e2a67c6a7b06306199a78f52ff3e7b4217792d65
[zfs.git] / lib / libzfs / libzfs_pool.c
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25  * Copyright (c) 2012 by Delphix. All rights reserved.
26  */
27
28 #include <ctype.h>
29 #include <errno.h>
30 #include <devid.h>
31 #include <fcntl.h>
32 #include <libintl.h>
33 #include <stdio.h>
34 #include <stdlib.h>
35 #include <strings.h>
36 #include <unistd.h>
37 #include <zone.h>
38 #include <sys/stat.h>
39 #include <sys/efi_partition.h>
40 #include <sys/vtoc.h>
41 #include <sys/zfs_ioctl.h>
42 #include <dlfcn.h>
43
44 #include "zfs_namecheck.h"
45 #include "zfs_prop.h"
46 #include "libzfs_impl.h"
47 #include "zfs_comutil.h"
48
49 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
50
51 typedef struct prop_flags {
52         int create:1;   /* Validate property on creation */
53         int import:1;   /* Validate property on import */
54 } prop_flags_t;
55
56 /*
57  * ====================================================================
58  *   zpool property functions
59  * ====================================================================
60  */
61
62 static int
63 zpool_get_all_props(zpool_handle_t *zhp)
64 {
65         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
66         libzfs_handle_t *hdl = zhp->zpool_hdl;
67
68         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
69
70         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
71                 return (-1);
72
73         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
74                 if (errno == ENOMEM) {
75                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
76                                 zcmd_free_nvlists(&zc);
77                                 return (-1);
78                         }
79                 } else {
80                         zcmd_free_nvlists(&zc);
81                         return (-1);
82                 }
83         }
84
85         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
86                 zcmd_free_nvlists(&zc);
87                 return (-1);
88         }
89
90         zcmd_free_nvlists(&zc);
91
92         return (0);
93 }
94
95 static int
96 zpool_props_refresh(zpool_handle_t *zhp)
97 {
98         nvlist_t *old_props;
99
100         old_props = zhp->zpool_props;
101
102         if (zpool_get_all_props(zhp) != 0)
103                 return (-1);
104
105         nvlist_free(old_props);
106         return (0);
107 }
108
109 static char *
110 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
111     zprop_source_t *src)
112 {
113         nvlist_t *nv, *nvl;
114         uint64_t ival;
115         char *value;
116         zprop_source_t source;
117
118         nvl = zhp->zpool_props;
119         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
120                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
121                 source = ival;
122                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
123         } else {
124                 source = ZPROP_SRC_DEFAULT;
125                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
126                         value = "-";
127         }
128
129         if (src)
130                 *src = source;
131
132         return (value);
133 }
134
135 uint64_t
136 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
137 {
138         nvlist_t *nv, *nvl;
139         uint64_t value;
140         zprop_source_t source;
141
142         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
143                 /*
144                  * zpool_get_all_props() has most likely failed because
145                  * the pool is faulted, but if all we need is the top level
146                  * vdev's guid then get it from the zhp config nvlist.
147                  */
148                 if ((prop == ZPOOL_PROP_GUID) &&
149                     (nvlist_lookup_nvlist(zhp->zpool_config,
150                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
151                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
152                     == 0)) {
153                         return (value);
154                 }
155                 return (zpool_prop_default_numeric(prop));
156         }
157
158         nvl = zhp->zpool_props;
159         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
160                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
161                 source = value;
162                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
163         } else {
164                 source = ZPROP_SRC_DEFAULT;
165                 value = zpool_prop_default_numeric(prop);
166         }
167
168         if (src)
169                 *src = source;
170
171         return (value);
172 }
173
174 /*
175  * Map VDEV STATE to printed strings.
176  */
177 char *
178 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
179 {
180         switch (state) {
181         default:
182                 break;
183         case VDEV_STATE_CLOSED:
184         case VDEV_STATE_OFFLINE:
185                 return (gettext("OFFLINE"));
186         case VDEV_STATE_REMOVED:
187                 return (gettext("REMOVED"));
188         case VDEV_STATE_CANT_OPEN:
189                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
190                         return (gettext("FAULTED"));
191                 else if (aux == VDEV_AUX_SPLIT_POOL)
192                         return (gettext("SPLIT"));
193                 else
194                         return (gettext("UNAVAIL"));
195         case VDEV_STATE_FAULTED:
196                 return (gettext("FAULTED"));
197         case VDEV_STATE_DEGRADED:
198                 return (gettext("DEGRADED"));
199         case VDEV_STATE_HEALTHY:
200                 return (gettext("ONLINE"));
201         }
202
203         return (gettext("UNKNOWN"));
204 }
205
206 /*
207  * Get a zpool property value for 'prop' and return the value in
208  * a pre-allocated buffer.
209  */
210 int
211 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
212     zprop_source_t *srctype)
213 {
214         uint64_t intval;
215         const char *strval;
216         zprop_source_t src = ZPROP_SRC_NONE;
217         nvlist_t *nvroot;
218         vdev_stat_t *vs;
219         uint_t vsc;
220
221         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
222                 switch (prop) {
223                 case ZPOOL_PROP_NAME:
224                         (void) strlcpy(buf, zpool_get_name(zhp), len);
225                         break;
226
227                 case ZPOOL_PROP_HEALTH:
228                         (void) strlcpy(buf, "FAULTED", len);
229                         break;
230
231                 case ZPOOL_PROP_GUID:
232                         intval = zpool_get_prop_int(zhp, prop, &src);
233                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
234                         break;
235
236                 case ZPOOL_PROP_ALTROOT:
237                 case ZPOOL_PROP_CACHEFILE:
238                 case ZPOOL_PROP_COMMENT:
239                         if (zhp->zpool_props != NULL ||
240                             zpool_get_all_props(zhp) == 0) {
241                                 (void) strlcpy(buf,
242                                     zpool_get_prop_string(zhp, prop, &src),
243                                     len);
244                                 if (srctype != NULL)
245                                         *srctype = src;
246                                 return (0);
247                         }
248                         /* FALLTHROUGH */
249                 default:
250                         (void) strlcpy(buf, "-", len);
251                         break;
252                 }
253
254                 if (srctype != NULL)
255                         *srctype = src;
256                 return (0);
257         }
258
259         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
260             prop != ZPOOL_PROP_NAME)
261                 return (-1);
262
263         switch (zpool_prop_get_type(prop)) {
264         case PROP_TYPE_STRING:
265                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
266                     len);
267                 break;
268
269         case PROP_TYPE_NUMBER:
270                 intval = zpool_get_prop_int(zhp, prop, &src);
271
272                 switch (prop) {
273                 case ZPOOL_PROP_SIZE:
274                 case ZPOOL_PROP_ALLOCATED:
275                 case ZPOOL_PROP_FREE:
276                 case ZPOOL_PROP_EXPANDSZ:
277                 case ZPOOL_PROP_ASHIFT:
278                         (void) zfs_nicenum(intval, buf, len);
279                         break;
280
281                 case ZPOOL_PROP_CAPACITY:
282                         (void) snprintf(buf, len, "%llu%%",
283                             (u_longlong_t)intval);
284                         break;
285
286                 case ZPOOL_PROP_DEDUPRATIO:
287                         (void) snprintf(buf, len, "%llu.%02llux",
288                             (u_longlong_t)(intval / 100),
289                             (u_longlong_t)(intval % 100));
290                         break;
291
292                 case ZPOOL_PROP_HEALTH:
293                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
294                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
295                         verify(nvlist_lookup_uint64_array(nvroot,
296                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
297                             == 0);
298
299                         (void) strlcpy(buf, zpool_state_to_name(intval,
300                             vs->vs_aux), len);
301                         break;
302                 default:
303                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
304                 }
305                 break;
306
307         case PROP_TYPE_INDEX:
308                 intval = zpool_get_prop_int(zhp, prop, &src);
309                 if (zpool_prop_index_to_string(prop, intval, &strval)
310                     != 0)
311                         return (-1);
312                 (void) strlcpy(buf, strval, len);
313                 break;
314
315         default:
316                 abort();
317         }
318
319         if (srctype)
320                 *srctype = src;
321
322         return (0);
323 }
324
325 /*
326  * Check if the bootfs name has the same pool name as it is set to.
327  * Assuming bootfs is a valid dataset name.
328  */
329 static boolean_t
330 bootfs_name_valid(const char *pool, char *bootfs)
331 {
332         int len = strlen(pool);
333
334         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
335                 return (B_FALSE);
336
337         if (strncmp(pool, bootfs, len) == 0 &&
338             (bootfs[len] == '/' || bootfs[len] == '\0'))
339                 return (B_TRUE);
340
341         return (B_FALSE);
342 }
343
344 #if defined(__sun__) || defined(__sun)
345 /*
346  * Inspect the configuration to determine if any of the devices contain
347  * an EFI label.
348  */
349 static boolean_t
350 pool_uses_efi(nvlist_t *config)
351 {
352         nvlist_t **child;
353         uint_t c, children;
354
355         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
356             &child, &children) != 0)
357                 return (read_efi_label(config, NULL) >= 0);
358
359         for (c = 0; c < children; c++) {
360                 if (pool_uses_efi(child[c]))
361                         return (B_TRUE);
362         }
363         return (B_FALSE);
364 }
365 #endif
366
367 boolean_t
368 zpool_is_bootable(zpool_handle_t *zhp)
369 {
370         char bootfs[ZPOOL_MAXNAMELEN];
371
372         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
373             sizeof (bootfs), NULL) == 0 && strncmp(bootfs, "-",
374             sizeof (bootfs)) != 0);
375 }
376
377
378 /*
379  * Given an nvlist of zpool properties to be set, validate that they are
380  * correct, and parse any numeric properties (index, boolean, etc) if they are
381  * specified as strings.
382  */
383 static nvlist_t *
384 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
385     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
386 {
387         nvpair_t *elem;
388         nvlist_t *retprops;
389         zpool_prop_t prop;
390         char *strval;
391         uint64_t intval;
392         char *slash, *check;
393         struct stat64 statbuf;
394         zpool_handle_t *zhp;
395         nvlist_t *nvroot;
396
397         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
398                 (void) no_memory(hdl);
399                 return (NULL);
400         }
401
402         elem = NULL;
403         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
404                 const char *propname = nvpair_name(elem);
405
406                 /*
407                  * Make sure this property is valid and applies to this type.
408                  */
409                 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) {
410                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
411                             "invalid property '%s'"), propname);
412                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
413                         goto error;
414                 }
415
416                 if (zpool_prop_readonly(prop)) {
417                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
418                             "is readonly"), propname);
419                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
420                         goto error;
421                 }
422
423                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
424                     &strval, &intval, errbuf) != 0)
425                         goto error;
426
427                 /*
428                  * Perform additional checking for specific properties.
429                  */
430                 switch (prop) {
431                 default:
432                         break;
433                 case ZPOOL_PROP_VERSION:
434                         if (intval < version || intval > SPA_VERSION) {
435                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
436                                     "property '%s' number %d is invalid."),
437                                     propname, intval);
438                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
439                                 goto error;
440                         }
441                         break;
442
443                 case ZPOOL_PROP_ASHIFT:
444                         if (!flags.create) {
445                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
446                                     "property '%s' can only be set at "
447                                     "creation time"), propname);
448                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
449                                 goto error;
450                         }
451
452                         if (intval != 0 && (intval < 9 || intval > 13)) {
453                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
454                                     "property '%s' number %d is invalid."),
455                                     propname, intval);
456                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
457                                 goto error;
458                         }
459                         break;
460
461                 case ZPOOL_PROP_BOOTFS:
462                         if (flags.create || flags.import) {
463                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
464                                     "property '%s' cannot be set at creation "
465                                     "or import time"), propname);
466                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
467                                 goto error;
468                         }
469
470                         if (version < SPA_VERSION_BOOTFS) {
471                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
472                                     "pool must be upgraded to support "
473                                     "'%s' property"), propname);
474                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
475                                 goto error;
476                         }
477
478                         /*
479                          * bootfs property value has to be a dataset name and
480                          * the dataset has to be in the same pool as it sets to.
481                          */
482                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
483                             strval)) {
484                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
485                                     "is an invalid name"), strval);
486                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
487                                 goto error;
488                         }
489
490                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
491                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
492                                     "could not open pool '%s'"), poolname);
493                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
494                                 goto error;
495                         }
496                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
497                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
498
499 #if defined(__sun__) || defined(__sun)
500                         /*
501                          * bootfs property cannot be set on a disk which has
502                          * been EFI labeled.
503                          */
504                         if (pool_uses_efi(nvroot)) {
505                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
506                                     "property '%s' not supported on "
507                                     "EFI labeled devices"), propname);
508                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
509                                 zpool_close(zhp);
510                                 goto error;
511                         }
512 #endif
513                         zpool_close(zhp);
514                         break;
515
516                 case ZPOOL_PROP_ALTROOT:
517                         if (!flags.create && !flags.import) {
518                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
519                                     "property '%s' can only be set during pool "
520                                     "creation or import"), propname);
521                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
522                                 goto error;
523                         }
524
525                         if (strval[0] != '/') {
526                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
527                                     "bad alternate root '%s'"), strval);
528                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
529                                 goto error;
530                         }
531                         break;
532
533                 case ZPOOL_PROP_CACHEFILE:
534                         if (strval[0] == '\0')
535                                 break;
536
537                         if (strcmp(strval, "none") == 0)
538                                 break;
539
540                         if (strval[0] != '/') {
541                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
542                                     "property '%s' must be empty, an "
543                                     "absolute path, or 'none'"), propname);
544                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
545                                 goto error;
546                         }
547
548                         slash = strrchr(strval, '/');
549
550                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
551                             strcmp(slash, "/..") == 0) {
552                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
553                                     "'%s' is not a valid file"), strval);
554                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
555                                 goto error;
556                         }
557
558                         *slash = '\0';
559
560                         if (strval[0] != '\0' &&
561                             (stat64(strval, &statbuf) != 0 ||
562                             !S_ISDIR(statbuf.st_mode))) {
563                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
564                                     "'%s' is not a valid directory"),
565                                     strval);
566                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
567                                 goto error;
568                         }
569
570                         *slash = '/';
571                         break;
572
573                 case ZPOOL_PROP_COMMENT:
574                         for (check = strval; *check != '\0'; check++) {
575                                 if (!isprint(*check)) {
576                                         zfs_error_aux(hdl,
577                                             dgettext(TEXT_DOMAIN,
578                                             "comment may only have printable "
579                                             "characters"));
580                                         (void) zfs_error(hdl, EZFS_BADPROP,
581                                             errbuf);
582                                         goto error;
583                                 }
584                         }
585                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
586                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
587                                     "comment must not exceed %d characters"),
588                                     ZPROP_MAX_COMMENT);
589                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
590                                 goto error;
591                         }
592                         break;
593                 case ZPOOL_PROP_READONLY:
594                         if (!flags.import) {
595                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
596                                     "property '%s' can only be set at "
597                                     "import time"), propname);
598                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
599                                 goto error;
600                         }
601                         break;
602                 }
603         }
604
605         return (retprops);
606 error:
607         nvlist_free(retprops);
608         return (NULL);
609 }
610
611 /*
612  * Set zpool property : propname=propval.
613  */
614 int
615 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
616 {
617         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
618         int ret = -1;
619         char errbuf[1024];
620         nvlist_t *nvl = NULL;
621         nvlist_t *realprops;
622         uint64_t version;
623         prop_flags_t flags = { 0 };
624
625         (void) snprintf(errbuf, sizeof (errbuf),
626             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
627             zhp->zpool_name);
628
629         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
630                 return (no_memory(zhp->zpool_hdl));
631
632         if (nvlist_add_string(nvl, propname, propval) != 0) {
633                 nvlist_free(nvl);
634                 return (no_memory(zhp->zpool_hdl));
635         }
636
637         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
638         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
639             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
640                 nvlist_free(nvl);
641                 return (-1);
642         }
643
644         nvlist_free(nvl);
645         nvl = realprops;
646
647         /*
648          * Execute the corresponding ioctl() to set this property.
649          */
650         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
651
652         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
653                 nvlist_free(nvl);
654                 return (-1);
655         }
656
657         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
658
659         zcmd_free_nvlists(&zc);
660         nvlist_free(nvl);
661
662         if (ret)
663                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
664         else
665                 (void) zpool_props_refresh(zhp);
666
667         return (ret);
668 }
669
670 int
671 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
672 {
673         libzfs_handle_t *hdl = zhp->zpool_hdl;
674         zprop_list_t *entry;
675         char buf[ZFS_MAXPROPLEN];
676
677         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
678                 return (-1);
679
680         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
681
682                 if (entry->pl_fixed)
683                         continue;
684
685                 if (entry->pl_prop != ZPROP_INVAL &&
686                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
687                     NULL) == 0) {
688                         if (strlen(buf) > entry->pl_width)
689                                 entry->pl_width = strlen(buf);
690                 }
691         }
692
693         return (0);
694 }
695
696
697 /*
698  * Don't start the slice at the default block of 34; many storage
699  * devices will use a stripe width of 128k, other vendors prefer a 1m
700  * alignment.  It is best to play it safe and ensure a 1m alignment
701  * given 512B blocks.  When the block size is larger by a power of 2
702  * we will still be 1m aligned.  Some devices are sensitive to the
703  * partition ending alignment as well.
704  */
705 #define NEW_START_BLOCK         2048
706 #define PARTITION_END_ALIGNMENT 2048
707
708 /*
709  * Validate the given pool name, optionally putting an extended error message in
710  * 'buf'.
711  */
712 boolean_t
713 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
714 {
715         namecheck_err_t why;
716         char what;
717         int ret;
718
719         ret = pool_namecheck(pool, &why, &what);
720
721         /*
722          * The rules for reserved pool names were extended at a later point.
723          * But we need to support users with existing pools that may now be
724          * invalid.  So we only check for this expanded set of names during a
725          * create (or import), and only in userland.
726          */
727         if (ret == 0 && !isopen &&
728             (strncmp(pool, "mirror", 6) == 0 ||
729             strncmp(pool, "raidz", 5) == 0 ||
730             strncmp(pool, "spare", 5) == 0 ||
731             strcmp(pool, "log") == 0)) {
732                 if (hdl != NULL)
733                         zfs_error_aux(hdl,
734                             dgettext(TEXT_DOMAIN, "name is reserved"));
735                 return (B_FALSE);
736         }
737
738
739         if (ret != 0) {
740                 if (hdl != NULL) {
741                         switch (why) {
742                         case NAME_ERR_TOOLONG:
743                                 zfs_error_aux(hdl,
744                                     dgettext(TEXT_DOMAIN, "name is too long"));
745                                 break;
746
747                         case NAME_ERR_INVALCHAR:
748                                 zfs_error_aux(hdl,
749                                     dgettext(TEXT_DOMAIN, "invalid character "
750                                     "'%c' in pool name"), what);
751                                 break;
752
753                         case NAME_ERR_NOLETTER:
754                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
755                                     "name must begin with a letter"));
756                                 break;
757
758                         case NAME_ERR_RESERVED:
759                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
760                                     "name is reserved"));
761                                 break;
762
763                         case NAME_ERR_DISKLIKE:
764                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
765                                     "pool name is reserved"));
766                                 break;
767
768                         case NAME_ERR_LEADING_SLASH:
769                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
770                                     "leading slash in name"));
771                                 break;
772
773                         case NAME_ERR_EMPTY_COMPONENT:
774                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
775                                     "empty component in name"));
776                                 break;
777
778                         case NAME_ERR_TRAILING_SLASH:
779                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
780                                     "trailing slash in name"));
781                                 break;
782
783                         case NAME_ERR_MULTIPLE_AT:
784                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
785                                     "multiple '@' delimiters in name"));
786                                 break;
787                         case NAME_ERR_NO_AT:
788                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
789                                     "permission set is missing '@'"));
790                                 break;
791                         }
792                 }
793                 return (B_FALSE);
794         }
795
796         return (B_TRUE);
797 }
798
799 /*
800  * Open a handle to the given pool, even if the pool is currently in the FAULTED
801  * state.
802  */
803 zpool_handle_t *
804 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
805 {
806         zpool_handle_t *zhp;
807         boolean_t missing;
808
809         /*
810          * Make sure the pool name is valid.
811          */
812         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
813                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
814                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
815                     pool);
816                 return (NULL);
817         }
818
819         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
820                 return (NULL);
821
822         zhp->zpool_hdl = hdl;
823         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
824
825         if (zpool_refresh_stats(zhp, &missing) != 0) {
826                 zpool_close(zhp);
827                 return (NULL);
828         }
829
830         if (missing) {
831                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
832                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
833                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
834                 zpool_close(zhp);
835                 return (NULL);
836         }
837
838         return (zhp);
839 }
840
841 /*
842  * Like the above, but silent on error.  Used when iterating over pools (because
843  * the configuration cache may be out of date).
844  */
845 int
846 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
847 {
848         zpool_handle_t *zhp;
849         boolean_t missing;
850
851         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
852                 return (-1);
853
854         zhp->zpool_hdl = hdl;
855         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
856
857         if (zpool_refresh_stats(zhp, &missing) != 0) {
858                 zpool_close(zhp);
859                 return (-1);
860         }
861
862         if (missing) {
863                 zpool_close(zhp);
864                 *ret = NULL;
865                 return (0);
866         }
867
868         *ret = zhp;
869         return (0);
870 }
871
872 /*
873  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
874  * state.
875  */
876 zpool_handle_t *
877 zpool_open(libzfs_handle_t *hdl, const char *pool)
878 {
879         zpool_handle_t *zhp;
880
881         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
882                 return (NULL);
883
884         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
885                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
886                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
887                 zpool_close(zhp);
888                 return (NULL);
889         }
890
891         return (zhp);
892 }
893
894 /*
895  * Close the handle.  Simply frees the memory associated with the handle.
896  */
897 void
898 zpool_close(zpool_handle_t *zhp)
899 {
900         if (zhp->zpool_config)
901                 nvlist_free(zhp->zpool_config);
902         if (zhp->zpool_old_config)
903                 nvlist_free(zhp->zpool_old_config);
904         if (zhp->zpool_props)
905                 nvlist_free(zhp->zpool_props);
906         free(zhp);
907 }
908
909 /*
910  * Return the name of the pool.
911  */
912 const char *
913 zpool_get_name(zpool_handle_t *zhp)
914 {
915         return (zhp->zpool_name);
916 }
917
918
919 /*
920  * Return the state of the pool (ACTIVE or UNAVAILABLE)
921  */
922 int
923 zpool_get_state(zpool_handle_t *zhp)
924 {
925         return (zhp->zpool_state);
926 }
927
928 /*
929  * Create the named pool, using the provided vdev list.  It is assumed
930  * that the consumer has already validated the contents of the nvlist, so we
931  * don't have to worry about error semantics.
932  */
933 int
934 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
935     nvlist_t *props, nvlist_t *fsprops)
936 {
937         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
938         nvlist_t *zc_fsprops = NULL;
939         nvlist_t *zc_props = NULL;
940         char msg[1024];
941         char *altroot;
942         int ret = -1;
943
944         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
945             "cannot create '%s'"), pool);
946
947         if (!zpool_name_valid(hdl, B_FALSE, pool))
948                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
949
950         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
951                 return (-1);
952
953         if (props) {
954                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
955
956                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
957                     SPA_VERSION_1, flags, msg)) == NULL) {
958                         goto create_failed;
959                 }
960         }
961
962         if (fsprops) {
963                 uint64_t zoned;
964                 char *zonestr;
965
966                 zoned = ((nvlist_lookup_string(fsprops,
967                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
968                     strcmp(zonestr, "on") == 0);
969
970                 if ((zc_fsprops = zfs_valid_proplist(hdl,
971                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
972                         goto create_failed;
973                 }
974                 if (!zc_props &&
975                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
976                         goto create_failed;
977                 }
978                 if (nvlist_add_nvlist(zc_props,
979                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
980                         goto create_failed;
981                 }
982         }
983
984         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
985                 goto create_failed;
986
987         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
988
989         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
990
991                 zcmd_free_nvlists(&zc);
992                 nvlist_free(zc_props);
993                 nvlist_free(zc_fsprops);
994
995                 switch (errno) {
996                 case EBUSY:
997                         /*
998                          * This can happen if the user has specified the same
999                          * device multiple times.  We can't reliably detect this
1000                          * until we try to add it and see we already have a
1001                          * label.  This can also happen under if the device is
1002                          * part of an active md or lvm device.
1003                          */
1004                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1005                             "one or more vdevs refer to the same device, or one of\n"
1006                             "the devices is part of an active md or lvm device"));
1007                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1008
1009                 case EOVERFLOW:
1010                         /*
1011                          * This occurs when one of the devices is below
1012                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1013                          * device was the problem device since there's no
1014                          * reliable way to determine device size from userland.
1015                          */
1016                         {
1017                                 char buf[64];
1018
1019                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1020
1021                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1022                                     "one or more devices is less than the "
1023                                     "minimum size (%s)"), buf);
1024                         }
1025                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1026
1027                 case ENOSPC:
1028                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1029                             "one or more devices is out of space"));
1030                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1031
1032                 case ENOTBLK:
1033                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1034                             "cache device must be a disk or disk slice"));
1035                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1036
1037                 default:
1038                         return (zpool_standard_error(hdl, errno, msg));
1039                 }
1040         }
1041
1042         /*
1043          * If this is an alternate root pool, then we automatically set the
1044          * mountpoint of the root dataset to be '/'.
1045          */
1046         if (nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT),
1047             &altroot) == 0) {
1048                 zfs_handle_t *zhp;
1049
1050                 verify((zhp = zfs_open(hdl, pool, ZFS_TYPE_DATASET)) != NULL);
1051                 verify(zfs_prop_set(zhp, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
1052                     "/") == 0);
1053
1054                 zfs_close(zhp);
1055         }
1056
1057 create_failed:
1058         zcmd_free_nvlists(&zc);
1059         nvlist_free(zc_props);
1060         nvlist_free(zc_fsprops);
1061         return (ret);
1062 }
1063
1064 /*
1065  * Destroy the given pool.  It is up to the caller to ensure that there are no
1066  * datasets left in the pool.
1067  */
1068 int
1069 zpool_destroy(zpool_handle_t *zhp)
1070 {
1071         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1072         zfs_handle_t *zfp = NULL;
1073         libzfs_handle_t *hdl = zhp->zpool_hdl;
1074         char msg[1024];
1075
1076         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1077             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1078                 return (-1);
1079
1080         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1081
1082         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1083                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1084                     "cannot destroy '%s'"), zhp->zpool_name);
1085
1086                 if (errno == EROFS) {
1087                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1088                             "one or more devices is read only"));
1089                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1090                 } else {
1091                         (void) zpool_standard_error(hdl, errno, msg);
1092                 }
1093
1094                 if (zfp)
1095                         zfs_close(zfp);
1096                 return (-1);
1097         }
1098
1099         if (zfp) {
1100                 remove_mountpoint(zfp);
1101                 zfs_close(zfp);
1102         }
1103
1104         return (0);
1105 }
1106
1107 /*
1108  * Add the given vdevs to the pool.  The caller must have already performed the
1109  * necessary verification to ensure that the vdev specification is well-formed.
1110  */
1111 int
1112 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1113 {
1114         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1115         int ret;
1116         libzfs_handle_t *hdl = zhp->zpool_hdl;
1117         char msg[1024];
1118         nvlist_t **spares, **l2cache;
1119         uint_t nspares, nl2cache;
1120
1121         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1122             "cannot add to '%s'"), zhp->zpool_name);
1123
1124         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1125             SPA_VERSION_SPARES &&
1126             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1127             &spares, &nspares) == 0) {
1128                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1129                     "upgraded to add hot spares"));
1130                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1131         }
1132
1133 #if defined(__sun__) || defined(__sun)
1134         if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1135             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1136                 uint64_t s;
1137
1138                 for (s = 0; s < nspares; s++) {
1139                         char *path;
1140
1141                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1142                             &path) == 0 && pool_uses_efi(spares[s])) {
1143                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1144                                     "device '%s' contains an EFI label and "
1145                                     "cannot be used on root pools."),
1146                                     zpool_vdev_name(hdl, NULL, spares[s],
1147                                     B_FALSE));
1148                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1149                         }
1150                 }
1151         }
1152 #endif
1153
1154         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1155             SPA_VERSION_L2CACHE &&
1156             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1157             &l2cache, &nl2cache) == 0) {
1158                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1159                     "upgraded to add cache devices"));
1160                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1161         }
1162
1163         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1164                 return (-1);
1165         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1166
1167         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1168                 switch (errno) {
1169                 case EBUSY:
1170                         /*
1171                          * This can happen if the user has specified the same
1172                          * device multiple times.  We can't reliably detect this
1173                          * until we try to add it and see we already have a
1174                          * label.
1175                          */
1176                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1177                             "one or more vdevs refer to the same device"));
1178                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1179                         break;
1180
1181                 case EOVERFLOW:
1182                         /*
1183                          * This occurrs when one of the devices is below
1184                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1185                          * device was the problem device since there's no
1186                          * reliable way to determine device size from userland.
1187                          */
1188                         {
1189                                 char buf[64];
1190
1191                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1192
1193                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1194                                     "device is less than the minimum "
1195                                     "size (%s)"), buf);
1196                         }
1197                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1198                         break;
1199
1200                 case ENOTSUP:
1201                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1202                             "pool must be upgraded to add these vdevs"));
1203                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1204                         break;
1205
1206                 case EDOM:
1207                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1208                             "root pool can not have multiple vdevs"
1209                             " or separate logs"));
1210                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1211                         break;
1212
1213                 case ENOTBLK:
1214                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1215                             "cache device must be a disk or disk slice"));
1216                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1217                         break;
1218
1219                 default:
1220                         (void) zpool_standard_error(hdl, errno, msg);
1221                 }
1222
1223                 ret = -1;
1224         } else {
1225                 ret = 0;
1226         }
1227
1228         zcmd_free_nvlists(&zc);
1229
1230         return (ret);
1231 }
1232
1233 /*
1234  * Exports the pool from the system.  The caller must ensure that there are no
1235  * mounted datasets in the pool.
1236  */
1237 int
1238 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce)
1239 {
1240         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1241         char msg[1024];
1242
1243         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1244             "cannot export '%s'"), zhp->zpool_name);
1245
1246         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1247         zc.zc_cookie = force;
1248         zc.zc_guid = hardforce;
1249
1250         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1251                 switch (errno) {
1252                 case EXDEV:
1253                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1254                             "use '-f' to override the following errors:\n"
1255                             "'%s' has an active shared spare which could be"
1256                             " used by other pools once '%s' is exported."),
1257                             zhp->zpool_name, zhp->zpool_name);
1258                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1259                             msg));
1260                 default:
1261                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1262                             msg));
1263                 }
1264         }
1265
1266         return (0);
1267 }
1268
1269 int
1270 zpool_export(zpool_handle_t *zhp, boolean_t force)
1271 {
1272         return (zpool_export_common(zhp, force, B_FALSE));
1273 }
1274
1275 int
1276 zpool_export_force(zpool_handle_t *zhp)
1277 {
1278         return (zpool_export_common(zhp, B_TRUE, B_TRUE));
1279 }
1280
1281 static void
1282 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1283     nvlist_t *config)
1284 {
1285         nvlist_t *nv = NULL;
1286         uint64_t rewindto;
1287         int64_t loss = -1;
1288         struct tm t;
1289         char timestr[128];
1290
1291         if (!hdl->libzfs_printerr || config == NULL)
1292                 return;
1293
1294         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0)
1295                 return;
1296
1297         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1298                 return;
1299         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1300
1301         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1302             strftime(timestr, 128, "%c", &t) != 0) {
1303                 if (dryrun) {
1304                         (void) printf(dgettext(TEXT_DOMAIN,
1305                             "Would be able to return %s "
1306                             "to its state as of %s.\n"),
1307                             name, timestr);
1308                 } else {
1309                         (void) printf(dgettext(TEXT_DOMAIN,
1310                             "Pool %s returned to its state as of %s.\n"),
1311                             name, timestr);
1312                 }
1313                 if (loss > 120) {
1314                         (void) printf(dgettext(TEXT_DOMAIN,
1315                             "%s approximately %lld "),
1316                             dryrun ? "Would discard" : "Discarded",
1317                             ((longlong_t)loss + 30) / 60);
1318                         (void) printf(dgettext(TEXT_DOMAIN,
1319                             "minutes of transactions.\n"));
1320                 } else if (loss > 0) {
1321                         (void) printf(dgettext(TEXT_DOMAIN,
1322                             "%s approximately %lld "),
1323                             dryrun ? "Would discard" : "Discarded",
1324                             (longlong_t)loss);
1325                         (void) printf(dgettext(TEXT_DOMAIN,
1326                             "seconds of transactions.\n"));
1327                 }
1328         }
1329 }
1330
1331 void
1332 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1333     nvlist_t *config)
1334 {
1335         nvlist_t *nv = NULL;
1336         int64_t loss = -1;
1337         uint64_t edata = UINT64_MAX;
1338         uint64_t rewindto;
1339         struct tm t;
1340         char timestr[128];
1341
1342         if (!hdl->libzfs_printerr)
1343                 return;
1344
1345         if (reason >= 0)
1346                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1347         else
1348                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1349
1350         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1351         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1352             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1353                 goto no_info;
1354
1355         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1356         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1357             &edata);
1358
1359         (void) printf(dgettext(TEXT_DOMAIN,
1360             "Recovery is possible, but will result in some data loss.\n"));
1361
1362         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1363             strftime(timestr, 128, "%c", &t) != 0) {
1364                 (void) printf(dgettext(TEXT_DOMAIN,
1365                     "\tReturning the pool to its state as of %s\n"
1366                     "\tshould correct the problem.  "),
1367                     timestr);
1368         } else {
1369                 (void) printf(dgettext(TEXT_DOMAIN,
1370                     "\tReverting the pool to an earlier state "
1371                     "should correct the problem.\n\t"));
1372         }
1373
1374         if (loss > 120) {
1375                 (void) printf(dgettext(TEXT_DOMAIN,
1376                     "Approximately %lld minutes of data\n"
1377                     "\tmust be discarded, irreversibly.  "),
1378                     ((longlong_t)loss + 30) / 60);
1379         } else if (loss > 0) {
1380                 (void) printf(dgettext(TEXT_DOMAIN,
1381                     "Approximately %lld seconds of data\n"
1382                     "\tmust be discarded, irreversibly.  "),
1383                     (longlong_t)loss);
1384         }
1385         if (edata != 0 && edata != UINT64_MAX) {
1386                 if (edata == 1) {
1387                         (void) printf(dgettext(TEXT_DOMAIN,
1388                             "After rewind, at least\n"
1389                             "\tone persistent user-data error will remain.  "));
1390                 } else {
1391                         (void) printf(dgettext(TEXT_DOMAIN,
1392                             "After rewind, several\n"
1393                             "\tpersistent user-data errors will remain.  "));
1394                 }
1395         }
1396         (void) printf(dgettext(TEXT_DOMAIN,
1397             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1398             reason >= 0 ? "clear" : "import", name);
1399
1400         (void) printf(dgettext(TEXT_DOMAIN,
1401             "A scrub of the pool\n"
1402             "\tis strongly recommended after recovery.\n"));
1403         return;
1404
1405 no_info:
1406         (void) printf(dgettext(TEXT_DOMAIN,
1407             "Destroy and re-create the pool from\n\ta backup source.\n"));
1408 }
1409
1410 /*
1411  * zpool_import() is a contracted interface. Should be kept the same
1412  * if possible.
1413  *
1414  * Applications should use zpool_import_props() to import a pool with
1415  * new properties value to be set.
1416  */
1417 int
1418 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1419     char *altroot)
1420 {
1421         nvlist_t *props = NULL;
1422         int ret;
1423
1424         if (altroot != NULL) {
1425                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1426                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1427                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1428                             newname));
1429                 }
1430
1431                 if (nvlist_add_string(props,
1432                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1433                     nvlist_add_string(props,
1434                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1435                         nvlist_free(props);
1436                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1437                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1438                             newname));
1439                 }
1440         }
1441
1442         ret = zpool_import_props(hdl, config, newname, props,
1443             ZFS_IMPORT_NORMAL);
1444         if (props)
1445                 nvlist_free(props);
1446         return (ret);
1447 }
1448
1449 static void
1450 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1451     int indent)
1452 {
1453         nvlist_t **child;
1454         uint_t c, children;
1455         char *vname;
1456         uint64_t is_log = 0;
1457
1458         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1459             &is_log);
1460
1461         if (name != NULL)
1462                 (void) printf("\t%*s%s%s\n", indent, "", name,
1463                     is_log ? " [log]" : "");
1464
1465         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1466             &child, &children) != 0)
1467                 return;
1468
1469         for (c = 0; c < children; c++) {
1470                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1471                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1472                 free(vname);
1473         }
1474 }
1475
1476 /*
1477  * Import the given pool using the known configuration and a list of
1478  * properties to be set. The configuration should have come from
1479  * zpool_find_import(). The 'newname' parameters control whether the pool
1480  * is imported with a different name.
1481  */
1482 int
1483 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1484     nvlist_t *props, int flags)
1485 {
1486         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1487         zpool_rewind_policy_t policy;
1488         nvlist_t *nv = NULL;
1489         nvlist_t *nvinfo = NULL;
1490         nvlist_t *missing = NULL;
1491         char *thename;
1492         char *origname;
1493         int ret;
1494         int error = 0;
1495         char errbuf[1024];
1496
1497         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1498             &origname) == 0);
1499
1500         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1501             "cannot import pool '%s'"), origname);
1502
1503         if (newname != NULL) {
1504                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1505                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1506                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1507                             newname));
1508                 thename = (char *)newname;
1509         } else {
1510                 thename = origname;
1511         }
1512
1513         if (props) {
1514                 uint64_t version;
1515                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1516
1517                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1518                     &version) == 0);
1519
1520                 if ((props = zpool_valid_proplist(hdl, origname,
1521                     props, version, flags, errbuf)) == NULL) {
1522                         return (-1);
1523                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1524                         nvlist_free(props);
1525                         return (-1);
1526                 }
1527         }
1528
1529         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1530
1531         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1532             &zc.zc_guid) == 0);
1533
1534         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1535                 nvlist_free(props);
1536                 return (-1);
1537         }
1538         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1539                 nvlist_free(props);
1540                 return (-1);
1541         }
1542
1543         zc.zc_cookie = flags;
1544         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1545             errno == ENOMEM) {
1546                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1547                         zcmd_free_nvlists(&zc);
1548                         return (-1);
1549                 }
1550         }
1551         if (ret != 0)
1552                 error = errno;
1553
1554         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1555         zpool_get_rewind_policy(config, &policy);
1556
1557         if (error) {
1558                 char desc[1024];
1559
1560                 /*
1561                  * Dry-run failed, but we print out what success
1562                  * looks like if we found a best txg
1563                  */
1564                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1565                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1566                             B_TRUE, nv);
1567                         nvlist_free(nv);
1568                         return (-1);
1569                 }
1570
1571                 if (newname == NULL)
1572                         (void) snprintf(desc, sizeof (desc),
1573                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1574                             thename);
1575                 else
1576                         (void) snprintf(desc, sizeof (desc),
1577                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1578                             origname, thename);
1579
1580                 switch (error) {
1581                 case ENOTSUP:
1582                         /*
1583                          * Unsupported version.
1584                          */
1585                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1586                         break;
1587
1588                 case EINVAL:
1589                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1590                         break;
1591
1592                 case EROFS:
1593                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1594                             "one or more devices is read only"));
1595                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1596                         break;
1597
1598                 case ENXIO:
1599                         if (nv && nvlist_lookup_nvlist(nv,
1600                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1601                             nvlist_lookup_nvlist(nvinfo,
1602                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1603                                 (void) printf(dgettext(TEXT_DOMAIN,
1604                                     "The devices below are missing, use "
1605                                     "'-m' to import the pool anyway:\n"));
1606                                 print_vdev_tree(hdl, NULL, missing, 2);
1607                                 (void) printf("\n");
1608                         }
1609                         (void) zpool_standard_error(hdl, error, desc);
1610                         break;
1611
1612                 case EEXIST:
1613                         (void) zpool_standard_error(hdl, error, desc);
1614                         break;
1615
1616                 case EBUSY:
1617                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1618                             "one or more devices are already in use\n"));
1619                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1620                         break;
1621
1622                 default:
1623                         (void) zpool_standard_error(hdl, error, desc);
1624                         zpool_explain_recover(hdl,
1625                             newname ? origname : thename, -error, nv);
1626                         break;
1627                 }
1628
1629                 nvlist_free(nv);
1630                 ret = -1;
1631         } else {
1632                 zpool_handle_t *zhp;
1633
1634                 /*
1635                  * This should never fail, but play it safe anyway.
1636                  */
1637                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1638                         ret = -1;
1639                 else if (zhp != NULL)
1640                         zpool_close(zhp);
1641                 if (policy.zrp_request &
1642                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1643                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1644                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1645                 }
1646                 nvlist_free(nv);
1647                 return (0);
1648         }
1649
1650         zcmd_free_nvlists(&zc);
1651         nvlist_free(props);
1652
1653         return (ret);
1654 }
1655
1656 /*
1657  * Scan the pool.
1658  */
1659 int
1660 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1661 {
1662         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1663         char msg[1024];
1664         libzfs_handle_t *hdl = zhp->zpool_hdl;
1665
1666         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1667         zc.zc_cookie = func;
1668
1669         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1670             (errno == ENOENT && func != POOL_SCAN_NONE))
1671                 return (0);
1672
1673         if (func == POOL_SCAN_SCRUB) {
1674                 (void) snprintf(msg, sizeof (msg),
1675                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1676         } else if (func == POOL_SCAN_NONE) {
1677                 (void) snprintf(msg, sizeof (msg),
1678                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1679                     zc.zc_name);
1680         } else {
1681                 assert(!"unexpected result");
1682         }
1683
1684         if (errno == EBUSY) {
1685                 nvlist_t *nvroot;
1686                 pool_scan_stat_t *ps = NULL;
1687                 uint_t psc;
1688
1689                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1690                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1691                 (void) nvlist_lookup_uint64_array(nvroot,
1692                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1693                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1694                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1695                 else
1696                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1697         } else if (errno == ENOENT) {
1698                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1699         } else {
1700                 return (zpool_standard_error(hdl, errno, msg));
1701         }
1702 }
1703
1704 /*
1705  * Find a vdev that matches the search criteria specified. We use the
1706  * the nvpair name to determine how we should look for the device.
1707  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1708  * spare; but FALSE if its an INUSE spare.
1709  */
1710 static nvlist_t *
1711 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1712     boolean_t *l2cache, boolean_t *log)
1713 {
1714         uint_t c, children;
1715         nvlist_t **child;
1716         nvlist_t *ret;
1717         uint64_t is_log;
1718         char *srchkey;
1719         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1720
1721         /* Nothing to look for */
1722         if (search == NULL || pair == NULL)
1723                 return (NULL);
1724
1725         /* Obtain the key we will use to search */
1726         srchkey = nvpair_name(pair);
1727
1728         switch (nvpair_type(pair)) {
1729         case DATA_TYPE_UINT64:
1730                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1731                         uint64_t srchval, theguid;
1732
1733                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1734                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1735                             &theguid) == 0);
1736                         if (theguid == srchval)
1737                                 return (nv);
1738                 }
1739                 break;
1740
1741         case DATA_TYPE_STRING: {
1742                 char *srchval, *val;
1743
1744                 verify(nvpair_value_string(pair, &srchval) == 0);
1745                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1746                         break;
1747
1748                 /*
1749                  * Search for the requested value. Special cases:
1750                  *
1751                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1752                  *   "-part1", or "p1".  The suffix is hidden from the user,
1753                  *   but included in the string, so this matches around it.
1754                  * - ZPOOL_CONFIG_PATH for short names zfs_strcmp_shortname()
1755                  *   is used to check all possible expanded paths.
1756                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1757                  *
1758                  * Otherwise, all other searches are simple string compares.
1759                  */
1760                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0) {
1761                         uint64_t wholedisk = 0;
1762
1763                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1764                             &wholedisk);
1765                         if (zfs_strcmp_pathname(srchval, val, wholedisk) == 0)
1766                                 return (nv);
1767
1768                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
1769                         char *type, *idx, *end, *p;
1770                         uint64_t id, vdev_id;
1771
1772                         /*
1773                          * Determine our vdev type, keeping in mind
1774                          * that the srchval is composed of a type and
1775                          * vdev id pair (i.e. mirror-4).
1776                          */
1777                         if ((type = strdup(srchval)) == NULL)
1778                                 return (NULL);
1779
1780                         if ((p = strrchr(type, '-')) == NULL) {
1781                                 free(type);
1782                                 break;
1783                         }
1784                         idx = p + 1;
1785                         *p = '\0';
1786
1787                         /*
1788                          * If the types don't match then keep looking.
1789                          */
1790                         if (strncmp(val, type, strlen(val)) != 0) {
1791                                 free(type);
1792                                 break;
1793                         }
1794
1795                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
1796                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1797                             strncmp(type, VDEV_TYPE_MIRROR,
1798                             strlen(VDEV_TYPE_MIRROR)) == 0);
1799                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
1800                             &id) == 0);
1801
1802                         errno = 0;
1803                         vdev_id = strtoull(idx, &end, 10);
1804
1805                         free(type);
1806                         if (errno != 0)
1807                                 return (NULL);
1808
1809                         /*
1810                          * Now verify that we have the correct vdev id.
1811                          */
1812                         if (vdev_id == id)
1813                                 return (nv);
1814                 }
1815
1816                 /*
1817                  * Common case
1818                  */
1819                 if (strcmp(srchval, val) == 0)
1820                         return (nv);
1821                 break;
1822         }
1823
1824         default:
1825                 break;
1826         }
1827
1828         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1829             &child, &children) != 0)
1830                 return (NULL);
1831
1832         for (c = 0; c < children; c++) {
1833                 if ((ret = vdev_to_nvlist_iter(child[c], search,
1834                     avail_spare, l2cache, NULL)) != NULL) {
1835                         /*
1836                          * The 'is_log' value is only set for the toplevel
1837                          * vdev, not the leaf vdevs.  So we always lookup the
1838                          * log device from the root of the vdev tree (where
1839                          * 'log' is non-NULL).
1840                          */
1841                         if (log != NULL &&
1842                             nvlist_lookup_uint64(child[c],
1843                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
1844                             is_log) {
1845                                 *log = B_TRUE;
1846                         }
1847                         return (ret);
1848                 }
1849         }
1850
1851         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1852             &child, &children) == 0) {
1853                 for (c = 0; c < children; c++) {
1854                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1855                             avail_spare, l2cache, NULL)) != NULL) {
1856                                 *avail_spare = B_TRUE;
1857                                 return (ret);
1858                         }
1859                 }
1860         }
1861
1862         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1863             &child, &children) == 0) {
1864                 for (c = 0; c < children; c++) {
1865                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1866                             avail_spare, l2cache, NULL)) != NULL) {
1867                                 *l2cache = B_TRUE;
1868                                 return (ret);
1869                         }
1870                 }
1871         }
1872
1873         return (NULL);
1874 }
1875
1876 /*
1877  * Given a physical path (minus the "/devices" prefix), find the
1878  * associated vdev.
1879  */
1880 nvlist_t *
1881 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
1882     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
1883 {
1884         nvlist_t *search, *nvroot, *ret;
1885
1886         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1887         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
1888
1889         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1890             &nvroot) == 0);
1891
1892         *avail_spare = B_FALSE;
1893         *l2cache = B_FALSE;
1894         if (log != NULL)
1895                 *log = B_FALSE;
1896         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1897         nvlist_free(search);
1898
1899         return (ret);
1900 }
1901
1902 /*
1903  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
1904  */
1905 boolean_t
1906 zpool_vdev_is_interior(const char *name)
1907 {
1908         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1909             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
1910                 return (B_TRUE);
1911         return (B_FALSE);
1912 }
1913
1914 nvlist_t *
1915 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
1916     boolean_t *l2cache, boolean_t *log)
1917 {
1918         char *end;
1919         nvlist_t *nvroot, *search, *ret;
1920         uint64_t guid;
1921
1922         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1923
1924         guid = strtoull(path, &end, 10);
1925         if (guid != 0 && *end == '\0') {
1926                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
1927         } else if (zpool_vdev_is_interior(path)) {
1928                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
1929         } else {
1930                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
1931         }
1932
1933         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1934             &nvroot) == 0);
1935
1936         *avail_spare = B_FALSE;
1937         *l2cache = B_FALSE;
1938         if (log != NULL)
1939                 *log = B_FALSE;
1940         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1941         nvlist_free(search);
1942
1943         return (ret);
1944 }
1945
1946 static int
1947 vdev_online(nvlist_t *nv)
1948 {
1949         uint64_t ival;
1950
1951         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
1952             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
1953             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
1954                 return (0);
1955
1956         return (1);
1957 }
1958
1959 /*
1960  * Helper function for zpool_get_physpaths().
1961  */
1962 static int
1963 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
1964     size_t *bytes_written)
1965 {
1966         size_t bytes_left, pos, rsz;
1967         char *tmppath;
1968         const char *format;
1969
1970         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
1971             &tmppath) != 0)
1972                 return (EZFS_NODEVICE);
1973
1974         pos = *bytes_written;
1975         bytes_left = physpath_size - pos;
1976         format = (pos == 0) ? "%s" : " %s";
1977
1978         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
1979         *bytes_written += rsz;
1980
1981         if (rsz >= bytes_left) {
1982                 /* if physpath was not copied properly, clear it */
1983                 if (bytes_left != 0) {
1984                         physpath[pos] = 0;
1985                 }
1986                 return (EZFS_NOSPC);
1987         }
1988         return (0);
1989 }
1990
1991 static int
1992 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
1993     size_t *rsz, boolean_t is_spare)
1994 {
1995         char *type;
1996         int ret;
1997
1998         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
1999                 return (EZFS_INVALCONFIG);
2000
2001         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2002                 /*
2003                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2004                  * For a spare vdev, we only want to boot from the active
2005                  * spare device.
2006                  */
2007                 if (is_spare) {
2008                         uint64_t spare = 0;
2009                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2010                             &spare);
2011                         if (!spare)
2012                                 return (EZFS_INVALCONFIG);
2013                 }
2014
2015                 if (vdev_online(nv)) {
2016                         if ((ret = vdev_get_one_physpath(nv, physpath,
2017                             phypath_size, rsz)) != 0)
2018                                 return (ret);
2019                 }
2020         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2021             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2022             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2023                 nvlist_t **child;
2024                 uint_t count;
2025                 int i, ret;
2026
2027                 if (nvlist_lookup_nvlist_array(nv,
2028                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2029                         return (EZFS_INVALCONFIG);
2030
2031                 for (i = 0; i < count; i++) {
2032                         ret = vdev_get_physpaths(child[i], physpath,
2033                             phypath_size, rsz, is_spare);
2034                         if (ret == EZFS_NOSPC)
2035                                 return (ret);
2036                 }
2037         }
2038
2039         return (EZFS_POOL_INVALARG);
2040 }
2041
2042 /*
2043  * Get phys_path for a root pool config.
2044  * Return 0 on success; non-zero on failure.
2045  */
2046 static int
2047 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2048 {
2049         size_t rsz;
2050         nvlist_t *vdev_root;
2051         nvlist_t **child;
2052         uint_t count;
2053         char *type;
2054
2055         rsz = 0;
2056
2057         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2058             &vdev_root) != 0)
2059                 return (EZFS_INVALCONFIG);
2060
2061         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2062             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2063             &child, &count) != 0)
2064                 return (EZFS_INVALCONFIG);
2065
2066 #if defined(__sun__) || defined(__sun)
2067         /*
2068          * root pool can not have EFI labeled disks and can only have
2069          * a single top-level vdev.
2070          */
2071         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2072             pool_uses_efi(vdev_root))
2073                 return (EZFS_POOL_INVALARG);
2074 #endif
2075
2076         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2077             B_FALSE);
2078
2079         /* No online devices */
2080         if (rsz == 0)
2081                 return (EZFS_NODEVICE);
2082
2083         return (0);
2084 }
2085
2086 /*
2087  * Get phys_path for a root pool
2088  * Return 0 on success; non-zero on failure.
2089  */
2090 int
2091 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2092 {
2093         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2094             phypath_size));
2095 }
2096
2097 /*
2098  * If the device has being dynamically expanded then we need to relabel
2099  * the disk to use the new unallocated space.
2100  */
2101 static int
2102 zpool_relabel_disk(libzfs_handle_t *hdl, const char *path, const char *msg)
2103 {
2104         int fd, error;
2105
2106         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0) {
2107                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2108                     "relabel '%s': unable to open device: %d"), path, errno);
2109                 return (zfs_error(hdl, EZFS_OPENFAILED, msg));
2110         }
2111
2112         /*
2113          * It's possible that we might encounter an error if the device
2114          * does not have any unallocated space left. If so, we simply
2115          * ignore that error and continue on.
2116          *
2117          * Also, we don't call efi_rescan() - that would just return EBUSY.
2118          * The module will do it for us in vdev_disk_open().
2119          */
2120         error = efi_use_whole_disk(fd);
2121         (void) close(fd);
2122         if (error && error != VT_ENOSPC) {
2123                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2124                     "relabel '%s': unable to read disk capacity"), path);
2125                 return (zfs_error(hdl, EZFS_NOCAP, msg));
2126         }
2127         return (0);
2128 }
2129
2130 /*
2131  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2132  * ZFS_ONLINE_* flags.
2133  */
2134 int
2135 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2136     vdev_state_t *newstate)
2137 {
2138         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2139         char msg[1024];
2140         nvlist_t *tgt;
2141         boolean_t avail_spare, l2cache, islog;
2142         libzfs_handle_t *hdl = zhp->zpool_hdl;
2143         int error;
2144
2145         if (flags & ZFS_ONLINE_EXPAND) {
2146                 (void) snprintf(msg, sizeof (msg),
2147                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2148         } else {
2149                 (void) snprintf(msg, sizeof (msg),
2150                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2151         }
2152
2153         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2154         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2155             &islog)) == NULL)
2156                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2157
2158         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2159
2160         if (avail_spare)
2161                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2162
2163         if (flags & ZFS_ONLINE_EXPAND ||
2164             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2165                 uint64_t wholedisk = 0;
2166
2167                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2168                     &wholedisk);
2169
2170                 /*
2171                  * XXX - L2ARC 1.0 devices can't support expansion.
2172                  */
2173                 if (l2cache) {
2174                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2175                             "cannot expand cache devices"));
2176                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2177                 }
2178
2179                 if (wholedisk) {
2180                         const char *fullpath = path;
2181                         char buf[MAXPATHLEN];
2182
2183                         if (path[0] != '/') {
2184                                 error = zfs_resolve_shortname(path, buf,
2185                                     sizeof(buf));
2186                                 if (error != 0)
2187                                         return (zfs_error(hdl, EZFS_NODEVICE,
2188                                             msg));
2189
2190                                 fullpath = buf;
2191                         }
2192
2193                         error = zpool_relabel_disk(hdl, fullpath, msg);
2194                         if (error != 0)
2195                                 return (error);
2196                 }
2197         }
2198
2199         zc.zc_cookie = VDEV_STATE_ONLINE;
2200         zc.zc_obj = flags;
2201
2202         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2203                 if (errno == EINVAL) {
2204                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2205                             "from this pool into a new one.  Use '%s' "
2206                             "instead"), "zpool detach");
2207                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2208                 }
2209                 return (zpool_standard_error(hdl, errno, msg));
2210         }
2211
2212         *newstate = zc.zc_cookie;
2213         return (0);
2214 }
2215
2216 /*
2217  * Take the specified vdev offline
2218  */
2219 int
2220 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2221 {
2222         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2223         char msg[1024];
2224         nvlist_t *tgt;
2225         boolean_t avail_spare, l2cache;
2226         libzfs_handle_t *hdl = zhp->zpool_hdl;
2227
2228         (void) snprintf(msg, sizeof (msg),
2229             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2230
2231         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2232         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2233             NULL)) == NULL)
2234                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2235
2236         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2237
2238         if (avail_spare)
2239                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2240
2241         zc.zc_cookie = VDEV_STATE_OFFLINE;
2242         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2243
2244         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2245                 return (0);
2246
2247         switch (errno) {
2248         case EBUSY:
2249
2250                 /*
2251                  * There are no other replicas of this device.
2252                  */
2253                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2254
2255         case EEXIST:
2256                 /*
2257                  * The log device has unplayed logs
2258                  */
2259                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2260
2261         default:
2262                 return (zpool_standard_error(hdl, errno, msg));
2263         }
2264 }
2265
2266 /*
2267  * Mark the given vdev faulted.
2268  */
2269 int
2270 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2271 {
2272         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2273         char msg[1024];
2274         libzfs_handle_t *hdl = zhp->zpool_hdl;
2275
2276         (void) snprintf(msg, sizeof (msg),
2277            dgettext(TEXT_DOMAIN, "cannot fault %llu"), (u_longlong_t)guid);
2278
2279         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2280         zc.zc_guid = guid;
2281         zc.zc_cookie = VDEV_STATE_FAULTED;
2282         zc.zc_obj = aux;
2283
2284         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2285                 return (0);
2286
2287         switch (errno) {
2288         case EBUSY:
2289
2290                 /*
2291                  * There are no other replicas of this device.
2292                  */
2293                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2294
2295         default:
2296                 return (zpool_standard_error(hdl, errno, msg));
2297         }
2298
2299 }
2300
2301 /*
2302  * Mark the given vdev degraded.
2303  */
2304 int
2305 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2306 {
2307         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2308         char msg[1024];
2309         libzfs_handle_t *hdl = zhp->zpool_hdl;
2310
2311         (void) snprintf(msg, sizeof (msg),
2312            dgettext(TEXT_DOMAIN, "cannot degrade %llu"), (u_longlong_t)guid);
2313
2314         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2315         zc.zc_guid = guid;
2316         zc.zc_cookie = VDEV_STATE_DEGRADED;
2317         zc.zc_obj = aux;
2318
2319         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2320                 return (0);
2321
2322         return (zpool_standard_error(hdl, errno, msg));
2323 }
2324
2325 /*
2326  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2327  * a hot spare.
2328  */
2329 static boolean_t
2330 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2331 {
2332         nvlist_t **child;
2333         uint_t c, children;
2334         char *type;
2335
2336         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2337             &children) == 0) {
2338                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2339                     &type) == 0);
2340
2341                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2342                     children == 2 && child[which] == tgt)
2343                         return (B_TRUE);
2344
2345                 for (c = 0; c < children; c++)
2346                         if (is_replacing_spare(child[c], tgt, which))
2347                                 return (B_TRUE);
2348         }
2349
2350         return (B_FALSE);
2351 }
2352
2353 /*
2354  * Attach new_disk (fully described by nvroot) to old_disk.
2355  * If 'replacing' is specified, the new disk will replace the old one.
2356  */
2357 int
2358 zpool_vdev_attach(zpool_handle_t *zhp,
2359     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2360 {
2361         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2362         char msg[1024];
2363         int ret;
2364         nvlist_t *tgt;
2365         boolean_t avail_spare, l2cache, islog;
2366         uint64_t val;
2367         char *newname;
2368         nvlist_t **child;
2369         uint_t children;
2370         nvlist_t *config_root;
2371         libzfs_handle_t *hdl = zhp->zpool_hdl;
2372         boolean_t rootpool = zpool_is_bootable(zhp);
2373
2374         if (replacing)
2375                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2376                     "cannot replace %s with %s"), old_disk, new_disk);
2377         else
2378                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2379                     "cannot attach %s to %s"), new_disk, old_disk);
2380
2381 #if defined(__sun__) || defined(__sun)
2382         /*
2383          * If this is a root pool, make sure that we're not attaching an
2384          * EFI labeled device.
2385          */
2386         if (rootpool && pool_uses_efi(nvroot)) {
2387                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2388                     "EFI labeled devices are not supported on root pools."));
2389                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2390         }
2391 #endif
2392
2393         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2394         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2395             &islog)) == 0)
2396                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2397
2398         if (avail_spare)
2399                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2400
2401         if (l2cache)
2402                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2403
2404         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2405         zc.zc_cookie = replacing;
2406
2407         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2408             &child, &children) != 0 || children != 1) {
2409                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2410                     "new device must be a single disk"));
2411                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2412         }
2413
2414         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2415             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2416
2417         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2418                 return (-1);
2419
2420         /*
2421          * If the target is a hot spare that has been swapped in, we can only
2422          * replace it with another hot spare.
2423          */
2424         if (replacing &&
2425             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2426             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2427             NULL) == NULL || !avail_spare) &&
2428             is_replacing_spare(config_root, tgt, 1)) {
2429                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2430                     "can only be replaced by another hot spare"));
2431                 free(newname);
2432                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2433         }
2434
2435         free(newname);
2436
2437         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2438                 return (-1);
2439
2440         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2441
2442         zcmd_free_nvlists(&zc);
2443
2444         if (ret == 0) {
2445                 if (rootpool) {
2446                         /*
2447                          * XXX need a better way to prevent user from
2448                          * booting up a half-baked vdev.
2449                          */
2450                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2451                             "sure to wait until resilver is done "
2452                             "before rebooting.\n"));
2453                 }
2454                 return (0);
2455         }
2456
2457         switch (errno) {
2458         case ENOTSUP:
2459                 /*
2460                  * Can't attach to or replace this type of vdev.
2461                  */
2462                 if (replacing) {
2463                         uint64_t version = zpool_get_prop_int(zhp,
2464                             ZPOOL_PROP_VERSION, NULL);
2465
2466                         if (islog)
2467                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2468                                     "cannot replace a log with a spare"));
2469                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2470                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2471                                     "already in replacing/spare config; wait "
2472                                     "for completion or use 'zpool detach'"));
2473                         else
2474                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2475                                     "cannot replace a replacing device"));
2476                 } else {
2477                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2478                             "can only attach to mirrors and top-level "
2479                             "disks"));
2480                 }
2481                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2482                 break;
2483
2484         case EINVAL:
2485                 /*
2486                  * The new device must be a single disk.
2487                  */
2488                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2489                     "new device must be a single disk"));
2490                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2491                 break;
2492
2493         case EBUSY:
2494                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2495                     new_disk);
2496                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2497                 break;
2498
2499         case EOVERFLOW:
2500                 /*
2501                  * The new device is too small.
2502                  */
2503                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2504                     "device is too small"));
2505                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2506                 break;
2507
2508         case EDOM:
2509                 /*
2510                  * The new device has a different alignment requirement.
2511                  */
2512                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2513                     "devices have different sector alignment"));
2514                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2515                 break;
2516
2517         case ENAMETOOLONG:
2518                 /*
2519                  * The resulting top-level vdev spec won't fit in the label.
2520                  */
2521                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2522                 break;
2523
2524         default:
2525                 (void) zpool_standard_error(hdl, errno, msg);
2526         }
2527
2528         return (-1);
2529 }
2530
2531 /*
2532  * Detach the specified device.
2533  */
2534 int
2535 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2536 {
2537         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2538         char msg[1024];
2539         nvlist_t *tgt;
2540         boolean_t avail_spare, l2cache;
2541         libzfs_handle_t *hdl = zhp->zpool_hdl;
2542
2543         (void) snprintf(msg, sizeof (msg),
2544             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2545
2546         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2547         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2548             NULL)) == 0)
2549                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2550
2551         if (avail_spare)
2552                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2553
2554         if (l2cache)
2555                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2556
2557         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2558
2559         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2560                 return (0);
2561
2562         switch (errno) {
2563
2564         case ENOTSUP:
2565                 /*
2566                  * Can't detach from this type of vdev.
2567                  */
2568                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2569                     "applicable to mirror and replacing vdevs"));
2570                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2571                 break;
2572
2573         case EBUSY:
2574                 /*
2575                  * There are no other replicas of this device.
2576                  */
2577                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2578                 break;
2579
2580         default:
2581                 (void) zpool_standard_error(hdl, errno, msg);
2582         }
2583
2584         return (-1);
2585 }
2586
2587 /*
2588  * Find a mirror vdev in the source nvlist.
2589  *
2590  * The mchild array contains a list of disks in one of the top-level mirrors
2591  * of the source pool.  The schild array contains a list of disks that the
2592  * user specified on the command line.  We loop over the mchild array to
2593  * see if any entry in the schild array matches.
2594  *
2595  * If a disk in the mchild array is found in the schild array, we return
2596  * the index of that entry.  Otherwise we return -1.
2597  */
2598 static int
2599 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2600     nvlist_t **schild, uint_t schildren)
2601 {
2602         uint_t mc;
2603
2604         for (mc = 0; mc < mchildren; mc++) {
2605                 uint_t sc;
2606                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2607                     mchild[mc], B_FALSE);
2608
2609                 for (sc = 0; sc < schildren; sc++) {
2610                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2611                             schild[sc], B_FALSE);
2612                         boolean_t result = (strcmp(mpath, spath) == 0);
2613
2614                         free(spath);
2615                         if (result) {
2616                                 free(mpath);
2617                                 return (mc);
2618                         }
2619                 }
2620
2621                 free(mpath);
2622         }
2623
2624         return (-1);
2625 }
2626
2627 /*
2628  * Split a mirror pool.  If newroot points to null, then a new nvlist
2629  * is generated and it is the responsibility of the caller to free it.
2630  */
2631 int
2632 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2633     nvlist_t *props, splitflags_t flags)
2634 {
2635         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2636         char msg[1024];
2637         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2638         nvlist_t **varray = NULL, *zc_props = NULL;
2639         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2640         libzfs_handle_t *hdl = zhp->zpool_hdl;
2641         uint64_t vers;
2642         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2643         int retval = 0;
2644
2645         (void) snprintf(msg, sizeof (msg),
2646             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2647
2648         if (!zpool_name_valid(hdl, B_FALSE, newname))
2649                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2650
2651         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2652                 (void) fprintf(stderr, gettext("Internal error: unable to "
2653                     "retrieve pool configuration\n"));
2654                 return (-1);
2655         }
2656
2657         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2658             == 0);
2659         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2660
2661         if (props) {
2662                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2663                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2664                     props, vers, flags, msg)) == NULL)
2665                         return (-1);
2666         }
2667
2668         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2669             &children) != 0) {
2670                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2671                     "Source pool is missing vdev tree"));
2672                 if (zc_props)
2673                         nvlist_free(zc_props);
2674                 return (-1);
2675         }
2676
2677         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2678         vcount = 0;
2679
2680         if (*newroot == NULL ||
2681             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2682             &newchild, &newchildren) != 0)
2683                 newchildren = 0;
2684
2685         for (c = 0; c < children; c++) {
2686                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2687                 char *type;
2688                 nvlist_t **mchild, *vdev;
2689                 uint_t mchildren;
2690                 int entry;
2691
2692                 /*
2693                  * Unlike cache & spares, slogs are stored in the
2694                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2695                  */
2696                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2697                     &is_log);
2698                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2699                     &is_hole);
2700                 if (is_log || is_hole) {
2701                         /*
2702                          * Create a hole vdev and put it in the config.
2703                          */
2704                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2705                                 goto out;
2706                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2707                             VDEV_TYPE_HOLE) != 0)
2708                                 goto out;
2709                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2710                             1) != 0)
2711                                 goto out;
2712                         if (lastlog == 0)
2713                                 lastlog = vcount;
2714                         varray[vcount++] = vdev;
2715                         continue;
2716                 }
2717                 lastlog = 0;
2718                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2719                     == 0);
2720                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2721                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2722                             "Source pool must be composed only of mirrors\n"));
2723                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2724                         goto out;
2725                 }
2726
2727                 verify(nvlist_lookup_nvlist_array(child[c],
2728                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2729
2730                 /* find or add an entry for this top-level vdev */
2731                 if (newchildren > 0 &&
2732                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2733                     newchild, newchildren)) >= 0) {
2734                         /* We found a disk that the user specified. */
2735                         vdev = mchild[entry];
2736                         ++found;
2737                 } else {
2738                         /* User didn't specify a disk for this vdev. */
2739                         vdev = mchild[mchildren - 1];
2740                 }
2741
2742                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2743                         goto out;
2744         }
2745
2746         /* did we find every disk the user specified? */
2747         if (found != newchildren) {
2748                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
2749                     "include at most one disk from each mirror"));
2750                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2751                 goto out;
2752         }
2753
2754         /* Prepare the nvlist for populating. */
2755         if (*newroot == NULL) {
2756                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
2757                         goto out;
2758                 freelist = B_TRUE;
2759                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
2760                     VDEV_TYPE_ROOT) != 0)
2761                         goto out;
2762         } else {
2763                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
2764         }
2765
2766         /* Add all the children we found */
2767         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
2768             lastlog == 0 ? vcount : lastlog) != 0)
2769                 goto out;
2770
2771         /*
2772          * If we're just doing a dry run, exit now with success.
2773          */
2774         if (flags.dryrun) {
2775                 memory_err = B_FALSE;
2776                 freelist = B_FALSE;
2777                 goto out;
2778         }
2779
2780         /* now build up the config list & call the ioctl */
2781         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
2782                 goto out;
2783
2784         if (nvlist_add_nvlist(newconfig,
2785             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
2786             nvlist_add_string(newconfig,
2787             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
2788             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
2789                 goto out;
2790
2791         /*
2792          * The new pool is automatically part of the namespace unless we
2793          * explicitly export it.
2794          */
2795         if (!flags.import)
2796                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
2797         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2798         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
2799         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
2800                 goto out;
2801         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
2802                 goto out;
2803
2804         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
2805                 retval = zpool_standard_error(hdl, errno, msg);
2806                 goto out;
2807         }
2808
2809         freelist = B_FALSE;
2810         memory_err = B_FALSE;
2811
2812 out:
2813         if (varray != NULL) {
2814                 int v;
2815
2816                 for (v = 0; v < vcount; v++)
2817                         nvlist_free(varray[v]);
2818                 free(varray);
2819         }
2820         zcmd_free_nvlists(&zc);
2821         if (zc_props)
2822                 nvlist_free(zc_props);
2823         if (newconfig)
2824                 nvlist_free(newconfig);
2825         if (freelist) {
2826                 nvlist_free(*newroot);
2827                 *newroot = NULL;
2828         }
2829
2830         if (retval != 0)
2831                 return (retval);
2832
2833         if (memory_err)
2834                 return (no_memory(hdl));
2835
2836         return (0);
2837 }
2838
2839 /*
2840  * Remove the given device.  Currently, this is supported only for hot spares
2841  * and level 2 cache devices.
2842  */
2843 int
2844 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
2845 {
2846         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2847         char msg[1024];
2848         nvlist_t *tgt;
2849         boolean_t avail_spare, l2cache, islog;
2850         libzfs_handle_t *hdl = zhp->zpool_hdl;
2851         uint64_t version;
2852
2853         (void) snprintf(msg, sizeof (msg),
2854             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
2855
2856         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2857         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2858             &islog)) == 0)
2859                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2860         /*
2861          * XXX - this should just go away.
2862          */
2863         if (!avail_spare && !l2cache && !islog) {
2864                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2865                     "only inactive hot spares, cache, top-level, "
2866                     "or log devices can be removed"));
2867                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2868         }
2869
2870         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
2871         if (islog && version < SPA_VERSION_HOLES) {
2872                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2873                     "pool must be upgrade to support log removal"));
2874                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
2875         }
2876
2877         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2878
2879         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
2880                 return (0);
2881
2882         return (zpool_standard_error(hdl, errno, msg));
2883 }
2884
2885 /*
2886  * Clear the errors for the pool, or the particular device if specified.
2887  */
2888 int
2889 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
2890 {
2891         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2892         char msg[1024];
2893         nvlist_t *tgt;
2894         zpool_rewind_policy_t policy;
2895         boolean_t avail_spare, l2cache;
2896         libzfs_handle_t *hdl = zhp->zpool_hdl;
2897         nvlist_t *nvi = NULL;
2898         int error;
2899
2900         if (path)
2901                 (void) snprintf(msg, sizeof (msg),
2902                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2903                     path);
2904         else
2905                 (void) snprintf(msg, sizeof (msg),
2906                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2907                     zhp->zpool_name);
2908
2909         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2910         if (path) {
2911                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
2912                     &l2cache, NULL)) == 0)
2913                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
2914
2915                 /*
2916                  * Don't allow error clearing for hot spares.  Do allow
2917                  * error clearing for l2cache devices.
2918                  */
2919                 if (avail_spare)
2920                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
2921
2922                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
2923                     &zc.zc_guid) == 0);
2924         }
2925
2926         zpool_get_rewind_policy(rewindnvl, &policy);
2927         zc.zc_cookie = policy.zrp_request;
2928
2929         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
2930                 return (-1);
2931
2932         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
2933                 return (-1);
2934
2935         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
2936             errno == ENOMEM) {
2937                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
2938                         zcmd_free_nvlists(&zc);
2939                         return (-1);
2940                 }
2941         }
2942
2943         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
2944             errno != EPERM && errno != EACCES)) {
2945                 if (policy.zrp_request &
2946                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
2947                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
2948                         zpool_rewind_exclaim(hdl, zc.zc_name,
2949                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
2950                             nvi);
2951                         nvlist_free(nvi);
2952                 }
2953                 zcmd_free_nvlists(&zc);
2954                 return (0);
2955         }
2956
2957         zcmd_free_nvlists(&zc);
2958         return (zpool_standard_error(hdl, errno, msg));
2959 }
2960
2961 /*
2962  * Similar to zpool_clear(), but takes a GUID (used by fmd).
2963  */
2964 int
2965 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
2966 {
2967         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2968         char msg[1024];
2969         libzfs_handle_t *hdl = zhp->zpool_hdl;
2970
2971         (void) snprintf(msg, sizeof (msg),
2972             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
2973            (u_longlong_t)guid);
2974
2975         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2976         zc.zc_guid = guid;
2977         zc.zc_cookie = ZPOOL_NO_REWIND;
2978
2979         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
2980                 return (0);
2981
2982         return (zpool_standard_error(hdl, errno, msg));
2983 }
2984
2985 /*
2986  * Change the GUID for a pool.
2987  */
2988 int
2989 zpool_reguid(zpool_handle_t *zhp)
2990 {
2991         char msg[1024];
2992         libzfs_handle_t *hdl = zhp->zpool_hdl;
2993         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2994
2995         (void) snprintf(msg, sizeof (msg),
2996             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
2997
2998         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2999         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3000                 return (0);
3001
3002         return (zpool_standard_error(hdl, errno, msg));
3003 }
3004
3005 /*
3006  * Reopen the pool.
3007  */
3008 int
3009 zpool_reopen(zpool_handle_t *zhp)
3010 {
3011         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3012         char msg[1024];
3013         libzfs_handle_t *hdl = zhp->zpool_hdl;
3014
3015         (void) snprintf(msg, sizeof (msg),
3016             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3017             zhp->zpool_name);
3018
3019         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3020         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3021                 return (0);
3022         return (zpool_standard_error(hdl, errno, msg));
3023 }
3024
3025 /*
3026  * Convert from a devid string to a path.
3027  */
3028 static char *
3029 devid_to_path(char *devid_str)
3030 {
3031         ddi_devid_t devid;
3032         char *minor;
3033         char *path;
3034         devid_nmlist_t *list = NULL;
3035         int ret;
3036
3037         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3038                 return (NULL);
3039
3040         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3041
3042         devid_str_free(minor);
3043         devid_free(devid);
3044
3045         if (ret != 0)
3046                 return (NULL);
3047
3048         if ((path = strdup(list[0].devname)) == NULL)
3049                 return (NULL);
3050
3051         devid_free_nmlist(list);
3052
3053         return (path);
3054 }
3055
3056 /*
3057  * Convert from a path to a devid string.
3058  */
3059 static char *
3060 path_to_devid(const char *path)
3061 {
3062         int fd;
3063         ddi_devid_t devid;
3064         char *minor, *ret;
3065
3066         if ((fd = open(path, O_RDONLY)) < 0)
3067                 return (NULL);
3068
3069         minor = NULL;
3070         ret = NULL;
3071         if (devid_get(fd, &devid) == 0) {
3072                 if (devid_get_minor_name(fd, &minor) == 0)
3073                         ret = devid_str_encode(devid, minor);
3074                 if (minor != NULL)
3075                         devid_str_free(minor);
3076                 devid_free(devid);
3077         }
3078         (void) close(fd);
3079
3080         return (ret);
3081 }
3082
3083 /*
3084  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3085  * ignore any failure here, since a common case is for an unprivileged user to
3086  * type 'zpool status', and we'll display the correct information anyway.
3087  */
3088 static void
3089 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3090 {
3091         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3092
3093         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3094         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3095         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3096             &zc.zc_guid) == 0);
3097
3098         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3099 }
3100
3101 /*
3102  * Remove partition suffix from a vdev path.  Partition suffixes may take three
3103  * forms: "-partX", "pX", or "X", where X is a string of digits.  The second
3104  * case only occurs when the suffix is preceded by a digit, i.e. "md0p0" The
3105  * third case only occurs when preceded by a string matching the regular
3106  * expression "^[hs]d[a-z]+", i.e. a scsi or ide disk.
3107  */
3108 static char *
3109 strip_partition(libzfs_handle_t *hdl, char *path)
3110 {
3111         char *tmp = zfs_strdup(hdl, path);
3112         char *part = NULL, *d = NULL;
3113
3114         if ((part = strstr(tmp, "-part")) && part != tmp) {
3115                 d = part + 5;
3116         } else if ((part = strrchr(tmp, 'p')) &&
3117             part > tmp + 1 && isdigit(*(part-1))) {
3118                 d = part + 1;
3119         } else if ((tmp[0] == 'h' || tmp[0] == 's') && tmp[1] == 'd') {
3120                 for (d = &tmp[2]; isalpha(*d); part = ++d);
3121         }
3122         if (part && d && *d != '\0') {
3123                 for (; isdigit(*d); d++);
3124                 if (*d == '\0')
3125                         *part = '\0';
3126         }
3127         return (tmp);
3128 }
3129
3130 #define PATH_BUF_LEN    64
3131
3132 /*
3133  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3134  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3135  * We also check if this is a whole disk, in which case we strip off the
3136  * trailing 's0' slice name.
3137  *
3138  * This routine is also responsible for identifying when disks have been
3139  * reconfigured in a new location.  The kernel will have opened the device by
3140  * devid, but the path will still refer to the old location.  To catch this, we
3141  * first do a path -> devid translation (which is fast for the common case).  If
3142  * the devid matches, we're done.  If not, we do a reverse devid -> path
3143  * translation and issue the appropriate ioctl() to update the path of the vdev.
3144  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3145  * of these checks.
3146  */
3147 char *
3148 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3149     boolean_t verbose)
3150 {
3151         char *path, *devid, *type;
3152         uint64_t value;
3153         char buf[PATH_BUF_LEN];
3154         char tmpbuf[PATH_BUF_LEN];
3155         vdev_stat_t *vs;
3156         uint_t vsc;
3157
3158         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
3159             &value) == 0) {
3160                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3161                     &value) == 0);
3162                 (void) snprintf(buf, sizeof (buf), "%llu",
3163                     (u_longlong_t)value);
3164                 path = buf;
3165         } else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
3166                 /*
3167                  * If the device is dead (faulted, offline, etc) then don't
3168                  * bother opening it.  Otherwise we may be forcing the user to
3169                  * open a misbehaving device, which can have undesirable
3170                  * effects.
3171                  */
3172                 if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3173                     (uint64_t **)&vs, &vsc) != 0 ||
3174                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3175                     zhp != NULL &&
3176                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3177                         /*
3178                          * Determine if the current path is correct.
3179                          */
3180                         char *newdevid = path_to_devid(path);
3181
3182                         if (newdevid == NULL ||
3183                             strcmp(devid, newdevid) != 0) {
3184                                 char *newpath;
3185
3186                                 if ((newpath = devid_to_path(devid)) != NULL) {
3187                                         /*
3188                                          * Update the path appropriately.
3189                                          */
3190                                         set_path(zhp, nv, newpath);
3191                                         if (nvlist_add_string(nv,
3192                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3193                                                 verify(nvlist_lookup_string(nv,
3194                                                     ZPOOL_CONFIG_PATH,
3195                                                     &path) == 0);
3196                                         free(newpath);
3197                                 }
3198                         }
3199
3200                         if (newdevid)
3201                                 devid_str_free(newdevid);
3202                 }
3203
3204                 /*
3205                  * For a block device only use the name.
3206                  */
3207                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
3208                 if (strcmp(type, VDEV_TYPE_DISK) == 0) {
3209                         path = strrchr(path, '/');
3210                         path++;
3211                 }
3212
3213                 /*
3214                  * Remove the partition from the path it this is a whole disk.
3215                  */
3216                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3217                     &value) == 0 && value) {
3218                         return strip_partition(hdl, path);
3219                 }
3220         } else {
3221                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3222
3223                 /*
3224                  * If it's a raidz device, we need to stick in the parity level.
3225                  */
3226                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3227
3228                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3229                             &value) == 0);
3230                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3231                             (u_longlong_t)value);
3232                         path = buf;
3233                 }
3234
3235                 /*
3236                  * We identify each top-level vdev by using a <type-id>
3237                  * naming convention.
3238                  */
3239                 if (verbose) {
3240                         uint64_t id;
3241
3242                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3243                             &id) == 0);
3244                         (void) snprintf(tmpbuf, sizeof (tmpbuf), "%s-%llu",
3245                             path, (u_longlong_t)id);
3246                         path = tmpbuf;
3247                 }
3248         }
3249
3250         return (zfs_strdup(hdl, path));
3251 }
3252
3253 static int
3254 zbookmark_compare(const void *a, const void *b)
3255 {
3256         return (memcmp(a, b, sizeof (zbookmark_t)));
3257 }
3258
3259 /*
3260  * Retrieve the persistent error log, uniquify the members, and return to the
3261  * caller.
3262  */
3263 int
3264 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3265 {
3266         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3267         uint64_t count;
3268         zbookmark_t *zb = NULL;
3269         int i;
3270
3271         /*
3272          * Retrieve the raw error list from the kernel.  If the number of errors
3273          * has increased, allocate more space and continue until we get the
3274          * entire list.
3275          */
3276         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3277             &count) == 0);
3278         if (count == 0)
3279                 return (0);
3280         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3281             count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
3282                 return (-1);
3283         zc.zc_nvlist_dst_size = count;
3284         (void) strcpy(zc.zc_name, zhp->zpool_name);
3285         for (;;) {
3286                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3287                     &zc) != 0) {
3288                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3289                         if (errno == ENOMEM) {
3290                                 count = zc.zc_nvlist_dst_size;
3291                                 if ((zc.zc_nvlist_dst = (uintptr_t)
3292                                     zfs_alloc(zhp->zpool_hdl, count *
3293                                     sizeof (zbookmark_t))) == (uintptr_t)NULL)
3294                                         return (-1);
3295                         } else {
3296                                 return (-1);
3297                         }
3298                 } else {
3299                         break;
3300                 }
3301         }
3302
3303         /*
3304          * Sort the resulting bookmarks.  This is a little confusing due to the
3305          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3306          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3307          * _not_ copied as part of the process.  So we point the start of our
3308          * array appropriate and decrement the total number of elements.
3309          */
3310         zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
3311             zc.zc_nvlist_dst_size;
3312         count -= zc.zc_nvlist_dst_size;
3313
3314         qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
3315
3316         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3317
3318         /*
3319          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3320          */
3321         for (i = 0; i < count; i++) {
3322                 nvlist_t *nv;
3323
3324                 /* ignoring zb_blkid and zb_level for now */
3325                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3326                     zb[i-1].zb_object == zb[i].zb_object)
3327                         continue;
3328
3329                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3330                         goto nomem;
3331                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3332                     zb[i].zb_objset) != 0) {
3333                         nvlist_free(nv);
3334                         goto nomem;
3335                 }
3336                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3337                     zb[i].zb_object) != 0) {
3338                         nvlist_free(nv);
3339                         goto nomem;
3340                 }
3341                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3342                         nvlist_free(nv);
3343                         goto nomem;
3344                 }
3345                 nvlist_free(nv);
3346         }
3347
3348         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3349         return (0);
3350
3351 nomem:
3352         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3353         return (no_memory(zhp->zpool_hdl));
3354 }
3355
3356 /*
3357  * Upgrade a ZFS pool to the latest on-disk version.
3358  */
3359 int
3360 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3361 {
3362         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3363         libzfs_handle_t *hdl = zhp->zpool_hdl;
3364
3365         (void) strcpy(zc.zc_name, zhp->zpool_name);
3366         zc.zc_cookie = new_version;
3367
3368         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3369                 return (zpool_standard_error_fmt(hdl, errno,
3370                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3371                     zhp->zpool_name));
3372         return (0);
3373 }
3374
3375 void
3376 zpool_set_history_str(const char *subcommand, int argc, char **argv,
3377     char *history_str)
3378 {
3379         int i;
3380
3381         (void) strlcpy(history_str, subcommand, HIS_MAX_RECORD_LEN);
3382         for (i = 1; i < argc; i++) {
3383                 if (strlen(history_str) + 1 + strlen(argv[i]) >
3384                     HIS_MAX_RECORD_LEN)
3385                         break;
3386                 (void) strlcat(history_str, " ", HIS_MAX_RECORD_LEN);
3387                 (void) strlcat(history_str, argv[i], HIS_MAX_RECORD_LEN);
3388         }
3389 }
3390
3391 /*
3392  * Stage command history for logging.
3393  */
3394 int
3395 zpool_stage_history(libzfs_handle_t *hdl, const char *history_str)
3396 {
3397         if (history_str == NULL)
3398                 return (EINVAL);
3399
3400         if (strlen(history_str) > HIS_MAX_RECORD_LEN)
3401                 return (EINVAL);
3402
3403         if (hdl->libzfs_log_str != NULL)
3404                 free(hdl->libzfs_log_str);
3405
3406         if ((hdl->libzfs_log_str = strdup(history_str)) == NULL)
3407                 return (no_memory(hdl));
3408
3409         return (0);
3410 }
3411
3412 /*
3413  * Perform ioctl to get some command history of a pool.
3414  *
3415  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3416  * logical offset of the history buffer to start reading from.
3417  *
3418  * Upon return, 'off' is the next logical offset to read from and
3419  * 'len' is the actual amount of bytes read into 'buf'.
3420  */
3421 static int
3422 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3423 {
3424         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3425         libzfs_handle_t *hdl = zhp->zpool_hdl;
3426
3427         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3428
3429         zc.zc_history = (uint64_t)(uintptr_t)buf;
3430         zc.zc_history_len = *len;
3431         zc.zc_history_offset = *off;
3432
3433         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3434                 switch (errno) {
3435                 case EPERM:
3436                         return (zfs_error_fmt(hdl, EZFS_PERM,
3437                             dgettext(TEXT_DOMAIN,
3438                             "cannot show history for pool '%s'"),
3439                             zhp->zpool_name));
3440                 case ENOENT:
3441                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3442                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3443                             "'%s'"), zhp->zpool_name));
3444                 case ENOTSUP:
3445                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3446                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3447                             "'%s', pool must be upgraded"), zhp->zpool_name));
3448                 default:
3449                         return (zpool_standard_error_fmt(hdl, errno,
3450                             dgettext(TEXT_DOMAIN,
3451                             "cannot get history for '%s'"), zhp->zpool_name));
3452                 }
3453         }
3454
3455         *len = zc.zc_history_len;
3456         *off = zc.zc_history_offset;
3457
3458         return (0);
3459 }
3460
3461 /*
3462  * Process the buffer of nvlists, unpacking and storing each nvlist record
3463  * into 'records'.  'leftover' is set to the number of bytes that weren't
3464  * processed as there wasn't a complete record.
3465  */
3466 int
3467 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3468     nvlist_t ***records, uint_t *numrecords)
3469 {
3470         uint64_t reclen;
3471         nvlist_t *nv;
3472         int i;
3473
3474         while (bytes_read > sizeof (reclen)) {
3475
3476                 /* get length of packed record (stored as little endian) */
3477                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3478                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3479
3480                 if (bytes_read < sizeof (reclen) + reclen)
3481                         break;
3482
3483                 /* unpack record */
3484                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3485                         return (ENOMEM);
3486                 bytes_read -= sizeof (reclen) + reclen;
3487                 buf += sizeof (reclen) + reclen;
3488
3489                 /* add record to nvlist array */
3490                 (*numrecords)++;
3491                 if (ISP2(*numrecords + 1)) {
3492                         *records = realloc(*records,
3493                             *numrecords * 2 * sizeof (nvlist_t *));
3494                 }
3495                 (*records)[*numrecords - 1] = nv;
3496         }
3497
3498         *leftover = bytes_read;
3499         return (0);
3500 }
3501
3502 #define HIS_BUF_LEN     (128*1024)
3503
3504 /*
3505  * Retrieve the command history of a pool.
3506  */
3507 int
3508 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3509 {
3510         char buf[HIS_BUF_LEN];
3511         uint64_t off = 0;
3512         nvlist_t **records = NULL;
3513         uint_t numrecords = 0;
3514         int err, i;
3515
3516         do {
3517                 uint64_t bytes_read = sizeof (buf);
3518                 uint64_t leftover;
3519
3520                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3521                         break;
3522
3523                 /* if nothing else was read in, we're at EOF, just return */
3524                 if (!bytes_read)
3525                         break;
3526
3527                 if ((err = zpool_history_unpack(buf, bytes_read,
3528                     &leftover, &records, &numrecords)) != 0)
3529                         break;
3530                 off -= leftover;
3531
3532                 /* CONSTCOND */
3533         } while (1);
3534
3535         if (!err) {
3536                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3537                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3538                     records, numrecords) == 0);
3539         }
3540         for (i = 0; i < numrecords; i++)
3541                 nvlist_free(records[i]);
3542         free(records);
3543
3544         return (err);
3545 }
3546
3547 /*
3548  * Retrieve the next event.  If there is a new event available 'nvp' will
3549  * contain a newly allocated nvlist and 'dropped' will be set to the number
3550  * of missed events since the last call to this function.  When 'nvp' is
3551  * set to NULL it indicates no new events are available.  In either case
3552  * the function returns 0 and it is up to the caller to free 'nvp'.  In
3553  * the case of a fatal error the function will return a non-zero value.
3554  * When the function is called in blocking mode it will not return until
3555  * a new event is available.
3556  */
3557 int
3558 zpool_events_next(libzfs_handle_t *hdl, nvlist_t **nvp,
3559     int *dropped, int block, int cleanup_fd)
3560 {
3561         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3562         int error = 0;
3563
3564         *nvp = NULL;
3565         *dropped = 0;
3566         zc.zc_cleanup_fd = cleanup_fd;
3567
3568         if (!block)
3569                 zc.zc_guid = ZEVENT_NONBLOCK;
3570
3571         if (zcmd_alloc_dst_nvlist(hdl, &zc, ZEVENT_SIZE) != 0)
3572                 return (-1);
3573
3574 retry:
3575         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_NEXT, &zc) != 0) {
3576                 switch (errno) {
3577                 case ESHUTDOWN:
3578                         error = zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
3579                             dgettext(TEXT_DOMAIN, "zfs shutdown"));
3580                         goto out;
3581                 case ENOENT:
3582                         /* Blocking error case should not occur */
3583                         if (block)
3584                                 error = zpool_standard_error_fmt(hdl, errno,
3585                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3586
3587                         goto out;
3588                 case ENOMEM:
3589                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3590                                 error = zfs_error_fmt(hdl, EZFS_NOMEM,
3591                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3592                                 goto out;
3593                         } else {
3594                                 goto retry;
3595                         }
3596                 default:
3597                         error = zpool_standard_error_fmt(hdl, errno,
3598                             dgettext(TEXT_DOMAIN, "cannot get event"));
3599                         goto out;
3600                 }
3601         }
3602
3603         error = zcmd_read_dst_nvlist(hdl, &zc, nvp);
3604         if (error != 0)
3605                 goto out;
3606
3607         *dropped = (int)zc.zc_cookie;
3608 out:
3609         zcmd_free_nvlists(&zc);
3610
3611         return (error);
3612 }
3613
3614 /*
3615  * Clear all events.
3616  */
3617 int
3618 zpool_events_clear(libzfs_handle_t *hdl, int *count)
3619 {
3620         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3621         char msg[1024];
3622
3623         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
3624             "cannot clear events"));
3625
3626         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_CLEAR, &zc) != 0)
3627                 return (zpool_standard_error_fmt(hdl, errno, msg));
3628
3629         if (count != NULL)
3630                 *count = (int)zc.zc_cookie; /* # of events cleared */
3631
3632         return (0);
3633 }
3634
3635 void
3636 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3637     char *pathname, size_t len)
3638 {
3639         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3640         boolean_t mounted = B_FALSE;
3641         char *mntpnt = NULL;
3642         char dsname[MAXNAMELEN];
3643
3644         if (dsobj == 0) {
3645                 /* special case for the MOS */
3646                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", (longlong_t)obj);
3647                 return;
3648         }
3649
3650         /* get the dataset's name */
3651         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3652         zc.zc_obj = dsobj;
3653         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3654             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3655                 /* just write out a path of two object numbers */
3656                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3657                     (longlong_t)dsobj, (longlong_t)obj);
3658                 return;
3659         }
3660         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3661
3662         /* find out if the dataset is mounted */
3663         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3664
3665         /* get the corrupted object's path */
3666         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3667         zc.zc_obj = obj;
3668         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3669             &zc) == 0) {
3670                 if (mounted) {
3671                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3672                             zc.zc_value);
3673                 } else {
3674                         (void) snprintf(pathname, len, "%s:%s",
3675                             dsname, zc.zc_value);
3676                 }
3677         } else {
3678                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, (longlong_t)obj);
3679         }
3680         free(mntpnt);
3681 }
3682
3683 /*
3684  * Read the EFI label from the config, if a label does not exist then
3685  * pass back the error to the caller. If the caller has passed a non-NULL
3686  * diskaddr argument then we set it to the starting address of the EFI
3687  * partition.
3688  */
3689 static int
3690 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3691 {
3692         char *path;
3693         int fd;
3694         char diskname[MAXPATHLEN];
3695         int err = -1;
3696
3697         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3698                 return (err);
3699
3700         (void) snprintf(diskname, sizeof (diskname), "%s%s", DISK_ROOT,
3701             strrchr(path, '/'));
3702         if ((fd = open(diskname, O_RDWR|O_DIRECT)) >= 0) {
3703                 struct dk_gpt *vtoc;
3704
3705                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3706                         if (sb != NULL)
3707                                 *sb = vtoc->efi_parts[0].p_start;
3708                         efi_free(vtoc);
3709                 }
3710                 (void) close(fd);
3711         }
3712         return (err);
3713 }
3714
3715 /*
3716  * determine where a partition starts on a disk in the current
3717  * configuration
3718  */
3719 static diskaddr_t
3720 find_start_block(nvlist_t *config)
3721 {
3722         nvlist_t **child;
3723         uint_t c, children;
3724         diskaddr_t sb = MAXOFFSET_T;
3725         uint64_t wholedisk;
3726
3727         if (nvlist_lookup_nvlist_array(config,
3728             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3729                 if (nvlist_lookup_uint64(config,
3730                     ZPOOL_CONFIG_WHOLE_DISK,
3731                     &wholedisk) != 0 || !wholedisk) {
3732                         return (MAXOFFSET_T);
3733                 }
3734                 if (read_efi_label(config, &sb) < 0)
3735                         sb = MAXOFFSET_T;
3736                 return (sb);
3737         }
3738
3739         for (c = 0; c < children; c++) {
3740                 sb = find_start_block(child[c]);
3741                 if (sb != MAXOFFSET_T) {
3742                         return (sb);
3743                 }
3744         }
3745         return (MAXOFFSET_T);
3746 }
3747
3748 int
3749 zpool_label_disk_wait(char *path, int timeout)
3750 {
3751         struct stat64 statbuf;
3752         int i;
3753
3754         /*
3755          * Wait timeout miliseconds for a newly created device to be available
3756          * from the given path.  There is a small window when a /dev/ device
3757          * will exist and the udev link will not, so we must wait for the
3758          * symlink.  Depending on the udev rules this may take a few seconds.
3759          */
3760         for (i = 0; i < timeout; i++) {
3761                 usleep(1000);
3762
3763                 errno = 0;
3764                 if ((stat64(path, &statbuf) == 0) && (errno == 0))
3765                         return (0);
3766         }
3767
3768         return (ENOENT);
3769 }
3770
3771 int
3772 zpool_label_disk_check(char *path)
3773 {
3774         struct dk_gpt *vtoc;
3775         int fd, err;
3776
3777         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0)
3778                 return errno;
3779
3780         if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
3781                 (void) close(fd);
3782                 return err;
3783         }
3784
3785         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
3786                 efi_free(vtoc);
3787                 (void) close(fd);
3788                 return EIDRM;
3789         }
3790
3791         efi_free(vtoc);
3792         (void) close(fd);
3793         return 0;
3794 }
3795
3796 /*
3797  * Label an individual disk.  The name provided is the short name,
3798  * stripped of any leading /dev path.
3799  */
3800 int
3801 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
3802 {
3803         char path[MAXPATHLEN];
3804         struct dk_gpt *vtoc;
3805         int rval, fd;
3806         size_t resv = EFI_MIN_RESV_SIZE;
3807         uint64_t slice_size;
3808         diskaddr_t start_block;
3809         char errbuf[1024];
3810
3811         /* prepare an error message just in case */
3812         (void) snprintf(errbuf, sizeof (errbuf),
3813             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3814
3815         if (zhp) {
3816                 nvlist_t *nvroot;
3817
3818 #if defined(__sun__) || defined(__sun)
3819                 if (zpool_is_bootable(zhp)) {
3820                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3821                             "EFI labeled devices are not supported on root "
3822                             "pools."));
3823                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3824                 }
3825 #endif
3826
3827                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3828                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3829
3830                 if (zhp->zpool_start_block == 0)
3831                         start_block = find_start_block(nvroot);
3832                 else
3833                         start_block = zhp->zpool_start_block;
3834                 zhp->zpool_start_block = start_block;
3835         } else {
3836                 /* new pool */
3837                 start_block = NEW_START_BLOCK;
3838         }
3839
3840         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
3841
3842         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0) {
3843                 /*
3844                  * This shouldn't happen.  We've long since verified that this
3845                  * is a valid device.
3846                  */
3847                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
3848                     "label '%s': unable to open device: %d"), path, errno);
3849                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3850         }
3851
3852         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3853                 /*
3854                  * The only way this can fail is if we run out of memory, or we
3855                  * were unable to read the disk's capacity
3856                  */
3857                 if (errno == ENOMEM)
3858                         (void) no_memory(hdl);
3859
3860                 (void) close(fd);
3861                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
3862                     "label '%s': unable to read disk capacity"), path);
3863
3864                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3865         }
3866
3867         slice_size = vtoc->efi_last_u_lba + 1;
3868         slice_size -= EFI_MIN_RESV_SIZE;
3869         if (start_block == MAXOFFSET_T)
3870                 start_block = NEW_START_BLOCK;
3871         slice_size -= start_block;
3872         slice_size = P2ALIGN(slice_size, PARTITION_END_ALIGNMENT);
3873
3874         vtoc->efi_parts[0].p_start = start_block;
3875         vtoc->efi_parts[0].p_size = slice_size;
3876
3877         /*
3878          * Why we use V_USR: V_BACKUP confuses users, and is considered
3879          * disposable by some EFI utilities (since EFI doesn't have a backup
3880          * slice).  V_UNASSIGNED is supposed to be used only for zero size
3881          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3882          * etc. were all pretty specific.  V_USR is as close to reality as we
3883          * can get, in the absence of V_OTHER.
3884          */
3885         vtoc->efi_parts[0].p_tag = V_USR;
3886         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3887
3888         vtoc->efi_parts[8].p_start = slice_size + start_block;
3889         vtoc->efi_parts[8].p_size = resv;
3890         vtoc->efi_parts[8].p_tag = V_RESERVED;
3891
3892         if ((rval = efi_write(fd, vtoc)) != 0 || (rval = efi_rescan(fd)) != 0) {
3893                 /*
3894                  * Some block drivers (like pcata) may not support EFI
3895                  * GPT labels.  Print out a helpful error message dir-
3896                  * ecting the user to manually label the disk and give
3897                  * a specific slice.
3898                  */
3899                 (void) close(fd);
3900                 efi_free(vtoc);
3901
3902                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "try using "
3903                     "parted(8) and then provide a specific slice: %d"), rval);
3904                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3905         }
3906
3907         (void) close(fd);
3908         efi_free(vtoc);
3909
3910         /* Wait for the first expected partition to appear. */
3911
3912         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
3913         (void) zfs_append_partition(path, MAXPATHLEN);
3914
3915         rval = zpool_label_disk_wait(path, 3000);
3916         if (rval) {
3917                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "failed to "
3918                     "detect device partitions on '%s': %d"), path, rval);
3919                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3920         }
3921
3922         /* We can't be to paranoid.  Read the label back and verify it. */
3923         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
3924         rval = zpool_label_disk_check(path);
3925         if (rval) {
3926                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "freshly written "
3927                     "EFI label on '%s' is damaged.  Ensure\nthis device "
3928                     "is not in in use, and is functioning properly: %d"),
3929                     path, rval);
3930                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3931         }
3932
3933         return 0;
3934 }