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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
  25  * Copyright 2015 RackTop Systems.
  26  * Copyright 2016 Nexenta Systems, Inc.
  27  */
  28
  29 /*
  30  * Pool import support functions.
  31  *
  32  * To import a pool, we rely on reading the configuration information from the
  33  * ZFS label of each device.  If we successfully read the label, then we
  34  * organize the configuration information in the following hierarchy:
  35  *
  36  *      pool guid -> toplevel vdev guid -> label txg
  37  *
  38  * Duplicate entries matching this same tuple will be discarded.  Once we have
  39  * examined every device, we pick the best label txg config for each toplevel
  40  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
  41  * update any paths that have changed.  Finally, we attempt to import the pool
  42  * using our derived config, and record the results.
  43  */
  44
  45 #include <aio.h>
  46 #include <ctype.h>
  47 #include <devid.h>
  48 #include <dirent.h>
  49 #include <errno.h>
  50 #include <libintl.h>
  51 #include <stddef.h>
  52 #include <stdlib.h>
  53 #include <string.h>
  54 #include <sys/stat.h>
  55 #include <unistd.h>
  56 #include <fcntl.h>
  57 #include <thread_pool.h>
  58 #include <libgeom.h>
  59
  60 #include <sys/vdev_impl.h>
  61
  62 #include "libzfs.h"
  63 #include "libzfs_impl.h"
  64
  65 /*
  66  * Intermediate structures used to gather configuration information.
  67  */
  68 typedef struct config_entry {
  69         uint64_t                ce_txg;
  70         nvlist_t                *ce_config;
  71         struct config_entry     *ce_next;
  72 } config_entry_t;
  73
  74 typedef struct vdev_entry {
  75         uint64_t                ve_guid;
  76         config_entry_t          *ve_configs;
  77         struct vdev_entry       *ve_next;
  78 } vdev_entry_t;
  79
  80 typedef struct pool_entry {
  81         uint64_t                pe_guid;
  82         vdev_entry_t            *pe_vdevs;
  83         struct pool_entry       *pe_next;
  84 } pool_entry_t;
  85
  86 typedef struct name_entry {
  87         char                    *ne_name;
  88         uint64_t                ne_guid;
  89         struct name_entry       *ne_next;
  90 } name_entry_t;
  91
  92 typedef struct pool_list {
  93         pool_entry_t            *pools;
  94         name_entry_t            *names;
  95 } pool_list_t;
  96
  97 static char *
  98 get_devid(const char *path)
  99 {
 100 #ifdef have_devid
 101         int fd;
 102         ddi_devid_t devid;
 103         char *minor, *ret;
 104
 105         if ((fd = open(path, O_RDONLY)) < 0)
 106                 return (NULL);
 107
 108         minor = NULL;
 109         ret = NULL;
 110         if (devid_get(fd, &devid) == 0) {
 111                 if (devid_get_minor_name(fd, &minor) == 0)
 112                         ret = devid_str_encode(devid, minor);
 113                 if (minor != NULL)
 114                         devid_str_free(minor);
 115                 devid_free(devid);
 116         }
 117         (void) close(fd);
 118
 119         return (ret);
 120 #else
 121         return (NULL);
 122 #endif
 123 }
 124
 125
 126 /*
 127  * Go through and fix up any path and/or devid information for the given vdev
 128  * configuration.
 129  */
 130 static int
 131 fix_paths(nvlist_t *nv, name_entry_t *names)
 132 {
 133         nvlist_t **child;
 134         uint_t c, children;
 135         uint64_t guid;
 136         name_entry_t *ne, *best;
 137         char *path, *devid;
 138         int matched;
 139
 140         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 141             &child, &children) == 0) {
 142                 for (c = 0; c < children; c++)
 143                         if (fix_paths(child[c], names) != 0)
 144                                 return (-1);
 145                 return (0);
 146         }
 147
 148         /*
 149          * This is a leaf (file or disk) vdev.  In either case, go through
 150          * the name list and see if we find a matching guid.  If so, replace
 151          * the path and see if we can calculate a new devid.
 152          *
 153          * There may be multiple names associated with a particular guid, in
 154          * which case we have overlapping slices or multiple paths to the same
 155          * disk.  If this is the case, then we want to pick the path that is
 156          * the most similar to the original, where "most similar" is the number
 157          * of matching characters starting from the end of the path.  This will
 158          * preserve slice numbers even if the disks have been reorganized, and
 159          * will also catch preferred disk names if multiple paths exist.
 160          */
 161         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
 162         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
 163                 path = NULL;
 164
 165         matched = 0;
 166         best = NULL;
 167         for (ne = names; ne != NULL; ne = ne->ne_next) {
 168                 if (ne->ne_guid == guid) {
 169                         const char *src, *dst;
 170                         int count;
 171
 172                         if (path == NULL) {
 173                                 best = ne;
 174                                 break;
 175                         }
 176
 177                         src = ne->ne_name + strlen(ne->ne_name) - 1;
 178                         dst = path + strlen(path) - 1;
 179                         for (count = 0; src >= ne->ne_name && dst >= path;
 180                             src--, dst--, count++)
 181                                 if (*src != *dst)
 182                                         break;
 183
 184                         /*
 185                          * At this point, 'count' is the number of characters
 186                          * matched from the end.
 187                          */
 188                         if (count > matched || best == NULL) {
 189                                 best = ne;
 190                                 matched = count;
 191                         }
 192                 }
 193         }
 194
 195         if (best == NULL)
 196                 return (0);
 197
 198         if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 199                 return (-1);
 200
 201         if ((devid = get_devid(best->ne_name)) == NULL) {
 202                 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
 203         } else {
 204                 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
 205                         devid_str_free(devid);
 206                         return (-1);
 207                 }
 208                 devid_str_free(devid);
 209         }
 210
 211         return (0);
 212 }
 213
 214 /*
 215  * Add the given configuration to the list of known devices.
 216  */
 217 static int
 218 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
 219     nvlist_t *config)
 220 {
 221         uint64_t pool_guid, vdev_guid, top_guid, txg, state;
 222         pool_entry_t *pe;
 223         vdev_entry_t *ve;
 224         config_entry_t *ce;
 225         name_entry_t *ne;
 226
 227         /*
 228          * If this is a hot spare not currently in use or level 2 cache
 229          * device, add it to the list of names to translate, but don't do
 230          * anything else.
 231          */
 232         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 233             &state) == 0 &&
 234             (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
 235             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
 236                 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 237                         return (-1);
 238
 239                 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 240                         free(ne);
 241                         return (-1);
 242                 }
 243
 244                 ne->ne_guid = vdev_guid;
 245                 ne->ne_next = pl->names;
 246                 pl->names = ne;
 247
 248                 return (0);
 249         }
 250
 251         /*
 252          * If we have a valid config but cannot read any of these fields, then
 253          * it means we have a half-initialized label.  In vdev_label_init()
 254          * we write a label with txg == 0 so that we can identify the device
 255          * in case the user refers to the same disk later on.  If we fail to
 256          * create the pool, we'll be left with a label in this state
 257          * which should not be considered part of a valid pool.
 258          */
 259         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 260             &pool_guid) != 0 ||
 261             nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
 262             &vdev_guid) != 0 ||
 263             nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 264             &top_guid) != 0 ||
 265             nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 266             &txg) != 0 || txg == 0) {
 267                 return (0);
 268         }
 269
 270         /*
 271          * First, see if we know about this pool.  If not, then add it to the
 272          * list of known pools.
 273          */
 274         for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
 275                 if (pe->pe_guid == pool_guid)
 276                         break;
 277         }
 278
 279         if (pe == NULL) {
 280                 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
 281                         return (-1);
 282                 }
 283                 pe->pe_guid = pool_guid;
 284                 pe->pe_next = pl->pools;
 285                 pl->pools = pe;
 286         }
 287
 288         /*
 289          * Second, see if we know about this toplevel vdev.  Add it if its
 290          * missing.
 291          */
 292         for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
 293                 if (ve->ve_guid == top_guid)
 294                         break;
 295         }
 296
 297         if (ve == NULL) {
 298                 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
 299                         return (-1);
 300                 }
 301                 ve->ve_guid = top_guid;
 302                 ve->ve_next = pe->pe_vdevs;
 303                 pe->pe_vdevs = ve;
 304         }
 305
 306         /*
 307          * Third, see if we have a config with a matching transaction group.  If
 308          * so, then we do nothing.  Otherwise, add it to the list of known
 309          * configs.
 310          */
 311         for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
 312                 if (ce->ce_txg == txg)
 313                         break;
 314         }
 315
 316         if (ce == NULL) {
 317                 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
 318                         return (-1);
 319                 }
 320                 ce->ce_txg = txg;
 321                 ce->ce_config = fnvlist_dup(config);
 322                 ce->ce_next = ve->ve_configs;
 323                 ve->ve_configs = ce;
 324         }
 325
 326         /*
 327          * At this point we've successfully added our config to the list of
 328          * known configs.  The last thing to do is add the vdev guid -> path
 329          * mappings so that we can fix up the configuration as necessary before
 330          * doing the import.
 331          */
 332         if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
 333                 return (-1);
 334
 335         if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
 336                 free(ne);
 337                 return (-1);
 338         }
 339
 340         ne->ne_guid = vdev_guid;
 341         ne->ne_next = pl->names;
 342         pl->names = ne;
 343
 344         return (0);
 345 }
 346
 347 /*
 348  * Returns true if the named pool matches the given GUID.
 349  */
 350 static int
 351 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
 352     boolean_t *isactive)
 353 {
 354         zpool_handle_t *zhp;
 355         uint64_t theguid;
 356
 357         if (zpool_open_silent(hdl, name, &zhp) != 0)
 358                 return (-1);
 359
 360         if (zhp == NULL) {
 361                 *isactive = B_FALSE;
 362                 return (0);
 363         }
 364
 365         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
 366             &theguid) == 0);
 367
 368         zpool_close(zhp);
 369
 370         *isactive = (theguid == guid);
 371         return (0);
 372 }
 373
 374 static nvlist_t *
 375 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 376 {
 377         nvlist_t *nvl;
 378         zfs_cmd_t zc = { 0 };
 379         int err, dstbuf_size;
 380
 381         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
 382                 return (NULL);
 383
 384         dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
 385
 386         if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
 387                 zcmd_free_nvlists(&zc);
 388                 return (NULL);
 389         }
 390
 391         while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
 392             &zc)) != 0 && errno == ENOMEM) {
 393                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
 394                         zcmd_free_nvlists(&zc);
 395                         return (NULL);
 396                 }
 397         }
 398
 399         if (err) {
 400                 zcmd_free_nvlists(&zc);
 401                 return (NULL);
 402         }
 403
 404         if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
 405                 zcmd_free_nvlists(&zc);
 406                 return (NULL);
 407         }
 408
 409         zcmd_free_nvlists(&zc);
 410         return (nvl);
 411 }
 412
 413 /*
 414  * Determine if the vdev id is a hole in the namespace.
 415  */
 416 boolean_t
 417 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
 418 {
 419         for (int c = 0; c < holes; c++) {
 420
 421                 /* Top-level is a hole */
 422                 if (hole_array[c] == id)
 423                         return (B_TRUE);
 424         }
 425         return (B_FALSE);
 426 }
 427
 428 /*
 429  * Convert our list of pools into the definitive set of configurations.  We
 430  * start by picking the best config for each toplevel vdev.  Once that's done,
 431  * we assemble the toplevel vdevs into a full config for the pool.  We make a
 432  * pass to fix up any incorrect paths, and then add it to the main list to
 433  * return to the user.
 434  */
 435 static nvlist_t *
 436 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok,
 437     nvlist_t *policy)
 438 {
 439         pool_entry_t *pe;
 440         vdev_entry_t *ve;
 441         config_entry_t *ce;
 442         nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *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 = NULL;
 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 (policy != NULL) {
 772                         if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY,
 773                             policy) != 0)
 774                                 goto nomem;
 775                 }
 776
 777                 if ((nvl = refresh_config(hdl, config)) == NULL) {
 778                         nvlist_free(config);
 779                         config = NULL;
 780                         continue;
 781                 }
 782
 783                 nvlist_free(config);
 784                 config = nvl;
 785
 786                 /*
 787                  * Go through and update the paths for spares, now that we have
 788                  * them.
 789                  */
 790                 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 791                     &nvroot) == 0);
 792                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
 793                     &spares, &nspares) == 0) {
 794                         for (i = 0; i < nspares; i++) {
 795                                 if (fix_paths(spares[i], pl->names) != 0)
 796                                         goto nomem;
 797                         }
 798                 }
 799
 800                 /*
 801                  * Update the paths for l2cache devices.
 802                  */
 803                 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 804                     &l2cache, &nl2cache) == 0) {
 805                         for (i = 0; i < nl2cache; i++) {
 806                                 if (fix_paths(l2cache[i], pl->names) != 0)
 807                                         goto nomem;
 808                         }
 809                 }
 810
 811                 /*
 812                  * Restore the original information read from the actual label.
 813                  */
 814                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
 815                     DATA_TYPE_UINT64);
 816                 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
 817                     DATA_TYPE_STRING);
 818                 if (hostid != 0) {
 819                         verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 820                             hostid) == 0);
 821                         verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 822                             hostname) == 0);
 823                 }
 824
 825 add_pool:
 826                 /*
 827                  * Add this pool to the list of configs.
 828                  */
 829                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
 830                     &name) == 0);
 831                 if (nvlist_add_nvlist(ret, name, config) != 0)
 832                         goto nomem;
 833
 834                 found_one = B_TRUE;
 835                 nvlist_free(config);
 836                 config = NULL;
 837         }
 838
 839         if (!found_one) {
 840                 nvlist_free(ret);
 841                 ret = NULL;
 842         }
 843
 844         return (ret);
 845
 846 nomem:
 847         (void) no_memory(hdl);
 848 error:
 849         nvlist_free(config);
 850         nvlist_free(ret);
 851         for (c = 0; c < children; c++)
 852                 nvlist_free(child[c]);
 853         free(child);
 854
 855         return (NULL);
 856 }
 857
 858 /*
 859  * Return the offset of the given label.
 860  */
 861 static uint64_t
 862 label_offset(uint64_t size, int l)
 863 {
 864         ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
 865         return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 866             0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
 867 }
 868
 869 /*
 870  * Given a file descriptor, read the label information and return an nvlist
 871  * describing the configuration, if there is one.
 872  * Return 0 on success, or -1 on failure
 873  */
 874 int
 875 zpool_read_label(int fd, nvlist_t **config)
 876 {
 877         struct stat64 statbuf;
 878         int l;
 879         vdev_label_t *label;
 880         uint64_t state, txg, size;
 881
 882         *config = NULL;
 883
 884         if (fstat64(fd, &statbuf) == -1)
 885                 return (-1);
 886         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 887
 888         if ((label = malloc(sizeof (vdev_label_t))) == NULL)
 889                 return (-1);
 890
 891         for (l = 0; l < VDEV_LABELS; l++) {
 892                 if (pread64(fd, label, sizeof (vdev_label_t),
 893                     label_offset(size, l)) != sizeof (vdev_label_t))
 894                         continue;
 895
 896                 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
 897                     sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
 898                         continue;
 899
 900                 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
 901                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 902                         nvlist_free(*config);
 903                         continue;
 904                 }
 905
 906                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 907                     (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
 908                     &txg) != 0 || txg == 0)) {
 909                         nvlist_free(*config);
 910                         continue;
 911                 }
 912
 913                 free(label);
 914                 return (0);
 915         }
 916
 917         free(label);
 918         *config = NULL;
 919         errno = ENOENT;
 920         return (-1);
 921 }
 922
 923 /*
 924  * Given a file descriptor, read the label information and return an nvlist
 925  * describing the configuration, if there is one.
 926  * returns the number of valid labels found
 927  * If a label is found, returns it via config.  The caller is responsible for
 928  * freeing it.
 929  */
 930 int
 931 zpool_read_all_labels(int fd, nvlist_t **config)
 932 {
 933         struct stat64 statbuf;
 934         struct aiocb aiocbs[VDEV_LABELS];
 935         struct aiocb *aiocbps[VDEV_LABELS];
 936         int l;
 937         vdev_phys_t *labels;
 938         uint64_t state, txg, size;
 939         int nlabels = 0;
 940
 941         *config = NULL;
 942
 943         if (fstat64(fd, &statbuf) == -1)
 944                 return (0);
 945         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
 946
 947         if ((labels = calloc(VDEV_LABELS, sizeof (vdev_phys_t))) == NULL)
 948                 return (0);
 949
 950         memset(aiocbs, 0, sizeof(aiocbs));
 951         for (l = 0; l < VDEV_LABELS; l++) {
 952                 aiocbs[l].aio_fildes = fd;
 953                 aiocbs[l].aio_offset = label_offset(size, l) + VDEV_SKIP_SIZE;
 954                 aiocbs[l].aio_buf = &labels[l];
 955                 aiocbs[l].aio_nbytes = sizeof(vdev_phys_t);
 956                 aiocbs[l].aio_lio_opcode = LIO_READ;
 957                 aiocbps[l] = &aiocbs[l];
 958         }
 959
 960         if (lio_listio(LIO_WAIT, aiocbps, VDEV_LABELS, NULL) != 0) {
 961                 if (errno == EAGAIN || errno == EINTR || errno == EIO) {
 962                         for (l = 0; l < VDEV_LABELS; l++) {
 963                                 errno = 0;
 964                                 int r = aio_error(&aiocbs[l]);
 965                                 if (r != EINVAL)
 966                                         (void)aio_return(&aiocbs[l]);
 967                         }
 968                 }
 969                 free(labels);
 970                 return (0);
 971         }
 972
 973         for (l = 0; l < VDEV_LABELS; l++) {
 974                 nvlist_t *temp = NULL;
 975
 976                 if (aio_return(&aiocbs[l]) != sizeof(vdev_phys_t))
 977                         continue;
 978
 979                 if (nvlist_unpack(labels[l].vp_nvlist,
 980                     sizeof (labels[l].vp_nvlist), &temp, 0) != 0)
 981                         continue;
 982
 983                 if (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_STATE,
 984                     &state) != 0 || state > POOL_STATE_L2CACHE) {
 985                         nvlist_free(temp);
 986                         temp = NULL;
 987                         continue;
 988                 }
 989
 990                 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 991                     (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_TXG,
 992                     &txg) != 0 || txg == 0)) {
 993                         nvlist_free(temp);
 994                         temp = NULL;
 995                         continue;
 996                 }
 997                 if (temp)
 998                         *config = temp;
 999
1000                 nlabels++;
1001         }
1002
1003         free(labels);
1004         return (nlabels);
1005 }
1006
1007 typedef struct rdsk_node {
1008         char *rn_name;
1009         int rn_dfd;
1010         libzfs_handle_t *rn_hdl;
1011         nvlist_t *rn_config;
1012         avl_tree_t *rn_avl;
1013         avl_node_t rn_node;
1014         boolean_t rn_nozpool;
1015 } rdsk_node_t;
1016
1017 static int
1018 slice_cache_compare(const void *arg1, const void *arg2)
1019 {
1020         const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
1021         const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
1022         char *nm1slice, *nm2slice;
1023         int rv;
1024
1025         /*
1026          * slices zero and two are the most likely to provide results,
1027          * so put those first
1028          */
1029         nm1slice = strstr(nm1, "s0");
1030         nm2slice = strstr(nm2, "s0");
1031         if (nm1slice && !nm2slice) {
1032                 return (-1);
1033         }
1034         if (!nm1slice && nm2slice) {
1035                 return (1);
1036         }
1037         nm1slice = strstr(nm1, "s2");
1038         nm2slice = strstr(nm2, "s2");
1039         if (nm1slice && !nm2slice) {
1040                 return (-1);
1041         }
1042         if (!nm1slice && nm2slice) {
1043                 return (1);
1044         }
1045
1046         rv = strcmp(nm1, nm2);
1047         if (rv == 0)
1048                 return (0);
1049         return (rv > 0 ? 1 : -1);
1050 }
1051
1052 #ifdef illumos
1053 static void
1054 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
1055     diskaddr_t size, uint_t blksz)
1056 {
1057         rdsk_node_t tmpnode;
1058         rdsk_node_t *node;
1059         char sname[MAXNAMELEN];
1060
1061         tmpnode.rn_name = &sname[0];
1062         (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
1063             diskname, partno);
1064         /*
1065          * protect against division by zero for disk labels that
1066          * contain a bogus sector size
1067          */
1068         if (blksz == 0)
1069                 blksz = DEV_BSIZE;
1070         /* too small to contain a zpool? */
1071         if ((size < (SPA_MINDEVSIZE / blksz)) &&
1072             (node = avl_find(r, &tmpnode, NULL)))
1073                 node->rn_nozpool = B_TRUE;
1074 }
1075 #endif  /* illumos */
1076
1077 static void
1078 nozpool_all_slices(avl_tree_t *r, const char *sname)
1079 {
1080 #ifdef illumos
1081         char diskname[MAXNAMELEN];
1082         char *ptr;
1083         int i;
1084
1085         (void) strncpy(diskname, sname, MAXNAMELEN);
1086         if (((ptr = strrchr(diskname, 's')) == NULL) &&
1087             ((ptr = strrchr(diskname, 'p')) == NULL))
1088                 return;
1089         ptr[0] = 's';
1090         ptr[1] = '\0';
1091         for (i = 0; i < NDKMAP; i++)
1092                 check_one_slice(r, diskname, i, 0, 1);
1093         ptr[0] = 'p';
1094         for (i = 0; i <= FD_NUMPART; i++)
1095                 check_one_slice(r, diskname, i, 0, 1);
1096 #endif  /* illumos */
1097 }
1098
1099 #ifdef illumos
1100 static void
1101 check_slices(avl_tree_t *r, int fd, const char *sname)
1102 {
1103         struct extvtoc vtoc;
1104         struct dk_gpt *gpt;
1105         char diskname[MAXNAMELEN];
1106         char *ptr;
1107         int i;
1108
1109         (void) strncpy(diskname, sname, MAXNAMELEN);
1110         if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1111                 return;
1112         ptr[1] = '\0';
1113
1114         if (read_extvtoc(fd, &vtoc) >= 0) {
1115                 for (i = 0; i < NDKMAP; i++)
1116                         check_one_slice(r, diskname, i,
1117                             vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1118         } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1119                 /*
1120                  * on x86 we'll still have leftover links that point
1121                  * to slices s[9-15], so use NDKMAP instead
1122                  */
1123                 for (i = 0; i < NDKMAP; i++)
1124                         check_one_slice(r, diskname, i,
1125                             gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1126                 /* nodes p[1-4] are never used with EFI labels */
1127                 ptr[0] = 'p';
1128                 for (i = 1; i <= FD_NUMPART; i++)
1129                         check_one_slice(r, diskname, i, 0, 1);
1130                 efi_free(gpt);
1131         }
1132 }
1133 #endif  /* illumos */
1134
1135 static void
1136 zpool_open_func(void *arg)
1137 {
1138         rdsk_node_t *rn = arg;
1139         struct stat64 statbuf;
1140         nvlist_t *config;
1141         int fd;
1142
1143         if (rn->rn_nozpool)
1144                 return;
1145         if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1146                 /* symlink to a device that's no longer there */
1147                 if (errno == ENOENT)
1148                         nozpool_all_slices(rn->rn_avl, rn->rn_name);
1149                 return;
1150         }
1151         /*
1152          * Ignore failed stats.  We only want regular
1153          * files, character devs and block devs.
1154          */
1155         if (fstat64(fd, &statbuf) != 0 ||
1156             (!S_ISREG(statbuf.st_mode) &&
1157             !S_ISCHR(statbuf.st_mode) &&
1158             !S_ISBLK(statbuf.st_mode))) {
1159                 (void) close(fd);
1160                 return;
1161         }
1162         /* this file is too small to hold a zpool */
1163 #ifdef illumos
1164         if (S_ISREG(statbuf.st_mode) &&
1165             statbuf.st_size < SPA_MINDEVSIZE) {
1166                 (void) close(fd);
1167                 return;
1168         } else if (!S_ISREG(statbuf.st_mode)) {
1169                 /*
1170                  * Try to read the disk label first so we don't have to
1171                  * open a bunch of minor nodes that can't have a zpool.
1172                  */
1173                 check_slices(rn->rn_avl, fd, rn->rn_name);
1174         }
1175 #else   /* !illumos */
1176         if (statbuf.st_size < SPA_MINDEVSIZE) {
1177                 (void) close(fd);
1178                 return;
1179         }
1180 #endif  /* illumos */
1181
1182         if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1183                 (void) close(fd);
1184                 (void) no_memory(rn->rn_hdl);
1185                 return;
1186         }
1187         (void) close(fd);
1188
1189         rn->rn_config = config;
1190 }
1191
1192 /*
1193  * Given a file descriptor, clear (zero) the label information.
1194  */
1195 int
1196 zpool_clear_label(int fd)
1197 {
1198         struct stat64 statbuf;
1199         int l;
1200         vdev_label_t *label;
1201         uint64_t size;
1202
1203         if (fstat64(fd, &statbuf) == -1)
1204                 return (0);
1205         size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1206
1207         if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1208                 return (-1);
1209
1210         for (l = 0; l < VDEV_LABELS; l++) {
1211                 if (pwrite64(fd, label, sizeof (vdev_label_t),
1212                     label_offset(size, l)) != sizeof (vdev_label_t)) {
1213                         free(label);
1214                         return (-1);
1215                 }
1216         }
1217
1218         free(label);
1219         return (0);
1220 }
1221
1222 /*
1223  * Given a list of directories to search, find all pools stored on disk.  This
1224  * includes partial pools which are not available to import.  If no args are
1225  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1226  * poolname or guid (but not both) are provided by the caller when trying
1227  * to import a specific pool.
1228  */
1229 static nvlist_t *
1230 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1231 {
1232         int i, dirs = iarg->paths;
1233         struct dirent64 *dp;
1234         char path[MAXPATHLEN];
1235         char *end, **dir = iarg->path;
1236         size_t pathleft;
1237         nvlist_t *ret = NULL;
1238         static char *default_dir = "/dev";
1239         pool_list_t pools = { 0 };
1240         pool_entry_t *pe, *penext;
1241         vdev_entry_t *ve, *venext;
1242         config_entry_t *ce, *cenext;
1243         name_entry_t *ne, *nenext;
1244         avl_tree_t slice_cache;
1245         rdsk_node_t *slice;
1246         void *cookie;
1247         boolean_t skip_zvols = B_FALSE;
1248         int value;
1249         size_t size = sizeof(value);
1250
1251         if (dirs == 0) {
1252                 dirs = 1;
1253                 dir = &default_dir;
1254         }
1255
1256         if (sysctlbyname("vfs.zfs.vol.recursive", &value, &size, NULL, 0) == 0
1257             && value == 0) {
1258                 skip_zvols = B_TRUE;
1259         }
1260
1261         /*
1262          * Go through and read the label configuration information from every
1263          * possible device, organizing the information according to pool GUID
1264          * and toplevel GUID.
1265          */
1266         for (i = 0; i < dirs; i++) {
1267                 tpool_t *t;
1268                 char rdsk[MAXPATHLEN];
1269                 int dfd;
1270                 boolean_t config_failed = B_FALSE;
1271                 DIR *dirp;
1272
1273                 /* use realpath to normalize the path */
1274                 if (realpath(dir[i], path) == 0) {
1275                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1276                             dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1277                         goto error;
1278                 }
1279                 end = &path[strlen(path)];
1280                 *end++ = '/';
1281                 *end = 0;
1282                 pathleft = &path[sizeof (path)] - end;
1283
1284 #ifdef illumos
1285                 /*
1286                  * Using raw devices instead of block devices when we're
1287                  * reading the labels skips a bunch of slow operations during
1288                  * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1289                  */
1290                 if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1291                         (void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1292                 else
1293 #endif
1294                         (void) strlcpy(rdsk, path, sizeof (rdsk));
1295
1296                 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1297                     (dirp = fdopendir(dfd)) == NULL) {
1298                         if (dfd >= 0)
1299                                 (void) close(dfd);
1300                         zfs_error_aux(hdl, strerror(errno));
1301                         (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1302                             dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1303                             rdsk);
1304                         goto error;
1305                 }
1306
1307                 avl_create(&slice_cache, slice_cache_compare,
1308                     sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1309
1310                 if (strcmp(rdsk, "/dev/") == 0) {
1311                         struct gmesh mesh;
1312                         struct gclass *mp;
1313                         struct ggeom *gp;
1314                         struct gprovider *pp;
1315
1316                         errno = geom_gettree(&mesh);
1317                         if (errno != 0) {
1318                                 zfs_error_aux(hdl, strerror(errno));
1319                                 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1320                                     dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1321                                 goto error;
1322                         }
1323
1324                         LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1325                                 if (skip_zvols &&
1326                                     strcmp(mp->lg_name, "ZFS::ZVOL") == 0) {
1327                                         continue;
1328                                 }
1329                                 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1330                                         LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1331                                                 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1332                                                 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1333                                                 slice->rn_avl = &slice_cache;
1334                                                 slice->rn_dfd = dfd;
1335                                                 slice->rn_hdl = hdl;
1336                                                 slice->rn_nozpool = B_FALSE;
1337                                                 avl_add(&slice_cache, slice);
1338                                         }
1339                                 }
1340                         }
1341
1342                         geom_deletetree(&mesh);
1343                         goto skipdir;
1344                 }
1345
1346                 /*
1347                  * This is not MT-safe, but we have no MT consumers of libzfs
1348                  */
1349                 while ((dp = readdir64(dirp)) != NULL) {
1350                         const char *name = dp->d_name;
1351                         if (name[0] == '.' &&
1352                             (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1353                                 continue;
1354
1355                         slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1356                         slice->rn_name = zfs_strdup(hdl, name);
1357                         slice->rn_avl = &slice_cache;
1358                         slice->rn_dfd = dfd;
1359                         slice->rn_hdl = hdl;
1360                         slice->rn_nozpool = B_FALSE;
1361                         avl_add(&slice_cache, slice);
1362                 }
1363 skipdir:
1364                 /*
1365                  * create a thread pool to do all of this in parallel;
1366                  * rn_nozpool is not protected, so this is racy in that
1367                  * multiple tasks could decide that the same slice can
1368                  * not hold a zpool, which is benign.  Also choose
1369                  * double the number of processors; we hold a lot of
1370                  * locks in the kernel, so going beyond this doesn't
1371                  * buy us much.
1372                  */
1373                 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1374                     0, NULL);
1375                 for (slice = avl_first(&slice_cache); slice;
1376                     (slice = avl_walk(&slice_cache, slice,
1377                     AVL_AFTER)))
1378                         (void) tpool_dispatch(t, zpool_open_func, slice);
1379                 tpool_wait(t);
1380                 tpool_destroy(t);
1381
1382                 cookie = NULL;
1383                 while ((slice = avl_destroy_nodes(&slice_cache,
1384                     &cookie)) != NULL) {
1385                         if (slice->rn_config != NULL && !config_failed) {
1386                                 nvlist_t *config = slice->rn_config;
1387                                 boolean_t matched = B_TRUE;
1388
1389                                 if (iarg->poolname != NULL) {
1390                                         char *pname;
1391
1392                                         matched = nvlist_lookup_string(config,
1393                                             ZPOOL_CONFIG_POOL_NAME,
1394                                             &pname) == 0 &&
1395                                             strcmp(iarg->poolname, pname) == 0;
1396                                 } else if (iarg->guid != 0) {
1397                                         uint64_t this_guid;
1398
1399                                         matched = nvlist_lookup_uint64(config,
1400                                             ZPOOL_CONFIG_POOL_GUID,
1401                                             &this_guid) == 0 &&
1402                                             iarg->guid == this_guid;
1403                                 }
1404                                 if (matched) {
1405                                         /*
1406                                          * use the non-raw path for the config
1407                                          */
1408                                         (void) strlcpy(end, slice->rn_name,
1409                                             pathleft);
1410                                         if (add_config(hdl, &pools, path,
1411                                             config) != 0)
1412                                                 config_failed = B_TRUE;
1413                                 }
1414                                 nvlist_free(config);
1415                         }
1416                         free(slice->rn_name);
1417                         free(slice);
1418                 }
1419                 avl_destroy(&slice_cache);
1420
1421                 (void) closedir(dirp);
1422
1423                 if (config_failed)
1424                         goto error;
1425         }
1426
1427         ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy);
1428
1429 error:
1430         for (pe = pools.pools; pe != NULL; pe = penext) {
1431                 penext = pe->pe_next;
1432                 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1433                         venext = ve->ve_next;
1434                         for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1435                                 cenext = ce->ce_next;
1436                                 nvlist_free(ce->ce_config);
1437                                 free(ce);
1438                         }
1439                         free(ve);
1440                 }
1441                 free(pe);
1442         }
1443
1444         for (ne = pools.names; ne != NULL; ne = nenext) {
1445                 nenext = ne->ne_next;
1446                 free(ne->ne_name);
1447                 free(ne);
1448         }
1449
1450         return (ret);
1451 }
1452
1453 nvlist_t *
1454 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1455 {
1456         importargs_t iarg = { 0 };
1457
1458         iarg.paths = argc;
1459         iarg.path = argv;
1460
1461         return (zpool_find_import_impl(hdl, &iarg));
1462 }
1463
1464 /*
1465  * Given a cache file, return the contents as a list of importable pools.
1466  * poolname or guid (but not both) are provided by the caller when trying
1467  * to import a specific pool.
1468  */
1469 nvlist_t *
1470 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1471     char *poolname, uint64_t guid)
1472 {
1473         char *buf;
1474         int fd;
1475         struct stat64 statbuf;
1476         nvlist_t *raw, *src, *dst;
1477         nvlist_t *pools;
1478         nvpair_t *elem;
1479         char *name;
1480         uint64_t this_guid;
1481         boolean_t active;
1482
1483         verify(poolname == NULL || guid == 0);
1484
1485         if ((fd = open(cachefile, O_RDONLY)) < 0) {
1486                 zfs_error_aux(hdl, "%s", strerror(errno));
1487                 (void) zfs_error(hdl, EZFS_BADCACHE,
1488                     dgettext(TEXT_DOMAIN, "failed to open cache file"));
1489                 return (NULL);
1490         }
1491
1492         if (fstat64(fd, &statbuf) != 0) {
1493                 zfs_error_aux(hdl, "%s", strerror(errno));
1494                 (void) close(fd);
1495                 (void) zfs_error(hdl, EZFS_BADCACHE,
1496                     dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1497                 return (NULL);
1498         }
1499
1500         if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1501                 (void) close(fd);
1502                 return (NULL);
1503         }
1504
1505         if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1506                 (void) close(fd);
1507                 free(buf);
1508                 (void) zfs_error(hdl, EZFS_BADCACHE,
1509                     dgettext(TEXT_DOMAIN,
1510                     "failed to read cache file contents"));
1511                 return (NULL);
1512         }
1513
1514         (void) close(fd);
1515
1516         if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1517                 free(buf);
1518                 (void) zfs_error(hdl, EZFS_BADCACHE,
1519                     dgettext(TEXT_DOMAIN,
1520                     "invalid or corrupt cache file contents"));
1521                 return (NULL);
1522         }
1523
1524         free(buf);
1525
1526         /*
1527          * Go through and get the current state of the pools and refresh their
1528          * state.
1529          */
1530         if (nvlist_alloc(&pools, 0, 0) != 0) {
1531                 (void) no_memory(hdl);
1532                 nvlist_free(raw);
1533                 return (NULL);
1534         }
1535
1536         elem = NULL;
1537         while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1538                 src = fnvpair_value_nvlist(elem);
1539
1540                 name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1541                 if (poolname != NULL && strcmp(poolname, name) != 0)
1542                         continue;
1543
1544                 this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1545                 if (guid != 0 && guid != this_guid)
1546                         continue;
1547
1548                 if (pool_active(hdl, name, this_guid, &active) != 0) {
1549                         nvlist_free(raw);
1550                         nvlist_free(pools);
1551                         return (NULL);
1552                 }
1553
1554                 if (active)
1555                         continue;
1556
1557                 if (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE,
1558                     cachefile) != 0) {
1559                         (void) no_memory(hdl);
1560                         nvlist_free(raw);
1561                         nvlist_free(pools);
1562                         return (NULL);
1563                 }
1564
1565                 if ((dst = refresh_config(hdl, src)) == NULL) {
1566                         nvlist_free(raw);
1567                         nvlist_free(pools);
1568                         return (NULL);
1569                 }
1570
1571                 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1572                         (void) no_memory(hdl);
1573                         nvlist_free(dst);
1574                         nvlist_free(raw);
1575                         nvlist_free(pools);
1576                         return (NULL);
1577                 }
1578                 nvlist_free(dst);
1579         }
1580
1581         nvlist_free(raw);
1582         return (pools);
1583 }
1584
1585 static int
1586 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1587 {
1588         importargs_t *import = data;
1589         int found = 0;
1590
1591         if (import->poolname != NULL) {
1592                 char *pool_name;
1593
1594                 verify(nvlist_lookup_string(zhp->zpool_config,
1595                     ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1596                 if (strcmp(pool_name, import->poolname) == 0)
1597                         found = 1;
1598         } else {
1599                 uint64_t pool_guid;
1600
1601                 verify(nvlist_lookup_uint64(zhp->zpool_config,
1602                     ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1603                 if (pool_guid == import->guid)
1604                         found = 1;
1605         }
1606
1607         zpool_close(zhp);
1608         return (found);
1609 }
1610
1611 nvlist_t *
1612 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1613 {
1614         nvlist_t *pools = NULL;
1615
1616         verify(import->poolname == NULL || import->guid == 0);
1617
1618         if (import->unique)
1619                 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1620
1621         if (import->cachefile != NULL)
1622                 pools = zpool_find_import_cached(hdl, import->cachefile,
1623                     import->poolname, import->guid);
1624         else
1625                 pools = zpool_find_import_impl(hdl, import);
1626
1627         return (pools);
1628 }
1629
1630 static boolean_t
1631 pool_match(nvlist_t *cfg, char *tgt)
1632 {
1633         uint64_t v, guid = strtoull(tgt, NULL, 0);
1634         char *s;
1635
1636         if (guid != 0) {
1637                 if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &v) == 0)
1638                         return (v == guid);
1639         } else {
1640                 if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &s) == 0)
1641                         return (strcmp(s, tgt) == 0);
1642         }
1643         return (B_FALSE);
1644 }
1645
1646 int
1647 zpool_tryimport(libzfs_handle_t *hdl, char *target, nvlist_t **configp,
1648     importargs_t *args)
1649 {
1650         nvlist_t *pools;
1651         nvlist_t *match = NULL;
1652         nvlist_t *config = NULL;
1653         char *sepp = NULL;
1654         int count = 0;
1655         char *targetdup = strdup(target);
1656
1657         *configp = NULL;
1658
1659         if ((sepp = strpbrk(targetdup, "/@")) != NULL) {
1660                 *sepp = '\0';
1661         }
1662
1663         pools = zpool_search_import(hdl, args);
1664
1665         if (pools != NULL) {
1666                 nvpair_t *elem = NULL;
1667                 while ((elem = nvlist_next_nvpair(pools, elem)) != NULL) {
1668                         VERIFY0(nvpair_value_nvlist(elem, &config));
1669                         if (pool_match(config, targetdup)) {
1670                                 count++;
1671                                 if (match != NULL) {
1672                                         /* multiple matches found */
1673                                         continue;
1674                                 } else {
1675                                         match = config;
1676                                 }
1677                         }
1678                 }
1679         }
1680
1681         if (count == 0) {
1682                 free(targetdup);
1683                 return (ENOENT);
1684         }
1685
1686         if (count > 1) {
1687                 free(targetdup);
1688                 return (EINVAL);
1689         }
1690
1691         *configp = match;
1692         free(targetdup);
1693
1694         return (0);
1695 }
1696
1697 boolean_t
1698 find_guid(nvlist_t *nv, uint64_t guid)
1699 {
1700         uint64_t tmp;
1701         nvlist_t **child;
1702         uint_t c, children;
1703
1704         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1705         if (tmp == guid)
1706                 return (B_TRUE);
1707
1708         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1709             &child, &children) == 0) {
1710                 for (c = 0; c < children; c++)
1711                         if (find_guid(child[c], guid))
1712                                 return (B_TRUE);
1713         }
1714
1715         return (B_FALSE);
1716 }
1717
1718 typedef struct aux_cbdata {
1719         const char      *cb_type;
1720         uint64_t        cb_guid;
1721         zpool_handle_t  *cb_zhp;
1722 } aux_cbdata_t;
1723
1724 static int
1725 find_aux(zpool_handle_t *zhp, void *data)
1726 {
1727         aux_cbdata_t *cbp = data;
1728         nvlist_t **list;
1729         uint_t i, count;
1730         uint64_t guid;
1731         nvlist_t *nvroot;
1732
1733         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1734             &nvroot) == 0);
1735
1736         if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1737             &list, &count) == 0) {
1738                 for (i = 0; i < count; i++) {
1739                         verify(nvlist_lookup_uint64(list[i],
1740                             ZPOOL_CONFIG_GUID, &guid) == 0);
1741                         if (guid == cbp->cb_guid) {
1742                                 cbp->cb_zhp = zhp;
1743                                 return (1);
1744                         }
1745                 }
1746         }
1747
1748         zpool_close(zhp);
1749         return (0);
1750 }
1751
1752 /*
1753  * Determines if the pool is in use.  If so, it returns true and the state of
1754  * the pool as well as the name of the pool.  Both strings are allocated and
1755  * must be freed by the caller.
1756  */
1757 int
1758 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1759     boolean_t *inuse)
1760 {
1761         nvlist_t *config;
1762         char *name;
1763         boolean_t ret;
1764         uint64_t guid, vdev_guid;
1765         zpool_handle_t *zhp;
1766         nvlist_t *pool_config;
1767         uint64_t stateval, isspare;
1768         aux_cbdata_t cb = { 0 };
1769         boolean_t isactive;
1770
1771         *inuse = B_FALSE;
1772
1773         if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1774                 (void) no_memory(hdl);
1775                 return (-1);
1776         }
1777
1778         if (config == NULL)
1779                 return (0);
1780
1781         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1782             &stateval) == 0);
1783         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1784             &vdev_guid) == 0);
1785
1786         if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1787                 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1788                     &name) == 0);
1789                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1790                     &guid) == 0);
1791         }
1792
1793         switch (stateval) {
1794         case POOL_STATE_EXPORTED:
1795                 /*
1796                  * A pool with an exported state may in fact be imported
1797                  * read-only, so check the in-core state to see if it's
1798                  * active and imported read-only.  If it is, set
1799                  * its state to active.
1800                  */
1801                 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1802                     (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1803                         if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1804                                 stateval = POOL_STATE_ACTIVE;
1805
1806                         /*
1807                          * All we needed the zpool handle for is the
1808                          * readonly prop check.
1809                          */
1810                         zpool_close(zhp);
1811                 }
1812
1813                 ret = B_TRUE;
1814                 break;
1815
1816         case POOL_STATE_ACTIVE:
1817                 /*
1818                  * For an active pool, we have to determine if it's really part
1819                  * of a currently active pool (in which case the pool will exist
1820                  * and the guid will be the same), or whether it's part of an
1821                  * active pool that was disconnected without being explicitly
1822                  * exported.
1823                  */
1824                 if (pool_active(hdl, name, guid, &isactive) != 0) {
1825                         nvlist_free(config);
1826                         return (-1);
1827                 }
1828
1829                 if (isactive) {
1830                         /*
1831                          * Because the device may have been removed while
1832                          * offlined, we only report it as active if the vdev is
1833                          * still present in the config.  Otherwise, pretend like
1834                          * it's not in use.
1835                          */
1836                         if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1837                             (pool_config = zpool_get_config(zhp, NULL))
1838                             != NULL) {
1839                                 nvlist_t *nvroot;
1840
1841                                 verify(nvlist_lookup_nvlist(pool_config,
1842                                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1843                                 ret = find_guid(nvroot, vdev_guid);
1844                         } else {
1845                                 ret = B_FALSE;
1846                         }
1847
1848                         /*
1849                          * If this is an active spare within another pool, we
1850                          * treat it like an unused hot spare.  This allows the
1851                          * user to create a pool with a hot spare that currently
1852                          * in use within another pool.  Since we return B_TRUE,
1853                          * libdiskmgt will continue to prevent generic consumers
1854                          * from using the device.
1855                          */
1856                         if (ret && nvlist_lookup_uint64(config,
1857                             ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1858                                 stateval = POOL_STATE_SPARE;
1859
1860                         if (zhp != NULL)
1861                                 zpool_close(zhp);
1862                 } else {
1863                         stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1864                         ret = B_TRUE;
1865                 }
1866                 break;
1867
1868         case POOL_STATE_SPARE:
1869                 /*
1870                  * For a hot spare, it can be either definitively in use, or
1871                  * potentially active.  To determine if it's in use, we iterate
1872                  * over all pools in the system and search for one with a spare
1873                  * with a matching guid.
1874                  *
1875                  * Due to the shared nature of spares, we don't actually report
1876                  * the potentially active case as in use.  This means the user
1877                  * can freely create pools on the hot spares of exported pools,
1878                  * but to do otherwise makes the resulting code complicated, and
1879                  * we end up having to deal with this case anyway.
1880                  */
1881                 cb.cb_zhp = NULL;
1882                 cb.cb_guid = vdev_guid;
1883                 cb.cb_type = ZPOOL_CONFIG_SPARES;
1884                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1885                         name = (char *)zpool_get_name(cb.cb_zhp);
1886                         ret = B_TRUE;
1887                 } else {
1888                         ret = B_FALSE;
1889                 }
1890                 break;
1891
1892         case POOL_STATE_L2CACHE:
1893
1894                 /*
1895                  * Check if any pool is currently using this l2cache device.
1896                  */
1897                 cb.cb_zhp = NULL;
1898                 cb.cb_guid = vdev_guid;
1899                 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1900                 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1901                         name = (char *)zpool_get_name(cb.cb_zhp);
1902                         ret = B_TRUE;
1903                 } else {
1904                         ret = B_FALSE;
1905                 }
1906                 break;
1907
1908         default:
1909                 ret = B_FALSE;
1910         }
1911
1912
1913         if (ret) {
1914                 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1915                         if (cb.cb_zhp)
1916                                 zpool_close(cb.cb_zhp);
1917                         nvlist_free(config);
1918                         return (-1);
1919                 }
1920                 *state = (pool_state_t)stateval;
1921         }
1922
1923         if (cb.cb_zhp)
1924                 zpool_close(cb.cb_zhp);
1925
1926         nvlist_free(config);
1927         *inuse = ret;
1928         return (0);
1929 }