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