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) 2012 by Delphix. All rights reserved.
  25  */
  26
  27 /*
  28  * Pool import support functions.
  29  *
  30  * To import a pool, we rely on reading the configuration information from the
  31  * ZFS label of each device.  If we successfully read the label, then we
  32  * organize the configuration information in the following hierarchy:
  33  *
  34  *      pool guid -> toplevel vdev guid -> label txg
  35  *
  36  * Duplicate entries matching this same tuple will be discarded.  Once we have
  37  * examined every device, we pick the best label txg config for each toplevel
  38  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  39  * update any paths that have changed.  Finally, we attempt to import the pool
  40  * using our derived config, and record the results.
  41  */
  42
  43 #include <ctype.h>
  44 #include <devid.h>
  45 #include <dirent.h>
  46 #include <errno.h>
  47 #include <libintl.h>
  48 #include <stddef.h>
  49 #include <stdlib.h>
  50 #include <string.h>
  51 #include <sys/stat.h>
  52 #include <unistd.h>
  53 #include <fcntl.h>
  54 #include <sys/vtoc.h>
  55 #include <sys/dktp/fdisk.h>
  56 #include <sys/efi_partition.h>
  57
  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 = NULL;
 438         uint64_t 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                                  *      pool txg (if available)
 527                                  *      comment (if available)
 528                                  *      pool state
 529                                  *      hostid (if available)
 530                                  *      hostname (if available)
 531                                  */
 532                                 uint64_t state, version, pool_txg;
 533                                 char *comment = NULL;
 534
 535                                 version = fnvlist_lookup_uint64(tmp,
 536                                     ZPOOL_CONFIG_VERSION);
 537                                 fnvlist_add_uint64(config,
 538                                     ZPOOL_CONFIG_VERSION, version);
 539                                 guid = fnvlist_lookup_uint64(tmp,
 540                                     ZPOOL_CONFIG_POOL_GUID);
 541                                 fnvlist_add_uint64(config,
 542                                     ZPOOL_CONFIG_POOL_GUID, guid);
 543                                 name = fnvlist_lookup_string(tmp,
 544                                     ZPOOL_CONFIG_POOL_NAME);
 545                                 fnvlist_add_string(config,
 546                                     ZPOOL_CONFIG_POOL_NAME, name);
 547
 548                                 if (nvlist_lookup_uint64(tmp,
 549                                     ZPOOL_CONFIG_POOL_TXG, &pool_txg) == 0)
 550                                         fnvlist_add_uint64(config,
 551                                             ZPOOL_CONFIG_POOL_TXG, pool_txg);
 552
 553                                 if (nvlist_lookup_string(tmp,
 554                                     ZPOOL_CONFIG_COMMENT, &comment) == 0)
 555                                         fnvlist_add_string(config,
 556                                             ZPOOL_CONFIG_COMMENT, comment);
 557
 558                                 state = fnvlist_lookup_uint64(tmp,
 559                                     ZPOOL_CONFIG_POOL_STATE);
 560                                 fnvlist_add_uint64(config,
 561                                     ZPOOL_CONFIG_POOL_STATE, state);
 562
 563                                 hostid = 0;
 564                                 if (nvlist_lookup_uint64(tmp,
 565                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 566                                         fnvlist_add_uint64(config,
 567                                             ZPOOL_CONFIG_HOSTID, hostid);
 568                                         hostname = fnvlist_lookup_string(tmp,
 569                                             ZPOOL_CONFIG_HOSTNAME);
 570                                         fnvlist_add_string(config,
 571                                             ZPOOL_CONFIG_HOSTNAME, hostname);
 572                                 }
 573
 574                                 config_seen = B_TRUE;
 575                         }
 576
 577                         /*
 578                          * Add this top-level vdev to the child array.
 579                          */
 580                         verify(nvlist_lookup_nvlist(tmp,
 581                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 582                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 583                             &id) == 0);
 584
 585                         if (id >= children) {
 586                                 nvlist_t **newchild;
 587
 588                                 newchild = zfs_alloc(hdl, (id + 1) *
 589                                     sizeof (nvlist_t *));
 590                                 if (newchild == NULL)
 591                                         goto nomem;
 592
 593                                 for (c = 0; c < children; c++)
 594                                         newchild[c] = child[c];
 595
 596                                 free(child);
 597                                 child = newchild;
 598                                 children = id + 1;
 599                         }
 600                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 601                                 goto nomem;
 602
 603                 }
 604
 605                 /*
 606                  * If we have information about all the top-levels then
 607                  * clean up the nvlist which we've constructed. This
 608                  * means removing any extraneous devices that are
 609                  * beyond the valid range or adding devices to the end
 610                  * of our array which appear to be missing.
 611                  */
 612                 if (valid_top_config) {
 613                         if (max_id < children) {
 614                                 for (c = max_id; c < children; c++)
 615                                         nvlist_free(child[c]);
 616                                 children = max_id;
 617                         } else if (max_id > children) {
 618                                 nvlist_t **newchild;
 619
 620                                 newchild = zfs_alloc(hdl, (max_id) *
 621                                     sizeof (nvlist_t *));
 622                                 if (newchild == NULL)
 623                                         goto nomem;
 624
 625                                 for (c = 0; c < children; c++)
 626                                         newchild[c] = child[c];
 627
 628                                 free(child);
 629                                 child = newchild;
 630                                 children = max_id;
 631                         }
 632                 }
 633
 634                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 635                     &guid) == 0);
 636
 637                 /*
 638                  * The vdev namespace may contain holes as a result of
 639                  * device removal. We must add them back into the vdev
 640                  * tree before we process any missing devices.
 641                  */
 642                 if (holes > 0) {
 643                         ASSERT(valid_top_config);
 644
 645                         for (c = 0; c < children; c++) {
 646                                 nvlist_t *holey;
 647
 648                                 if (child[c] != NULL ||
 649                                     !vdev_is_hole(hole_array, holes, c))
 650                                         continue;
 651
 652                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 653                                     0) != 0)
 654                                         goto nomem;
 655
 656                                 /*
 657                                  * Holes in the namespace are treated as
 658                                  * "hole" top-level vdevs and have a
 659                                  * special flag set on them.
 660                                  */
 661                                 if (nvlist_add_string(holey,
 662                                     ZPOOL_CONFIG_TYPE,
 663                                     VDEV_TYPE_HOLE) != 0 ||
 664                                     nvlist_add_uint64(holey,
 665                                     ZPOOL_CONFIG_ID, c) != 0 ||
 666                                     nvlist_add_uint64(holey,
 667                                     ZPOOL_CONFIG_GUID, 0ULL) != 0)
 668                                         goto nomem;
 669                                 child[c] = holey;
 670                         }
 671                 }
 672
 673                 /*
 674                  * Look for any missing top-level vdevs.  If this is the case,
 675                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 676                  * simply compress the child array, because the kernel performs
 677                  * certain checks to make sure the vdev IDs match their location
 678                  * in the configuration.
 679                  */
 680                 for (c = 0; c < children; c++) {
 681                         if (child[c] == NULL) {
 682                                 nvlist_t *missing;
 683                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 684                                     0) != 0)
 685                                         goto nomem;
 686                                 if (nvlist_add_string(missing,
 687                                     ZPOOL_CONFIG_TYPE,
 688                                     VDEV_TYPE_MISSING) != 0 ||
 689                                     nvlist_add_uint64(missing,
 690                                     ZPOOL_CONFIG_ID, c) != 0 ||
 691                                     nvlist_add_uint64(missing,
 692                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 693                                         nvlist_free(missing);
 694                                         goto nomem;
 695                                 }
 696                                 child[c] = missing;
 697                         }
 698                 }
 699
 700                 /*
 701                  * Put all of this pool's top-level vdevs into a root vdev.
 702                  */
 703                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 704                         goto nomem;
 705                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 706                     VDEV_TYPE_ROOT) != 0 ||
 707                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 708                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 709                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 710                     child, children) != 0) {
 711                         nvlist_free(nvroot);
 712                         goto nomem;
 713                 }
 714
 715                 for (c = 0; c < children; c++)
 716                         nvlist_free(child[c]);
 717                 free(child);
 718                 children = 0;
 719                 child = NULL;
 720
 721                 /*
 722                  * Go through and fix up any paths and/or devids based on our
 723                  * known list of vdev GUID -> path mappings.
 724                  */
 725                 if (fix_paths(nvroot, pl->names) != 0) {
 726                         nvlist_free(nvroot);
 727                         goto nomem;
 728                 }
 729
 730                 /*
 731                  * Add the root vdev to this pool's configuration.
 732                  */
 733                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 734                     nvroot) != 0) {
 735                         nvlist_free(nvroot);
 736                         goto nomem;
 737                 }
 738                 nvlist_free(nvroot);
 739
 740                 /*
 741                  * zdb uses this path to report on active pools that were
 742                  * imported or created using -R.
 743                  */
 744                 if (active_ok)
 745                         goto add_pool;
 746
 747                 /*
 748                  * Determine if this pool is currently active, in which case we
 749                  * can't actually import it.
 750                  */
 751                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 752                     &name) == 0);
 753                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 754                     &guid) == 0);
 755
 756                 if (pool_active(hdl, name, guid, &isactive) != 0)
 757                         goto error;
 758
 759                 if (isactive) {
 760                         nvlist_free(config);
 761                         config = NULL;
 762                         continue;
 763                 }
 764
 765                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 766                         nvlist_free(config);
 767                         config = NULL;
 768                         continue;
 769                 }
 770
 771                 nvlist_free(config);
 772                 config = nvl;
 773
 774                 /*
 775                  * Go through and update the paths for spares, now that we have
 776                  * them.
 777                  */
 778                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 779                     &nvroot) == 0);
 780                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 781                     &spares, &nspares) == 0) {
 782                         for (i = 0; i < nspares; i++) {
 783                                 if (fix_paths(spares[i], pl->names) != 0)
 784                                         goto nomem;
 785                         }
 786                 }
 787
 788                 /*
 789                  * Update the paths for l2cache devices.
 790                  */
 791                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 792                     &l2cache, &nl2cache) == 0) {
 793                         for (i = 0; i < nl2cache; i++) {
 794                                 if (fix_paths(l2cache[i], pl->names) != 0)
 795                                         goto nomem;
 796                         }
 797                 }
 798
 799                 /*
 800                  * Restore the original information read from the actual label.
 801                  */
 802                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 803                     DATA_TYPE_UINT64);
 804                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 805                     DATA_TYPE_STRING);
 806                 if (hostid != 0) {
 807                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 808                             hostid) == 0);
 809                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 810                             hostname) == 0);
 811                 }
 812
 813 add_pool:
 814                 /*
 815                  * Add this pool to the list of configs.
 816                  */
 817                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 818                     &name) == 0);
 819                 if (nvlist_add_nvlist(ret, name, config) != 0)
 820                         goto nomem;
 821
 822                 found_one = B_TRUE;
 823                 nvlist_free(config);
 824                 config = NULL;
 825         }
 826
 827         if (!found_one) {
 828                 nvlist_free(ret);
 829                 ret = NULL;
 830         }
 831
 832         return (ret);
 833
 834 nomem:
 835         (void) no_memory(hdl);
 836 error:
 837         nvlist_free(config);
 838         nvlist_free(ret);
 839         for (c = 0; c < children; c++)
 840                 nvlist_free(child[c]);
 841         free(child);
 842
 843         return (NULL);
 844 }
 845
 846 /*
 847  * Return the offset of the given label.
 848  */
 849 static uint64_t
 850 label_offset(uint64_t size, int l)
 851 {
 852         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 853         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 854             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 855 }
 856
 857 /*
 858  * Given a file descriptor, read the label information and return an nvlist
 859  * describing the configuration, if there is one.
 860  */
 861 int
 862 zpool_read_label(int fd, nvlist_t **config)
 863 {
 864         struct stat64 statbuf;
 865         int l;
 866         vdev_label_t *label;
 867         uint64_t state, txg, size;
 868
 869         *config = NULL;
 870
 871         if (fstat64(fd, &statbuf) == -1)
 872                 return (0);
 873         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 874
 875         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 876                 return (-1);
 877
 878         for (l = 0; l < VDEV_LABELS; l++) {
 879                 if (pread64(fd, label, sizeof (vdev_label_t),
 880                     label_offset(size, l)) != sizeof (vdev_label_t))
 881                         continue;
 882
 883                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 884                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 885                         continue;
 886
 887                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 888                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 889                         nvlist_free(*config);
 890                         continue;
 891                 }
 892
 893                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 894                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 895                     &txg) != 0 || txg == 0)) {
 896                         nvlist_free(*config);
 897                         continue;
 898                 }
 899
 900                 free(label);
 901                 return (0);
 902         }
 903
 904         free(label);
 905         *config = NULL;
 906         return (0);
 907 }
 908
 909 #ifdef HAVE_LIBBLKID
 910 /*
 911  * Use libblkid to quickly search for zfs devices
 912  */
 913 static int
 914 zpool_find_import_blkid(libzfs_handle_t *hdl, pool_list_t *pools)
 915 {
 916         blkid_cache cache;
 917         blkid_dev_iterate iter;
 918         blkid_dev dev;
 919         const char *devname;
 920         nvlist_t *config;
 921         int fd, err;
 922
 923         err = blkid_get_cache(&cache, NULL);
 924         if (err != 0) {
 925                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 926                     dgettext(TEXT_DOMAIN, "blkid_get_cache() %d"), err);
 927                 goto err_blkid1;
 928         }
 929
 930         err = blkid_probe_all(cache);
 931         if (err != 0) {
 932                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 933                     dgettext(TEXT_DOMAIN, "blkid_probe_all() %d"), err);
 934                 goto err_blkid2;
 935         }
 936
 937         iter = blkid_dev_iterate_begin(cache);
 938         if (iter == NULL) {
 939                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 940                     dgettext(TEXT_DOMAIN, "blkid_dev_iterate_begin()"));
 941                 goto err_blkid2;
 942         }
 943
 944         err = blkid_dev_set_search(iter, "TYPE", "zfs");
 945         if (err != 0) {
 946                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 947                     dgettext(TEXT_DOMAIN, "blkid_dev_set_search() %d"), err);
 948                 goto err_blkid3;
 949         }
 950
 951         while (blkid_dev_next(iter, &dev) == 0) {
 952                 devname = blkid_dev_devname(dev);
 953                 if ((fd = open64(devname, O_RDONLY)) < 0)
 954                         continue;
 955
 956                 err = zpool_read_label(fd, &config);
 957                 (void) close(fd);
 958
 959                 if (err != 0) {
 960                         (void) no_memory(hdl);
 961                         goto err_blkid3;
 962                 }
 963
 964                 if (config != NULL) {
 965                         err = add_config(hdl, pools, devname, 0, config);
 966                         if (err != 0)
 967                                 goto err_blkid3;
 968                 }
 969         }
 970
 971 err_blkid3:
 972         blkid_dev_iterate_end(iter);
 973 err_blkid2:
 974         blkid_put_cache(cache);
 975 err_blkid1:
 976         return err;
 977 }
 978 #endif /* HAVE_LIBBLKID */
 979
 980 char *
 981 zpool_default_import_path[DEFAULT_IMPORT_PATH_SIZE] = {
 982         "/dev/disk/by-vdev",    /* Custom rules, use first if they exist */
 983         "/dev/disk/zpool",      /* Custom rules, use first if they exist */
 984         "/dev/mapper",          /* Use multipath devices before components */
 985         "/dev/disk/by-uuid",    /* Single unique entry and persistent */
 986         "/dev/disk/by-id",      /* May be multiple entries and persistent */
 987         "/dev/disk/by-path",    /* Encodes physical location and persistent */
 988         "/dev/disk/by-label",   /* Custom persistent labels */
 989         "/dev"                  /* UNSAFE device names will change */
 990 };
 991
 992 /*
 993  * Given a list of directories to search, find all pools stored on disk.  This
 994  * includes partial pools which are not available to import.  If no args are
 995  * given (argc is 0), then the default directory (/dev/dsk) is searched.
 996  * poolname or guid (but not both) are provided by the caller when trying
 997  * to import a specific pool.
 998  */
 999 static nvlist_t *
1000 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1001 {
1002         int i, dirs = iarg->paths;
1003         DIR *dirp = NULL;
1004         struct dirent64 *dp;
1005         char path[MAXPATHLEN];
1006         char *end, **dir = iarg->path;
1007         size_t pathleft;
1008         struct stat64 statbuf;
1009         nvlist_t *ret = NULL, *config;
1010         int fd;
1011         pool_list_t pools = { 0 };
1012         pool_entry_t *pe, *penext;
1013         vdev_entry_t *ve, *venext;
1014         config_entry_t *ce, *cenext;
1015         name_entry_t *ne, *nenext;
1016
1017         verify(iarg->poolname == NULL || iarg->guid == 0);
1018
1019         if (dirs == 0) {
1020 #ifdef HAVE_LIBBLKID
1021                 /* Use libblkid to scan all device for their type */
1022                 if (zpool_find_import_blkid(hdl, &pools) == 0)
1023                         goto skip_scanning;
1024
1025                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1026                     dgettext(TEXT_DOMAIN, "blkid failure falling back "
1027                     "to manual probing"));
1028 #endif /* HAVE_LIBBLKID */
1029
1030                 dir = zpool_default_import_path;
1031                 dirs = DEFAULT_IMPORT_PATH_SIZE;
1032         }
1033
1034         /*
1035          * Go through and read the label configuration information from every
1036          * possible device, organizing the information according to pool GUID
1037          * and toplevel GUID.
1038          */
1039         for (i = 0; i < dirs; i++) {
1040                 char *rdsk;
1041                 int dfd;
1042
1043                 /* use realpath to normalize the path */
1044                 if (realpath(dir[i], path) == 0) {
1045
1046                         /* it is safe to skip missing search paths */
1047                         if (errno == ENOENT)
1048                                 continue;
1049
1050                         zfs_error_aux(hdl, strerror(errno));
1051                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1052                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1053                         goto error;
1054                 }
1055                 end = &path[strlen(path)];
1056                 *end++ = '/';
1057                 *end = 0;
1058                 pathleft = &path[sizeof (path)] - end;
1059
1060                 /*
1061                  * Using raw devices instead of block devices when we're
1062                  * reading the labels skips a bunch of slow operations during
1063                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1064                  */
1065                 if (strcmp(path, "/dev/dsk/") == 0)
1066                         rdsk = "/dev/rdsk/";
1067                 else
1068                         rdsk = path;
1069
1070                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1071                     (dirp = fdopendir(dfd)) == NULL) {
1072                         zfs_error_aux(hdl, strerror(errno));
1073                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1074                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1075                             rdsk);
1076                         goto error;
1077                 }
1078
1079                 /*
1080                  * This is not MT-safe, but we have no MT consumers of libzfs
1081                  */
1082                 while ((dp = readdir64(dirp)) != NULL) {
1083                         const char *name = dp->d_name;
1084                         if (name[0] == '.' &&
1085                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1086                                 continue;
1087
1088                         /*
1089                          * Skip checking devices with well known prefixes:
1090                          * watchdog - A special close is required to avoid
1091                          *            triggering it and resetting the system.
1092                          * fuse     - Fuse control device.
1093                          * ppp      - Generic PPP driver.
1094                          * tty*     - Generic serial interface.
1095                          * vcs*     - Virtual console memory.
1096                          * parport* - Parallel port interface.
1097                          * lp*      - Printer interface.
1098                          * fd*      - Floppy interface.
1099                          * hpet     - High Precision Event Timer, crashes qemu
1100                          *            when accessed from a virtual machine.
1101                          * core     - Symlink to /proc/kcore, causes a crash
1102                          *            when access from Xen dom0.
1103                          */
1104                         if ((strncmp(name, "watchdog", 8) == 0) ||
1105                             (strncmp(name, "fuse", 4) == 0)     ||
1106                             (strncmp(name, "ppp", 3) == 0)      ||
1107                             (strncmp(name, "tty", 3) == 0)      ||
1108                             (strncmp(name, "vcs", 3) == 0)      ||
1109                             (strncmp(name, "parport", 7) == 0)  ||
1110                             (strncmp(name, "lp", 2) == 0)       ||
1111                             (strncmp(name, "fd", 2) == 0)       ||
1112                             (strncmp(name, "hpet", 4) == 0)     ||
1113                             (strncmp(name, "core", 4) == 0))
1114                                 continue;
1115
1116                         /*
1117                          * Ignore failed stats.  We only want regular
1118                          * files and block devices.
1119                          */
1120                         if ((fstatat64(dfd, name, &statbuf, 0) != 0) ||
1121                             (!S_ISREG(statbuf.st_mode) &&
1122                             !S_ISBLK(statbuf.st_mode)))
1123                                 continue;
1124
1125                         if ((fd = openat64(dfd, name, O_RDONLY)) < 0)
1126                                 continue;
1127
1128                         if ((zpool_read_label(fd, &config)) != 0) {
1129                                 (void) close(fd);
1130                                 (void) no_memory(hdl);
1131                                 goto error;
1132                         }
1133
1134                         (void) close(fd);
1135
1136                         if (config != NULL) {
1137                                 boolean_t matched = B_TRUE;
1138                                 char *pname;
1139
1140                                 if ((iarg->poolname != NULL) &&
1141                                     (nvlist_lookup_string(config,
1142                                     ZPOOL_CONFIG_POOL_NAME, &pname) == 0)) {
1143
1144                                         if (strcmp(iarg->poolname, pname))
1145                                                matched = B_FALSE;
1146
1147                                 } else if (iarg->guid != 0) {
1148                                         uint64_t this_guid;
1149
1150                                         matched = nvlist_lookup_uint64(config,
1151                                             ZPOOL_CONFIG_POOL_GUID,
1152                                             &this_guid) == 0 &&
1153                                             iarg->guid == this_guid;
1154                                 }
1155                                 if (!matched) {
1156                                         nvlist_free(config);
1157                                         config = NULL;
1158                                         continue;
1159                                 }
1160                                 /* use the non-raw path for the config */
1161                                 (void) strlcpy(end, name, pathleft);
1162                                 if (add_config(hdl, &pools, path, i+1, config))
1163                                         goto error;
1164                         }
1165                 }
1166
1167                 (void) closedir(dirp);
1168                 dirp = NULL;
1169         }
1170
1171 #ifdef HAVE_LIBBLKID
1172 skip_scanning:
1173 #endif
1174         ret = get_configs(hdl, &pools, iarg->can_be_active);
1175
1176 error:
1177         for (pe = pools.pools; pe != NULL; pe = penext) {
1178                 penext = pe->pe_next;
1179                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1180                         venext = ve->ve_next;
1181                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1182                                 cenext = ce->ce_next;
1183                                 if (ce->ce_config)
1184                                         nvlist_free(ce->ce_config);
1185                                 free(ce);
1186                         }
1187                         free(ve);
1188                 }
1189                 free(pe);
1190         }
1191
1192         for (ne = pools.names; ne != NULL; ne = nenext) {
1193                 nenext = ne->ne_next;
1194                 if (ne->ne_name)
1195                         free(ne->ne_name);
1196                 free(ne);
1197         }
1198
1199         if (dirp)
1200                 (void) closedir(dirp);
1201
1202         return (ret);
1203 }
1204
1205 nvlist_t *
1206 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1207 {
1208         importargs_t iarg = { 0 };
1209
1210         iarg.paths = argc;
1211         iarg.path = argv;
1212
1213         return (zpool_find_import_impl(hdl, &iarg));
1214 }
1215
1216 /*
1217  * Given a cache file, return the contents as a list of importable pools.
1218  * poolname or guid (but not both) are provided by the caller when trying
1219  * to import a specific pool.
1220  */
1221 nvlist_t *
1222 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1223     char *poolname, uint64_t guid)
1224 {
1225         char *buf;
1226         int fd;
1227         struct stat64 statbuf;
1228         nvlist_t *raw, *src, *dst;
1229         nvlist_t *pools;
1230         nvpair_t *elem;
1231         char *name;
1232         uint64_t this_guid;
1233         boolean_t active;
1234
1235         verify(poolname == NULL || guid == 0);
1236
1237         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1238                 zfs_error_aux(hdl, "%s", strerror(errno));
1239                 (void) zfs_error(hdl, EZFS_BADCACHE,
1240                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1241                 return (NULL);
1242         }
1243
1244         if (fstat64(fd, &statbuf) != 0) {
1245                 zfs_error_aux(hdl, "%s", strerror(errno));
1246                 (void) close(fd);
1247                 (void) zfs_error(hdl, EZFS_BADCACHE,
1248                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1249                 return (NULL);
1250         }
1251
1252         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1253                 (void) close(fd);
1254                 return (NULL);
1255         }
1256
1257         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1258                 (void) close(fd);
1259                 free(buf);
1260                 (void) zfs_error(hdl, EZFS_BADCACHE,
1261                     dgettext(TEXT_DOMAIN,
1262                     "failed to read cache file contents"));
1263                 return (NULL);
1264         }
1265
1266         (void) close(fd);
1267
1268         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1269                 free(buf);
1270                 (void) zfs_error(hdl, EZFS_BADCACHE,
1271                     dgettext(TEXT_DOMAIN,
1272                     "invalid or corrupt cache file contents"));
1273                 return (NULL);
1274         }
1275
1276         free(buf);
1277
1278         /*
1279          * Go through and get the current state of the pools and refresh their
1280          * state.
1281          */
1282         if (nvlist_alloc(&pools, 0, 0) != 0) {
1283                 (void) no_memory(hdl);
1284                 nvlist_free(raw);
1285                 return (NULL);
1286         }
1287
1288         elem = NULL;
1289         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1290                 verify(nvpair_value_nvlist(elem, &src) == 0);
1291
1292                 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1293                     &name) == 0);
1294                 if (poolname != NULL && strcmp(poolname, name) != 0)
1295                         continue;
1296
1297                 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1298                     &this_guid) == 0);
1299                 if (guid != 0) {
1300                         verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1301                             &this_guid) == 0);
1302                         if (guid != this_guid)
1303                                 continue;
1304                 }
1305
1306                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1307                         nvlist_free(raw);
1308                         nvlist_free(pools);
1309                         return (NULL);
1310                 }
1311
1312                 if (active)
1313                         continue;
1314
1315                 if ((dst = refresh_config(hdl, src)) == NULL) {
1316                         nvlist_free(raw);
1317                         nvlist_free(pools);
1318                         return (NULL);
1319                 }
1320
1321                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1322                         (void) no_memory(hdl);
1323                         nvlist_free(dst);
1324                         nvlist_free(raw);
1325                         nvlist_free(pools);
1326                         return (NULL);
1327                 }
1328                 nvlist_free(dst);
1329         }
1330
1331         nvlist_free(raw);
1332         return (pools);
1333 }
1334
1335 static int
1336 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1337 {
1338         importargs_t *import = data;
1339         int found = 0;
1340
1341         if (import->poolname != NULL) {
1342                 char *pool_name;
1343
1344                 verify(nvlist_lookup_string(zhp->zpool_config,
1345                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1346                 if (strcmp(pool_name, import->poolname) == 0)
1347                         found = 1;
1348         } else {
1349                 uint64_t pool_guid;
1350
1351                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1352                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1353                 if (pool_guid == import->guid)
1354                         found = 1;
1355         }
1356
1357         zpool_close(zhp);
1358         return (found);
1359 }
1360
1361 nvlist_t *
1362 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1363 {
1364         verify(import->poolname == NULL || import->guid == 0);
1365
1366         if (import->unique)
1367                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1368
1369         if (import->cachefile != NULL)
1370                 return (zpool_find_import_cached(hdl, import->cachefile,
1371                     import->poolname, import->guid));
1372
1373         return (zpool_find_import_impl(hdl, import));
1374 }
1375
1376 boolean_t
1377 find_guid(nvlist_t *nv, uint64_t guid)
1378 {
1379         uint64_t tmp;
1380         nvlist_t **child;
1381         uint_t c, children;
1382
1383         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1384         if (tmp == guid)
1385                 return (B_TRUE);
1386
1387         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1388             &child, &children) == 0) {
1389                 for (c = 0; c < children; c++)
1390                         if (find_guid(child[c], guid))
1391                                 return (B_TRUE);
1392         }
1393
1394         return (B_FALSE);
1395 }
1396
1397 typedef struct aux_cbdata {
1398         const char      *cb_type;
1399         uint64_t        cb_guid;
1400         zpool_handle_t  *cb_zhp;
1401 } aux_cbdata_t;
1402
1403 static int
1404 find_aux(zpool_handle_t *zhp, void *data)
1405 {
1406         aux_cbdata_t *cbp = data;
1407         nvlist_t **list;
1408         uint_t i, count;
1409         uint64_t guid;
1410         nvlist_t *nvroot;
1411
1412         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1413             &nvroot) == 0);
1414
1415         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1416             &list, &count) == 0) {
1417                 for (i = 0; i < count; i++) {
1418                         verify(nvlist_lookup_uint64(list[i],
1419                             ZPOOL_CONFIG_GUID, &guid) == 0);
1420                         if (guid == cbp->cb_guid) {
1421                                 cbp->cb_zhp = zhp;
1422                                 return (1);
1423                         }
1424                 }
1425         }
1426
1427         zpool_close(zhp);
1428         return (0);
1429 }
1430
1431 /*
1432  * Determines if the pool is in use.  If so, it returns true and the state of
1433  * the pool as well as the name of the pool.  Both strings are allocated and
1434  * must be freed by the caller.
1435  */
1436 int
1437 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1438     boolean_t *inuse)
1439 {
1440         nvlist_t *config;
1441         char *name;
1442         boolean_t ret;
1443         uint64_t guid, vdev_guid;
1444         zpool_handle_t *zhp;
1445         nvlist_t *pool_config;
1446         uint64_t stateval, isspare;
1447         aux_cbdata_t cb = { 0 };
1448         boolean_t isactive;
1449
1450         *inuse = B_FALSE;
1451
1452         if (zpool_read_label(fd, &config) != 0) {
1453                 (void) no_memory(hdl);
1454                 return (-1);
1455         }
1456
1457         if (config == NULL)
1458                 return (0);
1459
1460         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1461             &stateval) == 0);
1462         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1463             &vdev_guid) == 0);
1464
1465         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1466                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1467                     &name) == 0);
1468                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1469                     &guid) == 0);
1470         }
1471
1472         switch (stateval) {
1473         case POOL_STATE_EXPORTED:
1474                 /*
1475                  * A pool with an exported state may in fact be imported
1476                  * read-only, so check the in-core state to see if it's
1477                  * active and imported read-only.  If it is, set
1478                  * its state to active.
1479                  */
1480                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1481                     (zhp = zpool_open_canfail(hdl, name)) != NULL &&
1482                     zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1483                         stateval = POOL_STATE_ACTIVE;
1484
1485                 ret = B_TRUE;
1486                 break;
1487
1488         case POOL_STATE_ACTIVE:
1489                 /*
1490                  * For an active pool, we have to determine if it's really part
1491                  * of a currently active pool (in which case the pool will exist
1492                  * and the guid will be the same), or whether it's part of an
1493                  * active pool that was disconnected without being explicitly
1494                  * exported.
1495                  */
1496                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1497                         nvlist_free(config);
1498                         return (-1);
1499                 }
1500
1501                 if (isactive) {
1502                         /*
1503                          * Because the device may have been removed while
1504                          * offlined, we only report it as active if the vdev is
1505                          * still present in the config.  Otherwise, pretend like
1506                          * it's not in use.
1507                          */
1508                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1509                             (pool_config = zpool_get_config(zhp, NULL))
1510                             != NULL) {
1511                                 nvlist_t *nvroot;
1512
1513                                 verify(nvlist_lookup_nvlist(pool_config,
1514                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1515                                 ret = find_guid(nvroot, vdev_guid);
1516                         } else {
1517                                 ret = B_FALSE;
1518                         }
1519
1520                         /*
1521                          * If this is an active spare within another pool, we
1522                          * treat it like an unused hot spare.  This allows the
1523                          * user to create a pool with a hot spare that currently
1524                          * in use within another pool.  Since we return B_TRUE,
1525                          * libdiskmgt will continue to prevent generic consumers
1526                          * from using the device.
1527                          */
1528                         if (ret && nvlist_lookup_uint64(config,
1529                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1530                                 stateval = POOL_STATE_SPARE;
1531
1532                         if (zhp != NULL)
1533                                 zpool_close(zhp);
1534                 } else {
1535                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1536                         ret = B_TRUE;
1537                 }
1538                 break;
1539
1540         case POOL_STATE_SPARE:
1541                 /*
1542                  * For a hot spare, it can be either definitively in use, or
1543                  * potentially active.  To determine if it's in use, we iterate
1544                  * over all pools in the system and search for one with a spare
1545                  * with a matching guid.
1546                  *
1547                  * Due to the shared nature of spares, we don't actually report
1548                  * the potentially active case as in use.  This means the user
1549                  * can freely create pools on the hot spares of exported pools,
1550                  * but to do otherwise makes the resulting code complicated, and
1551                  * we end up having to deal with this case anyway.
1552                  */
1553                 cb.cb_zhp = NULL;
1554                 cb.cb_guid = vdev_guid;
1555                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1556                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1557                         name = (char *)zpool_get_name(cb.cb_zhp);
1558                         ret = TRUE;
1559                 } else {
1560                         ret = FALSE;
1561                 }
1562                 break;
1563
1564         case POOL_STATE_L2CACHE:
1565
1566                 /*
1567                  * Check if any pool is currently using this l2cache device.
1568                  */
1569                 cb.cb_zhp = NULL;
1570                 cb.cb_guid = vdev_guid;
1571                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1572                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1573                         name = (char *)zpool_get_name(cb.cb_zhp);
1574                         ret = TRUE;
1575                 } else {
1576                         ret = FALSE;
1577                 }
1578                 break;
1579
1580         default:
1581                 ret = B_FALSE;
1582         }
1583
1584
1585         if (ret) {
1586                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1587                         if (cb.cb_zhp)
1588                                 zpool_close(cb.cb_zhp);
1589                         nvlist_free(config);
1590                         return (-1);
1591                 }
1592                 *state = (pool_state_t)stateval;
1593         }
1594
1595         if (cb.cb_zhp)
1596                 zpool_close(cb.cb_zhp);
1597
1598         nvlist_free(config);
1599         *inuse = ret;
1600         return (0);
1601 }