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) 2013 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 #ifdef have_devid
 100         int fd;
 101         ddi_devid_t devid;
 102         char *minor, *ret;
 103
 104         if ((fd = open(path, O_RDONLY)) < 0)
 105                 return (NULL);
 106
 107         minor = NULL;
 108         ret = NULL;
 109         if (devid_get(fd, &devid) == 0) {
 110                 if (devid_get_minor_name(fd, &minor) == 0)
 111                         ret = devid_str_encode(devid, minor);
 112                 if (minor != NULL)
 113                         devid_str_free(minor);
 114                 devid_free(devid);
 115         }
 116         (void) close(fd);
 117
 118         return (ret);
 119 #else
 120         return (NULL);
 121 #endif
 122 }
 123
 124
 125 /*
 126  * Go through and fix up any path and/or devid information for the given vdev
 127  * configuration.
 128  */
 129 static int
 130 fix_paths(nvlist_t *nv, name_entry_t *names)
 131 {
 132         nvlist_t **child;
 133         uint_t c, children;
 134         uint64_t guid;
 135         name_entry_t *ne, *best;
 136         char *path, *devid;
 137         int matched;
 138
 139         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 140             &child, &children) == 0) {
 141                 for (c = 0; c < children; c++)
 142                         if (fix_paths(child[c], names) != 0)
 143                                 return (-1);
 144                 return (0);
 145         }
 146
 147         /*
 148          * This is a leaf (file or disk) vdev.  In either case, go through
 149          * the name list and see if we find a matching guid.  If so, replace
 150          * the path and see if we can calculate a new devid.
 151          *
 152          * There may be multiple names associated with a particular guid, in
 153          * which case we have overlapping slices or multiple paths to the same
 154          * disk.  If this is the case, then we want to pick the path that is
 155          * the most similar to the original, where "most similar" is the number
 156          * of matching characters starting from the end of the path.  This will
 157          * preserve slice numbers even if the disks have been reorganized, and
 158          * will also catch preferred disk names if multiple paths exist.
 159          */
 160         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 161         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 162                 path = NULL;
 163
 164         matched = 0;
 165         best = NULL;
 166         for (ne = names; ne != NULL; ne = ne->ne_next) {
 167                 if (ne->ne_guid == guid) {
 168                         const char *src, *dst;
 169                         int count;
 170
 171                         if (path == NULL) {
 172                                 best = ne;
 173                                 break;
 174                         }
 175
 176                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 177                         dst = path + strlen(path) - 1;
 178                         for (count = 0; src >= ne->ne_name && dst >= path;
 179                             src--, dst--, count++)
 180                                 if (*src != *dst)
 181                                         break;
 182
 183                         /*
 184                          * At this point, 'count' is the number of characters
 185                          * matched from the end.
 186                          */
 187                         if (count > matched || best == NULL) {
 188                                 best = ne;
 189                                 matched = count;
 190                         }
 191                 }
 192         }
 193
 194         if (best == NULL)
 195                 return (0);
 196
 197         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 198                 return (-1);
 199
 200         if ((devid = get_devid(best->ne_name)) == NULL) {
 201                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 202         } else {
 203                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
 204                         devid_str_free(devid);
 205                         return (-1);
 206                 }
 207                 devid_str_free(devid);
 208         }
 209
 210         return (0);
 211 }
 212
 213 /*
 214  * Add the given configuration to the list of known devices.
 215  */
 216 static int
 217 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 218     nvlist_t *config)
 219 {
 220         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 221         pool_entry_t *pe;
 222         vdev_entry_t *ve;
 223         config_entry_t *ce;
 224         name_entry_t *ne;
 225
 226         /*
 227          * If this is a hot spare not currently in use or level 2 cache
 228          * device, add it to the list of names to translate, but don't do
 229          * anything else.
 230          */
 231         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 232             &state) == 0 &&
 233             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 234             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 235                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 236                         return (-1);
 237
 238                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 239                         free(ne);
 240                         return (-1);
 241                 }
 242                 ne->ne_guid = vdev_guid;
 243                 ne->ne_next = pl->names;
 244                 pl->names = ne;
 245                 return (0);
 246         }
 247
 248         /*
 249          * If we have a valid config but cannot read any of these fields, then
 250          * it means we have a half-initialized label.  In vdev_label_init()
 251          * we write a label with txg == 0 so that we can identify the device
 252          * in case the user refers to the same disk later on.  If we fail to
 253          * create the pool, we'll be left with a label in this state
 254          * which should not be considered part of a valid pool.
 255          */
 256         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 257             &pool_guid) != 0 ||
 258             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 259             &vdev_guid) != 0 ||
 260             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 261             &top_guid) != 0 ||
 262             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 263             &txg) != 0 || txg == 0) {
 264                 nvlist_free(config);
 265                 return (0);
 266         }
 267
 268         /*
 269          * First, see if we know about this pool.  If not, then add it to the
 270          * list of known pools.
 271          */
 272         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 273                 if (pe->pe_guid == pool_guid)
 274                         break;
 275         }
 276
 277         if (pe == NULL) {
 278                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 279                         nvlist_free(config);
 280                         return (-1);
 281                 }
 282                 pe->pe_guid = pool_guid;
 283                 pe->pe_next = pl->pools;
 284                 pl->pools = pe;
 285         }
 286
 287         /*
 288          * Second, see if we know about this toplevel vdev.  Add it if its
 289          * missing.
 290          */
 291         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 292                 if (ve->ve_guid == top_guid)
 293                         break;
 294         }
 295
 296         if (ve == NULL) {
 297                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 298                         nvlist_free(config);
 299                         return (-1);
 300                 }
 301                 ve->ve_guid = top_guid;
 302                 ve->ve_next = pe->pe_vdevs;
 303                 pe->pe_vdevs = ve;
 304         }
 305
 306         /*
 307          * Third, see if we have a config with a matching transaction group.  If
 308          * so, then we do nothing.  Otherwise, add it to the list of known
 309          * configs.
 310          */
 311         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 312                 if (ce->ce_txg == txg)
 313                         break;
 314         }
 315
 316         if (ce == NULL) {
 317                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 318                         nvlist_free(config);
 319                         return (-1);
 320                 }
 321                 ce->ce_txg = txg;
 322                 ce->ce_config = config;
 323                 ce->ce_next = ve->ve_configs;
 324                 ve->ve_configs = ce;
 325         } else {
 326                 nvlist_free(config);
 327         }
 328
 329         /*
 330          * At this point we've successfully added our config to the list of
 331          * known configs.  The last thing to do is add the vdev guid -> path
 332          * mappings so that we can fix up the configuration as necessary before
 333          * doing the import.
 334          */
 335         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 336                 return (-1);
 337
 338         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 339                 free(ne);
 340                 return (-1);
 341         }
 342
 343         ne->ne_guid = vdev_guid;
 344         ne->ne_next = pl->names;
 345         pl->names = ne;
 346
 347         return (0);
 348 }
 349
 350 /*
 351  * Returns true if the named pool matches the given GUID.
 352  */
 353 static int
 354 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 355     boolean_t *isactive)
 356 {
 357         zpool_handle_t *zhp;
 358         uint64_t theguid;
 359
 360         if (zpool_open_silent(hdl, name, &zhp) != 0)
 361                 return (-1);
 362
 363         if (zhp == NULL) {
 364                 *isactive = B_FALSE;
 365                 return (0);
 366         }
 367
 368         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 369             &theguid) == 0);
 370
 371         zpool_close(zhp);
 372
 373         *isactive = (theguid == guid);
 374         return (0);
 375 }
 376
 377 static nvlist_t *
 378 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 379 {
 380         nvlist_t *nvl;
 381         zfs_cmd_t zc = { 0 };
 382         int err;
 383
 384         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 385                 return (NULL);
 386
 387         if (zcmd_alloc_dst_nvlist(hdl, &zc,
 388             zc.zc_nvlist_conf_size * 2) != 0) {
 389                 zcmd_free_nvlists(&zc);
 390                 return (NULL);
 391         }
 392
 393         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 394             &zc)) != 0 && errno == ENOMEM) {
 395                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 396                         zcmd_free_nvlists(&zc);
 397                         return (NULL);
 398                 }
 399         }
 400
 401         if (err) {
 402                 zcmd_free_nvlists(&zc);
 403                 return (NULL);
 404         }
 405
 406         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 407                 zcmd_free_nvlists(&zc);
 408                 return (NULL);
 409         }
 410
 411         zcmd_free_nvlists(&zc);
 412         return (nvl);
 413 }
 414
 415 /*
 416  * Determine if the vdev id is a hole in the namespace.
 417  */
 418 boolean_t
 419 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
 420 {
 421         for (int c = 0; c < holes; c++) {
 422
 423                 /* Top-level is a hole */
 424                 if (hole_array[c] == id)
 425                         return (B_TRUE);
 426         }
 427         return (B_FALSE);
 428 }
 429
 430 /*
 431  * Convert our list of pools into the definitive set of configurations.  We
 432  * start by picking the best config for each toplevel vdev.  Once that's done,
 433  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 434  * pass to fix up any incorrect paths, and then add it to the main list to
 435  * return to the user.
 436  */
 437 static nvlist_t *
 438 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 439 {
 440         pool_entry_t *pe;
 441         vdev_entry_t *ve;
 442         config_entry_t *ce;
 443         nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
 444         nvlist_t **spares, **l2cache;
 445         uint_t i, nspares, nl2cache;
 446         boolean_t config_seen;
 447         uint64_t best_txg;
 448         char *name, *hostname;
 449         uint64_t guid;
 450         uint_t children = 0;
 451         nvlist_t **child = NULL;
 452         uint_t holes;
 453         uint64_t *hole_array, max_id;
 454         uint_t c;
 455         boolean_t isactive;
 456         uint64_t hostid;
 457         nvlist_t *nvl;
 458         boolean_t found_one = B_FALSE;
 459         boolean_t valid_top_config = B_FALSE;
 460
 461         if (nvlist_alloc(&ret, 0, 0) != 0)
 462                 goto nomem;
 463
 464         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 465                 uint64_t id, max_txg = 0;
 466
 467                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 468                         goto nomem;
 469                 config_seen = B_FALSE;
 470
 471                 /*
 472                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 473                  * from the first one we find, and then go through the rest and
 474                  * add them as necessary to the 'vdevs' member of the config.
 475                  */
 476                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 477
 478                         /*
 479                          * Determine the best configuration for this vdev by
 480                          * selecting the config with the latest transaction
 481                          * group.
 482                          */
 483                         best_txg = 0;
 484                         for (ce = ve->ve_configs; ce != NULL;
 485                             ce = ce->ce_next) {
 486
 487                                 if (ce->ce_txg > best_txg) {
 488                                         tmp = ce->ce_config;
 489                                         best_txg = ce->ce_txg;
 490                                 }
 491                         }
 492
 493                         /*
 494                          * We rely on the fact that the max txg for the
 495                          * pool will contain the most up-to-date information
 496                          * about the valid top-levels in the vdev namespace.
 497                          */
 498                         if (best_txg > max_txg) {
 499                                 (void) nvlist_remove(config,
 500                                     ZPOOL_CONFIG_VDEV_CHILDREN,
 501                                     DATA_TYPE_UINT64);
 502                                 (void) nvlist_remove(config,
 503                                     ZPOOL_CONFIG_HOLE_ARRAY,
 504                                     DATA_TYPE_UINT64_ARRAY);
 505
 506                                 max_txg = best_txg;
 507                                 hole_array = NULL;
 508                                 holes = 0;
 509                                 max_id = 0;
 510                                 valid_top_config = B_FALSE;
 511
 512                                 if (nvlist_lookup_uint64(tmp,
 513                                     ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
 514                                         verify(nvlist_add_uint64(config,
 515                                             ZPOOL_CONFIG_VDEV_CHILDREN,
 516                                             max_id) == 0);
 517                                         valid_top_config = B_TRUE;
 518                                 }
 519
 520                                 if (nvlist_lookup_uint64_array(tmp,
 521                                     ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
 522                                     &holes) == 0) {
 523                                         verify(nvlist_add_uint64_array(config,
 524                                             ZPOOL_CONFIG_HOLE_ARRAY,
 525                                             hole_array, holes) == 0);
 526                                 }
 527                         }
 528
 529                         if (!config_seen) {
 530                                 /*
 531                                  * Copy the relevant pieces of data to the pool
 532                                  * configuration:
 533                                  *
 534                                  *      version
 535                                  *      pool guid
 536                                  *      name
 537                                  *      comment (if available)
 538                                  *      pool state
 539                                  *      hostid (if available)
 540                                  *      hostname (if available)
 541                                  */
 542                                 uint64_t state, version;
 543                                 char *comment = NULL;
 544
 545                                 version = fnvlist_lookup_uint64(tmp,
 546                                     ZPOOL_CONFIG_VERSION);
 547                                 fnvlist_add_uint64(config,
 548                                     ZPOOL_CONFIG_VERSION, version);
 549                                 guid = fnvlist_lookup_uint64(tmp,
 550                                     ZPOOL_CONFIG_POOL_GUID);
 551                                 fnvlist_add_uint64(config,
 552                                     ZPOOL_CONFIG_POOL_GUID, guid);
 553                                 name = fnvlist_lookup_string(tmp,
 554                                     ZPOOL_CONFIG_POOL_NAME);
 555                                 fnvlist_add_string(config,
 556                                     ZPOOL_CONFIG_POOL_NAME, name);
 557
 558                                 if (nvlist_lookup_string(tmp,
 559                                     ZPOOL_CONFIG_COMMENT, &comment) == 0)
 560                                         fnvlist_add_string(config,
 561                                             ZPOOL_CONFIG_COMMENT, comment);
 562
 563                                 state = fnvlist_lookup_uint64(tmp,
 564                                     ZPOOL_CONFIG_POOL_STATE);
 565                                 fnvlist_add_uint64(config,
 566                                     ZPOOL_CONFIG_POOL_STATE, state);
 567
 568                                 hostid = 0;
 569                                 if (nvlist_lookup_uint64(tmp,
 570                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 571                                         fnvlist_add_uint64(config,
 572                                             ZPOOL_CONFIG_HOSTID, hostid);
 573                                         hostname = fnvlist_lookup_string(tmp,
 574                                             ZPOOL_CONFIG_HOSTNAME);
 575                                         fnvlist_add_string(config,
 576                                             ZPOOL_CONFIG_HOSTNAME, hostname);
 577                                 }
 578
 579                                 config_seen = B_TRUE;
 580                         }
 581
 582                         /*
 583                          * Add this top-level vdev to the child array.
 584                          */
 585                         verify(nvlist_lookup_nvlist(tmp,
 586                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 587                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 588                             &id) == 0);
 589
 590                         if (id >= children) {
 591                                 nvlist_t **newchild;
 592
 593                                 newchild = zfs_alloc(hdl, (id + 1) *
 594                                     sizeof (nvlist_t *));
 595                                 if (newchild == NULL)
 596                                         goto nomem;
 597
 598                                 for (c = 0; c < children; c++)
 599                                         newchild[c] = child[c];
 600
 601                                 free(child);
 602                                 child = newchild;
 603                                 children = id + 1;
 604                         }
 605                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 606                                 goto nomem;
 607
 608                 }
 609
 610                 /*
 611                  * If we have information about all the top-levels then
 612                  * clean up the nvlist which we've constructed. This
 613                  * means removing any extraneous devices that are
 614                  * beyond the valid range or adding devices to the end
 615                  * of our array which appear to be missing.
 616                  */
 617                 if (valid_top_config) {
 618                         if (max_id < children) {
 619                                 for (c = max_id; c < children; c++)
 620                                         nvlist_free(child[c]);
 621                                 children = max_id;
 622                         } else if (max_id > children) {
 623                                 nvlist_t **newchild;
 624
 625                                 newchild = zfs_alloc(hdl, (max_id) *
 626                                     sizeof (nvlist_t *));
 627                                 if (newchild == NULL)
 628                                         goto nomem;
 629
 630                                 for (c = 0; c < children; c++)
 631                                         newchild[c] = child[c];
 632
 633                                 free(child);
 634                                 child = newchild;
 635                                 children = max_id;
 636                         }
 637                 }
 638
 639                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 640                     &guid) == 0);
 641
 642                 /*
 643                  * The vdev namespace may contain holes as a result of
 644                  * device removal. We must add them back into the vdev
 645                  * tree before we process any missing devices.
 646                  */
 647                 if (holes > 0) {
 648                         ASSERT(valid_top_config);
 649
 650                         for (c = 0; c < children; c++) {
 651                                 nvlist_t *holey;
 652
 653                                 if (child[c] != NULL ||
 654                                     !vdev_is_hole(hole_array, holes, c))
 655                                         continue;
 656
 657                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 658                                     0) != 0)
 659                                         goto nomem;
 660
 661                                 /*
 662                                  * Holes in the namespace are treated as
 663                                  * "hole" top-level vdevs and have a
 664                                  * special flag set on them.
 665                                  */
 666                                 if (nvlist_add_string(holey,
 667                                     ZPOOL_CONFIG_TYPE,
 668                                     VDEV_TYPE_HOLE) != 0 ||
 669                                     nvlist_add_uint64(holey,
 670                                     ZPOOL_CONFIG_ID, c) != 0 ||
 671                                     nvlist_add_uint64(holey,
 672                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 673                                         nvlist_free(holey);
 674                                         goto nomem;
 675                                 }
 676                                 child[c] = holey;
 677                         }
 678                 }
 679
 680                 /*
 681                  * Look for any missing top-level vdevs.  If this is the case,
 682                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 683                  * simply compress the child array, because the kernel performs
 684                  * certain checks to make sure the vdev IDs match their location
 685                  * in the configuration.
 686                  */
 687                 for (c = 0; c < children; c++) {
 688                         if (child[c] == NULL) {
 689                                 nvlist_t *missing;
 690                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 691                                     0) != 0)
 692                                         goto nomem;
 693                                 if (nvlist_add_string(missing,
 694                                     ZPOOL_CONFIG_TYPE,
 695                                     VDEV_TYPE_MISSING) != 0 ||
 696                                     nvlist_add_uint64(missing,
 697                                     ZPOOL_CONFIG_ID, c) != 0 ||
 698                                     nvlist_add_uint64(missing,
 699                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 700                                         nvlist_free(missing);
 701                                         goto nomem;
 702                                 }
 703                                 child[c] = missing;
 704                         }
 705                 }
 706
 707                 /*
 708                  * Put all of this pool's top-level vdevs into a root vdev.
 709                  */
 710                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 711                         goto nomem;
 712                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 713                     VDEV_TYPE_ROOT) != 0 ||
 714                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 715                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 716                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 717                     child, children) != 0) {
 718                         nvlist_free(nvroot);
 719                         goto nomem;
 720                 }
 721
 722                 for (c = 0; c < children; c++)
 723                         nvlist_free(child[c]);
 724                 free(child);
 725                 children = 0;
 726                 child = NULL;
 727
 728                 /*
 729                  * Go through and fix up any paths and/or devids based on our
 730                  * known list of vdev GUID -> path mappings.
 731                  */
 732                 if (fix_paths(nvroot, pl->names) != 0) {
 733                         nvlist_free(nvroot);
 734                         goto nomem;
 735                 }
 736
 737                 /*
 738                  * Add the root vdev to this pool's configuration.
 739                  */
 740                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 741                     nvroot) != 0) {
 742                         nvlist_free(nvroot);
 743                         goto nomem;
 744                 }
 745                 nvlist_free(nvroot);
 746
 747                 /*
 748                  * zdb uses this path to report on active pools that were
 749                  * imported or created using -R.
 750                  */
 751                 if (active_ok)
 752                         goto add_pool;
 753
 754                 /*
 755                  * Determine if this pool is currently active, in which case we
 756                  * can't actually import it.
 757                  */
 758                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 759                     &name) == 0);
 760                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 761                     &guid) == 0);
 762
 763                 if (pool_active(hdl, name, guid, &isactive) != 0)
 764                         goto error;
 765
 766                 if (isactive) {
 767                         nvlist_free(config);
 768                         config = NULL;
 769                         continue;
 770                 }
 771
 772                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 773                         nvlist_free(config);
 774                         config = NULL;
 775                         continue;
 776                 }
 777
 778                 nvlist_free(config);
 779                 config = nvl;
 780
 781                 /*
 782                  * Go through and update the paths for spares, now that we have
 783                  * them.
 784                  */
 785                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 786                     &nvroot) == 0);
 787                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 788                     &spares, &nspares) == 0) {
 789                         for (i = 0; i < nspares; i++) {
 790                                 if (fix_paths(spares[i], pl->names) != 0)
 791                                         goto nomem;
 792                         }
 793                 }
 794
 795                 /*
 796                  * Update the paths for l2cache devices.
 797                  */
 798                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 799                     &l2cache, &nl2cache) == 0) {
 800                         for (i = 0; i < nl2cache; i++) {
 801                                 if (fix_paths(l2cache[i], pl->names) != 0)
 802                                         goto nomem;
 803                         }
 804                 }
 805
 806                 /*
 807                  * Restore the original information read from the actual label.
 808                  */
 809                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 810                     DATA_TYPE_UINT64);
 811                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 812                     DATA_TYPE_STRING);
 813                 if (hostid != 0) {
 814                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 815                             hostid) == 0);
 816                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 817                             hostname) == 0);
 818                 }
 819
 820 add_pool:
 821                 /*
 822                  * Add this pool to the list of configs.
 823                  */
 824                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 825                     &name) == 0);
 826                 if (nvlist_add_nvlist(ret, name, config) != 0)
 827                         goto nomem;
 828
 829                 found_one = B_TRUE;
 830                 nvlist_free(config);
 831                 config = NULL;
 832         }
 833
 834         if (!found_one) {
 835                 nvlist_free(ret);
 836                 ret = NULL;
 837         }
 838
 839         return (ret);
 840
 841 nomem:
 842         (void) no_memory(hdl);
 843 error:
 844         nvlist_free(config);
 845         nvlist_free(ret);
 846         for (c = 0; c < children; c++)
 847                 nvlist_free(child[c]);
 848         free(child);
 849
 850         return (NULL);
 851 }
 852
 853 /*
 854  * Return the offset of the given label.
 855  */
 856 static uint64_t
 857 label_offset(uint64_t size, int l)
 858 {
 859         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 860         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 861             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 862 }
 863
 864 /*
 865  * Given a file descriptor, read the label information and return an nvlist
 866  * describing the configuration, if there is one.
 867  */
 868 int
 869 zpool_read_label(int fd, nvlist_t **config)
 870 {
 871         struct stat64 statbuf;
 872         int l;
 873         vdev_label_t *label;
 874         uint64_t state, txg, size;
 875
 876         *config = NULL;
 877
 878         if (fstat64(fd, &statbuf) == -1)
 879                 return (0);
 880         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 881
 882         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 883                 return (-1);
 884
 885         for (l = 0; l < VDEV_LABELS; l++) {
 886                 if (pread64(fd, label, sizeof (vdev_label_t),
 887                     label_offset(size, l)) != sizeof (vdev_label_t))
 888                         continue;
 889
 890                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 891                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 892                         continue;
 893
 894                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 895                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 896                         nvlist_free(*config);
 897                         continue;
 898                 }
 899
 900                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 901                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 902                     &txg) != 0 || txg == 0)) {
 903                         nvlist_free(*config);
 904                         continue;
 905                 }
 906
 907                 free(label);
 908                 return (0);
 909         }
 910
 911         free(label);
 912         *config = NULL;
 913         return (0);
 914 }
 915
 916 typedef struct rdsk_node {
 917         char *rn_name;
 918         int rn_dfd;
 919         libzfs_handle_t *rn_hdl;
 920         nvlist_t *rn_config;
 921         avl_tree_t *rn_avl;
 922         avl_node_t rn_node;
 923         boolean_t rn_nozpool;
 924 } rdsk_node_t;
 925
 926 static int
 927 slice_cache_compare(const void *arg1, const void *arg2)
 928 {
 929         const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
 930         const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
 931         char *nm1slice, *nm2slice;
 932         int rv;
 933
 934         /*
 935          * slices zero and two are the most likely to provide results,
 936          * so put those first
 937          */
 938         nm1slice = strstr(nm1, "s0");
 939         nm2slice = strstr(nm2, "s0");
 940         if (nm1slice && !nm2slice) {
 941                 return (-1);
 942         }
 943         if (!nm1slice && nm2slice) {
 944                 return (1);
 945         }
 946         nm1slice = strstr(nm1, "s2");
 947         nm2slice = strstr(nm2, "s2");
 948         if (nm1slice && !nm2slice) {
 949                 return (-1);
 950         }
 951         if (!nm1slice && nm2slice) {
 952                 return (1);
 953         }
 954
 955         rv = strcmp(nm1, nm2);
 956         if (rv == 0)
 957                 return (0);
 958         return (rv > 0 ? 1 : -1);
 959 }
 960
 961 #ifdef sun
 962 static void
 963 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
 964     diskaddr_t size, uint_t blksz)
 965 {
 966         rdsk_node_t tmpnode;
 967         rdsk_node_t *node;
 968         char sname[MAXNAMELEN];
 969
 970         tmpnode.rn_name = &sname[0];
 971         (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
 972             diskname, partno);
 973         /*
 974          * protect against division by zero for disk labels that
 975          * contain a bogus sector size
 976          */
 977         if (blksz == 0)
 978                 blksz = DEV_BSIZE;
 979         /* too small to contain a zpool? */
 980         if ((size < (SPA_MINDEVSIZE / blksz)) &&
 981             (node = avl_find(r, &tmpnode, NULL)))
 982                 node->rn_nozpool = B_TRUE;
 983 }
 984 #endif  /* sun */
 985
 986 static void
 987 nozpool_all_slices(avl_tree_t *r, const char *sname)
 988 {
 989 #ifdef sun
 990         char diskname[MAXNAMELEN];
 991         char *ptr;
 992         int i;
 993
 994         (void) strncpy(diskname, sname, MAXNAMELEN);
 995         if (((ptr = strrchr(diskname, 's')) == NULL) &&
 996             ((ptr = strrchr(diskname, 'p')) == NULL))
 997                 return;
 998         ptr[0] = 's';
 999         ptr[1] = '\0';
1000         for (i = 0; i < NDKMAP; i++)
1001                 check_one_slice(r, diskname, i, 0, 1);
1002         ptr[0] = 'p';
1003         for (i = 0; i <= FD_NUMPART; i++)
1004                 check_one_slice(r, diskname, i, 0, 1);
1005 #endif  /* sun */
1006 }
1007
1008 #ifdef sun
1009 static void
1010 check_slices(avl_tree_t *r, int fd, const char *sname)
1011 {
1012         struct extvtoc vtoc;
1013         struct dk_gpt *gpt;
1014         char diskname[MAXNAMELEN];
1015         char *ptr;
1016         int i;
1017
1018         (void) strncpy(diskname, sname, MAXNAMELEN);
1019         if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1020                 return;
1021         ptr[1] = '\0';
1022
1023         if (read_extvtoc(fd, &vtoc) >= 0) {
1024                 for (i = 0; i < NDKMAP; i++)
1025                         check_one_slice(r, diskname, i,
1026                             vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1027         } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1028                 /*
1029                  * on x86 we'll still have leftover links that point
1030                  * to slices s[9-15], so use NDKMAP instead
1031                  */
1032                 for (i = 0; i < NDKMAP; i++)
1033                         check_one_slice(r, diskname, i,
1034                             gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1035                 /* nodes p[1-4] are never used with EFI labels */
1036                 ptr[0] = 'p';
1037                 for (i = 1; i <= FD_NUMPART; i++)
1038                         check_one_slice(r, diskname, i, 0, 1);
1039                 efi_free(gpt);
1040         }
1041 }
1042 #endif  /* sun */
1043
1044 static void
1045 zpool_open_func(void *arg)
1046 {
1047         rdsk_node_t *rn = arg;
1048         struct stat64 statbuf;
1049         nvlist_t *config;
1050         int fd;
1051
1052         if (rn->rn_nozpool)
1053                 return;
1054         if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1055                 /* symlink to a device that's no longer there */
1056                 if (errno == ENOENT)
1057                         nozpool_all_slices(rn->rn_avl, rn->rn_name);
1058                 return;
1059         }
1060         /*
1061          * Ignore failed stats.  We only want regular
1062          * files, character devs and block devs.
1063          */
1064         if (fstat64(fd, &statbuf) != 0 ||
1065             (!S_ISREG(statbuf.st_mode) &&
1066             !S_ISCHR(statbuf.st_mode) &&
1067             !S_ISBLK(statbuf.st_mode))) {
1068                 (void) close(fd);
1069                 return;
1070         }
1071         /* this file is too small to hold a zpool */
1072 #ifdef sun
1073         if (S_ISREG(statbuf.st_mode) &&
1074             statbuf.st_size < SPA_MINDEVSIZE) {
1075                 (void) close(fd);
1076                 return;
1077         } else if (!S_ISREG(statbuf.st_mode)) {
1078                 /*
1079                  * Try to read the disk label first so we don't have to
1080                  * open a bunch of minor nodes that can't have a zpool.
1081                  */
1082                 check_slices(rn->rn_avl, fd, rn->rn_name);
1083         }
1084 #else   /* !sun */
1085         if (statbuf.st_size < SPA_MINDEVSIZE) {
1086                 (void) close(fd);
1087                 return;
1088         }
1089 #endif  /* sun */
1090
1091         if ((zpool_read_label(fd, &config)) != 0) {
1092                 (void) close(fd);
1093                 (void) no_memory(rn->rn_hdl);
1094                 return;
1095         }
1096         (void) close(fd);
1097
1098         rn->rn_config = config;
1099 }
1100
1101 /*
1102  * Given a file descriptor, clear (zero) the label information.  This function
1103  * is used in the appliance stack as part of the ZFS sysevent module and
1104  * to implement the "zpool labelclear" command.
1105  */
1106 int
1107 zpool_clear_label(int fd)
1108 {
1109         struct stat64 statbuf;
1110         int l;
1111         vdev_label_t *label;
1112         uint64_t size;
1113
1114         if (fstat64(fd, &statbuf) == -1)
1115                 return (0);
1116         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1117
1118         if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1119                 return (-1);
1120
1121         for (l = 0; l < VDEV_LABELS; l++) {
1122                 if (pwrite64(fd, label, sizeof (vdev_label_t),
1123                     label_offset(size, l)) != sizeof (vdev_label_t)) {
1124                         free(label);
1125                         return (-1);
1126                 }
1127         }
1128
1129         free(label);
1130         return (0);
1131 }
1132
1133 /*
1134  * Given a list of directories to search, find all pools stored on disk.  This
1135  * includes partial pools which are not available to import.  If no args are
1136  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1137  * poolname or guid (but not both) are provided by the caller when trying
1138  * to import a specific pool.
1139  */
1140 static nvlist_t *
1141 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1142 {
1143         int i, dirs = iarg->paths;
1144         struct dirent64 *dp;
1145         char path[MAXPATHLEN];
1146         char *end, **dir = iarg->path;
1147         size_t pathleft;
1148         nvlist_t *ret = NULL;
1149         static char *default_dir = "/dev";
1150         pool_list_t pools = { 0 };
1151         pool_entry_t *pe, *penext;
1152         vdev_entry_t *ve, *venext;
1153         config_entry_t *ce, *cenext;
1154         name_entry_t *ne, *nenext;
1155         avl_tree_t slice_cache;
1156         rdsk_node_t *slice;
1157         void *cookie;
1158
1159         if (dirs == 0) {
1160                 dirs = 1;
1161                 dir = &default_dir;
1162         }
1163
1164         /*
1165          * Go through and read the label configuration information from every
1166          * possible device, organizing the information according to pool GUID
1167          * and toplevel GUID.
1168          */
1169         for (i = 0; i < dirs; i++) {
1170                 tpool_t *t;
1171                 char *rdsk;
1172                 int dfd;
1173                 boolean_t config_failed = B_FALSE;
1174                 DIR *dirp;
1175
1176                 /* use realpath to normalize the path */
1177                 if (realpath(dir[i], path) == 0) {
1178                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1179                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1180                         goto error;
1181                 }
1182                 end = &path[strlen(path)];
1183                 *end++ = '/';
1184                 *end = 0;
1185                 pathleft = &path[sizeof (path)] - end;
1186
1187                 /*
1188                  * Using raw devices instead of block devices when we're
1189                  * reading the labels skips a bunch of slow operations during
1190                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1191                  */
1192                 if (strcmp(path, "/dev/dsk/") == 0)
1193                         rdsk = "/dev/";
1194                 else
1195                         rdsk = path;
1196
1197                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1198                     (dirp = fdopendir(dfd)) == NULL) {
1199                         if (dfd >= 0)
1200                                 (void) close(dfd);
1201                         zfs_error_aux(hdl, strerror(errno));
1202                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1203                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1204                             rdsk);
1205                         goto error;
1206                 }
1207
1208                 avl_create(&slice_cache, slice_cache_compare,
1209                     sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1210
1211                 if (strcmp(rdsk, "/dev/") == 0) {
1212                         struct gmesh mesh;
1213                         struct gclass *mp;
1214                         struct ggeom *gp;
1215                         struct gprovider *pp;
1216
1217                         errno = geom_gettree(&mesh);
1218                         if (errno != 0) {
1219                                 zfs_error_aux(hdl, strerror(errno));
1220                                 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1221                                     dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1222                                 goto error;
1223                         }
1224
1225                         LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1226                                 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1227                                         LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1228                                                 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1229                                                 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1230                                                 slice->rn_avl = &slice_cache;
1231                                                 slice->rn_dfd = dfd;
1232                                                 slice->rn_hdl = hdl;
1233                                                 slice->rn_nozpool = B_FALSE;
1234                                                 avl_add(&slice_cache, slice);
1235                                         }
1236                                 }
1237                         }
1238
1239                         geom_deletetree(&mesh);
1240                         goto skipdir;
1241                 }
1242
1243                 /*
1244                  * This is not MT-safe, but we have no MT consumers of libzfs
1245                  */
1246                 while ((dp = readdir64(dirp)) != NULL) {
1247                         const char *name = dp->d_name;
1248                         if (name[0] == '.' &&
1249                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1250                                 continue;
1251
1252                         slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1253                         slice->rn_name = zfs_strdup(hdl, name);
1254                         slice->rn_avl = &slice_cache;
1255                         slice->rn_dfd = dfd;
1256                         slice->rn_hdl = hdl;
1257                         slice->rn_nozpool = B_FALSE;
1258                         avl_add(&slice_cache, slice);
1259                 }
1260 skipdir:
1261                 /*
1262                  * create a thread pool to do all of this in parallel;
1263                  * rn_nozpool is not protected, so this is racy in that
1264                  * multiple tasks could decide that the same slice can
1265                  * not hold a zpool, which is benign.  Also choose
1266                  * double the number of processors; we hold a lot of
1267                  * locks in the kernel, so going beyond this doesn't
1268                  * buy us much.
1269                  */
1270                 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1271                     0, NULL);
1272                 for (slice = avl_first(&slice_cache); slice;
1273                     (slice = avl_walk(&slice_cache, slice,
1274                     AVL_AFTER)))
1275                         (void) tpool_dispatch(t, zpool_open_func, slice);
1276                 tpool_wait(t);
1277                 tpool_destroy(t);
1278
1279                 cookie = NULL;
1280                 while ((slice = avl_destroy_nodes(&slice_cache,
1281                     &cookie)) != NULL) {
1282                         if (slice->rn_config != NULL && !config_failed) {
1283                                 nvlist_t *config = slice->rn_config;
1284                                 boolean_t matched = B_TRUE;
1285
1286                                 if (iarg->poolname != NULL) {
1287                                         char *pname;
1288
1289                                         matched = nvlist_lookup_string(config,
1290                                             ZPOOL_CONFIG_POOL_NAME,
1291                                             &pname) == 0 &&
1292                                             strcmp(iarg->poolname, pname) == 0;
1293                                 } else if (iarg->guid != 0) {
1294                                         uint64_t this_guid;
1295
1296                                         matched = nvlist_lookup_uint64(config,
1297                                             ZPOOL_CONFIG_POOL_GUID,
1298                                             &this_guid) == 0 &&
1299                                             iarg->guid == this_guid;
1300                                 }
1301                                 if (!matched) {
1302                                         nvlist_free(config);
1303                                 } else {
1304                                         /*
1305                                          * use the non-raw path for the config
1306                                          */
1307                                         (void) strlcpy(end, slice->rn_name,
1308                                             pathleft);
1309                                         if (add_config(hdl, &pools, path,
1310                                             config) != 0)
1311                                                 config_failed = B_TRUE;
1312                                 }
1313                         }
1314                         free(slice->rn_name);
1315                         free(slice);
1316                 }
1317                 avl_destroy(&slice_cache);
1318
1319                 (void) closedir(dirp);
1320
1321                 if (config_failed)
1322                         goto error;
1323         }
1324
1325         ret = get_configs(hdl, &pools, iarg->can_be_active);
1326
1327 error:
1328         for (pe = pools.pools; pe != NULL; pe = penext) {
1329                 penext = pe->pe_next;
1330                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1331                         venext = ve->ve_next;
1332                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1333                                 cenext = ce->ce_next;
1334                                 if (ce->ce_config)
1335                                         nvlist_free(ce->ce_config);
1336                                 free(ce);
1337                         }
1338                         free(ve);
1339                 }
1340                 free(pe);
1341         }
1342
1343         for (ne = pools.names; ne != NULL; ne = nenext) {
1344                 nenext = ne->ne_next;
1345                 free(ne->ne_name);
1346                 free(ne);
1347         }
1348
1349         return (ret);
1350 }
1351
1352 nvlist_t *
1353 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1354 {
1355         importargs_t iarg = { 0 };
1356
1357         iarg.paths = argc;
1358         iarg.path = argv;
1359
1360         return (zpool_find_import_impl(hdl, &iarg));
1361 }
1362
1363 /*
1364  * Given a cache file, return the contents as a list of importable pools.
1365  * poolname or guid (but not both) are provided by the caller when trying
1366  * to import a specific pool.
1367  */
1368 nvlist_t *
1369 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1370     char *poolname, uint64_t guid)
1371 {
1372         char *buf;
1373         int fd;
1374         struct stat64 statbuf;
1375         nvlist_t *raw, *src, *dst;
1376         nvlist_t *pools;
1377         nvpair_t *elem;
1378         char *name;
1379         uint64_t this_guid;
1380         boolean_t active;
1381
1382         verify(poolname == NULL || guid == 0);
1383
1384         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1385                 zfs_error_aux(hdl, "%s", strerror(errno));
1386                 (void) zfs_error(hdl, EZFS_BADCACHE,
1387                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1388                 return (NULL);
1389         }
1390
1391         if (fstat64(fd, &statbuf) != 0) {
1392                 zfs_error_aux(hdl, "%s", strerror(errno));
1393                 (void) close(fd);
1394                 (void) zfs_error(hdl, EZFS_BADCACHE,
1395                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1396                 return (NULL);
1397         }
1398
1399         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1400                 (void) close(fd);
1401                 return (NULL);
1402         }
1403
1404         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1405                 (void) close(fd);
1406                 free(buf);
1407                 (void) zfs_error(hdl, EZFS_BADCACHE,
1408                     dgettext(TEXT_DOMAIN,
1409                     "failed to read cache file contents"));
1410                 return (NULL);
1411         }
1412
1413         (void) close(fd);
1414
1415         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1416                 free(buf);
1417                 (void) zfs_error(hdl, EZFS_BADCACHE,
1418                     dgettext(TEXT_DOMAIN,
1419                     "invalid or corrupt cache file contents"));
1420                 return (NULL);
1421         }
1422
1423         free(buf);
1424
1425         /*
1426          * Go through and get the current state of the pools and refresh their
1427          * state.
1428          */
1429         if (nvlist_alloc(&pools, 0, 0) != 0) {
1430                 (void) no_memory(hdl);
1431                 nvlist_free(raw);
1432                 return (NULL);
1433         }
1434
1435         elem = NULL;
1436         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1437                 src = fnvpair_value_nvlist(elem);
1438
1439                 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1440                 if (poolname != NULL && strcmp(poolname, name) != 0)
1441                         continue;
1442
1443                 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1444                 if (guid != 0 && guid != this_guid)
1445                         continue;
1446
1447                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1448                         nvlist_free(raw);
1449                         nvlist_free(pools);
1450                         return (NULL);
1451                 }
1452
1453                 if (active)
1454                         continue;
1455
1456                 if ((dst = refresh_config(hdl, src)) == NULL) {
1457                         nvlist_free(raw);
1458                         nvlist_free(pools);
1459                         return (NULL);
1460                 }
1461
1462                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1463                         (void) no_memory(hdl);
1464                         nvlist_free(dst);
1465                         nvlist_free(raw);
1466                         nvlist_free(pools);
1467                         return (NULL);
1468                 }
1469                 nvlist_free(dst);
1470         }
1471
1472         nvlist_free(raw);
1473         return (pools);
1474 }
1475
1476 static int
1477 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1478 {
1479         importargs_t *import = data;
1480         int found = 0;
1481
1482         if (import->poolname != NULL) {
1483                 char *pool_name;
1484
1485                 verify(nvlist_lookup_string(zhp->zpool_config,
1486                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1487                 if (strcmp(pool_name, import->poolname) == 0)
1488                         found = 1;
1489         } else {
1490                 uint64_t pool_guid;
1491
1492                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1493                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1494                 if (pool_guid == import->guid)
1495                         found = 1;
1496         }
1497
1498         zpool_close(zhp);
1499         return (found);
1500 }
1501
1502 nvlist_t *
1503 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1504 {
1505         verify(import->poolname == NULL || import->guid == 0);
1506
1507         if (import->unique)
1508                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1509
1510         if (import->cachefile != NULL)
1511                 return (zpool_find_import_cached(hdl, import->cachefile,
1512                     import->poolname, import->guid));
1513
1514         return (zpool_find_import_impl(hdl, import));
1515 }
1516
1517 boolean_t
1518 find_guid(nvlist_t *nv, uint64_t guid)
1519 {
1520         uint64_t tmp;
1521         nvlist_t **child;
1522         uint_t c, children;
1523
1524         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1525         if (tmp == guid)
1526                 return (B_TRUE);
1527
1528         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1529             &child, &children) == 0) {
1530                 for (c = 0; c < children; c++)
1531                         if (find_guid(child[c], guid))
1532                                 return (B_TRUE);
1533         }
1534
1535         return (B_FALSE);
1536 }
1537
1538 typedef struct aux_cbdata {
1539         const char      *cb_type;
1540         uint64_t        cb_guid;
1541         zpool_handle_t  *cb_zhp;
1542 } aux_cbdata_t;
1543
1544 static int
1545 find_aux(zpool_handle_t *zhp, void *data)
1546 {
1547         aux_cbdata_t *cbp = data;
1548         nvlist_t **list;
1549         uint_t i, count;
1550         uint64_t guid;
1551         nvlist_t *nvroot;
1552
1553         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1554             &nvroot) == 0);
1555
1556         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1557             &list, &count) == 0) {
1558                 for (i = 0; i < count; i++) {
1559                         verify(nvlist_lookup_uint64(list[i],
1560                             ZPOOL_CONFIG_GUID, &guid) == 0);
1561                         if (guid == cbp->cb_guid) {
1562                                 cbp->cb_zhp = zhp;
1563                                 return (1);
1564                         }
1565                 }
1566         }
1567
1568         zpool_close(zhp);
1569         return (0);
1570 }
1571
1572 /*
1573  * Determines if the pool is in use.  If so, it returns true and the state of
1574  * the pool as well as the name of the pool.  Both strings are allocated and
1575  * must be freed by the caller.
1576  */
1577 int
1578 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1579     boolean_t *inuse)
1580 {
1581         nvlist_t *config;
1582         char *name;
1583         boolean_t ret;
1584         uint64_t guid, vdev_guid;
1585         zpool_handle_t *zhp;
1586         nvlist_t *pool_config;
1587         uint64_t stateval, isspare;
1588         aux_cbdata_t cb = { 0 };
1589         boolean_t isactive;
1590
1591         *inuse = B_FALSE;
1592
1593         if (zpool_read_label(fd, &config) != 0) {
1594                 (void) no_memory(hdl);
1595                 return (-1);
1596         }
1597
1598         if (config == NULL)
1599                 return (0);
1600
1601         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1602             &stateval) == 0);
1603         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1604             &vdev_guid) == 0);
1605
1606         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1607                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1608                     &name) == 0);
1609                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1610                     &guid) == 0);
1611         }
1612
1613         switch (stateval) {
1614         case POOL_STATE_EXPORTED:
1615                 /*
1616                  * A pool with an exported state may in fact be imported
1617                  * read-only, so check the in-core state to see if it's
1618                  * active and imported read-only.  If it is, set
1619                  * its state to active.
1620                  */
1621                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1622                     (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1623                         if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1624                                 stateval = POOL_STATE_ACTIVE;
1625
1626                         /*
1627                          * All we needed the zpool handle for is the
1628                          * readonly prop check.
1629                          */
1630                         zpool_close(zhp);
1631                 }
1632
1633                 ret = B_TRUE;
1634                 break;
1635
1636         case POOL_STATE_ACTIVE:
1637                 /*
1638                  * For an active pool, we have to determine if it's really part
1639                  * of a currently active pool (in which case the pool will exist
1640                  * and the guid will be the same), or whether it's part of an
1641                  * active pool that was disconnected without being explicitly
1642                  * exported.
1643                  */
1644                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1645                         nvlist_free(config);
1646                         return (-1);
1647                 }
1648
1649                 if (isactive) {
1650                         /*
1651                          * Because the device may have been removed while
1652                          * offlined, we only report it as active if the vdev is
1653                          * still present in the config.  Otherwise, pretend like
1654                          * it's not in use.
1655                          */
1656                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1657                             (pool_config = zpool_get_config(zhp, NULL))
1658                             != NULL) {
1659                                 nvlist_t *nvroot;
1660
1661                                 verify(nvlist_lookup_nvlist(pool_config,
1662                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1663                                 ret = find_guid(nvroot, vdev_guid);
1664                         } else {
1665                                 ret = B_FALSE;
1666                         }
1667
1668                         /*
1669                          * If this is an active spare within another pool, we
1670                          * treat it like an unused hot spare.  This allows the
1671                          * user to create a pool with a hot spare that currently
1672                          * in use within another pool.  Since we return B_TRUE,
1673                          * libdiskmgt will continue to prevent generic consumers
1674                          * from using the device.
1675                          */
1676                         if (ret && nvlist_lookup_uint64(config,
1677                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1678                                 stateval = POOL_STATE_SPARE;
1679
1680                         if (zhp != NULL)
1681                                 zpool_close(zhp);
1682                 } else {
1683                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1684                         ret = B_TRUE;
1685                 }
1686                 break;
1687
1688         case POOL_STATE_SPARE:
1689                 /*
1690                  * For a hot spare, it can be either definitively in use, or
1691                  * potentially active.  To determine if it's in use, we iterate
1692                  * over all pools in the system and search for one with a spare
1693                  * with a matching guid.
1694                  *
1695                  * Due to the shared nature of spares, we don't actually report
1696                  * the potentially active case as in use.  This means the user
1697                  * can freely create pools on the hot spares of exported pools,
1698                  * but to do otherwise makes the resulting code complicated, and
1699                  * we end up having to deal with this case anyway.
1700                  */
1701                 cb.cb_zhp = NULL;
1702                 cb.cb_guid = vdev_guid;
1703                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1704                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1705                         name = (char *)zpool_get_name(cb.cb_zhp);
1706                         ret = B_TRUE;
1707                 } else {
1708                         ret = B_FALSE;
1709                 }
1710                 break;
1711
1712         case POOL_STATE_L2CACHE:
1713
1714                 /*
1715                  * Check if any pool is currently using this l2cache device.
1716                  */
1717                 cb.cb_zhp = NULL;
1718                 cb.cb_guid = vdev_guid;
1719                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1720                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1721                         name = (char *)zpool_get_name(cb.cb_zhp);
1722                         ret = B_TRUE;
1723                 } else {
1724                         ret = B_FALSE;
1725                 }
1726                 break;
1727
1728         default:
1729                 ret = B_FALSE;
1730         }
1731
1732
1733         if (ret) {
1734                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1735                         if (cb.cb_zhp)
1736                                 zpool_close(cb.cb_zhp);
1737                         nvlist_free(config);
1738                         return (-1);
1739                 }
1740                 *state = (pool_state_t)stateval;
1741         }
1742
1743         if (cb.cb_zhp)
1744                 zpool_close(cb.cb_zhp);
1745
1746         nvlist_free(config);
1747         *inuse = ret;
1748         return (0);
1749 }