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		return (0);
221	}
222
223	/*
224	 * If we have a valid config but cannot read any of these fields, then
225	 * it means we have a half-initialized label.  In vdev_label_init()
226	 * we write a label with txg == 0 so that we can identify the device
227	 * in case the user refers to the same disk later on.  If we fail to
228	 * create the pool, we'll be left with a label in this state
229	 * which should not be considered part of a valid pool.
230	 */
231	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
232	    &pool_guid) != 0 ||
233	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
234	    &vdev_guid) != 0 ||
235	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
236	    &top_guid) != 0 ||
237	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
238	    &txg) != 0 || txg == 0) {
239		return (0);
240	}
241
242	/*
243	 * First, see if we know about this pool.  If not, then add it to the
244	 * list of known pools.
245	 */
246	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
247		if (pe->pe_guid == pool_guid)
248			break;
249	}
250
251	if (pe == NULL) {
252		if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
253			return (-1);
254		}
255		pe->pe_guid = pool_guid;
256		pe->pe_next = pl->pools;
257		pl->pools = pe;
258	}
259
260	/*
261	 * Second, see if we know about this toplevel vdev.  Add it if its
262	 * missing.
263	 */
264	for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
265		if (ve->ve_guid == top_guid)
266			break;
267	}
268
269	if (ve == NULL) {
270		if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
271			return (-1);
272		}
273		ve->ve_guid = top_guid;
274		ve->ve_next = pe->pe_vdevs;
275		pe->pe_vdevs = ve;
276	}
277
278	/*
279	 * Third, see if we have a config with a matching transaction group.  If
280	 * so, then we do nothing.  Otherwise, add it to the list of known
281	 * configs.
282	 */
283	for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
284		if (ce->ce_txg == txg)
285			break;
286	}
287
288	if (ce == NULL) {
289		if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
290			return (-1);
291		}
292		ce->ce_txg = txg;
293		ce->ce_config = fnvlist_dup(config);
294		ce->ce_next = ve->ve_configs;
295		ve->ve_configs = ce;
296	}
297
298	/*
299	 * At this point we've successfully added our config to the list of
300	 * known configs.  The last thing to do is add the vdev guid -> path
301	 * mappings so that we can fix up the configuration as necessary before
302	 * doing the import.
303	 */
304	if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
305		return (-1);
306
307	if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
308		free(ne);
309		return (-1);
310	}
311
312	ne->ne_guid = vdev_guid;
313	ne->ne_next = pl->names;
314	pl->names = ne;
315
316	return (0);
317}
318
319/*
320 * Returns true if the named pool matches the given GUID.
321 */
322static int
323pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
324    boolean_t *isactive)
325{
326	zpool_handle_t *zhp;
327	uint64_t theguid;
328
329	if (zpool_open_silent(hdl, name, &zhp) != 0)
330		return (-1);
331
332	if (zhp == NULL) {
333		*isactive = B_FALSE;
334		return (0);
335	}
336
337	verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
338	    &theguid) == 0);
339
340	zpool_close(zhp);
341
342	*isactive = (theguid == guid);
343	return (0);
344}
345
346static nvlist_t *
347refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
348{
349	nvlist_t *nvl;
350	zfs_cmd_t zc = { 0 };
351	int err, dstbuf_size;
352
353	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
354		return (NULL);
355
356	dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
357
358	if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
359		zcmd_free_nvlists(&zc);
360		return (NULL);
361	}
362
363	while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
364	    &zc)) != 0 && errno == ENOMEM) {
365		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
366			zcmd_free_nvlists(&zc);
367			return (NULL);
368		}
369	}
370
371	if (err) {
372		zcmd_free_nvlists(&zc);
373		return (NULL);
374	}
375
376	if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
377		zcmd_free_nvlists(&zc);
378		return (NULL);
379	}
380
381	zcmd_free_nvlists(&zc);
382	return (nvl);
383}
384
385/*
386 * Determine if the vdev id is a hole in the namespace.
387 */
388boolean_t
389vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
390{
391	for (int c = 0; c < holes; c++) {
392
393		/* Top-level is a hole */
394		if (hole_array[c] == id)
395			return (B_TRUE);
396	}
397	return (B_FALSE);
398}
399
400/*
401 * Convert our list of pools into the definitive set of configurations.  We
402 * start by picking the best config for each toplevel vdev.  Once that's done,
403 * we assemble the toplevel vdevs into a full config for the pool.  We make a
404 * pass to fix up any incorrect paths, and then add it to the main list to
405 * return to the user.
406 */
407static nvlist_t *
408get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok,
409    nvlist_t *policy)
410{
411	pool_entry_t *pe;
412	vdev_entry_t *ve;
413	config_entry_t *ce;
414	nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
415	nvlist_t **spares, **l2cache;
416	uint_t i, nspares, nl2cache;
417	boolean_t config_seen;
418	uint64_t best_txg;
419	char *name, *hostname = NULL;
420	uint64_t guid;
421	uint_t children = 0;
422	nvlist_t **child = NULL;
423	uint_t holes;
424	uint64_t *hole_array, max_id;
425	uint_t c;
426	boolean_t isactive;
427	uint64_t hostid;
428	nvlist_t *nvl;
429	boolean_t found_one = B_FALSE;
430	boolean_t valid_top_config = B_FALSE;
431
432	if (nvlist_alloc(&ret, 0, 0) != 0)
433		goto nomem;
434
435	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
436		uint64_t id, max_txg = 0;
437
438		if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
439			goto nomem;
440		config_seen = B_FALSE;
441
442		/*
443		 * Iterate over all toplevel vdevs.  Grab the pool configuration
444		 * from the first one we find, and then go through the rest and
445		 * add them as necessary to the 'vdevs' member of the config.
446		 */
447		for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
448
449			/*
450			 * Determine the best configuration for this vdev by
451			 * selecting the config with the latest transaction
452			 * group.
453			 */
454			best_txg = 0;
455			for (ce = ve->ve_configs; ce != NULL;
456			    ce = ce->ce_next) {
457
458				if (ce->ce_txg > best_txg) {
459					tmp = ce->ce_config;
460					best_txg = ce->ce_txg;
461				}
462			}
463
464			/*
465			 * We rely on the fact that the max txg for the
466			 * pool will contain the most up-to-date information
467			 * about the valid top-levels in the vdev namespace.
468			 */
469			if (best_txg > max_txg) {
470				(void) nvlist_remove(config,
471				    ZPOOL_CONFIG_VDEV_CHILDREN,
472				    DATA_TYPE_UINT64);
473				(void) nvlist_remove(config,
474				    ZPOOL_CONFIG_HOLE_ARRAY,
475				    DATA_TYPE_UINT64_ARRAY);
476
477				max_txg = best_txg;
478				hole_array = NULL;
479				holes = 0;
480				max_id = 0;
481				valid_top_config = B_FALSE;
482
483				if (nvlist_lookup_uint64(tmp,
484				    ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
485					verify(nvlist_add_uint64(config,
486					    ZPOOL_CONFIG_VDEV_CHILDREN,
487					    max_id) == 0);
488					valid_top_config = B_TRUE;
489				}
490
491				if (nvlist_lookup_uint64_array(tmp,
492				    ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
493				    &holes) == 0) {
494					verify(nvlist_add_uint64_array(config,
495					    ZPOOL_CONFIG_HOLE_ARRAY,
496					    hole_array, holes) == 0);
497				}
498			}
499
500			if (!config_seen) {
501				/*
502				 * Copy the relevant pieces of data to the pool
503				 * configuration:
504				 *
505				 *	version
506				 *	pool guid
507				 *	name
508				 *	comment (if available)
509				 *	pool state
510				 *	hostid (if available)
511				 *	hostname (if available)
512				 */
513				uint64_t state, version;
514				char *comment = NULL;
515
516				version = fnvlist_lookup_uint64(tmp,
517				    ZPOOL_CONFIG_VERSION);
518				fnvlist_add_uint64(config,
519				    ZPOOL_CONFIG_VERSION, version);
520				guid = fnvlist_lookup_uint64(tmp,
521				    ZPOOL_CONFIG_POOL_GUID);
522				fnvlist_add_uint64(config,
523				    ZPOOL_CONFIG_POOL_GUID, guid);
524				name = fnvlist_lookup_string(tmp,
525				    ZPOOL_CONFIG_POOL_NAME);
526				fnvlist_add_string(config,
527				    ZPOOL_CONFIG_POOL_NAME, name);
528
529				if (nvlist_lookup_string(tmp,
530				    ZPOOL_CONFIG_COMMENT, &comment) == 0)
531					fnvlist_add_string(config,
532					    ZPOOL_CONFIG_COMMENT, comment);
533
534				state = fnvlist_lookup_uint64(tmp,
535				    ZPOOL_CONFIG_POOL_STATE);
536				fnvlist_add_uint64(config,
537				    ZPOOL_CONFIG_POOL_STATE, state);
538
539				hostid = 0;
540				if (nvlist_lookup_uint64(tmp,
541				    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
542					fnvlist_add_uint64(config,
543					    ZPOOL_CONFIG_HOSTID, hostid);
544					hostname = fnvlist_lookup_string(tmp,
545					    ZPOOL_CONFIG_HOSTNAME);
546					fnvlist_add_string(config,
547					    ZPOOL_CONFIG_HOSTNAME, hostname);
548				}
549
550				config_seen = B_TRUE;
551			}
552
553			/*
554			 * Add this top-level vdev to the child array.
555			 */
556			verify(nvlist_lookup_nvlist(tmp,
557			    ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
558			verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
559			    &id) == 0);
560
561			if (id >= children) {
562				nvlist_t **newchild;
563
564				newchild = zfs_alloc(hdl, (id + 1) *
565				    sizeof (nvlist_t *));
566				if (newchild == NULL)
567					goto nomem;
568
569				for (c = 0; c < children; c++)
570					newchild[c] = child[c];
571
572				free(child);
573				child = newchild;
574				children = id + 1;
575			}
576			if (nvlist_dup(nvtop, &child[id], 0) != 0)
577				goto nomem;
578
579		}
580
581		/*
582		 * If we have information about all the top-levels then
583		 * clean up the nvlist which we've constructed. This
584		 * means removing any extraneous devices that are
585		 * beyond the valid range or adding devices to the end
586		 * of our array which appear to be missing.
587		 */
588		if (valid_top_config) {
589			if (max_id < children) {
590				for (c = max_id; c < children; c++)
591					nvlist_free(child[c]);
592				children = max_id;
593			} else if (max_id > children) {
594				nvlist_t **newchild;
595
596				newchild = zfs_alloc(hdl, (max_id) *
597				    sizeof (nvlist_t *));
598				if (newchild == NULL)
599					goto nomem;
600
601				for (c = 0; c < children; c++)
602					newchild[c] = child[c];
603
604				free(child);
605				child = newchild;
606				children = max_id;
607			}
608		}
609
610		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
611		    &guid) == 0);
612
613		/*
614		 * The vdev namespace may contain holes as a result of
615		 * device removal. We must add them back into the vdev
616		 * tree before we process any missing devices.
617		 */
618		if (holes > 0) {
619			ASSERT(valid_top_config);
620
621			for (c = 0; c < children; c++) {
622				nvlist_t *holey;
623
624				if (child[c] != NULL ||
625				    !vdev_is_hole(hole_array, holes, c))
626					continue;
627
628				if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
629				    0) != 0)
630					goto nomem;
631
632				/*
633				 * Holes in the namespace are treated as
634				 * "hole" top-level vdevs and have a
635				 * special flag set on them.
636				 */
637				if (nvlist_add_string(holey,
638				    ZPOOL_CONFIG_TYPE,
639				    VDEV_TYPE_HOLE) != 0 ||
640				    nvlist_add_uint64(holey,
641				    ZPOOL_CONFIG_ID, c) != 0 ||
642				    nvlist_add_uint64(holey,
643				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
644					nvlist_free(holey);
645					goto nomem;
646				}
647				child[c] = holey;
648			}
649		}
650
651		/*
652		 * Look for any missing top-level vdevs.  If this is the case,
653		 * create a faked up 'missing' vdev as a placeholder.  We cannot
654		 * simply compress the child array, because the kernel performs
655		 * certain checks to make sure the vdev IDs match their location
656		 * in the configuration.
657		 */
658		for (c = 0; c < children; c++) {
659			if (child[c] == NULL) {
660				nvlist_t *missing;
661				if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
662				    0) != 0)
663					goto nomem;
664				if (nvlist_add_string(missing,
665				    ZPOOL_CONFIG_TYPE,
666				    VDEV_TYPE_MISSING) != 0 ||
667				    nvlist_add_uint64(missing,
668				    ZPOOL_CONFIG_ID, c) != 0 ||
669				    nvlist_add_uint64(missing,
670				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
671					nvlist_free(missing);
672					goto nomem;
673				}
674				child[c] = missing;
675			}
676		}
677
678		/*
679		 * Put all of this pool's top-level vdevs into a root vdev.
680		 */
681		if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
682			goto nomem;
683		if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
684		    VDEV_TYPE_ROOT) != 0 ||
685		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
686		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
687		    nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
688		    child, children) != 0) {
689			nvlist_free(nvroot);
690			goto nomem;
691		}
692
693		for (c = 0; c < children; c++)
694			nvlist_free(child[c]);
695		free(child);
696		children = 0;
697		child = NULL;
698
699		/*
700		 * Go through and fix up any paths and/or devids based on our
701		 * known list of vdev GUID -> path mappings.
702		 */
703		if (fix_paths(nvroot, pl->names) != 0) {
704			nvlist_free(nvroot);
705			goto nomem;
706		}
707
708		/*
709		 * Add the root vdev to this pool's configuration.
710		 */
711		if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
712		    nvroot) != 0) {
713			nvlist_free(nvroot);
714			goto nomem;
715		}
716		nvlist_free(nvroot);
717
718		/*
719		 * zdb uses this path to report on active pools that were
720		 * imported or created using -R.
721		 */
722		if (active_ok)
723			goto add_pool;
724
725		/*
726		 * Determine if this pool is currently active, in which case we
727		 * can't actually import it.
728		 */
729		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
730		    &name) == 0);
731		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
732		    &guid) == 0);
733
734		if (pool_active(hdl, name, guid, &isactive) != 0)
735			goto error;
736
737		if (isactive) {
738			nvlist_free(config);
739			config = NULL;
740			continue;
741		}
742
743		if (policy != NULL) {
744			if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY,
745			    policy) != 0)
746				goto nomem;
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	errno = ENOENT;
892	return (-1);
893}
894
895typedef struct rdsk_node {
896	char *rn_name;
897	int rn_dfd;
898	libzfs_handle_t *rn_hdl;
899	nvlist_t *rn_config;
900	avl_tree_t *rn_avl;
901	avl_node_t rn_node;
902	boolean_t rn_nozpool;
903} rdsk_node_t;
904
905static int
906slice_cache_compare(const void *arg1, const void *arg2)
907{
908	const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
909	const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
910	char *nm1slice, *nm2slice;
911	int rv;
912
913	/*
914	 * slices zero and two are the most likely to provide results,
915	 * so put those first
916	 */
917	nm1slice = strstr(nm1, "s0");
918	nm2slice = strstr(nm2, "s0");
919	if (nm1slice && !nm2slice) {
920		return (-1);
921	}
922	if (!nm1slice && nm2slice) {
923		return (1);
924	}
925	nm1slice = strstr(nm1, "s2");
926	nm2slice = strstr(nm2, "s2");
927	if (nm1slice && !nm2slice) {
928		return (-1);
929	}
930	if (!nm1slice && nm2slice) {
931		return (1);
932	}
933
934	rv = strcmp(nm1, nm2);
935	if (rv == 0)
936		return (0);
937	return (rv > 0 ? 1 : -1);
938}
939
940static void
941check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
942    diskaddr_t size, uint_t blksz)
943{
944	rdsk_node_t tmpnode;
945	rdsk_node_t *node;
946	char sname[MAXNAMELEN];
947
948	tmpnode.rn_name = &sname[0];
949	(void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
950	    diskname, partno);
951	/*
952	 * protect against division by zero for disk labels that
953	 * contain a bogus sector size
954	 */
955	if (blksz == 0)
956		blksz = DEV_BSIZE;
957	/* too small to contain a zpool? */
958	if ((size < (SPA_MINDEVSIZE / blksz)) &&
959	    (node = avl_find(r, &tmpnode, NULL)))
960		node->rn_nozpool = B_TRUE;
961}
962
963static void
964nozpool_all_slices(avl_tree_t *r, const char *sname)
965{
966	char diskname[MAXNAMELEN];
967	char *ptr;
968	int i;
969
970	(void) strncpy(diskname, sname, MAXNAMELEN);
971	if (((ptr = strrchr(diskname, 's')) == NULL) &&
972	    ((ptr = strrchr(diskname, 'p')) == NULL))
973		return;
974	ptr[0] = 's';
975	ptr[1] = '\0';
976	for (i = 0; i < NDKMAP; i++)
977		check_one_slice(r, diskname, i, 0, 1);
978	ptr[0] = 'p';
979	for (i = 0; i <= FD_NUMPART; i++)
980		check_one_slice(r, diskname, i, 0, 1);
981}
982
983static void
984check_slices(avl_tree_t *r, int fd, const char *sname)
985{
986	struct extvtoc vtoc;
987	struct dk_gpt *gpt;
988	char diskname[MAXNAMELEN];
989	char *ptr;
990	int i;
991
992	(void) strncpy(diskname, sname, MAXNAMELEN);
993	if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
994		return;
995	ptr[1] = '\0';
996
997	if (read_extvtoc(fd, &vtoc) >= 0) {
998		for (i = 0; i < NDKMAP; i++)
999			check_one_slice(r, diskname, i,
1000			    vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1001	} else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1002		/*
1003		 * on x86 we'll still have leftover links that point
1004		 * to slices s[9-15], so use NDKMAP instead
1005		 */
1006		for (i = 0; i < NDKMAP; i++)
1007			check_one_slice(r, diskname, i,
1008			    gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1009		/* nodes p[1-4] are never used with EFI labels */
1010		ptr[0] = 'p';
1011		for (i = 1; i <= FD_NUMPART; i++)
1012			check_one_slice(r, diskname, i, 0, 1);
1013		efi_free(gpt);
1014	}
1015}
1016
1017static void
1018zpool_open_func(void *arg)
1019{
1020	rdsk_node_t *rn = arg;
1021	struct stat64 statbuf;
1022	nvlist_t *config;
1023	int fd;
1024
1025	if (rn->rn_nozpool)
1026		return;
1027	if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1028		/* symlink to a device that's no longer there */
1029		if (errno == ENOENT)
1030			nozpool_all_slices(rn->rn_avl, rn->rn_name);
1031		return;
1032	}
1033	/*
1034	 * Ignore failed stats.  We only want regular
1035	 * files, character devs and block devs.
1036	 */
1037	if (fstat64(fd, &statbuf) != 0 ||
1038	    (!S_ISREG(statbuf.st_mode) &&
1039	    !S_ISCHR(statbuf.st_mode) &&
1040	    !S_ISBLK(statbuf.st_mode))) {
1041		(void) close(fd);
1042		return;
1043	}
1044	/* this file is too small to hold a zpool */
1045	if (S_ISREG(statbuf.st_mode) &&
1046	    statbuf.st_size < SPA_MINDEVSIZE) {
1047		(void) close(fd);
1048		return;
1049	} else if (!S_ISREG(statbuf.st_mode)) {
1050		/*
1051		 * Try to read the disk label first so we don't have to
1052		 * open a bunch of minor nodes that can't have a zpool.
1053		 */
1054		check_slices(rn->rn_avl, fd, rn->rn_name);
1055	}
1056
1057	if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1058		(void) close(fd);
1059		(void) no_memory(rn->rn_hdl);
1060		return;
1061	}
1062	(void) close(fd);
1063
1064	rn->rn_config = config;
1065}
1066
1067/*
1068 * Given a file descriptor, clear (zero) the label information.
1069 */
1070int
1071zpool_clear_label(int fd)
1072{
1073	struct stat64 statbuf;
1074	int l;
1075	vdev_label_t *label;
1076	uint64_t size;
1077
1078	if (fstat64(fd, &statbuf) == -1)
1079		return (0);
1080	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1081
1082	if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1083		return (-1);
1084
1085	for (l = 0; l < VDEV_LABELS; l++) {
1086		if (pwrite64(fd, label, sizeof (vdev_label_t),
1087		    label_offset(size, l)) != sizeof (vdev_label_t)) {
1088			free(label);
1089			return (-1);
1090		}
1091	}
1092
1093	free(label);
1094	return (0);
1095}
1096
1097/*
1098 * Given a list of directories to search, find all pools stored on disk.  This
1099 * includes partial pools which are not available to import.  If no args are
1100 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1101 * poolname or guid (but not both) are provided by the caller when trying
1102 * to import a specific pool.
1103 */
1104static nvlist_t *
1105zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1106{
1107	int i, dirs = iarg->paths;
1108	struct dirent64 *dp;
1109	char path[MAXPATHLEN];
1110	char *end, **dir = iarg->path;
1111	size_t pathleft;
1112	nvlist_t *ret = NULL;
1113	static char *default_dir = ZFS_DISK_ROOT;
1114	pool_list_t pools = { 0 };
1115	pool_entry_t *pe, *penext;
1116	vdev_entry_t *ve, *venext;
1117	config_entry_t *ce, *cenext;
1118	name_entry_t *ne, *nenext;
1119	avl_tree_t slice_cache;
1120	rdsk_node_t *slice;
1121	void *cookie;
1122
1123	if (dirs == 0) {
1124		dirs = 1;
1125		dir = &default_dir;
1126	}
1127
1128	/*
1129	 * Go through and read the label configuration information from every
1130	 * possible device, organizing the information according to pool GUID
1131	 * and toplevel GUID.
1132	 */
1133	for (i = 0; i < dirs; i++) {
1134		tpool_t *t;
1135		char rdsk[MAXPATHLEN];
1136		int dfd;
1137		boolean_t config_failed = B_FALSE;
1138		DIR *dirp;
1139
1140		/* use realpath to normalize the path */
1141		if (realpath(dir[i], path) == 0) {
1142			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1143			    dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1144			goto error;
1145		}
1146		end = &path[strlen(path)];
1147		*end++ = '/';
1148		*end = 0;
1149		pathleft = &path[sizeof (path)] - end;
1150
1151		/*
1152		 * Using raw devices instead of block devices when we're
1153		 * reading the labels skips a bunch of slow operations during
1154		 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1155		 */
1156		if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1157			(void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1158		else
1159			(void) strlcpy(rdsk, path, sizeof (rdsk));
1160
1161		if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1162		    (dirp = fdopendir(dfd)) == NULL) {
1163			if (dfd >= 0)
1164				(void) close(dfd);
1165			zfs_error_aux(hdl, strerror(errno));
1166			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1167			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1168			    rdsk);
1169			goto error;
1170		}
1171
1172		avl_create(&slice_cache, slice_cache_compare,
1173		    sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1174		/*
1175		 * This is not MT-safe, but we have no MT consumers of libzfs
1176		 */
1177		while ((dp = readdir64(dirp)) != NULL) {
1178			const char *name = dp->d_name;
1179			if (name[0] == '.' &&
1180			    (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1181				continue;
1182
1183			slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1184			slice->rn_name = zfs_strdup(hdl, name);
1185			slice->rn_avl = &slice_cache;
1186			slice->rn_dfd = dfd;
1187			slice->rn_hdl = hdl;
1188			slice->rn_nozpool = B_FALSE;
1189			avl_add(&slice_cache, slice);
1190		}
1191		/*
1192		 * create a thread pool to do all of this in parallel;
1193		 * rn_nozpool is not protected, so this is racy in that
1194		 * multiple tasks could decide that the same slice can
1195		 * not hold a zpool, which is benign.  Also choose
1196		 * double the number of processors; we hold a lot of
1197		 * locks in the kernel, so going beyond this doesn't
1198		 * buy us much.
1199		 */
1200		t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1201		    0, NULL);
1202		for (slice = avl_first(&slice_cache); slice;
1203		    (slice = avl_walk(&slice_cache, slice,
1204		    AVL_AFTER)))
1205			(void) tpool_dispatch(t, zpool_open_func, slice);
1206		tpool_wait(t);
1207		tpool_destroy(t);
1208
1209		cookie = NULL;
1210		while ((slice = avl_destroy_nodes(&slice_cache,
1211		    &cookie)) != NULL) {
1212			if (slice->rn_config != NULL && !config_failed) {
1213				nvlist_t *config = slice->rn_config;
1214				boolean_t matched = B_TRUE;
1215
1216				if (iarg->poolname != NULL) {
1217					char *pname;
1218
1219					matched = nvlist_lookup_string(config,
1220					    ZPOOL_CONFIG_POOL_NAME,
1221					    &pname) == 0 &&
1222					    strcmp(iarg->poolname, pname) == 0;
1223				} else if (iarg->guid != 0) {
1224					uint64_t this_guid;
1225
1226					matched = nvlist_lookup_uint64(config,
1227					    ZPOOL_CONFIG_POOL_GUID,
1228					    &this_guid) == 0 &&
1229					    iarg->guid == this_guid;
1230				}
1231				if (matched) {
1232					/*
1233					 * use the non-raw path for the config
1234					 */
1235					(void) strlcpy(end, slice->rn_name,
1236					    pathleft);
1237					if (add_config(hdl, &pools, path,
1238					    config) != 0)
1239						config_failed = B_TRUE;
1240				}
1241				nvlist_free(config);
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, iarg->policy);
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 (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE,
1385		    cachefile) != 0) {
1386			(void) no_memory(hdl);
1387			nvlist_free(raw);
1388			nvlist_free(pools);
1389			return (NULL);
1390		}
1391
1392		if ((dst = refresh_config(hdl, src)) == NULL) {
1393			nvlist_free(raw);
1394			nvlist_free(pools);
1395			return (NULL);
1396		}
1397
1398		if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1399			(void) no_memory(hdl);
1400			nvlist_free(dst);
1401			nvlist_free(raw);
1402			nvlist_free(pools);
1403			return (NULL);
1404		}
1405		nvlist_free(dst);
1406	}
1407
1408	nvlist_free(raw);
1409	return (pools);
1410}
1411
1412static int
1413name_or_guid_exists(zpool_handle_t *zhp, void *data)
1414{
1415	importargs_t *import = data;
1416	int found = 0;
1417
1418	if (import->poolname != NULL) {
1419		char *pool_name;
1420
1421		verify(nvlist_lookup_string(zhp->zpool_config,
1422		    ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1423		if (strcmp(pool_name, import->poolname) == 0)
1424			found = 1;
1425	} else {
1426		uint64_t pool_guid;
1427
1428		verify(nvlist_lookup_uint64(zhp->zpool_config,
1429		    ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1430		if (pool_guid == import->guid)
1431			found = 1;
1432	}
1433
1434	zpool_close(zhp);
1435	return (found);
1436}
1437
1438nvlist_t *
1439zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1440{
1441	nvlist_t *pools = NULL;
1442
1443	verify(import->poolname == NULL || import->guid == 0);
1444
1445	if (import->unique)
1446		import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1447
1448	if (import->cachefile != NULL)
1449		pools = zpool_find_import_cached(hdl, import->cachefile,
1450		    import->poolname, import->guid);
1451	else
1452		pools = zpool_find_import_impl(hdl, import);
1453
1454	return (pools);
1455}
1456
1457static boolean_t
1458pool_match(nvlist_t *cfg, char *tgt)
1459{
1460	uint64_t v, guid = strtoull(tgt, NULL, 0);
1461	char *s;
1462
1463	if (guid != 0) {
1464		if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &v) == 0)
1465			return (v == guid);
1466	} else {
1467		if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &s) == 0)
1468			return (strcmp(s, tgt) == 0);
1469	}
1470	return (B_FALSE);
1471}
1472
1473int
1474zpool_tryimport(libzfs_handle_t *hdl, char *target, nvlist_t **configp,
1475    importargs_t *args)
1476{
1477	nvlist_t *pools;
1478	nvlist_t *match = NULL;
1479	nvlist_t *config = NULL;
1480	char *sepp = NULL;
1481	int count = 0;
1482	char *targetdup = strdup(target);
1483
1484	*configp = NULL;
1485
1486	if ((sepp = strpbrk(targetdup, "/@")) != NULL) {
1487		*sepp = '\0';
1488	}
1489
1490	pools = zpool_search_import(hdl, args);
1491
1492	if (pools != NULL) {
1493		nvpair_t *elem = NULL;
1494		while ((elem = nvlist_next_nvpair(pools, elem)) != NULL) {
1495			VERIFY0(nvpair_value_nvlist(elem, &config));
1496			if (pool_match(config, targetdup)) {
1497				count++;
1498				if (match != NULL) {
1499					/* multiple matches found */
1500					continue;
1501				} else {
1502					match = config;
1503				}
1504			}
1505		}
1506	}
1507
1508	if (count == 0) {
1509		free(targetdup);
1510		return (ENOENT);
1511	}
1512
1513	if (count > 1) {
1514		free(targetdup);
1515		return (EINVAL);
1516	}
1517
1518	*configp = match;
1519	free(targetdup);
1520
1521	return (0);
1522}
1523
1524boolean_t
1525find_guid(nvlist_t *nv, uint64_t guid)
1526{
1527	uint64_t tmp;
1528	nvlist_t **child;
1529	uint_t c, children;
1530
1531	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1532	if (tmp == guid)
1533		return (B_TRUE);
1534
1535	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1536	    &child, &children) == 0) {
1537		for (c = 0; c < children; c++)
1538			if (find_guid(child[c], guid))
1539				return (B_TRUE);
1540	}
1541
1542	return (B_FALSE);
1543}
1544
1545typedef struct aux_cbdata {
1546	const char	*cb_type;
1547	uint64_t	cb_guid;
1548	zpool_handle_t	*cb_zhp;
1549} aux_cbdata_t;
1550
1551static int
1552find_aux(zpool_handle_t *zhp, void *data)
1553{
1554	aux_cbdata_t *cbp = data;
1555	nvlist_t **list;
1556	uint_t i, count;
1557	uint64_t guid;
1558	nvlist_t *nvroot;
1559
1560	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1561	    &nvroot) == 0);
1562
1563	if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1564	    &list, &count) == 0) {
1565		for (i = 0; i < count; i++) {
1566			verify(nvlist_lookup_uint64(list[i],
1567			    ZPOOL_CONFIG_GUID, &guid) == 0);
1568			if (guid == cbp->cb_guid) {
1569				cbp->cb_zhp = zhp;
1570				return (1);
1571			}
1572		}
1573	}
1574
1575	zpool_close(zhp);
1576	return (0);
1577}
1578
1579/*
1580 * Determines if the pool is in use.  If so, it returns true and the state of
1581 * the pool as well as the name of the pool.  Both strings are allocated and
1582 * must be freed by the caller.
1583 */
1584int
1585zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1586    boolean_t *inuse)
1587{
1588	nvlist_t *config;
1589	char *name;
1590	boolean_t ret;
1591	uint64_t guid, vdev_guid;
1592	zpool_handle_t *zhp;
1593	nvlist_t *pool_config;
1594	uint64_t stateval, isspare;
1595	aux_cbdata_t cb = { 0 };
1596	boolean_t isactive;
1597
1598	*inuse = B_FALSE;
1599
1600	if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1601		(void) no_memory(hdl);
1602		return (-1);
1603	}
1604
1605	if (config == NULL)
1606		return (0);
1607
1608	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1609	    &stateval) == 0);
1610	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1611	    &vdev_guid) == 0);
1612
1613	if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1614		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1615		    &name) == 0);
1616		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1617		    &guid) == 0);
1618	}
1619
1620	switch (stateval) {
1621	case POOL_STATE_EXPORTED:
1622		/*
1623		 * A pool with an exported state may in fact be imported
1624		 * read-only, so check the in-core state to see if it's
1625		 * active and imported read-only.  If it is, set
1626		 * its state to active.
1627		 */
1628		if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1629		    (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1630			if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1631				stateval = POOL_STATE_ACTIVE;
1632
1633			/*
1634			 * All we needed the zpool handle for is the
1635			 * readonly prop check.
1636			 */
1637			zpool_close(zhp);
1638		}
1639
1640		ret = B_TRUE;
1641		break;
1642
1643	case POOL_STATE_ACTIVE:
1644		/*
1645		 * For an active pool, we have to determine if it's really part
1646		 * of a currently active pool (in which case the pool will exist
1647		 * and the guid will be the same), or whether it's part of an
1648		 * active pool that was disconnected without being explicitly
1649		 * exported.
1650		 */
1651		if (pool_active(hdl, name, guid, &isactive) != 0) {
1652			nvlist_free(config);
1653			return (-1);
1654		}
1655
1656		if (isactive) {
1657			/*
1658			 * Because the device may have been removed while
1659			 * offlined, we only report it as active if the vdev is
1660			 * still present in the config.  Otherwise, pretend like
1661			 * it's not in use.
1662			 */
1663			if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1664			    (pool_config = zpool_get_config(zhp, NULL))
1665			    != NULL) {
1666				nvlist_t *nvroot;
1667
1668				verify(nvlist_lookup_nvlist(pool_config,
1669				    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1670				ret = find_guid(nvroot, vdev_guid);
1671			} else {
1672				ret = B_FALSE;
1673			}
1674
1675			/*
1676			 * If this is an active spare within another pool, we
1677			 * treat it like an unused hot spare.  This allows the
1678			 * user to create a pool with a hot spare that currently
1679			 * in use within another pool.  Since we return B_TRUE,
1680			 * libdiskmgt will continue to prevent generic consumers
1681			 * from using the device.
1682			 */
1683			if (ret && nvlist_lookup_uint64(config,
1684			    ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1685				stateval = POOL_STATE_SPARE;
1686
1687			if (zhp != NULL)
1688				zpool_close(zhp);
1689		} else {
1690			stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1691			ret = B_TRUE;
1692		}
1693		break;
1694
1695	case POOL_STATE_SPARE:
1696		/*
1697		 * For a hot spare, it can be either definitively in use, or
1698		 * potentially active.  To determine if it's in use, we iterate
1699		 * over all pools in the system and search for one with a spare
1700		 * with a matching guid.
1701		 *
1702		 * Due to the shared nature of spares, we don't actually report
1703		 * the potentially active case as in use.  This means the user
1704		 * can freely create pools on the hot spares of exported pools,
1705		 * but to do otherwise makes the resulting code complicated, and
1706		 * we end up having to deal with this case anyway.
1707		 */
1708		cb.cb_zhp = NULL;
1709		cb.cb_guid = vdev_guid;
1710		cb.cb_type = ZPOOL_CONFIG_SPARES;
1711		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1712			name = (char *)zpool_get_name(cb.cb_zhp);
1713			ret = B_TRUE;
1714		} else {
1715			ret = B_FALSE;
1716		}
1717		break;
1718
1719	case POOL_STATE_L2CACHE:
1720
1721		/*
1722		 * Check if any pool is currently using this l2cache device.
1723		 */
1724		cb.cb_zhp = NULL;
1725		cb.cb_guid = vdev_guid;
1726		cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1727		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1728			name = (char *)zpool_get_name(cb.cb_zhp);
1729			ret = B_TRUE;
1730		} else {
1731			ret = B_FALSE;
1732		}
1733		break;
1734
1735	default:
1736		ret = B_FALSE;
1737	}
1738
1739
1740	if (ret) {
1741		if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1742			if (cb.cb_zhp)
1743				zpool_close(cb.cb_zhp);
1744			nvlist_free(config);
1745			return (-1);
1746		}
1747		*state = (pool_state_t)stateval;
1748	}
1749
1750	if (cb.cb_zhp)
1751		zpool_close(cb.cb_zhp);
1752
1753	nvlist_free(config);
1754	*inuse = ret;
1755	return (0);
1756}
1757