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