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 2017 Nexenta Systems, Inc.
24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
26  * Copyright 2015 RackTop Systems.
27  * Copyright (c) 2016, Intel Corporation.
28  * Copyright 2020 Joyent, Inc.
29  */
30 
31 /*
32  * Pool import support functions.
33  *
34  * Used by zpool, ztest, zdb, and zhack to locate importable configs. Since
35  * these commands are expected to run in the global zone, we can assume
36  * that the devices are all readable when called.
37  *
38  * To import a pool, we rely on reading the configuration information from the
39  * ZFS label of each device.  If we successfully read the label, then we
40  * organize the configuration information in the following hierarchy:
41  *
42  *	pool guid -> toplevel vdev guid -> label txg
43  *
44  * Duplicate entries matching this same tuple will be discarded.  Once we have
45  * examined every device, we pick the best label txg config for each toplevel
46  * vdev.  We then arrange these toplevel vdevs into a complete pool config, and
47  * update any paths that have changed.  Finally, we attempt to import the pool
48  * using our derived config, and record the results.
49  */
50 
51 #include <stdio.h>
52 #include <stdarg.h>
53 #include <assert.h>
54 #include <ctype.h>
55 #include <devid.h>
56 #include <dirent.h>
57 #include <errno.h>
58 #include <libintl.h>
59 #include <stddef.h>
60 #include <stdlib.h>
61 #include <string.h>
62 #include <sys/stat.h>
63 #include <unistd.h>
64 #include <fcntl.h>
65 #include <sys/vtoc.h>
66 #include <sys/dktp/fdisk.h>
67 #include <sys/efi_partition.h>
68 #include <sys/vdev_impl.h>
69 #include <sys/fs/zfs.h>
70 
71 #include <thread_pool.h>
72 #include <libzutil.h>
73 #include <libnvpair.h>
74 
75 #include "zutil_import.h"
76 
77 #ifdef NDEBUG
78 #define	verify(EX)	((void)(EX))
79 #else
80 #define	verify(EX)	assert(EX)
81 #endif
82 
83 /*PRINTFLIKE2*/
84 static void
zutil_error_aux(libpc_handle_t * hdl,const char * fmt,...)85 zutil_error_aux(libpc_handle_t *hdl, const char *fmt, ...)
86 {
87 	va_list ap;
88 
89 	va_start(ap, fmt);
90 
91 	(void) vsnprintf(hdl->lpc_desc, sizeof (hdl->lpc_desc), fmt, ap);
92 	hdl->lpc_desc_active = B_TRUE;
93 
94 	va_end(ap);
95 }
96 
97 static void
zutil_verror(libpc_handle_t * hdl,const char * error,const char * fmt,va_list ap)98 zutil_verror(libpc_handle_t *hdl, const char *error, const char *fmt,
99     va_list ap)
100 {
101 	char action[1024];
102 
103 	(void) vsnprintf(action, sizeof (action), fmt, ap);
104 
105 	if (hdl->lpc_desc_active)
106 		hdl->lpc_desc_active = B_FALSE;
107 	else
108 		hdl->lpc_desc[0] = '\0';
109 
110 	if (hdl->lpc_printerr) {
111 		if (hdl->lpc_desc[0] != '\0')
112 			error = hdl->lpc_desc;
113 
114 		(void) fprintf(stderr, "%s: %s\n", action, error);
115 	}
116 }
117 
118 /*PRINTFLIKE3*/
119 static int
zutil_error_fmt(libpc_handle_t * hdl,const char * error,const char * fmt,...)120 zutil_error_fmt(libpc_handle_t *hdl, const char *error, const char *fmt, ...)
121 {
122 	va_list ap;
123 
124 	va_start(ap, fmt);
125 
126 	zutil_verror(hdl, error, fmt, ap);
127 
128 	va_end(ap);
129 
130 	return (-1);
131 }
132 
133 static int
zutil_error(libpc_handle_t * hdl,const char * error,const char * msg)134 zutil_error(libpc_handle_t *hdl, const char *error, const char *msg)
135 {
136 	return (zutil_error_fmt(hdl, error, "%s", msg));
137 }
138 
139 static int
zutil_no_memory(libpc_handle_t * hdl)140 zutil_no_memory(libpc_handle_t *hdl)
141 {
142 	(void) zutil_error(hdl, EZFS_NOMEM, "internal error");
143 	exit(1);
144 }
145 
146 void *
zutil_alloc(libpc_handle_t * hdl,size_t size)147 zutil_alloc(libpc_handle_t *hdl, size_t size)
148 {
149 	void *data;
150 
151 	if ((data = calloc(1, size)) == NULL)
152 		(void) zutil_no_memory(hdl);
153 
154 	return (data);
155 }
156 
157 char *
zutil_strdup(libpc_handle_t * hdl,const char * str)158 zutil_strdup(libpc_handle_t *hdl, const char *str)
159 {
160 	char *ret;
161 
162 	if ((ret = strdup(str)) == NULL)
163 		(void) zutil_no_memory(hdl);
164 
165 	return (ret);
166 }
167 
168 /*
169  * Intermediate structures used to gather configuration information.
170  */
171 typedef struct config_entry {
172 	uint64_t		ce_txg;
173 	nvlist_t		*ce_config;
174 	struct config_entry	*ce_next;
175 } config_entry_t;
176 
177 typedef struct vdev_entry {
178 	uint64_t		ve_guid;
179 	config_entry_t		*ve_configs;
180 	struct vdev_entry	*ve_next;
181 } vdev_entry_t;
182 
183 typedef struct pool_entry {
184 	uint64_t		pe_guid;
185 	vdev_entry_t		*pe_vdevs;
186 	struct pool_entry	*pe_next;
187 } pool_entry_t;
188 
189 typedef struct name_entry {
190 	char			*ne_name;
191 	uint64_t		ne_guid;
192 	struct name_entry	*ne_next;
193 } name_entry_t;
194 
195 typedef struct pool_list {
196 	pool_entry_t		*pools;
197 	name_entry_t		*names;
198 } pool_list_t;
199 
200 /*
201  * Go through and fix up any path and/or devid information for the given vdev
202  * configuration.
203  */
204 static int
fix_paths(nvlist_t * nv,name_entry_t * names)205 fix_paths(nvlist_t *nv, name_entry_t *names)
206 {
207 	nvlist_t **child;
208 	uint_t c, children;
209 	uint64_t guid;
210 	name_entry_t *ne, *best;
211 	char *path, *devid;
212 	int matched;
213 
214 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
215 	    &child, &children) == 0) {
216 		for (c = 0; c < children; c++)
217 			if (fix_paths(child[c], names) != 0)
218 				return (-1);
219 		return (0);
220 	}
221 
222 	/*
223 	 * This is a leaf (file or disk) vdev.  In either case, go through
224 	 * the name list and see if we find a matching guid.  If so, replace
225 	 * the path and see if we can calculate a new devid.
226 	 *
227 	 * There may be multiple names associated with a particular guid, in
228 	 * which case we have overlapping slices or multiple paths to the same
229 	 * disk.  If this is the case, then we want to pick the path that is
230 	 * the most similar to the original, where "most similar" is the number
231 	 * of matching characters starting from the end of the path.  This will
232 	 * preserve slice numbers even if the disks have been reorganized, and
233 	 * will also catch preferred disk names if multiple paths exist.
234 	 */
235 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
236 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
237 		path = NULL;
238 
239 	matched = 0;
240 	best = NULL;
241 	for (ne = names; ne != NULL; ne = ne->ne_next) {
242 		if (ne->ne_guid == guid) {
243 			const char *src, *dst;
244 			int count;
245 
246 			if (path == NULL) {
247 				best = ne;
248 				break;
249 			}
250 
251 			src = ne->ne_name + strlen(ne->ne_name) - 1;
252 			dst = path + strlen(path) - 1;
253 			for (count = 0; src >= ne->ne_name && dst >= path;
254 			    src--, dst--, count++)
255 				if (*src != *dst)
256 					break;
257 
258 			/*
259 			 * At this point, 'count' is the number of characters
260 			 * matched from the end.
261 			 */
262 			if (count > matched || best == NULL) {
263 				best = ne;
264 				matched = count;
265 			}
266 		}
267 	}
268 
269 	if (best == NULL)
270 		return (0);
271 
272 	if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
273 		return (-1);
274 
275 	if ((devid = devid_str_from_path(best->ne_name)) == NULL) {
276 		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
277 	} else {
278 		if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
279 			devid_str_free(devid);
280 			return (-1);
281 		}
282 		devid_str_free(devid);
283 	}
284 
285 	return (0);
286 }
287 
288 /*
289  * Add the given configuration to the list of known devices.
290  */
291 static int
add_config(libpc_handle_t * hdl,pool_list_t * pl,const char * path,int order,int num_labels,nvlist_t * config)292 add_config(libpc_handle_t *hdl, pool_list_t *pl, const char *path,
293     int order, int num_labels, nvlist_t *config)
294 {
295 	uint64_t pool_guid, vdev_guid, top_guid, txg, state;
296 	pool_entry_t *pe;
297 	vdev_entry_t *ve;
298 	config_entry_t *ce;
299 	name_entry_t *ne;
300 
301 	/*
302 	 * If this is a hot spare not currently in use or level 2 cache
303 	 * device, add it to the list of names to translate, but don't do
304 	 * anything else.
305 	 */
306 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
307 	    &state) == 0 &&
308 	    (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
309 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
310 		if ((ne = zutil_alloc(hdl, sizeof (name_entry_t))) == NULL)
311 			return (-1);
312 
313 		if ((ne->ne_name = zutil_strdup(hdl, path)) == NULL) {
314 			free(ne);
315 			return (-1);
316 		}
317 
318 		ne->ne_guid = vdev_guid;
319 		ne->ne_next = pl->names;
320 		pl->names = ne;
321 
322 		return (0);
323 	}
324 
325 	/*
326 	 * If we have a valid config but cannot read any of these fields, then
327 	 * it means we have a half-initialized label.  In vdev_label_init()
328 	 * we write a label with txg == 0 so that we can identify the device
329 	 * in case the user refers to the same disk later on.  If we fail to
330 	 * create the pool, we'll be left with a label in this state
331 	 * which should not be considered part of a valid pool.
332 	 */
333 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
334 	    &pool_guid) != 0 ||
335 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
336 	    &vdev_guid) != 0 ||
337 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
338 	    &top_guid) != 0 ||
339 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
340 	    &txg) != 0 || txg == 0) {
341 		return (0);
342 	}
343 
344 	/*
345 	 * First, see if we know about this pool.  If not, then add it to the
346 	 * list of known pools.
347 	 */
348 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
349 		if (pe->pe_guid == pool_guid)
350 			break;
351 	}
352 
353 	if (pe == NULL) {
354 		if ((pe = zutil_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
355 			return (-1);
356 		}
357 		pe->pe_guid = pool_guid;
358 		pe->pe_next = pl->pools;
359 		pl->pools = pe;
360 	}
361 
362 	/*
363 	 * Second, see if we know about this toplevel vdev.  Add it if its
364 	 * missing.
365 	 */
366 	for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
367 		if (ve->ve_guid == top_guid)
368 			break;
369 	}
370 
371 	if (ve == NULL) {
372 		if ((ve = zutil_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
373 			return (-1);
374 		}
375 		ve->ve_guid = top_guid;
376 		ve->ve_next = pe->pe_vdevs;
377 		pe->pe_vdevs = ve;
378 	}
379 
380 	/*
381 	 * Third, see if we have a config with a matching transaction group.  If
382 	 * so, then we do nothing.  Otherwise, add it to the list of known
383 	 * configs.
384 	 */
385 	for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
386 		if (ce->ce_txg == txg)
387 			break;
388 	}
389 
390 	if (ce == NULL) {
391 		if ((ce = zutil_alloc(hdl, sizeof (config_entry_t))) == NULL) {
392 			return (-1);
393 		}
394 		ce->ce_txg = txg;
395 		ce->ce_config = fnvlist_dup(config);
396 		ce->ce_next = ve->ve_configs;
397 		ve->ve_configs = ce;
398 	}
399 
400 	/*
401 	 * At this point we've successfully added our config to the list of
402 	 * known configs.  The last thing to do is add the vdev guid -> path
403 	 * mappings so that we can fix up the configuration as necessary before
404 	 * doing the import.
405 	 */
406 	if ((ne = zutil_alloc(hdl, sizeof (name_entry_t))) == NULL)
407 		return (-1);
408 
409 	if ((ne->ne_name = zutil_strdup(hdl, path)) == NULL) {
410 		free(ne);
411 		return (-1);
412 	}
413 
414 	ne->ne_guid = vdev_guid;
415 	ne->ne_next = pl->names;
416 	pl->names = ne;
417 
418 	return (0);
419 }
420 
421 /*
422  * Returns true if the named pool matches the given GUID.
423  */
424 static int
zutil_pool_active(libpc_handle_t * hdl,const char * name,uint64_t guid,boolean_t * isactive)425 zutil_pool_active(libpc_handle_t *hdl, const char *name, uint64_t guid,
426     boolean_t *isactive)
427 {
428 	ASSERT(hdl->lpc_ops->pco_pool_active != NULL);
429 
430 	int error = hdl->lpc_ops->pco_pool_active(hdl->lpc_lib_handle, name,
431 	    guid, isactive);
432 
433 	return (error);
434 }
435 
436 static nvlist_t *
zutil_refresh_config(libpc_handle_t * hdl,nvlist_t * tryconfig)437 zutil_refresh_config(libpc_handle_t *hdl, nvlist_t *tryconfig)
438 {
439 	ASSERT(hdl->lpc_ops->pco_refresh_config != NULL);
440 
441 	return (hdl->lpc_ops->pco_refresh_config(hdl->lpc_lib_handle,
442 	    tryconfig));
443 }
444 
445 /*
446  * Determine if the vdev id is a hole in the namespace.
447  */
448 static boolean_t
vdev_is_hole(uint64_t * hole_array,uint_t holes,uint_t id)449 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
450 {
451 	for (int c = 0; c < holes; c++) {
452 
453 		/* Top-level is a hole */
454 		if (hole_array[c] == id)
455 			return (B_TRUE);
456 	}
457 	return (B_FALSE);
458 }
459 
460 /*
461  * Convert our list of pools into the definitive set of configurations.  We
462  * start by picking the best config for each toplevel vdev.  Once that's done,
463  * we assemble the toplevel vdevs into a full config for the pool.  We make a
464  * pass to fix up any incorrect paths, and then add it to the main list to
465  * return to the user.
466  */
467 static nvlist_t *
get_configs(libpc_handle_t * hdl,pool_list_t * pl,boolean_t active_ok,nvlist_t * policy)468 get_configs(libpc_handle_t *hdl, pool_list_t *pl, boolean_t active_ok,
469     nvlist_t *policy)
470 {
471 	pool_entry_t *pe;
472 	vdev_entry_t *ve;
473 	config_entry_t *ce;
474 	nvlist_t *ret = NULL, *config = NULL, *tmp = NULL, *nvtop, *nvroot;
475 	nvlist_t **spares, **l2cache;
476 	uint_t i, nspares, nl2cache;
477 	boolean_t config_seen;
478 	uint64_t best_txg;
479 	char *name, *hostname = NULL;
480 	uint64_t guid;
481 	uint_t children = 0;
482 	nvlist_t **child = NULL;
483 	uint_t holes;
484 	uint64_t *hole_array, max_id;
485 	uint_t c;
486 	boolean_t isactive;
487 	uint64_t hostid;
488 	nvlist_t *nvl;
489 	boolean_t found_one = B_FALSE;
490 	boolean_t valid_top_config = B_FALSE;
491 
492 	if (nvlist_alloc(&ret, 0, 0) != 0)
493 		goto nomem;
494 
495 	for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
496 		uint64_t id, max_txg = 0;
497 
498 		if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
499 			goto nomem;
500 		config_seen = B_FALSE;
501 
502 		/*
503 		 * Iterate over all toplevel vdevs.  Grab the pool configuration
504 		 * from the first one we find, and then go through the rest and
505 		 * add them as necessary to the 'vdevs' member of the config.
506 		 */
507 		for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
508 
509 			/*
510 			 * Determine the best configuration for this vdev by
511 			 * selecting the config with the latest transaction
512 			 * group.
513 			 */
514 			best_txg = 0;
515 			for (ce = ve->ve_configs; ce != NULL;
516 			    ce = ce->ce_next) {
517 
518 				if (ce->ce_txg > best_txg) {
519 					tmp = ce->ce_config;
520 					best_txg = ce->ce_txg;
521 				}
522 			}
523 
524 			/*
525 			 * We rely on the fact that the max txg for the
526 			 * pool will contain the most up-to-date information
527 			 * about the valid top-levels in the vdev namespace.
528 			 */
529 			if (best_txg > max_txg) {
530 				(void) nvlist_remove(config,
531 				    ZPOOL_CONFIG_VDEV_CHILDREN,
532 				    DATA_TYPE_UINT64);
533 				(void) nvlist_remove(config,
534 				    ZPOOL_CONFIG_HOLE_ARRAY,
535 				    DATA_TYPE_UINT64_ARRAY);
536 
537 				max_txg = best_txg;
538 				hole_array = NULL;
539 				holes = 0;
540 				max_id = 0;
541 				valid_top_config = B_FALSE;
542 
543 				if (nvlist_lookup_uint64(tmp,
544 				    ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
545 					verify(nvlist_add_uint64(config,
546 					    ZPOOL_CONFIG_VDEV_CHILDREN,
547 					    max_id) == 0);
548 					valid_top_config = B_TRUE;
549 				}
550 
551 				if (nvlist_lookup_uint64_array(tmp,
552 				    ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
553 				    &holes) == 0) {
554 					verify(nvlist_add_uint64_array(config,
555 					    ZPOOL_CONFIG_HOLE_ARRAY,
556 					    hole_array, holes) == 0);
557 				}
558 			}
559 
560 			if (!config_seen) {
561 				/*
562 				 * Copy the relevant pieces of data to the pool
563 				 * configuration:
564 				 *
565 				 *	version
566 				 *	pool guid
567 				 *	name
568 				 *	comment (if available)
569 				 *	pool state
570 				 *	hostid (if available)
571 				 *	hostname (if available)
572 				 */
573 				uint64_t state, version;
574 				char *comment = NULL;
575 
576 				version = fnvlist_lookup_uint64(tmp,
577 				    ZPOOL_CONFIG_VERSION);
578 				fnvlist_add_uint64(config,
579 				    ZPOOL_CONFIG_VERSION, version);
580 				guid = fnvlist_lookup_uint64(tmp,
581 				    ZPOOL_CONFIG_POOL_GUID);
582 				fnvlist_add_uint64(config,
583 				    ZPOOL_CONFIG_POOL_GUID, guid);
584 				name = fnvlist_lookup_string(tmp,
585 				    ZPOOL_CONFIG_POOL_NAME);
586 				fnvlist_add_string(config,
587 				    ZPOOL_CONFIG_POOL_NAME, name);
588 
589 				if (nvlist_lookup_string(tmp,
590 				    ZPOOL_CONFIG_COMMENT, &comment) == 0)
591 					fnvlist_add_string(config,
592 					    ZPOOL_CONFIG_COMMENT, comment);
593 
594 				state = fnvlist_lookup_uint64(tmp,
595 				    ZPOOL_CONFIG_POOL_STATE);
596 				fnvlist_add_uint64(config,
597 				    ZPOOL_CONFIG_POOL_STATE, state);
598 
599 				hostid = 0;
600 				if (nvlist_lookup_uint64(tmp,
601 				    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
602 					fnvlist_add_uint64(config,
603 					    ZPOOL_CONFIG_HOSTID, hostid);
604 					hostname = fnvlist_lookup_string(tmp,
605 					    ZPOOL_CONFIG_HOSTNAME);
606 					fnvlist_add_string(config,
607 					    ZPOOL_CONFIG_HOSTNAME, hostname);
608 				}
609 
610 				config_seen = B_TRUE;
611 			}
612 
613 			/*
614 			 * Add this top-level vdev to the child array.
615 			 */
616 			verify(nvlist_lookup_nvlist(tmp,
617 			    ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
618 			verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
619 			    &id) == 0);
620 
621 			if (id >= children) {
622 				nvlist_t **newchild;
623 
624 				newchild = zutil_alloc(hdl, (id + 1) *
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 = id + 1;
635 			}
636 			if (nvlist_dup(nvtop, &child[id], 0) != 0)
637 				goto nomem;
638 
639 		}
640 
641 		/*
642 		 * If we have information about all the top-levels then
643 		 * clean up the nvlist which we've constructed. This
644 		 * means removing any extraneous devices that are
645 		 * beyond the valid range or adding devices to the end
646 		 * of our array which appear to be missing.
647 		 */
648 		if (valid_top_config) {
649 			if (max_id < children) {
650 				for (c = max_id; c < children; c++)
651 					nvlist_free(child[c]);
652 				children = max_id;
653 			} else if (max_id > children) {
654 				nvlist_t **newchild;
655 
656 				newchild = zutil_alloc(hdl, (max_id) *
657 				    sizeof (nvlist_t *));
658 				if (newchild == NULL)
659 					goto nomem;
660 
661 				for (c = 0; c < children; c++)
662 					newchild[c] = child[c];
663 
664 				free(child);
665 				child = newchild;
666 				children = max_id;
667 			}
668 		}
669 
670 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
671 		    &guid) == 0);
672 
673 		/*
674 		 * The vdev namespace may contain holes as a result of
675 		 * device removal. We must add them back into the vdev
676 		 * tree before we process any missing devices.
677 		 */
678 		if (holes > 0) {
679 			ASSERT(valid_top_config);
680 
681 			for (c = 0; c < children; c++) {
682 				nvlist_t *holey;
683 
684 				if (child[c] != NULL ||
685 				    !vdev_is_hole(hole_array, holes, c))
686 					continue;
687 
688 				if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
689 				    0) != 0)
690 					goto nomem;
691 
692 				/*
693 				 * Holes in the namespace are treated as
694 				 * "hole" top-level vdevs and have a
695 				 * special flag set on them.
696 				 */
697 				if (nvlist_add_string(holey,
698 				    ZPOOL_CONFIG_TYPE,
699 				    VDEV_TYPE_HOLE) != 0 ||
700 				    nvlist_add_uint64(holey,
701 				    ZPOOL_CONFIG_ID, c) != 0 ||
702 				    nvlist_add_uint64(holey,
703 				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
704 					nvlist_free(holey);
705 					goto nomem;
706 				}
707 				child[c] = holey;
708 			}
709 		}
710 
711 		/*
712 		 * Look for any missing top-level vdevs.  If this is the case,
713 		 * create a faked up 'missing' vdev as a placeholder.  We cannot
714 		 * simply compress the child array, because the kernel performs
715 		 * certain checks to make sure the vdev IDs match their location
716 		 * in the configuration.
717 		 */
718 		for (c = 0; c < children; c++) {
719 			if (child[c] == NULL) {
720 				nvlist_t *missing;
721 				if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
722 				    0) != 0)
723 					goto nomem;
724 				if (nvlist_add_string(missing,
725 				    ZPOOL_CONFIG_TYPE,
726 				    VDEV_TYPE_MISSING) != 0 ||
727 				    nvlist_add_uint64(missing,
728 				    ZPOOL_CONFIG_ID, c) != 0 ||
729 				    nvlist_add_uint64(missing,
730 				    ZPOOL_CONFIG_GUID, 0ULL) != 0) {
731 					nvlist_free(missing);
732 					goto nomem;
733 				}
734 				child[c] = missing;
735 			}
736 		}
737 
738 		/*
739 		 * Put all of this pool's top-level vdevs into a root vdev.
740 		 */
741 		if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
742 			goto nomem;
743 		if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
744 		    VDEV_TYPE_ROOT) != 0 ||
745 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
746 		    nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
747 		    nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
748 		    child, children) != 0) {
749 			nvlist_free(nvroot);
750 			goto nomem;
751 		}
752 
753 		for (c = 0; c < children; c++)
754 			nvlist_free(child[c]);
755 		free(child);
756 		children = 0;
757 		child = NULL;
758 
759 		/*
760 		 * Go through and fix up any paths and/or devids based on our
761 		 * known list of vdev GUID -> path mappings.
762 		 */
763 		if (fix_paths(nvroot, pl->names) != 0) {
764 			nvlist_free(nvroot);
765 			goto nomem;
766 		}
767 
768 		/*
769 		 * Add the root vdev to this pool's configuration.
770 		 */
771 		if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
772 		    nvroot) != 0) {
773 			nvlist_free(nvroot);
774 			goto nomem;
775 		}
776 		nvlist_free(nvroot);
777 
778 		/*
779 		 * zdb uses this path to report on active pools that were
780 		 * imported or created using -R.
781 		 */
782 		if (active_ok)
783 			goto add_pool;
784 
785 		/*
786 		 * Determine if this pool is currently active, in which case we
787 		 * can't actually import it.
788 		 */
789 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
790 		    &name) == 0);
791 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
792 		    &guid) == 0);
793 
794 		if (zutil_pool_active(hdl, name, guid, &isactive) != 0)
795 			goto error;
796 
797 		if (isactive) {
798 			nvlist_free(config);
799 			config = NULL;
800 			continue;
801 		}
802 
803 		if (policy != NULL) {
804 			if (nvlist_add_nvlist(config, ZPOOL_LOAD_POLICY,
805 			    policy) != 0)
806 				goto nomem;
807 		}
808 
809 		if ((nvl = zutil_refresh_config(hdl, config)) == NULL) {
810 			nvlist_free(config);
811 			config = NULL;
812 			continue;
813 		}
814 
815 		nvlist_free(config);
816 		config = nvl;
817 
818 		/*
819 		 * Go through and update the paths for spares, now that we have
820 		 * them.
821 		 */
822 		verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
823 		    &nvroot) == 0);
824 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
825 		    &spares, &nspares) == 0) {
826 			for (i = 0; i < nspares; i++) {
827 				if (fix_paths(spares[i], pl->names) != 0)
828 					goto nomem;
829 			}
830 		}
831 
832 		/*
833 		 * Update the paths for l2cache devices.
834 		 */
835 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
836 		    &l2cache, &nl2cache) == 0) {
837 			for (i = 0; i < nl2cache; i++) {
838 				if (fix_paths(l2cache[i], pl->names) != 0)
839 					goto nomem;
840 			}
841 		}
842 
843 		/*
844 		 * Restore the original information read from the actual label.
845 		 */
846 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
847 		    DATA_TYPE_UINT64);
848 		(void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
849 		    DATA_TYPE_STRING);
850 		if (hostid != 0) {
851 			verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
852 			    hostid) == 0);
853 			verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
854 			    hostname) == 0);
855 		}
856 
857 add_pool:
858 		/*
859 		 * Add this pool to the list of configs.
860 		 */
861 		verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
862 		    &name) == 0);
863 		if (nvlist_add_nvlist(ret, name, config) != 0)
864 			goto nomem;
865 
866 		found_one = B_TRUE;
867 		nvlist_free(config);
868 		config = NULL;
869 	}
870 
871 	if (!found_one) {
872 		nvlist_free(ret);
873 		ret = NULL;
874 	}
875 
876 	return (ret);
877 
878 nomem:
879 	(void) zutil_no_memory(hdl);
880 error:
881 	nvlist_free(config);
882 	nvlist_free(ret);
883 	for (c = 0; c < children; c++)
884 		nvlist_free(child[c]);
885 	free(child);
886 
887 	return (NULL);
888 }
889 
890 /*
891  * Return the offset of the given label.
892  */
893 static uint64_t
label_offset(uint64_t size,int l)894 label_offset(uint64_t size, int l)
895 {
896 	ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
897 	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
898 	    0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
899 }
900 
901 /*
902  * Given a file descriptor, read the label information and return an nvlist
903  * describing the configuration, if there is one. The number of valid
904  * labels found will be returned in num_labels when non-NULL.
905  */
906 int
zpool_read_label(int fd,nvlist_t ** config,int * num_labels)907 zpool_read_label(int fd, nvlist_t **config, int *num_labels)
908 {
909 	struct stat64 statbuf;
910 	int l, count = 0;
911 	vdev_label_t *label;
912 	nvlist_t *expected_config = NULL;
913 	uint64_t expected_guid = 0, size;
914 
915 	*config = NULL;
916 
917 	if (num_labels != NULL)
918 		*num_labels = 0;
919 
920 	if (fstat64(fd, &statbuf) == -1)
921 		return (0);
922 	size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
923 
924 	if ((label = malloc(sizeof (vdev_label_t))) == NULL)
925 		return (-1);
926 
927 	for (l = 0; l < VDEV_LABELS; l++) {
928 		uint64_t state, guid, txg;
929 
930 		if (pread64(fd, label, sizeof (vdev_label_t),
931 		    label_offset(size, l)) != sizeof (vdev_label_t))
932 			continue;
933 
934 		if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
935 		    sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
936 			continue;
937 
938 		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_GUID,
939 		    &guid) != 0 || guid == 0) {
940 			nvlist_free(*config);
941 			continue;
942 		}
943 
944 		if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
945 		    &state) != 0 || state > POOL_STATE_L2CACHE) {
946 			nvlist_free(*config);
947 			continue;
948 		}
949 
950 		if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
951 		    (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
952 		    &txg) != 0 || txg == 0)) {
953 			nvlist_free(*config);
954 			continue;
955 		}
956 
957 		if (expected_guid) {
958 			if (expected_guid == guid)
959 				count++;
960 
961 			nvlist_free(*config);
962 		} else {
963 			expected_config = *config;
964 			expected_guid = guid;
965 			count++;
966 		}
967 	}
968 
969 	if (num_labels != NULL)
970 		*num_labels = count;
971 
972 	free(label);
973 	*config = expected_config;
974 
975 	return (0);
976 }
977 
978 static int
slice_cache_compare(const void * arg1,const void * arg2)979 slice_cache_compare(const void *arg1, const void *arg2)
980 {
981 	const char  *nm1 = ((rdsk_node_t *)arg1)->rn_name;
982 	const char  *nm2 = ((rdsk_node_t *)arg2)->rn_name;
983 	char *nm1slice, *nm2slice;
984 	int rv;
985 
986 	/*
987 	 * slices zero and two are the most likely to provide results,
988 	 * so put those first
989 	 */
990 	nm1slice = strstr(nm1, "s0");
991 	nm2slice = strstr(nm2, "s0");
992 	if (nm1slice && !nm2slice) {
993 		return (-1);
994 	}
995 	if (!nm1slice && nm2slice) {
996 		return (1);
997 	}
998 	nm1slice = strstr(nm1, "s2");
999 	nm2slice = strstr(nm2, "s2");
1000 	if (nm1slice && !nm2slice) {
1001 		return (-1);
1002 	}
1003 	if (!nm1slice && nm2slice) {
1004 		return (1);
1005 	}
1006 
1007 	rv = strcmp(nm1, nm2);
1008 	if (rv == 0)
1009 		return (0);
1010 	return (rv > 0 ? 1 : -1);
1011 }
1012 
1013 static void
check_one_slice(avl_tree_t * r,char * diskname,uint_t partno,diskaddr_t size,uint_t blksz)1014 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
1015     diskaddr_t size, uint_t blksz)
1016 {
1017 	rdsk_node_t tmpnode;
1018 	rdsk_node_t *node;
1019 	char sname[MAXNAMELEN];
1020 
1021 	tmpnode.rn_name = &sname[0];
1022 	(void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
1023 	    diskname, partno);
1024 	/*
1025 	 * protect against division by zero for disk labels that
1026 	 * contain a bogus sector size
1027 	 */
1028 	if (blksz == 0)
1029 		blksz = DEV_BSIZE;
1030 	/* too small to contain a zpool? */
1031 	if ((size < (SPA_MINDEVSIZE / blksz)) &&
1032 	    (node = avl_find(r, &tmpnode, NULL)))
1033 		node->rn_nozpool = B_TRUE;
1034 }
1035 
1036 static void
nozpool_all_slices(avl_tree_t * r,const char * sname)1037 nozpool_all_slices(avl_tree_t *r, const char *sname)
1038 {
1039 	char diskname[MAXNAMELEN];
1040 	char *ptr;
1041 	int i;
1042 
1043 	(void) strncpy(diskname, sname, MAXNAMELEN);
1044 	if (((ptr = strrchr(diskname, 's')) == NULL) &&
1045 	    ((ptr = strrchr(diskname, 'p')) == NULL))
1046 		return;
1047 	ptr[0] = 's';
1048 	ptr[1] = '\0';
1049 	for (i = 0; i < NDKMAP; i++)
1050 		check_one_slice(r, diskname, i, 0, 1);
1051 	ptr[0] = 'p';
1052 	for (i = 0; i <= FD_NUMPART; i++)
1053 		check_one_slice(r, diskname, i, 0, 1);
1054 }
1055 
1056 static void
check_slices(avl_tree_t * r,int fd,const char * sname)1057 check_slices(avl_tree_t *r, int fd, const char *sname)
1058 {
1059 	struct extvtoc vtoc;
1060 	struct dk_gpt *gpt;
1061 	char diskname[MAXNAMELEN];
1062 	char *ptr;
1063 	int i;
1064 
1065 	(void) strncpy(diskname, sname, MAXNAMELEN);
1066 	if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1067 		return;
1068 	ptr[1] = '\0';
1069 
1070 	if (read_extvtoc(fd, &vtoc) >= 0) {
1071 		for (i = 0; i < NDKMAP; i++)
1072 			check_one_slice(r, diskname, i,
1073 			    vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1074 	} else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1075 		/*
1076 		 * on x86 we'll still have leftover links that point
1077 		 * to slices s[9-15], so use NDKMAP instead
1078 		 */
1079 		for (i = 0; i < NDKMAP; i++)
1080 			check_one_slice(r, diskname, i,
1081 			    gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1082 		/* nodes p[1-4] are never used with EFI labels */
1083 		ptr[0] = 'p';
1084 		for (i = 1; i <= FD_NUMPART; i++)
1085 			check_one_slice(r, diskname, i, 0, 1);
1086 		efi_free(gpt);
1087 	}
1088 }
1089 
1090 void
zpool_open_func(void * arg)1091 zpool_open_func(void *arg)
1092 {
1093 	rdsk_node_t *rn = arg;
1094 	struct stat64 statbuf;
1095 	nvlist_t *config;
1096 	int error;
1097 	int num_labels = 0;
1098 	int fd;
1099 
1100 	if (rn->rn_nozpool)
1101 		return;
1102 	if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1103 		/* symlink to a device that's no longer there */
1104 		if (errno == ENOENT)
1105 			nozpool_all_slices(rn->rn_avl, rn->rn_name);
1106 		return;
1107 	}
1108 	/*
1109 	 * Ignore failed stats.  We only want regular
1110 	 * files, character devs and block devs.
1111 	 */
1112 	if (fstat64(fd, &statbuf) != 0 ||
1113 	    (!S_ISREG(statbuf.st_mode) &&
1114 	    !S_ISCHR(statbuf.st_mode) &&
1115 	    !S_ISBLK(statbuf.st_mode))) {
1116 		(void) close(fd);
1117 		return;
1118 	}
1119 	/* this file is too small to hold a zpool */
1120 	if (S_ISREG(statbuf.st_mode) &&
1121 	    statbuf.st_size < SPA_MINDEVSIZE) {
1122 		(void) close(fd);
1123 		return;
1124 	} else if (!S_ISREG(statbuf.st_mode)) {
1125 		/*
1126 		 * Try to read the disk label first so we don't have to
1127 		 * open a bunch of minor nodes that can't have a zpool.
1128 		 */
1129 		check_slices(rn->rn_avl, fd, rn->rn_name);
1130 	}
1131 
1132 	error = zpool_read_label(fd, &config, &num_labels);
1133 	if (error != 0) {
1134 		(void) close(fd);
1135 		return;
1136 	}
1137 
1138 	if (num_labels == 0) {
1139 		(void) close(fd);
1140 		nvlist_free(config);
1141 		return;
1142 	}
1143 
1144 	(void) close(fd);
1145 
1146 	rn->rn_config = config;
1147 	rn->rn_num_labels = num_labels;
1148 }
1149 
1150 /*
1151  * Given a list of directories to search, find all pools stored on disk.  This
1152  * includes partial pools which are not available to import.  If no args are
1153  * given (argc is 0), then the default directory (/dev/dsk) is searched.
1154  * poolname or guid (but not both) are provided by the caller when trying
1155  * to import a specific pool.
1156  */
1157 static nvlist_t *
zpool_find_import_impl(libpc_handle_t * hdl,importargs_t * iarg)1158 zpool_find_import_impl(libpc_handle_t *hdl, importargs_t *iarg)
1159 {
1160 	int i, dirs = iarg->paths;
1161 	struct dirent64 *dp;
1162 	char path[MAXPATHLEN];
1163 	char *end, **dir = iarg->path;
1164 	size_t pathleft;
1165 	nvlist_t *ret = NULL;
1166 	static char *default_dir = ZFS_DISK_ROOT;
1167 	pool_list_t pools = { 0 };
1168 	pool_entry_t *pe, *penext;
1169 	vdev_entry_t *ve, *venext;
1170 	config_entry_t *ce, *cenext;
1171 	name_entry_t *ne, *nenext;
1172 	avl_tree_t slice_cache;
1173 	rdsk_node_t *slice;
1174 	void *cookie;
1175 
1176 	if (dirs == 0) {
1177 		dirs = 1;
1178 		dir = &default_dir;
1179 	}
1180 
1181 	/*
1182 	 * Go through and read the label configuration information from every
1183 	 * possible device, organizing the information according to pool GUID
1184 	 * and toplevel GUID.
1185 	 */
1186 	for (i = 0; i < dirs; i++) {
1187 		tpool_t *t;
1188 		char rdsk[MAXPATHLEN];
1189 		int dfd;
1190 		boolean_t config_failed = B_FALSE;
1191 		DIR *dirp;
1192 
1193 		/* use realpath to normalize the path */
1194 		if (realpath(dir[i], path) == 0) {
1195 			(void) zutil_error_fmt(hdl, EZFS_BADPATH,
1196 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1197 			goto error;
1198 		}
1199 		end = &path[strlen(path)];
1200 		*end++ = '/';
1201 		*end = 0;
1202 		pathleft = &path[sizeof (path)] - end;
1203 
1204 		/*
1205 		 * Using raw devices instead of block devices when we're
1206 		 * reading the labels skips a bunch of slow operations during
1207 		 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1208 		 */
1209 		if (strcmp(path, ZFS_DISK_ROOTD) == 0)
1210 			(void) strlcpy(rdsk, ZFS_RDISK_ROOTD, sizeof (rdsk));
1211 		else
1212 			(void) strlcpy(rdsk, path, sizeof (rdsk));
1213 
1214 		if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1215 		    (dirp = fdopendir(dfd)) == NULL) {
1216 			if (dfd >= 0)
1217 				(void) close(dfd);
1218 			zutil_error_aux(hdl, strerror(errno));
1219 			(void) zutil_error_fmt(hdl, EZFS_BADPATH,
1220 			    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1221 			    rdsk);
1222 			goto error;
1223 		}
1224 
1225 		avl_create(&slice_cache, slice_cache_compare,
1226 		    sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1227 		/*
1228 		 * This is not MT-safe, but we have no MT consumers of libzutil
1229 		 */
1230 		while ((dp = readdir64(dirp)) != NULL) {
1231 			const char *name = dp->d_name;
1232 			if (name[0] == '.' &&
1233 			    (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1234 				continue;
1235 
1236 			slice = zutil_alloc(hdl, sizeof (rdsk_node_t));
1237 			slice->rn_name = zutil_strdup(hdl, name);
1238 			slice->rn_avl = &slice_cache;
1239 			slice->rn_dfd = dfd;
1240 			slice->rn_hdl = hdl;
1241 			slice->rn_nozpool = B_FALSE;
1242 			avl_add(&slice_cache, slice);
1243 		}
1244 		/*
1245 		 * create a thread pool to do all of this in parallel;
1246 		 * rn_nozpool is not protected, so this is racy in that
1247 		 * multiple tasks could decide that the same slice can
1248 		 * not hold a zpool, which is benign.  Also choose
1249 		 * double the number of processors; we hold a lot of
1250 		 * locks in the kernel, so going beyond this doesn't
1251 		 * buy us much.
1252 		 */
1253 		t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1254 		    0, NULL);
1255 		for (slice = avl_first(&slice_cache); slice;
1256 		    (slice = avl_walk(&slice_cache, slice,
1257 		    AVL_AFTER)))
1258 			(void) tpool_dispatch(t, zpool_open_func, slice);
1259 		tpool_wait(t);
1260 		tpool_destroy(t);
1261 
1262 		cookie = NULL;
1263 		while ((slice = avl_destroy_nodes(&slice_cache,
1264 		    &cookie)) != NULL) {
1265 			if (slice->rn_config != NULL && !config_failed) {
1266 				nvlist_t *config = slice->rn_config;
1267 				boolean_t matched = B_TRUE;
1268 
1269 				if (iarg->poolname != NULL) {
1270 					char *pname;
1271 
1272 					matched = nvlist_lookup_string(config,
1273 					    ZPOOL_CONFIG_POOL_NAME,
1274 					    &pname) == 0 &&
1275 					    strcmp(iarg->poolname, pname) == 0;
1276 				} else if (iarg->guid != 0) {
1277 					uint64_t this_guid;
1278 
1279 					matched = nvlist_lookup_uint64(config,
1280 					    ZPOOL_CONFIG_POOL_GUID,
1281 					    &this_guid) == 0 &&
1282 					    iarg->guid == this_guid;
1283 				}
1284 				if (matched) {
1285 					/*
1286 					 * use the non-raw path for the config
1287 					 */
1288 					(void) strlcpy(end, slice->rn_name,
1289 					    pathleft);
1290 					(void) add_config(hdl, &pools,
1291 					    path, slice->rn_order,
1292 					    slice->rn_num_labels, config);
1293 				}
1294 				nvlist_free(config);
1295 			}
1296 			free(slice->rn_name);
1297 			free(slice);
1298 		}
1299 		avl_destroy(&slice_cache);
1300 
1301 		(void) closedir(dirp);
1302 
1303 		if (config_failed)
1304 			goto error;
1305 	}
1306 
1307 	ret = get_configs(hdl, &pools, iarg->can_be_active, iarg->policy);
1308 
1309 error:
1310 	for (pe = pools.pools; pe != NULL; pe = penext) {
1311 		penext = pe->pe_next;
1312 		for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1313 			venext = ve->ve_next;
1314 			for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1315 				cenext = ce->ce_next;
1316 				nvlist_free(ce->ce_config);
1317 				free(ce);
1318 			}
1319 			free(ve);
1320 		}
1321 		free(pe);
1322 	}
1323 
1324 	for (ne = pools.names; ne != NULL; ne = nenext) {
1325 		nenext = ne->ne_next;
1326 		free(ne->ne_name);
1327 		free(ne);
1328 	}
1329 
1330 	return (ret);
1331 }
1332 
1333 /*
1334  * Given a cache file, return the contents as a list of importable pools.
1335  * poolname or guid (but not both) are provided by the caller when trying
1336  * to import a specific pool.
1337  */
1338 static nvlist_t *
zpool_find_import_cached(libpc_handle_t * hdl,const char * cachefile,const char * poolname,uint64_t guid)1339 zpool_find_import_cached(libpc_handle_t *hdl, const char *cachefile,
1340     const char *poolname, uint64_t guid)
1341 {
1342 	char *buf;
1343 	int fd;
1344 	struct stat64 statbuf;
1345 	nvlist_t *raw, *src, *dst;
1346 	nvlist_t *pools;
1347 	nvpair_t *elem;
1348 	char *name;
1349 	uint64_t this_guid;
1350 	boolean_t active;
1351 
1352 	verify(poolname == NULL || guid == 0);
1353 
1354 	if ((fd = open(cachefile, O_RDONLY)) < 0) {
1355 		zutil_error_aux(hdl, "%s", strerror(errno));
1356 		(void) zutil_error(hdl, EZFS_BADCACHE,
1357 		    dgettext(TEXT_DOMAIN, "failed to open cache file"));
1358 		return (NULL);
1359 	}
1360 
1361 	if (fstat64(fd, &statbuf) != 0) {
1362 		zutil_error_aux(hdl, "%s", strerror(errno));
1363 		(void) close(fd);
1364 		(void) zutil_error(hdl, EZFS_BADCACHE,
1365 		    dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1366 		return (NULL);
1367 	}
1368 
1369 	if ((buf = zutil_alloc(hdl, statbuf.st_size)) == NULL) {
1370 		(void) close(fd);
1371 		return (NULL);
1372 	}
1373 
1374 	if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1375 		(void) close(fd);
1376 		free(buf);
1377 		(void) zutil_error(hdl, EZFS_BADCACHE,
1378 		    dgettext(TEXT_DOMAIN,
1379 		    "failed to read cache file contents"));
1380 		return (NULL);
1381 	}
1382 
1383 	(void) close(fd);
1384 
1385 	if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1386 		free(buf);
1387 		(void) zutil_error(hdl, EZFS_BADCACHE,
1388 		    dgettext(TEXT_DOMAIN,
1389 		    "invalid or corrupt cache file contents"));
1390 		return (NULL);
1391 	}
1392 
1393 	free(buf);
1394 
1395 	/*
1396 	 * Go through and get the current state of the pools and refresh their
1397 	 * state.
1398 	 */
1399 	if (nvlist_alloc(&pools, 0, 0) != 0) {
1400 		(void) zutil_no_memory(hdl);
1401 		nvlist_free(raw);
1402 		return (NULL);
1403 	}
1404 
1405 	elem = NULL;
1406 	while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1407 		src = fnvpair_value_nvlist(elem);
1408 
1409 		name = fnvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME);
1410 		if (poolname != NULL && strcmp(poolname, name) != 0)
1411 			continue;
1412 
1413 		this_guid = fnvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID);
1414 		if (guid != 0 && guid != this_guid)
1415 			continue;
1416 
1417 		if (zutil_pool_active(hdl, name, this_guid, &active) != 0) {
1418 			nvlist_free(raw);
1419 			nvlist_free(pools);
1420 			return (NULL);
1421 		}
1422 
1423 		if (active)
1424 			continue;
1425 
1426 		if (nvlist_add_string(src, ZPOOL_CONFIG_CACHEFILE,
1427 		    cachefile) != 0) {
1428 			(void) zutil_no_memory(hdl);
1429 			nvlist_free(raw);
1430 			nvlist_free(pools);
1431 			return (NULL);
1432 		}
1433 
1434 		if ((dst = zutil_refresh_config(hdl, src)) == NULL) {
1435 			nvlist_free(raw);
1436 			nvlist_free(pools);
1437 			return (NULL);
1438 		}
1439 
1440 		if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1441 			(void) zutil_no_memory(hdl);
1442 			nvlist_free(dst);
1443 			nvlist_free(raw);
1444 			nvlist_free(pools);
1445 			return (NULL);
1446 		}
1447 		nvlist_free(dst);
1448 	}
1449 
1450 	nvlist_free(raw);
1451 	return (pools);
1452 }
1453 
1454 nvlist_t *
zpool_search_import(void * hdl,importargs_t * import,const pool_config_ops_t * pco)1455 zpool_search_import(void *hdl, importargs_t *import,
1456     const pool_config_ops_t *pco)
1457 {
1458 	libpc_handle_t handle = { 0 };
1459 	nvlist_t *pools = NULL;
1460 
1461 	handle.lpc_lib_handle = hdl;
1462 	handle.lpc_ops = pco;
1463 	handle.lpc_printerr = B_TRUE;
1464 
1465 	verify(import->poolname == NULL || import->guid == 0);
1466 
1467 	if (import->cachefile != NULL)
1468 		pools = zpool_find_import_cached(&handle, import->cachefile,
1469 		    import->poolname, import->guid);
1470 	else
1471 		pools = zpool_find_import_impl(&handle, import);
1472 
1473 	if ((pools == NULL || nvlist_empty(pools)) &&
1474 	    handle.lpc_open_access_error && geteuid() != 0) {
1475 		(void) zutil_error(&handle, EZFS_EACESS, dgettext(TEXT_DOMAIN,
1476 		    "no pools found"));
1477 	}
1478 
1479 	return (pools);
1480 }
1481 
1482 static boolean_t
pool_match(nvlist_t * cfg,char * tgt)1483 pool_match(nvlist_t *cfg, char *tgt)
1484 {
1485 	uint64_t v, guid = strtoull(tgt, NULL, 0);
1486 	char *s;
1487 
1488 	if (guid != 0) {
1489 		if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &v) == 0)
1490 			return (v == guid);
1491 	} else {
1492 		if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &s) == 0)
1493 			return (strcmp(s, tgt) == 0);
1494 	}
1495 	return (B_FALSE);
1496 }
1497 
1498 int
zpool_find_config(void * hdl,const char * target,nvlist_t ** configp,importargs_t * args,const pool_config_ops_t * pco)1499 zpool_find_config(void *hdl, const char *target, nvlist_t **configp,
1500     importargs_t *args, const pool_config_ops_t *pco)
1501 {
1502 	nvlist_t *pools;
1503 	nvlist_t *match = NULL;
1504 	nvlist_t *config = NULL;
1505 	char *sepp = NULL;
1506 	int count = 0;
1507 	char *targetdup = strdup(target);
1508 
1509 	*configp = NULL;
1510 
1511 	if ((sepp = strpbrk(targetdup, "/@")) != NULL) {
1512 		*sepp = '\0';
1513 	}
1514 
1515 	pools = zpool_search_import(hdl, args, pco);
1516 
1517 	if (pools != NULL) {
1518 		nvpair_t *elem = NULL;
1519 		while ((elem = nvlist_next_nvpair(pools, elem)) != NULL) {
1520 			VERIFY0(nvpair_value_nvlist(elem, &config));
1521 			if (pool_match(config, targetdup)) {
1522 				count++;
1523 				if (match != NULL) {
1524 					/* multiple matches found */
1525 					continue;
1526 				} else {
1527 					match = config;
1528 				}
1529 			}
1530 		}
1531 	}
1532 
1533 	if (count == 0) {
1534 		free(targetdup);
1535 		return (ENOENT);
1536 	}
1537 
1538 	if (count > 1) {
1539 		free(targetdup);
1540 		return (EINVAL);
1541 	}
1542 
1543 	*configp = match;
1544 	free(targetdup);
1545 
1546 	return (0);
1547 }
1548