xref: /netbsd-src/external/cddl/osnet/dist/lib/libzfs/common/libzfs_import.c (revision 549613bcf606e7defcaebe98e6987013ec4a108f)
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, 2015 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 <ctype.h>
46 #include <devid.h>
47 #include <dirent.h>
48 #include <errno.h>
49 #include <libintl.h>
50 #include <stddef.h>
51 #include <stdlib.h>
52 #include <string.h>
53 #include <sys/stat.h>
54 #include <unistd.h>
55 #include <fcntl.h>
56 #include <thread_pool.h>
57 #ifdef __FreeBSD__
58 #include <libgeom.h>
59 #endif
60 #ifdef __NetBSD__
61 #include <util.h>
62 static int native_ioctl(int fd, unsigned long cmd, void *arg);
63 #endif
64 
65 #include <sys/vdev_impl.h>
66 
67 #include "libzfs.h"
68 #include "libzfs_impl.h"
69 
70 /*
71  * Intermediate structures used to gather configuration information.
72  */
73 typedef struct config_entry {
74 	uint64_t		ce_txg;
75 	nvlist_t		*ce_config;
76 	struct config_entry	*ce_next;
77 } config_entry_t;
78 
79 typedef struct vdev_entry {
80 	uint64_t		ve_guid;
81 	config_entry_t		*ve_configs;
82 	struct vdev_entry	*ve_next;
83 } vdev_entry_t;
84 
85 typedef struct pool_entry {
86 	uint64_t		pe_guid;
87 	vdev_entry_t		*pe_vdevs;
88 	struct pool_entry	*pe_next;
89 } pool_entry_t;
90 
91 typedef struct name_entry {
92 	char			*ne_name;
93 	uint64_t		ne_guid;
94 	struct name_entry	*ne_next;
95 } name_entry_t;
96 
97 typedef struct pool_list {
98 	pool_entry_t		*pools;
99 	name_entry_t		*names;
100 } pool_list_t;
101 
102 static char *
get_devid(const char * path)103 get_devid(const char *path)
104 {
105 #ifdef have_devid
106 	int fd;
107 	ddi_devid_t devid;
108 	char *minor, *ret;
109 
110 	if ((fd = open(path, O_RDONLY)) < 0)
111 		return (NULL);
112 
113 	minor = NULL;
114 	ret = NULL;
115 	if (devid_get(fd, &devid) == 0) {
116 		if (devid_get_minor_name(fd, &minor) == 0)
117 			ret = devid_str_encode(devid, minor);
118 		if (minor != NULL)
119 			devid_str_free(minor);
120 		devid_free(devid);
121 	}
122 	(void) close(fd);
123 
124 	return (ret);
125 #else
126 	return (NULL);
127 #endif
128 }
129 
130 
131 /*
132  * Go through and fix up any path and/or devid information for the given vdev
133  * configuration.
134  */
135 static int
fix_paths(nvlist_t * nv,name_entry_t * names)136 fix_paths(nvlist_t *nv, name_entry_t *names)
137 {
138 	nvlist_t **child;
139 	uint_t c, children;
140 	uint64_t guid;
141 	name_entry_t *ne, *best;
142 	char *path, *devid;
143 	int matched;
144 
145 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
146 	    &child, &children) == 0) {
147 		for (c = 0; c < children; c++)
148 			if (fix_paths(child[c], names) != 0)
149 				return (-1);
150 		return (0);
151 	}
152 
153 	/*
154 	 * This is a leaf (file or disk) vdev.  In either case, go through
155 	 * the name list and see if we find a matching guid.  If so, replace
156 	 * the path and see if we can calculate a new devid.
157 	 *
158 	 * There may be multiple names associated with a particular guid, in
159 	 * which case we have overlapping slices or multiple paths to the same
160 	 * disk.  If this is the case, then we want to pick the path that is
161 	 * the most similar to the original, where "most similar" is the number
162 	 * of matching characters starting from the end of the path.  This will
163 	 * preserve slice numbers even if the disks have been reorganized, and
164 	 * will also catch preferred disk names if multiple paths exist.
165 	 */
166 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
167 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
168 		path = NULL;
169 
170 	matched = 0;
171 	best = NULL;
172 	for (ne = names; ne != NULL; ne = ne->ne_next) {
173 		if (ne->ne_guid == guid) {
174 			const char *src, *dst;
175 			int count;
176 
177 			if (path == NULL) {
178 				best = ne;
179 				break;
180 			}
181 
182 			src = ne->ne_name + strlen(ne->ne_name) - 1;
183 			dst = path + strlen(path) - 1;
184 			for (count = 0; src >= ne->ne_name && dst >= path;
185 			    src--, dst--, count++)
186 				if (*src != *dst)
187 					break;
188 
189 			/*
190 			 * At this point, 'count' is the number of characters
191 			 * matched from the end.
192 			 */
193 			if (count > matched || best == NULL) {
194 				best = ne;
195 				matched = count;
196 			}
197 		}
198 	}
199 
200 	if (best == NULL)
201 		return (0);
202 
203 	if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
204 		return (-1);
205 
206 	if ((devid = get_devid(best->ne_name)) == NULL) {
207 		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
208 	} else {
209 		if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
210 			devid_str_free(devid);
211 			return (-1);
212 		}
213 		devid_str_free(devid);
214 	}
215 
216 	return (0);
217 }
218 
219 /*
220  * Add the given configuration to the list of known devices.
221  */
222 static int
add_config(libzfs_handle_t * hdl,pool_list_t * pl,const char * path,nvlist_t * config)223 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
224     nvlist_t *config)
225 {
226 	uint64_t pool_guid, vdev_guid, top_guid, txg, state;
227 	pool_entry_t *pe;
228 	vdev_entry_t *ve;
229 	config_entry_t *ce;
230 	name_entry_t *ne;
231 
232 	/*
233 	 * If this is a hot spare not currently in use or level 2 cache
234 	 * device, add it to the list of names to translate, but don't do
235 	 * anything else.
236 	 */
237 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
238 	    &state) == 0 &&
239 	    (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
240 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
241 		if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
242 			return (-1);
243 
244 		if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
245 			free(ne);
246 			return (-1);
247 		}
248 		ne->ne_guid = vdev_guid;
249 		ne->ne_next = pl->names;
250 		pl->names = ne;
251 		return (0);
252 	}
253 
254 	/*
255 	 * If we have a valid config but cannot read any of these fields, then
256 	 * it means we have a half-initialized label.  In vdev_label_init()
257 	 * we write a label with txg == 0 so that we can identify the device
258 	 * in case the user refers to the same disk later on.  If we fail to
259 	 * create the pool, we'll be left with a label in this state
260 	 * which should not be considered part of a valid pool.
261 	 */
262 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
263 	    &pool_guid) != 0 ||
264 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
265 	    &vdev_guid) != 0 ||
266 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
267 	    &top_guid) != 0 ||
268 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
269 	    &txg) != 0 || txg == 0) {
270 		nvlist_free(config);
271 		return (0);
272 	}
273 
274 	/*
275 	 * First, see if we know about this pool.  If not, then add it to the
276 	 * list of known pools.
277 	 */
278 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
279 		if (pe->pe_guid == pool_guid)
280 			break;
281 	}
282 
283 	if (pe == NULL) {
284 		if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
285 			nvlist_free(config);
286 			return (-1);
287 		}
288 		pe->pe_guid = pool_guid;
289 		pe->pe_next = pl->pools;
290 		pl->pools = pe;
291 	}
292 
293 	/*
294 	 * Second, see if we know about this toplevel vdev.  Add it if its
295 	 * missing.
296 	 */
297 	for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
298 		if (ve->ve_guid == top_guid)
299 			break;
300 	}
301 
302 	if (ve == NULL) {
303 		if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
304 			nvlist_free(config);
305 			return (-1);
306 		}
307 		ve->ve_guid = top_guid;
308 		ve->ve_next = pe->pe_vdevs;
309 		pe->pe_vdevs = ve;
310 	}
311 
312 	/*
313 	 * Third, see if we have a config with a matching transaction group.  If
314 	 * so, then we do nothing.  Otherwise, add it to the list of known
315 	 * configs.
316 	 */
317 	for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
318 		if (ce->ce_txg == txg)
319 			break;
320 	}
321 
322 	if (ce == NULL) {
323 		if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
324 			nvlist_free(config);
325 			return (-1);
326 		}
327 		ce->ce_txg = txg;
328 		ce->ce_config = config;
329 		ce->ce_next = ve->ve_configs;
330 		ve->ve_configs = ce;
331 	} else {
332 		nvlist_free(config);
333 	}
334 
335 	/*
336 	 * At this point we've successfully added our config to the list of
337 	 * known configs.  The last thing to do is add the vdev guid -> path
338 	 * mappings so that we can fix up the configuration as necessary before
339 	 * doing the import.
340 	 */
341 	if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
342 		return (-1);
343 
344 	if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
345 		free(ne);
346 		return (-1);
347 	}
348 
349 	ne->ne_guid = vdev_guid;
350 	ne->ne_next = pl->names;
351 	pl->names = ne;
352 
353 	return (0);
354 }
355 
356 /*
357  * Returns true if the named pool matches the given GUID.
358  */
359 static int
pool_active(libzfs_handle_t * hdl,const char * name,uint64_t guid,boolean_t * isactive)360 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
361     boolean_t *isactive)
362 {
363 	zpool_handle_t *zhp;
364 	uint64_t theguid;
365 
366 	if (zpool_open_silent(hdl, name, &zhp) != 0)
367 		return (-1);
368 
369 	if (zhp == NULL) {
370 		*isactive = B_FALSE;
371 		return (0);
372 	}
373 
374 	verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
375 	    &theguid) == 0);
376 
377 	zpool_close(zhp);
378 
379 	*isactive = (theguid == guid);
380 	return (0);
381 }
382 
383 static nvlist_t *
refresh_config(libzfs_handle_t * hdl,nvlist_t * config)384 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
385 {
386 	nvlist_t *nvl;
387 	zfs_cmd_t zc = { 0 };
388 	int err;
389 
390 	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
391 		return (NULL);
392 
393 	if (zcmd_alloc_dst_nvlist(hdl, &zc,
394 	    zc.zc_nvlist_conf_size * 2) != 0) {
395 		zcmd_free_nvlists(&zc);
396 		return (NULL);
397 	}
398 
399 	while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
400 	    &zc)) != 0 && errno == ENOMEM) {
401 		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
402 			zcmd_free_nvlists(&zc);
403 			return (NULL);
404 		}
405 	}
406 
407 	if (err) {
408 		zcmd_free_nvlists(&zc);
409 		return (NULL);
410 	}
411 
412 	if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
413 		zcmd_free_nvlists(&zc);
414 		return (NULL);
415 	}
416 
417 	zcmd_free_nvlists(&zc);
418 	return (nvl);
419 }
420 
421 /*
422  * Determine if the vdev id is a hole in the namespace.
423  */
424 boolean_t
vdev_is_hole(uint64_t * hole_array,uint_t holes,uint_t id)425 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
426 {
427 	for (int c = 0; c < holes; c++) {
428 
429 		/* Top-level is a hole */
430 		if (hole_array[c] == id)
431 			return (B_TRUE);
432 	}
433 	return (B_FALSE);
434 }
435 
436 /*
437  * Convert our list of pools into the definitive set of configurations.  We
438  * start by picking the best config for each toplevel vdev.  Once that's done,
439  * we assemble the toplevel vdevs into a full config for the pool.  We make a
440  * pass to fix up any incorrect paths, and then add it to the main list to
441  * return to the user.
442  */
443 static nvlist_t *
get_configs(libzfs_handle_t * hdl,pool_list_t * pl,boolean_t active_ok)444 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
445 {
446 	pool_entry_t *pe;
447 	vdev_entry_t *ve;
448 	config_entry_t *ce;
449 	nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
450 	nvlist_t **spares, **l2cache;
451 	uint_t i, nspares, nl2cache;
452 	boolean_t config_seen;
453 	uint64_t best_txg;
454 	char *name, *hostname = NULL;
455 	uint64_t guid;
456 	uint_t children = 0;
457 	nvlist_t **child = NULL;
458 	uint_t holes;
459 	uint64_t *hole_array, max_id;
460 	uint_t c;
461 	boolean_t isactive;
462 	uint64_t hostid;
463 	nvlist_t *nvl;
464 	boolean_t found_one = B_FALSE;
465 	boolean_t valid_top_config = B_FALSE;
466 
467 	if (nvlist_alloc(&ret, 0, 0) != 0)
468 		goto nomem;
469 
470 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
471 		uint64_t id, max_txg = 0;
472 
473 		if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
474 			goto nomem;
475 		config_seen = B_FALSE;
476 
477 		/*
478 		 * Iterate over all toplevel vdevs.  Grab the pool configuration
479 		 * from the first one we find, and then go through the rest and
480 		 * add them as necessary to the 'vdevs' member of the config.
481 		 */
482 		for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
483 
484 			/*
485 			 * Determine the best configuration for this vdev by
486 			 * selecting the config with the latest transaction
487 			 * group.
488 			 */
489 			best_txg = 0;
490 			for (ce = ve->ve_configs; ce != NULL;
491 			    ce = ce->ce_next) {
492 
493 				if (ce->ce_txg > best_txg) {
494 					tmp = ce->ce_config;
495 					best_txg = ce->ce_txg;
496 				}
497 			}
498 
499 			/*
500 			 * We rely on the fact that the max txg for the
501 			 * pool will contain the most up-to-date information
502 			 * about the valid top-levels in the vdev namespace.
503 			 */
504 			if (best_txg > max_txg) {
505 				(void) nvlist_remove(config,
506 				    ZPOOL_CONFIG_VDEV_CHILDREN,
507 				    DATA_TYPE_UINT64);
508 				(void) nvlist_remove(config,
509 				    ZPOOL_CONFIG_HOLE_ARRAY,
510 				    DATA_TYPE_UINT64_ARRAY);
511 
512 				max_txg = best_txg;
513 				hole_array = NULL;
514 				holes = 0;
515 				max_id = 0;
516 				valid_top_config = B_FALSE;
517 
518 				if (nvlist_lookup_uint64(tmp,
519 				    ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
520 					verify(nvlist_add_uint64(config,
521 					    ZPOOL_CONFIG_VDEV_CHILDREN,
522 					    max_id) == 0);
523 					valid_top_config = B_TRUE;
524 				}
525 
526 				if (nvlist_lookup_uint64_array(tmp,
527 				    ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
528 				    &holes) == 0) {
529 					verify(nvlist_add_uint64_array(config,
530 					    ZPOOL_CONFIG_HOLE_ARRAY,
531 					    hole_array, holes) == 0);
532 				}
533 			}
534 
535 			if (!config_seen) {
536 				/*
537 				 * Copy the relevant pieces of data to the pool
538 				 * configuration:
539 				 *
540 				 *	version
541 				 *	pool guid
542 				 *	name
543 				 *	comment (if available)
544 				 *	pool state
545 				 *	hostid (if available)
546 				 *	hostname (if available)
547 				 */
548 				uint64_t state, version;
549 				char *comment = NULL;
550 
551 				version = fnvlist_lookup_uint64(tmp,
552 				    ZPOOL_CONFIG_VERSION);
553 				fnvlist_add_uint64(config,
554 				    ZPOOL_CONFIG_VERSION, version);
555 				guid = fnvlist_lookup_uint64(tmp,
556 				    ZPOOL_CONFIG_POOL_GUID);
557 				fnvlist_add_uint64(config,
558 				    ZPOOL_CONFIG_POOL_GUID, guid);
559 				name = fnvlist_lookup_string(tmp,
560 				    ZPOOL_CONFIG_POOL_NAME);
561 				fnvlist_add_string(config,
562 				    ZPOOL_CONFIG_POOL_NAME, name);
563 
564 				if (nvlist_lookup_string(tmp,
565 				    ZPOOL_CONFIG_COMMENT, &comment) == 0)
566 					fnvlist_add_string(config,
567 					    ZPOOL_CONFIG_COMMENT, comment);
568 
569 				state = fnvlist_lookup_uint64(tmp,
570 				    ZPOOL_CONFIG_POOL_STATE);
571 				fnvlist_add_uint64(config,
572 				    ZPOOL_CONFIG_POOL_STATE, state);
573 
574 				hostid = 0;
575 				if (nvlist_lookup_uint64(tmp,
576 				    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
577 					fnvlist_add_uint64(config,
578 					    ZPOOL_CONFIG_HOSTID, hostid);
579 					hostname = fnvlist_lookup_string(tmp,
580 					    ZPOOL_CONFIG_HOSTNAME);
581 					fnvlist_add_string(config,
582 					    ZPOOL_CONFIG_HOSTNAME, hostname);
583 				}
584 
585 				config_seen = B_TRUE;
586 			}
587 
588 			/*
589 			 * Add this top-level vdev to the child array.
590 			 */
591 			verify(nvlist_lookup_nvlist(tmp,
592 			    ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
593 			verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
594 			    &id) == 0);
595 
596 			if (id >= children) {
597 				nvlist_t **newchild;
598 
599 				newchild = zfs_alloc(hdl, (id + 1) *
600 				    sizeof (nvlist_t *));
601 				if (newchild == NULL)
602 					goto nomem;
603 
604 				for (c = 0; c < children; c++)
605 					newchild[c] = child[c];
606 
607 				free(child);
608 				child = newchild;
609 				children = id + 1;
610 			}
611 			if (nvlist_dup(nvtop, &child[id], 0) != 0)
612 				goto nomem;
613 
614 		}
615 
616 		/*
617 		 * If we have information about all the top-levels then
618 		 * clean up the nvlist which we've constructed. This
619 		 * means removing any extraneous devices that are
620 		 * beyond the valid range or adding devices to the end
621 		 * of our array which appear to be missing.
622 		 */
623 		if (valid_top_config) {
624 			if (max_id < children) {
625 				for (c = max_id; c < children; c++)
626 					nvlist_free(child[c]);
627 				children = max_id;
628 			} else if (max_id > children) {
629 				nvlist_t **newchild;
630 
631 				newchild = zfs_alloc(hdl, (max_id) *
632 				    sizeof (nvlist_t *));
633 				if (newchild == NULL)
634 					goto nomem;
635 
636 				for (c = 0; c < children; c++)
637 					newchild[c] = child[c];
638 
639 				free(child);
640 				child = newchild;
641 				children = max_id;
642 			}
643 		}
644 
645 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
646 		    &guid) == 0);
647 
648 		/*
649 		 * The vdev namespace may contain holes as a result of
650 		 * device removal. We must add them back into the vdev
651 		 * tree before we process any missing devices.
652 		 */
653 		if (holes > 0) {
654 			ASSERT(valid_top_config);
655 
656 			for (c = 0; c < children; c++) {
657 				nvlist_t *holey;
658 
659 				if (child[c] != NULL ||
660 				    !vdev_is_hole(hole_array, holes, c))
661 					continue;
662 
663 				if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
664 				    0) != 0)
665 					goto nomem;
666 
667 				/*
668 				 * Holes in the namespace are treated as
669 				 * "hole" top-level vdevs and have a
670 				 * special flag set on them.
671 				 */
672 				if (nvlist_add_string(holey,
673 				    ZPOOL_CONFIG_TYPE,
674 				    VDEV_TYPE_HOLE) != 0 ||
675 				    nvlist_add_uint64(holey,
676 				    ZPOOL_CONFIG_ID, c) != 0 ||
677 				    nvlist_add_uint64(holey,
678 				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
679 					nvlist_free(holey);
680 					goto nomem;
681 				}
682 				child[c] = holey;
683 			}
684 		}
685 
686 		/*
687 		 * Look for any missing top-level vdevs.  If this is the case,
688 		 * create a faked up 'missing' vdev as a placeholder.  We cannot
689 		 * simply compress the child array, because the kernel performs
690 		 * certain checks to make sure the vdev IDs match their location
691 		 * in the configuration.
692 		 */
693 		for (c = 0; c < children; c++) {
694 			if (child[c] == NULL) {
695 				nvlist_t *missing;
696 				if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
697 				    0) != 0)
698 					goto nomem;
699 				if (nvlist_add_string(missing,
700 				    ZPOOL_CONFIG_TYPE,
701 				    VDEV_TYPE_MISSING) != 0 ||
702 				    nvlist_add_uint64(missing,
703 				    ZPOOL_CONFIG_ID, c) != 0 ||
704 				    nvlist_add_uint64(missing,
705 				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
706 					nvlist_free(missing);
707 					goto nomem;
708 				}
709 				child[c] = missing;
710 			}
711 		}
712 
713 		/*
714 		 * Put all of this pool's top-level vdevs into a root vdev.
715 		 */
716 		if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
717 			goto nomem;
718 		if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
719 		    VDEV_TYPE_ROOT) != 0 ||
720 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
721 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
722 		    nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
723 		    child, children) != 0) {
724 			nvlist_free(nvroot);
725 			goto nomem;
726 		}
727 
728 		for (c = 0; c < children; c++)
729 			nvlist_free(child[c]);
730 		free(child);
731 		children = 0;
732 		child = NULL;
733 
734 		/*
735 		 * Go through and fix up any paths and/or devids based on our
736 		 * known list of vdev GUID -> path mappings.
737 		 */
738 		if (fix_paths(nvroot, pl->names) != 0) {
739 			nvlist_free(nvroot);
740 			goto nomem;
741 		}
742 
743 		/*
744 		 * Add the root vdev to this pool's configuration.
745 		 */
746 		if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
747 		    nvroot) != 0) {
748 			nvlist_free(nvroot);
749 			goto nomem;
750 		}
751 		nvlist_free(nvroot);
752 
753 		/*
754 		 * zdb uses this path to report on active pools that were
755 		 * imported or created using -R.
756 		 */
757 		if (active_ok)
758 			goto add_pool;
759 
760 		/*
761 		 * Determine if this pool is currently active, in which case we
762 		 * can't actually import it.
763 		 */
764 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
765 		    &name) == 0);
766 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
767 		    &guid) == 0);
768 
769 		if (pool_active(hdl, name, guid, &isactive) != 0)
770 			goto error;
771 
772 		if (isactive) {
773 			nvlist_free(config);
774 			config = NULL;
775 			continue;
776 		}
777 
778 		if ((nvl = refresh_config(hdl, config)) == NULL) {
779 			nvlist_free(config);
780 			config = NULL;
781 			continue;
782 		}
783 
784 		nvlist_free(config);
785 		config = nvl;
786 
787 		/*
788 		 * Go through and update the paths for spares, now that we have
789 		 * them.
790 		 */
791 		verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
792 		    &nvroot) == 0);
793 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
794 		    &spares, &nspares) == 0) {
795 			for (i = 0; i < nspares; i++) {
796 				if (fix_paths(spares[i], pl->names) != 0)
797 					goto nomem;
798 			}
799 		}
800 
801 		/*
802 		 * Update the paths for l2cache devices.
803 		 */
804 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
805 		    &l2cache, &nl2cache) == 0) {
806 			for (i = 0; i < nl2cache; i++) {
807 				if (fix_paths(l2cache[i], pl->names) != 0)
808 					goto nomem;
809 			}
810 		}
811 
812 		/*
813 		 * Restore the original information read from the actual label.
814 		 */
815 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
816 		    DATA_TYPE_UINT64);
817 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
818 		    DATA_TYPE_STRING);
819 		if (hostid != 0) {
820 			verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
821 			    hostid) == 0);
822 			verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
823 			    hostname) == 0);
824 		}
825 
826 add_pool:
827 		/*
828 		 * Add this pool to the list of configs.
829 		 */
830 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
831 		    &name) == 0);
832 		if (nvlist_add_nvlist(ret, name, config) != 0)
833 			goto nomem;
834 
835 		found_one = B_TRUE;
836 		nvlist_free(config);
837 		config = NULL;
838 	}
839 
840 	if (!found_one) {
841 		nvlist_free(ret);
842 		ret = NULL;
843 	}
844 
845 	return (ret);
846 
847 nomem:
848 	(void) no_memory(hdl);
849 error:
850 	nvlist_free(config);
851 	nvlist_free(ret);
852 	for (c = 0; c < children; c++)
853 		nvlist_free(child[c]);
854 	free(child);
855 
856 	return (NULL);
857 }
858 
859 /*
860  * Return the offset of the given label.
861  */
862 static uint64_t
label_offset(uint64_t size,int l)863 label_offset(uint64_t size, int l)
864 {
865 	ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
866 	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
867 	    0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
868 }
869 
870 /*
871  * Given a file descriptor, read the label information and return an nvlist
872  * describing the configuration, if there is one.
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 (0);
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 	return (0);
920 }
921 
922 typedef struct rdsk_node {
923 	char *rn_name;
924 	int rn_dfd;
925 	libzfs_handle_t *rn_hdl;
926 	nvlist_t *rn_config;
927 	avl_tree_t *rn_avl;
928 	avl_node_t rn_node;
929 	boolean_t rn_nozpool;
930 } rdsk_node_t;
931 
932 static int
slice_cache_compare(const void * arg1,const void * arg2)933 slice_cache_compare(const void *arg1, const void *arg2)
934 {
935 	const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
936 	const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
937 	char *nm1slice, *nm2slice;
938 	int rv;
939 
940 	/*
941 	 * slices zero and two are the most likely to provide results,
942 	 * so put those first
943 	 */
944 	nm1slice = strstr(nm1, "s0");
945 	nm2slice = strstr(nm2, "s0");
946 	if (nm1slice && !nm2slice) {
947 		return (-1);
948 	}
949 	if (!nm1slice && nm2slice) {
950 		return (1);
951 	}
952 	nm1slice = strstr(nm1, "s2");
953 	nm2slice = strstr(nm2, "s2");
954 	if (nm1slice && !nm2slice) {
955 		return (-1);
956 	}
957 	if (!nm1slice && nm2slice) {
958 		return (1);
959 	}
960 
961 	rv = strcmp(nm1, nm2);
962 	if (rv == 0)
963 		return (0);
964 	return (rv > 0 ? 1 : -1);
965 }
966 
967 #ifdef illumos
968 static void
check_one_slice(avl_tree_t * r,char * diskname,uint_t partno,diskaddr_t size,uint_t blksz)969 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
970     diskaddr_t size, uint_t blksz)
971 {
972 	rdsk_node_t tmpnode;
973 	rdsk_node_t *node;
974 	char sname[MAXNAMELEN];
975 
976 	tmpnode.rn_name = &sname[0];
977 	(void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
978 	    diskname, partno);
979 	/*
980 	 * protect against division by zero for disk labels that
981 	 * contain a bogus sector size
982 	 */
983 	if (blksz == 0)
984 		blksz = DEV_BSIZE;
985 	/* too small to contain a zpool? */
986 	if ((size < (SPA_MINDEVSIZE / blksz)) &&
987 	    (node = avl_find(r, &tmpnode, NULL)))
988 		node->rn_nozpool = B_TRUE;
989 }
990 #endif	/* illumos */
991 
992 static void
nozpool_all_slices(avl_tree_t * r,const char * sname)993 nozpool_all_slices(avl_tree_t *r, const char *sname)
994 {
995 #ifdef illumos
996 	char diskname[MAXNAMELEN];
997 	char *ptr;
998 	int i;
999 
1000 	(void) strncpy(diskname, sname, MAXNAMELEN);
1001 	if (((ptr = strrchr(diskname, 's')) == NULL) &&
1002 	    ((ptr = strrchr(diskname, 'p')) == NULL))
1003 		return;
1004 	ptr[0] = 's';
1005 	ptr[1] = '\0';
1006 	for (i = 0; i < NDKMAP; i++)
1007 		check_one_slice(r, diskname, i, 0, 1);
1008 	ptr[0] = 'p';
1009 	for (i = 0; i <= FD_NUMPART; i++)
1010 		check_one_slice(r, diskname, i, 0, 1);
1011 #endif	/* illumos */
1012 }
1013 
1014 #ifdef illumos
1015 static void
check_slices(avl_tree_t * r,int fd,const char * sname)1016 check_slices(avl_tree_t *r, int fd, const char *sname)
1017 {
1018 	struct extvtoc vtoc;
1019 	struct dk_gpt *gpt;
1020 	char diskname[MAXNAMELEN];
1021 	char *ptr;
1022 	int i;
1023 
1024 	(void) strncpy(diskname, sname, MAXNAMELEN);
1025 	if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1026 		return;
1027 	ptr[1] = '\0';
1028 
1029 	if (read_extvtoc(fd, &vtoc) >= 0) {
1030 		for (i = 0; i < NDKMAP; i++)
1031 			check_one_slice(r, diskname, i,
1032 			    vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1033 	} else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1034 		/*
1035 		 * on x86 we'll still have leftover links that point
1036 		 * to slices s[9-15], so use NDKMAP instead
1037 		 */
1038 		for (i = 0; i < NDKMAP; i++)
1039 			check_one_slice(r, diskname, i,
1040 			    gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1041 		/* nodes p[1-4] are never used with EFI labels */
1042 		ptr[0] = 'p';
1043 		for (i = 1; i <= FD_NUMPART; i++)
1044 			check_one_slice(r, diskname, i, 0, 1);
1045 		efi_free(gpt);
1046 	}
1047 }
1048 #endif	/* illumos */
1049 
1050 static void
zpool_open_func(void * arg)1051 zpool_open_func(void *arg)
1052 {
1053 	rdsk_node_t *rn = arg;
1054 	struct stat64 statbuf;
1055 	nvlist_t *config;
1056 	int fd;
1057 
1058 	if (rn->rn_nozpool)
1059 		return;
1060 	if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1061 		/* symlink to a device that's no longer there */
1062 		if (errno == ENOENT)
1063 			nozpool_all_slices(rn->rn_avl, rn->rn_name);
1064 		return;
1065 	}
1066 	/*
1067 	 * Ignore failed stats.  We only want regular
1068 	 * files, character devs and block devs.
1069 	 */
1070 	if (fstat64(fd, &statbuf) != 0 ||
1071 	    (!S_ISREG(statbuf.st_mode) &&
1072 	    !S_ISCHR(statbuf.st_mode) &&
1073 	    !S_ISBLK(statbuf.st_mode))) {
1074 		(void) close(fd);
1075 		return;
1076 	}
1077 	/* this file is too small to hold a zpool */
1078 #ifdef illumos
1079 	if (S_ISREG(statbuf.st_mode) &&
1080 	    statbuf.st_size < SPA_MINDEVSIZE) {
1081 		(void) close(fd);
1082 		return;
1083 	} else if (!S_ISREG(statbuf.st_mode)) {
1084 		/*
1085 		 * Try to read the disk label first so we don't have to
1086 		 * open a bunch of minor nodes that can't have a zpool.
1087 		 */
1088 		check_slices(rn->rn_avl, fd, rn->rn_name);
1089 	}
1090 #endif /* illumos */
1091 #ifdef __FreeBSD__
1092 	if (statbuf.st_size < SPA_MINDEVSIZE) {
1093 		(void) close(fd);
1094 		return;
1095 	}
1096 #endif /* __FreeBSD__ */
1097 #ifdef __NetBSD__
1098 	struct dkwedge_list dkwl;
1099 	off_t size;
1100 
1101 	/* skip devices with wedges */
1102 	memset(&dkwl, 0, sizeof(dkwl));
1103 	if (native_ioctl(fd, DIOCLWEDGES, &dkwl) == 0 &&
1104 	    dkwl.dkwl_nwedges > 0) {
1105 		(void) close(fd);
1106 		return;
1107 	}
1108 
1109 	if (native_ioctl(fd, DIOCGMEDIASIZE, &size) < 0 ||
1110 	    size < SPA_MINDEVSIZE) {
1111 		(void) close(fd);
1112 		return;
1113 	}
1114 #endif
1115 
1116 	if ((zpool_read_label(fd, &config)) != 0) {
1117 		(void) close(fd);
1118 		(void) no_memory(rn->rn_hdl);
1119 		return;
1120 	}
1121 	(void) close(fd);
1122 
1123 	rn->rn_config = config;
1124 }
1125 
1126 /*
1127  * Given a file descriptor, clear (zero) the label information.
1128  */
1129 int
zpool_clear_label(int fd)1130 zpool_clear_label(int fd)
1131 {
1132 	struct stat64 statbuf;
1133 	int l;
1134 	vdev_label_t *label;
1135 	uint64_t size;
1136 
1137 	if (fstat64(fd, &statbuf) == -1)
1138 		return (0);
1139 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1140 
1141 	if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1142 		return (-1);
1143 
1144 	for (l = 0; l < VDEV_LABELS; l++) {
1145 		if (pwrite64(fd, label, sizeof (vdev_label_t),
1146 		    label_offset(size, l)) != sizeof (vdev_label_t)) {
1147 			free(label);
1148 			return (-1);
1149 		}
1150 	}
1151 
1152 	free(label);
1153 	return (0);
1154 }
1155 
1156 /*
1157  * Given a list of directories to search, find all pools stored on disk.  This
1158  * includes partial pools which are not available to import.  If no args are
1159  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1160  * poolname or guid (but not both) are provided by the caller when trying
1161  * to import a specific pool.
1162  */
1163 static nvlist_t *
zpool_find_import_impl(libzfs_handle_t * hdl,importargs_t * iarg)1164 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1165 {
1166 	int i, dirs = iarg->paths;
1167 	struct dirent64 *dp;
1168 	char path[MAXPATHLEN];
1169 	char *end, **dir = iarg->path;
1170 	size_t pathleft;
1171 	nvlist_t *ret = NULL;
1172 	static char *default_dir = "/dev";
1173 	pool_list_t pools = { 0 };
1174 	pool_entry_t *pe, *penext;
1175 	vdev_entry_t *ve, *venext;
1176 	config_entry_t *ce, *cenext;
1177 	name_entry_t *ne, *nenext;
1178 	avl_tree_t slice_cache;
1179 	rdsk_node_t *slice;
1180 	void *cookie;
1181 
1182 	if (dirs == 0) {
1183 		dirs = 1;
1184 		dir = &default_dir;
1185 	}
1186 
1187 	/*
1188 	 * Go through and read the label configuration information from every
1189 	 * possible device, organizing the information according to pool GUID
1190 	 * and toplevel GUID.
1191 	 */
1192 	for (i = 0; i < dirs; i++) {
1193 		tpool_t *t;
1194 		char rdsk[MAXPATHLEN];
1195 		int dfd;
1196 		boolean_t config_failed = B_FALSE;
1197 		DIR *dirp;
1198 
1199 		/* use realpath to normalize the path */
1200 		if (realpath(dir[i], path) == 0) {
1201 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1202 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1203 			goto error;
1204 		}
1205 		end = &path[strlen(path)];
1206 		*end++ = '/';
1207 		*end = 0;
1208 		pathleft = &path[sizeof (path)] - end;
1209 
1210 #ifdef illumos
1211 		/*
1212 		 * Using raw devices instead of block devices when we're
1213 		 * reading the labels skips a bunch of slow operations during
1214 		 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1215 		 */
1216 		if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1217 			(void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1218 		else
1219 #endif
1220 			(void) strlcpy(rdsk, path, sizeof (rdsk));
1221 
1222 		if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1223 		    (dirp = fdopendir(dfd)) == NULL) {
1224 			if (dfd >= 0)
1225 				(void) close(dfd);
1226 			zfs_error_aux(hdl, strerror(errno));
1227 			(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1228 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1229 			    rdsk);
1230 			goto error;
1231 		}
1232 
1233 		avl_create(&slice_cache, slice_cache_compare,
1234 		    sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1235 
1236 #ifdef __FreeBSD__
1237 		if (strcmp(rdsk, "/dev/") == 0) {
1238 			struct gmesh mesh;
1239 			struct gclass *mp;
1240 			struct ggeom *gp;
1241 			struct gprovider *pp;
1242 
1243 			errno = geom_gettree(&mesh);
1244 			if (errno != 0) {
1245 				zfs_error_aux(hdl, strerror(errno));
1246 				(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1247 				    dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1248 				goto error;
1249 			}
1250 
1251 			LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1252 		        	LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1253 					LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1254 						slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1255 						slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1256 						slice->rn_avl = &slice_cache;
1257 						slice->rn_dfd = dfd;
1258 						slice->rn_hdl = hdl;
1259 						slice->rn_nozpool = B_FALSE;
1260 						avl_add(&slice_cache, slice);
1261 					}
1262 				}
1263 			}
1264 
1265 			geom_deletetree(&mesh);
1266 			goto skipdir;
1267 		}
1268 #endif
1269 #ifdef __NetBSD__
1270 		if (strcmp(rdsk, "/dev/") == 0) {
1271 			static const char mib_name[] = "hw.disknames";
1272 			size_t len;
1273 			char *disknames, *last, *name;
1274 
1275 			if (sysctlbyname(mib_name, NULL, &len, NULL, 0) == -1) {
1276 				zfs_error_aux(hdl, strerror(errno));
1277 				(void) zfs_error_fmt(hdl, EZFS_BADPATH,
1278 				    dgettext(TEXT_DOMAIN, "cannot get hw.disknames list"));
1279 
1280 				avl_destroy(&slice_cache);
1281 				(void) closedir(dirp);
1282 				goto error;
1283 			}
1284 			disknames = zfs_alloc(hdl, len + 2);
1285 			(void)sysctlbyname(mib_name, disknames, &len, NULL, 0);
1286 
1287 			for ((name = strtok_r(disknames, " ", &last)); name;
1288 			    (name = strtok_r(NULL, " ", &last))) {
1289 				slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1290 				slice->rn_name = zfs_strdup(hdl, name);
1291 				slice->rn_avl = &slice_cache;
1292 				slice->rn_dfd = dfd;
1293 				slice->rn_hdl = hdl;
1294 				slice->rn_nozpool = B_FALSE;
1295 				avl_add(&slice_cache, slice);
1296 			}
1297 			free(disknames);
1298 
1299 			goto skipdir;
1300 		}
1301 #endif
1302 
1303 		/*
1304 		 * This is not MT-safe, but we have no MT consumers of libzfs
1305 		 */
1306 		while ((dp = readdir64(dirp)) != NULL) {
1307 			const char *name = dp->d_name;
1308 			if (name[0] == '.' &&
1309 			    (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1310 				continue;
1311 
1312 			slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1313 			slice->rn_name = zfs_strdup(hdl, name);
1314 			slice->rn_avl = &slice_cache;
1315 			slice->rn_dfd = dfd;
1316 			slice->rn_hdl = hdl;
1317 			slice->rn_nozpool = B_FALSE;
1318 			avl_add(&slice_cache, slice);
1319 		}
1320 skipdir:
1321 		/*
1322 		 * create a thread pool to do all of this in parallel;
1323 		 * rn_nozpool is not protected, so this is racy in that
1324 		 * multiple tasks could decide that the same slice can
1325 		 * not hold a zpool, which is benign.  Also choose
1326 		 * double the number of processors; we hold a lot of
1327 		 * locks in the kernel, so going beyond this doesn't
1328 		 * buy us much.
1329 		 */
1330 		t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1331 		    0, NULL);
1332 		for (slice = avl_first(&slice_cache); slice;
1333 		    (slice = avl_walk(&slice_cache, slice,
1334 		    AVL_AFTER)))
1335 			(void) tpool_dispatch(t, zpool_open_func, slice);
1336 		tpool_wait(t);
1337 		tpool_destroy(t);
1338 
1339 		cookie = NULL;
1340 		while ((slice = avl_destroy_nodes(&slice_cache,
1341 		    &cookie)) != NULL) {
1342 			if (slice->rn_config != NULL && !config_failed) {
1343 				nvlist_t *config = slice->rn_config;
1344 				boolean_t matched = B_TRUE;
1345 
1346 				if (iarg->poolname != NULL) {
1347 					char *pname;
1348 
1349 					matched = nvlist_lookup_string(config,
1350 					    ZPOOL_CONFIG_POOL_NAME,
1351 					    &pname) == 0 &&
1352 					    strcmp(iarg->poolname, pname) == 0;
1353 				} else if (iarg->guid != 0) {
1354 					uint64_t this_guid;
1355 
1356 					matched = nvlist_lookup_uint64(config,
1357 					    ZPOOL_CONFIG_POOL_GUID,
1358 					    &this_guid) == 0 &&
1359 					    iarg->guid == this_guid;
1360 				}
1361 				if (!matched) {
1362 					nvlist_free(config);
1363 				} else {
1364 					/*
1365 					 * use the non-raw path for the config
1366 					 */
1367 					(void) strlcpy(end, slice->rn_name,
1368 					    pathleft);
1369 					if (add_config(hdl, &pools, path,
1370 					    config) != 0)
1371 						config_failed = B_TRUE;
1372 				}
1373 			}
1374 			free(slice->rn_name);
1375 			free(slice);
1376 		}
1377 		avl_destroy(&slice_cache);
1378 
1379 		(void) closedir(dirp);
1380 
1381 		if (config_failed)
1382 			goto error;
1383 	}
1384 
1385 	ret = get_configs(hdl, &pools, iarg->can_be_active);
1386 
1387 error:
1388 	for (pe = pools.pools; pe != NULL; pe = penext) {
1389 		penext = pe->pe_next;
1390 		for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1391 			venext = ve->ve_next;
1392 			for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1393 				cenext = ce->ce_next;
1394 				nvlist_free(ce->ce_config);
1395 				free(ce);
1396 			}
1397 			free(ve);
1398 		}
1399 		free(pe);
1400 	}
1401 
1402 	for (ne = pools.names; ne != NULL; ne = nenext) {
1403 		nenext = ne->ne_next;
1404 		free(ne->ne_name);
1405 		free(ne);
1406 	}
1407 
1408 	return (ret);
1409 }
1410 
1411 nvlist_t *
zpool_find_import(libzfs_handle_t * hdl,int argc,char ** argv)1412 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1413 {
1414 	importargs_t iarg = { 0 };
1415 
1416 	iarg.paths = argc;
1417 	iarg.path = argv;
1418 
1419 	return (zpool_find_import_impl(hdl, &iarg));
1420 }
1421 
1422 /*
1423  * Given a cache file, return the contents as a list of importable pools.
1424  * poolname or guid (but not both) are provided by the caller when trying
1425  * to import a specific pool.
1426  */
1427 nvlist_t *
zpool_find_import_cached(libzfs_handle_t * hdl,const char * cachefile,char * poolname,uint64_t guid)1428 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1429     char *poolname, uint64_t guid)
1430 {
1431 	char *buf;
1432 	int fd;
1433 	struct stat64 statbuf;
1434 	nvlist_t *raw, *src, *dst;
1435 	nvlist_t *pools;
1436 	nvpair_t *elem;
1437 	char *name;
1438 	uint64_t this_guid;
1439 	boolean_t active;
1440 
1441 	verify(poolname == NULL || guid == 0);
1442 
1443 	if ((fd = open(cachefile, O_RDONLY)) < 0) {
1444 		zfs_error_aux(hdl, "%s", strerror(errno));
1445 		(void) zfs_error(hdl, EZFS_BADCACHE,
1446 		    dgettext(TEXT_DOMAIN, "failed to open cache file"));
1447 		return (NULL);
1448 	}
1449 
1450 	if (fstat64(fd, &statbuf) != 0) {
1451 		zfs_error_aux(hdl, "%s", strerror(errno));
1452 		(void) close(fd);
1453 		(void) zfs_error(hdl, EZFS_BADCACHE,
1454 		    dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1455 		return (NULL);
1456 	}
1457 
1458 	if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1459 		(void) close(fd);
1460 		return (NULL);
1461 	}
1462 
1463 	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1464 		(void) close(fd);
1465 		free(buf);
1466 		(void) zfs_error(hdl, EZFS_BADCACHE,
1467 		    dgettext(TEXT_DOMAIN,
1468 		    "failed to read cache file contents"));
1469 		return (NULL);
1470 	}
1471 
1472 	(void) close(fd);
1473 
1474 	if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1475 		free(buf);
1476 		(void) zfs_error(hdl, EZFS_BADCACHE,
1477 		    dgettext(TEXT_DOMAIN,
1478 		    "invalid or corrupt cache file contents"));
1479 		return (NULL);
1480 	}
1481 
1482 	free(buf);
1483 
1484 	/*
1485 	 * Go through and get the current state of the pools and refresh their
1486 	 * state.
1487 	 */
1488 	if (nvlist_alloc(&pools, 0, 0) != 0) {
1489 		(void) no_memory(hdl);
1490 		nvlist_free(raw);
1491 		return (NULL);
1492 	}
1493 
1494 	elem = NULL;
1495 	while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1496 		src = fnvpair_value_nvlist(elem);
1497 
1498 		name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1499 		if (poolname != NULL && strcmp(poolname, name) != 0)
1500 			continue;
1501 
1502 		this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1503 		if (guid != 0 && guid != this_guid)
1504 			continue;
1505 
1506 		if (pool_active(hdl, name, this_guid, &active) != 0) {
1507 			nvlist_free(raw);
1508 			nvlist_free(pools);
1509 			return (NULL);
1510 		}
1511 
1512 		if (active)
1513 			continue;
1514 
1515 		if ((dst = refresh_config(hdl, src)) == NULL) {
1516 			nvlist_free(raw);
1517 			nvlist_free(pools);
1518 			return (NULL);
1519 		}
1520 
1521 		if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1522 			(void) no_memory(hdl);
1523 			nvlist_free(dst);
1524 			nvlist_free(raw);
1525 			nvlist_free(pools);
1526 			return (NULL);
1527 		}
1528 		nvlist_free(dst);
1529 	}
1530 
1531 	nvlist_free(raw);
1532 	return (pools);
1533 }
1534 
1535 static int
name_or_guid_exists(zpool_handle_t * zhp,void * data)1536 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1537 {
1538 	importargs_t *import = data;
1539 	int found = 0;
1540 
1541 	if (import->poolname != NULL) {
1542 		char *pool_name;
1543 
1544 		verify(nvlist_lookup_string(zhp->zpool_config,
1545 		    ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1546 		if (strcmp(pool_name, import->poolname) == 0)
1547 			found = 1;
1548 	} else {
1549 		uint64_t pool_guid;
1550 
1551 		verify(nvlist_lookup_uint64(zhp->zpool_config,
1552 		    ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1553 		if (pool_guid == import->guid)
1554 			found = 1;
1555 	}
1556 
1557 	zpool_close(zhp);
1558 	return (found);
1559 }
1560 
1561 nvlist_t *
zpool_search_import(libzfs_handle_t * hdl,importargs_t * import)1562 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1563 {
1564 	verify(import->poolname == NULL || import->guid == 0);
1565 
1566 	if (import->unique)
1567 		import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1568 
1569 	if (import->cachefile != NULL)
1570 		return (zpool_find_import_cached(hdl, import->cachefile,
1571 		    import->poolname, import->guid));
1572 
1573 	return (zpool_find_import_impl(hdl, import));
1574 }
1575 
1576 boolean_t
find_guid(nvlist_t * nv,uint64_t guid)1577 find_guid(nvlist_t *nv, uint64_t guid)
1578 {
1579 	uint64_t tmp;
1580 	nvlist_t **child;
1581 	uint_t c, children;
1582 
1583 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1584 	if (tmp == guid)
1585 		return (B_TRUE);
1586 
1587 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1588 	    &child, &children) == 0) {
1589 		for (c = 0; c < children; c++)
1590 			if (find_guid(child[c], guid))
1591 				return (B_TRUE);
1592 	}
1593 
1594 	return (B_FALSE);
1595 }
1596 
1597 typedef struct aux_cbdata {
1598 	const char	*cb_type;
1599 	uint64_t	cb_guid;
1600 	zpool_handle_t	*cb_zhp;
1601 } aux_cbdata_t;
1602 
1603 static int
find_aux(zpool_handle_t * zhp,void * data)1604 find_aux(zpool_handle_t *zhp, void *data)
1605 {
1606 	aux_cbdata_t *cbp = data;
1607 	nvlist_t **list;
1608 	uint_t i, count;
1609 	uint64_t guid;
1610 	nvlist_t *nvroot;
1611 
1612 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1613 	    &nvroot) == 0);
1614 
1615 	if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1616 	    &list, &count) == 0) {
1617 		for (i = 0; i < count; i++) {
1618 			verify(nvlist_lookup_uint64(list[i],
1619 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1620 			if (guid == cbp->cb_guid) {
1621 				cbp->cb_zhp = zhp;
1622 				return (1);
1623 			}
1624 		}
1625 	}
1626 
1627 	zpool_close(zhp);
1628 	return (0);
1629 }
1630 
1631 /*
1632  * Determines if the pool is in use.  If so, it returns true and the state of
1633  * the pool as well as the name of the pool.  Both strings are allocated and
1634  * must be freed by the caller.
1635  */
1636 int
zpool_in_use(libzfs_handle_t * hdl,int fd,pool_state_t * state,char ** namestr,boolean_t * inuse)1637 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1638     boolean_t *inuse)
1639 {
1640 	nvlist_t *config;
1641 	char *name;
1642 	boolean_t ret;
1643 	uint64_t guid, vdev_guid;
1644 	zpool_handle_t *zhp;
1645 	nvlist_t *pool_config;
1646 	uint64_t stateval, isspare;
1647 	aux_cbdata_t cb = { 0 };
1648 	boolean_t isactive;
1649 
1650 	*inuse = B_FALSE;
1651 
1652 	if (zpool_read_label(fd, &config) != 0) {
1653 		(void) no_memory(hdl);
1654 		return (-1);
1655 	}
1656 
1657 	if (config == NULL)
1658 		return (0);
1659 
1660 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1661 	    &stateval) == 0);
1662 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1663 	    &vdev_guid) == 0);
1664 
1665 	if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1666 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1667 		    &name) == 0);
1668 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1669 		    &guid) == 0);
1670 	}
1671 
1672 	switch (stateval) {
1673 	case POOL_STATE_EXPORTED:
1674 		/*
1675 		 * A pool with an exported state may in fact be imported
1676 		 * read-only, so check the in-core state to see if it's
1677 		 * active and imported read-only.  If it is, set
1678 		 * its state to active.
1679 		 */
1680 		if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1681 		    (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1682 			if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1683 				stateval = POOL_STATE_ACTIVE;
1684 
1685 			/*
1686 			 * All we needed the zpool handle for is the
1687 			 * readonly prop check.
1688 			 */
1689 			zpool_close(zhp);
1690 		}
1691 
1692 		ret = B_TRUE;
1693 		break;
1694 
1695 	case POOL_STATE_ACTIVE:
1696 		/*
1697 		 * For an active pool, we have to determine if it's really part
1698 		 * of a currently active pool (in which case the pool will exist
1699 		 * and the guid will be the same), or whether it's part of an
1700 		 * active pool that was disconnected without being explicitly
1701 		 * exported.
1702 		 */
1703 		if (pool_active(hdl, name, guid, &isactive) != 0) {
1704 			nvlist_free(config);
1705 			return (-1);
1706 		}
1707 
1708 		if (isactive) {
1709 			/*
1710 			 * Because the device may have been removed while
1711 			 * offlined, we only report it as active if the vdev is
1712 			 * still present in the config.  Otherwise, pretend like
1713 			 * it's not in use.
1714 			 */
1715 			if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1716 			    (pool_config = zpool_get_config(zhp, NULL))
1717 			    != NULL) {
1718 				nvlist_t *nvroot;
1719 
1720 				verify(nvlist_lookup_nvlist(pool_config,
1721 				    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1722 				ret = find_guid(nvroot, vdev_guid);
1723 			} else {
1724 				ret = B_FALSE;
1725 			}
1726 
1727 			/*
1728 			 * If this is an active spare within another pool, we
1729 			 * treat it like an unused hot spare.  This allows the
1730 			 * user to create a pool with a hot spare that currently
1731 			 * in use within another pool.  Since we return B_TRUE,
1732 			 * libdiskmgt will continue to prevent generic consumers
1733 			 * from using the device.
1734 			 */
1735 			if (ret && nvlist_lookup_uint64(config,
1736 			    ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1737 				stateval = POOL_STATE_SPARE;
1738 
1739 			if (zhp != NULL)
1740 				zpool_close(zhp);
1741 		} else {
1742 			stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1743 			ret = B_TRUE;
1744 		}
1745 		break;
1746 
1747 	case POOL_STATE_SPARE:
1748 		/*
1749 		 * For a hot spare, it can be either definitively in use, or
1750 		 * potentially active.  To determine if it's in use, we iterate
1751 		 * over all pools in the system and search for one with a spare
1752 		 * with a matching guid.
1753 		 *
1754 		 * Due to the shared nature of spares, we don't actually report
1755 		 * the potentially active case as in use.  This means the user
1756 		 * can freely create pools on the hot spares of exported pools,
1757 		 * but to do otherwise makes the resulting code complicated, and
1758 		 * we end up having to deal with this case anyway.
1759 		 */
1760 		cb.cb_zhp = NULL;
1761 		cb.cb_guid = vdev_guid;
1762 		cb.cb_type = ZPOOL_CONFIG_SPARES;
1763 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1764 			name = (char *)zpool_get_name(cb.cb_zhp);
1765 			ret = B_TRUE;
1766 		} else {
1767 			ret = B_FALSE;
1768 		}
1769 		break;
1770 
1771 	case POOL_STATE_L2CACHE:
1772 
1773 		/*
1774 		 * Check if any pool is currently using this l2cache device.
1775 		 */
1776 		cb.cb_zhp = NULL;
1777 		cb.cb_guid = vdev_guid;
1778 		cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1779 		if (zpool_iter(hdl, find_aux, &cb) == 1) {
1780 			name = (char *)zpool_get_name(cb.cb_zhp);
1781 			ret = B_TRUE;
1782 		} else {
1783 			ret = B_FALSE;
1784 		}
1785 		break;
1786 
1787 	default:
1788 		ret = B_FALSE;
1789 	}
1790 
1791 
1792 	if (ret) {
1793 		if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1794 			if (cb.cb_zhp)
1795 				zpool_close(cb.cb_zhp);
1796 			nvlist_free(config);
1797 			return (-1);
1798 		}
1799 		*state = (pool_state_t)stateval;
1800 	}
1801 
1802 	if (cb.cb_zhp)
1803 		zpool_close(cb.cb_zhp);
1804 
1805 	nvlist_free(config);
1806 	*inuse = ret;
1807 	return (0);
1808 }
1809 
1810 #ifdef __NetBSD__
1811 /*
1812  * This needs to be at the end of the file so that we can #undef ioctl
1813  * without affecting anything else.
1814  */
1815 #undef ioctl
1816 
1817 static int
native_ioctl(int fd,unsigned long cmd,void * arg)1818 native_ioctl(int fd, unsigned long cmd, void *arg)
1819 {
1820 
1821 	return ioctl(fd, cmd, arg);
1822 }
1823 #endif
1824