libzfs_import.c revision ae5ee1bd5e54409b4589217b160c8cf4b81e2fb8
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 2017 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 <ctype.h>
46#include <devid.h>
47#include <dirent.h>
48#include <errno.h>
49#include <libintl.h>
50#include <stddef.h>
51#include <stdlib.h>
52#include <string.h>
53#include <sys/stat.h>
54#include <unistd.h>
55#include <fcntl.h>
56#include <sys/vtoc.h>
57#include <sys/dktp/fdisk.h>
58#include <sys/efi_partition.h>
59#include <thread_pool.h>
60
61#include <sys/vdev_impl.h>
62
63#include "libzfs.h"
64#include "libzfs_impl.h"
65
66/*
67 * Intermediate structures used to gather configuration information.
68 */
69typedef struct config_entry {
70	uint64_t		ce_txg;
71	nvlist_t		*ce_config;
72	struct config_entry	*ce_next;
73} config_entry_t;
74
75typedef struct vdev_entry {
76	uint64_t		ve_guid;
77	config_entry_t		*ve_configs;
78	struct vdev_entry	*ve_next;
79} vdev_entry_t;
80
81typedef struct pool_entry {
82	uint64_t		pe_guid;
83	vdev_entry_t		*pe_vdevs;
84	struct pool_entry	*pe_next;
85} pool_entry_t;
86
87typedef struct name_entry {
88	char			*ne_name;
89	uint64_t		ne_guid;
90	struct name_entry	*ne_next;
91} name_entry_t;
92
93typedef struct pool_list {
94	pool_entry_t		*pools;
95	name_entry_t		*names;
96} pool_list_t;
97
98/*
99 * Go through and fix up any path and/or devid information for the given vdev
100 * configuration.
101 */
102static int
103fix_paths(nvlist_t *nv, name_entry_t *names)
104{
105	nvlist_t **child;
106	uint_t c, children;
107	uint64_t guid;
108	name_entry_t *ne, *best;
109	char *path, *devid;
110	int matched;
111
112	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
113	    &child, &children) == 0) {
114		for (c = 0; c < children; c++)
115			if (fix_paths(child[c], names) != 0)
116				return (-1);
117		return (0);
118	}
119
120	/*
121	 * This is a leaf (file or disk) vdev.  In either case, go through
122	 * the name list and see if we find a matching guid.  If so, replace
123	 * the path and see if we can calculate a new devid.
124	 *
125	 * There may be multiple names associated with a particular guid, in
126	 * which case we have overlapping slices or multiple paths to the same
127	 * disk.  If this is the case, then we want to pick the path that is
128	 * the most similar to the original, where "most similar" is the number
129	 * of matching characters starting from the end of the path.  This will
130	 * preserve slice numbers even if the disks have been reorganized, and
131	 * will also catch preferred disk names if multiple paths exist.
132	 */
133	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
134	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
135		path = NULL;
136
137	matched = 0;
138	best = NULL;
139	for (ne = names; ne != NULL; ne = ne->ne_next) {
140		if (ne->ne_guid == guid) {
141			const char *src, *dst;
142			int count;
143
144			if (path == NULL) {
145				best = ne;
146				break;
147			}
148
149			src = ne->ne_name + strlen(ne->ne_name) - 1;
150			dst = path + strlen(path) - 1;
151			for (count = 0; src >= ne->ne_name && dst >= path;
152			    src--, dst--, count++)
153				if (*src != *dst)
154					break;
155
156			/*
157			 * At this point, 'count' is the number of characters
158			 * matched from the end.
159			 */
160			if (count > matched || best == NULL) {
161				best = ne;
162				matched = count;
163			}
164		}
165	}
166
167	if (best == NULL)
168		return (0);
169
170	if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
171		return (-1);
172
173	if ((devid = devid_str_from_path(best->ne_name)) == NULL) {
174		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
175	} else {
176		if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
177			devid_str_free(devid);
178			return (-1);
179		}
180		devid_str_free(devid);
181	}
182
183	return (0);
184}
185
186/*
187 * Add the given configuration to the list of known devices.
188 */
189static int
190add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
191    nvlist_t *config)
192{
193	uint64_t pool_guid, vdev_guid, top_guid, txg, state;
194	pool_entry_t *pe;
195	vdev_entry_t *ve;
196	config_entry_t *ce;
197	name_entry_t *ne;
198
199	/*
200	 * If this is a hot spare not currently in use or level 2 cache
201	 * device, add it to the list of names to translate, but don't do
202	 * anything else.
203	 */
204	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
205	    &state) == 0 &&
206	    (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
207	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
208		if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
209			return (-1);
210
211		if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
212			free(ne);
213			return (-1);
214		}
215
216		ne->ne_guid = vdev_guid;
217		ne->ne_next = pl->names;
218		pl->names = ne;
219
220		nvlist_free(config);
221		return (0);
222	}
223
224	/*
225	 * If we have a valid config but cannot read any of these fields, then
226	 * it means we have a half-initialized label.  In vdev_label_init()
227	 * we write a label with txg == 0 so that we can identify the device
228	 * in case the user refers to the same disk later on.  If we fail to
229	 * create the pool, we'll be left with a label in this state
230	 * which should not be considered part of a valid pool.
231	 */
232	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
233	    &pool_guid) != 0 ||
234	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
235	    &vdev_guid) != 0 ||
236	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
237	    &top_guid) != 0 ||
238	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
239	    &txg) != 0 || txg == 0) {
240		nvlist_free(config);
241		return (0);
242	}
243
244	/*
245	 * First, see if we know about this pool.  If not, then add it to the
246	 * list of known pools.
247	 */
248	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
249		if (pe->pe_guid == pool_guid)
250			break;
251	}
252
253	if (pe == NULL) {
254		if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
255			nvlist_free(config);
256			return (-1);
257		}
258		pe->pe_guid = pool_guid;
259		pe->pe_next = pl->pools;
260		pl->pools = pe;
261	}
262
263	/*
264	 * Second, see if we know about this toplevel vdev.  Add it if its
265	 * missing.
266	 */
267	for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
268		if (ve->ve_guid == top_guid)
269			break;
270	}
271
272	if (ve == NULL) {
273		if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
274			nvlist_free(config);
275			return (-1);
276		}
277		ve->ve_guid = top_guid;
278		ve->ve_next = pe->pe_vdevs;
279		pe->pe_vdevs = ve;
280	}
281
282	/*
283	 * Third, see if we have a config with a matching transaction group.  If
284	 * so, then we do nothing.  Otherwise, add it to the list of known
285	 * configs.
286	 */
287	for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
288		if (ce->ce_txg == txg)
289			break;
290	}
291
292	if (ce == NULL) {
293		if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
294			nvlist_free(config);
295			return (-1);
296		}
297		ce->ce_txg = txg;
298		ce->ce_config = config;
299		ce->ce_next = ve->ve_configs;
300		ve->ve_configs = ce;
301	} else {
302		nvlist_free(config);
303	}
304
305	/*
306	 * At this point we've successfully added our config to the list of
307	 * known configs.  The last thing to do is add the vdev guid -> path
308	 * mappings so that we can fix up the configuration as necessary before
309	 * doing the import.
310	 */
311	if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
312		return (-1);
313
314	if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
315		free(ne);
316		return (-1);
317	}
318
319	ne->ne_guid = vdev_guid;
320	ne->ne_next = pl->names;
321	pl->names = ne;
322
323	return (0);
324}
325
326/*
327 * Returns true if the named pool matches the given GUID.
328 */
329static int
330pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
331    boolean_t *isactive)
332{
333	zpool_handle_t *zhp;
334	uint64_t theguid;
335
336	if (zpool_open_silent(hdl, name, &zhp) != 0)
337		return (-1);
338
339	if (zhp == NULL) {
340		*isactive = B_FALSE;
341		return (0);
342	}
343
344	verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
345	    &theguid) == 0);
346
347	zpool_close(zhp);
348
349	*isactive = (theguid == guid);
350	return (0);
351}
352
353static nvlist_t *
354refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
355{
356	nvlist_t *nvl;
357	zfs_cmd_t zc = { 0 };
358	int err, dstbuf_size;
359
360	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
361		return (NULL);
362
363	dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
364
365	if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
366		zcmd_free_nvlists(&zc);
367		return (NULL);
368	}
369
370	while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
371	    &zc)) != 0 && errno == ENOMEM) {
372		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
373			zcmd_free_nvlists(&zc);
374			return (NULL);
375		}
376	}
377
378	if (err) {
379		zcmd_free_nvlists(&zc);
380		return (NULL);
381	}
382
383	if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
384		zcmd_free_nvlists(&zc);
385		return (NULL);
386	}
387
388	zcmd_free_nvlists(&zc);
389	return (nvl);
390}
391
392/*
393 * Determine if the vdev id is a hole in the namespace.
394 */
395boolean_t
396vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
397{
398	for (int c = 0; c < holes; c++) {
399
400		/* Top-level is a hole */
401		if (hole_array[c] == id)
402			return (B_TRUE);
403	}
404	return (B_FALSE);
405}
406
407/*
408 * Convert our list of pools into the definitive set of configurations.  We
409 * start by picking the best config for each toplevel vdev.  Once that's done,
410 * we assemble the toplevel vdevs into a full config for the pool.  We make a
411 * pass to fix up any incorrect paths, and then add it to the main list to
412 * return to the user.
413 */
414static nvlist_t *
415get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok,
416    nvlist_t *policy)
417{
418	pool_entry_t *pe;
419	vdev_entry_t *ve;
420	config_entry_t *ce;
421	nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
422	nvlist_t **spares, **l2cache;
423	uint_t i, nspares, nl2cache;
424	boolean_t config_seen;
425	uint64_t best_txg;
426	char *name, *hostname = NULL;
427	uint64_t guid;
428	uint_t children = 0;
429	nvlist_t **child = NULL;
430	uint_t holes;
431	uint64_t *hole_array, max_id;
432	uint_t c;
433	boolean_t isactive;
434	uint64_t hostid;
435	nvlist_t *nvl;
436	boolean_t found_one = B_FALSE;
437	boolean_t valid_top_config = B_FALSE;
438
439	if (nvlist_alloc(&ret, 0, 0) != 0)
440		goto nomem;
441
442	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
443		uint64_t id, max_txg = 0;
444
445		if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
446			goto nomem;
447		config_seen = B_FALSE;
448
449		/*
450		 * Iterate over all toplevel vdevs.  Grab the pool configuration
451		 * from the first one we find, and then go through the rest and
452		 * add them as necessary to the 'vdevs' member of the config.
453		 */
454		for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
455
456			/*
457			 * Determine the best configuration for this vdev by
458			 * selecting the config with the latest transaction
459			 * group.
460			 */
461			best_txg = 0;
462			for (ce = ve->ve_configs; ce != NULL;
463			    ce = ce->ce_next) {
464
465				if (ce->ce_txg > best_txg) {
466					tmp = ce->ce_config;
467					best_txg = ce->ce_txg;
468				}
469			}
470
471			/*
472			 * We rely on the fact that the max txg for the
473			 * pool will contain the most up-to-date information
474			 * about the valid top-levels in the vdev namespace.
475			 */
476			if (best_txg > max_txg) {
477				(void) nvlist_remove(config,
478				    ZPOOL_CONFIG_VDEV_CHILDREN,
479				    DATA_TYPE_UINT64);
480				(void) nvlist_remove(config,
481				    ZPOOL_CONFIG_HOLE_ARRAY,
482				    DATA_TYPE_UINT64_ARRAY);
483
484				max_txg = best_txg;
485				hole_array = NULL;
486				holes = 0;
487				max_id = 0;
488				valid_top_config = B_FALSE;
489
490				if (nvlist_lookup_uint64(tmp,
491				    ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
492					verify(nvlist_add_uint64(config,
493					    ZPOOL_CONFIG_VDEV_CHILDREN,
494					    max_id) == 0);
495					valid_top_config = B_TRUE;
496				}
497
498				if (nvlist_lookup_uint64_array(tmp,
499				    ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
500				    &holes) == 0) {
501					verify(nvlist_add_uint64_array(config,
502					    ZPOOL_CONFIG_HOLE_ARRAY,
503					    hole_array, holes) == 0);
504				}
505			}
506
507			if (!config_seen) {
508				/*
509				 * Copy the relevant pieces of data to the pool
510				 * configuration:
511				 *
512				 *	version
513				 *	pool guid
514				 *	name
515				 *	comment (if available)
516				 *	pool state
517				 *	hostid (if available)
518				 *	hostname (if available)
519				 */
520				uint64_t state, version;
521				char *comment = NULL;
522
523				version = fnvlist_lookup_uint64(tmp,
524				    ZPOOL_CONFIG_VERSION);
525				fnvlist_add_uint64(config,
526				    ZPOOL_CONFIG_VERSION, version);
527				guid = fnvlist_lookup_uint64(tmp,
528				    ZPOOL_CONFIG_POOL_GUID);
529				fnvlist_add_uint64(config,
530				    ZPOOL_CONFIG_POOL_GUID, guid);
531				name = fnvlist_lookup_string(tmp,
532				    ZPOOL_CONFIG_POOL_NAME);
533				fnvlist_add_string(config,
534				    ZPOOL_CONFIG_POOL_NAME, name);
535
536				if (nvlist_lookup_string(tmp,
537				    ZPOOL_CONFIG_COMMENT, &comment) == 0)
538					fnvlist_add_string(config,
539					    ZPOOL_CONFIG_COMMENT, comment);
540
541				state = fnvlist_lookup_uint64(tmp,
542				    ZPOOL_CONFIG_POOL_STATE);
543				fnvlist_add_uint64(config,
544				    ZPOOL_CONFIG_POOL_STATE, state);
545
546				hostid = 0;
547				if (nvlist_lookup_uint64(tmp,
548				    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
549					fnvlist_add_uint64(config,
550					    ZPOOL_CONFIG_HOSTID, hostid);
551					hostname = fnvlist_lookup_string(tmp,
552					    ZPOOL_CONFIG_HOSTNAME);
553					fnvlist_add_string(config,
554					    ZPOOL_CONFIG_HOSTNAME, hostname);
555				}
556
557				config_seen = B_TRUE;
558			}
559
560			/*
561			 * Add this top-level vdev to the child array.
562			 */
563			verify(nvlist_lookup_nvlist(tmp,
564			    ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
565			verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
566			    &id) == 0);
567
568			if (id >= children) {
569				nvlist_t **newchild;
570
571				newchild = zfs_alloc(hdl, (id + 1) *
572				    sizeof (nvlist_t *));
573				if (newchild == NULL)
574					goto nomem;
575
576				for (c = 0; c < children; c++)
577					newchild[c] = child[c];
578
579				free(child);
580				child = newchild;
581				children = id + 1;
582			}
583			if (nvlist_dup(nvtop, &child[id], 0) != 0)
584				goto nomem;
585
586		}
587
588		/*
589		 * If we have information about all the top-levels then
590		 * clean up the nvlist which we've constructed. This
591		 * means removing any extraneous devices that are
592		 * beyond the valid range or adding devices to the end
593		 * of our array which appear to be missing.
594		 */
595		if (valid_top_config) {
596			if (max_id < children) {
597				for (c = max_id; c < children; c++)
598					nvlist_free(child[c]);
599				children = max_id;
600			} else if (max_id > children) {
601				nvlist_t **newchild;
602
603				newchild = zfs_alloc(hdl, (max_id) *
604				    sizeof (nvlist_t *));
605				if (newchild == NULL)
606					goto nomem;
607
608				for (c = 0; c < children; c++)
609					newchild[c] = child[c];
610
611				free(child);
612				child = newchild;
613				children = max_id;
614			}
615		}
616
617		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
618		    &guid) == 0);
619
620		/*
621		 * The vdev namespace may contain holes as a result of
622		 * device removal. We must add them back into the vdev
623		 * tree before we process any missing devices.
624		 */
625		if (holes > 0) {
626			ASSERT(valid_top_config);
627
628			for (c = 0; c < children; c++) {
629				nvlist_t *holey;
630
631				if (child[c] != NULL ||
632				    !vdev_is_hole(hole_array, holes, c))
633					continue;
634
635				if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
636				    0) != 0)
637					goto nomem;
638
639				/*
640				 * Holes in the namespace are treated as
641				 * "hole" top-level vdevs and have a
642				 * special flag set on them.
643				 */
644				if (nvlist_add_string(holey,
645				    ZPOOL_CONFIG_TYPE,
646				    VDEV_TYPE_HOLE) != 0 ||
647				    nvlist_add_uint64(holey,
648				    ZPOOL_CONFIG_ID, c) != 0 ||
649				    nvlist_add_uint64(holey,
650				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
651					nvlist_free(holey);
652					goto nomem;
653				}
654				child[c] = holey;
655			}
656		}
657
658		/*
659		 * Look for any missing top-level vdevs.  If this is the case,
660		 * create a faked up 'missing' vdev as a placeholder.  We cannot
661		 * simply compress the child array, because the kernel performs
662		 * certain checks to make sure the vdev IDs match their location
663		 * in the configuration.
664		 */
665		for (c = 0; c < children; c++) {
666			if (child[c] == NULL) {
667				nvlist_t *missing;
668				if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
669				    0) != 0)
670					goto nomem;
671				if (nvlist_add_string(missing,
672				    ZPOOL_CONFIG_TYPE,
673				    VDEV_TYPE_MISSING) != 0 ||
674				    nvlist_add_uint64(missing,
675				    ZPOOL_CONFIG_ID, c) != 0 ||
676				    nvlist_add_uint64(missing,
677				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
678					nvlist_free(missing);
679					goto nomem;
680				}
681				child[c] = missing;
682			}
683		}
684
685		/*
686		 * Put all of this pool's top-level vdevs into a root vdev.
687		 */
688		if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
689			goto nomem;
690		if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
691		    VDEV_TYPE_ROOT) != 0 ||
692		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
693		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
694		    nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
695		    child, children) != 0) {
696			nvlist_free(nvroot);
697			goto nomem;
698		}
699
700		for (c = 0; c < children; c++)
701			nvlist_free(child[c]);
702		free(child);
703		children = 0;
704		child = NULL;
705
706		/*
707		 * Go through and fix up any paths and/or devids based on our
708		 * known list of vdev GUID -> path mappings.
709		 */
710		if (fix_paths(nvroot, pl->names) != 0) {
711			nvlist_free(nvroot);
712			goto nomem;
713		}
714
715		/*
716		 * Add the root vdev to this pool's configuration.
717		 */
718		if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
719		    nvroot) != 0) {
720			nvlist_free(nvroot);
721			goto nomem;
722		}
723		nvlist_free(nvroot);
724
725		/*
726		 * zdb uses this path to report on active pools that were
727		 * imported or created using -R.
728		 */
729		if (active_ok)
730			goto add_pool;
731
732		/*
733		 * Determine if this pool is currently active, in which case we
734		 * can't actually import it.
735		 */
736		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
737		    &name) == 0);
738		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
739		    &guid) == 0);
740
741		if (pool_active(hdl, name, guid, &isactive) != 0)
742			goto error;
743
744		if (isactive) {
745			nvlist_free(config);
746			config = NULL;
747			continue;
748		}
749
750		if (policy != NULL) {
751			if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY,
752			    policy) != 0)
753				goto nomem;
754		}
755
756		if ((nvl = refresh_config(hdl, config)) == NULL) {
757			nvlist_free(config);
758			config = NULL;
759			continue;
760		}
761
762		nvlist_free(config);
763		config = nvl;
764
765		/*
766		 * Go through and update the paths for spares, now that we have
767		 * them.
768		 */
769		verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
770		    &nvroot) == 0);
771		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
772		    &spares, &nspares) == 0) {
773			for (i = 0; i < nspares; i++) {
774				if (fix_paths(spares[i], pl->names) != 0)
775					goto nomem;
776			}
777		}
778
779		/*
780		 * Update the paths for l2cache devices.
781		 */
782		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
783		    &l2cache, &nl2cache) == 0) {
784			for (i = 0; i < nl2cache; i++) {
785				if (fix_paths(l2cache[i], pl->names) != 0)
786					goto nomem;
787			}
788		}
789
790		/*
791		 * Restore the original information read from the actual label.
792		 */
793		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
794		    DATA_TYPE_UINT64);
795		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
796		    DATA_TYPE_STRING);
797		if (hostid != 0) {
798			verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
799			    hostid) == 0);
800			verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
801			    hostname) == 0);
802		}
803
804add_pool:
805		/*
806		 * Add this pool to the list of configs.
807		 */
808		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
809		    &name) == 0);
810		if (nvlist_add_nvlist(ret, name, config) != 0)
811			goto nomem;
812
813		found_one = B_TRUE;
814		nvlist_free(config);
815		config = NULL;
816	}
817
818	if (!found_one) {
819		nvlist_free(ret);
820		ret = NULL;
821	}
822
823	return (ret);
824
825nomem:
826	(void) no_memory(hdl);
827error:
828	nvlist_free(config);
829	nvlist_free(ret);
830	for (c = 0; c < children; c++)
831		nvlist_free(child[c]);
832	free(child);
833
834	return (NULL);
835}
836
837/*
838 * Return the offset of the given label.
839 */
840static uint64_t
841label_offset(uint64_t size, int l)
842{
843	ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
844	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
845	    0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
846}
847
848/*
849 * Given a file descriptor, read the label information and return an nvlist
850 * describing the configuration, if there is one.
851 * Return 0 on success, or -1 on failure
852 */
853int
854zpool_read_label(int fd, nvlist_t **config)
855{
856	struct stat64 statbuf;
857	int l;
858	vdev_label_t *label;
859	uint64_t state, txg, size;
860
861	*config = NULL;
862
863	if (fstat64(fd, &statbuf) == -1)
864		return (-1);
865	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
866
867	if ((label = malloc(sizeof (vdev_label_t))) == NULL)
868		return (-1);
869
870	for (l = 0; l < VDEV_LABELS; l++) {
871		if (pread64(fd, label, sizeof (vdev_label_t),
872		    label_offset(size, l)) != sizeof (vdev_label_t))
873			continue;
874
875		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
876		    sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
877			continue;
878
879		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
880		    &state) != 0 || state > POOL_STATE_L2CACHE) {
881			nvlist_free(*config);
882			continue;
883		}
884
885		if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
886		    (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
887		    &txg) != 0 || txg == 0)) {
888			nvlist_free(*config);
889			continue;
890		}
891
892		free(label);
893		return (0);
894	}
895
896	free(label);
897	*config = NULL;
898	errno = ENOENT;
899	return (-1);
900}
901
902typedef struct rdsk_node {
903	char *rn_name;
904	int rn_dfd;
905	libzfs_handle_t *rn_hdl;
906	nvlist_t *rn_config;
907	avl_tree_t *rn_avl;
908	avl_node_t rn_node;
909	boolean_t rn_nozpool;
910} rdsk_node_t;
911
912static int
913slice_cache_compare(const void *arg1, const void *arg2)
914{
915	const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
916	const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
917	char *nm1slice, *nm2slice;
918	int rv;
919
920	/*
921	 * slices zero and two are the most likely to provide results,
922	 * so put those first
923	 */
924	nm1slice = strstr(nm1, "s0");
925	nm2slice = strstr(nm2, "s0");
926	if (nm1slice && !nm2slice) {
927		return (-1);
928	}
929	if (!nm1slice && nm2slice) {
930		return (1);
931	}
932	nm1slice = strstr(nm1, "s2");
933	nm2slice = strstr(nm2, "s2");
934	if (nm1slice && !nm2slice) {
935		return (-1);
936	}
937	if (!nm1slice && nm2slice) {
938		return (1);
939	}
940
941	rv = strcmp(nm1, nm2);
942	if (rv == 0)
943		return (0);
944	return (rv > 0 ? 1 : -1);
945}
946
947static void
948check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
949    diskaddr_t size, uint_t blksz)
950{
951	rdsk_node_t tmpnode;
952	rdsk_node_t *node;
953	char sname[MAXNAMELEN];
954
955	tmpnode.rn_name = &sname[0];
956	(void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
957	    diskname, partno);
958	/*
959	 * protect against division by zero for disk labels that
960	 * contain a bogus sector size
961	 */
962	if (blksz == 0)
963		blksz = DEV_BSIZE;
964	/* too small to contain a zpool? */
965	if ((size < (SPA_MINDEVSIZE / blksz)) &&
966	    (node = avl_find(r, &tmpnode, NULL)))
967		node->rn_nozpool = B_TRUE;
968}
969
970static void
971nozpool_all_slices(avl_tree_t *r, const char *sname)
972{
973	char diskname[MAXNAMELEN];
974	char *ptr;
975	int i;
976
977	(void) strncpy(diskname, sname, MAXNAMELEN);
978	if (((ptr = strrchr(diskname, 's')) == NULL) &&
979	    ((ptr = strrchr(diskname, 'p')) == NULL))
980		return;
981	ptr[0] = 's';
982	ptr[1] = '\0';
983	for (i = 0; i < NDKMAP; i++)
984		check_one_slice(r, diskname, i, 0, 1);
985	ptr[0] = 'p';
986	for (i = 0; i <= FD_NUMPART; i++)
987		check_one_slice(r, diskname, i, 0, 1);
988}
989
990static void
991check_slices(avl_tree_t *r, int fd, const char *sname)
992{
993	struct extvtoc vtoc;
994	struct dk_gpt *gpt;
995	char diskname[MAXNAMELEN];
996	char *ptr;
997	int i;
998
999	(void) strncpy(diskname, sname, MAXNAMELEN);
1000	if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1001		return;
1002	ptr[1] = '\0';
1003
1004	if (read_extvtoc(fd, &vtoc) >= 0) {
1005		for (i = 0; i < NDKMAP; i++)
1006			check_one_slice(r, diskname, i,
1007			    vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1008	} else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1009		/*
1010		 * on x86 we'll still have leftover links that point
1011		 * to slices s[9-15], so use NDKMAP instead
1012		 */
1013		for (i = 0; i < NDKMAP; i++)
1014			check_one_slice(r, diskname, i,
1015			    gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1016		/* nodes p[1-4] are never used with EFI labels */
1017		ptr[0] = 'p';
1018		for (i = 1; i <= FD_NUMPART; i++)
1019			check_one_slice(r, diskname, i, 0, 1);
1020		efi_free(gpt);
1021	}
1022}
1023
1024static void
1025zpool_open_func(void *arg)
1026{
1027	rdsk_node_t *rn = arg;
1028	struct stat64 statbuf;
1029	nvlist_t *config;
1030	int fd;
1031
1032	if (rn->rn_nozpool)
1033		return;
1034	if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1035		/* symlink to a device that's no longer there */
1036		if (errno == ENOENT)
1037			nozpool_all_slices(rn->rn_avl, rn->rn_name);
1038		return;
1039	}
1040	/*
1041	 * Ignore failed stats.  We only want regular
1042	 * files, character devs and block devs.
1043	 */
1044	if (fstat64(fd, &statbuf) != 0 ||
1045	    (!S_ISREG(statbuf.st_mode) &&
1046	    !S_ISCHR(statbuf.st_mode) &&
1047	    !S_ISBLK(statbuf.st_mode))) {
1048		(void) close(fd);
1049		return;
1050	}
1051	/* this file is too small to hold a zpool */
1052	if (S_ISREG(statbuf.st_mode) &&
1053	    statbuf.st_size < SPA_MINDEVSIZE) {
1054		(void) close(fd);
1055		return;
1056	} else if (!S_ISREG(statbuf.st_mode)) {
1057		/*
1058		 * Try to read the disk label first so we don't have to
1059		 * open a bunch of minor nodes that can't have a zpool.
1060		 */
1061		check_slices(rn->rn_avl, fd, rn->rn_name);
1062	}
1063
1064	if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1065		(void) close(fd);
1066		(void) no_memory(rn->rn_hdl);
1067		return;
1068	}
1069	(void) close(fd);
1070
1071	rn->rn_config = config;
1072}
1073
1074/*
1075 * Given a file descriptor, clear (zero) the label information.
1076 */
1077int
1078zpool_clear_label(int fd)
1079{
1080	struct stat64 statbuf;
1081	int l;
1082	vdev_label_t *label;
1083	uint64_t size;
1084
1085	if (fstat64(fd, &statbuf) == -1)
1086		return (0);
1087	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1088
1089	if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1090		return (-1);
1091
1092	for (l = 0; l < VDEV_LABELS; l++) {
1093		if (pwrite64(fd, label, sizeof (vdev_label_t),
1094		    label_offset(size, l)) != sizeof (vdev_label_t)) {
1095			free(label);
1096			return (-1);
1097		}
1098	}
1099
1100	free(label);
1101	return (0);
1102}
1103
1104/*
1105 * Given a list of directories to search, find all pools stored on disk.  This
1106 * includes partial pools which are not available to import.  If no args are
1107 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1108 * poolname or guid (but not both) are provided by the caller when trying
1109 * to import a specific pool.
1110 */
1111static nvlist_t *
1112zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1113{
1114	int i, dirs = iarg->paths;
1115	struct dirent64 *dp;
1116	char path[MAXPATHLEN];
1117	char *end, **dir = iarg->path;
1118	size_t pathleft;
1119	nvlist_t *ret = NULL;
1120	static char *default_dir = ZFS_DISK_ROOT;
1121	pool_list_t pools = { 0 };
1122	pool_entry_t *pe, *penext;
1123	vdev_entry_t *ve, *venext;
1124	config_entry_t *ce, *cenext;
1125	name_entry_t *ne, *nenext;
1126	avl_tree_t slice_cache;
1127	rdsk_node_t *slice;
1128	void *cookie;
1129
1130	if (dirs == 0) {
1131		dirs = 1;
1132		dir = &default_dir;
1133	}
1134
1135	/*
1136	 * Go through and read the label configuration information from every
1137	 * possible device, organizing the information according to pool GUID
1138	 * and toplevel GUID.
1139	 */
1140	for (i = 0; i < dirs; i++) {
1141		tpool_t *t;
1142		char rdsk[MAXPATHLEN];
1143		int dfd;
1144		boolean_t config_failed = B_FALSE;
1145		DIR *dirp;
1146
1147		/* use realpath to normalize the path */
1148		if (realpath(dir[i], path) == 0) {
1149			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1150			    dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1151			goto error;
1152		}
1153		end = &path[strlen(path)];
1154		*end++ = '/';
1155		*end = 0;
1156		pathleft = &path[sizeof (path)] - end;
1157
1158		/*
1159		 * Using raw devices instead of block devices when we're
1160		 * reading the labels skips a bunch of slow operations during
1161		 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1162		 */
1163		if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1164			(void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1165		else
1166			(void) strlcpy(rdsk, path, sizeof (rdsk));
1167
1168		if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1169		    (dirp = fdopendir(dfd)) == NULL) {
1170			if (dfd >= 0)
1171				(void) close(dfd);
1172			zfs_error_aux(hdl, strerror(errno));
1173			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1174			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1175			    rdsk);
1176			goto error;
1177		}
1178
1179		avl_create(&slice_cache, slice_cache_compare,
1180		    sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1181		/*
1182		 * This is not MT-safe, but we have no MT consumers of libzfs
1183		 */
1184		while ((dp = readdir64(dirp)) != NULL) {
1185			const char *name = dp->d_name;
1186			if (name[0] == '.' &&
1187			    (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1188				continue;
1189
1190			slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1191			slice->rn_name = zfs_strdup(hdl, name);
1192			slice->rn_avl = &slice_cache;
1193			slice->rn_dfd = dfd;
1194			slice->rn_hdl = hdl;
1195			slice->rn_nozpool = B_FALSE;
1196			avl_add(&slice_cache, slice);
1197		}
1198		/*
1199		 * create a thread pool to do all of this in parallel;
1200		 * rn_nozpool is not protected, so this is racy in that
1201		 * multiple tasks could decide that the same slice can
1202		 * not hold a zpool, which is benign.  Also choose
1203		 * double the number of processors; we hold a lot of
1204		 * locks in the kernel, so going beyond this doesn't
1205		 * buy us much.
1206		 */
1207		t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1208		    0, NULL);
1209		for (slice = avl_first(&slice_cache); slice;
1210		    (slice = avl_walk(&slice_cache, slice,
1211		    AVL_AFTER)))
1212			(void) tpool_dispatch(t, zpool_open_func, slice);
1213		tpool_wait(t);
1214		tpool_destroy(t);
1215
1216		cookie = NULL;
1217		while ((slice = avl_destroy_nodes(&slice_cache,
1218		    &cookie)) != NULL) {
1219			if (slice->rn_config != NULL && !config_failed) {
1220				nvlist_t *config = slice->rn_config;
1221				boolean_t matched = B_TRUE;
1222
1223				if (iarg->poolname != NULL) {
1224					char *pname;
1225
1226					matched = nvlist_lookup_string(config,
1227					    ZPOOL_CONFIG_POOL_NAME,
1228					    &pname) == 0 &&
1229					    strcmp(iarg->poolname, pname) == 0;
1230				} else if (iarg->guid != 0) {
1231					uint64_t this_guid;
1232
1233					matched = nvlist_lookup_uint64(config,
1234					    ZPOOL_CONFIG_POOL_GUID,
1235					    &this_guid) == 0 &&
1236					    iarg->guid == this_guid;
1237				}
1238				if (!matched) {
1239					nvlist_free(config);
1240				} else {
1241					/*
1242					 * use the non-raw path for the config
1243					 */
1244					(void) strlcpy(end, slice->rn_name,
1245					    pathleft);
1246					if (add_config(hdl, &pools, path,
1247					    config) != 0)
1248						config_failed = B_TRUE;
1249				}
1250			}
1251			free(slice->rn_name);
1252			free(slice);
1253		}
1254		avl_destroy(&slice_cache);
1255
1256		(void) closedir(dirp);
1257
1258		if (config_failed)
1259			goto error;
1260	}
1261
1262	ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy);
1263
1264error:
1265	for (pe = pools.pools; pe != NULL; pe = penext) {
1266		penext = pe->pe_next;
1267		for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1268			venext = ve->ve_next;
1269			for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1270				cenext = ce->ce_next;
1271				nvlist_free(ce->ce_config);
1272				free(ce);
1273			}
1274			free(ve);
1275		}
1276		free(pe);
1277	}
1278
1279	for (ne = pools.names; ne != NULL; ne = nenext) {
1280		nenext = ne->ne_next;
1281		free(ne->ne_name);
1282		free(ne);
1283	}
1284
1285	return (ret);
1286}
1287
1288nvlist_t *
1289zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1290{
1291	importargs_t iarg = { 0 };
1292
1293	iarg.paths = argc;
1294	iarg.path = argv;
1295
1296	return (zpool_find_import_impl(hdl, &iarg));
1297}
1298
1299/*
1300 * Given a cache file, return the contents as a list of importable pools.
1301 * poolname or guid (but not both) are provided by the caller when trying
1302 * to import a specific pool.
1303 */
1304nvlist_t *
1305zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1306    char *poolname, uint64_t guid)
1307{
1308	char *buf;
1309	int fd;
1310	struct stat64 statbuf;
1311	nvlist_t *raw, *src, *dst;
1312	nvlist_t *pools;
1313	nvpair_t *elem;
1314	char *name;
1315	uint64_t this_guid;
1316	boolean_t active;
1317
1318	verify(poolname == NULL || guid == 0);
1319
1320	if ((fd = open(cachefile, O_RDONLY)) < 0) {
1321		zfs_error_aux(hdl, "%s", strerror(errno));
1322		(void) zfs_error(hdl, EZFS_BADCACHE,
1323		    dgettext(TEXT_DOMAIN, "failed to open cache file"));
1324		return (NULL);
1325	}
1326
1327	if (fstat64(fd, &statbuf) != 0) {
1328		zfs_error_aux(hdl, "%s", strerror(errno));
1329		(void) close(fd);
1330		(void) zfs_error(hdl, EZFS_BADCACHE,
1331		    dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1332		return (NULL);
1333	}
1334
1335	if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1336		(void) close(fd);
1337		return (NULL);
1338	}
1339
1340	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1341		(void) close(fd);
1342		free(buf);
1343		(void) zfs_error(hdl, EZFS_BADCACHE,
1344		    dgettext(TEXT_DOMAIN,
1345		    "failed to read cache file contents"));
1346		return (NULL);
1347	}
1348
1349	(void) close(fd);
1350
1351	if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1352		free(buf);
1353		(void) zfs_error(hdl, EZFS_BADCACHE,
1354		    dgettext(TEXT_DOMAIN,
1355		    "invalid or corrupt cache file contents"));
1356		return (NULL);
1357	}
1358
1359	free(buf);
1360
1361	/*
1362	 * Go through and get the current state of the pools and refresh their
1363	 * state.
1364	 */
1365	if (nvlist_alloc(&pools, 0, 0) != 0) {
1366		(void) no_memory(hdl);
1367		nvlist_free(raw);
1368		return (NULL);
1369	}
1370
1371	elem = NULL;
1372	while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1373		src = fnvpair_value_nvlist(elem);
1374
1375		name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1376		if (poolname != NULL && strcmp(poolname, name) != 0)
1377			continue;
1378
1379		this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1380		if (guid != 0 && guid != this_guid)
1381			continue;
1382
1383		if (pool_active(hdl, name, this_guid, &active) != 0) {
1384			nvlist_free(raw);
1385			nvlist_free(pools);
1386			return (NULL);
1387		}
1388
1389		if (active)
1390			continue;
1391
1392		if (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE,
1393		    cachefile) != 0) {
1394			(void) no_memory(hdl);
1395			nvlist_free(raw);
1396			nvlist_free(pools);
1397			return (NULL);
1398		}
1399
1400		if ((dst = refresh_config(hdl, src)) == NULL) {
1401			nvlist_free(raw);
1402			nvlist_free(pools);
1403			return (NULL);
1404		}
1405
1406		if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1407			(void) no_memory(hdl);
1408			nvlist_free(dst);
1409			nvlist_free(raw);
1410			nvlist_free(pools);
1411			return (NULL);
1412		}
1413		nvlist_free(dst);
1414	}
1415
1416	nvlist_free(raw);
1417	return (pools);
1418}
1419
1420static int
1421name_or_guid_exists(zpool_handle_t *zhp, void *data)
1422{
1423	importargs_t *import = data;
1424	int found = 0;
1425
1426	if (import->poolname != NULL) {
1427		char *pool_name;
1428
1429		verify(nvlist_lookup_string(zhp->zpool_config,
1430		    ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1431		if (strcmp(pool_name, import->poolname) == 0)
1432			found = 1;
1433	} else {
1434		uint64_t pool_guid;
1435
1436		verify(nvlist_lookup_uint64(zhp->zpool_config,
1437		    ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1438		if (pool_guid == import->guid)
1439			found = 1;
1440	}
1441
1442	zpool_close(zhp);
1443	return (found);
1444}
1445
1446nvlist_t *
1447zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1448{
1449	verify(import->poolname == NULL || import->guid == 0);
1450
1451	if (import->unique)
1452		import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1453
1454	if (import->cachefile != NULL)
1455		return (zpool_find_import_cached(hdl, import->cachefile,
1456		    import->poolname, import->guid));
1457
1458	return (zpool_find_import_impl(hdl, import));
1459}
1460
1461boolean_t
1462find_guid(nvlist_t *nv, uint64_t guid)
1463{
1464	uint64_t tmp;
1465	nvlist_t **child;
1466	uint_t c, children;
1467
1468	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1469	if (tmp == guid)
1470		return (B_TRUE);
1471
1472	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1473	    &child, &children) == 0) {
1474		for (c = 0; c < children; c++)
1475			if (find_guid(child[c], guid))
1476				return (B_TRUE);
1477	}
1478
1479	return (B_FALSE);
1480}
1481
1482typedef struct aux_cbdata {
1483	const char	*cb_type;
1484	uint64_t	cb_guid;
1485	zpool_handle_t	*cb_zhp;
1486} aux_cbdata_t;
1487
1488static int
1489find_aux(zpool_handle_t *zhp, void *data)
1490{
1491	aux_cbdata_t *cbp = data;
1492	nvlist_t **list;
1493	uint_t i, count;
1494	uint64_t guid;
1495	nvlist_t *nvroot;
1496
1497	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1498	    &nvroot) == 0);
1499
1500	if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1501	    &list, &count) == 0) {
1502		for (i = 0; i < count; i++) {
1503			verify(nvlist_lookup_uint64(list[i],
1504			    ZPOOL_CONFIG_GUID, &guid) == 0);
1505			if (guid == cbp->cb_guid) {
1506				cbp->cb_zhp = zhp;
1507				return (1);
1508			}
1509		}
1510	}
1511
1512	zpool_close(zhp);
1513	return (0);
1514}
1515
1516/*
1517 * Determines if the pool is in use.  If so, it returns true and the state of
1518 * the pool as well as the name of the pool.  Both strings are allocated and
1519 * must be freed by the caller.
1520 */
1521int
1522zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1523    boolean_t *inuse)
1524{
1525	nvlist_t *config;
1526	char *name;
1527	boolean_t ret;
1528	uint64_t guid, vdev_guid;
1529	zpool_handle_t *zhp;
1530	nvlist_t *pool_config;
1531	uint64_t stateval, isspare;
1532	aux_cbdata_t cb = { 0 };
1533	boolean_t isactive;
1534
1535	*inuse = B_FALSE;
1536
1537	if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1538		(void) no_memory(hdl);
1539		return (-1);
1540	}
1541
1542	if (config == NULL)
1543		return (0);
1544
1545	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1546	    &stateval) == 0);
1547	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1548	    &vdev_guid) == 0);
1549
1550	if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1551		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1552		    &name) == 0);
1553		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1554		    &guid) == 0);
1555	}
1556
1557	switch (stateval) {
1558	case POOL_STATE_EXPORTED:
1559		/*
1560		 * A pool with an exported state may in fact be imported
1561		 * read-only, so check the in-core state to see if it's
1562		 * active and imported read-only.  If it is, set
1563		 * its state to active.
1564		 */
1565		if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1566		    (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1567			if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1568				stateval = POOL_STATE_ACTIVE;
1569
1570			/*
1571			 * All we needed the zpool handle for is the
1572			 * readonly prop check.
1573			 */
1574			zpool_close(zhp);
1575		}
1576
1577		ret = B_TRUE;
1578		break;
1579
1580	case POOL_STATE_ACTIVE:
1581		/*
1582		 * For an active pool, we have to determine if it's really part
1583		 * of a currently active pool (in which case the pool will exist
1584		 * and the guid will be the same), or whether it's part of an
1585		 * active pool that was disconnected without being explicitly
1586		 * exported.
1587		 */
1588		if (pool_active(hdl, name, guid, &isactive) != 0) {
1589			nvlist_free(config);
1590			return (-1);
1591		}
1592
1593		if (isactive) {
1594			/*
1595			 * Because the device may have been removed while
1596			 * offlined, we only report it as active if the vdev is
1597			 * still present in the config.  Otherwise, pretend like
1598			 * it's not in use.
1599			 */
1600			if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1601			    (pool_config = zpool_get_config(zhp, NULL))
1602			    != NULL) {
1603				nvlist_t *nvroot;
1604
1605				verify(nvlist_lookup_nvlist(pool_config,
1606				    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1607				ret = find_guid(nvroot, vdev_guid);
1608			} else {
1609				ret = B_FALSE;
1610			}
1611
1612			/*
1613			 * If this is an active spare within another pool, we
1614			 * treat it like an unused hot spare.  This allows the
1615			 * user to create a pool with a hot spare that currently
1616			 * in use within another pool.  Since we return B_TRUE,
1617			 * libdiskmgt will continue to prevent generic consumers
1618			 * from using the device.
1619			 */
1620			if (ret && nvlist_lookup_uint64(config,
1621			    ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1622				stateval = POOL_STATE_SPARE;
1623
1624			if (zhp != NULL)
1625				zpool_close(zhp);
1626		} else {
1627			stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1628			ret = B_TRUE;
1629		}
1630		break;
1631
1632	case POOL_STATE_SPARE:
1633		/*
1634		 * For a hot spare, it can be either definitively in use, or
1635		 * potentially active.  To determine if it's in use, we iterate
1636		 * over all pools in the system and search for one with a spare
1637		 * with a matching guid.
1638		 *
1639		 * Due to the shared nature of spares, we don't actually report
1640		 * the potentially active case as in use.  This means the user
1641		 * can freely create pools on the hot spares of exported pools,
1642		 * but to do otherwise makes the resulting code complicated, and
1643		 * we end up having to deal with this case anyway.
1644		 */
1645		cb.cb_zhp = NULL;
1646		cb.cb_guid = vdev_guid;
1647		cb.cb_type = ZPOOL_CONFIG_SPARES;
1648		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1649			name = (char *)zpool_get_name(cb.cb_zhp);
1650			ret = B_TRUE;
1651		} else {
1652			ret = B_FALSE;
1653		}
1654		break;
1655
1656	case POOL_STATE_L2CACHE:
1657
1658		/*
1659		 * Check if any pool is currently using this l2cache device.
1660		 */
1661		cb.cb_zhp = NULL;
1662		cb.cb_guid = vdev_guid;
1663		cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1664		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1665			name = (char *)zpool_get_name(cb.cb_zhp);
1666			ret = B_TRUE;
1667		} else {
1668			ret = B_FALSE;
1669		}
1670		break;
1671
1672	default:
1673		ret = B_FALSE;
1674	}
1675
1676
1677	if (ret) {
1678		if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1679			if (cb.cb_zhp)
1680				zpool_close(cb.cb_zhp);
1681			nvlist_free(config);
1682			return (-1);
1683		}
1684		*state = (pool_state_t)stateval;
1685	}
1686
1687	if (cb.cb_zhp)
1688		zpool_close(cb.cb_zhp);
1689
1690	nvlist_free(config);
1691	*inuse = ret;
1692	return (0);
1693}
1694