lib/libzfs/libzfs_import.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  24  * Copyright (c) 2011 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
  58 #include <sys/vdev_impl.h>
  59 #ifdef HAVE_LIBBLKID
  60 #include <blkid/blkid.h>
  61 #endif
  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         uint64_t                ne_order;
  91         struct name_entry       *ne_next;
  92 } name_entry_t;
  93
  94 typedef struct pool_list {
  95         pool_entry_t            *pools;
  96         name_entry_t            *names;
  97 } pool_list_t;
  98
  99 static char *
 100 get_devid(const char *path)
 101 {
 102         int fd;
 103         ddi_devid_t devid;
 104         char *minor, *ret;
 105
 106         if ((fd = open(path, O_RDONLY)) < 0)
 107                 return (NULL);
 108
 109         minor = NULL;
 110         ret = NULL;
 111         if (devid_get(fd, &devid) == 0) {
 112                 if (devid_get_minor_name(fd, &minor) == 0)
 113                         ret = devid_str_encode(devid, minor);
 114                 if (minor != NULL)
 115                         devid_str_free(minor);
 116                 devid_free(devid);
 117         }
 118         (void) close(fd);
 119
 120         return (ret);
 121 }
 122
 123
 124 /*
 125  * Go through and fix up any path and/or devid information for the given vdev
 126  * configuration.
 127  */
 128 static int
 129 fix_paths(nvlist_t *nv, name_entry_t *names)
 130 {
 131         nvlist_t **child;
 132         uint_t c, children;
 133         uint64_t guid;
 134         name_entry_t *ne, *best;
 135         char *path, *devid;
 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 partitions or multiple paths to the
 152          * same disk.  In this case we prefer to use the path name which
 153          * matches the ZPOOL_CONFIG_PATH.  If no matching entry is found we
 154          * use the lowest order device which corresponds to the first match
 155          * while traversing the ZPOOL_IMPORT_PATH search path.
 156          */
 157         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 158         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 159                 path = NULL;
 160
 161         best = NULL;
 162         for (ne = names; ne != NULL; ne = ne->ne_next) {
 163                 if (ne->ne_guid == guid) {
 164
 165                         if (path == NULL) {
 166                                 best = ne;
 167                                 break;
 168                         }
 169
 170                         if ((strlen(path) == strlen(ne->ne_name)) &&
 171                             !strncmp(path, ne->ne_name, strlen(path))) {
 172                                 best = ne;
 173                                 break;
 174                         }
 175
 176                         if (best == NULL || ne->ne_order < best->ne_order)
 177                                 best = ne;
 178                 }
 179         }
 180
 181         if (best == NULL)
 182                 return (0);
 183
 184         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 185                 return (-1);
 186
 187         if ((devid = get_devid(best->ne_name)) == NULL) {
 188                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 189         } else {
 190                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
 191                         return (-1);
 192                 devid_str_free(devid);
 193         }
 194
 195         return (0);
 196 }
 197
 198 /*
 199  * Add the given configuration to the list of known devices.
 200  */
 201 static int
 202 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 203     int order, nvlist_t *config)
 204 {
 205         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 206         pool_entry_t *pe;
 207         vdev_entry_t *ve;
 208         config_entry_t *ce;
 209         name_entry_t *ne;
 210
 211         /*
 212          * If this is a hot spare not currently in use or level 2 cache
 213          * device, add it to the list of names to translate, but don't do
 214          * anything else.
 215          */
 216         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 217             &state) == 0 &&
 218             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 219             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 220                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 221                         return (-1);
 222
 223                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 224                         free(ne);
 225                         return (-1);
 226                 }
 227                 ne->ne_guid = vdev_guid;
 228                 ne->ne_order = order;
 229                 ne->ne_next = pl->names;
 230                 pl->names = ne;
 231                 return (0);
 232         }
 233
 234         /*
 235          * If we have a valid config but cannot read any of these fields, then
 236          * it means we have a half-initialized label.  In vdev_label_init()
 237          * we write a label with txg == 0 so that we can identify the device
 238          * in case the user refers to the same disk later on.  If we fail to
 239          * create the pool, we'll be left with a label in this state
 240          * which should not be considered part of a valid pool.
 241          */
 242         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 243             &pool_guid) != 0 ||
 244             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 245             &vdev_guid) != 0 ||
 246             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 247             &top_guid) != 0 ||
 248             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 249             &txg) != 0 || txg == 0) {
 250                 nvlist_free(config);
 251                 return (0);
 252         }
 253
 254         /*
 255          * First, see if we know about this pool.  If not, then add it to the
 256          * list of known pools.
 257          */
 258         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 259                 if (pe->pe_guid == pool_guid)
 260                         break;
 261         }
 262
 263         if (pe == NULL) {
 264                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 265                         nvlist_free(config);
 266                         return (-1);
 267                 }
 268                 pe->pe_guid = pool_guid;
 269                 pe->pe_next = pl->pools;
 270                 pl->pools = pe;
 271         }
 272
 273         /*
 274          * Second, see if we know about this toplevel vdev.  Add it if its
 275          * missing.
 276          */
 277         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 278                 if (ve->ve_guid == top_guid)
 279                         break;
 280         }
 281
 282         if (ve == NULL) {
 283                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 284                         nvlist_free(config);
 285                         return (-1);
 286                 }
 287                 ve->ve_guid = top_guid;
 288                 ve->ve_next = pe->pe_vdevs;
 289                 pe->pe_vdevs = ve;
 290         }
 291
 292         /*
 293          * Third, see if we have a config with a matching transaction group.  If
 294          * so, then we do nothing.  Otherwise, add it to the list of known
 295          * configs.
 296          */
 297         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 298                 if (ce->ce_txg == txg)
 299                         break;
 300         }
 301
 302         if (ce == NULL) {
 303                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 304                         nvlist_free(config);
 305                         return (-1);
 306                 }
 307                 ce->ce_txg = txg;
 308                 ce->ce_config = config;
 309                 ce->ce_next = ve->ve_configs;
 310                 ve->ve_configs = ce;
 311         } else {
 312                 nvlist_free(config);
 313         }
 314
 315         /*
 316          * At this point we've successfully added our config to the list of
 317          * known configs.  The last thing to do is add the vdev guid -> path
 318          * mappings so that we can fix up the configuration as necessary before
 319          * doing the import.
 320          */
 321         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 322                 return (-1);
 323
 324         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 325                 free(ne);
 326                 return (-1);
 327         }
 328
 329         ne->ne_guid = vdev_guid;
 330         ne->ne_order = order;
 331         ne->ne_next = pl->names;
 332         pl->names = ne;
 333
 334         return (0);
 335 }
 336
 337 /*
 338  * Returns true if the named pool matches the given GUID.
 339  */
 340 static int
 341 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 342     boolean_t *isactive)
 343 {
 344         zpool_handle_t *zhp;
 345         uint64_t theguid;
 346
 347         if (zpool_open_silent(hdl, name, &zhp) != 0)
 348                 return (-1);
 349
 350         if (zhp == NULL) {
 351                 *isactive = B_FALSE;
 352                 return (0);
 353         }
 354
 355         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 356             &theguid) == 0);
 357
 358         zpool_close(zhp);
 359
 360         *isactive = (theguid == guid);
 361         return (0);
 362 }
 363
 364 static nvlist_t *
 365 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 366 {
 367         nvlist_t *nvl;
 368         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
 369         int err;
 370
 371         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 372                 return (NULL);
 373
 374         if (zcmd_alloc_dst_nvlist(hdl, &zc,
 375             zc.zc_nvlist_conf_size * 2) != 0) {
 376                 zcmd_free_nvlists(&zc);
 377                 return (NULL);
 378         }
 379
 380         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 381             &zc)) != 0 && errno == ENOMEM) {
 382                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 383                         zcmd_free_nvlists(&zc);
 384                         return (NULL);
 385                 }
 386         }
 387
 388         if (err) {
 389                 zcmd_free_nvlists(&zc);
 390                 return (NULL);
 391         }
 392
 393         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 394                 zcmd_free_nvlists(&zc);
 395                 return (NULL);
 396         }
 397
 398         zcmd_free_nvlists(&zc);
 399         return (nvl);
 400 }
 401
 402 /*
 403  * Determine if the vdev id is a hole in the namespace.
 404  */
 405 boolean_t
 406 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
 407 {
 408         int c;
 409
 410         for (c = 0; c < holes; c++) {
 411
 412                 /* Top-level is a hole */
 413                 if (hole_array[c] == id)
 414                         return (B_TRUE);
 415         }
 416         return (B_FALSE);
 417 }
 418
 419 /*
 420  * Convert our list of pools into the definitive set of configurations.  We
 421  * start by picking the best config for each toplevel vdev.  Once that's done,
 422  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 423  * pass to fix up any incorrect paths, and then add it to the main list to
 424  * return to the user.
 425  */
 426 static nvlist_t *
 427 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 428 {
 429         pool_entry_t *pe;
 430         vdev_entry_t *ve;
 431         config_entry_t *ce;
 432         nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
 433         nvlist_t **spares, **l2cache;
 434         uint_t i, nspares, nl2cache;
 435         boolean_t config_seen;
 436         uint64_t best_txg;
 437         char *name, *hostname, *comment;
 438         uint64_t version, guid;
 439         uint_t children = 0;
 440         nvlist_t **child = NULL;
 441         uint_t holes;
 442         uint64_t *hole_array, max_id;
 443         uint_t c;
 444         boolean_t isactive;
 445         uint64_t hostid;
 446         nvlist_t *nvl;
 447         boolean_t found_one = B_FALSE;
 448         boolean_t valid_top_config = B_FALSE;
 449
 450         if (nvlist_alloc(&ret, 0, 0) != 0)
 451                 goto nomem;
 452
 453         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 454                 uint64_t id, max_txg = 0;
 455
 456                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 457                         goto nomem;
 458                 config_seen = B_FALSE;
 459
 460                 /*
 461                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 462                  * from the first one we find, and then go through the rest and
 463                  * add them as necessary to the 'vdevs' member of the config.
 464                  */
 465                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 466
 467                         /*
 468                          * Determine the best configuration for this vdev by
 469                          * selecting the config with the latest transaction
 470                          * group.
 471                          */
 472                         best_txg = 0;
 473                         for (ce = ve->ve_configs; ce != NULL;
 474                             ce = ce->ce_next) {
 475
 476                                 if (ce->ce_txg > best_txg) {
 477                                         tmp = ce->ce_config;
 478                                         best_txg = ce->ce_txg;
 479                                 }
 480                         }
 481
 482                         /*
 483                          * We rely on the fact that the max txg for the
 484                          * pool will contain the most up-to-date information
 485                          * about the valid top-levels in the vdev namespace.
 486                          */
 487                         if (best_txg > max_txg) {
 488                                 (void) nvlist_remove(config,
 489                                     ZPOOL_CONFIG_VDEV_CHILDREN,
 490                                     DATA_TYPE_UINT64);
 491                                 (void) nvlist_remove(config,
 492                                     ZPOOL_CONFIG_HOLE_ARRAY,
 493                                     DATA_TYPE_UINT64_ARRAY);
 494
 495                                 max_txg = best_txg;
 496                                 hole_array = NULL;
 497                                 holes = 0;
 498                                 max_id = 0;
 499                                 valid_top_config = B_FALSE;
 500
 501                                 if (nvlist_lookup_uint64(tmp,
 502                                     ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
 503                                         verify(nvlist_add_uint64(config,
 504                                             ZPOOL_CONFIG_VDEV_CHILDREN,
 505                                             max_id) == 0);
 506                                         valid_top_config = B_TRUE;
 507                                 }
 508
 509                                 if (nvlist_lookup_uint64_array(tmp,
 510                                     ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
 511                                     &holes) == 0) {
 512                                         verify(nvlist_add_uint64_array(config,
 513                                             ZPOOL_CONFIG_HOLE_ARRAY,
 514                                             hole_array, holes) == 0);
 515                                 }
 516                         }
 517
 518                         if (!config_seen) {
 519                                 /*
 520                                  * Copy the relevant pieces of data to the pool
 521                                  * configuration:
 522                                  *
 523                                  *      version
 524                                  *      pool guid
 525                                  *      name
 526                                  *      comment (if available)
 527                                  *      pool state
 528                                  *      hostid (if available)
 529                                  *      hostname (if available)
 530                                  */
 531                                 uint64_t state;
 532
 533                                 verify(nvlist_lookup_uint64(tmp,
 534                                     ZPOOL_CONFIG_VERSION, &version) == 0);
 535                                 if (nvlist_add_uint64(config,
 536                                     ZPOOL_CONFIG_VERSION, version) != 0)
 537                                         goto nomem;
 538                                 verify(nvlist_lookup_uint64(tmp,
 539                                     ZPOOL_CONFIG_POOL_GUID, &guid) == 0);
 540                                 if (nvlist_add_uint64(config,
 541                                     ZPOOL_CONFIG_POOL_GUID, guid) != 0)
 542                                         goto nomem;
 543                                 verify(nvlist_lookup_string(tmp,
 544                                     ZPOOL_CONFIG_POOL_NAME, &name) == 0);
 545                                 if (nvlist_add_string(config,
 546                                     ZPOOL_CONFIG_POOL_NAME, name) != 0)
 547                                         goto nomem;
 548
 549                                 /*
 550                                  * COMMENT is optional, don't bail if it's not
 551                                  * there, instead, set it to NULL.
 552                                  */
 553                                 if (nvlist_lookup_string(tmp,
 554                                     ZPOOL_CONFIG_COMMENT, &comment) != 0)
 555                                         comment = NULL;
 556                                 else if (nvlist_add_string(config,
 557                                     ZPOOL_CONFIG_COMMENT, comment) != 0)
 558                                         goto nomem;
 559
 560                                 verify(nvlist_lookup_uint64(tmp,
 561                                     ZPOOL_CONFIG_POOL_STATE, &state) == 0);
 562                                 if (nvlist_add_uint64(config,
 563                                     ZPOOL_CONFIG_POOL_STATE, state) != 0)
 564                                         goto nomem;
 565
 566                                 hostid = 0;
 567                                 if (nvlist_lookup_uint64(tmp,
 568                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 569                                         if (nvlist_add_uint64(config,
 570                                             ZPOOL_CONFIG_HOSTID, hostid) != 0)
 571                                                 goto nomem;
 572                                         verify(nvlist_lookup_string(tmp,
 573                                             ZPOOL_CONFIG_HOSTNAME,
 574                                             &hostname) == 0);
 575                                         if (nvlist_add_string(config,
 576                                             ZPOOL_CONFIG_HOSTNAME,
 577                                             hostname) != 0)
 578                                                 goto nomem;
 579                                 }
 580
 581                                 config_seen = B_TRUE;
 582                         }
 583
 584                         /*
 585                          * Add this top-level vdev to the child array.
 586                          */
 587                         verify(nvlist_lookup_nvlist(tmp,
 588                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 589                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 590                             &id) == 0);
 591
 592                         if (id >= children) {
 593                                 nvlist_t **newchild;
 594
 595                                 newchild = zfs_alloc(hdl, (id + 1) *
 596                                     sizeof (nvlist_t *));
 597                                 if (newchild == NULL)
 598                                         goto nomem;
 599
 600                                 for (c = 0; c < children; c++)
 601                                         newchild[c] = child[c];
 602
 603                                 free(child);
 604                                 child = newchild;
 605                                 children = id + 1;
 606                         }
 607                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 608                                 goto nomem;
 609
 610                 }
 611
 612                 /*
 613                  * If we have information about all the top-levels then
 614                  * clean up the nvlist which we've constructed. This
 615                  * means removing any extraneous devices that are
 616                  * beyond the valid range or adding devices to the end
 617                  * of our array which appear to be missing.
 618                  */
 619                 if (valid_top_config) {
 620                         if (max_id < children) {
 621                                 for (c = max_id; c < children; c++)
 622                                         nvlist_free(child[c]);
 623                                 children = max_id;
 624                         } else if (max_id > children) {
 625                                 nvlist_t **newchild;
 626
 627                                 newchild = zfs_alloc(hdl, (max_id) *
 628                                     sizeof (nvlist_t *));
 629                                 if (newchild == NULL)
 630                                         goto nomem;
 631
 632                                 for (c = 0; c < children; c++)
 633                                         newchild[c] = child[c];
 634
 635                                 free(child);
 636                                 child = newchild;
 637                                 children = max_id;
 638                         }
 639                 }
 640
 641                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 642                     &guid) == 0);
 643
 644                 /*
 645                  * The vdev namespace may contain holes as a result of
 646                  * device removal. We must add them back into the vdev
 647                  * tree before we process any missing devices.
 648                  */
 649                 if (holes > 0) {
 650                         ASSERT(valid_top_config);
 651
 652                         for (c = 0; c < children; c++) {
 653                                 nvlist_t *holey;
 654
 655                                 if (child[c] != NULL ||
 656                                     !vdev_is_hole(hole_array, holes, c))
 657                                         continue;
 658
 659                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 660                                     0) != 0)
 661                                         goto nomem;
 662
 663                                 /*
 664                                  * Holes in the namespace are treated as
 665                                  * "hole" top-level vdevs and have a
 666                                  * special flag set on them.
 667                                  */
 668                                 if (nvlist_add_string(holey,
 669                                     ZPOOL_CONFIG_TYPE,
 670                                     VDEV_TYPE_HOLE) != 0 ||
 671                                     nvlist_add_uint64(holey,
 672                                     ZPOOL_CONFIG_ID, c) != 0 ||
 673                                     nvlist_add_uint64(holey,
 674                                     ZPOOL_CONFIG_GUID, 0ULL) != 0)
 675                                         goto nomem;
 676                                 child[c] = holey;
 677                         }
 678                 }
 679
 680                 /*
 681                  * Look for any missing top-level vdevs.  If this is the case,
 682                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 683                  * simply compress the child array, because the kernel performs
 684                  * certain checks to make sure the vdev IDs match their location
 685                  * in the configuration.
 686                  */
 687                 for (c = 0; c < children; c++) {
 688                         if (child[c] == NULL) {
 689                                 nvlist_t *missing;
 690                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 691                                     0) != 0)
 692                                         goto nomem;
 693                                 if (nvlist_add_string(missing,
 694                                     ZPOOL_CONFIG_TYPE,
 695                                     VDEV_TYPE_MISSING) != 0 ||
 696                                     nvlist_add_uint64(missing,
 697                                     ZPOOL_CONFIG_ID, c) != 0 ||
 698                                     nvlist_add_uint64(missing,
 699                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 700                                         nvlist_free(missing);
 701                                         goto nomem;
 702                                 }
 703                                 child[c] = missing;
 704                         }
 705                 }
 706
 707                 /*
 708                  * Put all of this pool's top-level vdevs into a root vdev.
 709                  */
 710                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 711                         goto nomem;
 712                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 713                     VDEV_TYPE_ROOT) != 0 ||
 714                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 715                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 716                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 717                     child, children) != 0) {
 718                         nvlist_free(nvroot);
 719                         goto nomem;
 720                 }
 721
 722                 for (c = 0; c < children; c++)
 723                         nvlist_free(child[c]);
 724                 free(child);
 725                 children = 0;
 726                 child = NULL;
 727
 728                 /*
 729                  * Go through and fix up any paths and/or devids based on our
 730                  * known list of vdev GUID -> path mappings.
 731                  */
 732                 if (fix_paths(nvroot, pl->names) != 0) {
 733                         nvlist_free(nvroot);
 734                         goto nomem;
 735                 }
 736
 737                 /*
 738                  * Add the root vdev to this pool's configuration.
 739                  */
 740                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 741                     nvroot) != 0) {
 742                         nvlist_free(nvroot);
 743                         goto nomem;
 744                 }
 745                 nvlist_free(nvroot);
 746
 747                 /*
 748                  * zdb uses this path to report on active pools that were
 749                  * imported or created using -R.
 750                  */
 751                 if (active_ok)
 752                         goto add_pool;
 753
 754                 /*
 755                  * Determine if this pool is currently active, in which case we
 756                  * can't actually import it.
 757                  */
 758                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 759                     &name) == 0);
 760                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 761                     &guid) == 0);
 762
 763                 if (pool_active(hdl, name, guid, &isactive) != 0)
 764                         goto error;
 765
 766                 if (isactive) {
 767                         nvlist_free(config);
 768                         config = NULL;
 769                         continue;
 770                 }
 771
 772                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 773                         nvlist_free(config);
 774                         config = NULL;
 775                         continue;
 776                 }
 777
 778                 nvlist_free(config);
 779                 config = nvl;
 780
 781                 /*
 782                  * Go through and update the paths for spares, now that we have
 783                  * them.
 784                  */
 785                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 786                     &nvroot) == 0);
 787                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 788                     &spares, &nspares) == 0) {
 789                         for (i = 0; i < nspares; i++) {
 790                                 if (fix_paths(spares[i], pl->names) != 0)
 791                                         goto nomem;
 792                         }
 793                 }
 794
 795                 /*
 796                  * Update the paths for l2cache devices.
 797                  */
 798                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 799                     &l2cache, &nl2cache) == 0) {
 800                         for (i = 0; i < nl2cache; i++) {
 801                                 if (fix_paths(l2cache[i], pl->names) != 0)
 802                                         goto nomem;
 803                         }
 804                 }
 805
 806                 /*
 807                  * Restore the original information read from the actual label.
 808                  */
 809                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 810                     DATA_TYPE_UINT64);
 811                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 812                     DATA_TYPE_STRING);
 813                 if (hostid != 0) {
 814                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 815                             hostid) == 0);
 816                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 817                             hostname) == 0);
 818                 }
 819
 820 add_pool:
 821                 /*
 822                  * Add this pool to the list of configs.
 823                  */
 824                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 825                     &name) == 0);
 826                 if (nvlist_add_nvlist(ret, name, config) != 0)
 827                         goto nomem;
 828
 829                 found_one = B_TRUE;
 830                 nvlist_free(config);
 831                 config = NULL;
 832         }
 833
 834         if (!found_one) {
 835                 nvlist_free(ret);
 836                 ret = NULL;
 837         }
 838
 839         return (ret);
 840
 841 nomem:
 842         (void) no_memory(hdl);
 843 error:
 844         nvlist_free(config);
 845         nvlist_free(ret);
 846         for (c = 0; c < children; c++)
 847                 nvlist_free(child[c]);
 848         free(child);
 849
 850         return (NULL);
 851 }
 852
 853 /*
 854  * Return the offset of the given label.
 855  */
 856 static uint64_t
 857 label_offset(uint64_t size, int l)
 858 {
 859         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 860         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 861             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 862 }
 863
 864 /*
 865  * Given a file descriptor, read the label information and return an nvlist
 866  * describing the configuration, if there is one.
 867  */
 868 int
 869 zpool_read_label(int fd, nvlist_t **config)
 870 {
 871         struct stat64 statbuf;
 872         int l;
 873         vdev_label_t *label;
 874         uint64_t state, txg, size;
 875
 876         *config = NULL;
 877
 878         if (fstat64(fd, &statbuf) == -1)
 879                 return (0);
 880         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 881
 882         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 883                 return (-1);
 884
 885         for (l = 0; l < VDEV_LABELS; l++) {
 886                 if (pread64(fd, label, sizeof (vdev_label_t),
 887                     label_offset(size, l)) != sizeof (vdev_label_t))
 888                         continue;
 889
 890                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 891                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 892                         continue;
 893
 894                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 895                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 896                         nvlist_free(*config);
 897                         continue;
 898                 }
 899
 900                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 901                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 902                     &txg) != 0 || txg == 0)) {
 903                         nvlist_free(*config);
 904                         continue;
 905                 }
 906
 907                 free(label);
 908                 return (0);
 909         }
 910
 911         free(label);
 912         *config = NULL;
 913         return (0);
 914 }
 915
 916 #ifdef HAVE_LIBBLKID
 917 /*
 918  * Use libblkid to quickly search for zfs devices
 919  */
 920 static int
 921 zpool_find_import_blkid(libzfs_handle_t *hdl, pool_list_t *pools)
 922 {
 923         blkid_cache cache;
 924         blkid_dev_iterate iter;
 925         blkid_dev dev;
 926         const char *devname;
 927         nvlist_t *config;
 928         int fd, err;
 929
 930         err = blkid_get_cache(&cache, NULL);
 931         if (err != 0) {
 932                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 933                     dgettext(TEXT_DOMAIN, "blkid_get_cache() %d"), err);
 934                 goto err_blkid1;
 935         }
 936
 937         err = blkid_probe_all(cache);
 938         if (err != 0) {
 939                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 940                     dgettext(TEXT_DOMAIN, "blkid_probe_all() %d"), err);
 941                 goto err_blkid2;
 942         }
 943
 944         iter = blkid_dev_iterate_begin(cache);
 945         if (iter == NULL) {
 946                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 947                     dgettext(TEXT_DOMAIN, "blkid_dev_iterate_begin()"));
 948                 goto err_blkid2;
 949         }
 950
 951         err = blkid_dev_set_search(iter, "TYPE", "zfs");
 952         if (err != 0) {
 953                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 954                     dgettext(TEXT_DOMAIN, "blkid_dev_set_search() %d"), err);
 955                 goto err_blkid3;
 956         }
 957
 958         while (blkid_dev_next(iter, &dev) == 0) {
 959                 devname = blkid_dev_devname(dev);
 960                 if ((fd = open64(devname, O_RDONLY)) < 0)
 961                         continue;
 962
 963                 err = zpool_read_label(fd, &config);
 964                 (void) close(fd);
 965
 966                 if (err != 0) {
 967                         (void) no_memory(hdl);
 968                         goto err_blkid3;
 969                 }
 970
 971                 if (config != NULL) {
 972                         err = add_config(hdl, pools, devname, 0, config);
 973                         if (err != 0)
 974                                 goto err_blkid3;
 975                 }
 976         }
 977
 978 err_blkid3:
 979         blkid_dev_iterate_end(iter);
 980 err_blkid2:
 981         blkid_put_cache(cache);
 982 err_blkid1:
 983         return err;
 984 }
 985 #endif /* HAVE_LIBBLKID */
 986
 987 char *
 988 zpool_default_import_path[DEFAULT_IMPORT_PATH_SIZE] = {
 989         "/dev/disk/by-vdev",    /* Custom rules, use first if they exist */
 990         "/dev/disk/zpool",      /* Custom rules, use first if they exist */
 991         "/dev/mapper",          /* Use multipath devices before components */
 992         "/dev/disk/by-uuid",    /* Single unique entry and persistent */
 993         "/dev/disk/by-id",      /* May be multiple entries and persistent */
 994         "/dev/disk/by-path",    /* Encodes physical location and persistent */
 995         "/dev/disk/by-label",   /* Custom persistent labels */
 996         "/dev"                  /* UNSAFE device names will change */
 997 };
 998
 999 /*
1000  * Given a list of directories to search, find all pools stored on disk.  This
1001  * includes partial pools which are not available to import.  If no args are
1002  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1003  * poolname or guid (but not both) are provided by the caller when trying
1004  * to import a specific pool.
1005  */
1006 static nvlist_t *
1007 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1008 {
1009         int i, dirs = iarg->paths;
1010         DIR *dirp = NULL;
1011         struct dirent64 *dp;
1012         char path[MAXPATHLEN];
1013         char *end, **dir = iarg->path;
1014         size_t pathleft;
1015         struct stat64 statbuf;
1016         nvlist_t *ret = NULL, *config;
1017         int fd;
1018         pool_list_t pools = { 0 };
1019         pool_entry_t *pe, *penext;
1020         vdev_entry_t *ve, *venext;
1021         config_entry_t *ce, *cenext;
1022         name_entry_t *ne, *nenext;
1023
1024         verify(iarg->poolname == NULL || iarg->guid == 0);
1025
1026         if (dirs == 0) {
1027 #ifdef HAVE_LIBBLKID
1028                 /* Use libblkid to scan all device for their type */
1029                 if (zpool_find_import_blkid(hdl, &pools) == 0)
1030                         goto skip_scanning;
1031
1032                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1033                     dgettext(TEXT_DOMAIN, "blkid failure falling back "
1034                     "to manual probing"));
1035 #endif /* HAVE_LIBBLKID */
1036
1037                 dir = zpool_default_import_path;
1038                 dirs = DEFAULT_IMPORT_PATH_SIZE;
1039         }
1040
1041         /*
1042          * Go through and read the label configuration information from every
1043          * possible device, organizing the information according to pool GUID
1044          * and toplevel GUID.
1045          */
1046         for (i = 0; i < dirs; i++) {
1047                 char *rdsk;
1048                 int dfd;
1049
1050                 /* use realpath to normalize the path */
1051                 if (realpath(dir[i], path) == 0) {
1052
1053                         /* it is safe to skip missing search paths */
1054                         if (errno == ENOENT)
1055                                 continue;
1056
1057                         zfs_error_aux(hdl, strerror(errno));
1058                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1059                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1060                         goto error;
1061                 }
1062                 end = &path[strlen(path)];
1063                 *end++ = '/';
1064                 *end = 0;
1065                 pathleft = &path[sizeof (path)] - end;
1066
1067                 /*
1068                  * Using raw devices instead of block devices when we're
1069                  * reading the labels skips a bunch of slow operations during
1070                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1071                  */
1072                 if (strcmp(path, "/dev/dsk/") == 0)
1073                         rdsk = "/dev/rdsk/";
1074                 else
1075                         rdsk = path;
1076
1077                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1078                     (dirp = fdopendir(dfd)) == NULL) {
1079                         zfs_error_aux(hdl, strerror(errno));
1080                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1081                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1082                             rdsk);
1083                         goto error;
1084                 }
1085
1086                 /*
1087                  * This is not MT-safe, but we have no MT consumers of libzfs
1088                  */
1089                 while ((dp = readdir64(dirp)) != NULL) {
1090                         const char *name = dp->d_name;
1091                         if (name[0] == '.' &&
1092                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1093                                 continue;
1094
1095                         /*
1096                          * Skip checking devices with well known prefixes:
1097                          * watchdog - A special close is required to avoid
1098                          *            triggering it and resetting the system.
1099                          * fuse     - Fuse control device.
1100                          * ppp      - Generic PPP driver.
1101                          * tty*     - Generic serial interface.
1102                          * vcs*     - Virtual console memory.
1103                          * parport* - Parallel port interface.
1104                          * lp*      - Printer interface.
1105                          * fd*      - Floppy interface.
1106                          * hpet     - High Precision Event Timer, crashes qemu
1107                          *            when accessed from a virtual machine.
1108                          * core     - Symlink to /proc/kcore, causes a crash
1109                          *            when access from Xen dom0.
1110                          */
1111                         if ((strncmp(name, "watchdog", 8) == 0) ||
1112                             (strncmp(name, "fuse", 4) == 0)     ||
1113                             (strncmp(name, "ppp", 3) == 0)      ||
1114                             (strncmp(name, "tty", 3) == 0)      ||
1115                             (strncmp(name, "vcs", 3) == 0)      ||
1116                             (strncmp(name, "parport", 7) == 0)  ||
1117                             (strncmp(name, "lp", 2) == 0)       ||
1118                             (strncmp(name, "fd", 2) == 0)       ||
1119                             (strncmp(name, "hpet", 4) == 0)     ||
1120                             (strncmp(name, "core", 4) == 0))
1121                                 continue;
1122
1123                         /*
1124                          * Ignore failed stats.  We only want regular
1125                          * files and block devices.
1126                          */
1127                         if ((fstatat64(dfd, name, &statbuf, 0) != 0) ||
1128                             (!S_ISREG(statbuf.st_mode) &&
1129                             !S_ISBLK(statbuf.st_mode)))
1130                                 continue;
1131
1132                         if ((fd = openat64(dfd, name, O_RDONLY)) < 0)
1133                                 continue;
1134
1135                         if ((zpool_read_label(fd, &config)) != 0) {
1136                                 (void) close(fd);
1137                                 (void) no_memory(hdl);
1138                                 goto error;
1139                         }
1140
1141                         (void) close(fd);
1142
1143                         if (config != NULL) {
1144                                 boolean_t matched = B_TRUE;
1145                                 char *pname;
1146
1147                                 if ((iarg->poolname != NULL) &&
1148                                     (nvlist_lookup_string(config,
1149                                     ZPOOL_CONFIG_POOL_NAME, &pname) == 0)) {
1150
1151                                         if (strcmp(iarg->poolname, pname))
1152                                                matched = B_FALSE;
1153
1154                                 } else if (iarg->guid != 0) {
1155                                         uint64_t this_guid;
1156
1157                                         matched = nvlist_lookup_uint64(config,
1158                                             ZPOOL_CONFIG_POOL_GUID,
1159                                             &this_guid) == 0 &&
1160                                             iarg->guid == this_guid;
1161                                 }
1162                                 if (!matched) {
1163                                         nvlist_free(config);
1164                                         config = NULL;
1165                                         continue;
1166                                 }
1167                                 /* use the non-raw path for the config */
1168                                 (void) strlcpy(end, name, pathleft);
1169                                 if (add_config(hdl, &pools, path, i+1, config))
1170                                         goto error;
1171                         }
1172                 }
1173
1174                 (void) closedir(dirp);
1175                 dirp = NULL;
1176         }
1177
1178 #ifdef HAVE_LIBBLKID
1179 skip_scanning:
1180 #endif
1181         ret = get_configs(hdl, &pools, iarg->can_be_active);
1182
1183 error:
1184         for (pe = pools.pools; pe != NULL; pe = penext) {
1185                 penext = pe->pe_next;
1186                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1187                         venext = ve->ve_next;
1188                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1189                                 cenext = ce->ce_next;
1190                                 if (ce->ce_config)
1191                                         nvlist_free(ce->ce_config);
1192                                 free(ce);
1193                         }
1194                         free(ve);
1195                 }
1196                 free(pe);
1197         }
1198
1199         for (ne = pools.names; ne != NULL; ne = nenext) {
1200                 nenext = ne->ne_next;
1201                 if (ne->ne_name)
1202                         free(ne->ne_name);
1203                 free(ne);
1204         }
1205
1206         if (dirp)
1207                 (void) closedir(dirp);
1208
1209         return (ret);
1210 }
1211
1212 nvlist_t *
1213 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1214 {
1215         importargs_t iarg = { 0 };
1216
1217         iarg.paths = argc;
1218         iarg.path = argv;
1219
1220         return (zpool_find_import_impl(hdl, &iarg));
1221 }
1222
1223 /*
1224  * Given a cache file, return the contents as a list of importable pools.
1225  * poolname or guid (but not both) are provided by the caller when trying
1226  * to import a specific pool.
1227  */
1228 nvlist_t *
1229 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1230     char *poolname, uint64_t guid)
1231 {
1232         char *buf;
1233         int fd;
1234         struct stat64 statbuf;
1235         nvlist_t *raw, *src, *dst;
1236         nvlist_t *pools;
1237         nvpair_t *elem;
1238         char *name;
1239         uint64_t this_guid;
1240         boolean_t active;
1241
1242         verify(poolname == NULL || guid == 0);
1243
1244         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1245                 zfs_error_aux(hdl, "%s", strerror(errno));
1246                 (void) zfs_error(hdl, EZFS_BADCACHE,
1247                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1248                 return (NULL);
1249         }
1250
1251         if (fstat64(fd, &statbuf) != 0) {
1252                 zfs_error_aux(hdl, "%s", strerror(errno));
1253                 (void) close(fd);
1254                 (void) zfs_error(hdl, EZFS_BADCACHE,
1255                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1256                 return (NULL);
1257         }
1258
1259         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1260                 (void) close(fd);
1261                 return (NULL);
1262         }
1263
1264         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1265                 (void) close(fd);
1266                 free(buf);
1267                 (void) zfs_error(hdl, EZFS_BADCACHE,
1268                     dgettext(TEXT_DOMAIN,
1269                     "failed to read cache file contents"));
1270                 return (NULL);
1271         }
1272
1273         (void) close(fd);
1274
1275         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1276                 free(buf);
1277                 (void) zfs_error(hdl, EZFS_BADCACHE,
1278                     dgettext(TEXT_DOMAIN,
1279                     "invalid or corrupt cache file contents"));
1280                 return (NULL);
1281         }
1282
1283         free(buf);
1284
1285         /*
1286          * Go through and get the current state of the pools and refresh their
1287          * state.
1288          */
1289         if (nvlist_alloc(&pools, 0, 0) != 0) {
1290                 (void) no_memory(hdl);
1291                 nvlist_free(raw);
1292                 return (NULL);
1293         }
1294
1295         elem = NULL;
1296         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1297                 verify(nvpair_value_nvlist(elem, &src) == 0);
1298
1299                 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1300                     &name) == 0);
1301                 if (poolname != NULL && strcmp(poolname, name) != 0)
1302                         continue;
1303
1304                 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1305                     &this_guid) == 0);
1306                 if (guid != 0) {
1307                         verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1308                             &this_guid) == 0);
1309                         if (guid != this_guid)
1310                                 continue;
1311                 }
1312
1313                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1314                         nvlist_free(raw);
1315                         nvlist_free(pools);
1316                         return (NULL);
1317                 }
1318
1319                 if (active)
1320                         continue;
1321
1322                 if ((dst = refresh_config(hdl, src)) == NULL) {
1323                         nvlist_free(raw);
1324                         nvlist_free(pools);
1325                         return (NULL);
1326                 }
1327
1328                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1329                         (void) no_memory(hdl);
1330                         nvlist_free(dst);
1331                         nvlist_free(raw);
1332                         nvlist_free(pools);
1333                         return (NULL);
1334                 }
1335                 nvlist_free(dst);
1336         }
1337
1338         nvlist_free(raw);
1339         return (pools);
1340 }
1341
1342 static int
1343 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1344 {
1345         importargs_t *import = data;
1346         int found = 0;
1347
1348         if (import->poolname != NULL) {
1349                 char *pool_name;
1350
1351                 verify(nvlist_lookup_string(zhp->zpool_config,
1352                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1353                 if (strcmp(pool_name, import->poolname) == 0)
1354                         found = 1;
1355         } else {
1356                 uint64_t pool_guid;
1357
1358                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1359                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1360                 if (pool_guid == import->guid)
1361                         found = 1;
1362         }
1363
1364         zpool_close(zhp);
1365         return (found);
1366 }
1367
1368 nvlist_t *
1369 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1370 {
1371         verify(import->poolname == NULL || import->guid == 0);
1372
1373         if (import->unique)
1374                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1375
1376         if (import->cachefile != NULL)
1377                 return (zpool_find_import_cached(hdl, import->cachefile,
1378                     import->poolname, import->guid));
1379
1380         return (zpool_find_import_impl(hdl, import));
1381 }
1382
1383 boolean_t
1384 find_guid(nvlist_t *nv, uint64_t guid)
1385 {
1386         uint64_t tmp;
1387         nvlist_t **child;
1388         uint_t c, children;
1389
1390         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1391         if (tmp == guid)
1392                 return (B_TRUE);
1393
1394         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1395             &child, &children) == 0) {
1396                 for (c = 0; c < children; c++)
1397                         if (find_guid(child[c], guid))
1398                                 return (B_TRUE);
1399         }
1400
1401         return (B_FALSE);
1402 }
1403
1404 typedef struct aux_cbdata {
1405         const char      *cb_type;
1406         uint64_t        cb_guid;
1407         zpool_handle_t  *cb_zhp;
1408 } aux_cbdata_t;
1409
1410 static int
1411 find_aux(zpool_handle_t *zhp, void *data)
1412 {
1413         aux_cbdata_t *cbp = data;
1414         nvlist_t **list;
1415         uint_t i, count;
1416         uint64_t guid;
1417         nvlist_t *nvroot;
1418
1419         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1420             &nvroot) == 0);
1421
1422         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1423             &list, &count) == 0) {
1424                 for (i = 0; i < count; i++) {
1425                         verify(nvlist_lookup_uint64(list[i],
1426                             ZPOOL_CONFIG_GUID, &guid) == 0);
1427                         if (guid == cbp->cb_guid) {
1428                                 cbp->cb_zhp = zhp;
1429                                 return (1);
1430                         }
1431                 }
1432         }
1433
1434         zpool_close(zhp);
1435         return (0);
1436 }
1437
1438 /*
1439  * Determines if the pool is in use.  If so, it returns true and the state of
1440  * the pool as well as the name of the pool.  Both strings are allocated and
1441  * must be freed by the caller.
1442  */
1443 int
1444 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1445     boolean_t *inuse)
1446 {
1447         nvlist_t *config;
1448         char *name;
1449         boolean_t ret;
1450         uint64_t guid, vdev_guid;
1451         zpool_handle_t *zhp;
1452         nvlist_t *pool_config;
1453         uint64_t stateval, isspare;
1454         aux_cbdata_t cb = { 0 };
1455         boolean_t isactive;
1456
1457         *inuse = B_FALSE;
1458
1459         if (zpool_read_label(fd, &config) != 0) {
1460                 (void) no_memory(hdl);
1461                 return (-1);
1462         }
1463
1464         if (config == NULL)
1465                 return (0);
1466
1467         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1468             &stateval) == 0);
1469         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1470             &vdev_guid) == 0);
1471
1472         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1473                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1474                     &name) == 0);
1475                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1476                     &guid) == 0);
1477         }
1478
1479         switch (stateval) {
1480         case POOL_STATE_EXPORTED:
1481                 /*
1482                  * A pool with an exported state may in fact be imported
1483                  * read-only, so check the in-core state to see if it's
1484                  * active and imported read-only.  If it is, set
1485                  * its state to active.
1486                  */
1487                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1488                     (zhp = zpool_open_canfail(hdl, name)) != NULL &&
1489                     zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1490                         stateval = POOL_STATE_ACTIVE;
1491
1492                 ret = B_TRUE;
1493                 break;
1494
1495         case POOL_STATE_ACTIVE:
1496                 /*
1497                  * For an active pool, we have to determine if it's really part
1498                  * of a currently active pool (in which case the pool will exist
1499                  * and the guid will be the same), or whether it's part of an
1500                  * active pool that was disconnected without being explicitly
1501                  * exported.
1502                  */
1503                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1504                         nvlist_free(config);
1505                         return (-1);
1506                 }
1507
1508                 if (isactive) {
1509                         /*
1510                          * Because the device may have been removed while
1511                          * offlined, we only report it as active if the vdev is
1512                          * still present in the config.  Otherwise, pretend like
1513                          * it's not in use.
1514                          */
1515                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1516                             (pool_config = zpool_get_config(zhp, NULL))
1517                             != NULL) {
1518                                 nvlist_t *nvroot;
1519
1520                                 verify(nvlist_lookup_nvlist(pool_config,
1521                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1522                                 ret = find_guid(nvroot, vdev_guid);
1523                         } else {
1524                                 ret = B_FALSE;
1525                         }
1526
1527                         /*
1528                          * If this is an active spare within another pool, we
1529                          * treat it like an unused hot spare.  This allows the
1530                          * user to create a pool with a hot spare that currently
1531                          * in use within another pool.  Since we return B_TRUE,
1532                          * libdiskmgt will continue to prevent generic consumers
1533                          * from using the device.
1534                          */
1535                         if (ret && nvlist_lookup_uint64(config,
1536                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1537                                 stateval = POOL_STATE_SPARE;
1538
1539                         if (zhp != NULL)
1540                                 zpool_close(zhp);
1541                 } else {
1542                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1543                         ret = B_TRUE;
1544                 }
1545                 break;
1546
1547         case POOL_STATE_SPARE:
1548                 /*
1549                  * For a hot spare, it can be either definitively in use, or
1550                  * potentially active.  To determine if it's in use, we iterate
1551                  * over all pools in the system and search for one with a spare
1552                  * with a matching guid.
1553                  *
1554                  * Due to the shared nature of spares, we don't actually report
1555                  * the potentially active case as in use.  This means the user
1556                  * can freely create pools on the hot spares of exported pools,
1557                  * but to do otherwise makes the resulting code complicated, and
1558                  * we end up having to deal with this case anyway.
1559                  */
1560                 cb.cb_zhp = NULL;
1561                 cb.cb_guid = vdev_guid;
1562                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1563                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1564                         name = (char *)zpool_get_name(cb.cb_zhp);
1565                         ret = TRUE;
1566                 } else {
1567                         ret = FALSE;
1568                 }
1569                 break;
1570
1571         case POOL_STATE_L2CACHE:
1572
1573                 /*
1574                  * Check if any pool is currently using this l2cache device.
1575                  */
1576                 cb.cb_zhp = NULL;
1577                 cb.cb_guid = vdev_guid;
1578                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1579                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1580                         name = (char *)zpool_get_name(cb.cb_zhp);
1581                         ret = TRUE;
1582                 } else {
1583                         ret = FALSE;
1584                 }
1585                 break;
1586
1587         default:
1588                 ret = B_FALSE;
1589         }
1590
1591
1592         if (ret) {
1593                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1594                         if (cb.cb_zhp)
1595                                 zpool_close(cb.cb_zhp);
1596                         nvlist_free(config);
1597                         return (-1);
1598                 }
1599                 *state = (pool_state_t)stateval;
1600         }
1601
1602         if (cb.cb_zhp)
1603                 zpool_close(cb.cb_zhp);
1604
1605         nvlist_free(config);
1606         *inuse = ret;
1607         return (0);
1608 }