1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
25  * Copyright 2015 RackTop Systems.
26  * Copyright 2016 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 <aio.h>
46 #include <ctype.h>
47 #include <devid.h>
48 #include <dirent.h>
49 #include <errno.h>
50 #include <libintl.h>
51 #include <stddef.h>
52 #include <stdlib.h>
53 #include <string.h>
54 #include <sys/stat.h>
55 #include <unistd.h>
56 #include <fcntl.h>
57 #include <thread_pool.h>
58 #include <libgeom.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 *
get_devid(const char * path)98 get_devid(const char *path)
99 {
100 #ifdef have_devid
101 	int fd;
102 	ddi_devid_t devid;
103 	char *minor, *ret;
104 
105 	if ((fd = open(path, O_RDONLY)) < 0)
106 		return (NULL);
107 
108 	minor = NULL;
109 	ret = NULL;
110 	if (devid_get(fd, &devid) == 0) {
111 		if (devid_get_minor_name(fd, &minor) == 0)
112 			ret = devid_str_encode(devid, minor);
113 		if (minor != NULL)
114 			devid_str_free(minor);
115 		devid_free(devid);
116 	}
117 	(void) close(fd);
118 
119 	return (ret);
120 #else
121 	return (NULL);
122 #endif
123 }
124 
125 
126 /*
127  * Go through and fix up any path and/or devid information for the given vdev
128  * configuration.
129  */
130 static int
fix_paths(nvlist_t * nv,name_entry_t * names)131 fix_paths(nvlist_t *nv, name_entry_t *names)
132 {
133 	nvlist_t **child;
134 	uint_t c, children;
135 	uint64_t guid;
136 	name_entry_t *ne, *best;
137 	char *path, *devid;
138 	int matched;
139 
140 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
141 	    &child, &children) == 0) {
142 		for (c = 0; c < children; c++)
143 			if (fix_paths(child[c], names) != 0)
144 				return (-1);
145 		return (0);
146 	}
147 
148 	/*
149 	 * This is a leaf (file or disk) vdev.  In either case, go through
150 	 * the name list and see if we find a matching guid.  If so, replace
151 	 * the path and see if we can calculate a new devid.
152 	 *
153 	 * There may be multiple names associated with a particular guid, in
154 	 * which case we have overlapping slices or multiple paths to the same
155 	 * disk.  If this is the case, then we want to pick the path that is
156 	 * the most similar to the original, where "most similar" is the number
157 	 * of matching characters starting from the end of the path.  This will
158 	 * preserve slice numbers even if the disks have been reorganized, and
159 	 * will also catch preferred disk names if multiple paths exist.
160 	 */
161 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
162 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
163 		path = NULL;
164 
165 	matched = 0;
166 	best = NULL;
167 	for (ne = names; ne != NULL; ne = ne->ne_next) {
168 		if (ne->ne_guid == guid) {
169 			const char *src, *dst;
170 			int count;
171 
172 			if (path == NULL) {
173 				best = ne;
174 				break;
175 			}
176 
177 			src = ne->ne_name + strlen(ne->ne_name) - 1;
178 			dst = path + strlen(path) - 1;
179 			for (count = 0; src >= ne->ne_name && dst >= path;
180 			    src--, dst--, count++)
181 				if (*src != *dst)
182 					break;
183 
184 			/*
185 			 * At this point, 'count' is the number of characters
186 			 * matched from the end.
187 			 */
188 			if (count > matched || best == NULL) {
189 				best = ne;
190 				matched = count;
191 			}
192 		}
193 	}
194 
195 	if (best == NULL)
196 		return (0);
197 
198 	if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
199 		return (-1);
200 
201 	if ((devid = get_devid(best->ne_name)) == NULL) {
202 		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
203 	} else {
204 		if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
205 			devid_str_free(devid);
206 			return (-1);
207 		}
208 		devid_str_free(devid);
209 	}
210 
211 	return (0);
212 }
213 
214 /*
215  * Add the given configuration to the list of known devices.
216  */
217 static int
add_config(libzfs_handle_t * hdl,pool_list_t * pl,const char * path,nvlist_t * config)218 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
219     nvlist_t *config)
220 {
221 	uint64_t pool_guid, vdev_guid, top_guid, txg, state;
222 	pool_entry_t *pe;
223 	vdev_entry_t *ve;
224 	config_entry_t *ce;
225 	name_entry_t *ne;
226 
227 	/*
228 	 * If this is a hot spare not currently in use or level 2 cache
229 	 * device, add it to the list of names to translate, but don't do
230 	 * anything else.
231 	 */
232 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
233 	    &state) == 0 &&
234 	    (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
235 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
236 		if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
237 			return (-1);
238 
239 		if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
240 			free(ne);
241 			return (-1);
242 		}
243 
244 		ne->ne_guid = vdev_guid;
245 		ne->ne_next = pl->names;
246 		pl->names = ne;
247 
248 		return (0);
249 	}
250 
251 	/*
252 	 * If we have a valid config but cannot read any of these fields, then
253 	 * it means we have a half-initialized label.  In vdev_label_init()
254 	 * we write a label with txg == 0 so that we can identify the device
255 	 * in case the user refers to the same disk later on.  If we fail to
256 	 * create the pool, we'll be left with a label in this state
257 	 * which should not be considered part of a valid pool.
258 	 */
259 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
260 	    &pool_guid) != 0 ||
261 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
262 	    &vdev_guid) != 0 ||
263 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
264 	    &top_guid) != 0 ||
265 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
266 	    &txg) != 0 || txg == 0) {
267 		return (0);
268 	}
269 
270 	/*
271 	 * First, see if we know about this pool.  If not, then add it to the
272 	 * list of known pools.
273 	 */
274 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
275 		if (pe->pe_guid == pool_guid)
276 			break;
277 	}
278 
279 	if (pe == NULL) {
280 		if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
281 			return (-1);
282 		}
283 		pe->pe_guid = pool_guid;
284 		pe->pe_next = pl->pools;
285 		pl->pools = pe;
286 	}
287 
288 	/*
289 	 * Second, see if we know about this toplevel vdev.  Add it if its
290 	 * missing.
291 	 */
292 	for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
293 		if (ve->ve_guid == top_guid)
294 			break;
295 	}
296 
297 	if (ve == NULL) {
298 		if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
299 			return (-1);
300 		}
301 		ve->ve_guid = top_guid;
302 		ve->ve_next = pe->pe_vdevs;
303 		pe->pe_vdevs = ve;
304 	}
305 
306 	/*
307 	 * Third, see if we have a config with a matching transaction group.  If
308 	 * so, then we do nothing.  Otherwise, add it to the list of known
309 	 * configs.
310 	 */
311 	for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
312 		if (ce->ce_txg == txg)
313 			break;
314 	}
315 
316 	if (ce == NULL) {
317 		if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
318 			return (-1);
319 		}
320 		ce->ce_txg = txg;
321 		ce->ce_config = fnvlist_dup(config);
322 		ce->ce_next = ve->ve_configs;
323 		ve->ve_configs = ce;
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
pool_active(libzfs_handle_t * hdl,const char * name,uint64_t guid,boolean_t * isactive)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 *
refresh_config(libzfs_handle_t * hdl,nvlist_t * config)375 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
376 {
377 	nvlist_t *nvl;
378 	zfs_cmd_t zc = { 0 };
379 	int err, dstbuf_size;
380 
381 	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
382 		return (NULL);
383 
384 	dstbuf_size = MAX(CONFIG_BUF_MINSIZE, zc.zc_nvlist_conf_size * 4);
385 
386 	if (zcmd_alloc_dst_nvlist(hdl, &zc, dstbuf_size) != 0) {
387 		zcmd_free_nvlists(&zc);
388 		return (NULL);
389 	}
390 
391 	while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
392 	    &zc)) != 0 && errno == ENOMEM) {
393 		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
394 			zcmd_free_nvlists(&zc);
395 			return (NULL);
396 		}
397 	}
398 
399 	if (err) {
400 		zcmd_free_nvlists(&zc);
401 		return (NULL);
402 	}
403 
404 	if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
405 		zcmd_free_nvlists(&zc);
406 		return (NULL);
407 	}
408 
409 	zcmd_free_nvlists(&zc);
410 	return (nvl);
411 }
412 
413 /*
414  * Determine if the vdev id is a hole in the namespace.
415  */
416 boolean_t
vdev_is_hole(uint64_t * hole_array,uint_t holes,uint_t id)417 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
418 {
419 	for (int c = 0; c < holes; c++) {
420 
421 		/* Top-level is a hole */
422 		if (hole_array[c] == id)
423 			return (B_TRUE);
424 	}
425 	return (B_FALSE);
426 }
427 
428 /*
429  * Convert our list of pools into the definitive set of configurations.  We
430  * start by picking the best config for each toplevel vdev.  Once that's done,
431  * we assemble the toplevel vdevs into a full config for the pool.  We make a
432  * pass to fix up any incorrect paths, and then add it to the main list to
433  * return to the user.
434  */
435 static nvlist_t *
get_configs(libzfs_handle_t * hdl,pool_list_t * pl,boolean_t active_ok,nvlist_t * policy)436 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok,
437     nvlist_t *policy)
438 {
439 	pool_entry_t *pe;
440 	vdev_entry_t *ve;
441 	config_entry_t *ce;
442 	nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
443 	nvlist_t **spares, **l2cache;
444 	uint_t i, nspares, nl2cache;
445 	boolean_t config_seen;
446 	uint64_t best_txg;
447 	char *name, *hostname = NULL;
448 	uint64_t guid;
449 	uint_t children = 0;
450 	nvlist_t **child = NULL;
451 	uint_t holes;
452 	uint64_t *hole_array, max_id;
453 	uint_t c;
454 	boolean_t isactive;
455 	uint64_t hostid;
456 	nvlist_t *nvl;
457 	boolean_t found_one = B_FALSE;
458 	boolean_t valid_top_config = B_FALSE;
459 
460 	if (nvlist_alloc(&ret, 0, 0) != 0)
461 		goto nomem;
462 
463 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
464 		uint64_t id, max_txg = 0;
465 
466 		if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
467 			goto nomem;
468 		config_seen = B_FALSE;
469 
470 		/*
471 		 * Iterate over all toplevel vdevs.  Grab the pool configuration
472 		 * from the first one we find, and then go through the rest and
473 		 * add them as necessary to the 'vdevs' member of the config.
474 		 */
475 		for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
476 
477 			/*
478 			 * Determine the best configuration for this vdev by
479 			 * selecting the config with the latest transaction
480 			 * group.
481 			 */
482 			best_txg = 0;
483 			for (ce = ve->ve_configs; ce != NULL;
484 			    ce = ce->ce_next) {
485 
486 				if (ce->ce_txg > best_txg) {
487 					tmp = ce->ce_config;
488 					best_txg = ce->ce_txg;
489 				}
490 			}
491 
492 			/*
493 			 * We rely on the fact that the max txg for the
494 			 * pool will contain the most up-to-date information
495 			 * about the valid top-levels in the vdev namespace.
496 			 */
497 			if (best_txg > max_txg) {
498 				(void) nvlist_remove(config,
499 				    ZPOOL_CONFIG_VDEV_CHILDREN,
500 				    DATA_TYPE_UINT64);
501 				(void) nvlist_remove(config,
502 				    ZPOOL_CONFIG_HOLE_ARRAY,
503 				    DATA_TYPE_UINT64_ARRAY);
504 
505 				max_txg = best_txg;
506 				hole_array = NULL;
507 				holes = 0;
508 				max_id = 0;
509 				valid_top_config = B_FALSE;
510 
511 				if (nvlist_lookup_uint64(tmp,
512 				    ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
513 					verify(nvlist_add_uint64(config,
514 					    ZPOOL_CONFIG_VDEV_CHILDREN,
515 					    max_id) == 0);
516 					valid_top_config = B_TRUE;
517 				}
518 
519 				if (nvlist_lookup_uint64_array(tmp,
520 				    ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
521 				    &holes) == 0) {
522 					verify(nvlist_add_uint64_array(config,
523 					    ZPOOL_CONFIG_HOLE_ARRAY,
524 					    hole_array, holes) == 0);
525 				}
526 			}
527 
528 			if (!config_seen) {
529 				/*
530 				 * Copy the relevant pieces of data to the pool
531 				 * configuration:
532 				 *
533 				 *	version
534 				 *	pool guid
535 				 *	name
536 				 *	comment (if available)
537 				 *	pool state
538 				 *	hostid (if available)
539 				 *	hostname (if available)
540 				 */
541 				uint64_t state, version;
542 				char *comment = NULL;
543 
544 				version = fnvlist_lookup_uint64(tmp,
545 				    ZPOOL_CONFIG_VERSION);
546 				fnvlist_add_uint64(config,
547 				    ZPOOL_CONFIG_VERSION, version);
548 				guid = fnvlist_lookup_uint64(tmp,
549 				    ZPOOL_CONFIG_POOL_GUID);
550 				fnvlist_add_uint64(config,
551 				    ZPOOL_CONFIG_POOL_GUID, guid);
552 				name = fnvlist_lookup_string(tmp,
553 				    ZPOOL_CONFIG_POOL_NAME);
554 				fnvlist_add_string(config,
555 				    ZPOOL_CONFIG_POOL_NAME, name);
556 
557 				if (nvlist_lookup_string(tmp,
558 				    ZPOOL_CONFIG_COMMENT, &comment) == 0)
559 					fnvlist_add_string(config,
560 					    ZPOOL_CONFIG_COMMENT, comment);
561 
562 				state = fnvlist_lookup_uint64(tmp,
563 				    ZPOOL_CONFIG_POOL_STATE);
564 				fnvlist_add_uint64(config,
565 				    ZPOOL_CONFIG_POOL_STATE, state);
566 
567 				hostid = 0;
568 				if (nvlist_lookup_uint64(tmp,
569 				    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
570 					fnvlist_add_uint64(config,
571 					    ZPOOL_CONFIG_HOSTID, hostid);
572 					hostname = fnvlist_lookup_string(tmp,
573 					    ZPOOL_CONFIG_HOSTNAME);
574 					fnvlist_add_string(config,
575 					    ZPOOL_CONFIG_HOSTNAME, hostname);
576 				}
577 
578 				config_seen = B_TRUE;
579 			}
580 
581 			/*
582 			 * Add this top-level vdev to the child array.
583 			 */
584 			verify(nvlist_lookup_nvlist(tmp,
585 			    ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
586 			verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
587 			    &id) == 0);
588 
589 			if (id >= children) {
590 				nvlist_t **newchild;
591 
592 				newchild = zfs_alloc(hdl, (id + 1) *
593 				    sizeof (nvlist_t *));
594 				if (newchild == NULL)
595 					goto nomem;
596 
597 				for (c = 0; c < children; c++)
598 					newchild[c] = child[c];
599 
600 				free(child);
601 				child = newchild;
602 				children = id + 1;
603 			}
604 			if (nvlist_dup(nvtop, &child[id], 0) != 0)
605 				goto nomem;
606 
607 		}
608 
609 		/*
610 		 * If we have information about all the top-levels then
611 		 * clean up the nvlist which we've constructed. This
612 		 * means removing any extraneous devices that are
613 		 * beyond the valid range or adding devices to the end
614 		 * of our array which appear to be missing.
615 		 */
616 		if (valid_top_config) {
617 			if (max_id < children) {
618 				for (c = max_id; c < children; c++)
619 					nvlist_free(child[c]);
620 				children = max_id;
621 			} else if (max_id > children) {
622 				nvlist_t **newchild;
623 
624 				newchild = zfs_alloc(hdl, (max_id) *
625 				    sizeof (nvlist_t *));
626 				if (newchild == NULL)
627 					goto nomem;
628 
629 				for (c = 0; c < children; c++)
630 					newchild[c] = child[c];
631 
632 				free(child);
633 				child = newchild;
634 				children = max_id;
635 			}
636 		}
637 
638 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
639 		    &guid) == 0);
640 
641 		/*
642 		 * The vdev namespace may contain holes as a result of
643 		 * device removal. We must add them back into the vdev
644 		 * tree before we process any missing devices.
645 		 */
646 		if (holes > 0) {
647 			ASSERT(valid_top_config);
648 
649 			for (c = 0; c < children; c++) {
650 				nvlist_t *holey;
651 
652 				if (child[c] != NULL ||
653 				    !vdev_is_hole(hole_array, holes, c))
654 					continue;
655 
656 				if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
657 				    0) != 0)
658 					goto nomem;
659 
660 				/*
661 				 * Holes in the namespace are treated as
662 				 * "hole" top-level vdevs and have a
663 				 * special flag set on them.
664 				 */
665 				if (nvlist_add_string(holey,
666 				    ZPOOL_CONFIG_TYPE,
667 				    VDEV_TYPE_HOLE) != 0 ||
668 				    nvlist_add_uint64(holey,
669 				    ZPOOL_CONFIG_ID, c) != 0 ||
670 				    nvlist_add_uint64(holey,
671 				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
672 					nvlist_free(holey);
673 					goto nomem;
674 				}
675 				child[c] = holey;
676 			}
677 		}
678 
679 		/*
680 		 * Look for any missing top-level vdevs.  If this is the case,
681 		 * create a faked up 'missing' vdev as a placeholder.  We cannot
682 		 * simply compress the child array, because the kernel performs
683 		 * certain checks to make sure the vdev IDs match their location
684 		 * in the configuration.
685 		 */
686 		for (c = 0; c < children; c++) {
687 			if (child[c] == NULL) {
688 				nvlist_t *missing;
689 				if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
690 				    0) != 0)
691 					goto nomem;
692 				if (nvlist_add_string(missing,
693 				    ZPOOL_CONFIG_TYPE,
694 				    VDEV_TYPE_MISSING) != 0 ||
695 				    nvlist_add_uint64(missing,
696 				    ZPOOL_CONFIG_ID, c) != 0 ||
697 				    nvlist_add_uint64(missing,
698 				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
699 					nvlist_free(missing);
700 					goto nomem;
701 				}
702 				child[c] = missing;
703 			}
704 		}
705 
706 		/*
707 		 * Put all of this pool's top-level vdevs into a root vdev.
708 		 */
709 		if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
710 			goto nomem;
711 		if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
712 		    VDEV_TYPE_ROOT) != 0 ||
713 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
714 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
715 		    nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
716 		    child, children) != 0) {
717 			nvlist_free(nvroot);
718 			goto nomem;
719 		}
720 
721 		for (c = 0; c < children; c++)
722 			nvlist_free(child[c]);
723 		free(child);
724 		children = 0;
725 		child = NULL;
726 
727 		/*
728 		 * Go through and fix up any paths and/or devids based on our
729 		 * known list of vdev GUID -> path mappings.
730 		 */
731 		if (fix_paths(nvroot, pl->names) != 0) {
732 			nvlist_free(nvroot);
733 			goto nomem;
734 		}
735 
736 		/*
737 		 * Add the root vdev to this pool's configuration.
738 		 */
739 		if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
740 		    nvroot) != 0) {
741 			nvlist_free(nvroot);
742 			goto nomem;
743 		}
744 		nvlist_free(nvroot);
745 
746 		/*
747 		 * zdb uses this path to report on active pools that were
748 		 * imported or created using -R.
749 		 */
750 		if (active_ok)
751 			goto add_pool;
752 
753 		/*
754 		 * Determine if this pool is currently active, in which case we
755 		 * can't actually import it.
756 		 */
757 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
758 		    &name) == 0);
759 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
760 		    &guid) == 0);
761 
762 		if (pool_active(hdl, name, guid, &isactive) != 0)
763 			goto error;
764 
765 		if (isactive) {
766 			nvlist_free(config);
767 			config = NULL;
768 			continue;
769 		}
770 
771 		if (policy != NULL) {
772 			if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY,
773 			    policy) != 0)
774 				goto nomem;
775 		}
776 
777 		if ((nvl = refresh_config(hdl, config)) == NULL) {
778 			nvlist_free(config);
779 			config = NULL;
780 			continue;
781 		}
782 
783 		nvlist_free(config);
784 		config = nvl;
785 
786 		/*
787 		 * Go through and update the paths for spares, now that we have
788 		 * them.
789 		 */
790 		verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
791 		    &nvroot) == 0);
792 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
793 		    &spares, &nspares) == 0) {
794 			for (i = 0; i < nspares; i++) {
795 				if (fix_paths(spares[i], pl->names) != 0)
796 					goto nomem;
797 			}
798 		}
799 
800 		/*
801 		 * Update the paths for l2cache devices.
802 		 */
803 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
804 		    &l2cache, &nl2cache) == 0) {
805 			for (i = 0; i < nl2cache; i++) {
806 				if (fix_paths(l2cache[i], pl->names) != 0)
807 					goto nomem;
808 			}
809 		}
810 
811 		/*
812 		 * Restore the original information read from the actual label.
813 		 */
814 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
815 		    DATA_TYPE_UINT64);
816 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
817 		    DATA_TYPE_STRING);
818 		if (hostid != 0) {
819 			verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
820 			    hostid) == 0);
821 			verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
822 			    hostname) == 0);
823 		}
824 
825 add_pool:
826 		/*
827 		 * Add this pool to the list of configs.
828 		 */
829 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
830 		    &name) == 0);
831 		if (nvlist_add_nvlist(ret, name, config) != 0)
832 			goto nomem;
833 
834 		found_one = B_TRUE;
835 		nvlist_free(config);
836 		config = NULL;
837 	}
838 
839 	if (!found_one) {
840 		nvlist_free(ret);
841 		ret = NULL;
842 	}
843 
844 	return (ret);
845 
846 nomem:
847 	(void) no_memory(hdl);
848 error:
849 	nvlist_free(config);
850 	nvlist_free(ret);
851 	for (c = 0; c < children; c++)
852 		nvlist_free(child[c]);
853 	free(child);
854 
855 	return (NULL);
856 }
857 
858 /*
859  * Return the offset of the given label.
860  */
861 static uint64_t
label_offset(uint64_t size,int l)862 label_offset(uint64_t size, int l)
863 {
864 	ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
865 	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
866 	    0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
867 }
868 
869 /*
870  * Given a file descriptor, read the label information and return an nvlist
871  * describing the configuration, if there is one.
872  * Return 0 on success, or -1 on failure
873  */
874 int
zpool_read_label(int fd,nvlist_t ** config)875 zpool_read_label(int fd, nvlist_t **config)
876 {
877 	struct stat64 statbuf;
878 	int l;
879 	vdev_label_t *label;
880 	uint64_t state, txg, size;
881 
882 	*config = NULL;
883 
884 	if (fstat64(fd, &statbuf) == -1)
885 		return (-1);
886 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
887 
888 	if ((label = malloc(sizeof (vdev_label_t))) == NULL)
889 		return (-1);
890 
891 	for (l = 0; l < VDEV_LABELS; l++) {
892 		if (pread64(fd, label, sizeof (vdev_label_t),
893 		    label_offset(size, l)) != sizeof (vdev_label_t))
894 			continue;
895 
896 		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
897 		    sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
898 			continue;
899 
900 		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
901 		    &state) != 0 || state > POOL_STATE_L2CACHE) {
902 			nvlist_free(*config);
903 			continue;
904 		}
905 
906 		if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
907 		    (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
908 		    &txg) != 0 || txg == 0)) {
909 			nvlist_free(*config);
910 			continue;
911 		}
912 
913 		free(label);
914 		return (0);
915 	}
916 
917 	free(label);
918 	*config = NULL;
919 	errno = ENOENT;
920 	return (-1);
921 }
922 
923 /*
924  * Given a file descriptor, read the label information and return an nvlist
925  * describing the configuration, if there is one.
926  * returns the number of valid labels found
927  * If a label is found, returns it via config.  The caller is responsible for
928  * freeing it.
929  */
930 int
zpool_read_all_labels(int fd,nvlist_t ** config)931 zpool_read_all_labels(int fd, nvlist_t **config)
932 {
933 	struct stat64 statbuf;
934 	struct aiocb aiocbs[VDEV_LABELS];
935 	struct aiocb *aiocbps[VDEV_LABELS];
936 	int l;
937 	vdev_phys_t *labels;
938 	uint64_t state, txg, size;
939 	int nlabels = 0;
940 
941 	*config = NULL;
942 
943 	if (fstat64(fd, &statbuf) == -1)
944 		return (0);
945 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
946 
947 	if ((labels = calloc(VDEV_LABELS, sizeof (vdev_phys_t))) == NULL)
948 		return (0);
949 
950 	memset(aiocbs, 0, sizeof(aiocbs));
951 	for (l = 0; l < VDEV_LABELS; l++) {
952 		aiocbs[l].aio_fildes = fd;
953 		aiocbs[l].aio_offset = label_offset(size, l) + VDEV_SKIP_SIZE;
954 		aiocbs[l].aio_buf = &labels[l];
955 		aiocbs[l].aio_nbytes = sizeof(vdev_phys_t);
956 		aiocbs[l].aio_lio_opcode = LIO_READ;
957 		aiocbps[l] = &aiocbs[l];
958 	}
959 
960 	if (lio_listio(LIO_WAIT, aiocbps, VDEV_LABELS, NULL) != 0) {
961 		if (errno == EAGAIN || errno == EINTR || errno == EIO) {
962 			for (l = 0; l < VDEV_LABELS; l++) {
963 				errno = 0;
964 				int r = aio_error(&aiocbs[l]);
965 				if (r != EINVAL)
966 					(void)aio_return(&aiocbs[l]);
967 			}
968 		}
969 		free(labels);
970 		return (0);
971 	}
972 
973 	for (l = 0; l < VDEV_LABELS; l++) {
974 		nvlist_t *temp = NULL;
975 
976 		if (aio_return(&aiocbs[l]) != sizeof(vdev_phys_t))
977 			continue;
978 
979 		if (nvlist_unpack(labels[l].vp_nvlist,
980 		    sizeof (labels[l].vp_nvlist), &temp, 0) != 0)
981 			continue;
982 
983 		if (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_STATE,
984 		    &state) != 0 || state > POOL_STATE_L2CACHE) {
985 			nvlist_free(temp);
986 			temp = NULL;
987 			continue;
988 		}
989 
990 		if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
991 		    (nvlist_lookup_uint64(temp, ZPOOL_CONFIG_POOL_TXG,
992 		    &txg) != 0 || txg == 0)) {
993 			nvlist_free(temp);
994 			temp = NULL;
995 			continue;
996 		}
997 		if (temp)
998 			*config = temp;
999 
1000 		nlabels++;
1001 	}
1002 
1003 	free(labels);
1004 	return (nlabels);
1005 }
1006 
1007 typedef struct rdsk_node {
1008 	char *rn_name;
1009 	int rn_dfd;
1010 	libzfs_handle_t *rn_hdl;
1011 	nvlist_t *rn_config;
1012 	avl_tree_t *rn_avl;
1013 	avl_node_t rn_node;
1014 	boolean_t rn_nozpool;
1015 } rdsk_node_t;
1016 
1017 static int
slice_cache_compare(const void * arg1,const void * arg2)1018 slice_cache_compare(const void *arg1, const void *arg2)
1019 {
1020 	const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
1021 	const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
1022 	char *nm1slice, *nm2slice;
1023 	int rv;
1024 
1025 	/*
1026 	 * slices zero and two are the most likely to provide results,
1027 	 * so put those first
1028 	 */
1029 	nm1slice = strstr(nm1, "s0");
1030 	nm2slice = strstr(nm2, "s0");
1031 	if (nm1slice && !nm2slice) {
1032 		return (-1);
1033 	}
1034 	if (!nm1slice && nm2slice) {
1035 		return (1);
1036 	}
1037 	nm1slice = strstr(nm1, "s2");
1038 	nm2slice = strstr(nm2, "s2");
1039 	if (nm1slice && !nm2slice) {
1040 		return (-1);
1041 	}
1042 	if (!nm1slice && nm2slice) {
1043 		return (1);
1044 	}
1045 
1046 	rv = strcmp(nm1, nm2);
1047 	if (rv == 0)
1048 		return (0);
1049 	return (rv > 0 ? 1 : -1);
1050 }
1051 
1052 #ifdef illumos
1053 static void
check_one_slice(avl_tree_t * r,char * diskname,uint_t partno,diskaddr_t size,uint_t blksz)1054 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
1055     diskaddr_t size, uint_t blksz)
1056 {
1057 	rdsk_node_t tmpnode;
1058 	rdsk_node_t *node;
1059 	char sname[MAXNAMELEN];
1060 
1061 	tmpnode.rn_name = &sname[0];
1062 	(void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
1063 	    diskname, partno);
1064 	/*
1065 	 * protect against division by zero for disk labels that
1066 	 * contain a bogus sector size
1067 	 */
1068 	if (blksz == 0)
1069 		blksz = DEV_BSIZE;
1070 	/* too small to contain a zpool? */
1071 	if ((size < (SPA_MINDEVSIZE / blksz)) &&
1072 	    (node = avl_find(r, &tmpnode, NULL)))
1073 		node->rn_nozpool = B_TRUE;
1074 }
1075 #endif	/* illumos */
1076 
1077 static void
nozpool_all_slices(avl_tree_t * r,const char * sname)1078 nozpool_all_slices(avl_tree_t *r, const char *sname)
1079 {
1080 #ifdef illumos
1081 	char diskname[MAXNAMELEN];
1082 	char *ptr;
1083 	int i;
1084 
1085 	(void) strncpy(diskname, sname, MAXNAMELEN);
1086 	if (((ptr = strrchr(diskname, 's')) == NULL) &&
1087 	    ((ptr = strrchr(diskname, 'p')) == NULL))
1088 		return;
1089 	ptr[0] = 's';
1090 	ptr[1] = '\0';
1091 	for (i = 0; i < NDKMAP; i++)
1092 		check_one_slice(r, diskname, i, 0, 1);
1093 	ptr[0] = 'p';
1094 	for (i = 0; i <= FD_NUMPART; i++)
1095 		check_one_slice(r, diskname, i, 0, 1);
1096 #endif	/* illumos */
1097 }
1098 
1099 #ifdef illumos
1100 static void
check_slices(avl_tree_t * r,int fd,const char * sname)1101 check_slices(avl_tree_t *r, int fd, const char *sname)
1102 {
1103 	struct extvtoc vtoc;
1104 	struct dk_gpt *gpt;
1105 	char diskname[MAXNAMELEN];
1106 	char *ptr;
1107 	int i;
1108 
1109 	(void) strncpy(diskname, sname, MAXNAMELEN);
1110 	if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1111 		return;
1112 	ptr[1] = '\0';
1113 
1114 	if (read_extvtoc(fd, &vtoc) >= 0) {
1115 		for (i = 0; i < NDKMAP; i++)
1116 			check_one_slice(r, diskname, i,
1117 			    vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1118 	} else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1119 		/*
1120 		 * on x86 we'll still have leftover links that point
1121 		 * to slices s[9-15], so use NDKMAP instead
1122 		 */
1123 		for (i = 0; i < NDKMAP; i++)
1124 			check_one_slice(r, diskname, i,
1125 			    gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1126 		/* nodes p[1-4] are never used with EFI labels */
1127 		ptr[0] = 'p';
1128 		for (i = 1; i <= FD_NUMPART; i++)
1129 			check_one_slice(r, diskname, i, 0, 1);
1130 		efi_free(gpt);
1131 	}
1132 }
1133 #endif	/* illumos */
1134 
1135 static void
zpool_open_func(void * arg)1136 zpool_open_func(void *arg)
1137 {
1138 	rdsk_node_t *rn = arg;
1139 	struct stat64 statbuf;
1140 	nvlist_t *config;
1141 	int fd;
1142 
1143 	if (rn->rn_nozpool)
1144 		return;
1145 	if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1146 		/* symlink to a device that's no longer there */
1147 		if (errno == ENOENT)
1148 			nozpool_all_slices(rn->rn_avl, rn->rn_name);
1149 		return;
1150 	}
1151 	/*
1152 	 * Ignore failed stats.  We only want regular
1153 	 * files, character devs and block devs.
1154 	 */
1155 	if (fstat64(fd, &statbuf) != 0 ||
1156 	    (!S_ISREG(statbuf.st_mode) &&
1157 	    !S_ISCHR(statbuf.st_mode) &&
1158 	    !S_ISBLK(statbuf.st_mode))) {
1159 		(void) close(fd);
1160 		return;
1161 	}
1162 	/* this file is too small to hold a zpool */
1163 #ifdef illumos
1164 	if (S_ISREG(statbuf.st_mode) &&
1165 	    statbuf.st_size < SPA_MINDEVSIZE) {
1166 		(void) close(fd);
1167 		return;
1168 	} else if (!S_ISREG(statbuf.st_mode)) {
1169 		/*
1170 		 * Try to read the disk label first so we don't have to
1171 		 * open a bunch of minor nodes that can't have a zpool.
1172 		 */
1173 		check_slices(rn->rn_avl, fd, rn->rn_name);
1174 	}
1175 #else	/* !illumos */
1176 	if (statbuf.st_size < SPA_MINDEVSIZE) {
1177 		(void) close(fd);
1178 		return;
1179 	}
1180 #endif	/* illumos */
1181 
1182 	if ((zpool_read_label(fd, &config)) != 0 && errno == ENOMEM) {
1183 		(void) close(fd);
1184 		(void) no_memory(rn->rn_hdl);
1185 		return;
1186 	}
1187 	(void) close(fd);
1188 
1189 	rn->rn_config = config;
1190 }
1191 
1192 /*
1193  * Given a file descriptor, clear (zero) the label information.
1194  */
1195 int
zpool_clear_label(int fd)1196 zpool_clear_label(int fd)
1197 {
1198 	struct stat64 statbuf;
1199 	int l;
1200 	vdev_label_t *label;
1201 	uint64_t size;
1202 
1203 	if (fstat64(fd, &statbuf) == -1)
1204 		return (0);
1205 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1206 
1207 	if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1208 		return (-1);
1209 
1210 	for (l = 0; l < VDEV_LABELS; l++) {
1211 		if (pwrite64(fd, label, sizeof (vdev_label_t),
1212 		    label_offset(size, l)) != sizeof (vdev_label_t)) {
1213 			free(label);
1214 			return (-1);
1215 		}
1216 	}
1217 
1218 	free(label);
1219 	return (0);
1220 }
1221 
1222 /*
1223  * Given a list of directories to search, find all pools stored on disk.  This
1224  * includes partial pools which are not available to import.  If no args are
1225  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1226  * poolname or guid (but not both) are provided by the caller when trying
1227  * to import a specific pool.
1228  */
1229 static nvlist_t *
zpool_find_import_impl(libzfs_handle_t * hdl,importargs_t * iarg)1230 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1231 {
1232 	int i, dirs = iarg->paths;
1233 	struct dirent64 *dp;
1234 	char path[MAXPATHLEN];
1235 	char *end, **dir = iarg->path;
1236 	size_t pathleft;
1237 	nvlist_t *ret = NULL;
1238 	static char *default_dir = "/dev";
1239 	pool_list_t pools = { 0 };
1240 	pool_entry_t *pe, *penext;
1241 	vdev_entry_t *ve, *venext;
1242 	config_entry_t *ce, *cenext;
1243 	name_entry_t *ne, *nenext;
1244 	avl_tree_t slice_cache;
1245 	rdsk_node_t *slice;
1246 	void *cookie;
1247 
1248 	if (dirs == 0) {
1249 		dirs = 1;
1250 		dir = &default_dir;
1251 	}
1252 
1253 	/*
1254 	 * Go through and read the label configuration information from every
1255 	 * possible device, organizing the information according to pool GUID
1256 	 * and toplevel GUID.
1257 	 */
1258 	for (i = 0; i < dirs; i++) {
1259 		tpool_t *t;
1260 		char rdsk[MAXPATHLEN];
1261 		int dfd;
1262 		boolean_t config_failed = B_FALSE;
1263 		DIR *dirp;
1264 
1265 		/* use realpath to normalize the path */
1266 		if (realpath(dir[i], path) == 0) {
1267 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1268 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1269 			goto error;
1270 		}
1271 		end = &path[strlen(path)];
1272 		*end++ = '/';
1273 		*end = 0;
1274 		pathleft = &path[sizeof (path)] - end;
1275 
1276 #ifdef illumos
1277 		/*
1278 		 * Using raw devices instead of block devices when we're
1279 		 * reading the labels skips a bunch of slow operations during
1280 		 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1281 		 */
1282 		if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1283 			(void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1284 		else
1285 #endif
1286 			(void) strlcpy(rdsk, path, sizeof (rdsk));
1287 
1288 		if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1289 		    (dirp = fdopendir(dfd)) == NULL) {
1290 			if (dfd >= 0)
1291 				(void) close(dfd);
1292 			zfs_error_aux(hdl, strerror(errno));
1293 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1294 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1295 			    rdsk);
1296 			goto error;
1297 		}
1298 
1299 		avl_create(&slice_cache, slice_cache_compare,
1300 		    sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1301 
1302 		if (strcmp(rdsk, "/dev/") == 0) {
1303 			struct gmesh mesh;
1304 			struct gclass *mp;
1305 			struct ggeom *gp;
1306 			struct gprovider *pp;
1307 
1308 			errno = geom_gettree(&mesh);
1309 			if (errno != 0) {
1310 				zfs_error_aux(hdl, strerror(errno));
1311 				(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1312 				    dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1313 				goto error;
1314 			}
1315 
1316 			LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1317 		        	LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1318 					LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1319 						slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1320 						slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1321 						slice->rn_avl = &slice_cache;
1322 						slice->rn_dfd = dfd;
1323 						slice->rn_hdl = hdl;
1324 						slice->rn_nozpool = B_FALSE;
1325 						avl_add(&slice_cache, slice);
1326 					}
1327 				}
1328 			}
1329 
1330 			geom_deletetree(&mesh);
1331 			goto skipdir;
1332 		}
1333 
1334 		/*
1335 		 * This is not MT-safe, but we have no MT consumers of libzfs
1336 		 */
1337 		while ((dp = readdir64(dirp)) != NULL) {
1338 			const char *name = dp->d_name;
1339 			if (name[0] == '.' &&
1340 			    (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1341 				continue;
1342 
1343 			slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1344 			slice->rn_name = zfs_strdup(hdl, name);
1345 			slice->rn_avl = &slice_cache;
1346 			slice->rn_dfd = dfd;
1347 			slice->rn_hdl = hdl;
1348 			slice->rn_nozpool = B_FALSE;
1349 			avl_add(&slice_cache, slice);
1350 		}
1351 skipdir:
1352 		/*
1353 		 * create a thread pool to do all of this in parallel;
1354 		 * rn_nozpool is not protected, so this is racy in that
1355 		 * multiple tasks could decide that the same slice can
1356 		 * not hold a zpool, which is benign.  Also choose
1357 		 * double the number of processors; we hold a lot of
1358 		 * locks in the kernel, so going beyond this doesn't
1359 		 * buy us much.
1360 		 */
1361 		t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1362 		    0, NULL);
1363 		for (slice = avl_first(&slice_cache); slice;
1364 		    (slice = avl_walk(&slice_cache, slice,
1365 		    AVL_AFTER)))
1366 			(void) tpool_dispatch(t, zpool_open_func, slice);
1367 		tpool_wait(t);
1368 		tpool_destroy(t);
1369 
1370 		cookie = NULL;
1371 		while ((slice = avl_destroy_nodes(&slice_cache,
1372 		    &cookie)) != NULL) {
1373 			if (slice->rn_config != NULL && !config_failed) {
1374 				nvlist_t *config = slice->rn_config;
1375 				boolean_t matched = B_TRUE;
1376 
1377 				if (iarg->poolname != NULL) {
1378 					char *pname;
1379 
1380 					matched = nvlist_lookup_string(config,
1381 					    ZPOOL_CONFIG_POOL_NAME,
1382 					    &pname) == 0 &&
1383 					    strcmp(iarg->poolname, pname) == 0;
1384 				} else if (iarg->guid != 0) {
1385 					uint64_t this_guid;
1386 
1387 					matched = nvlist_lookup_uint64(config,
1388 					    ZPOOL_CONFIG_POOL_GUID,
1389 					    &this_guid) == 0 &&
1390 					    iarg->guid == this_guid;
1391 				}
1392 				if (matched) {
1393 					/*
1394 					 * use the non-raw path for the config
1395 					 */
1396 					(void) strlcpy(end, slice->rn_name,
1397 					    pathleft);
1398 					if (add_config(hdl, &pools, path,
1399 					    config) != 0)
1400 						config_failed = B_TRUE;
1401 				}
1402 				nvlist_free(config);
1403 			}
1404 			free(slice->rn_name);
1405 			free(slice);
1406 		}
1407 		avl_destroy(&slice_cache);
1408 
1409 		(void) closedir(dirp);
1410 
1411 		if (config_failed)
1412 			goto error;
1413 	}
1414 
1415 	ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy);
1416 
1417 error:
1418 	for (pe = pools.pools; pe != NULL; pe = penext) {
1419 		penext = pe->pe_next;
1420 		for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1421 			venext = ve->ve_next;
1422 			for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1423 				cenext = ce->ce_next;
1424 				nvlist_free(ce->ce_config);
1425 				free(ce);
1426 			}
1427 			free(ve);
1428 		}
1429 		free(pe);
1430 	}
1431 
1432 	for (ne = pools.names; ne != NULL; ne = nenext) {
1433 		nenext = ne->ne_next;
1434 		free(ne->ne_name);
1435 		free(ne);
1436 	}
1437 
1438 	return (ret);
1439 }
1440 
1441 nvlist_t *
zpool_find_import(libzfs_handle_t * hdl,int argc,char ** argv)1442 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1443 {
1444 	importargs_t iarg = { 0 };
1445 
1446 	iarg.paths = argc;
1447 	iarg.path = argv;
1448 
1449 	return (zpool_find_import_impl(hdl, &iarg));
1450 }
1451 
1452 /*
1453  * Given a cache file, return the contents as a list of importable pools.
1454  * poolname or guid (but not both) are provided by the caller when trying
1455  * to import a specific pool.
1456  */
1457 nvlist_t *
zpool_find_import_cached(libzfs_handle_t * hdl,const char * cachefile,char * poolname,uint64_t guid)1458 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1459     char *poolname, uint64_t guid)
1460 {
1461 	char *buf;
1462 	int fd;
1463 	struct stat64 statbuf;
1464 	nvlist_t *raw, *src, *dst;
1465 	nvlist_t *pools;
1466 	nvpair_t *elem;
1467 	char *name;
1468 	uint64_t this_guid;
1469 	boolean_t active;
1470 
1471 	verify(poolname == NULL || guid == 0);
1472 
1473 	if ((fd = open(cachefile, O_RDONLY)) < 0) {
1474 		zfs_error_aux(hdl, "%s", strerror(errno));
1475 		(void) zfs_error(hdl, EZFS_BADCACHE,
1476 		    dgettext(TEXT_DOMAIN, "failed to open cache file"));
1477 		return (NULL);
1478 	}
1479 
1480 	if (fstat64(fd, &statbuf) != 0) {
1481 		zfs_error_aux(hdl, "%s", strerror(errno));
1482 		(void) close(fd);
1483 		(void) zfs_error(hdl, EZFS_BADCACHE,
1484 		    dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1485 		return (NULL);
1486 	}
1487 
1488 	if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1489 		(void) close(fd);
1490 		return (NULL);
1491 	}
1492 
1493 	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1494 		(void) close(fd);
1495 		free(buf);
1496 		(void) zfs_error(hdl, EZFS_BADCACHE,
1497 		    dgettext(TEXT_DOMAIN,
1498 		    "failed to read cache file contents"));
1499 		return (NULL);
1500 	}
1501 
1502 	(void) close(fd);
1503 
1504 	if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1505 		free(buf);
1506 		(void) zfs_error(hdl, EZFS_BADCACHE,
1507 		    dgettext(TEXT_DOMAIN,
1508 		    "invalid or corrupt cache file contents"));
1509 		return (NULL);
1510 	}
1511 
1512 	free(buf);
1513 
1514 	/*
1515 	 * Go through and get the current state of the pools and refresh their
1516 	 * state.
1517 	 */
1518 	if (nvlist_alloc(&pools, 0, 0) != 0) {
1519 		(void) no_memory(hdl);
1520 		nvlist_free(raw);
1521 		return (NULL);
1522 	}
1523 
1524 	elem = NULL;
1525 	while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1526 		src = fnvpair_value_nvlist(elem);
1527 
1528 		name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1529 		if (poolname != NULL && strcmp(poolname, name) != 0)
1530 			continue;
1531 
1532 		this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1533 		if (guid != 0 && guid != this_guid)
1534 			continue;
1535 
1536 		if (pool_active(hdl, name, this_guid, &active) != 0) {
1537 			nvlist_free(raw);
1538 			nvlist_free(pools);
1539 			return (NULL);
1540 		}
1541 
1542 		if (active)
1543 			continue;
1544 
1545 		if (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE,
1546 		    cachefile) != 0) {
1547 			(void) no_memory(hdl);
1548 			nvlist_free(raw);
1549 			nvlist_free(pools);
1550 			return (NULL);
1551 		}
1552 
1553 		if ((dst = refresh_config(hdl, src)) == NULL) {
1554 			nvlist_free(raw);
1555 			nvlist_free(pools);
1556 			return (NULL);
1557 		}
1558 
1559 		if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1560 			(void) no_memory(hdl);
1561 			nvlist_free(dst);
1562 			nvlist_free(raw);
1563 			nvlist_free(pools);
1564 			return (NULL);
1565 		}
1566 		nvlist_free(dst);
1567 	}
1568 
1569 	nvlist_free(raw);
1570 	return (pools);
1571 }
1572 
1573 static int
name_or_guid_exists(zpool_handle_t * zhp,void * data)1574 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1575 {
1576 	importargs_t *import = data;
1577 	int found = 0;
1578 
1579 	if (import->poolname != NULL) {
1580 		char *pool_name;
1581 
1582 		verify(nvlist_lookup_string(zhp->zpool_config,
1583 		    ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1584 		if (strcmp(pool_name, import->poolname) == 0)
1585 			found = 1;
1586 	} else {
1587 		uint64_t pool_guid;
1588 
1589 		verify(nvlist_lookup_uint64(zhp->zpool_config,
1590 		    ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1591 		if (pool_guid == import->guid)
1592 			found = 1;
1593 	}
1594 
1595 	zpool_close(zhp);
1596 	return (found);
1597 }
1598 
1599 nvlist_t *
zpool_search_import(libzfs_handle_t * hdl,importargs_t * import)1600 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1601 {
1602 	verify(import->poolname == NULL || import->guid == 0);
1603 
1604 	if (import->unique)
1605 		import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1606 
1607 	if (import->cachefile != NULL)
1608 		return (zpool_find_import_cached(hdl, import->cachefile,
1609 		    import->poolname, import->guid));
1610 
1611 	return (zpool_find_import_impl(hdl, import));
1612 }
1613 
1614 boolean_t
find_guid(nvlist_t * nv,uint64_t guid)1615 find_guid(nvlist_t *nv, uint64_t guid)
1616 {
1617 	uint64_t tmp;
1618 	nvlist_t **child;
1619 	uint_t c, children;
1620 
1621 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1622 	if (tmp == guid)
1623 		return (B_TRUE);
1624 
1625 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1626 	    &child, &children) == 0) {
1627 		for (c = 0; c < children; c++)
1628 			if (find_guid(child[c], guid))
1629 				return (B_TRUE);
1630 	}
1631 
1632 	return (B_FALSE);
1633 }
1634 
1635 typedef struct aux_cbdata {
1636 	const char	*cb_type;
1637 	uint64_t	cb_guid;
1638 	zpool_handle_t	*cb_zhp;
1639 } aux_cbdata_t;
1640 
1641 static int
find_aux(zpool_handle_t * zhp,void * data)1642 find_aux(zpool_handle_t *zhp, void *data)
1643 {
1644 	aux_cbdata_t *cbp = data;
1645 	nvlist_t **list;
1646 	uint_t i, count;
1647 	uint64_t guid;
1648 	nvlist_t *nvroot;
1649 
1650 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1651 	    &nvroot) == 0);
1652 
1653 	if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1654 	    &list, &count) == 0) {
1655 		for (i = 0; i < count; i++) {
1656 			verify(nvlist_lookup_uint64(list[i],
1657 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1658 			if (guid == cbp->cb_guid) {
1659 				cbp->cb_zhp = zhp;
1660 				return (1);
1661 			}
1662 		}
1663 	}
1664 
1665 	zpool_close(zhp);
1666 	return (0);
1667 }
1668 
1669 /*
1670  * Determines if the pool is in use.  If so, it returns true and the state of
1671  * the pool as well as the name of the pool.  Both strings are allocated and
1672  * must be freed by the caller.
1673  */
1674 int
zpool_in_use(libzfs_handle_t * hdl,int fd,pool_state_t * state,char ** namestr,boolean_t * inuse)1675 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1676     boolean_t *inuse)
1677 {
1678 	nvlist_t *config;
1679 	char *name;
1680 	boolean_t ret;
1681 	uint64_t guid, vdev_guid;
1682 	zpool_handle_t *zhp;
1683 	nvlist_t *pool_config;
1684 	uint64_t stateval, isspare;
1685 	aux_cbdata_t cb = { 0 };
1686 	boolean_t isactive;
1687 
1688 	*inuse = B_FALSE;
1689 
1690 	if (zpool_read_label(fd, &config) != 0 && errno == ENOMEM) {
1691 		(void) no_memory(hdl);
1692 		return (-1);
1693 	}
1694 
1695 	if (config == NULL)
1696 		return (0);
1697 
1698 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1699 	    &stateval) == 0);
1700 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1701 	    &vdev_guid) == 0);
1702 
1703 	if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1704 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1705 		    &name) == 0);
1706 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1707 		    &guid) == 0);
1708 	}
1709 
1710 	switch (stateval) {
1711 	case POOL_STATE_EXPORTED:
1712 		/*
1713 		 * A pool with an exported state may in fact be imported
1714 		 * read-only, so check the in-core state to see if it's
1715 		 * active and imported read-only.  If it is, set
1716 		 * its state to active.
1717 		 */
1718 		if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1719 		    (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1720 			if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1721 				stateval = POOL_STATE_ACTIVE;
1722 
1723 			/*
1724 			 * All we needed the zpool handle for is the
1725 			 * readonly prop check.
1726 			 */
1727 			zpool_close(zhp);
1728 		}
1729 
1730 		ret = B_TRUE;
1731 		break;
1732 
1733 	case POOL_STATE_ACTIVE:
1734 		/*
1735 		 * For an active pool, we have to determine if it's really part
1736 		 * of a currently active pool (in which case the pool will exist
1737 		 * and the guid will be the same), or whether it's part of an
1738 		 * active pool that was disconnected without being explicitly
1739 		 * exported.
1740 		 */
1741 		if (pool_active(hdl, name, guid, &isactive) != 0) {
1742 			nvlist_free(config);
1743 			return (-1);
1744 		}
1745 
1746 		if (isactive) {
1747 			/*
1748 			 * Because the device may have been removed while
1749 			 * offlined, we only report it as active if the vdev is
1750 			 * still present in the config.  Otherwise, pretend like
1751 			 * it's not in use.
1752 			 */
1753 			if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1754 			    (pool_config = zpool_get_config(zhp, NULL))
1755 			    != NULL) {
1756 				nvlist_t *nvroot;
1757 
1758 				verify(nvlist_lookup_nvlist(pool_config,
1759 				    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1760 				ret = find_guid(nvroot, vdev_guid);
1761 			} else {
1762 				ret = B_FALSE;
1763 			}
1764 
1765 			/*
1766 			 * If this is an active spare within another pool, we
1767 			 * treat it like an unused hot spare.  This allows the
1768 			 * user to create a pool with a hot spare that currently
1769 			 * in use within another pool.  Since we return B_TRUE,
1770 			 * libdiskmgt will continue to prevent generic consumers
1771 			 * from using the device.
1772 			 */
1773 			if (ret && nvlist_lookup_uint64(config,
1774 			    ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1775 				stateval = POOL_STATE_SPARE;
1776 
1777 			if (zhp != NULL)
1778 				zpool_close(zhp);
1779 		} else {
1780 			stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1781 			ret = B_TRUE;
1782 		}
1783 		break;
1784 
1785 	case POOL_STATE_SPARE:
1786 		/*
1787 		 * For a hot spare, it can be either definitively in use, or
1788 		 * potentially active.  To determine if it's in use, we iterate
1789 		 * over all pools in the system and search for one with a spare
1790 		 * with a matching guid.
1791 		 *
1792 		 * Due to the shared nature of spares, we don't actually report
1793 		 * the potentially active case as in use.  This means the user
1794 		 * can freely create pools on the hot spares of exported pools,
1795 		 * but to do otherwise makes the resulting code complicated, and
1796 		 * we end up having to deal with this case anyway.
1797 		 */
1798 		cb.cb_zhp = NULL;
1799 		cb.cb_guid = vdev_guid;
1800 		cb.cb_type = ZPOOL_CONFIG_SPARES;
1801 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1802 			name = (char *)zpool_get_name(cb.cb_zhp);
1803 			ret = B_TRUE;
1804 		} else {
1805 			ret = B_FALSE;
1806 		}
1807 		break;
1808 
1809 	case POOL_STATE_L2CACHE:
1810 
1811 		/*
1812 		 * Check if any pool is currently using this l2cache device.
1813 		 */
1814 		cb.cb_zhp = NULL;
1815 		cb.cb_guid = vdev_guid;
1816 		cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1817 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1818 			name = (char *)zpool_get_name(cb.cb_zhp);
1819 			ret = B_TRUE;
1820 		} else {
1821 			ret = B_FALSE;
1822 		}
1823 		break;
1824 
1825 	default:
1826 		ret = B_FALSE;
1827 	}
1828 
1829 
1830 	if (ret) {
1831 		if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1832 			if (cb.cb_zhp)
1833 				zpool_close(cb.cb_zhp);
1834 			nvlist_free(config);
1835 			return (-1);
1836 		}
1837 		*state = (pool_state_t)stateval;
1838 	}
1839 
1840 	if (cb.cb_zhp)
1841 		zpool_close(cb.cb_zhp);
1842 
1843 	nvlist_free(config);
1844 	*inuse = ret;
1845 	return (0);
1846 }
1847