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) 2011 by Delphix. All rights reserved.
  25  */
  26
  27 /*
  28  * Pool import support functions.
  29  *
  30  * To import a pool, we rely on reading the configuration information from the
  31  * ZFS label of each device.  If we successfully read the label, then we
  32  * organize the configuration information in the following hierarchy:
  33  *
  34  *      pool guid -> toplevel vdev guid -> label txg
  35  *
  36  * Duplicate entries matching this same tuple will be discarded.  Once we have
  37  * examined every device, we pick the best label txg config for each toplevel
  38  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  39  * update any paths that have changed.  Finally, we attempt to import the pool
  40  * using our derived config, and record the results.
  41  */
  42
  43 #include <ctype.h>
  44 #include <devid.h>
  45 #include <dirent.h>
  46 #include <errno.h>
  47 #include <libintl.h>
  48 #include <stddef.h>
  49 #include <stdlib.h>
  50 #include <string.h>
  51 #include <sys/stat.h>
  52 #include <unistd.h>
  53 #include <fcntl.h>
  54 #include <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, *comment;
 441         uint64_t version, 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;
 535
 536                                 verify(nvlist_lookup_uint64(tmp,
 537                                     ZPOOL_CONFIG_VERSION, &version) == 0);
 538                                 if (nvlist_add_uint64(config,
 539                                     ZPOOL_CONFIG_VERSION, version) != 0)
 540                                         goto nomem;
 541                                 verify(nvlist_lookup_uint64(tmp,
 542                                     ZPOOL_CONFIG_POOL_GUID, &guid) == 0);
 543                                 if (nvlist_add_uint64(config,
 544                                     ZPOOL_CONFIG_POOL_GUID, guid) != 0)
 545                                         goto nomem;
 546                                 verify(nvlist_lookup_string(tmp,
 547                                     ZPOOL_CONFIG_POOL_NAME, &name) == 0);
 548                                 if (nvlist_add_string(config,
 549                                     ZPOOL_CONFIG_POOL_NAME, name) != 0)
 550                                         goto nomem;
 551
 552                                 /*
 553                                  * COMMENT is optional, don't bail if it's not
 554                                  * there, instead, set it to NULL.
 555                                  */
 556                                 if (nvlist_lookup_string(tmp,
 557                                     ZPOOL_CONFIG_COMMENT, &comment) != 0)
 558                                         comment = NULL;
 559                                 else if (nvlist_add_string(config,
 560                                     ZPOOL_CONFIG_COMMENT, comment) != 0)
 561                                         goto nomem;
 562
 563                                 verify(nvlist_lookup_uint64(tmp,
 564                                     ZPOOL_CONFIG_POOL_STATE, &state) == 0);
 565                                 if (nvlist_add_uint64(config,
 566                                     ZPOOL_CONFIG_POOL_STATE, state) != 0)
 567                                         goto nomem;
 568
 569                                 hostid = 0;
 570                                 if (nvlist_lookup_uint64(tmp,
 571                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 572                                         if (nvlist_add_uint64(config,
 573                                             ZPOOL_CONFIG_HOSTID, hostid) != 0)
 574                                                 goto nomem;
 575                                         verify(nvlist_lookup_string(tmp,
 576                                             ZPOOL_CONFIG_HOSTNAME,
 577                                             &hostname) == 0);
 578                                         if (nvlist_add_string(config,
 579                                             ZPOOL_CONFIG_HOSTNAME,
 580                                             hostname) != 0)
 581                                                 goto nomem;
 582                                 }
 583
 584                                 config_seen = B_TRUE;
 585                         }
 586
 587                         /*
 588                          * Add this top-level vdev to the child array.
 589                          */
 590                         verify(nvlist_lookup_nvlist(tmp,
 591                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 592                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 593                             &id) == 0);
 594
 595                         if (id >= children) {
 596                                 nvlist_t **newchild;
 597
 598                                 newchild = zfs_alloc(hdl, (id + 1) *
 599                                     sizeof (nvlist_t *));
 600                                 if (newchild == NULL)
 601                                         goto nomem;
 602
 603                                 for (c = 0; c < children; c++)
 604                                         newchild[c] = child[c];
 605
 606                                 free(child);
 607                                 child = newchild;
 608                                 children = id + 1;
 609                         }
 610                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 611                                 goto nomem;
 612
 613                 }
 614
 615                 /*
 616                  * If we have information about all the top-levels then
 617                  * clean up the nvlist which we've constructed. This
 618                  * means removing any extraneous devices that are
 619                  * beyond the valid range or adding devices to the end
 620                  * of our array which appear to be missing.
 621                  */
 622                 if (valid_top_config) {
 623                         if (max_id < children) {
 624                                 for (c = max_id; c < children; c++)
 625                                         nvlist_free(child[c]);
 626                                 children = max_id;
 627                         } else if (max_id > children) {
 628                                 nvlist_t **newchild;
 629
 630                                 newchild = zfs_alloc(hdl, (max_id) *
 631                                     sizeof (nvlist_t *));
 632                                 if (newchild == NULL)
 633                                         goto nomem;
 634
 635                                 for (c = 0; c < children; c++)
 636                                         newchild[c] = child[c];
 637
 638                                 free(child);
 639                                 child = newchild;
 640                                 children = max_id;
 641                         }
 642                 }
 643
 644                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 645                     &guid) == 0);
 646
 647                 /*
 648                  * The vdev namespace may contain holes as a result of
 649                  * device removal. We must add them back into the vdev
 650                  * tree before we process any missing devices.
 651                  */
 652                 if (holes > 0) {
 653                         ASSERT(valid_top_config);
 654
 655                         for (c = 0; c < children; c++) {
 656                                 nvlist_t *holey;
 657
 658                                 if (child[c] != NULL ||
 659                                     !vdev_is_hole(hole_array, holes, c))
 660                                         continue;
 661
 662                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 663                                     0) != 0)
 664                                         goto nomem;
 665
 666                                 /*
 667                                  * Holes in the namespace are treated as
 668                                  * "hole" top-level vdevs and have a
 669                                  * special flag set on them.
 670                                  */
 671                                 if (nvlist_add_string(holey,
 672                                     ZPOOL_CONFIG_TYPE,
 673                                     VDEV_TYPE_HOLE) != 0 ||
 674                                     nvlist_add_uint64(holey,
 675                                     ZPOOL_CONFIG_ID, c) != 0 ||
 676                                     nvlist_add_uint64(holey,
 677                                     ZPOOL_CONFIG_GUID, 0ULL) != 0)
 678                                         goto nomem;
 679                                 child[c] = holey;
 680                         }
 681                 }
 682
 683                 /*
 684                  * Look for any missing top-level vdevs.  If this is the case,
 685                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 686                  * simply compress the child array, because the kernel performs
 687                  * certain checks to make sure the vdev IDs match their location
 688                  * in the configuration.
 689                  */
 690                 for (c = 0; c < children; c++) {
 691                         if (child[c] == NULL) {
 692                                 nvlist_t *missing;
 693                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 694                                     0) != 0)
 695                                         goto nomem;
 696                                 if (nvlist_add_string(missing,
 697                                     ZPOOL_CONFIG_TYPE,
 698                                     VDEV_TYPE_MISSING) != 0 ||
 699                                     nvlist_add_uint64(missing,
 700                                     ZPOOL_CONFIG_ID, c) != 0 ||
 701                                     nvlist_add_uint64(missing,
 702                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 703                                         nvlist_free(missing);
 704                                         goto nomem;
 705                                 }
 706                                 child[c] = missing;
 707                         }
 708                 }
 709
 710                 /*
 711                  * Put all of this pool's top-level vdevs into a root vdev.
 712                  */
 713                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 714                         goto nomem;
 715                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 716                     VDEV_TYPE_ROOT) != 0 ||
 717                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 718                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 719                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 720                     child, children) != 0) {
 721                         nvlist_free(nvroot);
 722                         goto nomem;
 723                 }
 724
 725                 for (c = 0; c < children; c++)
 726                         nvlist_free(child[c]);
 727                 free(child);
 728                 children = 0;
 729                 child = NULL;
 730
 731                 /*
 732                  * Go through and fix up any paths and/or devids based on our
 733                  * known list of vdev GUID -> path mappings.
 734                  */
 735                 if (fix_paths(nvroot, pl->names) != 0) {
 736                         nvlist_free(nvroot);
 737                         goto nomem;
 738                 }
 739
 740                 /*
 741                  * Add the root vdev to this pool's configuration.
 742                  */
 743                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 744                     nvroot) != 0) {
 745                         nvlist_free(nvroot);
 746                         goto nomem;
 747                 }
 748                 nvlist_free(nvroot);
 749
 750                 /*
 751                  * zdb uses this path to report on active pools that were
 752                  * imported or created using -R.
 753                  */
 754                 if (active_ok)
 755                         goto add_pool;
 756
 757                 /*
 758                  * Determine if this pool is currently active, in which case we
 759                  * can't actually import it.
 760                  */
 761                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 762                     &name) == 0);
 763                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 764                     &guid) == 0);
 765
 766                 if (pool_active(hdl, name, guid, &isactive) != 0)
 767                         goto error;
 768
 769                 if (isactive) {
 770                         nvlist_free(config);
 771                         config = NULL;
 772                         continue;
 773                 }
 774
 775                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 776                         nvlist_free(config);
 777                         config = NULL;
 778                         continue;
 779                 }
 780
 781                 nvlist_free(config);
 782                 config = nvl;
 783
 784                 /*
 785                  * Go through and update the paths for spares, now that we have
 786                  * them.
 787                  */
 788                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 789                     &nvroot) == 0);
 790                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 791                     &spares, &nspares) == 0) {
 792                         for (i = 0; i < nspares; i++) {
 793                                 if (fix_paths(spares[i], pl->names) != 0)
 794                                         goto nomem;
 795                         }
 796                 }
 797
 798                 /*
 799                  * Update the paths for l2cache devices.
 800                  */
 801                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 802                     &l2cache, &nl2cache) == 0) {
 803                         for (i = 0; i < nl2cache; i++) {
 804                                 if (fix_paths(l2cache[i], pl->names) != 0)
 805                                         goto nomem;
 806                         }
 807                 }
 808
 809                 /*
 810                  * Restore the original information read from the actual label.
 811                  */
 812                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 813                     DATA_TYPE_UINT64);
 814                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 815                     DATA_TYPE_STRING);
 816                 if (hostid != 0) {
 817                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 818                             hostid) == 0);
 819                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 820                             hostname) == 0);
 821                 }
 822
 823 add_pool:
 824                 /*
 825                  * Add this pool to the list of configs.
 826                  */
 827                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 828                     &name) == 0);
 829                 if (nvlist_add_nvlist(ret, name, config) != 0)
 830                         goto nomem;
 831
 832                 found_one = B_TRUE;
 833                 nvlist_free(config);
 834                 config = NULL;
 835         }
 836
 837         if (!found_one) {
 838                 nvlist_free(ret);
 839                 ret = NULL;
 840         }
 841
 842         return (ret);
 843
 844 nomem:
 845         (void) no_memory(hdl);
 846 error:
 847         nvlist_free(config);
 848         nvlist_free(ret);
 849         for (c = 0; c < children; c++)
 850                 nvlist_free(child[c]);
 851         free(child);
 852
 853         return (NULL);
 854 }
 855
 856 /*
 857  * Return the offset of the given label.
 858  */
 859 static uint64_t
 860 label_offset(uint64_t size, int l)
 861 {
 862         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 863         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 864             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 865 }
 866
 867 /*
 868  * Given a file descriptor, read the label information and return an nvlist
 869  * describing the configuration, if there is one.
 870  */
 871 int
 872 zpool_read_label(int fd, nvlist_t **config)
 873 {
 874         struct stat64 statbuf;
 875         int l;
 876         vdev_label_t *label;
 877         uint64_t state, txg, size;
 878
 879         *config = NULL;
 880
 881         if (fstat64(fd, &statbuf) == -1)
 882                 return (0);
 883         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 884
 885         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 886                 return (-1);
 887
 888         for (l = 0; l < VDEV_LABELS; l++) {
 889                 if (pread64(fd, label, sizeof (vdev_label_t),
 890                     label_offset(size, l)) != sizeof (vdev_label_t))
 891                         continue;
 892
 893                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 894                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 895                         continue;
 896
 897                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 898                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 899                         nvlist_free(*config);
 900                         continue;
 901                 }
 902
 903                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 904                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 905                     &txg) != 0 || txg == 0)) {
 906                         nvlist_free(*config);
 907                         continue;
 908                 }
 909
 910                 free(label);
 911                 return (0);
 912         }
 913
 914         free(label);
 915         *config = NULL;
 916         return (0);
 917 }
 918
 919 typedef struct rdsk_node {
 920         char *rn_name;
 921         int rn_dfd;
 922         libzfs_handle_t *rn_hdl;
 923         nvlist_t *rn_config;
 924         avl_tree_t *rn_avl;
 925         avl_node_t rn_node;
 926         boolean_t rn_nozpool;
 927 } rdsk_node_t;
 928
 929 static int
 930 slice_cache_compare(const void *arg1, const void *arg2)
 931 {
 932         const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
 933         const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
 934         char *nm1slice, *nm2slice;
 935         int rv;
 936
 937         /*
 938          * slices zero and two are the most likely to provide results,
 939          * so put those first
 940          */
 941         nm1slice = strstr(nm1, "s0");
 942         nm2slice = strstr(nm2, "s0");
 943         if (nm1slice && !nm2slice) {
 944                 return (-1);
 945         }
 946         if (!nm1slice && nm2slice) {
 947                 return (1);
 948         }
 949         nm1slice = strstr(nm1, "s2");
 950         nm2slice = strstr(nm2, "s2");
 951         if (nm1slice && !nm2slice) {
 952                 return (-1);
 953         }
 954         if (!nm1slice && nm2slice) {
 955                 return (1);
 956         }
 957
 958         rv = strcmp(nm1, nm2);
 959         if (rv == 0)
 960                 return (0);
 961         return (rv > 0 ? 1 : -1);
 962 }
 963
 964 #ifdef sun
 965 static void
 966 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
 967     diskaddr_t size, uint_t blksz)
 968 {
 969         rdsk_node_t tmpnode;
 970         rdsk_node_t *node;
 971         char sname[MAXNAMELEN];
 972
 973         tmpnode.rn_name = &sname[0];
 974         (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
 975             diskname, partno);
 976         /*
 977          * protect against division by zero for disk labels that
 978          * contain a bogus sector size
 979          */
 980         if (blksz == 0)
 981                 blksz = DEV_BSIZE;
 982         /* too small to contain a zpool? */
 983         if ((size < (SPA_MINDEVSIZE / blksz)) &&
 984             (node = avl_find(r, &tmpnode, NULL)))
 985                 node->rn_nozpool = B_TRUE;
 986 }
 987 #endif  /* sun */
 988
 989 static void
 990 nozpool_all_slices(avl_tree_t *r, const char *sname)
 991 {
 992 #ifdef sun
 993         char diskname[MAXNAMELEN];
 994         char *ptr;
 995         int i;
 996
 997         (void) strncpy(diskname, sname, MAXNAMELEN);
 998         if (((ptr = strrchr(diskname, 's')) == NULL) &&
 999             ((ptr = strrchr(diskname, 'p')) == NULL))
1000                 return;
1001         ptr[0] = 's';
1002         ptr[1] = '\0';
1003         for (i = 0; i < NDKMAP; i++)
1004                 check_one_slice(r, diskname, i, 0, 1);
1005         ptr[0] = 'p';
1006         for (i = 0; i <= FD_NUMPART; i++)
1007                 check_one_slice(r, diskname, i, 0, 1);
1008 #endif  /* sun */
1009 }
1010
1011 static void
1012 check_slices(avl_tree_t *r, int fd, const char *sname)
1013 {
1014 #ifdef sun
1015         struct extvtoc vtoc;
1016         struct dk_gpt *gpt;
1017         char diskname[MAXNAMELEN];
1018         char *ptr;
1019         int i;
1020
1021         (void) strncpy(diskname, sname, MAXNAMELEN);
1022         if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1023                 return;
1024         ptr[1] = '\0';
1025
1026         if (read_extvtoc(fd, &vtoc) >= 0) {
1027                 for (i = 0; i < NDKMAP; i++)
1028                         check_one_slice(r, diskname, i,
1029                             vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1030         } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1031                 /*
1032                  * on x86 we'll still have leftover links that point
1033                  * to slices s[9-15], so use NDKMAP instead
1034                  */
1035                 for (i = 0; i < NDKMAP; i++)
1036                         check_one_slice(r, diskname, i,
1037                             gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1038                 /* nodes p[1-4] are never used with EFI labels */
1039                 ptr[0] = 'p';
1040                 for (i = 1; i <= FD_NUMPART; i++)
1041                         check_one_slice(r, diskname, i, 0, 1);
1042                 efi_free(gpt);
1043         }
1044 #endif  /* sun */
1045 }
1046
1047 static void
1048 zpool_open_func(void *arg)
1049 {
1050         rdsk_node_t *rn = arg;
1051         struct stat64 statbuf;
1052         nvlist_t *config;
1053         int fd;
1054
1055         if (rn->rn_nozpool)
1056                 return;
1057         if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1058                 /* symlink to a device that's no longer there */
1059                 if (errno == ENOENT)
1060                         nozpool_all_slices(rn->rn_avl, rn->rn_name);
1061                 return;
1062         }
1063         /*
1064          * Ignore failed stats.  We only want regular
1065          * files, character devs and block devs.
1066          */
1067         if (fstat64(fd, &statbuf) != 0 ||
1068             (!S_ISREG(statbuf.st_mode) &&
1069             !S_ISCHR(statbuf.st_mode) &&
1070             !S_ISBLK(statbuf.st_mode))) {
1071                 (void) close(fd);
1072                 return;
1073         }
1074         /* this file is too small to hold a zpool */
1075         if (S_ISREG(statbuf.st_mode) &&
1076             statbuf.st_size < SPA_MINDEVSIZE) {
1077                 (void) close(fd);
1078                 return;
1079         } else if (!S_ISREG(statbuf.st_mode)) {
1080                 /*
1081                  * Try to read the disk label first so we don't have to
1082                  * open a bunch of minor nodes that can't have a zpool.
1083                  */
1084                 check_slices(rn->rn_avl, fd, rn->rn_name);
1085         }
1086
1087         if ((zpool_read_label(fd, &config)) != 0) {
1088                 (void) close(fd);
1089                 (void) no_memory(rn->rn_hdl);
1090                 return;
1091         }
1092         (void) close(fd);
1093
1094
1095         rn->rn_config = config;
1096         if (config != NULL) {
1097                 assert(rn->rn_nozpool == B_FALSE);
1098         }
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                         return (-1);
1125         }
1126
1127         free(label);
1128         return (0);
1129 }
1130
1131 /*
1132  * Given a list of directories to search, find all pools stored on disk.  This
1133  * includes partial pools which are not available to import.  If no args are
1134  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1135  * poolname or guid (but not both) are provided by the caller when trying
1136  * to import a specific pool.
1137  */
1138 static nvlist_t *
1139 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1140 {
1141         int i, dirs = iarg->paths;
1142         DIR *dirp = NULL;
1143         struct dirent64 *dp;
1144         char path[MAXPATHLEN];
1145         char *end, **dir = iarg->path;
1146         size_t pathleft;
1147         nvlist_t *ret = NULL;
1148         static char *default_dir = "/dev";
1149         pool_list_t pools = { 0 };
1150         pool_entry_t *pe, *penext;
1151         vdev_entry_t *ve, *venext;
1152         config_entry_t *ce, *cenext;
1153         name_entry_t *ne, *nenext;
1154         avl_tree_t slice_cache;
1155         rdsk_node_t *slice;
1156         void *cookie;
1157
1158         if (dirs == 0) {
1159                 dirs = 1;
1160                 dir = &default_dir;
1161         }
1162
1163         /*
1164          * Go through and read the label configuration information from every
1165          * possible device, organizing the information according to pool GUID
1166          * and toplevel GUID.
1167          */
1168         for (i = 0; i < dirs; i++) {
1169                 tpool_t *t;
1170                 char *rdsk;
1171                 int dfd;
1172
1173                 /* use realpath to normalize the path */
1174                 if (realpath(dir[i], path) == 0) {
1175                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1176                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1177                         goto error;
1178                 }
1179                 end = &path[strlen(path)];
1180                 *end++ = '/';
1181                 *end = 0;
1182                 pathleft = &path[sizeof (path)] - end;
1183
1184                 /*
1185                  * Using raw devices instead of block devices when we're
1186                  * reading the labels skips a bunch of slow operations during
1187                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1188                  */
1189                 if (strcmp(path, "/dev/dsk/") == 0)
1190                         rdsk = "/dev/";
1191                 else
1192                         rdsk = path;
1193
1194                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1195                     (dirp = fdopendir(dfd)) == NULL) {
1196                         zfs_error_aux(hdl, strerror(errno));
1197                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1198                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1199                             rdsk);
1200                         goto error;
1201                 }
1202
1203                 avl_create(&slice_cache, slice_cache_compare,
1204                     sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1205
1206                 if (strcmp(rdsk, "/dev/") == 0) {
1207                         struct gmesh mesh;
1208                         struct gclass *mp;
1209                         struct ggeom *gp;
1210                         struct gprovider *pp;
1211
1212                         errno = geom_gettree(&mesh);
1213                         if (errno != 0) {
1214                                 zfs_error_aux(hdl, strerror(errno));
1215                                 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1216                                     dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1217                                 goto error;
1218                         }
1219
1220                         LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1221                                 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1222                                         LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1223                                                 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1224                                                 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1225                                                 slice->rn_avl = &slice_cache;
1226                                                 slice->rn_dfd = dfd;
1227                                                 slice->rn_hdl = hdl;
1228                                                 slice->rn_nozpool = B_FALSE;
1229                                                 avl_add(&slice_cache, slice);
1230                                         }
1231                                 }
1232                         }
1233
1234                         geom_deletetree(&mesh);
1235                         goto skipdir;
1236                 }
1237
1238                 /*
1239                  * This is not MT-safe, but we have no MT consumers of libzfs
1240                  */
1241                 while ((dp = readdir64(dirp)) != NULL) {
1242                         const char *name = dp->d_name;
1243                         if (name[0] == '.' &&
1244                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1245                                 continue;
1246
1247                         slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1248                         slice->rn_name = zfs_strdup(hdl, name);
1249                         slice->rn_avl = &slice_cache;
1250                         slice->rn_dfd = dfd;
1251                         slice->rn_hdl = hdl;
1252                         slice->rn_nozpool = B_FALSE;
1253                         avl_add(&slice_cache, slice);
1254                 }
1255 skipdir:
1256                 /*
1257                  * create a thread pool to do all of this in parallel;
1258                  * rn_nozpool is not protected, so this is racy in that
1259                  * multiple tasks could decide that the same slice can
1260                  * not hold a zpool, which is benign.  Also choose
1261                  * double the number of processors; we hold a lot of
1262                  * locks in the kernel, so going beyond this doesn't
1263                  * buy us much.
1264                  */
1265                 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1266                     0, NULL);
1267                 for (slice = avl_first(&slice_cache); slice;
1268                     (slice = avl_walk(&slice_cache, slice,
1269                     AVL_AFTER)))
1270                         (void) tpool_dispatch(t, zpool_open_func, slice);
1271                 tpool_wait(t);
1272                 tpool_destroy(t);
1273
1274                 cookie = NULL;
1275                 while ((slice = avl_destroy_nodes(&slice_cache,
1276                     &cookie)) != NULL) {
1277                         if (slice->rn_config != NULL) {
1278                                 nvlist_t *config = slice->rn_config;
1279                                 boolean_t matched = B_TRUE;
1280
1281                                 if (iarg->poolname != NULL) {
1282                                         char *pname;
1283
1284                                         matched = nvlist_lookup_string(config,
1285                                             ZPOOL_CONFIG_POOL_NAME,
1286                                             &pname) == 0 &&
1287                                             strcmp(iarg->poolname, pname) == 0;
1288                                 } else if (iarg->guid != 0) {
1289                                         uint64_t this_guid;
1290
1291                                         matched = nvlist_lookup_uint64(config,
1292                                             ZPOOL_CONFIG_POOL_GUID,
1293                                             &this_guid) == 0 &&
1294                                             iarg->guid == this_guid;
1295                                 }
1296                                 if (!matched) {
1297                                         nvlist_free(config);
1298                                         config = NULL;
1299                                         continue;
1300                                 }
1301                                 /* use the non-raw path for the config */
1302                                 (void) strlcpy(end, slice->rn_name, pathleft);
1303                                 if (add_config(hdl, &pools, path, config) != 0)
1304                                         goto error;
1305                         }
1306                         free(slice->rn_name);
1307                         free(slice);
1308                 }
1309                 avl_destroy(&slice_cache);
1310
1311                 (void) closedir(dirp);
1312                 dirp = NULL;
1313         }
1314
1315         ret = get_configs(hdl, &pools, iarg->can_be_active);
1316
1317 error:
1318         for (pe = pools.pools; pe != NULL; pe = penext) {
1319                 penext = pe->pe_next;
1320                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1321                         venext = ve->ve_next;
1322                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1323                                 cenext = ce->ce_next;
1324                                 if (ce->ce_config)
1325                                         nvlist_free(ce->ce_config);
1326                                 free(ce);
1327                         }
1328                         free(ve);
1329                 }
1330                 free(pe);
1331         }
1332
1333         for (ne = pools.names; ne != NULL; ne = nenext) {
1334                 nenext = ne->ne_next;
1335                 if (ne->ne_name)
1336                         free(ne->ne_name);
1337                 free(ne);
1338         }
1339
1340         if (dirp)
1341                 (void) closedir(dirp);
1342
1343         return (ret);
1344 }
1345
1346 nvlist_t *
1347 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1348 {
1349         importargs_t iarg = { 0 };
1350
1351         iarg.paths = argc;
1352         iarg.path = argv;
1353
1354         return (zpool_find_import_impl(hdl, &iarg));
1355 }
1356
1357 /*
1358  * Given a cache file, return the contents as a list of importable pools.
1359  * poolname or guid (but not both) are provided by the caller when trying
1360  * to import a specific pool.
1361  */
1362 nvlist_t *
1363 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1364     char *poolname, uint64_t guid)
1365 {
1366         char *buf;
1367         int fd;
1368         struct stat64 statbuf;
1369         nvlist_t *raw, *src, *dst;
1370         nvlist_t *pools;
1371         nvpair_t *elem;
1372         char *name;
1373         uint64_t this_guid;
1374         boolean_t active;
1375
1376         verify(poolname == NULL || guid == 0);
1377
1378         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1379                 zfs_error_aux(hdl, "%s", strerror(errno));
1380                 (void) zfs_error(hdl, EZFS_BADCACHE,
1381                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1382                 return (NULL);
1383         }
1384
1385         if (fstat64(fd, &statbuf) != 0) {
1386                 zfs_error_aux(hdl, "%s", strerror(errno));
1387                 (void) close(fd);
1388                 (void) zfs_error(hdl, EZFS_BADCACHE,
1389                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1390                 return (NULL);
1391         }
1392
1393         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1394                 (void) close(fd);
1395                 return (NULL);
1396         }
1397
1398         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1399                 (void) close(fd);
1400                 free(buf);
1401                 (void) zfs_error(hdl, EZFS_BADCACHE,
1402                     dgettext(TEXT_DOMAIN,
1403                     "failed to read cache file contents"));
1404                 return (NULL);
1405         }
1406
1407         (void) close(fd);
1408
1409         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1410                 free(buf);
1411                 (void) zfs_error(hdl, EZFS_BADCACHE,
1412                     dgettext(TEXT_DOMAIN,
1413                     "invalid or corrupt cache file contents"));
1414                 return (NULL);
1415         }
1416
1417         free(buf);
1418
1419         /*
1420          * Go through and get the current state of the pools and refresh their
1421          * state.
1422          */
1423         if (nvlist_alloc(&pools, 0, 0) != 0) {
1424                 (void) no_memory(hdl);
1425                 nvlist_free(raw);
1426                 return (NULL);
1427         }
1428
1429         elem = NULL;
1430         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1431                 verify(nvpair_value_nvlist(elem, &src) == 0);
1432
1433                 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1434                     &name) == 0);
1435                 if (poolname != NULL && strcmp(poolname, name) != 0)
1436                         continue;
1437
1438                 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1439                     &this_guid) == 0);
1440                 if (guid != 0) {
1441                         verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1442                             &this_guid) == 0);
1443                         if (guid != this_guid)
1444                                 continue;
1445                 }
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                     zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1624                         stateval = POOL_STATE_ACTIVE;
1625
1626                 ret = B_TRUE;
1627                 break;
1628
1629         case POOL_STATE_ACTIVE:
1630                 /*
1631                  * For an active pool, we have to determine if it's really part
1632                  * of a currently active pool (in which case the pool will exist
1633                  * and the guid will be the same), or whether it's part of an
1634                  * active pool that was disconnected without being explicitly
1635                  * exported.
1636                  */
1637                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1638                         nvlist_free(config);
1639                         return (-1);
1640                 }
1641
1642                 if (isactive) {
1643                         /*
1644                          * Because the device may have been removed while
1645                          * offlined, we only report it as active if the vdev is
1646                          * still present in the config.  Otherwise, pretend like
1647                          * it's not in use.
1648                          */
1649                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1650                             (pool_config = zpool_get_config(zhp, NULL))
1651                             != NULL) {
1652                                 nvlist_t *nvroot;
1653
1654                                 verify(nvlist_lookup_nvlist(pool_config,
1655                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1656                                 ret = find_guid(nvroot, vdev_guid);
1657                         } else {
1658                                 ret = B_FALSE;
1659                         }
1660
1661                         /*
1662                          * If this is an active spare within another pool, we
1663                          * treat it like an unused hot spare.  This allows the
1664                          * user to create a pool with a hot spare that currently
1665                          * in use within another pool.  Since we return B_TRUE,
1666                          * libdiskmgt will continue to prevent generic consumers
1667                          * from using the device.
1668                          */
1669                         if (ret && nvlist_lookup_uint64(config,
1670                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1671                                 stateval = POOL_STATE_SPARE;
1672
1673                         if (zhp != NULL)
1674                                 zpool_close(zhp);
1675                 } else {
1676                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1677                         ret = B_TRUE;
1678                 }
1679                 break;
1680
1681         case POOL_STATE_SPARE:
1682                 /*
1683                  * For a hot spare, it can be either definitively in use, or
1684                  * potentially active.  To determine if it's in use, we iterate
1685                  * over all pools in the system and search for one with a spare
1686                  * with a matching guid.
1687                  *
1688                  * Due to the shared nature of spares, we don't actually report
1689                  * the potentially active case as in use.  This means the user
1690                  * can freely create pools on the hot spares of exported pools,
1691                  * but to do otherwise makes the resulting code complicated, and
1692                  * we end up having to deal with this case anyway.
1693                  */
1694                 cb.cb_zhp = NULL;
1695                 cb.cb_guid = vdev_guid;
1696                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1697                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1698                         name = (char *)zpool_get_name(cb.cb_zhp);
1699                         ret = TRUE;
1700                 } else {
1701                         ret = FALSE;
1702                 }
1703                 break;
1704
1705         case POOL_STATE_L2CACHE:
1706
1707                 /*
1708                  * Check if any pool is currently using this l2cache device.
1709                  */
1710                 cb.cb_zhp = NULL;
1711                 cb.cb_guid = vdev_guid;
1712                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1713                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1714                         name = (char *)zpool_get_name(cb.cb_zhp);
1715                         ret = TRUE;
1716                 } else {
1717                         ret = FALSE;
1718                 }
1719                 break;
1720
1721         default:
1722                 ret = B_FALSE;
1723         }
1724
1725
1726         if (ret) {
1727                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1728                         if (cb.cb_zhp)
1729                                 zpool_close(cb.cb_zhp);
1730                         nvlist_free(config);
1731                         return (-1);
1732                 }
1733                 *state = (pool_state_t)stateval;
1734         }
1735
1736         if (cb.cb_zhp)
1737                 zpool_close(cb.cb_zhp);
1738
1739         nvlist_free(config);
1740         *inuse = ret;
1741         return (0);
1742 }