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 2015 Nexenta Systems, Inc.  All rights reserved.
  24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  25  * Copyright (c) 2013 by Delphix. All rights reserved.
  26  * Copyright 2015 RackTop Systems.
  27  */
  28 
  29 /*
  30  * Pool import support functions.
  31  *
  32  * To import a pool, we rely on reading the configuration information from the
  33  * ZFS label of each device.  If we successfully read the label, then we
  34  * organize the configuration information in the following hierarchy:
  35  *
  36  *      pool guid -> toplevel vdev guid -> label txg
  37  *
  38  * Duplicate entries matching this same tuple will be discarded.  Once we have
  39  * examined every device, we pick the best label txg config for each toplevel
  40  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  41  * update any paths that have changed.  Finally, we attempt to import the pool
  42  * using our derived config, and record the results.
  43  */
  44 
  45 #include <ctype.h>
  46 #include <devid.h>
  47 #include <dirent.h>
  48 #include <errno.h>
  49 #include <libintl.h>
  50 #include <stddef.h>
  51 #include <stdlib.h>
  52 #include <string.h>
  53 #include <sys/stat.h>
  54 #include <unistd.h>
  55 #include <fcntl.h>
  56 #include <sys/vtoc.h>
  57 #include <sys/dktp/fdisk.h>
  58 #include <sys/efi_partition.h>
  59 #include <thread_pool.h>
  60 
  61 #include <sys/vdev_impl.h>
  62 
  63 #include "libzfs.h"
  64 #include "libzfs_impl.h"
  65 
  66 /*
  67  * Intermediate structures used to gather configuration information.
  68  */
  69 typedef struct config_entry {
  70         uint64_t                ce_txg;
  71         nvlist_t                *ce_config;
  72         struct config_entry     *ce_next;
  73 } config_entry_t;
  74 
  75 typedef struct vdev_entry {
  76         uint64_t                ve_guid;
  77         config_entry_t          *ve_configs;
  78         struct vdev_entry       *ve_next;
  79 } vdev_entry_t;
  80 
  81 typedef struct pool_entry {
  82         uint64_t                pe_guid;
  83         vdev_entry_t            *pe_vdevs;
  84         struct pool_entry       *pe_next;
  85 } pool_entry_t;
  86 
  87 typedef struct name_entry {
  88         char                    *ne_name;
  89         uint64_t                ne_guid;
  90         struct name_entry       *ne_next;
  91 } name_entry_t;
  92 
  93 typedef struct pool_list {
  94         pool_entry_t            *pools;
  95         name_entry_t            *names;
  96 } pool_list_t;
  97 
  98 static char *
  99 get_devid(const char *path)
 100 {
 101         int fd;
 102         ddi_devid_t devid;
 103         char *minor, *ret;
 104 
 105         if ((fd = open(path, O_RDONLY)) < 0)
 106                 return (NULL);
 107 
 108         minor = NULL;
 109         ret = NULL;
 110         if (devid_get(fd, &devid) == 0) {
 111                 if (devid_get_minor_name(fd, &minor) == 0)
 112                         ret = devid_str_encode(devid, minor);
 113                 if (minor != NULL)
 114                         devid_str_free(minor);
 115                 devid_free(devid);
 116         }
 117         (void) close(fd);
 118 
 119         return (ret);
 120 }
 121 
 122 
 123 /*
 124  * Go through and fix up any path and/or devid information for the given vdev
 125  * configuration.
 126  */
 127 static int
 128 fix_paths(nvlist_t *nv, name_entry_t *names)
 129 {
 130         nvlist_t **child;
 131         uint_t c, children;
 132         uint64_t guid;
 133         name_entry_t *ne, *best;
 134         char *path, *devid;
 135         int matched;
 136 
 137         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 138             &child, &children) == 0) {
 139                 for (c = 0; c < children; c++)
 140                         if (fix_paths(child[c], names) != 0)
 141                                 return (-1);
 142                 return (0);
 143         }
 144 
 145         /*
 146          * This is a leaf (file or disk) vdev.  In either case, go through
 147          * the name list and see if we find a matching guid.  If so, replace
 148          * the path and see if we can calculate a new devid.
 149          *
 150          * There may be multiple names associated with a particular guid, in
 151          * which case we have overlapping slices or multiple paths to the same
 152          * disk.  If this is the case, then we want to pick the path that is
 153          * the most similar to the original, where "most similar" is the number
 154          * of matching characters starting from the end of the path.  This will
 155          * preserve slice numbers even if the disks have been reorganized, and
 156          * will also catch preferred disk names if multiple paths exist.
 157          */
 158         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 159         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 160                 path = NULL;
 161 
 162         matched = 0;
 163         best = NULL;
 164         for (ne = names; ne != NULL; ne = ne->ne_next) {
 165                 if (ne->ne_guid == guid) {
 166                         const char *src, *dst;
 167                         int count;
 168 
 169                         if (path == NULL) {
 170                                 best = ne;
 171                                 break;
 172                         }
 173 
 174                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 175                         dst = path + strlen(path) - 1;
 176                         for (count = 0; src >= ne->ne_name && dst >= path;
 177                             src--, dst--, count++)
 178                                 if (*src != *dst)
 179                                         break;
 180 
 181                         /*
 182                          * At this point, 'count' is the number of characters
 183                          * matched from the end.
 184                          */
 185                         if (count > matched || best == NULL) {
 186                                 best = ne;
 187                                 matched = count;
 188                         }
 189                 }
 190         }
 191 
 192         if (best == NULL)
 193                 return (0);
 194 
 195         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 196                 return (-1);
 197 
 198         if ((devid = get_devid(best->ne_name)) == NULL) {
 199                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 200         } else {
 201                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
 202                         devid_str_free(devid);
 203                         return (-1);
 204                 }
 205                 devid_str_free(devid);
 206         }
 207 
 208         return (0);
 209 }
 210 
 211 /*
 212  * Add the given configuration to the list of known devices.
 213  */
 214 static int
 215 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 216     nvlist_t *config)
 217 {
 218         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 219         pool_entry_t *pe;
 220         vdev_entry_t *ve;
 221         config_entry_t *ce;
 222         name_entry_t *ne;
 223 
 224         /*
 225          * If this is a hot spare not currently in use or level 2 cache
 226          * device, add it to the list of names to translate, but don't do
 227          * anything else.
 228          */
 229         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 230             &state) == 0 &&
 231             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 232             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 233                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 234                         return (-1);
 235 
 236                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 237                         free(ne);
 238                         return (-1);
 239                 }
 240                 ne->ne_guid = vdev_guid;
 241                 ne->ne_next = pl->names;
 242                 pl->names = ne;
 243                 return (0);
 244         }
 245 
 246         /*
 247          * If we have a valid config but cannot read any of these fields, then
 248          * it means we have a half-initialized label.  In vdev_label_init()
 249          * we write a label with txg == 0 so that we can identify the device
 250          * in case the user refers to the same disk later on.  If we fail to
 251          * create the pool, we'll be left with a label in this state
 252          * which should not be considered part of a valid pool.
 253          */
 254         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 255             &pool_guid) != 0 ||
 256             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 257             &vdev_guid) != 0 ||
 258             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 259             &top_guid) != 0 ||
 260             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 261             &txg) != 0 || txg == 0) {
 262                 nvlist_free(config);
 263                 return (0);
 264         }
 265 
 266         /*
 267          * First, see if we know about this pool.  If not, then add it to the
 268          * list of known pools.
 269          */
 270         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 271                 if (pe->pe_guid == pool_guid)
 272                         break;
 273         }
 274 
 275         if (pe == NULL) {
 276                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 277                         nvlist_free(config);
 278                         return (-1);
 279                 }
 280                 pe->pe_guid = pool_guid;
 281                 pe->pe_next = pl->pools;
 282                 pl->pools = pe;
 283         }
 284 
 285         /*
 286          * Second, see if we know about this toplevel vdev.  Add it if its
 287          * missing.
 288          */
 289         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 290                 if (ve->ve_guid == top_guid)
 291                         break;
 292         }
 293 
 294         if (ve == NULL) {
 295                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 296                         nvlist_free(config);
 297                         return (-1);
 298                 }
 299                 ve->ve_guid = top_guid;
 300                 ve->ve_next = pe->pe_vdevs;
 301                 pe->pe_vdevs = ve;
 302         }
 303 
 304         /*
 305          * Third, see if we have a config with a matching transaction group.  If
 306          * so, then we do nothing.  Otherwise, add it to the list of known
 307          * configs.
 308          */
 309         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 310                 if (ce->ce_txg == txg)
 311                         break;
 312         }
 313 
 314         if (ce == NULL) {
 315                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 316                         nvlist_free(config);
 317                         return (-1);
 318                 }
 319                 ce->ce_txg = txg;
 320                 ce->ce_config = config;
 321                 ce->ce_next = ve->ve_configs;
 322                 ve->ve_configs = ce;
 323         } else {
 324                 nvlist_free(config);
 325         }
 326 
 327         /*
 328          * At this point we've successfully added our config to the list of
 329          * known configs.  The last thing to do is add the vdev guid -> path
 330          * mappings so that we can fix up the configuration as necessary before
 331          * doing the import.
 332          */
 333         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 334                 return (-1);
 335 
 336         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 337                 free(ne);
 338                 return (-1);
 339         }
 340 
 341         ne->ne_guid = vdev_guid;
 342         ne->ne_next = pl->names;
 343         pl->names = ne;
 344 
 345         return (0);
 346 }
 347 
 348 /*
 349  * Returns true if the named pool matches the given GUID.
 350  */
 351 static int
 352 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 353     boolean_t *isactive)
 354 {
 355         zpool_handle_t *zhp;
 356         uint64_t theguid;
 357 
 358         if (zpool_open_silent(hdl, name, &zhp) != 0)
 359                 return (-1);
 360 
 361         if (zhp == NULL) {
 362                 *isactive = B_FALSE;
 363                 return (0);
 364         }
 365 
 366         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 367             &theguid) == 0);
 368 
 369         zpool_close(zhp);
 370 
 371         *isactive = (theguid == guid);
 372         return (0);
 373 }
 374 
 375 static nvlist_t *
 376 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 377 {
 378         nvlist_t *nvl;
 379         zfs_cmd_t zc = { 0 };
 380         int err;
 381 
 382         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 383                 return (NULL);
 384 
 385         if (zcmd_alloc_dst_nvlist(hdl, &zc,
 386             zc.zc_nvlist_conf_size * 2) != 0) {
 387                 zcmd_free_nvlists(&zc);
 388                 return (NULL);
 389         }
 390 
 391         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 392             &zc)) != 0 && errno == ENOMEM) {
 393                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 394                         zcmd_free_nvlists(&zc);
 395                         return (NULL);
 396                 }
 397         }
 398 
 399         if (err) {
 400                 zcmd_free_nvlists(&zc);
 401                 return (NULL);
 402         }
 403 
 404         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 405                 zcmd_free_nvlists(&zc);
 406                 return (NULL);
 407         }
 408 
 409         zcmd_free_nvlists(&zc);
 410         return (nvl);
 411 }
 412 
 413 /*
 414  * Determine if the vdev id is a hole in the namespace.
 415  */
 416 boolean_t
 417 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
 418 {
 419         for (int c = 0; c < holes; c++) {
 420 
 421                 /* Top-level is a hole */
 422                 if (hole_array[c] == id)
 423                         return (B_TRUE);
 424         }
 425         return (B_FALSE);
 426 }
 427 
 428 /*
 429  * Convert our list of pools into the definitive set of configurations.  We
 430  * start by picking the best config for each toplevel vdev.  Once that's done,
 431  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 432  * pass to fix up any incorrect paths, and then add it to the main list to
 433  * return to the user.
 434  */
 435 static nvlist_t *
 436 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 437 {
 438         pool_entry_t *pe;
 439         vdev_entry_t *ve;
 440         config_entry_t *ce;
 441         nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
 442         nvlist_t **spares, **l2cache;
 443         uint_t i, nspares, nl2cache;
 444         boolean_t config_seen;
 445         uint64_t best_txg;
 446         char *name, *hostname;
 447         uint64_t guid;
 448         uint_t children = 0;
 449         nvlist_t **child = NULL;
 450         uint_t holes;
 451         uint64_t *hole_array, max_id;
 452         uint_t c;
 453         boolean_t isactive;
 454         uint64_t hostid;
 455         nvlist_t *nvl;
 456         boolean_t found_one = B_FALSE;
 457         boolean_t valid_top_config = B_FALSE;
 458 
 459         if (nvlist_alloc(&ret, 0, 0) != 0)
 460                 goto nomem;
 461 
 462         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 463                 uint64_t id, max_txg = 0;
 464 
 465                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 466                         goto nomem;
 467                 config_seen = B_FALSE;
 468 
 469                 /*
 470                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 471                  * from the first one we find, and then go through the rest and
 472                  * add them as necessary to the 'vdevs' member of the config.
 473                  */
 474                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 475 
 476                         /*
 477                          * Determine the best configuration for this vdev by
 478                          * selecting the config with the latest transaction
 479                          * group.
 480                          */
 481                         best_txg = 0;
 482                         for (ce = ve->ve_configs; ce != NULL;
 483                             ce = ce->ce_next) {
 484 
 485                                 if (ce->ce_txg > best_txg) {
 486                                         tmp = ce->ce_config;
 487                                         best_txg = ce->ce_txg;
 488                                 }
 489                         }
 490 
 491                         /*
 492                          * We rely on the fact that the max txg for the
 493                          * pool will contain the most up-to-date information
 494                          * about the valid top-levels in the vdev namespace.
 495                          */
 496                         if (best_txg > max_txg) {
 497                                 (void) nvlist_remove(config,
 498                                     ZPOOL_CONFIG_VDEV_CHILDREN,
 499                                     DATA_TYPE_UINT64);
 500                                 (void) nvlist_remove(config,
 501                                     ZPOOL_CONFIG_HOLE_ARRAY,
 502                                     DATA_TYPE_UINT64_ARRAY);
 503 
 504                                 max_txg = best_txg;
 505                                 hole_array = NULL;
 506                                 holes = 0;
 507                                 max_id = 0;
 508                                 valid_top_config = B_FALSE;
 509 
 510                                 if (nvlist_lookup_uint64(tmp,
 511                                     ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
 512                                         verify(nvlist_add_uint64(config,
 513                                             ZPOOL_CONFIG_VDEV_CHILDREN,
 514                                             max_id) == 0);
 515                                         valid_top_config = B_TRUE;
 516                                 }
 517 
 518                                 if (nvlist_lookup_uint64_array(tmp,
 519                                     ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
 520                                     &holes) == 0) {
 521                                         verify(nvlist_add_uint64_array(config,
 522                                             ZPOOL_CONFIG_HOLE_ARRAY,
 523                                             hole_array, holes) == 0);
 524                                 }
 525                         }
 526 
 527                         if (!config_seen) {
 528                                 /*
 529                                  * Copy the relevant pieces of data to the pool
 530                                  * configuration:
 531                                  *
 532                                  *      version
 533                                  *      pool guid
 534                                  *      name
 535                                  *      comment (if available)
 536                                  *      pool state
 537                                  *      hostid (if available)
 538                                  *      hostname (if available)
 539                                  */
 540                                 uint64_t state, version;
 541                                 char *comment = NULL;
 542 
 543                                 version = fnvlist_lookup_uint64(tmp,
 544                                     ZPOOL_CONFIG_VERSION);
 545                                 fnvlist_add_uint64(config,
 546                                     ZPOOL_CONFIG_VERSION, version);
 547                                 guid = fnvlist_lookup_uint64(tmp,
 548                                     ZPOOL_CONFIG_POOL_GUID);
 549                                 fnvlist_add_uint64(config,
 550                                     ZPOOL_CONFIG_POOL_GUID, guid);
 551                                 name = fnvlist_lookup_string(tmp,
 552                                     ZPOOL_CONFIG_POOL_NAME);
 553                                 fnvlist_add_string(config,
 554                                     ZPOOL_CONFIG_POOL_NAME, name);
 555 
 556                                 if (nvlist_lookup_string(tmp,
 557                                     ZPOOL_CONFIG_COMMENT, &comment) == 0)
 558                                         fnvlist_add_string(config,
 559                                             ZPOOL_CONFIG_COMMENT, comment);
 560 
 561                                 state = fnvlist_lookup_uint64(tmp,
 562                                     ZPOOL_CONFIG_POOL_STATE);
 563                                 fnvlist_add_uint64(config,
 564                                     ZPOOL_CONFIG_POOL_STATE, state);
 565 
 566                                 hostid = 0;
 567                                 if (nvlist_lookup_uint64(tmp,
 568                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 569                                         fnvlist_add_uint64(config,
 570                                             ZPOOL_CONFIG_HOSTID, hostid);
 571                                         hostname = fnvlist_lookup_string(tmp,
 572                                             ZPOOL_CONFIG_HOSTNAME);
 573                                         fnvlist_add_string(config,
 574                                             ZPOOL_CONFIG_HOSTNAME, hostname);
 575                                 }
 576 
 577                                 config_seen = B_TRUE;
 578                         }
 579 
 580                         /*
 581                          * Add this top-level vdev to the child array.
 582                          */
 583                         verify(nvlist_lookup_nvlist(tmp,
 584                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 585                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 586                             &id) == 0);
 587 
 588                         if (id >= children) {
 589                                 nvlist_t **newchild;
 590 
 591                                 newchild = zfs_alloc(hdl, (id + 1) *
 592                                     sizeof (nvlist_t *));
 593                                 if (newchild == NULL)
 594                                         goto nomem;
 595 
 596                                 for (c = 0; c < children; c++)
 597                                         newchild[c] = child[c];
 598 
 599                                 free(child);
 600                                 child = newchild;
 601                                 children = id + 1;
 602                         }
 603                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 604                                 goto nomem;
 605 
 606                 }
 607 
 608                 /*
 609                  * If we have information about all the top-levels then
 610                  * clean up the nvlist which we've constructed. This
 611                  * means removing any extraneous devices that are
 612                  * beyond the valid range or adding devices to the end
 613                  * of our array which appear to be missing.
 614                  */
 615                 if (valid_top_config) {
 616                         if (max_id < children) {
 617                                 for (c = max_id; c < children; c++)
 618                                         nvlist_free(child[c]);
 619                                 children = max_id;
 620                         } else if (max_id > children) {
 621                                 nvlist_t **newchild;
 622 
 623                                 newchild = zfs_alloc(hdl, (max_id) *
 624                                     sizeof (nvlist_t *));
 625                                 if (newchild == NULL)
 626                                         goto nomem;
 627 
 628                                 for (c = 0; c < children; c++)
 629                                         newchild[c] = child[c];
 630 
 631                                 free(child);
 632                                 child = newchild;
 633                                 children = max_id;
 634                         }
 635                 }
 636 
 637                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 638                     &guid) == 0);
 639 
 640                 /*
 641                  * The vdev namespace may contain holes as a result of
 642                  * device removal. We must add them back into the vdev
 643                  * tree before we process any missing devices.
 644                  */
 645                 if (holes > 0) {
 646                         ASSERT(valid_top_config);
 647 
 648                         for (c = 0; c < children; c++) {
 649                                 nvlist_t *holey;
 650 
 651                                 if (child[c] != NULL ||
 652                                     !vdev_is_hole(hole_array, holes, c))
 653                                         continue;
 654 
 655                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 656                                     0) != 0)
 657                                         goto nomem;
 658 
 659                                 /*
 660                                  * Holes in the namespace are treated as
 661                                  * "hole" top-level vdevs and have a
 662                                  * special flag set on them.
 663                                  */
 664                                 if (nvlist_add_string(holey,
 665                                     ZPOOL_CONFIG_TYPE,
 666                                     VDEV_TYPE_HOLE) != 0 ||
 667                                     nvlist_add_uint64(holey,
 668                                     ZPOOL_CONFIG_ID, c) != 0 ||
 669                                     nvlist_add_uint64(holey,
 670                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 671                                         nvlist_free(holey);
 672                                         goto nomem;
 673                                 }
 674                                 child[c] = holey;
 675                         }
 676                 }
 677 
 678                 /*
 679                  * Look for any missing top-level vdevs.  If this is the case,
 680                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 681                  * simply compress the child array, because the kernel performs
 682                  * certain checks to make sure the vdev IDs match their location
 683                  * in the configuration.
 684                  */
 685                 for (c = 0; c < children; c++) {
 686                         if (child[c] == NULL) {
 687                                 nvlist_t *missing;
 688                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 689                                     0) != 0)
 690                                         goto nomem;
 691                                 if (nvlist_add_string(missing,
 692                                     ZPOOL_CONFIG_TYPE,
 693                                     VDEV_TYPE_MISSING) != 0 ||
 694                                     nvlist_add_uint64(missing,
 695                                     ZPOOL_CONFIG_ID, c) != 0 ||
 696                                     nvlist_add_uint64(missing,
 697                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 698                                         nvlist_free(missing);
 699                                         goto nomem;
 700                                 }
 701                                 child[c] = missing;
 702                         }
 703                 }
 704 
 705                 /*
 706                  * Put all of this pool's top-level vdevs into a root vdev.
 707                  */
 708                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 709                         goto nomem;
 710                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 711                     VDEV_TYPE_ROOT) != 0 ||
 712                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 713                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 714                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 715                     child, children) != 0) {
 716                         nvlist_free(nvroot);
 717                         goto nomem;
 718                 }
 719 
 720                 for (c = 0; c < children; c++)
 721                         nvlist_free(child[c]);
 722                 free(child);
 723                 children = 0;
 724                 child = NULL;
 725 
 726                 /*
 727                  * Go through and fix up any paths and/or devids based on our
 728                  * known list of vdev GUID -> path mappings.
 729                  */
 730                 if (fix_paths(nvroot, pl->names) != 0) {
 731                         nvlist_free(nvroot);
 732                         goto nomem;
 733                 }
 734 
 735                 /*
 736                  * Add the root vdev to this pool's configuration.
 737                  */
 738                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 739                     nvroot) != 0) {
 740                         nvlist_free(nvroot);
 741                         goto nomem;
 742                 }
 743                 nvlist_free(nvroot);
 744 
 745                 /*
 746                  * zdb uses this path to report on active pools that were
 747                  * imported or created using -R.
 748                  */
 749                 if (active_ok)
 750                         goto add_pool;
 751 
 752                 /*
 753                  * Determine if this pool is currently active, in which case we
 754                  * can't actually import it.
 755                  */
 756                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 757                     &name) == 0);
 758                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 759                     &guid) == 0);
 760 
 761                 if (pool_active(hdl, name, guid, &isactive) != 0)
 762                         goto error;
 763 
 764                 if (isactive) {
 765                         nvlist_free(config);
 766                         config = NULL;
 767                         continue;
 768                 }
 769 
 770                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 771                         nvlist_free(config);
 772                         config = NULL;
 773                         continue;
 774                 }
 775 
 776                 nvlist_free(config);
 777                 config = nvl;
 778 
 779                 /*
 780                  * Go through and update the paths for spares, now that we have
 781                  * them.
 782                  */
 783                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 784                     &nvroot) == 0);
 785                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 786                     &spares, &nspares) == 0) {
 787                         for (i = 0; i < nspares; i++) {
 788                                 if (fix_paths(spares[i], pl->names) != 0)
 789                                         goto nomem;
 790                         }
 791                 }
 792 
 793                 /*
 794                  * Update the paths for l2cache devices.
 795                  */
 796                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 797                     &l2cache, &nl2cache) == 0) {
 798                         for (i = 0; i < nl2cache; i++) {
 799                                 if (fix_paths(l2cache[i], pl->names) != 0)
 800                                         goto nomem;
 801                         }
 802                 }
 803 
 804                 /*
 805                  * Restore the original information read from the actual label.
 806                  */
 807                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 808                     DATA_TYPE_UINT64);
 809                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 810                     DATA_TYPE_STRING);
 811                 if (hostid != 0) {
 812                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 813                             hostid) == 0);
 814                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 815                             hostname) == 0);
 816                 }
 817 
 818 add_pool:
 819                 /*
 820                  * Add this pool to the list of configs.
 821                  */
 822                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 823                     &name) == 0);
 824                 if (nvlist_add_nvlist(ret, name, config) != 0)
 825                         goto nomem;
 826 
 827                 found_one = B_TRUE;
 828                 nvlist_free(config);
 829                 config = NULL;
 830         }
 831 
 832         if (!found_one) {
 833                 nvlist_free(ret);
 834                 ret = NULL;
 835         }
 836 
 837         return (ret);
 838 
 839 nomem:
 840         (void) no_memory(hdl);
 841 error:
 842         nvlist_free(config);
 843         nvlist_free(ret);
 844         for (c = 0; c < children; c++)
 845                 nvlist_free(child[c]);
 846         free(child);
 847 
 848         return (NULL);
 849 }
 850 
 851 /*
 852  * Return the offset of the given label.
 853  */
 854 static uint64_t
 855 label_offset(uint64_t size, int l)
 856 {
 857         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 858         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 859             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 860 }
 861 
 862 /*
 863  * Given a file descriptor, read the label information and return an nvlist
 864  * describing the configuration, if there is one.
 865  */
 866 int
 867 zpool_read_label(int fd, nvlist_t **config)
 868 {
 869         struct stat64 statbuf;
 870         int l;
 871         vdev_label_t *label;
 872         uint64_t state, txg, size;
 873 
 874         *config = NULL;
 875 
 876         if (fstat64(fd, &statbuf) == -1)
 877                 return (0);
 878         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 879 
 880         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 881                 return (-1);
 882 
 883         for (l = 0; l < VDEV_LABELS; l++) {
 884                 if (pread64(fd, label, sizeof (vdev_label_t),
 885                     label_offset(size, l)) != sizeof (vdev_label_t))
 886                         continue;
 887 
 888                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 889                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 890                         continue;
 891 
 892                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 893                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 894                         nvlist_free(*config);
 895                         continue;
 896                 }
 897 
 898                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 899                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 900                     &txg) != 0 || txg == 0)) {
 901                         nvlist_free(*config);
 902                         continue;
 903                 }
 904 
 905                 free(label);
 906                 return (0);
 907         }
 908 
 909         free(label);
 910         *config = NULL;
 911         return (0);
 912 }
 913 
 914 typedef struct rdsk_node {
 915         char *rn_name;
 916         int rn_dfd;
 917         libzfs_handle_t *rn_hdl;
 918         nvlist_t *rn_config;
 919         avl_tree_t *rn_avl;
 920         avl_node_t rn_node;
 921         boolean_t rn_nozpool;
 922 } rdsk_node_t;
 923 
 924 static int
 925 slice_cache_compare(const void *arg1, const void *arg2)
 926 {
 927         const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
 928         const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
 929         char *nm1slice, *nm2slice;
 930         int rv;
 931 
 932         /*
 933          * slices zero and two are the most likely to provide results,
 934          * so put those first
 935          */
 936         nm1slice = strstr(nm1, "s0");
 937         nm2slice = strstr(nm2, "s0");
 938         if (nm1slice && !nm2slice) {
 939                 return (-1);
 940         }
 941         if (!nm1slice && nm2slice) {
 942                 return (1);
 943         }
 944         nm1slice = strstr(nm1, "s2");
 945         nm2slice = strstr(nm2, "s2");
 946         if (nm1slice && !nm2slice) {
 947                 return (-1);
 948         }
 949         if (!nm1slice && nm2slice) {
 950                 return (1);
 951         }
 952 
 953         rv = strcmp(nm1, nm2);
 954         if (rv == 0)
 955                 return (0);
 956         return (rv > 0 ? 1 : -1);
 957 }
 958 
 959 static void
 960 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
 961     diskaddr_t size, uint_t blksz)
 962 {
 963         rdsk_node_t tmpnode;
 964         rdsk_node_t *node;
 965         char sname[MAXNAMELEN];
 966 
 967         tmpnode.rn_name = &sname[0];
 968         (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
 969             diskname, partno);
 970         /*
 971          * protect against division by zero for disk labels that
 972          * contain a bogus sector size
 973          */
 974         if (blksz == 0)
 975                 blksz = DEV_BSIZE;
 976         /* too small to contain a zpool? */
 977         if ((size < (SPA_MINDEVSIZE / blksz)) &&
 978             (node = avl_find(r, &tmpnode, NULL)))
 979                 node->rn_nozpool = B_TRUE;
 980 }
 981 
 982 static void
 983 nozpool_all_slices(avl_tree_t *r, const char *sname)
 984 {
 985         char diskname[MAXNAMELEN];
 986         char *ptr;
 987         int i;
 988 
 989         (void) strncpy(diskname, sname, MAXNAMELEN);
 990         if (((ptr = strrchr(diskname, 's')) == NULL) &&
 991             ((ptr = strrchr(diskname, 'p')) == NULL))
 992                 return;
 993         ptr[0] = 's';
 994         ptr[1] = '\0';
 995         for (i = 0; i < NDKMAP; i++)
 996                 check_one_slice(r, diskname, i, 0, 1);
 997         ptr[0] = 'p';
 998         for (i = 0; i <= FD_NUMPART; i++)
 999                 check_one_slice(r, diskname, i, 0, 1);
1000 }
1001 
1002 static void
1003 check_slices(avl_tree_t *r, int fd, const char *sname)
1004 {
1005         struct extvtoc vtoc;
1006         struct dk_gpt *gpt;
1007         char diskname[MAXNAMELEN];
1008         char *ptr;
1009         int i;
1010 
1011         (void) strncpy(diskname, sname, MAXNAMELEN);
1012         if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1013                 return;
1014         ptr[1] = '\0';
1015 
1016         if (read_extvtoc(fd, &vtoc) >= 0) {
1017                 for (i = 0; i < NDKMAP; i++)
1018                         check_one_slice(r, diskname, i,
1019                             vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1020         } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1021                 /*
1022                  * on x86 we'll still have leftover links that point
1023                  * to slices s[9-15], so use NDKMAP instead
1024                  */
1025                 for (i = 0; i < NDKMAP; i++)
1026                         check_one_slice(r, diskname, i,
1027                             gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1028                 /* nodes p[1-4] are never used with EFI labels */
1029                 ptr[0] = 'p';
1030                 for (i = 1; i <= FD_NUMPART; i++)
1031                         check_one_slice(r, diskname, i, 0, 1);
1032                 efi_free(gpt);
1033         }
1034 }
1035 
1036 static void
1037 zpool_open_func(void *arg)
1038 {
1039         rdsk_node_t *rn = arg;
1040         struct stat64 statbuf;
1041         nvlist_t *config;
1042         int fd;
1043 
1044         if (rn->rn_nozpool)
1045                 return;
1046         if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1047                 /* symlink to a device that's no longer there */
1048                 if (errno == ENOENT)
1049                         nozpool_all_slices(rn->rn_avl, rn->rn_name);
1050                 return;
1051         }
1052         /*
1053          * Ignore failed stats.  We only want regular
1054          * files, character devs and block devs.
1055          */
1056         if (fstat64(fd, &statbuf) != 0 ||
1057             (!S_ISREG(statbuf.st_mode) &&
1058             !S_ISCHR(statbuf.st_mode) &&
1059             !S_ISBLK(statbuf.st_mode))) {
1060                 (void) close(fd);
1061                 return;
1062         }
1063         /* this file is too small to hold a zpool */
1064         if (S_ISREG(statbuf.st_mode) &&
1065             statbuf.st_size < SPA_MINDEVSIZE) {
1066                 (void) close(fd);
1067                 return;
1068         } else if (!S_ISREG(statbuf.st_mode)) {
1069                 /*
1070                  * Try to read the disk label first so we don't have to
1071                  * open a bunch of minor nodes that can't have a zpool.
1072                  */
1073                 check_slices(rn->rn_avl, fd, rn->rn_name);
1074         }
1075 
1076         if ((zpool_read_label(fd, &config)) != 0) {
1077                 (void) close(fd);
1078                 (void) no_memory(rn->rn_hdl);
1079                 return;
1080         }
1081         (void) close(fd);
1082 
1083         rn->rn_config = config;
1084 }
1085 
1086 /*
1087  * Given a file descriptor, clear (zero) the label information.  This function
1088  * is currently only used in the appliance stack as part of the ZFS sysevent
1089  * module.
1090  */
1091 int
1092 zpool_clear_label(int fd)
1093 {
1094         struct stat64 statbuf;
1095         int l;
1096         vdev_label_t *label;
1097         uint64_t size;
1098 
1099         if (fstat64(fd, &statbuf) == -1)
1100                 return (0);
1101         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1102 
1103         if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1104                 return (-1);
1105 
1106         for (l = 0; l < VDEV_LABELS; l++) {
1107                 if (pwrite64(fd, label, sizeof (vdev_label_t),
1108                     label_offset(size, l)) != sizeof (vdev_label_t)) {
1109                         free(label);
1110                         return (-1);
1111                 }
1112         }
1113 
1114         free(label);
1115         return (0);
1116 }
1117 
1118 /*
1119  * Given a list of directories to search, find all pools stored on disk.  This
1120  * includes partial pools which are not available to import.  If no args are
1121  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1122  * poolname or guid (but not both) are provided by the caller when trying
1123  * to import a specific pool.
1124  */
1125 static nvlist_t *
1126 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1127 {
1128         int i, dirs = iarg->paths;
1129         struct dirent64 *dp;
1130         char path[MAXPATHLEN];
1131         char *end, **dir = iarg->path;
1132         size_t pathleft;
1133         nvlist_t *ret = NULL;
1134         static char *default_dir = "/dev/dsk";
1135         pool_list_t pools = { 0 };
1136         pool_entry_t *pe, *penext;
1137         vdev_entry_t *ve, *venext;
1138         config_entry_t *ce, *cenext;
1139         name_entry_t *ne, *nenext;
1140         avl_tree_t slice_cache;
1141         rdsk_node_t *slice;
1142         void *cookie;
1143 
1144         if (dirs == 0) {
1145                 dirs = 1;
1146                 dir = &default_dir;
1147         }
1148 
1149         /*
1150          * Go through and read the label configuration information from every
1151          * possible device, organizing the information according to pool GUID
1152          * and toplevel GUID.
1153          */
1154         for (i = 0; i < dirs; i++) {
1155                 tpool_t *t;
1156                 char *rdsk;
1157                 int dfd;
1158                 boolean_t config_failed = B_FALSE;
1159                 DIR *dirp;
1160 
1161                 /* use realpath to normalize the path */
1162                 if (realpath(dir[i], path) == 0) {
1163                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1164                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1165                         goto error;
1166                 }
1167                 end = &path[strlen(path)];
1168                 *end++ = '/';
1169                 *end = 0;
1170                 pathleft = &path[sizeof (path)] - end;
1171 
1172                 /*
1173                  * Using raw devices instead of block devices when we're
1174                  * reading the labels skips a bunch of slow operations during
1175                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1176                  */
1177                 if (strcmp(path, "/dev/dsk/") == 0)
1178                         rdsk = "/dev/rdsk/";
1179                 else
1180                         rdsk = path;
1181 
1182                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1183                     (dirp = fdopendir(dfd)) == NULL) {
1184                         if (dfd >= 0)
1185                                 (void) close(dfd);
1186                         zfs_error_aux(hdl, strerror(errno));
1187                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1188                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1189                             rdsk);
1190                         goto error;
1191                 }
1192 
1193                 avl_create(&slice_cache, slice_cache_compare,
1194                     sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1195                 /*
1196                  * This is not MT-safe, but we have no MT consumers of libzfs
1197                  */
1198                 while ((dp = readdir64(dirp)) != NULL) {
1199                         const char *name = dp->d_name;
1200                         if (name[0] == '.' &&
1201                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1202                                 continue;
1203 
1204                         slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1205                         slice->rn_name = zfs_strdup(hdl, name);
1206                         slice->rn_avl = &slice_cache;
1207                         slice->rn_dfd = dfd;
1208                         slice->rn_hdl = hdl;
1209                         slice->rn_nozpool = B_FALSE;
1210                         avl_add(&slice_cache, slice);
1211                 }
1212                 /*
1213                  * create a thread pool to do all of this in parallel;
1214                  * rn_nozpool is not protected, so this is racy in that
1215                  * multiple tasks could decide that the same slice can
1216                  * not hold a zpool, which is benign.  Also choose
1217                  * double the number of processors; we hold a lot of
1218                  * locks in the kernel, so going beyond this doesn't
1219                  * buy us much.
1220                  */
1221                 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1222                     0, NULL);
1223                 for (slice = avl_first(&slice_cache); slice;
1224                     (slice = avl_walk(&slice_cache, slice,
1225                     AVL_AFTER)))
1226                         (void) tpool_dispatch(t, zpool_open_func, slice);
1227                 tpool_wait(t);
1228                 tpool_destroy(t);
1229 
1230                 cookie = NULL;
1231                 while ((slice = avl_destroy_nodes(&slice_cache,
1232                     &cookie)) != NULL) {
1233                         if (slice->rn_config != NULL && !config_failed) {
1234                                 nvlist_t *config = slice->rn_config;
1235                                 boolean_t matched = B_TRUE;
1236 
1237                                 if (iarg->poolname != NULL) {
1238                                         char *pname;
1239 
1240                                         matched = nvlist_lookup_string(config,
1241                                             ZPOOL_CONFIG_POOL_NAME,
1242                                             &pname) == 0 &&
1243                                             strcmp(iarg->poolname, pname) == 0;
1244                                 } else if (iarg->guid != 0) {
1245                                         uint64_t this_guid;
1246 
1247                                         matched = nvlist_lookup_uint64(config,
1248                                             ZPOOL_CONFIG_POOL_GUID,
1249                                             &this_guid) == 0 &&
1250                                             iarg->guid == this_guid;
1251                                 }
1252                                 if (!matched) {
1253                                         nvlist_free(config);
1254                                 } else {
1255                                         /*
1256                                          * use the non-raw path for the config
1257                                          */
1258                                         (void) strlcpy(end, slice->rn_name,
1259                                             pathleft);
1260                                         if (add_config(hdl, &pools, path,
1261                                             config) != 0)
1262                                                 config_failed = B_TRUE;
1263                                 }
1264                         }
1265                         free(slice->rn_name);
1266                         free(slice);
1267                 }
1268                 avl_destroy(&slice_cache);
1269 
1270                 (void) closedir(dirp);
1271 
1272                 if (config_failed)
1273                         goto error;
1274         }
1275 
1276         ret = get_configs(hdl, &pools, iarg->can_be_active);
1277 
1278 error:
1279         for (pe = pools.pools; pe != NULL; pe = penext) {
1280                 penext = pe->pe_next;
1281                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1282                         venext = ve->ve_next;
1283                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1284                                 cenext = ce->ce_next;
1285                                 if (ce->ce_config)
1286                                         nvlist_free(ce->ce_config);
1287                                 free(ce);
1288                         }
1289                         free(ve);
1290                 }
1291                 free(pe);
1292         }
1293 
1294         for (ne = pools.names; ne != NULL; ne = nenext) {
1295                 nenext = ne->ne_next;
1296                 free(ne->ne_name);
1297                 free(ne);
1298         }
1299 
1300         return (ret);
1301 }
1302 
1303 nvlist_t *
1304 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1305 {
1306         importargs_t iarg = { 0 };
1307 
1308         iarg.paths = argc;
1309         iarg.path = argv;
1310 
1311         return (zpool_find_import_impl(hdl, &iarg));
1312 }
1313 
1314 /*
1315  * Given a cache file, return the contents as a list of importable pools.
1316  * poolname or guid (but not both) are provided by the caller when trying
1317  * to import a specific pool.
1318  */
1319 nvlist_t *
1320 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1321     char *poolname, uint64_t guid)
1322 {
1323         char *buf;
1324         int fd;
1325         struct stat64 statbuf;
1326         nvlist_t *raw, *src, *dst;
1327         nvlist_t *pools;
1328         nvpair_t *elem;
1329         char *name;
1330         uint64_t this_guid;
1331         boolean_t active;
1332 
1333         verify(poolname == NULL || guid == 0);
1334 
1335         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1336                 zfs_error_aux(hdl, "%s", strerror(errno));
1337                 (void) zfs_error(hdl, EZFS_BADCACHE,
1338                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1339                 return (NULL);
1340         }
1341 
1342         if (fstat64(fd, &statbuf) != 0) {
1343                 zfs_error_aux(hdl, "%s", strerror(errno));
1344                 (void) close(fd);
1345                 (void) zfs_error(hdl, EZFS_BADCACHE,
1346                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1347                 return (NULL);
1348         }
1349 
1350         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1351                 (void) close(fd);
1352                 return (NULL);
1353         }
1354 
1355         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1356                 (void) close(fd);
1357                 free(buf);
1358                 (void) zfs_error(hdl, EZFS_BADCACHE,
1359                     dgettext(TEXT_DOMAIN,
1360                     "failed to read cache file contents"));
1361                 return (NULL);
1362         }
1363 
1364         (void) close(fd);
1365 
1366         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1367                 free(buf);
1368                 (void) zfs_error(hdl, EZFS_BADCACHE,
1369                     dgettext(TEXT_DOMAIN,
1370                     "invalid or corrupt cache file contents"));
1371                 return (NULL);
1372         }
1373 
1374         free(buf);
1375 
1376         /*
1377          * Go through and get the current state of the pools and refresh their
1378          * state.
1379          */
1380         if (nvlist_alloc(&pools, 0, 0) != 0) {
1381                 (void) no_memory(hdl);
1382                 nvlist_free(raw);
1383                 return (NULL);
1384         }
1385 
1386         elem = NULL;
1387         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1388                 src = fnvpair_value_nvlist(elem);
1389 
1390                 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1391                 if (poolname != NULL && strcmp(poolname, name) != 0)
1392                         continue;
1393 
1394                 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1395                 if (guid != 0 && guid != this_guid)
1396                         continue;
1397 
1398                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1399                         nvlist_free(raw);
1400                         nvlist_free(pools);
1401                         return (NULL);
1402                 }
1403 
1404                 if (active)
1405                         continue;
1406 
1407                 if ((dst = refresh_config(hdl, src)) == NULL) {
1408                         nvlist_free(raw);
1409                         nvlist_free(pools);
1410                         return (NULL);
1411                 }
1412 
1413                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1414                         (void) no_memory(hdl);
1415                         nvlist_free(dst);
1416                         nvlist_free(raw);
1417                         nvlist_free(pools);
1418                         return (NULL);
1419                 }
1420                 nvlist_free(dst);
1421         }
1422 
1423         nvlist_free(raw);
1424         return (pools);
1425 }
1426 
1427 static int
1428 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1429 {
1430         importargs_t *import = data;
1431         int found = 0;
1432 
1433         if (import->poolname != NULL) {
1434                 char *pool_name;
1435 
1436                 verify(nvlist_lookup_string(zhp->zpool_config,
1437                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1438                 if (strcmp(pool_name, import->poolname) == 0)
1439                         found = 1;
1440         } else {
1441                 uint64_t pool_guid;
1442 
1443                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1444                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1445                 if (pool_guid == import->guid)
1446                         found = 1;
1447         }
1448 
1449         zpool_close(zhp);
1450         return (found);
1451 }
1452 
1453 nvlist_t *
1454 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1455 {
1456         verify(import->poolname == NULL || import->guid == 0);
1457 
1458         if (import->unique)
1459                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1460 
1461         if (import->cachefile != NULL)
1462                 return (zpool_find_import_cached(hdl, import->cachefile,
1463                     import->poolname, import->guid));
1464 
1465         return (zpool_find_import_impl(hdl, import));
1466 }
1467 
1468 boolean_t
1469 find_guid(nvlist_t *nv, uint64_t guid)
1470 {
1471         uint64_t tmp;
1472         nvlist_t **child;
1473         uint_t c, children;
1474 
1475         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1476         if (tmp == guid)
1477                 return (B_TRUE);
1478 
1479         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1480             &child, &children) == 0) {
1481                 for (c = 0; c < children; c++)
1482                         if (find_guid(child[c], guid))
1483                                 return (B_TRUE);
1484         }
1485 
1486         return (B_FALSE);
1487 }
1488 
1489 typedef struct aux_cbdata {
1490         const char      *cb_type;
1491         uint64_t        cb_guid;
1492         zpool_handle_t  *cb_zhp;
1493 } aux_cbdata_t;
1494 
1495 static int
1496 find_aux(zpool_handle_t *zhp, void *data)
1497 {
1498         aux_cbdata_t *cbp = data;
1499         nvlist_t **list;
1500         uint_t i, count;
1501         uint64_t guid;
1502         nvlist_t *nvroot;
1503 
1504         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1505             &nvroot) == 0);
1506 
1507         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1508             &list, &count) == 0) {
1509                 for (i = 0; i < count; i++) {
1510                         verify(nvlist_lookup_uint64(list[i],
1511                             ZPOOL_CONFIG_GUID, &guid) == 0);
1512                         if (guid == cbp->cb_guid) {
1513                                 cbp->cb_zhp = zhp;
1514                                 return (1);
1515                         }
1516                 }
1517         }
1518 
1519         zpool_close(zhp);
1520         return (0);
1521 }
1522 
1523 /*
1524  * Determines if the pool is in use.  If so, it returns true and the state of
1525  * the pool as well as the name of the pool.  Both strings are allocated and
1526  * must be freed by the caller.
1527  */
1528 int
1529 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1530     boolean_t *inuse)
1531 {
1532         nvlist_t *config;
1533         char *name;
1534         boolean_t ret;
1535         uint64_t guid, vdev_guid;
1536         zpool_handle_t *zhp;
1537         nvlist_t *pool_config;
1538         uint64_t stateval, isspare;
1539         aux_cbdata_t cb = { 0 };
1540         boolean_t isactive;
1541 
1542         *inuse = B_FALSE;
1543 
1544         if (zpool_read_label(fd, &config) != 0) {
1545                 (void) no_memory(hdl);
1546                 return (-1);
1547         }
1548 
1549         if (config == NULL)
1550                 return (0);
1551 
1552         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1553             &stateval) == 0);
1554         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1555             &vdev_guid) == 0);
1556 
1557         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1558                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1559                     &name) == 0);
1560                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1561                     &guid) == 0);
1562         }
1563 
1564         switch (stateval) {
1565         case POOL_STATE_EXPORTED:
1566                 /*
1567                  * A pool with an exported state may in fact be imported
1568                  * read-only, so check the in-core state to see if it's
1569                  * active and imported read-only.  If it is, set
1570                  * its state to active.
1571                  */
1572                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1573                     (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1574                         if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1575                                 stateval = POOL_STATE_ACTIVE;
1576 
1577                         /*
1578                          * All we needed the zpool handle for is the
1579                          * readonly prop check.
1580                          */
1581                         zpool_close(zhp);
1582                 }
1583 
1584                 ret = B_TRUE;
1585                 break;
1586 
1587         case POOL_STATE_ACTIVE:
1588                 /*
1589                  * For an active pool, we have to determine if it's really part
1590                  * of a currently active pool (in which case the pool will exist
1591                  * and the guid will be the same), or whether it's part of an
1592                  * active pool that was disconnected without being explicitly
1593                  * exported.
1594                  */
1595                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1596                         nvlist_free(config);
1597                         return (-1);
1598                 }
1599 
1600                 if (isactive) {
1601                         /*
1602                          * Because the device may have been removed while
1603                          * offlined, we only report it as active if the vdev is
1604                          * still present in the config.  Otherwise, pretend like
1605                          * it's not in use.
1606                          */
1607                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1608                             (pool_config = zpool_get_config(zhp, NULL))
1609                             != NULL) {
1610                                 nvlist_t *nvroot;
1611 
1612                                 verify(nvlist_lookup_nvlist(pool_config,
1613                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1614                                 ret = find_guid(nvroot, vdev_guid);
1615                         } else {
1616                                 ret = B_FALSE;
1617                         }
1618 
1619                         /*
1620                          * If this is an active spare within another pool, we
1621                          * treat it like an unused hot spare.  This allows the
1622                          * user to create a pool with a hot spare that currently
1623                          * in use within another pool.  Since we return B_TRUE,
1624                          * libdiskmgt will continue to prevent generic consumers
1625                          * from using the device.
1626                          */
1627                         if (ret && nvlist_lookup_uint64(config,
1628                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1629                                 stateval = POOL_STATE_SPARE;
1630 
1631                         if (zhp != NULL)
1632                                 zpool_close(zhp);
1633                 } else {
1634                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1635                         ret = B_TRUE;
1636                 }
1637                 break;
1638 
1639         case POOL_STATE_SPARE:
1640                 /*
1641                  * For a hot spare, it can be either definitively in use, or
1642                  * potentially active.  To determine if it's in use, we iterate
1643                  * over all pools in the system and search for one with a spare
1644                  * with a matching guid.
1645                  *
1646                  * Due to the shared nature of spares, we don't actually report
1647                  * the potentially active case as in use.  This means the user
1648                  * can freely create pools on the hot spares of exported pools,
1649                  * but to do otherwise makes the resulting code complicated, and
1650                  * we end up having to deal with this case anyway.
1651                  */
1652                 cb.cb_zhp = NULL;
1653                 cb.cb_guid = vdev_guid;
1654                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1655                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1656                         name = (char *)zpool_get_name(cb.cb_zhp);
1657                         ret = B_TRUE;
1658                 } else {
1659                         ret = B_FALSE;
1660                 }
1661                 break;
1662 
1663         case POOL_STATE_L2CACHE:
1664 
1665                 /*
1666                  * Check if any pool is currently using this l2cache device.
1667                  */
1668                 cb.cb_zhp = NULL;
1669                 cb.cb_guid = vdev_guid;
1670                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1671                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1672                         name = (char *)zpool_get_name(cb.cb_zhp);
1673                         ret = B_TRUE;
1674                 } else {
1675                         ret = B_FALSE;
1676                 }
1677                 break;
1678 
1679         default:
1680                 ret = B_FALSE;
1681         }
1682 
1683 
1684         if (ret) {
1685                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1686                         if (cb.cb_zhp)
1687                                 zpool_close(cb.cb_zhp);
1688                         nvlist_free(config);
1689                         return (-1);
1690                 }
1691                 *state = (pool_state_t)stateval;
1692         }
1693 
1694         if (cb.cb_zhp)
1695                 zpool_close(cb.cb_zhp);
1696 
1697         nvlist_free(config);
1698         *inuse = ret;
1699         return (0);
1700 }