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