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  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  24  * Copyright (c) 2012 by Delphix. All rights reserved.
  25  */
  26
  27 /*
  28  * Pool import support functions.
  29  *
  30  * To import a pool, we rely on reading the configuration information from the
  31  * ZFS label of each device.  If we successfully read the label, then we
  32  * organize the configuration information in the following hierarchy:
  33  *
  34  *      pool guid -> toplevel vdev guid -> label txg
  35  *
  36  * Duplicate entries matching this same tuple will be discarded.  Once we have
  37  * examined every device, we pick the best label txg config for each toplevel
  38  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  39  * update any paths that have changed.  Finally, we attempt to import the pool
  40  * using our derived config, and record the results.
  41  */
  42
  43 #include <ctype.h>
  44 #include <devid.h>
  45 #include <dirent.h>
  46 #include <errno.h>
  47 #include <libintl.h>
  48 #include <stddef.h>
  49 #include <stdlib.h>
  50 #include <string.h>
  51 #include <sys/stat.h>
  52 #include <unistd.h>
  53 #include <fcntl.h>
  54 #include <thread_pool.h>
  55 #include <libgeom.h>
  56
  57 #include <sys/vdev_impl.h>
  58
  59 #include "libzfs.h"
  60 #include "libzfs_impl.h"
  61
  62 /*
  63  * Intermediate structures used to gather configuration information.
  64  */
  65 typedef struct config_entry {
  66         uint64_t                ce_txg;
  67         nvlist_t                *ce_config;
  68         struct config_entry     *ce_next;
  69 } config_entry_t;
  70
  71 typedef struct vdev_entry {
  72         uint64_t                ve_guid;
  73         config_entry_t          *ve_configs;
  74         struct vdev_entry       *ve_next;
  75 } vdev_entry_t;
  76
  77 typedef struct pool_entry {
  78         uint64_t                pe_guid;
  79         vdev_entry_t            *pe_vdevs;
  80         struct pool_entry       *pe_next;
  81 } pool_entry_t;
  82
  83 typedef struct name_entry {
  84         char                    *ne_name;
  85         uint64_t                ne_guid;
  86         struct name_entry       *ne_next;
  87 } name_entry_t;
  88
  89 typedef struct pool_list {
  90         pool_entry_t            *pools;
  91         name_entry_t            *names;
  92 } pool_list_t;
  93
  94 static char *
  95 get_devid(const char *path)
  96 {
  97         int fd;
  98         ddi_devid_t devid;
  99         char *minor, *ret;
 100
 101         if ((fd = open(path, O_RDONLY)) < 0)
 102                 return (NULL);
 103
 104         minor = NULL;
 105         ret = NULL;
 106         if (devid_get(fd, &devid) == 0) {
 107                 if (devid_get_minor_name(fd, &minor) == 0)
 108                         ret = devid_str_encode(devid, minor);
 109                 if (minor != NULL)
 110                         devid_str_free(minor);
 111                 devid_free(devid);
 112         }
 113         (void) close(fd);
 114
 115         return (ret);
 116 }
 117
 118
 119 /*
 120  * Go through and fix up any path and/or devid information for the given vdev
 121  * configuration.
 122  */
 123 static int
 124 fix_paths(nvlist_t *nv, name_entry_t *names)
 125 {
 126         nvlist_t **child;
 127         uint_t c, children;
 128         uint64_t guid;
 129         name_entry_t *ne, *best;
 130         char *path, *devid;
 131         int matched;
 132
 133         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 134             &child, &children) == 0) {
 135                 for (c = 0; c < children; c++)
 136                         if (fix_paths(child[c], names) != 0)
 137                                 return (-1);
 138                 return (0);
 139         }
 140
 141         /*
 142          * This is a leaf (file or disk) vdev.  In either case, go through
 143          * the name list and see if we find a matching guid.  If so, replace
 144          * the path and see if we can calculate a new devid.
 145          *
 146          * There may be multiple names associated with a particular guid, in
 147          * which case we have overlapping slices or multiple paths to the same
 148          * disk.  If this is the case, then we want to pick the path that is
 149          * the most similar to the original, where "most similar" is the number
 150          * of matching characters starting from the end of the path.  This will
 151          * preserve slice numbers even if the disks have been reorganized, and
 152          * will also catch preferred disk names if multiple paths exist.
 153          */
 154         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 155         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 156                 path = NULL;
 157
 158         matched = 0;
 159         best = NULL;
 160         for (ne = names; ne != NULL; ne = ne->ne_next) {
 161                 if (ne->ne_guid == guid) {
 162                         const char *src, *dst;
 163                         int count;
 164
 165                         if (path == NULL) {
 166                                 best = ne;
 167                                 break;
 168                         }
 169
 170                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 171                         dst = path + strlen(path) - 1;
 172                         for (count = 0; src >= ne->ne_name && dst >= path;
 173                             src--, dst--, count++)
 174                                 if (*src != *dst)
 175                                         break;
 176
 177                         /*
 178                          * At this point, 'count' is the number of characters
 179                          * matched from the end.
 180                          */
 181                         if (count > matched || best == NULL) {
 182                                 best = ne;
 183                                 matched = count;
 184                         }
 185                 }
 186         }
 187
 188         if (best == NULL)
 189                 return (0);
 190
 191         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 192                 return (-1);
 193
 194         if ((devid = get_devid(best->ne_name)) == NULL) {
 195                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 196         } else {
 197                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
 198                         return (-1);
 199                 devid_str_free(devid);
 200         }
 201
 202         return (0);
 203 }
 204
 205 /*
 206  * Add the given configuration to the list of known devices.
 207  */
 208 static int
 209 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 210     nvlist_t *config)
 211 {
 212         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 213         pool_entry_t *pe;
 214         vdev_entry_t *ve;
 215         config_entry_t *ce;
 216         name_entry_t *ne;
 217
 218         /*
 219          * If this is a hot spare not currently in use or level 2 cache
 220          * device, add it to the list of names to translate, but don't do
 221          * anything else.
 222          */
 223         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 224             &state) == 0 &&
 225             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 226             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 227                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 228                         return (-1);
 229
 230                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 231                         free(ne);
 232                         return (-1);
 233                 }
 234                 ne->ne_guid = vdev_guid;
 235                 ne->ne_next = pl->names;
 236                 pl->names = ne;
 237                 return (0);
 238         }
 239
 240         /*
 241          * If we have a valid config but cannot read any of these fields, then
 242          * it means we have a half-initialized label.  In vdev_label_init()
 243          * we write a label with txg == 0 so that we can identify the device
 244          * in case the user refers to the same disk later on.  If we fail to
 245          * create the pool, we'll be left with a label in this state
 246          * which should not be considered part of a valid pool.
 247          */
 248         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 249             &pool_guid) != 0 ||
 250             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 251             &vdev_guid) != 0 ||
 252             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 253             &top_guid) != 0 ||
 254             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 255             &txg) != 0 || txg == 0) {
 256                 nvlist_free(config);
 257                 return (0);
 258         }
 259
 260         /*
 261          * First, see if we know about this pool.  If not, then add it to the
 262          * list of known pools.
 263          */
 264         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 265                 if (pe->pe_guid == pool_guid)
 266                         break;
 267         }
 268
 269         if (pe == NULL) {
 270                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 271                         nvlist_free(config);
 272                         return (-1);
 273                 }
 274                 pe->pe_guid = pool_guid;
 275                 pe->pe_next = pl->pools;
 276                 pl->pools = pe;
 277         }
 278
 279         /*
 280          * Second, see if we know about this toplevel vdev.  Add it if its
 281          * missing.
 282          */
 283         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 284                 if (ve->ve_guid == top_guid)
 285                         break;
 286         }
 287
 288         if (ve == NULL) {
 289                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 290                         nvlist_free(config);
 291                         return (-1);
 292                 }
 293                 ve->ve_guid = top_guid;
 294                 ve->ve_next = pe->pe_vdevs;
 295                 pe->pe_vdevs = ve;
 296         }
 297
 298         /*
 299          * Third, see if we have a config with a matching transaction group.  If
 300          * so, then we do nothing.  Otherwise, add it to the list of known
 301          * configs.
 302          */
 303         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 304                 if (ce->ce_txg == txg)
 305                         break;
 306         }
 307
 308         if (ce == NULL) {
 309                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 310                         nvlist_free(config);
 311                         return (-1);
 312                 }
 313                 ce->ce_txg = txg;
 314                 ce->ce_config = config;
 315                 ce->ce_next = ve->ve_configs;
 316                 ve->ve_configs = ce;
 317         } else {
 318                 nvlist_free(config);
 319         }
 320
 321         /*
 322          * At this point we've successfully added our config to the list of
 323          * known configs.  The last thing to do is add the vdev guid -> path
 324          * mappings so that we can fix up the configuration as necessary before
 325          * doing the import.
 326          */
 327         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 328                 return (-1);
 329
 330         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 331                 free(ne);
 332                 return (-1);
 333         }
 334
 335         ne->ne_guid = vdev_guid;
 336         ne->ne_next = pl->names;
 337         pl->names = ne;
 338
 339         return (0);
 340 }
 341
 342 /*
 343  * Returns true if the named pool matches the given GUID.
 344  */
 345 static int
 346 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 347     boolean_t *isactive)
 348 {
 349         zpool_handle_t *zhp;
 350         uint64_t theguid;
 351
 352         if (zpool_open_silent(hdl, name, &zhp) != 0)
 353                 return (-1);
 354
 355         if (zhp == NULL) {
 356                 *isactive = B_FALSE;
 357                 return (0);
 358         }
 359
 360         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 361             &theguid) == 0);
 362
 363         zpool_close(zhp);
 364
 365         *isactive = (theguid == guid);
 366         return (0);
 367 }
 368
 369 static nvlist_t *
 370 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 371 {
 372         nvlist_t *nvl;
 373         zfs_cmd_t zc = { 0 };
 374         int err;
 375
 376         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 377                 return (NULL);
 378
 379         if (zcmd_alloc_dst_nvlist(hdl, &zc,
 380             zc.zc_nvlist_conf_size * 2) != 0) {
 381                 zcmd_free_nvlists(&zc);
 382                 return (NULL);
 383         }
 384
 385         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 386             &zc)) != 0 && errno == ENOMEM) {
 387                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 388                         zcmd_free_nvlists(&zc);
 389                         return (NULL);
 390                 }
 391         }
 392
 393         if (err) {
 394                 zcmd_free_nvlists(&zc);
 395                 return (NULL);
 396         }
 397
 398         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 399                 zcmd_free_nvlists(&zc);
 400                 return (NULL);
 401         }
 402
 403         zcmd_free_nvlists(&zc);
 404         return (nvl);
 405 }
 406
 407 /*
 408  * Determine if the vdev id is a hole in the namespace.
 409  */
 410 boolean_t
 411 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
 412 {
 413         for (int c = 0; c < holes; c++) {
 414
 415                 /* Top-level is a hole */
 416                 if (hole_array[c] == id)
 417                         return (B_TRUE);
 418         }
 419         return (B_FALSE);
 420 }
 421
 422 /*
 423  * Convert our list of pools into the definitive set of configurations.  We
 424  * start by picking the best config for each toplevel vdev.  Once that's done,
 425  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 426  * pass to fix up any incorrect paths, and then add it to the main list to
 427  * return to the user.
 428  */
 429 static nvlist_t *
 430 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 431 {
 432         pool_entry_t *pe;
 433         vdev_entry_t *ve;
 434         config_entry_t *ce;
 435         nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
 436         nvlist_t **spares, **l2cache;
 437         uint_t i, nspares, nl2cache;
 438         boolean_t config_seen;
 439         uint64_t best_txg;
 440         char *name, *hostname;
 441         uint64_t guid;
 442         uint_t children = 0;
 443         nvlist_t **child = NULL;
 444         uint_t holes;
 445         uint64_t *hole_array, max_id;
 446         uint_t c;
 447         boolean_t isactive;
 448         uint64_t hostid;
 449         nvlist_t *nvl;
 450         boolean_t found_one = B_FALSE;
 451         boolean_t valid_top_config = B_FALSE;
 452
 453         if (nvlist_alloc(&ret, 0, 0) != 0)
 454                 goto nomem;
 455
 456         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 457                 uint64_t id, max_txg = 0;
 458
 459                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 460                         goto nomem;
 461                 config_seen = B_FALSE;
 462
 463                 /*
 464                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 465                  * from the first one we find, and then go through the rest and
 466                  * add them as necessary to the 'vdevs' member of the config.
 467                  */
 468                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 469
 470                         /*
 471                          * Determine the best configuration for this vdev by
 472                          * selecting the config with the latest transaction
 473                          * group.
 474                          */
 475                         best_txg = 0;
 476                         for (ce = ve->ve_configs; ce != NULL;
 477                             ce = ce->ce_next) {
 478
 479                                 if (ce->ce_txg > best_txg) {
 480                                         tmp = ce->ce_config;
 481                                         best_txg = ce->ce_txg;
 482                                 }
 483                         }
 484
 485                         /*
 486                          * We rely on the fact that the max txg for the
 487                          * pool will contain the most up-to-date information
 488                          * about the valid top-levels in the vdev namespace.
 489                          */
 490                         if (best_txg > max_txg) {
 491                                 (void) nvlist_remove(config,
 492                                     ZPOOL_CONFIG_VDEV_CHILDREN,
 493                                     DATA_TYPE_UINT64);
 494                                 (void) nvlist_remove(config,
 495                                     ZPOOL_CONFIG_HOLE_ARRAY,
 496                                     DATA_TYPE_UINT64_ARRAY);
 497
 498                                 max_txg = best_txg;
 499                                 hole_array = NULL;
 500                                 holes = 0;
 501                                 max_id = 0;
 502                                 valid_top_config = B_FALSE;
 503
 504                                 if (nvlist_lookup_uint64(tmp,
 505                                     ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
 506                                         verify(nvlist_add_uint64(config,
 507                                             ZPOOL_CONFIG_VDEV_CHILDREN,
 508                                             max_id) == 0);
 509                                         valid_top_config = B_TRUE;
 510                                 }
 511
 512                                 if (nvlist_lookup_uint64_array(tmp,
 513                                     ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
 514                                     &holes) == 0) {
 515                                         verify(nvlist_add_uint64_array(config,
 516                                             ZPOOL_CONFIG_HOLE_ARRAY,
 517                                             hole_array, holes) == 0);
 518                                 }
 519                         }
 520
 521                         if (!config_seen) {
 522                                 /*
 523                                  * Copy the relevant pieces of data to the pool
 524                                  * configuration:
 525                                  *
 526                                  *      version
 527                                  *      pool guid
 528                                  *      name
 529                                  *      comment (if available)
 530                                  *      pool state
 531                                  *      hostid (if available)
 532                                  *      hostname (if available)
 533                                  */
 534                                 uint64_t state, version;
 535                                 char *comment = NULL;
 536
 537                                 version = fnvlist_lookup_uint64(tmp,
 538                                     ZPOOL_CONFIG_VERSION);
 539                                 fnvlist_add_uint64(config,
 540                                     ZPOOL_CONFIG_VERSION, version);
 541                                 guid = fnvlist_lookup_uint64(tmp,
 542                                     ZPOOL_CONFIG_POOL_GUID);
 543                                 fnvlist_add_uint64(config,
 544                                     ZPOOL_CONFIG_POOL_GUID, guid);
 545                                 name = fnvlist_lookup_string(tmp,
 546                                     ZPOOL_CONFIG_POOL_NAME);
 547                                 fnvlist_add_string(config,
 548                                     ZPOOL_CONFIG_POOL_NAME, name);
 549
 550                                 if (nvlist_lookup_string(tmp,
 551                                     ZPOOL_CONFIG_COMMENT, &comment) == 0)
 552                                         fnvlist_add_string(config,
 553                                             ZPOOL_CONFIG_COMMENT, comment);
 554
 555                                 state = fnvlist_lookup_uint64(tmp,
 556                                     ZPOOL_CONFIG_POOL_STATE);
 557                                 fnvlist_add_uint64(config,
 558                                     ZPOOL_CONFIG_POOL_STATE, state);
 559
 560                                 hostid = 0;
 561                                 if (nvlist_lookup_uint64(tmp,
 562                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 563                                         fnvlist_add_uint64(config,
 564                                             ZPOOL_CONFIG_HOSTID, hostid);
 565                                         hostname = fnvlist_lookup_string(tmp,
 566                                             ZPOOL_CONFIG_HOSTNAME);
 567                                         fnvlist_add_string(config,
 568                                             ZPOOL_CONFIG_HOSTNAME, hostname);
 569                                 }
 570
 571                                 config_seen = B_TRUE;
 572                         }
 573
 574                         /*
 575                          * Add this top-level vdev to the child array.
 576                          */
 577                         verify(nvlist_lookup_nvlist(tmp,
 578                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 579                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 580                             &id) == 0);
 581
 582                         if (id >= children) {
 583                                 nvlist_t **newchild;
 584
 585                                 newchild = zfs_alloc(hdl, (id + 1) *
 586                                     sizeof (nvlist_t *));
 587                                 if (newchild == NULL)
 588                                         goto nomem;
 589
 590                                 for (c = 0; c < children; c++)
 591                                         newchild[c] = child[c];
 592
 593                                 free(child);
 594                                 child = newchild;
 595                                 children = id + 1;
 596                         }
 597                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 598                                 goto nomem;
 599
 600                 }
 601
 602                 /*
 603                  * If we have information about all the top-levels then
 604                  * clean up the nvlist which we've constructed. This
 605                  * means removing any extraneous devices that are
 606                  * beyond the valid range or adding devices to the end
 607                  * of our array which appear to be missing.
 608                  */
 609                 if (valid_top_config) {
 610                         if (max_id < children) {
 611                                 for (c = max_id; c < children; c++)
 612                                         nvlist_free(child[c]);
 613                                 children = max_id;
 614                         } else if (max_id > children) {
 615                                 nvlist_t **newchild;
 616
 617                                 newchild = zfs_alloc(hdl, (max_id) *
 618                                     sizeof (nvlist_t *));
 619                                 if (newchild == NULL)
 620                                         goto nomem;
 621
 622                                 for (c = 0; c < children; c++)
 623                                         newchild[c] = child[c];
 624
 625                                 free(child);
 626                                 child = newchild;
 627                                 children = max_id;
 628                         }
 629                 }
 630
 631                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 632                     &guid) == 0);
 633
 634                 /*
 635                  * The vdev namespace may contain holes as a result of
 636                  * device removal. We must add them back into the vdev
 637                  * tree before we process any missing devices.
 638                  */
 639                 if (holes > 0) {
 640                         ASSERT(valid_top_config);
 641
 642                         for (c = 0; c < children; c++) {
 643                                 nvlist_t *holey;
 644
 645                                 if (child[c] != NULL ||
 646                                     !vdev_is_hole(hole_array, holes, c))
 647                                         continue;
 648
 649                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 650                                     0) != 0)
 651                                         goto nomem;
 652
 653                                 /*
 654                                  * Holes in the namespace are treated as
 655                                  * "hole" top-level vdevs and have a
 656                                  * special flag set on them.
 657                                  */
 658                                 if (nvlist_add_string(holey,
 659                                     ZPOOL_CONFIG_TYPE,
 660                                     VDEV_TYPE_HOLE) != 0 ||
 661                                     nvlist_add_uint64(holey,
 662                                     ZPOOL_CONFIG_ID, c) != 0 ||
 663                                     nvlist_add_uint64(holey,
 664                                     ZPOOL_CONFIG_GUID, 0ULL) != 0)
 665                                         goto nomem;
 666                                 child[c] = holey;
 667                         }
 668                 }
 669
 670                 /*
 671                  * Look for any missing top-level vdevs.  If this is the case,
 672                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 673                  * simply compress the child array, because the kernel performs
 674                  * certain checks to make sure the vdev IDs match their location
 675                  * in the configuration.
 676                  */
 677                 for (c = 0; c < children; c++) {
 678                         if (child[c] == NULL) {
 679                                 nvlist_t *missing;
 680                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 681                                     0) != 0)
 682                                         goto nomem;
 683                                 if (nvlist_add_string(missing,
 684                                     ZPOOL_CONFIG_TYPE,
 685                                     VDEV_TYPE_MISSING) != 0 ||
 686                                     nvlist_add_uint64(missing,
 687                                     ZPOOL_CONFIG_ID, c) != 0 ||
 688                                     nvlist_add_uint64(missing,
 689                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 690                                         nvlist_free(missing);
 691                                         goto nomem;
 692                                 }
 693                                 child[c] = missing;
 694                         }
 695                 }
 696
 697                 /*
 698                  * Put all of this pool's top-level vdevs into a root vdev.
 699                  */
 700                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 701                         goto nomem;
 702                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 703                     VDEV_TYPE_ROOT) != 0 ||
 704                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 705                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 706                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 707                     child, children) != 0) {
 708                         nvlist_free(nvroot);
 709                         goto nomem;
 710                 }
 711
 712                 for (c = 0; c < children; c++)
 713                         nvlist_free(child[c]);
 714                 free(child);
 715                 children = 0;
 716                 child = NULL;
 717
 718                 /*
 719                  * Go through and fix up any paths and/or devids based on our
 720                  * known list of vdev GUID -> path mappings.
 721                  */
 722                 if (fix_paths(nvroot, pl->names) != 0) {
 723                         nvlist_free(nvroot);
 724                         goto nomem;
 725                 }
 726
 727                 /*
 728                  * Add the root vdev to this pool's configuration.
 729                  */
 730                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 731                     nvroot) != 0) {
 732                         nvlist_free(nvroot);
 733                         goto nomem;
 734                 }
 735                 nvlist_free(nvroot);
 736
 737                 /*
 738                  * zdb uses this path to report on active pools that were
 739                  * imported or created using -R.
 740                  */
 741                 if (active_ok)
 742                         goto add_pool;
 743
 744                 /*
 745                  * Determine if this pool is currently active, in which case we
 746                  * can't actually import it.
 747                  */
 748                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 749                     &name) == 0);
 750                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 751                     &guid) == 0);
 752
 753                 if (pool_active(hdl, name, guid, &isactive) != 0)
 754                         goto error;
 755
 756                 if (isactive) {
 757                         nvlist_free(config);
 758                         config = NULL;
 759                         continue;
 760                 }
 761
 762                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 763                         nvlist_free(config);
 764                         config = NULL;
 765                         continue;
 766                 }
 767
 768                 nvlist_free(config);
 769                 config = nvl;
 770
 771                 /*
 772                  * Go through and update the paths for spares, now that we have
 773                  * them.
 774                  */
 775                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 776                     &nvroot) == 0);
 777                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 778                     &spares, &nspares) == 0) {
 779                         for (i = 0; i < nspares; i++) {
 780                                 if (fix_paths(spares[i], pl->names) != 0)
 781                                         goto nomem;
 782                         }
 783                 }
 784
 785                 /*
 786                  * Update the paths for l2cache devices.
 787                  */
 788                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 789                     &l2cache, &nl2cache) == 0) {
 790                         for (i = 0; i < nl2cache; i++) {
 791                                 if (fix_paths(l2cache[i], pl->names) != 0)
 792                                         goto nomem;
 793                         }
 794                 }
 795
 796                 /*
 797                  * Restore the original information read from the actual label.
 798                  */
 799                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 800                     DATA_TYPE_UINT64);
 801                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 802                     DATA_TYPE_STRING);
 803                 if (hostid != 0) {
 804                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 805                             hostid) == 0);
 806                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 807                             hostname) == 0);
 808                 }
 809
 810 add_pool:
 811                 /*
 812                  * Add this pool to the list of configs.
 813                  */
 814                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 815                     &name) == 0);
 816                 if (nvlist_add_nvlist(ret, name, config) != 0)
 817                         goto nomem;
 818
 819                 found_one = B_TRUE;
 820                 nvlist_free(config);
 821                 config = NULL;
 822         }
 823
 824         if (!found_one) {
 825                 nvlist_free(ret);
 826                 ret = NULL;
 827         }
 828
 829         return (ret);
 830
 831 nomem:
 832         (void) no_memory(hdl);
 833 error:
 834         nvlist_free(config);
 835         nvlist_free(ret);
 836         for (c = 0; c < children; c++)
 837                 nvlist_free(child[c]);
 838         free(child);
 839
 840         return (NULL);
 841 }
 842
 843 /*
 844  * Return the offset of the given label.
 845  */
 846 static uint64_t
 847 label_offset(uint64_t size, int l)
 848 {
 849         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 850         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 851             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 852 }
 853
 854 /*
 855  * Given a file descriptor, read the label information and return an nvlist
 856  * describing the configuration, if there is one.
 857  */
 858 int
 859 zpool_read_label(int fd, nvlist_t **config)
 860 {
 861         struct stat64 statbuf;
 862         int l;
 863         vdev_label_t *label;
 864         uint64_t state, txg, size;
 865
 866         *config = NULL;
 867
 868         if (fstat64(fd, &statbuf) == -1)
 869                 return (0);
 870         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 871
 872         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 873                 return (-1);
 874
 875         for (l = 0; l < VDEV_LABELS; l++) {
 876                 if (pread64(fd, label, sizeof (vdev_label_t),
 877                     label_offset(size, l)) != sizeof (vdev_label_t))
 878                         continue;
 879
 880                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 881                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 882                         continue;
 883
 884                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 885                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 886                         nvlist_free(*config);
 887                         continue;
 888                 }
 889
 890                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 891                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 892                     &txg) != 0 || txg == 0)) {
 893                         nvlist_free(*config);
 894                         continue;
 895                 }
 896
 897                 free(label);
 898                 return (0);
 899         }
 900
 901         free(label);
 902         *config = NULL;
 903         return (0);
 904 }
 905
 906 typedef struct rdsk_node {
 907         char *rn_name;
 908         int rn_dfd;
 909         libzfs_handle_t *rn_hdl;
 910         nvlist_t *rn_config;
 911         avl_tree_t *rn_avl;
 912         avl_node_t rn_node;
 913         boolean_t rn_nozpool;
 914 } rdsk_node_t;
 915
 916 static int
 917 slice_cache_compare(const void *arg1, const void *arg2)
 918 {
 919         const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
 920         const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
 921         char *nm1slice, *nm2slice;
 922         int rv;
 923
 924         /*
 925          * slices zero and two are the most likely to provide results,
 926          * so put those first
 927          */
 928         nm1slice = strstr(nm1, "s0");
 929         nm2slice = strstr(nm2, "s0");
 930         if (nm1slice && !nm2slice) {
 931                 return (-1);
 932         }
 933         if (!nm1slice && nm2slice) {
 934                 return (1);
 935         }
 936         nm1slice = strstr(nm1, "s2");
 937         nm2slice = strstr(nm2, "s2");
 938         if (nm1slice && !nm2slice) {
 939                 return (-1);
 940         }
 941         if (!nm1slice && nm2slice) {
 942                 return (1);
 943         }
 944
 945         rv = strcmp(nm1, nm2);
 946         if (rv == 0)
 947                 return (0);
 948         return (rv > 0 ? 1 : -1);
 949 }
 950
 951 #ifdef sun
 952 static void
 953 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
 954     diskaddr_t size, uint_t blksz)
 955 {
 956         rdsk_node_t tmpnode;
 957         rdsk_node_t *node;
 958         char sname[MAXNAMELEN];
 959
 960         tmpnode.rn_name = &sname[0];
 961         (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
 962             diskname, partno);
 963         /*
 964          * protect against division by zero for disk labels that
 965          * contain a bogus sector size
 966          */
 967         if (blksz == 0)
 968                 blksz = DEV_BSIZE;
 969         /* too small to contain a zpool? */
 970         if ((size < (SPA_MINDEVSIZE / blksz)) &&
 971             (node = avl_find(r, &tmpnode, NULL)))
 972                 node->rn_nozpool = B_TRUE;
 973 }
 974 #endif  /* sun */
 975
 976 static void
 977 nozpool_all_slices(avl_tree_t *r, const char *sname)
 978 {
 979 #ifdef sun
 980         char diskname[MAXNAMELEN];
 981         char *ptr;
 982         int i;
 983
 984         (void) strncpy(diskname, sname, MAXNAMELEN);
 985         if (((ptr = strrchr(diskname, 's')) == NULL) &&
 986             ((ptr = strrchr(diskname, 'p')) == NULL))
 987                 return;
 988         ptr[0] = 's';
 989         ptr[1] = '\0';
 990         for (i = 0; i < NDKMAP; i++)
 991                 check_one_slice(r, diskname, i, 0, 1);
 992         ptr[0] = 'p';
 993         for (i = 0; i <= FD_NUMPART; i++)
 994                 check_one_slice(r, diskname, i, 0, 1);
 995 #endif  /* sun */
 996 }
 997
 998 static void
 999 check_slices(avl_tree_t *r, int fd, const char *sname)
1000 {
1001 #ifdef sun
1002         struct extvtoc vtoc;
1003         struct dk_gpt *gpt;
1004         char diskname[MAXNAMELEN];
1005         char *ptr;
1006         int i;
1007
1008         (void) strncpy(diskname, sname, MAXNAMELEN);
1009         if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1010                 return;
1011         ptr[1] = '\0';
1012
1013         if (read_extvtoc(fd, &vtoc) >= 0) {
1014                 for (i = 0; i < NDKMAP; i++)
1015                         check_one_slice(r, diskname, i,
1016                             vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1017         } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1018                 /*
1019                  * on x86 we'll still have leftover links that point
1020                  * to slices s[9-15], so use NDKMAP instead
1021                  */
1022                 for (i = 0; i < NDKMAP; i++)
1023                         check_one_slice(r, diskname, i,
1024                             gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1025                 /* nodes p[1-4] are never used with EFI labels */
1026                 ptr[0] = 'p';
1027                 for (i = 1; i <= FD_NUMPART; i++)
1028                         check_one_slice(r, diskname, i, 0, 1);
1029                 efi_free(gpt);
1030         }
1031 #endif  /* sun */
1032 }
1033
1034 static void
1035 zpool_open_func(void *arg)
1036 {
1037         rdsk_node_t *rn = arg;
1038         struct stat64 statbuf;
1039         nvlist_t *config;
1040         int fd;
1041
1042         if (rn->rn_nozpool)
1043                 return;
1044         if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1045                 /* symlink to a device that's no longer there */
1046                 if (errno == ENOENT)
1047                         nozpool_all_slices(rn->rn_avl, rn->rn_name);
1048                 return;
1049         }
1050         /*
1051          * Ignore failed stats.  We only want regular
1052          * files, character devs and block devs.
1053          */
1054         if (fstat64(fd, &statbuf) != 0 ||
1055             (!S_ISREG(statbuf.st_mode) &&
1056             !S_ISCHR(statbuf.st_mode) &&
1057             !S_ISBLK(statbuf.st_mode))) {
1058                 (void) close(fd);
1059                 return;
1060         }
1061         /* this file is too small to hold a zpool */
1062         if (S_ISREG(statbuf.st_mode) &&
1063             statbuf.st_size < SPA_MINDEVSIZE) {
1064                 (void) close(fd);
1065                 return;
1066         } else if (!S_ISREG(statbuf.st_mode)) {
1067                 /*
1068                  * Try to read the disk label first so we don't have to
1069                  * open a bunch of minor nodes that can't have a zpool.
1070                  */
1071                 check_slices(rn->rn_avl, fd, rn->rn_name);
1072         }
1073
1074         if ((zpool_read_label(fd, &config)) != 0) {
1075                 (void) close(fd);
1076                 (void) no_memory(rn->rn_hdl);
1077                 return;
1078         }
1079         (void) close(fd);
1080
1081
1082         rn->rn_config = config;
1083         if (config != NULL) {
1084                 assert(rn->rn_nozpool == B_FALSE);
1085         }
1086 }
1087
1088 /*
1089  * Given a file descriptor, clear (zero) the label information.  This function
1090  * is used in the appliance stack as part of the ZFS sysevent module and
1091  * to implement the "zpool labelclear" command.
1092  */
1093 int
1094 zpool_clear_label(int fd)
1095 {
1096         struct stat64 statbuf;
1097         int l;
1098         vdev_label_t *label;
1099         uint64_t size;
1100
1101         if (fstat64(fd, &statbuf) == -1)
1102                 return (0);
1103         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1104
1105         if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1106                 return (-1);
1107
1108         for (l = 0; l < VDEV_LABELS; l++) {
1109                 if (pwrite64(fd, label, sizeof (vdev_label_t),
1110                     label_offset(size, l)) != sizeof (vdev_label_t))
1111                         return (-1);
1112         }
1113
1114         free(label);
1115         return (0);
1116 }
1117
1118 /*
1119  * Given a list of directories to search, find all pools stored on disk.  This
1120  * includes partial pools which are not available to import.  If no args are
1121  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1122  * poolname or guid (but not both) are provided by the caller when trying
1123  * to import a specific pool.
1124  */
1125 static nvlist_t *
1126 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1127 {
1128         int i, dirs = iarg->paths;
1129         DIR *dirp = NULL;
1130         struct dirent64 *dp;
1131         char path[MAXPATHLEN];
1132         char *end, **dir = iarg->path;
1133         size_t pathleft;
1134         nvlist_t *ret = NULL;
1135         static char *default_dir = "/dev";
1136         pool_list_t pools = { 0 };
1137         pool_entry_t *pe, *penext;
1138         vdev_entry_t *ve, *venext;
1139         config_entry_t *ce, *cenext;
1140         name_entry_t *ne, *nenext;
1141         avl_tree_t slice_cache;
1142         rdsk_node_t *slice;
1143         void *cookie;
1144
1145         if (dirs == 0) {
1146                 dirs = 1;
1147                 dir = &default_dir;
1148         }
1149
1150         /*
1151          * Go through and read the label configuration information from every
1152          * possible device, organizing the information according to pool GUID
1153          * and toplevel GUID.
1154          */
1155         for (i = 0; i < dirs; i++) {
1156                 tpool_t *t;
1157                 char *rdsk;
1158                 int dfd;
1159
1160                 /* use realpath to normalize the path */
1161                 if (realpath(dir[i], path) == 0) {
1162                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1163                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1164                         goto error;
1165                 }
1166                 end = &path[strlen(path)];
1167                 *end++ = '/';
1168                 *end = 0;
1169                 pathleft = &path[sizeof (path)] - end;
1170
1171                 /*
1172                  * Using raw devices instead of block devices when we're
1173                  * reading the labels skips a bunch of slow operations during
1174                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1175                  */
1176                 if (strcmp(path, "/dev/dsk/") == 0)
1177                         rdsk = "/dev/";
1178                 else
1179                         rdsk = path;
1180
1181                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1182                     (dirp = fdopendir(dfd)) == NULL) {
1183                         zfs_error_aux(hdl, strerror(errno));
1184                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1185                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1186                             rdsk);
1187                         goto error;
1188                 }
1189
1190                 avl_create(&slice_cache, slice_cache_compare,
1191                     sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1192
1193                 if (strcmp(rdsk, "/dev/") == 0) {
1194                         struct gmesh mesh;
1195                         struct gclass *mp;
1196                         struct ggeom *gp;
1197                         struct gprovider *pp;
1198
1199                         errno = geom_gettree(&mesh);
1200                         if (errno != 0) {
1201                                 zfs_error_aux(hdl, strerror(errno));
1202                                 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1203                                     dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1204                                 goto error;
1205                         }
1206
1207                         LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1208                                 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1209                                         LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1210                                                 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1211                                                 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1212                                                 slice->rn_avl = &slice_cache;
1213                                                 slice->rn_dfd = dfd;
1214                                                 slice->rn_hdl = hdl;
1215                                                 slice->rn_nozpool = B_FALSE;
1216                                                 avl_add(&slice_cache, slice);
1217                                         }
1218                                 }
1219                         }
1220
1221                         geom_deletetree(&mesh);
1222                         goto skipdir;
1223                 }
1224
1225                 /*
1226                  * This is not MT-safe, but we have no MT consumers of libzfs
1227                  */
1228                 while ((dp = readdir64(dirp)) != NULL) {
1229                         const char *name = dp->d_name;
1230                         if (name[0] == '.' &&
1231                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1232                                 continue;
1233
1234                         slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1235                         slice->rn_name = zfs_strdup(hdl, name);
1236                         slice->rn_avl = &slice_cache;
1237                         slice->rn_dfd = dfd;
1238                         slice->rn_hdl = hdl;
1239                         slice->rn_nozpool = B_FALSE;
1240                         avl_add(&slice_cache, slice);
1241                 }
1242 skipdir:
1243                 /*
1244                  * create a thread pool to do all of this in parallel;
1245                  * rn_nozpool is not protected, so this is racy in that
1246                  * multiple tasks could decide that the same slice can
1247                  * not hold a zpool, which is benign.  Also choose
1248                  * double the number of processors; we hold a lot of
1249                  * locks in the kernel, so going beyond this doesn't
1250                  * buy us much.
1251                  */
1252                 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1253                     0, NULL);
1254                 for (slice = avl_first(&slice_cache); slice;
1255                     (slice = avl_walk(&slice_cache, slice,
1256                     AVL_AFTER)))
1257                         (void) tpool_dispatch(t, zpool_open_func, slice);
1258                 tpool_wait(t);
1259                 tpool_destroy(t);
1260
1261                 cookie = NULL;
1262                 while ((slice = avl_destroy_nodes(&slice_cache,
1263                     &cookie)) != NULL) {
1264                         if (slice->rn_config != NULL) {
1265                                 nvlist_t *config = slice->rn_config;
1266                                 boolean_t matched = B_TRUE;
1267
1268                                 if (iarg->poolname != NULL) {
1269                                         char *pname;
1270
1271                                         matched = nvlist_lookup_string(config,
1272                                             ZPOOL_CONFIG_POOL_NAME,
1273                                             &pname) == 0 &&
1274                                             strcmp(iarg->poolname, pname) == 0;
1275                                 } else if (iarg->guid != 0) {
1276                                         uint64_t this_guid;
1277
1278                                         matched = nvlist_lookup_uint64(config,
1279                                             ZPOOL_CONFIG_POOL_GUID,
1280                                             &this_guid) == 0 &&
1281                                             iarg->guid == this_guid;
1282                                 }
1283                                 if (!matched) {
1284                                         nvlist_free(config);
1285                                         config = NULL;
1286                                         continue;
1287                                 }
1288                                 /* use the non-raw path for the config */
1289                                 (void) strlcpy(end, slice->rn_name, pathleft);
1290                                 if (add_config(hdl, &pools, path, config) != 0)
1291                                         goto error;
1292                         }
1293                         free(slice->rn_name);
1294                         free(slice);
1295                 }
1296                 avl_destroy(&slice_cache);
1297
1298                 (void) closedir(dirp);
1299                 dirp = NULL;
1300         }
1301
1302         ret = get_configs(hdl, &pools, iarg->can_be_active);
1303
1304 error:
1305         for (pe = pools.pools; pe != NULL; pe = penext) {
1306                 penext = pe->pe_next;
1307                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1308                         venext = ve->ve_next;
1309                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1310                                 cenext = ce->ce_next;
1311                                 if (ce->ce_config)
1312                                         nvlist_free(ce->ce_config);
1313                                 free(ce);
1314                         }
1315                         free(ve);
1316                 }
1317                 free(pe);
1318         }
1319
1320         for (ne = pools.names; ne != NULL; ne = nenext) {
1321                 nenext = ne->ne_next;
1322                 if (ne->ne_name)
1323                         free(ne->ne_name);
1324                 free(ne);
1325         }
1326
1327         if (dirp)
1328                 (void) closedir(dirp);
1329
1330         return (ret);
1331 }
1332
1333 nvlist_t *
1334 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1335 {
1336         importargs_t iarg = { 0 };
1337
1338         iarg.paths = argc;
1339         iarg.path = argv;
1340
1341         return (zpool_find_import_impl(hdl, &iarg));
1342 }
1343
1344 /*
1345  * Given a cache file, return the contents as a list of importable pools.
1346  * poolname or guid (but not both) are provided by the caller when trying
1347  * to import a specific pool.
1348  */
1349 nvlist_t *
1350 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1351     char *poolname, uint64_t guid)
1352 {
1353         char *buf;
1354         int fd;
1355         struct stat64 statbuf;
1356         nvlist_t *raw, *src, *dst;
1357         nvlist_t *pools;
1358         nvpair_t *elem;
1359         char *name;
1360         uint64_t this_guid;
1361         boolean_t active;
1362
1363         verify(poolname == NULL || guid == 0);
1364
1365         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1366                 zfs_error_aux(hdl, "%s", strerror(errno));
1367                 (void) zfs_error(hdl, EZFS_BADCACHE,
1368                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1369                 return (NULL);
1370         }
1371
1372         if (fstat64(fd, &statbuf) != 0) {
1373                 zfs_error_aux(hdl, "%s", strerror(errno));
1374                 (void) close(fd);
1375                 (void) zfs_error(hdl, EZFS_BADCACHE,
1376                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1377                 return (NULL);
1378         }
1379
1380         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1381                 (void) close(fd);
1382                 return (NULL);
1383         }
1384
1385         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1386                 (void) close(fd);
1387                 free(buf);
1388                 (void) zfs_error(hdl, EZFS_BADCACHE,
1389                     dgettext(TEXT_DOMAIN,
1390                     "failed to read cache file contents"));
1391                 return (NULL);
1392         }
1393
1394         (void) close(fd);
1395
1396         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1397                 free(buf);
1398                 (void) zfs_error(hdl, EZFS_BADCACHE,
1399                     dgettext(TEXT_DOMAIN,
1400                     "invalid or corrupt cache file contents"));
1401                 return (NULL);
1402         }
1403
1404         free(buf);
1405
1406         /*
1407          * Go through and get the current state of the pools and refresh their
1408          * state.
1409          */
1410         if (nvlist_alloc(&pools, 0, 0) != 0) {
1411                 (void) no_memory(hdl);
1412                 nvlist_free(raw);
1413                 return (NULL);
1414         }
1415
1416         elem = NULL;
1417         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1418                 verify(nvpair_value_nvlist(elem, &src) == 0);
1419
1420                 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1421                     &name) == 0);
1422                 if (poolname != NULL && strcmp(poolname, name) != 0)
1423                         continue;
1424
1425                 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1426                     &this_guid) == 0);
1427                 if (guid != 0) {
1428                         verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1429                             &this_guid) == 0);
1430                         if (guid != this_guid)
1431                                 continue;
1432                 }
1433
1434                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1435                         nvlist_free(raw);
1436                         nvlist_free(pools);
1437                         return (NULL);
1438                 }
1439
1440                 if (active)
1441                         continue;
1442
1443                 if ((dst = refresh_config(hdl, src)) == NULL) {
1444                         nvlist_free(raw);
1445                         nvlist_free(pools);
1446                         return (NULL);
1447                 }
1448
1449                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1450                         (void) no_memory(hdl);
1451                         nvlist_free(dst);
1452                         nvlist_free(raw);
1453                         nvlist_free(pools);
1454                         return (NULL);
1455                 }
1456                 nvlist_free(dst);
1457         }
1458
1459         nvlist_free(raw);
1460         return (pools);
1461 }
1462
1463 static int
1464 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1465 {
1466         importargs_t *import = data;
1467         int found = 0;
1468
1469         if (import->poolname != NULL) {
1470                 char *pool_name;
1471
1472                 verify(nvlist_lookup_string(zhp->zpool_config,
1473                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1474                 if (strcmp(pool_name, import->poolname) == 0)
1475                         found = 1;
1476         } else {
1477                 uint64_t pool_guid;
1478
1479                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1480                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1481                 if (pool_guid == import->guid)
1482                         found = 1;
1483         }
1484
1485         zpool_close(zhp);
1486         return (found);
1487 }
1488
1489 nvlist_t *
1490 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1491 {
1492         verify(import->poolname == NULL || import->guid == 0);
1493
1494         if (import->unique)
1495                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1496
1497         if (import->cachefile != NULL)
1498                 return (zpool_find_import_cached(hdl, import->cachefile,
1499                     import->poolname, import->guid));
1500
1501         return (zpool_find_import_impl(hdl, import));
1502 }
1503
1504 boolean_t
1505 find_guid(nvlist_t *nv, uint64_t guid)
1506 {
1507         uint64_t tmp;
1508         nvlist_t **child;
1509         uint_t c, children;
1510
1511         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1512         if (tmp == guid)
1513                 return (B_TRUE);
1514
1515         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1516             &child, &children) == 0) {
1517                 for (c = 0; c < children; c++)
1518                         if (find_guid(child[c], guid))
1519                                 return (B_TRUE);
1520         }
1521
1522         return (B_FALSE);
1523 }
1524
1525 typedef struct aux_cbdata {
1526         const char      *cb_type;
1527         uint64_t        cb_guid;
1528         zpool_handle_t  *cb_zhp;
1529 } aux_cbdata_t;
1530
1531 static int
1532 find_aux(zpool_handle_t *zhp, void *data)
1533 {
1534         aux_cbdata_t *cbp = data;
1535         nvlist_t **list;
1536         uint_t i, count;
1537         uint64_t guid;
1538         nvlist_t *nvroot;
1539
1540         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1541             &nvroot) == 0);
1542
1543         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1544             &list, &count) == 0) {
1545                 for (i = 0; i < count; i++) {
1546                         verify(nvlist_lookup_uint64(list[i],
1547                             ZPOOL_CONFIG_GUID, &guid) == 0);
1548                         if (guid == cbp->cb_guid) {
1549                                 cbp->cb_zhp = zhp;
1550                                 return (1);
1551                         }
1552                 }
1553         }
1554
1555         zpool_close(zhp);
1556         return (0);
1557 }
1558
1559 /*
1560  * Determines if the pool is in use.  If so, it returns true and the state of
1561  * the pool as well as the name of the pool.  Both strings are allocated and
1562  * must be freed by the caller.
1563  */
1564 int
1565 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1566     boolean_t *inuse)
1567 {
1568         nvlist_t *config;
1569         char *name;
1570         boolean_t ret;
1571         uint64_t guid, vdev_guid;
1572         zpool_handle_t *zhp;
1573         nvlist_t *pool_config;
1574         uint64_t stateval, isspare;
1575         aux_cbdata_t cb = { 0 };
1576         boolean_t isactive;
1577
1578         *inuse = B_FALSE;
1579
1580         if (zpool_read_label(fd, &config) != 0) {
1581                 (void) no_memory(hdl);
1582                 return (-1);
1583         }
1584
1585         if (config == NULL)
1586                 return (0);
1587
1588         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1589             &stateval) == 0);
1590         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1591             &vdev_guid) == 0);
1592
1593         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1594                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1595                     &name) == 0);
1596                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1597                     &guid) == 0);
1598         }
1599
1600         switch (stateval) {
1601         case POOL_STATE_EXPORTED:
1602                 /*
1603                  * A pool with an exported state may in fact be imported
1604                  * read-only, so check the in-core state to see if it's
1605                  * active and imported read-only.  If it is, set
1606                  * its state to active.
1607                  */
1608                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1609                     (zhp = zpool_open_canfail(hdl, name)) != NULL &&
1610                     zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1611                         stateval = POOL_STATE_ACTIVE;
1612
1613                 ret = B_TRUE;
1614                 break;
1615
1616         case POOL_STATE_ACTIVE:
1617                 /*
1618                  * For an active pool, we have to determine if it's really part
1619                  * of a currently active pool (in which case the pool will exist
1620                  * and the guid will be the same), or whether it's part of an
1621                  * active pool that was disconnected without being explicitly
1622                  * exported.
1623                  */
1624                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1625                         nvlist_free(config);
1626                         return (-1);
1627                 }
1628
1629                 if (isactive) {
1630                         /*
1631                          * Because the device may have been removed while
1632                          * offlined, we only report it as active if the vdev is
1633                          * still present in the config.  Otherwise, pretend like
1634                          * it's not in use.
1635                          */
1636                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1637                             (pool_config = zpool_get_config(zhp, NULL))
1638                             != NULL) {
1639                                 nvlist_t *nvroot;
1640
1641                                 verify(nvlist_lookup_nvlist(pool_config,
1642                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1643                                 ret = find_guid(nvroot, vdev_guid);
1644                         } else {
1645                                 ret = B_FALSE;
1646                         }
1647
1648                         /*
1649                          * If this is an active spare within another pool, we
1650                          * treat it like an unused hot spare.  This allows the
1651                          * user to create a pool with a hot spare that currently
1652                          * in use within another pool.  Since we return B_TRUE,
1653                          * libdiskmgt will continue to prevent generic consumers
1654                          * from using the device.
1655                          */
1656                         if (ret && nvlist_lookup_uint64(config,
1657                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1658                                 stateval = POOL_STATE_SPARE;
1659
1660                         if (zhp != NULL)
1661                                 zpool_close(zhp);
1662                 } else {
1663                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1664                         ret = B_TRUE;
1665                 }
1666                 break;
1667
1668         case POOL_STATE_SPARE:
1669                 /*
1670                  * For a hot spare, it can be either definitively in use, or
1671                  * potentially active.  To determine if it's in use, we iterate
1672                  * over all pools in the system and search for one with a spare
1673                  * with a matching guid.
1674                  *
1675                  * Due to the shared nature of spares, we don't actually report
1676                  * the potentially active case as in use.  This means the user
1677                  * can freely create pools on the hot spares of exported pools,
1678                  * but to do otherwise makes the resulting code complicated, and
1679                  * we end up having to deal with this case anyway.
1680                  */
1681                 cb.cb_zhp = NULL;
1682                 cb.cb_guid = vdev_guid;
1683                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1684                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1685                         name = (char *)zpool_get_name(cb.cb_zhp);
1686                         ret = TRUE;
1687                 } else {
1688                         ret = FALSE;
1689                 }
1690                 break;
1691
1692         case POOL_STATE_L2CACHE:
1693
1694                 /*
1695                  * Check if any pool is currently using this l2cache device.
1696                  */
1697                 cb.cb_zhp = NULL;
1698                 cb.cb_guid = vdev_guid;
1699                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1700                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1701                         name = (char *)zpool_get_name(cb.cb_zhp);
1702                         ret = TRUE;
1703                 } else {
1704                         ret = FALSE;
1705                 }
1706                 break;
1707
1708         default:
1709                 ret = B_FALSE;
1710         }
1711
1712
1713         if (ret) {
1714                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1715                         if (cb.cb_zhp)
1716                                 zpool_close(cb.cb_zhp);
1717                         nvlist_free(config);
1718                         return (-1);
1719                 }
1720                 *state = (pool_state_t)stateval;
1721         }
1722
1723         if (cb.cb_zhp)
1724                 zpool_close(cb.cb_zhp);
1725
1726         nvlist_free(config);
1727         *inuse = ret;
1728         return (0);
1729 }