xref: /illumos-gate/usr/src/lib/libzfs/common/libzfs_import.c (revision d1aea6f139360e9e7f1504facb24f8521047b15c)
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  */
69 typedef struct config_entry {
70 	uint64_t		ce_txg;
71 	nvlist_t		*ce_config;
72 	struct config_entry	*ce_next;
73 } config_entry_t;
74 
75 typedef 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 
81 typedef 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 
87 typedef struct name_entry {
88 	char			*ne_name;
89 	uint64_t		ne_guid;
90 	struct name_entry	*ne_next;
91 } name_entry_t;
92 
93 typedef 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  */
102 static int
103 fix_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  */
189 static int
190 add_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  */
329 static int
330 pool_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 
353 static nvlist_t *
354 refresh_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  */
395 boolean_t
396 vdev_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  */
414 static nvlist_t *
415 get_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 
797 add_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 
818 nomem:
819 	(void) no_memory(hdl);
820 error:
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  */
833 static uint64_t
834 label_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  */
845 int
846 zpool_read_label(int fd, nvlist_t **config)
847 {
848 	struct stat64 statbuf;
849 	int l;
850 	vdev_label_t *label;
851 	uint64_t state, txg, size;
852 
853 	*config = NULL;
854 
855 	if (fstat64(fd, &statbuf) == -1)
856 		return (0);
857 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
858 
859 	if ((label = malloc(sizeof (vdev_label_t))) == NULL)
860 		return (-1);
861 
862 	for (l = 0; l < VDEV_LABELS; l++) {
863 		if (pread64(fd, label, sizeof (vdev_label_t),
864 		    label_offset(size, l)) != sizeof (vdev_label_t))
865 			continue;
866 
867 		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
868 		    sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
869 			continue;
870 
871 		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
872 		    &state) != 0 || state > POOL_STATE_L2CACHE) {
873 			nvlist_free(*config);
874 			continue;
875 		}
876 
877 		if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
878 		    (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
879 		    &txg) != 0 || txg == 0)) {
880 			nvlist_free(*config);
881 			continue;
882 		}
883 
884 		free(label);
885 		return (0);
886 	}
887 
888 	free(label);
889 	*config = NULL;
890 	return (0);
891 }
892 
893 typedef struct rdsk_node {
894 	char *rn_name;
895 	int rn_dfd;
896 	libzfs_handle_t *rn_hdl;
897 	nvlist_t *rn_config;
898 	avl_tree_t *rn_avl;
899 	avl_node_t rn_node;
900 	boolean_t rn_nozpool;
901 } rdsk_node_t;
902 
903 static int
904 slice_cache_compare(const void *arg1, const void *arg2)
905 {
906 	const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
907 	const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
908 	char *nm1slice, *nm2slice;
909 	int rv;
910 
911 	/*
912 	 * slices zero and two are the most likely to provide results,
913 	 * so put those first
914 	 */
915 	nm1slice = strstr(nm1, "s0");
916 	nm2slice = strstr(nm2, "s0");
917 	if (nm1slice && !nm2slice) {
918 		return (-1);
919 	}
920 	if (!nm1slice && nm2slice) {
921 		return (1);
922 	}
923 	nm1slice = strstr(nm1, "s2");
924 	nm2slice = strstr(nm2, "s2");
925 	if (nm1slice && !nm2slice) {
926 		return (-1);
927 	}
928 	if (!nm1slice && nm2slice) {
929 		return (1);
930 	}
931 
932 	rv = strcmp(nm1, nm2);
933 	if (rv == 0)
934 		return (0);
935 	return (rv > 0 ? 1 : -1);
936 }
937 
938 static void
939 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
940     diskaddr_t size, uint_t blksz)
941 {
942 	rdsk_node_t tmpnode;
943 	rdsk_node_t *node;
944 	char sname[MAXNAMELEN];
945 
946 	tmpnode.rn_name = &sname[0];
947 	(void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
948 	    diskname, partno);
949 	/*
950 	 * protect against division by zero for disk labels that
951 	 * contain a bogus sector size
952 	 */
953 	if (blksz == 0)
954 		blksz = DEV_BSIZE;
955 	/* too small to contain a zpool? */
956 	if ((size < (SPA_MINDEVSIZE / blksz)) &&
957 	    (node = avl_find(r, &tmpnode, NULL)))
958 		node->rn_nozpool = B_TRUE;
959 }
960 
961 static void
962 nozpool_all_slices(avl_tree_t *r, const char *sname)
963 {
964 	char diskname[MAXNAMELEN];
965 	char *ptr;
966 	int i;
967 
968 	(void) strncpy(diskname, sname, MAXNAMELEN);
969 	if (((ptr = strrchr(diskname, 's')) == NULL) &&
970 	    ((ptr = strrchr(diskname, 'p')) == NULL))
971 		return;
972 	ptr[0] = 's';
973 	ptr[1] = '\0';
974 	for (i = 0; i < NDKMAP; i++)
975 		check_one_slice(r, diskname, i, 0, 1);
976 	ptr[0] = 'p';
977 	for (i = 0; i <= FD_NUMPART; i++)
978 		check_one_slice(r, diskname, i, 0, 1);
979 }
980 
981 static void
982 check_slices(avl_tree_t *r, int fd, const char *sname)
983 {
984 	struct extvtoc vtoc;
985 	struct dk_gpt *gpt;
986 	char diskname[MAXNAMELEN];
987 	char *ptr;
988 	int i;
989 
990 	(void) strncpy(diskname, sname, MAXNAMELEN);
991 	if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
992 		return;
993 	ptr[1] = '\0';
994 
995 	if (read_extvtoc(fd, &vtoc) >= 0) {
996 		for (i = 0; i < NDKMAP; i++)
997 			check_one_slice(r, diskname, i,
998 			    vtoc.v_part[i].p_size, vtoc.v_sectorsz);
999 	} else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1000 		/*
1001 		 * on x86 we'll still have leftover links that point
1002 		 * to slices s[9-15], so use NDKMAP instead
1003 		 */
1004 		for (i = 0; i < NDKMAP; i++)
1005 			check_one_slice(r, diskname, i,
1006 			    gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1007 		/* nodes p[1-4] are never used with EFI labels */
1008 		ptr[0] = 'p';
1009 		for (i = 1; i <= FD_NUMPART; i++)
1010 			check_one_slice(r, diskname, i, 0, 1);
1011 		efi_free(gpt);
1012 	}
1013 }
1014 
1015 static void
1016 zpool_open_func(void *arg)
1017 {
1018 	rdsk_node_t *rn = arg;
1019 	struct stat64 statbuf;
1020 	nvlist_t *config;
1021 	int fd;
1022 
1023 	if (rn->rn_nozpool)
1024 		return;
1025 	if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1026 		/* symlink to a device that's no longer there */
1027 		if (errno == ENOENT)
1028 			nozpool_all_slices(rn->rn_avl, rn->rn_name);
1029 		return;
1030 	}
1031 	/*
1032 	 * Ignore failed stats.  We only want regular
1033 	 * files, character devs and block devs.
1034 	 */
1035 	if (fstat64(fd, &statbuf) != 0 ||
1036 	    (!S_ISREG(statbuf.st_mode) &&
1037 	    !S_ISCHR(statbuf.st_mode) &&
1038 	    !S_ISBLK(statbuf.st_mode))) {
1039 		(void) close(fd);
1040 		return;
1041 	}
1042 	/* this file is too small to hold a zpool */
1043 	if (S_ISREG(statbuf.st_mode) &&
1044 	    statbuf.st_size < SPA_MINDEVSIZE) {
1045 		(void) close(fd);
1046 		return;
1047 	} else if (!S_ISREG(statbuf.st_mode)) {
1048 		/*
1049 		 * Try to read the disk label first so we don't have to
1050 		 * open a bunch of minor nodes that can't have a zpool.
1051 		 */
1052 		check_slices(rn->rn_avl, fd, rn->rn_name);
1053 	}
1054 
1055 	if ((zpool_read_label(fd, &config)) != 0) {
1056 		(void) close(fd);
1057 		(void) no_memory(rn->rn_hdl);
1058 		return;
1059 	}
1060 	(void) close(fd);
1061 
1062 	rn->rn_config = config;
1063 }
1064 
1065 /*
1066  * Given a file descriptor, clear (zero) the label information.
1067  */
1068 int
1069 zpool_clear_label(int fd)
1070 {
1071 	struct stat64 statbuf;
1072 	int l;
1073 	vdev_label_t *label;
1074 	uint64_t size;
1075 
1076 	if (fstat64(fd, &statbuf) == -1)
1077 		return (0);
1078 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1079 
1080 	if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1081 		return (-1);
1082 
1083 	for (l = 0; l < VDEV_LABELS; l++) {
1084 		if (pwrite64(fd, label, sizeof (vdev_label_t),
1085 		    label_offset(size, l)) != sizeof (vdev_label_t)) {
1086 			free(label);
1087 			return (-1);
1088 		}
1089 	}
1090 
1091 	free(label);
1092 	return (0);
1093 }
1094 
1095 /*
1096  * Given a list of directories to search, find all pools stored on disk.  This
1097  * includes partial pools which are not available to import.  If no args are
1098  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1099  * poolname or guid (but not both) are provided by the caller when trying
1100  * to import a specific pool.
1101  */
1102 static nvlist_t *
1103 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1104 {
1105 	int i, dirs = iarg->paths;
1106 	struct dirent64 *dp;
1107 	char path[MAXPATHLEN];
1108 	char *end, **dir = iarg->path;
1109 	size_t pathleft;
1110 	nvlist_t *ret = NULL;
1111 	static char *default_dir = ZFS_DISK_ROOT;
1112 	pool_list_t pools = { 0 };
1113 	pool_entry_t *pe, *penext;
1114 	vdev_entry_t *ve, *venext;
1115 	config_entry_t *ce, *cenext;
1116 	name_entry_t *ne, *nenext;
1117 	avl_tree_t slice_cache;
1118 	rdsk_node_t *slice;
1119 	void *cookie;
1120 
1121 	if (dirs == 0) {
1122 		dirs = 1;
1123 		dir = &default_dir;
1124 	}
1125 
1126 	/*
1127 	 * Go through and read the label configuration information from every
1128 	 * possible device, organizing the information according to pool GUID
1129 	 * and toplevel GUID.
1130 	 */
1131 	for (i = 0; i < dirs; i++) {
1132 		tpool_t *t;
1133 		char rdsk[MAXPATHLEN];
1134 		int dfd;
1135 		boolean_t config_failed = B_FALSE;
1136 		DIR *dirp;
1137 
1138 		/* use realpath to normalize the path */
1139 		if (realpath(dir[i], path) == 0) {
1140 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1141 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1142 			goto error;
1143 		}
1144 		end = &path[strlen(path)];
1145 		*end++ = '/';
1146 		*end = 0;
1147 		pathleft = &path[sizeof (path)] - end;
1148 
1149 		/*
1150 		 * Using raw devices instead of block devices when we're
1151 		 * reading the labels skips a bunch of slow operations during
1152 		 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1153 		 */
1154 		if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1155 			(void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1156 		else
1157 			(void) strlcpy(rdsk, path, sizeof (rdsk));
1158 
1159 		if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1160 		    (dirp = fdopendir(dfd)) == NULL) {
1161 			if (dfd >= 0)
1162 				(void) close(dfd);
1163 			zfs_error_aux(hdl, strerror(errno));
1164 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1165 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1166 			    rdsk);
1167 			goto error;
1168 		}
1169 
1170 		avl_create(&slice_cache, slice_cache_compare,
1171 		    sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1172 		/*
1173 		 * This is not MT-safe, but we have no MT consumers of libzfs
1174 		 */
1175 		while ((dp = readdir64(dirp)) != NULL) {
1176 			const char *name = dp->d_name;
1177 			if (name[0] == '.' &&
1178 			    (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1179 				continue;
1180 
1181 			slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1182 			slice->rn_name = zfs_strdup(hdl, name);
1183 			slice->rn_avl = &slice_cache;
1184 			slice->rn_dfd = dfd;
1185 			slice->rn_hdl = hdl;
1186 			slice->rn_nozpool = B_FALSE;
1187 			avl_add(&slice_cache, slice);
1188 		}
1189 		/*
1190 		 * create a thread pool to do all of this in parallel;
1191 		 * rn_nozpool is not protected, so this is racy in that
1192 		 * multiple tasks could decide that the same slice can
1193 		 * not hold a zpool, which is benign.  Also choose
1194 		 * double the number of processors; we hold a lot of
1195 		 * locks in the kernel, so going beyond this doesn't
1196 		 * buy us much.
1197 		 */
1198 		t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1199 		    0, NULL);
1200 		for (slice = avl_first(&slice_cache); slice;
1201 		    (slice = avl_walk(&slice_cache, slice,
1202 		    AVL_AFTER)))
1203 			(void) tpool_dispatch(t, zpool_open_func, slice);
1204 		tpool_wait(t);
1205 		tpool_destroy(t);
1206 
1207 		cookie = NULL;
1208 		while ((slice = avl_destroy_nodes(&slice_cache,
1209 		    &cookie)) != NULL) {
1210 			if (slice->rn_config != NULL && !config_failed) {
1211 				nvlist_t *config = slice->rn_config;
1212 				boolean_t matched = B_TRUE;
1213 
1214 				if (iarg->poolname != NULL) {
1215 					char *pname;
1216 
1217 					matched = nvlist_lookup_string(config,
1218 					    ZPOOL_CONFIG_POOL_NAME,
1219 					    &pname) == 0 &&
1220 					    strcmp(iarg->poolname, pname) == 0;
1221 				} else if (iarg->guid != 0) {
1222 					uint64_t this_guid;
1223 
1224 					matched = nvlist_lookup_uint64(config,
1225 					    ZPOOL_CONFIG_POOL_GUID,
1226 					    &this_guid) == 0 &&
1227 					    iarg->guid == this_guid;
1228 				}
1229 				if (!matched) {
1230 					nvlist_free(config);
1231 				} else {
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 			}
1242 			free(slice->rn_name);
1243 			free(slice);
1244 		}
1245 		avl_destroy(&slice_cache);
1246 
1247 		(void) closedir(dirp);
1248 
1249 		if (config_failed)
1250 			goto error;
1251 	}
1252 
1253 	ret = get_configs(hdl, &pools, iarg->can_be_active);
1254 
1255 error:
1256 	for (pe = pools.pools; pe != NULL; pe = penext) {
1257 		penext = pe->pe_next;
1258 		for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1259 			venext = ve->ve_next;
1260 			for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1261 				cenext = ce->ce_next;
1262 				nvlist_free(ce->ce_config);
1263 				free(ce);
1264 			}
1265 			free(ve);
1266 		}
1267 		free(pe);
1268 	}
1269 
1270 	for (ne = pools.names; ne != NULL; ne = nenext) {
1271 		nenext = ne->ne_next;
1272 		free(ne->ne_name);
1273 		free(ne);
1274 	}
1275 
1276 	return (ret);
1277 }
1278 
1279 nvlist_t *
1280 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1281 {
1282 	importargs_t iarg = { 0 };
1283 
1284 	iarg.paths = argc;
1285 	iarg.path = argv;
1286 
1287 	return (zpool_find_import_impl(hdl, &iarg));
1288 }
1289 
1290 /*
1291  * Given a cache file, return the contents as a list of importable pools.
1292  * poolname or guid (but not both) are provided by the caller when trying
1293  * to import a specific pool.
1294  */
1295 nvlist_t *
1296 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1297     char *poolname, uint64_t guid)
1298 {
1299 	char *buf;
1300 	int fd;
1301 	struct stat64 statbuf;
1302 	nvlist_t *raw, *src, *dst;
1303 	nvlist_t *pools;
1304 	nvpair_t *elem;
1305 	char *name;
1306 	uint64_t this_guid;
1307 	boolean_t active;
1308 
1309 	verify(poolname == NULL || guid == 0);
1310 
1311 	if ((fd = open(cachefile, O_RDONLY)) < 0) {
1312 		zfs_error_aux(hdl, "%s", strerror(errno));
1313 		(void) zfs_error(hdl, EZFS_BADCACHE,
1314 		    dgettext(TEXT_DOMAIN, "failed to open cache file"));
1315 		return (NULL);
1316 	}
1317 
1318 	if (fstat64(fd, &statbuf) != 0) {
1319 		zfs_error_aux(hdl, "%s", strerror(errno));
1320 		(void) close(fd);
1321 		(void) zfs_error(hdl, EZFS_BADCACHE,
1322 		    dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1323 		return (NULL);
1324 	}
1325 
1326 	if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1327 		(void) close(fd);
1328 		return (NULL);
1329 	}
1330 
1331 	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1332 		(void) close(fd);
1333 		free(buf);
1334 		(void) zfs_error(hdl, EZFS_BADCACHE,
1335 		    dgettext(TEXT_DOMAIN,
1336 		    "failed to read cache file contents"));
1337 		return (NULL);
1338 	}
1339 
1340 	(void) close(fd);
1341 
1342 	if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1343 		free(buf);
1344 		(void) zfs_error(hdl, EZFS_BADCACHE,
1345 		    dgettext(TEXT_DOMAIN,
1346 		    "invalid or corrupt cache file contents"));
1347 		return (NULL);
1348 	}
1349 
1350 	free(buf);
1351 
1352 	/*
1353 	 * Go through and get the current state of the pools and refresh their
1354 	 * state.
1355 	 */
1356 	if (nvlist_alloc(&pools, 0, 0) != 0) {
1357 		(void) no_memory(hdl);
1358 		nvlist_free(raw);
1359 		return (NULL);
1360 	}
1361 
1362 	elem = NULL;
1363 	while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1364 		src = fnvpair_value_nvlist(elem);
1365 
1366 		name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1367 		if (poolname != NULL && strcmp(poolname, name) != 0)
1368 			continue;
1369 
1370 		this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1371 		if (guid != 0 && guid != this_guid)
1372 			continue;
1373 
1374 		if (pool_active(hdl, name, this_guid, &active) != 0) {
1375 			nvlist_free(raw);
1376 			nvlist_free(pools);
1377 			return (NULL);
1378 		}
1379 
1380 		if (active)
1381 			continue;
1382 
1383 		if ((dst = refresh_config(hdl, src)) == NULL) {
1384 			nvlist_free(raw);
1385 			nvlist_free(pools);
1386 			return (NULL);
1387 		}
1388 
1389 		if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1390 			(void) no_memory(hdl);
1391 			nvlist_free(dst);
1392 			nvlist_free(raw);
1393 			nvlist_free(pools);
1394 			return (NULL);
1395 		}
1396 		nvlist_free(dst);
1397 	}
1398 
1399 	nvlist_free(raw);
1400 	return (pools);
1401 }
1402 
1403 static int
1404 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1405 {
1406 	importargs_t *import = data;
1407 	int found = 0;
1408 
1409 	if (import->poolname != NULL) {
1410 		char *pool_name;
1411 
1412 		verify(nvlist_lookup_string(zhp->zpool_config,
1413 		    ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1414 		if (strcmp(pool_name, import->poolname) == 0)
1415 			found = 1;
1416 	} else {
1417 		uint64_t pool_guid;
1418 
1419 		verify(nvlist_lookup_uint64(zhp->zpool_config,
1420 		    ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1421 		if (pool_guid == import->guid)
1422 			found = 1;
1423 	}
1424 
1425 	zpool_close(zhp);
1426 	return (found);
1427 }
1428 
1429 nvlist_t *
1430 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1431 {
1432 	verify(import->poolname == NULL || import->guid == 0);
1433 
1434 	if (import->unique)
1435 		import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1436 
1437 	if (import->cachefile != NULL)
1438 		return (zpool_find_import_cached(hdl, import->cachefile,
1439 		    import->poolname, import->guid));
1440 
1441 	return (zpool_find_import_impl(hdl, import));
1442 }
1443 
1444 boolean_t
1445 find_guid(nvlist_t *nv, uint64_t guid)
1446 {
1447 	uint64_t tmp;
1448 	nvlist_t **child;
1449 	uint_t c, children;
1450 
1451 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1452 	if (tmp == guid)
1453 		return (B_TRUE);
1454 
1455 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1456 	    &child, &children) == 0) {
1457 		for (c = 0; c < children; c++)
1458 			if (find_guid(child[c], guid))
1459 				return (B_TRUE);
1460 	}
1461 
1462 	return (B_FALSE);
1463 }
1464 
1465 typedef struct aux_cbdata {
1466 	const char	*cb_type;
1467 	uint64_t	cb_guid;
1468 	zpool_handle_t	*cb_zhp;
1469 } aux_cbdata_t;
1470 
1471 static int
1472 find_aux(zpool_handle_t *zhp, void *data)
1473 {
1474 	aux_cbdata_t *cbp = data;
1475 	nvlist_t **list;
1476 	uint_t i, count;
1477 	uint64_t guid;
1478 	nvlist_t *nvroot;
1479 
1480 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1481 	    &nvroot) == 0);
1482 
1483 	if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1484 	    &list, &count) == 0) {
1485 		for (i = 0; i < count; i++) {
1486 			verify(nvlist_lookup_uint64(list[i],
1487 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1488 			if (guid == cbp->cb_guid) {
1489 				cbp->cb_zhp = zhp;
1490 				return (1);
1491 			}
1492 		}
1493 	}
1494 
1495 	zpool_close(zhp);
1496 	return (0);
1497 }
1498 
1499 /*
1500  * Determines if the pool is in use.  If so, it returns true and the state of
1501  * the pool as well as the name of the pool.  Both strings are allocated and
1502  * must be freed by the caller.
1503  */
1504 int
1505 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1506     boolean_t *inuse)
1507 {
1508 	nvlist_t *config;
1509 	char *name;
1510 	boolean_t ret;
1511 	uint64_t guid, vdev_guid;
1512 	zpool_handle_t *zhp;
1513 	nvlist_t *pool_config;
1514 	uint64_t stateval, isspare;
1515 	aux_cbdata_t cb = { 0 };
1516 	boolean_t isactive;
1517 
1518 	*inuse = B_FALSE;
1519 
1520 	if (zpool_read_label(fd, &config) != 0) {
1521 		(void) no_memory(hdl);
1522 		return (-1);
1523 	}
1524 
1525 	if (config == NULL)
1526 		return (0);
1527 
1528 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1529 	    &stateval) == 0);
1530 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1531 	    &vdev_guid) == 0);
1532 
1533 	if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1534 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1535 		    &name) == 0);
1536 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1537 		    &guid) == 0);
1538 	}
1539 
1540 	switch (stateval) {
1541 	case POOL_STATE_EXPORTED:
1542 		/*
1543 		 * A pool with an exported state may in fact be imported
1544 		 * read-only, so check the in-core state to see if it's
1545 		 * active and imported read-only.  If it is, set
1546 		 * its state to active.
1547 		 */
1548 		if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1549 		    (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1550 			if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1551 				stateval = POOL_STATE_ACTIVE;
1552 
1553 			/*
1554 			 * All we needed the zpool handle for is the
1555 			 * readonly prop check.
1556 			 */
1557 			zpool_close(zhp);
1558 		}
1559 
1560 		ret = B_TRUE;
1561 		break;
1562 
1563 	case POOL_STATE_ACTIVE:
1564 		/*
1565 		 * For an active pool, we have to determine if it's really part
1566 		 * of a currently active pool (in which case the pool will exist
1567 		 * and the guid will be the same), or whether it's part of an
1568 		 * active pool that was disconnected without being explicitly
1569 		 * exported.
1570 		 */
1571 		if (pool_active(hdl, name, guid, &isactive) != 0) {
1572 			nvlist_free(config);
1573 			return (-1);
1574 		}
1575 
1576 		if (isactive) {
1577 			/*
1578 			 * Because the device may have been removed while
1579 			 * offlined, we only report it as active if the vdev is
1580 			 * still present in the config.  Otherwise, pretend like
1581 			 * it's not in use.
1582 			 */
1583 			if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1584 			    (pool_config = zpool_get_config(zhp, NULL))
1585 			    != NULL) {
1586 				nvlist_t *nvroot;
1587 
1588 				verify(nvlist_lookup_nvlist(pool_config,
1589 				    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1590 				ret = find_guid(nvroot, vdev_guid);
1591 			} else {
1592 				ret = B_FALSE;
1593 			}
1594 
1595 			/*
1596 			 * If this is an active spare within another pool, we
1597 			 * treat it like an unused hot spare.  This allows the
1598 			 * user to create a pool with a hot spare that currently
1599 			 * in use within another pool.  Since we return B_TRUE,
1600 			 * libdiskmgt will continue to prevent generic consumers
1601 			 * from using the device.
1602 			 */
1603 			if (ret && nvlist_lookup_uint64(config,
1604 			    ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1605 				stateval = POOL_STATE_SPARE;
1606 
1607 			if (zhp != NULL)
1608 				zpool_close(zhp);
1609 		} else {
1610 			stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1611 			ret = B_TRUE;
1612 		}
1613 		break;
1614 
1615 	case POOL_STATE_SPARE:
1616 		/*
1617 		 * For a hot spare, it can be either definitively in use, or
1618 		 * potentially active.  To determine if it's in use, we iterate
1619 		 * over all pools in the system and search for one with a spare
1620 		 * with a matching guid.
1621 		 *
1622 		 * Due to the shared nature of spares, we don't actually report
1623 		 * the potentially active case as in use.  This means the user
1624 		 * can freely create pools on the hot spares of exported pools,
1625 		 * but to do otherwise makes the resulting code complicated, and
1626 		 * we end up having to deal with this case anyway.
1627 		 */
1628 		cb.cb_zhp = NULL;
1629 		cb.cb_guid = vdev_guid;
1630 		cb.cb_type = ZPOOL_CONFIG_SPARES;
1631 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1632 			name = (char *)zpool_get_name(cb.cb_zhp);
1633 			ret = B_TRUE;
1634 		} else {
1635 			ret = B_FALSE;
1636 		}
1637 		break;
1638 
1639 	case POOL_STATE_L2CACHE:
1640 
1641 		/*
1642 		 * Check if any pool is currently using this l2cache device.
1643 		 */
1644 		cb.cb_zhp = NULL;
1645 		cb.cb_guid = vdev_guid;
1646 		cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1647 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1648 			name = (char *)zpool_get_name(cb.cb_zhp);
1649 			ret = B_TRUE;
1650 		} else {
1651 			ret = B_FALSE;
1652 		}
1653 		break;
1654 
1655 	default:
1656 		ret = B_FALSE;
1657 	}
1658 
1659 
1660 	if (ret) {
1661 		if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1662 			if (cb.cb_zhp)
1663 				zpool_close(cb.cb_zhp);
1664 			nvlist_free(config);
1665 			return (-1);
1666 		}
1667 		*state = (pool_state_t)stateval;
1668 	}
1669 
1670 	if (cb.cb_zhp)
1671 		zpool_close(cb.cb_zhp);
1672 
1673 	nvlist_free(config);
1674 	*inuse = ret;
1675 	return (0);
1676 }
1677