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 https://opensource.org/licenses/CDDL-1.0.
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 	unsigned int create:1;	/* Validate property on creation */
64 	unsigned int import:1;	/* Validate property on import */
65 	unsigned 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 	const 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 = 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 		case ZPOOL_PROP_BCLONESAVED:
343 		case ZPOOL_PROP_BCLONEUSED:
344 			if (literal)
345 				(void) snprintf(buf, len, "%llu",
346 				    (u_longlong_t)intval);
347 			else
348 				(void) zfs_nicenum(intval, buf, len);
349 			break;
350 
351 		case ZPOOL_PROP_EXPANDSZ:
352 		case ZPOOL_PROP_CHECKPOINT:
353 			if (intval == 0) {
354 				(void) strlcpy(buf, "-", len);
355 			} else if (literal) {
356 				(void) snprintf(buf, len, "%llu",
357 				    (u_longlong_t)intval);
358 			} else {
359 				(void) zfs_nicebytes(intval, buf, len);
360 			}
361 			break;
362 
363 		case ZPOOL_PROP_CAPACITY:
364 			if (literal) {
365 				(void) snprintf(buf, len, "%llu",
366 				    (u_longlong_t)intval);
367 			} else {
368 				(void) snprintf(buf, len, "%llu%%",
369 				    (u_longlong_t)intval);
370 			}
371 			break;
372 
373 		case ZPOOL_PROP_FRAGMENTATION:
374 			if (intval == UINT64_MAX) {
375 				(void) strlcpy(buf, "-", len);
376 			} else if (literal) {
377 				(void) snprintf(buf, len, "%llu",
378 				    (u_longlong_t)intval);
379 			} else {
380 				(void) snprintf(buf, len, "%llu%%",
381 				    (u_longlong_t)intval);
382 			}
383 			break;
384 
385 		case ZPOOL_PROP_BCLONERATIO:
386 		case ZPOOL_PROP_DEDUPRATIO:
387 			if (literal)
388 				(void) snprintf(buf, len, "%llu.%02llu",
389 				    (u_longlong_t)(intval / 100),
390 				    (u_longlong_t)(intval % 100));
391 			else
392 				(void) snprintf(buf, len, "%llu.%02llux",
393 				    (u_longlong_t)(intval / 100),
394 				    (u_longlong_t)(intval % 100));
395 			break;
396 
397 		case ZPOOL_PROP_HEALTH:
398 			(void) strlcpy(buf, zpool_get_state_str(zhp), len);
399 			break;
400 		case ZPOOL_PROP_VERSION:
401 			if (intval >= SPA_VERSION_FEATURES) {
402 				(void) snprintf(buf, len, "-");
403 				break;
404 			}
405 			zfs_fallthrough;
406 		default:
407 			(void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
408 		}
409 		break;
410 
411 	case PROP_TYPE_INDEX:
412 		intval = zpool_get_prop_int(zhp, prop, &src);
413 		if (zpool_prop_index_to_string(prop, intval, &strval)
414 		    != 0)
415 			return (-1);
416 		(void) strlcpy(buf, strval, len);
417 		break;
418 
419 	default:
420 		abort();
421 	}
422 
423 	if (srctype)
424 		*srctype = src;
425 
426 	return (0);
427 }
428 
429 /*
430  * Check if the bootfs name has the same pool name as it is set to.
431  * Assuming bootfs is a valid dataset name.
432  */
433 static boolean_t
434 bootfs_name_valid(const char *pool, const char *bootfs)
435 {
436 	int len = strlen(pool);
437 	if (bootfs[0] == '\0')
438 		return (B_TRUE);
439 
440 	if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
441 		return (B_FALSE);
442 
443 	if (strncmp(pool, bootfs, len) == 0 &&
444 	    (bootfs[len] == '/' || bootfs[len] == '\0'))
445 		return (B_TRUE);
446 
447 	return (B_FALSE);
448 }
449 
450 /*
451  * Given an nvlist of zpool properties to be set, validate that they are
452  * correct, and parse any numeric properties (index, boolean, etc) if they are
453  * specified as strings.
454  */
455 static nvlist_t *
456 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
457     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
458 {
459 	nvpair_t *elem;
460 	nvlist_t *retprops;
461 	zpool_prop_t prop;
462 	const char *strval;
463 	uint64_t intval;
464 	const char *slash, *check;
465 	struct stat64 statbuf;
466 	zpool_handle_t *zhp;
467 	char report[1024];
468 
469 	if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
470 		(void) no_memory(hdl);
471 		return (NULL);
472 	}
473 
474 	elem = NULL;
475 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
476 		const char *propname = nvpair_name(elem);
477 
478 		if (flags.vdevprop && zpool_prop_vdev(propname)) {
479 			vdev_prop_t vprop = vdev_name_to_prop(propname);
480 
481 			if (vdev_prop_readonly(vprop)) {
482 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
483 				    "is readonly"), propname);
484 				(void) zfs_error(hdl, EZFS_PROPREADONLY,
485 				    errbuf);
486 				goto error;
487 			}
488 
489 			if (zprop_parse_value(hdl, elem, vprop, ZFS_TYPE_VDEV,
490 			    retprops, &strval, &intval, errbuf) != 0)
491 				goto error;
492 
493 			continue;
494 		} else if (flags.vdevprop && vdev_prop_user(propname)) {
495 			if (nvlist_add_nvpair(retprops, elem) != 0) {
496 				(void) no_memory(hdl);
497 				goto error;
498 			}
499 			continue;
500 		} else if (flags.vdevprop) {
501 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
502 			    "invalid property: '%s'"), propname);
503 			(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
504 			goto error;
505 		}
506 
507 		prop = zpool_name_to_prop(propname);
508 		if (prop == ZPOOL_PROP_INVAL && zpool_prop_feature(propname)) {
509 			int err;
510 			char *fname = strchr(propname, '@') + 1;
511 
512 			err = zfeature_lookup_name(fname, NULL);
513 			if (err != 0) {
514 				ASSERT3U(err, ==, ENOENT);
515 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
516 				    "feature '%s' unsupported by kernel"),
517 				    fname);
518 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
519 				goto error;
520 			}
521 
522 			if (nvpair_type(elem) != DATA_TYPE_STRING) {
523 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
524 				    "'%s' must be a string"), propname);
525 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
526 				goto error;
527 			}
528 
529 			(void) nvpair_value_string(elem, &strval);
530 			if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0 &&
531 			    strcmp(strval, ZFS_FEATURE_DISABLED) != 0) {
532 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
533 				    "property '%s' can only be set to "
534 				    "'enabled' or 'disabled'"), propname);
535 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
536 				goto error;
537 			}
538 
539 			if (!flags.create &&
540 			    strcmp(strval, ZFS_FEATURE_DISABLED) == 0) {
541 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
542 				    "property '%s' can only be set to "
543 				    "'disabled' at creation time"), propname);
544 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
545 				goto error;
546 			}
547 
548 			if (nvlist_add_uint64(retprops, propname, 0) != 0) {
549 				(void) no_memory(hdl);
550 				goto error;
551 			}
552 			continue;
553 		}
554 
555 		/*
556 		 * Make sure this property is valid and applies to this type.
557 		 */
558 		if (prop == ZPOOL_PROP_INVAL) {
559 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
560 			    "invalid property '%s'"), propname);
561 			(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
562 			goto error;
563 		}
564 
565 		if (zpool_prop_readonly(prop)) {
566 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
567 			    "is readonly"), propname);
568 			(void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
569 			goto error;
570 		}
571 
572 		if (!flags.create && zpool_prop_setonce(prop)) {
573 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
574 			    "property '%s' can only be set at "
575 			    "creation time"), propname);
576 			(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
577 			goto error;
578 		}
579 
580 		if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
581 		    &strval, &intval, errbuf) != 0)
582 			goto error;
583 
584 		/*
585 		 * Perform additional checking for specific properties.
586 		 */
587 		switch (prop) {
588 		case ZPOOL_PROP_VERSION:
589 			if (intval < version ||
590 			    !SPA_VERSION_IS_SUPPORTED(intval)) {
591 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
592 				    "property '%s' number %llu is invalid."),
593 				    propname, (unsigned long long)intval);
594 				(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
595 				goto error;
596 			}
597 			break;
598 
599 		case ZPOOL_PROP_ASHIFT:
600 			if (intval != 0 &&
601 			    (intval < ASHIFT_MIN || intval > ASHIFT_MAX)) {
602 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
603 				    "property '%s' number %llu is invalid, "
604 				    "only values between %" PRId32 " and %"
605 				    PRId32 " are allowed."),
606 				    propname, (unsigned long long)intval,
607 				    ASHIFT_MIN, ASHIFT_MAX);
608 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
609 				goto error;
610 			}
611 			break;
612 
613 		case ZPOOL_PROP_BOOTFS:
614 			if (flags.create || flags.import) {
615 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
616 				    "property '%s' cannot be set at creation "
617 				    "or import time"), propname);
618 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
619 				goto error;
620 			}
621 
622 			if (version < SPA_VERSION_BOOTFS) {
623 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
624 				    "pool must be upgraded to support "
625 				    "'%s' property"), propname);
626 				(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
627 				goto error;
628 			}
629 
630 			/*
631 			 * bootfs property value has to be a dataset name and
632 			 * the dataset has to be in the same pool as it sets to.
633 			 */
634 			if (!bootfs_name_valid(poolname, strval)) {
635 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
636 				    "is an invalid name"), strval);
637 				(void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
638 				goto error;
639 			}
640 
641 			if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
642 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
643 				    "could not open pool '%s'"), poolname);
644 				(void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
645 				goto error;
646 			}
647 			zpool_close(zhp);
648 			break;
649 
650 		case ZPOOL_PROP_ALTROOT:
651 			if (!flags.create && !flags.import) {
652 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
653 				    "property '%s' can only be set during pool "
654 				    "creation or import"), propname);
655 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
656 				goto error;
657 			}
658 
659 			if (strval[0] != '/') {
660 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
661 				    "bad alternate root '%s'"), strval);
662 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
663 				goto error;
664 			}
665 			break;
666 
667 		case ZPOOL_PROP_CACHEFILE:
668 			if (strval[0] == '\0')
669 				break;
670 
671 			if (strcmp(strval, "none") == 0)
672 				break;
673 
674 			if (strval[0] != '/') {
675 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
676 				    "property '%s' must be empty, an "
677 				    "absolute path, or 'none'"), propname);
678 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
679 				goto error;
680 			}
681 
682 			slash = strrchr(strval, '/');
683 
684 			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
685 			    strcmp(slash, "/..") == 0) {
686 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
687 				    "'%s' is not a valid file"), strval);
688 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
689 				goto error;
690 			}
691 
692 			*(char *)slash = '\0';
693 
694 			if (strval[0] != '\0' &&
695 			    (stat64(strval, &statbuf) != 0 ||
696 			    !S_ISDIR(statbuf.st_mode))) {
697 				*(char *)slash = '/';
698 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
699 				    "'%s' is not a valid directory"),
700 				    strval);
701 				(void) zfs_error(hdl, EZFS_BADPATH, errbuf);
702 				goto error;
703 			}
704 
705 			*(char *)slash = '/';
706 			break;
707 
708 		case ZPOOL_PROP_COMPATIBILITY:
709 			switch (zpool_load_compat(strval, NULL, report, 1024)) {
710 			case ZPOOL_COMPATIBILITY_OK:
711 			case ZPOOL_COMPATIBILITY_WARNTOKEN:
712 				break;
713 			case ZPOOL_COMPATIBILITY_BADFILE:
714 			case ZPOOL_COMPATIBILITY_BADTOKEN:
715 			case ZPOOL_COMPATIBILITY_NOFILES:
716 				zfs_error_aux(hdl, "%s", report);
717 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
718 				goto error;
719 			}
720 			break;
721 
722 		case ZPOOL_PROP_COMMENT:
723 			for (check = strval; *check != '\0'; check++) {
724 				if (!isprint(*check)) {
725 					zfs_error_aux(hdl,
726 					    dgettext(TEXT_DOMAIN,
727 					    "comment may only have printable "
728 					    "characters"));
729 					(void) zfs_error(hdl, EZFS_BADPROP,
730 					    errbuf);
731 					goto error;
732 				}
733 			}
734 			if (strlen(strval) > ZPROP_MAX_COMMENT) {
735 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
736 				    "comment must not exceed %d characters"),
737 				    ZPROP_MAX_COMMENT);
738 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
739 				goto error;
740 			}
741 			break;
742 		case ZPOOL_PROP_READONLY:
743 			if (!flags.import) {
744 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
745 				    "property '%s' can only be set at "
746 				    "import time"), propname);
747 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
748 				goto error;
749 			}
750 			break;
751 		case ZPOOL_PROP_MULTIHOST:
752 			if (get_system_hostid() == 0) {
753 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
754 				    "requires a non-zero system hostid"));
755 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
756 				goto error;
757 			}
758 			break;
759 		case ZPOOL_PROP_DEDUPDITTO:
760 			printf("Note: property '%s' no longer has "
761 			    "any effect\n", propname);
762 			break;
763 
764 		default:
765 			break;
766 		}
767 	}
768 
769 	return (retprops);
770 error:
771 	nvlist_free(retprops);
772 	return (NULL);
773 }
774 
775 /*
776  * Set zpool property : propname=propval.
777  */
778 int
779 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
780 {
781 	zfs_cmd_t zc = {"\0"};
782 	int ret = -1;
783 	char errbuf[ERRBUFLEN];
784 	nvlist_t *nvl = NULL;
785 	nvlist_t *realprops;
786 	uint64_t version;
787 	prop_flags_t flags = { 0 };
788 
789 	(void) snprintf(errbuf, sizeof (errbuf),
790 	    dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
791 	    zhp->zpool_name);
792 
793 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
794 		return (no_memory(zhp->zpool_hdl));
795 
796 	if (nvlist_add_string(nvl, propname, propval) != 0) {
797 		nvlist_free(nvl);
798 		return (no_memory(zhp->zpool_hdl));
799 	}
800 
801 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
802 	if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
803 	    zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
804 		nvlist_free(nvl);
805 		return (-1);
806 	}
807 
808 	nvlist_free(nvl);
809 	nvl = realprops;
810 
811 	/*
812 	 * Execute the corresponding ioctl() to set this property.
813 	 */
814 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
815 
816 	zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl);
817 
818 	ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
819 
820 	zcmd_free_nvlists(&zc);
821 	nvlist_free(nvl);
822 
823 	if (ret)
824 		(void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
825 	else
826 		(void) zpool_props_refresh(zhp);
827 
828 	return (ret);
829 }
830 
831 int
832 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp,
833     zfs_type_t type, boolean_t literal)
834 {
835 	libzfs_handle_t *hdl = zhp->zpool_hdl;
836 	zprop_list_t *entry;
837 	char buf[ZFS_MAXPROPLEN];
838 	nvlist_t *features = NULL;
839 	nvpair_t *nvp;
840 	zprop_list_t **last;
841 	boolean_t firstexpand = (NULL == *plp);
842 	int i;
843 
844 	if (zprop_expand_list(hdl, plp, type) != 0)
845 		return (-1);
846 
847 	if (type == ZFS_TYPE_VDEV)
848 		return (0);
849 
850 	last = plp;
851 	while (*last != NULL)
852 		last = &(*last)->pl_next;
853 
854 	if ((*plp)->pl_all)
855 		features = zpool_get_features(zhp);
856 
857 	if ((*plp)->pl_all && firstexpand) {
858 		for (i = 0; i < SPA_FEATURES; i++) {
859 			zprop_list_t *entry = zfs_alloc(hdl,
860 			    sizeof (zprop_list_t));
861 			entry->pl_prop = ZPROP_USERPROP;
862 			entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
863 			    spa_feature_table[i].fi_uname);
864 			entry->pl_width = strlen(entry->pl_user_prop);
865 			entry->pl_all = B_TRUE;
866 
867 			*last = entry;
868 			last = &entry->pl_next;
869 		}
870 	}
871 
872 	/* add any unsupported features */
873 	for (nvp = nvlist_next_nvpair(features, NULL);
874 	    nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
875 		char *propname;
876 		boolean_t found;
877 		zprop_list_t *entry;
878 
879 		if (zfeature_is_supported(nvpair_name(nvp)))
880 			continue;
881 
882 		propname = zfs_asprintf(hdl, "unsupported@%s",
883 		    nvpair_name(nvp));
884 
885 		/*
886 		 * Before adding the property to the list make sure that no
887 		 * other pool already added the same property.
888 		 */
889 		found = B_FALSE;
890 		entry = *plp;
891 		while (entry != NULL) {
892 			if (entry->pl_user_prop != NULL &&
893 			    strcmp(propname, entry->pl_user_prop) == 0) {
894 				found = B_TRUE;
895 				break;
896 			}
897 			entry = entry->pl_next;
898 		}
899 		if (found) {
900 			free(propname);
901 			continue;
902 		}
903 
904 		entry = zfs_alloc(hdl, sizeof (zprop_list_t));
905 		entry->pl_prop = ZPROP_USERPROP;
906 		entry->pl_user_prop = propname;
907 		entry->pl_width = strlen(entry->pl_user_prop);
908 		entry->pl_all = B_TRUE;
909 
910 		*last = entry;
911 		last = &entry->pl_next;
912 	}
913 
914 	for (entry = *plp; entry != NULL; entry = entry->pl_next) {
915 		if (entry->pl_fixed && !literal)
916 			continue;
917 
918 		if (entry->pl_prop != ZPROP_USERPROP &&
919 		    zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
920 		    NULL, literal) == 0) {
921 			if (strlen(buf) > entry->pl_width)
922 				entry->pl_width = strlen(buf);
923 		}
924 	}
925 
926 	return (0);
927 }
928 
929 int
930 vdev_expand_proplist(zpool_handle_t *zhp, const char *vdevname,
931     zprop_list_t **plp)
932 {
933 	zprop_list_t *entry;
934 	char buf[ZFS_MAXPROPLEN];
935 	const char *strval = NULL;
936 	int err = 0;
937 	nvpair_t *elem = NULL;
938 	nvlist_t *vprops = NULL;
939 	nvlist_t *propval = NULL;
940 	const char *propname;
941 	vdev_prop_t prop;
942 	zprop_list_t **last;
943 
944 	for (entry = *plp; entry != NULL; entry = entry->pl_next) {
945 		if (entry->pl_fixed)
946 			continue;
947 
948 		if (zpool_get_vdev_prop(zhp, vdevname, entry->pl_prop,
949 		    entry->pl_user_prop, buf, sizeof (buf), NULL,
950 		    B_FALSE) == 0) {
951 			if (strlen(buf) > entry->pl_width)
952 				entry->pl_width = strlen(buf);
953 		}
954 		if (entry->pl_prop == VDEV_PROP_NAME &&
955 		    strlen(vdevname) > entry->pl_width)
956 			entry->pl_width = strlen(vdevname);
957 	}
958 
959 	/* Handle the all properties case */
960 	last = plp;
961 	if (*last != NULL && (*last)->pl_all == B_TRUE) {
962 		while (*last != NULL)
963 			last = &(*last)->pl_next;
964 
965 		err = zpool_get_all_vdev_props(zhp, vdevname, &vprops);
966 		if (err != 0)
967 			return (err);
968 
969 		while ((elem = nvlist_next_nvpair(vprops, elem)) != NULL) {
970 			propname = nvpair_name(elem);
971 
972 			/* Skip properties that are not user defined */
973 			if ((prop = vdev_name_to_prop(propname)) !=
974 			    VDEV_PROP_USERPROP)
975 				continue;
976 
977 			if (nvpair_value_nvlist(elem, &propval) != 0)
978 				continue;
979 
980 			strval = fnvlist_lookup_string(propval, ZPROP_VALUE);
981 
982 			entry = zfs_alloc(zhp->zpool_hdl,
983 			    sizeof (zprop_list_t));
984 			entry->pl_prop = prop;
985 			entry->pl_user_prop = zfs_strdup(zhp->zpool_hdl,
986 			    propname);
987 			entry->pl_width = strlen(strval);
988 			entry->pl_all = B_TRUE;
989 			*last = entry;
990 			last = &entry->pl_next;
991 		}
992 	}
993 
994 	return (0);
995 }
996 
997 /*
998  * Get the state for the given feature on the given ZFS pool.
999  */
1000 int
1001 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
1002     size_t len)
1003 {
1004 	uint64_t refcount;
1005 	boolean_t found = B_FALSE;
1006 	nvlist_t *features = zpool_get_features(zhp);
1007 	boolean_t supported;
1008 	const char *feature = strchr(propname, '@') + 1;
1009 
1010 	supported = zpool_prop_feature(propname);
1011 	ASSERT(supported || zpool_prop_unsupported(propname));
1012 
1013 	/*
1014 	 * Convert from feature name to feature guid. This conversion is
1015 	 * unnecessary for unsupported@... properties because they already
1016 	 * use guids.
1017 	 */
1018 	if (supported) {
1019 		int ret;
1020 		spa_feature_t fid;
1021 
1022 		ret = zfeature_lookup_name(feature, &fid);
1023 		if (ret != 0) {
1024 			(void) strlcpy(buf, "-", len);
1025 			return (ENOTSUP);
1026 		}
1027 		feature = spa_feature_table[fid].fi_guid;
1028 	}
1029 
1030 	if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
1031 		found = B_TRUE;
1032 
1033 	if (supported) {
1034 		if (!found) {
1035 			(void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
1036 		} else  {
1037 			if (refcount == 0)
1038 				(void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
1039 			else
1040 				(void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
1041 		}
1042 	} else {
1043 		if (found) {
1044 			if (refcount == 0) {
1045 				(void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
1046 			} else {
1047 				(void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
1048 			}
1049 		} else {
1050 			(void) strlcpy(buf, "-", len);
1051 			return (ENOTSUP);
1052 		}
1053 	}
1054 
1055 	return (0);
1056 }
1057 
1058 /*
1059  * Validate the given pool name, optionally putting an extended error message in
1060  * 'buf'.
1061  */
1062 boolean_t
1063 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
1064 {
1065 	namecheck_err_t why;
1066 	char what;
1067 	int ret;
1068 
1069 	ret = pool_namecheck(pool, &why, &what);
1070 
1071 	/*
1072 	 * The rules for reserved pool names were extended at a later point.
1073 	 * But we need to support users with existing pools that may now be
1074 	 * invalid.  So we only check for this expanded set of names during a
1075 	 * create (or import), and only in userland.
1076 	 */
1077 	if (ret == 0 && !isopen &&
1078 	    (strncmp(pool, "mirror", 6) == 0 ||
1079 	    strncmp(pool, "raidz", 5) == 0 ||
1080 	    strncmp(pool, "draid", 5) == 0 ||
1081 	    strncmp(pool, "spare", 5) == 0 ||
1082 	    strcmp(pool, "log") == 0)) {
1083 		if (hdl != NULL)
1084 			zfs_error_aux(hdl,
1085 			    dgettext(TEXT_DOMAIN, "name is reserved"));
1086 		return (B_FALSE);
1087 	}
1088 
1089 
1090 	if (ret != 0) {
1091 		if (hdl != NULL) {
1092 			switch (why) {
1093 			case NAME_ERR_TOOLONG:
1094 				zfs_error_aux(hdl,
1095 				    dgettext(TEXT_DOMAIN, "name is too long"));
1096 				break;
1097 
1098 			case NAME_ERR_INVALCHAR:
1099 				zfs_error_aux(hdl,
1100 				    dgettext(TEXT_DOMAIN, "invalid character "
1101 				    "'%c' in pool name"), what);
1102 				break;
1103 
1104 			case NAME_ERR_NOLETTER:
1105 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1106 				    "name must begin with a letter"));
1107 				break;
1108 
1109 			case NAME_ERR_RESERVED:
1110 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1111 				    "name is reserved"));
1112 				break;
1113 
1114 			case NAME_ERR_DISKLIKE:
1115 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1116 				    "pool name is reserved"));
1117 				break;
1118 
1119 			case NAME_ERR_LEADING_SLASH:
1120 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1121 				    "leading slash in name"));
1122 				break;
1123 
1124 			case NAME_ERR_EMPTY_COMPONENT:
1125 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1126 				    "empty component in name"));
1127 				break;
1128 
1129 			case NAME_ERR_TRAILING_SLASH:
1130 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1131 				    "trailing slash in name"));
1132 				break;
1133 
1134 			case NAME_ERR_MULTIPLE_DELIMITERS:
1135 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1136 				    "multiple '@' and/or '#' delimiters in "
1137 				    "name"));
1138 				break;
1139 
1140 			case NAME_ERR_NO_AT:
1141 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1142 				    "permission set is missing '@'"));
1143 				break;
1144 
1145 			default:
1146 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1147 				    "(%d) not defined"), why);
1148 				break;
1149 			}
1150 		}
1151 		return (B_FALSE);
1152 	}
1153 
1154 	return (B_TRUE);
1155 }
1156 
1157 /*
1158  * Open a handle to the given pool, even if the pool is currently in the FAULTED
1159  * state.
1160  */
1161 zpool_handle_t *
1162 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
1163 {
1164 	zpool_handle_t *zhp;
1165 	boolean_t missing;
1166 
1167 	/*
1168 	 * Make sure the pool name is valid.
1169 	 */
1170 	if (!zpool_name_valid(hdl, B_TRUE, pool)) {
1171 		(void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1172 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1173 		    pool);
1174 		return (NULL);
1175 	}
1176 
1177 	zhp = zfs_alloc(hdl, sizeof (zpool_handle_t));
1178 
1179 	zhp->zpool_hdl = hdl;
1180 	(void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1181 
1182 	if (zpool_refresh_stats(zhp, &missing) != 0) {
1183 		zpool_close(zhp);
1184 		return (NULL);
1185 	}
1186 
1187 	if (missing) {
1188 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1189 		(void) zfs_error_fmt(hdl, EZFS_NOENT,
1190 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1191 		zpool_close(zhp);
1192 		return (NULL);
1193 	}
1194 
1195 	return (zhp);
1196 }
1197 
1198 /*
1199  * Like the above, but silent on error.  Used when iterating over pools (because
1200  * the configuration cache may be out of date).
1201  */
1202 int
1203 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1204 {
1205 	zpool_handle_t *zhp;
1206 	boolean_t missing;
1207 
1208 	zhp = zfs_alloc(hdl, sizeof (zpool_handle_t));
1209 
1210 	zhp->zpool_hdl = hdl;
1211 	(void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1212 
1213 	if (zpool_refresh_stats(zhp, &missing) != 0) {
1214 		zpool_close(zhp);
1215 		return (-1);
1216 	}
1217 
1218 	if (missing) {
1219 		zpool_close(zhp);
1220 		*ret = NULL;
1221 		return (0);
1222 	}
1223 
1224 	*ret = zhp;
1225 	return (0);
1226 }
1227 
1228 /*
1229  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1230  * state.
1231  */
1232 zpool_handle_t *
1233 zpool_open(libzfs_handle_t *hdl, const char *pool)
1234 {
1235 	zpool_handle_t *zhp;
1236 
1237 	if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1238 		return (NULL);
1239 
1240 	if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1241 		(void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1242 		    dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1243 		zpool_close(zhp);
1244 		return (NULL);
1245 	}
1246 
1247 	return (zhp);
1248 }
1249 
1250 /*
1251  * Close the handle.  Simply frees the memory associated with the handle.
1252  */
1253 void
1254 zpool_close(zpool_handle_t *zhp)
1255 {
1256 	nvlist_free(zhp->zpool_config);
1257 	nvlist_free(zhp->zpool_old_config);
1258 	nvlist_free(zhp->zpool_props);
1259 	free(zhp);
1260 }
1261 
1262 /*
1263  * Return the name of the pool.
1264  */
1265 const char *
1266 zpool_get_name(zpool_handle_t *zhp)
1267 {
1268 	return (zhp->zpool_name);
1269 }
1270 
1271 
1272 /*
1273  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1274  */
1275 int
1276 zpool_get_state(zpool_handle_t *zhp)
1277 {
1278 	return (zhp->zpool_state);
1279 }
1280 
1281 /*
1282  * Check if vdev list contains a special vdev
1283  */
1284 static boolean_t
1285 zpool_has_special_vdev(nvlist_t *nvroot)
1286 {
1287 	nvlist_t **child;
1288 	uint_t children;
1289 
1290 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &child,
1291 	    &children) == 0) {
1292 		for (uint_t c = 0; c < children; c++) {
1293 			const char *bias;
1294 
1295 			if (nvlist_lookup_string(child[c],
1296 			    ZPOOL_CONFIG_ALLOCATION_BIAS, &bias) == 0 &&
1297 			    strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) {
1298 				return (B_TRUE);
1299 			}
1300 		}
1301 	}
1302 	return (B_FALSE);
1303 }
1304 
1305 /*
1306  * Check if vdev list contains a dRAID vdev
1307  */
1308 static boolean_t
1309 zpool_has_draid_vdev(nvlist_t *nvroot)
1310 {
1311 	nvlist_t **child;
1312 	uint_t children;
1313 
1314 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1315 	    &child, &children) == 0) {
1316 		for (uint_t c = 0; c < children; c++) {
1317 			const char *type;
1318 
1319 			if (nvlist_lookup_string(child[c],
1320 			    ZPOOL_CONFIG_TYPE, &type) == 0 &&
1321 			    strcmp(type, VDEV_TYPE_DRAID) == 0) {
1322 				return (B_TRUE);
1323 			}
1324 		}
1325 	}
1326 	return (B_FALSE);
1327 }
1328 
1329 /*
1330  * Output a dRAID top-level vdev name in to the provided buffer.
1331  */
1332 static char *
1333 zpool_draid_name(char *name, int len, uint64_t data, uint64_t parity,
1334     uint64_t spares, uint64_t children)
1335 {
1336 	snprintf(name, len, "%s%llu:%llud:%lluc:%llus",
1337 	    VDEV_TYPE_DRAID, (u_longlong_t)parity, (u_longlong_t)data,
1338 	    (u_longlong_t)children, (u_longlong_t)spares);
1339 
1340 	return (name);
1341 }
1342 
1343 /*
1344  * Return B_TRUE if the provided name is a dRAID spare name.
1345  */
1346 boolean_t
1347 zpool_is_draid_spare(const char *name)
1348 {
1349 	uint64_t spare_id, parity, vdev_id;
1350 
1351 	if (sscanf(name, VDEV_TYPE_DRAID "%llu-%llu-%llu",
1352 	    (u_longlong_t *)&parity, (u_longlong_t *)&vdev_id,
1353 	    (u_longlong_t *)&spare_id) == 3) {
1354 		return (B_TRUE);
1355 	}
1356 
1357 	return (B_FALSE);
1358 }
1359 
1360 /*
1361  * Create the named pool, using the provided vdev list.  It is assumed
1362  * that the consumer has already validated the contents of the nvlist, so we
1363  * don't have to worry about error semantics.
1364  */
1365 int
1366 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1367     nvlist_t *props, nvlist_t *fsprops)
1368 {
1369 	zfs_cmd_t zc = {"\0"};
1370 	nvlist_t *zc_fsprops = NULL;
1371 	nvlist_t *zc_props = NULL;
1372 	nvlist_t *hidden_args = NULL;
1373 	uint8_t *wkeydata = NULL;
1374 	uint_t wkeylen = 0;
1375 	char errbuf[ERRBUFLEN];
1376 	int ret = -1;
1377 
1378 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1379 	    "cannot create '%s'"), pool);
1380 
1381 	if (!zpool_name_valid(hdl, B_FALSE, pool))
1382 		return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
1383 
1384 	zcmd_write_conf_nvlist(hdl, &zc, nvroot);
1385 
1386 	if (props) {
1387 		prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1388 
1389 		if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1390 		    SPA_VERSION_1, flags, errbuf)) == NULL) {
1391 			goto create_failed;
1392 		}
1393 	}
1394 
1395 	if (fsprops) {
1396 		uint64_t zoned;
1397 		const char *zonestr;
1398 
1399 		zoned = ((nvlist_lookup_string(fsprops,
1400 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1401 		    strcmp(zonestr, "on") == 0);
1402 
1403 		if ((zc_fsprops = zfs_valid_proplist(hdl, ZFS_TYPE_FILESYSTEM,
1404 		    fsprops, zoned, NULL, NULL, B_TRUE, errbuf)) == NULL) {
1405 			goto create_failed;
1406 		}
1407 
1408 		if (nvlist_exists(zc_fsprops,
1409 		    zfs_prop_to_name(ZFS_PROP_SPECIAL_SMALL_BLOCKS)) &&
1410 		    !zpool_has_special_vdev(nvroot)) {
1411 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1412 			    "%s property requires a special vdev"),
1413 			    zfs_prop_to_name(ZFS_PROP_SPECIAL_SMALL_BLOCKS));
1414 			(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
1415 			goto create_failed;
1416 		}
1417 
1418 		if (!zc_props &&
1419 		    (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1420 			goto create_failed;
1421 		}
1422 		if (zfs_crypto_create(hdl, NULL, zc_fsprops, props, B_TRUE,
1423 		    &wkeydata, &wkeylen) != 0) {
1424 			zfs_error(hdl, EZFS_CRYPTOFAILED, errbuf);
1425 			goto create_failed;
1426 		}
1427 		if (nvlist_add_nvlist(zc_props,
1428 		    ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1429 			goto create_failed;
1430 		}
1431 		if (wkeydata != NULL) {
1432 			if (nvlist_alloc(&hidden_args, NV_UNIQUE_NAME, 0) != 0)
1433 				goto create_failed;
1434 
1435 			if (nvlist_add_uint8_array(hidden_args, "wkeydata",
1436 			    wkeydata, wkeylen) != 0)
1437 				goto create_failed;
1438 
1439 			if (nvlist_add_nvlist(zc_props, ZPOOL_HIDDEN_ARGS,
1440 			    hidden_args) != 0)
1441 				goto create_failed;
1442 		}
1443 	}
1444 
1445 	if (zc_props)
1446 		zcmd_write_src_nvlist(hdl, &zc, zc_props);
1447 
1448 	(void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1449 
1450 	if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1451 
1452 		zcmd_free_nvlists(&zc);
1453 		nvlist_free(zc_props);
1454 		nvlist_free(zc_fsprops);
1455 		nvlist_free(hidden_args);
1456 		if (wkeydata != NULL)
1457 			free(wkeydata);
1458 
1459 		switch (errno) {
1460 		case EBUSY:
1461 			/*
1462 			 * This can happen if the user has specified the same
1463 			 * device multiple times.  We can't reliably detect this
1464 			 * until we try to add it and see we already have a
1465 			 * label.  This can also happen under if the device is
1466 			 * part of an active md or lvm device.
1467 			 */
1468 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1469 			    "one or more vdevs refer to the same device, or "
1470 			    "one of\nthe devices is part of an active md or "
1471 			    "lvm device"));
1472 			return (zfs_error(hdl, EZFS_BADDEV, errbuf));
1473 
1474 		case ERANGE:
1475 			/*
1476 			 * This happens if the record size is smaller or larger
1477 			 * than the allowed size range, or not a power of 2.
1478 			 *
1479 			 * NOTE: although zfs_valid_proplist is called earlier,
1480 			 * this case may have slipped through since the
1481 			 * pool does not exist yet and it is therefore
1482 			 * impossible to read properties e.g. max blocksize
1483 			 * from the pool.
1484 			 */
1485 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1486 			    "record size invalid"));
1487 			return (zfs_error(hdl, EZFS_BADPROP, errbuf));
1488 
1489 		case EOVERFLOW:
1490 			/*
1491 			 * This occurs when one of the devices is below
1492 			 * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1493 			 * device was the problem device since there's no
1494 			 * reliable way to determine device size from userland.
1495 			 */
1496 			{
1497 				char buf[64];
1498 
1499 				zfs_nicebytes(SPA_MINDEVSIZE, buf,
1500 				    sizeof (buf));
1501 
1502 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1503 				    "one or more devices is less than the "
1504 				    "minimum size (%s)"), buf);
1505 			}
1506 			return (zfs_error(hdl, EZFS_BADDEV, errbuf));
1507 
1508 		case ENOSPC:
1509 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1510 			    "one or more devices is out of space"));
1511 			return (zfs_error(hdl, EZFS_BADDEV, errbuf));
1512 
1513 		case EINVAL:
1514 			if (zpool_has_draid_vdev(nvroot) &&
1515 			    zfeature_lookup_name("draid", NULL) != 0) {
1516 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1517 				    "dRAID vdevs are unsupported by the "
1518 				    "kernel"));
1519 				return (zfs_error(hdl, EZFS_BADDEV, errbuf));
1520 			} else {
1521 				return (zpool_standard_error(hdl, errno,
1522 				    errbuf));
1523 			}
1524 
1525 		default:
1526 			return (zpool_standard_error(hdl, errno, errbuf));
1527 		}
1528 	}
1529 
1530 create_failed:
1531 	zcmd_free_nvlists(&zc);
1532 	nvlist_free(zc_props);
1533 	nvlist_free(zc_fsprops);
1534 	nvlist_free(hidden_args);
1535 	if (wkeydata != NULL)
1536 		free(wkeydata);
1537 	return (ret);
1538 }
1539 
1540 /*
1541  * Destroy the given pool.  It is up to the caller to ensure that there are no
1542  * datasets left in the pool.
1543  */
1544 int
1545 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1546 {
1547 	zfs_cmd_t zc = {"\0"};
1548 	zfs_handle_t *zfp = NULL;
1549 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1550 	char errbuf[ERRBUFLEN];
1551 
1552 	if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1553 	    (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1554 		return (-1);
1555 
1556 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1557 	zc.zc_history = (uint64_t)(uintptr_t)log_str;
1558 
1559 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1560 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1561 		    "cannot destroy '%s'"), zhp->zpool_name);
1562 
1563 		if (errno == EROFS) {
1564 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1565 			    "one or more devices is read only"));
1566 			(void) zfs_error(hdl, EZFS_BADDEV, errbuf);
1567 		} else {
1568 			(void) zpool_standard_error(hdl, errno, errbuf);
1569 		}
1570 
1571 		if (zfp)
1572 			zfs_close(zfp);
1573 		return (-1);
1574 	}
1575 
1576 	if (zfp) {
1577 		remove_mountpoint(zfp);
1578 		zfs_close(zfp);
1579 	}
1580 
1581 	return (0);
1582 }
1583 
1584 /*
1585  * Create a checkpoint in the given pool.
1586  */
1587 int
1588 zpool_checkpoint(zpool_handle_t *zhp)
1589 {
1590 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1591 	char errbuf[ERRBUFLEN];
1592 	int error;
1593 
1594 	error = lzc_pool_checkpoint(zhp->zpool_name);
1595 	if (error != 0) {
1596 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1597 		    "cannot checkpoint '%s'"), zhp->zpool_name);
1598 		(void) zpool_standard_error(hdl, error, errbuf);
1599 		return (-1);
1600 	}
1601 
1602 	return (0);
1603 }
1604 
1605 /*
1606  * Discard the checkpoint from the given pool.
1607  */
1608 int
1609 zpool_discard_checkpoint(zpool_handle_t *zhp)
1610 {
1611 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1612 	char errbuf[ERRBUFLEN];
1613 	int error;
1614 
1615 	error = lzc_pool_checkpoint_discard(zhp->zpool_name);
1616 	if (error != 0) {
1617 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1618 		    "cannot discard checkpoint in '%s'"), zhp->zpool_name);
1619 		(void) zpool_standard_error(hdl, error, errbuf);
1620 		return (-1);
1621 	}
1622 
1623 	return (0);
1624 }
1625 
1626 /*
1627  * Add the given vdevs to the pool.  The caller must have already performed the
1628  * necessary verification to ensure that the vdev specification is well-formed.
1629  */
1630 int
1631 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1632 {
1633 	zfs_cmd_t zc = {"\0"};
1634 	int ret;
1635 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1636 	char errbuf[ERRBUFLEN];
1637 	nvlist_t **spares, **l2cache;
1638 	uint_t nspares, nl2cache;
1639 
1640 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1641 	    "cannot add to '%s'"), zhp->zpool_name);
1642 
1643 	if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1644 	    SPA_VERSION_SPARES &&
1645 	    nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1646 	    &spares, &nspares) == 0) {
1647 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1648 		    "upgraded to add hot spares"));
1649 		return (zfs_error(hdl, EZFS_BADVERSION, errbuf));
1650 	}
1651 
1652 	if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1653 	    SPA_VERSION_L2CACHE &&
1654 	    nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1655 	    &l2cache, &nl2cache) == 0) {
1656 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1657 		    "upgraded to add cache devices"));
1658 		return (zfs_error(hdl, EZFS_BADVERSION, errbuf));
1659 	}
1660 
1661 	zcmd_write_conf_nvlist(hdl, &zc, nvroot);
1662 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1663 
1664 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1665 		switch (errno) {
1666 		case EBUSY:
1667 			/*
1668 			 * This can happen if the user has specified the same
1669 			 * device multiple times.  We can't reliably detect this
1670 			 * until we try to add it and see we already have a
1671 			 * label.
1672 			 */
1673 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1674 			    "one or more vdevs refer to the same device"));
1675 			(void) zfs_error(hdl, EZFS_BADDEV, errbuf);
1676 			break;
1677 
1678 		case EINVAL:
1679 
1680 			if (zpool_has_draid_vdev(nvroot) &&
1681 			    zfeature_lookup_name("draid", NULL) != 0) {
1682 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1683 				    "dRAID vdevs are unsupported by the "
1684 				    "kernel"));
1685 			} else {
1686 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1687 				    "invalid config; a pool with removing/"
1688 				    "removed vdevs does not support adding "
1689 				    "raidz or dRAID vdevs"));
1690 			}
1691 
1692 			(void) zfs_error(hdl, EZFS_BADDEV, errbuf);
1693 			break;
1694 
1695 		case EOVERFLOW:
1696 			/*
1697 			 * This occurs when one of the devices is below
1698 			 * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1699 			 * device was the problem device since there's no
1700 			 * reliable way to determine device size from userland.
1701 			 */
1702 			{
1703 				char buf[64];
1704 
1705 				zfs_nicebytes(SPA_MINDEVSIZE, buf,
1706 				    sizeof (buf));
1707 
1708 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1709 				    "device is less than the minimum "
1710 				    "size (%s)"), buf);
1711 			}
1712 			(void) zfs_error(hdl, EZFS_BADDEV, errbuf);
1713 			break;
1714 
1715 		case ENOTSUP:
1716 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1717 			    "pool must be upgraded to add these vdevs"));
1718 			(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
1719 			break;
1720 
1721 		default:
1722 			(void) zpool_standard_error(hdl, errno, errbuf);
1723 		}
1724 
1725 		ret = -1;
1726 	} else {
1727 		ret = 0;
1728 	}
1729 
1730 	zcmd_free_nvlists(&zc);
1731 
1732 	return (ret);
1733 }
1734 
1735 /*
1736  * Exports the pool from the system.  The caller must ensure that there are no
1737  * mounted datasets in the pool.
1738  */
1739 static int
1740 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1741     const char *log_str)
1742 {
1743 	zfs_cmd_t zc = {"\0"};
1744 
1745 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1746 	zc.zc_cookie = force;
1747 	zc.zc_guid = hardforce;
1748 	zc.zc_history = (uint64_t)(uintptr_t)log_str;
1749 
1750 	if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1751 		switch (errno) {
1752 		case EXDEV:
1753 			zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1754 			    "use '-f' to override the following errors:\n"
1755 			    "'%s' has an active shared spare which could be"
1756 			    " used by other pools once '%s' is exported."),
1757 			    zhp->zpool_name, zhp->zpool_name);
1758 			return (zfs_error_fmt(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1759 			    dgettext(TEXT_DOMAIN, "cannot export '%s'"),
1760 			    zhp->zpool_name));
1761 		default:
1762 			return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1763 			    dgettext(TEXT_DOMAIN, "cannot export '%s'"),
1764 			    zhp->zpool_name));
1765 		}
1766 	}
1767 
1768 	return (0);
1769 }
1770 
1771 int
1772 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1773 {
1774 	return (zpool_export_common(zhp, force, B_FALSE, log_str));
1775 }
1776 
1777 int
1778 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1779 {
1780 	return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1781 }
1782 
1783 static void
1784 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1785     nvlist_t *config)
1786 {
1787 	nvlist_t *nv = NULL;
1788 	uint64_t rewindto;
1789 	int64_t loss = -1;
1790 	struct tm t;
1791 	char timestr[128];
1792 
1793 	if (!hdl->libzfs_printerr || config == NULL)
1794 		return;
1795 
1796 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1797 	    nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1798 		return;
1799 	}
1800 
1801 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1802 		return;
1803 	(void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1804 
1805 	if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1806 	    strftime(timestr, 128, "%c", &t) != 0) {
1807 		if (dryrun) {
1808 			(void) printf(dgettext(TEXT_DOMAIN,
1809 			    "Would be able to return %s "
1810 			    "to its state as of %s.\n"),
1811 			    name, timestr);
1812 		} else {
1813 			(void) printf(dgettext(TEXT_DOMAIN,
1814 			    "Pool %s returned to its state as of %s.\n"),
1815 			    name, timestr);
1816 		}
1817 		if (loss > 120) {
1818 			(void) printf(dgettext(TEXT_DOMAIN,
1819 			    "%s approximately %lld "),
1820 			    dryrun ? "Would discard" : "Discarded",
1821 			    ((longlong_t)loss + 30) / 60);
1822 			(void) printf(dgettext(TEXT_DOMAIN,
1823 			    "minutes of transactions.\n"));
1824 		} else if (loss > 0) {
1825 			(void) printf(dgettext(TEXT_DOMAIN,
1826 			    "%s approximately %lld "),
1827 			    dryrun ? "Would discard" : "Discarded",
1828 			    (longlong_t)loss);
1829 			(void) printf(dgettext(TEXT_DOMAIN,
1830 			    "seconds of transactions.\n"));
1831 		}
1832 	}
1833 }
1834 
1835 void
1836 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1837     nvlist_t *config)
1838 {
1839 	nvlist_t *nv = NULL;
1840 	int64_t loss = -1;
1841 	uint64_t edata = UINT64_MAX;
1842 	uint64_t rewindto;
1843 	struct tm t;
1844 	char timestr[128];
1845 
1846 	if (!hdl->libzfs_printerr)
1847 		return;
1848 
1849 	if (reason >= 0)
1850 		(void) printf(dgettext(TEXT_DOMAIN, "action: "));
1851 	else
1852 		(void) printf(dgettext(TEXT_DOMAIN, "\t"));
1853 
1854 	/* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1855 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1856 	    nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1857 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1858 		goto no_info;
1859 
1860 	(void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1861 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1862 	    &edata);
1863 
1864 	(void) printf(dgettext(TEXT_DOMAIN,
1865 	    "Recovery is possible, but will result in some data loss.\n"));
1866 
1867 	if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1868 	    strftime(timestr, 128, "%c", &t) != 0) {
1869 		(void) printf(dgettext(TEXT_DOMAIN,
1870 		    "\tReturning the pool to its state as of %s\n"
1871 		    "\tshould correct the problem.  "),
1872 		    timestr);
1873 	} else {
1874 		(void) printf(dgettext(TEXT_DOMAIN,
1875 		    "\tReverting the pool to an earlier state "
1876 		    "should correct the problem.\n\t"));
1877 	}
1878 
1879 	if (loss > 120) {
1880 		(void) printf(dgettext(TEXT_DOMAIN,
1881 		    "Approximately %lld minutes of data\n"
1882 		    "\tmust be discarded, irreversibly.  "),
1883 		    ((longlong_t)loss + 30) / 60);
1884 	} else if (loss > 0) {
1885 		(void) printf(dgettext(TEXT_DOMAIN,
1886 		    "Approximately %lld seconds of data\n"
1887 		    "\tmust be discarded, irreversibly.  "),
1888 		    (longlong_t)loss);
1889 	}
1890 	if (edata != 0 && edata != UINT64_MAX) {
1891 		if (edata == 1) {
1892 			(void) printf(dgettext(TEXT_DOMAIN,
1893 			    "After rewind, at least\n"
1894 			    "\tone persistent user-data error will remain.  "));
1895 		} else {
1896 			(void) printf(dgettext(TEXT_DOMAIN,
1897 			    "After rewind, several\n"
1898 			    "\tpersistent user-data errors will remain.  "));
1899 		}
1900 	}
1901 	(void) printf(dgettext(TEXT_DOMAIN,
1902 	    "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1903 	    reason >= 0 ? "clear" : "import", name);
1904 
1905 	(void) printf(dgettext(TEXT_DOMAIN,
1906 	    "A scrub of the pool\n"
1907 	    "\tis strongly recommended after recovery.\n"));
1908 	return;
1909 
1910 no_info:
1911 	(void) printf(dgettext(TEXT_DOMAIN,
1912 	    "Destroy and re-create the pool from\n\ta backup source.\n"));
1913 }
1914 
1915 /*
1916  * zpool_import() is a contracted interface. Should be kept the same
1917  * if possible.
1918  *
1919  * Applications should use zpool_import_props() to import a pool with
1920  * new properties value to be set.
1921  */
1922 int
1923 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1924     char *altroot)
1925 {
1926 	nvlist_t *props = NULL;
1927 	int ret;
1928 
1929 	if (altroot != NULL) {
1930 		if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1931 			return (zfs_error_fmt(hdl, EZFS_NOMEM,
1932 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1933 			    newname));
1934 		}
1935 
1936 		if (nvlist_add_string(props,
1937 		    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1938 		    nvlist_add_string(props,
1939 		    zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1940 			nvlist_free(props);
1941 			return (zfs_error_fmt(hdl, EZFS_NOMEM,
1942 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1943 			    newname));
1944 		}
1945 	}
1946 
1947 	ret = zpool_import_props(hdl, config, newname, props,
1948 	    ZFS_IMPORT_NORMAL);
1949 	nvlist_free(props);
1950 	return (ret);
1951 }
1952 
1953 static void
1954 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1955     int indent)
1956 {
1957 	nvlist_t **child;
1958 	uint_t c, children;
1959 	char *vname;
1960 	uint64_t is_log = 0;
1961 
1962 	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1963 	    &is_log);
1964 
1965 	if (name != NULL)
1966 		(void) printf("\t%*s%s%s\n", indent, "", name,
1967 		    is_log ? " [log]" : "");
1968 
1969 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1970 	    &child, &children) != 0)
1971 		return;
1972 
1973 	for (c = 0; c < children; c++) {
1974 		vname = zpool_vdev_name(hdl, NULL, child[c], VDEV_NAME_TYPE_ID);
1975 		print_vdev_tree(hdl, vname, child[c], indent + 2);
1976 		free(vname);
1977 	}
1978 }
1979 
1980 void
1981 zpool_print_unsup_feat(nvlist_t *config)
1982 {
1983 	nvlist_t *nvinfo, *unsup_feat;
1984 
1985 	nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
1986 	unsup_feat = fnvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT);
1987 
1988 	for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL);
1989 	    nvp != NULL; nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1990 		const char *desc = fnvpair_value_string(nvp);
1991 		if (strlen(desc) > 0)
1992 			(void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1993 		else
1994 			(void) printf("\t%s\n", nvpair_name(nvp));
1995 	}
1996 }
1997 
1998 /*
1999  * Import the given pool using the known configuration and a list of
2000  * properties to be set. The configuration should have come from
2001  * zpool_find_import(). The 'newname' parameters control whether the pool
2002  * is imported with a different name.
2003  */
2004 int
2005 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
2006     nvlist_t *props, int flags)
2007 {
2008 	zfs_cmd_t zc = {"\0"};
2009 	zpool_load_policy_t policy;
2010 	nvlist_t *nv = NULL;
2011 	nvlist_t *nvinfo = NULL;
2012 	nvlist_t *missing = NULL;
2013 	const char *thename;
2014 	const char *origname;
2015 	int ret;
2016 	int error = 0;
2017 	char errbuf[ERRBUFLEN];
2018 
2019 	origname = fnvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME);
2020 
2021 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2022 	    "cannot import pool '%s'"), origname);
2023 
2024 	if (newname != NULL) {
2025 		if (!zpool_name_valid(hdl, B_FALSE, newname))
2026 			return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
2027 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
2028 			    newname));
2029 		thename = newname;
2030 	} else {
2031 		thename = origname;
2032 	}
2033 
2034 	if (props != NULL) {
2035 		uint64_t version;
2036 		prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2037 
2038 		version = fnvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION);
2039 
2040 		if ((props = zpool_valid_proplist(hdl, origname,
2041 		    props, version, flags, errbuf)) == NULL)
2042 			return (-1);
2043 		zcmd_write_src_nvlist(hdl, &zc, props);
2044 		nvlist_free(props);
2045 	}
2046 
2047 	(void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
2048 
2049 	zc.zc_guid = fnvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID);
2050 
2051 	zcmd_write_conf_nvlist(hdl, &zc, config);
2052 	zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2);
2053 
2054 	zc.zc_cookie = flags;
2055 	while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
2056 	    errno == ENOMEM)
2057 		zcmd_expand_dst_nvlist(hdl, &zc);
2058 	if (ret != 0)
2059 		error = errno;
2060 
2061 	(void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
2062 
2063 	zcmd_free_nvlists(&zc);
2064 
2065 	zpool_get_load_policy(config, &policy);
2066 
2067 	if (error) {
2068 		char desc[1024];
2069 		char aux[256];
2070 
2071 		/*
2072 		 * Dry-run failed, but we print out what success
2073 		 * looks like if we found a best txg
2074 		 */
2075 		if (policy.zlp_rewind & ZPOOL_TRY_REWIND) {
2076 			zpool_rewind_exclaim(hdl, newname ? origname : thename,
2077 			    B_TRUE, nv);
2078 			nvlist_free(nv);
2079 			return (-1);
2080 		}
2081 
2082 		if (newname == NULL)
2083 			(void) snprintf(desc, sizeof (desc),
2084 			    dgettext(TEXT_DOMAIN, "cannot import '%s'"),
2085 			    thename);
2086 		else
2087 			(void) snprintf(desc, sizeof (desc),
2088 			    dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
2089 			    origname, thename);
2090 
2091 		switch (error) {
2092 		case ENOTSUP:
2093 			if (nv != NULL && nvlist_lookup_nvlist(nv,
2094 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
2095 			    nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
2096 				(void) printf(dgettext(TEXT_DOMAIN, "This "
2097 				    "pool uses the following feature(s) not "
2098 				    "supported by this system:\n"));
2099 				zpool_print_unsup_feat(nv);
2100 				if (nvlist_exists(nvinfo,
2101 				    ZPOOL_CONFIG_CAN_RDONLY)) {
2102 					(void) printf(dgettext(TEXT_DOMAIN,
2103 					    "All unsupported features are only "
2104 					    "required for writing to the pool."
2105 					    "\nThe pool can be imported using "
2106 					    "'-o readonly=on'.\n"));
2107 				}
2108 			}
2109 			/*
2110 			 * Unsupported version.
2111 			 */
2112 			(void) zfs_error(hdl, EZFS_BADVERSION, desc);
2113 			break;
2114 
2115 		case EREMOTEIO:
2116 			if (nv != NULL && nvlist_lookup_nvlist(nv,
2117 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0) {
2118 				const char *hostname = "<unknown>";
2119 				uint64_t hostid = 0;
2120 				mmp_state_t mmp_state;
2121 
2122 				mmp_state = fnvlist_lookup_uint64(nvinfo,
2123 				    ZPOOL_CONFIG_MMP_STATE);
2124 
2125 				if (nvlist_exists(nvinfo,
2126 				    ZPOOL_CONFIG_MMP_HOSTNAME))
2127 					hostname = fnvlist_lookup_string(nvinfo,
2128 					    ZPOOL_CONFIG_MMP_HOSTNAME);
2129 
2130 				if (nvlist_exists(nvinfo,
2131 				    ZPOOL_CONFIG_MMP_HOSTID))
2132 					hostid = fnvlist_lookup_uint64(nvinfo,
2133 					    ZPOOL_CONFIG_MMP_HOSTID);
2134 
2135 				if (mmp_state == MMP_STATE_ACTIVE) {
2136 					(void) snprintf(aux, sizeof (aux),
2137 					    dgettext(TEXT_DOMAIN, "pool is imp"
2138 					    "orted on host '%s' (hostid=%lx).\n"
2139 					    "Export the pool on the other "
2140 					    "system, then run 'zpool import'."),
2141 					    hostname, (unsigned long) hostid);
2142 				} else if (mmp_state == MMP_STATE_NO_HOSTID) {
2143 					(void) snprintf(aux, sizeof (aux),
2144 					    dgettext(TEXT_DOMAIN, "pool has "
2145 					    "the multihost property on and "
2146 					    "the\nsystem's hostid is not set. "
2147 					    "Set a unique system hostid with "
2148 					    "the zgenhostid(8) command.\n"));
2149 				}
2150 
2151 				(void) zfs_error_aux(hdl, "%s", aux);
2152 			}
2153 			(void) zfs_error(hdl, EZFS_ACTIVE_POOL, desc);
2154 			break;
2155 
2156 		case EINVAL:
2157 			(void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
2158 			break;
2159 
2160 		case EROFS:
2161 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2162 			    "one or more devices is read only"));
2163 			(void) zfs_error(hdl, EZFS_BADDEV, desc);
2164 			break;
2165 
2166 		case ENXIO:
2167 			if (nv && nvlist_lookup_nvlist(nv,
2168 			    ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
2169 			    nvlist_lookup_nvlist(nvinfo,
2170 			    ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
2171 				(void) printf(dgettext(TEXT_DOMAIN,
2172 				    "The devices below are missing or "
2173 				    "corrupted, use '-m' to import the pool "
2174 				    "anyway:\n"));
2175 				print_vdev_tree(hdl, NULL, missing, 2);
2176 				(void) printf("\n");
2177 			}
2178 			(void) zpool_standard_error(hdl, error, desc);
2179 			break;
2180 
2181 		case EEXIST:
2182 			(void) zpool_standard_error(hdl, error, desc);
2183 			break;
2184 
2185 		case EBUSY:
2186 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2187 			    "one or more devices are already in use\n"));
2188 			(void) zfs_error(hdl, EZFS_BADDEV, desc);
2189 			break;
2190 		case ENAMETOOLONG:
2191 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2192 			    "new name of at least one dataset is longer than "
2193 			    "the maximum allowable length"));
2194 			(void) zfs_error(hdl, EZFS_NAMETOOLONG, desc);
2195 			break;
2196 		default:
2197 			(void) zpool_standard_error(hdl, error, desc);
2198 			zpool_explain_recover(hdl,
2199 			    newname ? origname : thename, -error, nv);
2200 			break;
2201 		}
2202 
2203 		nvlist_free(nv);
2204 		ret = -1;
2205 	} else {
2206 		zpool_handle_t *zhp;
2207 
2208 		/*
2209 		 * This should never fail, but play it safe anyway.
2210 		 */
2211 		if (zpool_open_silent(hdl, thename, &zhp) != 0)
2212 			ret = -1;
2213 		else if (zhp != NULL)
2214 			zpool_close(zhp);
2215 		if (policy.zlp_rewind &
2216 		    (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
2217 			zpool_rewind_exclaim(hdl, newname ? origname : thename,
2218 			    ((policy.zlp_rewind & ZPOOL_TRY_REWIND) != 0), nv);
2219 		}
2220 		nvlist_free(nv);
2221 	}
2222 
2223 	return (ret);
2224 }
2225 
2226 /*
2227  * Translate vdev names to guids.  If a vdev_path is determined to be
2228  * unsuitable then a vd_errlist is allocated and the vdev path and errno
2229  * are added to it.
2230  */
2231 static int
2232 zpool_translate_vdev_guids(zpool_handle_t *zhp, nvlist_t *vds,
2233     nvlist_t *vdev_guids, nvlist_t *guids_to_paths, nvlist_t **vd_errlist)
2234 {
2235 	nvlist_t *errlist = NULL;
2236 	int error = 0;
2237 
2238 	for (nvpair_t *elem = nvlist_next_nvpair(vds, NULL); elem != NULL;
2239 	    elem = nvlist_next_nvpair(vds, elem)) {
2240 		boolean_t spare, cache;
2241 
2242 		const char *vd_path = nvpair_name(elem);
2243 		nvlist_t *tgt = zpool_find_vdev(zhp, vd_path, &spare, &cache,
2244 		    NULL);
2245 
2246 		if ((tgt == NULL) || cache || spare) {
2247 			if (errlist == NULL) {
2248 				errlist = fnvlist_alloc();
2249 				error = EINVAL;
2250 			}
2251 
2252 			uint64_t err = (tgt == NULL) ? EZFS_NODEVICE :
2253 			    (spare ? EZFS_ISSPARE : EZFS_ISL2CACHE);
2254 			fnvlist_add_int64(errlist, vd_path, err);
2255 			continue;
2256 		}
2257 
2258 		uint64_t guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
2259 		fnvlist_add_uint64(vdev_guids, vd_path, guid);
2260 
2261 		char msg[MAXNAMELEN];
2262 		(void) snprintf(msg, sizeof (msg), "%llu", (u_longlong_t)guid);
2263 		fnvlist_add_string(guids_to_paths, msg, vd_path);
2264 	}
2265 
2266 	if (error != 0) {
2267 		verify(errlist != NULL);
2268 		if (vd_errlist != NULL)
2269 			*vd_errlist = errlist;
2270 		else
2271 			fnvlist_free(errlist);
2272 	}
2273 
2274 	return (error);
2275 }
2276 
2277 static int
2278 xlate_init_err(int err)
2279 {
2280 	switch (err) {
2281 	case ENODEV:
2282 		return (EZFS_NODEVICE);
2283 	case EINVAL:
2284 	case EROFS:
2285 		return (EZFS_BADDEV);
2286 	case EBUSY:
2287 		return (EZFS_INITIALIZING);
2288 	case ESRCH:
2289 		return (EZFS_NO_INITIALIZE);
2290 	}
2291 	return (err);
2292 }
2293 
2294 /*
2295  * Begin, suspend, or cancel the initialization (initializing of all free
2296  * blocks) for the given vdevs in the given pool.
2297  */
2298 static int
2299 zpool_initialize_impl(zpool_handle_t *zhp, pool_initialize_func_t cmd_type,
2300     nvlist_t *vds, boolean_t wait)
2301 {
2302 	int err;
2303 
2304 	nvlist_t *vdev_guids = fnvlist_alloc();
2305 	nvlist_t *guids_to_paths = fnvlist_alloc();
2306 	nvlist_t *vd_errlist = NULL;
2307 	nvlist_t *errlist;
2308 	nvpair_t *elem;
2309 
2310 	err = zpool_translate_vdev_guids(zhp, vds, vdev_guids,
2311 	    guids_to_paths, &vd_errlist);
2312 
2313 	if (err != 0) {
2314 		verify(vd_errlist != NULL);
2315 		goto list_errors;
2316 	}
2317 
2318 	err = lzc_initialize(zhp->zpool_name, cmd_type,
2319 	    vdev_guids, &errlist);
2320 
2321 	if (err != 0) {
2322 		if (errlist != NULL) {
2323 			vd_errlist = fnvlist_lookup_nvlist(errlist,
2324 			    ZPOOL_INITIALIZE_VDEVS);
2325 			goto list_errors;
2326 		}
2327 		(void) zpool_standard_error(zhp->zpool_hdl, err,
2328 		    dgettext(TEXT_DOMAIN, "operation failed"));
2329 		goto out;
2330 	}
2331 
2332 	if (wait) {
2333 		for (elem = nvlist_next_nvpair(vdev_guids, NULL); elem != NULL;
2334 		    elem = nvlist_next_nvpair(vdev_guids, elem)) {
2335 
2336 			uint64_t guid = fnvpair_value_uint64(elem);
2337 
2338 			err = lzc_wait_tag(zhp->zpool_name,
2339 			    ZPOOL_WAIT_INITIALIZE, guid, NULL);
2340 			if (err != 0) {
2341 				(void) zpool_standard_error_fmt(zhp->zpool_hdl,
2342 				    err, dgettext(TEXT_DOMAIN, "error "
2343 				    "waiting for '%s' to initialize"),
2344 				    nvpair_name(elem));
2345 
2346 				goto out;
2347 			}
2348 		}
2349 	}
2350 	goto out;
2351 
2352 list_errors:
2353 	for (elem = nvlist_next_nvpair(vd_errlist, NULL); elem != NULL;
2354 	    elem = nvlist_next_nvpair(vd_errlist, elem)) {
2355 		int64_t vd_error = xlate_init_err(fnvpair_value_int64(elem));
2356 		const char *path;
2357 
2358 		if (nvlist_lookup_string(guids_to_paths, nvpair_name(elem),
2359 		    &path) != 0)
2360 			path = nvpair_name(elem);
2361 
2362 		(void) zfs_error_fmt(zhp->zpool_hdl, vd_error,
2363 		    "cannot initialize '%s'", path);
2364 	}
2365 
2366 out:
2367 	fnvlist_free(vdev_guids);
2368 	fnvlist_free(guids_to_paths);
2369 
2370 	if (vd_errlist != NULL)
2371 		fnvlist_free(vd_errlist);
2372 
2373 	return (err == 0 ? 0 : -1);
2374 }
2375 
2376 int
2377 zpool_initialize(zpool_handle_t *zhp, pool_initialize_func_t cmd_type,
2378     nvlist_t *vds)
2379 {
2380 	return (zpool_initialize_impl(zhp, cmd_type, vds, B_FALSE));
2381 }
2382 
2383 int
2384 zpool_initialize_wait(zpool_handle_t *zhp, pool_initialize_func_t cmd_type,
2385     nvlist_t *vds)
2386 {
2387 	return (zpool_initialize_impl(zhp, cmd_type, vds, B_TRUE));
2388 }
2389 
2390 static int
2391 xlate_trim_err(int err)
2392 {
2393 	switch (err) {
2394 	case ENODEV:
2395 		return (EZFS_NODEVICE);
2396 	case EINVAL:
2397 	case EROFS:
2398 		return (EZFS_BADDEV);
2399 	case EBUSY:
2400 		return (EZFS_TRIMMING);
2401 	case ESRCH:
2402 		return (EZFS_NO_TRIM);
2403 	case EOPNOTSUPP:
2404 		return (EZFS_TRIM_NOTSUP);
2405 	}
2406 	return (err);
2407 }
2408 
2409 static int
2410 zpool_trim_wait(zpool_handle_t *zhp, nvlist_t *vdev_guids)
2411 {
2412 	int err;
2413 	nvpair_t *elem;
2414 
2415 	for (elem = nvlist_next_nvpair(vdev_guids, NULL); elem != NULL;
2416 	    elem = nvlist_next_nvpair(vdev_guids, elem)) {
2417 
2418 		uint64_t guid = fnvpair_value_uint64(elem);
2419 
2420 		err = lzc_wait_tag(zhp->zpool_name,
2421 		    ZPOOL_WAIT_TRIM, guid, NULL);
2422 		if (err != 0) {
2423 			(void) zpool_standard_error_fmt(zhp->zpool_hdl,
2424 			    err, dgettext(TEXT_DOMAIN, "error "
2425 			    "waiting to trim '%s'"), nvpair_name(elem));
2426 
2427 			return (err);
2428 		}
2429 	}
2430 	return (0);
2431 }
2432 
2433 /*
2434  * Check errlist and report any errors, omitting ones which should be
2435  * suppressed. Returns B_TRUE if any errors were reported.
2436  */
2437 static boolean_t
2438 check_trim_errs(zpool_handle_t *zhp, trimflags_t *trim_flags,
2439     nvlist_t *guids_to_paths, nvlist_t *vds, nvlist_t *errlist)
2440 {
2441 	nvpair_t *elem;
2442 	boolean_t reported_errs = B_FALSE;
2443 	int num_vds = 0;
2444 	int num_suppressed_errs = 0;
2445 
2446 	for (elem = nvlist_next_nvpair(vds, NULL);
2447 	    elem != NULL; elem = nvlist_next_nvpair(vds, elem)) {
2448 		num_vds++;
2449 	}
2450 
2451 	for (elem = nvlist_next_nvpair(errlist, NULL);
2452 	    elem != NULL; elem = nvlist_next_nvpair(errlist, elem)) {
2453 		int64_t vd_error = xlate_trim_err(fnvpair_value_int64(elem));
2454 		const char *path;
2455 
2456 		/*
2457 		 * If only the pool was specified, and it was not a secure
2458 		 * trim then suppress warnings for individual vdevs which
2459 		 * do not support trimming.
2460 		 */
2461 		if (vd_error == EZFS_TRIM_NOTSUP &&
2462 		    trim_flags->fullpool &&
2463 		    !trim_flags->secure) {
2464 			num_suppressed_errs++;
2465 			continue;
2466 		}
2467 
2468 		reported_errs = B_TRUE;
2469 		if (nvlist_lookup_string(guids_to_paths, nvpair_name(elem),
2470 		    &path) != 0)
2471 			path = nvpair_name(elem);
2472 
2473 		(void) zfs_error_fmt(zhp->zpool_hdl, vd_error,
2474 		    "cannot trim '%s'", path);
2475 	}
2476 
2477 	if (num_suppressed_errs == num_vds) {
2478 		(void) zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
2479 		    "no devices in pool support trim operations"));
2480 		(void) (zfs_error(zhp->zpool_hdl, EZFS_TRIM_NOTSUP,
2481 		    dgettext(TEXT_DOMAIN, "cannot trim")));
2482 		reported_errs = B_TRUE;
2483 	}
2484 
2485 	return (reported_errs);
2486 }
2487 
2488 /*
2489  * Begin, suspend, or cancel the TRIM (discarding of all free blocks) for
2490  * the given vdevs in the given pool.
2491  */
2492 int
2493 zpool_trim(zpool_handle_t *zhp, pool_trim_func_t cmd_type, nvlist_t *vds,
2494     trimflags_t *trim_flags)
2495 {
2496 	int err;
2497 	int retval = 0;
2498 
2499 	nvlist_t *vdev_guids = fnvlist_alloc();
2500 	nvlist_t *guids_to_paths = fnvlist_alloc();
2501 	nvlist_t *errlist = NULL;
2502 
2503 	err = zpool_translate_vdev_guids(zhp, vds, vdev_guids,
2504 	    guids_to_paths, &errlist);
2505 	if (err != 0) {
2506 		check_trim_errs(zhp, trim_flags, guids_to_paths, vds, errlist);
2507 		retval = -1;
2508 		goto out;
2509 	}
2510 
2511 	err = lzc_trim(zhp->zpool_name, cmd_type, trim_flags->rate,
2512 	    trim_flags->secure, vdev_guids, &errlist);
2513 	if (err != 0) {
2514 		nvlist_t *vd_errlist;
2515 		if (errlist != NULL && nvlist_lookup_nvlist(errlist,
2516 		    ZPOOL_TRIM_VDEVS, &vd_errlist) == 0) {
2517 			if (check_trim_errs(zhp, trim_flags, guids_to_paths,
2518 			    vds, vd_errlist)) {
2519 				retval = -1;
2520 				goto out;
2521 			}
2522 		} else {
2523 			char errbuf[ERRBUFLEN];
2524 
2525 			(void) snprintf(errbuf, sizeof (errbuf),
2526 			    dgettext(TEXT_DOMAIN, "operation failed"));
2527 			zpool_standard_error(zhp->zpool_hdl, err, errbuf);
2528 			retval = -1;
2529 			goto out;
2530 		}
2531 	}
2532 
2533 
2534 	if (trim_flags->wait)
2535 		retval = zpool_trim_wait(zhp, vdev_guids);
2536 
2537 out:
2538 	if (errlist != NULL)
2539 		fnvlist_free(errlist);
2540 	fnvlist_free(vdev_guids);
2541 	fnvlist_free(guids_to_paths);
2542 	return (retval);
2543 }
2544 
2545 /*
2546  * Scan the pool.
2547  */
2548 int
2549 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func, pool_scrub_cmd_t cmd)
2550 {
2551 	zfs_cmd_t zc = {"\0"};
2552 	char errbuf[ERRBUFLEN];
2553 	int err;
2554 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2555 
2556 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2557 	zc.zc_cookie = func;
2558 	zc.zc_flags = cmd;
2559 
2560 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0)
2561 		return (0);
2562 
2563 	err = errno;
2564 
2565 	/* ECANCELED on a scrub means we resumed a paused scrub */
2566 	if (err == ECANCELED && func == POOL_SCAN_SCRUB &&
2567 	    cmd == POOL_SCRUB_NORMAL)
2568 		return (0);
2569 
2570 	if (err == ENOENT && func != POOL_SCAN_NONE && cmd == POOL_SCRUB_NORMAL)
2571 		return (0);
2572 
2573 	if (func == POOL_SCAN_SCRUB) {
2574 		if (cmd == POOL_SCRUB_PAUSE) {
2575 			(void) snprintf(errbuf, sizeof (errbuf),
2576 			    dgettext(TEXT_DOMAIN, "cannot pause scrubbing %s"),
2577 			    zc.zc_name);
2578 		} else {
2579 			assert(cmd == POOL_SCRUB_NORMAL);
2580 			(void) snprintf(errbuf, sizeof (errbuf),
2581 			    dgettext(TEXT_DOMAIN, "cannot scrub %s"),
2582 			    zc.zc_name);
2583 		}
2584 	} else if (func == POOL_SCAN_RESILVER) {
2585 		assert(cmd == POOL_SCRUB_NORMAL);
2586 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2587 		    "cannot restart resilver on %s"), zc.zc_name);
2588 	} else if (func == POOL_SCAN_NONE) {
2589 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2590 		    "cannot cancel scrubbing %s"), zc.zc_name);
2591 	} else {
2592 		assert(!"unexpected result");
2593 	}
2594 
2595 	if (err == EBUSY) {
2596 		nvlist_t *nvroot;
2597 		pool_scan_stat_t *ps = NULL;
2598 		uint_t psc;
2599 
2600 		nvroot = fnvlist_lookup_nvlist(zhp->zpool_config,
2601 		    ZPOOL_CONFIG_VDEV_TREE);
2602 		(void) nvlist_lookup_uint64_array(nvroot,
2603 		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
2604 		if (ps && ps->pss_func == POOL_SCAN_SCRUB &&
2605 		    ps->pss_state == DSS_SCANNING) {
2606 			if (cmd == POOL_SCRUB_PAUSE)
2607 				return (zfs_error(hdl, EZFS_SCRUB_PAUSED,
2608 				    errbuf));
2609 			else
2610 				return (zfs_error(hdl, EZFS_SCRUBBING, errbuf));
2611 		} else {
2612 			return (zfs_error(hdl, EZFS_RESILVERING, errbuf));
2613 		}
2614 	} else if (err == ENOENT) {
2615 		return (zfs_error(hdl, EZFS_NO_SCRUB, errbuf));
2616 	} else if (err == ENOTSUP && func == POOL_SCAN_RESILVER) {
2617 		return (zfs_error(hdl, EZFS_NO_RESILVER_DEFER, errbuf));
2618 	} else {
2619 		return (zpool_standard_error(hdl, err, errbuf));
2620 	}
2621 }
2622 
2623 /*
2624  * Find a vdev that matches the search criteria specified. We use the
2625  * the nvpair name to determine how we should look for the device.
2626  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
2627  * spare; but FALSE if its an INUSE spare.
2628  */
2629 static nvlist_t *
2630 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
2631     boolean_t *l2cache, boolean_t *log)
2632 {
2633 	uint_t c, children;
2634 	nvlist_t **child;
2635 	nvlist_t *ret;
2636 	uint64_t is_log;
2637 	const char *srchkey;
2638 	nvpair_t *pair = nvlist_next_nvpair(search, NULL);
2639 
2640 	/* Nothing to look for */
2641 	if (search == NULL || pair == NULL)
2642 		return (NULL);
2643 
2644 	/* Obtain the key we will use to search */
2645 	srchkey = nvpair_name(pair);
2646 
2647 	switch (nvpair_type(pair)) {
2648 	case DATA_TYPE_UINT64:
2649 		if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
2650 			uint64_t srchval = fnvpair_value_uint64(pair);
2651 			uint64_t theguid = fnvlist_lookup_uint64(nv,
2652 			    ZPOOL_CONFIG_GUID);
2653 			if (theguid == srchval)
2654 				return (nv);
2655 		}
2656 		break;
2657 
2658 	case DATA_TYPE_STRING: {
2659 		const char *srchval, *val;
2660 
2661 		srchval = fnvpair_value_string(pair);
2662 		if (nvlist_lookup_string(nv, srchkey, &val) != 0)
2663 			break;
2664 
2665 		/*
2666 		 * Search for the requested value. Special cases:
2667 		 *
2668 		 * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
2669 		 *   "-part1", or "p1".  The suffix is hidden from the user,
2670 		 *   but included in the string, so this matches around it.
2671 		 * - ZPOOL_CONFIG_PATH for short names zfs_strcmp_shortname()
2672 		 *   is used to check all possible expanded paths.
2673 		 * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
2674 		 *
2675 		 * Otherwise, all other searches are simple string compares.
2676 		 */
2677 		if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0) {
2678 			uint64_t wholedisk = 0;
2679 
2680 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2681 			    &wholedisk);
2682 			if (zfs_strcmp_pathname(srchval, val, wholedisk) == 0)
2683 				return (nv);
2684 
2685 		} else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0) {
2686 			char *type, *idx, *end, *p;
2687 			uint64_t id, vdev_id;
2688 
2689 			/*
2690 			 * Determine our vdev type, keeping in mind
2691 			 * that the srchval is composed of a type and
2692 			 * vdev id pair (i.e. mirror-4).
2693 			 */
2694 			if ((type = strdup(srchval)) == NULL)
2695 				return (NULL);
2696 
2697 			if ((p = strrchr(type, '-')) == NULL) {
2698 				free(type);
2699 				break;
2700 			}
2701 			idx = p + 1;
2702 			*p = '\0';
2703 
2704 			/*
2705 			 * If the types don't match then keep looking.
2706 			 */
2707 			if (strncmp(val, type, strlen(val)) != 0) {
2708 				free(type);
2709 				break;
2710 			}
2711 
2712 			verify(zpool_vdev_is_interior(type));
2713 
2714 			id = fnvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID);
2715 			errno = 0;
2716 			vdev_id = strtoull(idx, &end, 10);
2717 
2718 			/*
2719 			 * If we are looking for a raidz and a parity is
2720 			 * specified, make sure it matches.
2721 			 */
2722 			int rzlen = strlen(VDEV_TYPE_RAIDZ);
2723 			assert(rzlen == strlen(VDEV_TYPE_DRAID));
2724 			int typlen = strlen(type);
2725 			if ((strncmp(type, VDEV_TYPE_RAIDZ, rzlen) == 0 ||
2726 			    strncmp(type, VDEV_TYPE_DRAID, rzlen) == 0) &&
2727 			    typlen != rzlen) {
2728 				uint64_t vdev_parity;
2729 				int parity = *(type + rzlen) - '0';
2730 
2731 				if (parity <= 0 || parity > 3 ||
2732 				    (typlen - rzlen) != 1) {
2733 					/*
2734 					 * Nonsense parity specified, can
2735 					 * never match
2736 					 */
2737 					free(type);
2738 					return (NULL);
2739 				}
2740 				vdev_parity = fnvlist_lookup_uint64(nv,
2741 				    ZPOOL_CONFIG_NPARITY);
2742 				if ((int)vdev_parity != parity) {
2743 					free(type);
2744 					break;
2745 				}
2746 			}
2747 
2748 			free(type);
2749 			if (errno != 0)
2750 				return (NULL);
2751 
2752 			/*
2753 			 * Now verify that we have the correct vdev id.
2754 			 */
2755 			if (vdev_id == id)
2756 				return (nv);
2757 		}
2758 
2759 		/*
2760 		 * Common case
2761 		 */
2762 		if (strcmp(srchval, val) == 0)
2763 			return (nv);
2764 		break;
2765 	}
2766 
2767 	default:
2768 		break;
2769 	}
2770 
2771 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2772 	    &child, &children) != 0)
2773 		return (NULL);
2774 
2775 	for (c = 0; c < children; c++) {
2776 		if ((ret = vdev_to_nvlist_iter(child[c], search,
2777 		    avail_spare, l2cache, NULL)) != NULL) {
2778 			/*
2779 			 * The 'is_log' value is only set for the toplevel
2780 			 * vdev, not the leaf vdevs.  So we always lookup the
2781 			 * log device from the root of the vdev tree (where
2782 			 * 'log' is non-NULL).
2783 			 */
2784 			if (log != NULL &&
2785 			    nvlist_lookup_uint64(child[c],
2786 			    ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2787 			    is_log) {
2788 				*log = B_TRUE;
2789 			}
2790 			return (ret);
2791 		}
2792 	}
2793 
2794 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2795 	    &child, &children) == 0) {
2796 		for (c = 0; c < children; c++) {
2797 			if ((ret = vdev_to_nvlist_iter(child[c], search,
2798 			    avail_spare, l2cache, NULL)) != NULL) {
2799 				*avail_spare = B_TRUE;
2800 				return (ret);
2801 			}
2802 		}
2803 	}
2804 
2805 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2806 	    &child, &children) == 0) {
2807 		for (c = 0; c < children; c++) {
2808 			if ((ret = vdev_to_nvlist_iter(child[c], search,
2809 			    avail_spare, l2cache, NULL)) != NULL) {
2810 				*l2cache = B_TRUE;
2811 				return (ret);
2812 			}
2813 		}
2814 	}
2815 
2816 	return (NULL);
2817 }
2818 
2819 /*
2820  * Given a physical path or guid, find the associated vdev.
2821  */
2822 nvlist_t *
2823 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2824     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2825 {
2826 	nvlist_t *search, *nvroot, *ret;
2827 	uint64_t guid;
2828 	char *end;
2829 
2830 	search = fnvlist_alloc();
2831 
2832 	guid = strtoull(ppath, &end, 0);
2833 	if (guid != 0 && *end == '\0') {
2834 		fnvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid);
2835 	} else {
2836 		fnvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath);
2837 	}
2838 
2839 	nvroot = fnvlist_lookup_nvlist(zhp->zpool_config,
2840 	    ZPOOL_CONFIG_VDEV_TREE);
2841 
2842 	*avail_spare = B_FALSE;
2843 	*l2cache = B_FALSE;
2844 	if (log != NULL)
2845 		*log = B_FALSE;
2846 	ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2847 	fnvlist_free(search);
2848 
2849 	return (ret);
2850 }
2851 
2852 /*
2853  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2854  */
2855 static boolean_t
2856 zpool_vdev_is_interior(const char *name)
2857 {
2858 	if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2859 	    strncmp(name, VDEV_TYPE_SPARE, strlen(VDEV_TYPE_SPARE)) == 0 ||
2860 	    strncmp(name,
2861 	    VDEV_TYPE_REPLACING, strlen(VDEV_TYPE_REPLACING)) == 0 ||
2862 	    strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2863 		return (B_TRUE);
2864 
2865 	if (strncmp(name, VDEV_TYPE_DRAID, strlen(VDEV_TYPE_DRAID)) == 0 &&
2866 	    !zpool_is_draid_spare(name))
2867 		return (B_TRUE);
2868 
2869 	return (B_FALSE);
2870 }
2871 
2872 nvlist_t *
2873 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2874     boolean_t *l2cache, boolean_t *log)
2875 {
2876 	char *end;
2877 	nvlist_t *nvroot, *search, *ret;
2878 	uint64_t guid;
2879 
2880 	search = fnvlist_alloc();
2881 
2882 	guid = strtoull(path, &end, 0);
2883 	if (guid != 0 && *end == '\0') {
2884 		fnvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid);
2885 	} else if (zpool_vdev_is_interior(path)) {
2886 		fnvlist_add_string(search, ZPOOL_CONFIG_TYPE, path);
2887 	} else {
2888 		fnvlist_add_string(search, ZPOOL_CONFIG_PATH, path);
2889 	}
2890 
2891 	nvroot = fnvlist_lookup_nvlist(zhp->zpool_config,
2892 	    ZPOOL_CONFIG_VDEV_TREE);
2893 
2894 	*avail_spare = B_FALSE;
2895 	*l2cache = B_FALSE;
2896 	if (log != NULL)
2897 		*log = B_FALSE;
2898 	ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2899 	fnvlist_free(search);
2900 
2901 	return (ret);
2902 }
2903 
2904 /*
2905  * Convert a vdev path to a GUID.  Returns GUID or 0 on error.
2906  *
2907  * If is_spare, is_l2cache, or is_log is non-NULL, then store within it
2908  * if the VDEV is a spare, l2cache, or log device.  If they're NULL then
2909  * ignore them.
2910  */
2911 static uint64_t
2912 zpool_vdev_path_to_guid_impl(zpool_handle_t *zhp, const char *path,
2913     boolean_t *is_spare, boolean_t *is_l2cache, boolean_t *is_log)
2914 {
2915 	boolean_t spare = B_FALSE, l2cache = B_FALSE, log = B_FALSE;
2916 	nvlist_t *tgt;
2917 
2918 	if ((tgt = zpool_find_vdev(zhp, path, &spare, &l2cache,
2919 	    &log)) == NULL)
2920 		return (0);
2921 
2922 	if (is_spare != NULL)
2923 		*is_spare = spare;
2924 	if (is_l2cache != NULL)
2925 		*is_l2cache = l2cache;
2926 	if (is_log != NULL)
2927 		*is_log = log;
2928 
2929 	return (fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID));
2930 }
2931 
2932 /* Convert a vdev path to a GUID.  Returns GUID or 0 on error. */
2933 uint64_t
2934 zpool_vdev_path_to_guid(zpool_handle_t *zhp, const char *path)
2935 {
2936 	return (zpool_vdev_path_to_guid_impl(zhp, path, NULL, NULL, NULL));
2937 }
2938 
2939 /*
2940  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2941  * ZFS_ONLINE_* flags.
2942  */
2943 int
2944 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2945     vdev_state_t *newstate)
2946 {
2947 	zfs_cmd_t zc = {"\0"};
2948 	char errbuf[ERRBUFLEN];
2949 	nvlist_t *tgt;
2950 	boolean_t avail_spare, l2cache, islog;
2951 	libzfs_handle_t *hdl = zhp->zpool_hdl;
2952 
2953 	if (flags & ZFS_ONLINE_EXPAND) {
2954 		(void) snprintf(errbuf, sizeof (errbuf),
2955 		    dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2956 	} else {
2957 		(void) snprintf(errbuf, sizeof (errbuf),
2958 		    dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2959 	}
2960 
2961 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2962 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2963 	    &islog)) == NULL)
2964 		return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
2965 
2966 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
2967 
2968 	if (!(flags & ZFS_ONLINE_SPARE) && avail_spare)
2969 		return (zfs_error(hdl, EZFS_ISSPARE, errbuf));
2970 
2971 #ifndef __FreeBSD__
2972 	const char *pathname;
2973 	if ((flags & ZFS_ONLINE_EXPAND ||
2974 	    zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) &&
2975 	    nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH, &pathname) == 0) {
2976 		uint64_t wholedisk = 0;
2977 
2978 		(void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2979 		    &wholedisk);
2980 
2981 		/*
2982 		 * XXX - L2ARC 1.0 devices can't support expansion.
2983 		 */
2984 		if (l2cache) {
2985 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2986 			    "cannot expand cache devices"));
2987 			return (zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf));
2988 		}
2989 
2990 		if (wholedisk) {
2991 			const char *fullpath = path;
2992 			char buf[MAXPATHLEN];
2993 			int error;
2994 
2995 			if (path[0] != '/') {
2996 				error = zfs_resolve_shortname(path, buf,
2997 				    sizeof (buf));
2998 				if (error != 0)
2999 					return (zfs_error(hdl, EZFS_NODEVICE,
3000 					    errbuf));
3001 
3002 				fullpath = buf;
3003 			}
3004 
3005 			error = zpool_relabel_disk(hdl, fullpath, errbuf);
3006 			if (error != 0)
3007 				return (error);
3008 		}
3009 	}
3010 #endif
3011 
3012 	zc.zc_cookie = VDEV_STATE_ONLINE;
3013 	zc.zc_obj = flags;
3014 
3015 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
3016 		if (errno == EINVAL) {
3017 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
3018 			    "from this pool into a new one.  Use '%s' "
3019 			    "instead"), "zpool detach");
3020 			return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, errbuf));
3021 		}
3022 		return (zpool_standard_error(hdl, errno, errbuf));
3023 	}
3024 
3025 	*newstate = zc.zc_cookie;
3026 	return (0);
3027 }
3028 
3029 /*
3030  * Take the specified vdev offline
3031  */
3032 int
3033 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
3034 {
3035 	zfs_cmd_t zc = {"\0"};
3036 	char errbuf[ERRBUFLEN];
3037 	nvlist_t *tgt;
3038 	boolean_t avail_spare, l2cache;
3039 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3040 
3041 	(void) snprintf(errbuf, sizeof (errbuf),
3042 	    dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
3043 
3044 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3045 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3046 	    NULL)) == NULL)
3047 		return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
3048 
3049 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3050 
3051 	if (avail_spare)
3052 		return (zfs_error(hdl, EZFS_ISSPARE, errbuf));
3053 
3054 	zc.zc_cookie = VDEV_STATE_OFFLINE;
3055 	zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
3056 
3057 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
3058 		return (0);
3059 
3060 	switch (errno) {
3061 	case EBUSY:
3062 
3063 		/*
3064 		 * There are no other replicas of this device.
3065 		 */
3066 		return (zfs_error(hdl, EZFS_NOREPLICAS, errbuf));
3067 
3068 	case EEXIST:
3069 		/*
3070 		 * The log device has unplayed logs
3071 		 */
3072 		return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, errbuf));
3073 
3074 	default:
3075 		return (zpool_standard_error(hdl, errno, errbuf));
3076 	}
3077 }
3078 
3079 /*
3080  * Remove the specified vdev asynchronously from the configuration, so
3081  * that it may come ONLINE if reinserted. This is called from zed on
3082  * Udev remove event.
3083  * Note: We also have a similar function zpool_vdev_remove() that
3084  * removes the vdev from the pool.
3085  */
3086 int
3087 zpool_vdev_remove_wanted(zpool_handle_t *zhp, const char *path)
3088 {
3089 	zfs_cmd_t zc = {"\0"};
3090 	char errbuf[ERRBUFLEN];
3091 	nvlist_t *tgt;
3092 	boolean_t avail_spare, l2cache;
3093 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3094 
3095 	(void) snprintf(errbuf, sizeof (errbuf),
3096 	    dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3097 
3098 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3099 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3100 	    NULL)) == NULL)
3101 		return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
3102 
3103 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3104 
3105 	zc.zc_cookie = VDEV_STATE_REMOVED;
3106 
3107 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
3108 		return (0);
3109 
3110 	return (zpool_standard_error(hdl, errno, errbuf));
3111 }
3112 
3113 /*
3114  * Mark the given vdev faulted.
3115  */
3116 int
3117 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
3118 {
3119 	zfs_cmd_t zc = {"\0"};
3120 	char errbuf[ERRBUFLEN];
3121 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3122 
3123 	(void) snprintf(errbuf, sizeof (errbuf),
3124 	    dgettext(TEXT_DOMAIN, "cannot fault %llu"), (u_longlong_t)guid);
3125 
3126 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3127 	zc.zc_guid = guid;
3128 	zc.zc_cookie = VDEV_STATE_FAULTED;
3129 	zc.zc_obj = aux;
3130 
3131 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
3132 		return (0);
3133 
3134 	switch (errno) {
3135 	case EBUSY:
3136 
3137 		/*
3138 		 * There are no other replicas of this device.
3139 		 */
3140 		return (zfs_error(hdl, EZFS_NOREPLICAS, errbuf));
3141 
3142 	default:
3143 		return (zpool_standard_error(hdl, errno, errbuf));
3144 	}
3145 
3146 }
3147 
3148 /*
3149  * Mark the given vdev degraded.
3150  */
3151 int
3152 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
3153 {
3154 	zfs_cmd_t zc = {"\0"};
3155 	char errbuf[ERRBUFLEN];
3156 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3157 
3158 	(void) snprintf(errbuf, sizeof (errbuf),
3159 	    dgettext(TEXT_DOMAIN, "cannot degrade %llu"), (u_longlong_t)guid);
3160 
3161 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3162 	zc.zc_guid = guid;
3163 	zc.zc_cookie = VDEV_STATE_DEGRADED;
3164 	zc.zc_obj = aux;
3165 
3166 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
3167 		return (0);
3168 
3169 	return (zpool_standard_error(hdl, errno, errbuf));
3170 }
3171 
3172 /*
3173  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
3174  * a hot spare.
3175  */
3176 static boolean_t
3177 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
3178 {
3179 	nvlist_t **child;
3180 	uint_t c, children;
3181 
3182 	if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
3183 	    &children) == 0) {
3184 		const char *type = fnvlist_lookup_string(search,
3185 		    ZPOOL_CONFIG_TYPE);
3186 		if ((strcmp(type, VDEV_TYPE_SPARE) == 0 ||
3187 		    strcmp(type, VDEV_TYPE_DRAID_SPARE) == 0) &&
3188 		    children == 2 && child[which] == tgt)
3189 			return (B_TRUE);
3190 
3191 		for (c = 0; c < children; c++)
3192 			if (is_replacing_spare(child[c], tgt, which))
3193 				return (B_TRUE);
3194 	}
3195 
3196 	return (B_FALSE);
3197 }
3198 
3199 /*
3200  * Attach new_disk (fully described by nvroot) to old_disk.
3201  * If 'replacing' is specified, the new disk will replace the old one.
3202  */
3203 int
3204 zpool_vdev_attach(zpool_handle_t *zhp, const char *old_disk,
3205     const char *new_disk, nvlist_t *nvroot, int replacing, boolean_t rebuild)
3206 {
3207 	zfs_cmd_t zc = {"\0"};
3208 	char errbuf[ERRBUFLEN];
3209 	int ret;
3210 	nvlist_t *tgt;
3211 	boolean_t avail_spare, l2cache, islog;
3212 	uint64_t val;
3213 	char *newname;
3214 	nvlist_t **child;
3215 	uint_t children;
3216 	nvlist_t *config_root;
3217 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3218 
3219 	if (replacing)
3220 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3221 		    "cannot replace %s with %s"), old_disk, new_disk);
3222 	else
3223 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3224 		    "cannot attach %s to %s"), new_disk, old_disk);
3225 
3226 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3227 	if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
3228 	    &islog)) == NULL)
3229 		return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
3230 
3231 	if (avail_spare)
3232 		return (zfs_error(hdl, EZFS_ISSPARE, errbuf));
3233 
3234 	if (l2cache)
3235 		return (zfs_error(hdl, EZFS_ISL2CACHE, errbuf));
3236 
3237 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3238 	zc.zc_cookie = replacing;
3239 	zc.zc_simple = rebuild;
3240 
3241 	if (rebuild &&
3242 	    zfeature_lookup_guid("org.openzfs:device_rebuild", NULL) != 0) {
3243 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3244 		    "the loaded zfs module doesn't support device rebuilds"));
3245 		return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3246 	}
3247 
3248 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3249 	    &child, &children) != 0 || children != 1) {
3250 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3251 		    "new device must be a single disk"));
3252 		return (zfs_error(hdl, EZFS_INVALCONFIG, errbuf));
3253 	}
3254 
3255 	config_root = fnvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
3256 	    ZPOOL_CONFIG_VDEV_TREE);
3257 
3258 	if ((newname = zpool_vdev_name(NULL, NULL, child[0], 0)) == NULL)
3259 		return (-1);
3260 
3261 	/*
3262 	 * If the target is a hot spare that has been swapped in, we can only
3263 	 * replace it with another hot spare.
3264 	 */
3265 	if (replacing &&
3266 	    nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
3267 	    (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
3268 	    NULL) == NULL || !avail_spare) &&
3269 	    is_replacing_spare(config_root, tgt, 1)) {
3270 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3271 		    "can only be replaced by another hot spare"));
3272 		free(newname);
3273 		return (zfs_error(hdl, EZFS_BADTARGET, errbuf));
3274 	}
3275 
3276 	free(newname);
3277 
3278 	zcmd_write_conf_nvlist(hdl, &zc, nvroot);
3279 
3280 	ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
3281 
3282 	zcmd_free_nvlists(&zc);
3283 
3284 	if (ret == 0)
3285 		return (0);
3286 
3287 	switch (errno) {
3288 	case ENOTSUP:
3289 		/*
3290 		 * Can't attach to or replace this type of vdev.
3291 		 */
3292 		if (replacing) {
3293 			uint64_t version = zpool_get_prop_int(zhp,
3294 			    ZPOOL_PROP_VERSION, NULL);
3295 
3296 			if (islog) {
3297 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3298 				    "cannot replace a log with a spare"));
3299 			} else if (rebuild) {
3300 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3301 				    "only mirror and dRAID vdevs support "
3302 				    "sequential reconstruction"));
3303 			} else if (zpool_is_draid_spare(new_disk)) {
3304 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3305 				    "dRAID spares can only replace child "
3306 				    "devices in their parent's dRAID vdev"));
3307 			} else if (version >= SPA_VERSION_MULTI_REPLACE) {
3308 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3309 				    "already in replacing/spare config; wait "
3310 				    "for completion or use 'zpool detach'"));
3311 			} else {
3312 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3313 				    "cannot replace a replacing device"));
3314 			}
3315 		} else {
3316 			char status[64] = {0};
3317 			zpool_prop_get_feature(zhp,
3318 			    "feature@device_rebuild", status, 63);
3319 			if (rebuild &&
3320 			    strncmp(status, ZFS_FEATURE_DISABLED, 64) == 0) {
3321 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3322 				    "device_rebuild feature must be enabled "
3323 				    "in order to use sequential "
3324 				    "reconstruction"));
3325 			} else {
3326 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3327 				    "can only attach to mirrors and top-level "
3328 				    "disks"));
3329 			}
3330 		}
3331 		(void) zfs_error(hdl, EZFS_BADTARGET, errbuf);
3332 		break;
3333 
3334 	case EINVAL:
3335 		/*
3336 		 * The new device must be a single disk.
3337 		 */
3338 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3339 		    "new device must be a single disk"));
3340 		(void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
3341 		break;
3342 
3343 	case EBUSY:
3344 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy, "
3345 		    "or device removal is in progress"),
3346 		    new_disk);
3347 		(void) zfs_error(hdl, EZFS_BADDEV, errbuf);
3348 		break;
3349 
3350 	case EOVERFLOW:
3351 		/*
3352 		 * The new device is too small.
3353 		 */
3354 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3355 		    "device is too small"));
3356 		(void) zfs_error(hdl, EZFS_BADDEV, errbuf);
3357 		break;
3358 
3359 	case EDOM:
3360 		/*
3361 		 * The new device has a different optimal sector size.
3362 		 */
3363 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3364 		    "new device has a different optimal sector size; use the "
3365 		    "option '-o ashift=N' to override the optimal size"));
3366 		(void) zfs_error(hdl, EZFS_BADDEV, errbuf);
3367 		break;
3368 
3369 	case ENAMETOOLONG:
3370 		/*
3371 		 * The resulting top-level vdev spec won't fit in the label.
3372 		 */
3373 		(void) zfs_error(hdl, EZFS_DEVOVERFLOW, errbuf);
3374 		break;
3375 
3376 	default:
3377 		(void) zpool_standard_error(hdl, errno, errbuf);
3378 	}
3379 
3380 	return (-1);
3381 }
3382 
3383 /*
3384  * Detach the specified device.
3385  */
3386 int
3387 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
3388 {
3389 	zfs_cmd_t zc = {"\0"};
3390 	char errbuf[ERRBUFLEN];
3391 	nvlist_t *tgt;
3392 	boolean_t avail_spare, l2cache;
3393 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3394 
3395 	(void) snprintf(errbuf, sizeof (errbuf),
3396 	    dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
3397 
3398 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3399 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3400 	    NULL)) == NULL)
3401 		return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
3402 
3403 	if (avail_spare)
3404 		return (zfs_error(hdl, EZFS_ISSPARE, errbuf));
3405 
3406 	if (l2cache)
3407 		return (zfs_error(hdl, EZFS_ISL2CACHE, errbuf));
3408 
3409 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3410 
3411 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
3412 		return (0);
3413 
3414 	switch (errno) {
3415 
3416 	case ENOTSUP:
3417 		/*
3418 		 * Can't detach from this type of vdev.
3419 		 */
3420 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
3421 		    "applicable to mirror and replacing vdevs"));
3422 		(void) zfs_error(hdl, EZFS_BADTARGET, errbuf);
3423 		break;
3424 
3425 	case EBUSY:
3426 		/*
3427 		 * There are no other replicas of this device.
3428 		 */
3429 		(void) zfs_error(hdl, EZFS_NOREPLICAS, errbuf);
3430 		break;
3431 
3432 	default:
3433 		(void) zpool_standard_error(hdl, errno, errbuf);
3434 	}
3435 
3436 	return (-1);
3437 }
3438 
3439 /*
3440  * Find a mirror vdev in the source nvlist.
3441  *
3442  * The mchild array contains a list of disks in one of the top-level mirrors
3443  * of the source pool.  The schild array contains a list of disks that the
3444  * user specified on the command line.  We loop over the mchild array to
3445  * see if any entry in the schild array matches.
3446  *
3447  * If a disk in the mchild array is found in the schild array, we return
3448  * the index of that entry.  Otherwise we return -1.
3449  */
3450 static int
3451 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
3452     nvlist_t **schild, uint_t schildren)
3453 {
3454 	uint_t mc;
3455 
3456 	for (mc = 0; mc < mchildren; mc++) {
3457 		uint_t sc;
3458 		char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
3459 		    mchild[mc], 0);
3460 
3461 		for (sc = 0; sc < schildren; sc++) {
3462 			char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
3463 			    schild[sc], 0);
3464 			boolean_t result = (strcmp(mpath, spath) == 0);
3465 
3466 			free(spath);
3467 			if (result) {
3468 				free(mpath);
3469 				return (mc);
3470 			}
3471 		}
3472 
3473 		free(mpath);
3474 	}
3475 
3476 	return (-1);
3477 }
3478 
3479 /*
3480  * Split a mirror pool.  If newroot points to null, then a new nvlist
3481  * is generated and it is the responsibility of the caller to free it.
3482  */
3483 int
3484 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
3485     nvlist_t *props, splitflags_t flags)
3486 {
3487 	zfs_cmd_t zc = {"\0"};
3488 	char errbuf[ERRBUFLEN];
3489 	const char *bias;
3490 	nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
3491 	nvlist_t **varray = NULL, *zc_props = NULL;
3492 	uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
3493 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3494 	uint64_t vers, readonly = B_FALSE;
3495 	boolean_t freelist = B_FALSE, memory_err = B_TRUE;
3496 	int retval = 0;
3497 
3498 	(void) snprintf(errbuf, sizeof (errbuf),
3499 	    dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
3500 
3501 	if (!zpool_name_valid(hdl, B_FALSE, newname))
3502 		return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3503 
3504 	if ((config = zpool_get_config(zhp, NULL)) == NULL) {
3505 		(void) fprintf(stderr, gettext("Internal error: unable to "
3506 		    "retrieve pool configuration\n"));
3507 		return (-1);
3508 	}
3509 
3510 	tree = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE);
3511 	vers = fnvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION);
3512 
3513 	if (props) {
3514 		prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
3515 		if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
3516 		    props, vers, flags, errbuf)) == NULL)
3517 			return (-1);
3518 		(void) nvlist_lookup_uint64(zc_props,
3519 		    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3520 		if (readonly) {
3521 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3522 			    "property %s can only be set at import time"),
3523 			    zpool_prop_to_name(ZPOOL_PROP_READONLY));
3524 			return (-1);
3525 		}
3526 	}
3527 
3528 	if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
3529 	    &children) != 0) {
3530 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3531 		    "Source pool is missing vdev tree"));
3532 		nvlist_free(zc_props);
3533 		return (-1);
3534 	}
3535 
3536 	varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
3537 	vcount = 0;
3538 
3539 	if (*newroot == NULL ||
3540 	    nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
3541 	    &newchild, &newchildren) != 0)
3542 		newchildren = 0;
3543 
3544 	for (c = 0; c < children; c++) {
3545 		uint64_t is_log = B_FALSE, is_hole = B_FALSE;
3546 		boolean_t is_special = B_FALSE, is_dedup = B_FALSE;
3547 		const char *type;
3548 		nvlist_t **mchild, *vdev;
3549 		uint_t mchildren;
3550 		int entry;
3551 
3552 		/*
3553 		 * Unlike cache & spares, slogs are stored in the
3554 		 * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
3555 		 */
3556 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
3557 		    &is_log);
3558 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
3559 		    &is_hole);
3560 		if (is_log || is_hole) {
3561 			/*
3562 			 * Create a hole vdev and put it in the config.
3563 			 */
3564 			if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
3565 				goto out;
3566 			if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
3567 			    VDEV_TYPE_HOLE) != 0)
3568 				goto out;
3569 			if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
3570 			    1) != 0)
3571 				goto out;
3572 			if (lastlog == 0)
3573 				lastlog = vcount;
3574 			varray[vcount++] = vdev;
3575 			continue;
3576 		}
3577 		lastlog = 0;
3578 		type = fnvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE);
3579 
3580 		if (strcmp(type, VDEV_TYPE_INDIRECT) == 0) {
3581 			vdev = child[c];
3582 			if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3583 				goto out;
3584 			continue;
3585 		} else if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
3586 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3587 			    "Source pool must be composed only of mirrors\n"));
3588 			retval = zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
3589 			goto out;
3590 		}
3591 
3592 		if (nvlist_lookup_string(child[c],
3593 		    ZPOOL_CONFIG_ALLOCATION_BIAS, &bias) == 0) {
3594 			if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
3595 				is_special = B_TRUE;
3596 			else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
3597 				is_dedup = B_TRUE;
3598 		}
3599 		verify(nvlist_lookup_nvlist_array(child[c],
3600 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3601 
3602 		/* find or add an entry for this top-level vdev */
3603 		if (newchildren > 0 &&
3604 		    (entry = find_vdev_entry(zhp, mchild, mchildren,
3605 		    newchild, newchildren)) >= 0) {
3606 			/* We found a disk that the user specified. */
3607 			vdev = mchild[entry];
3608 			++found;
3609 		} else {
3610 			/* User didn't specify a disk for this vdev. */
3611 			vdev = mchild[mchildren - 1];
3612 		}
3613 
3614 		if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3615 			goto out;
3616 
3617 		if (flags.dryrun != 0) {
3618 			if (is_dedup == B_TRUE) {
3619 				if (nvlist_add_string(varray[vcount - 1],
3620 				    ZPOOL_CONFIG_ALLOCATION_BIAS,
3621 				    VDEV_ALLOC_BIAS_DEDUP) != 0)
3622 					goto out;
3623 			} else if (is_special == B_TRUE) {
3624 				if (nvlist_add_string(varray[vcount - 1],
3625 				    ZPOOL_CONFIG_ALLOCATION_BIAS,
3626 				    VDEV_ALLOC_BIAS_SPECIAL) != 0)
3627 					goto out;
3628 			}
3629 		}
3630 	}
3631 
3632 	/* did we find every disk the user specified? */
3633 	if (found != newchildren) {
3634 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3635 		    "include at most one disk from each mirror"));
3636 		retval = zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
3637 		goto out;
3638 	}
3639 
3640 	/* Prepare the nvlist for populating. */
3641 	if (*newroot == NULL) {
3642 		if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3643 			goto out;
3644 		freelist = B_TRUE;
3645 		if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3646 		    VDEV_TYPE_ROOT) != 0)
3647 			goto out;
3648 	} else {
3649 		verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3650 	}
3651 
3652 	/* Add all the children we found */
3653 	if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
3654 	    (const nvlist_t **)varray, lastlog == 0 ? vcount : lastlog) != 0)
3655 		goto out;
3656 
3657 	/*
3658 	 * If we're just doing a dry run, exit now with success.
3659 	 */
3660 	if (flags.dryrun) {
3661 		memory_err = B_FALSE;
3662 		freelist = B_FALSE;
3663 		goto out;
3664 	}
3665 
3666 	/* now build up the config list & call the ioctl */
3667 	if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3668 		goto out;
3669 
3670 	if (nvlist_add_nvlist(newconfig,
3671 	    ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3672 	    nvlist_add_string(newconfig,
3673 	    ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3674 	    nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3675 		goto out;
3676 
3677 	/*
3678 	 * The new pool is automatically part of the namespace unless we
3679 	 * explicitly export it.
3680 	 */
3681 	if (!flags.import)
3682 		zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3683 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3684 	(void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3685 	zcmd_write_conf_nvlist(hdl, &zc, newconfig);
3686 	if (zc_props != NULL)
3687 		zcmd_write_src_nvlist(hdl, &zc, zc_props);
3688 
3689 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3690 		retval = zpool_standard_error(hdl, errno, errbuf);
3691 		goto out;
3692 	}
3693 
3694 	freelist = B_FALSE;
3695 	memory_err = B_FALSE;
3696 
3697 out:
3698 	if (varray != NULL) {
3699 		int v;
3700 
3701 		for (v = 0; v < vcount; v++)
3702 			nvlist_free(varray[v]);
3703 		free(varray);
3704 	}
3705 	zcmd_free_nvlists(&zc);
3706 	nvlist_free(zc_props);
3707 	nvlist_free(newconfig);
3708 	if (freelist) {
3709 		nvlist_free(*newroot);
3710 		*newroot = NULL;
3711 	}
3712 
3713 	if (retval != 0)
3714 		return (retval);
3715 
3716 	if (memory_err)
3717 		return (no_memory(hdl));
3718 
3719 	return (0);
3720 }
3721 
3722 /*
3723  * Remove the given device.
3724  */
3725 int
3726 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3727 {
3728 	zfs_cmd_t zc = {"\0"};
3729 	char errbuf[ERRBUFLEN];
3730 	nvlist_t *tgt;
3731 	boolean_t avail_spare, l2cache, islog;
3732 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3733 	uint64_t version;
3734 
3735 	(void) snprintf(errbuf, sizeof (errbuf),
3736 	    dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3737 
3738 	if (zpool_is_draid_spare(path)) {
3739 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3740 		    "dRAID spares cannot be removed"));
3741 		return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
3742 	}
3743 
3744 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3745 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3746 	    &islog)) == NULL)
3747 		return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
3748 
3749 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3750 	if (islog && version < SPA_VERSION_HOLES) {
3751 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3752 		    "pool must be upgraded to support log removal"));
3753 		return (zfs_error(hdl, EZFS_BADVERSION, errbuf));
3754 	}
3755 
3756 	zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3757 
3758 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3759 		return (0);
3760 
3761 	switch (errno) {
3762 
3763 	case EINVAL:
3764 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3765 		    "invalid config; all top-level vdevs must "
3766 		    "have the same sector size and not be raidz."));
3767 		(void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
3768 		break;
3769 
3770 	case EBUSY:
3771 		if (islog) {
3772 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3773 			    "Mount encrypted datasets to replay logs."));
3774 		} else {
3775 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3776 			    "Pool busy; removal may already be in progress"));
3777 		}
3778 		(void) zfs_error(hdl, EZFS_BUSY, errbuf);
3779 		break;
3780 
3781 	case EACCES:
3782 		if (islog) {
3783 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3784 			    "Mount encrypted datasets to replay logs."));
3785 			(void) zfs_error(hdl, EZFS_BUSY, errbuf);
3786 		} else {
3787 			(void) zpool_standard_error(hdl, errno, errbuf);
3788 		}
3789 		break;
3790 
3791 	default:
3792 		(void) zpool_standard_error(hdl, errno, errbuf);
3793 	}
3794 	return (-1);
3795 }
3796 
3797 int
3798 zpool_vdev_remove_cancel(zpool_handle_t *zhp)
3799 {
3800 	zfs_cmd_t zc = {{0}};
3801 	char errbuf[ERRBUFLEN];
3802 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3803 
3804 	(void) snprintf(errbuf, sizeof (errbuf),
3805 	    dgettext(TEXT_DOMAIN, "cannot cancel removal"));
3806 
3807 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3808 	zc.zc_cookie = 1;
3809 
3810 	if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3811 		return (0);
3812 
3813 	return (zpool_standard_error(hdl, errno, errbuf));
3814 }
3815 
3816 int
3817 zpool_vdev_indirect_size(zpool_handle_t *zhp, const char *path,
3818     uint64_t *sizep)
3819 {
3820 	char errbuf[ERRBUFLEN];
3821 	nvlist_t *tgt;
3822 	boolean_t avail_spare, l2cache, islog;
3823 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3824 
3825 	(void) snprintf(errbuf, sizeof (errbuf),
3826 	    dgettext(TEXT_DOMAIN, "cannot determine indirect size of %s"),
3827 	    path);
3828 
3829 	if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3830 	    &islog)) == NULL)
3831 		return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
3832 
3833 	if (avail_spare || l2cache || islog) {
3834 		*sizep = 0;
3835 		return (0);
3836 	}
3837 
3838 	if (nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_INDIRECT_SIZE, sizep) != 0) {
3839 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3840 		    "indirect size not available"));
3841 		return (zfs_error(hdl, EINVAL, errbuf));
3842 	}
3843 	return (0);
3844 }
3845 
3846 /*
3847  * Clear the errors for the pool, or the particular device if specified.
3848  */
3849 int
3850 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3851 {
3852 	zfs_cmd_t zc = {"\0"};
3853 	char errbuf[ERRBUFLEN];
3854 	nvlist_t *tgt;
3855 	zpool_load_policy_t policy;
3856 	boolean_t avail_spare, l2cache;
3857 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3858 	nvlist_t *nvi = NULL;
3859 	int error;
3860 
3861 	if (path)
3862 		(void) snprintf(errbuf, sizeof (errbuf),
3863 		    dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3864 		    path);
3865 	else
3866 		(void) snprintf(errbuf, sizeof (errbuf),
3867 		    dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3868 		    zhp->zpool_name);
3869 
3870 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3871 	if (path) {
3872 		if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3873 		    &l2cache, NULL)) == NULL)
3874 			return (zfs_error(hdl, EZFS_NODEVICE, errbuf));
3875 
3876 		/*
3877 		 * Don't allow error clearing for hot spares.  Do allow
3878 		 * error clearing for l2cache devices.
3879 		 */
3880 		if (avail_spare)
3881 			return (zfs_error(hdl, EZFS_ISSPARE, errbuf));
3882 
3883 		zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3884 	}
3885 
3886 	zpool_get_load_policy(rewindnvl, &policy);
3887 	zc.zc_cookie = policy.zlp_rewind;
3888 
3889 	zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2);
3890 	zcmd_write_src_nvlist(hdl, &zc, rewindnvl);
3891 
3892 	while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3893 	    errno == ENOMEM)
3894 		zcmd_expand_dst_nvlist(hdl, &zc);
3895 
3896 	if (!error || ((policy.zlp_rewind & ZPOOL_TRY_REWIND) &&
3897 	    errno != EPERM && errno != EACCES)) {
3898 		if (policy.zlp_rewind &
3899 		    (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3900 			(void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3901 			zpool_rewind_exclaim(hdl, zc.zc_name,
3902 			    ((policy.zlp_rewind & ZPOOL_TRY_REWIND) != 0),
3903 			    nvi);
3904 			nvlist_free(nvi);
3905 		}
3906 		zcmd_free_nvlists(&zc);
3907 		return (0);
3908 	}
3909 
3910 	zcmd_free_nvlists(&zc);
3911 	return (zpool_standard_error(hdl, errno, errbuf));
3912 }
3913 
3914 /*
3915  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3916  */
3917 int
3918 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3919 {
3920 	zfs_cmd_t zc = {"\0"};
3921 	char errbuf[ERRBUFLEN];
3922 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3923 
3924 	(void) snprintf(errbuf, sizeof (errbuf),
3925 	    dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3926 	    (u_longlong_t)guid);
3927 
3928 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3929 	zc.zc_guid = guid;
3930 	zc.zc_cookie = ZPOOL_NO_REWIND;
3931 
3932 	if (zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc) == 0)
3933 		return (0);
3934 
3935 	return (zpool_standard_error(hdl, errno, errbuf));
3936 }
3937 
3938 /*
3939  * Change the GUID for a pool.
3940  */
3941 int
3942 zpool_reguid(zpool_handle_t *zhp)
3943 {
3944 	char errbuf[ERRBUFLEN];
3945 	libzfs_handle_t *hdl = zhp->zpool_hdl;
3946 	zfs_cmd_t zc = {"\0"};
3947 
3948 	(void) snprintf(errbuf, sizeof (errbuf),
3949 	    dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3950 
3951 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3952 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3953 		return (0);
3954 
3955 	return (zpool_standard_error(hdl, errno, errbuf));
3956 }
3957 
3958 /*
3959  * Reopen the pool.
3960  */
3961 int
3962 zpool_reopen_one(zpool_handle_t *zhp, void *data)
3963 {
3964 	libzfs_handle_t *hdl = zpool_get_handle(zhp);
3965 	const char *pool_name = zpool_get_name(zhp);
3966 	boolean_t *scrub_restart = data;
3967 	int error;
3968 
3969 	error = lzc_reopen(pool_name, *scrub_restart);
3970 	if (error) {
3971 		return (zpool_standard_error_fmt(hdl, error,
3972 		    dgettext(TEXT_DOMAIN, "cannot reopen '%s'"), pool_name));
3973 	}
3974 
3975 	return (0);
3976 }
3977 
3978 /* call into libzfs_core to execute the sync IOCTL per pool */
3979 int
3980 zpool_sync_one(zpool_handle_t *zhp, void *data)
3981 {
3982 	int ret;
3983 	libzfs_handle_t *hdl = zpool_get_handle(zhp);
3984 	const char *pool_name = zpool_get_name(zhp);
3985 	boolean_t *force = data;
3986 	nvlist_t *innvl = fnvlist_alloc();
3987 
3988 	fnvlist_add_boolean_value(innvl, "force", *force);
3989 	if ((ret = lzc_sync(pool_name, innvl, NULL)) != 0) {
3990 		nvlist_free(innvl);
3991 		return (zpool_standard_error_fmt(hdl, ret,
3992 		    dgettext(TEXT_DOMAIN, "sync '%s' failed"), pool_name));
3993 	}
3994 	nvlist_free(innvl);
3995 
3996 	return (0);
3997 }
3998 
3999 #define	PATH_BUF_LEN	64
4000 
4001 /*
4002  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
4003  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
4004  * We also check if this is a whole disk, in which case we strip off the
4005  * trailing 's0' slice name.
4006  *
4007  * This routine is also responsible for identifying when disks have been
4008  * reconfigured in a new location.  The kernel will have opened the device by
4009  * devid, but the path will still refer to the old location.  To catch this, we
4010  * first do a path -> devid translation (which is fast for the common case).  If
4011  * the devid matches, we're done.  If not, we do a reverse devid -> path
4012  * translation and issue the appropriate ioctl() to update the path of the vdev.
4013  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
4014  * of these checks.
4015  */
4016 char *
4017 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
4018     int name_flags)
4019 {
4020 	const char *type, *tpath;
4021 	const char *path;
4022 	uint64_t value;
4023 	char buf[PATH_BUF_LEN];
4024 	char tmpbuf[PATH_BUF_LEN * 2];
4025 
4026 	/*
4027 	 * vdev_name will be "root"/"root-0" for the root vdev, but it is the
4028 	 * zpool name that will be displayed to the user.
4029 	 */
4030 	type = fnvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE);
4031 	if (zhp != NULL && strcmp(type, "root") == 0)
4032 		return (zfs_strdup(hdl, zpool_get_name(zhp)));
4033 
4034 	if (libzfs_envvar_is_set("ZPOOL_VDEV_NAME_PATH"))
4035 		name_flags |= VDEV_NAME_PATH;
4036 	if (libzfs_envvar_is_set("ZPOOL_VDEV_NAME_GUID"))
4037 		name_flags |= VDEV_NAME_GUID;
4038 	if (libzfs_envvar_is_set("ZPOOL_VDEV_NAME_FOLLOW_LINKS"))
4039 		name_flags |= VDEV_NAME_FOLLOW_LINKS;
4040 
4041 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
4042 	    name_flags & VDEV_NAME_GUID) {
4043 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value);
4044 		(void) snprintf(buf, sizeof (buf), "%llu", (u_longlong_t)value);
4045 		path = buf;
4046 	} else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &tpath) == 0) {
4047 		path = tpath;
4048 
4049 		if (name_flags & VDEV_NAME_FOLLOW_LINKS) {
4050 			char *rp = realpath(path, NULL);
4051 			if (rp) {
4052 				strlcpy(buf, rp, sizeof (buf));
4053 				path = buf;
4054 				free(rp);
4055 			}
4056 		}
4057 
4058 		/*
4059 		 * For a block device only use the name.
4060 		 */
4061 		if ((strcmp(type, VDEV_TYPE_DISK) == 0) &&
4062 		    !(name_flags & VDEV_NAME_PATH)) {
4063 			path = zfs_strip_path(path);
4064 		}
4065 
4066 		/*
4067 		 * Remove the partition from the path if this is a whole disk.
4068 		 */
4069 		if (strcmp(type, VDEV_TYPE_DRAID_SPARE) != 0 &&
4070 		    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &value)
4071 		    == 0 && value && !(name_flags & VDEV_NAME_PATH)) {
4072 			return (zfs_strip_partition(path));
4073 		}
4074 	} else {
4075 		path = type;
4076 
4077 		/*
4078 		 * If it's a raidz device, we need to stick in the parity level.
4079 		 */
4080 		if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
4081 			value = fnvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY);
4082 			(void) snprintf(buf, sizeof (buf), "%s%llu", path,
4083 			    (u_longlong_t)value);
4084 			path = buf;
4085 		}
4086 
4087 		/*
4088 		 * If it's a dRAID device, we add parity, groups, and spares.
4089 		 */
4090 		if (strcmp(path, VDEV_TYPE_DRAID) == 0) {
4091 			uint64_t ndata, nparity, nspares;
4092 			nvlist_t **child;
4093 			uint_t children;
4094 
4095 			verify(nvlist_lookup_nvlist_array(nv,
4096 			    ZPOOL_CONFIG_CHILDREN, &child, &children) == 0);
4097 			nparity = fnvlist_lookup_uint64(nv,
4098 			    ZPOOL_CONFIG_NPARITY);
4099 			ndata = fnvlist_lookup_uint64(nv,
4100 			    ZPOOL_CONFIG_DRAID_NDATA);
4101 			nspares = fnvlist_lookup_uint64(nv,
4102 			    ZPOOL_CONFIG_DRAID_NSPARES);
4103 
4104 			path = zpool_draid_name(buf, sizeof (buf), ndata,
4105 			    nparity, nspares, children);
4106 		}
4107 
4108 		/*
4109 		 * We identify each top-level vdev by using a <type-id>
4110 		 * naming convention.
4111 		 */
4112 		if (name_flags & VDEV_NAME_TYPE_ID) {
4113 			uint64_t id = fnvlist_lookup_uint64(nv,
4114 			    ZPOOL_CONFIG_ID);
4115 			(void) snprintf(tmpbuf, sizeof (tmpbuf), "%s-%llu",
4116 			    path, (u_longlong_t)id);
4117 			path = tmpbuf;
4118 		}
4119 	}
4120 
4121 	return (zfs_strdup(hdl, path));
4122 }
4123 
4124 static int
4125 zbookmark_mem_compare(const void *a, const void *b)
4126 {
4127 	return (memcmp(a, b, sizeof (zbookmark_phys_t)));
4128 }
4129 
4130 /*
4131  * Retrieve the persistent error log, uniquify the members, and return to the
4132  * caller.
4133  */
4134 int
4135 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
4136 {
4137 	zfs_cmd_t zc = {"\0"};
4138 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4139 	zbookmark_phys_t *buf;
4140 	uint64_t buflen = 10000; /* approx. 1MB of RAM */
4141 
4142 	if (fnvlist_lookup_uint64(zhp->zpool_config,
4143 	    ZPOOL_CONFIG_ERRCOUNT) == 0)
4144 		return (0);
4145 
4146 	/*
4147 	 * Retrieve the raw error list from the kernel.  If it doesn't fit,
4148 	 * allocate a larger buffer and retry.
4149 	 */
4150 	(void) strcpy(zc.zc_name, zhp->zpool_name);
4151 	for (;;) {
4152 		buf = zfs_alloc(zhp->zpool_hdl,
4153 		    buflen * sizeof (zbookmark_phys_t));
4154 		zc.zc_nvlist_dst = (uintptr_t)buf;
4155 		zc.zc_nvlist_dst_size = buflen;
4156 		if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_ERROR_LOG,
4157 		    &zc) != 0) {
4158 			free(buf);
4159 			if (errno == ENOMEM) {
4160 				buflen *= 2;
4161 			} else {
4162 				return (zpool_standard_error_fmt(hdl, errno,
4163 				    dgettext(TEXT_DOMAIN, "errors: List of "
4164 				    "errors unavailable")));
4165 			}
4166 		} else {
4167 			break;
4168 		}
4169 	}
4170 
4171 	/*
4172 	 * Sort the resulting bookmarks.  This is a little confusing due to the
4173 	 * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
4174 	 * to first, and 'zc_nvlist_dst_size' indicates the number of bookmarks
4175 	 * _not_ copied as part of the process.  So we point the start of our
4176 	 * array appropriate and decrement the total number of elements.
4177 	 */
4178 	zbookmark_phys_t *zb = buf + zc.zc_nvlist_dst_size;
4179 	uint64_t zblen = buflen - zc.zc_nvlist_dst_size;
4180 
4181 	qsort(zb, zblen, sizeof (zbookmark_phys_t), zbookmark_mem_compare);
4182 
4183 	verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
4184 
4185 	/*
4186 	 * Fill in the nverrlistp with nvlist's of dataset and object numbers.
4187 	 */
4188 	for (uint64_t i = 0; i < zblen; i++) {
4189 		nvlist_t *nv;
4190 
4191 		/* ignoring zb_blkid and zb_level for now */
4192 		if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
4193 		    zb[i-1].zb_object == zb[i].zb_object)
4194 			continue;
4195 
4196 		if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
4197 			goto nomem;
4198 		if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
4199 		    zb[i].zb_objset) != 0) {
4200 			nvlist_free(nv);
4201 			goto nomem;
4202 		}
4203 		if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
4204 		    zb[i].zb_object) != 0) {
4205 			nvlist_free(nv);
4206 			goto nomem;
4207 		}
4208 		if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
4209 			nvlist_free(nv);
4210 			goto nomem;
4211 		}
4212 		nvlist_free(nv);
4213 	}
4214 
4215 	free(buf);
4216 	return (0);
4217 
4218 nomem:
4219 	free(buf);
4220 	return (no_memory(zhp->zpool_hdl));
4221 }
4222 
4223 /*
4224  * Upgrade a ZFS pool to the latest on-disk version.
4225  */
4226 int
4227 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
4228 {
4229 	zfs_cmd_t zc = {"\0"};
4230 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4231 
4232 	(void) strcpy(zc.zc_name, zhp->zpool_name);
4233 	zc.zc_cookie = new_version;
4234 
4235 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
4236 		return (zpool_standard_error_fmt(hdl, errno,
4237 		    dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
4238 		    zhp->zpool_name));
4239 	return (0);
4240 }
4241 
4242 void
4243 zfs_save_arguments(int argc, char **argv, char *string, int len)
4244 {
4245 	int i;
4246 
4247 	(void) strlcpy(string, zfs_basename(argv[0]), len);
4248 	for (i = 1; i < argc; i++) {
4249 		(void) strlcat(string, " ", len);
4250 		(void) strlcat(string, argv[i], len);
4251 	}
4252 }
4253 
4254 int
4255 zpool_log_history(libzfs_handle_t *hdl, const char *message)
4256 {
4257 	zfs_cmd_t zc = {"\0"};
4258 	nvlist_t *args;
4259 
4260 	args = fnvlist_alloc();
4261 	fnvlist_add_string(args, "message", message);
4262 	zcmd_write_src_nvlist(hdl, &zc, args);
4263 	int err = zfs_ioctl(hdl, ZFS_IOC_LOG_HISTORY, &zc);
4264 	nvlist_free(args);
4265 	zcmd_free_nvlists(&zc);
4266 	return (err);
4267 }
4268 
4269 /*
4270  * Perform ioctl to get some command history of a pool.
4271  *
4272  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
4273  * logical offset of the history buffer to start reading from.
4274  *
4275  * Upon return, 'off' is the next logical offset to read from and
4276  * 'len' is the actual amount of bytes read into 'buf'.
4277  */
4278 static int
4279 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
4280 {
4281 	zfs_cmd_t zc = {"\0"};
4282 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4283 
4284 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4285 
4286 	zc.zc_history = (uint64_t)(uintptr_t)buf;
4287 	zc.zc_history_len = *len;
4288 	zc.zc_history_offset = *off;
4289 
4290 	if (zfs_ioctl(hdl, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
4291 		switch (errno) {
4292 		case EPERM:
4293 			return (zfs_error_fmt(hdl, EZFS_PERM,
4294 			    dgettext(TEXT_DOMAIN,
4295 			    "cannot show history for pool '%s'"),
4296 			    zhp->zpool_name));
4297 		case ENOENT:
4298 			return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
4299 			    dgettext(TEXT_DOMAIN, "cannot get history for pool "
4300 			    "'%s'"), zhp->zpool_name));
4301 		case ENOTSUP:
4302 			return (zfs_error_fmt(hdl, EZFS_BADVERSION,
4303 			    dgettext(TEXT_DOMAIN, "cannot get history for pool "
4304 			    "'%s', pool must be upgraded"), zhp->zpool_name));
4305 		default:
4306 			return (zpool_standard_error_fmt(hdl, errno,
4307 			    dgettext(TEXT_DOMAIN,
4308 			    "cannot get history for '%s'"), zhp->zpool_name));
4309 		}
4310 	}
4311 
4312 	*len = zc.zc_history_len;
4313 	*off = zc.zc_history_offset;
4314 
4315 	return (0);
4316 }
4317 
4318 /*
4319  * Retrieve the command history of a pool.
4320  */
4321 int
4322 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp, uint64_t *off,
4323     boolean_t *eof)
4324 {
4325 	libzfs_handle_t *hdl = zhp->zpool_hdl;
4326 	char *buf;
4327 	int buflen = 128 * 1024;
4328 	nvlist_t **records = NULL;
4329 	uint_t numrecords = 0;
4330 	int err = 0, i;
4331 	uint64_t start = *off;
4332 
4333 	buf = zfs_alloc(hdl, buflen);
4334 
4335 	/* process about 1MiB a time */
4336 	while (*off - start < 1024 * 1024) {
4337 		uint64_t bytes_read = buflen;
4338 		uint64_t leftover;
4339 
4340 		if ((err = get_history(zhp, buf, off, &bytes_read)) != 0)
4341 			break;
4342 
4343 		/* if nothing else was read in, we're at EOF, just return */
4344 		if (!bytes_read) {
4345 			*eof = B_TRUE;
4346 			break;
4347 		}
4348 
4349 		if ((err = zpool_history_unpack(buf, bytes_read,
4350 		    &leftover, &records, &numrecords)) != 0) {
4351 			zpool_standard_error_fmt(hdl, err,
4352 			    dgettext(TEXT_DOMAIN,
4353 			    "cannot get history for '%s'"), zhp->zpool_name);
4354 			break;
4355 		}
4356 		*off -= leftover;
4357 		if (leftover == bytes_read) {
4358 			/*
4359 			 * no progress made, because buffer is not big enough
4360 			 * to hold this record; resize and retry.
4361 			 */
4362 			buflen *= 2;
4363 			free(buf);
4364 			buf = zfs_alloc(hdl, buflen);
4365 		}
4366 	}
4367 
4368 	free(buf);
4369 
4370 	if (!err) {
4371 		*nvhisp = fnvlist_alloc();
4372 		fnvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
4373 		    (const nvlist_t **)records, numrecords);
4374 	}
4375 	for (i = 0; i < numrecords; i++)
4376 		nvlist_free(records[i]);
4377 	free(records);
4378 
4379 	return (err);
4380 }
4381 
4382 /*
4383  * Retrieve the next event given the passed 'zevent_fd' file descriptor.
4384  * If there is a new event available 'nvp' will contain a newly allocated
4385  * nvlist and 'dropped' will be set to the number of missed events since
4386  * the last call to this function.  When 'nvp' is set to NULL it indicates
4387  * no new events are available.  In either case the function returns 0 and
4388  * it is up to the caller to free 'nvp'.  In the case of a fatal error the
4389  * function will return a non-zero value.  When the function is called in
4390  * blocking mode (the default, unless the ZEVENT_NONBLOCK flag is passed),
4391  * it will not return until a new event is available.
4392  */
4393 int
4394 zpool_events_next(libzfs_handle_t *hdl, nvlist_t **nvp,
4395     int *dropped, unsigned flags, int zevent_fd)
4396 {
4397 	zfs_cmd_t zc = {"\0"};
4398 	int error = 0;
4399 
4400 	*nvp = NULL;
4401 	*dropped = 0;
4402 	zc.zc_cleanup_fd = zevent_fd;
4403 
4404 	if (flags & ZEVENT_NONBLOCK)
4405 		zc.zc_guid = ZEVENT_NONBLOCK;
4406 
4407 	zcmd_alloc_dst_nvlist(hdl, &zc, ZEVENT_SIZE);
4408 
4409 retry:
4410 	if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_NEXT, &zc) != 0) {
4411 		switch (errno) {
4412 		case ESHUTDOWN:
4413 			error = zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
4414 			    dgettext(TEXT_DOMAIN, "zfs shutdown"));
4415 			goto out;
4416 		case ENOENT:
4417 			/* Blocking error case should not occur */
4418 			if (!(flags & ZEVENT_NONBLOCK))
4419 				error = zpool_standard_error_fmt(hdl, errno,
4420 				    dgettext(TEXT_DOMAIN, "cannot get event"));
4421 
4422 			goto out;
4423 		case ENOMEM:
4424 			zcmd_expand_dst_nvlist(hdl, &zc);
4425 			goto retry;
4426 		default:
4427 			error = zpool_standard_error_fmt(hdl, errno,
4428 			    dgettext(TEXT_DOMAIN, "cannot get event"));
4429 			goto out;
4430 		}
4431 	}
4432 
4433 	error = zcmd_read_dst_nvlist(hdl, &zc, nvp);
4434 	if (error != 0)
4435 		goto out;
4436 
4437 	*dropped = (int)zc.zc_cookie;
4438 out:
4439 	zcmd_free_nvlists(&zc);
4440 
4441 	return (error);
4442 }
4443 
4444 /*
4445  * Clear all events.
4446  */
4447 int
4448 zpool_events_clear(libzfs_handle_t *hdl, int *count)
4449 {
4450 	zfs_cmd_t zc = {"\0"};
4451 
4452 	if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_CLEAR, &zc) != 0)
4453 		return (zpool_standard_error(hdl, errno,
4454 		    dgettext(TEXT_DOMAIN, "cannot clear events")));
4455 
4456 	if (count != NULL)
4457 		*count = (int)zc.zc_cookie; /* # of events cleared */
4458 
4459 	return (0);
4460 }
4461 
4462 /*
4463  * Seek to a specific EID, ZEVENT_SEEK_START, or ZEVENT_SEEK_END for
4464  * the passed zevent_fd file handle.  On success zero is returned,
4465  * otherwise -1 is returned and hdl->libzfs_error is set to the errno.
4466  */
4467 int
4468 zpool_events_seek(libzfs_handle_t *hdl, uint64_t eid, int zevent_fd)
4469 {
4470 	zfs_cmd_t zc = {"\0"};
4471 	int error = 0;
4472 
4473 	zc.zc_guid = eid;
4474 	zc.zc_cleanup_fd = zevent_fd;
4475 
4476 	if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_SEEK, &zc) != 0) {
4477 		switch (errno) {
4478 		case ENOENT:
4479 			error = zfs_error_fmt(hdl, EZFS_NOENT,
4480 			    dgettext(TEXT_DOMAIN, "cannot get event"));
4481 			break;
4482 
4483 		case ENOMEM:
4484 			error = zfs_error_fmt(hdl, EZFS_NOMEM,
4485 			    dgettext(TEXT_DOMAIN, "cannot get event"));
4486 			break;
4487 
4488 		default:
4489 			error = zpool_standard_error_fmt(hdl, errno,
4490 			    dgettext(TEXT_DOMAIN, "cannot get event"));
4491 			break;
4492 		}
4493 	}
4494 
4495 	return (error);
4496 }
4497 
4498 static void
4499 zpool_obj_to_path_impl(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4500     char *pathname, size_t len, boolean_t always_unmounted)
4501 {
4502 	zfs_cmd_t zc = {"\0"};
4503 	boolean_t mounted = B_FALSE;
4504 	char *mntpnt = NULL;
4505 	char dsname[ZFS_MAX_DATASET_NAME_LEN];
4506 
4507 	if (dsobj == 0) {
4508 		/* special case for the MOS */
4509 		(void) snprintf(pathname, len, "<metadata>:<0x%llx>",
4510 		    (longlong_t)obj);
4511 		return;
4512 	}
4513 
4514 	/* get the dataset's name */
4515 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4516 	zc.zc_obj = dsobj;
4517 	if (zfs_ioctl(zhp->zpool_hdl,
4518 	    ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
4519 		/* just write out a path of two object numbers */
4520 		(void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
4521 		    (longlong_t)dsobj, (longlong_t)obj);
4522 		return;
4523 	}
4524 	(void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
4525 
4526 	/* find out if the dataset is mounted */
4527 	mounted = !always_unmounted && is_mounted(zhp->zpool_hdl, dsname,
4528 	    &mntpnt);
4529 
4530 	/* get the corrupted object's path */
4531 	(void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
4532 	zc.zc_obj = obj;
4533 	if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_OBJ_TO_PATH,
4534 	    &zc) == 0) {
4535 		if (mounted) {
4536 			(void) snprintf(pathname, len, "%s%s", mntpnt,
4537 			    zc.zc_value);
4538 		} else {
4539 			(void) snprintf(pathname, len, "%s:%s",
4540 			    dsname, zc.zc_value);
4541 		}
4542 	} else {
4543 		(void) snprintf(pathname, len, "%s:<0x%llx>", dsname,
4544 		    (longlong_t)obj);
4545 	}
4546 	free(mntpnt);
4547 }
4548 
4549 void
4550 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4551     char *pathname, size_t len)
4552 {
4553 	zpool_obj_to_path_impl(zhp, dsobj, obj, pathname, len, B_FALSE);
4554 }
4555 
4556 void
4557 zpool_obj_to_path_ds(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4558     char *pathname, size_t len)
4559 {
4560 	zpool_obj_to_path_impl(zhp, dsobj, obj, pathname, len, B_TRUE);
4561 }
4562 /*
4563  * Wait while the specified activity is in progress in the pool.
4564  */
4565 int
4566 zpool_wait(zpool_handle_t *zhp, zpool_wait_activity_t activity)
4567 {
4568 	boolean_t missing;
4569 
4570 	int error = zpool_wait_status(zhp, activity, &missing, NULL);
4571 
4572 	if (missing) {
4573 		(void) zpool_standard_error_fmt(zhp->zpool_hdl, ENOENT,
4574 		    dgettext(TEXT_DOMAIN, "error waiting in pool '%s'"),
4575 		    zhp->zpool_name);
4576 		return (ENOENT);
4577 	} else {
4578 		return (error);
4579 	}
4580 }
4581 
4582 /*
4583  * Wait for the given activity and return the status of the wait (whether or not
4584  * any waiting was done) in the 'waited' parameter. Non-existent pools are
4585  * reported via the 'missing' parameter, rather than by printing an error
4586  * message. This is convenient when this function is called in a loop over a
4587  * long period of time (as it is, for example, by zpool's wait cmd). In that
4588  * scenario, a pool being exported or destroyed should be considered a normal
4589  * event, so we don't want to print an error when we find that the pool doesn't
4590  * exist.
4591  */
4592 int
4593 zpool_wait_status(zpool_handle_t *zhp, zpool_wait_activity_t activity,
4594     boolean_t *missing, boolean_t *waited)
4595 {
4596 	int error = lzc_wait(zhp->zpool_name, activity, waited);
4597 	*missing = (error == ENOENT);
4598 	if (*missing)
4599 		return (0);
4600 
4601 	if (error != 0) {
4602 		(void) zpool_standard_error_fmt(zhp->zpool_hdl, error,
4603 		    dgettext(TEXT_DOMAIN, "error waiting in pool '%s'"),
4604 		    zhp->zpool_name);
4605 	}
4606 
4607 	return (error);
4608 }
4609 
4610 int
4611 zpool_set_bootenv(zpool_handle_t *zhp, const nvlist_t *envmap)
4612 {
4613 	int error = lzc_set_bootenv(zhp->zpool_name, envmap);
4614 	if (error != 0) {
4615 		(void) zpool_standard_error_fmt(zhp->zpool_hdl, error,
4616 		    dgettext(TEXT_DOMAIN,
4617 		    "error setting bootenv in pool '%s'"), zhp->zpool_name);
4618 	}
4619 
4620 	return (error);
4621 }
4622 
4623 int
4624 zpool_get_bootenv(zpool_handle_t *zhp, nvlist_t **nvlp)
4625 {
4626 	nvlist_t *nvl;
4627 	int error;
4628 
4629 	nvl = NULL;
4630 	error = lzc_get_bootenv(zhp->zpool_name, &nvl);
4631 	if (error != 0) {
4632 		(void) zpool_standard_error_fmt(zhp->zpool_hdl, error,
4633 		    dgettext(TEXT_DOMAIN,
4634 		    "error getting bootenv in pool '%s'"), zhp->zpool_name);
4635 	} else {
4636 		*nvlp = nvl;
4637 	}
4638 
4639 	return (error);
4640 }
4641 
4642 /*
4643  * Attempt to read and parse feature file(s) (from "compatibility" property).
4644  * Files contain zpool feature names, comma or whitespace-separated.
4645  * Comments (# character to next newline) are discarded.
4646  *
4647  * Arguments:
4648  *  compatibility : string containing feature filenames
4649  *  features : either NULL or pointer to array of boolean
4650  *  report : either NULL or pointer to string buffer
4651  *  rlen : length of "report" buffer
4652  *
4653  * compatibility is NULL (unset), "", "off", "legacy", or list of
4654  * comma-separated filenames. filenames should either be absolute,
4655  * or relative to:
4656  *   1) ZPOOL_SYSCONF_COMPAT_D (eg: /etc/zfs/compatibility.d) or
4657  *   2) ZPOOL_DATA_COMPAT_D (eg: /usr/share/zfs/compatibility.d).
4658  * (Unset), "" or "off" => enable all features
4659  * "legacy" => disable all features
4660  *
4661  * Any feature names read from files which match unames in spa_feature_table
4662  * will have the corresponding boolean set in the features array (if non-NULL).
4663  * If more than one feature set specified, only features present in *all* of
4664  * them will be set.
4665  *
4666  * "report" if not NULL will be populated with a suitable status message.
4667  *
4668  * Return values:
4669  *   ZPOOL_COMPATIBILITY_OK : files read and parsed ok
4670  *   ZPOOL_COMPATIBILITY_BADFILE : file too big or not a text file
4671  *   ZPOOL_COMPATIBILITY_BADTOKEN : SYSCONF file contains invalid feature name
4672  *   ZPOOL_COMPATIBILITY_WARNTOKEN : DATA file contains invalid feature name
4673  *   ZPOOL_COMPATIBILITY_NOFILES : no feature files found
4674  */
4675 zpool_compat_status_t
4676 zpool_load_compat(const char *compat, boolean_t *features, char *report,
4677     size_t rlen)
4678 {
4679 	int sdirfd, ddirfd, featfd;
4680 	struct stat fs;
4681 	char *fc;
4682 	char *ps, *ls, *ws;
4683 	char *file, *line, *word;
4684 
4685 	char l_compat[ZFS_MAXPROPLEN];
4686 
4687 	boolean_t ret_nofiles = B_TRUE;
4688 	boolean_t ret_badfile = B_FALSE;
4689 	boolean_t ret_badtoken = B_FALSE;
4690 	boolean_t ret_warntoken = B_FALSE;
4691 
4692 	/* special cases (unset), "" and "off" => enable all features */
4693 	if (compat == NULL || compat[0] == '\0' ||
4694 	    strcmp(compat, ZPOOL_COMPAT_OFF) == 0) {
4695 		if (features != NULL)
4696 			for (uint_t i = 0; i < SPA_FEATURES; i++)
4697 				features[i] = B_TRUE;
4698 		if (report != NULL)
4699 			strlcpy(report, gettext("all features enabled"), rlen);
4700 		return (ZPOOL_COMPATIBILITY_OK);
4701 	}
4702 
4703 	/* Final special case "legacy" => disable all features */
4704 	if (strcmp(compat, ZPOOL_COMPAT_LEGACY) == 0) {
4705 		if (features != NULL)
4706 			for (uint_t i = 0; i < SPA_FEATURES; i++)
4707 				features[i] = B_FALSE;
4708 		if (report != NULL)
4709 			strlcpy(report, gettext("all features disabled"), rlen);
4710 		return (ZPOOL_COMPATIBILITY_OK);
4711 	}
4712 
4713 	/*
4714 	 * Start with all true; will be ANDed with results from each file
4715 	 */
4716 	if (features != NULL)
4717 		for (uint_t i = 0; i < SPA_FEATURES; i++)
4718 			features[i] = B_TRUE;
4719 
4720 	char err_badfile[ZFS_MAXPROPLEN] = "";
4721 	char err_badtoken[ZFS_MAXPROPLEN] = "";
4722 
4723 	/*
4724 	 * We ignore errors from the directory open()
4725 	 * as they're only needed if the filename is relative
4726 	 * which will be checked during the openat().
4727 	 */
4728 
4729 /* O_PATH safer than O_RDONLY if system allows it */
4730 #if defined(O_PATH)
4731 #define	ZC_DIR_FLAGS (O_DIRECTORY | O_CLOEXEC | O_PATH)
4732 #else
4733 #define	ZC_DIR_FLAGS (O_DIRECTORY | O_CLOEXEC | O_RDONLY)
4734 #endif
4735 
4736 	sdirfd = open(ZPOOL_SYSCONF_COMPAT_D, ZC_DIR_FLAGS);
4737 	ddirfd = open(ZPOOL_DATA_COMPAT_D, ZC_DIR_FLAGS);
4738 
4739 	(void) strlcpy(l_compat, compat, ZFS_MAXPROPLEN);
4740 
4741 	for (file = strtok_r(l_compat, ",", &ps);
4742 	    file != NULL;
4743 	    file = strtok_r(NULL, ",", &ps)) {
4744 
4745 		boolean_t l_features[SPA_FEATURES];
4746 
4747 		enum { Z_SYSCONF, Z_DATA } source;
4748 
4749 		/* try sysconfdir first, then datadir */
4750 		source = Z_SYSCONF;
4751 		if ((featfd = openat(sdirfd, file, O_RDONLY | O_CLOEXEC)) < 0) {
4752 			featfd = openat(ddirfd, file, O_RDONLY | O_CLOEXEC);
4753 			source = Z_DATA;
4754 		}
4755 
4756 		/* File readable and correct size? */
4757 		if (featfd < 0 ||
4758 		    fstat(featfd, &fs) < 0 ||
4759 		    fs.st_size < 1 ||
4760 		    fs.st_size > ZPOOL_COMPAT_MAXSIZE) {
4761 			(void) close(featfd);
4762 			strlcat(err_badfile, file, ZFS_MAXPROPLEN);
4763 			strlcat(err_badfile, " ", ZFS_MAXPROPLEN);
4764 			ret_badfile = B_TRUE;
4765 			continue;
4766 		}
4767 
4768 /* Prefault the file if system allows */
4769 #if defined(MAP_POPULATE)
4770 #define	ZC_MMAP_FLAGS (MAP_PRIVATE | MAP_POPULATE)
4771 #elif defined(MAP_PREFAULT_READ)
4772 #define	ZC_MMAP_FLAGS (MAP_PRIVATE | MAP_PREFAULT_READ)
4773 #else
4774 #define	ZC_MMAP_FLAGS (MAP_PRIVATE)
4775 #endif
4776 
4777 		/* private mmap() so we can strtok safely */
4778 		fc = (char *)mmap(NULL, fs.st_size, PROT_READ | PROT_WRITE,
4779 		    ZC_MMAP_FLAGS, featfd, 0);
4780 		(void) close(featfd);
4781 
4782 		/* map ok, and last character == newline? */
4783 		if (fc == MAP_FAILED || fc[fs.st_size - 1] != '\n') {
4784 			(void) munmap((void *) fc, fs.st_size);
4785 			strlcat(err_badfile, file, ZFS_MAXPROPLEN);
4786 			strlcat(err_badfile, " ", ZFS_MAXPROPLEN);
4787 			ret_badfile = B_TRUE;
4788 			continue;
4789 		}
4790 
4791 		ret_nofiles = B_FALSE;
4792 
4793 		for (uint_t i = 0; i < SPA_FEATURES; i++)
4794 			l_features[i] = B_FALSE;
4795 
4796 		/* replace final newline with NULL to ensure string ends */
4797 		fc[fs.st_size - 1] = '\0';
4798 
4799 		for (line = strtok_r(fc, "\n", &ls);
4800 		    line != NULL;
4801 		    line = strtok_r(NULL, "\n", &ls)) {
4802 			/* discard comments */
4803 			char *r = strchr(line, '#');
4804 			if (r != NULL)
4805 				*r = '\0';
4806 
4807 			for (word = strtok_r(line, ", \t", &ws);
4808 			    word != NULL;
4809 			    word = strtok_r(NULL, ", \t", &ws)) {
4810 				/* Find matching feature name */
4811 				uint_t f;
4812 				for (f = 0; f < SPA_FEATURES; f++) {
4813 					zfeature_info_t *fi =
4814 					    &spa_feature_table[f];
4815 					if (strcmp(word, fi->fi_uname) == 0) {
4816 						l_features[f] = B_TRUE;
4817 						break;
4818 					}
4819 				}
4820 				if (f < SPA_FEATURES)
4821 					continue;
4822 
4823 				/* found an unrecognized word */
4824 				/* lightly sanitize it */
4825 				if (strlen(word) > 32)
4826 					word[32] = '\0';
4827 				for (char *c = word; *c != '\0'; c++)
4828 					if (!isprint(*c))
4829 						*c = '?';
4830 
4831 				strlcat(err_badtoken, word, ZFS_MAXPROPLEN);
4832 				strlcat(err_badtoken, " ", ZFS_MAXPROPLEN);
4833 				if (source == Z_SYSCONF)
4834 					ret_badtoken = B_TRUE;
4835 				else
4836 					ret_warntoken = B_TRUE;
4837 			}
4838 		}
4839 		(void) munmap((void *) fc, fs.st_size);
4840 
4841 		if (features != NULL)
4842 			for (uint_t i = 0; i < SPA_FEATURES; i++)
4843 				features[i] &= l_features[i];
4844 	}
4845 	(void) close(sdirfd);
4846 	(void) close(ddirfd);
4847 
4848 	/* Return the most serious error */
4849 	if (ret_badfile) {
4850 		if (report != NULL)
4851 			snprintf(report, rlen, gettext("could not read/"
4852 			    "parse feature file(s): %s"), err_badfile);
4853 		return (ZPOOL_COMPATIBILITY_BADFILE);
4854 	}
4855 	if (ret_nofiles) {
4856 		if (report != NULL)
4857 			strlcpy(report,
4858 			    gettext("no valid compatibility files specified"),
4859 			    rlen);
4860 		return (ZPOOL_COMPATIBILITY_NOFILES);
4861 	}
4862 	if (ret_badtoken) {
4863 		if (report != NULL)
4864 			snprintf(report, rlen, gettext("invalid feature "
4865 			    "name(s) in local compatibility files: %s"),
4866 			    err_badtoken);
4867 		return (ZPOOL_COMPATIBILITY_BADTOKEN);
4868 	}
4869 	if (ret_warntoken) {
4870 		if (report != NULL)
4871 			snprintf(report, rlen, gettext("unrecognized feature "
4872 			    "name(s) in distribution compatibility files: %s"),
4873 			    err_badtoken);
4874 		return (ZPOOL_COMPATIBILITY_WARNTOKEN);
4875 	}
4876 	if (report != NULL)
4877 		strlcpy(report, gettext("compatibility set ok"), rlen);
4878 	return (ZPOOL_COMPATIBILITY_OK);
4879 }
4880 
4881 static int
4882 zpool_vdev_guid(zpool_handle_t *zhp, const char *vdevname, uint64_t *vdev_guid)
4883 {
4884 	nvlist_t *tgt;
4885 	boolean_t avail_spare, l2cache;
4886 
4887 	verify(zhp != NULL);
4888 	if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
4889 		char errbuf[ERRBUFLEN];
4890 		(void) snprintf(errbuf, sizeof (errbuf),
4891 		    dgettext(TEXT_DOMAIN, "pool is in an unavailable state"));
4892 		return (zfs_error(zhp->zpool_hdl, EZFS_POOLUNAVAIL, errbuf));
4893 	}
4894 
4895 	if ((tgt = zpool_find_vdev(zhp, vdevname, &avail_spare, &l2cache,
4896 	    NULL)) == NULL) {
4897 		char errbuf[ERRBUFLEN];
4898 		(void) snprintf(errbuf, sizeof (errbuf),
4899 		    dgettext(TEXT_DOMAIN, "can not find %s in %s"),
4900 		    vdevname, zhp->zpool_name);
4901 		return (zfs_error(zhp->zpool_hdl, EZFS_NODEVICE, errbuf));
4902 	}
4903 
4904 	*vdev_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
4905 	return (0);
4906 }
4907 
4908 /*
4909  * Get a vdev property value for 'prop' and return the value in
4910  * a pre-allocated buffer.
4911  */
4912 int
4913 zpool_get_vdev_prop_value(nvlist_t *nvprop, vdev_prop_t prop, char *prop_name,
4914     char *buf, size_t len, zprop_source_t *srctype, boolean_t literal)
4915 {
4916 	nvlist_t *nv;
4917 	const char *strval;
4918 	uint64_t intval;
4919 	zprop_source_t src = ZPROP_SRC_NONE;
4920 
4921 	if (prop == VDEV_PROP_USERPROP) {
4922 		/* user property, prop_name must contain the property name */
4923 		assert(prop_name != NULL);
4924 		if (nvlist_lookup_nvlist(nvprop, prop_name, &nv) == 0) {
4925 			src = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
4926 			strval = fnvlist_lookup_string(nv, ZPROP_VALUE);
4927 		} else {
4928 			/* user prop not found */
4929 			return (-1);
4930 		}
4931 		(void) strlcpy(buf, strval, len);
4932 		if (srctype)
4933 			*srctype = src;
4934 		return (0);
4935 	}
4936 
4937 	if (prop_name == NULL)
4938 		prop_name = (char *)vdev_prop_to_name(prop);
4939 
4940 	switch (vdev_prop_get_type(prop)) {
4941 	case PROP_TYPE_STRING:
4942 		if (nvlist_lookup_nvlist(nvprop, prop_name, &nv) == 0) {
4943 			src = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
4944 			strval = fnvlist_lookup_string(nv, ZPROP_VALUE);
4945 		} else {
4946 			src = ZPROP_SRC_DEFAULT;
4947 			if ((strval = vdev_prop_default_string(prop)) == NULL)
4948 				strval = "-";
4949 		}
4950 		(void) strlcpy(buf, strval, len);
4951 		break;
4952 
4953 	case PROP_TYPE_NUMBER:
4954 		if (nvlist_lookup_nvlist(nvprop, prop_name, &nv) == 0) {
4955 			src = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
4956 			intval = fnvlist_lookup_uint64(nv, ZPROP_VALUE);
4957 		} else {
4958 			src = ZPROP_SRC_DEFAULT;
4959 			intval = vdev_prop_default_numeric(prop);
4960 		}
4961 
4962 		switch (prop) {
4963 		case VDEV_PROP_ASIZE:
4964 		case VDEV_PROP_PSIZE:
4965 		case VDEV_PROP_SIZE:
4966 		case VDEV_PROP_BOOTSIZE:
4967 		case VDEV_PROP_ALLOCATED:
4968 		case VDEV_PROP_FREE:
4969 		case VDEV_PROP_READ_ERRORS:
4970 		case VDEV_PROP_WRITE_ERRORS:
4971 		case VDEV_PROP_CHECKSUM_ERRORS:
4972 		case VDEV_PROP_INITIALIZE_ERRORS:
4973 		case VDEV_PROP_OPS_NULL:
4974 		case VDEV_PROP_OPS_READ:
4975 		case VDEV_PROP_OPS_WRITE:
4976 		case VDEV_PROP_OPS_FREE:
4977 		case VDEV_PROP_OPS_CLAIM:
4978 		case VDEV_PROP_OPS_TRIM:
4979 		case VDEV_PROP_BYTES_NULL:
4980 		case VDEV_PROP_BYTES_READ:
4981 		case VDEV_PROP_BYTES_WRITE:
4982 		case VDEV_PROP_BYTES_FREE:
4983 		case VDEV_PROP_BYTES_CLAIM:
4984 		case VDEV_PROP_BYTES_TRIM:
4985 			if (literal) {
4986 				(void) snprintf(buf, len, "%llu",
4987 				    (u_longlong_t)intval);
4988 			} else {
4989 				(void) zfs_nicenum(intval, buf, len);
4990 			}
4991 			break;
4992 		case VDEV_PROP_EXPANDSZ:
4993 			if (intval == 0) {
4994 				(void) strlcpy(buf, "-", len);
4995 			} else if (literal) {
4996 				(void) snprintf(buf, len, "%llu",
4997 				    (u_longlong_t)intval);
4998 			} else {
4999 				(void) zfs_nicenum(intval, buf, len);
5000 			}
5001 			break;
5002 		case VDEV_PROP_CAPACITY:
5003 			if (literal) {
5004 				(void) snprintf(buf, len, "%llu",
5005 				    (u_longlong_t)intval);
5006 			} else {
5007 				(void) snprintf(buf, len, "%llu%%",
5008 				    (u_longlong_t)intval);
5009 			}
5010 			break;
5011 		case VDEV_PROP_CHECKSUM_N:
5012 		case VDEV_PROP_CHECKSUM_T:
5013 		case VDEV_PROP_IO_N:
5014 		case VDEV_PROP_IO_T:
5015 			if (intval == UINT64_MAX) {
5016 				(void) strlcpy(buf, "-", len);
5017 			} else {
5018 				(void) snprintf(buf, len, "%llu",
5019 				    (u_longlong_t)intval);
5020 			}
5021 			break;
5022 		case VDEV_PROP_FRAGMENTATION:
5023 			if (intval == UINT64_MAX) {
5024 				(void) strlcpy(buf, "-", len);
5025 			} else {
5026 				(void) snprintf(buf, len, "%llu%%",
5027 				    (u_longlong_t)intval);
5028 			}
5029 			break;
5030 		case VDEV_PROP_STATE:
5031 			if (literal) {
5032 				(void) snprintf(buf, len, "%llu",
5033 				    (u_longlong_t)intval);
5034 			} else {
5035 				(void) strlcpy(buf, zpool_state_to_name(intval,
5036 				    VDEV_AUX_NONE), len);
5037 			}
5038 			break;
5039 		default:
5040 			(void) snprintf(buf, len, "%llu",
5041 			    (u_longlong_t)intval);
5042 		}
5043 		break;
5044 
5045 	case PROP_TYPE_INDEX:
5046 		if (nvlist_lookup_nvlist(nvprop, prop_name, &nv) == 0) {
5047 			src = fnvlist_lookup_uint64(nv, ZPROP_SOURCE);
5048 			intval = fnvlist_lookup_uint64(nv, ZPROP_VALUE);
5049 		} else {
5050 			src = ZPROP_SRC_DEFAULT;
5051 			intval = vdev_prop_default_numeric(prop);
5052 		}
5053 		if (vdev_prop_index_to_string(prop, intval,
5054 		    (const char **)&strval) != 0)
5055 			return (-1);
5056 		(void) strlcpy(buf, strval, len);
5057 		break;
5058 
5059 	default:
5060 		abort();
5061 	}
5062 
5063 	if (srctype)
5064 		*srctype = src;
5065 
5066 	return (0);
5067 }
5068 
5069 /*
5070  * Get a vdev property value for 'prop_name' and return the value in
5071  * a pre-allocated buffer.
5072  */
5073 int
5074 zpool_get_vdev_prop(zpool_handle_t *zhp, const char *vdevname, vdev_prop_t prop,
5075     char *prop_name, char *buf, size_t len, zprop_source_t *srctype,
5076     boolean_t literal)
5077 {
5078 	nvlist_t *reqnvl, *reqprops;
5079 	nvlist_t *retprops = NULL;
5080 	uint64_t vdev_guid = 0;
5081 	int ret;
5082 
5083 	if ((ret = zpool_vdev_guid(zhp, vdevname, &vdev_guid)) != 0)
5084 		return (ret);
5085 
5086 	if (nvlist_alloc(&reqnvl, NV_UNIQUE_NAME, 0) != 0)
5087 		return (no_memory(zhp->zpool_hdl));
5088 	if (nvlist_alloc(&reqprops, NV_UNIQUE_NAME, 0) != 0)
5089 		return (no_memory(zhp->zpool_hdl));
5090 
5091 	fnvlist_add_uint64(reqnvl, ZPOOL_VDEV_PROPS_GET_VDEV, vdev_guid);
5092 
5093 	if (prop != VDEV_PROP_USERPROP) {
5094 		/* prop_name overrides prop value */
5095 		if (prop_name != NULL)
5096 			prop = vdev_name_to_prop(prop_name);
5097 		else
5098 			prop_name = (char *)vdev_prop_to_name(prop);
5099 		assert(prop < VDEV_NUM_PROPS);
5100 	}
5101 
5102 	assert(prop_name != NULL);
5103 	if (nvlist_add_uint64(reqprops, prop_name, prop) != 0) {
5104 		nvlist_free(reqnvl);
5105 		nvlist_free(reqprops);
5106 		return (no_memory(zhp->zpool_hdl));
5107 	}
5108 
5109 	fnvlist_add_nvlist(reqnvl, ZPOOL_VDEV_PROPS_GET_PROPS, reqprops);
5110 
5111 	ret = lzc_get_vdev_prop(zhp->zpool_name, reqnvl, &retprops);
5112 
5113 	if (ret == 0) {
5114 		ret = zpool_get_vdev_prop_value(retprops, prop, prop_name, buf,
5115 		    len, srctype, literal);
5116 	} else {
5117 		char errbuf[ERRBUFLEN];
5118 		(void) snprintf(errbuf, sizeof (errbuf),
5119 		    dgettext(TEXT_DOMAIN, "cannot get vdev property %s from"
5120 		    " %s in %s"), prop_name, vdevname, zhp->zpool_name);
5121 		(void) zpool_standard_error(zhp->zpool_hdl, ret, errbuf);
5122 	}
5123 
5124 	nvlist_free(reqnvl);
5125 	nvlist_free(reqprops);
5126 	nvlist_free(retprops);
5127 
5128 	return (ret);
5129 }
5130 
5131 /*
5132  * Get all vdev properties
5133  */
5134 int
5135 zpool_get_all_vdev_props(zpool_handle_t *zhp, const char *vdevname,
5136     nvlist_t **outnvl)
5137 {
5138 	nvlist_t *nvl = NULL;
5139 	uint64_t vdev_guid = 0;
5140 	int ret;
5141 
5142 	if ((ret = zpool_vdev_guid(zhp, vdevname, &vdev_guid)) != 0)
5143 		return (ret);
5144 
5145 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
5146 		return (no_memory(zhp->zpool_hdl));
5147 
5148 	fnvlist_add_uint64(nvl, ZPOOL_VDEV_PROPS_GET_VDEV, vdev_guid);
5149 
5150 	ret = lzc_get_vdev_prop(zhp->zpool_name, nvl, outnvl);
5151 
5152 	nvlist_free(nvl);
5153 
5154 	if (ret) {
5155 		char errbuf[ERRBUFLEN];
5156 		(void) snprintf(errbuf, sizeof (errbuf),
5157 		    dgettext(TEXT_DOMAIN, "cannot get vdev properties for"
5158 		    " %s in %s"), vdevname, zhp->zpool_name);
5159 		(void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
5160 	}
5161 
5162 	return (ret);
5163 }
5164 
5165 /*
5166  * Set vdev property
5167  */
5168 int
5169 zpool_set_vdev_prop(zpool_handle_t *zhp, const char *vdevname,
5170     const char *propname, const char *propval)
5171 {
5172 	int ret;
5173 	nvlist_t *nvl = NULL;
5174 	nvlist_t *outnvl = NULL;
5175 	nvlist_t *props;
5176 	nvlist_t *realprops;
5177 	prop_flags_t flags = { 0 };
5178 	uint64_t version;
5179 	uint64_t vdev_guid;
5180 
5181 	if ((ret = zpool_vdev_guid(zhp, vdevname, &vdev_guid)) != 0)
5182 		return (ret);
5183 
5184 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
5185 		return (no_memory(zhp->zpool_hdl));
5186 	if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0)
5187 		return (no_memory(zhp->zpool_hdl));
5188 
5189 	fnvlist_add_uint64(nvl, ZPOOL_VDEV_PROPS_SET_VDEV, vdev_guid);
5190 
5191 	if (nvlist_add_string(props, propname, propval) != 0) {
5192 		nvlist_free(props);
5193 		return (no_memory(zhp->zpool_hdl));
5194 	}
5195 
5196 	char errbuf[ERRBUFLEN];
5197 	(void) snprintf(errbuf, sizeof (errbuf),
5198 	    dgettext(TEXT_DOMAIN, "cannot set property %s for %s on %s"),
5199 	    propname, vdevname, zhp->zpool_name);
5200 
5201 	flags.vdevprop = 1;
5202 	version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
5203 	if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
5204 	    zhp->zpool_name, props, version, flags, errbuf)) == NULL) {
5205 		nvlist_free(props);
5206 		nvlist_free(nvl);
5207 		return (-1);
5208 	}
5209 
5210 	nvlist_free(props);
5211 	props = realprops;
5212 
5213 	fnvlist_add_nvlist(nvl, ZPOOL_VDEV_PROPS_SET_PROPS, props);
5214 
5215 	ret = lzc_set_vdev_prop(zhp->zpool_name, nvl, &outnvl);
5216 
5217 	nvlist_free(props);
5218 	nvlist_free(nvl);
5219 	nvlist_free(outnvl);
5220 
5221 	if (ret)
5222 		(void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
5223 
5224 	return (ret);
5225 }
5226