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 2008 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  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 libdiskmgt, makes 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 <devid.h>
  65 #include <errno.h>
  66 #include <fcntl.h>
  67 #include <libintl.h>
  68 #include <libnvpair.h>
  69 #include <stdio.h>
  70 #include <string.h>
  71 #include <unistd.h>
  72 #include <paths.h>
  73 #include <sys/stat.h>
  74 #include <sys/disk.h>
  75 #include <sys/mntent.h>
  76 #include <libgeom.h>
  77
  78 #include "zpool_util.h"
  79
  80 /*
  81  * For any given vdev specification, we can have multiple errors.  The
  82  * vdev_error() function keeps track of whether we have seen an error yet, and
  83  * prints out a header if its the first error we've seen.
  84  */
  85 boolean_t error_seen;
  86 boolean_t is_force;
  87
  88 /*PRINTFLIKE1*/
  89 static void
  90 vdev_error(const char *fmt, ...)
  91 {
  92         va_list ap;
  93
  94         if (!error_seen) {
  95                 (void) fprintf(stderr, gettext("invalid vdev specification\n"));
  96                 if (!is_force)
  97                         (void) fprintf(stderr, gettext("use '-f' to override "
  98                             "the following errors:\n"));
  99                 else
 100                         (void) fprintf(stderr, gettext("the following errors "
 101                             "must be manually repaired:\n"));
 102                 error_seen = B_TRUE;
 103         }
 104
 105         va_start(ap, fmt);
 106         (void) vfprintf(stderr, fmt, ap);
 107         va_end(ap);
 108 }
 109
 110 /*
 111  * Check that a file is valid.  All we can do in this case is check that it's
 112  * not in use by another pool, and not in use by swap.
 113  */
 114 static int
 115 check_file(const char *file, boolean_t force, boolean_t isspare)
 116 {
 117         char  *name;
 118         int fd;
 119         int ret = 0;
 120         int err;
 121         pool_state_t state;
 122         boolean_t inuse;
 123
 124 #if 0
 125         if (dm_inuse_swap(file, &err)) {
 126                 if (err)
 127                         libdiskmgt_error(err);
 128                 else
 129                         vdev_error(gettext("%s is currently used by swap. "
 130                             "Please see swap(1M).\n"), file);
 131                 return (-1);
 132         }
 133 #endif
 134
 135         if ((fd = open(file, O_RDONLY)) < 0)
 136                 return (0);
 137
 138         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) == 0 && inuse) {
 139                 const char *desc;
 140
 141                 switch (state) {
 142                 case POOL_STATE_ACTIVE:
 143                         desc = gettext("active");
 144                         break;
 145
 146                 case POOL_STATE_EXPORTED:
 147                         desc = gettext("exported");
 148                         break;
 149
 150                 case POOL_STATE_POTENTIALLY_ACTIVE:
 151                         desc = gettext("potentially active");
 152                         break;
 153
 154                 default:
 155                         desc = gettext("unknown");
 156                         break;
 157                 }
 158
 159                 /*
 160                  * Allow hot spares to be shared between pools.
 161                  */
 162                 if (state == POOL_STATE_SPARE && isspare)
 163                         return (0);
 164
 165                 if (state == POOL_STATE_ACTIVE ||
 166                     state == POOL_STATE_SPARE || !force) {
 167                         switch (state) {
 168                         case POOL_STATE_SPARE:
 169                                 vdev_error(gettext("%s is reserved as a hot "
 170                                     "spare for pool %s\n"), file, name);
 171                                 break;
 172                         default:
 173                                 vdev_error(gettext("%s is part of %s pool "
 174                                     "'%s'\n"), file, desc, name);
 175                                 break;
 176                         }
 177                         ret = -1;
 178                 }
 179
 180                 free(name);
 181         }
 182
 183         (void) close(fd);
 184         return (ret);
 185 }
 186
 187 static int
 188 check_provider(const char *name, boolean_t force, boolean_t isspare)
 189 {
 190         char path[MAXPATHLEN];
 191
 192         if (strncmp(name, _PATH_DEV, sizeof(_PATH_DEV) - 1) != 0)
 193                 snprintf(path, sizeof(path), "%s%s", _PATH_DEV, name);
 194         else
 195                 strlcpy(path, name, sizeof(path));
 196
 197         return (check_file(path, force, isspare));
 198 }
 199
 200 /*
 201  * By "whole disk" we mean an entire physical disk (something we can
 202  * label, toggle the write cache on, etc.) as opposed to the full
 203  * capacity of a pseudo-device such as lofi or did.  We act as if we
 204  * are labeling the disk, which should be a pretty good test of whether
 205  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
 206  * it isn't.
 207  */
 208 static boolean_t
 209 is_whole_disk(const char *name)
 210 {
 211         int fd;
 212
 213         fd = g_open(name, 0);
 214         if (fd >= 0) {
 215                 g_close(fd);
 216                 return (B_TRUE);
 217         }
 218         return (B_FALSE);
 219 }
 220
 221 /*
 222  * Create a leaf vdev.  Determine if this is a GEOM provider.
 223  * Valid forms for a leaf vdev are:
 224  *
 225  *      /dev/xxx        Complete path to a GEOM provider
 226  *      xxx             Shorthand for /dev/xxx
 227  */
 228 static nvlist_t *
 229 make_leaf_vdev(const char *arg, uint64_t is_log)
 230 {
 231         char path[MAXPATHLEN];
 232         struct stat64 statbuf;
 233         nvlist_t *vdev = NULL;
 234         char *type = NULL;
 235         boolean_t wholedisk = B_FALSE;
 236
 237         /*
 238          * Determine what type of vdev this is, and put the full path into
 239          * 'path'.  We detect whether this is a device of file afterwards by
 240          * checking the st_mode of the file.
 241          */
 242         if (arg[0] == '/') {
 243                 /*
 244                  * Complete device or file path.  Exact type is determined by
 245                  * examining the file descriptor afterwards.
 246                  */
 247                 wholedisk = is_whole_disk(arg);
 248                 if (!wholedisk && (stat64(arg, &statbuf) != 0)) {
 249                         (void) fprintf(stderr,
 250                             gettext("cannot open '%s': %s\n"),
 251                             arg, strerror(errno));
 252                         return (NULL);
 253                 }
 254
 255                 (void) strlcpy(path, arg, sizeof (path));
 256         } else {
 257                 /*
 258                  * This may be a short path for a device, or it could be total
 259                  * gibberish.  Check to see if it's a known device in
 260                  * /dev/dsk/.  As part of this check, see if we've been given a
 261                  * an entire disk (minus the slice number).
 262                  */
 263                 if (strncmp(arg, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
 264                         strlcpy(path, arg, sizeof (path));
 265                 else
 266                         snprintf(path, sizeof (path), "%s%s", _PATH_DEV, arg);
 267                 wholedisk = is_whole_disk(path);
 268                 if (!wholedisk && (stat64(path, &statbuf) != 0)) {
 269                         /*
 270                          * If we got ENOENT, then the user gave us
 271                          * gibberish, so try to direct them with a
 272                          * reasonable error message.  Otherwise,
 273                          * regurgitate strerror() since it's the best we
 274                          * can do.
 275                          */
 276                         if (errno == ENOENT) {
 277                                 (void) fprintf(stderr,
 278                                     gettext("cannot open '%s': no such "
 279                                     "GEOM provider\n"), arg);
 280                                 (void) fprintf(stderr,
 281                                     gettext("must be a full path or "
 282                                     "shorthand device name\n"));
 283                                 return (NULL);
 284                         } else {
 285                                 (void) fprintf(stderr,
 286                                     gettext("cannot open '%s': %s\n"),
 287                                     path, strerror(errno));
 288                                 return (NULL);
 289                         }
 290                 }
 291         }
 292
 293         /*
 294          * Determine whether this is a device or a file.
 295          */
 296         if (wholedisk) {
 297                 type = VDEV_TYPE_DISK;
 298         } else if (S_ISREG(statbuf.st_mode)) {
 299                 type = VDEV_TYPE_FILE;
 300         } else {
 301                 (void) fprintf(stderr, gettext("cannot use '%s': must be a "
 302                     "GEOM provider or regular file\n"), path);
 303                 return (NULL);
 304         }
 305
 306         /*
 307          * Finally, we have the complete device or file, and we know that it is
 308          * acceptable to use.  Construct the nvlist to describe this vdev.  All
 309          * vdevs have a 'path' element, and devices also have a 'devid' element.
 310          */
 311         verify(nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) == 0);
 312         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 313         verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 314         verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 315         if (strcmp(type, VDEV_TYPE_DISK) == 0)
 316                 verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 317                     (uint64_t)B_FALSE) == 0);
 318
 319         /*
 320          * For a whole disk, defer getting its devid until after labeling it.
 321          */
 322         if (1 || (S_ISBLK(statbuf.st_mode) && !wholedisk)) {
 323                 /*
 324                  * Get the devid for the device.
 325                  */
 326                 int fd;
 327                 ddi_devid_t devid;
 328                 char *minor = NULL, *devid_str = NULL;
 329
 330                 if ((fd = open(path, O_RDONLY)) < 0) {
 331                         (void) fprintf(stderr, gettext("cannot open '%s': "
 332                             "%s\n"), path, strerror(errno));
 333                         nvlist_free(vdev);
 334                         return (NULL);
 335                 }
 336
 337                 if (devid_get(fd, &devid) == 0) {
 338                         if (devid_get_minor_name(fd, &minor) == 0 &&
 339                             (devid_str = devid_str_encode(devid, minor)) !=
 340                             NULL) {
 341                                 verify(nvlist_add_string(vdev,
 342                                     ZPOOL_CONFIG_DEVID, devid_str) == 0);
 343                         }
 344                         if (devid_str != NULL)
 345                                 devid_str_free(devid_str);
 346                         if (minor != NULL)
 347                                 devid_str_free(minor);
 348                         devid_free(devid);
 349                 }
 350
 351                 (void) close(fd);
 352         }
 353
 354         return (vdev);
 355 }
 356
 357 /*
 358  * Go through and verify the replication level of the pool is consistent.
 359  * Performs the following checks:
 360  *
 361  *      For the new spec, verifies that devices in mirrors and raidz are the
 362  *      same size.
 363  *
 364  *      If the current configuration already has inconsistent replication
 365  *      levels, ignore any other potential problems in the new spec.
 366  *
 367  *      Otherwise, make sure that the current spec (if there is one) and the new
 368  *      spec have consistent replication levels.
 369  */
 370 typedef struct replication_level {
 371         char *zprl_type;
 372         uint64_t zprl_children;
 373         uint64_t zprl_parity;
 374 } replication_level_t;
 375
 376 #define ZPOOL_FUZZ      (16 * 1024 * 1024)
 377
 378 /*
 379  * Given a list of toplevel vdevs, return the current replication level.  If
 380  * the config is inconsistent, then NULL is returned.  If 'fatal' is set, then
 381  * an error message will be displayed for each self-inconsistent vdev.
 382  */
 383 static replication_level_t *
 384 get_replication(nvlist_t *nvroot, boolean_t fatal)
 385 {
 386         nvlist_t **top;
 387         uint_t t, toplevels;
 388         nvlist_t **child;
 389         uint_t c, children;
 390         nvlist_t *nv;
 391         char *type;
 392         replication_level_t lastrep, rep, *ret;
 393         boolean_t dontreport;
 394
 395         ret = safe_malloc(sizeof (replication_level_t));
 396
 397         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 398             &top, &toplevels) == 0);
 399
 400         lastrep.zprl_type = NULL;
 401         for (t = 0; t < toplevels; t++) {
 402                 uint64_t is_log = B_FALSE;
 403
 404                 nv = top[t];
 405
 406                 /*
 407                  * For separate logs we ignore the top level vdev replication
 408                  * constraints.
 409                  */
 410                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
 411                 if (is_log)
 412                         continue;
 413
 414                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE,
 415                     &type) == 0);
 416                 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 417                     &child, &children) != 0) {
 418                         /*
 419                          * This is a 'file' or 'disk' vdev.
 420                          */
 421                         rep.zprl_type = type;
 422                         rep.zprl_children = 1;
 423                         rep.zprl_parity = 0;
 424                 } else {
 425                         uint64_t vdev_size;
 426
 427                         /*
 428                          * This is a mirror or RAID-Z vdev.  Go through and make
 429                          * sure the contents are all the same (files vs. disks),
 430                          * keeping track of the number of elements in the
 431                          * process.
 432                          *
 433                          * We also check that the size of each vdev (if it can
 434                          * be determined) is the same.
 435                          */
 436                         rep.zprl_type = type;
 437                         rep.zprl_children = 0;
 438
 439                         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 440                                 verify(nvlist_lookup_uint64(nv,
 441                                     ZPOOL_CONFIG_NPARITY,
 442                                     &rep.zprl_parity) == 0);
 443                                 assert(rep.zprl_parity != 0);
 444                         } else {
 445                                 rep.zprl_parity = 0;
 446                         }
 447
 448                         /*
 449                          * The 'dontreport' variable indicates that we've
 450                          * already reported an error for this spec, so don't
 451                          * bother doing it again.
 452                          */
 453                         type = NULL;
 454                         dontreport = 0;
 455                         vdev_size = -1ULL;
 456                         for (c = 0; c < children; c++) {
 457                                 nvlist_t *cnv = child[c];
 458                                 char *path;
 459                                 struct stat64 statbuf;
 460                                 uint64_t size = -1ULL;
 461                                 char *childtype;
 462                                 int fd, err;
 463
 464                                 rep.zprl_children++;
 465
 466                                 verify(nvlist_lookup_string(cnv,
 467                                     ZPOOL_CONFIG_TYPE, &childtype) == 0);
 468
 469                                 /*
 470                                  * If this is a replacing or spare vdev, then
 471                                  * get the real first child of the vdev.
 472                                  */
 473                                 if (strcmp(childtype,
 474                                     VDEV_TYPE_REPLACING) == 0 ||
 475                                     strcmp(childtype, VDEV_TYPE_SPARE) == 0) {
 476                                         nvlist_t **rchild;
 477                                         uint_t rchildren;
 478
 479                                         verify(nvlist_lookup_nvlist_array(cnv,
 480                                             ZPOOL_CONFIG_CHILDREN, &rchild,
 481                                             &rchildren) == 0);
 482                                         assert(rchildren == 2);
 483                                         cnv = rchild[0];
 484
 485                                         verify(nvlist_lookup_string(cnv,
 486                                             ZPOOL_CONFIG_TYPE,
 487                                             &childtype) == 0);
 488                                 }
 489
 490                                 verify(nvlist_lookup_string(cnv,
 491                                     ZPOOL_CONFIG_PATH, &path) == 0);
 492
 493                                 /*
 494                                  * If we have a raidz/mirror that combines disks
 495                                  * with files, report it as an error.
 496                                  */
 497                                 if (!dontreport && type != NULL &&
 498                                     strcmp(type, childtype) != 0) {
 499                                         if (ret != NULL)
 500                                                 free(ret);
 501                                         ret = NULL;
 502                                         if (fatal)
 503                                                 vdev_error(gettext(
 504                                                     "mismatched replication "
 505                                                     "level: %s contains both "
 506                                                     "files and devices\n"),
 507                                                     rep.zprl_type);
 508                                         else
 509                                                 return (NULL);
 510                                         dontreport = B_TRUE;
 511                                 }
 512
 513                                 /*
 514                                  * According to stat(2), the value of 'st_size'
 515                                  * is undefined for block devices and character
 516                                  * devices.  But there is no effective way to
 517                                  * determine the real size in userland.
 518                                  *
 519                                  * Instead, we'll take advantage of an
 520                                  * implementation detail of spec_size().  If the
 521                                  * device is currently open, then we (should)
 522                                  * return a valid size.
 523                                  *
 524                                  * If we still don't get a valid size (indicated
 525                                  * by a size of 0 or MAXOFFSET_T), then ignore
 526                                  * this device altogether.
 527                                  */
 528                                 if ((fd = open(path, O_RDONLY)) >= 0) {
 529                                         err = fstat64(fd, &statbuf);
 530                                         if (err == 0 &&
 531                                             S_ISCHR(statbuf.st_mode)) {
 532                                                 err = ioctl(fd, DIOCGMEDIASIZE,
 533                                                     &statbuf.st_size);
 534                                         }
 535                                         (void) close(fd);
 536                                 } else {
 537                                         err = stat64(path, &statbuf);
 538                                 }
 539                                 if (err != 0 || statbuf.st_size == 0)
 540                                         continue;
 541
 542                                 size = statbuf.st_size;
 543
 544                                 /*
 545                                  * Also make sure that devices and
 546                                  * slices have a consistent size.  If
 547                                  * they differ by a significant amount
 548                                  * (~16MB) then report an error.
 549                                  */
 550                                 if (!dontreport &&
 551                                     (vdev_size != -1ULL &&
 552                                     (labs(size - vdev_size) >
 553                                     ZPOOL_FUZZ))) {
 554                                         if (ret != NULL)
 555                                                 free(ret);
 556                                         ret = NULL;
 557                                         if (fatal)
 558                                                 vdev_error(gettext(
 559                                                     "%s contains devices of "
 560                                                     "different sizes\n"),
 561                                                     rep.zprl_type);
 562                                         else
 563                                                 return (NULL);
 564                                         dontreport = B_TRUE;
 565                                 }
 566
 567                                 type = childtype;
 568                                 vdev_size = size;
 569                         }
 570                 }
 571
 572                 /*
 573                  * At this point, we have the replication of the last toplevel
 574                  * vdev in 'rep'.  Compare it to 'lastrep' to see if its
 575                  * different.
 576                  */
 577                 if (lastrep.zprl_type != NULL) {
 578                         if (strcmp(lastrep.zprl_type, rep.zprl_type) != 0) {
 579                                 if (ret != NULL)
 580                                         free(ret);
 581                                 ret = NULL;
 582                                 if (fatal)
 583                                         vdev_error(gettext(
 584                                             "mismatched replication level: "
 585                                             "both %s and %s vdevs are "
 586                                             "present\n"),
 587                                             lastrep.zprl_type, rep.zprl_type);
 588                                 else
 589                                         return (NULL);
 590                         } else if (lastrep.zprl_parity != rep.zprl_parity) {
 591                                 if (ret)
 592                                         free(ret);
 593                                 ret = NULL;
 594                                 if (fatal)
 595                                         vdev_error(gettext(
 596                                             "mismatched replication level: "
 597                                             "both %llu and %llu device parity "
 598                                             "%s vdevs are present\n"),
 599                                             lastrep.zprl_parity,
 600                                             rep.zprl_parity,
 601                                             rep.zprl_type);
 602                                 else
 603                                         return (NULL);
 604                         } else if (lastrep.zprl_children != rep.zprl_children) {
 605                                 if (ret)
 606                                         free(ret);
 607                                 ret = NULL;
 608                                 if (fatal)
 609                                         vdev_error(gettext(
 610                                             "mismatched replication level: "
 611                                             "both %llu-way and %llu-way %s "
 612                                             "vdevs are present\n"),
 613                                             lastrep.zprl_children,
 614                                             rep.zprl_children,
 615                                             rep.zprl_type);
 616                                 else
 617                                         return (NULL);
 618                         }
 619                 }
 620                 lastrep = rep;
 621         }
 622
 623         if (ret != NULL)
 624                 *ret = rep;
 625
 626         return (ret);
 627 }
 628
 629 /*
 630  * Check the replication level of the vdev spec against the current pool.  Calls
 631  * get_replication() to make sure the new spec is self-consistent.  If the pool
 632  * has a consistent replication level, then we ignore any errors.  Otherwise,
 633  * report any difference between the two.
 634  */
 635 static int
 636 check_replication(nvlist_t *config, nvlist_t *newroot)
 637 {
 638         nvlist_t **child;
 639         uint_t  children;
 640         replication_level_t *current = NULL, *new;
 641         int ret;
 642
 643         /*
 644          * If we have a current pool configuration, check to see if it's
 645          * self-consistent.  If not, simply return success.
 646          */
 647         if (config != NULL) {
 648                 nvlist_t *nvroot;
 649
 650                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 651                     &nvroot) == 0);
 652                 if ((current = get_replication(nvroot, B_FALSE)) == NULL)
 653                         return (0);
 654         }
 655         /*
 656          * for spares there may be no children, and therefore no
 657          * replication level to check
 658          */
 659         if ((nvlist_lookup_nvlist_array(newroot, ZPOOL_CONFIG_CHILDREN,
 660             &child, &children) != 0) || (children == 0)) {
 661                 free(current);
 662                 return (0);
 663         }
 664
 665         /*
 666          * If all we have is logs then there's no replication level to check.
 667          */
 668         if (num_logs(newroot) == children) {
 669                 free(current);
 670                 return (0);
 671         }
 672
 673         /*
 674          * Get the replication level of the new vdev spec, reporting any
 675          * inconsistencies found.
 676          */
 677         if ((new = get_replication(newroot, B_TRUE)) == NULL) {
 678                 free(current);
 679                 return (-1);
 680         }
 681
 682         /*
 683          * Check to see if the new vdev spec matches the replication level of
 684          * the current pool.
 685          */
 686         ret = 0;
 687         if (current != NULL) {
 688                 if (strcmp(current->zprl_type, new->zprl_type) != 0) {
 689                         vdev_error(gettext(
 690                             "mismatched replication level: pool uses %s "
 691                             "and new vdev is %s\n"),
 692                             current->zprl_type, new->zprl_type);
 693                         ret = -1;
 694                 } else if (current->zprl_parity != new->zprl_parity) {
 695                         vdev_error(gettext(
 696                             "mismatched replication level: pool uses %llu "
 697                             "device parity and new vdev uses %llu\n"),
 698                             current->zprl_parity, new->zprl_parity);
 699                         ret = -1;
 700                 } else if (current->zprl_children != new->zprl_children) {
 701                         vdev_error(gettext(
 702                             "mismatched replication level: pool uses %llu-way "
 703                             "%s and new vdev uses %llu-way %s\n"),
 704                             current->zprl_children, current->zprl_type,
 705                             new->zprl_children, new->zprl_type);
 706                         ret = -1;
 707                 }
 708         }
 709
 710         free(new);
 711         if (current != NULL)
 712                 free(current);
 713
 714         return (ret);
 715 }
 716
 717 /*
 718  * Determine if the given path is a hot spare within the given configuration.
 719  */
 720 static boolean_t
 721 is_spare(nvlist_t *config, const char *path)
 722 {
 723         int fd;
 724         pool_state_t state;
 725         char *name = NULL;
 726         nvlist_t *label;
 727         uint64_t guid, spareguid;
 728         nvlist_t *nvroot;
 729         nvlist_t **spares;
 730         uint_t i, nspares;
 731         boolean_t inuse;
 732
 733         if ((fd = open(path, O_RDONLY)) < 0)
 734                 return (B_FALSE);
 735
 736         if (zpool_in_use(g_zfs, fd, &state, &name, &inuse) != 0 ||
 737             !inuse ||
 738             state != POOL_STATE_SPARE ||
 739             zpool_read_label(fd, &label) != 0) {
 740                 free(name);
 741                 (void) close(fd);
 742                 return (B_FALSE);
 743         }
 744         free(name);
 745
 746         (void) close(fd);
 747         verify(nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) == 0);
 748         nvlist_free(label);
 749
 750         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 751             &nvroot) == 0);
 752         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 753             &spares, &nspares) == 0) {
 754                 for (i = 0; i < nspares; i++) {
 755                         verify(nvlist_lookup_uint64(spares[i],
 756                             ZPOOL_CONFIG_GUID, &spareguid) == 0);
 757                         if (spareguid == guid)
 758                                 return (B_TRUE);
 759                 }
 760         }
 761
 762         return (B_FALSE);
 763 }
 764
 765 /*
 766  * Go through and find any devices that are in use.  We rely on libdiskmgt for
 767  * the majority of this task.
 768  */
 769 static int
 770 check_in_use(nvlist_t *config, nvlist_t *nv, int force, int isreplacing,
 771     int isspare)
 772 {
 773         nvlist_t **child;
 774         uint_t c, children;
 775         char *type, *path;
 776         int ret;
 777         char buf[MAXPATHLEN];
 778         uint64_t wholedisk;
 779
 780         verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
 781
 782         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 783             &child, &children) != 0) {
 784
 785                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0);
 786
 787                 /*
 788                  * As a generic check, we look to see if this is a replace of a
 789                  * hot spare within the same pool.  If so, we allow it
 790                  * regardless of what libdiskmgt or zpool_in_use() says.
 791                  */
 792                 if (isreplacing) {
 793                         (void) strlcpy(buf, path, sizeof (buf));
 794                         if (is_spare(config, buf))
 795                                 return (0);
 796                 }
 797
 798                 if (strcmp(type, VDEV_TYPE_DISK) == 0)
 799                         ret = check_provider(path, force, isspare);
 800
 801                 if (strcmp(type, VDEV_TYPE_FILE) == 0)
 802                         ret = check_file(path, force, isspare);
 803
 804                 return (ret);
 805         }
 806
 807         for (c = 0; c < children; c++)
 808                 if ((ret = check_in_use(config, child[c], force,
 809                     isreplacing, B_FALSE)) != 0)
 810                         return (ret);
 811
 812         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
 813             &child, &children) == 0)
 814                 for (c = 0; c < children; c++)
 815                         if ((ret = check_in_use(config, child[c], force,
 816                             isreplacing, B_TRUE)) != 0)
 817                                 return (ret);
 818
 819         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
 820             &child, &children) == 0)
 821                 for (c = 0; c < children; c++)
 822                         if ((ret = check_in_use(config, child[c], force,
 823                             isreplacing, B_FALSE)) != 0)
 824                                 return (ret);
 825
 826         return (0);
 827 }
 828
 829 static const char *
 830 is_grouping(const char *type, int *mindev)
 831 {
 832         if (strcmp(type, "raidz") == 0 || strcmp(type, "raidz1") == 0) {
 833                 if (mindev != NULL)
 834                         *mindev = 2;
 835                 return (VDEV_TYPE_RAIDZ);
 836         }
 837
 838         if (strcmp(type, "raidz2") == 0) {
 839                 if (mindev != NULL)
 840                         *mindev = 3;
 841                 return (VDEV_TYPE_RAIDZ);
 842         }
 843
 844         if (strcmp(type, "mirror") == 0) {
 845                 if (mindev != NULL)
 846                         *mindev = 2;
 847                 return (VDEV_TYPE_MIRROR);
 848         }
 849
 850         if (strcmp(type, "spare") == 0) {
 851                 if (mindev != NULL)
 852                         *mindev = 1;
 853                 return (VDEV_TYPE_SPARE);
 854         }
 855
 856         if (strcmp(type, "log") == 0) {
 857                 if (mindev != NULL)
 858                         *mindev = 1;
 859                 return (VDEV_TYPE_LOG);
 860         }
 861
 862         if (strcmp(type, "cache") == 0) {
 863                 if (mindev != NULL)
 864                         *mindev = 1;
 865                 return (VDEV_TYPE_L2CACHE);
 866         }
 867
 868         return (NULL);
 869 }
 870
 871 /*
 872  * Construct a syntactically valid vdev specification,
 873  * and ensure that all devices and files exist and can be opened.
 874  * Note: we don't bother freeing anything in the error paths
 875  * because the program is just going to exit anyway.
 876  */
 877 nvlist_t *
 878 construct_spec(int argc, char **argv)
 879 {
 880         nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
 881         int t, toplevels, mindev, nspares, nlogs, nl2cache;
 882         const char *type;
 883         uint64_t is_log;
 884         boolean_t seen_logs;
 885
 886         top = NULL;
 887         toplevels = 0;
 888         spares = NULL;
 889         l2cache = NULL;
 890         nspares = 0;
 891         nlogs = 0;
 892         nl2cache = 0;
 893         is_log = B_FALSE;
 894         seen_logs = B_FALSE;
 895
 896         while (argc > 0) {
 897                 nv = NULL;
 898
 899                 /*
 900                  * If it's a mirror or raidz, the subsequent arguments are
 901                  * its leaves -- until we encounter the next mirror or raidz.
 902                  */
 903                 if ((type = is_grouping(argv[0], &mindev)) != NULL) {
 904                         nvlist_t **child = NULL;
 905                         int c, children = 0;
 906
 907                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
 908                                 if (spares != NULL) {
 909                                         (void) fprintf(stderr,
 910                                             gettext("invalid vdev "
 911                                             "specification: 'spare' can be "
 912                                             "specified only once\n"));
 913                                         return (NULL);
 914                                 }
 915                                 is_log = B_FALSE;
 916                         }
 917
 918                         if (strcmp(type, VDEV_TYPE_LOG) == 0) {
 919                                 if (seen_logs) {
 920                                         (void) fprintf(stderr,
 921                                             gettext("invalid vdev "
 922                                             "specification: 'log' can be "
 923                                             "specified only once\n"));
 924                                         return (NULL);
 925                                 }
 926                                 seen_logs = B_TRUE;
 927                                 is_log = B_TRUE;
 928                                 argc--;
 929                                 argv++;
 930                                 /*
 931                                  * A log is not a real grouping device.
 932                                  * We just set is_log and continue.
 933                                  */
 934                                 continue;
 935                         }
 936
 937                         if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
 938                                 if (l2cache != NULL) {
 939                                         (void) fprintf(stderr,
 940                                             gettext("invalid vdev "
 941                                             "specification: 'cache' can be "
 942                                             "specified only once\n"));
 943                                         return (NULL);
 944                                 }
 945                                 is_log = B_FALSE;
 946                         }
 947
 948                         if (is_log) {
 949                                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
 950                                         (void) fprintf(stderr,
 951                                             gettext("invalid vdev "
 952                                             "specification: unsupported 'log' "
 953                                             "device: %s\n"), type);
 954                                         return (NULL);
 955                                 }
 956                                 nlogs++;
 957                         }
 958
 959                         for (c = 1; c < argc; c++) {
 960                                 if (is_grouping(argv[c], NULL) != NULL)
 961                                         break;
 962                                 children++;
 963                                 child = realloc(child,
 964                                     children * sizeof (nvlist_t *));
 965                                 if (child == NULL)
 966                                         zpool_no_memory();
 967                                 if ((nv = make_leaf_vdev(argv[c], B_FALSE))
 968                                     == NULL)
 969                                         return (NULL);
 970                                 child[children - 1] = nv;
 971                         }
 972
 973                         if (children < mindev) {
 974                                 (void) fprintf(stderr, gettext("invalid vdev "
 975                                     "specification: %s requires at least %d "
 976                                     "devices\n"), argv[0], mindev);
 977                                 return (NULL);
 978                         }
 979
 980                         argc -= c;
 981                         argv += c;
 982
 983                         if (strcmp(type, VDEV_TYPE_SPARE) == 0) {
 984                                 spares = child;
 985                                 nspares = children;
 986                                 continue;
 987                         } else if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
 988                                 l2cache = child;
 989                                 nl2cache = children;
 990                                 continue;
 991                         } else {
 992                                 verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
 993                                     0) == 0);
 994                                 verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
 995                                     type) == 0);
 996                                 verify(nvlist_add_uint64(nv,
 997                                     ZPOOL_CONFIG_IS_LOG, is_log) == 0);
 998                                 if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 999                                         verify(nvlist_add_uint64(nv,
1000                                             ZPOOL_CONFIG_NPARITY,
1001                                             mindev - 1) == 0);
1002                                 }
1003                                 verify(nvlist_add_nvlist_array(nv,
1004                                     ZPOOL_CONFIG_CHILDREN, child,
1005                                     children) == 0);
1006
1007                                 for (c = 0; c < children; c++)
1008                                         nvlist_free(child[c]);
1009                                 free(child);
1010                         }
1011                 } else {
1012                         /*
1013                          * We have a device.  Pass off to make_leaf_vdev() to
1014                          * construct the appropriate nvlist describing the vdev.
1015                          */
1016                         if ((nv = make_leaf_vdev(argv[0], is_log)) == NULL)
1017                                 return (NULL);
1018                         if (is_log)
1019                                 nlogs++;
1020                         argc--;
1021                         argv++;
1022                 }
1023
1024                 toplevels++;
1025                 top = realloc(top, toplevels * sizeof (nvlist_t *));
1026                 if (top == NULL)
1027                         zpool_no_memory();
1028                 top[toplevels - 1] = nv;
1029         }
1030
1031         if (toplevels == 0 && nspares == 0 && nl2cache == 0) {
1032                 (void) fprintf(stderr, gettext("invalid vdev "
1033                     "specification: at least one toplevel vdev must be "
1034                     "specified\n"));
1035                 return (NULL);
1036         }
1037
1038         if (seen_logs && nlogs == 0) {
1039                 (void) fprintf(stderr, gettext("invalid vdev specification: "
1040                     "log requires at least 1 device\n"));
1041                 return (NULL);
1042         }
1043
1044         /*
1045          * Finally, create nvroot and add all top-level vdevs to it.
1046          */
1047         verify(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) == 0);
1048         verify(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
1049             VDEV_TYPE_ROOT) == 0);
1050         verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1051             top, toplevels) == 0);
1052         if (nspares != 0)
1053                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1054                     spares, nspares) == 0);
1055         if (nl2cache != 0)
1056                 verify(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1057                     l2cache, nl2cache) == 0);
1058
1059         for (t = 0; t < toplevels; t++)
1060                 nvlist_free(top[t]);
1061         for (t = 0; t < nspares; t++)
1062                 nvlist_free(spares[t]);
1063         for (t = 0; t < nl2cache; t++)
1064                 nvlist_free(l2cache[t]);
1065         if (spares)
1066                 free(spares);
1067         if (l2cache)
1068                 free(l2cache);
1069         free(top);
1070
1071         return (nvroot);
1072 }
1073
1074
1075 /*
1076  * Get and validate the contents of the given vdev specification.  This ensures
1077  * that the nvlist returned is well-formed, that all the devices exist, and that
1078  * they are not currently in use by any other known consumer.  The 'poolconfig'
1079  * parameter is the current configuration of the pool when adding devices
1080  * existing pool, and is used to perform additional checks, such as changing the
1081  * replication level of the pool.  It can be 'NULL' to indicate that this is a
1082  * new pool.  The 'force' flag controls whether devices should be forcefully
1083  * added, even if they appear in use.
1084  */
1085 nvlist_t *
1086 make_root_vdev(zpool_handle_t *zhp, int force, int check_rep,
1087     boolean_t isreplacing, boolean_t dryrun, int argc, char **argv)
1088 {
1089         nvlist_t *newroot;
1090         nvlist_t *poolconfig = NULL;
1091         is_force = force;
1092
1093         /*
1094          * Construct the vdev specification.  If this is successful, we know
1095          * that we have a valid specification, and that all devices can be
1096          * opened.
1097          */
1098         if ((newroot = construct_spec(argc, argv)) == NULL)
1099                 return (NULL);
1100
1101         if (zhp && ((poolconfig = zpool_get_config(zhp, NULL)) == NULL))
1102                 return (NULL);
1103
1104         /*
1105          * Validate each device to make sure that its not shared with another
1106          * subsystem.  We do this even if 'force' is set, because there are some
1107          * uses (such as a dedicated dump device) that even '-f' cannot
1108          * override.
1109          */
1110         if (check_in_use(poolconfig, newroot, force, isreplacing,
1111             B_FALSE) != 0) {
1112                 nvlist_free(newroot);
1113                 return (NULL);
1114         }
1115
1116         /*
1117          * Check the replication level of the given vdevs and report any errors
1118          * found.  We include the existing pool spec, if any, as we need to
1119          * catch changes against the existing replication level.
1120          */
1121         if (check_rep && check_replication(poolconfig, newroot) != 0) {
1122                 nvlist_free(newroot);
1123                 return (NULL);
1124         }
1125
1126         return (newroot);
1127 }