Home | History | Annotate | Download | only in common
      1 /*
      2  * CDDL HEADER START
      3  *
      4  * The contents of this file are subject to the terms of the
      5  * Common Development and Distribution License (the "License").
      6  * You may not use this file except in compliance with the License.
      7  *
      8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
      9  * or http://www.opensolaris.org/os/licensing.
     10  * See the License for the specific language governing permissions
     11  * and limitations under the License.
     12  *
     13  * When distributing Covered Code, include this CDDL HEADER in each
     14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
     15  * If applicable, add the following below this CDDL HEADER, with the
     16  * fields enclosed by brackets "[]" replaced with your own identifying
     17  * information: Portions Copyright [yyyy] [name of copyright owner]
     18  *
     19  * CDDL HEADER END
     20  */
     21 /*
     22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
     23  * Use is subject to license terms.
     24  */
     25 
     26 /*
     27  * Pool import support functions.
     28  *
     29  * To import a pool, we rely on reading the configuration information from the
     30  * ZFS label of each device.  If we successfully read the label, then we
     31  * organize the configuration information in the following hierarchy:
     32  *
     33  * 	pool guid -> toplevel vdev guid -> label txg
     34  *
     35  * Duplicate entries matching this same tuple will be discarded.  Once we have
     36  * examined every device, we pick the best label txg config for each toplevel
     37  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
     38  * update any paths that have changed.  Finally, we attempt to import the pool
     39  * using our derived config, and record the results.
     40  */
     41 
     42 #include <ctype.h>
     43 #include <devid.h>
     44 #include <dirent.h>
     45 #include <errno.h>
     46 #include <libintl.h>
     47 #include <stddef.h>
     48 #include <stdlib.h>
     49 #include <string.h>
     50 #include <sys/stat.h>
     51 #include <unistd.h>
     52 #include <fcntl.h>
     53 #include <sys/vtoc.h>
     54 #include <sys/dktp/fdisk.h>
     55 #include <sys/efi_partition.h>
     56 #include <thread_pool.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 	int fd;
     99 	ddi_devid_t devid;
    100 	char *minor, *ret;
    101 
    102 	if ((fd = open(path, O_RDONLY)) < 0)
    103 		return (NULL);
    104 
    105 	minor = NULL;
    106 	ret = NULL;
    107 	if (devid_get(fd, &devid) == 0) {
    108 		if (devid_get_minor_name(fd, &minor) == 0)
    109 			ret = devid_str_encode(devid, minor);
    110 		if (minor != NULL)
    111 			devid_str_free(minor);
    112 		devid_free(devid);
    113 	}
    114 	(void) close(fd);
    115 
    116 	return (ret);
    117 }
    118 
    119 
    120 /*
    121  * Go through and fix up any path and/or devid information for the given vdev
    122  * configuration.
    123  */
    124 static int
    125 fix_paths(nvlist_t *nv, name_entry_t *names)
    126 {
    127 	nvlist_t **child;
    128 	uint_t c, children;
    129 	uint64_t guid;
    130 	name_entry_t *ne, *best;
    131 	char *path, *devid;
    132 	int matched;
    133 
    134 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
    135 	    &child, &children) == 0) {
    136 		for (c = 0; c < children; c++)
    137 			if (fix_paths(child[c], names) != 0)
    138 				return (-1);
    139 		return (0);
    140 	}
    141 
    142 	/*
    143 	 * This is a leaf (file or disk) vdev.  In either case, go through
    144 	 * the name list and see if we find a matching guid.  If so, replace
    145 	 * the path and see if we can calculate a new devid.
    146 	 *
    147 	 * There may be multiple names associated with a particular guid, in
    148 	 * which case we have overlapping slices or multiple paths to the same
    149 	 * disk.  If this is the case, then we want to pick the path that is
    150 	 * the most similar to the original, where "most similar" is the number
    151 	 * of matching characters starting from the end of the path.  This will
    152 	 * preserve slice numbers even if the disks have been reorganized, and
    153 	 * will also catch preferred disk names if multiple paths exist.
    154 	 */
    155 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
    156 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
    157 		path = NULL;
    158 
    159 	matched = 0;
    160 	best = NULL;
    161 	for (ne = names; ne != NULL; ne = ne->ne_next) {
    162 		if (ne->ne_guid == guid) {
    163 			const char *src, *dst;
    164 			int count;
    165 
    166 			if (path == NULL) {
    167 				best = ne;
    168 				break;
    169 			}
    170 
    171 			src = ne->ne_name + strlen(ne->ne_name) - 1;
    172 			dst = path + strlen(path) - 1;
    173 			for (count = 0; src >= ne->ne_name && dst >= path;
    174 			    src--, dst--, count++)
    175 				if (*src != *dst)
    176 					break;
    177 
    178 			/*
    179 			 * At this point, 'count' is the number of characters
    180 			 * matched from the end.
    181 			 */
    182 			if (count > matched || best == NULL) {
    183 				best = ne;
    184 				matched = count;
    185 			}
    186 		}
    187 	}
    188 
    189 	if (best == NULL)
    190 		return (0);
    191 
    192 	if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
    193 		return (-1);
    194 
    195 	if ((devid = get_devid(best->ne_name)) == NULL) {
    196 		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
    197 	} else {
    198 		if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
    199 			return (-1);
    200 		devid_str_free(devid);
    201 	}
    202 
    203 	return (0);
    204 }
    205 
    206 /*
    207  * Add the given configuration to the list of known devices.
    208  */
    209 static int
    210 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
    211     nvlist_t *config)
    212 {
    213 	uint64_t pool_guid, vdev_guid, top_guid, txg, state;
    214 	pool_entry_t *pe;
    215 	vdev_entry_t *ve;
    216 	config_entry_t *ce;
    217 	name_entry_t *ne;
    218 
    219 	/*
    220 	 * If this is a hot spare not currently in use or level 2 cache
    221 	 * device, add it to the list of names to translate, but don't do
    222 	 * anything else.
    223 	 */
    224 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
    225 	    &state) == 0 &&
    226 	    (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
    227 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
    228 		if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
    229 			return (-1);
    230 
    231 		if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
    232 			free(ne);
    233 			return (-1);
    234 		}
    235 		ne->ne_guid = vdev_guid;
    236 		ne->ne_next = pl->names;
    237 		pl->names = ne;
    238 		return (0);
    239 	}
    240 
    241 	/*
    242 	 * If we have a valid config but cannot read any of these fields, then
    243 	 * it means we have a half-initialized label.  In vdev_label_init()
    244 	 * we write a label with txg == 0 so that we can identify the device
    245 	 * in case the user refers to the same disk later on.  If we fail to
    246 	 * create the pool, we'll be left with a label in this state
    247 	 * which should not be considered part of a valid pool.
    248 	 */
    249 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
    250 	    &pool_guid) != 0 ||
    251 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
    252 	    &vdev_guid) != 0 ||
    253 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
    254 	    &top_guid) != 0 ||
    255 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
    256 	    &txg) != 0 || txg == 0) {
    257 		nvlist_free(config);
    258 		return (0);
    259 	}
    260 
    261 	/*
    262 	 * First, see if we know about this pool.  If not, then add it to the
    263 	 * list of known pools.
    264 	 */
    265 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
    266 		if (pe->pe_guid == pool_guid)
    267 			break;
    268 	}
    269 
    270 	if (pe == NULL) {
    271 		if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
    272 			nvlist_free(config);
    273 			return (-1);
    274 		}
    275 		pe->pe_guid = pool_guid;
    276 		pe->pe_next = pl->pools;
    277 		pl->pools = pe;
    278 	}
    279 
    280 	/*
    281 	 * Second, see if we know about this toplevel vdev.  Add it if its
    282 	 * missing.
    283 	 */
    284 	for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
    285 		if (ve->ve_guid == top_guid)
    286 			break;
    287 	}
    288 
    289 	if (ve == NULL) {
    290 		if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
    291 			nvlist_free(config);
    292 			return (-1);
    293 		}
    294 		ve->ve_guid = top_guid;
    295 		ve->ve_next = pe->pe_vdevs;
    296 		pe->pe_vdevs = ve;
    297 	}
    298 
    299 	/*
    300 	 * Third, see if we have a config with a matching transaction group.  If
    301 	 * so, then we do nothing.  Otherwise, add it to the list of known
    302 	 * configs.
    303 	 */
    304 	for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
    305 		if (ce->ce_txg == txg)
    306 			break;
    307 	}
    308 
    309 	if (ce == NULL) {
    310 		if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
    311 			nvlist_free(config);
    312 			return (-1);
    313 		}
    314 		ce->ce_txg = txg;
    315 		ce->ce_config = config;
    316 		ce->ce_next = ve->ve_configs;
    317 		ve->ve_configs = ce;
    318 	} else {
    319 		nvlist_free(config);
    320 	}
    321 
    322 	/*
    323 	 * At this point we've successfully added our config to the list of
    324 	 * known configs.  The last thing to do is add the vdev guid -> path
    325 	 * mappings so that we can fix up the configuration as necessary before
    326 	 * doing the import.
    327 	 */
    328 	if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
    329 		return (-1);
    330 
    331 	if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
    332 		free(ne);
    333 		return (-1);
    334 	}
    335 
    336 	ne->ne_guid = vdev_guid;
    337 	ne->ne_next = pl->names;
    338 	pl->names = ne;
    339 
    340 	return (0);
    341 }
    342 
    343 /*
    344  * Returns true if the named pool matches the given GUID.
    345  */
    346 static int
    347 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
    348     boolean_t *isactive)
    349 {
    350 	zpool_handle_t *zhp;
    351 	uint64_t theguid;
    352 
    353 	if (zpool_open_silent(hdl, name, &zhp) != 0)
    354 		return (-1);
    355 
    356 	if (zhp == NULL) {
    357 		*isactive = B_FALSE;
    358 		return (0);
    359 	}
    360 
    361 	verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
    362 	    &theguid) == 0);
    363 
    364 	zpool_close(zhp);
    365 
    366 	*isactive = (theguid == guid);
    367 	return (0);
    368 }
    369 
    370 static nvlist_t *
    371 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
    372 {
    373 	nvlist_t *nvl;
    374 	zfs_cmd_t zc = { 0 };
    375 	int err;
    376 
    377 	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
    378 		return (NULL);
    379 
    380 	if (zcmd_alloc_dst_nvlist(hdl, &zc,
    381 	    zc.zc_nvlist_conf_size * 2) != 0) {
    382 		zcmd_free_nvlists(&zc);
    383 		return (NULL);
    384 	}
    385 
    386 	while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
    387 	    &zc)) != 0 && errno == ENOMEM) {
    388 		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
    389 			zcmd_free_nvlists(&zc);
    390 			return (NULL);
    391 		}
    392 	}
    393 
    394 	if (err) {
    395 		zcmd_free_nvlists(&zc);
    396 		return (NULL);
    397 	}
    398 
    399 	if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
    400 		zcmd_free_nvlists(&zc);
    401 		return (NULL);
    402 	}
    403 
    404 	zcmd_free_nvlists(&zc);
    405 	return (nvl);
    406 }
    407 
    408 /*
    409  * Determine if the vdev id is a hole in the namespace.
    410  */
    411 boolean_t
    412 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
    413 {
    414 	for (int c = 0; c < holes; c++) {
    415 
    416 		/* Top-level is a hole */
    417 		if (hole_array[c] == id)
    418 			return (B_TRUE);
    419 	}
    420 	return (B_FALSE);
    421 }
    422 
    423 /*
    424  * Convert our list of pools into the definitive set of configurations.  We
    425  * start by picking the best config for each toplevel vdev.  Once that's done,
    426  * we assemble the toplevel vdevs into a full config for the pool.  We make a
    427  * pass to fix up any incorrect paths, and then add it to the main list to
    428  * return to the user.
    429  */
    430 static nvlist_t *
    431 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
    432 {
    433 	pool_entry_t *pe;
    434 	vdev_entry_t *ve;
    435 	config_entry_t *ce;
    436 	nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
    437 	nvlist_t **spares, **l2cache;
    438 	uint_t i, nspares, nl2cache;
    439 	boolean_t config_seen;
    440 	uint64_t best_txg;
    441 	char *name, *hostname;
    442 	uint64_t version, guid;
    443 	uint_t children = 0;
    444 	nvlist_t **child = NULL;
    445 	uint_t holes;
    446 	uint64_t *hole_array, max_id;
    447 	uint_t c;
    448 	boolean_t isactive;
    449 	uint64_t hostid;
    450 	nvlist_t *nvl;
    451 	boolean_t found_one = B_FALSE;
    452 	boolean_t valid_top_config = B_FALSE;
    453 
    454 	if (nvlist_alloc(&ret, 0, 0) != 0)
    455 		goto nomem;
    456 
    457 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
    458 		uint64_t id, max_txg = 0;
    459 
    460 		if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
    461 			goto nomem;
    462 		config_seen = B_FALSE;
    463 
    464 		/*
    465 		 * Iterate over all toplevel vdevs.  Grab the pool configuration
    466 		 * from the first one we find, and then go through the rest and
    467 		 * add them as necessary to the 'vdevs' member of the config.
    468 		 */
    469 		for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
    470 
    471 			/*
    472 			 * Determine the best configuration for this vdev by
    473 			 * selecting the config with the latest transaction
    474 			 * group.
    475 			 */
    476 			best_txg = 0;
    477 			for (ce = ve->ve_configs; ce != NULL;
    478 			    ce = ce->ce_next) {
    479 
    480 				if (ce->ce_txg > best_txg) {
    481 					tmp = ce->ce_config;
    482 					best_txg = ce->ce_txg;
    483 				}
    484 			}
    485 
    486 			/*
    487 			 * We rely on the fact that the max txg for the
    488 			 * pool will contain the most up-to-date information
    489 			 * about the valid top-levels in the vdev namespace.
    490 			 */
    491 			if (best_txg > max_txg) {
    492 				(void) nvlist_remove(config,
    493 				    ZPOOL_CONFIG_VDEV_CHILDREN,
    494 				    DATA_TYPE_UINT64);
    495 				(void) nvlist_remove(config,
    496 				    ZPOOL_CONFIG_HOLE_ARRAY,
    497 				    DATA_TYPE_UINT64_ARRAY);
    498 
    499 				max_txg = best_txg;
    500 				hole_array = NULL;
    501 				holes = 0;
    502 				max_id = 0;
    503 				valid_top_config = B_FALSE;
    504 
    505 				if (nvlist_lookup_uint64(tmp,
    506 				    ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
    507 					verify(nvlist_add_uint64(config,
    508 					    ZPOOL_CONFIG_VDEV_CHILDREN,
    509 					    max_id) == 0);
    510 					valid_top_config = B_TRUE;
    511 				}
    512 
    513 				if (nvlist_lookup_uint64_array(tmp,
    514 				    ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
    515 				    &holes) == 0) {
    516 					verify(nvlist_add_uint64_array(config,
    517 					    ZPOOL_CONFIG_HOLE_ARRAY,
    518 					    hole_array, holes) == 0);
    519 				}
    520 			}
    521 
    522 			if (!config_seen) {
    523 				/*
    524 				 * Copy the relevant pieces of data to the pool
    525 				 * configuration:
    526 				 *
    527 				 *	version
    528 				 * 	pool guid
    529 				 * 	name
    530 				 * 	pool state
    531 				 *	hostid (if available)
    532 				 *	hostname (if available)
    533 				 */
    534 				uint64_t state;
    535 
    536 				verify(nvlist_lookup_uint64(tmp,
    537 				    ZPOOL_CONFIG_VERSION, &version) == 0);
    538 				if (nvlist_add_uint64(config,
    539 				    ZPOOL_CONFIG_VERSION, version) != 0)
    540 					goto nomem;
    541 				verify(nvlist_lookup_uint64(tmp,
    542 				    ZPOOL_CONFIG_POOL_GUID, &guid) == 0);
    543 				if (nvlist_add_uint64(config,
    544 				    ZPOOL_CONFIG_POOL_GUID, guid) != 0)
    545 					goto nomem;
    546 				verify(nvlist_lookup_string(tmp,
    547 				    ZPOOL_CONFIG_POOL_NAME, &name) == 0);
    548 				if (nvlist_add_string(config,
    549 				    ZPOOL_CONFIG_POOL_NAME, name) != 0)
    550 					goto nomem;
    551 				verify(nvlist_lookup_uint64(tmp,
    552 				    ZPOOL_CONFIG_POOL_STATE, &state) == 0);
    553 				if (nvlist_add_uint64(config,
    554 				    ZPOOL_CONFIG_POOL_STATE, state) != 0)
    555 					goto nomem;
    556 				hostid = 0;
    557 				if (nvlist_lookup_uint64(tmp,
    558 				    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
    559 					if (nvlist_add_uint64(config,
    560 					    ZPOOL_CONFIG_HOSTID, hostid) != 0)
    561 						goto nomem;
    562 					verify(nvlist_lookup_string(tmp,
    563 					    ZPOOL_CONFIG_HOSTNAME,
    564 					    &hostname) == 0);
    565 					if (nvlist_add_string(config,
    566 					    ZPOOL_CONFIG_HOSTNAME,
    567 					    hostname) != 0)
    568 						goto nomem;
    569 				}
    570 
    571 				config_seen = B_TRUE;
    572 			}
    573 
    574 			/*
    575 			 * Add this top-level vdev to the child array.
    576 			 */
    577 			verify(nvlist_lookup_nvlist(tmp,
    578 			    ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
    579 			verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
    580 			    &id) == 0);
    581 
    582 			if (id >= children) {
    583 				nvlist_t **newchild;
    584 
    585 				newchild = zfs_alloc(hdl, (id + 1) *
    586 				    sizeof (nvlist_t *));
    587 				if (newchild == NULL)
    588 					goto nomem;
    589 
    590 				for (c = 0; c < children; c++)
    591 					newchild[c] = child[c];
    592 
    593 				free(child);
    594 				child = newchild;
    595 				children = id + 1;
    596 			}
    597 			if (nvlist_dup(nvtop, &child[id], 0) != 0)
    598 				goto nomem;
    599 
    600 		}
    601 
    602 		/*
    603 		 * If we have information about all the top-levels then
    604 		 * clean up the nvlist which we've constructed. This
    605 		 * means removing any extraneous devices that are
    606 		 * beyond the valid range or adding devices to the end
    607 		 * of our array which appear to be missing.
    608 		 */
    609 		if (valid_top_config) {
    610 			if (max_id < children) {
    611 				for (c = max_id; c < children; c++)
    612 					nvlist_free(child[c]);
    613 				children = max_id;
    614 			} else if (max_id > children) {
    615 				nvlist_t **newchild;
    616 
    617 				newchild = zfs_alloc(hdl, (max_id) *
    618 				    sizeof (nvlist_t *));
    619 				if (newchild == NULL)
    620 					goto nomem;
    621 
    622 				for (c = 0; c < children; c++)
    623 					newchild[c] = child[c];
    624 
    625 				free(child);
    626 				child = newchild;
    627 				children = max_id;
    628 			}
    629 		}
    630 
    631 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
    632 		    &guid) == 0);
    633 
    634 		/*
    635 		 * The vdev namespace may contain holes as a result of
    636 		 * device removal. We must add them back into the vdev
    637 		 * tree before we process any missing devices.
    638 		 */
    639 		if (holes > 0) {
    640 			ASSERT(valid_top_config);
    641 
    642 			for (c = 0; c < children; c++) {
    643 				nvlist_t *holey;
    644 
    645 				if (child[c] != NULL ||
    646 				    !vdev_is_hole(hole_array, holes, c))
    647 					continue;
    648 
    649 				if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
    650 				    0) != 0)
    651 					goto nomem;
    652 
    653 				/*
    654 				 * Holes in the namespace are treated as
    655 				 * "hole" top-level vdevs and have a
    656 				 * special flag set on them.
    657 				 */
    658 				if (nvlist_add_string(holey,
    659 				    ZPOOL_CONFIG_TYPE,
    660 				    VDEV_TYPE_HOLE) != 0 ||
    661 				    nvlist_add_uint64(holey,
    662 				    ZPOOL_CONFIG_ID, c) != 0 ||
    663 				    nvlist_add_uint64(holey,
    664 				    ZPOOL_CONFIG_GUID, 0ULL) != 0)
    665 					goto nomem;
    666 				child[c] = holey;
    667 			}
    668 		}
    669 
    670 		/*
    671 		 * Look for any missing top-level vdevs.  If this is the case,
    672 		 * create a faked up 'missing' vdev as a placeholder.  We cannot
    673 		 * simply compress the child array, because the kernel performs
    674 		 * certain checks to make sure the vdev IDs match their location
    675 		 * in the configuration.
    676 		 */
    677 		for (c = 0; c < children; c++) {
    678 			if (child[c] == NULL) {
    679 				nvlist_t *missing;
    680 				if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
    681 				    0) != 0)
    682 					goto nomem;
    683 				if (nvlist_add_string(missing,
    684 				    ZPOOL_CONFIG_TYPE,
    685 				    VDEV_TYPE_MISSING) != 0 ||
    686 				    nvlist_add_uint64(missing,
    687 				    ZPOOL_CONFIG_ID, c) != 0 ||
    688 				    nvlist_add_uint64(missing,
    689 				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
    690 					nvlist_free(missing);
    691 					goto nomem;
    692 				}
    693 				child[c] = missing;
    694 			}
    695 		}
    696 
    697 		/*
    698 		 * Put all of this pool's top-level vdevs into a root vdev.
    699 		 */
    700 		if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
    701 			goto nomem;
    702 		if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
    703 		    VDEV_TYPE_ROOT) != 0 ||
    704 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
    705 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
    706 		    nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
    707 		    child, children) != 0) {
    708 			nvlist_free(nvroot);
    709 			goto nomem;
    710 		}
    711 
    712 		for (c = 0; c < children; c++)
    713 			nvlist_free(child[c]);
    714 		free(child);
    715 		children = 0;
    716 		child = NULL;
    717 
    718 		/*
    719 		 * Go through and fix up any paths and/or devids based on our
    720 		 * known list of vdev GUID -> path mappings.
    721 		 */
    722 		if (fix_paths(nvroot, pl->names) != 0) {
    723 			nvlist_free(nvroot);
    724 			goto nomem;
    725 		}
    726 
    727 		/*
    728 		 * Add the root vdev to this pool's configuration.
    729 		 */
    730 		if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
    731 		    nvroot) != 0) {
    732 			nvlist_free(nvroot);
    733 			goto nomem;
    734 		}
    735 		nvlist_free(nvroot);
    736 
    737 		/*
    738 		 * zdb uses this path to report on active pools that were
    739 		 * imported or created using -R.
    740 		 */
    741 		if (active_ok)
    742 			goto add_pool;
    743 
    744 		/*
    745 		 * Determine if this pool is currently active, in which case we
    746 		 * can't actually import it.
    747 		 */
    748 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
    749 		    &name) == 0);
    750 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
    751 		    &guid) == 0);
    752 
    753 		if (pool_active(hdl, name, guid, &isactive) != 0)
    754 			goto error;
    755 
    756 		if (isactive) {
    757 			nvlist_free(config);
    758 			config = NULL;
    759 			continue;
    760 		}
    761 
    762 		if ((nvl = refresh_config(hdl, config)) == NULL) {
    763 			nvlist_free(config);
    764 			config = NULL;
    765 			continue;
    766 		}
    767 
    768 		nvlist_free(config);
    769 		config = nvl;
    770 
    771 		/*
    772 		 * Go through and update the paths for spares, now that we have
    773 		 * them.
    774 		 */
    775 		verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
    776 		    &nvroot) == 0);
    777 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
    778 		    &spares, &nspares) == 0) {
    779 			for (i = 0; i < nspares; i++) {
    780 				if (fix_paths(spares[i], pl->names) != 0)
    781 					goto nomem;
    782 			}
    783 		}
    784 
    785 		/*
    786 		 * Update the paths for l2cache devices.
    787 		 */
    788 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
    789 		    &l2cache, &nl2cache) == 0) {
    790 			for (i = 0; i < nl2cache; i++) {
    791 				if (fix_paths(l2cache[i], pl->names) != 0)
    792 					goto nomem;
    793 			}
    794 		}
    795 
    796 		/*
    797 		 * Restore the original information read from the actual label.
    798 		 */
    799 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
    800 		    DATA_TYPE_UINT64);
    801 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
    802 		    DATA_TYPE_STRING);
    803 		if (hostid != 0) {
    804 			verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
    805 			    hostid) == 0);
    806 			verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
    807 			    hostname) == 0);
    808 		}
    809 
    810 add_pool:
    811 		/*
    812 		 * Add this pool to the list of configs.
    813 		 */
    814 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
    815 		    &name) == 0);
    816 		if (nvlist_add_nvlist(ret, name, config) != 0)
    817 			goto nomem;
    818 
    819 		found_one = B_TRUE;
    820 		nvlist_free(config);
    821 		config = NULL;
    822 	}
    823 
    824 	if (!found_one) {
    825 		nvlist_free(ret);
    826 		ret = NULL;
    827 	}
    828 
    829 	return (ret);
    830 
    831 nomem:
    832 	(void) no_memory(hdl);
    833 error:
    834 	nvlist_free(config);
    835 	nvlist_free(ret);
    836 	for (c = 0; c < children; c++)
    837 		nvlist_free(child[c]);
    838 	free(child);
    839 
    840 	return (NULL);
    841 }
    842 
    843 /*
    844  * Return the offset of the given label.
    845  */
    846 static uint64_t
    847 label_offset(uint64_t size, int l)
    848 {
    849 	ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
    850 	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
    851 	    0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
    852 }
    853 
    854 /*
    855  * Given a file descriptor, read the label information and return an nvlist
    856  * describing the configuration, if there is one.
    857  */
    858 int
    859 zpool_read_label(int fd, nvlist_t **config)
    860 {
    861 	struct stat64 statbuf;
    862 	int l;
    863 	vdev_label_t *label;
    864 	uint64_t state, txg, size;
    865 
    866 	*config = NULL;
    867 
    868 	if (fstat64(fd, &statbuf) == -1)
    869 		return (0);
    870 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
    871 
    872 	if ((label = malloc(sizeof (vdev_label_t))) == NULL)
    873 		return (-1);
    874 
    875 	for (l = 0; l < VDEV_LABELS; l++) {
    876 		if (pread64(fd, label, sizeof (vdev_label_t),
    877 		    label_offset(size, l)) != sizeof (vdev_label_t))
    878 			continue;
    879 
    880 		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
    881 		    sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
    882 			continue;
    883 
    884 		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
    885 		    &state) != 0 || state > POOL_STATE_L2CACHE) {
    886 			nvlist_free(*config);
    887 			continue;
    888 		}
    889 
    890 		if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
    891 		    (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
    892 		    &txg) != 0 || txg == 0)) {
    893 			nvlist_free(*config);
    894 			continue;
    895 		}
    896 
    897 		free(label);
    898 		return (0);
    899 	}
    900 
    901 	free(label);
    902 	*config = NULL;
    903 	return (0);
    904 }
    905 
    906 typedef struct rdsk_node {
    907 	char *rn_name;
    908 	int rn_dfd;
    909 	libzfs_handle_t *rn_hdl;
    910 	nvlist_t *rn_config;
    911 	avl_tree_t *rn_avl;
    912 	avl_node_t rn_node;
    913 	boolean_t rn_nozpool;
    914 } rdsk_node_t;
    915 
    916 static int
    917 slice_cache_compare(const void *arg1, const void *arg2)
    918 {
    919 	const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
    920 	const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
    921 	char *nm1slice, *nm2slice;
    922 	int rv;
    923 
    924 	/*
    925 	 * slices zero and two are the most likely to provide results,
    926 	 * so put those first
    927 	 */
    928 	nm1slice = strstr(nm1, "s0");
    929 	nm2slice = strstr(nm2, "s0");
    930 	if (nm1slice && !nm2slice) {
    931 		return (-1);
    932 	}
    933 	if (!nm1slice && nm2slice) {
    934 		return (1);
    935 	}
    936 	nm1slice = strstr(nm1, "s2");
    937 	nm2slice = strstr(nm2, "s2");
    938 	if (nm1slice && !nm2slice) {
    939 		return (-1);
    940 	}
    941 	if (!nm1slice && nm2slice) {
    942 		return (1);
    943 	}
    944 
    945 	rv = strcmp(nm1, nm2);
    946 	if (rv == 0)
    947 		return (0);
    948 	return (rv > 0 ? 1 : -1);
    949 }
    950 
    951 static void
    952 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
    953     diskaddr_t size, uint_t blksz)
    954 {
    955 	rdsk_node_t tmpnode;
    956 	rdsk_node_t *node;
    957 	char sname[MAXNAMELEN];
    958 
    959 	tmpnode.rn_name = &sname[0];
    960 	(void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
    961 	    diskname, partno);
    962 	/* too small to contain a zpool? */
    963 	if ((size < (SPA_MINDEVSIZE / blksz)) &&
    964 	    (node = avl_find(r, &tmpnode, NULL)))
    965 		node->rn_nozpool = B_TRUE;
    966 }
    967 
    968 static void
    969 nozpool_all_slices(avl_tree_t *r, const char *sname)
    970 {
    971 	char diskname[MAXNAMELEN];
    972 	char *ptr;
    973 	int i;
    974 
    975 	(void) strncpy(diskname, sname, MAXNAMELEN);
    976 	if (((ptr = strrchr(diskname, 's')) == NULL) &&
    977 	    ((ptr = strrchr(diskname, 'p')) == NULL))
    978 		return;
    979 	ptr[0] = 's';
    980 	ptr[1] = '\0';
    981 	for (i = 0; i < NDKMAP; i++)
    982 		check_one_slice(r, diskname, i, 0, 1);
    983 	ptr[0] = 'p';
    984 	for (i = 0; i <= FD_NUMPART; i++)
    985 		check_one_slice(r, diskname, i, 0, 1);
    986 }
    987 
    988 static void
    989 check_slices(avl_tree_t *r, int fd, const char *sname)
    990 {
    991 	struct extvtoc vtoc;
    992 	struct dk_gpt *gpt;
    993 	char diskname[MAXNAMELEN];
    994 	char *ptr;
    995 	int i;
    996 
    997 	(void) strncpy(diskname, sname, MAXNAMELEN);
    998 	if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
    999 		return;
   1000 	ptr[1] = '\0';
   1001 
   1002 	if (read_extvtoc(fd, &vtoc) >= 0) {
   1003 		for (i = 0; i < NDKMAP; i++)
   1004 			check_one_slice(r, diskname, i,
   1005 			    vtoc.v_part[i].p_size, vtoc.v_sectorsz);
   1006 	} else if (efi_alloc_and_read(fd, &gpt) >= 0) {
   1007 		/*
   1008 		 * on x86 we'll still have leftover links that point
   1009 		 * to slices s[9-15], so use NDKMAP instead
   1010 		 */
   1011 		for (i = 0; i < NDKMAP; i++)
   1012 			check_one_slice(r, diskname, i,
   1013 			    gpt->efi_parts[i].p_size, gpt->efi_lbasize);
   1014 		/* nodes p[1-4] are never used with EFI labels */
   1015 		ptr[0] = 'p';
   1016 		for (i = 1; i <= FD_NUMPART; i++)
   1017 			check_one_slice(r, diskname, i, 0, 1);
   1018 		efi_free(gpt);
   1019 	}
   1020 }
   1021 
   1022 static void
   1023 zpool_open_func(void *arg)
   1024 {
   1025 	rdsk_node_t *rn = arg;
   1026 	struct stat64 statbuf;
   1027 	nvlist_t *config;
   1028 	int fd;
   1029 
   1030 	if (rn->rn_nozpool)
   1031 		return;
   1032 	if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
   1033 		/* symlink to a device that's no longer there */
   1034 		if (errno == ENOENT)
   1035 			nozpool_all_slices(rn->rn_avl, rn->rn_name);
   1036 		return;
   1037 	}
   1038 	/*
   1039 	 * Ignore failed stats.  We only want regular
   1040 	 * files, character devs and block devs.
   1041 	 */
   1042 	if (fstat64(fd, &statbuf) != 0 ||
   1043 	    (!S_ISREG(statbuf.st_mode) &&
   1044 	    !S_ISCHR(statbuf.st_mode) &&
   1045 	    !S_ISBLK(statbuf.st_mode))) {
   1046 		(void) close(fd);
   1047 		return;
   1048 	}
   1049 	/* this file is too small to hold a zpool */
   1050 	if (S_ISREG(statbuf.st_mode) &&
   1051 	    statbuf.st_size < SPA_MINDEVSIZE) {
   1052 		(void) close(fd);
   1053 		return;
   1054 	} else if (!S_ISREG(statbuf.st_mode)) {
   1055 		/*
   1056 		 * Try to read the disk label first so we don't have to
   1057 		 * open a bunch of minor nodes that can't have a zpool.
   1058 		 */
   1059 		check_slices(rn->rn_avl, fd, rn->rn_name);
   1060 	}
   1061 
   1062 	if ((zpool_read_label(fd, &config)) != 0) {
   1063 		(void) close(fd);
   1064 		(void) no_memory(rn->rn_hdl);
   1065 		return;
   1066 	}
   1067 	(void) close(fd);
   1068 
   1069 
   1070 	rn->rn_config = config;
   1071 	if (config != NULL) {
   1072 		assert(rn->rn_nozpool == B_FALSE);
   1073 	}
   1074 }
   1075 
   1076 /*
   1077  * Given a file descriptor, clear (zero) the label information.  This function
   1078  * is currently only used in the appliance stack as part of the ZFS sysevent
   1079  * module.
   1080  */
   1081 int
   1082 zpool_clear_label(int fd)
   1083 {
   1084 	struct stat64 statbuf;
   1085 	int l;
   1086 	vdev_label_t *label;
   1087 	uint64_t size;
   1088 
   1089 	if (fstat64(fd, &statbuf) == -1)
   1090 		return (0);
   1091 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
   1092 
   1093 	if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
   1094 		return (-1);
   1095 
   1096 	for (l = 0; l < VDEV_LABELS; l++) {
   1097 		if (pwrite64(fd, label, sizeof (vdev_label_t),
   1098 		    label_offset(size, l)) != sizeof (vdev_label_t))
   1099 			return (-1);
   1100 	}
   1101 
   1102 	free(label);
   1103 	return (0);
   1104 }
   1105 
   1106 /*
   1107  * Given a list of directories to search, find all pools stored on disk.  This
   1108  * includes partial pools which are not available to import.  If no args are
   1109  * given (argc is 0), then the default directory (/dev/dsk) is searched.
   1110  * poolname or guid (but not both) are provided by the caller when trying
   1111  * to import a specific pool.
   1112  */
   1113 static nvlist_t *
   1114 zpool_find_import_impl(libzfs_handle_t *hdl, int argc, char **argv,
   1115     boolean_t active_ok, char *poolname, uint64_t guid)
   1116 {
   1117 	int i;
   1118 	DIR *dirp = NULL;
   1119 	struct dirent64 *dp;
   1120 	char path[MAXPATHLEN];
   1121 	char *end;
   1122 	size_t pathleft;
   1123 	nvlist_t *ret = NULL;
   1124 	static char *default_dir = "/dev/dsk";
   1125 	pool_list_t pools = { 0 };
   1126 	pool_entry_t *pe, *penext;
   1127 	vdev_entry_t *ve, *venext;
   1128 	config_entry_t *ce, *cenext;
   1129 	name_entry_t *ne, *nenext;
   1130 	avl_tree_t slice_cache;
   1131 	rdsk_node_t *slice;
   1132 	void *cookie;
   1133 
   1134 	verify(poolname == NULL || guid == 0);
   1135 
   1136 	if (argc == 0) {
   1137 		argc = 1;
   1138 		argv = &default_dir;
   1139 	}
   1140 
   1141 	/*
   1142 	 * Go through and read the label configuration information from every
   1143 	 * possible device, organizing the information according to pool GUID
   1144 	 * and toplevel GUID.
   1145 	 */
   1146 	for (i = 0; i < argc; i++) {
   1147 		tpool_t *t;
   1148 		char *rdsk;
   1149 		int dfd;
   1150 
   1151 		/* use realpath to normalize the path */
   1152 		if (realpath(argv[i], path) == 0) {
   1153 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
   1154 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
   1155 			    argv[i]);
   1156 			goto error;
   1157 		}
   1158 		end = &path[strlen(path)];
   1159 		*end++ = '/';
   1160 		*end = 0;
   1161 		pathleft = &path[sizeof (path)] - end;
   1162 
   1163 		/*
   1164 		 * Using raw devices instead of block devices when we're
   1165 		 * reading the labels skips a bunch of slow operations during
   1166 		 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
   1167 		 */
   1168 		if (strcmp(path, "/dev/dsk/") == 0)
   1169 			rdsk = "/dev/rdsk/";
   1170 		else
   1171 			rdsk = path;
   1172 
   1173 		if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
   1174 		    (dirp = fdopendir(dfd)) == NULL) {
   1175 			zfs_error_aux(hdl, strerror(errno));
   1176 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
   1177 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
   1178 			    rdsk);
   1179 			goto error;
   1180 		}
   1181 
   1182 		avl_create(&slice_cache, slice_cache_compare,
   1183 		    sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
   1184 		/*
   1185 		 * This is not MT-safe, but we have no MT consumers of libzfs
   1186 		 */
   1187 		while ((dp = readdir64(dirp)) != NULL) {
   1188 			const char *name = dp->d_name;
   1189 			if (name[0] == '.' &&
   1190 			    (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
   1191 				continue;
   1192 
   1193 			slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
   1194 			slice->rn_name = zfs_strdup(hdl, name);
   1195 			slice->rn_avl = &slice_cache;
   1196 			slice->rn_dfd = dfd;
   1197 			slice->rn_hdl = hdl;
   1198 			slice->rn_nozpool = B_FALSE;
   1199 			avl_add(&slice_cache, slice);
   1200 		}
   1201 		/*
   1202 		 * create a thread pool to do all of this in parallel;
   1203 		 * rn_nozpool is not protected, so this is racy in that
   1204 		 * multiple tasks could decide that the same slice can
   1205 		 * not hold a zpool, which is benign.  Also choose
   1206 		 * double the number of processors; we hold a lot of
   1207 		 * locks in the kernel, so going beyond this doesn't
   1208 		 * buy us much.
   1209 		 */
   1210 		t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
   1211 		    0, NULL);
   1212 		for (slice = avl_first(&slice_cache); slice;
   1213 		    (slice = avl_walk(&slice_cache, slice,
   1214 		    AVL_AFTER)))
   1215 			(void) tpool_dispatch(t, zpool_open_func, slice);
   1216 		tpool_wait(t);
   1217 		tpool_destroy(t);
   1218 
   1219 		cookie = NULL;
   1220 		while ((slice = avl_destroy_nodes(&slice_cache,
   1221 		    &cookie)) != NULL) {
   1222 			if (slice->rn_config != NULL) {
   1223 				nvlist_t *config = slice->rn_config;
   1224 				boolean_t matched = B_TRUE;
   1225 
   1226 				if (poolname != NULL) {
   1227 					char *pname;
   1228 
   1229 					matched = nvlist_lookup_string(config,
   1230 					    ZPOOL_CONFIG_POOL_NAME,
   1231 					    &pname) == 0 &&
   1232 					    strcmp(poolname, pname) == 0;
   1233 				} else if (guid != 0) {
   1234 					uint64_t this_guid;
   1235 
   1236 					matched = nvlist_lookup_uint64(config,
   1237 					    ZPOOL_CONFIG_POOL_GUID,
   1238 					    &this_guid) == 0 &&
   1239 					    guid == this_guid;
   1240 				}
   1241 				if (!matched) {
   1242 					nvlist_free(config);
   1243 					config = NULL;
   1244 					continue;
   1245 				}
   1246 				/* use the non-raw path for the config */
   1247 				(void) strlcpy(end, slice->rn_name, pathleft);
   1248 				if (add_config(hdl, &pools, path, config) != 0)
   1249 					goto error;
   1250 			}
   1251 			free(slice->rn_name);
   1252 			free(slice);
   1253 		}
   1254 		avl_destroy(&slice_cache);
   1255 
   1256 		(void) closedir(dirp);
   1257 		dirp = NULL;
   1258 	}
   1259 
   1260 	ret = get_configs(hdl, &pools, active_ok);
   1261 
   1262 error:
   1263 	for (pe = pools.pools; pe != NULL; pe = penext) {
   1264 		penext = pe->pe_next;
   1265 		for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
   1266 			venext = ve->ve_next;
   1267 			for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
   1268 				cenext = ce->ce_next;
   1269 				if (ce->ce_config)
   1270 					nvlist_free(ce->ce_config);
   1271 				free(ce);
   1272 			}
   1273 			free(ve);
   1274 		}
   1275 		free(pe);
   1276 	}
   1277 
   1278 	for (ne = pools.names; ne != NULL; ne = nenext) {
   1279 		nenext = ne->ne_next;
   1280 		if (ne->ne_name)
   1281 			free(ne->ne_name);
   1282 		free(ne);
   1283 	}
   1284 
   1285 	if (dirp)
   1286 		(void) closedir(dirp);
   1287 
   1288 	return (ret);
   1289 }
   1290 
   1291 nvlist_t *
   1292 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
   1293 {
   1294 	return (zpool_find_import_impl(hdl, argc, argv, B_FALSE, NULL, 0));
   1295 }
   1296 
   1297 nvlist_t *
   1298 zpool_find_import_byname(libzfs_handle_t *hdl, int argc, char **argv,
   1299     char *pool)
   1300 {
   1301 	return (zpool_find_import_impl(hdl, argc, argv, B_FALSE, pool, 0));
   1302 }
   1303 
   1304 nvlist_t *
   1305 zpool_find_import_byguid(libzfs_handle_t *hdl, int argc, char **argv,
   1306     uint64_t guid)
   1307 {
   1308 	return (zpool_find_import_impl(hdl, argc, argv, B_FALSE, NULL, guid));
   1309 }
   1310 
   1311 nvlist_t *
   1312 zpool_find_import_activeok(libzfs_handle_t *hdl, int argc, char **argv)
   1313 {
   1314 	return (zpool_find_import_impl(hdl, argc, argv, B_TRUE, NULL, 0));
   1315 }
   1316 
   1317 /*
   1318  * Given a cache file, return the contents as a list of importable pools.
   1319  * poolname or guid (but not both) are provided by the caller when trying
   1320  * to import a specific pool.
   1321  */
   1322 nvlist_t *
   1323 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
   1324     char *poolname, uint64_t guid)
   1325 {
   1326 	char *buf;
   1327 	int fd;
   1328 	struct stat64 statbuf;
   1329 	nvlist_t *raw, *src, *dst;
   1330 	nvlist_t *pools;
   1331 	nvpair_t *elem;
   1332 	char *name;
   1333 	uint64_t this_guid;
   1334 	boolean_t active;
   1335 
   1336 	verify(poolname == NULL || guid == 0);
   1337 
   1338 	if ((fd = open(cachefile, O_RDONLY)) < 0) {
   1339 		zfs_error_aux(hdl, "%s", strerror(errno));
   1340 		(void) zfs_error(hdl, EZFS_BADCACHE,
   1341 		    dgettext(TEXT_DOMAIN, "failed to open cache file"));
   1342 		return (NULL);
   1343 	}
   1344 
   1345 	if (fstat64(fd, &statbuf) != 0) {
   1346 		zfs_error_aux(hdl, "%s", strerror(errno));
   1347 		(void) close(fd);
   1348 		(void) zfs_error(hdl, EZFS_BADCACHE,
   1349 		    dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
   1350 		return (NULL);
   1351 	}
   1352 
   1353 	if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
   1354 		(void) close(fd);
   1355 		return (NULL);
   1356 	}
   1357 
   1358 	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
   1359 		(void) close(fd);
   1360 		free(buf);
   1361 		(void) zfs_error(hdl, EZFS_BADCACHE,
   1362 		    dgettext(TEXT_DOMAIN,
   1363 		    "failed to read cache file contents"));
   1364 		return (NULL);
   1365 	}
   1366 
   1367 	(void) close(fd);
   1368 
   1369 	if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
   1370 		free(buf);
   1371 		(void) zfs_error(hdl, EZFS_BADCACHE,
   1372 		    dgettext(TEXT_DOMAIN,
   1373 		    "invalid or corrupt cache file contents"));
   1374 		return (NULL);
   1375 	}
   1376 
   1377 	free(buf);
   1378 
   1379 	/*
   1380 	 * Go through and get the current state of the pools and refresh their
   1381 	 * state.
   1382 	 */
   1383 	if (nvlist_alloc(&pools, 0, 0) != 0) {
   1384 		(void) no_memory(hdl);
   1385 		nvlist_free(raw);
   1386 		return (NULL);
   1387 	}
   1388 
   1389 	elem = NULL;
   1390 	while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
   1391 		verify(nvpair_value_nvlist(elem, &src) == 0);
   1392 
   1393 		verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
   1394 		    &name) == 0);
   1395 		if (poolname != NULL && strcmp(poolname, name) != 0)
   1396 			continue;
   1397 
   1398 		verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
   1399 		    &this_guid) == 0);
   1400 		if (guid != 0) {
   1401 			verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
   1402 			    &this_guid) == 0);
   1403 			if (guid != this_guid)
   1404 				continue;
   1405 		}
   1406 
   1407 		if (pool_active(hdl, name, this_guid, &active) != 0) {
   1408 			nvlist_free(raw);
   1409 			nvlist_free(pools);
   1410 			return (NULL);
   1411 		}
   1412 
   1413 		if (active)
   1414 			continue;
   1415 
   1416 		if ((dst = refresh_config(hdl, src)) == NULL) {
   1417 			nvlist_free(raw);
   1418 			nvlist_free(pools);
   1419 			return (NULL);
   1420 		}
   1421 
   1422 		if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
   1423 			(void) no_memory(hdl);
   1424 			nvlist_free(dst);
   1425 			nvlist_free(raw);
   1426 			nvlist_free(pools);
   1427 			return (NULL);
   1428 		}
   1429 		nvlist_free(dst);
   1430 	}
   1431 
   1432 	nvlist_free(raw);
   1433 	return (pools);
   1434 }
   1435 
   1436 
   1437 boolean_t
   1438 find_guid(nvlist_t *nv, uint64_t guid)
   1439 {
   1440 	uint64_t tmp;
   1441 	nvlist_t **child;
   1442 	uint_t c, children;
   1443 
   1444 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
   1445 	if (tmp == guid)
   1446 		return (B_TRUE);
   1447 
   1448 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
   1449 	    &child, &children) == 0) {
   1450 		for (c = 0; c < children; c++)
   1451 			if (find_guid(child[c], guid))
   1452 				return (B_TRUE);
   1453 	}
   1454 
   1455 	return (B_FALSE);
   1456 }
   1457 
   1458 typedef struct aux_cbdata {
   1459 	const char	*cb_type;
   1460 	uint64_t	cb_guid;
   1461 	zpool_handle_t	*cb_zhp;
   1462 } aux_cbdata_t;
   1463 
   1464 static int
   1465 find_aux(zpool_handle_t *zhp, void *data)
   1466 {
   1467 	aux_cbdata_t *cbp = data;
   1468 	nvlist_t **list;
   1469 	uint_t i, count;
   1470 	uint64_t guid;
   1471 	nvlist_t *nvroot;
   1472 
   1473 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
   1474 	    &nvroot) == 0);
   1475 
   1476 	if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
   1477 	    &list, &count) == 0) {
   1478 		for (i = 0; i < count; i++) {
   1479 			verify(nvlist_lookup_uint64(list[i],
   1480 			    ZPOOL_CONFIG_GUID, &guid) == 0);
   1481 			if (guid == cbp->cb_guid) {
   1482 				cbp->cb_zhp = zhp;
   1483 				return (1);
   1484 			}
   1485 		}
   1486 	}
   1487 
   1488 	zpool_close(zhp);
   1489 	return (0);
   1490 }
   1491 
   1492 /*
   1493  * Determines if the pool is in use.  If so, it returns true and the state of
   1494  * the pool as well as the name of the pool.  Both strings are allocated and
   1495  * must be freed by the caller.
   1496  */
   1497 int
   1498 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
   1499     boolean_t *inuse)
   1500 {
   1501 	nvlist_t *config;
   1502 	char *name;
   1503 	boolean_t ret;
   1504 	uint64_t guid, vdev_guid;
   1505 	zpool_handle_t *zhp;
   1506 	nvlist_t *pool_config;
   1507 	uint64_t stateval, isspare;
   1508 	aux_cbdata_t cb = { 0 };
   1509 	boolean_t isactive;
   1510 
   1511 	*inuse = B_FALSE;
   1512 
   1513 	if (zpool_read_label(fd, &config) != 0) {
   1514 		(void) no_memory(hdl);
   1515 		return (-1);
   1516 	}
   1517 
   1518 	if (config == NULL)
   1519 		return (0);
   1520 
   1521 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
   1522 	    &stateval) == 0);
   1523 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
   1524 	    &vdev_guid) == 0);
   1525 
   1526 	if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
   1527 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
   1528 		    &name) == 0);
   1529 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
   1530 		    &guid) == 0);
   1531 	}
   1532 
   1533 	switch (stateval) {
   1534 	case POOL_STATE_EXPORTED:
   1535 		ret = B_TRUE;
   1536 		break;
   1537 
   1538 	case POOL_STATE_ACTIVE:
   1539 		/*
   1540 		 * For an active pool, we have to determine if it's really part
   1541 		 * of a currently active pool (in which case the pool will exist
   1542 		 * and the guid will be the same), or whether it's part of an
   1543 		 * active pool that was disconnected without being explicitly
   1544 		 * exported.
   1545 		 */
   1546 		if (pool_active(hdl, name, guid, &isactive) != 0) {
   1547 			nvlist_free(config);
   1548 			return (-1);
   1549 		}
   1550 
   1551 		if (isactive) {
   1552 			/*
   1553 			 * Because the device may have been removed while
   1554 			 * offlined, we only report it as active if the vdev is
   1555 			 * still present in the config.  Otherwise, pretend like
   1556 			 * it's not in use.
   1557 			 */
   1558 			if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
   1559 			    (pool_config = zpool_get_config(zhp, NULL))
   1560 			    != NULL) {
   1561 				nvlist_t *nvroot;
   1562 
   1563 				verify(nvlist_lookup_nvlist(pool_config,
   1564 				    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
   1565 				ret = find_guid(nvroot, vdev_guid);
   1566 			} else {
   1567 				ret = B_FALSE;
   1568 			}
   1569 
   1570 			/*
   1571 			 * If this is an active spare within another pool, we
   1572 			 * treat it like an unused hot spare.  This allows the
   1573 			 * user to create a pool with a hot spare that currently
   1574 			 * in use within another pool.  Since we return B_TRUE,
   1575 			 * libdiskmgt will continue to prevent generic consumers
   1576 			 * from using the device.
   1577 			 */
   1578 			if (ret && nvlist_lookup_uint64(config,
   1579 			    ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
   1580 				stateval = POOL_STATE_SPARE;
   1581 
   1582 			if (zhp != NULL)
   1583 				zpool_close(zhp);
   1584 		} else {
   1585 			stateval = POOL_STATE_POTENTIALLY_ACTIVE;
   1586 			ret = B_TRUE;
   1587 		}
   1588 		break;
   1589 
   1590 	case POOL_STATE_SPARE:
   1591 		/*
   1592 		 * For a hot spare, it can be either definitively in use, or
   1593 		 * potentially active.  To determine if it's in use, we iterate
   1594 		 * over all pools in the system and search for one with a spare
   1595 		 * with a matching guid.
   1596 		 *
   1597 		 * Due to the shared nature of spares, we don't actually report
   1598 		 * the potentially active case as in use.  This means the user
   1599 		 * can freely create pools on the hot spares of exported pools,
   1600 		 * but to do otherwise makes the resulting code complicated, and
   1601 		 * we end up having to deal with this case anyway.
   1602 		 */
   1603 		cb.cb_zhp = NULL;
   1604 		cb.cb_guid = vdev_guid;
   1605 		cb.cb_type = ZPOOL_CONFIG_SPARES;
   1606 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
   1607 			name = (char *)zpool_get_name(cb.cb_zhp);
   1608 			ret = TRUE;
   1609 		} else {
   1610 			ret = FALSE;
   1611 		}
   1612 		break;
   1613 
   1614 	case POOL_STATE_L2CACHE:
   1615 
   1616 		/*
   1617 		 * Check if any pool is currently using this l2cache device.
   1618 		 */
   1619 		cb.cb_zhp = NULL;
   1620 		cb.cb_guid = vdev_guid;
   1621 		cb.cb_type = ZPOOL_CONFIG_L2CACHE;
   1622 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
   1623 			name = (char *)zpool_get_name(cb.cb_zhp);
   1624 			ret = TRUE;
   1625 		} else {
   1626 			ret = FALSE;
   1627 		}
   1628 		break;
   1629 
   1630 	default:
   1631 		ret = B_FALSE;
   1632 	}
   1633 
   1634 
   1635 	if (ret) {
   1636 		if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
   1637 			if (cb.cb_zhp)
   1638 				zpool_close(cb.cb_zhp);
   1639 			nvlist_free(config);
   1640 			return (-1);
   1641 		}
   1642 		*state = (pool_state_t)stateval;
   1643 	}
   1644 
   1645 	if (cb.cb_zhp)
   1646 		zpool_close(cb.cb_zhp);
   1647 
   1648 	nvlist_free(config);
   1649 	*inuse = ret;
   1650 	return (0);
   1651 }
   1652