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