xref: /titanic_51/usr/src/lib/libzfs/common/libzfs_pool.c (revision e18306b13ed357bd545696aa96b53617b64db4a3)
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 2011 Nexenta Systems, Inc. All rights reserved.
25  * Copyright (c) 2012 by Delphix. All rights reserved.
26  */
27 
28 #include <ctype.h>
29 #include <errno.h>
30 #include <devid.h>
31 #include <fcntl.h>
32 #include <libintl.h>
33 #include <stdio.h>
34 #include <stdlib.h>
35 #include <strings.h>
36 #include <unistd.h>
37 #include <sys/efi_partition.h>
38 #include <sys/vtoc.h>
39 #include <sys/zfs_ioctl.h>
40 #include <dlfcn.h>
41 
42 #include "zfs_namecheck.h"
43 #include "zfs_prop.h"
44 #include "libzfs_impl.h"
45 #include "zfs_comutil.h"
46 #include "zfeature_common.h"
47 
48 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
49 
50 #define	DISK_ROOT	"/dev/dsk"
51 #define	RDISK_ROOT	"/dev/rdsk"
52 #define	BACKUP_SLICE	"s2"
53 
54 typedef struct prop_flags {
55 	int create:1;	/* Validate property on creation */
56 	int import:1;	/* Validate property on import */
57 } prop_flags_t;
58 
59 /*
60  * ====================================================================
61  *   zpool property functions
62  * ====================================================================
63  */
64 
65 static int
66 zpool_get_all_props(zpool_handle_t *zhp)
67 {
68 	zfs_cmd_t zc = { 0 };
69 	libzfs_handle_t *hdl = zhp->zpool_hdl;
70 
71 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
72 
73 	if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
74 		return (-1);
75 
76 	while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
77 		if (errno == ENOMEM) {
78 			if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
79 				zcmd_free_nvlists(&zc);
80 				return (-1);
81 			}
82 		} else {
83 			zcmd_free_nvlists(&zc);
84 			return (-1);
85 		}
86 	}
87 
88 	if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
89 		zcmd_free_nvlists(&zc);
90 		return (-1);
91 	}
92 
93 	zcmd_free_nvlists(&zc);
94 
95 	return (0);
96 }
97 
98 static int
99 zpool_props_refresh(zpool_handle_t *zhp)
100 {
101 	nvlist_t *old_props;
102 
103 	old_props = zhp->zpool_props;
104 
105 	if (zpool_get_all_props(zhp) != 0)
106 		return (-1);
107 
108 	nvlist_free(old_props);
109 	return (0);
110 }
111 
112 static char *
113 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
114     zprop_source_t *src)
115 {
116 	nvlist_t *nv, *nvl;
117 	uint64_t ival;
118 	char *value;
119 	zprop_source_t source;
120 
121 	nvl = zhp->zpool_props;
122 	if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
123 		verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
124 		source = ival;
125 		verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
126 	} else {
127 		source = ZPROP_SRC_DEFAULT;
128 		if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
129 			value = "-";
130 	}
131 
132 	if (src)
133 		*src = source;
134 
135 	return (value);
136 }
137 
138 uint64_t
139 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
140 {
141 	nvlist_t *nv, *nvl;
142 	uint64_t value;
143 	zprop_source_t source;
144 
145 	if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
146 		/*
147 		 * zpool_get_all_props() has most likely failed because
148 		 * the pool is faulted, but if all we need is the top level
149 		 * vdev's guid then get it from the zhp config nvlist.
150 		 */
151 		if ((prop == ZPOOL_PROP_GUID) &&
152 		    (nvlist_lookup_nvlist(zhp->zpool_config,
153 		    ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
154 		    (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
155 		    == 0)) {
156 			return (value);
157 		}
158 		return (zpool_prop_default_numeric(prop));
159 	}
160 
161 	nvl = zhp->zpool_props;
162 	if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
163 		verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
164 		source = value;
165 		verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
166 	} else {
167 		source = ZPROP_SRC_DEFAULT;
168 		value = zpool_prop_default_numeric(prop);
169 	}
170 
171 	if (src)
172 		*src = source;
173 
174 	return (value);
175 }
176 
177 /*
178  * Map VDEV STATE to printed strings.
179  */
180 char *
181 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
182 {
183 	switch (state) {
184 	case VDEV_STATE_CLOSED:
185 	case VDEV_STATE_OFFLINE:
186 		return (gettext("OFFLINE"));
187 	case VDEV_STATE_REMOVED:
188 		return (gettext("REMOVED"));
189 	case VDEV_STATE_CANT_OPEN:
190 		if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
191 			return (gettext("FAULTED"));
192 		else if (aux == VDEV_AUX_SPLIT_POOL)
193 			return (gettext("SPLIT"));
194 		else
195 			return (gettext("UNAVAIL"));
196 	case VDEV_STATE_FAULTED:
197 		return (gettext("FAULTED"));
198 	case VDEV_STATE_DEGRADED:
199 		return (gettext("DEGRADED"));
200 	case VDEV_STATE_HEALTHY:
201 		return (gettext("ONLINE"));
202 	}
203 
204 	return (gettext("UNKNOWN"));
205 }
206 
207 /*
208  * Get a zpool property value for 'prop' and return the value in
209  * a pre-allocated buffer.
210  */
211 int
212 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
213     zprop_source_t *srctype)
214 {
215 	uint64_t intval;
216 	const char *strval;
217 	zprop_source_t src = ZPROP_SRC_NONE;
218 	nvlist_t *nvroot;
219 	vdev_stat_t *vs;
220 	uint_t vsc;
221 
222 	if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
223 		switch (prop) {
224 		case ZPOOL_PROP_NAME:
225 			(void) strlcpy(buf, zpool_get_name(zhp), len);
226 			break;
227 
228 		case ZPOOL_PROP_HEALTH:
229 			(void) strlcpy(buf, "FAULTED", len);
230 			break;
231 
232 		case ZPOOL_PROP_GUID:
233 			intval = zpool_get_prop_int(zhp, prop, &src);
234 			(void) snprintf(buf, len, "%llu", intval);
235 			break;
236 
237 		case ZPOOL_PROP_ALTROOT:
238 		case ZPOOL_PROP_CACHEFILE:
239 		case ZPOOL_PROP_COMMENT:
240 			if (zhp->zpool_props != NULL ||
241 			    zpool_get_all_props(zhp) == 0) {
242 				(void) strlcpy(buf,
243 				    zpool_get_prop_string(zhp, prop, &src),
244 				    len);
245 				if (srctype != NULL)
246 					*srctype = src;
247 				return (0);
248 			}
249 			/* FALLTHROUGH */
250 		default:
251 			(void) strlcpy(buf, "-", len);
252 			break;
253 		}
254 
255 		if (srctype != NULL)
256 			*srctype = src;
257 		return (0);
258 	}
259 
260 	if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
261 	    prop != ZPOOL_PROP_NAME)
262 		return (-1);
263 
264 	switch (zpool_prop_get_type(prop)) {
265 	case PROP_TYPE_STRING:
266 		(void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
267 		    len);
268 		break;
269 
270 	case PROP_TYPE_NUMBER:
271 		intval = zpool_get_prop_int(zhp, prop, &src);
272 
273 		switch (prop) {
274 		case ZPOOL_PROP_SIZE:
275 		case ZPOOL_PROP_ALLOCATED:
276 		case ZPOOL_PROP_FREE:
277 		case ZPOOL_PROP_FREEING:
278 		case ZPOOL_PROP_EXPANDSZ:
279 			(void) zfs_nicenum(intval, buf, len);
280 			break;
281 
282 		case ZPOOL_PROP_CAPACITY:
283 			(void) snprintf(buf, len, "%llu%%",
284 			    (u_longlong_t)intval);
285 			break;
286 
287 		case ZPOOL_PROP_DEDUPRATIO:
288 			(void) snprintf(buf, len, "%llu.%02llux",
289 			    (u_longlong_t)(intval / 100),
290 			    (u_longlong_t)(intval % 100));
291 			break;
292 
293 		case ZPOOL_PROP_HEALTH:
294 			verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
295 			    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
296 			verify(nvlist_lookup_uint64_array(nvroot,
297 			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
298 			    == 0);
299 
300 			(void) strlcpy(buf, zpool_state_to_name(intval,
301 			    vs->vs_aux), len);
302 			break;
303 		case ZPOOL_PROP_VERSION:
304 			if (intval >= SPA_VERSION_FEATURES) {
305 				(void) snprintf(buf, len, "-");
306 				break;
307 			}
308 			/* FALLTHROUGH */
309 		default:
310 			(void) snprintf(buf, len, "%llu", intval);
311 		}
312 		break;
313 
314 	case PROP_TYPE_INDEX:
315 		intval = zpool_get_prop_int(zhp, prop, &src);
316 		if (zpool_prop_index_to_string(prop, intval, &strval)
317 		    != 0)
318 			return (-1);
319 		(void) strlcpy(buf, strval, len);
320 		break;
321 
322 	default:
323 		abort();
324 	}
325 
326 	if (srctype)
327 		*srctype = src;
328 
329 	return (0);
330 }
331 
332 /*
333  * Check if the bootfs name has the same pool name as it is set to.
334  * Assuming bootfs is a valid dataset name.
335  */
336 static boolean_t
337 bootfs_name_valid(const char *pool, char *bootfs)
338 {
339 	int len = strlen(pool);
340 
341 	if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
342 		return (B_FALSE);
343 
344 	if (strncmp(pool, bootfs, len) == 0 &&
345 	    (bootfs[len] == '/' || bootfs[len] == '\0'))
346 		return (B_TRUE);
347 
348 	return (B_FALSE);
349 }
350 
351 /*
352  * Inspect the configuration to determine if any of the devices contain
353  * an EFI label.
354  */
355 static boolean_t
356 pool_uses_efi(nvlist_t *config)
357 {
358 	nvlist_t **child;
359 	uint_t c, children;
360 
361 	if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
362 	    &child, &children) != 0)
363 		return (read_efi_label(config, NULL) >= 0);
364 
365 	for (c = 0; c < children; c++) {
366 		if (pool_uses_efi(child[c]))
367 			return (B_TRUE);
368 	}
369 	return (B_FALSE);
370 }
371 
372 boolean_t
373 zpool_is_bootable(zpool_handle_t *zhp)
374 {
375 	char bootfs[ZPOOL_MAXNAMELEN];
376 
377 	return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
378 	    sizeof (bootfs), NULL) == 0 && strncmp(bootfs, "-",
379 	    sizeof (bootfs)) != 0);
380 }
381 
382 
383 /*
384  * Given an nvlist of zpool properties to be set, validate that they are
385  * correct, and parse any numeric properties (index, boolean, etc) if they are
386  * specified as strings.
387  */
388 static nvlist_t *
389 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
390     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
391 {
392 	nvpair_t *elem;
393 	nvlist_t *retprops;
394 	zpool_prop_t prop;
395 	char *strval;
396 	uint64_t intval;
397 	char *slash, *check;
398 	struct stat64 statbuf;
399 	zpool_handle_t *zhp;
400 	nvlist_t *nvroot;
401 
402 	if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
403 		(void) no_memory(hdl);
404 		return (NULL);
405 	}
406 
407 	elem = NULL;
408 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
409 		const char *propname = nvpair_name(elem);
410 
411 		prop = zpool_name_to_prop(propname);
412 		if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
413 			int err;
414 			zfeature_info_t *feature;
415 			char *fname = strchr(propname, '@') + 1;
416 
417 			err = zfeature_lookup_name(fname, &feature);
418 			if (err != 0) {
419 				ASSERT3U(err, ==, ENOENT);
420 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
421 				    "invalid feature '%s'"), fname);
422 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
423 				goto error;
424 			}
425 
426 			if (nvpair_type(elem) != DATA_TYPE_STRING) {
427 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
428 				    "'%s' must be a string"), propname);
429 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
430 				goto error;
431 			}
432 
433 			(void) nvpair_value_string(elem, &strval);
434 			if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
435 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
436 				    "property '%s' can only be set to "
437 				    "'enabled'"), propname);
438 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
439 				goto error;
440 			}
441 
442 			if (nvlist_add_uint64(retprops, propname, 0) != 0) {
443 				(void) no_memory(hdl);
444 				goto error;
445 			}
446 			continue;
447 		}
448 
449 		/*
450 		 * Make sure this property is valid and applies to this type.
451 		 */
452 		if (prop == ZPROP_INVAL) {
453 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
454 			    "invalid property '%s'"), propname);
455 			(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
456 			goto error;
457 		}
458 
459 		if (zpool_prop_readonly(prop)) {
460 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
461 			    "is readonly"), propname);
462 			(void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
463 			goto error;
464 		}
465 
466 		if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
467 		    &strval, &intval, errbuf) != 0)
468 			goto error;
469 
470 		/*
471 		 * Perform additional checking for specific properties.
472 		 */
473 		switch (prop) {
474 		case ZPOOL_PROP_VERSION:
475 			if (intval < version ||
476 			    !SPA_VERSION_IS_SUPPORTED(intval)) {
477 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
478 				    "property '%s' number %d is invalid."),
479 				    propname, intval);
480 				(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
481 				goto error;
482 			}
483 			break;
484 
485 		case ZPOOL_PROP_BOOTFS:
486 			if (flags.create || flags.import) {
487 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
488 				    "property '%s' cannot be set at creation "
489 				    "or import time"), propname);
490 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
491 				goto error;
492 			}
493 
494 			if (version < SPA_VERSION_BOOTFS) {
495 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
496 				    "pool must be upgraded to support "
497 				    "'%s' property"), propname);
498 				(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
499 				goto error;
500 			}
501 
502 			/*
503 			 * bootfs property value has to be a dataset name and
504 			 * the dataset has to be in the same pool as it sets to.
505 			 */
506 			if (strval[0] != '\0' && !bootfs_name_valid(poolname,
507 			    strval)) {
508 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
509 				    "is an invalid name"), strval);
510 				(void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
511 				goto error;
512 			}
513 
514 			if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
515 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
516 				    "could not open pool '%s'"), poolname);
517 				(void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
518 				goto error;
519 			}
520 			verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
521 			    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
522 
523 			/*
524 			 * bootfs property cannot be set on a disk which has
525 			 * been EFI labeled.
526 			 */
527 			if (pool_uses_efi(nvroot)) {
528 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
529 				    "property '%s' not supported on "
530 				    "EFI labeled devices"), propname);
531 				(void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
532 				zpool_close(zhp);
533 				goto error;
534 			}
535 			zpool_close(zhp);
536 			break;
537 
538 		case ZPOOL_PROP_ALTROOT:
539 			if (!flags.create && !flags.import) {
540 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
541 				    "property '%s' can only be set during pool "
542 				    "creation or import"), propname);
543 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
544 				goto error;
545 			}
546 
547 			if (strval[0] != '/') {
548 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
549 				    "bad alternate root '%s'"), strval);
550 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
551 				goto error;
552 			}
553 			break;
554 
555 		case ZPOOL_PROP_CACHEFILE:
556 			if (strval[0] == '\0')
557 				break;
558 
559 			if (strcmp(strval, "none") == 0)
560 				break;
561 
562 			if (strval[0] != '/') {
563 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
564 				    "property '%s' must be empty, an "
565 				    "absolute path, or 'none'"), propname);
566 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
567 				goto error;
568 			}
569 
570 			slash = strrchr(strval, '/');
571 
572 			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
573 			    strcmp(slash, "/..") == 0) {
574 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
575 				    "'%s' is not a valid file"), strval);
576 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
577 				goto error;
578 			}
579 
580 			*slash = '\0';
581 
582 			if (strval[0] != '\0' &&
583 			    (stat64(strval, &statbuf) != 0 ||
584 			    !S_ISDIR(statbuf.st_mode))) {
585 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
586 				    "'%s' is not a valid directory"),
587 				    strval);
588 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
589 				goto error;
590 			}
591 
592 			*slash = '/';
593 			break;
594 
595 		case ZPOOL_PROP_COMMENT:
596 			for (check = strval; *check != '\0'; check++) {
597 				if (!isprint(*check)) {
598 					zfs_error_aux(hdl,
599 					    dgettext(TEXT_DOMAIN,
600 					    "comment may only have printable "
601 					    "characters"));
602 					(void) zfs_error(hdl, EZFS_BADPROP,
603 					    errbuf);
604 					goto error;
605 				}
606 			}
607 			if (strlen(strval) > ZPROP_MAX_COMMENT) {
608 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
609 				    "comment must not exceed %d characters"),
610 				    ZPROP_MAX_COMMENT);
611 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
612 				goto error;
613 			}
614 			break;
615 		case ZPOOL_PROP_READONLY:
616 			if (!flags.import) {
617 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
618 				    "property '%s' can only be set at "
619 				    "import time"), propname);
620 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
621 				goto error;
622 			}
623 			break;
624 		}
625 	}
626 
627 	return (retprops);
628 error:
629 	nvlist_free(retprops);
630 	return (NULL);
631 }
632 
633 /*
634  * Set zpool property : propname=propval.
635  */
636 int
637 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
638 {
639 	zfs_cmd_t zc = { 0 };
640 	int ret = -1;
641 	char errbuf[1024];
642 	nvlist_t *nvl = NULL;
643 	nvlist_t *realprops;
644 	uint64_t version;
645 	prop_flags_t flags = { 0 };
646 
647 	(void) snprintf(errbuf, sizeof (errbuf),
648 	    dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
649 	    zhp->zpool_name);
650 
651 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
652 		return (no_memory(zhp->zpool_hdl));
653 
654 	if (nvlist_add_string(nvl, propname, propval) != 0) {
655 		nvlist_free(nvl);
656 		return (no_memory(zhp->zpool_hdl));
657 	}
658 
659 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
660 	if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
661 	    zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
662 		nvlist_free(nvl);
663 		return (-1);
664 	}
665 
666 	nvlist_free(nvl);
667 	nvl = realprops;
668 
669 	/*
670 	 * Execute the corresponding ioctl() to set this property.
671 	 */
672 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
673 
674 	if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
675 		nvlist_free(nvl);
676 		return (-1);
677 	}
678 
679 	ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
680 
681 	zcmd_free_nvlists(&zc);
682 	nvlist_free(nvl);
683 
684 	if (ret)
685 		(void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
686 	else
687 		(void) zpool_props_refresh(zhp);
688 
689 	return (ret);
690 }
691 
692 int
693 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
694 {
695 	libzfs_handle_t *hdl = zhp->zpool_hdl;
696 	zprop_list_t *entry;
697 	char buf[ZFS_MAXPROPLEN];
698 	nvlist_t *features = NULL;
699 	zprop_list_t **last;
700 	boolean_t firstexpand = (NULL == *plp);
701 
702 	if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
703 		return (-1);
704 
705 	last = plp;
706 	while (*last != NULL)
707 		last = &(*last)->pl_next;
708 
709 	if ((*plp)->pl_all)
710 		features = zpool_get_features(zhp);
711 
712 	if ((*plp)->pl_all && firstexpand) {
713 		for (int i = 0; i < SPA_FEATURES; i++) {
714 			zprop_list_t *entry = zfs_alloc(hdl,
715 			    sizeof (zprop_list_t));
716 			entry->pl_prop = ZPROP_INVAL;
717 			entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
718 			    spa_feature_table[i].fi_uname);
719 			entry->pl_width = strlen(entry->pl_user_prop);
720 			entry->pl_all = B_TRUE;
721 
722 			*last = entry;
723 			last = &entry->pl_next;
724 		}
725 	}
726 
727 	/* add any unsupported features */
728 	for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
729 	    nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
730 		char *propname;
731 		boolean_t found;
732 		zprop_list_t *entry;
733 
734 		if (zfeature_is_supported(nvpair_name(nvp)))
735 			continue;
736 
737 		propname = zfs_asprintf(hdl, "unsupported@%s",
738 		    nvpair_name(nvp));
739 
740 		/*
741 		 * Before adding the property to the list make sure that no
742 		 * other pool already added the same property.
743 		 */
744 		found = B_FALSE;
745 		entry = *plp;
746 		while (entry != NULL) {
747 			if (entry->pl_user_prop != NULL &&
748 			    strcmp(propname, entry->pl_user_prop) == 0) {
749 				found = B_TRUE;
750 				break;
751 			}
752 			entry = entry->pl_next;
753 		}
754 		if (found) {
755 			free(propname);
756 			continue;
757 		}
758 
759 		entry = zfs_alloc(hdl, sizeof (zprop_list_t));
760 		entry->pl_prop = ZPROP_INVAL;
761 		entry->pl_user_prop = propname;
762 		entry->pl_width = strlen(entry->pl_user_prop);
763 		entry->pl_all = B_TRUE;
764 
765 		*last = entry;
766 		last = &entry->pl_next;
767 	}
768 
769 	for (entry = *plp; entry != NULL; entry = entry->pl_next) {
770 
771 		if (entry->pl_fixed)
772 			continue;
773 
774 		if (entry->pl_prop != ZPROP_INVAL &&
775 		    zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
776 		    NULL) == 0) {
777 			if (strlen(buf) > entry->pl_width)
778 				entry->pl_width = strlen(buf);
779 		}
780 	}
781 
782 	return (0);
783 }
784 
785 /*
786  * Get the state for the given feature on the given ZFS pool.
787  */
788 int
789 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
790     size_t len)
791 {
792 	uint64_t refcount;
793 	boolean_t found = B_FALSE;
794 	nvlist_t *features = zpool_get_features(zhp);
795 	boolean_t supported;
796 	const char *feature = strchr(propname, '@') + 1;
797 
798 	supported = zpool_prop_feature(propname);
799 	ASSERT(supported || zfs_prop_unsupported(propname));
800 
801 	/*
802 	 * Convert from feature name to feature guid. This conversion is
803 	 * unecessary for unsupported@... properties because they already
804 	 * use guids.
805 	 */
806 	if (supported) {
807 		int ret;
808 		zfeature_info_t *fi;
809 
810 		ret = zfeature_lookup_name(feature, &fi);
811 		if (ret != 0) {
812 			(void) strlcpy(buf, "-", len);
813 			return (ENOTSUP);
814 		}
815 		feature = fi->fi_guid;
816 	}
817 
818 	if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
819 		found = B_TRUE;
820 
821 	if (supported) {
822 		if (!found) {
823 			(void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
824 		} else  {
825 			if (refcount == 0)
826 				(void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
827 			else
828 				(void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
829 		}
830 	} else {
831 		if (found) {
832 			if (refcount == 0) {
833 				(void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
834 			} else {
835 				(void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
836 			}
837 		} else {
838 			(void) strlcpy(buf, "-", len);
839 			return (ENOTSUP);
840 		}
841 	}
842 
843 	return (0);
844 }
845 
846 /*
847  * Don't start the slice at the default block of 34; many storage
848  * devices will use a stripe width of 128k, so start there instead.
849  */
850 #define	NEW_START_BLOCK	256
851 
852 /*
853  * Validate the given pool name, optionally putting an extended error message in
854  * 'buf'.
855  */
856 boolean_t
857 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
858 {
859 	namecheck_err_t why;
860 	char what;
861 	int ret;
862 
863 	ret = pool_namecheck(pool, &why, &what);
864 
865 	/*
866 	 * The rules for reserved pool names were extended at a later point.
867 	 * But we need to support users with existing pools that may now be
868 	 * invalid.  So we only check for this expanded set of names during a
869 	 * create (or import), and only in userland.
870 	 */
871 	if (ret == 0 && !isopen &&
872 	    (strncmp(pool, "mirror", 6) == 0 ||
873 	    strncmp(pool, "raidz", 5) == 0 ||
874 	    strncmp(pool, "spare", 5) == 0 ||
875 	    strcmp(pool, "log") == 0)) {
876 		if (hdl != NULL)
877 			zfs_error_aux(hdl,
878 			    dgettext(TEXT_DOMAIN, "name is reserved"));
879 		return (B_FALSE);
880 	}
881 
882 
883 	if (ret != 0) {
884 		if (hdl != NULL) {
885 			switch (why) {
886 			case NAME_ERR_TOOLONG:
887 				zfs_error_aux(hdl,
888 				    dgettext(TEXT_DOMAIN, "name is too long"));
889 				break;
890 
891 			case NAME_ERR_INVALCHAR:
892 				zfs_error_aux(hdl,
893 				    dgettext(TEXT_DOMAIN, "invalid character "
894 				    "'%c' in pool name"), what);
895 				break;
896 
897 			case NAME_ERR_NOLETTER:
898 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
899 				    "name must begin with a letter"));
900 				break;
901 
902 			case NAME_ERR_RESERVED:
903 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
904 				    "name is reserved"));
905 				break;
906 
907 			case NAME_ERR_DISKLIKE:
908 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
909 				    "pool name is reserved"));
910 				break;
911 
912 			case NAME_ERR_LEADING_SLASH:
913 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
914 				    "leading slash in name"));
915 				break;
916 
917 			case NAME_ERR_EMPTY_COMPONENT:
918 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
919 				    "empty component in name"));
920 				break;
921 
922 			case NAME_ERR_TRAILING_SLASH:
923 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
924 				    "trailing slash in name"));
925 				break;
926 
927 			case NAME_ERR_MULTIPLE_AT:
928 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
929 				    "multiple '@' delimiters in name"));
930 				break;
931 
932 			}
933 		}
934 		return (B_FALSE);
935 	}
936 
937 	return (B_TRUE);
938 }
939 
940 /*
941  * Open a handle to the given pool, even if the pool is currently in the FAULTED
942  * state.
943  */
944 zpool_handle_t *
945 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
946 {
947 	zpool_handle_t *zhp;
948 	boolean_t missing;
949 
950 	/*
951 	 * Make sure the pool name is valid.
952 	 */
953 	if (!zpool_name_valid(hdl, B_TRUE, pool)) {
954 		(void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
955 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
956 		    pool);
957 		return (NULL);
958 	}
959 
960 	if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
961 		return (NULL);
962 
963 	zhp->zpool_hdl = hdl;
964 	(void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
965 
966 	if (zpool_refresh_stats(zhp, &missing) != 0) {
967 		zpool_close(zhp);
968 		return (NULL);
969 	}
970 
971 	if (missing) {
972 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
973 		(void) zfs_error_fmt(hdl, EZFS_NOENT,
974 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
975 		zpool_close(zhp);
976 		return (NULL);
977 	}
978 
979 	return (zhp);
980 }
981 
982 /*
983  * Like the above, but silent on error.  Used when iterating over pools (because
984  * the configuration cache may be out of date).
985  */
986 int
987 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
988 {
989 	zpool_handle_t *zhp;
990 	boolean_t missing;
991 
992 	if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
993 		return (-1);
994 
995 	zhp->zpool_hdl = hdl;
996 	(void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
997 
998 	if (zpool_refresh_stats(zhp, &missing) != 0) {
999 		zpool_close(zhp);
1000 		return (-1);
1001 	}
1002 
1003 	if (missing) {
1004 		zpool_close(zhp);
1005 		*ret = NULL;
1006 		return (0);
1007 	}
1008 
1009 	*ret = zhp;
1010 	return (0);
1011 }
1012 
1013 /*
1014  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1015  * state.
1016  */
1017 zpool_handle_t *
1018 zpool_open(libzfs_handle_t *hdl, const char *pool)
1019 {
1020 	zpool_handle_t *zhp;
1021 
1022 	if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1023 		return (NULL);
1024 
1025 	if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1026 		(void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1027 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1028 		zpool_close(zhp);
1029 		return (NULL);
1030 	}
1031 
1032 	return (zhp);
1033 }
1034 
1035 /*
1036  * Close the handle.  Simply frees the memory associated with the handle.
1037  */
1038 void
1039 zpool_close(zpool_handle_t *zhp)
1040 {
1041 	if (zhp->zpool_config)
1042 		nvlist_free(zhp->zpool_config);
1043 	if (zhp->zpool_old_config)
1044 		nvlist_free(zhp->zpool_old_config);
1045 	if (zhp->zpool_props)
1046 		nvlist_free(zhp->zpool_props);
1047 	free(zhp);
1048 }
1049 
1050 /*
1051  * Return the name of the pool.
1052  */
1053 const char *
1054 zpool_get_name(zpool_handle_t *zhp)
1055 {
1056 	return (zhp->zpool_name);
1057 }
1058 
1059 
1060 /*
1061  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1062  */
1063 int
1064 zpool_get_state(zpool_handle_t *zhp)
1065 {
1066 	return (zhp->zpool_state);
1067 }
1068 
1069 /*
1070  * Create the named pool, using the provided vdev list.  It is assumed
1071  * that the consumer has already validated the contents of the nvlist, so we
1072  * don't have to worry about error semantics.
1073  */
1074 int
1075 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1076     nvlist_t *props, nvlist_t *fsprops)
1077 {
1078 	zfs_cmd_t zc = { 0 };
1079 	nvlist_t *zc_fsprops = NULL;
1080 	nvlist_t *zc_props = NULL;
1081 	char msg[1024];
1082 	char *altroot;
1083 	int ret = -1;
1084 
1085 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1086 	    "cannot create '%s'"), pool);
1087 
1088 	if (!zpool_name_valid(hdl, B_FALSE, pool))
1089 		return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1090 
1091 	if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1092 		return (-1);
1093 
1094 	if (props) {
1095 		prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1096 
1097 		if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1098 		    SPA_VERSION_1, flags, msg)) == NULL) {
1099 			goto create_failed;
1100 		}
1101 	}
1102 
1103 	if (fsprops) {
1104 		uint64_t zoned;
1105 		char *zonestr;
1106 
1107 		zoned = ((nvlist_lookup_string(fsprops,
1108 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1109 		    strcmp(zonestr, "on") == 0);
1110 
1111 		if ((zc_fsprops = zfs_valid_proplist(hdl,
1112 		    ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
1113 			goto create_failed;
1114 		}
1115 		if (!zc_props &&
1116 		    (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1117 			goto create_failed;
1118 		}
1119 		if (nvlist_add_nvlist(zc_props,
1120 		    ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1121 			goto create_failed;
1122 		}
1123 	}
1124 
1125 	if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1126 		goto create_failed;
1127 
1128 	(void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1129 
1130 	if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1131 
1132 		zcmd_free_nvlists(&zc);
1133 		nvlist_free(zc_props);
1134 		nvlist_free(zc_fsprops);
1135 
1136 		switch (errno) {
1137 		case EBUSY:
1138 			/*
1139 			 * This can happen if the user has specified the same
1140 			 * device multiple times.  We can't reliably detect this
1141 			 * until we try to add it and see we already have a
1142 			 * label.
1143 			 */
1144 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1145 			    "one or more vdevs refer to the same device"));
1146 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1147 
1148 		case EOVERFLOW:
1149 			/*
1150 			 * This occurs when one of the devices is below
1151 			 * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1152 			 * device was the problem device since there's no
1153 			 * reliable way to determine device size from userland.
1154 			 */
1155 			{
1156 				char buf[64];
1157 
1158 				zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1159 
1160 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1161 				    "one or more devices is less than the "
1162 				    "minimum size (%s)"), buf);
1163 			}
1164 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1165 
1166 		case ENOSPC:
1167 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1168 			    "one or more devices is out of space"));
1169 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1170 
1171 		case ENOTBLK:
1172 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1173 			    "cache device must be a disk or disk slice"));
1174 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1175 
1176 		default:
1177 			return (zpool_standard_error(hdl, errno, msg));
1178 		}
1179 	}
1180 
1181 	/*
1182 	 * If this is an alternate root pool, then we automatically set the
1183 	 * mountpoint of the root dataset to be '/'.
1184 	 */
1185 	if (nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT),
1186 	    &altroot) == 0) {
1187 		zfs_handle_t *zhp;
1188 
1189 		verify((zhp = zfs_open(hdl, pool, ZFS_TYPE_DATASET)) != NULL);
1190 		verify(zfs_prop_set(zhp, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
1191 		    "/") == 0);
1192 
1193 		zfs_close(zhp);
1194 	}
1195 
1196 create_failed:
1197 	zcmd_free_nvlists(&zc);
1198 	nvlist_free(zc_props);
1199 	nvlist_free(zc_fsprops);
1200 	return (ret);
1201 }
1202 
1203 /*
1204  * Destroy the given pool.  It is up to the caller to ensure that there are no
1205  * datasets left in the pool.
1206  */
1207 int
1208 zpool_destroy(zpool_handle_t *zhp)
1209 {
1210 	zfs_cmd_t zc = { 0 };
1211 	zfs_handle_t *zfp = NULL;
1212 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1213 	char msg[1024];
1214 
1215 	if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1216 	    (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1217 		return (-1);
1218 
1219 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1220 
1221 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1222 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1223 		    "cannot destroy '%s'"), zhp->zpool_name);
1224 
1225 		if (errno == EROFS) {
1226 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1227 			    "one or more devices is read only"));
1228 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1229 		} else {
1230 			(void) zpool_standard_error(hdl, errno, msg);
1231 		}
1232 
1233 		if (zfp)
1234 			zfs_close(zfp);
1235 		return (-1);
1236 	}
1237 
1238 	if (zfp) {
1239 		remove_mountpoint(zfp);
1240 		zfs_close(zfp);
1241 	}
1242 
1243 	return (0);
1244 }
1245 
1246 /*
1247  * Add the given vdevs to the pool.  The caller must have already performed the
1248  * necessary verification to ensure that the vdev specification is well-formed.
1249  */
1250 int
1251 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1252 {
1253 	zfs_cmd_t zc = { 0 };
1254 	int ret;
1255 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1256 	char msg[1024];
1257 	nvlist_t **spares, **l2cache;
1258 	uint_t nspares, nl2cache;
1259 
1260 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1261 	    "cannot add to '%s'"), zhp->zpool_name);
1262 
1263 	if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1264 	    SPA_VERSION_SPARES &&
1265 	    nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1266 	    &spares, &nspares) == 0) {
1267 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1268 		    "upgraded to add hot spares"));
1269 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
1270 	}
1271 
1272 	if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1273 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1274 		uint64_t s;
1275 
1276 		for (s = 0; s < nspares; s++) {
1277 			char *path;
1278 
1279 			if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1280 			    &path) == 0 && pool_uses_efi(spares[s])) {
1281 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1282 				    "device '%s' contains an EFI label and "
1283 				    "cannot be used on root pools."),
1284 				    zpool_vdev_name(hdl, NULL, spares[s],
1285 				    B_FALSE));
1286 				return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1287 			}
1288 		}
1289 	}
1290 
1291 	if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1292 	    SPA_VERSION_L2CACHE &&
1293 	    nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1294 	    &l2cache, &nl2cache) == 0) {
1295 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1296 		    "upgraded to add cache devices"));
1297 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
1298 	}
1299 
1300 	if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1301 		return (-1);
1302 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1303 
1304 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1305 		switch (errno) {
1306 		case EBUSY:
1307 			/*
1308 			 * This can happen if the user has specified the same
1309 			 * device multiple times.  We can't reliably detect this
1310 			 * until we try to add it and see we already have a
1311 			 * label.
1312 			 */
1313 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1314 			    "one or more vdevs refer to the same device"));
1315 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1316 			break;
1317 
1318 		case EOVERFLOW:
1319 			/*
1320 			 * This occurrs when one of the devices is below
1321 			 * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1322 			 * device was the problem device since there's no
1323 			 * reliable way to determine device size from userland.
1324 			 */
1325 			{
1326 				char buf[64];
1327 
1328 				zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1329 
1330 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1331 				    "device is less than the minimum "
1332 				    "size (%s)"), buf);
1333 			}
1334 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1335 			break;
1336 
1337 		case ENOTSUP:
1338 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1339 			    "pool must be upgraded to add these vdevs"));
1340 			(void) zfs_error(hdl, EZFS_BADVERSION, msg);
1341 			break;
1342 
1343 		case EDOM:
1344 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1345 			    "root pool can not have multiple vdevs"
1346 			    " or separate logs"));
1347 			(void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1348 			break;
1349 
1350 		case ENOTBLK:
1351 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1352 			    "cache device must be a disk or disk slice"));
1353 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1354 			break;
1355 
1356 		default:
1357 			(void) zpool_standard_error(hdl, errno, msg);
1358 		}
1359 
1360 		ret = -1;
1361 	} else {
1362 		ret = 0;
1363 	}
1364 
1365 	zcmd_free_nvlists(&zc);
1366 
1367 	return (ret);
1368 }
1369 
1370 /*
1371  * Exports the pool from the system.  The caller must ensure that there are no
1372  * mounted datasets in the pool.
1373  */
1374 int
1375 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce)
1376 {
1377 	zfs_cmd_t zc = { 0 };
1378 	char msg[1024];
1379 
1380 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1381 	    "cannot export '%s'"), zhp->zpool_name);
1382 
1383 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1384 	zc.zc_cookie = force;
1385 	zc.zc_guid = hardforce;
1386 
1387 	if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1388 		switch (errno) {
1389 		case EXDEV:
1390 			zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1391 			    "use '-f' to override the following errors:\n"
1392 			    "'%s' has an active shared spare which could be"
1393 			    " used by other pools once '%s' is exported."),
1394 			    zhp->zpool_name, zhp->zpool_name);
1395 			return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1396 			    msg));
1397 		default:
1398 			return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1399 			    msg));
1400 		}
1401 	}
1402 
1403 	return (0);
1404 }
1405 
1406 int
1407 zpool_export(zpool_handle_t *zhp, boolean_t force)
1408 {
1409 	return (zpool_export_common(zhp, force, B_FALSE));
1410 }
1411 
1412 int
1413 zpool_export_force(zpool_handle_t *zhp)
1414 {
1415 	return (zpool_export_common(zhp, B_TRUE, B_TRUE));
1416 }
1417 
1418 static void
1419 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1420     nvlist_t *config)
1421 {
1422 	nvlist_t *nv = NULL;
1423 	uint64_t rewindto;
1424 	int64_t loss = -1;
1425 	struct tm t;
1426 	char timestr[128];
1427 
1428 	if (!hdl->libzfs_printerr || config == NULL)
1429 		return;
1430 
1431 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1432 	    nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1433 		return;
1434 	}
1435 
1436 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1437 		return;
1438 	(void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1439 
1440 	if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1441 	    strftime(timestr, 128, 0, &t) != 0) {
1442 		if (dryrun) {
1443 			(void) printf(dgettext(TEXT_DOMAIN,
1444 			    "Would be able to return %s "
1445 			    "to its state as of %s.\n"),
1446 			    name, timestr);
1447 		} else {
1448 			(void) printf(dgettext(TEXT_DOMAIN,
1449 			    "Pool %s returned to its state as of %s.\n"),
1450 			    name, timestr);
1451 		}
1452 		if (loss > 120) {
1453 			(void) printf(dgettext(TEXT_DOMAIN,
1454 			    "%s approximately %lld "),
1455 			    dryrun ? "Would discard" : "Discarded",
1456 			    (loss + 30) / 60);
1457 			(void) printf(dgettext(TEXT_DOMAIN,
1458 			    "minutes of transactions.\n"));
1459 		} else if (loss > 0) {
1460 			(void) printf(dgettext(TEXT_DOMAIN,
1461 			    "%s approximately %lld "),
1462 			    dryrun ? "Would discard" : "Discarded", loss);
1463 			(void) printf(dgettext(TEXT_DOMAIN,
1464 			    "seconds of transactions.\n"));
1465 		}
1466 	}
1467 }
1468 
1469 void
1470 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1471     nvlist_t *config)
1472 {
1473 	nvlist_t *nv = NULL;
1474 	int64_t loss = -1;
1475 	uint64_t edata = UINT64_MAX;
1476 	uint64_t rewindto;
1477 	struct tm t;
1478 	char timestr[128];
1479 
1480 	if (!hdl->libzfs_printerr)
1481 		return;
1482 
1483 	if (reason >= 0)
1484 		(void) printf(dgettext(TEXT_DOMAIN, "action: "));
1485 	else
1486 		(void) printf(dgettext(TEXT_DOMAIN, "\t"));
1487 
1488 	/* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1489 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1490 	    nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1491 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1492 		goto no_info;
1493 
1494 	(void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1495 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1496 	    &edata);
1497 
1498 	(void) printf(dgettext(TEXT_DOMAIN,
1499 	    "Recovery is possible, but will result in some data loss.\n"));
1500 
1501 	if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1502 	    strftime(timestr, 128, 0, &t) != 0) {
1503 		(void) printf(dgettext(TEXT_DOMAIN,
1504 		    "\tReturning the pool to its state as of %s\n"
1505 		    "\tshould correct the problem.  "),
1506 		    timestr);
1507 	} else {
1508 		(void) printf(dgettext(TEXT_DOMAIN,
1509 		    "\tReverting the pool to an earlier state "
1510 		    "should correct the problem.\n\t"));
1511 	}
1512 
1513 	if (loss > 120) {
1514 		(void) printf(dgettext(TEXT_DOMAIN,
1515 		    "Approximately %lld minutes of data\n"
1516 		    "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1517 	} else if (loss > 0) {
1518 		(void) printf(dgettext(TEXT_DOMAIN,
1519 		    "Approximately %lld seconds of data\n"
1520 		    "\tmust be discarded, irreversibly.  "), loss);
1521 	}
1522 	if (edata != 0 && edata != UINT64_MAX) {
1523 		if (edata == 1) {
1524 			(void) printf(dgettext(TEXT_DOMAIN,
1525 			    "After rewind, at least\n"
1526 			    "\tone persistent user-data error will remain.  "));
1527 		} else {
1528 			(void) printf(dgettext(TEXT_DOMAIN,
1529 			    "After rewind, several\n"
1530 			    "\tpersistent user-data errors will remain.  "));
1531 		}
1532 	}
1533 	(void) printf(dgettext(TEXT_DOMAIN,
1534 	    "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1535 	    reason >= 0 ? "clear" : "import", name);
1536 
1537 	(void) printf(dgettext(TEXT_DOMAIN,
1538 	    "A scrub of the pool\n"
1539 	    "\tis strongly recommended after recovery.\n"));
1540 	return;
1541 
1542 no_info:
1543 	(void) printf(dgettext(TEXT_DOMAIN,
1544 	    "Destroy and re-create the pool from\n\ta backup source.\n"));
1545 }
1546 
1547 /*
1548  * zpool_import() is a contracted interface. Should be kept the same
1549  * if possible.
1550  *
1551  * Applications should use zpool_import_props() to import a pool with
1552  * new properties value to be set.
1553  */
1554 int
1555 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1556     char *altroot)
1557 {
1558 	nvlist_t *props = NULL;
1559 	int ret;
1560 
1561 	if (altroot != NULL) {
1562 		if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1563 			return (zfs_error_fmt(hdl, EZFS_NOMEM,
1564 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1565 			    newname));
1566 		}
1567 
1568 		if (nvlist_add_string(props,
1569 		    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1570 		    nvlist_add_string(props,
1571 		    zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1572 			nvlist_free(props);
1573 			return (zfs_error_fmt(hdl, EZFS_NOMEM,
1574 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1575 			    newname));
1576 		}
1577 	}
1578 
1579 	ret = zpool_import_props(hdl, config, newname, props,
1580 	    ZFS_IMPORT_NORMAL);
1581 	if (props)
1582 		nvlist_free(props);
1583 	return (ret);
1584 }
1585 
1586 static void
1587 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1588     int indent)
1589 {
1590 	nvlist_t **child;
1591 	uint_t c, children;
1592 	char *vname;
1593 	uint64_t is_log = 0;
1594 
1595 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1596 	    &is_log);
1597 
1598 	if (name != NULL)
1599 		(void) printf("\t%*s%s%s\n", indent, "", name,
1600 		    is_log ? " [log]" : "");
1601 
1602 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1603 	    &child, &children) != 0)
1604 		return;
1605 
1606 	for (c = 0; c < children; c++) {
1607 		vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1608 		print_vdev_tree(hdl, vname, child[c], indent + 2);
1609 		free(vname);
1610 	}
1611 }
1612 
1613 void
1614 zpool_print_unsup_feat(nvlist_t *config)
1615 {
1616 	nvlist_t *nvinfo, *unsup_feat;
1617 
1618 	verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1619 	    0);
1620 	verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1621 	    &unsup_feat) == 0);
1622 
1623 	for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1624 	    nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1625 		char *desc;
1626 
1627 		verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1628 		verify(nvpair_value_string(nvp, &desc) == 0);
1629 
1630 		if (strlen(desc) > 0)
1631 			(void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1632 		else
1633 			(void) printf("\t%s\n", nvpair_name(nvp));
1634 	}
1635 }
1636 
1637 /*
1638  * Import the given pool using the known configuration and a list of
1639  * properties to be set. The configuration should have come from
1640  * zpool_find_import(). The 'newname' parameters control whether the pool
1641  * is imported with a different name.
1642  */
1643 int
1644 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1645     nvlist_t *props, int flags)
1646 {
1647 	zfs_cmd_t zc = { 0 };
1648 	zpool_rewind_policy_t policy;
1649 	nvlist_t *nv = NULL;
1650 	nvlist_t *nvinfo = NULL;
1651 	nvlist_t *missing = NULL;
1652 	char *thename;
1653 	char *origname;
1654 	int ret;
1655 	int error = 0;
1656 	char errbuf[1024];
1657 
1658 	verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1659 	    &origname) == 0);
1660 
1661 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1662 	    "cannot import pool '%s'"), origname);
1663 
1664 	if (newname != NULL) {
1665 		if (!zpool_name_valid(hdl, B_FALSE, newname))
1666 			return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1667 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1668 			    newname));
1669 		thename = (char *)newname;
1670 	} else {
1671 		thename = origname;
1672 	}
1673 
1674 	if (props) {
1675 		uint64_t version;
1676 		prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1677 
1678 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1679 		    &version) == 0);
1680 
1681 		if ((props = zpool_valid_proplist(hdl, origname,
1682 		    props, version, flags, errbuf)) == NULL) {
1683 			return (-1);
1684 		} else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1685 			nvlist_free(props);
1686 			return (-1);
1687 		}
1688 	}
1689 
1690 	(void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1691 
1692 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1693 	    &zc.zc_guid) == 0);
1694 
1695 	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1696 		nvlist_free(props);
1697 		return (-1);
1698 	}
1699 	if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1700 		nvlist_free(props);
1701 		return (-1);
1702 	}
1703 
1704 	zc.zc_cookie = flags;
1705 	while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1706 	    errno == ENOMEM) {
1707 		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1708 			zcmd_free_nvlists(&zc);
1709 			return (-1);
1710 		}
1711 	}
1712 	if (ret != 0)
1713 		error = errno;
1714 
1715 	(void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1716 	zpool_get_rewind_policy(config, &policy);
1717 
1718 	if (error) {
1719 		char desc[1024];
1720 
1721 		/*
1722 		 * Dry-run failed, but we print out what success
1723 		 * looks like if we found a best txg
1724 		 */
1725 		if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1726 			zpool_rewind_exclaim(hdl, newname ? origname : thename,
1727 			    B_TRUE, nv);
1728 			nvlist_free(nv);
1729 			return (-1);
1730 		}
1731 
1732 		if (newname == NULL)
1733 			(void) snprintf(desc, sizeof (desc),
1734 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1735 			    thename);
1736 		else
1737 			(void) snprintf(desc, sizeof (desc),
1738 			    dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1739 			    origname, thename);
1740 
1741 		switch (error) {
1742 		case ENOTSUP:
1743 			if (nv != NULL && nvlist_lookup_nvlist(nv,
1744 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1745 			    nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1746 				(void) printf(dgettext(TEXT_DOMAIN, "This "
1747 				    "pool uses the following feature(s) not "
1748 				    "supported by this system:\n"));
1749 				zpool_print_unsup_feat(nv);
1750 				if (nvlist_exists(nvinfo,
1751 				    ZPOOL_CONFIG_CAN_RDONLY)) {
1752 					(void) printf(dgettext(TEXT_DOMAIN,
1753 					    "All unsupported features are only "
1754 					    "required for writing to the pool."
1755 					    "\nThe pool can be imported using "
1756 					    "'-o readonly=on'.\n"));
1757 				}
1758 			}
1759 			/*
1760 			 * Unsupported version.
1761 			 */
1762 			(void) zfs_error(hdl, EZFS_BADVERSION, desc);
1763 			break;
1764 
1765 		case EINVAL:
1766 			(void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1767 			break;
1768 
1769 		case EROFS:
1770 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1771 			    "one or more devices is read only"));
1772 			(void) zfs_error(hdl, EZFS_BADDEV, desc);
1773 			break;
1774 
1775 		case ENXIO:
1776 			if (nv && nvlist_lookup_nvlist(nv,
1777 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1778 			    nvlist_lookup_nvlist(nvinfo,
1779 			    ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1780 				(void) printf(dgettext(TEXT_DOMAIN,
1781 				    "The devices below are missing, use "
1782 				    "'-m' to import the pool anyway:\n"));
1783 				print_vdev_tree(hdl, NULL, missing, 2);
1784 				(void) printf("\n");
1785 			}
1786 			(void) zpool_standard_error(hdl, error, desc);
1787 			break;
1788 
1789 		case EEXIST:
1790 			(void) zpool_standard_error(hdl, error, desc);
1791 			break;
1792 
1793 		default:
1794 			(void) zpool_standard_error(hdl, error, desc);
1795 			zpool_explain_recover(hdl,
1796 			    newname ? origname : thename, -error, nv);
1797 			break;
1798 		}
1799 
1800 		nvlist_free(nv);
1801 		ret = -1;
1802 	} else {
1803 		zpool_handle_t *zhp;
1804 
1805 		/*
1806 		 * This should never fail, but play it safe anyway.
1807 		 */
1808 		if (zpool_open_silent(hdl, thename, &zhp) != 0)
1809 			ret = -1;
1810 		else if (zhp != NULL)
1811 			zpool_close(zhp);
1812 		if (policy.zrp_request &
1813 		    (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1814 			zpool_rewind_exclaim(hdl, newname ? origname : thename,
1815 			    ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1816 		}
1817 		nvlist_free(nv);
1818 		return (0);
1819 	}
1820 
1821 	zcmd_free_nvlists(&zc);
1822 	nvlist_free(props);
1823 
1824 	return (ret);
1825 }
1826 
1827 /*
1828  * Scan the pool.
1829  */
1830 int
1831 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1832 {
1833 	zfs_cmd_t zc = { 0 };
1834 	char msg[1024];
1835 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1836 
1837 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1838 	zc.zc_cookie = func;
1839 
1840 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1841 	    (errno == ENOENT && func != POOL_SCAN_NONE))
1842 		return (0);
1843 
1844 	if (func == POOL_SCAN_SCRUB) {
1845 		(void) snprintf(msg, sizeof (msg),
1846 		    dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1847 	} else if (func == POOL_SCAN_NONE) {
1848 		(void) snprintf(msg, sizeof (msg),
1849 		    dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1850 		    zc.zc_name);
1851 	} else {
1852 		assert(!"unexpected result");
1853 	}
1854 
1855 	if (errno == EBUSY) {
1856 		nvlist_t *nvroot;
1857 		pool_scan_stat_t *ps = NULL;
1858 		uint_t psc;
1859 
1860 		verify(nvlist_lookup_nvlist(zhp->zpool_config,
1861 		    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1862 		(void) nvlist_lookup_uint64_array(nvroot,
1863 		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1864 		if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1865 			return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1866 		else
1867 			return (zfs_error(hdl, EZFS_RESILVERING, msg));
1868 	} else if (errno == ENOENT) {
1869 		return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1870 	} else {
1871 		return (zpool_standard_error(hdl, errno, msg));
1872 	}
1873 }
1874 
1875 /*
1876  * This provides a very minimal check whether a given string is likely a
1877  * c#t#d# style string.  Users of this are expected to do their own
1878  * verification of the s# part.
1879  */
1880 #define	CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1881 
1882 /*
1883  * More elaborate version for ones which may start with "/dev/dsk/"
1884  * and the like.
1885  */
1886 static int
1887 ctd_check_path(char *str) {
1888 	/*
1889 	 * If it starts with a slash, check the last component.
1890 	 */
1891 	if (str && str[0] == '/') {
1892 		char *tmp = strrchr(str, '/');
1893 
1894 		/*
1895 		 * If it ends in "/old", check the second-to-last
1896 		 * component of the string instead.
1897 		 */
1898 		if (tmp != str && strcmp(tmp, "/old") == 0) {
1899 			for (tmp--; *tmp != '/'; tmp--)
1900 				;
1901 		}
1902 		str = tmp + 1;
1903 	}
1904 	return (CTD_CHECK(str));
1905 }
1906 
1907 /*
1908  * Find a vdev that matches the search criteria specified. We use the
1909  * the nvpair name to determine how we should look for the device.
1910  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1911  * spare; but FALSE if its an INUSE spare.
1912  */
1913 static nvlist_t *
1914 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1915     boolean_t *l2cache, boolean_t *log)
1916 {
1917 	uint_t c, children;
1918 	nvlist_t **child;
1919 	nvlist_t *ret;
1920 	uint64_t is_log;
1921 	char *srchkey;
1922 	nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1923 
1924 	/* Nothing to look for */
1925 	if (search == NULL || pair == NULL)
1926 		return (NULL);
1927 
1928 	/* Obtain the key we will use to search */
1929 	srchkey = nvpair_name(pair);
1930 
1931 	switch (nvpair_type(pair)) {
1932 	case DATA_TYPE_UINT64:
1933 		if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1934 			uint64_t srchval, theguid;
1935 
1936 			verify(nvpair_value_uint64(pair, &srchval) == 0);
1937 			verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1938 			    &theguid) == 0);
1939 			if (theguid == srchval)
1940 				return (nv);
1941 		}
1942 		break;
1943 
1944 	case DATA_TYPE_STRING: {
1945 		char *srchval, *val;
1946 
1947 		verify(nvpair_value_string(pair, &srchval) == 0);
1948 		if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1949 			break;
1950 
1951 		/*
1952 		 * Search for the requested value. Special cases:
1953 		 *
1954 		 * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1955 		 *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1956 		 *   but included in the string, so this matches around it.
1957 		 * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1958 		 *
1959 		 * Otherwise, all other searches are simple string compares.
1960 		 */
1961 		if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1962 		    ctd_check_path(val)) {
1963 			uint64_t wholedisk = 0;
1964 
1965 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1966 			    &wholedisk);
1967 			if (wholedisk) {
1968 				int slen = strlen(srchval);
1969 				int vlen = strlen(val);
1970 
1971 				if (slen != vlen - 2)
1972 					break;
1973 
1974 				/*
1975 				 * make_leaf_vdev() should only set
1976 				 * wholedisk for ZPOOL_CONFIG_PATHs which
1977 				 * will include "/dev/dsk/", giving plenty of
1978 				 * room for the indices used next.
1979 				 */
1980 				ASSERT(vlen >= 6);
1981 
1982 				/*
1983 				 * strings identical except trailing "s0"
1984 				 */
1985 				if (strcmp(&val[vlen - 2], "s0") == 0 &&
1986 				    strncmp(srchval, val, slen) == 0)
1987 					return (nv);
1988 
1989 				/*
1990 				 * strings identical except trailing "s0/old"
1991 				 */
1992 				if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
1993 				    strcmp(&srchval[slen - 4], "/old") == 0 &&
1994 				    strncmp(srchval, val, slen - 4) == 0)
1995 					return (nv);
1996 
1997 				break;
1998 			}
1999 		} else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2000 			char *type, *idx, *end, *p;
2001 			uint64_t id, vdev_id;
2002 
2003 			/*
2004 			 * Determine our vdev type, keeping in mind
2005 			 * that the srchval is composed of a type and
2006 			 * vdev id pair (i.e. mirror-4).
2007 			 */
2008 			if ((type = strdup(srchval)) == NULL)
2009 				return (NULL);
2010 
2011 			if ((p = strrchr(type, '-')) == NULL) {
2012 				free(type);
2013 				break;
2014 			}
2015 			idx = p + 1;
2016 			*p = '\0';
2017 
2018 			/*
2019 			 * If the types don't match then keep looking.
2020 			 */
2021 			if (strncmp(val, type, strlen(val)) != 0) {
2022 				free(type);
2023 				break;
2024 			}
2025 
2026 			verify(strncmp(type, VDEV_TYPE_RAIDZ,
2027 			    strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2028 			    strncmp(type, VDEV_TYPE_MIRROR,
2029 			    strlen(VDEV_TYPE_MIRROR)) == 0);
2030 			verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2031 			    &id) == 0);
2032 
2033 			errno = 0;
2034 			vdev_id = strtoull(idx, &end, 10);
2035 
2036 			free(type);
2037 			if (errno != 0)
2038 				return (NULL);
2039 
2040 			/*
2041 			 * Now verify that we have the correct vdev id.
2042 			 */
2043 			if (vdev_id == id)
2044 				return (nv);
2045 		}
2046 
2047 		/*
2048 		 * Common case
2049 		 */
2050 		if (strcmp(srchval, val) == 0)
2051 			return (nv);
2052 		break;
2053 	}
2054 
2055 	default:
2056 		break;
2057 	}
2058 
2059 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2060 	    &child, &children) != 0)
2061 		return (NULL);
2062 
2063 	for (c = 0; c < children; c++) {
2064 		if ((ret = vdev_to_nvlist_iter(child[c], search,
2065 		    avail_spare, l2cache, NULL)) != NULL) {
2066 			/*
2067 			 * The 'is_log' value is only set for the toplevel
2068 			 * vdev, not the leaf vdevs.  So we always lookup the
2069 			 * log device from the root of the vdev tree (where
2070 			 * 'log' is non-NULL).
2071 			 */
2072 			if (log != NULL &&
2073 			    nvlist_lookup_uint64(child[c],
2074 			    ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2075 			    is_log) {
2076 				*log = B_TRUE;
2077 			}
2078 			return (ret);
2079 		}
2080 	}
2081 
2082 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2083 	    &child, &children) == 0) {
2084 		for (c = 0; c < children; c++) {
2085 			if ((ret = vdev_to_nvlist_iter(child[c], search,
2086 			    avail_spare, l2cache, NULL)) != NULL) {
2087 				*avail_spare = B_TRUE;
2088 				return (ret);
2089 			}
2090 		}
2091 	}
2092 
2093 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2094 	    &child, &children) == 0) {
2095 		for (c = 0; c < children; c++) {
2096 			if ((ret = vdev_to_nvlist_iter(child[c], search,
2097 			    avail_spare, l2cache, NULL)) != NULL) {
2098 				*l2cache = B_TRUE;
2099 				return (ret);
2100 			}
2101 		}
2102 	}
2103 
2104 	return (NULL);
2105 }
2106 
2107 /*
2108  * Given a physical path (minus the "/devices" prefix), find the
2109  * associated vdev.
2110  */
2111 nvlist_t *
2112 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2113     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2114 {
2115 	nvlist_t *search, *nvroot, *ret;
2116 
2117 	verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2118 	verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2119 
2120 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2121 	    &nvroot) == 0);
2122 
2123 	*avail_spare = B_FALSE;
2124 	*l2cache = B_FALSE;
2125 	if (log != NULL)
2126 		*log = B_FALSE;
2127 	ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2128 	nvlist_free(search);
2129 
2130 	return (ret);
2131 }
2132 
2133 /*
2134  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2135  */
2136 boolean_t
2137 zpool_vdev_is_interior(const char *name)
2138 {
2139 	if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2140 	    strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2141 		return (B_TRUE);
2142 	return (B_FALSE);
2143 }
2144 
2145 nvlist_t *
2146 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2147     boolean_t *l2cache, boolean_t *log)
2148 {
2149 	char buf[MAXPATHLEN];
2150 	char *end;
2151 	nvlist_t *nvroot, *search, *ret;
2152 	uint64_t guid;
2153 
2154 	verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2155 
2156 	guid = strtoull(path, &end, 10);
2157 	if (guid != 0 && *end == '\0') {
2158 		verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2159 	} else if (zpool_vdev_is_interior(path)) {
2160 		verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2161 	} else if (path[0] != '/') {
2162 		(void) snprintf(buf, sizeof (buf), "%s%s", "/dev/dsk/", path);
2163 		verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2164 	} else {
2165 		verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2166 	}
2167 
2168 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2169 	    &nvroot) == 0);
2170 
2171 	*avail_spare = B_FALSE;
2172 	*l2cache = B_FALSE;
2173 	if (log != NULL)
2174 		*log = B_FALSE;
2175 	ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2176 	nvlist_free(search);
2177 
2178 	return (ret);
2179 }
2180 
2181 static int
2182 vdev_online(nvlist_t *nv)
2183 {
2184 	uint64_t ival;
2185 
2186 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2187 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2188 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2189 		return (0);
2190 
2191 	return (1);
2192 }
2193 
2194 /*
2195  * Helper function for zpool_get_physpaths().
2196  */
2197 static int
2198 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2199     size_t *bytes_written)
2200 {
2201 	size_t bytes_left, pos, rsz;
2202 	char *tmppath;
2203 	const char *format;
2204 
2205 	if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2206 	    &tmppath) != 0)
2207 		return (EZFS_NODEVICE);
2208 
2209 	pos = *bytes_written;
2210 	bytes_left = physpath_size - pos;
2211 	format = (pos == 0) ? "%s" : " %s";
2212 
2213 	rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2214 	*bytes_written += rsz;
2215 
2216 	if (rsz >= bytes_left) {
2217 		/* if physpath was not copied properly, clear it */
2218 		if (bytes_left != 0) {
2219 			physpath[pos] = 0;
2220 		}
2221 		return (EZFS_NOSPC);
2222 	}
2223 	return (0);
2224 }
2225 
2226 static int
2227 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2228     size_t *rsz, boolean_t is_spare)
2229 {
2230 	char *type;
2231 	int ret;
2232 
2233 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2234 		return (EZFS_INVALCONFIG);
2235 
2236 	if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2237 		/*
2238 		 * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2239 		 * For a spare vdev, we only want to boot from the active
2240 		 * spare device.
2241 		 */
2242 		if (is_spare) {
2243 			uint64_t spare = 0;
2244 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2245 			    &spare);
2246 			if (!spare)
2247 				return (EZFS_INVALCONFIG);
2248 		}
2249 
2250 		if (vdev_online(nv)) {
2251 			if ((ret = vdev_get_one_physpath(nv, physpath,
2252 			    phypath_size, rsz)) != 0)
2253 				return (ret);
2254 		}
2255 	} else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2256 	    strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2257 	    (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2258 		nvlist_t **child;
2259 		uint_t count;
2260 		int i, ret;
2261 
2262 		if (nvlist_lookup_nvlist_array(nv,
2263 		    ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2264 			return (EZFS_INVALCONFIG);
2265 
2266 		for (i = 0; i < count; i++) {
2267 			ret = vdev_get_physpaths(child[i], physpath,
2268 			    phypath_size, rsz, is_spare);
2269 			if (ret == EZFS_NOSPC)
2270 				return (ret);
2271 		}
2272 	}
2273 
2274 	return (EZFS_POOL_INVALARG);
2275 }
2276 
2277 /*
2278  * Get phys_path for a root pool config.
2279  * Return 0 on success; non-zero on failure.
2280  */
2281 static int
2282 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2283 {
2284 	size_t rsz;
2285 	nvlist_t *vdev_root;
2286 	nvlist_t **child;
2287 	uint_t count;
2288 	char *type;
2289 
2290 	rsz = 0;
2291 
2292 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2293 	    &vdev_root) != 0)
2294 		return (EZFS_INVALCONFIG);
2295 
2296 	if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2297 	    nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2298 	    &child, &count) != 0)
2299 		return (EZFS_INVALCONFIG);
2300 
2301 	/*
2302 	 * root pool can not have EFI labeled disks and can only have
2303 	 * a single top-level vdev.
2304 	 */
2305 	if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2306 	    pool_uses_efi(vdev_root))
2307 		return (EZFS_POOL_INVALARG);
2308 
2309 	(void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2310 	    B_FALSE);
2311 
2312 	/* No online devices */
2313 	if (rsz == 0)
2314 		return (EZFS_NODEVICE);
2315 
2316 	return (0);
2317 }
2318 
2319 /*
2320  * Get phys_path for a root pool
2321  * Return 0 on success; non-zero on failure.
2322  */
2323 int
2324 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2325 {
2326 	return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2327 	    phypath_size));
2328 }
2329 
2330 /*
2331  * If the device has being dynamically expanded then we need to relabel
2332  * the disk to use the new unallocated space.
2333  */
2334 static int
2335 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2336 {
2337 	char path[MAXPATHLEN];
2338 	char errbuf[1024];
2339 	int fd, error;
2340 	int (*_efi_use_whole_disk)(int);
2341 
2342 	if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2343 	    "efi_use_whole_disk")) == NULL)
2344 		return (-1);
2345 
2346 	(void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2347 
2348 	if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2349 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2350 		    "relabel '%s': unable to open device"), name);
2351 		return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2352 	}
2353 
2354 	/*
2355 	 * It's possible that we might encounter an error if the device
2356 	 * does not have any unallocated space left. If so, we simply
2357 	 * ignore that error and continue on.
2358 	 */
2359 	error = _efi_use_whole_disk(fd);
2360 	(void) close(fd);
2361 	if (error && error != VT_ENOSPC) {
2362 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2363 		    "relabel '%s': unable to read disk capacity"), name);
2364 		return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2365 	}
2366 	return (0);
2367 }
2368 
2369 /*
2370  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2371  * ZFS_ONLINE_* flags.
2372  */
2373 int
2374 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2375     vdev_state_t *newstate)
2376 {
2377 	zfs_cmd_t zc = { 0 };
2378 	char msg[1024];
2379 	nvlist_t *tgt;
2380 	boolean_t avail_spare, l2cache, islog;
2381 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2382 
2383 	if (flags & ZFS_ONLINE_EXPAND) {
2384 		(void) snprintf(msg, sizeof (msg),
2385 		    dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2386 	} else {
2387 		(void) snprintf(msg, sizeof (msg),
2388 		    dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2389 	}
2390 
2391 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2392 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2393 	    &islog)) == NULL)
2394 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
2395 
2396 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2397 
2398 	if (avail_spare)
2399 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
2400 
2401 	if (flags & ZFS_ONLINE_EXPAND ||
2402 	    zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2403 		char *pathname = NULL;
2404 		uint64_t wholedisk = 0;
2405 
2406 		(void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2407 		    &wholedisk);
2408 		verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2409 		    &pathname) == 0);
2410 
2411 		/*
2412 		 * XXX - L2ARC 1.0 devices can't support expansion.
2413 		 */
2414 		if (l2cache) {
2415 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2416 			    "cannot expand cache devices"));
2417 			return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2418 		}
2419 
2420 		if (wholedisk) {
2421 			pathname += strlen(DISK_ROOT) + 1;
2422 			(void) zpool_relabel_disk(hdl, pathname);
2423 		}
2424 	}
2425 
2426 	zc.zc_cookie = VDEV_STATE_ONLINE;
2427 	zc.zc_obj = flags;
2428 
2429 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2430 		if (errno == EINVAL) {
2431 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2432 			    "from this pool into a new one.  Use '%s' "
2433 			    "instead"), "zpool detach");
2434 			return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2435 		}
2436 		return (zpool_standard_error(hdl, errno, msg));
2437 	}
2438 
2439 	*newstate = zc.zc_cookie;
2440 	return (0);
2441 }
2442 
2443 /*
2444  * Take the specified vdev offline
2445  */
2446 int
2447 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2448 {
2449 	zfs_cmd_t zc = { 0 };
2450 	char msg[1024];
2451 	nvlist_t *tgt;
2452 	boolean_t avail_spare, l2cache;
2453 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2454 
2455 	(void) snprintf(msg, sizeof (msg),
2456 	    dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2457 
2458 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2459 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2460 	    NULL)) == NULL)
2461 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
2462 
2463 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2464 
2465 	if (avail_spare)
2466 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
2467 
2468 	zc.zc_cookie = VDEV_STATE_OFFLINE;
2469 	zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2470 
2471 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2472 		return (0);
2473 
2474 	switch (errno) {
2475 	case EBUSY:
2476 
2477 		/*
2478 		 * There are no other replicas of this device.
2479 		 */
2480 		return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2481 
2482 	case EEXIST:
2483 		/*
2484 		 * The log device has unplayed logs
2485 		 */
2486 		return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2487 
2488 	default:
2489 		return (zpool_standard_error(hdl, errno, msg));
2490 	}
2491 }
2492 
2493 /*
2494  * Mark the given vdev faulted.
2495  */
2496 int
2497 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2498 {
2499 	zfs_cmd_t zc = { 0 };
2500 	char msg[1024];
2501 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2502 
2503 	(void) snprintf(msg, sizeof (msg),
2504 	    dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2505 
2506 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2507 	zc.zc_guid = guid;
2508 	zc.zc_cookie = VDEV_STATE_FAULTED;
2509 	zc.zc_obj = aux;
2510 
2511 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2512 		return (0);
2513 
2514 	switch (errno) {
2515 	case EBUSY:
2516 
2517 		/*
2518 		 * There are no other replicas of this device.
2519 		 */
2520 		return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2521 
2522 	default:
2523 		return (zpool_standard_error(hdl, errno, msg));
2524 	}
2525 
2526 }
2527 
2528 /*
2529  * Mark the given vdev degraded.
2530  */
2531 int
2532 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2533 {
2534 	zfs_cmd_t zc = { 0 };
2535 	char msg[1024];
2536 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2537 
2538 	(void) snprintf(msg, sizeof (msg),
2539 	    dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2540 
2541 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2542 	zc.zc_guid = guid;
2543 	zc.zc_cookie = VDEV_STATE_DEGRADED;
2544 	zc.zc_obj = aux;
2545 
2546 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2547 		return (0);
2548 
2549 	return (zpool_standard_error(hdl, errno, msg));
2550 }
2551 
2552 /*
2553  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2554  * a hot spare.
2555  */
2556 static boolean_t
2557 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2558 {
2559 	nvlist_t **child;
2560 	uint_t c, children;
2561 	char *type;
2562 
2563 	if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2564 	    &children) == 0) {
2565 		verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2566 		    &type) == 0);
2567 
2568 		if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2569 		    children == 2 && child[which] == tgt)
2570 			return (B_TRUE);
2571 
2572 		for (c = 0; c < children; c++)
2573 			if (is_replacing_spare(child[c], tgt, which))
2574 				return (B_TRUE);
2575 	}
2576 
2577 	return (B_FALSE);
2578 }
2579 
2580 /*
2581  * Attach new_disk (fully described by nvroot) to old_disk.
2582  * If 'replacing' is specified, the new disk will replace the old one.
2583  */
2584 int
2585 zpool_vdev_attach(zpool_handle_t *zhp,
2586     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2587 {
2588 	zfs_cmd_t zc = { 0 };
2589 	char msg[1024];
2590 	int ret;
2591 	nvlist_t *tgt;
2592 	boolean_t avail_spare, l2cache, islog;
2593 	uint64_t val;
2594 	char *newname;
2595 	nvlist_t **child;
2596 	uint_t children;
2597 	nvlist_t *config_root;
2598 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2599 	boolean_t rootpool = zpool_is_bootable(zhp);
2600 
2601 	if (replacing)
2602 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2603 		    "cannot replace %s with %s"), old_disk, new_disk);
2604 	else
2605 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2606 		    "cannot attach %s to %s"), new_disk, old_disk);
2607 
2608 	/*
2609 	 * If this is a root pool, make sure that we're not attaching an
2610 	 * EFI labeled device.
2611 	 */
2612 	if (rootpool && pool_uses_efi(nvroot)) {
2613 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2614 		    "EFI labeled devices are not supported on root pools."));
2615 		return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2616 	}
2617 
2618 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2619 	if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2620 	    &islog)) == 0)
2621 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
2622 
2623 	if (avail_spare)
2624 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
2625 
2626 	if (l2cache)
2627 		return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2628 
2629 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2630 	zc.zc_cookie = replacing;
2631 
2632 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2633 	    &child, &children) != 0 || children != 1) {
2634 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2635 		    "new device must be a single disk"));
2636 		return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2637 	}
2638 
2639 	verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2640 	    ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2641 
2642 	if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2643 		return (-1);
2644 
2645 	/*
2646 	 * If the target is a hot spare that has been swapped in, we can only
2647 	 * replace it with another hot spare.
2648 	 */
2649 	if (replacing &&
2650 	    nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2651 	    (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2652 	    NULL) == NULL || !avail_spare) &&
2653 	    is_replacing_spare(config_root, tgt, 1)) {
2654 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2655 		    "can only be replaced by another hot spare"));
2656 		free(newname);
2657 		return (zfs_error(hdl, EZFS_BADTARGET, msg));
2658 	}
2659 
2660 	free(newname);
2661 
2662 	if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2663 		return (-1);
2664 
2665 	ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2666 
2667 	zcmd_free_nvlists(&zc);
2668 
2669 	if (ret == 0) {
2670 		if (rootpool) {
2671 			/*
2672 			 * XXX need a better way to prevent user from
2673 			 * booting up a half-baked vdev.
2674 			 */
2675 			(void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2676 			    "sure to wait until resilver is done "
2677 			    "before rebooting.\n"));
2678 		}
2679 		return (0);
2680 	}
2681 
2682 	switch (errno) {
2683 	case ENOTSUP:
2684 		/*
2685 		 * Can't attach to or replace this type of vdev.
2686 		 */
2687 		if (replacing) {
2688 			uint64_t version = zpool_get_prop_int(zhp,
2689 			    ZPOOL_PROP_VERSION, NULL);
2690 
2691 			if (islog)
2692 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2693 				    "cannot replace a log with a spare"));
2694 			else if (version >= SPA_VERSION_MULTI_REPLACE)
2695 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2696 				    "already in replacing/spare config; wait "
2697 				    "for completion or use 'zpool detach'"));
2698 			else
2699 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2700 				    "cannot replace a replacing device"));
2701 		} else {
2702 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2703 			    "can only attach to mirrors and top-level "
2704 			    "disks"));
2705 		}
2706 		(void) zfs_error(hdl, EZFS_BADTARGET, msg);
2707 		break;
2708 
2709 	case EINVAL:
2710 		/*
2711 		 * The new device must be a single disk.
2712 		 */
2713 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2714 		    "new device must be a single disk"));
2715 		(void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2716 		break;
2717 
2718 	case EBUSY:
2719 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2720 		    new_disk);
2721 		(void) zfs_error(hdl, EZFS_BADDEV, msg);
2722 		break;
2723 
2724 	case EOVERFLOW:
2725 		/*
2726 		 * The new device is too small.
2727 		 */
2728 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2729 		    "device is too small"));
2730 		(void) zfs_error(hdl, EZFS_BADDEV, msg);
2731 		break;
2732 
2733 	case EDOM:
2734 		/*
2735 		 * The new device has a different alignment requirement.
2736 		 */
2737 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2738 		    "devices have different sector alignment"));
2739 		(void) zfs_error(hdl, EZFS_BADDEV, msg);
2740 		break;
2741 
2742 	case ENAMETOOLONG:
2743 		/*
2744 		 * The resulting top-level vdev spec won't fit in the label.
2745 		 */
2746 		(void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2747 		break;
2748 
2749 	default:
2750 		(void) zpool_standard_error(hdl, errno, msg);
2751 	}
2752 
2753 	return (-1);
2754 }
2755 
2756 /*
2757  * Detach the specified device.
2758  */
2759 int
2760 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2761 {
2762 	zfs_cmd_t zc = { 0 };
2763 	char msg[1024];
2764 	nvlist_t *tgt;
2765 	boolean_t avail_spare, l2cache;
2766 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2767 
2768 	(void) snprintf(msg, sizeof (msg),
2769 	    dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2770 
2771 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2772 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2773 	    NULL)) == 0)
2774 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
2775 
2776 	if (avail_spare)
2777 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
2778 
2779 	if (l2cache)
2780 		return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2781 
2782 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2783 
2784 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2785 		return (0);
2786 
2787 	switch (errno) {
2788 
2789 	case ENOTSUP:
2790 		/*
2791 		 * Can't detach from this type of vdev.
2792 		 */
2793 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2794 		    "applicable to mirror and replacing vdevs"));
2795 		(void) zfs_error(hdl, EZFS_BADTARGET, msg);
2796 		break;
2797 
2798 	case EBUSY:
2799 		/*
2800 		 * There are no other replicas of this device.
2801 		 */
2802 		(void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2803 		break;
2804 
2805 	default:
2806 		(void) zpool_standard_error(hdl, errno, msg);
2807 	}
2808 
2809 	return (-1);
2810 }
2811 
2812 /*
2813  * Find a mirror vdev in the source nvlist.
2814  *
2815  * The mchild array contains a list of disks in one of the top-level mirrors
2816  * of the source pool.  The schild array contains a list of disks that the
2817  * user specified on the command line.  We loop over the mchild array to
2818  * see if any entry in the schild array matches.
2819  *
2820  * If a disk in the mchild array is found in the schild array, we return
2821  * the index of that entry.  Otherwise we return -1.
2822  */
2823 static int
2824 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2825     nvlist_t **schild, uint_t schildren)
2826 {
2827 	uint_t mc;
2828 
2829 	for (mc = 0; mc < mchildren; mc++) {
2830 		uint_t sc;
2831 		char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2832 		    mchild[mc], B_FALSE);
2833 
2834 		for (sc = 0; sc < schildren; sc++) {
2835 			char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2836 			    schild[sc], B_FALSE);
2837 			boolean_t result = (strcmp(mpath, spath) == 0);
2838 
2839 			free(spath);
2840 			if (result) {
2841 				free(mpath);
2842 				return (mc);
2843 			}
2844 		}
2845 
2846 		free(mpath);
2847 	}
2848 
2849 	return (-1);
2850 }
2851 
2852 /*
2853  * Split a mirror pool.  If newroot points to null, then a new nvlist
2854  * is generated and it is the responsibility of the caller to free it.
2855  */
2856 int
2857 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2858     nvlist_t *props, splitflags_t flags)
2859 {
2860 	zfs_cmd_t zc = { 0 };
2861 	char msg[1024];
2862 	nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2863 	nvlist_t **varray = NULL, *zc_props = NULL;
2864 	uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2865 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2866 	uint64_t vers;
2867 	boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2868 	int retval = 0;
2869 
2870 	(void) snprintf(msg, sizeof (msg),
2871 	    dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2872 
2873 	if (!zpool_name_valid(hdl, B_FALSE, newname))
2874 		return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2875 
2876 	if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2877 		(void) fprintf(stderr, gettext("Internal error: unable to "
2878 		    "retrieve pool configuration\n"));
2879 		return (-1);
2880 	}
2881 
2882 	verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2883 	    == 0);
2884 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2885 
2886 	if (props) {
2887 		prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2888 		if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2889 		    props, vers, flags, msg)) == NULL)
2890 			return (-1);
2891 	}
2892 
2893 	if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2894 	    &children) != 0) {
2895 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2896 		    "Source pool is missing vdev tree"));
2897 		if (zc_props)
2898 			nvlist_free(zc_props);
2899 		return (-1);
2900 	}
2901 
2902 	varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2903 	vcount = 0;
2904 
2905 	if (*newroot == NULL ||
2906 	    nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2907 	    &newchild, &newchildren) != 0)
2908 		newchildren = 0;
2909 
2910 	for (c = 0; c < children; c++) {
2911 		uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2912 		char *type;
2913 		nvlist_t **mchild, *vdev;
2914 		uint_t mchildren;
2915 		int entry;
2916 
2917 		/*
2918 		 * Unlike cache & spares, slogs are stored in the
2919 		 * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2920 		 */
2921 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2922 		    &is_log);
2923 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2924 		    &is_hole);
2925 		if (is_log || is_hole) {
2926 			/*
2927 			 * Create a hole vdev and put it in the config.
2928 			 */
2929 			if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2930 				goto out;
2931 			if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2932 			    VDEV_TYPE_HOLE) != 0)
2933 				goto out;
2934 			if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2935 			    1) != 0)
2936 				goto out;
2937 			if (lastlog == 0)
2938 				lastlog = vcount;
2939 			varray[vcount++] = vdev;
2940 			continue;
2941 		}
2942 		lastlog = 0;
2943 		verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2944 		    == 0);
2945 		if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2946 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2947 			    "Source pool must be composed only of mirrors\n"));
2948 			retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2949 			goto out;
2950 		}
2951 
2952 		verify(nvlist_lookup_nvlist_array(child[c],
2953 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2954 
2955 		/* find or add an entry for this top-level vdev */
2956 		if (newchildren > 0 &&
2957 		    (entry = find_vdev_entry(zhp, mchild, mchildren,
2958 		    newchild, newchildren)) >= 0) {
2959 			/* We found a disk that the user specified. */
2960 			vdev = mchild[entry];
2961 			++found;
2962 		} else {
2963 			/* User didn't specify a disk for this vdev. */
2964 			vdev = mchild[mchildren - 1];
2965 		}
2966 
2967 		if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2968 			goto out;
2969 	}
2970 
2971 	/* did we find every disk the user specified? */
2972 	if (found != newchildren) {
2973 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
2974 		    "include at most one disk from each mirror"));
2975 		retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2976 		goto out;
2977 	}
2978 
2979 	/* Prepare the nvlist for populating. */
2980 	if (*newroot == NULL) {
2981 		if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
2982 			goto out;
2983 		freelist = B_TRUE;
2984 		if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
2985 		    VDEV_TYPE_ROOT) != 0)
2986 			goto out;
2987 	} else {
2988 		verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
2989 	}
2990 
2991 	/* Add all the children we found */
2992 	if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
2993 	    lastlog == 0 ? vcount : lastlog) != 0)
2994 		goto out;
2995 
2996 	/*
2997 	 * If we're just doing a dry run, exit now with success.
2998 	 */
2999 	if (flags.dryrun) {
3000 		memory_err = B_FALSE;
3001 		freelist = B_FALSE;
3002 		goto out;
3003 	}
3004 
3005 	/* now build up the config list & call the ioctl */
3006 	if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3007 		goto out;
3008 
3009 	if (nvlist_add_nvlist(newconfig,
3010 	    ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3011 	    nvlist_add_string(newconfig,
3012 	    ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3013 	    nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3014 		goto out;
3015 
3016 	/*
3017 	 * The new pool is automatically part of the namespace unless we
3018 	 * explicitly export it.
3019 	 */
3020 	if (!flags.import)
3021 		zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3022 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3023 	(void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3024 	if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3025 		goto out;
3026 	if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3027 		goto out;
3028 
3029 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3030 		retval = zpool_standard_error(hdl, errno, msg);
3031 		goto out;
3032 	}
3033 
3034 	freelist = B_FALSE;
3035 	memory_err = B_FALSE;
3036 
3037 out:
3038 	if (varray != NULL) {
3039 		int v;
3040 
3041 		for (v = 0; v < vcount; v++)
3042 			nvlist_free(varray[v]);
3043 		free(varray);
3044 	}
3045 	zcmd_free_nvlists(&zc);
3046 	if (zc_props)
3047 		nvlist_free(zc_props);
3048 	if (newconfig)
3049 		nvlist_free(newconfig);
3050 	if (freelist) {
3051 		nvlist_free(*newroot);
3052 		*newroot = NULL;
3053 	}
3054 
3055 	if (retval != 0)
3056 		return (retval);
3057 
3058 	if (memory_err)
3059 		return (no_memory(hdl));
3060 
3061 	return (0);
3062 }
3063 
3064 /*
3065  * Remove the given device.  Currently, this is supported only for hot spares
3066  * and level 2 cache devices.
3067  */
3068 int
3069 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3070 {
3071 	zfs_cmd_t zc = { 0 };
3072 	char msg[1024];
3073 	nvlist_t *tgt;
3074 	boolean_t avail_spare, l2cache, islog;
3075 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3076 	uint64_t version;
3077 
3078 	(void) snprintf(msg, sizeof (msg),
3079 	    dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3080 
3081 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3082 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3083 	    &islog)) == 0)
3084 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3085 	/*
3086 	 * XXX - this should just go away.
3087 	 */
3088 	if (!avail_spare && !l2cache && !islog) {
3089 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3090 		    "only inactive hot spares, cache, top-level, "
3091 		    "or log devices can be removed"));
3092 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3093 	}
3094 
3095 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3096 	if (islog && version < SPA_VERSION_HOLES) {
3097 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3098 		    "pool must be upgrade to support log removal"));
3099 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
3100 	}
3101 
3102 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3103 
3104 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3105 		return (0);
3106 
3107 	return (zpool_standard_error(hdl, errno, msg));
3108 }
3109 
3110 /*
3111  * Clear the errors for the pool, or the particular device if specified.
3112  */
3113 int
3114 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3115 {
3116 	zfs_cmd_t zc = { 0 };
3117 	char msg[1024];
3118 	nvlist_t *tgt;
3119 	zpool_rewind_policy_t policy;
3120 	boolean_t avail_spare, l2cache;
3121 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3122 	nvlist_t *nvi = NULL;
3123 	int error;
3124 
3125 	if (path)
3126 		(void) snprintf(msg, sizeof (msg),
3127 		    dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3128 		    path);
3129 	else
3130 		(void) snprintf(msg, sizeof (msg),
3131 		    dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3132 		    zhp->zpool_name);
3133 
3134 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3135 	if (path) {
3136 		if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3137 		    &l2cache, NULL)) == 0)
3138 			return (zfs_error(hdl, EZFS_NODEVICE, msg));
3139 
3140 		/*
3141 		 * Don't allow error clearing for hot spares.  Do allow
3142 		 * error clearing for l2cache devices.
3143 		 */
3144 		if (avail_spare)
3145 			return (zfs_error(hdl, EZFS_ISSPARE, msg));
3146 
3147 		verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3148 		    &zc.zc_guid) == 0);
3149 	}
3150 
3151 	zpool_get_rewind_policy(rewindnvl, &policy);
3152 	zc.zc_cookie = policy.zrp_request;
3153 
3154 	if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3155 		return (-1);
3156 
3157 	if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3158 		return (-1);
3159 
3160 	while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3161 	    errno == ENOMEM) {
3162 		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3163 			zcmd_free_nvlists(&zc);
3164 			return (-1);
3165 		}
3166 	}
3167 
3168 	if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3169 	    errno != EPERM && errno != EACCES)) {
3170 		if (policy.zrp_request &
3171 		    (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3172 			(void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3173 			zpool_rewind_exclaim(hdl, zc.zc_name,
3174 			    ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3175 			    nvi);
3176 			nvlist_free(nvi);
3177 		}
3178 		zcmd_free_nvlists(&zc);
3179 		return (0);
3180 	}
3181 
3182 	zcmd_free_nvlists(&zc);
3183 	return (zpool_standard_error(hdl, errno, msg));
3184 }
3185 
3186 /*
3187  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3188  */
3189 int
3190 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3191 {
3192 	zfs_cmd_t zc = { 0 };
3193 	char msg[1024];
3194 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3195 
3196 	(void) snprintf(msg, sizeof (msg),
3197 	    dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3198 	    guid);
3199 
3200 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3201 	zc.zc_guid = guid;
3202 	zc.zc_cookie = ZPOOL_NO_REWIND;
3203 
3204 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3205 		return (0);
3206 
3207 	return (zpool_standard_error(hdl, errno, msg));
3208 }
3209 
3210 /*
3211  * Change the GUID for a pool.
3212  */
3213 int
3214 zpool_reguid(zpool_handle_t *zhp)
3215 {
3216 	char msg[1024];
3217 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3218 	zfs_cmd_t zc = { 0 };
3219 
3220 	(void) snprintf(msg, sizeof (msg),
3221 	    dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3222 
3223 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3224 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3225 		return (0);
3226 
3227 	return (zpool_standard_error(hdl, errno, msg));
3228 }
3229 
3230 /*
3231  * Reopen the pool.
3232  */
3233 int
3234 zpool_reopen(zpool_handle_t *zhp)
3235 {
3236 	zfs_cmd_t zc = { 0 };
3237 	char msg[1024];
3238 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3239 
3240 	(void) snprintf(msg, sizeof (msg),
3241 	    dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3242 	    zhp->zpool_name);
3243 
3244 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3245 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3246 		return (0);
3247 	return (zpool_standard_error(hdl, errno, msg));
3248 }
3249 
3250 /*
3251  * Convert from a devid string to a path.
3252  */
3253 static char *
3254 devid_to_path(char *devid_str)
3255 {
3256 	ddi_devid_t devid;
3257 	char *minor;
3258 	char *path;
3259 	devid_nmlist_t *list = NULL;
3260 	int ret;
3261 
3262 	if (devid_str_decode(devid_str, &devid, &minor) != 0)
3263 		return (NULL);
3264 
3265 	ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3266 
3267 	devid_str_free(minor);
3268 	devid_free(devid);
3269 
3270 	if (ret != 0)
3271 		return (NULL);
3272 
3273 	if ((path = strdup(list[0].devname)) == NULL)
3274 		return (NULL);
3275 
3276 	devid_free_nmlist(list);
3277 
3278 	return (path);
3279 }
3280 
3281 /*
3282  * Convert from a path to a devid string.
3283  */
3284 static char *
3285 path_to_devid(const char *path)
3286 {
3287 	int fd;
3288 	ddi_devid_t devid;
3289 	char *minor, *ret;
3290 
3291 	if ((fd = open(path, O_RDONLY)) < 0)
3292 		return (NULL);
3293 
3294 	minor = NULL;
3295 	ret = NULL;
3296 	if (devid_get(fd, &devid) == 0) {
3297 		if (devid_get_minor_name(fd, &minor) == 0)
3298 			ret = devid_str_encode(devid, minor);
3299 		if (minor != NULL)
3300 			devid_str_free(minor);
3301 		devid_free(devid);
3302 	}
3303 	(void) close(fd);
3304 
3305 	return (ret);
3306 }
3307 
3308 /*
3309  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3310  * ignore any failure here, since a common case is for an unprivileged user to
3311  * type 'zpool status', and we'll display the correct information anyway.
3312  */
3313 static void
3314 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3315 {
3316 	zfs_cmd_t zc = { 0 };
3317 
3318 	(void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3319 	(void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3320 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3321 	    &zc.zc_guid) == 0);
3322 
3323 	(void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3324 }
3325 
3326 /*
3327  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3328  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3329  * We also check if this is a whole disk, in which case we strip off the
3330  * trailing 's0' slice name.
3331  *
3332  * This routine is also responsible for identifying when disks have been
3333  * reconfigured in a new location.  The kernel will have opened the device by
3334  * devid, but the path will still refer to the old location.  To catch this, we
3335  * first do a path -> devid translation (which is fast for the common case).  If
3336  * the devid matches, we're done.  If not, we do a reverse devid -> path
3337  * translation and issue the appropriate ioctl() to update the path of the vdev.
3338  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3339  * of these checks.
3340  */
3341 char *
3342 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3343     boolean_t verbose)
3344 {
3345 	char *path, *devid;
3346 	uint64_t value;
3347 	char buf[64];
3348 	vdev_stat_t *vs;
3349 	uint_t vsc;
3350 
3351 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
3352 	    &value) == 0) {
3353 		verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3354 		    &value) == 0);
3355 		(void) snprintf(buf, sizeof (buf), "%llu",
3356 		    (u_longlong_t)value);
3357 		path = buf;
3358 	} else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
3359 
3360 		/*
3361 		 * If the device is dead (faulted, offline, etc) then don't
3362 		 * bother opening it.  Otherwise we may be forcing the user to
3363 		 * open a misbehaving device, which can have undesirable
3364 		 * effects.
3365 		 */
3366 		if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3367 		    (uint64_t **)&vs, &vsc) != 0 ||
3368 		    vs->vs_state >= VDEV_STATE_DEGRADED) &&
3369 		    zhp != NULL &&
3370 		    nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3371 			/*
3372 			 * Determine if the current path is correct.
3373 			 */
3374 			char *newdevid = path_to_devid(path);
3375 
3376 			if (newdevid == NULL ||
3377 			    strcmp(devid, newdevid) != 0) {
3378 				char *newpath;
3379 
3380 				if ((newpath = devid_to_path(devid)) != NULL) {
3381 					/*
3382 					 * Update the path appropriately.
3383 					 */
3384 					set_path(zhp, nv, newpath);
3385 					if (nvlist_add_string(nv,
3386 					    ZPOOL_CONFIG_PATH, newpath) == 0)
3387 						verify(nvlist_lookup_string(nv,
3388 						    ZPOOL_CONFIG_PATH,
3389 						    &path) == 0);
3390 					free(newpath);
3391 				}
3392 			}
3393 
3394 			if (newdevid)
3395 				devid_str_free(newdevid);
3396 		}
3397 
3398 		if (strncmp(path, "/dev/dsk/", 9) == 0)
3399 			path += 9;
3400 
3401 		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3402 		    &value) == 0 && value) {
3403 			int pathlen = strlen(path);
3404 			char *tmp = zfs_strdup(hdl, path);
3405 
3406 			/*
3407 			 * If it starts with c#, and ends with "s0", chop
3408 			 * the "s0" off, or if it ends with "s0/old", remove
3409 			 * the "s0" from the middle.
3410 			 */
3411 			if (CTD_CHECK(tmp)) {
3412 				if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3413 					tmp[pathlen - 2] = '\0';
3414 				} else if (pathlen > 6 &&
3415 				    strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3416 					(void) strcpy(&tmp[pathlen - 6],
3417 					    "/old");
3418 				}
3419 			}
3420 			return (tmp);
3421 		}
3422 	} else {
3423 		verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3424 
3425 		/*
3426 		 * If it's a raidz device, we need to stick in the parity level.
3427 		 */
3428 		if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3429 			verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3430 			    &value) == 0);
3431 			(void) snprintf(buf, sizeof (buf), "%s%llu", path,
3432 			    (u_longlong_t)value);
3433 			path = buf;
3434 		}
3435 
3436 		/*
3437 		 * We identify each top-level vdev by using a <type-id>
3438 		 * naming convention.
3439 		 */
3440 		if (verbose) {
3441 			uint64_t id;
3442 
3443 			verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3444 			    &id) == 0);
3445 			(void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3446 			    (u_longlong_t)id);
3447 			path = buf;
3448 		}
3449 	}
3450 
3451 	return (zfs_strdup(hdl, path));
3452 }
3453 
3454 static int
3455 zbookmark_compare(const void *a, const void *b)
3456 {
3457 	return (memcmp(a, b, sizeof (zbookmark_t)));
3458 }
3459 
3460 /*
3461  * Retrieve the persistent error log, uniquify the members, and return to the
3462  * caller.
3463  */
3464 int
3465 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3466 {
3467 	zfs_cmd_t zc = { 0 };
3468 	uint64_t count;
3469 	zbookmark_t *zb = NULL;
3470 	int i;
3471 
3472 	/*
3473 	 * Retrieve the raw error list from the kernel.  If the number of errors
3474 	 * has increased, allocate more space and continue until we get the
3475 	 * entire list.
3476 	 */
3477 	verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3478 	    &count) == 0);
3479 	if (count == 0)
3480 		return (0);
3481 	if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3482 	    count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
3483 		return (-1);
3484 	zc.zc_nvlist_dst_size = count;
3485 	(void) strcpy(zc.zc_name, zhp->zpool_name);
3486 	for (;;) {
3487 		if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3488 		    &zc) != 0) {
3489 			free((void *)(uintptr_t)zc.zc_nvlist_dst);
3490 			if (errno == ENOMEM) {
3491 				count = zc.zc_nvlist_dst_size;
3492 				if ((zc.zc_nvlist_dst = (uintptr_t)
3493 				    zfs_alloc(zhp->zpool_hdl, count *
3494 				    sizeof (zbookmark_t))) == (uintptr_t)NULL)
3495 					return (-1);
3496 			} else {
3497 				return (-1);
3498 			}
3499 		} else {
3500 			break;
3501 		}
3502 	}
3503 
3504 	/*
3505 	 * Sort the resulting bookmarks.  This is a little confusing due to the
3506 	 * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3507 	 * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3508 	 * _not_ copied as part of the process.  So we point the start of our
3509 	 * array appropriate and decrement the total number of elements.
3510 	 */
3511 	zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
3512 	    zc.zc_nvlist_dst_size;
3513 	count -= zc.zc_nvlist_dst_size;
3514 
3515 	qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
3516 
3517 	verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3518 
3519 	/*
3520 	 * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3521 	 */
3522 	for (i = 0; i < count; i++) {
3523 		nvlist_t *nv;
3524 
3525 		/* ignoring zb_blkid and zb_level for now */
3526 		if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3527 		    zb[i-1].zb_object == zb[i].zb_object)
3528 			continue;
3529 
3530 		if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3531 			goto nomem;
3532 		if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3533 		    zb[i].zb_objset) != 0) {
3534 			nvlist_free(nv);
3535 			goto nomem;
3536 		}
3537 		if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3538 		    zb[i].zb_object) != 0) {
3539 			nvlist_free(nv);
3540 			goto nomem;
3541 		}
3542 		if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3543 			nvlist_free(nv);
3544 			goto nomem;
3545 		}
3546 		nvlist_free(nv);
3547 	}
3548 
3549 	free((void *)(uintptr_t)zc.zc_nvlist_dst);
3550 	return (0);
3551 
3552 nomem:
3553 	free((void *)(uintptr_t)zc.zc_nvlist_dst);
3554 	return (no_memory(zhp->zpool_hdl));
3555 }
3556 
3557 /*
3558  * Upgrade a ZFS pool to the latest on-disk version.
3559  */
3560 int
3561 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3562 {
3563 	zfs_cmd_t zc = { 0 };
3564 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3565 
3566 	(void) strcpy(zc.zc_name, zhp->zpool_name);
3567 	zc.zc_cookie = new_version;
3568 
3569 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3570 		return (zpool_standard_error_fmt(hdl, errno,
3571 		    dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3572 		    zhp->zpool_name));
3573 	return (0);
3574 }
3575 
3576 void
3577 zpool_set_history_str(const char *subcommand, int argc, char **argv,
3578     char *history_str)
3579 {
3580 	int i;
3581 
3582 	(void) strlcpy(history_str, subcommand, HIS_MAX_RECORD_LEN);
3583 	for (i = 1; i < argc; i++) {
3584 		if (strlen(history_str) + 1 + strlen(argv[i]) >
3585 		    HIS_MAX_RECORD_LEN)
3586 			break;
3587 		(void) strlcat(history_str, " ", HIS_MAX_RECORD_LEN);
3588 		(void) strlcat(history_str, argv[i], HIS_MAX_RECORD_LEN);
3589 	}
3590 }
3591 
3592 /*
3593  * Stage command history for logging.
3594  */
3595 int
3596 zpool_stage_history(libzfs_handle_t *hdl, const char *history_str)
3597 {
3598 	if (history_str == NULL)
3599 		return (EINVAL);
3600 
3601 	if (strlen(history_str) > HIS_MAX_RECORD_LEN)
3602 		return (EINVAL);
3603 
3604 	if (hdl->libzfs_log_str != NULL)
3605 		free(hdl->libzfs_log_str);
3606 
3607 	if ((hdl->libzfs_log_str = strdup(history_str)) == NULL)
3608 		return (no_memory(hdl));
3609 
3610 	return (0);
3611 }
3612 
3613 /*
3614  * Perform ioctl to get some command history of a pool.
3615  *
3616  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3617  * logical offset of the history buffer to start reading from.
3618  *
3619  * Upon return, 'off' is the next logical offset to read from and
3620  * 'len' is the actual amount of bytes read into 'buf'.
3621  */
3622 static int
3623 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3624 {
3625 	zfs_cmd_t zc = { 0 };
3626 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3627 
3628 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3629 
3630 	zc.zc_history = (uint64_t)(uintptr_t)buf;
3631 	zc.zc_history_len = *len;
3632 	zc.zc_history_offset = *off;
3633 
3634 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3635 		switch (errno) {
3636 		case EPERM:
3637 			return (zfs_error_fmt(hdl, EZFS_PERM,
3638 			    dgettext(TEXT_DOMAIN,
3639 			    "cannot show history for pool '%s'"),
3640 			    zhp->zpool_name));
3641 		case ENOENT:
3642 			return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3643 			    dgettext(TEXT_DOMAIN, "cannot get history for pool "
3644 			    "'%s'"), zhp->zpool_name));
3645 		case ENOTSUP:
3646 			return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3647 			    dgettext(TEXT_DOMAIN, "cannot get history for pool "
3648 			    "'%s', pool must be upgraded"), zhp->zpool_name));
3649 		default:
3650 			return (zpool_standard_error_fmt(hdl, errno,
3651 			    dgettext(TEXT_DOMAIN,
3652 			    "cannot get history for '%s'"), zhp->zpool_name));
3653 		}
3654 	}
3655 
3656 	*len = zc.zc_history_len;
3657 	*off = zc.zc_history_offset;
3658 
3659 	return (0);
3660 }
3661 
3662 /*
3663  * Process the buffer of nvlists, unpacking and storing each nvlist record
3664  * into 'records'.  'leftover' is set to the number of bytes that weren't
3665  * processed as there wasn't a complete record.
3666  */
3667 int
3668 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3669     nvlist_t ***records, uint_t *numrecords)
3670 {
3671 	uint64_t reclen;
3672 	nvlist_t *nv;
3673 	int i;
3674 
3675 	while (bytes_read > sizeof (reclen)) {
3676 
3677 		/* get length of packed record (stored as little endian) */
3678 		for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3679 			reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3680 
3681 		if (bytes_read < sizeof (reclen) + reclen)
3682 			break;
3683 
3684 		/* unpack record */
3685 		if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3686 			return (ENOMEM);
3687 		bytes_read -= sizeof (reclen) + reclen;
3688 		buf += sizeof (reclen) + reclen;
3689 
3690 		/* add record to nvlist array */
3691 		(*numrecords)++;
3692 		if (ISP2(*numrecords + 1)) {
3693 			*records = realloc(*records,
3694 			    *numrecords * 2 * sizeof (nvlist_t *));
3695 		}
3696 		(*records)[*numrecords - 1] = nv;
3697 	}
3698 
3699 	*leftover = bytes_read;
3700 	return (0);
3701 }
3702 
3703 #define	HIS_BUF_LEN	(128*1024)
3704 
3705 /*
3706  * Retrieve the command history of a pool.
3707  */
3708 int
3709 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3710 {
3711 	char buf[HIS_BUF_LEN];
3712 	uint64_t off = 0;
3713 	nvlist_t **records = NULL;
3714 	uint_t numrecords = 0;
3715 	int err, i;
3716 
3717 	do {
3718 		uint64_t bytes_read = sizeof (buf);
3719 		uint64_t leftover;
3720 
3721 		if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3722 			break;
3723 
3724 		/* if nothing else was read in, we're at EOF, just return */
3725 		if (!bytes_read)
3726 			break;
3727 
3728 		if ((err = zpool_history_unpack(buf, bytes_read,
3729 		    &leftover, &records, &numrecords)) != 0)
3730 			break;
3731 		off -= leftover;
3732 
3733 		/* CONSTCOND */
3734 	} while (1);
3735 
3736 	if (!err) {
3737 		verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3738 		verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3739 		    records, numrecords) == 0);
3740 	}
3741 	for (i = 0; i < numrecords; i++)
3742 		nvlist_free(records[i]);
3743 	free(records);
3744 
3745 	return (err);
3746 }
3747 
3748 void
3749 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3750     char *pathname, size_t len)
3751 {
3752 	zfs_cmd_t zc = { 0 };
3753 	boolean_t mounted = B_FALSE;
3754 	char *mntpnt = NULL;
3755 	char dsname[MAXNAMELEN];
3756 
3757 	if (dsobj == 0) {
3758 		/* special case for the MOS */
3759 		(void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3760 		return;
3761 	}
3762 
3763 	/* get the dataset's name */
3764 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3765 	zc.zc_obj = dsobj;
3766 	if (ioctl(zhp->zpool_hdl->libzfs_fd,
3767 	    ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3768 		/* just write out a path of two object numbers */
3769 		(void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3770 		    dsobj, obj);
3771 		return;
3772 	}
3773 	(void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3774 
3775 	/* find out if the dataset is mounted */
3776 	mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3777 
3778 	/* get the corrupted object's path */
3779 	(void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3780 	zc.zc_obj = obj;
3781 	if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3782 	    &zc) == 0) {
3783 		if (mounted) {
3784 			(void) snprintf(pathname, len, "%s%s", mntpnt,
3785 			    zc.zc_value);
3786 		} else {
3787 			(void) snprintf(pathname, len, "%s:%s",
3788 			    dsname, zc.zc_value);
3789 		}
3790 	} else {
3791 		(void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3792 	}
3793 	free(mntpnt);
3794 }
3795 
3796 /*
3797  * Read the EFI label from the config, if a label does not exist then
3798  * pass back the error to the caller. If the caller has passed a non-NULL
3799  * diskaddr argument then we set it to the starting address of the EFI
3800  * partition.
3801  */
3802 static int
3803 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3804 {
3805 	char *path;
3806 	int fd;
3807 	char diskname[MAXPATHLEN];
3808 	int err = -1;
3809 
3810 	if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3811 		return (err);
3812 
3813 	(void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3814 	    strrchr(path, '/'));
3815 	if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3816 		struct dk_gpt *vtoc;
3817 
3818 		if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3819 			if (sb != NULL)
3820 				*sb = vtoc->efi_parts[0].p_start;
3821 			efi_free(vtoc);
3822 		}
3823 		(void) close(fd);
3824 	}
3825 	return (err);
3826 }
3827 
3828 /*
3829  * determine where a partition starts on a disk in the current
3830  * configuration
3831  */
3832 static diskaddr_t
3833 find_start_block(nvlist_t *config)
3834 {
3835 	nvlist_t **child;
3836 	uint_t c, children;
3837 	diskaddr_t sb = MAXOFFSET_T;
3838 	uint64_t wholedisk;
3839 
3840 	if (nvlist_lookup_nvlist_array(config,
3841 	    ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3842 		if (nvlist_lookup_uint64(config,
3843 		    ZPOOL_CONFIG_WHOLE_DISK,
3844 		    &wholedisk) != 0 || !wholedisk) {
3845 			return (MAXOFFSET_T);
3846 		}
3847 		if (read_efi_label(config, &sb) < 0)
3848 			sb = MAXOFFSET_T;
3849 		return (sb);
3850 	}
3851 
3852 	for (c = 0; c < children; c++) {
3853 		sb = find_start_block(child[c]);
3854 		if (sb != MAXOFFSET_T) {
3855 			return (sb);
3856 		}
3857 	}
3858 	return (MAXOFFSET_T);
3859 }
3860 
3861 /*
3862  * Label an individual disk.  The name provided is the short name,
3863  * stripped of any leading /dev path.
3864  */
3865 int
3866 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
3867 {
3868 	char path[MAXPATHLEN];
3869 	struct dk_gpt *vtoc;
3870 	int fd;
3871 	size_t resv = EFI_MIN_RESV_SIZE;
3872 	uint64_t slice_size;
3873 	diskaddr_t start_block;
3874 	char errbuf[1024];
3875 
3876 	/* prepare an error message just in case */
3877 	(void) snprintf(errbuf, sizeof (errbuf),
3878 	    dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3879 
3880 	if (zhp) {
3881 		nvlist_t *nvroot;
3882 
3883 		if (zpool_is_bootable(zhp)) {
3884 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3885 			    "EFI labeled devices are not supported on root "
3886 			    "pools."));
3887 			return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3888 		}
3889 
3890 		verify(nvlist_lookup_nvlist(zhp->zpool_config,
3891 		    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3892 
3893 		if (zhp->zpool_start_block == 0)
3894 			start_block = find_start_block(nvroot);
3895 		else
3896 			start_block = zhp->zpool_start_block;
3897 		zhp->zpool_start_block = start_block;
3898 	} else {
3899 		/* new pool */
3900 		start_block = NEW_START_BLOCK;
3901 	}
3902 
3903 	(void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3904 	    BACKUP_SLICE);
3905 
3906 	if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3907 		/*
3908 		 * This shouldn't happen.  We've long since verified that this
3909 		 * is a valid device.
3910 		 */
3911 		zfs_error_aux(hdl,
3912 		    dgettext(TEXT_DOMAIN, "unable to open device"));
3913 		return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3914 	}
3915 
3916 	if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3917 		/*
3918 		 * The only way this can fail is if we run out of memory, or we
3919 		 * were unable to read the disk's capacity
3920 		 */
3921 		if (errno == ENOMEM)
3922 			(void) no_memory(hdl);
3923 
3924 		(void) close(fd);
3925 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3926 		    "unable to read disk capacity"), name);
3927 
3928 		return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3929 	}
3930 
3931 	slice_size = vtoc->efi_last_u_lba + 1;
3932 	slice_size -= EFI_MIN_RESV_SIZE;
3933 	if (start_block == MAXOFFSET_T)
3934 		start_block = NEW_START_BLOCK;
3935 	slice_size -= start_block;
3936 
3937 	vtoc->efi_parts[0].p_start = start_block;
3938 	vtoc->efi_parts[0].p_size = slice_size;
3939 
3940 	/*
3941 	 * Why we use V_USR: V_BACKUP confuses users, and is considered
3942 	 * disposable by some EFI utilities (since EFI doesn't have a backup
3943 	 * slice).  V_UNASSIGNED is supposed to be used only for zero size
3944 	 * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3945 	 * etc. were all pretty specific.  V_USR is as close to reality as we
3946 	 * can get, in the absence of V_OTHER.
3947 	 */
3948 	vtoc->efi_parts[0].p_tag = V_USR;
3949 	(void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3950 
3951 	vtoc->efi_parts[8].p_start = slice_size + start_block;
3952 	vtoc->efi_parts[8].p_size = resv;
3953 	vtoc->efi_parts[8].p_tag = V_RESERVED;
3954 
3955 	if (efi_write(fd, vtoc) != 0) {
3956 		/*
3957 		 * Some block drivers (like pcata) may not support EFI
3958 		 * GPT labels.  Print out a helpful error message dir-
3959 		 * ecting the user to manually label the disk and give
3960 		 * a specific slice.
3961 		 */
3962 		(void) close(fd);
3963 		efi_free(vtoc);
3964 
3965 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3966 		    "try using fdisk(1M) and then provide a specific slice"));
3967 		return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3968 	}
3969 
3970 	(void) close(fd);
3971 	efi_free(vtoc);
3972 	return (0);
3973 }
3974 
3975 static boolean_t
3976 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
3977 {
3978 	char *type;
3979 	nvlist_t **child;
3980 	uint_t children, c;
3981 
3982 	verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
3983 	if (strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
3984 	    strcmp(type, VDEV_TYPE_FILE) == 0 ||
3985 	    strcmp(type, VDEV_TYPE_LOG) == 0 ||
3986 	    strcmp(type, VDEV_TYPE_HOLE) == 0 ||
3987 	    strcmp(type, VDEV_TYPE_MISSING) == 0) {
3988 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3989 		    "vdev type '%s' is not supported"), type);
3990 		(void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
3991 		return (B_FALSE);
3992 	}
3993 	if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
3994 	    &child, &children) == 0) {
3995 		for (c = 0; c < children; c++) {
3996 			if (!supported_dump_vdev_type(hdl, child[c], errbuf))
3997 				return (B_FALSE);
3998 		}
3999 	}
4000 	return (B_TRUE);
4001 }
4002 
4003 /*
4004  * check if this zvol is allowable for use as a dump device; zero if
4005  * it is, > 0 if it isn't, < 0 if it isn't a zvol
4006  */
4007 int
4008 zvol_check_dump_config(char *arg)
4009 {
4010 	zpool_handle_t *zhp = NULL;
4011 	nvlist_t *config, *nvroot;
4012 	char *p, *volname;
4013 	nvlist_t **top;
4014 	uint_t toplevels;
4015 	libzfs_handle_t *hdl;
4016 	char errbuf[1024];
4017 	char poolname[ZPOOL_MAXNAMELEN];
4018 	int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4019 	int ret = 1;
4020 
4021 	if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4022 		return (-1);
4023 	}
4024 
4025 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4026 	    "dump is not supported on device '%s'"), arg);
4027 
4028 	if ((hdl = libzfs_init()) == NULL)
4029 		return (1);
4030 	libzfs_print_on_error(hdl, B_TRUE);
4031 
4032 	volname = arg + pathlen;
4033 
4034 	/* check the configuration of the pool */
4035 	if ((p = strchr(volname, '/')) == NULL) {
4036 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4037 		    "malformed dataset name"));
4038 		(void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4039 		return (1);
4040 	} else if (p - volname >= ZFS_MAXNAMELEN) {
4041 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4042 		    "dataset name is too long"));
4043 		(void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4044 		return (1);
4045 	} else {
4046 		(void) strncpy(poolname, volname, p - volname);
4047 		poolname[p - volname] = '\0';
4048 	}
4049 
4050 	if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4051 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4052 		    "could not open pool '%s'"), poolname);
4053 		(void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4054 		goto out;
4055 	}
4056 	config = zpool_get_config(zhp, NULL);
4057 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4058 	    &nvroot) != 0) {
4059 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4060 		    "could not obtain vdev configuration for  '%s'"), poolname);
4061 		(void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4062 		goto out;
4063 	}
4064 
4065 	verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4066 	    &top, &toplevels) == 0);
4067 	if (toplevels != 1) {
4068 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4069 		    "'%s' has multiple top level vdevs"), poolname);
4070 		(void) zfs_error(hdl, EZFS_DEVOVERFLOW, errbuf);
4071 		goto out;
4072 	}
4073 
4074 	if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4075 		goto out;
4076 	}
4077 	ret = 0;
4078 
4079 out:
4080 	if (zhp)
4081 		zpool_close(zhp);
4082 	libzfs_fini(hdl);
4083 	return (ret);
4084 }
4085