cddl/contrib/opensolaris/lib/libzfs/common/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 /*
  23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  25  * Copyright (c) 2012 by Delphix. All rights reserved.
  26  * Copyright 2015 RackTop Systems.
  27  */
  28
  29 /*
  30  * Pool import support functions.
  31  *
  32  * To import a pool, we rely on reading the configuration information from the
  33  * ZFS label of each device.  If we successfully read the label, then we
  34  * organize the configuration information in the following hierarchy:
  35  *
  36  *      pool guid -> toplevel vdev guid -> label txg
  37  *
  38  * Duplicate entries matching this same tuple will be discarded.  Once we have
  39  * examined every device, we pick the best label txg config for each toplevel
  40  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  41  * update any paths that have changed.  Finally, we attempt to import the pool
  42  * using our derived config, and record the results.
  43  */
  44
  45 #include <ctype.h>
  46 #include <devid.h>
  47 #include <dirent.h>
  48 #include <errno.h>
  49 #include <libintl.h>
  50 #include <stddef.h>
  51 #include <stdlib.h>
  52 #include <string.h>
  53 #include <sys/stat.h>
  54 #include <unistd.h>
  55 #include <fcntl.h>
  56 #include <thread_pool.h>
  57 #include <libgeom.h>
  58
  59 #include <sys/vdev_impl.h>
  60
  61 #include "libzfs.h"
  62 #include "libzfs_impl.h"
  63
  64 /*
  65  * Intermediate structures used to gather configuration information.
  66  */
  67 typedef struct config_entry {
  68         uint64_t                ce_txg;
  69         nvlist_t                *ce_config;
  70         struct config_entry     *ce_next;
  71 } config_entry_t;
  72
  73 typedef struct vdev_entry {
  74         uint64_t                ve_guid;
  75         config_entry_t          *ve_configs;
  76         struct vdev_entry       *ve_next;
  77 } vdev_entry_t;
  78
  79 typedef struct pool_entry {
  80         uint64_t                pe_guid;
  81         vdev_entry_t            *pe_vdevs;
  82         struct pool_entry       *pe_next;
  83 } pool_entry_t;
  84
  85 typedef struct name_entry {
  86         char                    *ne_name;
  87         uint64_t                ne_guid;
  88         struct name_entry       *ne_next;
  89 } name_entry_t;
  90
  91 typedef struct pool_list {
  92         pool_entry_t            *pools;
  93         name_entry_t            *names;
  94 } pool_list_t;
  95
  96 static char *
  97 get_devid(const char *path)
  98 {
  99         int fd;
 100         ddi_devid_t devid;
 101         char *minor, *ret;
 102
 103         if ((fd = open(path, O_RDONLY)) < 0)
 104                 return (NULL);
 105
 106         minor = NULL;
 107         ret = NULL;
 108         if (devid_get(fd, &devid) == 0) {
 109                 if (devid_get_minor_name(fd, &minor) == 0)
 110                         ret = devid_str_encode(devid, minor);
 111                 if (minor != NULL)
 112                         devid_str_free(minor);
 113                 devid_free(devid);
 114         }
 115         (void) close(fd);
 116
 117         return (ret);
 118 }
 119
 120
 121 /*
 122  * Go through and fix up any path and/or devid information for the given vdev
 123  * configuration.
 124  */
 125 static int
 126 fix_paths(nvlist_t *nv, name_entry_t *names)
 127 {
 128         nvlist_t **child;
 129         uint_t c, children;
 130         uint64_t guid;
 131         name_entry_t *ne, *best;
 132         char *path, *devid;
 133         int matched;
 134
 135         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 136             &child, &children) == 0) {
 137                 for (c = 0; c < children; c++)
 138                         if (fix_paths(child[c], names) != 0)
 139                                 return (-1);
 140                 return (0);
 141         }
 142
 143         /*
 144          * This is a leaf (file or disk) vdev.  In either case, go through
 145          * the name list and see if we find a matching guid.  If so, replace
 146          * the path and see if we can calculate a new devid.
 147          *
 148          * There may be multiple names associated with a particular guid, in
 149          * which case we have overlapping slices or multiple paths to the same
 150          * disk.  If this is the case, then we want to pick the path that is
 151          * the most similar to the original, where "most similar" is the number
 152          * of matching characters starting from the end of the path.  This will
 153          * preserve slice numbers even if the disks have been reorganized, and
 154          * will also catch preferred disk names if multiple paths exist.
 155          */
 156         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 157         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 158                 path = NULL;
 159
 160         matched = 0;
 161         best = NULL;
 162         for (ne = names; ne != NULL; ne = ne->ne_next) {
 163                 if (ne->ne_guid == guid) {
 164                         const char *src, *dst;
 165                         int count;
 166
 167                         if (path == NULL) {
 168                                 best = ne;
 169                                 break;
 170                         }
 171
 172                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 173                         dst = path + strlen(path) - 1;
 174                         for (count = 0; src >= ne->ne_name && dst >= path;
 175                             src--, dst--, count++)
 176                                 if (*src != *dst)
 177                                         break;
 178
 179                         /*
 180                          * At this point, 'count' is the number of characters
 181                          * matched from the end.
 182                          */
 183                         if (count > matched || best == NULL) {
 184                                 best = ne;
 185                                 matched = count;
 186                         }
 187                 }
 188         }
 189
 190         if (best == NULL)
 191                 return (0);
 192
 193         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 194                 return (-1);
 195
 196         if ((devid = get_devid(best->ne_name)) == NULL) {
 197                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 198         } else {
 199                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
 200                         devid_str_free(devid);
 201                         return (-1);
 202                 }
 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 };
 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         for (int c = 0; c < holes; c++) {
 418
 419                 /* Top-level is a hole */
 420                 if (hole_array[c] == id)
 421                         return (B_TRUE);
 422         }
 423         return (B_FALSE);
 424 }
 425
 426 /*
 427  * Convert our list of pools into the definitive set of configurations.  We
 428  * start by picking the best config for each toplevel vdev.  Once that's done,
 429  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 430  * pass to fix up any incorrect paths, and then add it to the main list to
 431  * return to the user.
 432  */
 433 static nvlist_t *
 434 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 435 {
 436         pool_entry_t *pe;
 437         vdev_entry_t *ve;
 438         config_entry_t *ce;
 439         nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
 440         nvlist_t **spares, **l2cache;
 441         uint_t i, nspares, nl2cache;
 442         boolean_t config_seen;
 443         uint64_t best_txg;
 444         char *name, *hostname;
 445         uint64_t guid;
 446         uint_t children = 0;
 447         nvlist_t **child = NULL;
 448         uint_t holes;
 449         uint64_t *hole_array, max_id;
 450         uint_t c;
 451         boolean_t isactive;
 452         uint64_t hostid;
 453         nvlist_t *nvl;
 454         boolean_t found_one = B_FALSE;
 455         boolean_t valid_top_config = B_FALSE;
 456
 457         if (nvlist_alloc(&ret, 0, 0) != 0)
 458                 goto nomem;
 459
 460         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 461                 uint64_t id, max_txg = 0;
 462
 463                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 464                         goto nomem;
 465                 config_seen = B_FALSE;
 466
 467                 /*
 468                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 469                  * from the first one we find, and then go through the rest and
 470                  * add them as necessary to the 'vdevs' member of the config.
 471                  */
 472                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 473
 474                         /*
 475                          * Determine the best configuration for this vdev by
 476                          * selecting the config with the latest transaction
 477                          * group.
 478                          */
 479                         best_txg = 0;
 480                         for (ce = ve->ve_configs; ce != NULL;
 481                             ce = ce->ce_next) {
 482
 483                                 if (ce->ce_txg > best_txg) {
 484                                         tmp = ce->ce_config;
 485                                         best_txg = ce->ce_txg;
 486                                 }
 487                         }
 488
 489                         /*
 490                          * We rely on the fact that the max txg for the
 491                          * pool will contain the most up-to-date information
 492                          * about the valid top-levels in the vdev namespace.
 493                          */
 494                         if (best_txg > max_txg) {
 495                                 (void) nvlist_remove(config,
 496                                     ZPOOL_CONFIG_VDEV_CHILDREN,
 497                                     DATA_TYPE_UINT64);
 498                                 (void) nvlist_remove(config,
 499                                     ZPOOL_CONFIG_HOLE_ARRAY,
 500                                     DATA_TYPE_UINT64_ARRAY);
 501
 502                                 max_txg = best_txg;
 503                                 hole_array = NULL;
 504                                 holes = 0;
 505                                 max_id = 0;
 506                                 valid_top_config = B_FALSE;
 507
 508                                 if (nvlist_lookup_uint64(tmp,
 509                                     ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
 510                                         verify(nvlist_add_uint64(config,
 511                                             ZPOOL_CONFIG_VDEV_CHILDREN,
 512                                             max_id) == 0);
 513                                         valid_top_config = B_TRUE;
 514                                 }
 515
 516                                 if (nvlist_lookup_uint64_array(tmp,
 517                                     ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
 518                                     &holes) == 0) {
 519                                         verify(nvlist_add_uint64_array(config,
 520                                             ZPOOL_CONFIG_HOLE_ARRAY,
 521                                             hole_array, holes) == 0);
 522                                 }
 523                         }
 524
 525                         if (!config_seen) {
 526                                 /*
 527                                  * Copy the relevant pieces of data to the pool
 528                                  * configuration:
 529                                  *
 530                                  *      version
 531                                  *      pool guid
 532                                  *      name
 533                                  *      comment (if available)
 534                                  *      pool state
 535                                  *      hostid (if available)
 536                                  *      hostname (if available)
 537                                  */
 538                                 uint64_t state, version;
 539                                 char *comment = NULL;
 540
 541                                 version = fnvlist_lookup_uint64(tmp,
 542                                     ZPOOL_CONFIG_VERSION);
 543                                 fnvlist_add_uint64(config,
 544                                     ZPOOL_CONFIG_VERSION, version);
 545                                 guid = fnvlist_lookup_uint64(tmp,
 546                                     ZPOOL_CONFIG_POOL_GUID);
 547                                 fnvlist_add_uint64(config,
 548                                     ZPOOL_CONFIG_POOL_GUID, guid);
 549                                 name = fnvlist_lookup_string(tmp,
 550                                     ZPOOL_CONFIG_POOL_NAME);
 551                                 fnvlist_add_string(config,
 552                                     ZPOOL_CONFIG_POOL_NAME, name);
 553
 554                                 if (nvlist_lookup_string(tmp,
 555                                     ZPOOL_CONFIG_COMMENT, &comment) == 0)
 556                                         fnvlist_add_string(config,
 557                                             ZPOOL_CONFIG_COMMENT, comment);
 558
 559                                 state = fnvlist_lookup_uint64(tmp,
 560                                     ZPOOL_CONFIG_POOL_STATE);
 561                                 fnvlist_add_uint64(config,
 562                                     ZPOOL_CONFIG_POOL_STATE, state);
 563
 564                                 hostid = 0;
 565                                 if (nvlist_lookup_uint64(tmp,
 566                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 567                                         fnvlist_add_uint64(config,
 568                                             ZPOOL_CONFIG_HOSTID, hostid);
 569                                         hostname = fnvlist_lookup_string(tmp,
 570                                             ZPOOL_CONFIG_HOSTNAME);
 571                                         fnvlist_add_string(config,
 572                                             ZPOOL_CONFIG_HOSTNAME, hostname);
 573                                 }
 574
 575                                 config_seen = B_TRUE;
 576                         }
 577
 578                         /*
 579                          * Add this top-level vdev to the child array.
 580                          */
 581                         verify(nvlist_lookup_nvlist(tmp,
 582                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 583                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 584                             &id) == 0);
 585
 586                         if (id >= children) {
 587                                 nvlist_t **newchild;
 588
 589                                 newchild = zfs_alloc(hdl, (id + 1) *
 590                                     sizeof (nvlist_t *));
 591                                 if (newchild == NULL)
 592                                         goto nomem;
 593
 594                                 for (c = 0; c < children; c++)
 595                                         newchild[c] = child[c];
 596
 597                                 free(child);
 598                                 child = newchild;
 599                                 children = id + 1;
 600                         }
 601                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 602                                 goto nomem;
 603
 604                 }
 605
 606                 /*
 607                  * If we have information about all the top-levels then
 608                  * clean up the nvlist which we've constructed. This
 609                  * means removing any extraneous devices that are
 610                  * beyond the valid range or adding devices to the end
 611                  * of our array which appear to be missing.
 612                  */
 613                 if (valid_top_config) {
 614                         if (max_id < children) {
 615                                 for (c = max_id; c < children; c++)
 616                                         nvlist_free(child[c]);
 617                                 children = max_id;
 618                         } else if (max_id > children) {
 619                                 nvlist_t **newchild;
 620
 621                                 newchild = zfs_alloc(hdl, (max_id) *
 622                                     sizeof (nvlist_t *));
 623                                 if (newchild == NULL)
 624                                         goto nomem;
 625
 626                                 for (c = 0; c < children; c++)
 627                                         newchild[c] = child[c];
 628
 629                                 free(child);
 630                                 child = newchild;
 631                                 children = max_id;
 632                         }
 633                 }
 634
 635                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 636                     &guid) == 0);
 637
 638                 /*
 639                  * The vdev namespace may contain holes as a result of
 640                  * device removal. We must add them back into the vdev
 641                  * tree before we process any missing devices.
 642                  */
 643                 if (holes > 0) {
 644                         ASSERT(valid_top_config);
 645
 646                         for (c = 0; c < children; c++) {
 647                                 nvlist_t *holey;
 648
 649                                 if (child[c] != NULL ||
 650                                     !vdev_is_hole(hole_array, holes, c))
 651                                         continue;
 652
 653                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 654                                     0) != 0)
 655                                         goto nomem;
 656
 657                                 /*
 658                                  * Holes in the namespace are treated as
 659                                  * "hole" top-level vdevs and have a
 660                                  * special flag set on them.
 661                                  */
 662                                 if (nvlist_add_string(holey,
 663                                     ZPOOL_CONFIG_TYPE,
 664                                     VDEV_TYPE_HOLE) != 0 ||
 665                                     nvlist_add_uint64(holey,
 666                                     ZPOOL_CONFIG_ID, c) != 0 ||
 667                                     nvlist_add_uint64(holey,
 668                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 669                                         nvlist_free(holey);
 670                                         goto nomem;
 671                                 }
 672                                 child[c] = holey;
 673                         }
 674                 }
 675
 676                 /*
 677                  * Look for any missing top-level vdevs.  If this is the case,
 678                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 679                  * simply compress the child array, because the kernel performs
 680                  * certain checks to make sure the vdev IDs match their location
 681                  * in the configuration.
 682                  */
 683                 for (c = 0; c < children; c++) {
 684                         if (child[c] == NULL) {
 685                                 nvlist_t *missing;
 686                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 687                                     0) != 0)
 688                                         goto nomem;
 689                                 if (nvlist_add_string(missing,
 690                                     ZPOOL_CONFIG_TYPE,
 691                                     VDEV_TYPE_MISSING) != 0 ||
 692                                     nvlist_add_uint64(missing,
 693                                     ZPOOL_CONFIG_ID, c) != 0 ||
 694                                     nvlist_add_uint64(missing,
 695                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 696                                         nvlist_free(missing);
 697                                         goto nomem;
 698                                 }
 699                                 child[c] = missing;
 700                         }
 701                 }
 702
 703                 /*
 704                  * Put all of this pool's top-level vdevs into a root vdev.
 705                  */
 706                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 707                         goto nomem;
 708                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 709                     VDEV_TYPE_ROOT) != 0 ||
 710                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 711                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 712                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 713                     child, children) != 0) {
 714                         nvlist_free(nvroot);
 715                         goto nomem;
 716                 }
 717
 718                 for (c = 0; c < children; c++)
 719                         nvlist_free(child[c]);
 720                 free(child);
 721                 children = 0;
 722                 child = NULL;
 723
 724                 /*
 725                  * Go through and fix up any paths and/or devids based on our
 726                  * known list of vdev GUID -> path mappings.
 727                  */
 728                 if (fix_paths(nvroot, pl->names) != 0) {
 729                         nvlist_free(nvroot);
 730                         goto nomem;
 731                 }
 732
 733                 /*
 734                  * Add the root vdev to this pool's configuration.
 735                  */
 736                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 737                     nvroot) != 0) {
 738                         nvlist_free(nvroot);
 739                         goto nomem;
 740                 }
 741                 nvlist_free(nvroot);
 742
 743                 /*
 744                  * zdb uses this path to report on active pools that were
 745                  * imported or created using -R.
 746                  */
 747                 if (active_ok)
 748                         goto add_pool;
 749
 750                 /*
 751                  * Determine if this pool is currently active, in which case we
 752                  * can't actually import it.
 753                  */
 754                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 755                     &name) == 0);
 756                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 757                     &guid) == 0);
 758
 759                 if (pool_active(hdl, name, guid, &isactive) != 0)
 760                         goto error;
 761
 762                 if (isactive) {
 763                         nvlist_free(config);
 764                         config = NULL;
 765                         continue;
 766                 }
 767
 768                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 769                         nvlist_free(config);
 770                         config = NULL;
 771                         continue;
 772                 }
 773
 774                 nvlist_free(config);
 775                 config = nvl;
 776
 777                 /*
 778                  * Go through and update the paths for spares, now that we have
 779                  * them.
 780                  */
 781                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 782                     &nvroot) == 0);
 783                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 784                     &spares, &nspares) == 0) {
 785                         for (i = 0; i < nspares; i++) {
 786                                 if (fix_paths(spares[i], pl->names) != 0)
 787                                         goto nomem;
 788                         }
 789                 }
 790
 791                 /*
 792                  * Update the paths for l2cache devices.
 793                  */
 794                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 795                     &l2cache, &nl2cache) == 0) {
 796                         for (i = 0; i < nl2cache; i++) {
 797                                 if (fix_paths(l2cache[i], pl->names) != 0)
 798                                         goto nomem;
 799                         }
 800                 }
 801
 802                 /*
 803                  * Restore the original information read from the actual label.
 804                  */
 805                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 806                     DATA_TYPE_UINT64);
 807                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 808                     DATA_TYPE_STRING);
 809                 if (hostid != 0) {
 810                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 811                             hostid) == 0);
 812                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 813                             hostname) == 0);
 814                 }
 815
 816 add_pool:
 817                 /*
 818                  * Add this pool to the list of configs.
 819                  */
 820                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 821                     &name) == 0);
 822                 if (nvlist_add_nvlist(ret, name, config) != 0)
 823                         goto nomem;
 824
 825                 found_one = B_TRUE;
 826                 nvlist_free(config);
 827                 config = NULL;
 828         }
 829
 830         if (!found_one) {
 831                 nvlist_free(ret);
 832                 ret = NULL;
 833         }
 834
 835         return (ret);
 836
 837 nomem:
 838         (void) no_memory(hdl);
 839 error:
 840         nvlist_free(config);
 841         nvlist_free(ret);
 842         for (c = 0; c < children; c++)
 843                 nvlist_free(child[c]);
 844         free(child);
 845
 846         return (NULL);
 847 }
 848
 849 /*
 850  * Return the offset of the given label.
 851  */
 852 static uint64_t
 853 label_offset(uint64_t size, int l)
 854 {
 855         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 856         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 857             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 858 }
 859
 860 /*
 861  * Given a file descriptor, read the label information and return an nvlist
 862  * describing the configuration, if there is one.
 863  */
 864 int
 865 zpool_read_label(int fd, nvlist_t **config)
 866 {
 867         struct stat64 statbuf;
 868         int l;
 869         vdev_label_t *label;
 870         uint64_t state, txg, size;
 871
 872         *config = NULL;
 873
 874         if (fstat64(fd, &statbuf) == -1)
 875                 return (0);
 876         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 877
 878         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 879                 return (-1);
 880
 881         for (l = 0; l < VDEV_LABELS; l++) {
 882                 if (pread64(fd, label, sizeof (vdev_label_t),
 883                     label_offset(size, l)) != sizeof (vdev_label_t))
 884                         continue;
 885
 886                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 887                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 888                         continue;
 889
 890                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 891                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 892                         nvlist_free(*config);
 893                         continue;
 894                 }
 895
 896                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 897                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 898                     &txg) != 0 || txg == 0)) {
 899                         nvlist_free(*config);
 900                         continue;
 901                 }
 902
 903                 free(label);
 904                 return (0);
 905         }
 906
 907         free(label);
 908         *config = NULL;
 909         return (0);
 910 }
 911
 912 typedef struct rdsk_node {
 913         char *rn_name;
 914         int rn_dfd;
 915         libzfs_handle_t *rn_hdl;
 916         nvlist_t *rn_config;
 917         avl_tree_t *rn_avl;
 918         avl_node_t rn_node;
 919         boolean_t rn_nozpool;
 920 } rdsk_node_t;
 921
 922 static int
 923 slice_cache_compare(const void *arg1, const void *arg2)
 924 {
 925         const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
 926         const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
 927         char *nm1slice, *nm2slice;
 928         int rv;
 929
 930         /*
 931          * slices zero and two are the most likely to provide results,
 932          * so put those first
 933          */
 934         nm1slice = strstr(nm1, "s0");
 935         nm2slice = strstr(nm2, "s0");
 936         if (nm1slice && !nm2slice) {
 937                 return (-1);
 938         }
 939         if (!nm1slice && nm2slice) {
 940                 return (1);
 941         }
 942         nm1slice = strstr(nm1, "s2");
 943         nm2slice = strstr(nm2, "s2");
 944         if (nm1slice && !nm2slice) {
 945                 return (-1);
 946         }
 947         if (!nm1slice && nm2slice) {
 948                 return (1);
 949         }
 950
 951         rv = strcmp(nm1, nm2);
 952         if (rv == 0)
 953                 return (0);
 954         return (rv > 0 ? 1 : -1);
 955 }
 956
 957 #ifdef sun
 958 static void
 959 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
 960     diskaddr_t size, uint_t blksz)
 961 {
 962         rdsk_node_t tmpnode;
 963         rdsk_node_t *node;
 964         char sname[MAXNAMELEN];
 965
 966         tmpnode.rn_name = &sname[0];
 967         (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
 968             diskname, partno);
 969         /*
 970          * protect against division by zero for disk labels that
 971          * contain a bogus sector size
 972          */
 973         if (blksz == 0)
 974                 blksz = DEV_BSIZE;
 975         /* too small to contain a zpool? */
 976         if ((size < (SPA_MINDEVSIZE / blksz)) &&
 977             (node = avl_find(r, &tmpnode, NULL)))
 978                 node->rn_nozpool = B_TRUE;
 979 }
 980 #endif  /* sun */
 981
 982 static void
 983 nozpool_all_slices(avl_tree_t *r, const char *sname)
 984 {
 985 #ifdef sun
 986         char diskname[MAXNAMELEN];
 987         char *ptr;
 988         int i;
 989
 990         (void) strncpy(diskname, sname, MAXNAMELEN);
 991         if (((ptr = strrchr(diskname, 's')) == NULL) &&
 992             ((ptr = strrchr(diskname, 'p')) == NULL))
 993                 return;
 994         ptr[0] = 's';
 995         ptr[1] = '\0';
 996         for (i = 0; i < NDKMAP; i++)
 997                 check_one_slice(r, diskname, i, 0, 1);
 998         ptr[0] = 'p';
 999         for (i = 0; i <= FD_NUMPART; i++)
1000                 check_one_slice(r, diskname, i, 0, 1);
1001 #endif  /* sun */
1002 }
1003
1004 #ifdef sun
1005 static void
1006 check_slices(avl_tree_t *r, int fd, const char *sname)
1007 {
1008         struct extvtoc vtoc;
1009         struct dk_gpt *gpt;
1010         char diskname[MAXNAMELEN];
1011         char *ptr;
1012         int i;
1013
1014         (void) strncpy(diskname, sname, MAXNAMELEN);
1015         if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1016                 return;
1017         ptr[1] = '\0';
1018
1019         if (read_extvtoc(fd, &vtoc) >= 0) {
1020                 for (i = 0; i < NDKMAP; i++)
1021                         check_one_slice(r, diskname, i,
1022                             vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1023         } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1024                 /*
1025                  * on x86 we'll still have leftover links that point
1026                  * to slices s[9-15], so use NDKMAP instead
1027                  */
1028                 for (i = 0; i < NDKMAP; i++)
1029                         check_one_slice(r, diskname, i,
1030                             gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1031                 /* nodes p[1-4] are never used with EFI labels */
1032                 ptr[0] = 'p';
1033                 for (i = 1; i <= FD_NUMPART; i++)
1034                         check_one_slice(r, diskname, i, 0, 1);
1035                 efi_free(gpt);
1036         }
1037 }
1038 #endif  /* sun */
1039
1040 static void
1041 zpool_open_func(void *arg)
1042 {
1043         rdsk_node_t *rn = arg;
1044         struct stat64 statbuf;
1045         nvlist_t *config;
1046         int fd;
1047
1048         if (rn->rn_nozpool)
1049                 return;
1050         if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1051                 /* symlink to a device that's no longer there */
1052                 if (errno == ENOENT)
1053                         nozpool_all_slices(rn->rn_avl, rn->rn_name);
1054                 return;
1055         }
1056         /*
1057          * Ignore failed stats.  We only want regular
1058          * files, character devs and block devs.
1059          */
1060         if (fstat64(fd, &statbuf) != 0 ||
1061             (!S_ISREG(statbuf.st_mode) &&
1062             !S_ISCHR(statbuf.st_mode) &&
1063             !S_ISBLK(statbuf.st_mode))) {
1064                 (void) close(fd);
1065                 return;
1066         }
1067         /* this file is too small to hold a zpool */
1068 #ifdef sun
1069         if (S_ISREG(statbuf.st_mode) &&
1070             statbuf.st_size < SPA_MINDEVSIZE) {
1071                 (void) close(fd);
1072                 return;
1073         } else if (!S_ISREG(statbuf.st_mode)) {
1074                 /*
1075                  * Try to read the disk label first so we don't have to
1076                  * open a bunch of minor nodes that can't have a zpool.
1077                  */
1078                 check_slices(rn->rn_avl, fd, rn->rn_name);
1079         }
1080 #else   /* !sun */
1081         if (statbuf.st_size < SPA_MINDEVSIZE) {
1082                 (void) close(fd);
1083                 return;
1084         }
1085 #endif  /* sun */
1086
1087         if ((zpool_read_label(fd, &config)) != 0) {
1088                 (void) close(fd);
1089                 (void) no_memory(rn->rn_hdl);
1090                 return;
1091         }
1092         (void) close(fd);
1093
1094         rn->rn_config = config;
1095 }
1096
1097 /*
1098  * Given a file descriptor, clear (zero) the label information.  This function
1099  * is used in the appliance stack as part of the ZFS sysevent module and
1100  * to implement the "zpool labelclear" command.
1101  */
1102 int
1103 zpool_clear_label(int fd)
1104 {
1105         struct stat64 statbuf;
1106         int l;
1107         vdev_label_t *label;
1108         uint64_t size;
1109
1110         if (fstat64(fd, &statbuf) == -1)
1111                 return (0);
1112         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1113
1114         if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1115                 return (-1);
1116
1117         for (l = 0; l < VDEV_LABELS; l++) {
1118                 if (pwrite64(fd, label, sizeof (vdev_label_t),
1119                     label_offset(size, l)) != sizeof (vdev_label_t)) {
1120                         free(label);
1121                         return (-1);
1122                 }
1123         }
1124
1125         free(label);
1126         return (0);
1127 }
1128
1129 /*
1130  * Given a list of directories to search, find all pools stored on disk.  This
1131  * includes partial pools which are not available to import.  If no args are
1132  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1133  * poolname or guid (but not both) are provided by the caller when trying
1134  * to import a specific pool.
1135  */
1136 static nvlist_t *
1137 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1138 {
1139         int i, dirs = iarg->paths;
1140         struct dirent64 *dp;
1141         char path[MAXPATHLEN];
1142         char *end, **dir = iarg->path;
1143         size_t pathleft;
1144         nvlist_t *ret = NULL;
1145         static char *default_dir = "/dev";
1146         pool_list_t pools = { 0 };
1147         pool_entry_t *pe, *penext;
1148         vdev_entry_t *ve, *venext;
1149         config_entry_t *ce, *cenext;
1150         name_entry_t *ne, *nenext;
1151         avl_tree_t slice_cache;
1152         rdsk_node_t *slice;
1153         void *cookie;
1154
1155         if (dirs == 0) {
1156                 dirs = 1;
1157                 dir = &default_dir;
1158         }
1159
1160         /*
1161          * Go through and read the label configuration information from every
1162          * possible device, organizing the information according to pool GUID
1163          * and toplevel GUID.
1164          */
1165         for (i = 0; i < dirs; i++) {
1166                 tpool_t *t;
1167                 char *rdsk;
1168                 int dfd;
1169                 boolean_t config_failed = B_FALSE;
1170                 DIR *dirp;
1171
1172                 /* use realpath to normalize the path */
1173                 if (realpath(dir[i], path) == 0) {
1174                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1175                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1176                         goto error;
1177                 }
1178                 end = &path[strlen(path)];
1179                 *end++ = '/';
1180                 *end = 0;
1181                 pathleft = &path[sizeof (path)] - end;
1182
1183                 /*
1184                  * Using raw devices instead of block devices when we're
1185                  * reading the labels skips a bunch of slow operations during
1186                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1187                  */
1188                 if (strcmp(path, "/dev/dsk/") == 0)
1189                         rdsk = "/dev/";
1190                 else
1191                         rdsk = path;
1192
1193                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1194                     (dirp = fdopendir(dfd)) == NULL) {
1195                         if (dfd >= 0)
1196                                 (void) close(dfd);
1197                         zfs_error_aux(hdl, strerror(errno));
1198                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1199                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1200                             rdsk);
1201                         goto error;
1202                 }
1203
1204                 avl_create(&slice_cache, slice_cache_compare,
1205                     sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1206
1207                 if (strcmp(rdsk, "/dev/") == 0) {
1208                         struct gmesh mesh;
1209                         struct gclass *mp;
1210                         struct ggeom *gp;
1211                         struct gprovider *pp;
1212
1213                         errno = geom_gettree(&mesh);
1214                         if (errno != 0) {
1215                                 zfs_error_aux(hdl, strerror(errno));
1216                                 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1217                                     dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1218                                 goto error;
1219                         }
1220
1221                         LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1222                                 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1223                                         LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1224                                                 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1225                                                 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1226                                                 slice->rn_avl = &slice_cache;
1227                                                 slice->rn_dfd = dfd;
1228                                                 slice->rn_hdl = hdl;
1229                                                 slice->rn_nozpool = B_FALSE;
1230                                                 avl_add(&slice_cache, slice);
1231                                         }
1232                                 }
1233                         }
1234
1235                         geom_deletetree(&mesh);
1236                         goto skipdir;
1237                 }
1238
1239                 /*
1240                  * This is not MT-safe, but we have no MT consumers of libzfs
1241                  */
1242                 while ((dp = readdir64(dirp)) != NULL) {
1243                         const char *name = dp->d_name;
1244                         if (name[0] == '.' &&
1245                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1246                                 continue;
1247
1248                         slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1249                         slice->rn_name = zfs_strdup(hdl, name);
1250                         slice->rn_avl = &slice_cache;
1251                         slice->rn_dfd = dfd;
1252                         slice->rn_hdl = hdl;
1253                         slice->rn_nozpool = B_FALSE;
1254                         avl_add(&slice_cache, slice);
1255                 }
1256 skipdir:
1257                 /*
1258                  * create a thread pool to do all of this in parallel;
1259                  * rn_nozpool is not protected, so this is racy in that
1260                  * multiple tasks could decide that the same slice can
1261                  * not hold a zpool, which is benign.  Also choose
1262                  * double the number of processors; we hold a lot of
1263                  * locks in the kernel, so going beyond this doesn't
1264                  * buy us much.
1265                  */
1266                 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1267                     0, NULL);
1268                 for (slice = avl_first(&slice_cache); slice;
1269                     (slice = avl_walk(&slice_cache, slice,
1270                     AVL_AFTER)))
1271                         (void) tpool_dispatch(t, zpool_open_func, slice);
1272                 tpool_wait(t);
1273                 tpool_destroy(t);
1274
1275                 cookie = NULL;
1276                 while ((slice = avl_destroy_nodes(&slice_cache,
1277                     &cookie)) != NULL) {
1278                         if (slice->rn_config != NULL && !config_failed) {
1279                                 nvlist_t *config = slice->rn_config;
1280                                 boolean_t matched = B_TRUE;
1281
1282                                 if (iarg->poolname != NULL) {
1283                                         char *pname;
1284
1285                                         matched = nvlist_lookup_string(config,
1286                                             ZPOOL_CONFIG_POOL_NAME,
1287                                             &pname) == 0 &&
1288                                             strcmp(iarg->poolname, pname) == 0;
1289                                 } else if (iarg->guid != 0) {
1290                                         uint64_t this_guid;
1291
1292                                         matched = nvlist_lookup_uint64(config,
1293                                             ZPOOL_CONFIG_POOL_GUID,
1294                                             &this_guid) == 0 &&
1295                                             iarg->guid == this_guid;
1296                                 }
1297                                 if (!matched) {
1298                                         nvlist_free(config);
1299                                 } else {
1300                                         /*
1301                                          * use the non-raw path for the config
1302                                          */
1303                                         (void) strlcpy(end, slice->rn_name,
1304                                             pathleft);
1305                                         if (add_config(hdl, &pools, path,
1306                                             config) != 0)
1307                                                 config_failed = B_TRUE;
1308                                 }
1309                         }
1310                         free(slice->rn_name);
1311                         free(slice);
1312                 }
1313                 avl_destroy(&slice_cache);
1314
1315                 (void) closedir(dirp);
1316
1317                 if (config_failed)
1318                         goto error;
1319         }
1320
1321         ret = get_configs(hdl, &pools, iarg->can_be_active);
1322
1323 error:
1324         for (pe = pools.pools; pe != NULL; pe = penext) {
1325                 penext = pe->pe_next;
1326                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1327                         venext = ve->ve_next;
1328                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1329                                 cenext = ce->ce_next;
1330                                 if (ce->ce_config)
1331                                         nvlist_free(ce->ce_config);
1332                                 free(ce);
1333                         }
1334                         free(ve);
1335                 }
1336                 free(pe);
1337         }
1338
1339         for (ne = pools.names; ne != NULL; ne = nenext) {
1340                 nenext = ne->ne_next;
1341                 free(ne->ne_name);
1342                 free(ne);
1343         }
1344
1345         return (ret);
1346 }
1347
1348 nvlist_t *
1349 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1350 {
1351         importargs_t iarg = { 0 };
1352
1353         iarg.paths = argc;
1354         iarg.path = argv;
1355
1356         return (zpool_find_import_impl(hdl, &iarg));
1357 }
1358
1359 /*
1360  * Given a cache file, return the contents as a list of importable pools.
1361  * poolname or guid (but not both) are provided by the caller when trying
1362  * to import a specific pool.
1363  */
1364 nvlist_t *
1365 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1366     char *poolname, uint64_t guid)
1367 {
1368         char *buf;
1369         int fd;
1370         struct stat64 statbuf;
1371         nvlist_t *raw, *src, *dst;
1372         nvlist_t *pools;
1373         nvpair_t *elem;
1374         char *name;
1375         uint64_t this_guid;
1376         boolean_t active;
1377
1378         verify(poolname == NULL || guid == 0);
1379
1380         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1381                 zfs_error_aux(hdl, "%s", strerror(errno));
1382                 (void) zfs_error(hdl, EZFS_BADCACHE,
1383                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1384                 return (NULL);
1385         }
1386
1387         if (fstat64(fd, &statbuf) != 0) {
1388                 zfs_error_aux(hdl, "%s", strerror(errno));
1389                 (void) close(fd);
1390                 (void) zfs_error(hdl, EZFS_BADCACHE,
1391                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1392                 return (NULL);
1393         }
1394
1395         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1396                 (void) close(fd);
1397                 return (NULL);
1398         }
1399
1400         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1401                 (void) close(fd);
1402                 free(buf);
1403                 (void) zfs_error(hdl, EZFS_BADCACHE,
1404                     dgettext(TEXT_DOMAIN,
1405                     "failed to read cache file contents"));
1406                 return (NULL);
1407         }
1408
1409         (void) close(fd);
1410
1411         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1412                 free(buf);
1413                 (void) zfs_error(hdl, EZFS_BADCACHE,
1414                     dgettext(TEXT_DOMAIN,
1415                     "invalid or corrupt cache file contents"));
1416                 return (NULL);
1417         }
1418
1419         free(buf);
1420
1421         /*
1422          * Go through and get the current state of the pools and refresh their
1423          * state.
1424          */
1425         if (nvlist_alloc(&pools, 0, 0) != 0) {
1426                 (void) no_memory(hdl);
1427                 nvlist_free(raw);
1428                 return (NULL);
1429         }
1430
1431         elem = NULL;
1432         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1433                 verify(nvpair_value_nvlist(elem, &src) == 0);
1434
1435                 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1436                     &name) == 0);
1437                 if (poolname != NULL && strcmp(poolname, name) != 0)
1438                         continue;
1439
1440                 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1441                     &this_guid) == 0);
1442                 if (guid != 0) {
1443                         verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1444                             &this_guid) == 0);
1445                         if (guid != this_guid)
1446                                 continue;
1447                 }
1448
1449                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1450                         nvlist_free(raw);
1451                         nvlist_free(pools);
1452                         return (NULL);
1453                 }
1454
1455                 if (active)
1456                         continue;
1457
1458                 if ((dst = refresh_config(hdl, src)) == NULL) {
1459                         nvlist_free(raw);
1460                         nvlist_free(pools);
1461                         return (NULL);
1462                 }
1463
1464                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1465                         (void) no_memory(hdl);
1466                         nvlist_free(dst);
1467                         nvlist_free(raw);
1468                         nvlist_free(pools);
1469                         return (NULL);
1470                 }
1471                 nvlist_free(dst);
1472         }
1473
1474         nvlist_free(raw);
1475         return (pools);
1476 }
1477
1478 static int
1479 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1480 {
1481         importargs_t *import = data;
1482         int found = 0;
1483
1484         if (import->poolname != NULL) {
1485                 char *pool_name;
1486
1487                 verify(nvlist_lookup_string(zhp->zpool_config,
1488                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1489                 if (strcmp(pool_name, import->poolname) == 0)
1490                         found = 1;
1491         } else {
1492                 uint64_t pool_guid;
1493
1494                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1495                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1496                 if (pool_guid == import->guid)
1497                         found = 1;
1498         }
1499
1500         zpool_close(zhp);
1501         return (found);
1502 }
1503
1504 nvlist_t *
1505 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1506 {
1507         verify(import->poolname == NULL || import->guid == 0);
1508
1509         if (import->unique)
1510                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1511
1512         if (import->cachefile != NULL)
1513                 return (zpool_find_import_cached(hdl, import->cachefile,
1514                     import->poolname, import->guid));
1515
1516         return (zpool_find_import_impl(hdl, import));
1517 }
1518
1519 boolean_t
1520 find_guid(nvlist_t *nv, uint64_t guid)
1521 {
1522         uint64_t tmp;
1523         nvlist_t **child;
1524         uint_t c, children;
1525
1526         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1527         if (tmp == guid)
1528                 return (B_TRUE);
1529
1530         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1531             &child, &children) == 0) {
1532                 for (c = 0; c < children; c++)
1533                         if (find_guid(child[c], guid))
1534                                 return (B_TRUE);
1535         }
1536
1537         return (B_FALSE);
1538 }
1539
1540 typedef struct aux_cbdata {
1541         const char      *cb_type;
1542         uint64_t        cb_guid;
1543         zpool_handle_t  *cb_zhp;
1544 } aux_cbdata_t;
1545
1546 static int
1547 find_aux(zpool_handle_t *zhp, void *data)
1548 {
1549         aux_cbdata_t *cbp = data;
1550         nvlist_t **list;
1551         uint_t i, count;
1552         uint64_t guid;
1553         nvlist_t *nvroot;
1554
1555         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1556             &nvroot) == 0);
1557
1558         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1559             &list, &count) == 0) {
1560                 for (i = 0; i < count; i++) {
1561                         verify(nvlist_lookup_uint64(list[i],
1562                             ZPOOL_CONFIG_GUID, &guid) == 0);
1563                         if (guid == cbp->cb_guid) {
1564                                 cbp->cb_zhp = zhp;
1565                                 return (1);
1566                         }
1567                 }
1568         }
1569
1570         zpool_close(zhp);
1571         return (0);
1572 }
1573
1574 /*
1575  * Determines if the pool is in use.  If so, it returns true and the state of
1576  * the pool as well as the name of the pool.  Both strings are allocated and
1577  * must be freed by the caller.
1578  */
1579 int
1580 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1581     boolean_t *inuse)
1582 {
1583         nvlist_t *config;
1584         char *name;
1585         boolean_t ret;
1586         uint64_t guid, vdev_guid;
1587         zpool_handle_t *zhp;
1588         nvlist_t *pool_config;
1589         uint64_t stateval, isspare;
1590         aux_cbdata_t cb = { 0 };
1591         boolean_t isactive;
1592
1593         *inuse = B_FALSE;
1594
1595         if (zpool_read_label(fd, &config) != 0) {
1596                 (void) no_memory(hdl);
1597                 return (-1);
1598         }
1599
1600         if (config == NULL)
1601                 return (0);
1602
1603         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1604             &stateval) == 0);
1605         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1606             &vdev_guid) == 0);
1607
1608         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1609                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1610                     &name) == 0);
1611                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1612                     &guid) == 0);
1613         }
1614
1615         switch (stateval) {
1616         case POOL_STATE_EXPORTED:
1617                 /*
1618                  * A pool with an exported state may in fact be imported
1619                  * read-only, so check the in-core state to see if it's
1620                  * active and imported read-only.  If it is, set
1621                  * its state to active.
1622                  */
1623                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1624                     (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1625                         if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1626                                 stateval = POOL_STATE_ACTIVE;
1627
1628                         /*
1629                          * All we needed the zpool handle for is the
1630                          * readonly prop check.
1631                          */
1632                         zpool_close(zhp);
1633                 }
1634
1635                 ret = B_TRUE;
1636                 break;
1637
1638         case POOL_STATE_ACTIVE:
1639                 /*
1640                  * For an active pool, we have to determine if it's really part
1641                  * of a currently active pool (in which case the pool will exist
1642                  * and the guid will be the same), or whether it's part of an
1643                  * active pool that was disconnected without being explicitly
1644                  * exported.
1645                  */
1646                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1647                         nvlist_free(config);
1648                         return (-1);
1649                 }
1650
1651                 if (isactive) {
1652                         /*
1653                          * Because the device may have been removed while
1654                          * offlined, we only report it as active if the vdev is
1655                          * still present in the config.  Otherwise, pretend like
1656                          * it's not in use.
1657                          */
1658                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1659                             (pool_config = zpool_get_config(zhp, NULL))
1660                             != NULL) {
1661                                 nvlist_t *nvroot;
1662
1663                                 verify(nvlist_lookup_nvlist(pool_config,
1664                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1665                                 ret = find_guid(nvroot, vdev_guid);
1666                         } else {
1667                                 ret = B_FALSE;
1668                         }
1669
1670                         /*
1671                          * If this is an active spare within another pool, we
1672                          * treat it like an unused hot spare.  This allows the
1673                          * user to create a pool with a hot spare that currently
1674                          * in use within another pool.  Since we return B_TRUE,
1675                          * libdiskmgt will continue to prevent generic consumers
1676                          * from using the device.
1677                          */
1678                         if (ret && nvlist_lookup_uint64(config,
1679                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1680                                 stateval = POOL_STATE_SPARE;
1681
1682                         if (zhp != NULL)
1683                                 zpool_close(zhp);
1684                 } else {
1685                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1686                         ret = B_TRUE;
1687                 }
1688                 break;
1689
1690         case POOL_STATE_SPARE:
1691                 /*
1692                  * For a hot spare, it can be either definitively in use, or
1693                  * potentially active.  To determine if it's in use, we iterate
1694                  * over all pools in the system and search for one with a spare
1695                  * with a matching guid.
1696                  *
1697                  * Due to the shared nature of spares, we don't actually report
1698                  * the potentially active case as in use.  This means the user
1699                  * can freely create pools on the hot spares of exported pools,
1700                  * but to do otherwise makes the resulting code complicated, and
1701                  * we end up having to deal with this case anyway.
1702                  */
1703                 cb.cb_zhp = NULL;
1704                 cb.cb_guid = vdev_guid;
1705                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1706                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1707                         name = (char *)zpool_get_name(cb.cb_zhp);
1708                         ret = B_TRUE;
1709                 } else {
1710                         ret = B_FALSE;
1711                 }
1712                 break;
1713
1714         case POOL_STATE_L2CACHE:
1715
1716                 /*
1717                  * Check if any pool is currently using this l2cache device.
1718                  */
1719                 cb.cb_zhp = NULL;
1720                 cb.cb_guid = vdev_guid;
1721                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1722                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1723                         name = (char *)zpool_get_name(cb.cb_zhp);
1724                         ret = B_TRUE;
1725                 } else {
1726                         ret = B_FALSE;
1727                 }
1728                 break;
1729
1730         default:
1731                 ret = B_FALSE;
1732         }
1733
1734
1735         if (ret) {
1736                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1737                         if (cb.cb_zhp)
1738                                 zpool_close(cb.cb_zhp);
1739                         nvlist_free(config);
1740                         return (-1);
1741                 }
1742                 *state = (pool_state_t)stateval;
1743         }
1744
1745         if (cb.cb_zhp)
1746                 zpool_close(cb.cb_zhp);
1747
1748         nvlist_free(config);
1749         *inuse = ret;
1750         return (0);
1751 }