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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  25  * Copyright (c) 2013 by Delphix. All rights reserved.
  26  */
  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 <ctype.h>
  45 #include <devid.h>
  46 #include <dirent.h>
  47 #include <errno.h>
  48 #include <libintl.h>
  49 #include <stddef.h>
  50 #include <stdlib.h>
  51 #include <string.h>
  52 #include <sys/stat.h>
  53 #include <unistd.h>
  54 #include <fcntl.h>
  55 #include <thread_pool.h>
  56 #include <libgeom.h>
  57
  58 #include <sys/vdev_impl.h>
  59
  60 #include "libzfs.h"
  61 #include "libzfs_impl.h"
  62
  63 /*
  64  * Intermediate structures used to gather configuration information.
  65  */
  66 typedef struct config_entry {
  67         uint64_t                ce_txg;
  68         nvlist_t                *ce_config;
  69         struct config_entry     *ce_next;
  70 } config_entry_t;
  71
  72 typedef struct vdev_entry {
  73         uint64_t                ve_guid;
  74         config_entry_t          *ve_configs;
  75         struct vdev_entry       *ve_next;
  76 } vdev_entry_t;
  77
  78 typedef struct pool_entry {
  79         uint64_t                pe_guid;
  80         vdev_entry_t            *pe_vdevs;
  81         struct pool_entry       *pe_next;
  82 } pool_entry_t;
  83
  84 typedef struct name_entry {
  85         char                    *ne_name;
  86         uint64_t                ne_guid;
  87         struct name_entry       *ne_next;
  88 } name_entry_t;
  89
  90 typedef struct pool_list {
  91         pool_entry_t            *pools;
  92         name_entry_t            *names;
  93 } pool_list_t;
  94
  95 static char *
  96 get_devid(const char *path)
  97 {
  98 #ifdef have_devid
  99         int fd;
 100         ddi_devid_t devid;
 101         char *minor, *ret;
 102
 103         if ((fd = open(path, O_RDONLY)) < 0)
 104                 return (NULL);
 105
 106         minor = NULL;
 107         ret = NULL;
 108         if (devid_get(fd, &devid) == 0) {
 109                 if (devid_get_minor_name(fd, &minor) == 0)
 110                         ret = devid_str_encode(devid, minor);
 111                 if (minor != NULL)
 112                         devid_str_free(minor);
 113                 devid_free(devid);
 114         }
 115         (void) close(fd);
 116
 117         return (ret);
 118 #else
 119         return (NULL);
 120 #endif
 121 }
 122
 123
 124 /*
 125  * Go through and fix up any path and/or devid information for the given vdev
 126  * configuration.
 127  */
 128 static int
 129 fix_paths(nvlist_t *nv, name_entry_t *names)
 130 {
 131         nvlist_t **child;
 132         uint_t c, children;
 133         uint64_t guid;
 134         name_entry_t *ne, *best;
 135         char *path, *devid;
 136         int matched;
 137
 138         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 139             &child, &children) == 0) {
 140                 for (c = 0; c < children; c++)
 141                         if (fix_paths(child[c], names) != 0)
 142                                 return (-1);
 143                 return (0);
 144         }
 145
 146         /*
 147          * This is a leaf (file or disk) vdev.  In either case, go through
 148          * the name list and see if we find a matching guid.  If so, replace
 149          * the path and see if we can calculate a new devid.
 150          *
 151          * There may be multiple names associated with a particular guid, in
 152          * which case we have overlapping slices or multiple paths to the same
 153          * disk.  If this is the case, then we want to pick the path that is
 154          * the most similar to the original, where "most similar" is the number
 155          * of matching characters starting from the end of the path.  This will
 156          * preserve slice numbers even if the disks have been reorganized, and
 157          * will also catch preferred disk names if multiple paths exist.
 158          */
 159         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 160         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 161                 path = NULL;
 162
 163         matched = 0;
 164         best = NULL;
 165         for (ne = names; ne != NULL; ne = ne->ne_next) {
 166                 if (ne->ne_guid == guid) {
 167                         const char *src, *dst;
 168                         int count;
 169
 170                         if (path == NULL) {
 171                                 best = ne;
 172                                 break;
 173                         }
 174
 175                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 176                         dst = path + strlen(path) - 1;
 177                         for (count = 0; src >= ne->ne_name && dst >= path;
 178                             src--, dst--, count++)
 179                                 if (*src != *dst)
 180                                         break;
 181
 182                         /*
 183                          * At this point, 'count' is the number of characters
 184                          * matched from the end.
 185                          */
 186                         if (count > matched || best == NULL) {
 187                                 best = ne;
 188                                 matched = count;
 189                         }
 190                 }
 191         }
 192
 193         if (best == NULL)
 194                 return (0);
 195
 196         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 197                 return (-1);
 198
 199         if ((devid = get_devid(best->ne_name)) == NULL) {
 200                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 201         } else {
 202                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
 203                         devid_str_free(devid);
 204                         return (-1);
 205                 }
 206                 devid_str_free(devid);
 207         }
 208
 209         return (0);
 210 }
 211
 212 /*
 213  * Add the given configuration to the list of known devices.
 214  */
 215 static int
 216 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 217     nvlist_t *config)
 218 {
 219         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 220         pool_entry_t *pe;
 221         vdev_entry_t *ve;
 222         config_entry_t *ce;
 223         name_entry_t *ne;
 224
 225         /*
 226          * If this is a hot spare not currently in use or level 2 cache
 227          * device, add it to the list of names to translate, but don't do
 228          * anything else.
 229          */
 230         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 231             &state) == 0 &&
 232             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 233             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 234                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 235                         return (-1);
 236
 237                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 238                         free(ne);
 239                         return (-1);
 240                 }
 241                 ne->ne_guid = vdev_guid;
 242                 ne->ne_next = pl->names;
 243                 pl->names = ne;
 244                 return (0);
 245         }
 246
 247         /*
 248          * If we have a valid config but cannot read any of these fields, then
 249          * it means we have a half-initialized label.  In vdev_label_init()
 250          * we write a label with txg == 0 so that we can identify the device
 251          * in case the user refers to the same disk later on.  If we fail to
 252          * create the pool, we'll be left with a label in this state
 253          * which should not be considered part of a valid pool.
 254          */
 255         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 256             &pool_guid) != 0 ||
 257             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 258             &vdev_guid) != 0 ||
 259             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 260             &top_guid) != 0 ||
 261             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 262             &txg) != 0 || txg == 0) {
 263                 nvlist_free(config);
 264                 return (0);
 265         }
 266
 267         /*
 268          * First, see if we know about this pool.  If not, then add it to the
 269          * list of known pools.
 270          */
 271         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 272                 if (pe->pe_guid == pool_guid)
 273                         break;
 274         }
 275
 276         if (pe == NULL) {
 277                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 278                         nvlist_free(config);
 279                         return (-1);
 280                 }
 281                 pe->pe_guid = pool_guid;
 282                 pe->pe_next = pl->pools;
 283                 pl->pools = pe;
 284         }
 285
 286         /*
 287          * Second, see if we know about this toplevel vdev.  Add it if its
 288          * missing.
 289          */
 290         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 291                 if (ve->ve_guid == top_guid)
 292                         break;
 293         }
 294
 295         if (ve == NULL) {
 296                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 297                         nvlist_free(config);
 298                         return (-1);
 299                 }
 300                 ve->ve_guid = top_guid;
 301                 ve->ve_next = pe->pe_vdevs;
 302                 pe->pe_vdevs = ve;
 303         }
 304
 305         /*
 306          * Third, see if we have a config with a matching transaction group.  If
 307          * so, then we do nothing.  Otherwise, add it to the list of known
 308          * configs.
 309          */
 310         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 311                 if (ce->ce_txg == txg)
 312                         break;
 313         }
 314
 315         if (ce == NULL) {
 316                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 317                         nvlist_free(config);
 318                         return (-1);
 319                 }
 320                 ce->ce_txg = txg;
 321                 ce->ce_config = config;
 322                 ce->ce_next = ve->ve_configs;
 323                 ve->ve_configs = ce;
 324         } else {
 325                 nvlist_free(config);
 326         }
 327
 328         /*
 329          * At this point we've successfully added our config to the list of
 330          * known configs.  The last thing to do is add the vdev guid -> path
 331          * mappings so that we can fix up the configuration as necessary before
 332          * doing the import.
 333          */
 334         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 335                 return (-1);
 336
 337         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 338                 free(ne);
 339                 return (-1);
 340         }
 341
 342         ne->ne_guid = vdev_guid;
 343         ne->ne_next = pl->names;
 344         pl->names = ne;
 345
 346         return (0);
 347 }
 348
 349 /*
 350  * Returns true if the named pool matches the given GUID.
 351  */
 352 static int
 353 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 354     boolean_t *isactive)
 355 {
 356         zpool_handle_t *zhp;
 357         uint64_t theguid;
 358
 359         if (zpool_open_silent(hdl, name, &zhp) != 0)
 360                 return (-1);
 361
 362         if (zhp == NULL) {
 363                 *isactive = B_FALSE;
 364                 return (0);
 365         }
 366
 367         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 368             &theguid) == 0);
 369
 370         zpool_close(zhp);
 371
 372         *isactive = (theguid == guid);
 373         return (0);
 374 }
 375
 376 static nvlist_t *
 377 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 378 {
 379         nvlist_t *nvl;
 380         zfs_cmd_t zc = { 0 };
 381         int err;
 382
 383         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 384                 return (NULL);
 385
 386         if (zcmd_alloc_dst_nvlist(hdl, &zc,
 387             zc.zc_nvlist_conf_size * 2) != 0) {
 388                 zcmd_free_nvlists(&zc);
 389                 return (NULL);
 390         }
 391
 392         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 393             &zc)) != 0 && errno == ENOMEM) {
 394                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 395                         zcmd_free_nvlists(&zc);
 396                         return (NULL);
 397                 }
 398         }
 399
 400         if (err) {
 401                 zcmd_free_nvlists(&zc);
 402                 return (NULL);
 403         }
 404
 405         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 406                 zcmd_free_nvlists(&zc);
 407                 return (NULL);
 408         }
 409
 410         zcmd_free_nvlists(&zc);
 411         return (nvl);
 412 }
 413
 414 /*
 415  * Determine if the vdev id is a hole in the namespace.
 416  */
 417 boolean_t
 418 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
 419 {
 420         for (int c = 0; c < holes; c++) {
 421
 422                 /* Top-level is a hole */
 423                 if (hole_array[c] == id)
 424                         return (B_TRUE);
 425         }
 426         return (B_FALSE);
 427 }
 428
 429 /*
 430  * Convert our list of pools into the definitive set of configurations.  We
 431  * start by picking the best config for each toplevel vdev.  Once that's done,
 432  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 433  * pass to fix up any incorrect paths, and then add it to the main list to
 434  * return to the user.
 435  */
 436 static nvlist_t *
 437 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
 438 {
 439         pool_entry_t *pe;
 440         vdev_entry_t *ve;
 441         config_entry_t *ce;
 442         nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
 443         nvlist_t **spares, **l2cache;
 444         uint_t i, nspares, nl2cache;
 445         boolean_t config_seen;
 446         uint64_t best_txg;
 447         char *name, *hostname;
 448         uint64_t guid;
 449         uint_t children = 0;
 450         nvlist_t **child = NULL;
 451         uint_t holes;
 452         uint64_t *hole_array, max_id;
 453         uint_t c;
 454         boolean_t isactive;
 455         uint64_t hostid;
 456         nvlist_t *nvl;
 457         boolean_t found_one = B_FALSE;
 458         boolean_t valid_top_config = B_FALSE;
 459
 460         if (nvlist_alloc(&ret, 0, 0) != 0)
 461                 goto nomem;
 462
 463         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 464                 uint64_t id, max_txg = 0;
 465
 466                 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
 467                         goto nomem;
 468                 config_seen = B_FALSE;
 469
 470                 /*
 471                  * Iterate over all toplevel vdevs.  Grab the pool configuration
 472                  * from the first one we find, and then go through the rest and
 473                  * add them as necessary to the 'vdevs' member of the config.
 474                  */
 475                 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 476
 477                         /*
 478                          * Determine the best configuration for this vdev by
 479                          * selecting the config with the latest transaction
 480                          * group.
 481                          */
 482                         best_txg = 0;
 483                         for (ce = ve->ve_configs; ce != NULL;
 484                             ce = ce->ce_next) {
 485
 486                                 if (ce->ce_txg > best_txg) {
 487                                         tmp = ce->ce_config;
 488                                         best_txg = ce->ce_txg;
 489                                 }
 490                         }
 491
 492                         /*
 493                          * We rely on the fact that the max txg for the
 494                          * pool will contain the most up-to-date information
 495                          * about the valid top-levels in the vdev namespace.
 496                          */
 497                         if (best_txg > max_txg) {
 498                                 (void) nvlist_remove(config,
 499                                     ZPOOL_CONFIG_VDEV_CHILDREN,
 500                                     DATA_TYPE_UINT64);
 501                                 (void) nvlist_remove(config,
 502                                     ZPOOL_CONFIG_HOLE_ARRAY,
 503                                     DATA_TYPE_UINT64_ARRAY);
 504
 505                                 max_txg = best_txg;
 506                                 hole_array = NULL;
 507                                 holes = 0;
 508                                 max_id = 0;
 509                                 valid_top_config = B_FALSE;
 510
 511                                 if (nvlist_lookup_uint64(tmp,
 512                                     ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
 513                                         verify(nvlist_add_uint64(config,
 514                                             ZPOOL_CONFIG_VDEV_CHILDREN,
 515                                             max_id) == 0);
 516                                         valid_top_config = B_TRUE;
 517                                 }
 518
 519                                 if (nvlist_lookup_uint64_array(tmp,
 520                                     ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
 521                                     &holes) == 0) {
 522                                         verify(nvlist_add_uint64_array(config,
 523                                             ZPOOL_CONFIG_HOLE_ARRAY,
 524                                             hole_array, holes) == 0);
 525                                 }
 526                         }
 527
 528                         if (!config_seen) {
 529                                 /*
 530                                  * Copy the relevant pieces of data to the pool
 531                                  * configuration:
 532                                  *
 533                                  *      version
 534                                  *      pool guid
 535                                  *      name
 536                                  *      comment (if available)
 537                                  *      pool state
 538                                  *      hostid (if available)
 539                                  *      hostname (if available)
 540                                  */
 541                                 uint64_t state, version;
 542                                 char *comment = NULL;
 543
 544                                 version = fnvlist_lookup_uint64(tmp,
 545                                     ZPOOL_CONFIG_VERSION);
 546                                 fnvlist_add_uint64(config,
 547                                     ZPOOL_CONFIG_VERSION, version);
 548                                 guid = fnvlist_lookup_uint64(tmp,
 549                                     ZPOOL_CONFIG_POOL_GUID);
 550                                 fnvlist_add_uint64(config,
 551                                     ZPOOL_CONFIG_POOL_GUID, guid);
 552                                 name = fnvlist_lookup_string(tmp,
 553                                     ZPOOL_CONFIG_POOL_NAME);
 554                                 fnvlist_add_string(config,
 555                                     ZPOOL_CONFIG_POOL_NAME, name);
 556
 557                                 if (nvlist_lookup_string(tmp,
 558                                     ZPOOL_CONFIG_COMMENT, &comment) == 0)
 559                                         fnvlist_add_string(config,
 560                                             ZPOOL_CONFIG_COMMENT, comment);
 561
 562                                 state = fnvlist_lookup_uint64(tmp,
 563                                     ZPOOL_CONFIG_POOL_STATE);
 564                                 fnvlist_add_uint64(config,
 565                                     ZPOOL_CONFIG_POOL_STATE, state);
 566
 567                                 hostid = 0;
 568                                 if (nvlist_lookup_uint64(tmp,
 569                                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
 570                                         fnvlist_add_uint64(config,
 571                                             ZPOOL_CONFIG_HOSTID, hostid);
 572                                         hostname = fnvlist_lookup_string(tmp,
 573                                             ZPOOL_CONFIG_HOSTNAME);
 574                                         fnvlist_add_string(config,
 575                                             ZPOOL_CONFIG_HOSTNAME, hostname);
 576                                 }
 577
 578                                 config_seen = B_TRUE;
 579                         }
 580
 581                         /*
 582                          * Add this top-level vdev to the child array.
 583                          */
 584                         verify(nvlist_lookup_nvlist(tmp,
 585                             ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
 586                         verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
 587                             &id) == 0);
 588
 589                         if (id >= children) {
 590                                 nvlist_t **newchild;
 591
 592                                 newchild = zfs_alloc(hdl, (id + 1) *
 593                                     sizeof (nvlist_t *));
 594                                 if (newchild == NULL)
 595                                         goto nomem;
 596
 597                                 for (c = 0; c < children; c++)
 598                                         newchild[c] = child[c];
 599
 600                                 free(child);
 601                                 child = newchild;
 602                                 children = id + 1;
 603                         }
 604                         if (nvlist_dup(nvtop, &child[id], 0) != 0)
 605                                 goto nomem;
 606
 607                 }
 608
 609                 /*
 610                  * If we have information about all the top-levels then
 611                  * clean up the nvlist which we've constructed. This
 612                  * means removing any extraneous devices that are
 613                  * beyond the valid range or adding devices to the end
 614                  * of our array which appear to be missing.
 615                  */
 616                 if (valid_top_config) {
 617                         if (max_id < children) {
 618                                 for (c = max_id; c < children; c++)
 619                                         nvlist_free(child[c]);
 620                                 children = max_id;
 621                         } else if (max_id > children) {
 622                                 nvlist_t **newchild;
 623
 624                                 newchild = zfs_alloc(hdl, (max_id) *
 625                                     sizeof (nvlist_t *));
 626                                 if (newchild == NULL)
 627                                         goto nomem;
 628
 629                                 for (c = 0; c < children; c++)
 630                                         newchild[c] = child[c];
 631
 632                                 free(child);
 633                                 child = newchild;
 634                                 children = max_id;
 635                         }
 636                 }
 637
 638                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 639                     &guid) == 0);
 640
 641                 /*
 642                  * The vdev namespace may contain holes as a result of
 643                  * device removal. We must add them back into the vdev
 644                  * tree before we process any missing devices.
 645                  */
 646                 if (holes > 0) {
 647                         ASSERT(valid_top_config);
 648
 649                         for (c = 0; c < children; c++) {
 650                                 nvlist_t *holey;
 651
 652                                 if (child[c] != NULL ||
 653                                     !vdev_is_hole(hole_array, holes, c))
 654                                         continue;
 655
 656                                 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
 657                                     0) != 0)
 658                                         goto nomem;
 659
 660                                 /*
 661                                  * Holes in the namespace are treated as
 662                                  * "hole" top-level vdevs and have a
 663                                  * special flag set on them.
 664                                  */
 665                                 if (nvlist_add_string(holey,
 666                                     ZPOOL_CONFIG_TYPE,
 667                                     VDEV_TYPE_HOLE) != 0 ||
 668                                     nvlist_add_uint64(holey,
 669                                     ZPOOL_CONFIG_ID, c) != 0 ||
 670                                     nvlist_add_uint64(holey,
 671                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 672                                         nvlist_free(holey);
 673                                         goto nomem;
 674                                 }
 675                                 child[c] = holey;
 676                         }
 677                 }
 678
 679                 /*
 680                  * Look for any missing top-level vdevs.  If this is the case,
 681                  * create a faked up 'missing' vdev as a placeholder.  We cannot
 682                  * simply compress the child array, because the kernel performs
 683                  * certain checks to make sure the vdev IDs match their location
 684                  * in the configuration.
 685                  */
 686                 for (c = 0; c < children; c++) {
 687                         if (child[c] == NULL) {
 688                                 nvlist_t *missing;
 689                                 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
 690                                     0) != 0)
 691                                         goto nomem;
 692                                 if (nvlist_add_string(missing,
 693                                     ZPOOL_CONFIG_TYPE,
 694                                     VDEV_TYPE_MISSING) != 0 ||
 695                                     nvlist_add_uint64(missing,
 696                                     ZPOOL_CONFIG_ID, c) != 0 ||
 697                                     nvlist_add_uint64(missing,
 698                                     ZPOOL_CONFIG_GUID, 0ULL) != 0) {
 699                                         nvlist_free(missing);
 700                                         goto nomem;
 701                                 }
 702                                 child[c] = missing;
 703                         }
 704                 }
 705
 706                 /*
 707                  * Put all of this pool's top-level vdevs into a root vdev.
 708                  */
 709                 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
 710                         goto nomem;
 711                 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
 712                     VDEV_TYPE_ROOT) != 0 ||
 713                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
 714                     nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
 715                     nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
 716                     child, children) != 0) {
 717                         nvlist_free(nvroot);
 718                         goto nomem;
 719                 }
 720
 721                 for (c = 0; c < children; c++)
 722                         nvlist_free(child[c]);
 723                 free(child);
 724                 children = 0;
 725                 child = NULL;
 726
 727                 /*
 728                  * Go through and fix up any paths and/or devids based on our
 729                  * known list of vdev GUID -> path mappings.
 730                  */
 731                 if (fix_paths(nvroot, pl->names) != 0) {
 732                         nvlist_free(nvroot);
 733                         goto nomem;
 734                 }
 735
 736                 /*
 737                  * Add the root vdev to this pool's configuration.
 738                  */
 739                 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 740                     nvroot) != 0) {
 741                         nvlist_free(nvroot);
 742                         goto nomem;
 743                 }
 744                 nvlist_free(nvroot);
 745
 746                 /*
 747                  * zdb uses this path to report on active pools that were
 748                  * imported or created using -R.
 749                  */
 750                 if (active_ok)
 751                         goto add_pool;
 752
 753                 /*
 754                  * Determine if this pool is currently active, in which case we
 755                  * can't actually import it.
 756                  */
 757                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 758                     &name) == 0);
 759                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 760                     &guid) == 0);
 761
 762                 if (pool_active(hdl, name, guid, &isactive) != 0)
 763                         goto error;
 764
 765                 if (isactive) {
 766                         nvlist_free(config);
 767                         config = NULL;
 768                         continue;
 769                 }
 770
 771                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 772                         nvlist_free(config);
 773                         config = NULL;
 774                         continue;
 775                 }
 776
 777                 nvlist_free(config);
 778                 config = nvl;
 779
 780                 /*
 781                  * Go through and update the paths for spares, now that we have
 782                  * them.
 783                  */
 784                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 785                     &nvroot) == 0);
 786                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 787                     &spares, &nspares) == 0) {
 788                         for (i = 0; i < nspares; i++) {
 789                                 if (fix_paths(spares[i], pl->names) != 0)
 790                                         goto nomem;
 791                         }
 792                 }
 793
 794                 /*
 795                  * Update the paths for l2cache devices.
 796                  */
 797                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 798                     &l2cache, &nl2cache) == 0) {
 799                         for (i = 0; i < nl2cache; i++) {
 800                                 if (fix_paths(l2cache[i], pl->names) != 0)
 801                                         goto nomem;
 802                         }
 803                 }
 804
 805                 /*
 806                  * Restore the original information read from the actual label.
 807                  */
 808                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 809                     DATA_TYPE_UINT64);
 810                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 811                     DATA_TYPE_STRING);
 812                 if (hostid != 0) {
 813                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 814                             hostid) == 0);
 815                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 816                             hostname) == 0);
 817                 }
 818
 819 add_pool:
 820                 /*
 821                  * Add this pool to the list of configs.
 822                  */
 823                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 824                     &name) == 0);
 825                 if (nvlist_add_nvlist(ret, name, config) != 0)
 826                         goto nomem;
 827
 828                 found_one = B_TRUE;
 829                 nvlist_free(config);
 830                 config = NULL;
 831         }
 832
 833         if (!found_one) {
 834                 nvlist_free(ret);
 835                 ret = NULL;
 836         }
 837
 838         return (ret);
 839
 840 nomem:
 841         (void) no_memory(hdl);
 842 error:
 843         nvlist_free(config);
 844         nvlist_free(ret);
 845         for (c = 0; c < children; c++)
 846                 nvlist_free(child[c]);
 847         free(child);
 848
 849         return (NULL);
 850 }
 851
 852 /*
 853  * Return the offset of the given label.
 854  */
 855 static uint64_t
 856 label_offset(uint64_t size, int l)
 857 {
 858         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 859         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 860             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 861 }
 862
 863 /*
 864  * Given a file descriptor, read the label information and return an nvlist
 865  * describing the configuration, if there is one.
 866  */
 867 int
 868 zpool_read_label(int fd, nvlist_t **config)
 869 {
 870         struct stat64 statbuf;
 871         int l;
 872         vdev_label_t *label;
 873         uint64_t state, txg, size;
 874
 875         *config = NULL;
 876
 877         if (fstat64(fd, &statbuf) == -1)
 878                 return (0);
 879         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 880
 881         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 882                 return (-1);
 883
 884         for (l = 0; l < VDEV_LABELS; l++) {
 885                 if (pread64(fd, label, sizeof (vdev_label_t),
 886                     label_offset(size, l)) != sizeof (vdev_label_t))
 887                         continue;
 888
 889                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 890                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 891                         continue;
 892
 893                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 894                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 895                         nvlist_free(*config);
 896                         continue;
 897                 }
 898
 899                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 900                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 901                     &txg) != 0 || txg == 0)) {
 902                         nvlist_free(*config);
 903                         continue;
 904                 }
 905
 906                 free(label);
 907                 return (0);
 908         }
 909
 910         free(label);
 911         *config = NULL;
 912         return (0);
 913 }
 914
 915 typedef struct rdsk_node {
 916         char *rn_name;
 917         int rn_dfd;
 918         libzfs_handle_t *rn_hdl;
 919         nvlist_t *rn_config;
 920         avl_tree_t *rn_avl;
 921         avl_node_t rn_node;
 922         boolean_t rn_nozpool;
 923 } rdsk_node_t;
 924
 925 static int
 926 slice_cache_compare(const void *arg1, const void *arg2)
 927 {
 928         const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
 929         const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
 930         char *nm1slice, *nm2slice;
 931         int rv;
 932
 933         /*
 934          * slices zero and two are the most likely to provide results,
 935          * so put those first
 936          */
 937         nm1slice = strstr(nm1, "s0");
 938         nm2slice = strstr(nm2, "s0");
 939         if (nm1slice && !nm2slice) {
 940                 return (-1);
 941         }
 942         if (!nm1slice && nm2slice) {
 943                 return (1);
 944         }
 945         nm1slice = strstr(nm1, "s2");
 946         nm2slice = strstr(nm2, "s2");
 947         if (nm1slice && !nm2slice) {
 948                 return (-1);
 949         }
 950         if (!nm1slice && nm2slice) {
 951                 return (1);
 952         }
 953
 954         rv = strcmp(nm1, nm2);
 955         if (rv == 0)
 956                 return (0);
 957         return (rv > 0 ? 1 : -1);
 958 }
 959
 960 #ifdef sun
 961 static void
 962 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
 963     diskaddr_t size, uint_t blksz)
 964 {
 965         rdsk_node_t tmpnode;
 966         rdsk_node_t *node;
 967         char sname[MAXNAMELEN];
 968
 969         tmpnode.rn_name = &sname[0];
 970         (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
 971             diskname, partno);
 972         /*
 973          * protect against division by zero for disk labels that
 974          * contain a bogus sector size
 975          */
 976         if (blksz == 0)
 977                 blksz = DEV_BSIZE;
 978         /* too small to contain a zpool? */
 979         if ((size < (SPA_MINDEVSIZE / blksz)) &&
 980             (node = avl_find(r, &tmpnode, NULL)))
 981                 node->rn_nozpool = B_TRUE;
 982 }
 983 #endif  /* sun */
 984
 985 static void
 986 nozpool_all_slices(avl_tree_t *r, const char *sname)
 987 {
 988 #ifdef sun
 989         char diskname[MAXNAMELEN];
 990         char *ptr;
 991         int i;
 992
 993         (void) strncpy(diskname, sname, MAXNAMELEN);
 994         if (((ptr = strrchr(diskname, 's')) == NULL) &&
 995             ((ptr = strrchr(diskname, 'p')) == NULL))
 996                 return;
 997         ptr[0] = 's';
 998         ptr[1] = '\0';
 999         for (i = 0; i < NDKMAP; i++)
1000                 check_one_slice(r, diskname, i, 0, 1);
1001         ptr[0] = 'p';
1002         for (i = 0; i <= FD_NUMPART; i++)
1003                 check_one_slice(r, diskname, i, 0, 1);
1004 #endif  /* sun */
1005 }
1006
1007 #ifdef sun
1008 static void
1009 check_slices(avl_tree_t *r, int fd, const char *sname)
1010 {
1011         struct extvtoc vtoc;
1012         struct dk_gpt *gpt;
1013         char diskname[MAXNAMELEN];
1014         char *ptr;
1015         int i;
1016
1017         (void) strncpy(diskname, sname, MAXNAMELEN);
1018         if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1019                 return;
1020         ptr[1] = '\0';
1021
1022         if (read_extvtoc(fd, &vtoc) >= 0) {
1023                 for (i = 0; i < NDKMAP; i++)
1024                         check_one_slice(r, diskname, i,
1025                             vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1026         } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1027                 /*
1028                  * on x86 we'll still have leftover links that point
1029                  * to slices s[9-15], so use NDKMAP instead
1030                  */
1031                 for (i = 0; i < NDKMAP; i++)
1032                         check_one_slice(r, diskname, i,
1033                             gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1034                 /* nodes p[1-4] are never used with EFI labels */
1035                 ptr[0] = 'p';
1036                 for (i = 1; i <= FD_NUMPART; i++)
1037                         check_one_slice(r, diskname, i, 0, 1);
1038                 efi_free(gpt);
1039         }
1040 }
1041 #endif  /* sun */
1042
1043 static void
1044 zpool_open_func(void *arg)
1045 {
1046         rdsk_node_t *rn = arg;
1047         struct stat64 statbuf;
1048         nvlist_t *config;
1049         int fd;
1050
1051         if (rn->rn_nozpool)
1052                 return;
1053         if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1054                 /* symlink to a device that's no longer there */
1055                 if (errno == ENOENT)
1056                         nozpool_all_slices(rn->rn_avl, rn->rn_name);
1057                 return;
1058         }
1059         /*
1060          * Ignore failed stats.  We only want regular
1061          * files, character devs and block devs.
1062          */
1063         if (fstat64(fd, &statbuf) != 0 ||
1064             (!S_ISREG(statbuf.st_mode) &&
1065             !S_ISCHR(statbuf.st_mode) &&
1066             !S_ISBLK(statbuf.st_mode))) {
1067                 (void) close(fd);
1068                 return;
1069         }
1070         /* this file is too small to hold a zpool */
1071 #ifdef sun
1072         if (S_ISREG(statbuf.st_mode) &&
1073             statbuf.st_size < SPA_MINDEVSIZE) {
1074                 (void) close(fd);
1075                 return;
1076         } else if (!S_ISREG(statbuf.st_mode)) {
1077                 /*
1078                  * Try to read the disk label first so we don't have to
1079                  * open a bunch of minor nodes that can't have a zpool.
1080                  */
1081                 check_slices(rn->rn_avl, fd, rn->rn_name);
1082         }
1083 #else   /* !sun */
1084         if (statbuf.st_size < SPA_MINDEVSIZE) {
1085                 (void) close(fd);
1086                 return;
1087         }
1088 #endif  /* sun */
1089
1090         if ((zpool_read_label(fd, &config)) != 0) {
1091                 (void) close(fd);
1092                 (void) no_memory(rn->rn_hdl);
1093                 return;
1094         }
1095         (void) close(fd);
1096
1097
1098         rn->rn_config = config;
1099         if (config != NULL) {
1100                 assert(rn->rn_nozpool == B_FALSE);
1101         }
1102 }
1103
1104 /*
1105  * Given a file descriptor, clear (zero) the label information.  This function
1106  * is used in the appliance stack as part of the ZFS sysevent module and
1107  * to implement the "zpool labelclear" command.
1108  */
1109 int
1110 zpool_clear_label(int fd)
1111 {
1112         struct stat64 statbuf;
1113         int l;
1114         vdev_label_t *label;
1115         uint64_t size;
1116
1117         if (fstat64(fd, &statbuf) == -1)
1118                 return (0);
1119         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1120
1121         if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1122                 return (-1);
1123
1124         for (l = 0; l < VDEV_LABELS; l++) {
1125                 if (pwrite64(fd, label, sizeof (vdev_label_t),
1126                     label_offset(size, l)) != sizeof (vdev_label_t)) {
1127                         free(label);
1128                         return (-1);
1129                 }
1130         }
1131
1132         free(label);
1133         return (0);
1134 }
1135
1136 /*
1137  * Given a list of directories to search, find all pools stored on disk.  This
1138  * includes partial pools which are not available to import.  If no args are
1139  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1140  * poolname or guid (but not both) are provided by the caller when trying
1141  * to import a specific pool.
1142  */
1143 static nvlist_t *
1144 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1145 {
1146         int i, dirs = iarg->paths;
1147         struct dirent64 *dp;
1148         char path[MAXPATHLEN];
1149         char *end, **dir = iarg->path;
1150         size_t pathleft;
1151         nvlist_t *ret = NULL;
1152         static char *default_dir = "/dev";
1153         pool_list_t pools = { 0 };
1154         pool_entry_t *pe, *penext;
1155         vdev_entry_t *ve, *venext;
1156         config_entry_t *ce, *cenext;
1157         name_entry_t *ne, *nenext;
1158         avl_tree_t slice_cache;
1159         rdsk_node_t *slice;
1160         void *cookie;
1161
1162         if (dirs == 0) {
1163                 dirs = 1;
1164                 dir = &default_dir;
1165         }
1166
1167         /*
1168          * Go through and read the label configuration information from every
1169          * possible device, organizing the information according to pool GUID
1170          * and toplevel GUID.
1171          */
1172         for (i = 0; i < dirs; i++) {
1173                 tpool_t *t;
1174                 char *rdsk;
1175                 int dfd;
1176                 boolean_t config_failed = B_FALSE;
1177                 DIR *dirp;
1178
1179                 /* use realpath to normalize the path */
1180                 if (realpath(dir[i], path) == 0) {
1181                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1182                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1183                         goto error;
1184                 }
1185                 end = &path[strlen(path)];
1186                 *end++ = '/';
1187                 *end = 0;
1188                 pathleft = &path[sizeof (path)] - end;
1189
1190                 /*
1191                  * Using raw devices instead of block devices when we're
1192                  * reading the labels skips a bunch of slow operations during
1193                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1194                  */
1195                 if (strcmp(path, "/dev/dsk/") == 0)
1196                         rdsk = "/dev/";
1197                 else
1198                         rdsk = path;
1199
1200                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1201                     (dirp = fdopendir(dfd)) == NULL) {
1202                         if (dfd >= 0)
1203                                 (void) close(dfd);
1204                         zfs_error_aux(hdl, strerror(errno));
1205                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1206                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1207                             rdsk);
1208                         goto error;
1209                 }
1210
1211                 avl_create(&slice_cache, slice_cache_compare,
1212                     sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1213
1214                 if (strcmp(rdsk, "/dev/") == 0) {
1215                         struct gmesh mesh;
1216                         struct gclass *mp;
1217                         struct ggeom *gp;
1218                         struct gprovider *pp;
1219
1220                         errno = geom_gettree(&mesh);
1221                         if (errno != 0) {
1222                                 zfs_error_aux(hdl, strerror(errno));
1223                                 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1224                                     dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1225                                 goto error;
1226                         }
1227
1228                         LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1229                                 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1230                                         LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1231                                                 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1232                                                 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1233                                                 slice->rn_avl = &slice_cache;
1234                                                 slice->rn_dfd = dfd;
1235                                                 slice->rn_hdl = hdl;
1236                                                 slice->rn_nozpool = B_FALSE;
1237                                                 avl_add(&slice_cache, slice);
1238                                         }
1239                                 }
1240                         }
1241
1242                         geom_deletetree(&mesh);
1243                         goto skipdir;
1244                 }
1245
1246                 /*
1247                  * This is not MT-safe, but we have no MT consumers of libzfs
1248                  */
1249                 while ((dp = readdir64(dirp)) != NULL) {
1250                         const char *name = dp->d_name;
1251                         if (name[0] == '.' &&
1252                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1253                                 continue;
1254
1255                         slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1256                         slice->rn_name = zfs_strdup(hdl, name);
1257                         slice->rn_avl = &slice_cache;
1258                         slice->rn_dfd = dfd;
1259                         slice->rn_hdl = hdl;
1260                         slice->rn_nozpool = B_FALSE;
1261                         avl_add(&slice_cache, slice);
1262                 }
1263 skipdir:
1264                 /*
1265                  * create a thread pool to do all of this in parallel;
1266                  * rn_nozpool is not protected, so this is racy in that
1267                  * multiple tasks could decide that the same slice can
1268                  * not hold a zpool, which is benign.  Also choose
1269                  * double the number of processors; we hold a lot of
1270                  * locks in the kernel, so going beyond this doesn't
1271                  * buy us much.
1272                  */
1273                 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1274                     0, NULL);
1275                 for (slice = avl_first(&slice_cache); slice;
1276                     (slice = avl_walk(&slice_cache, slice,
1277                     AVL_AFTER)))
1278                         (void) tpool_dispatch(t, zpool_open_func, slice);
1279                 tpool_wait(t);
1280                 tpool_destroy(t);
1281
1282                 cookie = NULL;
1283                 while ((slice = avl_destroy_nodes(&slice_cache,
1284                     &cookie)) != NULL) {
1285                         if (slice->rn_config != NULL && !config_failed) {
1286                                 nvlist_t *config = slice->rn_config;
1287                                 boolean_t matched = B_TRUE;
1288
1289                                 if (iarg->poolname != NULL) {
1290                                         char *pname;
1291
1292                                         matched = nvlist_lookup_string(config,
1293                                             ZPOOL_CONFIG_POOL_NAME,
1294                                             &pname) == 0 &&
1295                                             strcmp(iarg->poolname, pname) == 0;
1296                                 } else if (iarg->guid != 0) {
1297                                         uint64_t this_guid;
1298
1299                                         matched = nvlist_lookup_uint64(config,
1300                                             ZPOOL_CONFIG_POOL_GUID,
1301                                             &this_guid) == 0 &&
1302                                             iarg->guid == this_guid;
1303                                 }
1304                                 if (!matched) {
1305                                         nvlist_free(config);
1306                                 } else {
1307                                         /*
1308                                          * use the non-raw path for the config
1309                                          */
1310                                         (void) strlcpy(end, slice->rn_name,
1311                                             pathleft);
1312                                         if (add_config(hdl, &pools, path,
1313                                             config) != 0)
1314                                                 config_failed = B_TRUE;
1315                                 }
1316                         }
1317                         free(slice->rn_name);
1318                         free(slice);
1319                 }
1320                 avl_destroy(&slice_cache);
1321
1322                 (void) closedir(dirp);
1323
1324                 if (config_failed)
1325                         goto error;
1326         }
1327
1328         ret = get_configs(hdl, &pools, iarg->can_be_active);
1329
1330 error:
1331         for (pe = pools.pools; pe != NULL; pe = penext) {
1332                 penext = pe->pe_next;
1333                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1334                         venext = ve->ve_next;
1335                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1336                                 cenext = ce->ce_next;
1337                                 if (ce->ce_config)
1338                                         nvlist_free(ce->ce_config);
1339                                 free(ce);
1340                         }
1341                         free(ve);
1342                 }
1343                 free(pe);
1344         }
1345
1346         for (ne = pools.names; ne != NULL; ne = nenext) {
1347                 nenext = ne->ne_next;
1348                 free(ne->ne_name);
1349                 free(ne);
1350         }
1351
1352         return (ret);
1353 }
1354
1355 nvlist_t *
1356 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1357 {
1358         importargs_t iarg = { 0 };
1359
1360         iarg.paths = argc;
1361         iarg.path = argv;
1362
1363         return (zpool_find_import_impl(hdl, &iarg));
1364 }
1365
1366 /*
1367  * Given a cache file, return the contents as a list of importable pools.
1368  * poolname or guid (but not both) are provided by the caller when trying
1369  * to import a specific pool.
1370  */
1371 nvlist_t *
1372 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1373     char *poolname, uint64_t guid)
1374 {
1375         char *buf;
1376         int fd;
1377         struct stat64 statbuf;
1378         nvlist_t *raw, *src, *dst;
1379         nvlist_t *pools;
1380         nvpair_t *elem;
1381         char *name;
1382         uint64_t this_guid;
1383         boolean_t active;
1384
1385         verify(poolname == NULL || guid == 0);
1386
1387         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1388                 zfs_error_aux(hdl, "%s", strerror(errno));
1389                 (void) zfs_error(hdl, EZFS_BADCACHE,
1390                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1391                 return (NULL);
1392         }
1393
1394         if (fstat64(fd, &statbuf) != 0) {
1395                 zfs_error_aux(hdl, "%s", strerror(errno));
1396                 (void) close(fd);
1397                 (void) zfs_error(hdl, EZFS_BADCACHE,
1398                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1399                 return (NULL);
1400         }
1401
1402         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1403                 (void) close(fd);
1404                 return (NULL);
1405         }
1406
1407         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1408                 (void) close(fd);
1409                 free(buf);
1410                 (void) zfs_error(hdl, EZFS_BADCACHE,
1411                     dgettext(TEXT_DOMAIN,
1412                     "failed to read cache file contents"));
1413                 return (NULL);
1414         }
1415
1416         (void) close(fd);
1417
1418         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1419                 free(buf);
1420                 (void) zfs_error(hdl, EZFS_BADCACHE,
1421                     dgettext(TEXT_DOMAIN,
1422                     "invalid or corrupt cache file contents"));
1423                 return (NULL);
1424         }
1425
1426         free(buf);
1427
1428         /*
1429          * Go through and get the current state of the pools and refresh their
1430          * state.
1431          */
1432         if (nvlist_alloc(&pools, 0, 0) != 0) {
1433                 (void) no_memory(hdl);
1434                 nvlist_free(raw);
1435                 return (NULL);
1436         }
1437
1438         elem = NULL;
1439         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1440                 src = fnvpair_value_nvlist(elem);
1441
1442                 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1443                 if (poolname != NULL && strcmp(poolname, name) != 0)
1444                         continue;
1445
1446                 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1447                 if (guid != 0 && guid != this_guid)
1448                         continue;
1449
1450                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1451                         nvlist_free(raw);
1452                         nvlist_free(pools);
1453                         return (NULL);
1454                 }
1455
1456                 if (active)
1457                         continue;
1458
1459                 if ((dst = refresh_config(hdl, src)) == NULL) {
1460                         nvlist_free(raw);
1461                         nvlist_free(pools);
1462                         return (NULL);
1463                 }
1464
1465                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1466                         (void) no_memory(hdl);
1467                         nvlist_free(dst);
1468                         nvlist_free(raw);
1469                         nvlist_free(pools);
1470                         return (NULL);
1471                 }
1472                 nvlist_free(dst);
1473         }
1474
1475         nvlist_free(raw);
1476         return (pools);
1477 }
1478
1479 static int
1480 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1481 {
1482         importargs_t *import = data;
1483         int found = 0;
1484
1485         if (import->poolname != NULL) {
1486                 char *pool_name;
1487
1488                 verify(nvlist_lookup_string(zhp->zpool_config,
1489                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1490                 if (strcmp(pool_name, import->poolname) == 0)
1491                         found = 1;
1492         } else {
1493                 uint64_t pool_guid;
1494
1495                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1496                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1497                 if (pool_guid == import->guid)
1498                         found = 1;
1499         }
1500
1501         zpool_close(zhp);
1502         return (found);
1503 }
1504
1505 nvlist_t *
1506 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1507 {
1508         verify(import->poolname == NULL || import->guid == 0);
1509
1510         if (import->unique)
1511                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1512
1513         if (import->cachefile != NULL)
1514                 return (zpool_find_import_cached(hdl, import->cachefile,
1515                     import->poolname, import->guid));
1516
1517         return (zpool_find_import_impl(hdl, import));
1518 }
1519
1520 boolean_t
1521 find_guid(nvlist_t *nv, uint64_t guid)
1522 {
1523         uint64_t tmp;
1524         nvlist_t **child;
1525         uint_t c, children;
1526
1527         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1528         if (tmp == guid)
1529                 return (B_TRUE);
1530
1531         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1532             &child, &children) == 0) {
1533                 for (c = 0; c < children; c++)
1534                         if (find_guid(child[c], guid))
1535                                 return (B_TRUE);
1536         }
1537
1538         return (B_FALSE);
1539 }
1540
1541 typedef struct aux_cbdata {
1542         const char      *cb_type;
1543         uint64_t        cb_guid;
1544         zpool_handle_t  *cb_zhp;
1545 } aux_cbdata_t;
1546
1547 static int
1548 find_aux(zpool_handle_t *zhp, void *data)
1549 {
1550         aux_cbdata_t *cbp = data;
1551         nvlist_t **list;
1552         uint_t i, count;
1553         uint64_t guid;
1554         nvlist_t *nvroot;
1555
1556         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1557             &nvroot) == 0);
1558
1559         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1560             &list, &count) == 0) {
1561                 for (i = 0; i < count; i++) {
1562                         verify(nvlist_lookup_uint64(list[i],
1563                             ZPOOL_CONFIG_GUID, &guid) == 0);
1564                         if (guid == cbp->cb_guid) {
1565                                 cbp->cb_zhp = zhp;
1566                                 return (1);
1567                         }
1568                 }
1569         }
1570
1571         zpool_close(zhp);
1572         return (0);
1573 }
1574
1575 /*
1576  * Determines if the pool is in use.  If so, it returns true and the state of
1577  * the pool as well as the name of the pool.  Both strings are allocated and
1578  * must be freed by the caller.
1579  */
1580 int
1581 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1582     boolean_t *inuse)
1583 {
1584         nvlist_t *config;
1585         char *name;
1586         boolean_t ret;
1587         uint64_t guid, vdev_guid;
1588         zpool_handle_t *zhp;
1589         nvlist_t *pool_config;
1590         uint64_t stateval, isspare;
1591         aux_cbdata_t cb = { 0 };
1592         boolean_t isactive;
1593
1594         *inuse = B_FALSE;
1595
1596         if (zpool_read_label(fd, &config) != 0) {
1597                 (void) no_memory(hdl);
1598                 return (-1);
1599         }
1600
1601         if (config == NULL)
1602                 return (0);
1603
1604         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1605             &stateval) == 0);
1606         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1607             &vdev_guid) == 0);
1608
1609         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1610                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1611                     &name) == 0);
1612                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1613                     &guid) == 0);
1614         }
1615
1616         switch (stateval) {
1617         case POOL_STATE_EXPORTED:
1618                 /*
1619                  * A pool with an exported state may in fact be imported
1620                  * read-only, so check the in-core state to see if it's
1621                  * active and imported read-only.  If it is, set
1622                  * its state to active.
1623                  */
1624                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1625                     (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1626                         if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1627                                 stateval = POOL_STATE_ACTIVE;
1628
1629                         /*
1630                          * All we needed the zpool handle for is the
1631                          * readonly prop check.
1632                          */
1633                         zpool_close(zhp);
1634                 }
1635
1636                 ret = B_TRUE;
1637                 break;
1638
1639         case POOL_STATE_ACTIVE:
1640                 /*
1641                  * For an active pool, we have to determine if it's really part
1642                  * of a currently active pool (in which case the pool will exist
1643                  * and the guid will be the same), or whether it's part of an
1644                  * active pool that was disconnected without being explicitly
1645                  * exported.
1646                  */
1647                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1648                         nvlist_free(config);
1649                         return (-1);
1650                 }
1651
1652                 if (isactive) {
1653                         /*
1654                          * Because the device may have been removed while
1655                          * offlined, we only report it as active if the vdev is
1656                          * still present in the config.  Otherwise, pretend like
1657                          * it's not in use.
1658                          */
1659                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1660                             (pool_config = zpool_get_config(zhp, NULL))
1661                             != NULL) {
1662                                 nvlist_t *nvroot;
1663
1664                                 verify(nvlist_lookup_nvlist(pool_config,
1665                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1666                                 ret = find_guid(nvroot, vdev_guid);
1667                         } else {
1668                                 ret = B_FALSE;
1669                         }
1670
1671                         /*
1672                          * If this is an active spare within another pool, we
1673                          * treat it like an unused hot spare.  This allows the
1674                          * user to create a pool with a hot spare that currently
1675                          * in use within another pool.  Since we return B_TRUE,
1676                          * libdiskmgt will continue to prevent generic consumers
1677                          * from using the device.
1678                          */
1679                         if (ret && nvlist_lookup_uint64(config,
1680                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1681                                 stateval = POOL_STATE_SPARE;
1682
1683                         if (zhp != NULL)
1684                                 zpool_close(zhp);
1685                 } else {
1686                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1687                         ret = B_TRUE;
1688                 }
1689                 break;
1690
1691         case POOL_STATE_SPARE:
1692                 /*
1693                  * For a hot spare, it can be either definitively in use, or
1694                  * potentially active.  To determine if it's in use, we iterate
1695                  * over all pools in the system and search for one with a spare
1696                  * with a matching guid.
1697                  *
1698                  * Due to the shared nature of spares, we don't actually report
1699                  * the potentially active case as in use.  This means the user
1700                  * can freely create pools on the hot spares of exported pools,
1701                  * but to do otherwise makes the resulting code complicated, and
1702                  * we end up having to deal with this case anyway.
1703                  */
1704                 cb.cb_zhp = NULL;
1705                 cb.cb_guid = vdev_guid;
1706                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1707                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1708                         name = (char *)zpool_get_name(cb.cb_zhp);
1709                         ret = B_TRUE;
1710                 } else {
1711                         ret = B_FALSE;
1712                 }
1713                 break;
1714
1715         case POOL_STATE_L2CACHE:
1716
1717                 /*
1718                  * Check if any pool is currently using this l2cache device.
1719                  */
1720                 cb.cb_zhp = NULL;
1721                 cb.cb_guid = vdev_guid;
1722                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1723                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1724                         name = (char *)zpool_get_name(cb.cb_zhp);
1725                         ret = B_TRUE;
1726                 } else {
1727                         ret = B_FALSE;
1728                 }
1729                 break;
1730
1731         default:
1732                 ret = B_FALSE;
1733         }
1734
1735
1736         if (ret) {
1737                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1738                         if (cb.cb_zhp)
1739                                 zpool_close(cb.cb_zhp);
1740                         nvlist_free(config);
1741                         return (-1);
1742                 }
1743                 *state = (pool_state_t)stateval;
1744         }
1745
1746         if (cb.cb_zhp)
1747                 zpool_close(cb.cb_zhp);
1748
1749         nvlist_free(config);
1750         *inuse = ret;
1751         return (0);
1752 }