libzfs_import.c revision c861bfbd77c4ae780a0341e9cb6926d8b74341cf
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, 2016 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{
417	pool_entry_t *pe;
418	vdev_entry_t *ve;
419	config_entry_t *ce;
420	nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
421	nvlist_t **spares, **l2cache;
422	uint_t i, nspares, nl2cache;
423	boolean_t config_seen;
424	uint64_t best_txg;
425	char *name, *hostname = NULL;
426	uint64_t guid;
427	uint_t children = 0;
428	nvlist_t **child = NULL;
429	uint_t holes;
430	uint64_t *hole_array, max_id;
431	uint_t c;
432	boolean_t isactive;
433	uint64_t hostid;
434	nvlist_t *nvl;
435	boolean_t found_one = B_FALSE;
436	boolean_t valid_top_config = B_FALSE;
437
438	if (nvlist_alloc(&ret, 0, 0) != 0)
439		goto nomem;
440
441	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
442		uint64_t id, max_txg = 0;
443
444		if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
445			goto nomem;
446		config_seen = B_FALSE;
447
448		/*
449		 * Iterate over all toplevel vdevs.  Grab the pool configuration
450		 * from the first one we find, and then go through the rest and
451		 * add them as necessary to the 'vdevs' member of the config.
452		 */
453		for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
454
455			/*
456			 * Determine the best configuration for this vdev by
457			 * selecting the config with the latest transaction
458			 * group.
459			 */
460			best_txg = 0;
461			for (ce = ve->ve_configs; ce != NULL;
462			    ce = ce->ce_next) {
463
464				if (ce->ce_txg > best_txg) {
465					tmp = ce->ce_config;
466					best_txg = ce->ce_txg;
467				}
468			}
469
470			/*
471			 * We rely on the fact that the max txg for the
472			 * pool will contain the most up-to-date information
473			 * about the valid top-levels in the vdev namespace.
474			 */
475			if (best_txg > max_txg) {
476				(void) nvlist_remove(config,
477				    ZPOOL_CONFIG_VDEV_CHILDREN,
478				    DATA_TYPE_UINT64);
479				(void) nvlist_remove(config,
480				    ZPOOL_CONFIG_HOLE_ARRAY,
481				    DATA_TYPE_UINT64_ARRAY);
482
483				max_txg = best_txg;
484				hole_array = NULL;
485				holes = 0;
486				max_id = 0;
487				valid_top_config = B_FALSE;
488
489				if (nvlist_lookup_uint64(tmp,
490				    ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
491					verify(nvlist_add_uint64(config,
492					    ZPOOL_CONFIG_VDEV_CHILDREN,
493					    max_id) == 0);
494					valid_top_config = B_TRUE;
495				}
496
497				if (nvlist_lookup_uint64_array(tmp,
498				    ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
499				    &holes) == 0) {
500					verify(nvlist_add_uint64_array(config,
501					    ZPOOL_CONFIG_HOLE_ARRAY,
502					    hole_array, holes) == 0);
503				}
504			}
505
506			if (!config_seen) {
507				/*
508				 * Copy the relevant pieces of data to the pool
509				 * configuration:
510				 *
511				 *	version
512				 *	pool guid
513				 *	name
514				 *	comment (if available)
515				 *	pool state
516				 *	hostid (if available)
517				 *	hostname (if available)
518				 */
519				uint64_t state, version;
520				char *comment = NULL;
521
522				version = fnvlist_lookup_uint64(tmp,
523				    ZPOOL_CONFIG_VERSION);
524				fnvlist_add_uint64(config,
525				    ZPOOL_CONFIG_VERSION, version);
526				guid = fnvlist_lookup_uint64(tmp,
527				    ZPOOL_CONFIG_POOL_GUID);
528				fnvlist_add_uint64(config,
529				    ZPOOL_CONFIG_POOL_GUID, guid);
530				name = fnvlist_lookup_string(tmp,
531				    ZPOOL_CONFIG_POOL_NAME);
532				fnvlist_add_string(config,
533				    ZPOOL_CONFIG_POOL_NAME, name);
534
535				if (nvlist_lookup_string(tmp,
536				    ZPOOL_CONFIG_COMMENT, &comment) == 0)
537					fnvlist_add_string(config,
538					    ZPOOL_CONFIG_COMMENT, comment);
539
540				state = fnvlist_lookup_uint64(tmp,
541				    ZPOOL_CONFIG_POOL_STATE);
542				fnvlist_add_uint64(config,
543				    ZPOOL_CONFIG_POOL_STATE, state);
544
545				hostid = 0;
546				if (nvlist_lookup_uint64(tmp,
547				    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
548					fnvlist_add_uint64(config,
549					    ZPOOL_CONFIG_HOSTID, hostid);
550					hostname = fnvlist_lookup_string(tmp,
551					    ZPOOL_CONFIG_HOSTNAME);
552					fnvlist_add_string(config,
553					    ZPOOL_CONFIG_HOSTNAME, hostname);
554				}
555
556				config_seen = B_TRUE;
557			}
558
559			/*
560			 * Add this top-level vdev to the child array.
561			 */
562			verify(nvlist_lookup_nvlist(tmp,
563			    ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
564			verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
565			    &id) == 0);
566
567			if (id >= children) {
568				nvlist_t **newchild;
569
570				newchild = zfs_alloc(hdl, (id + 1) *
571				    sizeof (nvlist_t *));
572				if (newchild == NULL)
573					goto nomem;
574
575				for (c = 0; c < children; c++)
576					newchild[c] = child[c];
577
578				free(child);
579				child = newchild;
580				children = id + 1;
581			}
582			if (nvlist_dup(nvtop, &child[id], 0) != 0)
583				goto nomem;
584
585		}
586
587		/*
588		 * If we have information about all the top-levels then
589		 * clean up the nvlist which we've constructed. This
590		 * means removing any extraneous devices that are
591		 * beyond the valid range or adding devices to the end
592		 * of our array which appear to be missing.
593		 */
594		if (valid_top_config) {
595			if (max_id < children) {
596				for (c = max_id; c < children; c++)
597					nvlist_free(child[c]);
598				children = max_id;
599			} else if (max_id > children) {
600				nvlist_t **newchild;
601
602				newchild = zfs_alloc(hdl, (max_id) *
603				    sizeof (nvlist_t *));
604				if (newchild == NULL)
605					goto nomem;
606
607				for (c = 0; c < children; c++)
608					newchild[c] = child[c];
609
610				free(child);
611				child = newchild;
612				children = max_id;
613			}
614		}
615
616		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
617		    &guid) == 0);
618
619		/*
620		 * The vdev namespace may contain holes as a result of
621		 * device removal. We must add them back into the vdev
622		 * tree before we process any missing devices.
623		 */
624		if (holes > 0) {
625			ASSERT(valid_top_config);
626
627			for (c = 0; c < children; c++) {
628				nvlist_t *holey;
629
630				if (child[c] != NULL ||
631				    !vdev_is_hole(hole_array, holes, c))
632					continue;
633
634				if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
635				    0) != 0)
636					goto nomem;
637
638				/*
639				 * Holes in the namespace are treated as
640				 * "hole" top-level vdevs and have a
641				 * special flag set on them.
642				 */
643				if (nvlist_add_string(holey,
644				    ZPOOL_CONFIG_TYPE,
645				    VDEV_TYPE_HOLE) != 0 ||
646				    nvlist_add_uint64(holey,
647				    ZPOOL_CONFIG_ID, c) != 0 ||
648				    nvlist_add_uint64(holey,
649				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
650					nvlist_free(holey);
651					goto nomem;
652				}
653				child[c] = holey;
654			}
655		}
656
657		/*
658		 * Look for any missing top-level vdevs.  If this is the case,
659		 * create a faked up 'missing' vdev as a placeholder.  We cannot
660		 * simply compress the child array, because the kernel performs
661		 * certain checks to make sure the vdev IDs match their location
662		 * in the configuration.
663		 */
664		for (c = 0; c < children; c++) {
665			if (child[c] == NULL) {
666				nvlist_t *missing;
667				if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
668				    0) != 0)
669					goto nomem;
670				if (nvlist_add_string(missing,
671				    ZPOOL_CONFIG_TYPE,
672				    VDEV_TYPE_MISSING) != 0 ||
673				    nvlist_add_uint64(missing,
674				    ZPOOL_CONFIG_ID, c) != 0 ||
675				    nvlist_add_uint64(missing,
676				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
677					nvlist_free(missing);
678					goto nomem;
679				}
680				child[c] = missing;
681			}
682		}
683
684		/*
685		 * Put all of this pool's top-level vdevs into a root vdev.
686		 */
687		if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
688			goto nomem;
689		if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
690		    VDEV_TYPE_ROOT) != 0 ||
691		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
692		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
693		    nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
694		    child, children) != 0) {
695			nvlist_free(nvroot);
696			goto nomem;
697		}
698
699		for (c = 0; c < children; c++)
700			nvlist_free(child[c]);
701		free(child);
702		children = 0;
703		child = NULL;
704
705		/*
706		 * Go through and fix up any paths and/or devids based on our
707		 * known list of vdev GUID -> path mappings.
708		 */
709		if (fix_paths(nvroot, pl->names) != 0) {
710			nvlist_free(nvroot);
711			goto nomem;
712		}
713
714		/*
715		 * Add the root vdev to this pool's configuration.
716		 */
717		if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
718		    nvroot) != 0) {
719			nvlist_free(nvroot);
720			goto nomem;
721		}
722		nvlist_free(nvroot);
723
724		/*
725		 * zdb uses this path to report on active pools that were
726		 * imported or created using -R.
727		 */
728		if (active_ok)
729			goto add_pool;
730
731		/*
732		 * Determine if this pool is currently active, in which case we
733		 * can't actually import it.
734		 */
735		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
736		    &name) == 0);
737		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
738		    &guid) == 0);
739
740		if (pool_active(hdl, name, guid, &isactive) != 0)
741			goto error;
742
743		if (isactive) {
744			nvlist_free(config);
745			config = NULL;
746			continue;
747		}
748
749		if ((nvl = refresh_config(hdl, config)) == NULL) {
750			nvlist_free(config);
751			config = NULL;
752			continue;
753		}
754
755		nvlist_free(config);
756		config = nvl;
757
758		/*
759		 * Go through and update the paths for spares, now that we have
760		 * them.
761		 */
762		verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
763		    &nvroot) == 0);
764		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
765		    &spares, &nspares) == 0) {
766			for (i = 0; i < nspares; i++) {
767				if (fix_paths(spares[i], pl->names) != 0)
768					goto nomem;
769			}
770		}
771
772		/*
773		 * Update the paths for l2cache devices.
774		 */
775		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
776		    &l2cache, &nl2cache) == 0) {
777			for (i = 0; i < nl2cache; i++) {
778				if (fix_paths(l2cache[i], pl->names) != 0)
779					goto nomem;
780			}
781		}
782
783		/*
784		 * Restore the original information read from the actual label.
785		 */
786		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
787		    DATA_TYPE_UINT64);
788		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
789		    DATA_TYPE_STRING);
790		if (hostid != 0) {
791			verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
792			    hostid) == 0);
793			verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
794			    hostname) == 0);
795		}
796
797add_pool:
798		/*
799		 * Add this pool to the list of configs.
800		 */
801		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
802		    &name) == 0);
803		if (nvlist_add_nvlist(ret, name, config) != 0)
804			goto nomem;
805
806		found_one = B_TRUE;
807		nvlist_free(config);
808		config = NULL;
809	}
810
811	if (!found_one) {
812		nvlist_free(ret);
813		ret = NULL;
814	}
815
816	return (ret);
817
818nomem:
819	(void) no_memory(hdl);
820error:
821	nvlist_free(config);
822	nvlist_free(ret);
823	for (c = 0; c < children; c++)
824		nvlist_free(child[c]);
825	free(child);
826
827	return (NULL);
828}
829
830/*
831 * Return the offset of the given label.
832 */
833static uint64_t
834label_offset(uint64_t size, int l)
835{
836	ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
837	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
838	    0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
839}
840
841/*
842 * Given a file descriptor, read the label information and return an nvlist
843 * describing the configuration, if there is one.
844 * Return 0 on success, or -1 on failure
845 */
846int
847zpool_read_label(int fd, nvlist_t **config)
848{
849	struct stat64 statbuf;
850	int l;
851	vdev_label_t *label;
852	uint64_t state, txg, size;
853
854	*config = NULL;
855
856	if (fstat64(fd, &statbuf) == -1)
857		return (-1);
858	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
859
860	if ((label = malloc(sizeof (vdev_label_t))) == NULL)
861		return (-1);
862
863	for (l = 0; l < VDEV_LABELS; l++) {
864		if (pread64(fd, label, sizeof (vdev_label_t),
865		    label_offset(size, l)) != sizeof (vdev_label_t))
866			continue;
867
868		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
869		    sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
870			continue;
871
872		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
873		    &state) != 0 || state > POOL_STATE_L2CACHE) {
874			nvlist_free(*config);
875			continue;
876		}
877
878		if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
879		    (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
880		    &txg) != 0 || txg == 0)) {
881			nvlist_free(*config);
882			continue;
883		}
884
885		free(label);
886		return (0);
887	}
888
889	free(label);
890	*config = NULL;
891	return (-1);
892}
893
894typedef struct rdsk_node {
895	char *rn_name;
896	int rn_dfd;
897	libzfs_handle_t *rn_hdl;
898	nvlist_t *rn_config;
899	avl_tree_t *rn_avl;
900	avl_node_t rn_node;
901	boolean_t rn_nozpool;
902} rdsk_node_t;
903
904static int
905slice_cache_compare(const void *arg1, const void *arg2)
906{
907	const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
908	const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
909	char *nm1slice, *nm2slice;
910	int rv;
911
912	/*
913	 * slices zero and two are the most likely to provide results,
914	 * so put those first
915	 */
916	nm1slice = strstr(nm1, "s0");
917	nm2slice = strstr(nm2, "s0");
918	if (nm1slice && !nm2slice) {
919		return (-1);
920	}
921	if (!nm1slice && nm2slice) {
922		return (1);
923	}
924	nm1slice = strstr(nm1, "s2");
925	nm2slice = strstr(nm2, "s2");
926	if (nm1slice && !nm2slice) {
927		return (-1);
928	}
929	if (!nm1slice && nm2slice) {
930		return (1);
931	}
932
933	rv = strcmp(nm1, nm2);
934	if (rv == 0)
935		return (0);
936	return (rv > 0 ? 1 : -1);
937}
938
939static void
940check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
941    diskaddr_t size, uint_t blksz)
942{
943	rdsk_node_t tmpnode;
944	rdsk_node_t *node;
945	char sname[MAXNAMELEN];
946
947	tmpnode.rn_name = &sname[0];
948	(void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
949	    diskname, partno);
950	/*
951	 * protect against division by zero for disk labels that
952	 * contain a bogus sector size
953	 */
954	if (blksz == 0)
955		blksz = DEV_BSIZE;
956	/* too small to contain a zpool? */
957	if ((size < (SPA_MINDEVSIZE / blksz)) &&
958	    (node = avl_find(r, &tmpnode, NULL)))
959		node->rn_nozpool = B_TRUE;
960}
961
962static void
963nozpool_all_slices(avl_tree_t *r, const char *sname)
964{
965	char diskname[MAXNAMELEN];
966	char *ptr;
967	int i;
968
969	(void) strncpy(diskname, sname, MAXNAMELEN);
970	if (((ptr = strrchr(diskname, 's')) == NULL) &&
971	    ((ptr = strrchr(diskname, 'p')) == NULL))
972		return;
973	ptr[0] = 's';
974	ptr[1] = '\0';
975	for (i = 0; i < NDKMAP; i++)
976		check_one_slice(r, diskname, i, 0, 1);
977	ptr[0] = 'p';
978	for (i = 0; i <= FD_NUMPART; i++)
979		check_one_slice(r, diskname, i, 0, 1);
980}
981
982static void
983check_slices(avl_tree_t *r, int fd, const char *sname)
984{
985	struct extvtoc vtoc;
986	struct dk_gpt *gpt;
987	char diskname[MAXNAMELEN];
988	char *ptr;
989	int i;
990
991	(void) strncpy(diskname, sname, MAXNAMELEN);
992	if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
993		return;
994	ptr[1] = '\0';
995
996	if (read_extvtoc(fd, &vtoc) >= 0) {
997		for (i = 0; i < NDKMAP; i++)
998			check_one_slice(r, diskname, i,
999			    vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1000	} else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1001		/*
1002		 * on x86 we'll still have leftover links that point
1003		 * to slices s[9-15], so use NDKMAP instead
1004		 */
1005		for (i = 0; i < NDKMAP; i++)
1006			check_one_slice(r, diskname, i,
1007			    gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1008		/* nodes p[1-4] are never used with EFI labels */
1009		ptr[0] = 'p';
1010		for (i = 1; i <= FD_NUMPART; i++)
1011			check_one_slice(r, diskname, i, 0, 1);
1012		efi_free(gpt);
1013	}
1014}
1015
1016static void
1017zpool_open_func(void *arg)
1018{
1019	rdsk_node_t *rn = arg;
1020	struct stat64 statbuf;
1021	nvlist_t *config;
1022	int fd;
1023
1024	if (rn->rn_nozpool)
1025		return;
1026	if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1027		/* symlink to a device that's no longer there */
1028		if (errno == ENOENT)
1029			nozpool_all_slices(rn->rn_avl, rn->rn_name);
1030		return;
1031	}
1032	/*
1033	 * Ignore failed stats.  We only want regular
1034	 * files, character devs and block devs.
1035	 */
1036	if (fstat64(fd, &statbuf) != 0 ||
1037	    (!S_ISREG(statbuf.st_mode) &&
1038	    !S_ISCHR(statbuf.st_mode) &&
1039	    !S_ISBLK(statbuf.st_mode))) {
1040		(void) close(fd);
1041		return;
1042	}
1043	/* this file is too small to hold a zpool */
1044	if (S_ISREG(statbuf.st_mode) &&
1045	    statbuf.st_size < SPA_MINDEVSIZE) {
1046		(void) close(fd);
1047		return;
1048	} else if (!S_ISREG(statbuf.st_mode)) {
1049		/*
1050		 * Try to read the disk label first so we don't have to
1051		 * open a bunch of minor nodes that can't have a zpool.
1052		 */
1053		check_slices(rn->rn_avl, fd, rn->rn_name);
1054	}
1055
1056	if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1057		(void) close(fd);
1058		(void) no_memory(rn->rn_hdl);
1059		return;
1060	}
1061	(void) close(fd);
1062
1063	rn->rn_config = config;
1064}
1065
1066/*
1067 * Given a file descriptor, clear (zero) the label information.
1068 */
1069int
1070zpool_clear_label(int fd)
1071{
1072	struct stat64 statbuf;
1073	int l;
1074	vdev_label_t *label;
1075	uint64_t size;
1076
1077	if (fstat64(fd, &statbuf) == -1)
1078		return (0);
1079	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1080
1081	if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1082		return (-1);
1083
1084	for (l = 0; l < VDEV_LABELS; l++) {
1085		if (pwrite64(fd, label, sizeof (vdev_label_t),
1086		    label_offset(size, l)) != sizeof (vdev_label_t)) {
1087			free(label);
1088			return (-1);
1089		}
1090	}
1091
1092	free(label);
1093	return (0);
1094}
1095
1096/*
1097 * Given a list of directories to search, find all pools stored on disk.  This
1098 * includes partial pools which are not available to import.  If no args are
1099 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1100 * poolname or guid (but not both) are provided by the caller when trying
1101 * to import a specific pool.
1102 */
1103static nvlist_t *
1104zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1105{
1106	int i, dirs = iarg->paths;
1107	struct dirent64 *dp;
1108	char path[MAXPATHLEN];
1109	char *end, **dir = iarg->path;
1110	size_t pathleft;
1111	nvlist_t *ret = NULL;
1112	static char *default_dir = ZFS_DISK_ROOT;
1113	pool_list_t pools = { 0 };
1114	pool_entry_t *pe, *penext;
1115	vdev_entry_t *ve, *venext;
1116	config_entry_t *ce, *cenext;
1117	name_entry_t *ne, *nenext;
1118	avl_tree_t slice_cache;
1119	rdsk_node_t *slice;
1120	void *cookie;
1121
1122	if (dirs == 0) {
1123		dirs = 1;
1124		dir = &default_dir;
1125	}
1126
1127	/*
1128	 * Go through and read the label configuration information from every
1129	 * possible device, organizing the information according to pool GUID
1130	 * and toplevel GUID.
1131	 */
1132	for (i = 0; i < dirs; i++) {
1133		tpool_t *t;
1134		char rdsk[MAXPATHLEN];
1135		int dfd;
1136		boolean_t config_failed = B_FALSE;
1137		DIR *dirp;
1138
1139		/* use realpath to normalize the path */
1140		if (realpath(dir[i], path) == 0) {
1141			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1142			    dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1143			goto error;
1144		}
1145		end = &path[strlen(path)];
1146		*end++ = '/';
1147		*end = 0;
1148		pathleft = &path[sizeof (path)] - end;
1149
1150		/*
1151		 * Using raw devices instead of block devices when we're
1152		 * reading the labels skips a bunch of slow operations during
1153		 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1154		 */
1155		if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1156			(void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1157		else
1158			(void) strlcpy(rdsk, path, sizeof (rdsk));
1159
1160		if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1161		    (dirp = fdopendir(dfd)) == NULL) {
1162			if (dfd >= 0)
1163				(void) close(dfd);
1164			zfs_error_aux(hdl, strerror(errno));
1165			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1166			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1167			    rdsk);
1168			goto error;
1169		}
1170
1171		avl_create(&slice_cache, slice_cache_compare,
1172		    sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1173		/*
1174		 * This is not MT-safe, but we have no MT consumers of libzfs
1175		 */
1176		while ((dp = readdir64(dirp)) != NULL) {
1177			const char *name = dp->d_name;
1178			if (name[0] == '.' &&
1179			    (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1180				continue;
1181
1182			slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1183			slice->rn_name = zfs_strdup(hdl, name);
1184			slice->rn_avl = &slice_cache;
1185			slice->rn_dfd = dfd;
1186			slice->rn_hdl = hdl;
1187			slice->rn_nozpool = B_FALSE;
1188			avl_add(&slice_cache, slice);
1189		}
1190		/*
1191		 * create a thread pool to do all of this in parallel;
1192		 * rn_nozpool is not protected, so this is racy in that
1193		 * multiple tasks could decide that the same slice can
1194		 * not hold a zpool, which is benign.  Also choose
1195		 * double the number of processors; we hold a lot of
1196		 * locks in the kernel, so going beyond this doesn't
1197		 * buy us much.
1198		 */
1199		t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1200		    0, NULL);
1201		for (slice = avl_first(&slice_cache); slice;
1202		    (slice = avl_walk(&slice_cache, slice,
1203		    AVL_AFTER)))
1204			(void) tpool_dispatch(t, zpool_open_func, slice);
1205		tpool_wait(t);
1206		tpool_destroy(t);
1207
1208		cookie = NULL;
1209		while ((slice = avl_destroy_nodes(&slice_cache,
1210		    &cookie)) != NULL) {
1211			if (slice->rn_config != NULL && !config_failed) {
1212				nvlist_t *config = slice->rn_config;
1213				boolean_t matched = B_TRUE;
1214
1215				if (iarg->poolname != NULL) {
1216					char *pname;
1217
1218					matched = nvlist_lookup_string(config,
1219					    ZPOOL_CONFIG_POOL_NAME,
1220					    &pname) == 0 &&
1221					    strcmp(iarg->poolname, pname) == 0;
1222				} else if (iarg->guid != 0) {
1223					uint64_t this_guid;
1224
1225					matched = nvlist_lookup_uint64(config,
1226					    ZPOOL_CONFIG_POOL_GUID,
1227					    &this_guid) == 0 &&
1228					    iarg->guid == this_guid;
1229				}
1230				if (!matched) {
1231					nvlist_free(config);
1232				} else {
1233					/*
1234					 * use the non-raw path for the config
1235					 */
1236					(void) strlcpy(end, slice->rn_name,
1237					    pathleft);
1238					if (add_config(hdl, &pools, path,
1239					    config) != 0)
1240						config_failed = B_TRUE;
1241				}
1242			}
1243			free(slice->rn_name);
1244			free(slice);
1245		}
1246		avl_destroy(&slice_cache);
1247
1248		(void) closedir(dirp);
1249
1250		if (config_failed)
1251			goto error;
1252	}
1253
1254	ret = get_configs(hdl, &pools, iarg->can_be_active);
1255
1256error:
1257	for (pe = pools.pools; pe != NULL; pe = penext) {
1258		penext = pe->pe_next;
1259		for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1260			venext = ve->ve_next;
1261			for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1262				cenext = ce->ce_next;
1263				nvlist_free(ce->ce_config);
1264				free(ce);
1265			}
1266			free(ve);
1267		}
1268		free(pe);
1269	}
1270
1271	for (ne = pools.names; ne != NULL; ne = nenext) {
1272		nenext = ne->ne_next;
1273		free(ne->ne_name);
1274		free(ne);
1275	}
1276
1277	return (ret);
1278}
1279
1280nvlist_t *
1281zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1282{
1283	importargs_t iarg = { 0 };
1284
1285	iarg.paths = argc;
1286	iarg.path = argv;
1287
1288	return (zpool_find_import_impl(hdl, &iarg));
1289}
1290
1291/*
1292 * Given a cache file, return the contents as a list of importable pools.
1293 * poolname or guid (but not both) are provided by the caller when trying
1294 * to import a specific pool.
1295 */
1296nvlist_t *
1297zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1298    char *poolname, uint64_t guid)
1299{
1300	char *buf;
1301	int fd;
1302	struct stat64 statbuf;
1303	nvlist_t *raw, *src, *dst;
1304	nvlist_t *pools;
1305	nvpair_t *elem;
1306	char *name;
1307	uint64_t this_guid;
1308	boolean_t active;
1309
1310	verify(poolname == NULL || guid == 0);
1311
1312	if ((fd = open(cachefile, O_RDONLY)) < 0) {
1313		zfs_error_aux(hdl, "%s", strerror(errno));
1314		(void) zfs_error(hdl, EZFS_BADCACHE,
1315		    dgettext(TEXT_DOMAIN, "failed to open cache file"));
1316		return (NULL);
1317	}
1318
1319	if (fstat64(fd, &statbuf) != 0) {
1320		zfs_error_aux(hdl, "%s", strerror(errno));
1321		(void) close(fd);
1322		(void) zfs_error(hdl, EZFS_BADCACHE,
1323		    dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1324		return (NULL);
1325	}
1326
1327	if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1328		(void) close(fd);
1329		return (NULL);
1330	}
1331
1332	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1333		(void) close(fd);
1334		free(buf);
1335		(void) zfs_error(hdl, EZFS_BADCACHE,
1336		    dgettext(TEXT_DOMAIN,
1337		    "failed to read cache file contents"));
1338		return (NULL);
1339	}
1340
1341	(void) close(fd);
1342
1343	if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1344		free(buf);
1345		(void) zfs_error(hdl, EZFS_BADCACHE,
1346		    dgettext(TEXT_DOMAIN,
1347		    "invalid or corrupt cache file contents"));
1348		return (NULL);
1349	}
1350
1351	free(buf);
1352
1353	/*
1354	 * Go through and get the current state of the pools and refresh their
1355	 * state.
1356	 */
1357	if (nvlist_alloc(&pools, 0, 0) != 0) {
1358		(void) no_memory(hdl);
1359		nvlist_free(raw);
1360		return (NULL);
1361	}
1362
1363	elem = NULL;
1364	while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1365		src = fnvpair_value_nvlist(elem);
1366
1367		name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1368		if (poolname != NULL && strcmp(poolname, name) != 0)
1369			continue;
1370
1371		this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1372		if (guid != 0 && guid != this_guid)
1373			continue;
1374
1375		if (pool_active(hdl, name, this_guid, &active) != 0) {
1376			nvlist_free(raw);
1377			nvlist_free(pools);
1378			return (NULL);
1379		}
1380
1381		if (active)
1382			continue;
1383
1384		if ((dst = refresh_config(hdl, src)) == NULL) {
1385			nvlist_free(raw);
1386			nvlist_free(pools);
1387			return (NULL);
1388		}
1389
1390		if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1391			(void) no_memory(hdl);
1392			nvlist_free(dst);
1393			nvlist_free(raw);
1394			nvlist_free(pools);
1395			return (NULL);
1396		}
1397		nvlist_free(dst);
1398	}
1399
1400	nvlist_free(raw);
1401	return (pools);
1402}
1403
1404static int
1405name_or_guid_exists(zpool_handle_t *zhp, void *data)
1406{
1407	importargs_t *import = data;
1408	int found = 0;
1409
1410	if (import->poolname != NULL) {
1411		char *pool_name;
1412
1413		verify(nvlist_lookup_string(zhp->zpool_config,
1414		    ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1415		if (strcmp(pool_name, import->poolname) == 0)
1416			found = 1;
1417	} else {
1418		uint64_t pool_guid;
1419
1420		verify(nvlist_lookup_uint64(zhp->zpool_config,
1421		    ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1422		if (pool_guid == import->guid)
1423			found = 1;
1424	}
1425
1426	zpool_close(zhp);
1427	return (found);
1428}
1429
1430nvlist_t *
1431zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1432{
1433	verify(import->poolname == NULL || import->guid == 0);
1434
1435	if (import->unique)
1436		import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1437
1438	if (import->cachefile != NULL)
1439		return (zpool_find_import_cached(hdl, import->cachefile,
1440		    import->poolname, import->guid));
1441
1442	return (zpool_find_import_impl(hdl, import));
1443}
1444
1445boolean_t
1446find_guid(nvlist_t *nv, uint64_t guid)
1447{
1448	uint64_t tmp;
1449	nvlist_t **child;
1450	uint_t c, children;
1451
1452	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1453	if (tmp == guid)
1454		return (B_TRUE);
1455
1456	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1457	    &child, &children) == 0) {
1458		for (c = 0; c < children; c++)
1459			if (find_guid(child[c], guid))
1460				return (B_TRUE);
1461	}
1462
1463	return (B_FALSE);
1464}
1465
1466typedef struct aux_cbdata {
1467	const char	*cb_type;
1468	uint64_t	cb_guid;
1469	zpool_handle_t	*cb_zhp;
1470} aux_cbdata_t;
1471
1472static int
1473find_aux(zpool_handle_t *zhp, void *data)
1474{
1475	aux_cbdata_t *cbp = data;
1476	nvlist_t **list;
1477	uint_t i, count;
1478	uint64_t guid;
1479	nvlist_t *nvroot;
1480
1481	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1482	    &nvroot) == 0);
1483
1484	if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1485	    &list, &count) == 0) {
1486		for (i = 0; i < count; i++) {
1487			verify(nvlist_lookup_uint64(list[i],
1488			    ZPOOL_CONFIG_GUID, &guid) == 0);
1489			if (guid == cbp->cb_guid) {
1490				cbp->cb_zhp = zhp;
1491				return (1);
1492			}
1493		}
1494	}
1495
1496	zpool_close(zhp);
1497	return (0);
1498}
1499
1500/*
1501 * Determines if the pool is in use.  If so, it returns true and the state of
1502 * the pool as well as the name of the pool.  Both strings are allocated and
1503 * must be freed by the caller.
1504 */
1505int
1506zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1507    boolean_t *inuse)
1508{
1509	nvlist_t *config;
1510	char *name;
1511	boolean_t ret;
1512	uint64_t guid, vdev_guid;
1513	zpool_handle_t *zhp;
1514	nvlist_t *pool_config;
1515	uint64_t stateval, isspare;
1516	aux_cbdata_t cb = { 0 };
1517	boolean_t isactive;
1518
1519	*inuse = B_FALSE;
1520
1521	if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1522		(void) no_memory(hdl);
1523		return (-1);
1524	}
1525
1526	if (config == NULL)
1527		return (0);
1528
1529	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1530	    &stateval) == 0);
1531	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1532	    &vdev_guid) == 0);
1533
1534	if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1535		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1536		    &name) == 0);
1537		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1538		    &guid) == 0);
1539	}
1540
1541	switch (stateval) {
1542	case POOL_STATE_EXPORTED:
1543		/*
1544		 * A pool with an exported state may in fact be imported
1545		 * read-only, so check the in-core state to see if it's
1546		 * active and imported read-only.  If it is, set
1547		 * its state to active.
1548		 */
1549		if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1550		    (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1551			if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1552				stateval = POOL_STATE_ACTIVE;
1553
1554			/*
1555			 * All we needed the zpool handle for is the
1556			 * readonly prop check.
1557			 */
1558			zpool_close(zhp);
1559		}
1560
1561		ret = B_TRUE;
1562		break;
1563
1564	case POOL_STATE_ACTIVE:
1565		/*
1566		 * For an active pool, we have to determine if it's really part
1567		 * of a currently active pool (in which case the pool will exist
1568		 * and the guid will be the same), or whether it's part of an
1569		 * active pool that was disconnected without being explicitly
1570		 * exported.
1571		 */
1572		if (pool_active(hdl, name, guid, &isactive) != 0) {
1573			nvlist_free(config);
1574			return (-1);
1575		}
1576
1577		if (isactive) {
1578			/*
1579			 * Because the device may have been removed while
1580			 * offlined, we only report it as active if the vdev is
1581			 * still present in the config.  Otherwise, pretend like
1582			 * it's not in use.
1583			 */
1584			if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1585			    (pool_config = zpool_get_config(zhp, NULL))
1586			    != NULL) {
1587				nvlist_t *nvroot;
1588
1589				verify(nvlist_lookup_nvlist(pool_config,
1590				    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1591				ret = find_guid(nvroot, vdev_guid);
1592			} else {
1593				ret = B_FALSE;
1594			}
1595
1596			/*
1597			 * If this is an active spare within another pool, we
1598			 * treat it like an unused hot spare.  This allows the
1599			 * user to create a pool with a hot spare that currently
1600			 * in use within another pool.  Since we return B_TRUE,
1601			 * libdiskmgt will continue to prevent generic consumers
1602			 * from using the device.
1603			 */
1604			if (ret && nvlist_lookup_uint64(config,
1605			    ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1606				stateval = POOL_STATE_SPARE;
1607
1608			if (zhp != NULL)
1609				zpool_close(zhp);
1610		} else {
1611			stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1612			ret = B_TRUE;
1613		}
1614		break;
1615
1616	case POOL_STATE_SPARE:
1617		/*
1618		 * For a hot spare, it can be either definitively in use, or
1619		 * potentially active.  To determine if it's in use, we iterate
1620		 * over all pools in the system and search for one with a spare
1621		 * with a matching guid.
1622		 *
1623		 * Due to the shared nature of spares, we don't actually report
1624		 * the potentially active case as in use.  This means the user
1625		 * can freely create pools on the hot spares of exported pools,
1626		 * but to do otherwise makes the resulting code complicated, and
1627		 * we end up having to deal with this case anyway.
1628		 */
1629		cb.cb_zhp = NULL;
1630		cb.cb_guid = vdev_guid;
1631		cb.cb_type = ZPOOL_CONFIG_SPARES;
1632		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1633			name = (char *)zpool_get_name(cb.cb_zhp);
1634			ret = B_TRUE;
1635		} else {
1636			ret = B_FALSE;
1637		}
1638		break;
1639
1640	case POOL_STATE_L2CACHE:
1641
1642		/*
1643		 * Check if any pool is currently using this l2cache device.
1644		 */
1645		cb.cb_zhp = NULL;
1646		cb.cb_guid = vdev_guid;
1647		cb.cb_type = ZPOOL_CONFIG_L2CACHE;
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	default:
1657		ret = B_FALSE;
1658	}
1659
1660
1661	if (ret) {
1662		if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1663			if (cb.cb_zhp)
1664				zpool_close(cb.cb_zhp);
1665			nvlist_free(config);
1666			return (-1);
1667		}
1668		*state = (pool_state_t)stateval;
1669	}
1670
1671	if (cb.cb_zhp)
1672		zpool_close(cb.cb_zhp);
1673
1674	nvlist_free(config);
1675	*inuse = ret;
1676	return (0);
1677}
1678