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 2008 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25
  26 #pragma ident   "%Z%%M% %I%     %E% SMI"
  27
  28 /*
  29  * Pool import support functions.
  30  *
  31  * To import a pool, we rely on reading the configuration information from the
  32  * ZFS label of each device.  If we successfully read the label, then we
  33  * organize the configuration information in the following hierarchy:
  34  *
  35  *      pool guid -> toplevel vdev guid -> label txg
  36  *
  37  * Duplicate entries matching this same tuple will be discarded.  Once we have
  38  * examined every device, we pick the best label txg config for each toplevel
  39  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  40  * update any paths that have changed.  Finally, we attempt to import the pool
  41  * using our derived config, and record the results.
  42  */
  43
  44 #include <devid.h>
  45 #include <dirent.h>
  46 #include <errno.h>
  47 #include <libintl.h>
  48 #include <stdlib.h>
  49 #include <string.h>
  50 #include <sys/stat.h>
  51 #include <unistd.h>
  52 #include <fcntl.h>
  53
  54 #include <sys/vdev_impl.h>
  55
  56 #include "libzfs.h"
  57 #include "libzfs_impl.h"
  58
  59 /*
  60  * Intermediate structures used to gather configuration information.
  61  */
  62 typedef struct config_entry {
  63         uint64_t                ce_txg;
  64         nvlist_t                *ce_config;
  65         struct config_entry     *ce_next;
  66 } config_entry_t;
  67
  68 typedef struct vdev_entry {
  69         uint64_t                ve_guid;
  70         config_entry_t          *ve_configs;
  71         struct vdev_entry       *ve_next;
  72 } vdev_entry_t;
  73
  74 typedef struct pool_entry {
  75         uint64_t                pe_guid;
  76         vdev_entry_t            *pe_vdevs;
  77         struct pool_entry       *pe_next;
  78 } pool_entry_t;
  79
  80 typedef struct name_entry {
  81         char                    *ne_name;
  82         uint64_t                ne_guid;
  83         struct name_entry       *ne_next;
  84 } name_entry_t;
  85
  86 typedef struct pool_list {
  87         pool_entry_t            *pools;
  88         name_entry_t            *names;
  89 } pool_list_t;
  90
  91 static char *
  92 get_devid(const char *path)
  93 {
  94         int fd;
  95         ddi_devid_t devid;
  96         char *minor, *ret;
  97
  98         if ((fd = open(path, O_RDONLY)) < 0)
  99                 return (NULL);
 100
 101         minor = NULL;
 102         ret = NULL;
 103         if (devid_get(fd, &devid) == 0) {
 104                 if (devid_get_minor_name(fd, &minor) == 0)
 105                         ret = devid_str_encode(devid, minor);
 106                 if (minor != NULL)
 107                         devid_str_free(minor);
 108                 devid_free(devid);
 109         }
 110         (void) close(fd);
 111
 112         return (ret);
 113 }
 114
 115
 116 /*
 117  * Go through and fix up any path and/or devid information for the given vdev
 118  * configuration.
 119  */
 120 static int
 121 fix_paths(nvlist_t *nv, name_entry_t *names)
 122 {
 123         nvlist_t **child;
 124         uint_t c, children;
 125         uint64_t guid;
 126         name_entry_t *ne, *best;
 127         char *path, *devid;
 128         int matched;
 129
 130         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 131             &child, &children) == 0) {
 132                 for (c = 0; c < children; c++)
 133                         if (fix_paths(child[c], names) != 0)
 134                                 return (-1);
 135                 return (0);
 136         }
 137
 138         /*
 139          * This is a leaf (file or disk) vdev.  In either case, go through
 140          * the name list and see if we find a matching guid.  If so, replace
 141          * the path and see if we can calculate a new devid.
 142          *
 143          * There may be multiple names associated with a particular guid, in
 144          * which case we have overlapping slices or multiple paths to the same
 145          * disk.  If this is the case, then we want to pick the path that is
 146          * the most similar to the original, where "most similar" is the number
 147          * of matching characters starting from the end of the path.  This will
 148          * preserve slice numbers even if the disks have been reorganized, and
 149          * will also catch preferred disk names if multiple paths exist.
 150          */
 151         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 152         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 153                 path = NULL;
 154
 155         matched = 0;
 156         best = NULL;
 157         for (ne = names; ne != NULL; ne = ne->ne_next) {
 158                 if (ne->ne_guid == guid) {
 159                         const char *src, *dst;
 160                         int count;
 161
 162                         if (path == NULL) {
 163                                 best = ne;
 164                                 break;
 165                         }
 166
 167                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 168                         dst = path + strlen(path) - 1;
 169                         for (count = 0; src >= ne->ne_name && dst >= path;
 170                             src--, dst--, count++)
 171                                 if (*src != *dst)
 172                                         break;
 173
 174                         /*
 175                          * At this point, 'count' is the number of characters
 176                          * matched from the end.
 177                          */
 178                         if (count > matched || best == NULL) {
 179                                 best = ne;
 180                                 matched = count;
 181                         }
 182                 }
 183         }
 184
 185         if (best == NULL)
 186                 return (0);
 187
 188         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 189                 return (-1);
 190
 191         if ((devid = get_devid(best->ne_name)) == NULL) {
 192                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 193         } else {
 194                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
 195                         return (-1);
 196                 devid_str_free(devid);
 197         }
 198
 199         return (0);
 200 }
 201
 202 /*
 203  * Add the given configuration to the list of known devices.
 204  */
 205 static int
 206 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 207     nvlist_t *config)
 208 {
 209         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 210         pool_entry_t *pe;
 211         vdev_entry_t *ve;
 212         config_entry_t *ce;
 213         name_entry_t *ne;
 214
 215         /*
 216          * If this is a hot spare not currently in use or level 2 cache
 217          * device, add it to the list of names to translate, but don't do
 218          * anything else.
 219          */
 220         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 221             &state) == 0 &&
 222             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 223             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 224                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 225                         return (-1);
 226
 227                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 228                         free(ne);
 229                         return (-1);
 230                 }
 231                 ne->ne_guid = vdev_guid;
 232                 ne->ne_next = pl->names;
 233                 pl->names = ne;
 234                 return (0);
 235         }
 236
 237         /*
 238          * If we have a valid config but cannot read any of these fields, then
 239          * it means we have a half-initialized label.  In vdev_label_init()
 240          * we write a label with txg == 0 so that we can identify the device
 241          * in case the user refers to the same disk later on.  If we fail to
 242          * create the pool, we'll be left with a label in this state
 243          * which should not be considered part of a valid pool.
 244          */
 245         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 246             &pool_guid) != 0 ||
 247             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 248             &vdev_guid) != 0 ||
 249             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 250             &top_guid) != 0 ||
 251             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 252             &txg) != 0 || txg == 0) {
 253                 nvlist_free(config);
 254                 return (0);
 255         }
 256
 257         /*
 258          * First, see if we know about this pool.  If not, then add it to the
 259          * list of known pools.
 260          */
 261         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 262                 if (pe->pe_guid == pool_guid)
 263                         break;
 264         }
 265
 266         if (pe == NULL) {
 267                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 268                         nvlist_free(config);
 269                         return (-1);
 270                 }
 271                 pe->pe_guid = pool_guid;
 272                 pe->pe_next = pl->pools;
 273                 pl->pools = pe;
 274         }
 275
 276         /*
 277          * Second, see if we know about this toplevel vdev.  Add it if its
 278          * missing.
 279          */
 280         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 281                 if (ve->ve_guid == top_guid)
 282                         break;
 283         }
 284
 285         if (ve == NULL) {
 286                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 287                         nvlist_free(config);
 288                         return (-1);
 289                 }
 290                 ve->ve_guid = top_guid;
 291                 ve->ve_next = pe->pe_vdevs;
 292                 pe->pe_vdevs = ve;
 293         }
 294
 295         /*
 296          * Third, see if we have a config with a matching transaction group.  If
 297          * so, then we do nothing.  Otherwise, add it to the list of known
 298          * configs.
 299          */
 300         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 301                 if (ce->ce_txg == txg)
 302                         break;
 303         }
 304
 305         if (ce == NULL) {
 306                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 307                         nvlist_free(config);
 308                         return (-1);
 309                 }
 310                 ce->ce_txg = txg;
 311                 ce->ce_config = config;
 312                 ce->ce_next = ve->ve_configs;
 313                 ve->ve_configs = ce;
 314         } else {
 315                 nvlist_free(config);
 316         }
 317
 318         /*
 319          * At this point we've successfully added our config to the list of
 320          * known configs.  The last thing to do is add the vdev guid -> path
 321          * mappings so that we can fix up the configuration as necessary before
 322          * doing the import.
 323          */
 324         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 325                 return (-1);
 326
 327         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 328                 free(ne);
 329                 return (-1);
 330         }
 331
 332         ne->ne_guid = vdev_guid;
 333         ne->ne_next = pl->names;
 334         pl->names = ne;
 335
 336         return (0);
 337 }
 338
 339 /*
 340  * Returns true if the named pool matches the given GUID.
 341  */
 342 static int
 343 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 344     boolean_t *isactive)
 345 {
 346         zpool_handle_t *zhp;
 347         uint64_t theguid;
 348
 349         if (zpool_open_silent(hdl, name, &zhp) != 0)
 350                 return (-1);
 351
 352         if (zhp == NULL) {
 353                 *isactive = B_FALSE;
 354                 return (0);
 355         }
 356
 357         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 358             &theguid) == 0);
 359
 360         zpool_close(zhp);
 361
 362         *isactive = (theguid == guid);
 363         return (0);
 364 }
 365
 366 static nvlist_t *
 367 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 368 {
 369         nvlist_t *nvl;
 370         zfs_cmd_t zc = { 0 };
 371         int err;
 372
 373         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 374                 return (NULL);
 375
 376         if (zcmd_alloc_dst_nvlist(hdl, &zc,
 377             zc.zc_nvlist_conf_size * 2) != 0) {
 378                 zcmd_free_nvlists(&zc);
 379                 return (NULL);
 380         }
 381
 382         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 383             &zc)) != 0 && errno == ENOMEM) {
 384                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 385                         zcmd_free_nvlists(&zc);
 386                         return (NULL);
 387                 }
 388         }
 389
 390         if (err) {
 391                 (void) zpool_standard_error(hdl, errno,
 392                     dgettext(TEXT_DOMAIN, "cannot discover pools"));
 393                 zcmd_free_nvlists(&zc);
 394                 return (NULL);
 395         }
 396
 397         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 398                 zcmd_free_nvlists(&zc);
 399                 return (NULL);
 400         }
 401
 402         zcmd_free_nvlists(&zc);
 403         return (nvl);
 404 }
 405
 406 /*
 407  * Convert our list of pools into the definitive set of configurations.  We
 408  * start by picking the best config for each toplevel vdev.  Once that's done,
 409  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 410  * pass to fix up any incorrect paths, and then add it to the main list to
 411  * return to the user.
 412  */
 413 static nvlist_t *
 414 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 415 {
 416         pool_entry_t *pe;
 417         vdev_entry_t *ve;
 418         config_entry_t *ce;
 419         nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
 420         nvlist_t **spares, **l2cache;
 421         uint_t i, nspares, nl2cache;
 422         boolean_t config_seen;
 423         uint64_t best_txg;
 424         char *name, *hostname;
 425         uint64_t version, guid;
 426         uint_t children = 0;
 427         nvlist_t **child = NULL;
 428         uint_t c;
 429         boolean_t isactive;
 430         uint64_t hostid;
 431         nvlist_t *nvl;
 432         boolean_t found_one = B_FALSE;
 433
 434         if (nvlist_alloc(&ret, 0, 0) != 0)
 435                 goto nomem;
 436
 437         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 438                 uint64_t id;
 439
 440                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 441                         goto nomem;
 442                 config_seen = B_FALSE;
 443
 444                 /*
 445                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 446                  * from the first one we find, and then go through the rest and
 447                  * add them as necessary to the 'vdevs' member of the config.
 448                  */
 449                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 450
 451                         /*
 452                          * Determine the best configuration for this vdev by
 453                          * selecting the config with the latest transaction
 454                          * group.
 455                          */
 456                         best_txg = 0;
 457                         for (ce = ve->ve_configs; ce != NULL;
 458                             ce = ce->ce_next) {
 459
 460                                 if (ce->ce_txg > best_txg) {
 461                                         tmp = ce->ce_config;
 462                                         best_txg = ce->ce_txg;
 463                                 }
 464                         }
 465
 466                         if (!config_seen) {
 467                                 /*
 468                                  * Copy the relevant pieces of data to the pool
 469                                  * configuration:
 470                                  *
 471                                  *      version
 472                                  *      pool guid
 473                                  *      name
 474                                  *      pool state
 475                                  *      hostid (if available)
 476                                  *      hostname (if available)
 477                                  */
 478                                 uint64_t state;
 479
 480                                 verify(nvlist_lookup_uint64(tmp,
 481                                     ZPOOL_CONFIG_VERSION, &version) == 0);
 482                                 if (nvlist_add_uint64(config,
 483                                     ZPOOL_CONFIG_VERSION, version) != 0)
 484                                         goto nomem;
 485                                 verify(nvlist_lookup_uint64(tmp,
 486                                     ZPOOL_CONFIG_POOL_GUID, &guid) == 0);
 487                                 if (nvlist_add_uint64(config,
 488                                     ZPOOL_CONFIG_POOL_GUID, guid) != 0)
 489                                         goto nomem;
 490                                 verify(nvlist_lookup_string(tmp,
 491                                     ZPOOL_CONFIG_POOL_NAME, &name) == 0);
 492                                 if (nvlist_add_string(config,
 493                                     ZPOOL_CONFIG_POOL_NAME, name) != 0)
 494                                         goto nomem;
 495                                 verify(nvlist_lookup_uint64(tmp,
 496                                     ZPOOL_CONFIG_POOL_STATE, &state) == 0);
 497                                 if (nvlist_add_uint64(config,
 498                                     ZPOOL_CONFIG_POOL_STATE, state) != 0)
 499                                         goto nomem;
 500                                 hostid = 0;
 501                                 if (nvlist_lookup_uint64(tmp,
 502                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 503                                         if (nvlist_add_uint64(config,
 504                                             ZPOOL_CONFIG_HOSTID, hostid) != 0)
 505                                                 goto nomem;
 506                                         verify(nvlist_lookup_string(tmp,
 507                                             ZPOOL_CONFIG_HOSTNAME,
 508                                             &hostname) == 0);
 509                                         if (nvlist_add_string(config,
 510                                             ZPOOL_CONFIG_HOSTNAME,
 511                                             hostname) != 0)
 512                                                 goto nomem;
 513                                 }
 514
 515                                 config_seen = B_TRUE;
 516                         }
 517
 518                         /*
 519                          * Add this top-level vdev to the child array.
 520                          */
 521                         verify(nvlist_lookup_nvlist(tmp,
 522                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 523                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 524                             &id) == 0);
 525                         if (id >= children) {
 526                                 nvlist_t **newchild;
 527
 528                                 newchild = zfs_alloc(hdl, (id + 1) *
 529                                     sizeof (nvlist_t *));
 530                                 if (newchild == NULL)
 531                                         goto nomem;
 532
 533                                 for (c = 0; c < children; c++)
 534                                         newchild[c] = child[c];
 535
 536                                 free(child);
 537                                 child = newchild;
 538                                 children = id + 1;
 539                         }
 540                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 541                                 goto nomem;
 542
 543                 }
 544
 545                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 546                     &guid) == 0);
 547
 548                 /*
 549                  * Look for any missing top-level vdevs.  If this is the case,
 550                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 551                  * simply compress the child array, because the kernel performs
 552                  * certain checks to make sure the vdev IDs match their location
 553                  * in the configuration.
 554                  */
 555                 for (c = 0; c < children; c++)
 556                         if (child[c] == NULL) {
 557                                 nvlist_t *missing;
 558                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 559                                     0) != 0)
 560                                         goto nomem;
 561                                 if (nvlist_add_string(missing,
 562                                     ZPOOL_CONFIG_TYPE,
 563                                     VDEV_TYPE_MISSING) != 0 ||
 564                                     nvlist_add_uint64(missing,
 565                                     ZPOOL_CONFIG_ID, c) != 0 ||
 566                                     nvlist_add_uint64(missing,
 567                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 568                                         nvlist_free(missing);
 569                                         goto nomem;
 570                                 }
 571                                 child[c] = missing;
 572                         }
 573
 574                 /*
 575                  * Put all of this pool's top-level vdevs into a root vdev.
 576                  */
 577                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 578                         goto nomem;
 579                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 580                     VDEV_TYPE_ROOT) != 0 ||
 581                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 582                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 583                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 584                     child, children) != 0) {
 585                         nvlist_free(nvroot);
 586                         goto nomem;
 587                 }
 588
 589                 for (c = 0; c < children; c++)
 590                         nvlist_free(child[c]);
 591                 free(child);
 592                 children = 0;
 593                 child = NULL;
 594
 595                 /*
 596                  * Go through and fix up any paths and/or devids based on our
 597                  * known list of vdev GUID -> path mappings.
 598                  */
 599                 if (fix_paths(nvroot, pl->names) != 0) {
 600                         nvlist_free(nvroot);
 601                         goto nomem;
 602                 }
 603
 604                 /*
 605                  * Add the root vdev to this pool's configuration.
 606                  */
 607                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 608                     nvroot) != 0) {
 609                         nvlist_free(nvroot);
 610                         goto nomem;
 611                 }
 612                 nvlist_free(nvroot);
 613
 614                 /*
 615                  * zdb uses this path to report on active pools that were
 616                  * imported or created using -R.
 617                  */
 618                 if (active_ok)
 619                         goto add_pool;
 620
 621                 /*
 622                  * Determine if this pool is currently active, in which case we
 623                  * can't actually import it.
 624                  */
 625                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 626                     &name) == 0);
 627                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 628                     &guid) == 0);
 629
 630                 if (pool_active(hdl, name, guid, &isactive) != 0)
 631                         goto error;
 632
 633                 if (isactive) {
 634                         nvlist_free(config);
 635                         config = NULL;
 636                         continue;
 637                 }
 638
 639                 if ((nvl = refresh_config(hdl, config)) == NULL)
 640                         goto error;
 641
 642                 nvlist_free(config);
 643                 config = nvl;
 644
 645                 /*
 646                  * Go through and update the paths for spares, now that we have
 647                  * them.
 648                  */
 649                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 650                     &nvroot) == 0);
 651                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 652                     &spares, &nspares) == 0) {
 653                         for (i = 0; i < nspares; i++) {
 654                                 if (fix_paths(spares[i], pl->names) != 0)
 655                                         goto nomem;
 656                         }
 657                 }
 658
 659                 /*
 660                  * Update the paths for l2cache devices.
 661                  */
 662                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 663                     &l2cache, &nl2cache) == 0) {
 664                         for (i = 0; i < nl2cache; i++) {
 665                                 if (fix_paths(l2cache[i], pl->names) != 0)
 666                                         goto nomem;
 667                         }
 668                 }
 669
 670                 /*
 671                  * Restore the original information read from the actual label.
 672                  */
 673                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 674                     DATA_TYPE_UINT64);
 675                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 676                     DATA_TYPE_STRING);
 677                 if (hostid != 0) {
 678                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 679                             hostid) == 0);
 680                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 681                             hostname) == 0);
 682                 }
 683
 684 add_pool:
 685                 /*
 686                  * Add this pool to the list of configs.
 687                  */
 688                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 689                     &name) == 0);
 690                 if (nvlist_add_nvlist(ret, name, config) != 0)
 691                         goto nomem;
 692
 693                 found_one = B_TRUE;
 694                 nvlist_free(config);
 695                 config = NULL;
 696         }
 697
 698         if (!found_one) {
 699                 nvlist_free(ret);
 700                 ret = NULL;
 701         }
 702
 703         return (ret);
 704
 705 nomem:
 706         (void) no_memory(hdl);
 707 error:
 708         nvlist_free(config);
 709         nvlist_free(ret);
 710         for (c = 0; c < children; c++)
 711                 nvlist_free(child[c]);
 712         free(child);
 713
 714         return (NULL);
 715 }
 716
 717 /*
 718  * Return the offset of the given label.
 719  */
 720 static uint64_t
 721 label_offset(uint64_t size, int l)
 722 {
 723         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 724         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 725             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 726 }
 727
 728 /*
 729  * Given a file descriptor, read the label information and return an nvlist
 730  * describing the configuration, if there is one.
 731  */
 732 int
 733 zpool_read_label(int fd, nvlist_t **config)
 734 {
 735         struct stat64 statbuf;
 736         int l;
 737         vdev_label_t *label;
 738         uint64_t state, txg, size;
 739
 740         *config = NULL;
 741
 742         if (fstat64(fd, &statbuf) == -1)
 743                 return (0);
 744         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 745
 746         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 747                 return (-1);
 748
 749         for (l = 0; l < VDEV_LABELS; l++) {
 750                 if (pread64(fd, label, sizeof (vdev_label_t),
 751                     label_offset(size, l)) != sizeof (vdev_label_t))
 752                         continue;
 753
 754                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 755                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 756                         continue;
 757
 758                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 759                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 760                         nvlist_free(*config);
 761                         continue;
 762                 }
 763
 764                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 765                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 766                     &txg) != 0 || txg == 0)) {
 767                         nvlist_free(*config);
 768                         continue;
 769                 }
 770
 771                 free(label);
 772                 return (0);
 773         }
 774
 775         free(label);
 776         *config = NULL;
 777         return (0);
 778 }
 779
 780 /*
 781  * Given a list of directories to search, find all pools stored on disk.  This
 782  * includes partial pools which are not available to import.  If no args are
 783  * given (argc is 0), then the default directory (/dev/dsk) is searched.
 784  * poolname or guid (but not both) are provided by the caller when trying
 785  * to import a specific pool.
 786  */
 787 static nvlist_t *
 788 zpool_find_import_impl(libzfs_handle_t *hdl, int argc, char **argv,
 789     boolean_t active_ok, char *poolname, uint64_t guid)
 790 {
 791         int i;
 792         DIR *dirp = NULL;
 793         struct dirent64 *dp;
 794         char path[MAXPATHLEN];
 795         char *end;
 796         size_t pathleft;
 797         struct stat64 statbuf;
 798         nvlist_t *ret = NULL, *config;
 799         static char *default_dir = "/dev/dsk";
 800         int fd;
 801         pool_list_t pools = { 0 };
 802         pool_entry_t *pe, *penext;
 803         vdev_entry_t *ve, *venext;
 804         config_entry_t *ce, *cenext;
 805         name_entry_t *ne, *nenext;
 806
 807         verify(poolname == NULL || guid == 0);
 808
 809         if (argc == 0) {
 810                 argc = 1;
 811                 argv = &default_dir;
 812         }
 813
 814         /*
 815          * Go through and read the label configuration information from every
 816          * possible device, organizing the information according to pool GUID
 817          * and toplevel GUID.
 818          */
 819         for (i = 0; i < argc; i++) {
 820                 char *rdsk;
 821                 int dfd;
 822
 823                 /* use realpath to normalize the path */
 824                 if (realpath(argv[i], path) == 0) {
 825                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
 826                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 827                             argv[i]);
 828                         goto error;
 829                 }
 830                 end = &path[strlen(path)];
 831                 *end++ = '/';
 832                 *end = 0;
 833                 pathleft = &path[sizeof (path)] - end;
 834
 835                 /*
 836                  * Using raw devices instead of block devices when we're
 837                  * reading the labels skips a bunch of slow operations during
 838                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
 839                  */
 840                 if (strcmp(path, "/dev/dsk/") == 0)
 841                         rdsk = "/dev/rdsk/";
 842                 else
 843                         rdsk = path;
 844
 845                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
 846                     (dirp = fdopendir(dfd)) == NULL) {
 847                         zfs_error_aux(hdl, strerror(errno));
 848                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
 849                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 850                             rdsk);
 851                         goto error;
 852                 }
 853
 854                 /*
 855                  * This is not MT-safe, but we have no MT consumers of libzfs
 856                  */
 857                 while ((dp = readdir64(dirp)) != NULL) {
 858                         const char *name = dp->d_name;
 859                         if (name[0] == '.' &&
 860                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
 861                                 continue;
 862
 863                         if ((fd = openat64(dfd, name, O_RDONLY)) < 0)
 864                                 continue;
 865
 866                         /*
 867                          * Ignore failed stats.  We only want regular
 868                          * files, character devs and block devs.
 869                          */
 870                         if (fstat64(fd, &statbuf) != 0 ||
 871                             (!S_ISREG(statbuf.st_mode) &&
 872                             !S_ISCHR(statbuf.st_mode) &&
 873                             !S_ISBLK(statbuf.st_mode))) {
 874                                 (void) close(fd);
 875                                 continue;
 876                         }
 877
 878                         if ((zpool_read_label(fd, &config)) != 0) {
 879                                 (void) close(fd);
 880                                 (void) no_memory(hdl);
 881                                 goto error;
 882                         }
 883
 884                         (void) close(fd);
 885
 886                         if (config != NULL) {
 887                                 boolean_t matched = B_TRUE;
 888
 889                                 if (poolname != NULL) {
 890                                         char *pname;
 891
 892                                         matched = nvlist_lookup_string(config,
 893                                             ZPOOL_CONFIG_POOL_NAME,
 894                                             &pname) == 0 &&
 895                                             strcmp(poolname, pname) == 0;
 896                                 } else if (guid != 0) {
 897                                         uint64_t this_guid;
 898
 899                                         matched = nvlist_lookup_uint64(config,
 900                                             ZPOOL_CONFIG_POOL_GUID,
 901                                             &this_guid) == 0 &&
 902                                             guid == this_guid;
 903                                 }
 904                                 if (!matched) {
 905                                         nvlist_free(config);
 906                                         config = NULL;
 907                                         continue;
 908                                 }
 909                                 /* use the non-raw path for the config */
 910                                 (void) strlcpy(end, name, pathleft);
 911                                 if (add_config(hdl, &pools, path, config) != 0)
 912                                         goto error;
 913                         }
 914                 }
 915
 916                 (void) closedir(dirp);
 917                 dirp = NULL;
 918         }
 919
 920         ret = get_configs(hdl, &pools, active_ok);
 921
 922 error:
 923         for (pe = pools.pools; pe != NULL; pe = penext) {
 924                 penext = pe->pe_next;
 925                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
 926                         venext = ve->ve_next;
 927                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
 928                                 cenext = ce->ce_next;
 929                                 if (ce->ce_config)
 930                                         nvlist_free(ce->ce_config);
 931                                 free(ce);
 932                         }
 933                         free(ve);
 934                 }
 935                 free(pe);
 936         }
 937
 938         for (ne = pools.names; ne != NULL; ne = nenext) {
 939                 nenext = ne->ne_next;
 940                 if (ne->ne_name)
 941                         free(ne->ne_name);
 942                 free(ne);
 943         }
 944
 945         if (dirp)
 946                 (void) closedir(dirp);
 947
 948         return (ret);
 949 }
 950
 951 nvlist_t *
 952 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
 953 {
 954         return (zpool_find_import_impl(hdl, argc, argv, B_FALSE, NULL, 0));
 955 }
 956
 957 nvlist_t *
 958 zpool_find_import_byname(libzfs_handle_t *hdl, int argc, char **argv,
 959     char *pool)
 960 {
 961         return (zpool_find_import_impl(hdl, argc, argv, B_FALSE, pool, 0));
 962 }
 963
 964 nvlist_t *
 965 zpool_find_import_byguid(libzfs_handle_t *hdl, int argc, char **argv,
 966     uint64_t guid)
 967 {
 968         return (zpool_find_import_impl(hdl, argc, argv, B_FALSE, NULL, guid));
 969 }
 970
 971 nvlist_t *
 972 zpool_find_import_activeok(libzfs_handle_t *hdl, int argc, char **argv)
 973 {
 974         return (zpool_find_import_impl(hdl, argc, argv, B_TRUE, NULL, 0));
 975 }
 976
 977 /*
 978  * Given a cache file, return the contents as a list of importable pools.
 979  * poolname or guid (but not both) are provided by the caller when trying
 980  * to import a specific pool.
 981  */
 982 nvlist_t *
 983 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
 984     char *poolname, uint64_t guid)
 985 {
 986         char *buf;
 987         int fd;
 988         struct stat64 statbuf;
 989         nvlist_t *raw, *src, *dst;
 990         nvlist_t *pools;
 991         nvpair_t *elem;
 992         char *name;
 993         uint64_t this_guid;
 994         boolean_t active;
 995
 996         verify(poolname == NULL || guid == 0);
 997
 998         if ((fd = open(cachefile, O_RDONLY)) < 0) {
 999                 zfs_error_aux(hdl, "%s", strerror(errno));
1000                 (void) zfs_error(hdl, EZFS_BADCACHE,
1001                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1002                 return (NULL);
1003         }
1004
1005         if (fstat64(fd, &statbuf) != 0) {
1006                 zfs_error_aux(hdl, "%s", strerror(errno));
1007                 (void) close(fd);
1008                 (void) zfs_error(hdl, EZFS_BADCACHE,
1009                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1010                 return (NULL);
1011         }
1012
1013         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1014                 (void) close(fd);
1015                 return (NULL);
1016         }
1017
1018         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1019                 (void) close(fd);
1020                 free(buf);
1021                 (void) zfs_error(hdl, EZFS_BADCACHE,
1022                     dgettext(TEXT_DOMAIN,
1023                     "failed to read cache file contents"));
1024                 return (NULL);
1025         }
1026
1027         (void) close(fd);
1028
1029         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1030                 free(buf);
1031                 (void) zfs_error(hdl, EZFS_BADCACHE,
1032                     dgettext(TEXT_DOMAIN,
1033                     "invalid or corrupt cache file contents"));
1034                 return (NULL);
1035         }
1036
1037         free(buf);
1038
1039         /*
1040          * Go through and get the current state of the pools and refresh their
1041          * state.
1042          */
1043         if (nvlist_alloc(&pools, 0, 0) != 0) {
1044                 (void) no_memory(hdl);
1045                 nvlist_free(raw);
1046                 return (NULL);
1047         }
1048
1049         elem = NULL;
1050         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1051                 verify(nvpair_value_nvlist(elem, &src) == 0);
1052
1053                 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1054                     &name) == 0);
1055                 if (poolname != NULL && strcmp(poolname, name) != 0)
1056                         continue;
1057
1058                 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1059                     &this_guid) == 0);
1060                 if (guid != 0) {
1061                         verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1062                             &this_guid) == 0);
1063                         if (guid != this_guid)
1064                                 continue;
1065                 }
1066
1067                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1068                         nvlist_free(raw);
1069                         nvlist_free(pools);
1070                         return (NULL);
1071                 }
1072
1073                 if (active)
1074                         continue;
1075
1076                 if ((dst = refresh_config(hdl, src)) == NULL) {
1077                         nvlist_free(raw);
1078                         nvlist_free(pools);
1079                         return (NULL);
1080                 }
1081
1082                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1083                         (void) no_memory(hdl);
1084                         nvlist_free(dst);
1085                         nvlist_free(raw);
1086                         nvlist_free(pools);
1087                         return (NULL);
1088                 }
1089                 nvlist_free(dst);
1090         }
1091
1092         nvlist_free(raw);
1093         return (pools);
1094 }
1095
1096
1097 boolean_t
1098 find_guid(nvlist_t *nv, uint64_t guid)
1099 {
1100         uint64_t tmp;
1101         nvlist_t **child;
1102         uint_t c, children;
1103
1104         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1105         if (tmp == guid)
1106                 return (B_TRUE);
1107
1108         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1109             &child, &children) == 0) {
1110                 for (c = 0; c < children; c++)
1111                         if (find_guid(child[c], guid))
1112                                 return (B_TRUE);
1113         }
1114
1115         return (B_FALSE);
1116 }
1117
1118 typedef struct aux_cbdata {
1119         const char      *cb_type;
1120         uint64_t        cb_guid;
1121         zpool_handle_t  *cb_zhp;
1122 } aux_cbdata_t;
1123
1124 static int
1125 find_aux(zpool_handle_t *zhp, void *data)
1126 {
1127         aux_cbdata_t *cbp = data;
1128         nvlist_t **list;
1129         uint_t i, count;
1130         uint64_t guid;
1131         nvlist_t *nvroot;
1132
1133         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1134             &nvroot) == 0);
1135
1136         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1137             &list, &count) == 0) {
1138                 for (i = 0; i < count; i++) {
1139                         verify(nvlist_lookup_uint64(list[i],
1140                             ZPOOL_CONFIG_GUID, &guid) == 0);
1141                         if (guid == cbp->cb_guid) {
1142                                 cbp->cb_zhp = zhp;
1143                                 return (1);
1144                         }
1145                 }
1146         }
1147
1148         zpool_close(zhp);
1149         return (0);
1150 }
1151
1152 /*
1153  * Determines if the pool is in use.  If so, it returns true and the state of
1154  * the pool as well as the name of the pool.  Both strings are allocated and
1155  * must be freed by the caller.
1156  */
1157 int
1158 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1159     boolean_t *inuse)
1160 {
1161         nvlist_t *config;
1162         char *name;
1163         boolean_t ret;
1164         uint64_t guid, vdev_guid;
1165         zpool_handle_t *zhp;
1166         nvlist_t *pool_config;
1167         uint64_t stateval, isspare;
1168         aux_cbdata_t cb = { 0 };
1169         boolean_t isactive;
1170
1171         *inuse = B_FALSE;
1172
1173         if (zpool_read_label(fd, &config) != 0) {
1174                 (void) no_memory(hdl);
1175                 return (-1);
1176         }
1177
1178         if (config == NULL)
1179                 return (0);
1180
1181         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1182             &stateval) == 0);
1183         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1184             &vdev_guid) == 0);
1185
1186         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1187                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1188                     &name) == 0);
1189                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1190                     &guid) == 0);
1191         }
1192
1193         switch (stateval) {
1194         case POOL_STATE_EXPORTED:
1195                 ret = B_TRUE;
1196                 break;
1197
1198         case POOL_STATE_ACTIVE:
1199                 /*
1200                  * For an active pool, we have to determine if it's really part
1201                  * of a currently active pool (in which case the pool will exist
1202                  * and the guid will be the same), or whether it's part of an
1203                  * active pool that was disconnected without being explicitly
1204                  * exported.
1205                  */
1206                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1207                         nvlist_free(config);
1208                         return (-1);
1209                 }
1210
1211                 if (isactive) {
1212                         /*
1213                          * Because the device may have been removed while
1214                          * offlined, we only report it as active if the vdev is
1215                          * still present in the config.  Otherwise, pretend like
1216                          * it's not in use.
1217                          */
1218                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1219                             (pool_config = zpool_get_config(zhp, NULL))
1220                             != NULL) {
1221                                 nvlist_t *nvroot;
1222
1223                                 verify(nvlist_lookup_nvlist(pool_config,
1224                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1225                                 ret = find_guid(nvroot, vdev_guid);
1226                         } else {
1227                                 ret = B_FALSE;
1228                         }
1229
1230                         /*
1231                          * If this is an active spare within another pool, we
1232                          * treat it like an unused hot spare.  This allows the
1233                          * user to create a pool with a hot spare that currently
1234                          * in use within another pool.  Since we return B_TRUE,
1235                          * libdiskmgt will continue to prevent generic consumers
1236                          * from using the device.
1237                          */
1238                         if (ret && nvlist_lookup_uint64(config,
1239                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1240                                 stateval = POOL_STATE_SPARE;
1241
1242                         if (zhp != NULL)
1243                                 zpool_close(zhp);
1244                 } else {
1245                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1246                         ret = B_TRUE;
1247                 }
1248                 break;
1249
1250         case POOL_STATE_SPARE:
1251                 /*
1252                  * For a hot spare, it can be either definitively in use, or
1253                  * potentially active.  To determine if it's in use, we iterate
1254                  * over all pools in the system and search for one with a spare
1255                  * with a matching guid.
1256                  *
1257                  * Due to the shared nature of spares, we don't actually report
1258                  * the potentially active case as in use.  This means the user
1259                  * can freely create pools on the hot spares of exported pools,
1260                  * but to do otherwise makes the resulting code complicated, and
1261                  * we end up having to deal with this case anyway.
1262                  */
1263                 cb.cb_zhp = NULL;
1264                 cb.cb_guid = vdev_guid;
1265                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1266                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1267                         name = (char *)zpool_get_name(cb.cb_zhp);
1268                         ret = TRUE;
1269                 } else {
1270                         ret = FALSE;
1271                 }
1272                 break;
1273
1274         case POOL_STATE_L2CACHE:
1275
1276                 /*
1277                  * Check if any pool is currently using this l2cache device.
1278                  */
1279                 cb.cb_zhp = NULL;
1280                 cb.cb_guid = vdev_guid;
1281                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1282                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1283                         name = (char *)zpool_get_name(cb.cb_zhp);
1284                         ret = TRUE;
1285                 } else {
1286                         ret = FALSE;
1287                 }
1288                 break;
1289
1290         default:
1291                 ret = B_FALSE;
1292         }
1293
1294
1295         if (ret) {
1296                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1297                         if (cb.cb_zhp)
1298                                 zpool_close(cb.cb_zhp);
1299                         nvlist_free(config);
1300                         return (-1);
1301                 }
1302                 *state = (pool_state_t)stateval;
1303         }
1304
1305         if (cb.cb_zhp)
1306                 zpool_close(cb.cb_zhp);
1307
1308         nvlist_free(config);
1309         *inuse = ret;
1310         return (0);
1311 }