xref: /freebsd/sys/contrib/openzfs/lib/libzfs/libzfs_pool.c (revision 809922b01004daf627ad4b8d92c7f98eb579043c)
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 2015 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
26  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
27  * Copyright (c) 2018 Datto Inc.
28  * Copyright (c) 2017 Open-E, Inc. All Rights Reserved.
29  * Copyright (c) 2017, Intel Corporation.
30  * Copyright (c) 2018, loli10K <ezomori.nozomu@gmail.com>
31  * Copyright (c) 2021, Colm Buckley <colm@tuatha.org>
32  * Copyright (c) 2021, Klara Inc.
33  */
34 
35 #include <errno.h>
36 #include <libintl.h>
37 #include <stdio.h>
38 #include <stdlib.h>
39 #include <strings.h>
40 #include <unistd.h>
41 #include <libgen.h>
42 #include <zone.h>
43 #include <sys/stat.h>
44 #include <sys/efi_partition.h>
45 #include <sys/systeminfo.h>
46 #include <sys/zfs_ioctl.h>
47 #include <sys/zfs_sysfs.h>
48 #include <sys/vdev_disk.h>
49 #include <sys/types.h>
50 #include <dlfcn.h>
51 #include <libzutil.h>
52 #include <fcntl.h>
53 
54 #include "zfs_namecheck.h"
55 #include "zfs_prop.h"
56 #include "libzfs_impl.h"
57 #include "zfs_comutil.h"
58 #include "zfeature_common.h"
59 
60 static boolean_t zpool_vdev_is_interior(const char *name);
61 
62 typedef struct prop_flags {
63 	int create:1;	/* Validate property on creation */
64 	int import:1;	/* Validate property on import */
65 	int vdevprop:1;	/* Validate property as a VDEV property */
66 } prop_flags_t;
67 
68 /*
69  * ====================================================================
70  *   zpool property functions
71  * ====================================================================
72  */
73 
74 static int
75 zpool_get_all_props(zpool_handle_t *zhp)
76 {
77 	zfs_cmd_t zc = {"\0"};
78 	libzfs_handle_t *hdl = zhp->zpool_hdl;
79 
80 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
81 
82 	zcmd_alloc_dst_nvlist(hdl, &zc, 0);
83 
84 	while (zfs_ioctl(hdl, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
85 		if (errno == ENOMEM)
86 			zcmd_expand_dst_nvlist(hdl, &zc);
87 		else {
88 			zcmd_free_nvlists(&zc);
89 			return (-1);
90 		}
91 	}
92 
93 	if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
94 		zcmd_free_nvlists(&zc);
95 		return (-1);
96 	}
97 
98 	zcmd_free_nvlists(&zc);
99 
100 	return (0);
101 }
102 
103 int
104 zpool_props_refresh(zpool_handle_t *zhp)
105 {
106 	nvlist_t *old_props;
107 
108 	old_props = zhp->zpool_props;
109 
110 	if (zpool_get_all_props(zhp) != 0)
111 		return (-1);
112 
113 	nvlist_free(old_props);
114 	return (0);
115 }
116 
117 static const char *
118 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
119     zprop_source_t *src)
120 {
121 	nvlist_t *nv, *nvl;
122 	char *value;
123 	zprop_source_t source;
124 
125 	nvl = zhp->zpool_props;
126 	if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
127 		source = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
128 		value = fnvlist_lookup_string(nv, ZPROP_VALUE);
129 	} else {
130 		source = ZPROP_SRC_DEFAULT;
131 		if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
132 			value = "-";
133 	}
134 
135 	if (src)
136 		*src = source;
137 
138 	return (value);
139 }
140 
141 uint64_t
142 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
143 {
144 	nvlist_t *nv, *nvl;
145 	uint64_t value;
146 	zprop_source_t source;
147 
148 	if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
149 		/*
150 		 * zpool_get_all_props() has most likely failed because
151 		 * the pool is faulted, but if all we need is the top level
152 		 * vdev's guid then get it from the zhp config nvlist.
153 		 */
154 		if ((prop == ZPOOL_PROP_GUID) &&
155 		    (nvlist_lookup_nvlist(zhp->zpool_config,
156 		    ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
157 		    (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
158 		    == 0)) {
159 			return (value);
160 		}
161 		return (zpool_prop_default_numeric(prop));
162 	}
163 
164 	nvl = zhp->zpool_props;
165 	if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
166 		source = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
167 		value = fnvlist_lookup_uint64(nv, ZPROP_VALUE);
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 const 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 	default:
206 		break;
207 	}
208 
209 	return (gettext("UNKNOWN"));
210 }
211 
212 /*
213  * Map POOL STATE to printed strings.
214  */
215 const char *
216 zpool_pool_state_to_name(pool_state_t state)
217 {
218 	switch (state) {
219 	default:
220 		break;
221 	case POOL_STATE_ACTIVE:
222 		return (gettext("ACTIVE"));
223 	case POOL_STATE_EXPORTED:
224 		return (gettext("EXPORTED"));
225 	case POOL_STATE_DESTROYED:
226 		return (gettext("DESTROYED"));
227 	case POOL_STATE_SPARE:
228 		return (gettext("SPARE"));
229 	case POOL_STATE_L2CACHE:
230 		return (gettext("L2CACHE"));
231 	case POOL_STATE_UNINITIALIZED:
232 		return (gettext("UNINITIALIZED"));
233 	case POOL_STATE_UNAVAIL:
234 		return (gettext("UNAVAIL"));
235 	case POOL_STATE_POTENTIALLY_ACTIVE:
236 		return (gettext("POTENTIALLY_ACTIVE"));
237 	}
238 
239 	return (gettext("UNKNOWN"));
240 }
241 
242 /*
243  * Given a pool handle, return the pool health string ("ONLINE", "DEGRADED",
244  * "SUSPENDED", etc).
245  */
246 const char *
247 zpool_get_state_str(zpool_handle_t *zhp)
248 {
249 	zpool_errata_t errata;
250 	zpool_status_t status;
251 	const char *str;
252 
253 	status = zpool_get_status(zhp, NULL, &errata);
254 
255 	if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
256 		str = gettext("FAULTED");
257 	} else if (status == ZPOOL_STATUS_IO_FAILURE_WAIT ||
258 	    status == ZPOOL_STATUS_IO_FAILURE_MMP) {
259 		str = gettext("SUSPENDED");
260 	} else {
261 		nvlist_t *nvroot = fnvlist_lookup_nvlist(
262 		    zpool_get_config(zhp, NULL), ZPOOL_CONFIG_VDEV_TREE);
263 		uint_t vsc;
264 		vdev_stat_t *vs = (vdev_stat_t *)fnvlist_lookup_uint64_array(
265 		    nvroot, ZPOOL_CONFIG_VDEV_STATS, &vsc);
266 		str = zpool_state_to_name(vs->vs_state, vs->vs_aux);
267 	}
268 	return (str);
269 }
270 
271 /*
272  * Get a zpool property value for 'prop' and return the value in
273  * a pre-allocated buffer.
274  */
275 int
276 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf,
277     size_t len, zprop_source_t *srctype, boolean_t literal)
278 {
279 	uint64_t intval;
280 	const char *strval;
281 	zprop_source_t src = ZPROP_SRC_NONE;
282 
283 	if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
284 		switch (prop) {
285 		case ZPOOL_PROP_NAME:
286 			(void) strlcpy(buf, zpool_get_name(zhp), len);
287 			break;
288 
289 		case ZPOOL_PROP_HEALTH:
290 			(void) strlcpy(buf, zpool_get_state_str(zhp), len);
291 			break;
292 
293 		case ZPOOL_PROP_GUID:
294 			intval = zpool_get_prop_int(zhp, prop, &src);
295 			(void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
296 			break;
297 
298 		case ZPOOL_PROP_ALTROOT:
299 		case ZPOOL_PROP_CACHEFILE:
300 		case ZPOOL_PROP_COMMENT:
301 		case ZPOOL_PROP_COMPATIBILITY:
302 			if (zhp->zpool_props != NULL ||
303 			    zpool_get_all_props(zhp) == 0) {
304 				(void) strlcpy(buf,
305 				    zpool_get_prop_string(zhp, prop, &src),
306 				    len);
307 				break;
308 			}
309 			zfs_fallthrough;
310 		default:
311 			(void) strlcpy(buf, "-", len);
312 			break;
313 		}
314 
315 		if (srctype != NULL)
316 			*srctype = src;
317 		return (0);
318 	}
319 
320 	if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
321 	    prop != ZPOOL_PROP_NAME)
322 		return (-1);
323 
324 	switch (zpool_prop_get_type(prop)) {
325 	case PROP_TYPE_STRING:
326 		(void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
327 		    len);
328 		break;
329 
330 	case PROP_TYPE_NUMBER:
331 		intval = zpool_get_prop_int(zhp, prop, &src);
332 
333 		switch (prop) {
334 		case ZPOOL_PROP_SIZE:
335 		case ZPOOL_PROP_ALLOCATED:
336 		case ZPOOL_PROP_FREE:
337 		case ZPOOL_PROP_FREEING:
338 		case ZPOOL_PROP_LEAKED:
339 		case ZPOOL_PROP_ASHIFT:
340 		case ZPOOL_PROP_MAXBLOCKSIZE:
341 		case ZPOOL_PROP_MAXDNODESIZE:
342 			if (literal)
343 				(void) snprintf(buf, len, "%llu",
344 				    (u_longlong_t)intval);
345 			else
346 				(void) zfs_nicenum(intval, buf, len);
347 			break;
348 
349 		case ZPOOL_PROP_EXPANDSZ:
350 		case ZPOOL_PROP_CHECKPOINT:
351 			if (intval == 0) {
352 				(void) strlcpy(buf, "-", len);
353 			} else if (literal) {
354 				(void) snprintf(buf, len, "%llu",
355 				    (u_longlong_t)intval);
356 			} else {
357 				(void) zfs_nicebytes(intval, buf, len);
358 			}
359 			break;
360 
361 		case ZPOOL_PROP_CAPACITY:
362 			if (literal) {
363 				(void) snprintf(buf, len, "%llu",
364 				    (u_longlong_t)intval);
365 			} else {
366 				(void) snprintf(buf, len, "%llu%%",
367 				    (u_longlong_t)intval);
368 			}
369 			break;
370 
371 		case ZPOOL_PROP_FRAGMENTATION:
372 			if (intval == UINT64_MAX) {
373 				(void) strlcpy(buf, "-", len);
374 			} else if (literal) {
375 				(void) snprintf(buf, len, "%llu",
376 				    (u_longlong_t)intval);
377 			} else {
378 				(void) snprintf(buf, len, "%llu%%",
379 				    (u_longlong_t)intval);
380 			}
381 			break;
382 
383 		case ZPOOL_PROP_DEDUPRATIO:
384 			if (literal)
385 				(void) snprintf(buf, len, "%llu.%02llu",
386 				    (u_longlong_t)(intval / 100),
387 				    (u_longlong_t)(intval % 100));
388 			else
389 				(void) snprintf(buf, len, "%llu.%02llux",
390 				    (u_longlong_t)(intval / 100),
391 				    (u_longlong_t)(intval % 100));
392 			break;
393 
394 		case ZPOOL_PROP_HEALTH:
395 			(void) strlcpy(buf, zpool_get_state_str(zhp), len);
396 			break;
397 		case ZPOOL_PROP_VERSION:
398 			if (intval >= SPA_VERSION_FEATURES) {
399 				(void) snprintf(buf, len, "-");
400 				break;
401 			}
402 			zfs_fallthrough;
403 		default:
404 			(void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
405 		}
406 		break;
407 
408 	case PROP_TYPE_INDEX:
409 		intval = zpool_get_prop_int(zhp, prop, &src);
410 		if (zpool_prop_index_to_string(prop, intval, &strval)
411 		    != 0)
412 			return (-1);
413 		(void) strlcpy(buf, strval, len);
414 		break;
415 
416 	default:
417 		abort();
418 	}
419 
420 	if (srctype)
421 		*srctype = src;
422 
423 	return (0);
424 }
425 
426 /*
427  * Check if the bootfs name has the same pool name as it is set to.
428  * Assuming bootfs is a valid dataset name.
429  */
430 static boolean_t
431 bootfs_name_valid(const char *pool, const char *bootfs)
432 {
433 	int len = strlen(pool);
434 	if (bootfs[0] == '\0')
435 		return (B_TRUE);
436 
437 	if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
438 		return (B_FALSE);
439 
440 	if (strncmp(pool, bootfs, len) == 0 &&
441 	    (bootfs[len] == '/' || bootfs[len] == '\0'))
442 		return (B_TRUE);
443 
444 	return (B_FALSE);
445 }
446 
447 /*
448  * Given an nvlist of zpool properties to be set, validate that they are
449  * correct, and parse any numeric properties (index, boolean, etc) if they are
450  * specified as strings.
451  */
452 static nvlist_t *
453 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
454     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
455 {
456 	nvpair_t *elem;
457 	nvlist_t *retprops;
458 	zpool_prop_t prop;
459 	char *strval;
460 	uint64_t intval;
461 	char *slash, *check;
462 	struct stat64 statbuf;
463 	zpool_handle_t *zhp;
464 	char report[1024];
465 
466 	if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
467 		(void) no_memory(hdl);
468 		return (NULL);
469 	}
470 
471 	elem = NULL;
472 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
473 		const char *propname = nvpair_name(elem);
474 
475 		if (flags.vdevprop && zpool_prop_vdev(propname)) {
476 			vdev_prop_t vprop = vdev_name_to_prop(propname);
477 
478 			if (vdev_prop_readonly(vprop)) {
479 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
480 				    "is readonly"), propname);
481 				(void) zfs_error(hdl, EZFS_PROPREADONLY,
482 				    errbuf);
483 				goto error;
484 			}
485 
486 			if (zprop_parse_value(hdl, elem, vprop, ZFS_TYPE_VDEV,
487 			    retprops, &strval, &intval, errbuf) != 0)
488 				goto error;
489 
490 			continue;
491 		} else if (flags.vdevprop && vdev_prop_user(propname)) {
492 			if (nvlist_add_nvpair(retprops, elem) != 0) {
493 				(void) no_memory(hdl);
494 				goto error;
495 			}
496 			continue;
497 		} else if (flags.vdevprop) {
498 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
499 			    "invalid property: '%s'"), propname);
500 			(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
501 			goto error;
502 		}
503 
504 		prop = zpool_name_to_prop(propname);
505 		if (prop == ZPOOL_PROP_INVAL && zpool_prop_feature(propname)) {
506 			int err;
507 			char *fname = strchr(propname, '@') + 1;
508 
509 			err = zfeature_lookup_name(fname, NULL);
510 			if (err != 0) {
511 				ASSERT3U(err, ==, ENOENT);
512 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
513 				    "feature '%s' unsupported by kernel"),
514 				    fname);
515 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
516 				goto error;
517 			}
518 
519 			if (nvpair_type(elem) != DATA_TYPE_STRING) {
520 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
521 				    "'%s' must be a string"), propname);
522 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
523 				goto error;
524 			}
525 
526 			(void) nvpair_value_string(elem, &strval);
527 			if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0 &&
528 			    strcmp(strval, ZFS_FEATURE_DISABLED) != 0) {
529 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
530 				    "property '%s' can only be set to "
531 				    "'enabled' or 'disabled'"), propname);
532 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
533 				goto error;
534 			}
535 
536 			if (!flags.create &&
537 			    strcmp(strval, ZFS_FEATURE_DISABLED) == 0) {
538 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
539 				    "property '%s' can only be set to "
540 				    "'disabled' at creation time"), propname);
541 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
542 				goto error;
543 			}
544 
545 			if (nvlist_add_uint64(retprops, propname, 0) != 0) {
546 				(void) no_memory(hdl);
547 				goto error;
548 			}
549 			continue;
550 		}
551 
552 		/*
553 		 * Make sure this property is valid and applies to this type.
554 		 */
555 		if (prop == ZPOOL_PROP_INVAL) {
556 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
557 			    "invalid property '%s'"), propname);
558 			(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
559 			goto error;
560 		}
561 
562 		if (zpool_prop_readonly(prop)) {
563 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
564 			    "is readonly"), propname);
565 			(void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
566 			goto error;
567 		}
568 
569 		if (!flags.create && zpool_prop_setonce(prop)) {
570 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
571 			    "property '%s' can only be set at "
572 			    "creation time"), propname);
573 			(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
574 			goto error;
575 		}
576 
577 		if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
578 		    &strval, &intval, errbuf) != 0)
579 			goto error;
580 
581 		/*
582 		 * Perform additional checking for specific properties.
583 		 */
584 		switch (prop) {
585 		case ZPOOL_PROP_VERSION:
586 			if (intval < version ||
587 			    !SPA_VERSION_IS_SUPPORTED(intval)) {
588 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
589 				    "property '%s' number %llu is invalid."),
590 				    propname, (unsigned long long)intval);
591 				(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
592 				goto error;
593 			}
594 			break;
595 
596 		case ZPOOL_PROP_ASHIFT:
597 			if (intval != 0 &&
598 			    (intval < ASHIFT_MIN || intval > ASHIFT_MAX)) {
599 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
600 				    "property '%s' number %llu is invalid, "
601 				    "only values between %" PRId32 " and %"
602 				    PRId32 " are allowed."),
603 				    propname, (unsigned long long)intval,
604 				    ASHIFT_MIN, ASHIFT_MAX);
605 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
606 				goto error;
607 			}
608 			break;
609 
610 		case ZPOOL_PROP_BOOTFS:
611 			if (flags.create || flags.import) {
612 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
613 				    "property '%s' cannot be set at creation "
614 				    "or import time"), propname);
615 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
616 				goto error;
617 			}
618 
619 			if (version < SPA_VERSION_BOOTFS) {
620 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
621 				    "pool must be upgraded to support "
622 				    "'%s' property"), propname);
623 				(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
624 				goto error;
625 			}
626 
627 			/*
628 			 * bootfs property value has to be a dataset name and
629 			 * the dataset has to be in the same pool as it sets to.
630 			 */
631 			if (!bootfs_name_valid(poolname, strval)) {
632 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
633 				    "is an invalid name"), strval);
634 				(void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
635 				goto error;
636 			}
637 
638 			if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
639 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
640 				    "could not open pool '%s'"), poolname);
641 				(void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
642 				goto error;
643 			}
644 			zpool_close(zhp);
645 			break;
646 
647 		case ZPOOL_PROP_ALTROOT:
648 			if (!flags.create && !flags.import) {
649 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
650 				    "property '%s' can only be set during pool "
651 				    "creation or import"), propname);
652 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
653 				goto error;
654 			}
655 
656 			if (strval[0] != '/') {
657 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
658 				    "bad alternate root '%s'"), strval);
659 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
660 				goto error;
661 			}
662 			break;
663 
664 		case ZPOOL_PROP_CACHEFILE:
665 			if (strval[0] == '\0')
666 				break;
667 
668 			if (strcmp(strval, "none") == 0)
669 				break;
670 
671 			if (strval[0] != '/') {
672 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
673 				    "property '%s' must be empty, an "
674 				    "absolute path, or 'none'"), propname);
675 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
676 				goto error;
677 			}
678 
679 			slash = strrchr(strval, '/');
680 
681 			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
682 			    strcmp(slash, "/..") == 0) {
683 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
684 				    "'%s' is not a valid file"), strval);
685 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
686 				goto error;
687 			}
688 
689 			*slash = '\0';
690 
691 			if (strval[0] != '\0' &&
692 			    (stat64(strval, &statbuf) != 0 ||
693 			    !S_ISDIR(statbuf.st_mode))) {
694 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
695 				    "'%s' is not a valid directory"),
696 				    strval);
697 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
698 				goto error;
699 			}
700 
701 			*slash = '/';
702 			break;
703 
704 		case ZPOOL_PROP_COMPATIBILITY:
705 			switch (zpool_load_compat(strval, NULL, report, 1024)) {
706 			case ZPOOL_COMPATIBILITY_OK:
707 			case ZPOOL_COMPATIBILITY_WARNTOKEN:
708 				break;
709 			case ZPOOL_COMPATIBILITY_BADFILE:
710 			case ZPOOL_COMPATIBILITY_BADTOKEN:
711 			case ZPOOL_COMPATIBILITY_NOFILES:
712 				zfs_error_aux(hdl, "%s", report);
713 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
714 				goto error;
715 			}
716 			break;
717 
718 		case ZPOOL_PROP_COMMENT:
719 			for (check = strval; *check != '\0'; check++) {
720 				if (!isprint(*check)) {
721 					zfs_error_aux(hdl,
722 					    dgettext(TEXT_DOMAIN,
723 					    "comment may only have printable "
724 					    "characters"));
725 					(void) zfs_error(hdl, EZFS_BADPROP,
726 					    errbuf);
727 					goto error;
728 				}
729 			}
730 			if (strlen(strval) > ZPROP_MAX_COMMENT) {
731 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
732 				    "comment must not exceed %d characters"),
733 				    ZPROP_MAX_COMMENT);
734 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
735 				goto error;
736 			}
737 			break;
738 		case ZPOOL_PROP_READONLY:
739 			if (!flags.import) {
740 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
741 				    "property '%s' can only be set at "
742 				    "import time"), propname);
743 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
744 				goto error;
745 			}
746 			break;
747 		case ZPOOL_PROP_MULTIHOST:
748 			if (get_system_hostid() == 0) {
749 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
750 				    "requires a non-zero system hostid"));
751 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
752 				goto error;
753 			}
754 			break;
755 		case ZPOOL_PROP_DEDUPDITTO:
756 			printf("Note: property '%s' no longer has "
757 			    "any effect\n", propname);
758 			break;
759 
760 		default:
761 			break;
762 		}
763 	}
764 
765 	return (retprops);
766 error:
767 	nvlist_free(retprops);
768 	return (NULL);
769 }
770 
771 /*
772  * Set zpool property : propname=propval.
773  */
774 int
775 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
776 {
777 	zfs_cmd_t zc = {"\0"};
778 	int ret = -1;
779 	char errbuf[1024];
780 	nvlist_t *nvl = NULL;
781 	nvlist_t *realprops;
782 	uint64_t version;
783 	prop_flags_t flags = { 0 };
784 
785 	(void) snprintf(errbuf, sizeof (errbuf),
786 	    dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
787 	    zhp->zpool_name);
788 
789 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
790 		return (no_memory(zhp->zpool_hdl));
791 
792 	if (nvlist_add_string(nvl, propname, propval) != 0) {
793 		nvlist_free(nvl);
794 		return (no_memory(zhp->zpool_hdl));
795 	}
796 
797 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
798 	if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
799 	    zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
800 		nvlist_free(nvl);
801 		return (-1);
802 	}
803 
804 	nvlist_free(nvl);
805 	nvl = realprops;
806 
807 	/*
808 	 * Execute the corresponding ioctl() to set this property.
809 	 */
810 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
811 
812 	zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl);
813 
814 	ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
815 
816 	zcmd_free_nvlists(&zc);
817 	nvlist_free(nvl);
818 
819 	if (ret)
820 		(void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
821 	else
822 		(void) zpool_props_refresh(zhp);
823 
824 	return (ret);
825 }
826 
827 int
828 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp,
829     zfs_type_t type, boolean_t literal)
830 {
831 	libzfs_handle_t *hdl = zhp->zpool_hdl;
832 	zprop_list_t *entry;
833 	char buf[ZFS_MAXPROPLEN];
834 	nvlist_t *features = NULL;
835 	nvpair_t *nvp;
836 	zprop_list_t **last;
837 	boolean_t firstexpand = (NULL == *plp);
838 	int i;
839 
840 	if (zprop_expand_list(hdl, plp, type) != 0)
841 		return (-1);
842 
843 	if (type == ZFS_TYPE_VDEV)
844 		return (0);
845 
846 	last = plp;
847 	while (*last != NULL)
848 		last = &(*last)->pl_next;
849 
850 	if ((*plp)->pl_all)
851 		features = zpool_get_features(zhp);
852 
853 	if ((*plp)->pl_all && firstexpand) {
854 		for (i = 0; i < SPA_FEATURES; i++) {
855 			zprop_list_t *entry = zfs_alloc(hdl,
856 			    sizeof (zprop_list_t));
857 			entry->pl_prop = ZPROP_INVAL;
858 			entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
859 			    spa_feature_table[i].fi_uname);
860 			entry->pl_width = strlen(entry->pl_user_prop);
861 			entry->pl_all = B_TRUE;
862 
863 			*last = entry;
864 			last = &entry->pl_next;
865 		}
866 	}
867 
868 	/* add any unsupported features */
869 	for (nvp = nvlist_next_nvpair(features, NULL);
870 	    nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
871 		char *propname;
872 		boolean_t found;
873 		zprop_list_t *entry;
874 
875 		if (zfeature_is_supported(nvpair_name(nvp)))
876 			continue;
877 
878 		propname = zfs_asprintf(hdl, "unsupported@%s",
879 		    nvpair_name(nvp));
880 
881 		/*
882 		 * Before adding the property to the list make sure that no
883 		 * other pool already added the same property.
884 		 */
885 		found = B_FALSE;
886 		entry = *plp;
887 		while (entry != NULL) {
888 			if (entry->pl_user_prop != NULL &&
889 			    strcmp(propname, entry->pl_user_prop) == 0) {
890 				found = B_TRUE;
891 				break;
892 			}
893 			entry = entry->pl_next;
894 		}
895 		if (found) {
896 			free(propname);
897 			continue;
898 		}
899 
900 		entry = zfs_alloc(hdl, sizeof (zprop_list_t));
901 		entry->pl_prop = ZPROP_INVAL;
902 		entry->pl_user_prop = propname;
903 		entry->pl_width = strlen(entry->pl_user_prop);
904 		entry->pl_all = B_TRUE;
905 
906 		*last = entry;
907 		last = &entry->pl_next;
908 	}
909 
910 	for (entry = *plp; entry != NULL; entry = entry->pl_next) {
911 		if (entry->pl_fixed && !literal)
912 			continue;
913 
914 		if (entry->pl_prop != ZPROP_INVAL &&
915 		    zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
916 		    NULL, literal) == 0) {
917 			if (strlen(buf) > entry->pl_width)
918 				entry->pl_width = strlen(buf);
919 		}
920 	}
921 
922 	return (0);
923 }
924 
925 int
926 vdev_expand_proplist(zpool_handle_t *zhp, const char *vdevname,
927     zprop_list_t **plp)
928 {
929 	zprop_list_t *entry;
930 	char buf[ZFS_MAXPROPLEN];
931 	char *strval = NULL;
932 	int err = 0;
933 	nvpair_t *elem = NULL;
934 	nvlist_t *vprops = NULL;
935 	nvlist_t *propval = NULL;
936 	const char *propname;
937 	vdev_prop_t prop;
938 	zprop_list_t **last;
939 
940 	for (entry = *plp; entry != NULL; entry = entry->pl_next) {
941 		if (entry->pl_fixed)
942 			continue;
943 
944 		if (zpool_get_vdev_prop(zhp, vdevname, entry->pl_prop,
945 		    entry->pl_user_prop, buf, sizeof (buf), NULL,
946 		    B_FALSE) == 0) {
947 			if (strlen(buf) > entry->pl_width)
948 				entry->pl_width = strlen(buf);
949 		}
950 		if (entry->pl_prop == VDEV_PROP_NAME &&
951 		    strlen(vdevname) > entry->pl_width)
952 			entry->pl_width = strlen(vdevname);
953 	}
954 
955 	/* Handle the all properties case */
956 	last = plp;
957 	if (*last != NULL && (*last)->pl_all == B_TRUE) {
958 		while (*last != NULL)
959 			last = &(*last)->pl_next;
960 
961 		err = zpool_get_all_vdev_props(zhp, vdevname, &vprops);
962 		if (err != 0)
963 			return (err);
964 
965 		while ((elem = nvlist_next_nvpair(vprops, elem)) != NULL) {
966 			propname = nvpair_name(elem);
967 
968 			/* Skip properties that are not user defined */
969 			if ((prop = vdev_name_to_prop(propname)) !=
970 			    VDEV_PROP_USER)
971 				continue;
972 
973 			if (nvpair_value_nvlist(elem, &propval) != 0)
974 				continue;
975 
976 			strval = fnvlist_lookup_string(propval, ZPROP_VALUE);
977 
978 			entry = zfs_alloc(zhp->zpool_hdl,
979 			    sizeof (zprop_list_t));
980 			entry->pl_prop = prop;
981 			entry->pl_user_prop = zfs_strdup(zhp->zpool_hdl,
982 			    propname);
983 			entry->pl_width = strlen(strval);
984 			entry->pl_all = B_TRUE;
985 			*last = entry;
986 			last = &entry->pl_next;
987 		}
988 	}
989 
990 	return (0);
991 }
992 
993 /*
994  * Get the state for the given feature on the given ZFS pool.
995  */
996 int
997 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
998     size_t len)
999 {
1000 	uint64_t refcount;
1001 	boolean_t found = B_FALSE;
1002 	nvlist_t *features = zpool_get_features(zhp);
1003 	boolean_t supported;
1004 	const char *feature = strchr(propname, '@') + 1;
1005 
1006 	supported = zpool_prop_feature(propname);
1007 	ASSERT(supported || zpool_prop_unsupported(propname));
1008 
1009 	/*
1010 	 * Convert from feature name to feature guid. This conversion is
1011 	 * unnecessary for unsupported@... properties because they already
1012 	 * use guids.
1013 	 */
1014 	if (supported) {
1015 		int ret;
1016 		spa_feature_t fid;
1017 
1018 		ret = zfeature_lookup_name(feature, &fid);
1019 		if (ret != 0) {
1020 			(void) strlcpy(buf, "-", len);
1021 			return (ENOTSUP);
1022 		}
1023 		feature = spa_feature_table[fid].fi_guid;
1024 	}
1025 
1026 	if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
1027 		found = B_TRUE;
1028 
1029 	if (supported) {
1030 		if (!found) {
1031 			(void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
1032 		} else  {
1033 			if (refcount == 0)
1034 				(void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
1035 			else
1036 				(void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
1037 		}
1038 	} else {
1039 		if (found) {
1040 			if (refcount == 0) {
1041 				(void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
1042 			} else {
1043 				(void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
1044 			}
1045 		} else {
1046 			(void) strlcpy(buf, "-", len);
1047 			return (ENOTSUP);
1048 		}
1049 	}
1050 
1051 	return (0);
1052 }
1053 
1054 /*
1055  * Validate the given pool name, optionally putting an extended error message in
1056  * 'buf'.
1057  */
1058 boolean_t
1059 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
1060 {
1061 	namecheck_err_t why;
1062 	char what;
1063 	int ret;
1064 
1065 	ret = pool_namecheck(pool, &why, &what);
1066 
1067 	/*
1068 	 * The rules for reserved pool names were extended at a later point.
1069 	 * But we need to support users with existing pools that may now be
1070 	 * invalid.  So we only check for this expanded set of names during a
1071 	 * create (or import), and only in userland.
1072 	 */
1073 	if (ret == 0 && !isopen &&
1074 	    (strncmp(pool, "mirror", 6) == 0 ||
1075 	    strncmp(pool, "raidz", 5) == 0 ||
1076 	    strncmp(pool, "draid", 5) == 0 ||
1077 	    strncmp(pool, "spare", 5) == 0 ||
1078 	    strcmp(pool, "log") == 0)) {
1079 		if (hdl != NULL)
1080 			zfs_error_aux(hdl,
1081 			    dgettext(TEXT_DOMAIN, "name is reserved"));
1082 		return (B_FALSE);
1083 	}
1084 
1085 
1086 	if (ret != 0) {
1087 		if (hdl != NULL) {
1088 			switch (why) {
1089 			case NAME_ERR_TOOLONG:
1090 				zfs_error_aux(hdl,
1091 				    dgettext(TEXT_DOMAIN, "name is too long"));
1092 				break;
1093 
1094 			case NAME_ERR_INVALCHAR:
1095 				zfs_error_aux(hdl,
1096 				    dgettext(TEXT_DOMAIN, "invalid character "
1097 				    "'%c' in pool name"), what);
1098 				break;
1099 
1100 			case NAME_ERR_NOLETTER:
1101 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1102 				    "name must begin with a letter"));
1103 				break;
1104 
1105 			case NAME_ERR_RESERVED:
1106 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1107 				    "name is reserved"));
1108 				break;
1109 
1110 			case NAME_ERR_DISKLIKE:
1111 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1112 				    "pool name is reserved"));
1113 				break;
1114 
1115 			case NAME_ERR_LEADING_SLASH:
1116 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1117 				    "leading slash in name"));
1118 				break;
1119 
1120 			case NAME_ERR_EMPTY_COMPONENT:
1121 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1122 				    "empty component in name"));
1123 				break;
1124 
1125 			case NAME_ERR_TRAILING_SLASH:
1126 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1127 				    "trailing slash in name"));
1128 				break;
1129 
1130 			case NAME_ERR_MULTIPLE_DELIMITERS:
1131 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1132 				    "multiple '@' and/or '#' delimiters in "
1133 				    "name"));
1134 				break;
1135 
1136 			case NAME_ERR_NO_AT:
1137 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1138 				    "permission set is missing '@'"));
1139 				break;
1140 
1141 			default:
1142 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1143 				    "(%d) not defined"), why);
1144 				break;
1145 			}
1146 		}
1147 		return (B_FALSE);
1148 	}
1149 
1150 	return (B_TRUE);
1151 }
1152 
1153 /*
1154  * Open a handle to the given pool, even if the pool is currently in the FAULTED
1155  * state.
1156  */
1157 zpool_handle_t *
1158 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
1159 {
1160 	zpool_handle_t *zhp;
1161 	boolean_t missing;
1162 
1163 	/*
1164 	 * Make sure the pool name is valid.
1165 	 */
1166 	if (!zpool_name_valid(hdl, B_TRUE, pool)) {
1167 		(void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1168 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1169 		    pool);
1170 		return (NULL);
1171 	}
1172 
1173 	zhp = zfs_alloc(hdl, sizeof (zpool_handle_t));
1174 
1175 	zhp->zpool_hdl = hdl;
1176 	(void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1177 
1178 	if (zpool_refresh_stats(zhp, &missing) != 0) {
1179 		zpool_close(zhp);
1180 		return (NULL);
1181 	}
1182 
1183 	if (missing) {
1184 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1185 		(void) zfs_error_fmt(hdl, EZFS_NOENT,
1186 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1187 		zpool_close(zhp);
1188 		return (NULL);
1189 	}
1190 
1191 	return (zhp);
1192 }
1193 
1194 /*
1195  * Like the above, but silent on error.  Used when iterating over pools (because
1196  * the configuration cache may be out of date).
1197  */
1198 int
1199 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1200 {
1201 	zpool_handle_t *zhp;
1202 	boolean_t missing;
1203 
1204 	zhp = zfs_alloc(hdl, sizeof (zpool_handle_t));
1205 
1206 	zhp->zpool_hdl = hdl;
1207 	(void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1208 
1209 	if (zpool_refresh_stats(zhp, &missing) != 0) {
1210 		zpool_close(zhp);
1211 		return (-1);
1212 	}
1213 
1214 	if (missing) {
1215 		zpool_close(zhp);
1216 		*ret = NULL;
1217 		return (0);
1218 	}
1219 
1220 	*ret = zhp;
1221 	return (0);
1222 }
1223 
1224 /*
1225  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1226  * state.
1227  */
1228 zpool_handle_t *
1229 zpool_open(libzfs_handle_t *hdl, const char *pool)
1230 {
1231 	zpool_handle_t *zhp;
1232 
1233 	if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1234 		return (NULL);
1235 
1236 	if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1237 		(void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1238 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1239 		zpool_close(zhp);
1240 		return (NULL);
1241 	}
1242 
1243 	return (zhp);
1244 }
1245 
1246 /*
1247  * Close the handle.  Simply frees the memory associated with the handle.
1248  */
1249 void
1250 zpool_close(zpool_handle_t *zhp)
1251 {
1252 	nvlist_free(zhp->zpool_config);
1253 	nvlist_free(zhp->zpool_old_config);
1254 	nvlist_free(zhp->zpool_props);
1255 	free(zhp);
1256 }
1257 
1258 /*
1259  * Return the name of the pool.
1260  */
1261 const char *
1262 zpool_get_name(zpool_handle_t *zhp)
1263 {
1264 	return (zhp->zpool_name);
1265 }
1266 
1267 
1268 /*
1269  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1270  */
1271 int
1272 zpool_get_state(zpool_handle_t *zhp)
1273 {
1274 	return (zhp->zpool_state);
1275 }
1276 
1277 /*
1278  * Check if vdev list contains a special vdev
1279  */
1280 static boolean_t
1281 zpool_has_special_vdev(nvlist_t *nvroot)
1282 {
1283 	nvlist_t **child;
1284 	uint_t children;
1285 
1286 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &child,
1287 	    &children) == 0) {
1288 		for (uint_t c = 0; c < children; c++) {
1289 			char *bias;
1290 
1291 			if (nvlist_lookup_string(child[c],
1292 			    ZPOOL_CONFIG_ALLOCATION_BIAS, &bias) == 0 &&
1293 			    strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) {
1294 				return (B_TRUE);
1295 			}
1296 		}
1297 	}
1298 	return (B_FALSE);
1299 }
1300 
1301 /*
1302  * Check if vdev list contains a dRAID vdev
1303  */
1304 static boolean_t
1305 zpool_has_draid_vdev(nvlist_t *nvroot)
1306 {
1307 	nvlist_t **child;
1308 	uint_t children;
1309 
1310 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1311 	    &child, &children) == 0) {
1312 		for (uint_t c = 0; c < children; c++) {
1313 			char *type;
1314 
1315 			if (nvlist_lookup_string(child[c],
1316 			    ZPOOL_CONFIG_TYPE, &type) == 0 &&
1317 			    strcmp(type, VDEV_TYPE_DRAID) == 0) {
1318 				return (B_TRUE);
1319 			}
1320 		}
1321 	}
1322 	return (B_FALSE);
1323 }
1324 
1325 /*
1326  * Output a dRAID top-level vdev name in to the provided buffer.
1327  */
1328 static char *
1329 zpool_draid_name(char *name, int len, uint64_t data, uint64_t parity,
1330     uint64_t spares, uint64_t children)
1331 {
1332 	snprintf(name, len, "%s%llu:%llud:%lluc:%llus",
1333 	    VDEV_TYPE_DRAID, (u_longlong_t)parity, (u_longlong_t)data,
1334 	    (u_longlong_t)children, (u_longlong_t)spares);
1335 
1336 	return (name);
1337 }
1338 
1339 /*
1340  * Return B_TRUE if the provided name is a dRAID spare name.
1341  */
1342 boolean_t
1343 zpool_is_draid_spare(const char *name)
1344 {
1345 	uint64_t spare_id, parity, vdev_id;
1346 
1347 	if (sscanf(name, VDEV_TYPE_DRAID "%llu-%llu-%llu",
1348 	    (u_longlong_t *)&parity, (u_longlong_t *)&vdev_id,
1349 	    (u_longlong_t *)&spare_id) == 3) {
1350 		return (B_TRUE);
1351 	}
1352 
1353 	return (B_FALSE);
1354 }
1355 
1356 /*
1357  * Create the named pool, using the provided vdev list.  It is assumed
1358  * that the consumer has already validated the contents of the nvlist, so we
1359  * don't have to worry about error semantics.
1360  */
1361 int
1362 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1363     nvlist_t *props, nvlist_t *fsprops)
1364 {
1365 	zfs_cmd_t zc = {"\0"};
1366 	nvlist_t *zc_fsprops = NULL;
1367 	nvlist_t *zc_props = NULL;
1368 	nvlist_t *hidden_args = NULL;
1369 	uint8_t *wkeydata = NULL;
1370 	uint_t wkeylen = 0;
1371 	char msg[1024];
1372 	int ret = -1;
1373 
1374 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1375 	    "cannot create '%s'"), pool);
1376 
1377 	if (!zpool_name_valid(hdl, B_FALSE, pool))
1378 		return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1379 
1380 	zcmd_write_conf_nvlist(hdl, &zc, nvroot);
1381 
1382 	if (props) {
1383 		prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1384 
1385 		if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1386 		    SPA_VERSION_1, flags, msg)) == NULL) {
1387 			goto create_failed;
1388 		}
1389 	}
1390 
1391 	if (fsprops) {
1392 		uint64_t zoned;
1393 		char *zonestr;
1394 
1395 		zoned = ((nvlist_lookup_string(fsprops,
1396 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1397 		    strcmp(zonestr, "on") == 0);
1398 
1399 		if ((zc_fsprops = zfs_valid_proplist(hdl, ZFS_TYPE_FILESYSTEM,
1400 		    fsprops, zoned, NULL, NULL, B_TRUE, msg)) == NULL) {
1401 			goto create_failed;
1402 		}
1403 
1404 		if (nvlist_exists(zc_fsprops,
1405 		    zfs_prop_to_name(ZFS_PROP_SPECIAL_SMALL_BLOCKS)) &&
1406 		    !zpool_has_special_vdev(nvroot)) {
1407 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1408 			    "%s property requires a special vdev"),
1409 			    zfs_prop_to_name(ZFS_PROP_SPECIAL_SMALL_BLOCKS));
1410 			(void) zfs_error(hdl, EZFS_BADPROP, msg);
1411 			goto create_failed;
1412 		}
1413 
1414 		if (!zc_props &&
1415 		    (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1416 			goto create_failed;
1417 		}
1418 		if (zfs_crypto_create(hdl, NULL, zc_fsprops, props, B_TRUE,
1419 		    &wkeydata, &wkeylen) != 0) {
1420 			zfs_error(hdl, EZFS_CRYPTOFAILED, msg);
1421 			goto create_failed;
1422 		}
1423 		if (nvlist_add_nvlist(zc_props,
1424 		    ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1425 			goto create_failed;
1426 		}
1427 		if (wkeydata != NULL) {
1428 			if (nvlist_alloc(&hidden_args, NV_UNIQUE_NAME, 0) != 0)
1429 				goto create_failed;
1430 
1431 			if (nvlist_add_uint8_array(hidden_args, "wkeydata",
1432 			    wkeydata, wkeylen) != 0)
1433 				goto create_failed;
1434 
1435 			if (nvlist_add_nvlist(zc_props, ZPOOL_HIDDEN_ARGS,
1436 			    hidden_args) != 0)
1437 				goto create_failed;
1438 		}
1439 	}
1440 
1441 	if (zc_props)
1442 		zcmd_write_src_nvlist(hdl, &zc, zc_props);
1443 
1444 	(void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1445 
1446 	if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1447 
1448 		zcmd_free_nvlists(&zc);
1449 		nvlist_free(zc_props);
1450 		nvlist_free(zc_fsprops);
1451 		nvlist_free(hidden_args);
1452 		if (wkeydata != NULL)
1453 			free(wkeydata);
1454 
1455 		switch (errno) {
1456 		case EBUSY:
1457 			/*
1458 			 * This can happen if the user has specified the same
1459 			 * device multiple times.  We can't reliably detect this
1460 			 * until we try to add it and see we already have a
1461 			 * label.  This can also happen under if the device is
1462 			 * part of an active md or lvm device.
1463 			 */
1464 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1465 			    "one or more vdevs refer to the same device, or "
1466 			    "one of\nthe devices is part of an active md or "
1467 			    "lvm device"));
1468 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1469 
1470 		case ERANGE:
1471 			/*
1472 			 * This happens if the record size is smaller or larger
1473 			 * than the allowed size range, or not a power of 2.
1474 			 *
1475 			 * NOTE: although zfs_valid_proplist is called earlier,
1476 			 * this case may have slipped through since the
1477 			 * pool does not exist yet and it is therefore
1478 			 * impossible to read properties e.g. max blocksize
1479 			 * from the pool.
1480 			 */
1481 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1482 			    "record size invalid"));
1483 			return (zfs_error(hdl, EZFS_BADPROP, msg));
1484 
1485 		case EOVERFLOW:
1486 			/*
1487 			 * This occurs when one of the devices is below
1488 			 * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1489 			 * device was the problem device since there's no
1490 			 * reliable way to determine device size from userland.
1491 			 */
1492 			{
1493 				char buf[64];
1494 
1495 				zfs_nicebytes(SPA_MINDEVSIZE, buf,
1496 				    sizeof (buf));
1497 
1498 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1499 				    "one or more devices is less than the "
1500 				    "minimum size (%s)"), buf);
1501 			}
1502 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1503 
1504 		case ENOSPC:
1505 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1506 			    "one or more devices is out of space"));
1507 			return (zfs_error(hdl, EZFS_BADDEV, msg));
1508 
1509 		case EINVAL:
1510 			if (zpool_has_draid_vdev(nvroot) &&
1511 			    zfeature_lookup_name("draid", NULL) != 0) {
1512 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1513 				    "dRAID vdevs are unsupported by the "
1514 				    "kernel"));
1515 				return (zfs_error(hdl, EZFS_BADDEV, msg));
1516 			} else {
1517 				return (zpool_standard_error(hdl, errno, msg));
1518 			}
1519 
1520 		default:
1521 			return (zpool_standard_error(hdl, errno, msg));
1522 		}
1523 	}
1524 
1525 create_failed:
1526 	zcmd_free_nvlists(&zc);
1527 	nvlist_free(zc_props);
1528 	nvlist_free(zc_fsprops);
1529 	nvlist_free(hidden_args);
1530 	if (wkeydata != NULL)
1531 		free(wkeydata);
1532 	return (ret);
1533 }
1534 
1535 /*
1536  * Destroy the given pool.  It is up to the caller to ensure that there are no
1537  * datasets left in the pool.
1538  */
1539 int
1540 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1541 {
1542 	zfs_cmd_t zc = {"\0"};
1543 	zfs_handle_t *zfp = NULL;
1544 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1545 	char msg[1024];
1546 
1547 	if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1548 	    (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1549 		return (-1);
1550 
1551 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1552 	zc.zc_history = (uint64_t)(uintptr_t)log_str;
1553 
1554 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1555 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1556 		    "cannot destroy '%s'"), zhp->zpool_name);
1557 
1558 		if (errno == EROFS) {
1559 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1560 			    "one or more devices is read only"));
1561 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1562 		} else {
1563 			(void) zpool_standard_error(hdl, errno, msg);
1564 		}
1565 
1566 		if (zfp)
1567 			zfs_close(zfp);
1568 		return (-1);
1569 	}
1570 
1571 	if (zfp) {
1572 		remove_mountpoint(zfp);
1573 		zfs_close(zfp);
1574 	}
1575 
1576 	return (0);
1577 }
1578 
1579 /*
1580  * Create a checkpoint in the given pool.
1581  */
1582 int
1583 zpool_checkpoint(zpool_handle_t *zhp)
1584 {
1585 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1586 	char msg[1024];
1587 	int error;
1588 
1589 	error = lzc_pool_checkpoint(zhp->zpool_name);
1590 	if (error != 0) {
1591 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1592 		    "cannot checkpoint '%s'"), zhp->zpool_name);
1593 		(void) zpool_standard_error(hdl, error, msg);
1594 		return (-1);
1595 	}
1596 
1597 	return (0);
1598 }
1599 
1600 /*
1601  * Discard the checkpoint from the given pool.
1602  */
1603 int
1604 zpool_discard_checkpoint(zpool_handle_t *zhp)
1605 {
1606 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1607 	char msg[1024];
1608 	int error;
1609 
1610 	error = lzc_pool_checkpoint_discard(zhp->zpool_name);
1611 	if (error != 0) {
1612 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1613 		    "cannot discard checkpoint in '%s'"), zhp->zpool_name);
1614 		(void) zpool_standard_error(hdl, error, msg);
1615 		return (-1);
1616 	}
1617 
1618 	return (0);
1619 }
1620 
1621 /*
1622  * Add the given vdevs to the pool.  The caller must have already performed the
1623  * necessary verification to ensure that the vdev specification is well-formed.
1624  */
1625 int
1626 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1627 {
1628 	zfs_cmd_t zc = {"\0"};
1629 	int ret;
1630 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1631 	char msg[1024];
1632 	nvlist_t **spares, **l2cache;
1633 	uint_t nspares, nl2cache;
1634 
1635 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1636 	    "cannot add to '%s'"), zhp->zpool_name);
1637 
1638 	if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1639 	    SPA_VERSION_SPARES &&
1640 	    nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1641 	    &spares, &nspares) == 0) {
1642 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1643 		    "upgraded to add hot spares"));
1644 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
1645 	}
1646 
1647 	if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1648 	    SPA_VERSION_L2CACHE &&
1649 	    nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1650 	    &l2cache, &nl2cache) == 0) {
1651 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1652 		    "upgraded to add cache devices"));
1653 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
1654 	}
1655 
1656 	zcmd_write_conf_nvlist(hdl, &zc, nvroot);
1657 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1658 
1659 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1660 		switch (errno) {
1661 		case EBUSY:
1662 			/*
1663 			 * This can happen if the user has specified the same
1664 			 * device multiple times.  We can't reliably detect this
1665 			 * until we try to add it and see we already have a
1666 			 * label.
1667 			 */
1668 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1669 			    "one or more vdevs refer to the same device"));
1670 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1671 			break;
1672 
1673 		case EINVAL:
1674 
1675 			if (zpool_has_draid_vdev(nvroot) &&
1676 			    zfeature_lookup_name("draid", NULL) != 0) {
1677 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1678 				    "dRAID vdevs are unsupported by the "
1679 				    "kernel"));
1680 			} else {
1681 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1682 				    "invalid config; a pool with removing/"
1683 				    "removed vdevs does not support adding "
1684 				    "raidz or dRAID vdevs"));
1685 			}
1686 
1687 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1688 			break;
1689 
1690 		case EOVERFLOW:
1691 			/*
1692 			 * This occurs when one of the devices is below
1693 			 * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1694 			 * device was the problem device since there's no
1695 			 * reliable way to determine device size from userland.
1696 			 */
1697 			{
1698 				char buf[64];
1699 
1700 				zfs_nicebytes(SPA_MINDEVSIZE, buf,
1701 				    sizeof (buf));
1702 
1703 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1704 				    "device is less than the minimum "
1705 				    "size (%s)"), buf);
1706 			}
1707 			(void) zfs_error(hdl, EZFS_BADDEV, msg);
1708 			break;
1709 
1710 		case ENOTSUP:
1711 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1712 			    "pool must be upgraded to add these vdevs"));
1713 			(void) zfs_error(hdl, EZFS_BADVERSION, msg);
1714 			break;
1715 
1716 		default:
1717 			(void) zpool_standard_error(hdl, errno, msg);
1718 		}
1719 
1720 		ret = -1;
1721 	} else {
1722 		ret = 0;
1723 	}
1724 
1725 	zcmd_free_nvlists(&zc);
1726 
1727 	return (ret);
1728 }
1729 
1730 /*
1731  * Exports the pool from the system.  The caller must ensure that there are no
1732  * mounted datasets in the pool.
1733  */
1734 static int
1735 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1736     const char *log_str)
1737 {
1738 	zfs_cmd_t zc = {"\0"};
1739 
1740 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1741 	zc.zc_cookie = force;
1742 	zc.zc_guid = hardforce;
1743 	zc.zc_history = (uint64_t)(uintptr_t)log_str;
1744 
1745 	if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1746 		switch (errno) {
1747 		case EXDEV:
1748 			zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1749 			    "use '-f' to override the following errors:\n"
1750 			    "'%s' has an active shared spare which could be"
1751 			    " used by other pools once '%s' is exported."),
1752 			    zhp->zpool_name, zhp->zpool_name);
1753 			return (zfs_error_fmt(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1754 			    dgettext(TEXT_DOMAIN, "cannot export '%s'"),
1755 			    zhp->zpool_name));
1756 		default:
1757 			return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1758 			    dgettext(TEXT_DOMAIN, "cannot export '%s'"),
1759 			    zhp->zpool_name));
1760 		}
1761 	}
1762 
1763 	return (0);
1764 }
1765 
1766 int
1767 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1768 {
1769 	return (zpool_export_common(zhp, force, B_FALSE, log_str));
1770 }
1771 
1772 int
1773 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1774 {
1775 	return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1776 }
1777 
1778 static void
1779 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1780     nvlist_t *config)
1781 {
1782 	nvlist_t *nv = NULL;
1783 	uint64_t rewindto;
1784 	int64_t loss = -1;
1785 	struct tm t;
1786 	char timestr[128];
1787 
1788 	if (!hdl->libzfs_printerr || config == NULL)
1789 		return;
1790 
1791 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1792 	    nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1793 		return;
1794 	}
1795 
1796 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1797 		return;
1798 	(void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1799 
1800 	if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1801 	    strftime(timestr, 128, "%c", &t) != 0) {
1802 		if (dryrun) {
1803 			(void) printf(dgettext(TEXT_DOMAIN,
1804 			    "Would be able to return %s "
1805 			    "to its state as of %s.\n"),
1806 			    name, timestr);
1807 		} else {
1808 			(void) printf(dgettext(TEXT_DOMAIN,
1809 			    "Pool %s returned to its state as of %s.\n"),
1810 			    name, timestr);
1811 		}
1812 		if (loss > 120) {
1813 			(void) printf(dgettext(TEXT_DOMAIN,
1814 			    "%s approximately %lld "),
1815 			    dryrun ? "Would discard" : "Discarded",
1816 			    ((longlong_t)loss + 30) / 60);
1817 			(void) printf(dgettext(TEXT_DOMAIN,
1818 			    "minutes of transactions.\n"));
1819 		} else if (loss > 0) {
1820 			(void) printf(dgettext(TEXT_DOMAIN,
1821 			    "%s approximately %lld "),
1822 			    dryrun ? "Would discard" : "Discarded",
1823 			    (longlong_t)loss);
1824 			(void) printf(dgettext(TEXT_DOMAIN,
1825 			    "seconds of transactions.\n"));
1826 		}
1827 	}
1828 }
1829 
1830 void
1831 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1832     nvlist_t *config)
1833 {
1834 	nvlist_t *nv = NULL;
1835 	int64_t loss = -1;
1836 	uint64_t edata = UINT64_MAX;
1837 	uint64_t rewindto;
1838 	struct tm t;
1839 	char timestr[128];
1840 
1841 	if (!hdl->libzfs_printerr)
1842 		return;
1843 
1844 	if (reason >= 0)
1845 		(void) printf(dgettext(TEXT_DOMAIN, "action: "));
1846 	else
1847 		(void) printf(dgettext(TEXT_DOMAIN, "\t"));
1848 
1849 	/* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1850 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1851 	    nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1852 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1853 		goto no_info;
1854 
1855 	(void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1856 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1857 	    &edata);
1858 
1859 	(void) printf(dgettext(TEXT_DOMAIN,
1860 	    "Recovery is possible, but will result in some data loss.\n"));
1861 
1862 	if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1863 	    strftime(timestr, 128, "%c", &t) != 0) {
1864 		(void) printf(dgettext(TEXT_DOMAIN,
1865 		    "\tReturning the pool to its state as of %s\n"
1866 		    "\tshould correct the problem.  "),
1867 		    timestr);
1868 	} else {
1869 		(void) printf(dgettext(TEXT_DOMAIN,
1870 		    "\tReverting the pool to an earlier state "
1871 		    "should correct the problem.\n\t"));
1872 	}
1873 
1874 	if (loss > 120) {
1875 		(void) printf(dgettext(TEXT_DOMAIN,
1876 		    "Approximately %lld minutes of data\n"
1877 		    "\tmust be discarded, irreversibly.  "),
1878 		    ((longlong_t)loss + 30) / 60);
1879 	} else if (loss > 0) {
1880 		(void) printf(dgettext(TEXT_DOMAIN,
1881 		    "Approximately %lld seconds of data\n"
1882 		    "\tmust be discarded, irreversibly.  "),
1883 		    (longlong_t)loss);
1884 	}
1885 	if (edata != 0 && edata != UINT64_MAX) {
1886 		if (edata == 1) {
1887 			(void) printf(dgettext(TEXT_DOMAIN,
1888 			    "After rewind, at least\n"
1889 			    "\tone persistent user-data error will remain.  "));
1890 		} else {
1891 			(void) printf(dgettext(TEXT_DOMAIN,
1892 			    "After rewind, several\n"
1893 			    "\tpersistent user-data errors will remain.  "));
1894 		}
1895 	}
1896 	(void) printf(dgettext(TEXT_DOMAIN,
1897 	    "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1898 	    reason >= 0 ? "clear" : "import", name);
1899 
1900 	(void) printf(dgettext(TEXT_DOMAIN,
1901 	    "A scrub of the pool\n"
1902 	    "\tis strongly recommended after recovery.\n"));
1903 	return;
1904 
1905 no_info:
1906 	(void) printf(dgettext(TEXT_DOMAIN,
1907 	    "Destroy and re-create the pool from\n\ta backup source.\n"));
1908 }
1909 
1910 /*
1911  * zpool_import() is a contracted interface. Should be kept the same
1912  * if possible.
1913  *
1914  * Applications should use zpool_import_props() to import a pool with
1915  * new properties value to be set.
1916  */
1917 int
1918 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1919     char *altroot)
1920 {
1921 	nvlist_t *props = NULL;
1922 	int ret;
1923 
1924 	if (altroot != NULL) {
1925 		if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1926 			return (zfs_error_fmt(hdl, EZFS_NOMEM,
1927 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1928 			    newname));
1929 		}
1930 
1931 		if (nvlist_add_string(props,
1932 		    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1933 		    nvlist_add_string(props,
1934 		    zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1935 			nvlist_free(props);
1936 			return (zfs_error_fmt(hdl, EZFS_NOMEM,
1937 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1938 			    newname));
1939 		}
1940 	}
1941 
1942 	ret = zpool_import_props(hdl, config, newname, props,
1943 	    ZFS_IMPORT_NORMAL);
1944 	nvlist_free(props);
1945 	return (ret);
1946 }
1947 
1948 static void
1949 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1950     int indent)
1951 {
1952 	nvlist_t **child;
1953 	uint_t c, children;
1954 	char *vname;
1955 	uint64_t is_log = 0;
1956 
1957 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1958 	    &is_log);
1959 
1960 	if (name != NULL)
1961 		(void) printf("\t%*s%s%s\n", indent, "", name,
1962 		    is_log ? " [log]" : "");
1963 
1964 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1965 	    &child, &children) != 0)
1966 		return;
1967 
1968 	for (c = 0; c < children; c++) {
1969 		vname = zpool_vdev_name(hdl, NULL, child[c], VDEV_NAME_TYPE_ID);
1970 		print_vdev_tree(hdl, vname, child[c], indent + 2);
1971 		free(vname);
1972 	}
1973 }
1974 
1975 void
1976 zpool_print_unsup_feat(nvlist_t *config)
1977 {
1978 	nvlist_t *nvinfo, *unsup_feat;
1979 
1980 	nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
1981 	unsup_feat = fnvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT);
1982 
1983 	for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL);
1984 	    nvp != NULL; nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1985 		char *desc = fnvpair_value_string(nvp);
1986 		if (strlen(desc) > 0)
1987 			(void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1988 		else
1989 			(void) printf("\t%s\n", nvpair_name(nvp));
1990 	}
1991 }
1992 
1993 /*
1994  * Import the given pool using the known configuration and a list of
1995  * properties to be set. The configuration should have come from
1996  * zpool_find_import(). The 'newname' parameters control whether the pool
1997  * is imported with a different name.
1998  */
1999 int
2000 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
2001     nvlist_t *props, int flags)
2002 {
2003 	zfs_cmd_t zc = {"\0"};
2004 	zpool_load_policy_t policy;
2005 	nvlist_t *nv = NULL;
2006 	nvlist_t *nvinfo = NULL;
2007 	nvlist_t *missing = NULL;
2008 	const char *thename;
2009 	char *origname;
2010 	int ret;
2011 	int error = 0;
2012 	char errbuf[1024];
2013 
2014 	origname = fnvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME);
2015 
2016 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2017 	    "cannot import pool '%s'"), origname);
2018 
2019 	if (newname != NULL) {
2020 		if (!zpool_name_valid(hdl, B_FALSE, newname))
2021 			return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
2022 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
2023 			    newname));
2024 		thename = newname;
2025 	} else {
2026 		thename = origname;
2027 	}
2028 
2029 	if (props != NULL) {
2030 		uint64_t version;
2031 		prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2032 
2033 		version = fnvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION);
2034 
2035 		if ((props = zpool_valid_proplist(hdl, origname,
2036 		    props, version, flags, errbuf)) == NULL)
2037 			return (-1);
2038 		zcmd_write_src_nvlist(hdl, &zc, props);
2039 		nvlist_free(props);
2040 	}
2041 
2042 	(void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
2043 
2044 	zc.zc_guid = fnvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID);
2045 
2046 	zcmd_write_conf_nvlist(hdl, &zc, config);
2047 	zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2);
2048 
2049 	zc.zc_cookie = flags;
2050 	while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
2051 	    errno == ENOMEM)
2052 		zcmd_expand_dst_nvlist(hdl, &zc);
2053 	if (ret != 0)
2054 		error = errno;
2055 
2056 	(void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
2057 
2058 	zcmd_free_nvlists(&zc);
2059 
2060 	zpool_get_load_policy(config, &policy);
2061 
2062 	if (error) {
2063 		char desc[1024];
2064 		char aux[256];
2065 
2066 		/*
2067 		 * Dry-run failed, but we print out what success
2068 		 * looks like if we found a best txg
2069 		 */
2070 		if (policy.zlp_rewind & ZPOOL_TRY_REWIND) {
2071 			zpool_rewind_exclaim(hdl, newname ? origname : thename,
2072 			    B_TRUE, nv);
2073 			nvlist_free(nv);
2074 			return (-1);
2075 		}
2076 
2077 		if (newname == NULL)
2078 			(void) snprintf(desc, sizeof (desc),
2079 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
2080 			    thename);
2081 		else
2082 			(void) snprintf(desc, sizeof (desc),
2083 			    dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
2084 			    origname, thename);
2085 
2086 		switch (error) {
2087 		case ENOTSUP:
2088 			if (nv != NULL && nvlist_lookup_nvlist(nv,
2089 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
2090 			    nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
2091 				(void) printf(dgettext(TEXT_DOMAIN, "This "
2092 				    "pool uses the following feature(s) not "
2093 				    "supported by this system:\n"));
2094 				zpool_print_unsup_feat(nv);
2095 				if (nvlist_exists(nvinfo,
2096 				    ZPOOL_CONFIG_CAN_RDONLY)) {
2097 					(void) printf(dgettext(TEXT_DOMAIN,
2098 					    "All unsupported features are only "
2099 					    "required for writing to the pool."
2100 					    "\nThe pool can be imported using "
2101 					    "'-o readonly=on'.\n"));
2102 				}
2103 			}
2104 			/*
2105 			 * Unsupported version.
2106 			 */
2107 			(void) zfs_error(hdl, EZFS_BADVERSION, desc);
2108 			break;
2109 
2110 		case EREMOTEIO:
2111 			if (nv != NULL && nvlist_lookup_nvlist(nv,
2112 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0) {
2113 				char *hostname = "<unknown>";
2114 				uint64_t hostid = 0;
2115 				mmp_state_t mmp_state;
2116 
2117 				mmp_state = fnvlist_lookup_uint64(nvinfo,
2118 				    ZPOOL_CONFIG_MMP_STATE);
2119 
2120 				if (nvlist_exists(nvinfo,
2121 				    ZPOOL_CONFIG_MMP_HOSTNAME))
2122 					hostname = fnvlist_lookup_string(nvinfo,
2123 					    ZPOOL_CONFIG_MMP_HOSTNAME);
2124 
2125 				if (nvlist_exists(nvinfo,
2126 				    ZPOOL_CONFIG_MMP_HOSTID))
2127 					hostid = fnvlist_lookup_uint64(nvinfo,
2128 					    ZPOOL_CONFIG_MMP_HOSTID);
2129 
2130 				if (mmp_state == MMP_STATE_ACTIVE) {
2131 					(void) snprintf(aux, sizeof (aux),
2132 					    dgettext(TEXT_DOMAIN, "pool is imp"
2133 					    "orted on host '%s' (hostid=%lx).\n"
2134 					    "Export the pool on the other "
2135 					    "system, then run 'zpool import'."),
2136 					    hostname, (unsigned long) hostid);
2137 				} else if (mmp_state == MMP_STATE_NO_HOSTID) {
2138 					(void) snprintf(aux, sizeof (aux),
2139 					    dgettext(TEXT_DOMAIN, "pool has "
2140 					    "the multihost property on and "
2141 					    "the\nsystem's hostid is not set. "
2142 					    "Set a unique system hostid with "
2143 					    "the zgenhostid(8) command.\n"));
2144 				}
2145 
2146 				(void) zfs_error_aux(hdl, "%s", aux);
2147 			}
2148 			(void) zfs_error(hdl, EZFS_ACTIVE_POOL, desc);
2149 			break;
2150 
2151 		case EINVAL:
2152 			(void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
2153 			break;
2154 
2155 		case EROFS:
2156 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2157 			    "one or more devices is read only"));
2158 			(void) zfs_error(hdl, EZFS_BADDEV, desc);
2159 			break;
2160 
2161 		case ENXIO:
2162 			if (nv && nvlist_lookup_nvlist(nv,
2163 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
2164 			    nvlist_lookup_nvlist(nvinfo,
2165 			    ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
2166 				(void) printf(dgettext(TEXT_DOMAIN,
2167 				    "The devices below are missing or "
2168 				    "corrupted, use '-m' to import the pool "
2169 				    "anyway:\n"));
2170 				print_vdev_tree(hdl, NULL, missing, 2);
2171 				(void) printf("\n");
2172 			}
2173 			(void) zpool_standard_error(hdl, error, desc);
2174 			break;
2175 
2176 		case EEXIST:
2177 			(void) zpool_standard_error(hdl, error, desc);
2178 			break;
2179 
2180 		case EBUSY:
2181 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2182 			    "one or more devices are already in use\n"));
2183 			(void) zfs_error(hdl, EZFS_BADDEV, desc);
2184 			break;
2185 		case ENAMETOOLONG:
2186 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2187 			    "new name of at least one dataset is longer than "
2188 			    "the maximum allowable length"));
2189 			(void) zfs_error(hdl, EZFS_NAMETOOLONG, desc);
2190 			break;
2191 		default:
2192 			(void) zpool_standard_error(hdl, error, desc);
2193 			zpool_explain_recover(hdl,
2194 			    newname ? origname : thename, -error, nv);
2195 			break;
2196 		}
2197 
2198 		nvlist_free(nv);
2199 		ret = -1;
2200 	} else {
2201 		zpool_handle_t *zhp;
2202 
2203 		/*
2204 		 * This should never fail, but play it safe anyway.
2205 		 */
2206 		if (zpool_open_silent(hdl, thename, &zhp) != 0)
2207 			ret = -1;
2208 		else if (zhp != NULL)
2209 			zpool_close(zhp);
2210 		if (policy.zlp_rewind &
2211 		    (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
2212 			zpool_rewind_exclaim(hdl, newname ? origname : thename,
2213 			    ((policy.zlp_rewind & ZPOOL_TRY_REWIND) != 0), nv);
2214 		}
2215 		nvlist_free(nv);
2216 		return (0);
2217 	}
2218 
2219 	return (ret);
2220 }
2221 
2222 /*
2223  * Translate vdev names to guids.  If a vdev_path is determined to be
2224  * unsuitable then a vd_errlist is allocated and the vdev path and errno
2225  * are added to it.
2226  */
2227 static int
2228 zpool_translate_vdev_guids(zpool_handle_t *zhp, nvlist_t *vds,
2229     nvlist_t *vdev_guids, nvlist_t *guids_to_paths, nvlist_t **vd_errlist)
2230 {
2231 	nvlist_t *errlist = NULL;
2232 	int error = 0;
2233 
2234 	for (nvpair_t *elem = nvlist_next_nvpair(vds, NULL); elem != NULL;
2235 	    elem = nvlist_next_nvpair(vds, elem)) {
2236 		boolean_t spare, cache;
2237 
2238 		char *vd_path = nvpair_name(elem);
2239 		nvlist_t *tgt = zpool_find_vdev(zhp, vd_path, &spare, &cache,
2240 		    NULL);
2241 
2242 		if ((tgt == NULL) || cache || spare) {
2243 			if (errlist == NULL) {
2244 				errlist = fnvlist_alloc();
2245 				error = EINVAL;
2246 			}
2247 
2248 			uint64_t err = (tgt == NULL) ? EZFS_NODEVICE :
2249 			    (spare ? EZFS_ISSPARE : EZFS_ISL2CACHE);
2250 			fnvlist_add_int64(errlist, vd_path, err);
2251 			continue;
2252 		}
2253 
2254 		uint64_t guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
2255 		fnvlist_add_uint64(vdev_guids, vd_path, guid);
2256 
2257 		char msg[MAXNAMELEN];
2258 		(void) snprintf(msg, sizeof (msg), "%llu", (u_longlong_t)guid);
2259 		fnvlist_add_string(guids_to_paths, msg, vd_path);
2260 	}
2261 
2262 	if (error != 0) {
2263 		verify(errlist != NULL);
2264 		if (vd_errlist != NULL)
2265 			*vd_errlist = errlist;
2266 		else
2267 			fnvlist_free(errlist);
2268 	}
2269 
2270 	return (error);
2271 }
2272 
2273 static int
2274 xlate_init_err(int err)
2275 {
2276 	switch (err) {
2277 	case ENODEV:
2278 		return (EZFS_NODEVICE);
2279 	case EINVAL:
2280 	case EROFS:
2281 		return (EZFS_BADDEV);
2282 	case EBUSY:
2283 		return (EZFS_INITIALIZING);
2284 	case ESRCH:
2285 		return (EZFS_NO_INITIALIZE);
2286 	}
2287 	return (err);
2288 }
2289 
2290 /*
2291  * Begin, suspend, or cancel the initialization (initializing of all free
2292  * blocks) for the given vdevs in the given pool.
2293  */
2294 static int
2295 zpool_initialize_impl(zpool_handle_t *zhp, pool_initialize_func_t cmd_type,
2296     nvlist_t *vds, boolean_t wait)
2297 {
2298 	int err;
2299 
2300 	nvlist_t *vdev_guids = fnvlist_alloc();
2301 	nvlist_t *guids_to_paths = fnvlist_alloc();
2302 	nvlist_t *vd_errlist = NULL;
2303 	nvlist_t *errlist;
2304 	nvpair_t *elem;
2305 
2306 	err = zpool_translate_vdev_guids(zhp, vds, vdev_guids,
2307 	    guids_to_paths, &vd_errlist);
2308 
2309 	if (err != 0) {
2310 		verify(vd_errlist != NULL);
2311 		goto list_errors;
2312 	}
2313 
2314 	err = lzc_initialize(zhp->zpool_name, cmd_type,
2315 	    vdev_guids, &errlist);
2316 
2317 	if (err != 0) {
2318 		if (errlist != NULL) {
2319 			vd_errlist = fnvlist_lookup_nvlist(errlist,
2320 			    ZPOOL_INITIALIZE_VDEVS);
2321 			goto list_errors;
2322 		}
2323 		(void) zpool_standard_error(zhp->zpool_hdl, err,
2324 		    dgettext(TEXT_DOMAIN, "operation failed"));
2325 		goto out;
2326 	}
2327 
2328 	if (wait) {
2329 		for (elem = nvlist_next_nvpair(vdev_guids, NULL); elem != NULL;
2330 		    elem = nvlist_next_nvpair(vdev_guids, elem)) {
2331 
2332 			uint64_t guid = fnvpair_value_uint64(elem);
2333 
2334 			err = lzc_wait_tag(zhp->zpool_name,
2335 			    ZPOOL_WAIT_INITIALIZE, guid, NULL);
2336 			if (err != 0) {
2337 				(void) zpool_standard_error_fmt(zhp->zpool_hdl,
2338 				    err, dgettext(TEXT_DOMAIN, "error "
2339 				    "waiting for '%s' to initialize"),
2340 				    nvpair_name(elem));
2341 
2342 				goto out;
2343 			}
2344 		}
2345 	}
2346 	goto out;
2347 
2348 list_errors:
2349 	for (elem = nvlist_next_nvpair(vd_errlist, NULL); elem != NULL;
2350 	    elem = nvlist_next_nvpair(vd_errlist, elem)) {
2351 		int64_t vd_error = xlate_init_err(fnvpair_value_int64(elem));
2352 		char *path;
2353 
2354 		if (nvlist_lookup_string(guids_to_paths, nvpair_name(elem),
2355 		    &path) != 0)
2356 			path = nvpair_name(elem);
2357 
2358 		(void) zfs_error_fmt(zhp->zpool_hdl, vd_error,
2359 		    "cannot initialize '%s'", path);
2360 	}
2361 
2362 out:
2363 	fnvlist_free(vdev_guids);
2364 	fnvlist_free(guids_to_paths);
2365 
2366 	if (vd_errlist != NULL)
2367 		fnvlist_free(vd_errlist);
2368 
2369 	return (err == 0 ? 0 : -1);
2370 }
2371 
2372 int
2373 zpool_initialize(zpool_handle_t *zhp, pool_initialize_func_t cmd_type,
2374     nvlist_t *vds)
2375 {
2376 	return (zpool_initialize_impl(zhp, cmd_type, vds, B_FALSE));
2377 }
2378 
2379 int
2380 zpool_initialize_wait(zpool_handle_t *zhp, pool_initialize_func_t cmd_type,
2381     nvlist_t *vds)
2382 {
2383 	return (zpool_initialize_impl(zhp, cmd_type, vds, B_TRUE));
2384 }
2385 
2386 static int
2387 xlate_trim_err(int err)
2388 {
2389 	switch (err) {
2390 	case ENODEV:
2391 		return (EZFS_NODEVICE);
2392 	case EINVAL:
2393 	case EROFS:
2394 		return (EZFS_BADDEV);
2395 	case EBUSY:
2396 		return (EZFS_TRIMMING);
2397 	case ESRCH:
2398 		return (EZFS_NO_TRIM);
2399 	case EOPNOTSUPP:
2400 		return (EZFS_TRIM_NOTSUP);
2401 	}
2402 	return (err);
2403 }
2404 
2405 static int
2406 zpool_trim_wait(zpool_handle_t *zhp, nvlist_t *vdev_guids)
2407 {
2408 	int err;
2409 	nvpair_t *elem;
2410 
2411 	for (elem = nvlist_next_nvpair(vdev_guids, NULL); elem != NULL;
2412 	    elem = nvlist_next_nvpair(vdev_guids, elem)) {
2413 
2414 		uint64_t guid = fnvpair_value_uint64(elem);
2415 
2416 		err = lzc_wait_tag(zhp->zpool_name,
2417 		    ZPOOL_WAIT_TRIM, guid, NULL);
2418 		if (err != 0) {
2419 			(void) zpool_standard_error_fmt(zhp->zpool_hdl,
2420 			    err, dgettext(TEXT_DOMAIN, "error "
2421 			    "waiting to trim '%s'"), nvpair_name(elem));
2422 
2423 			return (err);
2424 		}
2425 	}
2426 	return (0);
2427 }
2428 
2429 /*
2430  * Check errlist and report any errors, omitting ones which should be
2431  * suppressed. Returns B_TRUE if any errors were reported.
2432  */
2433 static boolean_t
2434 check_trim_errs(zpool_handle_t *zhp, trimflags_t *trim_flags,
2435     nvlist_t *guids_to_paths, nvlist_t *vds, nvlist_t *errlist)
2436 {
2437 	nvpair_t *elem;
2438 	boolean_t reported_errs = B_FALSE;
2439 	int num_vds = 0;
2440 	int num_suppressed_errs = 0;
2441 
2442 	for (elem = nvlist_next_nvpair(vds, NULL);
2443 	    elem != NULL; elem = nvlist_next_nvpair(vds, elem)) {
2444 		num_vds++;
2445 	}
2446 
2447 	for (elem = nvlist_next_nvpair(errlist, NULL);
2448 	    elem != NULL; elem = nvlist_next_nvpair(errlist, elem)) {
2449 		int64_t vd_error = xlate_trim_err(fnvpair_value_int64(elem));
2450 		char *path;
2451 
2452 		/*
2453 		 * If only the pool was specified, and it was not a secure
2454 		 * trim then suppress warnings for individual vdevs which
2455 		 * do not support trimming.
2456 		 */
2457 		if (vd_error == EZFS_TRIM_NOTSUP &&
2458 		    trim_flags->fullpool &&
2459 		    !trim_flags->secure) {
2460 			num_suppressed_errs++;
2461 			continue;
2462 		}
2463 
2464 		reported_errs = B_TRUE;
2465 		if (nvlist_lookup_string(guids_to_paths, nvpair_name(elem),
2466 		    &path) != 0)
2467 			path = nvpair_name(elem);
2468 
2469 		(void) zfs_error_fmt(zhp->zpool_hdl, vd_error,
2470 		    "cannot trim '%s'", path);
2471 	}
2472 
2473 	if (num_suppressed_errs == num_vds) {
2474 		(void) zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
2475 		    "no devices in pool support trim operations"));
2476 		(void) (zfs_error(zhp->zpool_hdl, EZFS_TRIM_NOTSUP,
2477 		    dgettext(TEXT_DOMAIN, "cannot trim")));
2478 		reported_errs = B_TRUE;
2479 	}
2480 
2481 	return (reported_errs);
2482 }
2483 
2484 /*
2485  * Begin, suspend, or cancel the TRIM (discarding of all free blocks) for
2486  * the given vdevs in the given pool.
2487  */
2488 int
2489 zpool_trim(zpool_handle_t *zhp, pool_trim_func_t cmd_type, nvlist_t *vds,
2490     trimflags_t *trim_flags)
2491 {
2492 	int err;
2493 	int retval = 0;
2494 
2495 	nvlist_t *vdev_guids = fnvlist_alloc();
2496 	nvlist_t *guids_to_paths = fnvlist_alloc();
2497 	nvlist_t *errlist = NULL;
2498 
2499 	err = zpool_translate_vdev_guids(zhp, vds, vdev_guids,
2500 	    guids_to_paths, &errlist);
2501 	if (err != 0) {
2502 		check_trim_errs(zhp, trim_flags, guids_to_paths, vds, errlist);
2503 		retval = -1;
2504 		goto out;
2505 	}
2506 
2507 	err = lzc_trim(zhp->zpool_name, cmd_type, trim_flags->rate,
2508 	    trim_flags->secure, vdev_guids, &errlist);
2509 	if (err != 0) {
2510 		nvlist_t *vd_errlist;
2511 		if (errlist != NULL && nvlist_lookup_nvlist(errlist,
2512 		    ZPOOL_TRIM_VDEVS, &vd_errlist) == 0) {
2513 			if (check_trim_errs(zhp, trim_flags, guids_to_paths,
2514 			    vds, vd_errlist)) {
2515 				retval = -1;
2516 				goto out;
2517 			}
2518 		} else {
2519 			char msg[1024];
2520 
2521 			(void) snprintf(msg, sizeof (msg),
2522 			    dgettext(TEXT_DOMAIN, "operation failed"));
2523 			zpool_standard_error(zhp->zpool_hdl, err, msg);
2524 			retval = -1;
2525 			goto out;
2526 		}
2527 	}
2528 
2529 
2530 	if (trim_flags->wait)
2531 		retval = zpool_trim_wait(zhp, vdev_guids);
2532 
2533 out:
2534 	if (errlist != NULL)
2535 		fnvlist_free(errlist);
2536 	fnvlist_free(vdev_guids);
2537 	fnvlist_free(guids_to_paths);
2538 	return (retval);
2539 }
2540 
2541 /*
2542  * Scan the pool.
2543  */
2544 int
2545 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func, pool_scrub_cmd_t cmd)
2546 {
2547 	zfs_cmd_t zc = {"\0"};
2548 	char msg[1024];
2549 	int err;
2550 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2551 
2552 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2553 	zc.zc_cookie = func;
2554 	zc.zc_flags = cmd;
2555 
2556 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0)
2557 		return (0);
2558 
2559 	err = errno;
2560 
2561 	/* ECANCELED on a scrub means we resumed a paused scrub */
2562 	if (err == ECANCELED && func == POOL_SCAN_SCRUB &&
2563 	    cmd == POOL_SCRUB_NORMAL)
2564 		return (0);
2565 
2566 	if (err == ENOENT && func != POOL_SCAN_NONE && cmd == POOL_SCRUB_NORMAL)
2567 		return (0);
2568 
2569 	if (func == POOL_SCAN_SCRUB) {
2570 		if (cmd == POOL_SCRUB_PAUSE) {
2571 			(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2572 			    "cannot pause scrubbing %s"), zc.zc_name);
2573 		} else {
2574 			assert(cmd == POOL_SCRUB_NORMAL);
2575 			(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2576 			    "cannot scrub %s"), zc.zc_name);
2577 		}
2578 	} else if (func == POOL_SCAN_RESILVER) {
2579 		assert(cmd == POOL_SCRUB_NORMAL);
2580 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2581 		    "cannot restart resilver on %s"), zc.zc_name);
2582 	} else if (func == POOL_SCAN_NONE) {
2583 		(void) snprintf(msg, sizeof (msg),
2584 		    dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
2585 		    zc.zc_name);
2586 	} else {
2587 		assert(!"unexpected result");
2588 	}
2589 
2590 	if (err == EBUSY) {
2591 		nvlist_t *nvroot;
2592 		pool_scan_stat_t *ps = NULL;
2593 		uint_t psc;
2594 
2595 		nvroot = fnvlist_lookup_nvlist(zhp->zpool_config,
2596 		    ZPOOL_CONFIG_VDEV_TREE);
2597 		(void) nvlist_lookup_uint64_array(nvroot,
2598 		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
2599 		if (ps && ps->pss_func == POOL_SCAN_SCRUB &&
2600 		    ps->pss_state == DSS_SCANNING) {
2601 			if (cmd == POOL_SCRUB_PAUSE)
2602 				return (zfs_error(hdl, EZFS_SCRUB_PAUSED, msg));
2603 			else
2604 				return (zfs_error(hdl, EZFS_SCRUBBING, msg));
2605 		} else {
2606 			return (zfs_error(hdl, EZFS_RESILVERING, msg));
2607 		}
2608 	} else if (err == ENOENT) {
2609 		return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
2610 	} else if (err == ENOTSUP && func == POOL_SCAN_RESILVER) {
2611 		return (zfs_error(hdl, EZFS_NO_RESILVER_DEFER, msg));
2612 	} else {
2613 		return (zpool_standard_error(hdl, err, msg));
2614 	}
2615 }
2616 
2617 /*
2618  * Find a vdev that matches the search criteria specified. We use the
2619  * the nvpair name to determine how we should look for the device.
2620  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
2621  * spare; but FALSE if its an INUSE spare.
2622  */
2623 static nvlist_t *
2624 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
2625     boolean_t *l2cache, boolean_t *log)
2626 {
2627 	uint_t c, children;
2628 	nvlist_t **child;
2629 	nvlist_t *ret;
2630 	uint64_t is_log;
2631 	char *srchkey;
2632 	nvpair_t *pair = nvlist_next_nvpair(search, NULL);
2633 
2634 	/* Nothing to look for */
2635 	if (search == NULL || pair == NULL)
2636 		return (NULL);
2637 
2638 	/* Obtain the key we will use to search */
2639 	srchkey = nvpair_name(pair);
2640 
2641 	switch (nvpair_type(pair)) {
2642 	case DATA_TYPE_UINT64:
2643 		if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
2644 			uint64_t srchval = fnvpair_value_uint64(pair);
2645 			uint64_t theguid = fnvlist_lookup_uint64(nv,
2646 			    ZPOOL_CONFIG_GUID);
2647 			if (theguid == srchval)
2648 				return (nv);
2649 		}
2650 		break;
2651 
2652 	case DATA_TYPE_STRING: {
2653 		char *srchval, *val;
2654 
2655 		srchval = fnvpair_value_string(pair);
2656 		if (nvlist_lookup_string(nv, srchkey, &val) != 0)
2657 			break;
2658 
2659 		/*
2660 		 * Search for the requested value. Special cases:
2661 		 *
2662 		 * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
2663 		 *   "-part1", or "p1".  The suffix is hidden from the user,
2664 		 *   but included in the string, so this matches around it.
2665 		 * - ZPOOL_CONFIG_PATH for short names zfs_strcmp_shortname()
2666 		 *   is used to check all possible expanded paths.
2667 		 * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
2668 		 *
2669 		 * Otherwise, all other searches are simple string compares.
2670 		 */
2671 		if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0) {
2672 			uint64_t wholedisk = 0;
2673 
2674 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2675 			    &wholedisk);
2676 			if (zfs_strcmp_pathname(srchval, val, wholedisk) == 0)
2677 				return (nv);
2678 
2679 		} else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2680 			char *type, *idx, *end, *p;
2681 			uint64_t id, vdev_id;
2682 
2683 			/*
2684 			 * Determine our vdev type, keeping in mind
2685 			 * that the srchval is composed of a type and
2686 			 * vdev id pair (i.e. mirror-4).
2687 			 */
2688 			if ((type = strdup(srchval)) == NULL)
2689 				return (NULL);
2690 
2691 			if ((p = strrchr(type, '-')) == NULL) {
2692 				free(type);
2693 				break;
2694 			}
2695 			idx = p + 1;
2696 			*p = '\0';
2697 
2698 			/*
2699 			 * If the types don't match then keep looking.
2700 			 */
2701 			if (strncmp(val, type, strlen(val)) != 0) {
2702 				free(type);
2703 				break;
2704 			}
2705 
2706 			verify(zpool_vdev_is_interior(type));
2707 
2708 			id = fnvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID);
2709 			errno = 0;
2710 			vdev_id = strtoull(idx, &end, 10);
2711 
2712 			/*
2713 			 * If we are looking for a raidz and a parity is
2714 			 * specified, make sure it matches.
2715 			 */
2716 			int rzlen = strlen(VDEV_TYPE_RAIDZ);
2717 			assert(rzlen == strlen(VDEV_TYPE_DRAID));
2718 			int typlen = strlen(type);
2719 			if ((strncmp(type, VDEV_TYPE_RAIDZ, rzlen) == 0 ||
2720 			    strncmp(type, VDEV_TYPE_DRAID, rzlen) == 0) &&
2721 			    typlen != rzlen) {
2722 				uint64_t vdev_parity;
2723 				int parity = *(type + rzlen) - '0';
2724 
2725 				if (parity <= 0 || parity > 3 ||
2726 				    (typlen - rzlen) != 1) {
2727 					/*
2728 					 * Nonsense parity specified, can
2729 					 * never match
2730 					 */
2731 					free(type);
2732 					return (NULL);
2733 				}
2734 				vdev_parity = fnvlist_lookup_uint64(nv,
2735 				    ZPOOL_CONFIG_NPARITY);
2736 				if ((int)vdev_parity != parity) {
2737 					free(type);
2738 					break;
2739 				}
2740 			}
2741 
2742 			free(type);
2743 			if (errno != 0)
2744 				return (NULL);
2745 
2746 			/*
2747 			 * Now verify that we have the correct vdev id.
2748 			 */
2749 			if (vdev_id == id)
2750 				return (nv);
2751 		}
2752 
2753 		/*
2754 		 * Common case
2755 		 */
2756 		if (strcmp(srchval, val) == 0)
2757 			return (nv);
2758 		break;
2759 	}
2760 
2761 	default:
2762 		break;
2763 	}
2764 
2765 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2766 	    &child, &children) != 0)
2767 		return (NULL);
2768 
2769 	for (c = 0; c < children; c++) {
2770 		if ((ret = vdev_to_nvlist_iter(child[c], search,
2771 		    avail_spare, l2cache, NULL)) != NULL) {
2772 			/*
2773 			 * The 'is_log' value is only set for the toplevel
2774 			 * vdev, not the leaf vdevs.  So we always lookup the
2775 			 * log device from the root of the vdev tree (where
2776 			 * 'log' is non-NULL).
2777 			 */
2778 			if (log != NULL &&
2779 			    nvlist_lookup_uint64(child[c],
2780 			    ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2781 			    is_log) {
2782 				*log = B_TRUE;
2783 			}
2784 			return (ret);
2785 		}
2786 	}
2787 
2788 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2789 	    &child, &children) == 0) {
2790 		for (c = 0; c < children; c++) {
2791 			if ((ret = vdev_to_nvlist_iter(child[c], search,
2792 			    avail_spare, l2cache, NULL)) != NULL) {
2793 				*avail_spare = B_TRUE;
2794 				return (ret);
2795 			}
2796 		}
2797 	}
2798 
2799 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2800 	    &child, &children) == 0) {
2801 		for (c = 0; c < children; c++) {
2802 			if ((ret = vdev_to_nvlist_iter(child[c], search,
2803 			    avail_spare, l2cache, NULL)) != NULL) {
2804 				*l2cache = B_TRUE;
2805 				return (ret);
2806 			}
2807 		}
2808 	}
2809 
2810 	return (NULL);
2811 }
2812 
2813 /*
2814  * Given a physical path or guid, find the associated vdev.
2815  */
2816 nvlist_t *
2817 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2818     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2819 {
2820 	nvlist_t *search, *nvroot, *ret;
2821 	uint64_t guid;
2822 	char *end;
2823 
2824 	search = fnvlist_alloc();
2825 
2826 	guid = strtoull(ppath, &end, 0);
2827 	if (guid != 0 && *end == '\0') {
2828 		fnvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid);
2829 	} else {
2830 		fnvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath);
2831 	}
2832 
2833 	nvroot = fnvlist_lookup_nvlist(zhp->zpool_config,
2834 	    ZPOOL_CONFIG_VDEV_TREE);
2835 
2836 	*avail_spare = B_FALSE;
2837 	*l2cache = B_FALSE;
2838 	if (log != NULL)
2839 		*log = B_FALSE;
2840 	ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2841 	fnvlist_free(search);
2842 
2843 	return (ret);
2844 }
2845 
2846 /*
2847  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2848  */
2849 static boolean_t
2850 zpool_vdev_is_interior(const char *name)
2851 {
2852 	if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2853 	    strncmp(name, VDEV_TYPE_SPARE, strlen(VDEV_TYPE_SPARE)) == 0 ||
2854 	    strncmp(name,
2855 	    VDEV_TYPE_REPLACING, strlen(VDEV_TYPE_REPLACING)) == 0 ||
2856 	    strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2857 		return (B_TRUE);
2858 
2859 	if (strncmp(name, VDEV_TYPE_DRAID, strlen(VDEV_TYPE_DRAID)) == 0 &&
2860 	    !zpool_is_draid_spare(name))
2861 		return (B_TRUE);
2862 
2863 	return (B_FALSE);
2864 }
2865 
2866 nvlist_t *
2867 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2868     boolean_t *l2cache, boolean_t *log)
2869 {
2870 	char *end;
2871 	nvlist_t *nvroot, *search, *ret;
2872 	uint64_t guid;
2873 
2874 	search = fnvlist_alloc();
2875 
2876 	guid = strtoull(path, &end, 0);
2877 	if (guid != 0 && *end == '\0') {
2878 		fnvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid);
2879 	} else if (zpool_vdev_is_interior(path)) {
2880 		fnvlist_add_string(search, ZPOOL_CONFIG_TYPE, path);
2881 	} else {
2882 		fnvlist_add_string(search, ZPOOL_CONFIG_PATH, path);
2883 	}
2884 
2885 	nvroot = fnvlist_lookup_nvlist(zhp->zpool_config,
2886 	    ZPOOL_CONFIG_VDEV_TREE);
2887 
2888 	*avail_spare = B_FALSE;
2889 	*l2cache = B_FALSE;
2890 	if (log != NULL)
2891 		*log = B_FALSE;
2892 	ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2893 	fnvlist_free(search);
2894 
2895 	return (ret);
2896 }
2897 
2898 static int
2899 vdev_is_online(nvlist_t *nv)
2900 {
2901 	uint64_t ival;
2902 
2903 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2904 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2905 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2906 		return (0);
2907 
2908 	return (1);
2909 }
2910 
2911 /*
2912  * Helper function for zpool_get_physpaths().
2913  */
2914 static int
2915 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2916     size_t *bytes_written)
2917 {
2918 	size_t bytes_left, pos, rsz;
2919 	char *tmppath;
2920 	const char *format;
2921 
2922 	if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2923 	    &tmppath) != 0)
2924 		return (EZFS_NODEVICE);
2925 
2926 	pos = *bytes_written;
2927 	bytes_left = physpath_size - pos;
2928 	format = (pos == 0) ? "%s" : " %s";
2929 
2930 	rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2931 	*bytes_written += rsz;
2932 
2933 	if (rsz >= bytes_left) {
2934 		/* if physpath was not copied properly, clear it */
2935 		if (bytes_left != 0) {
2936 			physpath[pos] = 0;
2937 		}
2938 		return (EZFS_NOSPC);
2939 	}
2940 	return (0);
2941 }
2942 
2943 static int
2944 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2945     size_t *rsz, boolean_t is_spare)
2946 {
2947 	char *type;
2948 	int ret;
2949 
2950 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2951 		return (EZFS_INVALCONFIG);
2952 
2953 	if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2954 		/*
2955 		 * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2956 		 * For a spare vdev, we only want to boot from the active
2957 		 * spare device.
2958 		 */
2959 		if (is_spare) {
2960 			uint64_t spare = 0;
2961 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2962 			    &spare);
2963 			if (!spare)
2964 				return (EZFS_INVALCONFIG);
2965 		}
2966 
2967 		if (vdev_is_online(nv)) {
2968 			if ((ret = vdev_get_one_physpath(nv, physpath,
2969 			    phypath_size, rsz)) != 0)
2970 				return (ret);
2971 		}
2972 	} else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2973 	    strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
2974 	    strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2975 	    (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2976 		nvlist_t **child;
2977 		uint_t count;
2978 		int i, ret;
2979 
2980 		if (nvlist_lookup_nvlist_array(nv,
2981 		    ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2982 			return (EZFS_INVALCONFIG);
2983 
2984 		for (i = 0; i < count; i++) {
2985 			ret = vdev_get_physpaths(child[i], physpath,
2986 			    phypath_size, rsz, is_spare);
2987 			if (ret == EZFS_NOSPC)
2988 				return (ret);
2989 		}
2990 	}
2991 
2992 	return (EZFS_POOL_INVALARG);
2993 }
2994 
2995 /*
2996  * Get phys_path for a root pool config.
2997  * Return 0 on success; non-zero on failure.
2998  */
2999 static int
3000 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
3001 {
3002 	size_t rsz;
3003 	nvlist_t *vdev_root;
3004 	nvlist_t **child;
3005 	uint_t count;
3006 	char *type;
3007 
3008 	rsz = 0;
3009 
3010 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3011 	    &vdev_root) != 0)
3012 		return (EZFS_INVALCONFIG);
3013 
3014 	if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
3015 	    nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
3016 	    &child, &count) != 0)
3017 		return (EZFS_INVALCONFIG);
3018 
3019 	/*
3020 	 * root pool can only have a single top-level vdev.
3021 	 */
3022 	if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1)
3023 		return (EZFS_POOL_INVALARG);
3024 
3025 	(void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
3026 	    B_FALSE);
3027 
3028 	/* No online devices */
3029 	if (rsz == 0)
3030 		return (EZFS_NODEVICE);
3031 
3032 	return (0);
3033 }
3034 
3035 /*
3036  * Get phys_path for a root pool
3037  * Return 0 on success; non-zero on failure.
3038  */
3039 int
3040 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
3041 {
3042 	return (zpool_get_config_physpath(zhp->zpool_config, physpath,
3043 	    phypath_size));
3044 }
3045 
3046 /*
3047  * Convert a vdev path to a GUID.  Returns GUID or 0 on error.
3048  *
3049  * If is_spare, is_l2cache, or is_log is non-NULL, then store within it
3050  * if the VDEV is a spare, l2cache, or log device.  If they're NULL then
3051  * ignore them.
3052  */
3053 static uint64_t
3054 zpool_vdev_path_to_guid_impl(zpool_handle_t *zhp, const char *path,
3055     boolean_t *is_spare, boolean_t *is_l2cache, boolean_t *is_log)
3056 {
3057 	boolean_t spare = B_FALSE, l2cache = B_FALSE, log = B_FALSE;
3058 	nvlist_t *tgt;
3059 
3060 	if ((tgt = zpool_find_vdev(zhp, path, &spare, &l2cache,
3061 	    &log)) == NULL)
3062 		return (0);
3063 
3064 	if (is_spare != NULL)
3065 		*is_spare = spare;
3066 	if (is_l2cache != NULL)
3067 		*is_l2cache = l2cache;
3068 	if (is_log != NULL)
3069 		*is_log = log;
3070 
3071 	return (fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID));
3072 }
3073 
3074 /* Convert a vdev path to a GUID.  Returns GUID or 0 on error. */
3075 uint64_t
3076 zpool_vdev_path_to_guid(zpool_handle_t *zhp, const char *path)
3077 {
3078 	return (zpool_vdev_path_to_guid_impl(zhp, path, NULL, NULL, NULL));
3079 }
3080 
3081 /*
3082  * Bring the specified vdev online.   The 'flags' parameter is a set of the
3083  * ZFS_ONLINE_* flags.
3084  */
3085 int
3086 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
3087     vdev_state_t *newstate)
3088 {
3089 	zfs_cmd_t zc = {"\0"};
3090 	char msg[1024];
3091 	nvlist_t *tgt;
3092 	boolean_t avail_spare, l2cache, islog;
3093 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3094 
3095 	if (flags & ZFS_ONLINE_EXPAND) {
3096 		(void) snprintf(msg, sizeof (msg),
3097 		    dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
3098 	} else {
3099 		(void) snprintf(msg, sizeof (msg),
3100 		    dgettext(TEXT_DOMAIN, "cannot online %s"), path);
3101 	}
3102 
3103 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3104 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3105 	    &islog)) == NULL)
3106 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3107 
3108 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3109 
3110 	if (avail_spare)
3111 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
3112 
3113 #ifndef __FreeBSD__
3114 	char *pathname;
3115 	if ((flags & ZFS_ONLINE_EXPAND ||
3116 	    zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) &&
3117 	    nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH, &pathname) == 0) {
3118 		uint64_t wholedisk = 0;
3119 
3120 		(void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
3121 		    &wholedisk);
3122 
3123 		/*
3124 		 * XXX - L2ARC 1.0 devices can't support expansion.
3125 		 */
3126 		if (l2cache) {
3127 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3128 			    "cannot expand cache devices"));
3129 			return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
3130 		}
3131 
3132 		if (wholedisk) {
3133 			const char *fullpath = path;
3134 			char buf[MAXPATHLEN];
3135 			int error;
3136 
3137 			if (path[0] != '/') {
3138 				error = zfs_resolve_shortname(path, buf,
3139 				    sizeof (buf));
3140 				if (error != 0)
3141 					return (zfs_error(hdl, EZFS_NODEVICE,
3142 					    msg));
3143 
3144 				fullpath = buf;
3145 			}
3146 
3147 			error = zpool_relabel_disk(hdl, fullpath, msg);
3148 			if (error != 0)
3149 				return (error);
3150 		}
3151 	}
3152 #endif
3153 
3154 	zc.zc_cookie = VDEV_STATE_ONLINE;
3155 	zc.zc_obj = flags;
3156 
3157 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
3158 		if (errno == EINVAL) {
3159 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
3160 			    "from this pool into a new one.  Use '%s' "
3161 			    "instead"), "zpool detach");
3162 			return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
3163 		}
3164 		return (zpool_standard_error(hdl, errno, msg));
3165 	}
3166 
3167 	*newstate = zc.zc_cookie;
3168 	return (0);
3169 }
3170 
3171 /*
3172  * Take the specified vdev offline
3173  */
3174 int
3175 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
3176 {
3177 	zfs_cmd_t zc = {"\0"};
3178 	char msg[1024];
3179 	nvlist_t *tgt;
3180 	boolean_t avail_spare, l2cache;
3181 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3182 
3183 	(void) snprintf(msg, sizeof (msg),
3184 	    dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
3185 
3186 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3187 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3188 	    NULL)) == NULL)
3189 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3190 
3191 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3192 
3193 	if (avail_spare)
3194 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
3195 
3196 	zc.zc_cookie = VDEV_STATE_OFFLINE;
3197 	zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
3198 
3199 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
3200 		return (0);
3201 
3202 	switch (errno) {
3203 	case EBUSY:
3204 
3205 		/*
3206 		 * There are no other replicas of this device.
3207 		 */
3208 		return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
3209 
3210 	case EEXIST:
3211 		/*
3212 		 * The log device has unplayed logs
3213 		 */
3214 		return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
3215 
3216 	default:
3217 		return (zpool_standard_error(hdl, errno, msg));
3218 	}
3219 }
3220 
3221 /*
3222  * Mark the given vdev faulted.
3223  */
3224 int
3225 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
3226 {
3227 	zfs_cmd_t zc = {"\0"};
3228 	char msg[1024];
3229 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3230 
3231 	(void) snprintf(msg, sizeof (msg),
3232 	    dgettext(TEXT_DOMAIN, "cannot fault %llu"), (u_longlong_t)guid);
3233 
3234 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3235 	zc.zc_guid = guid;
3236 	zc.zc_cookie = VDEV_STATE_FAULTED;
3237 	zc.zc_obj = aux;
3238 
3239 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
3240 		return (0);
3241 
3242 	switch (errno) {
3243 	case EBUSY:
3244 
3245 		/*
3246 		 * There are no other replicas of this device.
3247 		 */
3248 		return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
3249 
3250 	default:
3251 		return (zpool_standard_error(hdl, errno, msg));
3252 	}
3253 
3254 }
3255 
3256 /*
3257  * Mark the given vdev degraded.
3258  */
3259 int
3260 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
3261 {
3262 	zfs_cmd_t zc = {"\0"};
3263 	char msg[1024];
3264 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3265 
3266 	(void) snprintf(msg, sizeof (msg),
3267 	    dgettext(TEXT_DOMAIN, "cannot degrade %llu"), (u_longlong_t)guid);
3268 
3269 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3270 	zc.zc_guid = guid;
3271 	zc.zc_cookie = VDEV_STATE_DEGRADED;
3272 	zc.zc_obj = aux;
3273 
3274 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
3275 		return (0);
3276 
3277 	return (zpool_standard_error(hdl, errno, msg));
3278 }
3279 
3280 /*
3281  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
3282  * a hot spare.
3283  */
3284 static boolean_t
3285 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
3286 {
3287 	nvlist_t **child;
3288 	uint_t c, children;
3289 
3290 	if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
3291 	    &children) == 0) {
3292 		char *type = fnvlist_lookup_string(search, ZPOOL_CONFIG_TYPE);
3293 		if ((strcmp(type, VDEV_TYPE_SPARE) == 0 ||
3294 		    strcmp(type, VDEV_TYPE_DRAID_SPARE) == 0) &&
3295 		    children == 2 && child[which] == tgt)
3296 			return (B_TRUE);
3297 
3298 		for (c = 0; c < children; c++)
3299 			if (is_replacing_spare(child[c], tgt, which))
3300 				return (B_TRUE);
3301 	}
3302 
3303 	return (B_FALSE);
3304 }
3305 
3306 /*
3307  * Attach new_disk (fully described by nvroot) to old_disk.
3308  * If 'replacing' is specified, the new disk will replace the old one.
3309  */
3310 int
3311 zpool_vdev_attach(zpool_handle_t *zhp, const char *old_disk,
3312     const char *new_disk, nvlist_t *nvroot, int replacing, boolean_t rebuild)
3313 {
3314 	zfs_cmd_t zc = {"\0"};
3315 	char msg[1024];
3316 	int ret;
3317 	nvlist_t *tgt;
3318 	boolean_t avail_spare, l2cache, islog;
3319 	uint64_t val;
3320 	char *newname;
3321 	nvlist_t **child;
3322 	uint_t children;
3323 	nvlist_t *config_root;
3324 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3325 
3326 	if (replacing)
3327 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
3328 		    "cannot replace %s with %s"), old_disk, new_disk);
3329 	else
3330 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
3331 		    "cannot attach %s to %s"), new_disk, old_disk);
3332 
3333 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3334 	if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
3335 	    &islog)) == NULL)
3336 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3337 
3338 	if (avail_spare)
3339 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
3340 
3341 	if (l2cache)
3342 		return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
3343 
3344 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3345 	zc.zc_cookie = replacing;
3346 	zc.zc_simple = rebuild;
3347 
3348 	if (rebuild &&
3349 	    zfeature_lookup_guid("org.openzfs:device_rebuild", NULL) != 0) {
3350 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3351 		    "the loaded zfs module doesn't support device rebuilds"));
3352 		return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
3353 	}
3354 
3355 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3356 	    &child, &children) != 0 || children != 1) {
3357 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3358 		    "new device must be a single disk"));
3359 		return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
3360 	}
3361 
3362 	config_root = fnvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
3363 	    ZPOOL_CONFIG_VDEV_TREE);
3364 
3365 	if ((newname = zpool_vdev_name(NULL, NULL, child[0], 0)) == NULL)
3366 		return (-1);
3367 
3368 	/*
3369 	 * If the target is a hot spare that has been swapped in, we can only
3370 	 * replace it with another hot spare.
3371 	 */
3372 	if (replacing &&
3373 	    nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
3374 	    (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
3375 	    NULL) == NULL || !avail_spare) &&
3376 	    is_replacing_spare(config_root, tgt, 1)) {
3377 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3378 		    "can only be replaced by another hot spare"));
3379 		free(newname);
3380 		return (zfs_error(hdl, EZFS_BADTARGET, msg));
3381 	}
3382 
3383 	free(newname);
3384 
3385 	zcmd_write_conf_nvlist(hdl, &zc, nvroot);
3386 
3387 	ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
3388 
3389 	zcmd_free_nvlists(&zc);
3390 
3391 	if (ret == 0)
3392 		return (0);
3393 
3394 	switch (errno) {
3395 	case ENOTSUP:
3396 		/*
3397 		 * Can't attach to or replace this type of vdev.
3398 		 */
3399 		if (replacing) {
3400 			uint64_t version = zpool_get_prop_int(zhp,
3401 			    ZPOOL_PROP_VERSION, NULL);
3402 
3403 			if (islog) {
3404 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3405 				    "cannot replace a log with a spare"));
3406 			} else if (rebuild) {
3407 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3408 				    "only mirror and dRAID vdevs support "
3409 				    "sequential reconstruction"));
3410 			} else if (zpool_is_draid_spare(new_disk)) {
3411 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3412 				    "dRAID spares can only replace child "
3413 				    "devices in their parent's dRAID vdev"));
3414 			} else if (version >= SPA_VERSION_MULTI_REPLACE) {
3415 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3416 				    "already in replacing/spare config; wait "
3417 				    "for completion or use 'zpool detach'"));
3418 			} else {
3419 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3420 				    "cannot replace a replacing device"));
3421 			}
3422 		} else {
3423 			char status[64] = {0};
3424 			zpool_prop_get_feature(zhp,
3425 			    "feature@device_rebuild", status, 63);
3426 			if (rebuild &&
3427 			    strncmp(status, ZFS_FEATURE_DISABLED, 64) == 0) {
3428 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3429 				    "device_rebuild feature must be enabled "
3430 				    "in order to use sequential "
3431 				    "reconstruction"));
3432 			} else {
3433 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3434 				    "can only attach to mirrors and top-level "
3435 				    "disks"));
3436 			}
3437 		}
3438 		(void) zfs_error(hdl, EZFS_BADTARGET, msg);
3439 		break;
3440 
3441 	case EINVAL:
3442 		/*
3443 		 * The new device must be a single disk.
3444 		 */
3445 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3446 		    "new device must be a single disk"));
3447 		(void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
3448 		break;
3449 
3450 	case EBUSY:
3451 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy, "
3452 		    "or device removal is in progress"),
3453 		    new_disk);
3454 		(void) zfs_error(hdl, EZFS_BADDEV, msg);
3455 		break;
3456 
3457 	case EOVERFLOW:
3458 		/*
3459 		 * The new device is too small.
3460 		 */
3461 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3462 		    "device is too small"));
3463 		(void) zfs_error(hdl, EZFS_BADDEV, msg);
3464 		break;
3465 
3466 	case EDOM:
3467 		/*
3468 		 * The new device has a different optimal sector size.
3469 		 */
3470 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3471 		    "new device has a different optimal sector size; use the "
3472 		    "option '-o ashift=N' to override the optimal size"));
3473 		(void) zfs_error(hdl, EZFS_BADDEV, msg);
3474 		break;
3475 
3476 	case ENAMETOOLONG:
3477 		/*
3478 		 * The resulting top-level vdev spec won't fit in the label.
3479 		 */
3480 		(void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
3481 		break;
3482 
3483 	default:
3484 		(void) zpool_standard_error(hdl, errno, msg);
3485 	}
3486 
3487 	return (-1);
3488 }
3489 
3490 /*
3491  * Detach the specified device.
3492  */
3493 int
3494 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
3495 {
3496 	zfs_cmd_t zc = {"\0"};
3497 	char msg[1024];
3498 	nvlist_t *tgt;
3499 	boolean_t avail_spare, l2cache;
3500 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3501 
3502 	(void) snprintf(msg, sizeof (msg),
3503 	    dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
3504 
3505 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3506 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3507 	    NULL)) == NULL)
3508 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3509 
3510 	if (avail_spare)
3511 		return (zfs_error(hdl, EZFS_ISSPARE, msg));
3512 
3513 	if (l2cache)
3514 		return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
3515 
3516 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3517 
3518 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
3519 		return (0);
3520 
3521 	switch (errno) {
3522 
3523 	case ENOTSUP:
3524 		/*
3525 		 * Can't detach from this type of vdev.
3526 		 */
3527 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
3528 		    "applicable to mirror and replacing vdevs"));
3529 		(void) zfs_error(hdl, EZFS_BADTARGET, msg);
3530 		break;
3531 
3532 	case EBUSY:
3533 		/*
3534 		 * There are no other replicas of this device.
3535 		 */
3536 		(void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
3537 		break;
3538 
3539 	default:
3540 		(void) zpool_standard_error(hdl, errno, msg);
3541 	}
3542 
3543 	return (-1);
3544 }
3545 
3546 /*
3547  * Find a mirror vdev in the source nvlist.
3548  *
3549  * The mchild array contains a list of disks in one of the top-level mirrors
3550  * of the source pool.  The schild array contains a list of disks that the
3551  * user specified on the command line.  We loop over the mchild array to
3552  * see if any entry in the schild array matches.
3553  *
3554  * If a disk in the mchild array is found in the schild array, we return
3555  * the index of that entry.  Otherwise we return -1.
3556  */
3557 static int
3558 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
3559     nvlist_t **schild, uint_t schildren)
3560 {
3561 	uint_t mc;
3562 
3563 	for (mc = 0; mc < mchildren; mc++) {
3564 		uint_t sc;
3565 		char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
3566 		    mchild[mc], 0);
3567 
3568 		for (sc = 0; sc < schildren; sc++) {
3569 			char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
3570 			    schild[sc], 0);
3571 			boolean_t result = (strcmp(mpath, spath) == 0);
3572 
3573 			free(spath);
3574 			if (result) {
3575 				free(mpath);
3576 				return (mc);
3577 			}
3578 		}
3579 
3580 		free(mpath);
3581 	}
3582 
3583 	return (-1);
3584 }
3585 
3586 /*
3587  * Split a mirror pool.  If newroot points to null, then a new nvlist
3588  * is generated and it is the responsibility of the caller to free it.
3589  */
3590 int
3591 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
3592     nvlist_t *props, splitflags_t flags)
3593 {
3594 	zfs_cmd_t zc = {"\0"};
3595 	char msg[1024], *bias;
3596 	nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
3597 	nvlist_t **varray = NULL, *zc_props = NULL;
3598 	uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
3599 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3600 	uint64_t vers, readonly = B_FALSE;
3601 	boolean_t freelist = B_FALSE, memory_err = B_TRUE;
3602 	int retval = 0;
3603 
3604 	(void) snprintf(msg, sizeof (msg),
3605 	    dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
3606 
3607 	if (!zpool_name_valid(hdl, B_FALSE, newname))
3608 		return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
3609 
3610 	if ((config = zpool_get_config(zhp, NULL)) == NULL) {
3611 		(void) fprintf(stderr, gettext("Internal error: unable to "
3612 		    "retrieve pool configuration\n"));
3613 		return (-1);
3614 	}
3615 
3616 	tree = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE);
3617 	vers = fnvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION);
3618 
3619 	if (props) {
3620 		prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
3621 		if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
3622 		    props, vers, flags, msg)) == NULL)
3623 			return (-1);
3624 		(void) nvlist_lookup_uint64(zc_props,
3625 		    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3626 		if (readonly) {
3627 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3628 			    "property %s can only be set at import time"),
3629 			    zpool_prop_to_name(ZPOOL_PROP_READONLY));
3630 			return (-1);
3631 		}
3632 	}
3633 
3634 	if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
3635 	    &children) != 0) {
3636 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3637 		    "Source pool is missing vdev tree"));
3638 		nvlist_free(zc_props);
3639 		return (-1);
3640 	}
3641 
3642 	varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
3643 	vcount = 0;
3644 
3645 	if (*newroot == NULL ||
3646 	    nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
3647 	    &newchild, &newchildren) != 0)
3648 		newchildren = 0;
3649 
3650 	for (c = 0; c < children; c++) {
3651 		uint64_t is_log = B_FALSE, is_hole = B_FALSE;
3652 		boolean_t is_special = B_FALSE, is_dedup = B_FALSE;
3653 		char *type;
3654 		nvlist_t **mchild, *vdev;
3655 		uint_t mchildren;
3656 		int entry;
3657 
3658 		/*
3659 		 * Unlike cache & spares, slogs are stored in the
3660 		 * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
3661 		 */
3662 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
3663 		    &is_log);
3664 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
3665 		    &is_hole);
3666 		if (is_log || is_hole) {
3667 			/*
3668 			 * Create a hole vdev and put it in the config.
3669 			 */
3670 			if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
3671 				goto out;
3672 			if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
3673 			    VDEV_TYPE_HOLE) != 0)
3674 				goto out;
3675 			if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
3676 			    1) != 0)
3677 				goto out;
3678 			if (lastlog == 0)
3679 				lastlog = vcount;
3680 			varray[vcount++] = vdev;
3681 			continue;
3682 		}
3683 		lastlog = 0;
3684 		type = fnvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE);
3685 
3686 		if (strcmp(type, VDEV_TYPE_INDIRECT) == 0) {
3687 			vdev = child[c];
3688 			if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3689 				goto out;
3690 			continue;
3691 		} else if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
3692 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3693 			    "Source pool must be composed only of mirrors\n"));
3694 			retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3695 			goto out;
3696 		}
3697 
3698 		if (nvlist_lookup_string(child[c],
3699 		    ZPOOL_CONFIG_ALLOCATION_BIAS, &bias) == 0) {
3700 			if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
3701 				is_special = B_TRUE;
3702 			else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
3703 				is_dedup = B_TRUE;
3704 		}
3705 		verify(nvlist_lookup_nvlist_array(child[c],
3706 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3707 
3708 		/* find or add an entry for this top-level vdev */
3709 		if (newchildren > 0 &&
3710 		    (entry = find_vdev_entry(zhp, mchild, mchildren,
3711 		    newchild, newchildren)) >= 0) {
3712 			/* We found a disk that the user specified. */
3713 			vdev = mchild[entry];
3714 			++found;
3715 		} else {
3716 			/* User didn't specify a disk for this vdev. */
3717 			vdev = mchild[mchildren - 1];
3718 		}
3719 
3720 		if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3721 			goto out;
3722 
3723 		if (flags.dryrun != 0) {
3724 			if (is_dedup == B_TRUE) {
3725 				if (nvlist_add_string(varray[vcount - 1],
3726 				    ZPOOL_CONFIG_ALLOCATION_BIAS,
3727 				    VDEV_ALLOC_BIAS_DEDUP) != 0)
3728 					goto out;
3729 			} else if (is_special == B_TRUE) {
3730 				if (nvlist_add_string(varray[vcount - 1],
3731 				    ZPOOL_CONFIG_ALLOCATION_BIAS,
3732 				    VDEV_ALLOC_BIAS_SPECIAL) != 0)
3733 					goto out;
3734 			}
3735 		}
3736 	}
3737 
3738 	/* did we find every disk the user specified? */
3739 	if (found != newchildren) {
3740 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3741 		    "include at most one disk from each mirror"));
3742 		retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3743 		goto out;
3744 	}
3745 
3746 	/* Prepare the nvlist for populating. */
3747 	if (*newroot == NULL) {
3748 		if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3749 			goto out;
3750 		freelist = B_TRUE;
3751 		if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3752 		    VDEV_TYPE_ROOT) != 0)
3753 			goto out;
3754 	} else {
3755 		verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3756 	}
3757 
3758 	/* Add all the children we found */
3759 	if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
3760 	    (const nvlist_t **)varray, lastlog == 0 ? vcount : lastlog) != 0)
3761 		goto out;
3762 
3763 	/*
3764 	 * If we're just doing a dry run, exit now with success.
3765 	 */
3766 	if (flags.dryrun) {
3767 		memory_err = B_FALSE;
3768 		freelist = B_FALSE;
3769 		goto out;
3770 	}
3771 
3772 	/* now build up the config list & call the ioctl */
3773 	if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3774 		goto out;
3775 
3776 	if (nvlist_add_nvlist(newconfig,
3777 	    ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3778 	    nvlist_add_string(newconfig,
3779 	    ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3780 	    nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3781 		goto out;
3782 
3783 	/*
3784 	 * The new pool is automatically part of the namespace unless we
3785 	 * explicitly export it.
3786 	 */
3787 	if (!flags.import)
3788 		zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3789 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3790 	(void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3791 	zcmd_write_conf_nvlist(hdl, &zc, newconfig);
3792 	if (zc_props != NULL)
3793 		zcmd_write_src_nvlist(hdl, &zc, zc_props);
3794 
3795 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3796 		retval = zpool_standard_error(hdl, errno, msg);
3797 		goto out;
3798 	}
3799 
3800 	freelist = B_FALSE;
3801 	memory_err = B_FALSE;
3802 
3803 out:
3804 	if (varray != NULL) {
3805 		int v;
3806 
3807 		for (v = 0; v < vcount; v++)
3808 			nvlist_free(varray[v]);
3809 		free(varray);
3810 	}
3811 	zcmd_free_nvlists(&zc);
3812 	nvlist_free(zc_props);
3813 	nvlist_free(newconfig);
3814 	if (freelist) {
3815 		nvlist_free(*newroot);
3816 		*newroot = NULL;
3817 	}
3818 
3819 	if (retval != 0)
3820 		return (retval);
3821 
3822 	if (memory_err)
3823 		return (no_memory(hdl));
3824 
3825 	return (0);
3826 }
3827 
3828 /*
3829  * Remove the given device.
3830  */
3831 int
3832 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3833 {
3834 	zfs_cmd_t zc = {"\0"};
3835 	char msg[1024];
3836 	nvlist_t *tgt;
3837 	boolean_t avail_spare, l2cache, islog;
3838 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3839 	uint64_t version;
3840 
3841 	(void) snprintf(msg, sizeof (msg),
3842 	    dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3843 
3844 	if (zpool_is_draid_spare(path)) {
3845 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3846 		    "dRAID spares cannot be removed"));
3847 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3848 	}
3849 
3850 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3851 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3852 	    &islog)) == NULL)
3853 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3854 
3855 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3856 	if (islog && version < SPA_VERSION_HOLES) {
3857 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3858 		    "pool must be upgraded to support log removal"));
3859 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
3860 	}
3861 
3862 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3863 
3864 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3865 		return (0);
3866 
3867 	switch (errno) {
3868 
3869 	case EINVAL:
3870 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3871 		    "invalid config; all top-level vdevs must "
3872 		    "have the same sector size and not be raidz."));
3873 		(void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
3874 		break;
3875 
3876 	case EBUSY:
3877 		if (islog) {
3878 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3879 			    "Mount encrypted datasets to replay logs."));
3880 		} else {
3881 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3882 			    "Pool busy; removal may already be in progress"));
3883 		}
3884 		(void) zfs_error(hdl, EZFS_BUSY, msg);
3885 		break;
3886 
3887 	case EACCES:
3888 		if (islog) {
3889 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3890 			    "Mount encrypted datasets to replay logs."));
3891 			(void) zfs_error(hdl, EZFS_BUSY, msg);
3892 		} else {
3893 			(void) zpool_standard_error(hdl, errno, msg);
3894 		}
3895 		break;
3896 
3897 	default:
3898 		(void) zpool_standard_error(hdl, errno, msg);
3899 	}
3900 	return (-1);
3901 }
3902 
3903 int
3904 zpool_vdev_remove_cancel(zpool_handle_t *zhp)
3905 {
3906 	zfs_cmd_t zc = {{0}};
3907 	char msg[1024];
3908 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3909 
3910 	(void) snprintf(msg, sizeof (msg),
3911 	    dgettext(TEXT_DOMAIN, "cannot cancel removal"));
3912 
3913 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3914 	zc.zc_cookie = 1;
3915 
3916 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3917 		return (0);
3918 
3919 	return (zpool_standard_error(hdl, errno, msg));
3920 }
3921 
3922 int
3923 zpool_vdev_indirect_size(zpool_handle_t *zhp, const char *path,
3924     uint64_t *sizep)
3925 {
3926 	char msg[1024];
3927 	nvlist_t *tgt;
3928 	boolean_t avail_spare, l2cache, islog;
3929 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3930 
3931 	(void) snprintf(msg, sizeof (msg),
3932 	    dgettext(TEXT_DOMAIN, "cannot determine indirect size of %s"),
3933 	    path);
3934 
3935 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3936 	    &islog)) == NULL)
3937 		return (zfs_error(hdl, EZFS_NODEVICE, msg));
3938 
3939 	if (avail_spare || l2cache || islog) {
3940 		*sizep = 0;
3941 		return (0);
3942 	}
3943 
3944 	if (nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_INDIRECT_SIZE, sizep) != 0) {
3945 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3946 		    "indirect size not available"));
3947 		return (zfs_error(hdl, EINVAL, msg));
3948 	}
3949 	return (0);
3950 }
3951 
3952 /*
3953  * Clear the errors for the pool, or the particular device if specified.
3954  */
3955 int
3956 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3957 {
3958 	zfs_cmd_t zc = {"\0"};
3959 	char msg[1024];
3960 	nvlist_t *tgt;
3961 	zpool_load_policy_t policy;
3962 	boolean_t avail_spare, l2cache;
3963 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3964 	nvlist_t *nvi = NULL;
3965 	int error;
3966 
3967 	if (path)
3968 		(void) snprintf(msg, sizeof (msg),
3969 		    dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3970 		    path);
3971 	else
3972 		(void) snprintf(msg, sizeof (msg),
3973 		    dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3974 		    zhp->zpool_name);
3975 
3976 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3977 	if (path) {
3978 		if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3979 		    &l2cache, NULL)) == NULL)
3980 			return (zfs_error(hdl, EZFS_NODEVICE, msg));
3981 
3982 		/*
3983 		 * Don't allow error clearing for hot spares.  Do allow
3984 		 * error clearing for l2cache devices.
3985 		 */
3986 		if (avail_spare)
3987 			return (zfs_error(hdl, EZFS_ISSPARE, msg));
3988 
3989 		zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3990 	}
3991 
3992 	zpool_get_load_policy(rewindnvl, &policy);
3993 	zc.zc_cookie = policy.zlp_rewind;
3994 
3995 	zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2);
3996 	zcmd_write_src_nvlist(hdl, &zc, rewindnvl);
3997 
3998 	while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3999 	    errno == ENOMEM)
4000 		zcmd_expand_dst_nvlist(hdl, &zc);
4001 
4002 	if (!error || ((policy.zlp_rewind & ZPOOL_TRY_REWIND) &&
4003 	    errno != EPERM && errno != EACCES)) {
4004 		if (policy.zlp_rewind &
4005 		    (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
4006 			(void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
4007 			zpool_rewind_exclaim(hdl, zc.zc_name,
4008 			    ((policy.zlp_rewind & ZPOOL_TRY_REWIND) != 0),
4009 			    nvi);
4010 			nvlist_free(nvi);
4011 		}
4012 		zcmd_free_nvlists(&zc);
4013 		return (0);
4014 	}
4015 
4016 	zcmd_free_nvlists(&zc);
4017 	return (zpool_standard_error(hdl, errno, msg));
4018 }
4019 
4020 /*
4021  * Similar to zpool_clear(), but takes a GUID (used by fmd).
4022  */
4023 int
4024 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
4025 {
4026 	zfs_cmd_t zc = {"\0"};
4027 	char msg[1024];
4028 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4029 
4030 	(void) snprintf(msg, sizeof (msg),
4031 	    dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
4032 	    (u_longlong_t)guid);
4033 
4034 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4035 	zc.zc_guid = guid;
4036 	zc.zc_cookie = ZPOOL_NO_REWIND;
4037 
4038 	if (zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc) == 0)
4039 		return (0);
4040 
4041 	return (zpool_standard_error(hdl, errno, msg));
4042 }
4043 
4044 /*
4045  * Change the GUID for a pool.
4046  */
4047 int
4048 zpool_reguid(zpool_handle_t *zhp)
4049 {
4050 	char msg[1024];
4051 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4052 	zfs_cmd_t zc = {"\0"};
4053 
4054 	(void) snprintf(msg, sizeof (msg),
4055 	    dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
4056 
4057 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4058 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
4059 		return (0);
4060 
4061 	return (zpool_standard_error(hdl, errno, msg));
4062 }
4063 
4064 /*
4065  * Reopen the pool.
4066  */
4067 int
4068 zpool_reopen_one(zpool_handle_t *zhp, void *data)
4069 {
4070 	libzfs_handle_t *hdl = zpool_get_handle(zhp);
4071 	const char *pool_name = zpool_get_name(zhp);
4072 	boolean_t *scrub_restart = data;
4073 	int error;
4074 
4075 	error = lzc_reopen(pool_name, *scrub_restart);
4076 	if (error) {
4077 		return (zpool_standard_error_fmt(hdl, error,
4078 		    dgettext(TEXT_DOMAIN, "cannot reopen '%s'"), pool_name));
4079 	}
4080 
4081 	return (0);
4082 }
4083 
4084 /* call into libzfs_core to execute the sync IOCTL per pool */
4085 int
4086 zpool_sync_one(zpool_handle_t *zhp, void *data)
4087 {
4088 	int ret;
4089 	libzfs_handle_t *hdl = zpool_get_handle(zhp);
4090 	const char *pool_name = zpool_get_name(zhp);
4091 	boolean_t *force = data;
4092 	nvlist_t *innvl = fnvlist_alloc();
4093 
4094 	fnvlist_add_boolean_value(innvl, "force", *force);
4095 	if ((ret = lzc_sync(pool_name, innvl, NULL)) != 0) {
4096 		nvlist_free(innvl);
4097 		return (zpool_standard_error_fmt(hdl, ret,
4098 		    dgettext(TEXT_DOMAIN, "sync '%s' failed"), pool_name));
4099 	}
4100 	nvlist_free(innvl);
4101 
4102 	return (0);
4103 }
4104 
4105 #define	PATH_BUF_LEN	64
4106 
4107 /*
4108  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
4109  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
4110  * We also check if this is a whole disk, in which case we strip off the
4111  * trailing 's0' slice name.
4112  *
4113  * This routine is also responsible for identifying when disks have been
4114  * reconfigured in a new location.  The kernel will have opened the device by
4115  * devid, but the path will still refer to the old location.  To catch this, we
4116  * first do a path -> devid translation (which is fast for the common case).  If
4117  * the devid matches, we're done.  If not, we do a reverse devid -> path
4118  * translation and issue the appropriate ioctl() to update the path of the vdev.
4119  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
4120  * of these checks.
4121  */
4122 char *
4123 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
4124     int name_flags)
4125 {
4126 	char *type, *tpath;
4127 	const char *path;
4128 	uint64_t value;
4129 	char buf[PATH_BUF_LEN];
4130 	char tmpbuf[PATH_BUF_LEN * 2];
4131 
4132 	/*
4133 	 * vdev_name will be "root"/"root-0" for the root vdev, but it is the
4134 	 * zpool name that will be displayed to the user.
4135 	 */
4136 	type = fnvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE);
4137 	if (zhp != NULL && strcmp(type, "root") == 0)
4138 		return (zfs_strdup(hdl, zpool_get_name(zhp)));
4139 
4140 	if (libzfs_envvar_is_set("ZPOOL_VDEV_NAME_PATH"))
4141 		name_flags |= VDEV_NAME_PATH;
4142 	if (libzfs_envvar_is_set("ZPOOL_VDEV_NAME_GUID"))
4143 		name_flags |= VDEV_NAME_GUID;
4144 	if (libzfs_envvar_is_set("ZPOOL_VDEV_NAME_FOLLOW_LINKS"))
4145 		name_flags |= VDEV_NAME_FOLLOW_LINKS;
4146 
4147 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
4148 	    name_flags & VDEV_NAME_GUID) {
4149 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value);
4150 		(void) snprintf(buf, sizeof (buf), "%llu", (u_longlong_t)value);
4151 		path = buf;
4152 	} else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &tpath) == 0) {
4153 		path = tpath;
4154 
4155 		if (name_flags & VDEV_NAME_FOLLOW_LINKS) {
4156 			char *rp = realpath(path, NULL);
4157 			if (rp) {
4158 				strlcpy(buf, rp, sizeof (buf));
4159 				path = buf;
4160 				free(rp);
4161 			}
4162 		}
4163 
4164 		/*
4165 		 * For a block device only use the name.
4166 		 */
4167 		if ((strcmp(type, VDEV_TYPE_DISK) == 0) &&
4168 		    !(name_flags & VDEV_NAME_PATH)) {
4169 			path = zfs_strip_path(path);
4170 		}
4171 
4172 		/*
4173 		 * Remove the partition from the path if this is a whole disk.
4174 		 */
4175 		if (strcmp(type, VDEV_TYPE_DRAID_SPARE) != 0 &&
4176 		    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &value)
4177 		    == 0 && value && !(name_flags & VDEV_NAME_PATH)) {
4178 			return (zfs_strip_partition(path));
4179 		}
4180 	} else {
4181 		path = type;
4182 
4183 		/*
4184 		 * If it's a raidz device, we need to stick in the parity level.
4185 		 */
4186 		if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
4187 			value = fnvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY);
4188 			(void) snprintf(buf, sizeof (buf), "%s%llu", path,
4189 			    (u_longlong_t)value);
4190 			path = buf;
4191 		}
4192 
4193 		/*
4194 		 * If it's a dRAID device, we add parity, groups, and spares.
4195 		 */
4196 		if (strcmp(path, VDEV_TYPE_DRAID) == 0) {
4197 			uint64_t ndata, nparity, nspares;
4198 			nvlist_t **child;
4199 			uint_t children;
4200 
4201 			verify(nvlist_lookup_nvlist_array(nv,
4202 			    ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
4203 			nparity = fnvlist_lookup_uint64(nv,
4204 			    ZPOOL_CONFIG_NPARITY);
4205 			ndata = fnvlist_lookup_uint64(nv,
4206 			    ZPOOL_CONFIG_DRAID_NDATA);
4207 			nspares = fnvlist_lookup_uint64(nv,
4208 			    ZPOOL_CONFIG_DRAID_NSPARES);
4209 
4210 			path = zpool_draid_name(buf, sizeof (buf), ndata,
4211 			    nparity, nspares, children);
4212 		}
4213 
4214 		/*
4215 		 * We identify each top-level vdev by using a <type-id>
4216 		 * naming convention.
4217 		 */
4218 		if (name_flags & VDEV_NAME_TYPE_ID) {
4219 			uint64_t id = fnvlist_lookup_uint64(nv,
4220 			    ZPOOL_CONFIG_ID);
4221 			(void) snprintf(tmpbuf, sizeof (tmpbuf), "%s-%llu",
4222 			    path, (u_longlong_t)id);
4223 			path = tmpbuf;
4224 		}
4225 	}
4226 
4227 	return (zfs_strdup(hdl, path));
4228 }
4229 
4230 static int
4231 zbookmark_mem_compare(const void *a, const void *b)
4232 {
4233 	return (memcmp(a, b, sizeof (zbookmark_phys_t)));
4234 }
4235 
4236 /*
4237  * Retrieve the persistent error log, uniquify the members, and return to the
4238  * caller.
4239  */
4240 int
4241 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
4242 {
4243 	zfs_cmd_t zc = {"\0"};
4244 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4245 	uint64_t count;
4246 	zbookmark_phys_t *zb = NULL;
4247 	int i;
4248 
4249 	/*
4250 	 * Retrieve the raw error list from the kernel.  If the number of errors
4251 	 * has increased, allocate more space and continue until we get the
4252 	 * entire list.
4253 	 */
4254 	count = fnvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT);
4255 	if (count == 0)
4256 		return (0);
4257 	zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
4258 	    count * sizeof (zbookmark_phys_t));
4259 	zc.zc_nvlist_dst_size = count;
4260 	(void) strcpy(zc.zc_name, zhp->zpool_name);
4261 	for (;;) {
4262 		if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_ERROR_LOG,
4263 		    &zc) != 0) {
4264 			free((void *)(uintptr_t)zc.zc_nvlist_dst);
4265 			if (errno == ENOMEM) {
4266 				void *dst;
4267 
4268 				count = zc.zc_nvlist_dst_size;
4269 				dst = zfs_alloc(zhp->zpool_hdl, count *
4270 				    sizeof (zbookmark_phys_t));
4271 				zc.zc_nvlist_dst = (uintptr_t)dst;
4272 			} else {
4273 				return (zpool_standard_error_fmt(hdl, errno,
4274 				    dgettext(TEXT_DOMAIN, "errors: List of "
4275 				    "errors unavailable")));
4276 			}
4277 		} else {
4278 			break;
4279 		}
4280 	}
4281 
4282 	/*
4283 	 * Sort the resulting bookmarks.  This is a little confusing due to the
4284 	 * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
4285 	 * to first, and 'zc_nvlist_dst_size' indicates the number of bookmarks
4286 	 * _not_ copied as part of the process.  So we point the start of our
4287 	 * array appropriate and decrement the total number of elements.
4288 	 */
4289 	zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
4290 	    zc.zc_nvlist_dst_size;
4291 	count -= zc.zc_nvlist_dst_size;
4292 
4293 	qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_mem_compare);
4294 
4295 	verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
4296 
4297 	/*
4298 	 * Fill in the nverrlistp with nvlist's of dataset and object numbers.
4299 	 */
4300 	for (i = 0; i < count; i++) {
4301 		nvlist_t *nv;
4302 
4303 		/* ignoring zb_blkid and zb_level for now */
4304 		if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
4305 		    zb[i-1].zb_object == zb[i].zb_object)
4306 			continue;
4307 
4308 		if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
4309 			goto nomem;
4310 		if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
4311 		    zb[i].zb_objset) != 0) {
4312 			nvlist_free(nv);
4313 			goto nomem;
4314 		}
4315 		if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
4316 		    zb[i].zb_object) != 0) {
4317 			nvlist_free(nv);
4318 			goto nomem;
4319 		}
4320 		if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
4321 			nvlist_free(nv);
4322 			goto nomem;
4323 		}
4324 		nvlist_free(nv);
4325 	}
4326 
4327 	free((void *)(uintptr_t)zc.zc_nvlist_dst);
4328 	return (0);
4329 
4330 nomem:
4331 	free((void *)(uintptr_t)zc.zc_nvlist_dst);
4332 	return (no_memory(zhp->zpool_hdl));
4333 }
4334 
4335 /*
4336  * Upgrade a ZFS pool to the latest on-disk version.
4337  */
4338 int
4339 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
4340 {
4341 	zfs_cmd_t zc = {"\0"};
4342 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4343 
4344 	(void) strcpy(zc.zc_name, zhp->zpool_name);
4345 	zc.zc_cookie = new_version;
4346 
4347 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
4348 		return (zpool_standard_error_fmt(hdl, errno,
4349 		    dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
4350 		    zhp->zpool_name));
4351 	return (0);
4352 }
4353 
4354 void
4355 zfs_save_arguments(int argc, char **argv, char *string, int len)
4356 {
4357 	int i;
4358 
4359 	(void) strlcpy(string, zfs_basename(argv[0]), len);
4360 	for (i = 1; i < argc; i++) {
4361 		(void) strlcat(string, " ", len);
4362 		(void) strlcat(string, argv[i], len);
4363 	}
4364 }
4365 
4366 int
4367 zpool_log_history(libzfs_handle_t *hdl, const char *message)
4368 {
4369 	zfs_cmd_t zc = {"\0"};
4370 	nvlist_t *args;
4371 
4372 	args = fnvlist_alloc();
4373 	fnvlist_add_string(args, "message", message);
4374 	zcmd_write_src_nvlist(hdl, &zc, args);
4375 	int err = zfs_ioctl(hdl, ZFS_IOC_LOG_HISTORY, &zc);
4376 	nvlist_free(args);
4377 	zcmd_free_nvlists(&zc);
4378 	return (err);
4379 }
4380 
4381 /*
4382  * Perform ioctl to get some command history of a pool.
4383  *
4384  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
4385  * logical offset of the history buffer to start reading from.
4386  *
4387  * Upon return, 'off' is the next logical offset to read from and
4388  * 'len' is the actual amount of bytes read into 'buf'.
4389  */
4390 static int
4391 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
4392 {
4393 	zfs_cmd_t zc = {"\0"};
4394 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4395 
4396 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4397 
4398 	zc.zc_history = (uint64_t)(uintptr_t)buf;
4399 	zc.zc_history_len = *len;
4400 	zc.zc_history_offset = *off;
4401 
4402 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
4403 		switch (errno) {
4404 		case EPERM:
4405 			return (zfs_error_fmt(hdl, EZFS_PERM,
4406 			    dgettext(TEXT_DOMAIN,
4407 			    "cannot show history for pool '%s'"),
4408 			    zhp->zpool_name));
4409 		case ENOENT:
4410 			return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
4411 			    dgettext(TEXT_DOMAIN, "cannot get history for pool "
4412 			    "'%s'"), zhp->zpool_name));
4413 		case ENOTSUP:
4414 			return (zfs_error_fmt(hdl, EZFS_BADVERSION,
4415 			    dgettext(TEXT_DOMAIN, "cannot get history for pool "
4416 			    "'%s', pool must be upgraded"), zhp->zpool_name));
4417 		default:
4418 			return (zpool_standard_error_fmt(hdl, errno,
4419 			    dgettext(TEXT_DOMAIN,
4420 			    "cannot get history for '%s'"), zhp->zpool_name));
4421 		}
4422 	}
4423 
4424 	*len = zc.zc_history_len;
4425 	*off = zc.zc_history_offset;
4426 
4427 	return (0);
4428 }
4429 
4430 /*
4431  * Retrieve the command history of a pool.
4432  */
4433 int
4434 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp, uint64_t *off,
4435     boolean_t *eof)
4436 {
4437 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4438 	char *buf;
4439 	int buflen = 128 * 1024;
4440 	nvlist_t **records = NULL;
4441 	uint_t numrecords = 0;
4442 	int err = 0, i;
4443 	uint64_t start = *off;
4444 
4445 	buf = zfs_alloc(hdl, buflen);
4446 
4447 	/* process about 1MiB a time */
4448 	while (*off - start < 1024 * 1024) {
4449 		uint64_t bytes_read = buflen;
4450 		uint64_t leftover;
4451 
4452 		if ((err = get_history(zhp, buf, off, &bytes_read)) != 0)
4453 			break;
4454 
4455 		/* if nothing else was read in, we're at EOF, just return */
4456 		if (!bytes_read) {
4457 			*eof = B_TRUE;
4458 			break;
4459 		}
4460 
4461 		if ((err = zpool_history_unpack(buf, bytes_read,
4462 		    &leftover, &records, &numrecords)) != 0) {
4463 			zpool_standard_error_fmt(hdl, err,
4464 			    dgettext(TEXT_DOMAIN,
4465 			    "cannot get history for '%s'"), zhp->zpool_name);
4466 			break;
4467 		}
4468 		*off -= leftover;
4469 		if (leftover == bytes_read) {
4470 			/*
4471 			 * no progress made, because buffer is not big enough
4472 			 * to hold this record; resize and retry.
4473 			 */
4474 			buflen *= 2;
4475 			free(buf);
4476 			buf = zfs_alloc(hdl, buflen);
4477 		}
4478 	}
4479 
4480 	free(buf);
4481 
4482 	if (!err) {
4483 		*nvhisp = fnvlist_alloc();
4484 		fnvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
4485 		    (const nvlist_t **)records, numrecords);
4486 	}
4487 	for (i = 0; i < numrecords; i++)
4488 		nvlist_free(records[i]);
4489 	free(records);
4490 
4491 	return (err);
4492 }
4493 
4494 /*
4495  * Retrieve the next event given the passed 'zevent_fd' file descriptor.
4496  * If there is a new event available 'nvp' will contain a newly allocated
4497  * nvlist and 'dropped' will be set to the number of missed events since
4498  * the last call to this function.  When 'nvp' is set to NULL it indicates
4499  * no new events are available.  In either case the function returns 0 and
4500  * it is up to the caller to free 'nvp'.  In the case of a fatal error the
4501  * function will return a non-zero value.  When the function is called in
4502  * blocking mode (the default, unless the ZEVENT_NONBLOCK flag is passed),
4503  * it will not return until a new event is available.
4504  */
4505 int
4506 zpool_events_next(libzfs_handle_t *hdl, nvlist_t **nvp,
4507     int *dropped, unsigned flags, int zevent_fd)
4508 {
4509 	zfs_cmd_t zc = {"\0"};
4510 	int error = 0;
4511 
4512 	*nvp = NULL;
4513 	*dropped = 0;
4514 	zc.zc_cleanup_fd = zevent_fd;
4515 
4516 	if (flags & ZEVENT_NONBLOCK)
4517 		zc.zc_guid = ZEVENT_NONBLOCK;
4518 
4519 	zcmd_alloc_dst_nvlist(hdl, &zc, ZEVENT_SIZE);
4520 
4521 retry:
4522 	if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_NEXT, &zc) != 0) {
4523 		switch (errno) {
4524 		case ESHUTDOWN:
4525 			error = zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
4526 			    dgettext(TEXT_DOMAIN, "zfs shutdown"));
4527 			goto out;
4528 		case ENOENT:
4529 			/* Blocking error case should not occur */
4530 			if (!(flags & ZEVENT_NONBLOCK))
4531 				error = zpool_standard_error_fmt(hdl, errno,
4532 				    dgettext(TEXT_DOMAIN, "cannot get event"));
4533 
4534 			goto out;
4535 		case ENOMEM:
4536 			zcmd_expand_dst_nvlist(hdl, &zc);
4537 			goto retry;
4538 		default:
4539 			error = zpool_standard_error_fmt(hdl, errno,
4540 			    dgettext(TEXT_DOMAIN, "cannot get event"));
4541 			goto out;
4542 		}
4543 	}
4544 
4545 	error = zcmd_read_dst_nvlist(hdl, &zc, nvp);
4546 	if (error != 0)
4547 		goto out;
4548 
4549 	*dropped = (int)zc.zc_cookie;
4550 out:
4551 	zcmd_free_nvlists(&zc);
4552 
4553 	return (error);
4554 }
4555 
4556 /*
4557  * Clear all events.
4558  */
4559 int
4560 zpool_events_clear(libzfs_handle_t *hdl, int *count)
4561 {
4562 	zfs_cmd_t zc = {"\0"};
4563 
4564 	if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_CLEAR, &zc) != 0)
4565 		return (zpool_standard_error(hdl, errno,
4566 		    dgettext(TEXT_DOMAIN, "cannot clear events")));
4567 
4568 	if (count != NULL)
4569 		*count = (int)zc.zc_cookie; /* # of events cleared */
4570 
4571 	return (0);
4572 }
4573 
4574 /*
4575  * Seek to a specific EID, ZEVENT_SEEK_START, or ZEVENT_SEEK_END for
4576  * the passed zevent_fd file handle.  On success zero is returned,
4577  * otherwise -1 is returned and hdl->libzfs_error is set to the errno.
4578  */
4579 int
4580 zpool_events_seek(libzfs_handle_t *hdl, uint64_t eid, int zevent_fd)
4581 {
4582 	zfs_cmd_t zc = {"\0"};
4583 	int error = 0;
4584 
4585 	zc.zc_guid = eid;
4586 	zc.zc_cleanup_fd = zevent_fd;
4587 
4588 	if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_SEEK, &zc) != 0) {
4589 		switch (errno) {
4590 		case ENOENT:
4591 			error = zfs_error_fmt(hdl, EZFS_NOENT,
4592 			    dgettext(TEXT_DOMAIN, "cannot get event"));
4593 			break;
4594 
4595 		case ENOMEM:
4596 			error = zfs_error_fmt(hdl, EZFS_NOMEM,
4597 			    dgettext(TEXT_DOMAIN, "cannot get event"));
4598 			break;
4599 
4600 		default:
4601 			error = zpool_standard_error_fmt(hdl, errno,
4602 			    dgettext(TEXT_DOMAIN, "cannot get event"));
4603 			break;
4604 		}
4605 	}
4606 
4607 	return (error);
4608 }
4609 
4610 static void
4611 zpool_obj_to_path_impl(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4612     char *pathname, size_t len, boolean_t always_unmounted)
4613 {
4614 	zfs_cmd_t zc = {"\0"};
4615 	boolean_t mounted = B_FALSE;
4616 	char *mntpnt = NULL;
4617 	char dsname[ZFS_MAX_DATASET_NAME_LEN];
4618 
4619 	if (dsobj == 0) {
4620 		/* special case for the MOS */
4621 		(void) snprintf(pathname, len, "<metadata>:<0x%llx>",
4622 		    (longlong_t)obj);
4623 		return;
4624 	}
4625 
4626 	/* get the dataset's name */
4627 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4628 	zc.zc_obj = dsobj;
4629 	if (zfs_ioctl(zhp->zpool_hdl,
4630 	    ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
4631 		/* just write out a path of two object numbers */
4632 		(void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
4633 		    (longlong_t)dsobj, (longlong_t)obj);
4634 		return;
4635 	}
4636 	(void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
4637 
4638 	/* find out if the dataset is mounted */
4639 	mounted = !always_unmounted && is_mounted(zhp->zpool_hdl, dsname,
4640 	    &mntpnt);
4641 
4642 	/* get the corrupted object's path */
4643 	(void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
4644 	zc.zc_obj = obj;
4645 	if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_OBJ_TO_PATH,
4646 	    &zc) == 0) {
4647 		if (mounted) {
4648 			(void) snprintf(pathname, len, "%s%s", mntpnt,
4649 			    zc.zc_value);
4650 		} else {
4651 			(void) snprintf(pathname, len, "%s:%s",
4652 			    dsname, zc.zc_value);
4653 		}
4654 	} else {
4655 		(void) snprintf(pathname, len, "%s:<0x%llx>", dsname,
4656 		    (longlong_t)obj);
4657 	}
4658 	free(mntpnt);
4659 }
4660 
4661 void
4662 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4663     char *pathname, size_t len)
4664 {
4665 	zpool_obj_to_path_impl(zhp, dsobj, obj, pathname, len, B_FALSE);
4666 }
4667 
4668 void
4669 zpool_obj_to_path_ds(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4670     char *pathname, size_t len)
4671 {
4672 	zpool_obj_to_path_impl(zhp, dsobj, obj, pathname, len, B_TRUE);
4673 }
4674 /*
4675  * Wait while the specified activity is in progress in the pool.
4676  */
4677 int
4678 zpool_wait(zpool_handle_t *zhp, zpool_wait_activity_t activity)
4679 {
4680 	boolean_t missing;
4681 
4682 	int error = zpool_wait_status(zhp, activity, &missing, NULL);
4683 
4684 	if (missing) {
4685 		(void) zpool_standard_error_fmt(zhp->zpool_hdl, ENOENT,
4686 		    dgettext(TEXT_DOMAIN, "error waiting in pool '%s'"),
4687 		    zhp->zpool_name);
4688 		return (ENOENT);
4689 	} else {
4690 		return (error);
4691 	}
4692 }
4693 
4694 /*
4695  * Wait for the given activity and return the status of the wait (whether or not
4696  * any waiting was done) in the 'waited' parameter. Non-existent pools are
4697  * reported via the 'missing' parameter, rather than by printing an error
4698  * message. This is convenient when this function is called in a loop over a
4699  * long period of time (as it is, for example, by zpool's wait cmd). In that
4700  * scenario, a pool being exported or destroyed should be considered a normal
4701  * event, so we don't want to print an error when we find that the pool doesn't
4702  * exist.
4703  */
4704 int
4705 zpool_wait_status(zpool_handle_t *zhp, zpool_wait_activity_t activity,
4706     boolean_t *missing, boolean_t *waited)
4707 {
4708 	int error = lzc_wait(zhp->zpool_name, activity, waited);
4709 	*missing = (error == ENOENT);
4710 	if (*missing)
4711 		return (0);
4712 
4713 	if (error != 0) {
4714 		(void) zpool_standard_error_fmt(zhp->zpool_hdl, error,
4715 		    dgettext(TEXT_DOMAIN, "error waiting in pool '%s'"),
4716 		    zhp->zpool_name);
4717 	}
4718 
4719 	return (error);
4720 }
4721 
4722 int
4723 zpool_set_bootenv(zpool_handle_t *zhp, const nvlist_t *envmap)
4724 {
4725 	int error = lzc_set_bootenv(zhp->zpool_name, envmap);
4726 	if (error != 0) {
4727 		(void) zpool_standard_error_fmt(zhp->zpool_hdl, error,
4728 		    dgettext(TEXT_DOMAIN,
4729 		    "error setting bootenv in pool '%s'"), zhp->zpool_name);
4730 	}
4731 
4732 	return (error);
4733 }
4734 
4735 int
4736 zpool_get_bootenv(zpool_handle_t *zhp, nvlist_t **nvlp)
4737 {
4738 	nvlist_t *nvl;
4739 	int error;
4740 
4741 	nvl = NULL;
4742 	error = lzc_get_bootenv(zhp->zpool_name, &nvl);
4743 	if (error != 0) {
4744 		(void) zpool_standard_error_fmt(zhp->zpool_hdl, error,
4745 		    dgettext(TEXT_DOMAIN,
4746 		    "error getting bootenv in pool '%s'"), zhp->zpool_name);
4747 	} else {
4748 		*nvlp = nvl;
4749 	}
4750 
4751 	return (error);
4752 }
4753 
4754 /*
4755  * Attempt to read and parse feature file(s) (from "compatibility" property).
4756  * Files contain zpool feature names, comma or whitespace-separated.
4757  * Comments (# character to next newline) are discarded.
4758  *
4759  * Arguments:
4760  *  compatibility : string containing feature filenames
4761  *  features : either NULL or pointer to array of boolean
4762  *  report : either NULL or pointer to string buffer
4763  *  rlen : length of "report" buffer
4764  *
4765  * compatibility is NULL (unset), "", "off", "legacy", or list of
4766  * comma-separated filenames. filenames should either be absolute,
4767  * or relative to:
4768  *   1) ZPOOL_SYSCONF_COMPAT_D (eg: /etc/zfs/compatibility.d) or
4769  *   2) ZPOOL_DATA_COMPAT_D (eg: /usr/share/zfs/compatibility.d).
4770  * (Unset), "" or "off" => enable all features
4771  * "legacy" => disable all features
4772  *
4773  * Any feature names read from files which match unames in spa_feature_table
4774  * will have the corresponding boolean set in the features array (if non-NULL).
4775  * If more than one feature set specified, only features present in *all* of
4776  * them will be set.
4777  *
4778  * "report" if not NULL will be populated with a suitable status message.
4779  *
4780  * Return values:
4781  *   ZPOOL_COMPATIBILITY_OK : files read and parsed ok
4782  *   ZPOOL_COMPATIBILITY_BADFILE : file too big or not a text file
4783  *   ZPOOL_COMPATIBILITY_BADTOKEN : SYSCONF file contains invalid feature name
4784  *   ZPOOL_COMPATIBILITY_WARNTOKEN : DATA file contains invalid feature name
4785  *   ZPOOL_COMPATIBILITY_NOFILES : no feature files found
4786  */
4787 zpool_compat_status_t
4788 zpool_load_compat(const char *compat, boolean_t *features, char *report,
4789     size_t rlen)
4790 {
4791 	int sdirfd, ddirfd, featfd;
4792 	struct stat fs;
4793 	char *fc;
4794 	char *ps, *ls, *ws;
4795 	char *file, *line, *word;
4796 
4797 	char l_compat[ZFS_MAXPROPLEN];
4798 
4799 	boolean_t ret_nofiles = B_TRUE;
4800 	boolean_t ret_badfile = B_FALSE;
4801 	boolean_t ret_badtoken = B_FALSE;
4802 	boolean_t ret_warntoken = B_FALSE;
4803 
4804 	/* special cases (unset), "" and "off" => enable all features */
4805 	if (compat == NULL || compat[0] == '\0' ||
4806 	    strcmp(compat, ZPOOL_COMPAT_OFF) == 0) {
4807 		if (features != NULL)
4808 			for (uint_t i = 0; i < SPA_FEATURES; i++)
4809 				features[i] = B_TRUE;
4810 		if (report != NULL)
4811 			strlcpy(report, gettext("all features enabled"), rlen);
4812 		return (ZPOOL_COMPATIBILITY_OK);
4813 	}
4814 
4815 	/* Final special case "legacy" => disable all features */
4816 	if (strcmp(compat, ZPOOL_COMPAT_LEGACY) == 0) {
4817 		if (features != NULL)
4818 			for (uint_t i = 0; i < SPA_FEATURES; i++)
4819 				features[i] = B_FALSE;
4820 		if (report != NULL)
4821 			strlcpy(report, gettext("all features disabled"), rlen);
4822 		return (ZPOOL_COMPATIBILITY_OK);
4823 	}
4824 
4825 	/*
4826 	 * Start with all true; will be ANDed with results from each file
4827 	 */
4828 	if (features != NULL)
4829 		for (uint_t i = 0; i < SPA_FEATURES; i++)
4830 			features[i] = B_TRUE;
4831 
4832 	char err_badfile[1024] = "";
4833 	char err_badtoken[1024] = "";
4834 
4835 	/*
4836 	 * We ignore errors from the directory open()
4837 	 * as they're only needed if the filename is relative
4838 	 * which will be checked during the openat().
4839 	 */
4840 
4841 /* O_PATH safer than O_RDONLY if system allows it */
4842 #if defined(O_PATH)
4843 #define	ZC_DIR_FLAGS (O_DIRECTORY | O_CLOEXEC | O_PATH)
4844 #else
4845 #define	ZC_DIR_FLAGS (O_DIRECTORY | O_CLOEXEC | O_RDONLY)
4846 #endif
4847 
4848 	sdirfd = open(ZPOOL_SYSCONF_COMPAT_D, ZC_DIR_FLAGS);
4849 	ddirfd = open(ZPOOL_DATA_COMPAT_D, ZC_DIR_FLAGS);
4850 
4851 	(void) strlcpy(l_compat, compat, ZFS_MAXPROPLEN);
4852 
4853 	for (file = strtok_r(l_compat, ",", &ps);
4854 	    file != NULL;
4855 	    file = strtok_r(NULL, ",", &ps)) {
4856 
4857 		boolean_t l_features[SPA_FEATURES];
4858 
4859 		enum { Z_SYSCONF, Z_DATA } source;
4860 
4861 		/* try sysconfdir first, then datadir */
4862 		source = Z_SYSCONF;
4863 		if ((featfd = openat(sdirfd, file, O_RDONLY | O_CLOEXEC)) < 0) {
4864 			featfd = openat(ddirfd, file, O_RDONLY | O_CLOEXEC);
4865 			source = Z_DATA;
4866 		}
4867 
4868 		/* File readable and correct size? */
4869 		if (featfd < 0 ||
4870 		    fstat(featfd, &fs) < 0 ||
4871 		    fs.st_size < 1 ||
4872 		    fs.st_size > ZPOOL_COMPAT_MAXSIZE) {
4873 			(void) close(featfd);
4874 			strlcat(err_badfile, file, ZFS_MAXPROPLEN);
4875 			strlcat(err_badfile, " ", ZFS_MAXPROPLEN);
4876 			ret_badfile = B_TRUE;
4877 			continue;
4878 		}
4879 
4880 /* Prefault the file if system allows */
4881 #if defined(MAP_POPULATE)
4882 #define	ZC_MMAP_FLAGS (MAP_PRIVATE | MAP_POPULATE)
4883 #elif defined(MAP_PREFAULT_READ)
4884 #define	ZC_MMAP_FLAGS (MAP_PRIVATE | MAP_PREFAULT_READ)
4885 #else
4886 #define	ZC_MMAP_FLAGS (MAP_PRIVATE)
4887 #endif
4888 
4889 		/* private mmap() so we can strtok safely */
4890 		fc = (char *)mmap(NULL, fs.st_size, PROT_READ | PROT_WRITE,
4891 		    ZC_MMAP_FLAGS, featfd, 0);
4892 		(void) close(featfd);
4893 
4894 		/* map ok, and last character == newline? */
4895 		if (fc == MAP_FAILED || fc[fs.st_size - 1] != '\n') {
4896 			(void) munmap((void *) fc, fs.st_size);
4897 			strlcat(err_badfile, file, ZFS_MAXPROPLEN);
4898 			strlcat(err_badfile, " ", ZFS_MAXPROPLEN);
4899 			ret_badfile = B_TRUE;
4900 			continue;
4901 		}
4902 
4903 		ret_nofiles = B_FALSE;
4904 
4905 		for (uint_t i = 0; i < SPA_FEATURES; i++)
4906 			l_features[i] = B_FALSE;
4907 
4908 		/* replace final newline with NULL to ensure string ends */
4909 		fc[fs.st_size - 1] = '\0';
4910 
4911 		for (line = strtok_r(fc, "\n", &ls);
4912 		    line != NULL;
4913 		    line = strtok_r(NULL, "\n", &ls)) {
4914 			/* discard comments */
4915 			char *r = strchr(line, '#');
4916 			if (r != NULL)
4917 				*r = '\0';
4918 
4919 			for (word = strtok_r(line, ", \t", &ws);
4920 			    word != NULL;
4921 			    word = strtok_r(NULL, ", \t", &ws)) {
4922 				/* Find matching feature name */
4923 				uint_t f;
4924 				for (f = 0; f < SPA_FEATURES; f++) {
4925 					zfeature_info_t *fi =
4926 					    &spa_feature_table[f];
4927 					if (strcmp(word, fi->fi_uname) == 0) {
4928 						l_features[f] = B_TRUE;
4929 						break;
4930 					}
4931 				}
4932 				if (f < SPA_FEATURES)
4933 					continue;
4934 
4935 				/* found an unrecognized word */
4936 				/* lightly sanitize it */
4937 				if (strlen(word) > 32)
4938 					word[32] = '\0';
4939 				for (char *c = word; *c != '\0'; c++)
4940 					if (!isprint(*c))
4941 						*c = '?';
4942 
4943 				strlcat(err_badtoken, word, ZFS_MAXPROPLEN);
4944 				strlcat(err_badtoken, " ", ZFS_MAXPROPLEN);
4945 				if (source == Z_SYSCONF)
4946 					ret_badtoken = B_TRUE;
4947 				else
4948 					ret_warntoken = B_TRUE;
4949 			}
4950 		}
4951 		(void) munmap((void *) fc, fs.st_size);
4952 
4953 		if (features != NULL)
4954 			for (uint_t i = 0; i < SPA_FEATURES; i++)
4955 				features[i] &= l_features[i];
4956 	}
4957 	(void) close(sdirfd);
4958 	(void) close(ddirfd);
4959 
4960 	/* Return the most serious error */
4961 	if (ret_badfile) {
4962 		if (report != NULL)
4963 			snprintf(report, rlen, gettext("could not read/"
4964 			    "parse feature file(s): %s"), err_badfile);
4965 		return (ZPOOL_COMPATIBILITY_BADFILE);
4966 	}
4967 	if (ret_nofiles) {
4968 		if (report != NULL)
4969 			strlcpy(report,
4970 			    gettext("no valid compatibility files specified"),
4971 			    rlen);
4972 		return (ZPOOL_COMPATIBILITY_NOFILES);
4973 	}
4974 	if (ret_badtoken) {
4975 		if (report != NULL)
4976 			snprintf(report, rlen, gettext("invalid feature "
4977 			    "name(s) in local compatibility files: %s"),
4978 			    err_badtoken);
4979 		return (ZPOOL_COMPATIBILITY_BADTOKEN);
4980 	}
4981 	if (ret_warntoken) {
4982 		if (report != NULL)
4983 			snprintf(report, rlen, gettext("unrecognized feature "
4984 			    "name(s) in distribution compatibility files: %s"),
4985 			    err_badtoken);
4986 		return (ZPOOL_COMPATIBILITY_WARNTOKEN);
4987 	}
4988 	if (report != NULL)
4989 		strlcpy(report, gettext("compatibility set ok"), rlen);
4990 	return (ZPOOL_COMPATIBILITY_OK);
4991 }
4992 
4993 static int
4994 zpool_vdev_guid(zpool_handle_t *zhp, const char *vdevname, uint64_t *vdev_guid)
4995 {
4996 	nvlist_t *tgt;
4997 	boolean_t avail_spare, l2cache;
4998 
4999 	verify(zhp != NULL);
5000 	if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
5001 		char errbuf[1024];
5002 		(void) snprintf(errbuf, sizeof (errbuf),
5003 		    dgettext(TEXT_DOMAIN, "pool is in an unavailable state"));
5004 		return (zfs_error(zhp->zpool_hdl, EZFS_POOLUNAVAIL, errbuf));
5005 	}
5006 
5007 	if ((tgt = zpool_find_vdev(zhp, vdevname, &avail_spare, &l2cache,
5008 	    NULL)) == NULL) {
5009 		char errbuf[1024];
5010 		(void) snprintf(errbuf, sizeof (errbuf),
5011 		    dgettext(TEXT_DOMAIN, "can not find %s in %s"),
5012 		    vdevname, zhp->zpool_name);
5013 		return (zfs_error(zhp->zpool_hdl, EZFS_NODEVICE, errbuf));
5014 	}
5015 
5016 	*vdev_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
5017 	return (0);
5018 }
5019 
5020 /*
5021  * Get a vdev property value for 'prop' and return the value in
5022  * a pre-allocated buffer.
5023  */
5024 int
5025 zpool_get_vdev_prop_value(nvlist_t *nvprop, vdev_prop_t prop, char *prop_name,
5026     char *buf, size_t len, zprop_source_t *srctype, boolean_t literal)
5027 {
5028 	nvlist_t *nv;
5029 	char *strval;
5030 	uint64_t intval;
5031 	zprop_source_t src = ZPROP_SRC_NONE;
5032 
5033 	if (prop == VDEV_PROP_USER) {
5034 		/* user property, prop_name must contain the property name */
5035 		assert(prop_name != NULL);
5036 		if (nvlist_lookup_nvlist(nvprop, prop_name, &nv) == 0) {
5037 			src = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
5038 			strval = fnvlist_lookup_string(nv, ZPROP_VALUE);
5039 		} else {
5040 			/* user prop not found */
5041 			return (-1);
5042 		}
5043 		(void) strlcpy(buf, strval, len);
5044 		if (srctype)
5045 			*srctype = src;
5046 		return (0);
5047 	}
5048 
5049 	if (prop_name == NULL)
5050 		prop_name = (char *)vdev_prop_to_name(prop);
5051 
5052 	switch (vdev_prop_get_type(prop)) {
5053 	case PROP_TYPE_STRING:
5054 		if (nvlist_lookup_nvlist(nvprop, prop_name, &nv) == 0) {
5055 			src = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
5056 			strval = fnvlist_lookup_string(nv, ZPROP_VALUE);
5057 		} else {
5058 			src = ZPROP_SRC_DEFAULT;
5059 			if ((strval = (char *)vdev_prop_default_string(prop))
5060 			    == NULL)
5061 				strval = "-";
5062 		}
5063 		(void) strlcpy(buf, strval, len);
5064 		break;
5065 
5066 	case PROP_TYPE_NUMBER:
5067 		if (nvlist_lookup_nvlist(nvprop, prop_name, &nv) == 0) {
5068 			src = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
5069 			intval = fnvlist_lookup_uint64(nv, ZPROP_VALUE);
5070 		} else {
5071 			src = ZPROP_SRC_DEFAULT;
5072 			intval = vdev_prop_default_numeric(prop);
5073 		}
5074 
5075 		switch (prop) {
5076 		case VDEV_PROP_ASIZE:
5077 		case VDEV_PROP_PSIZE:
5078 		case VDEV_PROP_SIZE:
5079 		case VDEV_PROP_BOOTSIZE:
5080 		case VDEV_PROP_ALLOCATED:
5081 		case VDEV_PROP_FREE:
5082 		case VDEV_PROP_READ_ERRORS:
5083 		case VDEV_PROP_WRITE_ERRORS:
5084 		case VDEV_PROP_CHECKSUM_ERRORS:
5085 		case VDEV_PROP_INITIALIZE_ERRORS:
5086 		case VDEV_PROP_OPS_NULL:
5087 		case VDEV_PROP_OPS_READ:
5088 		case VDEV_PROP_OPS_WRITE:
5089 		case VDEV_PROP_OPS_FREE:
5090 		case VDEV_PROP_OPS_CLAIM:
5091 		case VDEV_PROP_OPS_TRIM:
5092 		case VDEV_PROP_BYTES_NULL:
5093 		case VDEV_PROP_BYTES_READ:
5094 		case VDEV_PROP_BYTES_WRITE:
5095 		case VDEV_PROP_BYTES_FREE:
5096 		case VDEV_PROP_BYTES_CLAIM:
5097 		case VDEV_PROP_BYTES_TRIM:
5098 			if (literal) {
5099 				(void) snprintf(buf, len, "%llu",
5100 				    (u_longlong_t)intval);
5101 			} else {
5102 				(void) zfs_nicenum(intval, buf, len);
5103 			}
5104 			break;
5105 		case VDEV_PROP_EXPANDSZ:
5106 			if (intval == 0) {
5107 				(void) strlcpy(buf, "-", len);
5108 			} else if (literal) {
5109 				(void) snprintf(buf, len, "%llu",
5110 				    (u_longlong_t)intval);
5111 			} else {
5112 				(void) zfs_nicenum(intval, buf, len);
5113 			}
5114 			break;
5115 		case VDEV_PROP_CAPACITY:
5116 			if (literal) {
5117 				(void) snprintf(buf, len, "%llu",
5118 				    (u_longlong_t)intval);
5119 			} else {
5120 				(void) snprintf(buf, len, "%llu%%",
5121 				    (u_longlong_t)intval);
5122 			}
5123 			break;
5124 		case VDEV_PROP_FRAGMENTATION:
5125 			if (intval == UINT64_MAX) {
5126 				(void) strlcpy(buf, "-", len);
5127 			} else {
5128 				(void) snprintf(buf, len, "%llu%%",
5129 				    (u_longlong_t)intval);
5130 			}
5131 			break;
5132 		case VDEV_PROP_STATE:
5133 			if (literal) {
5134 				(void) snprintf(buf, len, "%llu",
5135 				    (u_longlong_t)intval);
5136 			} else {
5137 				(void) strlcpy(buf, zpool_state_to_name(intval,
5138 				    VDEV_AUX_NONE), len);
5139 			}
5140 			break;
5141 		default:
5142 			(void) snprintf(buf, len, "%llu",
5143 			    (u_longlong_t)intval);
5144 		}
5145 		break;
5146 
5147 	case PROP_TYPE_INDEX:
5148 		if (nvlist_lookup_nvlist(nvprop, prop_name, &nv) == 0) {
5149 			src = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
5150 			intval = fnvlist_lookup_uint64(nv, ZPROP_VALUE);
5151 		} else {
5152 			src = ZPROP_SRC_DEFAULT;
5153 			intval = vdev_prop_default_numeric(prop);
5154 		}
5155 		if (vdev_prop_index_to_string(prop, intval,
5156 		    (const char **)&strval) != 0)
5157 			return (-1);
5158 		(void) strlcpy(buf, strval, len);
5159 		break;
5160 
5161 	default:
5162 		abort();
5163 	}
5164 
5165 	if (srctype)
5166 		*srctype = src;
5167 
5168 	return (0);
5169 }
5170 
5171 /*
5172  * Get a vdev property value for 'prop_name' and return the value in
5173  * a pre-allocated buffer.
5174  */
5175 int
5176 zpool_get_vdev_prop(zpool_handle_t *zhp, const char *vdevname, vdev_prop_t prop,
5177     char *prop_name, char *buf, size_t len, zprop_source_t *srctype,
5178     boolean_t literal)
5179 {
5180 	nvlist_t *reqnvl, *reqprops;
5181 	nvlist_t *retprops = NULL;
5182 	uint64_t vdev_guid = 0;
5183 	int ret;
5184 
5185 	if ((ret = zpool_vdev_guid(zhp, vdevname, &vdev_guid)) != 0)
5186 		return (ret);
5187 
5188 	if (nvlist_alloc(&reqnvl, NV_UNIQUE_NAME, 0) != 0)
5189 		return (no_memory(zhp->zpool_hdl));
5190 	if (nvlist_alloc(&reqprops, NV_UNIQUE_NAME, 0) != 0)
5191 		return (no_memory(zhp->zpool_hdl));
5192 
5193 	fnvlist_add_uint64(reqnvl, ZPOOL_VDEV_PROPS_GET_VDEV, vdev_guid);
5194 
5195 	if (prop != VDEV_PROP_USER) {
5196 		/* prop_name overrides prop value */
5197 		if (prop_name != NULL)
5198 			prop = vdev_name_to_prop(prop_name);
5199 		else
5200 			prop_name = (char *)vdev_prop_to_name(prop);
5201 		assert(prop < VDEV_NUM_PROPS);
5202 	}
5203 
5204 	assert(prop_name != NULL);
5205 	if (nvlist_add_uint64(reqprops, prop_name, prop) != 0) {
5206 		nvlist_free(reqnvl);
5207 		nvlist_free(reqprops);
5208 		return (no_memory(zhp->zpool_hdl));
5209 	}
5210 
5211 	fnvlist_add_nvlist(reqnvl, ZPOOL_VDEV_PROPS_GET_PROPS, reqprops);
5212 
5213 	ret = lzc_get_vdev_prop(zhp->zpool_name, reqnvl, &retprops);
5214 
5215 	if (ret == 0) {
5216 		ret = zpool_get_vdev_prop_value(retprops, prop, prop_name, buf,
5217 		    len, srctype, literal);
5218 	} else {
5219 		char errbuf[1024];
5220 		(void) snprintf(errbuf, sizeof (errbuf),
5221 		    dgettext(TEXT_DOMAIN, "cannot get vdev property %s from"
5222 		    " %s in %s"), prop_name, vdevname, zhp->zpool_name);
5223 		(void) zpool_standard_error(zhp->zpool_hdl, ret, errbuf);
5224 	}
5225 
5226 	nvlist_free(reqnvl);
5227 	nvlist_free(reqprops);
5228 	nvlist_free(retprops);
5229 
5230 	return (ret);
5231 }
5232 
5233 /*
5234  * Get all vdev properties
5235  */
5236 int
5237 zpool_get_all_vdev_props(zpool_handle_t *zhp, const char *vdevname,
5238     nvlist_t **outnvl)
5239 {
5240 	nvlist_t *nvl = NULL;
5241 	uint64_t vdev_guid = 0;
5242 	int ret;
5243 
5244 	if ((ret = zpool_vdev_guid(zhp, vdevname, &vdev_guid)) != 0)
5245 		return (ret);
5246 
5247 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
5248 		return (no_memory(zhp->zpool_hdl));
5249 
5250 	fnvlist_add_uint64(nvl, ZPOOL_VDEV_PROPS_GET_VDEV, vdev_guid);
5251 
5252 	ret = lzc_get_vdev_prop(zhp->zpool_name, nvl, outnvl);
5253 
5254 	nvlist_free(nvl);
5255 
5256 	if (ret) {
5257 		char errbuf[1024];
5258 		(void) snprintf(errbuf, sizeof (errbuf),
5259 		    dgettext(TEXT_DOMAIN, "cannot get vdev properties for"
5260 		    " %s in %s"), vdevname, zhp->zpool_name);
5261 		(void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
5262 	}
5263 
5264 	return (ret);
5265 }
5266 
5267 /*
5268  * Set vdev property
5269  */
5270 int
5271 zpool_set_vdev_prop(zpool_handle_t *zhp, const char *vdevname,
5272     const char *propname, const char *propval)
5273 {
5274 	int ret;
5275 	nvlist_t *nvl = NULL;
5276 	nvlist_t *outnvl = NULL;
5277 	nvlist_t *props;
5278 	nvlist_t *realprops;
5279 	prop_flags_t flags = { 0 };
5280 	uint64_t version;
5281 	uint64_t vdev_guid;
5282 
5283 	if ((ret = zpool_vdev_guid(zhp, vdevname, &vdev_guid)) != 0)
5284 		return (ret);
5285 
5286 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
5287 		return (no_memory(zhp->zpool_hdl));
5288 	if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0)
5289 		return (no_memory(zhp->zpool_hdl));
5290 
5291 	fnvlist_add_uint64(nvl, ZPOOL_VDEV_PROPS_SET_VDEV, vdev_guid);
5292 
5293 	if (nvlist_add_string(props, propname, propval) != 0) {
5294 		nvlist_free(props);
5295 		return (no_memory(zhp->zpool_hdl));
5296 	}
5297 
5298 	char errbuf[1024];
5299 	(void) snprintf(errbuf, sizeof (errbuf),
5300 	    dgettext(TEXT_DOMAIN, "cannot set property %s for %s on %s"),
5301 	    propname, vdevname, zhp->zpool_name);
5302 
5303 	flags.vdevprop = 1;
5304 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
5305 	if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
5306 	    zhp->zpool_name, props, version, flags, errbuf)) == NULL) {
5307 		nvlist_free(props);
5308 		nvlist_free(nvl);
5309 		return (-1);
5310 	}
5311 
5312 	nvlist_free(props);
5313 	props = realprops;
5314 
5315 	fnvlist_add_nvlist(nvl, ZPOOL_VDEV_PROPS_SET_PROPS, props);
5316 
5317 	ret = lzc_set_vdev_prop(zhp->zpool_name, nvl, &outnvl);
5318 
5319 	nvlist_free(props);
5320 	nvlist_free(nvl);
5321 	nvlist_free(outnvl);
5322 
5323 	if (ret)
5324 		(void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
5325 
5326 	return (ret);
5327 }
5328