lib/libzfs/libzfs_import.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  24  * Copyright (c) 2011 by Delphix. All rights reserved.
  25  */
  26
  27 /*
  28  * Pool import support functions.
  29  *
  30  * To import a pool, we rely on reading the configuration information from the
  31  * ZFS label of each device.  If we successfully read the label, then we
  32  * organize the configuration information in the following hierarchy:
  33  *
  34  *      pool guid -> toplevel vdev guid -> label txg
  35  *
  36  * Duplicate entries matching this same tuple will be discarded.  Once we have
  37  * examined every device, we pick the best label txg config for each toplevel
  38  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  39  * update any paths that have changed.  Finally, we attempt to import the pool
  40  * using our derived config, and record the results.
  41  */
  42
  43 #include <ctype.h>
  44 #include <devid.h>
  45 #include <dirent.h>
  46 #include <errno.h>
  47 #include <libintl.h>
  48 #include <stddef.h>
  49 #include <stdlib.h>
  50 #include <string.h>
  51 #include <sys/stat.h>
  52 #include <unistd.h>
  53 #include <fcntl.h>
  54 #include <sys/vtoc.h>
  55 #include <sys/dktp/fdisk.h>
  56 #include <sys/efi_partition.h>
  57
  58 #include <sys/vdev_impl.h>
  59 #ifdef HAVE_LIBBLKID
  60 #include <blkid/blkid.h>
  61 #endif
  62
  63 #include "libzfs.h"
  64 #include "libzfs_impl.h"
  65
  66 /*
  67  * Intermediate structures used to gather configuration information.
  68  */
  69 typedef struct config_entry {
  70         uint64_t                ce_txg;
  71         nvlist_t                *ce_config;
  72         struct config_entry     *ce_next;
  73 } config_entry_t;
  74
  75 typedef struct vdev_entry {
  76         uint64_t                ve_guid;
  77         config_entry_t          *ve_configs;
  78         struct vdev_entry       *ve_next;
  79 } vdev_entry_t;
  80
  81 typedef struct pool_entry {
  82         uint64_t                pe_guid;
  83         vdev_entry_t            *pe_vdevs;
  84         struct pool_entry       *pe_next;
  85 } pool_entry_t;
  86
  87 typedef struct name_entry {
  88         char                    *ne_name;
  89         uint64_t                ne_guid;
  90         struct name_entry       *ne_next;
  91 } name_entry_t;
  92
  93 typedef struct pool_list {
  94         pool_entry_t            *pools;
  95         name_entry_t            *names;
  96 } pool_list_t;
  97
  98 static char *
  99 get_devid(const char *path)
 100 {
 101         int fd;
 102         ddi_devid_t devid;
 103         char *minor, *ret;
 104
 105         if ((fd = open(path, O_RDONLY)) < 0)
 106                 return (NULL);
 107
 108         minor = NULL;
 109         ret = NULL;
 110         if (devid_get(fd, &devid) == 0) {
 111                 if (devid_get_minor_name(fd, &minor) == 0)
 112                         ret = devid_str_encode(devid, minor);
 113                 if (minor != NULL)
 114                         devid_str_free(minor);
 115                 devid_free(devid);
 116         }
 117         (void) close(fd);
 118
 119         return (ret);
 120 }
 121
 122
 123 /*
 124  * Go through and fix up any path and/or devid information for the given vdev
 125  * configuration.
 126  */
 127 static int
 128 fix_paths(nvlist_t *nv, name_entry_t *names)
 129 {
 130         nvlist_t **child;
 131         uint_t c, children;
 132         uint64_t guid;
 133         name_entry_t *ne, *best;
 134         char *path, *devid;
 135         int matched;
 136
 137         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 138             &child, &children) == 0) {
 139                 for (c = 0; c < children; c++)
 140                         if (fix_paths(child[c], names) != 0)
 141                                 return (-1);
 142                 return (0);
 143         }
 144
 145         /*
 146          * This is a leaf (file or disk) vdev.  In either case, go through
 147          * the name list and see if we find a matching guid.  If so, replace
 148          * the path and see if we can calculate a new devid.
 149          *
 150          * There may be multiple names associated with a particular guid, in
 151          * which case we have overlapping slices or multiple paths to the same
 152          * disk.  If this is the case, then we want to pick the path that is
 153          * the most similar to the original, where "most similar" is the number
 154          * of matching characters starting from the end of the path.  This will
 155          * preserve slice numbers even if the disks have been reorganized, and
 156          * will also catch preferred disk names if multiple paths exist.
 157          */
 158         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 159         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 160                 path = NULL;
 161
 162         matched = 0;
 163         best = NULL;
 164         for (ne = names; ne != NULL; ne = ne->ne_next) {
 165                 if (ne->ne_guid == guid) {
 166                         const char *src, *dst;
 167                         int count;
 168
 169                         if (path == NULL) {
 170                                 best = ne;
 171                                 break;
 172                         }
 173
 174                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 175                         dst = path + strlen(path) - 1;
 176                         for (count = 0; src >= ne->ne_name && dst >= path;
 177                             src--, dst--, count++)
 178                                 if (*src != *dst)
 179                                         break;
 180
 181                         /*
 182                          * At this point, 'count' is the number of characters
 183                          * matched from the end.
 184                          */
 185                         if (count > matched || best == NULL) {
 186                                 best = ne;
 187                                 matched = count;
 188                         }
 189                 }
 190         }
 191
 192         if (best == NULL)
 193                 return (0);
 194
 195         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 196                 return (-1);
 197
 198         if ((devid = get_devid(best->ne_name)) == NULL) {
 199                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 200         } else {
 201                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
 202                         return (-1);
 203                 devid_str_free(devid);
 204         }
 205
 206         return (0);
 207 }
 208
 209 /*
 210  * Add the given configuration to the list of known devices.
 211  */
 212 static int
 213 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 214     nvlist_t *config)
 215 {
 216         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 217         pool_entry_t *pe;
 218         vdev_entry_t *ve;
 219         config_entry_t *ce;
 220         name_entry_t *ne;
 221
 222         /*
 223          * If this is a hot spare not currently in use or level 2 cache
 224          * device, add it to the list of names to translate, but don't do
 225          * anything else.
 226          */
 227         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 228             &state) == 0 &&
 229             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 230             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 231                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 232                         return (-1);
 233
 234                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 235                         free(ne);
 236                         return (-1);
 237                 }
 238                 ne->ne_guid = vdev_guid;
 239                 ne->ne_next = pl->names;
 240                 pl->names = ne;
 241                 return (0);
 242         }
 243
 244         /*
 245          * If we have a valid config but cannot read any of these fields, then
 246          * it means we have a half-initialized label.  In vdev_label_init()
 247          * we write a label with txg == 0 so that we can identify the device
 248          * in case the user refers to the same disk later on.  If we fail to
 249          * create the pool, we'll be left with a label in this state
 250          * which should not be considered part of a valid pool.
 251          */
 252         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 253             &pool_guid) != 0 ||
 254             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 255             &vdev_guid) != 0 ||
 256             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 257             &top_guid) != 0 ||
 258             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 259             &txg) != 0 || txg == 0) {
 260                 nvlist_free(config);
 261                 return (0);
 262         }
 263
 264         /*
 265          * First, see if we know about this pool.  If not, then add it to the
 266          * list of known pools.
 267          */
 268         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 269                 if (pe->pe_guid == pool_guid)
 270                         break;
 271         }
 272
 273         if (pe == NULL) {
 274                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 275                         nvlist_free(config);
 276                         return (-1);
 277                 }
 278                 pe->pe_guid = pool_guid;
 279                 pe->pe_next = pl->pools;
 280                 pl->pools = pe;
 281         }
 282
 283         /*
 284          * Second, see if we know about this toplevel vdev.  Add it if its
 285          * missing.
 286          */
 287         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 288                 if (ve->ve_guid == top_guid)
 289                         break;
 290         }
 291
 292         if (ve == NULL) {
 293                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 294                         nvlist_free(config);
 295                         return (-1);
 296                 }
 297                 ve->ve_guid = top_guid;
 298                 ve->ve_next = pe->pe_vdevs;
 299                 pe->pe_vdevs = ve;
 300         }
 301
 302         /*
 303          * Third, see if we have a config with a matching transaction group.  If
 304          * so, then we do nothing.  Otherwise, add it to the list of known
 305          * configs.
 306          */
 307         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 308                 if (ce->ce_txg == txg)
 309                         break;
 310         }
 311
 312         if (ce == NULL) {
 313                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 314                         nvlist_free(config);
 315                         return (-1);
 316                 }
 317                 ce->ce_txg = txg;
 318                 ce->ce_config = config;
 319                 ce->ce_next = ve->ve_configs;
 320                 ve->ve_configs = ce;
 321         } else {
 322                 nvlist_free(config);
 323         }
 324
 325         /*
 326          * At this point we've successfully added our config to the list of
 327          * known configs.  The last thing to do is add the vdev guid -> path
 328          * mappings so that we can fix up the configuration as necessary before
 329          * doing the import.
 330          */
 331         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 332                 return (-1);
 333
 334         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 335                 free(ne);
 336                 return (-1);
 337         }
 338
 339         ne->ne_guid = vdev_guid;
 340         ne->ne_next = pl->names;
 341         pl->names = ne;
 342
 343         return (0);
 344 }
 345
 346 /*
 347  * Returns true if the named pool matches the given GUID.
 348  */
 349 static int
 350 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 351     boolean_t *isactive)
 352 {
 353         zpool_handle_t *zhp;
 354         uint64_t theguid;
 355
 356         if (zpool_open_silent(hdl, name, &zhp) != 0)
 357                 return (-1);
 358
 359         if (zhp == NULL) {
 360                 *isactive = B_FALSE;
 361                 return (0);
 362         }
 363
 364         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 365             &theguid) == 0);
 366
 367         zpool_close(zhp);
 368
 369         *isactive = (theguid == guid);
 370         return (0);
 371 }
 372
 373 static nvlist_t *
 374 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 375 {
 376         nvlist_t *nvl;
 377         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
 378         int err;
 379
 380         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 381                 return (NULL);
 382
 383         if (zcmd_alloc_dst_nvlist(hdl, &zc,
 384             zc.zc_nvlist_conf_size * 2) != 0) {
 385                 zcmd_free_nvlists(&zc);
 386                 return (NULL);
 387         }
 388
 389         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 390             &zc)) != 0 && errno == ENOMEM) {
 391                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 392                         zcmd_free_nvlists(&zc);
 393                         return (NULL);
 394                 }
 395         }
 396
 397         if (err) {
 398                 zcmd_free_nvlists(&zc);
 399                 return (NULL);
 400         }
 401
 402         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 403                 zcmd_free_nvlists(&zc);
 404                 return (NULL);
 405         }
 406
 407         zcmd_free_nvlists(&zc);
 408         return (nvl);
 409 }
 410
 411 /*
 412  * Determine if the vdev id is a hole in the namespace.
 413  */
 414 boolean_t
 415 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
 416 {
 417         int c;
 418
 419         for (c = 0; c < holes; c++) {
 420
 421                 /* Top-level is a hole */
 422                 if (hole_array[c] == id)
 423                         return (B_TRUE);
 424         }
 425         return (B_FALSE);
 426 }
 427
 428 /*
 429  * Convert our list of pools into the definitive set of configurations.  We
 430  * start by picking the best config for each toplevel vdev.  Once that's done,
 431  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 432  * pass to fix up any incorrect paths, and then add it to the main list to
 433  * return to the user.
 434  */
 435 static nvlist_t *
 436 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 437 {
 438         pool_entry_t *pe;
 439         vdev_entry_t *ve;
 440         config_entry_t *ce;
 441         nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
 442         nvlist_t **spares, **l2cache;
 443         uint_t i, nspares, nl2cache;
 444         boolean_t config_seen;
 445         uint64_t best_txg;
 446         char *name, *hostname, *comment;
 447         uint64_t version, guid;
 448         uint_t children = 0;
 449         nvlist_t **child = NULL;
 450         uint_t holes;
 451         uint64_t *hole_array, max_id;
 452         uint_t c;
 453         boolean_t isactive;
 454         uint64_t hostid;
 455         nvlist_t *nvl;
 456         boolean_t found_one = B_FALSE;
 457         boolean_t valid_top_config = B_FALSE;
 458
 459         if (nvlist_alloc(&ret, 0, 0) != 0)
 460                 goto nomem;
 461
 462         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 463                 uint64_t id, max_txg = 0;
 464
 465                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 466                         goto nomem;
 467                 config_seen = B_FALSE;
 468
 469                 /*
 470                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 471                  * from the first one we find, and then go through the rest and
 472                  * add them as necessary to the 'vdevs' member of the config.
 473                  */
 474                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 475
 476                         /*
 477                          * Determine the best configuration for this vdev by
 478                          * selecting the config with the latest transaction
 479                          * group.
 480                          */
 481                         best_txg = 0;
 482                         for (ce = ve->ve_configs; ce != NULL;
 483                             ce = ce->ce_next) {
 484
 485                                 if (ce->ce_txg > best_txg) {
 486                                         tmp = ce->ce_config;
 487                                         best_txg = ce->ce_txg;
 488                                 }
 489                         }
 490
 491                         /*
 492                          * We rely on the fact that the max txg for the
 493                          * pool will contain the most up-to-date information
 494                          * about the valid top-levels in the vdev namespace.
 495                          */
 496                         if (best_txg > max_txg) {
 497                                 (void) nvlist_remove(config,
 498                                     ZPOOL_CONFIG_VDEV_CHILDREN,
 499                                     DATA_TYPE_UINT64);
 500                                 (void) nvlist_remove(config,
 501                                     ZPOOL_CONFIG_HOLE_ARRAY,
 502                                     DATA_TYPE_UINT64_ARRAY);
 503
 504                                 max_txg = best_txg;
 505                                 hole_array = NULL;
 506                                 holes = 0;
 507                                 max_id = 0;
 508                                 valid_top_config = B_FALSE;
 509
 510                                 if (nvlist_lookup_uint64(tmp,
 511                                     ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
 512                                         verify(nvlist_add_uint64(config,
 513                                             ZPOOL_CONFIG_VDEV_CHILDREN,
 514                                             max_id) == 0);
 515                                         valid_top_config = B_TRUE;
 516                                 }
 517
 518                                 if (nvlist_lookup_uint64_array(tmp,
 519                                     ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
 520                                     &holes) == 0) {
 521                                         verify(nvlist_add_uint64_array(config,
 522                                             ZPOOL_CONFIG_HOLE_ARRAY,
 523                                             hole_array, holes) == 0);
 524                                 }
 525                         }
 526
 527                         if (!config_seen) {
 528                                 /*
 529                                  * Copy the relevant pieces of data to the pool
 530                                  * configuration:
 531                                  *
 532                                  *      version
 533                                  *      pool guid
 534                                  *      name
 535                                  *      comment (if available)
 536                                  *      pool state
 537                                  *      hostid (if available)
 538                                  *      hostname (if available)
 539                                  */
 540                                 uint64_t state;
 541
 542                                 verify(nvlist_lookup_uint64(tmp,
 543                                     ZPOOL_CONFIG_VERSION, &version) == 0);
 544                                 if (nvlist_add_uint64(config,
 545                                     ZPOOL_CONFIG_VERSION, version) != 0)
 546                                         goto nomem;
 547                                 verify(nvlist_lookup_uint64(tmp,
 548                                     ZPOOL_CONFIG_POOL_GUID, &guid) == 0);
 549                                 if (nvlist_add_uint64(config,
 550                                     ZPOOL_CONFIG_POOL_GUID, guid) != 0)
 551                                         goto nomem;
 552                                 verify(nvlist_lookup_string(tmp,
 553                                     ZPOOL_CONFIG_POOL_NAME, &name) == 0);
 554                                 if (nvlist_add_string(config,
 555                                     ZPOOL_CONFIG_POOL_NAME, name) != 0)
 556                                         goto nomem;
 557
 558                                 /*
 559                                  * COMMENT is optional, don't bail if it's not
 560                                  * there, instead, set it to NULL.
 561                                  */
 562                                 if (nvlist_lookup_string(tmp,
 563                                     ZPOOL_CONFIG_COMMENT, &comment) != 0)
 564                                         comment = NULL;
 565                                 else if (nvlist_add_string(config,
 566                                     ZPOOL_CONFIG_COMMENT, comment) != 0)
 567                                         goto nomem;
 568
 569                                 verify(nvlist_lookup_uint64(tmp,
 570                                     ZPOOL_CONFIG_POOL_STATE, &state) == 0);
 571                                 if (nvlist_add_uint64(config,
 572                                     ZPOOL_CONFIG_POOL_STATE, state) != 0)
 573                                         goto nomem;
 574
 575                                 hostid = 0;
 576                                 if (nvlist_lookup_uint64(tmp,
 577                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 578                                         if (nvlist_add_uint64(config,
 579                                             ZPOOL_CONFIG_HOSTID, hostid) != 0)
 580                                                 goto nomem;
 581                                         verify(nvlist_lookup_string(tmp,
 582                                             ZPOOL_CONFIG_HOSTNAME,
 583                                             &hostname) == 0);
 584                                         if (nvlist_add_string(config,
 585                                             ZPOOL_CONFIG_HOSTNAME,
 586                                             hostname) != 0)
 587                                                 goto nomem;
 588                                 }
 589
 590                                 config_seen = B_TRUE;
 591                         }
 592
 593                         /*
 594                          * Add this top-level vdev to the child array.
 595                          */
 596                         verify(nvlist_lookup_nvlist(tmp,
 597                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 598                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 599                             &id) == 0);
 600
 601                         if (id >= children) {
 602                                 nvlist_t **newchild;
 603
 604                                 newchild = zfs_alloc(hdl, (id + 1) *
 605                                     sizeof (nvlist_t *));
 606                                 if (newchild == NULL)
 607                                         goto nomem;
 608
 609                                 for (c = 0; c < children; c++)
 610                                         newchild[c] = child[c];
 611
 612                                 free(child);
 613                                 child = newchild;
 614                                 children = id + 1;
 615                         }
 616                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 617                                 goto nomem;
 618
 619                 }
 620
 621                 /*
 622                  * If we have information about all the top-levels then
 623                  * clean up the nvlist which we've constructed. This
 624                  * means removing any extraneous devices that are
 625                  * beyond the valid range or adding devices to the end
 626                  * of our array which appear to be missing.
 627                  */
 628                 if (valid_top_config) {
 629                         if (max_id < children) {
 630                                 for (c = max_id; c < children; c++)
 631                                         nvlist_free(child[c]);
 632                                 children = max_id;
 633                         } else if (max_id > children) {
 634                                 nvlist_t **newchild;
 635
 636                                 newchild = zfs_alloc(hdl, (max_id) *
 637                                     sizeof (nvlist_t *));
 638                                 if (newchild == NULL)
 639                                         goto nomem;
 640
 641                                 for (c = 0; c < children; c++)
 642                                         newchild[c] = child[c];
 643
 644                                 free(child);
 645                                 child = newchild;
 646                                 children = max_id;
 647                         }
 648                 }
 649
 650                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 651                     &guid) == 0);
 652
 653                 /*
 654                  * The vdev namespace may contain holes as a result of
 655                  * device removal. We must add them back into the vdev
 656                  * tree before we process any missing devices.
 657                  */
 658                 if (holes > 0) {
 659                         ASSERT(valid_top_config);
 660
 661                         for (c = 0; c < children; c++) {
 662                                 nvlist_t *holey;
 663
 664                                 if (child[c] != NULL ||
 665                                     !vdev_is_hole(hole_array, holes, c))
 666                                         continue;
 667
 668                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 669                                     0) != 0)
 670                                         goto nomem;
 671
 672                                 /*
 673                                  * Holes in the namespace are treated as
 674                                  * "hole" top-level vdevs and have a
 675                                  * special flag set on them.
 676                                  */
 677                                 if (nvlist_add_string(holey,
 678                                     ZPOOL_CONFIG_TYPE,
 679                                     VDEV_TYPE_HOLE) != 0 ||
 680                                     nvlist_add_uint64(holey,
 681                                     ZPOOL_CONFIG_ID, c) != 0 ||
 682                                     nvlist_add_uint64(holey,
 683                                     ZPOOL_CONFIG_GUID, 0ULL) != 0)
 684                                         goto nomem;
 685                                 child[c] = holey;
 686                         }
 687                 }
 688
 689                 /*
 690                  * Look for any missing top-level vdevs.  If this is the case,
 691                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 692                  * simply compress the child array, because the kernel performs
 693                  * certain checks to make sure the vdev IDs match their location
 694                  * in the configuration.
 695                  */
 696                 for (c = 0; c < children; c++) {
 697                         if (child[c] == NULL) {
 698                                 nvlist_t *missing;
 699                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 700                                     0) != 0)
 701                                         goto nomem;
 702                                 if (nvlist_add_string(missing,
 703                                     ZPOOL_CONFIG_TYPE,
 704                                     VDEV_TYPE_MISSING) != 0 ||
 705                                     nvlist_add_uint64(missing,
 706                                     ZPOOL_CONFIG_ID, c) != 0 ||
 707                                     nvlist_add_uint64(missing,
 708                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 709                                         nvlist_free(missing);
 710                                         goto nomem;
 711                                 }
 712                                 child[c] = missing;
 713                         }
 714                 }
 715
 716                 /*
 717                  * Put all of this pool's top-level vdevs into a root vdev.
 718                  */
 719                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 720                         goto nomem;
 721                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 722                     VDEV_TYPE_ROOT) != 0 ||
 723                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 724                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 725                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 726                     child, children) != 0) {
 727                         nvlist_free(nvroot);
 728                         goto nomem;
 729                 }
 730
 731                 for (c = 0; c < children; c++)
 732                         nvlist_free(child[c]);
 733                 free(child);
 734                 children = 0;
 735                 child = NULL;
 736
 737                 /*
 738                  * Go through and fix up any paths and/or devids based on our
 739                  * known list of vdev GUID -> path mappings.
 740                  */
 741                 if (fix_paths(nvroot, pl->names) != 0) {
 742                         nvlist_free(nvroot);
 743                         goto nomem;
 744                 }
 745
 746                 /*
 747                  * Add the root vdev to this pool's configuration.
 748                  */
 749                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 750                     nvroot) != 0) {
 751                         nvlist_free(nvroot);
 752                         goto nomem;
 753                 }
 754                 nvlist_free(nvroot);
 755
 756                 /*
 757                  * zdb uses this path to report on active pools that were
 758                  * imported or created using -R.
 759                  */
 760                 if (active_ok)
 761                         goto add_pool;
 762
 763                 /*
 764                  * Determine if this pool is currently active, in which case we
 765                  * can't actually import it.
 766                  */
 767                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 768                     &name) == 0);
 769                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 770                     &guid) == 0);
 771
 772                 if (pool_active(hdl, name, guid, &isactive) != 0)
 773                         goto error;
 774
 775                 if (isactive) {
 776                         nvlist_free(config);
 777                         config = NULL;
 778                         continue;
 779                 }
 780
 781                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 782                         nvlist_free(config);
 783                         config = NULL;
 784                         continue;
 785                 }
 786
 787                 nvlist_free(config);
 788                 config = nvl;
 789
 790                 /*
 791                  * Go through and update the paths for spares, now that we have
 792                  * them.
 793                  */
 794                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 795                     &nvroot) == 0);
 796                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 797                     &spares, &nspares) == 0) {
 798                         for (i = 0; i < nspares; i++) {
 799                                 if (fix_paths(spares[i], pl->names) != 0)
 800                                         goto nomem;
 801                         }
 802                 }
 803
 804                 /*
 805                  * Update the paths for l2cache devices.
 806                  */
 807                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 808                     &l2cache, &nl2cache) == 0) {
 809                         for (i = 0; i < nl2cache; i++) {
 810                                 if (fix_paths(l2cache[i], pl->names) != 0)
 811                                         goto nomem;
 812                         }
 813                 }
 814
 815                 /*
 816                  * Restore the original information read from the actual label.
 817                  */
 818                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 819                     DATA_TYPE_UINT64);
 820                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 821                     DATA_TYPE_STRING);
 822                 if (hostid != 0) {
 823                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 824                             hostid) == 0);
 825                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 826                             hostname) == 0);
 827                 }
 828
 829 add_pool:
 830                 /*
 831                  * Add this pool to the list of configs.
 832                  */
 833                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 834                     &name) == 0);
 835                 if (nvlist_add_nvlist(ret, name, config) != 0)
 836                         goto nomem;
 837
 838                 found_one = B_TRUE;
 839                 nvlist_free(config);
 840                 config = NULL;
 841         }
 842
 843         if (!found_one) {
 844                 nvlist_free(ret);
 845                 ret = NULL;
 846         }
 847
 848         return (ret);
 849
 850 nomem:
 851         (void) no_memory(hdl);
 852 error:
 853         nvlist_free(config);
 854         nvlist_free(ret);
 855         for (c = 0; c < children; c++)
 856                 nvlist_free(child[c]);
 857         free(child);
 858
 859         return (NULL);
 860 }
 861
 862 /*
 863  * Return the offset of the given label.
 864  */
 865 static uint64_t
 866 label_offset(uint64_t size, int l)
 867 {
 868         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 869         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 870             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 871 }
 872
 873 /*
 874  * Given a file descriptor, read the label information and return an nvlist
 875  * describing the configuration, if there is one.
 876  */
 877 int
 878 zpool_read_label(int fd, nvlist_t **config)
 879 {
 880         struct stat64 statbuf;
 881         int l;
 882         vdev_label_t *label;
 883         uint64_t state, txg, size;
 884
 885         *config = NULL;
 886
 887         if (fstat64(fd, &statbuf) == -1)
 888                 return (0);
 889         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 890
 891         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 892                 return (-1);
 893
 894         for (l = 0; l < VDEV_LABELS; l++) {
 895                 if (pread64(fd, label, sizeof (vdev_label_t),
 896                     label_offset(size, l)) != sizeof (vdev_label_t))
 897                         continue;
 898
 899                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 900                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 901                         continue;
 902
 903                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 904                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 905                         nvlist_free(*config);
 906                         continue;
 907                 }
 908
 909                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 910                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 911                     &txg) != 0 || txg == 0)) {
 912                         nvlist_free(*config);
 913                         continue;
 914                 }
 915
 916                 free(label);
 917                 return (0);
 918         }
 919
 920         free(label);
 921         *config = NULL;
 922         return (0);
 923 }
 924
 925 #ifdef HAVE_LIBBLKID
 926 /*
 927  * Use libblkid to quickly search for zfs devices
 928  */
 929 static int
 930 zpool_find_import_blkid(libzfs_handle_t *hdl, pool_list_t *pools)
 931 {
 932         blkid_cache cache;
 933         blkid_dev_iterate iter;
 934         blkid_dev dev;
 935         const char *devname;
 936         nvlist_t *config;
 937         int fd, err;
 938
 939         err = blkid_get_cache(&cache, NULL);
 940         if (err != 0) {
 941                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 942                     dgettext(TEXT_DOMAIN, "blkid_get_cache() %d"), err);
 943                 goto err_blkid1;
 944         }
 945
 946         err = blkid_probe_all(cache);
 947         if (err != 0) {
 948                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 949                     dgettext(TEXT_DOMAIN, "blkid_probe_all() %d"), err);
 950                 goto err_blkid2;
 951         }
 952
 953         iter = blkid_dev_iterate_begin(cache);
 954         if (iter == NULL) {
 955                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 956                     dgettext(TEXT_DOMAIN, "blkid_dev_iterate_begin()"));
 957                 goto err_blkid2;
 958         }
 959
 960         err = blkid_dev_set_search(iter, "TYPE", "zfs");
 961         if (err != 0) {
 962                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
 963                     dgettext(TEXT_DOMAIN, "blkid_dev_set_search() %d"), err);
 964                 goto err_blkid3;
 965         }
 966
 967         while (blkid_dev_next(iter, &dev) == 0) {
 968                 devname = blkid_dev_devname(dev);
 969                 if ((fd = open64(devname, O_RDONLY)) < 0)
 970                         continue;
 971
 972                 err = zpool_read_label(fd, &config);
 973                 (void) close(fd);
 974
 975                 if (err != 0) {
 976                         (void) no_memory(hdl);
 977                         goto err_blkid3;
 978                 }
 979
 980                 if (config != NULL) {
 981                         err = add_config(hdl, pools, devname, config);
 982                         if (err != 0)
 983                                 goto err_blkid3;
 984                 }
 985         }
 986
 987 err_blkid3:
 988         blkid_dev_iterate_end(iter);
 989 err_blkid2:
 990         blkid_put_cache(cache);
 991 err_blkid1:
 992         return err;
 993 }
 994 #endif /* HAVE_LIBBLKID */
 995
 996 /*
 997  * Given a list of directories to search, find all pools stored on disk.  This
 998  * includes partial pools which are not available to import.  If no args are
 999  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1000  * poolname or guid (but not both) are provided by the caller when trying
1001  * to import a specific pool.
1002  */
1003 static nvlist_t *
1004 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1005 {
1006         int i, dirs = iarg->paths;
1007         DIR *dirp = NULL;
1008         struct dirent64 *dp;
1009         char path[MAXPATHLEN];
1010         char *end, **dir = iarg->path;
1011         size_t pathleft;
1012         struct stat64 statbuf;
1013         nvlist_t *ret = NULL, *config;
1014         static char *default_dir = DISK_ROOT;
1015         int fd;
1016         pool_list_t pools = { 0 };
1017         pool_entry_t *pe, *penext;
1018         vdev_entry_t *ve, *venext;
1019         config_entry_t *ce, *cenext;
1020         name_entry_t *ne, *nenext;
1021
1022         verify(iarg->poolname == NULL || iarg->guid == 0);
1023
1024         if (dirs == 0) {
1025 #ifdef HAVE_LIBBLKID
1026                 /* Use libblkid to scan all device for their type */
1027                 if (zpool_find_import_blkid(hdl, &pools) == 0)
1028                         goto skip_scanning;
1029
1030                 (void) zfs_error_fmt(hdl, EZFS_BADCACHE,
1031                     dgettext(TEXT_DOMAIN, "blkid failure falling back "
1032                     "to manual probing"));
1033 #endif /* HAVE_LIBBLKID */
1034                 dirs = 1;
1035                 dir = &default_dir;
1036         }
1037
1038         /*
1039          * Go through and read the label configuration information from every
1040          * possible device, organizing the information according to pool GUID
1041          * and toplevel GUID.
1042          */
1043         for (i = 0; i < dirs; i++) {
1044                 char *rdsk;
1045                 int dfd;
1046
1047                 /* use realpath to normalize the path */
1048                 if (realpath(dir[i], path) == 0) {
1049                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1050                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1051                         goto error;
1052                 }
1053                 end = &path[strlen(path)];
1054                 *end++ = '/';
1055                 *end = 0;
1056                 pathleft = &path[sizeof (path)] - end;
1057
1058                 /*
1059                  * Using raw devices instead of block devices when we're
1060                  * reading the labels skips a bunch of slow operations during
1061                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1062                  */
1063                 if (strcmp(path, "/dev/dsk/") == 0)
1064                         rdsk = "/dev/rdsk/";
1065                 else
1066                         rdsk = path;
1067
1068                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1069                     (dirp = fdopendir(dfd)) == NULL) {
1070                         zfs_error_aux(hdl, strerror(errno));
1071                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1072                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1073                             rdsk);
1074                         goto error;
1075                 }
1076
1077                 /*
1078                  * This is not MT-safe, but we have no MT consumers of libzfs
1079                  */
1080                 while ((dp = readdir64(dirp)) != NULL) {
1081                         const char *name = dp->d_name;
1082                         if (name[0] == '.' &&
1083                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1084                                 continue;
1085
1086                         /*
1087                          * Skip checking devices with well known prefixes:
1088                          * watchdog - A special close is required to avoid
1089                          *            triggering it and resetting the system.
1090                          * fuse     - Fuse control device.
1091                          * ppp      - Generic PPP driver.
1092                          * tty*     - Generic serial interface.
1093                          * vcs*     - Virtual console memory.
1094                          * parport* - Parallel port interface.
1095                          * lp*      - Printer interface.
1096                          * fd*      - Floppy interface.
1097                          * hpet     - High Precision Event Timer, crashes qemu
1098                          *            when accessed from a virtual machine.
1099                          * core     - Symlink to /proc/kcore, causes a crash
1100                          *            when access from Xen dom0.
1101                          */
1102                         if ((strncmp(name, "watchdog", 8) == 0) ||
1103                             (strncmp(name, "fuse", 4) == 0)     ||
1104                             (strncmp(name, "ppp", 3) == 0)      ||
1105                             (strncmp(name, "tty", 3) == 0)      ||
1106                             (strncmp(name, "vcs", 3) == 0)      ||
1107                             (strncmp(name, "parport", 7) == 0)  ||
1108                             (strncmp(name, "lp", 2) == 0)       ||
1109                             (strncmp(name, "fd", 2) == 0)       ||
1110                             (strncmp(name, "hpet", 4) == 0)     ||
1111                             (strncmp(name, "core", 4) == 0))
1112                                 continue;
1113
1114                         /*
1115                          * Ignore failed stats.  We only want regular
1116                          * files and block devices.
1117                          */
1118                         if ((fstatat64(dfd, name, &statbuf, 0) != 0) ||
1119                             (!S_ISREG(statbuf.st_mode) &&
1120                             !S_ISBLK(statbuf.st_mode)))
1121                                 continue;
1122
1123                         if ((fd = openat64(dfd, name, O_RDONLY)) < 0)
1124                                 continue;
1125
1126                         if ((zpool_read_label(fd, &config)) != 0) {
1127                                 (void) close(fd);
1128                                 (void) no_memory(hdl);
1129                                 goto error;
1130                         }
1131
1132                         (void) close(fd);
1133
1134                         if (config != NULL) {
1135                                 boolean_t matched = B_TRUE;
1136
1137                                 if (iarg->poolname != NULL) {
1138                                         char *pname;
1139
1140                                         matched = nvlist_lookup_string(config,
1141                                             ZPOOL_CONFIG_POOL_NAME,
1142                                             &pname) == 0 &&
1143                                             strcmp(iarg->poolname, pname) == 0;
1144                                 } else if (iarg->guid != 0) {
1145                                         uint64_t this_guid;
1146
1147                                         matched = nvlist_lookup_uint64(config,
1148                                             ZPOOL_CONFIG_POOL_GUID,
1149                                             &this_guid) == 0 &&
1150                                             iarg->guid == this_guid;
1151                                 }
1152                                 if (!matched) {
1153                                         nvlist_free(config);
1154                                         config = NULL;
1155                                         continue;
1156                                 }
1157                                 /* use the non-raw path for the config */
1158                                 (void) strlcpy(end, name, pathleft);
1159                                 if (add_config(hdl, &pools, path, config) != 0)
1160                                         goto error;
1161                         }
1162                 }
1163
1164                 (void) closedir(dirp);
1165                 dirp = NULL;
1166         }
1167
1168 #ifdef HAVE_LIBBLKID
1169 skip_scanning:
1170 #endif
1171         ret = get_configs(hdl, &pools, iarg->can_be_active);
1172
1173 error:
1174         for (pe = pools.pools; pe != NULL; pe = penext) {
1175                 penext = pe->pe_next;
1176                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1177                         venext = ve->ve_next;
1178                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1179                                 cenext = ce->ce_next;
1180                                 if (ce->ce_config)
1181                                         nvlist_free(ce->ce_config);
1182                                 free(ce);
1183                         }
1184                         free(ve);
1185                 }
1186                 free(pe);
1187         }
1188
1189         for (ne = pools.names; ne != NULL; ne = nenext) {
1190                 nenext = ne->ne_next;
1191                 if (ne->ne_name)
1192                         free(ne->ne_name);
1193                 free(ne);
1194         }
1195
1196         if (dirp)
1197                 (void) closedir(dirp);
1198
1199         return (ret);
1200 }
1201
1202 nvlist_t *
1203 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1204 {
1205         importargs_t iarg = { 0 };
1206
1207         iarg.paths = argc;
1208         iarg.path = argv;
1209
1210         return (zpool_find_import_impl(hdl, &iarg));
1211 }
1212
1213 /*
1214  * Given a cache file, return the contents as a list of importable pools.
1215  * poolname or guid (but not both) are provided by the caller when trying
1216  * to import a specific pool.
1217  */
1218 nvlist_t *
1219 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1220     char *poolname, uint64_t guid)
1221 {
1222         char *buf;
1223         int fd;
1224         struct stat64 statbuf;
1225         nvlist_t *raw, *src, *dst;
1226         nvlist_t *pools;
1227         nvpair_t *elem;
1228         char *name;
1229         uint64_t this_guid;
1230         boolean_t active;
1231
1232         verify(poolname == NULL || guid == 0);
1233
1234         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1235                 zfs_error_aux(hdl, "%s", strerror(errno));
1236                 (void) zfs_error(hdl, EZFS_BADCACHE,
1237                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1238                 return (NULL);
1239         }
1240
1241         if (fstat64(fd, &statbuf) != 0) {
1242                 zfs_error_aux(hdl, "%s", strerror(errno));
1243                 (void) close(fd);
1244                 (void) zfs_error(hdl, EZFS_BADCACHE,
1245                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1246                 return (NULL);
1247         }
1248
1249         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1250                 (void) close(fd);
1251                 return (NULL);
1252         }
1253
1254         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1255                 (void) close(fd);
1256                 free(buf);
1257                 (void) zfs_error(hdl, EZFS_BADCACHE,
1258                     dgettext(TEXT_DOMAIN,
1259                     "failed to read cache file contents"));
1260                 return (NULL);
1261         }
1262
1263         (void) close(fd);
1264
1265         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1266                 free(buf);
1267                 (void) zfs_error(hdl, EZFS_BADCACHE,
1268                     dgettext(TEXT_DOMAIN,
1269                     "invalid or corrupt cache file contents"));
1270                 return (NULL);
1271         }
1272
1273         free(buf);
1274
1275         /*
1276          * Go through and get the current state of the pools and refresh their
1277          * state.
1278          */
1279         if (nvlist_alloc(&pools, 0, 0) != 0) {
1280                 (void) no_memory(hdl);
1281                 nvlist_free(raw);
1282                 return (NULL);
1283         }
1284
1285         elem = NULL;
1286         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1287                 verify(nvpair_value_nvlist(elem, &src) == 0);
1288
1289                 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1290                     &name) == 0);
1291                 if (poolname != NULL && strcmp(poolname, name) != 0)
1292                         continue;
1293
1294                 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1295                     &this_guid) == 0);
1296                 if (guid != 0) {
1297                         verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1298                             &this_guid) == 0);
1299                         if (guid != this_guid)
1300                                 continue;
1301                 }
1302
1303                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1304                         nvlist_free(raw);
1305                         nvlist_free(pools);
1306                         return (NULL);
1307                 }
1308
1309                 if (active)
1310                         continue;
1311
1312                 if ((dst = refresh_config(hdl, src)) == NULL) {
1313                         nvlist_free(raw);
1314                         nvlist_free(pools);
1315                         return (NULL);
1316                 }
1317
1318                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1319                         (void) no_memory(hdl);
1320                         nvlist_free(dst);
1321                         nvlist_free(raw);
1322                         nvlist_free(pools);
1323                         return (NULL);
1324                 }
1325                 nvlist_free(dst);
1326         }
1327
1328         nvlist_free(raw);
1329         return (pools);
1330 }
1331
1332 static int
1333 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1334 {
1335         importargs_t *import = data;
1336         int found = 0;
1337
1338         if (import->poolname != NULL) {
1339                 char *pool_name;
1340
1341                 verify(nvlist_lookup_string(zhp->zpool_config,
1342                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1343                 if (strcmp(pool_name, import->poolname) == 0)
1344                         found = 1;
1345         } else {
1346                 uint64_t pool_guid;
1347
1348                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1349                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1350                 if (pool_guid == import->guid)
1351                         found = 1;
1352         }
1353
1354         zpool_close(zhp);
1355         return (found);
1356 }
1357
1358 nvlist_t *
1359 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1360 {
1361         verify(import->poolname == NULL || import->guid == 0);
1362
1363         if (import->unique)
1364                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1365
1366         if (import->cachefile != NULL)
1367                 return (zpool_find_import_cached(hdl, import->cachefile,
1368                     import->poolname, import->guid));
1369
1370         return (zpool_find_import_impl(hdl, import));
1371 }
1372
1373 boolean_t
1374 find_guid(nvlist_t *nv, uint64_t guid)
1375 {
1376         uint64_t tmp;
1377         nvlist_t **child;
1378         uint_t c, children;
1379
1380         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1381         if (tmp == guid)
1382                 return (B_TRUE);
1383
1384         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1385             &child, &children) == 0) {
1386                 for (c = 0; c < children; c++)
1387                         if (find_guid(child[c], guid))
1388                                 return (B_TRUE);
1389         }
1390
1391         return (B_FALSE);
1392 }
1393
1394 typedef struct aux_cbdata {
1395         const char      *cb_type;
1396         uint64_t        cb_guid;
1397         zpool_handle_t  *cb_zhp;
1398 } aux_cbdata_t;
1399
1400 static int
1401 find_aux(zpool_handle_t *zhp, void *data)
1402 {
1403         aux_cbdata_t *cbp = data;
1404         nvlist_t **list;
1405         uint_t i, count;
1406         uint64_t guid;
1407         nvlist_t *nvroot;
1408
1409         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1410             &nvroot) == 0);
1411
1412         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1413             &list, &count) == 0) {
1414                 for (i = 0; i < count; i++) {
1415                         verify(nvlist_lookup_uint64(list[i],
1416                             ZPOOL_CONFIG_GUID, &guid) == 0);
1417                         if (guid == cbp->cb_guid) {
1418                                 cbp->cb_zhp = zhp;
1419                                 return (1);
1420                         }
1421                 }
1422         }
1423
1424         zpool_close(zhp);
1425         return (0);
1426 }
1427
1428 /*
1429  * Determines if the pool is in use.  If so, it returns true and the state of
1430  * the pool as well as the name of the pool.  Both strings are allocated and
1431  * must be freed by the caller.
1432  */
1433 int
1434 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1435     boolean_t *inuse)
1436 {
1437         nvlist_t *config;
1438         char *name;
1439         boolean_t ret;
1440         uint64_t guid, vdev_guid;
1441         zpool_handle_t *zhp;
1442         nvlist_t *pool_config;
1443         uint64_t stateval, isspare;
1444         aux_cbdata_t cb = { 0 };
1445         boolean_t isactive;
1446
1447         *inuse = B_FALSE;
1448
1449         if (zpool_read_label(fd, &config) != 0) {
1450                 (void) no_memory(hdl);
1451                 return (-1);
1452         }
1453
1454         if (config == NULL)
1455                 return (0);
1456
1457         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1458             &stateval) == 0);
1459         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1460             &vdev_guid) == 0);
1461
1462         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1463                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1464                     &name) == 0);
1465                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1466                     &guid) == 0);
1467         }
1468
1469         switch (stateval) {
1470         case POOL_STATE_EXPORTED:
1471                 /*
1472                  * A pool with an exported state may in fact be imported
1473                  * read-only, so check the in-core state to see if it's
1474                  * active and imported read-only.  If it is, set
1475                  * its state to active.
1476                  */
1477                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1478                     (zhp = zpool_open_canfail(hdl, name)) != NULL &&
1479                     zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1480                         stateval = POOL_STATE_ACTIVE;
1481
1482                 ret = B_TRUE;
1483                 break;
1484
1485         case POOL_STATE_ACTIVE:
1486                 /*
1487                  * For an active pool, we have to determine if it's really part
1488                  * of a currently active pool (in which case the pool will exist
1489                  * and the guid will be the same), or whether it's part of an
1490                  * active pool that was disconnected without being explicitly
1491                  * exported.
1492                  */
1493                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1494                         nvlist_free(config);
1495                         return (-1);
1496                 }
1497
1498                 if (isactive) {
1499                         /*
1500                          * Because the device may have been removed while
1501                          * offlined, we only report it as active if the vdev is
1502                          * still present in the config.  Otherwise, pretend like
1503                          * it's not in use.
1504                          */
1505                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1506                             (pool_config = zpool_get_config(zhp, NULL))
1507                             != NULL) {
1508                                 nvlist_t *nvroot;
1509
1510                                 verify(nvlist_lookup_nvlist(pool_config,
1511                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1512                                 ret = find_guid(nvroot, vdev_guid);
1513                         } else {
1514                                 ret = B_FALSE;
1515                         }
1516
1517                         /*
1518                          * If this is an active spare within another pool, we
1519                          * treat it like an unused hot spare.  This allows the
1520                          * user to create a pool with a hot spare that currently
1521                          * in use within another pool.  Since we return B_TRUE,
1522                          * libdiskmgt will continue to prevent generic consumers
1523                          * from using the device.
1524                          */
1525                         if (ret && nvlist_lookup_uint64(config,
1526                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1527                                 stateval = POOL_STATE_SPARE;
1528
1529                         if (zhp != NULL)
1530                                 zpool_close(zhp);
1531                 } else {
1532                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1533                         ret = B_TRUE;
1534                 }
1535                 break;
1536
1537         case POOL_STATE_SPARE:
1538                 /*
1539                  * For a hot spare, it can be either definitively in use, or
1540                  * potentially active.  To determine if it's in use, we iterate
1541                  * over all pools in the system and search for one with a spare
1542                  * with a matching guid.
1543                  *
1544                  * Due to the shared nature of spares, we don't actually report
1545                  * the potentially active case as in use.  This means the user
1546                  * can freely create pools on the hot spares of exported pools,
1547                  * but to do otherwise makes the resulting code complicated, and
1548                  * we end up having to deal with this case anyway.
1549                  */
1550                 cb.cb_zhp = NULL;
1551                 cb.cb_guid = vdev_guid;
1552                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1553                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1554                         name = (char *)zpool_get_name(cb.cb_zhp);
1555                         ret = TRUE;
1556                 } else {
1557                         ret = FALSE;
1558                 }
1559                 break;
1560
1561         case POOL_STATE_L2CACHE:
1562
1563                 /*
1564                  * Check if any pool is currently using this l2cache device.
1565                  */
1566                 cb.cb_zhp = NULL;
1567                 cb.cb_guid = vdev_guid;
1568                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1569                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1570                         name = (char *)zpool_get_name(cb.cb_zhp);
1571                         ret = TRUE;
1572                 } else {
1573                         ret = FALSE;
1574                 }
1575                 break;
1576
1577         default:
1578                 ret = B_FALSE;
1579         }
1580
1581
1582         if (ret) {
1583                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1584                         if (cb.cb_zhp)
1585                                 zpool_close(cb.cb_zhp);
1586                         nvlist_free(config);
1587                         return (-1);
1588                 }
1589                 *state = (pool_state_t)stateval;
1590         }
1591
1592         if (cb.cb_zhp)
1593                 zpool_close(cb.cb_zhp);
1594
1595         nvlist_free(config);
1596         *inuse = ret;
1597         return (0);
1598 }