
 594 }
 595
 596 /*
 597  * Given a list of toplevel vdevs, return the current replication level.  If
 598  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 599  * an error message will be displayed for each self-inconsistent vdev.
 600  */
 601 static replication_level_t *
 602 get_replication(nvlist_t *nvroot, boolean_t fatal)
 603 {
 604         nvlist_t **top;
 605         uint_t t, toplevels;
 606         nvlist_t **child;
 607         uint_t c, children;
 608         nvlist_t *nv;
 609         char *type;
 610         replication_level_t lastrep = {0};
 611         replication_level_t rep;
 612         replication_level_t *ret;
 613         replication_level_t *raidz, *mirror;
 614         boolean_t dontreport;
 615
 616         ret = safe_malloc(sizeof (replication_level_t));
 617
 618         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 619             &top, &toplevels) == 0);
 620
 621         for (t = 0; t < toplevels; t++) {
 622                 uint64_t is_log = B_FALSE;
 623
 624                 nv = top[t];
 625
 626                 /*
 627                  * For separate logs we ignore the top level vdev replication
 628                  * constraints.
 629                  */
 630                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 631                 if (is_log)
 632                         continue;
 633
 634                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 635                     &type) == 0);
 636                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 637                     &child, &children) != 0) {
 638                         /*
 639                          * This is a 'file' or 'disk' vdev.
 640                          */
 641                         rep.zprl_type = type;
 642                         rep.zprl_children = 1;
 643                         rep.zprl_parity = 0;
 644                 } else {
 645                         uint64_t vdev_size;
 646
 647                         /*
 648                          * This is a mirror or RAID-Z vdev.  Go through and make
 649                          * sure the contents are all the same (files vs. disks),
 650                          * keeping track of the number of elements in the
 651                          * process.
 652                          *
 653                          * We also check that the size of each vdev (if it can
 654                          * be determined) is the same.
 655                          */
 656                         rep.zprl_type = type;
 657                         rep.zprl_children = 0;
 658
 659                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 660                                 verify(nvlist_lookup_uint64(nv,
 661                                     ZPOOL_CONFIG_NPARITY,
 662                                     &rep.zprl_parity) == 0);
 663                                 assert(rep.zprl_parity != 0);
 664                         } else {
 665                                 rep.zprl_parity = 0;
 666                         }
 667
 668                         /*
 669                          * The 'dontreport' variable indicates that we've
 670                          * already reported an error for this spec, so don't
 671                          * bother doing it again.
 672                          */
 673                         type = NULL;
 674                         dontreport = 0;
 675                         vdev_size = -1ULL;
 676                         for (c = 0; c < children; c++) {
 677                                 boolean_t is_replacing, is_spare;
 678                                 nvlist_t *cnv = child[c];
 679                                 char *path;
 680                                 struct stat64 statbuf;
 681                                 uint64_t size = -1ULL;
 682                                 char *childtype;
 683                                 int fd, err;
 684
 685                                 rep.zprl_children++;
 686
 687                                 verify(nvlist_lookup_string(cnv,
 688                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 689
 690                                 /*
 691                                  * If this is a replacing or spare vdev, then
 692                                  * get the real first child of the vdev.
 693                                  */
 694                                 is_replacing = strcmp(childtype,
 695                                     VDEV_TYPE_REPLACING) == 0;
 696                                 is_spare = strcmp(childtype,
 697                                     VDEV_TYPE_SPARE) == 0;
 698                                 if (is_replacing || is_spare) {
 699                                         nvlist_t **rchild;
 700                                         uint_t rchildren;
 701
 702                                         verify(nvlist_lookup_nvlist_array(cnv,
 703                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 704                                             &rchildren) == 0);
 705                                         assert((is_replacing && rchildren == 2)
 706                                             || (is_spare && rchildren >= 2));
 707                                         cnv = rchild[0];
 708
 709                                         verify(nvlist_lookup_string(cnv,
 710                                             ZPOOL_CONFIG_TYPE,
 711                                             &childtype) == 0);
 712                                         if (strcmp(childtype,
 713                                             VDEV_TYPE_SPARE) == 0) {
 714                                                 /* We have a replacing vdev with
 715                                                  * a spare child.  Get the first
 716                                                  * real child of the spare
 717                                                  */
 718                                                 verify(
 719                                                     nvlist_lookup_nvlist_array(
 720                                                         cnv,
 721                                                         ZPOOL_CONFIG_CHILDREN,
 722                                                         &rchild,
 723                                                     &rchildren) == 0);
 724                                                 assert(rchildren >= 2);
 725                                                 cnv = rchild[0];
 726                                         }
 727                                 }
 728
 729                                 verify(nvlist_lookup_string(cnv,
 730                                     ZPOOL_CONFIG_PATH, &path) == 0);
 731
 732                                 /*
 733                                  * If we have a raidz/mirror that combines disks
 734                                  * with files, report it as an error.
 735                                  */
 736                                 if (!dontreport && type != NULL &&
 737                                     strcmp(type, childtype) != 0) {
 738                                         if (ret != NULL)
 739                                                 free(ret);
 740                                         ret = NULL;
 741                                         if (fatal)
 742                                                 vdev_error(gettext(
 743                                                     "mismatched replication "
 744                                                     "level: %s contains both "
 745                                                     "files and devices\n"),
 746                                                     rep.zprl_type);
 747                                         else
 748                                                 return (NULL);
 749                                         dontreport = B_TRUE;
 750                                 }
 751
 752                                 /*
 753                                  * According to stat(2), the value of 'st_size'
 754                                  * is undefined for block devices and character
 755                                  * devices.  But there is no effective way to
 756                                  * determine the real size in userland.
 757                                  *
 758                                  * Instead, we'll take advantage of an
 759                                  * implementation detail of spec_size().  If the
 760                                  * device is currently open, then we (should)
 761                                  * return a valid size.
 762                                  *
 763                                  * If we still don't get a valid size (indicated
 764                                  * by a size of 0 or MAXOFFSET_T), then ignore
 765                                  * this device altogether.
 766                                  */
 767                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 768                                         err = fstat64(fd, &statbuf);
 769                                         (void) close(fd);
 770                                 } else {
 771                                         err = stat64(path, &statbuf);
 772                                 }
 773
 774                                 if (err != 0 ||
 775                                     statbuf.st_size == 0 ||
 776                                     statbuf.st_size == MAXOFFSET_T)
 777                                         continue;
 778
 779                                 size = statbuf.st_size;
 780
 781                                 /*
 782                                  * Also make sure that devices and
 783                                  * slices have a consistent size.  If
 784                                  * they differ by a significant amount
 785                                  * (~16MB) then report an error.
 786                                  */
 787                                 if (!dontreport &&
 788                                     (vdev_size != -1ULL &&
 789                                     (labs(size - vdev_size) >
 790                                     ZPOOL_FUZZ))) {
 791                                         if (ret != NULL)
 792                                                 free(ret);
 793                                         ret = NULL;
 794                                         if (fatal)
 795                                                 vdev_error(gettext(
 796                                                     "%s contains devices of "
 797                                                     "different sizes\n"),
 798                                                     rep.zprl_type);
 799                                         else
 800                                                 return (NULL);
 801                                         dontreport = B_TRUE;
 802                                 }
 803
 804                                 type = childtype;
 805                                 vdev_size = size;
 806                         }
 807                 }
 808
 809                 /*
 810                  * At this point, we have the replication of the last toplevel
 811                  * vdev in 'rep'.  Compare it to 'lastrep' to see if it is
 812                  * different.
 813                  */
 814                 if (lastrep.zprl_type != NULL) {
 815                         if (is_raidz_mirror(&lastrep, &rep, &raidz, &mirror) ||
 816                             is_raidz_mirror(&rep, &lastrep, &raidz, &mirror)) {
 817                                 /*
 818                                  * Accepted raidz and mirror when they can
 819                                  * handle the same number of disk failures.
 820                                  */
 821                                 if (raidz->zprl_parity !=
 822                                     mirror->zprl_children - 1) {
 823                                         if (ret != NULL)
 824                                                 free(ret);
 825                                         ret = NULL;
 826                                         if (fatal)
 827                                                 vdev_error(gettext(
 828                                                     "mismatched replication "
 829                                                     "level: "
 830                                                     "%s and %s vdevs with "
 831                                                     "different redundancy, "
 832                                                     "%llu vs. %llu (%llu-way) "
 833                                                     "are present\n"),
 834                                                     raidz->zprl_type,
 835                                                     mirror->zprl_type,
 836                                                     raidz->zprl_parity,
 837                                                     mirror->zprl_children - 1,
 838                                                     mirror->zprl_children);
 839                                         else
 840                                                 return (NULL);
 841                                 }
 842                         } else if (strcmp(lastrep.zprl_type, rep.zprl_type) !=
 843                             0) {
 844                                 if (ret != NULL)
 845                                         free(ret);
 846                                 ret = NULL;
 847                                 if (fatal)
 848                                         vdev_error(gettext(
 849                                             "mismatched replication level: "
 850                                             "both %s and %s vdevs are "
 851                                             "present\n"),
 852                                             lastrep.zprl_type, rep.zprl_type);
 853                                 else
 854                                         return (NULL);
 855                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 856                                 if (ret)
 857                                         free(ret);
 858                                 ret = NULL;
 859                                 if (fatal)
 860                                         vdev_error(gettext(
 861                                             "mismatched replication level: "
 862                                             "both %llu and %llu device parity "
 863                                             "%s vdevs are present\n"),
 864                                             lastrep.zprl_parity,
 865                                             rep.zprl_parity,
 866                                             rep.zprl_type);
 867                                 else
 868                                         return (NULL);
 869                         } else if (lastrep.zprl_children != rep.zprl_children) {
 870                                 if (ret)
 871                                         free(ret);
 872                                 ret = NULL;
 873                                 if (fatal)
 874                                         vdev_error(gettext(
 875                                             "mismatched replication level: "
 876                                             "both %llu-way and %llu-way %s "
 877                                             "vdevs are present\n"),
 878                                             lastrep.zprl_children,
 879                                             rep.zprl_children,
 880                                             rep.zprl_type);
 881                                 else
 882                                         return (NULL);
 883                         }
 884                 }
 885                 lastrep = rep;
 886         }
 887
 888         if (ret != NULL)
 889                 *ret = rep;
 890
 891         return (ret);
 892 }
 893
 894 /*
 895  * Check the replication level of the vdev spec against the current pool.  Calls
 896  * get_replication() to make sure the new spec is self-consistent.  If the pool
 897  * has a consistent replication level, then we ignore any errors.  Otherwise,
 898  * report any difference between the two.
 899  */
 900 static int
 901 check_replication(nvlist_t *config, nvlist_t *newroot)
 902 {
 903         nvlist_t **child;
 904         uint_t  children;
 905         replication_level_t *current = NULL, *new;
 906         replication_level_t *raidz, *mirror;
 907         int ret;
 908
 909         /*
 910          * If we have a current pool configuration, check to see if it's
 911          * self-consistent.  If not, simply return success.
 912          */
 913         if (config != NULL) {
 914                 nvlist_t *nvroot;
 915
 916                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 917                     &nvroot) == 0);
 918                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
 919                         return (0);
 920         }
 921         /*
 922          * for spares there may be no children, and therefore no
 923          * replication level to check
 924          */
 925         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
 926             &child, &children) != 0) || (children == 0)) {
 927                 free(current);
 928                 return (0);
 929         }
 930
 931         /*
 932          * If all we have is logs then there's no replication level to check.
 933          */
 934         if (num_logs(newroot) == children) {
 935                 free(current);
 936                 return (0);
 937         }
 938
 939         /*
 940          * Get the replication level of the new vdev spec, reporting any
 941          * inconsistencies found.
 942          */
 943         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
 944                 free(current);
 945                 return (-1);
 946         }
 947
 948         /*
 949          * Check to see if the new vdev spec matches the replication level of
 950          * the current pool.
 951          */
 952         ret = 0;
 953         if (current != NULL) {
 954                 if (is_raidz_mirror(current, new, &raidz, &mirror) ||
 955                     is_raidz_mirror(new, current, &raidz, &mirror)) {
 956                         if (raidz->zprl_parity != mirror->zprl_children - 1) {
 957                                 vdev_error(gettext(
 958                                     "mismatched replication level: pool and "
 959                                     "new vdev with different redundancy, %s "
 960                                     "and %s vdevs, %llu vs. %llu (%llu-way)\n"),
 961                                     raidz->zprl_type,
 962                                     mirror->zprl_type,
 963                                     raidz->zprl_parity,
 964                                     mirror->zprl_children - 1,
 965                                     mirror->zprl_children);
 966                                 ret = -1;
 967                         }
 968                 } else if (strcmp(current->zprl_type, new->zprl_type) != 0) {
 969                         vdev_error(gettext(
 970                             "mismatched replication level: pool uses %s "
 971                             "and new vdev is %s\n"),
 972                             current->zprl_type, new->zprl_type);
 973                         ret = -1;
 974                 } else if (current->zprl_parity != new->zprl_parity) {
 975                         vdev_error(gettext(
 976                             "mismatched replication level: pool uses %llu "
 977                             "device parity and new vdev uses %llu\n"),
 978                             current->zprl_parity, new->zprl_parity);
 979                         ret = -1;
 980                 } else if (current->zprl_children != new->zprl_children) {
 981                         vdev_error(gettext(
 982                             "mismatched replication level: pool uses %llu-way "
 983                             "%s and new vdev uses %llu-way %s\n"),
 984                             current->zprl_children, current->zprl_type,
 985                             new->zprl_children, new->zprl_type);
 986                         ret = -1;
 987                 }
 988         }
 989
 990         free(new);
 991         if (current != NULL)
 992                 free(current);
 993
 994         return (ret);
 995 }
 996
 997 #ifdef illumos
 998 /*
 999  * Go through and find any whole disks in the vdev specification, labelling them
1000  * as appropriate.  When constructing the vdev spec, we were unable to open this
1001  * device in order to provide a devid.  Now that we have labelled the disk and
1002  * know the pool slice is valid, we can construct the devid now.
1003  *
1004  * If the disk was already labeled with an EFI label, we will have gotten the
1005  * devid already (because we were able to open the whole disk).  Otherwise, we
1006  * need to get the devid after we label the disk.
1007  */
1008 static int
1009 make_disks(zpool_handle_t *zhp, nvlist_t *nv, zpool_boot_label_t boot_type,
1010     uint64_t boot_size)
1011 {
1012         nvlist_t **child;
1013         uint_t c, children;
1014         char *type, *path, *diskname;
1015         char buf[MAXPATHLEN];
1016         uint64_t wholedisk;
1017         int fd;
1018         int ret;
1019         int slice;
1020         ddi_devid_t devid;
1021         char *minor = NULL, *devid_str = NULL;
1022
1023         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1024
1025         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1026             &child, &children) != 0) {
1027
1028                 if (strcmp(type, VDEV_TYPE_DISK) != 0)
1029                         return (0);
1030
1031                 /*
1032                  * We have a disk device.  Get the path to the device
1033                  * and see if it's a whole disk by appending the backup
1034                  * slice and stat()ing the device.
1035                  */
1036                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
1037
1038                 diskname = strrchr(path, '/');
1039                 assert(diskname != NULL);
1040                 diskname++;
1041
1042                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1043                     &wholedisk) != 0 || !wholedisk) {
1044                         /*
1045                          * This is not whole disk, return error if
1046                          * boot partition creation was requested
1047                          */
1048                         if (boot_type == ZPOOL_CREATE_BOOT_LABEL) {
1049                                 (void) fprintf(stderr,
1050                                     gettext("creating boot partition is only "
1051                                     "supported on whole disk vdevs: %s\n"),
1052                                     diskname);
1053                                 return (-1);
1054                         }
1055                         return (0);
1056                 }
1057
1058                 ret = zpool_label_disk(g_zfs, zhp, diskname, boot_type,
1059                     boot_size, &slice);
1060                 if (ret == -1)
1061                         return (ret);
1062
1063                 /*
1064                  * Fill in the devid, now that we've labeled the disk.
1065                  */
1066                 (void) snprintf(buf, sizeof (buf), "%ss%d", path, slice);
1067                 if ((fd = open(buf, O_RDONLY)) < 0) {
1068                         (void) fprintf(stderr,
1069                             gettext("cannot open '%s': %s\n"),
1070                             buf, strerror(errno));
1071                         return (-1);
1072                 }
1073
1074                 if (devid_get(fd, &devid) == 0) {
1075                         if (devid_get_minor_name(fd, &minor) == 0 &&
1076                             (devid_str = devid_str_encode(devid, minor)) !=
1077                             NULL) {
1078                                 verify(nvlist_add_string(nv,
1079                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
1080                         }
1081                         if (devid_str != NULL)
1082                                 devid_str_free(devid_str);
1083                         if (minor != NULL)
1084                                 devid_str_free(minor);
1085                         devid_free(devid);
1086                 }
1087
1088                 /*
1089                  * Update the path to refer to the pool slice.  The presence of
1090                  * the 'whole_disk' field indicates to the CLI that we should
1091                  * chop off the slice number when displaying the device in
1092                  * future output.
1093                  */
1094                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, buf) == 0);
1095
1096                 (void) close(fd);
1097
1098                 return (0);
1099         }
1100
1101         /* illumos kernel does not support booting from multi-vdev pools. */
1102         if ((boot_type == ZPOOL_CREATE_BOOT_LABEL)) {
1103                 if ((strcmp(type, VDEV_TYPE_ROOT) == 0) && children > 1) {
1104                         (void) fprintf(stderr, gettext("boot pool "
1105                             "can not have more than one vdev\n"));
1106                         return (-1);
1107                 }
1108         }
1109
1110         for (c = 0; c < children; c++) {
1111                 ret = make_disks(zhp, child[c], boot_type, boot_size);
1112                 if (ret != 0)
1113                         return (ret);
1114         }
1115
1116         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1117             &child, &children) == 0)
1118                 for (c = 0; c < children; c++) {
1119                         ret = make_disks(zhp, child[c], boot_type, boot_size);
1120                         if (ret != 0)
1121                                 return (ret);
1122                 }
1123
1124         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1125             &child, &children) == 0)
1126                 for (c = 0; c < children; c++) {
1127                         ret = make_disks(zhp, child[c], boot_type, boot_size);
1128                         if (ret != 0)
1129                                 return (ret);
1130                 }
1131
1132         return (0);
1133 }
1134 #endif  /* illumos */
1135
1136 /*
1137  * Determine if the given path is a hot spare within the given configuration.
1138  */
1139 static boolean_t
1140 is_spare(nvlist_t *config, const char *path)
1141 {
1142         int fd;
1143         pool_state_t state;
1144         char *name = NULL;
1145         nvlist_t *label;
1146         uint64_t guid, spareguid;
1147         nvlist_t *nvroot;
1148         nvlist_t **spares;
1149         uint_t i, nspares;
1150         boolean_t inuse;
1151
1152         if ((fd = open(path, O_RDONLY)) < 0)
1153                 return (B_FALSE);
1154
1155         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
1156             !inuse ||
1157             state != POOL_STATE_SPARE ||
1158             zpool_read_label(fd, &label) != 0) {
1159                 free(name);
1160                 (void) close(fd);
1161                 return (B_FALSE);
1162         }
1163         free(name);
1164         (void) close(fd);
1165
1166         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
1167         nvlist_free(label);
1168
1169         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1170             &nvroot) == 0);
1171         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1172             &spares, &nspares) == 0) {
1173                 for (i = 0; i < nspares; i++) {
1174                         verify(nvlist_lookup_uint64(spares[i],
1175                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
1176                         if (spareguid == guid)
1177                                 return (B_TRUE);
1178                 }
1179         }
1180
1181         return (B_FALSE);
1182 }
1183
1184 /*
1185  * Go through and find any devices that are in use.  We rely on libdiskmgt for
1186  * the majority of this task.
1187  */
1188 static boolean_t
1189 is_device_in_use(nvlist_t *config, nvlist_t *nv, boolean_t force,
1190     boolean_t replacing, boolean_t isspare)
1191 {
1192         nvlist_t **child;
1193         uint_t c, children;
1194         char *type, *path;
1195         int ret = 0;
1196         char buf[MAXPATHLEN];
1197         uint64_t wholedisk;
1198         boolean_t anyinuse = B_FALSE;
1199
1200         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
1201
1202         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1203             &child, &children) != 0) {
1204
1205                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
1206
1207                 /*
1208                  * As a generic check, we look to see if this is a replace of a
1209                  * hot spare within the same pool.  If so, we allow it
1210                  * regardless of what libdiskmgt or zpool_in_use() says.
1211                  */
1212                 if (replacing) {
1213 #ifdef illumos
1214                         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1215                             &wholedisk) == 0 && wholedisk)
1216                                 (void) snprintf(buf, sizeof (buf), "%ss0",
1217                                     path);
1218                         else
1219 #endif
1220                                 (void) strlcpy(buf, path, sizeof (buf));
1221
1222                         if (is_spare(config, buf))
1223                                 return (B_FALSE);
1224                 }
1225
1226                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
1227                         ret = check_device(path, force, isspare);
1228                 else if (strcmp(type, VDEV_TYPE_FILE) == 0)
1229                         ret = check_file(path, force, isspare);
1230
1231                 return (ret != 0);
1232         }
1233
1234         for (c = 0; c < children; c++)
1235                 if (is_device_in_use(config, child[c], force, replacing,
1236                     B_FALSE))
1237                         anyinuse = B_TRUE;
1238
1239         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1240             &child, &children) == 0)
1241                 for (c = 0; c < children; c++)
1242                         if (is_device_in_use(config, child[c], force, replacing,
1243                             B_TRUE))
1244                                 anyinuse = B_TRUE;
1245
1246         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1247             &child, &children) == 0)
1248                 for (c = 0; c < children; c++)
1249                         if (is_device_in_use(config, child[c], force, replacing,
1250                             B_FALSE))
1251                                 anyinuse = B_TRUE;
1252
1253         return (anyinuse);
1254 }
1255
1256 static const char *
1257 is_grouping(const char *type, int *mindev, int *maxdev)
1258 {
1259         if (strncmp(type, "raidz", 5) == 0) {
1260                 const char *p = type + 5;
1261                 char *end;
1262                 long nparity;
1263
1264                 if (*p == '\0') {
1265                         nparity = 1;
1266                 } else if (*p == '0') {
1267                         return (NULL); /* no zero prefixes allowed */
1268                 } else {
1269                         errno = 0;
1270                         nparity = strtol(p, &end, 10);
1271                         if (errno != 0 || nparity < 1 || nparity >= 255 ||
1272                             *end != '\0')
1273                                 return (NULL);
1274                 }
1275
1276                 if (mindev != NULL)
1277                         *mindev = nparity + 1;
1278                 if (maxdev != NULL)
1279                         *maxdev = 255;
1280                 return (VDEV_TYPE_RAIDZ);
1281         }
1282
1283         if (maxdev != NULL)
1284                 *maxdev = INT_MAX;
1285
1286         if (strcmp(type, "mirror") == 0) {
1287                 if (mindev != NULL)
1288                         *mindev = 2;
1289                 return (VDEV_TYPE_MIRROR);
1290         }
1291
1292         if (strcmp(type, "spare") == 0) {
1293                 if (mindev != NULL)
1294                         *mindev = 1;
1295                 return (VDEV_TYPE_SPARE);
1296         }
1297
1298         if (strcmp(type, "log") == 0) {
1299                 if (mindev != NULL)
1300                         *mindev = 1;
1301                 return (VDEV_TYPE_LOG);
1302         }
1303
1304         if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0 ||
1305             strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) {
1306                 if (mindev != NULL)
1307                         *mindev = 1;
1308                 return (type);
1309         }
1310
1311         if (strcmp(type, "cache") == 0) {
1312                 if (mindev != NULL)
1313                         *mindev = 1;
1314                 return (VDEV_TYPE_L2CACHE);
1315         }
1316
1317         return (NULL);
1318 }
1319
1320 /*
1321  * Construct a syntactically valid vdev specification,
1322  * and ensure that all devices and files exist and can be opened.
1323  * Note: we don't bother freeing anything in the error paths
1324  * because the program is just going to exit anyway.
1325  */
1326 nvlist_t *
1327 construct_spec(int argc, char **argv)
1328 {
1329         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
1330         int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
1331         const char *type;
1332         uint64_t is_log, is_special, is_dedup;
1333         boolean_t seen_logs;
1334
1335         top = NULL;
1336         toplevels = 0;
1337         spares = NULL;
1338         l2cache = NULL;
1339         nspares = 0;
1340         nlogs = 0;
1341         nl2cache = 0;
1342         is_log = is_special = is_dedup = B_FALSE;
1343         seen_logs = B_FALSE;
1344
1345         while (argc > 0) {
1346                 nv = NULL;
1347
1348                 /*
1349                  * If it's a mirror or raidz, the subsequent arguments are
1350                  * its leaves -- until we encounter the next mirror or raidz.
1351                  */
1352                 if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
1353                         nvlist_t **child = NULL;
1354                         int c, children = 0;
1355
1356                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1357                                 if (spares != NULL) {
1358                                         (void) fprintf(stderr,
1359                                             gettext("invalid vdev "
1360                                             "specification: 'spare' can be "
1361                                             "specified only once\n"));
1362                                         return (NULL);
1363                                 }
1364                                 is_log = is_special = is_dedup = B_FALSE;
1365                         }
1366
1367                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
1368                                 if (seen_logs) {
1369                                         (void) fprintf(stderr,
1370                                             gettext("invalid vdev "
1371                                             "specification: 'log' can be "
1372                                             "specified only once\n"));
1373                                         return (NULL);
1374                                 }
1375                                 seen_logs = B_TRUE;
1376                                 is_log = B_TRUE;
1377                                 is_special = B_FALSE;
1378                                 is_dedup = B_FALSE;
1379                                 argc--;
1380                                 argv++;
1381                                 /*
1382                                  * A log is not a real grouping device.
1383                                  * We just set is_log and continue.
1384                                  */
1385                                 continue;
1386                         }
1387
1388                         if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0) {
1389                                 is_special = B_TRUE;
1390                                 is_log = B_FALSE;
1391                                 is_dedup = B_FALSE;
1392                                 argc--;
1393                                 argv++;
1394                                 continue;
1395                         }
1396
1397                         if (strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) {
1398                                 is_dedup = B_TRUE;
1399                                 is_log = B_FALSE;
1400                                 is_special = B_FALSE;
1401                                 argc--;
1402                                 argv++;
1403                                 continue;
1404                         }
1405
1406                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1407                                 if (l2cache != NULL) {
1408                                         (void) fprintf(stderr,
1409                                             gettext("invalid vdev "
1410                                             "specification: 'cache' can be "
1411                                             "specified only once\n"));
1412                                         return (NULL);
1413                                 }
1414                                 is_log = is_special = is_dedup = B_FALSE;
1415                         }
1416
1417                         if (is_log || is_special || is_dedup) {
1418                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
1419                                         (void) fprintf(stderr,
1420                                             gettext("invalid vdev "
1421                                             "specification: unsupported '%s' "
1422                                             "device: %s\n"), is_log ? "log" :
1423                                             "special", type);
1424                                         return (NULL);
1425                                 }
1426                                 nlogs++;
1427                         }
1428
1429                         for (c = 1; c < argc; c++) {
1430                                 if (is_grouping(argv[c], NULL, NULL) != NULL)
1431                                         break;
1432                                 children++;
1433                                 child = realloc(child,
1434                                     children * sizeof (nvlist_t *));
1435                                 if (child == NULL)
1436                                         zpool_no_memory();
1437                                 if ((nv = make_leaf_vdev(argv[c], B_FALSE))
1438                                     == NULL)
1439                                         return (NULL);
1440                                 child[children - 1] = nv;
1441                         }
1442
1443                         if (children < mindev) {
1444                                 (void) fprintf(stderr, gettext("invalid vdev "
1445                                     "specification: %s requires at least %d "
1446                                     "devices\n"), argv[0], mindev);
1447                                 return (NULL);
1448                         }
1449
1450                         if (children > maxdev) {
1451                                 (void) fprintf(stderr, gettext("invalid vdev "
1452                                     "specification: %s supports no more than "
1453                                     "%d devices\n"), argv[0], maxdev);
1454                                 return (NULL);
1455                         }
1456
1457                         argc -= c;
1458                         argv += c;
1459
1460                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
1461                                 spares = child;
1462                                 nspares = children;
1463                                 continue;
1464                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
1465                                 l2cache = child;
1466                                 nl2cache = children;
1467                                 continue;
1468                         } else {
1469                                 /* create a top-level vdev with children */
1470                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
1471                                     0) == 0);
1472                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
1473                                     type) == 0);
1474                                 verify(nvlist_add_uint64(nv,
1475                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
1476                                 if (is_log)
1477                                         verify(nvlist_add_string(nv,
1478                                             ZPOOL_CONFIG_ALLOCATION_BIAS,
1479                                             VDEV_ALLOC_BIAS_LOG) == 0);
1480                                 if (is_special) {
1481                                         verify(nvlist_add_string(nv,
1482                                             ZPOOL_CONFIG_ALLOCATION_BIAS,
1483                                             VDEV_ALLOC_BIAS_SPECIAL) == 0);
1484                                 }
1485                                 if (is_dedup) {
1486                                         verify(nvlist_add_string(nv,
1487                                             ZPOOL_CONFIG_ALLOCATION_BIAS,
1488                                             VDEV_ALLOC_BIAS_DEDUP) == 0);
1489                                 }
1490                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
1491                                         verify(nvlist_add_uint64(nv,
1492                                             ZPOOL_CONFIG_NPARITY,
1493                                             mindev - 1) == 0);
1494                                 }
1495                                 verify(nvlist_add_nvlist_array(nv,
1496                                     ZPOOL_CONFIG_CHILDREN, child,
1497                                     children) == 0);
1498
1499                                 for (c = 0; c < children; c++)
1500                                         nvlist_free(child[c]);
1501                                 free(child);
1502                         }
1503                 } else {
1504                         /*
1505                          * We have a device.  Pass off to make_leaf_vdev() to
1506                          * construct the appropriate nvlist describing the vdev.
1507                          */
1508                         if ((nv = make_leaf_vdev(argv[0], is_log)) == NULL)
1509                                 return (NULL);
1510                         if (is_log)
1511                                 nlogs++;
1512                         if (is_special) {
1513                                 verify(nvlist_add_string(nv,
1514                                     ZPOOL_CONFIG_ALLOCATION_BIAS,
1515                                     VDEV_ALLOC_BIAS_SPECIAL) == 0);
1516                         }
1517                         if (is_dedup) {
1518                                 verify(nvlist_add_string(nv,
1519                                     ZPOOL_CONFIG_ALLOCATION_BIAS,
1520                                     VDEV_ALLOC_BIAS_DEDUP) == 0);
1521                         }
1522                         argc--;
1523                         argv++;
1524                 }
1525
1526                 toplevels++;
1527                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1528                 if (top == NULL)
1529                         zpool_no_memory();
1530                 top[toplevels - 1] = nv;
1531         }
1532
1533         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1534                 (void) fprintf(stderr, gettext("invalid vdev "
1535                     "specification: at least one toplevel vdev must be "
1536                     "specified\n"));
1537                 return (NULL);
1538         }
1539
1540         if (seen_logs && nlogs == 0) {
1541                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1542                     "log requires at least 1 device\n"));
1543                 return (NULL);
1544         }
1545
1546         /*
1547          * Finally, create nvroot and add all top-level vdevs to it.
1548          */
1549         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1550         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1551             VDEV_TYPE_ROOT) == 0);
1552         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1553             top, toplevels) == 0);
1554         if (nspares != 0)
1555                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1556                     spares, nspares) == 0);
1557         if (nl2cache != 0)
1558                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1559                     l2cache, nl2cache) == 0);
1560
1561         for (t = 0; t < toplevels; t++)
1562                 nvlist_free(top[t]);
1563         for (t = 0; t < nspares; t++)
1564                 nvlist_free(spares[t]);
1565         for (t = 0; t < nl2cache; t++)
1566                 nvlist_free(l2cache[t]);
1567         if (spares)
1568                 free(spares);
1569         if (l2cache)
1570                 free(l2cache);
1571         free(top);
1572
1573         return (nvroot);
1574 }
1575
1576 nvlist_t *
1577 split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
1578     splitflags_t flags, int argc, char **argv)
1579 {
1580         nvlist_t *newroot = NULL, **child;
1581         uint_t c, children;
1582 #ifdef illumos
1583         zpool_boot_label_t boot_type;
1584 #endif
1585
1586         if (argc > 0) {
1587                 if ((newroot = construct_spec(argc, argv)) == NULL) {
1588                         (void) fprintf(stderr, gettext("Unable to build a "
1589                             "pool from the specified devices\n"));
1590                         return (NULL);
1591                 }
1592
1593 #ifdef illumos
1594                 if (zpool_is_bootable(zhp))
1595                         boot_type = ZPOOL_COPY_BOOT_LABEL;
1596                 else
1597                         boot_type = ZPOOL_NO_BOOT_LABEL;
1598
1599                 if (!flags.dryrun &&
1600                     make_disks(zhp, newroot, boot_type, 0) != 0) {
1601                         nvlist_free(newroot);
1602                         return (NULL);
1603                 }
1604 #endif
1605
1606                 /* avoid any tricks in the spec */
1607                 verify(nvlist_lookup_nvlist_array(newroot,
1608                     ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
1609                 for (c = 0; c < children; c++) {
1610                         char *path;
1611                         const char *type;
1612                         int min, max;
1613
1614                         verify(nvlist_lookup_string(child[c],
1615                             ZPOOL_CONFIG_PATH, &path) == 0);
1616                         if ((type = is_grouping(path, &min, &max)) != NULL) {
1617                                 (void) fprintf(stderr, gettext("Cannot use "
1618                                     "'%s' as a device for splitting\n"), type);
1619                                 nvlist_free(newroot);
1620                                 return (NULL);
1621                         }
1622                 }
1623         }
1624
1625         if (zpool_vdev_split(zhp, newname, &newroot, props, flags) != 0) {
1626                 nvlist_free(newroot);
1627                 return (NULL);
1628         }
1629
1630         return (newroot);
1631 }
1632
1633 static int
1634 num_normal_vdevs(nvlist_t *nvroot)
1635 {
1636         nvlist_t **top;
1637         uint_t t, toplevels, normal = 0;
1638
1639         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1640             &top, &toplevels) == 0);
1641
1642         for (t = 0; t < toplevels; t++) {
1643                 uint64_t log = B_FALSE;
1644
1645                 (void) nvlist_lookup_uint64(top[t], ZPOOL_CONFIG_IS_LOG, &log);
1646                 if (log)
1647                         continue;
1648                 if (nvlist_exists(top[t], ZPOOL_CONFIG_ALLOCATION_BIAS))
1649                         continue;
1650
1651                 normal++;
1652         }
1653
1654         return (normal);
1655 }
1656
1657 /*
1658  * Get and validate the contents of the given vdev specification.  This ensures
1659  * that the nvlist returned is well-formed, that all the devices exist, and that
1660  * they are not currently in use by any other known consumer.  The 'poolconfig'
1661  * parameter is the current configuration of the pool when adding devices
1662  * existing pool, and is used to perform additional checks, such as changing the
1663  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1664  * new pool.  The 'force' flag controls whether devices should be forcefully
1665  * added, even if they appear in use.
1666  */
1667 nvlist_t *
1668 make_root_vdev(zpool_handle_t *zhp, int force, int check_rep,
1669     boolean_t replacing, boolean_t dryrun, zpool_boot_label_t boot_type,
1670     uint64_t boot_size, int argc, char **argv)
1671 {
1672         nvlist_t *newroot;
1673         nvlist_t *poolconfig = NULL;
1674         is_force = force;
1675
1676         /*
1677          * Construct the vdev specification.  If this is successful, we know
1678          * that we have a valid specification, and that all devices can be
1679          * opened.
1680          */
1681         if ((newroot = construct_spec(argc, argv)) == NULL)
1682                 return (NULL);
1683
1684         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
1685                 return (NULL);
1686
1687         /*
1688          * Validate each device to make sure that its not shared with another
1689          * subsystem.  We do this even if 'force' is set, because there are some
1690          * uses (such as a dedicated dump device) that even '-f' cannot
1691          * override.
1692          */
1693         if (is_device_in_use(poolconfig, newroot, force, replacing, B_FALSE)) {
1694                 nvlist_free(newroot);
1695                 return (NULL);
1696         }
1697
1698         /*
1699          * Check the replication level of the given vdevs and report any errors
1700          * found.  We include the existing pool spec, if any, as we need to
1701          * catch changes against the existing replication level.
1702          */
1703         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1704                 nvlist_free(newroot);
1705                 return (NULL);
1706         }
1707
1708 #ifdef illumos
1709         /*
1710          * On pool create the new vdev spec must have one normal vdev.
1711          */
1712         if (poolconfig == NULL && num_normal_vdevs(newroot) == 0) {
1713                 vdev_error(gettext("at least one general top-level vdev must "
1714                     "be specified\n"));
1715                 nvlist_free(newroot);
1716                 return (NULL);
1717         }
1718
1719         /*
1720          * Run through the vdev specification and label any whole disks found.
1721          */
1722         if (!dryrun && make_disks(zhp, newroot, boot_type, boot_size) != 0) {
1723                 nvlist_free(newroot);
1724                 return (NULL);
1725         }
1726 #endif
1727
1728         return (newroot);
1729 }