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  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  24  * Copyright (c) 2012 by Delphix. All rights reserved.
  25  */
  26 
  27 /*
  28  * Pool import support functions.
  29  *
  30  * To import a pool, we rely on reading the configuration information from the
  31  * ZFS label of each device.  If we successfully read the label, then we
  32  * organize the configuration information in the following hierarchy:
  33  *
  34  *      pool guid -> toplevel vdev guid -> label txg
  35  *
  36  * Duplicate entries matching this same tuple will be discarded.  Once we have
  37  * examined every device, we pick the best label txg config for each toplevel
  38  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  39  * update any paths that have changed.  Finally, we attempt to import the pool
  40  * using our derived config, and record the results.
  41  */
  42 
  43 #include <ctype.h>
  44 #include <devid.h>
  45 #include <dirent.h>
  46 #include <errno.h>
  47 #include <libintl.h>
  48 #include <stddef.h>
  49 #include <stdlib.h>
  50 #include <string.h>
  51 #include <sys/stat.h>
  52 #include <unistd.h>
  53 #include <fcntl.h>
  54 #include <sys/vtoc.h>
  55 #include <sys/dktp/fdisk.h>
  56 #include <sys/efi_partition.h>
  57 #include <thread_pool.h>
  58 
  59 #include <sys/vdev_impl.h>
  60 
  61 #include "libzfs.h"
  62 #include "libzfs_impl.h"
  63 
  64 /*
  65  * Intermediate structures used to gather configuration information.
  66  */
  67 typedef struct config_entry {
  68         uint64_t                ce_txg;
  69         nvlist_t                *ce_config;
  70         struct config_entry     *ce_next;
  71 } config_entry_t;
  72 
  73 typedef struct vdev_entry {
  74         uint64_t                ve_guid;
  75         config_entry_t          *ve_configs;
  76         struct vdev_entry       *ve_next;
  77 } vdev_entry_t;
  78 
  79 typedef struct pool_entry {
  80         uint64_t                pe_guid;
  81         vdev_entry_t            *pe_vdevs;
  82         struct pool_entry       *pe_next;
  83 } pool_entry_t;
  84 
  85 typedef struct name_entry {
  86         char                    *ne_name;
  87         uint64_t                ne_guid;
  88         struct name_entry       *ne_next;
  89 } name_entry_t;
  90 
  91 typedef struct pool_list {
  92         pool_entry_t            *pools;
  93         name_entry_t            *names;
  94 } pool_list_t;
  95 
  96 static char *
  97 get_devid(const char *path)
  98 {
  99         int fd;
 100         ddi_devid_t devid;
 101         char *minor, *ret;
 102 
 103         if ((fd = open(path, O_RDONLY)) < 0)
 104                 return (NULL);
 105 
 106         minor = NULL;
 107         ret = NULL;
 108         if (devid_get(fd, &devid) == 0) {
 109                 if (devid_get_minor_name(fd, &minor) == 0)
 110                         ret = devid_str_encode(devid, minor);
 111                 if (minor != NULL)
 112                         devid_str_free(minor);
 113                 devid_free(devid);
 114         }
 115         (void) close(fd);
 116 
 117         return (ret);
 118 }
 119 
 120 
 121 /*
 122  * Go through and fix up any path and/or devid information for the given vdev
 123  * configuration.
 124  */
 125 static int
 126 fix_paths(nvlist_t *nv, name_entry_t *names)
 127 {
 128         nvlist_t **child;
 129         uint_t c, children;
 130         uint64_t guid;
 131         name_entry_t *ne, *best;
 132         char *path, *devid;
 133         int matched;
 134 
 135         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 136             &child, &children) == 0) {
 137                 for (c = 0; c < children; c++)
 138                         if (fix_paths(child[c], names) != 0)
 139                                 return (-1);
 140                 return (0);
 141         }
 142 
 143         /*
 144          * This is a leaf (file or disk) vdev.  In either case, go through
 145          * the name list and see if we find a matching guid.  If so, replace
 146          * the path and see if we can calculate a new devid.
 147          *
 148          * There may be multiple names associated with a particular guid, in
 149          * which case we have overlapping slices or multiple paths to the same
 150          * disk.  If this is the case, then we want to pick the path that is
 151          * the most similar to the original, where "most similar" is the number
 152          * of matching characters starting from the end of the path.  This will
 153          * preserve slice numbers even if the disks have been reorganized, and
 154          * will also catch preferred disk names if multiple paths exist.
 155          */
 156         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 157         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 158                 path = NULL;
 159 
 160         matched = 0;
 161         best = NULL;
 162         for (ne = names; ne != NULL; ne = ne->ne_next) {
 163                 if (ne->ne_guid == guid) {
 164                         const char *src, *dst;
 165                         int count;
 166 
 167                         if (path == NULL) {
 168                                 best = ne;
 169                                 break;
 170                         }
 171 
 172                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 173                         dst = path + strlen(path) - 1;
 174                         for (count = 0; src >= ne->ne_name && dst >= path;
 175                             src--, dst--, count++)
 176                                 if (*src != *dst)
 177                                         break;
 178 
 179                         /*
 180                          * At this point, 'count' is the number of characters
 181                          * matched from the end.
 182                          */
 183                         if (count > matched || best == NULL) {
 184                                 best = ne;
 185                                 matched = count;
 186                         }
 187                 }
 188         }
 189 
 190         if (best == NULL)
 191                 return (0);
 192 
 193         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 194                 return (-1);
 195 
 196         if ((devid = get_devid(best->ne_name)) == NULL) {
 197                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 198         } else {
 199                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
 200                         return (-1);
 201                 devid_str_free(devid);
 202         }
 203 
 204         return (0);
 205 }
 206 
 207 /*
 208  * Add the given configuration to the list of known devices.
 209  */
 210 static int
 211 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 212     nvlist_t *config)
 213 {
 214         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 215         pool_entry_t *pe;
 216         vdev_entry_t *ve;
 217         config_entry_t *ce;
 218         name_entry_t *ne;
 219 
 220         /*
 221          * If this is a hot spare not currently in use or level 2 cache
 222          * device, add it to the list of names to translate, but don't do
 223          * anything else.
 224          */
 225         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 226             &state) == 0 &&
 227             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 228             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 229                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 230                         return (-1);
 231 
 232                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 233                         free(ne);
 234                         return (-1);
 235                 }
 236                 ne->ne_guid = vdev_guid;
 237                 ne->ne_next = pl->names;
 238                 pl->names = ne;
 239                 return (0);
 240         }
 241 
 242         /*
 243          * If we have a valid config but cannot read any of these fields, then
 244          * it means we have a half-initialized label.  In vdev_label_init()
 245          * we write a label with txg == 0 so that we can identify the device
 246          * in case the user refers to the same disk later on.  If we fail to
 247          * create the pool, we'll be left with a label in this state
 248          * which should not be considered part of a valid pool.
 249          */
 250         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 251             &pool_guid) != 0 ||
 252             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 253             &vdev_guid) != 0 ||
 254             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 255             &top_guid) != 0 ||
 256             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 257             &txg) != 0 || txg == 0) {
 258                 nvlist_free(config);
 259                 return (0);
 260         }
 261 
 262         /*
 263          * First, see if we know about this pool.  If not, then add it to the
 264          * list of known pools.
 265          */
 266         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 267                 if (pe->pe_guid == pool_guid)
 268                         break;
 269         }
 270 
 271         if (pe == NULL) {
 272                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 273                         nvlist_free(config);
 274                         return (-1);
 275                 }
 276                 pe->pe_guid = pool_guid;
 277                 pe->pe_next = pl->pools;
 278                 pl->pools = pe;
 279         }
 280 
 281         /*
 282          * Second, see if we know about this toplevel vdev.  Add it if its
 283          * missing.
 284          */
 285         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 286                 if (ve->ve_guid == top_guid)
 287                         break;
 288         }
 289 
 290         if (ve == NULL) {
 291                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 292                         nvlist_free(config);
 293                         return (-1);
 294                 }
 295                 ve->ve_guid = top_guid;
 296                 ve->ve_next = pe->pe_vdevs;
 297                 pe->pe_vdevs = ve;
 298         }
 299 
 300         /*
 301          * Third, see if we have a config with a matching transaction group.  If
 302          * so, then we do nothing.  Otherwise, add it to the list of known
 303          * configs.
 304          */
 305         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 306                 if (ce->ce_txg == txg)
 307                         break;
 308         }
 309 
 310         if (ce == NULL) {
 311                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 312                         nvlist_free(config);
 313                         return (-1);
 314                 }
 315                 ce->ce_txg = txg;
 316                 ce->ce_config = config;
 317                 ce->ce_next = ve->ve_configs;
 318                 ve->ve_configs = ce;
 319         } else {
 320                 nvlist_free(config);
 321         }
 322 
 323         /*
 324          * At this point we've successfully added our config to the list of
 325          * known configs.  The last thing to do is add the vdev guid -> path
 326          * mappings so that we can fix up the configuration as necessary before
 327          * doing the import.
 328          */
 329         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 330                 return (-1);
 331 
 332         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 333                 free(ne);
 334                 return (-1);
 335         }
 336 
 337         ne->ne_guid = vdev_guid;
 338         ne->ne_next = pl->names;
 339         pl->names = ne;
 340 
 341         return (0);
 342 }
 343 
 344 /*
 345  * Returns true if the named pool matches the given GUID.
 346  */
 347 static int
 348 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 349     boolean_t *isactive)
 350 {
 351         zpool_handle_t *zhp;
 352         uint64_t theguid;
 353 
 354         if (zpool_open_silent(hdl, name, &zhp) != 0)
 355                 return (-1);
 356 
 357         if (zhp == NULL) {
 358                 *isactive = B_FALSE;
 359                 return (0);
 360         }
 361 
 362         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 363             &theguid) == 0);
 364 
 365         zpool_close(zhp);
 366 
 367         *isactive = (theguid == guid);
 368         return (0);
 369 }
 370 
 371 static nvlist_t *
 372 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 373 {
 374         nvlist_t *nvl;
 375         zfs_cmd_t zc = { 0 };
 376         int err;
 377 
 378         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 379                 return (NULL);
 380 
 381         if (zcmd_alloc_dst_nvlist(hdl, &zc,
 382             zc.zc_nvlist_conf_size * 2) != 0) {
 383                 zcmd_free_nvlists(&zc);
 384                 return (NULL);
 385         }
 386 
 387         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 388             &zc)) != 0 && errno == ENOMEM) {
 389                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 390                         zcmd_free_nvlists(&zc);
 391                         return (NULL);
 392                 }
 393         }
 394 
 395         if (err) {
 396                 zcmd_free_nvlists(&zc);
 397                 return (NULL);
 398         }
 399 
 400         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 401                 zcmd_free_nvlists(&zc);
 402                 return (NULL);
 403         }
 404 
 405         zcmd_free_nvlists(&zc);
 406         return (nvl);
 407 }
 408 
 409 /*
 410  * Determine if the vdev id is a hole in the namespace.
 411  */
 412 boolean_t
 413 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
 414 {
 415         for (int c = 0; c < holes; c++) {
 416 
 417                 /* Top-level is a hole */
 418                 if (hole_array[c] == id)
 419                         return (B_TRUE);
 420         }
 421         return (B_FALSE);
 422 }
 423 
 424 /*
 425  * Convert our list of pools into the definitive set of configurations.  We
 426  * start by picking the best config for each toplevel vdev.  Once that's done,
 427  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 428  * pass to fix up any incorrect paths, and then add it to the main list to
 429  * return to the user.
 430  */
 431 static nvlist_t *
 432 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 433 {
 434         pool_entry_t *pe;
 435         vdev_entry_t *ve;
 436         config_entry_t *ce;
 437         nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
 438         nvlist_t **spares, **l2cache;
 439         uint_t i, nspares, nl2cache;
 440         boolean_t config_seen;
 441         uint64_t best_txg;
 442         char *name, *hostname;
 443         uint64_t guid;
 444         uint_t children = 0;
 445         nvlist_t **child = NULL;
 446         uint_t holes;
 447         uint64_t *hole_array, max_id;
 448         uint_t c;
 449         boolean_t isactive;
 450         uint64_t hostid;
 451         nvlist_t *nvl;
 452         boolean_t found_one = B_FALSE;
 453         boolean_t valid_top_config = B_FALSE;
 454 
 455         if (nvlist_alloc(&ret, 0, 0) != 0)
 456                 goto nomem;
 457 
 458         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 459                 uint64_t id, max_txg = 0;
 460 
 461                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 462                         goto nomem;
 463                 config_seen = B_FALSE;
 464 
 465                 /*
 466                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 467                  * from the first one we find, and then go through the rest and
 468                  * add them as necessary to the 'vdevs' member of the config.
 469                  */
 470                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 471 
 472                         /*
 473                          * Determine the best configuration for this vdev by
 474                          * selecting the config with the latest transaction
 475                          * group.
 476                          */
 477                         best_txg = 0;
 478                         for (ce = ve->ve_configs; ce != NULL;
 479                             ce = ce->ce_next) {
 480 
 481                                 if (ce->ce_txg > best_txg) {
 482                                         tmp = ce->ce_config;
 483                                         best_txg = ce->ce_txg;
 484                                 }
 485                         }
 486 
 487                         /*
 488                          * We rely on the fact that the max txg for the
 489                          * pool will contain the most up-to-date information
 490                          * about the valid top-levels in the vdev namespace.
 491                          */
 492                         if (best_txg > max_txg) {
 493                                 (void) nvlist_remove(config,
 494                                     ZPOOL_CONFIG_VDEV_CHILDREN,
 495                                     DATA_TYPE_UINT64);
 496                                 (void) nvlist_remove(config,
 497                                     ZPOOL_CONFIG_HOLE_ARRAY,
 498                                     DATA_TYPE_UINT64_ARRAY);
 499 
 500                                 max_txg = best_txg;
 501                                 hole_array = NULL;
 502                                 holes = 0;
 503                                 max_id = 0;
 504                                 valid_top_config = B_FALSE;
 505 
 506                                 if (nvlist_lookup_uint64(tmp,
 507                                     ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
 508                                         verify(nvlist_add_uint64(config,
 509                                             ZPOOL_CONFIG_VDEV_CHILDREN,
 510                                             max_id) == 0);
 511                                         valid_top_config = B_TRUE;
 512                                 }
 513 
 514                                 if (nvlist_lookup_uint64_array(tmp,
 515                                     ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
 516                                     &holes) == 0) {
 517                                         verify(nvlist_add_uint64_array(config,
 518                                             ZPOOL_CONFIG_HOLE_ARRAY,
 519                                             hole_array, holes) == 0);
 520                                 }
 521                         }
 522 
 523                         if (!config_seen) {
 524                                 /*
 525                                  * Copy the relevant pieces of data to the pool
 526                                  * configuration:
 527                                  *
 528                                  *      version
 529                                  *      pool guid
 530                                  *      name
 531                                  *      comment (if available)
 532                                  *      pool state
 533                                  *      hostid (if available)
 534                                  *      hostname (if available)
 535                                  */
 536                                 uint64_t state, version;
 537                                 char *comment = NULL;
 538 
 539                                 version = fnvlist_lookup_uint64(tmp,
 540                                     ZPOOL_CONFIG_VERSION);
 541                                 fnvlist_add_uint64(config,
 542                                     ZPOOL_CONFIG_VERSION, version);
 543                                 guid = fnvlist_lookup_uint64(tmp,
 544                                     ZPOOL_CONFIG_POOL_GUID);
 545                                 fnvlist_add_uint64(config,
 546                                     ZPOOL_CONFIG_POOL_GUID, guid);
 547                                 name = fnvlist_lookup_string(tmp,
 548                                     ZPOOL_CONFIG_POOL_NAME);
 549                                 fnvlist_add_string(config,
 550                                     ZPOOL_CONFIG_POOL_NAME, name);
 551 
 552                                 if (nvlist_lookup_string(tmp,
 553                                     ZPOOL_CONFIG_COMMENT, &comment) == 0)
 554                                         fnvlist_add_string(config,
 555                                             ZPOOL_CONFIG_COMMENT, comment);
 556 
 557                                 state = fnvlist_lookup_uint64(tmp,
 558                                     ZPOOL_CONFIG_POOL_STATE);
 559                                 fnvlist_add_uint64(config,
 560                                     ZPOOL_CONFIG_POOL_STATE, state);
 561 
 562                                 hostid = 0;
 563                                 if (nvlist_lookup_uint64(tmp,
 564                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 565                                         fnvlist_add_uint64(config,
 566                                             ZPOOL_CONFIG_HOSTID, hostid);
 567                                         hostname = fnvlist_lookup_string(tmp,
 568                                             ZPOOL_CONFIG_HOSTNAME);
 569                                         fnvlist_add_string(config,
 570                                             ZPOOL_CONFIG_HOSTNAME, hostname);
 571                                 }
 572 
 573                                 config_seen = B_TRUE;
 574                         }
 575 
 576                         /*
 577                          * Add this top-level vdev to the child array.
 578                          */
 579                         verify(nvlist_lookup_nvlist(tmp,
 580                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 581                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 582                             &id) == 0);
 583 
 584                         if (id >= children) {
 585                                 nvlist_t **newchild;
 586 
 587                                 newchild = zfs_alloc(hdl, (id + 1) *
 588                                     sizeof (nvlist_t *));
 589                                 if (newchild == NULL)
 590                                         goto nomem;
 591 
 592                                 for (c = 0; c < children; c++)
 593                                         newchild[c] = child[c];
 594 
 595                                 free(child);
 596                                 child = newchild;
 597                                 children = id + 1;
 598                         }
 599                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 600                                 goto nomem;
 601 
 602                 }
 603 
 604                 /*
 605                  * If we have information about all the top-levels then
 606                  * clean up the nvlist which we've constructed. This
 607                  * means removing any extraneous devices that are
 608                  * beyond the valid range or adding devices to the end
 609                  * of our array which appear to be missing.
 610                  */
 611                 if (valid_top_config) {
 612                         if (max_id < children) {
 613                                 for (c = max_id; c < children; c++)
 614                                         nvlist_free(child[c]);
 615                                 children = max_id;
 616                         } else if (max_id > children) {
 617                                 nvlist_t **newchild;
 618 
 619                                 newchild = zfs_alloc(hdl, (max_id) *
 620                                     sizeof (nvlist_t *));
 621                                 if (newchild == NULL)
 622                                         goto nomem;
 623 
 624                                 for (c = 0; c < children; c++)
 625                                         newchild[c] = child[c];
 626 
 627                                 free(child);
 628                                 child = newchild;
 629                                 children = max_id;
 630                         }
 631                 }
 632 
 633                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 634                     &guid) == 0);
 635 
 636                 /*
 637                  * The vdev namespace may contain holes as a result of
 638                  * device removal. We must add them back into the vdev
 639                  * tree before we process any missing devices.
 640                  */
 641                 if (holes > 0) {
 642                         ASSERT(valid_top_config);
 643 
 644                         for (c = 0; c < children; c++) {
 645                                 nvlist_t *holey;
 646 
 647                                 if (child[c] != NULL ||
 648                                     !vdev_is_hole(hole_array, holes, c))
 649                                         continue;
 650 
 651                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 652                                     0) != 0)
 653                                         goto nomem;
 654 
 655                                 /*
 656                                  * Holes in the namespace are treated as
 657                                  * "hole" top-level vdevs and have a
 658                                  * special flag set on them.
 659                                  */
 660                                 if (nvlist_add_string(holey,
 661                                     ZPOOL_CONFIG_TYPE,
 662                                     VDEV_TYPE_HOLE) != 0 ||
 663                                     nvlist_add_uint64(holey,
 664                                     ZPOOL_CONFIG_ID, c) != 0 ||
 665                                     nvlist_add_uint64(holey,
 666                                     ZPOOL_CONFIG_GUID, 0ULL) != 0)
 667                                         goto nomem;
 668                                 child[c] = holey;
 669                         }
 670                 }
 671 
 672                 /*
 673                  * Look for any missing top-level vdevs.  If this is the case,
 674                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 675                  * simply compress the child array, because the kernel performs
 676                  * certain checks to make sure the vdev IDs match their location
 677                  * in the configuration.
 678                  */
 679                 for (c = 0; c < children; c++) {
 680                         if (child[c] == NULL) {
 681                                 nvlist_t *missing;
 682                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 683                                     0) != 0)
 684                                         goto nomem;
 685                                 if (nvlist_add_string(missing,
 686                                     ZPOOL_CONFIG_TYPE,
 687                                     VDEV_TYPE_MISSING) != 0 ||
 688                                     nvlist_add_uint64(missing,
 689                                     ZPOOL_CONFIG_ID, c) != 0 ||
 690                                     nvlist_add_uint64(missing,
 691                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 692                                         nvlist_free(missing);
 693                                         goto nomem;
 694                                 }
 695                                 child[c] = missing;
 696                         }
 697                 }
 698 
 699                 /*
 700                  * Put all of this pool's top-level vdevs into a root vdev.
 701                  */
 702                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 703                         goto nomem;
 704                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 705                     VDEV_TYPE_ROOT) != 0 ||
 706                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 707                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 708                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 709                     child, children) != 0) {
 710                         nvlist_free(nvroot);
 711                         goto nomem;
 712                 }
 713 
 714                 for (c = 0; c < children; c++)
 715                         nvlist_free(child[c]);
 716                 free(child);
 717                 children = 0;
 718                 child = NULL;
 719 
 720                 /*
 721                  * Go through and fix up any paths and/or devids based on our
 722                  * known list of vdev GUID -> path mappings.
 723                  */
 724                 if (fix_paths(nvroot, pl->names) != 0) {
 725                         nvlist_free(nvroot);
 726                         goto nomem;
 727                 }
 728 
 729                 /*
 730                  * Add the root vdev to this pool's configuration.
 731                  */
 732                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 733                     nvroot) != 0) {
 734                         nvlist_free(nvroot);
 735                         goto nomem;
 736                 }
 737                 nvlist_free(nvroot);
 738 
 739                 /*
 740                  * zdb uses this path to report on active pools that were
 741                  * imported or created using -R.
 742                  */
 743                 if (active_ok)
 744                         goto add_pool;
 745 
 746                 /*
 747                  * Determine if this pool is currently active, in which case we
 748                  * can't actually import it.
 749                  */
 750                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 751                     &name) == 0);
 752                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 753                     &guid) == 0);
 754 
 755                 if (pool_active(hdl, name, guid, &isactive) != 0)
 756                         goto error;
 757 
 758                 if (isactive) {
 759                         nvlist_free(config);
 760                         config = NULL;
 761                         continue;
 762                 }
 763 
 764                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 765                         nvlist_free(config);
 766                         config = NULL;
 767                         continue;
 768                 }
 769 
 770                 nvlist_free(config);
 771                 config = nvl;
 772 
 773                 /*
 774                  * Go through and update the paths for spares, now that we have
 775                  * them.
 776                  */
 777                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 778                     &nvroot) == 0);
 779                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 780                     &spares, &nspares) == 0) {
 781                         for (i = 0; i < nspares; i++) {
 782                                 if (fix_paths(spares[i], pl->names) != 0)
 783                                         goto nomem;
 784                         }
 785                 }
 786 
 787                 /*
 788                  * Update the paths for l2cache devices.
 789                  */
 790                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 791                     &l2cache, &nl2cache) == 0) {
 792                         for (i = 0; i < nl2cache; i++) {
 793                                 if (fix_paths(l2cache[i], pl->names) != 0)
 794                                         goto nomem;
 795                         }
 796                 }
 797 
 798                 /*
 799                  * Restore the original information read from the actual label.
 800                  */
 801                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 802                     DATA_TYPE_UINT64);
 803                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 804                     DATA_TYPE_STRING);
 805                 if (hostid != 0) {
 806                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 807                             hostid) == 0);
 808                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 809                             hostname) == 0);
 810                 }
 811 
 812 add_pool:
 813                 /*
 814                  * Add this pool to the list of configs.
 815                  */
 816                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 817                     &name) == 0);
 818                 if (nvlist_add_nvlist(ret, name, config) != 0)
 819                         goto nomem;
 820 
 821                 found_one = B_TRUE;
 822                 nvlist_free(config);
 823                 config = NULL;
 824         }
 825 
 826         if (!found_one) {
 827                 nvlist_free(ret);
 828                 ret = NULL;
 829         }
 830 
 831         return (ret);
 832 
 833 nomem:
 834         (void) no_memory(hdl);
 835 error:
 836         nvlist_free(config);
 837         nvlist_free(ret);
 838         for (c = 0; c < children; c++)
 839                 nvlist_free(child[c]);
 840         free(child);
 841 
 842         return (NULL);
 843 }
 844 
 845 /*
 846  * Return the offset of the given label.
 847  */
 848 static uint64_t
 849 label_offset(uint64_t size, int l)
 850 {
 851         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 852         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 853             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 854 }
 855 
 856 /*
 857  * Given a file descriptor, read the label information and return an nvlist
 858  * describing the configuration, if there is one.
 859  */
 860 int
 861 zpool_read_label(int fd, nvlist_t **config)
 862 {
 863         struct stat64 statbuf;
 864         int l;
 865         vdev_label_t *label;
 866         uint64_t state, txg, size;
 867 
 868         *config = NULL;
 869 
 870         if (fstat64(fd, &statbuf) == -1)
 871                 return (0);
 872         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 873 
 874         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 875                 return (-1);
 876 
 877         for (l = 0; l < VDEV_LABELS; l++) {
 878                 if (pread64(fd, label, sizeof (vdev_label_t),
 879                     label_offset(size, l)) != sizeof (vdev_label_t))
 880                         continue;
 881 
 882                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 883                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 884                         continue;
 885 
 886                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 887                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 888                         nvlist_free(*config);
 889                         continue;
 890                 }
 891 
 892                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 893                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 894                     &txg) != 0 || txg == 0)) {
 895                         nvlist_free(*config);
 896                         continue;
 897                 }
 898 
 899                 free(label);
 900                 return (0);
 901         }
 902 
 903         free(label);
 904         *config = NULL;
 905         return (0);
 906 }
 907 
 908 typedef struct rdsk_node {
 909         char *rn_name;
 910         int rn_dfd;
 911         libzfs_handle_t *rn_hdl;
 912         nvlist_t *rn_config;
 913         avl_tree_t *rn_avl;
 914         avl_node_t rn_node;
 915         boolean_t rn_nozpool;
 916 } rdsk_node_t;
 917 
 918 static int
 919 slice_cache_compare(const void *arg1, const void *arg2)
 920 {
 921         const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
 922         const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
 923         char *nm1slice, *nm2slice;
 924         int rv;
 925 
 926         /*
 927          * slices zero and two are the most likely to provide results,
 928          * so put those first
 929          */
 930         nm1slice = strstr(nm1, "s0");
 931         nm2slice = strstr(nm2, "s0");
 932         if (nm1slice && !nm2slice) {
 933                 return (-1);
 934         }
 935         if (!nm1slice && nm2slice) {
 936                 return (1);
 937         }
 938         nm1slice = strstr(nm1, "s2");
 939         nm2slice = strstr(nm2, "s2");
 940         if (nm1slice && !nm2slice) {
 941                 return (-1);
 942         }
 943         if (!nm1slice && nm2slice) {
 944                 return (1);
 945         }
 946 
 947         rv = strcmp(nm1, nm2);
 948         if (rv == 0)
 949                 return (0);
 950         return (rv > 0 ? 1 : -1);
 951 }
 952 
 953 static void
 954 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
 955     diskaddr_t size, uint_t blksz)
 956 {
 957         rdsk_node_t tmpnode;
 958         rdsk_node_t *node;
 959         char sname[MAXNAMELEN];
 960 
 961         tmpnode.rn_name = &sname[0];
 962         (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
 963             diskname, partno);
 964         /*
 965          * protect against division by zero for disk labels that
 966          * contain a bogus sector size
 967          */
 968         if (blksz == 0)
 969                 blksz = DEV_BSIZE;
 970         /* too small to contain a zpool? */
 971         if ((size < (SPA_MINDEVSIZE / blksz)) &&
 972             (node = avl_find(r, &tmpnode, NULL)))
 973                 node->rn_nozpool = B_TRUE;
 974 }
 975 
 976 static void
 977 nozpool_all_slices(avl_tree_t *r, const char *sname)
 978 {
 979         char diskname[MAXNAMELEN];
 980         char *ptr;
 981         int i;
 982 
 983         (void) strncpy(diskname, sname, MAXNAMELEN);
 984         if (((ptr = strrchr(diskname, 's')) == NULL) &&
 985             ((ptr = strrchr(diskname, 'p')) == NULL))
 986                 return;
 987         ptr[0] = 's';
 988         ptr[1] = '\0';
 989         for (i = 0; i < NDKMAP; i++)
 990                 check_one_slice(r, diskname, i, 0, 1);
 991         ptr[0] = 'p';
 992         for (i = 0; i <= FD_NUMPART; i++)
 993                 check_one_slice(r, diskname, i, 0, 1);
 994 }
 995 
 996 static void
 997 check_slices(avl_tree_t *r, int fd, const char *sname)
 998 {
 999         struct extvtoc vtoc;
1000         struct dk_gpt *gpt;
1001         char diskname[MAXNAMELEN];
1002         char *ptr;
1003         int i;
1004 
1005         (void) strncpy(diskname, sname, MAXNAMELEN);
1006         if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1007                 return;
1008         ptr[1] = '\0';
1009 
1010         if (read_extvtoc(fd, &vtoc) >= 0) {
1011                 for (i = 0; i < NDKMAP; i++)
1012                         check_one_slice(r, diskname, i,
1013                             vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1014         } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1015                 /*
1016                  * on x86 we'll still have leftover links that point
1017                  * to slices s[9-15], so use NDKMAP instead
1018                  */
1019                 for (i = 0; i < NDKMAP; i++)
1020                         check_one_slice(r, diskname, i,
1021                             gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1022                 /* nodes p[1-4] are never used with EFI labels */
1023                 ptr[0] = 'p';
1024                 for (i = 1; i <= FD_NUMPART; i++)
1025                         check_one_slice(r, diskname, i, 0, 1);
1026                 efi_free(gpt);
1027         }
1028 }
1029 
1030 static void
1031 zpool_open_func(void *arg)
1032 {
1033         rdsk_node_t *rn = arg;
1034         struct stat64 statbuf;
1035         nvlist_t *config;
1036         int fd;
1037 
1038         if (rn->rn_nozpool)
1039                 return;
1040         if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1041                 /* symlink to a device that's no longer there */
1042                 if (errno == ENOENT)
1043                         nozpool_all_slices(rn->rn_avl, rn->rn_name);
1044                 return;
1045         }
1046         /*
1047          * Ignore failed stats.  We only want regular
1048          * files, character devs and block devs.
1049          */
1050         if (fstat64(fd, &statbuf) != 0 ||
1051             (!S_ISREG(statbuf.st_mode) &&
1052             !S_ISCHR(statbuf.st_mode) &&
1053             !S_ISBLK(statbuf.st_mode))) {
1054                 (void) close(fd);
1055                 return;
1056         }
1057         /* this file is too small to hold a zpool */
1058         if (S_ISREG(statbuf.st_mode) &&
1059             statbuf.st_size < SPA_MINDEVSIZE) {
1060                 (void) close(fd);
1061                 return;
1062         } else if (!S_ISREG(statbuf.st_mode)) {
1063                 /*
1064                  * Try to read the disk label first so we don't have to
1065                  * open a bunch of minor nodes that can't have a zpool.
1066                  */
1067                 check_slices(rn->rn_avl, fd, rn->rn_name);
1068         }
1069 
1070         if ((zpool_read_label(fd, &config)) != 0) {
1071                 (void) close(fd);
1072                 (void) no_memory(rn->rn_hdl);
1073                 return;
1074         }
1075         (void) close(fd);
1076 
1077 
1078         rn->rn_config = config;
1079         if (config != NULL) {
1080                 assert(rn->rn_nozpool == B_FALSE);
1081         }
1082 }
1083 
1084 /*
1085  * Given a file descriptor, clear (zero) the label information.  This function
1086  * is currently only used in the appliance stack as part of the ZFS sysevent
1087  * module.
1088  */
1089 int
1090 zpool_clear_label(int fd)
1091 {
1092         struct stat64 statbuf;
1093         int l;
1094         vdev_label_t *label;
1095         uint64_t size;
1096 
1097         if (fstat64(fd, &statbuf) == -1)
1098                 return (0);
1099         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1100 
1101         if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1102                 return (-1);
1103 
1104         for (l = 0; l < VDEV_LABELS; l++) {
1105                 if (pwrite64(fd, label, sizeof (vdev_label_t),
1106                     label_offset(size, l)) != sizeof (vdev_label_t))
1107                         return (-1);
1108         }
1109 
1110         free(label);
1111         return (0);
1112 }
1113 
1114 /*
1115  * Given a list of directories to search, find all pools stored on disk.  This
1116  * includes partial pools which are not available to import.  If no args are
1117  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1118  * poolname or guid (but not both) are provided by the caller when trying
1119  * to import a specific pool.
1120  */
1121 static nvlist_t *
1122 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1123 {
1124         int i, dirs = iarg->paths;
1125         DIR *dirp = NULL;
1126         struct dirent64 *dp;
1127         char path[MAXPATHLEN];
1128         char *end, **dir = iarg->path;
1129         size_t pathleft;
1130         nvlist_t *ret = NULL;
1131         static char *default_dir = "/dev/dsk";
1132         pool_list_t pools = { 0 };
1133         pool_entry_t *pe, *penext;
1134         vdev_entry_t *ve, *venext;
1135         config_entry_t *ce, *cenext;
1136         name_entry_t *ne, *nenext;
1137         avl_tree_t slice_cache;
1138         rdsk_node_t *slice;
1139         void *cookie;
1140 
1141         if (dirs == 0) {
1142                 dirs = 1;
1143                 dir = &default_dir;
1144         }
1145 
1146         /*
1147          * Go through and read the label configuration information from every
1148          * possible device, organizing the information according to pool GUID
1149          * and toplevel GUID.
1150          */
1151         for (i = 0; i < dirs; i++) {
1152                 tpool_t *t;
1153                 char *rdsk;
1154                 int dfd;
1155 
1156                 /* use realpath to normalize the path */
1157                 if (realpath(dir[i], path) == 0) {
1158                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1159                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1160                         goto error;
1161                 }
1162                 end = &path[strlen(path)];
1163                 *end++ = '/';
1164                 *end = 0;
1165                 pathleft = &path[sizeof (path)] - end;
1166 
1167                 /*
1168                  * Using raw devices instead of block devices when we're
1169                  * reading the labels skips a bunch of slow operations during
1170                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1171                  */
1172                 if (strcmp(path, "/dev/dsk/") == 0)
1173                         rdsk = "/dev/rdsk/";
1174                 else
1175                         rdsk = path;
1176 
1177                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1178                     (dirp = fdopendir(dfd)) == NULL) {
1179                         zfs_error_aux(hdl, strerror(errno));
1180                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1181                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1182                             rdsk);
1183                         goto error;
1184                 }
1185 
1186                 avl_create(&slice_cache, slice_cache_compare,
1187                     sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1188                 /*
1189                  * This is not MT-safe, but we have no MT consumers of libzfs
1190                  */
1191                 while ((dp = readdir64(dirp)) != NULL) {
1192                         const char *name = dp->d_name;
1193                         if (name[0] == '.' &&
1194                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1195                                 continue;
1196 
1197                         slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1198                         slice->rn_name = zfs_strdup(hdl, name);
1199                         slice->rn_avl = &slice_cache;
1200                         slice->rn_dfd = dfd;
1201                         slice->rn_hdl = hdl;
1202                         slice->rn_nozpool = B_FALSE;
1203                         avl_add(&slice_cache, slice);
1204                 }
1205                 /*
1206                  * create a thread pool to do all of this in parallel;
1207                  * rn_nozpool is not protected, so this is racy in that
1208                  * multiple tasks could decide that the same slice can
1209                  * not hold a zpool, which is benign.  Also choose
1210                  * double the number of processors; we hold a lot of
1211                  * locks in the kernel, so going beyond this doesn't
1212                  * buy us much.
1213                  */
1214                 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1215                     0, NULL);
1216                 for (slice = avl_first(&slice_cache); slice;
1217                     (slice = avl_walk(&slice_cache, slice,
1218                     AVL_AFTER)))
1219                         (void) tpool_dispatch(t, zpool_open_func, slice);
1220                 tpool_wait(t);
1221                 tpool_destroy(t);
1222 
1223                 cookie = NULL;
1224                 while ((slice = avl_destroy_nodes(&slice_cache,
1225                     &cookie)) != NULL) {
1226                         if (slice->rn_config != NULL) {
1227                                 nvlist_t *config = slice->rn_config;
1228                                 boolean_t matched = B_TRUE;
1229 
1230                                 if (iarg->poolname != NULL) {
1231                                         char *pname;
1232 
1233                                         matched = nvlist_lookup_string(config,
1234                                             ZPOOL_CONFIG_POOL_NAME,
1235                                             &pname) == 0 &&
1236                                             strcmp(iarg->poolname, pname) == 0;
1237                                 } else if (iarg->guid != 0) {
1238                                         uint64_t this_guid;
1239 
1240                                         matched = nvlist_lookup_uint64(config,
1241                                             ZPOOL_CONFIG_POOL_GUID,
1242                                             &this_guid) == 0 &&
1243                                             iarg->guid == this_guid;
1244                                 }
1245                                 if (!matched) {
1246                                         nvlist_free(config);
1247                                         config = NULL;
1248                                         continue;
1249                                 }
1250                                 /* use the non-raw path for the config */
1251                                 (void) strlcpy(end, slice->rn_name, pathleft);
1252                                 if (add_config(hdl, &pools, path, config) != 0)
1253                                         goto error;
1254                         }
1255                         free(slice->rn_name);
1256                         free(slice);
1257                 }
1258                 avl_destroy(&slice_cache);
1259 
1260                 (void) closedir(dirp);
1261                 dirp = NULL;
1262         }
1263 
1264         ret = get_configs(hdl, &pools, iarg->can_be_active);
1265 
1266 error:
1267         for (pe = pools.pools; pe != NULL; pe = penext) {
1268                 penext = pe->pe_next;
1269                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1270                         venext = ve->ve_next;
1271                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1272                                 cenext = ce->ce_next;
1273                                 if (ce->ce_config)
1274                                         nvlist_free(ce->ce_config);
1275                                 free(ce);
1276                         }
1277                         free(ve);
1278                 }
1279                 free(pe);
1280         }
1281 
1282         for (ne = pools.names; ne != NULL; ne = nenext) {
1283                 nenext = ne->ne_next;
1284                 if (ne->ne_name)
1285                         free(ne->ne_name);
1286                 free(ne);
1287         }
1288 
1289         if (dirp)
1290                 (void) closedir(dirp);
1291 
1292         return (ret);
1293 }
1294 
1295 nvlist_t *
1296 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1297 {
1298         importargs_t iarg = { 0 };
1299 
1300         iarg.paths = argc;
1301         iarg.path = argv;
1302 
1303         return (zpool_find_import_impl(hdl, &iarg));
1304 }
1305 
1306 /*
1307  * Given a cache file, return the contents as a list of importable pools.
1308  * poolname or guid (but not both) are provided by the caller when trying
1309  * to import a specific pool.
1310  */
1311 nvlist_t *
1312 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1313     char *poolname, uint64_t guid)
1314 {
1315         char *buf;
1316         int fd;
1317         struct stat64 statbuf;
1318         nvlist_t *raw, *src, *dst;
1319         nvlist_t *pools;
1320         nvpair_t *elem;
1321         char *name;
1322         uint64_t this_guid;
1323         boolean_t active;
1324 
1325         verify(poolname == NULL || guid == 0);
1326 
1327         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1328                 zfs_error_aux(hdl, "%s", strerror(errno));
1329                 (void) zfs_error(hdl, EZFS_BADCACHE,
1330                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1331                 return (NULL);
1332         }
1333 
1334         if (fstat64(fd, &statbuf) != 0) {
1335                 zfs_error_aux(hdl, "%s", strerror(errno));
1336                 (void) close(fd);
1337                 (void) zfs_error(hdl, EZFS_BADCACHE,
1338                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1339                 return (NULL);
1340         }
1341 
1342         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1343                 (void) close(fd);
1344                 return (NULL);
1345         }
1346 
1347         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1348                 (void) close(fd);
1349                 free(buf);
1350                 (void) zfs_error(hdl, EZFS_BADCACHE,
1351                     dgettext(TEXT_DOMAIN,
1352                     "failed to read cache file contents"));
1353                 return (NULL);
1354         }
1355 
1356         (void) close(fd);
1357 
1358         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1359                 free(buf);
1360                 (void) zfs_error(hdl, EZFS_BADCACHE,
1361                     dgettext(TEXT_DOMAIN,
1362                     "invalid or corrupt cache file contents"));
1363                 return (NULL);
1364         }
1365 
1366         free(buf);
1367 
1368         /*
1369          * Go through and get the current state of the pools and refresh their
1370          * state.
1371          */
1372         if (nvlist_alloc(&pools, 0, 0) != 0) {
1373                 (void) no_memory(hdl);
1374                 nvlist_free(raw);
1375                 return (NULL);
1376         }
1377 
1378         elem = NULL;
1379         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1380                 verify(nvpair_value_nvlist(elem, &src) == 0);
1381 
1382                 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1383                     &name) == 0);
1384                 if (poolname != NULL && strcmp(poolname, name) != 0)
1385                         continue;
1386 
1387                 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1388                     &this_guid) == 0);
1389                 if (guid != 0) {
1390                         verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1391                             &this_guid) == 0);
1392                         if (guid != this_guid)
1393                                 continue;
1394                 }
1395 
1396                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1397                         nvlist_free(raw);
1398                         nvlist_free(pools);
1399                         return (NULL);
1400                 }
1401 
1402                 if (active)
1403                         continue;
1404 
1405                 if ((dst = refresh_config(hdl, src)) == NULL) {
1406                         nvlist_free(raw);
1407                         nvlist_free(pools);
1408                         return (NULL);
1409                 }
1410 
1411                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1412                         (void) no_memory(hdl);
1413                         nvlist_free(dst);
1414                         nvlist_free(raw);
1415                         nvlist_free(pools);
1416                         return (NULL);
1417                 }
1418                 nvlist_free(dst);
1419         }
1420 
1421         nvlist_free(raw);
1422         return (pools);
1423 }
1424 
1425 static int
1426 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1427 {
1428         importargs_t *import = data;
1429         int found = 0;
1430 
1431         if (import->poolname != NULL) {
1432                 char *pool_name;
1433 
1434                 verify(nvlist_lookup_string(zhp->zpool_config,
1435                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1436                 if (strcmp(pool_name, import->poolname) == 0)
1437                         found = 1;
1438         } else {
1439                 uint64_t pool_guid;
1440 
1441                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1442                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1443                 if (pool_guid == import->guid)
1444                         found = 1;
1445         }
1446 
1447         zpool_close(zhp);
1448         return (found);
1449 }
1450 
1451 nvlist_t *
1452 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1453 {
1454         verify(import->poolname == NULL || import->guid == 0);
1455 
1456         if (import->unique)
1457                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1458 
1459         if (import->cachefile != NULL)
1460                 return (zpool_find_import_cached(hdl, import->cachefile,
1461                     import->poolname, import->guid));
1462 
1463         return (zpool_find_import_impl(hdl, import));
1464 }
1465 
1466 boolean_t
1467 find_guid(nvlist_t *nv, uint64_t guid)
1468 {
1469         uint64_t tmp;
1470         nvlist_t **child;
1471         uint_t c, children;
1472 
1473         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1474         if (tmp == guid)
1475                 return (B_TRUE);
1476 
1477         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1478             &child, &children) == 0) {
1479                 for (c = 0; c < children; c++)
1480                         if (find_guid(child[c], guid))
1481                                 return (B_TRUE);
1482         }
1483 
1484         return (B_FALSE);
1485 }
1486 
1487 typedef struct aux_cbdata {
1488         const char      *cb_type;
1489         uint64_t        cb_guid;
1490         zpool_handle_t  *cb_zhp;
1491 } aux_cbdata_t;
1492 
1493 static int
1494 find_aux(zpool_handle_t *zhp, void *data)
1495 {
1496         aux_cbdata_t *cbp = data;
1497         nvlist_t **list;
1498         uint_t i, count;
1499         uint64_t guid;
1500         nvlist_t *nvroot;
1501 
1502         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1503             &nvroot) == 0);
1504 
1505         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1506             &list, &count) == 0) {
1507                 for (i = 0; i < count; i++) {
1508                         verify(nvlist_lookup_uint64(list[i],
1509                             ZPOOL_CONFIG_GUID, &guid) == 0);
1510                         if (guid == cbp->cb_guid) {
1511                                 cbp->cb_zhp = zhp;
1512                                 return (1);
1513                         }
1514                 }
1515         }
1516 
1517         zpool_close(zhp);
1518         return (0);
1519 }
1520 
1521 /*
1522  * Determines if the pool is in use.  If so, it returns true and the state of
1523  * the pool as well as the name of the pool.  Both strings are allocated and
1524  * must be freed by the caller.
1525  */
1526 int
1527 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1528     boolean_t *inuse)
1529 {
1530         nvlist_t *config;
1531         char *name;
1532         boolean_t ret;
1533         uint64_t guid, vdev_guid;
1534         zpool_handle_t *zhp;
1535         nvlist_t *pool_config;
1536         uint64_t stateval, isspare;
1537         aux_cbdata_t cb = { 0 };
1538         boolean_t isactive;
1539 
1540         *inuse = B_FALSE;
1541 
1542         if (zpool_read_label(fd, &config) != 0) {
1543                 (void) no_memory(hdl);
1544                 return (-1);
1545         }
1546 
1547         if (config == NULL)
1548                 return (0);
1549 
1550         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1551             &stateval) == 0);
1552         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1553             &vdev_guid) == 0);
1554 
1555         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1556                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1557                     &name) == 0);
1558                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1559                     &guid) == 0);
1560         }
1561 
1562         switch (stateval) {
1563         case POOL_STATE_EXPORTED:
1564                 /*
1565                  * A pool with an exported state may in fact be imported
1566                  * read-only, so check the in-core state to see if it's
1567                  * active and imported read-only.  If it is, set
1568                  * its state to active.
1569                  */
1570                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1571                     (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1572                         if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1573                                 stateval = POOL_STATE_ACTIVE;
1574 
1575                         zpool_close(zhp);
1576                 }
1577 
1578                 ret = B_TRUE;
1579                 break;
1580 
1581         case POOL_STATE_ACTIVE:
1582                 /*
1583                  * For an active pool, we have to determine if it's really part
1584                  * of a currently active pool (in which case the pool will exist
1585                  * and the guid will be the same), or whether it's part of an
1586                  * active pool that was disconnected without being explicitly
1587                  * exported.
1588                  */
1589                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1590                         nvlist_free(config);
1591                         return (-1);
1592                 }
1593 
1594                 if (isactive) {
1595                         /*
1596                          * Because the device may have been removed while
1597                          * offlined, we only report it as active if the vdev is
1598                          * still present in the config.  Otherwise, pretend like
1599                          * it's not in use.
1600                          */
1601                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1602                             (pool_config = zpool_get_config(zhp, NULL))
1603                             != NULL) {
1604                                 nvlist_t *nvroot;
1605 
1606                                 verify(nvlist_lookup_nvlist(pool_config,
1607                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1608                                 ret = find_guid(nvroot, vdev_guid);
1609                         } else {
1610                                 ret = B_FALSE;
1611                         }
1612 
1613                         /*
1614                          * If this is an active spare within another pool, we
1615                          * treat it like an unused hot spare.  This allows the
1616                          * user to create a pool with a hot spare that currently
1617                          * in use within another pool.  Since we return B_TRUE,
1618                          * libdiskmgt will continue to prevent generic consumers
1619                          * from using the device.
1620                          */
1621                         if (ret && nvlist_lookup_uint64(config,
1622                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1623                                 stateval = POOL_STATE_SPARE;
1624 
1625                         if (zhp != NULL)
1626                                 zpool_close(zhp);
1627                 } else {
1628                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1629                         ret = B_TRUE;
1630                 }
1631                 break;
1632 
1633         case POOL_STATE_SPARE:
1634                 /*
1635                  * For a hot spare, it can be either definitively in use, or
1636                  * potentially active.  To determine if it's in use, we iterate
1637                  * over all pools in the system and search for one with a spare
1638                  * with a matching guid.
1639                  *
1640                  * Due to the shared nature of spares, we don't actually report
1641                  * the potentially active case as in use.  This means the user
1642                  * can freely create pools on the hot spares of exported pools,
1643                  * but to do otherwise makes the resulting code complicated, and
1644                  * we end up having to deal with this case anyway.
1645                  */
1646                 cb.cb_zhp = NULL;
1647                 cb.cb_guid = vdev_guid;
1648                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1649                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1650                         name = (char *)zpool_get_name(cb.cb_zhp);
1651                         ret = TRUE;
1652                 } else {
1653                         ret = FALSE;
1654                 }
1655                 break;
1656 
1657         case POOL_STATE_L2CACHE:
1658 
1659                 /*
1660                  * Check if any pool is currently using this l2cache device.
1661                  */
1662                 cb.cb_zhp = NULL;
1663                 cb.cb_guid = vdev_guid;
1664                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1665                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1666                         name = (char *)zpool_get_name(cb.cb_zhp);
1667                         ret = TRUE;
1668                 } else {
1669                         ret = FALSE;
1670                 }
1671                 break;
1672 
1673         default:
1674                 ret = B_FALSE;
1675         }
1676 
1677 
1678         if (ret) {
1679                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1680                         if (cb.cb_zhp)
1681                                 zpool_close(cb.cb_zhp);
1682                         nvlist_free(config);
1683                         return (-1);
1684                 }
1685                 *state = (pool_state_t)stateval;
1686         }
1687 
1688         if (cb.cb_zhp)
1689                 zpool_close(cb.cb_zhp);
1690 
1691         nvlist_free(config);
1692         *inuse = ret;
1693         return (0);
1694 }