cddl/contrib/opensolaris/lib/libzfs/common/libzfs_import.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
  25  * Copyright 2015 RackTop Systems.
  26  * Copyright 2016 Nexenta Systems, Inc.
  27  */
  28
  29 /*
  30  * Pool import support functions.
  31  *
  32  * To import a pool, we rely on reading the configuration information from the
  33  * ZFS label of each device.  If we successfully read the label, then we
  34  * organize the configuration information in the following hierarchy:
  35  *
  36  *      pool guid -> toplevel vdev guid -> label txg
  37  *
  38  * Duplicate entries matching this same tuple will be discarded.  Once we have
  39  * examined every device, we pick the best label txg config for each toplevel
  40  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  41  * update any paths that have changed.  Finally, we attempt to import the pool
  42  * using our derived config, and record the results.
  43  */
  44
  45 #include <aio.h>
  46 #include <ctype.h>
  47 #include <devid.h>
  48 #include <dirent.h>
  49 #include <errno.h>
  50 #include <libintl.h>
  51 #include <stddef.h>
  52 #include <stdlib.h>
  53 #include <string.h>
  54 #include <sys/stat.h>
  55 #include <unistd.h>
  56 #include <fcntl.h>
  57 #include <thread_pool.h>
  58 #include <libgeom.h>
  59
  60 #include <sys/vdev_impl.h>
  61
  62 #include "libzfs.h"
  63 #include "libzfs_impl.h"
  64
  65 /*
  66  * Intermediate structures used to gather configuration information.
  67  */
  68 typedef struct config_entry {
  69         uint64_t                ce_txg;
  70         nvlist_t                *ce_config;
  71         struct config_entry     *ce_next;
  72 } config_entry_t;
  73
  74 typedef struct vdev_entry {
  75         uint64_t                ve_guid;
  76         config_entry_t          *ve_configs;
  77         struct vdev_entry       *ve_next;
  78 } vdev_entry_t;
  79
  80 typedef struct pool_entry {
  81         uint64_t                pe_guid;
  82         vdev_entry_t            *pe_vdevs;
  83         struct pool_entry       *pe_next;
  84 } pool_entry_t;
  85
  86 typedef struct name_entry {
  87         char                    *ne_name;
  88         uint64_t                ne_guid;
  89         struct name_entry       *ne_next;
  90 } name_entry_t;
  91
  92 typedef struct pool_list {
  93         pool_entry_t            *pools;
  94         name_entry_t            *names;
  95 } pool_list_t;
  96
  97 static char *
  98 get_devid(const char *path)
  99 {
 100 #ifdef have_devid
 101         int fd;
 102         ddi_devid_t devid;
 103         char *minor, *ret;
 104
 105         if ((fd = open(path, O_RDONLY)) < 0)
 106                 return (NULL);
 107
 108         minor = NULL;
 109         ret = NULL;
 110         if (devid_get(fd, &devid) == 0) {
 111                 if (devid_get_minor_name(fd, &minor) == 0)
 112                         ret = devid_str_encode(devid, minor);
 113                 if (minor != NULL)
 114                         devid_str_free(minor);
 115                 devid_free(devid);
 116         }
 117         (void) close(fd);
 118
 119         return (ret);
 120 #else
 121         return (NULL);
 122 #endif
 123 }
 124
 125
 126 /*
 127  * Go through and fix up any path and/or devid information for the given vdev
 128  * configuration.
 129  */
 130 static int
 131 fix_paths(nvlist_t *nv, name_entry_t *names)
 132 {
 133         nvlist_t **child;
 134         uint_t c, children;
 135         uint64_t guid;
 136         name_entry_t *ne, *best;
 137         char *path, *devid;
 138         int matched;
 139
 140         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 141             &child, &children) == 0) {
 142                 for (c = 0; c < children; c++)
 143                         if (fix_paths(child[c], names) != 0)
 144                                 return (-1);
 145                 return (0);
 146         }
 147
 148         /*
 149          * This is a leaf (file or disk) vdev.  In either case, go through
 150          * the name list and see if we find a matching guid.  If so, replace
 151          * the path and see if we can calculate a new devid.
 152          *
 153          * There may be multiple names associated with a particular guid, in
 154          * which case we have overlapping slices or multiple paths to the same
 155          * disk.  If this is the case, then we want to pick the path that is
 156          * the most similar to the original, where "most similar" is the number
 157          * of matching characters starting from the end of the path.  This will
 158          * preserve slice numbers even if the disks have been reorganized, and
 159          * will also catch preferred disk names if multiple paths exist.
 160          */
 161         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 162         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 163                 path = NULL;
 164
 165         matched = 0;
 166         best = NULL;
 167         for (ne = names; ne != NULL; ne = ne->ne_next) {
 168                 if (ne->ne_guid == guid) {
 169                         const char *src, *dst;
 170                         int count;
 171
 172                         if (path == NULL) {
 173                                 best = ne;
 174                                 break;
 175                         }
 176
 177                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 178                         dst = path + strlen(path) - 1;
 179                         for (count = 0; src >= ne->ne_name && dst >= path;
 180                             src--, dst--, count++)
 181                                 if (*src != *dst)
 182                                         break;
 183
 184                         /*
 185                          * At this point, 'count' is the number of characters
 186                          * matched from the end.
 187                          */
 188                         if (count > matched || best == NULL) {
 189                                 best = ne;
 190                                 matched = count;
 191                         }
 192                 }
 193         }
 194
 195         if (best == NULL)
 196                 return (0);
 197
 198         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 199                 return (-1);
 200
 201         if ((devid = get_devid(best->ne_name)) == NULL) {
 202                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 203         } else {
 204                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
 205                         devid_str_free(devid);
 206                         return (-1);
 207                 }
 208                 devid_str_free(devid);
 209         }
 210
 211         return (0);
 212 }
 213
 214 /*
 215  * Add the given configuration to the list of known devices.
 216  */
 217 static int
 218 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 219     nvlist_t *config)
 220 {
 221         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 222         pool_entry_t *pe;
 223         vdev_entry_t *ve;
 224         config_entry_t *ce;
 225         name_entry_t *ne;
 226
 227         /*
 228          * If this is a hot spare not currently in use or level 2 cache
 229          * device, add it to the list of names to translate, but don't do
 230          * anything else.
 231          */
 232         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 233             &state) == 0 &&
 234             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 235             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 236                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 237                         return (-1);
 238
 239                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 240                         free(ne);
 241                         return (-1);
 242                 }
 243                 ne->ne_guid = vdev_guid;
 244                 ne->ne_next = pl->names;
 245                 pl->names = ne;
 246                 return (0);
 247         }
 248
 249         /*
 250          * If we have a valid config but cannot read any of these fields, then
 251          * it means we have a half-initialized label.  In vdev_label_init()
 252          * we write a label with txg == 0 so that we can identify the device
 253          * in case the user refers to the same disk later on.  If we fail to
 254          * create the pool, we'll be left with a label in this state
 255          * which should not be considered part of a valid pool.
 256          */
 257         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 258             &pool_guid) != 0 ||
 259             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 260             &vdev_guid) != 0 ||
 261             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 262             &top_guid) != 0 ||
 263             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 264             &txg) != 0 || txg == 0) {
 265                 nvlist_free(config);
 266                 return (0);
 267         }
 268
 269         /*
 270          * First, see if we know about this pool.  If not, then add it to the
 271          * list of known pools.
 272          */
 273         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 274                 if (pe->pe_guid == pool_guid)
 275                         break;
 276         }
 277
 278         if (pe == NULL) {
 279                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 280                         nvlist_free(config);
 281                         return (-1);
 282                 }
 283                 pe->pe_guid = pool_guid;
 284                 pe->pe_next = pl->pools;
 285                 pl->pools = pe;
 286         }
 287
 288         /*
 289          * Second, see if we know about this toplevel vdev.  Add it if its
 290          * missing.
 291          */
 292         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 293                 if (ve->ve_guid == top_guid)
 294                         break;
 295         }
 296
 297         if (ve == NULL) {
 298                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 299                         nvlist_free(config);
 300                         return (-1);
 301                 }
 302                 ve->ve_guid = top_guid;
 303                 ve->ve_next = pe->pe_vdevs;
 304                 pe->pe_vdevs = ve;
 305         }
 306
 307         /*
 308          * Third, see if we have a config with a matching transaction group.  If
 309          * so, then we do nothing.  Otherwise, add it to the list of known
 310          * configs.
 311          */
 312         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 313                 if (ce->ce_txg == txg)
 314                         break;
 315         }
 316
 317         if (ce == NULL) {
 318                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 319                         nvlist_free(config);
 320                         return (-1);
 321                 }
 322                 ce->ce_txg = txg;
 323                 ce->ce_config = config;
 324                 ce->ce_next = ve->ve_configs;
 325                 ve->ve_configs = ce;
 326         } else {
 327                 nvlist_free(config);
 328         }
 329
 330         /*
 331          * At this point we've successfully added our config to the list of
 332          * known configs.  The last thing to do is add the vdev guid -> path
 333          * mappings so that we can fix up the configuration as necessary before
 334          * doing the import.
 335          */
 336         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 337                 return (-1);
 338
 339         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 340                 free(ne);
 341                 return (-1);
 342         }
 343
 344         ne->ne_guid = vdev_guid;
 345         ne->ne_next = pl->names;
 346         pl->names = ne;
 347
 348         return (0);
 349 }
 350
 351 /*
 352  * Returns true if the named pool matches the given GUID.
 353  */
 354 static int
 355 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 356     boolean_t *isactive)
 357 {
 358         zpool_handle_t *zhp;
 359         uint64_t theguid;
 360
 361         if (zpool_open_silent(hdl, name, &zhp) != 0)
 362                 return (-1);
 363
 364         if (zhp == NULL) {
 365                 *isactive = B_FALSE;
 366                 return (0);
 367         }
 368
 369         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 370             &theguid) == 0);
 371
 372         zpool_close(zhp);
 373
 374         *isactive = (theguid == guid);
 375         return (0);
 376 }
 377
 378 static nvlist_t *
 379 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 380 {
 381         nvlist_t *nvl;
 382         zfs_cmd_t zc = { 0 };
 383         int err, dstbuf_size;
 384
 385         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 386                 return (NULL);
 387
 388         dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
 389
 390         if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
 391                 zcmd_free_nvlists(&zc);
 392                 return (NULL);
 393         }
 394
 395         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 396             &zc)) != 0 && errno == ENOMEM) {
 397                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 398                         zcmd_free_nvlists(&zc);
 399                         return (NULL);
 400                 }
 401         }
 402
 403         if (err) {
 404                 zcmd_free_nvlists(&zc);
 405                 return (NULL);
 406         }
 407
 408         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 409                 zcmd_free_nvlists(&zc);
 410                 return (NULL);
 411         }
 412
 413         zcmd_free_nvlists(&zc);
 414         return (nvl);
 415 }
 416
 417 /*
 418  * Determine if the vdev id is a hole in the namespace.
 419  */
 420 boolean_t
 421 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
 422 {
 423         for (int c = 0; c < holes; c++) {
 424
 425                 /* Top-level is a hole */
 426                 if (hole_array[c] == id)
 427                         return (B_TRUE);
 428         }
 429         return (B_FALSE);
 430 }
 431
 432 /*
 433  * Convert our list of pools into the definitive set of configurations.  We
 434  * start by picking the best config for each toplevel vdev.  Once that's done,
 435  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 436  * pass to fix up any incorrect paths, and then add it to the main list to
 437  * return to the user.
 438  */
 439 static nvlist_t *
 440 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 441 {
 442         pool_entry_t *pe;
 443         vdev_entry_t *ve;
 444         config_entry_t *ce;
 445         nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
 446         nvlist_t **spares, **l2cache;
 447         uint_t i, nspares, nl2cache;
 448         boolean_t config_seen;
 449         uint64_t best_txg;
 450         char *name, *hostname = NULL;
 451         uint64_t guid;
 452         uint_t children = 0;
 453         nvlist_t **child = NULL;
 454         uint_t holes;
 455         uint64_t *hole_array, max_id;
 456         uint_t c;
 457         boolean_t isactive;
 458         uint64_t hostid;
 459         nvlist_t *nvl;
 460         boolean_t found_one = B_FALSE;
 461         boolean_t valid_top_config = B_FALSE;
 462
 463         if (nvlist_alloc(&ret, 0, 0) != 0)
 464                 goto nomem;
 465
 466         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 467                 uint64_t id, max_txg = 0;
 468
 469                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 470                         goto nomem;
 471                 config_seen = B_FALSE;
 472
 473                 /*
 474                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 475                  * from the first one we find, and then go through the rest and
 476                  * add them as necessary to the 'vdevs' member of the config.
 477                  */
 478                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 479
 480                         /*
 481                          * Determine the best configuration for this vdev by
 482                          * selecting the config with the latest transaction
 483                          * group.
 484                          */
 485                         best_txg = 0;
 486                         for (ce = ve->ve_configs; ce != NULL;
 487                             ce = ce->ce_next) {
 488
 489                                 if (ce->ce_txg > best_txg) {
 490                                         tmp = ce->ce_config;
 491                                         best_txg = ce->ce_txg;
 492                                 }
 493                         }
 494
 495                         /*
 496                          * We rely on the fact that the max txg for the
 497                          * pool will contain the most up-to-date information
 498                          * about the valid top-levels in the vdev namespace.
 499                          */
 500                         if (best_txg > max_txg) {
 501                                 (void) nvlist_remove(config,
 502                                     ZPOOL_CONFIG_VDEV_CHILDREN,
 503                                     DATA_TYPE_UINT64);
 504                                 (void) nvlist_remove(config,
 505                                     ZPOOL_CONFIG_HOLE_ARRAY,
 506                                     DATA_TYPE_UINT64_ARRAY);
 507
 508                                 max_txg = best_txg;
 509                                 hole_array = NULL;
 510                                 holes = 0;
 511                                 max_id = 0;
 512                                 valid_top_config = B_FALSE;
 513
 514                                 if (nvlist_lookup_uint64(tmp,
 515                                     ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
 516                                         verify(nvlist_add_uint64(config,
 517                                             ZPOOL_CONFIG_VDEV_CHILDREN,
 518                                             max_id) == 0);
 519                                         valid_top_config = B_TRUE;
 520                                 }
 521
 522                                 if (nvlist_lookup_uint64_array(tmp,
 523                                     ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
 524                                     &holes) == 0) {
 525                                         verify(nvlist_add_uint64_array(config,
 526                                             ZPOOL_CONFIG_HOLE_ARRAY,
 527                                             hole_array, holes) == 0);
 528                                 }
 529                         }
 530
 531                         if (!config_seen) {
 532                                 /*
 533                                  * Copy the relevant pieces of data to the pool
 534                                  * configuration:
 535                                  *
 536                                  *      version
 537                                  *      pool guid
 538                                  *      name
 539                                  *      comment (if available)
 540                                  *      pool state
 541                                  *      hostid (if available)
 542                                  *      hostname (if available)
 543                                  */
 544                                 uint64_t state, version;
 545                                 char *comment = NULL;
 546
 547                                 version = fnvlist_lookup_uint64(tmp,
 548                                     ZPOOL_CONFIG_VERSION);
 549                                 fnvlist_add_uint64(config,
 550                                     ZPOOL_CONFIG_VERSION, version);
 551                                 guid = fnvlist_lookup_uint64(tmp,
 552                                     ZPOOL_CONFIG_POOL_GUID);
 553                                 fnvlist_add_uint64(config,
 554                                     ZPOOL_CONFIG_POOL_GUID, guid);
 555                                 name = fnvlist_lookup_string(tmp,
 556                                     ZPOOL_CONFIG_POOL_NAME);
 557                                 fnvlist_add_string(config,
 558                                     ZPOOL_CONFIG_POOL_NAME, name);
 559
 560                                 if (nvlist_lookup_string(tmp,
 561                                     ZPOOL_CONFIG_COMMENT, &comment) == 0)
 562                                         fnvlist_add_string(config,
 563                                             ZPOOL_CONFIG_COMMENT, comment);
 564
 565                                 state = fnvlist_lookup_uint64(tmp,
 566                                     ZPOOL_CONFIG_POOL_STATE);
 567                                 fnvlist_add_uint64(config,
 568                                     ZPOOL_CONFIG_POOL_STATE, state);
 569
 570                                 hostid = 0;
 571                                 if (nvlist_lookup_uint64(tmp,
 572                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 573                                         fnvlist_add_uint64(config,
 574                                             ZPOOL_CONFIG_HOSTID, hostid);
 575                                         hostname = fnvlist_lookup_string(tmp,
 576                                             ZPOOL_CONFIG_HOSTNAME);
 577                                         fnvlist_add_string(config,
 578                                             ZPOOL_CONFIG_HOSTNAME, hostname);
 579                                 }
 580
 581                                 config_seen = B_TRUE;
 582                         }
 583
 584                         /*
 585                          * Add this top-level vdev to the child array.
 586                          */
 587                         verify(nvlist_lookup_nvlist(tmp,
 588                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 589                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 590                             &id) == 0);
 591
 592                         if (id >= children) {
 593                                 nvlist_t **newchild;
 594
 595                                 newchild = zfs_alloc(hdl, (id + 1) *
 596                                     sizeof (nvlist_t *));
 597                                 if (newchild == NULL)
 598                                         goto nomem;
 599
 600                                 for (c = 0; c < children; c++)
 601                                         newchild[c] = child[c];
 602
 603                                 free(child);
 604                                 child = newchild;
 605                                 children = id + 1;
 606                         }
 607                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 608                                 goto nomem;
 609
 610                 }
 611
 612                 /*
 613                  * If we have information about all the top-levels then
 614                  * clean up the nvlist which we've constructed. This
 615                  * means removing any extraneous devices that are
 616                  * beyond the valid range or adding devices to the end
 617                  * of our array which appear to be missing.
 618                  */
 619                 if (valid_top_config) {
 620                         if (max_id < children) {
 621                                 for (c = max_id; c < children; c++)
 622                                         nvlist_free(child[c]);
 623                                 children = max_id;
 624                         } else if (max_id > children) {
 625                                 nvlist_t **newchild;
 626
 627                                 newchild = zfs_alloc(hdl, (max_id) *
 628                                     sizeof (nvlist_t *));
 629                                 if (newchild == NULL)
 630                                         goto nomem;
 631
 632                                 for (c = 0; c < children; c++)
 633                                         newchild[c] = child[c];
 634
 635                                 free(child);
 636                                 child = newchild;
 637                                 children = max_id;
 638                         }
 639                 }
 640
 641                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 642                     &guid) == 0);
 643
 644                 /*
 645                  * The vdev namespace may contain holes as a result of
 646                  * device removal. We must add them back into the vdev
 647                  * tree before we process any missing devices.
 648                  */
 649                 if (holes > 0) {
 650                         ASSERT(valid_top_config);
 651
 652                         for (c = 0; c < children; c++) {
 653                                 nvlist_t *holey;
 654
 655                                 if (child[c] != NULL ||
 656                                     !vdev_is_hole(hole_array, holes, c))
 657                                         continue;
 658
 659                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 660                                     0) != 0)
 661                                         goto nomem;
 662
 663                                 /*
 664                                  * Holes in the namespace are treated as
 665                                  * "hole" top-level vdevs and have a
 666                                  * special flag set on them.
 667                                  */
 668                                 if (nvlist_add_string(holey,
 669                                     ZPOOL_CONFIG_TYPE,
 670                                     VDEV_TYPE_HOLE) != 0 ||
 671                                     nvlist_add_uint64(holey,
 672                                     ZPOOL_CONFIG_ID, c) != 0 ||
 673                                     nvlist_add_uint64(holey,
 674                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 675                                         nvlist_free(holey);
 676                                         goto nomem;
 677                                 }
 678                                 child[c] = holey;
 679                         }
 680                 }
 681
 682                 /*
 683                  * Look for any missing top-level vdevs.  If this is the case,
 684                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 685                  * simply compress the child array, because the kernel performs
 686                  * certain checks to make sure the vdev IDs match their location
 687                  * in the configuration.
 688                  */
 689                 for (c = 0; c < children; c++) {
 690                         if (child[c] == NULL) {
 691                                 nvlist_t *missing;
 692                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 693                                     0) != 0)
 694                                         goto nomem;
 695                                 if (nvlist_add_string(missing,
 696                                     ZPOOL_CONFIG_TYPE,
 697                                     VDEV_TYPE_MISSING) != 0 ||
 698                                     nvlist_add_uint64(missing,
 699                                     ZPOOL_CONFIG_ID, c) != 0 ||
 700                                     nvlist_add_uint64(missing,
 701                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 702                                         nvlist_free(missing);
 703                                         goto nomem;
 704                                 }
 705                                 child[c] = missing;
 706                         }
 707                 }
 708
 709                 /*
 710                  * Put all of this pool's top-level vdevs into a root vdev.
 711                  */
 712                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 713                         goto nomem;
 714                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 715                     VDEV_TYPE_ROOT) != 0 ||
 716                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 717                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 718                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 719                     child, children) != 0) {
 720                         nvlist_free(nvroot);
 721                         goto nomem;
 722                 }
 723
 724                 for (c = 0; c < children; c++)
 725                         nvlist_free(child[c]);
 726                 free(child);
 727                 children = 0;
 728                 child = NULL;
 729
 730                 /*
 731                  * Go through and fix up any paths and/or devids based on our
 732                  * known list of vdev GUID -> path mappings.
 733                  */
 734                 if (fix_paths(nvroot, pl->names) != 0) {
 735                         nvlist_free(nvroot);
 736                         goto nomem;
 737                 }
 738
 739                 /*
 740                  * Add the root vdev to this pool's configuration.
 741                  */
 742                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 743                     nvroot) != 0) {
 744                         nvlist_free(nvroot);
 745                         goto nomem;
 746                 }
 747                 nvlist_free(nvroot);
 748
 749                 /*
 750                  * zdb uses this path to report on active pools that were
 751                  * imported or created using -R.
 752                  */
 753                 if (active_ok)
 754                         goto add_pool;
 755
 756                 /*
 757                  * Determine if this pool is currently active, in which case we
 758                  * can't actually import it.
 759                  */
 760                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 761                     &name) == 0);
 762                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 763                     &guid) == 0);
 764
 765                 if (pool_active(hdl, name, guid, &isactive) != 0)
 766                         goto error;
 767
 768                 if (isactive) {
 769                         nvlist_free(config);
 770                         config = NULL;
 771                         continue;
 772                 }
 773
 774                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 775                         nvlist_free(config);
 776                         config = NULL;
 777                         continue;
 778                 }
 779
 780                 nvlist_free(config);
 781                 config = nvl;
 782
 783                 /*
 784                  * Go through and update the paths for spares, now that we have
 785                  * them.
 786                  */
 787                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 788                     &nvroot) == 0);
 789                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 790                     &spares, &nspares) == 0) {
 791                         for (i = 0; i < nspares; i++) {
 792                                 if (fix_paths(spares[i], pl->names) != 0)
 793                                         goto nomem;
 794                         }
 795                 }
 796
 797                 /*
 798                  * Update the paths for l2cache devices.
 799                  */
 800                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 801                     &l2cache, &nl2cache) == 0) {
 802                         for (i = 0; i < nl2cache; i++) {
 803                                 if (fix_paths(l2cache[i], pl->names) != 0)
 804                                         goto nomem;
 805                         }
 806                 }
 807
 808                 /*
 809                  * Restore the original information read from the actual label.
 810                  */
 811                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 812                     DATA_TYPE_UINT64);
 813                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 814                     DATA_TYPE_STRING);
 815                 if (hostid != 0) {
 816                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 817                             hostid) == 0);
 818                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 819                             hostname) == 0);
 820                 }
 821
 822 add_pool:
 823                 /*
 824                  * Add this pool to the list of configs.
 825                  */
 826                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 827                     &name) == 0);
 828                 if (nvlist_add_nvlist(ret, name, config) != 0)
 829                         goto nomem;
 830
 831                 found_one = B_TRUE;
 832                 nvlist_free(config);
 833                 config = NULL;
 834         }
 835
 836         if (!found_one) {
 837                 nvlist_free(ret);
 838                 ret = NULL;
 839         }
 840
 841         return (ret);
 842
 843 nomem:
 844         (void) no_memory(hdl);
 845 error:
 846         nvlist_free(config);
 847         nvlist_free(ret);
 848         for (c = 0; c < children; c++)
 849                 nvlist_free(child[c]);
 850         free(child);
 851
 852         return (NULL);
 853 }
 854
 855 /*
 856  * Return the offset of the given label.
 857  */
 858 static uint64_t
 859 label_offset(uint64_t size, int l)
 860 {
 861         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 862         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 863             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 864 }
 865
 866 /*
 867  * Given a file descriptor, read the label information and return an nvlist
 868  * describing the configuration, if there is one.
 869  * Return 0 on success, or -1 on failure
 870  */
 871 int
 872 zpool_read_label(int fd, nvlist_t **config)
 873 {
 874         struct stat64 statbuf;
 875         int l;
 876         vdev_label_t *label;
 877         uint64_t state, txg, size;
 878
 879         *config = NULL;
 880
 881         if (fstat64(fd, &statbuf) == -1)
 882                 return (-1);
 883         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 884
 885         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 886                 return (-1);
 887
 888         for (l = 0; l < VDEV_LABELS; l++) {
 889                 if (pread64(fd, label, sizeof (vdev_label_t),
 890                     label_offset(size, l)) != sizeof (vdev_label_t))
 891                         continue;
 892
 893                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 894                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 895                         continue;
 896
 897                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 898                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 899                         nvlist_free(*config);
 900                         continue;
 901                 }
 902
 903                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 904                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 905                     &txg) != 0 || txg == 0)) {
 906                         nvlist_free(*config);
 907                         continue;
 908                 }
 909
 910                 free(label);
 911                 return (0);
 912         }
 913
 914         free(label);
 915         *config = NULL;
 916         return (-1);
 917 }
 918
 919 /*
 920  * Given a file descriptor, read the label information and return an nvlist
 921  * describing the configuration, if there is one.
 922  * returns the number of valid labels found
 923  * If a label is found, returns it via config.  The caller is responsible for
 924  * freeing it.
 925  */
 926 int
 927 zpool_read_all_labels(int fd, nvlist_t **config)
 928 {
 929         struct stat64 statbuf;
 930         struct aiocb aiocbs[VDEV_LABELS];
 931         struct aiocb *aiocbps[VDEV_LABELS];
 932         int l;
 933         vdev_phys_t *labels;
 934         uint64_t state, txg, size;
 935         int nlabels = 0;
 936
 937         *config = NULL;
 938
 939         if (fstat64(fd, &statbuf) == -1)
 940                 return (0);
 941         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 942
 943         if ((labels = calloc(VDEV_LABELS, sizeof (vdev_phys_t))) == NULL)
 944                 return (0);
 945
 946         memset(aiocbs, 0, sizeof(aiocbs));
 947         for (l = 0; l < VDEV_LABELS; l++) {
 948                 aiocbs[l].aio_fildes = fd;
 949                 aiocbs[l].aio_offset = label_offset(size, l) + VDEV_SKIP_SIZE;
 950                 aiocbs[l].aio_buf = &labels[l];
 951                 aiocbs[l].aio_nbytes = sizeof(vdev_phys_t);
 952                 aiocbs[l].aio_lio_opcode = LIO_READ;
 953                 aiocbps[l] = &aiocbs[l];
 954         }
 955
 956         if (lio_listio(LIO_WAIT, aiocbps, VDEV_LABELS, NULL) != 0) {
 957                 if (errno == EAGAIN || errno == EINTR || errno == EIO) {
 958                         for (l = 0; l < VDEV_LABELS; l++) {
 959                                 errno = 0;
 960                                 int r = aio_error(&aiocbs[l]);
 961                                 if (r != EINVAL)
 962                                         (void)aio_return(&aiocbs[l]);
 963                         }
 964                 }
 965                 free(labels);
 966                 return (0);
 967         }
 968
 969         for (l = 0; l < VDEV_LABELS; l++) {
 970                 nvlist_t *temp = NULL;
 971
 972                 if (aio_return(&aiocbs[l]) != sizeof(vdev_phys_t))
 973                         continue;
 974
 975                 if (nvlist_unpack(labels[l].vp_nvlist,
 976                     sizeof (labels[l].vp_nvlist), &temp, 0) != 0)
 977                         continue;
 978
 979                 if (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_STATE,
 980                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 981                         nvlist_free(temp);
 982                         temp = NULL;
 983                         continue;
 984                 }
 985
 986                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 987                     (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_TXG,
 988                     &txg) != 0 || txg == 0)) {
 989                         nvlist_free(temp);
 990                         temp = NULL;
 991                         continue;
 992                 }
 993                 if (temp)
 994                         *config = temp;
 995
 996                 nlabels++;
 997         }
 998
 999         free(labels);
1000         return (nlabels);
1001 }
1002
1003 typedef struct rdsk_node {
1004         char *rn_name;
1005         int rn_dfd;
1006         libzfs_handle_t *rn_hdl;
1007         nvlist_t *rn_config;
1008         avl_tree_t *rn_avl;
1009         avl_node_t rn_node;
1010         boolean_t rn_nozpool;
1011 } rdsk_node_t;
1012
1013 static int
1014 slice_cache_compare(const void *arg1, const void *arg2)
1015 {
1016         const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
1017         const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
1018         char *nm1slice, *nm2slice;
1019         int rv;
1020
1021         /*
1022          * slices zero and two are the most likely to provide results,
1023          * so put those first
1024          */
1025         nm1slice = strstr(nm1, "s0");
1026         nm2slice = strstr(nm2, "s0");
1027         if (nm1slice && !nm2slice) {
1028                 return (-1);
1029         }
1030         if (!nm1slice && nm2slice) {
1031                 return (1);
1032         }
1033         nm1slice = strstr(nm1, "s2");
1034         nm2slice = strstr(nm2, "s2");
1035         if (nm1slice && !nm2slice) {
1036                 return (-1);
1037         }
1038         if (!nm1slice && nm2slice) {
1039                 return (1);
1040         }
1041
1042         rv = strcmp(nm1, nm2);
1043         if (rv == 0)
1044                 return (0);
1045         return (rv > 0 ? 1 : -1);
1046 }
1047
1048 #ifdef illumos
1049 static void
1050 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
1051     diskaddr_t size, uint_t blksz)
1052 {
1053         rdsk_node_t tmpnode;
1054         rdsk_node_t *node;
1055         char sname[MAXNAMELEN];
1056
1057         tmpnode.rn_name = &sname[0];
1058         (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
1059             diskname, partno);
1060         /*
1061          * protect against division by zero for disk labels that
1062          * contain a bogus sector size
1063          */
1064         if (blksz == 0)
1065                 blksz = DEV_BSIZE;
1066         /* too small to contain a zpool? */
1067         if ((size < (SPA_MINDEVSIZE / blksz)) &&
1068             (node = avl_find(r, &tmpnode, NULL)))
1069                 node->rn_nozpool = B_TRUE;
1070 }
1071 #endif  /* illumos */
1072
1073 static void
1074 nozpool_all_slices(avl_tree_t *r, const char *sname)
1075 {
1076 #ifdef illumos
1077         char diskname[MAXNAMELEN];
1078         char *ptr;
1079         int i;
1080
1081         (void) strncpy(diskname, sname, MAXNAMELEN);
1082         if (((ptr = strrchr(diskname, 's')) == NULL) &&
1083             ((ptr = strrchr(diskname, 'p')) == NULL))
1084                 return;
1085         ptr[0] = 's';
1086         ptr[1] = '\0';
1087         for (i = 0; i < NDKMAP; i++)
1088                 check_one_slice(r, diskname, i, 0, 1);
1089         ptr[0] = 'p';
1090         for (i = 0; i <= FD_NUMPART; i++)
1091                 check_one_slice(r, diskname, i, 0, 1);
1092 #endif  /* illumos */
1093 }
1094
1095 #ifdef illumos
1096 static void
1097 check_slices(avl_tree_t *r, int fd, const char *sname)
1098 {
1099         struct extvtoc vtoc;
1100         struct dk_gpt *gpt;
1101         char diskname[MAXNAMELEN];
1102         char *ptr;
1103         int i;
1104
1105         (void) strncpy(diskname, sname, MAXNAMELEN);
1106         if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1107                 return;
1108         ptr[1] = '\0';
1109
1110         if (read_extvtoc(fd, &vtoc) >= 0) {
1111                 for (i = 0; i < NDKMAP; i++)
1112                         check_one_slice(r, diskname, i,
1113                             vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1114         } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1115                 /*
1116                  * on x86 we'll still have leftover links that point
1117                  * to slices s[9-15], so use NDKMAP instead
1118                  */
1119                 for (i = 0; i < NDKMAP; i++)
1120                         check_one_slice(r, diskname, i,
1121                             gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1122                 /* nodes p[1-4] are never used with EFI labels */
1123                 ptr[0] = 'p';
1124                 for (i = 1; i <= FD_NUMPART; i++)
1125                         check_one_slice(r, diskname, i, 0, 1);
1126                 efi_free(gpt);
1127         }
1128 }
1129 #endif  /* illumos */
1130
1131 static void
1132 zpool_open_func(void *arg)
1133 {
1134         rdsk_node_t *rn = arg;
1135         struct stat64 statbuf;
1136         nvlist_t *config;
1137         int fd;
1138
1139         if (rn->rn_nozpool)
1140                 return;
1141         if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1142                 /* symlink to a device that's no longer there */
1143                 if (errno == ENOENT)
1144                         nozpool_all_slices(rn->rn_avl, rn->rn_name);
1145                 return;
1146         }
1147         /*
1148          * Ignore failed stats.  We only want regular
1149          * files, character devs and block devs.
1150          */
1151         if (fstat64(fd, &statbuf) != 0 ||
1152             (!S_ISREG(statbuf.st_mode) &&
1153             !S_ISCHR(statbuf.st_mode) &&
1154             !S_ISBLK(statbuf.st_mode))) {
1155                 (void) close(fd);
1156                 return;
1157         }
1158         /* this file is too small to hold a zpool */
1159 #ifdef illumos
1160         if (S_ISREG(statbuf.st_mode) &&
1161             statbuf.st_size < SPA_MINDEVSIZE) {
1162                 (void) close(fd);
1163                 return;
1164         } else if (!S_ISREG(statbuf.st_mode)) {
1165                 /*
1166                  * Try to read the disk label first so we don't have to
1167                  * open a bunch of minor nodes that can't have a zpool.
1168                  */
1169                 check_slices(rn->rn_avl, fd, rn->rn_name);
1170         }
1171 #else   /* !illumos */
1172         if (statbuf.st_size < SPA_MINDEVSIZE) {
1173                 (void) close(fd);
1174                 return;
1175         }
1176 #endif  /* illumos */
1177
1178         if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1179                 (void) close(fd);
1180                 (void) no_memory(rn->rn_hdl);
1181                 return;
1182         }
1183         (void) close(fd);
1184
1185         rn->rn_config = config;
1186 }
1187
1188 /*
1189  * Given a file descriptor, clear (zero) the label information.
1190  */
1191 int
1192 zpool_clear_label(int fd)
1193 {
1194         struct stat64 statbuf;
1195         int l;
1196         vdev_label_t *label;
1197         uint64_t size;
1198
1199         if (fstat64(fd, &statbuf) == -1)
1200                 return (0);
1201         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1202
1203         if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1204                 return (-1);
1205
1206         for (l = 0; l < VDEV_LABELS; l++) {
1207                 if (pwrite64(fd, label, sizeof (vdev_label_t),
1208                     label_offset(size, l)) != sizeof (vdev_label_t)) {
1209                         free(label);
1210                         return (-1);
1211                 }
1212         }
1213
1214         free(label);
1215         return (0);
1216 }
1217
1218 /*
1219  * Given a list of directories to search, find all pools stored on disk.  This
1220  * includes partial pools which are not available to import.  If no args are
1221  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1222  * poolname or guid (but not both) are provided by the caller when trying
1223  * to import a specific pool.
1224  */
1225 static nvlist_t *
1226 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1227 {
1228         int i, dirs = iarg->paths;
1229         struct dirent64 *dp;
1230         char path[MAXPATHLEN];
1231         char *end, **dir = iarg->path;
1232         size_t pathleft;
1233         nvlist_t *ret = NULL;
1234         static char *default_dir = "/dev";
1235         pool_list_t pools = { 0 };
1236         pool_entry_t *pe, *penext;
1237         vdev_entry_t *ve, *venext;
1238         config_entry_t *ce, *cenext;
1239         name_entry_t *ne, *nenext;
1240         avl_tree_t slice_cache;
1241         rdsk_node_t *slice;
1242         void *cookie;
1243
1244         if (dirs == 0) {
1245                 dirs = 1;
1246                 dir = &default_dir;
1247         }
1248
1249         /*
1250          * Go through and read the label configuration information from every
1251          * possible device, organizing the information according to pool GUID
1252          * and toplevel GUID.
1253          */
1254         for (i = 0; i < dirs; i++) {
1255                 tpool_t *t;
1256                 char rdsk[MAXPATHLEN];
1257                 int dfd;
1258                 boolean_t config_failed = B_FALSE;
1259                 DIR *dirp;
1260
1261                 /* use realpath to normalize the path */
1262                 if (realpath(dir[i], path) == 0) {
1263                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1264                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1265                         goto error;
1266                 }
1267                 end = &path[strlen(path)];
1268                 *end++ = '/';
1269                 *end = 0;
1270                 pathleft = &path[sizeof (path)] - end;
1271
1272 #ifdef illumos
1273                 /*
1274                  * Using raw devices instead of block devices when we're
1275                  * reading the labels skips a bunch of slow operations during
1276                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1277                  */
1278                 if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1279                         (void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1280                 else
1281 #endif
1282                         (void) strlcpy(rdsk, path, sizeof (rdsk));
1283
1284                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1285                     (dirp = fdopendir(dfd)) == NULL) {
1286                         if (dfd >= 0)
1287                                 (void) close(dfd);
1288                         zfs_error_aux(hdl, strerror(errno));
1289                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1290                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1291                             rdsk);
1292                         goto error;
1293                 }
1294
1295                 avl_create(&slice_cache, slice_cache_compare,
1296                     sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1297
1298                 if (strcmp(rdsk, "/dev/") == 0) {
1299                         struct gmesh mesh;
1300                         struct gclass *mp;
1301                         struct ggeom *gp;
1302                         struct gprovider *pp;
1303
1304                         errno = geom_gettree(&mesh);
1305                         if (errno != 0) {
1306                                 zfs_error_aux(hdl, strerror(errno));
1307                                 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1308                                     dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1309                                 goto error;
1310                         }
1311
1312                         LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1313                                 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1314                                         LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1315                                                 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1316                                                 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1317                                                 slice->rn_avl = &slice_cache;
1318                                                 slice->rn_dfd = dfd;
1319                                                 slice->rn_hdl = hdl;
1320                                                 slice->rn_nozpool = B_FALSE;
1321                                                 avl_add(&slice_cache, slice);
1322                                         }
1323                                 }
1324                         }
1325
1326                         geom_deletetree(&mesh);
1327                         goto skipdir;
1328                 }
1329
1330                 /*
1331                  * This is not MT-safe, but we have no MT consumers of libzfs
1332                  */
1333                 while ((dp = readdir64(dirp)) != NULL) {
1334                         const char *name = dp->d_name;
1335                         if (name[0] == '.' &&
1336                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1337                                 continue;
1338
1339                         slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1340                         slice->rn_name = zfs_strdup(hdl, name);
1341                         slice->rn_avl = &slice_cache;
1342                         slice->rn_dfd = dfd;
1343                         slice->rn_hdl = hdl;
1344                         slice->rn_nozpool = B_FALSE;
1345                         avl_add(&slice_cache, slice);
1346                 }
1347 skipdir:
1348                 /*
1349                  * create a thread pool to do all of this in parallel;
1350                  * rn_nozpool is not protected, so this is racy in that
1351                  * multiple tasks could decide that the same slice can
1352                  * not hold a zpool, which is benign.  Also choose
1353                  * double the number of processors; we hold a lot of
1354                  * locks in the kernel, so going beyond this doesn't
1355                  * buy us much.
1356                  */
1357                 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1358                     0, NULL);
1359                 for (slice = avl_first(&slice_cache); slice;
1360                     (slice = avl_walk(&slice_cache, slice,
1361                     AVL_AFTER)))
1362                         (void) tpool_dispatch(t, zpool_open_func, slice);
1363                 tpool_wait(t);
1364                 tpool_destroy(t);
1365
1366                 cookie = NULL;
1367                 while ((slice = avl_destroy_nodes(&slice_cache,
1368                     &cookie)) != NULL) {
1369                         if (slice->rn_config != NULL && !config_failed) {
1370                                 nvlist_t *config = slice->rn_config;
1371                                 boolean_t matched = B_TRUE;
1372
1373                                 if (iarg->poolname != NULL) {
1374                                         char *pname;
1375
1376                                         matched = nvlist_lookup_string(config,
1377                                             ZPOOL_CONFIG_POOL_NAME,
1378                                             &pname) == 0 &&
1379                                             strcmp(iarg->poolname, pname) == 0;
1380                                 } else if (iarg->guid != 0) {
1381                                         uint64_t this_guid;
1382
1383                                         matched = nvlist_lookup_uint64(config,
1384                                             ZPOOL_CONFIG_POOL_GUID,
1385                                             &this_guid) == 0 &&
1386                                             iarg->guid == this_guid;
1387                                 }
1388                                 if (!matched) {
1389                                         nvlist_free(config);
1390                                 } else {
1391                                         /*
1392                                          * use the non-raw path for the config
1393                                          */
1394                                         (void) strlcpy(end, slice->rn_name,
1395                                             pathleft);
1396                                         if (add_config(hdl, &pools, path,
1397                                             config) != 0)
1398                                                 config_failed = B_TRUE;
1399                                 }
1400                         }
1401                         free(slice->rn_name);
1402                         free(slice);
1403                 }
1404                 avl_destroy(&slice_cache);
1405
1406                 (void) closedir(dirp);
1407
1408                 if (config_failed)
1409                         goto error;
1410         }
1411
1412         ret = get_configs(hdl, &pools, iarg->can_be_active);
1413
1414 error:
1415         for (pe = pools.pools; pe != NULL; pe = penext) {
1416                 penext = pe->pe_next;
1417                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1418                         venext = ve->ve_next;
1419                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1420                                 cenext = ce->ce_next;
1421                                 nvlist_free(ce->ce_config);
1422                                 free(ce);
1423                         }
1424                         free(ve);
1425                 }
1426                 free(pe);
1427         }
1428
1429         for (ne = pools.names; ne != NULL; ne = nenext) {
1430                 nenext = ne->ne_next;
1431                 free(ne->ne_name);
1432                 free(ne);
1433         }
1434
1435         return (ret);
1436 }
1437
1438 nvlist_t *
1439 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1440 {
1441         importargs_t iarg = { 0 };
1442
1443         iarg.paths = argc;
1444         iarg.path = argv;
1445
1446         return (zpool_find_import_impl(hdl, &iarg));
1447 }
1448
1449 /*
1450  * Given a cache file, return the contents as a list of importable pools.
1451  * poolname or guid (but not both) are provided by the caller when trying
1452  * to import a specific pool.
1453  */
1454 nvlist_t *
1455 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1456     char *poolname, uint64_t guid)
1457 {
1458         char *buf;
1459         int fd;
1460         struct stat64 statbuf;
1461         nvlist_t *raw, *src, *dst;
1462         nvlist_t *pools;
1463         nvpair_t *elem;
1464         char *name;
1465         uint64_t this_guid;
1466         boolean_t active;
1467
1468         verify(poolname == NULL || guid == 0);
1469
1470         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1471                 zfs_error_aux(hdl, "%s", strerror(errno));
1472                 (void) zfs_error(hdl, EZFS_BADCACHE,
1473                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1474                 return (NULL);
1475         }
1476
1477         if (fstat64(fd, &statbuf) != 0) {
1478                 zfs_error_aux(hdl, "%s", strerror(errno));
1479                 (void) close(fd);
1480                 (void) zfs_error(hdl, EZFS_BADCACHE,
1481                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1482                 return (NULL);
1483         }
1484
1485         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1486                 (void) close(fd);
1487                 return (NULL);
1488         }
1489
1490         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1491                 (void) close(fd);
1492                 free(buf);
1493                 (void) zfs_error(hdl, EZFS_BADCACHE,
1494                     dgettext(TEXT_DOMAIN,
1495                     "failed to read cache file contents"));
1496                 return (NULL);
1497         }
1498
1499         (void) close(fd);
1500
1501         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1502                 free(buf);
1503                 (void) zfs_error(hdl, EZFS_BADCACHE,
1504                     dgettext(TEXT_DOMAIN,
1505                     "invalid or corrupt cache file contents"));
1506                 return (NULL);
1507         }
1508
1509         free(buf);
1510
1511         /*
1512          * Go through and get the current state of the pools and refresh their
1513          * state.
1514          */
1515         if (nvlist_alloc(&pools, 0, 0) != 0) {
1516                 (void) no_memory(hdl);
1517                 nvlist_free(raw);
1518                 return (NULL);
1519         }
1520
1521         elem = NULL;
1522         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1523                 src = fnvpair_value_nvlist(elem);
1524
1525                 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1526                 if (poolname != NULL && strcmp(poolname, name) != 0)
1527                         continue;
1528
1529                 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1530                 if (guid != 0 && guid != this_guid)
1531                         continue;
1532
1533                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1534                         nvlist_free(raw);
1535                         nvlist_free(pools);
1536                         return (NULL);
1537                 }
1538
1539                 if (active)
1540                         continue;
1541
1542                 if ((dst = refresh_config(hdl, src)) == NULL) {
1543                         nvlist_free(raw);
1544                         nvlist_free(pools);
1545                         return (NULL);
1546                 }
1547
1548                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1549                         (void) no_memory(hdl);
1550                         nvlist_free(dst);
1551                         nvlist_free(raw);
1552                         nvlist_free(pools);
1553                         return (NULL);
1554                 }
1555                 nvlist_free(dst);
1556         }
1557
1558         nvlist_free(raw);
1559         return (pools);
1560 }
1561
1562 static int
1563 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1564 {
1565         importargs_t *import = data;
1566         int found = 0;
1567
1568         if (import->poolname != NULL) {
1569                 char *pool_name;
1570
1571                 verify(nvlist_lookup_string(zhp->zpool_config,
1572                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1573                 if (strcmp(pool_name, import->poolname) == 0)
1574                         found = 1;
1575         } else {
1576                 uint64_t pool_guid;
1577
1578                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1579                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1580                 if (pool_guid == import->guid)
1581                         found = 1;
1582         }
1583
1584         zpool_close(zhp);
1585         return (found);
1586 }
1587
1588 nvlist_t *
1589 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1590 {
1591         verify(import->poolname == NULL || import->guid == 0);
1592
1593         if (import->unique)
1594                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1595
1596         if (import->cachefile != NULL)
1597                 return (zpool_find_import_cached(hdl, import->cachefile,
1598                     import->poolname, import->guid));
1599
1600         return (zpool_find_import_impl(hdl, import));
1601 }
1602
1603 boolean_t
1604 find_guid(nvlist_t *nv, uint64_t guid)
1605 {
1606         uint64_t tmp;
1607         nvlist_t **child;
1608         uint_t c, children;
1609
1610         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1611         if (tmp == guid)
1612                 return (B_TRUE);
1613
1614         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1615             &child, &children) == 0) {
1616                 for (c = 0; c < children; c++)
1617                         if (find_guid(child[c], guid))
1618                                 return (B_TRUE);
1619         }
1620
1621         return (B_FALSE);
1622 }
1623
1624 typedef struct aux_cbdata {
1625         const char      *cb_type;
1626         uint64_t        cb_guid;
1627         zpool_handle_t  *cb_zhp;
1628 } aux_cbdata_t;
1629
1630 static int
1631 find_aux(zpool_handle_t *zhp, void *data)
1632 {
1633         aux_cbdata_t *cbp = data;
1634         nvlist_t **list;
1635         uint_t i, count;
1636         uint64_t guid;
1637         nvlist_t *nvroot;
1638
1639         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1640             &nvroot) == 0);
1641
1642         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1643             &list, &count) == 0) {
1644                 for (i = 0; i < count; i++) {
1645                         verify(nvlist_lookup_uint64(list[i],
1646                             ZPOOL_CONFIG_GUID, &guid) == 0);
1647                         if (guid == cbp->cb_guid) {
1648                                 cbp->cb_zhp = zhp;
1649                                 return (1);
1650                         }
1651                 }
1652         }
1653
1654         zpool_close(zhp);
1655         return (0);
1656 }
1657
1658 /*
1659  * Determines if the pool is in use.  If so, it returns true and the state of
1660  * the pool as well as the name of the pool.  Both strings are allocated and
1661  * must be freed by the caller.
1662  */
1663 int
1664 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1665     boolean_t *inuse)
1666 {
1667         nvlist_t *config;
1668         char *name;
1669         boolean_t ret;
1670         uint64_t guid, vdev_guid;
1671         zpool_handle_t *zhp;
1672         nvlist_t *pool_config;
1673         uint64_t stateval, isspare;
1674         aux_cbdata_t cb = { 0 };
1675         boolean_t isactive;
1676
1677         *inuse = B_FALSE;
1678
1679         if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1680                 (void) no_memory(hdl);
1681                 return (-1);
1682         }
1683
1684         if (config == NULL)
1685                 return (0);
1686
1687         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1688             &stateval) == 0);
1689         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1690             &vdev_guid) == 0);
1691
1692         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1693                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1694                     &name) == 0);
1695                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1696                     &guid) == 0);
1697         }
1698
1699         switch (stateval) {
1700         case POOL_STATE_EXPORTED:
1701                 /*
1702                  * A pool with an exported state may in fact be imported
1703                  * read-only, so check the in-core state to see if it's
1704                  * active and imported read-only.  If it is, set
1705                  * its state to active.
1706                  */
1707                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1708                     (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1709                         if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1710                                 stateval = POOL_STATE_ACTIVE;
1711
1712                         /*
1713                          * All we needed the zpool handle for is the
1714                          * readonly prop check.
1715                          */
1716                         zpool_close(zhp);
1717                 }
1718
1719                 ret = B_TRUE;
1720                 break;
1721
1722         case POOL_STATE_ACTIVE:
1723                 /*
1724                  * For an active pool, we have to determine if it's really part
1725                  * of a currently active pool (in which case the pool will exist
1726                  * and the guid will be the same), or whether it's part of an
1727                  * active pool that was disconnected without being explicitly
1728                  * exported.
1729                  */
1730                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1731                         nvlist_free(config);
1732                         return (-1);
1733                 }
1734
1735                 if (isactive) {
1736                         /*
1737                          * Because the device may have been removed while
1738                          * offlined, we only report it as active if the vdev is
1739                          * still present in the config.  Otherwise, pretend like
1740                          * it's not in use.
1741                          */
1742                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1743                             (pool_config = zpool_get_config(zhp, NULL))
1744                             != NULL) {
1745                                 nvlist_t *nvroot;
1746
1747                                 verify(nvlist_lookup_nvlist(pool_config,
1748                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1749                                 ret = find_guid(nvroot, vdev_guid);
1750                         } else {
1751                                 ret = B_FALSE;
1752                         }
1753
1754                         /*
1755                          * If this is an active spare within another pool, we
1756                          * treat it like an unused hot spare.  This allows the
1757                          * user to create a pool with a hot spare that currently
1758                          * in use within another pool.  Since we return B_TRUE,
1759                          * libdiskmgt will continue to prevent generic consumers
1760                          * from using the device.
1761                          */
1762                         if (ret && nvlist_lookup_uint64(config,
1763                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1764                                 stateval = POOL_STATE_SPARE;
1765
1766                         if (zhp != NULL)
1767                                 zpool_close(zhp);
1768                 } else {
1769                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1770                         ret = B_TRUE;
1771                 }
1772                 break;
1773
1774         case POOL_STATE_SPARE:
1775                 /*
1776                  * For a hot spare, it can be either definitively in use, or
1777                  * potentially active.  To determine if it's in use, we iterate
1778                  * over all pools in the system and search for one with a spare
1779                  * with a matching guid.
1780                  *
1781                  * Due to the shared nature of spares, we don't actually report
1782                  * the potentially active case as in use.  This means the user
1783                  * can freely create pools on the hot spares of exported pools,
1784                  * but to do otherwise makes the resulting code complicated, and
1785                  * we end up having to deal with this case anyway.
1786                  */
1787                 cb.cb_zhp = NULL;
1788                 cb.cb_guid = vdev_guid;
1789                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1790                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1791                         name = (char *)zpool_get_name(cb.cb_zhp);
1792                         ret = B_TRUE;
1793                 } else {
1794                         ret = B_FALSE;
1795                 }
1796                 break;
1797
1798         case POOL_STATE_L2CACHE:
1799
1800                 /*
1801                  * Check if any pool is currently using this l2cache device.
1802                  */
1803                 cb.cb_zhp = NULL;
1804                 cb.cb_guid = vdev_guid;
1805                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1806                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1807                         name = (char *)zpool_get_name(cb.cb_zhp);
1808                         ret = B_TRUE;
1809                 } else {
1810                         ret = B_FALSE;
1811                 }
1812                 break;
1813
1814         default:
1815                 ret = B_FALSE;
1816         }
1817
1818
1819         if (ret) {
1820                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1821                         if (cb.cb_zhp)
1822                                 zpool_close(cb.cb_zhp);
1823                         nvlist_free(config);
1824                         return (-1);
1825                 }
1826                 *state = (pool_state_t)stateval;
1827         }
1828
1829         if (cb.cb_zhp)
1830                 zpool_close(cb.cb_zhp);
1831
1832         nvlist_free(config);
1833         *inuse = ret;
1834         return (0);
1835 }