xref: /titanic_41/usr/src/lib/libzfs/common/libzfs_pool.c (revision 1b7435644ebe3112993d73003e61f1bf7125440b)
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, 2014 by Delphix. All rights reserved.
26  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
27  */
28 
29 #include <ctype.h>
30 #include <errno.h>
31 #include <devid.h>
32 #include <fcntl.h>
33 #include <libintl.h>
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <strings.h>
37 #include <unistd.h>
38 #include <libgen.h>
39 #include <sys/efi_partition.h>
40 #include <sys/vtoc.h>
41 #include <sys/zfs_ioctl.h>
42 #include <dlfcn.h>
43 
44 #include "zfs_namecheck.h"
45 #include "zfs_prop.h"
46 #include "libzfs_impl.h"
47 #include "zfs_comutil.h"
48 #include "zfeature_common.h"
49 
50 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
51 
52 #define	DISK_ROOT	"/dev/dsk"
53 #define	RDISK_ROOT	"/dev/rdsk"
54 #define	BACKUP_SLICE	"s2"
55 
56 typedef struct prop_flags {
57 	int create:1;	/* Validate property on creation */
58 	int import:1;	/* Validate property on import */
59 } prop_flags_t;
60 
61 /*
62  * ====================================================================
63  *   zpool property functions
64  * ====================================================================
65  */
66 
67 static int
68 zpool_get_all_props(zpool_handle_t *zhp)
69 {
70 	zfs_cmd_t zc = { 0 };
71 	libzfs_handle_t *hdl = zhp->zpool_hdl;
72 
73 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
74 
75 	if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
76 		return (-1);
77 
78 	while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
79 		if (errno == ENOMEM) {
80 			if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
81 				zcmd_free_nvlists(&zc);
82 				return (-1);
83 			}
84 		} else {
85 			zcmd_free_nvlists(&zc);
86 			return (-1);
87 		}
88 	}
89 
90 	if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
91 		zcmd_free_nvlists(&zc);
92 		return (-1);
93 	}
94 
95 	zcmd_free_nvlists(&zc);
96 
97 	return (0);
98 }
99 
100 static int
101 zpool_props_refresh(zpool_handle_t *zhp)
102 {
103 	nvlist_t *old_props;
104 
105 	old_props = zhp->zpool_props;
106 
107 	if (zpool_get_all_props(zhp) != 0)
108 		return (-1);
109 
110 	nvlist_free(old_props);
111 	return (0);
112 }
113 
114 static char *
115 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
116     zprop_source_t *src)
117 {
118 	nvlist_t *nv, *nvl;
119 	uint64_t ival;
120 	char *value;
121 	zprop_source_t source;
122 
123 	nvl = zhp->zpool_props;
124 	if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
125 		verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
126 		source = ival;
127 		verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
128 	} else {
129 		source = ZPROP_SRC_DEFAULT;
130 		if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
131 			value = "-";
132 	}
133 
134 	if (src)
135 		*src = source;
136 
137 	return (value);
138 }
139 
140 uint64_t
141 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
142 {
143 	nvlist_t *nv, *nvl;
144 	uint64_t value;
145 	zprop_source_t source;
146 
147 	if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
148 		/*
149 		 * zpool_get_all_props() has most likely failed because
150 		 * the pool is faulted, but if all we need is the top level
151 		 * vdev's guid then get it from the zhp config nvlist.
152 		 */
153 		if ((prop == ZPOOL_PROP_GUID) &&
154 		    (nvlist_lookup_nvlist(zhp->zpool_config,
155 		    ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
156 		    (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
157 		    == 0)) {
158 			return (value);
159 		}
160 		return (zpool_prop_default_numeric(prop));
161 	}
162 
163 	nvl = zhp->zpool_props;
164 	if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
165 		verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
166 		source = value;
167 		verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
168 	} else {
169 		source = ZPROP_SRC_DEFAULT;
170 		value = zpool_prop_default_numeric(prop);
171 	}
172 
173 	if (src)
174 		*src = source;
175 
176 	return (value);
177 }
178 
179 /*
180  * Map VDEV STATE to printed strings.
181  */
182 char *
183 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
184 {
185 	switch (state) {
186 	case VDEV_STATE_CLOSED:
187 	case VDEV_STATE_OFFLINE:
188 		return (gettext("OFFLINE"));
189 	case VDEV_STATE_REMOVED:
190 		return (gettext("REMOVED"));
191 	case VDEV_STATE_CANT_OPEN:
192 		if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
193 			return (gettext("FAULTED"));
194 		else if (aux == VDEV_AUX_SPLIT_POOL)
195 			return (gettext("SPLIT"));
196 		else
197 			return (gettext("UNAVAIL"));
198 	case VDEV_STATE_FAULTED:
199 		return (gettext("FAULTED"));
200 	case VDEV_STATE_DEGRADED:
201 		return (gettext("DEGRADED"));
202 	case VDEV_STATE_HEALTHY:
203 		return (gettext("ONLINE"));
204 	}
205 
206 	return (gettext("UNKNOWN"));
207 }
208 
209 /*
210  * Get a zpool property value for 'prop' and return the value in
211  * a pre-allocated buffer.
212  */
213 int
214 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
215     zprop_source_t *srctype, boolean_t literal)
216 {
217 	uint64_t intval;
218 	const char *strval;
219 	zprop_source_t src = ZPROP_SRC_NONE;
220 	nvlist_t *nvroot;
221 	vdev_stat_t *vs;
222 	uint_t vsc;
223 
224 	if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
225 		switch (prop) {
226 		case ZPOOL_PROP_NAME:
227 			(void) strlcpy(buf, zpool_get_name(zhp), len);
228 			break;
229 
230 		case ZPOOL_PROP_HEALTH:
231 			(void) strlcpy(buf, "FAULTED", len);
232 			break;
233 
234 		case ZPOOL_PROP_GUID:
235 			intval = zpool_get_prop_int(zhp, prop, &src);
236 			(void) snprintf(buf, len, "%llu", intval);
237 			break;
238 
239 		case ZPOOL_PROP_ALTROOT:
240 		case ZPOOL_PROP_CACHEFILE:
241 		case ZPOOL_PROP_COMMENT:
242 			if (zhp->zpool_props != NULL ||
243 			    zpool_get_all_props(zhp) == 0) {
244 				(void) strlcpy(buf,
245 				    zpool_get_prop_string(zhp, prop, &src),
246 				    len);
247 				break;
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_LEAKED:
279 		case ZPOOL_PROP_EXPANDSZ:
280 			if (literal) {
281 				(void) snprintf(buf, len, "%llu",
282 				    (u_longlong_t)intval);
283 			} else {
284 				(void) zfs_nicenum(intval, buf, len);
285 			}
286 			break;
287 
288 		case ZPOOL_PROP_CAPACITY:
289 			if (literal) {
290 				(void) snprintf(buf, len, "%llu",
291 				    (u_longlong_t)intval);
292 			} else {
293 				(void) snprintf(buf, len, "%llu%%",
294 				    (u_longlong_t)intval);
295 			}
296 			break;
297 		case ZPOOL_PROP_FRAGMENTATION:
298 			if (intval == UINT64_MAX) {
299 				(void) strlcpy(buf, "-", len);
300 			} else {
301 				(void) snprintf(buf, len, "%llu%%",
302 				    (u_longlong_t)intval);
303 			}
304 			break;
305 
306 		case ZPOOL_PROP_DEDUPRATIO:
307 			(void) snprintf(buf, len, "%llu.%02llux",
308 			    (u_longlong_t)(intval / 100),
309 			    (u_longlong_t)(intval % 100));
310 			break;
311 
312 		case ZPOOL_PROP_HEALTH:
313 			verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
314 			    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
315 			verify(nvlist_lookup_uint64_array(nvroot,
316 			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
317 			    == 0);
318 
319 			(void) strlcpy(buf, zpool_state_to_name(intval,
320 			    vs->vs_aux), len);
321 			break;
322 		case ZPOOL_PROP_VERSION:
323 			if (intval >= SPA_VERSION_FEATURES) {
324 				(void) snprintf(buf, len, "-");
325 				break;
326 			}
327 			/* FALLTHROUGH */
328 		default:
329 			(void) snprintf(buf, len, "%llu", intval);
330 		}
331 		break;
332 
333 	case PROP_TYPE_INDEX:
334 		intval = zpool_get_prop_int(zhp, prop, &src);
335 		if (zpool_prop_index_to_string(prop, intval, &strval)
336 		    != 0)
337 			return (-1);
338 		(void) strlcpy(buf, strval, len);
339 		break;
340 
341 	default:
342 		abort();
343 	}
344 
345 	if (srctype)
346 		*srctype = src;
347 
348 	return (0);
349 }
350 
351 /*
352  * Check if the bootfs name has the same pool name as it is set to.
353  * Assuming bootfs is a valid dataset name.
354  */
355 static boolean_t
356 bootfs_name_valid(const char *pool, char *bootfs)
357 {
358 	int len = strlen(pool);
359 
360 	if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
361 		return (B_FALSE);
362 
363 	if (strncmp(pool, bootfs, len) == 0 &&
364 	    (bootfs[len] == '/' || bootfs[len] == '\0'))
365 		return (B_TRUE);
366 
367 	return (B_FALSE);
368 }
369 
370 /*
371  * Inspect the configuration to determine if any of the devices contain
372  * an EFI label.
373  */
374 static boolean_t
375 pool_uses_efi(nvlist_t *config)
376 {
377 	nvlist_t **child;
378 	uint_t c, children;
379 
380 	if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
381 	    &child, &children) != 0)
382 		return (read_efi_label(config, NULL) >= 0);
383 
384 	for (c = 0; c < children; c++) {
385 		if (pool_uses_efi(child[c]))
386 			return (B_TRUE);
387 	}
388 	return (B_FALSE);
389 }
390 
391 boolean_t
392 zpool_is_bootable(zpool_handle_t *zhp)
393 {
394 	char bootfs[ZPOOL_MAXNAMELEN];
395 
396 	return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
397 	    sizeof (bootfs), NULL, B_FALSE) == 0 && strncmp(bootfs, "-",
398 	    sizeof (bootfs)) != 0);
399 }
400 
401 
402 /*
403  * Given an nvlist of zpool properties to be set, validate that they are
404  * correct, and parse any numeric properties (index, boolean, etc) if they are
405  * specified as strings.
406  */
407 static nvlist_t *
408 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
409     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
410 {
411 	nvpair_t *elem;
412 	nvlist_t *retprops;
413 	zpool_prop_t prop;
414 	char *strval;
415 	uint64_t intval;
416 	char *slash, *check;
417 	struct stat64 statbuf;
418 	zpool_handle_t *zhp;
419 	nvlist_t *nvroot;
420 
421 	if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
422 		(void) no_memory(hdl);
423 		return (NULL);
424 	}
425 
426 	elem = NULL;
427 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
428 		const char *propname = nvpair_name(elem);
429 
430 		prop = zpool_name_to_prop(propname);
431 		if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
432 			int err;
433 			char *fname = strchr(propname, '@') + 1;
434 
435 			err = zfeature_lookup_name(fname, NULL);
436 			if (err != 0) {
437 				ASSERT3U(err, ==, ENOENT);
438 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
439 				    "invalid feature '%s'"), fname);
440 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
441 				goto error;
442 			}
443 
444 			if (nvpair_type(elem) != DATA_TYPE_STRING) {
445 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
446 				    "'%s' must be a string"), propname);
447 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
448 				goto error;
449 			}
450 
451 			(void) nvpair_value_string(elem, &strval);
452 			if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
453 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
454 				    "property '%s' can only be set to "
455 				    "'enabled'"), propname);
456 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
457 				goto error;
458 			}
459 
460 			if (nvlist_add_uint64(retprops, propname, 0) != 0) {
461 				(void) no_memory(hdl);
462 				goto error;
463 			}
464 			continue;
465 		}
466 
467 		/*
468 		 * Make sure this property is valid and applies to this type.
469 		 */
470 		if (prop == ZPROP_INVAL) {
471 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
472 			    "invalid property '%s'"), propname);
473 			(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
474 			goto error;
475 		}
476 
477 		if (zpool_prop_readonly(prop)) {
478 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
479 			    "is readonly"), propname);
480 			(void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
481 			goto error;
482 		}
483 
484 		if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
485 		    &strval, &intval, errbuf) != 0)
486 			goto error;
487 
488 		/*
489 		 * Perform additional checking for specific properties.
490 		 */
491 		switch (prop) {
492 		case ZPOOL_PROP_VERSION:
493 			if (intval < version ||
494 			    !SPA_VERSION_IS_SUPPORTED(intval)) {
495 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
496 				    "property '%s' number %d is invalid."),
497 				    propname, intval);
498 				(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
499 				goto error;
500 			}
501 			break;
502 
503 		case ZPOOL_PROP_BOOTFS:
504 			if (flags.create || flags.import) {
505 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
506 				    "property '%s' cannot be set at creation "
507 				    "or import time"), propname);
508 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
509 				goto error;
510 			}
511 
512 			if (version < SPA_VERSION_BOOTFS) {
513 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
514 				    "pool must be upgraded to support "
515 				    "'%s' property"), propname);
516 				(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
517 				goto error;
518 			}
519 
520 			/*
521 			 * bootfs property value has to be a dataset name and
522 			 * the dataset has to be in the same pool as it sets to.
523 			 */
524 			if (strval[0] != '\0' && !bootfs_name_valid(poolname,
525 			    strval)) {
526 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
527 				    "is an invalid name"), strval);
528 				(void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
529 				goto error;
530 			}
531 
532 			if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
533 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
534 				    "could not open pool '%s'"), poolname);
535 				(void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
536 				goto error;
537 			}
538 			verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
539 			    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
540 
541 			/*
542 			 * bootfs property cannot be set on a disk which has
543 			 * been EFI labeled.
544 			 */
545 			if (pool_uses_efi(nvroot)) {
546 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
547 				    "property '%s' not supported on "
548 				    "EFI labeled devices"), propname);
549 				(void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
550 				zpool_close(zhp);
551 				goto error;
552 			}
553 			zpool_close(zhp);
554 			break;
555 
556 		case ZPOOL_PROP_ALTROOT:
557 			if (!flags.create && !flags.import) {
558 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
559 				    "property '%s' can only be set during pool "
560 				    "creation or import"), propname);
561 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
562 				goto error;
563 			}
564 
565 			if (strval[0] != '/') {
566 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
567 				    "bad alternate root '%s'"), strval);
568 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
569 				goto error;
570 			}
571 			break;
572 
573 		case ZPOOL_PROP_CACHEFILE:
574 			if (strval[0] == '\0')
575 				break;
576 
577 			if (strcmp(strval, "none") == 0)
578 				break;
579 
580 			if (strval[0] != '/') {
581 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
582 				    "property '%s' must be empty, an "
583 				    "absolute path, or 'none'"), propname);
584 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
585 				goto error;
586 			}
587 
588 			slash = strrchr(strval, '/');
589 
590 			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
591 			    strcmp(slash, "/..") == 0) {
592 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
593 				    "'%s' is not a valid file"), strval);
594 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
595 				goto error;
596 			}
597 
598 			*slash = '\0';
599 
600 			if (strval[0] != '\0' &&
601 			    (stat64(strval, &statbuf) != 0 ||
602 			    !S_ISDIR(statbuf.st_mode))) {
603 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
604 				    "'%s' is not a valid directory"),
605 				    strval);
606 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
607 				goto error;
608 			}
609 
610 			*slash = '/';
611 			break;
612 
613 		case ZPOOL_PROP_COMMENT:
614 			for (check = strval; *check != '\0'; check++) {
615 				if (!isprint(*check)) {
616 					zfs_error_aux(hdl,
617 					    dgettext(TEXT_DOMAIN,
618 					    "comment may only have printable "
619 					    "characters"));
620 					(void) zfs_error(hdl, EZFS_BADPROP,
621 					    errbuf);
622 					goto error;
623 				}
624 			}
625 			if (strlen(strval) > ZPROP_MAX_COMMENT) {
626 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
627 				    "comment must not exceed %d characters"),
628 				    ZPROP_MAX_COMMENT);
629 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
630 				goto error;
631 			}
632 			break;
633 		case ZPOOL_PROP_READONLY:
634 			if (!flags.import) {
635 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
636 				    "property '%s' can only be set at "
637 				    "import time"), propname);
638 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
639 				goto error;
640 			}
641 			break;
642 		}
643 	}
644 
645 	return (retprops);
646 error:
647 	nvlist_free(retprops);
648 	return (NULL);
649 }
650 
651 /*
652  * Set zpool property : propname=propval.
653  */
654 int
655 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
656 {
657 	zfs_cmd_t zc = { 0 };
658 	int ret = -1;
659 	char errbuf[1024];
660 	nvlist_t *nvl = NULL;
661 	nvlist_t *realprops;
662 	uint64_t version;
663 	prop_flags_t flags = { 0 };
664 
665 	(void) snprintf(errbuf, sizeof (errbuf),
666 	    dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
667 	    zhp->zpool_name);
668 
669 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
670 		return (no_memory(zhp->zpool_hdl));
671 
672 	if (nvlist_add_string(nvl, propname, propval) != 0) {
673 		nvlist_free(nvl);
674 		return (no_memory(zhp->zpool_hdl));
675 	}
676 
677 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
678 	if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
679 	    zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
680 		nvlist_free(nvl);
681 		return (-1);
682 	}
683 
684 	nvlist_free(nvl);
685 	nvl = realprops;
686 
687 	/*
688 	 * Execute the corresponding ioctl() to set this property.
689 	 */
690 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
691 
692 	if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
693 		nvlist_free(nvl);
694 		return (-1);
695 	}
696 
697 	ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
698 
699 	zcmd_free_nvlists(&zc);
700 	nvlist_free(nvl);
701 
702 	if (ret)
703 		(void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
704 	else
705 		(void) zpool_props_refresh(zhp);
706 
707 	return (ret);
708 }
709 
710 int
711 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
712 {
713 	libzfs_handle_t *hdl = zhp->zpool_hdl;
714 	zprop_list_t *entry;
715 	char buf[ZFS_MAXPROPLEN];
716 	nvlist_t *features = NULL;
717 	zprop_list_t **last;
718 	boolean_t firstexpand = (NULL == *plp);
719 
720 	if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
721 		return (-1);
722 
723 	last = plp;
724 	while (*last != NULL)
725 		last = &(*last)->pl_next;
726 
727 	if ((*plp)->pl_all)
728 		features = zpool_get_features(zhp);
729 
730 	if ((*plp)->pl_all && firstexpand) {
731 		for (int i = 0; i < SPA_FEATURES; i++) {
732 			zprop_list_t *entry = zfs_alloc(hdl,
733 			    sizeof (zprop_list_t));
734 			entry->pl_prop = ZPROP_INVAL;
735 			entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
736 			    spa_feature_table[i].fi_uname);
737 			entry->pl_width = strlen(entry->pl_user_prop);
738 			entry->pl_all = B_TRUE;
739 
740 			*last = entry;
741 			last = &entry->pl_next;
742 		}
743 	}
744 
745 	/* add any unsupported features */
746 	for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
747 	    nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
748 		char *propname;
749 		boolean_t found;
750 		zprop_list_t *entry;
751 
752 		if (zfeature_is_supported(nvpair_name(nvp)))
753 			continue;
754 
755 		propname = zfs_asprintf(hdl, "unsupported@%s",
756 		    nvpair_name(nvp));
757 
758 		/*
759 		 * Before adding the property to the list make sure that no
760 		 * other pool already added the same property.
761 		 */
762 		found = B_FALSE;
763 		entry = *plp;
764 		while (entry != NULL) {
765 			if (entry->pl_user_prop != NULL &&
766 			    strcmp(propname, entry->pl_user_prop) == 0) {
767 				found = B_TRUE;
768 				break;
769 			}
770 			entry = entry->pl_next;
771 		}
772 		if (found) {
773 			free(propname);
774 			continue;
775 		}
776 
777 		entry = zfs_alloc(hdl, sizeof (zprop_list_t));
778 		entry->pl_prop = ZPROP_INVAL;
779 		entry->pl_user_prop = propname;
780 		entry->pl_width = strlen(entry->pl_user_prop);
781 		entry->pl_all = B_TRUE;
782 
783 		*last = entry;
784 		last = &entry->pl_next;
785 	}
786 
787 	for (entry = *plp; entry != NULL; entry = entry->pl_next) {
788 
789 		if (entry->pl_fixed)
790 			continue;
791 
792 		if (entry->pl_prop != ZPROP_INVAL &&
793 		    zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
794 		    NULL, B_FALSE) == 0) {
795 			if (strlen(buf) > entry->pl_width)
796 				entry->pl_width = strlen(buf);
797 		}
798 	}
799 
800 	return (0);
801 }
802 
803 /*
804  * Get the state for the given feature on the given ZFS pool.
805  */
806 int
807 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
808     size_t len)
809 {
810 	uint64_t refcount;
811 	boolean_t found = B_FALSE;
812 	nvlist_t *features = zpool_get_features(zhp);
813 	boolean_t supported;
814 	const char *feature = strchr(propname, '@') + 1;
815 
816 	supported = zpool_prop_feature(propname);
817 	ASSERT(supported || zfs_prop_unsupported(propname));
818 
819 	/*
820 	 * Convert from feature name to feature guid. This conversion is
821 	 * unecessary for unsupported@... properties because they already
822 	 * use guids.
823 	 */
824 	if (supported) {
825 		int ret;
826 		spa_feature_t fid;
827 
828 		ret = zfeature_lookup_name(feature, &fid);
829 		if (ret != 0) {
830 			(void) strlcpy(buf, "-", len);
831 			return (ENOTSUP);
832 		}
833 		feature = spa_feature_table[fid].fi_guid;
834 	}
835 
836 	if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
837 		found = B_TRUE;
838 
839 	if (supported) {
840 		if (!found) {
841 			(void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
842 		} else  {
843 			if (refcount == 0)
844 				(void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
845 			else
846 				(void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
847 		}
848 	} else {
849 		if (found) {
850 			if (refcount == 0) {
851 				(void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
852 			} else {
853 				(void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
854 			}
855 		} else {
856 			(void) strlcpy(buf, "-", len);
857 			return (ENOTSUP);
858 		}
859 	}
860 
861 	return (0);
862 }
863 
864 /*
865  * Don't start the slice at the default block of 34; many storage
866  * devices will use a stripe width of 128k, so start there instead.
867  */
868 #define	NEW_START_BLOCK	256
869 
870 /*
871  * Validate the given pool name, optionally putting an extended error message in
872  * 'buf'.
873  */
874 boolean_t
875 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
876 {
877 	namecheck_err_t why;
878 	char what;
879 	int ret;
880 
881 	ret = pool_namecheck(pool, &why, &what);
882 
883 	/*
884 	 * The rules for reserved pool names were extended at a later point.
885 	 * But we need to support users with existing pools that may now be
886 	 * invalid.  So we only check for this expanded set of names during a
887 	 * create (or import), and only in userland.
888 	 */
889 	if (ret == 0 && !isopen &&
890 	    (strncmp(pool, "mirror", 6) == 0 ||
891 	    strncmp(pool, "raidz", 5) == 0 ||
892 	    strncmp(pool, "spare", 5) == 0 ||
893 	    strcmp(pool, "log") == 0)) {
894 		if (hdl != NULL)
895 			zfs_error_aux(hdl,
896 			    dgettext(TEXT_DOMAIN, "name is reserved"));
897 		return (B_FALSE);
898 	}
899 
900 
901 	if (ret != 0) {
902 		if (hdl != NULL) {
903 			switch (why) {
904 			case NAME_ERR_TOOLONG:
905 				zfs_error_aux(hdl,
906 				    dgettext(TEXT_DOMAIN, "name is too long"));
907 				break;
908 
909 			case NAME_ERR_INVALCHAR:
910 				zfs_error_aux(hdl,
911 				    dgettext(TEXT_DOMAIN, "invalid character "
912 				    "'%c' in pool name"), what);
913 				break;
914 
915 			case NAME_ERR_NOLETTER:
916 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
917 				    "name must begin with a letter"));
918 				break;
919 
920 			case NAME_ERR_RESERVED:
921 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
922 				    "name is reserved"));
923 				break;
924 
925 			case NAME_ERR_DISKLIKE:
926 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
927 				    "pool name is reserved"));
928 				break;
929 
930 			case NAME_ERR_LEADING_SLASH:
931 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
932 				    "leading slash in name"));
933 				break;
934 
935 			case NAME_ERR_EMPTY_COMPONENT:
936 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
937 				    "empty component in name"));
938 				break;
939 
940 			case NAME_ERR_TRAILING_SLASH:
941 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
942 				    "trailing slash in name"));
943 				break;
944 
945 			case NAME_ERR_MULTIPLE_AT:
946 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
947 				    "multiple '@' delimiters in name"));
948 				break;
949 
950 			}
951 		}
952 		return (B_FALSE);
953 	}
954 
955 	return (B_TRUE);
956 }
957 
958 /*
959  * Open a handle to the given pool, even if the pool is currently in the FAULTED
960  * state.
961  */
962 zpool_handle_t *
963 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
964 {
965 	zpool_handle_t *zhp;
966 	boolean_t missing;
967 
968 	/*
969 	 * Make sure the pool name is valid.
970 	 */
971 	if (!zpool_name_valid(hdl, B_TRUE, pool)) {
972 		(void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
973 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
974 		    pool);
975 		return (NULL);
976 	}
977 
978 	if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
979 		return (NULL);
980 
981 	zhp->zpool_hdl = hdl;
982 	(void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
983 
984 	if (zpool_refresh_stats(zhp, &missing) != 0) {
985 		zpool_close(zhp);
986 		return (NULL);
987 	}
988 
989 	if (missing) {
990 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
991 		(void) zfs_error_fmt(hdl, EZFS_NOENT,
992 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
993 		zpool_close(zhp);
994 		return (NULL);
995 	}
996 
997 	return (zhp);
998 }
999 
1000 /*
1001  * Like the above, but silent on error.  Used when iterating over pools (because
1002  * the configuration cache may be out of date).
1003  */
1004 int
1005 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1006 {
1007 	zpool_handle_t *zhp;
1008 	boolean_t missing;
1009 
1010 	if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1011 		return (-1);
1012 
1013 	zhp->zpool_hdl = hdl;
1014 	(void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1015 
1016 	if (zpool_refresh_stats(zhp, &missing) != 0) {
1017 		zpool_close(zhp);
1018 		return (-1);
1019 	}
1020 
1021 	if (missing) {
1022 		zpool_close(zhp);
1023 		*ret = NULL;
1024 		return (0);
1025 	}
1026 
1027 	*ret = zhp;
1028 	return (0);
1029 }
1030 
1031 /*
1032  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1033  * state.
1034  */
1035 zpool_handle_t *
1036 zpool_open(libzfs_handle_t *hdl, const char *pool)
1037 {
1038 	zpool_handle_t *zhp;
1039 
1040 	if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1041 		return (NULL);
1042 
1043 	if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1044 		(void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1045 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1046 		zpool_close(zhp);
1047 		return (NULL);
1048 	}
1049 
1050 	return (zhp);
1051 }
1052 
1053 /*
1054  * Close the handle.  Simply frees the memory associated with the handle.
1055  */
1056 void
1057 zpool_close(zpool_handle_t *zhp)
1058 {
1059 	if (zhp->zpool_config)
1060 		nvlist_free(zhp->zpool_config);
1061 	if (zhp->zpool_old_config)
1062 		nvlist_free(zhp->zpool_old_config);
1063 	if (zhp->zpool_props)
1064 		nvlist_free(zhp->zpool_props);
1065 	free(zhp);
1066 }
1067 
1068 /*
1069  * Return the name of the pool.
1070  */
1071 const char *
1072 zpool_get_name(zpool_handle_t *zhp)
1073 {
1074 	return (zhp->zpool_name);
1075 }
1076 
1077 
1078 /*
1079  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1080  */
1081 int
1082 zpool_get_state(zpool_handle_t *zhp)
1083 {
1084 	return (zhp->zpool_state);
1085 }
1086 
1087 /*
1088  * Create the named pool, using the provided vdev list.  It is assumed
1089  * that the consumer has already validated the contents of the nvlist, so we
1090  * don't have to worry about error semantics.
1091  */
1092 int
1093 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1094     nvlist_t *props, nvlist_t *fsprops)
1095 {
1096 	zfs_cmd_t zc = { 0 };
1097 	nvlist_t *zc_fsprops = NULL;
1098 	nvlist_t *zc_props = NULL;
1099 	char msg[1024];
1100 	int ret = -1;
1101 
1102 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1103 	    "cannot create '%s'"), pool);
1104 
1105 	if (!zpool_name_valid(hdl, B_FALSE, pool))
1106 		return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1107 
1108 	if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1109 		return (-1);
1110 
1111 	if (props) {
1112 		prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1113 
1114 		if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1115 		    SPA_VERSION_1, flags, msg)) == NULL) {
1116 			goto create_failed;
1117 		}
1118 	}
1119 
1120 	if (fsprops) {
1121 		uint64_t zoned;
1122 		char *zonestr;
1123 
1124 		zoned = ((nvlist_lookup_string(fsprops,
1125 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1126 		    strcmp(zonestr, "on") == 0);
1127 
1128 		if ((zc_fsprops = zfs_valid_proplist(hdl,
1129 		    ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
1130 			goto create_failed;
1131 		}
1132 		if (!zc_props &&
1133 		    (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1134 			goto create_failed;
1135 		}
1136 		if (nvlist_add_nvlist(zc_props,
1137 		    ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1138 			goto create_failed;
1139 		}
1140 	}
1141 
1142 	if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1143 		goto create_failed;
1144 
1145 	(void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1146 
1147 	if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1148 
1149 		zcmd_free_nvlists(&zc);
1150 		nvlist_free(zc_props);
1151 		nvlist_free(zc_fsprops);
1152 
1153 		switch (errno) {
1154 		case EBUSY:
1155 			/*
1156 			 * This can happen if the user has specified the same
1157 			 * device multiple times.  We can't reliably detect this
1158 			 * until we try to add it and see we already have a
1159 			 * label.
1160 			 */
1161 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1162 			    "one or more vdevs refer to the same device"));
1163 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1164 
1165 		case EOVERFLOW:
1166 			/*
1167 			 * This occurs when one of the devices is below
1168 			 * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1169 			 * device was the problem device since there's no
1170 			 * reliable way to determine device size from userland.
1171 			 */
1172 			{
1173 				char buf[64];
1174 
1175 				zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1176 
1177 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1178 				    "one or more devices is less than the "
1179 				    "minimum size (%s)"), buf);
1180 			}
1181 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1182 
1183 		case ENOSPC:
1184 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1185 			    "one or more devices is out of space"));
1186 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1187 
1188 		case ENOTBLK:
1189 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1190 			    "cache device must be a disk or disk slice"));
1191 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1192 
1193 		default:
1194 			return (zpool_standard_error(hdl, errno, msg));
1195 		}
1196 	}
1197 
1198 create_failed:
1199 	zcmd_free_nvlists(&zc);
1200 	nvlist_free(zc_props);
1201 	nvlist_free(zc_fsprops);
1202 	return (ret);
1203 }
1204 
1205 /*
1206  * Destroy the given pool.  It is up to the caller to ensure that there are no
1207  * datasets left in the pool.
1208  */
1209 int
1210 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1211 {
1212 	zfs_cmd_t zc = { 0 };
1213 	zfs_handle_t *zfp = NULL;
1214 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1215 	char msg[1024];
1216 
1217 	if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1218 	    (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1219 		return (-1);
1220 
1221 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1222 	zc.zc_history = (uint64_t)(uintptr_t)log_str;
1223 
1224 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1225 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1226 		    "cannot destroy '%s'"), zhp->zpool_name);
1227 
1228 		if (errno == EROFS) {
1229 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1230 			    "one or more devices is read only"));
1231 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1232 		} else {
1233 			(void) zpool_standard_error(hdl, errno, msg);
1234 		}
1235 
1236 		if (zfp)
1237 			zfs_close(zfp);
1238 		return (-1);
1239 	}
1240 
1241 	if (zfp) {
1242 		remove_mountpoint(zfp);
1243 		zfs_close(zfp);
1244 	}
1245 
1246 	return (0);
1247 }
1248 
1249 /*
1250  * Add the given vdevs to the pool.  The caller must have already performed the
1251  * necessary verification to ensure that the vdev specification is well-formed.
1252  */
1253 int
1254 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1255 {
1256 	zfs_cmd_t zc = { 0 };
1257 	int ret;
1258 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1259 	char msg[1024];
1260 	nvlist_t **spares, **l2cache;
1261 	uint_t nspares, nl2cache;
1262 
1263 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1264 	    "cannot add to '%s'"), zhp->zpool_name);
1265 
1266 	if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1267 	    SPA_VERSION_SPARES &&
1268 	    nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1269 	    &spares, &nspares) == 0) {
1270 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1271 		    "upgraded to add hot spares"));
1272 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
1273 	}
1274 
1275 	if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1276 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1277 		uint64_t s;
1278 
1279 		for (s = 0; s < nspares; s++) {
1280 			char *path;
1281 
1282 			if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1283 			    &path) == 0 && pool_uses_efi(spares[s])) {
1284 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1285 				    "device '%s' contains an EFI label and "
1286 				    "cannot be used on root pools."),
1287 				    zpool_vdev_name(hdl, NULL, spares[s],
1288 				    B_FALSE));
1289 				return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1290 			}
1291 		}
1292 	}
1293 
1294 	if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1295 	    SPA_VERSION_L2CACHE &&
1296 	    nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1297 	    &l2cache, &nl2cache) == 0) {
1298 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1299 		    "upgraded to add cache devices"));
1300 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
1301 	}
1302 
1303 	if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1304 		return (-1);
1305 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1306 
1307 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1308 		switch (errno) {
1309 		case EBUSY:
1310 			/*
1311 			 * This can happen if the user has specified the same
1312 			 * device multiple times.  We can't reliably detect this
1313 			 * until we try to add it and see we already have a
1314 			 * label.
1315 			 */
1316 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1317 			    "one or more vdevs refer to the same device"));
1318 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1319 			break;
1320 
1321 		case EOVERFLOW:
1322 			/*
1323 			 * This occurrs when one of the devices is below
1324 			 * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1325 			 * device was the problem device since there's no
1326 			 * reliable way to determine device size from userland.
1327 			 */
1328 			{
1329 				char buf[64];
1330 
1331 				zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1332 
1333 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1334 				    "device is less than the minimum "
1335 				    "size (%s)"), buf);
1336 			}
1337 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1338 			break;
1339 
1340 		case ENOTSUP:
1341 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1342 			    "pool must be upgraded to add these vdevs"));
1343 			(void) zfs_error(hdl, EZFS_BADVERSION, msg);
1344 			break;
1345 
1346 		case EDOM:
1347 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1348 			    "root pool can not have multiple vdevs"
1349 			    " or separate logs"));
1350 			(void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1351 			break;
1352 
1353 		case ENOTBLK:
1354 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1355 			    "cache device must be a disk or disk slice"));
1356 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1357 			break;
1358 
1359 		default:
1360 			(void) zpool_standard_error(hdl, errno, msg);
1361 		}
1362 
1363 		ret = -1;
1364 	} else {
1365 		ret = 0;
1366 	}
1367 
1368 	zcmd_free_nvlists(&zc);
1369 
1370 	return (ret);
1371 }
1372 
1373 /*
1374  * Exports the pool from the system.  The caller must ensure that there are no
1375  * mounted datasets in the pool.
1376  */
1377 static int
1378 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1379     const char *log_str)
1380 {
1381 	zfs_cmd_t zc = { 0 };
1382 	char msg[1024];
1383 
1384 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1385 	    "cannot export '%s'"), zhp->zpool_name);
1386 
1387 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1388 	zc.zc_cookie = force;
1389 	zc.zc_guid = hardforce;
1390 	zc.zc_history = (uint64_t)(uintptr_t)log_str;
1391 
1392 	if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1393 		switch (errno) {
1394 		case EXDEV:
1395 			zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1396 			    "use '-f' to override the following errors:\n"
1397 			    "'%s' has an active shared spare which could be"
1398 			    " used by other pools once '%s' is exported."),
1399 			    zhp->zpool_name, zhp->zpool_name);
1400 			return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1401 			    msg));
1402 		default:
1403 			return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1404 			    msg));
1405 		}
1406 	}
1407 
1408 	return (0);
1409 }
1410 
1411 int
1412 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1413 {
1414 	return (zpool_export_common(zhp, force, B_FALSE, log_str));
1415 }
1416 
1417 int
1418 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1419 {
1420 	return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1421 }
1422 
1423 static void
1424 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1425     nvlist_t *config)
1426 {
1427 	nvlist_t *nv = NULL;
1428 	uint64_t rewindto;
1429 	int64_t loss = -1;
1430 	struct tm t;
1431 	char timestr[128];
1432 
1433 	if (!hdl->libzfs_printerr || config == NULL)
1434 		return;
1435 
1436 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1437 	    nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1438 		return;
1439 	}
1440 
1441 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1442 		return;
1443 	(void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1444 
1445 	if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1446 	    strftime(timestr, 128, 0, &t) != 0) {
1447 		if (dryrun) {
1448 			(void) printf(dgettext(TEXT_DOMAIN,
1449 			    "Would be able to return %s "
1450 			    "to its state as of %s.\n"),
1451 			    name, timestr);
1452 		} else {
1453 			(void) printf(dgettext(TEXT_DOMAIN,
1454 			    "Pool %s returned to its state as of %s.\n"),
1455 			    name, timestr);
1456 		}
1457 		if (loss > 120) {
1458 			(void) printf(dgettext(TEXT_DOMAIN,
1459 			    "%s approximately %lld "),
1460 			    dryrun ? "Would discard" : "Discarded",
1461 			    (loss + 30) / 60);
1462 			(void) printf(dgettext(TEXT_DOMAIN,
1463 			    "minutes of transactions.\n"));
1464 		} else if (loss > 0) {
1465 			(void) printf(dgettext(TEXT_DOMAIN,
1466 			    "%s approximately %lld "),
1467 			    dryrun ? "Would discard" : "Discarded", loss);
1468 			(void) printf(dgettext(TEXT_DOMAIN,
1469 			    "seconds of transactions.\n"));
1470 		}
1471 	}
1472 }
1473 
1474 void
1475 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1476     nvlist_t *config)
1477 {
1478 	nvlist_t *nv = NULL;
1479 	int64_t loss = -1;
1480 	uint64_t edata = UINT64_MAX;
1481 	uint64_t rewindto;
1482 	struct tm t;
1483 	char timestr[128];
1484 
1485 	if (!hdl->libzfs_printerr)
1486 		return;
1487 
1488 	if (reason >= 0)
1489 		(void) printf(dgettext(TEXT_DOMAIN, "action: "));
1490 	else
1491 		(void) printf(dgettext(TEXT_DOMAIN, "\t"));
1492 
1493 	/* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1494 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1495 	    nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1496 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1497 		goto no_info;
1498 
1499 	(void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1500 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1501 	    &edata);
1502 
1503 	(void) printf(dgettext(TEXT_DOMAIN,
1504 	    "Recovery is possible, but will result in some data loss.\n"));
1505 
1506 	if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1507 	    strftime(timestr, 128, 0, &t) != 0) {
1508 		(void) printf(dgettext(TEXT_DOMAIN,
1509 		    "\tReturning the pool to its state as of %s\n"
1510 		    "\tshould correct the problem.  "),
1511 		    timestr);
1512 	} else {
1513 		(void) printf(dgettext(TEXT_DOMAIN,
1514 		    "\tReverting the pool to an earlier state "
1515 		    "should correct the problem.\n\t"));
1516 	}
1517 
1518 	if (loss > 120) {
1519 		(void) printf(dgettext(TEXT_DOMAIN,
1520 		    "Approximately %lld minutes of data\n"
1521 		    "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1522 	} else if (loss > 0) {
1523 		(void) printf(dgettext(TEXT_DOMAIN,
1524 		    "Approximately %lld seconds of data\n"
1525 		    "\tmust be discarded, irreversibly.  "), loss);
1526 	}
1527 	if (edata != 0 && edata != UINT64_MAX) {
1528 		if (edata == 1) {
1529 			(void) printf(dgettext(TEXT_DOMAIN,
1530 			    "After rewind, at least\n"
1531 			    "\tone persistent user-data error will remain.  "));
1532 		} else {
1533 			(void) printf(dgettext(TEXT_DOMAIN,
1534 			    "After rewind, several\n"
1535 			    "\tpersistent user-data errors will remain.  "));
1536 		}
1537 	}
1538 	(void) printf(dgettext(TEXT_DOMAIN,
1539 	    "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1540 	    reason >= 0 ? "clear" : "import", name);
1541 
1542 	(void) printf(dgettext(TEXT_DOMAIN,
1543 	    "A scrub of the pool\n"
1544 	    "\tis strongly recommended after recovery.\n"));
1545 	return;
1546 
1547 no_info:
1548 	(void) printf(dgettext(TEXT_DOMAIN,
1549 	    "Destroy and re-create the pool from\n\ta backup source.\n"));
1550 }
1551 
1552 /*
1553  * zpool_import() is a contracted interface. Should be kept the same
1554  * if possible.
1555  *
1556  * Applications should use zpool_import_props() to import a pool with
1557  * new properties value to be set.
1558  */
1559 int
1560 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1561     char *altroot)
1562 {
1563 	nvlist_t *props = NULL;
1564 	int ret;
1565 
1566 	if (altroot != NULL) {
1567 		if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1568 			return (zfs_error_fmt(hdl, EZFS_NOMEM,
1569 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1570 			    newname));
1571 		}
1572 
1573 		if (nvlist_add_string(props,
1574 		    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1575 		    nvlist_add_string(props,
1576 		    zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1577 			nvlist_free(props);
1578 			return (zfs_error_fmt(hdl, EZFS_NOMEM,
1579 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1580 			    newname));
1581 		}
1582 	}
1583 
1584 	ret = zpool_import_props(hdl, config, newname, props,
1585 	    ZFS_IMPORT_NORMAL);
1586 	if (props)
1587 		nvlist_free(props);
1588 	return (ret);
1589 }
1590 
1591 static void
1592 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1593     int indent)
1594 {
1595 	nvlist_t **child;
1596 	uint_t c, children;
1597 	char *vname;
1598 	uint64_t is_log = 0;
1599 
1600 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1601 	    &is_log);
1602 
1603 	if (name != NULL)
1604 		(void) printf("\t%*s%s%s\n", indent, "", name,
1605 		    is_log ? " [log]" : "");
1606 
1607 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1608 	    &child, &children) != 0)
1609 		return;
1610 
1611 	for (c = 0; c < children; c++) {
1612 		vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1613 		print_vdev_tree(hdl, vname, child[c], indent + 2);
1614 		free(vname);
1615 	}
1616 }
1617 
1618 void
1619 zpool_print_unsup_feat(nvlist_t *config)
1620 {
1621 	nvlist_t *nvinfo, *unsup_feat;
1622 
1623 	verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1624 	    0);
1625 	verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1626 	    &unsup_feat) == 0);
1627 
1628 	for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1629 	    nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1630 		char *desc;
1631 
1632 		verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1633 		verify(nvpair_value_string(nvp, &desc) == 0);
1634 
1635 		if (strlen(desc) > 0)
1636 			(void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1637 		else
1638 			(void) printf("\t%s\n", nvpair_name(nvp));
1639 	}
1640 }
1641 
1642 /*
1643  * Import the given pool using the known configuration and a list of
1644  * properties to be set. The configuration should have come from
1645  * zpool_find_import(). The 'newname' parameters control whether the pool
1646  * is imported with a different name.
1647  */
1648 int
1649 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1650     nvlist_t *props, int flags)
1651 {
1652 	zfs_cmd_t zc = { 0 };
1653 	zpool_rewind_policy_t policy;
1654 	nvlist_t *nv = NULL;
1655 	nvlist_t *nvinfo = NULL;
1656 	nvlist_t *missing = NULL;
1657 	char *thename;
1658 	char *origname;
1659 	int ret;
1660 	int error = 0;
1661 	char errbuf[1024];
1662 
1663 	verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1664 	    &origname) == 0);
1665 
1666 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1667 	    "cannot import pool '%s'"), origname);
1668 
1669 	if (newname != NULL) {
1670 		if (!zpool_name_valid(hdl, B_FALSE, newname))
1671 			return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1672 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1673 			    newname));
1674 		thename = (char *)newname;
1675 	} else {
1676 		thename = origname;
1677 	}
1678 
1679 	if (props) {
1680 		uint64_t version;
1681 		prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1682 
1683 		verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1684 		    &version) == 0);
1685 
1686 		if ((props = zpool_valid_proplist(hdl, origname,
1687 		    props, version, flags, errbuf)) == NULL) {
1688 			return (-1);
1689 		} else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1690 			nvlist_free(props);
1691 			return (-1);
1692 		}
1693 	}
1694 
1695 	(void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1696 
1697 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1698 	    &zc.zc_guid) == 0);
1699 
1700 	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1701 		nvlist_free(props);
1702 		return (-1);
1703 	}
1704 	if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1705 		nvlist_free(props);
1706 		return (-1);
1707 	}
1708 
1709 	zc.zc_cookie = flags;
1710 	while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1711 	    errno == ENOMEM) {
1712 		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1713 			zcmd_free_nvlists(&zc);
1714 			return (-1);
1715 		}
1716 	}
1717 	if (ret != 0)
1718 		error = errno;
1719 
1720 	(void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1721 	zpool_get_rewind_policy(config, &policy);
1722 
1723 	if (error) {
1724 		char desc[1024];
1725 
1726 		/*
1727 		 * Dry-run failed, but we print out what success
1728 		 * looks like if we found a best txg
1729 		 */
1730 		if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1731 			zpool_rewind_exclaim(hdl, newname ? origname : thename,
1732 			    B_TRUE, nv);
1733 			nvlist_free(nv);
1734 			return (-1);
1735 		}
1736 
1737 		if (newname == NULL)
1738 			(void) snprintf(desc, sizeof (desc),
1739 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1740 			    thename);
1741 		else
1742 			(void) snprintf(desc, sizeof (desc),
1743 			    dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1744 			    origname, thename);
1745 
1746 		switch (error) {
1747 		case ENOTSUP:
1748 			if (nv != NULL && nvlist_lookup_nvlist(nv,
1749 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1750 			    nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1751 				(void) printf(dgettext(TEXT_DOMAIN, "This "
1752 				    "pool uses the following feature(s) not "
1753 				    "supported by this system:\n"));
1754 				zpool_print_unsup_feat(nv);
1755 				if (nvlist_exists(nvinfo,
1756 				    ZPOOL_CONFIG_CAN_RDONLY)) {
1757 					(void) printf(dgettext(TEXT_DOMAIN,
1758 					    "All unsupported features are only "
1759 					    "required for writing to the pool."
1760 					    "\nThe pool can be imported using "
1761 					    "'-o readonly=on'.\n"));
1762 				}
1763 			}
1764 			/*
1765 			 * Unsupported version.
1766 			 */
1767 			(void) zfs_error(hdl, EZFS_BADVERSION, desc);
1768 			break;
1769 
1770 		case EINVAL:
1771 			(void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1772 			break;
1773 
1774 		case EROFS:
1775 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1776 			    "one or more devices is read only"));
1777 			(void) zfs_error(hdl, EZFS_BADDEV, desc);
1778 			break;
1779 
1780 		case ENXIO:
1781 			if (nv && nvlist_lookup_nvlist(nv,
1782 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1783 			    nvlist_lookup_nvlist(nvinfo,
1784 			    ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1785 				(void) printf(dgettext(TEXT_DOMAIN,
1786 				    "The devices below are missing, use "
1787 				    "'-m' to import the pool anyway:\n"));
1788 				print_vdev_tree(hdl, NULL, missing, 2);
1789 				(void) printf("\n");
1790 			}
1791 			(void) zpool_standard_error(hdl, error, desc);
1792 			break;
1793 
1794 		case EEXIST:
1795 			(void) zpool_standard_error(hdl, error, desc);
1796 			break;
1797 
1798 		default:
1799 			(void) zpool_standard_error(hdl, error, desc);
1800 			zpool_explain_recover(hdl,
1801 			    newname ? origname : thename, -error, nv);
1802 			break;
1803 		}
1804 
1805 		nvlist_free(nv);
1806 		ret = -1;
1807 	} else {
1808 		zpool_handle_t *zhp;
1809 
1810 		/*
1811 		 * This should never fail, but play it safe anyway.
1812 		 */
1813 		if (zpool_open_silent(hdl, thename, &zhp) != 0)
1814 			ret = -1;
1815 		else if (zhp != NULL)
1816 			zpool_close(zhp);
1817 		if (policy.zrp_request &
1818 		    (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1819 			zpool_rewind_exclaim(hdl, newname ? origname : thename,
1820 			    ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1821 		}
1822 		nvlist_free(nv);
1823 		return (0);
1824 	}
1825 
1826 	zcmd_free_nvlists(&zc);
1827 	nvlist_free(props);
1828 
1829 	return (ret);
1830 }
1831 
1832 /*
1833  * Scan the pool.
1834  */
1835 int
1836 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1837 {
1838 	zfs_cmd_t zc = { 0 };
1839 	char msg[1024];
1840 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1841 
1842 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1843 	zc.zc_cookie = func;
1844 
1845 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1846 	    (errno == ENOENT && func != POOL_SCAN_NONE))
1847 		return (0);
1848 
1849 	if (func == POOL_SCAN_SCRUB) {
1850 		(void) snprintf(msg, sizeof (msg),
1851 		    dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1852 	} else if (func == POOL_SCAN_NONE) {
1853 		(void) snprintf(msg, sizeof (msg),
1854 		    dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1855 		    zc.zc_name);
1856 	} else {
1857 		assert(!"unexpected result");
1858 	}
1859 
1860 	if (errno == EBUSY) {
1861 		nvlist_t *nvroot;
1862 		pool_scan_stat_t *ps = NULL;
1863 		uint_t psc;
1864 
1865 		verify(nvlist_lookup_nvlist(zhp->zpool_config,
1866 		    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1867 		(void) nvlist_lookup_uint64_array(nvroot,
1868 		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1869 		if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1870 			return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1871 		else
1872 			return (zfs_error(hdl, EZFS_RESILVERING, msg));
1873 	} else if (errno == ENOENT) {
1874 		return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1875 	} else {
1876 		return (zpool_standard_error(hdl, errno, msg));
1877 	}
1878 }
1879 
1880 /*
1881  * This provides a very minimal check whether a given string is likely a
1882  * c#t#d# style string.  Users of this are expected to do their own
1883  * verification of the s# part.
1884  */
1885 #define	CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1886 
1887 /*
1888  * More elaborate version for ones which may start with "/dev/dsk/"
1889  * and the like.
1890  */
1891 static int
1892 ctd_check_path(char *str) {
1893 	/*
1894 	 * If it starts with a slash, check the last component.
1895 	 */
1896 	if (str && str[0] == '/') {
1897 		char *tmp = strrchr(str, '/');
1898 
1899 		/*
1900 		 * If it ends in "/old", check the second-to-last
1901 		 * component of the string instead.
1902 		 */
1903 		if (tmp != str && strcmp(tmp, "/old") == 0) {
1904 			for (tmp--; *tmp != '/'; tmp--)
1905 				;
1906 		}
1907 		str = tmp + 1;
1908 	}
1909 	return (CTD_CHECK(str));
1910 }
1911 
1912 /*
1913  * Find a vdev that matches the search criteria specified. We use the
1914  * the nvpair name to determine how we should look for the device.
1915  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1916  * spare; but FALSE if its an INUSE spare.
1917  */
1918 static nvlist_t *
1919 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1920     boolean_t *l2cache, boolean_t *log)
1921 {
1922 	uint_t c, children;
1923 	nvlist_t **child;
1924 	nvlist_t *ret;
1925 	uint64_t is_log;
1926 	char *srchkey;
1927 	nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1928 
1929 	/* Nothing to look for */
1930 	if (search == NULL || pair == NULL)
1931 		return (NULL);
1932 
1933 	/* Obtain the key we will use to search */
1934 	srchkey = nvpair_name(pair);
1935 
1936 	switch (nvpair_type(pair)) {
1937 	case DATA_TYPE_UINT64:
1938 		if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1939 			uint64_t srchval, theguid;
1940 
1941 			verify(nvpair_value_uint64(pair, &srchval) == 0);
1942 			verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1943 			    &theguid) == 0);
1944 			if (theguid == srchval)
1945 				return (nv);
1946 		}
1947 		break;
1948 
1949 	case DATA_TYPE_STRING: {
1950 		char *srchval, *val;
1951 
1952 		verify(nvpair_value_string(pair, &srchval) == 0);
1953 		if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1954 			break;
1955 
1956 		/*
1957 		 * Search for the requested value. Special cases:
1958 		 *
1959 		 * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1960 		 *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1961 		 *   but included in the string, so this matches around it.
1962 		 * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1963 		 *
1964 		 * Otherwise, all other searches are simple string compares.
1965 		 */
1966 		if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1967 		    ctd_check_path(val)) {
1968 			uint64_t wholedisk = 0;
1969 
1970 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1971 			    &wholedisk);
1972 			if (wholedisk) {
1973 				int slen = strlen(srchval);
1974 				int vlen = strlen(val);
1975 
1976 				if (slen != vlen - 2)
1977 					break;
1978 
1979 				/*
1980 				 * make_leaf_vdev() should only set
1981 				 * wholedisk for ZPOOL_CONFIG_PATHs which
1982 				 * will include "/dev/dsk/", giving plenty of
1983 				 * room for the indices used next.
1984 				 */
1985 				ASSERT(vlen >= 6);
1986 
1987 				/*
1988 				 * strings identical except trailing "s0"
1989 				 */
1990 				if (strcmp(&val[vlen - 2], "s0") == 0 &&
1991 				    strncmp(srchval, val, slen) == 0)
1992 					return (nv);
1993 
1994 				/*
1995 				 * strings identical except trailing "s0/old"
1996 				 */
1997 				if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
1998 				    strcmp(&srchval[slen - 4], "/old") == 0 &&
1999 				    strncmp(srchval, val, slen - 4) == 0)
2000 					return (nv);
2001 
2002 				break;
2003 			}
2004 		} else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2005 			char *type, *idx, *end, *p;
2006 			uint64_t id, vdev_id;
2007 
2008 			/*
2009 			 * Determine our vdev type, keeping in mind
2010 			 * that the srchval is composed of a type and
2011 			 * vdev id pair (i.e. mirror-4).
2012 			 */
2013 			if ((type = strdup(srchval)) == NULL)
2014 				return (NULL);
2015 
2016 			if ((p = strrchr(type, '-')) == NULL) {
2017 				free(type);
2018 				break;
2019 			}
2020 			idx = p + 1;
2021 			*p = '\0';
2022 
2023 			/*
2024 			 * If the types don't match then keep looking.
2025 			 */
2026 			if (strncmp(val, type, strlen(val)) != 0) {
2027 				free(type);
2028 				break;
2029 			}
2030 
2031 			verify(strncmp(type, VDEV_TYPE_RAIDZ,
2032 			    strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2033 			    strncmp(type, VDEV_TYPE_MIRROR,
2034 			    strlen(VDEV_TYPE_MIRROR)) == 0);
2035 			verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2036 			    &id) == 0);
2037 
2038 			errno = 0;
2039 			vdev_id = strtoull(idx, &end, 10);
2040 
2041 			free(type);
2042 			if (errno != 0)
2043 				return (NULL);
2044 
2045 			/*
2046 			 * Now verify that we have the correct vdev id.
2047 			 */
2048 			if (vdev_id == id)
2049 				return (nv);
2050 		}
2051 
2052 		/*
2053 		 * Common case
2054 		 */
2055 		if (strcmp(srchval, val) == 0)
2056 			return (nv);
2057 		break;
2058 	}
2059 
2060 	default:
2061 		break;
2062 	}
2063 
2064 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2065 	    &child, &children) != 0)
2066 		return (NULL);
2067 
2068 	for (c = 0; c < children; c++) {
2069 		if ((ret = vdev_to_nvlist_iter(child[c], search,
2070 		    avail_spare, l2cache, NULL)) != NULL) {
2071 			/*
2072 			 * The 'is_log' value is only set for the toplevel
2073 			 * vdev, not the leaf vdevs.  So we always lookup the
2074 			 * log device from the root of the vdev tree (where
2075 			 * 'log' is non-NULL).
2076 			 */
2077 			if (log != NULL &&
2078 			    nvlist_lookup_uint64(child[c],
2079 			    ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2080 			    is_log) {
2081 				*log = B_TRUE;
2082 			}
2083 			return (ret);
2084 		}
2085 	}
2086 
2087 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2088 	    &child, &children) == 0) {
2089 		for (c = 0; c < children; c++) {
2090 			if ((ret = vdev_to_nvlist_iter(child[c], search,
2091 			    avail_spare, l2cache, NULL)) != NULL) {
2092 				*avail_spare = B_TRUE;
2093 				return (ret);
2094 			}
2095 		}
2096 	}
2097 
2098 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2099 	    &child, &children) == 0) {
2100 		for (c = 0; c < children; c++) {
2101 			if ((ret = vdev_to_nvlist_iter(child[c], search,
2102 			    avail_spare, l2cache, NULL)) != NULL) {
2103 				*l2cache = B_TRUE;
2104 				return (ret);
2105 			}
2106 		}
2107 	}
2108 
2109 	return (NULL);
2110 }
2111 
2112 /*
2113  * Given a physical path (minus the "/devices" prefix), find the
2114  * associated vdev.
2115  */
2116 nvlist_t *
2117 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2118     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2119 {
2120 	nvlist_t *search, *nvroot, *ret;
2121 
2122 	verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2123 	verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2124 
2125 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2126 	    &nvroot) == 0);
2127 
2128 	*avail_spare = B_FALSE;
2129 	*l2cache = B_FALSE;
2130 	if (log != NULL)
2131 		*log = B_FALSE;
2132 	ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2133 	nvlist_free(search);
2134 
2135 	return (ret);
2136 }
2137 
2138 /*
2139  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2140  */
2141 boolean_t
2142 zpool_vdev_is_interior(const char *name)
2143 {
2144 	if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2145 	    strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2146 		return (B_TRUE);
2147 	return (B_FALSE);
2148 }
2149 
2150 nvlist_t *
2151 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2152     boolean_t *l2cache, boolean_t *log)
2153 {
2154 	char buf[MAXPATHLEN];
2155 	char *end;
2156 	nvlist_t *nvroot, *search, *ret;
2157 	uint64_t guid;
2158 
2159 	verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2160 
2161 	guid = strtoull(path, &end, 10);
2162 	if (guid != 0 && *end == '\0') {
2163 		verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2164 	} else if (zpool_vdev_is_interior(path)) {
2165 		verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2166 	} else if (path[0] != '/') {
2167 		(void) snprintf(buf, sizeof (buf), "%s%s", "/dev/dsk/", path);
2168 		verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2169 	} else {
2170 		verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2171 	}
2172 
2173 	verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2174 	    &nvroot) == 0);
2175 
2176 	*avail_spare = B_FALSE;
2177 	*l2cache = B_FALSE;
2178 	if (log != NULL)
2179 		*log = B_FALSE;
2180 	ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2181 	nvlist_free(search);
2182 
2183 	return (ret);
2184 }
2185 
2186 static int
2187 vdev_online(nvlist_t *nv)
2188 {
2189 	uint64_t ival;
2190 
2191 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2192 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2193 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2194 		return (0);
2195 
2196 	return (1);
2197 }
2198 
2199 /*
2200  * Helper function for zpool_get_physpaths().
2201  */
2202 static int
2203 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2204     size_t *bytes_written)
2205 {
2206 	size_t bytes_left, pos, rsz;
2207 	char *tmppath;
2208 	const char *format;
2209 
2210 	if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2211 	    &tmppath) != 0)
2212 		return (EZFS_NODEVICE);
2213 
2214 	pos = *bytes_written;
2215 	bytes_left = physpath_size - pos;
2216 	format = (pos == 0) ? "%s" : " %s";
2217 
2218 	rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2219 	*bytes_written += rsz;
2220 
2221 	if (rsz >= bytes_left) {
2222 		/* if physpath was not copied properly, clear it */
2223 		if (bytes_left != 0) {
2224 			physpath[pos] = 0;
2225 		}
2226 		return (EZFS_NOSPC);
2227 	}
2228 	return (0);
2229 }
2230 
2231 static int
2232 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2233     size_t *rsz, boolean_t is_spare)
2234 {
2235 	char *type;
2236 	int ret;
2237 
2238 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2239 		return (EZFS_INVALCONFIG);
2240 
2241 	if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2242 		/*
2243 		 * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2244 		 * For a spare vdev, we only want to boot from the active
2245 		 * spare device.
2246 		 */
2247 		if (is_spare) {
2248 			uint64_t spare = 0;
2249 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2250 			    &spare);
2251 			if (!spare)
2252 				return (EZFS_INVALCONFIG);
2253 		}
2254 
2255 		if (vdev_online(nv)) {
2256 			if ((ret = vdev_get_one_physpath(nv, physpath,
2257 			    phypath_size, rsz)) != 0)
2258 				return (ret);
2259 		}
2260 	} else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2261 	    strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2262 	    (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2263 		nvlist_t **child;
2264 		uint_t count;
2265 		int i, ret;
2266 
2267 		if (nvlist_lookup_nvlist_array(nv,
2268 		    ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2269 			return (EZFS_INVALCONFIG);
2270 
2271 		for (i = 0; i < count; i++) {
2272 			ret = vdev_get_physpaths(child[i], physpath,
2273 			    phypath_size, rsz, is_spare);
2274 			if (ret == EZFS_NOSPC)
2275 				return (ret);
2276 		}
2277 	}
2278 
2279 	return (EZFS_POOL_INVALARG);
2280 }
2281 
2282 /*
2283  * Get phys_path for a root pool config.
2284  * Return 0 on success; non-zero on failure.
2285  */
2286 static int
2287 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2288 {
2289 	size_t rsz;
2290 	nvlist_t *vdev_root;
2291 	nvlist_t **child;
2292 	uint_t count;
2293 	char *type;
2294 
2295 	rsz = 0;
2296 
2297 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2298 	    &vdev_root) != 0)
2299 		return (EZFS_INVALCONFIG);
2300 
2301 	if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2302 	    nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2303 	    &child, &count) != 0)
2304 		return (EZFS_INVALCONFIG);
2305 
2306 	/*
2307 	 * root pool can not have EFI labeled disks and can only have
2308 	 * a single top-level vdev.
2309 	 */
2310 	if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2311 	    pool_uses_efi(vdev_root))
2312 		return (EZFS_POOL_INVALARG);
2313 
2314 	(void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2315 	    B_FALSE);
2316 
2317 	/* No online devices */
2318 	if (rsz == 0)
2319 		return (EZFS_NODEVICE);
2320 
2321 	return (0);
2322 }
2323 
2324 /*
2325  * Get phys_path for a root pool
2326  * Return 0 on success; non-zero on failure.
2327  */
2328 int
2329 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2330 {
2331 	return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2332 	    phypath_size));
2333 }
2334 
2335 /*
2336  * If the device has being dynamically expanded then we need to relabel
2337  * the disk to use the new unallocated space.
2338  */
2339 static int
2340 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2341 {
2342 	char path[MAXPATHLEN];
2343 	char errbuf[1024];
2344 	int fd, error;
2345 	int (*_efi_use_whole_disk)(int);
2346 
2347 	if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2348 	    "efi_use_whole_disk")) == NULL)
2349 		return (-1);
2350 
2351 	(void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2352 
2353 	if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2354 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2355 		    "relabel '%s': unable to open device"), name);
2356 		return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2357 	}
2358 
2359 	/*
2360 	 * It's possible that we might encounter an error if the device
2361 	 * does not have any unallocated space left. If so, we simply
2362 	 * ignore that error and continue on.
2363 	 */
2364 	error = _efi_use_whole_disk(fd);
2365 	(void) close(fd);
2366 	if (error && error != VT_ENOSPC) {
2367 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2368 		    "relabel '%s': unable to read disk capacity"), name);
2369 		return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2370 	}
2371 	return (0);
2372 }
2373 
2374 /*
2375  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2376  * ZFS_ONLINE_* flags.
2377  */
2378 int
2379 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2380     vdev_state_t *newstate)
2381 {
2382 	zfs_cmd_t zc = { 0 };
2383 	char msg[1024];
2384 	nvlist_t *tgt;
2385 	boolean_t avail_spare, l2cache, islog;
2386 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2387 
2388 	if (flags & ZFS_ONLINE_EXPAND) {
2389 		(void) snprintf(msg, sizeof (msg),
2390 		    dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2391 	} else {
2392 		(void) snprintf(msg, sizeof (msg),
2393 		    dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2394 	}
2395 
2396 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2397 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2398 	    &islog)) == NULL)
2399 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
2400 
2401 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2402 
2403 	if (avail_spare)
2404 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
2405 
2406 	if (flags & ZFS_ONLINE_EXPAND ||
2407 	    zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2408 		char *pathname = NULL;
2409 		uint64_t wholedisk = 0;
2410 
2411 		(void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2412 		    &wholedisk);
2413 		verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2414 		    &pathname) == 0);
2415 
2416 		/*
2417 		 * XXX - L2ARC 1.0 devices can't support expansion.
2418 		 */
2419 		if (l2cache) {
2420 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2421 			    "cannot expand cache devices"));
2422 			return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2423 		}
2424 
2425 		if (wholedisk) {
2426 			pathname += strlen(DISK_ROOT) + 1;
2427 			(void) zpool_relabel_disk(hdl, pathname);
2428 		}
2429 	}
2430 
2431 	zc.zc_cookie = VDEV_STATE_ONLINE;
2432 	zc.zc_obj = flags;
2433 
2434 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2435 		if (errno == EINVAL) {
2436 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2437 			    "from this pool into a new one.  Use '%s' "
2438 			    "instead"), "zpool detach");
2439 			return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2440 		}
2441 		return (zpool_standard_error(hdl, errno, msg));
2442 	}
2443 
2444 	*newstate = zc.zc_cookie;
2445 	return (0);
2446 }
2447 
2448 /*
2449  * Take the specified vdev offline
2450  */
2451 int
2452 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2453 {
2454 	zfs_cmd_t zc = { 0 };
2455 	char msg[1024];
2456 	nvlist_t *tgt;
2457 	boolean_t avail_spare, l2cache;
2458 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2459 
2460 	(void) snprintf(msg, sizeof (msg),
2461 	    dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2462 
2463 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2464 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2465 	    NULL)) == NULL)
2466 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
2467 
2468 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2469 
2470 	if (avail_spare)
2471 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
2472 
2473 	zc.zc_cookie = VDEV_STATE_OFFLINE;
2474 	zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2475 
2476 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2477 		return (0);
2478 
2479 	switch (errno) {
2480 	case EBUSY:
2481 
2482 		/*
2483 		 * There are no other replicas of this device.
2484 		 */
2485 		return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2486 
2487 	case EEXIST:
2488 		/*
2489 		 * The log device has unplayed logs
2490 		 */
2491 		return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2492 
2493 	default:
2494 		return (zpool_standard_error(hdl, errno, msg));
2495 	}
2496 }
2497 
2498 /*
2499  * Mark the given vdev faulted.
2500  */
2501 int
2502 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2503 {
2504 	zfs_cmd_t zc = { 0 };
2505 	char msg[1024];
2506 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2507 
2508 	(void) snprintf(msg, sizeof (msg),
2509 	    dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2510 
2511 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2512 	zc.zc_guid = guid;
2513 	zc.zc_cookie = VDEV_STATE_FAULTED;
2514 	zc.zc_obj = aux;
2515 
2516 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2517 		return (0);
2518 
2519 	switch (errno) {
2520 	case EBUSY:
2521 
2522 		/*
2523 		 * There are no other replicas of this device.
2524 		 */
2525 		return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2526 
2527 	default:
2528 		return (zpool_standard_error(hdl, errno, msg));
2529 	}
2530 
2531 }
2532 
2533 /*
2534  * Mark the given vdev degraded.
2535  */
2536 int
2537 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2538 {
2539 	zfs_cmd_t zc = { 0 };
2540 	char msg[1024];
2541 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2542 
2543 	(void) snprintf(msg, sizeof (msg),
2544 	    dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2545 
2546 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2547 	zc.zc_guid = guid;
2548 	zc.zc_cookie = VDEV_STATE_DEGRADED;
2549 	zc.zc_obj = aux;
2550 
2551 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2552 		return (0);
2553 
2554 	return (zpool_standard_error(hdl, errno, msg));
2555 }
2556 
2557 /*
2558  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2559  * a hot spare.
2560  */
2561 static boolean_t
2562 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2563 {
2564 	nvlist_t **child;
2565 	uint_t c, children;
2566 	char *type;
2567 
2568 	if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2569 	    &children) == 0) {
2570 		verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2571 		    &type) == 0);
2572 
2573 		if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2574 		    children == 2 && child[which] == tgt)
2575 			return (B_TRUE);
2576 
2577 		for (c = 0; c < children; c++)
2578 			if (is_replacing_spare(child[c], tgt, which))
2579 				return (B_TRUE);
2580 	}
2581 
2582 	return (B_FALSE);
2583 }
2584 
2585 /*
2586  * Attach new_disk (fully described by nvroot) to old_disk.
2587  * If 'replacing' is specified, the new disk will replace the old one.
2588  */
2589 int
2590 zpool_vdev_attach(zpool_handle_t *zhp,
2591     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2592 {
2593 	zfs_cmd_t zc = { 0 };
2594 	char msg[1024];
2595 	int ret;
2596 	nvlist_t *tgt;
2597 	boolean_t avail_spare, l2cache, islog;
2598 	uint64_t val;
2599 	char *newname;
2600 	nvlist_t **child;
2601 	uint_t children;
2602 	nvlist_t *config_root;
2603 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2604 	boolean_t rootpool = zpool_is_bootable(zhp);
2605 
2606 	if (replacing)
2607 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2608 		    "cannot replace %s with %s"), old_disk, new_disk);
2609 	else
2610 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2611 		    "cannot attach %s to %s"), new_disk, old_disk);
2612 
2613 	/*
2614 	 * If this is a root pool, make sure that we're not attaching an
2615 	 * EFI labeled device.
2616 	 */
2617 	if (rootpool && pool_uses_efi(nvroot)) {
2618 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2619 		    "EFI labeled devices are not supported on root pools."));
2620 		return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2621 	}
2622 
2623 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2624 	if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2625 	    &islog)) == 0)
2626 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
2627 
2628 	if (avail_spare)
2629 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
2630 
2631 	if (l2cache)
2632 		return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2633 
2634 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2635 	zc.zc_cookie = replacing;
2636 
2637 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2638 	    &child, &children) != 0 || children != 1) {
2639 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2640 		    "new device must be a single disk"));
2641 		return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2642 	}
2643 
2644 	verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2645 	    ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2646 
2647 	if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2648 		return (-1);
2649 
2650 	/*
2651 	 * If the target is a hot spare that has been swapped in, we can only
2652 	 * replace it with another hot spare.
2653 	 */
2654 	if (replacing &&
2655 	    nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2656 	    (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2657 	    NULL) == NULL || !avail_spare) &&
2658 	    is_replacing_spare(config_root, tgt, 1)) {
2659 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2660 		    "can only be replaced by another hot spare"));
2661 		free(newname);
2662 		return (zfs_error(hdl, EZFS_BADTARGET, msg));
2663 	}
2664 
2665 	free(newname);
2666 
2667 	if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2668 		return (-1);
2669 
2670 	ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2671 
2672 	zcmd_free_nvlists(&zc);
2673 
2674 	if (ret == 0) {
2675 		if (rootpool) {
2676 			/*
2677 			 * XXX need a better way to prevent user from
2678 			 * booting up a half-baked vdev.
2679 			 */
2680 			(void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2681 			    "sure to wait until resilver is done "
2682 			    "before rebooting.\n"));
2683 		}
2684 		return (0);
2685 	}
2686 
2687 	switch (errno) {
2688 	case ENOTSUP:
2689 		/*
2690 		 * Can't attach to or replace this type of vdev.
2691 		 */
2692 		if (replacing) {
2693 			uint64_t version = zpool_get_prop_int(zhp,
2694 			    ZPOOL_PROP_VERSION, NULL);
2695 
2696 			if (islog)
2697 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2698 				    "cannot replace a log with a spare"));
2699 			else if (version >= SPA_VERSION_MULTI_REPLACE)
2700 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2701 				    "already in replacing/spare config; wait "
2702 				    "for completion or use 'zpool detach'"));
2703 			else
2704 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2705 				    "cannot replace a replacing device"));
2706 		} else {
2707 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2708 			    "can only attach to mirrors and top-level "
2709 			    "disks"));
2710 		}
2711 		(void) zfs_error(hdl, EZFS_BADTARGET, msg);
2712 		break;
2713 
2714 	case EINVAL:
2715 		/*
2716 		 * The new device must be a single disk.
2717 		 */
2718 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2719 		    "new device must be a single disk"));
2720 		(void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2721 		break;
2722 
2723 	case EBUSY:
2724 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2725 		    new_disk);
2726 		(void) zfs_error(hdl, EZFS_BADDEV, msg);
2727 		break;
2728 
2729 	case EOVERFLOW:
2730 		/*
2731 		 * The new device is too small.
2732 		 */
2733 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2734 		    "device is too small"));
2735 		(void) zfs_error(hdl, EZFS_BADDEV, msg);
2736 		break;
2737 
2738 	case EDOM:
2739 		/*
2740 		 * The new device has a different alignment requirement.
2741 		 */
2742 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2743 		    "devices have different sector alignment"));
2744 		(void) zfs_error(hdl, EZFS_BADDEV, msg);
2745 		break;
2746 
2747 	case ENAMETOOLONG:
2748 		/*
2749 		 * The resulting top-level vdev spec won't fit in the label.
2750 		 */
2751 		(void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2752 		break;
2753 
2754 	default:
2755 		(void) zpool_standard_error(hdl, errno, msg);
2756 	}
2757 
2758 	return (-1);
2759 }
2760 
2761 /*
2762  * Detach the specified device.
2763  */
2764 int
2765 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2766 {
2767 	zfs_cmd_t zc = { 0 };
2768 	char msg[1024];
2769 	nvlist_t *tgt;
2770 	boolean_t avail_spare, l2cache;
2771 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2772 
2773 	(void) snprintf(msg, sizeof (msg),
2774 	    dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2775 
2776 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2777 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2778 	    NULL)) == 0)
2779 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
2780 
2781 	if (avail_spare)
2782 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
2783 
2784 	if (l2cache)
2785 		return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2786 
2787 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2788 
2789 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2790 		return (0);
2791 
2792 	switch (errno) {
2793 
2794 	case ENOTSUP:
2795 		/*
2796 		 * Can't detach from this type of vdev.
2797 		 */
2798 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2799 		    "applicable to mirror and replacing vdevs"));
2800 		(void) zfs_error(hdl, EZFS_BADTARGET, msg);
2801 		break;
2802 
2803 	case EBUSY:
2804 		/*
2805 		 * There are no other replicas of this device.
2806 		 */
2807 		(void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2808 		break;
2809 
2810 	default:
2811 		(void) zpool_standard_error(hdl, errno, msg);
2812 	}
2813 
2814 	return (-1);
2815 }
2816 
2817 /*
2818  * Find a mirror vdev in the source nvlist.
2819  *
2820  * The mchild array contains a list of disks in one of the top-level mirrors
2821  * of the source pool.  The schild array contains a list of disks that the
2822  * user specified on the command line.  We loop over the mchild array to
2823  * see if any entry in the schild array matches.
2824  *
2825  * If a disk in the mchild array is found in the schild array, we return
2826  * the index of that entry.  Otherwise we return -1.
2827  */
2828 static int
2829 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2830     nvlist_t **schild, uint_t schildren)
2831 {
2832 	uint_t mc;
2833 
2834 	for (mc = 0; mc < mchildren; mc++) {
2835 		uint_t sc;
2836 		char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2837 		    mchild[mc], B_FALSE);
2838 
2839 		for (sc = 0; sc < schildren; sc++) {
2840 			char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2841 			    schild[sc], B_FALSE);
2842 			boolean_t result = (strcmp(mpath, spath) == 0);
2843 
2844 			free(spath);
2845 			if (result) {
2846 				free(mpath);
2847 				return (mc);
2848 			}
2849 		}
2850 
2851 		free(mpath);
2852 	}
2853 
2854 	return (-1);
2855 }
2856 
2857 /*
2858  * Split a mirror pool.  If newroot points to null, then a new nvlist
2859  * is generated and it is the responsibility of the caller to free it.
2860  */
2861 int
2862 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2863     nvlist_t *props, splitflags_t flags)
2864 {
2865 	zfs_cmd_t zc = { 0 };
2866 	char msg[1024];
2867 	nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2868 	nvlist_t **varray = NULL, *zc_props = NULL;
2869 	uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2870 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2871 	uint64_t vers;
2872 	boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2873 	int retval = 0;
2874 
2875 	(void) snprintf(msg, sizeof (msg),
2876 	    dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2877 
2878 	if (!zpool_name_valid(hdl, B_FALSE, newname))
2879 		return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2880 
2881 	if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2882 		(void) fprintf(stderr, gettext("Internal error: unable to "
2883 		    "retrieve pool configuration\n"));
2884 		return (-1);
2885 	}
2886 
2887 	verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2888 	    == 0);
2889 	verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2890 
2891 	if (props) {
2892 		prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2893 		if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2894 		    props, vers, flags, msg)) == NULL)
2895 			return (-1);
2896 	}
2897 
2898 	if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2899 	    &children) != 0) {
2900 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2901 		    "Source pool is missing vdev tree"));
2902 		if (zc_props)
2903 			nvlist_free(zc_props);
2904 		return (-1);
2905 	}
2906 
2907 	varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2908 	vcount = 0;
2909 
2910 	if (*newroot == NULL ||
2911 	    nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2912 	    &newchild, &newchildren) != 0)
2913 		newchildren = 0;
2914 
2915 	for (c = 0; c < children; c++) {
2916 		uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2917 		char *type;
2918 		nvlist_t **mchild, *vdev;
2919 		uint_t mchildren;
2920 		int entry;
2921 
2922 		/*
2923 		 * Unlike cache & spares, slogs are stored in the
2924 		 * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2925 		 */
2926 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2927 		    &is_log);
2928 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2929 		    &is_hole);
2930 		if (is_log || is_hole) {
2931 			/*
2932 			 * Create a hole vdev and put it in the config.
2933 			 */
2934 			if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2935 				goto out;
2936 			if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2937 			    VDEV_TYPE_HOLE) != 0)
2938 				goto out;
2939 			if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2940 			    1) != 0)
2941 				goto out;
2942 			if (lastlog == 0)
2943 				lastlog = vcount;
2944 			varray[vcount++] = vdev;
2945 			continue;
2946 		}
2947 		lastlog = 0;
2948 		verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2949 		    == 0);
2950 		if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2951 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2952 			    "Source pool must be composed only of mirrors\n"));
2953 			retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2954 			goto out;
2955 		}
2956 
2957 		verify(nvlist_lookup_nvlist_array(child[c],
2958 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2959 
2960 		/* find or add an entry for this top-level vdev */
2961 		if (newchildren > 0 &&
2962 		    (entry = find_vdev_entry(zhp, mchild, mchildren,
2963 		    newchild, newchildren)) >= 0) {
2964 			/* We found a disk that the user specified. */
2965 			vdev = mchild[entry];
2966 			++found;
2967 		} else {
2968 			/* User didn't specify a disk for this vdev. */
2969 			vdev = mchild[mchildren - 1];
2970 		}
2971 
2972 		if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2973 			goto out;
2974 	}
2975 
2976 	/* did we find every disk the user specified? */
2977 	if (found != newchildren) {
2978 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
2979 		    "include at most one disk from each mirror"));
2980 		retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2981 		goto out;
2982 	}
2983 
2984 	/* Prepare the nvlist for populating. */
2985 	if (*newroot == NULL) {
2986 		if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
2987 			goto out;
2988 		freelist = B_TRUE;
2989 		if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
2990 		    VDEV_TYPE_ROOT) != 0)
2991 			goto out;
2992 	} else {
2993 		verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
2994 	}
2995 
2996 	/* Add all the children we found */
2997 	if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
2998 	    lastlog == 0 ? vcount : lastlog) != 0)
2999 		goto out;
3000 
3001 	/*
3002 	 * If we're just doing a dry run, exit now with success.
3003 	 */
3004 	if (flags.dryrun) {
3005 		memory_err = B_FALSE;
3006 		freelist = B_FALSE;
3007 		goto out;
3008 	}
3009 
3010 	/* now build up the config list & call the ioctl */
3011 	if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3012 		goto out;
3013 
3014 	if (nvlist_add_nvlist(newconfig,
3015 	    ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3016 	    nvlist_add_string(newconfig,
3017 	    ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3018 	    nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3019 		goto out;
3020 
3021 	/*
3022 	 * The new pool is automatically part of the namespace unless we
3023 	 * explicitly export it.
3024 	 */
3025 	if (!flags.import)
3026 		zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3027 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3028 	(void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3029 	if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3030 		goto out;
3031 	if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3032 		goto out;
3033 
3034 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3035 		retval = zpool_standard_error(hdl, errno, msg);
3036 		goto out;
3037 	}
3038 
3039 	freelist = B_FALSE;
3040 	memory_err = B_FALSE;
3041 
3042 out:
3043 	if (varray != NULL) {
3044 		int v;
3045 
3046 		for (v = 0; v < vcount; v++)
3047 			nvlist_free(varray[v]);
3048 		free(varray);
3049 	}
3050 	zcmd_free_nvlists(&zc);
3051 	if (zc_props)
3052 		nvlist_free(zc_props);
3053 	if (newconfig)
3054 		nvlist_free(newconfig);
3055 	if (freelist) {
3056 		nvlist_free(*newroot);
3057 		*newroot = NULL;
3058 	}
3059 
3060 	if (retval != 0)
3061 		return (retval);
3062 
3063 	if (memory_err)
3064 		return (no_memory(hdl));
3065 
3066 	return (0);
3067 }
3068 
3069 /*
3070  * Remove the given device.  Currently, this is supported only for hot spares
3071  * and level 2 cache devices.
3072  */
3073 int
3074 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3075 {
3076 	zfs_cmd_t zc = { 0 };
3077 	char msg[1024];
3078 	nvlist_t *tgt;
3079 	boolean_t avail_spare, l2cache, islog;
3080 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3081 	uint64_t version;
3082 
3083 	(void) snprintf(msg, sizeof (msg),
3084 	    dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3085 
3086 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3087 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3088 	    &islog)) == 0)
3089 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3090 	/*
3091 	 * XXX - this should just go away.
3092 	 */
3093 	if (!avail_spare && !l2cache && !islog) {
3094 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3095 		    "only inactive hot spares, cache, top-level, "
3096 		    "or log devices can be removed"));
3097 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3098 	}
3099 
3100 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3101 	if (islog && version < SPA_VERSION_HOLES) {
3102 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3103 		    "pool must be upgrade to support log removal"));
3104 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
3105 	}
3106 
3107 	verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3108 
3109 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3110 		return (0);
3111 
3112 	return (zpool_standard_error(hdl, errno, msg));
3113 }
3114 
3115 /*
3116  * Clear the errors for the pool, or the particular device if specified.
3117  */
3118 int
3119 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3120 {
3121 	zfs_cmd_t zc = { 0 };
3122 	char msg[1024];
3123 	nvlist_t *tgt;
3124 	zpool_rewind_policy_t policy;
3125 	boolean_t avail_spare, l2cache;
3126 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3127 	nvlist_t *nvi = NULL;
3128 	int error;
3129 
3130 	if (path)
3131 		(void) snprintf(msg, sizeof (msg),
3132 		    dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3133 		    path);
3134 	else
3135 		(void) snprintf(msg, sizeof (msg),
3136 		    dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3137 		    zhp->zpool_name);
3138 
3139 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3140 	if (path) {
3141 		if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3142 		    &l2cache, NULL)) == 0)
3143 			return (zfs_error(hdl, EZFS_NODEVICE, msg));
3144 
3145 		/*
3146 		 * Don't allow error clearing for hot spares.  Do allow
3147 		 * error clearing for l2cache devices.
3148 		 */
3149 		if (avail_spare)
3150 			return (zfs_error(hdl, EZFS_ISSPARE, msg));
3151 
3152 		verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3153 		    &zc.zc_guid) == 0);
3154 	}
3155 
3156 	zpool_get_rewind_policy(rewindnvl, &policy);
3157 	zc.zc_cookie = policy.zrp_request;
3158 
3159 	if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3160 		return (-1);
3161 
3162 	if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3163 		return (-1);
3164 
3165 	while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3166 	    errno == ENOMEM) {
3167 		if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3168 			zcmd_free_nvlists(&zc);
3169 			return (-1);
3170 		}
3171 	}
3172 
3173 	if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3174 	    errno != EPERM && errno != EACCES)) {
3175 		if (policy.zrp_request &
3176 		    (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3177 			(void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3178 			zpool_rewind_exclaim(hdl, zc.zc_name,
3179 			    ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3180 			    nvi);
3181 			nvlist_free(nvi);
3182 		}
3183 		zcmd_free_nvlists(&zc);
3184 		return (0);
3185 	}
3186 
3187 	zcmd_free_nvlists(&zc);
3188 	return (zpool_standard_error(hdl, errno, msg));
3189 }
3190 
3191 /*
3192  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3193  */
3194 int
3195 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3196 {
3197 	zfs_cmd_t zc = { 0 };
3198 	char msg[1024];
3199 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3200 
3201 	(void) snprintf(msg, sizeof (msg),
3202 	    dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3203 	    guid);
3204 
3205 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3206 	zc.zc_guid = guid;
3207 	zc.zc_cookie = ZPOOL_NO_REWIND;
3208 
3209 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3210 		return (0);
3211 
3212 	return (zpool_standard_error(hdl, errno, msg));
3213 }
3214 
3215 /*
3216  * Change the GUID for a pool.
3217  */
3218 int
3219 zpool_reguid(zpool_handle_t *zhp)
3220 {
3221 	char msg[1024];
3222 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3223 	zfs_cmd_t zc = { 0 };
3224 
3225 	(void) snprintf(msg, sizeof (msg),
3226 	    dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3227 
3228 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3229 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3230 		return (0);
3231 
3232 	return (zpool_standard_error(hdl, errno, msg));
3233 }
3234 
3235 /*
3236  * Reopen the pool.
3237  */
3238 int
3239 zpool_reopen(zpool_handle_t *zhp)
3240 {
3241 	zfs_cmd_t zc = { 0 };
3242 	char msg[1024];
3243 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3244 
3245 	(void) snprintf(msg, sizeof (msg),
3246 	    dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3247 	    zhp->zpool_name);
3248 
3249 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3250 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3251 		return (0);
3252 	return (zpool_standard_error(hdl, errno, msg));
3253 }
3254 
3255 /*
3256  * Convert from a devid string to a path.
3257  */
3258 static char *
3259 devid_to_path(char *devid_str)
3260 {
3261 	ddi_devid_t devid;
3262 	char *minor;
3263 	char *path;
3264 	devid_nmlist_t *list = NULL;
3265 	int ret;
3266 
3267 	if (devid_str_decode(devid_str, &devid, &minor) != 0)
3268 		return (NULL);
3269 
3270 	ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3271 
3272 	devid_str_free(minor);
3273 	devid_free(devid);
3274 
3275 	if (ret != 0)
3276 		return (NULL);
3277 
3278 	if ((path = strdup(list[0].devname)) == NULL)
3279 		return (NULL);
3280 
3281 	devid_free_nmlist(list);
3282 
3283 	return (path);
3284 }
3285 
3286 /*
3287  * Convert from a path to a devid string.
3288  */
3289 static char *
3290 path_to_devid(const char *path)
3291 {
3292 	int fd;
3293 	ddi_devid_t devid;
3294 	char *minor, *ret;
3295 
3296 	if ((fd = open(path, O_RDONLY)) < 0)
3297 		return (NULL);
3298 
3299 	minor = NULL;
3300 	ret = NULL;
3301 	if (devid_get(fd, &devid) == 0) {
3302 		if (devid_get_minor_name(fd, &minor) == 0)
3303 			ret = devid_str_encode(devid, minor);
3304 		if (minor != NULL)
3305 			devid_str_free(minor);
3306 		devid_free(devid);
3307 	}
3308 	(void) close(fd);
3309 
3310 	return (ret);
3311 }
3312 
3313 /*
3314  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3315  * ignore any failure here, since a common case is for an unprivileged user to
3316  * type 'zpool status', and we'll display the correct information anyway.
3317  */
3318 static void
3319 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3320 {
3321 	zfs_cmd_t zc = { 0 };
3322 
3323 	(void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3324 	(void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3325 	verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3326 	    &zc.zc_guid) == 0);
3327 
3328 	(void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3329 }
3330 
3331 /*
3332  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3333  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3334  * We also check if this is a whole disk, in which case we strip off the
3335  * trailing 's0' slice name.
3336  *
3337  * This routine is also responsible for identifying when disks have been
3338  * reconfigured in a new location.  The kernel will have opened the device by
3339  * devid, but the path will still refer to the old location.  To catch this, we
3340  * first do a path -> devid translation (which is fast for the common case).  If
3341  * the devid matches, we're done.  If not, we do a reverse devid -> path
3342  * translation and issue the appropriate ioctl() to update the path of the vdev.
3343  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3344  * of these checks.
3345  */
3346 char *
3347 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3348     boolean_t verbose)
3349 {
3350 	char *path, *devid;
3351 	uint64_t value;
3352 	char buf[64];
3353 	vdev_stat_t *vs;
3354 	uint_t vsc;
3355 
3356 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
3357 	    &value) == 0) {
3358 		verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3359 		    &value) == 0);
3360 		(void) snprintf(buf, sizeof (buf), "%llu",
3361 		    (u_longlong_t)value);
3362 		path = buf;
3363 	} else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
3364 
3365 		/*
3366 		 * If the device is dead (faulted, offline, etc) then don't
3367 		 * bother opening it.  Otherwise we may be forcing the user to
3368 		 * open a misbehaving device, which can have undesirable
3369 		 * effects.
3370 		 */
3371 		if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3372 		    (uint64_t **)&vs, &vsc) != 0 ||
3373 		    vs->vs_state >= VDEV_STATE_DEGRADED) &&
3374 		    zhp != NULL &&
3375 		    nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3376 			/*
3377 			 * Determine if the current path is correct.
3378 			 */
3379 			char *newdevid = path_to_devid(path);
3380 
3381 			if (newdevid == NULL ||
3382 			    strcmp(devid, newdevid) != 0) {
3383 				char *newpath;
3384 
3385 				if ((newpath = devid_to_path(devid)) != NULL) {
3386 					/*
3387 					 * Update the path appropriately.
3388 					 */
3389 					set_path(zhp, nv, newpath);
3390 					if (nvlist_add_string(nv,
3391 					    ZPOOL_CONFIG_PATH, newpath) == 0)
3392 						verify(nvlist_lookup_string(nv,
3393 						    ZPOOL_CONFIG_PATH,
3394 						    &path) == 0);
3395 					free(newpath);
3396 				}
3397 			}
3398 
3399 			if (newdevid)
3400 				devid_str_free(newdevid);
3401 		}
3402 
3403 		if (strncmp(path, "/dev/dsk/", 9) == 0)
3404 			path += 9;
3405 
3406 		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3407 		    &value) == 0 && value) {
3408 			int pathlen = strlen(path);
3409 			char *tmp = zfs_strdup(hdl, path);
3410 
3411 			/*
3412 			 * If it starts with c#, and ends with "s0", chop
3413 			 * the "s0" off, or if it ends with "s0/old", remove
3414 			 * the "s0" from the middle.
3415 			 */
3416 			if (CTD_CHECK(tmp)) {
3417 				if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3418 					tmp[pathlen - 2] = '\0';
3419 				} else if (pathlen > 6 &&
3420 				    strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3421 					(void) strcpy(&tmp[pathlen - 6],
3422 					    "/old");
3423 				}
3424 			}
3425 			return (tmp);
3426 		}
3427 	} else {
3428 		verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3429 
3430 		/*
3431 		 * If it's a raidz device, we need to stick in the parity level.
3432 		 */
3433 		if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3434 			verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3435 			    &value) == 0);
3436 			(void) snprintf(buf, sizeof (buf), "%s%llu", path,
3437 			    (u_longlong_t)value);
3438 			path = buf;
3439 		}
3440 
3441 		/*
3442 		 * We identify each top-level vdev by using a <type-id>
3443 		 * naming convention.
3444 		 */
3445 		if (verbose) {
3446 			uint64_t id;
3447 
3448 			verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3449 			    &id) == 0);
3450 			(void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3451 			    (u_longlong_t)id);
3452 			path = buf;
3453 		}
3454 	}
3455 
3456 	return (zfs_strdup(hdl, path));
3457 }
3458 
3459 static int
3460 zbookmark_compare(const void *a, const void *b)
3461 {
3462 	return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3463 }
3464 
3465 /*
3466  * Retrieve the persistent error log, uniquify the members, and return to the
3467  * caller.
3468  */
3469 int
3470 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3471 {
3472 	zfs_cmd_t zc = { 0 };
3473 	uint64_t count;
3474 	zbookmark_phys_t *zb = NULL;
3475 	int i;
3476 
3477 	/*
3478 	 * Retrieve the raw error list from the kernel.  If the number of errors
3479 	 * has increased, allocate more space and continue until we get the
3480 	 * entire list.
3481 	 */
3482 	verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3483 	    &count) == 0);
3484 	if (count == 0)
3485 		return (0);
3486 	if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3487 	    count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3488 		return (-1);
3489 	zc.zc_nvlist_dst_size = count;
3490 	(void) strcpy(zc.zc_name, zhp->zpool_name);
3491 	for (;;) {
3492 		if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3493 		    &zc) != 0) {
3494 			free((void *)(uintptr_t)zc.zc_nvlist_dst);
3495 			if (errno == ENOMEM) {
3496 				void *dst;
3497 
3498 				count = zc.zc_nvlist_dst_size;
3499 				dst = zfs_alloc(zhp->zpool_hdl, count *
3500 				    sizeof (zbookmark_phys_t));
3501 				if (dst == NULL)
3502 					return (-1);
3503 				zc.zc_nvlist_dst = (uintptr_t)dst;
3504 			} else {
3505 				return (-1);
3506 			}
3507 		} else {
3508 			break;
3509 		}
3510 	}
3511 
3512 	/*
3513 	 * Sort the resulting bookmarks.  This is a little confusing due to the
3514 	 * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3515 	 * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3516 	 * _not_ copied as part of the process.  So we point the start of our
3517 	 * array appropriate and decrement the total number of elements.
3518 	 */
3519 	zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3520 	    zc.zc_nvlist_dst_size;
3521 	count -= zc.zc_nvlist_dst_size;
3522 
3523 	qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_compare);
3524 
3525 	verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3526 
3527 	/*
3528 	 * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3529 	 */
3530 	for (i = 0; i < count; i++) {
3531 		nvlist_t *nv;
3532 
3533 		/* ignoring zb_blkid and zb_level for now */
3534 		if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3535 		    zb[i-1].zb_object == zb[i].zb_object)
3536 			continue;
3537 
3538 		if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3539 			goto nomem;
3540 		if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3541 		    zb[i].zb_objset) != 0) {
3542 			nvlist_free(nv);
3543 			goto nomem;
3544 		}
3545 		if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3546 		    zb[i].zb_object) != 0) {
3547 			nvlist_free(nv);
3548 			goto nomem;
3549 		}
3550 		if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3551 			nvlist_free(nv);
3552 			goto nomem;
3553 		}
3554 		nvlist_free(nv);
3555 	}
3556 
3557 	free((void *)(uintptr_t)zc.zc_nvlist_dst);
3558 	return (0);
3559 
3560 nomem:
3561 	free((void *)(uintptr_t)zc.zc_nvlist_dst);
3562 	return (no_memory(zhp->zpool_hdl));
3563 }
3564 
3565 /*
3566  * Upgrade a ZFS pool to the latest on-disk version.
3567  */
3568 int
3569 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3570 {
3571 	zfs_cmd_t zc = { 0 };
3572 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3573 
3574 	(void) strcpy(zc.zc_name, zhp->zpool_name);
3575 	zc.zc_cookie = new_version;
3576 
3577 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3578 		return (zpool_standard_error_fmt(hdl, errno,
3579 		    dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3580 		    zhp->zpool_name));
3581 	return (0);
3582 }
3583 
3584 void
3585 zfs_save_arguments(int argc, char **argv, char *string, int len)
3586 {
3587 	(void) strlcpy(string, basename(argv[0]), len);
3588 	for (int i = 1; i < argc; i++) {
3589 		(void) strlcat(string, " ", len);
3590 		(void) strlcat(string, argv[i], len);
3591 	}
3592 }
3593 
3594 int
3595 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3596 {
3597 	zfs_cmd_t zc = { 0 };
3598 	nvlist_t *args;
3599 	int err;
3600 
3601 	args = fnvlist_alloc();
3602 	fnvlist_add_string(args, "message", message);
3603 	err = zcmd_write_src_nvlist(hdl, &zc, args);
3604 	if (err == 0)
3605 		err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3606 	nvlist_free(args);
3607 	zcmd_free_nvlists(&zc);
3608 	return (err);
3609 }
3610 
3611 /*
3612  * Perform ioctl to get some command history of a pool.
3613  *
3614  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3615  * logical offset of the history buffer to start reading from.
3616  *
3617  * Upon return, 'off' is the next logical offset to read from and
3618  * 'len' is the actual amount of bytes read into 'buf'.
3619  */
3620 static int
3621 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3622 {
3623 	zfs_cmd_t zc = { 0 };
3624 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3625 
3626 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3627 
3628 	zc.zc_history = (uint64_t)(uintptr_t)buf;
3629 	zc.zc_history_len = *len;
3630 	zc.zc_history_offset = *off;
3631 
3632 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3633 		switch (errno) {
3634 		case EPERM:
3635 			return (zfs_error_fmt(hdl, EZFS_PERM,
3636 			    dgettext(TEXT_DOMAIN,
3637 			    "cannot show history for pool '%s'"),
3638 			    zhp->zpool_name));
3639 		case ENOENT:
3640 			return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3641 			    dgettext(TEXT_DOMAIN, "cannot get history for pool "
3642 			    "'%s'"), zhp->zpool_name));
3643 		case ENOTSUP:
3644 			return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3645 			    dgettext(TEXT_DOMAIN, "cannot get history for pool "
3646 			    "'%s', pool must be upgraded"), zhp->zpool_name));
3647 		default:
3648 			return (zpool_standard_error_fmt(hdl, errno,
3649 			    dgettext(TEXT_DOMAIN,
3650 			    "cannot get history for '%s'"), zhp->zpool_name));
3651 		}
3652 	}
3653 
3654 	*len = zc.zc_history_len;
3655 	*off = zc.zc_history_offset;
3656 
3657 	return (0);
3658 }
3659 
3660 /*
3661  * Process the buffer of nvlists, unpacking and storing each nvlist record
3662  * into 'records'.  'leftover' is set to the number of bytes that weren't
3663  * processed as there wasn't a complete record.
3664  */
3665 int
3666 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3667     nvlist_t ***records, uint_t *numrecords)
3668 {
3669 	uint64_t reclen;
3670 	nvlist_t *nv;
3671 	int i;
3672 
3673 	while (bytes_read > sizeof (reclen)) {
3674 
3675 		/* get length of packed record (stored as little endian) */
3676 		for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3677 			reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3678 
3679 		if (bytes_read < sizeof (reclen) + reclen)
3680 			break;
3681 
3682 		/* unpack record */
3683 		if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3684 			return (ENOMEM);
3685 		bytes_read -= sizeof (reclen) + reclen;
3686 		buf += sizeof (reclen) + reclen;
3687 
3688 		/* add record to nvlist array */
3689 		(*numrecords)++;
3690 		if (ISP2(*numrecords + 1)) {
3691 			*records = realloc(*records,
3692 			    *numrecords * 2 * sizeof (nvlist_t *));
3693 		}
3694 		(*records)[*numrecords - 1] = nv;
3695 	}
3696 
3697 	*leftover = bytes_read;
3698 	return (0);
3699 }
3700 
3701 /*
3702  * Retrieve the command history of a pool.
3703  */
3704 int
3705 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3706 {
3707 	char *buf;
3708 	int buflen = 128 * 1024;
3709 	uint64_t off = 0;
3710 	nvlist_t **records = NULL;
3711 	uint_t numrecords = 0;
3712 	int err, i;
3713 
3714 	buf = malloc(buflen);
3715 	if (buf == NULL)
3716 		return (ENOMEM);
3717 	do {
3718 		uint64_t bytes_read = buflen;
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 		if (leftover == bytes_read) {
3733 			/*
3734 			 * no progress made, because buffer is not big enough
3735 			 * to hold this record; resize and retry.
3736 			 */
3737 			buflen *= 2;
3738 			free(buf);
3739 			buf = malloc(buflen);
3740 			if (buf == NULL)
3741 				return (ENOMEM);
3742 		}
3743 
3744 		/* CONSTCOND */
3745 	} while (1);
3746 
3747 	free(buf);
3748 
3749 	if (!err) {
3750 		verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3751 		verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3752 		    records, numrecords) == 0);
3753 	}
3754 	for (i = 0; i < numrecords; i++)
3755 		nvlist_free(records[i]);
3756 	free(records);
3757 
3758 	return (err);
3759 }
3760 
3761 void
3762 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3763     char *pathname, size_t len)
3764 {
3765 	zfs_cmd_t zc = { 0 };
3766 	boolean_t mounted = B_FALSE;
3767 	char *mntpnt = NULL;
3768 	char dsname[MAXNAMELEN];
3769 
3770 	if (dsobj == 0) {
3771 		/* special case for the MOS */
3772 		(void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3773 		return;
3774 	}
3775 
3776 	/* get the dataset's name */
3777 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3778 	zc.zc_obj = dsobj;
3779 	if (ioctl(zhp->zpool_hdl->libzfs_fd,
3780 	    ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3781 		/* just write out a path of two object numbers */
3782 		(void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3783 		    dsobj, obj);
3784 		return;
3785 	}
3786 	(void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3787 
3788 	/* find out if the dataset is mounted */
3789 	mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3790 
3791 	/* get the corrupted object's path */
3792 	(void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3793 	zc.zc_obj = obj;
3794 	if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3795 	    &zc) == 0) {
3796 		if (mounted) {
3797 			(void) snprintf(pathname, len, "%s%s", mntpnt,
3798 			    zc.zc_value);
3799 		} else {
3800 			(void) snprintf(pathname, len, "%s:%s",
3801 			    dsname, zc.zc_value);
3802 		}
3803 	} else {
3804 		(void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3805 	}
3806 	free(mntpnt);
3807 }
3808 
3809 /*
3810  * Read the EFI label from the config, if a label does not exist then
3811  * pass back the error to the caller. If the caller has passed a non-NULL
3812  * diskaddr argument then we set it to the starting address of the EFI
3813  * partition.
3814  */
3815 static int
3816 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3817 {
3818 	char *path;
3819 	int fd;
3820 	char diskname[MAXPATHLEN];
3821 	int err = -1;
3822 
3823 	if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3824 		return (err);
3825 
3826 	(void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3827 	    strrchr(path, '/'));
3828 	if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3829 		struct dk_gpt *vtoc;
3830 
3831 		if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3832 			if (sb != NULL)
3833 				*sb = vtoc->efi_parts[0].p_start;
3834 			efi_free(vtoc);
3835 		}
3836 		(void) close(fd);
3837 	}
3838 	return (err);
3839 }
3840 
3841 /*
3842  * determine where a partition starts on a disk in the current
3843  * configuration
3844  */
3845 static diskaddr_t
3846 find_start_block(nvlist_t *config)
3847 {
3848 	nvlist_t **child;
3849 	uint_t c, children;
3850 	diskaddr_t sb = MAXOFFSET_T;
3851 	uint64_t wholedisk;
3852 
3853 	if (nvlist_lookup_nvlist_array(config,
3854 	    ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3855 		if (nvlist_lookup_uint64(config,
3856 		    ZPOOL_CONFIG_WHOLE_DISK,
3857 		    &wholedisk) != 0 || !wholedisk) {
3858 			return (MAXOFFSET_T);
3859 		}
3860 		if (read_efi_label(config, &sb) < 0)
3861 			sb = MAXOFFSET_T;
3862 		return (sb);
3863 	}
3864 
3865 	for (c = 0; c < children; c++) {
3866 		sb = find_start_block(child[c]);
3867 		if (sb != MAXOFFSET_T) {
3868 			return (sb);
3869 		}
3870 	}
3871 	return (MAXOFFSET_T);
3872 }
3873 
3874 /*
3875  * Label an individual disk.  The name provided is the short name,
3876  * stripped of any leading /dev path.
3877  */
3878 int
3879 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
3880 {
3881 	char path[MAXPATHLEN];
3882 	struct dk_gpt *vtoc;
3883 	int fd;
3884 	size_t resv = EFI_MIN_RESV_SIZE;
3885 	uint64_t slice_size;
3886 	diskaddr_t start_block;
3887 	char errbuf[1024];
3888 
3889 	/* prepare an error message just in case */
3890 	(void) snprintf(errbuf, sizeof (errbuf),
3891 	    dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3892 
3893 	if (zhp) {
3894 		nvlist_t *nvroot;
3895 
3896 		if (zpool_is_bootable(zhp)) {
3897 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3898 			    "EFI labeled devices are not supported on root "
3899 			    "pools."));
3900 			return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3901 		}
3902 
3903 		verify(nvlist_lookup_nvlist(zhp->zpool_config,
3904 		    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3905 
3906 		if (zhp->zpool_start_block == 0)
3907 			start_block = find_start_block(nvroot);
3908 		else
3909 			start_block = zhp->zpool_start_block;
3910 		zhp->zpool_start_block = start_block;
3911 	} else {
3912 		/* new pool */
3913 		start_block = NEW_START_BLOCK;
3914 	}
3915 
3916 	(void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3917 	    BACKUP_SLICE);
3918 
3919 	if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3920 		/*
3921 		 * This shouldn't happen.  We've long since verified that this
3922 		 * is a valid device.
3923 		 */
3924 		zfs_error_aux(hdl,
3925 		    dgettext(TEXT_DOMAIN, "unable to open device"));
3926 		return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3927 	}
3928 
3929 	if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3930 		/*
3931 		 * The only way this can fail is if we run out of memory, or we
3932 		 * were unable to read the disk's capacity
3933 		 */
3934 		if (errno == ENOMEM)
3935 			(void) no_memory(hdl);
3936 
3937 		(void) close(fd);
3938 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3939 		    "unable to read disk capacity"), name);
3940 
3941 		return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3942 	}
3943 
3944 	slice_size = vtoc->efi_last_u_lba + 1;
3945 	slice_size -= EFI_MIN_RESV_SIZE;
3946 	if (start_block == MAXOFFSET_T)
3947 		start_block = NEW_START_BLOCK;
3948 	slice_size -= start_block;
3949 
3950 	vtoc->efi_parts[0].p_start = start_block;
3951 	vtoc->efi_parts[0].p_size = slice_size;
3952 
3953 	/*
3954 	 * Why we use V_USR: V_BACKUP confuses users, and is considered
3955 	 * disposable by some EFI utilities (since EFI doesn't have a backup
3956 	 * slice).  V_UNASSIGNED is supposed to be used only for zero size
3957 	 * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3958 	 * etc. were all pretty specific.  V_USR is as close to reality as we
3959 	 * can get, in the absence of V_OTHER.
3960 	 */
3961 	vtoc->efi_parts[0].p_tag = V_USR;
3962 	(void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3963 
3964 	vtoc->efi_parts[8].p_start = slice_size + start_block;
3965 	vtoc->efi_parts[8].p_size = resv;
3966 	vtoc->efi_parts[8].p_tag = V_RESERVED;
3967 
3968 	if (efi_write(fd, vtoc) != 0) {
3969 		/*
3970 		 * Some block drivers (like pcata) may not support EFI
3971 		 * GPT labels.  Print out a helpful error message dir-
3972 		 * ecting the user to manually label the disk and give
3973 		 * a specific slice.
3974 		 */
3975 		(void) close(fd);
3976 		efi_free(vtoc);
3977 
3978 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3979 		    "try using fdisk(1M) and then provide a specific slice"));
3980 		return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3981 	}
3982 
3983 	(void) close(fd);
3984 	efi_free(vtoc);
3985 	return (0);
3986 }
3987 
3988 static boolean_t
3989 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
3990 {
3991 	char *type;
3992 	nvlist_t **child;
3993 	uint_t children, c;
3994 
3995 	verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
3996 	if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
3997 	    strcmp(type, VDEV_TYPE_HOLE) == 0 ||
3998 	    strcmp(type, VDEV_TYPE_MISSING) == 0) {
3999 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4000 		    "vdev type '%s' is not supported"), type);
4001 		(void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4002 		return (B_FALSE);
4003 	}
4004 	if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4005 	    &child, &children) == 0) {
4006 		for (c = 0; c < children; c++) {
4007 			if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4008 				return (B_FALSE);
4009 		}
4010 	}
4011 	return (B_TRUE);
4012 }
4013 
4014 /*
4015  * Check if this zvol is allowable for use as a dump device; zero if
4016  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4017  *
4018  * Allowable storage configurations include mirrors, all raidz variants, and
4019  * pools with log, cache, and spare devices.  Pools which are backed by files or
4020  * have missing/hole vdevs are not suitable.
4021  */
4022 int
4023 zvol_check_dump_config(char *arg)
4024 {
4025 	zpool_handle_t *zhp = NULL;
4026 	nvlist_t *config, *nvroot;
4027 	char *p, *volname;
4028 	nvlist_t **top;
4029 	uint_t toplevels;
4030 	libzfs_handle_t *hdl;
4031 	char errbuf[1024];
4032 	char poolname[ZPOOL_MAXNAMELEN];
4033 	int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4034 	int ret = 1;
4035 
4036 	if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4037 		return (-1);
4038 	}
4039 
4040 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4041 	    "dump is not supported on device '%s'"), arg);
4042 
4043 	if ((hdl = libzfs_init()) == NULL)
4044 		return (1);
4045 	libzfs_print_on_error(hdl, B_TRUE);
4046 
4047 	volname = arg + pathlen;
4048 
4049 	/* check the configuration of the pool */
4050 	if ((p = strchr(volname, '/')) == NULL) {
4051 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4052 		    "malformed dataset name"));
4053 		(void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4054 		return (1);
4055 	} else if (p - volname >= ZFS_MAXNAMELEN) {
4056 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4057 		    "dataset name is too long"));
4058 		(void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4059 		return (1);
4060 	} else {
4061 		(void) strncpy(poolname, volname, p - volname);
4062 		poolname[p - volname] = '\0';
4063 	}
4064 
4065 	if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4066 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4067 		    "could not open pool '%s'"), poolname);
4068 		(void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4069 		goto out;
4070 	}
4071 	config = zpool_get_config(zhp, NULL);
4072 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4073 	    &nvroot) != 0) {
4074 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4075 		    "could not obtain vdev configuration for  '%s'"), poolname);
4076 		(void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4077 		goto out;
4078 	}
4079 
4080 	verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4081 	    &top, &toplevels) == 0);
4082 
4083 	if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4084 		goto out;
4085 	}
4086 	ret = 0;
4087 
4088 out:
4089 	if (zhp)
4090 		zpool_close(zhp);
4091 	libzfs_fini(hdl);
4092 	return (ret);
4093 }
4094