xref: /titanic_41/usr/src/lib/libzfs/common/libzfs_import.c (revision a9b5625ad449e28716053a1fd299e7e3271f750d)
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 2015 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2013 by Delphix. All rights reserved.
26  */
27 
28 /*
29  * Pool import support functions.
30  *
31  * To import a pool, we rely on reading the configuration information from the
32  * ZFS label of each device.  If we successfully read the label, then we
33  * organize the configuration information in the following hierarchy:
34  *
35  * 	pool guid -> toplevel vdev guid -> label txg
36  *
37  * Duplicate entries matching this same tuple will be discarded.  Once we have
38  * examined every device, we pick the best label txg config for each toplevel
39  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
40  * update any paths that have changed.  Finally, we attempt to import the pool
41  * using our derived config, and record the results.
42  */
43 
44 #include <ctype.h>
45 #include <devid.h>
46 #include <dirent.h>
47 #include <errno.h>
48 #include <libintl.h>
49 #include <stddef.h>
50 #include <stdlib.h>
51 #include <string.h>
52 #include <sys/stat.h>
53 #include <unistd.h>
54 #include <fcntl.h>
55 #include <sys/vtoc.h>
56 #include <sys/dktp/fdisk.h>
57 #include <sys/efi_partition.h>
58 #include <thread_pool.h>
59 
60 #include <sys/vdev_impl.h>
61 
62 #include "libzfs.h"
63 #include "libzfs_impl.h"
64 
65 /*
66  * Intermediate structures used to gather configuration information.
67  */
68 typedef struct config_entry {
69 	uint64_t		ce_txg;
70 	nvlist_t		*ce_config;
71 	struct config_entry	*ce_next;
72 } config_entry_t;
73 
74 typedef struct vdev_entry {
75 	uint64_t		ve_guid;
76 	config_entry_t		*ve_configs;
77 	struct vdev_entry	*ve_next;
78 } vdev_entry_t;
79 
80 typedef struct pool_entry {
81 	uint64_t		pe_guid;
82 	vdev_entry_t		*pe_vdevs;
83 	struct pool_entry	*pe_next;
84 } pool_entry_t;
85 
86 typedef struct name_entry {
87 	char			*ne_name;
88 	uint64_t		ne_guid;
89 	struct name_entry	*ne_next;
90 } name_entry_t;
91 
92 typedef struct pool_list {
93 	pool_entry_t		*pools;
94 	name_entry_t		*names;
95 } pool_list_t;
96 
97 static char *
98 get_devid(const char *path)
99 {
100 	int fd;
101 	ddi_devid_t devid;
102 	char *minor, *ret;
103 
104 	if ((fd = open(path, O_RDONLY)) < 0)
105 		return (NULL);
106 
107 	minor = NULL;
108 	ret = NULL;
109 	if (devid_get(fd, &devid) == 0) {
110 		if (devid_get_minor_name(fd, &minor) == 0)
111 			ret = devid_str_encode(devid, minor);
112 		if (minor != NULL)
113 			devid_str_free(minor);
114 		devid_free(devid);
115 	}
116 	(void) close(fd);
117 
118 	return (ret);
119 }
120 
121 
122 /*
123  * Go through and fix up any path and/or devid information for the given vdev
124  * configuration.
125  */
126 static int
127 fix_paths(nvlist_t *nv, name_entry_t *names)
128 {
129 	nvlist_t **child;
130 	uint_t c, children;
131 	uint64_t guid;
132 	name_entry_t *ne, *best;
133 	char *path, *devid;
134 	int matched;
135 
136 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
137 	    &child, &children) == 0) {
138 		for (c = 0; c < children; c++)
139 			if (fix_paths(child[c], names) != 0)
140 				return (-1);
141 		return (0);
142 	}
143 
144 	/*
145 	 * This is a leaf (file or disk) vdev.  In either case, go through
146 	 * the name list and see if we find a matching guid.  If so, replace
147 	 * the path and see if we can calculate a new devid.
148 	 *
149 	 * There may be multiple names associated with a particular guid, in
150 	 * which case we have overlapping slices or multiple paths to the same
151 	 * disk.  If this is the case, then we want to pick the path that is
152 	 * the most similar to the original, where "most similar" is the number
153 	 * of matching characters starting from the end of the path.  This will
154 	 * preserve slice numbers even if the disks have been reorganized, and
155 	 * will also catch preferred disk names if multiple paths exist.
156 	 */
157 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
158 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
159 		path = NULL;
160 
161 	matched = 0;
162 	best = NULL;
163 	for (ne = names; ne != NULL; ne = ne->ne_next) {
164 		if (ne->ne_guid == guid) {
165 			const char *src, *dst;
166 			int count;
167 
168 			if (path == NULL) {
169 				best = ne;
170 				break;
171 			}
172 
173 			src = ne->ne_name + strlen(ne->ne_name) - 1;
174 			dst = path + strlen(path) - 1;
175 			for (count = 0; src >= ne->ne_name && dst >= path;
176 			    src--, dst--, count++)
177 				if (*src != *dst)
178 					break;
179 
180 			/*
181 			 * At this point, 'count' is the number of characters
182 			 * matched from the end.
183 			 */
184 			if (count > matched || best == NULL) {
185 				best = ne;
186 				matched = count;
187 			}
188 		}
189 	}
190 
191 	if (best == NULL)
192 		return (0);
193 
194 	if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
195 		return (-1);
196 
197 	if ((devid = get_devid(best->ne_name)) == NULL) {
198 		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
199 	} else {
200 		if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
201 			devid_str_free(devid);
202 			return (-1);
203 		}
204 		devid_str_free(devid);
205 	}
206 
207 	return (0);
208 }
209 
210 /*
211  * Add the given configuration to the list of known devices.
212  */
213 static int
214 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
215     nvlist_t *config)
216 {
217 	uint64_t pool_guid, vdev_guid, top_guid, txg, state;
218 	pool_entry_t *pe;
219 	vdev_entry_t *ve;
220 	config_entry_t *ce;
221 	name_entry_t *ne;
222 
223 	/*
224 	 * If this is a hot spare not currently in use or level 2 cache
225 	 * device, add it to the list of names to translate, but don't do
226 	 * anything else.
227 	 */
228 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
229 	    &state) == 0 &&
230 	    (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
231 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
232 		if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
233 			return (-1);
234 
235 		if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
236 			free(ne);
237 			return (-1);
238 		}
239 		ne->ne_guid = vdev_guid;
240 		ne->ne_next = pl->names;
241 		pl->names = ne;
242 		return (0);
243 	}
244 
245 	/*
246 	 * If we have a valid config but cannot read any of these fields, then
247 	 * it means we have a half-initialized label.  In vdev_label_init()
248 	 * we write a label with txg == 0 so that we can identify the device
249 	 * in case the user refers to the same disk later on.  If we fail to
250 	 * create the pool, we'll be left with a label in this state
251 	 * which should not be considered part of a valid pool.
252 	 */
253 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
254 	    &pool_guid) != 0 ||
255 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
256 	    &vdev_guid) != 0 ||
257 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
258 	    &top_guid) != 0 ||
259 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
260 	    &txg) != 0 || txg == 0) {
261 		nvlist_free(config);
262 		return (0);
263 	}
264 
265 	/*
266 	 * First, see if we know about this pool.  If not, then add it to the
267 	 * list of known pools.
268 	 */
269 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
270 		if (pe->pe_guid == pool_guid)
271 			break;
272 	}
273 
274 	if (pe == NULL) {
275 		if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
276 			nvlist_free(config);
277 			return (-1);
278 		}
279 		pe->pe_guid = pool_guid;
280 		pe->pe_next = pl->pools;
281 		pl->pools = pe;
282 	}
283 
284 	/*
285 	 * Second, see if we know about this toplevel vdev.  Add it if its
286 	 * missing.
287 	 */
288 	for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
289 		if (ve->ve_guid == top_guid)
290 			break;
291 	}
292 
293 	if (ve == NULL) {
294 		if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
295 			nvlist_free(config);
296 			return (-1);
297 		}
298 		ve->ve_guid = top_guid;
299 		ve->ve_next = pe->pe_vdevs;
300 		pe->pe_vdevs = ve;
301 	}
302 
303 	/*
304 	 * Third, see if we have a config with a matching transaction group.  If
305 	 * so, then we do nothing.  Otherwise, add it to the list of known
306 	 * configs.
307 	 */
308 	for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
309 		if (ce->ce_txg == txg)
310 			break;
311 	}
312 
313 	if (ce == NULL) {
314 		if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
315 			nvlist_free(config);
316 			return (-1);
317 		}
318 		ce->ce_txg = txg;
319 		ce->ce_config = config;
320 		ce->ce_next = ve->ve_configs;
321 		ve->ve_configs = ce;
322 	} else {
323 		nvlist_free(config);
324 	}
325 
326 	/*
327 	 * At this point we've successfully added our config to the list of
328 	 * known configs.  The last thing to do is add the vdev guid -> path
329 	 * mappings so that we can fix up the configuration as necessary before
330 	 * doing the import.
331 	 */
332 	if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
333 		return (-1);
334 
335 	if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
336 		free(ne);
337 		return (-1);
338 	}
339 
340 	ne->ne_guid = vdev_guid;
341 	ne->ne_next = pl->names;
342 	pl->names = ne;
343 
344 	return (0);
345 }
346 
347 /*
348  * Returns true if the named pool matches the given GUID.
349  */
350 static int
351 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
352     boolean_t *isactive)
353 {
354 	zpool_handle_t *zhp;
355 	uint64_t theguid;
356 
357 	if (zpool_open_silent(hdl, name, &zhp) != 0)
358 		return (-1);
359 
360 	if (zhp == NULL) {
361 		*isactive = B_FALSE;
362 		return (0);
363 	}
364 
365 	verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
366 	    &theguid) == 0);
367 
368 	zpool_close(zhp);
369 
370 	*isactive = (theguid == guid);
371 	return (0);
372 }
373 
374 static nvlist_t *
375 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
376 {
377 	nvlist_t *nvl;
378 	zfs_cmd_t zc = { 0 };
379 	int err;
380 
381 	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
382 		return (NULL);
383 
384 	if (zcmd_alloc_dst_nvlist(hdl, &zc,
385 	    zc.zc_nvlist_conf_size * 2) != 0) {
386 		zcmd_free_nvlists(&zc);
387 		return (NULL);
388 	}
389 
390 	while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
391 	    &zc)) != 0 && errno == ENOMEM) {
392 		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
393 			zcmd_free_nvlists(&zc);
394 			return (NULL);
395 		}
396 	}
397 
398 	if (err) {
399 		zcmd_free_nvlists(&zc);
400 		return (NULL);
401 	}
402 
403 	if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
404 		zcmd_free_nvlists(&zc);
405 		return (NULL);
406 	}
407 
408 	zcmd_free_nvlists(&zc);
409 	return (nvl);
410 }
411 
412 /*
413  * Determine if the vdev id is a hole in the namespace.
414  */
415 boolean_t
416 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
417 {
418 	for (int c = 0; c < holes; c++) {
419 
420 		/* Top-level is a hole */
421 		if (hole_array[c] == id)
422 			return (B_TRUE);
423 	}
424 	return (B_FALSE);
425 }
426 
427 /*
428  * Convert our list of pools into the definitive set of configurations.  We
429  * start by picking the best config for each toplevel vdev.  Once that's done,
430  * we assemble the toplevel vdevs into a full config for the pool.  We make a
431  * pass to fix up any incorrect paths, and then add it to the main list to
432  * return to the user.
433  */
434 static nvlist_t *
435 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
436 {
437 	pool_entry_t *pe;
438 	vdev_entry_t *ve;
439 	config_entry_t *ce;
440 	nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
441 	nvlist_t **spares, **l2cache;
442 	uint_t i, nspares, nl2cache;
443 	boolean_t config_seen;
444 	uint64_t best_txg;
445 	char *name, *hostname;
446 	uint64_t guid;
447 	uint_t children = 0;
448 	nvlist_t **child = NULL;
449 	uint_t holes;
450 	uint64_t *hole_array, max_id;
451 	uint_t c;
452 	boolean_t isactive;
453 	uint64_t hostid;
454 	nvlist_t *nvl;
455 	boolean_t found_one = B_FALSE;
456 	boolean_t valid_top_config = B_FALSE;
457 
458 	if (nvlist_alloc(&ret, 0, 0) != 0)
459 		goto nomem;
460 
461 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
462 		uint64_t id, max_txg = 0;
463 
464 		if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
465 			goto nomem;
466 		config_seen = B_FALSE;
467 
468 		/*
469 		 * Iterate over all toplevel vdevs.  Grab the pool configuration
470 		 * from the first one we find, and then go through the rest and
471 		 * add them as necessary to the 'vdevs' member of the config.
472 		 */
473 		for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
474 
475 			/*
476 			 * Determine the best configuration for this vdev by
477 			 * selecting the config with the latest transaction
478 			 * group.
479 			 */
480 			best_txg = 0;
481 			for (ce = ve->ve_configs; ce != NULL;
482 			    ce = ce->ce_next) {
483 
484 				if (ce->ce_txg > best_txg) {
485 					tmp = ce->ce_config;
486 					best_txg = ce->ce_txg;
487 				}
488 			}
489 
490 			/*
491 			 * We rely on the fact that the max txg for the
492 			 * pool will contain the most up-to-date information
493 			 * about the valid top-levels in the vdev namespace.
494 			 */
495 			if (best_txg > max_txg) {
496 				(void) nvlist_remove(config,
497 				    ZPOOL_CONFIG_VDEV_CHILDREN,
498 				    DATA_TYPE_UINT64);
499 				(void) nvlist_remove(config,
500 				    ZPOOL_CONFIG_HOLE_ARRAY,
501 				    DATA_TYPE_UINT64_ARRAY);
502 
503 				max_txg = best_txg;
504 				hole_array = NULL;
505 				holes = 0;
506 				max_id = 0;
507 				valid_top_config = B_FALSE;
508 
509 				if (nvlist_lookup_uint64(tmp,
510 				    ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
511 					verify(nvlist_add_uint64(config,
512 					    ZPOOL_CONFIG_VDEV_CHILDREN,
513 					    max_id) == 0);
514 					valid_top_config = B_TRUE;
515 				}
516 
517 				if (nvlist_lookup_uint64_array(tmp,
518 				    ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
519 				    &holes) == 0) {
520 					verify(nvlist_add_uint64_array(config,
521 					    ZPOOL_CONFIG_HOLE_ARRAY,
522 					    hole_array, holes) == 0);
523 				}
524 			}
525 
526 			if (!config_seen) {
527 				/*
528 				 * Copy the relevant pieces of data to the pool
529 				 * configuration:
530 				 *
531 				 *	version
532 				 *	pool guid
533 				 *	name
534 				 *	comment (if available)
535 				 *	pool state
536 				 *	hostid (if available)
537 				 *	hostname (if available)
538 				 */
539 				uint64_t state, version;
540 				char *comment = NULL;
541 
542 				version = fnvlist_lookup_uint64(tmp,
543 				    ZPOOL_CONFIG_VERSION);
544 				fnvlist_add_uint64(config,
545 				    ZPOOL_CONFIG_VERSION, version);
546 				guid = fnvlist_lookup_uint64(tmp,
547 				    ZPOOL_CONFIG_POOL_GUID);
548 				fnvlist_add_uint64(config,
549 				    ZPOOL_CONFIG_POOL_GUID, guid);
550 				name = fnvlist_lookup_string(tmp,
551 				    ZPOOL_CONFIG_POOL_NAME);
552 				fnvlist_add_string(config,
553 				    ZPOOL_CONFIG_POOL_NAME, name);
554 
555 				if (nvlist_lookup_string(tmp,
556 				    ZPOOL_CONFIG_COMMENT, &comment) == 0)
557 					fnvlist_add_string(config,
558 					    ZPOOL_CONFIG_COMMENT, comment);
559 
560 				state = fnvlist_lookup_uint64(tmp,
561 				    ZPOOL_CONFIG_POOL_STATE);
562 				fnvlist_add_uint64(config,
563 				    ZPOOL_CONFIG_POOL_STATE, state);
564 
565 				hostid = 0;
566 				if (nvlist_lookup_uint64(tmp,
567 				    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
568 					fnvlist_add_uint64(config,
569 					    ZPOOL_CONFIG_HOSTID, hostid);
570 					hostname = fnvlist_lookup_string(tmp,
571 					    ZPOOL_CONFIG_HOSTNAME);
572 					fnvlist_add_string(config,
573 					    ZPOOL_CONFIG_HOSTNAME, hostname);
574 				}
575 
576 				config_seen = B_TRUE;
577 			}
578 
579 			/*
580 			 * Add this top-level vdev to the child array.
581 			 */
582 			verify(nvlist_lookup_nvlist(tmp,
583 			    ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
584 			verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
585 			    &id) == 0);
586 
587 			if (id >= children) {
588 				nvlist_t **newchild;
589 
590 				newchild = zfs_alloc(hdl, (id + 1) *
591 				    sizeof (nvlist_t *));
592 				if (newchild == NULL)
593 					goto nomem;
594 
595 				for (c = 0; c < children; c++)
596 					newchild[c] = child[c];
597 
598 				free(child);
599 				child = newchild;
600 				children = id + 1;
601 			}
602 			if (nvlist_dup(nvtop, &child[id], 0) != 0)
603 				goto nomem;
604 
605 		}
606 
607 		/*
608 		 * If we have information about all the top-levels then
609 		 * clean up the nvlist which we've constructed. This
610 		 * means removing any extraneous devices that are
611 		 * beyond the valid range or adding devices to the end
612 		 * of our array which appear to be missing.
613 		 */
614 		if (valid_top_config) {
615 			if (max_id < children) {
616 				for (c = max_id; c < children; c++)
617 					nvlist_free(child[c]);
618 				children = max_id;
619 			} else if (max_id > children) {
620 				nvlist_t **newchild;
621 
622 				newchild = zfs_alloc(hdl, (max_id) *
623 				    sizeof (nvlist_t *));
624 				if (newchild == NULL)
625 					goto nomem;
626 
627 				for (c = 0; c < children; c++)
628 					newchild[c] = child[c];
629 
630 				free(child);
631 				child = newchild;
632 				children = max_id;
633 			}
634 		}
635 
636 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
637 		    &guid) == 0);
638 
639 		/*
640 		 * The vdev namespace may contain holes as a result of
641 		 * device removal. We must add them back into the vdev
642 		 * tree before we process any missing devices.
643 		 */
644 		if (holes > 0) {
645 			ASSERT(valid_top_config);
646 
647 			for (c = 0; c < children; c++) {
648 				nvlist_t *holey;
649 
650 				if (child[c] != NULL ||
651 				    !vdev_is_hole(hole_array, holes, c))
652 					continue;
653 
654 				if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
655 				    0) != 0)
656 					goto nomem;
657 
658 				/*
659 				 * Holes in the namespace are treated as
660 				 * "hole" top-level vdevs and have a
661 				 * special flag set on them.
662 				 */
663 				if (nvlist_add_string(holey,
664 				    ZPOOL_CONFIG_TYPE,
665 				    VDEV_TYPE_HOLE) != 0 ||
666 				    nvlist_add_uint64(holey,
667 				    ZPOOL_CONFIG_ID, c) != 0 ||
668 				    nvlist_add_uint64(holey,
669 				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
670 					nvlist_free(holey);
671 					goto nomem;
672 				}
673 				child[c] = holey;
674 			}
675 		}
676 
677 		/*
678 		 * Look for any missing top-level vdevs.  If this is the case,
679 		 * create a faked up 'missing' vdev as a placeholder.  We cannot
680 		 * simply compress the child array, because the kernel performs
681 		 * certain checks to make sure the vdev IDs match their location
682 		 * in the configuration.
683 		 */
684 		for (c = 0; c < children; c++) {
685 			if (child[c] == NULL) {
686 				nvlist_t *missing;
687 				if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
688 				    0) != 0)
689 					goto nomem;
690 				if (nvlist_add_string(missing,
691 				    ZPOOL_CONFIG_TYPE,
692 				    VDEV_TYPE_MISSING) != 0 ||
693 				    nvlist_add_uint64(missing,
694 				    ZPOOL_CONFIG_ID, c) != 0 ||
695 				    nvlist_add_uint64(missing,
696 				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
697 					nvlist_free(missing);
698 					goto nomem;
699 				}
700 				child[c] = missing;
701 			}
702 		}
703 
704 		/*
705 		 * Put all of this pool's top-level vdevs into a root vdev.
706 		 */
707 		if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
708 			goto nomem;
709 		if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
710 		    VDEV_TYPE_ROOT) != 0 ||
711 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
712 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
713 		    nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
714 		    child, children) != 0) {
715 			nvlist_free(nvroot);
716 			goto nomem;
717 		}
718 
719 		for (c = 0; c < children; c++)
720 			nvlist_free(child[c]);
721 		free(child);
722 		children = 0;
723 		child = NULL;
724 
725 		/*
726 		 * Go through and fix up any paths and/or devids based on our
727 		 * known list of vdev GUID -> path mappings.
728 		 */
729 		if (fix_paths(nvroot, pl->names) != 0) {
730 			nvlist_free(nvroot);
731 			goto nomem;
732 		}
733 
734 		/*
735 		 * Add the root vdev to this pool's configuration.
736 		 */
737 		if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
738 		    nvroot) != 0) {
739 			nvlist_free(nvroot);
740 			goto nomem;
741 		}
742 		nvlist_free(nvroot);
743 
744 		/*
745 		 * zdb uses this path to report on active pools that were
746 		 * imported or created using -R.
747 		 */
748 		if (active_ok)
749 			goto add_pool;
750 
751 		/*
752 		 * Determine if this pool is currently active, in which case we
753 		 * can't actually import it.
754 		 */
755 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
756 		    &name) == 0);
757 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
758 		    &guid) == 0);
759 
760 		if (pool_active(hdl, name, guid, &isactive) != 0)
761 			goto error;
762 
763 		if (isactive) {
764 			nvlist_free(config);
765 			config = NULL;
766 			continue;
767 		}
768 
769 		if ((nvl = refresh_config(hdl, config)) == NULL) {
770 			nvlist_free(config);
771 			config = NULL;
772 			continue;
773 		}
774 
775 		nvlist_free(config);
776 		config = nvl;
777 
778 		/*
779 		 * Go through and update the paths for spares, now that we have
780 		 * them.
781 		 */
782 		verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
783 		    &nvroot) == 0);
784 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
785 		    &spares, &nspares) == 0) {
786 			for (i = 0; i < nspares; i++) {
787 				if (fix_paths(spares[i], pl->names) != 0)
788 					goto nomem;
789 			}
790 		}
791 
792 		/*
793 		 * Update the paths for l2cache devices.
794 		 */
795 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
796 		    &l2cache, &nl2cache) == 0) {
797 			for (i = 0; i < nl2cache; i++) {
798 				if (fix_paths(l2cache[i], pl->names) != 0)
799 					goto nomem;
800 			}
801 		}
802 
803 		/*
804 		 * Restore the original information read from the actual label.
805 		 */
806 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
807 		    DATA_TYPE_UINT64);
808 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
809 		    DATA_TYPE_STRING);
810 		if (hostid != 0) {
811 			verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
812 			    hostid) == 0);
813 			verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
814 			    hostname) == 0);
815 		}
816 
817 add_pool:
818 		/*
819 		 * Add this pool to the list of configs.
820 		 */
821 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
822 		    &name) == 0);
823 		if (nvlist_add_nvlist(ret, name, config) != 0)
824 			goto nomem;
825 
826 		found_one = B_TRUE;
827 		nvlist_free(config);
828 		config = NULL;
829 	}
830 
831 	if (!found_one) {
832 		nvlist_free(ret);
833 		ret = NULL;
834 	}
835 
836 	return (ret);
837 
838 nomem:
839 	(void) no_memory(hdl);
840 error:
841 	nvlist_free(config);
842 	nvlist_free(ret);
843 	for (c = 0; c < children; c++)
844 		nvlist_free(child[c]);
845 	free(child);
846 
847 	return (NULL);
848 }
849 
850 /*
851  * Return the offset of the given label.
852  */
853 static uint64_t
854 label_offset(uint64_t size, int l)
855 {
856 	ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
857 	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
858 	    0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
859 }
860 
861 /*
862  * Given a file descriptor, read the label information and return an nvlist
863  * describing the configuration, if there is one.
864  */
865 int
866 zpool_read_label(int fd, nvlist_t **config)
867 {
868 	struct stat64 statbuf;
869 	int l;
870 	vdev_label_t *label;
871 	uint64_t state, txg, size;
872 
873 	*config = NULL;
874 
875 	if (fstat64(fd, &statbuf) == -1)
876 		return (0);
877 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
878 
879 	if ((label = malloc(sizeof (vdev_label_t))) == NULL)
880 		return (-1);
881 
882 	for (l = 0; l < VDEV_LABELS; l++) {
883 		if (pread64(fd, label, sizeof (vdev_label_t),
884 		    label_offset(size, l)) != sizeof (vdev_label_t))
885 			continue;
886 
887 		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
888 		    sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
889 			continue;
890 
891 		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
892 		    &state) != 0 || state > POOL_STATE_L2CACHE) {
893 			nvlist_free(*config);
894 			continue;
895 		}
896 
897 		if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
898 		    (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
899 		    &txg) != 0 || txg == 0)) {
900 			nvlist_free(*config);
901 			continue;
902 		}
903 
904 		free(label);
905 		return (0);
906 	}
907 
908 	free(label);
909 	*config = NULL;
910 	return (0);
911 }
912 
913 typedef struct rdsk_node {
914 	char *rn_name;
915 	int rn_dfd;
916 	libzfs_handle_t *rn_hdl;
917 	nvlist_t *rn_config;
918 	avl_tree_t *rn_avl;
919 	avl_node_t rn_node;
920 	boolean_t rn_nozpool;
921 } rdsk_node_t;
922 
923 static int
924 slice_cache_compare(const void *arg1, const void *arg2)
925 {
926 	const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
927 	const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
928 	char *nm1slice, *nm2slice;
929 	int rv;
930 
931 	/*
932 	 * slices zero and two are the most likely to provide results,
933 	 * so put those first
934 	 */
935 	nm1slice = strstr(nm1, "s0");
936 	nm2slice = strstr(nm2, "s0");
937 	if (nm1slice && !nm2slice) {
938 		return (-1);
939 	}
940 	if (!nm1slice && nm2slice) {
941 		return (1);
942 	}
943 	nm1slice = strstr(nm1, "s2");
944 	nm2slice = strstr(nm2, "s2");
945 	if (nm1slice && !nm2slice) {
946 		return (-1);
947 	}
948 	if (!nm1slice && nm2slice) {
949 		return (1);
950 	}
951 
952 	rv = strcmp(nm1, nm2);
953 	if (rv == 0)
954 		return (0);
955 	return (rv > 0 ? 1 : -1);
956 }
957 
958 static void
959 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
960     diskaddr_t size, uint_t blksz)
961 {
962 	rdsk_node_t tmpnode;
963 	rdsk_node_t *node;
964 	char sname[MAXNAMELEN];
965 
966 	tmpnode.rn_name = &sname[0];
967 	(void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
968 	    diskname, partno);
969 	/*
970 	 * protect against division by zero for disk labels that
971 	 * contain a bogus sector size
972 	 */
973 	if (blksz == 0)
974 		blksz = DEV_BSIZE;
975 	/* too small to contain a zpool? */
976 	if ((size < (SPA_MINDEVSIZE / blksz)) &&
977 	    (node = avl_find(r, &tmpnode, NULL)))
978 		node->rn_nozpool = B_TRUE;
979 }
980 
981 static void
982 nozpool_all_slices(avl_tree_t *r, const char *sname)
983 {
984 	char diskname[MAXNAMELEN];
985 	char *ptr;
986 	int i;
987 
988 	(void) strncpy(diskname, sname, MAXNAMELEN);
989 	if (((ptr = strrchr(diskname, 's')) == NULL) &&
990 	    ((ptr = strrchr(diskname, 'p')) == NULL))
991 		return;
992 	ptr[0] = 's';
993 	ptr[1] = '\0';
994 	for (i = 0; i < NDKMAP; i++)
995 		check_one_slice(r, diskname, i, 0, 1);
996 	ptr[0] = 'p';
997 	for (i = 0; i <= FD_NUMPART; i++)
998 		check_one_slice(r, diskname, i, 0, 1);
999 }
1000 
1001 static void
1002 check_slices(avl_tree_t *r, int fd, const char *sname)
1003 {
1004 	struct extvtoc vtoc;
1005 	struct dk_gpt *gpt;
1006 	char diskname[MAXNAMELEN];
1007 	char *ptr;
1008 	int i;
1009 
1010 	(void) strncpy(diskname, sname, MAXNAMELEN);
1011 	if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1012 		return;
1013 	ptr[1] = '\0';
1014 
1015 	if (read_extvtoc(fd, &vtoc) >= 0) {
1016 		for (i = 0; i < NDKMAP; i++)
1017 			check_one_slice(r, diskname, i,
1018 			    vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1019 	} else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1020 		/*
1021 		 * on x86 we'll still have leftover links that point
1022 		 * to slices s[9-15], so use NDKMAP instead
1023 		 */
1024 		for (i = 0; i < NDKMAP; i++)
1025 			check_one_slice(r, diskname, i,
1026 			    gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1027 		/* nodes p[1-4] are never used with EFI labels */
1028 		ptr[0] = 'p';
1029 		for (i = 1; i <= FD_NUMPART; i++)
1030 			check_one_slice(r, diskname, i, 0, 1);
1031 		efi_free(gpt);
1032 	}
1033 }
1034 
1035 static void
1036 zpool_open_func(void *arg)
1037 {
1038 	rdsk_node_t *rn = arg;
1039 	struct stat64 statbuf;
1040 	nvlist_t *config;
1041 	int fd;
1042 
1043 	if (rn->rn_nozpool)
1044 		return;
1045 	if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1046 		/* symlink to a device that's no longer there */
1047 		if (errno == ENOENT)
1048 			nozpool_all_slices(rn->rn_avl, rn->rn_name);
1049 		return;
1050 	}
1051 	/*
1052 	 * Ignore failed stats.  We only want regular
1053 	 * files, character devs and block devs.
1054 	 */
1055 	if (fstat64(fd, &statbuf) != 0 ||
1056 	    (!S_ISREG(statbuf.st_mode) &&
1057 	    !S_ISCHR(statbuf.st_mode) &&
1058 	    !S_ISBLK(statbuf.st_mode))) {
1059 		(void) close(fd);
1060 		return;
1061 	}
1062 	/* this file is too small to hold a zpool */
1063 	if (S_ISREG(statbuf.st_mode) &&
1064 	    statbuf.st_size < SPA_MINDEVSIZE) {
1065 		(void) close(fd);
1066 		return;
1067 	} else if (!S_ISREG(statbuf.st_mode)) {
1068 		/*
1069 		 * Try to read the disk label first so we don't have to
1070 		 * open a bunch of minor nodes that can't have a zpool.
1071 		 */
1072 		check_slices(rn->rn_avl, fd, rn->rn_name);
1073 	}
1074 
1075 	if ((zpool_read_label(fd, &config)) != 0) {
1076 		(void) close(fd);
1077 		(void) no_memory(rn->rn_hdl);
1078 		return;
1079 	}
1080 	(void) close(fd);
1081 
1082 
1083 	rn->rn_config = config;
1084 	if (config != NULL) {
1085 		assert(rn->rn_nozpool == B_FALSE);
1086 	}
1087 }
1088 
1089 /*
1090  * Given a file descriptor, clear (zero) the label information.  This function
1091  * is currently only used in the appliance stack as part of the ZFS sysevent
1092  * module.
1093  */
1094 int
1095 zpool_clear_label(int fd)
1096 {
1097 	struct stat64 statbuf;
1098 	int l;
1099 	vdev_label_t *label;
1100 	uint64_t size;
1101 
1102 	if (fstat64(fd, &statbuf) == -1)
1103 		return (0);
1104 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1105 
1106 	if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1107 		return (-1);
1108 
1109 	for (l = 0; l < VDEV_LABELS; l++) {
1110 		if (pwrite64(fd, label, sizeof (vdev_label_t),
1111 		    label_offset(size, l)) != sizeof (vdev_label_t)) {
1112 			free(label);
1113 			return (-1);
1114 		}
1115 	}
1116 
1117 	free(label);
1118 	return (0);
1119 }
1120 
1121 /*
1122  * Given a list of directories to search, find all pools stored on disk.  This
1123  * includes partial pools which are not available to import.  If no args are
1124  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1125  * poolname or guid (but not both) are provided by the caller when trying
1126  * to import a specific pool.
1127  */
1128 static nvlist_t *
1129 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1130 {
1131 	int i, dirs = iarg->paths;
1132 	struct dirent64 *dp;
1133 	char path[MAXPATHLEN];
1134 	char *end, **dir = iarg->path;
1135 	size_t pathleft;
1136 	nvlist_t *ret = NULL;
1137 	static char *default_dir = "/dev/dsk";
1138 	pool_list_t pools = { 0 };
1139 	pool_entry_t *pe, *penext;
1140 	vdev_entry_t *ve, *venext;
1141 	config_entry_t *ce, *cenext;
1142 	name_entry_t *ne, *nenext;
1143 	avl_tree_t slice_cache;
1144 	rdsk_node_t *slice;
1145 	void *cookie;
1146 
1147 	if (dirs == 0) {
1148 		dirs = 1;
1149 		dir = &default_dir;
1150 	}
1151 
1152 	/*
1153 	 * Go through and read the label configuration information from every
1154 	 * possible device, organizing the information according to pool GUID
1155 	 * and toplevel GUID.
1156 	 */
1157 	for (i = 0; i < dirs; i++) {
1158 		tpool_t *t;
1159 		char *rdsk;
1160 		int dfd;
1161 		boolean_t config_failed = B_FALSE;
1162 		DIR *dirp;
1163 
1164 		/* use realpath to normalize the path */
1165 		if (realpath(dir[i], path) == 0) {
1166 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1167 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1168 			goto error;
1169 		}
1170 		end = &path[strlen(path)];
1171 		*end++ = '/';
1172 		*end = 0;
1173 		pathleft = &path[sizeof (path)] - end;
1174 
1175 		/*
1176 		 * Using raw devices instead of block devices when we're
1177 		 * reading the labels skips a bunch of slow operations during
1178 		 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1179 		 */
1180 		if (strcmp(path, "/dev/dsk/") == 0)
1181 			rdsk = "/dev/rdsk/";
1182 		else
1183 			rdsk = path;
1184 
1185 		if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1186 		    (dirp = fdopendir(dfd)) == NULL) {
1187 			if (dfd >= 0)
1188 				(void) close(dfd);
1189 			zfs_error_aux(hdl, strerror(errno));
1190 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1191 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1192 			    rdsk);
1193 			goto error;
1194 		}
1195 
1196 		avl_create(&slice_cache, slice_cache_compare,
1197 		    sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1198 		/*
1199 		 * This is not MT-safe, but we have no MT consumers of libzfs
1200 		 */
1201 		while ((dp = readdir64(dirp)) != NULL) {
1202 			const char *name = dp->d_name;
1203 			if (name[0] == '.' &&
1204 			    (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1205 				continue;
1206 
1207 			slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1208 			slice->rn_name = zfs_strdup(hdl, name);
1209 			slice->rn_avl = &slice_cache;
1210 			slice->rn_dfd = dfd;
1211 			slice->rn_hdl = hdl;
1212 			slice->rn_nozpool = B_FALSE;
1213 			avl_add(&slice_cache, slice);
1214 		}
1215 		/*
1216 		 * create a thread pool to do all of this in parallel;
1217 		 * rn_nozpool is not protected, so this is racy in that
1218 		 * multiple tasks could decide that the same slice can
1219 		 * not hold a zpool, which is benign.  Also choose
1220 		 * double the number of processors; we hold a lot of
1221 		 * locks in the kernel, so going beyond this doesn't
1222 		 * buy us much.
1223 		 */
1224 		t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1225 		    0, NULL);
1226 		for (slice = avl_first(&slice_cache); slice;
1227 		    (slice = avl_walk(&slice_cache, slice,
1228 		    AVL_AFTER)))
1229 			(void) tpool_dispatch(t, zpool_open_func, slice);
1230 		tpool_wait(t);
1231 		tpool_destroy(t);
1232 
1233 		cookie = NULL;
1234 		while ((slice = avl_destroy_nodes(&slice_cache,
1235 		    &cookie)) != NULL) {
1236 			if (slice->rn_config != NULL && !config_failed) {
1237 				nvlist_t *config = slice->rn_config;
1238 				boolean_t matched = B_TRUE;
1239 
1240 				if (iarg->poolname != NULL) {
1241 					char *pname;
1242 
1243 					matched = nvlist_lookup_string(config,
1244 					    ZPOOL_CONFIG_POOL_NAME,
1245 					    &pname) == 0 &&
1246 					    strcmp(iarg->poolname, pname) == 0;
1247 				} else if (iarg->guid != 0) {
1248 					uint64_t this_guid;
1249 
1250 					matched = nvlist_lookup_uint64(config,
1251 					    ZPOOL_CONFIG_POOL_GUID,
1252 					    &this_guid) == 0 &&
1253 					    iarg->guid == this_guid;
1254 				}
1255 				if (!matched) {
1256 					nvlist_free(config);
1257 				} else {
1258 					/*
1259 					 * use the non-raw path for the config
1260 					 */
1261 					(void) strlcpy(end, slice->rn_name,
1262 					    pathleft);
1263 					if (add_config(hdl, &pools, path,
1264 					    config) != 0)
1265 						config_failed = B_TRUE;
1266 				}
1267 			}
1268 			free(slice->rn_name);
1269 			free(slice);
1270 		}
1271 		avl_destroy(&slice_cache);
1272 
1273 		(void) closedir(dirp);
1274 
1275 		if (config_failed)
1276 			goto error;
1277 	}
1278 
1279 	ret = get_configs(hdl, &pools, iarg->can_be_active);
1280 
1281 error:
1282 	for (pe = pools.pools; pe != NULL; pe = penext) {
1283 		penext = pe->pe_next;
1284 		for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1285 			venext = ve->ve_next;
1286 			for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1287 				cenext = ce->ce_next;
1288 				if (ce->ce_config)
1289 					nvlist_free(ce->ce_config);
1290 				free(ce);
1291 			}
1292 			free(ve);
1293 		}
1294 		free(pe);
1295 	}
1296 
1297 	for (ne = pools.names; ne != NULL; ne = nenext) {
1298 		nenext = ne->ne_next;
1299 		free(ne->ne_name);
1300 		free(ne);
1301 	}
1302 
1303 	return (ret);
1304 }
1305 
1306 nvlist_t *
1307 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1308 {
1309 	importargs_t iarg = { 0 };
1310 
1311 	iarg.paths = argc;
1312 	iarg.path = argv;
1313 
1314 	return (zpool_find_import_impl(hdl, &iarg));
1315 }
1316 
1317 /*
1318  * Given a cache file, return the contents as a list of importable pools.
1319  * poolname or guid (but not both) are provided by the caller when trying
1320  * to import a specific pool.
1321  */
1322 nvlist_t *
1323 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1324     char *poolname, uint64_t guid)
1325 {
1326 	char *buf;
1327 	int fd;
1328 	struct stat64 statbuf;
1329 	nvlist_t *raw, *src, *dst;
1330 	nvlist_t *pools;
1331 	nvpair_t *elem;
1332 	char *name;
1333 	uint64_t this_guid;
1334 	boolean_t active;
1335 
1336 	verify(poolname == NULL || guid == 0);
1337 
1338 	if ((fd = open(cachefile, O_RDONLY)) < 0) {
1339 		zfs_error_aux(hdl, "%s", strerror(errno));
1340 		(void) zfs_error(hdl, EZFS_BADCACHE,
1341 		    dgettext(TEXT_DOMAIN, "failed to open cache file"));
1342 		return (NULL);
1343 	}
1344 
1345 	if (fstat64(fd, &statbuf) != 0) {
1346 		zfs_error_aux(hdl, "%s", strerror(errno));
1347 		(void) close(fd);
1348 		(void) zfs_error(hdl, EZFS_BADCACHE,
1349 		    dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1350 		return (NULL);
1351 	}
1352 
1353 	if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1354 		(void) close(fd);
1355 		return (NULL);
1356 	}
1357 
1358 	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1359 		(void) close(fd);
1360 		free(buf);
1361 		(void) zfs_error(hdl, EZFS_BADCACHE,
1362 		    dgettext(TEXT_DOMAIN,
1363 		    "failed to read cache file contents"));
1364 		return (NULL);
1365 	}
1366 
1367 	(void) close(fd);
1368 
1369 	if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1370 		free(buf);
1371 		(void) zfs_error(hdl, EZFS_BADCACHE,
1372 		    dgettext(TEXT_DOMAIN,
1373 		    "invalid or corrupt cache file contents"));
1374 		return (NULL);
1375 	}
1376 
1377 	free(buf);
1378 
1379 	/*
1380 	 * Go through and get the current state of the pools and refresh their
1381 	 * state.
1382 	 */
1383 	if (nvlist_alloc(&pools, 0, 0) != 0) {
1384 		(void) no_memory(hdl);
1385 		nvlist_free(raw);
1386 		return (NULL);
1387 	}
1388 
1389 	elem = NULL;
1390 	while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1391 		src = fnvpair_value_nvlist(elem);
1392 
1393 		name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1394 		if (poolname != NULL && strcmp(poolname, name) != 0)
1395 			continue;
1396 
1397 		this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1398 		if (guid != 0 && guid != this_guid)
1399 			continue;
1400 
1401 		if (pool_active(hdl, name, this_guid, &active) != 0) {
1402 			nvlist_free(raw);
1403 			nvlist_free(pools);
1404 			return (NULL);
1405 		}
1406 
1407 		if (active)
1408 			continue;
1409 
1410 		if ((dst = refresh_config(hdl, src)) == NULL) {
1411 			nvlist_free(raw);
1412 			nvlist_free(pools);
1413 			return (NULL);
1414 		}
1415 
1416 		if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1417 			(void) no_memory(hdl);
1418 			nvlist_free(dst);
1419 			nvlist_free(raw);
1420 			nvlist_free(pools);
1421 			return (NULL);
1422 		}
1423 		nvlist_free(dst);
1424 	}
1425 
1426 	nvlist_free(raw);
1427 	return (pools);
1428 }
1429 
1430 static int
1431 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1432 {
1433 	importargs_t *import = data;
1434 	int found = 0;
1435 
1436 	if (import->poolname != NULL) {
1437 		char *pool_name;
1438 
1439 		verify(nvlist_lookup_string(zhp->zpool_config,
1440 		    ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1441 		if (strcmp(pool_name, import->poolname) == 0)
1442 			found = 1;
1443 	} else {
1444 		uint64_t pool_guid;
1445 
1446 		verify(nvlist_lookup_uint64(zhp->zpool_config,
1447 		    ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1448 		if (pool_guid == import->guid)
1449 			found = 1;
1450 	}
1451 
1452 	zpool_close(zhp);
1453 	return (found);
1454 }
1455 
1456 nvlist_t *
1457 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1458 {
1459 	verify(import->poolname == NULL || import->guid == 0);
1460 
1461 	if (import->unique)
1462 		import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1463 
1464 	if (import->cachefile != NULL)
1465 		return (zpool_find_import_cached(hdl, import->cachefile,
1466 		    import->poolname, import->guid));
1467 
1468 	return (zpool_find_import_impl(hdl, import));
1469 }
1470 
1471 boolean_t
1472 find_guid(nvlist_t *nv, uint64_t guid)
1473 {
1474 	uint64_t tmp;
1475 	nvlist_t **child;
1476 	uint_t c, children;
1477 
1478 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1479 	if (tmp == guid)
1480 		return (B_TRUE);
1481 
1482 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1483 	    &child, &children) == 0) {
1484 		for (c = 0; c < children; c++)
1485 			if (find_guid(child[c], guid))
1486 				return (B_TRUE);
1487 	}
1488 
1489 	return (B_FALSE);
1490 }
1491 
1492 typedef struct aux_cbdata {
1493 	const char	*cb_type;
1494 	uint64_t	cb_guid;
1495 	zpool_handle_t	*cb_zhp;
1496 } aux_cbdata_t;
1497 
1498 static int
1499 find_aux(zpool_handle_t *zhp, void *data)
1500 {
1501 	aux_cbdata_t *cbp = data;
1502 	nvlist_t **list;
1503 	uint_t i, count;
1504 	uint64_t guid;
1505 	nvlist_t *nvroot;
1506 
1507 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1508 	    &nvroot) == 0);
1509 
1510 	if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1511 	    &list, &count) == 0) {
1512 		for (i = 0; i < count; i++) {
1513 			verify(nvlist_lookup_uint64(list[i],
1514 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1515 			if (guid == cbp->cb_guid) {
1516 				cbp->cb_zhp = zhp;
1517 				return (1);
1518 			}
1519 		}
1520 	}
1521 
1522 	zpool_close(zhp);
1523 	return (0);
1524 }
1525 
1526 /*
1527  * Determines if the pool is in use.  If so, it returns true and the state of
1528  * the pool as well as the name of the pool.  Both strings are allocated and
1529  * must be freed by the caller.
1530  */
1531 int
1532 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1533     boolean_t *inuse)
1534 {
1535 	nvlist_t *config;
1536 	char *name;
1537 	boolean_t ret;
1538 	uint64_t guid, vdev_guid;
1539 	zpool_handle_t *zhp;
1540 	nvlist_t *pool_config;
1541 	uint64_t stateval, isspare;
1542 	aux_cbdata_t cb = { 0 };
1543 	boolean_t isactive;
1544 
1545 	*inuse = B_FALSE;
1546 
1547 	if (zpool_read_label(fd, &config) != 0) {
1548 		(void) no_memory(hdl);
1549 		return (-1);
1550 	}
1551 
1552 	if (config == NULL)
1553 		return (0);
1554 
1555 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1556 	    &stateval) == 0);
1557 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1558 	    &vdev_guid) == 0);
1559 
1560 	if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1561 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1562 		    &name) == 0);
1563 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1564 		    &guid) == 0);
1565 	}
1566 
1567 	switch (stateval) {
1568 	case POOL_STATE_EXPORTED:
1569 		/*
1570 		 * A pool with an exported state may in fact be imported
1571 		 * read-only, so check the in-core state to see if it's
1572 		 * active and imported read-only.  If it is, set
1573 		 * its state to active.
1574 		 */
1575 		if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1576 		    (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1577 			if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1578 				stateval = POOL_STATE_ACTIVE;
1579 
1580 			/*
1581 			 * All we needed the zpool handle for is the
1582 			 * readonly prop check.
1583 			 */
1584 			zpool_close(zhp);
1585 		}
1586 
1587 		ret = B_TRUE;
1588 		break;
1589 
1590 	case POOL_STATE_ACTIVE:
1591 		/*
1592 		 * For an active pool, we have to determine if it's really part
1593 		 * of a currently active pool (in which case the pool will exist
1594 		 * and the guid will be the same), or whether it's part of an
1595 		 * active pool that was disconnected without being explicitly
1596 		 * exported.
1597 		 */
1598 		if (pool_active(hdl, name, guid, &isactive) != 0) {
1599 			nvlist_free(config);
1600 			return (-1);
1601 		}
1602 
1603 		if (isactive) {
1604 			/*
1605 			 * Because the device may have been removed while
1606 			 * offlined, we only report it as active if the vdev is
1607 			 * still present in the config.  Otherwise, pretend like
1608 			 * it's not in use.
1609 			 */
1610 			if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1611 			    (pool_config = zpool_get_config(zhp, NULL))
1612 			    != NULL) {
1613 				nvlist_t *nvroot;
1614 
1615 				verify(nvlist_lookup_nvlist(pool_config,
1616 				    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1617 				ret = find_guid(nvroot, vdev_guid);
1618 			} else {
1619 				ret = B_FALSE;
1620 			}
1621 
1622 			/*
1623 			 * If this is an active spare within another pool, we
1624 			 * treat it like an unused hot spare.  This allows the
1625 			 * user to create a pool with a hot spare that currently
1626 			 * in use within another pool.  Since we return B_TRUE,
1627 			 * libdiskmgt will continue to prevent generic consumers
1628 			 * from using the device.
1629 			 */
1630 			if (ret && nvlist_lookup_uint64(config,
1631 			    ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1632 				stateval = POOL_STATE_SPARE;
1633 
1634 			if (zhp != NULL)
1635 				zpool_close(zhp);
1636 		} else {
1637 			stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1638 			ret = B_TRUE;
1639 		}
1640 		break;
1641 
1642 	case POOL_STATE_SPARE:
1643 		/*
1644 		 * For a hot spare, it can be either definitively in use, or
1645 		 * potentially active.  To determine if it's in use, we iterate
1646 		 * over all pools in the system and search for one with a spare
1647 		 * with a matching guid.
1648 		 *
1649 		 * Due to the shared nature of spares, we don't actually report
1650 		 * the potentially active case as in use.  This means the user
1651 		 * can freely create pools on the hot spares of exported pools,
1652 		 * but to do otherwise makes the resulting code complicated, and
1653 		 * we end up having to deal with this case anyway.
1654 		 */
1655 		cb.cb_zhp = NULL;
1656 		cb.cb_guid = vdev_guid;
1657 		cb.cb_type = ZPOOL_CONFIG_SPARES;
1658 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1659 			name = (char *)zpool_get_name(cb.cb_zhp);
1660 			ret = B_TRUE;
1661 		} else {
1662 			ret = B_FALSE;
1663 		}
1664 		break;
1665 
1666 	case POOL_STATE_L2CACHE:
1667 
1668 		/*
1669 		 * Check if any pool is currently using this l2cache device.
1670 		 */
1671 		cb.cb_zhp = NULL;
1672 		cb.cb_guid = vdev_guid;
1673 		cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1674 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1675 			name = (char *)zpool_get_name(cb.cb_zhp);
1676 			ret = B_TRUE;
1677 		} else {
1678 			ret = B_FALSE;
1679 		}
1680 		break;
1681 
1682 	default:
1683 		ret = B_FALSE;
1684 	}
1685 
1686 
1687 	if (ret) {
1688 		if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1689 			if (cb.cb_zhp)
1690 				zpool_close(cb.cb_zhp);
1691 			nvlist_free(config);
1692 			return (-1);
1693 		}
1694 		*state = (pool_state_t)stateval;
1695 	}
1696 
1697 	if (cb.cb_zhp)
1698 		zpool_close(cb.cb_zhp);
1699 
1700 	nvlist_free(config);
1701 	*inuse = ret;
1702 	return (0);
1703 }
1704